Skip to content

[Linux] Uninitialized bytes being passed to write() #82

Description

@urlofmar

On Linux, valgrind complains about uninitialized bytes being passed to write():

==14689== Syscall param write(buf) points to uninitialised byte(s)
==14689==    at 0x5C6132F: __libc_write (write.c:26)
==14689==    by 0x5C6132F: write (write.c:24)
==14689==    by 0x5656A09: cppcoro::detail::linux::message_queue::enqueue_message(void*, cppcoro::detail::linux::message_type) (linux.cpp:78)
==14689==    by 0x564FB0B: cppcoro::io_service::post_wake_up_event() (io_service.cpp:748)
==14689==    by 0x564F564: cppcoro::io_service::stop() (io_service.cpp:454)
...
==14689==  Address 0x1ffefff044 is on thread 1's stack
==14689==  in frame #1, created by cppcoro::detail::linux::message_queue::enqueue_message(void*, cppcoro::detail::linux::message_type) (linux.cpp:74)

The write() call is in

bool message_queue::enqueue_message(void* msg, message_type type)
{
    message qmsg;
    qmsg.m_type = type;
    qmsg.m_ptr = msg;
    int status = write(m_pipefd[1], (const char*)&qmsg, sizeof(message));
    return status==-1?false:true;
}

where

struct message
{
    enum message_type m_type;
    void* m_ptr;
};

has padding bytes between its two members. These bytes are not initialized in qmsg, but still being written.

So the easiest solution would be to initialize all of qmsg's bytes:

    message qmsg;
+   std::memset(&qmsg, 0, sizeof(qmsg));
    qmsg.m_type = type;
    qmsg.m_ptr = msg;

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions