]> git.ipfire.org Git - thirdparty/kernel/linux.git/commit
drm/syncobj: Fix handle <-> fd ioctls with dirty stack
authorJulian Orth <ju.orth@gmail.com>
Sun, 1 Mar 2026 12:34:42 +0000 (13:34 +0100)
committerChristian König <christian.koenig@amd.com>
Mon, 2 Mar 2026 15:52:02 +0000 (16:52 +0100)
commit2e3649e237237258a08d75afef96648dd2b379f7
tree45c12efaeec8d9217e7d0c6588d329485409bb30
parentfb797a70108f3fda83fde6dea30bee4be7d5df8b
drm/syncobj: Fix handle <-> fd ioctls with dirty stack

Consider the following application:

    #include <fcntl.h>
    #include <string.h>
    #include <drm/drm.h>
    #include <sys/ioctl.h>

    int main(void) {
        int fd = open("/dev/dri/renderD128", O_RDWR);
        struct drm_syncobj_create arg1;
        ioctl(fd, DRM_IOCTL_SYNCOBJ_CREATE, &arg1);
        struct drm_syncobj_handle arg2;
        memset(&arg2, 1, sizeof(arg2)); // simulate dirty stack
        arg2.handle = arg1.handle;
        arg2.flags = 0;
        arg2.fd = 0;
        arg2.pad = 0;
        // arg2.point = 0; // userspace is required to set point to 0
        ioctl(fd, DRM_IOCTL_SYNCOBJ_HANDLE_TO_FD, &arg2);
    }

The last ioctl returns EINVAL because args->point is not 0. However,
userspace developed against older kernel versions is not aware of the
new point field and might therefore not initialize it.

The correct check would be

    if (args->flags & DRM_SYNCOBJ_FD_TO_HANDLE_FLAGS_TIMELINE)
        return -EINVAL;

However, there might already be userspace that relies on this not
returning an error as long as point == 0. Therefore use the more lenient
check.

Fixes: c2d3a7300695 ("drm/syncobj: Extend EXPORT_SYNC_FILE for timeline syncobjs")
Signed-off-by: Julian Orth <ju.orth@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Link: https://lore.kernel.org/r/20260301-point-v1-1-21fc5fd98614@gmail.com
drivers/gpu/drm/drm_syncobj.c