| Message ID | 20220412202942.386981-2-axelrasmussen@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [1/2] userfaultfd: add /dev/userfaultfd for fine grained access control | expand |
On Tue, 12 Apr 2022 13:29:42 -0700 Axel Rasmussen <axelrasmussen@google.com> wrote: > Prefer this new interface, but if using it fails for any reason just > fall back to using userfaultfd(2) as before. This seems a poor idea - the old interface will henceforth be untested. Why not tweak the code to test both interfaces?
Thanks for looking Andrew. And, fair criticism. In keeping with the status quo, I'm thinking of just adding a new command-line argument which toggles between the two modes. But, if I'm honest, it's starting to feel like the test has way too many arguments... I'm tempted to refactor the test to use the kselftest framework [1], get rid of all these command line arguments, and just always test everything. But, this seems like a big and perhaps controversial refactor, so I may take it up after this series... [1]: https://www.kernel.org/doc/html/latest/dev-tools/kselftest.html On Tue, Apr 12, 2022 at 1:42 PM Andrew Morton <akpm@linux-foundation.org> wrote: > > On Tue, 12 Apr 2022 13:29:42 -0700 Axel Rasmussen <axelrasmussen@google.com> wrote: > > > Prefer this new interface, but if using it fails for any reason just > > fall back to using userfaultfd(2) as before. > > This seems a poor idea - the old interface will henceforth be untested. > > Why not tweak the code to test both interfaces?
On Mon, 18 Apr 2022 15:16:02 -0700 Axel Rasmussen <axelrasmussen@google.com> wrote: > Thanks for looking Andrew. And, fair criticism. > > In keeping with the status quo, I'm thinking of just adding a new > command-line argument which toggles between the two modes. But I think you could tweak the test pretty simply to run itself twice. Once with the syscall then once with the /dev interface. I suppose that adding the commandline argument is equivalent, as long as the upper level script/makefile invokes the test program twice. > But, if I'm honest, it's starting to feel like the test has way too > many arguments... I'm tempted to refactor the test to use the > kselftest framework [1], get rid of all these command line arguments, > and just always test everything. But, this seems like a big and > perhaps controversial refactor, so I may take it up after this > series... Yes, that's a separable activity.
diff --git a/tools/testing/selftests/vm/userfaultfd.c b/tools/testing/selftests/vm/userfaultfd.c index 92a4516f8f0d..a50c430f036c 100644 --- a/tools/testing/selftests/vm/userfaultfd.c +++ b/tools/testing/selftests/vm/userfaultfd.c @@ -383,13 +383,32 @@ static void assert_expected_ioctls_present(uint64_t mode, uint64_t ioctls) } } +static void __userfaultfd_open_dev(void) +{ + int fd; + + uffd = -1; + fd = open("/dev/userfaultfd", O_RDWR | O_CLOEXEC); + if (fd < 0) + return; + + uffd = ioctl(fd, USERFAULTFD_IOC_NEW, + O_CLOEXEC | O_NONBLOCK | UFFD_USER_MODE_ONLY); + close(fd); +} + static void userfaultfd_open(uint64_t *features) { struct uffdio_api uffdio_api; - uffd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK | UFFD_USER_MODE_ONLY); + __userfaultfd_open_dev(); + if (uffd < 0) { + printf("/dev/userfaultfd failed, fallback to userfaultfd(2)\n"); + uffd = syscall(__NR_userfaultfd, + O_CLOEXEC | O_NONBLOCK | UFFD_USER_MODE_ONLY); + } if (uffd < 0) - err("userfaultfd syscall not available in this kernel"); + err("userfaultfd syscall failed"); uffd_flags = fcntl(uffd, F_GETFD, NULL); uffdio_api.api = UFFD_API;
Prefer this new interface, but if using it fails for any reason just fall back to using userfaultfd(2) as before. Signed-off-by: Axel Rasmussen <axelrasmussen@google.com> --- tools/testing/selftests/vm/userfaultfd.c | 23 +++++++++++++++++++++-- 1 file changed, 21 insertions(+), 2 deletions(-)