]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/commit
open: don't silently ignore unknown O-flags in openat2()
authorChristian Brauner <christian.brauner@ubuntu.com>
Fri, 28 May 2021 09:24:16 +0000 (11:24 +0200)
committerChristian Brauner <christian.brauner@ubuntu.com>
Fri, 28 May 2021 15:44:37 +0000 (17:44 +0200)
commitcfe80306a0dd6d363934913e47c3f30d71b721e5
treeabeb1cc793261aaeefa97d42be0aeb054f14be6f
parent5c350aa11b441b32baf3bfe4018168cb8d10cef7
open: don't silently ignore unknown O-flags in openat2()

The new openat2() syscall verifies that no unknown O-flag values are
set and returns an error to userspace if they are while the older open
syscalls like open() and openat() simply ignore unknown flag values:

  #define O_FLAG_CURRENTLY_INVALID (1 << 31)
  struct open_how how = {
          .flags = O_RDONLY | O_FLAG_CURRENTLY_INVALID,
          .resolve = 0,
  };

  /* fails */
  fd = openat2(-EBADF, "/dev/null", &how, sizeof(how));

  /* succeeds */
  fd = openat(-EBADF, "/dev/null", O_RDONLY | O_FLAG_CURRENTLY_INVALID);

However, openat2() silently truncates the upper 32 bits meaning:

  #define O_FLAG_CURRENTLY_INVALID_LOWER32 (1 << 31)
  #define O_FLAG_CURRENTLY_INVALID_UPPER32 (1 << 40)

  struct open_how how_lowe32 = {
          .flags = O_RDONLY | O_FLAG_CURRENTLY_INVALID_LOWER32,
  };

  struct open_how how_upper32 = {
          .flags = O_RDONLY | O_FLAG_CURRENTLY_INVALID_UPPER32,
  };

  /* fails */
  fd = openat2(-EBADF, "/dev/null", &how_lower32, sizeof(how_lower32));

  /* succeeds */
  fd = openat2(-EBADF, "/dev/null", &how_upper32, sizeof(how_upper32));

Fix this by preventing the immediate truncation in build_open_flags().

There's a snafu here though stripping FMODE_* directly from flags would
cause the upper 32 bits to be truncated as well due to integer promotion
rules since FMODE_* is unsigned int, O_* are signed ints (yuck).

In addition, struct open_flags currently defines flags to be 32 bit
which is reasonable. If we simply were to bump it to 64 bit we would
need to change a lot of code preemptively which doesn't seem worth it.
So simply add a compile-time check verifying that all currently known
O_* flags are within the 32 bit range and fail to build if they aren't
anymore.

This change shouldn't regress old open syscalls since they silently
truncate any unknown values anyway. It is a tiny semantic change for
openat2() but it is very unlikely people pass ing > 32 bit unknown flags
and the syscall is relatively new too.

Link: https://lore.kernel.org/r/20210528092417.3942079-3-brauner@kernel.org
Cc: Christoph Hellwig <hch@lst.de>
Cc: Aleksa Sarai <cyphar@cyphar.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Reported-by: Richard Guy Briggs <rgb@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Aleksa Sarai <cyphar@cyphar.com>
Reviewed-by: Richard Guy Briggs <rgb@redhat.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
fs/open.c