| Message ID | 20220309165222.2843651-8-tjmercier@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Proposal for a GPU cgroup controller | expand |
On Wed, Mar 9, 2022 at 8:52 AM T.J. Mercier <tjmercier@google.com> wrote: > > The kernel interface should use types that the kernel defines instead of > pid_t and uid_t, whose definiton is owned by libc. This fixes the header > so that it can be included without first including sys/types.h. > > Signed-off-by: T.J. Mercier <tjmercier@google.com> > --- > include/uapi/linux/android/binder.h | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/include/uapi/linux/android/binder.h b/include/uapi/linux/android/binder.h > index 169fd5069a1a..aa28454dbca3 100644 > --- a/include/uapi/linux/android/binder.h > +++ b/include/uapi/linux/android/binder.h > @@ -289,8 +289,8 @@ struct binder_transaction_data { > > /* General information about the transaction. */ > __u32 flags; > - pid_t sender_pid; > - uid_t sender_euid; > + __kernel_pid_t sender_pid; > + __kernel_uid_t sender_euid; Are we guaranteed that this does not affect the UAPI at all? Userspace code using this definition will have to run with kernels using the old definition and visa-versa. > binder_size_t data_size; /* number of bytes of data */ > binder_size_t offsets_size; /* number of bytes of offsets */ > > -- > 2.35.1.616.g0bdcbb4464-goog >
On Thu, Mar 10, 2022 at 11:33 AM Todd Kjos <tkjos@google.com> wrote: > > On Wed, Mar 9, 2022 at 8:52 AM T.J. Mercier <tjmercier@google.com> wrote: > > > > The kernel interface should use types that the kernel defines instead of > > pid_t and uid_t, whose definiton is owned by libc. This fixes the header > > so that it can be included without first including sys/types.h. > > > > Signed-off-by: T.J. Mercier <tjmercier@google.com> > > --- > > include/uapi/linux/android/binder.h | 4 ++-- > > 1 file changed, 2 insertions(+), 2 deletions(-) > > > > diff --git a/include/uapi/linux/android/binder.h b/include/uapi/linux/android/binder.h > > index 169fd5069a1a..aa28454dbca3 100644 > > --- a/include/uapi/linux/android/binder.h > > +++ b/include/uapi/linux/android/binder.h > > @@ -289,8 +289,8 @@ struct binder_transaction_data { > > > > /* General information about the transaction. */ > > __u32 flags; > > - pid_t sender_pid; > > - uid_t sender_euid; > > + __kernel_pid_t sender_pid; > > + __kernel_uid_t sender_euid; > > Are we guaranteed that this does not affect the UAPI at all? Userspace > code using this definition will have to run with kernels using the old > definition and visa-versa. A standards compliant userspace should be expecting a signed integer type here. So the only way I can think userspace would be affected is if: 1) pid_t is a long AND 2) sizeof(long) > sizeof(int) AND 3) Consumers of the pid_t definition actually attempt to mutate the result to make use of extra bits in the variable (which are not there) This seems extremely unlikely. For instance just on the topic of the first item, all of the C library implementations with pid_t definitions linked here use an int, except for Bionic which typdefs pid_t to __kernel_pid_t and Sortix which uses long. https://wiki.osdev.org/C_Library However I would argue this is already broken and should count as a bug fix since I can't do this: $ cat binder_include.c ; gcc binder_include.c #include <linux/android/binder.h> int main() {} In file included from binder_include.c:1: /usr/include/linux/android/binder.h:291:9: error: unknown type name ‘pid_t’ 291 | pid_t sender_pid; | ^~~~~ /usr/include/linux/android/binder.h:292:9: error: unknown type name ‘uid_t’ 292 | uid_t sender_euid; | ^~~~~ This is also the only occurrence of pid_t in all of include/uapi/linux. All 40+ other uses are __kernel_pid_t, and I don't see why the binder header should be different. > > > binder_size_t data_size; /* number of bytes of data */ > > binder_size_t offsets_size; /* number of bytes of offsets */ > > > > -- > > 2.35.1.616.g0bdcbb4464-goog > >
On Mon, Mar 14, 2022 at 4:45 PM T.J. Mercier <tjmercier@google.com> wrote: > > On Thu, Mar 10, 2022 at 11:33 AM Todd Kjos <tkjos@google.com> wrote: > > > > On Wed, Mar 9, 2022 at 8:52 AM T.J. Mercier <tjmercier@google.com> wrote: > > > > > > The kernel interface should use types that the kernel defines instead of > > > pid_t and uid_t, whose definiton is owned by libc. This fixes the header > > > so that it can be included without first including sys/types.h. > > > > > > Signed-off-by: T.J. Mercier <tjmercier@google.com> > > > --- > > > include/uapi/linux/android/binder.h | 4 ++-- > > > 1 file changed, 2 insertions(+), 2 deletions(-) > > > > > > diff --git a/include/uapi/linux/android/binder.h b/include/uapi/linux/android/binder.h > > > index 169fd5069a1a..aa28454dbca3 100644 > > > --- a/include/uapi/linux/android/binder.h > > > +++ b/include/uapi/linux/android/binder.h > > > @@ -289,8 +289,8 @@ struct binder_transaction_data { > > > > > > /* General information about the transaction. */ > > > __u32 flags; > > > - pid_t sender_pid; > > > - uid_t sender_euid; > > > + __kernel_pid_t sender_pid; > > > + __kernel_uid_t sender_euid; > > > > Are we guaranteed that this does not affect the UAPI at all? Userspace > > code using this definition will have to run with kernels using the old > > definition and visa-versa. > > A standards compliant userspace should be expecting a signed integer > type here. So the only way I can think userspace would be affected is > if: > 1) pid_t is a long AND > 2) sizeof(long) > sizeof(int) AND > 3) Consumers of the pid_t definition actually attempt to mutate the > result to make use of extra bits in the variable (which are not there) > > This seems extremely unlikely. For instance just on the topic of the > first item, all of the C library implementations with pid_t > definitions linked here use an int, except for Bionic which typdefs > pid_t to __kernel_pid_t and Sortix which uses long. > https://wiki.osdev.org/C_Library > > However I would argue this is already broken and should count as a bug > fix since I can't do this: > > $ cat binder_include.c ; gcc binder_include.c > #include <linux/android/binder.h> > int main() {} > In file included from binder_include.c:1: > /usr/include/linux/android/binder.h:291:9: error: unknown type name ‘pid_t’ > 291 | pid_t sender_pid; > | ^~~~~ > /usr/include/linux/android/binder.h:292:9: error: unknown type name ‘uid_t’ > 292 | uid_t sender_euid; > | ^~~~~ > > This is also the only occurrence of pid_t in all of > include/uapi/linux. All 40+ other uses are __kernel_pid_t, and I don't > see why the binder header should be different. It looks like those other cases used to be pid_t, but were changed to __kernel_pid_t. Acked-by: Todd Kjos <tkjos@google.com> > > > > > > > binder_size_t data_size; /* number of bytes of data */ > > > binder_size_t offsets_size; /* number of bytes of offsets */ > > > > > > -- > > > 2.35.1.616.g0bdcbb4464-goog > > >
From: T.J. Mercier > Sent: 14 March 2022 23:45 > > On Thu, Mar 10, 2022 at 11:33 AM Todd Kjos <tkjos@google.com> wrote: > > > > On Wed, Mar 9, 2022 at 8:52 AM T.J. Mercier <tjmercier@google.com> wrote: > > > > > > The kernel interface should use types that the kernel defines instead of > > > pid_t and uid_t, whose definiton is owned by libc. This fixes the header > > > so that it can be included without first including sys/types.h. > > > > > > Signed-off-by: T.J. Mercier <tjmercier@google.com> > > > --- > > > include/uapi/linux/android/binder.h | 4 ++-- > > > 1 file changed, 2 insertions(+), 2 deletions(-) > > > > > > diff --git a/include/uapi/linux/android/binder.h b/include/uapi/linux/android/binder.h > > > index 169fd5069a1a..aa28454dbca3 100644 > > > --- a/include/uapi/linux/android/binder.h > > > +++ b/include/uapi/linux/android/binder.h > > > @@ -289,8 +289,8 @@ struct binder_transaction_data { > > > > > > /* General information about the transaction. */ > > > __u32 flags; > > > - pid_t sender_pid; > > > - uid_t sender_euid; > > > + __kernel_pid_t sender_pid; > > > + __kernel_uid_t sender_euid; > > > > Are we guaranteed that this does not affect the UAPI at all? Userspace > > code using this definition will have to run with kernels using the old > > definition and visa-versa. > > A standards compliant userspace should be expecting a signed integer > type here. So the only way I can think userspace would be affected is > if: > 1) pid_t is a long AND > 2) sizeof(long) > sizeof(int) AND > 3) Consumers of the pid_t definition actually attempt to mutate the > result to make use of extra bits in the variable (which are not there) Or the userspace headers have a 16bit pid_t. I can't help feeling that uapi headers should only use explicit fixed sized types. There is no point indirecting the type names - the sizes still can't be changes. David - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
On Tue, Mar 15, 2022 at 12:56 AM David Laight <David.Laight@aculab.com> wrote: > > From: T.J. Mercier > > Sent: 14 March 2022 23:45 > > > > On Thu, Mar 10, 2022 at 11:33 AM Todd Kjos <tkjos@google.com> wrote: > > > > > > On Wed, Mar 9, 2022 at 8:52 AM T.J. Mercier <tjmercier@google.com> wrote: > > > > > > > > The kernel interface should use types that the kernel defines instead of > > > > pid_t and uid_t, whose definiton is owned by libc. This fixes the header > > > > so that it can be included without first including sys/types.h. > > > > > > > > Signed-off-by: T.J. Mercier <tjmercier@google.com> > > > > --- > > > > include/uapi/linux/android/binder.h | 4 ++-- > > > > 1 file changed, 2 insertions(+), 2 deletions(-) > > > > > > > > diff --git a/include/uapi/linux/android/binder.h b/include/uapi/linux/android/binder.h > > > > index 169fd5069a1a..aa28454dbca3 100644 > > > > --- a/include/uapi/linux/android/binder.h > > > > +++ b/include/uapi/linux/android/binder.h > > > > @@ -289,8 +289,8 @@ struct binder_transaction_data { > > > > > > > > /* General information about the transaction. */ > > > > __u32 flags; > > > > - pid_t sender_pid; > > > > - uid_t sender_euid; > > > > + __kernel_pid_t sender_pid; > > > > + __kernel_uid_t sender_euid; > > > > > > Are we guaranteed that this does not affect the UAPI at all? Userspace > > > code using this definition will have to run with kernels using the old > > > definition and visa-versa. > > > > A standards compliant userspace should be expecting a signed integer > > type here. So the only way I can think userspace would be affected is > > if: > > 1) pid_t is a long AND > > 2) sizeof(long) > sizeof(int) AND > > 3) Consumers of the pid_t definition actually attempt to mutate the > > result to make use of extra bits in the variable (which are not there) > > Or the userspace headers have a 16bit pid_t. Since the kernel uses an int for PIDs, wouldn't a 16 bit pid_t already be potentially broken (overflow) on systems where int is not 16 bits? On systems where int is 16 bits, there is no change here except to achieve uniform use of __kernel_pid_t in the kernel headers and fix the include problem. > > I can't help feeling that uapi headers should only use explicit > fixed sized types. > There is no point indirecting the type names - the sizes still > can't be changes. I think it's still unlikely to be an actual problem. For example there are other occasions where a switch like this was made: https://github.com/torvalds/linux/commit/694a58e29ef27c4c26f103a9decfd053f94dd34c https://github.com/torvalds/linux/commit/269b8fd5d058f2c0da01a42b20315ffc2640d99b And also since Binder's only known user is Android through Bionic which already expects the type of pid_t to be __kernel_pid_t. > > David > > - > Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK > Registration No: 1397386 (Wales)
diff --git a/include/uapi/linux/android/binder.h b/include/uapi/linux/android/binder.h index 169fd5069a1a..aa28454dbca3 100644 --- a/include/uapi/linux/android/binder.h +++ b/include/uapi/linux/android/binder.h @@ -289,8 +289,8 @@ struct binder_transaction_data { /* General information about the transaction. */ __u32 flags; - pid_t sender_pid; - uid_t sender_euid; + __kernel_pid_t sender_pid; + __kernel_uid_t sender_euid; binder_size_t data_size; /* number of bytes of data */ binder_size_t offsets_size; /* number of bytes of offsets */
The kernel interface should use types that the kernel defines instead of pid_t and uid_t, whose definiton is owned by libc. This fixes the header so that it can be included without first including sys/types.h. Signed-off-by: T.J. Mercier <tjmercier@google.com> --- include/uapi/linux/android/binder.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)