| Message ID | 20191009160907.10981-1-christian.brauner@ubuntu.com |
|---|---|
| Headers | show |
| Series | bpf: switch to new usercopy helpers | expand |
On Wed, Oct 9, 2019 at 9:09 AM Christian Brauner <christian.brauner@ubuntu.com> wrote: > > Hey everyone, > > In v5.4-rc2 we added two new helpers check_zeroed_user() and > copy_struct_from_user() including selftests (cf. [1]). It is a generic > interface designed to copy a struct from userspace. The helpers will be > especially useful for structs versioned by size of which we have quite a > few. > > The most obvious benefit is that this helper lets us get rid of > duplicate code. We've already switched over sched_setattr(), perf_event_open(), > and clone3(). More importantly it will also help to ensure that users > implementing versioning-by-size end up with the same core semantics. > > This point is especially crucial since we have at least one case where > versioning-by-size is used but with slighly different semantics: > sched_setattr(), perf_event_open(), and clone3() all do do similar > checks to copy_struct_from_user() while rt_sigprocmask(2) always rejects > differently-sized struct arguments. > > This little series switches over bpf codepaths that have hand-rolled > implementations of these helpers. check_zeroed_user() is not in bpf-next. we will let this set sit in patchworks for some time until bpf-next is merged back into net-next and we fast forward it. Then we can apply it (assuming no conflicts).
On Wed, Oct 09, 2019 at 04:06:18PM -0700, Alexei Starovoitov wrote: > On Wed, Oct 9, 2019 at 9:09 AM Christian Brauner > <christian.brauner@ubuntu.com> wrote: > > > > Hey everyone, > > > > In v5.4-rc2 we added two new helpers check_zeroed_user() and > > copy_struct_from_user() including selftests (cf. [1]). It is a generic > > interface designed to copy a struct from userspace. The helpers will be > > especially useful for structs versioned by size of which we have quite a > > few. > > > > The most obvious benefit is that this helper lets us get rid of > > duplicate code. We've already switched over sched_setattr(), perf_event_open(), > > and clone3(). More importantly it will also help to ensure that users > > implementing versioning-by-size end up with the same core semantics. > > > > This point is especially crucial since we have at least one case where > > versioning-by-size is used but with slighly different semantics: > > sched_setattr(), perf_event_open(), and clone3() all do do similar > > checks to copy_struct_from_user() while rt_sigprocmask(2) always rejects > > differently-sized struct arguments. > > > > This little series switches over bpf codepaths that have hand-rolled > > implementations of these helpers. > > check_zeroed_user() is not in bpf-next. > we will let this set sit in patchworks for some time until bpf-next > is merged back into net-next and we fast forward it. > Then we can apply it (assuming no conflicts). Sounds good to me. Just ping me when you need me to resend rebase onto bpf-next. Christian
On Thu, Oct 10, 2019 at 2:26 AM Christian Brauner <christian.brauner@ubuntu.com> wrote: > > On Wed, Oct 09, 2019 at 04:06:18PM -0700, Alexei Starovoitov wrote: > > On Wed, Oct 9, 2019 at 9:09 AM Christian Brauner > > <christian.brauner@ubuntu.com> wrote: > > > > > > Hey everyone, > > > > > > In v5.4-rc2 we added two new helpers check_zeroed_user() and > > > copy_struct_from_user() including selftests (cf. [1]). It is a generic > > > interface designed to copy a struct from userspace. The helpers will be > > > especially useful for structs versioned by size of which we have quite a > > > few. > > > > > > The most obvious benefit is that this helper lets us get rid of > > > duplicate code. We've already switched over sched_setattr(), perf_event_open(), > > > and clone3(). More importantly it will also help to ensure that users > > > implementing versioning-by-size end up with the same core semantics. > > > > > > This point is especially crucial since we have at least one case where > > > versioning-by-size is used but with slighly different semantics: > > > sched_setattr(), perf_event_open(), and clone3() all do do similar > > > checks to copy_struct_from_user() while rt_sigprocmask(2) always rejects > > > differently-sized struct arguments. > > > > > > This little series switches over bpf codepaths that have hand-rolled > > > implementations of these helpers. > > > > check_zeroed_user() is not in bpf-next. > > we will let this set sit in patchworks for some time until bpf-next > > is merged back into net-next and we fast forward it. > > Then we can apply it (assuming no conflicts). > > Sounds good to me. Just ping me when you need me to resend rebase onto > bpf-next. -rc1 is now in bpf-next. I took a look at patches and they look good overall. In patches 2 and 3 the zero init via "= {};" should be unnecessary anymore due to copy_struct_from_user() logic, right? Could you also convert all other case in kernel/bpf/, so bpf_check_uarg_tail_zero() can be removed ? Otherwise the half-way conversion will look odd.
On Tue, Oct 15, 2019 at 03:45:54PM -0700, Alexei Starovoitov wrote: > On Thu, Oct 10, 2019 at 2:26 AM Christian Brauner > <christian.brauner@ubuntu.com> wrote: > > > > On Wed, Oct 09, 2019 at 04:06:18PM -0700, Alexei Starovoitov wrote: > > > On Wed, Oct 9, 2019 at 9:09 AM Christian Brauner > > > <christian.brauner@ubuntu.com> wrote: > > > > > > > > Hey everyone, > > > > > > > > In v5.4-rc2 we added two new helpers check_zeroed_user() and > > > > copy_struct_from_user() including selftests (cf. [1]). It is a generic > > > > interface designed to copy a struct from userspace. The helpers will be > > > > especially useful for structs versioned by size of which we have quite a > > > > few. > > > > > > > > The most obvious benefit is that this helper lets us get rid of > > > > duplicate code. We've already switched over sched_setattr(), perf_event_open(), > > > > and clone3(). More importantly it will also help to ensure that users > > > > implementing versioning-by-size end up with the same core semantics. > > > > > > > > This point is especially crucial since we have at least one case where > > > > versioning-by-size is used but with slighly different semantics: > > > > sched_setattr(), perf_event_open(), and clone3() all do do similar > > > > checks to copy_struct_from_user() while rt_sigprocmask(2) always rejects > > > > differently-sized struct arguments. > > > > > > > > This little series switches over bpf codepaths that have hand-rolled > > > > implementations of these helpers. > > > > > > check_zeroed_user() is not in bpf-next. > > > we will let this set sit in patchworks for some time until bpf-next > > > is merged back into net-next and we fast forward it. > > > Then we can apply it (assuming no conflicts). > > > > Sounds good to me. Just ping me when you need me to resend rebase onto > > bpf-next. > > -rc1 is now in bpf-next. > I took a look at patches and they look good overall. > > In patches 2 and 3 the zero init via "= {};" > should be unnecessary anymore due to > copy_struct_from_user() logic, right? Right, I can remove them. > > Could you also convert all other case in kernel/bpf/, > so bpf_check_uarg_tail_zero() can be removed ? > Otherwise the half-way conversion will look odd. Hm, I thought I did that and concluded that bpf_check_uarg_tail_zero() can't be removed because sometimes it is called to verify whether a given struct is zeroed but nothing is actually copied from userspace but rather to userspace. See for example v5.4-rc3:kernel/bpf/syscall.c:bpf_map_get_info_by_fd() All call sites where something is actually copied from userspace I've switched to copy_struct_from_user(). Christian
On Tue, Oct 15, 2019 at 3:55 PM Christian Brauner <christian.brauner@ubuntu.com> wrote: > > On Tue, Oct 15, 2019 at 03:45:54PM -0700, Alexei Starovoitov wrote: > > On Thu, Oct 10, 2019 at 2:26 AM Christian Brauner > > <christian.brauner@ubuntu.com> wrote: > > > > > > On Wed, Oct 09, 2019 at 04:06:18PM -0700, Alexei Starovoitov wrote: > > > > On Wed, Oct 9, 2019 at 9:09 AM Christian Brauner > > > > <christian.brauner@ubuntu.com> wrote: > > > > > > > > > > Hey everyone, > > > > > > > > > > In v5.4-rc2 we added two new helpers check_zeroed_user() and > > > > > copy_struct_from_user() including selftests (cf. [1]). It is a generic > > > > > interface designed to copy a struct from userspace. The helpers will be > > > > > especially useful for structs versioned by size of which we have quite a > > > > > few. > > > > > > > > > > The most obvious benefit is that this helper lets us get rid of > > > > > duplicate code. We've already switched over sched_setattr(), perf_event_open(), > > > > > and clone3(). More importantly it will also help to ensure that users > > > > > implementing versioning-by-size end up with the same core semantics. > > > > > > > > > > This point is especially crucial since we have at least one case where > > > > > versioning-by-size is used but with slighly different semantics: > > > > > sched_setattr(), perf_event_open(), and clone3() all do do similar > > > > > checks to copy_struct_from_user() while rt_sigprocmask(2) always rejects > > > > > differently-sized struct arguments. > > > > > > > > > > This little series switches over bpf codepaths that have hand-rolled > > > > > implementations of these helpers. > > > > > > > > check_zeroed_user() is not in bpf-next. > > > > we will let this set sit in patchworks for some time until bpf-next > > > > is merged back into net-next and we fast forward it. > > > > Then we can apply it (assuming no conflicts). > > > > > > Sounds good to me. Just ping me when you need me to resend rebase onto > > > bpf-next. > > > > -rc1 is now in bpf-next. > > I took a look at patches and they look good overall. > > > > In patches 2 and 3 the zero init via "= {};" > > should be unnecessary anymore due to > > copy_struct_from_user() logic, right? > > Right, I can remove them. > > > > > Could you also convert all other case in kernel/bpf/, > > so bpf_check_uarg_tail_zero() can be removed ? > > Otherwise the half-way conversion will look odd. > > Hm, I thought I did that and concluded that bpf_check_uarg_tail_zero() > can't be removed because sometimes it is called to verify whether a > given struct is zeroed but nothing is actually copied from userspace but > rather to userspace. See for example > v5.4-rc3:kernel/bpf/syscall.c:bpf_map_get_info_by_fd() > All call sites where something is actually copied from userspace I've > switched to copy_struct_from_user(). I see. You're right. Could you update the comment in bpf_check_uarg_tail_zero() to clarify that copy_struct_from_user() should be used whenever possible instead ?
On Tue Oct 15, 2019 at 4:02 PM Alexei Starovoitov wrote: > On Tue, Oct 15, 2019 at 3:55 PM Christian Brauner > <christian.brauner@ubuntu.com> wrote: > > > > On Tue, Oct 15, 2019 at 03:45:54PM -0700, Alexei Starovoitov wrote: > > > On Thu, Oct 10, 2019 at 2:26 AM Christian Brauner > > > <christian.brauner@ubuntu.com> wrote: > > > > > > > > On Wed, Oct 09, 2019 at 04:06:18PM -0700, Alexei Starovoitov wrote: > > > > > On Wed, Oct 9, 2019 at 9:09 AM Christian Brauner > > > > > <christian.brauner@ubuntu.com> wrote: > > > > > > > > > > > > Hey everyone, > > > > > > > > > > > > In v5.4-rc2 we added two new helpers check_zeroed_user() and > > > > > > copy_struct_from_user() including selftests (cf. [1]). It is a generic > > > > > > interface designed to copy a struct from userspace. The helpers will be > > > > > > especially useful for structs versioned by size of which we have quite a > > > > > > few. > > > > > > > > > > > > The most obvious benefit is that this helper lets us get rid of > > > > > > duplicate code. We've already switched over sched_setattr(), perf_event_open(), > > > > > > and clone3(). More importantly it will also help to ensure that users > > > > > > implementing versioning-by-size end up with the same core semantics. > > > > > > > > > > > > This point is especially crucial since we have at least one case where > > > > > > versioning-by-size is used but with slighly different semantics: > > > > > > sched_setattr(), perf_event_open(), and clone3() all do do similar > > > > > > checks to copy_struct_from_user() while rt_sigprocmask(2) always rejects > > > > > > differently-sized struct arguments. > > > > > > > > > > > > This little series switches over bpf codepaths that have hand-rolled > > > > > > implementations of these helpers. > > > > > > > > > > check_zeroed_user() is not in bpf-next. > > > > > we will let this set sit in patchworks for some time until bpf-next > > > > > is merged back into net-next and we fast forward it. > > > > > Then we can apply it (assuming no conflicts). > > > > > > > > Sounds good to me. Just ping me when you need me to resend rebase onto > > > > bpf-next. > > > > > > -rc1 is now in bpf-next. > > > I took a look at patches and they look good overall. > > > > > > In patches 2 and 3 the zero init via "= {};" > > > should be unnecessary anymore due to > > > copy_struct_from_user() logic, right? > > > > Right, I can remove them. > > > > > > > > Could you also convert all other case in kernel/bpf/, > > > so bpf_check_uarg_tail_zero() can be removed ? > > > Otherwise the half-way conversion will look odd. > > > > Hm, I thought I did that and concluded that bpf_check_uarg_tail_zero() > > can't be removed because sometimes it is called to verify whether a > > given struct is zeroed but nothing is actually copied from userspace but > > rather to userspace. See for example > > v5.4-rc3:kernel/bpf/syscall.c:bpf_map_get_info_by_fd() > > All call sites where something is actually copied from userspace I've > > switched to copy_struct_from_user(). > > I see. You're right. > Could you update the comment in bpf_check_uarg_tail_zero() > to clarify that copy_struct_from_user() should be used whenever > possible instead ? Yup, can do. Christian
Hey everyone,
In v5.4-rc2 we added two new helpers check_zeroed_user() and
copy_struct_from_user() including selftests (cf. [1]). It is a generic
interface designed to copy a struct from userspace. The helpers will be
especially useful for structs versioned by size of which we have quite a
few.
The most obvious benefit is that this helper lets us get rid of
duplicate code. We've already switched over sched_setattr(), perf_event_open(),
and clone3(). More importantly it will also help to ensure that users
implementing versioning-by-size end up with the same core semantics.
This point is especially crucial since we have at least one case where
versioning-by-size is used but with slighly different semantics:
sched_setattr(), perf_event_open(), and clone3() all do do similar
checks to copy_struct_from_user() while rt_sigprocmask(2) always rejects
differently-sized struct arguments.
This little series switches over bpf codepaths that have hand-rolled
implementations of these helpers.
Thanks!
Christian
/* v1 */
Link: https://lore.kernel.org/r/20191009160907.10981-1-christian.brauner@ubuntu.com
/* v2 */
- rebase onto bpf-next
/* Reference */
[1]: f5a1a536fa14 ("lib: introduce copy_struct_from_user() helper")
Christian Brauner (3):
bpf: use check_zeroed_user() in bpf_check_uarg_tail_zero()
bpf: use copy_struct_from_user() in bpf_prog_get_info_by_fd()
bpf: use copy_struct_from_user() in bpf() syscall
kernel/bpf/syscall.c | 46 +++++++++++++++++---------------------------
1 file changed, 18 insertions(+), 28 deletions(-)
Hey everyone, In v5.4-rc2 we added two new helpers check_zeroed_user() and copy_struct_from_user() including selftests (cf. [1]). It is a generic interface designed to copy a struct from userspace. The helpers will be especially useful for structs versioned by size of which we have quite a few. The most obvious benefit is that this helper lets us get rid of duplicate code. We've already switched over sched_setattr(), perf_event_open(), and clone3(). More importantly it will also help to ensure that users implementing versioning-by-size end up with the same core semantics. This point is especially crucial since we have at least one case where versioning-by-size is used but with slighly different semantics: sched_setattr(), perf_event_open(), and clone3() all do do similar checks to copy_struct_from_user() while rt_sigprocmask(2) always rejects differently-sized struct arguments. This little series switches over bpf codepaths that have hand-rolled implementations of these helpers. Thanks! Christian /* Reference */ [1]: f5a1a536fa14 ("lib: introduce copy_struct_from_user() helper") Christian Brauner (3): bpf: use check_zeroed_user() in bpf_check_uarg_tail_zero() bpf: use copy_struct_from_user() in bpf_prog_get_info_by_fd() bpf: use copy_struct_from_user() in bpf() syscall kernel/bpf/syscall.c | 38 ++++++++++++-------------------------- 1 file changed, 12 insertions(+), 26 deletions(-)