KF_TRUSTED_ARGS kfuncs currently have a subtle and insidious bug in
validating pointers to scalars. Say that you have a kfunc like the
following, which takes an array as the first argument:
bool bpf_cpumask_empty(const struct cpumask *cpumask)
{
return cpumask_empty(cpumask);
}
...
BTF_ID_FLAGS(func, bpf_cpumask_empty, KF_TRUSTED_ARGS)
...
If a BPF program were to invoke the kfunc with a NULL argument, it would
crash the kernel. The reason is that struct cpumask is defined as a
bitmap, which is itself defined as an array, and is accessed as a memory
address by bitmap operations. So when the verifier analyzes the
register, it interprets it as a pointer to a scalar struct, which is an
array of size 8. check_mem_reg() then sees that the register is NULL and
returns 0, and the kfunc crashes when it passes it down to the cpumask
wrappers.
To fix this, this patch adds a check for KF_ARG_PTR_TO_MEM which
verifies that the register doesn't contain a possibly-NULL pointer if
the kfunc is KF_TRUSTED_ARGS.
Signed-off-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20230125143816.721952-2-void@manifault.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When validating BTF types for KF_TRUSTED_ARGS kfuncs, the verifier
currently enforces that the top-level type must match when calling
the kfunc. In other words, the verifier does not allow the BPF program
to pass a bitwise equivalent struct, despite it being allowed according
to the C standard.
For example, if you have the following type:
struct nf_conn___init {
struct nf_conn ct;
};
The C standard stipulates that it would be safe to pass a struct
nf_conn___init to a kfunc expecting a struct nf_conn. The verifier
currently disallows this, however, as semantically kfuncs may want to
enforce that structs that have equivalent types according to the C
standard, but have different BTF IDs, are not able to be passed to
kfuncs expecting one or the other. For example, struct nf_conn___init
may not be queried / looked up, as it is allocated but may not yet be
fully initialized.
On the other hand, being able to pass types that are equivalent
according to the C standard will be useful for other types of kfunc /
kptrs enabled by BPF. For example, in a follow-on patch, a series of
kfuncs will be added which allow programs to do bitwise queries on
cpumasks that are either allocated by the program (in which case they'll
be a 'struct bpf_cpumask' type that wraps a cpumask_t as its first
element), or a cpumask that was allocated by the main kernel (in which
case it will just be a straight cpumask_t, as in task->cpus_ptr).
Having the two types of cpumasks allows us to distinguish between the
two for when a cpumask is read-only vs. mutatable. A struct bpf_cpumask
can be mutated by e.g. bpf_cpumask_clear(), whereas a regular cpumask_t
cannot be. On the other hand, a struct bpf_cpumask can of course be
queried in the exact same manner as a cpumask_t, with e.g.
bpf_cpumask_test_cpu().
If we were to enforce that top level types match, then a user that's
passing a struct bpf_cpumask to a read-only cpumask_t argument would
have to cast with something like bpf_cast_to_kern_ctx() (which itself
would need to be updated to expect the alias, and currently it only
accommodates a single alias per prog type). Additionally, not specifying
KF_TRUSTED_ARGS is not an option, as some kfuncs take one argument as a
struct bpf_cpumask *, and another as a struct cpumask *
(i.e. cpumask_t).
In order to enable this, this patch relaxes the constraint that a
KF_TRUSTED_ARGS kfunc must have strict type matching, and instead only
enforces strict type matching if a type is observed to be a "no-cast
alias" (i.e., that the type names are equivalent, but one is suffixed
with ___init).
Additionally, in order to try and be conservative and match existing
behavior / expectations, this patch also enforces strict type checking
for acquire kfuncs. We were already enforcing it for release kfuncs, so
this should also improve the consistency of the semantics for kfuncs.
Signed-off-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20230120192523.3650503-3-void@manifault.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In kfuncs, a "trusted" pointer is a pointer that the kfunc can assume is
safe, and which the verifier will allow to be passed to a
KF_TRUSTED_ARGS kfunc. Currently, a KF_TRUSTED_ARGS kfunc disallows any
pointer to be passed at a nonzero offset, but sometimes this is in fact
safe if the "nested" pointer's lifetime is inherited from its parent.
For example, the const cpumask_t *cpus_ptr field in a struct task_struct
will remain valid until the task itself is destroyed, and thus would
also be safe to pass to a KF_TRUSTED_ARGS kfunc.
While it would be conceptually simple to enable this by using BTF tags,
gcc unfortunately does not yet support this. In the interim, this patch
enables support for this by using a type-naming convention. A new
BTF_TYPE_SAFE_NESTED macro is defined in verifier.c which allows a
developer to specify the nested fields of a type which are considered
trusted if its parent is also trusted. The verifier is also updated to
account for this. A patch with selftests will be added in a follow-on
change, along with documentation for this feature.
Signed-off-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20230120192523.3650503-2-void@manifault.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Andrii Nakryiko says:
====================
This patch set fixes and extends libbpf's bpf_tracing.h support for tracing
arguments of kprobes/uprobes, and syscall as a special case.
Depending on the architecture, anywhere between 3 and 8 arguments can be
passed to a function in registers (so relevant to kprobes and uprobes), but
before this patch set libbpf's macros in bpf_tracing.h only supported up to
5 arguments, which is limiting in practice. This patch set extends
bpf_tracing.h to support up to 8 arguments, if architecture allows. This
includes explicit PT_REGS_PARMx() macro family, as well as BPF_KPROBE() macro.
Now, with tracing syscall arguments situation is sometimes quite different.
For a lot of architectures syscall argument passing through registers differs
from function call sequence at least a little. For i386 it differs *a lot*.
This patch set addresses this issue across all currently supported
architectures and hopefully fixes existing issues. syscall(2) manpage defines
that either 6 or 7 arguments can be supported, depending on architecture, so
libbpf defines 6 or 7 registers per architecture to be used to fetch syscall
arguments.
Also, BPF_UPROBE and BPF_URETPROBE are introduced as part of this patch set.
They are aliases for BPF_KPROBE and BPF_KRETPROBE (as mechanics of argument
fetching of kernel functions and user-space functions are identical), but it
allows BPF users to have less confusing BPF-side code when working with
uprobes.
For both sets of changes selftests are extended to test these new register
definitions to architecture-defined limits. Unfortunately I don't have ability
to test it on all architectures, and BPF CI only tests 3 architecture (x86-64,
arm64, and s390x), so it would be greatly appreciated if people with access to
architectures other than above 3 helped review and test changes.
====================
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Each architecture supports at least 6 syscall argument registers, so now
that specs for each architecture is defined in bpf_tracing.h, remove
unnecessary macro overrides, which previously were required to keep
existing BPF_KSYSCALL() uses compiling and working.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230120200914.3008030-26-andrii@kernel.org
Set up generic support in bpf_tracing.h for up to 7 syscall arguments
tracing with BPF_KSYSCALL, which seems to be the limit according to
syscall(2) manpage. Also change the way that syscall convention is
specified to be more explicit. Subsequent patches will adjust and define
proper per-architecture syscall conventions.
__PT_PARM1_SYSCALL_REG through __PT_PARM6_SYSCALL_REG is added
temporarily to keep everything working before each architecture has
syscall reg tables defined. They will be removed afterwards.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Alan Maguire <alan.maguire@oracle.com> # arm64
Link: https://lore.kernel.org/bpf/20230120200914.3008030-13-andrii@kernel.org
Update uprobe_autoattach selftest to validate architecture-specific
argument passing through registers. Use new BPF_UPROBE and
BPF_URETPROBE, and construct both BPF-side and user-space side in such
a way that for different architectures we are fetching and checking
different number of arguments, matching architecture-specific limit of
how many registers are available for argument passing.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Alan Maguire <alan.maguire@oracle.com> # arm64
Tested-by: Ilya Leoshkevich <iii@linux.ibm.com> # s390x
Link: https://lore.kernel.org/bpf/20230120200914.3008030-12-andrii@kernel.org
Add BPF_UPROBE and BPF_URETPROBE macros, aliased to BPF_KPROBE and
BPF_KRETPROBE, respectively. This makes uprobe-based BPF program code
much less confusing, especially to people new to tracing, at no cost in
terms of maintainability. We'll use this macro in selftests in
subsequent patch.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230120200914.3008030-11-andrii@kernel.org
Add BPF_KPROBE() and PT_REGS_PARMx() support for up to 8 arguments, if
target architecture supports this. Currently all architectures are
limited to only 5 register-placed arguments, which is limiting even on
x86-64.
This patch adds generic macro machinery to support up to 8 arguments
both when explicitly fetching it from pt_regs through PT_REGS_PARMx()
macros, as well as more ergonomic access in BPF_KPROBE().
Also, for i386 architecture we now don't have to define fake PARM4 and
PARM5 definitions, they will be generically substituted, just like for
PARM6 through PARM8.
Subsequent patches will fill out architecture-specific definitions,
where appropriate.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Alan Maguire <alan.maguire@oracle.com> # arm64
Tested-by: Ilya Leoshkevich <iii@linux.ibm.com> # s390x
Link: https://lore.kernel.org/bpf/20230120200914.3008030-2-andrii@kernel.org
In commit 537c3f66ea ("selftests/bpf: add generic BPF program tester-loader"),
a new mechanism was added to the BPF selftest framework to allow testsuites to
use macros to define expected failing testcases.
This allows any testsuite which tests verification failure to remove a good
amount of boilerplate code. This patch updates the task_kfunc selftest suite
to use these new macros.
Signed-off-by: David Vernet <void@manifault.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20230120021844.3048244-1-void@manifault.com
To be used for verification of driver implementations. Note that
the skb path is gone from the series, but I'm still keeping the
implementation for any possible future work.
$ xdp_hw_metadata <ifname>
On the other machine:
$ echo -n xdp | nc -u -q1 <target> 9091 # for AF_XDP
$ echo -n skb | nc -u -q1 <target> 9092 # for skb
Sample output:
# xdp
xsk_ring_cons__peek: 1
0x19f9090: rx_desc[0]->addr=100000000008000 addr=8100 comp_addr=8000
rx_timestamp_supported: 1
rx_timestamp: 1667850075063948829
0x19f9090: complete idx=8 addr=8000
# skb
found skb hwtstamp = 1668314052.854274681
Decoding:
# xdp
rx_timestamp=1667850075.063948829
$ date -d @1667850075
Mon Nov 7 11:41:15 AM PST 2022
$ date
Mon Nov 7 11:42:05 AM PST 2022
# skb
$ date -d @1668314052
Sat Nov 12 08:34:12 PM PST 2022
$ date
Sat Nov 12 08:37:06 PM PST 2022
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Martin KaFai Lau <martin.lau@linux.dev>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Anatoly Burakov <anatoly.burakov@intel.com>
Cc: Alexander Lobakin <alexandr.lobakin@intel.com>
Cc: Magnus Karlsson <magnus.karlsson@gmail.com>
Cc: Maryam Tahhan <mtahhan@redhat.com>
Cc: xdp-hints@xdp-project.net
Cc: netdev@vger.kernel.org
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20230119221536.3349901-18-sdf@google.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Instead of rejecting the attaching of PROG_TYPE_EXT programs to XDP
programs that consume HW metadata, implement support for propagating the
offload information. The extension program doesn't need to set a flag or
ifindex, these will just be propagated from the target by the verifier.
We need to create a separate offload object for the extension program,
though, since it can be reattached to a different program later (which
means we can't just inherit the offload information from the target).
An additional check is added on attach that the new target is compatible
with the offload information in the extension prog.
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20230119221536.3349901-9-sdf@google.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Kumar Kartikeya Dwivedi says:
====================
This is part 2 of https://lore.kernel.org/bpf/20221018135920.726360-1-memxor@gmail.com.
Changelog:
----------
v4 -> v5
v5: https://lore.kernel.org/bpf/20230120070355.1983560-1-memxor@gmail.com
* Add comments, tests from Joanne
* Add Joanne's acks
v3 -> v4
v3: https://lore.kernel.org/bpf/20230120034314.1921848-1-memxor@gmail.com
* Adopt BPF ASM tests to more readable style (Alexei)
v2 -> v3
v2: https://lore.kernel.org/bpf/20230119021442.1465269-1-memxor@gmail.com
* Fix slice invalidation logic for unreferenced dynptrs (Joanne)
* Add selftests for precise slice invalidation on destruction
* Add Joanne's acks
v1 -> v2
v1: https://lore.kernel.org/bpf/20230101083403.332783-1-memxor@gmail.com
* Return error early in case of overwriting referenced dynptr slots (Andrii, Joanne)
* Rename destroy_stack_slots_dynptr to destroy_if_dynptr_stack_slot (Joanne)
* Invalidate dynptr slices associated with dynptr in destroy_if_dynptr_stack_slot (Joanne)
* Combine both dynptr_get_spi and is_spi_bounds_valid (Joanne)
* Compute spi once in process_dynptr_func and pass it as parameter instead of recomputing (Joanne)
* Add comments expanding REG_LIVE_WRITTEN marking in unmark_stack_slots_dynptr (Joanne)
* Add comments explaining why destroy_if_dynptr_stack_slot call needs to be done for both spi
and spi - 1 (Joanne)
* Port BPF assembly tests from test_verifier to test_progs framework (Andrii)
* Address misc feedback, rebase to bpf-next
Old v1 -> v1
Old v1: https://lore.kernel.org/bpf/20221018135920.726360-1-memxor@gmail.com
* Allow overwriting dynptr stack slots from dynptr init helpers
* Fix a bug in alignment check where reg->var_off.value was still not included
* Address other minor nits
Eduard Zingerman (1):
selftests/bpf: convenience macro for use with 'asm volatile' blocks
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
