netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH bpf-next v2] bpf, x86, arm64: enable jit by default when not built as always-on
@ 2019-12-09 15:08 Daniel Borkmann
  2019-12-09 17:06 ` Martin Lau
  0 siblings, 1 reply; 3+ messages in thread
From: Daniel Borkmann @ 2019-12-09 15:08 UTC (permalink / raw)
  To: ast; +Cc: netdev, bpf, will, Daniel Borkmann

After Spectre 2 fix via 290af86629b2 ("bpf: introduce BPF_JIT_ALWAYS_ON
config") most major distros use BPF_JIT_ALWAYS_ON configuration these days
which compiles out the BPF interpreter entirely and always enables the
JIT. Also given recent fix in e1608f3fa857 ("bpf: Avoid setting bpf insns
pages read-only when prog is jited"), we additionally avoid fragmenting
the direct map for the BPF insns pages sitting in the general data heap
since they are not used during execution. Latter is only needed when run
through the interpreter.

Since both x86 and arm64 JITs have seen a lot of exposure over the years,
are generally most up to date and maintained, there is more downside in
!BPF_JIT_ALWAYS_ON configurations to have the interpreter enabled by default
rather than the JIT. Add a ARCH_WANT_DEFAULT_BPF_JIT config which archs can
use to set the bpf_jit_{enable,kallsyms} to 1. Back in the days the
bpf_jit_kallsyms knob was set to 0 by default since major distros still
had /proc/kallsyms addresses exposed to unprivileged user space which is
not the case anymore. Hence both knobs are set via BPF_JIT_DEFAULT_ON which
is set to 'y' in case of BPF_JIT_ALWAYS_ON or ARCH_WANT_DEFAULT_BPF_JIT.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Will Deacon <will@kernel.org>
---
 [ Follow-up from https://lore.kernel.org/bpf/20191202200947.GA14353@pc-9.home/,
   applies to both bpf and bpf-next, but I think going via bpf-next is more
   appropriate. 

   v1 -> v2:
    - add depends on HAVE_EBPF_JIT && BPF_JIT to BPF_JIT_DEFAULT_ON (Will)
 ]

 arch/arm64/Kconfig | 1 +
 arch/x86/Kconfig   | 1 +
 init/Kconfig       | 7 +++++++
 kernel/bpf/core.c  | 4 ++--
 4 files changed, 11 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index b1b4476ddb83..29d03459de20 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -69,6 +69,7 @@ config ARM64
 	select ARCH_SUPPORTS_INT128 if CC_HAS_INT128 && (GCC_VERSION >= 50000 || CC_IS_CLANG)
 	select ARCH_SUPPORTS_NUMA_BALANCING
 	select ARCH_WANT_COMPAT_IPC_PARSE_VERSION if COMPAT
+	select ARCH_WANT_DEFAULT_BPF_JIT
 	select ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
 	select ARCH_WANT_FRAME_POINTERS
 	select ARCH_WANT_HUGE_PMD_SHARE if ARM64_4K_PAGES || (ARM64_16K_PAGES && !ARM64_VA_BITS_36)
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 5e8949953660..1f6a0388a65f 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -93,6 +93,7 @@ config X86
 	select ARCH_USE_QUEUED_RWLOCKS
 	select ARCH_USE_QUEUED_SPINLOCKS
 	select ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
+	select ARCH_WANT_DEFAULT_BPF_JIT	if X86_64
 	select ARCH_WANTS_DYNAMIC_TASK_STRUCT
 	select ARCH_WANT_HUGE_PMD_SHARE
 	select ARCH_WANTS_THP_SWAP		if X86_64
diff --git a/init/Kconfig b/init/Kconfig
index a34064a031a5..890aaa62efde 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -1604,6 +1604,9 @@ config BPF_SYSCALL
 	  Enable the bpf() system call that allows to manipulate eBPF
 	  programs and maps via file descriptors.
 
+config ARCH_WANT_DEFAULT_BPF_JIT
+	bool
+
 config BPF_JIT_ALWAYS_ON
 	bool "Permanently enable BPF JIT and remove BPF interpreter"
 	depends on BPF_SYSCALL && HAVE_EBPF_JIT && BPF_JIT
@@ -1611,6 +1614,10 @@ config BPF_JIT_ALWAYS_ON
 	  Enables BPF JIT and removes BPF interpreter to avoid
 	  speculative execution of BPF instructions by the interpreter
 
+config BPF_JIT_DEFAULT_ON
+	def_bool ARCH_WANT_DEFAULT_BPF_JIT || BPF_JIT_ALWAYS_ON
+	depends on HAVE_EBPF_JIT && BPF_JIT
+
 config USERFAULTFD
 	bool "Enable userfaultfd() system call"
 	depends on MMU
diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
index 49e32acad7d8..2ff01a716128 100644
--- a/kernel/bpf/core.c
+++ b/kernel/bpf/core.c
@@ -520,9 +520,9 @@ void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
 
 #ifdef CONFIG_BPF_JIT
 /* All BPF JIT sysctl knobs here. */
-int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
+int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
+int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
 int bpf_jit_harden   __read_mostly;
-int bpf_jit_kallsyms __read_mostly;
 long bpf_jit_limit   __read_mostly;
 
 static __always_inline void
-- 
2.21.0


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH bpf-next v2] bpf, x86, arm64: enable jit by default when not built as always-on
  2019-12-09 15:08 [PATCH bpf-next v2] bpf, x86, arm64: enable jit by default when not built as always-on Daniel Borkmann
@ 2019-12-09 17:06 ` Martin Lau
  2019-12-12  0:49   ` Alexei Starovoitov
  0 siblings, 1 reply; 3+ messages in thread
From: Martin Lau @ 2019-12-09 17:06 UTC (permalink / raw)
  To: Daniel Borkmann; +Cc: ast, netdev, bpf, will

On Mon, Dec 09, 2019 at 04:08:03PM +0100, Daniel Borkmann wrote:
> After Spectre 2 fix via 290af86629b2 ("bpf: introduce BPF_JIT_ALWAYS_ON
> config") most major distros use BPF_JIT_ALWAYS_ON configuration these days
> which compiles out the BPF interpreter entirely and always enables the
> JIT. Also given recent fix in e1608f3fa857 ("bpf: Avoid setting bpf insns
> pages read-only when prog is jited"), we additionally avoid fragmenting
> the direct map for the BPF insns pages sitting in the general data heap
> since they are not used during execution. Latter is only needed when run
> through the interpreter.
> 
> Since both x86 and arm64 JITs have seen a lot of exposure over the years,
> are generally most up to date and maintained, there is more downside in
> !BPF_JIT_ALWAYS_ON configurations to have the interpreter enabled by default
> rather than the JIT. Add a ARCH_WANT_DEFAULT_BPF_JIT config which archs can
> use to set the bpf_jit_{enable,kallsyms} to 1. Back in the days the
> bpf_jit_kallsyms knob was set to 0 by default since major distros still
> had /proc/kallsyms addresses exposed to unprivileged user space which is
> not the case anymore. Hence both knobs are set via BPF_JIT_DEFAULT_ON which
> is set to 'y' in case of BPF_JIT_ALWAYS_ON or ARCH_WANT_DEFAULT_BPF_JIT.
Acked-by: Martin KaFai Lau <kafai@fb.com>

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH bpf-next v2] bpf, x86, arm64: enable jit by default when not built as always-on
  2019-12-09 17:06 ` Martin Lau
@ 2019-12-12  0:49   ` Alexei Starovoitov
  0 siblings, 0 replies; 3+ messages in thread
From: Alexei Starovoitov @ 2019-12-12  0:49 UTC (permalink / raw)
  To: Martin Lau; +Cc: Daniel Borkmann, ast, netdev, bpf, will

On Mon, Dec 9, 2019 at 9:07 AM Martin Lau <kafai@fb.com> wrote:
>
> On Mon, Dec 09, 2019 at 04:08:03PM +0100, Daniel Borkmann wrote:
> > After Spectre 2 fix via 290af86629b2 ("bpf: introduce BPF_JIT_ALWAYS_ON
> > config") most major distros use BPF_JIT_ALWAYS_ON configuration these days
> > which compiles out the BPF interpreter entirely and always enables the
> > JIT. Also given recent fix in e1608f3fa857 ("bpf: Avoid setting bpf insns
> > pages read-only when prog is jited"), we additionally avoid fragmenting
> > the direct map for the BPF insns pages sitting in the general data heap
> > since they are not used during execution. Latter is only needed when run
> > through the interpreter.
> >
> > Since both x86 and arm64 JITs have seen a lot of exposure over the years,
> > are generally most up to date and maintained, there is more downside in
> > !BPF_JIT_ALWAYS_ON configurations to have the interpreter enabled by default
> > rather than the JIT. Add a ARCH_WANT_DEFAULT_BPF_JIT config which archs can
> > use to set the bpf_jit_{enable,kallsyms} to 1. Back in the days the
> > bpf_jit_kallsyms knob was set to 0 by default since major distros still
> > had /proc/kallsyms addresses exposed to unprivileged user space which is
> > not the case anymore. Hence both knobs are set via BPF_JIT_DEFAULT_ON which
> > is set to 'y' in case of BPF_JIT_ALWAYS_ON or ARCH_WANT_DEFAULT_BPF_JIT.
> Acked-by: Martin KaFai Lau <kafai@fb.com>

Applied. Thanks

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2019-12-12  0:49 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-12-09 15:08 [PATCH bpf-next v2] bpf, x86, arm64: enable jit by default when not built as always-on Daniel Borkmann
2019-12-09 17:06 ` Martin Lau
2019-12-12  0:49   ` Alexei Starovoitov

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).