[bpf] bpf: fix default unprivileged allocation limit

Message ID 20181206092732.4206-1-sandipan@linux.ibm.com
State Changes Requested
Delegated to: BPF Maintainers
Headers show
Series
  • [bpf] bpf: fix default unprivileged allocation limit
Related show

Commit Message

Sandipan Das Dec. 6, 2018, 9:27 a.m.
When using a large page size, the default value of the bpf_jit_limit
knob becomes invalid and users are not able to run unprivileged bpf
programs.

The bpf_jit_limit knob is represented internally as a 32-bit signed
integer because of which the default value, i.e. PAGE_SIZE * 40000,
overflows in case of an architecture like powerpc64 which uses 64K
as the default page size (i.e. CONFIG_PPC_64K_PAGES is set).

So, instead of depending on the page size, use a constant value.

Fixes: ede95a63b5e8 ("bpf: add bpf_jit_limit knob to restrict unpriv allocations")
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
---
 kernel/bpf/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Michael Roth Dec. 6, 2018, 4:58 p.m. | #1
Quoting Sandipan Das (2018-12-06 03:27:32)
> When using a large page size, the default value of the bpf_jit_limit
> knob becomes invalid and users are not able to run unprivileged bpf
> programs.
> 
> The bpf_jit_limit knob is represented internally as a 32-bit signed
> integer because of which the default value, i.e. PAGE_SIZE * 40000,
> overflows in case of an architecture like powerpc64 which uses 64K
> as the default page size (i.e. CONFIG_PPC_64K_PAGES is set).
> 
> So, instead of depending on the page size, use a constant value.
> 
> Fixes: ede95a63b5e8 ("bpf: add bpf_jit_limit knob to restrict unpriv allocations")

This also consistently caused a virtio-net KVM Ubuntu 18.04 guest to time out
on configuring networking during boot via systemd/netplan. A bisect
pointed to the same commit this patch addresses.

> Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
> ---
>  kernel/bpf/core.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> index b1a3545d0ec8..a81d097a17fb 100644
> --- a/kernel/bpf/core.c
> +++ b/kernel/bpf/core.c
> @@ -365,7 +365,7 @@ void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
>  }
> 
>  #ifdef CONFIG_BPF_JIT
> -# define BPF_JIT_LIMIT_DEFAULT (PAGE_SIZE * 40000)
> +# define BPF_JIT_LIMIT_DEFAULT (4096 * 40000)

This isn't quite right as we still use (bpf_jit_limit >> PAGE_SHIFT) to check
allocations in bpf_jit_charge_modmem(), so that should be fixed up as well.

Another alternative which is to clamp BPF_JIT_LIMIT_DEFAULT to INT_MAX,
which fixes the issue for me and is similar to what
bpf_jit_charge_init() does for kernels where MODULES_VADDR is defined.
I'll go ahead and send the patch in case that seems preferable.

> 
>  /* All BPF JIT sysctl knobs here. */
>  int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
> -- 
> 2.19.2
>

Patch

diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
index b1a3545d0ec8..a81d097a17fb 100644
--- a/kernel/bpf/core.c
+++ b/kernel/bpf/core.c
@@ -365,7 +365,7 @@  void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
 }
 
 #ifdef CONFIG_BPF_JIT
-# define BPF_JIT_LIMIT_DEFAULT	(PAGE_SIZE * 40000)
+# define BPF_JIT_LIMIT_DEFAULT	(4096 * 40000)
 
 /* All BPF JIT sysctl knobs here. */
 int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);