Reduce overhead of LSMs with static calls
Subsystems (2)
v14 cover Reduce overhead of LSMs with static calls
# Background
LSM hooks (callbacks) are currently invoked as indirect function calls. These callbacks are registered into a linked list at boot time as the order of the LSMs can be configured on the kernel command line with the "lsm=" command line parameter.
Indirect function calls have a high overhead due to retpoline mitigation for various speculative execution attacks.
Retpolines remain relevant even with newer generation CPUs as recently discovered speculative attacks, like Spectre BHB need Retpolines to mitigate against branch history injection and still need to be used in combination with newer mitigation features like eIBRS.
This overhead is especially significant for the "bpf" LSM which allows the user to implement LSM functionality with eBPF program. In order to facilitate this the "bpf" LSM provides a default callback for all LSM hooks. When enabled, the "bpf" LSM incurs an unnecessary / avoidable indirect call. This is especially bad in OS hot paths (e.g. in the networking stack). This overhead prevents the adoption of bpf LSM on performance critical systems, and also, in general, slows down all LSMs.
Since we know the address of the enabled LSM callbacks at compile time and only the order is determined at boot time, the LSM framework can allocate static calls for each of the possible LSM callbacks and these calls can be updated once the order is determined at boot.
This series is a respin of the RFC proposed by Paul Renauld ([email protected]) and Brendan Jackman ([email protected]) [1]
# Performance improvement
With this patch-set some syscalls with lots of LSM hooks in their path benefitted at an average of ~3% and I/O and Pipe based system calls benefitting the most.
Here are the results of the relevant Unixbench system benchmarks with BPF LSM and SELinux enabled with default policies enabled with and without these patches.
Benchmark Delta(%): (+ is better)
===============================================================================
Execl Throughput +1.9356 File Write 1024 bufsize 2000 maxblocks +6.5953 Pipe Throughput +9.5499 Pipe-based Context Switching +3.0209 Process Creation +2.3246 Shell Scripts (1 concurrent) +1.4975 System Call Overhead +2.7815 System Benchmarks Index Score (Partial Only): +3.4859
In the best case, some syscalls like eventfd_create benefitted to about ~10%. The full analysis can be viewed at https://kpsingh.ch/lsm-perf
More context in the excellent LWN article by Jonathan Corbet [2]
[1] https://lore.kernel.org/linux-security-module/[email protected]/
[2] https://lwn.net/Articles/979683/
# v13 to v14
* Dropped Patch 5 based on the ongoing discussion in
https://lore.kernel.org/linux-security-module/[email protected]/, BPF LSM will still have default callbacks enabled.
* Dropped Patch 4 as recommended by Paul, indirect calls will remain in some LSM hooks for now.
https://lore.kernel.org/linux-security-module/[email protected]/
* Fixed minor nits in Patch 3
# v12 to v13
* Fixed the GCC note when struct randomization is enabled
* Added missing link to BPF LSM side effects discussion in the commit message
# v11 to v12
* Casey's feedback on readability of patch 4
* Tetsuo's catch on the bug in Patch 5
* Changed LSM_HOOK_TOGGLEABLE to LSM_HOOK_RUNTIME, the bikeshed is blue now.
* Also, as security_toggle_hook relies on iterating over a
flat lsm_static_calls_table, I added the __packed attribute.
# v10 to v11
* bpf_lsm_toggle_hook to security_toggle_hook with LSM_HOOK_TOGGLEABLE limiting the hooks as the ones that can be toggled.
# v9 to v10
* Addressed Paul's comments for Patch 3. I did not remove the acks from this one as changes were minor.
* Moved BPF LSM specific hook toggling logic bpf_lsm_toggle_hook to s security_toggle_hook as a generic API. I removed the Ack's from this patch as it's worth another look.
* Refactored the non-standard hooks to use static calls.
# v8 to v9
Paul, I removed the 5th patch about CONFIG_SECURITY_HOOK_LIKELY and went through all the feedback. I believe it all should be addressed now.
But, please let me know if I missed anything.
The patches are based on https://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/lsm.git
(next branch as of 2024-02-07) and resolved a bunch of conflicts.
I also added Andrii's series ack to indidividual patches.
# v7 to v8
* Addressed Andrii's feedback
* Rebased (this seems to have removed the syscall changes). v7 has the required conflict resolution incase the conflicts need to be resolved again.
# v6 -> v7
* Rebased with latest LSM id changes merged
NOTE: The warning shown by the kernel test bot is spurious, there is no flex array and it seems to come from an older tool chain.
https://lore.kernel.org/bpf/[email protected]/
# v5 -> v6
* Fix a bug in BPF LSM hook toggle logic.
# v4 -> v5
* Rebase to linux-next/master
* Fixed the case where MAX_LSM_COUNT comes to zero when just CONFIG_SECURITY is compiled in without any other LSM enabled as reported here:
https://lore.kernel.org/bpf/[email protected]
# v3 -> v4
* Refactor LSM count macros to use COUNT_ARGS
* Change CONFIG_SECURITY_HOOK_LIKELY likely's default value to be based on the LSM enabled and have it depend on CONFIG_EXPERT. There are a lot of subtle options behind CONFIG_EXPERT and this should, hopefully alleviate concerns about yet another knob.
* __randomize_layout for struct lsm_static_call and, in addition to the cover letter add performance numbers to 3rd patch and some minor commit message updates.
* Rebase to linux-next.
# v2 -> v3
* Fixed a build issue on archs which don't have static calls and enable CONFIG_SECURITY.
* Updated the LSM_COUNT macros based on Andrii's suggestions.
* Changed the security_ prefix to lsm_prefix based on Casey's suggestion.
* Inlined static_branch_maybe into lsm_for_each_hook on Kees' feedback.
# v1 -> v2 (based on linux-next, next-20230614)
* Incorporated suggestions from Kees
* Changed the way MAX_LSMs are counted from a binary based generator to a clever header.
* Add CONFIG_SECURITY_HOOK_LIKELY to configure the likelihood of LSM hooks.
KP Singh (3):
kernel: Add helper macros for loop unrolling
security: Count the LSMs enabled at compile time
security: Replace indirect LSM hook calls with static calls
include/linux/args.h | 6 +- include/linux/lsm_count.h | 128 +++++++++++++++++++++++ include/linux/lsm_hooks.h | 53 ++++++++-- include/linux/unroll.h | 36 +++++++ security/security.c | 215 ++++++++++++++++++++++++++------------ 5 files changed, 362 insertions(+), 76 deletions(-) create mode 100644 include/linux/lsm_count.h create mode 100644 include/linux/unroll.h
Signature · 1 line
2.45.2.803.g4e1b14247a-goog
PATCH v14 1/3 kernel: Add helper macros for loop unrolling
This helps in easily initializing blocks of code (e.g. static calls and keys).
UNROLL(N, MACRO, __VA_ARGS__) calls MACRO N times with the first argument as the index of the iteration. This allows string pasting to create unique tokens for variable names, function calls etc.
As an example:
#include <linux/unroll.h>
#define MACRO(N, a, b) \
int add_##N(int a, int b) \
{ \
return a + b + N; \
}
UNROLL(2, MACRO, x, y)expands to:
int add_0(int x, int y)
{
return x + y + 0;
}
int add_1(int x, int y)
{
return x + y + 1;
}--- include/linux/unroll.h | 36 ++++++++++++++++++++++++++++++++++++ 1 file changed, 36 insertions(+) create mode 100644 include/linux/unroll.h
include/linux/unroll.h new file +36 -0
@@ -0,0 +1,36 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +/* +* Copyright (C) 2023 Google LLC. +*/ + +#ifndef __UNROLL_H +#define __UNROLL_H + +#include <linux/args.h> + +#define UNROLL(N, MACRO, args...) CONCATENATE(__UNROLL_, N)(MACRO, args) + +#define __UNROLL_0(MACRO, args...) +#define __UNROLL_1(MACRO, args...) __UNROLL_0(MACRO, args) MACRO(0, args) +#define __UNROLL_2(MACRO, args...) __UNROLL_1(MACRO, args) MACRO(1, args) +#define __UNROLL_3(MACRO, args...) __UNROLL_2(MACRO, args) MACRO(2, args) +#define __UNROLL_4(MACRO, args...) __UNROLL_3(MACRO, args) MACRO(3, args) +#define __UNROLL_5(MACRO, args...) __UNROLL_4(MACRO, args) MACRO(4, args) +#define __UNROLL_6(MACRO, args...) __UNROLL_5(MACRO, args) MACRO(5, args) +#define __UNROLL_7(MACRO, args...) __UNROLL_6(MACRO, args) MACRO(6, args) +#define __UNROLL_8(MACRO, args...) __UNROLL_7(MACRO, args) MACRO(7, args) +#define __UNROLL_9(MACRO, args...) __UNROLL_8(MACRO, args) MACRO(8, args) +#define __UNROLL_10(MACRO, args...) __UNROLL_9(MACRO, args) MACRO(9, args) +#define __UNROLL_11(MACRO, args...) __UNROLL_10(MACRO, args) MACRO(10, args) +#define __UNROLL_12(MACRO, args...) __UNROLL_11(MACRO, args) MACRO(11, args) +#define __UNROLL_13(MACRO, args...) __UNROLL_12(MACRO, args) MACRO(12, args) +#define __UNROLL_14(MACRO, args...) __UNROLL_13(MACRO, args) MACRO(13, args) +#define __UNROLL_15(MACRO, args...) __UNROLL_14(MACRO, args) MACRO(14, args) +#define __UNROLL_16(MACRO, args...) __UNROLL_15(MACRO, args) MACRO(15, args) +#define __UNROLL_17(MACRO, args...) __UNROLL_16(MACRO, args) MACRO(16, args) +#define __UNROLL_18(MACRO, args...) __UNROLL_17(MACRO, args) MACRO(17, args) +#define __UNROLL_19(MACRO, args...) __UNROLL_18(MACRO, args) MACRO(18, args) +#define __UNROLL_20(MACRO, args...) __UNROLL_19(MACRO, args) MACRO(19, args) + +#endif /* __UNROLL_H */
Signature · 1 line
2.45.2.803.g4e1b14247a-goog
PATCH v14 2/3 security: Count the LSMs enabled at compile time
In reply to KP SinghThese macros are a clever trick to determine a count of the number of LSMs that are enabled in the config to ascertain the maximum number of static calls that need to be configured per LSM hook.
Without this one would need to generate static calls for the total number of LSMs in the kernel (even if they are not compiled) times the number of LSM hooks which ends up being quite wasteful.
--- include/linux/args.h | 6 +- include/linux/lsm_count.h | 128 ++++++++++++++++++++++++++++++++++++++ 2 files changed, 131 insertions(+), 3 deletions(-) create mode 100644 include/linux/lsm_count.h
include/linux/args.h +3 -3
@@ -17,9 +17,9 @@* that as _n. */ -/* This counts to 12. Any more, it will return 13th argument. */ -#define __COUNT_ARGS(_0, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _n, X...) _n -#define COUNT_ARGS(X...) __COUNT_ARGS(, ##X, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0) +/* This counts to 15. Any more, it will return 16th argument. */ +#define __COUNT_ARGS(_0, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _n, X...) _n +#define COUNT_ARGS(X...) __COUNT_ARGS(, ##X, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0)/* Concatenate two parameters, but allow them to be expanded beforehand. */ #define __CONCAT (a, b) a ## b
include/linux/lsm_count.h new file +128 -0
128 lines not on this page · raw message
Signature · 1 line
2.45.2.803.g4e1b14247a-goog
PATCH v14 3/3 security: Replace indirect LSM hook calls with static calls
In reply to KP SinghLSM hooks are currently invoked from a linked list as indirect calls which are invoked using retpolines as a mitigation for speculative attacks (Branch History / Target injection) and add extra overhead which is especially bad in kernel hot paths:
security_file_ioctl:
0xff...0320 <+0>: endbr64
0xff...0324 <+4>: push %rbp 0xff...0325 <+5>: push %r15 0xff...0327 <+7>: push %r14 0xff...0329 <+9>: push %rbx 0xff...032a <+10>: mov %rdx,%rbx 0xff...032d <+13>: mov %esi,%ebp 0xff...032f <+15>: mov %rdi,%r14 0xff...0332 <+18>: mov $0xff...7030,%r15 0xff...0339 <+25>: mov (%r15),%r15 0xff...033c <+28>: test %r15,%r15 0xff...033f <+31>: je 0xff...0358 <security_file_ioctl+56> 0xff...0341 <+33>: mov 0x18(%r15),%r11 0xff...0345 <+37>: mov %r14,%rdi 0xff...0348 <+40>: mov %ebp,%esi 0xff...034a <+42>: mov %rbx,%rdx
0xff...034d <+45>: call 0xff...2e0 <__x86_indirect_thunk_array+352>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Indirect calls that use retpolines leading to overhead, not just due to extra instruction but also branch misses.
0xff...0352 <+50>: test %eax,%eax 0xff...0354 <+52>: je 0xff...0339 <security_file_ioctl+25> 0xff...0356 <+54>: jmp 0xff...035a <security_file_ioctl+58> 0xff...0358 <+56>: xor %eax,%eax 0xff...035a <+58>: pop %rbx 0xff...035b <+59>: pop %r14 0xff...035d <+61>: pop %r15 0xff...035f <+63>: pop %rbp 0xff...0360 <+64>: jmp 0xff...47c4 <__x86_return_thunk>
The indirect calls are not really needed as one knows the addresses of enabled LSM callbacks at boot time and only the order can possibly change at boot time with the lsm= kernel command line parameter.
An array of static calls is defined per LSM hook and the static calls are updated at boot time once the order has been determined.
A static key guards whether an LSM static call is enabled or not, without this static key, for LSM hooks that return an int, the presence of the hook that returns a default value can create side-effects which has resulted in bugs [1].
With the hook now exposed as a static call, one can see that the retpolines are no longer there and the LSM callbacks are invoked directly:
security_file_ioctl:
0xff...0ca0 <+0>: endbr64
0xff...0ca4 <+4>: nopl 0x0(%rax,%rax,1)
0xff...0ca9 <+9>: push %rbp
0xff...0caa <+10>: push %r14 0xff...0cac <+12>: push %rbx 0xff...0cad <+13>: mov %rdx,%rbx 0xff...0cb0 <+16>: mov %esi,%ebp 0xff...0cb2 <+18>: mov %rdi,%r14 0xff...0cb5 <+21>: jmp 0xff...0cc7 <security_file_ioctl+39>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Static key enabled for SELinux
0xffffffff818f0cb7 <+23>: jmp 0xff...0cde <security_file_ioctl+62>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Static key enabled for BPF LSM. This is something that is changed to default to false to avoid the existing side effect issues of BPF LSM
[1] in a subsequent patch.
0xff...0cb9 <+25>: xor %eax,%eax 0xff...0cbb <+27>: xchg %ax,%ax 0xff...0cbd <+29>: pop %rbx 0xff...0cbe <+30>: pop %r14 0xff...0cc0 <+32>: pop %rbp 0xff...0cc1 <+33>: cs jmp 0xff...0000 <__x86_return_thunk> 0xff...0cc7 <+39>: endbr64 0xff...0ccb <+43>: mov %r14,%rdi 0xff...0cce <+46>: mov %ebp,%esi 0xff...0cd0 <+48>: mov %rbx,%rdx 0xff...0cd3 <+51>: call 0xff...3230 <selinux_file_ioctl>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Direct call to SELinux.
0xff...0cd8 <+56>: test %eax,%eax 0xff...0cda <+58>: jne 0xff...0cbd <security_file_ioctl+29> 0xff...0cdc <+60>: jmp 0xff...0cb7 <security_file_ioctl+23> 0xff...0cde <+62>: endbr64 0xff...0ce2 <+66>: mov %r14,%rdi 0xff...0ce5 <+69>: mov %ebp,%esi 0xff...0ce7 <+71>: mov %rbx,%rdx 0xff...0cea <+74>: call 0xff...e220 <bpf_lsm_file_ioctl>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Direct call to BPF LSM.
0xff...0cef <+79>: test %eax,%eax 0xff...0cf1 <+81>: jne 0xff...0cbd <security_file_ioctl+29> 0xff...0cf3 <+83>: jmp 0xff...0cb9 <security_file_ioctl+25> 0xff...0cf5 <+85>: endbr64 0xff...0cf9 <+89>: mov %r14,%rdi 0xff...0cfc <+92>: mov %ebp,%esi 0xff...0cfe <+94>: mov %rbx,%rdx 0xff...0d01 <+97>: pop %rbx 0xff...0d02 <+98>: pop %r14
0xff...0d04 <+100>: pop %rbp
0xff...0d05 <+101>: ret
0xff...0d06 <+102>: int3
0xff...0d07 <+103>: int3
0xff...0d08 <+104>: int3
0xff...0d09 <+105>: int3
While this patch uses static_branch_unlikely indicating that an LSM hook is likely to be not present. In most cases this is still a better choice as even when an LSM with one hook is added, empty slots are created for all LSM hooks (especially when many LSMs that do not initialize most hooks are present on the system).
There are some hooks that don't use the call_int_hook or call_void_hook. These hooks are updated to use a new macro called lsm_for_each_hook where the lsm_callback is directly invoked as an indirect call.
Below are results of the relevant Unixbench system benchmarks with BPF LSM and SELinux enabled with default policies enabled with and without these patches.
Benchmark Delta(%): (+ is better)
===============================================================================
Execl Throughput +1.9356 File Write 1024 bufsize 2000 maxblocks +6.5953 Pipe Throughput +9.5499 Pipe-based Context Switching +3.0209 Process Creation +2.3246 Shell Scripts (1 concurrent) +1.4975 System Call Overhead +2.7815 System Benchmarks Index Score (Partial Only): +3.4859
In the best case, some syscalls like eventfd_create benefitted to about ~10%.
[1] https://lore.kernel.org/linux-security-module/[email protected]/
--- include/linux/lsm_hooks.h | 53 ++++++++-- security/security.c | 215 ++++++++++++++++++++++++++------------ 2 files changed, 195 insertions(+), 73 deletions(-)
include/linux/lsm_hooks.h +44 -9
@@ -30,19 +30,47 @@#include <linux/init.h> #include <linux/rculist.h> #include <linux/xattr.h> +#include <linux/static_call.h> +#include <linux/unroll.h> +#include <linux/jump_label.h> +#include <linux/lsm_count.h> union security_list_options {#define LSM_HOOK (RET ,DEFAULT ,NAME , ...) RET (*NAME)(__VA_ARGS__);#include "lsm_hook_defs.h"#undef LSM_HOOK +void *lsm_func_addr; }; -struct security_hook_heads { -#define LSM_HOOK (RET ,DEFAULT ,NAME , ...) struct hlist_head NAME; -#include "lsm_hook_defs.h" -#undef LSM_HOOK +/* +* @key: static call key as defined by STATIC_CALL_KEY +* @trampoline: static call trampoline as defined by STATIC_CALL_TRAMP +* @hl: The security_hook_list as initialized by the owning LSM. +* @active: Enabled when the static call has an LSM hook associated. +*/ +struct lsm_static_call { +struct static_call_key *key; +void *trampoline; +struct security_hook_list *hl; +/* this needs to be true or false based on what the key defaults to */ +struct static_key_false *active; } __randomize_layout; +/* +* Table of the static calls for each LSM hook. +* Once the LSMs are initialized, their callbacks will be copied to these +* tables such that the calls are filled backwards (from last to first). +* This way, we can jump directly to the first used static call, and execute +* all of them after. This essentially makes the entry point +* dynamic to adapt the number of static calls to the number of callbacks. +*/ +struct lsm_static_calls_table { +#define LSM_HOOK (RET ,DEFAULT ,NAME , ...) \ + struct lsm_static_call NAME[MAX_LSM_COUNT]; +#include <linux/lsm_hook_defs.h> +#undef LSM_HOOK +} __packed __randomize_layout; +/** * struct lsm_id - Identify a Linux Security Module. * @lsm: name of the LSM, must be approved by the LSM maintainers
@@ -58,10 +86,14 @@ struct lsm_id {/* * Security module hook list structure. * For use with generic list macros for common operations. +* +* struct security_hook_list - Contents of a cacheable, mappable object. +* @scalls: The beginning of the array of static calls assigned to this hook. +* @hook: The callback for the hook. +* @lsm: The name of the lsm that owns this hook. */ struct security_hook_list { -struct hlist_node list; -struct hlist_head *head; +struct lsm_static_call *scalls;union security_list_options hook;const struct lsm_id *lsmid; } __randomize_layout;
@@ -111,10 +143,12 @@ static inline struct xattr *lsm_get_xattr_slot(struct xattr *xattrs,* care of the common case and reduces the amount of * text involved. */ -#define LSM_HOOK_INIT (HEAD ,HOOK ) \ - { .head = &security_hook_heads.HEAD, .hook = { .HEAD = HOOK } } +#define LSM_HOOK_INIT (NAME ,HOOK ) \ + { \ + .scalls = static_calls_table.NAME, \ + .hook = { .NAME = HOOK } \ + } -extern struct security_hook_heads security_hook_heads;extern char *lsm_names; extern void security_add_hooks(struct security_hook_list *hooks, int count,
@@ -152,5 +186,6 @@ extern struct lsm_info __start_early_lsm_info[], __end_early_lsm_info[];__aligned(sizeof(unsigned long)) extern int lsm_inode_alloc (struct inode *inode); +extern struct lsm_static_calls_table static_calls_table__ro_after_init ;#endif /* ! __LINUX_LSM_HOOKS_H */
security/security.c +151 -64
378 lines not on this page · raw message
Signature · 1 line
2.45.2.803.g4e1b14247a-goog
Paul Moore
maintainerRe: [PATCH v14 3/3] security: Replace indirect LSM hook calls with static calls
Quoted 44 lines · On Jul 9, 2024 KP Singh <[email protected]> wrote:
LSM hooks are currently invoked from a linked list as indirect calls which are invoked using retpolines as a mitigation for speculative attacks (Branch History / Target injection) and add extra overhead which is especially bad in kernel hot paths:
security_file_ioctl:0xff...0320 <+0>: endbr64 0xff...0324 <+4>: push %rbp 0xff...0325 <+5>: push %r15 0xff...0327 <+7>: push %r14 0xff...0329 <+9>: push %rbx 0xff...032a <+10>: mov %rdx,%rbx 0xff...032d <+13>: mov %esi,%ebp 0xff...032f <+15>: mov %rdi,%r14 0xff...0332 <+18>: mov $0xff...7030,%r15 0xff...0339 <+25>: mov (%r15),%r15 0xff...033c <+28>: test %r15,%r15 0xff...033f <+31>: je 0xff...0358 <security_file_ioctl+56> 0xff...0341 <+33>: mov 0x18(%r15),%r11 0xff...0345 <+37>: mov %r14,%rdi 0xff...0348 <+40>: mov %ebp,%esi 0xff...034a <+42>: mov %rbx,%rdx 0xff...034d <+45>: call 0xff...2e0 <__x86_indirect_thunk_array+352> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Indirect calls that use retpolines leading to overhead, not just due to extra instruction but also branch misses. 0xff...0352 <+50>: test %eax,%eax 0xff...0354 <+52>: je 0xff...0339 <security_file_ioctl+25> 0xff...0356 <+54>: jmp 0xff...035a <security_file_ioctl+58> 0xff...0358 <+56>: xor %eax,%eax 0xff...035a <+58>: pop %rbx 0xff...035b <+59>: pop %r14 0xff...035d <+61>: pop %r15 0xff...035f <+63>: pop %rbp 0xff...0360 <+64>: jmp 0xff...47c4 <__x86_return_thunk>The indirect calls are not really needed as one knows the addresses of enabled LSM callbacks at boot time and only the order can possibly change at boot time with the lsm= kernel command line parameter.
Quoted 7 lines
An array of static calls is defined per LSM hook and the static calls are updated at boot time once the order has been determined.
A static key guards whether an LSM static call is enabled or not, without this static key, for LSM hooks that return an int, the presence of the hook that returns a default value can create side-effects which has resulted in bugs [1].
I don't want to rehash our previous discussions on this topic, but I do think we either need to simply delete the paragraph above or update it to indicate that all known side effects involving LSM callback return values have been addressed. Removal is likely easier if for no other reason than we don't have to go back and forth with edits, but I can understand if you would prefer to have the paragraph in the commit description, albeit in a revised form. If you want to go with the revised paragraph option, you don't need to keep resubmitting the patchset, once we agree on something I can do the paragraph swap when I merge the patchset.
Otherwise, this patchset looks okay, but as I mentioned earlier, given we are at -rc7 this isn't something that I'm comfortable sending up to Linus during the upcoming merge window. This is v6.12 material at this point.
Quoted 100 lines
With the hook now exposed as a static call, one can see that the retpolines are no longer there and the LSM callbacks are invoked directly:
security_file_ioctl:0xff...0ca0 <+0>: endbr64 0xff...0ca4 <+4>: nopl 0x0(%rax,%rax,1) 0xff...0ca9 <+9>: push %rbp 0xff...0caa <+10>: push %r14 0xff...0cac <+12>: push %rbx 0xff...0cad <+13>: mov %rdx,%rbx 0xff...0cb0 <+16>: mov %esi,%ebp 0xff...0cb2 <+18>: mov %rdi,%r14 0xff...0cb5 <+21>: jmp 0xff...0cc7 <security_file_ioctl+39>^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Static key enabled for SELinux0xffffffff818f0cb7 <+23>: jmp 0xff...0cde <security_file_ioctl+62> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Static key enabled for BPF LSM. This is something that is changed to default to false to avoid the existing side effect issues of BPF LSM [1] in a subsequent patch. 0xff...0cb9 <+25>: xor %eax,%eax 0xff...0cbb <+27>: xchg %ax,%ax 0xff...0cbd <+29>: pop %rbx 0xff...0cbe <+30>: pop %r14 0xff...0cc0 <+32>: pop %rbp 0xff...0cc1 <+33>: cs jmp 0xff...0000 <__x86_return_thunk> 0xff...0cc7 <+39>: endbr64 0xff...0ccb <+43>: mov %r14,%rdi 0xff...0cce <+46>: mov %ebp,%esi 0xff...0cd0 <+48>: mov %rbx,%rdx 0xff...0cd3 <+51>: call 0xff...3230 <selinux_file_ioctl> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Direct call to SELinux. 0xff...0cd8 <+56>: test %eax,%eax 0xff...0cda <+58>: jne 0xff...0cbd <security_file_ioctl+29> 0xff...0cdc <+60>: jmp 0xff...0cb7 <security_file_ioctl+23> 0xff...0cde <+62>: endbr64 0xff...0ce2 <+66>: mov %r14,%rdi 0xff...0ce5 <+69>: mov %ebp,%esi 0xff...0ce7 <+71>: mov %rbx,%rdx 0xff...0cea <+74>: call 0xff...e220 <bpf_lsm_file_ioctl> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Direct call to BPF LSM. 0xff...0cef <+79>: test %eax,%eax 0xff...0cf1 <+81>: jne 0xff...0cbd <security_file_ioctl+29> 0xff...0cf3 <+83>: jmp 0xff...0cb9 <security_file_ioctl+25> 0xff...0cf5 <+85>: endbr64 0xff...0cf9 <+89>: mov %r14,%rdi 0xff...0cfc <+92>: mov %ebp,%esi 0xff...0cfe <+94>: mov %rbx,%rdx 0xff...0d01 <+97>: pop %rbx 0xff...0d02 <+98>: pop %r14 0xff...0d04 <+100>: pop %rbp 0xff...0d05 <+101>: ret 0xff...0d06 <+102>: int3 0xff...0d07 <+103>: int3 0xff...0d08 <+104>: int3 0xff...0d09 <+105>: int3While this patch uses
static_branch_unlikelyindicating that an LSM hook is likely to be not present. In most cases this is still a better choice as even when an LSM with one hook is added, empty slots are created for all LSM hooks (especially when many LSMs that do not initialize most hooks are present on the system).There are some hooks that don't use the
call_int_hookorcall_void_hook. These hooks are updated to use a new macro calledlsm_for_each_hookwhere thelsm_callbackis directly invoked as an indirect call.Below are results of the relevant Unixbench system benchmarks with BPF LSM and SELinux enabled with default policies enabled with and without these patches.
Benchmark Delta(%): (+ is better)
===============================================================================Execl Throughput +1.9356 File Write 1024 bufsize 2000 maxblocks +6.5953 Pipe Throughput +9.5499 Pipe-based Context Switching +3.0209 Process Creation +2.3246 Shell Scripts (1 concurrent) +1.4975 System Call Overhead +2.7815 System Benchmarks Index Score (Partial Only): +3.4859In the best case, some syscalls like
eventfd_createbenefitted to about ~10%.[1] https://lore.kernel.org/linux-security-module/[email protected]/
Reviewed-by Casey Schaufler <[email protected]>Reviewed-by Kees Cook <[email protected]>Acked-by Song Liu <[email protected]>Acked-by Andrii Nakryiko <[email protected]>Signed-off-by KP Singh <[email protected]>
Quoted 4 lines
--- include/linux/lsm_hooks.h | 53 ++++++++-- security/security.c | 215 ++++++++++++++++++++++++++------------ 2 files changed, 195 insertions(+), 73 deletions(-)
--
paul-moore.com
Re: [PATCH v14 3/3] security: Replace indirect LSM hook calls with static calls
Quoted 54 lines · Paul Moore
Quoted 44 lines · On Jul 9, 2024 KP Singh <[email protected]> wrote:
LSM hooks are currently invoked from a linked list as indirect calls which are invoked using retpolines as a mitigation for speculative attacks (Branch History / Target injection) and add extra overhead which is especially bad in kernel hot paths:
security_file_ioctl:0xff...0320 <+0>: endbr64 0xff...0324 <+4>: push %rbp 0xff...0325 <+5>: push %r15 0xff...0327 <+7>: push %r14 0xff...0329 <+9>: push %rbx 0xff...032a <+10>: mov %rdx,%rbx 0xff...032d <+13>: mov %esi,%ebp 0xff...032f <+15>: mov %rdi,%r14 0xff...0332 <+18>: mov $0xff...7030,%r15 0xff...0339 <+25>: mov (%r15),%r15 0xff...033c <+28>: test %r15,%r15 0xff...033f <+31>: je 0xff...0358 <security_file_ioctl+56> 0xff...0341 <+33>: mov 0x18(%r15),%r11 0xff...0345 <+37>: mov %r14,%rdi 0xff...0348 <+40>: mov %ebp,%esi 0xff...034a <+42>: mov %rbx,%rdx0xff...034d <+45>: call 0xff...2e0 <__x86_indirect_thunk_array+352> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Indirect calls that use retpolines leading to overhead, not just due to extra instruction but also branch misses. 0xff...0352 <+50>: test %eax,%eax 0xff...0354 <+52>: je 0xff...0339 <security_file_ioctl+25> 0xff...0356 <+54>: jmp 0xff...035a <security_file_ioctl+58> 0xff...0358 <+56>: xor %eax,%eax 0xff...035a <+58>: pop %rbx 0xff...035b <+59>: pop %r14 0xff...035d <+61>: pop %r15 0xff...035f <+63>: pop %rbp 0xff...0360 <+64>: jmp 0xff...47c4 <__x86_return_thunk>The indirect calls are not really needed as one knows the addresses of enabled LSM callbacks at boot time and only the order can possibly change at boot time with the lsm= kernel command line parameter.
Quoted 7 lines
An array of static calls is defined per LSM hook and the static calls are updated at boot time once the order has been determined.
A static key guards whether an LSM static call is enabled or not, without this static key, for LSM hooks that return an int, the presence of the hook that returns a default value can create side-effects which has resulted in bugs [1].
Quoted 5 lines
I don't want to rehash our previous discussions on this topic, but I do think we either need to simply delete the paragraph above or update it to indicate that all known side effects involving LSM callback return values have been addressed. Removal is likely easier if for no other reason than we don't have to go back and forth with edits, but I can
Agreed, we can just delete this paragraph. Thanks!
- KP
Quoted 118 lines
understand if you would prefer to have the paragraph in the commit description, albeit in a revised form. If you want to go with the revised paragraph option, you don't need to keep resubmitting the patchset, once we agree on something I can do the paragraph swap when I merge the patchset.
Otherwise, this patchset looks okay, but as I mentioned earlier, given we are at -rc7 this isn't something that I'm comfortable sending up to Linus during the upcoming merge window. This is v6.12 material at this point.
Quoted 100 lines
With the hook now exposed as a static call, one can see that the retpolines are no longer there and the LSM callbacks are invoked directly:
security_file_ioctl:0xff...0ca0 <+0>: endbr64 0xff...0ca4 <+4>: nopl 0x0(%rax,%rax,1) 0xff...0ca9 <+9>: push %rbp 0xff...0caa <+10>: push %r14 0xff...0cac <+12>: push %rbx 0xff...0cad <+13>: mov %rdx,%rbx 0xff...0cb0 <+16>: mov %esi,%ebp 0xff...0cb2 <+18>: mov %rdi,%r14 0xff...0cb5 <+21>: jmp 0xff...0cc7 <security_file_ioctl+39> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^Static key enabled for SELinux
0xffffffff818f0cb7 <+23>: jmp 0xff...0cde <security_file_ioctl+62> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Static key enabled for BPF LSM. This is something that is changed to default to false to avoid the existing side effect issues of BPF LSM [1] in a subsequent patch. 0xff...0cb9 <+25>: xor %eax,%eax 0xff...0cbb <+27>: xchg %ax,%ax 0xff...0cbd <+29>: pop %rbx 0xff...0cbe <+30>: pop %r14 0xff...0cc0 <+32>: pop %rbp 0xff...0cc1 <+33>: cs jmp 0xff...0000 <__x86_return_thunk> 0xff...0cc7 <+39>: endbr64 0xff...0ccb <+43>: mov %r14,%rdi 0xff...0cce <+46>: mov %ebp,%esi 0xff...0cd0 <+48>: mov %rbx,%rdx 0xff...0cd3 <+51>: call 0xff...3230 <selinux_file_ioctl> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Direct call to SELinux. 0xff...0cd8 <+56>: test %eax,%eax 0xff...0cda <+58>: jne 0xff...0cbd <security_file_ioctl+29> 0xff...0cdc <+60>: jmp 0xff...0cb7 <security_file_ioctl+23> 0xff...0cde <+62>: endbr64 0xff...0ce2 <+66>: mov %r14,%rdi 0xff...0ce5 <+69>: mov %ebp,%esi 0xff...0ce7 <+71>: mov %rbx,%rdx 0xff...0cea <+74>: call 0xff...e220 <bpf_lsm_file_ioctl> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Direct call to BPF LSM. 0xff...0cef <+79>: test %eax,%eax 0xff...0cf1 <+81>: jne 0xff...0cbd <security_file_ioctl+29> 0xff...0cf3 <+83>: jmp 0xff...0cb9 <security_file_ioctl+25> 0xff...0cf5 <+85>: endbr64 0xff...0cf9 <+89>: mov %r14,%rdi 0xff...0cfc <+92>: mov %ebp,%esi 0xff...0cfe <+94>: mov %rbx,%rdx 0xff...0d01 <+97>: pop %rbx 0xff...0d02 <+98>: pop %r14 0xff...0d04 <+100>: pop %rbp 0xff...0d05 <+101>: ret 0xff...0d06 <+102>: int3 0xff...0d07 <+103>: int3 0xff...0d08 <+104>: int3 0xff...0d09 <+105>: int3While this patch uses
static_branch_unlikelyindicating that an LSM hook is likely to be not present. In most cases this is still a better choice as even when an LSM with one hook is added, empty slots are created for all LSM hooks (especially when many LSMs that do not initialize most hooks are present on the system).There are some hooks that don't use the
call_int_hookorcall_void_hook. These hooks are updated to use a new macro calledlsm_for_each_hookwhere thelsm_callbackis directly invoked as an indirect call.Below are results of the relevant Unixbench system benchmarks with BPF LSM and SELinux enabled with default policies enabled with and without these patches.
Benchmark Delta(%): (+ is better)
===============================================================================Execl Throughput +1.9356 File Write 1024 bufsize 2000 maxblocks +6.5953 Pipe Throughput +9.5499 Pipe-based Context Switching +3.0209 Process Creation +2.3246 Shell Scripts (1 concurrent) +1.4975 System Call Overhead +2.7815 System Benchmarks Index Score (Partial Only): +3.4859In the best case, some syscalls like
eventfd_createbenefitted to about ~10%.[1] https://lore.kernel.org/linux-security-module/[email protected]/
Reviewed-by Casey Schaufler <[email protected]>Reviewed-by Kees Cook <[email protected]>Acked-by Song Liu <[email protected]>Acked-by Andrii Nakryiko <[email protected]>Signed-off-by KP Singh <[email protected]>Quoted 4 lines
--- include/linux/lsm_hooks.h | 53 ++++++++-- security/security.c | 215 ++++++++++++++++++++++++++------------ 2 files changed, 195 insertions(+), 73 deletions(-)--
paul-moore.com
Paul Moore
maintainerRe: [PATCH v14 3/3] security: Replace indirect LSM hook calls with static calls
Quoted 2 lines · KP Singh
Quoted 1 line · Paul Moore
On Jul 9, 2024 KP Singh <[email protected]> wrote:
...
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
A static key guards whether an LSM static call is enabled or not, without this static key, for LSM hooks that return an int, the presence of the hook that returns a default value can create side-effects which has resulted in bugs [1].
Quoted 7 lines
Quoted 5 lines
I don't want to rehash our previous discussions on this topic, but I do think we either need to simply delete the paragraph above or update it to indicate that all known side effects involving LSM callback return values have been addressed. Removal is likely easier if for no other reason than we don't have to go back and forth with edits, but I can
Agreed, we can just delete this paragraph. Thanks!
Okay, I'll do that. I'll send another note when it is merged into lsm/dev, but as I said earlier, that is likely a few weeks out. This will likely end up in lsm/dev-staging before that for testing, etc.
Signature · 1 line
paul-moore.com
Paul Moore
maintainerRe: [PATCH v14 3/3] security: Replace indirect LSM hook calls with static calls
Quoted 17 lines · Paul Moore
Quoted 2 lines · KP Singh
Quoted 1 line · Paul Moore
On Jul 9, 2024 KP Singh <[email protected]> wrote:
...
Quoted 11 lines
Quoted 10 lines
Quoted 4 lines
A static key guards whether an LSM static call is enabled or not, without this static key, for LSM hooks that return an int, the presence of the hook that returns a default value can create side-effects which has resulted in bugs [1].
I don't want to rehash our previous discussions on this topic, but I do think we either need to simply delete the paragraph above or update it to indicate that all known side effects involving LSM callback return values have been addressed. Removal is likely easier if for no other reason than we don't have to go back and forth with edits, but I can
Quoted 5 lines
Quoted 1 line
Agreed, we can just delete this paragraph. Thanks!
Okay, I'll do that. I'll send another note when it is merged into lsm/dev, but as I said earlier, that is likely a few weeks out. This will likely end up in lsm/dev-staging before that for testing, etc.
Quick follow-up that these patches are now in lsm/dev-staging, I'll send another note when they are merged into lsm/{dev,next}.
Signature · 1 line
paul-moore.com
Re: [PATCH v14 3/3] security: Replace indirect LSM hook calls with static calls
In reply to KP SinghQuoted 144 lines · KP Singh
LSM hooks are currently invoked from a linked list as indirect calls which are invoked using retpolines as a mitigation for speculative attacks (Branch History / Target injection) and add extra overhead which is especially bad in kernel hot paths:
security_file_ioctl:0xff...0320 <+0>: endbr64 0xff...0324 <+4>: push %rbp 0xff...0325 <+5>: push %r15 0xff...0327 <+7>: push %r14 0xff...0329 <+9>: push %rbx 0xff...032a <+10>: mov %rdx,%rbx 0xff...032d <+13>: mov %esi,%ebp 0xff...032f <+15>: mov %rdi,%r14 0xff...0332 <+18>: mov $0xff...7030,%r15 0xff...0339 <+25>: mov (%r15),%r15 0xff...033c <+28>: test %r15,%r15 0xff...033f <+31>: je 0xff...0358 <security_file_ioctl+56> 0xff...0341 <+33>: mov 0x18(%r15),%r11 0xff...0345 <+37>: mov %r14,%rdi 0xff...0348 <+40>: mov %ebp,%esi 0xff...034a <+42>: mov %rbx,%rdx 0xff...034d <+45>: call 0xff...2e0 <__x86_indirect_thunk_array+352> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Indirect calls that use retpolines leading to overhead, not just due to extra instruction but also branch misses. 0xff...0352 <+50>: test %eax,%eax 0xff...0354 <+52>: je 0xff...0339 <security_file_ioctl+25> 0xff...0356 <+54>: jmp 0xff...035a <security_file_ioctl+58> 0xff...0358 <+56>: xor %eax,%eax 0xff...035a <+58>: pop %rbx 0xff...035b <+59>: pop %r14 0xff...035d <+61>: pop %r15 0xff...035f <+63>: pop %rbp 0xff...0360 <+64>: jmp 0xff...47c4 <__x86_return_thunk>The indirect calls are not really needed as one knows the addresses of enabled LSM callbacks at boot time and only the order can possibly change at boot time with the lsm= kernel command line parameter.
An array of static calls is defined per LSM hook and the static calls are updated at boot time once the order has been determined.
A static key guards whether an LSM static call is enabled or not, without this static key, for LSM hooks that return an int, the presence of the hook that returns a default value can create side-effects which has resulted in bugs [1].
With the hook now exposed as a static call, one can see that the retpolines are no longer there and the LSM callbacks are invoked directly:
security_file_ioctl:0xff...0ca0 <+0>: endbr64 0xff...0ca4 <+4>: nopl 0x0(%rax,%rax,1) 0xff...0ca9 <+9>: push %rbp 0xff...0caa <+10>: push %r14 0xff...0cac <+12>: push %rbx 0xff...0cad <+13>: mov %rdx,%rbx 0xff...0cb0 <+16>: mov %esi,%ebp 0xff...0cb2 <+18>: mov %rdi,%r14 0xff...0cb5 <+21>: jmp 0xff...0cc7 <security_file_ioctl+39>^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Static key enabled for SELinux0xffffffff818f0cb7 <+23>: jmp 0xff...0cde <security_file_ioctl+62> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Static key enabled for BPF LSM. This is something that is changed to default to false to avoid the existing side effect issues of BPF LSM [1] in a subsequent patch. 0xff...0cb9 <+25>: xor %eax,%eax 0xff...0cbb <+27>: xchg %ax,%ax 0xff...0cbd <+29>: pop %rbx 0xff...0cbe <+30>: pop %r14 0xff...0cc0 <+32>: pop %rbp 0xff...0cc1 <+33>: cs jmp 0xff...0000 <__x86_return_thunk> 0xff...0cc7 <+39>: endbr64 0xff...0ccb <+43>: mov %r14,%rdi 0xff...0cce <+46>: mov %ebp,%esi 0xff...0cd0 <+48>: mov %rbx,%rdx 0xff...0cd3 <+51>: call 0xff...3230 <selinux_file_ioctl> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Direct call to SELinux. 0xff...0cd8 <+56>: test %eax,%eax 0xff...0cda <+58>: jne 0xff...0cbd <security_file_ioctl+29> 0xff...0cdc <+60>: jmp 0xff...0cb7 <security_file_ioctl+23> 0xff...0cde <+62>: endbr64 0xff...0ce2 <+66>: mov %r14,%rdi 0xff...0ce5 <+69>: mov %ebp,%esi 0xff...0ce7 <+71>: mov %rbx,%rdx 0xff...0cea <+74>: call 0xff...e220 <bpf_lsm_file_ioctl> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Direct call to BPF LSM. 0xff...0cef <+79>: test %eax,%eax 0xff...0cf1 <+81>: jne 0xff...0cbd <security_file_ioctl+29> 0xff...0cf3 <+83>: jmp 0xff...0cb9 <security_file_ioctl+25> 0xff...0cf5 <+85>: endbr64 0xff...0cf9 <+89>: mov %r14,%rdi 0xff...0cfc <+92>: mov %ebp,%esi 0xff...0cfe <+94>: mov %rbx,%rdx 0xff...0d01 <+97>: pop %rbx 0xff...0d02 <+98>: pop %r14 0xff...0d04 <+100>: pop %rbp 0xff...0d05 <+101>: ret 0xff...0d06 <+102>: int3 0xff...0d07 <+103>: int3 0xff...0d08 <+104>: int3 0xff...0d09 <+105>: int3While this patch uses
static_branch_unlikelyindicating that an LSM hook is likely to be not present. In most cases this is still a better choice as even when an LSM with one hook is added, empty slots are created for all LSM hooks (especially when many LSMs that do not initialize most hooks are present on the system).There are some hooks that don't use the
call_int_hookorcall_void_hook. These hooks are updated to use a new macro calledlsm_for_each_hookwhere thelsm_callbackis directly invoked as an indirect call.Below are results of the relevant Unixbench system benchmarks with BPF LSM and SELinux enabled with default policies enabled with and without these patches.
Benchmark Delta(%): (+ is better)
===============================================================================Execl Throughput +1.9356 File Write 1024 bufsize 2000 maxblocks +6.5953 Pipe Throughput +9.5499 Pipe-based Context Switching +3.0209 Process Creation +2.3246 Shell Scripts (1 concurrent) +1.4975 System Call Overhead +2.7815 System Benchmarks Index Score (Partial Only): +3.4859In the best case, some syscalls like
eventfd_createbenefitted to about ~10%.
Quoted 7 lines
[1] https://lore.kernel.org/linux-security-module/[email protected]/
Reviewed-by Casey Schaufler <[email protected]>Reviewed-by Kees Cook <[email protected]>Acked-by Song Liu <[email protected]>Acked-by Andrii Nakryiko <[email protected]>Signed-off-by KP Singh <[email protected]>
I'm not against about use of static calls.
But I nack this series because of wrong assumption that ignores reality.
Paul Moore
maintainerRe: [PATCH v14 3/3] security: Replace indirect LSM hook calls with static calls
In reply to Paul MooreQuoted 18 lines · Paul Moore
Quoted 17 lines · Paul Moore
Quoted 2 lines · KP Singh
Quoted 1 line · Paul Moore
On Jul 9, 2024 KP Singh <[email protected]> wrote:
...
Quoted 11 lines
Quoted 10 lines
Quoted 4 lines
A static key guards whether an LSM static call is enabled or not, without this static key, for LSM hooks that return an int, the presence of the hook that returns a default value can create side-effects which has resulted in bugs [1].
I don't want to rehash our previous discussions on this topic, but I do think we either need to simply delete the paragraph above or update it to indicate that all known side effects involving LSM callback return values have been addressed. Removal is likely easier if for no other reason than we don't have to go back and forth with edits, but I can
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
Agreed, we can just delete this paragraph. Thanks!
Okay, I'll do that. I'll send another note when it is merged into lsm/dev, but as I said earlier, that is likely a few weeks out. This will likely end up in lsm/dev-staging before that for testing, etc.
Quick follow-up that these patches are now in lsm/dev-staging, I'll send another note when they are merged into lsm/{dev,next}.
One last update, these patches are now in lsm/dev and should go up to Linus during the next merge window.
Signature · 1 line
paul-moore.com