Reduce overhead of LSMs with static calls
Subsystems (4)
KP Singh
maintainerv5 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
[1] https://lore.kernel.org/linux-security-module/[email protected]/
# BPF LSM Side effects
Patch 4 of the series also addresses the issues with the side effects of the default value return values of the BPF LSM callbacks and also removes the overheads associated with them making it deployable at hyperscale.
# 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 (5):
kernel: Add helper macros for loop unrolling
security: Count the LSMs enabled at compile time
security: Replace indirect LSM hook calls with static calls
bpf: Only enable BPF LSM hooks when an LSM program is attached
security: Add CONFIG_SECURITY_HOOK_LIKELY
include/linux/bpf.h | 1 + include/linux/bpf_lsm.h | 5 + include/linux/lsm_count.h | 114 ++++++++++++++++++++ include/linux/lsm_hooks.h | 81 +++++++++++++-- include/linux/unroll.h | 36 +++++++ kernel/bpf/trampoline.c | 29 +++++- security/Kconfig | 11 ++ security/bpf/hooks.c | 25 ++++- security/security.c | 213 +++++++++++++++++++++++++------------- 9 files changed, 432 insertions(+), 83 deletions(-) create mode 100644 include/linux/lsm_count.h create mode 100644 include/linux/unroll.h
Signature · 1 line
2.42.0.582.g8ccd20d70d-goog
KP Singh
maintainerPATCH v5 1/5 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
2.42.0.582.g8ccd20d70d-goog
KP Singh
maintainerPATCH v5 2/5 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/lsm_count.h | 114 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 114 insertions(+) create mode 100644 include/linux/lsm_count.h
include/linux/lsm_count.h new file +114 -0
@@ -0,0 +1,114 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +/* +* Copyright (C) 2023 Google LLC. +*/ + +#ifndef __LINUX_LSM_COUNT_H +#define __LINUX_LSM_COUNT_H + +#include <linux/args.h> + +#ifdef CONFIG_SECURITY + +/* +* Macros to count the number of LSMs enabled in the kernel at compile time. +*/ + +/* +* Capabilities is enabled when CONFIG_SECURITY is enabled. +*/ +#if IS_ENABLED (CONFIG_SECURITY ) +#define CAPABILITIES_ENABLED 1, +#else +#define CAPABILITIES_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_SELINUX ) +#define SELINUX_ENABLED 1, +#else +#define SELINUX_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_SMACK ) +#define SMACK_ENABLED 1, +#else +#define SMACK_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_APPARMOR ) +#define APPARMOR_ENABLED 1, +#else +#define APPARMOR_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_TOMOYO ) +#define TOMOYO_ENABLED 1, +#else +#define TOMOYO_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_YAMA ) +#define YAMA_ENABLED 1, +#else +#define YAMA_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_LOADPIN ) +#define LOADPIN_ENABLED 1, +#else +#define LOADPIN_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_LOCKDOWN_LSM ) +#define LOCKDOWN_ENABLED 1, +#else +#define LOCKDOWN_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_SAFESETID ) +#define SAFESETID_ENABLED 1, +#else +#define SAFESETID_ENABLED +#endif + +#if IS_ENABLED (CONFIG_BPF_LSM ) +#define BPF_LSM_ENABLED 1, +#else +#define BPF_LSM_ENABLED +#endif + +#if IS_ENABLED (CONFIG_SECURITY_LANDLOCK ) +#define LANDLOCK_ENABLED 1, +#else +#define LANDLOCK_ENABLED +#endif + +/* +* There is a trailing comma that we need to be accounted for. This is done by +* using a skipped argument in __COUNT_LSMS +*/ +#define __COUNT_LSMS(skipped_arg, args...) COUNT_ARGS(args...) +#define COUNT_LSMS(args...) __COUNT_LSMS(args) + +#define MAX_LSM_COUNT \ + COUNT_LSMS( \ + CAPABILITIES_ENABLED \ + SELINUX_ENABLED \ + SMACK_ENABLED \ + APPARMOR_ENABLED \ + TOMOYO_ENABLED \ + YAMA_ENABLED \ + LOADPIN_ENABLED \ + LOCKDOWN_ENABLED \ + SAFESETID_ENABLED \ + BPF_LSM_ENABLED \ + LANDLOCK_ENABLED) + +#else + +#define MAX_LSM_COUNT 0 + +#endif /* CONFIG_SECURITY */ + +#endif /* __LINUX_LSM_COUNT_H */
Signature · 1 line
2.42.0.582.g8ccd20d70d-goog
Re: [PATCH v5 2/5] security: Count the LSMs enabled at compile time
Quoted 4 lines · KP Singh
These 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.
Quoted 8 lines
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.
Suggested-by Kui-Feng Lee <[email protected]>Suggested-by Andrii Nakryiko <[email protected]>Acked-by Song Liu <[email protected]>Signed-off-by KP Singh <[email protected]>
Thanks for doing the refactor with the existing macro!
Signature · 1 line
Kees Cook
KP Singh
maintainerPATCH v5 3/5 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:
0xffffffff814f0320 <+0>: endbr64
0xffffffff814f0324 <+4>: push %rbp 0xffffffff814f0325 <+5>: push %r15 0xffffffff814f0327 <+7>: push %r14 0xffffffff814f0329 <+9>: push %rbx 0xffffffff814f032a <+10>: mov %rdx,%rbx 0xffffffff814f032d <+13>: mov %esi,%ebp 0xffffffff814f032f <+15>: mov %rdi,%r14 0xffffffff814f0332 <+18>: mov $0xffffffff834a7030,%r15 0xffffffff814f0339 <+25>: mov (%r15),%r15 0xffffffff814f033c <+28>: test %r15,%r15 0xffffffff814f033f <+31>: je 0xffffffff814f0358 <security_file_ioctl+56> 0xffffffff814f0341 <+33>: mov 0x18(%r15),%r11 0xffffffff814f0345 <+37>: mov %r14,%rdi 0xffffffff814f0348 <+40>: mov %ebp,%esi 0xffffffff814f034a <+42>: mov %rbx,%rdx
0xffffffff814f034d <+45>: call 0xffffffff81f742e0 <__x86_indirect_thunk_array+352>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Indirect calls that use retpolines leading to overhead, not just due to extra instruction but also branch misses.
0xffffffff814f0352 <+50>: test %eax,%eax 0xffffffff814f0354 <+52>: je 0xffffffff814f0339 <security_file_ioctl+25> 0xffffffff814f0356 <+54>: jmp 0xffffffff814f035a <security_file_ioctl+58> 0xffffffff814f0358 <+56>: xor %eax,%eax 0xffffffff814f035a <+58>: pop %rbx 0xffffffff814f035b <+59>: pop %r14 0xffffffff814f035d <+61>: pop %r15 0xffffffff814f035f <+63>: pop %rbp 0xffffffff814f0360 <+64>: jmp 0xffffffff81f747c4 <__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:
0xffffffff818f0ca0 <+0>: endbr64
0xffffffff818f0ca4 <+4>: nopl 0x0(%rax,%rax,1)
0xffffffff818f0ca9 <+9>: push %rbp
0xffffffff818f0caa <+10>: push %r14 0xffffffff818f0cac <+12>: push %rbx 0xffffffff818f0cad <+13>: mov %rdx,%rbx 0xffffffff818f0cb0 <+16>: mov %esi,%ebp 0xffffffff818f0cb2 <+18>: mov %rdi,%r14 0xffffffff818f0cb5 <+21>: jmp 0xffffffff818f0cc7 <security_file_ioctl+39>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Static key enabled for SELinux
0xffffffff818f0cb7 <+23>: jmp 0xffffffff818f0cde <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.
0xffffffff818f0cb9 <+25>: xor %eax,%eax 0xffffffff818f0cbb <+27>: xchg %ax,%ax 0xffffffff818f0cbd <+29>: pop %rbx 0xffffffff818f0cbe <+30>: pop %r14 0xffffffff818f0cc0 <+32>: pop %rbp 0xffffffff818f0cc1 <+33>: cs jmp 0xffffffff82c00000 <__x86_return_thunk> 0xffffffff818f0cc7 <+39>: endbr64 0xffffffff818f0ccb <+43>: mov %r14,%rdi 0xffffffff818f0cce <+46>: mov %ebp,%esi 0xffffffff818f0cd0 <+48>: mov %rbx,%rdx 0xffffffff818f0cd3 <+51>: call 0xffffffff81903230 <selinux_file_ioctl>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Direct call to SELinux.
0xffffffff818f0cd8 <+56>: test %eax,%eax 0xffffffff818f0cda <+58>: jne 0xffffffff818f0cbd <security_file_ioctl+29> 0xffffffff818f0cdc <+60>: jmp 0xffffffff818f0cb7 <security_file_ioctl+23> 0xffffffff818f0cde <+62>: endbr64 0xffffffff818f0ce2 <+66>: mov %r14,%rdi 0xffffffff818f0ce5 <+69>: mov %ebp,%esi 0xffffffff818f0ce7 <+71>: mov %rbx,%rdx 0xffffffff818f0cea <+74>: call 0xffffffff8141e220 <bpf_lsm_file_ioctl>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Direct call to BPF LSM.
0xffffffff818f0cef <+79>: test %eax,%eax 0xffffffff818f0cf1 <+81>: jne 0xffffffff818f0cbd <security_file_ioctl+29> 0xffffffff818f0cf3 <+83>: jmp 0xffffffff818f0cb9 <security_file_ioctl+25> 0xffffffff818f0cf5 <+85>: endbr64 0xffffffff818f0cf9 <+89>: mov %r14,%rdi 0xffffffff818f0cfc <+92>: mov %ebp,%esi 0xffffffff818f0cfe <+94>: mov %rbx,%rdx 0xffffffff818f0d01 <+97>: pop %rbx 0xffffffff818f0d02 <+98>: pop %r14
0xffffffff818f0d04 <+100>: pop %rbp
0xffffffff818f0d05 <+101>: ret
0xffffffff818f0d06 <+102>: int3
0xffffffff818f0d07 <+103>: int3
0xffffffff818f0d08 <+104>: int3
0xffffffff818f0d09 <+105>: int3
While this patch uses static_branch_unlikely indicating that an LSM hook is likely to be not present, a subsequent makes it configurable. 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 and call_void_hook. These hooks are updated to use a new macro called security_for_each_hook where the lsm_callback is directly invoked as an indirect call. Currently, there are no performance sensitive hooks that use the security_for_each_hook macro. However, if, some performance sensitive hooks are discovered, these can be updated to use static calls with loop unrolling as well using a custom macro.
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%.
--- include/linux/lsm_hooks.h | 70 +++++++++++-- security/security.c | 208 +++++++++++++++++++++++++------------- 2 files changed, 199 insertions(+), 79 deletions(-)
include/linux/lsm_hooks.h +62 -8
@@ -29,26 +29,77 @@#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> + +#define SECURITY_HOOK_ACTIVE_KEY (HOOK ,IDX ) security_hook_active_##HOOK##_##IDX + +/* +* Identifier for the LSM static calls. +* HOOK is an LSM hook as defined in linux/lsm_hookdefs.h +* IDX is the index of the static call. 0 <= NUM < MAX_LSM_COUNT +*/ +#define LSM_STATIC_CALL (HOOK ,IDX ) lsm_static_call_##HOOK##_##IDX + +/* +* Call the macro M for each LSM hook MAX_LSM_COUNT times. +*/ +#define LSM_LOOP_UNROLL (M , ...) \ +do { \ + UNROLL(MAX_LSM_COUNT, M, __VA_ARGS__) \ +} while (0) + +#define LSM_DEFINE_UNROLL (M , ...) UNROLL(MAX_LSM_COUNT, M, __VA_ARGS__)union security_list_options {#define LSM_HOOK (RET ,DEFAULT ,NAME , ...) RET (*NAME)(__VA_ARGS__);#include "lsm_hook_defs.h"#undef LSM_HOOK +void *lsm_callback; }; -struct security_hook_heads { -#define LSM_HOOK (RET ,DEFAULT ,NAME , ...) struct hlist_head NAME; -#include "lsm_hook_defs.h" +/* +* @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 } __randomize_layout;/* * 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 char *lsm; } __randomize_layout;
@@ -97,10 +148,12 @@ 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 ,CALLBACK ) \ + { \ + .scalls = static_calls_table.NAME, \ + .hook = { .NAME = CALLBACK } \ + } -extern struct security_hook_heads security_hook_heads;extern char *lsm_names; extern void security_add_hooks(struct security_hook_list *hooks, int count,
@@ -138,5 +191,6 @@ 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 +137 -71
392 lines not on this page · raw message
Signature · 1 line
2.42.0.582.g8ccd20d70d-goog
kernel test robot
build botRe: [PATCH v5 3/5] security: Replace indirect LSM hook calls with static calls
kernel test robot noticed the following build errors:
Hi KP,
kernel test robot noticed the following build errors:
[auto build test ERROR on bpf-next/master]
[also build test ERROR on bpf/master pcmoore-selinux/next linus/master v6.6-rc3 next-20230929] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/KP-Singh/kernel-Add-helper-macros-for-loop-unrolling/20230929-042610 base: https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git master patch link: https://lore.kernel.org/r/20230928202410.3765062-4-kpsingh%40kernel.org patch subject: [PATCH v5 3/5] security: Replace indirect LSM hook calls with static calls config: i386-randconfig-001-20230930 (https://download.01.org/0day-ci/archive/20230930/[email protected]/config) compiler: gcc-9 (Debian 9.3.0-22) 9.3.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20230930/[email protected]/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
All errors (new ones prefixed by >>):
>> security/security.c:139:1: error: Only string constants are supported as initializers for randomized structures with flexible arrays
139 | };
| ^vim +139 security/security.c
118
119 /*
120 * Initialise a table of static calls for each LSM hook.
121 * DEFINE_STATIC_CALL_NULL invocation above generates a key (STATIC_CALL_KEY)
122 * and a trampoline (STATIC_CALL_TRAMP) which are used to call
123 * __static_call_update when updating the static call.
124 */
125 struct lsm_static_calls_table static_calls_table __ro_after_init = {
126 #define INIT_LSM_STATIC_CALL(NUM, NAME) \
127 (struct lsm_static_call) { \
128 .key = &STATIC_CALL_KEY(LSM_STATIC_CALL(NAME, NUM)), \
129 .trampoline = LSM_HOOK_TRAMP(NAME, NUM), \
130 .active = &SECURITY_HOOK_ACTIVE_KEY(NAME, NUM), \
131 },
132 #define LSM_HOOK(RET, DEFAULT, NAME, ...) \
133 .NAME = { \
134 LSM_DEFINE_UNROLL(INIT_LSM_STATIC_CALL, NAME) \
135 },
136 #include <linux/lsm_hook_defs.h>
137 #undef LSM_HOOK
138 #undef INIT_LSM_STATIC_CALL
> 139 };
140
Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
Re: [PATCH v5 3/5] security: Replace indirect LSM hook calls with static calls
Quoted 23 lines · kernel test robot
Hi KP,
kernel test robot noticed the following build errors:
[auto build test ERROR on bpf-next/master]
[also build test ERROR on bpf/master pcmoore-selinux/next linus/master v6.6-rc3 next-20230929] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]url: https://github.com/intel-lab-lkp/linux/commits/KP-Singh/kernel-Add-helper-macros-for-loop-unrolling/20230929-042610 base: https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git master patch link: https://lore.kernel.org/r/20230928202410.3765062-4-kpsingh%40kernel.org patch subject: [PATCH v5 3/5] security: Replace indirect LSM hook calls with static calls config: i386-randconfig-001-20230930 (https://download.01.org/0day-ci/archive/20230930/[email protected]/config) compiler: gcc-9 (Debian 9.3.0-22) 9.3.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20230930/[email protected]/reproduce)If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
Quoted 5 lines
All errors (new ones prefixed by >>):
Quoted 1 line
Quoted 1 line
security/security.c:139:1: error: Only string constants are supported as initializers for randomized structures with flexible arrays139 | }; | ^
Uuh, where is there a flexible array here?
Quoted 23 lines
vim +139
security/security.c118 119 /* 120 * Initialise a table of static calls for each LSM hook. 121 * DEFINE_STATIC_CALL_NULL invocation above generates a key (STATIC_CALL_KEY) 122 * and a trampoline (STATIC_CALL_TRAMP) which are used to call 123 * __static_call_update when updating the static call. 124 */ 125 struct lsm_static_calls_table static_calls_table __ro_after_init = { 126 #define INIT_LSM_STATIC_CALL(NUM, NAME) \ 127 (struct lsm_static_call) { \ 128 .key = &STATIC_CALL_KEY(LSM_STATIC_CALL(NAME, NUM)), \ 129 .trampoline = LSM_HOOK_TRAMP(NAME, NUM), \ 130 .active = &SECURITY_HOOK_ACTIVE_KEY(NAME, NUM), \ 131 }, 132 #define LSM_HOOK(RET, DEFAULT, NAME, ...) \ 133 .NAME = { \ 134 LSM_DEFINE_UNROLL(INIT_LSM_STATIC_CALL, NAME) \ 135 }, 136 #include <linux/lsm_hook_defs.h> 137 #undef LSM_HOOK 138 #undef INIT_LSM_STATIC_CALL
Quoted 2 lines
Quoted 1 line
139 };
140
*confused*
-Kees
Signature · 1 line
Kees Cook
KP Singh
maintainerRe: [PATCH v5 3/5] security: Replace indirect LSM hook calls with static calls
Quoted 25 lines · Kees Cook
Quoted 23 lines · kernel test robot
Hi KP,
kernel test robot noticed the following build errors:
[auto build test ERROR on bpf-next/master]
[also build test ERROR on bpf/master pcmoore-selinux/next linus/master v6.6-rc3 next-20230929] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]url: https://github.com/intel-lab-lkp/linux/commits/KP-Singh/kernel-Add-helper-macros-for-loop-unrolling/20230929-042610 base: https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git master patch link: https://lore.kernel.org/r/20230928202410.3765062-4-kpsingh%40kernel.org patch subject: [PATCH v5 3/5] security: Replace indirect LSM hook calls with static calls config: i386-randconfig-001-20230930 (https://download.01.org/0day-ci/archive/20230930/[email protected]/config) compiler: gcc-9 (Debian 9.3.0-22) 9.3.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20230930/[email protected]/reproduce)If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
Quoted 7 lines
Quoted 5 lines
All errors (new ones prefixed by >>):
Quoted 1 line
Quoted 1 line
security/security.c:139:1: error: Only string constants are supported as initializers for randomized structures with flexible arrays139 | }; | ^Uuh, where is there a flexible array here?
I cannot seem to reproduce this with my gcc
wget https://download.01.org/0day-ci/archive/20230930/[email protected]/config
mkdir build_dir && cp config build_dir/.config
make -j 64 W=1 O=build_dir ARCH=i386 olddefconfig
make -j 64 W=1 O=build_dir ARCH=i386 SHELL=/bin/bash
history
on bpf-next/master builds the kernel for me:
Kernel: arch/x86/boot/bzImage is ready (#1)
The reproducer does not seem to come with a tool chain, and I even hacked the make.cross from another similar reply for gcc-9 and it still seems to work. Not sure what happened in the CI there.
Quoted 33 lines
Quoted 24 lines
vim +139
security/security.c118 119 /* 120 * Initialise a table of static calls for each LSM hook. 121 * DEFINE_STATIC_CALL_NULL invocation above generates a key (STATIC_CALL_KEY) 122 * and a trampoline (STATIC_CALL_TRAMP) which are used to call 123 * __static_call_update when updating the static call. 124 */ 125 struct lsm_static_calls_table static_calls_table __ro_after_init = { 126 #define INIT_LSM_STATIC_CALL(NUM, NAME) \ 127 (struct lsm_static_call) { \ 128 .key = &STATIC_CALL_KEY(LSM_STATIC_CALL(NAME, NUM)), \ 129 .trampoline = LSM_HOOK_TRAMP(NAME, NUM), \ 130 .active = &SECURITY_HOOK_ACTIVE_KEY(NAME, NUM), \ 131 }, 132 #define LSM_HOOK(RET, DEFAULT, NAME, ...) \ 133 .NAME = { \ 134 LSM_DEFINE_UNROLL(INIT_LSM_STATIC_CALL, NAME) \ 135 }, 136 #include <linux/lsm_hook_defs.h> 137 #undef LSM_HOOK 138 #undef INIT_LSM_STATIC_CALLQuoted 2 lines
Quoted 1 line
139 };
140
*confused*
-Kees
--
Kees Cook
KP Singh
maintainerPATCH v5 4/5 bpf: Only enable BPF LSM hooks when an LSM program is attached
In reply to KP SinghBPF LSM hooks have side-effects (even when a default value is returned), as some hooks end up behaving differently due to the very presence of the hook.
The static keys guarding the BPF LSM hooks are disabled by default and enabled only when a BPF program is attached implementing the hook logic. This avoids the issue of the side-effects and also the minor overhead associated with the empty callback.
security_file_ioctl:
0xffffffff818f0e30 <+0>: endbr64
0xffffffff818f0e34 <+4>: nopl 0x0(%rax,%rax,1)
0xffffffff818f0e39 <+9>: push %rbp
0xffffffff818f0e3a <+10>: push %r14 0xffffffff818f0e3c <+12>: push %rbx 0xffffffff818f0e3d <+13>: mov %rdx,%rbx 0xffffffff818f0e40 <+16>: mov %esi,%ebp 0xffffffff818f0e42 <+18>: mov %rdi,%r14 0xffffffff818f0e45 <+21>: jmp 0xffffffff818f0e57 <security_file_ioctl+39>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Static key enabled for SELinux
0xffffffff818f0e47 <+23>: xchg %ax,%ax
^^^^^^^^^^^^^^
Static key disabled for BPF. This gets patched when a BPF LSM program is attached
0xffffffff818f0e49 <+25>: xor %eax,%eax 0xffffffff818f0e4b <+27>: xchg %ax,%ax 0xffffffff818f0e4d <+29>: pop %rbx 0xffffffff818f0e4e <+30>: pop %r14 0xffffffff818f0e50 <+32>: pop %rbp 0xffffffff818f0e51 <+33>: cs jmp 0xffffffff82c00000 <__x86_return_thunk> 0xffffffff818f0e57 <+39>: endbr64 0xffffffff818f0e5b <+43>: mov %r14,%rdi 0xffffffff818f0e5e <+46>: mov %ebp,%esi 0xffffffff818f0e60 <+48>: mov %rbx,%rdx 0xffffffff818f0e63 <+51>: call 0xffffffff819033c0 <selinux_file_ioctl> 0xffffffff818f0e68 <+56>: test %eax,%eax 0xffffffff818f0e6a <+58>: jne 0xffffffff818f0e4d <security_file_ioctl+29> 0xffffffff818f0e6c <+60>: jmp 0xffffffff818f0e47 <security_file_ioctl+23> 0xffffffff818f0e6e <+62>: endbr64 0xffffffff818f0e72 <+66>: mov %r14,%rdi 0xffffffff818f0e75 <+69>: mov %ebp,%esi 0xffffffff818f0e77 <+71>: mov %rbx,%rdx 0xffffffff818f0e7a <+74>: call 0xffffffff8141e3b0 <bpf_lsm_file_ioctl> 0xffffffff818f0e7f <+79>: test %eax,%eax 0xffffffff818f0e81 <+81>: jne 0xffffffff818f0e4d <security_file_ioctl+29> 0xffffffff818f0e83 <+83>: jmp 0xffffffff818f0e49 <security_file_ioctl+25> 0xffffffff818f0e85 <+85>: endbr64 0xffffffff818f0e89 <+89>: mov %r14,%rdi 0xffffffff818f0e8c <+92>: mov %ebp,%esi 0xffffffff818f0e8e <+94>: mov %rbx,%rdx 0xffffffff818f0e91 <+97>: pop %rbx 0xffffffff818f0e92 <+98>: pop %r14
0xffffffff818f0e94 <+100>: pop %rbp
0xffffffff818f0e95 <+101>: ret
--- include/linux/bpf.h | 1 + include/linux/bpf_lsm.h | 5 +++++ include/linux/lsm_hooks.h | 13 ++++++++++++- kernel/bpf/trampoline.c | 29 +++++++++++++++++++++++++++-- security/bpf/hooks.c | 25 ++++++++++++++++++++++++- security/security.c | 3 ++- 6 files changed, 71 insertions(+), 5 deletions(-)
include/linux/bpf.h +1 -0
@@ -1159,6 +1159,7 @@ struct bpf_attach_target_info {struct module *tgt_mod;const char *tgt_name;const struct btf_type *tgt_type; +bool is_lsm_target; };#define BPF_DISPATCHER_MAX 48/* Fits in 2048B */
include/linux/bpf_lsm.h +5 -0
@@ -29,6 +29,7 @@ int bpf_lsm_verify_prog(struct bpf_verifier_log *vlog,bool bpf_lsm_is_sleepable_hook (u32 btf_id);bool bpf_lsm_is_trusted (const struct bpf_prog *prog); +void bpf_lsm_toggle_hook (void *addr,bool value);static inline struct bpf_storage_blob *bpf_inode (const struct inode *inode)
@@ -78,6 +79,10 @@ static inline void bpf_lsm_find_cgroup_shim(const struct bpf_prog *prog, { } +static inline void bpf_lsm_toggle_hook (void *addr,bool value) +{ +} +#endif /* CONFIG_BPF_LSM */ #endif /* _LINUX_BPF_LSM_H */
include/linux/lsm_hooks.h +12 -1
@@ -97,11 +97,14 @@ struct lsm_static_calls_table {* @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. +* @default_state: The state of the LSM hook when initialized. If set to false, +* the static key guarding the hook will be set to disabled. */ struct security_hook_list {struct lsm_static_call *scalls;union security_list_options hook;const char *lsm; +bool default_state; } __randomize_layout;/*
@@ -151,7 +154,15 @@ static inline struct xattr *lsm_get_xattr_slot(struct xattr *xattrs,#define LSM_HOOK_INIT (NAME ,CALLBACK ) \ { \ .scalls = static_calls_table.NAME, \ - .hook = { .NAME = CALLBACK } \ + .hook = { .NAME = CALLBACK }, \ + .default_state = true \ + } + +#define LSM_HOOK_INIT_DISABLED (NAME ,CALLBACK ) \ + { \ + .scalls = static_calls_table.NAME, \ + .hook = { .NAME = CALLBACK }, \ + .default_state = false \ }extern char *lsm_names;
kernel/bpf/trampoline.c +27 -2
@@ -13,6 +13,7 @@#include <linux/bpf_verifier.h> #include <linux/bpf_lsm.h> #include <linux/delay.h> +#include <linux/bpf_lsm.h> /* dummy _ops. The verifier will operate on target program's ops. */ const struct bpf_verifier_ops bpf_extension_verifier_ops = {
@@ -514,7 +515,7 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr {enum bpf_tramp_prog_type kind;struct bpf_tramp_link *link_exiting; -int err =0 ; +int err =0 , num_lsm_progs =0 ;int cnt =0 , i; kind =bpf_attach_type_to_tramp (link->link.prog);
@@ -545,8 +546,14 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trcontinue ;/* prog already linked */ return -EBUSY ; + +if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) + num_lsm_progs++; } +if (!num_lsm_progs && link->link.prog->type ==BPF_PROG_TYPE_LSM ) +bpf_lsm_toggle_hook (tr->func.addr, true); +hlist_add_head (&link->tramp_hlist, &tr->progs_hlist[kind]); tr->progs_cnt[kind]++; err =bpf_trampoline_update (tr, true/* lock_direct_mutex */ );
@@ -569,8 +576,10 @@ int bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trampolinestatic int __bpf_trampoline_unlink_prog (struct bpf_tramp_link *link,struct bpf_trampoline *tr) { +struct bpf_tramp_link *link_exiting;enum bpf_tramp_prog_type kind; -int err; +bool lsm_link_found = false; +int err, num_lsm_progs =0 ; kind =bpf_attach_type_to_tramp (link->link.prog);if (kind ==BPF_TRAMP_REPLACE ) {
@@ -580,8 +589,24 @@ static int __bpf_trampoline_unlink_prog(struct bpf_tramp_link *link, struct bpf_ tr->extension_prog =NULL ;return err; } + +if (link->link.prog->type ==BPF_PROG_TYPE_LSM ) { +hlist_for_each_entry (link_exiting, &tr->progs_hlist[kind], + tramp_hlist) { +if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) + num_lsm_progs++; + +if (link_exiting->link.prog == link->link.prog) + lsm_link_found = true; + } + } +hlist_del_init (&link->tramp_hlist); tr->progs_cnt[kind]--; + +if (lsm_link_found && num_lsm_progs ==1 ) +bpf_lsm_toggle_hook (tr->func.addr, false); +return bpf_trampoline_update (tr, true/* lock_direct_mutex */ ); }
security/bpf/hooks.c +24 -1
@@ -8,7 +8,7 @@static struct security_hook_list bpf_lsm_hooks[]__ro_after_init = { #define LSM_HOOK(RET, DEFAULT, NAME, ...) \ - LSM_HOOK_INIT(NAME, bpf_lsm_##NAME), + LSM_HOOK_INIT_DISABLED(NAME, bpf_lsm_##NAME), #include <linux/lsm_hook_defs.h> #undef LSM_HOOK LSM_HOOK_INIT(inode_free_security, bpf_inode_storage_free),
@@ -32,3 +32,26 @@ DEFINE_LSM(bpf) = { .init = bpf_lsm_init, .blobs = &bpf_lsm_blob_sizes }; + +void bpf_lsm_toggle_hook (void *addr,bool value) +{ +struct lsm_static_call *scalls; +struct security_hook_list *h; +int i, j; + +for (i =0 ; i <ARRAY_SIZE (bpf_lsm_hooks); i++) { + h = &bpf_lsm_hooks[i]; + scalls = h->scalls; +if (h->hook.lsm_callback == addr) +continue ; + +for (j =0 ; j <MAX_LSM_COUNT ; j++) { +if (scalls[j].hl != h) +continue ; +if (value) +static_branch_enable (scalls[j].active); +else +static_branch_disable (scalls[j].active); + } + } +}
security/security.c +2 -1
@@ -382,7 +382,8 @@ static void __init lsm_static_call_init(struct security_hook_list *hl)__static_call_update (scall->key, scall->trampoline, hl->hook.lsm_callback); scall->hl = hl; -static_branch_enable (scall->active); +if (hl->default_state) +static_branch_enable (scall->active);return ; } scall++;
Signature · 1 line
2.42.0.582.g8ccd20d70d-goog
Re: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
On Thu, Sep 28, 2023 at 10:24:09PM +0200, KP Singh wrote:
SNIP
Quoted 22 lines
kernel/bpf/trampoline.c +2
-1@@ -13,6 +13,7 @@#include <linux/bpf_verifier.h> #include <linux/bpf_lsm.h> #include <linux/delay.h> +#include <linux/bpf_lsm.h> /* dummy _ops. The verifier will operate on target program's ops. */ const struct bpf_verifier_ops bpf_extension_verifier_ops = {@@ -514,7 +515,7 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr {enum bpf_tramp_prog_type kind;struct bpf_tramp_link *link_exiting; -int err =0 ; +int err =0 , num_lsm_progs =0 ;int cnt =0 , i; kind =bpf_attach_type_to_tramp (link->link.prog);
kernel/bpf/trampoline.c static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr · line 551
Quoted 6 lines
kernel/bpf/trampoline.c +3
-0@@ -545,8 +546,14 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trcontinue ;/* prog already linked */ return -EBUSY ; + +if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) + num_lsm_progs++;
this looks wrong, it's never reached.. seems like we should add separate hlist_for_each_entry loop over trampoline's links for this check/init of num_lsm_progs ?
jirka
Quoted 103 lines
kernel/bpf/trampoline.c +22
-1} +if (!num_lsm_progs && link->link.prog->type ==BPF_PROG_TYPE_LSM ) +bpf_lsm_toggle_hook (tr->func.addr, true); +hlist_add_head (&link->tramp_hlist, &tr->progs_hlist[kind]); tr->progs_cnt[kind]++; err =bpf_trampoline_update (tr, true/* lock_direct_mutex */ );@@ -569,8 +576,10 @@ int bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trampolinestatic int __bpf_trampoline_unlink_prog (struct bpf_tramp_link *link,struct bpf_trampoline *tr) { +struct bpf_tramp_link *link_exiting;enum bpf_tramp_prog_type kind; -int err; +bool lsm_link_found = false; +int err, num_lsm_progs =0 ; kind =bpf_attach_type_to_tramp (link->link.prog);if (kind ==BPF_TRAMP_REPLACE ) {@@ -580,8 +589,24 @@ static int __bpf_trampoline_unlink_prog(struct bpf_tramp_link *link, struct bpf_ tr->extension_prog =NULL ;return err; } + +if (link->link.prog->type ==BPF_PROG_TYPE_LSM ) { +hlist_for_each_entry (link_exiting, &tr->progs_hlist[kind], + tramp_hlist) { +if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) + num_lsm_progs++; + +if (link_exiting->link.prog == link->link.prog) + lsm_link_found = true; + } + } +hlist_del_init (&link->tramp_hlist); tr->progs_cnt[kind]--; + +if (lsm_link_found && num_lsm_progs ==1 ) +bpf_lsm_toggle_hook (tr->func.addr, false); +return bpf_trampoline_update (tr, true/* lock_direct_mutex */ ); }security/bpf/hooks.c +24
-1@@ -8,7 +8,7 @@static struct security_hook_list bpf_lsm_hooks[]__ro_after_init = { #define LSM_HOOK(RET, DEFAULT, NAME, ...) \ - LSM_HOOK_INIT(NAME, bpf_lsm_##NAME), + LSM_HOOK_INIT_DISABLED(NAME, bpf_lsm_##NAME), #include <linux/lsm_hook_defs.h> #undef LSM_HOOK LSM_HOOK_INIT(inode_free_security, bpf_inode_storage_free),@@ -32,3 +32,26 @@ DEFINE_LSM(bpf) = { .init = bpf_lsm_init, .blobs = &bpf_lsm_blob_sizes }; + +void bpf_lsm_toggle_hook (void *addr,bool value) +{ +struct lsm_static_call *scalls; +struct security_hook_list *h; +int i, j; + +for (i =0 ; i <ARRAY_SIZE (bpf_lsm_hooks); i++) { + h = &bpf_lsm_hooks[i]; + scalls = h->scalls; +if (h->hook.lsm_callback == addr) +continue ; + +for (j =0 ; j <MAX_LSM_COUNT ; j++) { +if (scalls[j].hl != h) +continue ; +if (value) +static_branch_enable (scalls[j].active); +else +static_branch_disable (scalls[j].active); + } + } +}security/security.c +2
-1@@ -382,7 +382,8 @@ static void __init lsm_static_call_init(struct security_hook_list *hl)__static_call_update (scall->key, scall->trampoline, hl->hook.lsm_callback); scall->hl = hl; -static_branch_enable (scall->active); +if (hl->default_state) +static_branch_enable (scall->active);return ; } scall++;Signature · 1 line
2.42.0.582.g8ccd20d70d-goog
KP Singh
maintainerRe: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
Quoted 38 lines · Jiri Olsa
On Thu, Sep 28, 2023 at 10:24:09PM +0200, KP Singh wrote:
SNIP
Quoted 22 lines
kernel/bpf/trampoline.c +2
-1@@ -13,6 +13,7 @@#include <linux/bpf_verifier.h> #include <linux/bpf_lsm.h> #include <linux/delay.h> +#include <linux/bpf_lsm.h> /* dummy _ops. The verifier will operate on target program's ops. */ const struct bpf_verifier_ops bpf_extension_verifier_ops = {@@ -514,7 +515,7 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr {enum bpf_tramp_prog_type kind;struct bpf_tramp_link *link_exiting; -int err =0 ; +int err =0 , num_lsm_progs =0 ;int cnt =0 , i; kind =bpf_attach_type_to_tramp (link->link.prog);kernel/bpf/trampoline.c static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr · line 551
Quoted 6 lines
kernel/bpf/trampoline.c +3
-0@@ -545,8 +546,14 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trcontinue ;/* prog already linked */ return -EBUSY ; + +if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) + num_lsm_progs++;this looks wrong, it's never reached.. seems like we should add separate
hlist_for_each_entryloop over trampoline's links for this check/init ofnum_lsm_progs?
Quoted 1 line
jirka
Good catch, I missed this during my rebase, after
https://lore.kernel.org/bpf/[email protected]/
this condition is basically never reached. I will do a general loop over to count LSM programs and toggle the hook to true (and same for unlink).
- KP
[...]
KP Singh
maintainerRe: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
Quoted 17 lines · KP Singh
Quoted 15 lines · Jiri Olsa
On Thu, Sep 28, 2023 at 10:24:09PM +0200, KP Singh wrote:
SNIP
Quoted 10 lines
kernel/bpf/trampoline.c +1
-0@@ -13,6 +13,7 @@#include <linux/bpf_verifier.h> #include <linux/bpf_lsm.h> #include <linux/delay.h> +#include <linux/bpf_lsm.h>
Quoted 6 lines
Quoted 6 lines
kernel/bpf/trampoline.c static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr · line 517
Quoted 6 lines
kernel/bpf/trampoline.c +0
-0/* dummy _ops. The verifier will operate on target program's ops. */ const struct bpf_verifier_ops bpf_extension_verifier_ops = {@@ -514,7 +515,7 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr {enum bpf_tramp_prog_type kind;struct bpf_tramp_link *link_exiting;
I think this is a typo here. It should be existing, no?
Quoted 28 lines
Quoted 17 lines
kernel/bpf/trampoline.c static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr · line 551
Quoted 12 lines
kernel/bpf/trampoline.c +4
-1-int err =0 ; +int err =0 , num_lsm_progs =0 ;int cnt =0 , i; kind =bpf_attach_type_to_tramp (link->link.prog);@@ -545,8 +546,14 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trcontinue ;/* prog already linked */ return -EBUSY ; + +if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) + num_lsm_progs++;this looks wrong, it's never reached.. seems like we should add separate
hlist_for_each_entryloop over trampoline's links for this check/init ofnum_lsm_progs?Quoted 1 line
jirka
Good catch, I missed this during my rebase, after
https://lore.kernel.org/bpf/[email protected]/
this condition is basically never reached. I will do a general loop over to count LSM programs and toggle the hook to true (and same for unlink).- KP
[...]
Re: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
Quoted 18 lines · KP Singh
Quoted 17 lines · KP Singh
Quoted 15 lines · Jiri Olsa
On Thu, Sep 28, 2023 at 10:24:09PM +0200, KP Singh wrote:
SNIP
Quoted 10 lines
kernel/bpf/trampoline.c +1
-0@@ -13,6 +13,7 @@#include <linux/bpf_verifier.h> #include <linux/bpf_lsm.h> #include <linux/delay.h> +#include <linux/bpf_lsm.h>
Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
kernel/bpf/trampoline.c static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr · line 517
Quoted 6 lines
kernel/bpf/trampoline.c +0
-0/* dummy _ops. The verifier will operate on target program's ops. */ const struct bpf_verifier_ops bpf_extension_verifier_ops = {@@ -514,7 +515,7 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr {enum bpf_tramp_prog_type kind;struct bpf_tramp_link *link_exiting;I think this is a typo here. It should be existing, no?
yes, I was wondering about that as well ;-)
jirka
Quoted 29 lines
Quoted 29 lines
Quoted 18 lines
kernel/bpf/trampoline.c static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr · line 551
Quoted 13 lines
kernel/bpf/trampoline.c +4
-1-int err =0 ; +int err =0 , num_lsm_progs =0 ;int cnt =0 , i; kind =bpf_attach_type_to_tramp (link->link.prog);@@ -545,8 +546,14 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trcontinue ;/* prog already linked */ return -EBUSY ; + +if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) + num_lsm_progs++;this looks wrong, it's never reached.. seems like we should add separate
hlist_for_each_entryloop over trampoline's links for this check/init ofnum_lsm_progs?Quoted 1 line
jirka
Good catch, I missed this during my rebase, after
https://lore.kernel.org/bpf/[email protected]/
this condition is basically never reached. I will do a general loop over to count LSM programs and toggle the hook to true (and same for unlink).- KP
[...]
KP Singh
maintainerRe: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
Quoted 29 lines · Jiri Olsa
Quoted 18 lines · KP Singh
Quoted 17 lines · KP Singh
Quoted 15 lines · Jiri Olsa
On Thu, Sep 28, 2023 at 10:24:09PM +0200, KP Singh wrote:
SNIP
Quoted 10 lines
kernel/bpf/trampoline.c +1
-0@@ -13,6 +13,7 @@#include <linux/bpf_verifier.h> #include <linux/bpf_lsm.h> #include <linux/delay.h> +#include <linux/bpf_lsm.h> Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
kernel/bpf/trampoline.c static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr · line 517
Quoted 6 lines
kernel/bpf/trampoline.c +0
-0/* dummy _ops. The verifier will operate on target program's ops. */ const struct bpf_verifier_ops bpf_extension_verifier_ops = {@@ -514,7 +515,7 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr {enum bpf_tramp_prog_type kind;struct bpf_tramp_link *link_exiting;I think this is a typo here. It should be existing, no?
Quoted 29 lines
yes, I was wondering about that as well ;-)
jirka
Quoted 25 lines
Quoted 25 lines
Quoted 18 lines
kernel/bpf/trampoline.c static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_tr · line 551
Quoted 13 lines
kernel/bpf/trampoline.c +4
-1-int err =0 ; +int err =0 , num_lsm_progs =0 ;int cnt =0 , i; kind =bpf_attach_type_to_tramp (link->link.prog);@@ -545,8 +546,14 @@ static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link, struct bpf_trcontinue ;/* prog already linked */ return -EBUSY ; + +if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) + num_lsm_progs++;this looks wrong, it's never reached.. seems like we should add separate
hlist_for_each_entryloop over trampoline's links for this check/init ofnum_lsm_progs?Quoted 1 line
jirka
Good catch, I missed this during my rebase, after
https://lore.kernel.org/bpf/[email protected]/
this condition is basically never reached. I will do a general loop over to count LSM programs and toggle the hook to true (and same for unlink).
So, there is something that is unclear about this code, i.e. what happens when there is an error from bpf_trampoline_update fails and the link and unlink seem to have different expectations:
* link seems to go back to the linked list and removes the trampoline and restores the refcount:
[...]
err = bpf_trampoline_update(tr, true /* lock_direct_mutex */);
if (err) {
hlist_del_init(&link->tramp_hlist);
tr->progs_cnt[kind]--; }
return err;
}
* unlink does restore the side effect (i.e. it does not put the removed trampoline back and increments the refcount).
hlist_del_init(&link->tramp_hlist); tr->progs_cnt[kind]--;
return bpf_trampoline_update(tr, true /* lock_direct_mutex */);
However, I think I will make it simpler and enforce the invariant that if an LSM program is attached, the hook is enabled and vice versa. How about:
kernel/bpf/trampoline.c +0 -0
@@ -511,11 +511,30 @@ static enum bpf_tramp_prog_type
bpf_attach_type_to_tramp(struct bpf_prog *prog)
}
}
+static void bpf_trampoline_toggle_lsm(struct bpf_trampoline *tr,
+ enum bpf_tramp_prog_type kind)
+{
+ struct bpf_tramp_link *link;
+ volatile bool found = false;
+
+ /* Loop through the links and if any LSM program is attached, ensure
+ * that the hook is enabled.
+ */
+ hlist_for_each_entry(link, &tr->progs_hlist[kind], tramp_hlist) {
+ if (link->link.prog->type == BPF_PROG_TYPE_LSM) {
+ found = true;
+ break;
+ }
+ }
+
+ bpf_lsm_toggle_hook(tr->func.addr, found);
+}
+
static int __bpf_trampoline_link_prog(struct bpf_tramp_link *link,
struct bpf_trampoline *tr)
{
enum bpf_tramp_prog_type kind;
struct bpf_tramp_link *link_exiting;
- int err = 0, num_lsm_progs = 0;
+ int err = 0;
int cnt = 0, i;
kind = bpf_attach_type_to_tramp(link->link.prog);
@@ -547,15 +566,14 @@ static int __bpf_trampoline_link_prog(struct
bpf_tramp_link *link, struct bpf_tr
/* prog already linked */
return -EBUSY;- if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM)
- num_lsm_progs++;
}- if (!num_lsm_progs && link->link.prog->type == BPF_PROG_TYPE_LSM)
- bpf_lsm_toggle_hook(tr->func.addr, true);
-
hlist_add_head(&link->tramp_hlist, &tr->progs_hlist[kind]);
tr->progs_cnt[kind]++;+
+ if (link->link.prog->type == BPF_PROG_TYPE_LSM)
+ bpf_trampoline_toggle_lsm(tr, kind);
+
err = bpf_trampoline_update(tr, true /* lock_direct_mutex */);
if (err) {
hlist_del_init(&link->tramp_hlist);
@@ -578,7 +596,6 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link *link, struct bpf_
{
struct bpf_tramp_link *link_exiting;
enum bpf_tramp_prog_type kind;
- bool lsm_link_found = false;
int err, num_lsm_progs = 0;
kind = bpf_attach_type_to_tramp(link->link.prog);
@@ -595,18 +612,14 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link *link, struct bpf_
tramp_hlist) {
if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM)
num_lsm_progs++;
-
- if (link_exiting->link.prog == link->link.prog)
- lsm_link_found = true;
}
}
hlist_del_init(&link->tramp_hlist);
tr->progs_cnt[kind]--;- if (lsm_link_found && num_lsm_progs == 1)
- bpf_lsm_toggle_hook(tr->func.addr, false);
-
+ if (link->link.prog->type == BPF_PROG_TYPE_LSM)
+ bpf_trampoline_toggle_lsm(tr, kind);
return bpf_trampoline_update(tr, true /* lock_direct_mutex */);}
- KP
Re: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
On Thu, Oct 05, 2023 at 06:07:28PM +0200, KP Singh wrote:
SNIP
Quoted 45 lines
kernel/bpf/trampoline.c +20
-1@@ -511,11 +511,30 @@ static enum bpf_tramp_prog_type bpf_attach_type_to_tramp(struct bpf_prog *prog) } } +static void bpf_trampoline_toggle_lsm (struct bpf_trampoline *tr, +enum bpf_tramp_prog_type kind) +{ +struct bpf_tramp_link *link; +volatile bool found = false; + +/* Loop through the links and if any LSM program is attached, ensure +* that the hook is enabled. +*/ +hlist_for_each_entry (link, &tr->progs_hlist[kind], tramp_hlist) { +if (link->link.prog->type ==BPF_PROG_TYPE_LSM ) { + found = true; +break ; + } + } + +bpf_lsm_toggle_hook (tr->func.addr, found); +} +static int __bpf_trampoline_link_prog (struct bpf_tramp_link *link,struct bpf_trampoline *tr) {enum bpf_tramp_prog_type kind;struct bpf_tramp_link *link_exiting; -int err =0 , num_lsm_progs =0 ; +int err =0 ;int cnt =0 , i;kind = bpf_attach_type_to_tramp(link->link.prog);kernel/bpf/trampoline.c +0
-0@@ -547,15 +566,14 @@ static int __bpf_trampoline_link_prog(struct
bpf_tramp_link*link, structbpf_tr/* prog already linked */ return -EBUSY;kernel/bpf/trampoline.c +0
-2-if (link_exiting->link.prog->type ==BPF_PROG_TYPE_LSM ) - num_lsm_progs++;
kernel/bpf/trampoline.c
Quoted 10 lines
kernel/bpf/trampoline.c +3
-3} -if (!num_lsm_progs && link->link.prog->type ==BPF_PROG_TYPE_LSM ) -bpf_lsm_toggle_hook (tr->func.addr, true); -hlist_add_head (&link->tramp_hlist, &tr->progs_hlist[kind]); tr->progs_cnt[kind]++; + +if (link->link.prog->type ==BPF_PROG_TYPE_LSM ) +bpf_trampoline_toggle_lsm (tr, kind);
how about keeping BPF_PROG_TYPE_LSM progs type count of attached programs in bpf_trampoline and toggle lsm on first coming in and last going out?
also the trampoline attach is actually made in bpf_trampoline_update, so I wonder it'd make more sense to put it in there, but it's already complicated, so it actually might be easier in here
jirka
Quoted 37 lines
kernel/bpf/trampoline.c +1
-0+ err =bpf_trampoline_update (tr, true/* lock_direct_mutex */ );if (err) {hlist_del_init (&link->tramp_hlist);@@ -578,7 +596,6 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link*link, structbpf_{ struct bpf_tramp_link *link_exiting; enum bpf_tramp_prog_type kind;kernel/bpf/trampoline.c +0
-1-bool lsm_link_found = false;int err, num_lsm_progs =0 ; kind =bpf_attach_type_to_tramp (link->link.prog);@@ -595,18 +612,14 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link*link, structbpf_tramp_hlist) { if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM) num_lsm_progs++;kernel/bpf/trampoline.c +2
-7- -if (link_exiting->link.prog == link->link.prog) - lsm_link_found = true; } }hlist_del_init (&link->tramp_hlist); tr->progs_cnt[kind]--; -if (lsm_link_found && num_lsm_progs ==1 ) -bpf_lsm_toggle_hook (tr->func.addr, false); - +if (link->link.prog->type ==BPF_PROG_TYPE_LSM ) +bpf_trampoline_toggle_lsm (tr, kind);return bpf_trampoline_update (tr, true/* lock_direct_mutex */ ); } -KP
Re: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
On Fri, Oct 06, 2023 at 09:27:57AM +0200, Jiri Olsa wrote:
SNIP
Quoted 19 lines
Quoted 19 lines
static int __bpf_trampoline_link_prog(struct
bpf_tramp_link*link,
structbpf_trampoline*tr){ enum bpf_tramp_prog_type kind; struct bpf_tramp_link *link_exiting;(quoted diff) +1
-1- int err = 0, num_lsm_progs = 0; + int err = 0; int cnt = 0, i; kind = bpf_attach_type_to_tramp(link->link.prog);@@ -547,15 +566,14 @@ static int __bpf_trampoline_link_prog(struct
bpf_tramp_link*link, structbpf_tr/* prog already linked */ return -EBUSY;(quoted diff) +0
-2- if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM) - num_lsm_progs++; }
Quoted 11 lines
Review
Quoted 8 lines
(quoted diff) +3
-3- if (!num_lsm_progs && link->link.prog->type == BPF_PROG_TYPE_LSM) - bpf_lsm_toggle_hook(tr->func.addr, true); - hlist_add_head(&link->tramp_hlist, &tr->progs_hlist[kind]); tr->progs_cnt[kind]++; + + if (link->link.prog->type == BPF_PROG_TYPE_LSM) + bpf_trampoline_toggle_lsm(tr, kind);how about keeping
BPF_PROG_TYPE_LSMprogs type count of attached programs inbpf_trampolineand toggle lsm on first coming in and last going out?
hm we actually allow other tracing program types to attach to bpf_lsm_* functions, so I wonder we should toggle the lsm hook for each program type (for bpf_lsm_* trampolines) because they'd expect the hook is called
but I'm not sure it's a valid use case to have like normal fentry program attached to bpf_lsm_XXX function
jirka
Quoted 44 lines
also the trampoline attach is actually made in
bpf_trampoline_update, so I wonder it'd make more sense to put it in there, but it's already complicated, so it actually might be easier in herejirka
Quoted 37 lines
(quoted diff) +1
-0+ err = bpf_trampoline_update(tr, true /* lock_direct_mutex */); if (err) { hlist_del_init(&link->tramp_hlist);@@ -578,7 +596,6 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link*link, structbpf_{ struct bpf_tramp_link *link_exiting; enum bpf_tramp_prog_type kind;(quoted diff) +0
-1- bool lsm_link_found = false; int err, num_lsm_progs = 0; kind = bpf_attach_type_to_tramp(link->link.prog);@@ -595,18 +612,14 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link*link, structbpf_tramp_hlist) { if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM) num_lsm_progs++;(quoted diff) +2
-7- - if (link_exiting->link.prog == link->link.prog) - lsm_link_found = true; } } hlist_del_init(&link->tramp_hlist); tr->progs_cnt[kind]--; - if (lsm_link_found && num_lsm_progs == 1) - bpf_lsm_toggle_hook(tr->func.addr, false); - + if (link->link.prog->type == BPF_PROG_TYPE_LSM) + bpf_trampoline_toggle_lsm(tr, kind); return bpf_trampoline_update(tr, true /* lock_direct_mutex */); } - KP
KP Singh
maintainerRe: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
Quoted 36 lines · Jiri Olsa
On Fri, Oct 06, 2023 at 09:27:57AM +0200, Jiri Olsa wrote:
SNIP
Quoted 19 lines
Quoted 19 lines
static int __bpf_trampoline_link_prog(struct
bpf_tramp_link*link,
structbpf_trampoline*tr){ enum bpf_tramp_prog_type kind; struct bpf_tramp_link *link_exiting;(quoted diff) +1
-1- int err = 0, num_lsm_progs = 0; + int err = 0; int cnt = 0, i; kind = bpf_attach_type_to_tramp(link->link.prog);@@ -547,15 +566,14 @@ static int __bpf_trampoline_link_prog(struct
bpf_tramp_link*link, structbpf_tr/* prog already linked */ return -EBUSY;(quoted diff) +0
-2- if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM) - num_lsm_progs++; }Quoted 11 lines
Review
Quoted 8 lines
(quoted diff) +3
-3- if (!num_lsm_progs && link->link.prog->type == BPF_PROG_TYPE_LSM) - bpf_lsm_toggle_hook(tr->func.addr, true); - hlist_add_head(&link->tramp_hlist, &tr->progs_hlist[kind]); tr->progs_cnt[kind]++; + + if (link->link.prog->type == BPF_PROG_TYPE_LSM) + bpf_trampoline_toggle_lsm(tr, kind);how about keeping
BPF_PROG_TYPE_LSMprogs type count of attached programs inbpf_trampolineand toggle lsm on first coming in and last going out?
Quoted 3 lines
hm we actually allow other tracing program types to attach to bpf_lsm_* functions, so I wonder we should toggle the lsm hook for each program type (for bpf_lsm_* trampolines) because they'd expect the hook is called
Tracing is about tracing, attaching a tracing program to bpf_lsm_ that changes the actual trace here is not something I would recommend. Infact, I have used tracing programs to figure out whether bpf_lsm_* is being called to debug stuff. Tracing users can always attach to security_* if they like.
- KP
Quoted 51 lines
but I'm not sure it's a valid use case to have like normal fentry program attached to
bpf_lsm_XXXfunctionjirka
Quoted 44 lines
also the trampoline attach is actually made in
bpf_trampoline_update, so I wonder it'd make more sense to put it in there, but it's already complicated, so it actually might be easier in herejirka
Quoted 37 lines
(quoted diff) +1
-0+ err = bpf_trampoline_update(tr, true /* lock_direct_mutex */); if (err) { hlist_del_init(&link->tramp_hlist);@@ -578,7 +596,6 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link*link, structbpf_{ struct bpf_tramp_link *link_exiting; enum bpf_tramp_prog_type kind;(quoted diff) +0
-1- bool lsm_link_found = false; int err, num_lsm_progs = 0; kind = bpf_attach_type_to_tramp(link->link.prog);@@ -595,18 +612,14 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link*link, structbpf_tramp_hlist) { if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM) num_lsm_progs++;(quoted diff) +2
-7- - if (link_exiting->link.prog == link->link.prog) - lsm_link_found = true; } } hlist_del_init(&link->tramp_hlist); tr->progs_cnt[kind]--; - if (lsm_link_found && num_lsm_progs == 1) - bpf_lsm_toggle_hook(tr->func.addr, false); - + if (link->link.prog->type == BPF_PROG_TYPE_LSM) + bpf_trampoline_toggle_lsm(tr, kind); return bpf_trampoline_update(tr, true /* lock_direct_mutex */); } - KP
KP Singh
maintainerRe: [PATCH v5 4/5] bpf: Only enable BPF LSM hooks when an LSM program is attached
Quoted 42 lines · KP Singh
Quoted 36 lines · Jiri Olsa
On Fri, Oct 06, 2023 at 09:27:57AM +0200, Jiri Olsa wrote:
SNIP
Quoted 19 lines
Quoted 19 lines
static int __bpf_trampoline_link_prog(struct
bpf_tramp_link*link,
structbpf_trampoline*tr){ enum bpf_tramp_prog_type kind; struct bpf_tramp_link *link_exiting;(quoted diff) +1
-1- int err = 0, num_lsm_progs = 0; + int err = 0; int cnt = 0, i; kind = bpf_attach_type_to_tramp(link->link.prog);@@ -547,15 +566,14 @@ static int __bpf_trampoline_link_prog(struct
bpf_tramp_link*link, structbpf_tr/* prog already linked */ return -EBUSY;(quoted diff) +0
-2- if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM) - num_lsm_progs++; }Quoted 11 lines
Review
Quoted 8 lines
(quoted diff) +3
-3- if (!num_lsm_progs && link->link.prog->type == BPF_PROG_TYPE_LSM) - bpf_lsm_toggle_hook(tr->func.addr, true); - hlist_add_head(&link->tramp_hlist, &tr->progs_hlist[kind]); tr->progs_cnt[kind]++; + + if (link->link.prog->type == BPF_PROG_TYPE_LSM) + bpf_trampoline_toggle_lsm(tr, kind);how about keeping
BPF_PROG_TYPE_LSMprogs type count of attached programs inbpf_trampolineand toggle lsm on first coming in and last going out?Quoted 3 lines
hm we actually allow other tracing program types to attach to bpf_lsm_* functions, so I wonder we should toggle the lsm hook for each program type (for bpf_lsm_* trampolines) because they'd expect the hook is called
Quoted 6 lines
Tracing is about tracing, attaching a tracing program to
bpf_lsm_that changes the actual trace here is not something I would recommend. Infact, I have used tracing programs to figure out whether bpf_lsm_* is being called to debug stuff. Tracing users can always attach to security_* if they like.
I will rev up with this fix and send it out as I will be unavailable for the next 2 weeks.
- KP
Quoted 53 lines
- KP
Quoted 51 lines
but I'm not sure it's a valid use case to have like normal fentry program attached to
bpf_lsm_XXXfunctionjirka
Quoted 44 lines
also the trampoline attach is actually made in
bpf_trampoline_update, so I wonder it'd make more sense to put it in there, but it's already complicated, so it actually might be easier in herejirka
Quoted 37 lines
(quoted diff) +1
-0+ err = bpf_trampoline_update(tr, true /* lock_direct_mutex */); if (err) { hlist_del_init(&link->tramp_hlist);@@ -578,7 +596,6 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link*link, structbpf_{ struct bpf_tramp_link *link_exiting; enum bpf_tramp_prog_type kind;(quoted diff) +0
-1- bool lsm_link_found = false; int err, num_lsm_progs = 0; kind = bpf_attach_type_to_tramp(link->link.prog);@@ -595,18 +612,14 @@ static int __bpf_trampoline_unlink_prog(struct
bpf_tramp_link*link, structbpf_tramp_hlist) { if (link_exiting->link.prog->type == BPF_PROG_TYPE_LSM) num_lsm_progs++;(quoted diff) +2
-7- - if (link_exiting->link.prog == link->link.prog) - lsm_link_found = true; } } hlist_del_init(&link->tramp_hlist); tr->progs_cnt[kind]--; - if (lsm_link_found && num_lsm_progs == 1) - bpf_lsm_toggle_hook(tr->func.addr, false); - + if (link->link.prog->type == BPF_PROG_TYPE_LSM) + bpf_trampoline_toggle_lsm(tr, kind); return bpf_trampoline_update(tr, true /* lock_direct_mutex */); } - KP
KP Singh
maintainerPATCH v5 5/5 security: Add CONFIG_SECURITY_HOOK_LIKELY
In reply to KP SinghThis config influences the nature of the static key that guards the static call for LSM hooks.
When enabled, it indicates that an LSM static call slot is more likely to be initialized. When disabled, it optimizes for the case when static call slot is more likely to be not initialized.
When a major LSM like (SELinux, AppArmor, Smack etc) is active on a system the system would benefit from enabling the config. However there are other cases which would benefit from the config being disabled (e.g. a system with a BPF LSM with no hooks enabled by default, or an LSM like loadpin / yama). Ultimately, there is no one-size fits all solution.
with CONFIG_SECURITY_HOOK_LIKELY enabled, the inactive / uninitialized case is penalized with a direct jmp (still better than an indirect jmp):
function security_file_ioctl:
0xffffffff818f0c80 <+0>: endbr64
0xffffffff818f0c84 <+4>: nopl 0x0(%rax,%rax,1)
0xffffffff818f0c89 <+9>: push %rbp
0xffffffff818f0c8a <+10>: push %r14 0xffffffff818f0c8c <+12>: push %rbx 0xffffffff818f0c8d <+13>: mov %rdx,%rbx 0xffffffff818f0c90 <+16>: mov %esi,%ebp 0xffffffff818f0c92 <+18>: mov %rdi,%r14 0xffffffff818f0c95 <+21>: jmp 0xffffffff818f0ca8 <security_file_ioctl+40>
jump to skip the inactive BPF LSM hook.
0xffffffff818f0c97 <+23>: mov %r14,%rdi 0xffffffff818f0c9a <+26>: mov %ebp,%esi 0xffffffff818f0c9c <+28>: mov %rbx,%rdx 0xffffffff818f0c9f <+31>: call 0xffffffff8141e3b0 <bpf_lsm_file_ioctl> 0xffffffff818f0ca4 <+36>: test %eax,%eax 0xffffffff818f0ca6 <+38>: jne 0xffffffff818f0cbf <security_file_ioctl+63>
0xffffffff818f0ca8 <+40>: endbr64
0xffffffff818f0cac <+44>: jmp 0xffffffff818f0ccd <security_file_ioctl+77>
jump to skip the empty slot.
0xffffffff818f0cae <+46>: mov %r14,%rdi 0xffffffff818f0cb1 <+49>: mov %ebp,%esi 0xffffffff818f0cb3 <+51>: mov %rbx,%rdx 0xffffffff818f0cb6 <+54>: nopl 0x0(%rax,%rax,1)
^^^^^^^^^^^^^^^^^^^^^^^ Empty slot
0xffffffff818f0cbb <+59>: test %eax,%eax
0xffffffff818f0cbd <+61>: je 0xffffffff818f0ccd <security_file_ioctl+77>
0xffffffff818f0cbf <+63>: endbr64
0xffffffff818f0cc3 <+67>: pop %rbx 0xffffffff818f0cc4 <+68>: pop %r14 0xffffffff818f0cc6 <+70>: pop %rbp 0xffffffff818f0cc7 <+71>: cs jmp 0xffffffff82c00000 <__x86_return_thunk> 0xffffffff818f0ccd <+77>: endbr64 0xffffffff818f0cd1 <+81>: xor %eax,%eax 0xffffffff818f0cd3 <+83>: jmp 0xffffffff818f0cbf <security_file_ioctl+63> 0xffffffff818f0cd5 <+85>: mov %r14,%rdi 0xffffffff818f0cd8 <+88>: mov %ebp,%esi 0xffffffff818f0cda <+90>: mov %rbx,%rdx 0xffffffff818f0cdd <+93>: pop %rbx 0xffffffff818f0cde <+94>: pop %r14 0xffffffff818f0ce0 <+96>: pop %rbp
0xffffffff818f0ce1 <+97>: ret
When the config is disabled, the case optimizes the scenario above.
security_file_ioctl:
0xffffffff818f0e30 <+0>: endbr64
0xffffffff818f0e34 <+4>: nopl 0x0(%rax,%rax,1)
0xffffffff818f0e39 <+9>: push %rbp
0xffffffff818f0e3a <+10>: push %r14 0xffffffff818f0e3c <+12>: push %rbx 0xffffffff818f0e3d <+13>: mov %rdx,%rbx 0xffffffff818f0e40 <+16>: mov %esi,%ebp 0xffffffff818f0e42 <+18>: mov %rdi,%r14 0xffffffff818f0e45 <+21>: xchg %ax,%ax 0xffffffff818f0e47 <+23>: xchg %ax,%ax
The static keys in their disabled state do not create jumps leading to faster code.
0xffffffff818f0e49 <+25>: xor %eax,%eax 0xffffffff818f0e4b <+27>: xchg %ax,%ax 0xffffffff818f0e4d <+29>: pop %rbx 0xffffffff818f0e4e <+30>: pop %r14 0xffffffff818f0e50 <+32>: pop %rbp 0xffffffff818f0e51 <+33>: cs jmp 0xffffffff82c00000 <__x86_return_thunk> 0xffffffff818f0e57 <+39>: endbr64 0xffffffff818f0e5b <+43>: mov %r14,%rdi 0xffffffff818f0e5e <+46>: mov %ebp,%esi 0xffffffff818f0e60 <+48>: mov %rbx,%rdx 0xffffffff818f0e63 <+51>: call 0xffffffff8141e3b0 <bpf_lsm_file_ioctl> 0xffffffff818f0e68 <+56>: test %eax,%eax 0xffffffff818f0e6a <+58>: jne 0xffffffff818f0e4d <security_file_ioctl+29> 0xffffffff818f0e6c <+60>: jmp 0xffffffff818f0e47 <security_file_ioctl+23> 0xffffffff818f0e6e <+62>: endbr64 0xffffffff818f0e72 <+66>: mov %r14,%rdi 0xffffffff818f0e75 <+69>: mov %ebp,%esi 0xffffffff818f0e77 <+71>: mov %rbx,%rdx 0xffffffff818f0e7a <+74>: nopl 0x0(%rax,%rax,1) 0xffffffff818f0e7f <+79>: test %eax,%eax 0xffffffff818f0e81 <+81>: jne 0xffffffff818f0e4d <security_file_ioctl+29> 0xffffffff818f0e83 <+83>: jmp 0xffffffff818f0e49 <security_file_ioctl+25> 0xffffffff818f0e85 <+85>: endbr64 0xffffffff818f0e89 <+89>: mov %r14,%rdi 0xffffffff818f0e8c <+92>: mov %ebp,%esi 0xffffffff818f0e8e <+94>: mov %rbx,%rdx 0xffffffff818f0e91 <+97>: pop %rbx 0xffffffff818f0e92 <+98>: pop %r14
0xffffffff818f0e94 <+100>: pop %rbp
0xffffffff818f0e95 <+101>: ret
--- security/Kconfig | 11 +++++++++++ security/security.c | 6 ++++-- 2 files changed, 15 insertions(+), 2 deletions(-)
security/Kconfig +11 -0
@@ -32,6 +32,17 @@ config SECURITY If you are unsure how to answer this question, answer N. +config SECURITY_HOOK_LIKELY + bool "LSM hooks are likely to be initialized" + depends on SECURITY && EXPERT + default SECURITY_SELINUX || SECURITY_SMACK || SECURITY_TOMOYO || SECURITY_APPARMOR + help + This controls the behaviour of the static keys that guard LSM hooks. + If LSM hooks are likely to be initialized by LSMs, then one gets + better performance by enabling this option. However, if the system is + using an LSM where hooks are much likely to be disabled, one gets + better performance by disabling this config. + config SECURITYFS bool "Enable the securityfs filesystem" help
security/security.c +4 -2
@@ -825,7 +825,8 @@ static int lsm_superblock_alloc(struct super_block *sb)*/ #define __CALL_STATIC_VOID (NUM ,HOOK , ...) \ do { \ - if (static_branch_unlikely(&SECURITY_HOOK_ACTIVE_KEY(HOOK, NUM))) { \ + if (static_branch_maybe(CONFIG_SECURITY_HOOK_LIKELY, \ + &SECURITY_HOOK_ACTIVE_KEY(HOOK, NUM))) { \ static_call(LSM_STATIC_CALL(HOOK, NUM))(__VA_ARGS__); \ } \ } while (0);
@@ -837,7 +838,8 @@ do { \#define __CALL_STATIC_INT (NUM ,R ,HOOK ,LABEL , ...) \ do { \ - if (static_branch_unlikely(&SECURITY_HOOK_ACTIVE_KEY(HOOK, NUM))) { \ + if (static_branch_maybe(CONFIG_SECURITY_HOOK_LIKELY, \ + &SECURITY_HOOK_ACTIVE_KEY(HOOK, NUM))) { \ R = static_call(LSM_STATIC_CALL(HOOK, NUM))(__VA_ARGS__); \ if (R != 0) \ goto LABEL; \
Signature · 1 line
2.42.0.582.g8ccd20d70d-goog
Re: [PATCH v5 5/5] security: Add CONFIG_SECURITY_HOOK_LIKELY
Quoted 115 lines · KP Singh
This config influences the nature of the static key that guards the static call for LSM hooks.
When enabled, it indicates that an LSM static call slot is more likely to be initialized. When disabled, it optimizes for the case when static call slot is more likely to be not initialized.
When a major LSM like (SELinux, AppArmor, Smack etc) is active on a system the system would benefit from enabling the config. However there are other cases which would benefit from the config being disabled (e.g. a system with a BPF LSM with no hooks enabled by default, or an LSM like loadpin / yama). Ultimately, there is no one-size fits all solution.
with
CONFIG_SECURITY_HOOK_LIKELYenabled, the inactive / uninitialized case is penalized with a direct jmp (still better than an indirect jmp):function
security_file_ioctl:0xffffffff818f0c80 <+0>: endbr64 0xffffffff818f0c84 <+4>: nopl 0x0(%rax,%rax,1) 0xffffffff818f0c89 <+9>: push %rbp 0xffffffff818f0c8a <+10>: push %r14 0xffffffff818f0c8c <+12>: push %rbx 0xffffffff818f0c8d <+13>: mov %rdx,%rbx 0xffffffff818f0c90 <+16>: mov %esi,%ebp 0xffffffff818f0c92 <+18>: mov %rdi,%r14 0xffffffff818f0c95 <+21>: jmp 0xffffffff818f0ca8 <security_file_ioctl+40> jump to skip the inactive BPF LSM hook. 0xffffffff818f0c97 <+23>: mov %r14,%rdi 0xffffffff818f0c9a <+26>: mov %ebp,%esi 0xffffffff818f0c9c <+28>: mov %rbx,%rdx 0xffffffff818f0c9f <+31>: call 0xffffffff8141e3b0 <bpf_lsm_file_ioctl> 0xffffffff818f0ca4 <+36>: test %eax,%eax 0xffffffff818f0ca6 <+38>: jne 0xffffffff818f0cbf <security_file_ioctl+63> 0xffffffff818f0ca8 <+40>: endbr64 0xffffffff818f0cac <+44>: jmp 0xffffffff818f0ccd <security_file_ioctl+77> jump to skip the empty slot. 0xffffffff818f0cae <+46>: mov %r14,%rdi 0xffffffff818f0cb1 <+49>: mov %ebp,%esi 0xffffffff818f0cb3 <+51>: mov %rbx,%rdx 0xffffffff818f0cb6 <+54>: nopl 0x0(%rax,%rax,1)^^^^^^^^^^^^^^^^^^^^^^^ Empty slot
0xffffffff818f0cbb <+59>: test %eax,%eax 0xffffffff818f0cbd <+61>: je 0xffffffff818f0ccd <security_file_ioctl+77> 0xffffffff818f0cbf <+63>: endbr64 0xffffffff818f0cc3 <+67>: pop %rbx 0xffffffff818f0cc4 <+68>: pop %r14 0xffffffff818f0cc6 <+70>: pop %rbp 0xffffffff818f0cc7 <+71>: cs jmp 0xffffffff82c00000 <__x86_return_thunk> 0xffffffff818f0ccd <+77>: endbr64 0xffffffff818f0cd1 <+81>: xor %eax,%eax 0xffffffff818f0cd3 <+83>: jmp 0xffffffff818f0cbf <security_file_ioctl+63> 0xffffffff818f0cd5 <+85>: mov %r14,%rdi 0xffffffff818f0cd8 <+88>: mov %ebp,%esi 0xffffffff818f0cda <+90>: mov %rbx,%rdx 0xffffffff818f0cdd <+93>: pop %rbx 0xffffffff818f0cde <+94>: pop %r14 0xffffffff818f0ce0 <+96>: pop %rbp 0xffffffff818f0ce1 <+97>: retWhen the config is disabled, the case optimizes the scenario above.
security_file_ioctl:0xffffffff818f0e30 <+0>: endbr64 0xffffffff818f0e34 <+4>: nopl 0x0(%rax,%rax,1) 0xffffffff818f0e39 <+9>: push %rbp 0xffffffff818f0e3a <+10>: push %r14 0xffffffff818f0e3c <+12>: push %rbx 0xffffffff818f0e3d <+13>: mov %rdx,%rbx 0xffffffff818f0e40 <+16>: mov %esi,%ebp 0xffffffff818f0e42 <+18>: mov %rdi,%r14 0xffffffff818f0e45 <+21>: xchg %ax,%ax 0xffffffff818f0e47 <+23>: xchg %ax,%ax The static keys in their disabled state do not create jumps leading to faster code. 0xffffffff818f0e49 <+25>: xor %eax,%eax 0xffffffff818f0e4b <+27>: xchg %ax,%ax 0xffffffff818f0e4d <+29>: pop %rbx 0xffffffff818f0e4e <+30>: pop %r14 0xffffffff818f0e50 <+32>: pop %rbp 0xffffffff818f0e51 <+33>: cs jmp 0xffffffff82c00000 <__x86_return_thunk> 0xffffffff818f0e57 <+39>: endbr64 0xffffffff818f0e5b <+43>: mov %r14,%rdi 0xffffffff818f0e5e <+46>: mov %ebp,%esi 0xffffffff818f0e60 <+48>: mov %rbx,%rdx 0xffffffff818f0e63 <+51>: call 0xffffffff8141e3b0 <bpf_lsm_file_ioctl> 0xffffffff818f0e68 <+56>: test %eax,%eax 0xffffffff818f0e6a <+58>: jne 0xffffffff818f0e4d <security_file_ioctl+29> 0xffffffff818f0e6c <+60>: jmp 0xffffffff818f0e47 <security_file_ioctl+23> 0xffffffff818f0e6e <+62>: endbr64 0xffffffff818f0e72 <+66>: mov %r14,%rdi 0xffffffff818f0e75 <+69>: mov %ebp,%esi 0xffffffff818f0e77 <+71>: mov %rbx,%rdx 0xffffffff818f0e7a <+74>: nopl 0x0(%rax,%rax,1) 0xffffffff818f0e7f <+79>: test %eax,%eax 0xffffffff818f0e81 <+81>: jne 0xffffffff818f0e4d <security_file_ioctl+29> 0xffffffff818f0e83 <+83>: jmp 0xffffffff818f0e49 <security_file_ioctl+25> 0xffffffff818f0e85 <+85>: endbr64 0xffffffff818f0e89 <+89>: mov %r14,%rdi 0xffffffff818f0e8c <+92>: mov %ebp,%esi 0xffffffff818f0e8e <+94>: mov %rbx,%rdx 0xffffffff818f0e91 <+97>: pop %rbx 0xffffffff818f0e92 <+98>: pop %r14 0xffffffff818f0e94 <+100>: pop %rbp 0xffffffff818f0e95 <+101>: ret
Quoted 2 lines
Acked-by Song Liu <[email protected]>Signed-off-by KP Singh <[email protected]>
This looks excellent, and gives us the right balance automatically. :)
Signature · 1 line
Kees Cook
Quoted 5 lines · KP Singh
# 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.
Paul, FWIW, I think this series is ready to land in -next. I'd like it to get some bake time there just to see if anything unexpected shows up. It's quite happy in all my local testing, though.
-Kees
Signature · 1 line
Kees Cook
Quoted 43 lines · KP Singh
# 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)
===============================================================================
Quoted 8 lines
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
FTR, I also measure a ~3% tput improvement in UDP stream test over loopback.
@KP Singh, I would have appreciated being cc-ed here, since I provided feedback on a previous revision (as soon as I learned of this effort).
Cheers,
Paolo
KP Singh
maintainerQuoted 54 lines · Paolo Abeni
Quoted 43 lines · KP Singh
# 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)
===============================================================================Quoted 8 lines
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
Quoted 3 lines
FTR, I also measure a ~3% tput improvement in UDP stream test over loopback.
Thanks for running the numbers and testing these patches, greatly appreciated!
Quoted 1 line
@KP Singh, I would have appreciated being cc-ed here, since I provided
Definitely, a miss on my part. Will keep you Cc'ed in any future revisions.
I think we can also add a Tested-by: tag on the main patch and add your performance numbers to the commit as well.
- KP
Quoted 6 lines
feedback on a previous revision (as soon as I learned of this effort).
Cheers,
Paolo
Quoted 58 lines · KP Singh
Quoted 53 lines · Paolo Abeni
Quoted 43 lines · KP Singh
# 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)
===============================================================================Quoted 8 lines
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): +3.4859Quoted 3 lines
FTR, I also measure a ~3% tput improvement in UDP stream test over loopback.
Quoted 5 lines
Thanks for running the numbers and testing these patches, greatly appreciated!
Quoted 1 line
@KP Singh, I would have appreciated being cc-ed here, since I provided
Definitely, a miss on my part. Will keep you Cc'ed in any future revisions.
Thanks!
Quoted 2 lines
I think we can also add a Tested-by: tag on the main patch and add your performance numbers to the commit as well.
Feel free to include that, even if my testing is limited to the performance test described above.
Cheers,
Paolo