Syscall User Redirection
v4 cover Syscall User Redirection
Hi,
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscalls is not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.
Also, I didn't expose tests for PR_GET as that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.
Finally, one question: Which tree would this go through?
Gabriel Krisman Bertazi (2):
kernel: Implement selective syscall userspace redirection
selftests: Add kselftest for syscall user dispatch
arch/Kconfig | 20 ++ arch/x86/Kconfig | 1 + arch/x86/entry/common.c | 5 + arch/x86/include/asm/thread_info.h | 4 +- arch/x86/kernel/signal_compat.c | 2 +- fs/exec.c | 2 + include/linux/sched.h | 3 + include/linux/syscall_user_dispatch.h | 50 ++++ include/uapi/asm-generic/siginfo.h | 3 +- include/uapi/linux/prctl.h | 5 + kernel/Makefile | 1 + kernel/fork.c | 1 + kernel/sys.c | 5 + kernel/syscall_user_dispatch.c | 92 +++++++ tools/testing/selftests/Makefile | 1 + .../syscall_user_dispatch/.gitignore | 2 + .../selftests/syscall_user_dispatch/Makefile | 9 + .../selftests/syscall_user_dispatch/config | 1 + .../syscall_user_dispatch.c | 256 ++++++++++++++++++ 19 files changed, 460 insertions(+), 3 deletions(-) create mode 100644 include/linux/syscall_user_dispatch.h create mode 100644 kernel/syscall_user_dispatch.c create mode 100644 tools/testing/selftests/syscall_user_dispatch/.gitignore create mode 100644 tools/testing/selftests/syscall_user_dispatch/Makefile create mode 100644 tools/testing/selftests/syscall_user_dispatch/config create mode 100644 tools/testing/selftests/syscall_user_dispatch/syscall_user_dispatch.c
Signature · 1 line
2.27.0
PATCH v4 1/2 kernel: Implement selective syscall userspace redirection
Introduce a mechanism to quickly disable/enable syscall handling for a specific process and redirect to userspace via SIGSYS. This is useful for processes with parts that require syscall redirection and parts that don't, but who need to perform this boundary crossing really fast, without paying the cost of a system call to reconfigure syscall handling on each boundary transition. This is particularly important for Windows games running over Wine.
The proposed interface looks like this:
prctl(PR_SET_SYSCALL_USER_DISPATCH, <op>, <start_addr>, <end_addr>, [selector])
The range [<start_addr>,<end_addr>] is a part of the process memory map that is allowed to by-pass the redirection code and dispatch syscalls directly, such that in fast paths a process doesn't need to disable the trap nor the kernel has to check the selector. This is essential to return from SIGSYS to a blocked area without triggering another SIGSYS from rt_sigreturn.
selector is an optional pointer to a char-sized userspace memory region that has a key switch for the mechanism. This key switch is set to either PR_SYS_DISPATCH_ON, PR_SYS_DISPATCH_OFF to enable and disable the redirection without calling the kernel.
The feature is meant to be set per-thread and it is disabled on fork/clone/execv.
Internally, this doesn't add overhead to the syscall hot path, and it requires very little per-architecture support. I avoided using seccomp, even though it duplicates some functionality, due to previous feedback that maybe it shouldn't mix with seccomp since it is not a security mechanism. And obviously, this should never be considered a security mechanism, since any part of the program can by-pass it by using the syscall dispatcher.
For the sysinfo benchmark, which measures the overhead added to executing a native syscall that doesn't require interception, the overhead using only the direct dispatcher region to issue syscalls is pretty much irrelevant. The overhead of using the selector goes around 40ns for a native (unredirected) syscall in my system, and it is (as expected) dominated by the supervisor-mode user-address access. In fact, with SMAP off, the overhead is consistently less than 5ns on my test box.
Right now, it is only supported by x86_64 and x86, but it should be easily enabled for other architectures.
An example code using this interface can be found at:
https://gitlab.collabora.com/krisman/syscall-disable-personality
Changes since v2:
(Matthew Wilcox suggestions)
- Drop __user on non-ptr type.
- Move #define closer to similar defs
- Allow a memory region that can dispatch directly (Kees Cook suggestions)
- Improve kconfig summary line
- Move flag cleanup on execve to begin_new_exec
- Hint branch predictor in the syscall path (Me)
- Convert selector to char
Changes since RFC:
(Kees Cook suggestions)
- Don't mention personality while explaining the feature
- Use syscall_get_nr
- Remove header guard on several places
- Convert WARN_ON to WARN_ON_ONCE
- Explicit check for state values
- Rename to syscall user dispatcher
--- arch/Kconfig | 20 ++++++ arch/x86/Kconfig | 1 + arch/x86/entry/common.c | 5 ++ arch/x86/include/asm/thread_info.h | 4 +- arch/x86/kernel/signal_compat.c | 2 +- fs/exec.c | 2 + include/linux/sched.h | 3 + include/linux/syscall_user_dispatch.h | 50 +++++++++++++++ include/uapi/asm-generic/siginfo.h | 3 +- include/uapi/linux/prctl.h | 5 ++ kernel/Makefile | 1 + kernel/fork.c | 1 + kernel/sys.c | 5 ++ kernel/syscall_user_dispatch.c | 92 +++++++++++++++++++++++++++ 14 files changed, 191 insertions(+), 3 deletions(-) create mode 100644 include/linux/syscall_user_dispatch.h create mode 100644 kernel/syscall_user_dispatch.c
arch/Kconfig +20 -0
@@ -465,6 +465,26 @@ config SECCOMP_FILTER See Documentation/userspace-api/seccomp_filter.rst for details. +config HAVE_ARCH_SYSCALL_USER_DISPATCH + bool + help + An arch should select this symbol if it provides all of these things: + - TIF_SYSCALL_USER_DISPATCH + - syscall_get_arch + - syscall_rollback + - syscall_get_nr + - SIGSYS siginfo_t support + +config SYSCALL_USER_DISPATCH + bool "Support syscall redirection to userspace dispatcher" + depends on HAVE_ARCH_SYSCALL_USER_DISPATCH + help + Enable tasks to ask the kernel to redirect syscalls not + issued from a predefined dispatcher back to userspace, + depending on a userspace memory selector. + + This option is useful to optimize games running over Wine. + config HAVE_ARCH_STACKLEAK bool help
arch/x86/Kconfig +1 -0
@@ -149,6 +149,7 @@ config X86 select HAVE_ARCH_COMPAT_MMAP_BASES if MMU && COMPAT select HAVE_ARCH_PREL32_RELOCATIONS select HAVE_ARCH_SECCOMP_FILTER + select HAVE_ARCH_SYSCALL_USER_DISPATCH select HAVE_ARCH_THREAD_STRUCT_WHITELIST select HAVE_ARCH_STACKLEAK select HAVE_ARCH_TRACEHOOK
arch/x86/entry/common.c +5 -0
@@ -138,6 +138,11 @@ static long syscall_trace_enter(struct pt_regs *regs)return -1L ; } +if (work & _TIF_SYSCALL_USER_DISPATCH) { +if (do_syscall_user_dispatch (regs)) +return -1L ; + } +#ifdef CONFIG_SECCOMP /* * Do seccomp after ptrace, to catch any tracer changes.
arch/x86/include/asm/thread_info.h +3 -1
@@ -93,6 +93,7 @@ struct thread_info {#define TIF_NOTSC 16/* TSC is not accessible in userland */ #define TIF_IA32 17/* IA32 compatibility process */ #define TIF_SLD 18/* Restore split lock detection on context switch */ +#define TIF_SYSCALL_USER_DISPATCH 19/* Redirect syscall for userspace handling */ #define TIF_MEMDIE 20/* is terminating due to OOM killer */ #define TIF_POLLING_NRFLAG 21/* idle is polling for TIF_NEED_RESCHED */ #define TIF_IO_BITMAP 22/* uses I/O bitmap */
@@ -123,6 +124,7 @@ struct thread_info {#define _TIF_NOTSC (1 << TIF_NOTSC)#define _TIF_IA32 (1 << TIF_IA32)#define _TIF_SLD (1 << TIF_SLD) +#define _TIF_SYSCALL_USER_DISPATCH (1 << TIF_SYSCALL_USER_DISPATCH)#define _TIF_POLLING_NRFLAG (1 << TIF_POLLING_NRFLAG)#define _TIF_IO_BITMAP (1 << TIF_IO_BITMAP)#define _TIF_FORCED_TF (1 << TIF_FORCED_TF)
@@ -136,7 +138,7 @@ struct thread_info {/* Work to do before invoking the actual syscall. */ #define _TIF_WORK_SYSCALL_ENTRY \ (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_EMU | _TIF_SYSCALL_AUDIT | \ - _TIF_SECCOMP | _TIF_SYSCALL_TRACEPOINT) + _TIF_SECCOMP | _TIF_SYSCALL_TRACEPOINT | _TIF_SYSCALL_USER_DISPATCH)/* flags to check in __switch_to() */ #define _TIF_WORK_CTXSW_BASE \
arch/x86/kernel/signal_compat.c +1 -1
@@ -31,7 +31,7 @@ static inline void signal_compat_build_tests(void)BUILD_BUG_ON (NSIGBUS !=5 );BUILD_BUG_ON (NSIGTRAP !=5 );BUILD_BUG_ON (NSIGCHLD !=6 ); -BUILD_BUG_ON (NSIGSYS !=1 ); +BUILD_BUG_ON (NSIGSYS !=2 );/* This is part of the ABI and can never change in size: */ BUILD_BUG_ON (sizeof (compat_siginfo_t) !=128 );
fs/exec.c +2 -0
@@ -1386,6 +1386,8 @@ int begin_new_exec(struct linux_binprm * bprm)flush_thread (); me->personality &= ~bprm->per_clear; +clear_tsk_syscall_user_dispatch (me); +/* * We have to apply CLOEXEC before we change whether the process is * dumpable (in setup_new_exec) to avoid a race with a process in userspace
include/linux/sched.h +3 -0
@@ -32,6 +32,7 @@#include <linux/posix-timers.h> #include <linux/rseq.h> #include <linux/kcsan.h> +#include <linux/syscall_user_dispatch.h> /* task_struct member predeclarations (sorted alphabetically): */ struct audit_context ;
@@ -953,6 +954,8 @@ struct task_struct {#endif struct seccomp seccomp; +struct syscall_user_dispatch syscall_dispatch; +/* Thread group tracking: */ u64 parent_exec_id;u64 self_exec_id;
include/linux/syscall_user_dispatch.h new file +50 -0
50 lines not on this page · raw message
include/uapi/asm-generic/siginfo.h +2 -1
@@ -284,7 +284,8 @@ typedef struct siginfo {* SIGSYS si_codes */ #define SYS_SECCOMP 1/* seccomp triggered */ -#define NSIGSYS 1 +#define SYS_USER_DISPATCH 2/* syscall user dispatch triggered */ +#define NSIGSYS 2/* * SIGEMT si_codes
include/uapi/linux/prctl.h +5 -0
@@ -238,4 +238,9 @@ struct prctl_mm_map {#define PR_SET_IO_FLUSHER 57#define PR_GET_IO_FLUSHER 58 +/* Dispatch syscalls to a userspace handler */ +#define PR_SET_SYSCALL_USER_DISPATCH 59 +# define PR_SYS_DISPATCH_OFF 0 +# define PR_SYS_DISPATCH_ON 1 +#endif /* _LINUX_PRCTL_H */
kernel/Makefile +1 -0
@@ -92,6 +92,7 @@ obj-$(CONFIG_DETECT_HUNG_TASK) += hung_task.o obj-$(CONFIG_LOCKUP_DETECTOR) += watchdog.o obj-$(CONFIG_HARDLOCKUP_DETECTOR_PERF) += watchdog_hld.o obj-$(CONFIG_SECCOMP) += seccomp.o +obj-$(CONFIG_SYSCALL_USER_DISPATCH) += syscall_user_dispatch.o obj-$(CONFIG_RELAY) += relay.o obj-$(CONFIG_SYSCTL) += utsname_sysctl.o obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o
kernel/fork.c +1 -0
@@ -921,6 +921,7 @@ static struct task_struct *dup_task_struct(struct task_struct *orig, int node)clear_user_return_notifier (tsk);clear_tsk_need_resched (tsk);set_task_stack_end_magic (tsk); +clear_tsk_syscall_user_dispatch (tsk);#ifdef CONFIG_STACKPROTECTOR tsk->stack_canary =get_random_canary ();
kernel/sys.c +5 -0
@@ -42,6 +42,7 @@#include <linux/syscore_ops.h> #include <linux/version.h> #include <linux/ctype.h> +#include <linux/syscall_user_dispatch.h> #include <linux/compat.h> #include <linux/syscalls.h>
@@ -2527,6 +2528,10 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, error = (current->flags &PR_IO_FLUSHER ) ==PR_IO_FLUSHER ;break ; +case PR_SET_SYSCALL_USER_DISPATCH : + error =set_syscall_user_dispatch ((int ) arg2, arg3, arg4, + (char __user *) arg5); +break ;default : error = -EINVAL ;break ;
kernel/syscall_user_dispatch.c new file +92 -0
92 lines not on this page · raw message
Signature · 1 line
2.27.0
Re: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
Quoted 7 lines · Gabriel Krisman Bertazi
selector is an optional pointer to a char-sized userspace memory region that has a key switch for the mechanism. This key switch is set to either
PR_SYS_DISPATCH_ON,PR_SYS_DISPATCH_OFFto enable and disable the redirection without calling the kernel.The feature is meant to be set per-thread and it is disabled on fork/clone/execv.
Disabled on exec. Disabled in the child on clone/fork (and vfork, I think).
That means we don't need to worry about it interacting badly with a setuid program, right?
Re: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
Quoted 6 lines · Matthew Wilcox
Quoted 5 lines · Gabriel Krisman Bertazi
selector is an optional pointer to a char-sized userspace memory region that has a key switch for the mechanism. This key switch is set to either
PR_SYS_DISPATCH_ON,PR_SYS_DISPATCH_OFFto enable and disable the redirection without calling the kernel.
Quoted 8 lines
Quoted 2 lines
The feature is meant to be set per-thread and it is disabled on fork/clone/execv.
Disabled on exec. Disabled in the child on clone/fork (and vfork, I think).
That means we don't need to worry about it interacting badly with a setuid program, right?
Right, that's the intention.
Signature · 1 line
Kees Cook
Andy Lutomirski
maintainerRe: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
In reply to Gabriel Krisman BertaziQuoted 1 line · Gabriel Krisman Bertazi
This is quite nice. I have a few comments, though:
You mentioned rt_sigreturn(). Should this automatically exempt the kernel-provided signal restorer on architectures (e.g. x86_32) that provide one?
The amount of syscall entry wiring that arches need to do is IMO already a bit out of hand. Should we instead rename TIF_SECCOMP to TIF_SYSCALL_INTERCEPTION and have one generic callback that handles seccomp and this new thing?
Quoted 8 lines
kernel/syscall_user_dispatch.c new file +8
-0+int do_syscall_user_dispatch (struct pt_regs *regs) +{ +struct syscall_user_dispatch *sd = ¤t->syscall_dispatch; +unsigned long ip =instruction_pointer (regs); +char state; + +if (likely (ip >= sd->dispatcher_start && ip <= sd->dispatcher_end)) +return 0 ;
kernel/syscall_user_dispatch.c line 48
Quoted 10 lines
kernel/syscall_user_dispatch.c new file +10
-0+ +if (likely (sd->selector)) { +if (unlikely (__get_user (state, sd->selector))) +do_exit (SIGSEGV ); + +if (likely (state ==0 )) +return 0 ; + +if (state !=1 ) +do_exit (SIGSEGV );
This seems a bit extreme and hard to debug if it ever happens.
Re: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
Quoted 9 lines · Andy Lutomirski
Quoted 1 line · Gabriel Krisman Bertazi
This is quite nice. I have a few comments, though:
You mentioned
rt_sigreturn(). Should this automatically exempt the kernel-provided signal restorer on architectures (e.g.x86_32) that provide one?
That seems reasonable. Not sure how easy it is to do it, though.
Quoted 4 lines
The amount of syscall entry wiring that arches need to do is IMO already a bit out of hand. Should we instead rename
TIF_SECCOMPtoTIF_SYSCALL_INTERCEPTIONand have one generic callback that handles seccomp and this new thing?
Considering the previous suggestion from Kees to hide it inside the tracehook and Thomas rework of this path, I'm not sure what is the best solution here, but some rework of these flags is due. Thomas suggested expanding these flags to 64 bits and having some arch specific and arch-agnostic flags. With the storage expansion and arch-agnostic flags, would this still be desirable?
Quoted 12 lines
Quoted 12 lines
kernel/syscall_user_dispatch.c new file +12
-0+int do_syscall_user_dispatch (struct pt_regs *regs) +{ +struct syscall_user_dispatch *sd = ¤t->syscall_dispatch; +unsigned long ip =instruction_pointer (regs); +char state; + +if (likely (ip >= sd->dispatcher_start && ip <= sd->dispatcher_end)) +return 0 ; + +if (likely (sd->selector)) { +if (unlikely (__get_user (state, sd->selector))) +do_exit (SIGSEGV );
Quoted 8 lines
kernel/syscall_user_dispatch.c line 48
Quoted 6 lines
kernel/syscall_user_dispatch.c new file +6
-0+ +if (likely (state ==0 )) +return 0 ; + +if (state !=1 ) +do_exit (SIGSEGV );This seems a bit extreme and hard to debug if it ever happens.
Makes sense, but I don't see a better way to return the error here. Maybe a SIGSYS with a different si_errno? Alternatively, we could revert to the previous behavior of allowing syscalls on state != 0, that existed in v1. What do you think?
Signature · 1 line
Gabriel Krisman Bertazi
Andy Lutomirski
maintainerRe: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
Quoted 7 lines · Gabriel Krisman Bertazi
Quoted 4 lines · Andy Lutomirski
Quoted 1 line · Gabriel Krisman Bertazi
Quoted 7 lines
Quoted 5 lines
This is quite nice. I have a few comments, though:
You mentioned
rt_sigreturn(). Should this automatically exempt the kernel-provided signal restorer on architectures (e.g.x86_32) that provide one?That seems reasonable. Not sure how easy it is to do it, though.
For better or for worse, it's currently straightforward because the code is:
__kernel_sigreturn:
.LSTART_sigreturn:
popl %eax /* XXX does this mean it needs unwind info? */
movl $__NR_sigreturn, %eax
SYSCALL_ENTER_KERNEL
and SYSCALL_ENTER_KERNEL is hardwired as int $0x80. (The latter is probably my fault, for better or for worse.) So this would change to:
__vdso32_sigreturn_syscall:
SYSCALL_ENTER_KERNEL
and vdso2c would wire up __vdso32_sigreturn_syscall. Then there would be something like:
bool arch_syscall_is_vdso_sigreturn(struct pt_regs *regs);
and that would be that. Does anyone have an opinion as to whether this is a good idea? Modern glibc shouldn't be using this mechanism, I think, but I won't swear to it.
Quoted 6 lines
Quoted 5 lines
The amount of syscall entry wiring that arches need to do is IMO already a bit out of hand. Should we instead rename
TIF_SECCOMPtoTIF_SYSCALL_INTERCEPTIONand have one generic callback that handles seccomp and this new thing?
Quoted 6 lines
Considering the previous suggestion from Kees to hide it inside the tracehook and Thomas rework of this path, I'm not sure what is the best solution here, but some rework of these flags is due. Thomas suggested expanding these flags to 64 bits and having some arch specific and arch-agnostic flags. With the storage expansion and arch-agnostic flags, would this still be desirable?
I think it would be desirable to consolidate this to avoid having multiple arches need to separately wire up all of these mechanisms. I'm not sure that the initial upstream implementation needs this, but it might be nice to support this out of the box on all arches with seccomp support.
Quoted 22 lines
Quoted 13 lines
Quoted 13 lines
kernel/syscall_user_dispatch.c new file +12
-0+int do_syscall_user_dispatch (struct pt_regs *regs) +{ +struct syscall_user_dispatch *sd = ¤t->syscall_dispatch; +unsigned long ip =instruction_pointer (regs); +char state; + +if (likely (ip >= sd->dispatcher_start && ip <= sd->dispatcher_end)) +return 0 ; + +if (likely (sd->selector)) { +if (unlikely (__get_user (state, sd->selector))) +do_exit (SIGSEGV );Quoted 8 lines
kernel/syscall_user_dispatch.c line 48
Quoted 6 lines
kernel/syscall_user_dispatch.c new file +6
-0+ +if (likely (state ==0 )) +return 0 ; + +if (state !=1 ) +do_exit (SIGSEGV );This seems a bit extreme and hard to debug if it ever happens.
Quoted 5 lines
Makes sense, but I don't see a better way to return the error here. Maybe a SIGSYS with a different
si_errno? Alternatively, we could revert to the previous behavior of allowing syscalls on state != 0, that existed in v1. What do you think?
I don't have a strong opinion. SIGSYS with different si_errno is probably reasonable.
--Andy
Re: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
Quoted 31 lines · Andy Lutomirski
Quoted 7 lines · Gabriel Krisman Bertazi
Quoted 4 lines · Andy Lutomirski
Quoted 1 line · Gabriel Krisman Bertazi
Quoted 7 lines
Quoted 5 lines
This is quite nice. I have a few comments, though:
You mentioned
rt_sigreturn(). Should this automatically exempt the kernel-provided signal restorer on architectures (e.g.x86_32) that provide one?That seems reasonable. Not sure how easy it is to do it, though.
For better or for worse, it's currently straightforward because the code is:
__kernel_sigreturn:
.LSTART_sigreturn:
popl %eax /* XXX does this mean it needs unwind info? */
movl $__NR_sigreturn, %eax
SYSCALL_ENTER_KERNELand
SYSCALL_ENTER_KERNELis hardwired as int $0x80. (The latter is probably my fault, for better or for worse.) So this would change to:
__vdso32_sigreturn_syscall:
SYSCALL_ENTER_KERNEL
Quoted 8 lines
and vdso2c would wire up
__vdso32_sigreturn_syscall. Then there would be something like:bool
arch_syscall_is_vdso_sigreturn(struct pt_regs *regs);and that would be that. Does anyone have an opinion as to whether this is a good idea? Modern glibc shouldn't be using this mechanism, I think, but I won't swear to it.
On arm64 sigreturn is always through the vdso, so IIUC we'd certainly need something like this. Otherwise it'd be the user's responsibility to register the vdso sigtramp range when making the prctl, and flip the selector in each signal handler, which sounds both painful and fragile.
Mark.
Re: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
In reply to Andy LutomirskiQuoted 6 lines · Andy Lutomirski
On Thu, Jul 16, 2020 at 12:31 PM Gabriel Krisman Bertazi <[email protected]> wrote:
The amount of syscall entry wiring that arches need to do is IMO already a bit out of hand. Should we instead renameTIF_SECCOMPtoTIF_SYSCALL_INTERCEPTIONand have one generic callback that handles seccomp and this new thing?
The right way to go is to consolidate all the stupidly different entry/exit work handling implementations and have exactly one in generic code, i.e. what I posted a few days ago.
Then we can make new features only available in the generic version by hiding the new functionality in the core code and not exposing the functions to architecture implementations.
Making it easy for architectures to keep their own variant forever just proliferates the mess we have right now.
Thanks,
tglx
Re: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
Quoted 12 lines · Thomas Gleixner
Quoted 6 lines · Andy Lutomirski
On Thu, Jul 16, 2020 at 12:31 PM Gabriel Krisman Bertazi <[email protected]> wrote:
The amount of syscall entry wiring that arches need to do is IMO already a bit out of hand. Should we instead renameTIF_SECCOMPtoTIF_SYSCALL_INTERCEPTIONand have one generic callback that handles seccomp and this new thing?The right way to go is to consolidate all the stupidly different entry/exit work handling implementations and have exactly one in generic code, i.e. what I posted a few days ago.
Quoted 6 lines
Then we can make new features only available in the generic version by hiding the new functionality in the core code and not exposing the functions to architecture implementations.
Making it easy for architectures to keep their own variant forever just proliferates the mess we have right now.
Couldn't agree more. We recently added PTRACE_SYSEMU to arm64 and I deeply regret doing that now that yet another way to rewrite the syscall number has come along. I only just untangled some of the mess in our entry code for that, so I can't say I'm looking forward to opening it right back up to support this new feature. Much better to do it in the core code instead.
Will
Re: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
In reply to Gabriel Krisman BertaziGabriel,
Quoted 10 lines · Gabriel Krisman Bertazi
Introduce a mechanism to quickly disable/enable syscall handling for a specific process and redirect to userspace via SIGSYS. This is useful for processes with parts that require syscall redirection and parts that don't, but who need to perform this boundary crossing really fast, without paying the cost of a system call to reconfigure syscall handling on each boundary transition. This is particularly important for Windows games running over Wine.
The proposed interface looks like this:
Quoted 8 lines
prctl(PR_SET_SYSCALL_USER_DISPATCH, <op>, <start_addr>, <end_addr>, [selector])The range [<start_addr>,<end_addr>] is a part of the process memory map that is allowed to by-pass the redirection code and dispatch syscalls directly, such that in fast paths a process doesn't need to disable the trap nor the kernel has to check the selector. This is essential to return from SIGSYS to a blocked area without triggering another SIGSYS from
rt_sigreturn.
Why isn't rt_sigreturn() exempt from that redirection in the first place?
Quoted 7 lines
--- arch/Kconfig | 20 ++++++ arch/x86/Kconfig | 1 + arch/x86/entry/common.c | 5 ++ arch/x86/include/asm/thread_info.h | 4 +- arch/x86/kernel/signal_compat.c | 2 +- fs/exec.c | 2 +
Quoted 8 lines
include/linux/sched.h | 3 + include/linux/syscall_user_dispatch.h | 50 +++++++++++++++ include/uapi/asm-generic/siginfo.h | 3 +- include/uapi/linux/prctl.h | 5 ++ kernel/Makefile | 1 + kernel/fork.c | 1 + kernel/sys.c | 5 ++ kernel/syscall_user_dispatch.c | 92 +++++++++++++++++++++++++++
A big combo patch is not how we do that. Please split it up into the core part and a patch enabling it for a particular architexture.
As I said in my reply to Andy, this wants to go on top of the generic entry/exit work stuff:
https://lore.kernel.org/r/[email protected]
and then syscall_user_dispatch.c ends up in kernel/entry/ and the dispatching function is not exposed outside of that directory.
I'm going to post a new version later today. Will cc you.
arch/x86/include/asm/thread_info.h struct thread_info { · line 96
Quoted 7 lines
arch/x86/include/asm/thread_info.h +1
-0@@ -93,6 +93,7 @@ struct thread_info {#define TIF_NOTSC 16/* TSC is not accessible in userland */ #define TIF_IA32 17/* IA32 compatibility process */ #define TIF_SLD 18/* Restore split lock detection on context switch */ +#define TIF_SYSCALL_USER_DISPATCH 19/* Redirect syscall for userspace handling */
There are two other things out there which compete about the last TIF bits on x86, so we need to clean that up first.
Quoted 16 lines
arch/x86/include/asm/thread_info.h +16
-0+static void trigger_sigsys (struct pt_regs *regs) +{ +struct kernel_siginfo info; + +clear_siginfo (&info); + info.si_signo =SIGSYS ; + info.si_code =SYS_USER_DISPATCH ; + info.si_call_addr = (void __user *)KSTK_EIP (current); + info.si_errno =0 ; + info.si_arch =syscall_get_arch (current); + info.si_syscall =syscall_get_nr (current, regs); + +force_sig_info (&info); +} + +int do_syscall_user_dispatch (struct pt_regs *regs)
arch/x86/include/asm/thread_info.h struct thread_info {
Quoted 10 lines
arch/x86/include/asm/thread_info.h +10
-0+{ +struct syscall_user_dispatch *sd = ¤t->syscall_dispatch; +unsigned long ip =instruction_pointer (regs); +char state; + +if (likely (ip >= sd->dispatcher_start && ip <= sd->dispatcher_end)) +return 0 ; + +if (likely (sd->selector)) { +if (unlikely (__get_user (state, sd->selector)))
__get_user() mandates an explicit access_ok() which happened in the prctl(). So this wants a comment why there is none right here.
arch/x86/include/asm/thread_info.h
Quoted 7 lines
arch/x86/include/asm/thread_info.h +7
-0+do_exit (SIGSEGV ); + +if (likely (state ==0 )) +return 0 ; + +if (state !=1 ) +do_exit (SIGSEGV );
If that happens its going to be quite interesting to debug.
Also please use proper defines which are exposed to user space instead of 0/1.
Quoted 16 lines
arch/x86/include/asm/thread_info.h +16
-0+ } + +syscall_rollback (current, regs); +trigger_sigsys (regs); + +return 1 ; +} + +int set_syscall_user_dispatch (int mode,unsigned long dispatcher_start, +unsigned long dispatcher_end,char __user *selector) +{ +switch (mode) { +case PR_SYS_DISPATCH_OFF : +if (dispatcher_start || dispatcher_end || selector) +return -EINVAL ; +break ;
arch/x86/include/asm/thread_info.h
Quoted 10 lines
arch/x86/include/asm/thread_info.h +10
-0+case PR_SYS_DISPATCH_ON : +/* +* Validate the direct dispatcher region just for basic +* sanity. If the user is able to submit a syscall from +* an address, that address is obviously valid. +*/ +if (dispatcher_end < dispatcher_start) +return -EINVAL ; + +if (selector && !access_ok (selector,1 ))
sizeof(*selector)
Thanks,
tglx
Re: [PATCH v4 1/2] kernel: Implement selective syscall userspace redirection
Hi Thomas,
Thanks for the valuable feedback!
Quoted 20 lines · Thomas Gleixner
Quoted 10 lines · Gabriel Krisman Bertazi
Introduce a mechanism to quickly disable/enable syscall handling for a specific process and redirect to userspace via SIGSYS. This is useful for processes with parts that require syscall redirection and parts that don't, but who need to perform this boundary crossing really fast, without paying the cost of a system call to reconfigure syscall handling on each boundary transition. This is particularly important for Windows games running over Wine.
The proposed interface looks like this:
Quoted 8 lines
prctl(PR_SET_SYSCALL_USER_DISPATCH, <op>, <start_addr>, <end_addr>, [selector])The range [<start_addr>,<end_addr>] is a part of the process memory map that is allowed to by-pass the redirection code and dispatch syscalls directly, such that in fast paths a process doesn't need to disable the trap nor the kernel has to check the selector. This is essential to return from SIGSYS to a blocked area without triggering another SIGSYS from
rt_sigreturn.
Quoted 2 lines
Why isn't
rt_sigreturn()exempt from that redirection in the first place?
This was actually a design decision for me.
The main use case I'm considering is emulation of applications written for other OSs (games over wine), which means this dispatcher code is exposed to applications built against different ABIs, who trigger syscalls with bogus parameters (from a linux perspective)
In this emulation scenario, I cannot really trust the syscall number means rt_sigreturn, so I try to only base the dispatcher decision on the memory region and selector variable.
I think the best we can do is what Andy said: to exempt rt_sigreturn when it comes from the vdso, for architectures that do it that way.
Quoted 23 lines
Quoted 16 lines
--- arch/Kconfig | 20 ++++++ arch/x86/Kconfig | 1 + arch/x86/entry/common.c | 5 ++ arch/x86/include/asm/thread_info.h | 4 +- arch/x86/kernel/signal_compat.c | 2 +- fs/exec.c | 2 + include/linux/sched.h | 3 + include/linux/syscall_user_dispatch.h | 50 +++++++++++++++ include/uapi/asm-generic/siginfo.h | 3 +- include/uapi/linux/prctl.h | 5 ++ kernel/Makefile | 1 + kernel/fork.c | 1 + kernel/sys.c | 5 ++ kernel/syscall_user_dispatch.c | 92 +++++++++++++++++++++++++++A big combo patch is not how we do that. Please split it up into the core part and a patch enabling it for a particular architexture.
As I said in my reply to Andy, this wants to go on top of the generic entry/exit work stuff:
Quoted 6 lines
https://lore.kernel.org/r/[email protected]
and then
syscall_user_dispatch.cends up inkernel/entry/and the dispatching function is not exposed outside of that directory.I'm going to post a new version later today. Will cc you.
Thanks. Will do!
Signature · 1 line
Gabriel Krisman Bertazi
PATCH v4 2/2 selftests: Add kselftest for syscall user dispatch
In reply to Gabriel Krisman BertaziImplement functionality tests for syscall user dispatch. In order to make the test portable, refrain from open coding syscall dispatchers and calculating glibc memory ranges.
changes since v3:
- Sort entry in Makefile
- Add SPDX header
- Use __NR_syscalls if available
--- tools/testing/selftests/Makefile | 1 + .../syscall_user_dispatch/.gitignore | 2 + .../selftests/syscall_user_dispatch/Makefile | 9 + .../selftests/syscall_user_dispatch/config | 1 + .../syscall_user_dispatch.c | 256 ++++++++++++++++++ 5 files changed, 269 insertions(+) create mode 100644 tools/testing/selftests/syscall_user_dispatch/.gitignore create mode 100644 tools/testing/selftests/syscall_user_dispatch/Makefile create mode 100644 tools/testing/selftests/syscall_user_dispatch/config create mode 100644 tools/testing/selftests/syscall_user_dispatch/syscall_user_dispatch.c
tools/testing/selftests/Makefile +1 -0
@@ -53,6 +53,7 @@ TARGETS += sparc64 TARGETS += splice TARGETS += static_keys TARGETS += sync +TARGETS += syscall_user_dispatch TARGETS += sysctl TARGETS += timens ifneq (1, $(quicktest))
tools/testing/selftests/syscall_user_dispatch/.gitignore new file +2 -0
@@ -0,0 +1,2 @@ +# SPDX-License-Identifier: GPL-2.0-only +syscall_user_dispatch
tools/testing/selftests/syscall_user_dispatch/Makefile new file +9 -0
@@ -0,0 +1,9 @@ +# SPDX-License-Identifier: GPL-2.0 +top_srcdir = ../../../.. +INSTALL_HDR_PATH = $(top_srcdir)/usr +LINUX_HDR_PATH = $(INSTALL_HDR_PATH)/include/ + +CFLAGS += -Wall -I$(LINUX_HDR_PATH) + +TEST_GEN_PROGS := syscall_user_dispatch +include ../lib.mk
tools/testing/selftests/syscall_user_dispatch/config new file +1 -0
@@ -0,0 +1 @@ +CONFIG_SYSCALL_USER_DISPATCH=y
tools/testing/selftests/syscall_user_dispatch/syscall_user_dispatch.c new file +256 -0
256 lines not on this page · raw message
Signature · 1 line
2.27.0
Quoted 11 lines · Gabriel Krisman Bertazi
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscallsis not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.Also, I didn't expose tests for
PR_GETas that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.
Thanks! That all looks good to me.
Quoted 1 line
Finally, one question: Which tree would this go through?
I haven't heard from several other x86 maintainers yet (which is where I would normally expect this series to land), but I would be comfortable taking this through my seccomp tree if I got Acks/Reviews at least from Andy and Matthew.
Andy, Matthew, what do you think of this?
Signature · 1 line
Kees Cook
Quoted 9 lines · Kees Cook
Quoted 8 lines · Gabriel Krisman Bertazi
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscallsis not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.
Quoted 5 lines
Quoted 3 lines
Also, I didn't expose tests for
PR_GETas that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.Thanks! That all looks good to me.
Don't have any problem with this but did this ever get exposure on linux-api? This is the first time I see this pop up.
Christian
Quoted 10 lines · Christian Brauner
Quoted 9 lines · Kees Cook
Quoted 8 lines · Gabriel Krisman Bertazi
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscallsis not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
Also, I didn't expose tests for
PR_GETas that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.Thanks! That all looks good to me.
Don't have any problem with this but did this ever get exposure on linux-api? This is the first time I see this pop up.
I thought I'd added it to CC in the past, but that might have been other recent unrelated threads. Does this need a full repost there too, you think?
Signature · 1 line
Kees Cook
Quoted 15 lines · Kees Cook
Quoted 14 lines · Christian Brauner
Quoted 13 lines · Kees Cook
Quoted 11 lines · Gabriel Krisman Bertazi
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscallsis not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.Also, I didn't expose tests for
PR_GETas that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.
Quoted 8 lines
Quoted 4 lines
Quoted 1 line
Thanks! That all looks good to me.
Don't have any problem with this but did this ever get exposure on linux-api? This is the first time I see this pop up.
I thought I'd added it to CC in the past, but that might have been other recent unrelated threads. Does this need a full repost there too, you think?
Nah, wasn't my intention to force a repost. Seems that several people have looked this over. :) Just curious why it didn't get to linux-api and we know quite some people who only do look at linux-api (for sanity). :)
Christian
Quoted 21 lines · Christian Brauner
Quoted 20 lines · Kees Cook
Quoted 18 lines · Christian Brauner
Quoted 9 lines · Kees Cook
Quoted 8 lines · Gabriel Krisman Bertazi
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscallsis not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.Quoted 5 lines
Quoted 3 lines
Also, I didn't expose tests for
PR_GETas that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.Thanks! That all looks good to me.
Don't have any problem with this but did this ever get exposure on linux-api? This is the first time I see this pop up.
Quoted 7 lines
Quoted 3 lines
I thought I'd added it to CC in the past, but that might have been other recent unrelated threads. Does this need a full repost there too, you think?
Nah, wasn't my intention to force a repost. Seems that several people have looked this over. :) Just curious why it didn't get to linux-api and we know quite some people who only do look at linux-api (for sanity). :)
That's my mistake. I didn't think about it when submitting :(
If this get re-spinned again I will make sure to CC linux-api.
Signature · 1 line
Gabriel Krisman Bertazi
Quoted 27 lines · Gabriel Krisman Bertazi
Quoted 25 lines · Christian Brauner
Quoted 23 lines · Kees Cook
Quoted 10 lines · Christian Brauner
Quoted 9 lines · Kees Cook
Quoted 8 lines · Gabriel Krisman Bertazi
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscallsis not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
Also, I didn't expose tests for
PR_GETas that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.Thanks! That all looks good to me.
Don't have any problem with this but did this ever get exposure on linux-api? This is the first time I see this pop up.
I thought I'd added it to CC in the past, but that might have been other recent unrelated threads. Does this need a full repost there too, you think?
Quoted 7 lines
Quoted 3 lines
Nah, wasn't my intention to force a repost. Seems that several people have looked this over. :) Just curious why it didn't get to linux-api and we know quite some people who only do look at linux-api (for sanity). :)
That's my mistake. I didn't think about it when submitting :(
If this get re-spinned again I will make sure to CC linux-api.
Thank you! It helps C library implementors stay up to date and comment on changes that impact userspace ABIs and APIs. This patch set was new to me. Interesting new feature.
Signature · 2 lines
Cheers, Carlos.
Hi!
Quoted 8 lines
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscallsis not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.
Quoted 5 lines
Also, I didn't expose tests for
PR_GETas that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.Finally, one question: Which tree would this go through?
Should it come with Documentation?
How does it interact with ptrace, seccomp and similar?
Pavel
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
Quoted 10 lines · Pavel Machek
Hi!
Quoted 8 lines
This is v4 of Syscall User Redirection. The implementation itself is not modified from v3, it only applies the latest round of reviews to the selftests.
__NR_syscallsis not really exported in header files other than asm-generic for every architecture, so it felt safer to optionally expose it with a fallback to a high value.
Quoted 7 lines
Quoted 5 lines
Also, I didn't expose tests for
PR_GETas that is not currently implemented. If possible, I'd have it supported by a future patchset, since it is not immediately necessary to support this feature.Finally, one question: Which tree would this go through?
Should it come with Documentation?
Hi,
Thanks for the review.
I will prepare it for the next iteration.
Quoted 1 line
How does it interact with ptrace, seccomp and similar?
That is a very good question.
Regarding seccomp, this must take precedence, since the use case is emulation (it can be invoked with a different ABI) such that seccomp filtering by syscall number doesn't make sense in the first place. In addition, either the syscall is dispatched back to userspace, in which case there is no resource for seccomp to protect, or the syscall will be executed, and seccomp will execute next.
Regarding ptrace, I experimented with before and after, and while the same ABI argument applies, I felt it was easier to debug if I let ptrace happen for syscalls that are dispatched back to userspace. In addition, doing it after ptrace makes the code in syscall_exit_work slightly simpler, since it doesn't require special handling for this feature.
For PTRACE_SYSEMU in particular, either placing this before or after is a bit odd. I don't think there is a right answer for this one, but I don't see anyone wanting to use these features at the same time. I think as long as it is documented behavior, it should be fine either way.
Signature · 1 line
Gabriel Krisman Bertazi