From 0051202f6ad5fd9c04d220343e66d1eb890f7b81 Mon Sep 17 00:00:00 2001 From: Andy Lutomirski Date: Thu, 7 Apr 2016 17:31:44 -0700 Subject: [PATCH] selftests/x86: Test the FSBASE/GSBASE API and context switching This catches two distinct bugs in the current code. I'll fix them. Signed-off-by: Andy Lutomirski Reviewed-by: Borislav Petkov Cc: Andy Lutomirski Cc: Borislav Petkov Cc: Brian Gerst Cc: Denys Vlasenko Cc: H. Peter Anvin Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Rudolf Marek Cc: Thomas Gleixner Link: http://lkml.kernel.org/r/7e5941148d1e2199f070dadcdf7355959f5f8e85.1460075211.git.luto@kernel.org Signed-off-by: Ingo Molnar --- tools/testing/selftests/x86/Makefile | 1 + tools/testing/selftests/x86/fsgsbase.c | 398 +++++++++++++++++++++++++ 2 files changed, 399 insertions(+) create mode 100644 tools/testing/selftests/x86/fsgsbase.c diff --git a/tools/testing/selftests/x86/Makefile b/tools/testing/selftests/x86/Makefile index b47ebd170690..c73425de3cfe 100644 --- a/tools/testing/selftests/x86/Makefile +++ b/tools/testing/selftests/x86/Makefile @@ -9,6 +9,7 @@ TARGETS_C_BOTHBITS := single_step_syscall sysret_ss_attrs syscall_nt ptrace_sysc TARGETS_C_32BIT_ONLY := entry_from_vm86 syscall_arg_fault test_syscall_vdso unwind_vdso \ test_FCMOV test_FCOMI test_FISTTP \ vdso_restorer +TARGETS_C_64BIT_ONLY := fsgsbase TARGETS_C_32BIT_ALL := $(TARGETS_C_BOTHBITS) $(TARGETS_C_32BIT_ONLY) TARGETS_C_64BIT_ALL := $(TARGETS_C_BOTHBITS) $(TARGETS_C_64BIT_ONLY) diff --git a/tools/testing/selftests/x86/fsgsbase.c b/tools/testing/selftests/x86/fsgsbase.c new file mode 100644 index 000000000000..5b2b4b3c634c --- /dev/null +++ b/tools/testing/selftests/x86/fsgsbase.c @@ -0,0 +1,398 @@ +/* + * fsgsbase.c, an fsgsbase test + * Copyright (c) 2014-2016 Andy Lutomirski + * GPL v2 + */ + +#define _GNU_SOURCE +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#ifndef __x86_64__ +# error This test is 64-bit only +#endif + +static volatile sig_atomic_t want_segv; +static volatile unsigned long segv_addr; + +static int nerrs; + +static void sethandler(int sig, void (*handler)(int, siginfo_t *, void *), + int flags) +{ + struct sigaction sa; + memset(&sa, 0, sizeof(sa)); + sa.sa_sigaction = handler; + sa.sa_flags = SA_SIGINFO | flags; + sigemptyset(&sa.sa_mask); + if (sigaction(sig, &sa, 0)) + err(1, "sigaction"); +} + +static void clearhandler(int sig) +{ + struct sigaction sa; + memset(&sa, 0, sizeof(sa)); + sa.sa_handler = SIG_DFL; + sigemptyset(&sa.sa_mask); + if (sigaction(sig, &sa, 0)) + err(1, "sigaction"); +} + +static void sigsegv(int sig, siginfo_t *si, void *ctx_void) +{ + ucontext_t *ctx = (ucontext_t*)ctx_void; + + if (!want_segv) { + clearhandler(SIGSEGV); + return; /* Crash cleanly. */ + } + + want_segv = false; + segv_addr = (unsigned long)si->si_addr; + + ctx->uc_mcontext.gregs[REG_RIP] += 4; /* Skip the faulting mov */ + +} + +enum which_base { FS, GS }; + +static unsigned long read_base(enum which_base which) +{ + unsigned long offset; + /* + * Unless we have FSGSBASE, there's no direct way to do this from + * user mode. We can get at it indirectly using signals, though. + */ + + want_segv = true; + + offset = 0; + if (which == FS) { + /* Use a constant-length instruction here. */ + asm volatile ("mov %%fs:(%%rcx), %%rax" : : "c" (offset) : "rax"); + } else { + asm volatile ("mov %%gs:(%%rcx), %%rax" : : "c" (offset) : "rax"); + } + if (!want_segv) + return segv_addr + offset; + + /* + * If that didn't segfault, try the other end of the address space. + * Unless we get really unlucky and run into the vsyscall page, this + * is guaranteed to segfault. + */ + + offset = (ULONG_MAX >> 1) + 1; + if (which == FS) { + asm volatile ("mov %%fs:(%%rcx), %%rax" + : : "c" (offset) : "rax"); + } else { + asm volatile ("mov %%gs:(%%rcx), %%rax" + : : "c" (offset) : "rax"); + } + if (!want_segv) + return segv_addr + offset; + + abort(); +} + +static void check_gs_value(unsigned long value) +{ + unsigned long base; + unsigned short sel; + + printf("[RUN]\tARCH_SET_GS to 0x%lx\n", value); + if (syscall(SYS_arch_prctl, ARCH_SET_GS, value) != 0) + err(1, "ARCH_SET_GS"); + + asm volatile ("mov %%gs, %0" : "=rm" (sel)); + base = read_base(GS); + if (base == value) { + printf("[OK]\tGSBASE was set as expected (selector 0x%hx)\n", + sel); + } else { + nerrs++; + printf("[FAIL]\tGSBASE was not as expected: got 0x%lx (selector 0x%hx)\n", + base, sel); + } + + if (syscall(SYS_arch_prctl, ARCH_GET_GS, &base) != 0) + err(1, "ARCH_GET_GS"); + if (base == value) { + printf("[OK]\tARCH_GET_GS worked as expected (selector 0x%hx)\n", + sel); + } else { + nerrs++; + printf("[FAIL]\tARCH_GET_GS was not as expected: got 0x%lx (selector 0x%hx)\n", + base, sel); + } +} + +static void mov_0_gs(unsigned long initial_base, bool schedule) +{ + unsigned long base, arch_base; + + printf("[RUN]\tARCH_SET_GS to 0x%lx then mov 0 to %%gs%s\n", initial_base, schedule ? " and schedule " : ""); + if (syscall(SYS_arch_prctl, ARCH_SET_GS, initial_base) != 0) + err(1, "ARCH_SET_GS"); + + if (schedule) + usleep(10); + + asm volatile ("mov %0, %%gs" : : "rm" (0)); + base = read_base(GS); + if (syscall(SYS_arch_prctl, ARCH_GET_GS, &arch_base) != 0) + err(1, "ARCH_GET_GS"); + if (base == arch_base) { + printf("[OK]\tGSBASE is 0x%lx\n", base); + } else { + nerrs++; + printf("[FAIL]\tGSBASE changed to 0x%lx but kernel reports 0x%lx\n", base, arch_base); + } +} + +static volatile unsigned long remote_base; +static volatile bool remote_hard_zero; +static volatile unsigned int ftx; + +/* + * ARCH_SET_FS/GS(0) may or may not program a selector of zero. HARD_ZERO + * means to force the selector to zero to improve test coverage. + */ +#define HARD_ZERO 0xa1fa5f343cb85fa4 + +static void do_remote_base() +{ + unsigned long to_set = remote_base; + bool hard_zero = false; + if (to_set == HARD_ZERO) { + to_set = 0; + hard_zero = true; + } + + if (syscall(SYS_arch_prctl, ARCH_SET_GS, to_set) != 0) + err(1, "ARCH_SET_GS"); + + if (hard_zero) + asm volatile ("mov %0, %%gs" : : "rm" ((unsigned short)0)); + + unsigned short sel; + asm volatile ("mov %%gs, %0" : "=rm" (sel)); + printf("\tother thread: ARCH_SET_GS(0x%lx)%s -- sel is 0x%hx\n", + to_set, hard_zero ? " and clear gs" : "", sel); +} + +void do_unexpected_base(void) +{ + /* + * The goal here is to try to arrange for GS == 0, GSBASE != + * 0, and for the the kernel the think that GSBASE == 0. + * + * To make the test as reliable as possible, this uses + * explicit descriptorss. (This is not the only way. This + * could use ARCH_SET_GS with a low, nonzero base, but the + * relevant side effect of ARCH_SET_GS could change.) + */ + + /* Step 1: tell the kernel that we have GSBASE == 0. */ + if (syscall(SYS_arch_prctl, ARCH_SET_GS, 0) != 0) + err(1, "ARCH_SET_GS"); + + /* Step 2: change GSBASE without telling the kernel. */ + struct user_desc desc = { + .entry_number = 0, + .base_addr = 0xBAADF00D, + .limit = 0xfffff, + .seg_32bit = 1, + .contents = 0, /* Data, grow-up */ + .read_exec_only = 0, + .limit_in_pages = 1, + .seg_not_present = 0, + .useable = 0 + }; + if (syscall(SYS_modify_ldt, 1, &desc, sizeof(desc)) == 0) { + printf("\tother thread: using LDT slot 0\n"); + asm volatile ("mov %0, %%gs" : : "rm" ((unsigned short)0x7)); + } else { + /* No modify_ldt for us (configured out, perhaps) */ + + struct user_desc *low_desc = mmap( + NULL, sizeof(desc), + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS | MAP_32BIT, -1, 0); + memcpy(low_desc, &desc, sizeof(desc)); + + low_desc->entry_number = -1; + + /* 32-bit set_thread_area */ + long ret; + asm volatile ("int $0x80" + : "=a" (ret) : "a" (243), "b" (low_desc) + : "flags"); + memcpy(&desc, low_desc, sizeof(desc)); + munmap(low_desc, sizeof(desc)); + + if (ret != 0) { + printf("[NOTE]\tcould not create a segment -- test won't do anything\n"); + return; + } + printf("\tother thread: using GDT slot %d\n", desc.entry_number); + asm volatile ("mov %0, %%gs" : : "rm" ((unsigned short)((desc.entry_number << 3) | 0x3))); + } + + /* + * Step 3: set the selector back to zero. On AMD chips, this will + * preserve GSBASE. + */ + + asm volatile ("mov %0, %%gs" : : "rm" ((unsigned short)0)); +} + +static void *threadproc(void *ctx) +{ + while (1) { + while (ftx == 0) + syscall(SYS_futex, &ftx, FUTEX_WAIT, 0, NULL, NULL, 0); + if (ftx == 3) + return NULL; + + if (ftx == 1) + do_remote_base(); + else if (ftx == 2) + do_unexpected_base(); + else + errx(1, "helper thread got bad command"); + + ftx = 0; + syscall(SYS_futex, &ftx, FUTEX_WAKE, 0, NULL, NULL, 0); + } +} + +static void set_gs_and_switch_to(unsigned long local, unsigned long remote) +{ + unsigned long base; + + bool hard_zero = false; + if (local == HARD_ZERO) { + hard_zero = true; + local = 0; + } + + printf("[RUN]\tARCH_SET_GS(0x%lx)%s, then schedule to 0x%lx\n", + local, hard_zero ? " and clear gs" : "", remote); + if (syscall(SYS_arch_prctl, ARCH_SET_GS, local) != 0) + err(1, "ARCH_SET_GS"); + if (hard_zero) + asm volatile ("mov %0, %%gs" : : "rm" ((unsigned short)0)); + + if (read_base(GS) != local) { + nerrs++; + printf("[FAIL]\tGSBASE wasn't set as expected\n"); + } + + remote_base = remote; + ftx = 1; + syscall(SYS_futex, &ftx, FUTEX_WAKE, 0, NULL, NULL, 0); + while (ftx != 0) + syscall(SYS_futex, &ftx, FUTEX_WAIT, 1, NULL, NULL, 0); + + base = read_base(GS); + if (base == local) { + printf("[OK]\tGSBASE remained 0x%lx\n", local); + } else { + nerrs++; + printf("[FAIL]\tGSBASE changed to 0x%lx\n", base); + } +} + +static void test_unexpected_base(void) +{ + unsigned long base; + + printf("[RUN]\tARCH_SET_GS(0), clear gs, then manipulate GSBASE in a different thread\n"); + if (syscall(SYS_arch_prctl, ARCH_SET_GS, 0) != 0) + err(1, "ARCH_SET_GS"); + asm volatile ("mov %0, %%gs" : : "rm" ((unsigned short)0)); + + ftx = 2; + syscall(SYS_futex, &ftx, FUTEX_WAKE, 0, NULL, NULL, 0); + while (ftx != 0) + syscall(SYS_futex, &ftx, FUTEX_WAIT, 1, NULL, NULL, 0); + + base = read_base(GS); + if (base == 0) { + printf("[OK]\tGSBASE remained 0\n"); + } else { + nerrs++; + printf("[FAIL]\tGSBASE changed to 0x%lx\n", base); + } +} + +int main() +{ + pthread_t thread; + + sethandler(SIGSEGV, sigsegv, 0); + + check_gs_value(0); + check_gs_value(1); + check_gs_value(0x200000000); + check_gs_value(0); + check_gs_value(0x200000000); + check_gs_value(1); + + for (int sched = 0; sched < 2; sched++) { + mov_0_gs(0, !!sched); + mov_0_gs(1, !!sched); + mov_0_gs(0x200000000, !!sched); + } + + /* Set up for multithreading. */ + + cpu_set_t cpuset; + CPU_ZERO(&cpuset); + CPU_SET(0, &cpuset); + if (sched_setaffinity(0, sizeof(cpuset), &cpuset) != 0) + err(1, "sched_setaffinity to CPU 0"); /* should never fail */ + + if (pthread_create(&thread, 0, threadproc, 0) != 0) + err(1, "pthread_create"); + + static unsigned long bases_with_hard_zero[] = { + 0, HARD_ZERO, 1, 0x200000000, + }; + + for (int local = 0; local < 4; local++) { + for (int remote = 0; remote < 4; remote++) { + set_gs_and_switch_to(bases_with_hard_zero[local], + bases_with_hard_zero[remote]); + } + } + + test_unexpected_base(); + + ftx = 3; /* Kill the thread. */ + syscall(SYS_futex, &ftx, FUTEX_WAKE, 0, NULL, NULL, 0); + + if (pthread_join(thread, NULL) != 0) + err(1, "pthread_join"); + + return nerrs == 0 ? 0 : 1; +} -- 2.20.1