| Message ID | 20190301001806.154271-1-smuckle@google.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | selftests: x86: add version check in test_syscall_vdso | expand |
> On Feb 28, 2019, at 4:18 PM, Steve Muckle <smuckle@google.com> wrote: > > Since 4.17 registers r8-r11 are not clobbered/zeroed by a 64-bit kernel > handling a 32-bit syscall and this behavior is enforced by the > test_syscall_vdso testcase. See commit 8bb2610bc496 > ("x86/entry/64/compat: Preserve r8-r11 in int $0x80"). > > Permit the old behavior in the testcase for kernels prior to 4.17. NAK. If you want an old buggy kernel to pass a test, please either patch the kernel or run an old test. > > Signed-off-by: Steve Muckle <smuckle@google.com> > --- > .../testing/selftests/x86/test_syscall_vdso.c | 30 +++++++++++++++++++ > 1 file changed, 30 insertions(+) > > diff --git a/tools/testing/selftests/x86/test_syscall_vdso.c b/tools/testing/selftests/x86/test_syscall_vdso.c > index c9c3281077bc..f7284dc4c32b 100644 > --- a/tools/testing/selftests/x86/test_syscall_vdso.c > +++ b/tools/testing/selftests/x86/test_syscall_vdso.c > @@ -30,6 +30,7 @@ > #include <sys/time.h> > #include <elf.h> > #include <sys/ptrace.h> > +#include <sys/utsname.h> > #include <sys/wait.h> > > #if !defined(__i386__) > @@ -71,6 +72,7 @@ struct regs64 { > }; > struct regs64 regs64; > int kernel_is_64bit; > +int clobber_ok; > > asm ( > " .pushsection .text\n" > @@ -130,6 +132,28 @@ void print_regs64(void) > printf("12:%016llx 13:%016llx 14:%016llx 15:%016llx\n", regs64.r12, regs64.r13, regs64.r14, regs64.r15); > } > > +static void get_kernel_version(int *version, int *patchlevel) > +{ > + int ret, sublevel; > + struct utsname utsname; > + > + ret = uname(&utsname); > + if (ret) { > + perror("uname"); > + exit(1); > + } > + > + ret = sscanf(utsname.release, "%d.%d.%d", version, patchlevel, > + &sublevel); > + if (ret < 0) { > + perror("sscanf"); > + exit(1); > + } else if (ret != 3) { > + printf("Malformed kernel version %s\n", &utsname.release); > + exit(1); > + } > +} > + > int check_regs64(void) > { > int err = 0; > @@ -166,6 +190,8 @@ int check_regs64(void) > * Historically (and probably unintentionally), they > * were clobbered or zeroed. > */ > + if (clobber_ok && *r64 == 0 && num <= 11) > + continue; > } > printf("[FAIL]\tR%d has changed:%016llx\n", num, *r64); > err++; > @@ -385,6 +411,7 @@ int main(int argc, char **argv, char **envp) > { > int exitcode = 0; > int cs; > + int version, patchlevel; > > asm("\n" > " movl %%cs, %%eax\n" > @@ -394,6 +421,9 @@ int main(int argc, char **argv, char **envp) > if (!kernel_is_64bit) > printf("[NOTE]\tNot a 64-bit kernel, won't test R8..R15 leaks\n"); > > + get_kernel_version(&version, &patchlevel); > + clobber_ok = version < 4 || (version == 4 && patchlevel < 17); > + > /* This only works for non-static builds: > * syscall_addr = dlsym(dlopen("linux-gate.so.1", RTLD_NOW), "__kernel_vsyscall"); > */ > -- > 2.21.0.352.gf09ad66450-goog >
On Fri, Mar 01, 2019 at 11:59:24AM -0800, Andy Lutomirski wrote: > > On Feb 28, 2019, at 4:18 PM, Steve Muckle <smuckle@google.com> wrote: > > > > Since 4.17 registers r8-r11 are not clobbered/zeroed by a 64-bit kernel > > handling a 32-bit syscall and this behavior is enforced by the > > test_syscall_vdso testcase. See commit 8bb2610bc496 > > ("x86/entry/64/compat: Preserve r8-r11 in int $0x80"). > > > > Permit the old behavior in the testcase for kernels prior to 4.17. > > NAK. If you want an old buggy kernel to pass a test, please either > patch the kernel or run an old test. Yeah, this isn't ok, especially as you might have 4.14 or 4.9 kernels that can pass the test if the needed patches were backported. Checking kernel version numbers should never be used for anything. greg k-h
On 3/2/19 2:10 AM, Greg KH wrote: > On Fri, Mar 01, 2019 at 11:59:24AM -0800, Andy Lutomirski wrote: >>> On Feb 28, 2019, at 4:18 PM, Steve Muckle <smuckle@google.com> wrote: >>> >>> Since 4.17 registers r8-r11 are not clobbered/zeroed by a 64-bit kernel >>> handling a 32-bit syscall and this behavior is enforced by the >>> test_syscall_vdso testcase. See commit 8bb2610bc496 >>> ("x86/entry/64/compat: Preserve r8-r11 in int $0x80"). >>> >>> Permit the old behavior in the testcase for kernels prior to 4.17. >> >> NAK. If you want an old buggy kernel to pass a test, please either >> patch the kernel or run an old test. > > Yeah, this isn't ok, especially as you might have 4.14 or 4.9 kernels > that can pass the test if the needed patches were backported. Checking > kernel version numbers should never be used for anything. > Also, this test failure can tell us if we should backport important fixes. I have had requests to add kernel version checking in tests before and the recommendation is "don't check kernel versions and reply on feature type checks". In the case of fixes like this one, it is better for the test to fail than make us feel warm and fuzzy by making it pass. thanks, -- Shuah
diff --git a/tools/testing/selftests/x86/test_syscall_vdso.c b/tools/testing/selftests/x86/test_syscall_vdso.c index c9c3281077bc..f7284dc4c32b 100644 --- a/tools/testing/selftests/x86/test_syscall_vdso.c +++ b/tools/testing/selftests/x86/test_syscall_vdso.c @@ -30,6 +30,7 @@ #include <sys/time.h> #include <elf.h> #include <sys/ptrace.h> +#include <sys/utsname.h> #include <sys/wait.h> #if !defined(__i386__) @@ -71,6 +72,7 @@ struct regs64 { }; struct regs64 regs64; int kernel_is_64bit; +int clobber_ok; asm ( " .pushsection .text\n" @@ -130,6 +132,28 @@ void print_regs64(void) printf("12:%016llx 13:%016llx 14:%016llx 15:%016llx\n", regs64.r12, regs64.r13, regs64.r14, regs64.r15); } +static void get_kernel_version(int *version, int *patchlevel) +{ + int ret, sublevel; + struct utsname utsname; + + ret = uname(&utsname); + if (ret) { + perror("uname"); + exit(1); + } + + ret = sscanf(utsname.release, "%d.%d.%d", version, patchlevel, + &sublevel); + if (ret < 0) { + perror("sscanf"); + exit(1); + } else if (ret != 3) { + printf("Malformed kernel version %s\n", &utsname.release); + exit(1); + } +} + int check_regs64(void) { int err = 0; @@ -166,6 +190,8 @@ int check_regs64(void) * Historically (and probably unintentionally), they * were clobbered or zeroed. */ + if (clobber_ok && *r64 == 0 && num <= 11) + continue; } printf("[FAIL]\tR%d has changed:%016llx\n", num, *r64); err++; @@ -385,6 +411,7 @@ int main(int argc, char **argv, char **envp) { int exitcode = 0; int cs; + int version, patchlevel; asm("\n" " movl %%cs, %%eax\n" @@ -394,6 +421,9 @@ int main(int argc, char **argv, char **envp) if (!kernel_is_64bit) printf("[NOTE]\tNot a 64-bit kernel, won't test R8..R15 leaks\n"); + get_kernel_version(&version, &patchlevel); + clobber_ok = version < 4 || (version == 4 && patchlevel < 17); + /* This only works for non-static builds: * syscall_addr = dlsym(dlopen("linux-gate.so.1", RTLD_NOW), "__kernel_vsyscall"); */
Since 4.17 registers r8-r11 are not clobbered/zeroed by a 64-bit kernel handling a 32-bit syscall and this behavior is enforced by the test_syscall_vdso testcase. See commit 8bb2610bc496 ("x86/entry/64/compat: Preserve r8-r11 in int $0x80"). Permit the old behavior in the testcase for kernels prior to 4.17. Signed-off-by: Steve Muckle <smuckle@google.com> --- .../testing/selftests/x86/test_syscall_vdso.c | 30 +++++++++++++++++++ 1 file changed, 30 insertions(+)