| Message ID | 20180325034024.31743-1-liwang@redhat.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | mprotect04: get the right function entry of big endian powerpc system | expand |
Li Wang <liwang@redhat.com> writes: > On Big endian powerpc ABI, function ptrs are basically pointers to > function descriptors. The testcase copies functions which results > in function descriptors getting copied. So easily the access was > denied by memory protection key in that address when performing it. > > 10000000-10020000 r-xp 00000000 fd:00 167223 mprotect04 > 10020000-10030000 r--p 00010000 fd:00 167223 mprotect04 > 10030000-10040000 rw-p 00020000 fd:00 167223 mprotect04 > 1001a380000-1001a3b0000 rw-p 00000000 00:00 0 [heap] > 7fffa6c60000-7fffa6c80000 --xp 00000000 00:00 0 > > &exec_func = 0x10030170 > &func = 0x7fffa6c60170 > > While perform the (*func)(); we get segmentation fault. > > strace log: > ----------- > mprotect(0x7fffaed00000, 131072, PROT_EXEC) = 0 > rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0 > --- SIGSEGV {si_signo=SIGSEGV, si_code=SEGV_PKUERR, si_addr=0x7fffaed00170} --- > > Reported-and-tested-by: Li Wang <liwang@redhat.com> > Signed-off-by: Ram Pai <linuxram@us.ibm.com> > Signed-off-by: Li Wang <liwang@redhat.com> > CC: Michael Ellerman <mpe@ellerman.id.au> Thanks for fixing this. Some comments below ... > diff --git a/testcases/kernel/syscalls/mprotect/mprotect04.c b/testcases/kernel/syscalls/mprotect/mprotect04.c > index 1173afd..5705cb0 100644 > --- a/testcases/kernel/syscalls/mprotect/mprotect04.c > +++ b/testcases/kernel/syscalls/mprotect/mprotect04.c > @@ -189,6 +191,14 @@ static void clear_cache(void *start, int len) > #endif > } > > +#if (defined (__powerpc__) || (__powerpc64__)) && __BYTE_ORDER == __BIG_ENDIAN That's not quite the right condition. 32-bit powerpc big endian doesn't use function descriptors, but your check for __powerpc__ will match on 32-bit. Also the check for __powerpc64__ is missing defined(). So this would work: #if defined(__powerpc64__) && __BYTE_ORDER == __BIG_ENDIAN More correct would be to check for the ABI version, because ppc64le can technically use function descriptors (it just doesn't in any configuration you're likely to see). That would be: #if defined(__powerpc64__) && (!defined(_CALL_ELF) || _CALL_ELF < 2) Given you need that in a few places you should probably make it a define, eg: #if defined(__powerpc64__) && (!defined(_CALL_ELF) || _CALL_ELF < 2) #define USE_FUNCTION_DESCRIPTORS #endif #ifdef USE_FUNCTION_DESCRIPTORS typedef struct { uintptr_t entry; uintptr_t toc; uintptr_t env; } func_descr_t; #endif cheers
On Mon, Mar 26, 2018 at 8:27 AM, Michael Ellerman <mpe@ellerman.id.au> wrote: > Li Wang <liwang@redhat.com> writes: > > > On Big endian powerpc ABI, function ptrs are basically pointers to > > function descriptors. The testcase copies functions which results > > in function descriptors getting copied. So easily the access was > > denied by memory protection key in that address when performing it. > > > > 10000000-10020000 r-xp 00000000 fd:00 167223 mprotect04 > > 10020000-10030000 r--p 00010000 fd:00 167223 mprotect04 > > 10030000-10040000 rw-p 00020000 fd:00 167223 mprotect04 > > 1001a380000-1001a3b0000 rw-p 00000000 00:00 0 [heap] > > 7fffa6c60000-7fffa6c80000 --xp 00000000 00:00 0 > > > > &exec_func = 0x10030170 > > &func = 0x7fffa6c60170 > > > > While perform the (*func)(); we get segmentation fault. > > > > strace log: > > ----------- > > mprotect(0x7fffaed00000, 131072, PROT_EXEC) = 0 > > rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0 > > --- SIGSEGV {si_signo=SIGSEGV, si_code=SEGV_PKUERR, > si_addr=0x7fffaed00170} --- > > > > Reported-and-tested-by: Li Wang <liwang@redhat.com> > > Signed-off-by: Ram Pai <linuxram@us.ibm.com> > > Signed-off-by: Li Wang <liwang@redhat.com> > > CC: Michael Ellerman <mpe@ellerman.id.au> > > Thanks for fixing this. > > Some comments below ... > > > diff --git a/testcases/kernel/syscalls/mprotect/mprotect04.c > b/testcases/kernel/syscalls/mprotect/mprotect04.c > > index 1173afd..5705cb0 100644 > > --- a/testcases/kernel/syscalls/mprotect/mprotect04.c > > +++ b/testcases/kernel/syscalls/mprotect/mprotect04.c > > @@ -189,6 +191,14 @@ static void clear_cache(void *start, int len) > > #endif > > } > > > > +#if (defined (__powerpc__) || (__powerpc64__)) && __BYTE_ORDER == > __BIG_ENDIAN > > That's not quite the right condition. > > 32-bit powerpc big endian doesn't use function descriptors, but your > check for __powerpc__ will match on 32-bit. > Also the check for __powerpc64__ is missing defined(). > > So this would work: > > #if defined(__powerpc64__) && __BYTE_ORDER == __BIG_ENDIAN > > More correct would be to check for the ABI version, because ppc64le can > technically use function descriptors (it just doesn't in any > configuration you're likely to see). > Thanks for your correction, very detail and clear. I'd like to take use of it in PATCH V2 directly. > > That would be: > > #if defined(__powerpc64__) && (!defined(_CALL_ELF) || _CALL_ELF < 2) > > Given you need that in a few places you should probably make it a > define, eg: > > #if defined(__powerpc64__) && (!defined(_CALL_ELF) || _CALL_ELF < 2) > #define USE_FUNCTION_DESCRIPTORS > #endif > > #ifdef USE_FUNCTION_DESCRIPTORS > typedef struct { > uintptr_t entry; > uintptr_t toc; > uintptr_t env; > } func_descr_t; > #endif > > > cheers > >
diff --git a/testcases/kernel/syscalls/mprotect/mprotect04.c b/testcases/kernel/syscalls/mprotect/mprotect04.c index 1173afd..5705cb0 100644 --- a/testcases/kernel/syscalls/mprotect/mprotect04.c +++ b/testcases/kernel/syscalls/mprotect/mprotect04.c @@ -35,6 +35,7 @@ #include <string.h> #include <sys/mman.h> #include <stdlib.h> +#include <endian.h> #include "test.h" #include "safe_macros.h" @@ -56,6 +57,7 @@ int TST_TOTAL = ARRAY_SIZE(testfunc); static volatile int sig_caught; static sigjmp_buf env; static unsigned int copy_sz; +typedef void (*func_ptr_t)(void); int main(int ac, char **av) { @@ -189,6 +191,14 @@ static void clear_cache(void *start, int len) #endif } +#if (defined (__powerpc__) || (__powerpc64__)) && __BYTE_ORDER == __BIG_ENDIAN +typedef struct { + uintptr_t entry; + uintptr_t toc; + uintptr_t env; +} func_descr_t; +#endif + /* * Copy page where &exec_func resides. Also try to copy subsequent page * in case exec_func is close to page boundary. @@ -197,11 +207,21 @@ static void *get_func(void *mem) { uintptr_t page_sz = getpagesize(); uintptr_t page_mask = ~(page_sz - 1); - uintptr_t func_page_offset = (uintptr_t)&exec_func & (page_sz - 1); - void *func_copy_start = mem + func_page_offset; - void *page_to_copy = (void *)((uintptr_t)&exec_func & page_mask); + uintptr_t func_page_offset; + void *func_copy_start, *page_to_copy; void *mem_start = mem; +#if (defined (__powerpc__) || (__powerpc64__)) && __BYTE_ORDER == __BIG_ENDIAN + func_descr_t *opd = (func_descr_t *)&exec_func; + func_page_offset = (uintptr_t)opd->entry & (page_sz - 1); + func_copy_start = mem + func_page_offset; + page_to_copy = (void *)((uintptr_t)opd->entry & page_mask); +#else + func_page_offset = (uintptr_t)&exec_func & (page_sz - 1); + func_copy_start = mem + func_page_offset; + page_to_copy = (void *)((uintptr_t)&exec_func & page_mask); +#endif + /* copy 1st page, if it's not present something is wrong */ if (!page_present(page_to_copy)) { tst_resm(TINFO, "exec_func: %p, page_to_copy: %p\n", @@ -228,7 +248,7 @@ static void *get_func(void *mem) static void testfunc_protexec(void) { - void (*func)(void); + func_ptr_t func; void *p; sig_caught = 0; @@ -236,7 +256,13 @@ static void testfunc_protexec(void) p = SAFE_MMAP(cleanup, 0, copy_sz, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); +#if (defined (__powerpc__) || (__powerpc64__)) && __BYTE_ORDER == __BIG_ENDIAN + func_descr_t opd; + opd.entry = (uintptr_t)get_func(p); + func = (func_ptr_t)&opd; +#else func = get_func(p); +#endif /* Change the protection to PROT_EXEC. */ TEST(mprotect(p, copy_sz, PROT_EXEC));
On Big endian powerpc ABI, function ptrs are basically pointers to function descriptors. The testcase copies functions which results in function descriptors getting copied. So easily the access was denied by memory protection key in that address when performing it. 10000000-10020000 r-xp 00000000 fd:00 167223 mprotect04 10020000-10030000 r--p 00010000 fd:00 167223 mprotect04 10030000-10040000 rw-p 00020000 fd:00 167223 mprotect04 1001a380000-1001a3b0000 rw-p 00000000 00:00 0 [heap] 7fffa6c60000-7fffa6c80000 --xp 00000000 00:00 0 &exec_func = 0x10030170 &func = 0x7fffa6c60170 While perform the (*func)(); we get segmentation fault. strace log: ----------- mprotect(0x7fffaed00000, 131072, PROT_EXEC) = 0 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0 --- SIGSEGV {si_signo=SIGSEGV, si_code=SEGV_PKUERR, si_addr=0x7fffaed00170} --- Reported-and-tested-by: Li Wang <liwang@redhat.com> Signed-off-by: Ram Pai <linuxram@us.ibm.com> Signed-off-by: Li Wang <liwang@redhat.com> CC: Michael Ellerman <mpe@ellerman.id.au> --- testcases/kernel/syscalls/mprotect/mprotect04.c | 34 ++++++++++++++++++++++--- 1 file changed, 30 insertions(+), 4 deletions(-)