/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ /* vim: set ts=8 sts=2 et sw=2 tw=80: */ /* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ /* Code in this file needs to be kept in sync with code in nsPresArena.cpp. * * We want to use a fixed address for frame poisoning so that it is readily * identifiable in crash dumps. Whether such an address is available * without any special setup depends on the system configuration. * * All current 64-bit CPUs (with the possible exception of PowerPC64) * reserve the vast majority of the virtual address space for future * hardware extensions; valid addresses must be below some break point * between 2**48 and 2**54, depending on exactly which chip you have. Some * chips (notably amd64) also allow the use of the *highest* 2**48 -- 2**54 * addresses. Thus, if user space pointers are 64 bits wide, we can just * use an address outside this range, and no more is required. To * accommodate the chips that allow very high addresses to be valid, the * value chosen is close to 2**63 (that is, in the middle of the space). * * In most cases, a purely 32-bit operating system must reserve some * fraction of the address space for its own use. Contemporary 32-bit OSes * tend to take the high gigabyte or so (0xC000_0000 on up). If we can * prove that high addresses are reserved to the kernel, we can use an * address in that region. Unfortunately, not all 32-bit OSes do this; * OSX 10.4 might not, and it is unclear what mobile OSes are like * (some 32-bit CPUs make it very easy for the kernel to exist in its own * private address space). * * Furthermore, when a 32-bit user space process is running on a 64-bit * kernel, the operating system has no need to reserve any of the space that * the process can see, and generally does not do so. This is the scenario * of greatest concern, since it covers all contemporary OSX iterations * (10.5+) as well as Windows Vista and 7 on newer amd64 hardware. Linux on * amd64 is generally run as a pure 64-bit environment, but its 32-bit * compatibility mode also has this property. * * Thus, when user space pointers are 32 bits wide, we need to validate * our chosen address, and possibly *make* it a good poison address by * allocating a page around it and marking it inaccessible. The algorithm * for this is: * * 1. Attempt to make the page surrounding the poison address a reserved, * inaccessible memory region using OS primitives. On Windows, this is * done with VirtualAlloc(MEM_RESERVE); on Unix, mmap(PROT_NONE). * * 2. If mmap/VirtualAlloc failed, there are two possible reasons: either * the region is reserved to the kernel and no further action is * required, or there is already usable memory in this area and we have * to pick a different address. The tricky part is knowing which case * we have, without attempting to access the region. On Windows, we * rely on GetSystemInfo()'s reported upper and lower bounds of the * application memory area. On Unix, there is nothing devoted to the * purpose, but seeing if madvise() fails is close enough (it *might* * disrupt someone else's use of the memory region, but not by as much * as anything else available). * * Be aware of these gotchas: * * 1. We cannot use mmap() with MAP_FIXED. MAP_FIXED is defined to * _replace_ any existing mapping in the region, if necessary to satisfy * the request. Obviously, as we are blindly attempting to acquire a * page at a constant address, we must not do this, lest we overwrite * someone else's allocation. * * 2. For the same reason, we cannot blindly use mprotect() if mmap() fails. * * 3. madvise() may fail when applied to a 'magic' memory region provided as * a kernel/user interface. Fortunately, the only such case I know about * is the "vsyscall" area (not to be confused with the "vdso" area) for * *64*-bit processes on Linux - and we don't even run this code for * 64-bit processes. * * 4. VirtualQuery() does not produce any useful information if * applied to kernel memory - in fact, it doesn't write its output * at all. Thus, it is not used here. */ // MAP_ANON(YMOUS) is not in any standard. Add defines as necessary. #define _GNU_SOURCE 1 #define _DARWIN_C_SOURCE 1 #include #include #include #include #include #ifdef _WIN32 # include #else # include # include # include # include # ifndef MAP_ANON # ifdef MAP_ANONYMOUS # define MAP_ANON MAP_ANONYMOUS # else # error "Don't know how to get anonymous memory" # endif # endif #endif #define SIZxPTR ((int)(sizeof(uintptr_t) * 2)) /* This program assumes that a whole number of return instructions fit into * 32 bits, and that 32-bit alignment is sufficient for a branch destination. * For architectures where this is not true, fiddling with RETURN_INSTR_TYPE * can be enough. */ #if defined __i386__ || defined __x86_64__ || defined __i386 || \ defined __x86_64 || defined _M_IX86 || defined _M_AMD64 # define RETURN_INSTR 0xC3C3C3C3 /* ret; ret; ret; ret */ #elif defined __arm__ || defined _M_ARM # define RETURN_INSTR 0xE12FFF1E /* bx lr */ // PPC has its own style of CPU-id #defines. There is no Windows for // PPC as far as I know, so no _M_ variant. #elif defined _ARCH_PPC || defined _ARCH_PWR || defined _ARCH_PWR2 # define RETURN_INSTR 0x4E800020 /* blr */ #elif defined __m68k__ # define RETURN_INSTR 0x4E754E75 /* rts; rts */ #elif defined __riscv # define RETURN_INSTR 0x80828082 /* ret; ret */ #elif defined __sparc || defined __sparcv9 # define RETURN_INSTR 0x81c3e008 /* retl */ #elif defined __alpha # define RETURN_INSTR 0x6bfa8001 /* ret */ #elif defined __hppa # define RETURN_INSTR 0xe840c002 /* bv,n r0(rp) */ #elif defined __mips # define RETURN_INSTR 0x03e00008 /* jr ra */ # ifdef __MIPSEL /* On mipsel, jr ra needs to be followed by a nop. 0x03e00008 as a 64 bits integer just does that */ # define RETURN_INSTR_TYPE uint64_t # endif #elif defined __s390__ # define RETURN_INSTR 0x07fe0000 /* br %r14 */ #elif defined __sh__ # define RETURN_INSTR 0x0b000b00 /* rts; rts */ #elif defined __aarch64__ || defined _M_ARM64 # define RETURN_INSTR 0xd65f03c0 /* ret */ #elif defined __loongarch64 # define RETURN_INSTR 0x4c000020 /* jirl zero, ra, 0 */ #elif defined __ia64 struct ia64_instr { uint32_t mI[4]; }; static const ia64_instr _return_instr = { {0x00000011, 0x00000001, 0x80000200, 0x00840008}}; /* br.ret.sptk.many b0 */ # define RETURN_INSTR _return_instr # define RETURN_INSTR_TYPE ia64_instr #else # error "Need return instruction for this architecture" #endif #ifndef RETURN_INSTR_TYPE # define RETURN_INSTR_TYPE uint32_t #endif // Miscellaneous Windows/Unix portability gumph #ifdef _WIN32 // Uses of this function deliberately leak the string. static LPSTR StrW32Error(DWORD aErrcode) { LPSTR errmsg; FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, nullptr, aErrcode, MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), (LPSTR)&errmsg, 0, nullptr); // FormatMessage puts an unwanted newline at the end of the string size_t n = strlen(errmsg) - 1; while (errmsg[n] == '\r' || errmsg[n] == '\n') { n--; } errmsg[n + 1] = '\0'; return errmsg; } # define LastErrMsg() (StrW32Error(GetLastError())) // Because we use VirtualAlloc in MEM_RESERVE mode, the "page size" we want // is the allocation granularity. static SYSTEM_INFO sInfo_; static inline uint32_t PageSize() { return sInfo_.dwAllocationGranularity; } static void* ReserveRegion(uintptr_t aRequest, bool aAccessible) { return VirtualAlloc((void*)aRequest, PageSize(), aAccessible ? MEM_RESERVE | MEM_COMMIT : MEM_RESERVE, aAccessible ? PAGE_EXECUTE_READWRITE : PAGE_NOACCESS); } static void ReleaseRegion(void* aPage) { VirtualFree(aPage, PageSize(), MEM_RELEASE); } static bool ProbeRegion(uintptr_t aPage) { return aPage >= (uintptr_t)sInfo_.lpMaximumApplicationAddress && aPage + PageSize() >= (uintptr_t)sInfo_.lpMaximumApplicationAddress; } static bool MakeRegionExecutable(void*) { return false; } # undef MAP_FAILED # define MAP_FAILED 0 #else // Unix # define LastErrMsg() (strerror(errno)) static unsigned long gUnixPageSize; static inline unsigned long PageSize() { return gUnixPageSize; } static void* ReserveRegion(uintptr_t aRequest, bool aAccessible) { return mmap(reinterpret_cast(aRequest), PageSize(), aAccessible ? PROT_READ | PROT_WRITE : PROT_NONE, MAP_PRIVATE | MAP_ANON, -1, 0); } static void ReleaseRegion(void* aPage) { munmap(aPage, PageSize()); } static bool ProbeRegion(uintptr_t aPage) { # ifdef XP_SOLARIS return !!posix_madvise(reinterpret_cast(aPage), PageSize(), POSIX_MADV_NORMAL); # else return !!madvise(reinterpret_cast(aPage), PageSize(), MADV_NORMAL); # endif } static int MakeRegionExecutable(void* aPage) { return mprotect((caddr_t)aPage, PageSize(), PROT_READ | PROT_WRITE | PROT_EXEC); } #endif static uintptr_t ReservePoisonArea() { if (sizeof(uintptr_t) == 8) { // Use the hardware-inaccessible region. // We have to avoid 64-bit constants and shifts by 32 bits, since this // code is compiled in 32-bit mode, although it is never executed there. uintptr_t result = (((uintptr_t(0x7FFFFFFFu) << 31) << 1 | uintptr_t(0xF0DEAFFFu)) & ~uintptr_t(PageSize() - 1)); printf("INFO | poison area assumed at 0x%.*" PRIxPTR "\n", SIZxPTR, result); return result; } // First see if we can allocate the preferred poison address from the OS. uintptr_t candidate = (0xF0DEAFFF & ~(PageSize() - 1)); void* result = ReserveRegion(candidate, false); if (result == reinterpret_cast(candidate)) { // success - inaccessible page allocated printf("INFO | poison area allocated at 0x%.*" PRIxPTR " (preferred addr)\n", SIZxPTR, reinterpret_cast(result)); return candidate; } // That didn't work, so see if the preferred address is within a range // of permanently inacessible memory. if (ProbeRegion(candidate)) { // success - selected page cannot be usable memory if (result != MAP_FAILED) { ReleaseRegion(result); } printf("INFO | poison area assumed at 0x%.*" PRIxPTR " (preferred addr)\n", SIZxPTR, candidate); return candidate; } // The preferred address is already in use. Did the OS give us a // consolation prize? if (result != MAP_FAILED) { uintptr_t ures = reinterpret_cast(result); printf("INFO | poison area allocated at 0x%.*" PRIxPTR " (consolation prize)\n", SIZxPTR, ures); return ures; } // It didn't, so try to allocate again, without any constraint on // the address. result = ReserveRegion(0, false); if (result != MAP_FAILED) { uintptr_t ures = reinterpret_cast(result); printf("INFO | poison area allocated at 0x%.*" PRIxPTR " (fallback)\n", SIZxPTR, ures); return ures; } printf("ERROR | no usable poison area found\n"); return 0; } /* The "positive control" area confirms that we can allocate a page with the * proper characteristics. */ static uintptr_t ReservePositiveControl() { void* result = ReserveRegion(0, false); if (result == MAP_FAILED) { printf("ERROR | allocating positive control | %s\n", LastErrMsg()); return 0; } printf("INFO | positive control allocated at 0x%.*" PRIxPTR "\n", SIZxPTR, (uintptr_t)result); return (uintptr_t)result; } /* The "negative control" area confirms that our probe logic does detect a * page that is readable, writable, or executable. */ static uintptr_t ReserveNegativeControl() { void* result = ReserveRegion(0, true); if (result == MAP_FAILED) { printf("ERROR | allocating negative control | %s\n", LastErrMsg()); return 0; } // Fill the page with return instructions. RETURN_INSTR_TYPE* p = reinterpret_cast(result); RETURN_INSTR_TYPE* limit = reinterpret_cast( reinterpret_cast(result) + PageSize()); while (p < limit) { *p++ = RETURN_INSTR; } // Now mark it executable as well as readable and writable. // (mmap(PROT_EXEC) may fail when applied to anonymous memory.) if (MakeRegionExecutable(result)) { printf("ERROR | making negative control executable | %s\n", LastErrMsg()); return 0; } printf("INFO | negative control allocated at 0x%.*" PRIxPTR "\n", SIZxPTR, (uintptr_t)result); return (uintptr_t)result; } #ifndef _WIN32 static void JumpTo(uintptr_t aOpaddr) { # ifdef __ia64 struct func_call { uintptr_t mFunc; uintptr_t mGp; } call = { aOpaddr, }; ((void (*)()) & call)(); # else ((void (*)())aOpaddr)(); # endif } #endif /* Test each page. */ static bool TestPage(const char* aPageLabel, uintptr_t aPageAddr, int aShouldSucceed) { const char* oplabel; uintptr_t opaddr; bool failed = false; for (unsigned int test = 0; test < 3; test++) { switch (test) { // The execute test must be done before the write test, because the // write test will clobber memory at the target address. case 0: oplabel = "reading"; opaddr = aPageAddr + PageSize() / 2 - 1; break; case 1: oplabel = "executing"; opaddr = aPageAddr + PageSize() / 2; break; case 2: oplabel = "writing"; opaddr = aPageAddr + PageSize() / 2 - 1; break; default: abort(); } #ifdef _WIN32 bool badptr = true; MEMORY_BASIC_INFORMATION mbi = {}; if (VirtualQuery((LPCVOID)opaddr, &mbi, sizeof(mbi)) && mbi.State == MEM_COMMIT) { switch (test) { case 0: // read badptr = !(mbi.Protect & (PAGE_EXECUTE_READ | PAGE_EXECUTE_READWRITE | PAGE_READONLY | PAGE_READWRITE)); break; case 1: // execute badptr = !(mbi.Protect & (PAGE_EXECUTE_READ | PAGE_EXECUTE_READWRITE)); break; case 2: // write badptr = !(mbi.Protect & (PAGE_READWRITE | PAGE_EXECUTE_READWRITE)); break; default: abort(); } } if (badptr) { if (aShouldSucceed) { printf("TEST-UNEXPECTED-FAIL | %s %s\n", oplabel, aPageLabel); failed = true; } else { printf("TEST-PASS | %s %s\n", oplabel, aPageLabel); } } else { // if control reaches this point the probe succeeded if (aShouldSucceed) { printf("TEST-PASS | %s %s\n", oplabel, aPageLabel); } else { printf("TEST-UNEXPECTED-FAIL | %s %s\n", oplabel, aPageLabel); failed = true; } } #else pid_t pid = fork(); if (pid == -1) { printf("ERROR | %s %s | fork=%s\n", oplabel, aPageLabel, LastErrMsg()); exit(2); } else if (pid == 0) { volatile unsigned char scratch; switch (test) { case 0: scratch = *(volatile unsigned char*)opaddr; break; case 1: JumpTo(opaddr); break; case 2: *(volatile unsigned char*)opaddr = 0; break; default: abort(); } (void)scratch; _exit(0); } else { int status; if (waitpid(pid, &status, 0) != pid) { printf("ERROR | %s %s | wait=%s\n", oplabel, aPageLabel, LastErrMsg()); exit(2); } if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { if (aShouldSucceed) { printf("TEST-PASS | %s %s\n", oplabel, aPageLabel); } else { printf("TEST-UNEXPECTED-FAIL | %s %s | unexpected successful exit\n", oplabel, aPageLabel); failed = true; } } else if (WIFEXITED(status)) { printf("ERROR | %s %s | unexpected exit code %d\n", oplabel, aPageLabel, WEXITSTATUS(status)); exit(2); } else if (WIFSIGNALED(status)) { if (aShouldSucceed) { printf("TEST-UNEXPECTED-FAIL | %s %s | unexpected signal %d\n", oplabel, aPageLabel, WTERMSIG(status)); failed = true; } else { printf("TEST-PASS | %s %s | signal %d (as expected)\n", oplabel, aPageLabel, WTERMSIG(status)); } } else { printf("ERROR | %s %s | unexpected exit status %d\n", oplabel, aPageLabel, status); exit(2); } } #endif } return failed; } int main() { #ifdef _WIN32 GetSystemInfo(&sInfo_); #else gUnixPageSize = sysconf(_SC_PAGESIZE); #endif uintptr_t ncontrol = ReserveNegativeControl(); uintptr_t pcontrol = ReservePositiveControl(); uintptr_t poison = ReservePoisonArea(); if (!ncontrol || !pcontrol || !poison) { return 2; } bool failed = false; failed |= TestPage("negative control", ncontrol, 1); failed |= TestPage("positive control", pcontrol, 0); failed |= TestPage("poison area", poison, 0); return failed ? 1 : 0; }