Merge "Remove boilerplate."
diff --git a/benchmarks/stdlib_benchmark.cpp b/benchmarks/stdlib_benchmark.cpp
index b6ea58d..14b380a 100644
--- a/benchmarks/stdlib_benchmark.cpp
+++ b/benchmarks/stdlib_benchmark.cpp
@@ -155,88 +155,71 @@
BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_malloc_multiple_8192_allocs_decay1, "AT_SMALL_SIZES");
#endif
-static void BM_stdlib_mbstowcs(benchmark::State& state) {
- const size_t buf_alignment = state.range(0);
- const size_t widebuf_alignment = state.range(1);
-
- std::vector<char> buf;
- std::vector<wchar_t> widebuf;
-
- setlocale(LC_CTYPE, "C.UTF-8")
- || setlocale(LC_CTYPE, "en_US.UTF-8")
- || setlocale(LC_CTYPE, "en_GB.UTF-8")
- || setlocale(LC_CTYPE, "en.UTF-8")
- || setlocale(LC_CTYPE, "de_DE-8")
- || setlocale(LC_CTYPE, "fr_FR-8");
- if (strcmp(nl_langinfo(CODESET), "UTF-8")) {
- errx(1, "ERROR: unable to set locale in BM_stdlib_mbstowcs");
- }
-
- char* buf_aligned = GetAlignedPtr(&buf, buf_alignment, 500000);
- wchar_t* widebuf_aligned = GetAlignedPtr(&widebuf, widebuf_alignment, 500000);
- size_t i, j, k, l;
- l = 0;
- for (i=0xc3; i<0xe0; i++)
- for (j=0x80; j<0xc0; j++)
- buf[l++] = i, buf[l++] = j;
- for (i=0xe1; i<0xed; i++)
- for (j=0x80; j<0xc0; j++)
- for (k=0x80; k<0xc0; k++)
- buf[l++] = i, buf[l++] = j, buf[l++] = k;
- for (i=0xf1; i<0xf4; i++)
- for (j=0x80; j<0xc0; j++)
- for (k=0x80; k<0xc0; k++)
- buf[l++] = i, buf[l++] = j, buf[l++] = 0x80, buf[l++] = k;
- buf[l++] = 0;
+static void BM_stdlib_mbstowcs_ascii(benchmark::State& state) {
+ // It doesn't really matter what ASCII character we pick.
+ // The flow through the fast path is the same regardless.
+ const size_t count = 500000;
+ std::vector<char> mbs(count, 'e');
+ std::vector<wchar_t> wcs(count);
for (auto _ : state) {
- benchmark::DoNotOptimize(mbstowcs(widebuf_aligned, buf_aligned, 500000));
+ benchmark::DoNotOptimize(mbstowcs(&wcs[0], &mbs[0], wcs.size()));
}
- state.SetBytesProcessed(uint64_t(state.iterations()) * uint64_t(500000));
+ state.SetBytesProcessed(uint64_t(state.iterations()) * uint64_t(wcs.size()));
}
-BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_mbstowcs, "0 0");
+BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_mbstowcs_ascii, "");
-static void BM_stdlib_mbrtowc(benchmark::State& state) {
- const size_t buf_alignment = state.range(0);
-
- std::vector<char> buf;
-
- setlocale(LC_CTYPE, "C.UTF-8")
- || setlocale(LC_CTYPE, "en_US.UTF-8")
- || setlocale(LC_CTYPE, "en_GB.UTF-8")
- || setlocale(LC_CTYPE, "en.UTF-8")
- || setlocale(LC_CTYPE, "de_DE-8")
- || setlocale(LC_CTYPE, "fr_FR-8");
- if (strcmp(nl_langinfo(CODESET), "UTF-8")) {
- errx(1, "ERROR: unable to set locale in BM_stdlib_mbrtowc");
+static void BM_stdlib_mbstowcs_wide(benchmark::State& state) {
+ // It doesn't matter much what wide character we pick.
+ // A three-byte character seems pretty representative, and all three byte
+ // characters are the same from the code's perspective.
+ const size_t count = 500000;
+ std::string mbs;
+ for (size_t i = 0; i < count; i++) {
+ mbs += "\xe5\xb1\xb1";
}
+ std::vector<wchar_t> wcs(count);
- char* buf_aligned = GetAlignedPtr(&buf, buf_alignment, 500000);
- size_t i, j, k, l;
- l = 0;
- for (i=0xc3; i<0xe0; i++)
- for (j=0x80; j<0xc0; j++)
- buf[l++] = i, buf[l++] = j;
- for (i=0xe1; i<0xed; i++)
- for (j=0x80; j<0xc0; j++)
- for (k=0x80; k<0xc0; k++)
- buf[l++] = i, buf[l++] = j, buf[l++] = k;
- for (i=0xf1; i<0xf4; i++)
- for (j=0x80; j<0xc0; j++)
- for (k=0x80; k<0xc0; k++)
- buf[l++] = i, buf[l++] = j, buf[l++] = 0x80, buf[l++] = k;
- buf[l++] = 0;
-
- wchar_t wc = 0;
for (auto _ : state) {
- for (j = 0; buf_aligned[j]; j+=mbrtowc(&wc, buf_aligned + j, 4, nullptr)) {
- }
+ benchmark::DoNotOptimize(mbstowcs(&wcs[0], &mbs[0], wcs.size()));
}
- state.SetBytesProcessed(uint64_t(state.iterations()) * uint64_t(500000));
+ state.SetBytesProcessed(uint64_t(state.iterations()) * uint64_t(wcs.size()));
}
-BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_mbrtowc, "0");
+BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_mbstowcs_wide, "");
+
+static void BM_stdlib_mbrtowc_1(benchmark::State& state) {
+ wchar_t wc;
+ for (auto _ : state) {
+ benchmark::DoNotOptimize(mbrtowc(&wc, "e", 1, nullptr));
+ }
+}
+BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_mbrtowc_1, "");
+
+static void BM_stdlib_mbrtowc_2(benchmark::State& state) {
+ wchar_t wc;
+ for (auto _ : state) {
+ benchmark::DoNotOptimize(mbrtowc(&wc, "\xc3\x9f", 3, nullptr));
+ }
+}
+BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_mbrtowc_2, "");
+
+static void BM_stdlib_mbrtowc_3(benchmark::State& state) {
+ wchar_t wc;
+ for (auto _ : state) {
+ benchmark::DoNotOptimize(mbrtowc(&wc, "\xe5\xb1\xb1", 3, nullptr));
+ }
+}
+BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_mbrtowc_3, "");
+
+static void BM_stdlib_mbrtowc_4(benchmark::State& state) {
+ wchar_t wc;
+ for (auto _ : state) {
+ benchmark::DoNotOptimize(mbrtowc(&wc, "\xf0\xa4\xad\xa2", 4, nullptr));
+ }
+}
+BIONIC_BENCHMARK_WITH_ARG(BM_stdlib_mbrtowc_4, "");
BIONIC_TRIVIAL_BENCHMARK(BM_stdlib_atoi, atoi(" -123"));
BIONIC_TRIVIAL_BENCHMARK(BM_stdlib_atol, atol(" -123"));
diff --git a/libc/bionic/c16rtomb.cpp b/libc/bionic/c16rtomb.cpp
index 2d6ae93..e052d04 100644
--- a/libc/bionic/c16rtomb.cpp
+++ b/libc/bionic/c16rtomb.cpp
@@ -43,7 +43,7 @@
size_t c16rtomb(char* s, char16_t c16, mbstate_t* ps) {
static mbstate_t __private_state;
mbstate_t* state = (ps == nullptr) ? &__private_state : ps;
- if (mbsinit(state)) {
+ if (mbstate_is_initial(state)) {
if (is_high_surrogate(c16)) {
char32_t c32 = (c16 & ~0xd800) << 10;
mbstate_set_byte(state, 3, (c32 & 0xff0000) >> 16);
diff --git a/libc/bionic/c32rtomb.cpp b/libc/bionic/c32rtomb.cpp
index 2909d8b..d2519b9 100644
--- a/libc/bionic/c32rtomb.cpp
+++ b/libc/bionic/c32rtomb.cpp
@@ -50,7 +50,7 @@
return mbstate_reset_and_return(1, state);
}
- if (!mbsinit(state)) {
+ if (!mbstate_is_initial(state)) {
return mbstate_reset_and_return_illegal(EILSEQ, state);
}
diff --git a/libc/bionic/mbrtoc16.cpp b/libc/bionic/mbrtoc16.cpp
index acea426..154b8a3 100644
--- a/libc/bionic/mbrtoc16.cpp
+++ b/libc/bionic/mbrtoc16.cpp
@@ -77,12 +77,15 @@
return nconv;
} else if (nconv == 0) {
return mbstate_reset_and_return(nconv, state);
- } else if (c32 > 0x10ffff) {
- // Input cannot be encoded as UTF-16.
- return mbstate_reset_and_return_illegal(EILSEQ, state);
} else if (c32 < 0x10000) {
*pc16 = static_cast<char16_t>(c32);
return mbstate_reset_and_return(nconv, state);
+ } else if (c32 > 0x10ffff) {
+ // This case is currently handled by mbrtoc32() returning an error, but
+ // if that function is extended to cover 5-byte sequences (which are
+ // illegal at the moment), we'd need to explicitly handle the case of
+ // codepoints that can't be represented as a surrogate pair here.
+ return mbstate_reset_and_return_illegal(EILSEQ, state);
} else {
return begin_surrogate(c32, pc16, nconv, state);
}
diff --git a/libc/bionic/mbrtoc32.cpp b/libc/bionic/mbrtoc32.cpp
index 644e542..21603a1 100644
--- a/libc/bionic/mbrtoc32.cpp
+++ b/libc/bionic/mbrtoc32.cpp
@@ -55,7 +55,7 @@
}
uint8_t ch;
- if (mbsinit(state) && (((ch = static_cast<uint8_t>(*s)) & ~0x7f) == 0)) {
+ if (mbstate_is_initial(state) && (((ch = static_cast<uint8_t>(*s)) & ~0x7f) == 0)) {
// Fast path for plain ASCII characters.
if (pc32 != nullptr) {
*pc32 = ch;
@@ -105,7 +105,7 @@
size_t bytes_wanted = length - bytes_so_far;
size_t i;
for (i = 0; i < MIN(bytes_wanted, n); i++) {
- if (!mbsinit(state) && ((*s & 0xc0) != 0x80)) {
+ if (!mbstate_is_initial(state) && ((*s & 0xc0) != 0x80)) {
// Malformed input; bad characters in the middle of a character.
return mbstate_reset_and_return_illegal(EILSEQ, state);
}
diff --git a/libc/bionic/wchar.cpp b/libc/bionic/wchar.cpp
index dabe824..bd9a45e 100644
--- a/libc/bionic/wchar.cpp
+++ b/libc/bionic/wchar.cpp
@@ -54,7 +54,7 @@
//
int mbsinit(const mbstate_t* ps) {
- return (ps == nullptr || (*(reinterpret_cast<const uint32_t*>(ps->__seq)) == 0));
+ return ps == nullptr || mbstate_is_initial(ps);
}
size_t mbrtowc(wchar_t* pwc, const char* s, size_t n, mbstate_t* ps) {
@@ -148,7 +148,7 @@
static mbstate_t __private_state;
mbstate_t* state = (ps == nullptr) ? &__private_state : ps;
- if (!mbsinit(state)) {
+ if (!mbstate_is_initial(state)) {
return mbstate_reset_and_return_illegal(EILSEQ, state);
}
diff --git a/libc/private/bionic_mbstate.h b/libc/private/bionic_mbstate.h
index 352115a..243b220 100644
--- a/libc/private/bionic_mbstate.h
+++ b/libc/private/bionic_mbstate.h
@@ -44,6 +44,10 @@
#define __MB_IS_ERR(rv) (rv == __MB_ERR_ILLEGAL_SEQUENCE || \
rv == __MB_ERR_INCOMPLETE_SEQUENCE)
+static inline __wur bool mbstate_is_initial(const mbstate_t* ps) {
+ return *(reinterpret_cast<const uint32_t*>(ps->__seq)) == 0;
+}
+
static inline __wur size_t mbstate_bytes_so_far(const mbstate_t* ps) {
return
(ps->__seq[2] != 0) ? 3 :
diff --git a/tests/malloc_test.cpp b/tests/malloc_test.cpp
index b5948c3..5e8fb9f 100644
--- a/tests/malloc_test.cpp
+++ b/tests/malloc_test.cpp
@@ -1430,3 +1430,25 @@
GTEST_SKIP() << "bionic extension";
#endif
}
+
+// Regression test for b/206701345 -- scudo bug, MTE only.
+// Fix: https://reviews.llvm.org/D105261
+// Fix: https://android-review.googlesource.com/c/platform/external/scudo/+/1763655
+TEST(malloc, realloc_mte_crash_b206701345) {
+ // We want to hit in-place realloc at the very end of an mmap-ed region. Not
+ // all size classes allow such placement - mmap size has to be divisible by
+ // the block size. At the time of writing this could only be reproduced with
+ // 64 byte size class (i.e. 48 byte allocations), but that may change in the
+ // future. Try several different classes at the lower end.
+ std::vector<void*> ptrs(10000);
+ for (int i = 1; i < 32; ++i) {
+ size_t sz = 16 * i - 1;
+ for (void*& p : ptrs) {
+ p = realloc(malloc(sz), sz + 1);
+ }
+
+ for (void* p : ptrs) {
+ free(p);
+ }
+ }
+}