/* * Copyright (C) 2012 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #include #include "SignalUtils.h" #include "utils.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "private/get_cpu_count_from_string.h" #if defined(__BIONIC__) #include "bionic/pthread_internal.h" #endif #if defined(NOFORTIFY) #define UNISTD_TEST unistd_nofortify #define UNISTD_DEATHTEST unistd_nofortify_DeathTest #else #define UNISTD_TEST unistd #define UNISTD_DEATHTEST unistd_DeathTest #endif using UNISTD_DEATHTEST = SilentDeathTest; using namespace std::chrono_literals; static void* get_brk() { return sbrk(0); } static void* page_align(uintptr_t addr) { uintptr_t mask = sysconf(_SC_PAGE_SIZE) - 1; return reinterpret_cast((addr + mask) & ~mask); } TEST(UNISTD_TEST, brk) { void* initial_break = get_brk(); void* new_break = reinterpret_cast(reinterpret_cast(initial_break) + 1); int ret = brk(new_break); if (ret == -1) { ASSERT_EQ(errno, ENOMEM); } else { ASSERT_EQ(0, ret); ASSERT_GE(get_brk(), new_break); } // Expand by a full page to force the mapping to expand new_break = page_align(reinterpret_cast(initial_break) + sysconf(_SC_PAGE_SIZE)); ret = brk(new_break); if (ret == -1) { ASSERT_EQ(errno, ENOMEM); } else { ASSERT_EQ(0, ret); ASSERT_EQ(get_brk(), new_break); } } TEST(UNISTD_TEST, brk_ENOMEM) { ASSERT_EQ(-1, brk(reinterpret_cast(-1))); ASSERT_EQ(ENOMEM, errno); } #if defined(__GLIBC__) #define SBRK_MIN INTPTR_MIN #define SBRK_MAX INTPTR_MAX #else #define SBRK_MIN PTRDIFF_MIN #define SBRK_MAX PTRDIFF_MAX #endif TEST(UNISTD_TEST, sbrk_ENOMEM) { #if defined(__BIONIC__) && !defined(__LP64__) // There is no way to guarantee that all overflow conditions can be tested // without manipulating the underlying values of the current break. extern void* __bionic_brk; class ScopedBrk { public: ScopedBrk() : saved_brk_(__bionic_brk) {} virtual ~ScopedBrk() { __bionic_brk = saved_brk_; } private: void* saved_brk_; }; ScopedBrk scope_brk; // Set the current break to a point that will cause an overflow. __bionic_brk = reinterpret_cast(static_cast(PTRDIFF_MAX) + 2); // Can't increase by so much that we'd overflow. ASSERT_EQ(reinterpret_cast(-1), sbrk(PTRDIFF_MAX)); ASSERT_EQ(ENOMEM, errno); // Set the current break to a point that will cause an overflow. __bionic_brk = reinterpret_cast(static_cast(PTRDIFF_MAX)); ASSERT_EQ(reinterpret_cast(-1), sbrk(PTRDIFF_MIN)); ASSERT_EQ(ENOMEM, errno); __bionic_brk = reinterpret_cast(static_cast(PTRDIFF_MAX) - 1); ASSERT_EQ(reinterpret_cast(-1), sbrk(PTRDIFF_MIN + 1)); ASSERT_EQ(ENOMEM, errno); #else class ScopedBrk { public: ScopedBrk() : saved_brk_(get_brk()) {} virtual ~ScopedBrk() { brk(saved_brk_); } private: void* saved_brk_; }; ScopedBrk scope_brk; uintptr_t cur_brk = reinterpret_cast(get_brk()); if (cur_brk < static_cast(-(SBRK_MIN+1))) { // Do the overflow test for a max negative increment. ASSERT_EQ(reinterpret_cast(-1), sbrk(SBRK_MIN)); #if defined(__BIONIC__) // GLIBC does not set errno in overflow case. ASSERT_EQ(ENOMEM, errno); #endif } uintptr_t overflow_brk = static_cast(SBRK_MAX) + 2; if (cur_brk < overflow_brk) { // Try and move the value to PTRDIFF_MAX + 2. cur_brk = reinterpret_cast(sbrk(overflow_brk)); } if (cur_brk >= overflow_brk) { ASSERT_EQ(reinterpret_cast(-1), sbrk(SBRK_MAX)); #if defined(__BIONIC__) // GLIBC does not set errno in overflow case. ASSERT_EQ(ENOMEM, errno); #endif } #endif } TEST(UNISTD_TEST, truncate) { TemporaryFile tf; ASSERT_EQ(0, close(tf.fd)); ASSERT_EQ(0, truncate(tf.path, 123)); struct stat sb; ASSERT_EQ(0, stat(tf.path, &sb)); ASSERT_EQ(123, sb.st_size); } TEST(UNISTD_TEST, truncate64_smoke) { TemporaryFile tf; ASSERT_EQ(0, close(tf.fd)); ASSERT_EQ(0, truncate64(tf.path, 123)); struct stat sb; ASSERT_EQ(0, stat(tf.path, &sb)); ASSERT_EQ(123, sb.st_size); } TEST(UNISTD_TEST, ftruncate) { TemporaryFile tf; ASSERT_EQ(0, ftruncate(tf.fd, 123)); ASSERT_EQ(0, close(tf.fd)); struct stat sb; ASSERT_EQ(0, stat(tf.path, &sb)); ASSERT_EQ(123, sb.st_size); } TEST(UNISTD_TEST, ftruncate64_smoke) { TemporaryFile tf; ASSERT_EQ(0, ftruncate64(tf.fd, 123)); ASSERT_EQ(0, close(tf.fd)); struct stat sb; ASSERT_EQ(0, stat(tf.path, &sb)); ASSERT_EQ(123, sb.st_size); } TEST(UNISTD_TEST, ftruncate_negative) { TemporaryFile tf; errno = 0; ASSERT_EQ(-1, ftruncate(tf.fd, -123)); ASSERT_EQ(EINVAL, errno); } static bool g_pause_test_flag = false; static void PauseTestSignalHandler(int) { g_pause_test_flag = true; } TEST(UNISTD_TEST, pause) { ScopedSignalHandler handler(SIGALRM, PauseTestSignalHandler); alarm(1); ASSERT_FALSE(g_pause_test_flag); ASSERT_EQ(-1, pause()); ASSERT_TRUE(g_pause_test_flag); } TEST(UNISTD_TEST, read) { int fd = open("/proc/version", O_RDONLY); ASSERT_TRUE(fd != -1); char buf[5]; ASSERT_EQ(5, read(fd, buf, 5)); ASSERT_EQ(buf[0], 'L'); ASSERT_EQ(buf[1], 'i'); ASSERT_EQ(buf[2], 'n'); ASSERT_EQ(buf[3], 'u'); ASSERT_EQ(buf[4], 'x'); close(fd); } TEST(UNISTD_TEST, read_EBADF) { // read returns ssize_t which is 64-bits on LP64, so it's worth explicitly checking that // our syscall stubs correctly return a 64-bit -1. char buf[1]; ASSERT_EQ(-1, read(-1, buf, sizeof(buf))); ASSERT_EQ(EBADF, errno); } TEST(UNISTD_TEST, syscall_long) { // Check that syscall(3) correctly returns long results. // https://code.google.com/p/android/issues/detail?id=73952 // We assume that the break is > 4GiB, but this is potentially flaky. uintptr_t p = reinterpret_cast(sbrk(0)); ASSERT_EQ(p, static_cast(syscall(__NR_brk, 0))); } TEST(UNISTD_TEST, alarm) { ASSERT_EQ(0U, alarm(0)); } TEST(UNISTD_TEST, _exit) { pid_t pid = fork(); ASSERT_NE(-1, pid) << strerror(errno); if (pid == 0) { _exit(99); } AssertChildExited(pid, 99); } TEST(UNISTD_TEST, getenv_unsetenv) { ASSERT_EQ(0, setenv("test-variable", "hello", 1)); ASSERT_STREQ("hello", getenv("test-variable")); ASSERT_EQ(0, unsetenv("test-variable")); ASSERT_TRUE(getenv("test-variable") == nullptr); } TEST(UNISTD_TEST, unsetenv_EINVAL) { EXPECT_EQ(-1, unsetenv("")); EXPECT_EQ(EINVAL, errno); EXPECT_EQ(-1, unsetenv("a=b")); EXPECT_EQ(EINVAL, errno); } TEST(UNISTD_TEST, setenv_EINVAL) { EXPECT_EQ(-1, setenv(nullptr, "value", 0)); EXPECT_EQ(EINVAL, errno); EXPECT_EQ(-1, setenv(nullptr, "value", 1)); EXPECT_EQ(EINVAL, errno); EXPECT_EQ(-1, setenv("", "value", 0)); EXPECT_EQ(EINVAL, errno); EXPECT_EQ(-1, setenv("", "value", 1)); EXPECT_EQ(EINVAL, errno); EXPECT_EQ(-1, setenv("a=b", "value", 0)); EXPECT_EQ(EINVAL, errno); EXPECT_EQ(-1, setenv("a=b", "value", 1)); EXPECT_EQ(EINVAL, errno); } TEST(UNISTD_TEST, setenv) { ASSERT_EQ(0, unsetenv("test-variable")); char a[] = "a"; char b[] = "b"; char c[] = "c"; // New value. EXPECT_EQ(0, setenv("test-variable", a, 0)); EXPECT_STREQ(a, getenv("test-variable")); // Existing value, no overwrite. EXPECT_EQ(0, setenv("test-variable", b, 0)); EXPECT_STREQ(a, getenv("test-variable")); // Existing value, overwrite. EXPECT_EQ(0, setenv("test-variable", c, 1)); EXPECT_STREQ(c, getenv("test-variable")); // But the arrays backing the values are unchanged. EXPECT_EQ('a', a[0]); EXPECT_EQ('b', b[0]); EXPECT_EQ('c', c[0]); ASSERT_EQ(0, unsetenv("test-variable")); } TEST(UNISTD_TEST, putenv) { ASSERT_EQ(0, unsetenv("a")); char* s1 = strdup("a=b"); ASSERT_EQ(0, putenv(s1)); ASSERT_STREQ("b", getenv("a")); s1[2] = 'c'; ASSERT_STREQ("c", getenv("a")); char* s2 = strdup("a=b"); ASSERT_EQ(0, putenv(s2)); ASSERT_STREQ("b", getenv("a")); ASSERT_EQ('c', s1[2]); ASSERT_EQ(0, unsetenv("a")); free(s1); free(s2); } TEST(UNISTD_TEST, clearenv) { extern char** environ; // Guarantee that environ is not initially empty... ASSERT_EQ(0, setenv("test-variable", "a", 1)); // Stash a copy. std::vector old_environ; for (size_t i = 0; environ[i] != nullptr; ++i) { old_environ.push_back(strdup(environ[i])); } ASSERT_EQ(0, clearenv()); EXPECT_TRUE(environ == nullptr || environ[0] == nullptr); EXPECT_EQ(nullptr, getenv("test-variable")); EXPECT_EQ(0, setenv("test-variable", "post-clear", 1)); EXPECT_STREQ("post-clear", getenv("test-variable")); // Put the old environment back. for (size_t i = 0; i < old_environ.size(); ++i) { EXPECT_EQ(0, putenv(old_environ[i])); } // Check it wasn't overwritten. EXPECT_STREQ("a", getenv("test-variable")); EXPECT_EQ(0, unsetenv("test-variable")); } static void TestSyncFunction(int (*fn)(int)) { int fd; // Can't sync an invalid fd. errno = 0; EXPECT_EQ(-1, fn(-1)); EXPECT_EQ(EBADF, errno); // It doesn't matter whether you've opened a file for write or not. TemporaryFile tf; ASSERT_NE(-1, tf.fd); EXPECT_EQ(0, fn(tf.fd)); ASSERT_NE(-1, fd = open(tf.path, O_RDONLY)); EXPECT_EQ(0, fn(fd)); close(fd); ASSERT_NE(-1, fd = open(tf.path, O_RDWR)); EXPECT_EQ(0, fn(fd)); close(fd); // The fd can even be a directory. ASSERT_NE(-1, fd = open("/data/local/tmp", O_RDONLY)); EXPECT_EQ(0, fn(fd)); close(fd); } static void TestFsyncFunction(int (*fn)(int)) { TestSyncFunction(fn); // But some file systems are fussy about fsync/fdatasync... errno = 0; int fd = open("/proc/version", O_RDONLY); ASSERT_NE(-1, fd); EXPECT_EQ(-1, fn(fd)); EXPECT_EQ(EINVAL, errno); close(fd); } TEST(UNISTD_TEST, fdatasync) { TestFsyncFunction(fdatasync); } TEST(UNISTD_TEST, fsync) { TestFsyncFunction(fsync); } TEST(UNISTD_TEST, syncfs) { TestSyncFunction(syncfs); } TEST(UNISTD_TEST, vfork) { #if defined(__BIONIC__) pthread_internal_t* self = __get_thread(); pid_t cached_pid; ASSERT_TRUE(self->get_cached_pid(&cached_pid)); ASSERT_EQ(syscall(__NR_getpid), cached_pid); ASSERT_FALSE(self->is_vforked()); pid_t rc = vfork(); ASSERT_NE(-1, rc); if (rc == 0) { if (self->get_cached_pid(&cached_pid)) { const char* error = "__get_thread()->cached_pid_ set after vfork\n"; write(STDERR_FILENO, error, strlen(error)); _exit(1); } if (!self->is_vforked()) { const char* error = "__get_thread()->vforked_ not set after vfork\n"; write(STDERR_FILENO, error, strlen(error)); _exit(1); } _exit(0); } else { ASSERT_TRUE(self->get_cached_pid(&cached_pid)); ASSERT_EQ(syscall(__NR_getpid), cached_pid); ASSERT_FALSE(self->is_vforked()); int status; pid_t wait_result = waitpid(rc, &status, 0); ASSERT_EQ(wait_result, rc); ASSERT_TRUE(WIFEXITED(status)); ASSERT_EQ(0, WEXITSTATUS(status)); } #endif } static void AssertGetPidCorrect() { // The loop is just to make manual testing/debugging with strace easier. pid_t getpid_syscall_result = syscall(__NR_getpid); for (size_t i = 0; i < 128; ++i) { ASSERT_EQ(getpid_syscall_result, getpid()); } } static void TestGetPidCachingWithFork(int (*fork_fn)(), void (*exit_fn)(int)) { pid_t parent_pid = getpid(); ASSERT_EQ(syscall(__NR_getpid), parent_pid); pid_t fork_result = fork_fn(); ASSERT_NE(fork_result, -1); if (fork_result == 0) { // We're the child. ASSERT_NO_FATAL_FAILURE(AssertGetPidCorrect()); ASSERT_EQ(parent_pid, getppid()); exit_fn(123); } else { // We're the parent. ASSERT_EQ(parent_pid, getpid()); AssertChildExited(fork_result, 123); } } // gettid() is marked as __attribute_const__, which will have the compiler // optimize out multiple calls to gettid in the same function. This wrapper // defeats that optimization. static __attribute__((__noinline__)) pid_t GetTidForTest() { __asm__(""); return gettid(); } static void AssertGetTidCorrect() { // The loop is just to make manual testing/debugging with strace easier. pid_t gettid_syscall_result = syscall(__NR_gettid); for (size_t i = 0; i < 128; ++i) { ASSERT_EQ(gettid_syscall_result, GetTidForTest()); } } static void TestGetTidCachingWithFork(int (*fork_fn)(), void (*exit_fn)(int)) { pid_t parent_tid = GetTidForTest(); ASSERT_EQ(syscall(__NR_gettid), parent_tid); pid_t fork_result = fork_fn(); ASSERT_NE(fork_result, -1); if (fork_result == 0) { // We're the child. EXPECT_EQ(syscall(__NR_getpid), syscall(__NR_gettid)); EXPECT_EQ(getpid(), GetTidForTest()) << "real tid is " << syscall(__NR_gettid) << ", pid is " << syscall(__NR_getpid); ASSERT_NO_FATAL_FAILURE(AssertGetTidCorrect()); exit_fn(123); } else { // We're the parent. ASSERT_EQ(parent_tid, GetTidForTest()); AssertChildExited(fork_result, 123); } } TEST(UNISTD_TEST, getpid_caching_and_fork) { TestGetPidCachingWithFork(fork, exit); } TEST(UNISTD_TEST, gettid_caching_and_fork) { TestGetTidCachingWithFork(fork, exit); } TEST(UNISTD_TEST, getpid_caching_and_vfork) { TestGetPidCachingWithFork(vfork, _exit); } static int CloneLikeFork() { return clone(nullptr, nullptr, SIGCHLD, nullptr); } TEST(UNISTD_TEST, getpid_caching_and_clone_process) { TestGetPidCachingWithFork(CloneLikeFork, exit); } TEST(UNISTD_TEST, gettid_caching_and_clone_process) { TestGetTidCachingWithFork(CloneLikeFork, exit); } static int CloneAndSetTid() { pid_t child_tid = 0; pid_t parent_tid = GetTidForTest(); int rv = clone(nullptr, nullptr, CLONE_CHILD_SETTID | SIGCHLD, nullptr, nullptr, nullptr, &child_tid); EXPECT_NE(-1, rv); if (rv == 0) { // Child. EXPECT_EQ(child_tid, GetTidForTest()); EXPECT_NE(child_tid, parent_tid); } else { EXPECT_NE(child_tid, GetTidForTest()); EXPECT_NE(child_tid, parent_tid); EXPECT_EQ(GetTidForTest(), parent_tid); } return rv; } TEST(UNISTD_TEST, gettid_caching_and_clone_process_settid) { TestGetTidCachingWithFork(CloneAndSetTid, exit); } static int CloneStartRoutine(int (*start_routine)(void*)) { void* child_stack[1024]; return clone(start_routine, untag_address(&child_stack[1024]), SIGCHLD, nullptr); } static int GetPidCachingCloneStartRoutine(void*) { AssertGetPidCorrect(); return 123; } TEST(UNISTD_TEST, getpid_caching_and_clone) { pid_t parent_pid = getpid(); ASSERT_EQ(syscall(__NR_getpid), parent_pid); int clone_result = CloneStartRoutine(GetPidCachingCloneStartRoutine); ASSERT_NE(clone_result, -1); ASSERT_EQ(parent_pid, getpid()); AssertChildExited(clone_result, 123); } static int GetTidCachingCloneStartRoutine(void*) { AssertGetTidCorrect(); return 123; } TEST(UNISTD_TEST, gettid_caching_and_clone) { pid_t parent_tid = GetTidForTest(); ASSERT_EQ(syscall(__NR_gettid), parent_tid); int clone_result = CloneStartRoutine(GetTidCachingCloneStartRoutine); ASSERT_NE(clone_result, -1); ASSERT_EQ(parent_tid, GetTidForTest()); AssertChildExited(clone_result, 123); } static int CloneChildExit(void*) { AssertGetPidCorrect(); AssertGetTidCorrect(); exit(33); } TEST(UNISTD_TEST, clone_fn_and_exit) { int clone_result = CloneStartRoutine(CloneChildExit); ASSERT_NE(-1, clone_result); AssertGetPidCorrect(); AssertGetTidCorrect(); AssertChildExited(clone_result, 33); } static void* GetPidCachingPthreadStartRoutine(void*) { AssertGetPidCorrect(); return nullptr; } TEST(UNISTD_TEST, getpid_caching_and_pthread_create) { pid_t parent_pid = getpid(); pthread_t t; ASSERT_EQ(0, pthread_create(&t, nullptr, GetPidCachingPthreadStartRoutine, nullptr)); ASSERT_EQ(parent_pid, getpid()); void* result; ASSERT_EQ(0, pthread_join(t, &result)); ASSERT_EQ(nullptr, result); } static void* GetTidCachingPthreadStartRoutine(void*) { AssertGetTidCorrect(); uint64_t tid = GetTidForTest(); return reinterpret_cast(tid); } TEST(UNISTD_TEST, gettid_caching_and_pthread_create) { pid_t parent_tid = GetTidForTest(); pthread_t t; ASSERT_EQ(0, pthread_create(&t, nullptr, GetTidCachingPthreadStartRoutine, &parent_tid)); ASSERT_EQ(parent_tid, GetTidForTest()); void* result; ASSERT_EQ(0, pthread_join(t, &result)); ASSERT_NE(static_cast(parent_tid), reinterpret_cast(result)); } __attribute__((noinline)) static void HwasanVforkTestChild() { // Allocate a tagged region on stack and leave it there. char x[10000]; DoNotOptimize(x); _exit(0); } __attribute__((noinline)) static void HwasanReadMemory(const char* p, size_t size) { // Read memory byte-by-byte. This will blow up if the pointer tag in p does not match any memory // tag in [p, p+size). char z; for (size_t i = 0; i < size; ++i) { DoNotOptimize(z = p[i]); } } __attribute__((noinline, no_sanitize("hwaddress"))) static void HwasanVforkTestParent() { // Allocate a region on stack, but don't tag it (see the function attribute). // This depends on unallocated stack space at current function entry being untagged. char x[10000]; DoNotOptimize(x); // Verify that contents of x[] are untagged. HwasanReadMemory(x, sizeof(x)); } TEST(UNISTD_TEST, hwasan_vfork) { // Test hwasan annotation in vfork. This test is only interesting when built with hwasan, but it // is supposed to work correctly either way. if (vfork()) { HwasanVforkTestParent(); } else { HwasanVforkTestChild(); } } TEST_F(UNISTD_DEATHTEST, abort) { ASSERT_EXIT(abort(), testing::KilledBySignal(SIGABRT), ""); } TEST(UNISTD_TEST, sethostname) { // The permissions check happens before the argument check, so this will // fail for a different reason if you're running as root than if you're // not, but it'll fail either way. Checking that we have the symbol is about // all we can do for sethostname(2). ASSERT_EQ(-1, sethostname("", -1)); } TEST(UNISTD_TEST, gethostname) { char hostname[HOST_NAME_MAX + 1]; memset(hostname, 0, sizeof(hostname)); // Can we get the hostname with a big buffer? ASSERT_EQ(0, gethostname(hostname, HOST_NAME_MAX)); // Can we get the hostname with a right-sized buffer? errno = 0; ASSERT_EQ(0, gethostname(hostname, strlen(hostname) + 1)); // Does uname(2) agree? utsname buf; ASSERT_EQ(0, uname(&buf)); ASSERT_EQ(0, strncmp(hostname, buf.nodename, sizeof(buf.nodename))); ASSERT_GT(strlen(hostname), 0U); // Do we correctly detect truncation? errno = 0; ASSERT_EQ(-1, gethostname(hostname, strlen(hostname))); ASSERT_EQ(ENAMETOOLONG, errno); } TEST(UNISTD_TEST, pathconf_fpathconf) { TemporaryFile tf; long rc = 0L; // As a file system's block size is always power of 2, the configure values // for ALLOC and XFER should be power of 2 as well. rc = pathconf(tf.path, _PC_ALLOC_SIZE_MIN); ASSERT_TRUE(rc > 0 && powerof2(rc)); rc = pathconf(tf.path, _PC_REC_MIN_XFER_SIZE); ASSERT_TRUE(rc > 0 && powerof2(rc)); rc = pathconf(tf.path, _PC_REC_XFER_ALIGN); ASSERT_TRUE(rc > 0 && powerof2(rc)); rc = fpathconf(tf.fd, _PC_ALLOC_SIZE_MIN); ASSERT_TRUE(rc > 0 && powerof2(rc)); rc = fpathconf(tf.fd, _PC_REC_MIN_XFER_SIZE); ASSERT_TRUE(rc > 0 && powerof2(rc)); rc = fpathconf(tf.fd, _PC_REC_XFER_ALIGN); ASSERT_TRUE(rc > 0 && powerof2(rc)); } TEST(UNISTD_TEST, _POSIX_constants) { // Make a tight verification of _POSIX_* / _POSIX2_* / _XOPEN_* macros, to prevent change by mistake. // Verify according to POSIX.1-2008. EXPECT_EQ(200809L, _POSIX_VERSION); EXPECT_EQ(2, _POSIX_AIO_LISTIO_MAX); EXPECT_EQ(1, _POSIX_AIO_MAX); EXPECT_EQ(4096, _POSIX_ARG_MAX); EXPECT_EQ(25, _POSIX_CHILD_MAX); EXPECT_EQ(20000000, _POSIX_CLOCKRES_MIN); EXPECT_EQ(32, _POSIX_DELAYTIMER_MAX); EXPECT_EQ(255, _POSIX_HOST_NAME_MAX); EXPECT_EQ(8, _POSIX_LINK_MAX); EXPECT_EQ(9, _POSIX_LOGIN_NAME_MAX); EXPECT_EQ(255, _POSIX_MAX_CANON); EXPECT_EQ(255, _POSIX_MAX_INPUT); EXPECT_EQ(8, _POSIX_MQ_OPEN_MAX); EXPECT_EQ(32, _POSIX_MQ_PRIO_MAX); EXPECT_EQ(14, _POSIX_NAME_MAX); EXPECT_EQ(8, _POSIX_NGROUPS_MAX); EXPECT_EQ(20, _POSIX_OPEN_MAX); EXPECT_EQ(256, _POSIX_PATH_MAX); EXPECT_EQ(512, _POSIX_PIPE_BUF); EXPECT_EQ(255, _POSIX_RE_DUP_MAX); EXPECT_EQ(8, _POSIX_RTSIG_MAX); EXPECT_EQ(256, _POSIX_SEM_NSEMS_MAX); EXPECT_EQ(32767, _POSIX_SEM_VALUE_MAX); EXPECT_EQ(32, _POSIX_SIGQUEUE_MAX); EXPECT_EQ(32767, _POSIX_SSIZE_MAX); EXPECT_EQ(8, _POSIX_STREAM_MAX); #if !defined(__GLIBC__) EXPECT_EQ(4, _POSIX_SS_REPL_MAX); #endif EXPECT_EQ(255, _POSIX_SYMLINK_MAX); EXPECT_EQ(8, _POSIX_SYMLOOP_MAX); EXPECT_EQ(4, _POSIX_THREAD_DESTRUCTOR_ITERATIONS); EXPECT_EQ(128, _POSIX_THREAD_KEYS_MAX); EXPECT_EQ(64, _POSIX_THREAD_THREADS_MAX); EXPECT_EQ(32, _POSIX_TIMER_MAX); #if !defined(__GLIBC__) EXPECT_EQ(30, _POSIX_TRACE_EVENT_NAME_MAX); EXPECT_EQ(8, _POSIX_TRACE_NAME_MAX); EXPECT_EQ(8, _POSIX_TRACE_SYS_MAX); EXPECT_EQ(32, _POSIX_TRACE_USER_EVENT_MAX); #endif EXPECT_EQ(9, _POSIX_TTY_NAME_MAX); EXPECT_EQ(6, _POSIX_TZNAME_MAX); EXPECT_EQ(99, _POSIX2_BC_BASE_MAX); EXPECT_EQ(2048, _POSIX2_BC_DIM_MAX); EXPECT_EQ(99, _POSIX2_BC_SCALE_MAX); EXPECT_EQ(1000, _POSIX2_BC_STRING_MAX); EXPECT_EQ(14, _POSIX2_CHARCLASS_NAME_MAX); EXPECT_EQ(2, _POSIX2_COLL_WEIGHTS_MAX); EXPECT_EQ(32, _POSIX2_EXPR_NEST_MAX); EXPECT_EQ(2048, _POSIX2_LINE_MAX); EXPECT_EQ(255, _POSIX2_RE_DUP_MAX); EXPECT_EQ(16, _XOPEN_IOV_MAX); #if !defined(__GLIBC__) EXPECT_EQ(255, _XOPEN_NAME_MAX); EXPECT_EQ(1024, _XOPEN_PATH_MAX); #endif } TEST(UNISTD_TEST, _POSIX_options) { EXPECT_EQ(_POSIX_VERSION, _POSIX_ADVISORY_INFO); EXPECT_GT(_POSIX_BARRIERS, 0); EXPECT_GT(_POSIX_SPIN_LOCKS, 0); EXPECT_NE(_POSIX_CHOWN_RESTRICTED, -1); EXPECT_EQ(_POSIX_VERSION, _POSIX_CLOCK_SELECTION); #if !defined(__GLIBC__) // glibc supports ancient kernels. EXPECT_EQ(_POSIX_VERSION, _POSIX_CPUTIME); #endif EXPECT_EQ(_POSIX_VERSION, _POSIX_FSYNC); EXPECT_EQ(_POSIX_VERSION, _POSIX_IPV6); EXPECT_GT(_POSIX_JOB_CONTROL, 0); EXPECT_EQ(_POSIX_VERSION, _POSIX_MAPPED_FILES); EXPECT_EQ(_POSIX_VERSION, _POSIX_MEMLOCK); EXPECT_EQ(_POSIX_VERSION, _POSIX_MEMLOCK_RANGE); EXPECT_EQ(_POSIX_VERSION, _POSIX_MEMORY_PROTECTION); #if !defined(__GLIBC__) // glibc supports ancient kernels. EXPECT_EQ(_POSIX_VERSION, _POSIX_MONOTONIC_CLOCK); #endif EXPECT_GT(_POSIX_NO_TRUNC, 0); #if !defined(ANDROID_HOST_MUSL) EXPECT_EQ(_POSIX_VERSION, _POSIX_PRIORITY_SCHEDULING); #endif EXPECT_EQ(_POSIX_VERSION, _POSIX_RAW_SOCKETS); EXPECT_EQ(_POSIX_VERSION, _POSIX_READER_WRITER_LOCKS); EXPECT_EQ(_POSIX_VERSION, _POSIX_REALTIME_SIGNALS); EXPECT_GT(_POSIX_REGEXP, 0); EXPECT_GT(_POSIX_SAVED_IDS, 0); EXPECT_EQ(_POSIX_VERSION, _POSIX_SEMAPHORES); EXPECT_GT(_POSIX_SHELL, 0); EXPECT_EQ(_POSIX_VERSION, _POSIX_SPAWN); #if !defined(ANDROID_HOST_MUSL) EXPECT_EQ(-1, _POSIX_SPORADIC_SERVER); EXPECT_EQ(_POSIX_VERSION, _POSIX_SYNCHRONIZED_IO); #endif EXPECT_EQ(_POSIX_VERSION, _POSIX_THREADS); EXPECT_EQ(_POSIX_VERSION, _POSIX_THREAD_ATTR_STACKADDR); EXPECT_EQ(_POSIX_VERSION, _POSIX_THREAD_ATTR_STACKSIZE); #if !defined(__GLIBC__) // glibc supports ancient kernels. EXPECT_EQ(_POSIX_VERSION, _POSIX_THREAD_CPUTIME); #endif EXPECT_EQ(_POSIX_VERSION, _POSIX_THREAD_PRIORITY_SCHEDULING); EXPECT_EQ(_POSIX_VERSION, _POSIX_THREAD_PROCESS_SHARED); #if !defined(ANDROID_HOST_MUSL) EXPECT_EQ(-1, _POSIX_THREAD_ROBUST_PRIO_PROTECT); #endif EXPECT_EQ(_POSIX_VERSION, _POSIX_THREAD_SAFE_FUNCTIONS); #if !defined(ANDROID_HOST_MUSL) EXPECT_EQ(-1, _POSIX_THREAD_SPORADIC_SERVER); #endif EXPECT_EQ(_POSIX_VERSION, _POSIX_TIMEOUTS); EXPECT_EQ(_POSIX_VERSION, _POSIX_TIMERS); #if !defined(ANDROID_HOST_MUSL) EXPECT_EQ(-1, _POSIX_TRACE); EXPECT_EQ(-1, _POSIX_TRACE_EVENT_FILTER); EXPECT_EQ(-1, _POSIX_TRACE_INHERIT); EXPECT_EQ(-1, _POSIX_TRACE_LOG); EXPECT_EQ(-1, _POSIX_TYPED_MEMORY_OBJECTS); #endif EXPECT_NE(-1, _POSIX_VDISABLE); EXPECT_EQ(_POSIX_VERSION, _POSIX2_VERSION); EXPECT_EQ(_POSIX_VERSION, _POSIX2_C_BIND); #if !defined(ANDROID_HOST_MUSL) EXPECT_EQ(_POSIX_VERSION, _POSIX2_CHAR_TERM); #endif EXPECT_EQ(700, _XOPEN_VERSION); EXPECT_EQ(1, _XOPEN_ENH_I18N); #if !defined(ANDROID_HOST_MUSL) EXPECT_EQ(1, _XOPEN_REALTIME); EXPECT_EQ(1, _XOPEN_REALTIME_THREADS); EXPECT_EQ(1, _XOPEN_SHM); #endif EXPECT_EQ(1, _XOPEN_UNIX); #if defined(__BIONIC__) // These tests only pass on bionic, as bionic and glibc has different support on these macros. // Macros like _POSIX_ASYNCHRONOUS_IO are not supported on bionic yet. EXPECT_EQ(-1, _POSIX_ASYNCHRONOUS_IO); EXPECT_EQ(-1, _POSIX_MESSAGE_PASSING); EXPECT_EQ(-1, _POSIX_PRIORITIZED_IO); EXPECT_EQ(-1, _POSIX_SHARED_MEMORY_OBJECTS); EXPECT_EQ(-1, _POSIX_THREAD_PRIO_INHERIT); EXPECT_EQ(-1, _POSIX_THREAD_PRIO_PROTECT); EXPECT_EQ(-1, _POSIX_THREAD_ROBUST_PRIO_INHERIT); EXPECT_EQ(-1, _POSIX2_C_DEV); EXPECT_EQ(-1, _POSIX2_FORT_DEV); EXPECT_EQ(-1, _POSIX2_FORT_RUN); EXPECT_EQ(-1, _POSIX2_LOCALEDEF); EXPECT_EQ(-1, _POSIX2_SW_DEV); EXPECT_EQ(-1, _POSIX2_UPE); EXPECT_EQ(-1, _XOPEN_CRYPT); EXPECT_EQ(-1, _XOPEN_LEGACY); EXPECT_EQ(-1, _XOPEN_STREAMS); #endif // defined(__BIONIC__) } #define VERIFY_SYSCONF_UNKNOWN(name) \ VerifySysconf(name, #name, [](long v){return v == -1 && errno == EINVAL;}) #define VERIFY_SYSCONF_UNSUPPORTED(name) \ VerifySysconf(name, #name, [](long v){return v == -1 && errno == 0;}) // sysconf() means unlimited when it returns -1 with errno unchanged. #define VERIFY_SYSCONF_POSITIVE(name) \ VerifySysconf(name, #name, [](long v){return (v > 0 || v == -1) && errno == 0;}) #define VERIFY_SYSCONF_POSIX_VERSION(name) \ VerifySysconf(name, #name, [](long v){return v == _POSIX_VERSION && errno == 0;}) static void VerifySysconf(int option, const char *option_name, bool (*verify)(long)) { errno = 0; long ret = sysconf(option); EXPECT_TRUE(verify(ret)) << "name = " << option_name << ", ret = " << ret <<", Error Message: " << strerror(errno); } TEST(UNISTD_TEST, sysconf) { VERIFY_SYSCONF_POSIX_VERSION(_SC_ADVISORY_INFO); VERIFY_SYSCONF_POSITIVE(_SC_ARG_MAX); VERIFY_SYSCONF_POSIX_VERSION(_SC_BARRIERS); VERIFY_SYSCONF_POSITIVE(_SC_BC_BASE_MAX); VERIFY_SYSCONF_POSITIVE(_SC_BC_DIM_MAX); VERIFY_SYSCONF_POSITIVE(_SC_BC_SCALE_MAX); VERIFY_SYSCONF_POSITIVE(_SC_CHILD_MAX); VERIFY_SYSCONF_POSITIVE(_SC_CLK_TCK); VERIFY_SYSCONF_POSITIVE(_SC_COLL_WEIGHTS_MAX); VERIFY_SYSCONF_POSIX_VERSION(_SC_CPUTIME); VERIFY_SYSCONF_POSITIVE(_SC_EXPR_NEST_MAX); VERIFY_SYSCONF_POSITIVE(_SC_LINE_MAX); VERIFY_SYSCONF_POSITIVE(_SC_NGROUPS_MAX); VERIFY_SYSCONF_POSITIVE(_SC_OPEN_MAX); VERIFY_SYSCONF_POSITIVE(_SC_PASS_MAX); VERIFY_SYSCONF_POSIX_VERSION(_SC_2_C_BIND); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_FORT_DEV); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_FORT_RUN); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_UPE); VERIFY_SYSCONF_POSIX_VERSION(_SC_2_VERSION); VERIFY_SYSCONF_POSITIVE(_SC_JOB_CONTROL); VERIFY_SYSCONF_POSITIVE(_SC_SAVED_IDS); VERIFY_SYSCONF_POSIX_VERSION(_SC_VERSION); VERIFY_SYSCONF_POSITIVE(_SC_RE_DUP_MAX); VERIFY_SYSCONF_POSITIVE(_SC_STREAM_MAX); VERIFY_SYSCONF_POSITIVE(_SC_TZNAME_MAX); VerifySysconf(_SC_XOPEN_VERSION, "_SC_XOPEN_VERSION", [](long v){return v == _XOPEN_VERSION && errno == 0;}); VERIFY_SYSCONF_POSITIVE(_SC_ATEXIT_MAX); VERIFY_SYSCONF_POSITIVE(_SC_IOV_MAX); VERIFY_SYSCONF_POSITIVE(_SC_UIO_MAXIOV); EXPECT_EQ(sysconf(_SC_IOV_MAX), sysconf(_SC_UIO_MAXIOV)); VERIFY_SYSCONF_POSITIVE(_SC_PAGESIZE); VERIFY_SYSCONF_POSITIVE(_SC_PAGE_SIZE); VerifySysconf(_SC_PAGE_SIZE, "_SC_PAGE_SIZE", [](long v){return v == sysconf(_SC_PAGESIZE) && errno == 0 && v == getpagesize();}); VERIFY_SYSCONF_POSITIVE(_SC_XOPEN_UNIX); VERIFY_SYSCONF_POSITIVE(_SC_AIO_LISTIO_MAX); VERIFY_SYSCONF_POSITIVE(_SC_AIO_MAX); VerifySysconf(_SC_AIO_PRIO_DELTA_MAX, "_SC_AIO_PRIO_DELTA_MAX", [](long v){return v >= 0 && errno == 0;}); VERIFY_SYSCONF_POSITIVE(_SC_DELAYTIMER_MAX); VERIFY_SYSCONF_POSITIVE(_SC_MQ_OPEN_MAX); VERIFY_SYSCONF_POSITIVE(_SC_MQ_PRIO_MAX); VERIFY_SYSCONF_POSITIVE(_SC_RTSIG_MAX); VERIFY_SYSCONF_POSITIVE(_SC_SEM_NSEMS_MAX); VERIFY_SYSCONF_POSITIVE(_SC_SEM_VALUE_MAX); VERIFY_SYSCONF_POSIX_VERSION(_SC_SPIN_LOCKS); VERIFY_SYSCONF_POSITIVE(_SC_TIMER_MAX); VERIFY_SYSCONF_POSIX_VERSION(_SC_FSYNC); VERIFY_SYSCONF_POSIX_VERSION(_SC_MAPPED_FILES); VERIFY_SYSCONF_POSIX_VERSION(_SC_MEMLOCK); VERIFY_SYSCONF_POSIX_VERSION(_SC_MEMLOCK_RANGE); VERIFY_SYSCONF_POSIX_VERSION(_SC_MEMORY_PROTECTION); VERIFY_SYSCONF_POSIX_VERSION(_SC_PRIORITY_SCHEDULING); VERIFY_SYSCONF_POSIX_VERSION(_SC_REALTIME_SIGNALS); VERIFY_SYSCONF_POSIX_VERSION(_SC_SEMAPHORES); VERIFY_SYSCONF_POSIX_VERSION(_SC_SYNCHRONIZED_IO); VERIFY_SYSCONF_POSIX_VERSION(_SC_TIMERS); VERIFY_SYSCONF_POSITIVE(_SC_GETGR_R_SIZE_MAX); VERIFY_SYSCONF_POSITIVE(_SC_GETPW_R_SIZE_MAX); VERIFY_SYSCONF_POSITIVE(_SC_LOGIN_NAME_MAX); VERIFY_SYSCONF_POSITIVE(_SC_THREAD_DESTRUCTOR_ITERATIONS); VERIFY_SYSCONF_POSITIVE(_SC_THREAD_KEYS_MAX); VERIFY_SYSCONF_POSITIVE(_SC_THREAD_STACK_MIN); VERIFY_SYSCONF_POSITIVE(_SC_THREAD_THREADS_MAX); VERIFY_SYSCONF_POSITIVE(_SC_TTY_NAME_MAX); VERIFY_SYSCONF_POSIX_VERSION(_SC_THREADS); VERIFY_SYSCONF_POSIX_VERSION(_SC_THREAD_ATTR_STACKADDR); VERIFY_SYSCONF_POSIX_VERSION(_SC_THREAD_ATTR_STACKSIZE); VERIFY_SYSCONF_POSIX_VERSION(_SC_THREAD_PRIORITY_SCHEDULING); VERIFY_SYSCONF_UNSUPPORTED(_SC_THREAD_PRIO_INHERIT); VERIFY_SYSCONF_UNSUPPORTED(_SC_THREAD_PRIO_PROTECT); VERIFY_SYSCONF_POSIX_VERSION(_SC_THREAD_SAFE_FUNCTIONS); VERIFY_SYSCONF_POSITIVE(_SC_NPROCESSORS_CONF); VERIFY_SYSCONF_POSITIVE(_SC_NPROCESSORS_ONLN); VERIFY_SYSCONF_POSITIVE(_SC_PHYS_PAGES); VERIFY_SYSCONF_POSITIVE(_SC_AVPHYS_PAGES); VERIFY_SYSCONF_POSIX_VERSION(_SC_MONOTONIC_CLOCK); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_PBS); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_PBS_ACCOUNTING); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_PBS_CHECKPOINT); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_PBS_LOCATE); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_PBS_MESSAGE); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_PBS_TRACK); VERIFY_SYSCONF_POSIX_VERSION(_SC_CLOCK_SELECTION); VERIFY_SYSCONF_POSITIVE(_SC_HOST_NAME_MAX); VERIFY_SYSCONF_POSIX_VERSION(_SC_IPV6); VERIFY_SYSCONF_POSIX_VERSION(_SC_RAW_SOCKETS); VERIFY_SYSCONF_POSIX_VERSION(_SC_READER_WRITER_LOCKS); VERIFY_SYSCONF_POSITIVE(_SC_REGEXP); VERIFY_SYSCONF_POSITIVE(_SC_SHELL); VERIFY_SYSCONF_POSIX_VERSION(_SC_SPAWN); VERIFY_SYSCONF_UNSUPPORTED(_SC_SPORADIC_SERVER); VERIFY_SYSCONF_POSITIVE(_SC_SYMLOOP_MAX); VERIFY_SYSCONF_POSIX_VERSION(_SC_THREAD_CPUTIME); VERIFY_SYSCONF_POSIX_VERSION(_SC_THREAD_PROCESS_SHARED); VERIFY_SYSCONF_UNSUPPORTED(_SC_THREAD_SPORADIC_SERVER); VERIFY_SYSCONF_POSIX_VERSION(_SC_TIMEOUTS); VERIFY_SYSCONF_UNSUPPORTED(_SC_TRACE); VERIFY_SYSCONF_UNSUPPORTED(_SC_TRACE_EVENT_FILTER); VERIFY_SYSCONF_UNSUPPORTED(_SC_TRACE_EVENT_NAME_MAX); VERIFY_SYSCONF_UNSUPPORTED(_SC_TRACE_INHERIT); VERIFY_SYSCONF_UNSUPPORTED(_SC_TRACE_LOG); VERIFY_SYSCONF_UNSUPPORTED(_SC_TRACE_NAME_MAX); VERIFY_SYSCONF_UNSUPPORTED(_SC_TRACE_SYS_MAX); VERIFY_SYSCONF_UNSUPPORTED(_SC_TRACE_USER_EVENT_MAX); VERIFY_SYSCONF_UNSUPPORTED(_SC_TYPED_MEMORY_OBJECTS); VERIFY_SYSCONF_UNSUPPORTED(_SC_XOPEN_STREAMS); #if defined(__LP64__) VERIFY_SYSCONF_UNSUPPORTED(_SC_V7_ILP32_OFF32); VERIFY_SYSCONF_UNSUPPORTED(_SC_V7_ILP32_OFFBIG); VERIFY_SYSCONF_POSITIVE(_SC_V7_LP64_OFF64); VERIFY_SYSCONF_POSITIVE(_SC_V7_LPBIG_OFFBIG); #else VERIFY_SYSCONF_POSITIVE(_SC_V7_ILP32_OFF32); #if defined(__BIONIC__) // bionic does not support 64 bits off_t type on 32bit machine. VERIFY_SYSCONF_UNSUPPORTED(_SC_V7_ILP32_OFFBIG); #endif VERIFY_SYSCONF_UNSUPPORTED(_SC_V7_LP64_OFF64); VERIFY_SYSCONF_UNSUPPORTED(_SC_V7_LPBIG_OFFBIG); #endif #if defined(__BIONIC__) // Tests can only run on bionic, as bionic and glibc have different support for these options. // Below options are not supported on bionic yet. VERIFY_SYSCONF_UNSUPPORTED(_SC_ASYNCHRONOUS_IO); VERIFY_SYSCONF_UNSUPPORTED(_SC_MESSAGE_PASSING); VERIFY_SYSCONF_UNSUPPORTED(_SC_PRIORITIZED_IO); VERIFY_SYSCONF_UNSUPPORTED(_SC_SHARED_MEMORY_OBJECTS); VERIFY_SYSCONF_UNSUPPORTED(_SC_THREAD_ROBUST_PRIO_INHERIT); VERIFY_SYSCONF_UNSUPPORTED(_SC_THREAD_ROBUST_PRIO_PROTECT); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_C_DEV); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_LOCALEDEF); VERIFY_SYSCONF_UNSUPPORTED(_SC_2_SW_DEV); VERIFY_SYSCONF_UNSUPPORTED(_SC_XOPEN_CRYPT); VERIFY_SYSCONF_UNSUPPORTED(_SC_XOPEN_LEGACY); VERIFY_SYSCONF_UNSUPPORTED(_SC_XOPEN_UUCP); #endif // defined(__BIONIC__) } TEST(UNISTD_TEST, get_cpu_count_from_string) { ASSERT_EQ(0, GetCpuCountFromString(" ")); ASSERT_EQ(1, GetCpuCountFromString("0")); ASSERT_EQ(40, GetCpuCountFromString("0-39")); ASSERT_EQ(4, GetCpuCountFromString("0, 1-2, 4\n")); } TEST(UNISTD_TEST, sysconf_SC_NPROCESSORS_ONLN) { std::string line; ASSERT_TRUE(android::base::ReadFileToString("/sys/devices/system/cpu/online", &line)); long online_cpus = 0; for (const std::string& s : android::base::Split(line, ",")) { std::vector numbers = android::base::Split(s, "-"); if (numbers.size() == 1u) { online_cpus++; } else { online_cpus += atoi(numbers[1].c_str()) - atoi(numbers[0].c_str()) + 1; } } ASSERT_EQ(online_cpus, sysconf(_SC_NPROCESSORS_ONLN)); } TEST(UNISTD_TEST, sysconf_SC_ARG_MAX) { // Since Linux 2.6.23, ARG_MAX isn't a constant and depends on RLIMIT_STACK. // See prepare_arg_pages() in the kernel for the gory details: // https://elixir.bootlin.com/linux/v5.3.11/source/fs/exec.c#L451 // Get our current limit, and set things up so we restore the limit. rlimit rl; ASSERT_EQ(0, getrlimit(RLIMIT_STACK, &rl)); uint64_t original_rlim_cur = rl.rlim_cur; if (rl.rlim_cur == RLIM_INFINITY) { rl.rlim_cur = 8 * 1024 * 1024; // Bionic reports unlimited stacks as 8MiB. } auto guard = android::base::make_scope_guard([&rl, original_rlim_cur]() { rl.rlim_cur = original_rlim_cur; ASSERT_EQ(0, setrlimit(RLIMIT_STACK, &rl)); }); // _SC_ARG_MAX should be 1/4 the stack size. EXPECT_EQ(static_cast(rl.rlim_cur / 4), sysconf(_SC_ARG_MAX)); // If you have a really small stack, the kernel still guarantees "32 pages" (see fs/exec.c). rl.rlim_cur = 1024; rl.rlim_max = RLIM_INFINITY; ASSERT_EQ(0, setrlimit(RLIMIT_STACK, &rl)); EXPECT_EQ(static_cast(32 * sysconf(_SC_PAGE_SIZE)), sysconf(_SC_ARG_MAX)); // With a 128-page stack limit, we know exactly what _SC_ARG_MAX should be... rl.rlim_cur = 128 * sysconf(_SC_PAGE_SIZE); rl.rlim_max = RLIM_INFINITY; ASSERT_EQ(0, setrlimit(RLIMIT_STACK, &rl)); EXPECT_EQ(static_cast((128 * sysconf(_SC_PAGE_SIZE)) / 4), sysconf(_SC_ARG_MAX)); } TEST(UNISTD_TEST, sysconf_unknown) { VERIFY_SYSCONF_UNKNOWN(-1); VERIFY_SYSCONF_UNKNOWN(666); } TEST(UNISTD_TEST, dup2_same) { // POSIX says of dup2: // If fildes2 is already a valid open file descriptor ... // [and] fildes is equal to fildes2 ... dup2() shall return // fildes2 without closing it. // This isn't true of dup3(2), so we need to manually implement that. // Equal and valid. int fd = open("/proc/version", O_RDONLY); ASSERT_TRUE(fd != -1); ASSERT_EQ(fd, dup2(fd, fd)); ASSERT_EQ(0, close(fd)); // Check that dup2 didn't close fd. // Equal, but invalid. errno = 0; ASSERT_EQ(-1, dup2(fd, fd)); ASSERT_EQ(EBADF, errno); } TEST(UNISTD_TEST, dup3) { int fd = open("/proc/version", O_RDONLY); ASSERT_EQ(666, dup3(fd, 666, 0)); ASSERT_FALSE(CloseOnExec(666)); close(666); ASSERT_EQ(667, dup3(fd, 667, O_CLOEXEC)); ASSERT_TRUE(CloseOnExec(667)); close(667); close(fd); } TEST(UNISTD_TEST, lockf_smoke) { constexpr off64_t file_size = 32*1024LL; TemporaryFile tf; ASSERT_EQ(0, ftruncate(tf.fd, file_size)); // Lock everything. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_LOCK, file_size)); // Try-lock everything, this should succeed too. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_TLOCK, file_size)); // Check status. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_TEST, file_size)); // Unlock file. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_ULOCK, file_size)); } TEST(UNISTD_TEST, lockf_zero) { constexpr off64_t file_size = 32*1024LL; TemporaryFile tf; ASSERT_EQ(0, ftruncate(tf.fd, file_size)); // Lock everything by specifying a size of 0 (meaning "to the end, even if it changes"). ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_LOCK, 0)); // Check that it's locked. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_TEST, file_size)); // Move the end. ASSERT_EQ(0, ftruncate(tf.fd, 2*file_size)); // Check that the new section is locked too. ASSERT_EQ(file_size, lseek64(tf.fd, file_size, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_TEST, 2*file_size)); } TEST(UNISTD_TEST, lockf_negative) { constexpr off64_t file_size = 32*1024LL; TemporaryFile tf; ASSERT_EQ(0, ftruncate(tf.fd, file_size)); // Lock everything, but specifying the range in reverse. ASSERT_EQ(file_size, lseek64(tf.fd, file_size, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_LOCK, -file_size)); // Check that it's locked. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_TEST, file_size)); } TEST(UNISTD_TEST, lockf_with_child) { constexpr off64_t file_size = 32*1024LL; TemporaryFile tf; ASSERT_EQ(0, ftruncate(tf.fd, file_size)); // Lock everything. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_LOCK, file_size)); // Fork a child process pid_t pid = fork(); ASSERT_NE(-1, pid); if (pid == 0) { // Check that the child cannot lock the file. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(-1, lockf64(tf.fd, F_TLOCK, file_size)); ASSERT_EQ(EAGAIN, errno); // Check also that it reports itself as locked. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(-1, lockf64(tf.fd, F_TEST, file_size)); ASSERT_EQ(EACCES, errno); _exit(0); } AssertChildExited(pid, 0); } TEST(UNISTD_TEST, lockf_partial_with_child) { constexpr off64_t file_size = 32*1024LL; TemporaryFile tf; ASSERT_EQ(0, ftruncate(tf.fd, file_size)); // Lock the first half of the file. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_LOCK, file_size/2)); // Fork a child process. pid_t pid = fork(); ASSERT_NE(-1, pid); if (pid == 0) { // Check that the child can lock the other half. ASSERT_EQ(file_size/2, lseek64(tf.fd, file_size/2, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_TLOCK, file_size/2)); // Check that the child cannot lock the first half. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(-1, lockf64(tf.fd, F_TEST, file_size/2)); ASSERT_EQ(EACCES, errno); // Check also that it reports itself as locked. ASSERT_EQ(0, lseek64(tf.fd, 0, SEEK_SET)); ASSERT_EQ(-1, lockf64(tf.fd, F_TEST, file_size/2)); ASSERT_EQ(EACCES, errno); _exit(0); } AssertChildExited(pid, 0); // The second half was locked by the child, but the lock disappeared // when the process exited, so check it can be locked now. ASSERT_EQ(file_size/2, lseek64(tf.fd, file_size/2, SEEK_SET)); ASSERT_EQ(0, lockf64(tf.fd, F_TLOCK, file_size/2)); } TEST(UNISTD_TEST, getdomainname) { struct utsname u; ASSERT_EQ(0, uname(&u)); char buf[sizeof(u.domainname)]; ASSERT_EQ(0, getdomainname(buf, sizeof(buf))); EXPECT_STREQ(u.domainname, buf); #if defined(__BIONIC__) // bionic and glibc have different behaviors when len is too small ASSERT_EQ(-1, getdomainname(buf, strlen(u.domainname))); EXPECT_EQ(EINVAL, errno); #endif } TEST(UNISTD_TEST, setdomainname) { __user_cap_header_struct header; memset(&header, 0, sizeof(header)); header.version = _LINUX_CAPABILITY_VERSION_3; __user_cap_data_struct old_caps[_LINUX_CAPABILITY_U32S_3]; ASSERT_EQ(0, capget(&header, &old_caps[0])); auto admin_idx = CAP_TO_INDEX(CAP_SYS_ADMIN); auto admin_mask = CAP_TO_MASK(CAP_SYS_ADMIN); bool has_admin = old_caps[admin_idx].effective & admin_mask; if (has_admin) { __user_cap_data_struct new_caps[_LINUX_CAPABILITY_U32S_3]; memcpy(new_caps, old_caps, sizeof(new_caps)); new_caps[admin_idx].effective &= ~admin_mask; ASSERT_EQ(0, capset(&header, &new_caps[0])) << "failed to drop admin privileges"; } const char* name = "newdomainname"; ASSERT_EQ(-1, setdomainname(name, strlen(name))); ASSERT_EQ(EPERM, errno); if (has_admin) { ASSERT_EQ(0, capset(&header, &old_caps[0])) << "failed to restore admin privileges"; } } TEST(UNISTD_TEST, execve_failure) { ExecTestHelper eth; errno = 0; ASSERT_EQ(-1, execve("/", eth.GetArgs(), eth.GetEnv())); ASSERT_EQ(EACCES, errno); } static void append_llvm_cov_env_var(std::string& env_str) { if (getenv("LLVM_PROFILE_FILE") != nullptr) env_str.append("__LLVM_PROFILE_RT_INIT_ONCE=__LLVM_PROFILE_RT_INIT_ONCE\n"); } TEST(UNISTD_TEST, execve_args) { // int execve(const char* path, char* argv[], char* envp[]); // Test basic argument passing. ExecTestHelper eth; eth.SetArgs({"echo", "hello", "world", nullptr}); eth.Run([&]() { execve(BIN_DIR "echo", eth.GetArgs(), eth.GetEnv()); }, 0, "hello world\n"); // Test environment variable setting too. eth.SetArgs({"printenv", nullptr}); eth.SetEnv({"A=B", nullptr}); std::string expected_output("A=B\n"); append_llvm_cov_env_var(expected_output); eth.Run([&]() { execve(BIN_DIR "printenv", eth.GetArgs(), eth.GetEnv()); }, 0, expected_output.c_str()); } TEST(UNISTD_TEST, execl_failure) { errno = 0; ASSERT_EQ(-1, execl("/", "/", nullptr)); ASSERT_EQ(EACCES, errno); } TEST(UNISTD_TEST, execl) { ExecTestHelper eth; // int execl(const char* path, const char* arg, ...); eth.Run([&]() { execl(BIN_DIR "echo", "echo", "hello", "world", nullptr); }, 0, "hello world\n"); } TEST(UNISTD_TEST, execle_failure) { ExecTestHelper eth; errno = 0; ASSERT_EQ(-1, execle("/", "/", nullptr, eth.GetEnv())); ASSERT_EQ(EACCES, errno); } TEST(UNISTD_TEST, execle) { ExecTestHelper eth; eth.SetEnv({"A=B", nullptr}); std::string expected_output("A=B\n"); append_llvm_cov_env_var(expected_output); // int execle(const char* path, const char* arg, ..., char* envp[]); eth.Run([&]() { execle(BIN_DIR "printenv", "printenv", nullptr, eth.GetEnv()); }, 0, expected_output.c_str()); } TEST(UNISTD_TEST, execv_failure) { ExecTestHelper eth; errno = 0; ASSERT_EQ(-1, execv("/", eth.GetArgs())); ASSERT_EQ(EACCES, errno); } TEST(UNISTD_TEST, execv) { ExecTestHelper eth; eth.SetArgs({"echo", "hello", "world", nullptr}); // int execv(const char* path, char* argv[]); eth.Run([&]() { execv(BIN_DIR "echo", eth.GetArgs()); }, 0, "hello world\n"); } TEST(UNISTD_TEST, execlp_failure) { errno = 0; ASSERT_EQ(-1, execlp("/", "/", nullptr)); ASSERT_EQ(EACCES, errno); } TEST(UNISTD_TEST, execlp) { ExecTestHelper eth; // int execlp(const char* file, const char* arg, ...); eth.Run([&]() { execlp("echo", "echo", "hello", "world", nullptr); }, 0, "hello world\n"); } TEST(UNISTD_TEST, execvp_failure) { ExecTestHelper eth; eth.SetArgs({nullptr}); errno = 0; ASSERT_EQ(-1, execvp("/", eth.GetArgs())); ASSERT_EQ(EACCES, errno); } TEST(UNISTD_TEST, execvp) { ExecTestHelper eth; eth.SetArgs({"echo", "hello", "world", nullptr}); // int execvp(const char* file, char* argv[]); eth.Run([&]() { execvp("echo", eth.GetArgs()); }, 0, "hello world\n"); } TEST(UNISTD_TEST, execvpe_failure) { ExecTestHelper eth; errno = 0; ASSERT_EQ(-1, execvpe("this-does-not-exist", eth.GetArgs(), eth.GetEnv())); // Running in CTS we might not even be able to search all directories in $PATH. ASSERT_TRUE(errno == ENOENT || errno == EACCES); } TEST(UNISTD_TEST, execvpe) { // int execvpe(const char* file, char* argv[], char* envp[]); // Test basic argument passing. ExecTestHelper eth; eth.SetArgs({"echo", "hello", "world", nullptr}); eth.Run([&]() { execvpe("echo", eth.GetArgs(), eth.GetEnv()); }, 0, "hello world\n"); // Test environment variable setting too. eth.SetArgs({"printenv", nullptr}); eth.SetEnv({"A=B", nullptr}); std::string expected_output("A=B\n"); append_llvm_cov_env_var(expected_output); eth.Run([&]() { execvpe("printenv", eth.GetArgs(), eth.GetEnv()); }, 0, expected_output.c_str()); } TEST(UNISTD_TEST, execvpe_ENOEXEC) { // Create a shell script with #!. TemporaryFile tf; ASSERT_TRUE(android::base::WriteStringToFile("#!" BIN_DIR "sh\necho script\n", tf.path)); // Set $PATH so we can find it. setenv("PATH", dirname(tf.path), 1); ExecTestHelper eth; eth.SetArgs({basename(tf.path), nullptr}); // It's not inherently executable. errno = 0; ASSERT_EQ(-1, execvpe(basename(tf.path), eth.GetArgs(), eth.GetEnv())); ASSERT_EQ(EACCES, errno); // Make it executable (and keep it writable because we're going to rewrite it below). ASSERT_EQ(0, chmod(tf.path, 0777)); // TemporaryFile will have a writable fd, so we can test ETXTBSY while we're here... errno = 0; ASSERT_EQ(-1, execvpe(basename(tf.path), eth.GetArgs(), eth.GetEnv())); ASSERT_EQ(ETXTBSY, errno); // 1. The simplest test: the kernel should handle this. ASSERT_EQ(0, close(tf.fd)); eth.Run([&]() { execvpe(basename(tf.path), eth.GetArgs(), eth.GetEnv()); }, 0, "script\n"); // 2. Try again without a #!. We should have to handle this ourselves. ASSERT_TRUE(android::base::WriteStringToFile("echo script\n", tf.path)); eth.Run([&]() { execvpe(basename(tf.path), eth.GetArgs(), eth.GetEnv()); }, 0, "script\n"); // 3. Again without a #!, but also with a leading '/', since that's a special case in the // implementation. eth.Run([&]() { execvpe(tf.path, eth.GetArgs(), eth.GetEnv()); }, 0, "script\n"); } TEST(UNISTD_TEST, execvp_libcore_test_55017) { ExecTestHelper eth; eth.SetArgs({"/system/bin/does-not-exist", nullptr}); errno = 0; ASSERT_EQ(-1, execvp("/system/bin/does-not-exist", eth.GetArgs())); ASSERT_EQ(ENOENT, errno); } TEST(UNISTD_TEST, exec_argv0_null) { // http://b/33276926 and http://b/227498625. // // With old kernels, bionic will see the null pointer and use "" but // with new (5.18+) kernels, the kernel will already have substituted the // empty string, so we don't make any assertion here about what (if anything) // comes before the first ':'. // // If this ever causes trouble, we could change bionic to replace _either_ the // null pointer or the empty string. We could also use the actual name from // readlink() on /proc/self/exe if we ever had reason to disallow programs // from trying to hide like this. char* args[] = {nullptr}; char* envs[] = {nullptr}; ASSERT_EXIT(execve("/system/bin/run-as", args, envs), testing::ExitedWithCode(1), ": usage: run-as"); } TEST(UNISTD_TEST, fexecve_failure) { ExecTestHelper eth; errno = 0; int fd = open("/", O_RDONLY); ASSERT_NE(-1, fd); ASSERT_EQ(-1, fexecve(fd, eth.GetArgs(), eth.GetEnv())); ASSERT_EQ(EACCES, errno); close(fd); } TEST(UNISTD_TEST, fexecve_bad_fd) { ExecTestHelper eth; errno = 0; ASSERT_EQ(-1, fexecve(-1, eth.GetArgs(), eth.GetEnv())); ASSERT_EQ(EBADF, errno); } TEST(UNISTD_TEST, fexecve_args) { // Test basic argument passing. int echo_fd = open(BIN_DIR "echo", O_RDONLY | O_CLOEXEC); ASSERT_NE(-1, echo_fd); ExecTestHelper eth; eth.SetArgs({"echo", "hello", "world", nullptr}); eth.Run([&]() { fexecve(echo_fd, eth.GetArgs(), eth.GetEnv()); }, 0, "hello world\n"); close(echo_fd); // Test environment variable setting too. int printenv_fd = open(BIN_DIR "printenv", O_RDONLY | O_CLOEXEC); ASSERT_NE(-1, printenv_fd); eth.SetArgs({"printenv", nullptr}); eth.SetEnv({"A=B", nullptr}); std::string expected_output("A=B\n"); append_llvm_cov_env_var(expected_output); eth.Run([&]() { fexecve(printenv_fd, eth.GetArgs(), eth.GetEnv()); }, 0, expected_output.c_str()); close(printenv_fd); } TEST(UNISTD_TEST, getlogin_r) { char buf[LOGIN_NAME_MAX] = {}; EXPECT_EQ(ERANGE, getlogin_r(buf, 0)); EXPECT_EQ(0, getlogin_r(buf, sizeof(buf))); EXPECT_STREQ(getlogin(), buf); } TEST(UNISTD_TEST, swab) { // POSIX: "The swab() function shall copy nbytes bytes, which are pointed to by src, // to the object pointed to by dest, exchanging adjacent bytes." char buf[BUFSIZ]; memset(buf, 'x', sizeof(buf)); swab("ehll oowlr\0d", buf, 12); ASSERT_STREQ("hello world", buf); } TEST(UNISTD_TEST, swab_odd_byte_count) { // POSIX: "If nbytes is odd, swab() copies and exchanges nbytes-1 bytes and the disposition // of the last byte is unspecified." // ...but it seems unreasonable to not just leave the last byte alone. char buf[BUFSIZ]; memset(buf, 'x', sizeof(buf)); swab("012345", buf, 3); ASSERT_EQ('1', buf[0]); ASSERT_EQ('0', buf[1]); ASSERT_EQ('x', buf[2]); } TEST(UNISTD_TEST, swab_overlap) { // POSIX: "If copying takes place between objects that overlap, the behavior is undefined." // ...but it seems unreasonable to not just do the right thing. char buf[] = "012345"; swab(buf, buf, 4); ASSERT_EQ('1', buf[0]); ASSERT_EQ('0', buf[1]); ASSERT_EQ('3', buf[2]); ASSERT_EQ('2', buf[3]); ASSERT_EQ('4', buf[4]); ASSERT_EQ('5', buf[5]); ASSERT_EQ(0, buf[6]); } TEST(UNISTD_TEST, swab_negative_byte_count) { // POSIX: "If nbytes is negative, swab() does nothing." char buf[BUFSIZ]; memset(buf, 'x', sizeof(buf)); swab("hello", buf, -1); ASSERT_EQ('x', buf[0]); } TEST(UNISTD_TEST, usleep) { auto t0 = std::chrono::steady_clock::now(); ASSERT_EQ(0, usleep(5000)); auto t1 = std::chrono::steady_clock::now(); ASSERT_GE(t1-t0, 5000us); } TEST(UNISTD_TEST, sleep) { auto t0 = std::chrono::steady_clock::now(); ASSERT_EQ(0U, sleep(1)); auto t1 = std::chrono::steady_clock::now(); ASSERT_GE(t1-t0, 1s); } TEST(UNISTD_TEST, close_range) { #if defined(__GLIBC__) GTEST_SKIP() << "glibc too old"; #elif defined(ANDROID_HOST_MUSL) GTEST_SKIP() << "musl does not have close_range"; #else // __GLIBC__ int fd = open("/proc/version", O_RDONLY); ASSERT_GE(fd, 0); // Try to close the file descriptor (this requires a 5.9+ kernel) if (close_range(fd, fd, 0) == 0) { // we can't close it *again* ASSERT_EQ(close(fd), -1); ASSERT_EQ(errno, EBADF); } else { ASSERT_EQ(errno, ENOSYS); // since close_range() failed, we can close it normally ASSERT_EQ(close(fd), 0); } #endif // __GLIBC__ } TEST(UNISTD_TEST, copy_file_range) { #if defined(__GLIBC__) GTEST_SKIP() << "glibc too old"; #else // __GLIBC__ TemporaryFile tf; ASSERT_TRUE(android::base::WriteStringToFd("hello world", tf.fd)); ASSERT_EQ(0, lseek(tf.fd, SEEK_SET, 0)); TemporaryFile tf2; ASSERT_EQ(11, copy_file_range(tf.fd, NULL, tf2.fd, NULL, 11, 0)); ASSERT_EQ(0, lseek(tf2.fd, SEEK_SET, 0)); std::string content; ASSERT_TRUE(android::base::ReadFdToString(tf2.fd, &content)); ASSERT_EQ("hello world", content); #endif // __GLIBC__ }