com_android_internal_os_Zygote.cpp revision a103ebed6e9a44a406b31636791d2970c882ec44
1/* 2 * Copyright (C) 2008 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17#define LOG_TAG "Zygote" 18 19// sys/mount.h has to come before linux/fs.h due to redefinition of MS_RDONLY, MS_BIND, etc 20#include <sys/mount.h> 21#include <linux/fs.h> 22 23#include <grp.h> 24#include <fcntl.h> 25#include <paths.h> 26#include <signal.h> 27#include <stdlib.h> 28#include <unistd.h> 29#include <sys/capability.h> 30#include <sys/personality.h> 31#include <sys/prctl.h> 32#include <sys/resource.h> 33#include <sys/stat.h> 34#include <sys/types.h> 35#include <sys/utsname.h> 36#include <sys/wait.h> 37 38 39#include <cutils/fs.h> 40#include <cutils/multiuser.h> 41#include <cutils/sched_policy.h> 42#include <private/android_filesystem_config.h> 43#include <utils/String8.h> 44#include <selinux/android.h> 45#include <processgroup/processgroup.h> 46 47#include "android_runtime/AndroidRuntime.h" 48#include "JNIHelp.h" 49#include "ScopedLocalRef.h" 50#include "ScopedPrimitiveArray.h" 51#include "ScopedUtfChars.h" 52 53#include "nativebridge/native_bridge.h" 54 55namespace { 56 57using android::String8; 58 59static pid_t gSystemServerPid = 0; 60 61static const char kZygoteClassName[] = "com/android/internal/os/Zygote"; 62static jclass gZygoteClass; 63static jmethodID gCallPostForkChildHooks; 64 65// Must match values in com.android.internal.os.Zygote. 66enum MountExternalKind { 67 MOUNT_EXTERNAL_NONE = 0, 68 MOUNT_EXTERNAL_SINGLEUSER = 1, 69 MOUNT_EXTERNAL_MULTIUSER = 2, 70 MOUNT_EXTERNAL_MULTIUSER_ALL = 3, 71}; 72 73static void RuntimeAbort(JNIEnv* env) { 74 env->FatalError("RuntimeAbort"); 75} 76 77// This signal handler is for zygote mode, since the zygote must reap its children 78static void SigChldHandler(int /*signal_number*/) { 79 pid_t pid; 80 int status; 81 82 while ((pid = waitpid(-1, &status, WNOHANG)) > 0) { 83 // Log process-death status that we care about. In general it is 84 // not safe to call LOG(...) from a signal handler because of 85 // possible reentrancy. However, we know a priori that the 86 // current implementation of LOG() is safe to call from a SIGCHLD 87 // handler in the zygote process. If the LOG() implementation 88 // changes its locking strategy or its use of syscalls within the 89 // lazy-init critical section, its use here may become unsafe. 90 if (WIFEXITED(status)) { 91 if (WEXITSTATUS(status)) { 92 ALOGI("Process %d exited cleanly (%d)", pid, WEXITSTATUS(status)); 93 } 94 } else if (WIFSIGNALED(status)) { 95 if (WTERMSIG(status) != SIGKILL) { 96 ALOGI("Process %d exited due to signal (%d)", pid, WTERMSIG(status)); 97 } 98 if (WCOREDUMP(status)) { 99 ALOGI("Process %d dumped core.", pid); 100 } 101 } 102 103 // If the just-crashed process is the system_server, bring down zygote 104 // so that it is restarted by init and system server will be restarted 105 // from there. 106 if (pid == gSystemServerPid) { 107 ALOGE("Exit zygote because system server (%d) has terminated"); 108 kill(getpid(), SIGKILL); 109 } 110 } 111 112 // Note that we shouldn't consider ECHILD an error because 113 // the secondary zygote might have no children left to wait for. 114 if (pid < 0 && errno != ECHILD) { 115 ALOGW("Zygote SIGCHLD error in waitpid: %s", strerror(errno)); 116 } 117} 118 119// Configures the SIGCHLD handler for the zygote process. This is configured 120// very late, because earlier in the runtime we may fork() and exec() 121// other processes, and we want to waitpid() for those rather than 122// have them be harvested immediately. 123// 124// This ends up being called repeatedly before each fork(), but there's 125// no real harm in that. 126static void SetSigChldHandler() { 127 struct sigaction sa; 128 memset(&sa, 0, sizeof(sa)); 129 sa.sa_handler = SigChldHandler; 130 131 int err = sigaction(SIGCHLD, &sa, NULL); 132 if (err < 0) { 133 ALOGW("Error setting SIGCHLD handler: %d", errno); 134 } 135} 136 137// Sets the SIGCHLD handler back to default behavior in zygote children. 138static void UnsetSigChldHandler() { 139 struct sigaction sa; 140 memset(&sa, 0, sizeof(sa)); 141 sa.sa_handler = SIG_DFL; 142 143 int err = sigaction(SIGCHLD, &sa, NULL); 144 if (err < 0) { 145 ALOGW("Error unsetting SIGCHLD handler: %d", errno); 146 } 147} 148 149// Calls POSIX setgroups() using the int[] object as an argument. 150// A NULL argument is tolerated. 151static void SetGids(JNIEnv* env, jintArray javaGids) { 152 if (javaGids == NULL) { 153 return; 154 } 155 156 ScopedIntArrayRO gids(env, javaGids); 157 if (gids.get() == NULL) { 158 RuntimeAbort(env); 159 } 160 int rc = setgroups(gids.size(), reinterpret_cast<const gid_t*>(&gids[0])); 161 if (rc == -1) { 162 ALOGE("setgroups failed"); 163 RuntimeAbort(env); 164 } 165} 166 167// Sets the resource limits via setrlimit(2) for the values in the 168// two-dimensional array of integers that's passed in. The second dimension 169// contains a tuple of length 3: (resource, rlim_cur, rlim_max). NULL is 170// treated as an empty array. 171static void SetRLimits(JNIEnv* env, jobjectArray javaRlimits) { 172 if (javaRlimits == NULL) { 173 return; 174 } 175 176 rlimit rlim; 177 memset(&rlim, 0, sizeof(rlim)); 178 179 for (int i = 0; i < env->GetArrayLength(javaRlimits); ++i) { 180 ScopedLocalRef<jobject> javaRlimitObject(env, env->GetObjectArrayElement(javaRlimits, i)); 181 ScopedIntArrayRO javaRlimit(env, reinterpret_cast<jintArray>(javaRlimitObject.get())); 182 if (javaRlimit.size() != 3) { 183 ALOGE("rlimits array must have a second dimension of size 3"); 184 RuntimeAbort(env); 185 } 186 187 rlim.rlim_cur = javaRlimit[1]; 188 rlim.rlim_max = javaRlimit[2]; 189 190 int rc = setrlimit(javaRlimit[0], &rlim); 191 if (rc == -1) { 192 ALOGE("setrlimit(%d, {%d, %d}) failed", javaRlimit[0], rlim.rlim_cur, rlim.rlim_max); 193 RuntimeAbort(env); 194 } 195 } 196} 197 198// The debug malloc library needs to know whether it's the zygote or a child. 199extern "C" int gMallocLeakZygoteChild; 200 201static void EnableKeepCapabilities(JNIEnv* env) { 202 int rc = prctl(PR_SET_KEEPCAPS, 1, 0, 0, 0); 203 if (rc == -1) { 204 ALOGE("prctl(PR_SET_KEEPCAPS) failed"); 205 RuntimeAbort(env); 206 } 207} 208 209static void DropCapabilitiesBoundingSet(JNIEnv* env) { 210 for (int i = 0; prctl(PR_CAPBSET_READ, i, 0, 0, 0) >= 0; i++) { 211 int rc = prctl(PR_CAPBSET_DROP, i, 0, 0, 0); 212 if (rc == -1) { 213 if (errno == EINVAL) { 214 ALOGE("prctl(PR_CAPBSET_DROP) failed with EINVAL. Please verify " 215 "your kernel is compiled with file capabilities support"); 216 } else { 217 ALOGE("prctl(PR_CAPBSET_DROP) failed"); 218 RuntimeAbort(env); 219 } 220 } 221 } 222} 223 224static void SetCapabilities(JNIEnv* env, int64_t permitted, int64_t effective) { 225 __user_cap_header_struct capheader; 226 memset(&capheader, 0, sizeof(capheader)); 227 capheader.version = _LINUX_CAPABILITY_VERSION_3; 228 capheader.pid = 0; 229 230 __user_cap_data_struct capdata[2]; 231 memset(&capdata, 0, sizeof(capdata)); 232 capdata[0].effective = effective; 233 capdata[1].effective = effective >> 32; 234 capdata[0].permitted = permitted; 235 capdata[1].permitted = permitted >> 32; 236 237 if (capset(&capheader, &capdata[0]) == -1) { 238 ALOGE("capset(%lld, %lld) failed", permitted, effective); 239 RuntimeAbort(env); 240 } 241} 242 243static void SetSchedulerPolicy(JNIEnv* env) { 244 errno = -set_sched_policy(0, SP_DEFAULT); 245 if (errno != 0) { 246 ALOGE("set_sched_policy(0, SP_DEFAULT) failed"); 247 RuntimeAbort(env); 248 } 249} 250 251// Create a private mount namespace and bind mount appropriate emulated 252// storage for the given user. 253static bool MountEmulatedStorage(uid_t uid, jint mount_mode, bool force_mount_namespace) { 254 if (mount_mode == MOUNT_EXTERNAL_NONE && !force_mount_namespace) { 255 return true; 256 } 257 258 // Create a second private mount namespace for our process 259 if (unshare(CLONE_NEWNS) == -1) { 260 ALOGW("Failed to unshare(): %d", errno); 261 return false; 262 } 263 264 if (mount_mode == MOUNT_EXTERNAL_NONE) { 265 return true; 266 } 267 268 // See storage config details at http://source.android.com/tech/storage/ 269 userid_t user_id = multiuser_get_user_id(uid); 270 271 // Create bind mounts to expose external storage 272 if (mount_mode == MOUNT_EXTERNAL_MULTIUSER || mount_mode == MOUNT_EXTERNAL_MULTIUSER_ALL) { 273 // These paths must already be created by init.rc 274 const char* source = getenv("EMULATED_STORAGE_SOURCE"); 275 const char* target = getenv("EMULATED_STORAGE_TARGET"); 276 const char* legacy = getenv("EXTERNAL_STORAGE"); 277 if (source == NULL || target == NULL || legacy == NULL) { 278 ALOGW("Storage environment undefined; unable to provide external storage"); 279 return false; 280 } 281 282 // Prepare source paths 283 284 // /mnt/shell/emulated/0 285 const String8 source_user(String8::format("%s/%d", source, user_id)); 286 // /storage/emulated/0 287 const String8 target_user(String8::format("%s/%d", target, user_id)); 288 289 if (fs_prepare_dir(source_user.string(), 0000, 0, 0) == -1 290 || fs_prepare_dir(target_user.string(), 0000, 0, 0) == -1) { 291 return false; 292 } 293 294 if (mount_mode == MOUNT_EXTERNAL_MULTIUSER_ALL) { 295 // Mount entire external storage tree for all users 296 if (TEMP_FAILURE_RETRY(mount(source, target, NULL, MS_BIND, NULL)) == -1) { 297 ALOGW("Failed to mount %s to %s :%d", source, target, errno); 298 return false; 299 } 300 } else { 301 // Only mount user-specific external storage 302 if (TEMP_FAILURE_RETRY( 303 mount(source_user.string(), target_user.string(), NULL, MS_BIND, NULL)) == -1) { 304 ALOGW("Failed to mount %s to %s: %d", source_user.string(), target_user.string(), errno); 305 return false; 306 } 307 } 308 309 if (fs_prepare_dir(legacy, 0000, 0, 0) == -1) { 310 return false; 311 } 312 313 // Finally, mount user-specific path into place for legacy users 314 if (TEMP_FAILURE_RETRY( 315 mount(target_user.string(), legacy, NULL, MS_BIND | MS_REC, NULL)) == -1) { 316 ALOGW("Failed to mount %s to %s: %d", target_user.string(), legacy, errno); 317 return false; 318 } 319 } else { 320 ALOGW("Mount mode %d unsupported", mount_mode); 321 return false; 322 } 323 324 return true; 325} 326 327static bool NeedsNoRandomizeWorkaround() { 328#if !defined(__arm__) 329 return false; 330#else 331 int major; 332 int minor; 333 struct utsname uts; 334 if (uname(&uts) == -1) { 335 return false; 336 } 337 338 if (sscanf(uts.release, "%d.%d", &major, &minor) != 2) { 339 return false; 340 } 341 342 // Kernels before 3.4.* need the workaround. 343 return (major < 3) || ((major == 3) && (minor < 4)); 344#endif 345} 346 347// Utility to close down the Zygote socket file descriptors while 348// the child is still running as root with Zygote's privileges. Each 349// descriptor (if any) is closed via dup2(), replacing it with a valid 350// (open) descriptor to /dev/null. 351 352static void DetachDescriptors(JNIEnv* env, jintArray fdsToClose) { 353 if (!fdsToClose) { 354 return; 355 } 356 jsize count = env->GetArrayLength(fdsToClose); 357 jint *ar = env->GetIntArrayElements(fdsToClose, 0); 358 if (!ar) { 359 ALOGE("Bad fd array"); 360 RuntimeAbort(env); 361 } 362 jsize i; 363 int devnull; 364 for (i = 0; i < count; i++) { 365 devnull = open("/dev/null", O_RDWR); 366 if (devnull < 0) { 367 ALOGE("Failed to open /dev/null"); 368 RuntimeAbort(env); 369 continue; 370 } 371 ALOGV("Switching descriptor %d to /dev/null: %d", ar[i], errno); 372 if (dup2(devnull, ar[i]) < 0) { 373 ALOGE("Failed dup2() on descriptor %d", ar[i]); 374 RuntimeAbort(env); 375 } 376 close(devnull); 377 } 378} 379 380void SetThreadName(const char* thread_name) { 381 bool hasAt = false; 382 bool hasDot = false; 383 const char* s = thread_name; 384 while (*s) { 385 if (*s == '.') { 386 hasDot = true; 387 } else if (*s == '@') { 388 hasAt = true; 389 } 390 s++; 391 } 392 const int len = s - thread_name; 393 if (len < 15 || hasAt || !hasDot) { 394 s = thread_name; 395 } else { 396 s = thread_name + len - 15; 397 } 398 // pthread_setname_np fails rather than truncating long strings. 399 char buf[16]; // MAX_TASK_COMM_LEN=16 is hard-coded into bionic 400 strlcpy(buf, s, sizeof(buf)-1); 401 errno = pthread_setname_np(pthread_self(), buf); 402 if (errno != 0) { 403 ALOGW("Unable to set the name of current thread to '%s'", buf); 404 } 405} 406 407// Utility routine to fork zygote and specialize the child process. 408static pid_t ForkAndSpecializeCommon(JNIEnv* env, uid_t uid, gid_t gid, jintArray javaGids, 409 jint debug_flags, jobjectArray javaRlimits, 410 jlong permittedCapabilities, jlong effectiveCapabilities, 411 jint mount_external, 412 jstring java_se_info, jstring java_se_name, 413 bool is_system_server, jintArray fdsToClose, 414 jstring instructionSet, jstring dataDir) { 415 SetSigChldHandler(); 416 417 pid_t pid = fork(); 418 419 if (pid == 0) { 420 // The child process. 421 gMallocLeakZygoteChild = 1; 422 423 // Clean up any descriptors which must be closed immediately 424 DetachDescriptors(env, fdsToClose); 425 426 // Keep capabilities across UID change, unless we're staying root. 427 if (uid != 0) { 428 EnableKeepCapabilities(env); 429 } 430 431 DropCapabilitiesBoundingSet(env); 432 433 bool need_native_bridge = false; 434 if (instructionSet != NULL) { 435 ScopedUtfChars isa_string(env, instructionSet); 436 need_native_bridge = android::NeedsNativeBridge(isa_string.c_str()); 437 } 438 439 if (!MountEmulatedStorage(uid, mount_external, need_native_bridge)) { 440 ALOGW("Failed to mount emulated storage: %d", errno); 441 if (errno == ENOTCONN || errno == EROFS) { 442 // When device is actively encrypting, we get ENOTCONN here 443 // since FUSE was mounted before the framework restarted. 444 // When encrypted device is booting, we get EROFS since 445 // FUSE hasn't been created yet by init. 446 // In either case, continue without external storage. 447 } else { 448 ALOGE("Cannot continue without emulated storage"); 449 RuntimeAbort(env); 450 } 451 } 452 453 if (!is_system_server) { 454 int rc = createProcessGroup(uid, getpid()); 455 if (rc != 0) { 456 if (rc == -EROFS) { 457 ALOGW("createProcessGroup failed, kernel missing CONFIG_CGROUP_CPUACCT?"); 458 } else { 459 ALOGE("createProcessGroup(%d, %d) failed: %s", uid, pid, strerror(-rc)); 460 } 461 } 462 } 463 464 SetGids(env, javaGids); 465 466 SetRLimits(env, javaRlimits); 467 468 if (!is_system_server && need_native_bridge) { 469 // Set the environment for the apps running with native bridge. 470 ScopedUtfChars isa_string(env, instructionSet); // Known non-null because of need_native_... 471 if (dataDir == NULL) { 472 android::PreInitializeNativeBridge(NULL, isa_string.c_str()); 473 } else { 474 ScopedUtfChars data_dir(env, dataDir); 475 android::PreInitializeNativeBridge(data_dir.c_str(), isa_string.c_str()); 476 } 477 } 478 479 int rc = setresgid(gid, gid, gid); 480 if (rc == -1) { 481 ALOGE("setresgid(%d) failed", gid); 482 RuntimeAbort(env); 483 } 484 485 rc = setresuid(uid, uid, uid); 486 if (rc == -1) { 487 ALOGE("setresuid(%d) failed", uid); 488 RuntimeAbort(env); 489 } 490 491 if (NeedsNoRandomizeWorkaround()) { 492 // Work around ARM kernel ASLR lossage (http://b/5817320). 493 int old_personality = personality(0xffffffff); 494 int new_personality = personality(old_personality | ADDR_NO_RANDOMIZE); 495 if (new_personality == -1) { 496 ALOGW("personality(%d) failed", new_personality); 497 } 498 } 499 500 SetCapabilities(env, permittedCapabilities, effectiveCapabilities); 501 502 SetSchedulerPolicy(env); 503 504 const char* se_info_c_str = NULL; 505 ScopedUtfChars* se_info = NULL; 506 if (java_se_info != NULL) { 507 se_info = new ScopedUtfChars(env, java_se_info); 508 se_info_c_str = se_info->c_str(); 509 if (se_info_c_str == NULL) { 510 ALOGE("se_info_c_str == NULL"); 511 RuntimeAbort(env); 512 } 513 } 514 const char* se_name_c_str = NULL; 515 ScopedUtfChars* se_name = NULL; 516 if (java_se_name != NULL) { 517 se_name = new ScopedUtfChars(env, java_se_name); 518 se_name_c_str = se_name->c_str(); 519 if (se_name_c_str == NULL) { 520 ALOGE("se_name_c_str == NULL"); 521 RuntimeAbort(env); 522 } 523 } 524 rc = selinux_android_setcontext(uid, is_system_server, se_info_c_str, se_name_c_str); 525 if (rc == -1) { 526 ALOGE("selinux_android_setcontext(%d, %d, \"%s\", \"%s\") failed", uid, 527 is_system_server, se_info_c_str, se_name_c_str); 528 RuntimeAbort(env); 529 } 530 531 // Make it easier to debug audit logs by setting the main thread's name to the 532 // nice name rather than "app_process". 533 if (se_info_c_str == NULL && is_system_server) { 534 se_name_c_str = "system_server"; 535 } 536 if (se_info_c_str != NULL) { 537 SetThreadName(se_name_c_str); 538 } 539 540 delete se_info; 541 delete se_name; 542 543 UnsetSigChldHandler(); 544 545 env->CallStaticVoidMethod(gZygoteClass, gCallPostForkChildHooks, debug_flags, 546 is_system_server ? NULL : instructionSet); 547 if (env->ExceptionCheck()) { 548 ALOGE("Error calling post fork hooks."); 549 RuntimeAbort(env); 550 } 551 } else if (pid > 0) { 552 // the parent process 553 } 554 return pid; 555} 556} // anonymous namespace 557 558namespace android { 559 560static jint com_android_internal_os_Zygote_nativeForkAndSpecialize( 561 JNIEnv* env, jclass, jint uid, jint gid, jintArray gids, 562 jint debug_flags, jobjectArray rlimits, 563 jint mount_external, jstring se_info, jstring se_name, 564 jintArray fdsToClose, jstring instructionSet, jstring appDataDir) { 565 // Grant CAP_WAKE_ALARM to the Bluetooth process. 566 jlong capabilities = 0; 567 if (uid == AID_BLUETOOTH) { 568 capabilities |= (1LL << CAP_WAKE_ALARM); 569 } 570 571 return ForkAndSpecializeCommon(env, uid, gid, gids, debug_flags, 572 rlimits, capabilities, capabilities, mount_external, se_info, 573 se_name, false, fdsToClose, instructionSet, appDataDir); 574} 575 576static jint com_android_internal_os_Zygote_nativeForkSystemServer( 577 JNIEnv* env, jclass, uid_t uid, gid_t gid, jintArray gids, 578 jint debug_flags, jobjectArray rlimits, jlong permittedCapabilities, 579 jlong effectiveCapabilities) { 580 pid_t pid = ForkAndSpecializeCommon(env, uid, gid, gids, 581 debug_flags, rlimits, 582 permittedCapabilities, effectiveCapabilities, 583 MOUNT_EXTERNAL_NONE, NULL, NULL, true, NULL, 584 NULL, NULL); 585 if (pid > 0) { 586 // The zygote process checks whether the child process has died or not. 587 ALOGI("System server process %d has been created", pid); 588 gSystemServerPid = pid; 589 // There is a slight window that the system server process has crashed 590 // but it went unnoticed because we haven't published its pid yet. So 591 // we recheck here just to make sure that all is well. 592 int status; 593 if (waitpid(pid, &status, WNOHANG) == pid) { 594 ALOGE("System server process %d has died. Restarting Zygote!", pid); 595 RuntimeAbort(env); 596 } 597 } 598 return pid; 599} 600 601static JNINativeMethod gMethods[] = { 602 { "nativeForkAndSpecialize", 603 "(II[II[[IILjava/lang/String;Ljava/lang/String;[ILjava/lang/String;Ljava/lang/String;)I", 604 (void *) com_android_internal_os_Zygote_nativeForkAndSpecialize }, 605 { "nativeForkSystemServer", "(II[II[[IJJ)I", 606 (void *) com_android_internal_os_Zygote_nativeForkSystemServer } 607}; 608 609int register_com_android_internal_os_Zygote(JNIEnv* env) { 610 gZygoteClass = (jclass) env->NewGlobalRef(env->FindClass(kZygoteClassName)); 611 if (gZygoteClass == NULL) { 612 RuntimeAbort(env); 613 } 614 gCallPostForkChildHooks = env->GetStaticMethodID(gZygoteClass, "callPostForkChildHooks", 615 "(ILjava/lang/String;)V"); 616 617 return AndroidRuntime::registerNativeMethods(env, "com/android/internal/os/Zygote", 618 gMethods, NELEM(gMethods)); 619} 620} // namespace android 621 622