com_android_internal_os_Zygote.cpp revision c0276a54ec20d83d0aa662c610ff5d24bc75fd33
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 <list> 24#include <sstream> 25#include <string> 26 27#include <fcntl.h> 28#include <grp.h> 29#include <inttypes.h> 30#include <malloc.h> 31#include <mntent.h> 32#include <paths.h> 33#include <signal.h> 34#include <stdlib.h> 35#include <sys/capability.h> 36#include <sys/cdefs.h> 37#include <sys/personality.h> 38#include <sys/prctl.h> 39#include <sys/resource.h> 40#include <sys/stat.h> 41#include <sys/time.h> 42#include <sys/types.h> 43#include <sys/utsname.h> 44#include <sys/wait.h> 45#include <unistd.h> 46 47#include "android-base/logging.h" 48#include <android-base/file.h> 49#include <android-base/stringprintf.h> 50#include <cutils/fs.h> 51#include <cutils/multiuser.h> 52#include <cutils/sched_policy.h> 53#include <private/android_filesystem_config.h> 54#include <utils/String8.h> 55#include <selinux/android.h> 56#include <seccomp_policy.h> 57#include <processgroup/processgroup.h> 58 59#include "core_jni_helpers.h" 60#include <nativehelper/JNIHelp.h> 61#include <nativehelper/ScopedLocalRef.h> 62#include <nativehelper/ScopedPrimitiveArray.h> 63#include <nativehelper/ScopedUtfChars.h> 64#include "fd_utils.h" 65 66#include "nativebridge/native_bridge.h" 67 68namespace { 69 70using android::String8; 71using android::base::StringPrintf; 72using android::base::WriteStringToFile; 73 74static pid_t gSystemServerPid = 0; 75 76static const char kZygoteClassName[] = "com/android/internal/os/Zygote"; 77static jclass gZygoteClass; 78static jmethodID gCallPostForkChildHooks; 79 80static bool g_is_security_enforced = true; 81 82// Must match values in com.android.internal.os.Zygote. 83enum MountExternalKind { 84 MOUNT_EXTERNAL_NONE = 0, 85 MOUNT_EXTERNAL_DEFAULT = 1, 86 MOUNT_EXTERNAL_READ = 2, 87 MOUNT_EXTERNAL_WRITE = 3, 88}; 89 90static void RuntimeAbort(JNIEnv* env, int line, const char* msg) { 91 std::ostringstream oss; 92 oss << __FILE__ << ":" << line << ": " << msg; 93 env->FatalError(oss.str().c_str()); 94} 95 96// This signal handler is for zygote mode, since the zygote must reap its children 97static void SigChldHandler(int /*signal_number*/) { 98 pid_t pid; 99 int status; 100 101 // It's necessary to save and restore the errno during this function. 102 // Since errno is stored per thread, changing it here modifies the errno 103 // on the thread on which this signal handler executes. If a signal occurs 104 // between a call and an errno check, it's possible to get the errno set 105 // here. 106 // See b/23572286 for extra information. 107 int saved_errno = errno; 108 109 while ((pid = waitpid(-1, &status, WNOHANG)) > 0) { 110 // Log process-death status that we care about. In general it is 111 // not safe to call LOG(...) from a signal handler because of 112 // possible reentrancy. However, we know a priori that the 113 // current implementation of LOG() is safe to call from a SIGCHLD 114 // handler in the zygote process. If the LOG() implementation 115 // changes its locking strategy or its use of syscalls within the 116 // lazy-init critical section, its use here may become unsafe. 117 if (WIFEXITED(status)) { 118 ALOGI("Process %d exited cleanly (%d)", pid, WEXITSTATUS(status)); 119 } else if (WIFSIGNALED(status)) { 120 ALOGI("Process %d exited due to signal (%d)", pid, WTERMSIG(status)); 121 if (WCOREDUMP(status)) { 122 ALOGI("Process %d dumped core.", pid); 123 } 124 } 125 126 // If the just-crashed process is the system_server, bring down zygote 127 // so that it is restarted by init and system server will be restarted 128 // from there. 129 if (pid == gSystemServerPid) { 130 ALOGE("Exit zygote because system server (%d) has terminated", pid); 131 kill(getpid(), SIGKILL); 132 } 133 } 134 135 // Note that we shouldn't consider ECHILD an error because 136 // the secondary zygote might have no children left to wait for. 137 if (pid < 0 && errno != ECHILD) { 138 ALOGW("Zygote SIGCHLD error in waitpid: %s", strerror(errno)); 139 } 140 141 errno = saved_errno; 142} 143 144// Configures the SIGCHLD/SIGHUP handlers for the zygote process. This is 145// configured very late, because earlier in the runtime we may fork() and 146// exec() other processes, and we want to waitpid() for those rather than 147// have them be harvested immediately. 148// 149// Ignore SIGHUP because all processes forked by the zygote are in the same 150// process group as the zygote and we don't want to be notified if we become 151// an orphaned group and have one or more stopped processes. This is not a 152// theoretical concern : 153// - we can become an orphaned group if one of our direct descendants forks 154// and is subsequently killed before its children. 155// - crash_dump routinely STOPs the process it's tracing. 156// 157// See issues b/71965619 and b/25567761 for further details. 158// 159// This ends up being called repeatedly before each fork(), but there's 160// no real harm in that. 161static void SetSignalHandlers() { 162 struct sigaction sig_chld = {}; 163 sig_chld.sa_handler = SigChldHandler; 164 165 if (sigaction(SIGCHLD, &sig_chld, NULL) < 0) { 166 ALOGW("Error setting SIGCHLD handler: %s", strerror(errno)); 167 } 168 169 struct sigaction sig_hup = {}; 170 sig_hup.sa_handler = SIG_IGN; 171 if (sigaction(SIGHUP, &sig_hup, NULL) < 0) { 172 ALOGW("Error setting SIGHUP handler: %s", strerror(errno)); 173 } 174} 175 176// Sets the SIGCHLD handler back to default behavior in zygote children. 177static void UnsetChldSignalHandler() { 178 struct sigaction sa; 179 memset(&sa, 0, sizeof(sa)); 180 sa.sa_handler = SIG_DFL; 181 182 if (sigaction(SIGCHLD, &sa, NULL) < 0) { 183 ALOGW("Error unsetting SIGCHLD handler: %s", strerror(errno)); 184 } 185} 186 187// Calls POSIX setgroups() using the int[] object as an argument. 188// A NULL argument is tolerated. 189static void SetGids(JNIEnv* env, jintArray javaGids) { 190 if (javaGids == NULL) { 191 return; 192 } 193 194 ScopedIntArrayRO gids(env, javaGids); 195 if (gids.get() == NULL) { 196 RuntimeAbort(env, __LINE__, "Getting gids int array failed"); 197 } 198 int rc = setgroups(gids.size(), reinterpret_cast<const gid_t*>(&gids[0])); 199 if (rc == -1) { 200 std::ostringstream oss; 201 oss << "setgroups failed: " << strerror(errno) << ", gids.size=" << gids.size(); 202 RuntimeAbort(env, __LINE__, oss.str().c_str()); 203 } 204} 205 206// Sets the resource limits via setrlimit(2) for the values in the 207// two-dimensional array of integers that's passed in. The second dimension 208// contains a tuple of length 3: (resource, rlim_cur, rlim_max). NULL is 209// treated as an empty array. 210static void SetRLimits(JNIEnv* env, jobjectArray javaRlimits) { 211 if (javaRlimits == NULL) { 212 return; 213 } 214 215 rlimit rlim; 216 memset(&rlim, 0, sizeof(rlim)); 217 218 for (int i = 0; i < env->GetArrayLength(javaRlimits); ++i) { 219 ScopedLocalRef<jobject> javaRlimitObject(env, env->GetObjectArrayElement(javaRlimits, i)); 220 ScopedIntArrayRO javaRlimit(env, reinterpret_cast<jintArray>(javaRlimitObject.get())); 221 if (javaRlimit.size() != 3) { 222 RuntimeAbort(env, __LINE__, "rlimits array must have a second dimension of size 3"); 223 } 224 225 rlim.rlim_cur = javaRlimit[1]; 226 rlim.rlim_max = javaRlimit[2]; 227 228 int rc = setrlimit(javaRlimit[0], &rlim); 229 if (rc == -1) { 230 ALOGE("setrlimit(%d, {%ld, %ld}) failed", javaRlimit[0], rlim.rlim_cur, 231 rlim.rlim_max); 232 RuntimeAbort(env, __LINE__, "setrlimit failed"); 233 } 234 } 235} 236 237// The debug malloc library needs to know whether it's the zygote or a child. 238extern "C" int gMallocLeakZygoteChild; 239 240static void PreApplicationInit() { 241 // The child process sets this to indicate it's not the zygote. 242 gMallocLeakZygoteChild = 1; 243 244 // Set the jemalloc decay time to 1. 245 mallopt(M_DECAY_TIME, 1); 246} 247 248static void SetUpSeccompFilter(uid_t uid) { 249 if (!g_is_security_enforced) { 250 ALOGI("seccomp disabled by setenforce 0"); 251 return; 252 } 253 254 // Apply system or app filter based on uid. 255 if (getuid() >= AID_APP_START) { 256 set_app_seccomp_filter(); 257 } else { 258 set_system_seccomp_filter(); 259 } 260} 261 262static void EnableKeepCapabilities(JNIEnv* env) { 263 int rc = prctl(PR_SET_KEEPCAPS, 1, 0, 0, 0); 264 if (rc == -1) { 265 RuntimeAbort(env, __LINE__, "prctl(PR_SET_KEEPCAPS) failed"); 266 } 267} 268 269static void DropCapabilitiesBoundingSet(JNIEnv* env) { 270 for (int i = 0; prctl(PR_CAPBSET_READ, i, 0, 0, 0) >= 0; i++) { 271 int rc = prctl(PR_CAPBSET_DROP, i, 0, 0, 0); 272 if (rc == -1) { 273 if (errno == EINVAL) { 274 ALOGE("prctl(PR_CAPBSET_DROP) failed with EINVAL. Please verify " 275 "your kernel is compiled with file capabilities support"); 276 } else { 277 ALOGE("prctl(PR_CAPBSET_DROP, %d) failed: %s", i, strerror(errno)); 278 RuntimeAbort(env, __LINE__, "prctl(PR_CAPBSET_DROP) failed"); 279 } 280 } 281 } 282} 283 284static void SetInheritable(JNIEnv* env, uint64_t inheritable) { 285 __user_cap_header_struct capheader; 286 memset(&capheader, 0, sizeof(capheader)); 287 capheader.version = _LINUX_CAPABILITY_VERSION_3; 288 capheader.pid = 0; 289 290 __user_cap_data_struct capdata[2]; 291 if (capget(&capheader, &capdata[0]) == -1) { 292 ALOGE("capget failed: %s", strerror(errno)); 293 RuntimeAbort(env, __LINE__, "capget failed"); 294 } 295 296 capdata[0].inheritable = inheritable; 297 capdata[1].inheritable = inheritable >> 32; 298 299 if (capset(&capheader, &capdata[0]) == -1) { 300 ALOGE("capset(inh=%" PRIx64 ") failed: %s", inheritable, strerror(errno)); 301 RuntimeAbort(env, __LINE__, "capset failed"); 302 } 303} 304 305static void SetCapabilities(JNIEnv* env, uint64_t permitted, uint64_t effective, 306 uint64_t inheritable) { 307 __user_cap_header_struct capheader; 308 memset(&capheader, 0, sizeof(capheader)); 309 capheader.version = _LINUX_CAPABILITY_VERSION_3; 310 capheader.pid = 0; 311 312 __user_cap_data_struct capdata[2]; 313 memset(&capdata, 0, sizeof(capdata)); 314 capdata[0].effective = effective; 315 capdata[1].effective = effective >> 32; 316 capdata[0].permitted = permitted; 317 capdata[1].permitted = permitted >> 32; 318 capdata[0].inheritable = inheritable; 319 capdata[1].inheritable = inheritable >> 32; 320 321 if (capset(&capheader, &capdata[0]) == -1) { 322 ALOGE("capset(perm=%" PRIx64 ", eff=%" PRIx64 ", inh=%" PRIx64 ") failed: %s", permitted, 323 effective, inheritable, strerror(errno)); 324 RuntimeAbort(env, __LINE__, "capset failed"); 325 } 326} 327 328static void SetSchedulerPolicy(JNIEnv* env) { 329 errno = -set_sched_policy(0, SP_DEFAULT); 330 if (errno != 0) { 331 ALOGE("set_sched_policy(0, SP_DEFAULT) failed"); 332 RuntimeAbort(env, __LINE__, "set_sched_policy(0, SP_DEFAULT) failed"); 333 } 334} 335 336static int UnmountTree(const char* path) { 337 size_t path_len = strlen(path); 338 339 FILE* fp = setmntent("/proc/mounts", "r"); 340 if (fp == NULL) { 341 ALOGE("Error opening /proc/mounts: %s", strerror(errno)); 342 return -errno; 343 } 344 345 // Some volumes can be stacked on each other, so force unmount in 346 // reverse order to give us the best chance of success. 347 std::list<std::string> toUnmount; 348 mntent* mentry; 349 while ((mentry = getmntent(fp)) != NULL) { 350 if (strncmp(mentry->mnt_dir, path, path_len) == 0) { 351 toUnmount.push_front(std::string(mentry->mnt_dir)); 352 } 353 } 354 endmntent(fp); 355 356 for (auto path : toUnmount) { 357 if (umount2(path.c_str(), MNT_DETACH)) { 358 ALOGW("Failed to unmount %s: %s", path.c_str(), strerror(errno)); 359 } 360 } 361 return 0; 362} 363 364// Create a private mount namespace and bind mount appropriate emulated 365// storage for the given user. 366static bool MountEmulatedStorage(uid_t uid, jint mount_mode, 367 bool force_mount_namespace) { 368 // See storage config details at http://source.android.com/tech/storage/ 369 370 String8 storageSource; 371 if (mount_mode == MOUNT_EXTERNAL_DEFAULT) { 372 storageSource = "/mnt/runtime/default"; 373 } else if (mount_mode == MOUNT_EXTERNAL_READ) { 374 storageSource = "/mnt/runtime/read"; 375 } else if (mount_mode == MOUNT_EXTERNAL_WRITE) { 376 storageSource = "/mnt/runtime/write"; 377 } else if (!force_mount_namespace) { 378 // Sane default of no storage visible 379 return true; 380 } 381 382 // Create a second private mount namespace for our process 383 if (unshare(CLONE_NEWNS) == -1) { 384 ALOGW("Failed to unshare(): %s", strerror(errno)); 385 return false; 386 } 387 388 // Handle force_mount_namespace with MOUNT_EXTERNAL_NONE. 389 if (mount_mode == MOUNT_EXTERNAL_NONE) { 390 return true; 391 } 392 393 if (TEMP_FAILURE_RETRY(mount(storageSource.string(), "/storage", 394 NULL, MS_BIND | MS_REC | MS_SLAVE, NULL)) == -1) { 395 ALOGW("Failed to mount %s to /storage: %s", storageSource.string(), strerror(errno)); 396 return false; 397 } 398 399 // Mount user-specific symlink helper into place 400 userid_t user_id = multiuser_get_user_id(uid); 401 const String8 userSource(String8::format("/mnt/user/%d", user_id)); 402 if (fs_prepare_dir(userSource.string(), 0751, 0, 0) == -1) { 403 return false; 404 } 405 if (TEMP_FAILURE_RETRY(mount(userSource.string(), "/storage/self", 406 NULL, MS_BIND, NULL)) == -1) { 407 ALOGW("Failed to mount %s to /storage/self: %s", userSource.string(), strerror(errno)); 408 return false; 409 } 410 411 return true; 412} 413 414static bool NeedsNoRandomizeWorkaround() { 415#if !defined(__arm__) 416 return false; 417#else 418 int major; 419 int minor; 420 struct utsname uts; 421 if (uname(&uts) == -1) { 422 return false; 423 } 424 425 if (sscanf(uts.release, "%d.%d", &major, &minor) != 2) { 426 return false; 427 } 428 429 // Kernels before 3.4.* need the workaround. 430 return (major < 3) || ((major == 3) && (minor < 4)); 431#endif 432} 433 434// Utility to close down the Zygote socket file descriptors while 435// the child is still running as root with Zygote's privileges. Each 436// descriptor (if any) is closed via dup2(), replacing it with a valid 437// (open) descriptor to /dev/null. 438 439static void DetachDescriptors(JNIEnv* env, jintArray fdsToClose) { 440 if (!fdsToClose) { 441 return; 442 } 443 jsize count = env->GetArrayLength(fdsToClose); 444 ScopedIntArrayRO ar(env, fdsToClose); 445 if (ar.get() == NULL) { 446 RuntimeAbort(env, __LINE__, "Bad fd array"); 447 } 448 jsize i; 449 int devnull; 450 for (i = 0; i < count; i++) { 451 devnull = open("/dev/null", O_RDWR); 452 if (devnull < 0) { 453 ALOGE("Failed to open /dev/null: %s", strerror(errno)); 454 RuntimeAbort(env, __LINE__, "Failed to open /dev/null"); 455 continue; 456 } 457 ALOGV("Switching descriptor %d to /dev/null: %s", ar[i], strerror(errno)); 458 if (dup2(devnull, ar[i]) < 0) { 459 ALOGE("Failed dup2() on descriptor %d: %s", ar[i], strerror(errno)); 460 RuntimeAbort(env, __LINE__, "Failed dup2()"); 461 } 462 close(devnull); 463 } 464} 465 466void SetThreadName(const char* thread_name) { 467 bool hasAt = false; 468 bool hasDot = false; 469 const char* s = thread_name; 470 while (*s) { 471 if (*s == '.') { 472 hasDot = true; 473 } else if (*s == '@') { 474 hasAt = true; 475 } 476 s++; 477 } 478 const int len = s - thread_name; 479 if (len < 15 || hasAt || !hasDot) { 480 s = thread_name; 481 } else { 482 s = thread_name + len - 15; 483 } 484 // pthread_setname_np fails rather than truncating long strings. 485 char buf[16]; // MAX_TASK_COMM_LEN=16 is hard-coded into bionic 486 strlcpy(buf, s, sizeof(buf)-1); 487 errno = pthread_setname_np(pthread_self(), buf); 488 if (errno != 0) { 489 ALOGW("Unable to set the name of current thread to '%s': %s", buf, strerror(errno)); 490 } 491} 492 493// The list of open zygote file descriptors. 494static FileDescriptorTable* gOpenFdTable = NULL; 495 496static void FillFileDescriptorVector(JNIEnv* env, 497 jintArray java_fds, 498 std::vector<int>* fds) { 499 CHECK(fds != nullptr); 500 if (java_fds != nullptr) { 501 ScopedIntArrayRO ar(env, java_fds); 502 if (ar.get() == nullptr) { 503 RuntimeAbort(env, __LINE__, "Bad fd array"); 504 } 505 fds->reserve(ar.size()); 506 for (size_t i = 0; i < ar.size(); ++i) { 507 fds->push_back(ar[i]); 508 } 509 } 510} 511 512// Utility routine to fork zygote and specialize the child process. 513static pid_t ForkAndSpecializeCommon(JNIEnv* env, uid_t uid, gid_t gid, jintArray javaGids, 514 jint runtime_flags, jobjectArray javaRlimits, 515 jlong permittedCapabilities, jlong effectiveCapabilities, 516 jint mount_external, 517 jstring java_se_info, jstring java_se_name, 518 bool is_system_server, jintArray fdsToClose, 519 jintArray fdsToIgnore, bool is_child_zygote, 520 jstring instructionSet, jstring dataDir) { 521 SetSignalHandlers(); 522 523 sigset_t sigchld; 524 sigemptyset(&sigchld); 525 sigaddset(&sigchld, SIGCHLD); 526 527 // Temporarily block SIGCHLD during forks. The SIGCHLD handler might 528 // log, which would result in the logging FDs we close being reopened. 529 // This would cause failures because the FDs are not whitelisted. 530 // 531 // Note that the zygote process is single threaded at this point. 532 if (sigprocmask(SIG_BLOCK, &sigchld, nullptr) == -1) { 533 ALOGE("sigprocmask(SIG_SETMASK, { SIGCHLD }) failed: %s", strerror(errno)); 534 RuntimeAbort(env, __LINE__, "Call to sigprocmask(SIG_BLOCK, { SIGCHLD }) failed."); 535 } 536 537 // Close any logging related FDs before we start evaluating the list of 538 // file descriptors. 539 __android_log_close(); 540 541 // If this is the first fork for this zygote, create the open FD table. 542 // If it isn't, we just need to check whether the list of open files has 543 // changed (and it shouldn't in the normal case). 544 std::vector<int> fds_to_ignore; 545 FillFileDescriptorVector(env, fdsToIgnore, &fds_to_ignore); 546 if (gOpenFdTable == NULL) { 547 gOpenFdTable = FileDescriptorTable::Create(fds_to_ignore); 548 if (gOpenFdTable == NULL) { 549 RuntimeAbort(env, __LINE__, "Unable to construct file descriptor table."); 550 } 551 } else if (!gOpenFdTable->Restat(fds_to_ignore)) { 552 RuntimeAbort(env, __LINE__, "Unable to restat file descriptor table."); 553 } 554 555 pid_t pid = fork(); 556 557 if (pid == 0) { 558 PreApplicationInit(); 559 560 // Clean up any descriptors which must be closed immediately 561 DetachDescriptors(env, fdsToClose); 562 563 // Re-open all remaining open file descriptors so that they aren't shared 564 // with the zygote across a fork. 565 if (!gOpenFdTable->ReopenOrDetach()) { 566 RuntimeAbort(env, __LINE__, "Unable to reopen whitelisted descriptors."); 567 } 568 569 if (sigprocmask(SIG_UNBLOCK, &sigchld, nullptr) == -1) { 570 ALOGE("sigprocmask(SIG_SETMASK, { SIGCHLD }) failed: %s", strerror(errno)); 571 RuntimeAbort(env, __LINE__, "Call to sigprocmask(SIG_UNBLOCK, { SIGCHLD }) failed."); 572 } 573 574 // Must be called when the new process still has CAP_SYS_ADMIN. The other alternative is to 575 // call prctl(PR_SET_NO_NEW_PRIVS, 1) afterward, but that breaks SELinux domain transition (see 576 // b/71859146). 577 SetUpSeccompFilter(uid); 578 579 // Keep capabilities across UID change, unless we're staying root. 580 if (uid != 0) { 581 EnableKeepCapabilities(env); 582 } 583 584 SetInheritable(env, permittedCapabilities); 585 DropCapabilitiesBoundingSet(env); 586 587 bool use_native_bridge = !is_system_server && (instructionSet != NULL) 588 && android::NativeBridgeAvailable(); 589 if (use_native_bridge) { 590 ScopedUtfChars isa_string(env, instructionSet); 591 use_native_bridge = android::NeedsNativeBridge(isa_string.c_str()); 592 } 593 if (use_native_bridge && dataDir == NULL) { 594 // dataDir should never be null if we need to use a native bridge. 595 // In general, dataDir will never be null for normal applications. It can only happen in 596 // special cases (for isolated processes which are not associated with any app). These are 597 // launched by the framework and should not be emulated anyway. 598 use_native_bridge = false; 599 ALOGW("Native bridge will not be used because dataDir == NULL."); 600 } 601 602 if (!MountEmulatedStorage(uid, mount_external, use_native_bridge)) { 603 ALOGW("Failed to mount emulated storage: %s", strerror(errno)); 604 if (errno == ENOTCONN || errno == EROFS) { 605 // When device is actively encrypting, we get ENOTCONN here 606 // since FUSE was mounted before the framework restarted. 607 // When encrypted device is booting, we get EROFS since 608 // FUSE hasn't been created yet by init. 609 // In either case, continue without external storage. 610 } else { 611 RuntimeAbort(env, __LINE__, "Cannot continue without emulated storage"); 612 } 613 } 614 615 if (!is_system_server) { 616 int rc = createProcessGroup(uid, getpid()); 617 if (rc != 0) { 618 if (rc == -EROFS) { 619 ALOGW("createProcessGroup failed, kernel missing CONFIG_CGROUP_CPUACCT?"); 620 } else { 621 ALOGE("createProcessGroup(%d, %d) failed: %s", uid, pid, strerror(-rc)); 622 } 623 } 624 } 625 626 SetGids(env, javaGids); 627 628 SetRLimits(env, javaRlimits); 629 630 if (use_native_bridge) { 631 ScopedUtfChars isa_string(env, instructionSet); 632 ScopedUtfChars data_dir(env, dataDir); 633 android::PreInitializeNativeBridge(data_dir.c_str(), isa_string.c_str()); 634 } 635 636 int rc = setresgid(gid, gid, gid); 637 if (rc == -1) { 638 ALOGE("setresgid(%d) failed: %s", gid, strerror(errno)); 639 RuntimeAbort(env, __LINE__, "setresgid failed"); 640 } 641 642 rc = setresuid(uid, uid, uid); 643 if (rc == -1) { 644 ALOGE("setresuid(%d) failed: %s", uid, strerror(errno)); 645 RuntimeAbort(env, __LINE__, "setresuid failed"); 646 } 647 648 if (NeedsNoRandomizeWorkaround()) { 649 // Work around ARM kernel ASLR lossage (http://b/5817320). 650 int old_personality = personality(0xffffffff); 651 int new_personality = personality(old_personality | ADDR_NO_RANDOMIZE); 652 if (new_personality == -1) { 653 ALOGW("personality(%d) failed: %s", new_personality, strerror(errno)); 654 } 655 } 656 657 SetCapabilities(env, permittedCapabilities, effectiveCapabilities, permittedCapabilities); 658 659 SetSchedulerPolicy(env); 660 661 const char* se_info_c_str = NULL; 662 ScopedUtfChars* se_info = NULL; 663 if (java_se_info != NULL) { 664 se_info = new ScopedUtfChars(env, java_se_info); 665 se_info_c_str = se_info->c_str(); 666 if (se_info_c_str == NULL) { 667 RuntimeAbort(env, __LINE__, "se_info_c_str == NULL"); 668 } 669 } 670 const char* se_name_c_str = NULL; 671 ScopedUtfChars* se_name = NULL; 672 if (java_se_name != NULL) { 673 se_name = new ScopedUtfChars(env, java_se_name); 674 se_name_c_str = se_name->c_str(); 675 if (se_name_c_str == NULL) { 676 RuntimeAbort(env, __LINE__, "se_name_c_str == NULL"); 677 } 678 } 679 rc = selinux_android_setcontext(uid, is_system_server, se_info_c_str, se_name_c_str); 680 if (rc == -1) { 681 ALOGE("selinux_android_setcontext(%d, %d, \"%s\", \"%s\") failed", uid, 682 is_system_server, se_info_c_str, se_name_c_str); 683 RuntimeAbort(env, __LINE__, "selinux_android_setcontext failed"); 684 } 685 686 // Make it easier to debug audit logs by setting the main thread's name to the 687 // nice name rather than "app_process". 688 if (se_info_c_str == NULL && is_system_server) { 689 se_name_c_str = "system_server"; 690 } 691 if (se_info_c_str != NULL) { 692 SetThreadName(se_name_c_str); 693 } 694 695 delete se_info; 696 delete se_name; 697 698 // Unset the SIGCHLD handler, but keep ignoring SIGHUP (rationale in SetSignalHandlers). 699 UnsetChldSignalHandler(); 700 701 env->CallStaticVoidMethod(gZygoteClass, gCallPostForkChildHooks, runtime_flags, 702 is_system_server, is_child_zygote, instructionSet); 703 if (env->ExceptionCheck()) { 704 RuntimeAbort(env, __LINE__, "Error calling post fork hooks."); 705 } 706 } else if (pid > 0) { 707 // the parent process 708 709 // We blocked SIGCHLD prior to a fork, we unblock it here. 710 if (sigprocmask(SIG_UNBLOCK, &sigchld, nullptr) == -1) { 711 ALOGE("sigprocmask(SIG_SETMASK, { SIGCHLD }) failed: %s", strerror(errno)); 712 RuntimeAbort(env, __LINE__, "Call to sigprocmask(SIG_UNBLOCK, { SIGCHLD }) failed."); 713 } 714 } 715 return pid; 716} 717 718static uint64_t GetEffectiveCapabilityMask(JNIEnv* env) { 719 __user_cap_header_struct capheader; 720 memset(&capheader, 0, sizeof(capheader)); 721 capheader.version = _LINUX_CAPABILITY_VERSION_3; 722 capheader.pid = 0; 723 724 __user_cap_data_struct capdata[2]; 725 if (capget(&capheader, &capdata[0]) == -1) { 726 ALOGE("capget failed: %s", strerror(errno)); 727 RuntimeAbort(env, __LINE__, "capget failed"); 728 } 729 730 return capdata[0].effective | 731 (static_cast<uint64_t>(capdata[1].effective) << 32); 732} 733} // anonymous namespace 734 735namespace android { 736 737static void com_android_internal_os_Zygote_nativeSecurityInit(JNIEnv*, jclass) { 738 // security_getenforce is not allowed on app process. Initialize and cache the value before 739 // zygote forks. 740 g_is_security_enforced = security_getenforce(); 741} 742 743static void com_android_internal_os_Zygote_nativePreApplicationInit(JNIEnv*, jclass) { 744 PreApplicationInit(); 745} 746 747static jint com_android_internal_os_Zygote_nativeForkAndSpecialize( 748 JNIEnv* env, jclass, jint uid, jint gid, jintArray gids, 749 jint runtime_flags, jobjectArray rlimits, 750 jint mount_external, jstring se_info, jstring se_name, 751 jintArray fdsToClose, jintArray fdsToIgnore, jboolean is_child_zygote, 752 jstring instructionSet, jstring appDataDir) { 753 jlong capabilities = 0; 754 755 // Grant CAP_WAKE_ALARM to the Bluetooth process. 756 // Additionally, allow bluetooth to open packet sockets so it can start the DHCP client. 757 // Grant CAP_SYS_NICE to allow Bluetooth to set RT priority for 758 // audio-related threads. 759 // TODO: consider making such functionality an RPC to netd. 760 if (multiuser_get_app_id(uid) == AID_BLUETOOTH) { 761 capabilities |= (1LL << CAP_WAKE_ALARM); 762 capabilities |= (1LL << CAP_NET_RAW); 763 capabilities |= (1LL << CAP_NET_BIND_SERVICE); 764 capabilities |= (1LL << CAP_SYS_NICE); 765 } 766 767 // Grant CAP_BLOCK_SUSPEND to processes that belong to GID "wakelock" 768 bool gid_wakelock_found = false; 769 if (gid == AID_WAKELOCK) { 770 gid_wakelock_found = true; 771 } else if (gids != NULL) { 772 jsize gids_num = env->GetArrayLength(gids); 773 ScopedIntArrayRO ar(env, gids); 774 if (ar.get() == NULL) { 775 RuntimeAbort(env, __LINE__, "Bad gids array"); 776 } 777 for (int i = 0; i < gids_num; i++) { 778 if (ar[i] == AID_WAKELOCK) { 779 gid_wakelock_found = true; 780 break; 781 } 782 } 783 } 784 if (gid_wakelock_found) { 785 capabilities |= (1LL << CAP_BLOCK_SUSPEND); 786 } 787 788 // If forking a child zygote process, that zygote will need to be able to change 789 // the UID and GID of processes it forks, as well as drop those capabilities. 790 if (is_child_zygote) { 791 capabilities |= (1LL << CAP_SETUID); 792 capabilities |= (1LL << CAP_SETGID); 793 capabilities |= (1LL << CAP_SETPCAP); 794 } 795 796 // Containers run without some capabilities, so drop any caps that are not 797 // available. 798 capabilities &= GetEffectiveCapabilityMask(env); 799 800 return ForkAndSpecializeCommon(env, uid, gid, gids, runtime_flags, 801 rlimits, capabilities, capabilities, mount_external, se_info, 802 se_name, false, fdsToClose, fdsToIgnore, is_child_zygote == JNI_TRUE, 803 instructionSet, appDataDir); 804} 805 806static jint com_android_internal_os_Zygote_nativeForkSystemServer( 807 JNIEnv* env, jclass, uid_t uid, gid_t gid, jintArray gids, 808 jint runtime_flags, jobjectArray rlimits, jlong permittedCapabilities, 809 jlong effectiveCapabilities) { 810 pid_t pid = ForkAndSpecializeCommon(env, uid, gid, gids, 811 runtime_flags, rlimits, 812 permittedCapabilities, effectiveCapabilities, 813 MOUNT_EXTERNAL_DEFAULT, NULL, NULL, true, NULL, 814 NULL, false, NULL, NULL); 815 if (pid > 0) { 816 // The zygote process checks whether the child process has died or not. 817 ALOGI("System server process %d has been created", pid); 818 gSystemServerPid = pid; 819 // There is a slight window that the system server process has crashed 820 // but it went unnoticed because we haven't published its pid yet. So 821 // we recheck here just to make sure that all is well. 822 int status; 823 if (waitpid(pid, &status, WNOHANG) == pid) { 824 ALOGE("System server process %d has died. Restarting Zygote!", pid); 825 RuntimeAbort(env, __LINE__, "System server process has died. Restarting Zygote!"); 826 } 827 828 // Assign system_server to the correct memory cgroup. 829 // Not all devices mount /dev/memcg so check for the file first 830 // to avoid unnecessarily printing errors and denials in the logs. 831 if (!access("/dev/memcg/system/tasks", F_OK) && 832 !WriteStringToFile(StringPrintf("%d", pid), "/dev/memcg/system/tasks")) { 833 ALOGE("couldn't write %d to /dev/memcg/system/tasks", pid); 834 } 835 } 836 return pid; 837} 838 839static void com_android_internal_os_Zygote_nativeAllowFileAcrossFork( 840 JNIEnv* env, jclass, jstring path) { 841 ScopedUtfChars path_native(env, path); 842 const char* path_cstr = path_native.c_str(); 843 if (!path_cstr) { 844 RuntimeAbort(env, __LINE__, "path_cstr == NULL"); 845 } 846 FileDescriptorWhitelist::Get()->Allow(path_cstr); 847} 848 849static void com_android_internal_os_Zygote_nativeUnmountStorageOnInit(JNIEnv* env, jclass) { 850 // Zygote process unmount root storage space initially before every child processes are forked. 851 // Every forked child processes (include SystemServer) only mount their own root storage space 852 // and no need unmount storage operation in MountEmulatedStorage method. 853 // Zygote process does not utilize root storage spaces and unshares its mount namespace below. 854 855 // See storage config details at http://source.android.com/tech/storage/ 856 // Create private mount namespace shared by all children 857 if (unshare(CLONE_NEWNS) == -1) { 858 RuntimeAbort(env, __LINE__, "Failed to unshare()"); 859 return; 860 } 861 862 // Mark rootfs as being a slave so that changes from default 863 // namespace only flow into our children. 864 if (mount("rootfs", "/", nullptr, (MS_SLAVE | MS_REC), nullptr) == -1) { 865 RuntimeAbort(env, __LINE__, "Failed to mount() rootfs as MS_SLAVE"); 866 return; 867 } 868 869 // Create a staging tmpfs that is shared by our children; they will 870 // bind mount storage into their respective private namespaces, which 871 // are isolated from each other. 872 const char* target_base = getenv("EMULATED_STORAGE_TARGET"); 873 if (target_base != nullptr) { 874#define STRINGIFY_UID(x) __STRING(x) 875 if (mount("tmpfs", target_base, "tmpfs", MS_NOSUID | MS_NODEV, 876 "uid=0,gid=" STRINGIFY_UID(AID_SDCARD_R) ",mode=0751") == -1) { 877 ALOGE("Failed to mount tmpfs to %s", target_base); 878 RuntimeAbort(env, __LINE__, "Failed to mount tmpfs"); 879 return; 880 } 881#undef STRINGIFY_UID 882 } 883 884 UnmountTree("/storage"); 885} 886 887static const JNINativeMethod gMethods[] = { 888 { "nativeSecurityInit", "()V", 889 (void *) com_android_internal_os_Zygote_nativeSecurityInit }, 890 { "nativeForkAndSpecialize", 891 "(II[II[[IILjava/lang/String;Ljava/lang/String;[I[IZLjava/lang/String;Ljava/lang/String;)I", 892 (void *) com_android_internal_os_Zygote_nativeForkAndSpecialize }, 893 { "nativeForkSystemServer", "(II[II[[IJJ)I", 894 (void *) com_android_internal_os_Zygote_nativeForkSystemServer }, 895 { "nativeAllowFileAcrossFork", "(Ljava/lang/String;)V", 896 (void *) com_android_internal_os_Zygote_nativeAllowFileAcrossFork }, 897 { "nativeUnmountStorageOnInit", "()V", 898 (void *) com_android_internal_os_Zygote_nativeUnmountStorageOnInit }, 899 { "nativePreApplicationInit", "()V", 900 (void *) com_android_internal_os_Zygote_nativePreApplicationInit } 901}; 902 903int register_com_android_internal_os_Zygote(JNIEnv* env) { 904 gZygoteClass = MakeGlobalRefOrDie(env, FindClassOrDie(env, kZygoteClassName)); 905 gCallPostForkChildHooks = GetStaticMethodIDOrDie(env, gZygoteClass, "callPostForkChildHooks", 906 "(IZZLjava/lang/String;)V"); 907 908 return RegisterMethodsOrDie(env, "com/android/internal/os/Zygote", gMethods, NELEM(gMethods)); 909} 910} // namespace android 911