com_android_internal_os_Zygote.cpp revision 54e387ddbe6a0462bc8e9e15c7c7b3463adfcb24
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 <mntent.h>
31#include <paths.h>
32#include <signal.h>
33#include <stdlib.h>
34#include <sys/capability.h>
35#include <sys/cdefs.h>
36#include <sys/personality.h>
37#include <sys/prctl.h>
38#include <sys/resource.h>
39#include <sys/stat.h>
40#include <sys/time.h>
41#include <sys/types.h>
42#include <sys/utsname.h>
43#include <sys/wait.h>
44#include <unistd.h>
45
46#include <cutils/fs.h>
47#include <cutils/multiuser.h>
48#include <cutils/sched_policy.h>
49#include <private/android_filesystem_config.h>
50#include <utils/String8.h>
51#include <selinux/android.h>
52#include <processgroup/processgroup.h>
53
54#include "core_jni_helpers.h"
55#include "JNIHelp.h"
56#include "ScopedLocalRef.h"
57#include "ScopedPrimitiveArray.h"
58#include "ScopedUtfChars.h"
59#include "fd_utils.h"
60
61#include "nativebridge/native_bridge.h"
62
63namespace {
64
65using android::String8;
66
67static pid_t gSystemServerPid = 0;
68
69static const char kZygoteClassName[] = "com/android/internal/os/Zygote";
70static jclass gZygoteClass;
71static jmethodID gCallPostForkChildHooks;
72
73// Must match values in com.android.internal.os.Zygote.
74enum MountExternalKind {
75  MOUNT_EXTERNAL_NONE = 0,
76  MOUNT_EXTERNAL_DEFAULT = 1,
77  MOUNT_EXTERNAL_READ = 2,
78  MOUNT_EXTERNAL_WRITE = 3,
79};
80
81static void RuntimeAbort(JNIEnv* env, int line, const char* msg) {
82  std::ostringstream oss;
83  oss << __FILE__ << ":" << line << ": " << msg;
84  env->FatalError(oss.str().c_str());
85}
86
87// This signal handler is for zygote mode, since the zygote must reap its children
88static void SigChldHandler(int /*signal_number*/) {
89  pid_t pid;
90  int status;
91
92  // It's necessary to save and restore the errno during this function.
93  // Since errno is stored per thread, changing it here modifies the errno
94  // on the thread on which this signal handler executes. If a signal occurs
95  // between a call and an errno check, it's possible to get the errno set
96  // here.
97  // See b/23572286 for extra information.
98  int saved_errno = errno;
99
100  while ((pid = waitpid(-1, &status, WNOHANG)) > 0) {
101     // Log process-death status that we care about.  In general it is
102     // not safe to call LOG(...) from a signal handler because of
103     // possible reentrancy.  However, we know a priori that the
104     // current implementation of LOG() is safe to call from a SIGCHLD
105     // handler in the zygote process.  If the LOG() implementation
106     // changes its locking strategy or its use of syscalls within the
107     // lazy-init critical section, its use here may become unsafe.
108    if (WIFEXITED(status)) {
109      if (WEXITSTATUS(status)) {
110        ALOGI("Process %d exited cleanly (%d)", pid, WEXITSTATUS(status));
111      }
112    } else if (WIFSIGNALED(status)) {
113      if (WTERMSIG(status) != SIGKILL) {
114        ALOGI("Process %d exited due to signal (%d)", pid, WTERMSIG(status));
115      }
116      if (WCOREDUMP(status)) {
117        ALOGI("Process %d dumped core.", pid);
118      }
119    }
120
121    // If the just-crashed process is the system_server, bring down zygote
122    // so that it is restarted by init and system server will be restarted
123    // from there.
124    if (pid == gSystemServerPid) {
125      ALOGE("Exit zygote because system server (%d) has terminated", pid);
126      kill(getpid(), SIGKILL);
127    }
128  }
129
130  // Note that we shouldn't consider ECHILD an error because
131  // the secondary zygote might have no children left to wait for.
132  if (pid < 0 && errno != ECHILD) {
133    ALOGW("Zygote SIGCHLD error in waitpid: %s", strerror(errno));
134  }
135
136  errno = saved_errno;
137}
138
139// Configures the SIGCHLD handler for the zygote process. This is configured
140// very late, because earlier in the runtime we may fork() and exec()
141// other processes, and we want to waitpid() for those rather than
142// have them be harvested immediately.
143//
144// This ends up being called repeatedly before each fork(), but there's
145// no real harm in that.
146static void SetSigChldHandler() {
147  struct sigaction sa;
148  memset(&sa, 0, sizeof(sa));
149  sa.sa_handler = SigChldHandler;
150
151  int err = sigaction(SIGCHLD, &sa, NULL);
152  if (err < 0) {
153    ALOGW("Error setting SIGCHLD handler: %s", strerror(errno));
154  }
155}
156
157// Resets nice priority for zygote process. Zygote priority can be set
158// to high value during boot phase to speed it up. We want to ensure
159// zygote is running at normal priority before childs are forked from it.
160//
161// This ends up being called repeatedly before each fork(), but there's
162// no real harm in that.
163static void ResetNicePriority(JNIEnv* env) {
164  errno = 0;
165  int prio = getpriority(PRIO_PROCESS, 0);
166  if (prio == -1 && errno != 0) {
167    ALOGW("getpriority failed: %s\n", strerror(errno));
168  }
169  if (prio != 0 && setpriority(PRIO_PROCESS, 0, 0) != 0) {
170    ALOGE("setpriority(%d, 0, 0) failed: %s", PRIO_PROCESS, strerror(errno));
171    RuntimeAbort(env, __LINE__, "setpriority failed");
172  }
173}
174
175// Sets the SIGCHLD handler back to default behavior in zygote children.
176static void UnsetSigChldHandler() {
177  struct sigaction sa;
178  memset(&sa, 0, sizeof(sa));
179  sa.sa_handler = SIG_DFL;
180
181  int err = sigaction(SIGCHLD, &sa, NULL);
182  if (err < 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 EnableKeepCapabilities(JNIEnv* env) {
241  int rc = prctl(PR_SET_KEEPCAPS, 1, 0, 0, 0);
242  if (rc == -1) {
243    RuntimeAbort(env, __LINE__, "prctl(PR_SET_KEEPCAPS) failed");
244  }
245}
246
247static void DropCapabilitiesBoundingSet(JNIEnv* env) {
248  for (int i = 0; prctl(PR_CAPBSET_READ, i, 0, 0, 0) >= 0; i++) {
249    int rc = prctl(PR_CAPBSET_DROP, i, 0, 0, 0);
250    if (rc == -1) {
251      if (errno == EINVAL) {
252        ALOGE("prctl(PR_CAPBSET_DROP) failed with EINVAL. Please verify "
253              "your kernel is compiled with file capabilities support");
254      } else {
255        RuntimeAbort(env, __LINE__, "prctl(PR_CAPBSET_DROP) failed");
256      }
257    }
258  }
259}
260
261static void SetCapabilities(JNIEnv* env, int64_t permitted, int64_t effective) {
262  __user_cap_header_struct capheader;
263  memset(&capheader, 0, sizeof(capheader));
264  capheader.version = _LINUX_CAPABILITY_VERSION_3;
265  capheader.pid = 0;
266
267  __user_cap_data_struct capdata[2];
268  memset(&capdata, 0, sizeof(capdata));
269  capdata[0].effective = effective;
270  capdata[1].effective = effective >> 32;
271  capdata[0].permitted = permitted;
272  capdata[1].permitted = permitted >> 32;
273
274  if (capset(&capheader, &capdata[0]) == -1) {
275    ALOGE("capset(%" PRId64 ", %" PRId64 ") failed", permitted, effective);
276    RuntimeAbort(env, __LINE__, "capset failed");
277  }
278}
279
280static void SetSchedulerPolicy(JNIEnv* env) {
281  errno = -set_sched_policy(0, SP_DEFAULT);
282  if (errno != 0) {
283    ALOGE("set_sched_policy(0, SP_DEFAULT) failed");
284    RuntimeAbort(env, __LINE__, "set_sched_policy(0, SP_DEFAULT) failed");
285  }
286}
287
288static int UnmountTree(const char* path) {
289    size_t path_len = strlen(path);
290
291    FILE* fp = setmntent("/proc/mounts", "r");
292    if (fp == NULL) {
293        ALOGE("Error opening /proc/mounts: %s", strerror(errno));
294        return -errno;
295    }
296
297    // Some volumes can be stacked on each other, so force unmount in
298    // reverse order to give us the best chance of success.
299    std::list<std::string> toUnmount;
300    mntent* mentry;
301    while ((mentry = getmntent(fp)) != NULL) {
302        if (strncmp(mentry->mnt_dir, path, path_len) == 0) {
303            toUnmount.push_front(std::string(mentry->mnt_dir));
304        }
305    }
306    endmntent(fp);
307
308    for (auto path : toUnmount) {
309        if (umount2(path.c_str(), MNT_DETACH)) {
310            ALOGW("Failed to unmount %s: %s", path.c_str(), strerror(errno));
311        }
312    }
313    return 0;
314}
315
316// Create a private mount namespace and bind mount appropriate emulated
317// storage for the given user.
318static bool MountEmulatedStorage(uid_t uid, jint mount_mode,
319        bool force_mount_namespace) {
320    // See storage config details at http://source.android.com/tech/storage/
321
322    String8 storageSource;
323    if (mount_mode == MOUNT_EXTERNAL_DEFAULT) {
324        storageSource = "/mnt/runtime/default";
325    } else if (mount_mode == MOUNT_EXTERNAL_READ) {
326        storageSource = "/mnt/runtime/read";
327    } else if (mount_mode == MOUNT_EXTERNAL_WRITE) {
328        storageSource = "/mnt/runtime/write";
329    } else if (!force_mount_namespace) {
330        // Sane default of no storage visible
331        return true;
332    }
333
334    // Create a second private mount namespace for our process
335    if (unshare(CLONE_NEWNS) == -1) {
336        ALOGW("Failed to unshare(): %s", strerror(errno));
337        return false;
338    }
339
340    if (TEMP_FAILURE_RETRY(mount(storageSource.string(), "/storage",
341            NULL, MS_BIND | MS_REC | MS_SLAVE, NULL)) == -1) {
342        ALOGW("Failed to mount %s to /storage: %s", storageSource.string(), strerror(errno));
343        return false;
344    }
345
346    // Mount user-specific symlink helper into place
347    userid_t user_id = multiuser_get_user_id(uid);
348    const String8 userSource(String8::format("/mnt/user/%d", user_id));
349    if (fs_prepare_dir(userSource.string(), 0751, 0, 0) == -1) {
350        return false;
351    }
352    if (TEMP_FAILURE_RETRY(mount(userSource.string(), "/storage/self",
353            NULL, MS_BIND, NULL)) == -1) {
354        ALOGW("Failed to mount %s to /storage/self: %s", userSource.string(), strerror(errno));
355        return false;
356    }
357
358    return true;
359}
360
361static bool NeedsNoRandomizeWorkaround() {
362#if !defined(__arm__)
363    return false;
364#else
365    int major;
366    int minor;
367    struct utsname uts;
368    if (uname(&uts) == -1) {
369        return false;
370    }
371
372    if (sscanf(uts.release, "%d.%d", &major, &minor) != 2) {
373        return false;
374    }
375
376    // Kernels before 3.4.* need the workaround.
377    return (major < 3) || ((major == 3) && (minor < 4));
378#endif
379}
380
381// Utility to close down the Zygote socket file descriptors while
382// the child is still running as root with Zygote's privileges.  Each
383// descriptor (if any) is closed via dup2(), replacing it with a valid
384// (open) descriptor to /dev/null.
385
386static void DetachDescriptors(JNIEnv* env, jintArray fdsToClose) {
387  if (!fdsToClose) {
388    return;
389  }
390  jsize count = env->GetArrayLength(fdsToClose);
391  ScopedIntArrayRO ar(env, fdsToClose);
392  if (ar.get() == NULL) {
393      RuntimeAbort(env, __LINE__, "Bad fd array");
394  }
395  jsize i;
396  int devnull;
397  for (i = 0; i < count; i++) {
398    devnull = open("/dev/null", O_RDWR);
399    if (devnull < 0) {
400      ALOGE("Failed to open /dev/null: %s", strerror(errno));
401      RuntimeAbort(env, __LINE__, "Failed to open /dev/null");
402      continue;
403    }
404    ALOGV("Switching descriptor %d to /dev/null: %s", ar[i], strerror(errno));
405    if (dup2(devnull, ar[i]) < 0) {
406      ALOGE("Failed dup2() on descriptor %d: %s", ar[i], strerror(errno));
407      RuntimeAbort(env, __LINE__, "Failed dup2()");
408    }
409    close(devnull);
410  }
411}
412
413void SetThreadName(const char* thread_name) {
414  bool hasAt = false;
415  bool hasDot = false;
416  const char* s = thread_name;
417  while (*s) {
418    if (*s == '.') {
419      hasDot = true;
420    } else if (*s == '@') {
421      hasAt = true;
422    }
423    s++;
424  }
425  const int len = s - thread_name;
426  if (len < 15 || hasAt || !hasDot) {
427    s = thread_name;
428  } else {
429    s = thread_name + len - 15;
430  }
431  // pthread_setname_np fails rather than truncating long strings.
432  char buf[16];       // MAX_TASK_COMM_LEN=16 is hard-coded into bionic
433  strlcpy(buf, s, sizeof(buf)-1);
434  errno = pthread_setname_np(pthread_self(), buf);
435  if (errno != 0) {
436    ALOGW("Unable to set the name of current thread to '%s': %s", buf, strerror(errno));
437  }
438}
439
440// The list of open zygote file descriptors.
441static FileDescriptorTable* gOpenFdTable = NULL;
442
443// Utility routine to fork zygote and specialize the child process.
444static pid_t ForkAndSpecializeCommon(JNIEnv* env, uid_t uid, gid_t gid, jintArray javaGids,
445                                     jint debug_flags, jobjectArray javaRlimits,
446                                     jlong permittedCapabilities, jlong effectiveCapabilities,
447                                     jint mount_external,
448                                     jstring java_se_info, jstring java_se_name,
449                                     bool is_system_server, jintArray fdsToClose,
450                                     jstring instructionSet, jstring dataDir) {
451  SetSigChldHandler();
452
453  // Close any logging related FDs before we start evaluating the list of
454  // file descriptors.
455  __android_log_close();
456
457  // If this is the first fork for this zygote, create the open FD table.
458  // If it isn't, we just need to check whether the list of open files has
459  // changed (and it shouldn't in the normal case).
460  if (gOpenFdTable == NULL) {
461    gOpenFdTable = FileDescriptorTable::Create();
462    if (gOpenFdTable == NULL) {
463      RuntimeAbort(env, __LINE__, "Unable to construct file descriptor table.");
464    }
465  } else if (!gOpenFdTable->Restat()) {
466    RuntimeAbort(env, __LINE__, "Unable to restat file descriptor table.");
467  }
468
469  ResetNicePriority(env);
470
471  pid_t pid = fork();
472
473  if (pid == 0) {
474    // The child process.
475    gMallocLeakZygoteChild = 1;
476
477    // Clean up any descriptors which must be closed immediately
478    DetachDescriptors(env, fdsToClose);
479
480    // Re-open all remaining open file descriptors so that they aren't shared
481    // with the zygote across a fork.
482    if (!gOpenFdTable->ReopenOrDetach()) {
483      RuntimeAbort(env, __LINE__, "Unable to reopen whitelisted descriptors.");
484    }
485
486    // Keep capabilities across UID change, unless we're staying root.
487    if (uid != 0) {
488      EnableKeepCapabilities(env);
489    }
490
491    DropCapabilitiesBoundingSet(env);
492
493    bool use_native_bridge = !is_system_server && (instructionSet != NULL)
494        && android::NativeBridgeAvailable();
495    if (use_native_bridge) {
496      ScopedUtfChars isa_string(env, instructionSet);
497      use_native_bridge = android::NeedsNativeBridge(isa_string.c_str());
498    }
499    if (use_native_bridge && dataDir == NULL) {
500      // dataDir should never be null if we need to use a native bridge.
501      // In general, dataDir will never be null for normal applications. It can only happen in
502      // special cases (for isolated processes which are not associated with any app). These are
503      // launched by the framework and should not be emulated anyway.
504      use_native_bridge = false;
505      ALOGW("Native bridge will not be used because dataDir == NULL.");
506    }
507
508    if (!MountEmulatedStorage(uid, mount_external, use_native_bridge)) {
509      ALOGW("Failed to mount emulated storage: %s", strerror(errno));
510      if (errno == ENOTCONN || errno == EROFS) {
511        // When device is actively encrypting, we get ENOTCONN here
512        // since FUSE was mounted before the framework restarted.
513        // When encrypted device is booting, we get EROFS since
514        // FUSE hasn't been created yet by init.
515        // In either case, continue without external storage.
516      } else {
517        RuntimeAbort(env, __LINE__, "Cannot continue without emulated storage");
518      }
519    }
520
521    if (!is_system_server) {
522        int rc = createProcessGroup(uid, getpid());
523        if (rc != 0) {
524            if (rc == -EROFS) {
525                ALOGW("createProcessGroup failed, kernel missing CONFIG_CGROUP_CPUACCT?");
526            } else {
527                ALOGE("createProcessGroup(%d, %d) failed: %s", uid, pid, strerror(-rc));
528            }
529        }
530    }
531
532    SetGids(env, javaGids);
533
534    SetRLimits(env, javaRlimits);
535
536    if (use_native_bridge) {
537      ScopedUtfChars isa_string(env, instructionSet);
538      ScopedUtfChars data_dir(env, dataDir);
539      android::PreInitializeNativeBridge(data_dir.c_str(), isa_string.c_str());
540    }
541
542    int rc = setresgid(gid, gid, gid);
543    if (rc == -1) {
544      ALOGE("setresgid(%d) failed: %s", gid, strerror(errno));
545      RuntimeAbort(env, __LINE__, "setresgid failed");
546    }
547
548    rc = setresuid(uid, uid, uid);
549    if (rc == -1) {
550      ALOGE("setresuid(%d) failed: %s", uid, strerror(errno));
551      RuntimeAbort(env, __LINE__, "setresuid failed");
552    }
553
554    if (NeedsNoRandomizeWorkaround()) {
555        // Work around ARM kernel ASLR lossage (http://b/5817320).
556        int old_personality = personality(0xffffffff);
557        int new_personality = personality(old_personality | ADDR_NO_RANDOMIZE);
558        if (new_personality == -1) {
559            ALOGW("personality(%d) failed: %s", new_personality, strerror(errno));
560        }
561    }
562
563    SetCapabilities(env, permittedCapabilities, effectiveCapabilities);
564
565    SetSchedulerPolicy(env);
566
567    const char* se_info_c_str = NULL;
568    ScopedUtfChars* se_info = NULL;
569    if (java_se_info != NULL) {
570        se_info = new ScopedUtfChars(env, java_se_info);
571        se_info_c_str = se_info->c_str();
572        if (se_info_c_str == NULL) {
573          RuntimeAbort(env, __LINE__, "se_info_c_str == NULL");
574        }
575    }
576    const char* se_name_c_str = NULL;
577    ScopedUtfChars* se_name = NULL;
578    if (java_se_name != NULL) {
579        se_name = new ScopedUtfChars(env, java_se_name);
580        se_name_c_str = se_name->c_str();
581        if (se_name_c_str == NULL) {
582          RuntimeAbort(env, __LINE__, "se_name_c_str == NULL");
583        }
584    }
585    rc = selinux_android_setcontext(uid, is_system_server, se_info_c_str, se_name_c_str);
586    if (rc == -1) {
587      ALOGE("selinux_android_setcontext(%d, %d, \"%s\", \"%s\") failed", uid,
588            is_system_server, se_info_c_str, se_name_c_str);
589      RuntimeAbort(env, __LINE__, "selinux_android_setcontext failed");
590    }
591
592    // Make it easier to debug audit logs by setting the main thread's name to the
593    // nice name rather than "app_process".
594    if (se_info_c_str == NULL && is_system_server) {
595      se_name_c_str = "system_server";
596    }
597    if (se_info_c_str != NULL) {
598      SetThreadName(se_name_c_str);
599    }
600
601    delete se_info;
602    delete se_name;
603
604    UnsetSigChldHandler();
605
606    env->CallStaticVoidMethod(gZygoteClass, gCallPostForkChildHooks, debug_flags,
607                              is_system_server, instructionSet);
608    if (env->ExceptionCheck()) {
609      RuntimeAbort(env, __LINE__, "Error calling post fork hooks.");
610    }
611  } else if (pid > 0) {
612    // the parent process
613  }
614  return pid;
615}
616}  // anonymous namespace
617
618namespace android {
619
620static jint com_android_internal_os_Zygote_nativeForkAndSpecialize(
621        JNIEnv* env, jclass, jint uid, jint gid, jintArray gids,
622        jint debug_flags, jobjectArray rlimits,
623        jint mount_external, jstring se_info, jstring se_name,
624        jintArray fdsToClose, jstring instructionSet, jstring appDataDir) {
625    jlong capabilities = 0;
626
627    // Grant CAP_WAKE_ALARM to the Bluetooth process.
628    // Additionally, allow bluetooth to open packet sockets so it can start the DHCP client.
629    // TODO: consider making such functionality an RPC to netd.
630    if (multiuser_get_app_id(uid) == AID_BLUETOOTH) {
631      capabilities |= (1LL << CAP_WAKE_ALARM);
632      capabilities |= (1LL << CAP_NET_RAW);
633      capabilities |= (1LL << CAP_NET_BIND_SERVICE);
634    }
635
636    // Grant CAP_BLOCK_SUSPEND to processes that belong to GID "wakelock"
637    bool gid_wakelock_found = false;
638    if (gid == AID_WAKELOCK) {
639      gid_wakelock_found = true;
640    } else if (gids != NULL) {
641      jsize gids_num = env->GetArrayLength(gids);
642      ScopedIntArrayRO ar(env, gids);
643      if (ar.get() == NULL) {
644        RuntimeAbort(env, __LINE__, "Bad gids array");
645      }
646      for (int i = 0; i < gids_num; i++) {
647        if (ar[i] == AID_WAKELOCK) {
648          gid_wakelock_found = true;
649          break;
650        }
651      }
652    }
653    if (gid_wakelock_found) {
654      capabilities |= (1LL << CAP_BLOCK_SUSPEND);
655    }
656
657    return ForkAndSpecializeCommon(env, uid, gid, gids, debug_flags,
658            rlimits, capabilities, capabilities, mount_external, se_info,
659            se_name, false, fdsToClose, instructionSet, appDataDir);
660}
661
662static jint com_android_internal_os_Zygote_nativeForkSystemServer(
663        JNIEnv* env, jclass, uid_t uid, gid_t gid, jintArray gids,
664        jint debug_flags, jobjectArray rlimits, jlong permittedCapabilities,
665        jlong effectiveCapabilities) {
666  pid_t pid = ForkAndSpecializeCommon(env, uid, gid, gids,
667                                      debug_flags, rlimits,
668                                      permittedCapabilities, effectiveCapabilities,
669                                      MOUNT_EXTERNAL_DEFAULT, NULL, NULL, true, NULL,
670                                      NULL, NULL);
671  if (pid > 0) {
672      // The zygote process checks whether the child process has died or not.
673      ALOGI("System server process %d has been created", pid);
674      gSystemServerPid = pid;
675      // There is a slight window that the system server process has crashed
676      // but it went unnoticed because we haven't published its pid yet. So
677      // we recheck here just to make sure that all is well.
678      int status;
679      if (waitpid(pid, &status, WNOHANG) == pid) {
680          ALOGE("System server process %d has died. Restarting Zygote!", pid);
681          RuntimeAbort(env, __LINE__, "System server process has died. Restarting Zygote!");
682      }
683  }
684  return pid;
685}
686
687static void com_android_internal_os_Zygote_nativeAllowFileAcrossFork(
688        JNIEnv* env, jclass, jstring path) {
689    ScopedUtfChars path_native(env, path);
690    const char* path_cstr = path_native.c_str();
691    if (!path_cstr) {
692        RuntimeAbort(env, __LINE__, "path_cstr == NULL");
693    }
694    FileDescriptorWhitelist::Get()->Allow(path_cstr);
695}
696
697static void com_android_internal_os_Zygote_nativeUnmountStorageOnInit(JNIEnv* env, jclass) {
698    // Zygote process unmount root storage space initially before every child processes are forked.
699    // Every forked child processes (include SystemServer) only mount their own root storage space
700    // and no need unmount storage operation in MountEmulatedStorage method.
701    // Zygote process does not utilize root storage spaces and unshares its mount namespace below.
702
703    // See storage config details at http://source.android.com/tech/storage/
704    // Create private mount namespace shared by all children
705    if (unshare(CLONE_NEWNS) == -1) {
706        RuntimeAbort(env, __LINE__, "Failed to unshare()");
707        return;
708    }
709
710    // Mark rootfs as being a slave so that changes from default
711    // namespace only flow into our children.
712    if (mount("rootfs", "/", nullptr, (MS_SLAVE | MS_REC), nullptr) == -1) {
713        RuntimeAbort(env, __LINE__, "Failed to mount() rootfs as MS_SLAVE");
714        return;
715    }
716
717    // Create a staging tmpfs that is shared by our children; they will
718    // bind mount storage into their respective private namespaces, which
719    // are isolated from each other.
720    const char* target_base = getenv("EMULATED_STORAGE_TARGET");
721    if (target_base != nullptr) {
722#define STRINGIFY_UID(x) __STRING(x)
723        if (mount("tmpfs", target_base, "tmpfs", MS_NOSUID | MS_NODEV,
724                  "uid=0,gid=" STRINGIFY_UID(AID_SDCARD_R) ",mode=0751") == -1) {
725            ALOGE("Failed to mount tmpfs to %s", target_base);
726            RuntimeAbort(env, __LINE__, "Failed to mount tmpfs");
727            return;
728        }
729#undef STRINGIFY_UID
730    }
731
732    UnmountTree("/storage");
733}
734
735static const JNINativeMethod gMethods[] = {
736    { "nativeForkAndSpecialize",
737      "(II[II[[IILjava/lang/String;Ljava/lang/String;[ILjava/lang/String;Ljava/lang/String;)I",
738      (void *) com_android_internal_os_Zygote_nativeForkAndSpecialize },
739    { "nativeForkSystemServer", "(II[II[[IJJ)I",
740      (void *) com_android_internal_os_Zygote_nativeForkSystemServer },
741    { "nativeAllowFileAcrossFork", "(Ljava/lang/String;)V",
742      (void *) com_android_internal_os_Zygote_nativeAllowFileAcrossFork },
743    { "nativeUnmountStorageOnInit", "()V",
744      (void *) com_android_internal_os_Zygote_nativeUnmountStorageOnInit }
745};
746
747int register_com_android_internal_os_Zygote(JNIEnv* env) {
748  gZygoteClass = MakeGlobalRefOrDie(env, FindClassOrDie(env, kZygoteClassName));
749  gCallPostForkChildHooks = GetStaticMethodIDOrDie(env, gZygoteClass, "callPostForkChildHooks",
750                                                   "(IZLjava/lang/String;)V");
751
752  return RegisterMethodsOrDie(env, "com/android/internal/os/Zygote", gMethods, NELEM(gMethods));
753}
754}  // namespace android
755