com_android_internal_os_Zygote.cpp revision 8dfa178efbbb155657639bd526e9a8579fce3886
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 "android-base/logging.h"
47#include <cutils/fs.h>
48#include <cutils/multiuser.h>
49#include <cutils/sched_policy.h>
50#include <private/android_filesystem_config.h>
51#include <utils/String8.h>
52#include <selinux/android.h>
53#include <processgroup/processgroup.h>
54
55#include "core_jni_helpers.h"
56#include "JNIHelp.h"
57#include "ScopedLocalRef.h"
58#include "ScopedPrimitiveArray.h"
59#include "ScopedUtfChars.h"
60#include "fd_utils.h"
61
62#include "nativebridge/native_bridge.h"
63
64namespace {
65
66using android::String8;
67
68static pid_t gSystemServerPid = 0;
69
70static const char kZygoteClassName[] = "com/android/internal/os/Zygote";
71static jclass gZygoteClass;
72static jmethodID gCallPostForkChildHooks;
73
74// Must match values in com.android.internal.os.Zygote.
75enum MountExternalKind {
76  MOUNT_EXTERNAL_NONE = 0,
77  MOUNT_EXTERNAL_DEFAULT = 1,
78  MOUNT_EXTERNAL_READ = 2,
79  MOUNT_EXTERNAL_WRITE = 3,
80};
81
82static void RuntimeAbort(JNIEnv* env, int line, const char* msg) {
83  std::ostringstream oss;
84  oss << __FILE__ << ":" << line << ": " << msg;
85  env->FatalError(oss.str().c_str());
86}
87
88// This signal handler is for zygote mode, since the zygote must reap its children
89static void SigChldHandler(int /*signal_number*/) {
90  pid_t pid;
91  int status;
92
93  // It's necessary to save and restore the errno during this function.
94  // Since errno is stored per thread, changing it here modifies the errno
95  // on the thread on which this signal handler executes. If a signal occurs
96  // between a call and an errno check, it's possible to get the errno set
97  // here.
98  // See b/23572286 for extra information.
99  int saved_errno = errno;
100
101  while ((pid = waitpid(-1, &status, WNOHANG)) > 0) {
102     // Log process-death status that we care about.  In general it is
103     // not safe to call LOG(...) from a signal handler because of
104     // possible reentrancy.  However, we know a priori that the
105     // current implementation of LOG() is safe to call from a SIGCHLD
106     // handler in the zygote process.  If the LOG() implementation
107     // changes its locking strategy or its use of syscalls within the
108     // lazy-init critical section, its use here may become unsafe.
109    if (WIFEXITED(status)) {
110      if (WEXITSTATUS(status)) {
111        ALOGI("Process %d exited cleanly (%d)", pid, WEXITSTATUS(status));
112      }
113    } else if (WIFSIGNALED(status)) {
114      if (WTERMSIG(status) != SIGKILL) {
115        ALOGI("Process %d exited due to signal (%d)", pid, WTERMSIG(status));
116      }
117      if (WCOREDUMP(status)) {
118        ALOGI("Process %d dumped core.", pid);
119      }
120    }
121
122    // If the just-crashed process is the system_server, bring down zygote
123    // so that it is restarted by init and system server will be restarted
124    // from there.
125    if (pid == gSystemServerPid) {
126      ALOGE("Exit zygote because system server (%d) has terminated", pid);
127      kill(getpid(), SIGKILL);
128    }
129  }
130
131  // Note that we shouldn't consider ECHILD an error because
132  // the secondary zygote might have no children left to wait for.
133  if (pid < 0 && errno != ECHILD) {
134    ALOGW("Zygote SIGCHLD error in waitpid: %s", strerror(errno));
135  }
136
137  errno = saved_errno;
138}
139
140// Configures the SIGCHLD handler for the zygote process. This is configured
141// very late, because earlier in the runtime we may fork() and exec()
142// other processes, and we want to waitpid() for those rather than
143// have them be harvested immediately.
144//
145// This ends up being called repeatedly before each fork(), but there's
146// no real harm in that.
147static void SetSigChldHandler() {
148  struct sigaction sa;
149  memset(&sa, 0, sizeof(sa));
150  sa.sa_handler = SigChldHandler;
151
152  int err = sigaction(SIGCHLD, &sa, NULL);
153  if (err < 0) {
154    ALOGW("Error setting SIGCHLD handler: %s", strerror(errno));
155  }
156}
157
158// Resets nice priority for zygote process. Zygote priority can be set
159// to high value during boot phase to speed it up. We want to ensure
160// zygote is running at normal priority before childs are forked from it.
161//
162// This ends up being called repeatedly before each fork(), but there's
163// no real harm in that.
164static void ResetNicePriority(JNIEnv* env) {
165  errno = 0;
166  int prio = getpriority(PRIO_PROCESS, 0);
167  if (prio == -1 && errno != 0) {
168    ALOGW("getpriority failed: %s\n", strerror(errno));
169  }
170  if (prio != 0 && setpriority(PRIO_PROCESS, 0, 0) != 0) {
171    ALOGE("setpriority(%d, 0, 0) failed: %s", PRIO_PROCESS, strerror(errno));
172    RuntimeAbort(env, __LINE__, "setpriority failed");
173  }
174}
175
176// Sets the SIGCHLD handler back to default behavior in zygote children.
177static void UnsetSigChldHandler() {
178  struct sigaction sa;
179  memset(&sa, 0, sizeof(sa));
180  sa.sa_handler = SIG_DFL;
181
182  int err = sigaction(SIGCHLD, &sa, NULL);
183  if (err < 0) {
184    ALOGW("Error unsetting SIGCHLD handler: %s", strerror(errno));
185  }
186}
187
188// Calls POSIX setgroups() using the int[] object as an argument.
189// A NULL argument is tolerated.
190static void SetGids(JNIEnv* env, jintArray javaGids) {
191  if (javaGids == NULL) {
192    return;
193  }
194
195  ScopedIntArrayRO gids(env, javaGids);
196  if (gids.get() == NULL) {
197    RuntimeAbort(env, __LINE__, "Getting gids int array failed");
198  }
199  int rc = setgroups(gids.size(), reinterpret_cast<const gid_t*>(&gids[0]));
200  if (rc == -1) {
201    std::ostringstream oss;
202    oss << "setgroups failed: " << strerror(errno) << ", gids.size=" << gids.size();
203    RuntimeAbort(env, __LINE__, oss.str().c_str());
204  }
205}
206
207// Sets the resource limits via setrlimit(2) for the values in the
208// two-dimensional array of integers that's passed in. The second dimension
209// contains a tuple of length 3: (resource, rlim_cur, rlim_max). NULL is
210// treated as an empty array.
211static void SetRLimits(JNIEnv* env, jobjectArray javaRlimits) {
212  if (javaRlimits == NULL) {
213    return;
214  }
215
216  rlimit rlim;
217  memset(&rlim, 0, sizeof(rlim));
218
219  for (int i = 0; i < env->GetArrayLength(javaRlimits); ++i) {
220    ScopedLocalRef<jobject> javaRlimitObject(env, env->GetObjectArrayElement(javaRlimits, i));
221    ScopedIntArrayRO javaRlimit(env, reinterpret_cast<jintArray>(javaRlimitObject.get()));
222    if (javaRlimit.size() != 3) {
223      RuntimeAbort(env, __LINE__, "rlimits array must have a second dimension of size 3");
224    }
225
226    rlim.rlim_cur = javaRlimit[1];
227    rlim.rlim_max = javaRlimit[2];
228
229    int rc = setrlimit(javaRlimit[0], &rlim);
230    if (rc == -1) {
231      ALOGE("setrlimit(%d, {%ld, %ld}) failed", javaRlimit[0], rlim.rlim_cur,
232            rlim.rlim_max);
233      RuntimeAbort(env, __LINE__, "setrlimit failed");
234    }
235  }
236}
237
238// The debug malloc library needs to know whether it's the zygote or a child.
239extern "C" int gMallocLeakZygoteChild;
240
241static void EnableKeepCapabilities(JNIEnv* env) {
242  int rc = prctl(PR_SET_KEEPCAPS, 1, 0, 0, 0);
243  if (rc == -1) {
244    RuntimeAbort(env, __LINE__, "prctl(PR_SET_KEEPCAPS) failed");
245  }
246}
247
248static void DropCapabilitiesBoundingSet(JNIEnv* env) {
249  for (int i = 0; prctl(PR_CAPBSET_READ, i, 0, 0, 0) >= 0; i++) {
250    int rc = prctl(PR_CAPBSET_DROP, i, 0, 0, 0);
251    if (rc == -1) {
252      if (errno == EINVAL) {
253        ALOGE("prctl(PR_CAPBSET_DROP) failed with EINVAL. Please verify "
254              "your kernel is compiled with file capabilities support");
255      } else {
256        RuntimeAbort(env, __LINE__, "prctl(PR_CAPBSET_DROP) failed");
257      }
258    }
259  }
260}
261
262static void SetCapabilities(JNIEnv* env, int64_t permitted, int64_t effective) {
263  __user_cap_header_struct capheader;
264  memset(&capheader, 0, sizeof(capheader));
265  capheader.version = _LINUX_CAPABILITY_VERSION_3;
266  capheader.pid = 0;
267
268  __user_cap_data_struct capdata[2];
269  memset(&capdata, 0, sizeof(capdata));
270  capdata[0].effective = effective;
271  capdata[1].effective = effective >> 32;
272  capdata[0].permitted = permitted;
273  capdata[1].permitted = permitted >> 32;
274
275  if (capset(&capheader, &capdata[0]) == -1) {
276    ALOGE("capset(%" PRId64 ", %" PRId64 ") failed", permitted, effective);
277    RuntimeAbort(env, __LINE__, "capset failed");
278  }
279}
280
281static void SetSchedulerPolicy(JNIEnv* env) {
282  errno = -set_sched_policy(0, SP_DEFAULT);
283  if (errno != 0) {
284    ALOGE("set_sched_policy(0, SP_DEFAULT) failed");
285    RuntimeAbort(env, __LINE__, "set_sched_policy(0, SP_DEFAULT) failed");
286  }
287}
288
289static int UnmountTree(const char* path) {
290    size_t path_len = strlen(path);
291
292    FILE* fp = setmntent("/proc/mounts", "r");
293    if (fp == NULL) {
294        ALOGE("Error opening /proc/mounts: %s", strerror(errno));
295        return -errno;
296    }
297
298    // Some volumes can be stacked on each other, so force unmount in
299    // reverse order to give us the best chance of success.
300    std::list<std::string> toUnmount;
301    mntent* mentry;
302    while ((mentry = getmntent(fp)) != NULL) {
303        if (strncmp(mentry->mnt_dir, path, path_len) == 0) {
304            toUnmount.push_front(std::string(mentry->mnt_dir));
305        }
306    }
307    endmntent(fp);
308
309    for (auto path : toUnmount) {
310        if (umount2(path.c_str(), MNT_DETACH)) {
311            ALOGW("Failed to unmount %s: %s", path.c_str(), strerror(errno));
312        }
313    }
314    return 0;
315}
316
317// Create a private mount namespace and bind mount appropriate emulated
318// storage for the given user.
319static bool MountEmulatedStorage(uid_t uid, jint mount_mode,
320        bool force_mount_namespace) {
321    // See storage config details at http://source.android.com/tech/storage/
322
323    String8 storageSource;
324    if (mount_mode == MOUNT_EXTERNAL_DEFAULT) {
325        storageSource = "/mnt/runtime/default";
326    } else if (mount_mode == MOUNT_EXTERNAL_READ) {
327        storageSource = "/mnt/runtime/read";
328    } else if (mount_mode == MOUNT_EXTERNAL_WRITE) {
329        storageSource = "/mnt/runtime/write";
330    } else if (!force_mount_namespace) {
331        // Sane default of no storage visible
332        return true;
333    }
334
335    // Create a second private mount namespace for our process
336    if (unshare(CLONE_NEWNS) == -1) {
337        ALOGW("Failed to unshare(): %s", strerror(errno));
338        return false;
339    }
340
341    if (TEMP_FAILURE_RETRY(mount(storageSource.string(), "/storage",
342            NULL, MS_BIND | MS_REC | MS_SLAVE, NULL)) == -1) {
343        ALOGW("Failed to mount %s to /storage: %s", storageSource.string(), strerror(errno));
344        return false;
345    }
346
347    // Mount user-specific symlink helper into place
348    userid_t user_id = multiuser_get_user_id(uid);
349    const String8 userSource(String8::format("/mnt/user/%d", user_id));
350    if (fs_prepare_dir(userSource.string(), 0751, 0, 0) == -1) {
351        return false;
352    }
353    if (TEMP_FAILURE_RETRY(mount(userSource.string(), "/storage/self",
354            NULL, MS_BIND, NULL)) == -1) {
355        ALOGW("Failed to mount %s to /storage/self: %s", userSource.string(), strerror(errno));
356        return false;
357    }
358
359    return true;
360}
361
362static bool NeedsNoRandomizeWorkaround() {
363#if !defined(__arm__)
364    return false;
365#else
366    int major;
367    int minor;
368    struct utsname uts;
369    if (uname(&uts) == -1) {
370        return false;
371    }
372
373    if (sscanf(uts.release, "%d.%d", &major, &minor) != 2) {
374        return false;
375    }
376
377    // Kernels before 3.4.* need the workaround.
378    return (major < 3) || ((major == 3) && (minor < 4));
379#endif
380}
381
382// Utility to close down the Zygote socket file descriptors while
383// the child is still running as root with Zygote's privileges.  Each
384// descriptor (if any) is closed via dup2(), replacing it with a valid
385// (open) descriptor to /dev/null.
386
387static void DetachDescriptors(JNIEnv* env, jintArray fdsToClose) {
388  if (!fdsToClose) {
389    return;
390  }
391  jsize count = env->GetArrayLength(fdsToClose);
392  ScopedIntArrayRO ar(env, fdsToClose);
393  if (ar.get() == NULL) {
394      RuntimeAbort(env, __LINE__, "Bad fd array");
395  }
396  jsize i;
397  int devnull;
398  for (i = 0; i < count; i++) {
399    devnull = open("/dev/null", O_RDWR);
400    if (devnull < 0) {
401      ALOGE("Failed to open /dev/null: %s", strerror(errno));
402      RuntimeAbort(env, __LINE__, "Failed to open /dev/null");
403      continue;
404    }
405    ALOGV("Switching descriptor %d to /dev/null: %s", ar[i], strerror(errno));
406    if (dup2(devnull, ar[i]) < 0) {
407      ALOGE("Failed dup2() on descriptor %d: %s", ar[i], strerror(errno));
408      RuntimeAbort(env, __LINE__, "Failed dup2()");
409    }
410    close(devnull);
411  }
412}
413
414void SetThreadName(const char* thread_name) {
415  bool hasAt = false;
416  bool hasDot = false;
417  const char* s = thread_name;
418  while (*s) {
419    if (*s == '.') {
420      hasDot = true;
421    } else if (*s == '@') {
422      hasAt = true;
423    }
424    s++;
425  }
426  const int len = s - thread_name;
427  if (len < 15 || hasAt || !hasDot) {
428    s = thread_name;
429  } else {
430    s = thread_name + len - 15;
431  }
432  // pthread_setname_np fails rather than truncating long strings.
433  char buf[16];       // MAX_TASK_COMM_LEN=16 is hard-coded into bionic
434  strlcpy(buf, s, sizeof(buf)-1);
435  errno = pthread_setname_np(pthread_self(), buf);
436  if (errno != 0) {
437    ALOGW("Unable to set the name of current thread to '%s': %s", buf, strerror(errno));
438  }
439}
440
441// The list of open zygote file descriptors.
442static FileDescriptorTable* gOpenFdTable = NULL;
443
444static void FillFileDescriptorVector(JNIEnv* env,
445                                     jintArray java_fds,
446                                     std::vector<int>* fds) {
447  CHECK(fds != nullptr);
448  if (java_fds != nullptr) {
449    ScopedIntArrayRO ar(env, java_fds);
450    if (ar.get() == nullptr) {
451      RuntimeAbort(env, __LINE__, "Bad fd array");
452    }
453    fds->reserve(ar.size());
454    for (size_t i = 0; i < ar.size(); ++i) {
455      fds->push_back(ar[i]);
456    }
457  }
458}
459
460// Utility routine to fork zygote and specialize the child process.
461static pid_t ForkAndSpecializeCommon(JNIEnv* env, uid_t uid, gid_t gid, jintArray javaGids,
462                                     jint debug_flags, jobjectArray javaRlimits,
463                                     jlong permittedCapabilities, jlong effectiveCapabilities,
464                                     jint mount_external,
465                                     jstring java_se_info, jstring java_se_name,
466                                     bool is_system_server, jintArray fdsToClose,
467                                     jintArray fdsToIgnore,
468                                     jstring instructionSet, jstring dataDir) {
469  SetSigChldHandler();
470
471  // Close any logging related FDs before we start evaluating the list of
472  // file descriptors.
473  __android_log_close();
474
475  // If this is the first fork for this zygote, create the open FD table.
476  // If it isn't, we just need to check whether the list of open files has
477  // changed (and it shouldn't in the normal case).
478  std::vector<int> fds_to_ignore;
479  FillFileDescriptorVector(env, fdsToIgnore, &fds_to_ignore);
480  if (gOpenFdTable == NULL) {
481    gOpenFdTable = FileDescriptorTable::Create(fds_to_ignore);
482    if (gOpenFdTable == NULL) {
483      RuntimeAbort(env, __LINE__, "Unable to construct file descriptor table.");
484    }
485  } else if (!gOpenFdTable->Restat(fds_to_ignore)) {
486    RuntimeAbort(env, __LINE__, "Unable to restat file descriptor table.");
487  }
488
489  ResetNicePriority(env);
490
491  pid_t pid = fork();
492
493  if (pid == 0) {
494    // The child process.
495    gMallocLeakZygoteChild = 1;
496
497    // Clean up any descriptors which must be closed immediately
498    DetachDescriptors(env, fdsToClose);
499
500    // Re-open all remaining open file descriptors so that they aren't shared
501    // with the zygote across a fork.
502    if (!gOpenFdTable->ReopenOrDetach()) {
503      RuntimeAbort(env, __LINE__, "Unable to reopen whitelisted descriptors.");
504    }
505
506    // Keep capabilities across UID change, unless we're staying root.
507    if (uid != 0) {
508      EnableKeepCapabilities(env);
509    }
510
511    DropCapabilitiesBoundingSet(env);
512
513    bool use_native_bridge = !is_system_server && (instructionSet != NULL)
514        && android::NativeBridgeAvailable();
515    if (use_native_bridge) {
516      ScopedUtfChars isa_string(env, instructionSet);
517      use_native_bridge = android::NeedsNativeBridge(isa_string.c_str());
518    }
519    if (use_native_bridge && dataDir == NULL) {
520      // dataDir should never be null if we need to use a native bridge.
521      // In general, dataDir will never be null for normal applications. It can only happen in
522      // special cases (for isolated processes which are not associated with any app). These are
523      // launched by the framework and should not be emulated anyway.
524      use_native_bridge = false;
525      ALOGW("Native bridge will not be used because dataDir == NULL.");
526    }
527
528    if (!MountEmulatedStorage(uid, mount_external, use_native_bridge)) {
529      ALOGW("Failed to mount emulated storage: %s", strerror(errno));
530      if (errno == ENOTCONN || errno == EROFS) {
531        // When device is actively encrypting, we get ENOTCONN here
532        // since FUSE was mounted before the framework restarted.
533        // When encrypted device is booting, we get EROFS since
534        // FUSE hasn't been created yet by init.
535        // In either case, continue without external storage.
536      } else {
537        RuntimeAbort(env, __LINE__, "Cannot continue without emulated storage");
538      }
539    }
540
541    if (!is_system_server) {
542        int rc = createProcessGroup(uid, getpid());
543        if (rc != 0) {
544            if (rc == -EROFS) {
545                ALOGW("createProcessGroup failed, kernel missing CONFIG_CGROUP_CPUACCT?");
546            } else {
547                ALOGE("createProcessGroup(%d, %d) failed: %s", uid, pid, strerror(-rc));
548            }
549        }
550    }
551
552    SetGids(env, javaGids);
553
554    SetRLimits(env, javaRlimits);
555
556    if (use_native_bridge) {
557      ScopedUtfChars isa_string(env, instructionSet);
558      ScopedUtfChars data_dir(env, dataDir);
559      android::PreInitializeNativeBridge(data_dir.c_str(), isa_string.c_str());
560    }
561
562    int rc = setresgid(gid, gid, gid);
563    if (rc == -1) {
564      ALOGE("setresgid(%d) failed: %s", gid, strerror(errno));
565      RuntimeAbort(env, __LINE__, "setresgid failed");
566    }
567
568    rc = setresuid(uid, uid, uid);
569    if (rc == -1) {
570      ALOGE("setresuid(%d) failed: %s", uid, strerror(errno));
571      RuntimeAbort(env, __LINE__, "setresuid failed");
572    }
573
574    if (NeedsNoRandomizeWorkaround()) {
575        // Work around ARM kernel ASLR lossage (http://b/5817320).
576        int old_personality = personality(0xffffffff);
577        int new_personality = personality(old_personality | ADDR_NO_RANDOMIZE);
578        if (new_personality == -1) {
579            ALOGW("personality(%d) failed: %s", new_personality, strerror(errno));
580        }
581    }
582
583    SetCapabilities(env, permittedCapabilities, effectiveCapabilities);
584
585    SetSchedulerPolicy(env);
586
587    const char* se_info_c_str = NULL;
588    ScopedUtfChars* se_info = NULL;
589    if (java_se_info != NULL) {
590        se_info = new ScopedUtfChars(env, java_se_info);
591        se_info_c_str = se_info->c_str();
592        if (se_info_c_str == NULL) {
593          RuntimeAbort(env, __LINE__, "se_info_c_str == NULL");
594        }
595    }
596    const char* se_name_c_str = NULL;
597    ScopedUtfChars* se_name = NULL;
598    if (java_se_name != NULL) {
599        se_name = new ScopedUtfChars(env, java_se_name);
600        se_name_c_str = se_name->c_str();
601        if (se_name_c_str == NULL) {
602          RuntimeAbort(env, __LINE__, "se_name_c_str == NULL");
603        }
604    }
605    rc = selinux_android_setcontext(uid, is_system_server, se_info_c_str, se_name_c_str);
606    if (rc == -1) {
607      ALOGE("selinux_android_setcontext(%d, %d, \"%s\", \"%s\") failed", uid,
608            is_system_server, se_info_c_str, se_name_c_str);
609      RuntimeAbort(env, __LINE__, "selinux_android_setcontext failed");
610    }
611
612    // Make it easier to debug audit logs by setting the main thread's name to the
613    // nice name rather than "app_process".
614    if (se_info_c_str == NULL && is_system_server) {
615      se_name_c_str = "system_server";
616    }
617    if (se_info_c_str != NULL) {
618      SetThreadName(se_name_c_str);
619    }
620
621    delete se_info;
622    delete se_name;
623
624    UnsetSigChldHandler();
625
626    env->CallStaticVoidMethod(gZygoteClass, gCallPostForkChildHooks, debug_flags,
627                              is_system_server, instructionSet);
628    if (env->ExceptionCheck()) {
629      RuntimeAbort(env, __LINE__, "Error calling post fork hooks.");
630    }
631  } else if (pid > 0) {
632    // the parent process
633  }
634  return pid;
635}
636}  // anonymous namespace
637
638namespace android {
639
640static jint com_android_internal_os_Zygote_nativeForkAndSpecialize(
641        JNIEnv* env, jclass, jint uid, jint gid, jintArray gids,
642        jint debug_flags, jobjectArray rlimits,
643        jint mount_external, jstring se_info, jstring se_name,
644        jintArray fdsToClose,
645        jintArray fdsToIgnore,
646        jstring instructionSet, jstring appDataDir) {
647    jlong capabilities = 0;
648
649    // Grant CAP_WAKE_ALARM to the Bluetooth process.
650    // Additionally, allow bluetooth to open packet sockets so it can start the DHCP client.
651    // TODO: consider making such functionality an RPC to netd.
652    if (multiuser_get_app_id(uid) == AID_BLUETOOTH) {
653      capabilities |= (1LL << CAP_WAKE_ALARM);
654      capabilities |= (1LL << CAP_NET_RAW);
655      capabilities |= (1LL << CAP_NET_BIND_SERVICE);
656    }
657
658    // Grant CAP_BLOCK_SUSPEND to processes that belong to GID "wakelock"
659    bool gid_wakelock_found = false;
660    if (gid == AID_WAKELOCK) {
661      gid_wakelock_found = true;
662    } else if (gids != NULL) {
663      jsize gids_num = env->GetArrayLength(gids);
664      ScopedIntArrayRO ar(env, gids);
665      if (ar.get() == NULL) {
666        RuntimeAbort(env, __LINE__, "Bad gids array");
667      }
668      for (int i = 0; i < gids_num; i++) {
669        if (ar[i] == AID_WAKELOCK) {
670          gid_wakelock_found = true;
671          break;
672        }
673      }
674    }
675    if (gid_wakelock_found) {
676      capabilities |= (1LL << CAP_BLOCK_SUSPEND);
677    }
678
679    return ForkAndSpecializeCommon(env, uid, gid, gids, debug_flags,
680            rlimits, capabilities, capabilities, mount_external, se_info,
681            se_name, false, fdsToClose, fdsToIgnore, instructionSet, appDataDir);
682}
683
684static jint com_android_internal_os_Zygote_nativeForkSystemServer(
685        JNIEnv* env, jclass, uid_t uid, gid_t gid, jintArray gids,
686        jint debug_flags, jobjectArray rlimits, jlong permittedCapabilities,
687        jlong effectiveCapabilities) {
688  pid_t pid = ForkAndSpecializeCommon(env, uid, gid, gids,
689                                      debug_flags, rlimits,
690                                      permittedCapabilities, effectiveCapabilities,
691                                      MOUNT_EXTERNAL_DEFAULT, NULL, NULL, true, NULL,
692                                      NULL, NULL, NULL);
693  if (pid > 0) {
694      // The zygote process checks whether the child process has died or not.
695      ALOGI("System server process %d has been created", pid);
696      gSystemServerPid = pid;
697      // There is a slight window that the system server process has crashed
698      // but it went unnoticed because we haven't published its pid yet. So
699      // we recheck here just to make sure that all is well.
700      int status;
701      if (waitpid(pid, &status, WNOHANG) == pid) {
702          ALOGE("System server process %d has died. Restarting Zygote!", pid);
703          RuntimeAbort(env, __LINE__, "System server process has died. Restarting Zygote!");
704      }
705  }
706  return pid;
707}
708
709static void com_android_internal_os_Zygote_nativeAllowFileAcrossFork(
710        JNIEnv* env, jclass, jstring path) {
711    ScopedUtfChars path_native(env, path);
712    const char* path_cstr = path_native.c_str();
713    if (!path_cstr) {
714        RuntimeAbort(env, __LINE__, "path_cstr == NULL");
715    }
716    FileDescriptorWhitelist::Get()->Allow(path_cstr);
717}
718
719static void com_android_internal_os_Zygote_nativeUnmountStorageOnInit(JNIEnv* env, jclass) {
720    // Zygote process unmount root storage space initially before every child processes are forked.
721    // Every forked child processes (include SystemServer) only mount their own root storage space
722    // and no need unmount storage operation in MountEmulatedStorage method.
723    // Zygote process does not utilize root storage spaces and unshares its mount namespace below.
724
725    // See storage config details at http://source.android.com/tech/storage/
726    // Create private mount namespace shared by all children
727    if (unshare(CLONE_NEWNS) == -1) {
728        RuntimeAbort(env, __LINE__, "Failed to unshare()");
729        return;
730    }
731
732    // Mark rootfs as being a slave so that changes from default
733    // namespace only flow into our children.
734    if (mount("rootfs", "/", nullptr, (MS_SLAVE | MS_REC), nullptr) == -1) {
735        RuntimeAbort(env, __LINE__, "Failed to mount() rootfs as MS_SLAVE");
736        return;
737    }
738
739    // Create a staging tmpfs that is shared by our children; they will
740    // bind mount storage into their respective private namespaces, which
741    // are isolated from each other.
742    const char* target_base = getenv("EMULATED_STORAGE_TARGET");
743    if (target_base != nullptr) {
744#define STRINGIFY_UID(x) __STRING(x)
745        if (mount("tmpfs", target_base, "tmpfs", MS_NOSUID | MS_NODEV,
746                  "uid=0,gid=" STRINGIFY_UID(AID_SDCARD_R) ",mode=0751") == -1) {
747            ALOGE("Failed to mount tmpfs to %s", target_base);
748            RuntimeAbort(env, __LINE__, "Failed to mount tmpfs");
749            return;
750        }
751#undef STRINGIFY_UID
752    }
753
754    UnmountTree("/storage");
755}
756
757static const JNINativeMethod gMethods[] = {
758    { "nativeForkAndSpecialize",
759      "(II[II[[IILjava/lang/String;Ljava/lang/String;[I[ILjava/lang/String;Ljava/lang/String;)I",
760      (void *) com_android_internal_os_Zygote_nativeForkAndSpecialize },
761    { "nativeForkSystemServer", "(II[II[[IJJ)I",
762      (void *) com_android_internal_os_Zygote_nativeForkSystemServer },
763    { "nativeAllowFileAcrossFork", "(Ljava/lang/String;)V",
764      (void *) com_android_internal_os_Zygote_nativeAllowFileAcrossFork },
765    { "nativeUnmountStorageOnInit", "()V",
766      (void *) com_android_internal_os_Zygote_nativeUnmountStorageOnInit }
767};
768
769int register_com_android_internal_os_Zygote(JNIEnv* env) {
770  gZygoteClass = MakeGlobalRefOrDie(env, FindClassOrDie(env, kZygoteClassName));
771  gCallPostForkChildHooks = GetStaticMethodIDOrDie(env, gZygoteClass, "callPostForkChildHooks",
772                                                   "(IZLjava/lang/String;)V");
773
774  return RegisterMethodsOrDie(env, "com/android/internal/os/Zygote", gMethods, NELEM(gMethods));
775}
776}  // namespace android
777