utils.cpp revision 770410dcf2f72cf4e806442263e63719a0be0f5a 
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#include <dirent.h>
18#include <errno.h>
19#include <fcntl.h>
20#include <limits.h>
21#include <poll.h>
22#include <signal.h>
23#include <stdarg.h>
24#include <stdio.h>
25#include <stdlib.h>
26#include <string>
27#include <string.h>
28#include <sys/inotify.h>
29#include <sys/stat.h>
30#include <sys/time.h>
31#include <sys/wait.h>
32#include <sys/klog.h>
33#include <time.h>
34#include <unistd.h>
35#include <vector>
36#include <sys/prctl.h>
37
38#define LOG_TAG "dumpstate"
39#include <cutils/debugger.h>
40#include <cutils/log.h>
41#include <cutils/properties.h>
42#include <cutils/sockets.h>
43#include <private/android_filesystem_config.h>
44
45#include <selinux/android.h>
46
47#include "dumpstate.h"
48
49static const int64_t NANOS_PER_SEC = 1000000000;
50
51/* list of native processes to include in the native dumps */
52static const char* native_processes_to_dump[] = {
53        "/system/bin/audioserver",
54        "/system/bin/drmserver",
55        "/system/bin/mediaserver",
56        "/system/bin/sdcard",
57        "/system/bin/surfaceflinger",
58        "/system/bin/vehicle_network_service",
59        NULL,
60};
61
62DurationReporter::DurationReporter(const char *title) {
63    title_ = title;
64    if (title) {
65        started_ = DurationReporter::nanotime();
66    }
67}
68
69DurationReporter::~DurationReporter() {
70    if (title_) {
71        uint64_t elapsed = DurationReporter::nanotime() - started_;
72        // Use "Yoda grammar" to make it easier to grep|sort sections.
73        printf("------ %.3fs was the duration of '%s' ------\n",
74               (float) elapsed / NANOS_PER_SEC, title_);
75    }
76}
77
78uint64_t DurationReporter::DurationReporter::nanotime() {
79    struct timespec ts;
80    clock_gettime(CLOCK_MONOTONIC, &ts);
81    return (uint64_t) ts.tv_sec * NANOS_PER_SEC + ts.tv_nsec;
82}
83
84void for_each_userid(void (*func)(int), const char *header) {
85    ON_DRY_RUN_RETURN();
86    DIR *d;
87    struct dirent *de;
88
89    if (header) printf("\n------ %s ------\n", header);
90    func(0);
91
92    if (!(d = opendir("/data/system/users"))) {
93        printf("Failed to open /data/system/users (%s)\n", strerror(errno));
94        return;
95    }
96
97    while ((de = readdir(d))) {
98        int userid;
99        if (de->d_type != DT_DIR || !(userid = atoi(de->d_name))) {
100            continue;
101        }
102        func(userid);
103    }
104
105    closedir(d);
106}
107
108static void __for_each_pid(void (*helper)(int, const char *, void *), const char *header, void *arg) {
109    DIR *d;
110    struct dirent *de;
111
112    if (!(d = opendir("/proc"))) {
113        printf("Failed to open /proc (%s)\n", strerror(errno));
114        return;
115    }
116
117    if (header) printf("\n------ %s ------\n", header);
118    while ((de = readdir(d))) {
119        int pid;
120        int fd;
121        char cmdpath[255];
122        char cmdline[255];
123
124        if (!(pid = atoi(de->d_name))) {
125            continue;
126        }
127
128        sprintf(cmdpath,"/proc/%d/cmdline", pid);
129        memset(cmdline, 0, sizeof(cmdline));
130        if ((fd = TEMP_FAILURE_RETRY(open(cmdpath, O_RDONLY | O_CLOEXEC))) < 0) {
131            strcpy(cmdline, "N/A");
132        } else {
133            read(fd, cmdline, sizeof(cmdline) - 1);
134            close(fd);
135        }
136        helper(pid, cmdline, arg);
137    }
138
139    closedir(d);
140}
141
142static void for_each_pid_helper(int pid, const char *cmdline, void *arg) {
143    for_each_pid_func *func = (for_each_pid_func*) arg;
144    func(pid, cmdline);
145}
146
147void for_each_pid(for_each_pid_func func, const char *header) {
148    ON_DRY_RUN_RETURN();
149  __for_each_pid(for_each_pid_helper, header, (void *)func);
150}
151
152static void for_each_tid_helper(int pid, const char *cmdline, void *arg) {
153    DIR *d;
154    struct dirent *de;
155    char taskpath[255];
156    for_each_tid_func *func = (for_each_tid_func *) arg;
157
158    sprintf(taskpath, "/proc/%d/task", pid);
159
160    if (!(d = opendir(taskpath))) {
161        printf("Failed to open %s (%s)\n", taskpath, strerror(errno));
162        return;
163    }
164
165    func(pid, pid, cmdline);
166
167    while ((de = readdir(d))) {
168        int tid;
169        int fd;
170        char commpath[255];
171        char comm[255];
172
173        if (!(tid = atoi(de->d_name))) {
174            continue;
175        }
176
177        if (tid == pid)
178            continue;
179
180        sprintf(commpath,"/proc/%d/comm", tid);
181        memset(comm, 0, sizeof(comm));
182        if ((fd = TEMP_FAILURE_RETRY(open(commpath, O_RDONLY | O_CLOEXEC))) < 0) {
183            strcpy(comm, "N/A");
184        } else {
185            char *c;
186            read(fd, comm, sizeof(comm) - 1);
187            close(fd);
188
189            c = strrchr(comm, '\n');
190            if (c) {
191                *c = '\0';
192            }
193        }
194        func(pid, tid, comm);
195    }
196
197    closedir(d);
198}
199
200void for_each_tid(for_each_tid_func func, const char *header) {
201    ON_DRY_RUN_RETURN();
202    __for_each_pid(for_each_tid_helper, header, (void *) func);
203}
204
205void show_wchan(int pid, int tid, const char *name) {
206    ON_DRY_RUN_RETURN();
207    char path[255];
208    char buffer[255];
209    int fd;
210    char name_buffer[255];
211
212    memset(buffer, 0, sizeof(buffer));
213
214    sprintf(path, "/proc/%d/wchan", tid);
215    if ((fd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_CLOEXEC))) < 0) {
216        printf("Failed to open '%s' (%s)\n", path, strerror(errno));
217        return;
218    }
219
220    if (read(fd, buffer, sizeof(buffer)) < 0) {
221        printf("Failed to read '%s' (%s)\n", path, strerror(errno));
222        goto out_close;
223    }
224
225    snprintf(name_buffer, sizeof(name_buffer), "%*s%s",
226             pid == tid ? 0 : 3, "", name);
227
228    printf("%-7d %-32s %s\n", tid, name_buffer, buffer);
229
230out_close:
231    close(fd);
232    return;
233}
234
235void do_dmesg() {
236    const char *title = "KERNEL LOG (dmesg)";
237    DurationReporter duration_reporter(title);
238    printf("------ %s ------\n", title);
239
240    ON_DRY_RUN_RETURN();
241    /* Get size of kernel buffer */
242    int size = klogctl(KLOG_SIZE_BUFFER, NULL, 0);
243    if (size <= 0) {
244        printf("Unexpected klogctl return value: %d\n\n", size);
245        return;
246    }
247    char *buf = (char *) malloc(size + 1);
248    if (buf == NULL) {
249        printf("memory allocation failed\n\n");
250        return;
251    }
252    int retval = klogctl(KLOG_READ_ALL, buf, size);
253    if (retval < 0) {
254        printf("klogctl failure\n\n");
255        free(buf);
256        return;
257    }
258    buf[retval] = '\0';
259    printf("%s\n\n", buf);
260    free(buf);
261    return;
262}
263
264void do_showmap(int pid, const char *name) {
265    char title[255];
266    char arg[255];
267
268    sprintf(title, "SHOW MAP %d (%s)", pid, name);
269    sprintf(arg, "%d", pid);
270    run_command(title, 10, SU_PATH, "root", "showmap", arg, NULL);
271}
272
273static int _dump_file_from_fd(const char *title, const char *path, int fd) {
274    if (title) {
275        printf("------ %s (%s", title, path);
276
277        struct stat st;
278        // Only show the modification time of non-device files.
279        size_t path_len = strlen(path);
280        if ((path_len < 6 || memcmp(path, "/proc/", 6)) &&
281                (path_len < 5 || memcmp(path, "/sys/", 5)) &&
282                (path_len < 3 || memcmp(path, "/d/", 3)) &&
283                !fstat(fd, &st)) {
284            char stamp[80];
285            time_t mtime = st.st_mtime;
286            strftime(stamp, sizeof(stamp), "%Y-%m-%d %H:%M:%S", localtime(&mtime));
287            printf(": %s", stamp);
288        }
289        printf(") ------\n");
290    }
291    ON_DRY_RUN({ update_progress(WEIGHT_FILE); close(fd); return 0; });
292
293    bool newline = false;
294    fd_set read_set;
295    struct timeval tm;
296    while (1) {
297        FD_ZERO(&read_set);
298        FD_SET(fd, &read_set);
299        /* Timeout if no data is read for 30 seconds. */
300        tm.tv_sec = 30;
301        tm.tv_usec = 0;
302        uint64_t elapsed = DurationReporter::nanotime();
303        int ret = TEMP_FAILURE_RETRY(select(fd + 1, &read_set, NULL, NULL, &tm));
304        if (ret == -1) {
305            printf("*** %s: select failed: %s\n", path, strerror(errno));
306            newline = true;
307            break;
308        } else if (ret == 0) {
309            elapsed = DurationReporter::nanotime() - elapsed;
310            printf("*** %s: Timed out after %.3fs\n", path,
311                   (float) elapsed / NANOS_PER_SEC);
312            newline = true;
313            break;
314        } else {
315            char buffer[65536];
316            ssize_t bytes_read = TEMP_FAILURE_RETRY(read(fd, buffer, sizeof(buffer)));
317            if (bytes_read > 0) {
318                fwrite(buffer, bytes_read, 1, stdout);
319                newline = (buffer[bytes_read-1] == '\n');
320            } else {
321                if (bytes_read == -1) {
322                    printf("*** %s: Failed to read from fd: %s", path, strerror(errno));
323                    newline = true;
324                }
325                break;
326            }
327        }
328    }
329    update_progress(WEIGHT_FILE);
330    close(fd);
331
332    if (!newline) printf("\n");
333    if (title) printf("\n");
334    return 0;
335}
336
337/* prints the contents of a file */
338int dump_file(const char *title, const char *path) {
339    DurationReporter duration_reporter(title);
340    int fd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_NONBLOCK | O_CLOEXEC));
341    if (fd < 0) {
342        int err = errno;
343        printf("*** %s: %s\n", path, strerror(err));
344        if (title) printf("\n");
345        return -1;
346    }
347    return _dump_file_from_fd(title, path, fd);
348}
349
350/* calls skip to gate calling dump_from_fd recursively
351 * in the specified directory. dump_from_fd defaults to
352 * dump_file_from_fd above when set to NULL. skip defaults
353 * to false when set to NULL. dump_from_fd will always be
354 * called with title NULL.
355 */
356int dump_files(const char *title, const char *dir,
357        bool (*skip)(const char *path),
358        int (*dump_from_fd)(const char *title, const char *path, int fd)) {
359    DurationReporter duration_reporter(title);
360    DIR *dirp;
361    struct dirent *d;
362    char *newpath = NULL;
363    const char *slash = "/";
364    int fd, retval = 0;
365
366    if (title) {
367        printf("------ %s (%s) ------\n", title, dir);
368    }
369    ON_DRY_RUN_RETURN(0);
370
371    if (dir[strlen(dir) - 1] == '/') {
372        ++slash;
373    }
374    dirp = opendir(dir);
375    if (dirp == NULL) {
376        retval = -errno;
377        fprintf(stderr, "%s: %s\n", dir, strerror(errno));
378        return retval;
379    }
380
381    if (!dump_from_fd) {
382        dump_from_fd = dump_file_from_fd;
383    }
384    for (; ((d = readdir(dirp))); free(newpath), newpath = NULL) {
385        if ((d->d_name[0] == '.')
386         && (((d->d_name[1] == '.') && (d->d_name[2] == '\0'))
387          || (d->d_name[1] == '\0'))) {
388            continue;
389        }
390        asprintf(&newpath, "%s%s%s%s", dir, slash, d->d_name,
391                 (d->d_type == DT_DIR) ? "/" : "");
392        if (!newpath) {
393            retval = -errno;
394            continue;
395        }
396        if (skip && (*skip)(newpath)) {
397            continue;
398        }
399        if (d->d_type == DT_DIR) {
400            int ret = dump_files(NULL, newpath, skip, dump_from_fd);
401            if (ret < 0) {
402                retval = ret;
403            }
404            continue;
405        }
406        fd = TEMP_FAILURE_RETRY(open(newpath, O_RDONLY | O_NONBLOCK | O_CLOEXEC));
407        if (fd < 0) {
408            retval = fd;
409            printf("*** %s: %s\n", newpath, strerror(errno));
410            continue;
411        }
412        (*dump_from_fd)(NULL, newpath, fd);
413    }
414    closedir(dirp);
415    if (title) {
416        printf("\n");
417    }
418    return retval;
419}
420
421/* fd must have been opened with the flag O_NONBLOCK. With this flag set,
422 * it's possible to avoid issues where opening the file itself can get
423 * stuck.
424 */
425int dump_file_from_fd(const char *title, const char *path, int fd) {
426    int flags = fcntl(fd, F_GETFL);
427    if (flags == -1) {
428        printf("*** %s: failed to get flags on fd %d: %s\n", path, fd, strerror(errno));
429        close(fd);
430        return -1;
431    } else if (!(flags & O_NONBLOCK)) {
432        printf("*** %s: fd must have O_NONBLOCK set.\n", path);
433        close(fd);
434        return -1;
435    }
436    return _dump_file_from_fd(title, path, fd);
437}
438
439bool waitpid_with_timeout(pid_t pid, int timeout_seconds, int* status) {
440    sigset_t child_mask, old_mask;
441    sigemptyset(&child_mask);
442    sigaddset(&child_mask, SIGCHLD);
443
444    if (sigprocmask(SIG_BLOCK, &child_mask, &old_mask) == -1) {
445        printf("*** sigprocmask failed: %s\n", strerror(errno));
446        return false;
447    }
448
449    struct timespec ts;
450    ts.tv_sec = timeout_seconds;
451    ts.tv_nsec = 0;
452    int ret = TEMP_FAILURE_RETRY(sigtimedwait(&child_mask, NULL, &ts));
453    int saved_errno = errno;
454    // Set the signals back the way they were.
455    if (sigprocmask(SIG_SETMASK, &old_mask, NULL) == -1) {
456        printf("*** sigprocmask failed: %s\n", strerror(errno));
457        if (ret == 0) {
458            return false;
459        }
460    }
461    if (ret == -1) {
462        errno = saved_errno;
463        if (errno == EAGAIN) {
464            errno = ETIMEDOUT;
465        } else {
466            printf("*** sigtimedwait failed: %s\n", strerror(errno));
467        }
468        return false;
469    }
470
471    pid_t child_pid = waitpid(pid, status, WNOHANG);
472    if (child_pid != pid) {
473        if (child_pid != -1) {
474            printf("*** Waiting for pid %d, got pid %d instead\n", pid, child_pid);
475        } else {
476            printf("*** waitpid failed: %s\n", strerror(errno));
477        }
478        return false;
479    }
480    return true;
481}
482
483int run_command(const char *title, int timeout_seconds, const char *command, ...) {
484    DurationReporter duration_reporter(title);
485    fflush(stdout);
486
487    const char *args[1024] = {command};
488    size_t arg;
489    va_list ap;
490    va_start(ap, command);
491    if (title) printf("------ %s (%s", title, command);
492    for (arg = 1; arg < sizeof(args) / sizeof(args[0]); ++arg) {
493        args[arg] = va_arg(ap, const char *);
494        if (args[arg] == NULL) break;
495        if (title) printf(" %s", args[arg]);
496    }
497    if (title) printf(") ------\n");
498    fflush(stdout);
499
500    ON_DRY_RUN({ update_progress(timeout_seconds); va_end(ap); return 0; });
501
502    int status = run_command_always(title, timeout_seconds, args);
503    va_end(ap);
504    return status;
505}
506
507/* forks a command and waits for it to finish */
508int run_command_always(const char *title, int timeout_seconds, const char *args[]) {
509    /* TODO: for now we're simplifying the progress calculation by using the timeout as the weight.
510     * It's a good approximation for most cases, except when calling dumpsys, where its weight
511     * should be much higher proportionally to its timeout. */
512    int weight = timeout_seconds;
513
514    const char *command = args[0];
515    uint64_t start = DurationReporter::nanotime();
516    pid_t pid = fork();
517
518    /* handle error case */
519    if (pid < 0) {
520        printf("*** fork: %s\n", strerror(errno));
521        return pid;
522    }
523
524    /* handle child case */
525    if (pid == 0) {
526
527        /* make sure the child dies when dumpstate dies */
528        prctl(PR_SET_PDEATHSIG, SIGKILL);
529
530        /* just ignore SIGPIPE, will go down with parent's */
531        struct sigaction sigact;
532        memset(&sigact, 0, sizeof(sigact));
533        sigact.sa_handler = SIG_IGN;
534        sigaction(SIGPIPE, &sigact, NULL);
535
536        execvp(command, (char**) args);
537        printf("*** exec(%s): %s\n", command, strerror(errno));
538        fflush(stdout);
539        _exit(-1);
540    }
541
542    /* handle parent case */
543    int status;
544    bool ret = waitpid_with_timeout(pid, timeout_seconds, &status);
545    uint64_t elapsed = DurationReporter::nanotime() - start;
546    if (!ret) {
547        if (errno == ETIMEDOUT) {
548            printf("*** %s: Timed out after %.3fs (killing pid %d)\n", command,
549                   (float) elapsed / NANOS_PER_SEC, pid);
550        } else {
551            printf("*** %s: Error after %.4fs (killing pid %d)\n", command,
552                   (float) elapsed / NANOS_PER_SEC, pid);
553        }
554        kill(pid, SIGTERM);
555        if (!waitpid_with_timeout(pid, 5, NULL)) {
556            kill(pid, SIGKILL);
557            if (!waitpid_with_timeout(pid, 5, NULL)) {
558                printf("*** %s: Cannot kill %d even with SIGKILL.\n", command, pid);
559            }
560        }
561        return -1;
562    }
563
564    if (WIFSIGNALED(status)) {
565        printf("*** %s: Killed by signal %d\n", command, WTERMSIG(status));
566    } else if (WIFEXITED(status) && WEXITSTATUS(status) > 0) {
567        printf("*** %s: Exit code %d\n", command, WEXITSTATUS(status));
568    }
569
570    if (weight > 0) {
571        update_progress(weight);
572    }
573    return status;
574}
575
576void send_broadcast(const std::string& action, const std::vector<std::string>& args) {
577    if (args.size() > 1000) {
578        fprintf(stderr, "send_broadcast: too many arguments (%d)\n", (int) args.size());
579        return;
580    }
581    const char *am_args[1024] = { "/system/bin/am", "broadcast", "--user", "0",
582                                  "-a", action.c_str() };
583    size_t am_index = 5; // Starts at the index of last initial value above.
584    for (const std::string& arg : args) {
585        am_args[++am_index] = arg.c_str();
586    }
587    // Always terminate with NULL.
588    am_args[am_index + 1] = NULL;
589    run_command_always(NULL, 5, am_args);
590}
591
592size_t num_props = 0;
593static char* props[2000];
594
595static void print_prop(const char *key, const char *name, void *user) {
596    (void) user;
597    if (num_props < sizeof(props) / sizeof(props[0])) {
598        char buf[PROPERTY_KEY_MAX + PROPERTY_VALUE_MAX + 10];
599        snprintf(buf, sizeof(buf), "[%s]: [%s]\n", key, name);
600        props[num_props++] = strdup(buf);
601    }
602}
603
604static int compare_prop(const void *a, const void *b) {
605    return strcmp(*(char * const *) a, *(char * const *) b);
606}
607
608/* prints all the system properties */
609void print_properties() {
610    const char* title = "SYSTEM PROPERTIES";
611    DurationReporter duration_reporter(title);
612    printf("------ %s ------\n", title);
613    ON_DRY_RUN_RETURN();
614    size_t i;
615    num_props = 0;
616    property_list(print_prop, NULL);
617    qsort(&props, num_props, sizeof(props[0]), compare_prop);
618
619    for (i = 0; i < num_props; ++i) {
620        fputs(props[i], stdout);
621        free(props[i]);
622    }
623    printf("\n");
624}
625
626/* redirect output to a service control socket */
627void redirect_to_socket(FILE *redirect, const char *service) {
628    int s = android_get_control_socket(service);
629    if (s < 0) {
630        fprintf(stderr, "android_get_control_socket(%s): %s\n", service, strerror(errno));
631        exit(1);
632    }
633    fcntl(s, F_SETFD, FD_CLOEXEC);
634    if (listen(s, 4) < 0) {
635        fprintf(stderr, "listen(control socket): %s\n", strerror(errno));
636        exit(1);
637    }
638
639    struct sockaddr addr;
640    socklen_t alen = sizeof(addr);
641    int fd = accept(s, &addr, &alen);
642    if (fd < 0) {
643        fprintf(stderr, "accept(control socket): %s\n", strerror(errno));
644        exit(1);
645    }
646
647    fflush(redirect);
648    dup2(fd, fileno(redirect));
649    close(fd);
650}
651
652/* redirect output to a file */
653void redirect_to_file(FILE *redirect, char *path) {
654    char *chp = path;
655
656    /* skip initial slash */
657    if (chp[0] == '/')
658        chp++;
659
660    /* create leading directories, if necessary */
661    while (chp && chp[0]) {
662        chp = strchr(chp, '/');
663        if (chp) {
664            *chp = 0;
665            mkdir(path, 0770);  /* drwxrwx--- */
666            *chp++ = '/';
667        }
668    }
669
670    int fd = TEMP_FAILURE_RETRY(open(path, O_WRONLY | O_CREAT | O_TRUNC | O_CLOEXEC,
671                                     S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH));
672    if (fd < 0) {
673        fprintf(stderr, "%s: %s\n", path, strerror(errno));
674        exit(1);
675    }
676
677    TEMP_FAILURE_RETRY(dup2(fd, fileno(redirect)));
678    close(fd);
679}
680
681static bool should_dump_native_traces(const char* path) {
682    for (const char** p = native_processes_to_dump; *p; p++) {
683        if (!strcmp(*p, path)) {
684            return true;
685        }
686    }
687    return false;
688}
689
690/* dump Dalvik and native stack traces, return the trace file location (NULL if none) */
691const char *dump_traces() {
692    DurationReporter duration_reporter("DUMP TRACES");
693    ON_DRY_RUN_RETURN(NULL);
694    const char* result = NULL;
695
696    char traces_path[PROPERTY_VALUE_MAX] = "";
697    property_get("dalvik.vm.stack-trace-file", traces_path, "");
698    if (!traces_path[0]) return NULL;
699
700    /* move the old traces.txt (if any) out of the way temporarily */
701    char anr_traces_path[PATH_MAX];
702    strlcpy(anr_traces_path, traces_path, sizeof(anr_traces_path));
703    strlcat(anr_traces_path, ".anr", sizeof(anr_traces_path));
704    if (rename(traces_path, anr_traces_path) && errno != ENOENT) {
705        fprintf(stderr, "rename(%s, %s): %s\n", traces_path, anr_traces_path, strerror(errno));
706        return NULL;  // Can't rename old traces.txt -- no permission? -- leave it alone instead
707    }
708
709    /* create a new, empty traces.txt file to receive stack dumps */
710    int fd = TEMP_FAILURE_RETRY(open(traces_path, O_CREAT | O_WRONLY | O_TRUNC | O_NOFOLLOW | O_CLOEXEC,
711                                     0666));  /* -rw-rw-rw- */
712    if (fd < 0) {
713        fprintf(stderr, "%s: %s\n", traces_path, strerror(errno));
714        return NULL;
715    }
716    int chmod_ret = fchmod(fd, 0666);
717    if (chmod_ret < 0) {
718        fprintf(stderr, "fchmod on %s failed: %s\n", traces_path, strerror(errno));
719        close(fd);
720        return NULL;
721    }
722
723    /* Variables below must be initialized before 'goto' statements */
724    int dalvik_found = 0;
725    int ifd, wfd = -1;
726
727    /* walk /proc and kill -QUIT all Dalvik processes */
728    DIR *proc = opendir("/proc");
729    if (proc == NULL) {
730        fprintf(stderr, "/proc: %s\n", strerror(errno));
731        goto error_close_fd;
732    }
733
734    /* use inotify to find when processes are done dumping */
735    ifd = inotify_init();
736    if (ifd < 0) {
737        fprintf(stderr, "inotify_init: %s\n", strerror(errno));
738        goto error_close_fd;
739    }
740
741    wfd = inotify_add_watch(ifd, traces_path, IN_CLOSE_WRITE);
742    if (wfd < 0) {
743        fprintf(stderr, "inotify_add_watch(%s): %s\n", traces_path, strerror(errno));
744        goto error_close_ifd;
745    }
746
747    struct dirent *d;
748    while ((d = readdir(proc))) {
749        int pid = atoi(d->d_name);
750        if (pid <= 0) continue;
751
752        char path[PATH_MAX];
753        char data[PATH_MAX];
754        snprintf(path, sizeof(path), "/proc/%d/exe", pid);
755        ssize_t len = readlink(path, data, sizeof(data) - 1);
756        if (len <= 0) {
757            continue;
758        }
759        data[len] = '\0';
760
761        if (!strncmp(data, "/system/bin/app_process", strlen("/system/bin/app_process"))) {
762            /* skip zygote -- it won't dump its stack anyway */
763            snprintf(path, sizeof(path), "/proc/%d/cmdline", pid);
764            int cfd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_CLOEXEC));
765            len = read(cfd, data, sizeof(data) - 1);
766            close(cfd);
767            if (len <= 0) {
768                continue;
769            }
770            data[len] = '\0';
771            if (!strncmp(data, "zygote", strlen("zygote"))) {
772                continue;
773            }
774
775            ++dalvik_found;
776            uint64_t start = DurationReporter::nanotime();
777            if (kill(pid, SIGQUIT)) {
778                fprintf(stderr, "kill(%d, SIGQUIT): %s\n", pid, strerror(errno));
779                continue;
780            }
781
782            /* wait for the writable-close notification from inotify */
783            struct pollfd pfd = { ifd, POLLIN, 0 };
784            int ret = poll(&pfd, 1, 5000);  /* 5 sec timeout */
785            if (ret < 0) {
786                fprintf(stderr, "poll: %s\n", strerror(errno));
787            } else if (ret == 0) {
788                fprintf(stderr, "warning: timed out dumping pid %d\n", pid);
789            } else {
790                struct inotify_event ie;
791                read(ifd, &ie, sizeof(ie));
792            }
793
794            if (lseek(fd, 0, SEEK_END) < 0) {
795                fprintf(stderr, "lseek: %s\n", strerror(errno));
796            } else {
797                dprintf(fd, "[dump dalvik stack %d: %.3fs elapsed]\n",
798                        pid, (float)(DurationReporter::nanotime() - start) / NANOS_PER_SEC);
799            }
800        } else if (should_dump_native_traces(data)) {
801            /* dump native process if appropriate */
802            if (lseek(fd, 0, SEEK_END) < 0) {
803                fprintf(stderr, "lseek: %s\n", strerror(errno));
804            } else {
805                static uint16_t timeout_failures = 0;
806                uint64_t start = DurationReporter::nanotime();
807
808                /* If 3 backtrace dumps fail in a row, consider debuggerd dead. */
809                if (timeout_failures == 3) {
810                    dprintf(fd, "too many stack dump failures, skipping...\n");
811                } else if (dump_backtrace_to_file_timeout(pid, fd, 20) == -1) {
812                    dprintf(fd, "dumping failed, likely due to a timeout\n");
813                    timeout_failures++;
814                } else {
815                    timeout_failures = 0;
816                }
817                dprintf(fd, "[dump native stack %d: %.3fs elapsed]\n",
818                        pid, (float)(DurationReporter::nanotime() - start) / NANOS_PER_SEC);
819            }
820        }
821    }
822
823    if (dalvik_found == 0) {
824        fprintf(stderr, "Warning: no Dalvik processes found to dump stacks\n");
825    }
826
827    static char dump_traces_path[PATH_MAX];
828    strlcpy(dump_traces_path, traces_path, sizeof(dump_traces_path));
829    strlcat(dump_traces_path, ".bugreport", sizeof(dump_traces_path));
830    if (rename(traces_path, dump_traces_path)) {
831        fprintf(stderr, "rename(%s, %s): %s\n", traces_path, dump_traces_path, strerror(errno));
832        goto error_close_ifd;
833    }
834    result = dump_traces_path;
835
836    /* replace the saved [ANR] traces.txt file */
837    rename(anr_traces_path, traces_path);
838
839error_close_ifd:
840    close(ifd);
841error_close_fd:
842    close(fd);
843    return result;
844}
845
846void dump_route_tables() {
847    DurationReporter duration_reporter("DUMP ROUTE TABLES");
848    ON_DRY_RUN_RETURN();
849    const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables";
850    dump_file("RT_TABLES", RT_TABLES_PATH);
851    FILE* fp = fopen(RT_TABLES_PATH, "re");
852    if (!fp) {
853        printf("*** %s: %s\n", RT_TABLES_PATH, strerror(errno));
854        return;
855    }
856    char table[16];
857    // Each line has an integer (the table number), a space, and a string (the table name). We only
858    // need the table number. It's a 32-bit unsigned number, so max 10 chars. Skip the table name.
859    // Add a fixed max limit so this doesn't go awry.
860    for (int i = 0; i < 64 && fscanf(fp, " %10s %*s", table) == 1; ++i) {
861        run_command("ROUTE TABLE IPv4", 10, "ip", "-4", "route", "show", "table", table, NULL);
862        run_command("ROUTE TABLE IPv6", 10, "ip", "-6", "route", "show", "table", table, NULL);
863    }
864    fclose(fp);
865}
866
867/* overall progress */
868int progress = 0;
869int do_update_progress = 0; // Set by dumpstate.cpp
870int weight_total = WEIGHT_TOTAL;
871
872// TODO: make this function thread safe if sections are generated in parallel.
873void update_progress(int delta) {
874    if (!do_update_progress) return;
875
876    progress += delta;
877
878    char key[PROPERTY_KEY_MAX];
879    char value[PROPERTY_VALUE_MAX];
880
881    // adjusts max on the fly
882    if (progress > weight_total) {
883        int new_total = weight_total * 1.2;
884        fprintf(stderr, "Adjusting total weight from %d to %d\n", weight_total, new_total);
885        weight_total = new_total;
886        sprintf(key, "dumpstate.%d.max", getpid());
887        sprintf(value, "%d", weight_total);
888        int status = property_set(key, value);
889        if (status) {
890            ALOGW("Could not update max weight by setting system property %s to %s: %d\n",
891                    key, value, status);
892        }
893    }
894
895    sprintf(key, "dumpstate.%d.progress", getpid());
896    sprintf(value, "%d", progress);
897
898    // stderr is ignored on normal invocations, but useful when calling /system/bin/dumpstate
899    // directly for debuggging.
900    fprintf(stderr, "Setting progress (%s): %s/%d\n", key, value, weight_total);
901
902    int status = property_set(key, value);
903    if (status) {
904        ALOGW("Could not update progress by setting system property %s to %s: %d\n",
905                key, value, status);
906    }
907}
908
909void take_screenshot(const std::string& path) {
910    const char *args[] = { "/system/bin/screencap", "-p", path.c_str(), NULL };
911    run_command_always(NULL, 10, args);
912}
913
914void vibrate(FILE* vibrator, int ms) {
915    fprintf(vibrator, "%d\n", ms);
916    fflush(vibrator);
917}
918
919bool is_dir(const char* pathname) {
920    struct stat info;
921    if (stat(pathname, &info) == -1) {
922        return false;
923    }
924    return S_ISDIR(info.st_mode);
925}
926
927time_t get_mtime(int fd, time_t default_mtime) {
928    struct stat info;
929    if (fstat(fd, &info) == -1) {
930        return default_mtime;
931    }
932    return info.st_mtime;
933}
934
935void dump_emmc_ecsd(const char *ext_csd_path) {
936    static const size_t EXT_CSD_REV = 192;
937    static const size_t EXT_PRE_EOL_INFO = 267;
938    static const size_t EXT_DEVICE_LIFE_TIME_EST_TYP_A = 268;
939    static const size_t EXT_DEVICE_LIFE_TIME_EST_TYP_B = 269;
940    struct hex {
941        char str[2];
942    } buffer[512];
943    int fd, ext_csd_rev, ext_pre_eol_info;
944    ssize_t bytes_read;
945    static const char *ver_str[] = {
946        "4.0", "4.1", "4.2", "4.3", "Obsolete", "4.41", "4.5", "5.0"
947    };
948    static const char *eol_str[] = {
949        "Undefined",
950        "Normal",
951        "Warning (consumed 80% of reserve)",
952        "Urgent (consumed 90% of reserve)"
953    };
954
955    printf("------ %s Extended CSD ------\n", ext_csd_path);
956
957    fd = TEMP_FAILURE_RETRY(open(ext_csd_path,
958                                 O_RDONLY | O_NONBLOCK | O_CLOEXEC));
959    if (fd < 0) {
960        printf("*** %s: %s\n\n", ext_csd_path, strerror(errno));
961        return;
962    }
963
964    bytes_read = TEMP_FAILURE_RETRY(read(fd, buffer, sizeof(buffer)));
965    close(fd);
966    if (bytes_read < 0) {
967        printf("*** %s: %s\n\n", ext_csd_path, strerror(errno));
968        return;
969    }
970    if (bytes_read < (ssize_t)(EXT_CSD_REV * sizeof(struct hex))) {
971        printf("*** %s: truncated content %zd\n\n", ext_csd_path, bytes_read);
972        return;
973    }
974
975    ext_csd_rev = 0;
976    if (sscanf(buffer[EXT_CSD_REV].str, "%02x", &ext_csd_rev) != 1) {
977        printf("*** %s: EXT_CSD_REV parse error \"%.2s\"\n\n",
978               ext_csd_path, buffer[EXT_CSD_REV].str);
979        return;
980    }
981
982    printf("rev 1.%d (MMC %s)\n",
983           ext_csd_rev,
984           (ext_csd_rev < (int)(sizeof(ver_str) / sizeof(ver_str[0]))) ?
985               ver_str[ext_csd_rev] :
986               "Unknown");
987    if (ext_csd_rev < 7) {
988        printf("\n");
989        return;
990    }
991
992    if (bytes_read < (ssize_t)(EXT_PRE_EOL_INFO * sizeof(struct hex))) {
993        printf("*** %s: truncated content %zd\n\n", ext_csd_path, bytes_read);
994        return;
995    }
996
997    ext_pre_eol_info = 0;
998    if (sscanf(buffer[EXT_PRE_EOL_INFO].str, "%02x", &ext_pre_eol_info) != 1) {
999        printf("*** %s: PRE_EOL_INFO parse error \"%.2s\"\n\n",
1000               ext_csd_path, buffer[EXT_PRE_EOL_INFO].str);
1001        return;
1002    }
1003    printf("PRE_EOL_INFO %d (MMC %s)\n",
1004           ext_pre_eol_info,
1005           eol_str[(ext_pre_eol_info < (int)
1006                       (sizeof(eol_str) / sizeof(eol_str[0]))) ?
1007                           ext_pre_eol_info : 0]);
1008
1009    for (size_t lifetime = EXT_DEVICE_LIFE_TIME_EST_TYP_A;
1010            lifetime <= EXT_DEVICE_LIFE_TIME_EST_TYP_B;
1011            ++lifetime) {
1012        int ext_device_life_time_est;
1013        static const char *est_str[] = {
1014            "Undefined",
1015            "0-10% of device lifetime used",
1016            "10-20% of device lifetime used",
1017            "20-30% of device lifetime used",
1018            "30-40% of device lifetime used",
1019            "40-50% of device lifetime used",
1020            "50-60% of device lifetime used",
1021            "60-70% of device lifetime used",
1022            "70-80% of device lifetime used",
1023            "80-90% of device lifetime used",
1024            "90-100% of device lifetime used",
1025            "Exceeded the maximum estimated device lifetime",
1026        };
1027
1028        if (bytes_read < (ssize_t)(lifetime * sizeof(struct hex))) {
1029            printf("*** %s: truncated content %zd\n", ext_csd_path, bytes_read);
1030            break;
1031        }
1032
1033        ext_device_life_time_est = 0;
1034        if (sscanf(buffer[lifetime].str, "%02x", &ext_device_life_time_est) != 1) {
1035            printf("*** %s: DEVICE_LIFE_TIME_EST_TYP_%c parse error \"%.2s\"\n",
1036                   ext_csd_path,
1037                   (unsigned)(lifetime - EXT_DEVICE_LIFE_TIME_EST_TYP_A) + 'A',
1038                   buffer[lifetime].str);
1039            continue;
1040        }
1041        printf("DEVICE_LIFE_TIME_EST_TYP_%c %d (MMC %s)\n",
1042               (unsigned)(lifetime - EXT_DEVICE_LIFE_TIME_EST_TYP_A) + 'A',
1043               ext_device_life_time_est,
1044               est_str[(ext_device_life_time_est < (int)
1045                           (sizeof(est_str) / sizeof(est_str[0]))) ?
1046                               ext_device_life_time_est : 0]);
1047    }
1048
1049    printf("\n");
1050}
1051