utils.cpp revision 3634a1e3459cb609da709646107e246cafbc01f9 
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/drmserver",
54        "/system/bin/mediaserver",
55        "/system/bin/sdcard",
56        "/system/bin/surfaceflinger",
57        "/system/bin/vehicle_network_service",
58        NULL,
59};
60
61static uint64_t nanotime() {
62    struct timespec ts;
63    clock_gettime(CLOCK_MONOTONIC, &ts);
64    return (uint64_t)ts.tv_sec * NANOS_PER_SEC + ts.tv_nsec;
65}
66
67void for_each_userid(void (*func)(int), const char *header) {
68    ON_DRY_RUN_RETURN();
69    DIR *d;
70    struct dirent *de;
71
72    if (header) printf("\n------ %s ------\n", header);
73    func(0);
74
75    if (!(d = opendir("/data/system/users"))) {
76        printf("Failed to open /data/system/users (%s)\n", strerror(errno));
77        return;
78    }
79
80    while ((de = readdir(d))) {
81        int userid;
82        if (de->d_type != DT_DIR || !(userid = atoi(de->d_name))) {
83            continue;
84        }
85        func(userid);
86    }
87
88    closedir(d);
89}
90
91static void __for_each_pid(void (*helper)(int, const char *, void *), const char *header, void *arg) {
92    DIR *d;
93    struct dirent *de;
94
95    if (!(d = opendir("/proc"))) {
96        printf("Failed to open /proc (%s)\n", strerror(errno));
97        return;
98    }
99
100    printf("\n------ %s ------\n", header);
101    while ((de = readdir(d))) {
102        int pid;
103        int fd;
104        char cmdpath[255];
105        char cmdline[255];
106
107        if (!(pid = atoi(de->d_name))) {
108            continue;
109        }
110
111        sprintf(cmdpath,"/proc/%d/cmdline", pid);
112        memset(cmdline, 0, sizeof(cmdline));
113        if ((fd = TEMP_FAILURE_RETRY(open(cmdpath, O_RDONLY | O_CLOEXEC))) < 0) {
114            strcpy(cmdline, "N/A");
115        } else {
116            read(fd, cmdline, sizeof(cmdline) - 1);
117            close(fd);
118        }
119        helper(pid, cmdline, arg);
120    }
121
122    closedir(d);
123}
124
125static void for_each_pid_helper(int pid, const char *cmdline, void *arg) {
126    for_each_pid_func *func = (for_each_pid_func*) arg;
127    func(pid, cmdline);
128}
129
130void for_each_pid(for_each_pid_func func, const char *header) {
131    ON_DRY_RUN_RETURN();
132  __for_each_pid(for_each_pid_helper, header, (void *)func);
133}
134
135static void for_each_tid_helper(int pid, const char *cmdline, void *arg) {
136    DIR *d;
137    struct dirent *de;
138    char taskpath[255];
139    for_each_tid_func *func = (for_each_tid_func *) arg;
140
141    sprintf(taskpath, "/proc/%d/task", pid);
142
143    if (!(d = opendir(taskpath))) {
144        printf("Failed to open %s (%s)\n", taskpath, strerror(errno));
145        return;
146    }
147
148    func(pid, pid, cmdline);
149
150    while ((de = readdir(d))) {
151        int tid;
152        int fd;
153        char commpath[255];
154        char comm[255];
155
156        if (!(tid = atoi(de->d_name))) {
157            continue;
158        }
159
160        if (tid == pid)
161            continue;
162
163        sprintf(commpath,"/proc/%d/comm", tid);
164        memset(comm, 0, sizeof(comm));
165        if ((fd = TEMP_FAILURE_RETRY(open(commpath, O_RDONLY | O_CLOEXEC))) < 0) {
166            strcpy(comm, "N/A");
167        } else {
168            char *c;
169            read(fd, comm, sizeof(comm) - 1);
170            close(fd);
171
172            c = strrchr(comm, '\n');
173            if (c) {
174                *c = '\0';
175            }
176        }
177        func(pid, tid, comm);
178    }
179
180    closedir(d);
181}
182
183void for_each_tid(for_each_tid_func func, const char *header) {
184    ON_DRY_RUN_RETURN();
185    __for_each_pid(for_each_tid_helper, header, (void *) func);
186}
187
188void show_wchan(int pid, int tid, const char *name) {
189    ON_DRY_RUN_RETURN();
190    char path[255];
191    char buffer[255];
192    int fd;
193    char name_buffer[255];
194
195    memset(buffer, 0, sizeof(buffer));
196
197    sprintf(path, "/proc/%d/wchan", tid);
198    if ((fd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_CLOEXEC))) < 0) {
199        printf("Failed to open '%s' (%s)\n", path, strerror(errno));
200        return;
201    }
202
203    if (read(fd, buffer, sizeof(buffer)) < 0) {
204        printf("Failed to read '%s' (%s)\n", path, strerror(errno));
205        goto out_close;
206    }
207
208    snprintf(name_buffer, sizeof(name_buffer), "%*s%s",
209             pid == tid ? 0 : 3, "", name);
210
211    printf("%-7d %-32s %s\n", tid, name_buffer, buffer);
212
213out_close:
214    close(fd);
215    return;
216}
217
218void do_dmesg() {
219    printf("------ KERNEL LOG (dmesg) ------\n");
220    ON_DRY_RUN_RETURN();
221    /* Get size of kernel buffer */
222    int size = klogctl(KLOG_SIZE_BUFFER, NULL, 0);
223    if (size <= 0) {
224        printf("Unexpected klogctl return value: %d\n\n", size);
225        return;
226    }
227    char *buf = (char *) malloc(size + 1);
228    if (buf == NULL) {
229        printf("memory allocation failed\n\n");
230        return;
231    }
232    int retval = klogctl(KLOG_READ_ALL, buf, size);
233    if (retval < 0) {
234        printf("klogctl failure\n\n");
235        free(buf);
236        return;
237    }
238    buf[retval] = '\0';
239    printf("%s\n\n", buf);
240    free(buf);
241    return;
242}
243
244void do_showmap(int pid, const char *name) {
245    char title[255];
246    char arg[255];
247
248    sprintf(title, "SHOW MAP %d (%s)", pid, name);
249    sprintf(arg, "%d", pid);
250    run_command(title, 10, SU_PATH, "root", "showmap", arg, NULL);
251}
252
253static int _dump_file_from_fd(const char *title, const char *path, int fd) {
254    if (title) {
255        printf("------ %s (%s", title, path);
256
257        struct stat st;
258        // Only show the modification time of non-device files.
259        size_t path_len = strlen(path);
260        if ((path_len < 6 || memcmp(path, "/proc/", 6)) &&
261                (path_len < 5 || memcmp(path, "/sys/", 5)) &&
262                (path_len < 3 || memcmp(path, "/d/", 3)) &&
263                !fstat(fd, &st)) {
264            char stamp[80];
265            time_t mtime = st.st_mtime;
266            strftime(stamp, sizeof(stamp), "%Y-%m-%d %H:%M:%S", localtime(&mtime));
267            printf(": %s", stamp);
268        }
269        printf(") ------\n");
270    }
271    ON_DRY_RUN({ update_progress(WEIGHT_FILE); close(fd); return 0; });
272
273    bool newline = false;
274    fd_set read_set;
275    struct timeval tm;
276    while (1) {
277        FD_ZERO(&read_set);
278        FD_SET(fd, &read_set);
279        /* Timeout if no data is read for 30 seconds. */
280        tm.tv_sec = 30;
281        tm.tv_usec = 0;
282        uint64_t elapsed = nanotime();
283        int ret = TEMP_FAILURE_RETRY(select(fd + 1, &read_set, NULL, NULL, &tm));
284        if (ret == -1) {
285            printf("*** %s: select failed: %s\n", path, strerror(errno));
286            newline = true;
287            break;
288        } else if (ret == 0) {
289            elapsed = nanotime() - elapsed;
290            printf("*** %s: Timed out after %.3fs\n", path,
291                   (float) elapsed / NANOS_PER_SEC);
292            newline = true;
293            break;
294        } else {
295            char buffer[65536];
296            ssize_t bytes_read = TEMP_FAILURE_RETRY(read(fd, buffer, sizeof(buffer)));
297            if (bytes_read > 0) {
298                fwrite(buffer, bytes_read, 1, stdout);
299                newline = (buffer[bytes_read-1] == '\n');
300            } else {
301                if (bytes_read == -1) {
302                    printf("*** %s: Failed to read from fd: %s", path, strerror(errno));
303                    newline = true;
304                }
305                break;
306            }
307        }
308    }
309    update_progress(WEIGHT_FILE);
310    close(fd);
311
312    if (!newline) printf("\n");
313    if (title) printf("\n");
314    return 0;
315}
316
317/* prints the contents of a file */
318int dump_file(const char *title, const char *path) {
319    int fd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_NONBLOCK | O_CLOEXEC));
320    if (fd < 0) {
321        int err = errno;
322        printf("*** %s: %s\n", path, strerror(err));
323        if (title) printf("\n");
324        return -1;
325    }
326    return _dump_file_from_fd(title, path, fd);
327}
328
329/* calls skip to gate calling dump_from_fd recursively
330 * in the specified directory. dump_from_fd defaults to
331 * dump_file_from_fd above when set to NULL. skip defaults
332 * to false when set to NULL. dump_from_fd will always be
333 * called with title NULL.
334 */
335int dump_files(const char *title, const char *dir,
336        bool (*skip)(const char *path),
337        int (*dump_from_fd)(const char *title, const char *path, int fd)) {
338    DIR *dirp;
339    struct dirent *d;
340    char *newpath = NULL;
341    const char *slash = "/";
342    int fd, retval = 0;
343
344    if (title) {
345        printf("------ %s (%s) ------\n", title, dir);
346    }
347    ON_DRY_RUN_RETURN(0);
348
349    if (dir[strlen(dir) - 1] == '/') {
350        ++slash;
351    }
352    dirp = opendir(dir);
353    if (dirp == NULL) {
354        retval = -errno;
355        fprintf(stderr, "%s: %s\n", dir, strerror(errno));
356        return retval;
357    }
358
359    if (!dump_from_fd) {
360        dump_from_fd = dump_file_from_fd;
361    }
362    for (; ((d = readdir(dirp))); free(newpath), newpath = NULL) {
363        if ((d->d_name[0] == '.')
364         && (((d->d_name[1] == '.') && (d->d_name[2] == '\0'))
365          || (d->d_name[1] == '\0'))) {
366            continue;
367        }
368        asprintf(&newpath, "%s%s%s%s", dir, slash, d->d_name,
369                 (d->d_type == DT_DIR) ? "/" : "");
370        if (!newpath) {
371            retval = -errno;
372            continue;
373        }
374        if (skip && (*skip)(newpath)) {
375            continue;
376        }
377        if (d->d_type == DT_DIR) {
378            int ret = dump_files(NULL, newpath, skip, dump_from_fd);
379            if (ret < 0) {
380                retval = ret;
381            }
382            continue;
383        }
384        fd = TEMP_FAILURE_RETRY(open(newpath, O_RDONLY | O_NONBLOCK | O_CLOEXEC));
385        if (fd < 0) {
386            retval = fd;
387            printf("*** %s: %s\n", newpath, strerror(errno));
388            continue;
389        }
390        (*dump_from_fd)(NULL, newpath, fd);
391    }
392    closedir(dirp);
393    if (title) {
394        printf("\n");
395    }
396    return retval;
397}
398
399/* fd must have been opened with the flag O_NONBLOCK. With this flag set,
400 * it's possible to avoid issues where opening the file itself can get
401 * stuck.
402 */
403int dump_file_from_fd(const char *title, const char *path, int fd) {
404    int flags = fcntl(fd, F_GETFL);
405    if (flags == -1) {
406        printf("*** %s: failed to get flags on fd %d: %s\n", path, fd, strerror(errno));
407        close(fd);
408        return -1;
409    } else if (!(flags & O_NONBLOCK)) {
410        printf("*** %s: fd must have O_NONBLOCK set.\n", path);
411        close(fd);
412        return -1;
413    }
414    return _dump_file_from_fd(title, path, fd);
415}
416
417bool waitpid_with_timeout(pid_t pid, int timeout_seconds, int* status) {
418    sigset_t child_mask, old_mask;
419    sigemptyset(&child_mask);
420    sigaddset(&child_mask, SIGCHLD);
421
422    if (sigprocmask(SIG_BLOCK, &child_mask, &old_mask) == -1) {
423        printf("*** sigprocmask failed: %s\n", strerror(errno));
424        return false;
425    }
426
427    struct timespec ts;
428    ts.tv_sec = timeout_seconds;
429    ts.tv_nsec = 0;
430    int ret = TEMP_FAILURE_RETRY(sigtimedwait(&child_mask, NULL, &ts));
431    int saved_errno = errno;
432    // Set the signals back the way they were.
433    if (sigprocmask(SIG_SETMASK, &old_mask, NULL) == -1) {
434        printf("*** sigprocmask failed: %s\n", strerror(errno));
435        if (ret == 0) {
436            return false;
437        }
438    }
439    if (ret == -1) {
440        errno = saved_errno;
441        if (errno == EAGAIN) {
442            errno = ETIMEDOUT;
443        } else {
444            printf("*** sigtimedwait failed: %s\n", strerror(errno));
445        }
446        return false;
447    }
448
449    pid_t child_pid = waitpid(pid, status, WNOHANG);
450    if (child_pid != pid) {
451        if (child_pid != -1) {
452            printf("*** Waiting for pid %d, got pid %d instead\n", pid, child_pid);
453        } else {
454            printf("*** waitpid failed: %s\n", strerror(errno));
455        }
456        return false;
457    }
458    return true;
459}
460
461int run_command(const char *title, int timeout_seconds, const char *command, ...) {
462    fflush(stdout);
463
464    const char *args[1024] = {command};
465    size_t arg;
466    va_list ap;
467    va_start(ap, command);
468    if (title) printf("------ %s (%s", title, command);
469    for (arg = 1; arg < sizeof(args) / sizeof(args[0]); ++arg) {
470        args[arg] = va_arg(ap, const char *);
471        if (args[arg] == NULL) break;
472        if (title) printf(" %s", args[arg]);
473    }
474    if (title) printf(") ------\n");
475    fflush(stdout);
476
477    ON_DRY_RUN({ update_progress(timeout_seconds); va_end(ap); return 0; });
478
479    int status = run_command_always(title, timeout_seconds, args);
480    va_end(ap);
481    return status;
482}
483
484/* forks a command and waits for it to finish */
485int run_command_always(const char *title, int timeout_seconds, const char *args[]) {
486    /* TODO: for now we're simplifying the progress calculation by using the timeout as the weight.
487     * It's a good approximation for most cases, except when calling dumpsys, where its weight
488     * should be much higher proportionally to its timeout. */
489    int weight = timeout_seconds;
490
491    const char *command = args[0];
492    uint64_t start = nanotime();
493    pid_t pid = fork();
494
495    /* handle error case */
496    if (pid < 0) {
497        printf("*** fork: %s\n", strerror(errno));
498        return pid;
499    }
500
501    /* handle child case */
502    if (pid == 0) {
503
504        /* make sure the child dies when dumpstate dies */
505        prctl(PR_SET_PDEATHSIG, SIGKILL);
506
507        /* just ignore SIGPIPE, will go down with parent's */
508        struct sigaction sigact;
509        memset(&sigact, 0, sizeof(sigact));
510        sigact.sa_handler = SIG_IGN;
511        sigaction(SIGPIPE, &sigact, NULL);
512
513        execvp(command, (char**) args);
514        printf("*** exec(%s): %s\n", command, strerror(errno));
515        fflush(stdout);
516        _exit(-1);
517    }
518
519    /* handle parent case */
520    int status;
521    bool ret = waitpid_with_timeout(pid, timeout_seconds, &status);
522    uint64_t elapsed = nanotime() - start;
523    if (!ret) {
524        if (errno == ETIMEDOUT) {
525            printf("*** %s: Timed out after %.3fs (killing pid %d)\n", command,
526                   (float) elapsed / NANOS_PER_SEC, pid);
527        } else {
528            printf("*** %s: Error after %.4fs (killing pid %d)\n", command,
529                   (float) elapsed / NANOS_PER_SEC, pid);
530        }
531        kill(pid, SIGTERM);
532        if (!waitpid_with_timeout(pid, 5, NULL)) {
533            kill(pid, SIGKILL);
534            if (!waitpid_with_timeout(pid, 5, NULL)) {
535                printf("*** %s: Cannot kill %d even with SIGKILL.\n", command, pid);
536            }
537        }
538        return -1;
539    }
540
541    if (WIFSIGNALED(status)) {
542        printf("*** %s: Killed by signal %d\n", command, WTERMSIG(status));
543    } else if (WIFEXITED(status) && WEXITSTATUS(status) > 0) {
544        printf("*** %s: Exit code %d\n", command, WEXITSTATUS(status));
545    }
546    if (title) printf("[%s: %.3fs elapsed]\n\n", command, (float)elapsed / NANOS_PER_SEC);
547
548    if (weight > 0) {
549        update_progress(weight);
550    }
551    return status;
552}
553
554void send_broadcast(const std::string& action, const std::vector<std::string>& args) {
555    if (args.size() > 1000) {
556        fprintf(stderr, "send_broadcast: too many arguments (%d)\n", (int) args.size());
557        return;
558    }
559    const char *am_args[1024] = { "/system/bin/am", "broadcast", "--user", "0",
560                                  "-a", action.c_str() };
561    size_t am_index = 5; // Starts at the index of last initial value above.
562    for (const std::string& arg : args) {
563        am_args[++am_index] = arg.c_str();
564    }
565    // Always terminate with NULL.
566    am_args[am_index + 1] = NULL;
567    run_command_always(NULL, 5, am_args);
568}
569
570size_t num_props = 0;
571static char* props[2000];
572
573static void print_prop(const char *key, const char *name, void *user) {
574    (void) user;
575    if (num_props < sizeof(props) / sizeof(props[0])) {
576        char buf[PROPERTY_KEY_MAX + PROPERTY_VALUE_MAX + 10];
577        snprintf(buf, sizeof(buf), "[%s]: [%s]\n", key, name);
578        props[num_props++] = strdup(buf);
579    }
580}
581
582static int compare_prop(const void *a, const void *b) {
583    return strcmp(*(char * const *) a, *(char * const *) b);
584}
585
586/* prints all the system properties */
587void print_properties() {
588    printf("------ SYSTEM PROPERTIES ------\n");
589    ON_DRY_RUN_RETURN();
590    size_t i;
591    num_props = 0;
592    property_list(print_prop, NULL);
593    qsort(&props, num_props, sizeof(props[0]), compare_prop);
594
595    for (i = 0; i < num_props; ++i) {
596        fputs(props[i], stdout);
597        free(props[i]);
598    }
599    printf("\n");
600}
601
602/* redirect output to a service control socket */
603void redirect_to_socket(FILE *redirect, const char *service) {
604    int s = android_get_control_socket(service);
605    if (s < 0) {
606        fprintf(stderr, "android_get_control_socket(%s): %s\n", service, strerror(errno));
607        exit(1);
608    }
609    fcntl(s, F_SETFD, FD_CLOEXEC);
610    if (listen(s, 4) < 0) {
611        fprintf(stderr, "listen(control socket): %s\n", strerror(errno));
612        exit(1);
613    }
614
615    struct sockaddr addr;
616    socklen_t alen = sizeof(addr);
617    int fd = accept(s, &addr, &alen);
618    if (fd < 0) {
619        fprintf(stderr, "accept(control socket): %s\n", strerror(errno));
620        exit(1);
621    }
622
623    fflush(redirect);
624    dup2(fd, fileno(redirect));
625    close(fd);
626}
627
628/* redirect output to a file */
629void redirect_to_file(FILE *redirect, char *path) {
630    char *chp = path;
631
632    /* skip initial slash */
633    if (chp[0] == '/')
634        chp++;
635
636    /* create leading directories, if necessary */
637    while (chp && chp[0]) {
638        chp = strchr(chp, '/');
639        if (chp) {
640            *chp = 0;
641            mkdir(path, 0770);  /* drwxrwx--- */
642            *chp++ = '/';
643        }
644    }
645
646    int fd = TEMP_FAILURE_RETRY(open(path, O_WRONLY | O_CREAT | O_TRUNC | O_CLOEXEC,
647                                     S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH));
648    if (fd < 0) {
649        fprintf(stderr, "%s: %s\n", path, strerror(errno));
650        exit(1);
651    }
652
653    TEMP_FAILURE_RETRY(dup2(fd, fileno(redirect)));
654    close(fd);
655}
656
657static bool should_dump_native_traces(const char* path) {
658    for (const char** p = native_processes_to_dump; *p; p++) {
659        if (!strcmp(*p, path)) {
660            return true;
661        }
662    }
663    return false;
664}
665
666/* dump Dalvik and native stack traces, return the trace file location (NULL if none) */
667const char *dump_traces() {
668    ON_DRY_RUN_RETURN(NULL);
669    const char* result = NULL;
670
671    char traces_path[PROPERTY_VALUE_MAX] = "";
672    property_get("dalvik.vm.stack-trace-file", traces_path, "");
673    if (!traces_path[0]) return NULL;
674
675    /* move the old traces.txt (if any) out of the way temporarily */
676    char anr_traces_path[PATH_MAX];
677    strlcpy(anr_traces_path, traces_path, sizeof(anr_traces_path));
678    strlcat(anr_traces_path, ".anr", sizeof(anr_traces_path));
679    if (rename(traces_path, anr_traces_path) && errno != ENOENT) {
680        fprintf(stderr, "rename(%s, %s): %s\n", traces_path, anr_traces_path, strerror(errno));
681        return NULL;  // Can't rename old traces.txt -- no permission? -- leave it alone instead
682    }
683
684    /* create a new, empty traces.txt file to receive stack dumps */
685    int fd = TEMP_FAILURE_RETRY(open(traces_path, O_CREAT | O_WRONLY | O_TRUNC | O_NOFOLLOW | O_CLOEXEC,
686                                     0666));  /* -rw-rw-rw- */
687    if (fd < 0) {
688        fprintf(stderr, "%s: %s\n", traces_path, strerror(errno));
689        return NULL;
690    }
691    int chmod_ret = fchmod(fd, 0666);
692    if (chmod_ret < 0) {
693        fprintf(stderr, "fchmod on %s failed: %s\n", traces_path, strerror(errno));
694        close(fd);
695        return NULL;
696    }
697
698    /* Variables below must be initialized before 'goto' statements */
699    int dalvik_found = 0;
700    int ifd, wfd = -1;
701
702    /* walk /proc and kill -QUIT all Dalvik processes */
703    DIR *proc = opendir("/proc");
704    if (proc == NULL) {
705        fprintf(stderr, "/proc: %s\n", strerror(errno));
706        goto error_close_fd;
707    }
708
709    /* use inotify to find when processes are done dumping */
710    ifd = inotify_init();
711    if (ifd < 0) {
712        fprintf(stderr, "inotify_init: %s\n", strerror(errno));
713        goto error_close_fd;
714    }
715
716    wfd = inotify_add_watch(ifd, traces_path, IN_CLOSE_WRITE);
717    if (wfd < 0) {
718        fprintf(stderr, "inotify_add_watch(%s): %s\n", traces_path, strerror(errno));
719        goto error_close_ifd;
720    }
721
722    struct dirent *d;
723    while ((d = readdir(proc))) {
724        int pid = atoi(d->d_name);
725        if (pid <= 0) continue;
726
727        char path[PATH_MAX];
728        char data[PATH_MAX];
729        snprintf(path, sizeof(path), "/proc/%d/exe", pid);
730        ssize_t len = readlink(path, data, sizeof(data) - 1);
731        if (len <= 0) {
732            continue;
733        }
734        data[len] = '\0';
735
736        if (!strncmp(data, "/system/bin/app_process", strlen("/system/bin/app_process"))) {
737            /* skip zygote -- it won't dump its stack anyway */
738            snprintf(path, sizeof(path), "/proc/%d/cmdline", pid);
739            int cfd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_CLOEXEC));
740            len = read(cfd, data, sizeof(data) - 1);
741            close(cfd);
742            if (len <= 0) {
743                continue;
744            }
745            data[len] = '\0';
746            if (!strncmp(data, "zygote", strlen("zygote"))) {
747                continue;
748            }
749
750            ++dalvik_found;
751            uint64_t start = nanotime();
752            if (kill(pid, SIGQUIT)) {
753                fprintf(stderr, "kill(%d, SIGQUIT): %s\n", pid, strerror(errno));
754                continue;
755            }
756
757            /* wait for the writable-close notification from inotify */
758            struct pollfd pfd = { ifd, POLLIN, 0 };
759            int ret = poll(&pfd, 1, 5000);  /* 5 sec timeout */
760            if (ret < 0) {
761                fprintf(stderr, "poll: %s\n", strerror(errno));
762            } else if (ret == 0) {
763                fprintf(stderr, "warning: timed out dumping pid %d\n", pid);
764            } else {
765                struct inotify_event ie;
766                read(ifd, &ie, sizeof(ie));
767            }
768
769            if (lseek(fd, 0, SEEK_END) < 0) {
770                fprintf(stderr, "lseek: %s\n", strerror(errno));
771            } else {
772                dprintf(fd, "[dump dalvik stack %d: %.3fs elapsed]\n",
773                        pid, (float)(nanotime() - start) / NANOS_PER_SEC);
774            }
775        } else if (should_dump_native_traces(data)) {
776            /* dump native process if appropriate */
777            if (lseek(fd, 0, SEEK_END) < 0) {
778                fprintf(stderr, "lseek: %s\n", strerror(errno));
779            } else {
780                static uint16_t timeout_failures = 0;
781                uint64_t start = nanotime();
782
783                /* If 3 backtrace dumps fail in a row, consider debuggerd dead. */
784                if (timeout_failures == 3) {
785                    dprintf(fd, "too many stack dump failures, skipping...\n");
786                } else if (dump_backtrace_to_file_timeout(pid, fd, 20) == -1) {
787                    dprintf(fd, "dumping failed, likely due to a timeout\n");
788                    timeout_failures++;
789                } else {
790                    timeout_failures = 0;
791                }
792                dprintf(fd, "[dump native stack %d: %.3fs elapsed]\n",
793                        pid, (float)(nanotime() - start) / NANOS_PER_SEC);
794            }
795        }
796    }
797
798    if (dalvik_found == 0) {
799        fprintf(stderr, "Warning: no Dalvik processes found to dump stacks\n");
800    }
801
802    static char dump_traces_path[PATH_MAX];
803    strlcpy(dump_traces_path, traces_path, sizeof(dump_traces_path));
804    strlcat(dump_traces_path, ".bugreport", sizeof(dump_traces_path));
805    if (rename(traces_path, dump_traces_path)) {
806        fprintf(stderr, "rename(%s, %s): %s\n", traces_path, dump_traces_path, strerror(errno));
807        goto error_close_ifd;
808    }
809    result = dump_traces_path;
810
811    /* replace the saved [ANR] traces.txt file */
812    rename(anr_traces_path, traces_path);
813
814error_close_ifd:
815    close(ifd);
816error_close_fd:
817    close(fd);
818    return result;
819}
820
821void dump_route_tables() {
822    ON_DRY_RUN_RETURN();
823    const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables";
824    dump_file("RT_TABLES", RT_TABLES_PATH);
825    FILE* fp = fopen(RT_TABLES_PATH, "re");
826    if (!fp) {
827        printf("*** %s: %s\n", RT_TABLES_PATH, strerror(errno));
828        return;
829    }
830    char table[16];
831    // Each line has an integer (the table number), a space, and a string (the table name). We only
832    // need the table number. It's a 32-bit unsigned number, so max 10 chars. Skip the table name.
833    // Add a fixed max limit so this doesn't go awry.
834    for (int i = 0; i < 64 && fscanf(fp, " %10s %*s", table) == 1; ++i) {
835        run_command("ROUTE TABLE IPv4", 10, "ip", "-4", "route", "show", "table", table, NULL);
836        run_command("ROUTE TABLE IPv6", 10, "ip", "-6", "route", "show", "table", table, NULL);
837    }
838    fclose(fp);
839}
840
841/* overall progress */
842int progress = 0;
843int do_update_progress = 0; // Set by dumpstate.cpp
844int weight_total = WEIGHT_TOTAL;
845
846// TODO: make this function thread safe if sections are generated in parallel.
847void update_progress(int delta) {
848    if (!do_update_progress) return;
849
850    progress += delta;
851
852    char key[PROPERTY_KEY_MAX];
853    char value[PROPERTY_VALUE_MAX];
854
855    // adjusts max on the fly
856    if (progress > weight_total) {
857        int new_total = weight_total * 1.2;
858        fprintf(stderr, "Adjusting total weight from %d to %d\n", weight_total, new_total);
859        weight_total = new_total;
860        sprintf(key, "dumpstate.%d.max", getpid());
861        sprintf(value, "%d", weight_total);
862        int status = property_set(key, value);
863        if (status) {
864            ALOGW("Could not update max weight by setting system property %s to %s: %d\n",
865                    key, value, status);
866        }
867    }
868
869    sprintf(key, "dumpstate.%d.progress", getpid());
870    sprintf(value, "%d", progress);
871
872    // stderr is ignored on normal invocations, but useful when calling /system/bin/dumpstate
873    // directly for debuggging.
874    fprintf(stderr, "Setting progress (%s): %s/%d\n", key, value, weight_total);
875
876    int status = property_set(key, value);
877    if (status) {
878        ALOGW("Could not update progress by setting system property %s to %s: %d\n",
879                key, value, status);
880    }
881}
882
883void take_screenshot(const std::string& path) {
884    const char *args[] = { "/system/bin/screencap", "-p", path.c_str(), NULL };
885    run_command_always(NULL, 10, args);
886}
887