Android OTA 升级之四：进入根文件系统
作者： 宋立新
Email：zjujoe@yahoo.com
前言
从bootloader 进入Recovery 模式后，首先也是运行Linux内核，该内核跟普通模式没有区别(减轻了BSP开发者的任务)。区别从执行文件系统开始。 Recovery 模式的细节就隐藏在其根文件系统中。
下面，我们就看看进入Recovery 根文件系统都干些啥。
init.rc
和正常启动一样，内核进入文件系统会执行/init, init 的配置文件就是 /init.rc, 前面文章讲过，这个文件来自：bootable/recovery/etc/init.rc，下面，我们看看它的内容。
1
2 on init
3 export PATH /sbin
4 export ANDROID_ROOT /system
5 export ANDROID_DATA /data
6 export EXTERNAL_STORAGE /sdcard
7
8 symlink /system/etc /etc
9
10 mkdir /sdcard
11 mkdir /system
12 mkdir /data
13 mkdir /cache
14 mount /tmp /tmp tmpfs
15
16 on boot
17
18 ifup lo
19 hostname localhost
20 domainname localdomain
21
22 class_start default
23
24
25 service recovery /sbin/recovery
26
27 service adbd /sbin/adbd recovery
28 disabled
29
30 on property:persist.service.adb.enable=1
31 start adbd
32
33 on property:persist.service.adb.enable=0
34 stop adbd
可以看到，它很非常简单：
1） 设置几个环境变量。备用。
2） 建立 etc 链接。
3） 造几个目录。备用。
4） Mount /tmp 目录为内存文件系统 tmpfs，后面会用到。
5） Trival 设置,不必关心。
6） 启动 recovery主程序。
7） 如果是eng模式（此时persist.service.adb.enable），启动adb
当然，init主程序还会装载属性配置文件 /default.prop, 它包含了很多系统属性设置，比如，ro.build.*， 等等。
很明显，这里最重要的就是recovery主程序，下面，我们分析它。
先看一段注释
Recovery.c 中，作者写了一段注释，对我们理解recovery的实现很有帮助，下面看一下：（我就不翻译了）
89 /*
90 * The recovery tool communicates with the main system through /cache files.
91 * /cache/recovery/command - INPUT - command line for tool, one arg per line
92 * /cache/recovery/log - OUTPUT - combined log file from recovery run(s)
93 * /cache/recovery/intent - OUTPUT - intent that was passed in
94 *
95 * The arguments which may be supplied in the recovery.command file:
96 * --send_intent=anystring - write the text out to recovery.intent
97 * --update_package=root:path - verify install an OTA package file
98 * --wipe_data - erase user data (and cache), then reboot
99 * --wipe_cache - wipe cache (but not user data), then reboot
100 * --set_encrypted_filesystem=on|off - enables / diasables encrypted fs
101 *
102 * After completing, we remove /cache/recovery/command and reboot.
103 * Arguments may also be supplied in the bootloader control block (BCB).
104 * These important scenarios must be safely restartable at any point:
105 *
106 * FACTORY RESET
107 * 1. user selects "factory reset"
108 * 2. main system writes "--wipe_data" to /cache/recovery/command
109 * 3. main system reboots into recovery
110 * 4. get_args() writes BCB with "boot-recovery" and "--wipe_data"
111 * -- after this, rebooting will restart the erase --
112 * 5. erase_root() reformats /data
113 * 6. erase_root() reformats /cache
114 * 7. finish_recovery() erases BCB
115 * -- after this, rebooting will restart the main system --
116 * 8. main() calls reboot() to boot main system
117 *
118 * OTA INSTALL
119 * 1. main system downloads OTA package to /cache/some-filename.zip
120 * 2. main system writes "--update_package=CACHE:some-filename.zip"
121 * 3. main system reboots into recovery
122 * 4. get_args() writes BCB with "boot-recovery" and "--update_package=..."
123 * -- after this, rebooting will attempt to reinstall the update --
124 * 5. install_package() attempts to install the update
125 * NOTE: the package install must itself be restartable from any point
126 * 6. finish_recovery() erases BCB
127 * -- after this, rebooting will (try to) restart the main system --
128 * 7. ** if install failed **
129 * 7a. prompt_and_wait() shows an error icon and waits for the user
130 * 7b; the user reboots (pulling the battery, etc) into the main system
131 * 8. main() calls maybe_install_firmware_update()
132 * ** if the update contained radio/hboot firmware **:
133 * 8a. m_i_f_u() writes BCB with "boot-recovery" and "--wipe_cache"
134 * -- after this, rebooting will reformat cache & restart main system --
135 * 8b. m_i_f_u() writes firmware image into raw cache partition
136 * 8c. m_i_f_u() writes BCB with "update-radio/hboot" and "--wipe_cache"
137 * -- after this, rebooting will attempt to reinstall firmware --
138 * 8d. bootloader tries to flash firmware
139 * 8e. bootloader writes BCB with "boot-recovery" (keeping "--wipe_cache")
140 * -- after this, rebooting will reformat cache & restart main system --
141 * 8f. erase_root() reformats /cache
142 * 8g. finish_recovery() erases BCB
143 * -- after this, rebooting will (try to) restart the main system --
144 * 9. main() calls reboot() to boot main system
145 *
146 * ENCRYPTED FILE SYSTEMS ENABLE/DISABLE
147 * 1. user selects "enable encrypted file systems"
148 * 2. main system writes "--set_encrypted_filesystem=on|off" to
149 * /cache/recovery/command
150 * 3. main system reboots into recovery
151 * 4. get_args() writes BCB with "boot-recovery" and
152 * "--set_encrypted_filesystems=on|off"
153 * -- after this, rebooting will restart the transition --
154 * 5. read_encrypted_fs_info() retrieves encrypted file systems settings from /data
155 * Settings include: property to specify the Encrypted FS istatus and
156 * FS encryption key if enabled (not yet implemented)
157 * 6. erase_root() reformats /data
158 * 7. erase_root() reformats /cache
159 * 8. restore_encrypted_fs_info() writes required encrypted file systems settings to /data
160 * Settings include: property to specify the Encrypted FS status and
161 * FS encryption key if enabled (not yet implemented)
162 * 9. finish_recovery() erases BCB
163 * -- after this, rebooting will restart the main system --
164 * 10. main() calls reboot() to boot main system
165 */
recovery 主程序
559 int
560 main(int argc, char **argv)
561 {
562 time_t start = time(NULL);
563
564 // If these fail, there's not really anywhere to complain...
565 freopen(TEMPORARY_LOG_FILE, "a", stdout); setbuf(stdout, NULL);
566 freopen(TEMPORARY_LOG_FILE, "a", stderr); setbuf(stderr, NULL);
567 fprintf(stderr, "Starting recovery on %s", ctime(&start));
568
将标准输出和标准错误输出重定位到"/tmp/recovery.log",如果是eng模式，就可以通过adb pull /tmp/recovery.log, 看到当前的log信息，这为我们提供了有效的调试手段。后面还会看到，recovery模式运行完毕后，会将其拷贝到cache分区，以便后续分析。
569 ui_init();
Recovery 使用了一个简单的基于framebuffer的ui系统，叫miniui,这里，进行了简单的初始化（主要是图形部分以及事件部分），并启动了一个 event 线程用于响应用户按键。
570 get_args(&argc, &argv);
从misc 分区以及 CACHE:recovery/command 文件中读入参数，写入到argc, argv ,并且，如果有必要，回写入misc分区。这样，如果recovery没有操作成功（比如，升级还没有结束，就拔电池），系统会一直进入recovery模式。提醒用户继续升级，直到成功。
572 int previous_runs = 0;
573 const char *send_intent = NULL;
574 const char *update_package = NULL;
575 int wipe_data = 0, wipe_cache = 0;
576
577 int arg;
578 while ((arg = getopt_long(argc, argv, "", OPTIONS, NULL)) != -1) {
579 switch (arg) {
580 case 'p': previous_runs = atoi(optarg); break;
581 case 's': send_intent = optarg; break;
582 case 'u': update_package = optarg; break;
583 case 'w': wipe_data = wipe_cache = 1; break;
584 case 'c': wipe_cache = 1; break;
585 case '?':
586 LOGE("Invalid command argument/n");
587 continue;
588 }
589 }
590
解析参数，p: previous_runs没有用到，其它含义见前面注释。
591 device_recovery_start();
这个函数没干什么。看名字，它給设备制造商提供了一个调用机会，可写入设备相关初始化代码。
592
593 fprintf(stderr, "Command:");
594 for (arg = 0; arg < argc; arg++) {
595 fprintf(stderr, " /"%s/"", argv[arg]);
596 }
597 fprintf(stderr, "/n/n");
598
打印出命令，比如，正常启动进入recovery模式，会打印：Command: "/sbin/recovery"
599 property_list(print_property, NULL);
600 fprintf(stderr, "/n");
601
打印出所有的系统属性（from default.prop）到log文件。
602 int status = INSTALL_SUCCESS;
603
604 if (update_package != NULL) {
605 status = install_package(update_package);
606 if (status != INSTALL_SUCCESS) ui_print("Installation aborted./n");
607 } else if (wipe_data) {
608 if (device_wipe_data()) status = INSTALL_ERROR;
609 if (erase_root("DATA:")) status = INSTALL_ERROR;
610 if (wipe_cache && erase_root("CACHE:")) status = INSTALL_ERROR;
611 if (status != INSTALL_SUCCESS) ui_print("Data wipe failed./n");
612 } else if (wipe_cache) {
613 if (wipe_cache && erase_root("CACHE:")) status = INSTALL_ERROR;
614 if (status != INSTALL_SUCCESS) ui_print("Cache wipe failed./n");
615 } else {
616 status = INSTALL_ERROR; // No command specified
617 }
根据用户提供参数，调用各项功能，比如，安装一个升级包，擦除cache分区, 擦除user data分区，install_package比较复杂，后面专门分析，其它都很简单。忽略。
618
619 if (status != INSTALL_SUCCESS) ui_set_background(BACKGROUND_ICON_ERROR);
622 if (status != INSTALL_SUCCESS) prompt_and_wait();
如果前面已经做了某项操作并且成功，则进入重启流程。否则，等待用户选择具体操作。
而用户可选操作为： reboot, 安装update.zip，除cache分区, 擦除user data分区,如前所述，只有安装package 比较复杂，其它简单。
623
624 // Otherwise, get ready to boot the main system...
625 finish_recovery(send_intent);
它的功能如下：
1）将前面定义的intent字符串写入（如果有的话）：CACHE:recovery/command
2）将 /tmp/recovery.log 复制到 "CACHE:recovery/log";
3）清空 misc 分区，这样重启就不会进入recovery模式
4)删除command 文件：CACHE:recovery/command;
626 ui_print("Rebooting.../n");
627 sync();
628 reboot(RB_AUTOBOOT);
629 return EXIT_SUCCESS;
630 }
重启。
下面我们分析核心函数 install_package
install_package
289 int
290 install_package(const char *root_path)
291 {
292 ui_set_background(BACKGROUND_ICON_INSTALLING);
294 ui_print("Finding update package.../n");
295 LOGI("Finding update package.../n");
296 ui_show_indeterminate_progress();
297 LOGI("Update location: %s/n", root_path);
298
更新 UI 显示
299 if (ensure_root_path_mounted(root_path) != 0) {
300 LOGE("Can't mount %s/n", root_path);
301 reset_mark_block();
302 return INSTALL_CORRUPT;
303 }
304
确保升级包所在分区已经mount,通常为 cache 分区或者 SD 分区
305 char path[PATH_MAX] = "";
306 if (translate_root_path(root_path, path, sizeof(path)) == NULL) {
307 LOGE("Bad path %s/n", root_path);
308 reset_mark_block();
309 return INSTALL_CORRUPT;
310 }
将根分区转化为具体分区信息.这些信息来自：全局数组：g_roots
313 ui_print("Opening update package.../n");
314 LOGI("Opening update package.../n");
315 LOGI("Update file path: %s/n", path);
316
317 int numKeys;
318 RSAPublicKey* loadedKeys = load_keys(PUBLIC_KEYS_FILE, &numKeys);
319 if (loadedKeys == NULL) {
320 LOGE("Failed to load keys/n");
321 reset_mark_block();
322 return INSTALL_CORRUPT;
323 }
324 LOGI("%d key(s) loaded from %s/n", numKeys, PUBLIC_KEYS_FILE);
从/res/keys中装载公钥。
326 // Give verification half the progress bar...
328 ui_print("Verifying update package.../n");
329 LOGI("Verifying update package.../n");
330 ui_show_progress(
331 VERIFICATION_PROGRESS_FRACTION,
332 VERIFICATION_PROGRESS_TIME);
333
334 int err;
335 err = verify_file(path, loadedKeys, numKeys);
336 free(loadedKeys);
337 LOGI("verify_file returned %d/n", err);
338 if (err != VERIFY_SUCCESS) {
339 LOGE("signature verification failed/n");
340 reset_mark_block();
341 return INSTALL_CORRUPT;
342 }
根据公钥验证升级包verify_file的注释说的很明白：
// Look for an RSA signature embedded in the .ZIP file comment given
// the path to the zip. Verify it matches one of the given public
// keys.
344 /* Try to open the package.
345 */
346 ZipArchive zip;
347 err = mzOpenZipArchive(path, &zip);
348 if (err != 0) {
349 LOGE("Can't open %s/n(%s)/n", path, err != -1 ? strerror(err) : "bad");
350 reset_mark_block();
351 return INSTALL_CORRUPT;
352 }
打开升级包，将相关信息存到ZuoArchive数据机构中，便于后面处理。
354 /* Verify and install the contents of the package.
355 */
356 int status = handle_update_package(path, &zip);
处理函数，我们后面继续分析。
357 mzCloseZipArchive(&zip);
358 return status;
359 }
关闭zip包，结束处理。
handle_update_package
204 static int
205 handle_update_package(const char *path, ZipArchive *zip)
206 {
207 // Update should take the rest of the progress bar.
208 ui_print("Installing update.../n");
209
210 int result = try_update_binary(path, zip);
211 register_package_root(NULL, NULL); // Unregister package root
212 return result;
213 }
它主要调用函数try_update_binary完成功能。
try_update_binary
84 // If the package contains an update binary, extract it and run it.
85 static int
86 try_update_binary(const char *path, ZipArchive *zip) {
87 const ZipEntry* binary_entry =
88 mzFindZipEntry(zip, ASSUMED_UPDATE_BINARY_NAME);
89 if (binary_entry == NULL) {
90 return INSTALL_CORRUPT;
91 }
93 char* binary = "/tmp/update_binary";
95 int fd = creat(binary, 0755);
97 LOGE("Can't make %s/n", binary);
98 return 1;
99 }
100 bool ok = mzExtractZipEntryToFile(zip, binary_entry, fd);
104 LOGE("Can't copy %s/n", ASSUMED_UPDATE_BINARY_NAME);
105 return 1;
106 }
将升级包内文件META-INF/com/google/android/update-binary 复制为/tmp/update_binary
108 int pipefd[2];
111 // When executing the update binary contained in the package, the
112 // arguments passed are:
113 //
114 // - the version number for this interface
115 //
116 // - an fd to which the program can write in order to update the
117 // progress bar. The program can write single-line commands:
118 //
119 // progress <frac> <secs>
120 // fill up the next <frac> part of of the progress bar
121 // over <secs> seconds. If <secs> is zero, use
122 // set_progress commands to manually control the
123 // progress of this segment of the bar
124 //
125 // set_progress <frac>
126 // <frac> should be between 0.0 and 1.0; sets the
127 // progress bar within the segment defined by the most
128 // recent progress command.
129 //
130 // firmware <"hboot"|"radio"> <filename>
131 // arrange to install the contents of <filename> in the
132 // given partition on reboot.
133 //
134 // (API v2: <filename> may start with "PACKAGE:" to
135 // indicate taking a file from the OTA package.)
136 //
137 // (API v3: this command no longer exists.)
138 //
139 // ui_print <string>
140 // display <string> on the screen.
141 //
142 // - the name of the package zip file.
143 //
注意看这段注释，它解释了以下代码的行为。结合代码，可知：
1） 将会创建新的进程，执行：/tmp/update_binary
2） 同时，会给该进程传入一些参数，其中最重要的就是一个管道fd，供新进程与原进程通信。
3） 新进程诞生后，原进程就变成了一个服务进程，它提供若干UI更新服务：
a) progress
b) set_progress
c) ui_print
这样，新进程就可以通过老进程的UI系统完成显示任务。而其他功能就靠它自己了。
145 char** args = malloc(sizeof(char*) * 5);
147 args[1] = EXPAND(RECOVERY_API_VERSION); // defined in Android.mk
149 sprintf(args[2], "%d", pipefd[1]);
157 fprintf(stderr, "E:Can't run %s (%s)/n", binary, strerror(errno));
159 }
163 FILE* from_child = fdopen(pipefd[0], "r");
164 while (fgets(buffer, sizeof(buffer), from_child) != NULL) {
165 char* command = strtok(buffer, " /n");
167 continue;
168 } else if (strcmp(command, "progress") == 0) {
169 char* fraction_s = strtok(NULL, " /n");
170 char* seconds_s = strtok(NULL, " /n");
172 float fraction = strtof(fraction_s, NULL);
173 int seconds = strtol(seconds_s, NULL, 10);
175 ui_show_progress(fraction * (1-VERIFICATION_PROGRESS_FRACTION),
177 } else if (strcmp(command, "set_progress") == 0) {
178 char* fraction_s = strtok(NULL, " /n");
179 float fraction = strtof(fraction_s, NULL);
180 ui_set_progress(fraction);
181 } else if (strcmp(command, "ui_print") == 0) {
182 char* str = strtok(NULL, "/n");
185 } else {
187 }
188 } else {
189 LOGE("unknown command [%s]/n", command);
190 }
191 }
196 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
197 LOGE("Error in %s/n(Status %d)/n", path, WEXITSTATUS(status));
198 return INSTALL_ERROR;
199 }
201 return INSTALL_SUCCESS;
202 }
这样，我们又回到了升级包中的文件：META-INF/com/google/android/update-binary，下回继续研究。
2623
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
