Use fsck.f2fs -a instead of -f for faster boot

and run fsck with -f on clean shutdown instead.

With -f, fsck.f2fs always performs a full scan of the /data
partition regardless of whether the partition is clean or not.
The full scan takes more than 2 seconds on volantis-userdebug
and delays the OS boot.

With -a, the command does almost nothing when the partition
is clean and finishes within 20-30ms on volantis-userdebug.
When the partition has an error or its check point has
CP_FSCK_FLAG (aka "need_fsck"), the command does exactly the
same full scan as -f to fix it.

Bug: 21853106
Change-Id: I126263caf34c0f5bb8f5e6794454d4e72526ce38

5 files changed, 236 insertions, 38 deletions

diff --git a/fs_mgr/fs_mgr.c b/fs_mgr/fs_mgr.c
index 6e20560..e78b581 100644
--- a/fs_mgr/fs_mgr.c
+++ b/fs_mgr/fs_mgr.c
@@ -163,10 +163,10 @@ static void check_fs(char *blk_device, char *fs_type, char *target)
     } else if (!strcmp(fs_type, "f2fs")) {
             char *f2fs_fsck_argv[] = {
                     F2FS_FSCK_BIN,
-                    "-f",
+                    "-a",
                     blk_device
             };
-        INFO("Running %s -f %s\n", F2FS_FSCK_BIN, blk_device);
+        INFO("Running %s -a %s\n", F2FS_FSCK_BIN, blk_device);
 
         ret = android_fork_execvp_ext(ARRAY_SIZE(f2fs_fsck_argv), f2fs_fsck_argv,
                                       &status, true, LOG_KLOG | LOG_FILE,
diff --git a/include/cutils/android_reboot.h b/include/cutils/android_reboot.h
index 85e1b7e..a3861a0 100644
--- a/include/cutils/android_reboot.h
+++ b/include/cutils/android_reboot.h
@@ -17,6 +17,8 @@
 #ifndef __CUTILS_ANDROID_REBOOT_H__
 #define __CUTILS_ANDROID_REBOOT_H__
 
+#include <mntent.h>
+
 __BEGIN_DECLS
 
 /* Commands */
@@ -28,6 +30,9 @@ __BEGIN_DECLS
 #define ANDROID_RB_PROPERTY "sys.powerctl"
 
 int android_reboot(int cmd, int flags, const char *arg);
+int android_reboot_with_callback(
+    int cmd, int flags, const char *arg,
+    void (*cb_on_remount)(const struct mntent*));
 
 __END_DECLS
 
diff --git a/init/builtins.cpp b/init/builtins.cpp
index b290ce3..d0dbecc 100644
--- a/init/builtins.cpp
+++ b/init/builtins.cpp
@@ -16,7 +16,9 @@
 
 #include <errno.h>
 #include <fcntl.h>
+#include <mntent.h>
 #include <net/if.h>
+#include <signal.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -38,6 +40,7 @@
 #include <base/stringprintf.h>
 #include <cutils/partition_utils.h>
 #include <cutils/android_reboot.h>
+#include <logwrap/logwrap.h>
 #include <private/android_filesystem_config.h>
 
 #include "init.h"
@@ -49,6 +52,8 @@
 #include "log.h"
 
 #define chmod DO_NOT_USE_CHMOD_USE_FCHMODAT_SYMLINK_NOFOLLOW
+#define UNMOUNT_CHECK_MS 5000
+#define UNMOUNT_CHECK_TIMES 10
 
 int add_environment(const char *name, const char *value);
 
@@ -109,6 +114,67 @@ static void service_start_if_not_disabled(struct service *svc)
     }
 }
 
+static void unmount_and_fsck(const struct mntent *entry)
+{
+    if (strcmp(entry->mnt_type, "f2fs") && strcmp(entry->mnt_type, "ext4"))
+        return;
+
+    /* First, lazily unmount the directory. This unmount request finishes when
+     * all processes that open a file or directory in |entry->mnt_dir| exit.
+     */
+    TEMP_FAILURE_RETRY(umount2(entry->mnt_dir, MNT_DETACH));
+
+    /* Next, kill all processes except init, kthreadd, and kthreadd's
+     * children to finish the lazy unmount. Killing all processes here is okay
+     * because this callback function is only called right before reboot().
+     * It might be cleaner to selectively kill processes that actually use
+     * |entry->mnt_dir| rather than killing all, probably by reusing a function
+     * like killProcessesWithOpenFiles() in vold/, but the selinux policy does
+     * not allow init to scan /proc/<pid> files which the utility function
+     * heavily relies on. The policy does not allow the process to execute
+     * killall/pkill binaries either. Note that some processes might
+     * automatically restart after kill(), but that is not really a problem
+     * because |entry->mnt_dir| is no longer visible to such new processes.
+     */
+    service_for_each(service_stop);
+    TEMP_FAILURE_RETRY(kill(-1, SIGKILL));
+
+    int count = 0;
+    while (count++ < UNMOUNT_CHECK_TIMES) {
+        int fd = TEMP_FAILURE_RETRY(open(entry->mnt_fsname, O_RDONLY | O_EXCL));
+        if (fd >= 0) {
+            /* |entry->mnt_dir| has sucessfully been unmounted. */
+            close(fd);
+            break;
+        } else if (errno == EBUSY) {
+            /* Some processes using |entry->mnt_dir| are still alive. Wait for a
+             * while then retry.
+             */
+            TEMP_FAILURE_RETRY(
+                usleep(UNMOUNT_CHECK_MS * 1000 / UNMOUNT_CHECK_TIMES));
+            continue;
+        } else {
+            /* Cannot open the device. Give up. */
+            return;
+        }
+    }
+
+    int st;
+    if (!strcmp(entry->mnt_type, "f2fs")) {
+        const char *f2fs_argv[] = {
+            "/system/bin/fsck.f2fs", "-f", entry->mnt_fsname,
+        };
+        android_fork_execvp_ext(ARRAY_SIZE(f2fs_argv), (char **)f2fs_argv,
+                                &st, true, LOG_KLOG, true, NULL);
+    } else if (!strcmp(entry->mnt_type, "ext4")) {
+        const char *ext4_argv[] = {
+            "/system/bin/e2fsck", "-f", "-y", entry->mnt_fsname,
+        };
+        android_fork_execvp_ext(ARRAY_SIZE(ext4_argv), (char **)ext4_argv,
+                                &st, true, LOG_KLOG, true, NULL);
+    }
+}
+
 int do_class_start(int nargs, char **args)
 {
     char prop[PROP_NAME_MAX];
@@ -577,6 +643,7 @@ int do_powerctl(int nargs, char **args)
     int len = 0;
     int cmd = 0;
     const char *reboot_target;
+    void (*callback_on_ro_remount)(const struct mntent*) = NULL;
 
     res = expand_props(command, args[1], sizeof(command));
     if (res) {
@@ -590,6 +657,7 @@ int do_powerctl(int nargs, char **args)
         }
         cmd = ANDROID_RB_POWEROFF;
         len = 8;
+        callback_on_ro_remount = unmount_and_fsck;
     } else if (strncmp(command, "reboot", 6) == 0) {
         cmd = ANDROID_RB_RESTART2;
         len = 6;
@@ -614,7 +682,8 @@ int do_powerctl(int nargs, char **args)
         return -EINVAL;
     }
 
-    return android_reboot(cmd, 0, reboot_target);
+    return android_reboot_with_callback(cmd, 0, reboot_target,
+                                        callback_on_ro_remount);
 }
 
 int do_trigger(int nargs, char **args)
diff --git a/libcutils/android_reboot.c b/libcutils/android_reboot.c
index 6ae23c1..af7e189 100644
--- a/libcutils/android_reboot.c
+++ b/libcutils/android_reboot.c
@@ -14,43 +14,108 @@
  * limitations under the License.
  */
 
-#include <unistd.h>
-#include <sys/reboot.h>
-#include <sys/syscall.h>
-#include <sys/types.h>
-#include <sys/stat.h>
+#include <errno.h>
 #include <fcntl.h>
 #include <mntent.h>
+#include <stdbool.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
+#include <sys/cdefs.h>
+#include <sys/mount.h>
+#include <sys/reboot.h>
+#include <sys/stat.h>
+#include <sys/syscall.h>
+#include <sys/types.h>
+#include <unistd.h>
 
 #include <cutils/android_reboot.h>
+#include <cutils/klog.h>
+#include <cutils/list.h>
 
-#define UNUSED __attribute__((unused))
+#define TAG "android_reboot"
+#define READONLY_CHECK_MS 5000
+#define READONLY_CHECK_TIMES 50
 
-/* Check to see if /proc/mounts contains any writeable filesystems
- * backed by a block device.
- * Return true if none found, else return false.
+typedef struct {
+    struct listnode list;
+    struct mntent entry;
+} mntent_list;
+
+static bool has_mount_option(const char* opts, const char* opt_to_find)
+{
+  bool ret = false;
+  char* copy = NULL;
+  char* opt;
+  char* rem;
+
+  while ((opt = strtok_r(copy ? NULL : (copy = strdup(opts)), ",", &rem))) {
+      if (!strcmp(opt, opt_to_find)) {
+          ret = true;
+          break;
+      }
+  }
+
+  free(copy);
+  return ret;
+}
+
+static bool is_block_device(const char* fsname)
+{
+    return !strncmp(fsname, "/dev/block", 10);
+}
+
+/* Find all read+write block devices in /proc/mounts and add them to
+ * |rw_entries|.
  */
-static int remount_ro_done(void)
+static void find_rw(struct listnode* rw_entries)
 {
     FILE* fp;
     struct mntent* mentry;
-    int found_rw_fs = 0;
 
     if ((fp = setmntent("/proc/mounts", "r")) == NULL) {
-        /* If we can't read /proc/mounts, just give up. */
-        return 1;
+        KLOG_WARNING(TAG, "Failed to open /proc/mounts.\n");
+        return;
     }
     while ((mentry = getmntent(fp)) != NULL) {
-        if (!strncmp(mentry->mnt_fsname, "/dev/block", 10) && strstr(mentry->mnt_opts, "rw,")) {
-            found_rw_fs = 1;
-            break;
+        if (is_block_device(mentry->mnt_fsname) &&
+            has_mount_option(mentry->mnt_opts, "rw")) {
+            mntent_list* item = (mntent_list*)calloc(1, sizeof(mntent_list));
+            item->entry = *mentry;
+            item->entry.mnt_fsname = strdup(mentry->mnt_fsname);
+            item->entry.mnt_dir = strdup(mentry->mnt_dir);
+            item->entry.mnt_type = strdup(mentry->mnt_type);
+            item->entry.mnt_opts = strdup(mentry->mnt_opts);
+            list_add_tail(rw_entries, &item->list);
         }
     }
     endmntent(fp);
+}
+
+static void free_entries(struct listnode* entries)
+{
+    struct listnode* node;
+    struct listnode* n;
+    list_for_each_safe(node, n, entries) {
+        mntent_list* item = node_to_item(node, mntent_list, list);
+        free(item->entry.mnt_fsname);
+        free(item->entry.mnt_dir);
+        free(item->entry.mnt_type);
+        free(item->entry.mnt_opts);
+        free(item);
+    }
+}
 
-    return !found_rw_fs;
+static mntent_list* find_item(struct listnode* rw_entries, const char* fsname_to_find)
+{
+    struct listnode* node;
+    list_for_each(node, rw_entries) {
+        mntent_list* item = node_to_item(node, mntent_list, list);
+        if (!strcmp(item->entry.mnt_fsname, fsname_to_find)) {
+            return item;
+        }
+    }
+    return NULL;
 }
 
 /* Remounting filesystems read-only is difficult when there are files
@@ -64,38 +129,92 @@ static int remount_ro_done(void)
  * repeatedly until there are no more writable filesystems mounted on
  * block devices.
  */
-static void remount_ro(void)
+static void remount_ro(void (*cb_on_remount)(const struct mntent*))
 {
-    int fd, cnt = 0;
+    int fd, cnt;
+    FILE* fp;
+    struct mntent* mentry;
+    struct listnode* node;
+
+    list_declare(rw_entries);
+    list_declare(ro_entries);
+
+    sync();
+    find_rw(&rw_entries);
 
     /* Trigger the remount of the filesystems as read-only,
      * which also marks them clean.
      */
-    fd = open("/proc/sysrq-trigger", O_WRONLY);
+    fd = TEMP_FAILURE_RETRY(open("/proc/sysrq-trigger", O_WRONLY));
     if (fd < 0) {
-        return;
+        KLOG_WARNING(TAG, "Failed to open sysrq-trigger.\n");
+        /* TODO: Try to remount each rw parition manually in readonly mode.
+         * This may succeed if no process is using the partition.
+         */
+        goto out;
+    }
+    if (TEMP_FAILURE_RETRY(write(fd, "u", 1)) != 1) {
+        close(fd);
+        KLOG_WARNING(TAG, "Failed to write to sysrq-trigger.\n");
+        /* TODO: The same. Manually remount the paritions. */
+        goto out;
     }
-    write(fd, "u", 1);
     close(fd);
 
-
     /* Now poll /proc/mounts till it's done */
-    while (!remount_ro_done() && (cnt < 50)) {
-        usleep(100000);
+    cnt = 0;
+    while (cnt < READONLY_CHECK_TIMES) {
+        if ((fp = setmntent("/proc/mounts", "r")) == NULL) {
+            /* If we can't read /proc/mounts, just give up. */
+            KLOG_WARNING(TAG, "Failed to open /proc/mounts.\n");
+            goto out;
+        }
+        while ((mentry = getmntent(fp)) != NULL) {
+            if (!is_block_device(mentry->mnt_fsname) ||
+                !has_mount_option(mentry->mnt_opts, "ro")) {
+                continue;
+            }
+            mntent_list* item = find_item(&rw_entries, mentry->mnt_fsname);
+            if (item) {
+                /* |item| has now been ro remounted. */
+                list_remove(&item->list);
+                list_add_tail(&ro_entries, &item->list);
+            }
+        }
+        endmntent(fp);
+        if (list_empty(&rw_entries)) {
+            /* All rw block devices are now readonly. */
+            break;
+        }
+        TEMP_FAILURE_RETRY(
+            usleep(READONLY_CHECK_MS * 1000 / READONLY_CHECK_TIMES));
         cnt++;
     }
 
-    return;
-}
+    list_for_each(node, &rw_entries) {
+        mntent_list* item = node_to_item(node, mntent_list, list);
+        KLOG_WARNING(TAG, "Failed to remount %s in readonly mode.\n",
+                     item->entry.mnt_fsname);
+    }
+
+    if (cb_on_remount) {
+        list_for_each(node, &ro_entries) {
+            mntent_list* item = node_to_item(node, mntent_list, list);
+            cb_on_remount(&item->entry);
+        }
+    }
 
+out:
+    free_entries(&rw_entries);
+    free_entries(&ro_entries);
+}
 
-int android_reboot(int cmd, int flags UNUSED, const char *arg)
+int android_reboot_with_callback(
+    int cmd, int flags __unused, const char *arg,
+    void (*cb_on_remount)(const struct mntent*))
 {
     int ret;
-
-    sync();
-    remount_ro();
-
+    remount_ro(cb_on_remount);
     switch (cmd) {
         case ANDROID_RB_RESTART:
             ret = reboot(RB_AUTOBOOT);
@@ -117,3 +236,7 @@ int android_reboot(int cmd, int flags UNUSED, const char *arg)
     return ret;
 }
 
+int android_reboot(int cmd, int flags, const char *arg)
+{
+    return android_reboot_with_callback(cmd, flags, arg, NULL);
+}
diff --git a/logwrapper/logwrap.c b/logwrapper/logwrap.c
index 83576fb..44455d1 100644
--- a/logwrapper/logwrap.c
+++ b/logwrapper/logwrap.c
@@ -355,7 +355,8 @@ static int parent(const char *tag, int parent_read, pid_t pid,
         }
 
         if (poll_fds[0].revents & POLLIN) {
-            sz = read(parent_read, &buffer[b], sizeof(buffer) - 1 - b);
+            sz = TEMP_FAILURE_RETRY(
+                read(parent_read, &buffer[b], sizeof(buffer) - 1 - b));
 
             sz += b;
             // Log one line at a time
@@ -490,7 +491,7 @@ int android_fork_execvp_ext(int argc, char* argv[], int *status, bool ignore_int
     }
 
     /* Use ptty instead of socketpair so that STDOUT is not buffered */
-    parent_ptty = open("/dev/ptmx", O_RDWR);
+    parent_ptty = TEMP_FAILURE_RETRY(open("/dev/ptmx", O_RDWR));
     if (parent_ptty < 0) {
         ERROR("Cannot create parent ptty\n");
         rc = -1;
@@ -505,7 +506,7 @@ int android_fork_execvp_ext(int argc, char* argv[], int *status, bool ignore_int
         goto err_ptty;
     }
 
-    child_ptty = open(child_devname, O_RDWR);
+    child_ptty = TEMP_FAILURE_RETRY(open(child_devname, O_RDWR));
     if (child_ptty < 0) {
         ERROR("Cannot open child_ptty\n");
         rc = -1;