Magisk/jni/daemon/bootstages.c

755 lines
19 KiB
C
Raw Normal View History

2017-04-30 17:58:52 +00:00
/* post_fs_data.c - post-fs-data actions
*/
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
#include <dirent.h>
#include <linux/loop.h>
#include <sys/ioctl.h>
#include <sys/mount.h>
#include <sys/wait.h>
2017-05-03 18:39:53 +00:00
#include <selinux/selinux.h>
2017-04-30 17:58:52 +00:00
#include "magisk.h"
#include "utils.h"
#include "daemon.h"
#include "resetprop.h"
static char *buf, *buf2;
static struct vector module_list;
/******************
* Node structure *
******************/
// Precedence: MODULE > SKEL > DUMMY
2017-04-30 17:58:52 +00:00
#define DO_NOTHING 0x0 /* intermediate node */
#define IS_DUMMY 0x1 /* mount from mirror */
#define IS_SKEL 0x2 /* mount from skeleton */
#define IS_MODULE 0x4 /* mount from module */
2017-04-30 17:58:52 +00:00
struct node_entry {
const char *module; /* Only used when status & IS_MODULE */
2017-04-30 17:58:52 +00:00
char *name;
uint8_t type;
uint8_t status;
struct node_entry *parent;
struct vector *children;
};
#define IS_DIR(n) (n->type == DT_DIR)
#define IS_LNK(n) (n->type == DT_LNK)
#define IS_REG(n) (n->type == DT_REG)
2017-04-30 17:58:52 +00:00
/******************
* Image handling *
******************/
static char *loopsetup(const char *img) {
char device[20];
struct loop_info64 info;
int i, lfd, ffd;
memset(&info, 0, sizeof(info));
// First get an empty loop device
for (i = 0; i <= 7; ++i) {
sprintf(device, "/dev/block/loop%d", i);
lfd = xopen(device, O_RDWR);
if (ioctl(lfd, LOOP_GET_STATUS64, &info) == -1)
break;
close(lfd);
}
if (i == 8) return NULL;
ffd = xopen(img, O_RDWR);
if (ioctl(lfd, LOOP_SET_FD, ffd) == -1)
return NULL;
strcpy((char *) info.lo_file_name, img);
ioctl(lfd, LOOP_SET_STATUS64, &info);
2017-05-01 20:55:55 +00:00
close(lfd);
close(ffd);
2017-04-30 17:58:52 +00:00
return strdup(device);
}
static char *mount_image(const char *img, const char *target) {
char *device = loopsetup(img);
if (device)
xmount(device, target, "ext4", 0, NULL);
return device;
}
2017-05-03 17:13:04 +00:00
static void umount_image(const char *target, const char *device) {
xumount(target);
int fd = xopen(device, O_RDWR);
ioctl(fd, LOOP_CLR_FD);
close(fd);
}
static int get_img_size(const char *img, int *used, int *total) {
char buffer[PATH_MAX];
snprintf(buffer, sizeof(buffer), "e2fsck -n %s 2>/dev/null | tail -n 1", img);
char *const command[] = { "sh", "-c", buffer, NULL };
int pid, fd;
pid = run_command(&fd, "/system/bin/sh", command);
fdgets(buffer, sizeof(buffer), fd);
close(fd);
if (pid == -1)
return 1;
waitpid(pid, NULL, 0);
char *tok;
tok = strtok(buffer, ",");
while(tok != NULL) {
if (strstr(tok, "blocks"))
break;
tok = strtok(NULL, ",");
}
sscanf(tok, "%d/%d", used, total);
*used = *used / 256 + 1;
*total /= 256;
return 0;
}
2017-05-26 14:05:04 +00:00
#define round_size(a) ((((a) / 32) + 2) * 32)
2017-05-03 17:13:04 +00:00
static int resize_img(const char *img, int size) {
LOGI("resize %s to %dM\n", img, size);
2017-05-04 20:16:00 +00:00
snprintf(buf, PATH_MAX, "e2fsck -yf %s; resize2fs %s %dM;", img, img, size);
return system(buf);
2017-05-03 17:13:04 +00:00
}
static int merge_img(const char *source, const char *target) {
if (access(source, F_OK) == -1)
return 0;
if (access(target, F_OK) == -1) {
rename(source, target);
return 0;
}
// resize target to worst case
int s_used, s_total, t_used, t_total, n_total;
get_img_size(source, &s_used, &s_total);
get_img_size(target, &t_used, &t_total);
n_total = round_size(s_used + t_used);
if (n_total != t_total && resize_img(target, n_total))
return 1;
xmkdir("/cache/source", 0755);
xmkdir("/cache/target", 0755);
char *s_loop, *t_loop;
s_loop = mount_image(source, "/cache/source");
if (s_loop == NULL) return 1;
t_loop = mount_image(target, "/cache/target");
if (t_loop == NULL) return 1;
DIR *dir;
struct dirent *entry;
2017-05-03 18:39:53 +00:00
if (!(dir = opendir("/cache/source")))
2017-05-03 17:13:04 +00:00
return 1;
while ((entry = xreaddir(dir))) {
if (entry->d_type == DT_DIR) {
if (strcmp(entry->d_name, ".") == 0 ||
strcmp(entry->d_name, "..") == 0 ||
strcmp(entry->d_name, ".core") == 0 ||
strcmp(entry->d_name, "lost+found") == 0)
continue;
// Cleanup old module
snprintf(buf, PATH_MAX, "/cache/target/%s", entry->d_name);
if (access(buf, F_OK) == 0) {
LOGI("merge module %s\n", entry->d_name);
rm_rf(buf);
}
}
}
closedir(dir);
clone_dir("/cache/source", "/cache/target");
// Unmount all loop devices
umount_image("/cache/source", s_loop);
umount_image("/cache/target", t_loop);
rmdir("/cache/source");
rmdir("/cache/target");
free(s_loop);
free(t_loop);
unlink(source);
return 0;
}
static void trim_img(const char *img) {
int used, total, new_size;
get_img_size(img, &used, &total);
new_size = round_size(used);
2017-05-26 14:05:04 +00:00
if (new_size != total)
2017-05-03 17:13:04 +00:00
resize_img(img, new_size);
}
2017-04-30 17:58:52 +00:00
/***********
* Scripts *
***********/
void exec_common_script(const char* stage) {
DIR *dir;
struct dirent *entry;
2017-05-03 19:05:37 +00:00
snprintf(buf, PATH_MAX, "%s/%s.d", COREDIR, stage);
2017-04-30 17:58:52 +00:00
if (!(dir = opendir(buf)))
return;
while ((entry = xreaddir(dir))) {
if (entry->d_type == DT_REG) {
2017-05-04 20:16:00 +00:00
snprintf(buf2, PATH_MAX, "%s/%s", buf, entry->d_name);
if (access(buf2, X_OK) == -1)
2017-04-30 17:58:52 +00:00
continue;
LOGI("%s.d: exec [%s]\n", stage, entry->d_name);
2017-05-04 20:16:00 +00:00
char *const command[] = { "sh", buf2, NULL };
2017-04-30 17:58:52 +00:00
int pid = run_command(NULL, "/system/bin/sh", command);
if (pid != -1)
waitpid(pid, NULL, 0);
}
}
closedir(dir);
}
void exec_module_script(const char* stage) {
char *module;
vec_for_each(&module_list, module) {
2017-05-03 17:13:04 +00:00
snprintf(buf, PATH_MAX, "%s/%s/%s.sh", MOUNTPOINT, module, stage);
2017-04-30 17:58:52 +00:00
if (access(buf, F_OK) == -1)
continue;
LOGI("%s: exec [%s.sh]\n", module, stage);
char *const command[] = { "sh", buf, NULL };
int pid = run_command(NULL, "/system/bin/sh", command);
if (pid != -1)
waitpid(pid, NULL, 0);
}
}
/***************
* Magic Mount *
***************/
static char *get_full_path(struct node_entry *node) {
char buffer[PATH_MAX], temp[PATH_MAX];
// Concat the paths
struct node_entry *cur = node;
strcpy(buffer, node->name);
while (cur->parent) {
strcpy(temp, buffer);
snprintf(buffer, sizeof(buffer), "%s/%s", cur->parent->name, temp);
cur = cur->parent;
}
return strdup(buffer);
}
// Free the node
static void destroy_node(struct node_entry *node) {
free(node->name);
vec_destroy(node->children);
free(node->children);
free(node);
}
2017-05-04 20:16:00 +00:00
// Free the node and all children recursively
2017-04-30 17:58:52 +00:00
static void destroy_subtree(struct node_entry *node) {
// Never free parent, since it shall be freed by themselves
struct node_entry *e;
vec_for_each(node->children, e) {
destroy_subtree(e);
}
destroy_node(node);
2017-04-30 17:58:52 +00:00
}
// Return the child
static struct node_entry *insert_child(struct node_entry *p, struct node_entry *c) {
2017-04-30 17:58:52 +00:00
c->parent = p;
if (p->children == NULL) {
p->children = xmalloc(sizeof(struct vector));
vec_init(p->children);
}
struct node_entry *e;
vec_for_each(p->children, e) {
if (strcmp(e->name, c->name) == 0) {
// Exist duplicate
if (c->status > e->status) {
// Precedence is higher, replace with new node
c->children = e->children; // Preserve all children
free(e->name);
free(e);
vec_entry(p->children)[_] = c;
return c;
} else {
// Free the new entry, return old
destroy_node(c);
return e;
}
2017-04-30 17:58:52 +00:00
}
}
// New entry, push back
vec_push_back(p->children, c);
return c;
2017-04-30 17:58:52 +00:00
}
static void construct_tree(const char *module, struct node_entry *parent) {
2017-04-30 17:58:52 +00:00
DIR *dir;
struct dirent *entry;
struct node_entry *node;
char *parent_path = get_full_path(parent);
snprintf(buf, PATH_MAX, "%s/%s%s", MOUNTPOINT, module, parent_path);
2017-04-30 17:58:52 +00:00
2017-05-03 18:39:53 +00:00
if (!(dir = opendir(buf)))
goto cleanup;
2017-04-30 17:58:52 +00:00
while ((entry = xreaddir(dir))) {
if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0)
continue;
// Create new node
node = xcalloc(sizeof(*node), 1);
node->module = module;
node->name = strdup(entry->d_name);
node->type = entry->d_type;
snprintf(buf, PATH_MAX, "%s/%s", parent_path, node->name);
2017-04-30 17:58:52 +00:00
// Check if the entry has a correspond target
if (IS_LNK(node) || access(buf, F_OK) == -1) {
2017-04-30 17:58:52 +00:00
// Mark the parent folder as a skeleton
parent->status |= IS_SKEL;
node->status |= IS_MODULE;
} else if (IS_DIR(node)) {
// Check if marked as replace
snprintf(buf2, PATH_MAX, "%s/%s%s/.replace", MOUNTPOINT, module, buf);
if (access(buf2, F_OK) == 0) {
// Replace everything, mark as leaf
node->status |= IS_MODULE;
2017-04-30 17:58:52 +00:00
}
} else if (IS_REG(node)) {
// This is a leaf, mark as target
node->status |= IS_MODULE;
}
node = insert_child(parent, node);
if (node->status == DO_NOTHING) {
// Intermediate node, travel deeper
construct_tree(module, node);
2017-04-30 17:58:52 +00:00
}
}
closedir(dir);
cleanup:
free(parent_path);
2017-04-30 17:58:52 +00:00
}
static void clone_skeleton(struct node_entry *node) {
2017-04-30 17:58:52 +00:00
DIR *dir;
struct dirent *entry;
struct node_entry *dummy, *child;
// Clone the structure
char *full_path = get_full_path(node);
if (!(dir = opendir(full_path)))
goto cleanup;
2017-04-30 17:58:52 +00:00
while ((entry = xreaddir(dir))) {
if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0)
continue;
// Create dummy node
dummy = xcalloc(sizeof(*dummy), 1);
dummy->name = strdup(entry->d_name);
dummy->type = entry->d_type;
dummy->status |= IS_DUMMY;
insert_child(node, dummy);
2017-04-30 17:58:52 +00:00
}
2017-05-01 20:55:55 +00:00
closedir(dir);
2017-04-30 17:58:52 +00:00
snprintf(buf, PATH_MAX, "%s%s", DUMMDIR, full_path);
2017-05-04 20:16:00 +00:00
mkdir_p(buf, 0755);
clone_attr(full_path, buf);
bind_mount(buf, full_path);
2017-04-30 17:58:52 +00:00
vec_for_each(node->children, child) {
snprintf(buf, PATH_MAX, "%s%s/%s", DUMMDIR, full_path, child->name);
// Create the dummy file/directory
if (IS_DIR(child))
2017-04-30 17:58:52 +00:00
xmkdir(buf, 0755);
else if (IS_REG(child))
2017-04-30 17:58:52 +00:00
close(open_new(buf));
if (child->status & IS_MODULE) {
2017-04-30 17:58:52 +00:00
// Mount from module file to dummy file
snprintf(buf2, PATH_MAX, "%s/%s%s/%s", MOUNTPOINT, child->module, full_path, child->name);
} else if (child->status & IS_SKEL) {
2017-04-30 17:58:52 +00:00
// It's another skeleton, recursive call and end
clone_skeleton(child);
2017-04-30 17:58:52 +00:00
continue;
} else if (child->status & IS_DUMMY) {
// Mount from mirror to dummy file
snprintf(buf2, PATH_MAX, "%s%s/%s", MIRRDIR, full_path, child->name);
2017-04-30 17:58:52 +00:00
}
if (IS_LNK(child)) {
2017-04-30 17:58:52 +00:00
// Symlink special treatments
char *temp = xmalloc(PATH_MAX);
xreadlink(buf2, temp, PATH_MAX);
symlink(temp, buf);
free(temp);
2017-05-03 18:39:53 +00:00
LOGI("cplink: %s -> %s\n", buf2, buf);
2017-04-30 17:58:52 +00:00
} else {
snprintf(buf, PATH_MAX, "%s/%s", full_path, child->name);
2017-04-30 17:58:52 +00:00
bind_mount(buf2, buf);
}
}
cleanup:
free(full_path);
2017-04-30 17:58:52 +00:00
}
static void magic_mount(struct node_entry *node) {
char *real_path;
struct node_entry *child;
2017-05-03 17:13:04 +00:00
if (strcmp(node->name, "vendor") == 0 && strcmp(node->parent->name, "/system") == 0) {
snprintf(buf, PATH_MAX, "%s/%s/system/vendor", MOUNTPOINT, node->module);
snprintf(buf2, PATH_MAX, "%s/%s/vendor", MOUNTPOINT, node->module);
unlink(buf2);
symlink(buf, buf2);
2017-04-30 17:58:52 +00:00
return;
2017-05-03 17:13:04 +00:00
}
2017-04-30 17:58:52 +00:00
if (node->status) {
if (node->status & IS_MODULE) {
// The real deal, mount module item
real_path = get_full_path(node);
2017-05-03 17:13:04 +00:00
snprintf(buf, PATH_MAX, "%s/%s%s", MOUNTPOINT, node->module, real_path);
2017-04-30 17:58:52 +00:00
bind_mount(buf, real_path);
free(real_path);
} else if (node->status & IS_SKEL) {
// The node is labeled to be cloned with skeleton, lets do it
clone_skeleton(node);
2017-04-30 17:58:52 +00:00
}
// We should not see dummies, so no need to handle it here
} else {
// It's an intermediate node, travel deeper
vec_for_each(node->children, child)
magic_mount(child);
2017-04-30 17:58:52 +00:00
}
}
2017-05-03 18:39:53 +00:00
/****************
* Simple Mount *
****************/
static void simple_mount(const char *path) {
DIR *dir;
struct dirent *entry;
snprintf(buf, PATH_MAX, "%s%s", CACHEMOUNT, path);
if (!(dir = opendir(buf)))
return;
while ((entry = xreaddir(dir))) {
if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0)
continue;
// Target file path
snprintf(buf2, PATH_MAX, "%s/%s", path, entry->d_name);
// Only mount existing file
if (access(buf2, F_OK) == -1)
continue;
if (entry->d_type == DT_DIR) {
char *new_path = strdup(buf2);
simple_mount(new_path);
free(new_path);
} else if (entry->d_type == DT_REG) {
// Actual file path
snprintf(buf, PATH_MAX, "%s/%s", buf, entry->d_name);
// Clone all attributes
2017-05-04 20:16:00 +00:00
clone_attr(buf2, buf);
2017-05-03 18:39:53 +00:00
// Finally, mount the file
bind_mount(buf, buf2);
}
}
closedir(dir);
}
2017-04-30 17:58:52 +00:00
/****************
* Entry points *
****************/
static void *start_magisk_hide(void *args) {
launch_magiskhide(-1);
return NULL;
}
2017-05-03 17:13:04 +00:00
static void unblock_boot_process() {
close(open(UNBLOCKFILE, O_RDONLY | O_CREAT));
pthread_exit(NULL);
}
2017-04-30 17:58:52 +00:00
void post_fs(int client) {
2017-05-03 17:13:04 +00:00
// Error handler
err_handler = unblock_boot_process;
2017-04-30 17:58:52 +00:00
LOGI("** post-fs mode running\n");
// ack
write_int(client, 0);
close(client);
2017-05-03 17:13:04 +00:00
// Uninstall or core only mode
if (access(UNINSTALLER, F_OK) == 0 || access(DISABLEFILE, F_OK) == 0)
goto unblock;
2017-04-30 17:58:52 +00:00
// Allocate buffer
buf = xmalloc(PATH_MAX);
buf2 = xmalloc(PATH_MAX);
2017-05-03 18:39:53 +00:00
simple_mount("/system");
simple_mount("/vendor");
2017-04-30 17:58:52 +00:00
unblock:
unblock_boot_process();
}
void post_fs_data(int client) {
2017-05-03 17:13:04 +00:00
// Error handler
err_handler = unblock_boot_process;
2017-04-30 17:58:52 +00:00
// ack
write_int(client, 0);
close(client);
if (!check_data())
goto unblock;
2017-05-03 17:13:04 +00:00
LOGI("** post-fs-data mode running\n");
// uninstaller
if (access(UNINSTALLER, F_OK) == 0) {
close(open(UNBLOCKFILE, O_RDONLY | O_CREAT));
char *const command[] = { "sh", UNBLOCKFILE, NULL };
run_command(NULL, "/system/bin/sh", command);
return;
}
2017-04-30 17:58:52 +00:00
// Allocate buffer
if (buf == NULL) buf = xmalloc(PATH_MAX);
if (buf2 == NULL) buf2 = xmalloc(PATH_MAX);
2017-05-03 17:13:04 +00:00
// Cache support
if (access("/cache/data_bin", F_OK) == 0) {
rm_rf(DATABIN);
rename("/cache/data_bin", DATABIN);
system("mv /cache/stock_boot* /data"); // Lazy.... use bash glob....
}
// Merge images
if (merge_img("/cache/magisk.img", MAINIMG))
goto unblock;
if (merge_img("/data/magisk_merge.img", MAINIMG))
goto unblock;
LOGI("* Mounting " MAINIMG "\n");
2017-04-30 17:58:52 +00:00
// Mounting magisk image
2017-05-03 17:13:04 +00:00
char *magiskloop = mount_image(MAINIMG, MOUNTPOINT);
if (magiskloop == NULL)
goto unblock;
2017-04-30 17:58:52 +00:00
// Run common scripts
LOGI("* Running post-fs-data.d scripts\n");
exec_common_script("post-fs-data");
2017-05-03 17:13:04 +00:00
// Core only mode
if (access(DISABLEFILE, F_OK) == 0)
goto unblock;
2017-04-30 17:58:52 +00:00
DIR *dir;
struct dirent *entry;
char *module;
struct node_entry *sys_root, *ven_root = NULL, *child;
2017-05-03 18:39:53 +00:00
dir = xopendir(MOUNTPOINT);
2017-04-30 17:58:52 +00:00
// Create the system root entry
sys_root = xcalloc(sizeof(*sys_root), 1);
sys_root->name = strdup("/system");
int has_modules = 0;
// Travel through each modules
vec_init(&module_list);
LOGI("* Loading modules\n");
while ((entry = xreaddir(dir))) {
if (entry->d_type == DT_DIR) {
if (strcmp(entry->d_name, ".") == 0 ||
strcmp(entry->d_name, "..") == 0 ||
strcmp(entry->d_name, ".core") == 0 ||
strcmp(entry->d_name, "lost+found") == 0)
continue;
2017-05-03 17:13:04 +00:00
snprintf(buf, PATH_MAX, "%s/%s", MOUNTPOINT, entry->d_name);
// Check whether remove
snprintf(buf2, PATH_MAX, "%s/remove", buf);
if (access(buf2, F_OK) == 0) {
rm_rf(buf);
continue;
}
2017-04-30 17:58:52 +00:00
// Check whether disable
2017-05-03 17:13:04 +00:00
snprintf(buf2, PATH_MAX, "%s/disable", buf);
if (access(buf2, F_OK) == 0)
2017-04-30 17:58:52 +00:00
continue;
// Add the module to list
module = strdup(entry->d_name);
vec_push_back(&module_list, module);
// Read props
2017-05-03 17:13:04 +00:00
snprintf(buf2, PATH_MAX, "%s/system.prop", buf);
if (access(buf2, F_OK) == 0) {
2017-04-30 17:58:52 +00:00
LOGI("%s: loading [system.prop]\n", module);
2017-05-03 17:13:04 +00:00
read_prop_file(buf2, 0);
2017-04-30 17:58:52 +00:00
}
// Check whether enable auto_mount
2017-05-03 17:13:04 +00:00
snprintf(buf2, PATH_MAX, "%s/auto_mount", buf);
if (access(buf2, F_OK) == -1)
2017-04-30 17:58:52 +00:00
continue;
// Double check whether the system folder exists
2017-05-03 17:13:04 +00:00
snprintf(buf2, PATH_MAX, "%s/system", buf);
if (access(buf2, F_OK) == -1)
2017-04-30 17:58:52 +00:00
continue;
2017-05-03 17:13:04 +00:00
2017-04-30 17:58:52 +00:00
// Construct structure
has_modules = 1;
LOGI("%s: constructing magic mount structure\n", module);
construct_tree(module, sys_root);
2017-04-30 17:58:52 +00:00
}
}
closedir(dir);
2017-05-03 17:13:04 +00:00
// Trim image
umount_image(MOUNTPOINT, magiskloop);
free(magiskloop);
trim_img(MAINIMG);
// Remount them back :)
magiskloop = mount_image(MAINIMG, MOUNTPOINT);
free(magiskloop);
2017-04-30 17:58:52 +00:00
if (has_modules) {
// Mount mirrors
LOGI("* Mounting system/vendor mirrors");
int seperate_vendor = 0;
struct vector mounts;
vec_init(&mounts);
file_to_vector("/proc/mounts", &mounts);
char *line;
vec_for_each(&mounts, line) {
if (strstr(line, " /system ")) {
2017-05-03 17:13:04 +00:00
sscanf(line, "%s", buf);
snprintf(buf2, PATH_MAX, "%s/system", MIRRDIR);
xmkdir_p(buf2, 0755);
xmount(buf, buf2, "ext4", MS_RDONLY, NULL);
2017-05-03 18:39:53 +00:00
LOGI("mount: %s -> %s\n", buf, buf2);
2017-04-30 17:58:52 +00:00
continue;
}
if (strstr(line, " /vendor ")) {
seperate_vendor = 1;
2017-05-03 17:13:04 +00:00
sscanf(line, "%s", buf);
snprintf(buf2, PATH_MAX, "%s/vendor", MIRRDIR);
xmkdir_p(buf2, 0755);
xmount(buf, buf2, "ext4", MS_RDONLY, NULL);
2017-05-03 18:39:53 +00:00
LOGI("mount: %s -> %s\n", buf, buf2);
2017-04-30 17:58:52 +00:00
continue;
}
}
vec_deep_destroy(&mounts);
if (!seperate_vendor) {
2017-05-03 17:13:04 +00:00
snprintf(buf, PATH_MAX, "%s/system/vendor", MIRRDIR);
snprintf(buf2, PATH_MAX, "%s/vendor", MIRRDIR);
symlink(buf, buf2);
2017-05-03 18:39:53 +00:00
LOGI("link: %s -> %s\n", buf, buf2);
2017-04-30 17:58:52 +00:00
}
// Magic!!
magic_mount(sys_root);
// Get the vendor node if exists
vec_for_each(sys_root->children, child) {
if (strcmp(child->name, "vendor") == 0) {
ven_root = child;
free(ven_root->name);
ven_root->name = strdup("/vendor");
ven_root->parent = NULL;
break;
}
}
if (ven_root)
magic_mount(ven_root);
}
// Cleanup memory
destroy_subtree(sys_root);
// Execute module scripts
LOGI("* Running module post-fs-data scripts\n");
exec_module_script("post-fs-data");
2017-05-03 19:05:37 +00:00
// Systemless hosts
if (access(HOSTSFILE, F_OK) == 0) {
LOGI("* Enabling systemless hosts file support");
bind_mount(HOSTSFILE, "/system/etc/hosts");
}
2017-04-30 17:58:52 +00:00
// Start magiskhide if enabled
2017-05-26 14:05:04 +00:00
char *hide_prop = getprop(MAGISKHIDE_PROP);
2017-04-30 17:58:52 +00:00
if (hide_prop) {
if (strcmp(hide_prop, "1") == 0) {
pthread_t thread;
xpthread_create(&thread, NULL, start_magisk_hide, NULL);
2017-05-04 20:16:00 +00:00
pthread_detach(thread);
2017-04-30 17:58:52 +00:00
}
free(hide_prop);
}
unblock:
unblock_boot_process();
}
void late_start(int client) {
LOGI("** late_start service mode running\n");
// ack
write_int(client, 0);
close(client);
2017-05-04 20:16:00 +00:00
// Allocate buffer
if (buf == NULL) buf = xmalloc(PATH_MAX);
if (buf2 == NULL) buf2 = xmalloc(PATH_MAX);
2017-04-30 17:58:52 +00:00
// Wait till the full patch is done
pthread_join(sepol_patch, NULL);
// Run scripts after full patch, most reliable way to run scripts
LOGI("* Running service.d scripts\n");
exec_common_script("service");
LOGI("* Running module service scripts\n");
exec_module_script("service");
// Install Magisk Manager if exists
if (access(MANAGERAPK, F_OK) == 0) {
while(1) {
sleep(5);
char *const command[] = { "sh", "-c",
"CLASSPATH=/system/framework/pm.jar "
"/system/bin/app_process /system/bin "
"com.android.commands.pm.Pm install -r " MANAGERAPK, NULL };
int apk_res, pid;
pid = run_command(&apk_res, "/system/bin/sh", command);
waitpid(pid, NULL, 0);
fdgets(buf, PATH_MAX, apk_res);
close(apk_res);
// Keep trying until pm is started
if (strstr(buf, "Error:") == NULL)
break;
}
unlink(MANAGERAPK);
}
2017-04-30 17:58:52 +00:00
// All boot stage done, cleanup everything
free(buf);
free(buf2);
vec_deep_destroy(&module_list);
}