Introduce new boot flow to handle SAR 2SI

The existing method for handling legacy SAR is:
1. Mount /sbin tmpfs overlay
2. Dump all patched/new files into /sbin
3. Magic mount root dir and re-exec patched stock init

With Android 11 removing the /sbin folder, it is quite obvious that
things completely break down right in step 1.

To overcome this issue, we have to find a way to swap out the init
binary AFTER we re-exec stock init. This is where 2SI comes to rescue!

2SI normal boot procedure is:
1st stage -> Load sepolicy -> 2nd stage -> boot continue...

2SI Magisk boot procedure is:
MagiskInit 1st stage -> Stock 1st stage -> MagiskInit 2nd Stage ->
-> Stock init load sepolicy -> Stock 2nd stage -> boot continue...

As you can see, the trick is to make stock 1st stage init re-exec back
into MagiskInit so we can do our setup. This is possible by manipulating
some ramdisk files on initramfs based 2SI devices (old ass non SAR
devices AND super modern devices like Pixel 3/4), but not possible
on device that are stuck using legacy SAR (device that are not that
modern but not too old, like Pixel 1/2. Fucking Google logic!!)

This commit introduces a new way to intercept stock init re-exec flow:
ptrace init with forked tracer, monitor PTRACE_EVENT_EXEC, then swap
out the init file with bind mounts right before execv returns!

Going through this flow however will lose some necessary backup files,
so some bookkeeping has to be done by making the tracer hold these
files in memory and act as a daemon. 2nd stage MagiskInit will ack the
daemon to release these files at the correct time.

It just works™  ¯\_(ツ)_/¯

native/jni/init/init.cpp

@@ -217,16 +217,19 @@ int main(int argc, char *argv[]) {
 		// This will also mount /sys and /proc
 		load_kernel_info(&cmd);
 
-		if (cmd.force_normal_boot) {
-			init = make_unique<ABFirstStageInit>(argv, &cmd);
+		if (access("/apex", F_OK) == 0) {
+			if (cmd.force_normal_boot)
+				init = make_unique<ForcedFirstStageInit>(argv, &cmd);
+			else if (cmd.skip_initramfs)
+				init = make_unique<SARFirstStageInit>(argv, &cmd);
+			else
+				init = make_unique<FirstStageInit>(argv, &cmd);
 		} else if (cmd.skip_initramfs) {
 			init = make_unique<SARInit>(argv, &cmd);
 		} else {
 			decompress_ramdisk();
 			if (access("/sbin/recovery", F_OK) == 0 || access("/system/bin/recovery", F_OK) == 0)
 				init = make_unique<RecoveryInit>(argv, &cmd);
-			else if (access("/apex", F_OK) == 0)
-				init = make_unique<AFirstStageInit>(argv, &cmd);
 			else
 				init = make_unique<RootFSInit>(argv, &cmd);
 		}

native/jni/init/init.hpp

@@ -1,4 +1,6 @@
 #include <sys/mount.h>
+#include <sys/socket.h>
+#include <sys/un.h>
 #include <unistd.h>
 #include <stdlib.h>
 #include <vector>
@@ -66,9 +68,11 @@ public:
 class SARBase : public MagiskInit {
 protected:
 	raw_data config;
+	std::vector<raw_file> overlays;
 
-	void backup_files(const char *self_path);
+	void backup_files();
 	void patch_rootdir();
+	void mount_system_root();
 public:
 	SARBase(char *argv[], cmdline *cmd) : MagiskInit(argv, cmd) {
 		persist_dir = MIRRDIR "/persist/magisk";
@@ -84,28 +88,41 @@ public:
  * 2 Stage Init
  * *************/
 
-class ABFirstStageInit : public BaseInit {
+class ForcedFirstStageInit : public BaseInit {
 private:
 	void prepare();
 public:
-	ABFirstStageInit(char *argv[], cmdline *cmd) : BaseInit(argv, cmd) {};
+	ForcedFirstStageInit(char *argv[], cmdline *cmd) : BaseInit(argv, cmd) {};
 	void start() override {
 		prepare();
 		exec_init("/system/bin/init");
 	}
 };
 
-class AFirstStageInit : public BaseInit {
+class FirstStageInit : public BaseInit {
 private:
 	void prepare();
 public:
-	AFirstStageInit(char *argv[], cmdline *cmd) : BaseInit(argv, cmd) {};
+	FirstStageInit(char *argv[], cmdline *cmd) : BaseInit(argv, cmd) {};
 	void start() override {
 		prepare();
 		exec_init();
 	}
 };
 
+class SARFirstStageInit : public SARBase {
+private:
+	void traced_exec_init();
+protected:
+	void early_mount() override;
+public:
+	SARFirstStageInit(char *argv[], cmdline *cmd) : SARBase(argv, cmd) {};
+	void start() override {
+		early_mount();
+		traced_exec_init();
+	}
+};
+
 class SecondStageInit : public SARBase {
 protected:
 	void early_mount() override;
@@ -152,3 +169,4 @@ int dump_magisk(const char *path, mode_t mode);
 int magisk_proxy_main(int argc, char *argv[]);
 void setup_klog();
 void mount_sbin();
+socklen_t setup_sockaddr(struct sockaddr_un *sun);

native/jni/init/mount.cpp

@@ -183,27 +183,15 @@ void RootFSInit::early_mount() {
 	mount_list.emplace_back("/dev/mnt/cache");
 }
 
-void SARBase::backup_files(const char *self_path) {
+void SARBase::backup_files() {
 	if (access("/overlay.d", F_OK) == 0)
-		cp_afc("/overlay.d", "/dev/overlay.d");
+		backup_folder("/overlay.d", overlays);
 
-	full_read(self_path, self.buf, self.sz);
+	full_read("/proc/self/exe", self.buf, self.sz);
 	full_read("/.backup/.magisk", config.buf, config.sz);
 }
 
-void SARInit::early_mount() {
-	// Make dev writable
-	xmkdir("/dev", 0755);
-	xmount("tmpfs", "/dev", "tmpfs", 0, "mode=755");
-	mount_list.emplace_back("/dev");
-
-	backup_files("/init");
-
-	LOGD("Cleaning rootfs\n");
-	int root = xopen("/", O_RDONLY | O_CLOEXEC);
-	frm_rf(root, { "proc", "sys", "dev" });
-	close(root);
-
+void SARBase::mount_system_root() {
 	LOGD("Early mount system_root\n");
 	sprintf(partname, "system%s", cmd->slot);
 	strcpy(block_dev, "/dev/root");
@@ -221,6 +209,22 @@ void SARInit::early_mount() {
 	xmkdir("/system_root", 0755);
 	if (xmount("/dev/root", "/system_root", "ext4", MS_RDONLY, nullptr))
 		xmount("/dev/root", "/system_root", "erofs", MS_RDONLY, nullptr);
+}
+
+void SARInit::early_mount() {
+	// Make dev writable
+	xmkdir("/dev", 0755);
+	xmount("tmpfs", "/dev", "tmpfs", 0, "mode=755");
+	mount_list.emplace_back("/dev");
+
+	backup_files();
+
+	LOGD("Cleaning rootfs\n");
+	int root = xopen("/", O_RDONLY | O_CLOEXEC);
+	frm_rf(root, { "proc", "sys", "dev" });
+	close(root);
+
+	mount_system_root();
 	switch_root("/system_root");
 
 	mount_root(vendor);
@@ -228,15 +232,22 @@ void SARInit::early_mount() {
 	mount_root(odm);
 }
 
-void SecondStageInit::early_mount() {
-	// Early mounts should already be done by first stage init
+void SARFirstStageInit::early_mount() {
+	backup_files();
+	mount_system_root();
+	switch_root("/system_root");
+}
 
-	backup_files("/system/bin/init");
+void SecondStageInit::early_mount() {
+	backup_files();
 	rm_rf("/system");
 	rm_rf("/.backup");
 	rm_rf("/overlay.d");
 
-	switch_root("/system_root");
+	umount2("/system/bin/init", MNT_DETACH);
+
+	if (access("/system_root", F_OK) == 0)
+		switch_root("/system_root");
 }
 
 void BaseInit::cleanup() {

native/jni/init/rootdir.cpp

@@ -67,22 +67,24 @@ static void patch_init_rc(FILE *rc) {
 	fprintf(rc, magiskrc, pfd_svc, pfd_svc, ls_svc, bc_svc, bc_svc);
 }
 
-static void load_overlay_rc(int dirfd) {
+static void load_overlay_rc(const char *overlay) {
+	auto dir = open_dir(overlay);
+	if (!dir) return;
+
+	int dfd = dirfd(dir.get());
 	// Do not allow overwrite init.rc
-	unlinkat(dirfd, "init.rc", 0);
-	DIR *dir = fdopendir(dirfd);
-	for (dirent *entry; (entry = readdir(dir));) {
+	unlinkat(dfd, "init.rc", 0);
+	for (dirent *entry; (entry = readdir(dir.get()));) {
 		if (strend(entry->d_name, ".rc") == 0) {
 			LOGD("Found rc script [%s]\n", entry->d_name);
-			int rc = xopenat(dirfd, entry->d_name, O_RDONLY | O_CLOEXEC);
+			int rc = xopenat(dfd, entry->d_name, O_RDONLY | O_CLOEXEC);
 			raw_data data;
 			fd_full_read(rc, data.buf, data.sz);
 			close(rc);
 			rc_list.push_back(std::move(data));
-			unlinkat(dirfd, entry->d_name, 0);
+			unlinkat(dfd, entry->d_name, 0);
 		}
 	}
-	rewinddir(dir);
 }
 
 void RootFSInit::setup_rootfs() {
@@ -102,13 +104,10 @@ void RootFSInit::setup_rootfs() {
 	}
 
 	// Handle overlays
-	int fd = open("/overlay.d", O_RDONLY | O_CLOEXEC);
-	if (fd >= 0) {
+	if (access("/overlay.d", F_OK) == 0) {
 		LOGD("Merge overlay.d\n");
-		load_overlay_rc(fd);
-		mv_dir(fd, root);
-		close(fd);
-		rmdir("/overlay.d");
+		load_overlay_rc("/overlay.d");
+		mv_f("/overlay.d", "/");
 	}
 
 	// Patch init.rc
@@ -127,7 +126,7 @@ void RootFSInit::setup_rootfs() {
 	close(sbin);
 
 	// Dump magiskinit as magisk
-	fd = xopen("/sbin/magisk", O_WRONLY | O_CREAT, 0755);
+	int fd = xopen("/sbin/magisk", O_WRONLY | O_CREAT, 0755);
 	write(fd, self.buf, self.sz);
 	close(fd);
 }
@@ -336,21 +335,26 @@ void SARBase::patch_rootdir() {
 	patch_sepolicy(PATCHPOLICY);
 
 	// Handle overlay
-	if ((src = xopen("/dev/overlay.d", O_RDONLY | O_CLOEXEC)) >= 0) {
-		load_overlay_rc(src);
-		if (int fd = xopen("/dev/overlay.d/sbin", O_RDONLY | O_CLOEXEC); fd >= 0) {
-			dest = xopen("/sbin", O_RDONLY | O_CLOEXEC);
-			clone_dir(fd, dest);
-			close(fd);
-			close(dest);
-			xmkdir(ROOTOVL "/sbin", 0);  // Prevent copying
-		}
-		dest = xopen(ROOTOVL, O_RDONLY | O_CLOEXEC);
-		clone_dir(src, dest, false);
-		rmdir(ROOTOVL "/sbin");
-		close(src);
-		close(dest);
-		rm_rf("/dev/overlay.d");
+	struct sockaddr_un sun{};
+	socklen_t len = setup_sockaddr(&sun);
+	int socketfd = xsocket(AF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0);
+	if (connect(socketfd, (struct sockaddr*) &sun, len) == 0) {
+		LOGD("ACK init tracer to write backup files\n");
+		int ack;
+		// Wait for init tracer finish copying files
+		read(socketfd, &ack, sizeof(ack));
+	} else {
+		LOGD("Restore backup files locally\n");
+		restore_folder(ROOTOVL, overlays);
+		overlays.clear();
+	}
+	close(socketfd);
+	if (access(ROOTOVL "/sbin", F_OK) == 0) {
+		file_attr a;
+		getattr("/sbin", &a);
+		cp_afc(ROOTOVL "/sbin", "/sbin");
+		rm_rf(ROOTOVL "/sbin");
+		setattr("/sbin", &a);
 	}
 
 	// Patch init.rc
@@ -366,90 +370,6 @@ void SARBase::patch_rootdir() {
 	close(dest);
 }
 
-static void patch_fstab(const string &fstab) {
-	string patched = fstab + ".p";
-	FILE *fp = xfopen(patched.data(), "we");
-	file_readline(fstab.data(), [=](string_view l) -> bool {
-		if (l[0] == '#' || l.length() == 1)
-			return true;
-		char *line = (char *) l.data();
-		int src0, src1, mnt0, mnt1, type0, type1, opt0, opt1, flag0, flag1;
-		sscanf(line, "%n%*s%n %n%*s%n %n%*s%n %n%*s%n %n%*s%n",
-			   &src0, &src1, &mnt0, &mnt1, &type0, &type1, &opt0, &opt1, &flag0, &flag1);
-		const char *src, *mnt, *type, *opt, *flag;
-		src = &line[src0];
-		line[src1] = '\0';
-		mnt = &line[mnt0];
-		line[mnt1] = '\0';
-		type = &line[type0];
-		line[type1] = '\0';
-		opt = &line[opt0];
-		line[opt1] = '\0';
-		flag = &line[flag0];
-		line[flag1] = '\0';
-
-		// Redirect system to system_root
-		if (mnt == "/system"sv)
-			mnt = "/system_root";
-
-		fprintf(fp, "%s %s %s %s %s\n", src, mnt, type, opt, flag);
-		return true;
-	});
-	fclose(fp);
-
-	// Replace old fstab
-	clone_attr(fstab.data(), patched.data());
-	rename(patched.data(), fstab.data());
-}
-
-#define FSR "/first_stage_ramdisk"
-
-void ABFirstStageInit::prepare() {
-	// It is actually possible to NOT have FSR, create it just in case
-	xmkdir(FSR, 0755);
-
-	if (auto dir = xopen_dir(FSR); dir) {
-		string fstab(FSR "/");
-		for (dirent *de; (de = xreaddir(dir.get()));) {
-			if (strstr(de->d_name, "fstab")) {
-				fstab += de->d_name;
-				break;
-			}
-		}
-		if (fstab.length() == sizeof(FSR))
-			return;
-
-		patch_fstab(fstab);
-	} else {
-		return;
-	}
-
-	// Move stuffs for next stage
-	xmkdir(FSR "/system", 0755);
-	xmkdir(FSR "/system/bin", 0755);
-	rename("/init", FSR "/system/bin/init");
-	symlink("/system/bin/init", FSR "/init");
-	xmkdir(FSR "/.backup", 0);
-	rename("/.backup/.magisk", FSR "/.backup/.magisk");
-	rename("/overlay.d", FSR "/overlay.d");
-}
-
-void AFirstStageInit::prepare() {
-	auto dir = xopen_dir("/");
-	for (dirent *de; (de = xreaddir(dir.get()));) {
-		if (strstr(de->d_name, "fstab")) {
-			patch_fstab(de->d_name);
-			break;
-		}
-	}
-
-	// Move stuffs for next stage
-	xmkdir("/system", 0755);
-	xmkdir("/system/bin", 0755);
-	rename("/init", "/system/bin/init");
-	rename("/.backup/init", "/init");
-}
-
 int magisk_proxy_main(int argc, char *argv[]) {
 	setup_klog();
 

native/jni/init/twostage.cpp

@@ -0,0 +1,182 @@
+#include <sys/ptrace.h>
+#include <sys/wait.h>
+
+#include <utils.hpp>
+#include <logging.hpp>
+
+#include "init.hpp"
+
+using namespace std;
+
+static void patch_fstab(const string &fstab) {
+	string patched = fstab + ".p";
+	FILE *fp = xfopen(patched.data(), "we");
+	file_readline(fstab.data(), [=](string_view l) -> bool {
+		if (l[0] == '#' || l.length() == 1)
+			return true;
+		char *line = (char *) l.data();
+		int src0, src1, mnt0, mnt1, type0, type1, opt0, opt1, flag0, flag1;
+		sscanf(line, "%n%*s%n %n%*s%n %n%*s%n %n%*s%n %n%*s%n",
+			   &src0, &src1, &mnt0, &mnt1, &type0, &type1, &opt0, &opt1, &flag0, &flag1);
+		const char *src, *mnt, *type, *opt, *flag;
+		src = &line[src0];
+		line[src1] = '\0';
+		mnt = &line[mnt0];
+		line[mnt1] = '\0';
+		type = &line[type0];
+		line[type1] = '\0';
+		opt = &line[opt0];
+		line[opt1] = '\0';
+		flag = &line[flag0];
+		line[flag1] = '\0';
+
+		// Redirect system to system_root
+		if (mnt == "/system"sv)
+			mnt = "/system_root";
+
+		fprintf(fp, "%s %s %s %s %s\n", src, mnt, type, opt, flag);
+		return true;
+	});
+	fclose(fp);
+
+	// Replace old fstab
+	clone_attr(fstab.data(), patched.data());
+	rename(patched.data(), fstab.data());
+}
+
+#define FSR "/first_stage_ramdisk"
+
+void ForcedFirstStageInit::prepare() {
+	// It is actually possible to NOT have FSR, create it just in case
+	xmkdir(FSR, 0755);
+
+	if (auto dir = xopen_dir(FSR); dir) {
+		string fstab(FSR "/");
+		for (dirent *de; (de = xreaddir(dir.get()));) {
+			if (strstr(de->d_name, "fstab")) {
+				fstab += de->d_name;
+				break;
+			}
+		}
+		if (fstab.length() == sizeof(FSR))
+			return;
+
+		patch_fstab(fstab);
+	} else {
+		return;
+	}
+
+	// Move stuffs for next stage
+	xmkdir(FSR "/system", 0755);
+	xmkdir(FSR "/system/bin", 0755);
+	rename("/init", FSR "/system/bin/init");
+	symlink("/system/bin/init", FSR "/init");
+	xmkdir(FSR "/.backup", 0);
+	rename("/.backup/.magisk", FSR "/.backup/.magisk");
+	rename("/overlay.d", FSR "/overlay.d");
+}
+
+void FirstStageInit::prepare() {
+	auto dir = xopen_dir("/");
+	for (dirent *de; (de = xreaddir(dir.get()));) {
+		if (strstr(de->d_name, "fstab")) {
+			patch_fstab(de->d_name);
+			break;
+		}
+	}
+
+	// Move stuffs for next stage
+	xmkdir("/system", 0755);
+	xmkdir("/system/bin", 0755);
+	rename("/init", "/system/bin/init");
+	rename("/.backup/init", "/init");
+}
+
+static inline long xptrace(int request, pid_t pid, void *addr, void *data) {
+	long ret = ptrace(request, pid, addr, data);
+	if (ret < 0)
+		PLOGE("ptrace %d", pid);
+	return ret;
+}
+
+static inline long xptrace(int request, pid_t pid, void *addr = nullptr, intptr_t data = 0) {
+	return xptrace(request, pid, addr, reinterpret_cast<void *>(data));
+}
+
+#define INIT_SOCKET "MAGISKINIT"
+
+socklen_t setup_sockaddr(struct sockaddr_un *sun) {
+	sun->sun_family = AF_LOCAL;
+	strcpy(sun->sun_path + 1, INIT_SOCKET);
+	return sizeof(sa_family_t) + sizeof(INIT_SOCKET);
+}
+
+void SARFirstStageInit::traced_exec_init() {
+	int pid = getpid();
+
+	// Block SIGUSR1
+	sigset_t block, old;
+	sigemptyset(&block);
+	sigaddset(&block, SIGUSR1);
+	sigprocmask(SIG_BLOCK, &block, &old);
+
+	if (int child = xfork(); child) {
+		LOGD("init tracer [%d]\n", child);
+		// Wait for children to attach
+		int sig;
+		sigwait(&block, &sig);
+
+		// Restore sigmask
+		sigprocmask(SIG_BLOCK, &old, nullptr);
+
+		// Re-exec init
+		exec_init();
+	} else {
+		// Close all file descriptors and stop logging
+		no_logging();
+		for (int i = 0; i < 20; ++i)
+			close(i);
+
+		// Attach to parent to trace exec
+		xptrace(PTRACE_ATTACH, pid);
+		waitpid(pid, nullptr, __WALL | __WNOTHREAD);
+		xptrace(PTRACE_SETOPTIONS, pid, nullptr, PTRACE_O_TRACEEXEC);
+		xptrace(PTRACE_CONT, pid, 0, SIGUSR1);
+
+		// Wait for execve
+		waitpid(pid, nullptr, __WALL | __WNOTHREAD);
+
+		// Swap out init with bind mount
+		xmount("tmpfs", "/dev", "tmpfs", 0, "mode=755");
+		int init = xopen("/dev/magisk", O_CREAT | O_WRONLY, 0750);
+		write(init, self.buf, self.sz);
+		close(init);
+		xmount("/dev/magisk", "/init", nullptr, MS_BIND, nullptr);
+		xumount2("/dev", MNT_DETACH);
+
+		xptrace(PTRACE_DETACH, pid);
+
+		// Start daemon for 2nd stage preparation
+		struct sockaddr_un sun{};
+		auto len = setup_sockaddr(&sun);
+		int sockfd = xsocket(AF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0);
+		xbind(sockfd, (struct sockaddr*) &sun, len);
+		xlisten(sockfd, 1);
+
+		// Wait for second stage ack
+		int client = xaccept4(sockfd, nullptr, nullptr, SOCK_CLOEXEC);
+
+		// Write backup files
+		int cfg = xopen(MAGISKTMP "/config", O_WRONLY | O_CREAT, 0000);
+		xwrite(cfg, config.buf, config.sz);
+		close(cfg);
+		restore_folder(ROOTOVL, overlays);
+
+		// Ack and bail out!
+		write(sockfd, &sockfd, sizeof(sockfd));
+		close(client);
+		close(sockfd);
+
+		exit(0);
+	}
+}