criu/0006-rseq-initial-support.patch

From e444c089ebfb03fb2b6d69a40322d31ab33c0597 Mon Sep 17 00:00:00 2001
From: bb-cat <ningyu9@huawei.com>
Date: Wed, 2 Mar 2022 14:52:35 +0800
Subject: [PATCH 06/16] rseq: initial support TODO: 1. properly handle case
 when the kernel has rseq() support but has no
 ptrace(PTRACE_GET_RSEQ_CONFIGURATION) support and user processes haven't used
 rseq(). 2. properly handle "transient" states, when CRIU comes during rseq
 was executed. We need test for this case with some "heavy" rseq + we need to
 properly handle RSEQ_CS_* flags.

Fixes: #1696

Reported-by: Radostin Stoyanov <radostin@redhat.com>
Suggested-by: Florian Weimer <fweimer@redhat.com>
Signed-off-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com>
---
 compel/include/uapi/ptrace.h |  16 +--
 criu/cr-dump.c               |  99 ++++++++++++++++
 criu/cr-restore.c            |  17 +++
 criu/include/linux/rseq.h    | 144 +++++++++++++++++++++++
 criu/include/parasite.h      |   7 ++
 criu/include/restorer.h      |   7 ++
 criu/kerndat.c               |   2 +-
 criu/parasite-syscall.c      |  11 ++
 criu/pie/parasite.c          |  99 ++++++++++++++++
 criu/pie/restorer.c          |  24 ++++
 images/Makefile              |   1 +
 images/core.proto            |   2 +
 images/rseq.proto            |   9 ++
 13 files changed, 429 insertions(+), 9 deletions(-)
 create mode 100644 criu/include/linux/rseq.h
 create mode 100644 images/rseq.proto

diff --git a/compel/include/uapi/ptrace.h b/compel/include/uapi/ptrace.h
index bfe28c7..d807a92 100644
--- a/compel/include/uapi/ptrace.h
+++ b/compel/include/uapi/ptrace.h
@@ -66,14 +66,14 @@ typedef struct {
 } seccomp_metadata_t;
 
 #ifndef PTRACE_GET_RSEQ_CONFIGURATION
-#define PTRACE_GET_RSEQ_CONFIGURATION	0x420f
-
-struct ptrace_rseq_configuration {
-	__u64 rseq_abi_pointer;
-	__u32 rseq_abi_size;
-	__u32 signature;
-	__u32 flags;
-	__u32 pad;
+#define PTRACE_GET_RSEQ_CONFIGURATION 0x420f
+
+struct __ptrace_rseq_configuration {
+	uint64_t rseq_abi_pointer;
+	uint32_t rseq_abi_size;
+	uint32_t signature;
+	uint32_t flags;
+	uint32_t pad;
 };
 #endif
 
diff --git a/criu/cr-dump.c b/criu/cr-dump.c
index f07fe6e..91dd08a 100644
--- a/criu/cr-dump.c
+++ b/criu/cr-dump.c
@@ -45,6 +45,7 @@
 #include "proc_parse.h"
 #include "parasite.h"
 #include "parasite-syscall.h"
+#include <compel/ptrace.h>
 #include "files.h"
 #include "files-reg.h"
 #include "shmem.h"
@@ -200,6 +201,25 @@ static int dump_sched_info(int pid, ThreadCoreEntry *tc)
 	return 0;
 }
 
+static int check_thread_rseq(pid_t tid, const struct parasite_check_rseq *ti_rseq, bool has_tc_rseq_entry)
+{
+	if (!kdat.has_rseq || kdat.has_ptrace_get_rseq_conf)
+		return 0;
+
+	pr_debug("%d has rseq_inited = %d\n", tid, ti_rseq->rseq_inited);
+
+	/*
+	 * We have no kdat.has_ptrace_get_rseq_conf and user
+	 * process has rseq() used, let's fail dump.
+	 */
+	if (ti_rseq->rseq_inited) {
+		pr_err("%d has rseq but kernel lacks get_rseq_conf feature\n", tid);
+		return -1;
+	}
+
+	return 0;
+}
+
 struct cr_imgset *glob_imgset;
 
 static int collect_fds(pid_t pid, struct parasite_drain_fd **dfds)
@@ -730,6 +750,17 @@ int dump_thread_core(int pid, CoreEntry *core, const struct parasite_dump_thread
 	if (!ret)
 		ret = seccomp_dump_thread(pid, tc);
 
+	/*
+	 * We are dumping rseq() in the dump_thread_rseq() function,
+	 * *before* processes gets infected (because of ptrace requests
+	 * API restriction). At this point, if the kernel lacks
+	 * kdat.has_ptrace_get_rseq_conf support we have to ensure
+	 * that dumpable processes haven't initialized rseq() or
+	 * fail dump if rseq() was used.
+	 */
+	if (!ret)
+		ret = check_thread_rseq(pid, &ti->rseq, !!tc->rseq_entry);
+
 	return ret;
 }
 
@@ -1016,6 +1047,68 @@ static int dump_task_signals(pid_t pid, struct pstree_item *item)
 	return 0;
 }
 
+static int dump_thread_rseq(pid_t tid, RseqEntry **rseqep)
+{
+	struct __ptrace_rseq_configuration rseq;
+	RseqEntry *rseqe = NULL;
+	int ret;
+
+	/*
+	 * If we are here it means that rseq() syscall is supported,
+	 * but ptrace(PTRACE_GET_RSEQ_CONFIGURATION) isn't supported,
+	 * we can just fail dump here. But this is bad idea, IMHO.
+	 *
+	 * So, we will try to detect if victim process was used rseq().
+	 * See check_rseq() and check_thread_rseq() functions.
+	 */
+	if (!kdat.has_ptrace_get_rseq_conf)
+		return 0;
+
+	ret = ptrace(PTRACE_GET_RSEQ_CONFIGURATION, tid, sizeof(rseq), &rseq);
+	if (ret != sizeof(rseq)) {
+		pr_perror("ptrace(PTRACE_GET_RSEQ_CONFIGURATION, %d) = %d", tid, ret);
+		return -1;
+	}
+
+	if (rseq.flags != 0) {
+		pr_err("something wrong with ptrace(PTRACE_GET_RSEQ_CONFIGURATION, %d) flags = 0x%x\n", tid,
+		       rseq.flags);
+		return -1;
+	}
+
+	pr_err("Dump rseq of %d: ptr = 0x%lx sign = 0x%x\n", tid, (unsigned long)rseq.rseq_abi_pointer, rseq.signature);
+
+	rseqe = xmalloc(sizeof(*rseqe));
+	if (!rseqe)
+		return -1;
+
+	rseq_entry__init(rseqe);
+
+	rseqe->rseq_abi_pointer = rseq.rseq_abi_pointer;
+	rseqe->rseq_abi_size = rseq.rseq_abi_size;
+	rseqe->signature = rseq.signature;
+
+	*rseqep = rseqe;
+
+	return 0;
+}
+
+static int dump_task_rseq(pid_t pid, struct pstree_item *item)
+{
+	int i;
+
+	/* if rseq() syscall isn't supported then nothing to dump */
+	if (!kdat.has_rseq)
+		return 0;
+
+	for (i = 0; i < item->nr_threads; i++) {
+		if (dump_thread_rseq(item->threads[i].real, &item->core[i]->thread_core->rseq_entry))
+			return -1;
+	}
+
+	return 0;
+}
+
 static struct proc_pid_stat pps_buf;
 
 static int dump_task_threads(struct parasite_ctl *parasite_ctl, const struct pstree_item *item)
@@ -1304,6 +1397,12 @@ static int dump_one_task(struct pstree_item *item, InventoryEntry *parent_ie)
 		goto err;
 	}
 
+	ret = dump_task_rseq(pid, item);
+	if (ret) {
+		pr_err("Dump %d rseq failed %d\n", pid, ret);
+		goto err;
+	}
+
 	parasite_ctl = parasite_infect_seized(pid, item, &vmas);
 	if (!parasite_ctl) {
 		pr_err("Can't infect (pid: %d) with parasite\n", pid);
diff --git a/criu/cr-restore.c b/criu/cr-restore.c
index 5b645c1..b2bd044 100644
--- a/criu/cr-restore.c
+++ b/criu/cr-restore.c
@@ -2975,6 +2975,19 @@ static int prep_sched_info(struct rst_sched_param *sp, ThreadCoreEntry *tc)
 	return 0;
 }
 
+static int prep_rseq(struct rst_rseq_param *rseq, ThreadCoreEntry *tc)
+{
+	/* compatibility with older CRIU versions */
+	if (!tc->rseq_entry)
+		return 0;
+
+	rseq->rseq_abi_pointer = tc->rseq_entry->rseq_abi_pointer;
+	rseq->rseq_abi_size = tc->rseq_entry->rseq_abi_size;
+	rseq->signature = tc->rseq_entry->signature;
+
+	return 0;
+}
+
 static rlim_t decode_rlim(rlim_t ival)
 {
 	return ival == -1 ? RLIM_INFINITY : ival;
@@ -3704,6 +3717,10 @@ static int sigreturn_restore(pid_t pid, struct task_restore_args *task_args, uns
 		thread_args[i].clear_tid_addr = CORE_THREAD_ARCH_INFO(tcore)->clear_tid_addr;
 		core_get_tls(tcore, &thread_args[i].tls);
 
+		ret = prep_rseq(&thread_args[i].rseq, tcore->thread_core);
+		if (ret)
+			goto err;
+
 		rst_reloc_creds(&thread_args[i], &creds_pos_next);
 
 		thread_args[i].futex_rla = tcore->thread_core->futex_rla;
diff --git a/criu/include/linux/rseq.h b/criu/include/linux/rseq.h
new file mode 100644
index 0000000..5c1706a
--- /dev/null
+++ b/criu/include/linux/rseq.h
@@ -0,0 +1,144 @@
+/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
+#ifndef _UAPI_LINUX_RSEQ_H
+#define _UAPI_LINUX_RSEQ_H
+
+/*
+ * linux/rseq.h
+ *
+ * Restartable sequences system call API
+ *
+ * Copyright (c) 2015-2018 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ */
+
+#include <linux/types.h>
+#include <asm/byteorder.h>
+
+enum rseq_cpu_id_state {
+	RSEQ_CPU_ID_UNINITIALIZED = -1,
+	RSEQ_CPU_ID_REGISTRATION_FAILED = -2,
+};
+
+enum rseq_flags {
+	RSEQ_FLAG_UNREGISTER = (1 << 0),
+};
+
+enum rseq_cs_flags_bit {
+	RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT = 0,
+	RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT = 1,
+	RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT = 2,
+};
+
+enum rseq_cs_flags {
+	RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT = (1U << RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT),
+	RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL = (1U << RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT),
+	RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE = (1U << RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT),
+};
+
+/*
+ * struct rseq_cs is aligned on 4 * 8 bytes to ensure it is always
+ * contained within a single cache-line. It is usually declared as
+ * link-time constant data.
+ */
+struct rseq_cs {
+	/* Version of this structure. */
+	__u32 version;
+	/* enum rseq_cs_flags */
+	__u32 flags;
+	__u64 start_ip;
+	/* Offset from start_ip. */
+	__u64 post_commit_offset;
+	__u64 abort_ip;
+} __attribute__((aligned(4 * sizeof(__u64))));
+
+/*
+ * struct rseq is aligned on 4 * 8 bytes to ensure it is always
+ * contained within a single cache-line.
+ *
+ * A single struct rseq per thread is allowed.
+ */
+struct rseq {
+	/*
+	 * Restartable sequences cpu_id_start field. Updated by the
+	 * kernel. Read by user-space with single-copy atomicity
+	 * semantics. This field should only be read by the thread which
+	 * registered this data structure. Aligned on 32-bit. Always
+	 * contains a value in the range of possible CPUs, although the
+	 * value may not be the actual current CPU (e.g. if rseq is not
+	 * initialized). This CPU number value should always be compared
+	 * against the value of the cpu_id field before performing a rseq
+	 * commit or returning a value read from a data structure indexed
+	 * using the cpu_id_start value.
+	 */
+	__u32 cpu_id_start;
+	/*
+	 * Restartable sequences cpu_id field. Updated by the kernel.
+	 * Read by user-space with single-copy atomicity semantics. This
+	 * field should only be read by the thread which registered this
+	 * data structure. Aligned on 32-bit. Values
+	 * RSEQ_CPU_ID_UNINITIALIZED and RSEQ_CPU_ID_REGISTRATION_FAILED
+	 * have a special semantic: the former means "rseq uninitialized",
+	 * and latter means "rseq initialization failed". This value is
+	 * meant to be read within rseq critical sections and compared
+	 * with the cpu_id_start value previously read, before performing
+	 * the commit instruction, or read and compared with the
+	 * cpu_id_start value before returning a value loaded from a data
+	 * structure indexed using the cpu_id_start value.
+	 */
+	__u32 cpu_id;
+	/*
+	 * Restartable sequences rseq_cs field.
+	 *
+	 * Contains NULL when no critical section is active for the current
+	 * thread, or holds a pointer to the currently active struct rseq_cs.
+	 *
+	 * Updated by user-space, which sets the address of the currently
+	 * active rseq_cs at the beginning of assembly instruction sequence
+	 * block, and set to NULL by the kernel when it restarts an assembly
+	 * instruction sequence block, as well as when the kernel detects that
+	 * it is preempting or delivering a signal outside of the range
+	 * targeted by the rseq_cs. Also needs to be set to NULL by user-space
+	 * before reclaiming memory that contains the targeted struct rseq_cs.
+	 *
+	 * Read and set by the kernel. Set by user-space with single-copy
+	 * atomicity semantics. This field should only be updated by the
+	 * thread which registered this data structure. Aligned on 64-bit.
+	 */
+	union {
+		__u64 ptr64;
+#ifdef __LP64__
+		__u64 ptr;
+#else
+		struct {
+#if (defined(__BYTE_ORDER) && (__BYTE_ORDER == __BIG_ENDIAN)) || defined(__BIG_ENDIAN)
+			__u32 padding; /* Initialized to zero. */
+			__u32 ptr32;
+#else  /* LITTLE */
+			__u32 ptr32;
+			__u32 padding; /* Initialized to zero. */
+#endif /* ENDIAN */
+		} ptr;
+#endif
+	} rseq_cs;
+
+	/*
+	 * Restartable sequences flags field.
+	 *
+	 * This field should only be updated by the thread which
+	 * registered this data structure. Read by the kernel.
+	 * Mainly used for single-stepping through rseq critical sections
+	 * with debuggers.
+	 *
+	 * - RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT
+	 *     Inhibit instruction sequence block restart on preemption
+	 *     for this thread.
+	 * - RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL
+	 *     Inhibit instruction sequence block restart on signal
+	 *     delivery for this thread.
+	 * - RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE
+	 *     Inhibit instruction sequence block restart on migration for
+	 *     this thread.
+	 */
+	__u32 flags;
+} __attribute__((aligned(4 * sizeof(__u64))));
+
+#endif /* _UAPI_LINUX_RSEQ_H */
diff --git a/criu/include/parasite.h b/criu/include/parasite.h
index 8107aa4..5fde809 100644
--- a/criu/include/parasite.h
+++ b/criu/include/parasite.h
@@ -164,10 +164,17 @@ struct parasite_dump_creds {
 	unsigned int groups[0];
 };
 
+struct parasite_check_rseq {
+	bool has_rseq;
+	bool has_ptrace_get_rseq_conf; /* no need to check if supported */
+	bool rseq_inited;
+};
+
 struct parasite_dump_thread {
 	unsigned int *tid_addr;
 	pid_t tid;
 	tls_t tls;
+	struct parasite_check_rseq rseq;
 	stack_t sas;
 	int pdeath_sig;
 	char comm[TASK_COMM_LEN];
diff --git a/criu/include/restorer.h b/criu/include/restorer.h
index c2ef8f0..c29d869 100644
--- a/criu/include/restorer.h
+++ b/criu/include/restorer.h
@@ -45,6 +45,12 @@ struct rst_sched_param {
 	int prio;
 };
 
+struct rst_rseq_param {
+	u64 rseq_abi_pointer;
+	u32 rseq_abi_size;
+	u32 signature;
+};
+
 struct restore_posix_timer {
 	struct str_posix_timer spt;
 	struct itimerspec val;
@@ -99,6 +105,7 @@ struct thread_restore_args {
 	struct task_restore_args *ta;
 
 	tls_t tls;
+	struct rst_rseq_param rseq;
 
 	siginfo_t *siginfo;
 	unsigned int siginfo_n;
diff --git a/criu/kerndat.c b/criu/kerndat.c
index 4841387..af7113a 100644
--- a/criu/kerndat.c
+++ b/criu/kerndat.c
@@ -837,7 +837,7 @@ static int kerndat_has_ptrace_get_rseq_conf(void)
 {
 	pid_t pid;
 	int len;
-	struct ptrace_rseq_configuration rseq;
+	struct __ptrace_rseq_configuration rseq;
 
 	pid = fork_and_ptrace_attach(NULL);
 	if (pid < 0)
diff --git a/criu/parasite-syscall.c b/criu/parasite-syscall.c
index 7175ade..ee4fa86 100644
--- a/criu/parasite-syscall.c
+++ b/criu/parasite-syscall.c
@@ -132,6 +132,13 @@ static int alloc_groups_copy_creds(CredsEntry *ce, struct parasite_dump_creds *c
 	return ce->groups ? 0 : -ENOMEM;
 }
 
+static void init_parasite_rseq_arg(struct parasite_check_rseq *rseq)
+{
+	rseq->has_rseq = kdat.has_rseq;
+	rseq->has_ptrace_get_rseq_conf = kdat.has_ptrace_get_rseq_conf;
+	rseq->rseq_inited = false;
+}
+
 int parasite_dump_thread_leader_seized(struct parasite_ctl *ctl, int pid, CoreEntry *core)
 {
 	ThreadCoreEntry *tc = core->thread_core;
@@ -144,6 +151,8 @@ int parasite_dump_thread_leader_seized(struct parasite_ctl *ctl, int pid, CoreEn
 	pc = args->creds;
 	pc->cap_last_cap = kdat.last_cap;
 
+	init_parasite_rseq_arg(&args->rseq);
+
 	ret = compel_rpc_call_sync(PARASITE_CMD_DUMP_THREAD, ctl);
 	if (ret < 0)
 		return ret;
@@ -197,6 +206,8 @@ int parasite_dump_thread_seized(struct parasite_thread_ctl *tctl, struct parasit
 
 	compel_arch_get_tls_thread(tctl, &args->tls);
 
+	init_parasite_rseq_arg(&args->rseq);
+
 	ret = compel_run_in_thread(tctl, PARASITE_CMD_DUMP_THREAD);
 	if (ret) {
 		pr_err("Can't init thread in parasite %d\n", pid);
diff --git a/criu/pie/parasite.c b/criu/pie/parasite.c
index bc0a33c..e49958b 100644
--- a/criu/pie/parasite.c
+++ b/criu/pie/parasite.c
@@ -8,6 +8,8 @@
 #include <sys/ioctl.h>
 #include <sys/uio.h>
 
+#include "linux/rseq.h"
+
 #include "common/config.h"
 #include "int.h"
 #include "types.h"
@@ -167,6 +169,7 @@ static int dump_posix_timers(struct parasite_dump_posix_timers_args *args)
 }
 
 static int dump_creds(struct parasite_dump_creds *args);
+static int check_rseq(struct parasite_check_rseq *rseq);
 
 static int dump_thread_common(struct parasite_dump_thread *ti)
 {
@@ -197,6 +200,12 @@ static int dump_thread_common(struct parasite_dump_thread *ti)
 		goto out;
 	}
 
+	ret = check_rseq(&ti->rseq);
+	if (ret) {
+		pr_err("Unable to check if rseq() is initialized: %d\n", ret);
+		goto out;
+	}
+
 	ret = dump_creds(ti->creds);
 out:
 	return ret;
@@ -313,6 +322,96 @@ grps_err:
 	return -1;
 }
 
+static int check_rseq(struct parasite_check_rseq *rseq)
+{
+	int ret;
+	unsigned long rseq_abi_pointer;
+	unsigned long rseq_abi_size;
+	uint32_t rseq_signature;
+	void *addr;
+
+	/* no need to do hacky check if we can get all info from ptrace() */
+	if (!rseq->has_rseq || rseq->has_ptrace_get_rseq_conf)
+		return 0;
+
+	/*
+	 * We need to determine if victim process has rseq()
+	 * initialized, but we have no *any* proper kernel interface
+	 * supported at this point.
+	 * Our plan:
+	 * 1. We know that if we call rseq() syscall and process already
+	 * has current->rseq filled, then we get:
+	 * -EINVAL if current->rseq != rseq || rseq_len != sizeof(*rseq),
+	 * -EPERM  if current->rseq_sig != sig),
+	 * -EBUSY  if current->rseq == rseq && rseq_len == sizeof(*rseq) &&
+	 *            current->rseq_sig != sig
+	 * if current->rseq == NULL (rseq() wasn't used) then we go to:
+	 * IS_ALIGNED(rseq ...) check, if we fail it we get -EINVAL and it
+	 * will be hard to distinguish case when rseq() was initialized or not.
+	 * Let's construct arguments payload
+	 * with:
+	 * 1. correct rseq_abi_size
+	 * 2. aligned and correct rseq_abi_pointer
+	 * And see what rseq() return to us.
+	 * If ret value is:
+	 * 0: it means that rseq *wasn't* used and we successfuly registered it,
+	 * -EINVAL or : it means that rseq is already initialized,
+	 * so we *have* to dump it. But as we have has_ptrace_get_rseq_conf = false,
+	 * we should just fail dump as it's unsafe to skip rseq() dump for processes
+	 * with rseq() initialized.
+	 * -EPERM or -EBUSY: should not happen as we take a fresh memory area for rseq
+	 */
+	addr = (void *)sys_mmap(NULL, sizeof(struct rseq), PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+	if (addr == MAP_FAILED) {
+		pr_err("mmap() failed for struct rseq ret = %lx\n", (unsigned long)addr);
+		return -1;
+	}
+
+	memset(addr, 0, sizeof(struct rseq));
+
+	/* sys_mmap returns page aligned addresses */
+	rseq_abi_pointer = (unsigned long)addr;
+	rseq_abi_size = (unsigned long)sizeof(struct rseq);
+	/* it's not so important to have unique signature for us,
+	 * because rseq_abi_pointer is guaranteed to be unique
+	 */
+	rseq_signature = 0x12345612;
+
+	pr_info("\ttrying sys_rseq(%lx, %lx, %x, %x)\n", rseq_abi_pointer, rseq_abi_size, 0, rseq_signature);
+	ret = sys_rseq((void *)rseq_abi_pointer, rseq_abi_size, 0, rseq_signature);
+	if (ret) {
+		if (ret == -EINVAL) {
+			pr_info("\trseq is initialized in the victim\n");
+			rseq->rseq_inited = true;
+
+			ret = 0;
+		} else {
+			pr_err("\tunexpected failure of sys_rseq(%lx, %lx, %x, %x) = %d\n", rseq_abi_pointer,
+			       rseq_abi_size, 0, rseq_signature, ret);
+
+			ret = -1;
+		}
+	} else {
+		ret = sys_rseq((void *)rseq_abi_pointer, sizeof(struct rseq), RSEQ_FLAG_UNREGISTER, rseq_signature);
+		if (ret) {
+			pr_err("\tfailed to unregister sys_rseq(%lx, %lx, %x, %x) = %d\n", rseq_abi_pointer,
+			       rseq_abi_size, RSEQ_FLAG_UNREGISTER, rseq_signature, ret);
+
+			ret = -1;
+			goto out;
+		}
+
+		pr_info("\tsys_rseq succeed, let's unregister it back... ok Error\n");
+		pr_info("\trseq is non-initialized in the victim Error\n");
+		rseq->rseq_inited = false;
+		ret = 0;
+	}
+
+out:
+	sys_munmap(addr, sizeof(struct rseq));
+	return ret;
+}
+
 static int fill_fds_fown(int fd, struct fd_opts *p)
 {
 	int flags, ret;
diff --git a/criu/pie/restorer.c b/criu/pie/restorer.c
index fbc89fe..368b5a0 100644
--- a/criu/pie/restorer.c
+++ b/criu/pie/restorer.c
@@ -459,6 +459,27 @@ static int restore_cpu_affinity(struct task_restore_args *args)
 	return 0;
 }
 
+static int restore_rseq(struct rst_rseq_param *rseq)
+{
+	int ret;
+
+	if (!rseq->rseq_abi_pointer) {
+		pr_debug("rseq: nothing to restore\n");
+		return 0;
+	}
+
+	pr_debug("rseq: rseq_abi_pointer = %lx signature = %x\n", (unsigned long)rseq->rseq_abi_pointer, rseq->signature);
+
+	ret = sys_rseq(decode_pointer(rseq->rseq_abi_pointer), rseq->rseq_abi_size, 0, rseq->signature);
+	if (ret) {
+		pr_err("failed sys_rseq(%lx, %lx, %x, %x) = %d\n", (unsigned long)rseq->rseq_abi_pointer,
+		       (unsigned long)rseq->rseq_abi_size, 0, rseq->signature, ret);
+		return -1;
+	}
+
+	return 0;
+}
+
 static int restore_seccomp_filter(pid_t tid, struct thread_restore_args *args)
 {
 	unsigned int flags = args->seccomp_force_tsync ? SECCOMP_FILTER_FLAG_TSYNC : 0;
@@ -583,6 +604,9 @@ static int restore_thread_common(struct thread_restore_args *args)
 
 	restore_tls(&args->tls);
 
+	if (restore_rseq(&args->rseq))
+		return -1;
+
 	return 0;
 }
 
diff --git a/images/Makefile b/images/Makefile
index 2eaeb7c..004e22e 100644
--- a/images/Makefile
+++ b/images/Makefile
@@ -71,6 +71,7 @@ proto-obj-y	+= img-streamer.o
 proto-obj-y	+= bpfmap-file.o
 proto-obj-y	+= bpfmap-data.o
 proto-obj-y	+= apparmor.o
+proto-obj-y	+= rseq.o
 
 CFLAGS		+= -iquote $(obj)/
 
diff --git a/images/core.proto b/images/core.proto
index 39e7f32..b66230e 100644
--- a/images/core.proto
+++ b/images/core.proto
@@ -14,6 +14,7 @@ import "timer.proto";
 import "creds.proto";
 import "sa.proto";
 import "siginfo.proto";
+import "rseq.proto";
 
 import "opts.proto";
 
@@ -106,6 +107,7 @@ message thread_core_entry {
 	optional string			comm		= 13;
 	optional uint64			blk_sigset_extended	= 14;
 	required thread_allowedcpus_entry 	allowed_cpus	= 15;
+	optional rseq_entry		rseq_entry	= 16;
 }
 
 message task_rlimits_entry {
diff --git a/images/rseq.proto b/images/rseq.proto
new file mode 100644
index 0000000..be28004
--- /dev/null
+++ b/images/rseq.proto
@@ -0,0 +1,9 @@
+// SPDX-License-Identifier: MIT
+
+syntax = "proto2";
+
+message rseq_entry {
+	required uint64 rseq_abi_pointer	= 1;
+	required uint32 rseq_abi_size		= 2;
+	required uint32 signature		= 3;
+}
-- 
2.30.0