infect: add SIGTSTP support

Add SIGTSTP signal dump and restore. Add a corresponding field
in the image, save it only if a task is in the stopped state.

Restore task state by sending desired stop signal if it is present
in the image. Fallback to SIGSTOP if it's absent.

Signed-off-by: Yuriy Vasiliev <yuriy.vasiliev@openvz.org>

Documentation/compel.txt

@@ -97,7 +97,10 @@ Following steps are performed to infect the victim process:
     - execute system call: *int compel_syscall(ctl, int syscall_nr, long *ret, int arg ...);*
     - infect victim: *int compel_infect(ctl, nr_thread, size_of_args_area);*
     - cure the victim: *int compel_cure(ctl);* //ctl pointer is freed by this call
-    - Resume victim: *int compel_resume_task(pid, orig_state, state);*
+    - Resume victim: *int compel_resume_task(pid, orig_state, state)* or
+    *int compel_resume_task_sig(pid, orig_state, state, stop_signo).*
+    //compel_resume_task_sig() could be used in case when victim is in stopped state.
+    stop_signo could be read by calling compel_parse_stop_signo().
 
 *ctl* must be configured with blob information by calling *PREFIX_setup_c_header()*, with ctl as its argument.
 *PREFIX* is the argument given to *-p* when calling hgen, else it is deduced from file name.

compel/include/uapi/infect.h

@@ -18,6 +18,7 @@ extern int __must_check compel_interrupt_task(int pid);
 struct seize_task_status {
 	unsigned long long sigpnd;
 	unsigned long long shdpnd;
+	unsigned long long sigblk;
 	char state;
 	int vpid;
 	int ppid;
@@ -30,7 +31,9 @@ extern int __must_check compel_wait_task(int pid, int ppid,
 					 struct seize_task_status *st, void *data);
 
 extern int __must_check compel_stop_task(int pid);
+extern int __must_check compel_parse_stop_signo(int pid);
 extern int compel_resume_task(pid_t pid, int orig_state, int state);
+extern int compel_resume_task_sig(pid_t pid, int orig_state, int state, int stop_signo);
 
 struct parasite_ctl;
 struct parasite_thread_ctl;

compel/src/lib/infect.c

@@ -92,6 +92,12 @@ static int parse_pid_status(int pid, struct seize_task_status *ss, void *data)
 
 			continue;
 		}
+		if (!strncmp(aux, "SigBlk:", 7)) {
+			if (sscanf(aux + 7, "%llx", &ss->sigblk) != 1)
+				goto err_parse;
+
+			continue;
+		}
 	}
 
 	fclose(f);
@@ -186,6 +192,29 @@ static int skip_sigstop(int pid, int nr_signals)
 	return 0;
 }
 
+#define SIG_MASK(sig) (1ULL << ((sig)-1))
+
+#define SIG_IN_MASK(sig, mask) ((sig) > 0 && (sig) <= SIGMAX && (SIG_MASK(sig) & (mask)))
+
+#define SUPPORTED_STOP_MASK ((1ULL << (SIGSTOP - 1)) | (1ULL << (SIGTSTP - 1)))
+
+static inline int sig_stop(int sig)
+{
+	return SIG_IN_MASK(sig, SUPPORTED_STOP_MASK);
+}
+
+int compel_parse_stop_signo(int pid)
+{
+	siginfo_t si;
+
+	if (ptrace(PTRACE_GETSIGINFO, pid, NULL, &si) < 0) {
+		pr_perror("SEIZE %d: can't parse stopped siginfo", pid);
+		return -1;
+	}
+
+	return si.si_signo;
+}
+
 /*
  * This routine seizes task putting it into a special
  * state where we can manipulate the task via ptrace
@@ -198,7 +227,7 @@ int compel_wait_task(int pid, int ppid, int (*get_status)(int pid, struct seize_
 		     void *data)
 {
 	siginfo_t si;
-	int status, nr_sigstop;
+	int status, nr_stopsig;
 	int ret = 0, ret2, wait_errno = 0;
 
 	/*
@@ -291,17 +320,32 @@ try_again:
 		goto err;
 	}
 
-	nr_sigstop = 0;
-	if (ss->sigpnd & (1 << (SIGSTOP - 1)))
-		nr_sigstop++;
-	if (ss->shdpnd & (1 << (SIGSTOP - 1)))
-		nr_sigstop++;
-	if (si.si_signo == SIGSTOP)
-		nr_sigstop++;
+	nr_stopsig = 0;
+	if (SIG_IN_MASK(SIGSTOP, ss->sigpnd))
+		nr_stopsig++;
+	if (SIG_IN_MASK(SIGSTOP, ss->shdpnd))
+		nr_stopsig++;
 
-	if (nr_sigstop) {
-		if (skip_sigstop(pid, nr_sigstop))
-			goto err_stop;
+	if (SIG_IN_MASK(SIGTSTP, ss->sigpnd) && !SIG_IN_MASK(SIGTSTP, ss->sigblk))
+		nr_stopsig++;
+	if (SIG_IN_MASK(SIGTSTP, ss->shdpnd) && !SIG_IN_MASK(SIGTSTP, ss->sigblk))
+		nr_stopsig++;
+
+	if (sig_stop(si.si_signo))
+		nr_stopsig++;
+
+	if (nr_stopsig) {
+		if (skip_sigstop(pid, nr_stopsig)) {
+			/*
+			 * Make sure that the task is stopped by a supported stop signal and
+			 * send it again to restore task state before criu intervention.
+			 */
+			if (sig_stop(si.si_signo))
+				kill(pid, si.si_signo);
+			else
+				kill(pid, SIGSTOP);
+			goto err;
+		}
 
 		return COMPEL_TASK_STOPPED;
 	}
@@ -313,8 +357,6 @@ try_again:
 		goto err;
 	}
 
-err_stop:
-	kill(pid, SIGSTOP);
 err:
 	if (ptrace(PTRACE_DETACH, pid, NULL, NULL))
 		pr_perror("Unable to detach from %d", pid);
@@ -322,6 +364,11 @@ err:
 }
 
 int compel_resume_task(pid_t pid, int orig_st, int st)
+{
+	return compel_resume_task_sig(pid, orig_st, st, SIGSTOP);
+}
+
+int compel_resume_task_sig(pid_t pid, int orig_st, int st, int stop_signo)
 {
 	int ret = 0;
 
@@ -345,8 +392,18 @@ int compel_resume_task(pid_t pid, int orig_st, int st)
 		 * task with STOP in queue that would get lost after
 		 * detach, so stop it again.
 		 */
-		if (orig_st == COMPEL_TASK_STOPPED)
-			kill(pid, SIGSTOP);
+		if (orig_st == COMPEL_TASK_STOPPED) {
+			/*
+			 * Check that stop_signo contain supported stop signal.
+			 * If it isn't, then send SIGSTOP. It makes sense in the case
+			 * when we get COMPEL_TASK_STOPPED from old image,
+			 * where stop_signo was not yet supported.
+			 */
+			if (sig_stop(stop_signo))
+				kill(pid, stop_signo);
+			else
+				kill(pid, SIGSTOP);
+		}
 	} else {
 		pr_err("Unknown final state %d\n", st);
 		ret = -1;

criu/cr-dump.c

@@ -781,6 +781,11 @@ static int dump_task_core_all(struct parasite_ctl *ctl, struct pstree_item *item
 	core->thread_core->creds->lsm_profile = dmpi(item)->thread_lsms[0]->profile;
 	core->thread_core->creds->lsm_sockcreate = dmpi(item)->thread_lsms[0]->sockcreate;
 
+	if (core->tc->task_state == TASK_STOPPED) {
+		core->tc->has_stop_signo = true;
+		core->tc->stop_signo = item->pid->stop_signo;
+	}
+
 	ret = parasite_dump_thread_leader_seized(ctl, pid, core);
 	if (ret)
 		goto err;

criu/cr-restore.c

@@ -1350,6 +1350,9 @@ static inline int fork_with_pid(struct pstree_item *item)
 		item->pid->state = ca.core->tc->task_state;
 		rsti(item)->cg_set = ca.core->tc->cg_set;
 
+		if (ca.core->tc->has_stop_signo)
+			item->pid->stop_signo = ca.core->tc->stop_signo;
+
 		if (item->pid->state != TASK_DEAD && !task_alive(item)) {
 			pr_err("Unknown task state %d\n", item->pid->state);
 			return -1;
@@ -2104,8 +2107,14 @@ static void finalize_restore(void)
 
 		xfree(ctl);
 
-		if ((item->pid->state == TASK_STOPPED) || (opts.final_state == TASK_STOPPED))
+		if (opts.final_state == TASK_STOPPED)
 			kill(item->pid->real, SIGSTOP);
+		else if (item->pid->state == TASK_STOPPED) {
+			if (item->pid->stop_signo > 0)
+				kill(item->pid->real, item->pid->stop_signo);
+			else
+				kill(item->pid->real, SIGSTOP);
+		}
 	}
 }
 

criu/include/pid.h

@@ -31,6 +31,10 @@ struct pid {
 	pid_t real;
 
 	int state; /* TASK_XXX constants */
+	/* If an item is in stopped state it has a signal number
+	 * that caused task to stop.
+	 */
+	int stop_signo;
 
 	/*
 	 * The @virt pid is one which used in the image itself and keeps

criu/proc_parse.c

@@ -1027,12 +1027,13 @@ int parse_pid_status(pid_t pid, struct seize_task_status *ss, void *data)
 
 	cr->s.sigpnd = 0;
 	cr->s.shdpnd = 0;
+	cr->s.sigblk = 0;
 	cr->s.seccomp_mode = SECCOMP_MODE_DISABLED;
 
 	if (bfdopenr(&f))
 		return -1;
 
-	while (done < 13) {
+	while (done < 14) {
 		str = breadline(&f);
 		if (str == NULL)
 			break;
@@ -1143,13 +1144,23 @@ int parse_pid_status(pid_t pid, struct seize_task_status *ss, void *data)
 				goto err_parse;
 			cr->s.sigpnd |= sigpnd;
 
+			done++;
+			continue;
+		}
+		if (!strncmp(str, "SigBlk:", 7)) {
+			unsigned long long sigblk = 0;
+
+			if (sscanf(str + 7, "%llx", &sigblk) != 1)
+				goto err_parse;
+			cr->s.sigblk |= sigblk;
+
 			done++;
 			continue;
 		}
 	}
 
 	/* seccomp and nspids are optional */
-	expected_done = (parsed_seccomp ? 11 : 10);
+	expected_done = (parsed_seccomp ? 12 : 11);
 	if (kdat.has_nspid)
 		expected_done++;
 	if (done == expected_done)

criu/pstree.c

@@ -222,6 +222,7 @@ struct pstree_item *__alloc_pstree_item(bool rst)
 	item->pid->ns[0].virt = -1;
 	item->pid->real = -1;
 	item->pid->state = TASK_UNDEF;
+	item->pid->stop_signo = -1;
 	item->born_sid = -1;
 	item->pid->item = item;
 	futex_init(&item->task_st);

criu/seize.c

@@ -615,6 +615,9 @@ static int collect_children(struct pstree_item *item)
 		else
 			processes_to_wait--;
 
+		if (ret == TASK_STOPPED)
+			c->pid->stop_signo = compel_parse_stop_signo(pid);
+
 		c->pid->real = pid;
 		c->parent = item;
 		c->pid->state = ret;
@@ -646,7 +649,7 @@ static void unseize_task_and_threads(const struct pstree_item *item, int st)
 	 * the item->state is the state task was in when we seized one.
 	 */
 
-	compel_resume_task(item->pid->real, item->pid->state, st);
+	compel_resume_task_sig(item->pid->real, item->pid->state, st, item->pid->stop_signo);
 
 	if (st == TASK_DEAD)
 		return;
@@ -950,6 +953,9 @@ int collect_pstree(void)
 	else
 		processes_to_wait--;
 
+	if (ret == TASK_STOPPED)
+		root_item->pid->stop_signo = compel_parse_stop_signo(pid);
+
 	pr_info("Seized task %d, state %d\n", pid, ret);
 	root_item->pid->state = ret;
 

images/core.proto

@@ -60,6 +60,8 @@ message task_core_entry {
 	// Reserved for container relative start time
 	//optional uint64		start_time	= 19;
 	optional uint64			blk_sigset_extended = 20[(criu).hex = true];
+
+	optional uint32			stop_signo = 21;
 }
 
 message task_kobj_ids_entry {