core: Support task scheduler policies and priorities

No magic here, just fetch info using getpriority and sched_getxxx calls.
Good news is that the mentioned syscalls take pid as argument and do work
with it, i.e. -- no need in parasite help here.

Restore is splitted into prep -- copy sched bits from image on restorer
args -- and the restore itself. It's done to avoid restoring tasks info
with IDLE priority ;) To make restorer not-fail sched bits are validated
for sanity on prep stage.

Minimal sanity test is also there.

Signed-off-by: Pavel Emelyanov <xemul@parallels.com>

cr-dump.c

@@ -18,6 +18,9 @@
 #include <sys/sendfile.h>
 #include <sys/mman.h>
 
+#include <sched.h>
+#include <sys/resource.h>
+
 #include <linux/major.h>
 
 #include "protobuf.h"
@@ -98,6 +101,54 @@ err:
 	return ret;
 }
 
+static int dump_sched_info(int pid, ThreadCoreEntry *tc)
+{
+	int ret;
+	struct sched_param sp;
+
+	BUILD_BUG_ON(SCHED_OTHER != 0); /* default in proto message */
+
+	ret = sched_getscheduler(pid);
+	if (ret < 0) {
+		pr_perror("Can't get sched policy for %d", pid);
+		return -1;
+	}
+
+	pr_info("%d has %d sched policy\n", pid, ret);
+	tc->has_sched_policy = true;
+	tc->sched_policy = ret;
+
+	if ((ret == SCHED_RR) || (ret == SCHED_FIFO)) {
+		ret = sched_getparam(pid, &sp);
+		if (ret < 0) {
+			pr_perror("Can't get sched param for %d", pid);
+			return -1;
+		}
+
+		pr_info("\tdumping %d prio for %d\n", sp.sched_priority, pid);
+		tc->has_sched_prio = true;
+		tc->sched_prio = sp.sched_priority;
+	}
+
+	/*
+	 * The nice is ignored for RT sched policies, but is stored
+	 * in kernel. Thus we have to take it with us in the image.
+	 */
+
+	errno = 0;
+	ret = getpriority(PRIO_PROCESS, pid);
+	if (errno) {
+		pr_perror("Can't get nice for %d", pid);
+		return -1;
+	}
+
+	pr_info("\tdumping %d nice for %d\n", ret, pid);
+	tc->has_sched_nice = true;
+	tc->sched_nice = ret;
+
+	return 0;
+}
+
 struct cr_fdset *glob_fdset;
 
 static int collect_fds(pid_t pid, struct parasite_drain_fd *dfds)
@@ -880,6 +931,10 @@ static int dump_task_core_all(pid_t pid, const struct proc_pid_stat *stat,
 	core->tc->task_state = TASK_ALIVE;
 	core->tc->exit_code = 0;
 
+	ret = dump_sched_info(pid, core->thread_core);
+	if (ret)
+		goto err_free;
+
 	ret = pb_write_one(fd_core, core, PB_CORE);
 	if (ret < 0) {
 		pr_info("ERROR\n");
@@ -1277,6 +1332,10 @@ static int dump_task_thread(struct parasite_ctl *parasite_ctl, struct pid *tid)
 	pr_info("%d: tid_address=%p\n", pid, taddr);
 	core->thread_info->clear_tid_addr = (u64) taddr;
 
+	ret = dump_sched_info(pid, core->thread_core);
+	if (ret)
+		goto err_free;
+
 	pr_info("OK\n");
 
 	fd_core = open_image(CR_FD_CORE, O_DUMP, tid->virt);

cr-restore.c

@@ -1258,6 +1258,45 @@ static int remap_restorer_blob(void *addr)
 	return 0;
 }
 
+static int validate_sched_parm(struct rst_sched_param *sp)
+{
+	if ((sp->nice < -20) || (sp->nice > 19))
+		return 0;
+
+	switch (sp->policy) {
+	case SCHED_RR:
+	case SCHED_FIFO:
+		return ((sp->prio > 0) && (sp->prio < 100));
+	case SCHED_IDLE:
+	case SCHED_OTHER:
+	case SCHED_BATCH:
+		return sp->prio == 0;
+	}
+
+	return 0;
+}
+
+static int prep_sched_info(struct rst_sched_param *sp, ThreadCoreEntry *tc)
+{
+	if (!tc->has_sched_policy) {
+		sp->policy = SCHED_OTHER;
+		sp->nice = 0;
+		return 0;
+	}
+
+	sp->policy = tc->sched_policy;
+	sp->nice = tc->sched_nice;
+	sp->prio = tc->sched_prio;
+
+	if (!validate_sched_parm(sp)) {
+		pr_err("Inconsistent sched params received (%d.%d.%d)\n",
+				sp->policy, sp->nice, sp->prio);
+		return -1;
+	}
+
+	return 0;
+}
+
 static int sigreturn_restore(pid_t pid, CoreEntry *core, struct list_head *tgt_vmas, int nr_vmas)
 {
 	long restore_task_vma_len;
@@ -1427,6 +1466,10 @@ static int sigreturn_restore(pid_t pid, CoreEntry *core, struct list_head *tgt_v
 		task_args->has_futex		= true;
 		task_args->futex_rla		= core->thread_core->futex_rla;
 		task_args->futex_rla_len	= core->thread_core->futex_rla_len;
+
+		ret = prep_sched_info(&task_args->sp, core->thread_core);
+		if (ret)
+			goto err;
 	}
 
 	/* No longer need it */
@@ -1494,6 +1537,10 @@ static int sigreturn_restore(pid_t pid, CoreEntry *core, struct list_head *tgt_v
 			thread_args[i].has_futex	= true;
 			thread_args[i].futex_rla	= core->thread_core->futex_rla;
 			thread_args[i].futex_rla_len	= core->thread_core->futex_rla_len;
+
+			ret = prep_sched_info(&thread_args[i].sp, core->thread_core);
+			if (ret)
+				goto err;
 		}
 
 		core_entry__free_unpacked(core, NULL);

include/restorer.h

@@ -56,6 +56,12 @@ struct restore_mem_zone {
 #define first_on_heap(ptr, heap)	((typeof(ptr))heap)
 #define next_on_heap(ptr, prev)		((typeof(ptr))((long)(prev) + sizeof(*(prev))))
 
+struct rst_sched_param {
+	int policy;
+	int nice;
+	int prio;
+};
+
 /* Make sure it's pow2 in size */
 struct thread_restore_args {
 	struct restore_mem_zone		mem_zone;
@@ -68,6 +74,8 @@ struct thread_restore_args {
 	bool				has_futex;
 	u64				futex_rla;
 	u32				futex_rla_len;
+
+	struct rst_sched_param		sp;
 } __aligned(sizeof(long));
 
 struct task_restore_core_args {
@@ -113,6 +121,8 @@ struct task_restore_core_args {
 
 	int				*rst_tcp_socks;
 	int				rst_tcp_socks_size;
+
+	struct rst_sched_param		sp;
 } __aligned(sizeof(long));
 
 struct pt_regs {

include/syscall-types.h

@@ -10,6 +10,7 @@
 
 #include <sys/time.h>
 #include <arpa/inet.h>
+#include <sched.h>
 
 #include "types.h"
 

include/syscall-x86-64.def

@@ -54,6 +54,8 @@ __NR_setfsgid		123		sys_setfsgid		(int fsgid)
 __NR_getsid		124		sys_getsid		(void)
 __NR_capset		126		sys_capset		(struct cap_header *h, struct cap_data *d)
 __NR_personality	135		sys_personality		(unsigned int personality)
+__NR_setpriority	141		sys_setpriority		(int which, int who, int nice)
+__NR_sched_setscheduler	144		sys_sched_setscheduler	(int pid, int policy, struct sched_param *p)
 __NR_prctl		157		sys_prctl		(int option, unsigned long arg2, unsigned long arg3, unsigned long arg4, unsigned long arg5)
 __NR_arch_prctl		158		sys_arch_prctl		(int option, unsigned long addr)
 __NR_mount		165		sys_mount		(char *dev_nmae, char *dir_name, char *type, unsigned long flags, void *data)

protobuf/core.proto

@@ -69,6 +69,9 @@ message thread_info_x86 {
 message thread_core_entry {
 	required uint64			futex_rla	= 1;
 	required uint32			futex_rla_len	= 2;
+	optional sint32			sched_nice	= 3;
+	optional uint32			sched_policy	= 4;
+	optional uint32			sched_prio	= 5;
 }
 
 message core_entry {

restorer.c

@@ -127,6 +127,21 @@ static void restore_creds(CredsEntry *ce)
 	sys_capset(&hdr, data);
 }
 
+static void restore_sched_info(struct rst_sched_param *p)
+{
+	struct sched_param parm;
+
+	if ((p->policy == SCHED_OTHER) && (p->nice == 0))
+		return;
+
+	pr_info("Restoring scheduler params %d.%d.%d\n",
+			p->policy, p->nice, p->prio);
+
+	sys_setpriority(PRIO_PROCESS, 0, p->nice);
+	parm.sched_priority = p->prio;
+	sys_sched_setscheduler(0, p->policy, &parm);
+}
+
 /*
  * Threads restoration via sigreturn. Note it's locked
  * routine and calls for unlock at the end.
@@ -152,6 +167,8 @@ long __export_restore_thread(struct thread_restore_args *args)
 		}
 	}
 
+	restore_sched_info(&args->sp);
+
 	rt_sigframe = (void *)args->mem_zone.rt_sigframe + 8;
 
 #define CPREGT1(d)	rt_sigframe->uc.uc_mcontext.d = args->gpregs.d
@@ -442,6 +459,8 @@ long __export_restore_task(struct task_restore_core_args *args)
 		}
 	}
 
+	restore_sched_info(&args->sp);
+
 	/*
 	 * We need to prepare a valid sigframe here, so
 	 * after sigreturn the kernel will pick up the

test/zdtm.sh

@@ -19,6 +19,7 @@ static/write_read02
 static/write_read10
 static/wait00
 static/vdso00
+static/sched_prio00
 static/file_shared
 static/timers
 static/futex

test/zdtm/live/static/Makefile

@@ -19,6 +19,7 @@ TST_NOFILE	=				\
 		mprotect00			\
 		timers				\
 		unbound_sock			\
+		sched_prio00			\
 		socket_listen			\
 		socket_udp			\
 		socket6_udp			\

test/zdtm/live/static/sched_prio00.c

@@ -0,0 +1,108 @@
+#include <errno.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <signal.h>
+#include <string.h>
+#include <sched.h>
+
+#include "zdtmtst.h"
+
+const char *test_doc	= "Check sched prios to be preserved";
+const char *test_author	= "Pavel Emelyanov <xemul@parallels.com>";
+
+#define NRTASKS	4
+
+static int do_nothing(void)
+{
+	while (1)
+		sleep(10);
+
+	return -1;
+}
+
+static void kill_all(int *pid, int n)
+{
+	int i;
+
+	for (i = 0; i < n; i++)
+		kill(pid[i], SIGKILL);
+}
+
+int main(int argc, char ** argv)
+{
+	int pid[NRTASKS], i, parm[NRTASKS], ret;
+
+	test_init(argc, argv);
+
+	/* first 3 -- normal */
+	parm[0] = -20;
+	parm[1] = 19;
+	parm[2] = 1;
+	parm[3] = 3;
+
+	/* next 1 -- RR */
+
+	for (i = 0; i < NRTASKS; i++) {
+		pid[i] = fork();
+		if (!pid[i])
+			return do_nothing();
+
+		if (i < 3) {
+			if (setpriority(PRIO_PROCESS, pid[i], parm[i])) {
+				err("Can't set prio %d", i);
+				kill_all(pid, i);
+				return -1;
+			}
+		} else {
+			struct sched_param p;
+
+			p.sched_priority = parm[i];
+			if (sched_setscheduler(pid[i], SCHED_RR, &p)) {
+				err("Can't set policy %d", i);
+				kill_all(pid, i);
+				return -1;
+			}
+		}
+	}
+
+	test_daemon();
+	test_waitsig();
+
+	for (i = 0; i < NRTASKS; i++) {
+		if (i < 3) {
+			errno = 0;
+			ret = getpriority(PRIO_PROCESS, pid[i]);
+			if (errno) {
+				fail("No prio for task %d", i);
+				break;
+			}
+
+			if (ret != parm[i]) {
+				fail("Broken nice for %d", i);
+				break;
+			}
+		} else {
+			struct sched_param p;
+
+			ret = sched_getscheduler(pid[i]);
+			if (ret != SCHED_RR) {
+				fail("Broken/No policy for %d", i);
+				break;
+			}
+
+			ret = sched_getparam(pid[i], &p);
+			if (ret < 0 || p.sched_priority != parm[i]) {
+				fail("Broken prio for %d", i);
+				break;
+			}
+		}
+	}
+
+	if (i == NRTASKS)
+		pass();
+
+	kill_all(pid, NRTASKS);
+	return 0;
+}