[RHEL7,COMMIT] ve/proc: Added separate start time field to task_struct to show in container

Submitted by Konstantin Khorenko on Jan. 30, 2020, 7:15 p.m.

Details

Message ID 202001301915.00UJFY5H015047@finist-ce7.sw.ru
State New
Series "Series without cover letter"
Headers show

Commit Message

Konstantin Khorenko Jan. 30, 2020, 7:15 p.m.
The commit is pushed to "branch-rh7-3.10.0-1062.7.1.vz7.130.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh7-3.10.0-1062.7.1.vz7.130.13
------>
commit a6d2403bcdf4a6a8c415743f3f802942cb91b3fb
Author: Valeriy Vdovin <valeriy.vdovin@virtuozzo.com>
Date:   Thu Jan 30 22:15:34 2020 +0300

    ve/proc: Added separate start time field to task_struct to show in container
    
    Introduced 'real_start_time_ct' field in task_struct.
    
    The value is READ:
    1. When the process lives inside of a ve group and any process
    inside of the same ve group wants to know it's start time by reading
    it's /proc/[pid]/stat file.
    2. At container suspend operation to store this value to a dump image.
    
    The value is WRITTEN:
    1. At creation time (copy_process function)
    1.1. If a process is being created outside of ve group / on host, then
    this value is initialized to 0
    1.2. If a process is being created by process already living in ve
    group, this value is calculated as host_uptime - ve_uptime.
    
    2. During attach to ve. (ve_attach function). The process can be created on
    a host and later attached to ve. It's container's start_time value has been
    already initialized to 0 at creation time. After the process enters the
    domain of a ve, the value should be initialized.
    Note that the process can be attached to a non-running container, in which
    case it's start_time value should not be calculated and left initialized to
    0.
    
    3. At container restore via prctl (prctl_set_task_ct_fields function).
    In this case the value is only settable outside of a container.
    During restore the processes would be created from the dump image.
    At restore step each process will execute prctl to set it's start_time
    value, read from the dump. This would only be permitted during
    pseudosuper ve mode. The value is set as is (read from the dump), without
    any calculations.
    
    https://jira.sw.ru/browse/PSBM-64123
    
    Signed-off-by: Valeriy Vdovin <valeriy.vdovin@virtuozzo.com>
---
 fs/proc/array.c            | 16 ++++------------
 include/linux/sched.h      |  5 +++++
 include/linux/ve.h         | 23 +++++++++++++++++++++++
 include/uapi/linux/prctl.h |  7 +++++++
 kernel/fork.c              | 13 +++++++++++++
 kernel/sys.c               | 23 +++++++++++++++++++++++
 kernel/ve/ve.c             |  2 ++
 7 files changed, 77 insertions(+), 12 deletions(-)

Patch hide | download patch | download mbox

diff --git a/fs/proc/array.c b/fs/proc/array.c
index 3aa8a7dad5e35..fb611b1723f26 100644
--- a/fs/proc/array.c
+++ b/fs/proc/array.c
@@ -611,19 +611,11 @@  static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
 		(unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
 				+ task->real_start_time.tv_nsec;
 #ifdef CONFIG_VE
-	if (!is_super) {
-		struct timespec *ve_start_ts =
-				&get_exec_env()->real_start_timespec;
-		start_time -=
-			(unsigned long long)ve_start_ts->tv_sec * NSEC_PER_SEC
-				+ ve_start_ts->tv_nsec;
-	}
-	/* tasks inside a CT can have negative start time e.g. if the CT was
-	 * migrated from another hw node, in which case we will report 0 in
-	 * order not to confuse userspace */
-	if ((s64)start_time < 0)
-		start_time = 0;
+	if (!is_super)
+		start_time = (unsigned long long)
+			timespec_to_ns(&task->real_start_time_ct);
 #endif
+
 	/* convert nsec -> ticks */
 	start_time = nsec_to_clock_t(start_time);
 
diff --git a/include/linux/sched.h b/include/linux/sched.h
index f0d976c3800a1..89982455835e5 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1928,6 +1928,11 @@  struct task_struct {
 	struct wake_q_node wake_q;
 	struct prev_cputime prev_cputime;
 	struct vtime vtime;
+	/*
+	 * this is a container-side copy of 'real_start_time' field
+	 * shown from inside of a container and modified by host.
+	 */
+	struct timespec real_start_time_ct;
 #endif /* __GENKSYMS__ */
 };
 
diff --git a/include/linux/ve.h b/include/linux/ve.h
index 9d60838677133..088e27455fea1 100644
--- a/include/linux/ve.h
+++ b/include/linux/ve.h
@@ -199,6 +199,29 @@  static inline struct ve_struct *cgroup_ve(struct cgroup *cgroup)
 			struct ve_struct, css);
 }
 
+static inline void ve_try_set_task_start_time(struct ve_struct *ve,
+	struct task_struct *t)
+{
+	struct timespec host_uptime;
+
+	/*
+	 * mitigate memory access reordering risks by doing double check,
+	 * 'is_running' could be read as 1 before we see
+	 * 'real_start_timespec' updated here. If it's still 0,
+	 * we know 'is_running' is being modified right NOW in
+	 * parallel so it's safe to say that start time is also 0
+	 */
+	if (!ve->is_running || !timespec_to_ns(&ve->real_start_timespec)) {
+		t->real_start_time_ct.tv_sec = 0;
+		t->real_start_time_ct.tv_nsec = 0;
+	} else {
+		do_posix_clock_monotonic_gettime(&host_uptime);
+		monotonic_to_bootbased(&host_uptime);
+		t->real_start_time_ct = timespec_sub(host_uptime,
+			ve->real_start_timespec);
+	}
+}
+
 extern unsigned long long ve_relative_clock(struct timespec * ts);
 extern void monotonic_abs_to_ve(clockid_t which_clock, struct timespec *tp);
 extern void monotonic_ve_to_abs(clockid_t which_clock, struct timespec *tp);
diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
index 02376de8e771d..b1851139b545d 100644
--- a/include/uapi/linux/prctl.h
+++ b/include/uapi/linux/prctl.h
@@ -204,5 +204,12 @@  struct prctl_mm_map {
 # define PR_SPEC_DISABLE		(1UL << 2)
 # define PR_SPEC_FORCE_DISABLE		(1UL << 3)
 # define PR_SPEC_DISABLE_NOEXEC		(1UL << 4)
+/* Set task container related fields */
+#define PR_SET_TASK_CT_FIELDS	1000
+#define PR_TASK_CT_FIELDS_START_TIME	(1ULL << 0)
+
+struct prctl_task_ct_fields {
+	__s64 real_start_time;
+};
 
 #endif /* _LINUX_PRCTL_H */
diff --git a/kernel/fork.c b/kernel/fork.c
index 3d74228c9ab29..2314eb886f417 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1350,6 +1350,10 @@  static struct task_struct *copy_process(unsigned long clone_flags,
 	struct task_struct *p;
 	void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
 
+#ifdef CONFIG_VE
+	struct ve_struct *ve = get_exec_env();
+#endif
+
 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 		return ERR_PTR(-EINVAL);
 
@@ -1472,6 +1476,15 @@  static struct task_struct *copy_process(unsigned long clone_flags,
 	do_posix_clock_monotonic_gettime(&p->start_time);
 	p->real_start_time = p->start_time;
 	monotonic_to_bootbased(&p->real_start_time);
+
+	p->real_start_time_ct.tv_sec = 0;
+	p->real_start_time_ct.tv_nsec = 0;
+
+#ifdef CONFIG_VE
+	if (!ve_is_super(ve))
+		ve_try_set_task_start_time(ve, p);
+#endif
+
 	p->io_context = NULL;
 	p->audit_context = NULL;
 	if (clone_flags & CLONE_THREAD)
diff --git a/kernel/sys.c b/kernel/sys.c
index c4d633efe44cc..2ce16c7d15cdc 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -2457,6 +2457,26 @@  static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
 }
 #endif
 
+static int prctl_set_task_ct_fields(struct task_struct *t, unsigned long arg,
+		unsigned long flags)
+{
+	struct prctl_task_ct_fields params;
+#ifdef CONFIG_VE
+	struct ve_struct *ve = t->task_ve;
+
+	if (!ve_is_super(ve) && !ve->is_pseudosuper)
+		return -EPERM;
+#endif
+
+	if (copy_from_user(&params, (const void __user *)arg, sizeof(params)))
+		return -EFAULT;
+
+	if (flags & PR_TASK_CT_FIELDS_START_TIME)
+		t->real_start_time_ct = ns_to_timespec(params.real_start_time);
+
+	return 0;
+}
+
 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
 {
 	return -EINVAL;
@@ -2684,6 +2704,9 @@  SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
 			return -EINVAL;
 		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
 		break;
+	case PR_SET_TASK_CT_FIELDS:
+		error = prctl_set_task_ct_fields(me, arg2, arg3);
+		break;
 	default:
 		error = -EINVAL;
 		break;
diff --git a/kernel/ve/ve.c b/kernel/ve/ve.c
index ad3a698ba9dee..f3970e8ef0f6a 100644
--- a/kernel/ve/ve.c
+++ b/kernel/ve/ve.c
@@ -850,6 +850,8 @@  static void ve_attach(struct cgroup *cg, struct cgroup_taskset *tset)
 		/* Leave parent exec domain */
 		task->parent_exec_id--;
 
+		ve_try_set_task_start_time(ve, task);
+
 		task->task_ve = ve;
 	}