kernel/kernel-rt/centos/patches/Notification-of-death-of-arbitrary-processes.patch

From 220edc1ccc6a0bc3dfb94a92946bf2b9a6cc0c61 Mon Sep 17 00:00:00 2001
From: Chris Friesen <chris.friesen@windriver.com>
Date: Thu, 7 Apr 2016 11:16:19 -0600
Subject: [PATCH] Notification of death of arbitrary processes

Note: this commit was copied from Titanium Cloud Rel2

This exposes a new feature which may be called to request
notification when an arbitrary process changes state. The
caller specifies a pid, signal number, and event mask, and
when that pid dies, or is stopped, or anything else that
would normally cause a SIGCHLD, the kernel will send the
specified signal to the caller if the event is in the event
mask originally passed down. The siginfo_t struct will
contain the same information as would be included with SIGCHLD.

This is exposed to userspace via the prctl() call with the
PR_DO_NOTIFY_TASK_STATE option.

Signed-off-by: Jim Somerville <Jim.Somerville@windriver.com>
Signed-off-by: Zhang Zhiguo <zhangzhg@neusoft.com>
Signed-off-by: Shuicheng Lin <shuicheng.lin@intel.com>
Signed-off-by: Austin Sun <austin.sun@intel.com>
---
 include/linux/init_task.h  |   9 ++
 include/linux/sched.h      |   6 +
 include/uapi/linux/prctl.h |  18 +++
 init/Kconfig               |  15 +++
 init/init_task.c           |   1 +
 kernel/Makefile            |   1 +
 kernel/death_notify.c      | 228 +++++++++++++++++++++++++++++++++++++
 kernel/death_notify.h      |  46 ++++++++
 kernel/exit.c              |   6 +
 kernel/fork.c              |   4 +
 kernel/signal.c            |  11 ++
 kernel/sys.c               |   8 ++
 12 files changed, 353 insertions(+)
 create mode 100644 kernel/death_notify.c
 create mode 100644 kernel/death_notify.h

diff --git a/include/linux/init_task.h b/include/linux/init_task.h
index a7083a45a..1ad2341b3 100644
--- a/include/linux/init_task.h
+++ b/include/linux/init_task.h
@@ -24,6 +24,15 @@
 extern struct files_struct init_files;
 extern struct fs_struct init_fs;
 extern struct nsproxy init_nsproxy;
+
+#ifdef CONFIG_SIGEXIT
+#define INIT_SIGEXIT(tsk) \
+	.notify		= LIST_HEAD_INIT(tsk.notify),			\
+	.monitor	= LIST_HEAD_INIT(tsk.monitor),
+#else
+#define INIT_SIGEXIT(tsk)
+#endif
+
 extern struct group_info init_groups;
 extern struct cred init_cred;
 
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 91596900f..5e34f5021 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1083,6 +1083,12 @@ struct task_struct {
 	short				il_prev;
 	short				pref_node_fork;
 #endif
+#ifdef CONFIG_SIGEXIT
+	/* list of processes to notify on death */
+	struct                          list_head notify;
+	/* list of outstanding monitor requests */
+	struct                          list_head monitor;
+#endif
 #ifdef CONFIG_NUMA_BALANCING
 	int				numa_scan_seq;
 	unsigned int			numa_scan_period;
diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
index 327319b88..708484598 100644
--- a/include/uapi/linux/prctl.h
+++ b/include/uapi/linux/prctl.h
@@ -63,6 +63,24 @@
 # define PR_ENDIAN_LITTLE	1	/* True little endian mode */
 # define PR_ENDIAN_PPC_LITTLE	2	/* "PowerPC" pseudo little endian */
 
+#ifdef CONFIG_SIGEXIT
+#define PR_DO_NOTIFY_TASK_STATE 17	/* Set/get notification for task
+					   state changes */
+
+/* This is the data structure for requestion process death
+ * (and other state change) information.  Sig of -1 means
+ * query, sig of 0 means deregistration, positive sig means
+ * that you want to set it.  sig and events are value-result
+ * and will be updated with the previous values on every
+ * successful call.
+ */
+struct task_state_notify_info {
+	pid_t pid;
+	int sig;
+	unsigned int events;
+};
+#endif
+
 /* Get/set process seccomp mode */
 #define PR_GET_SECCOMP	21
 #define PR_SET_SECCOMP	22
diff --git a/init/Kconfig b/init/Kconfig
index 7625bea9c..f88a36fef 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -1610,6 +1610,21 @@ config VM_EVENT_COUNTERS
 	  on EXPERT systems.  /proc/vmstat will only show page counts
 	  if VM event counters are disabled.
 
+config SIGEXIT
+	bool "Notification of death of arbitrary processes"
+	default n
+	help
+	  When enabled this exposes a new feature which may be called to request
+	  notification when an arbitrary process changes state.  The caller specifies
+	  a pid, signal number, and event mask, and when that pid dies, or is
+	  stopped, or anything else that would normally cause a SIGCHLD, the
+	  kernel will send the specified signal to the caller if the event is in
+	  the event mask originally passed down. The siginfo_t struct will
+	  contain the same information as would be included with SIGCHLD.
+
+	  This is exposed to userspace via the prctl()
+	  call with the PR_DO_NOTIFY_TASK_STATE option
+
 config SLUB_DEBUG
 	default y
 	bool "Enable SLUB debugging support" if EXPERT
diff --git a/init/init_task.c b/init/init_task.c
index f2d3b023f..1623a4273 100644
--- a/init/init_task.c
+++ b/init/init_task.c
@@ -123,6 +123,7 @@ struct task_struct init_task
 	.alloc_lock	= __SPIN_LOCK_UNLOCKED(init_task.alloc_lock),
 	.journal_info	= NULL,
 	INIT_CPU_TIMERS(init_task)
+	INIT_SIGEXIT(init_task)
 	.pi_lock	= __RAW_SPIN_LOCK_UNLOCKED(init_task.pi_lock),
 	.timer_slack_ns = 50000, /* 50 usec default slack */
 	INIT_TIMER_LIST
diff --git a/kernel/Makefile b/kernel/Makefile
index aebf6a889..2669df332 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -104,6 +104,7 @@ obj-$(CONFIG_TRACEPOINTS) += trace/
 obj-$(CONFIG_IRQ_WORK) += irq_work.o
 obj-$(CONFIG_CPU_PM) += cpu_pm.o
 obj-$(CONFIG_BPF) += bpf/
+obj-$(CONFIG_SIGEXIT) += death_notify.o
 
 obj-$(CONFIG_PERF_EVENTS) += events/
 
diff --git a/kernel/death_notify.c b/kernel/death_notify.c
new file mode 100644
index 000000000..5819d35a2
--- /dev/null
+++ b/kernel/death_notify.c
@@ -0,0 +1,228 @@
+/*
+ * kernel/death_notify.c, Process death notification support
+ *
+ * Copyright (c) 2006-2014 Wind River Systems, Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#include <linux/errno.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/sched/task.h>
+#include <linux/slab.h>
+#include <linux/prctl.h>
+#include <linux/uaccess.h>
+#include "death_notify.h"
+
+static void unlink_status_notifier(struct signotifier *n)
+{
+	list_del(&n->monitor_list);
+	list_del(&n->notify_list);
+	kfree(n);
+}
+
+static void handle_already_monitoring(struct signotifier *node,
+       struct task_state_notify_info *args,
+       struct task_state_notify_info *oldargs)
+{
+	/* Store the old values */
+	oldargs->sig = node->sig;
+	oldargs->events = node->events;
+
+	/* We know that args->sig is 0 or a valid signal. */
+	if (args->sig > 0) {
+		/* Update the new values */
+		node->sig = args->sig;
+		node->events = args->events;
+	} else if (!args->sig) {
+		/* args->sig of 0 means to deregister */
+		unlink_status_notifier(node);
+	}
+}
+
+static void setup_new_node(struct task_struct *p,
+	struct signotifier *node,
+	struct task_state_notify_info *args)
+{
+	node->notify_tsk = current;
+	node->sig = args->sig;
+	node->events = args->events;
+
+	/* Add this node to the list of notification requests
+	 * for the specified process.
+	 */
+	list_add_tail(&node->notify_list, &p->notify);
+
+	/* Also add this node to the list of monitor requests
+	 * for the current process.
+	 */
+	list_add_tail(&node->monitor_list, &current->monitor);
+}
+
+/* Returns 0 if arguments are valid, 1 if they are not. */
+static int invalid_args(struct task_state_notify_info *args)
+{
+	int ret = 1;
+
+	if (args->pid <= 0)
+		goto out;
+
+	/* Sig of -1 implies query, sig of 0 implies deregistration.
+	 * Otherwise sig must be positive and within range.
+	 */
+	if ((args->sig < -1) || (args->sig > _NSIG))
+		goto out;
+
+	/* If positive sig, must have valid events. */
+	if (args->sig > 0) {
+		if (!args->events || (args->events >= (1 << (NSIGCHLD+1))))
+			goto out;
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+/* Notify those registered for process state updates via do_notify_task_state().
+ * If "del" is nonzero, the process is dying and we want to free
+ * the nodes in the list as we go.
+ *
+ * Note: we only notify processes for events in which they have registered
+ * interest.
+ *
+ * Must be called holding a lock on tasklist_lock.
+ */
+void do_notify_others(struct task_struct *tsk, struct kernel_siginfo *info)
+{
+	struct signotifier *node;
+	unsigned int events;
+
+	/* This method of generating the event bit must be
+	 * matched in the userspace library.
+	 */
+	events = 1 << (info->si_code & 0xFF);
+
+	list_for_each_entry(node, &tsk->notify, notify_list) {
+		if (events & node->events) {
+			info->si_signo = node->sig;
+			group_send_sig_info(node->sig, info, node->notify_tsk, PIDTYPE_TGID);
+		}
+	}
+}
+
+void release_notify_others(struct task_struct *p)
+{
+	struct signotifier *n, *t;
+
+	/* Need to clean up any outstanding requests where we
+	 * wanted to be notified when others died.
+	 */
+	list_for_each_entry_safe(n, t, &p->monitor, monitor_list) {
+		unlink_status_notifier(n);
+	}
+
+	/* Also need to clean up any outstanding requests where others
+	 * wanted to be notified when we died.
+	 */
+	list_for_each_entry_safe(n, t, &p->notify, notify_list) {
+		unlink_status_notifier(n);
+	}
+}
+
+/* If the config is defined, then processes can call this routine
+ * to request notification when the specified task's state changes.
+ * On the death (or other state change) of the specified process,
+ * we will send them the specified signal if the event is listed
+ * in their event bitfield.
+ *
+ * A sig of 0 means that we want to deregister.
+ *
+ * The sig/events fields are value/result.  On success we update them
+ * to reflect what they were before the call.
+ *
+ * Returns error code on error, on success we return 0.
+ */
+int do_notify_task_state(unsigned long arg)
+{
+	int err;
+	struct task_struct *p;
+	struct signotifier *node, *tmp;
+	struct task_state_notify_info args, oldargs;
+
+	if (copy_from_user(&args, (struct task_state_notify_info __user *)arg,
+			sizeof(args)))
+		return -EFAULT;
+	oldargs.pid = args.pid;
+
+	/* Validate the arguments passed in. */
+	err = -EINVAL;
+	if (invalid_args(&args))
+		goto out;
+
+	/* We must hold a write lock on tasklist_lock to add the notification
+	 * later on, and we need some lock on tasklist_lock for
+	 * find_task_by_pid(), so may as well take the write lock now.
+	 * Must use write_lock_irq().
+	 */
+	write_lock_irq(&tasklist_lock);
+
+	err = -ESRCH;
+	p = find_task_by_vpid(args.pid);
+	if (!p)
+		goto unlock_out;
+
+	/* Now we know pid exists, unlikely to fail. */
+	err = 0;
+
+	/* Check if we're already monitoring the specified pid. If so, update
+	 * the monitoring parameters and return the old ones.
+	 */
+	list_for_each_entry(tmp, &p->notify, notify_list) {
+		if (tmp->notify_tsk == current) {
+			handle_already_monitoring(tmp, &args, &oldargs);
+			goto unlock_out;
+		}
+	}
+
+	/* If we get here, we're not currently monitoring the process. */
+	oldargs.sig = 0;
+	oldargs.events = 0;
+
+	/* If we wanted to set up a new monitor, do it now. If we didn't
+	 * manage to allocate memory for the new node, then we return
+	 * an appropriate error.
+	 */
+	if (args.sig > 0) {
+		node = kmalloc(sizeof(*node), GFP_ATOMIC);
+		if (node)
+			setup_new_node(p, node, &args);
+		else
+			err = -ENOMEM;
+	}
+
+unlock_out:
+	write_unlock_irq(&tasklist_lock);
+
+	/* Copy the old values back to caller. */
+	if (copy_to_user((struct task_state_notify_info __user *)arg,
+			&oldargs, sizeof(oldargs)))
+		err = -EFAULT;
+
+out:
+	return err;
+}
+
diff --git a/kernel/death_notify.h b/kernel/death_notify.h
new file mode 100644
index 000000000..14a0995b7
--- /dev/null
+++ b/kernel/death_notify.h
@@ -0,0 +1,46 @@
+/*
+ * kernel/death_notify.h, Process death notification support
+ *
+ * Copyright (c) 2006-2014 Wind River Systems, Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+#ifndef _KERNEL_DEATH_NOTIFY_H
+#define _KERNEL_DEATH_NOTIFY_H
+
+#ifdef CONFIG_SIGEXIT
+
+struct signotifier {
+	struct task_struct *notify_tsk;
+	struct list_head notify_list;
+	struct list_head monitor_list;
+	int sig;
+	unsigned int events;
+};
+
+extern int do_notify_task_state(unsigned long arg);
+extern void do_notify_others(struct task_struct *tsk,
+					struct kernel_siginfo *info);
+extern void release_notify_others(struct task_struct *p);
+
+#else /* !CONFIG_SIGEXIT */
+
+static inline void do_notify_others(struct task_struct *tsk,
+					struct kernel_siginfo *info) {}
+static inline void release_notify_others(struct task_struct *p) {}
+
+#endif /* CONFIG_SIGEXIT */
+#endif
+
diff --git a/kernel/exit.c b/kernel/exit.c
index 6199d4812..0467ef691 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -67,6 +67,9 @@
 #include <asm/unistd.h>
 #include <asm/pgtable.h>
 #include <asm/mmu_context.h>
+#ifdef CONFIG_SIGEXIT
+#include "death_notify.h"
+#endif
 
 static void __unhash_process(struct task_struct *p, bool group_dead)
 {
@@ -196,6 +199,9 @@ void release_task(struct task_struct *p)
 	proc_flush_task(p);
 
 	write_lock_irq(&tasklist_lock);
+#ifdef CONFIG_SIGEXIT
+	release_notify_others(p);
+#endif
 	ptrace_release_task(p);
 	__exit_signal(p);
 
diff --git a/kernel/fork.c b/kernel/fork.c
index 63245141d..c20efb927 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1953,6 +1953,10 @@ static __latent_entropy struct task_struct *copy_process(
 	p->sequential_io	= 0;
 	p->sequential_io_avg	= 0;
 #endif
+#ifdef CONFIG_SIGEXIT
+	INIT_LIST_HEAD(&p->notify);
+	INIT_LIST_HEAD(&p->monitor);
+#endif
 
 	/* Perform scheduler related setup. Assign this task to a CPU. */
 	retval = sched_fork(clone_flags, p);
diff --git a/kernel/signal.c b/kernel/signal.c
index f4d847cc4..c25b78c05 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -53,6 +53,9 @@
 #include <asm/siginfo.h>
 #include <asm/cacheflush.h>
 #include "audit.h"	/* audit_signal_info() */
+#ifdef CONFIG_SIGEXIT
+#include "death_notify.h"
+#endif
 
 /*
  * SLAB caches for signal bits.
@@ -1989,6 +1992,10 @@ bool do_notify_parent(struct task_struct *tsk, int sig)
 	__wake_up_parent(tsk, tsk->parent);
 	spin_unlock_irqrestore(&psig->siglock, flags);
 
+#ifdef CONFIG_SIGEXIT
+	do_notify_others(tsk, &info);
+#endif
+
 	return autoreap;
 }
 
@@ -2061,6 +2068,10 @@ static void do_notify_parent_cldstop(struct task_struct *tsk,
 	 */
 	__wake_up_parent(tsk, parent);
 	spin_unlock_irqrestore(&sighand->siglock, flags);
+
+#ifdef CONFIG_SIGEXIT
+	do_notify_others(tsk, &info);
+#endif
 }
 
 static inline bool may_ptrace_stop(void)
diff --git a/kernel/sys.c b/kernel/sys.c
index 8a0fab918..e957bd5a5 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -75,6 +75,9 @@
 #include <linux/nospec.h>
 
 #include "uid16.h"
+#ifdef CONFIG_SIGEXIT
+#include "death_notify.h"
+#endif
 
 #ifndef SET_UNALIGN_CTL
 # define SET_UNALIGN_CTL(a, b)	(-EINVAL)
@@ -2409,6 +2412,11 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
 		else
 			error = PR_MCE_KILL_DEFAULT;
 		break;
+#ifdef CONFIG_SIGEXIT
+	case PR_DO_NOTIFY_TASK_STATE:
+		error = do_notify_task_state(arg2);
+		break;
+#endif
 	case PR_SET_MM:
 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
 		break;
-- 
2.19.1