From 07c3c72d18e5c7da2109b5afa918966733039f13 Mon Sep 17 00:00:00 2001
From: Bixuan Cui <cuibixuan@linux.alibaba.com>
Date: Sun, 5 Jun 2022 02:10:24 +0800
Subject: [PATCH] rasdaemon: Add notification support when page goes offline for Memory Corrected Error

When the page goes offline, it may affect the user's processes.
The user needs to do some special actions (such as restarting the
process) before or after going offline.

So add page-ce-offline-pre-notice and page-ce-offline-post-notice
to env file of rasdaemon for notifying the user when doing page
offline.

Signed-off-by: Bixuan Cui <cuibixuan@linux.alibaba.com>
---
 Makefile.am                              |  2 +-
 misc/notices/page-ce-offline-post-notice | 17 +++++
 misc/notices/page-ce-offline-pre-notice  | 17 +++++
 misc/rasdaemon.env                       |  4 ++
 misc/rasdaemon.spec.in                   |  3 +
 ras-page-isolation.c                     | 90 ++++++++++++++++++++++++
 6 files changed, 132 insertions(+), 1 deletion(-)
 create mode 100755 misc/notices/page-ce-offline-post-notice
 create mode 100755 misc/notices/page-ce-offline-pre-notice

diff --git a/Makefile.am b/Makefile.am
index de76301..701b120 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -1,6 +1,6 @@
 ACLOCAL_AMFLAGS=-I m4
 SUBDIRS = libtrace util man
-SYSTEMD_SERVICES_IN = misc/rasdaemon.service.in misc/ras-mc-ctl.service.in misc/rasdaemon.env
+SYSTEMD_SERVICES_IN = misc/rasdaemon.service.in misc/ras-mc-ctl.service.in misc/rasdaemon.env misc/notices
 SYSTEMD_SERVICES = $(SYSTEMD_SERVICES_IN:.service.in=.service)
 EXTRA_DIST = $(SYSTEMD_SERVICES_IN)
 
diff --git a/misc/notices/page-ce-offline-post-notice b/misc/notices/page-ce-offline-post-notice
new file mode 100755
index 0000000..d78b1b0
--- /dev/null
+++ b/misc/notices/page-ce-offline-post-notice
@@ -0,0 +1,17 @@
+#!/bin/sh
+#  This shell script can be executed by rasdaemon after a page goes offline.
+
+cd `dirname $0`
+
+[ -x ./page-ce-offline-post-notice.local ] && . ./page-ce-offline-post-notice.local $1
+
+if [ -d page-ce-offline-post-notice.extern ]
+then
+    ls page-ce-offline-post-notice.extern |
+    while read item
+    do
+        [ -x ./page-ce-offline-post-notice.extern/$item ] && . ./page-ce-offline-post-notice.extern/$item $1
+    done
+fi
+
+exit 0
diff --git a/misc/notices/page-ce-offline-pre-notice b/misc/notices/page-ce-offline-pre-notice
new file mode 100755
index 0000000..d1038a3
--- /dev/null
+++ b/misc/notices/page-ce-offline-pre-notice
@@ -0,0 +1,17 @@
+#!/bin/sh
+#  This shell script can be executed by rasdaemon before a page goes offline.
+
+cd `dirname $0`
+
+[ -x ./page-ce-offline-pre-notice.local ] && . ./page-ce-offline-pre-notice.local $1
+
+if [ -d page-ce-offline-pre-notice.extern ]
+then
+    ls page-ce-offline-pre-notice.extern |
+    while read item
+    do
+        [ -x ./page-ce-offline-pre-notice.extern/$item ] && . ./page-ce-offline-pre-notice.extern/$item $1
+    done
+fi
+
+exit 0
diff --git a/misc/rasdaemon.env b/misc/rasdaemon.env
index 12fd766..713875a 100644
--- a/misc/rasdaemon.env
+++ b/misc/rasdaemon.env
@@ -27,3 +27,7 @@ PAGE_CE_THRESHOLD="50"
 # soft-then-hard   First try to soft offline, then try hard offlining.
 # Note: default offline choice is "soft".
 PAGE_CE_ACTION="soft"
+
+# Notices script when doing memory offline
+PAGE_CE_OFFLINE_PRE_NOTICE="page-ce-offline-pre-notice"
+PAGE_CE_OFFLINE_POST_NOTICE="page-ce-offline-post-notice"
diff --git a/misc/rasdaemon.spec.in b/misc/rasdaemon.spec.in
index eff9794..f690575 100644
--- a/misc/rasdaemon.spec.in
+++ b/misc/rasdaemon.spec.in
@@ -45,6 +45,8 @@ make install DESTDIR=%{buildroot}
 install -D -p -m 0644 misc/rasdaemon.env %{buildroot}%{_sysconfdir}/sysconfig/%{name}
 install -D -p -m 0644 misc/rasdaemon.service %{buildroot}/%{_unitdir}/rasdaemon.service
 install -D -p -m 0644 misc/ras-mc-ctl.service %{buildroot}%{_unitdir}/ras-mc-ctl.service
+install -d %{buildroot}%{_sysconfdir}/rasdaemon_notices/
+install -D -p -m 0755 misc/notices/* %{buildroot}%{_sysconfdir}/rasdaemon_notices/
 rm INSTALL %{buildroot}/usr/include/*.h
 
 %files
@@ -56,6 +58,7 @@ rm INSTALL %{buildroot}/usr/include/*.h
 %{_sharedstatedir}/rasdaemon
 %{_sysconfdir}/ras/dimm_labels.d
 %config(noreplace) %{_sysconfdir}/sysconfig/%{name}
+%config(noreplace) %{_sysconfdir}/rasdaemon_notices/*
 
 %changelog
 
diff --git a/ras-page-isolation.c b/ras-page-isolation.c
index 50e4406..f4f3bc1 100644
--- a/ras-page-isolation.c
+++ b/ras-page-isolation.c
@@ -17,9 +17,13 @@
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
+#include <assert.h>
+#include <sys/types.h>
+#include <sys/wait.h>
 #include "ras-logger.h"
 #include "ras-page-isolation.h"
 
+#define MAX_PATH_LEN 64
 #define PARSED_ENV_LEN 50
 static const struct config threshold_units[] = {
 	{ "m",	1000 },
@@ -73,6 +77,8 @@ static const char *page_state[] = {
 
 static enum otype offline = OFFLINE_SOFT;
 static struct rb_root page_records;
+static char pre_notice[MAX_PATH_LEN];
+static char post_notice[MAX_PATH_LEN];
 
 static void page_offline_init(void)
 {
@@ -202,16 +208,94 @@ static void page_isolation_init(void)
 			threshold_string, cycle_string);
 }
 
+static void page_notice_init(void)
+{
+	char *notice_root = "/etc/rasdaemon_notices";
+	char *pre_re = getenv("PAGE_CE_OFFLINE_PRE_NOTICE");
+	char *post_re = getenv("PAGE_CE_OFFLINE_POST_NOTICE");
+
+	if (offline <= OFFLINE_ACCOUNT)
+		return;
+
+	snprintf(pre_notice, sizeof(pre_notice), "%s/%s", notice_root, pre_re);
+	if (access(pre_notice, R_OK|X_OK) < 0)
+		log(TERM, LOG_ERR, "cannot access page notice '%s'\n", pre_notice);
+
+	snprintf(post_notice, sizeof(post_notice), "%s/%s", notice_root, post_re);
+	if (access(post_notice, R_OK|X_OK) < 0)
+		log(TERM, LOG_ERR, "cannot access page notice '%s'\n", post_notice);
+}
+
 void ras_page_account_init(void)
 {
 	page_offline_init();
 	page_isolation_init();
+	page_notice_init();
+}
+
+static void finish_child(pid_t child, int status)
+{
+	if (WIFEXITED(status) && WEXITSTATUS(status)) {
+		log(TERM, LOG_INFO, "notice exited with status %d\n", WEXITSTATUS(status));
+	} else if (WIFSIGNALED(status)) {
+	        log(TERM, LOG_INFO,"notice died with signal %s\n", strsignal(WTERMSIG(status)));
+	}
+
+	return;
+}
+
+static void __run_notice(char *argv[], char **env)
+{
+	pid_t child;
+	int status;
+
+	child = fork();
+	if (child < 0) {
+		log(TERM, LOG_ERR, "Cannot create process for offline notice");
+                return;
+        }
+        if (child == 0) {
+                execve(argv[0], argv, env);
+                _exit(127);
+        }
+	else {
+		waitpid(child, &status, 0);
+		finish_child(child, status);
+        }
+}
+
+static void run_notice(char *argv[])
+{
+	int MAX_ENV = 20;
+	char *env[MAX_ENV];
+	int ei = 0;
+	int i;
+
+	asprintf(&env[ei++], "PATH=%s", getenv("PATH") ?: "/sbin:/usr/sbin:/bin:/usr/bin");
+	env[ei] = NULL;
+	assert(ei < MAX_ENV);
+
+	__run_notice(argv, env);
+
+	for (i = 0; i < ei; i++)
+		free(env[i]);
 }
 
 static int do_page_offline(unsigned long long addr, enum otype type)
 {
 	FILE *offline_file;
 	int err;
+	char *args;
+	char *argv[] = {
+		NULL,
+		NULL,
+		NULL,
+	};
+
+	asprintf(&args, "%llu", addr);
+	argv[0] = (char*)&pre_notice;
+	argv[1] = args;
+	run_notice(argv);
 
 	offline_file = fopen(kernel_offline[type], "w");
 	if (!offline_file)
@@ -221,6 +305,11 @@ static int do_page_offline(unsigned long long addr, enum otype type)
 	err = ferror(offline_file) ? -1 : 0;
 	fclose(offline_file);
 
+	argv[0] = (char*)&post_notice;
+	run_notice(argv);
+
+	free(args);
+
 	return err;
 }
 
@@ -329,4 +418,5 @@ void ras_record_page_error(unsigned long long addr, unsigned count, time_t time)
 			pr->start = time;
 		page_record(pr, count, time);
 	}
+
 }
-- 
2.27.0