Patchwork [v2] watchdog: fix ping mode failure

login
register
mail settings
Submitter rongqing.li@windriver.com
Date July 18, 2012, 1:41 a.m.
Message ID <1342575703-10784-1-git-send-email-rongqing.li@windriver.com>
Download mbox | patch
Permalink /patch/32339/
State Accepted
Commit 90f3a90413aa1e08c3206b838dcaee0c1c640dc7
Headers show

Comments

rongqing.li@windriver.com - July 18, 2012, 1:41 a.m.
From: Roy.Li <rongqing.li@windriver.com>

[YOCTO #2755]

When watchdog works on ping mode, the system will be rebooted since
watchdog can not receive the expected ECOREPLY on a setting interval.

Ping mode uses a raw socket to send a ECO packet, then uses select()
to wait and recvfrom() to receive the ECOREPLY packet, if select()
shows the data is ready, and the data is not the expected ECOREPLY,
and waiting time is not overdue, it will continue use select() and
recvfrom().

Problem is that the raw socket can receive any icmp packets, if we do
not set filters, and there are many icmp packets on socket, this
program will not find its interested ECOREPLY packet in a special
interval, which makes the ping mode fail.

Other program is that watchdog sometime can not reach the call of
recvfrom to try to receive packets since tv_sec of struct timeval
of select parameter is 0.

The timeout of select() is the result of ping interval minusing the
time of calling gettimeofday spending, when ping interval is 1 second,
and the call of gettimeofday() spends several useconds, the tv_sec of
struct timeval of select parameter must be 0, at that condition, we
should think it is valid of tv_sec of struct timeval of select parameter
be 0

Signed-off-by: Roy.Li <rongqing.li@windriver.com>
---
 .../watchdog/files/fix-ping-failure.patch          |   74 ++++++++++++++++++++
 meta/recipes-extended/watchdog/watchdog_5.11.bb    |    6 +-
 2 files changed, 78 insertions(+), 2 deletions(-)
 create mode 100644 meta/recipes-extended/watchdog/files/fix-ping-failure.patch
Saul Wold - July 20, 2012, 3:57 a.m.
On 07/17/2012 06:41 PM, rongqing.li@windriver.com wrote:
> From: Roy.Li <rongqing.li@windriver.com>
>
> [YOCTO #2755]
>
> When watchdog works on ping mode, the system will be rebooted since
> watchdog can not receive the expected ECOREPLY on a setting interval.
>
> Ping mode uses a raw socket to send a ECO packet, then uses select()
> to wait and recvfrom() to receive the ECOREPLY packet, if select()
> shows the data is ready, and the data is not the expected ECOREPLY,
> and waiting time is not overdue, it will continue use select() and
> recvfrom().
>
> Problem is that the raw socket can receive any icmp packets, if we do
> not set filters, and there are many icmp packets on socket, this
> program will not find its interested ECOREPLY packet in a special
> interval, which makes the ping mode fail.
>
> Other program is that watchdog sometime can not reach the call of
> recvfrom to try to receive packets since tv_sec of struct timeval
> of select parameter is 0.
>
> The timeout of select() is the result of ping interval minusing the
> time of calling gettimeofday spending, when ping interval is 1 second,
> and the call of gettimeofday() spends several useconds, the tv_sec of
> struct timeval of select parameter must be 0, at that condition, we
> should think it is valid of tv_sec of struct timeval of select parameter
> be 0
>
> Signed-off-by: Roy.Li <rongqing.li@windriver.com>
> ---
>   .../watchdog/files/fix-ping-failure.patch          |   74 ++++++++++++++++++++
>   meta/recipes-extended/watchdog/watchdog_5.11.bb    |    6 +-
>   2 files changed, 78 insertions(+), 2 deletions(-)
>   create mode 100644 meta/recipes-extended/watchdog/files/fix-ping-failure.patch
>
> diff --git a/meta/recipes-extended/watchdog/files/fix-ping-failure.patch b/meta/recipes-extended/watchdog/files/fix-ping-failure.patch
> new file mode 100644
> index 0000000..c6ee843
> --- /dev/null
> +++ b/meta/recipes-extended/watchdog/files/fix-ping-failure.patch
> @@ -0,0 +1,74 @@
> +Fix ping mode failure
> +
> +Upstream-Status: Pending
> +
> +When watchdog works on ping mode, the system will be rebooted since
> +watchdog can not receive the expected ECOREPLY on a setting interval.
> +
> +Ping mode uses a raw socket to send a ECO packet, then uses select()
> +to wait and recvfrom() to receive the ECOREPLY packet, if select()
> +shows the data is ready, and the data is not the expected ECOREPLY,
> +and waiting time is not overdue, it will continue use select() and
> +recvfrom().
> +
> +Problem is that the raw socket can receive any icmp packets, if we do
> +not set filters, and there are many icmp packets on socket, this
> +program will not find its interested ECOREPLY packet in a special
> +interval, which makes the ping mode fail.
> +
> +
> +Other program is that watchdog sometime can not reach the call of
> +recvfrom to try to receive packets since tv_sec of struct timeval
> +of select parameter is 0.
> +
> +The timeout of select() is the result of ping interval minusing the
> +time of calling gettimeofday spending, when ping interval is 1 second,
> +and the call of gettimeofday() spends several useconds, the tv_sec of
> +struct timeval of select parameter must be 0, at that condition, we
> +should it is valid of tv_sec of struct timeval of select parameter be 0
> +
> +Signed-off-by: Roy.Li <rongqing.li@windriver.com>
> +---
> + src/net.c      |    2 +-
> + src/watchdog.c |    5 ++++-
> + 2 files changed, 5 insertions(+), 2 deletions(-)
> +
> +--- a/src/net.c
> ++++ b/src/net.c
> +@@ -118,7 +118,7 @@ int check_net(char *target, int sock_fp,
> + 		    dtimeout.tv_usec -= 1000000;
> + 		    dtimeout.tv_sec++;
> + 		}
> +-		if (dtimeout.tv_sec <= 0)
> ++		if (dtimeout.tv_sec < 0)
> + 		    break;
> + #if USE_SYSLOG
> + 		if (verbose && logtick && ticker == 1)
> +--- a/src/watchdog.c
> ++++ b/src/watchdog.c
> +@@ -28,6 +28,7 @@
> + #include <sys/types.h>
> + #include <sys/ioctl.h>
> + #include <linux/oom.h>
> ++#include <linux/icmp.h>
> + #include <linux/watchdog.h>
> + #include <string.h>
> +
> +@@ -567,6 +568,8 @@ int main(int argc, char *const argv[])
> +     pid_t child_pid;
> +     int oom_adjusted = 0;
> +     struct stat s;
> ++    struct icmp_filter filt;
> ++    filt.data = ~(1<<ICMP_ECHOREPLY);
> +
> + #if USE_SYSLOG
> +     char *opts = "d:i:n:Ffsvbql:p:t:c:r:m:a:";
> +@@ -703,7 +706,7 @@ int main(int argc, char *const argv[])
> + 		perror(progname);
> + 		exit(1);
> + 	    }
> +-	
> ++	    setsockopt(net->sock_fp, SOL_RAW, ICMP_FILTER, (char*)&filt, sizeof(filt));
> + 	    /* this is necessary for broadcast pings to work */
> + 	    (void) setsockopt(net->sock_fp, SOL_SOCKET, SO_BROADCAST, (char *)&hold, sizeof(hold));
> +
> diff --git a/meta/recipes-extended/watchdog/watchdog_5.11.bb b/meta/recipes-extended/watchdog/watchdog_5.11.bb
> index 3c4f77b..6a611c9 100644
> --- a/meta/recipes-extended/watchdog/watchdog_5.11.bb
> +++ b/meta/recipes-extended/watchdog/watchdog_5.11.bb
> @@ -8,9 +8,11 @@ BUGTRACKER = "http://sourceforge.net/tracker/?group_id=172030&atid=860194"
>   LICENSE = "GPL-1.0+"
>   LIC_FILES_CHKSUM = "file://COPYING;md5=8a7258c60a71a2f04b67fb01f495889c"
>
> -PR = "r0"
> +PR = "r1"
> +
> +SRC_URI = "${SOURCEFORGE_MIRROR}/watchdog/watchdog_${PV}.tar.gz \
> +	   file://fix-ping-failure.patch"
>
> -SRC_URI = "${SOURCEFORGE_MIRROR}/watchdog/watchdog_${PV}.tar.gz"
>
>   SRC_URI[md5sum] = "02c764219b3bdb2373091cbd67109eb6"
>   SRC_URI[sha256sum] = "723a7966e0c3d58e3f4df20943a5c9aa1553381f46aa0dbcf832016756e62792"
>

Merged into OE-Core

Thanks
	Sau!

Patch

diff --git a/meta/recipes-extended/watchdog/files/fix-ping-failure.patch b/meta/recipes-extended/watchdog/files/fix-ping-failure.patch
new file mode 100644
index 0000000..c6ee843
--- /dev/null
+++ b/meta/recipes-extended/watchdog/files/fix-ping-failure.patch
@@ -0,0 +1,74 @@ 
+Fix ping mode failure
+
+Upstream-Status: Pending  
+
+When watchdog works on ping mode, the system will be rebooted since
+watchdog can not receive the expected ECOREPLY on a setting interval.
+
+Ping mode uses a raw socket to send a ECO packet, then uses select()
+to wait and recvfrom() to receive the ECOREPLY packet, if select()
+shows the data is ready, and the data is not the expected ECOREPLY,
+and waiting time is not overdue, it will continue use select() and
+recvfrom().
+
+Problem is that the raw socket can receive any icmp packets, if we do
+not set filters, and there are many icmp packets on socket, this
+program will not find its interested ECOREPLY packet in a special
+interval, which makes the ping mode fail.
+
+
+Other program is that watchdog sometime can not reach the call of
+recvfrom to try to receive packets since tv_sec of struct timeval
+of select parameter is 0.
+
+The timeout of select() is the result of ping interval minusing the
+time of calling gettimeofday spending, when ping interval is 1 second,
+and the call of gettimeofday() spends several useconds, the tv_sec of
+struct timeval of select parameter must be 0, at that condition, we
+should it is valid of tv_sec of struct timeval of select parameter be 0 
+
+Signed-off-by: Roy.Li <rongqing.li@windriver.com>
+---
+ src/net.c      |    2 +-
+ src/watchdog.c |    5 ++++-
+ 2 files changed, 5 insertions(+), 2 deletions(-)
+
+--- a/src/net.c
++++ b/src/net.c
+@@ -118,7 +118,7 @@ int check_net(char *target, int sock_fp,
+ 		    dtimeout.tv_usec -= 1000000;
+ 		    dtimeout.tv_sec++;
+ 		}
+-		if (dtimeout.tv_sec <= 0)
++		if (dtimeout.tv_sec < 0)
+ 		    break;
+ #if USE_SYSLOG
+ 		if (verbose && logtick && ticker == 1)
+--- a/src/watchdog.c
++++ b/src/watchdog.c
+@@ -28,6 +28,7 @@
+ #include <sys/types.h>
+ #include <sys/ioctl.h>
+ #include <linux/oom.h>
++#include <linux/icmp.h>
+ #include <linux/watchdog.h>
+ #include <string.h>
+ 
+@@ -567,6 +568,8 @@ int main(int argc, char *const argv[])
+     pid_t child_pid;
+     int oom_adjusted = 0;
+     struct stat s;
++    struct icmp_filter filt;
++    filt.data = ~(1<<ICMP_ECHOREPLY);
+ 
+ #if USE_SYSLOG
+     char *opts = "d:i:n:Ffsvbql:p:t:c:r:m:a:";
+@@ -703,7 +706,7 @@ int main(int argc, char *const argv[])
+ 		perror(progname);
+ 		exit(1);
+ 	    }
+-	    
++	    setsockopt(net->sock_fp, SOL_RAW, ICMP_FILTER, (char*)&filt, sizeof(filt));
+ 	    /* this is necessary for broadcast pings to work */
+ 	    (void) setsockopt(net->sock_fp, SOL_SOCKET, SO_BROADCAST, (char *)&hold, sizeof(hold));
+     
diff --git a/meta/recipes-extended/watchdog/watchdog_5.11.bb b/meta/recipes-extended/watchdog/watchdog_5.11.bb
index 3c4f77b..6a611c9 100644
--- a/meta/recipes-extended/watchdog/watchdog_5.11.bb
+++ b/meta/recipes-extended/watchdog/watchdog_5.11.bb
@@ -8,9 +8,11 @@  BUGTRACKER = "http://sourceforge.net/tracker/?group_id=172030&atid=860194"
 LICENSE = "GPL-1.0+"
 LIC_FILES_CHKSUM = "file://COPYING;md5=8a7258c60a71a2f04b67fb01f495889c"
 
-PR = "r0"
+PR = "r1"
+
+SRC_URI = "${SOURCEFORGE_MIRROR}/watchdog/watchdog_${PV}.tar.gz \
+	   file://fix-ping-failure.patch"
 
-SRC_URI = "${SOURCEFORGE_MIRROR}/watchdog/watchdog_${PV}.tar.gz"
 
 SRC_URI[md5sum] = "02c764219b3bdb2373091cbd67109eb6"
 SRC_URI[sha256sum] = "723a7966e0c3d58e3f4df20943a5c9aa1553381f46aa0dbcf832016756e62792"