Re: [PATCH] multipath: enable getting uevents through libudev

[Benjamin Marzinski <bmarzins@redhat.com>]

On Wed, Apr 04, 2012 at 01:12:45PM +0200, Zdenek Kabelac wrote:
> Dne 4.4.2012 01:42, Benjamin Marzinski napsal(a):
>> udev is removing support for RUN+="socket:..." rules. For now, I've kept
>> all the existing uevent code, but I've added a new method for getting the
>> uevent information using libudev that will be tried first.
>>
>> Signed-off-by: Benjamin Marzinski<bmarzins@redhat.com>
>> ---
>>   libmultipath/Makefile |    2
>>   libmultipath/uevent.c |  151 ++++++++++++++++++++++++++++++++++++++++++++------
>>   2 files changed, 134 insertions(+), 19 deletions(-)
>>
>> Index: multipath-tools-120123/libmultipath/Makefile
>> ===================================================================
>> --- multipath-tools-120123.orig/libmultipath/Makefile
>> +++ multipath-tools-120123/libmultipath/Makefile
>> @@ -7,7 +7,7 @@ include ../Makefile.inc
>>   SONAME=0
>>   DEVLIB = libmultipath.so
>>   LIBS = $(DEVLIB).$(SONAME)
>> -LIBDEPS = -lpthread -ldl -ldevmapper
>> +LIBDEPS = -lpthread -ldl -ldevmapper -ludev
>>
>>   OBJS = memory.o parser.o vector.o devmapper.o callout.o \
>>          hwtable.o blacklist.o util.o dmparser.o config.o \
>> Index: multipath-tools-120123/libmultipath/uevent.c
>> ===================================================================
>> --- multipath-tools-120123.orig/libmultipath/uevent.c
>> +++ multipath-tools-120123/libmultipath/uevent.c
>> @@ -39,6 +39,7 @@
>>   #include<pthread.h>
>>   #include<limits.h>
>>   #include<sys/mman.h>
>> +#include<libudev.h>
>>   #include<errno.h>
>>
>>   #include "memory.h"
>> @@ -161,7 +162,7 @@ int uevent_dispatch(int (*uev_trigger)(s
>>   	return 0;
>>   }
>>
>> -int uevent_listen(void)
>> +int failback_listen(void)
>>   {
>>   	int sock;
>>   	struct sockaddr_nl snl;
>> @@ -173,20 +174,6 @@ int uevent_listen(void)
>>   	int rcvszsz = sizeof(rcvsz);
>>   	unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
>>   	const int feature_on = 1;
>> -
>> -	/*
>> -	 * Queue uevents for service by dedicated thread so that the uevent
>> -	 * listening thread does not block on multipathd locks (vecs->lock)
>> -	 * thereby not getting to empty the socket's receive buffer queue
>> -	 * often enough.
>> -	 */
>> -	INIT_LIST_HEAD(&uevq);
>> -
>> -	pthread_mutex_init(uevq_lockp, NULL);
>> -	pthread_cond_init(uev_condp, NULL);
>> -
>> -	pthread_cleanup_push(uevq_stop, NULL);
>> -
>>   	/*
>>   	 * First check whether we have a udev socket
>>   	 */
>> @@ -382,13 +369,141 @@ int uevent_listen(void)
>>
>>   exit:
>>   	close(sock);
>> +	return 1;
>> +}
>>
>> -	pthread_cleanup_pop(1);
>> +int uevent_listen(void)
>> +{
>> +	int err;
>> +	struct udev *udev = NULL;
>> +	struct udev_monitor *monitor = NULL;
>> +	int fd, socket_flags;
>> +	int need_failback = 0;
>> +	/*
>> +	 * Queue uevents for service by dedicated thread so that the uevent
>> +	 * listening thread does not block on multipathd locks (vecs->lock)
>> +	 * thereby not getting to empty the socket's receive buffer queue
>> +	 * often enough.
>> +	 */
>> +	INIT_LIST_HEAD(&uevq);
>> +
>> +	pthread_mutex_init(uevq_lockp, NULL);
>> +	pthread_cond_init(uev_condp, NULL);
>> +	pthread_cleanup_push(uevq_stop, NULL);
>> +
>> +	udev = udev_new();
>> +	if (!udev) {
>> +		condlog(2, "failed to create udev context");
>> +		need_failback = 1;
>> +		goto out;
>> +	}
>> +	monitor = udev_monitor_new_from_netlink(udev, "udev");
>> +	if (!monitor) {
>> +		condlog(2, "failed to create udev monitor");
>> +		need_failback = 1;
>> +		goto out;
>> +	}
>> +	if (udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024))
>> +		condlog(2, "failed to increase buffer size");
>> +	fd = udev_monitor_get_fd(monitor);
>> +	socket_flags = fcntl(fd, F_GETFL);
>> +	if (socket_flags<  0) {
>> +		condlog(2, "failed to get monitor socket flags : %s",
>> +			strerror(errno));
>> +		need_failback = 1;
>> +		goto out;
>> +	}
>> +	if (fcntl(fd, F_SETFL, socket_flags&  ~O_NONBLOCK)<  0) {
>> +		condlog(2, "failed to set monitor socket flags : %s",
>> +			strerror(errno));
>> +		need_failback = 1;
>> +		goto out;
>> +	}
>> +	err = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block",
>> +							      NULL);
>> +	if (err)
>> +		condlog(2, "failed to create filter : %s\n", strerror(-err));
>> +	err = udev_monitor_enable_receiving(monitor);
>> +	if (err) {
>> +		condlog(2, "failed to enable receiving : %s\n", strerror(-err));
>> +		need_failback = 1;
>> +		goto out;
>> +	}
>> +	while (1) {
>> +		int i = 0;
>> +		char *pos, *end;
>> +		struct uevent *uev;
>> +		struct udev_device *dev;
>> +                struct udev_list_entry *list_entry;
>> +
>> +		dev = udev_monitor_receive_device(monitor);
>> +		if (!dev) {
>> +			condlog(0, "failed getting udev device");
>> +			continue;
>> +		}
>>
>> +		uev = alloc_uevent();
>> +		if (!uev) {
>> +			condlog(1, "lost uevent, oom");
>> +			continue;
>> +		}
>> +		pos = uev->buffer;
>> +		end = pos + HOTPLUG_BUFFER_SIZE + OBJECT_SIZE - 1;
>> +		udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(dev)) {
>> +			const char *name, *value;
>> +			int bytes;
>> +
>> +			name = udev_list_entry_get_name(list_entry);
>> +			value = udev_list_entry_get_value(list_entry);
>> +			bytes = snprintf(pos, end - pos, "%s=%s", name,
>
> I'd recommend to validate all input going from libudev agains NULL pointers:
>
> (i.e. look at https://bugzilla.redhat.com/show_bug.cgi?id=809576)

Looking at the udev code and the Libudev reference manual, most of my
checking decisions seem reasonable. I do more checking that udevd and
udevadm do. But I suppose it doesn't hurt to do some more.

I don't see how, given my previous checking, udev_monitor_get_fd() could
fail, but that function can return a failure, and it's no pain to check it.

Also, udev_list_entry_get_name() and udev_list_entry_get_value() can
return NULL, although I don't think it's possible here given that I
check and have a reference on the device, udev_list_entry_foreach()
cannot return a NULL list_entry, and I'm looking at a list of properties
which AFAIK must have a value. None of the udev code checks for NULL
here itself. But again, there's no compelling reason not to check for
NULL.

I'll resend this patch with the additional checks. 

-Ben

>
> Zdenek