StorageWorks multipath support

StorageWorks multipath support

[Christophe Varoqui]

Hello,

here is a note describing the lego blocks needed to have a fully operational multipath stack with StorageWorks and Qlogic hardware.

Kernel :
========

2.6.12-rc6

Additional patches :
====================

Qlogic : LU scaning fix
http://marc.theaimsgroup.com/?l=linux-scsi&m=111836291000433&w=2

Qlogic : Postpone fc_rport block/unblock to scheduled work
http://marc.theaimsgroup.com/?l=linux-scsi&m=111868700022257&w=2

Mike Christie : export block layer function
https://www.redhat.com/archives/dm-devel/2005-June/msg00084.html

Hardware Handler module :
=========================

Mike Christie, Christophe Varoqui : 
http://christophe.varoqui.free.fr/multipath-tools/dm-hp-sw.tar.bz2

Multipath Tools :
=================

git HEAD up to commit fd9857d2eb9796d0a205887bdff879d56355eda2 at minimum :
http://kernel.org/pub/scm/linux/storage/multipath-tools/.git

Configlet :
===========

device {
        vendor                  "COMPAQ"
        product                 "HSV110 (C)COMPAQ"
        path_grouping_policy    group_by_serial
        path_checker            hp_sw
        features                "1 queue_if_no_path"
        hardware_handler        "1 hp_sw"
}


It would be good to have the kernel patches pushed to 2.6.12 before it comes out.

Happy testing,
cvaroqui

Re: StorageWorks multipath support

[Lars Marowsky-Bree]

On 2005-06-15T10:56:34, Christophe Varoqui <christophe.varoqui@free.fr> wrote:

> Hello,
> 
> here is a note describing the lego blocks needed to have a fully
> operational multipath stack with StorageWorks and Qlogic hardware.

Christophe, is the "ghost" denomination there to stay?

If so, the passive paths (ie, to the non-active Service Processor) of
the EMC CLARiiON also might wish to use this...


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"

Re: StorageWorks multipath support

[Christophe Varoqui]

On Wed, Jun 15, 2005 at 03:15:20PM +0200, Lars Marowsky-Bree wrote:
> On 2005-06-15T10:56:34, Christophe Varoqui <christophe.varoqui@free.fr> wrote:
> 
> > Hello,
> > 
> > here is a note describing the lego blocks needed to have a fully
> > operational multipath stack with StorageWorks and Qlogic hardware.
> 
> Christophe, is the "ghost" denomination there to stay?
> 
> If so, the passive paths (ie, to the non-active Service Processor) of
> the EMC CLARiiON also might wish to use this...
> 
I'm not attached to the ghost term, but what it describes is here to stay.

This state has a special treatment in the daemon : paths in this state are kept in active state in the DM, instead of proactively being failed. This is for the hwhandler to be given a chance to activate a PG with only ghost paths in it.

If you come with a better term, I'll sure consider it. Maybe "spare" ?

Regards,
cvaroqui

Re: StorageWorks multipath support

[Lars Marowsky-Bree]

On 2005-06-15T15:45:36, Christophe Varoqui <christophe.varoqui@free.fr> wrote:

> I'm not attached to the ghost term, but what it describes is here to
> stay.

OK.

> This state has a special treatment in the daemon : paths in this state
> are kept in active state in the DM, instead of proactively being
> failed. This is for the hwhandler to be given a chance to activate a
> PG with only ghost paths in it.

Well, before, those paths were simply kept as "active" / "healthy" (and
from the point of view of the kernel code they are, just that the PG
they are in might not be enabled).

But it makes the status output look nicer of course if those paths are
listed as "ghost" in the user-space.

> If you come with a better term, I'll sure consider it. Maybe "spare" ?

Well, "ghost" or "passive" or "spare", I don't care ;-) I was just
asking.

BTW, backporting uevent is a bit more annoying then I thought. How about
a way to manually prod the daemon? That might be useful in general, for
example if a hotplug event was lost due to whatever reason...


-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"

Re: StorageWorks multipath support

[Christophe Varoqui]

On Wed, Jun 15, 2005 at 03:53:37PM +0200, Lars Marowsky-Bree wrote:
> 
> Well, before, those paths were simply kept as "active" / "healthy" (and
> from the point of view of the kernel code they are, just that the PG
> they are in might not be enabled).
> 
Mmm, not anymore.

It was true till 0.4.4, but in HEAD there are patches to implement to proactive DM-failing of checker-failed paths. This was a request from Tran if I remember.

> But it makes the status output look nicer of course if those paths are
> listed as "ghost" in the user-space.
> 
Agreed

> 
> BTW, backporting uevent is a bit more annoying then I thought. How about
> a way to manually prod the daemon? That might be useful in general, for
> example if a hotplug event was lost due to whatever reason...
> 
Too bad.

Can you elaborate on what you have in mind ?

Regards,
cvaroqui

Re: StorageWorks multipath support

[Lars Marowsky-Bree]

On 2005-06-15T16:07:40, Christophe Varoqui <christophe.varoqui@free.fr> wrote:

> > Well, before, those paths were simply kept as "active" / "healthy" (and
> > from the point of view of the kernel code they are, just that the PG
> > they are in might not be enabled).
> > 
> Mmm, not anymore.
> 
> It was true till 0.4.4, but in HEAD there are patches to implement to proactive DM-failing of checker-failed paths. This was a request from Tran if I remember.

The emc_clariion checker considers the "ghost" paths to be healthy
because of that. (Which is true: they haven't failed, they are just
passive.)

That failed paths are not proactively failed is an orthogonal issue, and
a change I like ;-)

> > BTW, backporting uevent is a bit more annoying then I thought. How about
> > a way to manually prod the daemon? That might be useful in general, for
> > example if a hotplug event was lost due to whatever reason...
> > 
> Too bad.
> 
> Can you elaborate on what you have in mind ?

Well, basically I'd like it if a SIGHUP would still cause the daemon to
reload the configuration and rescan the mappings. (Which is a common use
of SIGHUP.) Then we'd get the best of both worlds, we could re-trigger
it on the old dists and use the more efficient uevent on newer kernels.


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"

Re: StorageWorks multipath support

[Alasdair G Kergon]

On Wed, Jun 15, 2005 at 04:07:40PM +0200, christophe varoqui wrote:
> Can you elaborate on what you have in mind ?
 
Listen on a socket in the filesystem?  [cf. what screen does]
And have a separate helper program that sends messages down it
similar to the uevent messages.

Alasdair
-- 
agk@redhat.com

Re: StorageWorks multipath support

[Lars Marowsky-Bree]

On 2005-06-15T15:20:52, Alasdair G Kergon <agk@redhat.com> wrote:

> > Can you elaborate on what you have in mind ?
> Listen on a socket in the filesystem?  [cf. what screen does]
> And have a separate helper program that sends messages down it
> similar to the uevent messages.

That would of course also work...



-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"

Re: StorageWorks multipath support

[Christophe Varoqui]

On Wed, Jun 15, 2005 at 04:13:02PM +0200, Lars Marowsky-Bree wrote:
> 
> The emc_clariion checker considers the "ghost" paths to be healthy
> because of that. (Which is true: they haven't failed, they are just
> passive.)
> 
ah ok, I missed that refinement in the emc checker.
Do you want to send the patch to move this checker to a tri-state like hp_sw ?

> > 
> > Can you elaborate on what you have in mind ?
> 
> Well, basically I'd like it if a SIGHUP would still cause the daemon to
> reload the configuration and rescan the mappings. (Which is a common use
> of SIGHUP.) Then we'd get the best of both worlds, we could re-trigger
> it on the old dists and use the more efficient uevent on newer kernels.
> 
Well, seems a reasonable feature indeed. I defered implementing this because you are almost equally served with a daemon restart.

Do you consider Alasdair's proposition (fs-socket) complementary, best, ... ?

Regards,
cvaroqui

Re: StorageWorks multipath support

[christophe varoqui]

[-- Attachment #1: Type: text/plain, Size: 1639 bytes --]

On mer, 2005-06-15 at 15:20 +0100, Alasdair G Kergon wrote:
> On Wed, Jun 15, 2005 at 04:07:40PM +0200, christophe varoqui wrote:
> > Can you elaborate on what you have in mind ?
>  
> Listen on a socket in the filesystem?  [cf. what screen does]
> And have a separate helper program that sends messages down it
> similar to the uevent messages.
> 
Would the following do what you want ?

The client is wrapped in the daemon.
You can call it through :
- multipathd -k"$cmd" : one shot
- multipathd -k : interactive

The synthax is somewhat rude :/

- lp : list paths
- lm : list maps
- ap $path : add path checker, $path is in the "sdb" form
- rp $path : remove path checker
- am $map : add map event thread, $map is in the "dm-0" form
- rm $map : remove map event thread

For your pleasure, here is a sample session ...

xa-s05:~/scm/test/multipathd# ./multipathd -k
lp

       sdc: state 2,   2/ 28 XX..........................
       sde: state 2,   2/ 28 XX..........................
       sdb: state 2,   9/ 28 XXXXXXXXX...................
       sdd: state 2,  27/ 28 XXXXXXXXXXXXXXXXXXXXXXXXXXX.

rp sdd
ok

lp

       sdc: state 2,  18/ 28 XXXXXXXXXXXXXXXXXX..........
       sde: state 2,  18/ 28 XXXXXXXXXXXXXXXXXX..........
       sdb: state 2,  25/ 28 XXXXXXXXXXXXXXXXXXXXXXXXX...

ap sdd
ok

lp

       sdc: state 2,  12/ 28 XXXXXXXXXXXX................
       sde: state 2,  12/ 28 XXXXXXXXXXXX................
       sdb: state 2,  19/ 28 XXXXXXXXXXXXXXXXXXX.........
       sdd: state 2,   7/  7 XXXXXXX



Let me know if I should commit this as a foundation.

Regards,
-- 
christophe varoqui <christophe.varoqui@free.fr>

[-- Attachment #2: uxsock.patch --]
[-- Type: text/x-patch, Size: 12363 bytes --]

libmultipath/Makefile: needs update
libmultipath/dmparser.c: needs update
libmultipath/util.c: needs update
libmultipath/util.h: needs update
multipathd/Makefile: needs update
multipathd/main.c: needs update
Index: libmultipath/Makefile
===================================================================
--- 3675ec950a3806374cce8caa81ff064c0513b382/libmultipath/Makefile  (mode:100644)
+++ uncommitted/libmultipath/Makefile  (mode:100644)
@@ -10,7 +10,7 @@
        hwtable.o blacklist.o util.o dmparser.o config.o \
        structs.o cache.o discovery.o propsel.o dict.o \
        pgpolicies.o debug.o regex.o defaults.o uevent.o \
-       switchgroup.o
+       switchgroup.o uxsock.o
 
 CFLAGS = -pipe -g -Wall -Wunused -Wstrict-prototypes
 
Index: libmultipath/dmparser.c
===================================================================
--- 3675ec950a3806374cce8caa81ff064c0513b382/libmultipath/dmparser.c  (mode:100644)
+++ uncommitted/libmultipath/dmparser.c  (mode:100644)
@@ -16,45 +16,6 @@
 #define WORD_SIZE 64
 
 static int
-get_word (char * sentence, char ** word)
-{
-	char * p;
-	int len;
-	int skip = 0;
-	
-	while (*sentence ==  ' ') {
-		sentence++;
-		skip++;
-	}
-	if (*sentence == '\0')
-		return 0;
-
-	p = sentence;
-
-	while (*p !=  ' ' && *p != '\0')
-		p++;
-
-	len = (int) (p - sentence);
-
-	if (!word)
-		return skip + len;
-
-	*word = MALLOC(len + 1);
-
-	if (!*word) {
-		condlog(0, "get_word : oom\n");
-		return 0;
-	}
-	strncpy(*word, sentence, len);
-	condlog(4, "*word = %s, len = %i", *word, len);
-
-	if (*p == '\0')
-		return 0;
-
-	return skip + len;
-}
-
-static int
 merge_words (char ** dst, char * word, int space)
 {
 	char * p;
Index: libmultipath/util.c
===================================================================
--- 3675ec950a3806374cce8caa81ff064c0513b382/libmultipath/util.c  (mode:100644)
+++ uncommitted/libmultipath/util.c  (mode:100644)
@@ -4,6 +4,9 @@
 #include <sys/stat.h>
 #include <unistd.h>
 
+#include "debug.h"
+#include "memory.h"
+
 #define PARAMS_SIZE 255
 
 int
@@ -49,3 +52,42 @@
 	return 0;
 }
 
+int
+get_word (char * sentence, char ** word)
+{
+	char * p;
+	int len;
+	int skip = 0;
+	
+	while (*sentence ==  ' ') {
+		sentence++;
+		skip++;
+	}
+	if (*sentence == '\0')
+		return 0;
+
+	p = sentence;
+
+	while (*p !=  ' ' && *p != '\0')
+		p++;
+
+	len = (int) (p - sentence);
+
+	if (!word)
+		return skip + len;
+
+	*word = MALLOC(len + 1);
+
+	if (!*word) {
+		condlog(0, "get_word : oom\n");
+		return 0;
+	}
+	strncpy(*word, sentence, len);
+	condlog(4, "*word = %s, len = %i", *word, len);
+
+	if (*p == '\0')
+		return 0;
+
+	return skip + len;
+}
+
Index: libmultipath/util.h
===================================================================
--- 3675ec950a3806374cce8caa81ff064c0513b382/libmultipath/util.h  (mode:100644)
+++ uncommitted/libmultipath/util.h  (mode:100644)
@@ -4,6 +4,8 @@
 int strcmp_chomp(char *, char *);
 void basename (char * src, char * dst);
 int filepresent (char * run);
+int get_word (char * sentence, char ** word);
+
 
 #define safe_sprintf(var, format, args...)	\
 	snprintf(var, sizeof(var), format, ##args) >= sizeof(var)
Index: multipathd/Makefile
===================================================================
--- 3675ec950a3806374cce8caa81ff064c0513b382/multipathd/Makefile  (mode:100644)
+++ uncommitted/multipathd/Makefile  (mode:100644)
@@ -26,9 +26,8 @@
 #
 # object files
 #
-OBJS = main.o copy.o log.o log_pthread.o pidfile.o \
-       $(MULTIPATHLIB)-glibc.a \
-       $(CHECKERSLIB)-glibc.a \
+OBJS = main.o copy.o log.o log_pthread.o pidfile.o uxlsnr.o uxclnt.o \
+       $(MULTIPATHLIB)-glibc.a $(CHECKERSLIB)-glibc.a \
 
 
 #
Index: multipathd/main.c
===================================================================
--- 3675ec950a3806374cce8caa81ff064c0513b382/multipathd/main.c  (mode:100644)
+++ uncommitted/multipathd/main.c  (mode:100644)
@@ -53,9 +53,12 @@
 #include "copy.h"
 #include "clone_platform.h"
 #include "pidfile.h"
+#include "uxlsnr.h"
+#include "uxclnt.h"
 
 #define FILE_NAME_SIZE 256
 #define CMDSIZE 160
+#define MAX_REPLY_LEN 1000
 
 #define CALLOUT_DIR "/var/cache/multipathd"
 
@@ -561,6 +564,115 @@
 	return 0;
 }
 
+static char *
+show_paths (struct paths * allpaths)
+{
+	int i, j, k;
+	struct path * pp;
+	char * c;
+	char * reply;
+
+	reply = MALLOC(MAX_REPLY_LEN);
+
+	if (!reply)
+		return NULL;
+
+	c = reply;
+	c += sprintf(c, "\n");
+
+	vector_foreach_slot(allpaths->pathvec, pp, i) {
+		c += sprintf(c, "%10s: ", pp->dev);
+		c += sprintf(c, "state %i, ", pp->state);
+
+		j = pp->tick;
+		k = pp->checkint - pp->tick;
+		c += sprintf(c, "%3i/%3i ", j, pp->checkint);
+
+		while (j-- > 0)
+			c += sprintf(c, "X");
+
+
+		while (k-- > 0)
+			c += sprintf(c, ".");
+
+		c += sprintf(c, "\n");
+	}
+
+	return reply;
+}
+
+static char *
+show_maps (struct paths * allpaths)
+{
+	int i, j, k;
+	struct multipath * mpp;
+	char * c;
+	char * reply;
+
+	reply = MALLOC(MAX_REPLY_LEN);
+
+	if (!reply)
+		return NULL;
+
+	c = reply;
+	c += sprintf(c, "\n");
+
+	vector_foreach_slot(allpaths->mpvec, mpp, i) {
+		c += sprintf(c, "%20s: ", mpp->alias);
+
+		j = mpp->failback_tick;
+		k = mpp->pgfailback - mpp->failback_tick;
+		c += sprintf(c, "%3i/%3i ", j, mpp->pgfailback);
+
+		while (j-- > 0)
+			c += sprintf(c, "X");
+
+
+		while (k-- > 0)
+			c += sprintf(c, ".");
+
+		c += sprintf(c, "\n");
+	}
+
+	return reply;
+}
+
+char *
+uxsock_trigger (char * str, void * trigger_data)
+{
+	struct paths * allpaths;
+	char * reply = NULL;
+
+	allpaths = (struct paths *)trigger_data;
+
+	lock(allpaths->lock);
+
+	if (*str == 'l' && *(str + 1) == 'p')
+		reply = show_paths(allpaths);
+
+	else if (*str == 'l' && *(str + 1) == 'm')
+		reply = show_maps(allpaths);
+
+	else if (*str == 'r' && *(str + 1) == 'p')
+		uev_remove_path(str + 3, allpaths);
+
+	else if (*str == 'a' && *(str + 1) == 'p')
+		uev_add_path(str + 3, allpaths);
+
+	else if (*str == 'r' && *(str + 1) == 'm')
+		uev_remove_map(str + 3, allpaths);
+
+	else if (*str == 'a' && *(str + 1) == 'm')
+		uev_add_map(str + 3, allpaths);
+
+	if (!reply)
+		asprintf(&reply, "ok\n");
+
+	unlock(allpaths->lock);
+
+	return reply;
+}
+
 int 
 uev_trigger (struct uevent * uev, void * trigger_data)
 {
@@ -622,6 +734,14 @@
 	return NULL;
 }
 
+static void *
+uxlsnrloop (void * ap)
+{
+	uxsock_listen(&uxsock_trigger, ap);
+
+	return NULL;
+}
+
 static void
 strvec_free (vector vec)
 {
@@ -1045,7 +1165,7 @@
 static int
 child (void * param)
 {
-	pthread_t check_thr, uevent_thr;
+	pthread_t check_thr, uevent_thr, uxlsnr_thr;
 	pthread_attr_t attr;
 	struct paths * allpaths;
 
@@ -1120,8 +1240,10 @@
 	
 	pthread_create(&check_thr, &attr, checkerloop, allpaths);
 	pthread_create(&uevent_thr, &attr, ueventloop, allpaths);
+	pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, allpaths);
 	pthread_join(check_thr, NULL);
 	pthread_join(uevent_thr, NULL);
+	pthread_join(uxlsnr_thr, NULL);
 
 	return 0;
 }
@@ -1156,7 +1278,7 @@
 	if (!conf)
 		exit(1);
 
-	while ((arg = getopt(argc, argv, ":dv:")) != EOF ) {
+	while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
 	switch(arg) {
 		case 'd':
 			logsink = 0;
@@ -1168,6 +1290,9 @@
 
 			conf->verbosity = atoi(optarg);
 			break;
+		case 'k':
+			uxclnt(optarg);
+			exit(0);
 		default:
 			;
 		}
Index: multipathd/uxclnt.c
===================================================================
--- /dev/null  (tree:3675ec950a3806374cce8caa81ff064c0513b382)
+++ uncommitted/multipathd/uxclnt.c  (mode:100644)
@@ -0,0 +1,69 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <stdarg.h>
+#include <fcntl.h>
+#include <sys/ioctl.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+#include <sys/poll.h>
+
+#include <uxsock.h>
+
+/*
+ * process the client 
+ */
+static void process(int fd)
+{
+	char line[1000];
+	char *reply;
+
+	while (fgets(line, sizeof(line), stdin)) {
+		size_t len = strlen(line);
+
+		if (line[len-1] == '\n') {
+			line[len-1] = 0;
+			len--;
+		}
+		
+		if (send_packet(fd, line, strlen(line)) != 0) break;
+		if (recv_packet(fd, &reply, &len) != 0) break;
+
+		printf("%*.*s\n", (int)len, (int)len, reply);
+		free(reply);
+	}
+}
+
+static void process_req(int fd, char * inbuf)
+{
+	char *reply;
+	size_t len;
+
+	send_packet(fd, inbuf, strlen(inbuf));
+	recv_packet(fd, &reply, &len);
+
+	printf("%*.*s\n", (int)len, (int)len, reply);
+	free(reply);
+}
+	
+/*
+ * entry point
+ */
+int uxclnt(char * inbuf)
+{
+	int fd;
+
+	fd = ux_socket_connect(SOCKET_NAME);
+	if (fd == -1) {
+		perror("ux_socket_connect");
+		exit(1);
+	}
+
+	if (inbuf)
+		process_req(fd, inbuf);
+	else
+		process(fd);
+	
+	return 0;
+}
Index: multipathd/uxclnt.h
===================================================================
--- /dev/null  (tree:3675ec950a3806374cce8caa81ff064c0513b382)
+++ uncommitted/multipathd/uxclnt.h  (mode:100644)
@@ -0,0 +1 @@
+int uxclnt(char * inbuf);
Index: multipathd/uxlsnr.c
===================================================================
--- /dev/null  (tree:3675ec950a3806374cce8caa81ff064c0513b382)
+++ uncommitted/multipathd/uxlsnr.c  (mode:100644)
@@ -0,0 +1,148 @@
+/*
+ * A simple domain socket listener
+ * Original author : tridge@samba.org, January 2002
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <stdarg.h>
+#include <fcntl.h>
+#include <sys/ioctl.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+#include <sys/poll.h>
+
+#include <memory.h>
+#include <debug.h>
+#include <vector.h>
+#include <structs.h>
+#include <uxsock.h>
+
+#define SLEEP_TIME 5000
+
+struct client {
+	int fd;
+	struct client *next, *prev;
+};
+
+static struct client *clients;
+static unsigned num_clients;
+
+/*
+ * handle a new client joining
+ */
+static void new_client(int ux_sock)
+{
+	struct client *c;
+	struct sockaddr addr;
+	socklen_t len = sizeof(addr);
+	int fd;
+
+	fd = accept(ux_sock, &addr, &len);
+	
+	if (fd == -1)
+		return;
+
+	/* put it in our linked list */
+	c = (struct client *)MALLOC(sizeof(*c));
+	memset(c, 0, sizeof(*c));
+	c->fd = fd;
+	c->next = clients;
+	clients = c;
+	num_clients++;
+}
+
+/*
+ * kill off a dead client
+ */
+static void dead_client(struct client *c)
+{
+	close(c->fd);
+	if (c->prev) c->prev->next = c->next;
+	if (c->next) c->next->prev = c->prev;
+	if (c == clients) clients = c->next;
+	FREE(c);
+	num_clients--;
+}
+
+/*
+ * entry point
+ */
+void * uxsock_listen(char * (*uxsock_trigger)(char *, void * trigger_data),
+			void * trigger_data)
+{
+	int ux_sock;
+	size_t len;
+	char *inbuf;
+	struct pollfd *polls = NULL;
+	char *reply;
+
+	ux_sock = ux_socket_listen(SOCKET_NAME);
+
+	if (ux_sock == -1) {
+		condlog(0, "ux_socket_listen error");
+		exit(1);
+	}
+
+	while (1) {
+		struct client *c;
+		int i, poll_count;
+
+		/* setup for a poll */
+		polls = REALLOC(polls, (1+num_clients) * sizeof(*polls));
+		polls[0].fd = ux_sock;
+		polls[0].events = POLLIN;
+
+		/* setup the clients */
+		for (i=1, c = clients; c; i++, c = c->next) {
+			polls[i].fd = c->fd;
+			polls[i].events = POLLIN;
+		}
+
+		/* most of our life is spent in this call */
+		poll_count = poll(polls, i, SLEEP_TIME);
+		
+		if (poll_count == -1) {
+			/* something went badly wrong! */
+			condlog(0, "poll");
+			exit(1);
+		}
+
+		if (poll_count == 0)
+			continue;
+
+		/* see if a client wants to speak to us */
+		for (i=1, c = clients; c; i++) {
+			struct client *next = c->next;
+
+			if (polls[i].revents & POLLIN) {
+				if (recv_packet(c->fd, &inbuf, &len) != 0) {
+					dead_client(c);
+				} else {
+					condlog(4, "Got request [%*.*s]",
+						(int)len, (int)len, inbuf);
+					reply = uxsock_trigger(inbuf,
+							trigger_data);
+
+					if (reply) {
+						if (send_packet(c->fd, reply,
+							strlen(reply)) != 0) {
+							dead_client(c);
+						}
+						FREE(reply);
+					}
+					FREE(inbuf);
+				}
+			}
+			c = next;
+		}
+
+		/* see if we got a new client */
+		if (polls[0].revents & POLLIN) {
+			new_client(ux_sock);
+		}
+	}
+
+	return NULL;
+}
Index: multipathd/uxlsnr.h
===================================================================
--- /dev/null  (tree:3675ec950a3806374cce8caa81ff064c0513b382)
+++ uncommitted/multipathd/uxlsnr.h  (mode:100644)
@@ -0,0 +1,4 @@
+void * uxsock_listen(char * (*uxsock_trigger)
+			(char *, void * trigger_data),
+			void * trigger_data);
+

[-- Attachment #3: Type: text/plain, Size: 0 bytes --]

Re: StorageWorks multipath support

[christophe varoqui]

On sam, 2005-06-18 at 19:35 +0200, christophe varoqui wrote:
> On mer, 2005-06-15 at 15:20 +0100, Alasdair G Kergon wrote:
> > On Wed, Jun 15, 2005 at 04:07:40PM +0200, christophe varoqui wrote:
> > > Can you elaborate on what you have in mind ?
> >  
> > Listen on a socket in the filesystem?  [cf. what screen does]
> > And have a separate helper program that sends messages down it
> > similar to the uevent messages.
> > 
> Would the following do what you want ?
> 
No response ? LinuxTag vielleicht ?
Well, it must mean yes then :)

I commited the thing, plus readline support and a few fixes for bugs
uncovered by this nice new feature.

The parser, and the synthax, are still ugly as hell. If someone cares,
he should point me to a nice one.

I'm also open to "lp" and "lm" output format/content suggestions.

Regards,
-- 
christophe varoqui <christophe.varoqui@free.fr>