xenstore.c

/******************************************************************************
 * xenstore.c
 *
 * Low-level kernel interface to the XenStore.
 *
 * Copyright (C) 2005 Rusty Russell, IBM Corporation
 * Copyright (C) 2009,2010 Spectra Logic Corporation
 *
 * This file may be distributed separately from the Linux kernel, or
 * incorporated into other software packages, subject to the following license:
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this source file (the "Software"), to deal in the Software without
 * restriction, including without limitation the rights to use, copy, modify,
 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
 * and to permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */


#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");

#include <sys/param.h>
#include <sys/bus.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/module.h>
#include <sys/mutex.h>
#include <sys/sx.h>
#include <sys/syslog.h>
#include <sys/malloc.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/kthread.h>
#include <sys/sbuf.h>
#include <sys/sysctl.h>
#include <sys/uio.h>
#include <sys/unistd.h>

#include <machine/xen/xen-os.h>
#include <machine/stdarg.h>

#include <xen/evtchn.h>
#include <xen/gnttab.h>
#include <xen/hypervisor.h>
#include <xen/xen_intr.h>

#include <xen/interface/hvm/params.h>

#include <xen/xenstore/xenstorevar.h>
#include <xen/xenstore/xenstore_internal.h>

#include <vm/vm.h>
#include <vm/pmap.h>

/**
 * \file xenstore.c
 * \brief XenStore interface
 *
 * The XenStore interface is a simple storage system that is a means of
 * communicating state and configuration data between the Xen Domain 0
 * and the various guest domains.  All configuration data other than
 * a small amount of essential information required during the early
 * boot process of launching a Xen aware guest, is managed using the
 * XenStore.
 *
 * The XenStore is ASCII string based, and has a structure and semantics
 * similar to a filesystem.  There are files and directories, the directories
 * able to contain files or other directories.  The depth of the hierachy
 * is only limited by the XenStore's maximum path length.
 *
 * The communication channel between the XenStore service and other
 * domains is via two, guest specific, ring buffers in a shared memory
 * area.  One ring buffer is used for communicating in each direction.
 * The grant table references for this shared memory are given to the
 * guest either via the xen_start_info structure for a fully para-
 * virtualized guest, or via HVM hypercalls for a hardware virtualized
 * guest.
 *
 * The XenStore communication relies on an event channel and thus
 * interrupts.  For this reason, the attachment of the XenStore
 * relies on an interrupt driven configuration hook to hold off
 * boot processing until communication with the XenStore service
 * can be established.
 *
 * Several Xen services depend on the XenStore, most notably the
 * XenBus used to discover and manage Xen devices.  These services
 * are implemented as NewBus child attachments to a bus exported
 * by this XenStore driver.
 */

static struct xs_watch *find_watch(const char *token);

MALLOC_DEFINE(M_XENSTORE, "xenstore", "XenStore data and results");

/**
 * Pointer to shared memory communication structures allowing us
 * to communicate with the XenStore service.
 *
 * When operating in full PV mode, this pointer is set early in kernel
 * startup from within xen_machdep.c.  In HVM mode, we use hypercalls
 * to get the guest frame number for the shared page and then map it
 * into kva.  See xs_init() for details.
 */
struct xenstore_domain_interface *xen_store;

/*-------------------------- Private Data Structures ------------------------*/

/**
 * Structure capturing messages received from the XenStore service.
 */
struct xs_stored_msg {
	TAILQ_ENTRY(xs_stored_msg) list;

	struct xsd_sockmsg hdr;

	union {
		/* Queued replies. */
		struct {
			char *body;
		} reply;

		/* Queued watch events. */
		struct {
			struct xs_watch *handle;
			const char **vec;
			u_int vec_size;
		} watch;
	} u;
};
TAILQ_HEAD(xs_stored_msg_list, xs_stored_msg);

/**
 * Container for all XenStore related state.
 */
struct xs_softc {
	/** Newbus device for the XenStore. */
	device_t xs_dev;

	/**
	 * Lock serializing access to ring producer/consumer
	 * indexes.  Use of this lock guarantees that wakeups
	 * of blocking readers/writers are not missed due to
	 * races with the XenStore service.
	 */
	struct mtx ring_lock;

	/*
	 * Mutex used to insure exclusive access to the outgoing
	 * communication ring.  We use a lock type that can be
	 * held while sleeping so that xs_write() can block waiting
	 * for space in the ring to free up, without allowing another
	 * writer to come in and corrupt a partial message write.
	 */
	struct sx request_mutex;

	/**
	 * A list of replies to our requests.
	 *
	 * The reply list is filled by xs_rcv_thread().  It
	 * is consumed by the context that issued the request
	 * to which a reply is made.  The requester blocks in
	 * xs_read_reply().
	 *
	 * /note Only one requesting context can be active at a time.
	 *       This is guaranteed by the request_mutex and insures
	 *	 that the requester sees replies matching the order
	 *	 of its requests.
	 */
	struct xs_stored_msg_list reply_list;

	/** Lock protecting the reply list. */
	struct mtx reply_lock;

	/**
	 * List of registered watches.
	 */
	struct xs_watch_list  registered_watches;

	/** Lock protecting the registered watches list. */
	struct mtx registered_watches_lock;

	/**
	 * List of pending watch callback events.
	 */
	struct xs_stored_msg_list watch_events;

	/** Lock protecting the watch calback list. */
	struct mtx watch_events_lock;

	/**
	 * Sleepable lock used to prevent VM suspension while a
	 * xenstore transaction is outstanding.
	 *
	 * Each active transaction holds a shared lock on the
	 * suspend mutex.  Our suspend method blocks waiting
	 * to acquire an exclusive lock.  This guarantees that
	 * suspend processing will only proceed once all active
	 * transactions have been retired.
	 */
	struct sx suspend_mutex;

	/**
	 * The processid of the xenwatch thread.
	 */
	pid_t xenwatch_pid;

	/**
	 * Sleepable mutex used to gate the execution of XenStore
	 * watch event callbacks.
	 *
	 * xenwatch_thread holds an exclusive lock on this mutex
	 * while delivering event callbacks, and xenstore_unregister_watch()
	 * uses an exclusive lock of this mutex to guarantee that no
	 * callbacks of the just unregistered watch are pending
	 * before returning to its caller.
	 */
	struct sx xenwatch_mutex;

#ifdef XENHVM
	/**
	 * The HVM guest pseudo-physical frame number.  This is Xen's mapping
	 * of the true machine frame number into our "physical address space".
	 */
	unsigned long gpfn;
#endif

	/**
	 * The event channel for communicating with the
	 * XenStore service.
	 */
	int evtchn;

	/** Interrupt number for our event channel. */
	u_int irq;

	/**
	 * Interrupt driven config hook allowing us to defer
	 * attaching children until interrupts (and thus communication
	 * with the XenStore service) are available.
	 */
	struct intr_config_hook xs_attachcb;
};

/*-------------------------------- Global Data ------------------------------*/
static struct xs_softc xs;

/*------------------------- Private Utility Functions -----------------------*/

/**
 * Count and optionally record pointers to a number of NUL terminated
 * strings in a buffer.
 *
 * \param strings  A pointer to a contiguous buffer of NUL terminated strings.
 * \param dest	   An array to store pointers to each string found in strings.
 * \param len	   The length of the buffer pointed to by strings.
 *
 * \return  A count of the number of strings found.
 */
static u_int
extract_strings(const char *strings, const char **dest, u_int len)
{
	u_int num;
	const char *p;

	for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1) {
		if (dest != NULL)
			*dest++ = p;
		num++;
	}

	return (num);
}

/**
 * Convert a contiguous buffer containing a series of NUL terminated
 * strings into an array of pointers to strings.
 *
 * The returned pointer references the array of string pointers which
 * is followed by the storage for the string data.  It is the client's
 * responsibility to free this storage.
 *
 * The storage addressed by strings is free'd prior to split returning.
 *
 * \param strings  A pointer to a contiguous buffer of NUL terminated strings.
 * \param len	   The length of the buffer pointed to by strings.
 * \param num	   The number of strings found and returned in the strings
 *                 array.
 *
 * \return  An array of pointers to the strings found in the input buffer.
 */
static const char **
split(char *strings, u_int len, u_int *num)
{
	const char **ret;

	/* Protect against unterminated buffers. */
	strings[len - 1] = '\0';

	/* Count the strings. */
	*num = extract_strings(strings, /*dest*/NULL, len);

	/* Transfer to one big alloc for easy freeing by the caller. */
	ret = malloc(*num * sizeof(char *) + len, M_XENSTORE, M_WAITOK);
	memcpy(&ret[*num], strings, len);
	free(strings, M_XENSTORE);

	/* Extract pointers to newly allocated array. */
	strings = (char *)&ret[*num];
	(void)extract_strings(strings, /*dest*/ret, len);

	return (ret);
}

/*------------------------- Public Utility Functions -------------------------*/
/*------- API comments for these methods can be found in xenstorevar.h -------*/
struct sbuf *
xs_join(const char *dir, const char *name)
{
	struct sbuf *sb;

	sb = sbuf_new_auto();
	sbuf_cat(sb, dir);
	if (name[0] != '\0') {
		sbuf_putc(sb, '/');
		sbuf_cat(sb, name);
	}
	sbuf_finish(sb);

	return (sb);
}

/*-------------------- Low Level Communication Management --------------------*/
/**
 * Interrupt handler for the XenStore event channel.
 *
 * XenStore reads and writes block on "xen_store" for buffer
 * space.  Wakeup any blocking operations when the XenStore
 * service has modified the queues.
 */
static void
xs_intr(void * arg __unused /*__attribute__((unused))*/)
{

	/*
	 * Hold ring lock across wakeup so that clients
	 * cannot miss a wakeup.
	 */
	mtx_lock(&xs.ring_lock);
	wakeup(xen_store);
	mtx_unlock(&xs.ring_lock);
}

/**
 * Verify that the indexes for a ring are valid.
 *
 * The difference between the producer and consumer cannot
 * exceed the size of the ring.
 *
 * \param cons  The consumer index for the ring to test.
 * \param prod  The producer index for the ring to test.
 *
 * \retval 1  If indexes are in range.
 * \retval 0  If the indexes are out of range.
 */
static int
xs_check_indexes(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod)
{

	return ((prod - cons) <= XENSTORE_RING_SIZE);
}

/**
 * Return a pointer to, and the length of, the contiguous
 * free region available for output in a ring buffer.
 *
 * \param cons  The consumer index for the ring.
 * \param prod  The producer index for the ring.
 * \param buf   The base address of the ring's storage.
 * \param len   The amount of contiguous storage available.
 *
 * \return  A pointer to the start location of the free region.
 */
static void *
xs_get_output_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
    char *buf, uint32_t *len)
{

	*len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(prod);
	if ((XENSTORE_RING_SIZE - (prod - cons)) < *len)
		*len = XENSTORE_RING_SIZE - (prod - cons);
	return (buf + MASK_XENSTORE_IDX(prod));
}

/**
 * Return a pointer to, and the length of, the contiguous
 * data available to read from a ring buffer.
 *
 * \param cons  The consumer index for the ring.
 * \param prod  The producer index for the ring.
 * \param buf   The base address of the ring's storage.
 * \param len   The amount of contiguous data available to read.
 *
 * \return  A pointer to the start location of the available data.
 */
static const void *
xs_get_input_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
    const char *buf, uint32_t *len)
{

	*len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(cons);
	if ((prod - cons) < *len)
		*len = prod - cons;
	return (buf + MASK_XENSTORE_IDX(cons));
}

/**
 * Transmit data to the XenStore service.
 *
 * \param tdata  A pointer to the contiguous data to send.
 * \param len    The amount of data to send.
 *
 * \return  On success 0, otherwise an errno value indicating the
 *          cause of failure.
 *
 * \invariant  Called from thread context.
 * \invariant  The buffer pointed to by tdata is at least len bytes
 *             in length.
 * \invariant  xs.request_mutex exclusively locked.
 */
static int
xs_write_store(const void *tdata, unsigned len)
{
	XENSTORE_RING_IDX cons, prod;
	const char *data = (const char *)tdata;
	int error;

	sx_assert(&xs.request_mutex, SX_XLOCKED);
	while (len != 0) {
		void *dst;
		u_int avail;

		/* Hold lock so we can't miss wakeups should we block. */
		mtx_lock(&xs.ring_lock);
		cons = xen_store->req_cons;
		prod = xen_store->req_prod;
		if ((prod - cons) == XENSTORE_RING_SIZE) {
			/*
			 * Output ring is full. Wait for a ring event.
			 *
			 * Note that the events from both queues
			 * are combined, so being woken does not
			 * guarantee that data exist in the read
			 * ring.
			 *
			 * To simplify error recovery and the retry,
			 * we specify PDROP so our lock is *not* held
			 * when msleep returns.
			 */
			error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
			     "xbwrite", /*timeout*/0);
			if (error && error != EWOULDBLOCK)
				return (error);

			/* Try again. */
			continue;
		}
		mtx_unlock(&xs.ring_lock);

		/* Verify queue sanity. */
		if (!xs_check_indexes(cons, prod)) {
			xen_store->req_cons = xen_store->req_prod = 0;
			return (EIO);
		}

		dst = xs_get_output_chunk(cons, prod, xen_store->req, &avail);
		if (avail > len)
			avail = len;

		memcpy(dst, data, avail);
		data += avail;
		len -= avail;

		/*
		 * The store to the producer index, which indicates
		 * to the other side that new data has arrived, must
		 * be visible only after our copy of the data into the
		 * ring has completed.
		 */
		wmb();
		xen_store->req_prod += avail;

		/*
		 * notify_remote_via_evtchn implies mb(). The other side
		 * will see the change to req_prod at the time of the
		 * interrupt.
		 */
		notify_remote_via_evtchn(xs.evtchn);
	}

	return (0);
}

/**
 * Receive data from the XenStore service.
 *
 * \param tdata  A pointer to the contiguous buffer to receive the data.
 * \param len    The amount of data to receive.
 *
 * \return  On success 0, otherwise an errno value indicating the
 *          cause of failure.
 *
 * \invariant  Called from thread context.
 * \invariant  The buffer pointed to by tdata is at least len bytes
 *             in length.
 *
 * \note xs_read does not perform any internal locking to guarantee
 *       serial access to the incoming ring buffer.  However, there
 *	 is only one context processing reads: xs_rcv_thread().
 */
static int
xs_read_store(void *tdata, unsigned len)
{
	XENSTORE_RING_IDX cons, prod;
	char *data = (char *)tdata;
	int error;

	while (len != 0) {
		u_int avail;
		const char *src;

		/* Hold lock so we can't miss wakeups should we block. */
		mtx_lock(&xs.ring_lock);
		cons = xen_store->rsp_cons;
		prod = xen_store->rsp_prod;
		if (cons == prod) {
			/*
			 * Nothing to read. Wait for a ring event.
			 *
			 * Note that the events from both queues
			 * are combined, so being woken does not
			 * guarantee that data exist in the read
			 * ring.
			 *
			 * To simplify error recovery and the retry,
			 * we specify PDROP so our lock is *not* held
			 * when msleep returns.
			 */
			error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
			    "xbread", /*timeout*/0);
			if (error && error != EWOULDBLOCK)
				return (error);
			continue;
		}
		mtx_unlock(&xs.ring_lock);

		/* Verify queue sanity. */
		if (!xs_check_indexes(cons, prod)) {
			xen_store->rsp_cons = xen_store->rsp_prod = 0;
			return (EIO);
		}

		src = xs_get_input_chunk(cons, prod, xen_store->rsp, &avail);
		if (avail > len)
			avail = len;

		/*
		 * Insure the data we read is related to the indexes
		 * we read above.
		 */
		rmb();

		memcpy(data, src, avail);
		data += avail;
		len -= avail;

		/*
		 * Insure that the producer of this ring does not see
		 * the ring space as free until after we have copied it
		 * out.
		 */
		mb();
		xen_store->rsp_cons += avail;

		/*
		 * notify_remote_via_evtchn implies mb(). The producer
		 * will see the updated consumer index when the event
		 * is delivered.
		 */
		notify_remote_via_evtchn(xs.evtchn);
	}

	return (0);
}

/*----------------------- Received Message Processing ------------------------*/
/**
 * Block reading the next message from the XenStore service and
 * process the result.
 *
 * \param type  The returned type of the XenStore message received.
 *
 * \return  0 on success.  Otherwise an errno value indicating the
 *          type of failure encountered.
 */
static int
xs_process_msg(enum xsd_sockmsg_type *type)
{
	struct xs_stored_msg *msg;
	char *body;
	int error;

	msg = malloc(sizeof(*msg), M_XENSTORE, M_WAITOK);
	error = xs_read_store(&msg->hdr, sizeof(msg->hdr));
	if (error) {
		free(msg, M_XENSTORE);
		return (error);
	}

	body = malloc(msg->hdr.len + 1, M_XENSTORE, M_WAITOK);
	error = xs_read_store(body, msg->hdr.len);
	if (error) {
		free(body, M_XENSTORE);
		free(msg, M_XENSTORE);
		return (error);
	}
	body[msg->hdr.len] = '\0';

	*type = msg->hdr.type;
	if (msg->hdr.type == XS_WATCH_EVENT) {
		msg->u.watch.vec = split(body, msg->hdr.len,
		    &msg->u.watch.vec_size);

		mtx_lock(&xs.registered_watches_lock);
		msg->u.watch.handle = find_watch(
		    msg->u.watch.vec[XS_WATCH_TOKEN]);
		if (msg->u.watch.handle != NULL) {
			mtx_lock(&xs.watch_events_lock);
			TAILQ_INSERT_TAIL(&xs.watch_events, msg, list);
			wakeup(&xs.watch_events);
			mtx_unlock(&xs.watch_events_lock);
		} else {
			free(msg->u.watch.vec, M_XENSTORE);
			free(msg, M_XENSTORE);
		}
		mtx_unlock(&xs.registered_watches_lock);
	} else {
		msg->u.reply.body = body;
		mtx_lock(&xs.reply_lock);
		TAILQ_INSERT_TAIL(&xs.reply_list, msg, list);
		wakeup(&xs.reply_list);
		mtx_unlock(&xs.reply_lock);
	}

	return (0);
}

/**
 * Thread body of the XenStore receive thread.
 *
 * This thread blocks waiting for data from the XenStore service
 * and processes and received messages.
 */
static void
xs_rcv_thread(void *arg __unused)
{
	int error;
	enum xsd_sockmsg_type type;

	for (;;) {
		error = xs_process_msg(&type);
		if (error)
			printf("XENSTORE error %d while reading message\n",
			    error);
	}
}

/*---------------- XenStore Message Request/Reply Processing -----------------*/
/**
 * Filter invoked before transmitting any message to the XenStore service.
 *
 * The role of the filter may expand, but currently serves to manage
 * the interactions of messages with transaction state.
 *
 * \param request_msg_type  The message type for the request.
 */
static inline void
xs_request_filter(uint32_t request_msg_type)
{
	if (request_msg_type == XS_TRANSACTION_START)
		sx_slock(&xs.suspend_mutex);
}

/**
 * Filter invoked after transmitting any message to the XenStore service.
 *
 * The role of the filter may expand, but currently serves to manage
 * the interactions of messages with transaction state.
 *
 * \param request_msg_type     The message type for the original request.
 * \param reply_msg_type       The message type for any received reply.
 * \param request_reply_error  The error status from the attempt to send
 *                             the request or retrieve the reply.
 */
static inline void
xs_reply_filter(uint32_t request_msg_type,
    uint32_t reply_msg_type, int request_reply_error)
{
	/*
	 * The count of transactions drops if we attempted
	 * to end a transaction (even if that attempt fails
	 * in error), we receive a transaction end acknowledgement,
	 * or if our attempt to begin a transaction fails.
	 */
	if (request_msg_type == XS_TRANSACTION_END
	 || (request_reply_error == 0 && reply_msg_type == XS_TRANSACTION_END)
	 || (request_msg_type == XS_TRANSACTION_START
	  && (request_reply_error != 0 || reply_msg_type == XS_ERROR)))
		sx_sunlock(&xs.suspend_mutex);

}

#define xsd_error_count	(sizeof(xsd_errors) / sizeof(xsd_errors[0]))

/**
 * Convert a XenStore error string into an errno number.
 *
 * \param errorstring  The error string to convert.
 *
 * \return  The errno best matching the input string.
 *
 * \note Unknown error strings are converted to EINVAL.
 */
static int
xs_get_error(const char *errorstring)
{
	u_int i;

	for (i = 0; i < xsd_error_count; i++) {
		if (!strcmp(errorstring, xsd_errors[i].errstring))
			return (xsd_errors[i].errnum);
	}
	bsd_log(LOG_WARNING, "XENSTORE xen store gave: unknown error %s",
	    errorstring);
	return (EINVAL);
}

/**
 * Block waiting for a reply to a message request.
 *
 * \param type	  The returned type of the reply.
 * \param len	  The returned body length of the reply.
 * \param result  The returned body of the reply.
 *
 * \return  0 on success.  Otherwise an errno indicating the
 *          cause of failure.
 */
static int
xs_read_reply(enum xsd_sockmsg_type *type, u_int *len, void **result)
{
	struct xs_stored_msg *msg;
	char *body;
	int error;

	mtx_lock(&xs.reply_lock);
	while (TAILQ_EMPTY(&xs.reply_list)) {
		error = mtx_sleep(&xs.reply_list, &xs.reply_lock,
		    PCATCH, "xswait", hz/10);
		if (error && error != EWOULDBLOCK) {
			mtx_unlock(&xs.reply_lock);
			return (error);
		}
	}
	msg = TAILQ_FIRST(&xs.reply_list);
	TAILQ_REMOVE(&xs.reply_list, msg, list);
	mtx_unlock(&xs.reply_lock);

	*type = msg->hdr.type;
	if (len)
		*len = msg->hdr.len;
	body = msg->u.reply.body;

	free(msg, M_XENSTORE);
	*result = body;
	return (0);
}

/**
 * Pass-thru interface for XenStore access by userland processes
 * via the XenStore device.
 *
 * Reply type and length data are returned by overwriting these
 * fields in the passed in request message.
 *
 * \param msg	  A properly formatted message to transmit to
 *		  the XenStore service.
 * \param result  The returned body of the reply.
 *
 * \return  0 on success.  Otherwise an errno indicating the cause
 *          of failure.
 *
 * \note The returned result is provided in malloced storage and thus
 *       must be free'd by the caller with 'free(result, M_XENSTORE);
 */
int
xs_dev_request_and_reply(struct xsd_sockmsg *msg, void **result)
{
	uint32_t request_type;
	int error;

	request_type = msg->type;
	xs_request_filter(request_type);

	sx_xlock(&xs.request_mutex);
	if ((error = xs_write_store(msg, sizeof(*msg) + msg->len)) == 0)
		error = xs_read_reply(&msg->type, &msg->len, result);
	sx_xunlock(&xs.request_mutex);

	xs_reply_filter(request_type, msg->type, error);

	return (error);
}

/**
 * Send a message with an optionally muti-part body to the XenStore service.
 *
 * \param t              The transaction to use for this request.
 * \param request_type   The type of message to send.
 * \param iovec          Pointers to the body sections of the request.
 * \param num_vecs       The number of body sections in the request.
 * \param len            The returned length of the reply.
 * \param result         The returned body of the reply.
 *
 * \return  0 on success.  Otherwise an errno indicating
 *          the cause of failure.
 *
 * \note The returned result is provided in malloced storage and thus
 *       must be free'd by the caller with 'free(*result, M_XENSTORE);
 */
static int
xs_talkv(struct xs_transaction t, enum xsd_sockmsg_type request_type,
    const struct iovec *iovec, u_int num_vecs, u_int *len, void **result)
{
	struct xsd_sockmsg msg;
	void *ret = NULL;
	u_int i;
	int error;

	msg.tx_id = t.id;
	msg.req_id = 0;
	msg.type = request_type;
	msg.len = 0;
	for (i = 0; i < num_vecs; i++)
		msg.len += iovec[i].iov_len;

	xs_request_filter(request_type);

	sx_xlock(&xs.request_mutex);
	error = xs_write_store(&msg, sizeof(msg));
	if (error) {
		printf("xs_talkv failed %d\n", error);
		goto error_lock_held;
	}

	for (i = 0; i < num_vecs; i++) {
		error = xs_write_store(iovec[i].iov_base, iovec[i].iov_len);
		if (error) {
			printf("xs_talkv failed %d\n", error);
			goto error_lock_held;
		}
	}

	error = xs_read_reply(&msg.type, len, &ret);

error_lock_held:
	sx_xunlock(&xs.request_mutex);
	xs_reply_filter(request_type, msg.type, error);
	if (error)
		return (error);

	if (msg.type == XS_ERROR) {
		error = xs_get_error(ret);
		free(ret, M_XENSTORE);
		return (error);
	}

	/* Reply is either error or an echo of our request message type. */
	KASSERT(msg.type == request_type, ("bad xenstore message type"));

	if (result)
		*result = ret;
	else
		free(ret, M_XENSTORE);

	return (0);
}

/**
 * Wrapper for xs_talkv allowing easy transmission of a message with
 * a single, contiguous, message body.
 *
 * \param t              The transaction to use for this request.
 * \param request_type   The type of message to send.
 * \param body           The body of the request.
 * \param len            The returned length of the reply.
 * \param result         The returned body of the reply.
 *
 * \return  0 on success.  Otherwise an errno indicating
 *          the cause of failure.
 *
 * \note The returned result is provided in malloced storage and thus
 *       must be free'd by the caller with 'free(*result, M_XENSTORE);
 */
static int
xs_single(struct xs_transaction t, enum xsd_sockmsg_type request_type,
    const char *body, u_int *len, void **result)
{
	struct iovec iovec;

	iovec.iov_base = (void *)(uintptr_t)body;
	iovec.iov_len = strlen(body) + 1;

	return (xs_talkv(t, request_type, &iovec, 1, len, result));
}

/*------------------------- XenStore Watch Support ---------------------------*/
/**
 * Transmit a watch request to the XenStore service.
 *
 * \param path    The path in the XenStore to watch.
 * \param tocken  A unique identifier for this watch.
 *
 * \return  0 on success.  Otherwise an errno indicating the
 *          cause of failure.
 */
static int
xs_watch(const char *path, const char *token)
{
	struct iovec iov[2];

	iov[0].iov_base = (void *)(uintptr_t) path;
	iov[0].iov_len = strlen(path) + 1;
	iov[1].iov_base = (void *)(uintptr_t) token;
	iov[1].iov_len = strlen(token) + 1;

	return (xs_talkv(XST_NIL, XS_WATCH, iov, 2, NULL, NULL));
}

/**
 * Transmit an uwatch request to the XenStore service.
 *
 * \param path    The path in the XenStore to watch.
 * \param tocken  A unique identifier for this watch.
 *
 * \return  0 on success.  Otherwise an errno indicating the
 *          cause of failure.
 */
static int
xs_unwatch(const char *path, const char *token)
{
	struct iovec iov[2];

	iov[0].iov_base = (void *)(uintptr_t) path;
	iov[0].iov_len = strlen(path) + 1;
	iov[1].iov_base = (void *)(uintptr_t) token;
	iov[1].iov_len = strlen(token) + 1;

	return (xs_talkv(XST_NIL, XS_UNWATCH, iov, 2, NULL, NULL));
}

/**
 * Convert from watch token (unique identifier) to the associated
 * internal tracking structure for this watch.
 *
 * \param tocken  The unique identifier for the watch to find.
 *
 * \return  A pointer to the found watch structure or NULL.
 */
static struct xs_watch *
find_watch(const char *token)
{
	struct xs_watch *i, *cmp;

	cmp = (void *)strtoul(token, NULL, 16);

	LIST_FOREACH(i, &xs.registered_watches, list)
		if (i == cmp)
			return (i);