1 /******************************************************************************
4 * Low-level kernel interface to the XenStore.
6 * Copyright (C) 2005 Rusty Russell, IBM Corporation
7 * Copyright (C) 2009,2010 Spectra Logic Corporation
9 * This file may be distributed separately from the Linux kernel, or
10 * incorporated into other software packages, subject to the following license:
12 * Permission is hereby granted, free of charge, to any person obtaining a copy
13 * of this source file (the "Software"), to deal in the Software without
14 * restriction, including without limitation the rights to use, copy, modify,
15 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
16 * and to permit persons to whom the Software is furnished to do so, subject to
17 * the following conditions:
19 * The above copyright notice and this permission notice shall be included in
20 * all copies or substantial portions of the Software.
22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
37 #include <sys/kernel.h>
39 #include <sys/module.h>
40 #include <sys/mutex.h>
42 #include <sys/syslog.h>
43 #include <sys/malloc.h>
44 #include <sys/systm.h>
46 #include <sys/kthread.h>
48 #include <sys/sysctl.h>
50 #include <sys/unistd.h>
52 #include <machine/xen/xen-os.h>
53 #include <machine/stdarg.h>
55 #include <xen/evtchn.h>
56 #include <xen/gnttab.h>
57 #include <xen/hypervisor.h>
58 #include <xen/xen_intr.h>
60 #include <xen/interface/hvm/params.h>
62 #include <xen/xenstore/xenstorevar.h>
63 #include <xen/xenstore/xenstore_internal.h>
70 * \brief XenStore interface
72 * The XenStore interface is a simple storage system that is a means of
73 * communicating state and configuration data between the Xen Domain 0
74 * and the various guest domains. All configuration data other than
75 * a small amount of essential information required during the early
76 * boot process of launching a Xen aware guest, is managed using the
79 * The XenStore is ASCII string based, and has a structure and semantics
80 * similar to a filesystem. There are files and directories, the directories
81 * able to contain files or other directories. The depth of the hierachy
82 * is only limited by the XenStore's maximum path length.
84 * The communication channel between the XenStore service and other
85 * domains is via two, guest specific, ring buffers in a shared memory
86 * area. One ring buffer is used for communicating in each direction.
87 * The grant table references for this shared memory are given to the
88 * guest either via the xen_start_info structure for a fully para-
89 * virtualized guest, or via HVM hypercalls for a hardware virtualized
92 * The XenStore communication relies on an event channel and thus
93 * interrupts. For this reason, the attachment of the XenStore
94 * relies on an interrupt driven configuration hook to hold off
95 * boot processing until communication with the XenStore service
98 * Several Xen services depend on the XenStore, most notably the
99 * XenBus used to discover and manage Xen devices. These services
100 * are implemented as NewBus child attachments to a bus exported
101 * by this XenStore driver.
104 static struct xs_watch *find_watch(const char *token);
106 MALLOC_DEFINE(M_XENSTORE, "xenstore", "XenStore data and results");
109 * Pointer to shared memory communication structures allowing us
110 * to communicate with the XenStore service.
112 * When operating in full PV mode, this pointer is set early in kernel
113 * startup from within xen_machdep.c. In HVM mode, we use hypercalls
114 * to get the guest frame number for the shared page and then map it
115 * into kva. See xs_init() for details.
117 struct xenstore_domain_interface *xen_store;
119 /*-------------------------- Private Data Structures ------------------------*/
122 * Structure capturing messages received from the XenStore service.
124 struct xs_stored_msg {
125 TAILQ_ENTRY(xs_stored_msg) list;
127 struct xsd_sockmsg hdr;
130 /* Queued replies. */
135 /* Queued watch events. */
137 struct xs_watch *handle;
143 TAILQ_HEAD(xs_stored_msg_list, xs_stored_msg);
146 * Container for all XenStore related state.
149 /** Newbus device for the XenStore. */
153 * Lock serializing access to ring producer/consumer
154 * indexes. Use of this lock guarantees that wakeups
155 * of blocking readers/writers are not missed due to
156 * races with the XenStore service.
158 struct mtx ring_lock;
161 * Mutex used to insure exclusive access to the outgoing
162 * communication ring. We use a lock type that can be
163 * held while sleeping so that xs_write() can block waiting
164 * for space in the ring to free up, without allowing another
165 * writer to come in and corrupt a partial message write.
167 struct sx request_mutex;
170 * A list of replies to our requests.
172 * The reply list is filled by xs_rcv_thread(). It
173 * is consumed by the context that issued the request
174 * to which a reply is made. The requester blocks in
177 * /note Only one requesting context can be active at a time.
178 * This is guaranteed by the request_mutex and insures
179 * that the requester sees replies matching the order
182 struct xs_stored_msg_list reply_list;
184 /** Lock protecting the reply list. */
185 struct mtx reply_lock;
188 * List of registered watches.
190 struct xs_watch_list registered_watches;
192 /** Lock protecting the registered watches list. */
193 struct mtx registered_watches_lock;
196 * List of pending watch callback events.
198 struct xs_stored_msg_list watch_events;
200 /** Lock protecting the watch calback list. */
201 struct mtx watch_events_lock;
204 * Sleepable lock used to prevent VM suspension while a
205 * xenstore transaction is outstanding.
207 * Each active transaction holds a shared lock on the
208 * suspend mutex. Our suspend method blocks waiting
209 * to acquire an exclusive lock. This guarantees that
210 * suspend processing will only proceed once all active
211 * transactions have been retired.
213 struct sx suspend_mutex;
216 * The processid of the xenwatch thread.
221 * Sleepable mutex used to gate the execution of XenStore
222 * watch event callbacks.
224 * xenwatch_thread holds an exclusive lock on this mutex
225 * while delivering event callbacks, and xenstore_unregister_watch()
226 * uses an exclusive lock of this mutex to guarantee that no
227 * callbacks of the just unregistered watch are pending
228 * before returning to its caller.
230 struct sx xenwatch_mutex;
234 * The HVM guest pseudo-physical frame number. This is Xen's mapping
235 * of the true machine frame number into our "physical address space".
241 * The event channel for communicating with the
246 /** Interrupt number for our event channel. */
250 * Interrupt driven config hook allowing us to defer
251 * attaching children until interrupts (and thus communication
252 * with the XenStore service) are available.
254 struct intr_config_hook xs_attachcb;
257 /*-------------------------------- Global Data ------------------------------*/
258 static struct xs_softc xs;
260 /*------------------------- Private Utility Functions -----------------------*/
263 * Count and optionally record pointers to a number of NUL terminated
264 * strings in a buffer.
266 * \param strings A pointer to a contiguous buffer of NUL terminated strings.
267 * \param dest An array to store pointers to each string found in strings.
268 * \param len The length of the buffer pointed to by strings.
270 * \return A count of the number of strings found.
273 extract_strings(const char *strings, const char **dest, u_int len)
278 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1) {
288 * Convert a contiguous buffer containing a series of NUL terminated
289 * strings into an array of pointers to strings.
291 * The returned pointer references the array of string pointers which
292 * is followed by the storage for the string data. It is the client's
293 * responsibility to free this storage.
295 * The storage addressed by strings is free'd prior to split returning.
297 * \param strings A pointer to a contiguous buffer of NUL terminated strings.
298 * \param len The length of the buffer pointed to by strings.
299 * \param num The number of strings found and returned in the strings
302 * \return An array of pointers to the strings found in the input buffer.
305 split(char *strings, u_int len, u_int *num)
309 /* Protect against unterminated buffers. */
310 strings[len - 1] = '\0';
312 /* Count the strings. */
313 *num = extract_strings(strings, /*dest*/NULL, len);
315 /* Transfer to one big alloc for easy freeing by the caller. */
316 ret = malloc(*num * sizeof(char *) + len, M_XENSTORE, M_WAITOK);
317 memcpy(&ret[*num], strings, len);
318 free(strings, M_XENSTORE);
320 /* Extract pointers to newly allocated array. */
321 strings = (char *)&ret[*num];
322 (void)extract_strings(strings, /*dest*/ret, len);
327 /*------------------------- Public Utility Functions -------------------------*/
328 /*------- API comments for these methods can be found in xenstorevar.h -------*/
330 xs_join(const char *dir, const char *name)
334 sb = sbuf_new_auto();
336 if (name[0] != '\0') {
345 /*-------------------- Low Level Communication Management --------------------*/
347 * Interrupt handler for the XenStore event channel.
349 * XenStore reads and writes block on "xen_store" for buffer
350 * space. Wakeup any blocking operations when the XenStore
351 * service has modified the queues.
354 xs_intr(void * arg __unused /*__attribute__((unused))*/)
358 * Hold ring lock across wakeup so that clients
359 * cannot miss a wakeup.
361 mtx_lock(&xs.ring_lock);
363 mtx_unlock(&xs.ring_lock);
367 * Verify that the indexes for a ring are valid.
369 * The difference between the producer and consumer cannot
370 * exceed the size of the ring.
372 * \param cons The consumer index for the ring to test.
373 * \param prod The producer index for the ring to test.
375 * \retval 1 If indexes are in range.
376 * \retval 0 If the indexes are out of range.
379 xs_check_indexes(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod)
382 return ((prod - cons) <= XENSTORE_RING_SIZE);
386 * Return a pointer to, and the length of, the contiguous
387 * free region available for output in a ring buffer.
389 * \param cons The consumer index for the ring.
390 * \param prod The producer index for the ring.
391 * \param buf The base address of the ring's storage.
392 * \param len The amount of contiguous storage available.
394 * \return A pointer to the start location of the free region.
397 xs_get_output_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
398 char *buf, uint32_t *len)
401 *len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(prod);
402 if ((XENSTORE_RING_SIZE - (prod - cons)) < *len)
403 *len = XENSTORE_RING_SIZE - (prod - cons);
404 return (buf + MASK_XENSTORE_IDX(prod));
408 * Return a pointer to, and the length of, the contiguous
409 * data available to read from a ring buffer.
411 * \param cons The consumer index for the ring.
412 * \param prod The producer index for the ring.
413 * \param buf The base address of the ring's storage.
414 * \param len The amount of contiguous data available to read.
416 * \return A pointer to the start location of the available data.
419 xs_get_input_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
420 const char *buf, uint32_t *len)
423 *len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(cons);
424 if ((prod - cons) < *len)
426 return (buf + MASK_XENSTORE_IDX(cons));
430 * Transmit data to the XenStore service.
432 * \param tdata A pointer to the contiguous data to send.
433 * \param len The amount of data to send.
435 * \return On success 0, otherwise an errno value indicating the
438 * \invariant Called from thread context.
439 * \invariant The buffer pointed to by tdata is at least len bytes
441 * \invariant xs.request_mutex exclusively locked.
444 xs_write_store(const void *tdata, unsigned len)
446 XENSTORE_RING_IDX cons, prod;
447 const char *data = (const char *)tdata;
450 sx_assert(&xs.request_mutex, SX_XLOCKED);
455 /* Hold lock so we can't miss wakeups should we block. */
456 mtx_lock(&xs.ring_lock);
457 cons = xen_store->req_cons;
458 prod = xen_store->req_prod;
459 if ((prod - cons) == XENSTORE_RING_SIZE) {
461 * Output ring is full. Wait for a ring event.
463 * Note that the events from both queues
464 * are combined, so being woken does not
465 * guarantee that data exist in the read
468 * To simplify error recovery and the retry,
469 * we specify PDROP so our lock is *not* held
470 * when msleep returns.
472 error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
473 "xbwrite", /*timeout*/0);
474 if (error && error != EWOULDBLOCK)
480 mtx_unlock(&xs.ring_lock);
482 /* Verify queue sanity. */
483 if (!xs_check_indexes(cons, prod)) {
484 xen_store->req_cons = xen_store->req_prod = 0;
488 dst = xs_get_output_chunk(cons, prod, xen_store->req, &avail);
492 memcpy(dst, data, avail);
497 * The store to the producer index, which indicates
498 * to the other side that new data has arrived, must
499 * be visible only after our copy of the data into the
500 * ring has completed.
503 xen_store->req_prod += avail;
506 * notify_remote_via_evtchn implies mb(). The other side
507 * will see the change to req_prod at the time of the
510 notify_remote_via_evtchn(xs.evtchn);
517 * Receive data from the XenStore service.
519 * \param tdata A pointer to the contiguous buffer to receive the data.
520 * \param len The amount of data to receive.
522 * \return On success 0, otherwise an errno value indicating the
525 * \invariant Called from thread context.
526 * \invariant The buffer pointed to by tdata is at least len bytes
529 * \note xs_read does not perform any internal locking to guarantee
530 * serial access to the incoming ring buffer. However, there
531 * is only one context processing reads: xs_rcv_thread().
534 xs_read_store(void *tdata, unsigned len)
536 XENSTORE_RING_IDX cons, prod;
537 char *data = (char *)tdata;
544 /* Hold lock so we can't miss wakeups should we block. */
545 mtx_lock(&xs.ring_lock);
546 cons = xen_store->rsp_cons;
547 prod = xen_store->rsp_prod;
550 * Nothing to read. Wait for a ring event.
552 * Note that the events from both queues
553 * are combined, so being woken does not
554 * guarantee that data exist in the read
557 * To simplify error recovery and the retry,
558 * we specify PDROP so our lock is *not* held
559 * when msleep returns.
561 error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
562 "xbread", /*timeout*/0);
563 if (error && error != EWOULDBLOCK)
567 mtx_unlock(&xs.ring_lock);
569 /* Verify queue sanity. */
570 if (!xs_check_indexes(cons, prod)) {
571 xen_store->rsp_cons = xen_store->rsp_prod = 0;
575 src = xs_get_input_chunk(cons, prod, xen_store->rsp, &avail);
580 * Insure the data we read is related to the indexes
585 memcpy(data, src, avail);
590 * Insure that the producer of this ring does not see
591 * the ring space as free until after we have copied it
595 xen_store->rsp_cons += avail;
598 * notify_remote_via_evtchn implies mb(). The producer
599 * will see the updated consumer index when the event
602 notify_remote_via_evtchn(xs.evtchn);
608 /*----------------------- Received Message Processing ------------------------*/
610 * Block reading the next message from the XenStore service and
611 * process the result.
613 * \param type The returned type of the XenStore message received.
615 * \return 0 on success. Otherwise an errno value indicating the
616 * type of failure encountered.
619 xs_process_msg(enum xsd_sockmsg_type *type)
621 struct xs_stored_msg *msg;
625 msg = malloc(sizeof(*msg), M_XENSTORE, M_WAITOK);
626 error = xs_read_store(&msg->hdr, sizeof(msg->hdr));
628 free(msg, M_XENSTORE);
632 body = malloc(msg->hdr.len + 1, M_XENSTORE, M_WAITOK);
633 error = xs_read_store(body, msg->hdr.len);
635 free(body, M_XENSTORE);
636 free(msg, M_XENSTORE);
639 body[msg->hdr.len] = '\0';
641 *type = msg->hdr.type;
642 if (msg->hdr.type == XS_WATCH_EVENT) {
643 msg->u.watch.vec = split(body, msg->hdr.len,
644 &msg->u.watch.vec_size);
646 mtx_lock(&xs.registered_watches_lock);
647 msg->u.watch.handle = find_watch(
648 msg->u.watch.vec[XS_WATCH_TOKEN]);
649 if (msg->u.watch.handle != NULL) {
650 mtx_lock(&xs.watch_events_lock);
651 TAILQ_INSERT_TAIL(&xs.watch_events, msg, list);
652 wakeup(&xs.watch_events);
653 mtx_unlock(&xs.watch_events_lock);
655 free(msg->u.watch.vec, M_XENSTORE);
656 free(msg, M_XENSTORE);
658 mtx_unlock(&xs.registered_watches_lock);
660 msg->u.reply.body = body;
661 mtx_lock(&xs.reply_lock);
662 TAILQ_INSERT_TAIL(&xs.reply_list, msg, list);
663 wakeup(&xs.reply_list);
664 mtx_unlock(&xs.reply_lock);
671 * Thread body of the XenStore receive thread.
673 * This thread blocks waiting for data from the XenStore service
674 * and processes and received messages.
677 xs_rcv_thread(void *arg __unused)
680 enum xsd_sockmsg_type type;
683 error = xs_process_msg(&type);
685 printf("XENSTORE error %d while reading message\n",
690 /*---------------- XenStore Message Request/Reply Processing -----------------*/
692 * Filter invoked before transmitting any message to the XenStore service.
694 * The role of the filter may expand, but currently serves to manage
695 * the interactions of messages with transaction state.
697 * \param request_msg_type The message type for the request.
700 xs_request_filter(uint32_t request_msg_type)
702 if (request_msg_type == XS_TRANSACTION_START)
703 sx_slock(&xs.suspend_mutex);
707 * Filter invoked after transmitting any message to the XenStore service.
709 * The role of the filter may expand, but currently serves to manage
710 * the interactions of messages with transaction state.
712 * \param request_msg_type The message type for the original request.
713 * \param reply_msg_type The message type for any received reply.
714 * \param request_reply_error The error status from the attempt to send
715 * the request or retrieve the reply.
718 xs_reply_filter(uint32_t request_msg_type,
719 uint32_t reply_msg_type, int request_reply_error)
722 * The count of transactions drops if we attempted
723 * to end a transaction (even if that attempt fails
724 * in error), we receive a transaction end acknowledgement,
725 * or if our attempt to begin a transaction fails.
727 if (request_msg_type == XS_TRANSACTION_END
728 || (request_reply_error == 0 && reply_msg_type == XS_TRANSACTION_END)
729 || (request_msg_type == XS_TRANSACTION_START
730 && (request_reply_error != 0 || reply_msg_type == XS_ERROR)))
731 sx_sunlock(&xs.suspend_mutex);
735 #define xsd_error_count (sizeof(xsd_errors) / sizeof(xsd_errors[0]))
738 * Convert a XenStore error string into an errno number.
740 * \param errorstring The error string to convert.
742 * \return The errno best matching the input string.
744 * \note Unknown error strings are converted to EINVAL.
747 xs_get_error(const char *errorstring)
751 for (i = 0; i < xsd_error_count; i++) {
752 if (!strcmp(errorstring, xsd_errors[i].errstring))
753 return (xsd_errors[i].errnum);
755 log(LOG_WARNING, "XENSTORE xen store gave: unknown error %s",
761 * Block waiting for a reply to a message request.
763 * \param type The returned type of the reply.
764 * \param len The returned body length of the reply.
765 * \param result The returned body of the reply.
767 * \return 0 on success. Otherwise an errno indicating the
771 xs_read_reply(enum xsd_sockmsg_type *type, u_int *len, void **result)
773 struct xs_stored_msg *msg;
777 mtx_lock(&xs.reply_lock);
778 while (TAILQ_EMPTY(&xs.reply_list)) {
779 error = mtx_sleep(&xs.reply_list, &xs.reply_lock,
780 PCATCH, "xswait", hz/10);
781 if (error && error != EWOULDBLOCK) {
782 mtx_unlock(&xs.reply_lock);
786 msg = TAILQ_FIRST(&xs.reply_list);
787 TAILQ_REMOVE(&xs.reply_list, msg, list);
788 mtx_unlock(&xs.reply_lock);
790 *type = msg->hdr.type;
793 body = msg->u.reply.body;
795 free(msg, M_XENSTORE);
801 * Pass-thru interface for XenStore access by userland processes
802 * via the XenStore device.
804 * Reply type and length data are returned by overwriting these
805 * fields in the passed in request message.
807 * \param msg A properly formatted message to transmit to
808 * the XenStore service.
809 * \param result The returned body of the reply.
811 * \return 0 on success. Otherwise an errno indicating the cause
814 * \note The returned result is provided in malloced storage and thus
815 * must be free'd by the caller with 'free(result, M_XENSTORE);
818 xs_dev_request_and_reply(struct xsd_sockmsg *msg, void **result)
820 uint32_t request_type;
823 request_type = msg->type;
824 xs_request_filter(request_type);
826 sx_xlock(&xs.request_mutex);
827 if ((error = xs_write_store(msg, sizeof(*msg) + msg->len)) == 0)
828 error = xs_read_reply(&msg->type, &msg->len, result);
829 sx_xunlock(&xs.request_mutex);
831 xs_reply_filter(request_type, msg->type, error);
837 * Send a message with an optionally muti-part body to the XenStore service.
839 * \param t The transaction to use for this request.
840 * \param request_type The type of message to send.
841 * \param iovec Pointers to the body sections of the request.
842 * \param num_vecs The number of body sections in the request.
843 * \param len The returned length of the reply.
844 * \param result The returned body of the reply.
846 * \return 0 on success. Otherwise an errno indicating
847 * the cause of failure.
849 * \note The returned result is provided in malloced storage and thus
850 * must be free'd by the caller with 'free(*result, M_XENSTORE);
853 xs_talkv(struct xs_transaction t, enum xsd_sockmsg_type request_type,
854 const struct iovec *iovec, u_int num_vecs, u_int *len, void **result)
856 struct xsd_sockmsg msg;
863 msg.type = request_type;
865 for (i = 0; i < num_vecs; i++)
866 msg.len += iovec[i].iov_len;
868 xs_request_filter(request_type);
870 sx_xlock(&xs.request_mutex);
871 error = xs_write_store(&msg, sizeof(msg));
873 printf("xs_talkv failed %d\n", error);
874 goto error_lock_held;
877 for (i = 0; i < num_vecs; i++) {
878 error = xs_write_store(iovec[i].iov_base, iovec[i].iov_len);
880 printf("xs_talkv failed %d\n", error);
881 goto error_lock_held;
885 error = xs_read_reply(&msg.type, len, &ret);
888 sx_xunlock(&xs.request_mutex);
889 xs_reply_filter(request_type, msg.type, error);
893 if (msg.type == XS_ERROR) {
894 error = xs_get_error(ret);
895 free(ret, M_XENSTORE);
899 /* Reply is either error or an echo of our request message type. */
900 KASSERT(msg.type == request_type, ("bad xenstore message type"));
905 free(ret, M_XENSTORE);
911 * Wrapper for xs_talkv allowing easy transmission of a message with
912 * a single, contiguous, message body.
914 * \param t The transaction to use for this request.
915 * \param request_type The type of message to send.
916 * \param body The body of the request.
917 * \param len The returned length of the reply.
918 * \param result The returned body of the reply.
920 * \return 0 on success. Otherwise an errno indicating
921 * the cause of failure.
923 * \note The returned result is provided in malloced storage and thus
924 * must be free'd by the caller with 'free(*result, M_XENSTORE);
927 xs_single(struct xs_transaction t, enum xsd_sockmsg_type request_type,
928 const char *body, u_int *len, void **result)
932 iovec.iov_base = (void *)(uintptr_t)body;
933 iovec.iov_len = strlen(body) + 1;
935 return (xs_talkv(t, request_type, &iovec, 1, len, result));
938 /*------------------------- XenStore Watch Support ---------------------------*/
940 * Transmit a watch request to the XenStore service.
942 * \param path The path in the XenStore to watch.
943 * \param tocken A unique identifier for this watch.
945 * \return 0 on success. Otherwise an errno indicating the
949 xs_watch(const char *path, const char *token)
953 iov[0].iov_base = (void *)(uintptr_t) path;
954 iov[0].iov_len = strlen(path) + 1;
955 iov[1].iov_base = (void *)(uintptr_t) token;
956 iov[1].iov_len = strlen(token) + 1;
958 return (xs_talkv(XST_NIL, XS_WATCH, iov, 2, NULL, NULL));
962 * Transmit an uwatch request to the XenStore service.
964 * \param path The path in the XenStore to watch.
965 * \param tocken A unique identifier for this watch.
967 * \return 0 on success. Otherwise an errno indicating the
971 xs_unwatch(const char *path, const char *token)
975 iov[0].iov_base = (void *)(uintptr_t) path;
976 iov[0].iov_len = strlen(path) + 1;
977 iov[1].iov_base = (void *)(uintptr_t) token;
978 iov[1].iov_len = strlen(token) + 1;
980 return (xs_talkv(XST_NIL, XS_UNWATCH, iov, 2, NULL, NULL));
984 * Convert from watch token (unique identifier) to the associated
985 * internal tracking structure for this watch.
987 * \param tocken The unique identifier for the watch to find.
989 * \return A pointer to the found watch structure or NULL.
991 static struct xs_watch *
992 find_watch(const char *token)
994 struct xs_watch *i, *cmp;
996 cmp = (void *)strtoul(token, NULL, 16);
998 LIST_FOREACH(i, &xs.registered_watches, list)
1006 * Thread body of the XenStore watch event dispatch thread.
1009 xenwatch_thread(void *unused)
1011 struct xs_stored_msg *msg;
1015 mtx_lock(&xs.watch_events_lock);
1016 while (TAILQ_EMPTY(&xs.watch_events))
1017 mtx_sleep(&xs.watch_events,
1018 &xs.watch_events_lock,
1019 PWAIT | PCATCH, "waitev", hz/10);
1021 mtx_unlock(&xs.watch_events_lock);
1022 sx_xlock(&xs.xenwatch_mutex);
1024 mtx_lock(&xs.watch_events_lock);
1025 msg = TAILQ_FIRST(&xs.watch_events);
1027 TAILQ_REMOVE(&xs.watch_events, msg, list);
1028 mtx_unlock(&xs.watch_events_lock);
1032 * XXX There are messages coming in with a NULL
1033 * XXX callback. This deserves further investigation;
1034 * XXX the workaround here simply prevents the kernel
1035 * XXX from panic'ing on startup.
1037 if (msg->u.watch.handle->callback != NULL)
1038 msg->u.watch.handle->callback(
1039 msg->u.watch.handle,
1040 (const char **)msg->u.watch.vec,
1041 msg->u.watch.vec_size);
1042 free(msg->u.watch.vec, M_XENSTORE);
1043 free(msg, M_XENSTORE);
1046 sx_xunlock(&xs.xenwatch_mutex);
1050 /*----------- XenStore Configuration, Initialization, and Control ------------*/
1052 * Setup communication channels with the XenStore service.
1054 * \return On success, 0. Otherwise an errno value indicating the
1062 if (xen_store->rsp_prod != xen_store->rsp_cons) {
1063 log(LOG_WARNING, "XENSTORE response ring is not quiescent "
1064 "(%08x:%08x): fixing up\n",
1065 xen_store->rsp_cons, xen_store->rsp_prod);
1066 xen_store->rsp_cons = xen_store->rsp_prod;
1070 unbind_from_irqhandler(xs.irq);
1072 error = bind_caller_port_to_irqhandler(xs.evtchn, "xenstore",
1073 xs_intr, NULL, INTR_TYPE_NET, &xs.irq);
1075 log(LOG_WARNING, "XENSTORE request irq failed %i\n", error);
1082 /*------------------ Private Device Attachment Functions --------------------*/
1084 xs_identify(driver_t *driver, device_t parent)
1087 BUS_ADD_CHILD(parent, 0, "xenstore", 0);
1091 * Probe for the existance of the XenStore.
1096 xs_probe(device_t dev)
1099 * We are either operating within a PV kernel or being probed
1100 * as the child of the successfully attached xenpci device.
1101 * Thus we are in a Xen environment and there will be a XenStore.
1102 * Unconditionally return success.
1104 device_set_desc(dev, "XenStore");
1109 xs_attach_deferred(void *arg)
1113 bus_generic_probe(xs.xs_dev);
1114 bus_generic_attach(xs.xs_dev);
1116 config_intrhook_disestablish(&xs.xs_attachcb);
1120 * Attach to the XenStore.
1122 * This routine also prepares for the probe/attach of drivers that rely
1126 xs_attach(device_t dev)
1130 /* Allow us to get device_t from softc and vice-versa. */
1132 device_set_softc(dev, &xs);
1135 * This seems to be a layering violation. The XenStore is just
1136 * one of many clients of the Grant Table facility. It happens
1137 * to be the first and a gating consumer to all other devices,
1138 * so this does work. A better place would be in the PV support
1139 * code for fully PV kernels and the xenpci driver for HVM kernels.
1141 error = gnttab_init();
1144 "XENSTORE: Error initializing grant tables: %d\n", error);
1148 /* Initialize the interface to xenstore. */
1152 xs.evtchn = hvm_get_parameter(HVM_PARAM_STORE_EVTCHN);
1153 xs.gpfn = hvm_get_parameter(HVM_PARAM_STORE_PFN);
1154 xen_store = pmap_mapdev(xs.gpfn * PAGE_SIZE, PAGE_SIZE);
1156 xs.evtchn = xen_start_info->store_evtchn;
1159 TAILQ_INIT(&xs.reply_list);
1160 TAILQ_INIT(&xs.watch_events);
1162 mtx_init(&xs.ring_lock, "ring lock", NULL, MTX_DEF);
1163 mtx_init(&xs.reply_lock, "reply lock", NULL, MTX_DEF);
1164 sx_init(&xs.xenwatch_mutex, "xenwatch");
1165 sx_init(&xs.request_mutex, "xenstore request");
1166 sx_init(&xs.suspend_mutex, "xenstore suspend");
1167 mtx_init(&xs.registered_watches_lock, "watches", NULL, MTX_DEF);
1168 mtx_init(&xs.watch_events_lock, "watch events", NULL, MTX_DEF);
1171 /* Initialize the shared memory rings to talk to xenstored */
1172 error = xs_init_comms();
1176 error = kproc_create(xenwatch_thread, NULL, &p, RFHIGHPID,
1180 xs.xenwatch_pid = p->p_pid;
1182 error = kproc_create(xs_rcv_thread, NULL, NULL,
1183 RFHIGHPID, 0, "xenstore_rcv");
1185 xs.xs_attachcb.ich_func = xs_attach_deferred;
1186 xs.xs_attachcb.ich_arg = NULL;
1187 config_intrhook_establish(&xs.xs_attachcb);
1193 * Prepare for suspension of this VM by halting XenStore access after
1194 * all transactions and individual requests have completed.
1197 xs_suspend(device_t dev)
1201 /* Suspend child Xen devices. */
1202 error = bus_generic_suspend(dev);
1206 sx_xlock(&xs.suspend_mutex);
1207 sx_xlock(&xs.request_mutex);
1213 * Resume XenStore operations after this VM is resumed.
1216 xs_resume(device_t dev __unused)
1218 struct xs_watch *watch;
1219 char token[sizeof(watch) * 2 + 1];
1223 sx_xunlock(&xs.request_mutex);
1226 * No need for registered_watches_lock: the suspend_mutex
1229 LIST_FOREACH(watch, &xs.registered_watches, list) {
1230 sprintf(token, "%lX", (long)watch);
1231 xs_watch(watch->node, token);
1234 sx_xunlock(&xs.suspend_mutex);
1236 /* Resume child Xen devices. */
1237 bus_generic_resume(dev);
1242 /*-------------------- Private Device Attachment Data -----------------------*/
1243 static device_method_t xenstore_methods[] = {
1244 /* Device interface */
1245 DEVMETHOD(device_identify, xs_identify),
1246 DEVMETHOD(device_probe, xs_probe),
1247 DEVMETHOD(device_attach, xs_attach),
1248 DEVMETHOD(device_detach, bus_generic_detach),
1249 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1250 DEVMETHOD(device_suspend, xs_suspend),
1251 DEVMETHOD(device_resume, xs_resume),
1254 DEVMETHOD(bus_add_child, bus_generic_add_child),
1255 DEVMETHOD(bus_alloc_resource, bus_generic_alloc_resource),
1256 DEVMETHOD(bus_release_resource, bus_generic_release_resource),
1257 DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
1258 DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
1263 DEFINE_CLASS_0(xenstore, xenstore_driver, xenstore_methods, 0);
1264 static devclass_t xenstore_devclass;
1267 DRIVER_MODULE(xenstore, xenpci, xenstore_driver, xenstore_devclass, 0, 0);
1269 DRIVER_MODULE(xenstore, nexus, xenstore_driver, xenstore_devclass, 0, 0);
1272 /*------------------------------- Sysctl Data --------------------------------*/
1273 /* XXX Shouldn't the node be somewhere else? */
1274 SYSCTL_NODE(_dev, OID_AUTO, xen, CTLFLAG_RD, NULL, "Xen");
1275 SYSCTL_INT(_dev_xen, OID_AUTO, xsd_port, CTLFLAG_RD, &xs.evtchn, 0, "");
1276 SYSCTL_ULONG(_dev_xen, OID_AUTO, xsd_kva, CTLFLAG_RD, (u_long *) &xen_store, 0, "");
1278 /*-------------------------------- Public API --------------------------------*/
1279 /*------- API comments for these methods can be found in xenstorevar.h -------*/
1281 xs_directory(struct xs_transaction t, const char *dir, const char *node,
1282 u_int *num, const char ***result)
1289 path = xs_join(dir, node);
1290 error = xs_single(t, XS_DIRECTORY, sbuf_data(path), &len,
1296 *result = split(strings, len, num);
1302 xs_exists(struct xs_transaction t, const char *dir, const char *node)
1307 error = xs_directory(t, dir, node, &dir_n, &d);
1310 free(d, M_XENSTORE);
1315 xs_read(struct xs_transaction t, const char *dir, const char *node,
1316 u_int *len, void **result)
1322 path = xs_join(dir, node);
1323 error = xs_single(t, XS_READ, sbuf_data(path), len, &ret);
1332 xs_write(struct xs_transaction t, const char *dir, const char *node,
1336 struct iovec iovec[2];
1339 path = xs_join(dir, node);
1341 iovec[0].iov_base = (void *)(uintptr_t) sbuf_data(path);
1342 iovec[0].iov_len = sbuf_len(path) + 1;
1343 iovec[1].iov_base = (void *)(uintptr_t) string;
1344 iovec[1].iov_len = strlen(string);
1346 error = xs_talkv(t, XS_WRITE, iovec, 2, NULL, NULL);
1353 xs_mkdir(struct xs_transaction t, const char *dir, const char *node)
1358 path = xs_join(dir, node);
1359 ret = xs_single(t, XS_MKDIR, sbuf_data(path), NULL, NULL);
1366 xs_rm(struct xs_transaction t, const char *dir, const char *node)
1371 path = xs_join(dir, node);
1372 ret = xs_single(t, XS_RM, sbuf_data(path), NULL, NULL);
1379 xs_rm_tree(struct xs_transaction xbt, const char *base, const char *node)
1381 struct xs_transaction local_xbt;
1382 struct sbuf *root_path_sbuf;
1383 struct sbuf *cur_path_sbuf;
1391 root_path_sbuf = xs_join(base, node);
1392 cur_path_sbuf = xs_join(base, node);
1393 root_path = sbuf_data(root_path_sbuf);
1394 cur_path = sbuf_data(cur_path_sbuf);
1399 error = xs_transaction_start(&local_xbt);
1410 error = xs_directory(xbt, cur_path, "", &count, &dir);
1414 for (i = 0; i < count; i++) {
1415 error = xs_rm(xbt, cur_path, dir[i]);
1416 if (error == ENOTEMPTY) {
1417 struct sbuf *push_dir;
1420 * Descend to clear out this sub directory.
1421 * We'll return to cur_dir once push_dir
1424 push_dir = xs_join(cur_path, dir[i]);
1425 sbuf_delete(cur_path_sbuf);
1426 cur_path_sbuf = push_dir;
1427 cur_path = sbuf_data(cur_path_sbuf);
1429 } else if (error != 0) {
1434 free(dir, M_XENSTORE);
1440 /* Directory is empty. It is now safe to remove. */
1441 error = xs_rm(xbt, cur_path, "");
1445 if (!strcmp(cur_path, root_path))
1448 /* Return to processing the parent directory. */
1449 last_slash = strrchr(cur_path, '/');
1450 KASSERT(last_slash != NULL,
1451 ("xs_rm_tree: mangled path %s", cur_path));
1457 sbuf_delete(cur_path_sbuf);
1458 sbuf_delete(root_path_sbuf);
1460 free(dir, M_XENSTORE);
1462 if (local_xbt.id != 0) {
1465 terror = xs_transaction_end(local_xbt, /*abort*/error != 0);
1467 if (terror == EAGAIN && error == 0)
1474 xs_transaction_start(struct xs_transaction *t)
1479 error = xs_single(XST_NIL, XS_TRANSACTION_START, "", NULL,
1482 t->id = strtoul(id_str, NULL, 0);
1483 free(id_str, M_XENSTORE);
1489 xs_transaction_end(struct xs_transaction t, int abort)
1494 strcpy(abortstr, "F");
1496 strcpy(abortstr, "T");
1498 return (xs_single(t, XS_TRANSACTION_END, abortstr, NULL, NULL));
1502 xs_scanf(struct xs_transaction t, const char *dir, const char *node,
1503 int *scancountp, const char *fmt, ...)
1509 error = xs_read(t, dir, node, NULL, (void **) &val);
1514 ns = vsscanf(val, fmt, ap);
1516 free(val, M_XENSTORE);
1517 /* Distinctive errno. */
1526 xs_vprintf(struct xs_transaction t,
1527 const char *dir, const char *node, const char *fmt, va_list ap)
1532 sb = sbuf_new_auto();
1533 sbuf_vprintf(sb, fmt, ap);
1535 error = xs_write(t, dir, node, sbuf_data(sb));
1542 xs_printf(struct xs_transaction t, const char *dir, const char *node,
1543 const char *fmt, ...)
1549 error = xs_vprintf(t, dir, node, fmt, ap);
1556 xs_gather(struct xs_transaction t, const char *dir, ...)
1564 while (error == 0 && (name = va_arg(ap, char *)) != NULL) {
1565 const char *fmt = va_arg(ap, char *);
1566 void *result = va_arg(ap, void *);
1569 error = xs_read(t, dir, name, NULL, (void **) &p);
1574 if (sscanf(p, fmt, result) == 0)
1576 free(p, M_XENSTORE);
1578 *(char **)result = p;
1586 xs_register_watch(struct xs_watch *watch)
1588 /* Pointer in ascii is the token. */
1589 char token[sizeof(watch) * 2 + 1];
1592 sprintf(token, "%lX", (long)watch);
1594 sx_slock(&xs.suspend_mutex);
1596 mtx_lock(&xs.registered_watches_lock);
1597 KASSERT(find_watch(token) == NULL, ("watch already registered"));
1598 LIST_INSERT_HEAD(&xs.registered_watches, watch, list);
1599 mtx_unlock(&xs.registered_watches_lock);
1601 error = xs_watch(watch->node, token);
1603 /* Ignore errors due to multiple registration. */
1604 if (error == EEXIST)
1608 mtx_lock(&xs.registered_watches_lock);
1609 LIST_REMOVE(watch, list);
1610 mtx_unlock(&xs.registered_watches_lock);
1613 sx_sunlock(&xs.suspend_mutex);
1619 xs_unregister_watch(struct xs_watch *watch)
1621 struct xs_stored_msg *msg, *tmp;
1622 char token[sizeof(watch) * 2 + 1];
1625 sprintf(token, "%lX", (long)watch);
1627 sx_slock(&xs.suspend_mutex);
1629 mtx_lock(&xs.registered_watches_lock);
1630 if (find_watch(token) == NULL) {
1631 mtx_unlock(&xs.registered_watches_lock);
1632 sx_sunlock(&xs.suspend_mutex);
1635 LIST_REMOVE(watch, list);
1636 mtx_unlock(&xs.registered_watches_lock);
1638 error = xs_unwatch(watch->node, token);
1640 log(LOG_WARNING, "XENSTORE Failed to release watch %s: %i\n",
1641 watch->node, error);
1643 sx_sunlock(&xs.suspend_mutex);
1645 /* Cancel pending watch events. */
1646 mtx_lock(&xs.watch_events_lock);
1647 TAILQ_FOREACH_SAFE(msg, &xs.watch_events, list, tmp) {
1648 if (msg->u.watch.handle != watch)
1650 TAILQ_REMOVE(&xs.watch_events, msg, list);
1651 free(msg->u.watch.vec, M_XENSTORE);
1652 free(msg, M_XENSTORE);
1654 mtx_unlock(&xs.watch_events_lock);
1656 /* Flush any currently-executing callback, unless we are it. :-) */
1657 if (curproc->p_pid != xs.xenwatch_pid) {
1658 sx_xlock(&xs.xenwatch_mutex);
1659 sx_xunlock(&xs.xenwatch_mutex);