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>
63 #include <xen/xenstore/xenstorevar.h>
64 #include <xen/xenstore/xenstore_internal.h>
71 * \brief XenStore interface
73 * The XenStore interface is a simple storage system that is a means of
74 * communicating state and configuration data between the Xen Domain 0
75 * and the various guest domains. All configuration data other than
76 * a small amount of essential information required during the early
77 * boot process of launching a Xen aware guest, is managed using the
80 * The XenStore is ASCII string based, and has a structure and semantics
81 * similar to a filesystem. There are files and directories, the directories
82 * able to contain files or other directories. The depth of the hierachy
83 * is only limited by the XenStore's maximum path length.
85 * The communication channel between the XenStore service and other
86 * domains is via two, guest specific, ring buffers in a shared memory
87 * area. One ring buffer is used for communicating in each direction.
88 * The grant table references for this shared memory are given to the
89 * guest either via the xen_start_info structure for a fully para-
90 * virtualized guest, or via HVM hypercalls for a hardware virtualized
93 * The XenStore communication relies on an event channel and thus
94 * interrupts. For this reason, the attachment of the XenStore
95 * relies on an interrupt driven configuration hook to hold off
96 * boot processing until communication with the XenStore service
99 * Several Xen services depend on the XenStore, most notably the
100 * XenBus used to discover and manage Xen devices. These services
101 * are implemented as NewBus child attachments to a bus exported
102 * by this XenStore driver.
105 static struct xs_watch *find_watch(const char *token);
107 MALLOC_DEFINE(M_XENSTORE, "xenstore", "XenStore data and results");
110 * Pointer to shared memory communication structures allowing us
111 * to communicate with the XenStore service.
113 * When operating in full PV mode, this pointer is set early in kernel
114 * startup from within xen_machdep.c. In HVM mode, we use hypercalls
115 * to get the guest frame number for the shared page and then map it
116 * into kva. See xs_init() for details.
118 struct xenstore_domain_interface *xen_store;
120 /*-------------------------- Private Data Structures ------------------------*/
123 * Structure capturing messages received from the XenStore service.
125 struct xs_stored_msg {
126 TAILQ_ENTRY(xs_stored_msg) list;
128 struct xsd_sockmsg hdr;
131 /* Queued replies. */
136 /* Queued watch events. */
138 struct xs_watch *handle;
144 TAILQ_HEAD(xs_stored_msg_list, xs_stored_msg);
147 * Container for all XenStore related state.
150 /** Newbus device for the XenStore. */
154 * Lock serializing access to ring producer/consumer
155 * indexes. Use of this lock guarantees that wakeups
156 * of blocking readers/writers are not missed due to
157 * races with the XenStore service.
159 struct mtx ring_lock;
162 * Mutex used to insure exclusive access to the outgoing
163 * communication ring. We use a lock type that can be
164 * held while sleeping so that xs_write() can block waiting
165 * for space in the ring to free up, without allowing another
166 * writer to come in and corrupt a partial message write.
168 struct sx request_mutex;
171 * A list of replies to our requests.
173 * The reply list is filled by xs_rcv_thread(). It
174 * is consumed by the context that issued the request
175 * to which a reply is made. The requester blocks in
178 * /note Only one requesting context can be active at a time.
179 * This is guaranteed by the request_mutex and insures
180 * that the requester sees replies matching the order
183 struct xs_stored_msg_list reply_list;
185 /** Lock protecting the reply list. */
186 struct mtx reply_lock;
189 * List of registered watches.
191 struct xs_watch_list registered_watches;
193 /** Lock protecting the registered watches list. */
194 struct mtx registered_watches_lock;
197 * List of pending watch callback events.
199 struct xs_stored_msg_list watch_events;
201 /** Lock protecting the watch calback list. */
202 struct mtx watch_events_lock;
205 * Sleepable lock used to prevent VM suspension while a
206 * xenstore transaction is outstanding.
208 * Each active transaction holds a shared lock on the
209 * suspend mutex. Our suspend method blocks waiting
210 * to acquire an exclusive lock. This guarantees that
211 * suspend processing will only proceed once all active
212 * transactions have been retired.
214 struct sx suspend_mutex;
217 * The processid of the xenwatch thread.
222 * Sleepable mutex used to gate the execution of XenStore
223 * watch event callbacks.
225 * xenwatch_thread holds an exclusive lock on this mutex
226 * while delivering event callbacks, and xenstore_unregister_watch()
227 * uses an exclusive lock of this mutex to guarantee that no
228 * callbacks of the just unregistered watch are pending
229 * before returning to its caller.
231 struct sx xenwatch_mutex;
235 * The HVM guest pseudo-physical frame number. This is Xen's mapping
236 * of the true machine frame number into our "physical address space".
242 * The event channel for communicating with the
247 /** Interrupt number for our event channel. */
251 * Interrupt driven config hook allowing us to defer
252 * attaching children until interrupts (and thus communication
253 * with the XenStore service) are available.
255 struct intr_config_hook xs_attachcb;
258 /*-------------------------------- Global Data ------------------------------*/
259 static struct xs_softc xs;
261 /*------------------------- Private Utility Functions -----------------------*/
264 * Count and optionally record pointers to a number of NUL terminated
265 * strings in a buffer.
267 * \param strings A pointer to a contiguous buffer of NUL terminated strings.
268 * \param dest An array to store pointers to each string found in strings.
269 * \param len The length of the buffer pointed to by strings.
271 * \return A count of the number of strings found.
274 extract_strings(const char *strings, const char **dest, u_int len)
279 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1) {
289 * Convert a contiguous buffer containing a series of NUL terminated
290 * strings into an array of pointers to strings.
292 * The returned pointer references the array of string pointers which
293 * is followed by the storage for the string data. It is the client's
294 * responsibility to free this storage.
296 * The storage addressed by strings is free'd prior to split returning.
298 * \param strings A pointer to a contiguous buffer of NUL terminated strings.
299 * \param len The length of the buffer pointed to by strings.
300 * \param num The number of strings found and returned in the strings
303 * \return An array of pointers to the strings found in the input buffer.
306 split(char *strings, u_int len, u_int *num)
310 /* Protect against unterminated buffers. */
312 strings[len - 1] = '\0';
314 /* Count the strings. */
315 *num = extract_strings(strings, /*dest*/NULL, len);
317 /* Transfer to one big alloc for easy freeing by the caller. */
318 ret = malloc(*num * sizeof(char *) + len, M_XENSTORE, M_WAITOK);
319 memcpy(&ret[*num], strings, len);
320 free(strings, M_XENSTORE);
322 /* Extract pointers to newly allocated array. */
323 strings = (char *)&ret[*num];
324 (void)extract_strings(strings, /*dest*/ret, len);
329 /*------------------------- Public Utility Functions -------------------------*/
330 /*------- API comments for these methods can be found in xenstorevar.h -------*/
332 xs_join(const char *dir, const char *name)
336 sb = sbuf_new_auto();
338 if (name[0] != '\0') {
347 /*-------------------- Low Level Communication Management --------------------*/
349 * Interrupt handler for the XenStore event channel.
351 * XenStore reads and writes block on "xen_store" for buffer
352 * space. Wakeup any blocking operations when the XenStore
353 * service has modified the queues.
356 xs_intr(void * arg __unused /*__attribute__((unused))*/)
360 * Hold ring lock across wakeup so that clients
361 * cannot miss a wakeup.
363 mtx_lock(&xs.ring_lock);
365 mtx_unlock(&xs.ring_lock);
369 * Verify that the indexes for a ring are valid.
371 * The difference between the producer and consumer cannot
372 * exceed the size of the ring.
374 * \param cons The consumer index for the ring to test.
375 * \param prod The producer index for the ring to test.
377 * \retval 1 If indexes are in range.
378 * \retval 0 If the indexes are out of range.
381 xs_check_indexes(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod)
384 return ((prod - cons) <= XENSTORE_RING_SIZE);
388 * Return a pointer to, and the length of, the contiguous
389 * free region available for output in a ring buffer.
391 * \param cons The consumer index for the ring.
392 * \param prod The producer index for the ring.
393 * \param buf The base address of the ring's storage.
394 * \param len The amount of contiguous storage available.
396 * \return A pointer to the start location of the free region.
399 xs_get_output_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
400 char *buf, uint32_t *len)
403 *len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(prod);
404 if ((XENSTORE_RING_SIZE - (prod - cons)) < *len)
405 *len = XENSTORE_RING_SIZE - (prod - cons);
406 return (buf + MASK_XENSTORE_IDX(prod));
410 * Return a pointer to, and the length of, the contiguous
411 * data available to read from a ring buffer.
413 * \param cons The consumer index for the ring.
414 * \param prod The producer index for the ring.
415 * \param buf The base address of the ring's storage.
416 * \param len The amount of contiguous data available to read.
418 * \return A pointer to the start location of the available data.
421 xs_get_input_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
422 const char *buf, uint32_t *len)
425 *len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(cons);
426 if ((prod - cons) < *len)
428 return (buf + MASK_XENSTORE_IDX(cons));
432 * Transmit data to the XenStore service.
434 * \param tdata A pointer to the contiguous data to send.
435 * \param len The amount of data to send.
437 * \return On success 0, otherwise an errno value indicating the
440 * \invariant Called from thread context.
441 * \invariant The buffer pointed to by tdata is at least len bytes
443 * \invariant xs.request_mutex exclusively locked.
446 xs_write_store(const void *tdata, unsigned len)
448 XENSTORE_RING_IDX cons, prod;
449 const char *data = (const char *)tdata;
452 sx_assert(&xs.request_mutex, SX_XLOCKED);
457 /* Hold lock so we can't miss wakeups should we block. */
458 mtx_lock(&xs.ring_lock);
459 cons = xen_store->req_cons;
460 prod = xen_store->req_prod;
461 if ((prod - cons) == XENSTORE_RING_SIZE) {
463 * Output ring is full. Wait for a ring event.
465 * Note that the events from both queues
466 * are combined, so being woken does not
467 * guarantee that data exist in the read
470 * To simplify error recovery and the retry,
471 * we specify PDROP so our lock is *not* held
472 * when msleep returns.
474 error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
475 "xbwrite", /*timeout*/0);
476 if (error && error != EWOULDBLOCK)
482 mtx_unlock(&xs.ring_lock);
484 /* Verify queue sanity. */
485 if (!xs_check_indexes(cons, prod)) {
486 xen_store->req_cons = xen_store->req_prod = 0;
490 dst = xs_get_output_chunk(cons, prod, xen_store->req, &avail);
494 memcpy(dst, data, avail);
499 * The store to the producer index, which indicates
500 * to the other side that new data has arrived, must
501 * be visible only after our copy of the data into the
502 * ring has completed.
505 xen_store->req_prod += avail;
508 * notify_remote_via_evtchn implies mb(). The other side
509 * will see the change to req_prod at the time of the
512 notify_remote_via_evtchn(xs.evtchn);
519 * Receive data from the XenStore service.
521 * \param tdata A pointer to the contiguous buffer to receive the data.
522 * \param len The amount of data to receive.
524 * \return On success 0, otherwise an errno value indicating the
527 * \invariant Called from thread context.
528 * \invariant The buffer pointed to by tdata is at least len bytes
531 * \note xs_read does not perform any internal locking to guarantee
532 * serial access to the incoming ring buffer. However, there
533 * is only one context processing reads: xs_rcv_thread().
536 xs_read_store(void *tdata, unsigned len)
538 XENSTORE_RING_IDX cons, prod;
539 char *data = (char *)tdata;
546 /* Hold lock so we can't miss wakeups should we block. */
547 mtx_lock(&xs.ring_lock);
548 cons = xen_store->rsp_cons;
549 prod = xen_store->rsp_prod;
552 * Nothing to read. Wait for a ring event.
554 * Note that the events from both queues
555 * are combined, so being woken does not
556 * guarantee that data exist in the read
559 * To simplify error recovery and the retry,
560 * we specify PDROP so our lock is *not* held
561 * when msleep returns.
563 error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
564 "xbread", /*timeout*/0);
565 if (error && error != EWOULDBLOCK)
569 mtx_unlock(&xs.ring_lock);
571 /* Verify queue sanity. */
572 if (!xs_check_indexes(cons, prod)) {
573 xen_store->rsp_cons = xen_store->rsp_prod = 0;
577 src = xs_get_input_chunk(cons, prod, xen_store->rsp, &avail);
582 * Insure the data we read is related to the indexes
587 memcpy(data, src, avail);
592 * Insure that the producer of this ring does not see
593 * the ring space as free until after we have copied it
597 xen_store->rsp_cons += avail;
600 * notify_remote_via_evtchn implies mb(). The producer
601 * will see the updated consumer index when the event
604 notify_remote_via_evtchn(xs.evtchn);
610 /*----------------------- Received Message Processing ------------------------*/
612 * Block reading the next message from the XenStore service and
613 * process the result.
615 * \param type The returned type of the XenStore message received.
617 * \return 0 on success. Otherwise an errno value indicating the
618 * type of failure encountered.
621 xs_process_msg(enum xsd_sockmsg_type *type)
623 struct xs_stored_msg *msg;
627 msg = malloc(sizeof(*msg), M_XENSTORE, M_WAITOK);
628 error = xs_read_store(&msg->hdr, sizeof(msg->hdr));
630 free(msg, M_XENSTORE);
634 body = malloc(msg->hdr.len + 1, M_XENSTORE, M_WAITOK);
635 error = xs_read_store(body, msg->hdr.len);
637 free(body, M_XENSTORE);
638 free(msg, M_XENSTORE);
641 body[msg->hdr.len] = '\0';
643 *type = msg->hdr.type;
644 if (msg->hdr.type == XS_WATCH_EVENT) {
645 msg->u.watch.vec = split(body, msg->hdr.len,
646 &msg->u.watch.vec_size);
648 mtx_lock(&xs.registered_watches_lock);
649 msg->u.watch.handle = find_watch(
650 msg->u.watch.vec[XS_WATCH_TOKEN]);
651 if (msg->u.watch.handle != NULL) {
652 mtx_lock(&xs.watch_events_lock);
653 TAILQ_INSERT_TAIL(&xs.watch_events, msg, list);
654 wakeup(&xs.watch_events);
655 mtx_unlock(&xs.watch_events_lock);
657 free(msg->u.watch.vec, M_XENSTORE);
658 free(msg, M_XENSTORE);
660 mtx_unlock(&xs.registered_watches_lock);
662 msg->u.reply.body = body;
663 mtx_lock(&xs.reply_lock);
664 TAILQ_INSERT_TAIL(&xs.reply_list, msg, list);
665 wakeup(&xs.reply_list);
666 mtx_unlock(&xs.reply_lock);
673 * Thread body of the XenStore receive thread.
675 * This thread blocks waiting for data from the XenStore service
676 * and processes and received messages.
679 xs_rcv_thread(void *arg __unused)
682 enum xsd_sockmsg_type type;
685 error = xs_process_msg(&type);
687 printf("XENSTORE error %d while reading message\n",
692 /*---------------- XenStore Message Request/Reply Processing -----------------*/
694 * Filter invoked before transmitting any message to the XenStore service.
696 * The role of the filter may expand, but currently serves to manage
697 * the interactions of messages with transaction state.
699 * \param request_msg_type The message type for the request.
702 xs_request_filter(uint32_t request_msg_type)
704 if (request_msg_type == XS_TRANSACTION_START)
705 sx_slock(&xs.suspend_mutex);
709 * Filter invoked after transmitting any message to the XenStore service.
711 * The role of the filter may expand, but currently serves to manage
712 * the interactions of messages with transaction state.
714 * \param request_msg_type The message type for the original request.
715 * \param reply_msg_type The message type for any received reply.
716 * \param request_reply_error The error status from the attempt to send
717 * the request or retrieve the reply.
720 xs_reply_filter(uint32_t request_msg_type,
721 uint32_t reply_msg_type, int request_reply_error)
724 * The count of transactions drops if we attempted
725 * to end a transaction (even if that attempt fails
726 * in error), we receive a transaction end acknowledgement,
727 * or if our attempt to begin a transaction fails.
729 if (request_msg_type == XS_TRANSACTION_END
730 || (request_reply_error == 0 && reply_msg_type == XS_TRANSACTION_END)
731 || (request_msg_type == XS_TRANSACTION_START
732 && (request_reply_error != 0 || reply_msg_type == XS_ERROR)))
733 sx_sunlock(&xs.suspend_mutex);
737 #define xsd_error_count (sizeof(xsd_errors) / sizeof(xsd_errors[0]))
740 * Convert a XenStore error string into an errno number.
742 * \param errorstring The error string to convert.
744 * \return The errno best matching the input string.
746 * \note Unknown error strings are converted to EINVAL.
749 xs_get_error(const char *errorstring)
753 for (i = 0; i < xsd_error_count; i++) {
754 if (!strcmp(errorstring, xsd_errors[i].errstring))
755 return (xsd_errors[i].errnum);
757 log(LOG_WARNING, "XENSTORE xen store gave: unknown error %s",
763 * Block waiting for a reply to a message request.
765 * \param type The returned type of the reply.
766 * \param len The returned body length of the reply.
767 * \param result The returned body of the reply.
769 * \return 0 on success. Otherwise an errno indicating the
773 xs_read_reply(enum xsd_sockmsg_type *type, u_int *len, void **result)
775 struct xs_stored_msg *msg;
779 mtx_lock(&xs.reply_lock);
780 while (TAILQ_EMPTY(&xs.reply_list)) {
781 error = mtx_sleep(&xs.reply_list, &xs.reply_lock,
782 PCATCH, "xswait", hz/10);
783 if (error && error != EWOULDBLOCK) {
784 mtx_unlock(&xs.reply_lock);
788 msg = TAILQ_FIRST(&xs.reply_list);
789 TAILQ_REMOVE(&xs.reply_list, msg, list);
790 mtx_unlock(&xs.reply_lock);
792 *type = msg->hdr.type;
795 body = msg->u.reply.body;
797 free(msg, M_XENSTORE);
803 * Pass-thru interface for XenStore access by userland processes
804 * via the XenStore device.
806 * Reply type and length data are returned by overwriting these
807 * fields in the passed in request message.
809 * \param msg A properly formatted message to transmit to
810 * the XenStore service.
811 * \param result The returned body of the reply.
813 * \return 0 on success. Otherwise an errno indicating the cause
816 * \note The returned result is provided in malloced storage and thus
817 * must be free'd by the caller with 'free(result, M_XENSTORE);
820 xs_dev_request_and_reply(struct xsd_sockmsg *msg, void **result)
822 uint32_t request_type;
825 request_type = msg->type;
826 xs_request_filter(request_type);
828 sx_xlock(&xs.request_mutex);
829 if ((error = xs_write_store(msg, sizeof(*msg) + msg->len)) == 0)
830 error = xs_read_reply(&msg->type, &msg->len, result);
831 sx_xunlock(&xs.request_mutex);
833 xs_reply_filter(request_type, msg->type, error);
839 * Send a message with an optionally muti-part body to the XenStore service.
841 * \param t The transaction to use for this request.
842 * \param request_type The type of message to send.
843 * \param iovec Pointers to the body sections of the request.
844 * \param num_vecs The number of body sections in the request.
845 * \param len The returned length of the reply.
846 * \param result The returned body of the reply.
848 * \return 0 on success. Otherwise an errno indicating
849 * the cause of failure.
851 * \note The returned result is provided in malloced storage and thus
852 * must be free'd by the caller with 'free(*result, M_XENSTORE);
855 xs_talkv(struct xs_transaction t, enum xsd_sockmsg_type request_type,
856 const struct iovec *iovec, u_int num_vecs, u_int *len, void **result)
858 struct xsd_sockmsg msg;
865 msg.type = request_type;
867 for (i = 0; i < num_vecs; i++)
868 msg.len += iovec[i].iov_len;
870 xs_request_filter(request_type);
872 sx_xlock(&xs.request_mutex);
873 error = xs_write_store(&msg, sizeof(msg));
875 printf("xs_talkv failed %d\n", error);
876 goto error_lock_held;
879 for (i = 0; i < num_vecs; i++) {
880 error = xs_write_store(iovec[i].iov_base, iovec[i].iov_len);
882 printf("xs_talkv failed %d\n", error);
883 goto error_lock_held;
887 error = xs_read_reply(&msg.type, len, &ret);
890 sx_xunlock(&xs.request_mutex);
891 xs_reply_filter(request_type, msg.type, error);
895 if (msg.type == XS_ERROR) {
896 error = xs_get_error(ret);
897 free(ret, M_XENSTORE);
901 /* Reply is either error or an echo of our request message type. */
902 KASSERT(msg.type == request_type, ("bad xenstore message type"));
907 free(ret, M_XENSTORE);
913 * Wrapper for xs_talkv allowing easy transmission of a message with
914 * a single, contiguous, message body.
916 * \param t The transaction to use for this request.
917 * \param request_type The type of message to send.
918 * \param body The body of the request.
919 * \param len The returned length of the reply.
920 * \param result The returned body of the reply.
922 * \return 0 on success. Otherwise an errno indicating
923 * the cause of failure.
925 * \note The returned result is provided in malloced storage and thus
926 * must be free'd by the caller with 'free(*result, M_XENSTORE);
929 xs_single(struct xs_transaction t, enum xsd_sockmsg_type request_type,
930 const char *body, u_int *len, void **result)
934 iovec.iov_base = (void *)(uintptr_t)body;
935 iovec.iov_len = strlen(body) + 1;
937 return (xs_talkv(t, request_type, &iovec, 1, len, result));
940 /*------------------------- XenStore Watch Support ---------------------------*/
942 * Transmit a watch request to the XenStore service.
944 * \param path The path in the XenStore to watch.
945 * \param tocken A unique identifier for this watch.
947 * \return 0 on success. Otherwise an errno indicating the
951 xs_watch(const char *path, const char *token)
955 iov[0].iov_base = (void *)(uintptr_t) path;
956 iov[0].iov_len = strlen(path) + 1;
957 iov[1].iov_base = (void *)(uintptr_t) token;
958 iov[1].iov_len = strlen(token) + 1;
960 return (xs_talkv(XST_NIL, XS_WATCH, iov, 2, NULL, NULL));
964 * Transmit an uwatch request to the XenStore service.
966 * \param path The path in the XenStore to watch.
967 * \param tocken A unique identifier for this watch.
969 * \return 0 on success. Otherwise an errno indicating the
973 xs_unwatch(const char *path, const char *token)
977 iov[0].iov_base = (void *)(uintptr_t) path;
978 iov[0].iov_len = strlen(path) + 1;
979 iov[1].iov_base = (void *)(uintptr_t) token;
980 iov[1].iov_len = strlen(token) + 1;
982 return (xs_talkv(XST_NIL, XS_UNWATCH, iov, 2, NULL, NULL));
986 * Convert from watch token (unique identifier) to the associated
987 * internal tracking structure for this watch.
989 * \param tocken The unique identifier for the watch to find.
991 * \return A pointer to the found watch structure or NULL.
993 static struct xs_watch *
994 find_watch(const char *token)
996 struct xs_watch *i, *cmp;
998 cmp = (void *)strtoul(token, NULL, 16);
1000 LIST_FOREACH(i, &xs.registered_watches, list)
1008 * Thread body of the XenStore watch event dispatch thread.
1011 xenwatch_thread(void *unused)
1013 struct xs_stored_msg *msg;
1017 mtx_lock(&xs.watch_events_lock);
1018 while (TAILQ_EMPTY(&xs.watch_events))
1019 mtx_sleep(&xs.watch_events,
1020 &xs.watch_events_lock,
1021 PWAIT | PCATCH, "waitev", hz/10);
1023 mtx_unlock(&xs.watch_events_lock);
1024 sx_xlock(&xs.xenwatch_mutex);
1026 mtx_lock(&xs.watch_events_lock);
1027 msg = TAILQ_FIRST(&xs.watch_events);
1029 TAILQ_REMOVE(&xs.watch_events, msg, list);
1030 mtx_unlock(&xs.watch_events_lock);
1034 * XXX There are messages coming in with a NULL
1035 * XXX callback. This deserves further investigation;
1036 * XXX the workaround here simply prevents the kernel
1037 * XXX from panic'ing on startup.
1039 if (msg->u.watch.handle->callback != NULL)
1040 msg->u.watch.handle->callback(
1041 msg->u.watch.handle,
1042 (const char **)msg->u.watch.vec,
1043 msg->u.watch.vec_size);
1044 free(msg->u.watch.vec, M_XENSTORE);
1045 free(msg, M_XENSTORE);
1048 sx_xunlock(&xs.xenwatch_mutex);
1052 /*----------- XenStore Configuration, Initialization, and Control ------------*/
1054 * Setup communication channels with the XenStore service.
1056 * \return On success, 0. Otherwise an errno value indicating the
1064 if (xen_store->rsp_prod != xen_store->rsp_cons) {
1065 log(LOG_WARNING, "XENSTORE response ring is not quiescent "
1066 "(%08x:%08x): fixing up\n",
1067 xen_store->rsp_cons, xen_store->rsp_prod);
1068 xen_store->rsp_cons = xen_store->rsp_prod;
1072 unbind_from_irqhandler(xs.irq);
1074 error = bind_caller_port_to_irqhandler(xs.evtchn, "xenstore",
1075 xs_intr, NULL, INTR_TYPE_NET, &xs.irq);
1077 log(LOG_WARNING, "XENSTORE request irq failed %i\n", error);
1084 /*------------------ Private Device Attachment Functions --------------------*/
1086 xs_identify(driver_t *driver, device_t parent)
1089 BUS_ADD_CHILD(parent, 0, "xenstore", 0);
1093 * Probe for the existance of the XenStore.
1098 xs_probe(device_t dev)
1101 * We are either operating within a PV kernel or being probed
1102 * as the child of the successfully attached xenpci device.
1103 * Thus we are in a Xen environment and there will be a XenStore.
1104 * Unconditionally return success.
1106 device_set_desc(dev, "XenStore");
1111 xs_attach_deferred(void *arg)
1115 bus_generic_probe(xs.xs_dev);
1116 bus_generic_attach(xs.xs_dev);
1118 config_intrhook_disestablish(&xs.xs_attachcb);
1122 * Attach to the XenStore.
1124 * This routine also prepares for the probe/attach of drivers that rely
1128 xs_attach(device_t dev)
1132 /* Allow us to get device_t from softc and vice-versa. */
1134 device_set_softc(dev, &xs);
1137 * This seems to be a layering violation. The XenStore is just
1138 * one of many clients of the Grant Table facility. It happens
1139 * to be the first and a gating consumer to all other devices,
1140 * so this does work. A better place would be in the PV support
1141 * code for fully PV kernels and the xenpci driver for HVM kernels.
1143 error = gnttab_init();
1146 "XENSTORE: Error initializing grant tables: %d\n", error);
1150 /* Initialize the interface to xenstore. */
1154 xs.evtchn = hvm_get_parameter(HVM_PARAM_STORE_EVTCHN);
1155 xs.gpfn = hvm_get_parameter(HVM_PARAM_STORE_PFN);
1156 xen_store = pmap_mapdev(xs.gpfn * PAGE_SIZE, PAGE_SIZE);
1158 xs.evtchn = xen_start_info->store_evtchn;
1161 TAILQ_INIT(&xs.reply_list);
1162 TAILQ_INIT(&xs.watch_events);
1164 mtx_init(&xs.ring_lock, "ring lock", NULL, MTX_DEF);
1165 mtx_init(&xs.reply_lock, "reply lock", NULL, MTX_DEF);
1166 sx_init(&xs.xenwatch_mutex, "xenwatch");
1167 sx_init(&xs.request_mutex, "xenstore request");
1168 sx_init(&xs.suspend_mutex, "xenstore suspend");
1169 mtx_init(&xs.registered_watches_lock, "watches", NULL, MTX_DEF);
1170 mtx_init(&xs.watch_events_lock, "watch events", NULL, MTX_DEF);
1173 /* Initialize the shared memory rings to talk to xenstored */
1174 error = xs_init_comms();
1178 error = kproc_create(xenwatch_thread, NULL, &p, RFHIGHPID,
1182 xs.xenwatch_pid = p->p_pid;
1184 error = kproc_create(xs_rcv_thread, NULL, NULL,
1185 RFHIGHPID, 0, "xenstore_rcv");
1187 xs.xs_attachcb.ich_func = xs_attach_deferred;
1188 xs.xs_attachcb.ich_arg = NULL;
1189 config_intrhook_establish(&xs.xs_attachcb);
1195 * Prepare for suspension of this VM by halting XenStore access after
1196 * all transactions and individual requests have completed.
1199 xs_suspend(device_t dev)
1203 /* Suspend child Xen devices. */
1204 error = bus_generic_suspend(dev);
1208 sx_xlock(&xs.suspend_mutex);
1209 sx_xlock(&xs.request_mutex);
1215 * Resume XenStore operations after this VM is resumed.
1218 xs_resume(device_t dev __unused)
1220 struct xs_watch *watch;
1221 char token[sizeof(watch) * 2 + 1];
1225 sx_xunlock(&xs.request_mutex);
1228 * No need for registered_watches_lock: the suspend_mutex
1231 LIST_FOREACH(watch, &xs.registered_watches, list) {
1232 sprintf(token, "%lX", (long)watch);
1233 xs_watch(watch->node, token);
1236 sx_xunlock(&xs.suspend_mutex);
1238 /* Resume child Xen devices. */
1239 bus_generic_resume(dev);
1244 /*-------------------- Private Device Attachment Data -----------------------*/
1245 static device_method_t xenstore_methods[] = {
1246 /* Device interface */
1247 DEVMETHOD(device_identify, xs_identify),
1248 DEVMETHOD(device_probe, xs_probe),
1249 DEVMETHOD(device_attach, xs_attach),
1250 DEVMETHOD(device_detach, bus_generic_detach),
1251 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1252 DEVMETHOD(device_suspend, xs_suspend),
1253 DEVMETHOD(device_resume, xs_resume),
1256 DEVMETHOD(bus_add_child, bus_generic_add_child),
1257 DEVMETHOD(bus_alloc_resource, bus_generic_alloc_resource),
1258 DEVMETHOD(bus_release_resource, bus_generic_release_resource),
1259 DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
1260 DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
1265 DEFINE_CLASS_0(xenstore, xenstore_driver, xenstore_methods, 0);
1266 static devclass_t xenstore_devclass;
1269 DRIVER_MODULE(xenstore, xenpci, xenstore_driver, xenstore_devclass, 0, 0);
1271 DRIVER_MODULE(xenstore, nexus, xenstore_driver, xenstore_devclass, 0, 0);
1274 /*------------------------------- Sysctl Data --------------------------------*/
1275 /* XXX Shouldn't the node be somewhere else? */
1276 SYSCTL_NODE(_dev, OID_AUTO, xen, CTLFLAG_RD, NULL, "Xen");
1277 SYSCTL_INT(_dev_xen, OID_AUTO, xsd_port, CTLFLAG_RD, &xs.evtchn, 0, "");
1278 SYSCTL_ULONG(_dev_xen, OID_AUTO, xsd_kva, CTLFLAG_RD, (u_long *) &xen_store, 0, "");
1280 /*-------------------------------- Public API --------------------------------*/
1281 /*------- API comments for these methods can be found in xenstorevar.h -------*/
1283 xs_directory(struct xs_transaction t, const char *dir, const char *node,
1284 u_int *num, const char ***result)
1291 path = xs_join(dir, node);
1292 error = xs_single(t, XS_DIRECTORY, sbuf_data(path), &len,
1298 *result = split(strings, len, num);
1304 xs_exists(struct xs_transaction t, const char *dir, const char *node)
1309 error = xs_directory(t, dir, node, &dir_n, &d);
1312 free(d, M_XENSTORE);
1317 xs_read(struct xs_transaction t, const char *dir, const char *node,
1318 u_int *len, void **result)
1324 path = xs_join(dir, node);
1325 error = xs_single(t, XS_READ, sbuf_data(path), len, &ret);
1334 xs_write(struct xs_transaction t, const char *dir, const char *node,
1338 struct iovec iovec[2];
1341 path = xs_join(dir, node);
1343 iovec[0].iov_base = (void *)(uintptr_t) sbuf_data(path);
1344 iovec[0].iov_len = sbuf_len(path) + 1;
1345 iovec[1].iov_base = (void *)(uintptr_t) string;
1346 iovec[1].iov_len = strlen(string);
1348 error = xs_talkv(t, XS_WRITE, iovec, 2, NULL, NULL);
1355 xs_mkdir(struct xs_transaction t, const char *dir, const char *node)
1360 path = xs_join(dir, node);
1361 ret = xs_single(t, XS_MKDIR, sbuf_data(path), NULL, NULL);
1368 xs_rm(struct xs_transaction t, const char *dir, const char *node)
1373 path = xs_join(dir, node);
1374 ret = xs_single(t, XS_RM, sbuf_data(path), NULL, NULL);
1381 xs_rm_tree(struct xs_transaction xbt, const char *base, const char *node)
1383 struct xs_transaction local_xbt;
1384 struct sbuf *root_path_sbuf;
1385 struct sbuf *cur_path_sbuf;
1393 root_path_sbuf = xs_join(base, node);
1394 cur_path_sbuf = xs_join(base, node);
1395 root_path = sbuf_data(root_path_sbuf);
1396 cur_path = sbuf_data(cur_path_sbuf);
1401 error = xs_transaction_start(&local_xbt);
1412 error = xs_directory(xbt, cur_path, "", &count, &dir);
1416 for (i = 0; i < count; i++) {
1417 error = xs_rm(xbt, cur_path, dir[i]);
1418 if (error == ENOTEMPTY) {
1419 struct sbuf *push_dir;
1422 * Descend to clear out this sub directory.
1423 * We'll return to cur_dir once push_dir
1426 push_dir = xs_join(cur_path, dir[i]);
1427 sbuf_delete(cur_path_sbuf);
1428 cur_path_sbuf = push_dir;
1429 cur_path = sbuf_data(cur_path_sbuf);
1431 } else if (error != 0) {
1436 free(dir, M_XENSTORE);
1442 /* Directory is empty. It is now safe to remove. */
1443 error = xs_rm(xbt, cur_path, "");
1447 if (!strcmp(cur_path, root_path))
1450 /* Return to processing the parent directory. */
1451 last_slash = strrchr(cur_path, '/');
1452 KASSERT(last_slash != NULL,
1453 ("xs_rm_tree: mangled path %s", cur_path));
1459 sbuf_delete(cur_path_sbuf);
1460 sbuf_delete(root_path_sbuf);
1462 free(dir, M_XENSTORE);
1464 if (local_xbt.id != 0) {
1467 terror = xs_transaction_end(local_xbt, /*abort*/error != 0);
1469 if (terror == EAGAIN && error == 0)
1476 xs_transaction_start(struct xs_transaction *t)
1481 error = xs_single(XST_NIL, XS_TRANSACTION_START, "", NULL,
1484 t->id = strtoul(id_str, NULL, 0);
1485 free(id_str, M_XENSTORE);
1491 xs_transaction_end(struct xs_transaction t, int abort)
1496 strcpy(abortstr, "F");
1498 strcpy(abortstr, "T");
1500 return (xs_single(t, XS_TRANSACTION_END, abortstr, NULL, NULL));
1504 xs_scanf(struct xs_transaction t, const char *dir, const char *node,
1505 int *scancountp, const char *fmt, ...)
1511 error = xs_read(t, dir, node, NULL, (void **) &val);
1516 ns = vsscanf(val, fmt, ap);
1518 free(val, M_XENSTORE);
1519 /* Distinctive errno. */
1528 xs_vprintf(struct xs_transaction t,
1529 const char *dir, const char *node, const char *fmt, va_list ap)
1534 sb = sbuf_new_auto();
1535 sbuf_vprintf(sb, fmt, ap);
1537 error = xs_write(t, dir, node, sbuf_data(sb));
1544 xs_printf(struct xs_transaction t, const char *dir, const char *node,
1545 const char *fmt, ...)
1551 error = xs_vprintf(t, dir, node, fmt, ap);
1558 xs_gather(struct xs_transaction t, const char *dir, ...)
1566 while (error == 0 && (name = va_arg(ap, char *)) != NULL) {
1567 const char *fmt = va_arg(ap, char *);
1568 void *result = va_arg(ap, void *);
1571 error = xs_read(t, dir, name, NULL, (void **) &p);
1576 if (sscanf(p, fmt, result) == 0)
1578 free(p, M_XENSTORE);
1580 *(char **)result = p;
1588 xs_register_watch(struct xs_watch *watch)
1590 /* Pointer in ascii is the token. */
1591 char token[sizeof(watch) * 2 + 1];
1594 sprintf(token, "%lX", (long)watch);
1596 sx_slock(&xs.suspend_mutex);
1598 mtx_lock(&xs.registered_watches_lock);
1599 KASSERT(find_watch(token) == NULL, ("watch already registered"));
1600 LIST_INSERT_HEAD(&xs.registered_watches, watch, list);
1601 mtx_unlock(&xs.registered_watches_lock);
1603 error = xs_watch(watch->node, token);
1605 /* Ignore errors due to multiple registration. */
1606 if (error == EEXIST)
1610 mtx_lock(&xs.registered_watches_lock);
1611 LIST_REMOVE(watch, list);
1612 mtx_unlock(&xs.registered_watches_lock);
1615 sx_sunlock(&xs.suspend_mutex);
1621 xs_unregister_watch(struct xs_watch *watch)
1623 struct xs_stored_msg *msg, *tmp;
1624 char token[sizeof(watch) * 2 + 1];
1627 sprintf(token, "%lX", (long)watch);
1629 sx_slock(&xs.suspend_mutex);
1631 mtx_lock(&xs.registered_watches_lock);
1632 if (find_watch(token) == NULL) {
1633 mtx_unlock(&xs.registered_watches_lock);
1634 sx_sunlock(&xs.suspend_mutex);
1637 LIST_REMOVE(watch, list);
1638 mtx_unlock(&xs.registered_watches_lock);
1640 error = xs_unwatch(watch->node, token);
1642 log(LOG_WARNING, "XENSTORE Failed to release watch %s: %i\n",
1643 watch->node, error);
1645 sx_sunlock(&xs.suspend_mutex);
1647 /* Cancel pending watch events. */
1648 mtx_lock(&xs.watch_events_lock);
1649 TAILQ_FOREACH_SAFE(msg, &xs.watch_events, list, tmp) {
1650 if (msg->u.watch.handle != watch)
1652 TAILQ_REMOVE(&xs.watch_events, msg, list);
1653 free(msg->u.watch.vec, M_XENSTORE);
1654 free(msg, M_XENSTORE);
1656 mtx_unlock(&xs.watch_events_lock);
1658 /* Flush any currently-executing callback, unless we are it. :-) */
1659 if (curproc->p_pid != xs.xenwatch_pid) {
1660 sx_xlock(&xs.xenwatch_mutex);
1661 sx_xunlock(&xs.xenwatch_mutex);