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>
51 #include <sys/queue.h>
52 #include <sys/taskqueue.h>
54 #include <machine/stdarg.h>
56 #include <xen/xen-os.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 hierarchy
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;
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".
240 * The event channel for communicating with the
245 /** Handle for XenStore interrupts. */
246 xen_intr_handle_t xen_intr_handle;
249 * Interrupt driven config hook allowing us to defer
250 * attaching children until interrupts (and thus communication
251 * with the XenStore service) are available.
253 struct intr_config_hook xs_attachcb;
256 * Xenstore is a user-space process that usually runs in Dom0,
257 * so if this domain is booting as Dom0, xenstore wont we accessible,
258 * and we have to defer the initialization of xenstore related
259 * devices to later (when xenstore is started).
264 * Task to run when xenstore is initialized (Dom0 only), will
265 * take care of attaching xenstore related devices.
267 struct task xs_late_init;
270 /*-------------------------------- Global Data ------------------------------*/
271 static struct xs_softc xs;
273 /*------------------------- Private Utility Functions -----------------------*/
276 * Count and optionally record pointers to a number of NUL terminated
277 * strings in a buffer.
279 * \param strings A pointer to a contiguous buffer of NUL terminated strings.
280 * \param dest An array to store pointers to each string found in strings.
281 * \param len The length of the buffer pointed to by strings.
283 * \return A count of the number of strings found.
286 extract_strings(const char *strings, const char **dest, u_int len)
291 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1) {
301 * Convert a contiguous buffer containing a series of NUL terminated
302 * strings into an array of pointers to strings.
304 * The returned pointer references the array of string pointers which
305 * is followed by the storage for the string data. It is the client's
306 * responsibility to free this storage.
308 * The storage addressed by strings is free'd prior to split returning.
310 * \param strings A pointer to a contiguous buffer of NUL terminated strings.
311 * \param len The length of the buffer pointed to by strings.
312 * \param num The number of strings found and returned in the strings
315 * \return An array of pointers to the strings found in the input buffer.
318 split(char *strings, u_int len, u_int *num)
322 /* Protect against unterminated buffers. */
324 strings[len - 1] = '\0';
326 /* Count the strings. */
327 *num = extract_strings(strings, /*dest*/NULL, len);
329 /* Transfer to one big alloc for easy freeing by the caller. */
330 ret = malloc(*num * sizeof(char *) + len, M_XENSTORE, M_WAITOK);
331 memcpy(&ret[*num], strings, len);
332 free(strings, M_XENSTORE);
334 /* Extract pointers to newly allocated array. */
335 strings = (char *)&ret[*num];
336 (void)extract_strings(strings, /*dest*/ret, len);
341 /*------------------------- Public Utility Functions -------------------------*/
342 /*------- API comments for these methods can be found in xenstorevar.h -------*/
344 xs_join(const char *dir, const char *name)
348 sb = sbuf_new_auto();
350 if (name[0] != '\0') {
359 /*-------------------- Low Level Communication Management --------------------*/
361 * Interrupt handler for the XenStore event channel.
363 * XenStore reads and writes block on "xen_store" for buffer
364 * space. Wakeup any blocking operations when the XenStore
365 * service has modified the queues.
368 xs_intr(void * arg __unused /*__attribute__((unused))*/)
371 /* If xenstore has not been initialized, initialize it now */
372 if (!xs.initialized) {
373 xs.initialized = true;
375 * Since this task is probing and attaching devices we
376 * have to hold the Giant lock.
378 taskqueue_enqueue(taskqueue_swi_giant, &xs.xs_late_init);
382 * Hold ring lock across wakeup so that clients
383 * cannot miss a wakeup.
385 mtx_lock(&xs.ring_lock);
387 mtx_unlock(&xs.ring_lock);
391 * Verify that the indexes for a ring are valid.
393 * The difference between the producer and consumer cannot
394 * exceed the size of the ring.
396 * \param cons The consumer index for the ring to test.
397 * \param prod The producer index for the ring to test.
399 * \retval 1 If indexes are in range.
400 * \retval 0 If the indexes are out of range.
403 xs_check_indexes(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod)
406 return ((prod - cons) <= XENSTORE_RING_SIZE);
410 * Return a pointer to, and the length of, the contiguous
411 * free region available for output in 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 storage available.
418 * \return A pointer to the start location of the free region.
421 xs_get_output_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
422 char *buf, uint32_t *len)
425 *len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(prod);
426 if ((XENSTORE_RING_SIZE - (prod - cons)) < *len)
427 *len = XENSTORE_RING_SIZE - (prod - cons);
428 return (buf + MASK_XENSTORE_IDX(prod));
432 * Return a pointer to, and the length of, the contiguous
433 * data available to read from a ring buffer.
435 * \param cons The consumer index for the ring.
436 * \param prod The producer index for the ring.
437 * \param buf The base address of the ring's storage.
438 * \param len The amount of contiguous data available to read.
440 * \return A pointer to the start location of the available data.
443 xs_get_input_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
444 const char *buf, uint32_t *len)
447 *len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(cons);
448 if ((prod - cons) < *len)
450 return (buf + MASK_XENSTORE_IDX(cons));
454 * Transmit data to the XenStore service.
456 * \param tdata A pointer to the contiguous data to send.
457 * \param len The amount of data to send.
459 * \return On success 0, otherwise an errno value indicating the
462 * \invariant Called from thread context.
463 * \invariant The buffer pointed to by tdata is at least len bytes
465 * \invariant xs.request_mutex exclusively locked.
468 xs_write_store(const void *tdata, unsigned len)
470 XENSTORE_RING_IDX cons, prod;
471 const char *data = (const char *)tdata;
474 sx_assert(&xs.request_mutex, SX_XLOCKED);
479 /* Hold lock so we can't miss wakeups should we block. */
480 mtx_lock(&xs.ring_lock);
481 cons = xen_store->req_cons;
482 prod = xen_store->req_prod;
483 if ((prod - cons) == XENSTORE_RING_SIZE) {
485 * Output ring is full. Wait for a ring event.
487 * Note that the events from both queues
488 * are combined, so being woken does not
489 * guarantee that data exist in the read
492 * To simplify error recovery and the retry,
493 * we specify PDROP so our lock is *not* held
494 * when msleep returns.
496 error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
497 "xbwrite", /*timeout*/0);
498 if (error && error != EWOULDBLOCK)
504 mtx_unlock(&xs.ring_lock);
506 /* Verify queue sanity. */
507 if (!xs_check_indexes(cons, prod)) {
508 xen_store->req_cons = xen_store->req_prod = 0;
512 dst = xs_get_output_chunk(cons, prod, xen_store->req, &avail);
516 memcpy(dst, data, avail);
521 * The store to the producer index, which indicates
522 * to the other side that new data has arrived, must
523 * be visible only after our copy of the data into the
524 * ring has completed.
527 xen_store->req_prod += avail;
530 * xen_intr_signal() implies mb(). The other side will see
531 * the change to req_prod at the time of the interrupt.
533 xen_intr_signal(xs.xen_intr_handle);
540 * Receive data from the XenStore service.
542 * \param tdata A pointer to the contiguous buffer to receive the data.
543 * \param len The amount of data to receive.
545 * \return On success 0, otherwise an errno value indicating the
548 * \invariant Called from thread context.
549 * \invariant The buffer pointed to by tdata is at least len bytes
552 * \note xs_read does not perform any internal locking to guarantee
553 * serial access to the incoming ring buffer. However, there
554 * is only one context processing reads: xs_rcv_thread().
557 xs_read_store(void *tdata, unsigned len)
559 XENSTORE_RING_IDX cons, prod;
560 char *data = (char *)tdata;
567 /* Hold lock so we can't miss wakeups should we block. */
568 mtx_lock(&xs.ring_lock);
569 cons = xen_store->rsp_cons;
570 prod = xen_store->rsp_prod;
573 * Nothing to read. Wait for a ring event.
575 * Note that the events from both queues
576 * are combined, so being woken does not
577 * guarantee that data exist in the read
580 * To simplify error recovery and the retry,
581 * we specify PDROP so our lock is *not* held
582 * when msleep returns.
584 error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
585 "xbread", /*timeout*/0);
586 if (error && error != EWOULDBLOCK)
590 mtx_unlock(&xs.ring_lock);
592 /* Verify queue sanity. */
593 if (!xs_check_indexes(cons, prod)) {
594 xen_store->rsp_cons = xen_store->rsp_prod = 0;
598 src = xs_get_input_chunk(cons, prod, xen_store->rsp, &avail);
603 * Insure the data we read is related to the indexes
608 memcpy(data, src, avail);
613 * Insure that the producer of this ring does not see
614 * the ring space as free until after we have copied it
618 xen_store->rsp_cons += avail;
621 * xen_intr_signal() implies mb(). The producer will see
622 * the updated consumer index when the event is delivered.
624 xen_intr_signal(xs.xen_intr_handle);
630 /*----------------------- Received Message Processing ------------------------*/
632 * Block reading the next message from the XenStore service and
633 * process the result.
635 * \param type The returned type of the XenStore message received.
637 * \return 0 on success. Otherwise an errno value indicating the
638 * type of failure encountered.
641 xs_process_msg(enum xsd_sockmsg_type *type)
643 struct xs_stored_msg *msg;
647 msg = malloc(sizeof(*msg), M_XENSTORE, M_WAITOK);
648 error = xs_read_store(&msg->hdr, sizeof(msg->hdr));
650 free(msg, M_XENSTORE);
654 body = malloc(msg->hdr.len + 1, M_XENSTORE, M_WAITOK);
655 error = xs_read_store(body, msg->hdr.len);
657 free(body, M_XENSTORE);
658 free(msg, M_XENSTORE);
661 body[msg->hdr.len] = '\0';
663 *type = msg->hdr.type;
664 if (msg->hdr.type == XS_WATCH_EVENT) {
665 msg->u.watch.vec = split(body, msg->hdr.len,
666 &msg->u.watch.vec_size);
668 mtx_lock(&xs.registered_watches_lock);
669 msg->u.watch.handle = find_watch(
670 msg->u.watch.vec[XS_WATCH_TOKEN]);
671 if (msg->u.watch.handle != NULL) {
672 mtx_lock(&xs.watch_events_lock);
673 TAILQ_INSERT_TAIL(&xs.watch_events, msg, list);
674 wakeup(&xs.watch_events);
675 mtx_unlock(&xs.watch_events_lock);
677 free(msg->u.watch.vec, M_XENSTORE);
678 free(msg, M_XENSTORE);
680 mtx_unlock(&xs.registered_watches_lock);
682 msg->u.reply.body = body;
683 mtx_lock(&xs.reply_lock);
684 TAILQ_INSERT_TAIL(&xs.reply_list, msg, list);
685 wakeup(&xs.reply_list);
686 mtx_unlock(&xs.reply_lock);
693 * Thread body of the XenStore receive thread.
695 * This thread blocks waiting for data from the XenStore service
696 * and processes and received messages.
699 xs_rcv_thread(void *arg __unused)
702 enum xsd_sockmsg_type type;
705 error = xs_process_msg(&type);
707 printf("XENSTORE error %d while reading message\n",
712 /*---------------- XenStore Message Request/Reply Processing -----------------*/
714 * Filter invoked before transmitting any message to the XenStore service.
716 * The role of the filter may expand, but currently serves to manage
717 * the interactions of messages with transaction state.
719 * \param request_msg_type The message type for the request.
722 xs_request_filter(uint32_t request_msg_type)
724 if (request_msg_type == XS_TRANSACTION_START)
725 sx_slock(&xs.suspend_mutex);
729 * Filter invoked after transmitting any message to the XenStore service.
731 * The role of the filter may expand, but currently serves to manage
732 * the interactions of messages with transaction state.
734 * \param request_msg_type The message type for the original request.
735 * \param reply_msg_type The message type for any received reply.
736 * \param request_reply_error The error status from the attempt to send
737 * the request or retrieve the reply.
740 xs_reply_filter(uint32_t request_msg_type,
741 uint32_t reply_msg_type, int request_reply_error)
744 * The count of transactions drops if we attempted
745 * to end a transaction (even if that attempt fails
746 * in error), we receive a transaction end acknowledgement,
747 * or if our attempt to begin a transaction fails.
749 if (request_msg_type == XS_TRANSACTION_END
750 || (request_reply_error == 0 && reply_msg_type == XS_TRANSACTION_END)
751 || (request_msg_type == XS_TRANSACTION_START
752 && (request_reply_error != 0 || reply_msg_type == XS_ERROR)))
753 sx_sunlock(&xs.suspend_mutex);
757 #define xsd_error_count (sizeof(xsd_errors) / sizeof(xsd_errors[0]))
760 * Convert a XenStore error string into an errno number.
762 * \param errorstring The error string to convert.
764 * \return The errno best matching the input string.
766 * \note Unknown error strings are converted to EINVAL.
769 xs_get_error(const char *errorstring)
773 for (i = 0; i < xsd_error_count; i++) {
774 if (!strcmp(errorstring, xsd_errors[i].errstring))
775 return (xsd_errors[i].errnum);
777 log(LOG_WARNING, "XENSTORE xen store gave: unknown error %s",
783 * Block waiting for a reply to a message request.
785 * \param type The returned type of the reply.
786 * \param len The returned body length of the reply.
787 * \param result The returned body of the reply.
789 * \return 0 on success. Otherwise an errno indicating the
793 xs_read_reply(enum xsd_sockmsg_type *type, u_int *len, void **result)
795 struct xs_stored_msg *msg;
799 mtx_lock(&xs.reply_lock);
800 while (TAILQ_EMPTY(&xs.reply_list)) {
801 error = mtx_sleep(&xs.reply_list, &xs.reply_lock,
802 PCATCH, "xswait", hz/10);
803 if (error && error != EWOULDBLOCK) {
804 mtx_unlock(&xs.reply_lock);
808 msg = TAILQ_FIRST(&xs.reply_list);
809 TAILQ_REMOVE(&xs.reply_list, msg, list);
810 mtx_unlock(&xs.reply_lock);
812 *type = msg->hdr.type;
815 body = msg->u.reply.body;
817 free(msg, M_XENSTORE);
823 * Pass-thru interface for XenStore access by userland processes
824 * via the XenStore device.
826 * Reply type and length data are returned by overwriting these
827 * fields in the passed in request message.
829 * \param msg A properly formatted message to transmit to
830 * the XenStore service.
831 * \param result The returned body of the reply.
833 * \return 0 on success. Otherwise an errno indicating the cause
836 * \note The returned result is provided in malloced storage and thus
837 * must be free'd by the caller with 'free(result, M_XENSTORE);
840 xs_dev_request_and_reply(struct xsd_sockmsg *msg, void **result)
842 uint32_t request_type;
845 request_type = msg->type;
846 xs_request_filter(request_type);
848 sx_xlock(&xs.request_mutex);
849 if ((error = xs_write_store(msg, sizeof(*msg) + msg->len)) == 0)
850 error = xs_read_reply(&msg->type, &msg->len, result);
851 sx_xunlock(&xs.request_mutex);
853 xs_reply_filter(request_type, msg->type, error);
859 * Send a message with an optionally muti-part body to the XenStore service.
861 * \param t The transaction to use for this request.
862 * \param request_type The type of message to send.
863 * \param iovec Pointers to the body sections of the request.
864 * \param num_vecs The number of body sections in the request.
865 * \param len The returned length of the reply.
866 * \param result The returned body of the reply.
868 * \return 0 on success. Otherwise an errno indicating
869 * the cause of failure.
871 * \note The returned result is provided in malloced storage and thus
872 * must be free'd by the caller with 'free(*result, M_XENSTORE);
875 xs_talkv(struct xs_transaction t, enum xsd_sockmsg_type request_type,
876 const struct iovec *iovec, u_int num_vecs, u_int *len, void **result)
878 struct xsd_sockmsg msg;
885 msg.type = request_type;
887 for (i = 0; i < num_vecs; i++)
888 msg.len += iovec[i].iov_len;
890 xs_request_filter(request_type);
892 sx_xlock(&xs.request_mutex);
893 error = xs_write_store(&msg, sizeof(msg));
895 printf("xs_talkv failed %d\n", error);
896 goto error_lock_held;
899 for (i = 0; i < num_vecs; i++) {
900 error = xs_write_store(iovec[i].iov_base, iovec[i].iov_len);
902 printf("xs_talkv failed %d\n", error);
903 goto error_lock_held;
907 error = xs_read_reply(&msg.type, len, &ret);
910 sx_xunlock(&xs.request_mutex);
911 xs_reply_filter(request_type, msg.type, error);
915 if (msg.type == XS_ERROR) {
916 error = xs_get_error(ret);
917 free(ret, M_XENSTORE);
921 /* Reply is either error or an echo of our request message type. */
922 KASSERT(msg.type == request_type, ("bad xenstore message type"));
927 free(ret, M_XENSTORE);
933 * Wrapper for xs_talkv allowing easy transmission of a message with
934 * a single, contiguous, message body.
936 * \param t The transaction to use for this request.
937 * \param request_type The type of message to send.
938 * \param body The body of the request.
939 * \param len The returned length of the reply.
940 * \param result The returned body of the reply.
942 * \return 0 on success. Otherwise an errno indicating
943 * the cause of failure.
945 * \note The returned result is provided in malloced storage and thus
946 * must be free'd by the caller with 'free(*result, M_XENSTORE);
949 xs_single(struct xs_transaction t, enum xsd_sockmsg_type request_type,
950 const char *body, u_int *len, void **result)
954 iovec.iov_base = (void *)(uintptr_t)body;
955 iovec.iov_len = strlen(body) + 1;
957 return (xs_talkv(t, request_type, &iovec, 1, len, result));
960 /*------------------------- XenStore Watch Support ---------------------------*/
962 * Transmit a watch 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_watch(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_WATCH, iov, 2, NULL, NULL));
984 * Transmit an uwatch request to the XenStore service.
986 * \param path The path in the XenStore to watch.
987 * \param tocken A unique identifier for this watch.
989 * \return 0 on success. Otherwise an errno indicating the
993 xs_unwatch(const char *path, const char *token)
997 iov[0].iov_base = (void *)(uintptr_t) path;
998 iov[0].iov_len = strlen(path) + 1;
999 iov[1].iov_base = (void *)(uintptr_t) token;
1000 iov[1].iov_len = strlen(token) + 1;
1002 return (xs_talkv(XST_NIL, XS_UNWATCH, iov, 2, NULL, NULL));
1006 * Convert from watch token (unique identifier) to the associated
1007 * internal tracking structure for this watch.
1009 * \param tocken The unique identifier for the watch to find.
1011 * \return A pointer to the found watch structure or NULL.
1013 static struct xs_watch *
1014 find_watch(const char *token)
1016 struct xs_watch *i, *cmp;
1018 cmp = (void *)strtoul(token, NULL, 16);
1020 LIST_FOREACH(i, &xs.registered_watches, list)
1028 * Thread body of the XenStore watch event dispatch thread.
1031 xenwatch_thread(void *unused)
1033 struct xs_stored_msg *msg;
1037 mtx_lock(&xs.watch_events_lock);
1038 while (TAILQ_EMPTY(&xs.watch_events))
1039 mtx_sleep(&xs.watch_events,
1040 &xs.watch_events_lock,
1041 PWAIT | PCATCH, "waitev", hz/10);
1043 mtx_unlock(&xs.watch_events_lock);
1044 sx_xlock(&xs.xenwatch_mutex);
1046 mtx_lock(&xs.watch_events_lock);
1047 msg = TAILQ_FIRST(&xs.watch_events);
1049 TAILQ_REMOVE(&xs.watch_events, msg, list);
1050 mtx_unlock(&xs.watch_events_lock);
1054 * XXX There are messages coming in with a NULL
1055 * XXX callback. This deserves further investigation;
1056 * XXX the workaround here simply prevents the kernel
1057 * XXX from panic'ing on startup.
1059 if (msg->u.watch.handle->callback != NULL)
1060 msg->u.watch.handle->callback(
1061 msg->u.watch.handle,
1062 (const char **)msg->u.watch.vec,
1063 msg->u.watch.vec_size);
1064 free(msg->u.watch.vec, M_XENSTORE);
1065 free(msg, M_XENSTORE);
1068 sx_xunlock(&xs.xenwatch_mutex);
1072 /*----------- XenStore Configuration, Initialization, and Control ------------*/
1074 * Setup communication channels with the XenStore service.
1076 * \return On success, 0. Otherwise an errno value indicating the
1084 if (xen_store->rsp_prod != xen_store->rsp_cons) {
1085 log(LOG_WARNING, "XENSTORE response ring is not quiescent "
1086 "(%08x:%08x): fixing up\n",
1087 xen_store->rsp_cons, xen_store->rsp_prod);
1088 xen_store->rsp_cons = xen_store->rsp_prod;
1091 xen_intr_unbind(&xs.xen_intr_handle);
1093 error = xen_intr_bind_local_port(xs.xs_dev, xs.evtchn,
1094 /*filter*/NULL, xs_intr, /*arg*/NULL, INTR_TYPE_NET|INTR_MPSAFE,
1095 &xs.xen_intr_handle);
1097 log(LOG_WARNING, "XENSTORE request irq failed %i\n", error);
1104 /*------------------ Private Device Attachment Functions --------------------*/
1106 xs_identify(driver_t *driver, device_t parent)
1109 BUS_ADD_CHILD(parent, 0, "xenstore", 0);
1113 * Probe for the existence of the XenStore.
1118 xs_probe(device_t dev)
1121 * We are either operating within a PV kernel or being probed
1122 * as the child of the successfully attached xenpci device.
1123 * Thus we are in a Xen environment and there will be a XenStore.
1124 * Unconditionally return success.
1126 device_set_desc(dev, "XenStore");
1127 return (BUS_PROBE_NOWILDCARD);
1131 xs_attach_deferred(void *arg)
1134 bus_generic_probe(xs.xs_dev);
1135 bus_generic_attach(xs.xs_dev);
1137 config_intrhook_disestablish(&xs.xs_attachcb);
1141 xs_attach_late(void *arg, int pending)
1144 KASSERT((pending == 1), ("xs late attach queued several times"));
1145 bus_generic_probe(xs.xs_dev);
1146 bus_generic_attach(xs.xs_dev);
1150 * Attach to the XenStore.
1152 * This routine also prepares for the probe/attach of drivers that rely
1156 xs_attach(device_t dev)
1160 /* Allow us to get device_t from softc and vice-versa. */
1162 device_set_softc(dev, &xs);
1164 /* Initialize the interface to xenstore. */
1167 xs.initialized = false;
1168 if (xen_hvm_domain()) {
1169 xs.evtchn = hvm_get_parameter(HVM_PARAM_STORE_EVTCHN);
1170 xs.gpfn = hvm_get_parameter(HVM_PARAM_STORE_PFN);
1171 xen_store = pmap_mapdev(xs.gpfn * PAGE_SIZE, PAGE_SIZE);
1172 xs.initialized = true;
1173 } else if (xen_pv_domain()) {
1174 if (HYPERVISOR_start_info->store_evtchn == 0) {
1175 struct evtchn_alloc_unbound alloc_unbound;
1177 /* Allocate a local event channel for xenstore */
1178 alloc_unbound.dom = DOMID_SELF;
1179 alloc_unbound.remote_dom = DOMID_SELF;
1180 error = HYPERVISOR_event_channel_op(
1181 EVTCHNOP_alloc_unbound, &alloc_unbound);
1184 "unable to alloc event channel for Dom0: %d",
1187 HYPERVISOR_start_info->store_evtchn =
1189 xs.evtchn = alloc_unbound.port;
1191 /* Allocate memory for the xs shared ring */
1192 xen_store = malloc(PAGE_SIZE, M_XENSTORE,
1195 xs.evtchn = HYPERVISOR_start_info->store_evtchn;
1196 xs.initialized = true;
1199 panic("Unknown domain type, cannot initialize xenstore.");
1202 TAILQ_INIT(&xs.reply_list);
1203 TAILQ_INIT(&xs.watch_events);
1205 mtx_init(&xs.ring_lock, "ring lock", NULL, MTX_DEF);
1206 mtx_init(&xs.reply_lock, "reply lock", NULL, MTX_DEF);
1207 sx_init(&xs.xenwatch_mutex, "xenwatch");
1208 sx_init(&xs.request_mutex, "xenstore request");
1209 sx_init(&xs.suspend_mutex, "xenstore suspend");
1210 mtx_init(&xs.registered_watches_lock, "watches", NULL, MTX_DEF);
1211 mtx_init(&xs.watch_events_lock, "watch events", NULL, MTX_DEF);
1213 /* Initialize the shared memory rings to talk to xenstored */
1214 error = xs_init_comms();
1218 error = kproc_create(xenwatch_thread, NULL, &p, RFHIGHPID,
1222 xs.xenwatch_pid = p->p_pid;
1224 error = kproc_create(xs_rcv_thread, NULL, NULL,
1225 RFHIGHPID, 0, "xenstore_rcv");
1227 xs.xs_attachcb.ich_func = xs_attach_deferred;
1228 xs.xs_attachcb.ich_arg = NULL;
1229 if (xs.initialized) {
1230 config_intrhook_establish(&xs.xs_attachcb);
1232 TASK_INIT(&xs.xs_late_init, 0, xs_attach_late, NULL);
1239 * Prepare for suspension of this VM by halting XenStore access after
1240 * all transactions and individual requests have completed.
1243 xs_suspend(device_t dev)
1247 /* Suspend child Xen devices. */
1248 error = bus_generic_suspend(dev);
1252 sx_xlock(&xs.suspend_mutex);
1253 sx_xlock(&xs.request_mutex);
1259 * Resume XenStore operations after this VM is resumed.
1262 xs_resume(device_t dev __unused)
1264 struct xs_watch *watch;
1265 char token[sizeof(watch) * 2 + 1];
1269 sx_xunlock(&xs.request_mutex);
1272 * No need for registered_watches_lock: the suspend_mutex
1275 LIST_FOREACH(watch, &xs.registered_watches, list) {
1276 sprintf(token, "%lX", (long)watch);
1277 xs_watch(watch->node, token);
1280 sx_xunlock(&xs.suspend_mutex);
1282 /* Resume child Xen devices. */
1283 bus_generic_resume(dev);
1288 /*-------------------- Private Device Attachment Data -----------------------*/
1289 static device_method_t xenstore_methods[] = {
1290 /* Device interface */
1291 DEVMETHOD(device_identify, xs_identify),
1292 DEVMETHOD(device_probe, xs_probe),
1293 DEVMETHOD(device_attach, xs_attach),
1294 DEVMETHOD(device_detach, bus_generic_detach),
1295 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1296 DEVMETHOD(device_suspend, xs_suspend),
1297 DEVMETHOD(device_resume, xs_resume),
1300 DEVMETHOD(bus_add_child, bus_generic_add_child),
1301 DEVMETHOD(bus_alloc_resource, bus_generic_alloc_resource),
1302 DEVMETHOD(bus_release_resource, bus_generic_release_resource),
1303 DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
1304 DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
1309 DEFINE_CLASS_0(xenstore, xenstore_driver, xenstore_methods, 0);
1310 static devclass_t xenstore_devclass;
1312 DRIVER_MODULE(xenstore, xenpv, xenstore_driver, xenstore_devclass, 0, 0);
1314 /*------------------------------- Sysctl Data --------------------------------*/
1315 /* XXX Shouldn't the node be somewhere else? */
1316 SYSCTL_NODE(_dev, OID_AUTO, xen, CTLFLAG_RD, NULL, "Xen");
1317 SYSCTL_INT(_dev_xen, OID_AUTO, xsd_port, CTLFLAG_RD, &xs.evtchn, 0, "");
1318 SYSCTL_ULONG(_dev_xen, OID_AUTO, xsd_kva, CTLFLAG_RD, (u_long *) &xen_store, 0, "");
1320 /*-------------------------------- Public API --------------------------------*/
1321 /*------- API comments for these methods can be found in xenstorevar.h -------*/
1323 xs_directory(struct xs_transaction t, const char *dir, const char *node,
1324 u_int *num, const char ***result)
1331 path = xs_join(dir, node);
1332 error = xs_single(t, XS_DIRECTORY, sbuf_data(path), &len,
1338 *result = split(strings, len, num);
1344 xs_exists(struct xs_transaction t, const char *dir, const char *node)
1349 error = xs_directory(t, dir, node, &dir_n, &d);
1352 free(d, M_XENSTORE);
1357 xs_read(struct xs_transaction t, const char *dir, const char *node,
1358 u_int *len, void **result)
1364 path = xs_join(dir, node);
1365 error = xs_single(t, XS_READ, sbuf_data(path), len, &ret);
1374 xs_write(struct xs_transaction t, const char *dir, const char *node,
1378 struct iovec iovec[2];
1381 path = xs_join(dir, node);
1383 iovec[0].iov_base = (void *)(uintptr_t) sbuf_data(path);
1384 iovec[0].iov_len = sbuf_len(path) + 1;
1385 iovec[1].iov_base = (void *)(uintptr_t) string;
1386 iovec[1].iov_len = strlen(string);
1388 error = xs_talkv(t, XS_WRITE, iovec, 2, NULL, NULL);
1395 xs_mkdir(struct xs_transaction t, const char *dir, const char *node)
1400 path = xs_join(dir, node);
1401 ret = xs_single(t, XS_MKDIR, sbuf_data(path), NULL, NULL);
1408 xs_rm(struct xs_transaction t, const char *dir, const char *node)
1413 path = xs_join(dir, node);
1414 ret = xs_single(t, XS_RM, sbuf_data(path), NULL, NULL);
1421 xs_rm_tree(struct xs_transaction xbt, const char *base, const char *node)
1423 struct xs_transaction local_xbt;
1424 struct sbuf *root_path_sbuf;
1425 struct sbuf *cur_path_sbuf;
1432 root_path_sbuf = xs_join(base, node);
1433 cur_path_sbuf = xs_join(base, node);
1434 root_path = sbuf_data(root_path_sbuf);
1435 cur_path = sbuf_data(cur_path_sbuf);
1440 error = xs_transaction_start(&local_xbt);
1450 error = xs_directory(xbt, cur_path, "", &count, &dir);
1454 for (i = 0; i < count; i++) {
1455 error = xs_rm(xbt, cur_path, dir[i]);
1456 if (error == ENOTEMPTY) {
1457 struct sbuf *push_dir;
1460 * Descend to clear out this sub directory.
1461 * We'll return to cur_dir once push_dir
1464 push_dir = xs_join(cur_path, dir[i]);
1465 sbuf_delete(cur_path_sbuf);
1466 cur_path_sbuf = push_dir;
1467 cur_path = sbuf_data(cur_path_sbuf);
1469 } else if (error != 0) {
1474 free(dir, M_XENSTORE);
1480 /* Directory is empty. It is now safe to remove. */
1481 error = xs_rm(xbt, cur_path, "");
1485 if (!strcmp(cur_path, root_path))
1488 /* Return to processing the parent directory. */
1489 last_slash = strrchr(cur_path, '/');
1490 KASSERT(last_slash != NULL,
1491 ("xs_rm_tree: mangled path %s", cur_path));
1497 sbuf_delete(cur_path_sbuf);
1498 sbuf_delete(root_path_sbuf);
1500 free(dir, M_XENSTORE);
1502 if (local_xbt.id != 0) {
1505 terror = xs_transaction_end(local_xbt, /*abort*/error != 0);
1507 if (terror == EAGAIN && error == 0)
1514 xs_transaction_start(struct xs_transaction *t)
1519 error = xs_single(XST_NIL, XS_TRANSACTION_START, "", NULL,
1522 t->id = strtoul(id_str, NULL, 0);
1523 free(id_str, M_XENSTORE);
1529 xs_transaction_end(struct xs_transaction t, int abort)
1534 strcpy(abortstr, "F");
1536 strcpy(abortstr, "T");
1538 return (xs_single(t, XS_TRANSACTION_END, abortstr, NULL, NULL));
1542 xs_scanf(struct xs_transaction t, const char *dir, const char *node,
1543 int *scancountp, const char *fmt, ...)
1549 error = xs_read(t, dir, node, NULL, (void **) &val);
1554 ns = vsscanf(val, fmt, ap);
1556 free(val, M_XENSTORE);
1557 /* Distinctive errno. */
1566 xs_vprintf(struct xs_transaction t,
1567 const char *dir, const char *node, const char *fmt, va_list ap)
1572 sb = sbuf_new_auto();
1573 sbuf_vprintf(sb, fmt, ap);
1575 error = xs_write(t, dir, node, sbuf_data(sb));
1582 xs_printf(struct xs_transaction t, const char *dir, const char *node,
1583 const char *fmt, ...)
1589 error = xs_vprintf(t, dir, node, fmt, ap);
1596 xs_gather(struct xs_transaction t, const char *dir, ...)
1604 while (error == 0 && (name = va_arg(ap, char *)) != NULL) {
1605 const char *fmt = va_arg(ap, char *);
1606 void *result = va_arg(ap, void *);
1609 error = xs_read(t, dir, name, NULL, (void **) &p);
1614 if (sscanf(p, fmt, result) == 0)
1616 free(p, M_XENSTORE);
1618 *(char **)result = p;
1626 xs_register_watch(struct xs_watch *watch)
1628 /* Pointer in ascii is the token. */
1629 char token[sizeof(watch) * 2 + 1];
1632 sprintf(token, "%lX", (long)watch);
1634 sx_slock(&xs.suspend_mutex);
1636 mtx_lock(&xs.registered_watches_lock);
1637 KASSERT(find_watch(token) == NULL, ("watch already registered"));
1638 LIST_INSERT_HEAD(&xs.registered_watches, watch, list);
1639 mtx_unlock(&xs.registered_watches_lock);
1641 error = xs_watch(watch->node, token);
1643 /* Ignore errors due to multiple registration. */
1644 if (error == EEXIST)
1648 mtx_lock(&xs.registered_watches_lock);
1649 LIST_REMOVE(watch, list);
1650 mtx_unlock(&xs.registered_watches_lock);
1653 sx_sunlock(&xs.suspend_mutex);
1659 xs_unregister_watch(struct xs_watch *watch)
1661 struct xs_stored_msg *msg, *tmp;
1662 char token[sizeof(watch) * 2 + 1];
1665 sprintf(token, "%lX", (long)watch);
1667 sx_slock(&xs.suspend_mutex);
1669 mtx_lock(&xs.registered_watches_lock);
1670 if (find_watch(token) == NULL) {
1671 mtx_unlock(&xs.registered_watches_lock);
1672 sx_sunlock(&xs.suspend_mutex);
1675 LIST_REMOVE(watch, list);
1676 mtx_unlock(&xs.registered_watches_lock);
1678 error = xs_unwatch(watch->node, token);
1680 log(LOG_WARNING, "XENSTORE Failed to release watch %s: %i\n",
1681 watch->node, error);
1683 sx_sunlock(&xs.suspend_mutex);
1685 /* Cancel pending watch events. */
1686 mtx_lock(&xs.watch_events_lock);
1687 TAILQ_FOREACH_SAFE(msg, &xs.watch_events, list, tmp) {
1688 if (msg->u.watch.handle != watch)
1690 TAILQ_REMOVE(&xs.watch_events, msg, list);
1691 free(msg->u.watch.vec, M_XENSTORE);
1692 free(msg, M_XENSTORE);
1694 mtx_unlock(&xs.watch_events_lock);
1696 /* Flush any currently-executing callback, unless we are it. :-) */
1697 if (curproc->p_pid != xs.xenwatch_pid) {
1698 sx_xlock(&xs.xenwatch_mutex);
1699 sx_xunlock(&xs.xenwatch_mutex);
1707 sx_xlock(&xs.request_mutex);
1715 sx_xunlock(&xs.request_mutex);