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
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
36 #include <sys/kernel.h>
38 #include <sys/module.h>
39 #include <sys/mutex.h>
41 #include <sys/syslog.h>
42 #include <sys/malloc.h>
43 #include <sys/systm.h>
45 #include <sys/kthread.h>
47 #include <sys/sysctl.h>
49 #include <sys/unistd.h>
50 #include <sys/queue.h>
51 #include <sys/taskqueue.h>
53 #include <machine/stdarg.h>
55 #include <xen/xen-os.h>
56 #include <xen/hypervisor.h>
57 #include <xen/xen_intr.h>
59 #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 hierarchy
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 static 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 * The processid of the xenwatch thread.
209 * Sleepable mutex used to gate the execution of XenStore
210 * watch event callbacks.
212 * xenwatch_thread holds an exclusive lock on this mutex
213 * while delivering event callbacks, and xenstore_unregister_watch()
214 * uses an exclusive lock of this mutex to guarantee that no
215 * callbacks of the just unregistered watch are pending
216 * before returning to its caller.
218 struct sx xenwatch_mutex;
221 * The HVM guest pseudo-physical frame number. This is Xen's mapping
222 * of the true machine frame number into our "physical address space".
227 * The event channel for communicating with the
232 /** Handle for XenStore interrupts. */
233 xen_intr_handle_t xen_intr_handle;
236 * Interrupt driven config hook allowing us to defer
237 * attaching children until interrupts (and thus communication
238 * with the XenStore service) are available.
240 struct intr_config_hook xs_attachcb;
243 * Xenstore is a user-space process that usually runs in Dom0,
244 * so if this domain is booting as Dom0, xenstore wont we accessible,
245 * and we have to defer the initialization of xenstore related
246 * devices to later (when xenstore is started).
251 * Task to run when xenstore is initialized (Dom0 only), will
252 * take care of attaching xenstore related devices.
254 struct task xs_late_init;
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. */
311 strings[len - 1] = '\0';
313 /* Count the strings. */
314 *num = extract_strings(strings, /*dest*/NULL, len);
316 /* Transfer to one big alloc for easy freeing by the caller. */
317 ret = malloc(*num * sizeof(char *) + len, M_XENSTORE, M_WAITOK);
318 memcpy(&ret[*num], strings, len);
319 free(strings, M_XENSTORE);
321 /* Extract pointers to newly allocated array. */
322 strings = (char *)&ret[*num];
323 (void)extract_strings(strings, /*dest*/ret, len);
328 /*------------------------- Public Utility Functions -------------------------*/
329 /*------- API comments for these methods can be found in xenstorevar.h -------*/
331 xs_join(const char *dir, const char *name)
335 sb = sbuf_new_auto();
337 if (name[0] != '\0') {
346 /*-------------------- Low Level Communication Management --------------------*/
348 * Interrupt handler for the XenStore event channel.
350 * XenStore reads and writes block on "xen_store" for buffer
351 * space. Wakeup any blocking operations when the XenStore
352 * service has modified the queues.
355 xs_intr(void * arg __unused /*__attribute__((unused))*/)
358 /* If xenstore has not been initialized, initialize it now */
359 if (!xs.initialized) {
360 xs.initialized = true;
362 * Since this task is probing and attaching devices we
363 * have to hold the Giant lock.
365 taskqueue_enqueue(taskqueue_swi_giant, &xs.xs_late_init);
369 * Hold ring lock across wakeup so that clients
370 * cannot miss a wakeup.
372 mtx_lock(&xs.ring_lock);
374 mtx_unlock(&xs.ring_lock);
378 * Verify that the indexes for a ring are valid.
380 * The difference between the producer and consumer cannot
381 * exceed the size of the ring.
383 * \param cons The consumer index for the ring to test.
384 * \param prod The producer index for the ring to test.
386 * \retval 1 If indexes are in range.
387 * \retval 0 If the indexes are out of range.
390 xs_check_indexes(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod)
393 return ((prod - cons) <= XENSTORE_RING_SIZE);
397 * Return a pointer to, and the length of, the contiguous
398 * free region available for output in a ring buffer.
400 * \param cons The consumer index for the ring.
401 * \param prod The producer index for the ring.
402 * \param buf The base address of the ring's storage.
403 * \param len The amount of contiguous storage available.
405 * \return A pointer to the start location of the free region.
408 xs_get_output_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
409 char *buf, uint32_t *len)
412 *len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(prod);
413 if ((XENSTORE_RING_SIZE - (prod - cons)) < *len)
414 *len = XENSTORE_RING_SIZE - (prod - cons);
415 return (buf + MASK_XENSTORE_IDX(prod));
419 * Return a pointer to, and the length of, the contiguous
420 * data available to read from a ring buffer.
422 * \param cons The consumer index for the ring.
423 * \param prod The producer index for the ring.
424 * \param buf The base address of the ring's storage.
425 * \param len The amount of contiguous data available to read.
427 * \return A pointer to the start location of the available data.
430 xs_get_input_chunk(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod,
431 const char *buf, uint32_t *len)
434 *len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(cons);
435 if ((prod - cons) < *len)
437 return (buf + MASK_XENSTORE_IDX(cons));
441 * Transmit data to the XenStore service.
443 * \param tdata A pointer to the contiguous data to send.
444 * \param len The amount of data to send.
446 * \return On success 0, otherwise an errno value indicating the
449 * \invariant Called from thread context.
450 * \invariant The buffer pointed to by tdata is at least len bytes
452 * \invariant xs.request_mutex exclusively locked.
455 xs_write_store(const void *tdata, unsigned len)
457 XENSTORE_RING_IDX cons, prod;
458 const char *data = (const char *)tdata;
461 sx_assert(&xs.request_mutex, SX_XLOCKED);
466 /* Hold lock so we can't miss wakeups should we block. */
467 mtx_lock(&xs.ring_lock);
468 cons = xen_store->req_cons;
469 prod = xen_store->req_prod;
470 if ((prod - cons) == XENSTORE_RING_SIZE) {
472 * Output ring is full. Wait for a ring event.
474 * Note that the events from both queues
475 * are combined, so being woken does not
476 * guarantee that data exist in the read
479 * To simplify error recovery and the retry,
480 * we specify PDROP so our lock is *not* held
481 * when msleep returns.
483 error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
484 "xbwrite", /*timeout*/0);
485 if (error && error != EWOULDBLOCK)
491 mtx_unlock(&xs.ring_lock);
493 /* Verify queue sanity. */
494 if (!xs_check_indexes(cons, prod)) {
495 xen_store->req_cons = xen_store->req_prod = 0;
499 dst = xs_get_output_chunk(cons, prod, xen_store->req, &avail);
503 memcpy(dst, data, avail);
508 * The store to the producer index, which indicates
509 * to the other side that new data has arrived, must
510 * be visible only after our copy of the data into the
511 * ring has completed.
514 xen_store->req_prod += avail;
517 * xen_intr_signal() implies mb(). The other side will see
518 * the change to req_prod at the time of the interrupt.
520 xen_intr_signal(xs.xen_intr_handle);
527 * Receive data from the XenStore service.
529 * \param tdata A pointer to the contiguous buffer to receive the data.
530 * \param len The amount of data to receive.
532 * \return On success 0, otherwise an errno value indicating the
535 * \invariant Called from thread context.
536 * \invariant The buffer pointed to by tdata is at least len bytes
539 * \note xs_read does not perform any internal locking to guarantee
540 * serial access to the incoming ring buffer. However, there
541 * is only one context processing reads: xs_rcv_thread().
544 xs_read_store(void *tdata, unsigned len)
546 XENSTORE_RING_IDX cons, prod;
547 char *data = (char *)tdata;
554 /* Hold lock so we can't miss wakeups should we block. */
555 mtx_lock(&xs.ring_lock);
556 cons = xen_store->rsp_cons;
557 prod = xen_store->rsp_prod;
560 * Nothing to read. Wait for a ring event.
562 * Note that the events from both queues
563 * are combined, so being woken does not
564 * guarantee that data exist in the read
567 * To simplify error recovery and the retry,
568 * we specify PDROP so our lock is *not* held
569 * when msleep returns.
571 error = msleep(xen_store, &xs.ring_lock, PCATCH|PDROP,
572 "xbread", /*timeout*/0);
573 if (error && error != EWOULDBLOCK)
577 mtx_unlock(&xs.ring_lock);
579 /* Verify queue sanity. */
580 if (!xs_check_indexes(cons, prod)) {
581 xen_store->rsp_cons = xen_store->rsp_prod = 0;
585 src = xs_get_input_chunk(cons, prod, xen_store->rsp, &avail);
590 * Insure the data we read is related to the indexes
595 memcpy(data, src, avail);
600 * Insure that the producer of this ring does not see
601 * the ring space as free until after we have copied it
605 xen_store->rsp_cons += avail;
608 * xen_intr_signal() implies mb(). The producer will see
609 * the updated consumer index when the event is delivered.
611 xen_intr_signal(xs.xen_intr_handle);
617 /*----------------------- Received Message Processing ------------------------*/
619 * Block reading the next message from the XenStore service and
620 * process the result.
622 * \param type The returned type of the XenStore message received.
624 * \return 0 on success. Otherwise an errno value indicating the
625 * type of failure encountered.
628 xs_process_msg(enum xsd_sockmsg_type *type)
630 struct xs_stored_msg *msg;
634 msg = malloc(sizeof(*msg), M_XENSTORE, M_WAITOK);
635 error = xs_read_store(&msg->hdr, sizeof(msg->hdr));
637 free(msg, M_XENSTORE);
641 body = malloc(msg->hdr.len + 1, M_XENSTORE, M_WAITOK);
642 error = xs_read_store(body, msg->hdr.len);
644 free(body, M_XENSTORE);
645 free(msg, M_XENSTORE);
648 body[msg->hdr.len] = '\0';
650 *type = msg->hdr.type;
651 if (msg->hdr.type == XS_WATCH_EVENT) {
652 msg->u.watch.vec = split(body, msg->hdr.len,
653 &msg->u.watch.vec_size);
655 mtx_lock(&xs.registered_watches_lock);
656 msg->u.watch.handle = find_watch(
657 msg->u.watch.vec[XS_WATCH_TOKEN]);
658 mtx_lock(&xs.watch_events_lock);
659 if (msg->u.watch.handle != NULL &&
660 (!msg->u.watch.handle->max_pending ||
661 msg->u.watch.handle->pending <
662 msg->u.watch.handle->max_pending)) {
663 msg->u.watch.handle->pending++;
664 TAILQ_INSERT_TAIL(&xs.watch_events, msg, list);
665 wakeup(&xs.watch_events);
666 mtx_unlock(&xs.watch_events_lock);
668 mtx_unlock(&xs.watch_events_lock);
669 free(msg->u.watch.vec, M_XENSTORE);
670 free(msg, M_XENSTORE);
672 mtx_unlock(&xs.registered_watches_lock);
674 msg->u.reply.body = body;
675 mtx_lock(&xs.reply_lock);
676 TAILQ_INSERT_TAIL(&xs.reply_list, msg, list);
677 wakeup(&xs.reply_list);
678 mtx_unlock(&xs.reply_lock);
685 * Thread body of the XenStore receive thread.
687 * This thread blocks waiting for data from the XenStore service
688 * and processes and received messages.
691 xs_rcv_thread(void *arg __unused)
694 enum xsd_sockmsg_type type;
697 error = xs_process_msg(&type);
699 printf("XENSTORE error %d while reading message\n",
704 /*---------------- XenStore Message Request/Reply Processing -----------------*/
705 #define xsd_error_count (sizeof(xsd_errors) / sizeof(xsd_errors[0]))
708 * Convert a XenStore error string into an errno number.
710 * \param errorstring The error string to convert.
712 * \return The errno best matching the input string.
714 * \note Unknown error strings are converted to EINVAL.
717 xs_get_error(const char *errorstring)
721 for (i = 0; i < xsd_error_count; i++) {
722 if (!strcmp(errorstring, xsd_errors[i].errstring))
723 return (xsd_errors[i].errnum);
725 log(LOG_WARNING, "XENSTORE xen store gave: unknown error %s",
731 * Block waiting for a reply to a message request.
733 * \param type The returned type of the reply.
734 * \param len The returned body length of the reply.
735 * \param result The returned body of the reply.
737 * \return 0 on success. Otherwise an errno indicating the
741 xs_read_reply(enum xsd_sockmsg_type *type, u_int *len, void **result)
743 struct xs_stored_msg *msg;
747 mtx_lock(&xs.reply_lock);
748 while (TAILQ_EMPTY(&xs.reply_list)) {
749 error = mtx_sleep(&xs.reply_list, &xs.reply_lock, 0, "xswait",
751 if (error && error != EWOULDBLOCK) {
752 mtx_unlock(&xs.reply_lock);
756 msg = TAILQ_FIRST(&xs.reply_list);
757 TAILQ_REMOVE(&xs.reply_list, msg, list);
758 mtx_unlock(&xs.reply_lock);
760 *type = msg->hdr.type;
763 body = msg->u.reply.body;
765 free(msg, M_XENSTORE);
771 * Pass-thru interface for XenStore access by userland processes
772 * via the XenStore device.
774 * Reply type and length data are returned by overwriting these
775 * fields in the passed in request message.
777 * \param msg A properly formatted message to transmit to
778 * the XenStore service.
779 * \param result The returned body of the reply.
781 * \return 0 on success. Otherwise an errno indicating the cause
784 * \note The returned result is provided in malloced storage and thus
785 * must be free'd by the caller with 'free(result, M_XENSTORE);
788 xs_dev_request_and_reply(struct xsd_sockmsg *msg, void **result)
790 uint32_t request_type;
793 request_type = msg->type;
795 sx_xlock(&xs.request_mutex);
796 if ((error = xs_write_store(msg, sizeof(*msg) + msg->len)) == 0)
797 error = xs_read_reply(&msg->type, &msg->len, result);
798 sx_xunlock(&xs.request_mutex);
804 * Send a message with an optionally muti-part body to the XenStore service.
806 * \param t The transaction to use for this request.
807 * \param request_type The type of message to send.
808 * \param iovec Pointers to the body sections of the request.
809 * \param num_vecs The number of body sections in the request.
810 * \param len The returned length of the reply.
811 * \param result The returned body of the reply.
813 * \return 0 on success. Otherwise an errno indicating
814 * the cause of failure.
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_talkv(struct xs_transaction t, enum xsd_sockmsg_type request_type,
821 const struct iovec *iovec, u_int num_vecs, u_int *len, void **result)
823 struct xsd_sockmsg msg;
830 msg.type = request_type;
832 for (i = 0; i < num_vecs; i++)
833 msg.len += iovec[i].iov_len;
835 sx_xlock(&xs.request_mutex);
836 error = xs_write_store(&msg, sizeof(msg));
838 printf("xs_talkv failed %d\n", error);
839 goto error_lock_held;
842 for (i = 0; i < num_vecs; i++) {
843 error = xs_write_store(iovec[i].iov_base, iovec[i].iov_len);
845 printf("xs_talkv failed %d\n", error);
846 goto error_lock_held;
850 error = xs_read_reply(&msg.type, len, &ret);
853 sx_xunlock(&xs.request_mutex);
857 if (msg.type == XS_ERROR) {
858 error = xs_get_error(ret);
859 free(ret, M_XENSTORE);
863 /* Reply is either error or an echo of our request message type. */
864 KASSERT(msg.type == request_type, ("bad xenstore message type"));
869 free(ret, M_XENSTORE);
875 * Wrapper for xs_talkv allowing easy transmission of a message with
876 * a single, contiguous, message body.
878 * \param t The transaction to use for this request.
879 * \param request_type The type of message to send.
880 * \param body The body of the request.
881 * \param len The returned length of the reply.
882 * \param result The returned body of the reply.
884 * \return 0 on success. Otherwise an errno indicating
885 * the cause of failure.
887 * \note The returned result is provided in malloced storage and thus
888 * must be free'd by the caller with 'free(*result, M_XENSTORE);
891 xs_single(struct xs_transaction t, enum xsd_sockmsg_type request_type,
892 const char *body, u_int *len, void **result)
896 iovec.iov_base = (void *)(uintptr_t)body;
897 iovec.iov_len = strlen(body) + 1;
899 return (xs_talkv(t, request_type, &iovec, 1, len, result));
902 /*------------------------- XenStore Watch Support ---------------------------*/
904 * Transmit a watch request to the XenStore service.
906 * \param path The path in the XenStore to watch.
907 * \param tocken A unique identifier for this watch.
909 * \return 0 on success. Otherwise an errno indicating the
913 xs_watch(const char *path, const char *token)
917 iov[0].iov_base = (void *)(uintptr_t) path;
918 iov[0].iov_len = strlen(path) + 1;
919 iov[1].iov_base = (void *)(uintptr_t) token;
920 iov[1].iov_len = strlen(token) + 1;
922 return (xs_talkv(XST_NIL, XS_WATCH, iov, 2, NULL, NULL));
926 * Transmit an uwatch request to the XenStore service.
928 * \param path The path in the XenStore to watch.
929 * \param tocken A unique identifier for this watch.
931 * \return 0 on success. Otherwise an errno indicating the
935 xs_unwatch(const char *path, const char *token)
939 iov[0].iov_base = (void *)(uintptr_t) path;
940 iov[0].iov_len = strlen(path) + 1;
941 iov[1].iov_base = (void *)(uintptr_t) token;
942 iov[1].iov_len = strlen(token) + 1;
944 return (xs_talkv(XST_NIL, XS_UNWATCH, iov, 2, NULL, NULL));
948 * Convert from watch token (unique identifier) to the associated
949 * internal tracking structure for this watch.
951 * \param tocken The unique identifier for the watch to find.
953 * \return A pointer to the found watch structure or NULL.
955 static struct xs_watch *
956 find_watch(const char *token)
958 struct xs_watch *i, *cmp;
960 cmp = (void *)strtoul(token, NULL, 16);
962 LIST_FOREACH(i, &xs.registered_watches, list)
970 * Thread body of the XenStore watch event dispatch thread.
973 xenwatch_thread(void *unused)
975 struct xs_stored_msg *msg;
978 mtx_lock(&xs.watch_events_lock);
979 while (TAILQ_EMPTY(&xs.watch_events))
980 mtx_sleep(&xs.watch_events,
981 &xs.watch_events_lock,
982 PWAIT | PCATCH, "waitev", hz/10);
984 mtx_unlock(&xs.watch_events_lock);
985 sx_xlock(&xs.xenwatch_mutex);
987 mtx_lock(&xs.watch_events_lock);
988 msg = TAILQ_FIRST(&xs.watch_events);
990 TAILQ_REMOVE(&xs.watch_events, msg, list);
991 msg->u.watch.handle->pending--;
993 mtx_unlock(&xs.watch_events_lock);
997 * XXX There are messages coming in with a NULL
998 * XXX callback. This deserves further investigation;
999 * XXX the workaround here simply prevents the kernel
1000 * XXX from panic'ing on startup.
1002 if (msg->u.watch.handle->callback != NULL)
1003 msg->u.watch.handle->callback(
1004 msg->u.watch.handle,
1005 (const char **)msg->u.watch.vec,
1006 msg->u.watch.vec_size);
1007 free(msg->u.watch.vec, M_XENSTORE);
1008 free(msg, M_XENSTORE);
1011 sx_xunlock(&xs.xenwatch_mutex);
1015 /*----------- XenStore Configuration, Initialization, and Control ------------*/
1017 * Setup communication channels with the XenStore service.
1019 * \return On success, 0. Otherwise an errno value indicating the
1027 if (xen_store->rsp_prod != xen_store->rsp_cons) {
1028 log(LOG_WARNING, "XENSTORE response ring is not quiescent "
1029 "(%08x:%08x): fixing up\n",
1030 xen_store->rsp_cons, xen_store->rsp_prod);
1031 xen_store->rsp_cons = xen_store->rsp_prod;
1034 xen_intr_unbind(&xs.xen_intr_handle);
1036 error = xen_intr_bind_local_port(xs.xs_dev, xs.evtchn,
1037 /*filter*/NULL, xs_intr, /*arg*/NULL, INTR_TYPE_NET|INTR_MPSAFE,
1038 &xs.xen_intr_handle);
1040 log(LOG_WARNING, "XENSTORE request irq failed %i\n", error);
1047 /*------------------ Private Device Attachment Functions --------------------*/
1049 xs_identify(driver_t *driver, device_t parent)
1052 BUS_ADD_CHILD(parent, 0, "xenstore", 0);
1056 * Probe for the existence of the XenStore.
1061 xs_probe(device_t dev)
1064 * We are either operating within a PV kernel or being probed
1065 * as the child of the successfully attached xenpci device.
1066 * Thus we are in a Xen environment and there will be a XenStore.
1067 * Unconditionally return success.
1069 device_set_desc(dev, "XenStore");
1070 return (BUS_PROBE_NOWILDCARD);
1074 xs_attach_deferred(void *arg)
1077 bus_generic_probe(xs.xs_dev);
1078 bus_generic_attach(xs.xs_dev);
1080 config_intrhook_disestablish(&xs.xs_attachcb);
1084 xs_attach_late(void *arg, int pending)
1087 KASSERT((pending == 1), ("xs late attach queued several times"));
1088 bus_generic_probe(xs.xs_dev);
1089 bus_generic_attach(xs.xs_dev);
1093 * Attach to the XenStore.
1095 * This routine also prepares for the probe/attach of drivers that rely
1099 xs_attach(device_t dev)
1103 /* Allow us to get device_t from softc and vice-versa. */
1105 device_set_softc(dev, &xs);
1107 /* Initialize the interface to xenstore. */
1110 xs.initialized = false;
1111 xs.evtchn = xen_get_xenstore_evtchn();
1112 if (xs.evtchn == 0) {
1113 struct evtchn_alloc_unbound alloc_unbound;
1115 /* Allocate a local event channel for xenstore */
1116 alloc_unbound.dom = DOMID_SELF;
1117 alloc_unbound.remote_dom = DOMID_SELF;
1118 error = HYPERVISOR_event_channel_op(
1119 EVTCHNOP_alloc_unbound, &alloc_unbound);
1122 "unable to alloc event channel for Dom0: %d",
1125 xs.evtchn = alloc_unbound.port;
1127 /* Allocate memory for the xs shared ring */
1128 xen_store = malloc(PAGE_SIZE, M_XENSTORE, M_WAITOK | M_ZERO);
1129 xs.gpfn = atop(pmap_kextract((vm_offset_t)xen_store));
1131 xs.gpfn = xen_get_xenstore_mfn();
1132 xen_store = pmap_mapdev_attr(ptoa(xs.gpfn), PAGE_SIZE,
1134 xs.initialized = true;
1137 TAILQ_INIT(&xs.reply_list);
1138 TAILQ_INIT(&xs.watch_events);
1140 mtx_init(&xs.ring_lock, "ring lock", NULL, MTX_DEF);
1141 mtx_init(&xs.reply_lock, "reply lock", NULL, MTX_DEF);
1142 sx_init(&xs.xenwatch_mutex, "xenwatch");
1143 sx_init(&xs.request_mutex, "xenstore request");
1144 mtx_init(&xs.registered_watches_lock, "watches", NULL, MTX_DEF);
1145 mtx_init(&xs.watch_events_lock, "watch events", NULL, MTX_DEF);
1147 /* Initialize the shared memory rings to talk to xenstored */
1148 error = xs_init_comms();
1152 error = kproc_create(xenwatch_thread, NULL, &p, RFHIGHPID,
1156 xs.xenwatch_pid = p->p_pid;
1158 error = kproc_create(xs_rcv_thread, NULL, NULL,
1159 RFHIGHPID, 0, "xenstore_rcv");
1161 xs.xs_attachcb.ich_func = xs_attach_deferred;
1162 xs.xs_attachcb.ich_arg = NULL;
1163 if (xs.initialized) {
1164 config_intrhook_establish(&xs.xs_attachcb);
1166 TASK_INIT(&xs.xs_late_init, 0, xs_attach_late, NULL);
1173 * Prepare for suspension of this VM by halting XenStore access after
1174 * all transactions and individual requests have completed.
1177 xs_suspend(device_t dev)
1181 /* Suspend child Xen devices. */
1182 error = bus_generic_suspend(dev);
1186 sx_xlock(&xs.request_mutex);
1192 * Resume XenStore operations after this VM is resumed.
1195 xs_resume(device_t dev __unused)
1197 struct xs_watch *watch;
1198 char token[sizeof(watch) * 2 + 1];
1202 sx_xunlock(&xs.request_mutex);
1205 * NB: since xenstore childs have not been resumed yet, there's
1206 * no need to hold any watch mutex. Having clients try to add or
1207 * remove watches at this point (before xenstore is resumed) is
1208 * clearly a violantion of the resume order.
1210 LIST_FOREACH(watch, &xs.registered_watches, list) {
1211 sprintf(token, "%lX", (long)watch);
1212 xs_watch(watch->node, token);
1215 /* Resume child Xen devices. */
1216 bus_generic_resume(dev);
1221 /*-------------------- Private Device Attachment Data -----------------------*/
1222 static device_method_t xenstore_methods[] = {
1223 /* Device interface */
1224 DEVMETHOD(device_identify, xs_identify),
1225 DEVMETHOD(device_probe, xs_probe),
1226 DEVMETHOD(device_attach, xs_attach),
1227 DEVMETHOD(device_detach, bus_generic_detach),
1228 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1229 DEVMETHOD(device_suspend, xs_suspend),
1230 DEVMETHOD(device_resume, xs_resume),
1233 DEVMETHOD(bus_add_child, bus_generic_add_child),
1234 DEVMETHOD(bus_alloc_resource, bus_generic_alloc_resource),
1235 DEVMETHOD(bus_release_resource, bus_generic_release_resource),
1236 DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
1237 DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
1242 DEFINE_CLASS_0(xenstore, xenstore_driver, xenstore_methods, 0);
1243 static devclass_t xenstore_devclass;
1245 DRIVER_MODULE(xenstore, xenpv, xenstore_driver, xenstore_devclass, 0, 0);
1247 /*------------------------------- Sysctl Data --------------------------------*/
1248 /* XXX Shouldn't the node be somewhere else? */
1249 SYSCTL_NODE(_dev, OID_AUTO, xen, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
1251 SYSCTL_INT(_dev_xen, OID_AUTO, xsd_port, CTLFLAG_RD, &xs.evtchn, 0, "");
1252 SYSCTL_ULONG(_dev_xen, OID_AUTO, xsd_kva, CTLFLAG_RD, (u_long *) &xen_store, 0, "");
1254 /*-------------------------------- Public API --------------------------------*/
1255 /*------- API comments for these methods can be found in xenstorevar.h -------*/
1257 xs_initialized(void)
1260 return (xs.initialized);
1274 return (ptoa(xs.gpfn));
1278 xs_directory(struct xs_transaction t, const char *dir, const char *node,
1279 u_int *num, const char ***result)
1286 path = xs_join(dir, node);
1287 error = xs_single(t, XS_DIRECTORY, sbuf_data(path), &len,
1293 *result = split(strings, len, num);
1299 xs_exists(struct xs_transaction t, const char *dir, const char *node)
1304 error = xs_directory(t, dir, node, &dir_n, &d);
1307 free(d, M_XENSTORE);
1312 xs_read(struct xs_transaction t, const char *dir, const char *node,
1313 u_int *len, void **result)
1319 path = xs_join(dir, node);
1320 error = xs_single(t, XS_READ, sbuf_data(path), len, &ret);
1329 xs_write(struct xs_transaction t, const char *dir, const char *node,
1333 struct iovec iovec[2];
1336 path = xs_join(dir, node);
1338 iovec[0].iov_base = (void *)(uintptr_t) sbuf_data(path);
1339 iovec[0].iov_len = sbuf_len(path) + 1;
1340 iovec[1].iov_base = (void *)(uintptr_t) string;
1341 iovec[1].iov_len = strlen(string);
1343 error = xs_talkv(t, XS_WRITE, iovec, 2, NULL, NULL);
1350 xs_mkdir(struct xs_transaction t, const char *dir, const char *node)
1355 path = xs_join(dir, node);
1356 ret = xs_single(t, XS_MKDIR, sbuf_data(path), NULL, NULL);
1363 xs_rm(struct xs_transaction t, const char *dir, const char *node)
1368 path = xs_join(dir, node);
1369 ret = xs_single(t, XS_RM, sbuf_data(path), NULL, NULL);
1376 xs_rm_tree(struct xs_transaction xbt, const char *base, const char *node)
1378 struct xs_transaction local_xbt;
1379 struct sbuf *root_path_sbuf;
1380 struct sbuf *cur_path_sbuf;
1387 root_path_sbuf = xs_join(base, node);
1388 cur_path_sbuf = xs_join(base, node);
1389 root_path = sbuf_data(root_path_sbuf);
1390 cur_path = sbuf_data(cur_path_sbuf);
1395 error = xs_transaction_start(&local_xbt);
1405 error = xs_directory(xbt, cur_path, "", &count, &dir);
1409 for (i = 0; i < count; i++) {
1410 error = xs_rm(xbt, cur_path, dir[i]);
1411 if (error == ENOTEMPTY) {
1412 struct sbuf *push_dir;
1415 * Descend to clear out this sub directory.
1416 * We'll return to cur_dir once push_dir
1419 push_dir = xs_join(cur_path, dir[i]);
1420 sbuf_delete(cur_path_sbuf);
1421 cur_path_sbuf = push_dir;
1422 cur_path = sbuf_data(cur_path_sbuf);
1424 } else if (error != 0) {
1429 free(dir, M_XENSTORE);
1435 /* Directory is empty. It is now safe to remove. */
1436 error = xs_rm(xbt, cur_path, "");
1440 if (!strcmp(cur_path, root_path))
1443 /* Return to processing the parent directory. */
1444 last_slash = strrchr(cur_path, '/');
1445 KASSERT(last_slash != NULL,
1446 ("xs_rm_tree: mangled path %s", cur_path));
1452 sbuf_delete(cur_path_sbuf);
1453 sbuf_delete(root_path_sbuf);
1455 free(dir, M_XENSTORE);
1457 if (local_xbt.id != 0) {
1460 terror = xs_transaction_end(local_xbt, /*abort*/error != 0);
1462 if (terror == EAGAIN && error == 0)
1469 xs_transaction_start(struct xs_transaction *t)
1474 error = xs_single(XST_NIL, XS_TRANSACTION_START, "", NULL,
1477 t->id = strtoul(id_str, NULL, 0);
1478 free(id_str, M_XENSTORE);
1484 xs_transaction_end(struct xs_transaction t, int abort)
1489 strcpy(abortstr, "F");
1491 strcpy(abortstr, "T");
1493 return (xs_single(t, XS_TRANSACTION_END, abortstr, NULL, NULL));
1497 xs_scanf(struct xs_transaction t, const char *dir, const char *node,
1498 int *scancountp, const char *fmt, ...)
1504 error = xs_read(t, dir, node, NULL, (void **) &val);
1509 ns = vsscanf(val, fmt, ap);
1511 free(val, M_XENSTORE);
1512 /* Distinctive errno. */
1521 xs_vprintf(struct xs_transaction t,
1522 const char *dir, const char *node, const char *fmt, va_list ap)
1527 sb = sbuf_new_auto();
1528 sbuf_vprintf(sb, fmt, ap);
1530 error = xs_write(t, dir, node, sbuf_data(sb));
1537 xs_printf(struct xs_transaction t, const char *dir, const char *node,
1538 const char *fmt, ...)
1544 error = xs_vprintf(t, dir, node, fmt, ap);
1551 xs_gather(struct xs_transaction t, const char *dir, ...)
1559 while (error == 0 && (name = va_arg(ap, char *)) != NULL) {
1560 const char *fmt = va_arg(ap, char *);
1561 void *result = va_arg(ap, void *);
1564 error = xs_read(t, dir, name, NULL, (void **) &p);
1569 if (sscanf(p, fmt, result) == 0)
1571 free(p, M_XENSTORE);
1573 *(char **)result = p;
1581 xs_register_watch(struct xs_watch *watch)
1583 /* Pointer in ascii is the token. */
1584 char token[sizeof(watch) * 2 + 1];
1588 sprintf(token, "%lX", (long)watch);
1590 mtx_lock(&xs.registered_watches_lock);
1591 KASSERT(find_watch(token) == NULL, ("watch already registered"));
1592 LIST_INSERT_HEAD(&xs.registered_watches, watch, list);
1593 mtx_unlock(&xs.registered_watches_lock);
1595 error = xs_watch(watch->node, token);
1597 /* Ignore errors due to multiple registration. */
1598 if (error == EEXIST)
1602 mtx_lock(&xs.registered_watches_lock);
1603 LIST_REMOVE(watch, list);
1604 mtx_unlock(&xs.registered_watches_lock);
1611 xs_unregister_watch(struct xs_watch *watch)
1613 struct xs_stored_msg *msg, *tmp;
1614 char token[sizeof(watch) * 2 + 1];
1617 sprintf(token, "%lX", (long)watch);
1619 mtx_lock(&xs.registered_watches_lock);
1620 if (find_watch(token) == NULL) {
1621 mtx_unlock(&xs.registered_watches_lock);
1624 LIST_REMOVE(watch, list);
1625 mtx_unlock(&xs.registered_watches_lock);
1627 error = xs_unwatch(watch->node, token);
1629 log(LOG_WARNING, "XENSTORE Failed to release watch %s: %i\n",
1630 watch->node, error);
1632 /* Cancel pending watch events. */
1633 mtx_lock(&xs.watch_events_lock);
1634 TAILQ_FOREACH_SAFE(msg, &xs.watch_events, list, tmp) {
1635 if (msg->u.watch.handle != watch)
1637 TAILQ_REMOVE(&xs.watch_events, msg, list);
1638 free(msg->u.watch.vec, M_XENSTORE);
1639 free(msg, M_XENSTORE);
1641 mtx_unlock(&xs.watch_events_lock);
1643 /* Flush any currently-executing callback, unless we are it. :-) */
1644 if (curproc->p_pid != xs.xenwatch_pid) {
1645 sx_xlock(&xs.xenwatch_mutex);
1646 sx_xunlock(&xs.xenwatch_mutex);
1654 sx_xlock(&xs.request_mutex);
1662 sx_xunlock(&xs.request_mutex);