xref: /freebsd/sys/dev/xen/blkfront/blkfront.c (revision 3fc36ee018bb836bd1796067cf4ef8683f166ebc)
1 /*
2  * XenBSD block device driver
3  *
4  * Copyright (c) 2010-2013 Spectra Logic Corporation
5  * Copyright (c) 2009 Scott Long, Yahoo!
6  * Copyright (c) 2009 Frank Suchomel, Citrix
7  * Copyright (c) 2009 Doug F. Rabson, Citrix
8  * Copyright (c) 2005 Kip Macy
9  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
10  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
11  *
12  *
13  * Permission is hereby granted, free of charge, to any person obtaining a copy
14  * of this software and associated documentation files (the "Software"), to
15  * deal in the Software without restriction, including without limitation the
16  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
17  * sell copies of the Software, and to permit persons to whom the Software is
18  * furnished to do so, subject to the following conditions:
19  *
20  * The above copyright notice and this permission notice shall be included in
21  * 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
28  * DEALINGS IN THE SOFTWARE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/malloc.h>
37 #include <sys/kernel.h>
38 #include <vm/vm.h>
39 #include <vm/pmap.h>
40 
41 #include <sys/bio.h>
42 #include <sys/bus.h>
43 #include <sys/conf.h>
44 #include <sys/module.h>
45 #include <sys/sysctl.h>
46 
47 #include <machine/bus.h>
48 #include <sys/rman.h>
49 #include <machine/resource.h>
50 #include <machine/intr_machdep.h>
51 #include <machine/vmparam.h>
52 #include <sys/bus_dma.h>
53 
54 #include <xen/xen-os.h>
55 #include <xen/hypervisor.h>
56 #include <xen/xen_intr.h>
57 #include <xen/gnttab.h>
58 #include <xen/interface/grant_table.h>
59 #include <xen/interface/io/protocols.h>
60 #include <xen/xenbus/xenbusvar.h>
61 
62 #include <machine/_inttypes.h>
63 
64 #include <geom/geom_disk.h>
65 
66 #include <dev/xen/blkfront/block.h>
67 
68 #include "xenbus_if.h"
69 
70 /*--------------------------- Forward Declarations ---------------------------*/
71 static void xbd_closing(device_t);
72 static void xbd_startio(struct xbd_softc *sc);
73 
74 /*---------------------------------- Macros ----------------------------------*/
75 #if 0
76 #define DPRINTK(fmt, args...) printf("[XEN] %s:%d: " fmt ".\n", __func__, __LINE__, ##args)
77 #else
78 #define DPRINTK(fmt, args...)
79 #endif
80 
81 #define XBD_SECTOR_SHFT		9
82 
83 /*---------------------------- Global Static Data ----------------------------*/
84 static MALLOC_DEFINE(M_XENBLOCKFRONT, "xbd", "Xen Block Front driver data");
85 
86 static int xbd_enable_indirect = 1;
87 SYSCTL_NODE(_hw, OID_AUTO, xbd, CTLFLAG_RD, 0, "xbd driver parameters");
88 SYSCTL_INT(_hw_xbd, OID_AUTO, xbd_enable_indirect, CTLFLAG_RDTUN,
89     &xbd_enable_indirect, 0, "Enable xbd indirect segments");
90 
91 /*---------------------------- Command Processing ----------------------------*/
92 static void
93 xbd_freeze(struct xbd_softc *sc, xbd_flag_t xbd_flag)
94 {
95 	if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) != 0)
96 		return;
97 
98 	sc->xbd_flags |= xbd_flag;
99 	sc->xbd_qfrozen_cnt++;
100 }
101 
102 static void
103 xbd_thaw(struct xbd_softc *sc, xbd_flag_t xbd_flag)
104 {
105 	if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) == 0)
106 		return;
107 
108 	if (sc->xbd_qfrozen_cnt == 0)
109 		panic("%s: Thaw with flag 0x%x while not frozen.",
110 		    __func__, xbd_flag);
111 
112 	sc->xbd_flags &= ~xbd_flag;
113 	sc->xbd_qfrozen_cnt--;
114 }
115 
116 static void
117 xbd_cm_freeze(struct xbd_softc *sc, struct xbd_command *cm, xbdc_flag_t cm_flag)
118 {
119 	if ((cm->cm_flags & XBDCF_FROZEN) != 0)
120 		return;
121 
122 	cm->cm_flags |= XBDCF_FROZEN|cm_flag;
123 	xbd_freeze(sc, XBDF_NONE);
124 }
125 
126 static void
127 xbd_cm_thaw(struct xbd_softc *sc, struct xbd_command *cm)
128 {
129 	if ((cm->cm_flags & XBDCF_FROZEN) == 0)
130 		return;
131 
132 	cm->cm_flags &= ~XBDCF_FROZEN;
133 	xbd_thaw(sc, XBDF_NONE);
134 }
135 
136 static inline void
137 xbd_flush_requests(struct xbd_softc *sc)
138 {
139 	int notify;
140 
141 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->xbd_ring, notify);
142 
143 	if (notify)
144 		xen_intr_signal(sc->xen_intr_handle);
145 }
146 
147 static void
148 xbd_free_command(struct xbd_command *cm)
149 {
150 
151 	KASSERT((cm->cm_flags & XBDCF_Q_MASK) == XBD_Q_NONE,
152 	    ("Freeing command that is still on queue %d.",
153 	    cm->cm_flags & XBDCF_Q_MASK));
154 
155 	cm->cm_flags = XBDCF_INITIALIZER;
156 	cm->cm_bp = NULL;
157 	cm->cm_complete = NULL;
158 	xbd_enqueue_cm(cm, XBD_Q_FREE);
159 	xbd_thaw(cm->cm_sc, XBDF_CM_SHORTAGE);
160 }
161 
162 static void
163 xbd_mksegarray(bus_dma_segment_t *segs, int nsegs,
164     grant_ref_t * gref_head, int otherend_id, int readonly,
165     grant_ref_t * sg_ref, struct blkif_request_segment *sg)
166 {
167 	struct blkif_request_segment *last_block_sg = sg + nsegs;
168 	vm_paddr_t buffer_ma;
169 	uint64_t fsect, lsect;
170 	int ref;
171 
172 	while (sg < last_block_sg) {
173 		KASSERT(segs->ds_addr % (1 << XBD_SECTOR_SHFT) == 0,
174 		    ("XEN disk driver I/O must be sector aligned"));
175 		KASSERT(segs->ds_len % (1 << XBD_SECTOR_SHFT) == 0,
176 		    ("XEN disk driver I/Os must be a multiple of "
177 		    "the sector length"));
178 		buffer_ma = segs->ds_addr;
179 		fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT;
180 		lsect = fsect + (segs->ds_len  >> XBD_SECTOR_SHFT) - 1;
181 
182 		KASSERT(lsect <= 7, ("XEN disk driver data cannot "
183 		    "cross a page boundary"));
184 
185 		/* install a grant reference. */
186 		ref = gnttab_claim_grant_reference(gref_head);
187 
188 		/*
189 		 * GNTTAB_LIST_END == 0xffffffff, but it is private
190 		 * to gnttab.c.
191 		 */
192 		KASSERT(ref != ~0, ("grant_reference failed"));
193 
194 		gnttab_grant_foreign_access_ref(
195 		    ref,
196 		    otherend_id,
197 		    buffer_ma >> PAGE_SHIFT,
198 		    readonly);
199 
200 		*sg_ref = ref;
201 		*sg = (struct blkif_request_segment) {
202 			.gref       = ref,
203 			.first_sect = fsect,
204 			.last_sect  = lsect
205 		};
206 		sg++;
207 		sg_ref++;
208 		segs++;
209 	}
210 }
211 
212 static void
213 xbd_queue_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
214 {
215 	struct xbd_softc *sc;
216 	struct xbd_command *cm;
217 	int op;
218 
219 	cm = arg;
220 	sc = cm->cm_sc;
221 
222 	if (error) {
223 		cm->cm_bp->bio_error = EIO;
224 		biodone(cm->cm_bp);
225 		xbd_free_command(cm);
226 		return;
227 	}
228 
229 	KASSERT(nsegs <= sc->xbd_max_request_segments,
230 	    ("Too many segments in a blkfront I/O"));
231 
232 	if (nsegs <= BLKIF_MAX_SEGMENTS_PER_REQUEST) {
233 		blkif_request_t	*ring_req;
234 
235 		/* Fill out a blkif_request_t structure. */
236 		ring_req = (blkif_request_t *)
237 		    RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt);
238 		sc->xbd_ring.req_prod_pvt++;
239 		ring_req->id = cm->cm_id;
240 		ring_req->operation = cm->cm_operation;
241 		ring_req->sector_number = cm->cm_sector_number;
242 		ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk;
243 		ring_req->nr_segments = nsegs;
244 		cm->cm_nseg = nsegs;
245 		xbd_mksegarray(segs, nsegs, &cm->cm_gref_head,
246 		    xenbus_get_otherend_id(sc->xbd_dev),
247 		    cm->cm_operation == BLKIF_OP_WRITE,
248 		    cm->cm_sg_refs, ring_req->seg);
249 	} else {
250 		blkif_request_indirect_t *ring_req;
251 
252 		/* Fill out a blkif_request_indirect_t structure. */
253 		ring_req = (blkif_request_indirect_t *)
254 		    RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt);
255 		sc->xbd_ring.req_prod_pvt++;
256 		ring_req->id = cm->cm_id;
257 		ring_req->operation = BLKIF_OP_INDIRECT;
258 		ring_req->indirect_op = cm->cm_operation;
259 		ring_req->sector_number = cm->cm_sector_number;
260 		ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk;
261 		ring_req->nr_segments = nsegs;
262 		cm->cm_nseg = nsegs;
263 		xbd_mksegarray(segs, nsegs, &cm->cm_gref_head,
264 		    xenbus_get_otherend_id(sc->xbd_dev),
265 		    cm->cm_operation == BLKIF_OP_WRITE,
266 		    cm->cm_sg_refs, cm->cm_indirectionpages);
267 		memcpy(ring_req->indirect_grefs, &cm->cm_indirectionrefs,
268 		    sizeof(grant_ref_t) * sc->xbd_max_request_indirectpages);
269 	}
270 
271 	if (cm->cm_operation == BLKIF_OP_READ)
272 		op = BUS_DMASYNC_PREREAD;
273 	else if (cm->cm_operation == BLKIF_OP_WRITE)
274 		op = BUS_DMASYNC_PREWRITE;
275 	else
276 		op = 0;
277 	bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op);
278 
279 	gnttab_free_grant_references(cm->cm_gref_head);
280 
281 	xbd_enqueue_cm(cm, XBD_Q_BUSY);
282 
283 	/*
284 	 * If bus dma had to asynchronously call us back to dispatch
285 	 * this command, we are no longer executing in the context of
286 	 * xbd_startio().  Thus we cannot rely on xbd_startio()'s call to
287 	 * xbd_flush_requests() to publish this command to the backend
288 	 * along with any other commands that it could batch.
289 	 */
290 	if ((cm->cm_flags & XBDCF_ASYNC_MAPPING) != 0)
291 		xbd_flush_requests(sc);
292 
293 	return;
294 }
295 
296 static int
297 xbd_queue_request(struct xbd_softc *sc, struct xbd_command *cm)
298 {
299 	int error;
300 
301 	if (cm->cm_bp != NULL)
302 		error = bus_dmamap_load_bio(sc->xbd_io_dmat, cm->cm_map,
303 		    cm->cm_bp, xbd_queue_cb, cm, 0);
304 	else
305 		error = bus_dmamap_load(sc->xbd_io_dmat, cm->cm_map,
306 		    cm->cm_data, cm->cm_datalen, xbd_queue_cb, cm, 0);
307 	if (error == EINPROGRESS) {
308 		/*
309 		 * Maintain queuing order by freezing the queue.  The next
310 		 * command may not require as many resources as the command
311 		 * we just attempted to map, so we can't rely on bus dma
312 		 * blocking for it too.
313 		 */
314 		xbd_cm_freeze(sc, cm, XBDCF_ASYNC_MAPPING);
315 		return (0);
316 	}
317 
318 	return (error);
319 }
320 
321 static void
322 xbd_restart_queue_callback(void *arg)
323 {
324 	struct xbd_softc *sc = arg;
325 
326 	mtx_lock(&sc->xbd_io_lock);
327 
328 	xbd_thaw(sc, XBDF_GNT_SHORTAGE);
329 
330 	xbd_startio(sc);
331 
332 	mtx_unlock(&sc->xbd_io_lock);
333 }
334 
335 static struct xbd_command *
336 xbd_bio_command(struct xbd_softc *sc)
337 {
338 	struct xbd_command *cm;
339 	struct bio *bp;
340 
341 	if (__predict_false(sc->xbd_state != XBD_STATE_CONNECTED))
342 		return (NULL);
343 
344 	bp = xbd_dequeue_bio(sc);
345 	if (bp == NULL)
346 		return (NULL);
347 
348 	if ((cm = xbd_dequeue_cm(sc, XBD_Q_FREE)) == NULL) {
349 		xbd_freeze(sc, XBDF_CM_SHORTAGE);
350 		xbd_requeue_bio(sc, bp);
351 		return (NULL);
352 	}
353 
354 	if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
355 	    &cm->cm_gref_head) != 0) {
356 		gnttab_request_free_callback(&sc->xbd_callback,
357 		    xbd_restart_queue_callback, sc,
358 		    sc->xbd_max_request_segments);
359 		xbd_freeze(sc, XBDF_GNT_SHORTAGE);
360 		xbd_requeue_bio(sc, bp);
361 		xbd_enqueue_cm(cm, XBD_Q_FREE);
362 		return (NULL);
363 	}
364 
365 	cm->cm_bp = bp;
366 	cm->cm_sector_number = (blkif_sector_t)bp->bio_pblkno;
367 
368 	switch (bp->bio_cmd) {
369 	case BIO_READ:
370 		cm->cm_operation = BLKIF_OP_READ;
371 		break;
372 	case BIO_WRITE:
373 		cm->cm_operation = BLKIF_OP_WRITE;
374 		if ((bp->bio_flags & BIO_ORDERED) != 0) {
375 			if ((sc->xbd_flags & XBDF_BARRIER) != 0) {
376 				cm->cm_operation = BLKIF_OP_WRITE_BARRIER;
377 			} else {
378 				/*
379 				 * Single step this command.
380 				 */
381 				cm->cm_flags |= XBDCF_Q_FREEZE;
382 				if (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
383 					/*
384 					 * Wait for in-flight requests to
385 					 * finish.
386 					 */
387 					xbd_freeze(sc, XBDF_WAIT_IDLE);
388 					xbd_requeue_cm(cm, XBD_Q_READY);
389 					return (NULL);
390 				}
391 			}
392 		}
393 		break;
394 	case BIO_FLUSH:
395 		if ((sc->xbd_flags & XBDF_FLUSH) != 0)
396 			cm->cm_operation = BLKIF_OP_FLUSH_DISKCACHE;
397 		else if ((sc->xbd_flags & XBDF_BARRIER) != 0)
398 			cm->cm_operation = BLKIF_OP_WRITE_BARRIER;
399 		else
400 			panic("flush request, but no flush support available");
401 		break;
402 	default:
403 		panic("unknown bio command %d", bp->bio_cmd);
404 	}
405 
406 	return (cm);
407 }
408 
409 /*
410  * Dequeue buffers and place them in the shared communication ring.
411  * Return when no more requests can be accepted or all buffers have
412  * been queued.
413  *
414  * Signal XEN once the ring has been filled out.
415  */
416 static void
417 xbd_startio(struct xbd_softc *sc)
418 {
419 	struct xbd_command *cm;
420 	int error, queued = 0;
421 
422 	mtx_assert(&sc->xbd_io_lock, MA_OWNED);
423 
424 	if (sc->xbd_state != XBD_STATE_CONNECTED)
425 		return;
426 
427 	while (!RING_FULL(&sc->xbd_ring)) {
428 
429 		if (sc->xbd_qfrozen_cnt != 0)
430 			break;
431 
432 		cm = xbd_dequeue_cm(sc, XBD_Q_READY);
433 
434 		if (cm == NULL)
435 		    cm = xbd_bio_command(sc);
436 
437 		if (cm == NULL)
438 			break;
439 
440 		if ((cm->cm_flags & XBDCF_Q_FREEZE) != 0) {
441 			/*
442 			 * Single step command.  Future work is
443 			 * held off until this command completes.
444 			 */
445 			xbd_cm_freeze(sc, cm, XBDCF_Q_FREEZE);
446 		}
447 
448 		if ((error = xbd_queue_request(sc, cm)) != 0) {
449 			printf("xbd_queue_request returned %d\n", error);
450 			break;
451 		}
452 		queued++;
453 	}
454 
455 	if (queued != 0)
456 		xbd_flush_requests(sc);
457 }
458 
459 static void
460 xbd_bio_complete(struct xbd_softc *sc, struct xbd_command *cm)
461 {
462 	struct bio *bp;
463 
464 	bp = cm->cm_bp;
465 
466 	if (__predict_false(cm->cm_status != BLKIF_RSP_OKAY)) {
467 		disk_err(bp, "disk error" , -1, 0);
468 		printf(" status: %x\n", cm->cm_status);
469 		bp->bio_flags |= BIO_ERROR;
470 	}
471 
472 	if (bp->bio_flags & BIO_ERROR)
473 		bp->bio_error = EIO;
474 	else
475 		bp->bio_resid = 0;
476 
477 	xbd_free_command(cm);
478 	biodone(bp);
479 }
480 
481 static void
482 xbd_int(void *xsc)
483 {
484 	struct xbd_softc *sc = xsc;
485 	struct xbd_command *cm;
486 	blkif_response_t *bret;
487 	RING_IDX i, rp;
488 	int op;
489 
490 	mtx_lock(&sc->xbd_io_lock);
491 
492 	if (__predict_false(sc->xbd_state == XBD_STATE_DISCONNECTED)) {
493 		mtx_unlock(&sc->xbd_io_lock);
494 		return;
495 	}
496 
497  again:
498 	rp = sc->xbd_ring.sring->rsp_prod;
499 	rmb(); /* Ensure we see queued responses up to 'rp'. */
500 
501 	for (i = sc->xbd_ring.rsp_cons; i != rp;) {
502 		bret = RING_GET_RESPONSE(&sc->xbd_ring, i);
503 		cm   = &sc->xbd_shadow[bret->id];
504 
505 		xbd_remove_cm(cm, XBD_Q_BUSY);
506 		gnttab_end_foreign_access_references(cm->cm_nseg,
507 		    cm->cm_sg_refs);
508 		i++;
509 
510 		if (cm->cm_operation == BLKIF_OP_READ)
511 			op = BUS_DMASYNC_POSTREAD;
512 		else if (cm->cm_operation == BLKIF_OP_WRITE ||
513 		    cm->cm_operation == BLKIF_OP_WRITE_BARRIER)
514 			op = BUS_DMASYNC_POSTWRITE;
515 		else
516 			op = 0;
517 		bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op);
518 		bus_dmamap_unload(sc->xbd_io_dmat, cm->cm_map);
519 
520 		/*
521 		 * Release any hold this command has on future command
522 		 * dispatch.
523 		 */
524 		xbd_cm_thaw(sc, cm);
525 
526 		/*
527 		 * Directly call the i/o complete routine to save an
528 		 * an indirection in the common case.
529 		 */
530 		cm->cm_status = bret->status;
531 		if (cm->cm_bp)
532 			xbd_bio_complete(sc, cm);
533 		else if (cm->cm_complete != NULL)
534 			cm->cm_complete(cm);
535 		else
536 			xbd_free_command(cm);
537 	}
538 
539 	sc->xbd_ring.rsp_cons = i;
540 
541 	if (i != sc->xbd_ring.req_prod_pvt) {
542 		int more_to_do;
543 		RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, more_to_do);
544 		if (more_to_do)
545 			goto again;
546 	} else {
547 		sc->xbd_ring.sring->rsp_event = i + 1;
548 	}
549 
550 	if (xbd_queue_length(sc, XBD_Q_BUSY) == 0)
551 		xbd_thaw(sc, XBDF_WAIT_IDLE);
552 
553 	xbd_startio(sc);
554 
555 	if (__predict_false(sc->xbd_state == XBD_STATE_SUSPENDED))
556 		wakeup(&sc->xbd_cm_q[XBD_Q_BUSY]);
557 
558 	mtx_unlock(&sc->xbd_io_lock);
559 }
560 
561 /*------------------------------- Dump Support -------------------------------*/
562 /**
563  * Quiesce the disk writes for a dump file before allowing the next buffer.
564  */
565 static void
566 xbd_quiesce(struct xbd_softc *sc)
567 {
568 	int mtd;
569 
570 	// While there are outstanding requests
571 	while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
572 		RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, mtd);
573 		if (mtd) {
574 			/* Received request completions, update queue. */
575 			xbd_int(sc);
576 		}
577 		if (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
578 			/*
579 			 * Still pending requests, wait for the disk i/o
580 			 * to complete.
581 			 */
582 			HYPERVISOR_yield();
583 		}
584 	}
585 }
586 
587 /* Kernel dump function for a paravirtualized disk device */
588 static void
589 xbd_dump_complete(struct xbd_command *cm)
590 {
591 
592 	xbd_enqueue_cm(cm, XBD_Q_COMPLETE);
593 }
594 
595 static int
596 xbd_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
597     size_t length)
598 {
599 	struct disk *dp = arg;
600 	struct xbd_softc *sc = dp->d_drv1;
601 	struct xbd_command *cm;
602 	size_t chunk;
603 	int sbp;
604 	int rc = 0;
605 
606 	if (length <= 0)
607 		return (rc);
608 
609 	xbd_quiesce(sc);	/* All quiet on the western front. */
610 
611 	/*
612 	 * If this lock is held, then this module is failing, and a
613 	 * successful kernel dump is highly unlikely anyway.
614 	 */
615 	mtx_lock(&sc->xbd_io_lock);
616 
617 	/* Split the 64KB block as needed */
618 	for (sbp=0; length > 0; sbp++) {
619 		cm = xbd_dequeue_cm(sc, XBD_Q_FREE);
620 		if (cm == NULL) {
621 			mtx_unlock(&sc->xbd_io_lock);
622 			device_printf(sc->xbd_dev, "dump: no more commands?\n");
623 			return (EBUSY);
624 		}
625 
626 		if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
627 		    &cm->cm_gref_head) != 0) {
628 			xbd_free_command(cm);
629 			mtx_unlock(&sc->xbd_io_lock);
630 			device_printf(sc->xbd_dev, "no more grant allocs?\n");
631 			return (EBUSY);
632 		}
633 
634 		chunk = length > sc->xbd_max_request_size ?
635 		    sc->xbd_max_request_size : length;
636 		cm->cm_data = virtual;
637 		cm->cm_datalen = chunk;
638 		cm->cm_operation = BLKIF_OP_WRITE;
639 		cm->cm_sector_number = offset / dp->d_sectorsize;
640 		cm->cm_complete = xbd_dump_complete;
641 
642 		xbd_enqueue_cm(cm, XBD_Q_READY);
643 
644 		length -= chunk;
645 		offset += chunk;
646 		virtual = (char *) virtual + chunk;
647 	}
648 
649 	/* Tell DOM0 to do the I/O */
650 	xbd_startio(sc);
651 	mtx_unlock(&sc->xbd_io_lock);
652 
653 	/* Poll for the completion. */
654 	xbd_quiesce(sc);	/* All quite on the eastern front */
655 
656 	/* If there were any errors, bail out... */
657 	while ((cm = xbd_dequeue_cm(sc, XBD_Q_COMPLETE)) != NULL) {
658 		if (cm->cm_status != BLKIF_RSP_OKAY) {
659 			device_printf(sc->xbd_dev,
660 			    "Dump I/O failed at sector %jd\n",
661 			    cm->cm_sector_number);
662 			rc = EIO;
663 		}
664 		xbd_free_command(cm);
665 	}
666 
667 	return (rc);
668 }
669 
670 /*----------------------------- Disk Entrypoints -----------------------------*/
671 static int
672 xbd_open(struct disk *dp)
673 {
674 	struct xbd_softc *sc = dp->d_drv1;
675 
676 	if (sc == NULL) {
677 		printf("xbd%d: not found", dp->d_unit);
678 		return (ENXIO);
679 	}
680 
681 	sc->xbd_flags |= XBDF_OPEN;
682 	sc->xbd_users++;
683 	return (0);
684 }
685 
686 static int
687 xbd_close(struct disk *dp)
688 {
689 	struct xbd_softc *sc = dp->d_drv1;
690 
691 	if (sc == NULL)
692 		return (ENXIO);
693 	sc->xbd_flags &= ~XBDF_OPEN;
694 	if (--(sc->xbd_users) == 0) {
695 		/*
696 		 * Check whether we have been instructed to close.  We will
697 		 * have ignored this request initially, as the device was
698 		 * still mounted.
699 		 */
700 		if (xenbus_get_otherend_state(sc->xbd_dev) ==
701 		    XenbusStateClosing)
702 			xbd_closing(sc->xbd_dev);
703 	}
704 	return (0);
705 }
706 
707 static int
708 xbd_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
709 {
710 	struct xbd_softc *sc = dp->d_drv1;
711 
712 	if (sc == NULL)
713 		return (ENXIO);
714 
715 	return (ENOTTY);
716 }
717 
718 /*
719  * Read/write routine for a buffer.  Finds the proper unit, place it on
720  * the sortq and kick the controller.
721  */
722 static void
723 xbd_strategy(struct bio *bp)
724 {
725 	struct xbd_softc *sc = bp->bio_disk->d_drv1;
726 
727 	/* bogus disk? */
728 	if (sc == NULL) {
729 		bp->bio_error = EINVAL;
730 		bp->bio_flags |= BIO_ERROR;
731 		bp->bio_resid = bp->bio_bcount;
732 		biodone(bp);
733 		return;
734 	}
735 
736 	/*
737 	 * Place it in the queue of disk activities for this disk
738 	 */
739 	mtx_lock(&sc->xbd_io_lock);
740 
741 	xbd_enqueue_bio(sc, bp);
742 	xbd_startio(sc);
743 
744 	mtx_unlock(&sc->xbd_io_lock);
745 	return;
746 }
747 
748 /*------------------------------ Ring Management -----------------------------*/
749 static int
750 xbd_alloc_ring(struct xbd_softc *sc)
751 {
752 	blkif_sring_t *sring;
753 	uintptr_t sring_page_addr;
754 	int error;
755 	int i;
756 
757 	sring = malloc(sc->xbd_ring_pages * PAGE_SIZE, M_XENBLOCKFRONT,
758 	    M_NOWAIT|M_ZERO);
759 	if (sring == NULL) {
760 		xenbus_dev_fatal(sc->xbd_dev, ENOMEM, "allocating shared ring");
761 		return (ENOMEM);
762 	}
763 	SHARED_RING_INIT(sring);
764 	FRONT_RING_INIT(&sc->xbd_ring, sring, sc->xbd_ring_pages * PAGE_SIZE);
765 
766 	for (i = 0, sring_page_addr = (uintptr_t)sring;
767 	     i < sc->xbd_ring_pages;
768 	     i++, sring_page_addr += PAGE_SIZE) {
769 
770 		error = xenbus_grant_ring(sc->xbd_dev,
771 		    (vtophys(sring_page_addr) >> PAGE_SHIFT),
772 		    &sc->xbd_ring_ref[i]);
773 		if (error) {
774 			xenbus_dev_fatal(sc->xbd_dev, error,
775 			    "granting ring_ref(%d)", i);
776 			return (error);
777 		}
778 	}
779 	if (sc->xbd_ring_pages == 1) {
780 		error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev),
781 		    "ring-ref", "%u", sc->xbd_ring_ref[0]);
782 		if (error) {
783 			xenbus_dev_fatal(sc->xbd_dev, error,
784 			    "writing %s/ring-ref",
785 			    xenbus_get_node(sc->xbd_dev));
786 			return (error);
787 		}
788 	} else {
789 		for (i = 0; i < sc->xbd_ring_pages; i++) {
790 			char ring_ref_name[]= "ring_refXX";
791 
792 			snprintf(ring_ref_name, sizeof(ring_ref_name),
793 			    "ring-ref%u", i);
794 			error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev),
795 			     ring_ref_name, "%u", sc->xbd_ring_ref[i]);
796 			if (error) {
797 				xenbus_dev_fatal(sc->xbd_dev, error,
798 				    "writing %s/%s",
799 				    xenbus_get_node(sc->xbd_dev),
800 				    ring_ref_name);
801 				return (error);
802 			}
803 		}
804 	}
805 
806 	error = xen_intr_alloc_and_bind_local_port(sc->xbd_dev,
807 	    xenbus_get_otherend_id(sc->xbd_dev), NULL, xbd_int, sc,
808 	    INTR_TYPE_BIO | INTR_MPSAFE, &sc->xen_intr_handle);
809 	if (error) {
810 		xenbus_dev_fatal(sc->xbd_dev, error,
811 		    "xen_intr_alloc_and_bind_local_port failed");
812 		return (error);
813 	}
814 
815 	return (0);
816 }
817 
818 static void
819 xbd_free_ring(struct xbd_softc *sc)
820 {
821 	int i;
822 
823 	if (sc->xbd_ring.sring == NULL)
824 		return;
825 
826 	for (i = 0; i < sc->xbd_ring_pages; i++) {
827 		if (sc->xbd_ring_ref[i] != GRANT_REF_INVALID) {
828 			gnttab_end_foreign_access_ref(sc->xbd_ring_ref[i]);
829 			sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
830 		}
831 	}
832 	free(sc->xbd_ring.sring, M_XENBLOCKFRONT);
833 	sc->xbd_ring.sring = NULL;
834 }
835 
836 /*-------------------------- Initialization/Teardown -------------------------*/
837 static int
838 xbd_feature_string(struct xbd_softc *sc, char *features, size_t len)
839 {
840 	struct sbuf sb;
841 	int feature_cnt;
842 
843 	sbuf_new(&sb, features, len, SBUF_FIXEDLEN);
844 
845 	feature_cnt = 0;
846 	if ((sc->xbd_flags & XBDF_FLUSH) != 0) {
847 		sbuf_printf(&sb, "flush");
848 		feature_cnt++;
849 	}
850 
851 	if ((sc->xbd_flags & XBDF_BARRIER) != 0) {
852 		if (feature_cnt != 0)
853 			sbuf_printf(&sb, ", ");
854 		sbuf_printf(&sb, "write_barrier");
855 		feature_cnt++;
856 	}
857 
858 	if ((sc->xbd_flags & XBDF_DISCARD) != 0) {
859 		if (feature_cnt != 0)
860 			sbuf_printf(&sb, ", ");
861 		sbuf_printf(&sb, "discard");
862 		feature_cnt++;
863 	}
864 
865 	if ((sc->xbd_flags & XBDF_PERSISTENT) != 0) {
866 		if (feature_cnt != 0)
867 			sbuf_printf(&sb, ", ");
868 		sbuf_printf(&sb, "persistent_grants");
869 		feature_cnt++;
870 	}
871 
872 	(void) sbuf_finish(&sb);
873 	return (sbuf_len(&sb));
874 }
875 
876 static int
877 xbd_sysctl_features(SYSCTL_HANDLER_ARGS)
878 {
879 	char features[80];
880 	struct xbd_softc *sc = arg1;
881 	int error;
882 	int len;
883 
884 	error = sysctl_wire_old_buffer(req, 0);
885 	if (error != 0)
886 		return (error);
887 
888 	len = xbd_feature_string(sc, features, sizeof(features));
889 
890 	/* len is -1 on error, which will make the SYSCTL_OUT a no-op. */
891 	return (SYSCTL_OUT(req, features, len + 1/*NUL*/));
892 }
893 
894 static void
895 xbd_setup_sysctl(struct xbd_softc *xbd)
896 {
897 	struct sysctl_ctx_list *sysctl_ctx = NULL;
898 	struct sysctl_oid *sysctl_tree = NULL;
899 	struct sysctl_oid_list *children;
900 
901 	sysctl_ctx = device_get_sysctl_ctx(xbd->xbd_dev);
902 	if (sysctl_ctx == NULL)
903 		return;
904 
905 	sysctl_tree = device_get_sysctl_tree(xbd->xbd_dev);
906 	if (sysctl_tree == NULL)
907 		return;
908 
909 	children = SYSCTL_CHILDREN(sysctl_tree);
910 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
911 	    "max_requests", CTLFLAG_RD, &xbd->xbd_max_requests, -1,
912 	    "maximum outstanding requests (negotiated)");
913 
914 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
915 	    "max_request_segments", CTLFLAG_RD,
916 	    &xbd->xbd_max_request_segments, 0,
917 	    "maximum number of pages per requests (negotiated)");
918 
919 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
920 	    "max_request_size", CTLFLAG_RD, &xbd->xbd_max_request_size, 0,
921 	    "maximum size in bytes of a request (negotiated)");
922 
923 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
924 	    "ring_pages", CTLFLAG_RD, &xbd->xbd_ring_pages, 0,
925 	    "communication channel pages (negotiated)");
926 
927 	SYSCTL_ADD_PROC(sysctl_ctx, children, OID_AUTO,
928 	    "features", CTLTYPE_STRING|CTLFLAG_RD, xbd, 0,
929 	    xbd_sysctl_features, "A", "protocol features (negotiated)");
930 }
931 
932 /*
933  * Translate Linux major/minor to an appropriate name and unit
934  * number. For HVM guests, this allows us to use the same drive names
935  * with blkfront as the emulated drives, easing transition slightly.
936  */
937 static void
938 xbd_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name)
939 {
940 	static struct vdev_info {
941 		int major;
942 		int shift;
943 		int base;
944 		const char *name;
945 	} info[] = {
946 		{3,	6,	0,	"ada"},	/* ide0 */
947 		{22,	6,	2,	"ada"},	/* ide1 */
948 		{33,	6,	4,	"ada"},	/* ide2 */
949 		{34,	6,	6,	"ada"},	/* ide3 */
950 		{56,	6,	8,	"ada"},	/* ide4 */
951 		{57,	6,	10,	"ada"},	/* ide5 */
952 		{88,	6,	12,	"ada"},	/* ide6 */
953 		{89,	6,	14,	"ada"},	/* ide7 */
954 		{90,	6,	16,	"ada"},	/* ide8 */
955 		{91,	6,	18,	"ada"},	/* ide9 */
956 
957 		{8,	4,	0,	"da"},	/* scsi disk0 */
958 		{65,	4,	16,	"da"},	/* scsi disk1 */
959 		{66,	4,	32,	"da"},	/* scsi disk2 */
960 		{67,	4,	48,	"da"},	/* scsi disk3 */
961 		{68,	4,	64,	"da"},	/* scsi disk4 */
962 		{69,	4,	80,	"da"},	/* scsi disk5 */
963 		{70,	4,	96,	"da"},	/* scsi disk6 */
964 		{71,	4,	112,	"da"},	/* scsi disk7 */
965 		{128,	4,	128,	"da"},	/* scsi disk8 */
966 		{129,	4,	144,	"da"},	/* scsi disk9 */
967 		{130,	4,	160,	"da"},	/* scsi disk10 */
968 		{131,	4,	176,	"da"},	/* scsi disk11 */
969 		{132,	4,	192,	"da"},	/* scsi disk12 */
970 		{133,	4,	208,	"da"},	/* scsi disk13 */
971 		{134,	4,	224,	"da"},	/* scsi disk14 */
972 		{135,	4,	240,	"da"},	/* scsi disk15 */
973 
974 		{202,	4,	0,	"xbd"},	/* xbd */
975 
976 		{0,	0,	0,	NULL},
977 	};
978 	int major = vdevice >> 8;
979 	int minor = vdevice & 0xff;
980 	int i;
981 
982 	if (vdevice & (1 << 28)) {
983 		*unit = (vdevice & ((1 << 28) - 1)) >> 8;
984 		*name = "xbd";
985 		return;
986 	}
987 
988 	for (i = 0; info[i].major; i++) {
989 		if (info[i].major == major) {
990 			*unit = info[i].base + (minor >> info[i].shift);
991 			*name = info[i].name;
992 			return;
993 		}
994 	}
995 
996 	*unit = minor >> 4;
997 	*name = "xbd";
998 }
999 
1000 int
1001 xbd_instance_create(struct xbd_softc *sc, blkif_sector_t sectors,
1002     int vdevice, uint16_t vdisk_info, unsigned long sector_size,
1003     unsigned long phys_sector_size)
1004 {
1005 	char features[80];
1006 	int unit, error = 0;
1007 	const char *name;
1008 
1009 	xbd_vdevice_to_unit(vdevice, &unit, &name);
1010 
1011 	sc->xbd_unit = unit;
1012 
1013 	if (strcmp(name, "xbd") != 0)
1014 		device_printf(sc->xbd_dev, "attaching as %s%d\n", name, unit);
1015 
1016 	if (xbd_feature_string(sc, features, sizeof(features)) > 0) {
1017 		device_printf(sc->xbd_dev, "features: %s\n",
1018 		    features);
1019 	}
1020 
1021 	sc->xbd_disk = disk_alloc();
1022 	sc->xbd_disk->d_unit = sc->xbd_unit;
1023 	sc->xbd_disk->d_open = xbd_open;
1024 	sc->xbd_disk->d_close = xbd_close;
1025 	sc->xbd_disk->d_ioctl = xbd_ioctl;
1026 	sc->xbd_disk->d_strategy = xbd_strategy;
1027 	sc->xbd_disk->d_dump = xbd_dump;
1028 	sc->xbd_disk->d_name = name;
1029 	sc->xbd_disk->d_drv1 = sc;
1030 	sc->xbd_disk->d_sectorsize = sector_size;
1031 	sc->xbd_disk->d_stripesize = phys_sector_size;
1032 	sc->xbd_disk->d_stripeoffset = 0;
1033 
1034 	sc->xbd_disk->d_mediasize = sectors * sector_size;
1035 	sc->xbd_disk->d_maxsize = sc->xbd_max_request_size;
1036 	sc->xbd_disk->d_flags = DISKFLAG_UNMAPPED_BIO;
1037 	if ((sc->xbd_flags & (XBDF_FLUSH|XBDF_BARRIER)) != 0) {
1038 		sc->xbd_disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
1039 		device_printf(sc->xbd_dev,
1040 		    "synchronize cache commands enabled.\n");
1041 	}
1042 	disk_create(sc->xbd_disk, DISK_VERSION);
1043 
1044 	return error;
1045 }
1046 
1047 static void
1048 xbd_free(struct xbd_softc *sc)
1049 {
1050 	int i;
1051 
1052 	/* Prevent new requests being issued until we fix things up. */
1053 	mtx_lock(&sc->xbd_io_lock);
1054 	sc->xbd_state = XBD_STATE_DISCONNECTED;
1055 	mtx_unlock(&sc->xbd_io_lock);
1056 
1057 	/* Free resources associated with old device channel. */
1058 	xbd_free_ring(sc);
1059 	if (sc->xbd_shadow) {
1060 
1061 		for (i = 0; i < sc->xbd_max_requests; i++) {
1062 			struct xbd_command *cm;
1063 
1064 			cm = &sc->xbd_shadow[i];
1065 			if (cm->cm_sg_refs != NULL) {
1066 				free(cm->cm_sg_refs, M_XENBLOCKFRONT);
1067 				cm->cm_sg_refs = NULL;
1068 			}
1069 
1070 			if (cm->cm_indirectionpages != NULL) {
1071 				gnttab_end_foreign_access_references(
1072 				    sc->xbd_max_request_indirectpages,
1073 				    &cm->cm_indirectionrefs[0]);
1074 				contigfree(cm->cm_indirectionpages, PAGE_SIZE *
1075 				    sc->xbd_max_request_indirectpages,
1076 				    M_XENBLOCKFRONT);
1077 				cm->cm_indirectionpages = NULL;
1078 			}
1079 
1080 			bus_dmamap_destroy(sc->xbd_io_dmat, cm->cm_map);
1081 		}
1082 		free(sc->xbd_shadow, M_XENBLOCKFRONT);
1083 		sc->xbd_shadow = NULL;
1084 
1085 		bus_dma_tag_destroy(sc->xbd_io_dmat);
1086 
1087 		xbd_initq_cm(sc, XBD_Q_FREE);
1088 		xbd_initq_cm(sc, XBD_Q_READY);
1089 		xbd_initq_cm(sc, XBD_Q_COMPLETE);
1090 	}
1091 
1092 	xen_intr_unbind(&sc->xen_intr_handle);
1093 
1094 }
1095 
1096 /*--------------------------- State Change Handlers --------------------------*/
1097 static void
1098 xbd_initialize(struct xbd_softc *sc)
1099 {
1100 	const char *otherend_path;
1101 	const char *node_path;
1102 	uint32_t max_ring_page_order;
1103 	int error;
1104 
1105 	if (xenbus_get_state(sc->xbd_dev) != XenbusStateInitialising) {
1106 		/* Initialization has already been performed. */
1107 		return;
1108 	}
1109 
1110 	/*
1111 	 * Protocol defaults valid even if negotiation for a
1112 	 * setting fails.
1113 	 */
1114 	max_ring_page_order = 0;
1115 	sc->xbd_ring_pages = 1;
1116 
1117 	/*
1118 	 * Protocol negotiation.
1119 	 *
1120 	 * \note xs_gather() returns on the first encountered error, so
1121 	 *       we must use independent calls in order to guarantee
1122 	 *       we don't miss information in a sparsly populated back-end
1123 	 *       tree.
1124 	 *
1125 	 * \note xs_scanf() does not update variables for unmatched
1126 	 *	 fields.
1127 	 */
1128 	otherend_path = xenbus_get_otherend_path(sc->xbd_dev);
1129 	node_path = xenbus_get_node(sc->xbd_dev);
1130 
1131 	/* Support both backend schemes for relaying ring page limits. */
1132 	(void)xs_scanf(XST_NIL, otherend_path,
1133 	    "max-ring-page-order", NULL, "%" PRIu32,
1134 	    &max_ring_page_order);
1135 	sc->xbd_ring_pages = 1 << max_ring_page_order;
1136 	(void)xs_scanf(XST_NIL, otherend_path,
1137 	    "max-ring-pages", NULL, "%" PRIu32,
1138 	    &sc->xbd_ring_pages);
1139 	if (sc->xbd_ring_pages < 1)
1140 		sc->xbd_ring_pages = 1;
1141 
1142 	if (sc->xbd_ring_pages > XBD_MAX_RING_PAGES) {
1143 		device_printf(sc->xbd_dev,
1144 		    "Back-end specified ring-pages of %u "
1145 		    "limited to front-end limit of %u.\n",
1146 		    sc->xbd_ring_pages, XBD_MAX_RING_PAGES);
1147 		sc->xbd_ring_pages = XBD_MAX_RING_PAGES;
1148 	}
1149 
1150 	if (powerof2(sc->xbd_ring_pages) == 0) {
1151 		uint32_t new_page_limit;
1152 
1153 		new_page_limit = 0x01 << (fls(sc->xbd_ring_pages) - 1);
1154 		device_printf(sc->xbd_dev,
1155 		    "Back-end specified ring-pages of %u "
1156 		    "is not a power of 2. Limited to %u.\n",
1157 		    sc->xbd_ring_pages, new_page_limit);
1158 		sc->xbd_ring_pages = new_page_limit;
1159 	}
1160 
1161 	sc->xbd_max_requests =
1162 	    BLKIF_MAX_RING_REQUESTS(sc->xbd_ring_pages * PAGE_SIZE);
1163 	if (sc->xbd_max_requests > XBD_MAX_REQUESTS) {
1164 		device_printf(sc->xbd_dev,
1165 		    "Back-end specified max_requests of %u "
1166 		    "limited to front-end limit of %zu.\n",
1167 		    sc->xbd_max_requests, XBD_MAX_REQUESTS);
1168 		sc->xbd_max_requests = XBD_MAX_REQUESTS;
1169 	}
1170 
1171 	if (xbd_alloc_ring(sc) != 0)
1172 		return;
1173 
1174 	/* Support both backend schemes for relaying ring page limits. */
1175 	if (sc->xbd_ring_pages > 1) {
1176 		error = xs_printf(XST_NIL, node_path,
1177 		    "num-ring-pages","%u",
1178 		    sc->xbd_ring_pages);
1179 		if (error) {
1180 			xenbus_dev_fatal(sc->xbd_dev, error,
1181 			    "writing %s/num-ring-pages",
1182 			    node_path);
1183 			return;
1184 		}
1185 
1186 		error = xs_printf(XST_NIL, node_path,
1187 		    "ring-page-order", "%u",
1188 		    fls(sc->xbd_ring_pages) - 1);
1189 		if (error) {
1190 			xenbus_dev_fatal(sc->xbd_dev, error,
1191 			    "writing %s/ring-page-order",
1192 			    node_path);
1193 			return;
1194 		}
1195 	}
1196 
1197 	error = xs_printf(XST_NIL, node_path, "event-channel",
1198 	    "%u", xen_intr_port(sc->xen_intr_handle));
1199 	if (error) {
1200 		xenbus_dev_fatal(sc->xbd_dev, error,
1201 		    "writing %s/event-channel",
1202 		    node_path);
1203 		return;
1204 	}
1205 
1206 	error = xs_printf(XST_NIL, node_path, "protocol",
1207 	    "%s", XEN_IO_PROTO_ABI_NATIVE);
1208 	if (error) {
1209 		xenbus_dev_fatal(sc->xbd_dev, error,
1210 		    "writing %s/protocol",
1211 		    node_path);
1212 		return;
1213 	}
1214 
1215 	xenbus_set_state(sc->xbd_dev, XenbusStateInitialised);
1216 }
1217 
1218 /*
1219  * Invoked when the backend is finally 'ready' (and has published
1220  * the details about the physical device - #sectors, size, etc).
1221  */
1222 static void
1223 xbd_connect(struct xbd_softc *sc)
1224 {
1225 	device_t dev = sc->xbd_dev;
1226 	unsigned long sectors, sector_size, phys_sector_size;
1227 	unsigned int binfo;
1228 	int err, feature_barrier, feature_flush;
1229 	int i, j;
1230 
1231 	if (sc->xbd_state == XBD_STATE_CONNECTED ||
1232 	    sc->xbd_state == XBD_STATE_SUSPENDED)
1233 		return;
1234 
1235 	DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev));
1236 
1237 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1238 	    "sectors", "%lu", &sectors,
1239 	    "info", "%u", &binfo,
1240 	    "sector-size", "%lu", &sector_size,
1241 	    NULL);
1242 	if (err) {
1243 		xenbus_dev_fatal(dev, err,
1244 		    "reading backend fields at %s",
1245 		    xenbus_get_otherend_path(dev));
1246 		return;
1247 	}
1248 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1249 	     "physical-sector-size", "%lu", &phys_sector_size,
1250 	     NULL);
1251 	if (err || phys_sector_size <= sector_size)
1252 		phys_sector_size = 0;
1253 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1254 	     "feature-barrier", "%lu", &feature_barrier,
1255 	     NULL);
1256 	if (err == 0 && feature_barrier != 0)
1257 		sc->xbd_flags |= XBDF_BARRIER;
1258 
1259 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1260 	     "feature-flush-cache", "%lu", &feature_flush,
1261 	     NULL);
1262 	if (err == 0 && feature_flush != 0)
1263 		sc->xbd_flags |= XBDF_FLUSH;
1264 
1265 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1266 	    "feature-max-indirect-segments", "%" PRIu32,
1267 	    &sc->xbd_max_request_segments, NULL);
1268 	if ((err != 0) || (xbd_enable_indirect == 0))
1269 		sc->xbd_max_request_segments = 0;
1270 	if (sc->xbd_max_request_segments > XBD_MAX_INDIRECT_SEGMENTS)
1271 		sc->xbd_max_request_segments = XBD_MAX_INDIRECT_SEGMENTS;
1272 	if (sc->xbd_max_request_segments > XBD_SIZE_TO_SEGS(MAXPHYS))
1273 		sc->xbd_max_request_segments = XBD_SIZE_TO_SEGS(MAXPHYS);
1274 	sc->xbd_max_request_indirectpages =
1275 	    XBD_INDIRECT_SEGS_TO_PAGES(sc->xbd_max_request_segments);
1276 	if (sc->xbd_max_request_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
1277 		sc->xbd_max_request_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1278 	sc->xbd_max_request_size =
1279 	    XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments);
1280 
1281 	/* Allocate datastructures based on negotiated values. */
1282 	err = bus_dma_tag_create(
1283 	    bus_get_dma_tag(sc->xbd_dev),	/* parent */
1284 	    512, PAGE_SIZE,			/* algnmnt, boundary */
1285 	    BUS_SPACE_MAXADDR,			/* lowaddr */
1286 	    BUS_SPACE_MAXADDR,			/* highaddr */
1287 	    NULL, NULL,				/* filter, filterarg */
1288 	    sc->xbd_max_request_size,
1289 	    sc->xbd_max_request_segments,
1290 	    PAGE_SIZE,				/* maxsegsize */
1291 	    BUS_DMA_ALLOCNOW,			/* flags */
1292 	    busdma_lock_mutex,			/* lockfunc */
1293 	    &sc->xbd_io_lock,			/* lockarg */
1294 	    &sc->xbd_io_dmat);
1295 	if (err != 0) {
1296 		xenbus_dev_fatal(sc->xbd_dev, err,
1297 		    "Cannot allocate parent DMA tag\n");
1298 		return;
1299 	}
1300 
1301 	/* Per-transaction data allocation. */
1302 	sc->xbd_shadow = malloc(sizeof(*sc->xbd_shadow) * sc->xbd_max_requests,
1303 	    M_XENBLOCKFRONT, M_NOWAIT|M_ZERO);
1304 	if (sc->xbd_shadow == NULL) {
1305 		bus_dma_tag_destroy(sc->xbd_io_dmat);
1306 		xenbus_dev_fatal(sc->xbd_dev, ENOMEM,
1307 		    "Cannot allocate request structures\n");
1308 		return;
1309 	}
1310 
1311 	for (i = 0; i < sc->xbd_max_requests; i++) {
1312 		struct xbd_command *cm;
1313 		void * indirectpages;
1314 
1315 		cm = &sc->xbd_shadow[i];
1316 		cm->cm_sg_refs = malloc(
1317 		    sizeof(grant_ref_t) * sc->xbd_max_request_segments,
1318 		    M_XENBLOCKFRONT, M_NOWAIT);
1319 		if (cm->cm_sg_refs == NULL)
1320 			break;
1321 		cm->cm_id = i;
1322 		cm->cm_flags = XBDCF_INITIALIZER;
1323 		cm->cm_sc = sc;
1324 		if (bus_dmamap_create(sc->xbd_io_dmat, 0, &cm->cm_map) != 0)
1325 			break;
1326 		if (sc->xbd_max_request_indirectpages > 0) {
1327 			indirectpages = contigmalloc(
1328 			    PAGE_SIZE * sc->xbd_max_request_indirectpages,
1329 			    M_XENBLOCKFRONT, M_ZERO, 0, ~0, PAGE_SIZE, 0);
1330 		} else {
1331 			indirectpages = NULL;
1332 		}
1333 		for (j = 0; j < sc->xbd_max_request_indirectpages; j++) {
1334 			if (gnttab_grant_foreign_access(
1335 			    xenbus_get_otherend_id(sc->xbd_dev),
1336 			    (vtophys(indirectpages) >> PAGE_SHIFT) + j,
1337 			    1 /* grant read-only access */,
1338 			    &cm->cm_indirectionrefs[j]))
1339 				break;
1340 		}
1341 		if (j < sc->xbd_max_request_indirectpages)
1342 			break;
1343 		cm->cm_indirectionpages = indirectpages;
1344 		xbd_free_command(cm);
1345 	}
1346 
1347 	if (sc->xbd_disk == NULL) {
1348 		device_printf(dev, "%juMB <%s> at %s",
1349 		    (uintmax_t) sectors / (1048576 / sector_size),
1350 		    device_get_desc(dev),
1351 		    xenbus_get_node(dev));
1352 		bus_print_child_footer(device_get_parent(dev), dev);
1353 
1354 		xbd_instance_create(sc, sectors, sc->xbd_vdevice, binfo,
1355 		    sector_size, phys_sector_size);
1356 	}
1357 
1358 	(void)xenbus_set_state(dev, XenbusStateConnected);
1359 
1360 	/* Kick pending requests. */
1361 	mtx_lock(&sc->xbd_io_lock);
1362 	sc->xbd_state = XBD_STATE_CONNECTED;
1363 	xbd_startio(sc);
1364 	sc->xbd_flags |= XBDF_READY;
1365 	mtx_unlock(&sc->xbd_io_lock);
1366 }
1367 
1368 /**
1369  * Handle the change of state of the backend to Closing.  We must delete our
1370  * device-layer structures now, to ensure that writes are flushed through to
1371  * the backend.  Once this is done, we can switch to Closed in
1372  * acknowledgement.
1373  */
1374 static void
1375 xbd_closing(device_t dev)
1376 {
1377 	struct xbd_softc *sc = device_get_softc(dev);
1378 
1379 	xenbus_set_state(dev, XenbusStateClosing);
1380 
1381 	DPRINTK("xbd_closing: %s removed\n", xenbus_get_node(dev));
1382 
1383 	if (sc->xbd_disk != NULL) {
1384 		disk_destroy(sc->xbd_disk);
1385 		sc->xbd_disk = NULL;
1386 	}
1387 
1388 	xenbus_set_state(dev, XenbusStateClosed);
1389 }
1390 
1391 /*---------------------------- NewBus Entrypoints ----------------------------*/
1392 static int
1393 xbd_probe(device_t dev)
1394 {
1395 	if (strcmp(xenbus_get_type(dev), "vbd") != 0)
1396 		return (ENXIO);
1397 
1398 	if (xen_hvm_domain() && xen_disable_pv_disks != 0)
1399 		return (ENXIO);
1400 
1401 	if (xen_hvm_domain()) {
1402 		int error;
1403 		char *type;
1404 
1405 		/*
1406 		 * When running in an HVM domain, IDE disk emulation is
1407 		 * disabled early in boot so that native drivers will
1408 		 * not see emulated hardware.  However, CDROM device
1409 		 * emulation cannot be disabled.
1410 		 *
1411 		 * Through use of FreeBSD's vm_guest and xen_hvm_domain()
1412 		 * APIs, we could modify the native CDROM driver to fail its
1413 		 * probe when running under Xen.  Unfortunatlely, the PV
1414 		 * CDROM support in XenServer (up through at least version
1415 		 * 6.2) isn't functional, so we instead rely on the emulated
1416 		 * CDROM instance, and fail to attach the PV one here in
1417 		 * the blkfront driver.
1418 		 */
1419 		error = xs_read(XST_NIL, xenbus_get_node(dev),
1420 		    "device-type", NULL, (void **) &type);
1421 		if (error)
1422 			return (ENXIO);
1423 
1424 		if (strncmp(type, "cdrom", 5) == 0) {
1425 			free(type, M_XENSTORE);
1426 			return (ENXIO);
1427 		}
1428 		free(type, M_XENSTORE);
1429 	}
1430 
1431 	device_set_desc(dev, "Virtual Block Device");
1432 	device_quiet(dev);
1433 	return (0);
1434 }
1435 
1436 /*
1437  * Setup supplies the backend dir, virtual device.  We place an event
1438  * channel and shared frame entries.  We watch backend to wait if it's
1439  * ok.
1440  */
1441 static int
1442 xbd_attach(device_t dev)
1443 {
1444 	struct xbd_softc *sc;
1445 	const char *name;
1446 	uint32_t vdevice;
1447 	int error;
1448 	int i;
1449 	int unit;
1450 
1451 	/* FIXME: Use dynamic device id if this is not set. */
1452 	error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1453 	    "virtual-device", NULL, "%" PRIu32, &vdevice);
1454 	if (error)
1455 		error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1456 		    "virtual-device-ext", NULL, "%" PRIu32, &vdevice);
1457 	if (error) {
1458 		xenbus_dev_fatal(dev, error, "reading virtual-device");
1459 		device_printf(dev, "Couldn't determine virtual device.\n");
1460 		return (error);
1461 	}
1462 
1463 	xbd_vdevice_to_unit(vdevice, &unit, &name);
1464 	if (!strcmp(name, "xbd"))
1465 		device_set_unit(dev, unit);
1466 
1467 	sc = device_get_softc(dev);
1468 	mtx_init(&sc->xbd_io_lock, "blkfront i/o lock", NULL, MTX_DEF);
1469 	xbd_initqs(sc);
1470 	for (i = 0; i < XBD_MAX_RING_PAGES; i++)
1471 		sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
1472 
1473 	sc->xbd_dev = dev;
1474 	sc->xbd_vdevice = vdevice;
1475 	sc->xbd_state = XBD_STATE_DISCONNECTED;
1476 
1477 	xbd_setup_sysctl(sc);
1478 
1479 	/* Wait for backend device to publish its protocol capabilities. */
1480 	xenbus_set_state(dev, XenbusStateInitialising);
1481 
1482 	return (0);
1483 }
1484 
1485 static int
1486 xbd_detach(device_t dev)
1487 {
1488 	struct xbd_softc *sc = device_get_softc(dev);
1489 
1490 	DPRINTK("%s: %s removed\n", __func__, xenbus_get_node(dev));
1491 
1492 	xbd_free(sc);
1493 	mtx_destroy(&sc->xbd_io_lock);
1494 
1495 	return 0;
1496 }
1497 
1498 static int
1499 xbd_suspend(device_t dev)
1500 {
1501 	struct xbd_softc *sc = device_get_softc(dev);
1502 	int retval;
1503 	int saved_state;
1504 
1505 	/* Prevent new requests being issued until we fix things up. */
1506 	mtx_lock(&sc->xbd_io_lock);
1507 	saved_state = sc->xbd_state;
1508 	sc->xbd_state = XBD_STATE_SUSPENDED;
1509 
1510 	/* Wait for outstanding I/O to drain. */
1511 	retval = 0;
1512 	while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
1513 		if (msleep(&sc->xbd_cm_q[XBD_Q_BUSY], &sc->xbd_io_lock,
1514 		    PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) {
1515 			retval = EBUSY;
1516 			break;
1517 		}
1518 	}
1519 	mtx_unlock(&sc->xbd_io_lock);
1520 
1521 	if (retval != 0)
1522 		sc->xbd_state = saved_state;
1523 
1524 	return (retval);
1525 }
1526 
1527 static int
1528 xbd_resume(device_t dev)
1529 {
1530 	struct xbd_softc *sc = device_get_softc(dev);
1531 
1532 	DPRINTK("xbd_resume: %s\n", xenbus_get_node(dev));
1533 
1534 	xbd_free(sc);
1535 	xbd_initialize(sc);
1536 	return (0);
1537 }
1538 
1539 /**
1540  * Callback received when the backend's state changes.
1541  */
1542 static void
1543 xbd_backend_changed(device_t dev, XenbusState backend_state)
1544 {
1545 	struct xbd_softc *sc = device_get_softc(dev);
1546 
1547 	DPRINTK("backend_state=%d\n", backend_state);
1548 
1549 	switch (backend_state) {
1550 	case XenbusStateUnknown:
1551 	case XenbusStateInitialising:
1552 	case XenbusStateReconfigured:
1553 	case XenbusStateReconfiguring:
1554 	case XenbusStateClosed:
1555 		break;
1556 
1557 	case XenbusStateInitWait:
1558 	case XenbusStateInitialised:
1559 		xbd_initialize(sc);
1560 		break;
1561 
1562 	case XenbusStateConnected:
1563 		xbd_initialize(sc);
1564 		xbd_connect(sc);
1565 		break;
1566 
1567 	case XenbusStateClosing:
1568 		if (sc->xbd_users > 0)
1569 			xenbus_dev_error(dev, -EBUSY,
1570 			    "Device in use; refusing to close");
1571 		else
1572 			xbd_closing(dev);
1573 		break;
1574 	}
1575 }
1576 
1577 /*---------------------------- NewBus Registration ---------------------------*/
1578 static device_method_t xbd_methods[] = {
1579 	/* Device interface */
1580 	DEVMETHOD(device_probe,         xbd_probe),
1581 	DEVMETHOD(device_attach,        xbd_attach),
1582 	DEVMETHOD(device_detach,        xbd_detach),
1583 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
1584 	DEVMETHOD(device_suspend,       xbd_suspend),
1585 	DEVMETHOD(device_resume,        xbd_resume),
1586 
1587 	/* Xenbus interface */
1588 	DEVMETHOD(xenbus_otherend_changed, xbd_backend_changed),
1589 
1590 	{ 0, 0 }
1591 };
1592 
1593 static driver_t xbd_driver = {
1594 	"xbd",
1595 	xbd_methods,
1596 	sizeof(struct xbd_softc),
1597 };
1598 devclass_t xbd_devclass;
1599 
1600 DRIVER_MODULE(xbd, xenbusb_front, xbd_driver, xbd_devclass, 0, 0);
1601