xref: /freebsd/sys/dev/xen/blkfront/blkfront.c (revision 349cc55c9796c4596a5b9904cd3281af295f878f)
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/vmparam.h>
51 
52 #include <xen/xen-os.h>
53 #include <xen/hypervisor.h>
54 #include <xen/xen_intr.h>
55 #include <xen/gnttab.h>
56 #include <contrib/xen/grant_table.h>
57 #include <contrib/xen/io/protocols.h>
58 #include <xen/xenbus/xenbusvar.h>
59 
60 #include <machine/_inttypes.h>
61 
62 #include <geom/geom_disk.h>
63 
64 #include <dev/xen/blkfront/block.h>
65 
66 #include "xenbus_if.h"
67 
68 /*--------------------------- Forward Declarations ---------------------------*/
69 static void xbd_closing(device_t);
70 static void xbd_startio(struct xbd_softc *sc);
71 
72 /*---------------------------------- Macros ----------------------------------*/
73 #if 0
74 #define DPRINTK(fmt, args...) printf("[XEN] %s:%d: " fmt ".\n", __func__, __LINE__, ##args)
75 #else
76 #define DPRINTK(fmt, args...)
77 #endif
78 
79 #define XBD_SECTOR_SHFT		9
80 
81 /*---------------------------- Global Static Data ----------------------------*/
82 static MALLOC_DEFINE(M_XENBLOCKFRONT, "xbd", "Xen Block Front driver data");
83 
84 static int xbd_enable_indirect = 1;
85 SYSCTL_NODE(_hw, OID_AUTO, xbd, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
86     "xbd driver parameters");
87 SYSCTL_INT(_hw_xbd, OID_AUTO, xbd_enable_indirect, CTLFLAG_RDTUN,
88     &xbd_enable_indirect, 0, "Enable xbd indirect segments");
89 
90 /*---------------------------- Command Processing ----------------------------*/
91 static void
92 xbd_freeze(struct xbd_softc *sc, xbd_flag_t xbd_flag)
93 {
94 	if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) != 0)
95 		return;
96 
97 	sc->xbd_flags |= xbd_flag;
98 	sc->xbd_qfrozen_cnt++;
99 }
100 
101 static void
102 xbd_thaw(struct xbd_softc *sc, xbd_flag_t xbd_flag)
103 {
104 	if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) == 0)
105 		return;
106 
107 	if (sc->xbd_qfrozen_cnt == 0)
108 		panic("%s: Thaw with flag 0x%x while not frozen.",
109 		    __func__, xbd_flag);
110 
111 	sc->xbd_flags &= ~xbd_flag;
112 	sc->xbd_qfrozen_cnt--;
113 }
114 
115 static void
116 xbd_cm_freeze(struct xbd_softc *sc, struct xbd_command *cm, xbdc_flag_t cm_flag)
117 {
118 	if ((cm->cm_flags & XBDCF_FROZEN) != 0)
119 		return;
120 
121 	cm->cm_flags |= XBDCF_FROZEN|cm_flag;
122 	xbd_freeze(sc, XBDF_NONE);
123 }
124 
125 static void
126 xbd_cm_thaw(struct xbd_softc *sc, struct xbd_command *cm)
127 {
128 	if ((cm->cm_flags & XBDCF_FROZEN) == 0)
129 		return;
130 
131 	cm->cm_flags &= ~XBDCF_FROZEN;
132 	xbd_thaw(sc, XBDF_NONE);
133 }
134 
135 static inline void
136 xbd_flush_requests(struct xbd_softc *sc)
137 {
138 	int notify;
139 
140 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->xbd_ring, notify);
141 
142 	if (notify)
143 		xen_intr_signal(sc->xen_intr_handle);
144 }
145 
146 static void
147 xbd_free_command(struct xbd_command *cm)
148 {
149 
150 	KASSERT((cm->cm_flags & XBDCF_Q_MASK) == XBD_Q_NONE,
151 	    ("Freeing command that is still on queue %d.",
152 	    cm->cm_flags & XBDCF_Q_MASK));
153 
154 	cm->cm_flags = XBDCF_INITIALIZER;
155 	cm->cm_bp = NULL;
156 	cm->cm_complete = NULL;
157 	xbd_enqueue_cm(cm, XBD_Q_FREE);
158 	xbd_thaw(cm->cm_sc, XBDF_CM_SHORTAGE);
159 }
160 
161 static void
162 xbd_mksegarray(bus_dma_segment_t *segs, int nsegs,
163     grant_ref_t * gref_head, int otherend_id, int readonly,
164     grant_ref_t * sg_ref, struct blkif_request_segment *sg)
165 {
166 	struct blkif_request_segment *last_block_sg = sg + nsegs;
167 	vm_paddr_t buffer_ma;
168 	uint64_t fsect, lsect;
169 	int ref;
170 
171 	while (sg < last_block_sg) {
172 		KASSERT(segs->ds_addr % (1 << XBD_SECTOR_SHFT) == 0,
173 		    ("XEN disk driver I/O must be sector aligned"));
174 		KASSERT(segs->ds_len % (1 << XBD_SECTOR_SHFT) == 0,
175 		    ("XEN disk driver I/Os must be a multiple of "
176 		    "the sector length"));
177 		buffer_ma = segs->ds_addr;
178 		fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT;
179 		lsect = fsect + (segs->ds_len  >> XBD_SECTOR_SHFT) - 1;
180 
181 		KASSERT(lsect <= 7, ("XEN disk driver data cannot "
182 		    "cross a page boundary"));
183 
184 		/* install a grant reference. */
185 		ref = gnttab_claim_grant_reference(gref_head);
186 
187 		/*
188 		 * GNTTAB_LIST_END == 0xffffffff, but it is private
189 		 * to gnttab.c.
190 		 */
191 		KASSERT(ref != ~0, ("grant_reference failed"));
192 
193 		gnttab_grant_foreign_access_ref(
194 		    ref,
195 		    otherend_id,
196 		    buffer_ma >> PAGE_SHIFT,
197 		    readonly);
198 
199 		*sg_ref = ref;
200 		*sg = (struct blkif_request_segment) {
201 			.gref       = ref,
202 			.first_sect = fsect,
203 			.last_sect  = lsect
204 		};
205 		sg++;
206 		sg_ref++;
207 		segs++;
208 	}
209 }
210 
211 static void
212 xbd_queue_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
213 {
214 	struct xbd_softc *sc;
215 	struct xbd_command *cm;
216 	int op;
217 
218 	cm = arg;
219 	sc = cm->cm_sc;
220 
221 	if (error) {
222 		cm->cm_bp->bio_error = EIO;
223 		biodone(cm->cm_bp);
224 		xbd_free_command(cm);
225 		return;
226 	}
227 
228 	KASSERT(nsegs <= sc->xbd_max_request_segments,
229 	    ("Too many segments in a blkfront I/O"));
230 
231 	if (nsegs <= BLKIF_MAX_SEGMENTS_PER_REQUEST) {
232 		blkif_request_t	*ring_req;
233 
234 		/* Fill out a blkif_request_t structure. */
235 		ring_req = (blkif_request_t *)
236 		    RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt);
237 		sc->xbd_ring.req_prod_pvt++;
238 		ring_req->id = cm->cm_id;
239 		ring_req->operation = cm->cm_operation;
240 		ring_req->sector_number = cm->cm_sector_number;
241 		ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk;
242 		ring_req->nr_segments = nsegs;
243 		cm->cm_nseg = nsegs;
244 		xbd_mksegarray(segs, nsegs, &cm->cm_gref_head,
245 		    xenbus_get_otherend_id(sc->xbd_dev),
246 		    cm->cm_operation == BLKIF_OP_WRITE,
247 		    cm->cm_sg_refs, ring_req->seg);
248 	} else {
249 		blkif_request_indirect_t *ring_req;
250 
251 		/* Fill out a blkif_request_indirect_t structure. */
252 		ring_req = (blkif_request_indirect_t *)
253 		    RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt);
254 		sc->xbd_ring.req_prod_pvt++;
255 		ring_req->id = cm->cm_id;
256 		ring_req->operation = BLKIF_OP_INDIRECT;
257 		ring_req->indirect_op = cm->cm_operation;
258 		ring_req->sector_number = cm->cm_sector_number;
259 		ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk;
260 		ring_req->nr_segments = nsegs;
261 		cm->cm_nseg = nsegs;
262 		xbd_mksegarray(segs, nsegs, &cm->cm_gref_head,
263 		    xenbus_get_otherend_id(sc->xbd_dev),
264 		    cm->cm_operation == BLKIF_OP_WRITE,
265 		    cm->cm_sg_refs, cm->cm_indirectionpages);
266 		memcpy(ring_req->indirect_grefs, &cm->cm_indirectionrefs,
267 		    sizeof(grant_ref_t) * sc->xbd_max_request_indirectpages);
268 	}
269 
270 	if (cm->cm_operation == BLKIF_OP_READ)
271 		op = BUS_DMASYNC_PREREAD;
272 	else if (cm->cm_operation == BLKIF_OP_WRITE)
273 		op = BUS_DMASYNC_PREWRITE;
274 	else
275 		op = 0;
276 	bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op);
277 
278 	gnttab_free_grant_references(cm->cm_gref_head);
279 
280 	xbd_enqueue_cm(cm, XBD_Q_BUSY);
281 
282 	/*
283 	 * If bus dma had to asynchronously call us back to dispatch
284 	 * this command, we are no longer executing in the context of
285 	 * xbd_startio().  Thus we cannot rely on xbd_startio()'s call to
286 	 * xbd_flush_requests() to publish this command to the backend
287 	 * along with any other commands that it could batch.
288 	 */
289 	if ((cm->cm_flags & XBDCF_ASYNC_MAPPING) != 0)
290 		xbd_flush_requests(sc);
291 
292 	return;
293 }
294 
295 static int
296 xbd_queue_request(struct xbd_softc *sc, struct xbd_command *cm)
297 {
298 	int error;
299 
300 	if (cm->cm_bp != NULL)
301 		error = bus_dmamap_load_bio(sc->xbd_io_dmat, cm->cm_map,
302 		    cm->cm_bp, xbd_queue_cb, cm, 0);
303 	else
304 		error = bus_dmamap_load(sc->xbd_io_dmat, cm->cm_map,
305 		    cm->cm_data, cm->cm_datalen, xbd_queue_cb, cm, 0);
306 	if (error == EINPROGRESS) {
307 		/*
308 		 * Maintain queuing order by freezing the queue.  The next
309 		 * command may not require as many resources as the command
310 		 * we just attempted to map, so we can't rely on bus dma
311 		 * blocking for it too.
312 		 */
313 		xbd_cm_freeze(sc, cm, XBDCF_ASYNC_MAPPING);
314 		return (0);
315 	}
316 
317 	return (error);
318 }
319 
320 static void
321 xbd_restart_queue_callback(void *arg)
322 {
323 	struct xbd_softc *sc = arg;
324 
325 	mtx_lock(&sc->xbd_io_lock);
326 
327 	xbd_thaw(sc, XBDF_GNT_SHORTAGE);
328 
329 	xbd_startio(sc);
330 
331 	mtx_unlock(&sc->xbd_io_lock);
332 }
333 
334 static struct xbd_command *
335 xbd_bio_command(struct xbd_softc *sc)
336 {
337 	struct xbd_command *cm;
338 	struct bio *bp;
339 
340 	if (__predict_false(sc->xbd_state != XBD_STATE_CONNECTED))
341 		return (NULL);
342 
343 	bp = xbd_dequeue_bio(sc);
344 	if (bp == NULL)
345 		return (NULL);
346 
347 	if ((cm = xbd_dequeue_cm(sc, XBD_Q_FREE)) == NULL) {
348 		xbd_freeze(sc, XBDF_CM_SHORTAGE);
349 		xbd_requeue_bio(sc, bp);
350 		return (NULL);
351 	}
352 
353 	if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
354 	    &cm->cm_gref_head) != 0) {
355 		gnttab_request_free_callback(&sc->xbd_callback,
356 		    xbd_restart_queue_callback, sc,
357 		    sc->xbd_max_request_segments);
358 		xbd_freeze(sc, XBDF_GNT_SHORTAGE);
359 		xbd_requeue_bio(sc, bp);
360 		xbd_enqueue_cm(cm, XBD_Q_FREE);
361 		return (NULL);
362 	}
363 
364 	cm->cm_bp = bp;
365 	cm->cm_sector_number = (blkif_sector_t)bp->bio_pblkno;
366 
367 	switch (bp->bio_cmd) {
368 	case BIO_READ:
369 		cm->cm_operation = BLKIF_OP_READ;
370 		break;
371 	case BIO_WRITE:
372 		cm->cm_operation = BLKIF_OP_WRITE;
373 		if ((bp->bio_flags & BIO_ORDERED) != 0) {
374 			if ((sc->xbd_flags & XBDF_BARRIER) != 0) {
375 				cm->cm_operation = BLKIF_OP_WRITE_BARRIER;
376 			} else {
377 				/*
378 				 * Single step this command.
379 				 */
380 				cm->cm_flags |= XBDCF_Q_FREEZE;
381 				if (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
382 					/*
383 					 * Wait for in-flight requests to
384 					 * finish.
385 					 */
386 					xbd_freeze(sc, XBDF_WAIT_IDLE);
387 					xbd_requeue_cm(cm, XBD_Q_READY);
388 					return (NULL);
389 				}
390 			}
391 		}
392 		break;
393 	case BIO_FLUSH:
394 		if ((sc->xbd_flags & XBDF_FLUSH) != 0)
395 			cm->cm_operation = BLKIF_OP_FLUSH_DISKCACHE;
396 		else if ((sc->xbd_flags & XBDF_BARRIER) != 0)
397 			cm->cm_operation = BLKIF_OP_WRITE_BARRIER;
398 		else
399 			panic("flush request, but no flush support available");
400 		break;
401 	default:
402 		biofinish(bp, NULL, EOPNOTSUPP);
403 		xbd_enqueue_cm(cm, XBD_Q_FREE);
404 		return (NULL);
405 	}
406 
407 	return (cm);
408 }
409 
410 /*
411  * Dequeue buffers and place them in the shared communication ring.
412  * Return when no more requests can be accepted or all buffers have
413  * been queued.
414  *
415  * Signal XEN once the ring has been filled out.
416  */
417 static void
418 xbd_startio(struct xbd_softc *sc)
419 {
420 	struct xbd_command *cm;
421 	int error, queued = 0;
422 
423 	mtx_assert(&sc->xbd_io_lock, MA_OWNED);
424 
425 	if (sc->xbd_state != XBD_STATE_CONNECTED)
426 		return;
427 
428 	while (!RING_FULL(&sc->xbd_ring)) {
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, off_t offset, size_t length)
597 {
598 	struct disk *dp = arg;
599 	struct xbd_softc *sc = dp->d_drv1;
600 	struct xbd_command *cm;
601 	size_t chunk;
602 	int sbp;
603 	int rc = 0;
604 
605 	if (length == 0)
606 		return (0);
607 
608 	xbd_quiesce(sc);	/* All quiet on the western front. */
609 
610 	/*
611 	 * If this lock is held, then this module is failing, and a
612 	 * successful kernel dump is highly unlikely anyway.
613 	 */
614 	mtx_lock(&sc->xbd_io_lock);
615 
616 	/* Split the 64KB block as needed */
617 	for (sbp=0; length > 0; sbp++) {
618 		cm = xbd_dequeue_cm(sc, XBD_Q_FREE);
619 		if (cm == NULL) {
620 			mtx_unlock(&sc->xbd_io_lock);
621 			device_printf(sc->xbd_dev, "dump: no more commands?\n");
622 			return (EBUSY);
623 		}
624 
625 		if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
626 		    &cm->cm_gref_head) != 0) {
627 			xbd_free_command(cm);
628 			mtx_unlock(&sc->xbd_io_lock);
629 			device_printf(sc->xbd_dev, "no more grant allocs?\n");
630 			return (EBUSY);
631 		}
632 
633 		chunk = length > sc->xbd_max_request_size ?
634 		    sc->xbd_max_request_size : length;
635 		cm->cm_data = virtual;
636 		cm->cm_datalen = chunk;
637 		cm->cm_operation = BLKIF_OP_WRITE;
638 		cm->cm_sector_number = offset / dp->d_sectorsize;
639 		cm->cm_complete = xbd_dump_complete;
640 
641 		xbd_enqueue_cm(cm, XBD_Q_READY);
642 
643 		length -= chunk;
644 		offset += chunk;
645 		virtual = (char *) virtual + chunk;
646 	}
647 
648 	/* Tell DOM0 to do the I/O */
649 	xbd_startio(sc);
650 	mtx_unlock(&sc->xbd_io_lock);
651 
652 	/* Poll for the completion. */
653 	xbd_quiesce(sc);	/* All quite on the eastern front */
654 
655 	/* If there were any errors, bail out... */
656 	while ((cm = xbd_dequeue_cm(sc, XBD_Q_COMPLETE)) != NULL) {
657 		if (cm->cm_status != BLKIF_RSP_OKAY) {
658 			device_printf(sc->xbd_dev,
659 			    "Dump I/O failed at sector %jd\n",
660 			    cm->cm_sector_number);
661 			rc = EIO;
662 		}
663 		xbd_free_command(cm);
664 	}
665 
666 	return (rc);
667 }
668 
669 /*----------------------------- Disk Entrypoints -----------------------------*/
670 static int
671 xbd_open(struct disk *dp)
672 {
673 	struct xbd_softc *sc = dp->d_drv1;
674 
675 	if (sc == NULL) {
676 		printf("xbd%d: not found", dp->d_unit);
677 		return (ENXIO);
678 	}
679 
680 	sc->xbd_flags |= XBDF_OPEN;
681 	sc->xbd_users++;
682 	return (0);
683 }
684 
685 static int
686 xbd_close(struct disk *dp)
687 {
688 	struct xbd_softc *sc = dp->d_drv1;
689 
690 	if (sc == NULL)
691 		return (ENXIO);
692 	sc->xbd_flags &= ~XBDF_OPEN;
693 	if (--(sc->xbd_users) == 0) {
694 		/*
695 		 * Check whether we have been instructed to close.  We will
696 		 * have ignored this request initially, as the device was
697 		 * still mounted.
698 		 */
699 		if (xenbus_get_otherend_state(sc->xbd_dev) ==
700 		    XenbusStateClosing)
701 			xbd_closing(sc->xbd_dev);
702 	}
703 	return (0);
704 }
705 
706 static int
707 xbd_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
708 {
709 	struct xbd_softc *sc = dp->d_drv1;
710 
711 	if (sc == NULL)
712 		return (ENXIO);
713 
714 	return (ENOTTY);
715 }
716 
717 /*
718  * Read/write routine for a buffer.  Finds the proper unit, place it on
719  * the sortq and kick the controller.
720  */
721 static void
722 xbd_strategy(struct bio *bp)
723 {
724 	struct xbd_softc *sc = bp->bio_disk->d_drv1;
725 
726 	/* bogus disk? */
727 	if (sc == NULL) {
728 		bp->bio_error = EINVAL;
729 		bp->bio_flags |= BIO_ERROR;
730 		bp->bio_resid = bp->bio_bcount;
731 		biodone(bp);
732 		return;
733 	}
734 
735 	/*
736 	 * Place it in the queue of disk activities for this disk
737 	 */
738 	mtx_lock(&sc->xbd_io_lock);
739 
740 	xbd_enqueue_bio(sc, bp);
741 	xbd_startio(sc);
742 
743 	mtx_unlock(&sc->xbd_io_lock);
744 	return;
745 }
746 
747 /*------------------------------ Ring Management -----------------------------*/
748 static int
749 xbd_alloc_ring(struct xbd_softc *sc)
750 {
751 	blkif_sring_t *sring;
752 	uintptr_t sring_page_addr;
753 	int error;
754 	int i;
755 
756 	sring = malloc(sc->xbd_ring_pages * PAGE_SIZE, M_XENBLOCKFRONT,
757 	    M_NOWAIT|M_ZERO);
758 	if (sring == NULL) {
759 		xenbus_dev_fatal(sc->xbd_dev, ENOMEM, "allocating shared ring");
760 		return (ENOMEM);
761 	}
762 	SHARED_RING_INIT(sring);
763 	FRONT_RING_INIT(&sc->xbd_ring, sring, sc->xbd_ring_pages * PAGE_SIZE);
764 
765 	for (i = 0, sring_page_addr = (uintptr_t)sring;
766 	     i < sc->xbd_ring_pages;
767 	     i++, sring_page_addr += PAGE_SIZE) {
768 		error = xenbus_grant_ring(sc->xbd_dev,
769 		    (vtophys(sring_page_addr) >> PAGE_SHIFT),
770 		    &sc->xbd_ring_ref[i]);
771 		if (error) {
772 			xenbus_dev_fatal(sc->xbd_dev, error,
773 			    "granting ring_ref(%d)", i);
774 			return (error);
775 		}
776 	}
777 	if (sc->xbd_ring_pages == 1) {
778 		error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev),
779 		    "ring-ref", "%u", sc->xbd_ring_ref[0]);
780 		if (error) {
781 			xenbus_dev_fatal(sc->xbd_dev, error,
782 			    "writing %s/ring-ref",
783 			    xenbus_get_node(sc->xbd_dev));
784 			return (error);
785 		}
786 	} else {
787 		for (i = 0; i < sc->xbd_ring_pages; i++) {
788 			char ring_ref_name[]= "ring_refXX";
789 
790 			snprintf(ring_ref_name, sizeof(ring_ref_name),
791 			    "ring-ref%u", i);
792 			error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev),
793 			     ring_ref_name, "%u", sc->xbd_ring_ref[i]);
794 			if (error) {
795 				xenbus_dev_fatal(sc->xbd_dev, error,
796 				    "writing %s/%s",
797 				    xenbus_get_node(sc->xbd_dev),
798 				    ring_ref_name);
799 				return (error);
800 			}
801 		}
802 	}
803 
804 	error = xen_intr_alloc_and_bind_local_port(sc->xbd_dev,
805 	    xenbus_get_otherend_id(sc->xbd_dev), NULL, xbd_int, sc,
806 	    INTR_TYPE_BIO | INTR_MPSAFE, &sc->xen_intr_handle);
807 	if (error) {
808 		xenbus_dev_fatal(sc->xbd_dev, error,
809 		    "xen_intr_alloc_and_bind_local_port failed");
810 		return (error);
811 	}
812 
813 	return (0);
814 }
815 
816 static void
817 xbd_free_ring(struct xbd_softc *sc)
818 {
819 	int i;
820 
821 	if (sc->xbd_ring.sring == NULL)
822 		return;
823 
824 	for (i = 0; i < sc->xbd_ring_pages; i++) {
825 		if (sc->xbd_ring_ref[i] != GRANT_REF_INVALID) {
826 			gnttab_end_foreign_access_ref(sc->xbd_ring_ref[i]);
827 			sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
828 		}
829 	}
830 	free(sc->xbd_ring.sring, M_XENBLOCKFRONT);
831 	sc->xbd_ring.sring = NULL;
832 }
833 
834 /*-------------------------- Initialization/Teardown -------------------------*/
835 static int
836 xbd_feature_string(struct xbd_softc *sc, char *features, size_t len)
837 {
838 	struct sbuf sb;
839 	int feature_cnt;
840 
841 	sbuf_new(&sb, features, len, SBUF_FIXEDLEN);
842 
843 	feature_cnt = 0;
844 	if ((sc->xbd_flags & XBDF_FLUSH) != 0) {
845 		sbuf_printf(&sb, "flush");
846 		feature_cnt++;
847 	}
848 
849 	if ((sc->xbd_flags & XBDF_BARRIER) != 0) {
850 		if (feature_cnt != 0)
851 			sbuf_printf(&sb, ", ");
852 		sbuf_printf(&sb, "write_barrier");
853 		feature_cnt++;
854 	}
855 
856 	if ((sc->xbd_flags & XBDF_DISCARD) != 0) {
857 		if (feature_cnt != 0)
858 			sbuf_printf(&sb, ", ");
859 		sbuf_printf(&sb, "discard");
860 		feature_cnt++;
861 	}
862 
863 	if ((sc->xbd_flags & XBDF_PERSISTENT) != 0) {
864 		if (feature_cnt != 0)
865 			sbuf_printf(&sb, ", ");
866 		sbuf_printf(&sb, "persistent_grants");
867 		feature_cnt++;
868 	}
869 
870 	(void) sbuf_finish(&sb);
871 	return (sbuf_len(&sb));
872 }
873 
874 static int
875 xbd_sysctl_features(SYSCTL_HANDLER_ARGS)
876 {
877 	char features[80];
878 	struct xbd_softc *sc = arg1;
879 	int error;
880 	int len;
881 
882 	error = sysctl_wire_old_buffer(req, 0);
883 	if (error != 0)
884 		return (error);
885 
886 	len = xbd_feature_string(sc, features, sizeof(features));
887 
888 	/* len is -1 on error, which will make the SYSCTL_OUT a no-op. */
889 	return (SYSCTL_OUT(req, features, len + 1/*NUL*/));
890 }
891 
892 static void
893 xbd_setup_sysctl(struct xbd_softc *xbd)
894 {
895 	struct sysctl_ctx_list *sysctl_ctx = NULL;
896 	struct sysctl_oid *sysctl_tree = NULL;
897 	struct sysctl_oid_list *children;
898 
899 	sysctl_ctx = device_get_sysctl_ctx(xbd->xbd_dev);
900 	if (sysctl_ctx == NULL)
901 		return;
902 
903 	sysctl_tree = device_get_sysctl_tree(xbd->xbd_dev);
904 	if (sysctl_tree == NULL)
905 		return;
906 
907 	children = SYSCTL_CHILDREN(sysctl_tree);
908 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
909 	    "max_requests", CTLFLAG_RD, &xbd->xbd_max_requests, -1,
910 	    "maximum outstanding requests (negotiated)");
911 
912 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
913 	    "max_request_segments", CTLFLAG_RD,
914 	    &xbd->xbd_max_request_segments, 0,
915 	    "maximum number of pages per requests (negotiated)");
916 
917 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
918 	    "max_request_size", CTLFLAG_RD, &xbd->xbd_max_request_size, 0,
919 	    "maximum size in bytes of a request (negotiated)");
920 
921 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
922 	    "ring_pages", CTLFLAG_RD, &xbd->xbd_ring_pages, 0,
923 	    "communication channel pages (negotiated)");
924 
925 	SYSCTL_ADD_PROC(sysctl_ctx, children, OID_AUTO,
926 	    "features", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, xbd,
927 	    0, xbd_sysctl_features, "A", "protocol features (negotiated)");
928 }
929 
930 /*
931  * Translate Linux major/minor to an appropriate name and unit
932  * number. For HVM guests, this allows us to use the same drive names
933  * with blkfront as the emulated drives, easing transition slightly.
934  */
935 static void
936 xbd_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name)
937 {
938 	static struct vdev_info {
939 		int major;
940 		int shift;
941 		int base;
942 		const char *name;
943 	} info[] = {
944 		{3,	6,	0,	"ada"},	/* ide0 */
945 		{22,	6,	2,	"ada"},	/* ide1 */
946 		{33,	6,	4,	"ada"},	/* ide2 */
947 		{34,	6,	6,	"ada"},	/* ide3 */
948 		{56,	6,	8,	"ada"},	/* ide4 */
949 		{57,	6,	10,	"ada"},	/* ide5 */
950 		{88,	6,	12,	"ada"},	/* ide6 */
951 		{89,	6,	14,	"ada"},	/* ide7 */
952 		{90,	6,	16,	"ada"},	/* ide8 */
953 		{91,	6,	18,	"ada"},	/* ide9 */
954 
955 		{8,	4,	0,	"da"},	/* scsi disk0 */
956 		{65,	4,	16,	"da"},	/* scsi disk1 */
957 		{66,	4,	32,	"da"},	/* scsi disk2 */
958 		{67,	4,	48,	"da"},	/* scsi disk3 */
959 		{68,	4,	64,	"da"},	/* scsi disk4 */
960 		{69,	4,	80,	"da"},	/* scsi disk5 */
961 		{70,	4,	96,	"da"},	/* scsi disk6 */
962 		{71,	4,	112,	"da"},	/* scsi disk7 */
963 		{128,	4,	128,	"da"},	/* scsi disk8 */
964 		{129,	4,	144,	"da"},	/* scsi disk9 */
965 		{130,	4,	160,	"da"},	/* scsi disk10 */
966 		{131,	4,	176,	"da"},	/* scsi disk11 */
967 		{132,	4,	192,	"da"},	/* scsi disk12 */
968 		{133,	4,	208,	"da"},	/* scsi disk13 */
969 		{134,	4,	224,	"da"},	/* scsi disk14 */
970 		{135,	4,	240,	"da"},	/* scsi disk15 */
971 
972 		{202,	4,	0,	"xbd"},	/* xbd */
973 
974 		{0,	0,	0,	NULL},
975 	};
976 	int major = vdevice >> 8;
977 	int minor = vdevice & 0xff;
978 	int i;
979 
980 	if (vdevice & (1 << 28)) {
981 		*unit = (vdevice & ((1 << 28) - 1)) >> 8;
982 		*name = "xbd";
983 		return;
984 	}
985 
986 	for (i = 0; info[i].major; i++) {
987 		if (info[i].major == major) {
988 			*unit = info[i].base + (minor >> info[i].shift);
989 			*name = info[i].name;
990 			return;
991 		}
992 	}
993 
994 	*unit = minor >> 4;
995 	*name = "xbd";
996 }
997 
998 int
999 xbd_instance_create(struct xbd_softc *sc, blkif_sector_t sectors,
1000     int vdevice, uint16_t vdisk_info, unsigned long sector_size,
1001     unsigned long phys_sector_size)
1002 {
1003 	char features[80];
1004 	int unit, error = 0;
1005 	const char *name;
1006 
1007 	xbd_vdevice_to_unit(vdevice, &unit, &name);
1008 
1009 	sc->xbd_unit = unit;
1010 
1011 	if (strcmp(name, "xbd") != 0)
1012 		device_printf(sc->xbd_dev, "attaching as %s%d\n", name, unit);
1013 
1014 	if (xbd_feature_string(sc, features, sizeof(features)) > 0) {
1015 		device_printf(sc->xbd_dev, "features: %s\n",
1016 		    features);
1017 	}
1018 
1019 	sc->xbd_disk = disk_alloc();
1020 	sc->xbd_disk->d_unit = sc->xbd_unit;
1021 	sc->xbd_disk->d_open = xbd_open;
1022 	sc->xbd_disk->d_close = xbd_close;
1023 	sc->xbd_disk->d_ioctl = xbd_ioctl;
1024 	sc->xbd_disk->d_strategy = xbd_strategy;
1025 	sc->xbd_disk->d_dump = xbd_dump;
1026 	sc->xbd_disk->d_name = name;
1027 	sc->xbd_disk->d_drv1 = sc;
1028 	sc->xbd_disk->d_sectorsize = sector_size;
1029 	sc->xbd_disk->d_stripesize = phys_sector_size;
1030 	sc->xbd_disk->d_stripeoffset = 0;
1031 
1032 	sc->xbd_disk->d_mediasize = sectors * sector_size;
1033 	sc->xbd_disk->d_maxsize = sc->xbd_max_request_size;
1034 	sc->xbd_disk->d_flags = DISKFLAG_UNMAPPED_BIO;
1035 	if ((sc->xbd_flags & (XBDF_FLUSH|XBDF_BARRIER)) != 0) {
1036 		sc->xbd_disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
1037 		device_printf(sc->xbd_dev,
1038 		    "synchronize cache commands enabled.\n");
1039 	}
1040 	disk_create(sc->xbd_disk, DISK_VERSION);
1041 
1042 	return error;
1043 }
1044 
1045 static void
1046 xbd_free(struct xbd_softc *sc)
1047 {
1048 	int i;
1049 
1050 	/* Prevent new requests being issued until we fix things up. */
1051 	mtx_lock(&sc->xbd_io_lock);
1052 	sc->xbd_state = XBD_STATE_DISCONNECTED;
1053 	mtx_unlock(&sc->xbd_io_lock);
1054 
1055 	/* Free resources associated with old device channel. */
1056 	xbd_free_ring(sc);
1057 	if (sc->xbd_shadow) {
1058 		for (i = 0; i < sc->xbd_max_requests; i++) {
1059 			struct xbd_command *cm;
1060 
1061 			cm = &sc->xbd_shadow[i];
1062 			if (cm->cm_sg_refs != NULL) {
1063 				free(cm->cm_sg_refs, M_XENBLOCKFRONT);
1064 				cm->cm_sg_refs = NULL;
1065 			}
1066 
1067 			if (cm->cm_indirectionpages != NULL) {
1068 				gnttab_end_foreign_access_references(
1069 				    sc->xbd_max_request_indirectpages,
1070 				    &cm->cm_indirectionrefs[0]);
1071 				contigfree(cm->cm_indirectionpages, PAGE_SIZE *
1072 				    sc->xbd_max_request_indirectpages,
1073 				    M_XENBLOCKFRONT);
1074 				cm->cm_indirectionpages = NULL;
1075 			}
1076 
1077 			bus_dmamap_destroy(sc->xbd_io_dmat, cm->cm_map);
1078 		}
1079 		free(sc->xbd_shadow, M_XENBLOCKFRONT);
1080 		sc->xbd_shadow = NULL;
1081 
1082 		bus_dma_tag_destroy(sc->xbd_io_dmat);
1083 
1084 		xbd_initq_cm(sc, XBD_Q_FREE);
1085 		xbd_initq_cm(sc, XBD_Q_READY);
1086 		xbd_initq_cm(sc, XBD_Q_COMPLETE);
1087 	}
1088 
1089 	xen_intr_unbind(&sc->xen_intr_handle);
1090 
1091 }
1092 
1093 /*--------------------------- State Change Handlers --------------------------*/
1094 static void
1095 xbd_initialize(struct xbd_softc *sc)
1096 {
1097 	const char *otherend_path;
1098 	const char *node_path;
1099 	uint32_t max_ring_page_order;
1100 	int error;
1101 
1102 	if (xenbus_get_state(sc->xbd_dev) != XenbusStateInitialising) {
1103 		/* Initialization has already been performed. */
1104 		return;
1105 	}
1106 
1107 	/*
1108 	 * Protocol defaults valid even if negotiation for a
1109 	 * setting fails.
1110 	 */
1111 	max_ring_page_order = 0;
1112 	sc->xbd_ring_pages = 1;
1113 
1114 	/*
1115 	 * Protocol negotiation.
1116 	 *
1117 	 * \note xs_gather() returns on the first encountered error, so
1118 	 *       we must use independent calls in order to guarantee
1119 	 *       we don't miss information in a sparsly populated back-end
1120 	 *       tree.
1121 	 *
1122 	 * \note xs_scanf() does not update variables for unmatched
1123 	 *	 fields.
1124 	 */
1125 	otherend_path = xenbus_get_otherend_path(sc->xbd_dev);
1126 	node_path = xenbus_get_node(sc->xbd_dev);
1127 
1128 	/* Support both backend schemes for relaying ring page limits. */
1129 	(void)xs_scanf(XST_NIL, otherend_path,
1130 	    "max-ring-page-order", NULL, "%" PRIu32,
1131 	    &max_ring_page_order);
1132 	sc->xbd_ring_pages = 1 << max_ring_page_order;
1133 	(void)xs_scanf(XST_NIL, otherend_path,
1134 	    "max-ring-pages", NULL, "%" PRIu32,
1135 	    &sc->xbd_ring_pages);
1136 	if (sc->xbd_ring_pages < 1)
1137 		sc->xbd_ring_pages = 1;
1138 
1139 	if (sc->xbd_ring_pages > XBD_MAX_RING_PAGES) {
1140 		device_printf(sc->xbd_dev,
1141 		    "Back-end specified ring-pages of %u "
1142 		    "limited to front-end limit of %u.\n",
1143 		    sc->xbd_ring_pages, XBD_MAX_RING_PAGES);
1144 		sc->xbd_ring_pages = XBD_MAX_RING_PAGES;
1145 	}
1146 
1147 	if (powerof2(sc->xbd_ring_pages) == 0) {
1148 		uint32_t new_page_limit;
1149 
1150 		new_page_limit = 0x01 << (fls(sc->xbd_ring_pages) - 1);
1151 		device_printf(sc->xbd_dev,
1152 		    "Back-end specified ring-pages of %u "
1153 		    "is not a power of 2. Limited to %u.\n",
1154 		    sc->xbd_ring_pages, new_page_limit);
1155 		sc->xbd_ring_pages = new_page_limit;
1156 	}
1157 
1158 	sc->xbd_max_requests =
1159 	    BLKIF_MAX_RING_REQUESTS(sc->xbd_ring_pages * PAGE_SIZE);
1160 	if (sc->xbd_max_requests > XBD_MAX_REQUESTS) {
1161 		device_printf(sc->xbd_dev,
1162 		    "Back-end specified max_requests of %u "
1163 		    "limited to front-end limit of %zu.\n",
1164 		    sc->xbd_max_requests, XBD_MAX_REQUESTS);
1165 		sc->xbd_max_requests = XBD_MAX_REQUESTS;
1166 	}
1167 
1168 	if (xbd_alloc_ring(sc) != 0)
1169 		return;
1170 
1171 	/* Support both backend schemes for relaying ring page limits. */
1172 	if (sc->xbd_ring_pages > 1) {
1173 		error = xs_printf(XST_NIL, node_path,
1174 		    "num-ring-pages","%u",
1175 		    sc->xbd_ring_pages);
1176 		if (error) {
1177 			xenbus_dev_fatal(sc->xbd_dev, error,
1178 			    "writing %s/num-ring-pages",
1179 			    node_path);
1180 			return;
1181 		}
1182 
1183 		error = xs_printf(XST_NIL, node_path,
1184 		    "ring-page-order", "%u",
1185 		    fls(sc->xbd_ring_pages) - 1);
1186 		if (error) {
1187 			xenbus_dev_fatal(sc->xbd_dev, error,
1188 			    "writing %s/ring-page-order",
1189 			    node_path);
1190 			return;
1191 		}
1192 	}
1193 
1194 	error = xs_printf(XST_NIL, node_path, "event-channel",
1195 	    "%u", xen_intr_port(sc->xen_intr_handle));
1196 	if (error) {
1197 		xenbus_dev_fatal(sc->xbd_dev, error,
1198 		    "writing %s/event-channel",
1199 		    node_path);
1200 		return;
1201 	}
1202 
1203 	error = xs_printf(XST_NIL, node_path, "protocol",
1204 	    "%s", XEN_IO_PROTO_ABI_NATIVE);
1205 	if (error) {
1206 		xenbus_dev_fatal(sc->xbd_dev, error,
1207 		    "writing %s/protocol",
1208 		    node_path);
1209 		return;
1210 	}
1211 
1212 	xenbus_set_state(sc->xbd_dev, XenbusStateInitialised);
1213 }
1214 
1215 /*
1216  * Invoked when the backend is finally 'ready' (and has published
1217  * the details about the physical device - #sectors, size, etc).
1218  */
1219 static void
1220 xbd_connect(struct xbd_softc *sc)
1221 {
1222 	device_t dev = sc->xbd_dev;
1223 	blkif_sector_t sectors;
1224 	unsigned long sector_size, phys_sector_size;
1225 	unsigned int binfo;
1226 	int err, feature_barrier, feature_flush;
1227 	int i, j;
1228 
1229 	DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev));
1230 
1231 	if (sc->xbd_state == XBD_STATE_SUSPENDED) {
1232 		return;
1233 	}
1234 
1235 	if (sc->xbd_state == XBD_STATE_CONNECTED) {
1236 		struct disk *disk;
1237 
1238 		disk = sc->xbd_disk;
1239 		if (disk == NULL) {
1240 			return;
1241 		}
1242 		err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1243 		    "sectors", "%"PRIu64, &sectors, NULL);
1244 		if (err != 0) {
1245 			xenbus_dev_error(dev, err,
1246 			    "reading sectors at %s",
1247 			    xenbus_get_otherend_path(dev));
1248 			return;
1249 		}
1250 		disk->d_mediasize = disk->d_sectorsize * sectors;
1251 		err = disk_resize(disk, M_NOWAIT);
1252 		if (err) {
1253 			xenbus_dev_error(dev, err,
1254 			    "unable to resize disk %s%u",
1255 			    disk->d_name, disk->d_unit);
1256 			return;
1257 		}
1258 		device_printf(sc->xbd_dev,
1259 		    "changed capacity to %jd\n",
1260 		    (intmax_t)disk->d_mediasize);
1261 		return;
1262 	}
1263 
1264 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1265 	    "sectors", "%"PRIu64, &sectors,
1266 	    "info", "%u", &binfo,
1267 	    "sector-size", "%lu", &sector_size,
1268 	    NULL);
1269 	if (err) {
1270 		xenbus_dev_fatal(dev, err,
1271 		    "reading backend fields at %s",
1272 		    xenbus_get_otherend_path(dev));
1273 		return;
1274 	}
1275 	if ((sectors == 0) || (sector_size == 0)) {
1276 		xenbus_dev_fatal(dev, 0,
1277 		    "invalid parameters from %s:"
1278 		    " sectors = %"PRIu64", sector_size = %lu",
1279 		    xenbus_get_otherend_path(dev),
1280 		    sectors, sector_size);
1281 		return;
1282 	}
1283 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1284 	     "physical-sector-size", "%lu", &phys_sector_size,
1285 	     NULL);
1286 	if (err || phys_sector_size <= sector_size)
1287 		phys_sector_size = 0;
1288 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1289 	     "feature-barrier", "%d", &feature_barrier,
1290 	     NULL);
1291 	if (err == 0 && feature_barrier != 0)
1292 		sc->xbd_flags |= XBDF_BARRIER;
1293 
1294 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1295 	     "feature-flush-cache", "%d", &feature_flush,
1296 	     NULL);
1297 	if (err == 0 && feature_flush != 0)
1298 		sc->xbd_flags |= XBDF_FLUSH;
1299 
1300 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1301 	    "feature-max-indirect-segments", "%" PRIu32,
1302 	    &sc->xbd_max_request_segments, NULL);
1303 	if ((err != 0) || (xbd_enable_indirect == 0))
1304 		sc->xbd_max_request_segments = 0;
1305 	if (sc->xbd_max_request_segments > XBD_MAX_INDIRECT_SEGMENTS)
1306 		sc->xbd_max_request_segments = XBD_MAX_INDIRECT_SEGMENTS;
1307 	if (sc->xbd_max_request_segments > XBD_SIZE_TO_SEGS(maxphys))
1308 		sc->xbd_max_request_segments = XBD_SIZE_TO_SEGS(maxphys);
1309 	sc->xbd_max_request_indirectpages =
1310 	    XBD_INDIRECT_SEGS_TO_PAGES(sc->xbd_max_request_segments);
1311 	if (sc->xbd_max_request_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
1312 		sc->xbd_max_request_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1313 	sc->xbd_max_request_size =
1314 	    XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments);
1315 
1316 	/* Allocate datastructures based on negotiated values. */
1317 	err = bus_dma_tag_create(
1318 	    bus_get_dma_tag(sc->xbd_dev),	/* parent */
1319 	    512, PAGE_SIZE,			/* algnmnt, boundary */
1320 	    BUS_SPACE_MAXADDR,			/* lowaddr */
1321 	    BUS_SPACE_MAXADDR,			/* highaddr */
1322 	    NULL, NULL,				/* filter, filterarg */
1323 	    sc->xbd_max_request_size,
1324 	    sc->xbd_max_request_segments,
1325 	    PAGE_SIZE,				/* maxsegsize */
1326 	    BUS_DMA_ALLOCNOW,			/* flags */
1327 	    busdma_lock_mutex,			/* lockfunc */
1328 	    &sc->xbd_io_lock,			/* lockarg */
1329 	    &sc->xbd_io_dmat);
1330 	if (err != 0) {
1331 		xenbus_dev_fatal(sc->xbd_dev, err,
1332 		    "Cannot allocate parent DMA tag\n");
1333 		return;
1334 	}
1335 
1336 	/* Per-transaction data allocation. */
1337 	sc->xbd_shadow = malloc(sizeof(*sc->xbd_shadow) * sc->xbd_max_requests,
1338 	    M_XENBLOCKFRONT, M_NOWAIT|M_ZERO);
1339 	if (sc->xbd_shadow == NULL) {
1340 		bus_dma_tag_destroy(sc->xbd_io_dmat);
1341 		xenbus_dev_fatal(sc->xbd_dev, ENOMEM,
1342 		    "Cannot allocate request structures\n");
1343 		return;
1344 	}
1345 
1346 	for (i = 0; i < sc->xbd_max_requests; i++) {
1347 		struct xbd_command *cm;
1348 		void * indirectpages;
1349 
1350 		cm = &sc->xbd_shadow[i];
1351 		cm->cm_sg_refs = malloc(
1352 		    sizeof(grant_ref_t) * sc->xbd_max_request_segments,
1353 		    M_XENBLOCKFRONT, M_NOWAIT);
1354 		if (cm->cm_sg_refs == NULL)
1355 			break;
1356 		cm->cm_id = i;
1357 		cm->cm_flags = XBDCF_INITIALIZER;
1358 		cm->cm_sc = sc;
1359 		if (bus_dmamap_create(sc->xbd_io_dmat, 0, &cm->cm_map) != 0)
1360 			break;
1361 		if (sc->xbd_max_request_indirectpages > 0) {
1362 			indirectpages = contigmalloc(
1363 			    PAGE_SIZE * sc->xbd_max_request_indirectpages,
1364 			    M_XENBLOCKFRONT, M_ZERO | M_NOWAIT, 0, ~0,
1365 			    PAGE_SIZE, 0);
1366 			if (indirectpages == NULL)
1367 				sc->xbd_max_request_indirectpages = 0;
1368 		} else {
1369 			indirectpages = NULL;
1370 		}
1371 		for (j = 0; j < sc->xbd_max_request_indirectpages; j++) {
1372 			if (gnttab_grant_foreign_access(
1373 			    xenbus_get_otherend_id(sc->xbd_dev),
1374 			    (vtophys(indirectpages) >> PAGE_SHIFT) + j,
1375 			    1 /* grant read-only access */,
1376 			    &cm->cm_indirectionrefs[j]))
1377 				break;
1378 		}
1379 		if (j < sc->xbd_max_request_indirectpages) {
1380 			contigfree(indirectpages,
1381 			    PAGE_SIZE * sc->xbd_max_request_indirectpages,
1382 			    M_XENBLOCKFRONT);
1383 			break;
1384 		}
1385 		cm->cm_indirectionpages = indirectpages;
1386 		xbd_free_command(cm);
1387 	}
1388 
1389 	if (sc->xbd_disk == NULL) {
1390 		device_printf(dev, "%juMB <%s> at %s",
1391 		    (uintmax_t) sectors / (1048576 / sector_size),
1392 		    device_get_desc(dev),
1393 		    xenbus_get_node(dev));
1394 		bus_print_child_footer(device_get_parent(dev), dev);
1395 
1396 		xbd_instance_create(sc, sectors, sc->xbd_vdevice, binfo,
1397 		    sector_size, phys_sector_size);
1398 	}
1399 
1400 	(void)xenbus_set_state(dev, XenbusStateConnected);
1401 
1402 	/* Kick pending requests. */
1403 	mtx_lock(&sc->xbd_io_lock);
1404 	sc->xbd_state = XBD_STATE_CONNECTED;
1405 	xbd_startio(sc);
1406 	sc->xbd_flags |= XBDF_READY;
1407 	mtx_unlock(&sc->xbd_io_lock);
1408 }
1409 
1410 /**
1411  * Handle the change of state of the backend to Closing.  We must delete our
1412  * device-layer structures now, to ensure that writes are flushed through to
1413  * the backend.  Once this is done, we can switch to Closed in
1414  * acknowledgement.
1415  */
1416 static void
1417 xbd_closing(device_t dev)
1418 {
1419 	struct xbd_softc *sc = device_get_softc(dev);
1420 
1421 	xenbus_set_state(dev, XenbusStateClosing);
1422 
1423 	DPRINTK("xbd_closing: %s removed\n", xenbus_get_node(dev));
1424 
1425 	if (sc->xbd_disk != NULL) {
1426 		disk_destroy(sc->xbd_disk);
1427 		sc->xbd_disk = NULL;
1428 	}
1429 
1430 	xenbus_set_state(dev, XenbusStateClosed);
1431 }
1432 
1433 /*---------------------------- NewBus Entrypoints ----------------------------*/
1434 static int
1435 xbd_probe(device_t dev)
1436 {
1437 	if (strcmp(xenbus_get_type(dev), "vbd") != 0)
1438 		return (ENXIO);
1439 
1440 	if (xen_pv_disks_disabled())
1441 		return (ENXIO);
1442 
1443 	if (xen_hvm_domain()) {
1444 		int error;
1445 		char *type;
1446 
1447 		/*
1448 		 * When running in an HVM domain, IDE disk emulation is
1449 		 * disabled early in boot so that native drivers will
1450 		 * not see emulated hardware.  However, CDROM device
1451 		 * emulation cannot be disabled.
1452 		 *
1453 		 * Through use of FreeBSD's vm_guest and xen_hvm_domain()
1454 		 * APIs, we could modify the native CDROM driver to fail its
1455 		 * probe when running under Xen.  Unfortunatlely, the PV
1456 		 * CDROM support in XenServer (up through at least version
1457 		 * 6.2) isn't functional, so we instead rely on the emulated
1458 		 * CDROM instance, and fail to attach the PV one here in
1459 		 * the blkfront driver.
1460 		 */
1461 		error = xs_read(XST_NIL, xenbus_get_node(dev),
1462 		    "device-type", NULL, (void **) &type);
1463 		if (error)
1464 			return (ENXIO);
1465 
1466 		if (strncmp(type, "cdrom", 5) == 0) {
1467 			free(type, M_XENSTORE);
1468 			return (ENXIO);
1469 		}
1470 		free(type, M_XENSTORE);
1471 	}
1472 
1473 	device_set_desc(dev, "Virtual Block Device");
1474 	device_quiet(dev);
1475 	return (0);
1476 }
1477 
1478 /*
1479  * Setup supplies the backend dir, virtual device.  We place an event
1480  * channel and shared frame entries.  We watch backend to wait if it's
1481  * ok.
1482  */
1483 static int
1484 xbd_attach(device_t dev)
1485 {
1486 	struct xbd_softc *sc;
1487 	const char *name;
1488 	uint32_t vdevice;
1489 	int error;
1490 	int i;
1491 	int unit;
1492 
1493 	/* FIXME: Use dynamic device id if this is not set. */
1494 	error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1495 	    "virtual-device", NULL, "%" PRIu32, &vdevice);
1496 	if (error)
1497 		error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1498 		    "virtual-device-ext", NULL, "%" PRIu32, &vdevice);
1499 	if (error) {
1500 		xenbus_dev_fatal(dev, error, "reading virtual-device");
1501 		device_printf(dev, "Couldn't determine virtual device.\n");
1502 		return (error);
1503 	}
1504 
1505 	xbd_vdevice_to_unit(vdevice, &unit, &name);
1506 	if (!strcmp(name, "xbd"))
1507 		device_set_unit(dev, unit);
1508 
1509 	sc = device_get_softc(dev);
1510 	mtx_init(&sc->xbd_io_lock, "blkfront i/o lock", NULL, MTX_DEF);
1511 	xbd_initqs(sc);
1512 	for (i = 0; i < XBD_MAX_RING_PAGES; i++)
1513 		sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
1514 
1515 	sc->xbd_dev = dev;
1516 	sc->xbd_vdevice = vdevice;
1517 	sc->xbd_state = XBD_STATE_DISCONNECTED;
1518 
1519 	xbd_setup_sysctl(sc);
1520 
1521 	/* Wait for backend device to publish its protocol capabilities. */
1522 	xenbus_set_state(dev, XenbusStateInitialising);
1523 
1524 	return (0);
1525 }
1526 
1527 static int
1528 xbd_detach(device_t dev)
1529 {
1530 	struct xbd_softc *sc = device_get_softc(dev);
1531 
1532 	DPRINTK("%s: %s removed\n", __func__, xenbus_get_node(dev));
1533 
1534 	xbd_free(sc);
1535 	mtx_destroy(&sc->xbd_io_lock);
1536 
1537 	return 0;
1538 }
1539 
1540 static int
1541 xbd_suspend(device_t dev)
1542 {
1543 	struct xbd_softc *sc = device_get_softc(dev);
1544 	int retval;
1545 	int saved_state;
1546 
1547 	/* Prevent new requests being issued until we fix things up. */
1548 	mtx_lock(&sc->xbd_io_lock);
1549 	saved_state = sc->xbd_state;
1550 	sc->xbd_state = XBD_STATE_SUSPENDED;
1551 
1552 	/* Wait for outstanding I/O to drain. */
1553 	retval = 0;
1554 	while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
1555 		if (msleep(&sc->xbd_cm_q[XBD_Q_BUSY], &sc->xbd_io_lock,
1556 		    PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) {
1557 			retval = EBUSY;
1558 			break;
1559 		}
1560 	}
1561 	mtx_unlock(&sc->xbd_io_lock);
1562 
1563 	if (retval != 0)
1564 		sc->xbd_state = saved_state;
1565 
1566 	return (retval);
1567 }
1568 
1569 static int
1570 xbd_resume(device_t dev)
1571 {
1572 	struct xbd_softc *sc = device_get_softc(dev);
1573 
1574 	if (xen_suspend_cancelled) {
1575 		sc->xbd_state = XBD_STATE_CONNECTED;
1576 		return (0);
1577 	}
1578 
1579 	DPRINTK("xbd_resume: %s\n", xenbus_get_node(dev));
1580 
1581 	xbd_free(sc);
1582 	xbd_initialize(sc);
1583 	return (0);
1584 }
1585 
1586 /**
1587  * Callback received when the backend's state changes.
1588  */
1589 static void
1590 xbd_backend_changed(device_t dev, XenbusState backend_state)
1591 {
1592 	struct xbd_softc *sc = device_get_softc(dev);
1593 
1594 	DPRINTK("backend_state=%d\n", backend_state);
1595 
1596 	switch (backend_state) {
1597 	case XenbusStateUnknown:
1598 	case XenbusStateInitialising:
1599 	case XenbusStateReconfigured:
1600 	case XenbusStateReconfiguring:
1601 	case XenbusStateClosed:
1602 		break;
1603 
1604 	case XenbusStateInitWait:
1605 	case XenbusStateInitialised:
1606 		xbd_initialize(sc);
1607 		break;
1608 
1609 	case XenbusStateConnected:
1610 		xbd_initialize(sc);
1611 		xbd_connect(sc);
1612 		break;
1613 
1614 	case XenbusStateClosing:
1615 		if (sc->xbd_users > 0) {
1616 			device_printf(dev, "detaching with pending users\n");
1617 			KASSERT(sc->xbd_disk != NULL,
1618 			    ("NULL disk with pending users\n"));
1619 			disk_gone(sc->xbd_disk);
1620 		} else {
1621 			xbd_closing(dev);
1622 		}
1623 		break;
1624 	}
1625 }
1626 
1627 /*---------------------------- NewBus Registration ---------------------------*/
1628 static device_method_t xbd_methods[] = {
1629 	/* Device interface */
1630 	DEVMETHOD(device_probe,         xbd_probe),
1631 	DEVMETHOD(device_attach,        xbd_attach),
1632 	DEVMETHOD(device_detach,        xbd_detach),
1633 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
1634 	DEVMETHOD(device_suspend,       xbd_suspend),
1635 	DEVMETHOD(device_resume,        xbd_resume),
1636 
1637 	/* Xenbus interface */
1638 	DEVMETHOD(xenbus_otherend_changed, xbd_backend_changed),
1639 
1640 	{ 0, 0 }
1641 };
1642 
1643 static driver_t xbd_driver = {
1644 	"xbd",
1645 	xbd_methods,
1646 	sizeof(struct xbd_softc),
1647 };
1648 
1649 DRIVER_MODULE(xbd, xenbusb_front, xbd_driver, 0, 0);
1650