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: stable/12/sys/dev/xen/blkfront/blkfront.c 372349 2022-07-29 18:51:03Z dim $");
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
53 #include <xen/xen-os.h>
54 #include <xen/hypervisor.h>
55 #include <xen/xen_intr.h>
56 #include <xen/gnttab.h>
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/protocols.h>
59 #include <xen/xenbus/xenbusvar.h>
60
61 #include <machine/_inttypes.h>
62
63 #include <geom/geom_disk.h>
64
65 #include <dev/xen/blkfront/block.h>
66
67 #include "xenbus_if.h"
68
69 /*--------------------------- Forward Declarations ---------------------------*/
70 static void xbd_closing(device_t);
71 static void xbd_startio(struct xbd_softc *sc);
72
73 /*---------------------------------- Macros ----------------------------------*/
74 #if 0
75 #define DPRINTK(fmt, args...) printf("[XEN] %s:%d: " fmt ".\n", __func__, __LINE__, ##args)
76 #else
77 #define DPRINTK(fmt, args...)
78 #endif
79
80 #define XBD_SECTOR_SHFT 9
81
82 /*---------------------------- Global Static Data ----------------------------*/
83 static MALLOC_DEFINE(M_XENBLOCKFRONT, "xbd", "Xen Block Front driver data");
84
85 static int xbd_enable_indirect = 1;
86 SYSCTL_NODE(_hw, OID_AUTO, xbd, CTLFLAG_RD, 0, "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
xbd_freeze(struct xbd_softc * sc,xbd_flag_t xbd_flag)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
xbd_thaw(struct xbd_softc * sc,xbd_flag_t xbd_flag)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
xbd_cm_freeze(struct xbd_softc * sc,struct xbd_command * cm,xbdc_flag_t cm_flag)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
xbd_cm_thaw(struct xbd_softc * sc,struct xbd_command * cm)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
xbd_flush_requests(struct xbd_softc * sc)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
xbd_free_command(struct xbd_command * cm)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
xbd_mksegarray(bus_dma_segment_t * segs,int nsegs,grant_ref_t * gref_head,int otherend_id,int readonly,grant_ref_t * sg_ref,struct blkif_request_segment * sg)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
xbd_queue_cb(void * arg,bus_dma_segment_t * segs,int nsegs,int error)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
xbd_queue_request(struct xbd_softc * sc,struct xbd_command * cm)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
xbd_restart_queue_callback(void * arg)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 *
xbd_bio_command(struct xbd_softc * sc)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 panic("unknown bio command %d", bp->bio_cmd);
403 }
404
405 return (cm);
406 }
407
408 /*
409 * Dequeue buffers and place them in the shared communication ring.
410 * Return when no more requests can be accepted or all buffers have
411 * been queued.
412 *
413 * Signal XEN once the ring has been filled out.
414 */
415 static void
xbd_startio(struct xbd_softc * sc)416 xbd_startio(struct xbd_softc *sc)
417 {
418 struct xbd_command *cm;
419 int error, queued = 0;
420
421 mtx_assert(&sc->xbd_io_lock, MA_OWNED);
422
423 if (sc->xbd_state != XBD_STATE_CONNECTED)
424 return;
425
426 while (!RING_FULL(&sc->xbd_ring)) {
427
428 if (sc->xbd_qfrozen_cnt != 0)
429 break;
430
431 cm = xbd_dequeue_cm(sc, XBD_Q_READY);
432
433 if (cm == NULL)
434 cm = xbd_bio_command(sc);
435
436 if (cm == NULL)
437 break;
438
439 if ((cm->cm_flags & XBDCF_Q_FREEZE) != 0) {
440 /*
441 * Single step command. Future work is
442 * held off until this command completes.
443 */
444 xbd_cm_freeze(sc, cm, XBDCF_Q_FREEZE);
445 }
446
447 if ((error = xbd_queue_request(sc, cm)) != 0) {
448 printf("xbd_queue_request returned %d\n", error);
449 break;
450 }
451 queued++;
452 }
453
454 if (queued != 0)
455 xbd_flush_requests(sc);
456 }
457
458 static void
xbd_bio_complete(struct xbd_softc * sc,struct xbd_command * cm)459 xbd_bio_complete(struct xbd_softc *sc, struct xbd_command *cm)
460 {
461 struct bio *bp;
462
463 bp = cm->cm_bp;
464
465 if (__predict_false(cm->cm_status != BLKIF_RSP_OKAY)) {
466 disk_err(bp, "disk error" , -1, 0);
467 printf(" status: %x\n", cm->cm_status);
468 bp->bio_flags |= BIO_ERROR;
469 }
470
471 if (bp->bio_flags & BIO_ERROR)
472 bp->bio_error = EIO;
473 else
474 bp->bio_resid = 0;
475
476 xbd_free_command(cm);
477 biodone(bp);
478 }
479
480 static void
xbd_int(void * xsc)481 xbd_int(void *xsc)
482 {
483 struct xbd_softc *sc = xsc;
484 struct xbd_command *cm;
485 blkif_response_t *bret;
486 RING_IDX i, rp;
487 int op;
488
489 mtx_lock(&sc->xbd_io_lock);
490
491 if (__predict_false(sc->xbd_state == XBD_STATE_DISCONNECTED)) {
492 mtx_unlock(&sc->xbd_io_lock);
493 return;
494 }
495
496 again:
497 rp = sc->xbd_ring.sring->rsp_prod;
498 rmb(); /* Ensure we see queued responses up to 'rp'. */
499
500 for (i = sc->xbd_ring.rsp_cons; i != rp;) {
501 bret = RING_GET_RESPONSE(&sc->xbd_ring, i);
502 cm = &sc->xbd_shadow[bret->id];
503
504 xbd_remove_cm(cm, XBD_Q_BUSY);
505 gnttab_end_foreign_access_references(cm->cm_nseg,
506 cm->cm_sg_refs);
507 i++;
508
509 if (cm->cm_operation == BLKIF_OP_READ)
510 op = BUS_DMASYNC_POSTREAD;
511 else if (cm->cm_operation == BLKIF_OP_WRITE ||
512 cm->cm_operation == BLKIF_OP_WRITE_BARRIER)
513 op = BUS_DMASYNC_POSTWRITE;
514 else
515 op = 0;
516 bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op);
517 bus_dmamap_unload(sc->xbd_io_dmat, cm->cm_map);
518
519 /*
520 * Release any hold this command has on future command
521 * dispatch.
522 */
523 xbd_cm_thaw(sc, cm);
524
525 /*
526 * Directly call the i/o complete routine to save an
527 * an indirection in the common case.
528 */
529 cm->cm_status = bret->status;
530 if (cm->cm_bp)
531 xbd_bio_complete(sc, cm);
532 else if (cm->cm_complete != NULL)
533 cm->cm_complete(cm);
534 else
535 xbd_free_command(cm);
536 }
537
538 sc->xbd_ring.rsp_cons = i;
539
540 if (i != sc->xbd_ring.req_prod_pvt) {
541 int more_to_do;
542 RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, more_to_do);
543 if (more_to_do)
544 goto again;
545 } else {
546 sc->xbd_ring.sring->rsp_event = i + 1;
547 }
548
549 if (xbd_queue_length(sc, XBD_Q_BUSY) == 0)
550 xbd_thaw(sc, XBDF_WAIT_IDLE);
551
552 xbd_startio(sc);
553
554 if (__predict_false(sc->xbd_state == XBD_STATE_SUSPENDED))
555 wakeup(&sc->xbd_cm_q[XBD_Q_BUSY]);
556
557 mtx_unlock(&sc->xbd_io_lock);
558 }
559
560 /*------------------------------- Dump Support -------------------------------*/
561 /**
562 * Quiesce the disk writes for a dump file before allowing the next buffer.
563 */
564 static void
xbd_quiesce(struct xbd_softc * sc)565 xbd_quiesce(struct xbd_softc *sc)
566 {
567 int mtd;
568
569 // While there are outstanding requests
570 while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
571 RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, mtd);
572 if (mtd) {
573 /* Received request completions, update queue. */
574 xbd_int(sc);
575 }
576 if (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
577 /*
578 * Still pending requests, wait for the disk i/o
579 * to complete.
580 */
581 HYPERVISOR_yield();
582 }
583 }
584 }
585
586 /* Kernel dump function for a paravirtualized disk device */
587 static void
xbd_dump_complete(struct xbd_command * cm)588 xbd_dump_complete(struct xbd_command *cm)
589 {
590
591 xbd_enqueue_cm(cm, XBD_Q_COMPLETE);
592 }
593
594 static int
xbd_dump(void * arg,void * virtual,vm_offset_t physical,off_t offset,size_t length)595 xbd_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
596 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 rc = 0;
603
604 if (length == 0)
605 return (0);
606
607 xbd_quiesce(sc); /* All quiet on the western front. */
608
609 /*
610 * If this lock is held, then this module is failing, and a
611 * successful kernel dump is highly unlikely anyway.
612 */
613 mtx_lock(&sc->xbd_io_lock);
614
615 /* Split the 64KB block as needed */
616 while (length > 0) {
617 cm = xbd_dequeue_cm(sc, XBD_Q_FREE);
618 if (cm == NULL) {
619 mtx_unlock(&sc->xbd_io_lock);
620 device_printf(sc->xbd_dev, "dump: no more commands?\n");
621 return (EBUSY);
622 }
623
624 if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
625 &cm->cm_gref_head) != 0) {
626 xbd_free_command(cm);
627 mtx_unlock(&sc->xbd_io_lock);
628 device_printf(sc->xbd_dev, "no more grant allocs?\n");
629 return (EBUSY);
630 }
631
632 chunk = length > sc->xbd_max_request_size ?
633 sc->xbd_max_request_size : length;
634 cm->cm_data = virtual;
635 cm->cm_datalen = chunk;
636 cm->cm_operation = BLKIF_OP_WRITE;
637 cm->cm_sector_number = offset / dp->d_sectorsize;
638 cm->cm_complete = xbd_dump_complete;
639
640 xbd_enqueue_cm(cm, XBD_Q_READY);
641
642 length -= chunk;
643 offset += chunk;
644 virtual = (char *) virtual + chunk;
645 }
646
647 /* Tell DOM0 to do the I/O */
648 xbd_startio(sc);
649 mtx_unlock(&sc->xbd_io_lock);
650
651 /* Poll for the completion. */
652 xbd_quiesce(sc); /* All quite on the eastern front */
653
654 /* If there were any errors, bail out... */
655 while ((cm = xbd_dequeue_cm(sc, XBD_Q_COMPLETE)) != NULL) {
656 if (cm->cm_status != BLKIF_RSP_OKAY) {
657 device_printf(sc->xbd_dev,
658 "Dump I/O failed at sector %jd\n",
659 cm->cm_sector_number);
660 rc = EIO;
661 }
662 xbd_free_command(cm);
663 }
664
665 return (rc);
666 }
667
668 /*----------------------------- Disk Entrypoints -----------------------------*/
669 static int
xbd_open(struct disk * dp)670 xbd_open(struct disk *dp)
671 {
672 struct xbd_softc *sc = dp->d_drv1;
673
674 if (sc == NULL) {
675 printf("xbd%d: not found", dp->d_unit);
676 return (ENXIO);
677 }
678
679 sc->xbd_flags |= XBDF_OPEN;
680 sc->xbd_users++;
681 return (0);
682 }
683
684 static int
xbd_close(struct disk * dp)685 xbd_close(struct disk *dp)
686 {
687 struct xbd_softc *sc = dp->d_drv1;
688
689 if (sc == NULL)
690 return (ENXIO);
691 sc->xbd_flags &= ~XBDF_OPEN;
692 if (--(sc->xbd_users) == 0) {
693 /*
694 * Check whether we have been instructed to close. We will
695 * have ignored this request initially, as the device was
696 * still mounted.
697 */
698 if (xenbus_get_otherend_state(sc->xbd_dev) ==
699 XenbusStateClosing)
700 xbd_closing(sc->xbd_dev);
701 }
702 return (0);
703 }
704
705 static int
xbd_ioctl(struct disk * dp,u_long cmd,void * addr,int flag,struct thread * td)706 xbd_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
707 {
708 struct xbd_softc *sc = dp->d_drv1;
709
710 if (sc == NULL)
711 return (ENXIO);
712
713 return (ENOTTY);
714 }
715
716 /*
717 * Read/write routine for a buffer. Finds the proper unit, place it on
718 * the sortq and kick the controller.
719 */
720 static void
xbd_strategy(struct bio * bp)721 xbd_strategy(struct bio *bp)
722 {
723 struct xbd_softc *sc = bp->bio_disk->d_drv1;
724
725 /* bogus disk? */
726 if (sc == NULL) {
727 bp->bio_error = EINVAL;
728 bp->bio_flags |= BIO_ERROR;
729 bp->bio_resid = bp->bio_bcount;
730 biodone(bp);
731 return;
732 }
733
734 /*
735 * Place it in the queue of disk activities for this disk
736 */
737 mtx_lock(&sc->xbd_io_lock);
738
739 xbd_enqueue_bio(sc, bp);
740 xbd_startio(sc);
741
742 mtx_unlock(&sc->xbd_io_lock);
743 return;
744 }
745
746 /*------------------------------ Ring Management -----------------------------*/
747 static int
xbd_alloc_ring(struct xbd_softc * sc)748 xbd_alloc_ring(struct xbd_softc *sc)
749 {
750 blkif_sring_t *sring;
751 uintptr_t sring_page_addr;
752 int error;
753 int i;
754
755 sring = malloc(sc->xbd_ring_pages * PAGE_SIZE, M_XENBLOCKFRONT,
756 M_NOWAIT|M_ZERO);
757 if (sring == NULL) {
758 xenbus_dev_fatal(sc->xbd_dev, ENOMEM, "allocating shared ring");
759 return (ENOMEM);
760 }
761 SHARED_RING_INIT(sring);
762 FRONT_RING_INIT(&sc->xbd_ring, sring, sc->xbd_ring_pages * PAGE_SIZE);
763
764 for (i = 0, sring_page_addr = (uintptr_t)sring;
765 i < sc->xbd_ring_pages;
766 i++, sring_page_addr += PAGE_SIZE) {
767
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
xbd_free_ring(struct xbd_softc * sc)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
xbd_feature_string(struct xbd_softc * sc,char * features,size_t len)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
xbd_sysctl_features(SYSCTL_HANDLER_ARGS)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
xbd_setup_sysctl(struct xbd_softc * xbd)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, xbd, 0,
927 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
xbd_vdevice_to_unit(uint32_t vdevice,int * unit,const char ** name)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
xbd_instance_create(struct xbd_softc * sc,blkif_sector_t sectors,int vdevice,uint16_t vdisk_info,unsigned long sector_size,unsigned long phys_sector_size)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
xbd_free(struct xbd_softc * sc)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
1059 for (i = 0; i < sc->xbd_max_requests; i++) {
1060 struct xbd_command *cm;
1061
1062 cm = &sc->xbd_shadow[i];
1063 if (cm->cm_sg_refs != NULL) {
1064 free(cm->cm_sg_refs, M_XENBLOCKFRONT);
1065 cm->cm_sg_refs = NULL;
1066 }
1067
1068 if (cm->cm_indirectionpages != NULL) {
1069 gnttab_end_foreign_access_references(
1070 sc->xbd_max_request_indirectpages,
1071 &cm->cm_indirectionrefs[0]);
1072 contigfree(cm->cm_indirectionpages, PAGE_SIZE *
1073 sc->xbd_max_request_indirectpages,
1074 M_XENBLOCKFRONT);
1075 cm->cm_indirectionpages = NULL;
1076 }
1077
1078 bus_dmamap_destroy(sc->xbd_io_dmat, cm->cm_map);
1079 }
1080 free(sc->xbd_shadow, M_XENBLOCKFRONT);
1081 sc->xbd_shadow = NULL;
1082
1083 bus_dma_tag_destroy(sc->xbd_io_dmat);
1084
1085 xbd_initq_cm(sc, XBD_Q_FREE);
1086 xbd_initq_cm(sc, XBD_Q_READY);
1087 xbd_initq_cm(sc, XBD_Q_COMPLETE);
1088 }
1089
1090 xen_intr_unbind(&sc->xen_intr_handle);
1091
1092 }
1093
1094 /*--------------------------- State Change Handlers --------------------------*/
1095 static void
xbd_initialize(struct xbd_softc * sc)1096 xbd_initialize(struct xbd_softc *sc)
1097 {
1098 const char *otherend_path;
1099 const char *node_path;
1100 uint32_t max_ring_page_order;
1101 int error;
1102
1103 if (xenbus_get_state(sc->xbd_dev) != XenbusStateInitialising) {
1104 /* Initialization has already been performed. */
1105 return;
1106 }
1107
1108 /*
1109 * Protocol defaults valid even if negotiation for a
1110 * setting fails.
1111 */
1112 max_ring_page_order = 0;
1113 sc->xbd_ring_pages = 1;
1114
1115 /*
1116 * Protocol negotiation.
1117 *
1118 * \note xs_gather() returns on the first encountered error, so
1119 * we must use independent calls in order to guarantee
1120 * we don't miss information in a sparsly populated back-end
1121 * tree.
1122 *
1123 * \note xs_scanf() does not update variables for unmatched
1124 * fields.
1125 */
1126 otherend_path = xenbus_get_otherend_path(sc->xbd_dev);
1127 node_path = xenbus_get_node(sc->xbd_dev);
1128
1129 /* Support both backend schemes for relaying ring page limits. */
1130 (void)xs_scanf(XST_NIL, otherend_path,
1131 "max-ring-page-order", NULL, "%" PRIu32,
1132 &max_ring_page_order);
1133 sc->xbd_ring_pages = 1 << max_ring_page_order;
1134 (void)xs_scanf(XST_NIL, otherend_path,
1135 "max-ring-pages", NULL, "%" PRIu32,
1136 &sc->xbd_ring_pages);
1137 if (sc->xbd_ring_pages < 1)
1138 sc->xbd_ring_pages = 1;
1139
1140 if (sc->xbd_ring_pages > XBD_MAX_RING_PAGES) {
1141 device_printf(sc->xbd_dev,
1142 "Back-end specified ring-pages of %u "
1143 "limited to front-end limit of %u.\n",
1144 sc->xbd_ring_pages, XBD_MAX_RING_PAGES);
1145 sc->xbd_ring_pages = XBD_MAX_RING_PAGES;
1146 }
1147
1148 if (powerof2(sc->xbd_ring_pages) == 0) {
1149 uint32_t new_page_limit;
1150
1151 new_page_limit = 0x01 << (fls(sc->xbd_ring_pages) - 1);
1152 device_printf(sc->xbd_dev,
1153 "Back-end specified ring-pages of %u "
1154 "is not a power of 2. Limited to %u.\n",
1155 sc->xbd_ring_pages, new_page_limit);
1156 sc->xbd_ring_pages = new_page_limit;
1157 }
1158
1159 sc->xbd_max_requests =
1160 BLKIF_MAX_RING_REQUESTS(sc->xbd_ring_pages * PAGE_SIZE);
1161 if (sc->xbd_max_requests > XBD_MAX_REQUESTS) {
1162 device_printf(sc->xbd_dev,
1163 "Back-end specified max_requests of %u "
1164 "limited to front-end limit of %zu.\n",
1165 sc->xbd_max_requests, XBD_MAX_REQUESTS);
1166 sc->xbd_max_requests = XBD_MAX_REQUESTS;
1167 }
1168
1169 if (xbd_alloc_ring(sc) != 0)
1170 return;
1171
1172 /* Support both backend schemes for relaying ring page limits. */
1173 if (sc->xbd_ring_pages > 1) {
1174 error = xs_printf(XST_NIL, node_path,
1175 "num-ring-pages","%u",
1176 sc->xbd_ring_pages);
1177 if (error) {
1178 xenbus_dev_fatal(sc->xbd_dev, error,
1179 "writing %s/num-ring-pages",
1180 node_path);
1181 return;
1182 }
1183
1184 error = xs_printf(XST_NIL, node_path,
1185 "ring-page-order", "%u",
1186 fls(sc->xbd_ring_pages) - 1);
1187 if (error) {
1188 xenbus_dev_fatal(sc->xbd_dev, error,
1189 "writing %s/ring-page-order",
1190 node_path);
1191 return;
1192 }
1193 }
1194
1195 error = xs_printf(XST_NIL, node_path, "event-channel",
1196 "%u", xen_intr_port(sc->xen_intr_handle));
1197 if (error) {
1198 xenbus_dev_fatal(sc->xbd_dev, error,
1199 "writing %s/event-channel",
1200 node_path);
1201 return;
1202 }
1203
1204 error = xs_printf(XST_NIL, node_path, "protocol",
1205 "%s", XEN_IO_PROTO_ABI_NATIVE);
1206 if (error) {
1207 xenbus_dev_fatal(sc->xbd_dev, error,
1208 "writing %s/protocol",
1209 node_path);
1210 return;
1211 }
1212
1213 xenbus_set_state(sc->xbd_dev, XenbusStateInitialised);
1214 }
1215
1216 /*
1217 * Invoked when the backend is finally 'ready' (and has published
1218 * the details about the physical device - #sectors, size, etc).
1219 */
1220 static void
xbd_connect(struct xbd_softc * sc)1221 xbd_connect(struct xbd_softc *sc)
1222 {
1223 device_t dev = sc->xbd_dev;
1224 unsigned long sectors, sector_size, phys_sector_size;
1225 unsigned int binfo;
1226 int err, feature_barrier, feature_flush;
1227 int i, j;
1228
1229 if (sc->xbd_state == XBD_STATE_CONNECTED ||
1230 sc->xbd_state == XBD_STATE_SUSPENDED)
1231 return;
1232
1233 DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev));
1234
1235 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1236 "sectors", "%lu", §ors,
1237 "info", "%u", &binfo,
1238 "sector-size", "%lu", §or_size,
1239 NULL);
1240 if (err) {
1241 xenbus_dev_fatal(dev, err,
1242 "reading backend fields at %s",
1243 xenbus_get_otherend_path(dev));
1244 return;
1245 }
1246 if ((sectors == 0) || (sector_size == 0)) {
1247 xenbus_dev_fatal(dev, 0,
1248 "invalid parameters from %s:"
1249 " sectors = %lu, sector_size = %lu",
1250 xenbus_get_otherend_path(dev),
1251 sectors, sector_size);
1252 return;
1253 }
1254 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1255 "physical-sector-size", "%lu", &phys_sector_size,
1256 NULL);
1257 if (err || phys_sector_size <= sector_size)
1258 phys_sector_size = 0;
1259 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1260 "feature-barrier", "%d", &feature_barrier,
1261 NULL);
1262 if (err == 0 && feature_barrier != 0)
1263 sc->xbd_flags |= XBDF_BARRIER;
1264
1265 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1266 "feature-flush-cache", "%d", &feature_flush,
1267 NULL);
1268 if (err == 0 && feature_flush != 0)
1269 sc->xbd_flags |= XBDF_FLUSH;
1270
1271 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1272 "feature-max-indirect-segments", "%" PRIu32,
1273 &sc->xbd_max_request_segments, NULL);
1274 if ((err != 0) || (xbd_enable_indirect == 0))
1275 sc->xbd_max_request_segments = 0;
1276 if (sc->xbd_max_request_segments > XBD_MAX_INDIRECT_SEGMENTS)
1277 sc->xbd_max_request_segments = XBD_MAX_INDIRECT_SEGMENTS;
1278 if (sc->xbd_max_request_segments > XBD_SIZE_TO_SEGS(MAXPHYS))
1279 sc->xbd_max_request_segments = XBD_SIZE_TO_SEGS(MAXPHYS);
1280 sc->xbd_max_request_indirectpages =
1281 XBD_INDIRECT_SEGS_TO_PAGES(sc->xbd_max_request_segments);
1282 if (sc->xbd_max_request_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
1283 sc->xbd_max_request_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1284 sc->xbd_max_request_size =
1285 XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments);
1286
1287 /* Allocate datastructures based on negotiated values. */
1288 err = bus_dma_tag_create(
1289 bus_get_dma_tag(sc->xbd_dev), /* parent */
1290 512, PAGE_SIZE, /* algnmnt, boundary */
1291 BUS_SPACE_MAXADDR, /* lowaddr */
1292 BUS_SPACE_MAXADDR, /* highaddr */
1293 NULL, NULL, /* filter, filterarg */
1294 sc->xbd_max_request_size,
1295 sc->xbd_max_request_segments,
1296 PAGE_SIZE, /* maxsegsize */
1297 BUS_DMA_ALLOCNOW, /* flags */
1298 busdma_lock_mutex, /* lockfunc */
1299 &sc->xbd_io_lock, /* lockarg */
1300 &sc->xbd_io_dmat);
1301 if (err != 0) {
1302 xenbus_dev_fatal(sc->xbd_dev, err,
1303 "Cannot allocate parent DMA tag\n");
1304 return;
1305 }
1306
1307 /* Per-transaction data allocation. */
1308 sc->xbd_shadow = malloc(sizeof(*sc->xbd_shadow) * sc->xbd_max_requests,
1309 M_XENBLOCKFRONT, M_NOWAIT|M_ZERO);
1310 if (sc->xbd_shadow == NULL) {
1311 bus_dma_tag_destroy(sc->xbd_io_dmat);
1312 xenbus_dev_fatal(sc->xbd_dev, ENOMEM,
1313 "Cannot allocate request structures\n");
1314 return;
1315 }
1316
1317 for (i = 0; i < sc->xbd_max_requests; i++) {
1318 struct xbd_command *cm;
1319 void * indirectpages;
1320
1321 cm = &sc->xbd_shadow[i];
1322 cm->cm_sg_refs = malloc(
1323 sizeof(grant_ref_t) * sc->xbd_max_request_segments,
1324 M_XENBLOCKFRONT, M_NOWAIT);
1325 if (cm->cm_sg_refs == NULL)
1326 break;
1327 cm->cm_id = i;
1328 cm->cm_flags = XBDCF_INITIALIZER;
1329 cm->cm_sc = sc;
1330 if (bus_dmamap_create(sc->xbd_io_dmat, 0, &cm->cm_map) != 0)
1331 break;
1332 if (sc->xbd_max_request_indirectpages > 0) {
1333 indirectpages = contigmalloc(
1334 PAGE_SIZE * sc->xbd_max_request_indirectpages,
1335 M_XENBLOCKFRONT, M_ZERO | M_NOWAIT, 0, ~0,
1336 PAGE_SIZE, 0);
1337 if (indirectpages == NULL)
1338 sc->xbd_max_request_indirectpages = 0;
1339 } else {
1340 indirectpages = NULL;
1341 }
1342 for (j = 0; j < sc->xbd_max_request_indirectpages; j++) {
1343 if (gnttab_grant_foreign_access(
1344 xenbus_get_otherend_id(sc->xbd_dev),
1345 (vtophys(indirectpages) >> PAGE_SHIFT) + j,
1346 1 /* grant read-only access */,
1347 &cm->cm_indirectionrefs[j]))
1348 break;
1349 }
1350 if (j < sc->xbd_max_request_indirectpages) {
1351 contigfree(indirectpages,
1352 PAGE_SIZE * sc->xbd_max_request_indirectpages,
1353 M_XENBLOCKFRONT);
1354 break;
1355 }
1356 cm->cm_indirectionpages = indirectpages;
1357 xbd_free_command(cm);
1358 }
1359
1360 if (sc->xbd_disk == NULL) {
1361 device_printf(dev, "%juMB <%s> at %s",
1362 (uintmax_t) sectors / (1048576 / sector_size),
1363 device_get_desc(dev),
1364 xenbus_get_node(dev));
1365 bus_print_child_footer(device_get_parent(dev), dev);
1366
1367 xbd_instance_create(sc, sectors, sc->xbd_vdevice, binfo,
1368 sector_size, phys_sector_size);
1369 }
1370
1371 (void)xenbus_set_state(dev, XenbusStateConnected);
1372
1373 /* Kick pending requests. */
1374 mtx_lock(&sc->xbd_io_lock);
1375 sc->xbd_state = XBD_STATE_CONNECTED;
1376 xbd_startio(sc);
1377 sc->xbd_flags |= XBDF_READY;
1378 mtx_unlock(&sc->xbd_io_lock);
1379 }
1380
1381 /**
1382 * Handle the change of state of the backend to Closing. We must delete our
1383 * device-layer structures now, to ensure that writes are flushed through to
1384 * the backend. Once this is done, we can switch to Closed in
1385 * acknowledgement.
1386 */
1387 static void
xbd_closing(device_t dev)1388 xbd_closing(device_t dev)
1389 {
1390 struct xbd_softc *sc = device_get_softc(dev);
1391
1392 xenbus_set_state(dev, XenbusStateClosing);
1393
1394 DPRINTK("xbd_closing: %s removed\n", xenbus_get_node(dev));
1395
1396 if (sc->xbd_disk != NULL) {
1397 disk_destroy(sc->xbd_disk);
1398 sc->xbd_disk = NULL;
1399 }
1400
1401 xenbus_set_state(dev, XenbusStateClosed);
1402 }
1403
1404 /*---------------------------- NewBus Entrypoints ----------------------------*/
1405 static int
xbd_probe(device_t dev)1406 xbd_probe(device_t dev)
1407 {
1408 if (strcmp(xenbus_get_type(dev), "vbd") != 0)
1409 return (ENXIO);
1410
1411 if (xen_hvm_domain() && xen_disable_pv_disks != 0)
1412 return (ENXIO);
1413
1414 if (xen_hvm_domain()) {
1415 int error;
1416 char *type;
1417
1418 /*
1419 * When running in an HVM domain, IDE disk emulation is
1420 * disabled early in boot so that native drivers will
1421 * not see emulated hardware. However, CDROM device
1422 * emulation cannot be disabled.
1423 *
1424 * Through use of FreeBSD's vm_guest and xen_hvm_domain()
1425 * APIs, we could modify the native CDROM driver to fail its
1426 * probe when running under Xen. Unfortunatlely, the PV
1427 * CDROM support in XenServer (up through at least version
1428 * 6.2) isn't functional, so we instead rely on the emulated
1429 * CDROM instance, and fail to attach the PV one here in
1430 * the blkfront driver.
1431 */
1432 error = xs_read(XST_NIL, xenbus_get_node(dev),
1433 "device-type", NULL, (void **) &type);
1434 if (error)
1435 return (ENXIO);
1436
1437 if (strncmp(type, "cdrom", 5) == 0) {
1438 free(type, M_XENSTORE);
1439 return (ENXIO);
1440 }
1441 free(type, M_XENSTORE);
1442 }
1443
1444 device_set_desc(dev, "Virtual Block Device");
1445 device_quiet(dev);
1446 return (0);
1447 }
1448
1449 /*
1450 * Setup supplies the backend dir, virtual device. We place an event
1451 * channel and shared frame entries. We watch backend to wait if it's
1452 * ok.
1453 */
1454 static int
xbd_attach(device_t dev)1455 xbd_attach(device_t dev)
1456 {
1457 struct xbd_softc *sc;
1458 const char *name;
1459 uint32_t vdevice;
1460 int error;
1461 int i;
1462 int unit;
1463
1464 /* FIXME: Use dynamic device id if this is not set. */
1465 error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1466 "virtual-device", NULL, "%" PRIu32, &vdevice);
1467 if (error)
1468 error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1469 "virtual-device-ext", NULL, "%" PRIu32, &vdevice);
1470 if (error) {
1471 xenbus_dev_fatal(dev, error, "reading virtual-device");
1472 device_printf(dev, "Couldn't determine virtual device.\n");
1473 return (error);
1474 }
1475
1476 xbd_vdevice_to_unit(vdevice, &unit, &name);
1477 if (!strcmp(name, "xbd"))
1478 device_set_unit(dev, unit);
1479
1480 sc = device_get_softc(dev);
1481 mtx_init(&sc->xbd_io_lock, "blkfront i/o lock", NULL, MTX_DEF);
1482 xbd_initqs(sc);
1483 for (i = 0; i < XBD_MAX_RING_PAGES; i++)
1484 sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
1485
1486 sc->xbd_dev = dev;
1487 sc->xbd_vdevice = vdevice;
1488 sc->xbd_state = XBD_STATE_DISCONNECTED;
1489
1490 xbd_setup_sysctl(sc);
1491
1492 /* Wait for backend device to publish its protocol capabilities. */
1493 xenbus_set_state(dev, XenbusStateInitialising);
1494
1495 return (0);
1496 }
1497
1498 static int
xbd_detach(device_t dev)1499 xbd_detach(device_t dev)
1500 {
1501 struct xbd_softc *sc = device_get_softc(dev);
1502
1503 DPRINTK("%s: %s removed\n", __func__, xenbus_get_node(dev));
1504
1505 xbd_free(sc);
1506 mtx_destroy(&sc->xbd_io_lock);
1507
1508 return 0;
1509 }
1510
1511 static int
xbd_suspend(device_t dev)1512 xbd_suspend(device_t dev)
1513 {
1514 struct xbd_softc *sc = device_get_softc(dev);
1515 int retval;
1516 int saved_state;
1517
1518 /* Prevent new requests being issued until we fix things up. */
1519 mtx_lock(&sc->xbd_io_lock);
1520 saved_state = sc->xbd_state;
1521 sc->xbd_state = XBD_STATE_SUSPENDED;
1522
1523 /* Wait for outstanding I/O to drain. */
1524 retval = 0;
1525 while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
1526 if (msleep(&sc->xbd_cm_q[XBD_Q_BUSY], &sc->xbd_io_lock,
1527 PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) {
1528 retval = EBUSY;
1529 break;
1530 }
1531 }
1532 mtx_unlock(&sc->xbd_io_lock);
1533
1534 if (retval != 0)
1535 sc->xbd_state = saved_state;
1536
1537 return (retval);
1538 }
1539
1540 static int
xbd_resume(device_t dev)1541 xbd_resume(device_t dev)
1542 {
1543 struct xbd_softc *sc = device_get_softc(dev);
1544
1545 if (xen_suspend_cancelled) {
1546 sc->xbd_state = XBD_STATE_CONNECTED;
1547 return (0);
1548 }
1549
1550 DPRINTK("xbd_resume: %s\n", xenbus_get_node(dev));
1551
1552 xbd_free(sc);
1553 xbd_initialize(sc);
1554 return (0);
1555 }
1556
1557 /**
1558 * Callback received when the backend's state changes.
1559 */
1560 static void
xbd_backend_changed(device_t dev,XenbusState backend_state)1561 xbd_backend_changed(device_t dev, XenbusState backend_state)
1562 {
1563 struct xbd_softc *sc = device_get_softc(dev);
1564
1565 DPRINTK("backend_state=%d\n", backend_state);
1566
1567 switch (backend_state) {
1568 case XenbusStateUnknown:
1569 case XenbusStateInitialising:
1570 case XenbusStateReconfigured:
1571 case XenbusStateReconfiguring:
1572 case XenbusStateClosed:
1573 break;
1574
1575 case XenbusStateInitWait:
1576 case XenbusStateInitialised:
1577 xbd_initialize(sc);
1578 break;
1579
1580 case XenbusStateConnected:
1581 xbd_initialize(sc);
1582 xbd_connect(sc);
1583 break;
1584
1585 case XenbusStateClosing:
1586 if (sc->xbd_users > 0) {
1587 device_printf(dev, "detaching with pending users\n");
1588 KASSERT(sc->xbd_disk != NULL,
1589 ("NULL disk with pending users\n"));
1590 disk_gone(sc->xbd_disk);
1591 } else {
1592 xbd_closing(dev);
1593 }
1594 break;
1595 }
1596 }
1597
1598 /*---------------------------- NewBus Registration ---------------------------*/
1599 static device_method_t xbd_methods[] = {
1600 /* Device interface */
1601 DEVMETHOD(device_probe, xbd_probe),
1602 DEVMETHOD(device_attach, xbd_attach),
1603 DEVMETHOD(device_detach, xbd_detach),
1604 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1605 DEVMETHOD(device_suspend, xbd_suspend),
1606 DEVMETHOD(device_resume, xbd_resume),
1607
1608 /* Xenbus interface */
1609 DEVMETHOD(xenbus_otherend_changed, xbd_backend_changed),
1610
1611 { 0, 0 }
1612 };
1613
1614 static driver_t xbd_driver = {
1615 "xbd",
1616 xbd_methods,
1617 sizeof(struct xbd_softc),
1618 };
1619 devclass_t xbd_devclass;
1620
1621 DRIVER_MODULE(xbd, xenbusb_front, xbd_driver, xbd_devclass, 0, 0);
1622