xref: /linux/drivers/scsi/xen-scsifront.c (revision a13926db3e5ea5bbac297f59e9f35061f52892d3)
1 /*
2  * Xen SCSI frontend driver
3  *
4  * Copyright (c) 2008, FUJITSU Limited
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License version 2
8  * as published by the Free Software Foundation; or, when distributed
9  * separately from the Linux kernel or incorporated into other
10  * software packages, subject to the following license:
11  *
12  * Permission is hereby granted, free of charge, to any person obtaining a copy
13  * of this source file (the "Software"), to deal in the Software without
14  * restriction, including without limitation the rights to use, copy, modify,
15  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
16  * and to permit persons to whom the Software is furnished to do so, subject to
17  * the following conditions:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28  * IN THE SOFTWARE.
29  */
30 
31 #include <linux/module.h>
32 #include <linux/kernel.h>
33 #include <linux/device.h>
34 #include <linux/wait.h>
35 #include <linux/interrupt.h>
36 #include <linux/mutex.h>
37 #include <linux/spinlock.h>
38 #include <linux/sched.h>
39 #include <linux/blkdev.h>
40 #include <linux/pfn.h>
41 #include <linux/slab.h>
42 #include <linux/bitops.h>
43 
44 #include <scsi/scsi_cmnd.h>
45 #include <scsi/scsi_device.h>
46 #include <scsi/scsi.h>
47 #include <scsi/scsi_host.h>
48 
49 #include <xen/xen.h>
50 #include <xen/xenbus.h>
51 #include <xen/grant_table.h>
52 #include <xen/events.h>
53 #include <xen/page.h>
54 
55 #include <xen/interface/grant_table.h>
56 #include <xen/interface/io/vscsiif.h>
57 #include <xen/interface/io/protocols.h>
58 
59 #include <asm/xen/hypervisor.h>
60 
61 
62 #define GRANT_INVALID_REF	0
63 
64 #define VSCSIFRONT_OP_ADD_LUN	1
65 #define VSCSIFRONT_OP_DEL_LUN	2
66 
67 /* Tuning point. */
68 #define VSCSIIF_DEFAULT_CMD_PER_LUN 10
69 #define VSCSIIF_MAX_TARGET          64
70 #define VSCSIIF_MAX_LUN             255
71 
72 #define VSCSIIF_RING_SIZE	__CONST_RING_SIZE(vscsiif, PAGE_SIZE)
73 #define VSCSIIF_MAX_REQS	VSCSIIF_RING_SIZE
74 
75 #define vscsiif_grants_sg(_sg)	(PFN_UP((_sg) *		\
76 				sizeof(struct scsiif_request_segment)))
77 
78 struct vscsifrnt_shadow {
79 	/* command between backend and frontend */
80 	unsigned char act;
81 	uint16_t rqid;
82 
83 	unsigned int nr_grants;		/* number of grants in gref[] */
84 	struct scsiif_request_segment *sg;	/* scatter/gather elements */
85 
86 	/* Do reset or abort function. */
87 	wait_queue_head_t wq_reset;	/* reset work queue           */
88 	int wait_reset;			/* reset work queue condition */
89 	int32_t rslt_reset;		/* reset response status:     */
90 					/* SUCCESS or FAILED or:      */
91 #define RSLT_RESET_WAITING	0
92 #define RSLT_RESET_ERR		-1
93 
94 	/* Requested struct scsi_cmnd is stored from kernel. */
95 	struct scsi_cmnd *sc;
96 	int gref[vscsiif_grants_sg(SG_ALL) + SG_ALL];
97 };
98 
99 struct vscsifrnt_info {
100 	struct xenbus_device *dev;
101 
102 	struct Scsi_Host *host;
103 	int host_active;
104 
105 	unsigned int evtchn;
106 	unsigned int irq;
107 
108 	grant_ref_t ring_ref;
109 	struct vscsiif_front_ring ring;
110 	struct vscsiif_response	ring_rsp;
111 
112 	spinlock_t shadow_lock;
113 	DECLARE_BITMAP(shadow_free_bitmap, VSCSIIF_MAX_REQS);
114 	struct vscsifrnt_shadow *shadow[VSCSIIF_MAX_REQS];
115 
116 	wait_queue_head_t wq_sync;
117 	unsigned int wait_ring_available:1;
118 
119 	char dev_state_path[64];
120 	struct task_struct *curr;
121 };
122 
123 static DEFINE_MUTEX(scsifront_mutex);
124 
125 static void scsifront_wake_up(struct vscsifrnt_info *info)
126 {
127 	info->wait_ring_available = 0;
128 	wake_up(&info->wq_sync);
129 }
130 
131 static int scsifront_get_rqid(struct vscsifrnt_info *info)
132 {
133 	unsigned long flags;
134 	int free;
135 
136 	spin_lock_irqsave(&info->shadow_lock, flags);
137 
138 	free = find_first_bit(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
139 	__clear_bit(free, info->shadow_free_bitmap);
140 
141 	spin_unlock_irqrestore(&info->shadow_lock, flags);
142 
143 	return free;
144 }
145 
146 static int _scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
147 {
148 	int empty = bitmap_empty(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
149 
150 	__set_bit(id, info->shadow_free_bitmap);
151 	info->shadow[id] = NULL;
152 
153 	return empty || info->wait_ring_available;
154 }
155 
156 static void scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
157 {
158 	unsigned long flags;
159 	int kick;
160 
161 	spin_lock_irqsave(&info->shadow_lock, flags);
162 	kick = _scsifront_put_rqid(info, id);
163 	spin_unlock_irqrestore(&info->shadow_lock, flags);
164 
165 	if (kick)
166 		scsifront_wake_up(info);
167 }
168 
169 static struct vscsiif_request *scsifront_pre_req(struct vscsifrnt_info *info)
170 {
171 	struct vscsiif_front_ring *ring = &(info->ring);
172 	struct vscsiif_request *ring_req;
173 	uint32_t id;
174 
175 	id = scsifront_get_rqid(info);	/* use id in response */
176 	if (id >= VSCSIIF_MAX_REQS)
177 		return NULL;
178 
179 	ring_req = RING_GET_REQUEST(&(info->ring), ring->req_prod_pvt);
180 
181 	ring->req_prod_pvt++;
182 
183 	ring_req->rqid = (uint16_t)id;
184 
185 	return ring_req;
186 }
187 
188 static void scsifront_do_request(struct vscsifrnt_info *info)
189 {
190 	struct vscsiif_front_ring *ring = &(info->ring);
191 	int notify;
192 
193 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(ring, notify);
194 	if (notify)
195 		notify_remote_via_irq(info->irq);
196 }
197 
198 static void scsifront_gnttab_done(struct vscsifrnt_info *info, uint32_t id)
199 {
200 	struct vscsifrnt_shadow *s = info->shadow[id];
201 	int i;
202 
203 	if (s->sc->sc_data_direction == DMA_NONE)
204 		return;
205 
206 	for (i = 0; i < s->nr_grants; i++) {
207 		if (unlikely(gnttab_query_foreign_access(s->gref[i]) != 0)) {
208 			shost_printk(KERN_ALERT, info->host, KBUILD_MODNAME
209 				     "grant still in use by backend\n");
210 			BUG();
211 		}
212 		gnttab_end_foreign_access(s->gref[i], 0, 0UL);
213 	}
214 
215 	kfree(s->sg);
216 }
217 
218 static void scsifront_cdb_cmd_done(struct vscsifrnt_info *info,
219 				   struct vscsiif_response *ring_rsp)
220 {
221 	struct scsi_cmnd *sc;
222 	uint32_t id;
223 	uint8_t sense_len;
224 
225 	id = ring_rsp->rqid;
226 	sc = info->shadow[id]->sc;
227 
228 	BUG_ON(sc == NULL);
229 
230 	scsifront_gnttab_done(info, id);
231 	scsifront_put_rqid(info, id);
232 
233 	sc->result = ring_rsp->rslt;
234 	scsi_set_resid(sc, ring_rsp->residual_len);
235 
236 	sense_len = min_t(uint8_t, VSCSIIF_SENSE_BUFFERSIZE,
237 			  ring_rsp->sense_len);
238 
239 	if (sense_len)
240 		memcpy(sc->sense_buffer, ring_rsp->sense_buffer, sense_len);
241 
242 	sc->scsi_done(sc);
243 }
244 
245 static void scsifront_sync_cmd_done(struct vscsifrnt_info *info,
246 				    struct vscsiif_response *ring_rsp)
247 {
248 	uint16_t id = ring_rsp->rqid;
249 	unsigned long flags;
250 	struct vscsifrnt_shadow *shadow = info->shadow[id];
251 	int kick;
252 
253 	spin_lock_irqsave(&info->shadow_lock, flags);
254 	shadow->wait_reset = 1;
255 	switch (shadow->rslt_reset) {
256 	case RSLT_RESET_WAITING:
257 		shadow->rslt_reset = ring_rsp->rslt;
258 		break;
259 	case RSLT_RESET_ERR:
260 		kick = _scsifront_put_rqid(info, id);
261 		spin_unlock_irqrestore(&info->shadow_lock, flags);
262 		kfree(shadow);
263 		if (kick)
264 			scsifront_wake_up(info);
265 		return;
266 	default:
267 		shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
268 			     "bad reset state %d, possibly leaking %u\n",
269 			     shadow->rslt_reset, id);
270 		break;
271 	}
272 	spin_unlock_irqrestore(&info->shadow_lock, flags);
273 
274 	wake_up(&shadow->wq_reset);
275 }
276 
277 static int scsifront_cmd_done(struct vscsifrnt_info *info)
278 {
279 	struct vscsiif_response *ring_rsp;
280 	RING_IDX i, rp;
281 	int more_to_do = 0;
282 	unsigned long flags;
283 
284 	spin_lock_irqsave(info->host->host_lock, flags);
285 
286 	rp = info->ring.sring->rsp_prod;
287 	rmb();	/* ordering required respective to dom0 */
288 	for (i = info->ring.rsp_cons; i != rp; i++) {
289 
290 		ring_rsp = RING_GET_RESPONSE(&info->ring, i);
291 
292 		if (WARN(ring_rsp->rqid >= VSCSIIF_MAX_REQS ||
293 			 test_bit(ring_rsp->rqid, info->shadow_free_bitmap),
294 			 "illegal rqid %u returned by backend!\n",
295 			 ring_rsp->rqid))
296 			continue;
297 
298 		if (info->shadow[ring_rsp->rqid]->act == VSCSIIF_ACT_SCSI_CDB)
299 			scsifront_cdb_cmd_done(info, ring_rsp);
300 		else
301 			scsifront_sync_cmd_done(info, ring_rsp);
302 	}
303 
304 	info->ring.rsp_cons = i;
305 
306 	if (i != info->ring.req_prod_pvt)
307 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
308 	else
309 		info->ring.sring->rsp_event = i + 1;
310 
311 	info->wait_ring_available = 0;
312 
313 	spin_unlock_irqrestore(info->host->host_lock, flags);
314 
315 	wake_up(&info->wq_sync);
316 
317 	return more_to_do;
318 }
319 
320 static irqreturn_t scsifront_irq_fn(int irq, void *dev_id)
321 {
322 	struct vscsifrnt_info *info = dev_id;
323 
324 	while (scsifront_cmd_done(info))
325 		/* Yield point for this unbounded loop. */
326 		cond_resched();
327 
328 	return IRQ_HANDLED;
329 }
330 
331 static int map_data_for_request(struct vscsifrnt_info *info,
332 				struct scsi_cmnd *sc,
333 				struct vscsiif_request *ring_req,
334 				struct vscsifrnt_shadow *shadow)
335 {
336 	grant_ref_t gref_head;
337 	struct page *page;
338 	int err, ref, ref_cnt = 0;
339 	int grant_ro = (sc->sc_data_direction == DMA_TO_DEVICE);
340 	unsigned int i, off, len, bytes;
341 	unsigned int data_len = scsi_bufflen(sc);
342 	unsigned int data_grants = 0, seg_grants = 0;
343 	struct scatterlist *sg;
344 	unsigned long mfn;
345 	struct scsiif_request_segment *seg;
346 
347 	ring_req->nr_segments = 0;
348 	if (sc->sc_data_direction == DMA_NONE || !data_len)
349 		return 0;
350 
351 	scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i)
352 		data_grants += PFN_UP(sg->offset + sg->length);
353 
354 	if (data_grants > VSCSIIF_SG_TABLESIZE) {
355 		if (data_grants > info->host->sg_tablesize) {
356 			shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
357 			     "Unable to map request_buffer for command!\n");
358 			return -E2BIG;
359 		}
360 		seg_grants = vscsiif_grants_sg(data_grants);
361 		shadow->sg = kcalloc(data_grants,
362 			sizeof(struct scsiif_request_segment), GFP_ATOMIC);
363 		if (!shadow->sg)
364 			return -ENOMEM;
365 	}
366 	seg = shadow->sg ? : ring_req->seg;
367 
368 	err = gnttab_alloc_grant_references(seg_grants + data_grants,
369 					    &gref_head);
370 	if (err) {
371 		kfree(shadow->sg);
372 		shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
373 			     "gnttab_alloc_grant_references() error\n");
374 		return -ENOMEM;
375 	}
376 
377 	if (seg_grants) {
378 		page = virt_to_page(seg);
379 		off = (unsigned long)seg & ~PAGE_MASK;
380 		len = sizeof(struct scsiif_request_segment) * data_grants;
381 		while (len > 0) {
382 			bytes = min_t(unsigned int, len, PAGE_SIZE - off);
383 
384 			ref = gnttab_claim_grant_reference(&gref_head);
385 			BUG_ON(ref == -ENOSPC);
386 
387 			mfn = pfn_to_mfn(page_to_pfn(page));
388 			gnttab_grant_foreign_access_ref(ref,
389 				info->dev->otherend_id, mfn, 1);
390 			shadow->gref[ref_cnt] = ref;
391 			ring_req->seg[ref_cnt].gref   = ref;
392 			ring_req->seg[ref_cnt].offset = (uint16_t)off;
393 			ring_req->seg[ref_cnt].length = (uint16_t)bytes;
394 
395 			page++;
396 			len -= bytes;
397 			off = 0;
398 			ref_cnt++;
399 		}
400 		BUG_ON(seg_grants < ref_cnt);
401 		seg_grants = ref_cnt;
402 	}
403 
404 	scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) {
405 		page = sg_page(sg);
406 		off = sg->offset;
407 		len = sg->length;
408 
409 		while (len > 0 && data_len > 0) {
410 			/*
411 			 * sg sends a scatterlist that is larger than
412 			 * the data_len it wants transferred for certain
413 			 * IO sizes.
414 			 */
415 			bytes = min_t(unsigned int, len, PAGE_SIZE - off);
416 			bytes = min(bytes, data_len);
417 
418 			ref = gnttab_claim_grant_reference(&gref_head);
419 			BUG_ON(ref == -ENOSPC);
420 
421 			mfn = pfn_to_mfn(page_to_pfn(page));
422 			gnttab_grant_foreign_access_ref(ref,
423 				info->dev->otherend_id, mfn, grant_ro);
424 
425 			shadow->gref[ref_cnt] = ref;
426 			seg->gref   = ref;
427 			seg->offset = (uint16_t)off;
428 			seg->length = (uint16_t)bytes;
429 
430 			page++;
431 			seg++;
432 			len -= bytes;
433 			data_len -= bytes;
434 			off = 0;
435 			ref_cnt++;
436 		}
437 	}
438 
439 	if (seg_grants)
440 		ring_req->nr_segments = VSCSIIF_SG_GRANT | seg_grants;
441 	else
442 		ring_req->nr_segments = (uint8_t)ref_cnt;
443 	shadow->nr_grants = ref_cnt;
444 
445 	return 0;
446 }
447 
448 static struct vscsiif_request *scsifront_command2ring(
449 		struct vscsifrnt_info *info, struct scsi_cmnd *sc,
450 		struct vscsifrnt_shadow *shadow)
451 {
452 	struct vscsiif_request *ring_req;
453 
454 	memset(shadow, 0, sizeof(*shadow));
455 
456 	ring_req = scsifront_pre_req(info);
457 	if (!ring_req)
458 		return NULL;
459 
460 	info->shadow[ring_req->rqid] = shadow;
461 	shadow->rqid = ring_req->rqid;
462 
463 	ring_req->id      = sc->device->id;
464 	ring_req->lun     = sc->device->lun;
465 	ring_req->channel = sc->device->channel;
466 	ring_req->cmd_len = sc->cmd_len;
467 
468 	BUG_ON(sc->cmd_len > VSCSIIF_MAX_COMMAND_SIZE);
469 
470 	memcpy(ring_req->cmnd, sc->cmnd, sc->cmd_len);
471 
472 	ring_req->sc_data_direction   = (uint8_t)sc->sc_data_direction;
473 	ring_req->timeout_per_command = sc->request->timeout / HZ;
474 
475 	return ring_req;
476 }
477 
478 static int scsifront_queuecommand(struct Scsi_Host *shost,
479 				  struct scsi_cmnd *sc)
480 {
481 	struct vscsifrnt_info *info = shost_priv(shost);
482 	struct vscsiif_request *ring_req;
483 	struct vscsifrnt_shadow *shadow = scsi_cmd_priv(sc);
484 	unsigned long flags;
485 	int err;
486 	uint16_t rqid;
487 
488 	spin_lock_irqsave(shost->host_lock, flags);
489 	if (RING_FULL(&info->ring))
490 		goto busy;
491 
492 	ring_req = scsifront_command2ring(info, sc, shadow);
493 	if (!ring_req)
494 		goto busy;
495 
496 	sc->result = 0;
497 
498 	rqid = ring_req->rqid;
499 	ring_req->act = VSCSIIF_ACT_SCSI_CDB;
500 
501 	shadow->sc  = sc;
502 	shadow->act = VSCSIIF_ACT_SCSI_CDB;
503 
504 	err = map_data_for_request(info, sc, ring_req, shadow);
505 	if (err < 0) {
506 		pr_debug("%s: err %d\n", __func__, err);
507 		scsifront_put_rqid(info, rqid);
508 		spin_unlock_irqrestore(shost->host_lock, flags);
509 		if (err == -ENOMEM)
510 			return SCSI_MLQUEUE_HOST_BUSY;
511 		sc->result = DID_ERROR << 16;
512 		sc->scsi_done(sc);
513 		return 0;
514 	}
515 
516 	scsifront_do_request(info);
517 	spin_unlock_irqrestore(shost->host_lock, flags);
518 
519 	return 0;
520 
521 busy:
522 	spin_unlock_irqrestore(shost->host_lock, flags);
523 	pr_debug("%s: busy\n", __func__);
524 	return SCSI_MLQUEUE_HOST_BUSY;
525 }
526 
527 /*
528  * Any exception handling (reset or abort) must be forwarded to the backend.
529  * We have to wait until an answer is returned. This answer contains the
530  * result to be returned to the requestor.
531  */
532 static int scsifront_action_handler(struct scsi_cmnd *sc, uint8_t act)
533 {
534 	struct Scsi_Host *host = sc->device->host;
535 	struct vscsifrnt_info *info = shost_priv(host);
536 	struct vscsifrnt_shadow *shadow, *s = scsi_cmd_priv(sc);
537 	struct vscsiif_request *ring_req;
538 	int err = 0;
539 
540 	shadow = kmalloc(sizeof(*shadow), GFP_NOIO);
541 	if (!shadow)
542 		return FAILED;
543 
544 	spin_lock_irq(host->host_lock);
545 
546 	for (;;) {
547 		if (!RING_FULL(&info->ring)) {
548 			ring_req = scsifront_command2ring(info, sc, shadow);
549 			if (ring_req)
550 				break;
551 		}
552 		if (err) {
553 			spin_unlock_irq(host->host_lock);
554 			kfree(shadow);
555 			return FAILED;
556 		}
557 		info->wait_ring_available = 1;
558 		spin_unlock_irq(host->host_lock);
559 		err = wait_event_interruptible(info->wq_sync,
560 					       !info->wait_ring_available);
561 		spin_lock_irq(host->host_lock);
562 	}
563 
564 	ring_req->act = act;
565 	ring_req->ref_rqid = s->rqid;
566 
567 	shadow->act = act;
568 	shadow->rslt_reset = RSLT_RESET_WAITING;
569 	init_waitqueue_head(&shadow->wq_reset);
570 
571 	ring_req->nr_segments = 0;
572 
573 	scsifront_do_request(info);
574 
575 	spin_unlock_irq(host->host_lock);
576 	err = wait_event_interruptible(shadow->wq_reset, shadow->wait_reset);
577 	spin_lock_irq(host->host_lock);
578 
579 	if (!err) {
580 		err = shadow->rslt_reset;
581 		scsifront_put_rqid(info, shadow->rqid);
582 		kfree(shadow);
583 	} else {
584 		spin_lock(&info->shadow_lock);
585 		shadow->rslt_reset = RSLT_RESET_ERR;
586 		spin_unlock(&info->shadow_lock);
587 		err = FAILED;
588 	}
589 
590 	spin_unlock_irq(host->host_lock);
591 	return err;
592 }
593 
594 static int scsifront_eh_abort_handler(struct scsi_cmnd *sc)
595 {
596 	pr_debug("%s\n", __func__);
597 	return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_ABORT);
598 }
599 
600 static int scsifront_dev_reset_handler(struct scsi_cmnd *sc)
601 {
602 	pr_debug("%s\n", __func__);
603 	return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_RESET);
604 }
605 
606 static int scsifront_sdev_configure(struct scsi_device *sdev)
607 {
608 	struct vscsifrnt_info *info = shost_priv(sdev->host);
609 
610 	if (info && current == info->curr)
611 		xenbus_printf(XBT_NIL, info->dev->nodename,
612 			      info->dev_state_path, "%d", XenbusStateConnected);
613 
614 	return 0;
615 }
616 
617 static void scsifront_sdev_destroy(struct scsi_device *sdev)
618 {
619 	struct vscsifrnt_info *info = shost_priv(sdev->host);
620 
621 	if (info && current == info->curr)
622 		xenbus_printf(XBT_NIL, info->dev->nodename,
623 			      info->dev_state_path, "%d", XenbusStateClosed);
624 }
625 
626 static struct scsi_host_template scsifront_sht = {
627 	.module			= THIS_MODULE,
628 	.name			= "Xen SCSI frontend driver",
629 	.queuecommand		= scsifront_queuecommand,
630 	.eh_abort_handler	= scsifront_eh_abort_handler,
631 	.eh_device_reset_handler = scsifront_dev_reset_handler,
632 	.slave_configure	= scsifront_sdev_configure,
633 	.slave_destroy		= scsifront_sdev_destroy,
634 	.cmd_per_lun		= VSCSIIF_DEFAULT_CMD_PER_LUN,
635 	.can_queue		= VSCSIIF_MAX_REQS,
636 	.this_id		= -1,
637 	.cmd_size		= sizeof(struct vscsifrnt_shadow),
638 	.sg_tablesize		= VSCSIIF_SG_TABLESIZE,
639 	.use_clustering		= DISABLE_CLUSTERING,
640 	.proc_name		= "scsifront",
641 };
642 
643 static int scsifront_alloc_ring(struct vscsifrnt_info *info)
644 {
645 	struct xenbus_device *dev = info->dev;
646 	struct vscsiif_sring *sring;
647 	int err = -ENOMEM;
648 
649 	/***** Frontend to Backend ring start *****/
650 	sring = (struct vscsiif_sring *)__get_free_page(GFP_KERNEL);
651 	if (!sring) {
652 		xenbus_dev_fatal(dev, err,
653 			"fail to allocate shared ring (Front to Back)");
654 		return err;
655 	}
656 	SHARED_RING_INIT(sring);
657 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
658 
659 	err = xenbus_grant_ring(dev, virt_to_mfn(sring));
660 	if (err < 0) {
661 		free_page((unsigned long)sring);
662 		xenbus_dev_fatal(dev, err,
663 			"fail to grant shared ring (Front to Back)");
664 		return err;
665 	}
666 	info->ring_ref = err;
667 
668 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
669 	if (err) {
670 		xenbus_dev_fatal(dev, err, "xenbus_alloc_evtchn");
671 		goto free_gnttab;
672 	}
673 
674 	err = bind_evtchn_to_irq(info->evtchn);
675 	if (err <= 0) {
676 		xenbus_dev_fatal(dev, err, "bind_evtchn_to_irq");
677 		goto free_gnttab;
678 	}
679 
680 	info->irq = err;
681 
682 	err = request_threaded_irq(info->irq, NULL, scsifront_irq_fn,
683 				   IRQF_ONESHOT, "scsifront", info);
684 	if (err) {
685 		xenbus_dev_fatal(dev, err, "request_threaded_irq");
686 		goto free_irq;
687 	}
688 
689 	return 0;
690 
691 /* free resource */
692 free_irq:
693 	unbind_from_irqhandler(info->irq, info);
694 free_gnttab:
695 	gnttab_end_foreign_access(info->ring_ref, 0,
696 				  (unsigned long)info->ring.sring);
697 
698 	return err;
699 }
700 
701 static int scsifront_init_ring(struct vscsifrnt_info *info)
702 {
703 	struct xenbus_device *dev = info->dev;
704 	struct xenbus_transaction xbt;
705 	int err;
706 
707 	pr_debug("%s\n", __func__);
708 
709 	err = scsifront_alloc_ring(info);
710 	if (err)
711 		return err;
712 	pr_debug("%s: %u %u\n", __func__, info->ring_ref, info->evtchn);
713 
714 again:
715 	err = xenbus_transaction_start(&xbt);
716 	if (err)
717 		xenbus_dev_fatal(dev, err, "starting transaction");
718 
719 	err = xenbus_printf(xbt, dev->nodename, "ring-ref", "%u",
720 			    info->ring_ref);
721 	if (err) {
722 		xenbus_dev_fatal(dev, err, "%s", "writing ring-ref");
723 		goto fail;
724 	}
725 
726 	err = xenbus_printf(xbt, dev->nodename, "event-channel", "%u",
727 			    info->evtchn);
728 
729 	if (err) {
730 		xenbus_dev_fatal(dev, err, "%s", "writing event-channel");
731 		goto fail;
732 	}
733 
734 	err = xenbus_transaction_end(xbt, 0);
735 	if (err) {
736 		if (err == -EAGAIN)
737 			goto again;
738 		xenbus_dev_fatal(dev, err, "completing transaction");
739 		goto free_sring;
740 	}
741 
742 	return 0;
743 
744 fail:
745 	xenbus_transaction_end(xbt, 1);
746 free_sring:
747 	unbind_from_irqhandler(info->irq, info);
748 	gnttab_end_foreign_access(info->ring_ref, 0,
749 				  (unsigned long)info->ring.sring);
750 
751 	return err;
752 }
753 
754 
755 static int scsifront_probe(struct xenbus_device *dev,
756 			   const struct xenbus_device_id *id)
757 {
758 	struct vscsifrnt_info *info;
759 	struct Scsi_Host *host;
760 	int err = -ENOMEM;
761 	char name[TASK_COMM_LEN];
762 
763 	host = scsi_host_alloc(&scsifront_sht, sizeof(*info));
764 	if (!host) {
765 		xenbus_dev_fatal(dev, err, "fail to allocate scsi host");
766 		return err;
767 	}
768 	info = (struct vscsifrnt_info *)host->hostdata;
769 
770 	dev_set_drvdata(&dev->dev, info);
771 	info->dev = dev;
772 
773 	bitmap_fill(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
774 
775 	err = scsifront_init_ring(info);
776 	if (err) {
777 		scsi_host_put(host);
778 		return err;
779 	}
780 
781 	init_waitqueue_head(&info->wq_sync);
782 	spin_lock_init(&info->shadow_lock);
783 
784 	snprintf(name, TASK_COMM_LEN, "vscsiif.%d", host->host_no);
785 
786 	host->max_id      = VSCSIIF_MAX_TARGET;
787 	host->max_channel = 0;
788 	host->max_lun     = VSCSIIF_MAX_LUN;
789 	host->max_sectors = (host->sg_tablesize - 1) * PAGE_SIZE / 512;
790 	host->max_cmd_len = VSCSIIF_MAX_COMMAND_SIZE;
791 
792 	err = scsi_add_host(host, &dev->dev);
793 	if (err) {
794 		dev_err(&dev->dev, "fail to add scsi host %d\n", err);
795 		goto free_sring;
796 	}
797 	info->host = host;
798 	info->host_active = 1;
799 
800 	xenbus_switch_state(dev, XenbusStateInitialised);
801 
802 	return 0;
803 
804 free_sring:
805 	unbind_from_irqhandler(info->irq, info);
806 	gnttab_end_foreign_access(info->ring_ref, 0,
807 				  (unsigned long)info->ring.sring);
808 	scsi_host_put(host);
809 	return err;
810 }
811 
812 static int scsifront_remove(struct xenbus_device *dev)
813 {
814 	struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
815 
816 	pr_debug("%s: %s removed\n", __func__, dev->nodename);
817 
818 	mutex_lock(&scsifront_mutex);
819 	if (info->host_active) {
820 		/* Scsi_host not yet removed */
821 		scsi_remove_host(info->host);
822 		info->host_active = 0;
823 	}
824 	mutex_unlock(&scsifront_mutex);
825 
826 	gnttab_end_foreign_access(info->ring_ref, 0,
827 				  (unsigned long)info->ring.sring);
828 	unbind_from_irqhandler(info->irq, info);
829 
830 	scsi_host_put(info->host);
831 
832 	return 0;
833 }
834 
835 static void scsifront_disconnect(struct vscsifrnt_info *info)
836 {
837 	struct xenbus_device *dev = info->dev;
838 	struct Scsi_Host *host = info->host;
839 
840 	pr_debug("%s: %s disconnect\n", __func__, dev->nodename);
841 
842 	/*
843 	 * When this function is executed, all devices of
844 	 * Frontend have been deleted.
845 	 * Therefore, it need not block I/O before remove_host.
846 	 */
847 
848 	mutex_lock(&scsifront_mutex);
849 	if (info->host_active) {
850 		scsi_remove_host(host);
851 		info->host_active = 0;
852 	}
853 	mutex_unlock(&scsifront_mutex);
854 
855 	xenbus_frontend_closed(dev);
856 }
857 
858 static void scsifront_do_lun_hotplug(struct vscsifrnt_info *info, int op)
859 {
860 	struct xenbus_device *dev = info->dev;
861 	int i, err = 0;
862 	char str[64];
863 	char **dir;
864 	unsigned int dir_n = 0;
865 	unsigned int device_state;
866 	unsigned int hst, chn, tgt, lun;
867 	struct scsi_device *sdev;
868 
869 	dir = xenbus_directory(XBT_NIL, dev->otherend, "vscsi-devs", &dir_n);
870 	if (IS_ERR(dir))
871 		return;
872 
873 	/* mark current task as the one allowed to modify device states */
874 	BUG_ON(info->curr);
875 	info->curr = current;
876 
877 	for (i = 0; i < dir_n; i++) {
878 		/* read status */
879 		snprintf(str, sizeof(str), "vscsi-devs/%s/state", dir[i]);
880 		err = xenbus_scanf(XBT_NIL, dev->otherend, str, "%u",
881 				   &device_state);
882 		if (XENBUS_EXIST_ERR(err))
883 			continue;
884 
885 		/* virtual SCSI device */
886 		snprintf(str, sizeof(str), "vscsi-devs/%s/v-dev", dir[i]);
887 		err = xenbus_scanf(XBT_NIL, dev->otherend, str,
888 				   "%u:%u:%u:%u", &hst, &chn, &tgt, &lun);
889 		if (XENBUS_EXIST_ERR(err))
890 			continue;
891 
892 		/*
893 		 * Front device state path, used in slave_configure called
894 		 * on successfull scsi_add_device, and in slave_destroy called
895 		 * on remove of a device.
896 		 */
897 		snprintf(info->dev_state_path, sizeof(info->dev_state_path),
898 			 "vscsi-devs/%s/state", dir[i]);
899 
900 		switch (op) {
901 		case VSCSIFRONT_OP_ADD_LUN:
902 			if (device_state != XenbusStateInitialised)
903 				break;
904 
905 			if (scsi_add_device(info->host, chn, tgt, lun)) {
906 				dev_err(&dev->dev, "scsi_add_device\n");
907 				xenbus_printf(XBT_NIL, dev->nodename,
908 					      info->dev_state_path,
909 					      "%d", XenbusStateClosed);
910 			}
911 			break;
912 		case VSCSIFRONT_OP_DEL_LUN:
913 			if (device_state != XenbusStateClosing)
914 				break;
915 
916 			sdev = scsi_device_lookup(info->host, chn, tgt, lun);
917 			if (sdev) {
918 				scsi_remove_device(sdev);
919 				scsi_device_put(sdev);
920 			}
921 			break;
922 		default:
923 			break;
924 		}
925 	}
926 
927 	info->curr = NULL;
928 
929 	kfree(dir);
930 }
931 
932 static void scsifront_read_backend_params(struct xenbus_device *dev,
933 					  struct vscsifrnt_info *info)
934 {
935 	unsigned int sg_grant;
936 	int ret;
937 	struct Scsi_Host *host = info->host;
938 
939 	ret = xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg-grant", "%u",
940 			   &sg_grant);
941 	if (ret == 1 && sg_grant) {
942 		sg_grant = min_t(unsigned int, sg_grant, SG_ALL);
943 		sg_grant = max_t(unsigned int, sg_grant, VSCSIIF_SG_TABLESIZE);
944 		host->sg_tablesize = min_t(unsigned int, sg_grant,
945 			VSCSIIF_SG_TABLESIZE * PAGE_SIZE /
946 			sizeof(struct scsiif_request_segment));
947 		host->max_sectors = (host->sg_tablesize - 1) * PAGE_SIZE / 512;
948 	}
949 	dev_info(&dev->dev, "using up to %d SG entries\n", host->sg_tablesize);
950 }
951 
952 static void scsifront_backend_changed(struct xenbus_device *dev,
953 				      enum xenbus_state backend_state)
954 {
955 	struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
956 
957 	pr_debug("%s: %p %u %u\n", __func__, dev, dev->state, backend_state);
958 
959 	switch (backend_state) {
960 	case XenbusStateUnknown:
961 	case XenbusStateInitialising:
962 	case XenbusStateInitWait:
963 	case XenbusStateInitialised:
964 		break;
965 
966 	case XenbusStateConnected:
967 		scsifront_read_backend_params(dev, info);
968 		if (xenbus_read_driver_state(dev->nodename) ==
969 		    XenbusStateInitialised)
970 			scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
971 
972 		if (dev->state != XenbusStateConnected)
973 			xenbus_switch_state(dev, XenbusStateConnected);
974 		break;
975 
976 	case XenbusStateClosed:
977 		if (dev->state == XenbusStateClosed)
978 			break;
979 		/* Missed the backend's Closing state -- fallthrough */
980 	case XenbusStateClosing:
981 		scsifront_disconnect(info);
982 		break;
983 
984 	case XenbusStateReconfiguring:
985 		scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_DEL_LUN);
986 		xenbus_switch_state(dev, XenbusStateReconfiguring);
987 		break;
988 
989 	case XenbusStateReconfigured:
990 		scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
991 		xenbus_switch_state(dev, XenbusStateConnected);
992 		break;
993 	}
994 }
995 
996 static const struct xenbus_device_id scsifront_ids[] = {
997 	{ "vscsi" },
998 	{ "" }
999 };
1000 
1001 static struct xenbus_driver scsifront_driver = {
1002 	.ids			= scsifront_ids,
1003 	.probe			= scsifront_probe,
1004 	.remove			= scsifront_remove,
1005 	.otherend_changed	= scsifront_backend_changed,
1006 };
1007 
1008 static int __init scsifront_init(void)
1009 {
1010 	if (!xen_domain())
1011 		return -ENODEV;
1012 
1013 	return xenbus_register_frontend(&scsifront_driver);
1014 }
1015 module_init(scsifront_init);
1016 
1017 static void __exit scsifront_exit(void)
1018 {
1019 	xenbus_unregister_driver(&scsifront_driver);
1020 }
1021 module_exit(scsifront_exit);
1022 
1023 MODULE_DESCRIPTION("Xen SCSI frontend driver");
1024 MODULE_LICENSE("GPL");
1025 MODULE_ALIAS("xen:vscsi");
1026 MODULE_AUTHOR("Juergen Gross <jgross@suse.com>");
1027