GNU Linux-libre 5.10.219-gnu1
[releases.git] / drivers / scsi / xen-scsifront.c
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 #define VSCSIFRONT_OP_READD_LUN 3
67
68 /* Tuning point. */
69 #define VSCSIIF_DEFAULT_CMD_PER_LUN 10
70 #define VSCSIIF_MAX_TARGET          64
71 #define VSCSIIF_MAX_LUN             255
72
73 #define VSCSIIF_RING_SIZE       __CONST_RING_SIZE(vscsiif, PAGE_SIZE)
74 #define VSCSIIF_MAX_REQS        VSCSIIF_RING_SIZE
75
76 #define vscsiif_grants_sg(_sg)  (PFN_UP((_sg) *         \
77                                 sizeof(struct scsiif_request_segment)))
78
79 struct vscsifrnt_shadow {
80         /* command between backend and frontend */
81         unsigned char act;
82         uint8_t nr_segments;
83         uint16_t rqid;
84         uint16_t ref_rqid;
85
86         unsigned int nr_grants;         /* number of grants in gref[] */
87         struct scsiif_request_segment *sg;      /* scatter/gather elements */
88         struct scsiif_request_segment seg[VSCSIIF_SG_TABLESIZE];
89
90         /* Do reset or abort function. */
91         wait_queue_head_t wq_reset;     /* reset work queue           */
92         int wait_reset;                 /* reset work queue condition */
93         int32_t rslt_reset;             /* reset response status:     */
94                                         /* SUCCESS or FAILED or:      */
95 #define RSLT_RESET_WAITING      0
96 #define RSLT_RESET_ERR          -1
97
98         /* Requested struct scsi_cmnd is stored from kernel. */
99         struct scsi_cmnd *sc;
100         int gref[vscsiif_grants_sg(SG_ALL) + SG_ALL];
101 };
102
103 struct vscsifrnt_info {
104         struct xenbus_device *dev;
105
106         struct Scsi_Host *host;
107         int host_active;
108
109         unsigned int evtchn;
110         unsigned int irq;
111
112         grant_ref_t ring_ref;
113         struct vscsiif_front_ring ring;
114         struct vscsiif_response ring_rsp;
115
116         spinlock_t shadow_lock;
117         DECLARE_BITMAP(shadow_free_bitmap, VSCSIIF_MAX_REQS);
118         struct vscsifrnt_shadow *shadow[VSCSIIF_MAX_REQS];
119
120         /* Following items are protected by the host lock. */
121         wait_queue_head_t wq_sync;
122         wait_queue_head_t wq_pause;
123         unsigned int wait_ring_available:1;
124         unsigned int waiting_pause:1;
125         unsigned int pause:1;
126         unsigned callers;
127
128         char dev_state_path[64];
129         struct task_struct *curr;
130 };
131
132 static DEFINE_MUTEX(scsifront_mutex);
133
134 static void scsifront_wake_up(struct vscsifrnt_info *info)
135 {
136         info->wait_ring_available = 0;
137         wake_up(&info->wq_sync);
138 }
139
140 static int scsifront_get_rqid(struct vscsifrnt_info *info)
141 {
142         unsigned long flags;
143         int free;
144
145         spin_lock_irqsave(&info->shadow_lock, flags);
146
147         free = find_first_bit(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
148         __clear_bit(free, info->shadow_free_bitmap);
149
150         spin_unlock_irqrestore(&info->shadow_lock, flags);
151
152         return free;
153 }
154
155 static int _scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
156 {
157         int empty = bitmap_empty(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
158
159         __set_bit(id, info->shadow_free_bitmap);
160         info->shadow[id] = NULL;
161
162         return empty || info->wait_ring_available;
163 }
164
165 static void scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
166 {
167         unsigned long flags;
168         int kick;
169
170         spin_lock_irqsave(&info->shadow_lock, flags);
171         kick = _scsifront_put_rqid(info, id);
172         spin_unlock_irqrestore(&info->shadow_lock, flags);
173
174         if (kick)
175                 scsifront_wake_up(info);
176 }
177
178 static int scsifront_do_request(struct vscsifrnt_info *info,
179                                 struct vscsifrnt_shadow *shadow)
180 {
181         struct vscsiif_front_ring *ring = &(info->ring);
182         struct vscsiif_request *ring_req;
183         struct scsi_cmnd *sc = shadow->sc;
184         uint32_t id;
185         int i, notify;
186
187         if (RING_FULL(&info->ring))
188                 return -EBUSY;
189
190         id = scsifront_get_rqid(info);  /* use id in response */
191         if (id >= VSCSIIF_MAX_REQS)
192                 return -EBUSY;
193
194         info->shadow[id] = shadow;
195         shadow->rqid = id;
196
197         ring_req = RING_GET_REQUEST(&(info->ring), ring->req_prod_pvt);
198         ring->req_prod_pvt++;
199
200         ring_req->rqid        = id;
201         ring_req->act         = shadow->act;
202         ring_req->ref_rqid    = shadow->ref_rqid;
203         ring_req->nr_segments = shadow->nr_segments;
204
205         ring_req->id      = sc->device->id;
206         ring_req->lun     = sc->device->lun;
207         ring_req->channel = sc->device->channel;
208         ring_req->cmd_len = sc->cmd_len;
209
210         BUG_ON(sc->cmd_len > VSCSIIF_MAX_COMMAND_SIZE);
211
212         memcpy(ring_req->cmnd, sc->cmnd, sc->cmd_len);
213
214         ring_req->sc_data_direction   = (uint8_t)sc->sc_data_direction;
215         ring_req->timeout_per_command = sc->request->timeout / HZ;
216
217         for (i = 0; i < (shadow->nr_segments & ~VSCSIIF_SG_GRANT); i++)
218                 ring_req->seg[i] = shadow->seg[i];
219
220         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(ring, notify);
221         if (notify)
222                 notify_remote_via_irq(info->irq);
223
224         return 0;
225 }
226
227 static void scsifront_gnttab_done(struct vscsifrnt_info *info,
228                                   struct vscsifrnt_shadow *shadow)
229 {
230         int i;
231
232         if (shadow->sc->sc_data_direction == DMA_NONE)
233                 return;
234
235         for (i = 0; i < shadow->nr_grants; i++) {
236                 if (unlikely(!gnttab_try_end_foreign_access(shadow->gref[i]))) {
237                         shost_printk(KERN_ALERT, info->host, KBUILD_MODNAME
238                                      "grant still in use by backend\n");
239                         BUG();
240                 }
241         }
242
243         kfree(shadow->sg);
244 }
245
246 static void scsifront_cdb_cmd_done(struct vscsifrnt_info *info,
247                                    struct vscsiif_response *ring_rsp)
248 {
249         struct vscsifrnt_shadow *shadow;
250         struct scsi_cmnd *sc;
251         uint32_t id;
252         uint8_t sense_len;
253
254         id = ring_rsp->rqid;
255         shadow = info->shadow[id];
256         sc = shadow->sc;
257
258         BUG_ON(sc == NULL);
259
260         scsifront_gnttab_done(info, shadow);
261         scsifront_put_rqid(info, id);
262
263         sc->result = ring_rsp->rslt;
264         scsi_set_resid(sc, ring_rsp->residual_len);
265
266         sense_len = min_t(uint8_t, VSCSIIF_SENSE_BUFFERSIZE,
267                           ring_rsp->sense_len);
268
269         if (sense_len)
270                 memcpy(sc->sense_buffer, ring_rsp->sense_buffer, sense_len);
271
272         sc->scsi_done(sc);
273 }
274
275 static void scsifront_sync_cmd_done(struct vscsifrnt_info *info,
276                                     struct vscsiif_response *ring_rsp)
277 {
278         uint16_t id = ring_rsp->rqid;
279         unsigned long flags;
280         struct vscsifrnt_shadow *shadow = info->shadow[id];
281         int kick;
282
283         spin_lock_irqsave(&info->shadow_lock, flags);
284         shadow->wait_reset = 1;
285         switch (shadow->rslt_reset) {
286         case RSLT_RESET_WAITING:
287                 shadow->rslt_reset = ring_rsp->rslt;
288                 break;
289         case RSLT_RESET_ERR:
290                 kick = _scsifront_put_rqid(info, id);
291                 spin_unlock_irqrestore(&info->shadow_lock, flags);
292                 kfree(shadow);
293                 if (kick)
294                         scsifront_wake_up(info);
295                 return;
296         default:
297                 shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
298                              "bad reset state %d, possibly leaking %u\n",
299                              shadow->rslt_reset, id);
300                 break;
301         }
302         spin_unlock_irqrestore(&info->shadow_lock, flags);
303
304         wake_up(&shadow->wq_reset);
305 }
306
307 static void scsifront_do_response(struct vscsifrnt_info *info,
308                                   struct vscsiif_response *ring_rsp)
309 {
310         if (WARN(ring_rsp->rqid >= VSCSIIF_MAX_REQS ||
311                  test_bit(ring_rsp->rqid, info->shadow_free_bitmap),
312                  "illegal rqid %u returned by backend!\n", ring_rsp->rqid))
313                 return;
314
315         if (info->shadow[ring_rsp->rqid]->act == VSCSIIF_ACT_SCSI_CDB)
316                 scsifront_cdb_cmd_done(info, ring_rsp);
317         else
318                 scsifront_sync_cmd_done(info, ring_rsp);
319 }
320
321 static int scsifront_ring_drain(struct vscsifrnt_info *info)
322 {
323         struct vscsiif_response *ring_rsp;
324         RING_IDX i, rp;
325         int more_to_do = 0;
326
327         rp = info->ring.sring->rsp_prod;
328         rmb();  /* ordering required respective to dom0 */
329         for (i = info->ring.rsp_cons; i != rp; i++) {
330                 ring_rsp = RING_GET_RESPONSE(&info->ring, i);
331                 scsifront_do_response(info, ring_rsp);
332         }
333
334         info->ring.rsp_cons = i;
335
336         if (i != info->ring.req_prod_pvt)
337                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
338         else
339                 info->ring.sring->rsp_event = i + 1;
340
341         return more_to_do;
342 }
343
344 static int scsifront_cmd_done(struct vscsifrnt_info *info)
345 {
346         int more_to_do;
347         unsigned long flags;
348
349         spin_lock_irqsave(info->host->host_lock, flags);
350
351         more_to_do = scsifront_ring_drain(info);
352
353         info->wait_ring_available = 0;
354
355         spin_unlock_irqrestore(info->host->host_lock, flags);
356
357         wake_up(&info->wq_sync);
358
359         return more_to_do;
360 }
361
362 static irqreturn_t scsifront_irq_fn(int irq, void *dev_id)
363 {
364         struct vscsifrnt_info *info = dev_id;
365
366         while (scsifront_cmd_done(info))
367                 /* Yield point for this unbounded loop. */
368                 cond_resched();
369
370         return IRQ_HANDLED;
371 }
372
373 static void scsifront_finish_all(struct vscsifrnt_info *info)
374 {
375         unsigned i;
376         struct vscsiif_response resp;
377
378         scsifront_ring_drain(info);
379
380         for (i = 0; i < VSCSIIF_MAX_REQS; i++) {
381                 if (test_bit(i, info->shadow_free_bitmap))
382                         continue;
383                 resp.rqid = i;
384                 resp.sense_len = 0;
385                 resp.rslt = DID_RESET << 16;
386                 resp.residual_len = 0;
387                 scsifront_do_response(info, &resp);
388         }
389 }
390
391 static int map_data_for_request(struct vscsifrnt_info *info,
392                                 struct scsi_cmnd *sc,
393                                 struct vscsifrnt_shadow *shadow)
394 {
395         grant_ref_t gref_head;
396         struct page *page;
397         int err, ref, ref_cnt = 0;
398         int grant_ro = (sc->sc_data_direction == DMA_TO_DEVICE);
399         unsigned int i, off, len, bytes;
400         unsigned int data_len = scsi_bufflen(sc);
401         unsigned int data_grants = 0, seg_grants = 0;
402         struct scatterlist *sg;
403         struct scsiif_request_segment *seg;
404
405         if (sc->sc_data_direction == DMA_NONE || !data_len)
406                 return 0;
407
408         scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i)
409                 data_grants += PFN_UP(sg->offset + sg->length);
410
411         if (data_grants > VSCSIIF_SG_TABLESIZE) {
412                 if (data_grants > info->host->sg_tablesize) {
413                         shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
414                              "Unable to map request_buffer for command!\n");
415                         return -E2BIG;
416                 }
417                 seg_grants = vscsiif_grants_sg(data_grants);
418                 shadow->sg = kcalloc(data_grants,
419                         sizeof(struct scsiif_request_segment), GFP_ATOMIC);
420                 if (!shadow->sg)
421                         return -ENOMEM;
422         }
423         seg = shadow->sg ? : shadow->seg;
424
425         err = gnttab_alloc_grant_references(seg_grants + data_grants,
426                                             &gref_head);
427         if (err) {
428                 kfree(shadow->sg);
429                 shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
430                              "gnttab_alloc_grant_references() error\n");
431                 return -ENOMEM;
432         }
433
434         if (seg_grants) {
435                 page = virt_to_page(seg);
436                 off = offset_in_page(seg);
437                 len = sizeof(struct scsiif_request_segment) * data_grants;
438                 while (len > 0) {
439                         bytes = min_t(unsigned int, len, PAGE_SIZE - off);
440
441                         ref = gnttab_claim_grant_reference(&gref_head);
442                         BUG_ON(ref == -ENOSPC);
443
444                         gnttab_grant_foreign_access_ref(ref,
445                                 info->dev->otherend_id,
446                                 xen_page_to_gfn(page), 1);
447                         shadow->gref[ref_cnt] = ref;
448                         shadow->seg[ref_cnt].gref   = ref;
449                         shadow->seg[ref_cnt].offset = (uint16_t)off;
450                         shadow->seg[ref_cnt].length = (uint16_t)bytes;
451
452                         page++;
453                         len -= bytes;
454                         off = 0;
455                         ref_cnt++;
456                 }
457                 BUG_ON(seg_grants < ref_cnt);
458                 seg_grants = ref_cnt;
459         }
460
461         scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) {
462                 page = sg_page(sg);
463                 off = sg->offset;
464                 len = sg->length;
465
466                 while (len > 0 && data_len > 0) {
467                         /*
468                          * sg sends a scatterlist that is larger than
469                          * the data_len it wants transferred for certain
470                          * IO sizes.
471                          */
472                         bytes = min_t(unsigned int, len, PAGE_SIZE - off);
473                         bytes = min(bytes, data_len);
474
475                         ref = gnttab_claim_grant_reference(&gref_head);
476                         BUG_ON(ref == -ENOSPC);
477
478                         gnttab_grant_foreign_access_ref(ref,
479                                 info->dev->otherend_id,
480                                 xen_page_to_gfn(page),
481                                 grant_ro);
482
483                         shadow->gref[ref_cnt] = ref;
484                         seg->gref   = ref;
485                         seg->offset = (uint16_t)off;
486                         seg->length = (uint16_t)bytes;
487
488                         page++;
489                         seg++;
490                         len -= bytes;
491                         data_len -= bytes;
492                         off = 0;
493                         ref_cnt++;
494                 }
495         }
496
497         if (seg_grants)
498                 shadow->nr_segments = VSCSIIF_SG_GRANT | seg_grants;
499         else
500                 shadow->nr_segments = (uint8_t)ref_cnt;
501         shadow->nr_grants = ref_cnt;
502
503         return 0;
504 }
505
506 static int scsifront_enter(struct vscsifrnt_info *info)
507 {
508         if (info->pause)
509                 return 1;
510         info->callers++;
511         return 0;
512 }
513
514 static void scsifront_return(struct vscsifrnt_info *info)
515 {
516         info->callers--;
517         if (info->callers)
518                 return;
519
520         if (!info->waiting_pause)
521                 return;
522
523         info->waiting_pause = 0;
524         wake_up(&info->wq_pause);
525 }
526
527 static int scsifront_queuecommand(struct Scsi_Host *shost,
528                                   struct scsi_cmnd *sc)
529 {
530         struct vscsifrnt_info *info = shost_priv(shost);
531         struct vscsifrnt_shadow *shadow = scsi_cmd_priv(sc);
532         unsigned long flags;
533         int err;
534
535         sc->result = 0;
536
537         shadow->sc  = sc;
538         shadow->act = VSCSIIF_ACT_SCSI_CDB;
539
540         spin_lock_irqsave(shost->host_lock, flags);
541         if (scsifront_enter(info)) {
542                 spin_unlock_irqrestore(shost->host_lock, flags);
543                 return SCSI_MLQUEUE_HOST_BUSY;
544         }
545
546         err = map_data_for_request(info, sc, shadow);
547         if (err < 0) {
548                 pr_debug("%s: err %d\n", __func__, err);
549                 scsifront_return(info);
550                 spin_unlock_irqrestore(shost->host_lock, flags);
551                 if (err == -ENOMEM)
552                         return SCSI_MLQUEUE_HOST_BUSY;
553                 sc->result = DID_ERROR << 16;
554                 sc->scsi_done(sc);
555                 return 0;
556         }
557
558         if (scsifront_do_request(info, shadow)) {
559                 scsifront_gnttab_done(info, shadow);
560                 goto busy;
561         }
562
563         scsifront_return(info);
564         spin_unlock_irqrestore(shost->host_lock, flags);
565
566         return 0;
567
568 busy:
569         scsifront_return(info);
570         spin_unlock_irqrestore(shost->host_lock, flags);
571         pr_debug("%s: busy\n", __func__);
572         return SCSI_MLQUEUE_HOST_BUSY;
573 }
574
575 /*
576  * Any exception handling (reset or abort) must be forwarded to the backend.
577  * We have to wait until an answer is returned. This answer contains the
578  * result to be returned to the requestor.
579  */
580 static int scsifront_action_handler(struct scsi_cmnd *sc, uint8_t act)
581 {
582         struct Scsi_Host *host = sc->device->host;
583         struct vscsifrnt_info *info = shost_priv(host);
584         struct vscsifrnt_shadow *shadow, *s = scsi_cmd_priv(sc);
585         int err = 0;
586
587         shadow = kzalloc(sizeof(*shadow), GFP_NOIO);
588         if (!shadow)
589                 return FAILED;
590
591         shadow->act = act;
592         shadow->rslt_reset = RSLT_RESET_WAITING;
593         shadow->sc = sc;
594         shadow->ref_rqid = s->rqid;
595         init_waitqueue_head(&shadow->wq_reset);
596
597         spin_lock_irq(host->host_lock);
598
599         for (;;) {
600                 if (scsifront_enter(info))
601                         goto fail;
602
603                 if (!scsifront_do_request(info, shadow))
604                         break;
605
606                 scsifront_return(info);
607                 if (err)
608                         goto fail;
609                 info->wait_ring_available = 1;
610                 spin_unlock_irq(host->host_lock);
611                 err = wait_event_interruptible(info->wq_sync,
612                                                !info->wait_ring_available);
613                 spin_lock_irq(host->host_lock);
614         }
615
616         spin_unlock_irq(host->host_lock);
617         err = wait_event_interruptible(shadow->wq_reset, shadow->wait_reset);
618         spin_lock_irq(host->host_lock);
619
620         if (!err) {
621                 err = shadow->rslt_reset;
622                 scsifront_put_rqid(info, shadow->rqid);
623                 kfree(shadow);
624         } else {
625                 spin_lock(&info->shadow_lock);
626                 shadow->rslt_reset = RSLT_RESET_ERR;
627                 spin_unlock(&info->shadow_lock);
628                 err = FAILED;
629         }
630
631         scsifront_return(info);
632         spin_unlock_irq(host->host_lock);
633         return err;
634
635 fail:
636         spin_unlock_irq(host->host_lock);
637         kfree(shadow);
638         return FAILED;
639 }
640
641 static int scsifront_eh_abort_handler(struct scsi_cmnd *sc)
642 {
643         pr_debug("%s\n", __func__);
644         return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_ABORT);
645 }
646
647 static int scsifront_dev_reset_handler(struct scsi_cmnd *sc)
648 {
649         pr_debug("%s\n", __func__);
650         return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_RESET);
651 }
652
653 static int scsifront_sdev_configure(struct scsi_device *sdev)
654 {
655         struct vscsifrnt_info *info = shost_priv(sdev->host);
656         int err;
657
658         if (info && current == info->curr) {
659                 err = xenbus_printf(XBT_NIL, info->dev->nodename,
660                               info->dev_state_path, "%d", XenbusStateConnected);
661                 if (err) {
662                         xenbus_dev_error(info->dev, err,
663                                 "%s: writing dev_state_path", __func__);
664                         return err;
665                 }
666         }
667
668         return 0;
669 }
670
671 static void scsifront_sdev_destroy(struct scsi_device *sdev)
672 {
673         struct vscsifrnt_info *info = shost_priv(sdev->host);
674         int err;
675
676         if (info && current == info->curr) {
677                 err = xenbus_printf(XBT_NIL, info->dev->nodename,
678                               info->dev_state_path, "%d", XenbusStateClosed);
679                 if (err)
680                         xenbus_dev_error(info->dev, err,
681                                 "%s: writing dev_state_path", __func__);
682         }
683 }
684
685 static struct scsi_host_template scsifront_sht = {
686         .module                 = THIS_MODULE,
687         .name                   = "Xen SCSI frontend driver",
688         .queuecommand           = scsifront_queuecommand,
689         .eh_abort_handler       = scsifront_eh_abort_handler,
690         .eh_device_reset_handler = scsifront_dev_reset_handler,
691         .slave_configure        = scsifront_sdev_configure,
692         .slave_destroy          = scsifront_sdev_destroy,
693         .cmd_per_lun            = VSCSIIF_DEFAULT_CMD_PER_LUN,
694         .can_queue              = VSCSIIF_MAX_REQS,
695         .this_id                = -1,
696         .cmd_size               = sizeof(struct vscsifrnt_shadow),
697         .sg_tablesize           = VSCSIIF_SG_TABLESIZE,
698         .proc_name              = "scsifront",
699 };
700
701 static int scsifront_alloc_ring(struct vscsifrnt_info *info)
702 {
703         struct xenbus_device *dev = info->dev;
704         struct vscsiif_sring *sring;
705         grant_ref_t gref;
706         int err = -ENOMEM;
707
708         /***** Frontend to Backend ring start *****/
709         sring = (struct vscsiif_sring *)__get_free_page(GFP_KERNEL);
710         if (!sring) {
711                 xenbus_dev_fatal(dev, err,
712                         "fail to allocate shared ring (Front to Back)");
713                 return err;
714         }
715         SHARED_RING_INIT(sring);
716         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
717
718         err = xenbus_grant_ring(dev, sring, 1, &gref);
719         if (err < 0) {
720                 free_page((unsigned long)sring);
721                 xenbus_dev_fatal(dev, err,
722                         "fail to grant shared ring (Front to Back)");
723                 return err;
724         }
725         info->ring_ref = gref;
726
727         err = xenbus_alloc_evtchn(dev, &info->evtchn);
728         if (err) {
729                 xenbus_dev_fatal(dev, err, "xenbus_alloc_evtchn");
730                 goto free_gnttab;
731         }
732
733         err = bind_evtchn_to_irq(info->evtchn);
734         if (err <= 0) {
735                 xenbus_dev_fatal(dev, err, "bind_evtchn_to_irq");
736                 goto free_gnttab;
737         }
738
739         info->irq = err;
740
741         err = request_threaded_irq(info->irq, NULL, scsifront_irq_fn,
742                                    IRQF_ONESHOT, "scsifront", info);
743         if (err) {
744                 xenbus_dev_fatal(dev, err, "request_threaded_irq");
745                 goto free_irq;
746         }
747
748         return 0;
749
750 /* free resource */
751 free_irq:
752         unbind_from_irqhandler(info->irq, info);
753 free_gnttab:
754         gnttab_end_foreign_access(info->ring_ref, 0,
755                                   (unsigned long)info->ring.sring);
756
757         return err;
758 }
759
760 static void scsifront_free_ring(struct vscsifrnt_info *info)
761 {
762         unbind_from_irqhandler(info->irq, info);
763         gnttab_end_foreign_access(info->ring_ref, 0,
764                                   (unsigned long)info->ring.sring);
765 }
766
767 static int scsifront_init_ring(struct vscsifrnt_info *info)
768 {
769         struct xenbus_device *dev = info->dev;
770         struct xenbus_transaction xbt;
771         int err;
772
773         pr_debug("%s\n", __func__);
774
775         err = scsifront_alloc_ring(info);
776         if (err)
777                 return err;
778         pr_debug("%s: %u %u\n", __func__, info->ring_ref, info->evtchn);
779
780 again:
781         err = xenbus_transaction_start(&xbt);
782         if (err)
783                 xenbus_dev_fatal(dev, err, "starting transaction");
784
785         err = xenbus_printf(xbt, dev->nodename, "ring-ref", "%u",
786                             info->ring_ref);
787         if (err) {
788                 xenbus_dev_fatal(dev, err, "%s", "writing ring-ref");
789                 goto fail;
790         }
791
792         err = xenbus_printf(xbt, dev->nodename, "event-channel", "%u",
793                             info->evtchn);
794
795         if (err) {
796                 xenbus_dev_fatal(dev, err, "%s", "writing event-channel");
797                 goto fail;
798         }
799
800         err = xenbus_transaction_end(xbt, 0);
801         if (err) {
802                 if (err == -EAGAIN)
803                         goto again;
804                 xenbus_dev_fatal(dev, err, "completing transaction");
805                 goto free_sring;
806         }
807
808         return 0;
809
810 fail:
811         xenbus_transaction_end(xbt, 1);
812 free_sring:
813         scsifront_free_ring(info);
814
815         return err;
816 }
817
818
819 static int scsifront_probe(struct xenbus_device *dev,
820                            const struct xenbus_device_id *id)
821 {
822         struct vscsifrnt_info *info;
823         struct Scsi_Host *host;
824         int err = -ENOMEM;
825         char name[TASK_COMM_LEN];
826
827         host = scsi_host_alloc(&scsifront_sht, sizeof(*info));
828         if (!host) {
829                 xenbus_dev_fatal(dev, err, "fail to allocate scsi host");
830                 return err;
831         }
832         info = (struct vscsifrnt_info *)host->hostdata;
833
834         dev_set_drvdata(&dev->dev, info);
835         info->dev = dev;
836
837         bitmap_fill(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
838
839         err = scsifront_init_ring(info);
840         if (err) {
841                 scsi_host_put(host);
842                 return err;
843         }
844
845         init_waitqueue_head(&info->wq_sync);
846         init_waitqueue_head(&info->wq_pause);
847         spin_lock_init(&info->shadow_lock);
848
849         snprintf(name, TASK_COMM_LEN, "vscsiif.%d", host->host_no);
850
851         host->max_id      = VSCSIIF_MAX_TARGET;
852         host->max_channel = 0;
853         host->max_lun     = VSCSIIF_MAX_LUN;
854         host->max_sectors = (host->sg_tablesize - 1) * PAGE_SIZE / 512;
855         host->max_cmd_len = VSCSIIF_MAX_COMMAND_SIZE;
856
857         err = scsi_add_host(host, &dev->dev);
858         if (err) {
859                 dev_err(&dev->dev, "fail to add scsi host %d\n", err);
860                 goto free_sring;
861         }
862         info->host = host;
863         info->host_active = 1;
864
865         xenbus_switch_state(dev, XenbusStateInitialised);
866
867         return 0;
868
869 free_sring:
870         scsifront_free_ring(info);
871         scsi_host_put(host);
872         return err;
873 }
874
875 static int scsifront_resume(struct xenbus_device *dev)
876 {
877         struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
878         struct Scsi_Host *host = info->host;
879         int err;
880
881         spin_lock_irq(host->host_lock);
882
883         /* Finish all still pending commands. */
884         scsifront_finish_all(info);
885
886         spin_unlock_irq(host->host_lock);
887
888         /* Reconnect to dom0. */
889         scsifront_free_ring(info);
890         err = scsifront_init_ring(info);
891         if (err) {
892                 dev_err(&dev->dev, "fail to resume %d\n", err);
893                 scsi_host_put(host);
894                 return err;
895         }
896
897         xenbus_switch_state(dev, XenbusStateInitialised);
898
899         return 0;
900 }
901
902 static int scsifront_suspend(struct xenbus_device *dev)
903 {
904         struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
905         struct Scsi_Host *host = info->host;
906         int err = 0;
907
908         /* No new commands for the backend. */
909         spin_lock_irq(host->host_lock);
910         info->pause = 1;
911         while (info->callers && !err) {
912                 info->waiting_pause = 1;
913                 info->wait_ring_available = 0;
914                 spin_unlock_irq(host->host_lock);
915                 wake_up(&info->wq_sync);
916                 err = wait_event_interruptible(info->wq_pause,
917                                                !info->waiting_pause);
918                 spin_lock_irq(host->host_lock);
919         }
920         spin_unlock_irq(host->host_lock);
921         return err;
922 }
923
924 static int scsifront_remove(struct xenbus_device *dev)
925 {
926         struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
927
928         pr_debug("%s: %s removed\n", __func__, dev->nodename);
929
930         mutex_lock(&scsifront_mutex);
931         if (info->host_active) {
932                 /* Scsi_host not yet removed */
933                 scsi_remove_host(info->host);
934                 info->host_active = 0;
935         }
936         mutex_unlock(&scsifront_mutex);
937
938         scsifront_free_ring(info);
939         scsi_host_put(info->host);
940
941         return 0;
942 }
943
944 static void scsifront_disconnect(struct vscsifrnt_info *info)
945 {
946         struct xenbus_device *dev = info->dev;
947         struct Scsi_Host *host = info->host;
948
949         pr_debug("%s: %s disconnect\n", __func__, dev->nodename);
950
951         /*
952          * When this function is executed, all devices of
953          * Frontend have been deleted.
954          * Therefore, it need not block I/O before remove_host.
955          */
956
957         mutex_lock(&scsifront_mutex);
958         if (info->host_active) {
959                 scsi_remove_host(host);
960                 info->host_active = 0;
961         }
962         mutex_unlock(&scsifront_mutex);
963
964         xenbus_frontend_closed(dev);
965 }
966
967 static void scsifront_do_lun_hotplug(struct vscsifrnt_info *info, int op)
968 {
969         struct xenbus_device *dev = info->dev;
970         int i, err = 0;
971         char str[64];
972         char **dir;
973         unsigned int dir_n = 0;
974         unsigned int device_state;
975         unsigned int hst, chn, tgt, lun;
976         struct scsi_device *sdev;
977
978         dir = xenbus_directory(XBT_NIL, dev->otherend, "vscsi-devs", &dir_n);
979         if (IS_ERR(dir))
980                 return;
981
982         /* mark current task as the one allowed to modify device states */
983         BUG_ON(info->curr);
984         info->curr = current;
985
986         for (i = 0; i < dir_n; i++) {
987                 /* read status */
988                 snprintf(str, sizeof(str), "vscsi-devs/%s/state", dir[i]);
989                 err = xenbus_scanf(XBT_NIL, dev->otherend, str, "%u",
990                                    &device_state);
991                 if (XENBUS_EXIST_ERR(err))
992                         continue;
993
994                 /* virtual SCSI device */
995                 snprintf(str, sizeof(str), "vscsi-devs/%s/v-dev", dir[i]);
996                 err = xenbus_scanf(XBT_NIL, dev->otherend, str,
997                                    "%u:%u:%u:%u", &hst, &chn, &tgt, &lun);
998                 if (XENBUS_EXIST_ERR(err))
999                         continue;
1000
1001                 /*
1002                  * Front device state path, used in slave_configure called
1003                  * on successfull scsi_add_device, and in slave_destroy called
1004                  * on remove of a device.
1005                  */
1006                 snprintf(info->dev_state_path, sizeof(info->dev_state_path),
1007                          "vscsi-devs/%s/state", dir[i]);
1008
1009                 switch (op) {
1010                 case VSCSIFRONT_OP_ADD_LUN:
1011                         if (device_state != XenbusStateInitialised)
1012                                 break;
1013
1014                         if (scsi_add_device(info->host, chn, tgt, lun)) {
1015                                 dev_err(&dev->dev, "scsi_add_device\n");
1016                                 err = xenbus_printf(XBT_NIL, dev->nodename,
1017                                               info->dev_state_path,
1018                                               "%d", XenbusStateClosed);
1019                                 if (err)
1020                                         xenbus_dev_error(dev, err,
1021                                                 "%s: writing dev_state_path", __func__);
1022                         }
1023                         break;
1024                 case VSCSIFRONT_OP_DEL_LUN:
1025                         if (device_state != XenbusStateClosing)
1026                                 break;
1027
1028                         sdev = scsi_device_lookup(info->host, chn, tgt, lun);
1029                         if (sdev) {
1030                                 scsi_remove_device(sdev);
1031                                 scsi_device_put(sdev);
1032                         }
1033                         break;
1034                 case VSCSIFRONT_OP_READD_LUN:
1035                         if (device_state == XenbusStateConnected) {
1036                                 err = xenbus_printf(XBT_NIL, dev->nodename,
1037                                               info->dev_state_path,
1038                                               "%d", XenbusStateConnected);
1039                                 if (err)
1040                                         xenbus_dev_error(dev, err,
1041                                                 "%s: writing dev_state_path", __func__);
1042                         }
1043                         break;
1044                 default:
1045                         break;
1046                 }
1047         }
1048
1049         info->curr = NULL;
1050
1051         kfree(dir);
1052 }
1053
1054 static void scsifront_read_backend_params(struct xenbus_device *dev,
1055                                           struct vscsifrnt_info *info)
1056 {
1057         unsigned int sg_grant, nr_segs;
1058         struct Scsi_Host *host = info->host;
1059
1060         sg_grant = xenbus_read_unsigned(dev->otherend, "feature-sg-grant", 0);
1061         nr_segs = min_t(unsigned int, sg_grant, SG_ALL);
1062         nr_segs = max_t(unsigned int, nr_segs, VSCSIIF_SG_TABLESIZE);
1063         nr_segs = min_t(unsigned int, nr_segs,
1064                         VSCSIIF_SG_TABLESIZE * PAGE_SIZE /
1065                         sizeof(struct scsiif_request_segment));
1066
1067         if (!info->pause && sg_grant)
1068                 dev_info(&dev->dev, "using up to %d SG entries\n", nr_segs);
1069         else if (info->pause && nr_segs < host->sg_tablesize)
1070                 dev_warn(&dev->dev,
1071                          "SG entries decreased from %d to %u - device may not work properly anymore\n",
1072                          host->sg_tablesize, nr_segs);
1073
1074         host->sg_tablesize = nr_segs;
1075         host->max_sectors = (nr_segs - 1) * PAGE_SIZE / 512;
1076 }
1077
1078 static void scsifront_backend_changed(struct xenbus_device *dev,
1079                                       enum xenbus_state backend_state)
1080 {
1081         struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
1082
1083         pr_debug("%s: %p %u %u\n", __func__, dev, dev->state, backend_state);
1084
1085         switch (backend_state) {
1086         case XenbusStateUnknown:
1087         case XenbusStateInitialising:
1088         case XenbusStateInitWait:
1089         case XenbusStateInitialised:
1090                 break;
1091
1092         case XenbusStateConnected:
1093                 scsifront_read_backend_params(dev, info);
1094
1095                 if (info->pause) {
1096                         scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_READD_LUN);
1097                         xenbus_switch_state(dev, XenbusStateConnected);
1098                         info->pause = 0;
1099                         return;
1100                 }
1101
1102                 if (xenbus_read_driver_state(dev->nodename) ==
1103                     XenbusStateInitialised)
1104                         scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
1105
1106                 if (dev->state != XenbusStateConnected)
1107                         xenbus_switch_state(dev, XenbusStateConnected);
1108                 break;
1109
1110         case XenbusStateClosed:
1111                 if (dev->state == XenbusStateClosed)
1112                         break;
1113                 fallthrough;    /* Missed the backend's Closing state */
1114         case XenbusStateClosing:
1115                 scsifront_disconnect(info);
1116                 break;
1117
1118         case XenbusStateReconfiguring:
1119                 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_DEL_LUN);
1120                 xenbus_switch_state(dev, XenbusStateReconfiguring);
1121                 break;
1122
1123         case XenbusStateReconfigured:
1124                 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
1125                 xenbus_switch_state(dev, XenbusStateConnected);
1126                 break;
1127         }
1128 }
1129
1130 static const struct xenbus_device_id scsifront_ids[] = {
1131         { "vscsi" },
1132         { "" }
1133 };
1134
1135 static struct xenbus_driver scsifront_driver = {
1136         .ids                    = scsifront_ids,
1137         .probe                  = scsifront_probe,
1138         .remove                 = scsifront_remove,
1139         .resume                 = scsifront_resume,
1140         .suspend                = scsifront_suspend,
1141         .otherend_changed       = scsifront_backend_changed,
1142 };
1143
1144 static int __init scsifront_init(void)
1145 {
1146         if (!xen_domain())
1147                 return -ENODEV;
1148
1149         return xenbus_register_frontend(&scsifront_driver);
1150 }
1151 module_init(scsifront_init);
1152
1153 static void __exit scsifront_exit(void)
1154 {
1155         xenbus_unregister_driver(&scsifront_driver);
1156 }
1157 module_exit(scsifront_exit);
1158
1159 MODULE_DESCRIPTION("Xen SCSI frontend driver");
1160 MODULE_LICENSE("GPL");
1161 MODULE_ALIAS("xen:vscsi");
1162 MODULE_AUTHOR("Juergen Gross <jgross@suse.com>");