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 #define VSCSIFRONT_OP_ADD_LUN   1
  62 #define VSCSIFRONT_OP_DEL_LUN   2
  63 #define VSCSIFRONT_OP_READD_LUN 3
  64
  65 /* Tuning point. */
  66 #define VSCSIIF_DEFAULT_CMD_PER_LUN 10
  67 #define VSCSIIF_MAX_TARGET          64
  68 #define VSCSIIF_MAX_LUN             255
  69
  70 #define VSCSIIF_RING_SIZE       __CONST_RING_SIZE(vscsiif, PAGE_SIZE)
  71 #define VSCSIIF_MAX_REQS        VSCSIIF_RING_SIZE
  72
  73 #define vscsiif_grants_sg(_sg)  (PFN_UP((_sg) *         \
  74                                 sizeof(struct scsiif_request_segment)))
  75
  76 struct vscsifrnt_shadow {
  77         /* command between backend and frontend */
  78         unsigned char act;
  79         uint8_t nr_segments;
  80         uint16_t rqid;
  81         uint16_t ref_rqid;
  82
  83         bool inflight;
  84
  85         unsigned int nr_grants;         /* number of grants in gref[] */
  86         struct scsiif_request_segment *sg;      /* scatter/gather elements */
  87         struct scsiif_request_segment seg[VSCSIIF_SG_TABLESIZE];
  88
  89         /* Do reset or abort function. */
  90         wait_queue_head_t wq_reset;     /* reset work queue           */
  91         int wait_reset;                 /* reset work queue condition */
  92         int32_t rslt_reset;             /* reset response status:     */
  93                                         /* SUCCESS or FAILED or:      */
  94 #define RSLT_RESET_WAITING      0
  95 #define RSLT_RESET_ERR          -1
  96
  97         /* Requested struct scsi_cmnd is stored from kernel. */
  98         struct scsi_cmnd *sc;
  99         int gref[vscsiif_grants_sg(SG_ALL) + SG_ALL];
 100 };
 101
 102 struct vscsifrnt_info {
 103         struct xenbus_device *dev;
 104
 105         struct Scsi_Host *host;
 106         enum {
 107                 STATE_INACTIVE,
 108                 STATE_ACTIVE,
 109                 STATE_ERROR
 110         }  host_active;
 111
 112         unsigned int evtchn;
 113         unsigned int irq;
 114
 115         grant_ref_t ring_ref;
 116         struct vscsiif_front_ring ring;
 117         struct vscsiif_response ring_rsp;
 118
 119         spinlock_t shadow_lock;
 120         DECLARE_BITMAP(shadow_free_bitmap, VSCSIIF_MAX_REQS);
 121         struct vscsifrnt_shadow *shadow[VSCSIIF_MAX_REQS];
 122
 123         /* Following items are protected by the host lock. */
 124         wait_queue_head_t wq_sync;
 125         wait_queue_head_t wq_pause;
 126         unsigned int wait_ring_available:1;
 127         unsigned int waiting_pause:1;
 128         unsigned int pause:1;
 129         unsigned callers;
 130
 131         char dev_state_path[64];
 132         struct task_struct *curr;
 133 };
 134
 135 static DEFINE_MUTEX(scsifront_mutex);
 136
 137 static void scsifront_wake_up(struct vscsifrnt_info *info)
 138 {
 139         info->wait_ring_available = 0;
 140         wake_up(&info->wq_sync);
 141 }
 142
 143 static int scsifront_get_rqid(struct vscsifrnt_info *info)
 144 {
 145         unsigned long flags;
 146         int free;
 147
 148         spin_lock_irqsave(&info->shadow_lock, flags);
 149
 150         free = find_first_bit(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
 151         __clear_bit(free, info->shadow_free_bitmap);
 152
 153         spin_unlock_irqrestore(&info->shadow_lock, flags);
 154
 155         return free;
 156 }
 157
 158 static int _scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
 159 {
 160         int empty = bitmap_empty(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
 161
 162         __set_bit(id, info->shadow_free_bitmap);
 163         info->shadow[id] = NULL;
 164
 165         return empty || info->wait_ring_available;
 166 }
 167
 168 static void scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
 169 {
 170         unsigned long flags;
 171         int kick;
 172
 173         spin_lock_irqsave(&info->shadow_lock, flags);
 174         kick = _scsifront_put_rqid(info, id);
 175         spin_unlock_irqrestore(&info->shadow_lock, flags);
 176
 177         if (kick)
 178                 scsifront_wake_up(info);
 179 }
 180
 181 static int scsifront_do_request(struct vscsifrnt_info *info,
 182                                 struct vscsifrnt_shadow *shadow)
 183 {
 184         struct vscsiif_front_ring *ring = &(info->ring);
 185         struct vscsiif_request *ring_req;
 186         struct scsi_cmnd *sc = shadow->sc;
 187         uint32_t id;
 188         int i, notify;
 189
 190         if (RING_FULL(&info->ring))
 191                 return -EBUSY;
 192
 193         id = scsifront_get_rqid(info);  /* use id in response */
 194         if (id >= VSCSIIF_MAX_REQS)
 195                 return -EBUSY;
 196
 197         info->shadow[id] = shadow;
 198         shadow->rqid = id;
 199
 200         ring_req = RING_GET_REQUEST(&(info->ring), ring->req_prod_pvt);
 201         ring->req_prod_pvt++;
 202
 203         ring_req->rqid        = id;
 204         ring_req->act         = shadow->act;
 205         ring_req->ref_rqid    = shadow->ref_rqid;
 206         ring_req->nr_segments = shadow->nr_segments;
 207
 208         ring_req->id      = sc->device->id;
 209         ring_req->lun     = sc->device->lun;
 210         ring_req->channel = sc->device->channel;
 211         ring_req->cmd_len = sc->cmd_len;
 212
 213         BUG_ON(sc->cmd_len > VSCSIIF_MAX_COMMAND_SIZE);
 214
 215         memcpy(ring_req->cmnd, sc->cmnd, sc->cmd_len);
 216
 217         ring_req->sc_data_direction   = (uint8_t)sc->sc_data_direction;
 218         ring_req->timeout_per_command = scsi_cmd_to_rq(sc)->timeout / HZ;
 219
 220         for (i = 0; i < (shadow->nr_segments & ~VSCSIIF_SG_GRANT); i++)
 221                 ring_req->seg[i] = shadow->seg[i];
 222
 223         shadow->inflight = true;
 224
 225         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(ring, notify);
 226         if (notify)
 227                 notify_remote_via_irq(info->irq);
 228
 229         return 0;
 230 }
 231
 232 static void scsifront_set_error(struct vscsifrnt_info *info, const char *msg)
 233 {
 234         shost_printk(KERN_ERR, info->host, KBUILD_MODNAME "%s\n"
 235                      "Disabling device for further use\n", msg);
 236         info->host_active = STATE_ERROR;
 237 }
 238
 239 static void scsifront_gnttab_done(struct vscsifrnt_info *info,
 240                                   struct vscsifrnt_shadow *shadow)
 241 {
 242         int i;
 243
 244         if (shadow->sc->sc_data_direction == DMA_NONE)
 245                 return;
 246
 247         for (i = 0; i < shadow->nr_grants; i++) {
 248                 if (unlikely(!gnttab_try_end_foreign_access(shadow->gref[i]))) {
 249                         scsifront_set_error(info, "grant still in use by backend");
 250                         return;
 251                 }
 252         }
 253
 254         kfree(shadow->sg);
 255 }
 256
 257 static unsigned int scsifront_host_byte(int32_t rslt)
 258 {
 259         switch (XEN_VSCSIIF_RSLT_HOST(rslt)) {
 260         case XEN_VSCSIIF_RSLT_HOST_OK:
 261                 return DID_OK;
 262         case XEN_VSCSIIF_RSLT_HOST_NO_CONNECT:
 263                 return DID_NO_CONNECT;
 264         case XEN_VSCSIIF_RSLT_HOST_BUS_BUSY:
 265                 return DID_BUS_BUSY;
 266         case XEN_VSCSIIF_RSLT_HOST_TIME_OUT:
 267                 return DID_TIME_OUT;
 268         case XEN_VSCSIIF_RSLT_HOST_BAD_TARGET:
 269                 return DID_BAD_TARGET;
 270         case XEN_VSCSIIF_RSLT_HOST_ABORT:
 271                 return DID_ABORT;
 272         case XEN_VSCSIIF_RSLT_HOST_PARITY:
 273                 return DID_PARITY;
 274         case XEN_VSCSIIF_RSLT_HOST_ERROR:
 275                 return DID_ERROR;
 276         case XEN_VSCSIIF_RSLT_HOST_RESET:
 277                 return DID_RESET;
 278         case XEN_VSCSIIF_RSLT_HOST_BAD_INTR:
 279                 return DID_BAD_INTR;
 280         case XEN_VSCSIIF_RSLT_HOST_PASSTHROUGH:
 281                 return DID_PASSTHROUGH;
 282         case XEN_VSCSIIF_RSLT_HOST_SOFT_ERROR:
 283                 return DID_SOFT_ERROR;
 284         case XEN_VSCSIIF_RSLT_HOST_IMM_RETRY:
 285                 return DID_IMM_RETRY;
 286         case XEN_VSCSIIF_RSLT_HOST_REQUEUE:
 287                 return DID_REQUEUE;
 288         case XEN_VSCSIIF_RSLT_HOST_TRANSPORT_DISRUPTED:
 289                 return DID_TRANSPORT_DISRUPTED;
 290         case XEN_VSCSIIF_RSLT_HOST_TRANSPORT_FAILFAST:
 291                 return DID_TRANSPORT_FAILFAST;
 292         case XEN_VSCSIIF_RSLT_HOST_TRANSPORT_MARGINAL:
 293                 return DID_TRANSPORT_MARGINAL;
 294         default:
 295                 return DID_ERROR;
 296         }
 297 }
 298
 299 static void scsifront_cdb_cmd_done(struct vscsifrnt_info *info,
 300                                    struct vscsiif_response *ring_rsp)
 301 {
 302         struct vscsifrnt_shadow *shadow;
 303         struct scsi_cmnd *sc;
 304         uint32_t id;
 305         uint8_t sense_len;
 306
 307         id = ring_rsp->rqid;
 308         shadow = info->shadow[id];
 309         sc = shadow->sc;
 310
 311         BUG_ON(sc == NULL);
 312
 313         scsifront_gnttab_done(info, shadow);
 314         if (info->host_active == STATE_ERROR)
 315                 return;
 316         scsifront_put_rqid(info, id);
 317
 318         set_host_byte(sc, scsifront_host_byte(ring_rsp->rslt));
 319         set_status_byte(sc, XEN_VSCSIIF_RSLT_STATUS(ring_rsp->rslt));
 320         scsi_set_resid(sc, ring_rsp->residual_len);
 321
 322         sense_len = min_t(uint8_t, VSCSIIF_SENSE_BUFFERSIZE,
 323                           ring_rsp->sense_len);
 324
 325         if (sense_len)
 326                 memcpy(sc->sense_buffer, ring_rsp->sense_buffer, sense_len);
 327
 328         scsi_done(sc);
 329 }
 330
 331 static void scsifront_sync_cmd_done(struct vscsifrnt_info *info,
 332                                     struct vscsiif_response *ring_rsp)
 333 {
 334         uint16_t id = ring_rsp->rqid;
 335         unsigned long flags;
 336         struct vscsifrnt_shadow *shadow = info->shadow[id];
 337         int kick;
 338
 339         spin_lock_irqsave(&info->shadow_lock, flags);
 340         shadow->wait_reset = 1;
 341         switch (shadow->rslt_reset) {
 342         case RSLT_RESET_WAITING:
 343                 if (ring_rsp->rslt == XEN_VSCSIIF_RSLT_RESET_SUCCESS)
 344                         shadow->rslt_reset = SUCCESS;
 345                 else
 346                         shadow->rslt_reset = FAILED;
 347                 break;
 348         case RSLT_RESET_ERR:
 349                 kick = _scsifront_put_rqid(info, id);
 350                 spin_unlock_irqrestore(&info->shadow_lock, flags);
 351                 kfree(shadow);
 352                 if (kick)
 353                         scsifront_wake_up(info);
 354                 return;
 355         default:
 356                 scsifront_set_error(info, "bad reset state");
 357                 break;
 358         }
 359         spin_unlock_irqrestore(&info->shadow_lock, flags);
 360
 361         wake_up(&shadow->wq_reset);
 362 }
 363
 364 static void scsifront_do_response(struct vscsifrnt_info *info,
 365                                   struct vscsiif_response *ring_rsp)
 366 {
 367         struct vscsifrnt_shadow *shadow;
 368
 369         if (ring_rsp->rqid >= VSCSIIF_MAX_REQS ||
 370             !info->shadow[ring_rsp->rqid]->inflight) {
 371                 scsifront_set_error(info, "illegal rqid returned by backend!");
 372                 return;
 373         }
 374         shadow = info->shadow[ring_rsp->rqid];
 375         shadow->inflight = false;
 376
 377         if (shadow->act == VSCSIIF_ACT_SCSI_CDB)
 378                 scsifront_cdb_cmd_done(info, ring_rsp);
 379         else
 380                 scsifront_sync_cmd_done(info, ring_rsp);
 381 }
 382
 383 static int scsifront_ring_drain(struct vscsifrnt_info *info,
 384                                 unsigned int *eoiflag)
 385 {
 386         struct vscsiif_response ring_rsp;
 387         RING_IDX i, rp;
 388         int more_to_do = 0;
 389
 390         rp = READ_ONCE(info->ring.sring->rsp_prod);
 391         virt_rmb();     /* ordering required respective to backend */
 392         if (RING_RESPONSE_PROD_OVERFLOW(&info->ring, rp)) {
 393                 scsifront_set_error(info, "illegal number of responses");
 394                 return 0;
 395         }
 396         for (i = info->ring.rsp_cons; i != rp; i++) {
 397                 RING_COPY_RESPONSE(&info->ring, i, &ring_rsp);
 398                 scsifront_do_response(info, &ring_rsp);
 399                 if (info->host_active == STATE_ERROR)
 400                         return 0;
 401                 *eoiflag &= ~XEN_EOI_FLAG_SPURIOUS;
 402         }
 403
 404         info->ring.rsp_cons = i;
 405
 406         if (i != info->ring.req_prod_pvt)
 407                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 408         else
 409                 info->ring.sring->rsp_event = i + 1;
 410
 411         return more_to_do;
 412 }
 413
 414 static int scsifront_cmd_done(struct vscsifrnt_info *info,
 415                               unsigned int *eoiflag)
 416 {
 417         int more_to_do;
 418         unsigned long flags;
 419
 420         spin_lock_irqsave(info->host->host_lock, flags);
 421
 422         more_to_do = scsifront_ring_drain(info, eoiflag);
 423
 424         info->wait_ring_available = 0;
 425
 426         spin_unlock_irqrestore(info->host->host_lock, flags);
 427
 428         wake_up(&info->wq_sync);
 429
 430         return more_to_do;
 431 }
 432
 433 static irqreturn_t scsifront_irq_fn(int irq, void *dev_id)
 434 {
 435         struct vscsifrnt_info *info = dev_id;
 436         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
 437
 438         if (info->host_active == STATE_ERROR) {
 439                 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
 440                 return IRQ_HANDLED;
 441         }
 442
 443         while (scsifront_cmd_done(info, &eoiflag))
 444                 /* Yield point for this unbounded loop. */
 445                 cond_resched();
 446
 447         xen_irq_lateeoi(irq, eoiflag);
 448
 449         return IRQ_HANDLED;
 450 }
 451
 452 static void scsifront_finish_all(struct vscsifrnt_info *info)
 453 {
 454         unsigned int i, dummy;
 455         struct vscsiif_response resp;
 456
 457         scsifront_ring_drain(info, &dummy);
 458
 459         for (i = 0; i < VSCSIIF_MAX_REQS; i++) {
 460                 if (test_bit(i, info->shadow_free_bitmap))
 461                         continue;
 462                 resp.rqid = i;
 463                 resp.sense_len = 0;
 464                 resp.rslt = DID_RESET << 16;
 465                 resp.residual_len = 0;
 466                 scsifront_do_response(info, &resp);
 467         }
 468 }
 469
 470 static int map_data_for_request(struct vscsifrnt_info *info,
 471                                 struct scsi_cmnd *sc,
 472                                 struct vscsifrnt_shadow *shadow)
 473 {
 474         grant_ref_t gref_head;
 475         struct page *page;
 476         int err, ref, ref_cnt = 0;
 477         int grant_ro = (sc->sc_data_direction == DMA_TO_DEVICE);
 478         unsigned int i, off, len, bytes;
 479         unsigned int data_len = scsi_bufflen(sc);
 480         unsigned int data_grants = 0, seg_grants = 0;
 481         struct scatterlist *sg;
 482         struct scsiif_request_segment *seg;
 483
 484         if (sc->sc_data_direction == DMA_NONE || !data_len)
 485                 return 0;
 486
 487         scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i)
 488                 data_grants += PFN_UP(sg->offset + sg->length);
 489
 490         if (data_grants > VSCSIIF_SG_TABLESIZE) {
 491                 if (data_grants > info->host->sg_tablesize) {
 492                         shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
 493                              "Unable to map request_buffer for command!\n");
 494                         return -E2BIG;
 495                 }
 496                 seg_grants = vscsiif_grants_sg(data_grants);
 497                 shadow->sg = kcalloc(data_grants,
 498                         sizeof(struct scsiif_request_segment), GFP_ATOMIC);
 499                 if (!shadow->sg)
 500                         return -ENOMEM;
 501         }
 502         seg = shadow->sg ? : shadow->seg;
 503
 504         err = gnttab_alloc_grant_references(seg_grants + data_grants,
 505                                             &gref_head);
 506         if (err) {
 507                 kfree(shadow->sg);
 508                 shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
 509                              "gnttab_alloc_grant_references() error\n");
 510                 return -ENOMEM;
 511         }
 512
 513         if (seg_grants) {
 514                 page = virt_to_page(seg);
 515                 off = offset_in_page(seg);
 516                 len = sizeof(struct scsiif_request_segment) * data_grants;
 517                 while (len > 0) {
 518                         bytes = min_t(unsigned int, len, PAGE_SIZE - off);
 519
 520                         ref = gnttab_claim_grant_reference(&gref_head);
 521                         BUG_ON(ref == -ENOSPC);
 522
 523                         gnttab_grant_foreign_access_ref(ref,
 524                                 info->dev->otherend_id,
 525                                 xen_page_to_gfn(page), 1);
 526                         shadow->gref[ref_cnt] = ref;
 527                         shadow->seg[ref_cnt].gref   = ref;
 528                         shadow->seg[ref_cnt].offset = (uint16_t)off;
 529                         shadow->seg[ref_cnt].length = (uint16_t)bytes;
 530
 531                         page++;
 532                         len -= bytes;
 533                         off = 0;
 534                         ref_cnt++;
 535                 }
 536                 BUG_ON(seg_grants < ref_cnt);
 537                 seg_grants = ref_cnt;
 538         }
 539
 540         scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) {
 541                 page = sg_page(sg);
 542                 off = sg->offset;
 543                 len = sg->length;
 544
 545                 while (len > 0 && data_len > 0) {
 546                         /*
 547                          * sg sends a scatterlist that is larger than
 548                          * the data_len it wants transferred for certain
 549                          * IO sizes.
 550                          */
 551                         bytes = min_t(unsigned int, len, PAGE_SIZE - off);
 552                         bytes = min(bytes, data_len);
 553
 554                         ref = gnttab_claim_grant_reference(&gref_head);
 555                         BUG_ON(ref == -ENOSPC);
 556
 557                         gnttab_grant_foreign_access_ref(ref,
 558                                 info->dev->otherend_id,
 559                                 xen_page_to_gfn(page),
 560                                 grant_ro);
 561
 562                         shadow->gref[ref_cnt] = ref;
 563                         seg->gref   = ref;
 564                         seg->offset = (uint16_t)off;
 565                         seg->length = (uint16_t)bytes;
 566
 567                         page++;
 568                         seg++;
 569                         len -= bytes;
 570                         data_len -= bytes;
 571                         off = 0;
 572                         ref_cnt++;
 573                 }
 574         }
 575
 576         if (seg_grants)
 577                 shadow->nr_segments = VSCSIIF_SG_GRANT | seg_grants;
 578         else
 579                 shadow->nr_segments = (uint8_t)ref_cnt;
 580         shadow->nr_grants = ref_cnt;
 581
 582         return 0;
 583 }
 584
 585 static int scsifront_enter(struct vscsifrnt_info *info)
 586 {
 587         if (info->pause)
 588                 return 1;
 589         info->callers++;
 590         return 0;
 591 }
 592
 593 static void scsifront_return(struct vscsifrnt_info *info)
 594 {
 595         info->callers--;
 596         if (info->callers)
 597                 return;
 598
 599         if (!info->waiting_pause)
 600                 return;
 601
 602         info->waiting_pause = 0;
 603         wake_up(&info->wq_pause);
 604 }
 605
 606 static int scsifront_queuecommand(struct Scsi_Host *shost,
 607                                   struct scsi_cmnd *sc)
 608 {
 609         struct vscsifrnt_info *info = shost_priv(shost);
 610         struct vscsifrnt_shadow *shadow = scsi_cmd_priv(sc);
 611         unsigned long flags;
 612         int err;
 613
 614         if (info->host_active == STATE_ERROR)
 615                 return SCSI_MLQUEUE_HOST_BUSY;
 616
 617         sc->result = 0;
 618
 619         shadow->sc  = sc;
 620         shadow->act = VSCSIIF_ACT_SCSI_CDB;
 621
 622         spin_lock_irqsave(shost->host_lock, flags);
 623         if (scsifront_enter(info)) {
 624                 spin_unlock_irqrestore(shost->host_lock, flags);
 625                 return SCSI_MLQUEUE_HOST_BUSY;
 626         }
 627
 628         err = map_data_for_request(info, sc, shadow);
 629         if (err < 0) {
 630                 pr_debug("%s: err %d\n", __func__, err);
 631                 scsifront_return(info);
 632                 spin_unlock_irqrestore(shost->host_lock, flags);
 633                 if (err == -ENOMEM)
 634                         return SCSI_MLQUEUE_HOST_BUSY;
 635                 sc->result = DID_ERROR << 16;
 636                 scsi_done(sc);
 637                 return 0;
 638         }
 639
 640         if (scsifront_do_request(info, shadow)) {
 641                 scsifront_gnttab_done(info, shadow);
 642                 goto busy;
 643         }
 644
 645         scsifront_return(info);
 646         spin_unlock_irqrestore(shost->host_lock, flags);
 647
 648         return 0;
 649
 650 busy:
 651         scsifront_return(info);
 652         spin_unlock_irqrestore(shost->host_lock, flags);
 653         pr_debug("%s: busy\n", __func__);
 654         return SCSI_MLQUEUE_HOST_BUSY;
 655 }
 656
 657 /*
 658  * Any exception handling (reset or abort) must be forwarded to the backend.
 659  * We have to wait until an answer is returned. This answer contains the
 660  * result to be returned to the requestor.
 661  */
 662 static int scsifront_action_handler(struct scsi_cmnd *sc, uint8_t act)
 663 {
 664         struct Scsi_Host *host = sc->device->host;
 665         struct vscsifrnt_info *info = shost_priv(host);
 666         struct vscsifrnt_shadow *shadow, *s = scsi_cmd_priv(sc);
 667         int err = 0;
 668
 669         if (info->host_active == STATE_ERROR)
 670                 return FAILED;
 671
 672         shadow = kzalloc(sizeof(*shadow), GFP_NOIO);
 673         if (!shadow)
 674                 return FAILED;
 675
 676         shadow->act = act;
 677         shadow->rslt_reset = RSLT_RESET_WAITING;
 678         shadow->sc = sc;
 679         shadow->ref_rqid = s->rqid;
 680         init_waitqueue_head(&shadow->wq_reset);
 681
 682         spin_lock_irq(host->host_lock);
 683
 684         for (;;) {
 685                 if (scsifront_enter(info))
 686                         goto fail;
 687
 688                 if (!scsifront_do_request(info, shadow))
 689                         break;
 690
 691                 scsifront_return(info);
 692                 if (err)
 693                         goto fail;
 694                 info->wait_ring_available = 1;
 695                 spin_unlock_irq(host->host_lock);
 696                 err = wait_event_interruptible(info->wq_sync,
 697                                                !info->wait_ring_available);
 698                 spin_lock_irq(host->host_lock);
 699         }
 700
 701         spin_unlock_irq(host->host_lock);
 702         err = wait_event_interruptible(shadow->wq_reset, shadow->wait_reset);
 703         spin_lock_irq(host->host_lock);
 704
 705         if (!err) {
 706                 err = shadow->rslt_reset;
 707                 scsifront_put_rqid(info, shadow->rqid);
 708                 kfree(shadow);
 709         } else {
 710                 spin_lock(&info->shadow_lock);
 711                 shadow->rslt_reset = RSLT_RESET_ERR;
 712                 spin_unlock(&info->shadow_lock);
 713                 err = FAILED;
 714         }
 715
 716         scsifront_return(info);
 717         spin_unlock_irq(host->host_lock);
 718         return err;
 719
 720 fail:
 721         spin_unlock_irq(host->host_lock);
 722         kfree(shadow);
 723         return FAILED;
 724 }
 725
 726 static int scsifront_eh_abort_handler(struct scsi_cmnd *sc)
 727 {
 728         pr_debug("%s\n", __func__);
 729         return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_ABORT);
 730 }
 731
 732 static int scsifront_dev_reset_handler(struct scsi_cmnd *sc)
 733 {
 734         pr_debug("%s\n", __func__);
 735         return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_RESET);
 736 }
 737
 738 static int scsifront_sdev_configure(struct scsi_device *sdev)
 739 {
 740         struct vscsifrnt_info *info = shost_priv(sdev->host);
 741         int err;
 742
 743         if (info->host_active == STATE_ERROR)
 744                 return -EIO;
 745
 746         if (current == info->curr) {
 747                 err = xenbus_printf(XBT_NIL, info->dev->nodename,
 748                               info->dev_state_path, "%d", XenbusStateConnected);
 749                 if (err) {
 750                         xenbus_dev_error(info->dev, err,
 751                                 "%s: writing dev_state_path", __func__);
 752                         return err;
 753                 }
 754         }
 755
 756         return 0;
 757 }
 758
 759 static void scsifront_sdev_destroy(struct scsi_device *sdev)
 760 {
 761         struct vscsifrnt_info *info = shost_priv(sdev->host);
 762         int err;
 763
 764         if (current == info->curr) {
 765                 err = xenbus_printf(XBT_NIL, info->dev->nodename,
 766                               info->dev_state_path, "%d", XenbusStateClosed);
 767                 if (err)
 768                         xenbus_dev_error(info->dev, err,
 769                                 "%s: writing dev_state_path", __func__);
 770         }
 771 }
 772
 773 static const struct scsi_host_template scsifront_sht = {
 774         .module                 = THIS_MODULE,
 775         .name                   = "Xen SCSI frontend driver",
 776         .queuecommand           = scsifront_queuecommand,
 777         .eh_abort_handler       = scsifront_eh_abort_handler,
 778         .eh_device_reset_handler = scsifront_dev_reset_handler,
 779         .slave_configure        = scsifront_sdev_configure,
 780         .slave_destroy          = scsifront_sdev_destroy,
 781         .cmd_per_lun            = VSCSIIF_DEFAULT_CMD_PER_LUN,
 782         .can_queue              = VSCSIIF_MAX_REQS,
 783         .this_id                = -1,
 784         .cmd_size               = sizeof(struct vscsifrnt_shadow),
 785         .sg_tablesize           = VSCSIIF_SG_TABLESIZE,
 786         .proc_name              = "scsifront",
 787 };
 788
 789 static int scsifront_alloc_ring(struct vscsifrnt_info *info)
 790 {
 791         struct xenbus_device *dev = info->dev;
 792         struct vscsiif_sring *sring;
 793         int err;
 794
 795         /***** Frontend to Backend ring start *****/
 796         err = xenbus_setup_ring(dev, GFP_KERNEL, (void **)&sring, 1,
 797                                 &info->ring_ref);
 798         if (err)
 799                 return err;
 800
 801         XEN_FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 802
 803         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 804         if (err) {
 805                 xenbus_dev_fatal(dev, err, "xenbus_alloc_evtchn");
 806                 goto free_gnttab;
 807         }
 808
 809         err = bind_evtchn_to_irq_lateeoi(info->evtchn);
 810         if (err <= 0) {
 811                 xenbus_dev_fatal(dev, err, "bind_evtchn_to_irq");
 812                 goto free_gnttab;
 813         }
 814
 815         info->irq = err;
 816
 817         err = request_threaded_irq(info->irq, NULL, scsifront_irq_fn,
 818                                    IRQF_ONESHOT, "scsifront", info);
 819         if (err) {
 820                 xenbus_dev_fatal(dev, err, "request_threaded_irq");
 821                 goto free_irq;
 822         }
 823
 824         return 0;
 825
 826 /* free resource */
 827 free_irq:
 828         unbind_from_irqhandler(info->irq, info);
 829 free_gnttab:
 830         xenbus_teardown_ring((void **)&sring, 1, &info->ring_ref);
 831
 832         return err;
 833 }
 834
 835 static void scsifront_free_ring(struct vscsifrnt_info *info)
 836 {
 837         unbind_from_irqhandler(info->irq, info);
 838         xenbus_teardown_ring((void **)&info->ring.sring, 1, &info->ring_ref);
 839 }
 840
 841 static int scsifront_init_ring(struct vscsifrnt_info *info)
 842 {
 843         struct xenbus_device *dev = info->dev;
 844         struct xenbus_transaction xbt;
 845         int err;
 846
 847         pr_debug("%s\n", __func__);
 848
 849         err = scsifront_alloc_ring(info);
 850         if (err)
 851                 return err;
 852         pr_debug("%s: %u %u\n", __func__, info->ring_ref, info->evtchn);
 853
 854 again:
 855         err = xenbus_transaction_start(&xbt);
 856         if (err)
 857                 xenbus_dev_fatal(dev, err, "starting transaction");
 858
 859         err = xenbus_printf(xbt, dev->nodename, "ring-ref", "%u",
 860                             info->ring_ref);
 861         if (err) {
 862                 xenbus_dev_fatal(dev, err, "%s", "writing ring-ref");
 863                 goto fail;
 864         }
 865
 866         err = xenbus_printf(xbt, dev->nodename, "event-channel", "%u",
 867                             info->evtchn);
 868
 869         if (err) {
 870                 xenbus_dev_fatal(dev, err, "%s", "writing event-channel");
 871                 goto fail;
 872         }
 873
 874         err = xenbus_transaction_end(xbt, 0);
 875         if (err) {
 876                 if (err == -EAGAIN)
 877                         goto again;
 878                 xenbus_dev_fatal(dev, err, "completing transaction");
 879                 goto free_sring;
 880         }
 881
 882         return 0;
 883
 884 fail:
 885         xenbus_transaction_end(xbt, 1);
 886 free_sring:
 887         scsifront_free_ring(info);
 888
 889         return err;
 890 }
 891
 892
 893 static int scsifront_probe(struct xenbus_device *dev,
 894                            const struct xenbus_device_id *id)
 895 {
 896         struct vscsifrnt_info *info;
 897         struct Scsi_Host *host;
 898         int err = -ENOMEM;
 899         char name[TASK_COMM_LEN];
 900
 901         host = scsi_host_alloc(&scsifront_sht, sizeof(*info));
 902         if (!host) {
 903                 xenbus_dev_fatal(dev, err, "fail to allocate scsi host");
 904                 return err;
 905         }
 906         info = shost_priv(host);
 907
 908         dev_set_drvdata(&dev->dev, info);
 909         info->dev = dev;
 910
 911         bitmap_fill(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
 912
 913         err = scsifront_init_ring(info);
 914         if (err) {
 915                 scsi_host_put(host);
 916                 return err;
 917         }
 918
 919         init_waitqueue_head(&info->wq_sync);
 920         init_waitqueue_head(&info->wq_pause);
 921         spin_lock_init(&info->shadow_lock);
 922
 923         snprintf(name, TASK_COMM_LEN, "vscsiif.%d", host->host_no);
 924
 925         host->max_id      = VSCSIIF_MAX_TARGET;
 926         host->max_channel = 0;
 927         host->max_lun     = VSCSIIF_MAX_LUN;
 928         host->max_sectors = (host->sg_tablesize - 1) * PAGE_SIZE / 512;
 929         host->max_cmd_len = VSCSIIF_MAX_COMMAND_SIZE;
 930
 931         err = scsi_add_host(host, &dev->dev);
 932         if (err) {
 933                 dev_err(&dev->dev, "fail to add scsi host %d\n", err);
 934                 goto free_sring;
 935         }
 936         info->host = host;
 937         info->host_active = STATE_ACTIVE;
 938
 939         xenbus_switch_state(dev, XenbusStateInitialised);
 940
 941         return 0;
 942
 943 free_sring:
 944         scsifront_free_ring(info);
 945         scsi_host_put(host);
 946         return err;
 947 }
 948
 949 static int scsifront_resume(struct xenbus_device *dev)
 950 {
 951         struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
 952         struct Scsi_Host *host = info->host;
 953         int err;
 954
 955         spin_lock_irq(host->host_lock);
 956
 957         /* Finish all still pending commands. */
 958         scsifront_finish_all(info);
 959
 960         spin_unlock_irq(host->host_lock);
 961
 962         /* Reconnect to dom0. */
 963         scsifront_free_ring(info);
 964         err = scsifront_init_ring(info);
 965         if (err) {
 966                 dev_err(&dev->dev, "fail to resume %d\n", err);
 967                 scsi_host_put(host);
 968                 return err;
 969         }
 970
 971         xenbus_switch_state(dev, XenbusStateInitialised);
 972
 973         return 0;
 974 }
 975
 976 static int scsifront_suspend(struct xenbus_device *dev)
 977 {
 978         struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
 979         struct Scsi_Host *host = info->host;
 980         int err = 0;
 981
 982         /* No new commands for the backend. */
 983         spin_lock_irq(host->host_lock);
 984         info->pause = 1;
 985         while (info->callers && !err) {
 986                 info->waiting_pause = 1;
 987                 info->wait_ring_available = 0;
 988                 spin_unlock_irq(host->host_lock);
 989                 wake_up(&info->wq_sync);
 990                 err = wait_event_interruptible(info->wq_pause,
 991                                                !info->waiting_pause);
 992                 spin_lock_irq(host->host_lock);
 993         }
 994         spin_unlock_irq(host->host_lock);
 995         return err;
 996 }
 997
 998 static void scsifront_remove(struct xenbus_device *dev)
 999 {
1000         struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
1001
1002         pr_debug("%s: %s removed\n", __func__, dev->nodename);
1003
1004         mutex_lock(&scsifront_mutex);
1005         if (info->host_active != STATE_INACTIVE) {
1006                 /* Scsi_host not yet removed */
1007                 scsi_remove_host(info->host);
1008                 info->host_active = STATE_INACTIVE;
1009         }
1010         mutex_unlock(&scsifront_mutex);
1011
1012         scsifront_free_ring(info);
1013         scsi_host_put(info->host);
1014 }
1015
1016 static void scsifront_disconnect(struct vscsifrnt_info *info)
1017 {
1018         struct xenbus_device *dev = info->dev;
1019         struct Scsi_Host *host = info->host;
1020
1021         pr_debug("%s: %s disconnect\n", __func__, dev->nodename);
1022
1023         /*
1024          * When this function is executed, all devices of
1025          * Frontend have been deleted.
1026          * Therefore, it need not block I/O before remove_host.
1027          */
1028
1029         mutex_lock(&scsifront_mutex);
1030         if (info->host_active != STATE_INACTIVE) {
1031                 scsi_remove_host(host);
1032                 info->host_active = STATE_INACTIVE;
1033         }
1034         mutex_unlock(&scsifront_mutex);
1035
1036         xenbus_frontend_closed(dev);
1037 }
1038
1039 static void scsifront_do_lun_hotplug(struct vscsifrnt_info *info, int op)
1040 {
1041         struct xenbus_device *dev = info->dev;
1042         int i, err = 0;
1043         char str[64];
1044         char **dir;
1045         unsigned int dir_n = 0;
1046         unsigned int device_state;
1047         unsigned int hst, chn, tgt, lun;
1048         struct scsi_device *sdev;
1049
1050         if (info->host_active == STATE_ERROR)
1051                 return;
1052
1053         dir = xenbus_directory(XBT_NIL, dev->otherend, "vscsi-devs", &dir_n);
1054         if (IS_ERR(dir))
1055                 return;
1056
1057         /* mark current task as the one allowed to modify device states */
1058         BUG_ON(info->curr);
1059         info->curr = current;
1060
1061         for (i = 0; i < dir_n; i++) {
1062                 /* read status */
1063                 snprintf(str, sizeof(str), "vscsi-devs/%s/state", dir[i]);
1064                 err = xenbus_scanf(XBT_NIL, dev->otherend, str, "%u",
1065                                    &device_state);
1066                 if (XENBUS_EXIST_ERR(err))
1067                         continue;
1068
1069                 /* virtual SCSI device */
1070                 snprintf(str, sizeof(str), "vscsi-devs/%s/v-dev", dir[i]);
1071                 err = xenbus_scanf(XBT_NIL, dev->otherend, str,
1072                                    "%u:%u:%u:%u", &hst, &chn, &tgt, &lun);
1073                 if (XENBUS_EXIST_ERR(err))
1074                         continue;
1075
1076                 /*
1077                  * Front device state path, used in slave_configure called
1078                  * on successfull scsi_add_device, and in slave_destroy called
1079                  * on remove of a device.
1080                  */
1081                 snprintf(info->dev_state_path, sizeof(info->dev_state_path),
1082                          "vscsi-devs/%s/state", dir[i]);
1083
1084                 switch (op) {
1085                 case VSCSIFRONT_OP_ADD_LUN:
1086                         if (device_state != XenbusStateInitialised)
1087                                 break;
1088
1089                         if (scsi_add_device(info->host, chn, tgt, lun)) {
1090                                 dev_err(&dev->dev, "scsi_add_device\n");
1091                                 err = xenbus_printf(XBT_NIL, dev->nodename,
1092                                               info->dev_state_path,
1093                                               "%d", XenbusStateClosed);
1094                                 if (err)
1095                                         xenbus_dev_error(dev, err,
1096                                                 "%s: writing dev_state_path", __func__);
1097                         }
1098                         break;
1099                 case VSCSIFRONT_OP_DEL_LUN:
1100                         if (device_state != XenbusStateClosing)
1101                                 break;
1102
1103                         sdev = scsi_device_lookup(info->host, chn, tgt, lun);
1104                         if (sdev) {
1105                                 scsi_remove_device(sdev);
1106                                 scsi_device_put(sdev);
1107                         }
1108                         break;
1109                 case VSCSIFRONT_OP_READD_LUN:
1110                         if (device_state == XenbusStateConnected) {
1111                                 err = xenbus_printf(XBT_NIL, dev->nodename,
1112                                               info->dev_state_path,
1113                                               "%d", XenbusStateConnected);
1114                                 if (err)
1115                                         xenbus_dev_error(dev, err,
1116                                                 "%s: writing dev_state_path", __func__);
1117                         }
1118                         break;
1119                 default:
1120                         break;
1121                 }
1122         }
1123
1124         info->curr = NULL;
1125
1126         kfree(dir);
1127 }
1128
1129 static void scsifront_read_backend_params(struct xenbus_device *dev,
1130                                           struct vscsifrnt_info *info)
1131 {
1132         unsigned int sg_grant, nr_segs;
1133         struct Scsi_Host *host = info->host;
1134
1135         sg_grant = xenbus_read_unsigned(dev->otherend, "feature-sg-grant", 0);
1136         nr_segs = min_t(unsigned int, sg_grant, SG_ALL);
1137         nr_segs = max_t(unsigned int, nr_segs, VSCSIIF_SG_TABLESIZE);
1138         nr_segs = min_t(unsigned int, nr_segs,
1139                         VSCSIIF_SG_TABLESIZE * PAGE_SIZE /
1140                         sizeof(struct scsiif_request_segment));
1141
1142         if (!info->pause && sg_grant)
1143                 dev_info(&dev->dev, "using up to %d SG entries\n", nr_segs);
1144         else if (info->pause && nr_segs < host->sg_tablesize)
1145                 dev_warn(&dev->dev,
1146                          "SG entries decreased from %d to %u - device may not work properly anymore\n",
1147                          host->sg_tablesize, nr_segs);
1148
1149         host->sg_tablesize = nr_segs;
1150         host->max_sectors = (nr_segs - 1) * PAGE_SIZE / 512;
1151 }
1152
1153 static void scsifront_backend_changed(struct xenbus_device *dev,
1154                                       enum xenbus_state backend_state)
1155 {
1156         struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
1157
1158         pr_debug("%s: %p %u %u\n", __func__, dev, dev->state, backend_state);
1159
1160         switch (backend_state) {
1161         case XenbusStateUnknown:
1162         case XenbusStateInitialising:
1163         case XenbusStateInitWait:
1164         case XenbusStateInitialised:
1165                 break;
1166
1167         case XenbusStateConnected:
1168                 scsifront_read_backend_params(dev, info);
1169
1170                 if (info->pause) {
1171                         scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_READD_LUN);
1172                         xenbus_switch_state(dev, XenbusStateConnected);
1173                         info->pause = 0;
1174                         return;
1175                 }
1176
1177                 if (xenbus_read_driver_state(dev->nodename) ==
1178                     XenbusStateInitialised)
1179                         scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
1180
1181                 if (dev->state != XenbusStateConnected)
1182                         xenbus_switch_state(dev, XenbusStateConnected);
1183                 break;
1184
1185         case XenbusStateClosed:
1186                 if (dev->state == XenbusStateClosed)
1187                         break;
1188                 fallthrough;    /* Missed the backend's Closing state */
1189         case XenbusStateClosing:
1190                 scsifront_disconnect(info);
1191                 break;
1192
1193         case XenbusStateReconfiguring:
1194                 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_DEL_LUN);
1195                 xenbus_switch_state(dev, XenbusStateReconfiguring);
1196                 break;
1197
1198         case XenbusStateReconfigured:
1199                 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
1200                 xenbus_switch_state(dev, XenbusStateConnected);
1201                 break;
1202         }
1203 }
1204
1205 static const struct xenbus_device_id scsifront_ids[] = {
1206         { "vscsi" },
1207         { "" }
1208 };
1209
1210 static struct xenbus_driver scsifront_driver = {
1211         .ids                    = scsifront_ids,
1212         .probe                  = scsifront_probe,
1213         .remove                 = scsifront_remove,
1214         .resume                 = scsifront_resume,
1215         .suspend                = scsifront_suspend,
1216         .otherend_changed       = scsifront_backend_changed,
1217 };
1218
1219 static int __init scsifront_init(void)
1220 {
1221         if (!xen_domain())
1222                 return -ENODEV;
1223
1224         return xenbus_register_frontend(&scsifront_driver);
1225 }
1226 module_init(scsifront_init);
1227
1228 static void __exit scsifront_exit(void)
1229 {
1230         xenbus_unregister_driver(&scsifront_driver);
1231 }
1232 module_exit(scsifront_exit);
1233
1234 MODULE_DESCRIPTION("Xen SCSI frontend driver");
1235 MODULE_LICENSE("GPL");
1236 MODULE_ALIAS("xen:vscsi");
1237 MODULE_AUTHOR("Juergen Gross <jgross@suse.com>");