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>"); 1238