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_TARGET_FAILURE: 293 return DID_TARGET_FAILURE; 294 case XEN_VSCSIIF_RSLT_HOST_NEXUS_FAILURE: 295 return DID_NEXUS_FAILURE; 296 case XEN_VSCSIIF_RSLT_HOST_ALLOC_FAILURE: 297 return DID_ALLOC_FAILURE; 298 case XEN_VSCSIIF_RSLT_HOST_MEDIUM_ERROR: 299 return DID_MEDIUM_ERROR; 300 case XEN_VSCSIIF_RSLT_HOST_TRANSPORT_MARGINAL: 301 return DID_TRANSPORT_MARGINAL; 302 default: 303 return DID_ERROR; 304 } 305 } 306 307 static void scsifront_cdb_cmd_done(struct vscsifrnt_info *info, 308 struct vscsiif_response *ring_rsp) 309 { 310 struct vscsifrnt_shadow *shadow; 311 struct scsi_cmnd *sc; 312 uint32_t id; 313 uint8_t sense_len; 314 315 id = ring_rsp->rqid; 316 shadow = info->shadow[id]; 317 sc = shadow->sc; 318 319 BUG_ON(sc == NULL); 320 321 scsifront_gnttab_done(info, shadow); 322 if (info->host_active == STATE_ERROR) 323 return; 324 scsifront_put_rqid(info, id); 325 326 set_host_byte(sc, scsifront_host_byte(ring_rsp->rslt)); 327 set_status_byte(sc, XEN_VSCSIIF_RSLT_STATUS(ring_rsp->rslt)); 328 scsi_set_resid(sc, ring_rsp->residual_len); 329 330 sense_len = min_t(uint8_t, VSCSIIF_SENSE_BUFFERSIZE, 331 ring_rsp->sense_len); 332 333 if (sense_len) 334 memcpy(sc->sense_buffer, ring_rsp->sense_buffer, sense_len); 335 336 scsi_done(sc); 337 } 338 339 static void scsifront_sync_cmd_done(struct vscsifrnt_info *info, 340 struct vscsiif_response *ring_rsp) 341 { 342 uint16_t id = ring_rsp->rqid; 343 unsigned long flags; 344 struct vscsifrnt_shadow *shadow = info->shadow[id]; 345 int kick; 346 347 spin_lock_irqsave(&info->shadow_lock, flags); 348 shadow->wait_reset = 1; 349 switch (shadow->rslt_reset) { 350 case RSLT_RESET_WAITING: 351 if (ring_rsp->rslt == XEN_VSCSIIF_RSLT_RESET_SUCCESS) 352 shadow->rslt_reset = SUCCESS; 353 else 354 shadow->rslt_reset = FAILED; 355 break; 356 case RSLT_RESET_ERR: 357 kick = _scsifront_put_rqid(info, id); 358 spin_unlock_irqrestore(&info->shadow_lock, flags); 359 kfree(shadow); 360 if (kick) 361 scsifront_wake_up(info); 362 return; 363 default: 364 scsifront_set_error(info, "bad reset state"); 365 break; 366 } 367 spin_unlock_irqrestore(&info->shadow_lock, flags); 368 369 wake_up(&shadow->wq_reset); 370 } 371 372 static void scsifront_do_response(struct vscsifrnt_info *info, 373 struct vscsiif_response *ring_rsp) 374 { 375 struct vscsifrnt_shadow *shadow; 376 377 if (ring_rsp->rqid >= VSCSIIF_MAX_REQS || 378 !info->shadow[ring_rsp->rqid]->inflight) { 379 scsifront_set_error(info, "illegal rqid returned by backend!"); 380 return; 381 } 382 shadow = info->shadow[ring_rsp->rqid]; 383 shadow->inflight = false; 384 385 if (shadow->act == VSCSIIF_ACT_SCSI_CDB) 386 scsifront_cdb_cmd_done(info, ring_rsp); 387 else 388 scsifront_sync_cmd_done(info, ring_rsp); 389 } 390 391 static int scsifront_ring_drain(struct vscsifrnt_info *info, 392 unsigned int *eoiflag) 393 { 394 struct vscsiif_response ring_rsp; 395 RING_IDX i, rp; 396 int more_to_do = 0; 397 398 rp = READ_ONCE(info->ring.sring->rsp_prod); 399 virt_rmb(); /* ordering required respective to backend */ 400 if (RING_RESPONSE_PROD_OVERFLOW(&info->ring, rp)) { 401 scsifront_set_error(info, "illegal number of responses"); 402 return 0; 403 } 404 for (i = info->ring.rsp_cons; i != rp; i++) { 405 RING_COPY_RESPONSE(&info->ring, i, &ring_rsp); 406 scsifront_do_response(info, &ring_rsp); 407 if (info->host_active == STATE_ERROR) 408 return 0; 409 *eoiflag &= ~XEN_EOI_FLAG_SPURIOUS; 410 } 411 412 info->ring.rsp_cons = i; 413 414 if (i != info->ring.req_prod_pvt) 415 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do); 416 else 417 info->ring.sring->rsp_event = i + 1; 418 419 return more_to_do; 420 } 421 422 static int scsifront_cmd_done(struct vscsifrnt_info *info, 423 unsigned int *eoiflag) 424 { 425 int more_to_do; 426 unsigned long flags; 427 428 spin_lock_irqsave(info->host->host_lock, flags); 429 430 more_to_do = scsifront_ring_drain(info, eoiflag); 431 432 info->wait_ring_available = 0; 433 434 spin_unlock_irqrestore(info->host->host_lock, flags); 435 436 wake_up(&info->wq_sync); 437 438 return more_to_do; 439 } 440 441 static irqreturn_t scsifront_irq_fn(int irq, void *dev_id) 442 { 443 struct vscsifrnt_info *info = dev_id; 444 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS; 445 446 if (info->host_active == STATE_ERROR) { 447 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS); 448 return IRQ_HANDLED; 449 } 450 451 while (scsifront_cmd_done(info, &eoiflag)) 452 /* Yield point for this unbounded loop. */ 453 cond_resched(); 454 455 xen_irq_lateeoi(irq, eoiflag); 456 457 return IRQ_HANDLED; 458 } 459 460 static void scsifront_finish_all(struct vscsifrnt_info *info) 461 { 462 unsigned int i, dummy; 463 struct vscsiif_response resp; 464 465 scsifront_ring_drain(info, &dummy); 466 467 for (i = 0; i < VSCSIIF_MAX_REQS; i++) { 468 if (test_bit(i, info->shadow_free_bitmap)) 469 continue; 470 resp.rqid = i; 471 resp.sense_len = 0; 472 resp.rslt = DID_RESET << 16; 473 resp.residual_len = 0; 474 scsifront_do_response(info, &resp); 475 } 476 } 477 478 static int map_data_for_request(struct vscsifrnt_info *info, 479 struct scsi_cmnd *sc, 480 struct vscsifrnt_shadow *shadow) 481 { 482 grant_ref_t gref_head; 483 struct page *page; 484 int err, ref, ref_cnt = 0; 485 int grant_ro = (sc->sc_data_direction == DMA_TO_DEVICE); 486 unsigned int i, off, len, bytes; 487 unsigned int data_len = scsi_bufflen(sc); 488 unsigned int data_grants = 0, seg_grants = 0; 489 struct scatterlist *sg; 490 struct scsiif_request_segment *seg; 491 492 if (sc->sc_data_direction == DMA_NONE || !data_len) 493 return 0; 494 495 scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) 496 data_grants += PFN_UP(sg->offset + sg->length); 497 498 if (data_grants > VSCSIIF_SG_TABLESIZE) { 499 if (data_grants > info->host->sg_tablesize) { 500 shost_printk(KERN_ERR, info->host, KBUILD_MODNAME 501 "Unable to map request_buffer for command!\n"); 502 return -E2BIG; 503 } 504 seg_grants = vscsiif_grants_sg(data_grants); 505 shadow->sg = kcalloc(data_grants, 506 sizeof(struct scsiif_request_segment), GFP_ATOMIC); 507 if (!shadow->sg) 508 return -ENOMEM; 509 } 510 seg = shadow->sg ? : shadow->seg; 511 512 err = gnttab_alloc_grant_references(seg_grants + data_grants, 513 &gref_head); 514 if (err) { 515 kfree(shadow->sg); 516 shost_printk(KERN_ERR, info->host, KBUILD_MODNAME 517 "gnttab_alloc_grant_references() error\n"); 518 return -ENOMEM; 519 } 520 521 if (seg_grants) { 522 page = virt_to_page(seg); 523 off = offset_in_page(seg); 524 len = sizeof(struct scsiif_request_segment) * data_grants; 525 while (len > 0) { 526 bytes = min_t(unsigned int, len, PAGE_SIZE - off); 527 528 ref = gnttab_claim_grant_reference(&gref_head); 529 BUG_ON(ref == -ENOSPC); 530 531 gnttab_grant_foreign_access_ref(ref, 532 info->dev->otherend_id, 533 xen_page_to_gfn(page), 1); 534 shadow->gref[ref_cnt] = ref; 535 shadow->seg[ref_cnt].gref = ref; 536 shadow->seg[ref_cnt].offset = (uint16_t)off; 537 shadow->seg[ref_cnt].length = (uint16_t)bytes; 538 539 page++; 540 len -= bytes; 541 off = 0; 542 ref_cnt++; 543 } 544 BUG_ON(seg_grants < ref_cnt); 545 seg_grants = ref_cnt; 546 } 547 548 scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) { 549 page = sg_page(sg); 550 off = sg->offset; 551 len = sg->length; 552 553 while (len > 0 && data_len > 0) { 554 /* 555 * sg sends a scatterlist that is larger than 556 * the data_len it wants transferred for certain 557 * IO sizes. 558 */ 559 bytes = min_t(unsigned int, len, PAGE_SIZE - off); 560 bytes = min(bytes, data_len); 561 562 ref = gnttab_claim_grant_reference(&gref_head); 563 BUG_ON(ref == -ENOSPC); 564 565 gnttab_grant_foreign_access_ref(ref, 566 info->dev->otherend_id, 567 xen_page_to_gfn(page), 568 grant_ro); 569 570 shadow->gref[ref_cnt] = ref; 571 seg->gref = ref; 572 seg->offset = (uint16_t)off; 573 seg->length = (uint16_t)bytes; 574 575 page++; 576 seg++; 577 len -= bytes; 578 data_len -= bytes; 579 off = 0; 580 ref_cnt++; 581 } 582 } 583 584 if (seg_grants) 585 shadow->nr_segments = VSCSIIF_SG_GRANT | seg_grants; 586 else 587 shadow->nr_segments = (uint8_t)ref_cnt; 588 shadow->nr_grants = ref_cnt; 589 590 return 0; 591 } 592 593 static int scsifront_enter(struct vscsifrnt_info *info) 594 { 595 if (info->pause) 596 return 1; 597 info->callers++; 598 return 0; 599 } 600 601 static void scsifront_return(struct vscsifrnt_info *info) 602 { 603 info->callers--; 604 if (info->callers) 605 return; 606 607 if (!info->waiting_pause) 608 return; 609 610 info->waiting_pause = 0; 611 wake_up(&info->wq_pause); 612 } 613 614 static int scsifront_queuecommand(struct Scsi_Host *shost, 615 struct scsi_cmnd *sc) 616 { 617 struct vscsifrnt_info *info = shost_priv(shost); 618 struct vscsifrnt_shadow *shadow = scsi_cmd_priv(sc); 619 unsigned long flags; 620 int err; 621 622 if (info->host_active == STATE_ERROR) 623 return SCSI_MLQUEUE_HOST_BUSY; 624 625 sc->result = 0; 626 627 shadow->sc = sc; 628 shadow->act = VSCSIIF_ACT_SCSI_CDB; 629 630 spin_lock_irqsave(shost->host_lock, flags); 631 if (scsifront_enter(info)) { 632 spin_unlock_irqrestore(shost->host_lock, flags); 633 return SCSI_MLQUEUE_HOST_BUSY; 634 } 635 636 err = map_data_for_request(info, sc, shadow); 637 if (err < 0) { 638 pr_debug("%s: err %d\n", __func__, err); 639 scsifront_return(info); 640 spin_unlock_irqrestore(shost->host_lock, flags); 641 if (err == -ENOMEM) 642 return SCSI_MLQUEUE_HOST_BUSY; 643 sc->result = DID_ERROR << 16; 644 scsi_done(sc); 645 return 0; 646 } 647 648 if (scsifront_do_request(info, shadow)) { 649 scsifront_gnttab_done(info, shadow); 650 goto busy; 651 } 652 653 scsifront_return(info); 654 spin_unlock_irqrestore(shost->host_lock, flags); 655 656 return 0; 657 658 busy: 659 scsifront_return(info); 660 spin_unlock_irqrestore(shost->host_lock, flags); 661 pr_debug("%s: busy\n", __func__); 662 return SCSI_MLQUEUE_HOST_BUSY; 663 } 664 665 /* 666 * Any exception handling (reset or abort) must be forwarded to the backend. 667 * We have to wait until an answer is returned. This answer contains the 668 * result to be returned to the requestor. 669 */ 670 static int scsifront_action_handler(struct scsi_cmnd *sc, uint8_t act) 671 { 672 struct Scsi_Host *host = sc->device->host; 673 struct vscsifrnt_info *info = shost_priv(host); 674 struct vscsifrnt_shadow *shadow, *s = scsi_cmd_priv(sc); 675 int err = 0; 676 677 if (info->host_active == STATE_ERROR) 678 return FAILED; 679 680 shadow = kzalloc(sizeof(*shadow), GFP_NOIO); 681 if (!shadow) 682 return FAILED; 683 684 shadow->act = act; 685 shadow->rslt_reset = RSLT_RESET_WAITING; 686 shadow->sc = sc; 687 shadow->ref_rqid = s->rqid; 688 init_waitqueue_head(&shadow->wq_reset); 689 690 spin_lock_irq(host->host_lock); 691 692 for (;;) { 693 if (scsifront_enter(info)) 694 goto fail; 695 696 if (!scsifront_do_request(info, shadow)) 697 break; 698 699 scsifront_return(info); 700 if (err) 701 goto fail; 702 info->wait_ring_available = 1; 703 spin_unlock_irq(host->host_lock); 704 err = wait_event_interruptible(info->wq_sync, 705 !info->wait_ring_available); 706 spin_lock_irq(host->host_lock); 707 } 708 709 spin_unlock_irq(host->host_lock); 710 err = wait_event_interruptible(shadow->wq_reset, shadow->wait_reset); 711 spin_lock_irq(host->host_lock); 712 713 if (!err) { 714 err = shadow->rslt_reset; 715 scsifront_put_rqid(info, shadow->rqid); 716 kfree(shadow); 717 } else { 718 spin_lock(&info->shadow_lock); 719 shadow->rslt_reset = RSLT_RESET_ERR; 720 spin_unlock(&info->shadow_lock); 721 err = FAILED; 722 } 723 724 scsifront_return(info); 725 spin_unlock_irq(host->host_lock); 726 return err; 727 728 fail: 729 spin_unlock_irq(host->host_lock); 730 kfree(shadow); 731 return FAILED; 732 } 733 734 static int scsifront_eh_abort_handler(struct scsi_cmnd *sc) 735 { 736 pr_debug("%s\n", __func__); 737 return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_ABORT); 738 } 739 740 static int scsifront_dev_reset_handler(struct scsi_cmnd *sc) 741 { 742 pr_debug("%s\n", __func__); 743 return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_RESET); 744 } 745 746 static int scsifront_sdev_configure(struct scsi_device *sdev) 747 { 748 struct vscsifrnt_info *info = shost_priv(sdev->host); 749 int err; 750 751 if (info->host_active == STATE_ERROR) 752 return -EIO; 753 754 if (info && current == info->curr) { 755 err = xenbus_printf(XBT_NIL, info->dev->nodename, 756 info->dev_state_path, "%d", XenbusStateConnected); 757 if (err) { 758 xenbus_dev_error(info->dev, err, 759 "%s: writing dev_state_path", __func__); 760 return err; 761 } 762 } 763 764 return 0; 765 } 766 767 static void scsifront_sdev_destroy(struct scsi_device *sdev) 768 { 769 struct vscsifrnt_info *info = shost_priv(sdev->host); 770 int err; 771 772 if (info && current == info->curr) { 773 err = xenbus_printf(XBT_NIL, info->dev->nodename, 774 info->dev_state_path, "%d", XenbusStateClosed); 775 if (err) 776 xenbus_dev_error(info->dev, err, 777 "%s: writing dev_state_path", __func__); 778 } 779 } 780 781 static struct scsi_host_template scsifront_sht = { 782 .module = THIS_MODULE, 783 .name = "Xen SCSI frontend driver", 784 .queuecommand = scsifront_queuecommand, 785 .eh_abort_handler = scsifront_eh_abort_handler, 786 .eh_device_reset_handler = scsifront_dev_reset_handler, 787 .slave_configure = scsifront_sdev_configure, 788 .slave_destroy = scsifront_sdev_destroy, 789 .cmd_per_lun = VSCSIIF_DEFAULT_CMD_PER_LUN, 790 .can_queue = VSCSIIF_MAX_REQS, 791 .this_id = -1, 792 .cmd_size = sizeof(struct vscsifrnt_shadow), 793 .sg_tablesize = VSCSIIF_SG_TABLESIZE, 794 .proc_name = "scsifront", 795 }; 796 797 static int scsifront_alloc_ring(struct vscsifrnt_info *info) 798 { 799 struct xenbus_device *dev = info->dev; 800 struct vscsiif_sring *sring; 801 grant_ref_t gref; 802 int err = -ENOMEM; 803 804 /***** Frontend to Backend ring start *****/ 805 sring = (struct vscsiif_sring *)__get_free_page(GFP_KERNEL); 806 if (!sring) { 807 xenbus_dev_fatal(dev, err, 808 "fail to allocate shared ring (Front to Back)"); 809 return err; 810 } 811 SHARED_RING_INIT(sring); 812 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE); 813 814 err = xenbus_grant_ring(dev, sring, 1, &gref); 815 if (err < 0) { 816 free_page((unsigned long)sring); 817 xenbus_dev_fatal(dev, err, 818 "fail to grant shared ring (Front to Back)"); 819 return err; 820 } 821 info->ring_ref = gref; 822 823 err = xenbus_alloc_evtchn(dev, &info->evtchn); 824 if (err) { 825 xenbus_dev_fatal(dev, err, "xenbus_alloc_evtchn"); 826 goto free_gnttab; 827 } 828 829 err = bind_evtchn_to_irq_lateeoi(info->evtchn); 830 if (err <= 0) { 831 xenbus_dev_fatal(dev, err, "bind_evtchn_to_irq"); 832 goto free_gnttab; 833 } 834 835 info->irq = err; 836 837 err = request_threaded_irq(info->irq, NULL, scsifront_irq_fn, 838 IRQF_ONESHOT, "scsifront", info); 839 if (err) { 840 xenbus_dev_fatal(dev, err, "request_threaded_irq"); 841 goto free_irq; 842 } 843 844 return 0; 845 846 /* free resource */ 847 free_irq: 848 unbind_from_irqhandler(info->irq, info); 849 free_gnttab: 850 gnttab_end_foreign_access(info->ring_ref, 851 (unsigned long)info->ring.sring); 852 853 return err; 854 } 855 856 static void scsifront_free_ring(struct vscsifrnt_info *info) 857 { 858 unbind_from_irqhandler(info->irq, info); 859 gnttab_end_foreign_access(info->ring_ref, 860 (unsigned long)info->ring.sring); 861 } 862 863 static int scsifront_init_ring(struct vscsifrnt_info *info) 864 { 865 struct xenbus_device *dev = info->dev; 866 struct xenbus_transaction xbt; 867 int err; 868 869 pr_debug("%s\n", __func__); 870 871 err = scsifront_alloc_ring(info); 872 if (err) 873 return err; 874 pr_debug("%s: %u %u\n", __func__, info->ring_ref, info->evtchn); 875 876 again: 877 err = xenbus_transaction_start(&xbt); 878 if (err) 879 xenbus_dev_fatal(dev, err, "starting transaction"); 880 881 err = xenbus_printf(xbt, dev->nodename, "ring-ref", "%u", 882 info->ring_ref); 883 if (err) { 884 xenbus_dev_fatal(dev, err, "%s", "writing ring-ref"); 885 goto fail; 886 } 887 888 err = xenbus_printf(xbt, dev->nodename, "event-channel", "%u", 889 info->evtchn); 890 891 if (err) { 892 xenbus_dev_fatal(dev, err, "%s", "writing event-channel"); 893 goto fail; 894 } 895 896 err = xenbus_transaction_end(xbt, 0); 897 if (err) { 898 if (err == -EAGAIN) 899 goto again; 900 xenbus_dev_fatal(dev, err, "completing transaction"); 901 goto free_sring; 902 } 903 904 return 0; 905 906 fail: 907 xenbus_transaction_end(xbt, 1); 908 free_sring: 909 scsifront_free_ring(info); 910 911 return err; 912 } 913 914 915 static int scsifront_probe(struct xenbus_device *dev, 916 const struct xenbus_device_id *id) 917 { 918 struct vscsifrnt_info *info; 919 struct Scsi_Host *host; 920 int err = -ENOMEM; 921 char name[TASK_COMM_LEN]; 922 923 host = scsi_host_alloc(&scsifront_sht, sizeof(*info)); 924 if (!host) { 925 xenbus_dev_fatal(dev, err, "fail to allocate scsi host"); 926 return err; 927 } 928 info = (struct vscsifrnt_info *)host->hostdata; 929 930 dev_set_drvdata(&dev->dev, info); 931 info->dev = dev; 932 933 bitmap_fill(info->shadow_free_bitmap, VSCSIIF_MAX_REQS); 934 935 err = scsifront_init_ring(info); 936 if (err) { 937 scsi_host_put(host); 938 return err; 939 } 940 941 init_waitqueue_head(&info->wq_sync); 942 init_waitqueue_head(&info->wq_pause); 943 spin_lock_init(&info->shadow_lock); 944 945 snprintf(name, TASK_COMM_LEN, "vscsiif.%d", host->host_no); 946 947 host->max_id = VSCSIIF_MAX_TARGET; 948 host->max_channel = 0; 949 host->max_lun = VSCSIIF_MAX_LUN; 950 host->max_sectors = (host->sg_tablesize - 1) * PAGE_SIZE / 512; 951 host->max_cmd_len = VSCSIIF_MAX_COMMAND_SIZE; 952 953 err = scsi_add_host(host, &dev->dev); 954 if (err) { 955 dev_err(&dev->dev, "fail to add scsi host %d\n", err); 956 goto free_sring; 957 } 958 info->host = host; 959 info->host_active = STATE_ACTIVE; 960 961 xenbus_switch_state(dev, XenbusStateInitialised); 962 963 return 0; 964 965 free_sring: 966 scsifront_free_ring(info); 967 scsi_host_put(host); 968 return err; 969 } 970 971 static int scsifront_resume(struct xenbus_device *dev) 972 { 973 struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev); 974 struct Scsi_Host *host = info->host; 975 int err; 976 977 spin_lock_irq(host->host_lock); 978 979 /* Finish all still pending commands. */ 980 scsifront_finish_all(info); 981 982 spin_unlock_irq(host->host_lock); 983 984 /* Reconnect to dom0. */ 985 scsifront_free_ring(info); 986 err = scsifront_init_ring(info); 987 if (err) { 988 dev_err(&dev->dev, "fail to resume %d\n", err); 989 scsi_host_put(host); 990 return err; 991 } 992 993 xenbus_switch_state(dev, XenbusStateInitialised); 994 995 return 0; 996 } 997 998 static int scsifront_suspend(struct xenbus_device *dev) 999 { 1000 struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev); 1001 struct Scsi_Host *host = info->host; 1002 int err = 0; 1003 1004 /* No new commands for the backend. */ 1005 spin_lock_irq(host->host_lock); 1006 info->pause = 1; 1007 while (info->callers && !err) { 1008 info->waiting_pause = 1; 1009 info->wait_ring_available = 0; 1010 spin_unlock_irq(host->host_lock); 1011 wake_up(&info->wq_sync); 1012 err = wait_event_interruptible(info->wq_pause, 1013 !info->waiting_pause); 1014 spin_lock_irq(host->host_lock); 1015 } 1016 spin_unlock_irq(host->host_lock); 1017 return err; 1018 } 1019 1020 static int scsifront_remove(struct xenbus_device *dev) 1021 { 1022 struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev); 1023 1024 pr_debug("%s: %s removed\n", __func__, dev->nodename); 1025 1026 mutex_lock(&scsifront_mutex); 1027 if (info->host_active != STATE_INACTIVE) { 1028 /* Scsi_host not yet removed */ 1029 scsi_remove_host(info->host); 1030 info->host_active = STATE_INACTIVE; 1031 } 1032 mutex_unlock(&scsifront_mutex); 1033 1034 scsifront_free_ring(info); 1035 scsi_host_put(info->host); 1036 1037 return 0; 1038 } 1039 1040 static void scsifront_disconnect(struct vscsifrnt_info *info) 1041 { 1042 struct xenbus_device *dev = info->dev; 1043 struct Scsi_Host *host = info->host; 1044 1045 pr_debug("%s: %s disconnect\n", __func__, dev->nodename); 1046 1047 /* 1048 * When this function is executed, all devices of 1049 * Frontend have been deleted. 1050 * Therefore, it need not block I/O before remove_host. 1051 */ 1052 1053 mutex_lock(&scsifront_mutex); 1054 if (info->host_active != STATE_INACTIVE) { 1055 scsi_remove_host(host); 1056 info->host_active = STATE_INACTIVE; 1057 } 1058 mutex_unlock(&scsifront_mutex); 1059 1060 xenbus_frontend_closed(dev); 1061 } 1062 1063 static void scsifront_do_lun_hotplug(struct vscsifrnt_info *info, int op) 1064 { 1065 struct xenbus_device *dev = info->dev; 1066 int i, err = 0; 1067 char str[64]; 1068 char **dir; 1069 unsigned int dir_n = 0; 1070 unsigned int device_state; 1071 unsigned int hst, chn, tgt, lun; 1072 struct scsi_device *sdev; 1073 1074 if (info->host_active == STATE_ERROR) 1075 return; 1076 1077 dir = xenbus_directory(XBT_NIL, dev->otherend, "vscsi-devs", &dir_n); 1078 if (IS_ERR(dir)) 1079 return; 1080 1081 /* mark current task as the one allowed to modify device states */ 1082 BUG_ON(info->curr); 1083 info->curr = current; 1084 1085 for (i = 0; i < dir_n; i++) { 1086 /* read status */ 1087 snprintf(str, sizeof(str), "vscsi-devs/%s/state", dir[i]); 1088 err = xenbus_scanf(XBT_NIL, dev->otherend, str, "%u", 1089 &device_state); 1090 if (XENBUS_EXIST_ERR(err)) 1091 continue; 1092 1093 /* virtual SCSI device */ 1094 snprintf(str, sizeof(str), "vscsi-devs/%s/v-dev", dir[i]); 1095 err = xenbus_scanf(XBT_NIL, dev->otherend, str, 1096 "%u:%u:%u:%u", &hst, &chn, &tgt, &lun); 1097 if (XENBUS_EXIST_ERR(err)) 1098 continue; 1099 1100 /* 1101 * Front device state path, used in slave_configure called 1102 * on successfull scsi_add_device, and in slave_destroy called 1103 * on remove of a device. 1104 */ 1105 snprintf(info->dev_state_path, sizeof(info->dev_state_path), 1106 "vscsi-devs/%s/state", dir[i]); 1107 1108 switch (op) { 1109 case VSCSIFRONT_OP_ADD_LUN: 1110 if (device_state != XenbusStateInitialised) 1111 break; 1112 1113 if (scsi_add_device(info->host, chn, tgt, lun)) { 1114 dev_err(&dev->dev, "scsi_add_device\n"); 1115 err = xenbus_printf(XBT_NIL, dev->nodename, 1116 info->dev_state_path, 1117 "%d", XenbusStateClosed); 1118 if (err) 1119 xenbus_dev_error(dev, err, 1120 "%s: writing dev_state_path", __func__); 1121 } 1122 break; 1123 case VSCSIFRONT_OP_DEL_LUN: 1124 if (device_state != XenbusStateClosing) 1125 break; 1126 1127 sdev = scsi_device_lookup(info->host, chn, tgt, lun); 1128 if (sdev) { 1129 scsi_remove_device(sdev); 1130 scsi_device_put(sdev); 1131 } 1132 break; 1133 case VSCSIFRONT_OP_READD_LUN: 1134 if (device_state == XenbusStateConnected) { 1135 err = xenbus_printf(XBT_NIL, dev->nodename, 1136 info->dev_state_path, 1137 "%d", XenbusStateConnected); 1138 if (err) 1139 xenbus_dev_error(dev, err, 1140 "%s: writing dev_state_path", __func__); 1141 } 1142 break; 1143 default: 1144 break; 1145 } 1146 } 1147 1148 info->curr = NULL; 1149 1150 kfree(dir); 1151 } 1152 1153 static void scsifront_read_backend_params(struct xenbus_device *dev, 1154 struct vscsifrnt_info *info) 1155 { 1156 unsigned int sg_grant, nr_segs; 1157 struct Scsi_Host *host = info->host; 1158 1159 sg_grant = xenbus_read_unsigned(dev->otherend, "feature-sg-grant", 0); 1160 nr_segs = min_t(unsigned int, sg_grant, SG_ALL); 1161 nr_segs = max_t(unsigned int, nr_segs, VSCSIIF_SG_TABLESIZE); 1162 nr_segs = min_t(unsigned int, nr_segs, 1163 VSCSIIF_SG_TABLESIZE * PAGE_SIZE / 1164 sizeof(struct scsiif_request_segment)); 1165 1166 if (!info->pause && sg_grant) 1167 dev_info(&dev->dev, "using up to %d SG entries\n", nr_segs); 1168 else if (info->pause && nr_segs < host->sg_tablesize) 1169 dev_warn(&dev->dev, 1170 "SG entries decreased from %d to %u - device may not work properly anymore\n", 1171 host->sg_tablesize, nr_segs); 1172 1173 host->sg_tablesize = nr_segs; 1174 host->max_sectors = (nr_segs - 1) * PAGE_SIZE / 512; 1175 } 1176 1177 static void scsifront_backend_changed(struct xenbus_device *dev, 1178 enum xenbus_state backend_state) 1179 { 1180 struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev); 1181 1182 pr_debug("%s: %p %u %u\n", __func__, dev, dev->state, backend_state); 1183 1184 switch (backend_state) { 1185 case XenbusStateUnknown: 1186 case XenbusStateInitialising: 1187 case XenbusStateInitWait: 1188 case XenbusStateInitialised: 1189 break; 1190 1191 case XenbusStateConnected: 1192 scsifront_read_backend_params(dev, info); 1193 1194 if (info->pause) { 1195 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_READD_LUN); 1196 xenbus_switch_state(dev, XenbusStateConnected); 1197 info->pause = 0; 1198 return; 1199 } 1200 1201 if (xenbus_read_driver_state(dev->nodename) == 1202 XenbusStateInitialised) 1203 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN); 1204 1205 if (dev->state != XenbusStateConnected) 1206 xenbus_switch_state(dev, XenbusStateConnected); 1207 break; 1208 1209 case XenbusStateClosed: 1210 if (dev->state == XenbusStateClosed) 1211 break; 1212 fallthrough; /* Missed the backend's Closing state */ 1213 case XenbusStateClosing: 1214 scsifront_disconnect(info); 1215 break; 1216 1217 case XenbusStateReconfiguring: 1218 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_DEL_LUN); 1219 xenbus_switch_state(dev, XenbusStateReconfiguring); 1220 break; 1221 1222 case XenbusStateReconfigured: 1223 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN); 1224 xenbus_switch_state(dev, XenbusStateConnected); 1225 break; 1226 } 1227 } 1228 1229 static const struct xenbus_device_id scsifront_ids[] = { 1230 { "vscsi" }, 1231 { "" } 1232 }; 1233 1234 static struct xenbus_driver scsifront_driver = { 1235 .ids = scsifront_ids, 1236 .probe = scsifront_probe, 1237 .remove = scsifront_remove, 1238 .resume = scsifront_resume, 1239 .suspend = scsifront_suspend, 1240 .otherend_changed = scsifront_backend_changed, 1241 }; 1242 1243 static int __init scsifront_init(void) 1244 { 1245 if (!xen_domain()) 1246 return -ENODEV; 1247 1248 return xenbus_register_frontend(&scsifront_driver); 1249 } 1250 module_init(scsifront_init); 1251 1252 static void __exit scsifront_exit(void) 1253 { 1254 xenbus_unregister_driver(&scsifront_driver); 1255 } 1256 module_exit(scsifront_exit); 1257 1258 MODULE_DESCRIPTION("Xen SCSI frontend driver"); 1259 MODULE_LICENSE("GPL"); 1260 MODULE_ALIAS("xen:vscsi"); 1261 MODULE_AUTHOR("Juergen Gross <jgross@suse.com>"); 1262