1 /* 2 * Xen SCSI frontend driver 3 * 4 * Copyright (c) 2008, FUJITSU Limited 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License version 2 8 * as published by the Free Software Foundation; or, when distributed 9 * separately from the Linux kernel or incorporated into other 10 * software packages, subject to the following license: 11 * 12 * Permission is hereby granted, free of charge, to any person obtaining a copy 13 * of this source file (the "Software"), to deal in the Software without 14 * restriction, including without limitation the rights to use, copy, modify, 15 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 16 * and to permit persons to whom the Software is furnished to do so, subject to 17 * the following conditions: 18 * 19 * The above copyright notice and this permission notice shall be included in 20 * all copies or substantial portions of the Software. 21 * 22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 25 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 27 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 28 * IN THE SOFTWARE. 29 */ 30 31 #include <linux/module.h> 32 #include <linux/kernel.h> 33 #include <linux/device.h> 34 #include <linux/wait.h> 35 #include <linux/interrupt.h> 36 #include <linux/mutex.h> 37 #include <linux/spinlock.h> 38 #include <linux/sched.h> 39 #include <linux/blkdev.h> 40 #include <linux/pfn.h> 41 #include <linux/slab.h> 42 #include <linux/bitops.h> 43 44 #include <scsi/scsi_cmnd.h> 45 #include <scsi/scsi_device.h> 46 #include <scsi/scsi.h> 47 #include <scsi/scsi_host.h> 48 49 #include <xen/xen.h> 50 #include <xen/xenbus.h> 51 #include <xen/grant_table.h> 52 #include <xen/events.h> 53 #include <xen/page.h> 54 55 #include <xen/interface/grant_table.h> 56 #include <xen/interface/io/vscsiif.h> 57 #include <xen/interface/io/protocols.h> 58 59 #include <asm/xen/hypervisor.h> 60 61 62 #define GRANT_INVALID_REF 0 63 64 #define VSCSIFRONT_OP_ADD_LUN 1 65 #define VSCSIFRONT_OP_DEL_LUN 2 66 #define VSCSIFRONT_OP_READD_LUN 3 67 68 /* Tuning point. */ 69 #define VSCSIIF_DEFAULT_CMD_PER_LUN 10 70 #define VSCSIIF_MAX_TARGET 64 71 #define VSCSIIF_MAX_LUN 255 72 73 #define VSCSIIF_RING_SIZE __CONST_RING_SIZE(vscsiif, PAGE_SIZE) 74 #define VSCSIIF_MAX_REQS VSCSIIF_RING_SIZE 75 76 #define vscsiif_grants_sg(_sg) (PFN_UP((_sg) * \ 77 sizeof(struct scsiif_request_segment))) 78 79 struct vscsifrnt_shadow { 80 /* command between backend and frontend */ 81 unsigned char act; 82 uint8_t nr_segments; 83 uint16_t rqid; 84 uint16_t ref_rqid; 85 86 unsigned int nr_grants; /* number of grants in gref[] */ 87 struct scsiif_request_segment *sg; /* scatter/gather elements */ 88 struct scsiif_request_segment seg[VSCSIIF_SG_TABLESIZE]; 89 90 /* Do reset or abort function. */ 91 wait_queue_head_t wq_reset; /* reset work queue */ 92 int wait_reset; /* reset work queue condition */ 93 int32_t rslt_reset; /* reset response status: */ 94 /* SUCCESS or FAILED or: */ 95 #define RSLT_RESET_WAITING 0 96 #define RSLT_RESET_ERR -1 97 98 /* Requested struct scsi_cmnd is stored from kernel. */ 99 struct scsi_cmnd *sc; 100 int gref[vscsiif_grants_sg(SG_ALL) + SG_ALL]; 101 }; 102 103 struct vscsifrnt_info { 104 struct xenbus_device *dev; 105 106 struct Scsi_Host *host; 107 int host_active; 108 109 unsigned int evtchn; 110 unsigned int irq; 111 112 grant_ref_t ring_ref; 113 struct vscsiif_front_ring ring; 114 struct vscsiif_response ring_rsp; 115 116 spinlock_t shadow_lock; 117 DECLARE_BITMAP(shadow_free_bitmap, VSCSIIF_MAX_REQS); 118 struct vscsifrnt_shadow *shadow[VSCSIIF_MAX_REQS]; 119 120 /* Following items are protected by the host lock. */ 121 wait_queue_head_t wq_sync; 122 wait_queue_head_t wq_pause; 123 unsigned int wait_ring_available:1; 124 unsigned int waiting_pause:1; 125 unsigned int pause:1; 126 unsigned callers; 127 128 char dev_state_path[64]; 129 struct task_struct *curr; 130 }; 131 132 static DEFINE_MUTEX(scsifront_mutex); 133 134 static void scsifront_wake_up(struct vscsifrnt_info *info) 135 { 136 info->wait_ring_available = 0; 137 wake_up(&info->wq_sync); 138 } 139 140 static int scsifront_get_rqid(struct vscsifrnt_info *info) 141 { 142 unsigned long flags; 143 int free; 144 145 spin_lock_irqsave(&info->shadow_lock, flags); 146 147 free = find_first_bit(info->shadow_free_bitmap, VSCSIIF_MAX_REQS); 148 __clear_bit(free, info->shadow_free_bitmap); 149 150 spin_unlock_irqrestore(&info->shadow_lock, flags); 151 152 return free; 153 } 154 155 static int _scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id) 156 { 157 int empty = bitmap_empty(info->shadow_free_bitmap, VSCSIIF_MAX_REQS); 158 159 __set_bit(id, info->shadow_free_bitmap); 160 info->shadow[id] = NULL; 161 162 return empty || info->wait_ring_available; 163 } 164 165 static void scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id) 166 { 167 unsigned long flags; 168 int kick; 169 170 spin_lock_irqsave(&info->shadow_lock, flags); 171 kick = _scsifront_put_rqid(info, id); 172 spin_unlock_irqrestore(&info->shadow_lock, flags); 173 174 if (kick) 175 scsifront_wake_up(info); 176 } 177 178 static int scsifront_do_request(struct vscsifrnt_info *info, 179 struct vscsifrnt_shadow *shadow) 180 { 181 struct vscsiif_front_ring *ring = &(info->ring); 182 struct vscsiif_request *ring_req; 183 struct scsi_cmnd *sc = shadow->sc; 184 uint32_t id; 185 int i, notify; 186 187 if (RING_FULL(&info->ring)) 188 return -EBUSY; 189 190 id = scsifront_get_rqid(info); /* use id in response */ 191 if (id >= VSCSIIF_MAX_REQS) 192 return -EBUSY; 193 194 info->shadow[id] = shadow; 195 shadow->rqid = id; 196 197 ring_req = RING_GET_REQUEST(&(info->ring), ring->req_prod_pvt); 198 ring->req_prod_pvt++; 199 200 ring_req->rqid = id; 201 ring_req->act = shadow->act; 202 ring_req->ref_rqid = shadow->ref_rqid; 203 ring_req->nr_segments = shadow->nr_segments; 204 205 ring_req->id = sc->device->id; 206 ring_req->lun = sc->device->lun; 207 ring_req->channel = sc->device->channel; 208 ring_req->cmd_len = sc->cmd_len; 209 210 BUG_ON(sc->cmd_len > VSCSIIF_MAX_COMMAND_SIZE); 211 212 memcpy(ring_req->cmnd, sc->cmnd, sc->cmd_len); 213 214 ring_req->sc_data_direction = (uint8_t)sc->sc_data_direction; 215 ring_req->timeout_per_command = sc->request->timeout / HZ; 216 217 for (i = 0; i < (shadow->nr_segments & ~VSCSIIF_SG_GRANT); i++) 218 ring_req->seg[i] = shadow->seg[i]; 219 220 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(ring, notify); 221 if (notify) 222 notify_remote_via_irq(info->irq); 223 224 return 0; 225 } 226 227 static void scsifront_gnttab_done(struct vscsifrnt_info *info, 228 struct vscsifrnt_shadow *shadow) 229 { 230 int i; 231 232 if (shadow->sc->sc_data_direction == DMA_NONE) 233 return; 234 235 for (i = 0; i < shadow->nr_grants; i++) { 236 if (unlikely(gnttab_query_foreign_access(shadow->gref[i]))) { 237 shost_printk(KERN_ALERT, info->host, KBUILD_MODNAME 238 "grant still in use by backend\n"); 239 BUG(); 240 } 241 gnttab_end_foreign_access(shadow->gref[i], 0, 0UL); 242 } 243 244 kfree(shadow->sg); 245 } 246 247 static void scsifront_cdb_cmd_done(struct vscsifrnt_info *info, 248 struct vscsiif_response *ring_rsp) 249 { 250 struct vscsifrnt_shadow *shadow; 251 struct scsi_cmnd *sc; 252 uint32_t id; 253 uint8_t sense_len; 254 255 id = ring_rsp->rqid; 256 shadow = info->shadow[id]; 257 sc = shadow->sc; 258 259 BUG_ON(sc == NULL); 260 261 scsifront_gnttab_done(info, shadow); 262 scsifront_put_rqid(info, id); 263 264 sc->result = ring_rsp->rslt; 265 scsi_set_resid(sc, ring_rsp->residual_len); 266 267 sense_len = min_t(uint8_t, VSCSIIF_SENSE_BUFFERSIZE, 268 ring_rsp->sense_len); 269 270 if (sense_len) 271 memcpy(sc->sense_buffer, ring_rsp->sense_buffer, sense_len); 272 273 sc->scsi_done(sc); 274 } 275 276 static void scsifront_sync_cmd_done(struct vscsifrnt_info *info, 277 struct vscsiif_response *ring_rsp) 278 { 279 uint16_t id = ring_rsp->rqid; 280 unsigned long flags; 281 struct vscsifrnt_shadow *shadow = info->shadow[id]; 282 int kick; 283 284 spin_lock_irqsave(&info->shadow_lock, flags); 285 shadow->wait_reset = 1; 286 switch (shadow->rslt_reset) { 287 case RSLT_RESET_WAITING: 288 shadow->rslt_reset = ring_rsp->rslt; 289 break; 290 case RSLT_RESET_ERR: 291 kick = _scsifront_put_rqid(info, id); 292 spin_unlock_irqrestore(&info->shadow_lock, flags); 293 kfree(shadow); 294 if (kick) 295 scsifront_wake_up(info); 296 return; 297 default: 298 shost_printk(KERN_ERR, info->host, KBUILD_MODNAME 299 "bad reset state %d, possibly leaking %u\n", 300 shadow->rslt_reset, id); 301 break; 302 } 303 spin_unlock_irqrestore(&info->shadow_lock, flags); 304 305 wake_up(&shadow->wq_reset); 306 } 307 308 static void scsifront_do_response(struct vscsifrnt_info *info, 309 struct vscsiif_response *ring_rsp) 310 { 311 if (WARN(ring_rsp->rqid >= VSCSIIF_MAX_REQS || 312 test_bit(ring_rsp->rqid, info->shadow_free_bitmap), 313 "illegal rqid %u returned by backend!\n", ring_rsp->rqid)) 314 return; 315 316 if (info->shadow[ring_rsp->rqid]->act == VSCSIIF_ACT_SCSI_CDB) 317 scsifront_cdb_cmd_done(info, ring_rsp); 318 else 319 scsifront_sync_cmd_done(info, ring_rsp); 320 } 321 322 static int scsifront_ring_drain(struct vscsifrnt_info *info) 323 { 324 struct vscsiif_response *ring_rsp; 325 RING_IDX i, rp; 326 int more_to_do = 0; 327 328 rp = info->ring.sring->rsp_prod; 329 rmb(); /* ordering required respective to dom0 */ 330 for (i = info->ring.rsp_cons; i != rp; i++) { 331 ring_rsp = RING_GET_RESPONSE(&info->ring, i); 332 scsifront_do_response(info, ring_rsp); 333 } 334 335 info->ring.rsp_cons = i; 336 337 if (i != info->ring.req_prod_pvt) 338 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do); 339 else 340 info->ring.sring->rsp_event = i + 1; 341 342 return more_to_do; 343 } 344 345 static int scsifront_cmd_done(struct vscsifrnt_info *info) 346 { 347 int more_to_do; 348 unsigned long flags; 349 350 spin_lock_irqsave(info->host->host_lock, flags); 351 352 more_to_do = scsifront_ring_drain(info); 353 354 info->wait_ring_available = 0; 355 356 spin_unlock_irqrestore(info->host->host_lock, flags); 357 358 wake_up(&info->wq_sync); 359 360 return more_to_do; 361 } 362 363 static irqreturn_t scsifront_irq_fn(int irq, void *dev_id) 364 { 365 struct vscsifrnt_info *info = dev_id; 366 367 while (scsifront_cmd_done(info)) 368 /* Yield point for this unbounded loop. */ 369 cond_resched(); 370 371 return IRQ_HANDLED; 372 } 373 374 static void scsifront_finish_all(struct vscsifrnt_info *info) 375 { 376 unsigned i; 377 struct vscsiif_response resp; 378 379 scsifront_ring_drain(info); 380 381 for (i = 0; i < VSCSIIF_MAX_REQS; i++) { 382 if (test_bit(i, info->shadow_free_bitmap)) 383 continue; 384 resp.rqid = i; 385 resp.sense_len = 0; 386 resp.rslt = DID_RESET << 16; 387 resp.residual_len = 0; 388 scsifront_do_response(info, &resp); 389 } 390 } 391 392 static int map_data_for_request(struct vscsifrnt_info *info, 393 struct scsi_cmnd *sc, 394 struct vscsifrnt_shadow *shadow) 395 { 396 grant_ref_t gref_head; 397 struct page *page; 398 int err, ref, ref_cnt = 0; 399 int grant_ro = (sc->sc_data_direction == DMA_TO_DEVICE); 400 unsigned int i, off, len, bytes; 401 unsigned int data_len = scsi_bufflen(sc); 402 unsigned int data_grants = 0, seg_grants = 0; 403 struct scatterlist *sg; 404 struct scsiif_request_segment *seg; 405 406 if (sc->sc_data_direction == DMA_NONE || !data_len) 407 return 0; 408 409 scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) 410 data_grants += PFN_UP(sg->offset + sg->length); 411 412 if (data_grants > VSCSIIF_SG_TABLESIZE) { 413 if (data_grants > info->host->sg_tablesize) { 414 shost_printk(KERN_ERR, info->host, KBUILD_MODNAME 415 "Unable to map request_buffer for command!\n"); 416 return -E2BIG; 417 } 418 seg_grants = vscsiif_grants_sg(data_grants); 419 shadow->sg = kcalloc(data_grants, 420 sizeof(struct scsiif_request_segment), GFP_ATOMIC); 421 if (!shadow->sg) 422 return -ENOMEM; 423 } 424 seg = shadow->sg ? : shadow->seg; 425 426 err = gnttab_alloc_grant_references(seg_grants + data_grants, 427 &gref_head); 428 if (err) { 429 kfree(shadow->sg); 430 shost_printk(KERN_ERR, info->host, KBUILD_MODNAME 431 "gnttab_alloc_grant_references() error\n"); 432 return -ENOMEM; 433 } 434 435 if (seg_grants) { 436 page = virt_to_page(seg); 437 off = offset_in_page(seg); 438 len = sizeof(struct scsiif_request_segment) * data_grants; 439 while (len > 0) { 440 bytes = min_t(unsigned int, len, PAGE_SIZE - off); 441 442 ref = gnttab_claim_grant_reference(&gref_head); 443 BUG_ON(ref == -ENOSPC); 444 445 gnttab_grant_foreign_access_ref(ref, 446 info->dev->otherend_id, 447 xen_page_to_gfn(page), 1); 448 shadow->gref[ref_cnt] = ref; 449 shadow->seg[ref_cnt].gref = ref; 450 shadow->seg[ref_cnt].offset = (uint16_t)off; 451 shadow->seg[ref_cnt].length = (uint16_t)bytes; 452 453 page++; 454 len -= bytes; 455 off = 0; 456 ref_cnt++; 457 } 458 BUG_ON(seg_grants < ref_cnt); 459 seg_grants = ref_cnt; 460 } 461 462 scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) { 463 page = sg_page(sg); 464 off = sg->offset; 465 len = sg->length; 466 467 while (len > 0 && data_len > 0) { 468 /* 469 * sg sends a scatterlist that is larger than 470 * the data_len it wants transferred for certain 471 * IO sizes. 472 */ 473 bytes = min_t(unsigned int, len, PAGE_SIZE - off); 474 bytes = min(bytes, data_len); 475 476 ref = gnttab_claim_grant_reference(&gref_head); 477 BUG_ON(ref == -ENOSPC); 478 479 gnttab_grant_foreign_access_ref(ref, 480 info->dev->otherend_id, 481 xen_page_to_gfn(page), 482 grant_ro); 483 484 shadow->gref[ref_cnt] = ref; 485 seg->gref = ref; 486 seg->offset = (uint16_t)off; 487 seg->length = (uint16_t)bytes; 488 489 page++; 490 seg++; 491 len -= bytes; 492 data_len -= bytes; 493 off = 0; 494 ref_cnt++; 495 } 496 } 497 498 if (seg_grants) 499 shadow->nr_segments = VSCSIIF_SG_GRANT | seg_grants; 500 else 501 shadow->nr_segments = (uint8_t)ref_cnt; 502 shadow->nr_grants = ref_cnt; 503 504 return 0; 505 } 506 507 static int scsifront_enter(struct vscsifrnt_info *info) 508 { 509 if (info->pause) 510 return 1; 511 info->callers++; 512 return 0; 513 } 514 515 static void scsifront_return(struct vscsifrnt_info *info) 516 { 517 info->callers--; 518 if (info->callers) 519 return; 520 521 if (!info->waiting_pause) 522 return; 523 524 info->waiting_pause = 0; 525 wake_up(&info->wq_pause); 526 } 527 528 static int scsifront_queuecommand(struct Scsi_Host *shost, 529 struct scsi_cmnd *sc) 530 { 531 struct vscsifrnt_info *info = shost_priv(shost); 532 struct vscsifrnt_shadow *shadow = scsi_cmd_priv(sc); 533 unsigned long flags; 534 int err; 535 536 sc->result = 0; 537 538 shadow->sc = sc; 539 shadow->act = VSCSIIF_ACT_SCSI_CDB; 540 541 spin_lock_irqsave(shost->host_lock, flags); 542 if (scsifront_enter(info)) { 543 spin_unlock_irqrestore(shost->host_lock, flags); 544 return SCSI_MLQUEUE_HOST_BUSY; 545 } 546 547 err = map_data_for_request(info, sc, shadow); 548 if (err < 0) { 549 pr_debug("%s: err %d\n", __func__, err); 550 scsifront_return(info); 551 spin_unlock_irqrestore(shost->host_lock, flags); 552 if (err == -ENOMEM) 553 return SCSI_MLQUEUE_HOST_BUSY; 554 sc->result = DID_ERROR << 16; 555 sc->scsi_done(sc); 556 return 0; 557 } 558 559 if (scsifront_do_request(info, shadow)) { 560 scsifront_gnttab_done(info, shadow); 561 goto busy; 562 } 563 564 scsifront_return(info); 565 spin_unlock_irqrestore(shost->host_lock, flags); 566 567 return 0; 568 569 busy: 570 scsifront_return(info); 571 spin_unlock_irqrestore(shost->host_lock, flags); 572 pr_debug("%s: busy\n", __func__); 573 return SCSI_MLQUEUE_HOST_BUSY; 574 } 575 576 /* 577 * Any exception handling (reset or abort) must be forwarded to the backend. 578 * We have to wait until an answer is returned. This answer contains the 579 * result to be returned to the requestor. 580 */ 581 static int scsifront_action_handler(struct scsi_cmnd *sc, uint8_t act) 582 { 583 struct Scsi_Host *host = sc->device->host; 584 struct vscsifrnt_info *info = shost_priv(host); 585 struct vscsifrnt_shadow *shadow, *s = scsi_cmd_priv(sc); 586 int err = 0; 587 588 shadow = kzalloc(sizeof(*shadow), GFP_NOIO); 589 if (!shadow) 590 return FAILED; 591 592 shadow->act = act; 593 shadow->rslt_reset = RSLT_RESET_WAITING; 594 shadow->sc = sc; 595 shadow->ref_rqid = s->rqid; 596 init_waitqueue_head(&shadow->wq_reset); 597 598 spin_lock_irq(host->host_lock); 599 600 for (;;) { 601 if (scsifront_enter(info)) 602 goto fail; 603 604 if (!scsifront_do_request(info, shadow)) 605 break; 606 607 scsifront_return(info); 608 if (err) 609 goto fail; 610 info->wait_ring_available = 1; 611 spin_unlock_irq(host->host_lock); 612 err = wait_event_interruptible(info->wq_sync, 613 !info->wait_ring_available); 614 spin_lock_irq(host->host_lock); 615 } 616 617 spin_unlock_irq(host->host_lock); 618 err = wait_event_interruptible(shadow->wq_reset, shadow->wait_reset); 619 spin_lock_irq(host->host_lock); 620 621 if (!err) { 622 err = shadow->rslt_reset; 623 scsifront_put_rqid(info, shadow->rqid); 624 kfree(shadow); 625 } else { 626 spin_lock(&info->shadow_lock); 627 shadow->rslt_reset = RSLT_RESET_ERR; 628 spin_unlock(&info->shadow_lock); 629 err = FAILED; 630 } 631 632 scsifront_return(info); 633 spin_unlock_irq(host->host_lock); 634 return err; 635 636 fail: 637 spin_unlock_irq(host->host_lock); 638 kfree(shadow); 639 return FAILED; 640 } 641 642 static int scsifront_eh_abort_handler(struct scsi_cmnd *sc) 643 { 644 pr_debug("%s\n", __func__); 645 return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_ABORT); 646 } 647 648 static int scsifront_dev_reset_handler(struct scsi_cmnd *sc) 649 { 650 pr_debug("%s\n", __func__); 651 return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_RESET); 652 } 653 654 static int scsifront_sdev_configure(struct scsi_device *sdev) 655 { 656 struct vscsifrnt_info *info = shost_priv(sdev->host); 657 int err; 658 659 if (info && current == info->curr) { 660 err = xenbus_printf(XBT_NIL, info->dev->nodename, 661 info->dev_state_path, "%d", XenbusStateConnected); 662 if (err) { 663 xenbus_dev_error(info->dev, err, 664 "%s: writing dev_state_path", __func__); 665 return err; 666 } 667 } 668 669 return 0; 670 } 671 672 static void scsifront_sdev_destroy(struct scsi_device *sdev) 673 { 674 struct vscsifrnt_info *info = shost_priv(sdev->host); 675 int err; 676 677 if (info && current == info->curr) { 678 err = xenbus_printf(XBT_NIL, info->dev->nodename, 679 info->dev_state_path, "%d", XenbusStateClosed); 680 if (err) 681 xenbus_dev_error(info->dev, err, 682 "%s: writing dev_state_path", __func__); 683 } 684 } 685 686 static struct scsi_host_template scsifront_sht = { 687 .module = THIS_MODULE, 688 .name = "Xen SCSI frontend driver", 689 .queuecommand = scsifront_queuecommand, 690 .eh_abort_handler = scsifront_eh_abort_handler, 691 .eh_device_reset_handler = scsifront_dev_reset_handler, 692 .slave_configure = scsifront_sdev_configure, 693 .slave_destroy = scsifront_sdev_destroy, 694 .cmd_per_lun = VSCSIIF_DEFAULT_CMD_PER_LUN, 695 .can_queue = VSCSIIF_MAX_REQS, 696 .this_id = -1, 697 .cmd_size = sizeof(struct vscsifrnt_shadow), 698 .sg_tablesize = VSCSIIF_SG_TABLESIZE, 699 .use_clustering = DISABLE_CLUSTERING, 700 .proc_name = "scsifront", 701 }; 702 703 static int scsifront_alloc_ring(struct vscsifrnt_info *info) 704 { 705 struct xenbus_device *dev = info->dev; 706 struct vscsiif_sring *sring; 707 grant_ref_t gref; 708 int err = -ENOMEM; 709 710 /***** Frontend to Backend ring start *****/ 711 sring = (struct vscsiif_sring *)__get_free_page(GFP_KERNEL); 712 if (!sring) { 713 xenbus_dev_fatal(dev, err, 714 "fail to allocate shared ring (Front to Back)"); 715 return err; 716 } 717 SHARED_RING_INIT(sring); 718 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE); 719 720 err = xenbus_grant_ring(dev, sring, 1, &gref); 721 if (err < 0) { 722 free_page((unsigned long)sring); 723 xenbus_dev_fatal(dev, err, 724 "fail to grant shared ring (Front to Back)"); 725 return err; 726 } 727 info->ring_ref = gref; 728 729 err = xenbus_alloc_evtchn(dev, &info->evtchn); 730 if (err) { 731 xenbus_dev_fatal(dev, err, "xenbus_alloc_evtchn"); 732 goto free_gnttab; 733 } 734 735 err = bind_evtchn_to_irq(info->evtchn); 736 if (err <= 0) { 737 xenbus_dev_fatal(dev, err, "bind_evtchn_to_irq"); 738 goto free_gnttab; 739 } 740 741 info->irq = err; 742 743 err = request_threaded_irq(info->irq, NULL, scsifront_irq_fn, 744 IRQF_ONESHOT, "scsifront", info); 745 if (err) { 746 xenbus_dev_fatal(dev, err, "request_threaded_irq"); 747 goto free_irq; 748 } 749 750 return 0; 751 752 /* free resource */ 753 free_irq: 754 unbind_from_irqhandler(info->irq, info); 755 free_gnttab: 756 gnttab_end_foreign_access(info->ring_ref, 0, 757 (unsigned long)info->ring.sring); 758 759 return err; 760 } 761 762 static void scsifront_free_ring(struct vscsifrnt_info *info) 763 { 764 unbind_from_irqhandler(info->irq, info); 765 gnttab_end_foreign_access(info->ring_ref, 0, 766 (unsigned long)info->ring.sring); 767 } 768 769 static int scsifront_init_ring(struct vscsifrnt_info *info) 770 { 771 struct xenbus_device *dev = info->dev; 772 struct xenbus_transaction xbt; 773 int err; 774 775 pr_debug("%s\n", __func__); 776 777 err = scsifront_alloc_ring(info); 778 if (err) 779 return err; 780 pr_debug("%s: %u %u\n", __func__, info->ring_ref, info->evtchn); 781 782 again: 783 err = xenbus_transaction_start(&xbt); 784 if (err) 785 xenbus_dev_fatal(dev, err, "starting transaction"); 786 787 err = xenbus_printf(xbt, dev->nodename, "ring-ref", "%u", 788 info->ring_ref); 789 if (err) { 790 xenbus_dev_fatal(dev, err, "%s", "writing ring-ref"); 791 goto fail; 792 } 793 794 err = xenbus_printf(xbt, dev->nodename, "event-channel", "%u", 795 info->evtchn); 796 797 if (err) { 798 xenbus_dev_fatal(dev, err, "%s", "writing event-channel"); 799 goto fail; 800 } 801 802 err = xenbus_transaction_end(xbt, 0); 803 if (err) { 804 if (err == -EAGAIN) 805 goto again; 806 xenbus_dev_fatal(dev, err, "completing transaction"); 807 goto free_sring; 808 } 809 810 return 0; 811 812 fail: 813 xenbus_transaction_end(xbt, 1); 814 free_sring: 815 scsifront_free_ring(info); 816 817 return err; 818 } 819 820 821 static int scsifront_probe(struct xenbus_device *dev, 822 const struct xenbus_device_id *id) 823 { 824 struct vscsifrnt_info *info; 825 struct Scsi_Host *host; 826 int err = -ENOMEM; 827 char name[TASK_COMM_LEN]; 828 829 host = scsi_host_alloc(&scsifront_sht, sizeof(*info)); 830 if (!host) { 831 xenbus_dev_fatal(dev, err, "fail to allocate scsi host"); 832 return err; 833 } 834 info = (struct vscsifrnt_info *)host->hostdata; 835 836 dev_set_drvdata(&dev->dev, info); 837 info->dev = dev; 838 839 bitmap_fill(info->shadow_free_bitmap, VSCSIIF_MAX_REQS); 840 841 err = scsifront_init_ring(info); 842 if (err) { 843 scsi_host_put(host); 844 return err; 845 } 846 847 init_waitqueue_head(&info->wq_sync); 848 init_waitqueue_head(&info->wq_pause); 849 spin_lock_init(&info->shadow_lock); 850 851 snprintf(name, TASK_COMM_LEN, "vscsiif.%d", host->host_no); 852 853 host->max_id = VSCSIIF_MAX_TARGET; 854 host->max_channel = 0; 855 host->max_lun = VSCSIIF_MAX_LUN; 856 host->max_sectors = (host->sg_tablesize - 1) * PAGE_SIZE / 512; 857 host->max_cmd_len = VSCSIIF_MAX_COMMAND_SIZE; 858 859 err = scsi_add_host(host, &dev->dev); 860 if (err) { 861 dev_err(&dev->dev, "fail to add scsi host %d\n", err); 862 goto free_sring; 863 } 864 info->host = host; 865 info->host_active = 1; 866 867 xenbus_switch_state(dev, XenbusStateInitialised); 868 869 return 0; 870 871 free_sring: 872 scsifront_free_ring(info); 873 scsi_host_put(host); 874 return err; 875 } 876 877 static int scsifront_resume(struct xenbus_device *dev) 878 { 879 struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev); 880 struct Scsi_Host *host = info->host; 881 int err; 882 883 spin_lock_irq(host->host_lock); 884 885 /* Finish all still pending commands. */ 886 scsifront_finish_all(info); 887 888 spin_unlock_irq(host->host_lock); 889 890 /* Reconnect to dom0. */ 891 scsifront_free_ring(info); 892 err = scsifront_init_ring(info); 893 if (err) { 894 dev_err(&dev->dev, "fail to resume %d\n", err); 895 scsi_host_put(host); 896 return err; 897 } 898 899 xenbus_switch_state(dev, XenbusStateInitialised); 900 901 return 0; 902 } 903 904 static int scsifront_suspend(struct xenbus_device *dev) 905 { 906 struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev); 907 struct Scsi_Host *host = info->host; 908 int err = 0; 909 910 /* No new commands for the backend. */ 911 spin_lock_irq(host->host_lock); 912 info->pause = 1; 913 while (info->callers && !err) { 914 info->waiting_pause = 1; 915 info->wait_ring_available = 0; 916 spin_unlock_irq(host->host_lock); 917 wake_up(&info->wq_sync); 918 err = wait_event_interruptible(info->wq_pause, 919 !info->waiting_pause); 920 spin_lock_irq(host->host_lock); 921 } 922 spin_unlock_irq(host->host_lock); 923 return err; 924 } 925 926 static int scsifront_remove(struct xenbus_device *dev) 927 { 928 struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev); 929 930 pr_debug("%s: %s removed\n", __func__, dev->nodename); 931 932 mutex_lock(&scsifront_mutex); 933 if (info->host_active) { 934 /* Scsi_host not yet removed */ 935 scsi_remove_host(info->host); 936 info->host_active = 0; 937 } 938 mutex_unlock(&scsifront_mutex); 939 940 scsifront_free_ring(info); 941 scsi_host_put(info->host); 942 943 return 0; 944 } 945 946 static void scsifront_disconnect(struct vscsifrnt_info *info) 947 { 948 struct xenbus_device *dev = info->dev; 949 struct Scsi_Host *host = info->host; 950 951 pr_debug("%s: %s disconnect\n", __func__, dev->nodename); 952 953 /* 954 * When this function is executed, all devices of 955 * Frontend have been deleted. 956 * Therefore, it need not block I/O before remove_host. 957 */ 958 959 mutex_lock(&scsifront_mutex); 960 if (info->host_active) { 961 scsi_remove_host(host); 962 info->host_active = 0; 963 } 964 mutex_unlock(&scsifront_mutex); 965 966 xenbus_frontend_closed(dev); 967 } 968 969 static void scsifront_do_lun_hotplug(struct vscsifrnt_info *info, int op) 970 { 971 struct xenbus_device *dev = info->dev; 972 int i, err = 0; 973 char str[64]; 974 char **dir; 975 unsigned int dir_n = 0; 976 unsigned int device_state; 977 unsigned int hst, chn, tgt, lun; 978 struct scsi_device *sdev; 979 980 dir = xenbus_directory(XBT_NIL, dev->otherend, "vscsi-devs", &dir_n); 981 if (IS_ERR(dir)) 982 return; 983 984 /* mark current task as the one allowed to modify device states */ 985 BUG_ON(info->curr); 986 info->curr = current; 987 988 for (i = 0; i < dir_n; i++) { 989 /* read status */ 990 snprintf(str, sizeof(str), "vscsi-devs/%s/state", dir[i]); 991 err = xenbus_scanf(XBT_NIL, dev->otherend, str, "%u", 992 &device_state); 993 if (XENBUS_EXIST_ERR(err)) 994 continue; 995 996 /* virtual SCSI device */ 997 snprintf(str, sizeof(str), "vscsi-devs/%s/v-dev", dir[i]); 998 err = xenbus_scanf(XBT_NIL, dev->otherend, str, 999 "%u:%u:%u:%u", &hst, &chn, &tgt, &lun); 1000 if (XENBUS_EXIST_ERR(err)) 1001 continue; 1002 1003 /* 1004 * Front device state path, used in slave_configure called 1005 * on successfull scsi_add_device, and in slave_destroy called 1006 * on remove of a device. 1007 */ 1008 snprintf(info->dev_state_path, sizeof(info->dev_state_path), 1009 "vscsi-devs/%s/state", dir[i]); 1010 1011 switch (op) { 1012 case VSCSIFRONT_OP_ADD_LUN: 1013 if (device_state != XenbusStateInitialised) 1014 break; 1015 1016 if (scsi_add_device(info->host, chn, tgt, lun)) { 1017 dev_err(&dev->dev, "scsi_add_device\n"); 1018 err = xenbus_printf(XBT_NIL, dev->nodename, 1019 info->dev_state_path, 1020 "%d", XenbusStateClosed); 1021 if (err) 1022 xenbus_dev_error(dev, err, 1023 "%s: writing dev_state_path", __func__); 1024 } 1025 break; 1026 case VSCSIFRONT_OP_DEL_LUN: 1027 if (device_state != XenbusStateClosing) 1028 break; 1029 1030 sdev = scsi_device_lookup(info->host, chn, tgt, lun); 1031 if (sdev) { 1032 scsi_remove_device(sdev); 1033 scsi_device_put(sdev); 1034 } 1035 break; 1036 case VSCSIFRONT_OP_READD_LUN: 1037 if (device_state == XenbusStateConnected) { 1038 err = xenbus_printf(XBT_NIL, dev->nodename, 1039 info->dev_state_path, 1040 "%d", XenbusStateConnected); 1041 if (err) 1042 xenbus_dev_error(dev, err, 1043 "%s: writing dev_state_path", __func__); 1044 } 1045 break; 1046 default: 1047 break; 1048 } 1049 } 1050 1051 info->curr = NULL; 1052 1053 kfree(dir); 1054 } 1055 1056 static void scsifront_read_backend_params(struct xenbus_device *dev, 1057 struct vscsifrnt_info *info) 1058 { 1059 unsigned int sg_grant, nr_segs; 1060 struct Scsi_Host *host = info->host; 1061 1062 sg_grant = xenbus_read_unsigned(dev->otherend, "feature-sg-grant", 0); 1063 nr_segs = min_t(unsigned int, sg_grant, SG_ALL); 1064 nr_segs = max_t(unsigned int, nr_segs, VSCSIIF_SG_TABLESIZE); 1065 nr_segs = min_t(unsigned int, nr_segs, 1066 VSCSIIF_SG_TABLESIZE * PAGE_SIZE / 1067 sizeof(struct scsiif_request_segment)); 1068 1069 if (!info->pause && sg_grant) 1070 dev_info(&dev->dev, "using up to %d SG entries\n", nr_segs); 1071 else if (info->pause && nr_segs < host->sg_tablesize) 1072 dev_warn(&dev->dev, 1073 "SG entries decreased from %d to %u - device may not work properly anymore\n", 1074 host->sg_tablesize, nr_segs); 1075 1076 host->sg_tablesize = nr_segs; 1077 host->max_sectors = (nr_segs - 1) * PAGE_SIZE / 512; 1078 } 1079 1080 static void scsifront_backend_changed(struct xenbus_device *dev, 1081 enum xenbus_state backend_state) 1082 { 1083 struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev); 1084 1085 pr_debug("%s: %p %u %u\n", __func__, dev, dev->state, backend_state); 1086 1087 switch (backend_state) { 1088 case XenbusStateUnknown: 1089 case XenbusStateInitialising: 1090 case XenbusStateInitWait: 1091 case XenbusStateInitialised: 1092 break; 1093 1094 case XenbusStateConnected: 1095 scsifront_read_backend_params(dev, info); 1096 1097 if (info->pause) { 1098 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_READD_LUN); 1099 xenbus_switch_state(dev, XenbusStateConnected); 1100 info->pause = 0; 1101 return; 1102 } 1103 1104 if (xenbus_read_driver_state(dev->nodename) == 1105 XenbusStateInitialised) 1106 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN); 1107 1108 if (dev->state != XenbusStateConnected) 1109 xenbus_switch_state(dev, XenbusStateConnected); 1110 break; 1111 1112 case XenbusStateClosed: 1113 if (dev->state == XenbusStateClosed) 1114 break; 1115 /* Missed the backend's Closing state -- fallthrough */ 1116 case XenbusStateClosing: 1117 scsifront_disconnect(info); 1118 break; 1119 1120 case XenbusStateReconfiguring: 1121 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_DEL_LUN); 1122 xenbus_switch_state(dev, XenbusStateReconfiguring); 1123 break; 1124 1125 case XenbusStateReconfigured: 1126 scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN); 1127 xenbus_switch_state(dev, XenbusStateConnected); 1128 break; 1129 } 1130 } 1131 1132 static const struct xenbus_device_id scsifront_ids[] = { 1133 { "vscsi" }, 1134 { "" } 1135 }; 1136 1137 static struct xenbus_driver scsifront_driver = { 1138 .ids = scsifront_ids, 1139 .probe = scsifront_probe, 1140 .remove = scsifront_remove, 1141 .resume = scsifront_resume, 1142 .suspend = scsifront_suspend, 1143 .otherend_changed = scsifront_backend_changed, 1144 }; 1145 1146 static int __init scsifront_init(void) 1147 { 1148 if (!xen_domain()) 1149 return -ENODEV; 1150 1151 return xenbus_register_frontend(&scsifront_driver); 1152 } 1153 module_init(scsifront_init); 1154 1155 static void __exit scsifront_exit(void) 1156 { 1157 xenbus_unregister_driver(&scsifront_driver); 1158 } 1159 module_exit(scsifront_exit); 1160 1161 MODULE_DESCRIPTION("Xen SCSI frontend driver"); 1162 MODULE_LICENSE("GPL"); 1163 MODULE_ALIAS("xen:vscsi"); 1164 MODULE_AUTHOR("Juergen Gross <jgross@suse.com>"); 1165