1 /* 2 * blkfront.c 3 * 4 * XenLinux virtual block device driver. 5 * 6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand 7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge 8 * Copyright (c) 2004, Christian Limpach 9 * Copyright (c) 2004, Andrew Warfield 10 * Copyright (c) 2005, Christopher Clark 11 * Copyright (c) 2005, XenSource Ltd 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License version 2 15 * as published by the Free Software Foundation; or, when distributed 16 * separately from the Linux kernel or incorporated into other 17 * software packages, subject to the following license: 18 * 19 * Permission is hereby granted, free of charge, to any person obtaining a copy 20 * of this source file (the "Software"), to deal in the Software without 21 * restriction, including without limitation the rights to use, copy, modify, 22 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 23 * and to permit persons to whom the Software is furnished to do so, subject to 24 * the following conditions: 25 * 26 * The above copyright notice and this permission notice shall be included in 27 * all copies or substantial portions of the Software. 28 * 29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 35 * IN THE SOFTWARE. 36 */ 37 38 #include <linux/interrupt.h> 39 #include <linux/blkdev.h> 40 #include <linux/hdreg.h> 41 #include <linux/cdrom.h> 42 #include <linux/module.h> 43 #include <linux/slab.h> 44 #include <linux/mutex.h> 45 #include <linux/scatterlist.h> 46 #include <linux/bitmap.h> 47 #include <linux/list.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 #include <xen/platform_pci.h> 55 56 #include <xen/interface/grant_table.h> 57 #include <xen/interface/io/blkif.h> 58 #include <xen/interface/io/protocols.h> 59 60 #include <asm/xen/hypervisor.h> 61 62 enum blkif_state { 63 BLKIF_STATE_DISCONNECTED, 64 BLKIF_STATE_CONNECTED, 65 BLKIF_STATE_SUSPENDED, 66 }; 67 68 struct grant { 69 grant_ref_t gref; 70 unsigned long pfn; 71 struct list_head node; 72 }; 73 74 struct blk_shadow { 75 struct blkif_request req; 76 struct request *request; 77 struct grant **grants_used; 78 struct grant **indirect_grants; 79 struct scatterlist *sg; 80 }; 81 82 struct split_bio { 83 struct bio *bio; 84 atomic_t pending; 85 int err; 86 }; 87 88 static DEFINE_MUTEX(blkfront_mutex); 89 static const struct block_device_operations xlvbd_block_fops; 90 91 /* 92 * Maximum number of segments in indirect requests, the actual value used by 93 * the frontend driver is the minimum of this value and the value provided 94 * by the backend driver. 95 */ 96 97 static unsigned int xen_blkif_max_segments = 32; 98 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO); 99 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)"); 100 101 /* 102 * Maximum order of pages to be used for the shared ring between front and 103 * backend, 4KB page granularity is used. 104 */ 105 static unsigned int xen_blkif_max_ring_order; 106 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO); 107 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring"); 108 109 #define BLK_RING_SIZE(info) __CONST_RING_SIZE(blkif, PAGE_SIZE * (info)->nr_ring_pages) 110 #define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * XENBUS_MAX_RING_PAGES) 111 /* 112 * ring-ref%i i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19 113 * characters are enough. Define to 20 to keep consist with backend. 114 */ 115 #define RINGREF_NAME_LEN (20) 116 117 /* 118 * We have one of these per vbd, whether ide, scsi or 'other'. They 119 * hang in private_data off the gendisk structure. We may end up 120 * putting all kinds of interesting stuff here :-) 121 */ 122 struct blkfront_info 123 { 124 spinlock_t io_lock; 125 struct mutex mutex; 126 struct xenbus_device *xbdev; 127 struct gendisk *gd; 128 int vdevice; 129 blkif_vdev_t handle; 130 enum blkif_state connected; 131 int ring_ref[XENBUS_MAX_RING_PAGES]; 132 unsigned int nr_ring_pages; 133 struct blkif_front_ring ring; 134 unsigned int evtchn, irq; 135 struct request_queue *rq; 136 struct work_struct work; 137 struct gnttab_free_callback callback; 138 struct blk_shadow shadow[BLK_MAX_RING_SIZE]; 139 struct list_head grants; 140 struct list_head indirect_pages; 141 unsigned int persistent_gnts_c; 142 unsigned long shadow_free; 143 unsigned int feature_flush; 144 unsigned int feature_discard:1; 145 unsigned int feature_secdiscard:1; 146 unsigned int discard_granularity; 147 unsigned int discard_alignment; 148 unsigned int feature_persistent:1; 149 unsigned int max_indirect_segments; 150 int is_ready; 151 }; 152 153 static unsigned int nr_minors; 154 static unsigned long *minors; 155 static DEFINE_SPINLOCK(minor_lock); 156 157 #define GRANT_INVALID_REF 0 158 159 #define PARTS_PER_DISK 16 160 #define PARTS_PER_EXT_DISK 256 161 162 #define BLKIF_MAJOR(dev) ((dev)>>8) 163 #define BLKIF_MINOR(dev) ((dev) & 0xff) 164 165 #define EXT_SHIFT 28 166 #define EXTENDED (1<<EXT_SHIFT) 167 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED)) 168 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED)) 169 #define EMULATED_HD_DISK_MINOR_OFFSET (0) 170 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256) 171 #define EMULATED_SD_DISK_MINOR_OFFSET (0) 172 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256) 173 174 #define DEV_NAME "xvd" /* name in /dev */ 175 176 #define SEGS_PER_INDIRECT_FRAME \ 177 (PAGE_SIZE/sizeof(struct blkif_request_segment)) 178 #define INDIRECT_GREFS(_segs) \ 179 ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME) 180 181 static int blkfront_setup_indirect(struct blkfront_info *info); 182 183 static int get_id_from_freelist(struct blkfront_info *info) 184 { 185 unsigned long free = info->shadow_free; 186 BUG_ON(free >= BLK_RING_SIZE(info)); 187 info->shadow_free = info->shadow[free].req.u.rw.id; 188 info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */ 189 return free; 190 } 191 192 static int add_id_to_freelist(struct blkfront_info *info, 193 unsigned long id) 194 { 195 if (info->shadow[id].req.u.rw.id != id) 196 return -EINVAL; 197 if (info->shadow[id].request == NULL) 198 return -EINVAL; 199 info->shadow[id].req.u.rw.id = info->shadow_free; 200 info->shadow[id].request = NULL; 201 info->shadow_free = id; 202 return 0; 203 } 204 205 static int fill_grant_buffer(struct blkfront_info *info, int num) 206 { 207 struct page *granted_page; 208 struct grant *gnt_list_entry, *n; 209 int i = 0; 210 211 while(i < num) { 212 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO); 213 if (!gnt_list_entry) 214 goto out_of_memory; 215 216 if (info->feature_persistent) { 217 granted_page = alloc_page(GFP_NOIO); 218 if (!granted_page) { 219 kfree(gnt_list_entry); 220 goto out_of_memory; 221 } 222 gnt_list_entry->pfn = page_to_pfn(granted_page); 223 } 224 225 gnt_list_entry->gref = GRANT_INVALID_REF; 226 list_add(&gnt_list_entry->node, &info->grants); 227 i++; 228 } 229 230 return 0; 231 232 out_of_memory: 233 list_for_each_entry_safe(gnt_list_entry, n, 234 &info->grants, node) { 235 list_del(&gnt_list_entry->node); 236 if (info->feature_persistent) 237 __free_page(pfn_to_page(gnt_list_entry->pfn)); 238 kfree(gnt_list_entry); 239 i--; 240 } 241 BUG_ON(i != 0); 242 return -ENOMEM; 243 } 244 245 static struct grant *get_grant(grant_ref_t *gref_head, 246 unsigned long pfn, 247 struct blkfront_info *info) 248 { 249 struct grant *gnt_list_entry; 250 unsigned long buffer_mfn; 251 252 BUG_ON(list_empty(&info->grants)); 253 gnt_list_entry = list_first_entry(&info->grants, struct grant, 254 node); 255 list_del(&gnt_list_entry->node); 256 257 if (gnt_list_entry->gref != GRANT_INVALID_REF) { 258 info->persistent_gnts_c--; 259 return gnt_list_entry; 260 } 261 262 /* Assign a gref to this page */ 263 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head); 264 BUG_ON(gnt_list_entry->gref == -ENOSPC); 265 if (!info->feature_persistent) { 266 BUG_ON(!pfn); 267 gnt_list_entry->pfn = pfn; 268 } 269 buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn); 270 gnttab_grant_foreign_access_ref(gnt_list_entry->gref, 271 info->xbdev->otherend_id, 272 buffer_mfn, 0); 273 return gnt_list_entry; 274 } 275 276 static const char *op_name(int op) 277 { 278 static const char *const names[] = { 279 [BLKIF_OP_READ] = "read", 280 [BLKIF_OP_WRITE] = "write", 281 [BLKIF_OP_WRITE_BARRIER] = "barrier", 282 [BLKIF_OP_FLUSH_DISKCACHE] = "flush", 283 [BLKIF_OP_DISCARD] = "discard" }; 284 285 if (op < 0 || op >= ARRAY_SIZE(names)) 286 return "unknown"; 287 288 if (!names[op]) 289 return "reserved"; 290 291 return names[op]; 292 } 293 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr) 294 { 295 unsigned int end = minor + nr; 296 int rc; 297 298 if (end > nr_minors) { 299 unsigned long *bitmap, *old; 300 301 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap), 302 GFP_KERNEL); 303 if (bitmap == NULL) 304 return -ENOMEM; 305 306 spin_lock(&minor_lock); 307 if (end > nr_minors) { 308 old = minors; 309 memcpy(bitmap, minors, 310 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap)); 311 minors = bitmap; 312 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG; 313 } else 314 old = bitmap; 315 spin_unlock(&minor_lock); 316 kfree(old); 317 } 318 319 spin_lock(&minor_lock); 320 if (find_next_bit(minors, end, minor) >= end) { 321 bitmap_set(minors, minor, nr); 322 rc = 0; 323 } else 324 rc = -EBUSY; 325 spin_unlock(&minor_lock); 326 327 return rc; 328 } 329 330 static void xlbd_release_minors(unsigned int minor, unsigned int nr) 331 { 332 unsigned int end = minor + nr; 333 334 BUG_ON(end > nr_minors); 335 spin_lock(&minor_lock); 336 bitmap_clear(minors, minor, nr); 337 spin_unlock(&minor_lock); 338 } 339 340 static void blkif_restart_queue_callback(void *arg) 341 { 342 struct blkfront_info *info = (struct blkfront_info *)arg; 343 schedule_work(&info->work); 344 } 345 346 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg) 347 { 348 /* We don't have real geometry info, but let's at least return 349 values consistent with the size of the device */ 350 sector_t nsect = get_capacity(bd->bd_disk); 351 sector_t cylinders = nsect; 352 353 hg->heads = 0xff; 354 hg->sectors = 0x3f; 355 sector_div(cylinders, hg->heads * hg->sectors); 356 hg->cylinders = cylinders; 357 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect) 358 hg->cylinders = 0xffff; 359 return 0; 360 } 361 362 static int blkif_ioctl(struct block_device *bdev, fmode_t mode, 363 unsigned command, unsigned long argument) 364 { 365 struct blkfront_info *info = bdev->bd_disk->private_data; 366 int i; 367 368 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n", 369 command, (long)argument); 370 371 switch (command) { 372 case CDROMMULTISESSION: 373 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n"); 374 for (i = 0; i < sizeof(struct cdrom_multisession); i++) 375 if (put_user(0, (char __user *)(argument + i))) 376 return -EFAULT; 377 return 0; 378 379 case CDROM_GET_CAPABILITY: { 380 struct gendisk *gd = info->gd; 381 if (gd->flags & GENHD_FL_CD) 382 return 0; 383 return -EINVAL; 384 } 385 386 default: 387 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n", 388 command);*/ 389 return -EINVAL; /* same return as native Linux */ 390 } 391 392 return 0; 393 } 394 395 /* 396 * Generate a Xen blkfront IO request from a blk layer request. Reads 397 * and writes are handled as expected. 398 * 399 * @req: a request struct 400 */ 401 static int blkif_queue_request(struct request *req) 402 { 403 struct blkfront_info *info = req->rq_disk->private_data; 404 struct blkif_request *ring_req; 405 unsigned long id; 406 unsigned int fsect, lsect; 407 int i, ref, n; 408 struct blkif_request_segment *segments = NULL; 409 410 /* 411 * Used to store if we are able to queue the request by just using 412 * existing persistent grants, or if we have to get new grants, 413 * as there are not sufficiently many free. 414 */ 415 bool new_persistent_gnts; 416 grant_ref_t gref_head; 417 struct grant *gnt_list_entry = NULL; 418 struct scatterlist *sg; 419 int nseg, max_grefs; 420 421 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) 422 return 1; 423 424 max_grefs = req->nr_phys_segments; 425 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST) 426 /* 427 * If we are using indirect segments we need to account 428 * for the indirect grefs used in the request. 429 */ 430 max_grefs += INDIRECT_GREFS(req->nr_phys_segments); 431 432 /* Check if we have enough grants to allocate a requests */ 433 if (info->persistent_gnts_c < max_grefs) { 434 new_persistent_gnts = 1; 435 if (gnttab_alloc_grant_references( 436 max_grefs - info->persistent_gnts_c, 437 &gref_head) < 0) { 438 gnttab_request_free_callback( 439 &info->callback, 440 blkif_restart_queue_callback, 441 info, 442 max_grefs); 443 return 1; 444 } 445 } else 446 new_persistent_gnts = 0; 447 448 /* Fill out a communications ring structure. */ 449 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt); 450 id = get_id_from_freelist(info); 451 info->shadow[id].request = req; 452 453 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) { 454 ring_req->operation = BLKIF_OP_DISCARD; 455 ring_req->u.discard.nr_sectors = blk_rq_sectors(req); 456 ring_req->u.discard.id = id; 457 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req); 458 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard) 459 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE; 460 else 461 ring_req->u.discard.flag = 0; 462 } else { 463 BUG_ON(info->max_indirect_segments == 0 && 464 req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST); 465 BUG_ON(info->max_indirect_segments && 466 req->nr_phys_segments > info->max_indirect_segments); 467 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg); 468 ring_req->u.rw.id = id; 469 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) { 470 /* 471 * The indirect operation can only be a BLKIF_OP_READ or 472 * BLKIF_OP_WRITE 473 */ 474 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA)); 475 ring_req->operation = BLKIF_OP_INDIRECT; 476 ring_req->u.indirect.indirect_op = rq_data_dir(req) ? 477 BLKIF_OP_WRITE : BLKIF_OP_READ; 478 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req); 479 ring_req->u.indirect.handle = info->handle; 480 ring_req->u.indirect.nr_segments = nseg; 481 } else { 482 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req); 483 ring_req->u.rw.handle = info->handle; 484 ring_req->operation = rq_data_dir(req) ? 485 BLKIF_OP_WRITE : BLKIF_OP_READ; 486 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) { 487 /* 488 * Ideally we can do an unordered flush-to-disk. In case the 489 * backend onlysupports barriers, use that. A barrier request 490 * a superset of FUA, so we can implement it the same 491 * way. (It's also a FLUSH+FUA, since it is 492 * guaranteed ordered WRT previous writes.) 493 */ 494 switch (info->feature_flush & 495 ((REQ_FLUSH|REQ_FUA))) { 496 case REQ_FLUSH|REQ_FUA: 497 ring_req->operation = 498 BLKIF_OP_WRITE_BARRIER; 499 break; 500 case REQ_FLUSH: 501 ring_req->operation = 502 BLKIF_OP_FLUSH_DISKCACHE; 503 break; 504 default: 505 ring_req->operation = 0; 506 } 507 } 508 ring_req->u.rw.nr_segments = nseg; 509 } 510 for_each_sg(info->shadow[id].sg, sg, nseg, i) { 511 fsect = sg->offset >> 9; 512 lsect = fsect + (sg->length >> 9) - 1; 513 514 if ((ring_req->operation == BLKIF_OP_INDIRECT) && 515 (i % SEGS_PER_INDIRECT_FRAME == 0)) { 516 unsigned long uninitialized_var(pfn); 517 518 if (segments) 519 kunmap_atomic(segments); 520 521 n = i / SEGS_PER_INDIRECT_FRAME; 522 if (!info->feature_persistent) { 523 struct page *indirect_page; 524 525 /* Fetch a pre-allocated page to use for indirect grefs */ 526 BUG_ON(list_empty(&info->indirect_pages)); 527 indirect_page = list_first_entry(&info->indirect_pages, 528 struct page, lru); 529 list_del(&indirect_page->lru); 530 pfn = page_to_pfn(indirect_page); 531 } 532 gnt_list_entry = get_grant(&gref_head, pfn, info); 533 info->shadow[id].indirect_grants[n] = gnt_list_entry; 534 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn)); 535 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref; 536 } 537 538 gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info); 539 ref = gnt_list_entry->gref; 540 541 info->shadow[id].grants_used[i] = gnt_list_entry; 542 543 if (rq_data_dir(req) && info->feature_persistent) { 544 char *bvec_data; 545 void *shared_data; 546 547 BUG_ON(sg->offset + sg->length > PAGE_SIZE); 548 549 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn)); 550 bvec_data = kmap_atomic(sg_page(sg)); 551 552 /* 553 * this does not wipe data stored outside the 554 * range sg->offset..sg->offset+sg->length. 555 * Therefore, blkback *could* see data from 556 * previous requests. This is OK as long as 557 * persistent grants are shared with just one 558 * domain. It may need refactoring if this 559 * changes 560 */ 561 memcpy(shared_data + sg->offset, 562 bvec_data + sg->offset, 563 sg->length); 564 565 kunmap_atomic(bvec_data); 566 kunmap_atomic(shared_data); 567 } 568 if (ring_req->operation != BLKIF_OP_INDIRECT) { 569 ring_req->u.rw.seg[i] = 570 (struct blkif_request_segment) { 571 .gref = ref, 572 .first_sect = fsect, 573 .last_sect = lsect }; 574 } else { 575 n = i % SEGS_PER_INDIRECT_FRAME; 576 segments[n] = 577 (struct blkif_request_segment) { 578 .gref = ref, 579 .first_sect = fsect, 580 .last_sect = lsect }; 581 } 582 } 583 if (segments) 584 kunmap_atomic(segments); 585 } 586 587 info->ring.req_prod_pvt++; 588 589 /* Keep a private copy so we can reissue requests when recovering. */ 590 info->shadow[id].req = *ring_req; 591 592 if (new_persistent_gnts) 593 gnttab_free_grant_references(gref_head); 594 595 return 0; 596 } 597 598 599 static inline void flush_requests(struct blkfront_info *info) 600 { 601 int notify; 602 603 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify); 604 605 if (notify) 606 notify_remote_via_irq(info->irq); 607 } 608 609 static inline bool blkif_request_flush_invalid(struct request *req, 610 struct blkfront_info *info) 611 { 612 return ((req->cmd_type != REQ_TYPE_FS) || 613 ((req->cmd_flags & REQ_FLUSH) && 614 !(info->feature_flush & REQ_FLUSH)) || 615 ((req->cmd_flags & REQ_FUA) && 616 !(info->feature_flush & REQ_FUA))); 617 } 618 619 /* 620 * do_blkif_request 621 * read a block; request is in a request queue 622 */ 623 static void do_blkif_request(struct request_queue *rq) 624 { 625 struct blkfront_info *info = NULL; 626 struct request *req; 627 int queued; 628 629 pr_debug("Entered do_blkif_request\n"); 630 631 queued = 0; 632 633 while ((req = blk_peek_request(rq)) != NULL) { 634 info = req->rq_disk->private_data; 635 636 if (RING_FULL(&info->ring)) 637 goto wait; 638 639 blk_start_request(req); 640 641 if (blkif_request_flush_invalid(req, info)) { 642 __blk_end_request_all(req, -EOPNOTSUPP); 643 continue; 644 } 645 646 pr_debug("do_blk_req %p: cmd %p, sec %lx, " 647 "(%u/%u) [%s]\n", 648 req, req->cmd, (unsigned long)blk_rq_pos(req), 649 blk_rq_cur_sectors(req), blk_rq_sectors(req), 650 rq_data_dir(req) ? "write" : "read"); 651 652 if (blkif_queue_request(req)) { 653 blk_requeue_request(rq, req); 654 wait: 655 /* Avoid pointless unplugs. */ 656 blk_stop_queue(rq); 657 break; 658 } 659 660 queued++; 661 } 662 663 if (queued != 0) 664 flush_requests(info); 665 } 666 667 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size, 668 unsigned int physical_sector_size, 669 unsigned int segments) 670 { 671 struct request_queue *rq; 672 struct blkfront_info *info = gd->private_data; 673 674 rq = blk_init_queue(do_blkif_request, &info->io_lock); 675 if (rq == NULL) 676 return -1; 677 678 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq); 679 680 if (info->feature_discard) { 681 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq); 682 blk_queue_max_discard_sectors(rq, get_capacity(gd)); 683 rq->limits.discard_granularity = info->discard_granularity; 684 rq->limits.discard_alignment = info->discard_alignment; 685 if (info->feature_secdiscard) 686 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq); 687 } 688 689 /* Hard sector size and max sectors impersonate the equiv. hardware. */ 690 blk_queue_logical_block_size(rq, sector_size); 691 blk_queue_physical_block_size(rq, physical_sector_size); 692 blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512); 693 694 /* Each segment in a request is up to an aligned page in size. */ 695 blk_queue_segment_boundary(rq, PAGE_SIZE - 1); 696 blk_queue_max_segment_size(rq, PAGE_SIZE); 697 698 /* Ensure a merged request will fit in a single I/O ring slot. */ 699 blk_queue_max_segments(rq, segments); 700 701 /* Make sure buffer addresses are sector-aligned. */ 702 blk_queue_dma_alignment(rq, 511); 703 704 /* Make sure we don't use bounce buffers. */ 705 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY); 706 707 gd->queue = rq; 708 709 return 0; 710 } 711 712 static const char *flush_info(unsigned int feature_flush) 713 { 714 switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) { 715 case REQ_FLUSH|REQ_FUA: 716 return "barrier: enabled;"; 717 case REQ_FLUSH: 718 return "flush diskcache: enabled;"; 719 default: 720 return "barrier or flush: disabled;"; 721 } 722 } 723 724 static void xlvbd_flush(struct blkfront_info *info) 725 { 726 blk_queue_flush(info->rq, info->feature_flush); 727 pr_info("blkfront: %s: %s %s %s %s %s\n", 728 info->gd->disk_name, flush_info(info->feature_flush), 729 "persistent grants:", info->feature_persistent ? 730 "enabled;" : "disabled;", "indirect descriptors:", 731 info->max_indirect_segments ? "enabled;" : "disabled;"); 732 } 733 734 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset) 735 { 736 int major; 737 major = BLKIF_MAJOR(vdevice); 738 *minor = BLKIF_MINOR(vdevice); 739 switch (major) { 740 case XEN_IDE0_MAJOR: 741 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET; 742 *minor = ((*minor / 64) * PARTS_PER_DISK) + 743 EMULATED_HD_DISK_MINOR_OFFSET; 744 break; 745 case XEN_IDE1_MAJOR: 746 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET; 747 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) + 748 EMULATED_HD_DISK_MINOR_OFFSET; 749 break; 750 case XEN_SCSI_DISK0_MAJOR: 751 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET; 752 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET; 753 break; 754 case XEN_SCSI_DISK1_MAJOR: 755 case XEN_SCSI_DISK2_MAJOR: 756 case XEN_SCSI_DISK3_MAJOR: 757 case XEN_SCSI_DISK4_MAJOR: 758 case XEN_SCSI_DISK5_MAJOR: 759 case XEN_SCSI_DISK6_MAJOR: 760 case XEN_SCSI_DISK7_MAJOR: 761 *offset = (*minor / PARTS_PER_DISK) + 762 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) + 763 EMULATED_SD_DISK_NAME_OFFSET; 764 *minor = *minor + 765 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) + 766 EMULATED_SD_DISK_MINOR_OFFSET; 767 break; 768 case XEN_SCSI_DISK8_MAJOR: 769 case XEN_SCSI_DISK9_MAJOR: 770 case XEN_SCSI_DISK10_MAJOR: 771 case XEN_SCSI_DISK11_MAJOR: 772 case XEN_SCSI_DISK12_MAJOR: 773 case XEN_SCSI_DISK13_MAJOR: 774 case XEN_SCSI_DISK14_MAJOR: 775 case XEN_SCSI_DISK15_MAJOR: 776 *offset = (*minor / PARTS_PER_DISK) + 777 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) + 778 EMULATED_SD_DISK_NAME_OFFSET; 779 *minor = *minor + 780 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) + 781 EMULATED_SD_DISK_MINOR_OFFSET; 782 break; 783 case XENVBD_MAJOR: 784 *offset = *minor / PARTS_PER_DISK; 785 break; 786 default: 787 printk(KERN_WARNING "blkfront: your disk configuration is " 788 "incorrect, please use an xvd device instead\n"); 789 return -ENODEV; 790 } 791 return 0; 792 } 793 794 static char *encode_disk_name(char *ptr, unsigned int n) 795 { 796 if (n >= 26) 797 ptr = encode_disk_name(ptr, n / 26 - 1); 798 *ptr = 'a' + n % 26; 799 return ptr + 1; 800 } 801 802 static int xlvbd_alloc_gendisk(blkif_sector_t capacity, 803 struct blkfront_info *info, 804 u16 vdisk_info, u16 sector_size, 805 unsigned int physical_sector_size) 806 { 807 struct gendisk *gd; 808 int nr_minors = 1; 809 int err; 810 unsigned int offset; 811 int minor; 812 int nr_parts; 813 char *ptr; 814 815 BUG_ON(info->gd != NULL); 816 BUG_ON(info->rq != NULL); 817 818 if ((info->vdevice>>EXT_SHIFT) > 1) { 819 /* this is above the extended range; something is wrong */ 820 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice); 821 return -ENODEV; 822 } 823 824 if (!VDEV_IS_EXTENDED(info->vdevice)) { 825 err = xen_translate_vdev(info->vdevice, &minor, &offset); 826 if (err) 827 return err; 828 nr_parts = PARTS_PER_DISK; 829 } else { 830 minor = BLKIF_MINOR_EXT(info->vdevice); 831 nr_parts = PARTS_PER_EXT_DISK; 832 offset = minor / nr_parts; 833 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4) 834 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with " 835 "emulated IDE disks,\n\t choose an xvd device name" 836 "from xvde on\n", info->vdevice); 837 } 838 if (minor >> MINORBITS) { 839 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n", 840 info->vdevice, minor); 841 return -ENODEV; 842 } 843 844 if ((minor % nr_parts) == 0) 845 nr_minors = nr_parts; 846 847 err = xlbd_reserve_minors(minor, nr_minors); 848 if (err) 849 goto out; 850 err = -ENODEV; 851 852 gd = alloc_disk(nr_minors); 853 if (gd == NULL) 854 goto release; 855 856 strcpy(gd->disk_name, DEV_NAME); 857 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset); 858 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN); 859 if (nr_minors > 1) 860 *ptr = 0; 861 else 862 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr, 863 "%d", minor & (nr_parts - 1)); 864 865 gd->major = XENVBD_MAJOR; 866 gd->first_minor = minor; 867 gd->fops = &xlvbd_block_fops; 868 gd->private_data = info; 869 gd->driverfs_dev = &(info->xbdev->dev); 870 set_capacity(gd, capacity); 871 872 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size, 873 info->max_indirect_segments ? : 874 BLKIF_MAX_SEGMENTS_PER_REQUEST)) { 875 del_gendisk(gd); 876 goto release; 877 } 878 879 info->rq = gd->queue; 880 info->gd = gd; 881 882 xlvbd_flush(info); 883 884 if (vdisk_info & VDISK_READONLY) 885 set_disk_ro(gd, 1); 886 887 if (vdisk_info & VDISK_REMOVABLE) 888 gd->flags |= GENHD_FL_REMOVABLE; 889 890 if (vdisk_info & VDISK_CDROM) 891 gd->flags |= GENHD_FL_CD; 892 893 return 0; 894 895 release: 896 xlbd_release_minors(minor, nr_minors); 897 out: 898 return err; 899 } 900 901 static void xlvbd_release_gendisk(struct blkfront_info *info) 902 { 903 unsigned int minor, nr_minors; 904 unsigned long flags; 905 906 if (info->rq == NULL) 907 return; 908 909 spin_lock_irqsave(&info->io_lock, flags); 910 911 /* No more blkif_request(). */ 912 blk_stop_queue(info->rq); 913 914 /* No more gnttab callback work. */ 915 gnttab_cancel_free_callback(&info->callback); 916 spin_unlock_irqrestore(&info->io_lock, flags); 917 918 /* Flush gnttab callback work. Must be done with no locks held. */ 919 flush_work(&info->work); 920 921 del_gendisk(info->gd); 922 923 minor = info->gd->first_minor; 924 nr_minors = info->gd->minors; 925 xlbd_release_minors(minor, nr_minors); 926 927 blk_cleanup_queue(info->rq); 928 info->rq = NULL; 929 930 put_disk(info->gd); 931 info->gd = NULL; 932 } 933 934 static void kick_pending_request_queues(struct blkfront_info *info) 935 { 936 if (!RING_FULL(&info->ring)) { 937 /* Re-enable calldowns. */ 938 blk_start_queue(info->rq); 939 /* Kick things off immediately. */ 940 do_blkif_request(info->rq); 941 } 942 } 943 944 static void blkif_restart_queue(struct work_struct *work) 945 { 946 struct blkfront_info *info = container_of(work, struct blkfront_info, work); 947 948 spin_lock_irq(&info->io_lock); 949 if (info->connected == BLKIF_STATE_CONNECTED) 950 kick_pending_request_queues(info); 951 spin_unlock_irq(&info->io_lock); 952 } 953 954 static void blkif_free(struct blkfront_info *info, int suspend) 955 { 956 struct grant *persistent_gnt; 957 struct grant *n; 958 int i, j, segs; 959 960 /* Prevent new requests being issued until we fix things up. */ 961 spin_lock_irq(&info->io_lock); 962 info->connected = suspend ? 963 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED; 964 /* No more blkif_request(). */ 965 if (info->rq) 966 blk_stop_queue(info->rq); 967 968 /* Remove all persistent grants */ 969 if (!list_empty(&info->grants)) { 970 list_for_each_entry_safe(persistent_gnt, n, 971 &info->grants, node) { 972 list_del(&persistent_gnt->node); 973 if (persistent_gnt->gref != GRANT_INVALID_REF) { 974 gnttab_end_foreign_access(persistent_gnt->gref, 975 0, 0UL); 976 info->persistent_gnts_c--; 977 } 978 if (info->feature_persistent) 979 __free_page(pfn_to_page(persistent_gnt->pfn)); 980 kfree(persistent_gnt); 981 } 982 } 983 BUG_ON(info->persistent_gnts_c != 0); 984 985 /* 986 * Remove indirect pages, this only happens when using indirect 987 * descriptors but not persistent grants 988 */ 989 if (!list_empty(&info->indirect_pages)) { 990 struct page *indirect_page, *n; 991 992 BUG_ON(info->feature_persistent); 993 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) { 994 list_del(&indirect_page->lru); 995 __free_page(indirect_page); 996 } 997 } 998 999 for (i = 0; i < BLK_RING_SIZE(info); i++) { 1000 /* 1001 * Clear persistent grants present in requests already 1002 * on the shared ring 1003 */ 1004 if (!info->shadow[i].request) 1005 goto free_shadow; 1006 1007 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ? 1008 info->shadow[i].req.u.indirect.nr_segments : 1009 info->shadow[i].req.u.rw.nr_segments; 1010 for (j = 0; j < segs; j++) { 1011 persistent_gnt = info->shadow[i].grants_used[j]; 1012 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL); 1013 if (info->feature_persistent) 1014 __free_page(pfn_to_page(persistent_gnt->pfn)); 1015 kfree(persistent_gnt); 1016 } 1017 1018 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT) 1019 /* 1020 * If this is not an indirect operation don't try to 1021 * free indirect segments 1022 */ 1023 goto free_shadow; 1024 1025 for (j = 0; j < INDIRECT_GREFS(segs); j++) { 1026 persistent_gnt = info->shadow[i].indirect_grants[j]; 1027 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL); 1028 __free_page(pfn_to_page(persistent_gnt->pfn)); 1029 kfree(persistent_gnt); 1030 } 1031 1032 free_shadow: 1033 kfree(info->shadow[i].grants_used); 1034 info->shadow[i].grants_used = NULL; 1035 kfree(info->shadow[i].indirect_grants); 1036 info->shadow[i].indirect_grants = NULL; 1037 kfree(info->shadow[i].sg); 1038 info->shadow[i].sg = NULL; 1039 } 1040 1041 /* No more gnttab callback work. */ 1042 gnttab_cancel_free_callback(&info->callback); 1043 spin_unlock_irq(&info->io_lock); 1044 1045 /* Flush gnttab callback work. Must be done with no locks held. */ 1046 flush_work(&info->work); 1047 1048 /* Free resources associated with old device channel. */ 1049 for (i = 0; i < info->nr_ring_pages; i++) { 1050 if (info->ring_ref[i] != GRANT_INVALID_REF) { 1051 gnttab_end_foreign_access(info->ring_ref[i], 0, 0); 1052 info->ring_ref[i] = GRANT_INVALID_REF; 1053 } 1054 } 1055 free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE)); 1056 info->ring.sring = NULL; 1057 1058 if (info->irq) 1059 unbind_from_irqhandler(info->irq, info); 1060 info->evtchn = info->irq = 0; 1061 1062 } 1063 1064 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info, 1065 struct blkif_response *bret) 1066 { 1067 int i = 0; 1068 struct scatterlist *sg; 1069 char *bvec_data; 1070 void *shared_data; 1071 int nseg; 1072 1073 nseg = s->req.operation == BLKIF_OP_INDIRECT ? 1074 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments; 1075 1076 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) { 1077 for_each_sg(s->sg, sg, nseg, i) { 1078 BUG_ON(sg->offset + sg->length > PAGE_SIZE); 1079 shared_data = kmap_atomic( 1080 pfn_to_page(s->grants_used[i]->pfn)); 1081 bvec_data = kmap_atomic(sg_page(sg)); 1082 memcpy(bvec_data + sg->offset, 1083 shared_data + sg->offset, 1084 sg->length); 1085 kunmap_atomic(bvec_data); 1086 kunmap_atomic(shared_data); 1087 } 1088 } 1089 /* Add the persistent grant into the list of free grants */ 1090 for (i = 0; i < nseg; i++) { 1091 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) { 1092 /* 1093 * If the grant is still mapped by the backend (the 1094 * backend has chosen to make this grant persistent) 1095 * we add it at the head of the list, so it will be 1096 * reused first. 1097 */ 1098 if (!info->feature_persistent) 1099 pr_alert_ratelimited("backed has not unmapped grant: %u\n", 1100 s->grants_used[i]->gref); 1101 list_add(&s->grants_used[i]->node, &info->grants); 1102 info->persistent_gnts_c++; 1103 } else { 1104 /* 1105 * If the grant is not mapped by the backend we end the 1106 * foreign access and add it to the tail of the list, 1107 * so it will not be picked again unless we run out of 1108 * persistent grants. 1109 */ 1110 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL); 1111 s->grants_used[i]->gref = GRANT_INVALID_REF; 1112 list_add_tail(&s->grants_used[i]->node, &info->grants); 1113 } 1114 } 1115 if (s->req.operation == BLKIF_OP_INDIRECT) { 1116 for (i = 0; i < INDIRECT_GREFS(nseg); i++) { 1117 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) { 1118 if (!info->feature_persistent) 1119 pr_alert_ratelimited("backed has not unmapped grant: %u\n", 1120 s->indirect_grants[i]->gref); 1121 list_add(&s->indirect_grants[i]->node, &info->grants); 1122 info->persistent_gnts_c++; 1123 } else { 1124 struct page *indirect_page; 1125 1126 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL); 1127 /* 1128 * Add the used indirect page back to the list of 1129 * available pages for indirect grefs. 1130 */ 1131 indirect_page = pfn_to_page(s->indirect_grants[i]->pfn); 1132 list_add(&indirect_page->lru, &info->indirect_pages); 1133 s->indirect_grants[i]->gref = GRANT_INVALID_REF; 1134 list_add_tail(&s->indirect_grants[i]->node, &info->grants); 1135 } 1136 } 1137 } 1138 } 1139 1140 static irqreturn_t blkif_interrupt(int irq, void *dev_id) 1141 { 1142 struct request *req; 1143 struct blkif_response *bret; 1144 RING_IDX i, rp; 1145 unsigned long flags; 1146 struct blkfront_info *info = (struct blkfront_info *)dev_id; 1147 int error; 1148 1149 spin_lock_irqsave(&info->io_lock, flags); 1150 1151 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) { 1152 spin_unlock_irqrestore(&info->io_lock, flags); 1153 return IRQ_HANDLED; 1154 } 1155 1156 again: 1157 rp = info->ring.sring->rsp_prod; 1158 rmb(); /* Ensure we see queued responses up to 'rp'. */ 1159 1160 for (i = info->ring.rsp_cons; i != rp; i++) { 1161 unsigned long id; 1162 1163 bret = RING_GET_RESPONSE(&info->ring, i); 1164 id = bret->id; 1165 /* 1166 * The backend has messed up and given us an id that we would 1167 * never have given to it (we stamp it up to BLK_RING_SIZE - 1168 * look in get_id_from_freelist. 1169 */ 1170 if (id >= BLK_RING_SIZE(info)) { 1171 WARN(1, "%s: response to %s has incorrect id (%ld)\n", 1172 info->gd->disk_name, op_name(bret->operation), id); 1173 /* We can't safely get the 'struct request' as 1174 * the id is busted. */ 1175 continue; 1176 } 1177 req = info->shadow[id].request; 1178 1179 if (bret->operation != BLKIF_OP_DISCARD) 1180 blkif_completion(&info->shadow[id], info, bret); 1181 1182 if (add_id_to_freelist(info, id)) { 1183 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n", 1184 info->gd->disk_name, op_name(bret->operation), id); 1185 continue; 1186 } 1187 1188 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO; 1189 switch (bret->operation) { 1190 case BLKIF_OP_DISCARD: 1191 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) { 1192 struct request_queue *rq = info->rq; 1193 printk(KERN_WARNING "blkfront: %s: %s op failed\n", 1194 info->gd->disk_name, op_name(bret->operation)); 1195 error = -EOPNOTSUPP; 1196 info->feature_discard = 0; 1197 info->feature_secdiscard = 0; 1198 queue_flag_clear(QUEUE_FLAG_DISCARD, rq); 1199 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq); 1200 } 1201 __blk_end_request_all(req, error); 1202 break; 1203 case BLKIF_OP_FLUSH_DISKCACHE: 1204 case BLKIF_OP_WRITE_BARRIER: 1205 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) { 1206 printk(KERN_WARNING "blkfront: %s: %s op failed\n", 1207 info->gd->disk_name, op_name(bret->operation)); 1208 error = -EOPNOTSUPP; 1209 } 1210 if (unlikely(bret->status == BLKIF_RSP_ERROR && 1211 info->shadow[id].req.u.rw.nr_segments == 0)) { 1212 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n", 1213 info->gd->disk_name, op_name(bret->operation)); 1214 error = -EOPNOTSUPP; 1215 } 1216 if (unlikely(error)) { 1217 if (error == -EOPNOTSUPP) 1218 error = 0; 1219 info->feature_flush = 0; 1220 xlvbd_flush(info); 1221 } 1222 /* fall through */ 1223 case BLKIF_OP_READ: 1224 case BLKIF_OP_WRITE: 1225 if (unlikely(bret->status != BLKIF_RSP_OKAY)) 1226 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data " 1227 "request: %x\n", bret->status); 1228 1229 __blk_end_request_all(req, error); 1230 break; 1231 default: 1232 BUG(); 1233 } 1234 } 1235 1236 info->ring.rsp_cons = i; 1237 1238 if (i != info->ring.req_prod_pvt) { 1239 int more_to_do; 1240 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do); 1241 if (more_to_do) 1242 goto again; 1243 } else 1244 info->ring.sring->rsp_event = i + 1; 1245 1246 kick_pending_request_queues(info); 1247 1248 spin_unlock_irqrestore(&info->io_lock, flags); 1249 1250 return IRQ_HANDLED; 1251 } 1252 1253 1254 static int setup_blkring(struct xenbus_device *dev, 1255 struct blkfront_info *info) 1256 { 1257 struct blkif_sring *sring; 1258 int err, i; 1259 unsigned long ring_size = info->nr_ring_pages * PAGE_SIZE; 1260 grant_ref_t gref[XENBUS_MAX_RING_PAGES]; 1261 1262 for (i = 0; i < info->nr_ring_pages; i++) 1263 info->ring_ref[i] = GRANT_INVALID_REF; 1264 1265 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH, 1266 get_order(ring_size)); 1267 if (!sring) { 1268 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring"); 1269 return -ENOMEM; 1270 } 1271 SHARED_RING_INIT(sring); 1272 FRONT_RING_INIT(&info->ring, sring, ring_size); 1273 1274 err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, gref); 1275 if (err < 0) { 1276 free_pages((unsigned long)sring, get_order(ring_size)); 1277 info->ring.sring = NULL; 1278 goto fail; 1279 } 1280 for (i = 0; i < info->nr_ring_pages; i++) 1281 info->ring_ref[i] = gref[i]; 1282 1283 err = xenbus_alloc_evtchn(dev, &info->evtchn); 1284 if (err) 1285 goto fail; 1286 1287 err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0, 1288 "blkif", info); 1289 if (err <= 0) { 1290 xenbus_dev_fatal(dev, err, 1291 "bind_evtchn_to_irqhandler failed"); 1292 goto fail; 1293 } 1294 info->irq = err; 1295 1296 return 0; 1297 fail: 1298 blkif_free(info, 0); 1299 return err; 1300 } 1301 1302 1303 /* Common code used when first setting up, and when resuming. */ 1304 static int talk_to_blkback(struct xenbus_device *dev, 1305 struct blkfront_info *info) 1306 { 1307 const char *message = NULL; 1308 struct xenbus_transaction xbt; 1309 int err, i; 1310 unsigned int max_page_order = 0; 1311 unsigned int ring_page_order = 0; 1312 1313 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, 1314 "max-ring-page-order", "%u", &max_page_order); 1315 if (err != 1) 1316 info->nr_ring_pages = 1; 1317 else { 1318 ring_page_order = min(xen_blkif_max_ring_order, max_page_order); 1319 info->nr_ring_pages = 1 << ring_page_order; 1320 } 1321 1322 /* Create shared ring, alloc event channel. */ 1323 err = setup_blkring(dev, info); 1324 if (err) 1325 goto out; 1326 1327 again: 1328 err = xenbus_transaction_start(&xbt); 1329 if (err) { 1330 xenbus_dev_fatal(dev, err, "starting transaction"); 1331 goto destroy_blkring; 1332 } 1333 1334 if (info->nr_ring_pages == 1) { 1335 err = xenbus_printf(xbt, dev->nodename, 1336 "ring-ref", "%u", info->ring_ref[0]); 1337 if (err) { 1338 message = "writing ring-ref"; 1339 goto abort_transaction; 1340 } 1341 } else { 1342 err = xenbus_printf(xbt, dev->nodename, 1343 "ring-page-order", "%u", ring_page_order); 1344 if (err) { 1345 message = "writing ring-page-order"; 1346 goto abort_transaction; 1347 } 1348 1349 for (i = 0; i < info->nr_ring_pages; i++) { 1350 char ring_ref_name[RINGREF_NAME_LEN]; 1351 1352 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i); 1353 err = xenbus_printf(xbt, dev->nodename, ring_ref_name, 1354 "%u", info->ring_ref[i]); 1355 if (err) { 1356 message = "writing ring-ref"; 1357 goto abort_transaction; 1358 } 1359 } 1360 } 1361 err = xenbus_printf(xbt, dev->nodename, 1362 "event-channel", "%u", info->evtchn); 1363 if (err) { 1364 message = "writing event-channel"; 1365 goto abort_transaction; 1366 } 1367 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s", 1368 XEN_IO_PROTO_ABI_NATIVE); 1369 if (err) { 1370 message = "writing protocol"; 1371 goto abort_transaction; 1372 } 1373 err = xenbus_printf(xbt, dev->nodename, 1374 "feature-persistent", "%u", 1); 1375 if (err) 1376 dev_warn(&dev->dev, 1377 "writing persistent grants feature to xenbus"); 1378 1379 err = xenbus_transaction_end(xbt, 0); 1380 if (err) { 1381 if (err == -EAGAIN) 1382 goto again; 1383 xenbus_dev_fatal(dev, err, "completing transaction"); 1384 goto destroy_blkring; 1385 } 1386 1387 for (i = 0; i < BLK_RING_SIZE(info); i++) 1388 info->shadow[i].req.u.rw.id = i+1; 1389 info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff; 1390 xenbus_switch_state(dev, XenbusStateInitialised); 1391 1392 return 0; 1393 1394 abort_transaction: 1395 xenbus_transaction_end(xbt, 1); 1396 if (message) 1397 xenbus_dev_fatal(dev, err, "%s", message); 1398 destroy_blkring: 1399 blkif_free(info, 0); 1400 out: 1401 return err; 1402 } 1403 1404 /** 1405 * Entry point to this code when a new device is created. Allocate the basic 1406 * structures and the ring buffer for communication with the backend, and 1407 * inform the backend of the appropriate details for those. Switch to 1408 * Initialised state. 1409 */ 1410 static int blkfront_probe(struct xenbus_device *dev, 1411 const struct xenbus_device_id *id) 1412 { 1413 int err, vdevice; 1414 struct blkfront_info *info; 1415 1416 /* FIXME: Use dynamic device id if this is not set. */ 1417 err = xenbus_scanf(XBT_NIL, dev->nodename, 1418 "virtual-device", "%i", &vdevice); 1419 if (err != 1) { 1420 /* go looking in the extended area instead */ 1421 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext", 1422 "%i", &vdevice); 1423 if (err != 1) { 1424 xenbus_dev_fatal(dev, err, "reading virtual-device"); 1425 return err; 1426 } 1427 } 1428 1429 if (xen_hvm_domain()) { 1430 char *type; 1431 int len; 1432 /* no unplug has been done: do not hook devices != xen vbds */ 1433 if (xen_has_pv_and_legacy_disk_devices()) { 1434 int major; 1435 1436 if (!VDEV_IS_EXTENDED(vdevice)) 1437 major = BLKIF_MAJOR(vdevice); 1438 else 1439 major = XENVBD_MAJOR; 1440 1441 if (major != XENVBD_MAJOR) { 1442 printk(KERN_INFO 1443 "%s: HVM does not support vbd %d as xen block device\n", 1444 __func__, vdevice); 1445 return -ENODEV; 1446 } 1447 } 1448 /* do not create a PV cdrom device if we are an HVM guest */ 1449 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len); 1450 if (IS_ERR(type)) 1451 return -ENODEV; 1452 if (strncmp(type, "cdrom", 5) == 0) { 1453 kfree(type); 1454 return -ENODEV; 1455 } 1456 kfree(type); 1457 } 1458 info = kzalloc(sizeof(*info), GFP_KERNEL); 1459 if (!info) { 1460 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure"); 1461 return -ENOMEM; 1462 } 1463 1464 mutex_init(&info->mutex); 1465 spin_lock_init(&info->io_lock); 1466 info->xbdev = dev; 1467 info->vdevice = vdevice; 1468 INIT_LIST_HEAD(&info->grants); 1469 INIT_LIST_HEAD(&info->indirect_pages); 1470 info->persistent_gnts_c = 0; 1471 info->connected = BLKIF_STATE_DISCONNECTED; 1472 INIT_WORK(&info->work, blkif_restart_queue); 1473 1474 /* Front end dir is a number, which is used as the id. */ 1475 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); 1476 dev_set_drvdata(&dev->dev, info); 1477 1478 return 0; 1479 } 1480 1481 static void split_bio_end(struct bio *bio, int error) 1482 { 1483 struct split_bio *split_bio = bio->bi_private; 1484 1485 if (error) 1486 split_bio->err = error; 1487 1488 if (atomic_dec_and_test(&split_bio->pending)) { 1489 split_bio->bio->bi_phys_segments = 0; 1490 bio_endio(split_bio->bio, split_bio->err); 1491 kfree(split_bio); 1492 } 1493 bio_put(bio); 1494 } 1495 1496 static int blkif_recover(struct blkfront_info *info) 1497 { 1498 int i; 1499 struct request *req, *n; 1500 struct blk_shadow *copy; 1501 int rc; 1502 struct bio *bio, *cloned_bio; 1503 struct bio_list bio_list, merge_bio; 1504 unsigned int segs, offset; 1505 int pending, size; 1506 struct split_bio *split_bio; 1507 struct list_head requests; 1508 1509 /* Stage 1: Make a safe copy of the shadow state. */ 1510 copy = kmemdup(info->shadow, sizeof(info->shadow), 1511 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH); 1512 if (!copy) 1513 return -ENOMEM; 1514 1515 /* Stage 2: Set up free list. */ 1516 memset(&info->shadow, 0, sizeof(info->shadow)); 1517 for (i = 0; i < BLK_RING_SIZE(info); i++) 1518 info->shadow[i].req.u.rw.id = i+1; 1519 info->shadow_free = info->ring.req_prod_pvt; 1520 info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff; 1521 1522 rc = blkfront_setup_indirect(info); 1523 if (rc) { 1524 kfree(copy); 1525 return rc; 1526 } 1527 1528 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST; 1529 blk_queue_max_segments(info->rq, segs); 1530 bio_list_init(&bio_list); 1531 INIT_LIST_HEAD(&requests); 1532 for (i = 0; i < BLK_RING_SIZE(info); i++) { 1533 /* Not in use? */ 1534 if (!copy[i].request) 1535 continue; 1536 1537 /* 1538 * Get the bios in the request so we can re-queue them. 1539 */ 1540 if (copy[i].request->cmd_flags & 1541 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) { 1542 /* 1543 * Flush operations don't contain bios, so 1544 * we need to requeue the whole request 1545 */ 1546 list_add(©[i].request->queuelist, &requests); 1547 continue; 1548 } 1549 merge_bio.head = copy[i].request->bio; 1550 merge_bio.tail = copy[i].request->biotail; 1551 bio_list_merge(&bio_list, &merge_bio); 1552 copy[i].request->bio = NULL; 1553 blk_end_request_all(copy[i].request, 0); 1554 } 1555 1556 kfree(copy); 1557 1558 /* 1559 * Empty the queue, this is important because we might have 1560 * requests in the queue with more segments than what we 1561 * can handle now. 1562 */ 1563 spin_lock_irq(&info->io_lock); 1564 while ((req = blk_fetch_request(info->rq)) != NULL) { 1565 if (req->cmd_flags & 1566 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) { 1567 list_add(&req->queuelist, &requests); 1568 continue; 1569 } 1570 merge_bio.head = req->bio; 1571 merge_bio.tail = req->biotail; 1572 bio_list_merge(&bio_list, &merge_bio); 1573 req->bio = NULL; 1574 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) 1575 pr_alert("diskcache flush request found!\n"); 1576 __blk_end_request_all(req, 0); 1577 } 1578 spin_unlock_irq(&info->io_lock); 1579 1580 xenbus_switch_state(info->xbdev, XenbusStateConnected); 1581 1582 spin_lock_irq(&info->io_lock); 1583 1584 /* Now safe for us to use the shared ring */ 1585 info->connected = BLKIF_STATE_CONNECTED; 1586 1587 /* Kick any other new requests queued since we resumed */ 1588 kick_pending_request_queues(info); 1589 1590 list_for_each_entry_safe(req, n, &requests, queuelist) { 1591 /* Requeue pending requests (flush or discard) */ 1592 list_del_init(&req->queuelist); 1593 BUG_ON(req->nr_phys_segments > segs); 1594 blk_requeue_request(info->rq, req); 1595 } 1596 spin_unlock_irq(&info->io_lock); 1597 1598 while ((bio = bio_list_pop(&bio_list)) != NULL) { 1599 /* Traverse the list of pending bios and re-queue them */ 1600 if (bio_segments(bio) > segs) { 1601 /* 1602 * This bio has more segments than what we can 1603 * handle, we have to split it. 1604 */ 1605 pending = (bio_segments(bio) + segs - 1) / segs; 1606 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO); 1607 BUG_ON(split_bio == NULL); 1608 atomic_set(&split_bio->pending, pending); 1609 split_bio->bio = bio; 1610 for (i = 0; i < pending; i++) { 1611 offset = (i * segs * PAGE_SIZE) >> 9; 1612 size = min((unsigned int)(segs * PAGE_SIZE) >> 9, 1613 (unsigned int)bio_sectors(bio) - offset); 1614 cloned_bio = bio_clone(bio, GFP_NOIO); 1615 BUG_ON(cloned_bio == NULL); 1616 bio_trim(cloned_bio, offset, size); 1617 cloned_bio->bi_private = split_bio; 1618 cloned_bio->bi_end_io = split_bio_end; 1619 submit_bio(cloned_bio->bi_rw, cloned_bio); 1620 } 1621 /* 1622 * Now we have to wait for all those smaller bios to 1623 * end, so we can also end the "parent" bio. 1624 */ 1625 continue; 1626 } 1627 /* We don't need to split this bio */ 1628 submit_bio(bio->bi_rw, bio); 1629 } 1630 1631 return 0; 1632 } 1633 1634 /** 1635 * We are reconnecting to the backend, due to a suspend/resume, or a backend 1636 * driver restart. We tear down our blkif structure and recreate it, but 1637 * leave the device-layer structures intact so that this is transparent to the 1638 * rest of the kernel. 1639 */ 1640 static int blkfront_resume(struct xenbus_device *dev) 1641 { 1642 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 1643 int err; 1644 1645 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename); 1646 1647 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED); 1648 1649 err = talk_to_blkback(dev, info); 1650 1651 /* 1652 * We have to wait for the backend to switch to 1653 * connected state, since we want to read which 1654 * features it supports. 1655 */ 1656 1657 return err; 1658 } 1659 1660 static void 1661 blkfront_closing(struct blkfront_info *info) 1662 { 1663 struct xenbus_device *xbdev = info->xbdev; 1664 struct block_device *bdev = NULL; 1665 1666 mutex_lock(&info->mutex); 1667 1668 if (xbdev->state == XenbusStateClosing) { 1669 mutex_unlock(&info->mutex); 1670 return; 1671 } 1672 1673 if (info->gd) 1674 bdev = bdget_disk(info->gd, 0); 1675 1676 mutex_unlock(&info->mutex); 1677 1678 if (!bdev) { 1679 xenbus_frontend_closed(xbdev); 1680 return; 1681 } 1682 1683 mutex_lock(&bdev->bd_mutex); 1684 1685 if (bdev->bd_openers) { 1686 xenbus_dev_error(xbdev, -EBUSY, 1687 "Device in use; refusing to close"); 1688 xenbus_switch_state(xbdev, XenbusStateClosing); 1689 } else { 1690 xlvbd_release_gendisk(info); 1691 xenbus_frontend_closed(xbdev); 1692 } 1693 1694 mutex_unlock(&bdev->bd_mutex); 1695 bdput(bdev); 1696 } 1697 1698 static void blkfront_setup_discard(struct blkfront_info *info) 1699 { 1700 int err; 1701 unsigned int discard_granularity; 1702 unsigned int discard_alignment; 1703 unsigned int discard_secure; 1704 1705 info->feature_discard = 1; 1706 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1707 "discard-granularity", "%u", &discard_granularity, 1708 "discard-alignment", "%u", &discard_alignment, 1709 NULL); 1710 if (!err) { 1711 info->discard_granularity = discard_granularity; 1712 info->discard_alignment = discard_alignment; 1713 } 1714 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1715 "discard-secure", "%d", &discard_secure, 1716 NULL); 1717 if (!err) 1718 info->feature_secdiscard = !!discard_secure; 1719 } 1720 1721 static int blkfront_setup_indirect(struct blkfront_info *info) 1722 { 1723 unsigned int indirect_segments, segs; 1724 int err, i; 1725 1726 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1727 "feature-max-indirect-segments", "%u", &indirect_segments, 1728 NULL); 1729 if (err) { 1730 info->max_indirect_segments = 0; 1731 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST; 1732 } else { 1733 info->max_indirect_segments = min(indirect_segments, 1734 xen_blkif_max_segments); 1735 segs = info->max_indirect_segments; 1736 } 1737 1738 err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE(info)); 1739 if (err) 1740 goto out_of_memory; 1741 1742 if (!info->feature_persistent && info->max_indirect_segments) { 1743 /* 1744 * We are using indirect descriptors but not persistent 1745 * grants, we need to allocate a set of pages that can be 1746 * used for mapping indirect grefs 1747 */ 1748 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE(info); 1749 1750 BUG_ON(!list_empty(&info->indirect_pages)); 1751 for (i = 0; i < num; i++) { 1752 struct page *indirect_page = alloc_page(GFP_NOIO); 1753 if (!indirect_page) 1754 goto out_of_memory; 1755 list_add(&indirect_page->lru, &info->indirect_pages); 1756 } 1757 } 1758 1759 for (i = 0; i < BLK_RING_SIZE(info); i++) { 1760 info->shadow[i].grants_used = kzalloc( 1761 sizeof(info->shadow[i].grants_used[0]) * segs, 1762 GFP_NOIO); 1763 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO); 1764 if (info->max_indirect_segments) 1765 info->shadow[i].indirect_grants = kzalloc( 1766 sizeof(info->shadow[i].indirect_grants[0]) * 1767 INDIRECT_GREFS(segs), 1768 GFP_NOIO); 1769 if ((info->shadow[i].grants_used == NULL) || 1770 (info->shadow[i].sg == NULL) || 1771 (info->max_indirect_segments && 1772 (info->shadow[i].indirect_grants == NULL))) 1773 goto out_of_memory; 1774 sg_init_table(info->shadow[i].sg, segs); 1775 } 1776 1777 1778 return 0; 1779 1780 out_of_memory: 1781 for (i = 0; i < BLK_RING_SIZE(info); i++) { 1782 kfree(info->shadow[i].grants_used); 1783 info->shadow[i].grants_used = NULL; 1784 kfree(info->shadow[i].sg); 1785 info->shadow[i].sg = NULL; 1786 kfree(info->shadow[i].indirect_grants); 1787 info->shadow[i].indirect_grants = NULL; 1788 } 1789 if (!list_empty(&info->indirect_pages)) { 1790 struct page *indirect_page, *n; 1791 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) { 1792 list_del(&indirect_page->lru); 1793 __free_page(indirect_page); 1794 } 1795 } 1796 return -ENOMEM; 1797 } 1798 1799 /* 1800 * Invoked when the backend is finally 'ready' (and has told produced 1801 * the details about the physical device - #sectors, size, etc). 1802 */ 1803 static void blkfront_connect(struct blkfront_info *info) 1804 { 1805 unsigned long long sectors; 1806 unsigned long sector_size; 1807 unsigned int physical_sector_size; 1808 unsigned int binfo; 1809 int err; 1810 int barrier, flush, discard, persistent; 1811 1812 switch (info->connected) { 1813 case BLKIF_STATE_CONNECTED: 1814 /* 1815 * Potentially, the back-end may be signalling 1816 * a capacity change; update the capacity. 1817 */ 1818 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, 1819 "sectors", "%Lu", §ors); 1820 if (XENBUS_EXIST_ERR(err)) 1821 return; 1822 printk(KERN_INFO "Setting capacity to %Lu\n", 1823 sectors); 1824 set_capacity(info->gd, sectors); 1825 revalidate_disk(info->gd); 1826 1827 return; 1828 case BLKIF_STATE_SUSPENDED: 1829 /* 1830 * If we are recovering from suspension, we need to wait 1831 * for the backend to announce it's features before 1832 * reconnecting, at least we need to know if the backend 1833 * supports indirect descriptors, and how many. 1834 */ 1835 blkif_recover(info); 1836 return; 1837 1838 default: 1839 break; 1840 } 1841 1842 dev_dbg(&info->xbdev->dev, "%s:%s.\n", 1843 __func__, info->xbdev->otherend); 1844 1845 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1846 "sectors", "%llu", §ors, 1847 "info", "%u", &binfo, 1848 "sector-size", "%lu", §or_size, 1849 NULL); 1850 if (err) { 1851 xenbus_dev_fatal(info->xbdev, err, 1852 "reading backend fields at %s", 1853 info->xbdev->otherend); 1854 return; 1855 } 1856 1857 /* 1858 * physcial-sector-size is a newer field, so old backends may not 1859 * provide this. Assume physical sector size to be the same as 1860 * sector_size in that case. 1861 */ 1862 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, 1863 "physical-sector-size", "%u", &physical_sector_size); 1864 if (err != 1) 1865 physical_sector_size = sector_size; 1866 1867 info->feature_flush = 0; 1868 1869 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1870 "feature-barrier", "%d", &barrier, 1871 NULL); 1872 1873 /* 1874 * If there's no "feature-barrier" defined, then it means 1875 * we're dealing with a very old backend which writes 1876 * synchronously; nothing to do. 1877 * 1878 * If there are barriers, then we use flush. 1879 */ 1880 if (!err && barrier) 1881 info->feature_flush = REQ_FLUSH | REQ_FUA; 1882 /* 1883 * And if there is "feature-flush-cache" use that above 1884 * barriers. 1885 */ 1886 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1887 "feature-flush-cache", "%d", &flush, 1888 NULL); 1889 1890 if (!err && flush) 1891 info->feature_flush = REQ_FLUSH; 1892 1893 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1894 "feature-discard", "%d", &discard, 1895 NULL); 1896 1897 if (!err && discard) 1898 blkfront_setup_discard(info); 1899 1900 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1901 "feature-persistent", "%u", &persistent, 1902 NULL); 1903 if (err) 1904 info->feature_persistent = 0; 1905 else 1906 info->feature_persistent = persistent; 1907 1908 err = blkfront_setup_indirect(info); 1909 if (err) { 1910 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s", 1911 info->xbdev->otherend); 1912 return; 1913 } 1914 1915 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size, 1916 physical_sector_size); 1917 if (err) { 1918 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s", 1919 info->xbdev->otherend); 1920 return; 1921 } 1922 1923 xenbus_switch_state(info->xbdev, XenbusStateConnected); 1924 1925 /* Kick pending requests. */ 1926 spin_lock_irq(&info->io_lock); 1927 info->connected = BLKIF_STATE_CONNECTED; 1928 kick_pending_request_queues(info); 1929 spin_unlock_irq(&info->io_lock); 1930 1931 add_disk(info->gd); 1932 1933 info->is_ready = 1; 1934 } 1935 1936 /** 1937 * Callback received when the backend's state changes. 1938 */ 1939 static void blkback_changed(struct xenbus_device *dev, 1940 enum xenbus_state backend_state) 1941 { 1942 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 1943 1944 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state); 1945 1946 switch (backend_state) { 1947 case XenbusStateInitWait: 1948 if (dev->state != XenbusStateInitialising) 1949 break; 1950 if (talk_to_blkback(dev, info)) { 1951 kfree(info); 1952 dev_set_drvdata(&dev->dev, NULL); 1953 break; 1954 } 1955 case XenbusStateInitialising: 1956 case XenbusStateInitialised: 1957 case XenbusStateReconfiguring: 1958 case XenbusStateReconfigured: 1959 case XenbusStateUnknown: 1960 break; 1961 1962 case XenbusStateConnected: 1963 blkfront_connect(info); 1964 break; 1965 1966 case XenbusStateClosed: 1967 if (dev->state == XenbusStateClosed) 1968 break; 1969 /* Missed the backend's Closing state -- fallthrough */ 1970 case XenbusStateClosing: 1971 blkfront_closing(info); 1972 break; 1973 } 1974 } 1975 1976 static int blkfront_remove(struct xenbus_device *xbdev) 1977 { 1978 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev); 1979 struct block_device *bdev = NULL; 1980 struct gendisk *disk; 1981 1982 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename); 1983 1984 blkif_free(info, 0); 1985 1986 mutex_lock(&info->mutex); 1987 1988 disk = info->gd; 1989 if (disk) 1990 bdev = bdget_disk(disk, 0); 1991 1992 info->xbdev = NULL; 1993 mutex_unlock(&info->mutex); 1994 1995 if (!bdev) { 1996 kfree(info); 1997 return 0; 1998 } 1999 2000 /* 2001 * The xbdev was removed before we reached the Closed 2002 * state. See if it's safe to remove the disk. If the bdev 2003 * isn't closed yet, we let release take care of it. 2004 */ 2005 2006 mutex_lock(&bdev->bd_mutex); 2007 info = disk->private_data; 2008 2009 dev_warn(disk_to_dev(disk), 2010 "%s was hot-unplugged, %d stale handles\n", 2011 xbdev->nodename, bdev->bd_openers); 2012 2013 if (info && !bdev->bd_openers) { 2014 xlvbd_release_gendisk(info); 2015 disk->private_data = NULL; 2016 kfree(info); 2017 } 2018 2019 mutex_unlock(&bdev->bd_mutex); 2020 bdput(bdev); 2021 2022 return 0; 2023 } 2024 2025 static int blkfront_is_ready(struct xenbus_device *dev) 2026 { 2027 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 2028 2029 return info->is_ready && info->xbdev; 2030 } 2031 2032 static int blkif_open(struct block_device *bdev, fmode_t mode) 2033 { 2034 struct gendisk *disk = bdev->bd_disk; 2035 struct blkfront_info *info; 2036 int err = 0; 2037 2038 mutex_lock(&blkfront_mutex); 2039 2040 info = disk->private_data; 2041 if (!info) { 2042 /* xbdev gone */ 2043 err = -ERESTARTSYS; 2044 goto out; 2045 } 2046 2047 mutex_lock(&info->mutex); 2048 2049 if (!info->gd) 2050 /* xbdev is closed */ 2051 err = -ERESTARTSYS; 2052 2053 mutex_unlock(&info->mutex); 2054 2055 out: 2056 mutex_unlock(&blkfront_mutex); 2057 return err; 2058 } 2059 2060 static void blkif_release(struct gendisk *disk, fmode_t mode) 2061 { 2062 struct blkfront_info *info = disk->private_data; 2063 struct block_device *bdev; 2064 struct xenbus_device *xbdev; 2065 2066 mutex_lock(&blkfront_mutex); 2067 2068 bdev = bdget_disk(disk, 0); 2069 2070 if (!bdev) { 2071 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name); 2072 goto out_mutex; 2073 } 2074 if (bdev->bd_openers) 2075 goto out; 2076 2077 /* 2078 * Check if we have been instructed to close. We will have 2079 * deferred this request, because the bdev was still open. 2080 */ 2081 2082 mutex_lock(&info->mutex); 2083 xbdev = info->xbdev; 2084 2085 if (xbdev && xbdev->state == XenbusStateClosing) { 2086 /* pending switch to state closed */ 2087 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n"); 2088 xlvbd_release_gendisk(info); 2089 xenbus_frontend_closed(info->xbdev); 2090 } 2091 2092 mutex_unlock(&info->mutex); 2093 2094 if (!xbdev) { 2095 /* sudden device removal */ 2096 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n"); 2097 xlvbd_release_gendisk(info); 2098 disk->private_data = NULL; 2099 kfree(info); 2100 } 2101 2102 out: 2103 bdput(bdev); 2104 out_mutex: 2105 mutex_unlock(&blkfront_mutex); 2106 } 2107 2108 static const struct block_device_operations xlvbd_block_fops = 2109 { 2110 .owner = THIS_MODULE, 2111 .open = blkif_open, 2112 .release = blkif_release, 2113 .getgeo = blkif_getgeo, 2114 .ioctl = blkif_ioctl, 2115 }; 2116 2117 2118 static const struct xenbus_device_id blkfront_ids[] = { 2119 { "vbd" }, 2120 { "" } 2121 }; 2122 2123 static struct xenbus_driver blkfront_driver = { 2124 .ids = blkfront_ids, 2125 .probe = blkfront_probe, 2126 .remove = blkfront_remove, 2127 .resume = blkfront_resume, 2128 .otherend_changed = blkback_changed, 2129 .is_ready = blkfront_is_ready, 2130 }; 2131 2132 static int __init xlblk_init(void) 2133 { 2134 int ret; 2135 2136 if (!xen_domain()) 2137 return -ENODEV; 2138 2139 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_PAGE_ORDER) { 2140 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n", 2141 xen_blkif_max_ring_order, XENBUS_MAX_RING_PAGE_ORDER); 2142 xen_blkif_max_ring_order = 0; 2143 } 2144 2145 if (!xen_has_pv_disk_devices()) 2146 return -ENODEV; 2147 2148 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) { 2149 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n", 2150 XENVBD_MAJOR, DEV_NAME); 2151 return -ENODEV; 2152 } 2153 2154 ret = xenbus_register_frontend(&blkfront_driver); 2155 if (ret) { 2156 unregister_blkdev(XENVBD_MAJOR, DEV_NAME); 2157 return ret; 2158 } 2159 2160 return 0; 2161 } 2162 module_init(xlblk_init); 2163 2164 2165 static void __exit xlblk_exit(void) 2166 { 2167 xenbus_unregister_driver(&blkfront_driver); 2168 unregister_blkdev(XENVBD_MAJOR, DEV_NAME); 2169 kfree(minors); 2170 } 2171 module_exit(xlblk_exit); 2172 2173 MODULE_DESCRIPTION("Xen virtual block device frontend"); 2174 MODULE_LICENSE("GPL"); 2175 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR); 2176 MODULE_ALIAS("xen:vbd"); 2177 MODULE_ALIAS("xenblk"); 2178