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 47 #include <xen/xen.h> 48 #include <xen/xenbus.h> 49 #include <xen/grant_table.h> 50 #include <xen/events.h> 51 #include <xen/page.h> 52 #include <xen/platform_pci.h> 53 54 #include <xen/interface/grant_table.h> 55 #include <xen/interface/io/blkif.h> 56 #include <xen/interface/io/protocols.h> 57 58 #include <asm/xen/hypervisor.h> 59 60 enum blkif_state { 61 BLKIF_STATE_DISCONNECTED, 62 BLKIF_STATE_CONNECTED, 63 BLKIF_STATE_SUSPENDED, 64 }; 65 66 struct blk_shadow { 67 struct blkif_request req; 68 struct request *request; 69 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 70 }; 71 72 static DEFINE_MUTEX(blkfront_mutex); 73 static const struct block_device_operations xlvbd_block_fops; 74 75 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE) 76 77 /* 78 * We have one of these per vbd, whether ide, scsi or 'other'. They 79 * hang in private_data off the gendisk structure. We may end up 80 * putting all kinds of interesting stuff here :-) 81 */ 82 struct blkfront_info 83 { 84 struct mutex mutex; 85 struct xenbus_device *xbdev; 86 struct gendisk *gd; 87 int vdevice; 88 blkif_vdev_t handle; 89 enum blkif_state connected; 90 int ring_ref; 91 struct blkif_front_ring ring; 92 struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 93 unsigned int evtchn, irq; 94 struct request_queue *rq; 95 struct work_struct work; 96 struct gnttab_free_callback callback; 97 struct blk_shadow shadow[BLK_RING_SIZE]; 98 unsigned long shadow_free; 99 unsigned int feature_flush; 100 int is_ready; 101 }; 102 103 static DEFINE_SPINLOCK(blkif_io_lock); 104 105 static unsigned int nr_minors; 106 static unsigned long *minors; 107 static DEFINE_SPINLOCK(minor_lock); 108 109 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \ 110 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE) 111 #define GRANT_INVALID_REF 0 112 113 #define PARTS_PER_DISK 16 114 #define PARTS_PER_EXT_DISK 256 115 116 #define BLKIF_MAJOR(dev) ((dev)>>8) 117 #define BLKIF_MINOR(dev) ((dev) & 0xff) 118 119 #define EXT_SHIFT 28 120 #define EXTENDED (1<<EXT_SHIFT) 121 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED)) 122 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED)) 123 #define EMULATED_HD_DISK_MINOR_OFFSET (0) 124 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256) 125 #define EMULATED_SD_DISK_MINOR_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET + (4 * 16)) 126 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_HD_DISK_NAME_OFFSET + 4) 127 128 #define DEV_NAME "xvd" /* name in /dev */ 129 130 static int get_id_from_freelist(struct blkfront_info *info) 131 { 132 unsigned long free = info->shadow_free; 133 BUG_ON(free >= BLK_RING_SIZE); 134 info->shadow_free = info->shadow[free].req.id; 135 info->shadow[free].req.id = 0x0fffffee; /* debug */ 136 return free; 137 } 138 139 static void add_id_to_freelist(struct blkfront_info *info, 140 unsigned long id) 141 { 142 info->shadow[id].req.id = info->shadow_free; 143 info->shadow[id].request = NULL; 144 info->shadow_free = id; 145 } 146 147 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr) 148 { 149 unsigned int end = minor + nr; 150 int rc; 151 152 if (end > nr_minors) { 153 unsigned long *bitmap, *old; 154 155 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap), 156 GFP_KERNEL); 157 if (bitmap == NULL) 158 return -ENOMEM; 159 160 spin_lock(&minor_lock); 161 if (end > nr_minors) { 162 old = minors; 163 memcpy(bitmap, minors, 164 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap)); 165 minors = bitmap; 166 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG; 167 } else 168 old = bitmap; 169 spin_unlock(&minor_lock); 170 kfree(old); 171 } 172 173 spin_lock(&minor_lock); 174 if (find_next_bit(minors, end, minor) >= end) { 175 for (; minor < end; ++minor) 176 __set_bit(minor, minors); 177 rc = 0; 178 } else 179 rc = -EBUSY; 180 spin_unlock(&minor_lock); 181 182 return rc; 183 } 184 185 static void xlbd_release_minors(unsigned int minor, unsigned int nr) 186 { 187 unsigned int end = minor + nr; 188 189 BUG_ON(end > nr_minors); 190 spin_lock(&minor_lock); 191 for (; minor < end; ++minor) 192 __clear_bit(minor, minors); 193 spin_unlock(&minor_lock); 194 } 195 196 static void blkif_restart_queue_callback(void *arg) 197 { 198 struct blkfront_info *info = (struct blkfront_info *)arg; 199 schedule_work(&info->work); 200 } 201 202 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg) 203 { 204 /* We don't have real geometry info, but let's at least return 205 values consistent with the size of the device */ 206 sector_t nsect = get_capacity(bd->bd_disk); 207 sector_t cylinders = nsect; 208 209 hg->heads = 0xff; 210 hg->sectors = 0x3f; 211 sector_div(cylinders, hg->heads * hg->sectors); 212 hg->cylinders = cylinders; 213 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect) 214 hg->cylinders = 0xffff; 215 return 0; 216 } 217 218 static int blkif_ioctl(struct block_device *bdev, fmode_t mode, 219 unsigned command, unsigned long argument) 220 { 221 struct blkfront_info *info = bdev->bd_disk->private_data; 222 int i; 223 224 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n", 225 command, (long)argument); 226 227 switch (command) { 228 case CDROMMULTISESSION: 229 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n"); 230 for (i = 0; i < sizeof(struct cdrom_multisession); i++) 231 if (put_user(0, (char __user *)(argument + i))) 232 return -EFAULT; 233 return 0; 234 235 case CDROM_GET_CAPABILITY: { 236 struct gendisk *gd = info->gd; 237 if (gd->flags & GENHD_FL_CD) 238 return 0; 239 return -EINVAL; 240 } 241 242 default: 243 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n", 244 command);*/ 245 return -EINVAL; /* same return as native Linux */ 246 } 247 248 return 0; 249 } 250 251 /* 252 * Generate a Xen blkfront IO request from a blk layer request. Reads 253 * and writes are handled as expected. Since we lack a loose flush 254 * request, we map flushes into a full ordered barrier. 255 * 256 * @req: a request struct 257 */ 258 static int blkif_queue_request(struct request *req) 259 { 260 struct blkfront_info *info = req->rq_disk->private_data; 261 unsigned long buffer_mfn; 262 struct blkif_request *ring_req; 263 unsigned long id; 264 unsigned int fsect, lsect; 265 int i, ref; 266 grant_ref_t gref_head; 267 struct scatterlist *sg; 268 269 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) 270 return 1; 271 272 if (gnttab_alloc_grant_references( 273 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) { 274 gnttab_request_free_callback( 275 &info->callback, 276 blkif_restart_queue_callback, 277 info, 278 BLKIF_MAX_SEGMENTS_PER_REQUEST); 279 return 1; 280 } 281 282 /* Fill out a communications ring structure. */ 283 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt); 284 id = get_id_from_freelist(info); 285 info->shadow[id].request = req; 286 287 ring_req->id = id; 288 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req); 289 ring_req->handle = info->handle; 290 291 ring_req->operation = rq_data_dir(req) ? 292 BLKIF_OP_WRITE : BLKIF_OP_READ; 293 294 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) { 295 /* 296 * Ideally we could just do an unordered 297 * flush-to-disk, but all we have is a full write 298 * barrier at the moment. However, a barrier write is 299 * a superset of FUA, so we can implement it the same 300 * way. (It's also a FLUSH+FUA, since it is 301 * guaranteed ordered WRT previous writes.) 302 */ 303 ring_req->operation = BLKIF_OP_WRITE_BARRIER; 304 } 305 306 ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg); 307 BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST); 308 309 for_each_sg(info->sg, sg, ring_req->nr_segments, i) { 310 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg))); 311 fsect = sg->offset >> 9; 312 lsect = fsect + (sg->length >> 9) - 1; 313 /* install a grant reference. */ 314 ref = gnttab_claim_grant_reference(&gref_head); 315 BUG_ON(ref == -ENOSPC); 316 317 gnttab_grant_foreign_access_ref( 318 ref, 319 info->xbdev->otherend_id, 320 buffer_mfn, 321 rq_data_dir(req) ); 322 323 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn); 324 ring_req->u.rw.seg[i] = 325 (struct blkif_request_segment) { 326 .gref = ref, 327 .first_sect = fsect, 328 .last_sect = lsect }; 329 } 330 331 info->ring.req_prod_pvt++; 332 333 /* Keep a private copy so we can reissue requests when recovering. */ 334 info->shadow[id].req = *ring_req; 335 336 gnttab_free_grant_references(gref_head); 337 338 return 0; 339 } 340 341 342 static inline void flush_requests(struct blkfront_info *info) 343 { 344 int notify; 345 346 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify); 347 348 if (notify) 349 notify_remote_via_irq(info->irq); 350 } 351 352 /* 353 * do_blkif_request 354 * read a block; request is in a request queue 355 */ 356 static void do_blkif_request(struct request_queue *rq) 357 { 358 struct blkfront_info *info = NULL; 359 struct request *req; 360 int queued; 361 362 pr_debug("Entered do_blkif_request\n"); 363 364 queued = 0; 365 366 while ((req = blk_peek_request(rq)) != NULL) { 367 info = req->rq_disk->private_data; 368 369 if (RING_FULL(&info->ring)) 370 goto wait; 371 372 blk_start_request(req); 373 374 if (req->cmd_type != REQ_TYPE_FS) { 375 __blk_end_request_all(req, -EIO); 376 continue; 377 } 378 379 pr_debug("do_blk_req %p: cmd %p, sec %lx, " 380 "(%u/%u) buffer:%p [%s]\n", 381 req, req->cmd, (unsigned long)blk_rq_pos(req), 382 blk_rq_cur_sectors(req), blk_rq_sectors(req), 383 req->buffer, rq_data_dir(req) ? "write" : "read"); 384 385 if (blkif_queue_request(req)) { 386 blk_requeue_request(rq, req); 387 wait: 388 /* Avoid pointless unplugs. */ 389 blk_stop_queue(rq); 390 break; 391 } 392 393 queued++; 394 } 395 396 if (queued != 0) 397 flush_requests(info); 398 } 399 400 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size) 401 { 402 struct request_queue *rq; 403 404 rq = blk_init_queue(do_blkif_request, &blkif_io_lock); 405 if (rq == NULL) 406 return -1; 407 408 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq); 409 410 /* Hard sector size and max sectors impersonate the equiv. hardware. */ 411 blk_queue_logical_block_size(rq, sector_size); 412 blk_queue_max_hw_sectors(rq, 512); 413 414 /* Each segment in a request is up to an aligned page in size. */ 415 blk_queue_segment_boundary(rq, PAGE_SIZE - 1); 416 blk_queue_max_segment_size(rq, PAGE_SIZE); 417 418 /* Ensure a merged request will fit in a single I/O ring slot. */ 419 blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST); 420 421 /* Make sure buffer addresses are sector-aligned. */ 422 blk_queue_dma_alignment(rq, 511); 423 424 /* Make sure we don't use bounce buffers. */ 425 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY); 426 427 gd->queue = rq; 428 429 return 0; 430 } 431 432 433 static void xlvbd_flush(struct blkfront_info *info) 434 { 435 blk_queue_flush(info->rq, info->feature_flush); 436 printk(KERN_INFO "blkfront: %s: barriers %s\n", 437 info->gd->disk_name, 438 info->feature_flush ? "enabled" : "disabled"); 439 } 440 441 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset) 442 { 443 int major; 444 major = BLKIF_MAJOR(vdevice); 445 *minor = BLKIF_MINOR(vdevice); 446 switch (major) { 447 case XEN_IDE0_MAJOR: 448 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET; 449 *minor = ((*minor / 64) * PARTS_PER_DISK) + 450 EMULATED_HD_DISK_MINOR_OFFSET; 451 break; 452 case XEN_IDE1_MAJOR: 453 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET; 454 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) + 455 EMULATED_HD_DISK_MINOR_OFFSET; 456 break; 457 case XEN_SCSI_DISK0_MAJOR: 458 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET; 459 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET; 460 break; 461 case XEN_SCSI_DISK1_MAJOR: 462 case XEN_SCSI_DISK2_MAJOR: 463 case XEN_SCSI_DISK3_MAJOR: 464 case XEN_SCSI_DISK4_MAJOR: 465 case XEN_SCSI_DISK5_MAJOR: 466 case XEN_SCSI_DISK6_MAJOR: 467 case XEN_SCSI_DISK7_MAJOR: 468 *offset = (*minor / PARTS_PER_DISK) + 469 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) + 470 EMULATED_SD_DISK_NAME_OFFSET; 471 *minor = *minor + 472 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) + 473 EMULATED_SD_DISK_MINOR_OFFSET; 474 break; 475 case XEN_SCSI_DISK8_MAJOR: 476 case XEN_SCSI_DISK9_MAJOR: 477 case XEN_SCSI_DISK10_MAJOR: 478 case XEN_SCSI_DISK11_MAJOR: 479 case XEN_SCSI_DISK12_MAJOR: 480 case XEN_SCSI_DISK13_MAJOR: 481 case XEN_SCSI_DISK14_MAJOR: 482 case XEN_SCSI_DISK15_MAJOR: 483 *offset = (*minor / PARTS_PER_DISK) + 484 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) + 485 EMULATED_SD_DISK_NAME_OFFSET; 486 *minor = *minor + 487 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) + 488 EMULATED_SD_DISK_MINOR_OFFSET; 489 break; 490 case XENVBD_MAJOR: 491 *offset = *minor / PARTS_PER_DISK; 492 break; 493 default: 494 printk(KERN_WARNING "blkfront: your disk configuration is " 495 "incorrect, please use an xvd device instead\n"); 496 return -ENODEV; 497 } 498 return 0; 499 } 500 501 static int xlvbd_alloc_gendisk(blkif_sector_t capacity, 502 struct blkfront_info *info, 503 u16 vdisk_info, u16 sector_size) 504 { 505 struct gendisk *gd; 506 int nr_minors = 1; 507 int err; 508 unsigned int offset; 509 int minor; 510 int nr_parts; 511 512 BUG_ON(info->gd != NULL); 513 BUG_ON(info->rq != NULL); 514 515 if ((info->vdevice>>EXT_SHIFT) > 1) { 516 /* this is above the extended range; something is wrong */ 517 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice); 518 return -ENODEV; 519 } 520 521 if (!VDEV_IS_EXTENDED(info->vdevice)) { 522 err = xen_translate_vdev(info->vdevice, &minor, &offset); 523 if (err) 524 return err; 525 nr_parts = PARTS_PER_DISK; 526 } else { 527 minor = BLKIF_MINOR_EXT(info->vdevice); 528 nr_parts = PARTS_PER_EXT_DISK; 529 offset = minor / nr_parts; 530 if (xen_hvm_domain() && offset <= EMULATED_HD_DISK_NAME_OFFSET + 4) 531 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with " 532 "emulated IDE disks,\n\t choose an xvd device name" 533 "from xvde on\n", info->vdevice); 534 } 535 err = -ENODEV; 536 537 if ((minor % nr_parts) == 0) 538 nr_minors = nr_parts; 539 540 err = xlbd_reserve_minors(minor, nr_minors); 541 if (err) 542 goto out; 543 err = -ENODEV; 544 545 gd = alloc_disk(nr_minors); 546 if (gd == NULL) 547 goto release; 548 549 if (nr_minors > 1) { 550 if (offset < 26) 551 sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset); 552 else 553 sprintf(gd->disk_name, "%s%c%c", DEV_NAME, 554 'a' + ((offset / 26)-1), 'a' + (offset % 26)); 555 } else { 556 if (offset < 26) 557 sprintf(gd->disk_name, "%s%c%d", DEV_NAME, 558 'a' + offset, 559 minor & (nr_parts - 1)); 560 else 561 sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME, 562 'a' + ((offset / 26) - 1), 563 'a' + (offset % 26), 564 minor & (nr_parts - 1)); 565 } 566 567 gd->major = XENVBD_MAJOR; 568 gd->first_minor = minor; 569 gd->fops = &xlvbd_block_fops; 570 gd->private_data = info; 571 gd->driverfs_dev = &(info->xbdev->dev); 572 set_capacity(gd, capacity); 573 574 if (xlvbd_init_blk_queue(gd, sector_size)) { 575 del_gendisk(gd); 576 goto release; 577 } 578 579 info->rq = gd->queue; 580 info->gd = gd; 581 582 xlvbd_flush(info); 583 584 if (vdisk_info & VDISK_READONLY) 585 set_disk_ro(gd, 1); 586 587 if (vdisk_info & VDISK_REMOVABLE) 588 gd->flags |= GENHD_FL_REMOVABLE; 589 590 if (vdisk_info & VDISK_CDROM) 591 gd->flags |= GENHD_FL_CD; 592 593 return 0; 594 595 release: 596 xlbd_release_minors(minor, nr_minors); 597 out: 598 return err; 599 } 600 601 static void xlvbd_release_gendisk(struct blkfront_info *info) 602 { 603 unsigned int minor, nr_minors; 604 unsigned long flags; 605 606 if (info->rq == NULL) 607 return; 608 609 spin_lock_irqsave(&blkif_io_lock, flags); 610 611 /* No more blkif_request(). */ 612 blk_stop_queue(info->rq); 613 614 /* No more gnttab callback work. */ 615 gnttab_cancel_free_callback(&info->callback); 616 spin_unlock_irqrestore(&blkif_io_lock, flags); 617 618 /* Flush gnttab callback work. Must be done with no locks held. */ 619 flush_work_sync(&info->work); 620 621 del_gendisk(info->gd); 622 623 minor = info->gd->first_minor; 624 nr_minors = info->gd->minors; 625 xlbd_release_minors(minor, nr_minors); 626 627 blk_cleanup_queue(info->rq); 628 info->rq = NULL; 629 630 put_disk(info->gd); 631 info->gd = NULL; 632 } 633 634 static void kick_pending_request_queues(struct blkfront_info *info) 635 { 636 if (!RING_FULL(&info->ring)) { 637 /* Re-enable calldowns. */ 638 blk_start_queue(info->rq); 639 /* Kick things off immediately. */ 640 do_blkif_request(info->rq); 641 } 642 } 643 644 static void blkif_restart_queue(struct work_struct *work) 645 { 646 struct blkfront_info *info = container_of(work, struct blkfront_info, work); 647 648 spin_lock_irq(&blkif_io_lock); 649 if (info->connected == BLKIF_STATE_CONNECTED) 650 kick_pending_request_queues(info); 651 spin_unlock_irq(&blkif_io_lock); 652 } 653 654 static void blkif_free(struct blkfront_info *info, int suspend) 655 { 656 /* Prevent new requests being issued until we fix things up. */ 657 spin_lock_irq(&blkif_io_lock); 658 info->connected = suspend ? 659 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED; 660 /* No more blkif_request(). */ 661 if (info->rq) 662 blk_stop_queue(info->rq); 663 /* No more gnttab callback work. */ 664 gnttab_cancel_free_callback(&info->callback); 665 spin_unlock_irq(&blkif_io_lock); 666 667 /* Flush gnttab callback work. Must be done with no locks held. */ 668 flush_work_sync(&info->work); 669 670 /* Free resources associated with old device channel. */ 671 if (info->ring_ref != GRANT_INVALID_REF) { 672 gnttab_end_foreign_access(info->ring_ref, 0, 673 (unsigned long)info->ring.sring); 674 info->ring_ref = GRANT_INVALID_REF; 675 info->ring.sring = NULL; 676 } 677 if (info->irq) 678 unbind_from_irqhandler(info->irq, info); 679 info->evtchn = info->irq = 0; 680 681 } 682 683 static void blkif_completion(struct blk_shadow *s) 684 { 685 int i; 686 for (i = 0; i < s->req.nr_segments; i++) 687 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL); 688 } 689 690 static irqreturn_t blkif_interrupt(int irq, void *dev_id) 691 { 692 struct request *req; 693 struct blkif_response *bret; 694 RING_IDX i, rp; 695 unsigned long flags; 696 struct blkfront_info *info = (struct blkfront_info *)dev_id; 697 int error; 698 699 spin_lock_irqsave(&blkif_io_lock, flags); 700 701 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) { 702 spin_unlock_irqrestore(&blkif_io_lock, flags); 703 return IRQ_HANDLED; 704 } 705 706 again: 707 rp = info->ring.sring->rsp_prod; 708 rmb(); /* Ensure we see queued responses up to 'rp'. */ 709 710 for (i = info->ring.rsp_cons; i != rp; i++) { 711 unsigned long id; 712 713 bret = RING_GET_RESPONSE(&info->ring, i); 714 id = bret->id; 715 req = info->shadow[id].request; 716 717 blkif_completion(&info->shadow[id]); 718 719 add_id_to_freelist(info, id); 720 721 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO; 722 switch (bret->operation) { 723 case BLKIF_OP_WRITE_BARRIER: 724 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) { 725 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n", 726 info->gd->disk_name); 727 error = -EOPNOTSUPP; 728 } 729 if (unlikely(bret->status == BLKIF_RSP_ERROR && 730 info->shadow[id].req.nr_segments == 0)) { 731 printk(KERN_WARNING "blkfront: %s: empty write barrier op failed\n", 732 info->gd->disk_name); 733 error = -EOPNOTSUPP; 734 } 735 if (unlikely(error)) { 736 if (error == -EOPNOTSUPP) 737 error = 0; 738 info->feature_flush = 0; 739 xlvbd_flush(info); 740 } 741 /* fall through */ 742 case BLKIF_OP_READ: 743 case BLKIF_OP_WRITE: 744 if (unlikely(bret->status != BLKIF_RSP_OKAY)) 745 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data " 746 "request: %x\n", bret->status); 747 748 __blk_end_request_all(req, error); 749 break; 750 default: 751 BUG(); 752 } 753 } 754 755 info->ring.rsp_cons = i; 756 757 if (i != info->ring.req_prod_pvt) { 758 int more_to_do; 759 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do); 760 if (more_to_do) 761 goto again; 762 } else 763 info->ring.sring->rsp_event = i + 1; 764 765 kick_pending_request_queues(info); 766 767 spin_unlock_irqrestore(&blkif_io_lock, flags); 768 769 return IRQ_HANDLED; 770 } 771 772 773 static int setup_blkring(struct xenbus_device *dev, 774 struct blkfront_info *info) 775 { 776 struct blkif_sring *sring; 777 int err; 778 779 info->ring_ref = GRANT_INVALID_REF; 780 781 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH); 782 if (!sring) { 783 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring"); 784 return -ENOMEM; 785 } 786 SHARED_RING_INIT(sring); 787 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE); 788 789 sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST); 790 791 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring)); 792 if (err < 0) { 793 free_page((unsigned long)sring); 794 info->ring.sring = NULL; 795 goto fail; 796 } 797 info->ring_ref = err; 798 799 err = xenbus_alloc_evtchn(dev, &info->evtchn); 800 if (err) 801 goto fail; 802 803 err = bind_evtchn_to_irqhandler(info->evtchn, 804 blkif_interrupt, 805 IRQF_SAMPLE_RANDOM, "blkif", info); 806 if (err <= 0) { 807 xenbus_dev_fatal(dev, err, 808 "bind_evtchn_to_irqhandler failed"); 809 goto fail; 810 } 811 info->irq = err; 812 813 return 0; 814 fail: 815 blkif_free(info, 0); 816 return err; 817 } 818 819 820 /* Common code used when first setting up, and when resuming. */ 821 static int talk_to_blkback(struct xenbus_device *dev, 822 struct blkfront_info *info) 823 { 824 const char *message = NULL; 825 struct xenbus_transaction xbt; 826 int err; 827 828 /* Create shared ring, alloc event channel. */ 829 err = setup_blkring(dev, info); 830 if (err) 831 goto out; 832 833 again: 834 err = xenbus_transaction_start(&xbt); 835 if (err) { 836 xenbus_dev_fatal(dev, err, "starting transaction"); 837 goto destroy_blkring; 838 } 839 840 err = xenbus_printf(xbt, dev->nodename, 841 "ring-ref", "%u", info->ring_ref); 842 if (err) { 843 message = "writing ring-ref"; 844 goto abort_transaction; 845 } 846 err = xenbus_printf(xbt, dev->nodename, 847 "event-channel", "%u", info->evtchn); 848 if (err) { 849 message = "writing event-channel"; 850 goto abort_transaction; 851 } 852 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s", 853 XEN_IO_PROTO_ABI_NATIVE); 854 if (err) { 855 message = "writing protocol"; 856 goto abort_transaction; 857 } 858 859 err = xenbus_transaction_end(xbt, 0); 860 if (err) { 861 if (err == -EAGAIN) 862 goto again; 863 xenbus_dev_fatal(dev, err, "completing transaction"); 864 goto destroy_blkring; 865 } 866 867 xenbus_switch_state(dev, XenbusStateInitialised); 868 869 return 0; 870 871 abort_transaction: 872 xenbus_transaction_end(xbt, 1); 873 if (message) 874 xenbus_dev_fatal(dev, err, "%s", message); 875 destroy_blkring: 876 blkif_free(info, 0); 877 out: 878 return err; 879 } 880 881 /** 882 * Entry point to this code when a new device is created. Allocate the basic 883 * structures and the ring buffer for communication with the backend, and 884 * inform the backend of the appropriate details for those. Switch to 885 * Initialised state. 886 */ 887 static int blkfront_probe(struct xenbus_device *dev, 888 const struct xenbus_device_id *id) 889 { 890 int err, vdevice, i; 891 struct blkfront_info *info; 892 893 /* FIXME: Use dynamic device id if this is not set. */ 894 err = xenbus_scanf(XBT_NIL, dev->nodename, 895 "virtual-device", "%i", &vdevice); 896 if (err != 1) { 897 /* go looking in the extended area instead */ 898 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext", 899 "%i", &vdevice); 900 if (err != 1) { 901 xenbus_dev_fatal(dev, err, "reading virtual-device"); 902 return err; 903 } 904 } 905 906 if (xen_hvm_domain()) { 907 char *type; 908 int len; 909 /* no unplug has been done: do not hook devices != xen vbds */ 910 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) { 911 int major; 912 913 if (!VDEV_IS_EXTENDED(vdevice)) 914 major = BLKIF_MAJOR(vdevice); 915 else 916 major = XENVBD_MAJOR; 917 918 if (major != XENVBD_MAJOR) { 919 printk(KERN_INFO 920 "%s: HVM does not support vbd %d as xen block device\n", 921 __FUNCTION__, vdevice); 922 return -ENODEV; 923 } 924 } 925 /* do not create a PV cdrom device if we are an HVM guest */ 926 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len); 927 if (IS_ERR(type)) 928 return -ENODEV; 929 if (strncmp(type, "cdrom", 5) == 0) { 930 kfree(type); 931 return -ENODEV; 932 } 933 kfree(type); 934 } 935 info = kzalloc(sizeof(*info), GFP_KERNEL); 936 if (!info) { 937 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure"); 938 return -ENOMEM; 939 } 940 941 mutex_init(&info->mutex); 942 info->xbdev = dev; 943 info->vdevice = vdevice; 944 info->connected = BLKIF_STATE_DISCONNECTED; 945 INIT_WORK(&info->work, blkif_restart_queue); 946 947 for (i = 0; i < BLK_RING_SIZE; i++) 948 info->shadow[i].req.id = i+1; 949 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff; 950 951 /* Front end dir is a number, which is used as the id. */ 952 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); 953 dev_set_drvdata(&dev->dev, info); 954 955 err = talk_to_blkback(dev, info); 956 if (err) { 957 kfree(info); 958 dev_set_drvdata(&dev->dev, NULL); 959 return err; 960 } 961 962 return 0; 963 } 964 965 966 static int blkif_recover(struct blkfront_info *info) 967 { 968 int i; 969 struct blkif_request *req; 970 struct blk_shadow *copy; 971 int j; 972 973 /* Stage 1: Make a safe copy of the shadow state. */ 974 copy = kmalloc(sizeof(info->shadow), 975 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH); 976 if (!copy) 977 return -ENOMEM; 978 memcpy(copy, info->shadow, sizeof(info->shadow)); 979 980 /* Stage 2: Set up free list. */ 981 memset(&info->shadow, 0, sizeof(info->shadow)); 982 for (i = 0; i < BLK_RING_SIZE; i++) 983 info->shadow[i].req.id = i+1; 984 info->shadow_free = info->ring.req_prod_pvt; 985 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff; 986 987 /* Stage 3: Find pending requests and requeue them. */ 988 for (i = 0; i < BLK_RING_SIZE; i++) { 989 /* Not in use? */ 990 if (!copy[i].request) 991 continue; 992 993 /* Grab a request slot and copy shadow state into it. */ 994 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt); 995 *req = copy[i].req; 996 997 /* We get a new request id, and must reset the shadow state. */ 998 req->id = get_id_from_freelist(info); 999 memcpy(&info->shadow[req->id], ©[i], sizeof(copy[i])); 1000 1001 /* Rewrite any grant references invalidated by susp/resume. */ 1002 for (j = 0; j < req->nr_segments; j++) 1003 gnttab_grant_foreign_access_ref( 1004 req->u.rw.seg[j].gref, 1005 info->xbdev->otherend_id, 1006 pfn_to_mfn(info->shadow[req->id].frame[j]), 1007 rq_data_dir(info->shadow[req->id].request)); 1008 info->shadow[req->id].req = *req; 1009 1010 info->ring.req_prod_pvt++; 1011 } 1012 1013 kfree(copy); 1014 1015 xenbus_switch_state(info->xbdev, XenbusStateConnected); 1016 1017 spin_lock_irq(&blkif_io_lock); 1018 1019 /* Now safe for us to use the shared ring */ 1020 info->connected = BLKIF_STATE_CONNECTED; 1021 1022 /* Send off requeued requests */ 1023 flush_requests(info); 1024 1025 /* Kick any other new requests queued since we resumed */ 1026 kick_pending_request_queues(info); 1027 1028 spin_unlock_irq(&blkif_io_lock); 1029 1030 return 0; 1031 } 1032 1033 /** 1034 * We are reconnecting to the backend, due to a suspend/resume, or a backend 1035 * driver restart. We tear down our blkif structure and recreate it, but 1036 * leave the device-layer structures intact so that this is transparent to the 1037 * rest of the kernel. 1038 */ 1039 static int blkfront_resume(struct xenbus_device *dev) 1040 { 1041 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 1042 int err; 1043 1044 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename); 1045 1046 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED); 1047 1048 err = talk_to_blkback(dev, info); 1049 if (info->connected == BLKIF_STATE_SUSPENDED && !err) 1050 err = blkif_recover(info); 1051 1052 return err; 1053 } 1054 1055 static void 1056 blkfront_closing(struct blkfront_info *info) 1057 { 1058 struct xenbus_device *xbdev = info->xbdev; 1059 struct block_device *bdev = NULL; 1060 1061 mutex_lock(&info->mutex); 1062 1063 if (xbdev->state == XenbusStateClosing) { 1064 mutex_unlock(&info->mutex); 1065 return; 1066 } 1067 1068 if (info->gd) 1069 bdev = bdget_disk(info->gd, 0); 1070 1071 mutex_unlock(&info->mutex); 1072 1073 if (!bdev) { 1074 xenbus_frontend_closed(xbdev); 1075 return; 1076 } 1077 1078 mutex_lock(&bdev->bd_mutex); 1079 1080 if (bdev->bd_openers) { 1081 xenbus_dev_error(xbdev, -EBUSY, 1082 "Device in use; refusing to close"); 1083 xenbus_switch_state(xbdev, XenbusStateClosing); 1084 } else { 1085 xlvbd_release_gendisk(info); 1086 xenbus_frontend_closed(xbdev); 1087 } 1088 1089 mutex_unlock(&bdev->bd_mutex); 1090 bdput(bdev); 1091 } 1092 1093 /* 1094 * Invoked when the backend is finally 'ready' (and has told produced 1095 * the details about the physical device - #sectors, size, etc). 1096 */ 1097 static void blkfront_connect(struct blkfront_info *info) 1098 { 1099 unsigned long long sectors; 1100 unsigned long sector_size; 1101 unsigned int binfo; 1102 int err; 1103 int barrier; 1104 1105 switch (info->connected) { 1106 case BLKIF_STATE_CONNECTED: 1107 /* 1108 * Potentially, the back-end may be signalling 1109 * a capacity change; update the capacity. 1110 */ 1111 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, 1112 "sectors", "%Lu", §ors); 1113 if (XENBUS_EXIST_ERR(err)) 1114 return; 1115 printk(KERN_INFO "Setting capacity to %Lu\n", 1116 sectors); 1117 set_capacity(info->gd, sectors); 1118 revalidate_disk(info->gd); 1119 1120 /* fall through */ 1121 case BLKIF_STATE_SUSPENDED: 1122 return; 1123 1124 default: 1125 break; 1126 } 1127 1128 dev_dbg(&info->xbdev->dev, "%s:%s.\n", 1129 __func__, info->xbdev->otherend); 1130 1131 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1132 "sectors", "%llu", §ors, 1133 "info", "%u", &binfo, 1134 "sector-size", "%lu", §or_size, 1135 NULL); 1136 if (err) { 1137 xenbus_dev_fatal(info->xbdev, err, 1138 "reading backend fields at %s", 1139 info->xbdev->otherend); 1140 return; 1141 } 1142 1143 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1144 "feature-barrier", "%lu", &barrier, 1145 NULL); 1146 1147 /* 1148 * If there's no "feature-barrier" defined, then it means 1149 * we're dealing with a very old backend which writes 1150 * synchronously; nothing to do. 1151 * 1152 * If there are barriers, then we use flush. 1153 */ 1154 info->feature_flush = 0; 1155 1156 if (!err && barrier) 1157 info->feature_flush = REQ_FLUSH | REQ_FUA; 1158 1159 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size); 1160 if (err) { 1161 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s", 1162 info->xbdev->otherend); 1163 return; 1164 } 1165 1166 xenbus_switch_state(info->xbdev, XenbusStateConnected); 1167 1168 /* Kick pending requests. */ 1169 spin_lock_irq(&blkif_io_lock); 1170 info->connected = BLKIF_STATE_CONNECTED; 1171 kick_pending_request_queues(info); 1172 spin_unlock_irq(&blkif_io_lock); 1173 1174 add_disk(info->gd); 1175 1176 info->is_ready = 1; 1177 } 1178 1179 /** 1180 * Callback received when the backend's state changes. 1181 */ 1182 static void blkback_changed(struct xenbus_device *dev, 1183 enum xenbus_state backend_state) 1184 { 1185 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 1186 1187 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state); 1188 1189 switch (backend_state) { 1190 case XenbusStateInitialising: 1191 case XenbusStateInitWait: 1192 case XenbusStateInitialised: 1193 case XenbusStateReconfiguring: 1194 case XenbusStateReconfigured: 1195 case XenbusStateUnknown: 1196 case XenbusStateClosed: 1197 break; 1198 1199 case XenbusStateConnected: 1200 blkfront_connect(info); 1201 break; 1202 1203 case XenbusStateClosing: 1204 blkfront_closing(info); 1205 break; 1206 } 1207 } 1208 1209 static int blkfront_remove(struct xenbus_device *xbdev) 1210 { 1211 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev); 1212 struct block_device *bdev = NULL; 1213 struct gendisk *disk; 1214 1215 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename); 1216 1217 blkif_free(info, 0); 1218 1219 mutex_lock(&info->mutex); 1220 1221 disk = info->gd; 1222 if (disk) 1223 bdev = bdget_disk(disk, 0); 1224 1225 info->xbdev = NULL; 1226 mutex_unlock(&info->mutex); 1227 1228 if (!bdev) { 1229 kfree(info); 1230 return 0; 1231 } 1232 1233 /* 1234 * The xbdev was removed before we reached the Closed 1235 * state. See if it's safe to remove the disk. If the bdev 1236 * isn't closed yet, we let release take care of it. 1237 */ 1238 1239 mutex_lock(&bdev->bd_mutex); 1240 info = disk->private_data; 1241 1242 dev_warn(disk_to_dev(disk), 1243 "%s was hot-unplugged, %d stale handles\n", 1244 xbdev->nodename, bdev->bd_openers); 1245 1246 if (info && !bdev->bd_openers) { 1247 xlvbd_release_gendisk(info); 1248 disk->private_data = NULL; 1249 kfree(info); 1250 } 1251 1252 mutex_unlock(&bdev->bd_mutex); 1253 bdput(bdev); 1254 1255 return 0; 1256 } 1257 1258 static int blkfront_is_ready(struct xenbus_device *dev) 1259 { 1260 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 1261 1262 return info->is_ready && info->xbdev; 1263 } 1264 1265 static int blkif_open(struct block_device *bdev, fmode_t mode) 1266 { 1267 struct gendisk *disk = bdev->bd_disk; 1268 struct blkfront_info *info; 1269 int err = 0; 1270 1271 mutex_lock(&blkfront_mutex); 1272 1273 info = disk->private_data; 1274 if (!info) { 1275 /* xbdev gone */ 1276 err = -ERESTARTSYS; 1277 goto out; 1278 } 1279 1280 mutex_lock(&info->mutex); 1281 1282 if (!info->gd) 1283 /* xbdev is closed */ 1284 err = -ERESTARTSYS; 1285 1286 mutex_unlock(&info->mutex); 1287 1288 out: 1289 mutex_unlock(&blkfront_mutex); 1290 return err; 1291 } 1292 1293 static int blkif_release(struct gendisk *disk, fmode_t mode) 1294 { 1295 struct blkfront_info *info = disk->private_data; 1296 struct block_device *bdev; 1297 struct xenbus_device *xbdev; 1298 1299 mutex_lock(&blkfront_mutex); 1300 1301 bdev = bdget_disk(disk, 0); 1302 bdput(bdev); 1303 1304 if (bdev->bd_openers) 1305 goto out; 1306 1307 /* 1308 * Check if we have been instructed to close. We will have 1309 * deferred this request, because the bdev was still open. 1310 */ 1311 1312 mutex_lock(&info->mutex); 1313 xbdev = info->xbdev; 1314 1315 if (xbdev && xbdev->state == XenbusStateClosing) { 1316 /* pending switch to state closed */ 1317 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n"); 1318 xlvbd_release_gendisk(info); 1319 xenbus_frontend_closed(info->xbdev); 1320 } 1321 1322 mutex_unlock(&info->mutex); 1323 1324 if (!xbdev) { 1325 /* sudden device removal */ 1326 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n"); 1327 xlvbd_release_gendisk(info); 1328 disk->private_data = NULL; 1329 kfree(info); 1330 } 1331 1332 out: 1333 mutex_unlock(&blkfront_mutex); 1334 return 0; 1335 } 1336 1337 static const struct block_device_operations xlvbd_block_fops = 1338 { 1339 .owner = THIS_MODULE, 1340 .open = blkif_open, 1341 .release = blkif_release, 1342 .getgeo = blkif_getgeo, 1343 .ioctl = blkif_ioctl, 1344 }; 1345 1346 1347 static const struct xenbus_device_id blkfront_ids[] = { 1348 { "vbd" }, 1349 { "" } 1350 }; 1351 1352 static struct xenbus_driver blkfront = { 1353 .name = "vbd", 1354 .owner = THIS_MODULE, 1355 .ids = blkfront_ids, 1356 .probe = blkfront_probe, 1357 .remove = blkfront_remove, 1358 .resume = blkfront_resume, 1359 .otherend_changed = blkback_changed, 1360 .is_ready = blkfront_is_ready, 1361 }; 1362 1363 static int __init xlblk_init(void) 1364 { 1365 if (!xen_domain()) 1366 return -ENODEV; 1367 1368 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) { 1369 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n", 1370 XENVBD_MAJOR, DEV_NAME); 1371 return -ENODEV; 1372 } 1373 1374 return xenbus_register_frontend(&blkfront); 1375 } 1376 module_init(xlblk_init); 1377 1378 1379 static void __exit xlblk_exit(void) 1380 { 1381 return xenbus_unregister_driver(&blkfront); 1382 } 1383 module_exit(xlblk_exit); 1384 1385 MODULE_DESCRIPTION("Xen virtual block device frontend"); 1386 MODULE_LICENSE("GPL"); 1387 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR); 1388 MODULE_ALIAS("xen:vbd"); 1389 MODULE_ALIAS("xenblk"); 1390