drivers/block/xen-blkfront.c

   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[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  78         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  79 };
  80
  81 static DEFINE_MUTEX(blkfront_mutex);
  82 static const struct block_device_operations xlvbd_block_fops;
  83
  84 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
  85
  86 /*
  87  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  88  * hang in private_data off the gendisk structure. We may end up
  89  * putting all kinds of interesting stuff here :-)
  90  */
  91 struct blkfront_info
  92 {
  93         spinlock_t io_lock;
  94         struct mutex mutex;
  95         struct xenbus_device *xbdev;
  96         struct gendisk *gd;
  97         int vdevice;
  98         blkif_vdev_t handle;
  99         enum blkif_state connected;
 100         int ring_ref;
 101         struct blkif_front_ring ring;
 102         unsigned int evtchn, irq;
 103         struct request_queue *rq;
 104         struct work_struct work;
 105         struct gnttab_free_callback callback;
 106         struct blk_shadow shadow[BLK_RING_SIZE];
 107         struct list_head persistent_gnts;
 108         unsigned int persistent_gnts_c;
 109         unsigned long shadow_free;
 110         unsigned int feature_flush;
 111         unsigned int flush_op;
 112         unsigned int feature_discard:1;
 113         unsigned int feature_secdiscard:1;
 114         unsigned int discard_granularity;
 115         unsigned int discard_alignment;
 116         unsigned int feature_persistent:1;
 117         int is_ready;
 118 };
 119
 120 static unsigned int nr_minors;
 121 static unsigned long *minors;
 122 static DEFINE_SPINLOCK(minor_lock);
 123
 124 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 125         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 126 #define GRANT_INVALID_REF       0
 127
 128 #define PARTS_PER_DISK          16
 129 #define PARTS_PER_EXT_DISK      256
 130
 131 #define BLKIF_MAJOR(dev) ((dev)>>8)
 132 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 133
 134 #define EXT_SHIFT 28
 135 #define EXTENDED (1<<EXT_SHIFT)
 136 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 137 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 138 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 139 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 140 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 141 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 142
 143 #define DEV_NAME        "xvd"   /* name in /dev */
 144
 145 static int get_id_from_freelist(struct blkfront_info *info)
 146 {
 147         unsigned long free = info->shadow_free;
 148         BUG_ON(free >= BLK_RING_SIZE);
 149         info->shadow_free = info->shadow[free].req.u.rw.id;
 150         info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 151         return free;
 152 }
 153
 154 static int add_id_to_freelist(struct blkfront_info *info,
 155                                unsigned long id)
 156 {
 157         if (info->shadow[id].req.u.rw.id != id)
 158                 return -EINVAL;
 159         if (info->shadow[id].request == NULL)
 160                 return -EINVAL;
 161         info->shadow[id].req.u.rw.id  = info->shadow_free;
 162         info->shadow[id].request = NULL;
 163         info->shadow_free = id;
 164         return 0;
 165 }
 166
 167 static int fill_grant_buffer(struct blkfront_info *info, int num)
 168 {
 169         struct page *granted_page;
 170         struct grant *gnt_list_entry, *n;
 171         int i = 0;
 172
 173         while(i < num) {
 174                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 175                 if (!gnt_list_entry)
 176                         goto out_of_memory;
 177
 178                 granted_page = alloc_page(GFP_NOIO);
 179                 if (!granted_page) {
 180                         kfree(gnt_list_entry);
 181                         goto out_of_memory;
 182                 }
 183
 184                 gnt_list_entry->pfn = page_to_pfn(granted_page);
 185                 gnt_list_entry->gref = GRANT_INVALID_REF;
 186                 list_add(&gnt_list_entry->node, &info->persistent_gnts);
 187                 i++;
 188         }
 189
 190         return 0;
 191
 192 out_of_memory:
 193         list_for_each_entry_safe(gnt_list_entry, n,
 194                                  &info->persistent_gnts, node) {
 195                 list_del(&gnt_list_entry->node);
 196                 __free_page(pfn_to_page(gnt_list_entry->pfn));
 197                 kfree(gnt_list_entry);
 198                 i--;
 199         }
 200         BUG_ON(i != 0);
 201         return -ENOMEM;
 202 }
 203
 204 static struct grant *get_grant(grant_ref_t *gref_head,
 205                                struct blkfront_info *info)
 206 {
 207         struct grant *gnt_list_entry;
 208         unsigned long buffer_mfn;
 209
 210         BUG_ON(list_empty(&info->persistent_gnts));
 211         gnt_list_entry = list_first_entry(&info->persistent_gnts, struct grant,
 212                                           node);
 213         list_del(&gnt_list_entry->node);
 214
 215         if (gnt_list_entry->gref != GRANT_INVALID_REF) {
 216                 info->persistent_gnts_c--;
 217                 return gnt_list_entry;
 218         }
 219
 220         /* Assign a gref to this page */
 221         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 222         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 223         buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
 224         gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 225                                         info->xbdev->otherend_id,
 226                                         buffer_mfn, 0);
 227         return gnt_list_entry;
 228 }
 229
 230 static const char *op_name(int op)
 231 {
 232         static const char *const names[] = {
 233                 [BLKIF_OP_READ] = "read",
 234                 [BLKIF_OP_WRITE] = "write",
 235                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 236                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 237                 [BLKIF_OP_DISCARD] = "discard" };
 238
 239         if (op < 0 || op >= ARRAY_SIZE(names))
 240                 return "unknown";
 241
 242         if (!names[op])
 243                 return "reserved";
 244
 245         return names[op];
 246 }
 247 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 248 {
 249         unsigned int end = minor + nr;
 250         int rc;
 251
 252         if (end > nr_minors) {
 253                 unsigned long *bitmap, *old;
 254
 255                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 256                                  GFP_KERNEL);
 257                 if (bitmap == NULL)
 258                         return -ENOMEM;
 259
 260                 spin_lock(&minor_lock);
 261                 if (end > nr_minors) {
 262                         old = minors;
 263                         memcpy(bitmap, minors,
 264                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 265                         minors = bitmap;
 266                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 267                 } else
 268                         old = bitmap;
 269                 spin_unlock(&minor_lock);
 270                 kfree(old);
 271         }
 272
 273         spin_lock(&minor_lock);
 274         if (find_next_bit(minors, end, minor) >= end) {
 275                 bitmap_set(minors, minor, nr);
 276                 rc = 0;
 277         } else
 278                 rc = -EBUSY;
 279         spin_unlock(&minor_lock);
 280
 281         return rc;
 282 }
 283
 284 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 285 {
 286         unsigned int end = minor + nr;
 287
 288         BUG_ON(end > nr_minors);
 289         spin_lock(&minor_lock);
 290         bitmap_clear(minors,  minor, nr);
 291         spin_unlock(&minor_lock);
 292 }
 293
 294 static void blkif_restart_queue_callback(void *arg)
 295 {
 296         struct blkfront_info *info = (struct blkfront_info *)arg;
 297         schedule_work(&info->work);
 298 }
 299
 300 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 301 {
 302         /* We don't have real geometry info, but let's at least return
 303            values consistent with the size of the device */
 304         sector_t nsect = get_capacity(bd->bd_disk);
 305         sector_t cylinders = nsect;
 306
 307         hg->heads = 0xff;
 308         hg->sectors = 0x3f;
 309         sector_div(cylinders, hg->heads * hg->sectors);
 310         hg->cylinders = cylinders;
 311         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 312                 hg->cylinders = 0xffff;
 313         return 0;
 314 }
 315
 316 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 317                        unsigned command, unsigned long argument)
 318 {
 319         struct blkfront_info *info = bdev->bd_disk->private_data;
 320         int i;
 321
 322         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 323                 command, (long)argument);
 324
 325         switch (command) {
 326         case CDROMMULTISESSION:
 327                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 328                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 329                         if (put_user(0, (char __user *)(argument + i)))
 330                                 return -EFAULT;
 331                 return 0;
 332
 333         case CDROM_GET_CAPABILITY: {
 334                 struct gendisk *gd = info->gd;
 335                 if (gd->flags & GENHD_FL_CD)
 336                         return 0;
 337                 return -EINVAL;
 338         }
 339
 340         default:
 341                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 342                   command);*/
 343                 return -EINVAL; /* same return as native Linux */
 344         }
 345
 346         return 0;
 347 }
 348
 349 /*
 350  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 351  * and writes are handled as expected.
 352  *
 353  * @req: a request struct
 354  */
 355 static int blkif_queue_request(struct request *req)
 356 {
 357         struct blkfront_info *info = req->rq_disk->private_data;
 358         struct blkif_request *ring_req;
 359         unsigned long id;
 360         unsigned int fsect, lsect;
 361         int i, ref;
 362
 363         /*
 364          * Used to store if we are able to queue the request by just using
 365          * existing persistent grants, or if we have to get new grants,
 366          * as there are not sufficiently many free.
 367          */
 368         bool new_persistent_gnts;
 369         grant_ref_t gref_head;
 370         struct grant *gnt_list_entry = NULL;
 371         struct scatterlist *sg;
 372
 373         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 374                 return 1;
 375
 376         /* Check if we have enought grants to allocate a requests */
 377         if (info->persistent_gnts_c < BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 378                 new_persistent_gnts = 1;
 379                 if (gnttab_alloc_grant_references(
 380                     BLKIF_MAX_SEGMENTS_PER_REQUEST - info->persistent_gnts_c,
 381                     &gref_head) < 0) {
 382                         gnttab_request_free_callback(
 383                                 &info->callback,
 384                                 blkif_restart_queue_callback,
 385                                 info,
 386                                 BLKIF_MAX_SEGMENTS_PER_REQUEST);
 387                         return 1;
 388                 }
 389         } else
 390                 new_persistent_gnts = 0;
 391
 392         /* Fill out a communications ring structure. */
 393         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 394         id = get_id_from_freelist(info);
 395         info->shadow[id].request = req;
 396
 397         ring_req->u.rw.id = id;
 398         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 399         ring_req->u.rw.handle = info->handle;
 400
 401         ring_req->operation = rq_data_dir(req) ?
 402                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 403
 404         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 405                 /*
 406                  * Ideally we can do an unordered flush-to-disk. In case the
 407                  * backend onlysupports barriers, use that. A barrier request
 408                  * a superset of FUA, so we can implement it the same
 409                  * way.  (It's also a FLUSH+FUA, since it is
 410                  * guaranteed ordered WRT previous writes.)
 411                  */
 412                 ring_req->operation = info->flush_op;
 413         }
 414
 415         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 416                 /* id, sector_number and handle are set above. */
 417                 ring_req->operation = BLKIF_OP_DISCARD;
 418                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 419                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 420                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 421                 else
 422                         ring_req->u.discard.flag = 0;
 423         } else {
 424                 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
 425                                                            info->shadow[id].sg);
 426                 BUG_ON(ring_req->u.rw.nr_segments >
 427                        BLKIF_MAX_SEGMENTS_PER_REQUEST);
 428
 429                 for_each_sg(info->shadow[id].sg, sg, ring_req->u.rw.nr_segments, i) {
 430                         fsect = sg->offset >> 9;
 431                         lsect = fsect + (sg->length >> 9) - 1;
 432
 433                         gnt_list_entry = get_grant(&gref_head, info);
 434                         ref = gnt_list_entry->gref;
 435
 436                         info->shadow[id].grants_used[i] = gnt_list_entry;
 437
 438                         if (rq_data_dir(req)) {
 439                                 char *bvec_data;
 440                                 void *shared_data;
 441
 442                                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 443
 444                                 shared_data = kmap_atomic(
 445                                         pfn_to_page(gnt_list_entry->pfn));
 446                                 bvec_data = kmap_atomic(sg_page(sg));
 447
 448                                 /*
 449                                  * this does not wipe data stored outside the
 450                                  * range sg->offset..sg->offset+sg->length.
 451                                  * Therefore, blkback *could* see data from
 452                                  * previous requests. This is OK as long as
 453                                  * persistent grants are shared with just one
 454                                  * domain. It may need refactoring if this
 455                                  * changes
 456                                  */
 457                                 memcpy(shared_data + sg->offset,
 458                                        bvec_data   + sg->offset,
 459                                        sg->length);
 460
 461                                 kunmap_atomic(bvec_data);
 462                                 kunmap_atomic(shared_data);
 463                         }
 464
 465                         ring_req->u.rw.seg[i] =
 466                                         (struct blkif_request_segment) {
 467                                                 .gref       = ref,
 468                                                 .first_sect = fsect,
 469                                                 .last_sect  = lsect };
 470                 }
 471         }
 472
 473         info->ring.req_prod_pvt++;
 474
 475         /* Keep a private copy so we can reissue requests when recovering. */
 476         info->shadow[id].req = *ring_req;
 477
 478         if (new_persistent_gnts)
 479                 gnttab_free_grant_references(gref_head);
 480
 481         return 0;
 482 }
 483
 484
 485 static inline void flush_requests(struct blkfront_info *info)
 486 {
 487         int notify;
 488
 489         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 490
 491         if (notify)
 492                 notify_remote_via_irq(info->irq);
 493 }
 494
 495 /*
 496  * do_blkif_request
 497  *  read a block; request is in a request queue
 498  */
 499 static void do_blkif_request(struct request_queue *rq)
 500 {
 501         struct blkfront_info *info = NULL;
 502         struct request *req;
 503         int queued;
 504
 505         pr_debug("Entered do_blkif_request\n");
 506
 507         queued = 0;
 508
 509         while ((req = blk_peek_request(rq)) != NULL) {
 510                 info = req->rq_disk->private_data;
 511
 512                 if (RING_FULL(&info->ring))
 513                         goto wait;
 514
 515                 blk_start_request(req);
 516
 517                 if ((req->cmd_type != REQ_TYPE_FS) ||
 518                     ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
 519                     !info->flush_op)) {
 520                         __blk_end_request_all(req, -EIO);
 521                         continue;
 522                 }
 523
 524                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 525                          "(%u/%u) buffer:%p [%s]\n",
 526                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 527                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 528                          req->buffer, rq_data_dir(req) ? "write" : "read");
 529
 530                 if (blkif_queue_request(req)) {
 531                         blk_requeue_request(rq, req);
 532 wait:
 533                         /* Avoid pointless unplugs. */
 534                         blk_stop_queue(rq);
 535                         break;
 536                 }
 537
 538                 queued++;
 539         }
 540
 541         if (queued != 0)
 542                 flush_requests(info);
 543 }
 544
 545 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 546 {
 547         struct request_queue *rq;
 548         struct blkfront_info *info = gd->private_data;
 549
 550         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 551         if (rq == NULL)
 552                 return -1;
 553
 554         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 555
 556         if (info->feature_discard) {
 557                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 558                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 559                 rq->limits.discard_granularity = info->discard_granularity;
 560                 rq->limits.discard_alignment = info->discard_alignment;
 561                 if (info->feature_secdiscard)
 562                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 563         }
 564
 565         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 566         blk_queue_logical_block_size(rq, sector_size);
 567         blk_queue_max_hw_sectors(rq, 512);
 568
 569         /* Each segment in a request is up to an aligned page in size. */
 570         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 571         blk_queue_max_segment_size(rq, PAGE_SIZE);
 572
 573         /* Ensure a merged request will fit in a single I/O ring slot. */
 574         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 575
 576         /* Make sure buffer addresses are sector-aligned. */
 577         blk_queue_dma_alignment(rq, 511);
 578
 579         /* Make sure we don't use bounce buffers. */
 580         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 581
 582         gd->queue = rq;
 583
 584         return 0;
 585 }
 586
 587
 588 static void xlvbd_flush(struct blkfront_info *info)
 589 {
 590         blk_queue_flush(info->rq, info->feature_flush);
 591         printk(KERN_INFO "blkfront: %s: %s: %s %s\n",
 592                info->gd->disk_name,
 593                info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 594                 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 595                 "flush diskcache" : "barrier or flush"),
 596                info->feature_flush ? "enabled" : "disabled",
 597                info->feature_persistent ? "using persistent grants" : "");
 598 }
 599
 600 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 601 {
 602         int major;
 603         major = BLKIF_MAJOR(vdevice);
 604         *minor = BLKIF_MINOR(vdevice);
 605         switch (major) {
 606                 case XEN_IDE0_MAJOR:
 607                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 608                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 609                                 EMULATED_HD_DISK_MINOR_OFFSET;
 610                         break;
 611                 case XEN_IDE1_MAJOR:
 612                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 613                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 614                                 EMULATED_HD_DISK_MINOR_OFFSET;
 615                         break;
 616                 case XEN_SCSI_DISK0_MAJOR:
 617                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 618                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 619                         break;
 620                 case XEN_SCSI_DISK1_MAJOR:
 621                 case XEN_SCSI_DISK2_MAJOR:
 622                 case XEN_SCSI_DISK3_MAJOR:
 623                 case XEN_SCSI_DISK4_MAJOR:
 624                 case XEN_SCSI_DISK5_MAJOR:
 625                 case XEN_SCSI_DISK6_MAJOR:
 626                 case XEN_SCSI_DISK7_MAJOR:
 627                         *offset = (*minor / PARTS_PER_DISK) +
 628                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 629                                 EMULATED_SD_DISK_NAME_OFFSET;
 630                         *minor = *minor +
 631                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 632                                 EMULATED_SD_DISK_MINOR_OFFSET;
 633                         break;
 634                 case XEN_SCSI_DISK8_MAJOR:
 635                 case XEN_SCSI_DISK9_MAJOR:
 636                 case XEN_SCSI_DISK10_MAJOR:
 637                 case XEN_SCSI_DISK11_MAJOR:
 638                 case XEN_SCSI_DISK12_MAJOR:
 639                 case XEN_SCSI_DISK13_MAJOR:
 640                 case XEN_SCSI_DISK14_MAJOR:
 641                 case XEN_SCSI_DISK15_MAJOR:
 642                         *offset = (*minor / PARTS_PER_DISK) +
 643                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 644                                 EMULATED_SD_DISK_NAME_OFFSET;
 645                         *minor = *minor +
 646                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 647                                 EMULATED_SD_DISK_MINOR_OFFSET;
 648                         break;
 649                 case XENVBD_MAJOR:
 650                         *offset = *minor / PARTS_PER_DISK;
 651                         break;
 652                 default:
 653                         printk(KERN_WARNING "blkfront: your disk configuration is "
 654                                         "incorrect, please use an xvd device instead\n");
 655                         return -ENODEV;
 656         }
 657         return 0;
 658 }
 659
 660 static char *encode_disk_name(char *ptr, unsigned int n)
 661 {
 662         if (n >= 26)
 663                 ptr = encode_disk_name(ptr, n / 26 - 1);
 664         *ptr = 'a' + n % 26;
 665         return ptr + 1;
 666 }
 667
 668 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 669                                struct blkfront_info *info,
 670                                u16 vdisk_info, u16 sector_size)
 671 {
 672         struct gendisk *gd;
 673         int nr_minors = 1;
 674         int err;
 675         unsigned int offset;
 676         int minor;
 677         int nr_parts;
 678         char *ptr;
 679
 680         BUG_ON(info->gd != NULL);
 681         BUG_ON(info->rq != NULL);
 682
 683         if ((info->vdevice>>EXT_SHIFT) > 1) {
 684                 /* this is above the extended range; something is wrong */
 685                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 686                 return -ENODEV;
 687         }
 688
 689         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 690                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 691                 if (err)
 692                         return err;
 693                 nr_parts = PARTS_PER_DISK;
 694         } else {
 695                 minor = BLKIF_MINOR_EXT(info->vdevice);
 696                 nr_parts = PARTS_PER_EXT_DISK;
 697                 offset = minor / nr_parts;
 698                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 699                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 700                                         "emulated IDE disks,\n\t choose an xvd device name"
 701                                         "from xvde on\n", info->vdevice);
 702         }
 703         if (minor >> MINORBITS) {
 704                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 705                         info->vdevice, minor);
 706                 return -ENODEV;
 707         }
 708
 709         if ((minor % nr_parts) == 0)
 710                 nr_minors = nr_parts;
 711
 712         err = xlbd_reserve_minors(minor, nr_minors);
 713         if (err)
 714                 goto out;
 715         err = -ENODEV;
 716
 717         gd = alloc_disk(nr_minors);
 718         if (gd == NULL)
 719                 goto release;
 720
 721         strcpy(gd->disk_name, DEV_NAME);
 722         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 723         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 724         if (nr_minors > 1)
 725                 *ptr = 0;
 726         else
 727                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 728                          "%d", minor & (nr_parts - 1));
 729
 730         gd->major = XENVBD_MAJOR;
 731         gd->first_minor = minor;
 732         gd->fops = &xlvbd_block_fops;
 733         gd->private_data = info;
 734         gd->driverfs_dev = &(info->xbdev->dev);
 735         set_capacity(gd, capacity);
 736
 737         if (xlvbd_init_blk_queue(gd, sector_size)) {
 738                 del_gendisk(gd);
 739                 goto release;
 740         }
 741
 742         info->rq = gd->queue;
 743         info->gd = gd;
 744
 745         xlvbd_flush(info);
 746
 747         if (vdisk_info & VDISK_READONLY)
 748                 set_disk_ro(gd, 1);
 749
 750         if (vdisk_info & VDISK_REMOVABLE)
 751                 gd->flags |= GENHD_FL_REMOVABLE;
 752
 753         if (vdisk_info & VDISK_CDROM)
 754                 gd->flags |= GENHD_FL_CD;
 755
 756         return 0;
 757
 758  release:
 759         xlbd_release_minors(minor, nr_minors);
 760  out:
 761         return err;
 762 }
 763
 764 static void xlvbd_release_gendisk(struct blkfront_info *info)
 765 {
 766         unsigned int minor, nr_minors;
 767         unsigned long flags;
 768
 769         if (info->rq == NULL)
 770                 return;
 771
 772         spin_lock_irqsave(&info->io_lock, flags);
 773
 774         /* No more blkif_request(). */
 775         blk_stop_queue(info->rq);
 776
 777         /* No more gnttab callback work. */
 778         gnttab_cancel_free_callback(&info->callback);
 779         spin_unlock_irqrestore(&info->io_lock, flags);
 780
 781         /* Flush gnttab callback work. Must be done with no locks held. */
 782         flush_work(&info->work);
 783
 784         del_gendisk(info->gd);
 785
 786         minor = info->gd->first_minor;
 787         nr_minors = info->gd->minors;
 788         xlbd_release_minors(minor, nr_minors);
 789
 790         blk_cleanup_queue(info->rq);
 791         info->rq = NULL;
 792
 793         put_disk(info->gd);
 794         info->gd = NULL;
 795 }
 796
 797 static void kick_pending_request_queues(struct blkfront_info *info)
 798 {
 799         if (!RING_FULL(&info->ring)) {
 800                 /* Re-enable calldowns. */
 801                 blk_start_queue(info->rq);
 802                 /* Kick things off immediately. */
 803                 do_blkif_request(info->rq);
 804         }
 805 }
 806
 807 static void blkif_restart_queue(struct work_struct *work)
 808 {
 809         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 810
 811         spin_lock_irq(&info->io_lock);
 812         if (info->connected == BLKIF_STATE_CONNECTED)
 813                 kick_pending_request_queues(info);
 814         spin_unlock_irq(&info->io_lock);
 815 }
 816
 817 static void blkif_free(struct blkfront_info *info, int suspend)
 818 {
 819         struct grant *persistent_gnt;
 820         struct grant *n;
 821
 822         /* Prevent new requests being issued until we fix things up. */
 823         spin_lock_irq(&info->io_lock);
 824         info->connected = suspend ?
 825                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 826         /* No more blkif_request(). */
 827         if (info->rq)
 828                 blk_stop_queue(info->rq);
 829
 830         /* Remove all persistent grants */
 831         if (!list_empty(&info->persistent_gnts)) {
 832                 list_for_each_entry_safe(persistent_gnt, n,
 833                                          &info->persistent_gnts, node) {
 834                         list_del(&persistent_gnt->node);
 835                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
 836                                 gnttab_end_foreign_access(persistent_gnt->gref,
 837                                                           0, 0UL);
 838                                 info->persistent_gnts_c--;
 839                         }
 840                         __free_page(pfn_to_page(persistent_gnt->pfn));
 841                         kfree(persistent_gnt);
 842                 }
 843         }
 844         BUG_ON(info->persistent_gnts_c != 0);
 845
 846         /* No more gnttab callback work. */
 847         gnttab_cancel_free_callback(&info->callback);
 848         spin_unlock_irq(&info->io_lock);
 849
 850         /* Flush gnttab callback work. Must be done with no locks held. */
 851         flush_work(&info->work);
 852
 853         /* Free resources associated with old device channel. */
 854         if (info->ring_ref != GRANT_INVALID_REF) {
 855                 gnttab_end_foreign_access(info->ring_ref, 0,
 856                                           (unsigned long)info->ring.sring);
 857                 info->ring_ref = GRANT_INVALID_REF;
 858                 info->ring.sring = NULL;
 859         }
 860         if (info->irq)
 861                 unbind_from_irqhandler(info->irq, info);
 862         info->evtchn = info->irq = 0;
 863
 864 }
 865
 866 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
 867                              struct blkif_response *bret)
 868 {
 869         int i = 0;
 870         struct scatterlist *sg;
 871         char *bvec_data;
 872         void *shared_data;
 873         int nseg;
 874
 875         nseg = s->req.u.rw.nr_segments;
 876
 877         if (bret->operation == BLKIF_OP_READ) {
 878                 /*
 879                  * Copy the data received from the backend into the bvec.
 880                  * Since bv_offset can be different than 0, and bv_len different
 881                  * than PAGE_SIZE, we have to keep track of the current offset,
 882                  * to be sure we are copying the data from the right shared page.
 883                  */
 884                 for_each_sg(s->sg, sg, nseg, i) {
 885                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 886                         shared_data = kmap_atomic(
 887                                 pfn_to_page(s->grants_used[i]->pfn));
 888                         bvec_data = kmap_atomic(sg_page(sg));
 889                         memcpy(bvec_data   + sg->offset,
 890                                shared_data + sg->offset,
 891                                sg->length);
 892                         kunmap_atomic(bvec_data);
 893                         kunmap_atomic(shared_data);
 894                 }
 895         }
 896         /* Add the persistent grant into the list of free grants */
 897         for (i = 0; i < s->req.u.rw.nr_segments; i++) {
 898                 list_add(&s->grants_used[i]->node, &info->persistent_gnts);
 899                 info->persistent_gnts_c++;
 900         }
 901 }
 902
 903 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 904 {
 905         struct request *req;
 906         struct blkif_response *bret;
 907         RING_IDX i, rp;
 908         unsigned long flags;
 909         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 910         int error;
 911
 912         spin_lock_irqsave(&info->io_lock, flags);
 913
 914         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 915                 spin_unlock_irqrestore(&info->io_lock, flags);
 916                 return IRQ_HANDLED;
 917         }
 918
 919  again:
 920         rp = info->ring.sring->rsp_prod;
 921         rmb(); /* Ensure we see queued responses up to 'rp'. */
 922
 923         for (i = info->ring.rsp_cons; i != rp; i++) {
 924                 unsigned long id;
 925
 926                 bret = RING_GET_RESPONSE(&info->ring, i);
 927                 id   = bret->id;
 928                 /*
 929                  * The backend has messed up and given us an id that we would
 930                  * never have given to it (we stamp it up to BLK_RING_SIZE -
 931                  * look in get_id_from_freelist.
 932                  */
 933                 if (id >= BLK_RING_SIZE) {
 934                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
 935                              info->gd->disk_name, op_name(bret->operation), id);
 936                         /* We can't safely get the 'struct request' as
 937                          * the id is busted. */
 938                         continue;
 939                 }
 940                 req  = info->shadow[id].request;
 941
 942                 if (bret->operation != BLKIF_OP_DISCARD)
 943                         blkif_completion(&info->shadow[id], info, bret);
 944
 945                 if (add_id_to_freelist(info, id)) {
 946                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
 947                              info->gd->disk_name, op_name(bret->operation), id);
 948                         continue;
 949                 }
 950
 951                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 952                 switch (bret->operation) {
 953                 case BLKIF_OP_DISCARD:
 954                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 955                                 struct request_queue *rq = info->rq;
 956                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
 957                                            info->gd->disk_name, op_name(bret->operation));
 958                                 error = -EOPNOTSUPP;
 959                                 info->feature_discard = 0;
 960                                 info->feature_secdiscard = 0;
 961                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
 962                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
 963                         }
 964                         __blk_end_request_all(req, error);
 965                         break;
 966                 case BLKIF_OP_FLUSH_DISKCACHE:
 967                 case BLKIF_OP_WRITE_BARRIER:
 968                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 969                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
 970                                        info->gd->disk_name, op_name(bret->operation));
 971                                 error = -EOPNOTSUPP;
 972                         }
 973                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
 974                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
 975                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
 976                                        info->gd->disk_name, op_name(bret->operation));
 977                                 error = -EOPNOTSUPP;
 978                         }
 979                         if (unlikely(error)) {
 980                                 if (error == -EOPNOTSUPP)
 981                                         error = 0;
 982                                 info->feature_flush = 0;
 983                                 info->flush_op = 0;
 984                                 xlvbd_flush(info);
 985                         }
 986                         /* fall through */
 987                 case BLKIF_OP_READ:
 988                 case BLKIF_OP_WRITE:
 989                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 990                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 991                                         "request: %x\n", bret->status);
 992
 993                         __blk_end_request_all(req, error);
 994                         break;
 995                 default:
 996                         BUG();
 997                 }
 998         }
 999
1000         info->ring.rsp_cons = i;
1001
1002         if (i != info->ring.req_prod_pvt) {
1003                 int more_to_do;
1004                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1005                 if (more_to_do)
1006                         goto again;
1007         } else
1008                 info->ring.sring->rsp_event = i + 1;
1009
1010         kick_pending_request_queues(info);
1011
1012         spin_unlock_irqrestore(&info->io_lock, flags);
1013
1014         return IRQ_HANDLED;
1015 }
1016
1017
1018 static int setup_blkring(struct xenbus_device *dev,
1019                          struct blkfront_info *info)
1020 {
1021         struct blkif_sring *sring;
1022         int err, i;
1023
1024         info->ring_ref = GRANT_INVALID_REF;
1025
1026         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1027         if (!sring) {
1028                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1029                 return -ENOMEM;
1030         }
1031         SHARED_RING_INIT(sring);
1032         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1033
1034         for (i = 0; i < BLK_RING_SIZE; i++)
1035                 sg_init_table(info->shadow[i].sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
1036
1037         /* Allocate memory for grants */
1038         err = fill_grant_buffer(info, BLK_RING_SIZE *
1039                                       BLKIF_MAX_SEGMENTS_PER_REQUEST);
1040         if (err)
1041                 goto fail;
1042
1043         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1044         if (err < 0) {
1045                 free_page((unsigned long)sring);
1046                 info->ring.sring = NULL;
1047                 goto fail;
1048         }
1049         info->ring_ref = err;
1050
1051         err = xenbus_alloc_evtchn(dev, &info->evtchn);
1052         if (err)
1053                 goto fail;
1054
1055         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1056                                         "blkif", info);
1057         if (err <= 0) {
1058                 xenbus_dev_fatal(dev, err,
1059                                  "bind_evtchn_to_irqhandler failed");
1060                 goto fail;
1061         }
1062         info->irq = err;
1063
1064         return 0;
1065 fail:
1066         blkif_free(info, 0);
1067         return err;
1068 }
1069
1070
1071 /* Common code used when first setting up, and when resuming. */
1072 static int talk_to_blkback(struct xenbus_device *dev,
1073                            struct blkfront_info *info)
1074 {
1075         const char *message = NULL;
1076         struct xenbus_transaction xbt;
1077         int err;
1078
1079         /* Create shared ring, alloc event channel. */
1080         err = setup_blkring(dev, info);
1081         if (err)
1082                 goto out;
1083
1084 again:
1085         err = xenbus_transaction_start(&xbt);
1086         if (err) {
1087                 xenbus_dev_fatal(dev, err, "starting transaction");
1088                 goto destroy_blkring;
1089         }
1090
1091         err = xenbus_printf(xbt, dev->nodename,
1092                             "ring-ref", "%u", info->ring_ref);
1093         if (err) {
1094                 message = "writing ring-ref";
1095                 goto abort_transaction;
1096         }
1097         err = xenbus_printf(xbt, dev->nodename,
1098                             "event-channel", "%u", info->evtchn);
1099         if (err) {
1100                 message = "writing event-channel";
1101                 goto abort_transaction;
1102         }
1103         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1104                             XEN_IO_PROTO_ABI_NATIVE);
1105         if (err) {
1106                 message = "writing protocol";
1107                 goto abort_transaction;
1108         }
1109         err = xenbus_printf(xbt, dev->nodename,
1110                             "feature-persistent", "%u", 1);
1111         if (err)
1112                 dev_warn(&dev->dev,
1113                          "writing persistent grants feature to xenbus");
1114
1115         err = xenbus_transaction_end(xbt, 0);
1116         if (err) {
1117                 if (err == -EAGAIN)
1118                         goto again;
1119                 xenbus_dev_fatal(dev, err, "completing transaction");
1120                 goto destroy_blkring;
1121         }
1122
1123         xenbus_switch_state(dev, XenbusStateInitialised);
1124
1125         return 0;
1126
1127  abort_transaction:
1128         xenbus_transaction_end(xbt, 1);
1129         if (message)
1130                 xenbus_dev_fatal(dev, err, "%s", message);
1131  destroy_blkring:
1132         blkif_free(info, 0);
1133  out:
1134         return err;
1135 }
1136
1137 /**
1138  * Entry point to this code when a new device is created.  Allocate the basic
1139  * structures and the ring buffer for communication with the backend, and
1140  * inform the backend of the appropriate details for those.  Switch to
1141  * Initialised state.
1142  */
1143 static int blkfront_probe(struct xenbus_device *dev,
1144                           const struct xenbus_device_id *id)
1145 {
1146         int err, vdevice, i;
1147         struct blkfront_info *info;
1148
1149         /* FIXME: Use dynamic device id if this is not set. */
1150         err = xenbus_scanf(XBT_NIL, dev->nodename,
1151                            "virtual-device", "%i", &vdevice);
1152         if (err != 1) {
1153                 /* go looking in the extended area instead */
1154                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1155                                    "%i", &vdevice);
1156                 if (err != 1) {
1157                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1158                         return err;
1159                 }
1160         }
1161
1162         if (xen_hvm_domain()) {
1163                 char *type;
1164                 int len;
1165                 /* no unplug has been done: do not hook devices != xen vbds */
1166                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
1167                         int major;
1168
1169                         if (!VDEV_IS_EXTENDED(vdevice))
1170                                 major = BLKIF_MAJOR(vdevice);
1171                         else
1172                                 major = XENVBD_MAJOR;
1173
1174                         if (major != XENVBD_MAJOR) {
1175                                 printk(KERN_INFO
1176                                                 "%s: HVM does not support vbd %d as xen block device\n",
1177                                                 __FUNCTION__, vdevice);
1178                                 return -ENODEV;
1179                         }
1180                 }
1181                 /* do not create a PV cdrom device if we are an HVM guest */
1182                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1183                 if (IS_ERR(type))
1184                         return -ENODEV;
1185                 if (strncmp(type, "cdrom", 5) == 0) {
1186                         kfree(type);
1187                         return -ENODEV;
1188                 }
1189                 kfree(type);
1190         }
1191         info = kzalloc(sizeof(*info), GFP_KERNEL);
1192         if (!info) {
1193                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1194                 return -ENOMEM;
1195         }
1196
1197         mutex_init(&info->mutex);
1198         spin_lock_init(&info->io_lock);
1199         info->xbdev = dev;
1200         info->vdevice = vdevice;
1201         INIT_LIST_HEAD(&info->persistent_gnts);
1202         info->persistent_gnts_c = 0;
1203         info->connected = BLKIF_STATE_DISCONNECTED;
1204         INIT_WORK(&info->work, blkif_restart_queue);
1205
1206         for (i = 0; i < BLK_RING_SIZE; i++)
1207                 info->shadow[i].req.u.rw.id = i+1;
1208         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1209
1210         /* Front end dir is a number, which is used as the id. */
1211         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1212         dev_set_drvdata(&dev->dev, info);
1213
1214         err = talk_to_blkback(dev, info);
1215         if (err) {
1216                 kfree(info);
1217                 dev_set_drvdata(&dev->dev, NULL);
1218                 return err;
1219         }
1220
1221         return 0;
1222 }
1223
1224
1225 static int blkif_recover(struct blkfront_info *info)
1226 {
1227         int i;
1228         struct blkif_request *req;
1229         struct blk_shadow *copy;
1230         int j;
1231
1232         /* Stage 1: Make a safe copy of the shadow state. */
1233         copy = kmemdup(info->shadow, sizeof(info->shadow),
1234                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1235         if (!copy)
1236                 return -ENOMEM;
1237
1238         /* Stage 2: Set up free list. */
1239         memset(&info->shadow, 0, sizeof(info->shadow));
1240         for (i = 0; i < BLK_RING_SIZE; i++)
1241                 info->shadow[i].req.u.rw.id = i+1;
1242         info->shadow_free = info->ring.req_prod_pvt;
1243         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1244
1245         /* Stage 3: Find pending requests and requeue them. */
1246         for (i = 0; i < BLK_RING_SIZE; i++) {
1247                 /* Not in use? */
1248                 if (!copy[i].request)
1249                         continue;
1250
1251                 /* Grab a request slot and copy shadow state into it. */
1252                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1253                 *req = copy[i].req;
1254
1255                 /* We get a new request id, and must reset the shadow state. */
1256                 req->u.rw.id = get_id_from_freelist(info);
1257                 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1258
1259                 if (req->operation != BLKIF_OP_DISCARD) {
1260                 /* Rewrite any grant references invalidated by susp/resume. */
1261                         for (j = 0; j < req->u.rw.nr_segments; j++)
1262                                 gnttab_grant_foreign_access_ref(
1263                                         req->u.rw.seg[j].gref,
1264                                         info->xbdev->otherend_id,
1265                                         pfn_to_mfn(copy[i].grants_used[j]->pfn),
1266                                         0);
1267                 }
1268                 info->shadow[req->u.rw.id].req = *req;
1269
1270                 info->ring.req_prod_pvt++;
1271         }
1272
1273         kfree(copy);
1274
1275         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1276
1277         spin_lock_irq(&info->io_lock);
1278
1279         /* Now safe for us to use the shared ring */
1280         info->connected = BLKIF_STATE_CONNECTED;
1281
1282         /* Send off requeued requests */
1283         flush_requests(info);
1284
1285         /* Kick any other new requests queued since we resumed */
1286         kick_pending_request_queues(info);
1287
1288         spin_unlock_irq(&info->io_lock);
1289
1290         return 0;
1291 }
1292
1293 /**
1294  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1295  * driver restart.  We tear down our blkif structure and recreate it, but
1296  * leave the device-layer structures intact so that this is transparent to the
1297  * rest of the kernel.
1298  */
1299 static int blkfront_resume(struct xenbus_device *dev)
1300 {
1301         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1302         int err;
1303
1304         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1305
1306         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1307
1308         err = talk_to_blkback(dev, info);
1309         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1310                 err = blkif_recover(info);
1311
1312         return err;
1313 }
1314
1315 static void
1316 blkfront_closing(struct blkfront_info *info)
1317 {
1318         struct xenbus_device *xbdev = info->xbdev;
1319         struct block_device *bdev = NULL;
1320
1321         mutex_lock(&info->mutex);
1322
1323         if (xbdev->state == XenbusStateClosing) {
1324                 mutex_unlock(&info->mutex);
1325                 return;
1326         }
1327
1328         if (info->gd)
1329                 bdev = bdget_disk(info->gd, 0);
1330
1331         mutex_unlock(&info->mutex);
1332
1333         if (!bdev) {
1334                 xenbus_frontend_closed(xbdev);
1335                 return;
1336         }
1337
1338         mutex_lock(&bdev->bd_mutex);
1339
1340         if (bdev->bd_openers) {
1341                 xenbus_dev_error(xbdev, -EBUSY,
1342                                  "Device in use; refusing to close");
1343                 xenbus_switch_state(xbdev, XenbusStateClosing);
1344         } else {
1345                 xlvbd_release_gendisk(info);
1346                 xenbus_frontend_closed(xbdev);
1347         }
1348
1349         mutex_unlock(&bdev->bd_mutex);
1350         bdput(bdev);
1351 }
1352
1353 static void blkfront_setup_discard(struct blkfront_info *info)
1354 {
1355         int err;
1356         char *type;
1357         unsigned int discard_granularity;
1358         unsigned int discard_alignment;
1359         unsigned int discard_secure;
1360
1361         type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1362         if (IS_ERR(type))
1363                 return;
1364
1365         info->feature_secdiscard = 0;
1366         if (strncmp(type, "phy", 3) == 0) {
1367                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1368                         "discard-granularity", "%u", &discard_granularity,
1369                         "discard-alignment", "%u", &discard_alignment,
1370                         NULL);
1371                 if (!err) {
1372                         info->feature_discard = 1;
1373                         info->discard_granularity = discard_granularity;
1374                         info->discard_alignment = discard_alignment;
1375                 }
1376                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1377                             "discard-secure", "%d", &discard_secure,
1378                             NULL);
1379                 if (!err)
1380                         info->feature_secdiscard = discard_secure;
1381
1382         } else if (strncmp(type, "file", 4) == 0)
1383                 info->feature_discard = 1;
1384
1385         kfree(type);
1386 }
1387
1388 /*
1389  * Invoked when the backend is finally 'ready' (and has told produced
1390  * the details about the physical device - #sectors, size, etc).
1391  */
1392 static void blkfront_connect(struct blkfront_info *info)
1393 {
1394         unsigned long long sectors;
1395         unsigned long sector_size;
1396         unsigned int binfo;
1397         int err;
1398         int barrier, flush, discard, persistent;
1399
1400         switch (info->connected) {
1401         case BLKIF_STATE_CONNECTED:
1402                 /*
1403                  * Potentially, the back-end may be signalling
1404                  * a capacity change; update the capacity.
1405                  */
1406                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1407                                    "sectors", "%Lu", &sectors);
1408                 if (XENBUS_EXIST_ERR(err))
1409                         return;
1410                 printk(KERN_INFO "Setting capacity to %Lu\n",
1411                        sectors);
1412                 set_capacity(info->gd, sectors);
1413                 revalidate_disk(info->gd);
1414
1415                 /* fall through */
1416         case BLKIF_STATE_SUSPENDED:
1417                 return;
1418
1419         default:
1420                 break;
1421         }
1422
1423         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1424                 __func__, info->xbdev->otherend);
1425
1426         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1427                             "sectors", "%llu", &sectors,
1428                             "info", "%u", &binfo,
1429                             "sector-size", "%lu", &sector_size,
1430                             NULL);
1431         if (err) {
1432                 xenbus_dev_fatal(info->xbdev, err,
1433                                  "reading backend fields at %s",
1434                                  info->xbdev->otherend);
1435                 return;
1436         }
1437
1438         info->feature_flush = 0;
1439         info->flush_op = 0;
1440
1441         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1442                             "feature-barrier", "%d", &barrier,
1443                             NULL);
1444
1445         /*
1446          * If there's no "feature-barrier" defined, then it means
1447          * we're dealing with a very old backend which writes
1448          * synchronously; nothing to do.
1449          *
1450          * If there are barriers, then we use flush.
1451          */
1452         if (!err && barrier) {
1453                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1454                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1455         }
1456         /*
1457          * And if there is "feature-flush-cache" use that above
1458          * barriers.
1459          */
1460         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1461                             "feature-flush-cache", "%d", &flush,
1462                             NULL);
1463
1464         if (!err && flush) {
1465                 info->feature_flush = REQ_FLUSH;
1466                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1467         }
1468
1469         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1470                             "feature-discard", "%d", &discard,
1471                             NULL);
1472
1473         if (!err && discard)
1474                 blkfront_setup_discard(info);
1475
1476         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1477                             "feature-persistent", "%u", &persistent,
1478                             NULL);
1479         if (err)
1480                 info->feature_persistent = 0;
1481         else
1482                 info->feature_persistent = persistent;
1483
1484         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1485         if (err) {
1486                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1487                                  info->xbdev->otherend);
1488                 return;
1489         }
1490
1491         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1492
1493         /* Kick pending requests. */
1494         spin_lock_irq(&info->io_lock);
1495         info->connected = BLKIF_STATE_CONNECTED;
1496         kick_pending_request_queues(info);
1497         spin_unlock_irq(&info->io_lock);
1498
1499         add_disk(info->gd);
1500
1501         info->is_ready = 1;
1502 }
1503
1504 /**
1505  * Callback received when the backend's state changes.
1506  */
1507 static void blkback_changed(struct xenbus_device *dev,
1508                             enum xenbus_state backend_state)
1509 {
1510         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1511
1512         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1513
1514         switch (backend_state) {
1515         case XenbusStateInitialising:
1516         case XenbusStateInitWait:
1517         case XenbusStateInitialised:
1518         case XenbusStateReconfiguring:
1519         case XenbusStateReconfigured:
1520         case XenbusStateUnknown:
1521         case XenbusStateClosed:
1522                 break;
1523
1524         case XenbusStateConnected:
1525                 blkfront_connect(info);
1526                 break;
1527
1528         case XenbusStateClosing:
1529                 blkfront_closing(info);
1530                 break;
1531         }
1532 }
1533
1534 static int blkfront_remove(struct xenbus_device *xbdev)
1535 {
1536         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1537         struct block_device *bdev = NULL;
1538         struct gendisk *disk;
1539
1540         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1541
1542         blkif_free(info, 0);
1543
1544         mutex_lock(&info->mutex);
1545
1546         disk = info->gd;
1547         if (disk)
1548                 bdev = bdget_disk(disk, 0);
1549
1550         info->xbdev = NULL;
1551         mutex_unlock(&info->mutex);
1552
1553         if (!bdev) {
1554                 kfree(info);
1555                 return 0;
1556         }
1557
1558         /*
1559          * The xbdev was removed before we reached the Closed
1560          * state. See if it's safe to remove the disk. If the bdev
1561          * isn't closed yet, we let release take care of it.
1562          */
1563
1564         mutex_lock(&bdev->bd_mutex);
1565         info = disk->private_data;
1566
1567         dev_warn(disk_to_dev(disk),
1568                  "%s was hot-unplugged, %d stale handles\n",
1569                  xbdev->nodename, bdev->bd_openers);
1570
1571         if (info && !bdev->bd_openers) {
1572                 xlvbd_release_gendisk(info);
1573                 disk->private_data = NULL;
1574                 kfree(info);
1575         }
1576
1577         mutex_unlock(&bdev->bd_mutex);
1578         bdput(bdev);
1579
1580         return 0;
1581 }
1582
1583 static int blkfront_is_ready(struct xenbus_device *dev)
1584 {
1585         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1586
1587         return info->is_ready && info->xbdev;
1588 }
1589
1590 static int blkif_open(struct block_device *bdev, fmode_t mode)
1591 {
1592         struct gendisk *disk = bdev->bd_disk;
1593         struct blkfront_info *info;
1594         int err = 0;
1595
1596         mutex_lock(&blkfront_mutex);
1597
1598         info = disk->private_data;
1599         if (!info) {
1600                 /* xbdev gone */
1601                 err = -ERESTARTSYS;
1602                 goto out;
1603         }
1604
1605         mutex_lock(&info->mutex);
1606
1607         if (!info->gd)
1608                 /* xbdev is closed */
1609                 err = -ERESTARTSYS;
1610
1611         mutex_unlock(&info->mutex);
1612
1613 out:
1614         mutex_unlock(&blkfront_mutex);
1615         return err;
1616 }
1617
1618 static void blkif_release(struct gendisk *disk, fmode_t mode)
1619 {
1620         struct blkfront_info *info = disk->private_data;
1621         struct block_device *bdev;
1622         struct xenbus_device *xbdev;
1623
1624         mutex_lock(&blkfront_mutex);
1625
1626         bdev = bdget_disk(disk, 0);
1627
1628         if (bdev->bd_openers)
1629                 goto out;
1630
1631         /*
1632          * Check if we have been instructed to close. We will have
1633          * deferred this request, because the bdev was still open.
1634          */
1635
1636         mutex_lock(&info->mutex);
1637         xbdev = info->xbdev;
1638
1639         if (xbdev && xbdev->state == XenbusStateClosing) {
1640                 /* pending switch to state closed */
1641                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1642                 xlvbd_release_gendisk(info);
1643                 xenbus_frontend_closed(info->xbdev);
1644         }
1645
1646         mutex_unlock(&info->mutex);
1647
1648         if (!xbdev) {
1649                 /* sudden device removal */
1650                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1651                 xlvbd_release_gendisk(info);
1652                 disk->private_data = NULL;
1653                 kfree(info);
1654         }
1655
1656 out:
1657         bdput(bdev);
1658         mutex_unlock(&blkfront_mutex);
1659 }
1660
1661 static const struct block_device_operations xlvbd_block_fops =
1662 {
1663         .owner = THIS_MODULE,
1664         .open = blkif_open,
1665         .release = blkif_release,
1666         .getgeo = blkif_getgeo,
1667         .ioctl = blkif_ioctl,
1668 };
1669
1670
1671 static const struct xenbus_device_id blkfront_ids[] = {
1672         { "vbd" },
1673         { "" }
1674 };
1675
1676 static DEFINE_XENBUS_DRIVER(blkfront, ,
1677         .probe = blkfront_probe,
1678         .remove = blkfront_remove,
1679         .resume = blkfront_resume,
1680         .otherend_changed = blkback_changed,
1681         .is_ready = blkfront_is_ready,
1682 );
1683
1684 static int __init xlblk_init(void)
1685 {
1686         int ret;
1687
1688         if (!xen_domain())
1689                 return -ENODEV;
1690
1691         if (xen_hvm_domain() && !xen_platform_pci_unplug)
1692                 return -ENODEV;
1693
1694         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1695                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1696                        XENVBD_MAJOR, DEV_NAME);
1697                 return -ENODEV;
1698         }
1699
1700         ret = xenbus_register_frontend(&blkfront_driver);
1701         if (ret) {
1702                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1703                 return ret;
1704         }
1705
1706         return 0;
1707 }
1708 module_init(xlblk_init);
1709
1710
1711 static void __exit xlblk_exit(void)
1712 {
1713         xenbus_unregister_driver(&blkfront_driver);
1714         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1715         kfree(minors);
1716 }
1717 module_exit(xlblk_exit);
1718
1719 MODULE_DESCRIPTION("Xen virtual block device frontend");
1720 MODULE_LICENSE("GPL");
1721 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1722 MODULE_ALIAS("xen:vbd");
1723 MODULE_ALIAS("xenblk");