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/llist.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 llist_node node;
  72 };
  73
  74 struct blk_shadow {
  75         struct blkif_request req;
  76         struct request *request;
  77         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  78         struct grant *grants_used[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         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 103         unsigned int evtchn, irq;
 104         struct request_queue *rq;
 105         struct work_struct work;
 106         struct gnttab_free_callback callback;
 107         struct blk_shadow shadow[BLK_RING_SIZE];
 108         struct llist_head persistent_gnts;
 109         unsigned int persistent_gnts_c;
 110         unsigned long shadow_free;
 111         unsigned int feature_flush;
 112         unsigned int flush_op;
 113         unsigned int feature_discard:1;
 114         unsigned int feature_secdiscard:1;
 115         unsigned int discard_granularity;
 116         unsigned int discard_alignment;
 117         unsigned int feature_persistent:1;
 118         int is_ready;
 119 };
 120
 121 static unsigned int nr_minors;
 122 static unsigned long *minors;
 123 static DEFINE_SPINLOCK(minor_lock);
 124
 125 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 126         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 127 #define GRANT_INVALID_REF       0
 128
 129 #define PARTS_PER_DISK          16
 130 #define PARTS_PER_EXT_DISK      256
 131
 132 #define BLKIF_MAJOR(dev) ((dev)>>8)
 133 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 134
 135 #define EXT_SHIFT 28
 136 #define EXTENDED (1<<EXT_SHIFT)
 137 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 138 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 139 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 140 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 141 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 142 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 143
 144 #define DEV_NAME        "xvd"   /* name in /dev */
 145
 146 static int get_id_from_freelist(struct blkfront_info *info)
 147 {
 148         unsigned long free = info->shadow_free;
 149         BUG_ON(free >= BLK_RING_SIZE);
 150         info->shadow_free = info->shadow[free].req.u.rw.id;
 151         info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 152         return free;
 153 }
 154
 155 static int add_id_to_freelist(struct blkfront_info *info,
 156                                unsigned long id)
 157 {
 158         if (info->shadow[id].req.u.rw.id != id)
 159                 return -EINVAL;
 160         if (info->shadow[id].request == NULL)
 161                 return -EINVAL;
 162         info->shadow[id].req.u.rw.id  = info->shadow_free;
 163         info->shadow[id].request = NULL;
 164         info->shadow_free = id;
 165         return 0;
 166 }
 167
 168 static const char *op_name(int op)
 169 {
 170         static const char *const names[] = {
 171                 [BLKIF_OP_READ] = "read",
 172                 [BLKIF_OP_WRITE] = "write",
 173                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 174                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 175                 [BLKIF_OP_DISCARD] = "discard" };
 176
 177         if (op < 0 || op >= ARRAY_SIZE(names))
 178                 return "unknown";
 179
 180         if (!names[op])
 181                 return "reserved";
 182
 183         return names[op];
 184 }
 185 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 186 {
 187         unsigned int end = minor + nr;
 188         int rc;
 189
 190         if (end > nr_minors) {
 191                 unsigned long *bitmap, *old;
 192
 193                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 194                                  GFP_KERNEL);
 195                 if (bitmap == NULL)
 196                         return -ENOMEM;
 197
 198                 spin_lock(&minor_lock);
 199                 if (end > nr_minors) {
 200                         old = minors;
 201                         memcpy(bitmap, minors,
 202                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 203                         minors = bitmap;
 204                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 205                 } else
 206                         old = bitmap;
 207                 spin_unlock(&minor_lock);
 208                 kfree(old);
 209         }
 210
 211         spin_lock(&minor_lock);
 212         if (find_next_bit(minors, end, minor) >= end) {
 213                 bitmap_set(minors, minor, nr);
 214                 rc = 0;
 215         } else
 216                 rc = -EBUSY;
 217         spin_unlock(&minor_lock);
 218
 219         return rc;
 220 }
 221
 222 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 223 {
 224         unsigned int end = minor + nr;
 225
 226         BUG_ON(end > nr_minors);
 227         spin_lock(&minor_lock);
 228         bitmap_clear(minors,  minor, nr);
 229         spin_unlock(&minor_lock);
 230 }
 231
 232 static void blkif_restart_queue_callback(void *arg)
 233 {
 234         struct blkfront_info *info = (struct blkfront_info *)arg;
 235         schedule_work(&info->work);
 236 }
 237
 238 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 239 {
 240         /* We don't have real geometry info, but let's at least return
 241            values consistent with the size of the device */
 242         sector_t nsect = get_capacity(bd->bd_disk);
 243         sector_t cylinders = nsect;
 244
 245         hg->heads = 0xff;
 246         hg->sectors = 0x3f;
 247         sector_div(cylinders, hg->heads * hg->sectors);
 248         hg->cylinders = cylinders;
 249         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 250                 hg->cylinders = 0xffff;
 251         return 0;
 252 }
 253
 254 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 255                        unsigned command, unsigned long argument)
 256 {
 257         struct blkfront_info *info = bdev->bd_disk->private_data;
 258         int i;
 259
 260         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 261                 command, (long)argument);
 262
 263         switch (command) {
 264         case CDROMMULTISESSION:
 265                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 266                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 267                         if (put_user(0, (char __user *)(argument + i)))
 268                                 return -EFAULT;
 269                 return 0;
 270
 271         case CDROM_GET_CAPABILITY: {
 272                 struct gendisk *gd = info->gd;
 273                 if (gd->flags & GENHD_FL_CD)
 274                         return 0;
 275                 return -EINVAL;
 276         }
 277
 278         default:
 279                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 280                   command);*/
 281                 return -EINVAL; /* same return as native Linux */
 282         }
 283
 284         return 0;
 285 }
 286
 287 /*
 288  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 289  * and writes are handled as expected.
 290  *
 291  * @req: a request struct
 292  */
 293 static int blkif_queue_request(struct request *req)
 294 {
 295         struct blkfront_info *info = req->rq_disk->private_data;
 296         unsigned long buffer_mfn;
 297         struct blkif_request *ring_req;
 298         unsigned long id;
 299         unsigned int fsect, lsect;
 300         int i, ref;
 301
 302         /*
 303          * Used to store if we are able to queue the request by just using
 304          * existing persistent grants, or if we have to get new grants,
 305          * as there are not sufficiently many free.
 306          */
 307         bool new_persistent_gnts;
 308         grant_ref_t gref_head;
 309         struct page *granted_page;
 310         struct grant *gnt_list_entry = NULL;
 311         struct scatterlist *sg;
 312
 313         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 314                 return 1;
 315
 316         /* Check if we have enought grants to allocate a requests */
 317         if (info->persistent_gnts_c < BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 318                 new_persistent_gnts = 1;
 319                 if (gnttab_alloc_grant_references(
 320                     BLKIF_MAX_SEGMENTS_PER_REQUEST - info->persistent_gnts_c,
 321                     &gref_head) < 0) {
 322                         gnttab_request_free_callback(
 323                                 &info->callback,
 324                                 blkif_restart_queue_callback,
 325                                 info,
 326                                 BLKIF_MAX_SEGMENTS_PER_REQUEST);
 327                         return 1;
 328                 }
 329         } else
 330                 new_persistent_gnts = 0;
 331
 332         /* Fill out a communications ring structure. */
 333         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 334         id = get_id_from_freelist(info);
 335         info->shadow[id].request = req;
 336
 337         ring_req->u.rw.id = id;
 338         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 339         ring_req->u.rw.handle = info->handle;
 340
 341         ring_req->operation = rq_data_dir(req) ?
 342                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 343
 344         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 345                 /*
 346                  * Ideally we can do an unordered flush-to-disk. In case the
 347                  * backend onlysupports barriers, use that. A barrier request
 348                  * a superset of FUA, so we can implement it the same
 349                  * way.  (It's also a FLUSH+FUA, since it is
 350                  * guaranteed ordered WRT previous writes.)
 351                  */
 352                 ring_req->operation = info->flush_op;
 353         }
 354
 355         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 356                 /* id, sector_number and handle are set above. */
 357                 ring_req->operation = BLKIF_OP_DISCARD;
 358                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 359                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 360                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 361                 else
 362                         ring_req->u.discard.flag = 0;
 363         } else {
 364                 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
 365                                                            info->sg);
 366                 BUG_ON(ring_req->u.rw.nr_segments >
 367                        BLKIF_MAX_SEGMENTS_PER_REQUEST);
 368
 369                 for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
 370                         fsect = sg->offset >> 9;
 371                         lsect = fsect + (sg->length >> 9) - 1;
 372
 373                         if (info->persistent_gnts_c) {
 374                                 BUG_ON(llist_empty(&info->persistent_gnts));
 375                                 gnt_list_entry = llist_entry(
 376                                         llist_del_first(&info->persistent_gnts),
 377                                         struct grant, node);
 378
 379                                 ref = gnt_list_entry->gref;
 380                                 buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
 381                                 info->persistent_gnts_c--;
 382                         } else {
 383                                 ref = gnttab_claim_grant_reference(&gref_head);
 384                                 BUG_ON(ref == -ENOSPC);
 385
 386                                 gnt_list_entry =
 387                                         kmalloc(sizeof(struct grant),
 388                                                          GFP_ATOMIC);
 389                                 if (!gnt_list_entry)
 390                                         return -ENOMEM;
 391
 392                                 granted_page = alloc_page(GFP_ATOMIC);
 393                                 if (!granted_page) {
 394                                         kfree(gnt_list_entry);
 395                                         return -ENOMEM;
 396                                 }
 397
 398                                 gnt_list_entry->pfn =
 399                                         page_to_pfn(granted_page);
 400                                 gnt_list_entry->gref = ref;
 401
 402                                 buffer_mfn = pfn_to_mfn(page_to_pfn(
 403                                                                 granted_page));
 404                                 gnttab_grant_foreign_access_ref(ref,
 405                                         info->xbdev->otherend_id,
 406                                         buffer_mfn, 0);
 407                         }
 408
 409                         info->shadow[id].grants_used[i] = gnt_list_entry;
 410
 411                         if (rq_data_dir(req)) {
 412                                 char *bvec_data;
 413                                 void *shared_data;
 414
 415                                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 416
 417                                 shared_data = kmap_atomic(
 418                                         pfn_to_page(gnt_list_entry->pfn));
 419                                 bvec_data = kmap_atomic(sg_page(sg));
 420
 421                                 /*
 422                                  * this does not wipe data stored outside the
 423                                  * range sg->offset..sg->offset+sg->length.
 424                                  * Therefore, blkback *could* see data from
 425                                  * previous requests. This is OK as long as
 426                                  * persistent grants are shared with just one
 427                                  * domain. It may need refactoring if this
 428                                  * changes
 429                                  */
 430                                 memcpy(shared_data + sg->offset,
 431                                        bvec_data   + sg->offset,
 432                                        sg->length);
 433
 434                                 kunmap_atomic(bvec_data);
 435                                 kunmap_atomic(shared_data);
 436                         }
 437
 438                         info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 439                         ring_req->u.rw.seg[i] =
 440                                         (struct blkif_request_segment) {
 441                                                 .gref       = ref,
 442                                                 .first_sect = fsect,
 443                                                 .last_sect  = lsect };
 444                 }
 445         }
 446
 447         info->ring.req_prod_pvt++;
 448
 449         /* Keep a private copy so we can reissue requests when recovering. */
 450         info->shadow[id].req = *ring_req;
 451
 452         if (new_persistent_gnts)
 453                 gnttab_free_grant_references(gref_head);
 454
 455         return 0;
 456 }
 457
 458
 459 static inline void flush_requests(struct blkfront_info *info)
 460 {
 461         int notify;
 462
 463         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 464
 465         if (notify)
 466                 notify_remote_via_irq(info->irq);
 467 }
 468
 469 /*
 470  * do_blkif_request
 471  *  read a block; request is in a request queue
 472  */
 473 static void do_blkif_request(struct request_queue *rq)
 474 {
 475         struct blkfront_info *info = NULL;
 476         struct request *req;
 477         int queued;
 478
 479         pr_debug("Entered do_blkif_request\n");
 480
 481         queued = 0;
 482
 483         while ((req = blk_peek_request(rq)) != NULL) {
 484                 info = req->rq_disk->private_data;
 485
 486                 if (RING_FULL(&info->ring))
 487                         goto wait;
 488
 489                 blk_start_request(req);
 490
 491                 if ((req->cmd_type != REQ_TYPE_FS) ||
 492                     ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
 493                     !info->flush_op)) {
 494                         __blk_end_request_all(req, -EIO);
 495                         continue;
 496                 }
 497
 498                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 499                          "(%u/%u) buffer:%p [%s]\n",
 500                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 501                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 502                          req->buffer, rq_data_dir(req) ? "write" : "read");
 503
 504                 if (blkif_queue_request(req)) {
 505                         blk_requeue_request(rq, req);
 506 wait:
 507                         /* Avoid pointless unplugs. */
 508                         blk_stop_queue(rq);
 509                         break;
 510                 }
 511
 512                 queued++;
 513         }
 514
 515         if (queued != 0)
 516                 flush_requests(info);
 517 }
 518
 519 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 520 {
 521         struct request_queue *rq;
 522         struct blkfront_info *info = gd->private_data;
 523
 524         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 525         if (rq == NULL)
 526                 return -1;
 527
 528         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 529
 530         if (info->feature_discard) {
 531                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 532                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 533                 rq->limits.discard_granularity = info->discard_granularity;
 534                 rq->limits.discard_alignment = info->discard_alignment;
 535                 if (info->feature_secdiscard)
 536                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 537         }
 538
 539         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 540         blk_queue_logical_block_size(rq, sector_size);
 541         blk_queue_max_hw_sectors(rq, 512);
 542
 543         /* Each segment in a request is up to an aligned page in size. */
 544         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 545         blk_queue_max_segment_size(rq, PAGE_SIZE);
 546
 547         /* Ensure a merged request will fit in a single I/O ring slot. */
 548         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 549
 550         /* Make sure buffer addresses are sector-aligned. */
 551         blk_queue_dma_alignment(rq, 511);
 552
 553         /* Make sure we don't use bounce buffers. */
 554         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 555
 556         gd->queue = rq;
 557
 558         return 0;
 559 }
 560
 561
 562 static void xlvbd_flush(struct blkfront_info *info)
 563 {
 564         blk_queue_flush(info->rq, info->feature_flush);
 565         printk(KERN_INFO "blkfront: %s: %s: %s %s\n",
 566                info->gd->disk_name,
 567                info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 568                 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 569                 "flush diskcache" : "barrier or flush"),
 570                info->feature_flush ? "enabled" : "disabled",
 571                info->feature_persistent ? "using persistent grants" : "");
 572 }
 573
 574 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 575 {
 576         int major;
 577         major = BLKIF_MAJOR(vdevice);
 578         *minor = BLKIF_MINOR(vdevice);
 579         switch (major) {
 580                 case XEN_IDE0_MAJOR:
 581                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 582                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 583                                 EMULATED_HD_DISK_MINOR_OFFSET;
 584                         break;
 585                 case XEN_IDE1_MAJOR:
 586                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 587                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 588                                 EMULATED_HD_DISK_MINOR_OFFSET;
 589                         break;
 590                 case XEN_SCSI_DISK0_MAJOR:
 591                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 592                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 593                         break;
 594                 case XEN_SCSI_DISK1_MAJOR:
 595                 case XEN_SCSI_DISK2_MAJOR:
 596                 case XEN_SCSI_DISK3_MAJOR:
 597                 case XEN_SCSI_DISK4_MAJOR:
 598                 case XEN_SCSI_DISK5_MAJOR:
 599                 case XEN_SCSI_DISK6_MAJOR:
 600                 case XEN_SCSI_DISK7_MAJOR:
 601                         *offset = (*minor / PARTS_PER_DISK) +
 602                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 603                                 EMULATED_SD_DISK_NAME_OFFSET;
 604                         *minor = *minor +
 605                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 606                                 EMULATED_SD_DISK_MINOR_OFFSET;
 607                         break;
 608                 case XEN_SCSI_DISK8_MAJOR:
 609                 case XEN_SCSI_DISK9_MAJOR:
 610                 case XEN_SCSI_DISK10_MAJOR:
 611                 case XEN_SCSI_DISK11_MAJOR:
 612                 case XEN_SCSI_DISK12_MAJOR:
 613                 case XEN_SCSI_DISK13_MAJOR:
 614                 case XEN_SCSI_DISK14_MAJOR:
 615                 case XEN_SCSI_DISK15_MAJOR:
 616                         *offset = (*minor / PARTS_PER_DISK) +
 617                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 618                                 EMULATED_SD_DISK_NAME_OFFSET;
 619                         *minor = *minor +
 620                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 621                                 EMULATED_SD_DISK_MINOR_OFFSET;
 622                         break;
 623                 case XENVBD_MAJOR:
 624                         *offset = *minor / PARTS_PER_DISK;
 625                         break;
 626                 default:
 627                         printk(KERN_WARNING "blkfront: your disk configuration is "
 628                                         "incorrect, please use an xvd device instead\n");
 629                         return -ENODEV;
 630         }
 631         return 0;
 632 }
 633
 634 static char *encode_disk_name(char *ptr, unsigned int n)
 635 {
 636         if (n >= 26)
 637                 ptr = encode_disk_name(ptr, n / 26 - 1);
 638         *ptr = 'a' + n % 26;
 639         return ptr + 1;
 640 }
 641
 642 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 643                                struct blkfront_info *info,
 644                                u16 vdisk_info, u16 sector_size)
 645 {
 646         struct gendisk *gd;
 647         int nr_minors = 1;
 648         int err;
 649         unsigned int offset;
 650         int minor;
 651         int nr_parts;
 652         char *ptr;
 653
 654         BUG_ON(info->gd != NULL);
 655         BUG_ON(info->rq != NULL);
 656
 657         if ((info->vdevice>>EXT_SHIFT) > 1) {
 658                 /* this is above the extended range; something is wrong */
 659                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 660                 return -ENODEV;
 661         }
 662
 663         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 664                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 665                 if (err)
 666                         return err;
 667                 nr_parts = PARTS_PER_DISK;
 668         } else {
 669                 minor = BLKIF_MINOR_EXT(info->vdevice);
 670                 nr_parts = PARTS_PER_EXT_DISK;
 671                 offset = minor / nr_parts;
 672                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 673                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 674                                         "emulated IDE disks,\n\t choose an xvd device name"
 675                                         "from xvde on\n", info->vdevice);
 676         }
 677         if (minor >> MINORBITS) {
 678                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 679                         info->vdevice, minor);
 680                 return -ENODEV;
 681         }
 682
 683         if ((minor % nr_parts) == 0)
 684                 nr_minors = nr_parts;
 685
 686         err = xlbd_reserve_minors(minor, nr_minors);
 687         if (err)
 688                 goto out;
 689         err = -ENODEV;
 690
 691         gd = alloc_disk(nr_minors);
 692         if (gd == NULL)
 693                 goto release;
 694
 695         strcpy(gd->disk_name, DEV_NAME);
 696         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 697         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 698         if (nr_minors > 1)
 699                 *ptr = 0;
 700         else
 701                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 702                          "%d", minor & (nr_parts - 1));
 703
 704         gd->major = XENVBD_MAJOR;
 705         gd->first_minor = minor;
 706         gd->fops = &xlvbd_block_fops;
 707         gd->private_data = info;
 708         gd->driverfs_dev = &(info->xbdev->dev);
 709         set_capacity(gd, capacity);
 710
 711         if (xlvbd_init_blk_queue(gd, sector_size)) {
 712                 del_gendisk(gd);
 713                 goto release;
 714         }
 715
 716         info->rq = gd->queue;
 717         info->gd = gd;
 718
 719         xlvbd_flush(info);
 720
 721         if (vdisk_info & VDISK_READONLY)
 722                 set_disk_ro(gd, 1);
 723
 724         if (vdisk_info & VDISK_REMOVABLE)
 725                 gd->flags |= GENHD_FL_REMOVABLE;
 726
 727         if (vdisk_info & VDISK_CDROM)
 728                 gd->flags |= GENHD_FL_CD;
 729
 730         return 0;
 731
 732  release:
 733         xlbd_release_minors(minor, nr_minors);
 734  out:
 735         return err;
 736 }
 737
 738 static void xlvbd_release_gendisk(struct blkfront_info *info)
 739 {
 740         unsigned int minor, nr_minors;
 741         unsigned long flags;
 742
 743         if (info->rq == NULL)
 744                 return;
 745
 746         spin_lock_irqsave(&info->io_lock, flags);
 747
 748         /* No more blkif_request(). */
 749         blk_stop_queue(info->rq);
 750
 751         /* No more gnttab callback work. */
 752         gnttab_cancel_free_callback(&info->callback);
 753         spin_unlock_irqrestore(&info->io_lock, flags);
 754
 755         /* Flush gnttab callback work. Must be done with no locks held. */
 756         flush_work(&info->work);
 757
 758         del_gendisk(info->gd);
 759
 760         minor = info->gd->first_minor;
 761         nr_minors = info->gd->minors;
 762         xlbd_release_minors(minor, nr_minors);
 763
 764         blk_cleanup_queue(info->rq);
 765         info->rq = NULL;
 766
 767         put_disk(info->gd);
 768         info->gd = NULL;
 769 }
 770
 771 static void kick_pending_request_queues(struct blkfront_info *info)
 772 {
 773         if (!RING_FULL(&info->ring)) {
 774                 /* Re-enable calldowns. */
 775                 blk_start_queue(info->rq);
 776                 /* Kick things off immediately. */
 777                 do_blkif_request(info->rq);
 778         }
 779 }
 780
 781 static void blkif_restart_queue(struct work_struct *work)
 782 {
 783         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 784
 785         spin_lock_irq(&info->io_lock);
 786         if (info->connected == BLKIF_STATE_CONNECTED)
 787                 kick_pending_request_queues(info);
 788         spin_unlock_irq(&info->io_lock);
 789 }
 790
 791 static void blkif_free(struct blkfront_info *info, int suspend)
 792 {
 793         struct llist_node *all_gnts;
 794         struct grant *persistent_gnt;
 795
 796         /* Prevent new requests being issued until we fix things up. */
 797         spin_lock_irq(&info->io_lock);
 798         info->connected = suspend ?
 799                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 800         /* No more blkif_request(). */
 801         if (info->rq)
 802                 blk_stop_queue(info->rq);
 803
 804         /* Remove all persistent grants */
 805         if (info->persistent_gnts_c) {
 806                 all_gnts = llist_del_all(&info->persistent_gnts);
 807                 llist_for_each_entry(persistent_gnt, all_gnts, node) {
 808                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
 809                         __free_page(pfn_to_page(persistent_gnt->pfn));
 810                         kfree(persistent_gnt);
 811                 }
 812                 info->persistent_gnts_c = 0;
 813         }
 814
 815         /* No more gnttab callback work. */
 816         gnttab_cancel_free_callback(&info->callback);
 817         spin_unlock_irq(&info->io_lock);
 818
 819         /* Flush gnttab callback work. Must be done with no locks held. */
 820         flush_work(&info->work);
 821
 822         /* Free resources associated with old device channel. */
 823         if (info->ring_ref != GRANT_INVALID_REF) {
 824                 gnttab_end_foreign_access(info->ring_ref, 0,
 825                                           (unsigned long)info->ring.sring);
 826                 info->ring_ref = GRANT_INVALID_REF;
 827                 info->ring.sring = NULL;
 828         }
 829         if (info->irq)
 830                 unbind_from_irqhandler(info->irq, info);
 831         info->evtchn = info->irq = 0;
 832
 833 }
 834
 835 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
 836                              struct blkif_response *bret)
 837 {
 838         int i;
 839         struct bio_vec *bvec;
 840         struct req_iterator iter;
 841         unsigned long flags;
 842         char *bvec_data;
 843         void *shared_data;
 844         unsigned int offset = 0;
 845
 846         if (bret->operation == BLKIF_OP_READ) {
 847                 /*
 848                  * Copy the data received from the backend into the bvec.
 849                  * Since bv_offset can be different than 0, and bv_len different
 850                  * than PAGE_SIZE, we have to keep track of the current offset,
 851                  * to be sure we are copying the data from the right shared page.
 852                  */
 853                 rq_for_each_segment(bvec, s->request, iter) {
 854                         BUG_ON((bvec->bv_offset + bvec->bv_len) > PAGE_SIZE);
 855                         i = offset >> PAGE_SHIFT;
 856                         BUG_ON(i >= s->req.u.rw.nr_segments);
 857                         shared_data = kmap_atomic(
 858                                 pfn_to_page(s->grants_used[i]->pfn));
 859                         bvec_data = bvec_kmap_irq(bvec, &flags);
 860                         memcpy(bvec_data, shared_data + bvec->bv_offset,
 861                                 bvec->bv_len);
 862                         bvec_kunmap_irq(bvec_data, &flags);
 863                         kunmap_atomic(shared_data);
 864                         offset += bvec->bv_len;
 865                 }
 866         }
 867         /* Add the persistent grant into the list of free grants */
 868         for (i = 0; i < s->req.u.rw.nr_segments; i++) {
 869                 llist_add(&s->grants_used[i]->node, &info->persistent_gnts);
 870                 info->persistent_gnts_c++;
 871         }
 872 }
 873
 874 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 875 {
 876         struct request *req;
 877         struct blkif_response *bret;
 878         RING_IDX i, rp;
 879         unsigned long flags;
 880         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 881         int error;
 882
 883         spin_lock_irqsave(&info->io_lock, flags);
 884
 885         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 886                 spin_unlock_irqrestore(&info->io_lock, flags);
 887                 return IRQ_HANDLED;
 888         }
 889
 890  again:
 891         rp = info->ring.sring->rsp_prod;
 892         rmb(); /* Ensure we see queued responses up to 'rp'. */
 893
 894         for (i = info->ring.rsp_cons; i != rp; i++) {
 895                 unsigned long id;
 896
 897                 bret = RING_GET_RESPONSE(&info->ring, i);
 898                 id   = bret->id;
 899                 /*
 900                  * The backend has messed up and given us an id that we would
 901                  * never have given to it (we stamp it up to BLK_RING_SIZE -
 902                  * look in get_id_from_freelist.
 903                  */
 904                 if (id >= BLK_RING_SIZE) {
 905                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
 906                              info->gd->disk_name, op_name(bret->operation), id);
 907                         /* We can't safely get the 'struct request' as
 908                          * the id is busted. */
 909                         continue;
 910                 }
 911                 req  = info->shadow[id].request;
 912
 913                 if (bret->operation != BLKIF_OP_DISCARD)
 914                         blkif_completion(&info->shadow[id], info, bret);
 915
 916                 if (add_id_to_freelist(info, id)) {
 917                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
 918                              info->gd->disk_name, op_name(bret->operation), id);
 919                         continue;
 920                 }
 921
 922                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 923                 switch (bret->operation) {
 924                 case BLKIF_OP_DISCARD:
 925                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 926                                 struct request_queue *rq = info->rq;
 927                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
 928                                            info->gd->disk_name, op_name(bret->operation));
 929                                 error = -EOPNOTSUPP;
 930                                 info->feature_discard = 0;
 931                                 info->feature_secdiscard = 0;
 932                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
 933                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
 934                         }
 935                         __blk_end_request_all(req, error);
 936                         break;
 937                 case BLKIF_OP_FLUSH_DISKCACHE:
 938                 case BLKIF_OP_WRITE_BARRIER:
 939                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 940                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
 941                                        info->gd->disk_name, op_name(bret->operation));
 942                                 error = -EOPNOTSUPP;
 943                         }
 944                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
 945                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
 946                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
 947                                        info->gd->disk_name, op_name(bret->operation));
 948                                 error = -EOPNOTSUPP;
 949                         }
 950                         if (unlikely(error)) {
 951                                 if (error == -EOPNOTSUPP)
 952                                         error = 0;
 953                                 info->feature_flush = 0;
 954                                 info->flush_op = 0;
 955                                 xlvbd_flush(info);
 956                         }
 957                         /* fall through */
 958                 case BLKIF_OP_READ:
 959                 case BLKIF_OP_WRITE:
 960                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 961                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 962                                         "request: %x\n", bret->status);
 963
 964                         __blk_end_request_all(req, error);
 965                         break;
 966                 default:
 967                         BUG();
 968                 }
 969         }
 970
 971         info->ring.rsp_cons = i;
 972
 973         if (i != info->ring.req_prod_pvt) {
 974                 int more_to_do;
 975                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 976                 if (more_to_do)
 977                         goto again;
 978         } else
 979                 info->ring.sring->rsp_event = i + 1;
 980
 981         kick_pending_request_queues(info);
 982
 983         spin_unlock_irqrestore(&info->io_lock, flags);
 984
 985         return IRQ_HANDLED;
 986 }
 987
 988
 989 static int setup_blkring(struct xenbus_device *dev,
 990                          struct blkfront_info *info)
 991 {
 992         struct blkif_sring *sring;
 993         int err;
 994
 995         info->ring_ref = GRANT_INVALID_REF;
 996
 997         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 998         if (!sring) {
 999                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1000                 return -ENOMEM;
1001         }
1002         SHARED_RING_INIT(sring);
1003         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1004
1005         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
1006
1007         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1008         if (err < 0) {
1009                 free_page((unsigned long)sring);
1010                 info->ring.sring = NULL;
1011                 goto fail;
1012         }
1013         info->ring_ref = err;
1014
1015         err = xenbus_alloc_evtchn(dev, &info->evtchn);
1016         if (err)
1017                 goto fail;
1018
1019         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1020                                         "blkif", info);
1021         if (err <= 0) {
1022                 xenbus_dev_fatal(dev, err,
1023                                  "bind_evtchn_to_irqhandler failed");
1024                 goto fail;
1025         }
1026         info->irq = err;
1027
1028         return 0;
1029 fail:
1030         blkif_free(info, 0);
1031         return err;
1032 }
1033
1034
1035 /* Common code used when first setting up, and when resuming. */
1036 static int talk_to_blkback(struct xenbus_device *dev,
1037                            struct blkfront_info *info)
1038 {
1039         const char *message = NULL;
1040         struct xenbus_transaction xbt;
1041         int err;
1042
1043         /* Create shared ring, alloc event channel. */
1044         err = setup_blkring(dev, info);
1045         if (err)
1046                 goto out;
1047
1048 again:
1049         err = xenbus_transaction_start(&xbt);
1050         if (err) {
1051                 xenbus_dev_fatal(dev, err, "starting transaction");
1052                 goto destroy_blkring;
1053         }
1054
1055         err = xenbus_printf(xbt, dev->nodename,
1056                             "ring-ref", "%u", info->ring_ref);
1057         if (err) {
1058                 message = "writing ring-ref";
1059                 goto abort_transaction;
1060         }
1061         err = xenbus_printf(xbt, dev->nodename,
1062                             "event-channel", "%u", info->evtchn);
1063         if (err) {
1064                 message = "writing event-channel";
1065                 goto abort_transaction;
1066         }
1067         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1068                             XEN_IO_PROTO_ABI_NATIVE);
1069         if (err) {
1070                 message = "writing protocol";
1071                 goto abort_transaction;
1072         }
1073         err = xenbus_printf(xbt, dev->nodename,
1074                             "feature-persistent", "%u", 1);
1075         if (err)
1076                 dev_warn(&dev->dev,
1077                          "writing persistent grants feature to xenbus");
1078
1079         err = xenbus_transaction_end(xbt, 0);
1080         if (err) {
1081                 if (err == -EAGAIN)
1082                         goto again;
1083                 xenbus_dev_fatal(dev, err, "completing transaction");
1084                 goto destroy_blkring;
1085         }
1086
1087         xenbus_switch_state(dev, XenbusStateInitialised);
1088
1089         return 0;
1090
1091  abort_transaction:
1092         xenbus_transaction_end(xbt, 1);
1093         if (message)
1094                 xenbus_dev_fatal(dev, err, "%s", message);
1095  destroy_blkring:
1096         blkif_free(info, 0);
1097  out:
1098         return err;
1099 }
1100
1101 /**
1102  * Entry point to this code when a new device is created.  Allocate the basic
1103  * structures and the ring buffer for communication with the backend, and
1104  * inform the backend of the appropriate details for those.  Switch to
1105  * Initialised state.
1106  */
1107 static int blkfront_probe(struct xenbus_device *dev,
1108                           const struct xenbus_device_id *id)
1109 {
1110         int err, vdevice, i;
1111         struct blkfront_info *info;
1112
1113         /* FIXME: Use dynamic device id if this is not set. */
1114         err = xenbus_scanf(XBT_NIL, dev->nodename,
1115                            "virtual-device", "%i", &vdevice);
1116         if (err != 1) {
1117                 /* go looking in the extended area instead */
1118                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1119                                    "%i", &vdevice);
1120                 if (err != 1) {
1121                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1122                         return err;
1123                 }
1124         }
1125
1126         if (xen_hvm_domain()) {
1127                 char *type;
1128                 int len;
1129                 /* no unplug has been done: do not hook devices != xen vbds */
1130                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
1131                         int major;
1132
1133                         if (!VDEV_IS_EXTENDED(vdevice))
1134                                 major = BLKIF_MAJOR(vdevice);
1135                         else
1136                                 major = XENVBD_MAJOR;
1137
1138                         if (major != XENVBD_MAJOR) {
1139                                 printk(KERN_INFO
1140                                                 "%s: HVM does not support vbd %d as xen block device\n",
1141                                                 __FUNCTION__, vdevice);
1142                                 return -ENODEV;
1143                         }
1144                 }
1145                 /* do not create a PV cdrom device if we are an HVM guest */
1146                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1147                 if (IS_ERR(type))
1148                         return -ENODEV;
1149                 if (strncmp(type, "cdrom", 5) == 0) {
1150                         kfree(type);
1151                         return -ENODEV;
1152                 }
1153                 kfree(type);
1154         }
1155         info = kzalloc(sizeof(*info), GFP_KERNEL);
1156         if (!info) {
1157                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1158                 return -ENOMEM;
1159         }
1160
1161         mutex_init(&info->mutex);
1162         spin_lock_init(&info->io_lock);
1163         info->xbdev = dev;
1164         info->vdevice = vdevice;
1165         init_llist_head(&info->persistent_gnts);
1166         info->persistent_gnts_c = 0;
1167         info->connected = BLKIF_STATE_DISCONNECTED;
1168         INIT_WORK(&info->work, blkif_restart_queue);
1169
1170         for (i = 0; i < BLK_RING_SIZE; i++)
1171                 info->shadow[i].req.u.rw.id = i+1;
1172         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1173
1174         /* Front end dir is a number, which is used as the id. */
1175         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1176         dev_set_drvdata(&dev->dev, info);
1177
1178         err = talk_to_blkback(dev, info);
1179         if (err) {
1180                 kfree(info);
1181                 dev_set_drvdata(&dev->dev, NULL);
1182                 return err;
1183         }
1184
1185         return 0;
1186 }
1187
1188
1189 static int blkif_recover(struct blkfront_info *info)
1190 {
1191         int i;
1192         struct blkif_request *req;
1193         struct blk_shadow *copy;
1194         int j;
1195
1196         /* Stage 1: Make a safe copy of the shadow state. */
1197         copy = kmalloc(sizeof(info->shadow),
1198                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1199         if (!copy)
1200                 return -ENOMEM;
1201         memcpy(copy, info->shadow, sizeof(info->shadow));
1202
1203         /* Stage 2: Set up free list. */
1204         memset(&info->shadow, 0, sizeof(info->shadow));
1205         for (i = 0; i < BLK_RING_SIZE; i++)
1206                 info->shadow[i].req.u.rw.id = i+1;
1207         info->shadow_free = info->ring.req_prod_pvt;
1208         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1209
1210         /* Stage 3: Find pending requests and requeue them. */
1211         for (i = 0; i < BLK_RING_SIZE; i++) {
1212                 /* Not in use? */
1213                 if (!copy[i].request)
1214                         continue;
1215
1216                 /* Grab a request slot and copy shadow state into it. */
1217                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1218                 *req = copy[i].req;
1219
1220                 /* We get a new request id, and must reset the shadow state. */
1221                 req->u.rw.id = get_id_from_freelist(info);
1222                 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1223
1224                 if (req->operation != BLKIF_OP_DISCARD) {
1225                 /* Rewrite any grant references invalidated by susp/resume. */
1226                         for (j = 0; j < req->u.rw.nr_segments; j++)
1227                                 gnttab_grant_foreign_access_ref(
1228                                         req->u.rw.seg[j].gref,
1229                                         info->xbdev->otherend_id,
1230                                         pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1231                                         0);
1232                 }
1233                 info->shadow[req->u.rw.id].req = *req;
1234
1235                 info->ring.req_prod_pvt++;
1236         }
1237
1238         kfree(copy);
1239
1240         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1241
1242         spin_lock_irq(&info->io_lock);
1243
1244         /* Now safe for us to use the shared ring */
1245         info->connected = BLKIF_STATE_CONNECTED;
1246
1247         /* Send off requeued requests */
1248         flush_requests(info);
1249
1250         /* Kick any other new requests queued since we resumed */
1251         kick_pending_request_queues(info);
1252
1253         spin_unlock_irq(&info->io_lock);
1254
1255         return 0;
1256 }
1257
1258 /**
1259  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1260  * driver restart.  We tear down our blkif structure and recreate it, but
1261  * leave the device-layer structures intact so that this is transparent to the
1262  * rest of the kernel.
1263  */
1264 static int blkfront_resume(struct xenbus_device *dev)
1265 {
1266         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1267         int err;
1268
1269         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1270
1271         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1272
1273         err = talk_to_blkback(dev, info);
1274         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1275                 err = blkif_recover(info);
1276
1277         return err;
1278 }
1279
1280 static void
1281 blkfront_closing(struct blkfront_info *info)
1282 {
1283         struct xenbus_device *xbdev = info->xbdev;
1284         struct block_device *bdev = NULL;
1285
1286         mutex_lock(&info->mutex);
1287
1288         if (xbdev->state == XenbusStateClosing) {
1289                 mutex_unlock(&info->mutex);
1290                 return;
1291         }
1292
1293         if (info->gd)
1294                 bdev = bdget_disk(info->gd, 0);
1295
1296         mutex_unlock(&info->mutex);
1297
1298         if (!bdev) {
1299                 xenbus_frontend_closed(xbdev);
1300                 return;
1301         }
1302
1303         mutex_lock(&bdev->bd_mutex);
1304
1305         if (bdev->bd_openers) {
1306                 xenbus_dev_error(xbdev, -EBUSY,
1307                                  "Device in use; refusing to close");
1308                 xenbus_switch_state(xbdev, XenbusStateClosing);
1309         } else {
1310                 xlvbd_release_gendisk(info);
1311                 xenbus_frontend_closed(xbdev);
1312         }
1313
1314         mutex_unlock(&bdev->bd_mutex);
1315         bdput(bdev);
1316 }
1317
1318 static void blkfront_setup_discard(struct blkfront_info *info)
1319 {
1320         int err;
1321         char *type;
1322         unsigned int discard_granularity;
1323         unsigned int discard_alignment;
1324         unsigned int discard_secure;
1325
1326         type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1327         if (IS_ERR(type))
1328                 return;
1329
1330         info->feature_secdiscard = 0;
1331         if (strncmp(type, "phy", 3) == 0) {
1332                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1333                         "discard-granularity", "%u", &discard_granularity,
1334                         "discard-alignment", "%u", &discard_alignment,
1335                         NULL);
1336                 if (!err) {
1337                         info->feature_discard = 1;
1338                         info->discard_granularity = discard_granularity;
1339                         info->discard_alignment = discard_alignment;
1340                 }
1341                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1342                             "discard-secure", "%d", &discard_secure,
1343                             NULL);
1344                 if (!err)
1345                         info->feature_secdiscard = discard_secure;
1346
1347         } else if (strncmp(type, "file", 4) == 0)
1348                 info->feature_discard = 1;
1349
1350         kfree(type);
1351 }
1352
1353 /*
1354  * Invoked when the backend is finally 'ready' (and has told produced
1355  * the details about the physical device - #sectors, size, etc).
1356  */
1357 static void blkfront_connect(struct blkfront_info *info)
1358 {
1359         unsigned long long sectors;
1360         unsigned long sector_size;
1361         unsigned int binfo;
1362         int err;
1363         int barrier, flush, discard, persistent;
1364
1365         switch (info->connected) {
1366         case BLKIF_STATE_CONNECTED:
1367                 /*
1368                  * Potentially, the back-end may be signalling
1369                  * a capacity change; update the capacity.
1370                  */
1371                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1372                                    "sectors", "%Lu", &sectors);
1373                 if (XENBUS_EXIST_ERR(err))
1374                         return;
1375                 printk(KERN_INFO "Setting capacity to %Lu\n",
1376                        sectors);
1377                 set_capacity(info->gd, sectors);
1378                 revalidate_disk(info->gd);
1379
1380                 /* fall through */
1381         case BLKIF_STATE_SUSPENDED:
1382                 return;
1383
1384         default:
1385                 break;
1386         }
1387
1388         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1389                 __func__, info->xbdev->otherend);
1390
1391         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1392                             "sectors", "%llu", &sectors,
1393                             "info", "%u", &binfo,
1394                             "sector-size", "%lu", &sector_size,
1395                             NULL);
1396         if (err) {
1397                 xenbus_dev_fatal(info->xbdev, err,
1398                                  "reading backend fields at %s",
1399                                  info->xbdev->otherend);
1400                 return;
1401         }
1402
1403         info->feature_flush = 0;
1404         info->flush_op = 0;
1405
1406         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1407                             "feature-barrier", "%d", &barrier,
1408                             NULL);
1409
1410         /*
1411          * If there's no "feature-barrier" defined, then it means
1412          * we're dealing with a very old backend which writes
1413          * synchronously; nothing to do.
1414          *
1415          * If there are barriers, then we use flush.
1416          */
1417         if (!err && barrier) {
1418                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1419                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1420         }
1421         /*
1422          * And if there is "feature-flush-cache" use that above
1423          * barriers.
1424          */
1425         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1426                             "feature-flush-cache", "%d", &flush,
1427                             NULL);
1428
1429         if (!err && flush) {
1430                 info->feature_flush = REQ_FLUSH;
1431                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1432         }
1433
1434         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1435                             "feature-discard", "%d", &discard,
1436                             NULL);
1437
1438         if (!err && discard)
1439                 blkfront_setup_discard(info);
1440
1441         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1442                             "feature-persistent", "%u", &persistent,
1443                             NULL);
1444         if (err)
1445                 info->feature_persistent = 0;
1446         else
1447                 info->feature_persistent = persistent;
1448
1449         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1450         if (err) {
1451                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1452                                  info->xbdev->otherend);
1453                 return;
1454         }
1455
1456         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1457
1458         /* Kick pending requests. */
1459         spin_lock_irq(&info->io_lock);
1460         info->connected = BLKIF_STATE_CONNECTED;
1461         kick_pending_request_queues(info);
1462         spin_unlock_irq(&info->io_lock);
1463
1464         add_disk(info->gd);
1465
1466         info->is_ready = 1;
1467 }
1468
1469 /**
1470  * Callback received when the backend's state changes.
1471  */
1472 static void blkback_changed(struct xenbus_device *dev,
1473                             enum xenbus_state backend_state)
1474 {
1475         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1476
1477         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1478
1479         switch (backend_state) {
1480         case XenbusStateInitialising:
1481         case XenbusStateInitWait:
1482         case XenbusStateInitialised:
1483         case XenbusStateReconfiguring:
1484         case XenbusStateReconfigured:
1485         case XenbusStateUnknown:
1486         case XenbusStateClosed:
1487                 break;
1488
1489         case XenbusStateConnected:
1490                 blkfront_connect(info);
1491                 break;
1492
1493         case XenbusStateClosing:
1494                 blkfront_closing(info);
1495                 break;
1496         }
1497 }
1498
1499 static int blkfront_remove(struct xenbus_device *xbdev)
1500 {
1501         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1502         struct block_device *bdev = NULL;
1503         struct gendisk *disk;
1504
1505         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1506
1507         blkif_free(info, 0);
1508
1509         mutex_lock(&info->mutex);
1510
1511         disk = info->gd;
1512         if (disk)
1513                 bdev = bdget_disk(disk, 0);
1514
1515         info->xbdev = NULL;
1516         mutex_unlock(&info->mutex);
1517
1518         if (!bdev) {
1519                 kfree(info);
1520                 return 0;
1521         }
1522
1523         /*
1524          * The xbdev was removed before we reached the Closed
1525          * state. See if it's safe to remove the disk. If the bdev
1526          * isn't closed yet, we let release take care of it.
1527          */
1528
1529         mutex_lock(&bdev->bd_mutex);
1530         info = disk->private_data;
1531
1532         dev_warn(disk_to_dev(disk),
1533                  "%s was hot-unplugged, %d stale handles\n",
1534                  xbdev->nodename, bdev->bd_openers);
1535
1536         if (info && !bdev->bd_openers) {
1537                 xlvbd_release_gendisk(info);
1538                 disk->private_data = NULL;
1539                 kfree(info);
1540         }
1541
1542         mutex_unlock(&bdev->bd_mutex);
1543         bdput(bdev);
1544
1545         return 0;
1546 }
1547
1548 static int blkfront_is_ready(struct xenbus_device *dev)
1549 {
1550         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1551
1552         return info->is_ready && info->xbdev;
1553 }
1554
1555 static int blkif_open(struct block_device *bdev, fmode_t mode)
1556 {
1557         struct gendisk *disk = bdev->bd_disk;
1558         struct blkfront_info *info;
1559         int err = 0;
1560
1561         mutex_lock(&blkfront_mutex);
1562
1563         info = disk->private_data;
1564         if (!info) {
1565                 /* xbdev gone */
1566                 err = -ERESTARTSYS;
1567                 goto out;
1568         }
1569
1570         mutex_lock(&info->mutex);
1571
1572         if (!info->gd)
1573                 /* xbdev is closed */
1574                 err = -ERESTARTSYS;
1575
1576         mutex_unlock(&info->mutex);
1577
1578 out:
1579         mutex_unlock(&blkfront_mutex);
1580         return err;
1581 }
1582
1583 static int blkif_release(struct gendisk *disk, fmode_t mode)
1584 {
1585         struct blkfront_info *info = disk->private_data;
1586         struct block_device *bdev;
1587         struct xenbus_device *xbdev;
1588
1589         mutex_lock(&blkfront_mutex);
1590
1591         bdev = bdget_disk(disk, 0);
1592
1593         if (bdev->bd_openers)
1594                 goto out;
1595
1596         /*
1597          * Check if we have been instructed to close. We will have
1598          * deferred this request, because the bdev was still open.
1599          */
1600
1601         mutex_lock(&info->mutex);
1602         xbdev = info->xbdev;
1603
1604         if (xbdev && xbdev->state == XenbusStateClosing) {
1605                 /* pending switch to state closed */
1606                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1607                 xlvbd_release_gendisk(info);
1608                 xenbus_frontend_closed(info->xbdev);
1609         }
1610
1611         mutex_unlock(&info->mutex);
1612
1613         if (!xbdev) {
1614                 /* sudden device removal */
1615                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1616                 xlvbd_release_gendisk(info);
1617                 disk->private_data = NULL;
1618                 kfree(info);
1619         }
1620
1621 out:
1622         bdput(bdev);
1623         mutex_unlock(&blkfront_mutex);
1624         return 0;
1625 }
1626
1627 static const struct block_device_operations xlvbd_block_fops =
1628 {
1629         .owner = THIS_MODULE,
1630         .open = blkif_open,
1631         .release = blkif_release,
1632         .getgeo = blkif_getgeo,
1633         .ioctl = blkif_ioctl,
1634 };
1635
1636
1637 static const struct xenbus_device_id blkfront_ids[] = {
1638         { "vbd" },
1639         { "" }
1640 };
1641
1642 static DEFINE_XENBUS_DRIVER(blkfront, ,
1643         .probe = blkfront_probe,
1644         .remove = blkfront_remove,
1645         .resume = blkfront_resume,
1646         .otherend_changed = blkback_changed,
1647         .is_ready = blkfront_is_ready,
1648 );
1649
1650 static int __init xlblk_init(void)
1651 {
1652         int ret;
1653
1654         if (!xen_domain())
1655                 return -ENODEV;
1656
1657         if (xen_hvm_domain() && !xen_platform_pci_unplug)
1658                 return -ENODEV;
1659
1660         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1661                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1662                        XENVBD_MAJOR, DEV_NAME);
1663                 return -ENODEV;
1664         }
1665
1666         ret = xenbus_register_frontend(&blkfront_driver);
1667         if (ret) {
1668                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1669                 return ret;
1670         }
1671
1672         return 0;
1673 }
1674 module_init(xlblk_init);
1675
1676
1677 static void __exit xlblk_exit(void)
1678 {
1679         xenbus_unregister_driver(&blkfront_driver);
1680         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1681         kfree(minors);
1682 }
1683 module_exit(xlblk_exit);
1684
1685 MODULE_DESCRIPTION("Xen virtual block device frontend");
1686 MODULE_LICENSE("GPL");
1687 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1688 MODULE_ALIAS("xen:vbd");
1689 MODULE_ALIAS("xenblk");