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/blk-mq.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/cdrom.h>
  43 #include <linux/module.h>
  44 #include <linux/slab.h>
  45 #include <linux/mutex.h>
  46 #include <linux/scatterlist.h>
  47 #include <linux/bitmap.h>
  48 #include <linux/list.h>
  49
  50 #include <xen/xen.h>
  51 #include <xen/xenbus.h>
  52 #include <xen/grant_table.h>
  53 #include <xen/events.h>
  54 #include <xen/page.h>
  55 #include <xen/platform_pci.h>
  56
  57 #include <xen/interface/grant_table.h>
  58 #include <xen/interface/io/blkif.h>
  59 #include <xen/interface/io/protocols.h>
  60
  61 #include <asm/xen/hypervisor.h>
  62
  63 enum blkif_state {
  64         BLKIF_STATE_DISCONNECTED,
  65         BLKIF_STATE_CONNECTED,
  66         BLKIF_STATE_SUSPENDED,
  67 };
  68
  69 struct grant {
  70         grant_ref_t gref;
  71         unsigned long pfn;
  72         struct list_head node;
  73 };
  74
  75 struct blk_shadow {
  76         struct blkif_request req;
  77         struct request *request;
  78         struct grant **grants_used;
  79         struct grant **indirect_grants;
  80         struct scatterlist *sg;
  81 };
  82
  83 struct split_bio {
  84         struct bio *bio;
  85         atomic_t pending;
  86 };
  87
  88 static DEFINE_MUTEX(blkfront_mutex);
  89 static const struct block_device_operations xlvbd_block_fops;
  90
  91 /*
  92  * Maximum number of segments in indirect requests, the actual value used by
  93  * the frontend driver is the minimum of this value and the value provided
  94  * by the backend driver.
  95  */
  96
  97 static unsigned int xen_blkif_max_segments = 32;
  98 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
  99 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
 100
 101 /*
 102  * Maximum order of pages to be used for the shared ring between front and
 103  * backend, 4KB page granularity is used.
 104  */
 105 static unsigned int xen_blkif_max_ring_order;
 106 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
 107 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 108
 109 #define BLK_RING_SIZE(info) __CONST_RING_SIZE(blkif, PAGE_SIZE * (info)->nr_ring_pages)
 110 #define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * XENBUS_MAX_RING_PAGES)
 111 /*
 112  * ring-ref%i i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 113  * characters are enough. Define to 20 to keep consist with backend.
 114  */
 115 #define RINGREF_NAME_LEN (20)
 116
 117 /*
 118  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 119  * hang in private_data off the gendisk structure. We may end up
 120  * putting all kinds of interesting stuff here :-)
 121  */
 122 struct blkfront_info
 123 {
 124         spinlock_t io_lock;
 125         struct mutex mutex;
 126         struct xenbus_device *xbdev;
 127         struct gendisk *gd;
 128         int vdevice;
 129         blkif_vdev_t handle;
 130         enum blkif_state connected;
 131         int ring_ref[XENBUS_MAX_RING_PAGES];
 132         unsigned int nr_ring_pages;
 133         struct blkif_front_ring ring;
 134         unsigned int evtchn, irq;
 135         struct request_queue *rq;
 136         struct work_struct work;
 137         struct gnttab_free_callback callback;
 138         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 139         struct list_head grants;
 140         struct list_head indirect_pages;
 141         unsigned int persistent_gnts_c;
 142         unsigned long shadow_free;
 143         unsigned int feature_flush;
 144         unsigned int feature_discard:1;
 145         unsigned int feature_secdiscard:1;
 146         unsigned int discard_granularity;
 147         unsigned int discard_alignment;
 148         unsigned int feature_persistent:1;
 149         unsigned int max_indirect_segments;
 150         int is_ready;
 151         struct blk_mq_tag_set tag_set;
 152 };
 153
 154 static unsigned int nr_minors;
 155 static unsigned long *minors;
 156 static DEFINE_SPINLOCK(minor_lock);
 157
 158 #define GRANT_INVALID_REF       0
 159
 160 #define PARTS_PER_DISK          16
 161 #define PARTS_PER_EXT_DISK      256
 162
 163 #define BLKIF_MAJOR(dev) ((dev)>>8)
 164 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 165
 166 #define EXT_SHIFT 28
 167 #define EXTENDED (1<<EXT_SHIFT)
 168 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 169 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 170 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 171 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 172 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 173 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 174
 175 #define DEV_NAME        "xvd"   /* name in /dev */
 176
 177 #define SEGS_PER_INDIRECT_FRAME \
 178         (PAGE_SIZE/sizeof(struct blkif_request_segment))
 179 #define INDIRECT_GREFS(_segs) \
 180         ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
 181
 182 static int blkfront_setup_indirect(struct blkfront_info *info);
 183 static int blkfront_gather_backend_features(struct blkfront_info *info);
 184
 185 static int get_id_from_freelist(struct blkfront_info *info)
 186 {
 187         unsigned long free = info->shadow_free;
 188         BUG_ON(free >= BLK_RING_SIZE(info));
 189         info->shadow_free = info->shadow[free].req.u.rw.id;
 190         info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 191         return free;
 192 }
 193
 194 static int add_id_to_freelist(struct blkfront_info *info,
 195                                unsigned long id)
 196 {
 197         if (info->shadow[id].req.u.rw.id != id)
 198                 return -EINVAL;
 199         if (info->shadow[id].request == NULL)
 200                 return -EINVAL;
 201         info->shadow[id].req.u.rw.id  = info->shadow_free;
 202         info->shadow[id].request = NULL;
 203         info->shadow_free = id;
 204         return 0;
 205 }
 206
 207 static int fill_grant_buffer(struct blkfront_info *info, int num)
 208 {
 209         struct page *granted_page;
 210         struct grant *gnt_list_entry, *n;
 211         int i = 0;
 212
 213         while(i < num) {
 214                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 215                 if (!gnt_list_entry)
 216                         goto out_of_memory;
 217
 218                 if (info->feature_persistent) {
 219                         granted_page = alloc_page(GFP_NOIO);
 220                         if (!granted_page) {
 221                                 kfree(gnt_list_entry);
 222                                 goto out_of_memory;
 223                         }
 224                         gnt_list_entry->pfn = page_to_pfn(granted_page);
 225                 }
 226
 227                 gnt_list_entry->gref = GRANT_INVALID_REF;
 228                 list_add(&gnt_list_entry->node, &info->grants);
 229                 i++;
 230         }
 231
 232         return 0;
 233
 234 out_of_memory:
 235         list_for_each_entry_safe(gnt_list_entry, n,
 236                                  &info->grants, node) {
 237                 list_del(&gnt_list_entry->node);
 238                 if (info->feature_persistent)
 239                         __free_page(pfn_to_page(gnt_list_entry->pfn));
 240                 kfree(gnt_list_entry);
 241                 i--;
 242         }
 243         BUG_ON(i != 0);
 244         return -ENOMEM;
 245 }
 246
 247 static struct grant *get_grant(grant_ref_t *gref_head,
 248                                unsigned long pfn,
 249                                struct blkfront_info *info)
 250 {
 251         struct grant *gnt_list_entry;
 252         unsigned long buffer_gfn;
 253
 254         BUG_ON(list_empty(&info->grants));
 255         gnt_list_entry = list_first_entry(&info->grants, struct grant,
 256                                           node);
 257         list_del(&gnt_list_entry->node);
 258
 259         if (gnt_list_entry->gref != GRANT_INVALID_REF) {
 260                 info->persistent_gnts_c--;
 261                 return gnt_list_entry;
 262         }
 263
 264         /* Assign a gref to this page */
 265         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 266         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 267         if (!info->feature_persistent) {
 268                 BUG_ON(!pfn);
 269                 gnt_list_entry->pfn = pfn;
 270         }
 271         buffer_gfn = pfn_to_gfn(gnt_list_entry->pfn);
 272         gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 273                                         info->xbdev->otherend_id,
 274                                         buffer_gfn, 0);
 275         return gnt_list_entry;
 276 }
 277
 278 static const char *op_name(int op)
 279 {
 280         static const char *const names[] = {
 281                 [BLKIF_OP_READ] = "read",
 282                 [BLKIF_OP_WRITE] = "write",
 283                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 284                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 285                 [BLKIF_OP_DISCARD] = "discard" };
 286
 287         if (op < 0 || op >= ARRAY_SIZE(names))
 288                 return "unknown";
 289
 290         if (!names[op])
 291                 return "reserved";
 292
 293         return names[op];
 294 }
 295 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 296 {
 297         unsigned int end = minor + nr;
 298         int rc;
 299
 300         if (end > nr_minors) {
 301                 unsigned long *bitmap, *old;
 302
 303                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 304                                  GFP_KERNEL);
 305                 if (bitmap == NULL)
 306                         return -ENOMEM;
 307
 308                 spin_lock(&minor_lock);
 309                 if (end > nr_minors) {
 310                         old = minors;
 311                         memcpy(bitmap, minors,
 312                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 313                         minors = bitmap;
 314                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 315                 } else
 316                         old = bitmap;
 317                 spin_unlock(&minor_lock);
 318                 kfree(old);
 319         }
 320
 321         spin_lock(&minor_lock);
 322         if (find_next_bit(minors, end, minor) >= end) {
 323                 bitmap_set(minors, minor, nr);
 324                 rc = 0;
 325         } else
 326                 rc = -EBUSY;
 327         spin_unlock(&minor_lock);
 328
 329         return rc;
 330 }
 331
 332 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 333 {
 334         unsigned int end = minor + nr;
 335
 336         BUG_ON(end > nr_minors);
 337         spin_lock(&minor_lock);
 338         bitmap_clear(minors,  minor, nr);
 339         spin_unlock(&minor_lock);
 340 }
 341
 342 static void blkif_restart_queue_callback(void *arg)
 343 {
 344         struct blkfront_info *info = (struct blkfront_info *)arg;
 345         schedule_work(&info->work);
 346 }
 347
 348 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 349 {
 350         /* We don't have real geometry info, but let's at least return
 351            values consistent with the size of the device */
 352         sector_t nsect = get_capacity(bd->bd_disk);
 353         sector_t cylinders = nsect;
 354
 355         hg->heads = 0xff;
 356         hg->sectors = 0x3f;
 357         sector_div(cylinders, hg->heads * hg->sectors);
 358         hg->cylinders = cylinders;
 359         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 360                 hg->cylinders = 0xffff;
 361         return 0;
 362 }
 363
 364 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 365                        unsigned command, unsigned long argument)
 366 {
 367         struct blkfront_info *info = bdev->bd_disk->private_data;
 368         int i;
 369
 370         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 371                 command, (long)argument);
 372
 373         switch (command) {
 374         case CDROMMULTISESSION:
 375                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 376                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 377                         if (put_user(0, (char __user *)(argument + i)))
 378                                 return -EFAULT;
 379                 return 0;
 380
 381         case CDROM_GET_CAPABILITY: {
 382                 struct gendisk *gd = info->gd;
 383                 if (gd->flags & GENHD_FL_CD)
 384                         return 0;
 385                 return -EINVAL;
 386         }
 387
 388         default:
 389                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 390                   command);*/
 391                 return -EINVAL; /* same return as native Linux */
 392         }
 393
 394         return 0;
 395 }
 396
 397 /*
 398  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 399  * and writes are handled as expected.
 400  *
 401  * @req: a request struct
 402  */
 403 static int blkif_queue_request(struct request *req)
 404 {
 405         struct blkfront_info *info = req->rq_disk->private_data;
 406         struct blkif_request *ring_req;
 407         unsigned long id;
 408         unsigned int fsect, lsect;
 409         int i, ref, n;
 410         struct blkif_request_segment *segments = NULL;
 411
 412         /*
 413          * Used to store if we are able to queue the request by just using
 414          * existing persistent grants, or if we have to get new grants,
 415          * as there are not sufficiently many free.
 416          */
 417         bool new_persistent_gnts;
 418         grant_ref_t gref_head;
 419         struct grant *gnt_list_entry = NULL;
 420         struct scatterlist *sg;
 421         int nseg, max_grefs;
 422
 423         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 424                 return 1;
 425
 426         max_grefs = req->nr_phys_segments;
 427         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 428                 /*
 429                  * If we are using indirect segments we need to account
 430                  * for the indirect grefs used in the request.
 431                  */
 432                 max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
 433
 434         /* Check if we have enough grants to allocate a requests */
 435         if (info->persistent_gnts_c < max_grefs) {
 436                 new_persistent_gnts = 1;
 437                 if (gnttab_alloc_grant_references(
 438                     max_grefs - info->persistent_gnts_c,
 439                     &gref_head) < 0) {
 440                         gnttab_request_free_callback(
 441                                 &info->callback,
 442                                 blkif_restart_queue_callback,
 443                                 info,
 444                                 max_grefs);
 445                         return 1;
 446                 }
 447         } else
 448                 new_persistent_gnts = 0;
 449
 450         /* Fill out a communications ring structure. */
 451         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 452         id = get_id_from_freelist(info);
 453         info->shadow[id].request = req;
 454
 455         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 456                 ring_req->operation = BLKIF_OP_DISCARD;
 457                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 458                 ring_req->u.discard.id = id;
 459                 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 460                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 461                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 462                 else
 463                         ring_req->u.discard.flag = 0;
 464         } else {
 465                 BUG_ON(info->max_indirect_segments == 0 &&
 466                        req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 467                 BUG_ON(info->max_indirect_segments &&
 468                        req->nr_phys_segments > info->max_indirect_segments);
 469                 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
 470                 ring_req->u.rw.id = id;
 471                 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 472                         /*
 473                          * The indirect operation can only be a BLKIF_OP_READ or
 474                          * BLKIF_OP_WRITE
 475                          */
 476                         BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
 477                         ring_req->operation = BLKIF_OP_INDIRECT;
 478                         ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 479                                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 480                         ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 481                         ring_req->u.indirect.handle = info->handle;
 482                         ring_req->u.indirect.nr_segments = nseg;
 483                 } else {
 484                         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 485                         ring_req->u.rw.handle = info->handle;
 486                         ring_req->operation = rq_data_dir(req) ?
 487                                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 488                         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 489                                 /*
 490                                  * Ideally we can do an unordered flush-to-disk. In case the
 491                                  * backend onlysupports barriers, use that. A barrier request
 492                                  * a superset of FUA, so we can implement it the same
 493                                  * way.  (It's also a FLUSH+FUA, since it is
 494                                  * guaranteed ordered WRT previous writes.)
 495                                  */
 496                                 switch (info->feature_flush &
 497                                         ((REQ_FLUSH|REQ_FUA))) {
 498                                 case REQ_FLUSH|REQ_FUA:
 499                                         ring_req->operation =
 500                                                 BLKIF_OP_WRITE_BARRIER;
 501                                         break;
 502                                 case REQ_FLUSH:
 503                                         ring_req->operation =
 504                                                 BLKIF_OP_FLUSH_DISKCACHE;
 505                                         break;
 506                                 default:
 507                                         ring_req->operation = 0;
 508                                 }
 509                         }
 510                         ring_req->u.rw.nr_segments = nseg;
 511                 }
 512                 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
 513                         fsect = sg->offset >> 9;
 514                         lsect = fsect + (sg->length >> 9) - 1;
 515
 516                         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 517                             (i % SEGS_PER_INDIRECT_FRAME == 0)) {
 518                                 unsigned long uninitialized_var(pfn);
 519
 520                                 if (segments)
 521                                         kunmap_atomic(segments);
 522
 523                                 n = i / SEGS_PER_INDIRECT_FRAME;
 524                                 if (!info->feature_persistent) {
 525                                         struct page *indirect_page;
 526
 527                                         /* Fetch a pre-allocated page to use for indirect grefs */
 528                                         BUG_ON(list_empty(&info->indirect_pages));
 529                                         indirect_page = list_first_entry(&info->indirect_pages,
 530                                                                          struct page, lru);
 531                                         list_del(&indirect_page->lru);
 532                                         pfn = page_to_pfn(indirect_page);
 533                                 }
 534                                 gnt_list_entry = get_grant(&gref_head, pfn, info);
 535                                 info->shadow[id].indirect_grants[n] = gnt_list_entry;
 536                                 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
 537                                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 538                         }
 539
 540                         gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
 541                         ref = gnt_list_entry->gref;
 542
 543                         info->shadow[id].grants_used[i] = gnt_list_entry;
 544
 545                         if (rq_data_dir(req) && info->feature_persistent) {
 546                                 char *bvec_data;
 547                                 void *shared_data;
 548
 549                                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 550
 551                                 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
 552                                 bvec_data = kmap_atomic(sg_page(sg));
 553
 554                                 /*
 555                                  * this does not wipe data stored outside the
 556                                  * range sg->offset..sg->offset+sg->length.
 557                                  * Therefore, blkback *could* see data from
 558                                  * previous requests. This is OK as long as
 559                                  * persistent grants are shared with just one
 560                                  * domain. It may need refactoring if this
 561                                  * changes
 562                                  */
 563                                 memcpy(shared_data + sg->offset,
 564                                        bvec_data   + sg->offset,
 565                                        sg->length);
 566
 567                                 kunmap_atomic(bvec_data);
 568                                 kunmap_atomic(shared_data);
 569                         }
 570                         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 571                                 ring_req->u.rw.seg[i] =
 572                                                 (struct blkif_request_segment) {
 573                                                         .gref       = ref,
 574                                                         .first_sect = fsect,
 575                                                         .last_sect  = lsect };
 576                         } else {
 577                                 n = i % SEGS_PER_INDIRECT_FRAME;
 578                                 segments[n] =
 579                                         (struct blkif_request_segment) {
 580                                                         .gref       = ref,
 581                                                         .first_sect = fsect,
 582                                                         .last_sect  = lsect };
 583                         }
 584                 }
 585                 if (segments)
 586                         kunmap_atomic(segments);
 587         }
 588
 589         info->ring.req_prod_pvt++;
 590
 591         /* Keep a private copy so we can reissue requests when recovering. */
 592         info->shadow[id].req = *ring_req;
 593
 594         if (new_persistent_gnts)
 595                 gnttab_free_grant_references(gref_head);
 596
 597         return 0;
 598 }
 599
 600
 601 static inline void flush_requests(struct blkfront_info *info)
 602 {
 603         int notify;
 604
 605         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 606
 607         if (notify)
 608                 notify_remote_via_irq(info->irq);
 609 }
 610
 611 static inline bool blkif_request_flush_invalid(struct request *req,
 612                                                struct blkfront_info *info)
 613 {
 614         return ((req->cmd_type != REQ_TYPE_FS) ||
 615                 ((req->cmd_flags & REQ_FLUSH) &&
 616                  !(info->feature_flush & REQ_FLUSH)) ||
 617                 ((req->cmd_flags & REQ_FUA) &&
 618                  !(info->feature_flush & REQ_FUA)));
 619 }
 620
 621 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 622                            const struct blk_mq_queue_data *qd)
 623 {
 624         struct blkfront_info *info = qd->rq->rq_disk->private_data;
 625
 626         blk_mq_start_request(qd->rq);
 627         spin_lock_irq(&info->io_lock);
 628         if (RING_FULL(&info->ring))
 629                 goto out_busy;
 630
 631         if (blkif_request_flush_invalid(qd->rq, info))
 632                 goto out_err;
 633
 634         if (blkif_queue_request(qd->rq))
 635                 goto out_busy;
 636
 637         flush_requests(info);
 638         spin_unlock_irq(&info->io_lock);
 639         return BLK_MQ_RQ_QUEUE_OK;
 640
 641 out_err:
 642         spin_unlock_irq(&info->io_lock);
 643         return BLK_MQ_RQ_QUEUE_ERROR;
 644
 645 out_busy:
 646         spin_unlock_irq(&info->io_lock);
 647         blk_mq_stop_hw_queue(hctx);
 648         return BLK_MQ_RQ_QUEUE_BUSY;
 649 }
 650
 651 static struct blk_mq_ops blkfront_mq_ops = {
 652         .queue_rq = blkif_queue_rq,
 653         .map_queue = blk_mq_map_queue,
 654 };
 655
 656 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 657                                 unsigned int physical_sector_size,
 658                                 unsigned int segments)
 659 {
 660         struct request_queue *rq;
 661         struct blkfront_info *info = gd->private_data;
 662
 663         memset(&info->tag_set, 0, sizeof(info->tag_set));
 664         info->tag_set.ops = &blkfront_mq_ops;
 665         info->tag_set.nr_hw_queues = 1;
 666         info->tag_set.queue_depth =  BLK_RING_SIZE(info);
 667         info->tag_set.numa_node = NUMA_NO_NODE;
 668         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
 669         info->tag_set.cmd_size = 0;
 670         info->tag_set.driver_data = info;
 671
 672         if (blk_mq_alloc_tag_set(&info->tag_set))
 673                 return -1;
 674         rq = blk_mq_init_queue(&info->tag_set);
 675         if (IS_ERR(rq)) {
 676                 blk_mq_free_tag_set(&info->tag_set);
 677                 return -1;
 678         }
 679
 680         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 681
 682         if (info->feature_discard) {
 683                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 684                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 685                 rq->limits.discard_granularity = info->discard_granularity;
 686                 rq->limits.discard_alignment = info->discard_alignment;
 687                 if (info->feature_secdiscard)
 688                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 689         }
 690
 691         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 692         blk_queue_logical_block_size(rq, sector_size);
 693         blk_queue_physical_block_size(rq, physical_sector_size);
 694         blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
 695
 696         /* Each segment in a request is up to an aligned page in size. */
 697         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 698         blk_queue_max_segment_size(rq, PAGE_SIZE);
 699
 700         /* Ensure a merged request will fit in a single I/O ring slot. */
 701         blk_queue_max_segments(rq, segments);
 702
 703         /* Make sure buffer addresses are sector-aligned. */
 704         blk_queue_dma_alignment(rq, 511);
 705
 706         /* Make sure we don't use bounce buffers. */
 707         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 708
 709         gd->queue = rq;
 710
 711         return 0;
 712 }
 713
 714 static const char *flush_info(unsigned int feature_flush)
 715 {
 716         switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
 717         case REQ_FLUSH|REQ_FUA:
 718                 return "barrier: enabled;";
 719         case REQ_FLUSH:
 720                 return "flush diskcache: enabled;";
 721         default:
 722                 return "barrier or flush: disabled;";
 723         }
 724 }
 725
 726 static void xlvbd_flush(struct blkfront_info *info)
 727 {
 728         blk_queue_flush(info->rq, info->feature_flush);
 729         pr_info("blkfront: %s: %s %s %s %s %s\n",
 730                 info->gd->disk_name, flush_info(info->feature_flush),
 731                 "persistent grants:", info->feature_persistent ?
 732                 "enabled;" : "disabled;", "indirect descriptors:",
 733                 info->max_indirect_segments ? "enabled;" : "disabled;");
 734 }
 735
 736 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 737 {
 738         int major;
 739         major = BLKIF_MAJOR(vdevice);
 740         *minor = BLKIF_MINOR(vdevice);
 741         switch (major) {
 742                 case XEN_IDE0_MAJOR:
 743                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 744                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 745                                 EMULATED_HD_DISK_MINOR_OFFSET;
 746                         break;
 747                 case XEN_IDE1_MAJOR:
 748                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 749                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 750                                 EMULATED_HD_DISK_MINOR_OFFSET;
 751                         break;
 752                 case XEN_SCSI_DISK0_MAJOR:
 753                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 754                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 755                         break;
 756                 case XEN_SCSI_DISK1_MAJOR:
 757                 case XEN_SCSI_DISK2_MAJOR:
 758                 case XEN_SCSI_DISK3_MAJOR:
 759                 case XEN_SCSI_DISK4_MAJOR:
 760                 case XEN_SCSI_DISK5_MAJOR:
 761                 case XEN_SCSI_DISK6_MAJOR:
 762                 case XEN_SCSI_DISK7_MAJOR:
 763                         *offset = (*minor / PARTS_PER_DISK) +
 764                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 765                                 EMULATED_SD_DISK_NAME_OFFSET;
 766                         *minor = *minor +
 767                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 768                                 EMULATED_SD_DISK_MINOR_OFFSET;
 769                         break;
 770                 case XEN_SCSI_DISK8_MAJOR:
 771                 case XEN_SCSI_DISK9_MAJOR:
 772                 case XEN_SCSI_DISK10_MAJOR:
 773                 case XEN_SCSI_DISK11_MAJOR:
 774                 case XEN_SCSI_DISK12_MAJOR:
 775                 case XEN_SCSI_DISK13_MAJOR:
 776                 case XEN_SCSI_DISK14_MAJOR:
 777                 case XEN_SCSI_DISK15_MAJOR:
 778                         *offset = (*minor / PARTS_PER_DISK) +
 779                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 780                                 EMULATED_SD_DISK_NAME_OFFSET;
 781                         *minor = *minor +
 782                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 783                                 EMULATED_SD_DISK_MINOR_OFFSET;
 784                         break;
 785                 case XENVBD_MAJOR:
 786                         *offset = *minor / PARTS_PER_DISK;
 787                         break;
 788                 default:
 789                         printk(KERN_WARNING "blkfront: your disk configuration is "
 790                                         "incorrect, please use an xvd device instead\n");
 791                         return -ENODEV;
 792         }
 793         return 0;
 794 }
 795
 796 static char *encode_disk_name(char *ptr, unsigned int n)
 797 {
 798         if (n >= 26)
 799                 ptr = encode_disk_name(ptr, n / 26 - 1);
 800         *ptr = 'a' + n % 26;
 801         return ptr + 1;
 802 }
 803
 804 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 805                                struct blkfront_info *info,
 806                                u16 vdisk_info, u16 sector_size,
 807                                unsigned int physical_sector_size)
 808 {
 809         struct gendisk *gd;
 810         int nr_minors = 1;
 811         int err;
 812         unsigned int offset;
 813         int minor;
 814         int nr_parts;
 815         char *ptr;
 816
 817         BUG_ON(info->gd != NULL);
 818         BUG_ON(info->rq != NULL);
 819
 820         if ((info->vdevice>>EXT_SHIFT) > 1) {
 821                 /* this is above the extended range; something is wrong */
 822                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 823                 return -ENODEV;
 824         }
 825
 826         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 827                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 828                 if (err)
 829                         return err;
 830                 nr_parts = PARTS_PER_DISK;
 831         } else {
 832                 minor = BLKIF_MINOR_EXT(info->vdevice);
 833                 nr_parts = PARTS_PER_EXT_DISK;
 834                 offset = minor / nr_parts;
 835                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 836                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 837                                         "emulated IDE disks,\n\t choose an xvd device name"
 838                                         "from xvde on\n", info->vdevice);
 839         }
 840         if (minor >> MINORBITS) {
 841                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 842                         info->vdevice, minor);
 843                 return -ENODEV;
 844         }
 845
 846         if ((minor % nr_parts) == 0)
 847                 nr_minors = nr_parts;
 848
 849         err = xlbd_reserve_minors(minor, nr_minors);
 850         if (err)
 851                 goto out;
 852         err = -ENODEV;
 853
 854         gd = alloc_disk(nr_minors);
 855         if (gd == NULL)
 856                 goto release;
 857
 858         strcpy(gd->disk_name, DEV_NAME);
 859         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 860         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 861         if (nr_minors > 1)
 862                 *ptr = 0;
 863         else
 864                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 865                          "%d", minor & (nr_parts - 1));
 866
 867         gd->major = XENVBD_MAJOR;
 868         gd->first_minor = minor;
 869         gd->fops = &xlvbd_block_fops;
 870         gd->private_data = info;
 871         gd->driverfs_dev = &(info->xbdev->dev);
 872         set_capacity(gd, capacity);
 873
 874         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
 875                                  info->max_indirect_segments ? :
 876                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 877                 del_gendisk(gd);
 878                 goto release;
 879         }
 880
 881         info->rq = gd->queue;
 882         info->gd = gd;
 883
 884         xlvbd_flush(info);
 885
 886         if (vdisk_info & VDISK_READONLY)
 887                 set_disk_ro(gd, 1);
 888
 889         if (vdisk_info & VDISK_REMOVABLE)
 890                 gd->flags |= GENHD_FL_REMOVABLE;
 891
 892         if (vdisk_info & VDISK_CDROM)
 893                 gd->flags |= GENHD_FL_CD;
 894
 895         return 0;
 896
 897  release:
 898         xlbd_release_minors(minor, nr_minors);
 899  out:
 900         return err;
 901 }
 902
 903 static void xlvbd_release_gendisk(struct blkfront_info *info)
 904 {
 905         unsigned int minor, nr_minors;
 906
 907         if (info->rq == NULL)
 908                 return;
 909
 910         /* No more blkif_request(). */
 911         blk_mq_stop_hw_queues(info->rq);
 912
 913         /* No more gnttab callback work. */
 914         gnttab_cancel_free_callback(&info->callback);
 915
 916         /* Flush gnttab callback work. Must be done with no locks held. */
 917         flush_work(&info->work);
 918
 919         del_gendisk(info->gd);
 920
 921         minor = info->gd->first_minor;
 922         nr_minors = info->gd->minors;
 923         xlbd_release_minors(minor, nr_minors);
 924
 925         blk_cleanup_queue(info->rq);
 926         blk_mq_free_tag_set(&info->tag_set);
 927         info->rq = NULL;
 928
 929         put_disk(info->gd);
 930         info->gd = NULL;
 931 }
 932
 933 /* Must be called with io_lock holded */
 934 static void kick_pending_request_queues(struct blkfront_info *info)
 935 {
 936         if (!RING_FULL(&info->ring))
 937                 blk_mq_start_stopped_hw_queues(info->rq, true);
 938 }
 939
 940 static void blkif_restart_queue(struct work_struct *work)
 941 {
 942         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 943
 944         spin_lock_irq(&info->io_lock);
 945         if (info->connected == BLKIF_STATE_CONNECTED)
 946                 kick_pending_request_queues(info);
 947         spin_unlock_irq(&info->io_lock);
 948 }
 949
 950 static void blkif_free(struct blkfront_info *info, int suspend)
 951 {
 952         struct grant *persistent_gnt;
 953         struct grant *n;
 954         int i, j, segs;
 955
 956         /* Prevent new requests being issued until we fix things up. */
 957         spin_lock_irq(&info->io_lock);
 958         info->connected = suspend ?
 959                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 960         /* No more blkif_request(). */
 961         if (info->rq)
 962                 blk_mq_stop_hw_queues(info->rq);
 963
 964         /* Remove all persistent grants */
 965         if (!list_empty(&info->grants)) {
 966                 list_for_each_entry_safe(persistent_gnt, n,
 967                                          &info->grants, node) {
 968                         list_del(&persistent_gnt->node);
 969                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
 970                                 gnttab_end_foreign_access(persistent_gnt->gref,
 971                                                           0, 0UL);
 972                                 info->persistent_gnts_c--;
 973                         }
 974                         if (info->feature_persistent)
 975                                 __free_page(pfn_to_page(persistent_gnt->pfn));
 976                         kfree(persistent_gnt);
 977                 }
 978         }
 979         BUG_ON(info->persistent_gnts_c != 0);
 980
 981         /*
 982          * Remove indirect pages, this only happens when using indirect
 983          * descriptors but not persistent grants
 984          */
 985         if (!list_empty(&info->indirect_pages)) {
 986                 struct page *indirect_page, *n;
 987
 988                 BUG_ON(info->feature_persistent);
 989                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
 990                         list_del(&indirect_page->lru);
 991                         __free_page(indirect_page);
 992                 }
 993         }
 994
 995         for (i = 0; i < BLK_RING_SIZE(info); i++) {
 996                 /*
 997                  * Clear persistent grants present in requests already
 998                  * on the shared ring
 999                  */
1000                 if (!info->shadow[i].request)
1001                         goto free_shadow;
1002
1003                 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1004                        info->shadow[i].req.u.indirect.nr_segments :
1005                        info->shadow[i].req.u.rw.nr_segments;
1006                 for (j = 0; j < segs; j++) {
1007                         persistent_gnt = info->shadow[i].grants_used[j];
1008                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1009                         if (info->feature_persistent)
1010                                 __free_page(pfn_to_page(persistent_gnt->pfn));
1011                         kfree(persistent_gnt);
1012                 }
1013
1014                 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1015                         /*
1016                          * If this is not an indirect operation don't try to
1017                          * free indirect segments
1018                          */
1019                         goto free_shadow;
1020
1021                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1022                         persistent_gnt = info->shadow[i].indirect_grants[j];
1023                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1024                         __free_page(pfn_to_page(persistent_gnt->pfn));
1025                         kfree(persistent_gnt);
1026                 }
1027
1028 free_shadow:
1029                 kfree(info->shadow[i].grants_used);
1030                 info->shadow[i].grants_used = NULL;
1031                 kfree(info->shadow[i].indirect_grants);
1032                 info->shadow[i].indirect_grants = NULL;
1033                 kfree(info->shadow[i].sg);
1034                 info->shadow[i].sg = NULL;
1035         }
1036
1037         /* No more gnttab callback work. */
1038         gnttab_cancel_free_callback(&info->callback);
1039         spin_unlock_irq(&info->io_lock);
1040
1041         /* Flush gnttab callback work. Must be done with no locks held. */
1042         flush_work(&info->work);
1043
1044         /* Free resources associated with old device channel. */
1045         for (i = 0; i < info->nr_ring_pages; i++) {
1046                 if (info->ring_ref[i] != GRANT_INVALID_REF) {
1047                         gnttab_end_foreign_access(info->ring_ref[i], 0, 0);
1048                         info->ring_ref[i] = GRANT_INVALID_REF;
1049                 }
1050         }
1051         free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1052         info->ring.sring = NULL;
1053
1054         if (info->irq)
1055                 unbind_from_irqhandler(info->irq, info);
1056         info->evtchn = info->irq = 0;
1057
1058 }
1059
1060 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1061                              struct blkif_response *bret)
1062 {
1063         int i = 0;
1064         struct scatterlist *sg;
1065         char *bvec_data;
1066         void *shared_data;
1067         int nseg;
1068
1069         nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1070                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1071
1072         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1073                 for_each_sg(s->sg, sg, nseg, i) {
1074                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1075                         shared_data = kmap_atomic(
1076                                 pfn_to_page(s->grants_used[i]->pfn));
1077                         bvec_data = kmap_atomic(sg_page(sg));
1078                         memcpy(bvec_data   + sg->offset,
1079                                shared_data + sg->offset,
1080                                sg->length);
1081                         kunmap_atomic(bvec_data);
1082                         kunmap_atomic(shared_data);
1083                 }
1084         }
1085         /* Add the persistent grant into the list of free grants */
1086         for (i = 0; i < nseg; i++) {
1087                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1088                         /*
1089                          * If the grant is still mapped by the backend (the
1090                          * backend has chosen to make this grant persistent)
1091                          * we add it at the head of the list, so it will be
1092                          * reused first.
1093                          */
1094                         if (!info->feature_persistent)
1095                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1096                                                      s->grants_used[i]->gref);
1097                         list_add(&s->grants_used[i]->node, &info->grants);
1098                         info->persistent_gnts_c++;
1099                 } else {
1100                         /*
1101                          * If the grant is not mapped by the backend we end the
1102                          * foreign access and add it to the tail of the list,
1103                          * so it will not be picked again unless we run out of
1104                          * persistent grants.
1105                          */
1106                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1107                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1108                         list_add_tail(&s->grants_used[i]->node, &info->grants);
1109                 }
1110         }
1111         if (s->req.operation == BLKIF_OP_INDIRECT) {
1112                 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1113                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1114                                 if (!info->feature_persistent)
1115                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1116                                                              s->indirect_grants[i]->gref);
1117                                 list_add(&s->indirect_grants[i]->node, &info->grants);
1118                                 info->persistent_gnts_c++;
1119                         } else {
1120                                 struct page *indirect_page;
1121
1122                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1123                                 /*
1124                                  * Add the used indirect page back to the list of
1125                                  * available pages for indirect grefs.
1126                                  */
1127                                 if (!info->feature_persistent) {
1128                                         indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1129                                         list_add(&indirect_page->lru, &info->indirect_pages);
1130                                 }
1131                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1132                                 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1133                         }
1134                 }
1135         }
1136 }
1137
1138 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1139 {
1140         struct request *req;
1141         struct blkif_response *bret;
1142         RING_IDX i, rp;
1143         unsigned long flags;
1144         struct blkfront_info *info = (struct blkfront_info *)dev_id;
1145         int error;
1146
1147         spin_lock_irqsave(&info->io_lock, flags);
1148
1149         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1150                 spin_unlock_irqrestore(&info->io_lock, flags);
1151                 return IRQ_HANDLED;
1152         }
1153
1154  again:
1155         rp = info->ring.sring->rsp_prod;
1156         rmb(); /* Ensure we see queued responses up to 'rp'. */
1157
1158         for (i = info->ring.rsp_cons; i != rp; i++) {
1159                 unsigned long id;
1160
1161                 bret = RING_GET_RESPONSE(&info->ring, i);
1162                 id   = bret->id;
1163                 /*
1164                  * The backend has messed up and given us an id that we would
1165                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1166                  * look in get_id_from_freelist.
1167                  */
1168                 if (id >= BLK_RING_SIZE(info)) {
1169                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1170                              info->gd->disk_name, op_name(bret->operation), id);
1171                         /* We can't safely get the 'struct request' as
1172                          * the id is busted. */
1173                         continue;
1174                 }
1175                 req  = info->shadow[id].request;
1176
1177                 if (bret->operation != BLKIF_OP_DISCARD)
1178                         blkif_completion(&info->shadow[id], info, bret);
1179
1180                 if (add_id_to_freelist(info, id)) {
1181                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1182                              info->gd->disk_name, op_name(bret->operation), id);
1183                         continue;
1184                 }
1185
1186                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1187                 switch (bret->operation) {
1188                 case BLKIF_OP_DISCARD:
1189                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1190                                 struct request_queue *rq = info->rq;
1191                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1192                                            info->gd->disk_name, op_name(bret->operation));
1193                                 error = -EOPNOTSUPP;
1194                                 info->feature_discard = 0;
1195                                 info->feature_secdiscard = 0;
1196                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1197                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1198                         }
1199                         blk_mq_complete_request(req, error);
1200                         break;
1201                 case BLKIF_OP_FLUSH_DISKCACHE:
1202                 case BLKIF_OP_WRITE_BARRIER:
1203                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1204                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1205                                        info->gd->disk_name, op_name(bret->operation));
1206                                 error = -EOPNOTSUPP;
1207                         }
1208                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1209                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
1210                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1211                                        info->gd->disk_name, op_name(bret->operation));
1212                                 error = -EOPNOTSUPP;
1213                         }
1214                         if (unlikely(error)) {
1215                                 if (error == -EOPNOTSUPP)
1216                                         error = 0;
1217                                 info->feature_flush = 0;
1218                                 xlvbd_flush(info);
1219                         }
1220                         /* fall through */
1221                 case BLKIF_OP_READ:
1222                 case BLKIF_OP_WRITE:
1223                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1224                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1225                                         "request: %x\n", bret->status);
1226
1227                         blk_mq_complete_request(req, error);
1228                         break;
1229                 default:
1230                         BUG();
1231                 }
1232         }
1233
1234         info->ring.rsp_cons = i;
1235
1236         if (i != info->ring.req_prod_pvt) {
1237                 int more_to_do;
1238                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1239                 if (more_to_do)
1240                         goto again;
1241         } else
1242                 info->ring.sring->rsp_event = i + 1;
1243
1244         kick_pending_request_queues(info);
1245
1246         spin_unlock_irqrestore(&info->io_lock, flags);
1247
1248         return IRQ_HANDLED;
1249 }
1250
1251
1252 static int setup_blkring(struct xenbus_device *dev,
1253                          struct blkfront_info *info)
1254 {
1255         struct blkif_sring *sring;
1256         int err, i;
1257         unsigned long ring_size = info->nr_ring_pages * PAGE_SIZE;
1258         grant_ref_t gref[XENBUS_MAX_RING_PAGES];
1259
1260         for (i = 0; i < info->nr_ring_pages; i++)
1261                 info->ring_ref[i] = GRANT_INVALID_REF;
1262
1263         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1264                                                        get_order(ring_size));
1265         if (!sring) {
1266                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1267                 return -ENOMEM;
1268         }
1269         SHARED_RING_INIT(sring);
1270         FRONT_RING_INIT(&info->ring, sring, ring_size);
1271
1272         err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, gref);
1273         if (err < 0) {
1274                 free_pages((unsigned long)sring, get_order(ring_size));
1275                 info->ring.sring = NULL;
1276                 goto fail;
1277         }
1278         for (i = 0; i < info->nr_ring_pages; i++)
1279                 info->ring_ref[i] = gref[i];
1280
1281         err = xenbus_alloc_evtchn(dev, &info->evtchn);
1282         if (err)
1283                 goto fail;
1284
1285         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1286                                         "blkif", info);
1287         if (err <= 0) {
1288                 xenbus_dev_fatal(dev, err,
1289                                  "bind_evtchn_to_irqhandler failed");
1290                 goto fail;
1291         }
1292         info->irq = err;
1293
1294         return 0;
1295 fail:
1296         blkif_free(info, 0);
1297         return err;
1298 }
1299
1300
1301 /* Common code used when first setting up, and when resuming. */
1302 static int talk_to_blkback(struct xenbus_device *dev,
1303                            struct blkfront_info *info)
1304 {
1305         const char *message = NULL;
1306         struct xenbus_transaction xbt;
1307         int err, i;
1308         unsigned int max_page_order = 0;
1309         unsigned int ring_page_order = 0;
1310
1311         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1312                            "max-ring-page-order", "%u", &max_page_order);
1313         if (err != 1)
1314                 info->nr_ring_pages = 1;
1315         else {
1316                 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1317                 info->nr_ring_pages = 1 << ring_page_order;
1318         }
1319
1320         /* Create shared ring, alloc event channel. */
1321         err = setup_blkring(dev, info);
1322         if (err)
1323                 goto out;
1324
1325 again:
1326         err = xenbus_transaction_start(&xbt);
1327         if (err) {
1328                 xenbus_dev_fatal(dev, err, "starting transaction");
1329                 goto destroy_blkring;
1330         }
1331
1332         if (info->nr_ring_pages == 1) {
1333                 err = xenbus_printf(xbt, dev->nodename,
1334                                     "ring-ref", "%u", info->ring_ref[0]);
1335                 if (err) {
1336                         message = "writing ring-ref";
1337                         goto abort_transaction;
1338                 }
1339         } else {
1340                 err = xenbus_printf(xbt, dev->nodename,
1341                                     "ring-page-order", "%u", ring_page_order);
1342                 if (err) {
1343                         message = "writing ring-page-order";
1344                         goto abort_transaction;
1345                 }
1346
1347                 for (i = 0; i < info->nr_ring_pages; i++) {
1348                         char ring_ref_name[RINGREF_NAME_LEN];
1349
1350                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1351                         err = xenbus_printf(xbt, dev->nodename, ring_ref_name,
1352                                             "%u", info->ring_ref[i]);
1353                         if (err) {
1354                                 message = "writing ring-ref";
1355                                 goto abort_transaction;
1356                         }
1357                 }
1358         }
1359         err = xenbus_printf(xbt, dev->nodename,
1360                             "event-channel", "%u", info->evtchn);
1361         if (err) {
1362                 message = "writing event-channel";
1363                 goto abort_transaction;
1364         }
1365         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1366                             XEN_IO_PROTO_ABI_NATIVE);
1367         if (err) {
1368                 message = "writing protocol";
1369                 goto abort_transaction;
1370         }
1371         err = xenbus_printf(xbt, dev->nodename,
1372                             "feature-persistent", "%u", 1);
1373         if (err)
1374                 dev_warn(&dev->dev,
1375                          "writing persistent grants feature to xenbus");
1376
1377         err = xenbus_transaction_end(xbt, 0);
1378         if (err) {
1379                 if (err == -EAGAIN)
1380                         goto again;
1381                 xenbus_dev_fatal(dev, err, "completing transaction");
1382                 goto destroy_blkring;
1383         }
1384
1385         for (i = 0; i < BLK_RING_SIZE(info); i++)
1386                 info->shadow[i].req.u.rw.id = i+1;
1387         info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1388         xenbus_switch_state(dev, XenbusStateInitialised);
1389
1390         return 0;
1391
1392  abort_transaction:
1393         xenbus_transaction_end(xbt, 1);
1394         if (message)
1395                 xenbus_dev_fatal(dev, err, "%s", message);
1396  destroy_blkring:
1397         blkif_free(info, 0);
1398  out:
1399         return err;
1400 }
1401
1402 /**
1403  * Entry point to this code when a new device is created.  Allocate the basic
1404  * structures and the ring buffer for communication with the backend, and
1405  * inform the backend of the appropriate details for those.  Switch to
1406  * Initialised state.
1407  */
1408 static int blkfront_probe(struct xenbus_device *dev,
1409                           const struct xenbus_device_id *id)
1410 {
1411         int err, vdevice;
1412         struct blkfront_info *info;
1413
1414         /* FIXME: Use dynamic device id if this is not set. */
1415         err = xenbus_scanf(XBT_NIL, dev->nodename,
1416                            "virtual-device", "%i", &vdevice);
1417         if (err != 1) {
1418                 /* go looking in the extended area instead */
1419                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1420                                    "%i", &vdevice);
1421                 if (err != 1) {
1422                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1423                         return err;
1424                 }
1425         }
1426
1427         if (xen_hvm_domain()) {
1428                 char *type;
1429                 int len;
1430                 /* no unplug has been done: do not hook devices != xen vbds */
1431                 if (xen_has_pv_and_legacy_disk_devices()) {
1432                         int major;
1433
1434                         if (!VDEV_IS_EXTENDED(vdevice))
1435                                 major = BLKIF_MAJOR(vdevice);
1436                         else
1437                                 major = XENVBD_MAJOR;
1438
1439                         if (major != XENVBD_MAJOR) {
1440                                 printk(KERN_INFO
1441                                                 "%s: HVM does not support vbd %d as xen block device\n",
1442                                                 __func__, vdevice);
1443                                 return -ENODEV;
1444                         }
1445                 }
1446                 /* do not create a PV cdrom device if we are an HVM guest */
1447                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1448                 if (IS_ERR(type))
1449                         return -ENODEV;
1450                 if (strncmp(type, "cdrom", 5) == 0) {
1451                         kfree(type);
1452                         return -ENODEV;
1453                 }
1454                 kfree(type);
1455         }
1456         info = kzalloc(sizeof(*info), GFP_KERNEL);
1457         if (!info) {
1458                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1459                 return -ENOMEM;
1460         }
1461
1462         mutex_init(&info->mutex);
1463         spin_lock_init(&info->io_lock);
1464         info->xbdev = dev;
1465         info->vdevice = vdevice;
1466         INIT_LIST_HEAD(&info->grants);
1467         INIT_LIST_HEAD(&info->indirect_pages);
1468         info->persistent_gnts_c = 0;
1469         info->connected = BLKIF_STATE_DISCONNECTED;
1470         INIT_WORK(&info->work, blkif_restart_queue);
1471
1472         /* Front end dir is a number, which is used as the id. */
1473         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1474         dev_set_drvdata(&dev->dev, info);
1475
1476         return 0;
1477 }
1478
1479 static void split_bio_end(struct bio *bio)
1480 {
1481         struct split_bio *split_bio = bio->bi_private;
1482
1483         if (atomic_dec_and_test(&split_bio->pending)) {
1484                 split_bio->bio->bi_phys_segments = 0;
1485                 split_bio->bio->bi_error = bio->bi_error;
1486                 bio_endio(split_bio->bio);
1487                 kfree(split_bio);
1488         }
1489         bio_put(bio);
1490 }
1491
1492 static int blkif_recover(struct blkfront_info *info)
1493 {
1494         int i;
1495         struct request *req, *n;
1496         struct blk_shadow *copy;
1497         int rc;
1498         struct bio *bio, *cloned_bio;
1499         struct bio_list bio_list, merge_bio;
1500         unsigned int segs, offset;
1501         int pending, size;
1502         struct split_bio *split_bio;
1503         struct list_head requests;
1504
1505         /* Stage 1: Make a safe copy of the shadow state. */
1506         copy = kmemdup(info->shadow, sizeof(info->shadow),
1507                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1508         if (!copy)
1509                 return -ENOMEM;
1510
1511         /* Stage 2: Set up free list. */
1512         memset(&info->shadow, 0, sizeof(info->shadow));
1513         for (i = 0; i < BLK_RING_SIZE(info); i++)
1514                 info->shadow[i].req.u.rw.id = i+1;
1515         info->shadow_free = info->ring.req_prod_pvt;
1516         info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1517
1518         rc = blkfront_gather_backend_features(info);
1519         if (rc) {
1520                 kfree(copy);
1521                 return rc;
1522         }
1523
1524         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1525         blk_queue_max_segments(info->rq, segs);
1526         bio_list_init(&bio_list);
1527         INIT_LIST_HEAD(&requests);
1528         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1529                 /* Not in use? */
1530                 if (!copy[i].request)
1531                         continue;
1532
1533                 /*
1534                  * Get the bios in the request so we can re-queue them.
1535                  */
1536                 if (copy[i].request->cmd_flags &
1537                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1538                         /*
1539                          * Flush operations don't contain bios, so
1540                          * we need to requeue the whole request
1541                          */
1542                         list_add(&copy[i].request->queuelist, &requests);
1543                         continue;
1544                 }
1545                 merge_bio.head = copy[i].request->bio;
1546                 merge_bio.tail = copy[i].request->biotail;
1547                 bio_list_merge(&bio_list, &merge_bio);
1548                 copy[i].request->bio = NULL;
1549                 blk_end_request_all(copy[i].request, 0);
1550         }
1551
1552         kfree(copy);
1553
1554         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1555
1556         spin_lock_irq(&info->io_lock);
1557
1558         /* Now safe for us to use the shared ring */
1559         info->connected = BLKIF_STATE_CONNECTED;
1560
1561         /* Kick any other new requests queued since we resumed */
1562         kick_pending_request_queues(info);
1563
1564         list_for_each_entry_safe(req, n, &requests, queuelist) {
1565                 /* Requeue pending requests (flush or discard) */
1566                 list_del_init(&req->queuelist);
1567                 BUG_ON(req->nr_phys_segments > segs);
1568                 blk_mq_requeue_request(req);
1569         }
1570         spin_unlock_irq(&info->io_lock);
1571         blk_mq_kick_requeue_list(info->rq);
1572
1573         while ((bio = bio_list_pop(&bio_list)) != NULL) {
1574                 /* Traverse the list of pending bios and re-queue them */
1575                 if (bio_segments(bio) > segs) {
1576                         /*
1577                          * This bio has more segments than what we can
1578                          * handle, we have to split it.
1579                          */
1580                         pending = (bio_segments(bio) + segs - 1) / segs;
1581                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1582                         BUG_ON(split_bio == NULL);
1583                         atomic_set(&split_bio->pending, pending);
1584                         split_bio->bio = bio;
1585                         for (i = 0; i < pending; i++) {
1586                                 offset = (i * segs * PAGE_SIZE) >> 9;
1587                                 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1588                                            (unsigned int)bio_sectors(bio) - offset);
1589                                 cloned_bio = bio_clone(bio, GFP_NOIO);
1590                                 BUG_ON(cloned_bio == NULL);
1591                                 bio_trim(cloned_bio, offset, size);
1592                                 cloned_bio->bi_private = split_bio;
1593                                 cloned_bio->bi_end_io = split_bio_end;
1594                                 submit_bio(cloned_bio->bi_rw, cloned_bio);
1595                         }
1596                         /*
1597                          * Now we have to wait for all those smaller bios to
1598                          * end, so we can also end the "parent" bio.
1599                          */
1600                         continue;
1601                 }
1602                 /* We don't need to split this bio */
1603                 submit_bio(bio->bi_rw, bio);
1604         }
1605
1606         return 0;
1607 }
1608
1609 /**
1610  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1611  * driver restart.  We tear down our blkif structure and recreate it, but
1612  * leave the device-layer structures intact so that this is transparent to the
1613  * rest of the kernel.
1614  */
1615 static int blkfront_resume(struct xenbus_device *dev)
1616 {
1617         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1618         int err;
1619
1620         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1621
1622         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1623
1624         err = talk_to_blkback(dev, info);
1625
1626         /*
1627          * We have to wait for the backend to switch to
1628          * connected state, since we want to read which
1629          * features it supports.
1630          */
1631
1632         return err;
1633 }
1634
1635 static void
1636 blkfront_closing(struct blkfront_info *info)
1637 {
1638         struct xenbus_device *xbdev = info->xbdev;
1639         struct block_device *bdev = NULL;
1640
1641         mutex_lock(&info->mutex);
1642
1643         if (xbdev->state == XenbusStateClosing) {
1644                 mutex_unlock(&info->mutex);
1645                 return;
1646         }
1647
1648         if (info->gd)
1649                 bdev = bdget_disk(info->gd, 0);
1650
1651         mutex_unlock(&info->mutex);
1652
1653         if (!bdev) {
1654                 xenbus_frontend_closed(xbdev);
1655                 return;
1656         }
1657
1658         mutex_lock(&bdev->bd_mutex);
1659
1660         if (bdev->bd_openers) {
1661                 xenbus_dev_error(xbdev, -EBUSY,
1662                                  "Device in use; refusing to close");
1663                 xenbus_switch_state(xbdev, XenbusStateClosing);
1664         } else {
1665                 xlvbd_release_gendisk(info);
1666                 xenbus_frontend_closed(xbdev);
1667         }
1668
1669         mutex_unlock(&bdev->bd_mutex);
1670         bdput(bdev);
1671 }
1672
1673 static void blkfront_setup_discard(struct blkfront_info *info)
1674 {
1675         int err;
1676         unsigned int discard_granularity;
1677         unsigned int discard_alignment;
1678         unsigned int discard_secure;
1679
1680         info->feature_discard = 1;
1681         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1682                 "discard-granularity", "%u", &discard_granularity,
1683                 "discard-alignment", "%u", &discard_alignment,
1684                 NULL);
1685         if (!err) {
1686                 info->discard_granularity = discard_granularity;
1687                 info->discard_alignment = discard_alignment;
1688         }
1689         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1690                     "discard-secure", "%d", &discard_secure,
1691                     NULL);
1692         if (!err)
1693                 info->feature_secdiscard = !!discard_secure;
1694 }
1695
1696 static int blkfront_setup_indirect(struct blkfront_info *info)
1697 {
1698         unsigned int segs;
1699         int err, i;
1700
1701         if (info->max_indirect_segments == 0)
1702                 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1703         else
1704                 segs = info->max_indirect_segments;
1705
1706         err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE(info));
1707         if (err)
1708                 goto out_of_memory;
1709
1710         if (!info->feature_persistent && info->max_indirect_segments) {
1711                 /*
1712                  * We are using indirect descriptors but not persistent
1713                  * grants, we need to allocate a set of pages that can be
1714                  * used for mapping indirect grefs
1715                  */
1716                 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE(info);
1717
1718                 BUG_ON(!list_empty(&info->indirect_pages));
1719                 for (i = 0; i < num; i++) {
1720                         struct page *indirect_page = alloc_page(GFP_NOIO);
1721                         if (!indirect_page)
1722                                 goto out_of_memory;
1723                         list_add(&indirect_page->lru, &info->indirect_pages);
1724                 }
1725         }
1726
1727         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1728                 info->shadow[i].grants_used = kzalloc(
1729                         sizeof(info->shadow[i].grants_used[0]) * segs,
1730                         GFP_NOIO);
1731                 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1732                 if (info->max_indirect_segments)
1733                         info->shadow[i].indirect_grants = kzalloc(
1734                                 sizeof(info->shadow[i].indirect_grants[0]) *
1735                                 INDIRECT_GREFS(segs),
1736                                 GFP_NOIO);
1737                 if ((info->shadow[i].grants_used == NULL) ||
1738                         (info->shadow[i].sg == NULL) ||
1739                      (info->max_indirect_segments &&
1740                      (info->shadow[i].indirect_grants == NULL)))
1741                         goto out_of_memory;
1742                 sg_init_table(info->shadow[i].sg, segs);
1743         }
1744
1745
1746         return 0;
1747
1748 out_of_memory:
1749         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1750                 kfree(info->shadow[i].grants_used);
1751                 info->shadow[i].grants_used = NULL;
1752                 kfree(info->shadow[i].sg);
1753                 info->shadow[i].sg = NULL;
1754                 kfree(info->shadow[i].indirect_grants);
1755                 info->shadow[i].indirect_grants = NULL;
1756         }
1757         if (!list_empty(&info->indirect_pages)) {
1758                 struct page *indirect_page, *n;
1759                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1760                         list_del(&indirect_page->lru);
1761                         __free_page(indirect_page);
1762                 }
1763         }
1764         return -ENOMEM;
1765 }
1766
1767 /*
1768  * Gather all backend feature-*
1769  */
1770 static int blkfront_gather_backend_features(struct blkfront_info *info)
1771 {
1772         int err;
1773         int barrier, flush, discard, persistent;
1774         unsigned int indirect_segments;
1775
1776         info->feature_flush = 0;
1777
1778         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1779                         "feature-barrier", "%d", &barrier,
1780                         NULL);
1781
1782         /*
1783          * If there's no "feature-barrier" defined, then it means
1784          * we're dealing with a very old backend which writes
1785          * synchronously; nothing to do.
1786          *
1787          * If there are barriers, then we use flush.
1788          */
1789         if (!err && barrier)
1790                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1791         /*
1792          * And if there is "feature-flush-cache" use that above
1793          * barriers.
1794          */
1795         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1796                         "feature-flush-cache", "%d", &flush,
1797                         NULL);
1798
1799         if (!err && flush)
1800                 info->feature_flush = REQ_FLUSH;
1801
1802         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1803                         "feature-discard", "%d", &discard,
1804                         NULL);
1805
1806         if (!err && discard)
1807                 blkfront_setup_discard(info);
1808
1809         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1810                         "feature-persistent", "%u", &persistent,
1811                         NULL);
1812         if (err)
1813                 info->feature_persistent = 0;
1814         else
1815                 info->feature_persistent = persistent;
1816
1817         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1818                             "feature-max-indirect-segments", "%u", &indirect_segments,
1819                             NULL);
1820         if (err)
1821                 info->max_indirect_segments = 0;
1822         else
1823                 info->max_indirect_segments = min(indirect_segments,
1824                                                   xen_blkif_max_segments);
1825
1826         return blkfront_setup_indirect(info);
1827 }
1828
1829 /*
1830  * Invoked when the backend is finally 'ready' (and has told produced
1831  * the details about the physical device - #sectors, size, etc).
1832  */
1833 static void blkfront_connect(struct blkfront_info *info)
1834 {
1835         unsigned long long sectors;
1836         unsigned long sector_size;
1837         unsigned int physical_sector_size;
1838         unsigned int binfo;
1839         int err;
1840
1841         switch (info->connected) {
1842         case BLKIF_STATE_CONNECTED:
1843                 /*
1844                  * Potentially, the back-end may be signalling
1845                  * a capacity change; update the capacity.
1846                  */
1847                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1848                                    "sectors", "%Lu", &sectors);
1849                 if (XENBUS_EXIST_ERR(err))
1850                         return;
1851                 printk(KERN_INFO "Setting capacity to %Lu\n",
1852                        sectors);
1853                 set_capacity(info->gd, sectors);
1854                 revalidate_disk(info->gd);
1855
1856                 return;
1857         case BLKIF_STATE_SUSPENDED:
1858                 /*
1859                  * If we are recovering from suspension, we need to wait
1860                  * for the backend to announce it's features before
1861                  * reconnecting, at least we need to know if the backend
1862                  * supports indirect descriptors, and how many.
1863                  */
1864                 blkif_recover(info);
1865                 return;
1866
1867         default:
1868                 break;
1869         }
1870
1871         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1872                 __func__, info->xbdev->otherend);
1873
1874         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1875                             "sectors", "%llu", &sectors,
1876                             "info", "%u", &binfo,
1877                             "sector-size", "%lu", &sector_size,
1878                             NULL);
1879         if (err) {
1880                 xenbus_dev_fatal(info->xbdev, err,
1881                                  "reading backend fields at %s",
1882                                  info->xbdev->otherend);
1883                 return;
1884         }
1885
1886         /*
1887          * physcial-sector-size is a newer field, so old backends may not
1888          * provide this. Assume physical sector size to be the same as
1889          * sector_size in that case.
1890          */
1891         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1892                            "physical-sector-size", "%u", &physical_sector_size);
1893         if (err != 1)
1894                 physical_sector_size = sector_size;
1895
1896         err = blkfront_gather_backend_features(info);
1897         if (err) {
1898                 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1899                                  info->xbdev->otherend);
1900                 return;
1901         }
1902
1903         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1904                                   physical_sector_size);
1905         if (err) {
1906                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1907                                  info->xbdev->otherend);
1908                 return;
1909         }
1910
1911         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1912
1913         /* Kick pending requests. */
1914         spin_lock_irq(&info->io_lock);
1915         info->connected = BLKIF_STATE_CONNECTED;
1916         kick_pending_request_queues(info);
1917         spin_unlock_irq(&info->io_lock);
1918
1919         add_disk(info->gd);
1920
1921         info->is_ready = 1;
1922 }
1923
1924 /**
1925  * Callback received when the backend's state changes.
1926  */
1927 static void blkback_changed(struct xenbus_device *dev,
1928                             enum xenbus_state backend_state)
1929 {
1930         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1931
1932         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1933
1934         switch (backend_state) {
1935         case XenbusStateInitWait:
1936                 if (dev->state != XenbusStateInitialising)
1937                         break;
1938                 if (talk_to_blkback(dev, info)) {
1939                         kfree(info);
1940                         dev_set_drvdata(&dev->dev, NULL);
1941                         break;
1942                 }
1943         case XenbusStateInitialising:
1944         case XenbusStateInitialised:
1945         case XenbusStateReconfiguring:
1946         case XenbusStateReconfigured:
1947         case XenbusStateUnknown:
1948                 break;
1949
1950         case XenbusStateConnected:
1951                 blkfront_connect(info);
1952                 break;
1953
1954         case XenbusStateClosed:
1955                 if (dev->state == XenbusStateClosed)
1956                         break;
1957                 /* Missed the backend's Closing state -- fallthrough */
1958         case XenbusStateClosing:
1959                 blkfront_closing(info);
1960                 break;
1961         }
1962 }
1963
1964 static int blkfront_remove(struct xenbus_device *xbdev)
1965 {
1966         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1967         struct block_device *bdev = NULL;
1968         struct gendisk *disk;
1969
1970         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1971
1972         blkif_free(info, 0);
1973
1974         mutex_lock(&info->mutex);
1975
1976         disk = info->gd;
1977         if (disk)
1978                 bdev = bdget_disk(disk, 0);
1979
1980         info->xbdev = NULL;
1981         mutex_unlock(&info->mutex);
1982
1983         if (!bdev) {
1984                 kfree(info);
1985                 return 0;
1986         }
1987
1988         /*
1989          * The xbdev was removed before we reached the Closed
1990          * state. See if it's safe to remove the disk. If the bdev
1991          * isn't closed yet, we let release take care of it.
1992          */
1993
1994         mutex_lock(&bdev->bd_mutex);
1995         info = disk->private_data;
1996
1997         dev_warn(disk_to_dev(disk),
1998                  "%s was hot-unplugged, %d stale handles\n",
1999                  xbdev->nodename, bdev->bd_openers);
2000
2001         if (info && !bdev->bd_openers) {
2002                 xlvbd_release_gendisk(info);
2003                 disk->private_data = NULL;
2004                 kfree(info);
2005         }
2006
2007         mutex_unlock(&bdev->bd_mutex);
2008         bdput(bdev);
2009
2010         return 0;
2011 }
2012
2013 static int blkfront_is_ready(struct xenbus_device *dev)
2014 {
2015         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2016
2017         return info->is_ready && info->xbdev;
2018 }
2019
2020 static int blkif_open(struct block_device *bdev, fmode_t mode)
2021 {
2022         struct gendisk *disk = bdev->bd_disk;
2023         struct blkfront_info *info;
2024         int err = 0;
2025
2026         mutex_lock(&blkfront_mutex);
2027
2028         info = disk->private_data;
2029         if (!info) {
2030                 /* xbdev gone */
2031                 err = -ERESTARTSYS;
2032                 goto out;
2033         }
2034
2035         mutex_lock(&info->mutex);
2036
2037         if (!info->gd)
2038                 /* xbdev is closed */
2039                 err = -ERESTARTSYS;
2040
2041         mutex_unlock(&info->mutex);
2042
2043 out:
2044         mutex_unlock(&blkfront_mutex);
2045         return err;
2046 }
2047
2048 static void blkif_release(struct gendisk *disk, fmode_t mode)
2049 {
2050         struct blkfront_info *info = disk->private_data;
2051         struct block_device *bdev;
2052         struct xenbus_device *xbdev;
2053
2054         mutex_lock(&blkfront_mutex);
2055
2056         bdev = bdget_disk(disk, 0);
2057
2058         if (!bdev) {
2059                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2060                 goto out_mutex;
2061         }
2062         if (bdev->bd_openers)
2063                 goto out;
2064
2065         /*
2066          * Check if we have been instructed to close. We will have
2067          * deferred this request, because the bdev was still open.
2068          */
2069
2070         mutex_lock(&info->mutex);
2071         xbdev = info->xbdev;
2072
2073         if (xbdev && xbdev->state == XenbusStateClosing) {
2074                 /* pending switch to state closed */
2075                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2076                 xlvbd_release_gendisk(info);
2077                 xenbus_frontend_closed(info->xbdev);
2078         }
2079
2080         mutex_unlock(&info->mutex);
2081
2082         if (!xbdev) {
2083                 /* sudden device removal */
2084                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2085                 xlvbd_release_gendisk(info);
2086                 disk->private_data = NULL;
2087                 kfree(info);
2088         }
2089
2090 out:
2091         bdput(bdev);
2092 out_mutex:
2093         mutex_unlock(&blkfront_mutex);
2094 }
2095
2096 static const struct block_device_operations xlvbd_block_fops =
2097 {
2098         .owner = THIS_MODULE,
2099         .open = blkif_open,
2100         .release = blkif_release,
2101         .getgeo = blkif_getgeo,
2102         .ioctl = blkif_ioctl,
2103 };
2104
2105
2106 static const struct xenbus_device_id blkfront_ids[] = {
2107         { "vbd" },
2108         { "" }
2109 };
2110
2111 static struct xenbus_driver blkfront_driver = {
2112         .ids  = blkfront_ids,
2113         .probe = blkfront_probe,
2114         .remove = blkfront_remove,
2115         .resume = blkfront_resume,
2116         .otherend_changed = blkback_changed,
2117         .is_ready = blkfront_is_ready,
2118 };
2119
2120 static int __init xlblk_init(void)
2121 {
2122         int ret;
2123
2124         if (!xen_domain())
2125                 return -ENODEV;
2126
2127         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_PAGE_ORDER) {
2128                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2129                         xen_blkif_max_ring_order, XENBUS_MAX_RING_PAGE_ORDER);
2130                 xen_blkif_max_ring_order = 0;
2131         }
2132
2133         if (!xen_has_pv_disk_devices())
2134                 return -ENODEV;
2135
2136         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2137                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2138                        XENVBD_MAJOR, DEV_NAME);
2139                 return -ENODEV;
2140         }
2141
2142         ret = xenbus_register_frontend(&blkfront_driver);
2143         if (ret) {
2144                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2145                 return ret;
2146         }
2147
2148         return 0;
2149 }
2150 module_init(xlblk_init);
2151
2152
2153 static void __exit xlblk_exit(void)
2154 {
2155         xenbus_unregister_driver(&blkfront_driver);
2156         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2157         kfree(minors);
2158 }
2159 module_exit(xlblk_exit);
2160
2161 MODULE_DESCRIPTION("Xen virtual block device frontend");
2162 MODULE_LICENSE("GPL");
2163 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2164 MODULE_ALIAS("xen:vbd");
2165 MODULE_ALIAS("xenblk");