drivers/block/xen-blkfront.c

   1 /*
   2  * blkfront.c
   3  *
   4  * XenLinux virtual block device driver.
   5  *
   6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8  * Copyright (c) 2004, Christian Limpach
   9  * Copyright (c) 2004, Andrew Warfield
  10  * Copyright (c) 2005, Christopher Clark
  11  * Copyright (c) 2005, XenSource Ltd
  12  *
  13  * This program is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU General Public License version 2
  15  * as published by the Free Software Foundation; or, when distributed
  16  * separately from the Linux kernel or incorporated into other
  17  * software packages, subject to the following license:
  18  *
  19  * Permission is hereby granted, free of charge, to any person obtaining a copy
  20  * of this source file (the "Software"), to deal in the Software without
  21  * restriction, including without limitation the rights to use, copy, modify,
  22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23  * and to permit persons to whom the Software is furnished to do so, subject to
  24  * the following conditions:
  25  *
  26  * The above copyright notice and this permission notice shall be included in
  27  * all copies or substantial portions of the Software.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35  * IN THE SOFTWARE.
  36  */
  37
  38 #include <linux/interrupt.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/cdrom.h>
  42 #include <linux/module.h>
  43 #include <linux/slab.h>
  44 #include <linux/mutex.h>
  45 #include <linux/scatterlist.h>
  46 #include <linux/bitmap.h>
  47 #include <linux/list.h>
  48
  49 #include <xen/xen.h>
  50 #include <xen/xenbus.h>
  51 #include <xen/grant_table.h>
  52 #include <xen/events.h>
  53 #include <xen/page.h>
  54 #include <xen/platform_pci.h>
  55
  56 #include <xen/interface/grant_table.h>
  57 #include <xen/interface/io/blkif.h>
  58 #include <xen/interface/io/protocols.h>
  59
  60 #include <asm/xen/hypervisor.h>
  61
  62 enum blkif_state {
  63         BLKIF_STATE_DISCONNECTED,
  64         BLKIF_STATE_CONNECTED,
  65         BLKIF_STATE_SUSPENDED,
  66 };
  67
  68 struct grant {
  69         grant_ref_t gref;
  70         unsigned long pfn;
  71         struct list_head node;
  72 };
  73
  74 struct blk_shadow {
  75         struct blkif_request req;
  76         struct request *request;
  77         struct grant **grants_used;
  78         struct grant **indirect_grants;
  79         struct scatterlist *sg;
  80 };
  81
  82 struct split_bio {
  83         struct bio *bio;
  84         atomic_t pending;
  85 };
  86
  87 static DEFINE_MUTEX(blkfront_mutex);
  88 static const struct block_device_operations xlvbd_block_fops;
  89
  90 /*
  91  * Maximum number of segments in indirect requests, the actual value used by
  92  * the frontend driver is the minimum of this value and the value provided
  93  * by the backend driver.
  94  */
  95
  96 static unsigned int xen_blkif_max_segments = 32;
  97 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
  98 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
  99
 100 /*
 101  * Maximum order of pages to be used for the shared ring between front and
 102  * backend, 4KB page granularity is used.
 103  */
 104 static unsigned int xen_blkif_max_ring_order;
 105 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
 106 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 107
 108 #define BLK_RING_SIZE(info) __CONST_RING_SIZE(blkif, PAGE_SIZE * (info)->nr_ring_pages)
 109 #define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * XENBUS_MAX_RING_PAGES)
 110 /*
 111  * ring-ref%i i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 112  * characters are enough. Define to 20 to keep consist with backend.
 113  */
 114 #define RINGREF_NAME_LEN (20)
 115
 116 /*
 117  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 118  * hang in private_data off the gendisk structure. We may end up
 119  * putting all kinds of interesting stuff here :-)
 120  */
 121 struct blkfront_info
 122 {
 123         spinlock_t io_lock;
 124         struct mutex mutex;
 125         struct xenbus_device *xbdev;
 126         struct gendisk *gd;
 127         int vdevice;
 128         blkif_vdev_t handle;
 129         enum blkif_state connected;
 130         int ring_ref[XENBUS_MAX_RING_PAGES];
 131         unsigned int nr_ring_pages;
 132         struct blkif_front_ring ring;
 133         unsigned int evtchn, irq;
 134         struct request_queue *rq;
 135         struct work_struct work;
 136         struct gnttab_free_callback callback;
 137         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 138         struct list_head grants;
 139         struct list_head indirect_pages;
 140         unsigned int persistent_gnts_c;
 141         unsigned long shadow_free;
 142         unsigned int feature_flush;
 143         unsigned int feature_discard:1;
 144         unsigned int feature_secdiscard:1;
 145         unsigned int discard_granularity;
 146         unsigned int discard_alignment;
 147         unsigned int feature_persistent:1;
 148         unsigned int max_indirect_segments;
 149         int is_ready;
 150 };
 151
 152 static unsigned int nr_minors;
 153 static unsigned long *minors;
 154 static DEFINE_SPINLOCK(minor_lock);
 155
 156 #define GRANT_INVALID_REF       0
 157
 158 #define PARTS_PER_DISK          16
 159 #define PARTS_PER_EXT_DISK      256
 160
 161 #define BLKIF_MAJOR(dev) ((dev)>>8)
 162 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 163
 164 #define EXT_SHIFT 28
 165 #define EXTENDED (1<<EXT_SHIFT)
 166 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 167 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 168 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 169 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 170 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 171 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 172
 173 #define DEV_NAME        "xvd"   /* name in /dev */
 174
 175 #define SEGS_PER_INDIRECT_FRAME \
 176         (PAGE_SIZE/sizeof(struct blkif_request_segment))
 177 #define INDIRECT_GREFS(_segs) \
 178         ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
 179
 180 static int blkfront_setup_indirect(struct blkfront_info *info);
 181 static int blkfront_gather_backend_features(struct blkfront_info *info);
 182
 183 static int get_id_from_freelist(struct blkfront_info *info)
 184 {
 185         unsigned long free = info->shadow_free;
 186         BUG_ON(free >= BLK_RING_SIZE(info));
 187         info->shadow_free = info->shadow[free].req.u.rw.id;
 188         info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 189         return free;
 190 }
 191
 192 static int add_id_to_freelist(struct blkfront_info *info,
 193                                unsigned long id)
 194 {
 195         if (info->shadow[id].req.u.rw.id != id)
 196                 return -EINVAL;
 197         if (info->shadow[id].request == NULL)
 198                 return -EINVAL;
 199         info->shadow[id].req.u.rw.id  = info->shadow_free;
 200         info->shadow[id].request = NULL;
 201         info->shadow_free = id;
 202         return 0;
 203 }
 204
 205 static int fill_grant_buffer(struct blkfront_info *info, int num)
 206 {
 207         struct page *granted_page;
 208         struct grant *gnt_list_entry, *n;
 209         int i = 0;
 210
 211         while(i < num) {
 212                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 213                 if (!gnt_list_entry)
 214                         goto out_of_memory;
 215
 216                 if (info->feature_persistent) {
 217                         granted_page = alloc_page(GFP_NOIO);
 218                         if (!granted_page) {
 219                                 kfree(gnt_list_entry);
 220                                 goto out_of_memory;
 221                         }
 222                         gnt_list_entry->pfn = page_to_pfn(granted_page);
 223                 }
 224
 225                 gnt_list_entry->gref = GRANT_INVALID_REF;
 226                 list_add(&gnt_list_entry->node, &info->grants);
 227                 i++;
 228         }
 229
 230         return 0;
 231
 232 out_of_memory:
 233         list_for_each_entry_safe(gnt_list_entry, n,
 234                                  &info->grants, node) {
 235                 list_del(&gnt_list_entry->node);
 236                 if (info->feature_persistent)
 237                         __free_page(pfn_to_page(gnt_list_entry->pfn));
 238                 kfree(gnt_list_entry);
 239                 i--;
 240         }
 241         BUG_ON(i != 0);
 242         return -ENOMEM;
 243 }
 244
 245 static struct grant *get_grant(grant_ref_t *gref_head,
 246                                unsigned long pfn,
 247                                struct blkfront_info *info)
 248 {
 249         struct grant *gnt_list_entry;
 250         unsigned long buffer_mfn;
 251
 252         BUG_ON(list_empty(&info->grants));
 253         gnt_list_entry = list_first_entry(&info->grants, struct grant,
 254                                           node);
 255         list_del(&gnt_list_entry->node);
 256
 257         if (gnt_list_entry->gref != GRANT_INVALID_REF) {
 258                 info->persistent_gnts_c--;
 259                 return gnt_list_entry;
 260         }
 261
 262         /* Assign a gref to this page */
 263         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 264         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 265         if (!info->feature_persistent) {
 266                 BUG_ON(!pfn);
 267                 gnt_list_entry->pfn = pfn;
 268         }
 269         buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
 270         gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 271                                         info->xbdev->otherend_id,
 272                                         buffer_mfn, 0);
 273         return gnt_list_entry;
 274 }
 275
 276 static const char *op_name(int op)
 277 {
 278         static const char *const names[] = {
 279                 [BLKIF_OP_READ] = "read",
 280                 [BLKIF_OP_WRITE] = "write",
 281                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 282                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 283                 [BLKIF_OP_DISCARD] = "discard" };
 284
 285         if (op < 0 || op >= ARRAY_SIZE(names))
 286                 return "unknown";
 287
 288         if (!names[op])
 289                 return "reserved";
 290
 291         return names[op];
 292 }
 293 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 294 {
 295         unsigned int end = minor + nr;
 296         int rc;
 297
 298         if (end > nr_minors) {
 299                 unsigned long *bitmap, *old;
 300
 301                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 302                                  GFP_KERNEL);
 303                 if (bitmap == NULL)
 304                         return -ENOMEM;
 305
 306                 spin_lock(&minor_lock);
 307                 if (end > nr_minors) {
 308                         old = minors;
 309                         memcpy(bitmap, minors,
 310                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 311                         minors = bitmap;
 312                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 313                 } else
 314                         old = bitmap;
 315                 spin_unlock(&minor_lock);
 316                 kfree(old);
 317         }
 318
 319         spin_lock(&minor_lock);
 320         if (find_next_bit(minors, end, minor) >= end) {
 321                 bitmap_set(minors, minor, nr);
 322                 rc = 0;
 323         } else
 324                 rc = -EBUSY;
 325         spin_unlock(&minor_lock);
 326
 327         return rc;
 328 }
 329
 330 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 331 {
 332         unsigned int end = minor + nr;
 333
 334         BUG_ON(end > nr_minors);
 335         spin_lock(&minor_lock);
 336         bitmap_clear(minors,  minor, nr);
 337         spin_unlock(&minor_lock);
 338 }
 339
 340 static void blkif_restart_queue_callback(void *arg)
 341 {
 342         struct blkfront_info *info = (struct blkfront_info *)arg;
 343         schedule_work(&info->work);
 344 }
 345
 346 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 347 {
 348         /* We don't have real geometry info, but let's at least return
 349            values consistent with the size of the device */
 350         sector_t nsect = get_capacity(bd->bd_disk);
 351         sector_t cylinders = nsect;
 352
 353         hg->heads = 0xff;
 354         hg->sectors = 0x3f;
 355         sector_div(cylinders, hg->heads * hg->sectors);
 356         hg->cylinders = cylinders;
 357         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 358                 hg->cylinders = 0xffff;
 359         return 0;
 360 }
 361
 362 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 363                        unsigned command, unsigned long argument)
 364 {
 365         struct blkfront_info *info = bdev->bd_disk->private_data;
 366         int i;
 367
 368         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 369                 command, (long)argument);
 370
 371         switch (command) {
 372         case CDROMMULTISESSION:
 373                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 374                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 375                         if (put_user(0, (char __user *)(argument + i)))
 376                                 return -EFAULT;
 377                 return 0;
 378
 379         case CDROM_GET_CAPABILITY: {
 380                 struct gendisk *gd = info->gd;
 381                 if (gd->flags & GENHD_FL_CD)
 382                         return 0;
 383                 return -EINVAL;
 384         }
 385
 386         default:
 387                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 388                   command);*/
 389                 return -EINVAL; /* same return as native Linux */
 390         }
 391
 392         return 0;
 393 }
 394
 395 /*
 396  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 397  * and writes are handled as expected.
 398  *
 399  * @req: a request struct
 400  */
 401 static int blkif_queue_request(struct request *req)
 402 {
 403         struct blkfront_info *info = req->rq_disk->private_data;
 404         struct blkif_request *ring_req;
 405         unsigned long id;
 406         unsigned int fsect, lsect;
 407         int i, ref, n;
 408         struct blkif_request_segment *segments = NULL;
 409
 410         /*
 411          * Used to store if we are able to queue the request by just using
 412          * existing persistent grants, or if we have to get new grants,
 413          * as there are not sufficiently many free.
 414          */
 415         bool new_persistent_gnts;
 416         grant_ref_t gref_head;
 417         struct grant *gnt_list_entry = NULL;
 418         struct scatterlist *sg;
 419         int nseg, max_grefs;
 420
 421         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 422                 return 1;
 423
 424         max_grefs = req->nr_phys_segments;
 425         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 426                 /*
 427                  * If we are using indirect segments we need to account
 428                  * for the indirect grefs used in the request.
 429                  */
 430                 max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
 431
 432         /* Check if we have enough grants to allocate a requests */
 433         if (info->persistent_gnts_c < max_grefs) {
 434                 new_persistent_gnts = 1;
 435                 if (gnttab_alloc_grant_references(
 436                     max_grefs - info->persistent_gnts_c,
 437                     &gref_head) < 0) {
 438                         gnttab_request_free_callback(
 439                                 &info->callback,
 440                                 blkif_restart_queue_callback,
 441                                 info,
 442                                 max_grefs);
 443                         return 1;
 444                 }
 445         } else
 446                 new_persistent_gnts = 0;
 447
 448         /* Fill out a communications ring structure. */
 449         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 450         id = get_id_from_freelist(info);
 451         info->shadow[id].request = req;
 452
 453         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 454                 ring_req->operation = BLKIF_OP_DISCARD;
 455                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 456                 ring_req->u.discard.id = id;
 457                 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 458                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 459                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 460                 else
 461                         ring_req->u.discard.flag = 0;
 462         } else {
 463                 BUG_ON(info->max_indirect_segments == 0 &&
 464                        req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 465                 BUG_ON(info->max_indirect_segments &&
 466                        req->nr_phys_segments > info->max_indirect_segments);
 467                 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
 468                 ring_req->u.rw.id = id;
 469                 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 470                         /*
 471                          * The indirect operation can only be a BLKIF_OP_READ or
 472                          * BLKIF_OP_WRITE
 473                          */
 474                         BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
 475                         ring_req->operation = BLKIF_OP_INDIRECT;
 476                         ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 477                                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 478                         ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 479                         ring_req->u.indirect.handle = info->handle;
 480                         ring_req->u.indirect.nr_segments = nseg;
 481                 } else {
 482                         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 483                         ring_req->u.rw.handle = info->handle;
 484                         ring_req->operation = rq_data_dir(req) ?
 485                                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 486                         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 487                                 /*
 488                                  * Ideally we can do an unordered flush-to-disk. In case the
 489                                  * backend onlysupports barriers, use that. A barrier request
 490                                  * a superset of FUA, so we can implement it the same
 491                                  * way.  (It's also a FLUSH+FUA, since it is
 492                                  * guaranteed ordered WRT previous writes.)
 493                                  */
 494                                 switch (info->feature_flush &
 495                                         ((REQ_FLUSH|REQ_FUA))) {
 496                                 case REQ_FLUSH|REQ_FUA:
 497                                         ring_req->operation =
 498                                                 BLKIF_OP_WRITE_BARRIER;
 499                                         break;
 500                                 case REQ_FLUSH:
 501                                         ring_req->operation =
 502                                                 BLKIF_OP_FLUSH_DISKCACHE;
 503                                         break;
 504                                 default:
 505                                         ring_req->operation = 0;
 506                                 }
 507                         }
 508                         ring_req->u.rw.nr_segments = nseg;
 509                 }
 510                 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
 511                         fsect = sg->offset >> 9;
 512                         lsect = fsect + (sg->length >> 9) - 1;
 513
 514                         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 515                             (i % SEGS_PER_INDIRECT_FRAME == 0)) {
 516                                 unsigned long uninitialized_var(pfn);
 517
 518                                 if (segments)
 519                                         kunmap_atomic(segments);
 520
 521                                 n = i / SEGS_PER_INDIRECT_FRAME;
 522                                 if (!info->feature_persistent) {
 523                                         struct page *indirect_page;
 524
 525                                         /* Fetch a pre-allocated page to use for indirect grefs */
 526                                         BUG_ON(list_empty(&info->indirect_pages));
 527                                         indirect_page = list_first_entry(&info->indirect_pages,
 528                                                                          struct page, lru);
 529                                         list_del(&indirect_page->lru);
 530                                         pfn = page_to_pfn(indirect_page);
 531                                 }
 532                                 gnt_list_entry = get_grant(&gref_head, pfn, info);
 533                                 info->shadow[id].indirect_grants[n] = gnt_list_entry;
 534                                 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
 535                                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 536                         }
 537
 538                         gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
 539                         ref = gnt_list_entry->gref;
 540
 541                         info->shadow[id].grants_used[i] = gnt_list_entry;
 542
 543                         if (rq_data_dir(req) && info->feature_persistent) {
 544                                 char *bvec_data;
 545                                 void *shared_data;
 546
 547                                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 548
 549                                 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
 550                                 bvec_data = kmap_atomic(sg_page(sg));
 551
 552                                 /*
 553                                  * this does not wipe data stored outside the
 554                                  * range sg->offset..sg->offset+sg->length.
 555                                  * Therefore, blkback *could* see data from
 556                                  * previous requests. This is OK as long as
 557                                  * persistent grants are shared with just one
 558                                  * domain. It may need refactoring if this
 559                                  * changes
 560                                  */
 561                                 memcpy(shared_data + sg->offset,
 562                                        bvec_data   + sg->offset,
 563                                        sg->length);
 564
 565                                 kunmap_atomic(bvec_data);
 566                                 kunmap_atomic(shared_data);
 567                         }
 568                         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 569                                 ring_req->u.rw.seg[i] =
 570                                                 (struct blkif_request_segment) {
 571                                                         .gref       = ref,
 572                                                         .first_sect = fsect,
 573                                                         .last_sect  = lsect };
 574                         } else {
 575                                 n = i % SEGS_PER_INDIRECT_FRAME;
 576                                 segments[n] =
 577                                         (struct blkif_request_segment) {
 578                                                         .gref       = ref,
 579                                                         .first_sect = fsect,
 580                                                         .last_sect  = lsect };
 581                         }
 582                 }
 583                 if (segments)
 584                         kunmap_atomic(segments);
 585         }
 586
 587         info->ring.req_prod_pvt++;
 588
 589         /* Keep a private copy so we can reissue requests when recovering. */
 590         info->shadow[id].req = *ring_req;
 591
 592         if (new_persistent_gnts)
 593                 gnttab_free_grant_references(gref_head);
 594
 595         return 0;
 596 }
 597
 598
 599 static inline void flush_requests(struct blkfront_info *info)
 600 {
 601         int notify;
 602
 603         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 604
 605         if (notify)
 606                 notify_remote_via_irq(info->irq);
 607 }
 608
 609 static inline bool blkif_request_flush_invalid(struct request *req,
 610                                                struct blkfront_info *info)
 611 {
 612         return ((req->cmd_type != REQ_TYPE_FS) ||
 613                 ((req->cmd_flags & REQ_FLUSH) &&
 614                  !(info->feature_flush & REQ_FLUSH)) ||
 615                 ((req->cmd_flags & REQ_FUA) &&
 616                  !(info->feature_flush & REQ_FUA)));
 617 }
 618
 619 /*
 620  * do_blkif_request
 621  *  read a block; request is in a request queue
 622  */
 623 static void do_blkif_request(struct request_queue *rq)
 624 {
 625         struct blkfront_info *info = NULL;
 626         struct request *req;
 627         int queued;
 628
 629         pr_debug("Entered do_blkif_request\n");
 630
 631         queued = 0;
 632
 633         while ((req = blk_peek_request(rq)) != NULL) {
 634                 info = req->rq_disk->private_data;
 635
 636                 if (RING_FULL(&info->ring))
 637                         goto wait;
 638
 639                 blk_start_request(req);
 640
 641                 if (blkif_request_flush_invalid(req, info)) {
 642                         __blk_end_request_all(req, -EOPNOTSUPP);
 643                         continue;
 644                 }
 645
 646                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 647                          "(%u/%u) [%s]\n",
 648                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 649                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 650                          rq_data_dir(req) ? "write" : "read");
 651
 652                 if (blkif_queue_request(req)) {
 653                         blk_requeue_request(rq, req);
 654 wait:
 655                         /* Avoid pointless unplugs. */
 656                         blk_stop_queue(rq);
 657                         break;
 658                 }
 659
 660                 queued++;
 661         }
 662
 663         if (queued != 0)
 664                 flush_requests(info);
 665 }
 666
 667 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 668                                 unsigned int physical_sector_size,
 669                                 unsigned int segments)
 670 {
 671         struct request_queue *rq;
 672         struct blkfront_info *info = gd->private_data;
 673
 674         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 675         if (rq == NULL)
 676                 return -1;
 677
 678         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 679
 680         if (info->feature_discard) {
 681                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 682                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 683                 rq->limits.discard_granularity = info->discard_granularity;
 684                 rq->limits.discard_alignment = info->discard_alignment;
 685                 if (info->feature_secdiscard)
 686                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 687         }
 688
 689         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 690         blk_queue_logical_block_size(rq, sector_size);
 691         blk_queue_physical_block_size(rq, physical_sector_size);
 692         blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
 693
 694         /* Each segment in a request is up to an aligned page in size. */
 695         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 696         blk_queue_max_segment_size(rq, PAGE_SIZE);
 697
 698         /* Ensure a merged request will fit in a single I/O ring slot. */
 699         blk_queue_max_segments(rq, segments);
 700
 701         /* Make sure buffer addresses are sector-aligned. */
 702         blk_queue_dma_alignment(rq, 511);
 703
 704         /* Make sure we don't use bounce buffers. */
 705         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 706
 707         gd->queue = rq;
 708
 709         return 0;
 710 }
 711
 712 static const char *flush_info(unsigned int feature_flush)
 713 {
 714         switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
 715         case REQ_FLUSH|REQ_FUA:
 716                 return "barrier: enabled;";
 717         case REQ_FLUSH:
 718                 return "flush diskcache: enabled;";
 719         default:
 720                 return "barrier or flush: disabled;";
 721         }
 722 }
 723
 724 static void xlvbd_flush(struct blkfront_info *info)
 725 {
 726         blk_queue_flush(info->rq, info->feature_flush);
 727         pr_info("blkfront: %s: %s %s %s %s %s\n",
 728                 info->gd->disk_name, flush_info(info->feature_flush),
 729                 "persistent grants:", info->feature_persistent ?
 730                 "enabled;" : "disabled;", "indirect descriptors:",
 731                 info->max_indirect_segments ? "enabled;" : "disabled;");
 732 }
 733
 734 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 735 {
 736         int major;
 737         major = BLKIF_MAJOR(vdevice);
 738         *minor = BLKIF_MINOR(vdevice);
 739         switch (major) {
 740                 case XEN_IDE0_MAJOR:
 741                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 742                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 743                                 EMULATED_HD_DISK_MINOR_OFFSET;
 744                         break;
 745                 case XEN_IDE1_MAJOR:
 746                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 747                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 748                                 EMULATED_HD_DISK_MINOR_OFFSET;
 749                         break;
 750                 case XEN_SCSI_DISK0_MAJOR:
 751                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 752                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 753                         break;
 754                 case XEN_SCSI_DISK1_MAJOR:
 755                 case XEN_SCSI_DISK2_MAJOR:
 756                 case XEN_SCSI_DISK3_MAJOR:
 757                 case XEN_SCSI_DISK4_MAJOR:
 758                 case XEN_SCSI_DISK5_MAJOR:
 759                 case XEN_SCSI_DISK6_MAJOR:
 760                 case XEN_SCSI_DISK7_MAJOR:
 761                         *offset = (*minor / PARTS_PER_DISK) +
 762                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 763                                 EMULATED_SD_DISK_NAME_OFFSET;
 764                         *minor = *minor +
 765                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 766                                 EMULATED_SD_DISK_MINOR_OFFSET;
 767                         break;
 768                 case XEN_SCSI_DISK8_MAJOR:
 769                 case XEN_SCSI_DISK9_MAJOR:
 770                 case XEN_SCSI_DISK10_MAJOR:
 771                 case XEN_SCSI_DISK11_MAJOR:
 772                 case XEN_SCSI_DISK12_MAJOR:
 773                 case XEN_SCSI_DISK13_MAJOR:
 774                 case XEN_SCSI_DISK14_MAJOR:
 775                 case XEN_SCSI_DISK15_MAJOR:
 776                         *offset = (*minor / PARTS_PER_DISK) +
 777                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 778                                 EMULATED_SD_DISK_NAME_OFFSET;
 779                         *minor = *minor +
 780                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 781                                 EMULATED_SD_DISK_MINOR_OFFSET;
 782                         break;
 783                 case XENVBD_MAJOR:
 784                         *offset = *minor / PARTS_PER_DISK;
 785                         break;
 786                 default:
 787                         printk(KERN_WARNING "blkfront: your disk configuration is "
 788                                         "incorrect, please use an xvd device instead\n");
 789                         return -ENODEV;
 790         }
 791         return 0;
 792 }
 793
 794 static char *encode_disk_name(char *ptr, unsigned int n)
 795 {
 796         if (n >= 26)
 797                 ptr = encode_disk_name(ptr, n / 26 - 1);
 798         *ptr = 'a' + n % 26;
 799         return ptr + 1;
 800 }
 801
 802 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 803                                struct blkfront_info *info,
 804                                u16 vdisk_info, u16 sector_size,
 805                                unsigned int physical_sector_size)
 806 {
 807         struct gendisk *gd;
 808         int nr_minors = 1;
 809         int err;
 810         unsigned int offset;
 811         int minor;
 812         int nr_parts;
 813         char *ptr;
 814
 815         BUG_ON(info->gd != NULL);
 816         BUG_ON(info->rq != NULL);
 817
 818         if ((info->vdevice>>EXT_SHIFT) > 1) {
 819                 /* this is above the extended range; something is wrong */
 820                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 821                 return -ENODEV;
 822         }
 823
 824         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 825                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 826                 if (err)
 827                         return err;
 828                 nr_parts = PARTS_PER_DISK;
 829         } else {
 830                 minor = BLKIF_MINOR_EXT(info->vdevice);
 831                 nr_parts = PARTS_PER_EXT_DISK;
 832                 offset = minor / nr_parts;
 833                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 834                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 835                                         "emulated IDE disks,\n\t choose an xvd device name"
 836                                         "from xvde on\n", info->vdevice);
 837         }
 838         if (minor >> MINORBITS) {
 839                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 840                         info->vdevice, minor);
 841                 return -ENODEV;
 842         }
 843
 844         if ((minor % nr_parts) == 0)
 845                 nr_minors = nr_parts;
 846
 847         err = xlbd_reserve_minors(minor, nr_minors);
 848         if (err)
 849                 goto out;
 850         err = -ENODEV;
 851
 852         gd = alloc_disk(nr_minors);
 853         if (gd == NULL)
 854                 goto release;
 855
 856         strcpy(gd->disk_name, DEV_NAME);
 857         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 858         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 859         if (nr_minors > 1)
 860                 *ptr = 0;
 861         else
 862                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 863                          "%d", minor & (nr_parts - 1));
 864
 865         gd->major = XENVBD_MAJOR;
 866         gd->first_minor = minor;
 867         gd->fops = &xlvbd_block_fops;
 868         gd->private_data = info;
 869         gd->driverfs_dev = &(info->xbdev->dev);
 870         set_capacity(gd, capacity);
 871
 872         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
 873                                  info->max_indirect_segments ? :
 874                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 875                 del_gendisk(gd);
 876                 goto release;
 877         }
 878
 879         info->rq = gd->queue;
 880         info->gd = gd;
 881
 882         xlvbd_flush(info);
 883
 884         if (vdisk_info & VDISK_READONLY)
 885                 set_disk_ro(gd, 1);
 886
 887         if (vdisk_info & VDISK_REMOVABLE)
 888                 gd->flags |= GENHD_FL_REMOVABLE;
 889
 890         if (vdisk_info & VDISK_CDROM)
 891                 gd->flags |= GENHD_FL_CD;
 892
 893         return 0;
 894
 895  release:
 896         xlbd_release_minors(minor, nr_minors);
 897  out:
 898         return err;
 899 }
 900
 901 static void xlvbd_release_gendisk(struct blkfront_info *info)
 902 {
 903         unsigned int minor, nr_minors;
 904         unsigned long flags;
 905
 906         if (info->rq == NULL)
 907                 return;
 908
 909         spin_lock_irqsave(&info->io_lock, flags);
 910
 911         /* No more blkif_request(). */
 912         blk_stop_queue(info->rq);
 913
 914         /* No more gnttab callback work. */
 915         gnttab_cancel_free_callback(&info->callback);
 916         spin_unlock_irqrestore(&info->io_lock, flags);
 917
 918         /* Flush gnttab callback work. Must be done with no locks held. */
 919         flush_work(&info->work);
 920
 921         del_gendisk(info->gd);
 922
 923         minor = info->gd->first_minor;
 924         nr_minors = info->gd->minors;
 925         xlbd_release_minors(minor, nr_minors);
 926
 927         blk_cleanup_queue(info->rq);
 928         info->rq = NULL;
 929
 930         put_disk(info->gd);
 931         info->gd = NULL;
 932 }
 933
 934 static void kick_pending_request_queues(struct blkfront_info *info)
 935 {
 936         if (!RING_FULL(&info->ring)) {
 937                 /* Re-enable calldowns. */
 938                 blk_start_queue(info->rq);
 939                 /* Kick things off immediately. */
 940                 do_blkif_request(info->rq);
 941         }
 942 }
 943
 944 static void blkif_restart_queue(struct work_struct *work)
 945 {
 946         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 947
 948         spin_lock_irq(&info->io_lock);
 949         if (info->connected == BLKIF_STATE_CONNECTED)
 950                 kick_pending_request_queues(info);
 951         spin_unlock_irq(&info->io_lock);
 952 }
 953
 954 static void blkif_free(struct blkfront_info *info, int suspend)
 955 {
 956         struct grant *persistent_gnt;
 957         struct grant *n;
 958         int i, j, segs;
 959
 960         /* Prevent new requests being issued until we fix things up. */
 961         spin_lock_irq(&info->io_lock);
 962         info->connected = suspend ?
 963                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 964         /* No more blkif_request(). */
 965         if (info->rq)
 966                 blk_stop_queue(info->rq);
 967
 968         /* Remove all persistent grants */
 969         if (!list_empty(&info->grants)) {
 970                 list_for_each_entry_safe(persistent_gnt, n,
 971                                          &info->grants, node) {
 972                         list_del(&persistent_gnt->node);
 973                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
 974                                 gnttab_end_foreign_access(persistent_gnt->gref,
 975                                                           0, 0UL);
 976                                 info->persistent_gnts_c--;
 977                         }
 978                         if (info->feature_persistent)
 979                                 __free_page(pfn_to_page(persistent_gnt->pfn));
 980                         kfree(persistent_gnt);
 981                 }
 982         }
 983         BUG_ON(info->persistent_gnts_c != 0);
 984
 985         /*
 986          * Remove indirect pages, this only happens when using indirect
 987          * descriptors but not persistent grants
 988          */
 989         if (!list_empty(&info->indirect_pages)) {
 990                 struct page *indirect_page, *n;
 991
 992                 BUG_ON(info->feature_persistent);
 993                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
 994                         list_del(&indirect_page->lru);
 995                         __free_page(indirect_page);
 996                 }
 997         }
 998
 999         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1000                 /*
1001                  * Clear persistent grants present in requests already
1002                  * on the shared ring
1003                  */
1004                 if (!info->shadow[i].request)
1005                         goto free_shadow;
1006
1007                 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1008                        info->shadow[i].req.u.indirect.nr_segments :
1009                        info->shadow[i].req.u.rw.nr_segments;
1010                 for (j = 0; j < segs; j++) {
1011                         persistent_gnt = info->shadow[i].grants_used[j];
1012                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1013                         if (info->feature_persistent)
1014                                 __free_page(pfn_to_page(persistent_gnt->pfn));
1015                         kfree(persistent_gnt);
1016                 }
1017
1018                 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1019                         /*
1020                          * If this is not an indirect operation don't try to
1021                          * free indirect segments
1022                          */
1023                         goto free_shadow;
1024
1025                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1026                         persistent_gnt = info->shadow[i].indirect_grants[j];
1027                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1028                         __free_page(pfn_to_page(persistent_gnt->pfn));
1029                         kfree(persistent_gnt);
1030                 }
1031
1032 free_shadow:
1033                 kfree(info->shadow[i].grants_used);
1034                 info->shadow[i].grants_used = NULL;
1035                 kfree(info->shadow[i].indirect_grants);
1036                 info->shadow[i].indirect_grants = NULL;
1037                 kfree(info->shadow[i].sg);
1038                 info->shadow[i].sg = NULL;
1039         }
1040
1041         /* No more gnttab callback work. */
1042         gnttab_cancel_free_callback(&info->callback);
1043         spin_unlock_irq(&info->io_lock);
1044
1045         /* Flush gnttab callback work. Must be done with no locks held. */
1046         flush_work(&info->work);
1047
1048         /* Free resources associated with old device channel. */
1049         for (i = 0; i < info->nr_ring_pages; i++) {
1050                 if (info->ring_ref[i] != GRANT_INVALID_REF) {
1051                         gnttab_end_foreign_access(info->ring_ref[i], 0, 0);
1052                         info->ring_ref[i] = GRANT_INVALID_REF;
1053                 }
1054         }
1055         free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1056         info->ring.sring = NULL;
1057
1058         if (info->irq)
1059                 unbind_from_irqhandler(info->irq, info);
1060         info->evtchn = info->irq = 0;
1061
1062 }
1063
1064 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1065                              struct blkif_response *bret)
1066 {
1067         int i = 0;
1068         struct scatterlist *sg;
1069         char *bvec_data;
1070         void *shared_data;
1071         int nseg;
1072
1073         nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1074                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1075
1076         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1077                 for_each_sg(s->sg, sg, nseg, i) {
1078                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1079                         shared_data = kmap_atomic(
1080                                 pfn_to_page(s->grants_used[i]->pfn));
1081                         bvec_data = kmap_atomic(sg_page(sg));
1082                         memcpy(bvec_data   + sg->offset,
1083                                shared_data + sg->offset,
1084                                sg->length);
1085                         kunmap_atomic(bvec_data);
1086                         kunmap_atomic(shared_data);
1087                 }
1088         }
1089         /* Add the persistent grant into the list of free grants */
1090         for (i = 0; i < nseg; i++) {
1091                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1092                         /*
1093                          * If the grant is still mapped by the backend (the
1094                          * backend has chosen to make this grant persistent)
1095                          * we add it at the head of the list, so it will be
1096                          * reused first.
1097                          */
1098                         if (!info->feature_persistent)
1099                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1100                                                      s->grants_used[i]->gref);
1101                         list_add(&s->grants_used[i]->node, &info->grants);
1102                         info->persistent_gnts_c++;
1103                 } else {
1104                         /*
1105                          * If the grant is not mapped by the backend we end the
1106                          * foreign access and add it to the tail of the list,
1107                          * so it will not be picked again unless we run out of
1108                          * persistent grants.
1109                          */
1110                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1111                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1112                         list_add_tail(&s->grants_used[i]->node, &info->grants);
1113                 }
1114         }
1115         if (s->req.operation == BLKIF_OP_INDIRECT) {
1116                 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1117                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1118                                 if (!info->feature_persistent)
1119                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1120                                                              s->indirect_grants[i]->gref);
1121                                 list_add(&s->indirect_grants[i]->node, &info->grants);
1122                                 info->persistent_gnts_c++;
1123                         } else {
1124                                 struct page *indirect_page;
1125
1126                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1127                                 /*
1128                                  * Add the used indirect page back to the list of
1129                                  * available pages for indirect grefs.
1130                                  */
1131                                 if (!info->feature_persistent) {
1132                                         indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1133                                         list_add(&indirect_page->lru, &info->indirect_pages);
1134                                 }
1135                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1136                                 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1137                         }
1138                 }
1139         }
1140 }
1141
1142 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1143 {
1144         struct request *req;
1145         struct blkif_response *bret;
1146         RING_IDX i, rp;
1147         unsigned long flags;
1148         struct blkfront_info *info = (struct blkfront_info *)dev_id;
1149         int error;
1150
1151         spin_lock_irqsave(&info->io_lock, flags);
1152
1153         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1154                 spin_unlock_irqrestore(&info->io_lock, flags);
1155                 return IRQ_HANDLED;
1156         }
1157
1158  again:
1159         rp = info->ring.sring->rsp_prod;
1160         rmb(); /* Ensure we see queued responses up to 'rp'. */
1161
1162         for (i = info->ring.rsp_cons; i != rp; i++) {
1163                 unsigned long id;
1164
1165                 bret = RING_GET_RESPONSE(&info->ring, i);
1166                 id   = bret->id;
1167                 /*
1168                  * The backend has messed up and given us an id that we would
1169                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1170                  * look in get_id_from_freelist.
1171                  */
1172                 if (id >= BLK_RING_SIZE(info)) {
1173                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1174                              info->gd->disk_name, op_name(bret->operation), id);
1175                         /* We can't safely get the 'struct request' as
1176                          * the id is busted. */
1177                         continue;
1178                 }
1179                 req  = info->shadow[id].request;
1180
1181                 if (bret->operation != BLKIF_OP_DISCARD)
1182                         blkif_completion(&info->shadow[id], info, bret);
1183
1184                 if (add_id_to_freelist(info, id)) {
1185                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1186                              info->gd->disk_name, op_name(bret->operation), id);
1187                         continue;
1188                 }
1189
1190                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1191                 switch (bret->operation) {
1192                 case BLKIF_OP_DISCARD:
1193                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1194                                 struct request_queue *rq = info->rq;
1195                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1196                                            info->gd->disk_name, op_name(bret->operation));
1197                                 error = -EOPNOTSUPP;
1198                                 info->feature_discard = 0;
1199                                 info->feature_secdiscard = 0;
1200                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1201                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1202                         }
1203                         __blk_end_request_all(req, error);
1204                         break;
1205                 case BLKIF_OP_FLUSH_DISKCACHE:
1206                 case BLKIF_OP_WRITE_BARRIER:
1207                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1208                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1209                                        info->gd->disk_name, op_name(bret->operation));
1210                                 error = -EOPNOTSUPP;
1211                         }
1212                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1213                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
1214                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1215                                        info->gd->disk_name, op_name(bret->operation));
1216                                 error = -EOPNOTSUPP;
1217                         }
1218                         if (unlikely(error)) {
1219                                 if (error == -EOPNOTSUPP)
1220                                         error = 0;
1221                                 info->feature_flush = 0;
1222                                 xlvbd_flush(info);
1223                         }
1224                         /* fall through */
1225                 case BLKIF_OP_READ:
1226                 case BLKIF_OP_WRITE:
1227                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1228                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1229                                         "request: %x\n", bret->status);
1230
1231                         __blk_end_request_all(req, error);
1232                         break;
1233                 default:
1234                         BUG();
1235                 }
1236         }
1237
1238         info->ring.rsp_cons = i;
1239
1240         if (i != info->ring.req_prod_pvt) {
1241                 int more_to_do;
1242                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1243                 if (more_to_do)
1244                         goto again;
1245         } else
1246                 info->ring.sring->rsp_event = i + 1;
1247
1248         kick_pending_request_queues(info);
1249
1250         spin_unlock_irqrestore(&info->io_lock, flags);
1251
1252         return IRQ_HANDLED;
1253 }
1254
1255
1256 static int setup_blkring(struct xenbus_device *dev,
1257                          struct blkfront_info *info)
1258 {
1259         struct blkif_sring *sring;
1260         int err, i;
1261         unsigned long ring_size = info->nr_ring_pages * PAGE_SIZE;
1262         grant_ref_t gref[XENBUS_MAX_RING_PAGES];
1263
1264         for (i = 0; i < info->nr_ring_pages; i++)
1265                 info->ring_ref[i] = GRANT_INVALID_REF;
1266
1267         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1268                                                        get_order(ring_size));
1269         if (!sring) {
1270                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1271                 return -ENOMEM;
1272         }
1273         SHARED_RING_INIT(sring);
1274         FRONT_RING_INIT(&info->ring, sring, ring_size);
1275
1276         err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, gref);
1277         if (err < 0) {
1278                 free_pages((unsigned long)sring, get_order(ring_size));
1279                 info->ring.sring = NULL;
1280                 goto fail;
1281         }
1282         for (i = 0; i < info->nr_ring_pages; i++)
1283                 info->ring_ref[i] = gref[i];
1284
1285         err = xenbus_alloc_evtchn(dev, &info->evtchn);
1286         if (err)
1287                 goto fail;
1288
1289         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1290                                         "blkif", info);
1291         if (err <= 0) {
1292                 xenbus_dev_fatal(dev, err,
1293                                  "bind_evtchn_to_irqhandler failed");
1294                 goto fail;
1295         }
1296         info->irq = err;
1297
1298         return 0;
1299 fail:
1300         blkif_free(info, 0);
1301         return err;
1302 }
1303
1304
1305 /* Common code used when first setting up, and when resuming. */
1306 static int talk_to_blkback(struct xenbus_device *dev,
1307                            struct blkfront_info *info)
1308 {
1309         const char *message = NULL;
1310         struct xenbus_transaction xbt;
1311         int err, i;
1312         unsigned int max_page_order = 0;
1313         unsigned int ring_page_order = 0;
1314
1315         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1316                            "max-ring-page-order", "%u", &max_page_order);
1317         if (err != 1)
1318                 info->nr_ring_pages = 1;
1319         else {
1320                 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1321                 info->nr_ring_pages = 1 << ring_page_order;
1322         }
1323
1324         /* Create shared ring, alloc event channel. */
1325         err = setup_blkring(dev, info);
1326         if (err)
1327                 goto out;
1328
1329 again:
1330         err = xenbus_transaction_start(&xbt);
1331         if (err) {
1332                 xenbus_dev_fatal(dev, err, "starting transaction");
1333                 goto destroy_blkring;
1334         }
1335
1336         if (info->nr_ring_pages == 1) {
1337                 err = xenbus_printf(xbt, dev->nodename,
1338                                     "ring-ref", "%u", info->ring_ref[0]);
1339                 if (err) {
1340                         message = "writing ring-ref";
1341                         goto abort_transaction;
1342                 }
1343         } else {
1344                 err = xenbus_printf(xbt, dev->nodename,
1345                                     "ring-page-order", "%u", ring_page_order);
1346                 if (err) {
1347                         message = "writing ring-page-order";
1348                         goto abort_transaction;
1349                 }
1350
1351                 for (i = 0; i < info->nr_ring_pages; i++) {
1352                         char ring_ref_name[RINGREF_NAME_LEN];
1353
1354                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1355                         err = xenbus_printf(xbt, dev->nodename, ring_ref_name,
1356                                             "%u", info->ring_ref[i]);
1357                         if (err) {
1358                                 message = "writing ring-ref";
1359                                 goto abort_transaction;
1360                         }
1361                 }
1362         }
1363         err = xenbus_printf(xbt, dev->nodename,
1364                             "event-channel", "%u", info->evtchn);
1365         if (err) {
1366                 message = "writing event-channel";
1367                 goto abort_transaction;
1368         }
1369         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1370                             XEN_IO_PROTO_ABI_NATIVE);
1371         if (err) {
1372                 message = "writing protocol";
1373                 goto abort_transaction;
1374         }
1375         err = xenbus_printf(xbt, dev->nodename,
1376                             "feature-persistent", "%u", 1);
1377         if (err)
1378                 dev_warn(&dev->dev,
1379                          "writing persistent grants feature to xenbus");
1380
1381         err = xenbus_transaction_end(xbt, 0);
1382         if (err) {
1383                 if (err == -EAGAIN)
1384                         goto again;
1385                 xenbus_dev_fatal(dev, err, "completing transaction");
1386                 goto destroy_blkring;
1387         }
1388
1389         for (i = 0; i < BLK_RING_SIZE(info); i++)
1390                 info->shadow[i].req.u.rw.id = i+1;
1391         info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1392         xenbus_switch_state(dev, XenbusStateInitialised);
1393
1394         return 0;
1395
1396  abort_transaction:
1397         xenbus_transaction_end(xbt, 1);
1398         if (message)
1399                 xenbus_dev_fatal(dev, err, "%s", message);
1400  destroy_blkring:
1401         blkif_free(info, 0);
1402  out:
1403         return err;
1404 }
1405
1406 /**
1407  * Entry point to this code when a new device is created.  Allocate the basic
1408  * structures and the ring buffer for communication with the backend, and
1409  * inform the backend of the appropriate details for those.  Switch to
1410  * Initialised state.
1411  */
1412 static int blkfront_probe(struct xenbus_device *dev,
1413                           const struct xenbus_device_id *id)
1414 {
1415         int err, vdevice;
1416         struct blkfront_info *info;
1417
1418         /* FIXME: Use dynamic device id if this is not set. */
1419         err = xenbus_scanf(XBT_NIL, dev->nodename,
1420                            "virtual-device", "%i", &vdevice);
1421         if (err != 1) {
1422                 /* go looking in the extended area instead */
1423                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1424                                    "%i", &vdevice);
1425                 if (err != 1) {
1426                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1427                         return err;
1428                 }
1429         }
1430
1431         if (xen_hvm_domain()) {
1432                 char *type;
1433                 int len;
1434                 /* no unplug has been done: do not hook devices != xen vbds */
1435                 if (xen_has_pv_and_legacy_disk_devices()) {
1436                         int major;
1437
1438                         if (!VDEV_IS_EXTENDED(vdevice))
1439                                 major = BLKIF_MAJOR(vdevice);
1440                         else
1441                                 major = XENVBD_MAJOR;
1442
1443                         if (major != XENVBD_MAJOR) {
1444                                 printk(KERN_INFO
1445                                                 "%s: HVM does not support vbd %d as xen block device\n",
1446                                                 __func__, vdevice);
1447                                 return -ENODEV;
1448                         }
1449                 }
1450                 /* do not create a PV cdrom device if we are an HVM guest */
1451                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1452                 if (IS_ERR(type))
1453                         return -ENODEV;
1454                 if (strncmp(type, "cdrom", 5) == 0) {
1455                         kfree(type);
1456                         return -ENODEV;
1457                 }
1458                 kfree(type);
1459         }
1460         info = kzalloc(sizeof(*info), GFP_KERNEL);
1461         if (!info) {
1462                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1463                 return -ENOMEM;
1464         }
1465
1466         mutex_init(&info->mutex);
1467         spin_lock_init(&info->io_lock);
1468         info->xbdev = dev;
1469         info->vdevice = vdevice;
1470         INIT_LIST_HEAD(&info->grants);
1471         INIT_LIST_HEAD(&info->indirect_pages);
1472         info->persistent_gnts_c = 0;
1473         info->connected = BLKIF_STATE_DISCONNECTED;
1474         INIT_WORK(&info->work, blkif_restart_queue);
1475
1476         /* Front end dir is a number, which is used as the id. */
1477         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1478         dev_set_drvdata(&dev->dev, info);
1479
1480         return 0;
1481 }
1482
1483 static void split_bio_end(struct bio *bio)
1484 {
1485         struct split_bio *split_bio = bio->bi_private;
1486
1487         if (atomic_dec_and_test(&split_bio->pending)) {
1488                 split_bio->bio->bi_phys_segments = 0;
1489                 split_bio->bio->bi_error = bio->bi_error;
1490                 bio_endio(split_bio->bio);
1491                 kfree(split_bio);
1492         }
1493         bio_put(bio);
1494 }
1495
1496 static int blkif_recover(struct blkfront_info *info)
1497 {
1498         int i;
1499         struct request *req, *n;
1500         struct blk_shadow *copy;
1501         int rc;
1502         struct bio *bio, *cloned_bio;
1503         struct bio_list bio_list, merge_bio;
1504         unsigned int segs, offset;
1505         int pending, size;
1506         struct split_bio *split_bio;
1507         struct list_head requests;
1508
1509         /* Stage 1: Make a safe copy of the shadow state. */
1510         copy = kmemdup(info->shadow, sizeof(info->shadow),
1511                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1512         if (!copy)
1513                 return -ENOMEM;
1514
1515         /* Stage 2: Set up free list. */
1516         memset(&info->shadow, 0, sizeof(info->shadow));
1517         for (i = 0; i < BLK_RING_SIZE(info); i++)
1518                 info->shadow[i].req.u.rw.id = i+1;
1519         info->shadow_free = info->ring.req_prod_pvt;
1520         info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1521
1522         rc = blkfront_gather_backend_features(info);
1523         if (rc) {
1524                 kfree(copy);
1525                 return rc;
1526         }
1527
1528         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1529         blk_queue_max_segments(info->rq, segs);
1530         bio_list_init(&bio_list);
1531         INIT_LIST_HEAD(&requests);
1532         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1533                 /* Not in use? */
1534                 if (!copy[i].request)
1535                         continue;
1536
1537                 /*
1538                  * Get the bios in the request so we can re-queue them.
1539                  */
1540                 if (copy[i].request->cmd_flags &
1541                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1542                         /*
1543                          * Flush operations don't contain bios, so
1544                          * we need to requeue the whole request
1545                          */
1546                         list_add(&copy[i].request->queuelist, &requests);
1547                         continue;
1548                 }
1549                 merge_bio.head = copy[i].request->bio;
1550                 merge_bio.tail = copy[i].request->biotail;
1551                 bio_list_merge(&bio_list, &merge_bio);
1552                 copy[i].request->bio = NULL;
1553                 blk_end_request_all(copy[i].request, 0);
1554         }
1555
1556         kfree(copy);
1557
1558         /*
1559          * Empty the queue, this is important because we might have
1560          * requests in the queue with more segments than what we
1561          * can handle now.
1562          */
1563         spin_lock_irq(&info->io_lock);
1564         while ((req = blk_fetch_request(info->rq)) != NULL) {
1565                 if (req->cmd_flags &
1566                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1567                         list_add(&req->queuelist, &requests);
1568                         continue;
1569                 }
1570                 merge_bio.head = req->bio;
1571                 merge_bio.tail = req->biotail;
1572                 bio_list_merge(&bio_list, &merge_bio);
1573                 req->bio = NULL;
1574                 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1575                         pr_alert("diskcache flush request found!\n");
1576                 __blk_end_request_all(req, 0);
1577         }
1578         spin_unlock_irq(&info->io_lock);
1579
1580         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1581
1582         spin_lock_irq(&info->io_lock);
1583
1584         /* Now safe for us to use the shared ring */
1585         info->connected = BLKIF_STATE_CONNECTED;
1586
1587         /* Kick any other new requests queued since we resumed */
1588         kick_pending_request_queues(info);
1589
1590         list_for_each_entry_safe(req, n, &requests, queuelist) {
1591                 /* Requeue pending requests (flush or discard) */
1592                 list_del_init(&req->queuelist);
1593                 BUG_ON(req->nr_phys_segments > segs);
1594                 blk_requeue_request(info->rq, req);
1595         }
1596         spin_unlock_irq(&info->io_lock);
1597
1598         while ((bio = bio_list_pop(&bio_list)) != NULL) {
1599                 /* Traverse the list of pending bios and re-queue them */
1600                 if (bio_segments(bio) > segs) {
1601                         /*
1602                          * This bio has more segments than what we can
1603                          * handle, we have to split it.
1604                          */
1605                         pending = (bio_segments(bio) + segs - 1) / segs;
1606                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1607                         BUG_ON(split_bio == NULL);
1608                         atomic_set(&split_bio->pending, pending);
1609                         split_bio->bio = bio;
1610                         for (i = 0; i < pending; i++) {
1611                                 offset = (i * segs * PAGE_SIZE) >> 9;
1612                                 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1613                                            (unsigned int)bio_sectors(bio) - offset);
1614                                 cloned_bio = bio_clone(bio, GFP_NOIO);
1615                                 BUG_ON(cloned_bio == NULL);
1616                                 bio_trim(cloned_bio, offset, size);
1617                                 cloned_bio->bi_private = split_bio;
1618                                 cloned_bio->bi_end_io = split_bio_end;
1619                                 submit_bio(cloned_bio->bi_rw, cloned_bio);
1620                         }
1621                         /*
1622                          * Now we have to wait for all those smaller bios to
1623                          * end, so we can also end the "parent" bio.
1624                          */
1625                         continue;
1626                 }
1627                 /* We don't need to split this bio */
1628                 submit_bio(bio->bi_rw, bio);
1629         }
1630
1631         return 0;
1632 }
1633
1634 /**
1635  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1636  * driver restart.  We tear down our blkif structure and recreate it, but
1637  * leave the device-layer structures intact so that this is transparent to the
1638  * rest of the kernel.
1639  */
1640 static int blkfront_resume(struct xenbus_device *dev)
1641 {
1642         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1643         int err;
1644
1645         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1646
1647         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1648
1649         err = talk_to_blkback(dev, info);
1650
1651         /*
1652          * We have to wait for the backend to switch to
1653          * connected state, since we want to read which
1654          * features it supports.
1655          */
1656
1657         return err;
1658 }
1659
1660 static void
1661 blkfront_closing(struct blkfront_info *info)
1662 {
1663         struct xenbus_device *xbdev = info->xbdev;
1664         struct block_device *bdev = NULL;
1665
1666         mutex_lock(&info->mutex);
1667
1668         if (xbdev->state == XenbusStateClosing) {
1669                 mutex_unlock(&info->mutex);
1670                 return;
1671         }
1672
1673         if (info->gd)
1674                 bdev = bdget_disk(info->gd, 0);
1675
1676         mutex_unlock(&info->mutex);
1677
1678         if (!bdev) {
1679                 xenbus_frontend_closed(xbdev);
1680                 return;
1681         }
1682
1683         mutex_lock(&bdev->bd_mutex);
1684
1685         if (bdev->bd_openers) {
1686                 xenbus_dev_error(xbdev, -EBUSY,
1687                                  "Device in use; refusing to close");
1688                 xenbus_switch_state(xbdev, XenbusStateClosing);
1689         } else {
1690                 xlvbd_release_gendisk(info);
1691                 xenbus_frontend_closed(xbdev);
1692         }
1693
1694         mutex_unlock(&bdev->bd_mutex);
1695         bdput(bdev);
1696 }
1697
1698 static void blkfront_setup_discard(struct blkfront_info *info)
1699 {
1700         int err;
1701         unsigned int discard_granularity;
1702         unsigned int discard_alignment;
1703         unsigned int discard_secure;
1704
1705         info->feature_discard = 1;
1706         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1707                 "discard-granularity", "%u", &discard_granularity,
1708                 "discard-alignment", "%u", &discard_alignment,
1709                 NULL);
1710         if (!err) {
1711                 info->discard_granularity = discard_granularity;
1712                 info->discard_alignment = discard_alignment;
1713         }
1714         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1715                     "discard-secure", "%d", &discard_secure,
1716                     NULL);
1717         if (!err)
1718                 info->feature_secdiscard = !!discard_secure;
1719 }
1720
1721 static int blkfront_setup_indirect(struct blkfront_info *info)
1722 {
1723         unsigned int segs;
1724         int err, i;
1725
1726         if (info->max_indirect_segments == 0)
1727                 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1728         else
1729                 segs = info->max_indirect_segments;
1730
1731         err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE(info));
1732         if (err)
1733                 goto out_of_memory;
1734
1735         if (!info->feature_persistent && info->max_indirect_segments) {
1736                 /*
1737                  * We are using indirect descriptors but not persistent
1738                  * grants, we need to allocate a set of pages that can be
1739                  * used for mapping indirect grefs
1740                  */
1741                 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE(info);
1742
1743                 BUG_ON(!list_empty(&info->indirect_pages));
1744                 for (i = 0; i < num; i++) {
1745                         struct page *indirect_page = alloc_page(GFP_NOIO);
1746                         if (!indirect_page)
1747                                 goto out_of_memory;
1748                         list_add(&indirect_page->lru, &info->indirect_pages);
1749                 }
1750         }
1751
1752         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1753                 info->shadow[i].grants_used = kzalloc(
1754                         sizeof(info->shadow[i].grants_used[0]) * segs,
1755                         GFP_NOIO);
1756                 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1757                 if (info->max_indirect_segments)
1758                         info->shadow[i].indirect_grants = kzalloc(
1759                                 sizeof(info->shadow[i].indirect_grants[0]) *
1760                                 INDIRECT_GREFS(segs),
1761                                 GFP_NOIO);
1762                 if ((info->shadow[i].grants_used == NULL) ||
1763                         (info->shadow[i].sg == NULL) ||
1764                      (info->max_indirect_segments &&
1765                      (info->shadow[i].indirect_grants == NULL)))
1766                         goto out_of_memory;
1767                 sg_init_table(info->shadow[i].sg, segs);
1768         }
1769
1770
1771         return 0;
1772
1773 out_of_memory:
1774         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1775                 kfree(info->shadow[i].grants_used);
1776                 info->shadow[i].grants_used = NULL;
1777                 kfree(info->shadow[i].sg);
1778                 info->shadow[i].sg = NULL;
1779                 kfree(info->shadow[i].indirect_grants);
1780                 info->shadow[i].indirect_grants = NULL;
1781         }
1782         if (!list_empty(&info->indirect_pages)) {
1783                 struct page *indirect_page, *n;
1784                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1785                         list_del(&indirect_page->lru);
1786                         __free_page(indirect_page);
1787                 }
1788         }
1789         return -ENOMEM;
1790 }
1791
1792 /*
1793  * Gather all backend feature-*
1794  */
1795 static int blkfront_gather_backend_features(struct blkfront_info *info)
1796 {
1797         int err;
1798         int barrier, flush, discard, persistent;
1799         unsigned int indirect_segments;
1800
1801         info->feature_flush = 0;
1802
1803         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1804                         "feature-barrier", "%d", &barrier,
1805                         NULL);
1806
1807         /*
1808          * If there's no "feature-barrier" defined, then it means
1809          * we're dealing with a very old backend which writes
1810          * synchronously; nothing to do.
1811          *
1812          * If there are barriers, then we use flush.
1813          */
1814         if (!err && barrier)
1815                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1816         /*
1817          * And if there is "feature-flush-cache" use that above
1818          * barriers.
1819          */
1820         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1821                         "feature-flush-cache", "%d", &flush,
1822                         NULL);
1823
1824         if (!err && flush)
1825                 info->feature_flush = REQ_FLUSH;
1826
1827         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1828                         "feature-discard", "%d", &discard,
1829                         NULL);
1830
1831         if (!err && discard)
1832                 blkfront_setup_discard(info);
1833
1834         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1835                         "feature-persistent", "%u", &persistent,
1836                         NULL);
1837         if (err)
1838                 info->feature_persistent = 0;
1839         else
1840                 info->feature_persistent = persistent;
1841
1842         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1843                             "feature-max-indirect-segments", "%u", &indirect_segments,
1844                             NULL);
1845         if (err)
1846                 info->max_indirect_segments = 0;
1847         else
1848                 info->max_indirect_segments = min(indirect_segments,
1849                                                   xen_blkif_max_segments);
1850
1851         return blkfront_setup_indirect(info);
1852 }
1853
1854 /*
1855  * Invoked when the backend is finally 'ready' (and has told produced
1856  * the details about the physical device - #sectors, size, etc).
1857  */
1858 static void blkfront_connect(struct blkfront_info *info)
1859 {
1860         unsigned long long sectors;
1861         unsigned long sector_size;
1862         unsigned int physical_sector_size;
1863         unsigned int binfo;
1864         int err;
1865
1866         switch (info->connected) {
1867         case BLKIF_STATE_CONNECTED:
1868                 /*
1869                  * Potentially, the back-end may be signalling
1870                  * a capacity change; update the capacity.
1871                  */
1872                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1873                                    "sectors", "%Lu", &sectors);
1874                 if (XENBUS_EXIST_ERR(err))
1875                         return;
1876                 printk(KERN_INFO "Setting capacity to %Lu\n",
1877                        sectors);
1878                 set_capacity(info->gd, sectors);
1879                 revalidate_disk(info->gd);
1880
1881                 return;
1882         case BLKIF_STATE_SUSPENDED:
1883                 /*
1884                  * If we are recovering from suspension, we need to wait
1885                  * for the backend to announce it's features before
1886                  * reconnecting, at least we need to know if the backend
1887                  * supports indirect descriptors, and how many.
1888                  */
1889                 blkif_recover(info);
1890                 return;
1891
1892         default:
1893                 break;
1894         }
1895
1896         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1897                 __func__, info->xbdev->otherend);
1898
1899         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1900                             "sectors", "%llu", &sectors,
1901                             "info", "%u", &binfo,
1902                             "sector-size", "%lu", &sector_size,
1903                             NULL);
1904         if (err) {
1905                 xenbus_dev_fatal(info->xbdev, err,
1906                                  "reading backend fields at %s",
1907                                  info->xbdev->otherend);
1908                 return;
1909         }
1910
1911         /*
1912          * physcial-sector-size is a newer field, so old backends may not
1913          * provide this. Assume physical sector size to be the same as
1914          * sector_size in that case.
1915          */
1916         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1917                            "physical-sector-size", "%u", &physical_sector_size);
1918         if (err != 1)
1919                 physical_sector_size = sector_size;
1920
1921         err = blkfront_gather_backend_features(info);
1922         if (err) {
1923                 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1924                                  info->xbdev->otherend);
1925                 return;
1926         }
1927
1928         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1929                                   physical_sector_size);
1930         if (err) {
1931                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1932                                  info->xbdev->otherend);
1933                 return;
1934         }
1935
1936         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1937
1938         /* Kick pending requests. */
1939         spin_lock_irq(&info->io_lock);
1940         info->connected = BLKIF_STATE_CONNECTED;
1941         kick_pending_request_queues(info);
1942         spin_unlock_irq(&info->io_lock);
1943
1944         add_disk(info->gd);
1945
1946         info->is_ready = 1;
1947 }
1948
1949 /**
1950  * Callback received when the backend's state changes.
1951  */
1952 static void blkback_changed(struct xenbus_device *dev,
1953                             enum xenbus_state backend_state)
1954 {
1955         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1956
1957         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1958
1959         switch (backend_state) {
1960         case XenbusStateInitWait:
1961                 if (dev->state != XenbusStateInitialising)
1962                         break;
1963                 if (talk_to_blkback(dev, info)) {
1964                         kfree(info);
1965                         dev_set_drvdata(&dev->dev, NULL);
1966                         break;
1967                 }
1968         case XenbusStateInitialising:
1969         case XenbusStateInitialised:
1970         case XenbusStateReconfiguring:
1971         case XenbusStateReconfigured:
1972         case XenbusStateUnknown:
1973                 break;
1974
1975         case XenbusStateConnected:
1976                 blkfront_connect(info);
1977                 break;
1978
1979         case XenbusStateClosed:
1980                 if (dev->state == XenbusStateClosed)
1981                         break;
1982                 /* Missed the backend's Closing state -- fallthrough */
1983         case XenbusStateClosing:
1984                 blkfront_closing(info);
1985                 break;
1986         }
1987 }
1988
1989 static int blkfront_remove(struct xenbus_device *xbdev)
1990 {
1991         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1992         struct block_device *bdev = NULL;
1993         struct gendisk *disk;
1994
1995         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1996
1997         blkif_free(info, 0);
1998
1999         mutex_lock(&info->mutex);
2000
2001         disk = info->gd;
2002         if (disk)
2003                 bdev = bdget_disk(disk, 0);
2004
2005         info->xbdev = NULL;
2006         mutex_unlock(&info->mutex);
2007
2008         if (!bdev) {
2009                 kfree(info);
2010                 return 0;
2011         }
2012
2013         /*
2014          * The xbdev was removed before we reached the Closed
2015          * state. See if it's safe to remove the disk. If the bdev
2016          * isn't closed yet, we let release take care of it.
2017          */
2018
2019         mutex_lock(&bdev->bd_mutex);
2020         info = disk->private_data;
2021
2022         dev_warn(disk_to_dev(disk),
2023                  "%s was hot-unplugged, %d stale handles\n",
2024                  xbdev->nodename, bdev->bd_openers);
2025
2026         if (info && !bdev->bd_openers) {
2027                 xlvbd_release_gendisk(info);
2028                 disk->private_data = NULL;
2029                 kfree(info);
2030         }
2031
2032         mutex_unlock(&bdev->bd_mutex);
2033         bdput(bdev);
2034
2035         return 0;
2036 }
2037
2038 static int blkfront_is_ready(struct xenbus_device *dev)
2039 {
2040         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2041
2042         return info->is_ready && info->xbdev;
2043 }
2044
2045 static int blkif_open(struct block_device *bdev, fmode_t mode)
2046 {
2047         struct gendisk *disk = bdev->bd_disk;
2048         struct blkfront_info *info;
2049         int err = 0;
2050
2051         mutex_lock(&blkfront_mutex);
2052
2053         info = disk->private_data;
2054         if (!info) {
2055                 /* xbdev gone */
2056                 err = -ERESTARTSYS;
2057                 goto out;
2058         }
2059
2060         mutex_lock(&info->mutex);
2061
2062         if (!info->gd)
2063                 /* xbdev is closed */
2064                 err = -ERESTARTSYS;
2065
2066         mutex_unlock(&info->mutex);
2067
2068 out:
2069         mutex_unlock(&blkfront_mutex);
2070         return err;
2071 }
2072
2073 static void blkif_release(struct gendisk *disk, fmode_t mode)
2074 {
2075         struct blkfront_info *info = disk->private_data;
2076         struct block_device *bdev;
2077         struct xenbus_device *xbdev;
2078
2079         mutex_lock(&blkfront_mutex);
2080
2081         bdev = bdget_disk(disk, 0);
2082
2083         if (!bdev) {
2084                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2085                 goto out_mutex;
2086         }
2087         if (bdev->bd_openers)
2088                 goto out;
2089
2090         /*
2091          * Check if we have been instructed to close. We will have
2092          * deferred this request, because the bdev was still open.
2093          */
2094
2095         mutex_lock(&info->mutex);
2096         xbdev = info->xbdev;
2097
2098         if (xbdev && xbdev->state == XenbusStateClosing) {
2099                 /* pending switch to state closed */
2100                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2101                 xlvbd_release_gendisk(info);
2102                 xenbus_frontend_closed(info->xbdev);
2103         }
2104
2105         mutex_unlock(&info->mutex);
2106
2107         if (!xbdev) {
2108                 /* sudden device removal */
2109                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2110                 xlvbd_release_gendisk(info);
2111                 disk->private_data = NULL;
2112                 kfree(info);
2113         }
2114
2115 out:
2116         bdput(bdev);
2117 out_mutex:
2118         mutex_unlock(&blkfront_mutex);
2119 }
2120
2121 static const struct block_device_operations xlvbd_block_fops =
2122 {
2123         .owner = THIS_MODULE,
2124         .open = blkif_open,
2125         .release = blkif_release,
2126         .getgeo = blkif_getgeo,
2127         .ioctl = blkif_ioctl,
2128 };
2129
2130
2131 static const struct xenbus_device_id blkfront_ids[] = {
2132         { "vbd" },
2133         { "" }
2134 };
2135
2136 static struct xenbus_driver blkfront_driver = {
2137         .ids  = blkfront_ids,
2138         .probe = blkfront_probe,
2139         .remove = blkfront_remove,
2140         .resume = blkfront_resume,
2141         .otherend_changed = blkback_changed,
2142         .is_ready = blkfront_is_ready,
2143 };
2144
2145 static int __init xlblk_init(void)
2146 {
2147         int ret;
2148
2149         if (!xen_domain())
2150                 return -ENODEV;
2151
2152         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_PAGE_ORDER) {
2153                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2154                         xen_blkif_max_ring_order, XENBUS_MAX_RING_PAGE_ORDER);
2155                 xen_blkif_max_ring_order = 0;
2156         }
2157
2158         if (!xen_has_pv_disk_devices())
2159                 return -ENODEV;
2160
2161         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2162                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2163                        XENVBD_MAJOR, DEV_NAME);
2164                 return -ENODEV;
2165         }
2166
2167         ret = xenbus_register_frontend(&blkfront_driver);
2168         if (ret) {
2169                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2170                 return ret;
2171         }
2172
2173         return 0;
2174 }
2175 module_init(xlblk_init);
2176
2177
2178 static void __exit xlblk_exit(void)
2179 {
2180         xenbus_unregister_driver(&blkfront_driver);
2181         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2182         kfree(minors);
2183 }
2184 module_exit(xlblk_exit);
2185
2186 MODULE_DESCRIPTION("Xen virtual block device frontend");
2187 MODULE_LICENSE("GPL");
2188 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2189 MODULE_ALIAS("xen:vbd");
2190 MODULE_ALIAS("xenblk");