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 /*
  64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  65  * When Linux is using a different page size than Xen, it may not be possible
  66  * to put all the data in a single segment.
  67  * This can happen when the backend doesn't support indirect descriptor and
  68  * therefore the maximum amount of data that a request can carry is
  69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  70  *
  71  * Note that we only support one extra request. So the Linux page size
  72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  73  * 88KB.
  74  */
  75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  76
  77 enum blkif_state {
  78         BLKIF_STATE_DISCONNECTED,
  79         BLKIF_STATE_CONNECTED,
  80         BLKIF_STATE_SUSPENDED,
  81 };
  82
  83 struct grant {
  84         grant_ref_t gref;
  85         struct page *page;
  86         struct list_head node;
  87 };
  88
  89 enum blk_req_status {
  90         REQ_WAITING,
  91         REQ_DONE,
  92         REQ_ERROR,
  93         REQ_EOPNOTSUPP,
  94 };
  95
  96 struct blk_shadow {
  97         struct blkif_request req;
  98         struct request *request;
  99         struct grant **grants_used;
 100         struct grant **indirect_grants;
 101         struct scatterlist *sg;
 102         unsigned int num_sg;
 103         enum blk_req_status status;
 104
 105         #define NO_ASSOCIATED_ID ~0UL
 106         /*
 107          * Id of the sibling if we ever need 2 requests when handling a
 108          * block I/O request
 109          */
 110         unsigned long associated_id;
 111 };
 112
 113 struct split_bio {
 114         struct bio *bio;
 115         atomic_t pending;
 116 };
 117
 118 static DEFINE_MUTEX(blkfront_mutex);
 119 static const struct block_device_operations xlvbd_block_fops;
 120
 121 /*
 122  * Maximum number of segments in indirect requests, the actual value used by
 123  * the frontend driver is the minimum of this value and the value provided
 124  * by the backend driver.
 125  */
 126
 127 static unsigned int xen_blkif_max_segments = 32;
 128 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
 129                    S_IRUGO);
 130 MODULE_PARM_DESC(max_indirect_segments,
 131                  "Maximum amount of segments in indirect requests (default is 32)");
 132
 133 static unsigned int xen_blkif_max_queues = 4;
 134 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
 135 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 136
 137 /*
 138  * Maximum order of pages to be used for the shared ring between front and
 139  * backend, 4KB page granularity is used.
 140  */
 141 static unsigned int xen_blkif_max_ring_order;
 142 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
 143 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 144
 145 #define BLK_RING_SIZE(info)     \
 146         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 147
 148 #define BLK_MAX_RING_SIZE       \
 149         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
 150
 151 /*
 152  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 153  * characters are enough. Define to 20 to keep consistent with backend.
 154  */
 155 #define RINGREF_NAME_LEN (20)
 156 /*
 157  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 158  */
 159 #define QUEUE_NAME_LEN (17)
 160
 161 /*
 162  *  Per-ring info.
 163  *  Every blkfront device can associate with one or more blkfront_ring_info,
 164  *  depending on how many hardware queues/rings to be used.
 165  */
 166 struct blkfront_ring_info {
 167         /* Lock to protect data in every ring buffer. */
 168         spinlock_t ring_lock;
 169         struct blkif_front_ring ring;
 170         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 171         unsigned int evtchn, irq;
 172         struct work_struct work;
 173         struct gnttab_free_callback callback;
 174         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 175         struct list_head indirect_pages;
 176         struct list_head grants;
 177         unsigned int persistent_gnts_c;
 178         unsigned long shadow_free;
 179         struct blkfront_info *dev_info;
 180 };
 181
 182 /*
 183  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 184  * hang in private_data off the gendisk structure. We may end up
 185  * putting all kinds of interesting stuff here :-)
 186  */
 187 struct blkfront_info
 188 {
 189         struct mutex mutex;
 190         struct xenbus_device *xbdev;
 191         struct gendisk *gd;
 192         int vdevice;
 193         blkif_vdev_t handle;
 194         enum blkif_state connected;
 195         /* Number of pages per ring buffer. */
 196         unsigned int nr_ring_pages;
 197         struct request_queue *rq;
 198         unsigned int feature_flush;
 199         unsigned int feature_discard:1;
 200         unsigned int feature_secdiscard:1;
 201         unsigned int discard_granularity;
 202         unsigned int discard_alignment;
 203         unsigned int feature_persistent:1;
 204         /* Number of 4KB segments handled */
 205         unsigned int max_indirect_segments;
 206         int is_ready;
 207         struct blk_mq_tag_set tag_set;
 208         struct blkfront_ring_info *rinfo;
 209         unsigned int nr_rings;
 210 };
 211
 212 static unsigned int nr_minors;
 213 static unsigned long *minors;
 214 static DEFINE_SPINLOCK(minor_lock);
 215
 216 #define GRANT_INVALID_REF       0
 217
 218 #define PARTS_PER_DISK          16
 219 #define PARTS_PER_EXT_DISK      256
 220
 221 #define BLKIF_MAJOR(dev) ((dev)>>8)
 222 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 223
 224 #define EXT_SHIFT 28
 225 #define EXTENDED (1<<EXT_SHIFT)
 226 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 227 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 228 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 229 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 230 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 231 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 232
 233 #define DEV_NAME        "xvd"   /* name in /dev */
 234
 235 /*
 236  * Grants are always the same size as a Xen page (i.e 4KB).
 237  * A physical segment is always the same size as a Linux page.
 238  * Number of grants per physical segment
 239  */
 240 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 241
 242 #define GRANTS_PER_INDIRECT_FRAME \
 243         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 244
 245 #define PSEGS_PER_INDIRECT_FRAME        \
 246         (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
 247
 248 #define INDIRECT_GREFS(_grants)         \
 249         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 250
 251 #define GREFS(_psegs)   ((_psegs) * GRANTS_PER_PSEG)
 252
 253 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 254 static void blkfront_gather_backend_features(struct blkfront_info *info);
 255
 256 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 257 {
 258         unsigned long free = rinfo->shadow_free;
 259
 260         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 261         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 262         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 263         return free;
 264 }
 265
 266 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 267                               unsigned long id)
 268 {
 269         if (rinfo->shadow[id].req.u.rw.id != id)
 270                 return -EINVAL;
 271         if (rinfo->shadow[id].request == NULL)
 272                 return -EINVAL;
 273         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 274         rinfo->shadow[id].request = NULL;
 275         rinfo->shadow_free = id;
 276         return 0;
 277 }
 278
 279 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 280 {
 281         struct blkfront_info *info = rinfo->dev_info;
 282         struct page *granted_page;
 283         struct grant *gnt_list_entry, *n;
 284         int i = 0;
 285
 286         while (i < num) {
 287                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 288                 if (!gnt_list_entry)
 289                         goto out_of_memory;
 290
 291                 if (info->feature_persistent) {
 292                         granted_page = alloc_page(GFP_NOIO);
 293                         if (!granted_page) {
 294                                 kfree(gnt_list_entry);
 295                                 goto out_of_memory;
 296                         }
 297                         gnt_list_entry->page = granted_page;
 298                 }
 299
 300                 gnt_list_entry->gref = GRANT_INVALID_REF;
 301                 list_add(&gnt_list_entry->node, &rinfo->grants);
 302                 i++;
 303         }
 304
 305         return 0;
 306
 307 out_of_memory:
 308         list_for_each_entry_safe(gnt_list_entry, n,
 309                                  &rinfo->grants, node) {
 310                 list_del(&gnt_list_entry->node);
 311                 if (info->feature_persistent)
 312                         __free_page(gnt_list_entry->page);
 313                 kfree(gnt_list_entry);
 314                 i--;
 315         }
 316         BUG_ON(i != 0);
 317         return -ENOMEM;
 318 }
 319
 320 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 321 {
 322         struct grant *gnt_list_entry;
 323
 324         BUG_ON(list_empty(&rinfo->grants));
 325         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 326                                           node);
 327         list_del(&gnt_list_entry->node);
 328
 329         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 330                 rinfo->persistent_gnts_c--;
 331
 332         return gnt_list_entry;
 333 }
 334
 335 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 336                                         const struct blkfront_info *info)
 337 {
 338         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 339                                                  info->xbdev->otherend_id,
 340                                                  gnt_list_entry->page,
 341                                                  0);
 342 }
 343
 344 static struct grant *get_grant(grant_ref_t *gref_head,
 345                                unsigned long gfn,
 346                                struct blkfront_ring_info *rinfo)
 347 {
 348         struct grant *gnt_list_entry = get_free_grant(rinfo);
 349         struct blkfront_info *info = rinfo->dev_info;
 350
 351         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 352                 return gnt_list_entry;
 353
 354         /* Assign a gref to this page */
 355         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 356         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 357         if (info->feature_persistent)
 358                 grant_foreign_access(gnt_list_entry, info);
 359         else {
 360                 /* Grant access to the GFN passed by the caller */
 361                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 362                                                 info->xbdev->otherend_id,
 363                                                 gfn, 0);
 364         }
 365
 366         return gnt_list_entry;
 367 }
 368
 369 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 370                                         struct blkfront_ring_info *rinfo)
 371 {
 372         struct grant *gnt_list_entry = get_free_grant(rinfo);
 373         struct blkfront_info *info = rinfo->dev_info;
 374
 375         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 376                 return gnt_list_entry;
 377
 378         /* Assign a gref to this page */
 379         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 380         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 381         if (!info->feature_persistent) {
 382                 struct page *indirect_page;
 383
 384                 /* Fetch a pre-allocated page to use for indirect grefs */
 385                 BUG_ON(list_empty(&rinfo->indirect_pages));
 386                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 387                                                  struct page, lru);
 388                 list_del(&indirect_page->lru);
 389                 gnt_list_entry->page = indirect_page;
 390         }
 391         grant_foreign_access(gnt_list_entry, info);
 392
 393         return gnt_list_entry;
 394 }
 395
 396 static const char *op_name(int op)
 397 {
 398         static const char *const names[] = {
 399                 [BLKIF_OP_READ] = "read",
 400                 [BLKIF_OP_WRITE] = "write",
 401                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 402                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 403                 [BLKIF_OP_DISCARD] = "discard" };
 404
 405         if (op < 0 || op >= ARRAY_SIZE(names))
 406                 return "unknown";
 407
 408         if (!names[op])
 409                 return "reserved";
 410
 411         return names[op];
 412 }
 413 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 414 {
 415         unsigned int end = minor + nr;
 416         int rc;
 417
 418         if (end > nr_minors) {
 419                 unsigned long *bitmap, *old;
 420
 421                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 422                                  GFP_KERNEL);
 423                 if (bitmap == NULL)
 424                         return -ENOMEM;
 425
 426                 spin_lock(&minor_lock);
 427                 if (end > nr_minors) {
 428                         old = minors;
 429                         memcpy(bitmap, minors,
 430                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 431                         minors = bitmap;
 432                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 433                 } else
 434                         old = bitmap;
 435                 spin_unlock(&minor_lock);
 436                 kfree(old);
 437         }
 438
 439         spin_lock(&minor_lock);
 440         if (find_next_bit(minors, end, minor) >= end) {
 441                 bitmap_set(minors, minor, nr);
 442                 rc = 0;
 443         } else
 444                 rc = -EBUSY;
 445         spin_unlock(&minor_lock);
 446
 447         return rc;
 448 }
 449
 450 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 451 {
 452         unsigned int end = minor + nr;
 453
 454         BUG_ON(end > nr_minors);
 455         spin_lock(&minor_lock);
 456         bitmap_clear(minors,  minor, nr);
 457         spin_unlock(&minor_lock);
 458 }
 459
 460 static void blkif_restart_queue_callback(void *arg)
 461 {
 462         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 463         schedule_work(&rinfo->work);
 464 }
 465
 466 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 467 {
 468         /* We don't have real geometry info, but let's at least return
 469            values consistent with the size of the device */
 470         sector_t nsect = get_capacity(bd->bd_disk);
 471         sector_t cylinders = nsect;
 472
 473         hg->heads = 0xff;
 474         hg->sectors = 0x3f;
 475         sector_div(cylinders, hg->heads * hg->sectors);
 476         hg->cylinders = cylinders;
 477         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 478                 hg->cylinders = 0xffff;
 479         return 0;
 480 }
 481
 482 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 483                        unsigned command, unsigned long argument)
 484 {
 485         struct blkfront_info *info = bdev->bd_disk->private_data;
 486         int i;
 487
 488         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 489                 command, (long)argument);
 490
 491         switch (command) {
 492         case CDROMMULTISESSION:
 493                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 494                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 495                         if (put_user(0, (char __user *)(argument + i)))
 496                                 return -EFAULT;
 497                 return 0;
 498
 499         case CDROM_GET_CAPABILITY: {
 500                 struct gendisk *gd = info->gd;
 501                 if (gd->flags & GENHD_FL_CD)
 502                         return 0;
 503                 return -EINVAL;
 504         }
 505
 506         default:
 507                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 508                   command);*/
 509                 return -EINVAL; /* same return as native Linux */
 510         }
 511
 512         return 0;
 513 }
 514
 515 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 516                                             struct request *req,
 517                                             struct blkif_request **ring_req)
 518 {
 519         unsigned long id;
 520
 521         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 522         rinfo->ring.req_prod_pvt++;
 523
 524         id = get_id_from_freelist(rinfo);
 525         rinfo->shadow[id].request = req;
 526         rinfo->shadow[id].status = REQ_WAITING;
 527         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 528
 529         (*ring_req)->u.rw.id = id;
 530
 531         return id;
 532 }
 533
 534 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 535 {
 536         struct blkfront_info *info = rinfo->dev_info;
 537         struct blkif_request *ring_req;
 538         unsigned long id;
 539
 540         /* Fill out a communications ring structure. */
 541         id = blkif_ring_get_request(rinfo, req, &ring_req);
 542
 543         ring_req->operation = BLKIF_OP_DISCARD;
 544         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 545         ring_req->u.discard.id = id;
 546         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 547         if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 548                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 549         else
 550                 ring_req->u.discard.flag = 0;
 551
 552         /* Keep a private copy so we can reissue requests when recovering. */
 553         rinfo->shadow[id].req = *ring_req;
 554
 555         return 0;
 556 }
 557
 558 struct setup_rw_req {
 559         unsigned int grant_idx;
 560         struct blkif_request_segment *segments;
 561         struct blkfront_ring_info *rinfo;
 562         struct blkif_request *ring_req;
 563         grant_ref_t gref_head;
 564         unsigned int id;
 565         /* Only used when persistent grant is used and it's a read request */
 566         bool need_copy;
 567         unsigned int bvec_off;
 568         char *bvec_data;
 569
 570         bool require_extra_req;
 571         struct blkif_request *extra_ring_req;
 572 };
 573
 574 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 575                                      unsigned int len, void *data)
 576 {
 577         struct setup_rw_req *setup = data;
 578         int n, ref;
 579         struct grant *gnt_list_entry;
 580         unsigned int fsect, lsect;
 581         /* Convenient aliases */
 582         unsigned int grant_idx = setup->grant_idx;
 583         struct blkif_request *ring_req = setup->ring_req;
 584         struct blkfront_ring_info *rinfo = setup->rinfo;
 585         /*
 586          * We always use the shadow of the first request to store the list
 587          * of grant associated to the block I/O request. This made the
 588          * completion more easy to handle even if the block I/O request is
 589          * split.
 590          */
 591         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 592
 593         if (unlikely(setup->require_extra_req &&
 594                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 595                 /*
 596                  * We are using the second request, setup grant_idx
 597                  * to be the index of the segment array.
 598                  */
 599                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 600                 ring_req = setup->extra_ring_req;
 601         }
 602
 603         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 604             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 605                 if (setup->segments)
 606                         kunmap_atomic(setup->segments);
 607
 608                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 609                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 610                 shadow->indirect_grants[n] = gnt_list_entry;
 611                 setup->segments = kmap_atomic(gnt_list_entry->page);
 612                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 613         }
 614
 615         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 616         ref = gnt_list_entry->gref;
 617         /*
 618          * All the grants are stored in the shadow of the first
 619          * request. Therefore we have to use the global index.
 620          */
 621         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 622
 623         if (setup->need_copy) {
 624                 void *shared_data;
 625
 626                 shared_data = kmap_atomic(gnt_list_entry->page);
 627                 /*
 628                  * this does not wipe data stored outside the
 629                  * range sg->offset..sg->offset+sg->length.
 630                  * Therefore, blkback *could* see data from
 631                  * previous requests. This is OK as long as
 632                  * persistent grants are shared with just one
 633                  * domain. It may need refactoring if this
 634                  * changes
 635                  */
 636                 memcpy(shared_data + offset,
 637                        setup->bvec_data + setup->bvec_off,
 638                        len);
 639
 640                 kunmap_atomic(shared_data);
 641                 setup->bvec_off += len;
 642         }
 643
 644         fsect = offset >> 9;
 645         lsect = fsect + (len >> 9) - 1;
 646         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 647                 ring_req->u.rw.seg[grant_idx] =
 648                         (struct blkif_request_segment) {
 649                                 .gref       = ref,
 650                                 .first_sect = fsect,
 651                                 .last_sect  = lsect };
 652         } else {
 653                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 654                         (struct blkif_request_segment) {
 655                                 .gref       = ref,
 656                                 .first_sect = fsect,
 657                                 .last_sect  = lsect };
 658         }
 659
 660         (setup->grant_idx)++;
 661 }
 662
 663 static void blkif_setup_extra_req(struct blkif_request *first,
 664                                   struct blkif_request *second)
 665 {
 666         uint16_t nr_segments = first->u.rw.nr_segments;
 667
 668         /*
 669          * The second request is only present when the first request uses
 670          * all its segments. It's always the continuity of the first one.
 671          */
 672         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 673
 674         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 675         second->u.rw.sector_number = first->u.rw.sector_number +
 676                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 677
 678         second->u.rw.handle = first->u.rw.handle;
 679         second->operation = first->operation;
 680 }
 681
 682 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 683 {
 684         struct blkfront_info *info = rinfo->dev_info;
 685         struct blkif_request *ring_req, *extra_ring_req = NULL;
 686         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 687         bool require_extra_req = false;
 688         int i;
 689         struct setup_rw_req setup = {
 690                 .grant_idx = 0,
 691                 .segments = NULL,
 692                 .rinfo = rinfo,
 693                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 694         };
 695
 696         /*
 697          * Used to store if we are able to queue the request by just using
 698          * existing persistent grants, or if we have to get new grants,
 699          * as there are not sufficiently many free.
 700          */
 701         struct scatterlist *sg;
 702         int num_sg, max_grefs, num_grant;
 703
 704         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 705         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 706                 /*
 707                  * If we are using indirect segments we need to account
 708                  * for the indirect grefs used in the request.
 709                  */
 710                 max_grefs += INDIRECT_GREFS(max_grefs);
 711
 712         /*
 713          * We have to reserve 'max_grefs' grants because persistent
 714          * grants are shared by all rings.
 715          */
 716         if (max_grefs > 0)
 717                 if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
 718                         gnttab_request_free_callback(
 719                                 &rinfo->callback,
 720                                 blkif_restart_queue_callback,
 721                                 rinfo,
 722                                 max_grefs);
 723                         return 1;
 724                 }
 725
 726         /* Fill out a communications ring structure. */
 727         id = blkif_ring_get_request(rinfo, req, &ring_req);
 728
 729         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 730         num_grant = 0;
 731         /* Calculate the number of grant used */
 732         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 733                num_grant += gnttab_count_grant(sg->offset, sg->length);
 734
 735         require_extra_req = info->max_indirect_segments == 0 &&
 736                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 737         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 738
 739         rinfo->shadow[id].num_sg = num_sg;
 740         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 741             likely(!require_extra_req)) {
 742                 /*
 743                  * The indirect operation can only be a BLKIF_OP_READ or
 744                  * BLKIF_OP_WRITE
 745                  */
 746                 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
 747                 ring_req->operation = BLKIF_OP_INDIRECT;
 748                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 749                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 750                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 751                 ring_req->u.indirect.handle = info->handle;
 752                 ring_req->u.indirect.nr_segments = num_grant;
 753         } else {
 754                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 755                 ring_req->u.rw.handle = info->handle;
 756                 ring_req->operation = rq_data_dir(req) ?
 757                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 758                 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 759                         /*
 760                          * Ideally we can do an unordered flush-to-disk.
 761                          * In case the backend onlysupports barriers, use that.
 762                          * A barrier request a superset of FUA, so we can
 763                          * implement it the same way.  (It's also a FLUSH+FUA,
 764                          * since it is guaranteed ordered WRT previous writes.)
 765                          */
 766                         switch (info->feature_flush &
 767                                 ((REQ_FLUSH|REQ_FUA))) {
 768                         case REQ_FLUSH|REQ_FUA:
 769                                 ring_req->operation =
 770                                         BLKIF_OP_WRITE_BARRIER;
 771                                 break;
 772                         case REQ_FLUSH:
 773                                 ring_req->operation =
 774                                         BLKIF_OP_FLUSH_DISKCACHE;
 775                                 break;
 776                         default:
 777                                 ring_req->operation = 0;
 778                         }
 779                 }
 780                 ring_req->u.rw.nr_segments = num_grant;
 781                 if (unlikely(require_extra_req)) {
 782                         extra_id = blkif_ring_get_request(rinfo, req,
 783                                                           &extra_ring_req);
 784                         /*
 785                          * Only the first request contains the scatter-gather
 786                          * list.
 787                          */
 788                         rinfo->shadow[extra_id].num_sg = 0;
 789
 790                         blkif_setup_extra_req(ring_req, extra_ring_req);
 791
 792                         /* Link the 2 requests together */
 793                         rinfo->shadow[extra_id].associated_id = id;
 794                         rinfo->shadow[id].associated_id = extra_id;
 795                 }
 796         }
 797
 798         setup.ring_req = ring_req;
 799         setup.id = id;
 800
 801         setup.require_extra_req = require_extra_req;
 802         if (unlikely(require_extra_req))
 803                 setup.extra_ring_req = extra_ring_req;
 804
 805         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 806                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 807
 808                 if (setup.need_copy) {
 809                         setup.bvec_off = sg->offset;
 810                         setup.bvec_data = kmap_atomic(sg_page(sg));
 811                 }
 812
 813                 gnttab_foreach_grant_in_range(sg_page(sg),
 814                                               sg->offset,
 815                                               sg->length,
 816                                               blkif_setup_rw_req_grant,
 817                                               &setup);
 818
 819                 if (setup.need_copy)
 820                         kunmap_atomic(setup.bvec_data);
 821         }
 822         if (setup.segments)
 823                 kunmap_atomic(setup.segments);
 824
 825         /* Keep a private copy so we can reissue requests when recovering. */
 826         rinfo->shadow[id].req = *ring_req;
 827         if (unlikely(require_extra_req))
 828                 rinfo->shadow[extra_id].req = *extra_ring_req;
 829
 830         if (max_grefs > 0)
 831                 gnttab_free_grant_references(setup.gref_head);
 832
 833         return 0;
 834 }
 835
 836 /*
 837  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 838  * and writes are handled as expected.
 839  *
 840  * @req: a request struct
 841  */
 842 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 843 {
 844         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 845                 return 1;
 846
 847         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE)))
 848                 return blkif_queue_discard_req(req, rinfo);
 849         else
 850                 return blkif_queue_rw_req(req, rinfo);
 851 }
 852
 853 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 854 {
 855         int notify;
 856
 857         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 858
 859         if (notify)
 860                 notify_remote_via_irq(rinfo->irq);
 861 }
 862
 863 static inline bool blkif_request_flush_invalid(struct request *req,
 864                                                struct blkfront_info *info)
 865 {
 866         return ((req->cmd_type != REQ_TYPE_FS) ||
 867                 ((req->cmd_flags & REQ_FLUSH) &&
 868                  !(info->feature_flush & REQ_FLUSH)) ||
 869                 ((req->cmd_flags & REQ_FUA) &&
 870                  !(info->feature_flush & REQ_FUA)));
 871 }
 872
 873 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 874                           const struct blk_mq_queue_data *qd)
 875 {
 876         unsigned long flags;
 877         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)hctx->driver_data;
 878
 879         blk_mq_start_request(qd->rq);
 880         spin_lock_irqsave(&rinfo->ring_lock, flags);
 881         if (RING_FULL(&rinfo->ring))
 882                 goto out_busy;
 883
 884         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 885                 goto out_err;
 886
 887         if (blkif_queue_request(qd->rq, rinfo))
 888                 goto out_busy;
 889
 890         flush_requests(rinfo);
 891         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 892         return BLK_MQ_RQ_QUEUE_OK;
 893
 894 out_err:
 895         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 896         return BLK_MQ_RQ_QUEUE_ERROR;
 897
 898 out_busy:
 899         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 900         blk_mq_stop_hw_queue(hctx);
 901         return BLK_MQ_RQ_QUEUE_BUSY;
 902 }
 903
 904 static int blk_mq_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
 905                             unsigned int index)
 906 {
 907         struct blkfront_info *info = (struct blkfront_info *)data;
 908
 909         BUG_ON(info->nr_rings <= index);
 910         hctx->driver_data = &info->rinfo[index];
 911         return 0;
 912 }
 913
 914 static struct blk_mq_ops blkfront_mq_ops = {
 915         .queue_rq = blkif_queue_rq,
 916         .map_queue = blk_mq_map_queue,
 917         .init_hctx = blk_mq_init_hctx,
 918 };
 919
 920 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 921                                 unsigned int physical_sector_size,
 922                                 unsigned int segments)
 923 {
 924         struct request_queue *rq;
 925         struct blkfront_info *info = gd->private_data;
 926
 927         memset(&info->tag_set, 0, sizeof(info->tag_set));
 928         info->tag_set.ops = &blkfront_mq_ops;
 929         info->tag_set.nr_hw_queues = info->nr_rings;
 930         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
 931                 /*
 932                  * When indirect descriptior is not supported, the I/O request
 933                  * will be split between multiple request in the ring.
 934                  * To avoid problems when sending the request, divide by
 935                  * 2 the depth of the queue.
 936                  */
 937                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
 938         } else
 939                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
 940         info->tag_set.numa_node = NUMA_NO_NODE;
 941         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
 942         info->tag_set.cmd_size = 0;
 943         info->tag_set.driver_data = info;
 944
 945         if (blk_mq_alloc_tag_set(&info->tag_set))
 946                 return -EINVAL;
 947         rq = blk_mq_init_queue(&info->tag_set);
 948         if (IS_ERR(rq)) {
 949                 blk_mq_free_tag_set(&info->tag_set);
 950                 return PTR_ERR(rq);
 951         }
 952
 953         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 954
 955         if (info->feature_discard) {
 956                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 957                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 958                 rq->limits.discard_granularity = info->discard_granularity;
 959                 rq->limits.discard_alignment = info->discard_alignment;
 960                 if (info->feature_secdiscard)
 961                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 962         }
 963
 964         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 965         blk_queue_logical_block_size(rq, sector_size);
 966         blk_queue_physical_block_size(rq, physical_sector_size);
 967         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 968
 969         /* Each segment in a request is up to an aligned page in size. */
 970         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 971         blk_queue_max_segment_size(rq, PAGE_SIZE);
 972
 973         /* Ensure a merged request will fit in a single I/O ring slot. */
 974         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 975
 976         /* Make sure buffer addresses are sector-aligned. */
 977         blk_queue_dma_alignment(rq, 511);
 978
 979         /* Make sure we don't use bounce buffers. */
 980         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 981
 982         gd->queue = rq;
 983
 984         return 0;
 985 }
 986
 987 static const char *flush_info(unsigned int feature_flush)
 988 {
 989         switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
 990         case REQ_FLUSH|REQ_FUA:
 991                 return "barrier: enabled;";
 992         case REQ_FLUSH:
 993                 return "flush diskcache: enabled;";
 994         default:
 995                 return "barrier or flush: disabled;";
 996         }
 997 }
 998
 999 static void xlvbd_flush(struct blkfront_info *info)
1000 {
1001         blk_queue_write_cache(info->rq, info->feature_flush & REQ_FLUSH,
1002                                 info->feature_flush & REQ_FUA);
1003         pr_info("blkfront: %s: %s %s %s %s %s\n",
1004                 info->gd->disk_name, flush_info(info->feature_flush),
1005                 "persistent grants:", info->feature_persistent ?
1006                 "enabled;" : "disabled;", "indirect descriptors:",
1007                 info->max_indirect_segments ? "enabled;" : "disabled;");
1008 }
1009
1010 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1011 {
1012         int major;
1013         major = BLKIF_MAJOR(vdevice);
1014         *minor = BLKIF_MINOR(vdevice);
1015         switch (major) {
1016                 case XEN_IDE0_MAJOR:
1017                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1018                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1019                                 EMULATED_HD_DISK_MINOR_OFFSET;
1020                         break;
1021                 case XEN_IDE1_MAJOR:
1022                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1023                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1024                                 EMULATED_HD_DISK_MINOR_OFFSET;
1025                         break;
1026                 case XEN_SCSI_DISK0_MAJOR:
1027                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1028                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1029                         break;
1030                 case XEN_SCSI_DISK1_MAJOR:
1031                 case XEN_SCSI_DISK2_MAJOR:
1032                 case XEN_SCSI_DISK3_MAJOR:
1033                 case XEN_SCSI_DISK4_MAJOR:
1034                 case XEN_SCSI_DISK5_MAJOR:
1035                 case XEN_SCSI_DISK6_MAJOR:
1036                 case XEN_SCSI_DISK7_MAJOR:
1037                         *offset = (*minor / PARTS_PER_DISK) +
1038                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1039                                 EMULATED_SD_DISK_NAME_OFFSET;
1040                         *minor = *minor +
1041                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1042                                 EMULATED_SD_DISK_MINOR_OFFSET;
1043                         break;
1044                 case XEN_SCSI_DISK8_MAJOR:
1045                 case XEN_SCSI_DISK9_MAJOR:
1046                 case XEN_SCSI_DISK10_MAJOR:
1047                 case XEN_SCSI_DISK11_MAJOR:
1048                 case XEN_SCSI_DISK12_MAJOR:
1049                 case XEN_SCSI_DISK13_MAJOR:
1050                 case XEN_SCSI_DISK14_MAJOR:
1051                 case XEN_SCSI_DISK15_MAJOR:
1052                         *offset = (*minor / PARTS_PER_DISK) +
1053                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1054                                 EMULATED_SD_DISK_NAME_OFFSET;
1055                         *minor = *minor +
1056                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1057                                 EMULATED_SD_DISK_MINOR_OFFSET;
1058                         break;
1059                 case XENVBD_MAJOR:
1060                         *offset = *minor / PARTS_PER_DISK;
1061                         break;
1062                 default:
1063                         printk(KERN_WARNING "blkfront: your disk configuration is "
1064                                         "incorrect, please use an xvd device instead\n");
1065                         return -ENODEV;
1066         }
1067         return 0;
1068 }
1069
1070 static char *encode_disk_name(char *ptr, unsigned int n)
1071 {
1072         if (n >= 26)
1073                 ptr = encode_disk_name(ptr, n / 26 - 1);
1074         *ptr = 'a' + n % 26;
1075         return ptr + 1;
1076 }
1077
1078 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1079                                struct blkfront_info *info,
1080                                u16 vdisk_info, u16 sector_size,
1081                                unsigned int physical_sector_size)
1082 {
1083         struct gendisk *gd;
1084         int nr_minors = 1;
1085         int err;
1086         unsigned int offset;
1087         int minor;
1088         int nr_parts;
1089         char *ptr;
1090
1091         BUG_ON(info->gd != NULL);
1092         BUG_ON(info->rq != NULL);
1093
1094         if ((info->vdevice>>EXT_SHIFT) > 1) {
1095                 /* this is above the extended range; something is wrong */
1096                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1097                 return -ENODEV;
1098         }
1099
1100         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1101                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1102                 if (err)
1103                         return err;
1104                 nr_parts = PARTS_PER_DISK;
1105         } else {
1106                 minor = BLKIF_MINOR_EXT(info->vdevice);
1107                 nr_parts = PARTS_PER_EXT_DISK;
1108                 offset = minor / nr_parts;
1109                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1110                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1111                                         "emulated IDE disks,\n\t choose an xvd device name"
1112                                         "from xvde on\n", info->vdevice);
1113         }
1114         if (minor >> MINORBITS) {
1115                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1116                         info->vdevice, minor);
1117                 return -ENODEV;
1118         }
1119
1120         if ((minor % nr_parts) == 0)
1121                 nr_minors = nr_parts;
1122
1123         err = xlbd_reserve_minors(minor, nr_minors);
1124         if (err)
1125                 goto out;
1126         err = -ENODEV;
1127
1128         gd = alloc_disk(nr_minors);
1129         if (gd == NULL)
1130                 goto release;
1131
1132         strcpy(gd->disk_name, DEV_NAME);
1133         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1134         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1135         if (nr_minors > 1)
1136                 *ptr = 0;
1137         else
1138                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1139                          "%d", minor & (nr_parts - 1));
1140
1141         gd->major = XENVBD_MAJOR;
1142         gd->first_minor = minor;
1143         gd->fops = &xlvbd_block_fops;
1144         gd->private_data = info;
1145         gd->driverfs_dev = &(info->xbdev->dev);
1146         set_capacity(gd, capacity);
1147
1148         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
1149                                  info->max_indirect_segments ? :
1150                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
1151                 del_gendisk(gd);
1152                 goto release;
1153         }
1154
1155         info->rq = gd->queue;
1156         info->gd = gd;
1157
1158         xlvbd_flush(info);
1159
1160         if (vdisk_info & VDISK_READONLY)
1161                 set_disk_ro(gd, 1);
1162
1163         if (vdisk_info & VDISK_REMOVABLE)
1164                 gd->flags |= GENHD_FL_REMOVABLE;
1165
1166         if (vdisk_info & VDISK_CDROM)
1167                 gd->flags |= GENHD_FL_CD;
1168
1169         return 0;
1170
1171  release:
1172         xlbd_release_minors(minor, nr_minors);
1173  out:
1174         return err;
1175 }
1176
1177 static void xlvbd_release_gendisk(struct blkfront_info *info)
1178 {
1179         unsigned int minor, nr_minors, i;
1180
1181         if (info->rq == NULL)
1182                 return;
1183
1184         /* No more blkif_request(). */
1185         blk_mq_stop_hw_queues(info->rq);
1186
1187         for (i = 0; i < info->nr_rings; i++) {
1188                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1189
1190                 /* No more gnttab callback work. */
1191                 gnttab_cancel_free_callback(&rinfo->callback);
1192
1193                 /* Flush gnttab callback work. Must be done with no locks held. */
1194                 flush_work(&rinfo->work);
1195         }
1196
1197         del_gendisk(info->gd);
1198
1199         minor = info->gd->first_minor;
1200         nr_minors = info->gd->minors;
1201         xlbd_release_minors(minor, nr_minors);
1202
1203         blk_cleanup_queue(info->rq);
1204         blk_mq_free_tag_set(&info->tag_set);
1205         info->rq = NULL;
1206
1207         put_disk(info->gd);
1208         info->gd = NULL;
1209 }
1210
1211 /* Already hold rinfo->ring_lock. */
1212 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1213 {
1214         if (!RING_FULL(&rinfo->ring))
1215                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1216 }
1217
1218 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1219 {
1220         unsigned long flags;
1221
1222         spin_lock_irqsave(&rinfo->ring_lock, flags);
1223         kick_pending_request_queues_locked(rinfo);
1224         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1225 }
1226
1227 static void blkif_restart_queue(struct work_struct *work)
1228 {
1229         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1230
1231         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1232                 kick_pending_request_queues(rinfo);
1233 }
1234
1235 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1236 {
1237         struct grant *persistent_gnt, *n;
1238         struct blkfront_info *info = rinfo->dev_info;
1239         int i, j, segs;
1240
1241         /*
1242          * Remove indirect pages, this only happens when using indirect
1243          * descriptors but not persistent grants
1244          */
1245         if (!list_empty(&rinfo->indirect_pages)) {
1246                 struct page *indirect_page, *n;
1247
1248                 BUG_ON(info->feature_persistent);
1249                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1250                         list_del(&indirect_page->lru);
1251                         __free_page(indirect_page);
1252                 }
1253         }
1254
1255         /* Remove all persistent grants. */
1256         if (!list_empty(&rinfo->grants)) {
1257                 list_for_each_entry_safe(persistent_gnt, n,
1258                                          &rinfo->grants, node) {
1259                         list_del(&persistent_gnt->node);
1260                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1261                                 gnttab_end_foreign_access(persistent_gnt->gref,
1262                                                           0, 0UL);
1263                                 rinfo->persistent_gnts_c--;
1264                         }
1265                         if (info->feature_persistent)
1266                                 __free_page(persistent_gnt->page);
1267                         kfree(persistent_gnt);
1268                 }
1269         }
1270         BUG_ON(rinfo->persistent_gnts_c != 0);
1271
1272         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1273                 /*
1274                  * Clear persistent grants present in requests already
1275                  * on the shared ring
1276                  */
1277                 if (!rinfo->shadow[i].request)
1278                         goto free_shadow;
1279
1280                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1281                        rinfo->shadow[i].req.u.indirect.nr_segments :
1282                        rinfo->shadow[i].req.u.rw.nr_segments;
1283                 for (j = 0; j < segs; j++) {
1284                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1285                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1286                         if (info->feature_persistent)
1287                                 __free_page(persistent_gnt->page);
1288                         kfree(persistent_gnt);
1289                 }
1290
1291                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1292                         /*
1293                          * If this is not an indirect operation don't try to
1294                          * free indirect segments
1295                          */
1296                         goto free_shadow;
1297
1298                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1299                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1300                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1301                         __free_page(persistent_gnt->page);
1302                         kfree(persistent_gnt);
1303                 }
1304
1305 free_shadow:
1306                 kfree(rinfo->shadow[i].grants_used);
1307                 rinfo->shadow[i].grants_used = NULL;
1308                 kfree(rinfo->shadow[i].indirect_grants);
1309                 rinfo->shadow[i].indirect_grants = NULL;
1310                 kfree(rinfo->shadow[i].sg);
1311                 rinfo->shadow[i].sg = NULL;
1312         }
1313
1314         /* No more gnttab callback work. */
1315         gnttab_cancel_free_callback(&rinfo->callback);
1316
1317         /* Flush gnttab callback work. Must be done with no locks held. */
1318         flush_work(&rinfo->work);
1319
1320         /* Free resources associated with old device channel. */
1321         for (i = 0; i < info->nr_ring_pages; i++) {
1322                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1323                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1324                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1325                 }
1326         }
1327         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1328         rinfo->ring.sring = NULL;
1329
1330         if (rinfo->irq)
1331                 unbind_from_irqhandler(rinfo->irq, rinfo);
1332         rinfo->evtchn = rinfo->irq = 0;
1333 }
1334
1335 static void blkif_free(struct blkfront_info *info, int suspend)
1336 {
1337         unsigned int i;
1338
1339         /* Prevent new requests being issued until we fix things up. */
1340         info->connected = suspend ?
1341                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1342         /* No more blkif_request(). */
1343         if (info->rq)
1344                 blk_mq_stop_hw_queues(info->rq);
1345
1346         for (i = 0; i < info->nr_rings; i++)
1347                 blkif_free_ring(&info->rinfo[i]);
1348
1349         kfree(info->rinfo);
1350         info->rinfo = NULL;
1351         info->nr_rings = 0;
1352 }
1353
1354 struct copy_from_grant {
1355         const struct blk_shadow *s;
1356         unsigned int grant_idx;
1357         unsigned int bvec_offset;
1358         char *bvec_data;
1359 };
1360
1361 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1362                                   unsigned int len, void *data)
1363 {
1364         struct copy_from_grant *info = data;
1365         char *shared_data;
1366         /* Convenient aliases */
1367         const struct blk_shadow *s = info->s;
1368
1369         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1370
1371         memcpy(info->bvec_data + info->bvec_offset,
1372                shared_data + offset, len);
1373
1374         info->bvec_offset += len;
1375         info->grant_idx++;
1376
1377         kunmap_atomic(shared_data);
1378 }
1379
1380 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1381 {
1382         switch (rsp)
1383         {
1384         case BLKIF_RSP_OKAY:
1385                 return REQ_DONE;
1386         case BLKIF_RSP_EOPNOTSUPP:
1387                 return REQ_EOPNOTSUPP;
1388         case BLKIF_RSP_ERROR:
1389                 /* Fallthrough. */
1390         default:
1391                 return REQ_ERROR;
1392         }
1393 }
1394
1395 /*
1396  * Get the final status of the block request based on two ring response
1397  */
1398 static int blkif_get_final_status(enum blk_req_status s1,
1399                                   enum blk_req_status s2)
1400 {
1401         BUG_ON(s1 == REQ_WAITING);
1402         BUG_ON(s2 == REQ_WAITING);
1403
1404         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1405                 return BLKIF_RSP_ERROR;
1406         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1407                 return BLKIF_RSP_EOPNOTSUPP;
1408         return BLKIF_RSP_OKAY;
1409 }
1410
1411 static bool blkif_completion(unsigned long *id,
1412                              struct blkfront_ring_info *rinfo,
1413                              struct blkif_response *bret)
1414 {
1415         int i = 0;
1416         struct scatterlist *sg;
1417         int num_sg, num_grant;
1418         struct blkfront_info *info = rinfo->dev_info;
1419         struct blk_shadow *s = &rinfo->shadow[*id];
1420         struct copy_from_grant data = {
1421                 .grant_idx = 0,
1422         };
1423
1424         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1425                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1426
1427         /* The I/O request may be split in two. */
1428         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1429                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1430
1431                 /* Keep the status of the current response in shadow. */
1432                 s->status = blkif_rsp_to_req_status(bret->status);
1433
1434                 /* Wait the second response if not yet here. */
1435                 if (s2->status == REQ_WAITING)
1436                         return 0;
1437
1438                 bret->status = blkif_get_final_status(s->status,
1439                                                       s2->status);
1440
1441                 /*
1442                  * All the grants is stored in the first shadow in order
1443                  * to make the completion code simpler.
1444                  */
1445                 num_grant += s2->req.u.rw.nr_segments;
1446
1447                 /*
1448                  * The two responses may not come in order. Only the
1449                  * first request will store the scatter-gather list.
1450                  */
1451                 if (s2->num_sg != 0) {
1452                         /* Update "id" with the ID of the first response. */
1453                         *id = s->associated_id;
1454                         s = s2;
1455                 }
1456
1457                 /*
1458                  * We don't need anymore the second request, so recycling
1459                  * it now.
1460                  */
1461                 if (add_id_to_freelist(rinfo, s->associated_id))
1462                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1463                              info->gd->disk_name, s->associated_id);
1464         }
1465
1466         data.s = s;
1467         num_sg = s->num_sg;
1468
1469         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1470                 for_each_sg(s->sg, sg, num_sg, i) {
1471                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1472
1473                         data.bvec_offset = sg->offset;
1474                         data.bvec_data = kmap_atomic(sg_page(sg));
1475
1476                         gnttab_foreach_grant_in_range(sg_page(sg),
1477                                                       sg->offset,
1478                                                       sg->length,
1479                                                       blkif_copy_from_grant,
1480                                                       &data);
1481
1482                         kunmap_atomic(data.bvec_data);
1483                 }
1484         }
1485         /* Add the persistent grant into the list of free grants */
1486         for (i = 0; i < num_grant; i++) {
1487                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1488                         /*
1489                          * If the grant is still mapped by the backend (the
1490                          * backend has chosen to make this grant persistent)
1491                          * we add it at the head of the list, so it will be
1492                          * reused first.
1493                          */
1494                         if (!info->feature_persistent)
1495                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1496                                                      s->grants_used[i]->gref);
1497                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1498                         rinfo->persistent_gnts_c++;
1499                 } else {
1500                         /*
1501                          * If the grant is not mapped by the backend we end the
1502                          * foreign access and add it to the tail of the list,
1503                          * so it will not be picked again unless we run out of
1504                          * persistent grants.
1505                          */
1506                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1507                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1508                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1509                 }
1510         }
1511         if (s->req.operation == BLKIF_OP_INDIRECT) {
1512                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1513                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1514                                 if (!info->feature_persistent)
1515                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1516                                                              s->indirect_grants[i]->gref);
1517                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1518                                 rinfo->persistent_gnts_c++;
1519                         } else {
1520                                 struct page *indirect_page;
1521
1522                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1523                                 /*
1524                                  * Add the used indirect page back to the list of
1525                                  * available pages for indirect grefs.
1526                                  */
1527                                 if (!info->feature_persistent) {
1528                                         indirect_page = s->indirect_grants[i]->page;
1529                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1530                                 }
1531                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1532                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1533                         }
1534                 }
1535         }
1536
1537         return 1;
1538 }
1539
1540 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1541 {
1542         struct request *req;
1543         struct blkif_response *bret;
1544         RING_IDX i, rp;
1545         unsigned long flags;
1546         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1547         struct blkfront_info *info = rinfo->dev_info;
1548         int error;
1549
1550         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1551                 return IRQ_HANDLED;
1552
1553         spin_lock_irqsave(&rinfo->ring_lock, flags);
1554  again:
1555         rp = rinfo->ring.sring->rsp_prod;
1556         rmb(); /* Ensure we see queued responses up to 'rp'. */
1557
1558         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1559                 unsigned long id;
1560
1561                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1562                 id   = bret->id;
1563                 /*
1564                  * The backend has messed up and given us an id that we would
1565                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1566                  * look in get_id_from_freelist.
1567                  */
1568                 if (id >= BLK_RING_SIZE(info)) {
1569                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1570                              info->gd->disk_name, op_name(bret->operation), id);
1571                         /* We can't safely get the 'struct request' as
1572                          * the id is busted. */
1573                         continue;
1574                 }
1575                 req  = rinfo->shadow[id].request;
1576
1577                 if (bret->operation != BLKIF_OP_DISCARD) {
1578                         /*
1579                          * We may need to wait for an extra response if the
1580                          * I/O request is split in 2
1581                          */
1582                         if (!blkif_completion(&id, rinfo, bret))
1583                                 continue;
1584                 }
1585
1586                 if (add_id_to_freelist(rinfo, id)) {
1587                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1588                              info->gd->disk_name, op_name(bret->operation), id);
1589                         continue;
1590                 }
1591
1592                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1593                 switch (bret->operation) {
1594                 case BLKIF_OP_DISCARD:
1595                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1596                                 struct request_queue *rq = info->rq;
1597                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1598                                            info->gd->disk_name, op_name(bret->operation));
1599                                 error = -EOPNOTSUPP;
1600                                 info->feature_discard = 0;
1601                                 info->feature_secdiscard = 0;
1602                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1603                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1604                         }
1605                         blk_mq_complete_request(req, error);
1606                         break;
1607                 case BLKIF_OP_FLUSH_DISKCACHE:
1608                 case BLKIF_OP_WRITE_BARRIER:
1609                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1610                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1611                                        info->gd->disk_name, op_name(bret->operation));
1612                                 error = -EOPNOTSUPP;
1613                         }
1614                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1615                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1616                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1617                                        info->gd->disk_name, op_name(bret->operation));
1618                                 error = -EOPNOTSUPP;
1619                         }
1620                         if (unlikely(error)) {
1621                                 if (error == -EOPNOTSUPP)
1622                                         error = 0;
1623                                 info->feature_flush = 0;
1624                                 xlvbd_flush(info);
1625                         }
1626                         /* fall through */
1627                 case BLKIF_OP_READ:
1628                 case BLKIF_OP_WRITE:
1629                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1630                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1631                                         "request: %x\n", bret->status);
1632
1633                         blk_mq_complete_request(req, error);
1634                         break;
1635                 default:
1636                         BUG();
1637                 }
1638         }
1639
1640         rinfo->ring.rsp_cons = i;
1641
1642         if (i != rinfo->ring.req_prod_pvt) {
1643                 int more_to_do;
1644                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1645                 if (more_to_do)
1646                         goto again;
1647         } else
1648                 rinfo->ring.sring->rsp_event = i + 1;
1649
1650         kick_pending_request_queues_locked(rinfo);
1651
1652         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1653
1654         return IRQ_HANDLED;
1655 }
1656
1657
1658 static int setup_blkring(struct xenbus_device *dev,
1659                          struct blkfront_ring_info *rinfo)
1660 {
1661         struct blkif_sring *sring;
1662         int err, i;
1663         struct blkfront_info *info = rinfo->dev_info;
1664         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1665         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1666
1667         for (i = 0; i < info->nr_ring_pages; i++)
1668                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1669
1670         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1671                                                        get_order(ring_size));
1672         if (!sring) {
1673                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1674                 return -ENOMEM;
1675         }
1676         SHARED_RING_INIT(sring);
1677         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1678
1679         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1680         if (err < 0) {
1681                 free_pages((unsigned long)sring, get_order(ring_size));
1682                 rinfo->ring.sring = NULL;
1683                 goto fail;
1684         }
1685         for (i = 0; i < info->nr_ring_pages; i++)
1686                 rinfo->ring_ref[i] = gref[i];
1687
1688         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1689         if (err)
1690                 goto fail;
1691
1692         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1693                                         "blkif", rinfo);
1694         if (err <= 0) {
1695                 xenbus_dev_fatal(dev, err,
1696                                  "bind_evtchn_to_irqhandler failed");
1697                 goto fail;
1698         }
1699         rinfo->irq = err;
1700
1701         return 0;
1702 fail:
1703         blkif_free(info, 0);
1704         return err;
1705 }
1706
1707 /*
1708  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1709  * ring buffer may have multi pages depending on ->nr_ring_pages.
1710  */
1711 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1712                                 struct blkfront_ring_info *rinfo, const char *dir)
1713 {
1714         int err;
1715         unsigned int i;
1716         const char *message = NULL;
1717         struct blkfront_info *info = rinfo->dev_info;
1718
1719         if (info->nr_ring_pages == 1) {
1720                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1721                 if (err) {
1722                         message = "writing ring-ref";
1723                         goto abort_transaction;
1724                 }
1725         } else {
1726                 for (i = 0; i < info->nr_ring_pages; i++) {
1727                         char ring_ref_name[RINGREF_NAME_LEN];
1728
1729                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1730                         err = xenbus_printf(xbt, dir, ring_ref_name,
1731                                             "%u", rinfo->ring_ref[i]);
1732                         if (err) {
1733                                 message = "writing ring-ref";
1734                                 goto abort_transaction;
1735                         }
1736                 }
1737         }
1738
1739         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1740         if (err) {
1741                 message = "writing event-channel";
1742                 goto abort_transaction;
1743         }
1744
1745         return 0;
1746
1747 abort_transaction:
1748         xenbus_transaction_end(xbt, 1);
1749         if (message)
1750                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1751
1752         return err;
1753 }
1754
1755 /* Common code used when first setting up, and when resuming. */
1756 static int talk_to_blkback(struct xenbus_device *dev,
1757                            struct blkfront_info *info)
1758 {
1759         const char *message = NULL;
1760         struct xenbus_transaction xbt;
1761         int err;
1762         unsigned int i, max_page_order = 0;
1763         unsigned int ring_page_order = 0;
1764
1765         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1766                            "max-ring-page-order", "%u", &max_page_order);
1767         if (err != 1)
1768                 info->nr_ring_pages = 1;
1769         else {
1770                 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1771                 info->nr_ring_pages = 1 << ring_page_order;
1772         }
1773
1774         for (i = 0; i < info->nr_rings; i++) {
1775                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1776
1777                 /* Create shared ring, alloc event channel. */
1778                 err = setup_blkring(dev, rinfo);
1779                 if (err)
1780                         goto destroy_blkring;
1781         }
1782
1783 again:
1784         err = xenbus_transaction_start(&xbt);
1785         if (err) {
1786                 xenbus_dev_fatal(dev, err, "starting transaction");
1787                 goto destroy_blkring;
1788         }
1789
1790         if (info->nr_ring_pages > 1) {
1791                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1792                                     ring_page_order);
1793                 if (err) {
1794                         message = "writing ring-page-order";
1795                         goto abort_transaction;
1796                 }
1797         }
1798
1799         /* We already got the number of queues/rings in _probe */
1800         if (info->nr_rings == 1) {
1801                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1802                 if (err)
1803                         goto destroy_blkring;
1804         } else {
1805                 char *path;
1806                 size_t pathsize;
1807
1808                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1809                                     info->nr_rings);
1810                 if (err) {
1811                         message = "writing multi-queue-num-queues";
1812                         goto abort_transaction;
1813                 }
1814
1815                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1816                 path = kmalloc(pathsize, GFP_KERNEL);
1817                 if (!path) {
1818                         err = -ENOMEM;
1819                         message = "ENOMEM while writing ring references";
1820                         goto abort_transaction;
1821                 }
1822
1823                 for (i = 0; i < info->nr_rings; i++) {
1824                         memset(path, 0, pathsize);
1825                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1826                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1827                         if (err) {
1828                                 kfree(path);
1829                                 goto destroy_blkring;
1830                         }
1831                 }
1832                 kfree(path);
1833         }
1834         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1835                             XEN_IO_PROTO_ABI_NATIVE);
1836         if (err) {
1837                 message = "writing protocol";
1838                 goto abort_transaction;
1839         }
1840         err = xenbus_printf(xbt, dev->nodename,
1841                             "feature-persistent", "%u", 1);
1842         if (err)
1843                 dev_warn(&dev->dev,
1844                          "writing persistent grants feature to xenbus");
1845
1846         err = xenbus_transaction_end(xbt, 0);
1847         if (err) {
1848                 if (err == -EAGAIN)
1849                         goto again;
1850                 xenbus_dev_fatal(dev, err, "completing transaction");
1851                 goto destroy_blkring;
1852         }
1853
1854         for (i = 0; i < info->nr_rings; i++) {
1855                 unsigned int j;
1856                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1857
1858                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1859                         rinfo->shadow[j].req.u.rw.id = j + 1;
1860                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1861         }
1862         xenbus_switch_state(dev, XenbusStateInitialised);
1863
1864         return 0;
1865
1866  abort_transaction:
1867         xenbus_transaction_end(xbt, 1);
1868         if (message)
1869                 xenbus_dev_fatal(dev, err, "%s", message);
1870  destroy_blkring:
1871         blkif_free(info, 0);
1872
1873         kfree(info);
1874         dev_set_drvdata(&dev->dev, NULL);
1875
1876         return err;
1877 }
1878
1879 static int negotiate_mq(struct blkfront_info *info)
1880 {
1881         unsigned int backend_max_queues = 0;
1882         int err;
1883         unsigned int i;
1884
1885         BUG_ON(info->nr_rings);
1886
1887         /* Check if backend supports multiple queues. */
1888         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1889                            "multi-queue-max-queues", "%u", &backend_max_queues);
1890         if (err < 0)
1891                 backend_max_queues = 1;
1892
1893         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1894         /* We need at least one ring. */
1895         if (!info->nr_rings)
1896                 info->nr_rings = 1;
1897
1898         info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1899         if (!info->rinfo) {
1900                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1901                 return -ENOMEM;
1902         }
1903
1904         for (i = 0; i < info->nr_rings; i++) {
1905                 struct blkfront_ring_info *rinfo;
1906
1907                 rinfo = &info->rinfo[i];
1908                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1909                 INIT_LIST_HEAD(&rinfo->grants);
1910                 rinfo->dev_info = info;
1911                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1912                 spin_lock_init(&rinfo->ring_lock);
1913         }
1914         return 0;
1915 }
1916 /**
1917  * Entry point to this code when a new device is created.  Allocate the basic
1918  * structures and the ring buffer for communication with the backend, and
1919  * inform the backend of the appropriate details for those.  Switch to
1920  * Initialised state.
1921  */
1922 static int blkfront_probe(struct xenbus_device *dev,
1923                           const struct xenbus_device_id *id)
1924 {
1925         int err, vdevice;
1926         struct blkfront_info *info;
1927
1928         /* FIXME: Use dynamic device id if this is not set. */
1929         err = xenbus_scanf(XBT_NIL, dev->nodename,
1930                            "virtual-device", "%i", &vdevice);
1931         if (err != 1) {
1932                 /* go looking in the extended area instead */
1933                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1934                                    "%i", &vdevice);
1935                 if (err != 1) {
1936                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1937                         return err;
1938                 }
1939         }
1940
1941         if (xen_hvm_domain()) {
1942                 char *type;
1943                 int len;
1944                 /* no unplug has been done: do not hook devices != xen vbds */
1945                 if (xen_has_pv_and_legacy_disk_devices()) {
1946                         int major;
1947
1948                         if (!VDEV_IS_EXTENDED(vdevice))
1949                                 major = BLKIF_MAJOR(vdevice);
1950                         else
1951                                 major = XENVBD_MAJOR;
1952
1953                         if (major != XENVBD_MAJOR) {
1954                                 printk(KERN_INFO
1955                                                 "%s: HVM does not support vbd %d as xen block device\n",
1956                                                 __func__, vdevice);
1957                                 return -ENODEV;
1958                         }
1959                 }
1960                 /* do not create a PV cdrom device if we are an HVM guest */
1961                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1962                 if (IS_ERR(type))
1963                         return -ENODEV;
1964                 if (strncmp(type, "cdrom", 5) == 0) {
1965                         kfree(type);
1966                         return -ENODEV;
1967                 }
1968                 kfree(type);
1969         }
1970         info = kzalloc(sizeof(*info), GFP_KERNEL);
1971         if (!info) {
1972                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1973                 return -ENOMEM;
1974         }
1975
1976         info->xbdev = dev;
1977         err = negotiate_mq(info);
1978         if (err) {
1979                 kfree(info);
1980                 return err;
1981         }
1982
1983         mutex_init(&info->mutex);
1984         info->vdevice = vdevice;
1985         info->connected = BLKIF_STATE_DISCONNECTED;
1986
1987         /* Front end dir is a number, which is used as the id. */
1988         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1989         dev_set_drvdata(&dev->dev, info);
1990
1991         return 0;
1992 }
1993
1994 static void split_bio_end(struct bio *bio)
1995 {
1996         struct split_bio *split_bio = bio->bi_private;
1997
1998         if (atomic_dec_and_test(&split_bio->pending)) {
1999                 split_bio->bio->bi_phys_segments = 0;
2000                 split_bio->bio->bi_error = bio->bi_error;
2001                 bio_endio(split_bio->bio);
2002                 kfree(split_bio);
2003         }
2004         bio_put(bio);
2005 }
2006
2007 static int blkif_recover(struct blkfront_info *info)
2008 {
2009         unsigned int i, r_index;
2010         struct request *req, *n;
2011         struct blk_shadow *copy;
2012         int rc;
2013         struct bio *bio, *cloned_bio;
2014         struct bio_list bio_list, merge_bio;
2015         unsigned int segs, offset;
2016         int pending, size;
2017         struct split_bio *split_bio;
2018         struct list_head requests;
2019
2020         blkfront_gather_backend_features(info);
2021         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2022         blk_queue_max_segments(info->rq, segs);
2023         bio_list_init(&bio_list);
2024         INIT_LIST_HEAD(&requests);
2025
2026         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2027                 struct blkfront_ring_info *rinfo;
2028
2029                 rinfo = &info->rinfo[r_index];
2030                 /* Stage 1: Make a safe copy of the shadow state. */
2031                 copy = kmemdup(rinfo->shadow, sizeof(rinfo->shadow),
2032                                GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
2033                 if (!copy)
2034                         return -ENOMEM;
2035
2036                 /* Stage 2: Set up free list. */
2037                 memset(&rinfo->shadow, 0, sizeof(rinfo->shadow));
2038                 for (i = 0; i < BLK_RING_SIZE(info); i++)
2039                         rinfo->shadow[i].req.u.rw.id = i+1;
2040                 rinfo->shadow_free = rinfo->ring.req_prod_pvt;
2041                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
2042
2043                 rc = blkfront_setup_indirect(rinfo);
2044                 if (rc) {
2045                         kfree(copy);
2046                         return rc;
2047                 }
2048
2049                 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2050                         /* Not in use? */
2051                         if (!copy[i].request)
2052                                 continue;
2053
2054                         /*
2055                          * Get the bios in the request so we can re-queue them.
2056                          */
2057                         if (copy[i].request->cmd_flags &
2058                             (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
2059                                 /*
2060                                  * Flush operations don't contain bios, so
2061                                  * we need to requeue the whole request
2062                                  */
2063                                 list_add(&copy[i].request->queuelist, &requests);
2064                                 continue;
2065                         }
2066                         merge_bio.head = copy[i].request->bio;
2067                         merge_bio.tail = copy[i].request->biotail;
2068                         bio_list_merge(&bio_list, &merge_bio);
2069                         copy[i].request->bio = NULL;
2070                         blk_end_request_all(copy[i].request, 0);
2071                 }
2072
2073                 kfree(copy);
2074         }
2075         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2076
2077         /* Now safe for us to use the shared ring */
2078         info->connected = BLKIF_STATE_CONNECTED;
2079
2080         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2081                 struct blkfront_ring_info *rinfo;
2082
2083                 rinfo = &info->rinfo[r_index];
2084                 /* Kick any other new requests queued since we resumed */
2085                 kick_pending_request_queues(rinfo);
2086         }
2087
2088         list_for_each_entry_safe(req, n, &requests, queuelist) {
2089                 /* Requeue pending requests (flush or discard) */
2090                 list_del_init(&req->queuelist);
2091                 BUG_ON(req->nr_phys_segments > segs);
2092                 blk_mq_requeue_request(req);
2093         }
2094         blk_mq_kick_requeue_list(info->rq);
2095
2096         while ((bio = bio_list_pop(&bio_list)) != NULL) {
2097                 /* Traverse the list of pending bios and re-queue them */
2098                 if (bio_segments(bio) > segs) {
2099                         /*
2100                          * This bio has more segments than what we can
2101                          * handle, we have to split it.
2102                          */
2103                         pending = (bio_segments(bio) + segs - 1) / segs;
2104                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2105                         BUG_ON(split_bio == NULL);
2106                         atomic_set(&split_bio->pending, pending);
2107                         split_bio->bio = bio;
2108                         for (i = 0; i < pending; i++) {
2109                                 offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2110                                 size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2111                                            (unsigned int)bio_sectors(bio) - offset);
2112                                 cloned_bio = bio_clone(bio, GFP_NOIO);
2113                                 BUG_ON(cloned_bio == NULL);
2114                                 bio_trim(cloned_bio, offset, size);
2115                                 cloned_bio->bi_private = split_bio;
2116                                 cloned_bio->bi_end_io = split_bio_end;
2117                                 submit_bio(cloned_bio->bi_rw, cloned_bio);
2118                         }
2119                         /*
2120                          * Now we have to wait for all those smaller bios to
2121                          * end, so we can also end the "parent" bio.
2122                          */
2123                         continue;
2124                 }
2125                 /* We don't need to split this bio */
2126                 submit_bio(bio->bi_rw, bio);
2127         }
2128
2129         return 0;
2130 }
2131
2132 /**
2133  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2134  * driver restart.  We tear down our blkif structure and recreate it, but
2135  * leave the device-layer structures intact so that this is transparent to the
2136  * rest of the kernel.
2137  */
2138 static int blkfront_resume(struct xenbus_device *dev)
2139 {
2140         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2141         int err = 0;
2142
2143         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2144
2145         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2146
2147         err = negotiate_mq(info);
2148         if (err)
2149                 return err;
2150
2151         err = talk_to_blkback(dev, info);
2152
2153         /*
2154          * We have to wait for the backend to switch to
2155          * connected state, since we want to read which
2156          * features it supports.
2157          */
2158
2159         return err;
2160 }
2161
2162 static void blkfront_closing(struct blkfront_info *info)
2163 {
2164         struct xenbus_device *xbdev = info->xbdev;
2165         struct block_device *bdev = NULL;
2166
2167         mutex_lock(&info->mutex);
2168
2169         if (xbdev->state == XenbusStateClosing) {
2170                 mutex_unlock(&info->mutex);
2171                 return;
2172         }
2173
2174         if (info->gd)
2175                 bdev = bdget_disk(info->gd, 0);
2176
2177         mutex_unlock(&info->mutex);
2178
2179         if (!bdev) {
2180                 xenbus_frontend_closed(xbdev);
2181                 return;
2182         }
2183
2184         mutex_lock(&bdev->bd_mutex);
2185
2186         if (bdev->bd_openers) {
2187                 xenbus_dev_error(xbdev, -EBUSY,
2188                                  "Device in use; refusing to close");
2189                 xenbus_switch_state(xbdev, XenbusStateClosing);
2190         } else {
2191                 xlvbd_release_gendisk(info);
2192                 xenbus_frontend_closed(xbdev);
2193         }
2194
2195         mutex_unlock(&bdev->bd_mutex);
2196         bdput(bdev);
2197 }
2198
2199 static void blkfront_setup_discard(struct blkfront_info *info)
2200 {
2201         int err;
2202         unsigned int discard_granularity;
2203         unsigned int discard_alignment;
2204         unsigned int discard_secure;
2205
2206         info->feature_discard = 1;
2207         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2208                 "discard-granularity", "%u", &discard_granularity,
2209                 "discard-alignment", "%u", &discard_alignment,
2210                 NULL);
2211         if (!err) {
2212                 info->discard_granularity = discard_granularity;
2213                 info->discard_alignment = discard_alignment;
2214         }
2215         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2216                     "discard-secure", "%d", &discard_secure,
2217                     NULL);
2218         if (!err)
2219                 info->feature_secdiscard = !!discard_secure;
2220 }
2221
2222 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2223 {
2224         unsigned int psegs, grants;
2225         int err, i;
2226         struct blkfront_info *info = rinfo->dev_info;
2227
2228         if (info->max_indirect_segments == 0) {
2229                 if (!HAS_EXTRA_REQ)
2230                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2231                 else {
2232                         /*
2233                          * When an extra req is required, the maximum
2234                          * grants supported is related to the size of the
2235                          * Linux block segment.
2236                          */
2237                         grants = GRANTS_PER_PSEG;
2238                 }
2239         }
2240         else
2241                 grants = info->max_indirect_segments;
2242         psegs = grants / GRANTS_PER_PSEG;
2243
2244         err = fill_grant_buffer(rinfo,
2245                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2246         if (err)
2247                 goto out_of_memory;
2248
2249         if (!info->feature_persistent && info->max_indirect_segments) {
2250                 /*
2251                  * We are using indirect descriptors but not persistent
2252                  * grants, we need to allocate a set of pages that can be
2253                  * used for mapping indirect grefs
2254                  */
2255                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2256
2257                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2258                 for (i = 0; i < num; i++) {
2259                         struct page *indirect_page = alloc_page(GFP_NOIO);
2260                         if (!indirect_page)
2261                                 goto out_of_memory;
2262                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2263                 }
2264         }
2265
2266         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2267                 rinfo->shadow[i].grants_used = kzalloc(
2268                         sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2269                         GFP_NOIO);
2270                 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2271                 if (info->max_indirect_segments)
2272                         rinfo->shadow[i].indirect_grants = kzalloc(
2273                                 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2274                                 INDIRECT_GREFS(grants),
2275                                 GFP_NOIO);
2276                 if ((rinfo->shadow[i].grants_used == NULL) ||
2277                         (rinfo->shadow[i].sg == NULL) ||
2278                      (info->max_indirect_segments &&
2279                      (rinfo->shadow[i].indirect_grants == NULL)))
2280                         goto out_of_memory;
2281                 sg_init_table(rinfo->shadow[i].sg, psegs);
2282         }
2283
2284
2285         return 0;
2286
2287 out_of_memory:
2288         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2289                 kfree(rinfo->shadow[i].grants_used);
2290                 rinfo->shadow[i].grants_used = NULL;
2291                 kfree(rinfo->shadow[i].sg);
2292                 rinfo->shadow[i].sg = NULL;
2293                 kfree(rinfo->shadow[i].indirect_grants);
2294                 rinfo->shadow[i].indirect_grants = NULL;
2295         }
2296         if (!list_empty(&rinfo->indirect_pages)) {
2297                 struct page *indirect_page, *n;
2298                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2299                         list_del(&indirect_page->lru);
2300                         __free_page(indirect_page);
2301                 }
2302         }
2303         return -ENOMEM;
2304 }
2305
2306 /*
2307  * Gather all backend feature-*
2308  */
2309 static void blkfront_gather_backend_features(struct blkfront_info *info)
2310 {
2311         int err;
2312         int barrier, flush, discard, persistent;
2313         unsigned int indirect_segments;
2314
2315         info->feature_flush = 0;
2316
2317         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2318                         "feature-barrier", "%d", &barrier,
2319                         NULL);
2320
2321         /*
2322          * If there's no "feature-barrier" defined, then it means
2323          * we're dealing with a very old backend which writes
2324          * synchronously; nothing to do.
2325          *
2326          * If there are barriers, then we use flush.
2327          */
2328         if (!err && barrier)
2329                 info->feature_flush = REQ_FLUSH | REQ_FUA;
2330         /*
2331          * And if there is "feature-flush-cache" use that above
2332          * barriers.
2333          */
2334         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2335                         "feature-flush-cache", "%d", &flush,
2336                         NULL);
2337
2338         if (!err && flush)
2339                 info->feature_flush = REQ_FLUSH;
2340
2341         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2342                         "feature-discard", "%d", &discard,
2343                         NULL);
2344
2345         if (!err && discard)
2346                 blkfront_setup_discard(info);
2347
2348         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2349                         "feature-persistent", "%u", &persistent,
2350                         NULL);
2351         if (err)
2352                 info->feature_persistent = 0;
2353         else
2354                 info->feature_persistent = persistent;
2355
2356         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2357                             "feature-max-indirect-segments", "%u", &indirect_segments,
2358                             NULL);
2359         if (err)
2360                 info->max_indirect_segments = 0;
2361         else
2362                 info->max_indirect_segments = min(indirect_segments,
2363                                                   xen_blkif_max_segments);
2364 }
2365
2366 /*
2367  * Invoked when the backend is finally 'ready' (and has told produced
2368  * the details about the physical device - #sectors, size, etc).
2369  */
2370 static void blkfront_connect(struct blkfront_info *info)
2371 {
2372         unsigned long long sectors;
2373         unsigned long sector_size;
2374         unsigned int physical_sector_size;
2375         unsigned int binfo;
2376         int err, i;
2377
2378         switch (info->connected) {
2379         case BLKIF_STATE_CONNECTED:
2380                 /*
2381                  * Potentially, the back-end may be signalling
2382                  * a capacity change; update the capacity.
2383                  */
2384                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2385                                    "sectors", "%Lu", &sectors);
2386                 if (XENBUS_EXIST_ERR(err))
2387                         return;
2388                 printk(KERN_INFO "Setting capacity to %Lu\n",
2389                        sectors);
2390                 set_capacity(info->gd, sectors);
2391                 revalidate_disk(info->gd);
2392
2393                 return;
2394         case BLKIF_STATE_SUSPENDED:
2395                 /*
2396                  * If we are recovering from suspension, we need to wait
2397                  * for the backend to announce it's features before
2398                  * reconnecting, at least we need to know if the backend
2399                  * supports indirect descriptors, and how many.
2400                  */
2401                 blkif_recover(info);
2402                 return;
2403
2404         default:
2405                 break;
2406         }
2407
2408         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2409                 __func__, info->xbdev->otherend);
2410
2411         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2412                             "sectors", "%llu", &sectors,
2413                             "info", "%u", &binfo,
2414                             "sector-size", "%lu", &sector_size,
2415                             NULL);
2416         if (err) {
2417                 xenbus_dev_fatal(info->xbdev, err,
2418                                  "reading backend fields at %s",
2419                                  info->xbdev->otherend);
2420                 return;
2421         }
2422
2423         /*
2424          * physcial-sector-size is a newer field, so old backends may not
2425          * provide this. Assume physical sector size to be the same as
2426          * sector_size in that case.
2427          */
2428         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2429                            "physical-sector-size", "%u", &physical_sector_size);
2430         if (err != 1)
2431                 physical_sector_size = sector_size;
2432
2433         blkfront_gather_backend_features(info);
2434         for (i = 0; i < info->nr_rings; i++) {
2435                 err = blkfront_setup_indirect(&info->rinfo[i]);
2436                 if (err) {
2437                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2438                                          info->xbdev->otherend);
2439                         blkif_free(info, 0);
2440                         break;
2441                 }
2442         }
2443
2444         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2445                                   physical_sector_size);
2446         if (err) {
2447                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2448                                  info->xbdev->otherend);
2449                 return;
2450         }
2451
2452         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2453
2454         /* Kick pending requests. */
2455         info->connected = BLKIF_STATE_CONNECTED;
2456         for (i = 0; i < info->nr_rings; i++)
2457                 kick_pending_request_queues(&info->rinfo[i]);
2458
2459         add_disk(info->gd);
2460
2461         info->is_ready = 1;
2462 }
2463
2464 /**
2465  * Callback received when the backend's state changes.
2466  */
2467 static void blkback_changed(struct xenbus_device *dev,
2468                             enum xenbus_state backend_state)
2469 {
2470         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2471
2472         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2473
2474         switch (backend_state) {
2475         case XenbusStateInitWait:
2476                 if (dev->state != XenbusStateInitialising)
2477                         break;
2478                 if (talk_to_blkback(dev, info))
2479                         break;
2480         case XenbusStateInitialising:
2481         case XenbusStateInitialised:
2482         case XenbusStateReconfiguring:
2483         case XenbusStateReconfigured:
2484         case XenbusStateUnknown:
2485                 break;
2486
2487         case XenbusStateConnected:
2488                 if (dev->state != XenbusStateInitialised) {
2489                         if (talk_to_blkback(dev, info))
2490                                 break;
2491                 }
2492                 blkfront_connect(info);
2493                 break;
2494
2495         case XenbusStateClosed:
2496                 if (dev->state == XenbusStateClosed)
2497                         break;
2498                 /* Missed the backend's Closing state -- fallthrough */
2499         case XenbusStateClosing:
2500                 if (info)
2501                         blkfront_closing(info);
2502                 break;
2503         }
2504 }
2505
2506 static int blkfront_remove(struct xenbus_device *xbdev)
2507 {
2508         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2509         struct block_device *bdev = NULL;
2510         struct gendisk *disk;
2511
2512         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2513
2514         blkif_free(info, 0);
2515
2516         mutex_lock(&info->mutex);
2517
2518         disk = info->gd;
2519         if (disk)
2520                 bdev = bdget_disk(disk, 0);
2521
2522         info->xbdev = NULL;
2523         mutex_unlock(&info->mutex);
2524
2525         if (!bdev) {
2526                 kfree(info);
2527                 return 0;
2528         }
2529
2530         /*
2531          * The xbdev was removed before we reached the Closed
2532          * state. See if it's safe to remove the disk. If the bdev
2533          * isn't closed yet, we let release take care of it.
2534          */
2535
2536         mutex_lock(&bdev->bd_mutex);
2537         info = disk->private_data;
2538
2539         dev_warn(disk_to_dev(disk),
2540                  "%s was hot-unplugged, %d stale handles\n",
2541                  xbdev->nodename, bdev->bd_openers);
2542
2543         if (info && !bdev->bd_openers) {
2544                 xlvbd_release_gendisk(info);
2545                 disk->private_data = NULL;
2546                 kfree(info);
2547         }
2548
2549         mutex_unlock(&bdev->bd_mutex);
2550         bdput(bdev);
2551
2552         return 0;
2553 }
2554
2555 static int blkfront_is_ready(struct xenbus_device *dev)
2556 {
2557         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2558
2559         return info->is_ready && info->xbdev;
2560 }
2561
2562 static int blkif_open(struct block_device *bdev, fmode_t mode)
2563 {
2564         struct gendisk *disk = bdev->bd_disk;
2565         struct blkfront_info *info;
2566         int err = 0;
2567
2568         mutex_lock(&blkfront_mutex);
2569
2570         info = disk->private_data;
2571         if (!info) {
2572                 /* xbdev gone */
2573                 err = -ERESTARTSYS;
2574                 goto out;
2575         }
2576
2577         mutex_lock(&info->mutex);
2578
2579         if (!info->gd)
2580                 /* xbdev is closed */
2581                 err = -ERESTARTSYS;
2582
2583         mutex_unlock(&info->mutex);
2584
2585 out:
2586         mutex_unlock(&blkfront_mutex);
2587         return err;
2588 }
2589
2590 static void blkif_release(struct gendisk *disk, fmode_t mode)
2591 {
2592         struct blkfront_info *info = disk->private_data;
2593         struct block_device *bdev;
2594         struct xenbus_device *xbdev;
2595
2596         mutex_lock(&blkfront_mutex);
2597
2598         bdev = bdget_disk(disk, 0);
2599
2600         if (!bdev) {
2601                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2602                 goto out_mutex;
2603         }
2604         if (bdev->bd_openers)
2605                 goto out;
2606
2607         /*
2608          * Check if we have been instructed to close. We will have
2609          * deferred this request, because the bdev was still open.
2610          */
2611
2612         mutex_lock(&info->mutex);
2613         xbdev = info->xbdev;
2614
2615         if (xbdev && xbdev->state == XenbusStateClosing) {
2616                 /* pending switch to state closed */
2617                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2618                 xlvbd_release_gendisk(info);
2619                 xenbus_frontend_closed(info->xbdev);
2620         }
2621
2622         mutex_unlock(&info->mutex);
2623
2624         if (!xbdev) {
2625                 /* sudden device removal */
2626                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2627                 xlvbd_release_gendisk(info);
2628                 disk->private_data = NULL;
2629                 kfree(info);
2630         }
2631
2632 out:
2633         bdput(bdev);
2634 out_mutex:
2635         mutex_unlock(&blkfront_mutex);
2636 }
2637
2638 static const struct block_device_operations xlvbd_block_fops =
2639 {
2640         .owner = THIS_MODULE,
2641         .open = blkif_open,
2642         .release = blkif_release,
2643         .getgeo = blkif_getgeo,
2644         .ioctl = blkif_ioctl,
2645 };
2646
2647
2648 static const struct xenbus_device_id blkfront_ids[] = {
2649         { "vbd" },
2650         { "" }
2651 };
2652
2653 static struct xenbus_driver blkfront_driver = {
2654         .ids  = blkfront_ids,
2655         .probe = blkfront_probe,
2656         .remove = blkfront_remove,
2657         .resume = blkfront_resume,
2658         .otherend_changed = blkback_changed,
2659         .is_ready = blkfront_is_ready,
2660 };
2661
2662 static int __init xlblk_init(void)
2663 {
2664         int ret;
2665         int nr_cpus = num_online_cpus();
2666
2667         if (!xen_domain())
2668                 return -ENODEV;
2669
2670         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2671                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2672                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2673                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2674         }
2675
2676         if (xen_blkif_max_queues > nr_cpus) {
2677                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2678                         xen_blkif_max_queues, nr_cpus);
2679                 xen_blkif_max_queues = nr_cpus;
2680         }
2681
2682         if (!xen_has_pv_disk_devices())
2683                 return -ENODEV;
2684
2685         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2686                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2687                        XENVBD_MAJOR, DEV_NAME);
2688                 return -ENODEV;
2689         }
2690
2691         ret = xenbus_register_frontend(&blkfront_driver);
2692         if (ret) {
2693                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2694                 return ret;
2695         }
2696
2697         return 0;
2698 }
2699 module_init(xlblk_init);
2700
2701
2702 static void __exit xlblk_exit(void)
2703 {
2704         xenbus_unregister_driver(&blkfront_driver);
2705         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2706         kfree(minors);
2707 }
2708 module_exit(xlblk_exit);
2709
2710 MODULE_DESCRIPTION("Xen virtual block device frontend");
2711 MODULE_LICENSE("GPL");
2712 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2713 MODULE_ALIAS("xen:vbd");
2714 MODULE_ALIAS("xenblk");