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 #include <linux/workqueue.h>
  50
  51 #include <xen/xen.h>
  52 #include <xen/xenbus.h>
  53 #include <xen/grant_table.h>
  54 #include <xen/events.h>
  55 #include <xen/page.h>
  56 #include <xen/platform_pci.h>
  57
  58 #include <xen/interface/grant_table.h>
  59 #include <xen/interface/io/blkif.h>
  60 #include <xen/interface/io/protocols.h>
  61
  62 #include <asm/xen/hypervisor.h>
  63
  64 /*
  65  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  66  * When Linux is using a different page size than Xen, it may not be possible
  67  * to put all the data in a single segment.
  68  * This can happen when the backend doesn't support indirect descriptor and
  69  * therefore the maximum amount of data that a request can carry is
  70  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  71  *
  72  * Note that we only support one extra request. So the Linux page size
  73  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  74  * 88KB.
  75  */
  76 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  77
  78 enum blkif_state {
  79         BLKIF_STATE_DISCONNECTED,
  80         BLKIF_STATE_CONNECTED,
  81         BLKIF_STATE_SUSPENDED,
  82 };
  83
  84 struct grant {
  85         grant_ref_t gref;
  86         struct page *page;
  87         struct list_head node;
  88 };
  89
  90 enum blk_req_status {
  91         REQ_WAITING,
  92         REQ_DONE,
  93         REQ_ERROR,
  94         REQ_EOPNOTSUPP,
  95 };
  96
  97 struct blk_shadow {
  98         struct blkif_request req;
  99         struct request *request;
 100         struct grant **grants_used;
 101         struct grant **indirect_grants;
 102         struct scatterlist *sg;
 103         unsigned int num_sg;
 104         enum blk_req_status status;
 105
 106         #define NO_ASSOCIATED_ID ~0UL
 107         /*
 108          * Id of the sibling if we ever need 2 requests when handling a
 109          * block I/O request
 110          */
 111         unsigned long associated_id;
 112 };
 113
 114 struct blkif_req {
 115         blk_status_t    error;
 116 };
 117
 118 static inline struct blkif_req *blkif_req(struct request *rq)
 119 {
 120         return blk_mq_rq_to_pdu(rq);
 121 }
 122
 123 static DEFINE_MUTEX(blkfront_mutex);
 124 static const struct block_device_operations xlvbd_block_fops;
 125 static struct delayed_work blkfront_work;
 126 static LIST_HEAD(info_list);
 127
 128 /*
 129  * Maximum number of segments in indirect requests, the actual value used by
 130  * the frontend driver is the minimum of this value and the value provided
 131  * by the backend driver.
 132  */
 133
 134 static unsigned int xen_blkif_max_segments = 32;
 135 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 136 MODULE_PARM_DESC(max_indirect_segments,
 137                  "Maximum amount of segments in indirect requests (default is 32)");
 138
 139 static unsigned int xen_blkif_max_queues = 4;
 140 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 141 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 142
 143 /*
 144  * Maximum order of pages to be used for the shared ring between front and
 145  * backend, 4KB page granularity is used.
 146  */
 147 static unsigned int xen_blkif_max_ring_order;
 148 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 149 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 150
 151 #define BLK_RING_SIZE(info)     \
 152         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 153
 154 /*
 155  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 156  * characters are enough. Define to 20 to keep consistent with backend.
 157  */
 158 #define RINGREF_NAME_LEN (20)
 159 /*
 160  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 161  */
 162 #define QUEUE_NAME_LEN (17)
 163
 164 /*
 165  *  Per-ring info.
 166  *  Every blkfront device can associate with one or more blkfront_ring_info,
 167  *  depending on how many hardware queues/rings to be used.
 168  */
 169 struct blkfront_ring_info {
 170         /* Lock to protect data in every ring buffer. */
 171         spinlock_t ring_lock;
 172         struct blkif_front_ring ring;
 173         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 174         unsigned int evtchn, irq;
 175         struct work_struct work;
 176         struct gnttab_free_callback callback;
 177         struct list_head indirect_pages;
 178         struct list_head grants;
 179         unsigned int persistent_gnts_c;
 180         unsigned long shadow_free;
 181         struct blkfront_info *dev_info;
 182         struct blk_shadow shadow[];
 183 };
 184
 185 /*
 186  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 187  * hang in private_data off the gendisk structure. We may end up
 188  * putting all kinds of interesting stuff here :-)
 189  */
 190 struct blkfront_info
 191 {
 192         struct mutex mutex;
 193         struct xenbus_device *xbdev;
 194         struct gendisk *gd;
 195         u16 sector_size;
 196         unsigned int physical_sector_size;
 197         int vdevice;
 198         blkif_vdev_t handle;
 199         enum blkif_state connected;
 200         /* Number of pages per ring buffer. */
 201         unsigned int nr_ring_pages;
 202         struct request_queue *rq;
 203         unsigned int feature_flush:1;
 204         unsigned int feature_fua:1;
 205         unsigned int feature_discard:1;
 206         unsigned int feature_secdiscard:1;
 207         unsigned int feature_persistent:1;
 208         unsigned int discard_granularity;
 209         unsigned int discard_alignment;
 210         /* Number of 4KB segments handled */
 211         unsigned int max_indirect_segments;
 212         int is_ready;
 213         struct blk_mq_tag_set tag_set;
 214         struct blkfront_ring_info *rinfo;
 215         unsigned int nr_rings;
 216         /* Save uncomplete reqs and bios for migration. */
 217         struct list_head requests;
 218         struct bio_list bio_list;
 219         struct list_head info_list;
 220 };
 221
 222 static unsigned int nr_minors;
 223 static unsigned long *minors;
 224 static DEFINE_SPINLOCK(minor_lock);
 225
 226 #define GRANT_INVALID_REF       0
 227
 228 #define PARTS_PER_DISK          16
 229 #define PARTS_PER_EXT_DISK      256
 230
 231 #define BLKIF_MAJOR(dev) ((dev)>>8)
 232 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 233
 234 #define EXT_SHIFT 28
 235 #define EXTENDED (1<<EXT_SHIFT)
 236 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 237 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 238 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 239 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 240 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 241 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 242
 243 #define DEV_NAME        "xvd"   /* name in /dev */
 244
 245 /*
 246  * Grants are always the same size as a Xen page (i.e 4KB).
 247  * A physical segment is always the same size as a Linux page.
 248  * Number of grants per physical segment
 249  */
 250 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 251
 252 #define GRANTS_PER_INDIRECT_FRAME \
 253         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 254
 255 #define INDIRECT_GREFS(_grants)         \
 256         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 257
 258 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 259 static void blkfront_gather_backend_features(struct blkfront_info *info);
 260 static int negotiate_mq(struct blkfront_info *info);
 261
 262 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 263 {
 264         unsigned long free = rinfo->shadow_free;
 265
 266         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 267         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 268         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 269         return free;
 270 }
 271
 272 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 273                               unsigned long id)
 274 {
 275         if (rinfo->shadow[id].req.u.rw.id != id)
 276                 return -EINVAL;
 277         if (rinfo->shadow[id].request == NULL)
 278                 return -EINVAL;
 279         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 280         rinfo->shadow[id].request = NULL;
 281         rinfo->shadow_free = id;
 282         return 0;
 283 }
 284
 285 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 286 {
 287         struct blkfront_info *info = rinfo->dev_info;
 288         struct page *granted_page;
 289         struct grant *gnt_list_entry, *n;
 290         int i = 0;
 291
 292         while (i < num) {
 293                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 294                 if (!gnt_list_entry)
 295                         goto out_of_memory;
 296
 297                 if (info->feature_persistent) {
 298                         granted_page = alloc_page(GFP_NOIO);
 299                         if (!granted_page) {
 300                                 kfree(gnt_list_entry);
 301                                 goto out_of_memory;
 302                         }
 303                         gnt_list_entry->page = granted_page;
 304                 }
 305
 306                 gnt_list_entry->gref = GRANT_INVALID_REF;
 307                 list_add(&gnt_list_entry->node, &rinfo->grants);
 308                 i++;
 309         }
 310
 311         return 0;
 312
 313 out_of_memory:
 314         list_for_each_entry_safe(gnt_list_entry, n,
 315                                  &rinfo->grants, node) {
 316                 list_del(&gnt_list_entry->node);
 317                 if (info->feature_persistent)
 318                         __free_page(gnt_list_entry->page);
 319                 kfree(gnt_list_entry);
 320                 i--;
 321         }
 322         BUG_ON(i != 0);
 323         return -ENOMEM;
 324 }
 325
 326 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 327 {
 328         struct grant *gnt_list_entry;
 329
 330         BUG_ON(list_empty(&rinfo->grants));
 331         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 332                                           node);
 333         list_del(&gnt_list_entry->node);
 334
 335         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 336                 rinfo->persistent_gnts_c--;
 337
 338         return gnt_list_entry;
 339 }
 340
 341 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 342                                         const struct blkfront_info *info)
 343 {
 344         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 345                                                  info->xbdev->otherend_id,
 346                                                  gnt_list_entry->page,
 347                                                  0);
 348 }
 349
 350 static struct grant *get_grant(grant_ref_t *gref_head,
 351                                unsigned long gfn,
 352                                struct blkfront_ring_info *rinfo)
 353 {
 354         struct grant *gnt_list_entry = get_free_grant(rinfo);
 355         struct blkfront_info *info = rinfo->dev_info;
 356
 357         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 358                 return gnt_list_entry;
 359
 360         /* Assign a gref to this page */
 361         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 362         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 363         if (info->feature_persistent)
 364                 grant_foreign_access(gnt_list_entry, info);
 365         else {
 366                 /* Grant access to the GFN passed by the caller */
 367                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 368                                                 info->xbdev->otherend_id,
 369                                                 gfn, 0);
 370         }
 371
 372         return gnt_list_entry;
 373 }
 374
 375 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 376                                         struct blkfront_ring_info *rinfo)
 377 {
 378         struct grant *gnt_list_entry = get_free_grant(rinfo);
 379         struct blkfront_info *info = rinfo->dev_info;
 380
 381         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 382                 return gnt_list_entry;
 383
 384         /* Assign a gref to this page */
 385         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 386         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 387         if (!info->feature_persistent) {
 388                 struct page *indirect_page;
 389
 390                 /* Fetch a pre-allocated page to use for indirect grefs */
 391                 BUG_ON(list_empty(&rinfo->indirect_pages));
 392                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 393                                                  struct page, lru);
 394                 list_del(&indirect_page->lru);
 395                 gnt_list_entry->page = indirect_page;
 396         }
 397         grant_foreign_access(gnt_list_entry, info);
 398
 399         return gnt_list_entry;
 400 }
 401
 402 static const char *op_name(int op)
 403 {
 404         static const char *const names[] = {
 405                 [BLKIF_OP_READ] = "read",
 406                 [BLKIF_OP_WRITE] = "write",
 407                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 408                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 409                 [BLKIF_OP_DISCARD] = "discard" };
 410
 411         if (op < 0 || op >= ARRAY_SIZE(names))
 412                 return "unknown";
 413
 414         if (!names[op])
 415                 return "reserved";
 416
 417         return names[op];
 418 }
 419 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 420 {
 421         unsigned int end = minor + nr;
 422         int rc;
 423
 424         if (end > nr_minors) {
 425                 unsigned long *bitmap, *old;
 426
 427                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 428                                  GFP_KERNEL);
 429                 if (bitmap == NULL)
 430                         return -ENOMEM;
 431
 432                 spin_lock(&minor_lock);
 433                 if (end > nr_minors) {
 434                         old = minors;
 435                         memcpy(bitmap, minors,
 436                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 437                         minors = bitmap;
 438                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 439                 } else
 440                         old = bitmap;
 441                 spin_unlock(&minor_lock);
 442                 kfree(old);
 443         }
 444
 445         spin_lock(&minor_lock);
 446         if (find_next_bit(minors, end, minor) >= end) {
 447                 bitmap_set(minors, minor, nr);
 448                 rc = 0;
 449         } else
 450                 rc = -EBUSY;
 451         spin_unlock(&minor_lock);
 452
 453         return rc;
 454 }
 455
 456 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 457 {
 458         unsigned int end = minor + nr;
 459
 460         BUG_ON(end > nr_minors);
 461         spin_lock(&minor_lock);
 462         bitmap_clear(minors,  minor, nr);
 463         spin_unlock(&minor_lock);
 464 }
 465
 466 static void blkif_restart_queue_callback(void *arg)
 467 {
 468         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 469         schedule_work(&rinfo->work);
 470 }
 471
 472 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 473 {
 474         /* We don't have real geometry info, but let's at least return
 475            values consistent with the size of the device */
 476         sector_t nsect = get_capacity(bd->bd_disk);
 477         sector_t cylinders = nsect;
 478
 479         hg->heads = 0xff;
 480         hg->sectors = 0x3f;
 481         sector_div(cylinders, hg->heads * hg->sectors);
 482         hg->cylinders = cylinders;
 483         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 484                 hg->cylinders = 0xffff;
 485         return 0;
 486 }
 487
 488 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 489                        unsigned command, unsigned long argument)
 490 {
 491         struct blkfront_info *info = bdev->bd_disk->private_data;
 492         int i;
 493
 494         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 495                 command, (long)argument);
 496
 497         switch (command) {
 498         case CDROMMULTISESSION:
 499                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 500                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 501                         if (put_user(0, (char __user *)(argument + i)))
 502                                 return -EFAULT;
 503                 return 0;
 504
 505         case CDROM_GET_CAPABILITY: {
 506                 struct gendisk *gd = info->gd;
 507                 if (gd->flags & GENHD_FL_CD)
 508                         return 0;
 509                 return -EINVAL;
 510         }
 511
 512         default:
 513                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 514                   command);*/
 515                 return -EINVAL; /* same return as native Linux */
 516         }
 517
 518         return 0;
 519 }
 520
 521 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 522                                             struct request *req,
 523                                             struct blkif_request **ring_req)
 524 {
 525         unsigned long id;
 526
 527         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 528         rinfo->ring.req_prod_pvt++;
 529
 530         id = get_id_from_freelist(rinfo);
 531         rinfo->shadow[id].request = req;
 532         rinfo->shadow[id].status = REQ_WAITING;
 533         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 534
 535         (*ring_req)->u.rw.id = id;
 536
 537         return id;
 538 }
 539
 540 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 541 {
 542         struct blkfront_info *info = rinfo->dev_info;
 543         struct blkif_request *ring_req;
 544         unsigned long id;
 545
 546         /* Fill out a communications ring structure. */
 547         id = blkif_ring_get_request(rinfo, req, &ring_req);
 548
 549         ring_req->operation = BLKIF_OP_DISCARD;
 550         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 551         ring_req->u.discard.id = id;
 552         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 553         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 554                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 555         else
 556                 ring_req->u.discard.flag = 0;
 557
 558         /* Keep a private copy so we can reissue requests when recovering. */
 559         rinfo->shadow[id].req = *ring_req;
 560
 561         return 0;
 562 }
 563
 564 struct setup_rw_req {
 565         unsigned int grant_idx;
 566         struct blkif_request_segment *segments;
 567         struct blkfront_ring_info *rinfo;
 568         struct blkif_request *ring_req;
 569         grant_ref_t gref_head;
 570         unsigned int id;
 571         /* Only used when persistent grant is used and it's a read request */
 572         bool need_copy;
 573         unsigned int bvec_off;
 574         char *bvec_data;
 575
 576         bool require_extra_req;
 577         struct blkif_request *extra_ring_req;
 578 };
 579
 580 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 581                                      unsigned int len, void *data)
 582 {
 583         struct setup_rw_req *setup = data;
 584         int n, ref;
 585         struct grant *gnt_list_entry;
 586         unsigned int fsect, lsect;
 587         /* Convenient aliases */
 588         unsigned int grant_idx = setup->grant_idx;
 589         struct blkif_request *ring_req = setup->ring_req;
 590         struct blkfront_ring_info *rinfo = setup->rinfo;
 591         /*
 592          * We always use the shadow of the first request to store the list
 593          * of grant associated to the block I/O request. This made the
 594          * completion more easy to handle even if the block I/O request is
 595          * split.
 596          */
 597         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 598
 599         if (unlikely(setup->require_extra_req &&
 600                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 601                 /*
 602                  * We are using the second request, setup grant_idx
 603                  * to be the index of the segment array.
 604                  */
 605                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 606                 ring_req = setup->extra_ring_req;
 607         }
 608
 609         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 610             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 611                 if (setup->segments)
 612                         kunmap_atomic(setup->segments);
 613
 614                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 615                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 616                 shadow->indirect_grants[n] = gnt_list_entry;
 617                 setup->segments = kmap_atomic(gnt_list_entry->page);
 618                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 619         }
 620
 621         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 622         ref = gnt_list_entry->gref;
 623         /*
 624          * All the grants are stored in the shadow of the first
 625          * request. Therefore we have to use the global index.
 626          */
 627         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 628
 629         if (setup->need_copy) {
 630                 void *shared_data;
 631
 632                 shared_data = kmap_atomic(gnt_list_entry->page);
 633                 /*
 634                  * this does not wipe data stored outside the
 635                  * range sg->offset..sg->offset+sg->length.
 636                  * Therefore, blkback *could* see data from
 637                  * previous requests. This is OK as long as
 638                  * persistent grants are shared with just one
 639                  * domain. It may need refactoring if this
 640                  * changes
 641                  */
 642                 memcpy(shared_data + offset,
 643                        setup->bvec_data + setup->bvec_off,
 644                        len);
 645
 646                 kunmap_atomic(shared_data);
 647                 setup->bvec_off += len;
 648         }
 649
 650         fsect = offset >> 9;
 651         lsect = fsect + (len >> 9) - 1;
 652         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 653                 ring_req->u.rw.seg[grant_idx] =
 654                         (struct blkif_request_segment) {
 655                                 .gref       = ref,
 656                                 .first_sect = fsect,
 657                                 .last_sect  = lsect };
 658         } else {
 659                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 660                         (struct blkif_request_segment) {
 661                                 .gref       = ref,
 662                                 .first_sect = fsect,
 663                                 .last_sect  = lsect };
 664         }
 665
 666         (setup->grant_idx)++;
 667 }
 668
 669 static void blkif_setup_extra_req(struct blkif_request *first,
 670                                   struct blkif_request *second)
 671 {
 672         uint16_t nr_segments = first->u.rw.nr_segments;
 673
 674         /*
 675          * The second request is only present when the first request uses
 676          * all its segments. It's always the continuity of the first one.
 677          */
 678         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 679
 680         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 681         second->u.rw.sector_number = first->u.rw.sector_number +
 682                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 683
 684         second->u.rw.handle = first->u.rw.handle;
 685         second->operation = first->operation;
 686 }
 687
 688 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 689 {
 690         struct blkfront_info *info = rinfo->dev_info;
 691         struct blkif_request *ring_req, *extra_ring_req = NULL;
 692         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 693         bool require_extra_req = false;
 694         int i;
 695         struct setup_rw_req setup = {
 696                 .grant_idx = 0,
 697                 .segments = NULL,
 698                 .rinfo = rinfo,
 699                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 700         };
 701
 702         /*
 703          * Used to store if we are able to queue the request by just using
 704          * existing persistent grants, or if we have to get new grants,
 705          * as there are not sufficiently many free.
 706          */
 707         bool new_persistent_gnts = false;
 708         struct scatterlist *sg;
 709         int num_sg, max_grefs, num_grant;
 710
 711         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 712         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 713                 /*
 714                  * If we are using indirect segments we need to account
 715                  * for the indirect grefs used in the request.
 716                  */
 717                 max_grefs += INDIRECT_GREFS(max_grefs);
 718
 719         /* Check if we have enough persistent grants to allocate a requests */
 720         if (rinfo->persistent_gnts_c < max_grefs) {
 721                 new_persistent_gnts = true;
 722
 723                 if (gnttab_alloc_grant_references(
 724                     max_grefs - rinfo->persistent_gnts_c,
 725                     &setup.gref_head) < 0) {
 726                         gnttab_request_free_callback(
 727                                 &rinfo->callback,
 728                                 blkif_restart_queue_callback,
 729                                 rinfo,
 730                                 max_grefs - rinfo->persistent_gnts_c);
 731                         return 1;
 732                 }
 733         }
 734
 735         /* Fill out a communications ring structure. */
 736         id = blkif_ring_get_request(rinfo, req, &ring_req);
 737
 738         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 739         num_grant = 0;
 740         /* Calculate the number of grant used */
 741         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 742                num_grant += gnttab_count_grant(sg->offset, sg->length);
 743
 744         require_extra_req = info->max_indirect_segments == 0 &&
 745                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 746         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 747
 748         rinfo->shadow[id].num_sg = num_sg;
 749         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 750             likely(!require_extra_req)) {
 751                 /*
 752                  * The indirect operation can only be a BLKIF_OP_READ or
 753                  * BLKIF_OP_WRITE
 754                  */
 755                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 756                 ring_req->operation = BLKIF_OP_INDIRECT;
 757                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 758                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 759                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 760                 ring_req->u.indirect.handle = info->handle;
 761                 ring_req->u.indirect.nr_segments = num_grant;
 762         } else {
 763                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 764                 ring_req->u.rw.handle = info->handle;
 765                 ring_req->operation = rq_data_dir(req) ?
 766                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 767                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 768                         /*
 769                          * Ideally we can do an unordered flush-to-disk.
 770                          * In case the backend onlysupports barriers, use that.
 771                          * A barrier request a superset of FUA, so we can
 772                          * implement it the same way.  (It's also a FLUSH+FUA,
 773                          * since it is guaranteed ordered WRT previous writes.)
 774                          */
 775                         if (info->feature_flush && info->feature_fua)
 776                                 ring_req->operation =
 777                                         BLKIF_OP_WRITE_BARRIER;
 778                         else if (info->feature_flush)
 779                                 ring_req->operation =
 780                                         BLKIF_OP_FLUSH_DISKCACHE;
 781                         else
 782                                 ring_req->operation = 0;
 783                 }
 784                 ring_req->u.rw.nr_segments = num_grant;
 785                 if (unlikely(require_extra_req)) {
 786                         extra_id = blkif_ring_get_request(rinfo, req,
 787                                                           &extra_ring_req);
 788                         /*
 789                          * Only the first request contains the scatter-gather
 790                          * list.
 791                          */
 792                         rinfo->shadow[extra_id].num_sg = 0;
 793
 794                         blkif_setup_extra_req(ring_req, extra_ring_req);
 795
 796                         /* Link the 2 requests together */
 797                         rinfo->shadow[extra_id].associated_id = id;
 798                         rinfo->shadow[id].associated_id = extra_id;
 799                 }
 800         }
 801
 802         setup.ring_req = ring_req;
 803         setup.id = id;
 804
 805         setup.require_extra_req = require_extra_req;
 806         if (unlikely(require_extra_req))
 807                 setup.extra_ring_req = extra_ring_req;
 808
 809         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 810                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 811
 812                 if (setup.need_copy) {
 813                         setup.bvec_off = sg->offset;
 814                         setup.bvec_data = kmap_atomic(sg_page(sg));
 815                 }
 816
 817                 gnttab_foreach_grant_in_range(sg_page(sg),
 818                                               sg->offset,
 819                                               sg->length,
 820                                               blkif_setup_rw_req_grant,
 821                                               &setup);
 822
 823                 if (setup.need_copy)
 824                         kunmap_atomic(setup.bvec_data);
 825         }
 826         if (setup.segments)
 827                 kunmap_atomic(setup.segments);
 828
 829         /* Keep a private copy so we can reissue requests when recovering. */
 830         rinfo->shadow[id].req = *ring_req;
 831         if (unlikely(require_extra_req))
 832                 rinfo->shadow[extra_id].req = *extra_ring_req;
 833
 834         if (new_persistent_gnts)
 835                 gnttab_free_grant_references(setup.gref_head);
 836
 837         return 0;
 838 }
 839
 840 /*
 841  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 842  * and writes are handled as expected.
 843  *
 844  * @req: a request struct
 845  */
 846 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 847 {
 848         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 849                 return 1;
 850
 851         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 852                      req_op(req) == REQ_OP_SECURE_ERASE))
 853                 return blkif_queue_discard_req(req, rinfo);
 854         else
 855                 return blkif_queue_rw_req(req, rinfo);
 856 }
 857
 858 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 859 {
 860         int notify;
 861
 862         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 863
 864         if (notify)
 865                 notify_remote_via_irq(rinfo->irq);
 866 }
 867
 868 static inline bool blkif_request_flush_invalid(struct request *req,
 869                                                struct blkfront_info *info)
 870 {
 871         return (blk_rq_is_passthrough(req) ||
 872                 ((req_op(req) == REQ_OP_FLUSH) &&
 873                  !info->feature_flush) ||
 874                 ((req->cmd_flags & REQ_FUA) &&
 875                  !info->feature_fua));
 876 }
 877
 878 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 879                           const struct blk_mq_queue_data *qd)
 880 {
 881         unsigned long flags;
 882         int qid = hctx->queue_num;
 883         struct blkfront_info *info = hctx->queue->queuedata;
 884         struct blkfront_ring_info *rinfo = NULL;
 885
 886         BUG_ON(info->nr_rings <= qid);
 887         rinfo = &info->rinfo[qid];
 888         blk_mq_start_request(qd->rq);
 889         spin_lock_irqsave(&rinfo->ring_lock, flags);
 890         if (RING_FULL(&rinfo->ring))
 891                 goto out_busy;
 892
 893         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 894                 goto out_err;
 895
 896         if (blkif_queue_request(qd->rq, rinfo))
 897                 goto out_busy;
 898
 899         flush_requests(rinfo);
 900         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 901         return BLK_STS_OK;
 902
 903 out_err:
 904         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 905         return BLK_STS_IOERR;
 906
 907 out_busy:
 908         blk_mq_stop_hw_queue(hctx);
 909         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 910         return BLK_STS_DEV_RESOURCE;
 911 }
 912
 913 static void blkif_complete_rq(struct request *rq)
 914 {
 915         blk_mq_end_request(rq, blkif_req(rq)->error);
 916 }
 917
 918 static const struct blk_mq_ops blkfront_mq_ops = {
 919         .queue_rq = blkif_queue_rq,
 920         .complete = blkif_complete_rq,
 921 };
 922
 923 static void blkif_set_queue_limits(struct blkfront_info *info)
 924 {
 925         struct request_queue *rq = info->rq;
 926         struct gendisk *gd = info->gd;
 927         unsigned int segments = info->max_indirect_segments ? :
 928                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 929
 930         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 931
 932         if (info->feature_discard) {
 933                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
 934                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 935                 rq->limits.discard_granularity = info->discard_granularity;
 936                 rq->limits.discard_alignment = info->discard_alignment;
 937                 if (info->feature_secdiscard)
 938                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
 939         }
 940
 941         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 942         blk_queue_logical_block_size(rq, info->sector_size);
 943         blk_queue_physical_block_size(rq, info->physical_sector_size);
 944         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 945
 946         /* Each segment in a request is up to an aligned page in size. */
 947         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 948         blk_queue_max_segment_size(rq, PAGE_SIZE);
 949
 950         /* Ensure a merged request will fit in a single I/O ring slot. */
 951         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 952
 953         /* Make sure buffer addresses are sector-aligned. */
 954         blk_queue_dma_alignment(rq, 511);
 955 }
 956
 957 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 958                                 unsigned int physical_sector_size)
 959 {
 960         struct request_queue *rq;
 961         struct blkfront_info *info = gd->private_data;
 962
 963         memset(&info->tag_set, 0, sizeof(info->tag_set));
 964         info->tag_set.ops = &blkfront_mq_ops;
 965         info->tag_set.nr_hw_queues = info->nr_rings;
 966         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
 967                 /*
 968                  * When indirect descriptior is not supported, the I/O request
 969                  * will be split between multiple request in the ring.
 970                  * To avoid problems when sending the request, divide by
 971                  * 2 the depth of the queue.
 972                  */
 973                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
 974         } else
 975                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
 976         info->tag_set.numa_node = NUMA_NO_NODE;
 977         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 978         info->tag_set.cmd_size = sizeof(struct blkif_req);
 979         info->tag_set.driver_data = info;
 980
 981         if (blk_mq_alloc_tag_set(&info->tag_set))
 982                 return -EINVAL;
 983         rq = blk_mq_init_queue(&info->tag_set);
 984         if (IS_ERR(rq)) {
 985                 blk_mq_free_tag_set(&info->tag_set);
 986                 return PTR_ERR(rq);
 987         }
 988
 989         rq->queuedata = info;
 990         info->rq = gd->queue = rq;
 991         info->gd = gd;
 992         info->sector_size = sector_size;
 993         info->physical_sector_size = physical_sector_size;
 994         blkif_set_queue_limits(info);
 995
 996         return 0;
 997 }
 998
 999 static const char *flush_info(struct blkfront_info *info)
1000 {
1001         if (info->feature_flush && info->feature_fua)
1002                 return "barrier: enabled;";
1003         else if (info->feature_flush)
1004                 return "flush diskcache: enabled;";
1005         else
1006                 return "barrier or flush: disabled;";
1007 }
1008
1009 static void xlvbd_flush(struct blkfront_info *info)
1010 {
1011         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1012                               info->feature_fua ? true : false);
1013         pr_info("blkfront: %s: %s %s %s %s %s\n",
1014                 info->gd->disk_name, flush_info(info),
1015                 "persistent grants:", info->feature_persistent ?
1016                 "enabled;" : "disabled;", "indirect descriptors:",
1017                 info->max_indirect_segments ? "enabled;" : "disabled;");
1018 }
1019
1020 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1021 {
1022         int major;
1023         major = BLKIF_MAJOR(vdevice);
1024         *minor = BLKIF_MINOR(vdevice);
1025         switch (major) {
1026                 case XEN_IDE0_MAJOR:
1027                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1028                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1029                                 EMULATED_HD_DISK_MINOR_OFFSET;
1030                         break;
1031                 case XEN_IDE1_MAJOR:
1032                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1033                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1034                                 EMULATED_HD_DISK_MINOR_OFFSET;
1035                         break;
1036                 case XEN_SCSI_DISK0_MAJOR:
1037                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1038                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1039                         break;
1040                 case XEN_SCSI_DISK1_MAJOR:
1041                 case XEN_SCSI_DISK2_MAJOR:
1042                 case XEN_SCSI_DISK3_MAJOR:
1043                 case XEN_SCSI_DISK4_MAJOR:
1044                 case XEN_SCSI_DISK5_MAJOR:
1045                 case XEN_SCSI_DISK6_MAJOR:
1046                 case XEN_SCSI_DISK7_MAJOR:
1047                         *offset = (*minor / PARTS_PER_DISK) +
1048                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1049                                 EMULATED_SD_DISK_NAME_OFFSET;
1050                         *minor = *minor +
1051                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1052                                 EMULATED_SD_DISK_MINOR_OFFSET;
1053                         break;
1054                 case XEN_SCSI_DISK8_MAJOR:
1055                 case XEN_SCSI_DISK9_MAJOR:
1056                 case XEN_SCSI_DISK10_MAJOR:
1057                 case XEN_SCSI_DISK11_MAJOR:
1058                 case XEN_SCSI_DISK12_MAJOR:
1059                 case XEN_SCSI_DISK13_MAJOR:
1060                 case XEN_SCSI_DISK14_MAJOR:
1061                 case XEN_SCSI_DISK15_MAJOR:
1062                         *offset = (*minor / PARTS_PER_DISK) +
1063                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1064                                 EMULATED_SD_DISK_NAME_OFFSET;
1065                         *minor = *minor +
1066                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1067                                 EMULATED_SD_DISK_MINOR_OFFSET;
1068                         break;
1069                 case XENVBD_MAJOR:
1070                         *offset = *minor / PARTS_PER_DISK;
1071                         break;
1072                 default:
1073                         printk(KERN_WARNING "blkfront: your disk configuration is "
1074                                         "incorrect, please use an xvd device instead\n");
1075                         return -ENODEV;
1076         }
1077         return 0;
1078 }
1079
1080 static char *encode_disk_name(char *ptr, unsigned int n)
1081 {
1082         if (n >= 26)
1083                 ptr = encode_disk_name(ptr, n / 26 - 1);
1084         *ptr = 'a' + n % 26;
1085         return ptr + 1;
1086 }
1087
1088 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1089                                struct blkfront_info *info,
1090                                u16 vdisk_info, u16 sector_size,
1091                                unsigned int physical_sector_size)
1092 {
1093         struct gendisk *gd;
1094         int nr_minors = 1;
1095         int err;
1096         unsigned int offset;
1097         int minor;
1098         int nr_parts;
1099         char *ptr;
1100
1101         BUG_ON(info->gd != NULL);
1102         BUG_ON(info->rq != NULL);
1103
1104         if ((info->vdevice>>EXT_SHIFT) > 1) {
1105                 /* this is above the extended range; something is wrong */
1106                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1107                 return -ENODEV;
1108         }
1109
1110         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1111                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1112                 if (err)
1113                         return err;
1114                 nr_parts = PARTS_PER_DISK;
1115         } else {
1116                 minor = BLKIF_MINOR_EXT(info->vdevice);
1117                 nr_parts = PARTS_PER_EXT_DISK;
1118                 offset = minor / nr_parts;
1119                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1120                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1121                                         "emulated IDE disks,\n\t choose an xvd device name"
1122                                         "from xvde on\n", info->vdevice);
1123         }
1124         if (minor >> MINORBITS) {
1125                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1126                         info->vdevice, minor);
1127                 return -ENODEV;
1128         }
1129
1130         if ((minor % nr_parts) == 0)
1131                 nr_minors = nr_parts;
1132
1133         err = xlbd_reserve_minors(minor, nr_minors);
1134         if (err)
1135                 goto out;
1136         err = -ENODEV;
1137
1138         gd = alloc_disk(nr_minors);
1139         if (gd == NULL)
1140                 goto release;
1141
1142         strcpy(gd->disk_name, DEV_NAME);
1143         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1144         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1145         if (nr_minors > 1)
1146                 *ptr = 0;
1147         else
1148                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1149                          "%d", minor & (nr_parts - 1));
1150
1151         gd->major = XENVBD_MAJOR;
1152         gd->first_minor = minor;
1153         gd->fops = &xlvbd_block_fops;
1154         gd->private_data = info;
1155         set_capacity(gd, capacity);
1156
1157         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1158                 del_gendisk(gd);
1159                 goto release;
1160         }
1161
1162         xlvbd_flush(info);
1163
1164         if (vdisk_info & VDISK_READONLY)
1165                 set_disk_ro(gd, 1);
1166
1167         if (vdisk_info & VDISK_REMOVABLE)
1168                 gd->flags |= GENHD_FL_REMOVABLE;
1169
1170         if (vdisk_info & VDISK_CDROM)
1171                 gd->flags |= GENHD_FL_CD;
1172
1173         return 0;
1174
1175  release:
1176         xlbd_release_minors(minor, nr_minors);
1177  out:
1178         return err;
1179 }
1180
1181 static void xlvbd_release_gendisk(struct blkfront_info *info)
1182 {
1183         unsigned int minor, nr_minors, i;
1184
1185         if (info->rq == NULL)
1186                 return;
1187
1188         /* No more blkif_request(). */
1189         blk_mq_stop_hw_queues(info->rq);
1190
1191         for (i = 0; i < info->nr_rings; i++) {
1192                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1193
1194                 /* No more gnttab callback work. */
1195                 gnttab_cancel_free_callback(&rinfo->callback);
1196
1197                 /* Flush gnttab callback work. Must be done with no locks held. */
1198                 flush_work(&rinfo->work);
1199         }
1200
1201         del_gendisk(info->gd);
1202
1203         minor = info->gd->first_minor;
1204         nr_minors = info->gd->minors;
1205         xlbd_release_minors(minor, nr_minors);
1206
1207         blk_cleanup_queue(info->rq);
1208         blk_mq_free_tag_set(&info->tag_set);
1209         info->rq = NULL;
1210
1211         put_disk(info->gd);
1212         info->gd = NULL;
1213 }
1214
1215 /* Already hold rinfo->ring_lock. */
1216 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1217 {
1218         if (!RING_FULL(&rinfo->ring))
1219                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1220 }
1221
1222 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1223 {
1224         unsigned long flags;
1225
1226         spin_lock_irqsave(&rinfo->ring_lock, flags);
1227         kick_pending_request_queues_locked(rinfo);
1228         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1229 }
1230
1231 static void blkif_restart_queue(struct work_struct *work)
1232 {
1233         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1234
1235         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1236                 kick_pending_request_queues(rinfo);
1237 }
1238
1239 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1240 {
1241         struct grant *persistent_gnt, *n;
1242         struct blkfront_info *info = rinfo->dev_info;
1243         int i, j, segs;
1244
1245         /*
1246          * Remove indirect pages, this only happens when using indirect
1247          * descriptors but not persistent grants
1248          */
1249         if (!list_empty(&rinfo->indirect_pages)) {
1250                 struct page *indirect_page, *n;
1251
1252                 BUG_ON(info->feature_persistent);
1253                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1254                         list_del(&indirect_page->lru);
1255                         __free_page(indirect_page);
1256                 }
1257         }
1258
1259         /* Remove all persistent grants. */
1260         if (!list_empty(&rinfo->grants)) {
1261                 list_for_each_entry_safe(persistent_gnt, n,
1262                                          &rinfo->grants, node) {
1263                         list_del(&persistent_gnt->node);
1264                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1265                                 gnttab_end_foreign_access(persistent_gnt->gref,
1266                                                           0, 0UL);
1267                                 rinfo->persistent_gnts_c--;
1268                         }
1269                         if (info->feature_persistent)
1270                                 __free_page(persistent_gnt->page);
1271                         kfree(persistent_gnt);
1272                 }
1273         }
1274         BUG_ON(rinfo->persistent_gnts_c != 0);
1275
1276         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1277                 /*
1278                  * Clear persistent grants present in requests already
1279                  * on the shared ring
1280                  */
1281                 if (!rinfo->shadow[i].request)
1282                         goto free_shadow;
1283
1284                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1285                        rinfo->shadow[i].req.u.indirect.nr_segments :
1286                        rinfo->shadow[i].req.u.rw.nr_segments;
1287                 for (j = 0; j < segs; j++) {
1288                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1289                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1290                         if (info->feature_persistent)
1291                                 __free_page(persistent_gnt->page);
1292                         kfree(persistent_gnt);
1293                 }
1294
1295                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1296                         /*
1297                          * If this is not an indirect operation don't try to
1298                          * free indirect segments
1299                          */
1300                         goto free_shadow;
1301
1302                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1303                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1304                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1305                         __free_page(persistent_gnt->page);
1306                         kfree(persistent_gnt);
1307                 }
1308
1309 free_shadow:
1310                 kvfree(rinfo->shadow[i].grants_used);
1311                 rinfo->shadow[i].grants_used = NULL;
1312                 kvfree(rinfo->shadow[i].indirect_grants);
1313                 rinfo->shadow[i].indirect_grants = NULL;
1314                 kvfree(rinfo->shadow[i].sg);
1315                 rinfo->shadow[i].sg = NULL;
1316         }
1317
1318         /* No more gnttab callback work. */
1319         gnttab_cancel_free_callback(&rinfo->callback);
1320
1321         /* Flush gnttab callback work. Must be done with no locks held. */
1322         flush_work(&rinfo->work);
1323
1324         /* Free resources associated with old device channel. */
1325         for (i = 0; i < info->nr_ring_pages; i++) {
1326                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1327                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1328                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1329                 }
1330         }
1331         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1332         rinfo->ring.sring = NULL;
1333
1334         if (rinfo->irq)
1335                 unbind_from_irqhandler(rinfo->irq, rinfo);
1336         rinfo->evtchn = rinfo->irq = 0;
1337 }
1338
1339 static void blkif_free(struct blkfront_info *info, int suspend)
1340 {
1341         unsigned int i;
1342
1343         /* Prevent new requests being issued until we fix things up. */
1344         info->connected = suspend ?
1345                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1346         /* No more blkif_request(). */
1347         if (info->rq)
1348                 blk_mq_stop_hw_queues(info->rq);
1349
1350         for (i = 0; i < info->nr_rings; i++)
1351                 blkif_free_ring(&info->rinfo[i]);
1352
1353         kvfree(info->rinfo);
1354         info->rinfo = NULL;
1355         info->nr_rings = 0;
1356 }
1357
1358 struct copy_from_grant {
1359         const struct blk_shadow *s;
1360         unsigned int grant_idx;
1361         unsigned int bvec_offset;
1362         char *bvec_data;
1363 };
1364
1365 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1366                                   unsigned int len, void *data)
1367 {
1368         struct copy_from_grant *info = data;
1369         char *shared_data;
1370         /* Convenient aliases */
1371         const struct blk_shadow *s = info->s;
1372
1373         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1374
1375         memcpy(info->bvec_data + info->bvec_offset,
1376                shared_data + offset, len);
1377
1378         info->bvec_offset += len;
1379         info->grant_idx++;
1380
1381         kunmap_atomic(shared_data);
1382 }
1383
1384 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1385 {
1386         switch (rsp)
1387         {
1388         case BLKIF_RSP_OKAY:
1389                 return REQ_DONE;
1390         case BLKIF_RSP_EOPNOTSUPP:
1391                 return REQ_EOPNOTSUPP;
1392         case BLKIF_RSP_ERROR:
1393                 /* Fallthrough. */
1394         default:
1395                 return REQ_ERROR;
1396         }
1397 }
1398
1399 /*
1400  * Get the final status of the block request based on two ring response
1401  */
1402 static int blkif_get_final_status(enum blk_req_status s1,
1403                                   enum blk_req_status s2)
1404 {
1405         BUG_ON(s1 == REQ_WAITING);
1406         BUG_ON(s2 == REQ_WAITING);
1407
1408         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1409                 return BLKIF_RSP_ERROR;
1410         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1411                 return BLKIF_RSP_EOPNOTSUPP;
1412         return BLKIF_RSP_OKAY;
1413 }
1414
1415 static bool blkif_completion(unsigned long *id,
1416                              struct blkfront_ring_info *rinfo,
1417                              struct blkif_response *bret)
1418 {
1419         int i = 0;
1420         struct scatterlist *sg;
1421         int num_sg, num_grant;
1422         struct blkfront_info *info = rinfo->dev_info;
1423         struct blk_shadow *s = &rinfo->shadow[*id];
1424         struct copy_from_grant data = {
1425                 .grant_idx = 0,
1426         };
1427
1428         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1429                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1430
1431         /* The I/O request may be split in two. */
1432         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1433                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1434
1435                 /* Keep the status of the current response in shadow. */
1436                 s->status = blkif_rsp_to_req_status(bret->status);
1437
1438                 /* Wait the second response if not yet here. */
1439                 if (s2->status == REQ_WAITING)
1440                         return false;
1441
1442                 bret->status = blkif_get_final_status(s->status,
1443                                                       s2->status);
1444
1445                 /*
1446                  * All the grants is stored in the first shadow in order
1447                  * to make the completion code simpler.
1448                  */
1449                 num_grant += s2->req.u.rw.nr_segments;
1450
1451                 /*
1452                  * The two responses may not come in order. Only the
1453                  * first request will store the scatter-gather list.
1454                  */
1455                 if (s2->num_sg != 0) {
1456                         /* Update "id" with the ID of the first response. */
1457                         *id = s->associated_id;
1458                         s = s2;
1459                 }
1460
1461                 /*
1462                  * We don't need anymore the second request, so recycling
1463                  * it now.
1464                  */
1465                 if (add_id_to_freelist(rinfo, s->associated_id))
1466                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1467                              info->gd->disk_name, s->associated_id);
1468         }
1469
1470         data.s = s;
1471         num_sg = s->num_sg;
1472
1473         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1474                 for_each_sg(s->sg, sg, num_sg, i) {
1475                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1476
1477                         data.bvec_offset = sg->offset;
1478                         data.bvec_data = kmap_atomic(sg_page(sg));
1479
1480                         gnttab_foreach_grant_in_range(sg_page(sg),
1481                                                       sg->offset,
1482                                                       sg->length,
1483                                                       blkif_copy_from_grant,
1484                                                       &data);
1485
1486                         kunmap_atomic(data.bvec_data);
1487                 }
1488         }
1489         /* Add the persistent grant into the list of free grants */
1490         for (i = 0; i < num_grant; i++) {
1491                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1492                         /*
1493                          * If the grant is still mapped by the backend (the
1494                          * backend has chosen to make this grant persistent)
1495                          * we add it at the head of the list, so it will be
1496                          * reused first.
1497                          */
1498                         if (!info->feature_persistent)
1499                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1500                                                      s->grants_used[i]->gref);
1501                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1502                         rinfo->persistent_gnts_c++;
1503                 } else {
1504                         /*
1505                          * If the grant is not mapped by the backend we end the
1506                          * foreign access and add it to the tail of the list,
1507                          * so it will not be picked again unless we run out of
1508                          * persistent grants.
1509                          */
1510                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1511                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1512                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1513                 }
1514         }
1515         if (s->req.operation == BLKIF_OP_INDIRECT) {
1516                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1517                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1518                                 if (!info->feature_persistent)
1519                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1520                                                              s->indirect_grants[i]->gref);
1521                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1522                                 rinfo->persistent_gnts_c++;
1523                         } else {
1524                                 struct page *indirect_page;
1525
1526                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1527                                 /*
1528                                  * Add the used indirect page back to the list of
1529                                  * available pages for indirect grefs.
1530                                  */
1531                                 if (!info->feature_persistent) {
1532                                         indirect_page = s->indirect_grants[i]->page;
1533                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1534                                 }
1535                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1536                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1537                         }
1538                 }
1539         }
1540
1541         return true;
1542 }
1543
1544 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1545 {
1546         struct request *req;
1547         struct blkif_response *bret;
1548         RING_IDX i, rp;
1549         unsigned long flags;
1550         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1551         struct blkfront_info *info = rinfo->dev_info;
1552
1553         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1554                 return IRQ_HANDLED;
1555
1556         spin_lock_irqsave(&rinfo->ring_lock, flags);
1557  again:
1558         rp = rinfo->ring.sring->rsp_prod;
1559         rmb(); /* Ensure we see queued responses up to 'rp'. */
1560
1561         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1562                 unsigned long id;
1563
1564                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1565                 id   = bret->id;
1566                 /*
1567                  * The backend has messed up and given us an id that we would
1568                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1569                  * look in get_id_from_freelist.
1570                  */
1571                 if (id >= BLK_RING_SIZE(info)) {
1572                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1573                              info->gd->disk_name, op_name(bret->operation), id);
1574                         /* We can't safely get the 'struct request' as
1575                          * the id is busted. */
1576                         continue;
1577                 }
1578                 req  = rinfo->shadow[id].request;
1579
1580                 if (bret->operation != BLKIF_OP_DISCARD) {
1581                         /*
1582                          * We may need to wait for an extra response if the
1583                          * I/O request is split in 2
1584                          */
1585                         if (!blkif_completion(&id, rinfo, bret))
1586                                 continue;
1587                 }
1588
1589                 if (add_id_to_freelist(rinfo, id)) {
1590                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1591                              info->gd->disk_name, op_name(bret->operation), id);
1592                         continue;
1593                 }
1594
1595                 if (bret->status == BLKIF_RSP_OKAY)
1596                         blkif_req(req)->error = BLK_STS_OK;
1597                 else
1598                         blkif_req(req)->error = BLK_STS_IOERR;
1599
1600                 switch (bret->operation) {
1601                 case BLKIF_OP_DISCARD:
1602                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1603                                 struct request_queue *rq = info->rq;
1604                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1605                                            info->gd->disk_name, op_name(bret->operation));
1606                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1607                                 info->feature_discard = 0;
1608                                 info->feature_secdiscard = 0;
1609                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1610                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1611                         }
1612                         break;
1613                 case BLKIF_OP_FLUSH_DISKCACHE:
1614                 case BLKIF_OP_WRITE_BARRIER:
1615                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1616                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1617                                        info->gd->disk_name, op_name(bret->operation));
1618                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1619                         }
1620                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1621                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1622                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1623                                        info->gd->disk_name, op_name(bret->operation));
1624                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1625                         }
1626                         if (unlikely(blkif_req(req)->error)) {
1627                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1628                                         blkif_req(req)->error = BLK_STS_OK;
1629                                 info->feature_fua = 0;
1630                                 info->feature_flush = 0;
1631                                 xlvbd_flush(info);
1632                         }
1633                         /* fall through */
1634                 case BLKIF_OP_READ:
1635                 case BLKIF_OP_WRITE:
1636                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1637                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1638                                         "request: %x\n", bret->status);
1639
1640                         break;
1641                 default:
1642                         BUG();
1643                 }
1644
1645                 blk_mq_complete_request(req);
1646         }
1647
1648         rinfo->ring.rsp_cons = i;
1649
1650         if (i != rinfo->ring.req_prod_pvt) {
1651                 int more_to_do;
1652                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1653                 if (more_to_do)
1654                         goto again;
1655         } else
1656                 rinfo->ring.sring->rsp_event = i + 1;
1657
1658         kick_pending_request_queues_locked(rinfo);
1659
1660         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1661
1662         return IRQ_HANDLED;
1663 }
1664
1665
1666 static int setup_blkring(struct xenbus_device *dev,
1667                          struct blkfront_ring_info *rinfo)
1668 {
1669         struct blkif_sring *sring;
1670         int err, i;
1671         struct blkfront_info *info = rinfo->dev_info;
1672         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1673         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1674
1675         for (i = 0; i < info->nr_ring_pages; i++)
1676                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1677
1678         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1679                                                        get_order(ring_size));
1680         if (!sring) {
1681                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1682                 return -ENOMEM;
1683         }
1684         SHARED_RING_INIT(sring);
1685         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1686
1687         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1688         if (err < 0) {
1689                 free_pages((unsigned long)sring, get_order(ring_size));
1690                 rinfo->ring.sring = NULL;
1691                 goto fail;
1692         }
1693         for (i = 0; i < info->nr_ring_pages; i++)
1694                 rinfo->ring_ref[i] = gref[i];
1695
1696         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1697         if (err)
1698                 goto fail;
1699
1700         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1701                                         "blkif", rinfo);
1702         if (err <= 0) {
1703                 xenbus_dev_fatal(dev, err,
1704                                  "bind_evtchn_to_irqhandler failed");
1705                 goto fail;
1706         }
1707         rinfo->irq = err;
1708
1709         return 0;
1710 fail:
1711         blkif_free(info, 0);
1712         return err;
1713 }
1714
1715 /*
1716  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1717  * ring buffer may have multi pages depending on ->nr_ring_pages.
1718  */
1719 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1720                                 struct blkfront_ring_info *rinfo, const char *dir)
1721 {
1722         int err;
1723         unsigned int i;
1724         const char *message = NULL;
1725         struct blkfront_info *info = rinfo->dev_info;
1726
1727         if (info->nr_ring_pages == 1) {
1728                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1729                 if (err) {
1730                         message = "writing ring-ref";
1731                         goto abort_transaction;
1732                 }
1733         } else {
1734                 for (i = 0; i < info->nr_ring_pages; i++) {
1735                         char ring_ref_name[RINGREF_NAME_LEN];
1736
1737                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1738                         err = xenbus_printf(xbt, dir, ring_ref_name,
1739                                             "%u", rinfo->ring_ref[i]);
1740                         if (err) {
1741                                 message = "writing ring-ref";
1742                                 goto abort_transaction;
1743                         }
1744                 }
1745         }
1746
1747         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1748         if (err) {
1749                 message = "writing event-channel";
1750                 goto abort_transaction;
1751         }
1752
1753         return 0;
1754
1755 abort_transaction:
1756         xenbus_transaction_end(xbt, 1);
1757         if (message)
1758                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1759
1760         return err;
1761 }
1762
1763 static void free_info(struct blkfront_info *info)
1764 {
1765         list_del(&info->info_list);
1766         kfree(info);
1767 }
1768
1769 /* Common code used when first setting up, and when resuming. */
1770 static int talk_to_blkback(struct xenbus_device *dev,
1771                            struct blkfront_info *info)
1772 {
1773         const char *message = NULL;
1774         struct xenbus_transaction xbt;
1775         int err;
1776         unsigned int i, max_page_order;
1777         unsigned int ring_page_order;
1778
1779         if (!info)
1780                 return -ENODEV;
1781
1782         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1783                                               "max-ring-page-order", 0);
1784         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1785         info->nr_ring_pages = 1 << ring_page_order;
1786
1787         err = negotiate_mq(info);
1788         if (err)
1789                 goto destroy_blkring;
1790
1791         for (i = 0; i < info->nr_rings; i++) {
1792                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1793
1794                 /* Create shared ring, alloc event channel. */
1795                 err = setup_blkring(dev, rinfo);
1796                 if (err)
1797                         goto destroy_blkring;
1798         }
1799
1800 again:
1801         err = xenbus_transaction_start(&xbt);
1802         if (err) {
1803                 xenbus_dev_fatal(dev, err, "starting transaction");
1804                 goto destroy_blkring;
1805         }
1806
1807         if (info->nr_ring_pages > 1) {
1808                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1809                                     ring_page_order);
1810                 if (err) {
1811                         message = "writing ring-page-order";
1812                         goto abort_transaction;
1813                 }
1814         }
1815
1816         /* We already got the number of queues/rings in _probe */
1817         if (info->nr_rings == 1) {
1818                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1819                 if (err)
1820                         goto destroy_blkring;
1821         } else {
1822                 char *path;
1823                 size_t pathsize;
1824
1825                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1826                                     info->nr_rings);
1827                 if (err) {
1828                         message = "writing multi-queue-num-queues";
1829                         goto abort_transaction;
1830                 }
1831
1832                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1833                 path = kmalloc(pathsize, GFP_KERNEL);
1834                 if (!path) {
1835                         err = -ENOMEM;
1836                         message = "ENOMEM while writing ring references";
1837                         goto abort_transaction;
1838                 }
1839
1840                 for (i = 0; i < info->nr_rings; i++) {
1841                         memset(path, 0, pathsize);
1842                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1843                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1844                         if (err) {
1845                                 kfree(path);
1846                                 goto destroy_blkring;
1847                         }
1848                 }
1849                 kfree(path);
1850         }
1851         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1852                             XEN_IO_PROTO_ABI_NATIVE);
1853         if (err) {
1854                 message = "writing protocol";
1855                 goto abort_transaction;
1856         }
1857         err = xenbus_printf(xbt, dev->nodename,
1858                             "feature-persistent", "%u", 1);
1859         if (err)
1860                 dev_warn(&dev->dev,
1861                          "writing persistent grants feature to xenbus");
1862
1863         err = xenbus_transaction_end(xbt, 0);
1864         if (err) {
1865                 if (err == -EAGAIN)
1866                         goto again;
1867                 xenbus_dev_fatal(dev, err, "completing transaction");
1868                 goto destroy_blkring;
1869         }
1870
1871         for (i = 0; i < info->nr_rings; i++) {
1872                 unsigned int j;
1873                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1874
1875                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1876                         rinfo->shadow[j].req.u.rw.id = j + 1;
1877                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1878         }
1879         xenbus_switch_state(dev, XenbusStateInitialised);
1880
1881         return 0;
1882
1883  abort_transaction:
1884         xenbus_transaction_end(xbt, 1);
1885         if (message)
1886                 xenbus_dev_fatal(dev, err, "%s", message);
1887  destroy_blkring:
1888         blkif_free(info, 0);
1889
1890         mutex_lock(&blkfront_mutex);
1891         free_info(info);
1892         mutex_unlock(&blkfront_mutex);
1893
1894         dev_set_drvdata(&dev->dev, NULL);
1895
1896         return err;
1897 }
1898
1899 static int negotiate_mq(struct blkfront_info *info)
1900 {
1901         unsigned int backend_max_queues;
1902         unsigned int i;
1903
1904         BUG_ON(info->nr_rings);
1905
1906         /* Check if backend supports multiple queues. */
1907         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1908                                                   "multi-queue-max-queues", 1);
1909         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1910         /* We need at least one ring. */
1911         if (!info->nr_rings)
1912                 info->nr_rings = 1;
1913
1914         info->rinfo = kvcalloc(info->nr_rings,
1915                                struct_size(info->rinfo, shadow,
1916                                            BLK_RING_SIZE(info)),
1917                                GFP_KERNEL);
1918         if (!info->rinfo) {
1919                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1920                 info->nr_rings = 0;
1921                 return -ENOMEM;
1922         }
1923
1924         for (i = 0; i < info->nr_rings; i++) {
1925                 struct blkfront_ring_info *rinfo;
1926
1927                 rinfo = &info->rinfo[i];
1928                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1929                 INIT_LIST_HEAD(&rinfo->grants);
1930                 rinfo->dev_info = info;
1931                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1932                 spin_lock_init(&rinfo->ring_lock);
1933         }
1934         return 0;
1935 }
1936 /**
1937  * Entry point to this code when a new device is created.  Allocate the basic
1938  * structures and the ring buffer for communication with the backend, and
1939  * inform the backend of the appropriate details for those.  Switch to
1940  * Initialised state.
1941  */
1942 static int blkfront_probe(struct xenbus_device *dev,
1943                           const struct xenbus_device_id *id)
1944 {
1945         int err, vdevice;
1946         struct blkfront_info *info;
1947
1948         /* FIXME: Use dynamic device id if this is not set. */
1949         err = xenbus_scanf(XBT_NIL, dev->nodename,
1950                            "virtual-device", "%i", &vdevice);
1951         if (err != 1) {
1952                 /* go looking in the extended area instead */
1953                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1954                                    "%i", &vdevice);
1955                 if (err != 1) {
1956                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1957                         return err;
1958                 }
1959         }
1960
1961         if (xen_hvm_domain()) {
1962                 char *type;
1963                 int len;
1964                 /* no unplug has been done: do not hook devices != xen vbds */
1965                 if (xen_has_pv_and_legacy_disk_devices()) {
1966                         int major;
1967
1968                         if (!VDEV_IS_EXTENDED(vdevice))
1969                                 major = BLKIF_MAJOR(vdevice);
1970                         else
1971                                 major = XENVBD_MAJOR;
1972
1973                         if (major != XENVBD_MAJOR) {
1974                                 printk(KERN_INFO
1975                                                 "%s: HVM does not support vbd %d as xen block device\n",
1976                                                 __func__, vdevice);
1977                                 return -ENODEV;
1978                         }
1979                 }
1980                 /* do not create a PV cdrom device if we are an HVM guest */
1981                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1982                 if (IS_ERR(type))
1983                         return -ENODEV;
1984                 if (strncmp(type, "cdrom", 5) == 0) {
1985                         kfree(type);
1986                         return -ENODEV;
1987                 }
1988                 kfree(type);
1989         }
1990         info = kzalloc(sizeof(*info), GFP_KERNEL);
1991         if (!info) {
1992                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1993                 return -ENOMEM;
1994         }
1995
1996         info->xbdev = dev;
1997
1998         mutex_init(&info->mutex);
1999         info->vdevice = vdevice;
2000         info->connected = BLKIF_STATE_DISCONNECTED;
2001
2002         /* Front end dir is a number, which is used as the id. */
2003         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2004         dev_set_drvdata(&dev->dev, info);
2005
2006         mutex_lock(&blkfront_mutex);
2007         list_add(&info->info_list, &info_list);
2008         mutex_unlock(&blkfront_mutex);
2009
2010         return 0;
2011 }
2012
2013 static int blkif_recover(struct blkfront_info *info)
2014 {
2015         unsigned int r_index;
2016         struct request *req, *n;
2017         int rc;
2018         struct bio *bio;
2019         unsigned int segs;
2020
2021         blkfront_gather_backend_features(info);
2022         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2023         blkif_set_queue_limits(info);
2024         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2025         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2026
2027         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2028                 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2029
2030                 rc = blkfront_setup_indirect(rinfo);
2031                 if (rc)
2032                         return rc;
2033         }
2034         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2035
2036         /* Now safe for us to use the shared ring */
2037         info->connected = BLKIF_STATE_CONNECTED;
2038
2039         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2040                 struct blkfront_ring_info *rinfo;
2041
2042                 rinfo = &info->rinfo[r_index];
2043                 /* Kick any other new requests queued since we resumed */
2044                 kick_pending_request_queues(rinfo);
2045         }
2046
2047         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2048                 /* Requeue pending requests (flush or discard) */
2049                 list_del_init(&req->queuelist);
2050                 BUG_ON(req->nr_phys_segments > segs);
2051                 blk_mq_requeue_request(req, false);
2052         }
2053         blk_mq_start_stopped_hw_queues(info->rq, true);
2054         blk_mq_kick_requeue_list(info->rq);
2055
2056         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2057                 /* Traverse the list of pending bios and re-queue them */
2058                 submit_bio(bio);
2059         }
2060
2061         return 0;
2062 }
2063
2064 /**
2065  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2066  * driver restart.  We tear down our blkif structure and recreate it, but
2067  * leave the device-layer structures intact so that this is transparent to the
2068  * rest of the kernel.
2069  */
2070 static int blkfront_resume(struct xenbus_device *dev)
2071 {
2072         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2073         int err = 0;
2074         unsigned int i, j;
2075
2076         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2077
2078         bio_list_init(&info->bio_list);
2079         INIT_LIST_HEAD(&info->requests);
2080         for (i = 0; i < info->nr_rings; i++) {
2081                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2082                 struct bio_list merge_bio;
2083                 struct blk_shadow *shadow = rinfo->shadow;
2084
2085                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2086                         /* Not in use? */
2087                         if (!shadow[j].request)
2088                                 continue;
2089
2090                         /*
2091                          * Get the bios in the request so we can re-queue them.
2092                          */
2093                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2094                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2095                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2096                             shadow[j].request->cmd_flags & REQ_FUA) {
2097                                 /*
2098                                  * Flush operations don't contain bios, so
2099                                  * we need to requeue the whole request
2100                                  *
2101                                  * XXX: but this doesn't make any sense for a
2102                                  * write with the FUA flag set..
2103                                  */
2104                                 list_add(&shadow[j].request->queuelist, &info->requests);
2105                                 continue;
2106                         }
2107                         merge_bio.head = shadow[j].request->bio;
2108                         merge_bio.tail = shadow[j].request->biotail;
2109                         bio_list_merge(&info->bio_list, &merge_bio);
2110                         shadow[j].request->bio = NULL;
2111                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2112                 }
2113         }
2114
2115         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2116
2117         err = talk_to_blkback(dev, info);
2118         if (!err)
2119                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2120
2121         /*
2122          * We have to wait for the backend to switch to
2123          * connected state, since we want to read which
2124          * features it supports.
2125          */
2126
2127         return err;
2128 }
2129
2130 static void blkfront_closing(struct blkfront_info *info)
2131 {
2132         struct xenbus_device *xbdev = info->xbdev;
2133         struct block_device *bdev = NULL;
2134
2135         mutex_lock(&info->mutex);
2136
2137         if (xbdev->state == XenbusStateClosing) {
2138                 mutex_unlock(&info->mutex);
2139                 return;
2140         }
2141
2142         if (info->gd)
2143                 bdev = bdget_disk(info->gd, 0);
2144
2145         mutex_unlock(&info->mutex);
2146
2147         if (!bdev) {
2148                 xenbus_frontend_closed(xbdev);
2149                 return;
2150         }
2151
2152         mutex_lock(&bdev->bd_mutex);
2153
2154         if (bdev->bd_openers) {
2155                 xenbus_dev_error(xbdev, -EBUSY,
2156                                  "Device in use; refusing to close");
2157                 xenbus_switch_state(xbdev, XenbusStateClosing);
2158         } else {
2159                 xlvbd_release_gendisk(info);
2160                 xenbus_frontend_closed(xbdev);
2161         }
2162
2163         mutex_unlock(&bdev->bd_mutex);
2164         bdput(bdev);
2165 }
2166
2167 static void blkfront_setup_discard(struct blkfront_info *info)
2168 {
2169         int err;
2170         unsigned int discard_granularity;
2171         unsigned int discard_alignment;
2172
2173         info->feature_discard = 1;
2174         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2175                 "discard-granularity", "%u", &discard_granularity,
2176                 "discard-alignment", "%u", &discard_alignment,
2177                 NULL);
2178         if (!err) {
2179                 info->discard_granularity = discard_granularity;
2180                 info->discard_alignment = discard_alignment;
2181         }
2182         info->feature_secdiscard =
2183                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2184                                        0);
2185 }
2186
2187 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2188 {
2189         unsigned int psegs, grants;
2190         int err, i;
2191         struct blkfront_info *info = rinfo->dev_info;
2192
2193         if (info->max_indirect_segments == 0) {
2194                 if (!HAS_EXTRA_REQ)
2195                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2196                 else {
2197                         /*
2198                          * When an extra req is required, the maximum
2199                          * grants supported is related to the size of the
2200                          * Linux block segment.
2201                          */
2202                         grants = GRANTS_PER_PSEG;
2203                 }
2204         }
2205         else
2206                 grants = info->max_indirect_segments;
2207         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2208
2209         err = fill_grant_buffer(rinfo,
2210                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2211         if (err)
2212                 goto out_of_memory;
2213
2214         if (!info->feature_persistent && info->max_indirect_segments) {
2215                 /*
2216                  * We are using indirect descriptors but not persistent
2217                  * grants, we need to allocate a set of pages that can be
2218                  * used for mapping indirect grefs
2219                  */
2220                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2221
2222                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2223                 for (i = 0; i < num; i++) {
2224                         struct page *indirect_page = alloc_page(GFP_NOIO);
2225                         if (!indirect_page)
2226                                 goto out_of_memory;
2227                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2228                 }
2229         }
2230
2231         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2232                 rinfo->shadow[i].grants_used =
2233                         kvcalloc(grants,
2234                                  sizeof(rinfo->shadow[i].grants_used[0]),
2235                                  GFP_NOIO);
2236                 rinfo->shadow[i].sg = kvcalloc(psegs,
2237                                                sizeof(rinfo->shadow[i].sg[0]),
2238                                                GFP_NOIO);
2239                 if (info->max_indirect_segments)
2240                         rinfo->shadow[i].indirect_grants =
2241                                 kvcalloc(INDIRECT_GREFS(grants),
2242                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2243                                          GFP_NOIO);
2244                 if ((rinfo->shadow[i].grants_used == NULL) ||
2245                         (rinfo->shadow[i].sg == NULL) ||
2246                      (info->max_indirect_segments &&
2247                      (rinfo->shadow[i].indirect_grants == NULL)))
2248                         goto out_of_memory;
2249                 sg_init_table(rinfo->shadow[i].sg, psegs);
2250         }
2251
2252
2253         return 0;
2254
2255 out_of_memory:
2256         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2257                 kvfree(rinfo->shadow[i].grants_used);
2258                 rinfo->shadow[i].grants_used = NULL;
2259                 kvfree(rinfo->shadow[i].sg);
2260                 rinfo->shadow[i].sg = NULL;
2261                 kvfree(rinfo->shadow[i].indirect_grants);
2262                 rinfo->shadow[i].indirect_grants = NULL;
2263         }
2264         if (!list_empty(&rinfo->indirect_pages)) {
2265                 struct page *indirect_page, *n;
2266                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2267                         list_del(&indirect_page->lru);
2268                         __free_page(indirect_page);
2269                 }
2270         }
2271         return -ENOMEM;
2272 }
2273
2274 /*
2275  * Gather all backend feature-*
2276  */
2277 static void blkfront_gather_backend_features(struct blkfront_info *info)
2278 {
2279         unsigned int indirect_segments;
2280
2281         info->feature_flush = 0;
2282         info->feature_fua = 0;
2283
2284         /*
2285          * If there's no "feature-barrier" defined, then it means
2286          * we're dealing with a very old backend which writes
2287          * synchronously; nothing to do.
2288          *
2289          * If there are barriers, then we use flush.
2290          */
2291         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2292                 info->feature_flush = 1;
2293                 info->feature_fua = 1;
2294         }
2295
2296         /*
2297          * And if there is "feature-flush-cache" use that above
2298          * barriers.
2299          */
2300         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2301                                  0)) {
2302                 info->feature_flush = 1;
2303                 info->feature_fua = 0;
2304         }
2305
2306         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2307                 blkfront_setup_discard(info);
2308
2309         info->feature_persistent =
2310                 !!xenbus_read_unsigned(info->xbdev->otherend,
2311                                        "feature-persistent", 0);
2312
2313         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2314                                         "feature-max-indirect-segments", 0);
2315         if (indirect_segments > xen_blkif_max_segments)
2316                 indirect_segments = xen_blkif_max_segments;
2317         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2318                 indirect_segments = 0;
2319         info->max_indirect_segments = indirect_segments;
2320
2321         if (info->feature_persistent) {
2322                 mutex_lock(&blkfront_mutex);
2323                 schedule_delayed_work(&blkfront_work, HZ * 10);
2324                 mutex_unlock(&blkfront_mutex);
2325         }
2326 }
2327
2328 /*
2329  * Invoked when the backend is finally 'ready' (and has told produced
2330  * the details about the physical device - #sectors, size, etc).
2331  */
2332 static void blkfront_connect(struct blkfront_info *info)
2333 {
2334         unsigned long long sectors;
2335         unsigned long sector_size;
2336         unsigned int physical_sector_size;
2337         unsigned int binfo;
2338         char *envp[] = { "RESIZE=1", NULL };
2339         int err, i;
2340
2341         switch (info->connected) {
2342         case BLKIF_STATE_CONNECTED:
2343                 /*
2344                  * Potentially, the back-end may be signalling
2345                  * a capacity change; update the capacity.
2346                  */
2347                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2348                                    "sectors", "%Lu", &sectors);
2349                 if (XENBUS_EXIST_ERR(err))
2350                         return;
2351                 printk(KERN_INFO "Setting capacity to %Lu\n",
2352                        sectors);
2353                 set_capacity(info->gd, sectors);
2354                 revalidate_disk(info->gd);
2355                 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2356                                    KOBJ_CHANGE, envp);
2357
2358                 return;
2359         case BLKIF_STATE_SUSPENDED:
2360                 /*
2361                  * If we are recovering from suspension, we need to wait
2362                  * for the backend to announce it's features before
2363                  * reconnecting, at least we need to know if the backend
2364                  * supports indirect descriptors, and how many.
2365                  */
2366                 blkif_recover(info);
2367                 return;
2368
2369         default:
2370                 break;
2371         }
2372
2373         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2374                 __func__, info->xbdev->otherend);
2375
2376         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2377                             "sectors", "%llu", &sectors,
2378                             "info", "%u", &binfo,
2379                             "sector-size", "%lu", &sector_size,
2380                             NULL);
2381         if (err) {
2382                 xenbus_dev_fatal(info->xbdev, err,
2383                                  "reading backend fields at %s",
2384                                  info->xbdev->otherend);
2385                 return;
2386         }
2387
2388         /*
2389          * physcial-sector-size is a newer field, so old backends may not
2390          * provide this. Assume physical sector size to be the same as
2391          * sector_size in that case.
2392          */
2393         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2394                                                     "physical-sector-size",
2395                                                     sector_size);
2396         blkfront_gather_backend_features(info);
2397         for (i = 0; i < info->nr_rings; i++) {
2398                 err = blkfront_setup_indirect(&info->rinfo[i]);
2399                 if (err) {
2400                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2401                                          info->xbdev->otherend);
2402                         blkif_free(info, 0);
2403                         break;
2404                 }
2405         }
2406
2407         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2408                                   physical_sector_size);
2409         if (err) {
2410                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2411                                  info->xbdev->otherend);
2412                 goto fail;
2413         }
2414
2415         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2416
2417         /* Kick pending requests. */
2418         info->connected = BLKIF_STATE_CONNECTED;
2419         for (i = 0; i < info->nr_rings; i++)
2420                 kick_pending_request_queues(&info->rinfo[i]);
2421
2422         device_add_disk(&info->xbdev->dev, info->gd, NULL);
2423
2424         info->is_ready = 1;
2425         return;
2426
2427 fail:
2428         blkif_free(info, 0);
2429         return;
2430 }
2431
2432 /**
2433  * Callback received when the backend's state changes.
2434  */
2435 static void blkback_changed(struct xenbus_device *dev,
2436                             enum xenbus_state backend_state)
2437 {
2438         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2439
2440         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2441
2442         switch (backend_state) {
2443         case XenbusStateInitWait:
2444                 if (dev->state != XenbusStateInitialising)
2445                         break;
2446                 if (talk_to_blkback(dev, info))
2447                         break;
2448         case XenbusStateInitialising:
2449         case XenbusStateInitialised:
2450         case XenbusStateReconfiguring:
2451         case XenbusStateReconfigured:
2452         case XenbusStateUnknown:
2453                 break;
2454
2455         case XenbusStateConnected:
2456                 /*
2457                  * talk_to_blkback sets state to XenbusStateInitialised
2458                  * and blkfront_connect sets it to XenbusStateConnected
2459                  * (if connection went OK).
2460                  *
2461                  * If the backend (or toolstack) decides to poke at backend
2462                  * state (and re-trigger the watch by setting the state repeatedly
2463                  * to XenbusStateConnected (4)) we need to deal with this.
2464                  * This is allowed as this is used to communicate to the guest
2465                  * that the size of disk has changed!
2466                  */
2467                 if ((dev->state != XenbusStateInitialised) &&
2468                     (dev->state != XenbusStateConnected)) {
2469                         if (talk_to_blkback(dev, info))
2470                                 break;
2471                 }
2472
2473                 blkfront_connect(info);
2474                 break;
2475
2476         case XenbusStateClosed:
2477                 if (dev->state == XenbusStateClosed)
2478                         break;
2479                 /* fall through */
2480         case XenbusStateClosing:
2481                 if (info)
2482                         blkfront_closing(info);
2483                 break;
2484         }
2485 }
2486
2487 static int blkfront_remove(struct xenbus_device *xbdev)
2488 {
2489         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2490         struct block_device *bdev = NULL;
2491         struct gendisk *disk;
2492
2493         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2494
2495         if (!info)
2496                 return 0;
2497
2498         blkif_free(info, 0);
2499
2500         mutex_lock(&info->mutex);
2501
2502         disk = info->gd;
2503         if (disk)
2504                 bdev = bdget_disk(disk, 0);
2505
2506         info->xbdev = NULL;
2507         mutex_unlock(&info->mutex);
2508
2509         if (!bdev) {
2510                 mutex_lock(&blkfront_mutex);
2511                 free_info(info);
2512                 mutex_unlock(&blkfront_mutex);
2513                 return 0;
2514         }
2515
2516         /*
2517          * The xbdev was removed before we reached the Closed
2518          * state. See if it's safe to remove the disk. If the bdev
2519          * isn't closed yet, we let release take care of it.
2520          */
2521
2522         mutex_lock(&bdev->bd_mutex);
2523         info = disk->private_data;
2524
2525         dev_warn(disk_to_dev(disk),
2526                  "%s was hot-unplugged, %d stale handles\n",
2527                  xbdev->nodename, bdev->bd_openers);
2528
2529         if (info && !bdev->bd_openers) {
2530                 xlvbd_release_gendisk(info);
2531                 disk->private_data = NULL;
2532                 mutex_lock(&blkfront_mutex);
2533                 free_info(info);
2534                 mutex_unlock(&blkfront_mutex);
2535         }
2536
2537         mutex_unlock(&bdev->bd_mutex);
2538         bdput(bdev);
2539
2540         return 0;
2541 }
2542
2543 static int blkfront_is_ready(struct xenbus_device *dev)
2544 {
2545         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2546
2547         return info->is_ready && info->xbdev;
2548 }
2549
2550 static int blkif_open(struct block_device *bdev, fmode_t mode)
2551 {
2552         struct gendisk *disk = bdev->bd_disk;
2553         struct blkfront_info *info;
2554         int err = 0;
2555
2556         mutex_lock(&blkfront_mutex);
2557
2558         info = disk->private_data;
2559         if (!info) {
2560                 /* xbdev gone */
2561                 err = -ERESTARTSYS;
2562                 goto out;
2563         }
2564
2565         mutex_lock(&info->mutex);
2566
2567         if (!info->gd)
2568                 /* xbdev is closed */
2569                 err = -ERESTARTSYS;
2570
2571         mutex_unlock(&info->mutex);
2572
2573 out:
2574         mutex_unlock(&blkfront_mutex);
2575         return err;
2576 }
2577
2578 static void blkif_release(struct gendisk *disk, fmode_t mode)
2579 {
2580         struct blkfront_info *info = disk->private_data;
2581         struct block_device *bdev;
2582         struct xenbus_device *xbdev;
2583
2584         mutex_lock(&blkfront_mutex);
2585
2586         bdev = bdget_disk(disk, 0);
2587
2588         if (!bdev) {
2589                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2590                 goto out_mutex;
2591         }
2592         if (bdev->bd_openers)
2593                 goto out;
2594
2595         /*
2596          * Check if we have been instructed to close. We will have
2597          * deferred this request, because the bdev was still open.
2598          */
2599
2600         mutex_lock(&info->mutex);
2601         xbdev = info->xbdev;
2602
2603         if (xbdev && xbdev->state == XenbusStateClosing) {
2604                 /* pending switch to state closed */
2605                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2606                 xlvbd_release_gendisk(info);
2607                 xenbus_frontend_closed(info->xbdev);
2608         }
2609
2610         mutex_unlock(&info->mutex);
2611
2612         if (!xbdev) {
2613                 /* sudden device removal */
2614                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2615                 xlvbd_release_gendisk(info);
2616                 disk->private_data = NULL;
2617                 free_info(info);
2618         }
2619
2620 out:
2621         bdput(bdev);
2622 out_mutex:
2623         mutex_unlock(&blkfront_mutex);
2624 }
2625
2626 static const struct block_device_operations xlvbd_block_fops =
2627 {
2628         .owner = THIS_MODULE,
2629         .open = blkif_open,
2630         .release = blkif_release,
2631         .getgeo = blkif_getgeo,
2632         .ioctl = blkif_ioctl,
2633         .compat_ioctl = blkdev_compat_ptr_ioctl,
2634 };
2635
2636
2637 static const struct xenbus_device_id blkfront_ids[] = {
2638         { "vbd" },
2639         { "" }
2640 };
2641
2642 static struct xenbus_driver blkfront_driver = {
2643         .ids  = blkfront_ids,
2644         .probe = blkfront_probe,
2645         .remove = blkfront_remove,
2646         .resume = blkfront_resume,
2647         .otherend_changed = blkback_changed,
2648         .is_ready = blkfront_is_ready,
2649 };
2650
2651 static void purge_persistent_grants(struct blkfront_info *info)
2652 {
2653         unsigned int i;
2654         unsigned long flags;
2655
2656         for (i = 0; i < info->nr_rings; i++) {
2657                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2658                 struct grant *gnt_list_entry, *tmp;
2659
2660                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2661
2662                 if (rinfo->persistent_gnts_c == 0) {
2663                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2664                         continue;
2665                 }
2666
2667                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2668                                          node) {
2669                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2670                             gnttab_query_foreign_access(gnt_list_entry->gref))
2671                                 continue;
2672
2673                         list_del(&gnt_list_entry->node);
2674                         gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2675                         rinfo->persistent_gnts_c--;
2676                         gnt_list_entry->gref = GRANT_INVALID_REF;
2677                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2678                 }
2679
2680                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2681         }
2682 }
2683
2684 static void blkfront_delay_work(struct work_struct *work)
2685 {
2686         struct blkfront_info *info;
2687         bool need_schedule_work = false;
2688
2689         mutex_lock(&blkfront_mutex);
2690
2691         list_for_each_entry(info, &info_list, info_list) {
2692                 if (info->feature_persistent) {
2693                         need_schedule_work = true;
2694                         mutex_lock(&info->mutex);
2695                         purge_persistent_grants(info);
2696                         mutex_unlock(&info->mutex);
2697                 }
2698         }
2699
2700         if (need_schedule_work)
2701                 schedule_delayed_work(&blkfront_work, HZ * 10);
2702
2703         mutex_unlock(&blkfront_mutex);
2704 }
2705
2706 static int __init xlblk_init(void)
2707 {
2708         int ret;
2709         int nr_cpus = num_online_cpus();
2710
2711         if (!xen_domain())
2712                 return -ENODEV;
2713
2714         if (!xen_has_pv_disk_devices())
2715                 return -ENODEV;
2716
2717         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2718                 pr_warn("xen_blk: can't get major %d with name %s\n",
2719                         XENVBD_MAJOR, DEV_NAME);
2720                 return -ENODEV;
2721         }
2722
2723         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2724                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2725
2726         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2727                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2728                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2729                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2730         }
2731
2732         if (xen_blkif_max_queues > nr_cpus) {
2733                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2734                         xen_blkif_max_queues, nr_cpus);
2735                 xen_blkif_max_queues = nr_cpus;
2736         }
2737
2738         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2739
2740         ret = xenbus_register_frontend(&blkfront_driver);
2741         if (ret) {
2742                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2743                 return ret;
2744         }
2745
2746         return 0;
2747 }
2748 module_init(xlblk_init);
2749
2750
2751 static void __exit xlblk_exit(void)
2752 {
2753         cancel_delayed_work_sync(&blkfront_work);
2754
2755         xenbus_unregister_driver(&blkfront_driver);
2756         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2757         kfree(minors);
2758 }
2759 module_exit(xlblk_exit);
2760
2761 MODULE_DESCRIPTION("Xen virtual block device frontend");
2762 MODULE_LICENSE("GPL");
2763 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2764 MODULE_ALIAS("xen:vbd");
2765 MODULE_ALIAS("xenblk");