drivers/block/xen-blkfront.c

   1 /*
   2  * blkfront.c
   3  *
   4  * XenLinux virtual block device driver.
   5  *
   6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8  * Copyright (c) 2004, Christian Limpach
   9  * Copyright (c) 2004, Andrew Warfield
  10  * Copyright (c) 2005, Christopher Clark
  11  * Copyright (c) 2005, XenSource Ltd
  12  *
  13  * This program is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU General Public License version 2
  15  * as published by the Free Software Foundation; or, when distributed
  16  * separately from the Linux kernel or incorporated into other
  17  * software packages, subject to the following license:
  18  *
  19  * Permission is hereby granted, free of charge, to any person obtaining a copy
  20  * of this source file (the "Software"), to deal in the Software without
  21  * restriction, including without limitation the rights to use, copy, modify,
  22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23  * and to permit persons to whom the Software is furnished to do so, subject to
  24  * the following conditions:
  25  *
  26  * The above copyright notice and this permission notice shall be included in
  27  * all copies or substantial portions of the Software.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35  * IN THE SOFTWARE.
  36  */
  37
  38 #include <linux/interrupt.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/blk-mq.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/cdrom.h>
  43 #include <linux/module.h>
  44 #include <linux/slab.h>
  45 #include <linux/mutex.h>
  46 #include <linux/scatterlist.h>
  47 #include <linux/bitmap.h>
  48 #include <linux/list.h>
  49
  50 #include <xen/xen.h>
  51 #include <xen/xenbus.h>
  52 #include <xen/grant_table.h>
  53 #include <xen/events.h>
  54 #include <xen/page.h>
  55 #include <xen/platform_pci.h>
  56
  57 #include <xen/interface/grant_table.h>
  58 #include <xen/interface/io/blkif.h>
  59 #include <xen/interface/io/protocols.h>
  60
  61 #include <asm/xen/hypervisor.h>
  62
  63 /*
  64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  65  * When Linux is using a different page size than Xen, it may not be possible
  66  * to put all the data in a single segment.
  67  * This can happen when the backend doesn't support indirect descriptor and
  68  * therefore the maximum amount of data that a request can carry is
  69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  70  *
  71  * Note that we only support one extra request. So the Linux page size
  72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  73  * 88KB.
  74  */
  75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  76
  77 enum blkif_state {
  78         BLKIF_STATE_DISCONNECTED,
  79         BLKIF_STATE_CONNECTED,
  80         BLKIF_STATE_SUSPENDED,
  81 };
  82
  83 struct grant {
  84         grant_ref_t gref;
  85         struct page *page;
  86         struct list_head node;
  87 };
  88
  89 enum blk_req_status {
  90         REQ_WAITING,
  91         REQ_DONE,
  92         REQ_ERROR,
  93         REQ_EOPNOTSUPP,
  94 };
  95
  96 struct blk_shadow {
  97         struct blkif_request req;
  98         struct request *request;
  99         struct grant **grants_used;
 100         struct grant **indirect_grants;
 101         struct scatterlist *sg;
 102         unsigned int num_sg;
 103         enum blk_req_status status;
 104
 105         #define NO_ASSOCIATED_ID ~0UL
 106         /*
 107          * Id of the sibling if we ever need 2 requests when handling a
 108          * block I/O request
 109          */
 110         unsigned long associated_id;
 111 };
 112
 113 struct split_bio {
 114         struct bio *bio;
 115         atomic_t pending;
 116 };
 117
 118 struct blkif_req {
 119         int     error;
 120 };
 121
 122 static inline struct blkif_req *blkif_req(struct request *rq)
 123 {
 124         return blk_mq_rq_to_pdu(rq);
 125 }
 126
 127 static DEFINE_MUTEX(blkfront_mutex);
 128 static const struct block_device_operations xlvbd_block_fops;
 129
 130 /*
 131  * Maximum number of segments in indirect requests, the actual value used by
 132  * the frontend driver is the minimum of this value and the value provided
 133  * by the backend driver.
 134  */
 135
 136 static unsigned int xen_blkif_max_segments = 32;
 137 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
 138                    S_IRUGO);
 139 MODULE_PARM_DESC(max_indirect_segments,
 140                  "Maximum amount of segments in indirect requests (default is 32)");
 141
 142 static unsigned int xen_blkif_max_queues = 4;
 143 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
 144 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 145
 146 /*
 147  * Maximum order of pages to be used for the shared ring between front and
 148  * backend, 4KB page granularity is used.
 149  */
 150 static unsigned int xen_blkif_max_ring_order;
 151 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
 152 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 153
 154 #define BLK_RING_SIZE(info)     \
 155         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 156
 157 #define BLK_MAX_RING_SIZE       \
 158         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
 159
 160 /*
 161  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 162  * characters are enough. Define to 20 to keep consistent with backend.
 163  */
 164 #define RINGREF_NAME_LEN (20)
 165 /*
 166  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 167  */
 168 #define QUEUE_NAME_LEN (17)
 169
 170 /*
 171  *  Per-ring info.
 172  *  Every blkfront device can associate with one or more blkfront_ring_info,
 173  *  depending on how many hardware queues/rings to be used.
 174  */
 175 struct blkfront_ring_info {
 176         /* Lock to protect data in every ring buffer. */
 177         spinlock_t ring_lock;
 178         struct blkif_front_ring ring;
 179         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 180         unsigned int evtchn, irq;
 181         struct work_struct work;
 182         struct gnttab_free_callback callback;
 183         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 184         struct list_head indirect_pages;
 185         struct list_head grants;
 186         unsigned int persistent_gnts_c;
 187         unsigned long shadow_free;
 188         struct blkfront_info *dev_info;
 189 };
 190
 191 /*
 192  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 193  * hang in private_data off the gendisk structure. We may end up
 194  * putting all kinds of interesting stuff here :-)
 195  */
 196 struct blkfront_info
 197 {
 198         struct mutex mutex;
 199         struct xenbus_device *xbdev;
 200         struct gendisk *gd;
 201         u16 sector_size;
 202         unsigned int physical_sector_size;
 203         int vdevice;
 204         blkif_vdev_t handle;
 205         enum blkif_state connected;
 206         /* Number of pages per ring buffer. */
 207         unsigned int nr_ring_pages;
 208         struct request_queue *rq;
 209         unsigned int feature_flush:1;
 210         unsigned int feature_fua:1;
 211         unsigned int feature_discard:1;
 212         unsigned int feature_secdiscard:1;
 213         unsigned int feature_persistent:1;
 214         unsigned int discard_granularity;
 215         unsigned int discard_alignment;
 216         /* Number of 4KB segments handled */
 217         unsigned int max_indirect_segments;
 218         int is_ready;
 219         struct blk_mq_tag_set tag_set;
 220         struct blkfront_ring_info *rinfo;
 221         unsigned int nr_rings;
 222         /* Save uncomplete reqs and bios for migration. */
 223         struct list_head requests;
 224         struct bio_list bio_list;
 225 };
 226
 227 static unsigned int nr_minors;
 228 static unsigned long *minors;
 229 static DEFINE_SPINLOCK(minor_lock);
 230
 231 #define GRANT_INVALID_REF       0
 232
 233 #define PARTS_PER_DISK          16
 234 #define PARTS_PER_EXT_DISK      256
 235
 236 #define BLKIF_MAJOR(dev) ((dev)>>8)
 237 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 238
 239 #define EXT_SHIFT 28
 240 #define EXTENDED (1<<EXT_SHIFT)
 241 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 242 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 243 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 244 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 245 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 246 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 247
 248 #define DEV_NAME        "xvd"   /* name in /dev */
 249
 250 /*
 251  * Grants are always the same size as a Xen page (i.e 4KB).
 252  * A physical segment is always the same size as a Linux page.
 253  * Number of grants per physical segment
 254  */
 255 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 256
 257 #define GRANTS_PER_INDIRECT_FRAME \
 258         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 259
 260 #define PSEGS_PER_INDIRECT_FRAME        \
 261         (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
 262
 263 #define INDIRECT_GREFS(_grants)         \
 264         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 265
 266 #define GREFS(_psegs)   ((_psegs) * GRANTS_PER_PSEG)
 267
 268 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 269 static void blkfront_gather_backend_features(struct blkfront_info *info);
 270
 271 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 272 {
 273         unsigned long free = rinfo->shadow_free;
 274
 275         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 276         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 277         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 278         return free;
 279 }
 280
 281 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 282                               unsigned long id)
 283 {
 284         if (rinfo->shadow[id].req.u.rw.id != id)
 285                 return -EINVAL;
 286         if (rinfo->shadow[id].request == NULL)
 287                 return -EINVAL;
 288         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 289         rinfo->shadow[id].request = NULL;
 290         rinfo->shadow_free = id;
 291         return 0;
 292 }
 293
 294 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 295 {
 296         struct blkfront_info *info = rinfo->dev_info;
 297         struct page *granted_page;
 298         struct grant *gnt_list_entry, *n;
 299         int i = 0;
 300
 301         while (i < num) {
 302                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 303                 if (!gnt_list_entry)
 304                         goto out_of_memory;
 305
 306                 if (info->feature_persistent) {
 307                         granted_page = alloc_page(GFP_NOIO);
 308                         if (!granted_page) {
 309                                 kfree(gnt_list_entry);
 310                                 goto out_of_memory;
 311                         }
 312                         gnt_list_entry->page = granted_page;
 313                 }
 314
 315                 gnt_list_entry->gref = GRANT_INVALID_REF;
 316                 list_add(&gnt_list_entry->node, &rinfo->grants);
 317                 i++;
 318         }
 319
 320         return 0;
 321
 322 out_of_memory:
 323         list_for_each_entry_safe(gnt_list_entry, n,
 324                                  &rinfo->grants, node) {
 325                 list_del(&gnt_list_entry->node);
 326                 if (info->feature_persistent)
 327                         __free_page(gnt_list_entry->page);
 328                 kfree(gnt_list_entry);
 329                 i--;
 330         }
 331         BUG_ON(i != 0);
 332         return -ENOMEM;
 333 }
 334
 335 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 336 {
 337         struct grant *gnt_list_entry;
 338
 339         BUG_ON(list_empty(&rinfo->grants));
 340         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 341                                           node);
 342         list_del(&gnt_list_entry->node);
 343
 344         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 345                 rinfo->persistent_gnts_c--;
 346
 347         return gnt_list_entry;
 348 }
 349
 350 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 351                                         const struct blkfront_info *info)
 352 {
 353         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 354                                                  info->xbdev->otherend_id,
 355                                                  gnt_list_entry->page,
 356                                                  0);
 357 }
 358
 359 static struct grant *get_grant(grant_ref_t *gref_head,
 360                                unsigned long gfn,
 361                                struct blkfront_ring_info *rinfo)
 362 {
 363         struct grant *gnt_list_entry = get_free_grant(rinfo);
 364         struct blkfront_info *info = rinfo->dev_info;
 365
 366         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 367                 return gnt_list_entry;
 368
 369         /* Assign a gref to this page */
 370         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 371         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 372         if (info->feature_persistent)
 373                 grant_foreign_access(gnt_list_entry, info);
 374         else {
 375                 /* Grant access to the GFN passed by the caller */
 376                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 377                                                 info->xbdev->otherend_id,
 378                                                 gfn, 0);
 379         }
 380
 381         return gnt_list_entry;
 382 }
 383
 384 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 385                                         struct blkfront_ring_info *rinfo)
 386 {
 387         struct grant *gnt_list_entry = get_free_grant(rinfo);
 388         struct blkfront_info *info = rinfo->dev_info;
 389
 390         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 391                 return gnt_list_entry;
 392
 393         /* Assign a gref to this page */
 394         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 395         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 396         if (!info->feature_persistent) {
 397                 struct page *indirect_page;
 398
 399                 /* Fetch a pre-allocated page to use for indirect grefs */
 400                 BUG_ON(list_empty(&rinfo->indirect_pages));
 401                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 402                                                  struct page, lru);
 403                 list_del(&indirect_page->lru);
 404                 gnt_list_entry->page = indirect_page;
 405         }
 406         grant_foreign_access(gnt_list_entry, info);
 407
 408         return gnt_list_entry;
 409 }
 410
 411 static const char *op_name(int op)
 412 {
 413         static const char *const names[] = {
 414                 [BLKIF_OP_READ] = "read",
 415                 [BLKIF_OP_WRITE] = "write",
 416                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 417                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 418                 [BLKIF_OP_DISCARD] = "discard" };
 419
 420         if (op < 0 || op >= ARRAY_SIZE(names))
 421                 return "unknown";
 422
 423         if (!names[op])
 424                 return "reserved";
 425
 426         return names[op];
 427 }
 428 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 429 {
 430         unsigned int end = minor + nr;
 431         int rc;
 432
 433         if (end > nr_minors) {
 434                 unsigned long *bitmap, *old;
 435
 436                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 437                                  GFP_KERNEL);
 438                 if (bitmap == NULL)
 439                         return -ENOMEM;
 440
 441                 spin_lock(&minor_lock);
 442                 if (end > nr_minors) {
 443                         old = minors;
 444                         memcpy(bitmap, minors,
 445                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 446                         minors = bitmap;
 447                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 448                 } else
 449                         old = bitmap;
 450                 spin_unlock(&minor_lock);
 451                 kfree(old);
 452         }
 453
 454         spin_lock(&minor_lock);
 455         if (find_next_bit(minors, end, minor) >= end) {
 456                 bitmap_set(minors, minor, nr);
 457                 rc = 0;
 458         } else
 459                 rc = -EBUSY;
 460         spin_unlock(&minor_lock);
 461
 462         return rc;
 463 }
 464
 465 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 466 {
 467         unsigned int end = minor + nr;
 468
 469         BUG_ON(end > nr_minors);
 470         spin_lock(&minor_lock);
 471         bitmap_clear(minors,  minor, nr);
 472         spin_unlock(&minor_lock);
 473 }
 474
 475 static void blkif_restart_queue_callback(void *arg)
 476 {
 477         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 478         schedule_work(&rinfo->work);
 479 }
 480
 481 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 482 {
 483         /* We don't have real geometry info, but let's at least return
 484            values consistent with the size of the device */
 485         sector_t nsect = get_capacity(bd->bd_disk);
 486         sector_t cylinders = nsect;
 487
 488         hg->heads = 0xff;
 489         hg->sectors = 0x3f;
 490         sector_div(cylinders, hg->heads * hg->sectors);
 491         hg->cylinders = cylinders;
 492         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 493                 hg->cylinders = 0xffff;
 494         return 0;
 495 }
 496
 497 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 498                        unsigned command, unsigned long argument)
 499 {
 500         struct blkfront_info *info = bdev->bd_disk->private_data;
 501         int i;
 502
 503         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 504                 command, (long)argument);
 505
 506         switch (command) {
 507         case CDROMMULTISESSION:
 508                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 509                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 510                         if (put_user(0, (char __user *)(argument + i)))
 511                                 return -EFAULT;
 512                 return 0;
 513
 514         case CDROM_GET_CAPABILITY: {
 515                 struct gendisk *gd = info->gd;
 516                 if (gd->flags & GENHD_FL_CD)
 517                         return 0;
 518                 return -EINVAL;
 519         }
 520
 521         default:
 522                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 523                   command);*/
 524                 return -EINVAL; /* same return as native Linux */
 525         }
 526
 527         return 0;
 528 }
 529
 530 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 531                                             struct request *req,
 532                                             struct blkif_request **ring_req)
 533 {
 534         unsigned long id;
 535
 536         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 537         rinfo->ring.req_prod_pvt++;
 538
 539         id = get_id_from_freelist(rinfo);
 540         rinfo->shadow[id].request = req;
 541         rinfo->shadow[id].status = REQ_WAITING;
 542         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 543
 544         (*ring_req)->u.rw.id = id;
 545
 546         return id;
 547 }
 548
 549 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 550 {
 551         struct blkfront_info *info = rinfo->dev_info;
 552         struct blkif_request *ring_req;
 553         unsigned long id;
 554
 555         /* Fill out a communications ring structure. */
 556         id = blkif_ring_get_request(rinfo, req, &ring_req);
 557
 558         ring_req->operation = BLKIF_OP_DISCARD;
 559         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 560         ring_req->u.discard.id = id;
 561         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 562         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 563                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 564         else
 565                 ring_req->u.discard.flag = 0;
 566
 567         /* Keep a private copy so we can reissue requests when recovering. */
 568         rinfo->shadow[id].req = *ring_req;
 569
 570         return 0;
 571 }
 572
 573 struct setup_rw_req {
 574         unsigned int grant_idx;
 575         struct blkif_request_segment *segments;
 576         struct blkfront_ring_info *rinfo;
 577         struct blkif_request *ring_req;
 578         grant_ref_t gref_head;
 579         unsigned int id;
 580         /* Only used when persistent grant is used and it's a read request */
 581         bool need_copy;
 582         unsigned int bvec_off;
 583         char *bvec_data;
 584
 585         bool require_extra_req;
 586         struct blkif_request *extra_ring_req;
 587 };
 588
 589 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 590                                      unsigned int len, void *data)
 591 {
 592         struct setup_rw_req *setup = data;
 593         int n, ref;
 594         struct grant *gnt_list_entry;
 595         unsigned int fsect, lsect;
 596         /* Convenient aliases */
 597         unsigned int grant_idx = setup->grant_idx;
 598         struct blkif_request *ring_req = setup->ring_req;
 599         struct blkfront_ring_info *rinfo = setup->rinfo;
 600         /*
 601          * We always use the shadow of the first request to store the list
 602          * of grant associated to the block I/O request. This made the
 603          * completion more easy to handle even if the block I/O request is
 604          * split.
 605          */
 606         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 607
 608         if (unlikely(setup->require_extra_req &&
 609                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 610                 /*
 611                  * We are using the second request, setup grant_idx
 612                  * to be the index of the segment array.
 613                  */
 614                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 615                 ring_req = setup->extra_ring_req;
 616         }
 617
 618         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 619             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 620                 if (setup->segments)
 621                         kunmap_atomic(setup->segments);
 622
 623                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 624                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 625                 shadow->indirect_grants[n] = gnt_list_entry;
 626                 setup->segments = kmap_atomic(gnt_list_entry->page);
 627                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 628         }
 629
 630         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 631         ref = gnt_list_entry->gref;
 632         /*
 633          * All the grants are stored in the shadow of the first
 634          * request. Therefore we have to use the global index.
 635          */
 636         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 637
 638         if (setup->need_copy) {
 639                 void *shared_data;
 640
 641                 shared_data = kmap_atomic(gnt_list_entry->page);
 642                 /*
 643                  * this does not wipe data stored outside the
 644                  * range sg->offset..sg->offset+sg->length.
 645                  * Therefore, blkback *could* see data from
 646                  * previous requests. This is OK as long as
 647                  * persistent grants are shared with just one
 648                  * domain. It may need refactoring if this
 649                  * changes
 650                  */
 651                 memcpy(shared_data + offset,
 652                        setup->bvec_data + setup->bvec_off,
 653                        len);
 654
 655                 kunmap_atomic(shared_data);
 656                 setup->bvec_off += len;
 657         }
 658
 659         fsect = offset >> 9;
 660         lsect = fsect + (len >> 9) - 1;
 661         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 662                 ring_req->u.rw.seg[grant_idx] =
 663                         (struct blkif_request_segment) {
 664                                 .gref       = ref,
 665                                 .first_sect = fsect,
 666                                 .last_sect  = lsect };
 667         } else {
 668                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 669                         (struct blkif_request_segment) {
 670                                 .gref       = ref,
 671                                 .first_sect = fsect,
 672                                 .last_sect  = lsect };
 673         }
 674
 675         (setup->grant_idx)++;
 676 }
 677
 678 static void blkif_setup_extra_req(struct blkif_request *first,
 679                                   struct blkif_request *second)
 680 {
 681         uint16_t nr_segments = first->u.rw.nr_segments;
 682
 683         /*
 684          * The second request is only present when the first request uses
 685          * all its segments. It's always the continuity of the first one.
 686          */
 687         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 688
 689         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 690         second->u.rw.sector_number = first->u.rw.sector_number +
 691                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 692
 693         second->u.rw.handle = first->u.rw.handle;
 694         second->operation = first->operation;
 695 }
 696
 697 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 698 {
 699         struct blkfront_info *info = rinfo->dev_info;
 700         struct blkif_request *ring_req, *extra_ring_req = NULL;
 701         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 702         bool require_extra_req = false;
 703         int i;
 704         struct setup_rw_req setup = {
 705                 .grant_idx = 0,
 706                 .segments = NULL,
 707                 .rinfo = rinfo,
 708                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 709         };
 710
 711         /*
 712          * Used to store if we are able to queue the request by just using
 713          * existing persistent grants, or if we have to get new grants,
 714          * as there are not sufficiently many free.
 715          */
 716         struct scatterlist *sg;
 717         int num_sg, max_grefs, num_grant;
 718
 719         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 720         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 721                 /*
 722                  * If we are using indirect segments we need to account
 723                  * for the indirect grefs used in the request.
 724                  */
 725                 max_grefs += INDIRECT_GREFS(max_grefs);
 726
 727         /*
 728          * We have to reserve 'max_grefs' grants because persistent
 729          * grants are shared by all rings.
 730          */
 731         if (max_grefs > 0)
 732                 if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
 733                         gnttab_request_free_callback(
 734                                 &rinfo->callback,
 735                                 blkif_restart_queue_callback,
 736                                 rinfo,
 737                                 max_grefs);
 738                         return 1;
 739                 }
 740
 741         /* Fill out a communications ring structure. */
 742         id = blkif_ring_get_request(rinfo, req, &ring_req);
 743
 744         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 745         num_grant = 0;
 746         /* Calculate the number of grant used */
 747         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 748                num_grant += gnttab_count_grant(sg->offset, sg->length);
 749
 750         require_extra_req = info->max_indirect_segments == 0 &&
 751                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 752         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 753
 754         rinfo->shadow[id].num_sg = num_sg;
 755         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 756             likely(!require_extra_req)) {
 757                 /*
 758                  * The indirect operation can only be a BLKIF_OP_READ or
 759                  * BLKIF_OP_WRITE
 760                  */
 761                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 762                 ring_req->operation = BLKIF_OP_INDIRECT;
 763                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 764                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 765                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 766                 ring_req->u.indirect.handle = info->handle;
 767                 ring_req->u.indirect.nr_segments = num_grant;
 768         } else {
 769                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 770                 ring_req->u.rw.handle = info->handle;
 771                 ring_req->operation = rq_data_dir(req) ?
 772                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 773                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 774                         /*
 775                          * Ideally we can do an unordered flush-to-disk.
 776                          * In case the backend onlysupports barriers, use that.
 777                          * A barrier request a superset of FUA, so we can
 778                          * implement it the same way.  (It's also a FLUSH+FUA,
 779                          * since it is guaranteed ordered WRT previous writes.)
 780                          */
 781                         if (info->feature_flush && info->feature_fua)
 782                                 ring_req->operation =
 783                                         BLKIF_OP_WRITE_BARRIER;
 784                         else if (info->feature_flush)
 785                                 ring_req->operation =
 786                                         BLKIF_OP_FLUSH_DISKCACHE;
 787                         else
 788                                 ring_req->operation = 0;
 789                 }
 790                 ring_req->u.rw.nr_segments = num_grant;
 791                 if (unlikely(require_extra_req)) {
 792                         extra_id = blkif_ring_get_request(rinfo, req,
 793                                                           &extra_ring_req);
 794                         /*
 795                          * Only the first request contains the scatter-gather
 796                          * list.
 797                          */
 798                         rinfo->shadow[extra_id].num_sg = 0;
 799
 800                         blkif_setup_extra_req(ring_req, extra_ring_req);
 801
 802                         /* Link the 2 requests together */
 803                         rinfo->shadow[extra_id].associated_id = id;
 804                         rinfo->shadow[id].associated_id = extra_id;
 805                 }
 806         }
 807
 808         setup.ring_req = ring_req;
 809         setup.id = id;
 810
 811         setup.require_extra_req = require_extra_req;
 812         if (unlikely(require_extra_req))
 813                 setup.extra_ring_req = extra_ring_req;
 814
 815         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 816                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 817
 818                 if (setup.need_copy) {
 819                         setup.bvec_off = sg->offset;
 820                         setup.bvec_data = kmap_atomic(sg_page(sg));
 821                 }
 822
 823                 gnttab_foreach_grant_in_range(sg_page(sg),
 824                                               sg->offset,
 825                                               sg->length,
 826                                               blkif_setup_rw_req_grant,
 827                                               &setup);
 828
 829                 if (setup.need_copy)
 830                         kunmap_atomic(setup.bvec_data);
 831         }
 832         if (setup.segments)
 833                 kunmap_atomic(setup.segments);
 834
 835         /* Keep a private copy so we can reissue requests when recovering. */
 836         rinfo->shadow[id].req = *ring_req;
 837         if (unlikely(require_extra_req))
 838                 rinfo->shadow[extra_id].req = *extra_ring_req;
 839
 840         if (max_grefs > 0)
 841                 gnttab_free_grant_references(setup.gref_head);
 842
 843         return 0;
 844 }
 845
 846 /*
 847  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 848  * and writes are handled as expected.
 849  *
 850  * @req: a request struct
 851  */
 852 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 853 {
 854         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 855                 return 1;
 856
 857         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 858                      req_op(req) == REQ_OP_SECURE_ERASE))
 859                 return blkif_queue_discard_req(req, rinfo);
 860         else
 861                 return blkif_queue_rw_req(req, rinfo);
 862 }
 863
 864 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 865 {
 866         int notify;
 867
 868         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 869
 870         if (notify)
 871                 notify_remote_via_irq(rinfo->irq);
 872 }
 873
 874 static inline bool blkif_request_flush_invalid(struct request *req,
 875                                                struct blkfront_info *info)
 876 {
 877         return (blk_rq_is_passthrough(req) ||
 878                 ((req_op(req) == REQ_OP_FLUSH) &&
 879                  !info->feature_flush) ||
 880                 ((req->cmd_flags & REQ_FUA) &&
 881                  !info->feature_fua));
 882 }
 883
 884 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 885                           const struct blk_mq_queue_data *qd)
 886 {
 887         unsigned long flags;
 888         int qid = hctx->queue_num;
 889         struct blkfront_info *info = hctx->queue->queuedata;
 890         struct blkfront_ring_info *rinfo = NULL;
 891
 892         BUG_ON(info->nr_rings <= qid);
 893         rinfo = &info->rinfo[qid];
 894         blk_mq_start_request(qd->rq);
 895         spin_lock_irqsave(&rinfo->ring_lock, flags);
 896         if (RING_FULL(&rinfo->ring))
 897                 goto out_busy;
 898
 899         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 900                 goto out_err;
 901
 902         if (blkif_queue_request(qd->rq, rinfo))
 903                 goto out_busy;
 904
 905         flush_requests(rinfo);
 906         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 907         return BLK_MQ_RQ_QUEUE_OK;
 908
 909 out_err:
 910         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 911         return BLK_MQ_RQ_QUEUE_ERROR;
 912
 913 out_busy:
 914         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 915         blk_mq_stop_hw_queue(hctx);
 916         return BLK_MQ_RQ_QUEUE_BUSY;
 917 }
 918
 919 static void blkif_complete_rq(struct request *rq)
 920 {
 921         blk_mq_end_request(rq, blkif_req(rq)->error);
 922 }
 923
 924 static const struct blk_mq_ops blkfront_mq_ops = {
 925         .queue_rq = blkif_queue_rq,
 926         .complete = blkif_complete_rq,
 927 };
 928
 929 static void blkif_set_queue_limits(struct blkfront_info *info)
 930 {
 931         struct request_queue *rq = info->rq;
 932         struct gendisk *gd = info->gd;
 933         unsigned int segments = info->max_indirect_segments ? :
 934                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 935
 936         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 937
 938         if (info->feature_discard) {
 939                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 940                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 941                 rq->limits.discard_granularity = info->discard_granularity;
 942                 rq->limits.discard_alignment = info->discard_alignment;
 943                 if (info->feature_secdiscard)
 944                         queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
 945         }
 946
 947         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 948         blk_queue_logical_block_size(rq, info->sector_size);
 949         blk_queue_physical_block_size(rq, info->physical_sector_size);
 950         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 951
 952         /* Each segment in a request is up to an aligned page in size. */
 953         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 954         blk_queue_max_segment_size(rq, PAGE_SIZE);
 955
 956         /* Ensure a merged request will fit in a single I/O ring slot. */
 957         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 958
 959         /* Make sure buffer addresses are sector-aligned. */
 960         blk_queue_dma_alignment(rq, 511);
 961
 962         /* Make sure we don't use bounce buffers. */
 963         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 964 }
 965
 966 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 967                                 unsigned int physical_sector_size)
 968 {
 969         struct request_queue *rq;
 970         struct blkfront_info *info = gd->private_data;
 971
 972         memset(&info->tag_set, 0, sizeof(info->tag_set));
 973         info->tag_set.ops = &blkfront_mq_ops;
 974         info->tag_set.nr_hw_queues = info->nr_rings;
 975         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
 976                 /*
 977                  * When indirect descriptior is not supported, the I/O request
 978                  * will be split between multiple request in the ring.
 979                  * To avoid problems when sending the request, divide by
 980                  * 2 the depth of the queue.
 981                  */
 982                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
 983         } else
 984                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
 985         info->tag_set.numa_node = NUMA_NO_NODE;
 986         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
 987         info->tag_set.cmd_size = sizeof(struct blkif_req);
 988         info->tag_set.driver_data = info;
 989
 990         if (blk_mq_alloc_tag_set(&info->tag_set))
 991                 return -EINVAL;
 992         rq = blk_mq_init_queue(&info->tag_set);
 993         if (IS_ERR(rq)) {
 994                 blk_mq_free_tag_set(&info->tag_set);
 995                 return PTR_ERR(rq);
 996         }
 997
 998         rq->queuedata = info;
 999         info->rq = gd->queue = rq;
1000         info->gd = gd;
1001         info->sector_size = sector_size;
1002         info->physical_sector_size = physical_sector_size;
1003         blkif_set_queue_limits(info);
1004
1005         return 0;
1006 }
1007
1008 static const char *flush_info(struct blkfront_info *info)
1009 {
1010         if (info->feature_flush && info->feature_fua)
1011                 return "barrier: enabled;";
1012         else if (info->feature_flush)
1013                 return "flush diskcache: enabled;";
1014         else
1015                 return "barrier or flush: disabled;";
1016 }
1017
1018 static void xlvbd_flush(struct blkfront_info *info)
1019 {
1020         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1021                               info->feature_fua ? true : false);
1022         pr_info("blkfront: %s: %s %s %s %s %s\n",
1023                 info->gd->disk_name, flush_info(info),
1024                 "persistent grants:", info->feature_persistent ?
1025                 "enabled;" : "disabled;", "indirect descriptors:",
1026                 info->max_indirect_segments ? "enabled;" : "disabled;");
1027 }
1028
1029 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1030 {
1031         int major;
1032         major = BLKIF_MAJOR(vdevice);
1033         *minor = BLKIF_MINOR(vdevice);
1034         switch (major) {
1035                 case XEN_IDE0_MAJOR:
1036                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1037                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1038                                 EMULATED_HD_DISK_MINOR_OFFSET;
1039                         break;
1040                 case XEN_IDE1_MAJOR:
1041                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1042                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1043                                 EMULATED_HD_DISK_MINOR_OFFSET;
1044                         break;
1045                 case XEN_SCSI_DISK0_MAJOR:
1046                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1047                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1048                         break;
1049                 case XEN_SCSI_DISK1_MAJOR:
1050                 case XEN_SCSI_DISK2_MAJOR:
1051                 case XEN_SCSI_DISK3_MAJOR:
1052                 case XEN_SCSI_DISK4_MAJOR:
1053                 case XEN_SCSI_DISK5_MAJOR:
1054                 case XEN_SCSI_DISK6_MAJOR:
1055                 case XEN_SCSI_DISK7_MAJOR:
1056                         *offset = (*minor / PARTS_PER_DISK) +
1057                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1058                                 EMULATED_SD_DISK_NAME_OFFSET;
1059                         *minor = *minor +
1060                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1061                                 EMULATED_SD_DISK_MINOR_OFFSET;
1062                         break;
1063                 case XEN_SCSI_DISK8_MAJOR:
1064                 case XEN_SCSI_DISK9_MAJOR:
1065                 case XEN_SCSI_DISK10_MAJOR:
1066                 case XEN_SCSI_DISK11_MAJOR:
1067                 case XEN_SCSI_DISK12_MAJOR:
1068                 case XEN_SCSI_DISK13_MAJOR:
1069                 case XEN_SCSI_DISK14_MAJOR:
1070                 case XEN_SCSI_DISK15_MAJOR:
1071                         *offset = (*minor / PARTS_PER_DISK) +
1072                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1073                                 EMULATED_SD_DISK_NAME_OFFSET;
1074                         *minor = *minor +
1075                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1076                                 EMULATED_SD_DISK_MINOR_OFFSET;
1077                         break;
1078                 case XENVBD_MAJOR:
1079                         *offset = *minor / PARTS_PER_DISK;
1080                         break;
1081                 default:
1082                         printk(KERN_WARNING "blkfront: your disk configuration is "
1083                                         "incorrect, please use an xvd device instead\n");
1084                         return -ENODEV;
1085         }
1086         return 0;
1087 }
1088
1089 static char *encode_disk_name(char *ptr, unsigned int n)
1090 {
1091         if (n >= 26)
1092                 ptr = encode_disk_name(ptr, n / 26 - 1);
1093         *ptr = 'a' + n % 26;
1094         return ptr + 1;
1095 }
1096
1097 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1098                                struct blkfront_info *info,
1099                                u16 vdisk_info, u16 sector_size,
1100                                unsigned int physical_sector_size)
1101 {
1102         struct gendisk *gd;
1103         int nr_minors = 1;
1104         int err;
1105         unsigned int offset;
1106         int minor;
1107         int nr_parts;
1108         char *ptr;
1109
1110         BUG_ON(info->gd != NULL);
1111         BUG_ON(info->rq != NULL);
1112
1113         if ((info->vdevice>>EXT_SHIFT) > 1) {
1114                 /* this is above the extended range; something is wrong */
1115                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1116                 return -ENODEV;
1117         }
1118
1119         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1120                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1121                 if (err)
1122                         return err;
1123                 nr_parts = PARTS_PER_DISK;
1124         } else {
1125                 minor = BLKIF_MINOR_EXT(info->vdevice);
1126                 nr_parts = PARTS_PER_EXT_DISK;
1127                 offset = minor / nr_parts;
1128                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1129                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1130                                         "emulated IDE disks,\n\t choose an xvd device name"
1131                                         "from xvde on\n", info->vdevice);
1132         }
1133         if (minor >> MINORBITS) {
1134                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1135                         info->vdevice, minor);
1136                 return -ENODEV;
1137         }
1138
1139         if ((minor % nr_parts) == 0)
1140                 nr_minors = nr_parts;
1141
1142         err = xlbd_reserve_minors(minor, nr_minors);
1143         if (err)
1144                 goto out;
1145         err = -ENODEV;
1146
1147         gd = alloc_disk(nr_minors);
1148         if (gd == NULL)
1149                 goto release;
1150
1151         strcpy(gd->disk_name, DEV_NAME);
1152         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1153         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1154         if (nr_minors > 1)
1155                 *ptr = 0;
1156         else
1157                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1158                          "%d", minor & (nr_parts - 1));
1159
1160         gd->major = XENVBD_MAJOR;
1161         gd->first_minor = minor;
1162         gd->fops = &xlvbd_block_fops;
1163         gd->private_data = info;
1164         set_capacity(gd, capacity);
1165
1166         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1167                 del_gendisk(gd);
1168                 goto release;
1169         }
1170
1171         xlvbd_flush(info);
1172
1173         if (vdisk_info & VDISK_READONLY)
1174                 set_disk_ro(gd, 1);
1175
1176         if (vdisk_info & VDISK_REMOVABLE)
1177                 gd->flags |= GENHD_FL_REMOVABLE;
1178
1179         if (vdisk_info & VDISK_CDROM)
1180                 gd->flags |= GENHD_FL_CD;
1181
1182         return 0;
1183
1184  release:
1185         xlbd_release_minors(minor, nr_minors);
1186  out:
1187         return err;
1188 }
1189
1190 static void xlvbd_release_gendisk(struct blkfront_info *info)
1191 {
1192         unsigned int minor, nr_minors, i;
1193
1194         if (info->rq == NULL)
1195                 return;
1196
1197         /* No more blkif_request(). */
1198         blk_mq_stop_hw_queues(info->rq);
1199
1200         for (i = 0; i < info->nr_rings; i++) {
1201                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1202
1203                 /* No more gnttab callback work. */
1204                 gnttab_cancel_free_callback(&rinfo->callback);
1205
1206                 /* Flush gnttab callback work. Must be done with no locks held. */
1207                 flush_work(&rinfo->work);
1208         }
1209
1210         del_gendisk(info->gd);
1211
1212         minor = info->gd->first_minor;
1213         nr_minors = info->gd->minors;
1214         xlbd_release_minors(minor, nr_minors);
1215
1216         blk_cleanup_queue(info->rq);
1217         blk_mq_free_tag_set(&info->tag_set);
1218         info->rq = NULL;
1219
1220         put_disk(info->gd);
1221         info->gd = NULL;
1222 }
1223
1224 /* Already hold rinfo->ring_lock. */
1225 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1226 {
1227         if (!RING_FULL(&rinfo->ring))
1228                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1229 }
1230
1231 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1232 {
1233         unsigned long flags;
1234
1235         spin_lock_irqsave(&rinfo->ring_lock, flags);
1236         kick_pending_request_queues_locked(rinfo);
1237         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1238 }
1239
1240 static void blkif_restart_queue(struct work_struct *work)
1241 {
1242         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1243
1244         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1245                 kick_pending_request_queues(rinfo);
1246 }
1247
1248 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1249 {
1250         struct grant *persistent_gnt, *n;
1251         struct blkfront_info *info = rinfo->dev_info;
1252         int i, j, segs;
1253
1254         /*
1255          * Remove indirect pages, this only happens when using indirect
1256          * descriptors but not persistent grants
1257          */
1258         if (!list_empty(&rinfo->indirect_pages)) {
1259                 struct page *indirect_page, *n;
1260
1261                 BUG_ON(info->feature_persistent);
1262                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1263                         list_del(&indirect_page->lru);
1264                         __free_page(indirect_page);
1265                 }
1266         }
1267
1268         /* Remove all persistent grants. */
1269         if (!list_empty(&rinfo->grants)) {
1270                 list_for_each_entry_safe(persistent_gnt, n,
1271                                          &rinfo->grants, node) {
1272                         list_del(&persistent_gnt->node);
1273                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1274                                 gnttab_end_foreign_access(persistent_gnt->gref,
1275                                                           0, 0UL);
1276                                 rinfo->persistent_gnts_c--;
1277                         }
1278                         if (info->feature_persistent)
1279                                 __free_page(persistent_gnt->page);
1280                         kfree(persistent_gnt);
1281                 }
1282         }
1283         BUG_ON(rinfo->persistent_gnts_c != 0);
1284
1285         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1286                 /*
1287                  * Clear persistent grants present in requests already
1288                  * on the shared ring
1289                  */
1290                 if (!rinfo->shadow[i].request)
1291                         goto free_shadow;
1292
1293                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1294                        rinfo->shadow[i].req.u.indirect.nr_segments :
1295                        rinfo->shadow[i].req.u.rw.nr_segments;
1296                 for (j = 0; j < segs; j++) {
1297                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1298                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1299                         if (info->feature_persistent)
1300                                 __free_page(persistent_gnt->page);
1301                         kfree(persistent_gnt);
1302                 }
1303
1304                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1305                         /*
1306                          * If this is not an indirect operation don't try to
1307                          * free indirect segments
1308                          */
1309                         goto free_shadow;
1310
1311                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1312                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1313                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1314                         __free_page(persistent_gnt->page);
1315                         kfree(persistent_gnt);
1316                 }
1317
1318 free_shadow:
1319                 kfree(rinfo->shadow[i].grants_used);
1320                 rinfo->shadow[i].grants_used = NULL;
1321                 kfree(rinfo->shadow[i].indirect_grants);
1322                 rinfo->shadow[i].indirect_grants = NULL;
1323                 kfree(rinfo->shadow[i].sg);
1324                 rinfo->shadow[i].sg = NULL;
1325         }
1326
1327         /* No more gnttab callback work. */
1328         gnttab_cancel_free_callback(&rinfo->callback);
1329
1330         /* Flush gnttab callback work. Must be done with no locks held. */
1331         flush_work(&rinfo->work);
1332
1333         /* Free resources associated with old device channel. */
1334         for (i = 0; i < info->nr_ring_pages; i++) {
1335                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1336                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1337                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1338                 }
1339         }
1340         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1341         rinfo->ring.sring = NULL;
1342
1343         if (rinfo->irq)
1344                 unbind_from_irqhandler(rinfo->irq, rinfo);
1345         rinfo->evtchn = rinfo->irq = 0;
1346 }
1347
1348 static void blkif_free(struct blkfront_info *info, int suspend)
1349 {
1350         unsigned int i;
1351
1352         /* Prevent new requests being issued until we fix things up. */
1353         info->connected = suspend ?
1354                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1355         /* No more blkif_request(). */
1356         if (info->rq)
1357                 blk_mq_stop_hw_queues(info->rq);
1358
1359         for (i = 0; i < info->nr_rings; i++)
1360                 blkif_free_ring(&info->rinfo[i]);
1361
1362         kfree(info->rinfo);
1363         info->rinfo = NULL;
1364         info->nr_rings = 0;
1365 }
1366
1367 struct copy_from_grant {
1368         const struct blk_shadow *s;
1369         unsigned int grant_idx;
1370         unsigned int bvec_offset;
1371         char *bvec_data;
1372 };
1373
1374 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1375                                   unsigned int len, void *data)
1376 {
1377         struct copy_from_grant *info = data;
1378         char *shared_data;
1379         /* Convenient aliases */
1380         const struct blk_shadow *s = info->s;
1381
1382         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1383
1384         memcpy(info->bvec_data + info->bvec_offset,
1385                shared_data + offset, len);
1386
1387         info->bvec_offset += len;
1388         info->grant_idx++;
1389
1390         kunmap_atomic(shared_data);
1391 }
1392
1393 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1394 {
1395         switch (rsp)
1396         {
1397         case BLKIF_RSP_OKAY:
1398                 return REQ_DONE;
1399         case BLKIF_RSP_EOPNOTSUPP:
1400                 return REQ_EOPNOTSUPP;
1401         case BLKIF_RSP_ERROR:
1402                 /* Fallthrough. */
1403         default:
1404                 return REQ_ERROR;
1405         }
1406 }
1407
1408 /*
1409  * Get the final status of the block request based on two ring response
1410  */
1411 static int blkif_get_final_status(enum blk_req_status s1,
1412                                   enum blk_req_status s2)
1413 {
1414         BUG_ON(s1 == REQ_WAITING);
1415         BUG_ON(s2 == REQ_WAITING);
1416
1417         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1418                 return BLKIF_RSP_ERROR;
1419         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1420                 return BLKIF_RSP_EOPNOTSUPP;
1421         return BLKIF_RSP_OKAY;
1422 }
1423
1424 static bool blkif_completion(unsigned long *id,
1425                              struct blkfront_ring_info *rinfo,
1426                              struct blkif_response *bret)
1427 {
1428         int i = 0;
1429         struct scatterlist *sg;
1430         int num_sg, num_grant;
1431         struct blkfront_info *info = rinfo->dev_info;
1432         struct blk_shadow *s = &rinfo->shadow[*id];
1433         struct copy_from_grant data = {
1434                 .grant_idx = 0,
1435         };
1436
1437         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1438                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1439
1440         /* The I/O request may be split in two. */
1441         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1442                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1443
1444                 /* Keep the status of the current response in shadow. */
1445                 s->status = blkif_rsp_to_req_status(bret->status);
1446
1447                 /* Wait the second response if not yet here. */
1448                 if (s2->status == REQ_WAITING)
1449                         return 0;
1450
1451                 bret->status = blkif_get_final_status(s->status,
1452                                                       s2->status);
1453
1454                 /*
1455                  * All the grants is stored in the first shadow in order
1456                  * to make the completion code simpler.
1457                  */
1458                 num_grant += s2->req.u.rw.nr_segments;
1459
1460                 /*
1461                  * The two responses may not come in order. Only the
1462                  * first request will store the scatter-gather list.
1463                  */
1464                 if (s2->num_sg != 0) {
1465                         /* Update "id" with the ID of the first response. */
1466                         *id = s->associated_id;
1467                         s = s2;
1468                 }
1469
1470                 /*
1471                  * We don't need anymore the second request, so recycling
1472                  * it now.
1473                  */
1474                 if (add_id_to_freelist(rinfo, s->associated_id))
1475                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1476                              info->gd->disk_name, s->associated_id);
1477         }
1478
1479         data.s = s;
1480         num_sg = s->num_sg;
1481
1482         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1483                 for_each_sg(s->sg, sg, num_sg, i) {
1484                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1485
1486                         data.bvec_offset = sg->offset;
1487                         data.bvec_data = kmap_atomic(sg_page(sg));
1488
1489                         gnttab_foreach_grant_in_range(sg_page(sg),
1490                                                       sg->offset,
1491                                                       sg->length,
1492                                                       blkif_copy_from_grant,
1493                                                       &data);
1494
1495                         kunmap_atomic(data.bvec_data);
1496                 }
1497         }
1498         /* Add the persistent grant into the list of free grants */
1499         for (i = 0; i < num_grant; i++) {
1500                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1501                         /*
1502                          * If the grant is still mapped by the backend (the
1503                          * backend has chosen to make this grant persistent)
1504                          * we add it at the head of the list, so it will be
1505                          * reused first.
1506                          */
1507                         if (!info->feature_persistent)
1508                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1509                                                      s->grants_used[i]->gref);
1510                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1511                         rinfo->persistent_gnts_c++;
1512                 } else {
1513                         /*
1514                          * If the grant is not mapped by the backend we end the
1515                          * foreign access and add it to the tail of the list,
1516                          * so it will not be picked again unless we run out of
1517                          * persistent grants.
1518                          */
1519                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1520                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1521                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1522                 }
1523         }
1524         if (s->req.operation == BLKIF_OP_INDIRECT) {
1525                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1526                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1527                                 if (!info->feature_persistent)
1528                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1529                                                              s->indirect_grants[i]->gref);
1530                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1531                                 rinfo->persistent_gnts_c++;
1532                         } else {
1533                                 struct page *indirect_page;
1534
1535                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1536                                 /*
1537                                  * Add the used indirect page back to the list of
1538                                  * available pages for indirect grefs.
1539                                  */
1540                                 if (!info->feature_persistent) {
1541                                         indirect_page = s->indirect_grants[i]->page;
1542                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1543                                 }
1544                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1545                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1546                         }
1547                 }
1548         }
1549
1550         return 1;
1551 }
1552
1553 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1554 {
1555         struct request *req;
1556         struct blkif_response *bret;
1557         RING_IDX i, rp;
1558         unsigned long flags;
1559         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1560         struct blkfront_info *info = rinfo->dev_info;
1561
1562         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1563                 return IRQ_HANDLED;
1564
1565         spin_lock_irqsave(&rinfo->ring_lock, flags);
1566  again:
1567         rp = rinfo->ring.sring->rsp_prod;
1568         rmb(); /* Ensure we see queued responses up to 'rp'. */
1569
1570         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1571                 unsigned long id;
1572
1573                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1574                 id   = bret->id;
1575                 /*
1576                  * The backend has messed up and given us an id that we would
1577                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1578                  * look in get_id_from_freelist.
1579                  */
1580                 if (id >= BLK_RING_SIZE(info)) {
1581                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1582                              info->gd->disk_name, op_name(bret->operation), id);
1583                         /* We can't safely get the 'struct request' as
1584                          * the id is busted. */
1585                         continue;
1586                 }
1587                 req  = rinfo->shadow[id].request;
1588
1589                 if (bret->operation != BLKIF_OP_DISCARD) {
1590                         /*
1591                          * We may need to wait for an extra response if the
1592                          * I/O request is split in 2
1593                          */
1594                         if (!blkif_completion(&id, rinfo, bret))
1595                                 continue;
1596                 }
1597
1598                 if (add_id_to_freelist(rinfo, id)) {
1599                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1600                              info->gd->disk_name, op_name(bret->operation), id);
1601                         continue;
1602                 }
1603
1604                 if (bret->status == BLKIF_RSP_OKAY)
1605                         blkif_req(req)->error = BLK_STS_OK;
1606                 else
1607                         blkif_req(req)->error = BLK_STS_IOERR;
1608
1609                 switch (bret->operation) {
1610                 case BLKIF_OP_DISCARD:
1611                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1612                                 struct request_queue *rq = info->rq;
1613                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1614                                            info->gd->disk_name, op_name(bret->operation));
1615                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1616                                 info->feature_discard = 0;
1617                                 info->feature_secdiscard = 0;
1618                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1619                                 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1620                         }
1621                         break;
1622                 case BLKIF_OP_FLUSH_DISKCACHE:
1623                 case BLKIF_OP_WRITE_BARRIER:
1624                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1625                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1626                                        info->gd->disk_name, op_name(bret->operation));
1627                                 blkif_req(req)->error = -EOPNOTSUPP;
1628                         }
1629                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1630                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1631                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1632                                        info->gd->disk_name, op_name(bret->operation));
1633                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1634                         }
1635                         if (unlikely(blkif_req(req)->error)) {
1636                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1637                                         blkif_req(req)->error = BLK_STS_OK;
1638                                 info->feature_fua = 0;
1639                                 info->feature_flush = 0;
1640                                 xlvbd_flush(info);
1641                         }
1642                         /* fall through */
1643                 case BLKIF_OP_READ:
1644                 case BLKIF_OP_WRITE:
1645                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1646                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1647                                         "request: %x\n", bret->status);
1648
1649                         break;
1650                 default:
1651                         BUG();
1652                 }
1653
1654                 blk_mq_complete_request(req);
1655         }
1656
1657         rinfo->ring.rsp_cons = i;
1658
1659         if (i != rinfo->ring.req_prod_pvt) {
1660                 int more_to_do;
1661                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1662                 if (more_to_do)
1663                         goto again;
1664         } else
1665                 rinfo->ring.sring->rsp_event = i + 1;
1666
1667         kick_pending_request_queues_locked(rinfo);
1668
1669         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1670
1671         return IRQ_HANDLED;
1672 }
1673
1674
1675 static int setup_blkring(struct xenbus_device *dev,
1676                          struct blkfront_ring_info *rinfo)
1677 {
1678         struct blkif_sring *sring;
1679         int err, i;
1680         struct blkfront_info *info = rinfo->dev_info;
1681         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1682         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1683
1684         for (i = 0; i < info->nr_ring_pages; i++)
1685                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1686
1687         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1688                                                        get_order(ring_size));
1689         if (!sring) {
1690                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1691                 return -ENOMEM;
1692         }
1693         SHARED_RING_INIT(sring);
1694         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1695
1696         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1697         if (err < 0) {
1698                 free_pages((unsigned long)sring, get_order(ring_size));
1699                 rinfo->ring.sring = NULL;
1700                 goto fail;
1701         }
1702         for (i = 0; i < info->nr_ring_pages; i++)
1703                 rinfo->ring_ref[i] = gref[i];
1704
1705         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1706         if (err)
1707                 goto fail;
1708
1709         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1710                                         "blkif", rinfo);
1711         if (err <= 0) {
1712                 xenbus_dev_fatal(dev, err,
1713                                  "bind_evtchn_to_irqhandler failed");
1714                 goto fail;
1715         }
1716         rinfo->irq = err;
1717
1718         return 0;
1719 fail:
1720         blkif_free(info, 0);
1721         return err;
1722 }
1723
1724 /*
1725  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1726  * ring buffer may have multi pages depending on ->nr_ring_pages.
1727  */
1728 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1729                                 struct blkfront_ring_info *rinfo, const char *dir)
1730 {
1731         int err;
1732         unsigned int i;
1733         const char *message = NULL;
1734         struct blkfront_info *info = rinfo->dev_info;
1735
1736         if (info->nr_ring_pages == 1) {
1737                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1738                 if (err) {
1739                         message = "writing ring-ref";
1740                         goto abort_transaction;
1741                 }
1742         } else {
1743                 for (i = 0; i < info->nr_ring_pages; i++) {
1744                         char ring_ref_name[RINGREF_NAME_LEN];
1745
1746                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1747                         err = xenbus_printf(xbt, dir, ring_ref_name,
1748                                             "%u", rinfo->ring_ref[i]);
1749                         if (err) {
1750                                 message = "writing ring-ref";
1751                                 goto abort_transaction;
1752                         }
1753                 }
1754         }
1755
1756         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1757         if (err) {
1758                 message = "writing event-channel";
1759                 goto abort_transaction;
1760         }
1761
1762         return 0;
1763
1764 abort_transaction:
1765         xenbus_transaction_end(xbt, 1);
1766         if (message)
1767                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1768
1769         return err;
1770 }
1771
1772 /* Common code used when first setting up, and when resuming. */
1773 static int talk_to_blkback(struct xenbus_device *dev,
1774                            struct blkfront_info *info)
1775 {
1776         const char *message = NULL;
1777         struct xenbus_transaction xbt;
1778         int err;
1779         unsigned int i, max_page_order;
1780         unsigned int ring_page_order;
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         for (i = 0; i < info->nr_rings; i++) {
1788                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1789
1790                 /* Create shared ring, alloc event channel. */
1791                 err = setup_blkring(dev, rinfo);
1792                 if (err)
1793                         goto destroy_blkring;
1794         }
1795
1796 again:
1797         err = xenbus_transaction_start(&xbt);
1798         if (err) {
1799                 xenbus_dev_fatal(dev, err, "starting transaction");
1800                 goto destroy_blkring;
1801         }
1802
1803         if (info->nr_ring_pages > 1) {
1804                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1805                                     ring_page_order);
1806                 if (err) {
1807                         message = "writing ring-page-order";
1808                         goto abort_transaction;
1809                 }
1810         }
1811
1812         /* We already got the number of queues/rings in _probe */
1813         if (info->nr_rings == 1) {
1814                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1815                 if (err)
1816                         goto destroy_blkring;
1817         } else {
1818                 char *path;
1819                 size_t pathsize;
1820
1821                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1822                                     info->nr_rings);
1823                 if (err) {
1824                         message = "writing multi-queue-num-queues";
1825                         goto abort_transaction;
1826                 }
1827
1828                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1829                 path = kmalloc(pathsize, GFP_KERNEL);
1830                 if (!path) {
1831                         err = -ENOMEM;
1832                         message = "ENOMEM while writing ring references";
1833                         goto abort_transaction;
1834                 }
1835
1836                 for (i = 0; i < info->nr_rings; i++) {
1837                         memset(path, 0, pathsize);
1838                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1839                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1840                         if (err) {
1841                                 kfree(path);
1842                                 goto destroy_blkring;
1843                         }
1844                 }
1845                 kfree(path);
1846         }
1847         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1848                             XEN_IO_PROTO_ABI_NATIVE);
1849         if (err) {
1850                 message = "writing protocol";
1851                 goto abort_transaction;
1852         }
1853         err = xenbus_printf(xbt, dev->nodename,
1854                             "feature-persistent", "%u", 1);
1855         if (err)
1856                 dev_warn(&dev->dev,
1857                          "writing persistent grants feature to xenbus");
1858
1859         err = xenbus_transaction_end(xbt, 0);
1860         if (err) {
1861                 if (err == -EAGAIN)
1862                         goto again;
1863                 xenbus_dev_fatal(dev, err, "completing transaction");
1864                 goto destroy_blkring;
1865         }
1866
1867         for (i = 0; i < info->nr_rings; i++) {
1868                 unsigned int j;
1869                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1870
1871                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1872                         rinfo->shadow[j].req.u.rw.id = j + 1;
1873                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1874         }
1875         xenbus_switch_state(dev, XenbusStateInitialised);
1876
1877         return 0;
1878
1879  abort_transaction:
1880         xenbus_transaction_end(xbt, 1);
1881         if (message)
1882                 xenbus_dev_fatal(dev, err, "%s", message);
1883  destroy_blkring:
1884         blkif_free(info, 0);
1885
1886         kfree(info);
1887         dev_set_drvdata(&dev->dev, NULL);
1888
1889         return err;
1890 }
1891
1892 static int negotiate_mq(struct blkfront_info *info)
1893 {
1894         unsigned int backend_max_queues;
1895         unsigned int i;
1896
1897         BUG_ON(info->nr_rings);
1898
1899         /* Check if backend supports multiple queues. */
1900         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1901                                                   "multi-queue-max-queues", 1);
1902         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1903         /* We need at least one ring. */
1904         if (!info->nr_rings)
1905                 info->nr_rings = 1;
1906
1907         info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1908         if (!info->rinfo) {
1909                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1910                 return -ENOMEM;
1911         }
1912
1913         for (i = 0; i < info->nr_rings; i++) {
1914                 struct blkfront_ring_info *rinfo;
1915
1916                 rinfo = &info->rinfo[i];
1917                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1918                 INIT_LIST_HEAD(&rinfo->grants);
1919                 rinfo->dev_info = info;
1920                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1921                 spin_lock_init(&rinfo->ring_lock);
1922         }
1923         return 0;
1924 }
1925 /**
1926  * Entry point to this code when a new device is created.  Allocate the basic
1927  * structures and the ring buffer for communication with the backend, and
1928  * inform the backend of the appropriate details for those.  Switch to
1929  * Initialised state.
1930  */
1931 static int blkfront_probe(struct xenbus_device *dev,
1932                           const struct xenbus_device_id *id)
1933 {
1934         int err, vdevice;
1935         struct blkfront_info *info;
1936
1937         /* FIXME: Use dynamic device id if this is not set. */
1938         err = xenbus_scanf(XBT_NIL, dev->nodename,
1939                            "virtual-device", "%i", &vdevice);
1940         if (err != 1) {
1941                 /* go looking in the extended area instead */
1942                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1943                                    "%i", &vdevice);
1944                 if (err != 1) {
1945                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1946                         return err;
1947                 }
1948         }
1949
1950         if (xen_hvm_domain()) {
1951                 char *type;
1952                 int len;
1953                 /* no unplug has been done: do not hook devices != xen vbds */
1954                 if (xen_has_pv_and_legacy_disk_devices()) {
1955                         int major;
1956
1957                         if (!VDEV_IS_EXTENDED(vdevice))
1958                                 major = BLKIF_MAJOR(vdevice);
1959                         else
1960                                 major = XENVBD_MAJOR;
1961
1962                         if (major != XENVBD_MAJOR) {
1963                                 printk(KERN_INFO
1964                                                 "%s: HVM does not support vbd %d as xen block device\n",
1965                                                 __func__, vdevice);
1966                                 return -ENODEV;
1967                         }
1968                 }
1969                 /* do not create a PV cdrom device if we are an HVM guest */
1970                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1971                 if (IS_ERR(type))
1972                         return -ENODEV;
1973                 if (strncmp(type, "cdrom", 5) == 0) {
1974                         kfree(type);
1975                         return -ENODEV;
1976                 }
1977                 kfree(type);
1978         }
1979         info = kzalloc(sizeof(*info), GFP_KERNEL);
1980         if (!info) {
1981                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1982                 return -ENOMEM;
1983         }
1984
1985         info->xbdev = dev;
1986         err = negotiate_mq(info);
1987         if (err) {
1988                 kfree(info);
1989                 return err;
1990         }
1991
1992         mutex_init(&info->mutex);
1993         info->vdevice = vdevice;
1994         info->connected = BLKIF_STATE_DISCONNECTED;
1995
1996         /* Front end dir is a number, which is used as the id. */
1997         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1998         dev_set_drvdata(&dev->dev, info);
1999
2000         return 0;
2001 }
2002
2003 static void split_bio_end(struct bio *bio)
2004 {
2005         struct split_bio *split_bio = bio->bi_private;
2006
2007         if (atomic_dec_and_test(&split_bio->pending)) {
2008                 split_bio->bio->bi_phys_segments = 0;
2009                 split_bio->bio->bi_error = bio->bi_error;
2010                 bio_endio(split_bio->bio);
2011                 kfree(split_bio);
2012         }
2013         bio_put(bio);
2014 }
2015
2016 static int blkif_recover(struct blkfront_info *info)
2017 {
2018         unsigned int i, r_index;
2019         struct request *req, *n;
2020         int rc;
2021         struct bio *bio, *cloned_bio;
2022         unsigned int segs, offset;
2023         int pending, size;
2024         struct split_bio *split_bio;
2025
2026         blkfront_gather_backend_features(info);
2027         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2028         blkif_set_queue_limits(info);
2029         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2030         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2031
2032         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2033                 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2034
2035                 rc = blkfront_setup_indirect(rinfo);
2036                 if (rc)
2037                         return rc;
2038         }
2039         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2040
2041         /* Now safe for us to use the shared ring */
2042         info->connected = BLKIF_STATE_CONNECTED;
2043
2044         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2045                 struct blkfront_ring_info *rinfo;
2046
2047                 rinfo = &info->rinfo[r_index];
2048                 /* Kick any other new requests queued since we resumed */
2049                 kick_pending_request_queues(rinfo);
2050         }
2051
2052         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2053                 /* Requeue pending requests (flush or discard) */
2054                 list_del_init(&req->queuelist);
2055                 BUG_ON(req->nr_phys_segments > segs);
2056                 blk_mq_requeue_request(req, false);
2057         }
2058         blk_mq_start_stopped_hw_queues(info->rq, true);
2059         blk_mq_kick_requeue_list(info->rq);
2060
2061         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2062                 /* Traverse the list of pending bios and re-queue them */
2063                 if (bio_segments(bio) > segs) {
2064                         /*
2065                          * This bio has more segments than what we can
2066                          * handle, we have to split it.
2067                          */
2068                         pending = (bio_segments(bio) + segs - 1) / segs;
2069                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2070                         BUG_ON(split_bio == NULL);
2071                         atomic_set(&split_bio->pending, pending);
2072                         split_bio->bio = bio;
2073                         for (i = 0; i < pending; i++) {
2074                                 offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2075                                 size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2076                                            (unsigned int)bio_sectors(bio) - offset);
2077                                 cloned_bio = bio_clone(bio, GFP_NOIO);
2078                                 BUG_ON(cloned_bio == NULL);
2079                                 bio_trim(cloned_bio, offset, size);
2080                                 cloned_bio->bi_private = split_bio;
2081                                 cloned_bio->bi_end_io = split_bio_end;
2082                                 submit_bio(cloned_bio);
2083                         }
2084                         /*
2085                          * Now we have to wait for all those smaller bios to
2086                          * end, so we can also end the "parent" bio.
2087                          */
2088                         continue;
2089                 }
2090                 /* We don't need to split this bio */
2091                 submit_bio(bio);
2092         }
2093
2094         return 0;
2095 }
2096
2097 /**
2098  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2099  * driver restart.  We tear down our blkif structure and recreate it, but
2100  * leave the device-layer structures intact so that this is transparent to the
2101  * rest of the kernel.
2102  */
2103 static int blkfront_resume(struct xenbus_device *dev)
2104 {
2105         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2106         int err = 0;
2107         unsigned int i, j;
2108
2109         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2110
2111         bio_list_init(&info->bio_list);
2112         INIT_LIST_HEAD(&info->requests);
2113         for (i = 0; i < info->nr_rings; i++) {
2114                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2115                 struct bio_list merge_bio;
2116                 struct blk_shadow *shadow = rinfo->shadow;
2117
2118                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2119                         /* Not in use? */
2120                         if (!shadow[j].request)
2121                                 continue;
2122
2123                         /*
2124                          * Get the bios in the request so we can re-queue them.
2125                          */
2126                         if (req_op(shadow[i].request) == REQ_OP_FLUSH ||
2127                             req_op(shadow[i].request) == REQ_OP_DISCARD ||
2128                             req_op(shadow[i].request) == REQ_OP_SECURE_ERASE ||
2129                             shadow[j].request->cmd_flags & REQ_FUA) {
2130                                 /*
2131                                  * Flush operations don't contain bios, so
2132                                  * we need to requeue the whole request
2133                                  *
2134                                  * XXX: but this doesn't make any sense for a
2135                                  * write with the FUA flag set..
2136                                  */
2137                                 list_add(&shadow[j].request->queuelist, &info->requests);
2138                                 continue;
2139                         }
2140                         merge_bio.head = shadow[j].request->bio;
2141                         merge_bio.tail = shadow[j].request->biotail;
2142                         bio_list_merge(&info->bio_list, &merge_bio);
2143                         shadow[j].request->bio = NULL;
2144                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2145                 }
2146         }
2147
2148         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2149
2150         err = negotiate_mq(info);
2151         if (err)
2152                 return err;
2153
2154         err = talk_to_blkback(dev, info);
2155         if (!err)
2156                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2157
2158         /*
2159          * We have to wait for the backend to switch to
2160          * connected state, since we want to read which
2161          * features it supports.
2162          */
2163
2164         return err;
2165 }
2166
2167 static void blkfront_closing(struct blkfront_info *info)
2168 {
2169         struct xenbus_device *xbdev = info->xbdev;
2170         struct block_device *bdev = NULL;
2171
2172         mutex_lock(&info->mutex);
2173
2174         if (xbdev->state == XenbusStateClosing) {
2175                 mutex_unlock(&info->mutex);
2176                 return;
2177         }
2178
2179         if (info->gd)
2180                 bdev = bdget_disk(info->gd, 0);
2181
2182         mutex_unlock(&info->mutex);
2183
2184         if (!bdev) {
2185                 xenbus_frontend_closed(xbdev);
2186                 return;
2187         }
2188
2189         mutex_lock(&bdev->bd_mutex);
2190
2191         if (bdev->bd_openers) {
2192                 xenbus_dev_error(xbdev, -EBUSY,
2193                                  "Device in use; refusing to close");
2194                 xenbus_switch_state(xbdev, XenbusStateClosing);
2195         } else {
2196                 xlvbd_release_gendisk(info);
2197                 xenbus_frontend_closed(xbdev);
2198         }
2199
2200         mutex_unlock(&bdev->bd_mutex);
2201         bdput(bdev);
2202 }
2203
2204 static void blkfront_setup_discard(struct blkfront_info *info)
2205 {
2206         int err;
2207         unsigned int discard_granularity;
2208         unsigned int discard_alignment;
2209
2210         info->feature_discard = 1;
2211         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2212                 "discard-granularity", "%u", &discard_granularity,
2213                 "discard-alignment", "%u", &discard_alignment,
2214                 NULL);
2215         if (!err) {
2216                 info->discard_granularity = discard_granularity;
2217                 info->discard_alignment = discard_alignment;
2218         }
2219         info->feature_secdiscard =
2220                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2221                                        0);
2222 }
2223
2224 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2225 {
2226         unsigned int psegs, grants;
2227         int err, i;
2228         struct blkfront_info *info = rinfo->dev_info;
2229
2230         if (info->max_indirect_segments == 0) {
2231                 if (!HAS_EXTRA_REQ)
2232                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2233                 else {
2234                         /*
2235                          * When an extra req is required, the maximum
2236                          * grants supported is related to the size of the
2237                          * Linux block segment.
2238                          */
2239                         grants = GRANTS_PER_PSEG;
2240                 }
2241         }
2242         else
2243                 grants = info->max_indirect_segments;
2244         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2245
2246         err = fill_grant_buffer(rinfo,
2247                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2248         if (err)
2249                 goto out_of_memory;
2250
2251         if (!info->feature_persistent && info->max_indirect_segments) {
2252                 /*
2253                  * We are using indirect descriptors but not persistent
2254                  * grants, we need to allocate a set of pages that can be
2255                  * used for mapping indirect grefs
2256                  */
2257                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2258
2259                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2260                 for (i = 0; i < num; i++) {
2261                         struct page *indirect_page = alloc_page(GFP_NOIO);
2262                         if (!indirect_page)
2263                                 goto out_of_memory;
2264                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2265                 }
2266         }
2267
2268         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2269                 rinfo->shadow[i].grants_used = kzalloc(
2270                         sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2271                         GFP_NOIO);
2272                 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2273                 if (info->max_indirect_segments)
2274                         rinfo->shadow[i].indirect_grants = kzalloc(
2275                                 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2276                                 INDIRECT_GREFS(grants),
2277                                 GFP_NOIO);
2278                 if ((rinfo->shadow[i].grants_used == NULL) ||
2279                         (rinfo->shadow[i].sg == NULL) ||
2280                      (info->max_indirect_segments &&
2281                      (rinfo->shadow[i].indirect_grants == NULL)))
2282                         goto out_of_memory;
2283                 sg_init_table(rinfo->shadow[i].sg, psegs);
2284         }
2285
2286
2287         return 0;
2288
2289 out_of_memory:
2290         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2291                 kfree(rinfo->shadow[i].grants_used);
2292                 rinfo->shadow[i].grants_used = NULL;
2293                 kfree(rinfo->shadow[i].sg);
2294                 rinfo->shadow[i].sg = NULL;
2295                 kfree(rinfo->shadow[i].indirect_grants);
2296                 rinfo->shadow[i].indirect_grants = NULL;
2297         }
2298         if (!list_empty(&rinfo->indirect_pages)) {
2299                 struct page *indirect_page, *n;
2300                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2301                         list_del(&indirect_page->lru);
2302                         __free_page(indirect_page);
2303                 }
2304         }
2305         return -ENOMEM;
2306 }
2307
2308 /*
2309  * Gather all backend feature-*
2310  */
2311 static void blkfront_gather_backend_features(struct blkfront_info *info)
2312 {
2313         unsigned int indirect_segments;
2314
2315         info->feature_flush = 0;
2316         info->feature_fua = 0;
2317
2318         /*
2319          * If there's no "feature-barrier" defined, then it means
2320          * we're dealing with a very old backend which writes
2321          * synchronously; nothing to do.
2322          *
2323          * If there are barriers, then we use flush.
2324          */
2325         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2326                 info->feature_flush = 1;
2327                 info->feature_fua = 1;
2328         }
2329
2330         /*
2331          * And if there is "feature-flush-cache" use that above
2332          * barriers.
2333          */
2334         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2335                                  0)) {
2336                 info->feature_flush = 1;
2337                 info->feature_fua = 0;
2338         }
2339
2340         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2341                 blkfront_setup_discard(info);
2342
2343         info->feature_persistent =
2344                 !!xenbus_read_unsigned(info->xbdev->otherend,
2345                                        "feature-persistent", 0);
2346
2347         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2348                                         "feature-max-indirect-segments", 0);
2349         if (indirect_segments > xen_blkif_max_segments)
2350                 indirect_segments = xen_blkif_max_segments;
2351         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2352                 indirect_segments = 0;
2353         info->max_indirect_segments = indirect_segments;
2354 }
2355
2356 /*
2357  * Invoked when the backend is finally 'ready' (and has told produced
2358  * the details about the physical device - #sectors, size, etc).
2359  */
2360 static void blkfront_connect(struct blkfront_info *info)
2361 {
2362         unsigned long long sectors;
2363         unsigned long sector_size;
2364         unsigned int physical_sector_size;
2365         unsigned int binfo;
2366         char *envp[] = { "RESIZE=1", NULL };
2367         int err, i;
2368
2369         switch (info->connected) {
2370         case BLKIF_STATE_CONNECTED:
2371                 /*
2372                  * Potentially, the back-end may be signalling
2373                  * a capacity change; update the capacity.
2374                  */
2375                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2376                                    "sectors", "%Lu", &sectors);
2377                 if (XENBUS_EXIST_ERR(err))
2378                         return;
2379                 printk(KERN_INFO "Setting capacity to %Lu\n",
2380                        sectors);
2381                 set_capacity(info->gd, sectors);
2382                 revalidate_disk(info->gd);
2383                 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2384                                    KOBJ_CHANGE, envp);
2385
2386                 return;
2387         case BLKIF_STATE_SUSPENDED:
2388                 /*
2389                  * If we are recovering from suspension, we need to wait
2390                  * for the backend to announce it's features before
2391                  * reconnecting, at least we need to know if the backend
2392                  * supports indirect descriptors, and how many.
2393                  */
2394                 blkif_recover(info);
2395                 return;
2396
2397         default:
2398                 break;
2399         }
2400
2401         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2402                 __func__, info->xbdev->otherend);
2403
2404         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2405                             "sectors", "%llu", &sectors,
2406                             "info", "%u", &binfo,
2407                             "sector-size", "%lu", &sector_size,
2408                             NULL);
2409         if (err) {
2410                 xenbus_dev_fatal(info->xbdev, err,
2411                                  "reading backend fields at %s",
2412                                  info->xbdev->otherend);
2413                 return;
2414         }
2415
2416         /*
2417          * physcial-sector-size is a newer field, so old backends may not
2418          * provide this. Assume physical sector size to be the same as
2419          * sector_size in that case.
2420          */
2421         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2422                                                     "physical-sector-size",
2423                                                     sector_size);
2424         blkfront_gather_backend_features(info);
2425         for (i = 0; i < info->nr_rings; i++) {
2426                 err = blkfront_setup_indirect(&info->rinfo[i]);
2427                 if (err) {
2428                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2429                                          info->xbdev->otherend);
2430                         blkif_free(info, 0);
2431                         break;
2432                 }
2433         }
2434
2435         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2436                                   physical_sector_size);
2437         if (err) {
2438                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2439                                  info->xbdev->otherend);
2440                 goto fail;
2441         }
2442
2443         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2444
2445         /* Kick pending requests. */
2446         info->connected = BLKIF_STATE_CONNECTED;
2447         for (i = 0; i < info->nr_rings; i++)
2448                 kick_pending_request_queues(&info->rinfo[i]);
2449
2450         device_add_disk(&info->xbdev->dev, info->gd);
2451
2452         info->is_ready = 1;
2453         return;
2454
2455 fail:
2456         blkif_free(info, 0);
2457         return;
2458 }
2459
2460 /**
2461  * Callback received when the backend's state changes.
2462  */
2463 static void blkback_changed(struct xenbus_device *dev,
2464                             enum xenbus_state backend_state)
2465 {
2466         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2467
2468         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2469
2470         switch (backend_state) {
2471         case XenbusStateInitWait:
2472                 if (dev->state != XenbusStateInitialising)
2473                         break;
2474                 if (talk_to_blkback(dev, info))
2475                         break;
2476         case XenbusStateInitialising:
2477         case XenbusStateInitialised:
2478         case XenbusStateReconfiguring:
2479         case XenbusStateReconfigured:
2480         case XenbusStateUnknown:
2481                 break;
2482
2483         case XenbusStateConnected:
2484                 /*
2485                  * talk_to_blkback sets state to XenbusStateInitialised
2486                  * and blkfront_connect sets it to XenbusStateConnected
2487                  * (if connection went OK).
2488                  *
2489                  * If the backend (or toolstack) decides to poke at backend
2490                  * state (and re-trigger the watch by setting the state repeatedly
2491                  * to XenbusStateConnected (4)) we need to deal with this.
2492                  * This is allowed as this is used to communicate to the guest
2493                  * that the size of disk has changed!
2494                  */
2495                 if ((dev->state != XenbusStateInitialised) &&
2496                     (dev->state != XenbusStateConnected)) {
2497                         if (talk_to_blkback(dev, info))
2498                                 break;
2499                 }
2500
2501                 blkfront_connect(info);
2502                 break;
2503
2504         case XenbusStateClosed:
2505                 if (dev->state == XenbusStateClosed)
2506                         break;
2507                 /* Missed the backend's Closing state -- fallthrough */
2508         case XenbusStateClosing:
2509                 if (info)
2510                         blkfront_closing(info);
2511                 break;
2512         }
2513 }
2514
2515 static int blkfront_remove(struct xenbus_device *xbdev)
2516 {
2517         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2518         struct block_device *bdev = NULL;
2519         struct gendisk *disk;
2520
2521         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2522
2523         blkif_free(info, 0);
2524
2525         mutex_lock(&info->mutex);
2526
2527         disk = info->gd;
2528         if (disk)
2529                 bdev = bdget_disk(disk, 0);
2530
2531         info->xbdev = NULL;
2532         mutex_unlock(&info->mutex);
2533
2534         if (!bdev) {
2535                 kfree(info);
2536                 return 0;
2537         }
2538
2539         /*
2540          * The xbdev was removed before we reached the Closed
2541          * state. See if it's safe to remove the disk. If the bdev
2542          * isn't closed yet, we let release take care of it.
2543          */
2544
2545         mutex_lock(&bdev->bd_mutex);
2546         info = disk->private_data;
2547
2548         dev_warn(disk_to_dev(disk),
2549                  "%s was hot-unplugged, %d stale handles\n",
2550                  xbdev->nodename, bdev->bd_openers);
2551
2552         if (info && !bdev->bd_openers) {
2553                 xlvbd_release_gendisk(info);
2554                 disk->private_data = NULL;
2555                 kfree(info);
2556         }
2557
2558         mutex_unlock(&bdev->bd_mutex);
2559         bdput(bdev);
2560
2561         return 0;
2562 }
2563
2564 static int blkfront_is_ready(struct xenbus_device *dev)
2565 {
2566         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2567
2568         return info->is_ready && info->xbdev;
2569 }
2570
2571 static int blkif_open(struct block_device *bdev, fmode_t mode)
2572 {
2573         struct gendisk *disk = bdev->bd_disk;
2574         struct blkfront_info *info;
2575         int err = 0;
2576
2577         mutex_lock(&blkfront_mutex);
2578
2579         info = disk->private_data;
2580         if (!info) {
2581                 /* xbdev gone */
2582                 err = -ERESTARTSYS;
2583                 goto out;
2584         }
2585
2586         mutex_lock(&info->mutex);
2587
2588         if (!info->gd)
2589                 /* xbdev is closed */
2590                 err = -ERESTARTSYS;
2591
2592         mutex_unlock(&info->mutex);
2593
2594 out:
2595         mutex_unlock(&blkfront_mutex);
2596         return err;
2597 }
2598
2599 static void blkif_release(struct gendisk *disk, fmode_t mode)
2600 {
2601         struct blkfront_info *info = disk->private_data;
2602         struct block_device *bdev;
2603         struct xenbus_device *xbdev;
2604
2605         mutex_lock(&blkfront_mutex);
2606
2607         bdev = bdget_disk(disk, 0);
2608
2609         if (!bdev) {
2610                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2611                 goto out_mutex;
2612         }
2613         if (bdev->bd_openers)
2614                 goto out;
2615
2616         /*
2617          * Check if we have been instructed to close. We will have
2618          * deferred this request, because the bdev was still open.
2619          */
2620
2621         mutex_lock(&info->mutex);
2622         xbdev = info->xbdev;
2623
2624         if (xbdev && xbdev->state == XenbusStateClosing) {
2625                 /* pending switch to state closed */
2626                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2627                 xlvbd_release_gendisk(info);
2628                 xenbus_frontend_closed(info->xbdev);
2629         }
2630
2631         mutex_unlock(&info->mutex);
2632
2633         if (!xbdev) {
2634                 /* sudden device removal */
2635                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2636                 xlvbd_release_gendisk(info);
2637                 disk->private_data = NULL;
2638                 kfree(info);
2639         }
2640
2641 out:
2642         bdput(bdev);
2643 out_mutex:
2644         mutex_unlock(&blkfront_mutex);
2645 }
2646
2647 static const struct block_device_operations xlvbd_block_fops =
2648 {
2649         .owner = THIS_MODULE,
2650         .open = blkif_open,
2651         .release = blkif_release,
2652         .getgeo = blkif_getgeo,
2653         .ioctl = blkif_ioctl,
2654 };
2655
2656
2657 static const struct xenbus_device_id blkfront_ids[] = {
2658         { "vbd" },
2659         { "" }
2660 };
2661
2662 static struct xenbus_driver blkfront_driver = {
2663         .ids  = blkfront_ids,
2664         .probe = blkfront_probe,
2665         .remove = blkfront_remove,
2666         .resume = blkfront_resume,
2667         .otherend_changed = blkback_changed,
2668         .is_ready = blkfront_is_ready,
2669 };
2670
2671 static int __init xlblk_init(void)
2672 {
2673         int ret;
2674         int nr_cpus = num_online_cpus();
2675
2676         if (!xen_domain())
2677                 return -ENODEV;
2678
2679         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2680                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2681
2682         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2683                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2684                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2685                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2686         }
2687
2688         if (xen_blkif_max_queues > nr_cpus) {
2689                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2690                         xen_blkif_max_queues, nr_cpus);
2691                 xen_blkif_max_queues = nr_cpus;
2692         }
2693
2694         if (!xen_has_pv_disk_devices())
2695                 return -ENODEV;
2696
2697         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2698                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2699                        XENVBD_MAJOR, DEV_NAME);
2700                 return -ENODEV;
2701         }
2702
2703         ret = xenbus_register_frontend(&blkfront_driver);
2704         if (ret) {
2705                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2706                 return ret;
2707         }
2708
2709         return 0;
2710 }
2711 module_init(xlblk_init);
2712
2713
2714 static void __exit xlblk_exit(void)
2715 {
2716         xenbus_unregister_driver(&blkfront_driver);
2717         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2718         kfree(minors);
2719 }
2720 module_exit(xlblk_exit);
2721
2722 MODULE_DESCRIPTION("Xen virtual block device frontend");
2723 MODULE_LICENSE("GPL");
2724 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2725 MODULE_ALIAS("xen:vbd");
2726 MODULE_ALIAS("xenblk");