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/major.h>
  46 #include <linux/mutex.h>
  47 #include <linux/scatterlist.h>
  48 #include <linux/bitmap.h>
  49 #include <linux/list.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/sched/mm.h>
  52
  53 #include <xen/xen.h>
  54 #include <xen/xenbus.h>
  55 #include <xen/grant_table.h>
  56 #include <xen/events.h>
  57 #include <xen/page.h>
  58 #include <xen/platform_pci.h>
  59
  60 #include <xen/interface/grant_table.h>
  61 #include <xen/interface/io/blkif.h>
  62 #include <xen/interface/io/protocols.h>
  63
  64 #include <asm/xen/hypervisor.h>
  65
  66 /*
  67  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  68  * When Linux is using a different page size than Xen, it may not be possible
  69  * to put all the data in a single segment.
  70  * This can happen when the backend doesn't support indirect descriptor and
  71  * therefore the maximum amount of data that a request can carry is
  72  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  73  *
  74  * Note that we only support one extra request. So the Linux page size
  75  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  76  * 88KB.
  77  */
  78 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  79
  80 enum blkif_state {
  81         BLKIF_STATE_DISCONNECTED,
  82         BLKIF_STATE_CONNECTED,
  83         BLKIF_STATE_SUSPENDED,
  84         BLKIF_STATE_ERROR,
  85 };
  86
  87 struct grant {
  88         grant_ref_t gref;
  89         struct page *page;
  90         struct list_head node;
  91 };
  92
  93 enum blk_req_status {
  94         REQ_PROCESSING,
  95         REQ_WAITING,
  96         REQ_DONE,
  97         REQ_ERROR,
  98         REQ_EOPNOTSUPP,
  99 };
 100
 101 struct blk_shadow {
 102         struct blkif_request req;
 103         struct request *request;
 104         struct grant **grants_used;
 105         struct grant **indirect_grants;
 106         struct scatterlist *sg;
 107         unsigned int num_sg;
 108         enum blk_req_status status;
 109
 110         #define NO_ASSOCIATED_ID ~0UL
 111         /*
 112          * Id of the sibling if we ever need 2 requests when handling a
 113          * block I/O request
 114          */
 115         unsigned long associated_id;
 116 };
 117
 118 struct blkif_req {
 119         blk_status_t    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 static struct delayed_work blkfront_work;
 130 static LIST_HEAD(info_list);
 131
 132 /*
 133  * Maximum number of segments in indirect requests, the actual value used by
 134  * the frontend driver is the minimum of this value and the value provided
 135  * by the backend driver.
 136  */
 137
 138 static unsigned int xen_blkif_max_segments = 32;
 139 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 140 MODULE_PARM_DESC(max_indirect_segments,
 141                  "Maximum amount of segments in indirect requests (default is 32)");
 142
 143 static unsigned int xen_blkif_max_queues = 4;
 144 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 145 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 146
 147 /*
 148  * Maximum order of pages to be used for the shared ring between front and
 149  * backend, 4KB page granularity is used.
 150  */
 151 static unsigned int xen_blkif_max_ring_order;
 152 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 153 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 154
 155 #define BLK_RING_SIZE(info)     \
 156         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 157
 158 /*
 159  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 160  * characters are enough. Define to 20 to keep consistent with backend.
 161  */
 162 #define RINGREF_NAME_LEN (20)
 163 /*
 164  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 165  */
 166 #define QUEUE_NAME_LEN (17)
 167
 168 /*
 169  *  Per-ring info.
 170  *  Every blkfront device can associate with one or more blkfront_ring_info,
 171  *  depending on how many hardware queues/rings to be used.
 172  */
 173 struct blkfront_ring_info {
 174         /* Lock to protect data in every ring buffer. */
 175         spinlock_t ring_lock;
 176         struct blkif_front_ring ring;
 177         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 178         unsigned int evtchn, irq;
 179         struct work_struct work;
 180         struct gnttab_free_callback callback;
 181         struct list_head indirect_pages;
 182         struct list_head grants;
 183         unsigned int persistent_gnts_c;
 184         unsigned long shadow_free;
 185         struct blkfront_info *dev_info;
 186         struct blk_shadow shadow[];
 187 };
 188
 189 /*
 190  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 191  * hang in private_data off the gendisk structure. We may end up
 192  * putting all kinds of interesting stuff here :-)
 193  */
 194 struct blkfront_info
 195 {
 196         struct mutex mutex;
 197         struct xenbus_device *xbdev;
 198         struct gendisk *gd;
 199         u16 sector_size;
 200         unsigned int physical_sector_size;
 201         int vdevice;
 202         blkif_vdev_t handle;
 203         enum blkif_state connected;
 204         /* Number of pages per ring buffer. */
 205         unsigned int nr_ring_pages;
 206         struct request_queue *rq;
 207         unsigned int feature_flush:1;
 208         unsigned int feature_fua:1;
 209         unsigned int feature_discard:1;
 210         unsigned int feature_secdiscard:1;
 211         unsigned int feature_persistent:1;
 212         unsigned int discard_granularity;
 213         unsigned int discard_alignment;
 214         /* Number of 4KB segments handled */
 215         unsigned int max_indirect_segments;
 216         int is_ready;
 217         struct blk_mq_tag_set tag_set;
 218         struct blkfront_ring_info *rinfo;
 219         unsigned int nr_rings;
 220         unsigned int rinfo_size;
 221         /* Save uncomplete reqs and bios for migration. */
 222         struct list_head requests;
 223         struct bio_list bio_list;
 224         struct list_head info_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 INDIRECT_GREFS(_grants)         \
 261         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 262
 263 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 264 static void blkfront_gather_backend_features(struct blkfront_info *info);
 265 static int negotiate_mq(struct blkfront_info *info);
 266
 267 #define for_each_rinfo(info, ptr, idx)                          \
 268         for ((ptr) = (info)->rinfo, (idx) = 0;                  \
 269              (idx) < (info)->nr_rings;                          \
 270              (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
 271
 272 static inline struct blkfront_ring_info *
 273 get_rinfo(const struct blkfront_info *info, unsigned int i)
 274 {
 275         BUG_ON(i >= info->nr_rings);
 276         return (void *)info->rinfo + i * info->rinfo_size;
 277 }
 278
 279 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 280 {
 281         unsigned long free = rinfo->shadow_free;
 282
 283         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 284         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 285         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 286         return free;
 287 }
 288
 289 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 290                               unsigned long id)
 291 {
 292         if (rinfo->shadow[id].req.u.rw.id != id)
 293                 return -EINVAL;
 294         if (rinfo->shadow[id].request == NULL)
 295                 return -EINVAL;
 296         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 297         rinfo->shadow[id].request = NULL;
 298         rinfo->shadow_free = id;
 299         return 0;
 300 }
 301
 302 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 303 {
 304         struct blkfront_info *info = rinfo->dev_info;
 305         struct page *granted_page;
 306         struct grant *gnt_list_entry, *n;
 307         int i = 0;
 308
 309         while (i < num) {
 310                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 311                 if (!gnt_list_entry)
 312                         goto out_of_memory;
 313
 314                 if (info->feature_persistent) {
 315                         granted_page = alloc_page(GFP_NOIO);
 316                         if (!granted_page) {
 317                                 kfree(gnt_list_entry);
 318                                 goto out_of_memory;
 319                         }
 320                         gnt_list_entry->page = granted_page;
 321                 }
 322
 323                 gnt_list_entry->gref = GRANT_INVALID_REF;
 324                 list_add(&gnt_list_entry->node, &rinfo->grants);
 325                 i++;
 326         }
 327
 328         return 0;
 329
 330 out_of_memory:
 331         list_for_each_entry_safe(gnt_list_entry, n,
 332                                  &rinfo->grants, node) {
 333                 list_del(&gnt_list_entry->node);
 334                 if (info->feature_persistent)
 335                         __free_page(gnt_list_entry->page);
 336                 kfree(gnt_list_entry);
 337                 i--;
 338         }
 339         BUG_ON(i != 0);
 340         return -ENOMEM;
 341 }
 342
 343 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 344 {
 345         struct grant *gnt_list_entry;
 346
 347         BUG_ON(list_empty(&rinfo->grants));
 348         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 349                                           node);
 350         list_del(&gnt_list_entry->node);
 351
 352         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 353                 rinfo->persistent_gnts_c--;
 354
 355         return gnt_list_entry;
 356 }
 357
 358 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 359                                         const struct blkfront_info *info)
 360 {
 361         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 362                                                  info->xbdev->otherend_id,
 363                                                  gnt_list_entry->page,
 364                                                  0);
 365 }
 366
 367 static struct grant *get_grant(grant_ref_t *gref_head,
 368                                unsigned long gfn,
 369                                struct blkfront_ring_info *rinfo)
 370 {
 371         struct grant *gnt_list_entry = get_free_grant(rinfo);
 372         struct blkfront_info *info = rinfo->dev_info;
 373
 374         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 375                 return gnt_list_entry;
 376
 377         /* Assign a gref to this page */
 378         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 379         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 380         if (info->feature_persistent)
 381                 grant_foreign_access(gnt_list_entry, info);
 382         else {
 383                 /* Grant access to the GFN passed by the caller */
 384                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 385                                                 info->xbdev->otherend_id,
 386                                                 gfn, 0);
 387         }
 388
 389         return gnt_list_entry;
 390 }
 391
 392 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 393                                         struct blkfront_ring_info *rinfo)
 394 {
 395         struct grant *gnt_list_entry = get_free_grant(rinfo);
 396         struct blkfront_info *info = rinfo->dev_info;
 397
 398         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 399                 return gnt_list_entry;
 400
 401         /* Assign a gref to this page */
 402         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 403         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 404         if (!info->feature_persistent) {
 405                 struct page *indirect_page;
 406
 407                 /* Fetch a pre-allocated page to use for indirect grefs */
 408                 BUG_ON(list_empty(&rinfo->indirect_pages));
 409                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 410                                                  struct page, lru);
 411                 list_del(&indirect_page->lru);
 412                 gnt_list_entry->page = indirect_page;
 413         }
 414         grant_foreign_access(gnt_list_entry, info);
 415
 416         return gnt_list_entry;
 417 }
 418
 419 static const char *op_name(int op)
 420 {
 421         static const char *const names[] = {
 422                 [BLKIF_OP_READ] = "read",
 423                 [BLKIF_OP_WRITE] = "write",
 424                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 425                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 426                 [BLKIF_OP_DISCARD] = "discard" };
 427
 428         if (op < 0 || op >= ARRAY_SIZE(names))
 429                 return "unknown";
 430
 431         if (!names[op])
 432                 return "reserved";
 433
 434         return names[op];
 435 }
 436 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 437 {
 438         unsigned int end = minor + nr;
 439         int rc;
 440
 441         if (end > nr_minors) {
 442                 unsigned long *bitmap, *old;
 443
 444                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 445                                  GFP_KERNEL);
 446                 if (bitmap == NULL)
 447                         return -ENOMEM;
 448
 449                 spin_lock(&minor_lock);
 450                 if (end > nr_minors) {
 451                         old = minors;
 452                         memcpy(bitmap, minors,
 453                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 454                         minors = bitmap;
 455                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 456                 } else
 457                         old = bitmap;
 458                 spin_unlock(&minor_lock);
 459                 kfree(old);
 460         }
 461
 462         spin_lock(&minor_lock);
 463         if (find_next_bit(minors, end, minor) >= end) {
 464                 bitmap_set(minors, minor, nr);
 465                 rc = 0;
 466         } else
 467                 rc = -EBUSY;
 468         spin_unlock(&minor_lock);
 469
 470         return rc;
 471 }
 472
 473 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 474 {
 475         unsigned int end = minor + nr;
 476
 477         BUG_ON(end > nr_minors);
 478         spin_lock(&minor_lock);
 479         bitmap_clear(minors,  minor, nr);
 480         spin_unlock(&minor_lock);
 481 }
 482
 483 static void blkif_restart_queue_callback(void *arg)
 484 {
 485         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 486         schedule_work(&rinfo->work);
 487 }
 488
 489 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 490 {
 491         /* We don't have real geometry info, but let's at least return
 492            values consistent with the size of the device */
 493         sector_t nsect = get_capacity(bd->bd_disk);
 494         sector_t cylinders = nsect;
 495
 496         hg->heads = 0xff;
 497         hg->sectors = 0x3f;
 498         sector_div(cylinders, hg->heads * hg->sectors);
 499         hg->cylinders = cylinders;
 500         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 501                 hg->cylinders = 0xffff;
 502         return 0;
 503 }
 504
 505 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 506                        unsigned command, unsigned long argument)
 507 {
 508         int i;
 509
 510         switch (command) {
 511         case CDROMMULTISESSION:
 512                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 513                         if (put_user(0, (char __user *)(argument + i)))
 514                                 return -EFAULT;
 515                 return 0;
 516         case CDROM_GET_CAPABILITY:
 517                 if (bdev->bd_disk->flags & GENHD_FL_CD)
 518                         return 0;
 519                 return -EINVAL;
 520         default:
 521                 return -EINVAL;
 522         }
 523 }
 524
 525 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 526                                             struct request *req,
 527                                             struct blkif_request **ring_req)
 528 {
 529         unsigned long id;
 530
 531         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 532         rinfo->ring.req_prod_pvt++;
 533
 534         id = get_id_from_freelist(rinfo);
 535         rinfo->shadow[id].request = req;
 536         rinfo->shadow[id].status = REQ_PROCESSING;
 537         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 538
 539         rinfo->shadow[id].req.u.rw.id = id;
 540
 541         return id;
 542 }
 543
 544 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 545 {
 546         struct blkfront_info *info = rinfo->dev_info;
 547         struct blkif_request *ring_req, *final_ring_req;
 548         unsigned long id;
 549
 550         /* Fill out a communications ring structure. */
 551         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 552         ring_req = &rinfo->shadow[id].req;
 553
 554         ring_req->operation = BLKIF_OP_DISCARD;
 555         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 556         ring_req->u.discard.id = id;
 557         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 558         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 559                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 560         else
 561                 ring_req->u.discard.flag = 0;
 562
 563         /* Copy the request to the ring page. */
 564         *final_ring_req = *ring_req;
 565         rinfo->shadow[id].status = REQ_WAITING;
 566
 567         return 0;
 568 }
 569
 570 struct setup_rw_req {
 571         unsigned int grant_idx;
 572         struct blkif_request_segment *segments;
 573         struct blkfront_ring_info *rinfo;
 574         struct blkif_request *ring_req;
 575         grant_ref_t gref_head;
 576         unsigned int id;
 577         /* Only used when persistent grant is used and it's a read request */
 578         bool need_copy;
 579         unsigned int bvec_off;
 580         char *bvec_data;
 581
 582         bool require_extra_req;
 583         struct blkif_request *extra_ring_req;
 584 };
 585
 586 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 587                                      unsigned int len, void *data)
 588 {
 589         struct setup_rw_req *setup = data;
 590         int n, ref;
 591         struct grant *gnt_list_entry;
 592         unsigned int fsect, lsect;
 593         /* Convenient aliases */
 594         unsigned int grant_idx = setup->grant_idx;
 595         struct blkif_request *ring_req = setup->ring_req;
 596         struct blkfront_ring_info *rinfo = setup->rinfo;
 597         /*
 598          * We always use the shadow of the first request to store the list
 599          * of grant associated to the block I/O request. This made the
 600          * completion more easy to handle even if the block I/O request is
 601          * split.
 602          */
 603         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 604
 605         if (unlikely(setup->require_extra_req &&
 606                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 607                 /*
 608                  * We are using the second request, setup grant_idx
 609                  * to be the index of the segment array.
 610                  */
 611                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 612                 ring_req = setup->extra_ring_req;
 613         }
 614
 615         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 616             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 617                 if (setup->segments)
 618                         kunmap_atomic(setup->segments);
 619
 620                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 621                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 622                 shadow->indirect_grants[n] = gnt_list_entry;
 623                 setup->segments = kmap_atomic(gnt_list_entry->page);
 624                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 625         }
 626
 627         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 628         ref = gnt_list_entry->gref;
 629         /*
 630          * All the grants are stored in the shadow of the first
 631          * request. Therefore we have to use the global index.
 632          */
 633         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 634
 635         if (setup->need_copy) {
 636                 void *shared_data;
 637
 638                 shared_data = kmap_atomic(gnt_list_entry->page);
 639                 /*
 640                  * this does not wipe data stored outside the
 641                  * range sg->offset..sg->offset+sg->length.
 642                  * Therefore, blkback *could* see data from
 643                  * previous requests. This is OK as long as
 644                  * persistent grants are shared with just one
 645                  * domain. It may need refactoring if this
 646                  * changes
 647                  */
 648                 memcpy(shared_data + offset,
 649                        setup->bvec_data + setup->bvec_off,
 650                        len);
 651
 652                 kunmap_atomic(shared_data);
 653                 setup->bvec_off += len;
 654         }
 655
 656         fsect = offset >> 9;
 657         lsect = fsect + (len >> 9) - 1;
 658         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 659                 ring_req->u.rw.seg[grant_idx] =
 660                         (struct blkif_request_segment) {
 661                                 .gref       = ref,
 662                                 .first_sect = fsect,
 663                                 .last_sect  = lsect };
 664         } else {
 665                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 666                         (struct blkif_request_segment) {
 667                                 .gref       = ref,
 668                                 .first_sect = fsect,
 669                                 .last_sect  = lsect };
 670         }
 671
 672         (setup->grant_idx)++;
 673 }
 674
 675 static void blkif_setup_extra_req(struct blkif_request *first,
 676                                   struct blkif_request *second)
 677 {
 678         uint16_t nr_segments = first->u.rw.nr_segments;
 679
 680         /*
 681          * The second request is only present when the first request uses
 682          * all its segments. It's always the continuity of the first one.
 683          */
 684         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 685
 686         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 687         second->u.rw.sector_number = first->u.rw.sector_number +
 688                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 689
 690         second->u.rw.handle = first->u.rw.handle;
 691         second->operation = first->operation;
 692 }
 693
 694 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 695 {
 696         struct blkfront_info *info = rinfo->dev_info;
 697         struct blkif_request *ring_req, *extra_ring_req = NULL;
 698         struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
 699         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 700         bool require_extra_req = false;
 701         int i;
 702         struct setup_rw_req setup = {
 703                 .grant_idx = 0,
 704                 .segments = NULL,
 705                 .rinfo = rinfo,
 706                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 707         };
 708
 709         /*
 710          * Used to store if we are able to queue the request by just using
 711          * existing persistent grants, or if we have to get new grants,
 712          * as there are not sufficiently many free.
 713          */
 714         bool new_persistent_gnts = false;
 715         struct scatterlist *sg;
 716         int num_sg, max_grefs, num_grant;
 717
 718         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 719         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 720                 /*
 721                  * If we are using indirect segments we need to account
 722                  * for the indirect grefs used in the request.
 723                  */
 724                 max_grefs += INDIRECT_GREFS(max_grefs);
 725
 726         /* Check if we have enough persistent grants to allocate a requests */
 727         if (rinfo->persistent_gnts_c < max_grefs) {
 728                 new_persistent_gnts = true;
 729
 730                 if (gnttab_alloc_grant_references(
 731                     max_grefs - rinfo->persistent_gnts_c,
 732                     &setup.gref_head) < 0) {
 733                         gnttab_request_free_callback(
 734                                 &rinfo->callback,
 735                                 blkif_restart_queue_callback,
 736                                 rinfo,
 737                                 max_grefs - rinfo->persistent_gnts_c);
 738                         return 1;
 739                 }
 740         }
 741
 742         /* Fill out a communications ring structure. */
 743         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 744         ring_req = &rinfo->shadow[id].req;
 745
 746         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 747         num_grant = 0;
 748         /* Calculate the number of grant used */
 749         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 750                num_grant += gnttab_count_grant(sg->offset, sg->length);
 751
 752         require_extra_req = info->max_indirect_segments == 0 &&
 753                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 754         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 755
 756         rinfo->shadow[id].num_sg = num_sg;
 757         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 758             likely(!require_extra_req)) {
 759                 /*
 760                  * The indirect operation can only be a BLKIF_OP_READ or
 761                  * BLKIF_OP_WRITE
 762                  */
 763                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 764                 ring_req->operation = BLKIF_OP_INDIRECT;
 765                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 766                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 767                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 768                 ring_req->u.indirect.handle = info->handle;
 769                 ring_req->u.indirect.nr_segments = num_grant;
 770         } else {
 771                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 772                 ring_req->u.rw.handle = info->handle;
 773                 ring_req->operation = rq_data_dir(req) ?
 774                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 775                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 776                         /*
 777                          * Ideally we can do an unordered flush-to-disk.
 778                          * In case the backend onlysupports barriers, use that.
 779                          * A barrier request a superset of FUA, so we can
 780                          * implement it the same way.  (It's also a FLUSH+FUA,
 781                          * since it is guaranteed ordered WRT previous writes.)
 782                          */
 783                         if (info->feature_flush && info->feature_fua)
 784                                 ring_req->operation =
 785                                         BLKIF_OP_WRITE_BARRIER;
 786                         else if (info->feature_flush)
 787                                 ring_req->operation =
 788                                         BLKIF_OP_FLUSH_DISKCACHE;
 789                         else
 790                                 ring_req->operation = 0;
 791                 }
 792                 ring_req->u.rw.nr_segments = num_grant;
 793                 if (unlikely(require_extra_req)) {
 794                         extra_id = blkif_ring_get_request(rinfo, req,
 795                                                           &final_extra_ring_req);
 796                         extra_ring_req = &rinfo->shadow[extra_id].req;
 797
 798                         /*
 799                          * Only the first request contains the scatter-gather
 800                          * list.
 801                          */
 802                         rinfo->shadow[extra_id].num_sg = 0;
 803
 804                         blkif_setup_extra_req(ring_req, extra_ring_req);
 805
 806                         /* Link the 2 requests together */
 807                         rinfo->shadow[extra_id].associated_id = id;
 808                         rinfo->shadow[id].associated_id = extra_id;
 809                 }
 810         }
 811
 812         setup.ring_req = ring_req;
 813         setup.id = id;
 814
 815         setup.require_extra_req = require_extra_req;
 816         if (unlikely(require_extra_req))
 817                 setup.extra_ring_req = extra_ring_req;
 818
 819         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 820                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 821
 822                 if (setup.need_copy) {
 823                         setup.bvec_off = sg->offset;
 824                         setup.bvec_data = kmap_atomic(sg_page(sg));
 825                 }
 826
 827                 gnttab_foreach_grant_in_range(sg_page(sg),
 828                                               sg->offset,
 829                                               sg->length,
 830                                               blkif_setup_rw_req_grant,
 831                                               &setup);
 832
 833                 if (setup.need_copy)
 834                         kunmap_atomic(setup.bvec_data);
 835         }
 836         if (setup.segments)
 837                 kunmap_atomic(setup.segments);
 838
 839         /* Copy request(s) to the ring page. */
 840         *final_ring_req = *ring_req;
 841         rinfo->shadow[id].status = REQ_WAITING;
 842         if (unlikely(require_extra_req)) {
 843                 *final_extra_ring_req = *extra_ring_req;
 844                 rinfo->shadow[extra_id].status = REQ_WAITING;
 845         }
 846
 847         if (new_persistent_gnts)
 848                 gnttab_free_grant_references(setup.gref_head);
 849
 850         return 0;
 851 }
 852
 853 /*
 854  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 855  * and writes are handled as expected.
 856  *
 857  * @req: a request struct
 858  */
 859 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 860 {
 861         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 862                 return 1;
 863
 864         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 865                      req_op(req) == REQ_OP_SECURE_ERASE))
 866                 return blkif_queue_discard_req(req, rinfo);
 867         else
 868                 return blkif_queue_rw_req(req, rinfo);
 869 }
 870
 871 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 872 {
 873         int notify;
 874
 875         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 876
 877         if (notify)
 878                 notify_remote_via_irq(rinfo->irq);
 879 }
 880
 881 static inline bool blkif_request_flush_invalid(struct request *req,
 882                                                struct blkfront_info *info)
 883 {
 884         return (blk_rq_is_passthrough(req) ||
 885                 ((req_op(req) == REQ_OP_FLUSH) &&
 886                  !info->feature_flush) ||
 887                 ((req->cmd_flags & REQ_FUA) &&
 888                  !info->feature_fua));
 889 }
 890
 891 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 892                           const struct blk_mq_queue_data *qd)
 893 {
 894         unsigned long flags;
 895         int qid = hctx->queue_num;
 896         struct blkfront_info *info = hctx->queue->queuedata;
 897         struct blkfront_ring_info *rinfo = NULL;
 898
 899         rinfo = get_rinfo(info, qid);
 900         blk_mq_start_request(qd->rq);
 901         spin_lock_irqsave(&rinfo->ring_lock, flags);
 902         if (RING_FULL(&rinfo->ring))
 903                 goto out_busy;
 904
 905         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 906                 goto out_err;
 907
 908         if (blkif_queue_request(qd->rq, rinfo))
 909                 goto out_busy;
 910
 911         flush_requests(rinfo);
 912         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 913         return BLK_STS_OK;
 914
 915 out_err:
 916         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 917         return BLK_STS_IOERR;
 918
 919 out_busy:
 920         blk_mq_stop_hw_queue(hctx);
 921         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 922         return BLK_STS_DEV_RESOURCE;
 923 }
 924
 925 static void blkif_complete_rq(struct request *rq)
 926 {
 927         blk_mq_end_request(rq, blkif_req(rq)->error);
 928 }
 929
 930 static const struct blk_mq_ops blkfront_mq_ops = {
 931         .queue_rq = blkif_queue_rq,
 932         .complete = blkif_complete_rq,
 933 };
 934
 935 static void blkif_set_queue_limits(struct blkfront_info *info)
 936 {
 937         struct request_queue *rq = info->rq;
 938         struct gendisk *gd = info->gd;
 939         unsigned int segments = info->max_indirect_segments ? :
 940                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 941
 942         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 943
 944         if (info->feature_discard) {
 945                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
 946                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 947                 rq->limits.discard_granularity = info->discard_granularity ?:
 948                                                  info->physical_sector_size;
 949                 rq->limits.discard_alignment = info->discard_alignment;
 950                 if (info->feature_secdiscard)
 951                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
 952         }
 953
 954         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 955         blk_queue_logical_block_size(rq, info->sector_size);
 956         blk_queue_physical_block_size(rq, info->physical_sector_size);
 957         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 958
 959         /* Each segment in a request is up to an aligned page in size. */
 960         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 961         blk_queue_max_segment_size(rq, PAGE_SIZE);
 962
 963         /* Ensure a merged request will fit in a single I/O ring slot. */
 964         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 965
 966         /* Make sure buffer addresses are sector-aligned. */
 967         blk_queue_dma_alignment(rq, 511);
 968 }
 969
 970 static const char *flush_info(struct blkfront_info *info)
 971 {
 972         if (info->feature_flush && info->feature_fua)
 973                 return "barrier: enabled;";
 974         else if (info->feature_flush)
 975                 return "flush diskcache: enabled;";
 976         else
 977                 return "barrier or flush: disabled;";
 978 }
 979
 980 static void xlvbd_flush(struct blkfront_info *info)
 981 {
 982         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
 983                               info->feature_fua ? true : false);
 984         pr_info("blkfront: %s: %s %s %s %s %s\n",
 985                 info->gd->disk_name, flush_info(info),
 986                 "persistent grants:", info->feature_persistent ?
 987                 "enabled;" : "disabled;", "indirect descriptors:",
 988                 info->max_indirect_segments ? "enabled;" : "disabled;");
 989 }
 990
 991 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 992 {
 993         int major;
 994         major = BLKIF_MAJOR(vdevice);
 995         *minor = BLKIF_MINOR(vdevice);
 996         switch (major) {
 997                 case XEN_IDE0_MAJOR:
 998                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 999                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1000                                 EMULATED_HD_DISK_MINOR_OFFSET;
1001                         break;
1002                 case XEN_IDE1_MAJOR:
1003                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1004                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1005                                 EMULATED_HD_DISK_MINOR_OFFSET;
1006                         break;
1007                 case XEN_SCSI_DISK0_MAJOR:
1008                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1009                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1010                         break;
1011                 case XEN_SCSI_DISK1_MAJOR:
1012                 case XEN_SCSI_DISK2_MAJOR:
1013                 case XEN_SCSI_DISK3_MAJOR:
1014                 case XEN_SCSI_DISK4_MAJOR:
1015                 case XEN_SCSI_DISK5_MAJOR:
1016                 case XEN_SCSI_DISK6_MAJOR:
1017                 case XEN_SCSI_DISK7_MAJOR:
1018                         *offset = (*minor / PARTS_PER_DISK) +
1019                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1020                                 EMULATED_SD_DISK_NAME_OFFSET;
1021                         *minor = *minor +
1022                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1023                                 EMULATED_SD_DISK_MINOR_OFFSET;
1024                         break;
1025                 case XEN_SCSI_DISK8_MAJOR:
1026                 case XEN_SCSI_DISK9_MAJOR:
1027                 case XEN_SCSI_DISK10_MAJOR:
1028                 case XEN_SCSI_DISK11_MAJOR:
1029                 case XEN_SCSI_DISK12_MAJOR:
1030                 case XEN_SCSI_DISK13_MAJOR:
1031                 case XEN_SCSI_DISK14_MAJOR:
1032                 case XEN_SCSI_DISK15_MAJOR:
1033                         *offset = (*minor / PARTS_PER_DISK) +
1034                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1035                                 EMULATED_SD_DISK_NAME_OFFSET;
1036                         *minor = *minor +
1037                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1038                                 EMULATED_SD_DISK_MINOR_OFFSET;
1039                         break;
1040                 case XENVBD_MAJOR:
1041                         *offset = *minor / PARTS_PER_DISK;
1042                         break;
1043                 default:
1044                         printk(KERN_WARNING "blkfront: your disk configuration is "
1045                                         "incorrect, please use an xvd device instead\n");
1046                         return -ENODEV;
1047         }
1048         return 0;
1049 }
1050
1051 static char *encode_disk_name(char *ptr, unsigned int n)
1052 {
1053         if (n >= 26)
1054                 ptr = encode_disk_name(ptr, n / 26 - 1);
1055         *ptr = 'a' + n % 26;
1056         return ptr + 1;
1057 }
1058
1059 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1060                                struct blkfront_info *info,
1061                                u16 vdisk_info, u16 sector_size,
1062                                unsigned int physical_sector_size)
1063 {
1064         struct gendisk *gd;
1065         int nr_minors = 1;
1066         int err;
1067         unsigned int offset;
1068         int minor;
1069         int nr_parts;
1070         char *ptr;
1071
1072         BUG_ON(info->gd != NULL);
1073         BUG_ON(info->rq != NULL);
1074
1075         if ((info->vdevice>>EXT_SHIFT) > 1) {
1076                 /* this is above the extended range; something is wrong */
1077                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1078                 return -ENODEV;
1079         }
1080
1081         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1082                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1083                 if (err)
1084                         return err;
1085                 nr_parts = PARTS_PER_DISK;
1086         } else {
1087                 minor = BLKIF_MINOR_EXT(info->vdevice);
1088                 nr_parts = PARTS_PER_EXT_DISK;
1089                 offset = minor / nr_parts;
1090                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1091                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1092                                         "emulated IDE disks,\n\t choose an xvd device name"
1093                                         "from xvde on\n", info->vdevice);
1094         }
1095         if (minor >> MINORBITS) {
1096                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1097                         info->vdevice, minor);
1098                 return -ENODEV;
1099         }
1100
1101         if ((minor % nr_parts) == 0)
1102                 nr_minors = nr_parts;
1103
1104         err = xlbd_reserve_minors(minor, nr_minors);
1105         if (err)
1106                 return err;
1107
1108         memset(&info->tag_set, 0, sizeof(info->tag_set));
1109         info->tag_set.ops = &blkfront_mq_ops;
1110         info->tag_set.nr_hw_queues = info->nr_rings;
1111         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1112                 /*
1113                  * When indirect descriptior is not supported, the I/O request
1114                  * will be split between multiple request in the ring.
1115                  * To avoid problems when sending the request, divide by
1116                  * 2 the depth of the queue.
1117                  */
1118                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
1119         } else
1120                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1121         info->tag_set.numa_node = NUMA_NO_NODE;
1122         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1123         info->tag_set.cmd_size = sizeof(struct blkif_req);
1124         info->tag_set.driver_data = info;
1125
1126         err = blk_mq_alloc_tag_set(&info->tag_set);
1127         if (err)
1128                 goto out_release_minors;
1129
1130         gd = blk_mq_alloc_disk(&info->tag_set, info);
1131         if (IS_ERR(gd)) {
1132                 err = PTR_ERR(gd);
1133                 goto out_free_tag_set;
1134         }
1135
1136         strcpy(gd->disk_name, DEV_NAME);
1137         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1138         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1139         if (nr_minors > 1)
1140                 *ptr = 0;
1141         else
1142                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1143                          "%d", minor & (nr_parts - 1));
1144
1145         gd->major = XENVBD_MAJOR;
1146         gd->first_minor = minor;
1147         gd->minors = nr_minors;
1148         gd->fops = &xlvbd_block_fops;
1149         gd->private_data = info;
1150         set_capacity(gd, capacity);
1151
1152         info->rq = gd->queue;
1153         info->gd = gd;
1154         info->sector_size = sector_size;
1155         info->physical_sector_size = physical_sector_size;
1156         blkif_set_queue_limits(info);
1157
1158         xlvbd_flush(info);
1159
1160         if (vdisk_info & VDISK_READONLY)
1161                 set_disk_ro(gd, 1);
1162
1163         if (vdisk_info & VDISK_REMOVABLE)
1164                 gd->flags |= GENHD_FL_REMOVABLE;
1165
1166         if (vdisk_info & VDISK_CDROM)
1167                 gd->flags |= GENHD_FL_CD;
1168
1169         return 0;
1170
1171 out_free_tag_set:
1172         blk_mq_free_tag_set(&info->tag_set);
1173 out_release_minors:
1174         xlbd_release_minors(minor, nr_minors);
1175         return err;
1176 }
1177
1178 /* Already hold rinfo->ring_lock. */
1179 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1180 {
1181         if (!RING_FULL(&rinfo->ring))
1182                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1183 }
1184
1185 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1186 {
1187         unsigned long flags;
1188
1189         spin_lock_irqsave(&rinfo->ring_lock, flags);
1190         kick_pending_request_queues_locked(rinfo);
1191         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1192 }
1193
1194 static void blkif_restart_queue(struct work_struct *work)
1195 {
1196         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1197
1198         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1199                 kick_pending_request_queues(rinfo);
1200 }
1201
1202 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1203 {
1204         struct grant *persistent_gnt, *n;
1205         struct blkfront_info *info = rinfo->dev_info;
1206         int i, j, segs;
1207
1208         /*
1209          * Remove indirect pages, this only happens when using indirect
1210          * descriptors but not persistent grants
1211          */
1212         if (!list_empty(&rinfo->indirect_pages)) {
1213                 struct page *indirect_page, *n;
1214
1215                 BUG_ON(info->feature_persistent);
1216                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1217                         list_del(&indirect_page->lru);
1218                         __free_page(indirect_page);
1219                 }
1220         }
1221
1222         /* Remove all persistent grants. */
1223         if (!list_empty(&rinfo->grants)) {
1224                 list_for_each_entry_safe(persistent_gnt, n,
1225                                          &rinfo->grants, node) {
1226                         list_del(&persistent_gnt->node);
1227                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1228                                 gnttab_end_foreign_access(persistent_gnt->gref,
1229                                                           0, 0UL);
1230                                 rinfo->persistent_gnts_c--;
1231                         }
1232                         if (info->feature_persistent)
1233                                 __free_page(persistent_gnt->page);
1234                         kfree(persistent_gnt);
1235                 }
1236         }
1237         BUG_ON(rinfo->persistent_gnts_c != 0);
1238
1239         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1240                 /*
1241                  * Clear persistent grants present in requests already
1242                  * on the shared ring
1243                  */
1244                 if (!rinfo->shadow[i].request)
1245                         goto free_shadow;
1246
1247                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1248                        rinfo->shadow[i].req.u.indirect.nr_segments :
1249                        rinfo->shadow[i].req.u.rw.nr_segments;
1250                 for (j = 0; j < segs; j++) {
1251                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1252                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1253                         if (info->feature_persistent)
1254                                 __free_page(persistent_gnt->page);
1255                         kfree(persistent_gnt);
1256                 }
1257
1258                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1259                         /*
1260                          * If this is not an indirect operation don't try to
1261                          * free indirect segments
1262                          */
1263                         goto free_shadow;
1264
1265                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1266                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1267                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1268                         __free_page(persistent_gnt->page);
1269                         kfree(persistent_gnt);
1270                 }
1271
1272 free_shadow:
1273                 kvfree(rinfo->shadow[i].grants_used);
1274                 rinfo->shadow[i].grants_used = NULL;
1275                 kvfree(rinfo->shadow[i].indirect_grants);
1276                 rinfo->shadow[i].indirect_grants = NULL;
1277                 kvfree(rinfo->shadow[i].sg);
1278                 rinfo->shadow[i].sg = NULL;
1279         }
1280
1281         /* No more gnttab callback work. */
1282         gnttab_cancel_free_callback(&rinfo->callback);
1283
1284         /* Flush gnttab callback work. Must be done with no locks held. */
1285         flush_work(&rinfo->work);
1286
1287         /* Free resources associated with old device channel. */
1288         for (i = 0; i < info->nr_ring_pages; i++) {
1289                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1290                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1291                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1292                 }
1293         }
1294         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1295         rinfo->ring.sring = NULL;
1296
1297         if (rinfo->irq)
1298                 unbind_from_irqhandler(rinfo->irq, rinfo);
1299         rinfo->evtchn = rinfo->irq = 0;
1300 }
1301
1302 static void blkif_free(struct blkfront_info *info, int suspend)
1303 {
1304         unsigned int i;
1305         struct blkfront_ring_info *rinfo;
1306
1307         /* Prevent new requests being issued until we fix things up. */
1308         info->connected = suspend ?
1309                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1310         /* No more blkif_request(). */
1311         if (info->rq)
1312                 blk_mq_stop_hw_queues(info->rq);
1313
1314         for_each_rinfo(info, rinfo, i)
1315                 blkif_free_ring(rinfo);
1316
1317         kvfree(info->rinfo);
1318         info->rinfo = NULL;
1319         info->nr_rings = 0;
1320 }
1321
1322 struct copy_from_grant {
1323         const struct blk_shadow *s;
1324         unsigned int grant_idx;
1325         unsigned int bvec_offset;
1326         char *bvec_data;
1327 };
1328
1329 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1330                                   unsigned int len, void *data)
1331 {
1332         struct copy_from_grant *info = data;
1333         char *shared_data;
1334         /* Convenient aliases */
1335         const struct blk_shadow *s = info->s;
1336
1337         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1338
1339         memcpy(info->bvec_data + info->bvec_offset,
1340                shared_data + offset, len);
1341
1342         info->bvec_offset += len;
1343         info->grant_idx++;
1344
1345         kunmap_atomic(shared_data);
1346 }
1347
1348 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1349 {
1350         switch (rsp)
1351         {
1352         case BLKIF_RSP_OKAY:
1353                 return REQ_DONE;
1354         case BLKIF_RSP_EOPNOTSUPP:
1355                 return REQ_EOPNOTSUPP;
1356         case BLKIF_RSP_ERROR:
1357         default:
1358                 return REQ_ERROR;
1359         }
1360 }
1361
1362 /*
1363  * Get the final status of the block request based on two ring response
1364  */
1365 static int blkif_get_final_status(enum blk_req_status s1,
1366                                   enum blk_req_status s2)
1367 {
1368         BUG_ON(s1 < REQ_DONE);
1369         BUG_ON(s2 < REQ_DONE);
1370
1371         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1372                 return BLKIF_RSP_ERROR;
1373         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1374                 return BLKIF_RSP_EOPNOTSUPP;
1375         return BLKIF_RSP_OKAY;
1376 }
1377
1378 static bool blkif_completion(unsigned long *id,
1379                              struct blkfront_ring_info *rinfo,
1380                              struct blkif_response *bret)
1381 {
1382         int i = 0;
1383         struct scatterlist *sg;
1384         int num_sg, num_grant;
1385         struct blkfront_info *info = rinfo->dev_info;
1386         struct blk_shadow *s = &rinfo->shadow[*id];
1387         struct copy_from_grant data = {
1388                 .grant_idx = 0,
1389         };
1390
1391         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1392                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1393
1394         /* The I/O request may be split in two. */
1395         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1396                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1397
1398                 /* Keep the status of the current response in shadow. */
1399                 s->status = blkif_rsp_to_req_status(bret->status);
1400
1401                 /* Wait the second response if not yet here. */
1402                 if (s2->status < REQ_DONE)
1403                         return false;
1404
1405                 bret->status = blkif_get_final_status(s->status,
1406                                                       s2->status);
1407
1408                 /*
1409                  * All the grants is stored in the first shadow in order
1410                  * to make the completion code simpler.
1411                  */
1412                 num_grant += s2->req.u.rw.nr_segments;
1413
1414                 /*
1415                  * The two responses may not come in order. Only the
1416                  * first request will store the scatter-gather list.
1417                  */
1418                 if (s2->num_sg != 0) {
1419                         /* Update "id" with the ID of the first response. */
1420                         *id = s->associated_id;
1421                         s = s2;
1422                 }
1423
1424                 /*
1425                  * We don't need anymore the second request, so recycling
1426                  * it now.
1427                  */
1428                 if (add_id_to_freelist(rinfo, s->associated_id))
1429                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1430                              info->gd->disk_name, s->associated_id);
1431         }
1432
1433         data.s = s;
1434         num_sg = s->num_sg;
1435
1436         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1437                 for_each_sg(s->sg, sg, num_sg, i) {
1438                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1439
1440                         data.bvec_offset = sg->offset;
1441                         data.bvec_data = kmap_atomic(sg_page(sg));
1442
1443                         gnttab_foreach_grant_in_range(sg_page(sg),
1444                                                       sg->offset,
1445                                                       sg->length,
1446                                                       blkif_copy_from_grant,
1447                                                       &data);
1448
1449                         kunmap_atomic(data.bvec_data);
1450                 }
1451         }
1452         /* Add the persistent grant into the list of free grants */
1453         for (i = 0; i < num_grant; i++) {
1454                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1455                         /*
1456                          * If the grant is still mapped by the backend (the
1457                          * backend has chosen to make this grant persistent)
1458                          * we add it at the head of the list, so it will be
1459                          * reused first.
1460                          */
1461                         if (!info->feature_persistent)
1462                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1463                                                      s->grants_used[i]->gref);
1464                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1465                         rinfo->persistent_gnts_c++;
1466                 } else {
1467                         /*
1468                          * If the grant is not mapped by the backend we end the
1469                          * foreign access and add it to the tail of the list,
1470                          * so it will not be picked again unless we run out of
1471                          * persistent grants.
1472                          */
1473                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1474                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1475                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1476                 }
1477         }
1478         if (s->req.operation == BLKIF_OP_INDIRECT) {
1479                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1480                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1481                                 if (!info->feature_persistent)
1482                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1483                                                              s->indirect_grants[i]->gref);
1484                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1485                                 rinfo->persistent_gnts_c++;
1486                         } else {
1487                                 struct page *indirect_page;
1488
1489                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1490                                 /*
1491                                  * Add the used indirect page back to the list of
1492                                  * available pages for indirect grefs.
1493                                  */
1494                                 if (!info->feature_persistent) {
1495                                         indirect_page = s->indirect_grants[i]->page;
1496                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1497                                 }
1498                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1499                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1500                         }
1501                 }
1502         }
1503
1504         return true;
1505 }
1506
1507 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1508 {
1509         struct request *req;
1510         struct blkif_response bret;
1511         RING_IDX i, rp;
1512         unsigned long flags;
1513         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1514         struct blkfront_info *info = rinfo->dev_info;
1515
1516         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1517                 return IRQ_HANDLED;
1518
1519         spin_lock_irqsave(&rinfo->ring_lock, flags);
1520  again:
1521         rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1522         virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1523         if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1524                 pr_alert("%s: illegal number of responses %u\n",
1525                          info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1526                 goto err;
1527         }
1528
1529         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1530                 unsigned long id;
1531                 unsigned int op;
1532
1533                 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1534                 id = bret.id;
1535
1536                 /*
1537                  * The backend has messed up and given us an id that we would
1538                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1539                  * look in get_id_from_freelist.
1540                  */
1541                 if (id >= BLK_RING_SIZE(info)) {
1542                         pr_alert("%s: response has incorrect id (%ld)\n",
1543                                  info->gd->disk_name, id);
1544                         goto err;
1545                 }
1546                 if (rinfo->shadow[id].status != REQ_WAITING) {
1547                         pr_alert("%s: response references no pending request\n",
1548                                  info->gd->disk_name);
1549                         goto err;
1550                 }
1551
1552                 rinfo->shadow[id].status = REQ_PROCESSING;
1553                 req  = rinfo->shadow[id].request;
1554
1555                 op = rinfo->shadow[id].req.operation;
1556                 if (op == BLKIF_OP_INDIRECT)
1557                         op = rinfo->shadow[id].req.u.indirect.indirect_op;
1558                 if (bret.operation != op) {
1559                         pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1560                                  info->gd->disk_name, bret.operation, op);
1561                         goto err;
1562                 }
1563
1564                 if (bret.operation != BLKIF_OP_DISCARD) {
1565                         /*
1566                          * We may need to wait for an extra response if the
1567                          * I/O request is split in 2
1568                          */
1569                         if (!blkif_completion(&id, rinfo, &bret))
1570                                 continue;
1571                 }
1572
1573                 if (add_id_to_freelist(rinfo, id)) {
1574                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1575                              info->gd->disk_name, op_name(bret.operation), id);
1576                         continue;
1577                 }
1578
1579                 if (bret.status == BLKIF_RSP_OKAY)
1580                         blkif_req(req)->error = BLK_STS_OK;
1581                 else
1582                         blkif_req(req)->error = BLK_STS_IOERR;
1583
1584                 switch (bret.operation) {
1585                 case BLKIF_OP_DISCARD:
1586                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1587                                 struct request_queue *rq = info->rq;
1588
1589                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1590                                            info->gd->disk_name, op_name(bret.operation));
1591                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1592                                 info->feature_discard = 0;
1593                                 info->feature_secdiscard = 0;
1594                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1595                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1596                         }
1597                         break;
1598                 case BLKIF_OP_FLUSH_DISKCACHE:
1599                 case BLKIF_OP_WRITE_BARRIER:
1600                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1601                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1602                                        info->gd->disk_name, op_name(bret.operation));
1603                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1604                         }
1605                         if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1606                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1607                                 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1608                                        info->gd->disk_name, op_name(bret.operation));
1609                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1610                         }
1611                         if (unlikely(blkif_req(req)->error)) {
1612                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1613                                         blkif_req(req)->error = BLK_STS_OK;
1614                                 info->feature_fua = 0;
1615                                 info->feature_flush = 0;
1616                                 xlvbd_flush(info);
1617                         }
1618                         fallthrough;
1619                 case BLKIF_OP_READ:
1620                 case BLKIF_OP_WRITE:
1621                         if (unlikely(bret.status != BLKIF_RSP_OKAY))
1622                                 dev_dbg_ratelimited(&info->xbdev->dev,
1623                                         "Bad return from blkdev data request: %#x\n",
1624                                         bret.status);
1625
1626                         break;
1627                 default:
1628                         BUG();
1629                 }
1630
1631                 if (likely(!blk_should_fake_timeout(req->q)))
1632                         blk_mq_complete_request(req);
1633         }
1634
1635         rinfo->ring.rsp_cons = i;
1636
1637         if (i != rinfo->ring.req_prod_pvt) {
1638                 int more_to_do;
1639                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1640                 if (more_to_do)
1641                         goto again;
1642         } else
1643                 rinfo->ring.sring->rsp_event = i + 1;
1644
1645         kick_pending_request_queues_locked(rinfo);
1646
1647         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1648
1649         return IRQ_HANDLED;
1650
1651  err:
1652         info->connected = BLKIF_STATE_ERROR;
1653
1654         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1655
1656         pr_alert("%s disabled for further use\n", info->gd->disk_name);
1657         return IRQ_HANDLED;
1658 }
1659
1660
1661 static int setup_blkring(struct xenbus_device *dev,
1662                          struct blkfront_ring_info *rinfo)
1663 {
1664         struct blkif_sring *sring;
1665         int err, i;
1666         struct blkfront_info *info = rinfo->dev_info;
1667         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1668         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1669
1670         for (i = 0; i < info->nr_ring_pages; i++)
1671                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1672
1673         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1674                                                        get_order(ring_size));
1675         if (!sring) {
1676                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1677                 return -ENOMEM;
1678         }
1679         SHARED_RING_INIT(sring);
1680         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1681
1682         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1683         if (err < 0) {
1684                 free_pages((unsigned long)sring, get_order(ring_size));
1685                 rinfo->ring.sring = NULL;
1686                 goto fail;
1687         }
1688         for (i = 0; i < info->nr_ring_pages; i++)
1689                 rinfo->ring_ref[i] = gref[i];
1690
1691         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1692         if (err)
1693                 goto fail;
1694
1695         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1696                                         "blkif", rinfo);
1697         if (err <= 0) {
1698                 xenbus_dev_fatal(dev, err,
1699                                  "bind_evtchn_to_irqhandler failed");
1700                 goto fail;
1701         }
1702         rinfo->irq = err;
1703
1704         return 0;
1705 fail:
1706         blkif_free(info, 0);
1707         return err;
1708 }
1709
1710 /*
1711  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1712  * ring buffer may have multi pages depending on ->nr_ring_pages.
1713  */
1714 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1715                                 struct blkfront_ring_info *rinfo, const char *dir)
1716 {
1717         int err;
1718         unsigned int i;
1719         const char *message = NULL;
1720         struct blkfront_info *info = rinfo->dev_info;
1721
1722         if (info->nr_ring_pages == 1) {
1723                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1724                 if (err) {
1725                         message = "writing ring-ref";
1726                         goto abort_transaction;
1727                 }
1728         } else {
1729                 for (i = 0; i < info->nr_ring_pages; i++) {
1730                         char ring_ref_name[RINGREF_NAME_LEN];
1731
1732                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1733                         err = xenbus_printf(xbt, dir, ring_ref_name,
1734                                             "%u", rinfo->ring_ref[i]);
1735                         if (err) {
1736                                 message = "writing ring-ref";
1737                                 goto abort_transaction;
1738                         }
1739                 }
1740         }
1741
1742         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1743         if (err) {
1744                 message = "writing event-channel";
1745                 goto abort_transaction;
1746         }
1747
1748         return 0;
1749
1750 abort_transaction:
1751         xenbus_transaction_end(xbt, 1);
1752         if (message)
1753                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1754
1755         return err;
1756 }
1757
1758 /* Common code used when first setting up, and when resuming. */
1759 static int talk_to_blkback(struct xenbus_device *dev,
1760                            struct blkfront_info *info)
1761 {
1762         const char *message = NULL;
1763         struct xenbus_transaction xbt;
1764         int err;
1765         unsigned int i, max_page_order;
1766         unsigned int ring_page_order;
1767         struct blkfront_ring_info *rinfo;
1768
1769         if (!info)
1770                 return -ENODEV;
1771
1772         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1773                                               "max-ring-page-order", 0);
1774         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1775         info->nr_ring_pages = 1 << ring_page_order;
1776
1777         err = negotiate_mq(info);
1778         if (err)
1779                 goto destroy_blkring;
1780
1781         for_each_rinfo(info, rinfo, i) {
1782                 /* Create shared ring, alloc event channel. */
1783                 err = setup_blkring(dev, rinfo);
1784                 if (err)
1785                         goto destroy_blkring;
1786         }
1787
1788 again:
1789         err = xenbus_transaction_start(&xbt);
1790         if (err) {
1791                 xenbus_dev_fatal(dev, err, "starting transaction");
1792                 goto destroy_blkring;
1793         }
1794
1795         if (info->nr_ring_pages > 1) {
1796                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1797                                     ring_page_order);
1798                 if (err) {
1799                         message = "writing ring-page-order";
1800                         goto abort_transaction;
1801                 }
1802         }
1803
1804         /* We already got the number of queues/rings in _probe */
1805         if (info->nr_rings == 1) {
1806                 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1807                 if (err)
1808                         goto destroy_blkring;
1809         } else {
1810                 char *path;
1811                 size_t pathsize;
1812
1813                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1814                                     info->nr_rings);
1815                 if (err) {
1816                         message = "writing multi-queue-num-queues";
1817                         goto abort_transaction;
1818                 }
1819
1820                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1821                 path = kmalloc(pathsize, GFP_KERNEL);
1822                 if (!path) {
1823                         err = -ENOMEM;
1824                         message = "ENOMEM while writing ring references";
1825                         goto abort_transaction;
1826                 }
1827
1828                 for_each_rinfo(info, rinfo, i) {
1829                         memset(path, 0, pathsize);
1830                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1831                         err = write_per_ring_nodes(xbt, rinfo, path);
1832                         if (err) {
1833                                 kfree(path);
1834                                 goto destroy_blkring;
1835                         }
1836                 }
1837                 kfree(path);
1838         }
1839         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1840                             XEN_IO_PROTO_ABI_NATIVE);
1841         if (err) {
1842                 message = "writing protocol";
1843                 goto abort_transaction;
1844         }
1845         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1846                         info->feature_persistent);
1847         if (err)
1848                 dev_warn(&dev->dev,
1849                          "writing persistent grants feature to xenbus");
1850
1851         err = xenbus_transaction_end(xbt, 0);
1852         if (err) {
1853                 if (err == -EAGAIN)
1854                         goto again;
1855                 xenbus_dev_fatal(dev, err, "completing transaction");
1856                 goto destroy_blkring;
1857         }
1858
1859         for_each_rinfo(info, rinfo, i) {
1860                 unsigned int j;
1861
1862                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1863                         rinfo->shadow[j].req.u.rw.id = j + 1;
1864                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1865         }
1866         xenbus_switch_state(dev, XenbusStateInitialised);
1867
1868         return 0;
1869
1870  abort_transaction:
1871         xenbus_transaction_end(xbt, 1);
1872         if (message)
1873                 xenbus_dev_fatal(dev, err, "%s", message);
1874  destroy_blkring:
1875         blkif_free(info, 0);
1876         return err;
1877 }
1878
1879 static int negotiate_mq(struct blkfront_info *info)
1880 {
1881         unsigned int backend_max_queues;
1882         unsigned int i;
1883         struct blkfront_ring_info *rinfo;
1884
1885         BUG_ON(info->nr_rings);
1886
1887         /* Check if backend supports multiple queues. */
1888         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1889                                                   "multi-queue-max-queues", 1);
1890         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1891         /* We need at least one ring. */
1892         if (!info->nr_rings)
1893                 info->nr_rings = 1;
1894
1895         info->rinfo_size = struct_size(info->rinfo, shadow,
1896                                        BLK_RING_SIZE(info));
1897         info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1898         if (!info->rinfo) {
1899                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1900                 info->nr_rings = 0;
1901                 return -ENOMEM;
1902         }
1903
1904         for_each_rinfo(info, rinfo, i) {
1905                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1906                 INIT_LIST_HEAD(&rinfo->grants);
1907                 rinfo->dev_info = info;
1908                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1909                 spin_lock_init(&rinfo->ring_lock);
1910         }
1911         return 0;
1912 }
1913
1914 /* Enable the persistent grants feature. */
1915 static bool feature_persistent = true;
1916 module_param(feature_persistent, bool, 0644);
1917 MODULE_PARM_DESC(feature_persistent,
1918                 "Enables the persistent grants feature");
1919
1920 /*
1921  * Entry point to this code when a new device is created.  Allocate the basic
1922  * structures and the ring buffer for communication with the backend, and
1923  * inform the backend of the appropriate details for those.  Switch to
1924  * Initialised state.
1925  */
1926 static int blkfront_probe(struct xenbus_device *dev,
1927                           const struct xenbus_device_id *id)
1928 {
1929         int err, vdevice;
1930         struct blkfront_info *info;
1931
1932         /* FIXME: Use dynamic device id if this is not set. */
1933         err = xenbus_scanf(XBT_NIL, dev->nodename,
1934                            "virtual-device", "%i", &vdevice);
1935         if (err != 1) {
1936                 /* go looking in the extended area instead */
1937                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1938                                    "%i", &vdevice);
1939                 if (err != 1) {
1940                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1941                         return err;
1942                 }
1943         }
1944
1945         if (xen_hvm_domain()) {
1946                 char *type;
1947                 int len;
1948                 /* no unplug has been done: do not hook devices != xen vbds */
1949                 if (xen_has_pv_and_legacy_disk_devices()) {
1950                         int major;
1951
1952                         if (!VDEV_IS_EXTENDED(vdevice))
1953                                 major = BLKIF_MAJOR(vdevice);
1954                         else
1955                                 major = XENVBD_MAJOR;
1956
1957                         if (major != XENVBD_MAJOR) {
1958                                 printk(KERN_INFO
1959                                                 "%s: HVM does not support vbd %d as xen block device\n",
1960                                                 __func__, vdevice);
1961                                 return -ENODEV;
1962                         }
1963                 }
1964                 /* do not create a PV cdrom device if we are an HVM guest */
1965                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1966                 if (IS_ERR(type))
1967                         return -ENODEV;
1968                 if (strncmp(type, "cdrom", 5) == 0) {
1969                         kfree(type);
1970                         return -ENODEV;
1971                 }
1972                 kfree(type);
1973         }
1974         info = kzalloc(sizeof(*info), GFP_KERNEL);
1975         if (!info) {
1976                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1977                 return -ENOMEM;
1978         }
1979
1980         info->xbdev = dev;
1981
1982         mutex_init(&info->mutex);
1983         info->vdevice = vdevice;
1984         info->connected = BLKIF_STATE_DISCONNECTED;
1985
1986         info->feature_persistent = feature_persistent;
1987
1988         /* Front end dir is a number, which is used as the id. */
1989         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1990         dev_set_drvdata(&dev->dev, info);
1991
1992         mutex_lock(&blkfront_mutex);
1993         list_add(&info->info_list, &info_list);
1994         mutex_unlock(&blkfront_mutex);
1995
1996         return 0;
1997 }
1998
1999 static int blkif_recover(struct blkfront_info *info)
2000 {
2001         unsigned int r_index;
2002         struct request *req, *n;
2003         int rc;
2004         struct bio *bio;
2005         unsigned int segs;
2006         struct blkfront_ring_info *rinfo;
2007
2008         blkfront_gather_backend_features(info);
2009         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2010         blkif_set_queue_limits(info);
2011         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2012         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2013
2014         for_each_rinfo(info, rinfo, r_index) {
2015                 rc = blkfront_setup_indirect(rinfo);
2016                 if (rc)
2017                         return rc;
2018         }
2019         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2020
2021         /* Now safe for us to use the shared ring */
2022         info->connected = BLKIF_STATE_CONNECTED;
2023
2024         for_each_rinfo(info, rinfo, r_index) {
2025                 /* Kick any other new requests queued since we resumed */
2026                 kick_pending_request_queues(rinfo);
2027         }
2028
2029         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2030                 /* Requeue pending requests (flush or discard) */
2031                 list_del_init(&req->queuelist);
2032                 BUG_ON(req->nr_phys_segments > segs);
2033                 blk_mq_requeue_request(req, false);
2034         }
2035         blk_mq_start_stopped_hw_queues(info->rq, true);
2036         blk_mq_kick_requeue_list(info->rq);
2037
2038         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2039                 /* Traverse the list of pending bios and re-queue them */
2040                 submit_bio(bio);
2041         }
2042
2043         return 0;
2044 }
2045
2046 /*
2047  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2048  * driver restart.  We tear down our blkif structure and recreate it, but
2049  * leave the device-layer structures intact so that this is transparent to the
2050  * rest of the kernel.
2051  */
2052 static int blkfront_resume(struct xenbus_device *dev)
2053 {
2054         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2055         int err = 0;
2056         unsigned int i, j;
2057         struct blkfront_ring_info *rinfo;
2058
2059         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2060
2061         bio_list_init(&info->bio_list);
2062         INIT_LIST_HEAD(&info->requests);
2063         for_each_rinfo(info, rinfo, i) {
2064                 struct bio_list merge_bio;
2065                 struct blk_shadow *shadow = rinfo->shadow;
2066
2067                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2068                         /* Not in use? */
2069                         if (!shadow[j].request)
2070                                 continue;
2071
2072                         /*
2073                          * Get the bios in the request so we can re-queue them.
2074                          */
2075                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2076                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2077                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2078                             shadow[j].request->cmd_flags & REQ_FUA) {
2079                                 /*
2080                                  * Flush operations don't contain bios, so
2081                                  * we need to requeue the whole request
2082                                  *
2083                                  * XXX: but this doesn't make any sense for a
2084                                  * write with the FUA flag set..
2085                                  */
2086                                 list_add(&shadow[j].request->queuelist, &info->requests);
2087                                 continue;
2088                         }
2089                         merge_bio.head = shadow[j].request->bio;
2090                         merge_bio.tail = shadow[j].request->biotail;
2091                         bio_list_merge(&info->bio_list, &merge_bio);
2092                         shadow[j].request->bio = NULL;
2093                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2094                 }
2095         }
2096
2097         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2098
2099         err = talk_to_blkback(dev, info);
2100         if (!err)
2101                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2102
2103         /*
2104          * We have to wait for the backend to switch to
2105          * connected state, since we want to read which
2106          * features it supports.
2107          */
2108
2109         return err;
2110 }
2111
2112 static void blkfront_closing(struct blkfront_info *info)
2113 {
2114         struct xenbus_device *xbdev = info->xbdev;
2115         struct blkfront_ring_info *rinfo;
2116         unsigned int i;
2117
2118         if (xbdev->state == XenbusStateClosing)
2119                 return;
2120
2121         /* No more blkif_request(). */
2122         blk_mq_stop_hw_queues(info->rq);
2123         blk_set_queue_dying(info->rq);
2124         set_capacity(info->gd, 0);
2125
2126         for_each_rinfo(info, rinfo, i) {
2127                 /* No more gnttab callback work. */
2128                 gnttab_cancel_free_callback(&rinfo->callback);
2129
2130                 /* Flush gnttab callback work. Must be done with no locks held. */
2131                 flush_work(&rinfo->work);
2132         }
2133
2134         xenbus_frontend_closed(xbdev);
2135 }
2136
2137 static void blkfront_setup_discard(struct blkfront_info *info)
2138 {
2139         info->feature_discard = 1;
2140         info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2141                                                          "discard-granularity",
2142                                                          0);
2143         info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2144                                                        "discard-alignment", 0);
2145         info->feature_secdiscard =
2146                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2147                                        0);
2148 }
2149
2150 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2151 {
2152         unsigned int psegs, grants, memflags;
2153         int err, i;
2154         struct blkfront_info *info = rinfo->dev_info;
2155
2156         memflags = memalloc_noio_save();
2157
2158         if (info->max_indirect_segments == 0) {
2159                 if (!HAS_EXTRA_REQ)
2160                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2161                 else {
2162                         /*
2163                          * When an extra req is required, the maximum
2164                          * grants supported is related to the size of the
2165                          * Linux block segment.
2166                          */
2167                         grants = GRANTS_PER_PSEG;
2168                 }
2169         }
2170         else
2171                 grants = info->max_indirect_segments;
2172         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2173
2174         err = fill_grant_buffer(rinfo,
2175                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2176         if (err)
2177                 goto out_of_memory;
2178
2179         if (!info->feature_persistent && info->max_indirect_segments) {
2180                 /*
2181                  * We are using indirect descriptors but not persistent
2182                  * grants, we need to allocate a set of pages that can be
2183                  * used for mapping indirect grefs
2184                  */
2185                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2186
2187                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2188                 for (i = 0; i < num; i++) {
2189                         struct page *indirect_page = alloc_page(GFP_KERNEL);
2190                         if (!indirect_page)
2191                                 goto out_of_memory;
2192                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2193                 }
2194         }
2195
2196         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2197                 rinfo->shadow[i].grants_used =
2198                         kvcalloc(grants,
2199                                  sizeof(rinfo->shadow[i].grants_used[0]),
2200                                  GFP_KERNEL);
2201                 rinfo->shadow[i].sg = kvcalloc(psegs,
2202                                                sizeof(rinfo->shadow[i].sg[0]),
2203                                                GFP_KERNEL);
2204                 if (info->max_indirect_segments)
2205                         rinfo->shadow[i].indirect_grants =
2206                                 kvcalloc(INDIRECT_GREFS(grants),
2207                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2208                                          GFP_KERNEL);
2209                 if ((rinfo->shadow[i].grants_used == NULL) ||
2210                         (rinfo->shadow[i].sg == NULL) ||
2211                      (info->max_indirect_segments &&
2212                      (rinfo->shadow[i].indirect_grants == NULL)))
2213                         goto out_of_memory;
2214                 sg_init_table(rinfo->shadow[i].sg, psegs);
2215         }
2216
2217         memalloc_noio_restore(memflags);
2218
2219         return 0;
2220
2221 out_of_memory:
2222         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2223                 kvfree(rinfo->shadow[i].grants_used);
2224                 rinfo->shadow[i].grants_used = NULL;
2225                 kvfree(rinfo->shadow[i].sg);
2226                 rinfo->shadow[i].sg = NULL;
2227                 kvfree(rinfo->shadow[i].indirect_grants);
2228                 rinfo->shadow[i].indirect_grants = NULL;
2229         }
2230         if (!list_empty(&rinfo->indirect_pages)) {
2231                 struct page *indirect_page, *n;
2232                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2233                         list_del(&indirect_page->lru);
2234                         __free_page(indirect_page);
2235                 }
2236         }
2237
2238         memalloc_noio_restore(memflags);
2239
2240         return -ENOMEM;
2241 }
2242
2243 /*
2244  * Gather all backend feature-*
2245  */
2246 static void blkfront_gather_backend_features(struct blkfront_info *info)
2247 {
2248         unsigned int indirect_segments;
2249
2250         info->feature_flush = 0;
2251         info->feature_fua = 0;
2252
2253         /*
2254          * If there's no "feature-barrier" defined, then it means
2255          * we're dealing with a very old backend which writes
2256          * synchronously; nothing to do.
2257          *
2258          * If there are barriers, then we use flush.
2259          */
2260         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2261                 info->feature_flush = 1;
2262                 info->feature_fua = 1;
2263         }
2264
2265         /*
2266          * And if there is "feature-flush-cache" use that above
2267          * barriers.
2268          */
2269         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2270                                  0)) {
2271                 info->feature_flush = 1;
2272                 info->feature_fua = 0;
2273         }
2274
2275         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2276                 blkfront_setup_discard(info);
2277
2278         if (info->feature_persistent)
2279                 info->feature_persistent =
2280                         !!xenbus_read_unsigned(info->xbdev->otherend,
2281                                                "feature-persistent", 0);
2282
2283         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2284                                         "feature-max-indirect-segments", 0);
2285         if (indirect_segments > xen_blkif_max_segments)
2286                 indirect_segments = xen_blkif_max_segments;
2287         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2288                 indirect_segments = 0;
2289         info->max_indirect_segments = indirect_segments;
2290
2291         if (info->feature_persistent) {
2292                 mutex_lock(&blkfront_mutex);
2293                 schedule_delayed_work(&blkfront_work, HZ * 10);
2294                 mutex_unlock(&blkfront_mutex);
2295         }
2296 }
2297
2298 /*
2299  * Invoked when the backend is finally 'ready' (and has told produced
2300  * the details about the physical device - #sectors, size, etc).
2301  */
2302 static void blkfront_connect(struct blkfront_info *info)
2303 {
2304         unsigned long long sectors;
2305         unsigned long sector_size;
2306         unsigned int physical_sector_size;
2307         unsigned int binfo;
2308         int err, i;
2309         struct blkfront_ring_info *rinfo;
2310
2311         switch (info->connected) {
2312         case BLKIF_STATE_CONNECTED:
2313                 /*
2314                  * Potentially, the back-end may be signalling
2315                  * a capacity change; update the capacity.
2316                  */
2317                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2318                                    "sectors", "%Lu", &sectors);
2319                 if (XENBUS_EXIST_ERR(err))
2320                         return;
2321                 printk(KERN_INFO "Setting capacity to %Lu\n",
2322                        sectors);
2323                 set_capacity_and_notify(info->gd, sectors);
2324
2325                 return;
2326         case BLKIF_STATE_SUSPENDED:
2327                 /*
2328                  * If we are recovering from suspension, we need to wait
2329                  * for the backend to announce it's features before
2330                  * reconnecting, at least we need to know if the backend
2331                  * supports indirect descriptors, and how many.
2332                  */
2333                 blkif_recover(info);
2334                 return;
2335
2336         default:
2337                 break;
2338         }
2339
2340         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2341                 __func__, info->xbdev->otherend);
2342
2343         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2344                             "sectors", "%llu", &sectors,
2345                             "info", "%u", &binfo,
2346                             "sector-size", "%lu", &sector_size,
2347                             NULL);
2348         if (err) {
2349                 xenbus_dev_fatal(info->xbdev, err,
2350                                  "reading backend fields at %s",
2351                                  info->xbdev->otherend);
2352                 return;
2353         }
2354
2355         /*
2356          * physical-sector-size is a newer field, so old backends may not
2357          * provide this. Assume physical sector size to be the same as
2358          * sector_size in that case.
2359          */
2360         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2361                                                     "physical-sector-size",
2362                                                     sector_size);
2363         blkfront_gather_backend_features(info);
2364         for_each_rinfo(info, rinfo, i) {
2365                 err = blkfront_setup_indirect(rinfo);
2366                 if (err) {
2367                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2368                                          info->xbdev->otherend);
2369                         blkif_free(info, 0);
2370                         break;
2371                 }
2372         }
2373
2374         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2375                                   physical_sector_size);
2376         if (err) {
2377                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2378                                  info->xbdev->otherend);
2379                 goto fail;
2380         }
2381
2382         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2383
2384         /* Kick pending requests. */
2385         info->connected = BLKIF_STATE_CONNECTED;
2386         for_each_rinfo(info, rinfo, i)
2387                 kick_pending_request_queues(rinfo);
2388
2389         err = device_add_disk(&info->xbdev->dev, info->gd, NULL);
2390         if (err) {
2391                 blk_cleanup_disk(info->gd);
2392                 blk_mq_free_tag_set(&info->tag_set);
2393                 info->rq = NULL;
2394                 goto fail;
2395         }
2396
2397         info->is_ready = 1;
2398         return;
2399
2400 fail:
2401         blkif_free(info, 0);
2402         return;
2403 }
2404
2405 /*
2406  * Callback received when the backend's state changes.
2407  */
2408 static void blkback_changed(struct xenbus_device *dev,
2409                             enum xenbus_state backend_state)
2410 {
2411         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2412
2413         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2414
2415         switch (backend_state) {
2416         case XenbusStateInitWait:
2417                 if (dev->state != XenbusStateInitialising)
2418                         break;
2419                 if (talk_to_blkback(dev, info))
2420                         break;
2421                 break;
2422         case XenbusStateInitialising:
2423         case XenbusStateInitialised:
2424         case XenbusStateReconfiguring:
2425         case XenbusStateReconfigured:
2426         case XenbusStateUnknown:
2427                 break;
2428
2429         case XenbusStateConnected:
2430                 /*
2431                  * talk_to_blkback sets state to XenbusStateInitialised
2432                  * and blkfront_connect sets it to XenbusStateConnected
2433                  * (if connection went OK).
2434                  *
2435                  * If the backend (or toolstack) decides to poke at backend
2436                  * state (and re-trigger the watch by setting the state repeatedly
2437                  * to XenbusStateConnected (4)) we need to deal with this.
2438                  * This is allowed as this is used to communicate to the guest
2439                  * that the size of disk has changed!
2440                  */
2441                 if ((dev->state != XenbusStateInitialised) &&
2442                     (dev->state != XenbusStateConnected)) {
2443                         if (talk_to_blkback(dev, info))
2444                                 break;
2445                 }
2446
2447                 blkfront_connect(info);
2448                 break;
2449
2450         case XenbusStateClosed:
2451                 if (dev->state == XenbusStateClosed)
2452                         break;
2453                 fallthrough;
2454         case XenbusStateClosing:
2455                 blkfront_closing(info);
2456                 break;
2457         }
2458 }
2459
2460 static int blkfront_remove(struct xenbus_device *xbdev)
2461 {
2462         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2463
2464         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2465
2466         del_gendisk(info->gd);
2467
2468         mutex_lock(&blkfront_mutex);
2469         list_del(&info->info_list);
2470         mutex_unlock(&blkfront_mutex);
2471
2472         blkif_free(info, 0);
2473         xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2474         blk_cleanup_disk(info->gd);
2475         blk_mq_free_tag_set(&info->tag_set);
2476
2477         kfree(info);
2478         return 0;
2479 }
2480
2481 static int blkfront_is_ready(struct xenbus_device *dev)
2482 {
2483         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2484
2485         return info->is_ready && info->xbdev;
2486 }
2487
2488 static const struct block_device_operations xlvbd_block_fops =
2489 {
2490         .owner = THIS_MODULE,
2491         .getgeo = blkif_getgeo,
2492         .ioctl = blkif_ioctl,
2493         .compat_ioctl = blkdev_compat_ptr_ioctl,
2494 };
2495
2496
2497 static const struct xenbus_device_id blkfront_ids[] = {
2498         { "vbd" },
2499         { "" }
2500 };
2501
2502 static struct xenbus_driver blkfront_driver = {
2503         .ids  = blkfront_ids,
2504         .probe = blkfront_probe,
2505         .remove = blkfront_remove,
2506         .resume = blkfront_resume,
2507         .otherend_changed = blkback_changed,
2508         .is_ready = blkfront_is_ready,
2509 };
2510
2511 static void purge_persistent_grants(struct blkfront_info *info)
2512 {
2513         unsigned int i;
2514         unsigned long flags;
2515         struct blkfront_ring_info *rinfo;
2516
2517         for_each_rinfo(info, rinfo, i) {
2518                 struct grant *gnt_list_entry, *tmp;
2519
2520                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2521
2522                 if (rinfo->persistent_gnts_c == 0) {
2523                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2524                         continue;
2525                 }
2526
2527                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2528                                          node) {
2529                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2530                             gnttab_query_foreign_access(gnt_list_entry->gref))
2531                                 continue;
2532
2533                         list_del(&gnt_list_entry->node);
2534                         gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2535                         rinfo->persistent_gnts_c--;
2536                         gnt_list_entry->gref = GRANT_INVALID_REF;
2537                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2538                 }
2539
2540                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2541         }
2542 }
2543
2544 static void blkfront_delay_work(struct work_struct *work)
2545 {
2546         struct blkfront_info *info;
2547         bool need_schedule_work = false;
2548
2549         mutex_lock(&blkfront_mutex);
2550
2551         list_for_each_entry(info, &info_list, info_list) {
2552                 if (info->feature_persistent) {
2553                         need_schedule_work = true;
2554                         mutex_lock(&info->mutex);
2555                         purge_persistent_grants(info);
2556                         mutex_unlock(&info->mutex);
2557                 }
2558         }
2559
2560         if (need_schedule_work)
2561                 schedule_delayed_work(&blkfront_work, HZ * 10);
2562
2563         mutex_unlock(&blkfront_mutex);
2564 }
2565
2566 static int __init xlblk_init(void)
2567 {
2568         int ret;
2569         int nr_cpus = num_online_cpus();
2570
2571         if (!xen_domain())
2572                 return -ENODEV;
2573
2574         if (!xen_has_pv_disk_devices())
2575                 return -ENODEV;
2576
2577         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2578                 pr_warn("xen_blk: can't get major %d with name %s\n",
2579                         XENVBD_MAJOR, DEV_NAME);
2580                 return -ENODEV;
2581         }
2582
2583         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2584                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2585
2586         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2587                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2588                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2589                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2590         }
2591
2592         if (xen_blkif_max_queues > nr_cpus) {
2593                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2594                         xen_blkif_max_queues, nr_cpus);
2595                 xen_blkif_max_queues = nr_cpus;
2596         }
2597
2598         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2599
2600         ret = xenbus_register_frontend(&blkfront_driver);
2601         if (ret) {
2602                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2603                 return ret;
2604         }
2605
2606         return 0;
2607 }
2608 module_init(xlblk_init);
2609
2610
2611 static void __exit xlblk_exit(void)
2612 {
2613         cancel_delayed_work_sync(&blkfront_work);
2614
2615         xenbus_unregister_driver(&blkfront_driver);
2616         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2617         kfree(minors);
2618 }
2619 module_exit(xlblk_exit);
2620
2621 MODULE_DESCRIPTION("Xen virtual block device frontend");
2622 MODULE_LICENSE("GPL");
2623 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2624 MODULE_ALIAS("xen:vbd");
2625 MODULE_ALIAS("xenblk");