drivers/block/xen-blkfront.c

   1 /*
   2  * blkfront.c
   3  *
   4  * XenLinux virtual block device driver.
   5  *
   6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8  * Copyright (c) 2004, Christian Limpach
   9  * Copyright (c) 2004, Andrew Warfield
  10  * Copyright (c) 2005, Christopher Clark
  11  * Copyright (c) 2005, XenSource Ltd
  12  *
  13  * This program is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU General Public License version 2
  15  * as published by the Free Software Foundation; or, when distributed
  16  * separately from the Linux kernel or incorporated into other
  17  * software packages, subject to the following license:
  18  *
  19  * Permission is hereby granted, free of charge, to any person obtaining a copy
  20  * of this source file (the "Software"), to deal in the Software without
  21  * restriction, including without limitation the rights to use, copy, modify,
  22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23  * and to permit persons to whom the Software is furnished to do so, subject to
  24  * the following conditions:
  25  *
  26  * The above copyright notice and this permission notice shall be included in
  27  * all copies or substantial portions of the Software.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35  * IN THE SOFTWARE.
  36  */
  37
  38 #include <linux/interrupt.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/blk-mq.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/cdrom.h>
  43 #include <linux/module.h>
  44 #include <linux/slab.h>
  45 #include <linux/mutex.h>
  46 #include <linux/scatterlist.h>
  47 #include <linux/bitmap.h>
  48 #include <linux/list.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/sched/mm.h>
  51
  52 #include <xen/xen.h>
  53 #include <xen/xenbus.h>
  54 #include <xen/grant_table.h>
  55 #include <xen/events.h>
  56 #include <xen/page.h>
  57 #include <xen/platform_pci.h>
  58
  59 #include <xen/interface/grant_table.h>
  60 #include <xen/interface/io/blkif.h>
  61 #include <xen/interface/io/protocols.h>
  62
  63 #include <asm/xen/hypervisor.h>
  64
  65 /*
  66  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  67  * When Linux is using a different page size than Xen, it may not be possible
  68  * to put all the data in a single segment.
  69  * This can happen when the backend doesn't support indirect descriptor and
  70  * therefore the maximum amount of data that a request can carry is
  71  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  72  *
  73  * Note that we only support one extra request. So the Linux page size
  74  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  75  * 88KB.
  76  */
  77 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  78
  79 enum blkif_state {
  80         BLKIF_STATE_DISCONNECTED,
  81         BLKIF_STATE_CONNECTED,
  82         BLKIF_STATE_SUSPENDED,
  83         BLKIF_STATE_ERROR,
  84 };
  85
  86 struct grant {
  87         grant_ref_t gref;
  88         struct page *page;
  89         struct list_head node;
  90 };
  91
  92 enum blk_req_status {
  93         REQ_PROCESSING,
  94         REQ_WAITING,
  95         REQ_DONE,
  96         REQ_ERROR,
  97         REQ_EOPNOTSUPP,
  98 };
  99
 100 struct blk_shadow {
 101         struct blkif_request req;
 102         struct request *request;
 103         struct grant **grants_used;
 104         struct grant **indirect_grants;
 105         struct scatterlist *sg;
 106         unsigned int num_sg;
 107         enum blk_req_status status;
 108
 109         #define NO_ASSOCIATED_ID ~0UL
 110         /*
 111          * Id of the sibling if we ever need 2 requests when handling a
 112          * block I/O request
 113          */
 114         unsigned long associated_id;
 115 };
 116
 117 struct blkif_req {
 118         blk_status_t    error;
 119 };
 120
 121 static inline struct blkif_req *blkif_req(struct request *rq)
 122 {
 123         return blk_mq_rq_to_pdu(rq);
 124 }
 125
 126 static DEFINE_MUTEX(blkfront_mutex);
 127 static const struct block_device_operations xlvbd_block_fops;
 128 static struct delayed_work blkfront_work;
 129 static LIST_HEAD(info_list);
 130
 131 /*
 132  * Maximum number of segments in indirect requests, the actual value used by
 133  * the frontend driver is the minimum of this value and the value provided
 134  * by the backend driver.
 135  */
 136
 137 static unsigned int xen_blkif_max_segments = 32;
 138 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 139 MODULE_PARM_DESC(max_indirect_segments,
 140                  "Maximum amount of segments in indirect requests (default is 32)");
 141
 142 static unsigned int xen_blkif_max_queues = 4;
 143 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 144 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 145
 146 /*
 147  * Maximum order of pages to be used for the shared ring between front and
 148  * backend, 4KB page granularity is used.
 149  */
 150 static unsigned int xen_blkif_max_ring_order;
 151 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 152 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 153
 154 #define BLK_RING_SIZE(info)     \
 155         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 156
 157 /*
 158  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 159  * characters are enough. Define to 20 to keep consistent with backend.
 160  */
 161 #define RINGREF_NAME_LEN (20)
 162 /*
 163  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 164  */
 165 #define QUEUE_NAME_LEN (17)
 166
 167 /*
 168  *  Per-ring info.
 169  *  Every blkfront device can associate with one or more blkfront_ring_info,
 170  *  depending on how many hardware queues/rings to be used.
 171  */
 172 struct blkfront_ring_info {
 173         /* Lock to protect data in every ring buffer. */
 174         spinlock_t ring_lock;
 175         struct blkif_front_ring ring;
 176         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 177         unsigned int evtchn, irq;
 178         struct work_struct work;
 179         struct gnttab_free_callback callback;
 180         struct list_head indirect_pages;
 181         struct list_head grants;
 182         unsigned int persistent_gnts_c;
 183         unsigned long shadow_free;
 184         struct blkfront_info *dev_info;
 185         struct blk_shadow shadow[];
 186 };
 187
 188 /*
 189  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 190  * hang in private_data off the gendisk structure. We may end up
 191  * putting all kinds of interesting stuff here :-)
 192  */
 193 struct blkfront_info
 194 {
 195         struct mutex mutex;
 196         struct xenbus_device *xbdev;
 197         struct gendisk *gd;
 198         u16 sector_size;
 199         unsigned int physical_sector_size;
 200         int vdevice;
 201         blkif_vdev_t handle;
 202         enum blkif_state connected;
 203         /* Number of pages per ring buffer. */
 204         unsigned int nr_ring_pages;
 205         struct request_queue *rq;
 206         unsigned int feature_flush:1;
 207         unsigned int feature_fua:1;
 208         unsigned int feature_discard:1;
 209         unsigned int feature_secdiscard:1;
 210         unsigned int feature_persistent:1;
 211         unsigned int discard_granularity;
 212         unsigned int discard_alignment;
 213         /* Number of 4KB segments handled */
 214         unsigned int max_indirect_segments;
 215         int is_ready;
 216         struct blk_mq_tag_set tag_set;
 217         struct blkfront_ring_info *rinfo;
 218         unsigned int nr_rings;
 219         unsigned int rinfo_size;
 220         /* Save uncomplete reqs and bios for migration. */
 221         struct list_head requests;
 222         struct bio_list bio_list;
 223         struct list_head info_list;
 224 };
 225
 226 static unsigned int nr_minors;
 227 static unsigned long *minors;
 228 static DEFINE_SPINLOCK(minor_lock);
 229
 230 #define GRANT_INVALID_REF       0
 231
 232 #define PARTS_PER_DISK          16
 233 #define PARTS_PER_EXT_DISK      256
 234
 235 #define BLKIF_MAJOR(dev) ((dev)>>8)
 236 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 237
 238 #define EXT_SHIFT 28
 239 #define EXTENDED (1<<EXT_SHIFT)
 240 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 241 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 242 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 243 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 244 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 245 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 246
 247 #define DEV_NAME        "xvd"   /* name in /dev */
 248
 249 /*
 250  * Grants are always the same size as a Xen page (i.e 4KB).
 251  * A physical segment is always the same size as a Linux page.
 252  * Number of grants per physical segment
 253  */
 254 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 255
 256 #define GRANTS_PER_INDIRECT_FRAME \
 257         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 258
 259 #define INDIRECT_GREFS(_grants)         \
 260         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 261
 262 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 263 static void blkfront_gather_backend_features(struct blkfront_info *info);
 264 static int negotiate_mq(struct blkfront_info *info);
 265
 266 #define for_each_rinfo(info, ptr, idx)                          \
 267         for ((ptr) = (info)->rinfo, (idx) = 0;                  \
 268              (idx) < (info)->nr_rings;                          \
 269              (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
 270
 271 static inline struct blkfront_ring_info *
 272 get_rinfo(const struct blkfront_info *info, unsigned int i)
 273 {
 274         BUG_ON(i >= info->nr_rings);
 275         return (void *)info->rinfo + i * info->rinfo_size;
 276 }
 277
 278 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 279 {
 280         unsigned long free = rinfo->shadow_free;
 281
 282         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 283         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 284         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 285         return free;
 286 }
 287
 288 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 289                               unsigned long id)
 290 {
 291         if (rinfo->shadow[id].req.u.rw.id != id)
 292                 return -EINVAL;
 293         if (rinfo->shadow[id].request == NULL)
 294                 return -EINVAL;
 295         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 296         rinfo->shadow[id].request = NULL;
 297         rinfo->shadow_free = id;
 298         return 0;
 299 }
 300
 301 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 302 {
 303         struct blkfront_info *info = rinfo->dev_info;
 304         struct page *granted_page;
 305         struct grant *gnt_list_entry, *n;
 306         int i = 0;
 307
 308         while (i < num) {
 309                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 310                 if (!gnt_list_entry)
 311                         goto out_of_memory;
 312
 313                 if (info->feature_persistent) {
 314                         granted_page = alloc_page(GFP_NOIO);
 315                         if (!granted_page) {
 316                                 kfree(gnt_list_entry);
 317                                 goto out_of_memory;
 318                         }
 319                         gnt_list_entry->page = granted_page;
 320                 }
 321
 322                 gnt_list_entry->gref = GRANT_INVALID_REF;
 323                 list_add(&gnt_list_entry->node, &rinfo->grants);
 324                 i++;
 325         }
 326
 327         return 0;
 328
 329 out_of_memory:
 330         list_for_each_entry_safe(gnt_list_entry, n,
 331                                  &rinfo->grants, node) {
 332                 list_del(&gnt_list_entry->node);
 333                 if (info->feature_persistent)
 334                         __free_page(gnt_list_entry->page);
 335                 kfree(gnt_list_entry);
 336                 i--;
 337         }
 338         BUG_ON(i != 0);
 339         return -ENOMEM;
 340 }
 341
 342 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 343 {
 344         struct grant *gnt_list_entry;
 345
 346         BUG_ON(list_empty(&rinfo->grants));
 347         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 348                                           node);
 349         list_del(&gnt_list_entry->node);
 350
 351         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 352                 rinfo->persistent_gnts_c--;
 353
 354         return gnt_list_entry;
 355 }
 356
 357 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 358                                         const struct blkfront_info *info)
 359 {
 360         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 361                                                  info->xbdev->otherend_id,
 362                                                  gnt_list_entry->page,
 363                                                  0);
 364 }
 365
 366 static struct grant *get_grant(grant_ref_t *gref_head,
 367                                unsigned long gfn,
 368                                struct blkfront_ring_info *rinfo)
 369 {
 370         struct grant *gnt_list_entry = get_free_grant(rinfo);
 371         struct blkfront_info *info = rinfo->dev_info;
 372
 373         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 374                 return gnt_list_entry;
 375
 376         /* Assign a gref to this page */
 377         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 378         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 379         if (info->feature_persistent)
 380                 grant_foreign_access(gnt_list_entry, info);
 381         else {
 382                 /* Grant access to the GFN passed by the caller */
 383                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 384                                                 info->xbdev->otherend_id,
 385                                                 gfn, 0);
 386         }
 387
 388         return gnt_list_entry;
 389 }
 390
 391 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 392                                         struct blkfront_ring_info *rinfo)
 393 {
 394         struct grant *gnt_list_entry = get_free_grant(rinfo);
 395         struct blkfront_info *info = rinfo->dev_info;
 396
 397         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 398                 return gnt_list_entry;
 399
 400         /* Assign a gref to this page */
 401         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 402         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 403         if (!info->feature_persistent) {
 404                 struct page *indirect_page;
 405
 406                 /* Fetch a pre-allocated page to use for indirect grefs */
 407                 BUG_ON(list_empty(&rinfo->indirect_pages));
 408                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 409                                                  struct page, lru);
 410                 list_del(&indirect_page->lru);
 411                 gnt_list_entry->page = indirect_page;
 412         }
 413         grant_foreign_access(gnt_list_entry, info);
 414
 415         return gnt_list_entry;
 416 }
 417
 418 static const char *op_name(int op)
 419 {
 420         static const char *const names[] = {
 421                 [BLKIF_OP_READ] = "read",
 422                 [BLKIF_OP_WRITE] = "write",
 423                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 424                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 425                 [BLKIF_OP_DISCARD] = "discard" };
 426
 427         if (op < 0 || op >= ARRAY_SIZE(names))
 428                 return "unknown";
 429
 430         if (!names[op])
 431                 return "reserved";
 432
 433         return names[op];
 434 }
 435 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 436 {
 437         unsigned int end = minor + nr;
 438         int rc;
 439
 440         if (end > nr_minors) {
 441                 unsigned long *bitmap, *old;
 442
 443                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 444                                  GFP_KERNEL);
 445                 if (bitmap == NULL)
 446                         return -ENOMEM;
 447
 448                 spin_lock(&minor_lock);
 449                 if (end > nr_minors) {
 450                         old = minors;
 451                         memcpy(bitmap, minors,
 452                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 453                         minors = bitmap;
 454                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 455                 } else
 456                         old = bitmap;
 457                 spin_unlock(&minor_lock);
 458                 kfree(old);
 459         }
 460
 461         spin_lock(&minor_lock);
 462         if (find_next_bit(minors, end, minor) >= end) {
 463                 bitmap_set(minors, minor, nr);
 464                 rc = 0;
 465         } else
 466                 rc = -EBUSY;
 467         spin_unlock(&minor_lock);
 468
 469         return rc;
 470 }
 471
 472 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 473 {
 474         unsigned int end = minor + nr;
 475
 476         BUG_ON(end > nr_minors);
 477         spin_lock(&minor_lock);
 478         bitmap_clear(minors,  minor, nr);
 479         spin_unlock(&minor_lock);
 480 }
 481
 482 static void blkif_restart_queue_callback(void *arg)
 483 {
 484         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 485         schedule_work(&rinfo->work);
 486 }
 487
 488 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 489 {
 490         /* We don't have real geometry info, but let's at least return
 491            values consistent with the size of the device */
 492         sector_t nsect = get_capacity(bd->bd_disk);
 493         sector_t cylinders = nsect;
 494
 495         hg->heads = 0xff;
 496         hg->sectors = 0x3f;
 497         sector_div(cylinders, hg->heads * hg->sectors);
 498         hg->cylinders = cylinders;
 499         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 500                 hg->cylinders = 0xffff;
 501         return 0;
 502 }
 503
 504 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 505                        unsigned command, unsigned long argument)
 506 {
 507         int i;
 508
 509         switch (command) {
 510         case CDROMMULTISESSION:
 511                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 512                         if (put_user(0, (char __user *)(argument + i)))
 513                                 return -EFAULT;
 514                 return 0;
 515         case CDROM_GET_CAPABILITY:
 516                 if (bdev->bd_disk->flags & GENHD_FL_CD)
 517                         return 0;
 518                 return -EINVAL;
 519         default:
 520                 return -EINVAL;
 521         }
 522 }
 523
 524 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 525                                             struct request *req,
 526                                             struct blkif_request **ring_req)
 527 {
 528         unsigned long id;
 529
 530         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 531         rinfo->ring.req_prod_pvt++;
 532
 533         id = get_id_from_freelist(rinfo);
 534         rinfo->shadow[id].request = req;
 535         rinfo->shadow[id].status = REQ_PROCESSING;
 536         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 537
 538         rinfo->shadow[id].req.u.rw.id = id;
 539
 540         return id;
 541 }
 542
 543 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 544 {
 545         struct blkfront_info *info = rinfo->dev_info;
 546         struct blkif_request *ring_req, *final_ring_req;
 547         unsigned long id;
 548
 549         /* Fill out a communications ring structure. */
 550         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 551         ring_req = &rinfo->shadow[id].req;
 552
 553         ring_req->operation = BLKIF_OP_DISCARD;
 554         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 555         ring_req->u.discard.id = id;
 556         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 557         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 558                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 559         else
 560                 ring_req->u.discard.flag = 0;
 561
 562         /* Copy the request to the ring page. */
 563         *final_ring_req = *ring_req;
 564         rinfo->shadow[id].status = REQ_WAITING;
 565
 566         return 0;
 567 }
 568
 569 struct setup_rw_req {
 570         unsigned int grant_idx;
 571         struct blkif_request_segment *segments;
 572         struct blkfront_ring_info *rinfo;
 573         struct blkif_request *ring_req;
 574         grant_ref_t gref_head;
 575         unsigned int id;
 576         /* Only used when persistent grant is used and it's a read request */
 577         bool need_copy;
 578         unsigned int bvec_off;
 579         char *bvec_data;
 580
 581         bool require_extra_req;
 582         struct blkif_request *extra_ring_req;
 583 };
 584
 585 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 586                                      unsigned int len, void *data)
 587 {
 588         struct setup_rw_req *setup = data;
 589         int n, ref;
 590         struct grant *gnt_list_entry;
 591         unsigned int fsect, lsect;
 592         /* Convenient aliases */
 593         unsigned int grant_idx = setup->grant_idx;
 594         struct blkif_request *ring_req = setup->ring_req;
 595         struct blkfront_ring_info *rinfo = setup->rinfo;
 596         /*
 597          * We always use the shadow of the first request to store the list
 598          * of grant associated to the block I/O request. This made the
 599          * completion more easy to handle even if the block I/O request is
 600          * split.
 601          */
 602         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 603
 604         if (unlikely(setup->require_extra_req &&
 605                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 606                 /*
 607                  * We are using the second request, setup grant_idx
 608                  * to be the index of the segment array.
 609                  */
 610                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 611                 ring_req = setup->extra_ring_req;
 612         }
 613
 614         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 615             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 616                 if (setup->segments)
 617                         kunmap_atomic(setup->segments);
 618
 619                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 620                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 621                 shadow->indirect_grants[n] = gnt_list_entry;
 622                 setup->segments = kmap_atomic(gnt_list_entry->page);
 623                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 624         }
 625
 626         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 627         ref = gnt_list_entry->gref;
 628         /*
 629          * All the grants are stored in the shadow of the first
 630          * request. Therefore we have to use the global index.
 631          */
 632         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 633
 634         if (setup->need_copy) {
 635                 void *shared_data;
 636
 637                 shared_data = kmap_atomic(gnt_list_entry->page);
 638                 /*
 639                  * this does not wipe data stored outside the
 640                  * range sg->offset..sg->offset+sg->length.
 641                  * Therefore, blkback *could* see data from
 642                  * previous requests. This is OK as long as
 643                  * persistent grants are shared with just one
 644                  * domain. It may need refactoring if this
 645                  * changes
 646                  */
 647                 memcpy(shared_data + offset,
 648                        setup->bvec_data + setup->bvec_off,
 649                        len);
 650
 651                 kunmap_atomic(shared_data);
 652                 setup->bvec_off += len;
 653         }
 654
 655         fsect = offset >> 9;
 656         lsect = fsect + (len >> 9) - 1;
 657         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 658                 ring_req->u.rw.seg[grant_idx] =
 659                         (struct blkif_request_segment) {
 660                                 .gref       = ref,
 661                                 .first_sect = fsect,
 662                                 .last_sect  = lsect };
 663         } else {
 664                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 665                         (struct blkif_request_segment) {
 666                                 .gref       = ref,
 667                                 .first_sect = fsect,
 668                                 .last_sect  = lsect };
 669         }
 670
 671         (setup->grant_idx)++;
 672 }
 673
 674 static void blkif_setup_extra_req(struct blkif_request *first,
 675                                   struct blkif_request *second)
 676 {
 677         uint16_t nr_segments = first->u.rw.nr_segments;
 678
 679         /*
 680          * The second request is only present when the first request uses
 681          * all its segments. It's always the continuity of the first one.
 682          */
 683         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 684
 685         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 686         second->u.rw.sector_number = first->u.rw.sector_number +
 687                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 688
 689         second->u.rw.handle = first->u.rw.handle;
 690         second->operation = first->operation;
 691 }
 692
 693 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 694 {
 695         struct blkfront_info *info = rinfo->dev_info;
 696         struct blkif_request *ring_req, *extra_ring_req = NULL;
 697         struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
 698         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 699         bool require_extra_req = false;
 700         int i;
 701         struct setup_rw_req setup = {
 702                 .grant_idx = 0,
 703                 .segments = NULL,
 704                 .rinfo = rinfo,
 705                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 706         };
 707
 708         /*
 709          * Used to store if we are able to queue the request by just using
 710          * existing persistent grants, or if we have to get new grants,
 711          * as there are not sufficiently many free.
 712          */
 713         bool new_persistent_gnts = false;
 714         struct scatterlist *sg;
 715         int num_sg, max_grefs, num_grant;
 716
 717         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 718         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 719                 /*
 720                  * If we are using indirect segments we need to account
 721                  * for the indirect grefs used in the request.
 722                  */
 723                 max_grefs += INDIRECT_GREFS(max_grefs);
 724
 725         /* Check if we have enough persistent grants to allocate a requests */
 726         if (rinfo->persistent_gnts_c < max_grefs) {
 727                 new_persistent_gnts = true;
 728
 729                 if (gnttab_alloc_grant_references(
 730                     max_grefs - rinfo->persistent_gnts_c,
 731                     &setup.gref_head) < 0) {
 732                         gnttab_request_free_callback(
 733                                 &rinfo->callback,
 734                                 blkif_restart_queue_callback,
 735                                 rinfo,
 736                                 max_grefs - rinfo->persistent_gnts_c);
 737                         return 1;
 738                 }
 739         }
 740
 741         /* Fill out a communications ring structure. */
 742         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 743         ring_req = &rinfo->shadow[id].req;
 744
 745         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 746         num_grant = 0;
 747         /* Calculate the number of grant used */
 748         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 749                num_grant += gnttab_count_grant(sg->offset, sg->length);
 750
 751         require_extra_req = info->max_indirect_segments == 0 &&
 752                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 753         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 754
 755         rinfo->shadow[id].num_sg = num_sg;
 756         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 757             likely(!require_extra_req)) {
 758                 /*
 759                  * The indirect operation can only be a BLKIF_OP_READ or
 760                  * BLKIF_OP_WRITE
 761                  */
 762                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 763                 ring_req->operation = BLKIF_OP_INDIRECT;
 764                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 765                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 766                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 767                 ring_req->u.indirect.handle = info->handle;
 768                 ring_req->u.indirect.nr_segments = num_grant;
 769         } else {
 770                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 771                 ring_req->u.rw.handle = info->handle;
 772                 ring_req->operation = rq_data_dir(req) ?
 773                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 774                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 775                         /*
 776                          * Ideally we can do an unordered flush-to-disk.
 777                          * In case the backend onlysupports barriers, use that.
 778                          * A barrier request a superset of FUA, so we can
 779                          * implement it the same way.  (It's also a FLUSH+FUA,
 780                          * since it is guaranteed ordered WRT previous writes.)
 781                          */
 782                         if (info->feature_flush && info->feature_fua)
 783                                 ring_req->operation =
 784                                         BLKIF_OP_WRITE_BARRIER;
 785                         else if (info->feature_flush)
 786                                 ring_req->operation =
 787                                         BLKIF_OP_FLUSH_DISKCACHE;
 788                         else
 789                                 ring_req->operation = 0;
 790                 }
 791                 ring_req->u.rw.nr_segments = num_grant;
 792                 if (unlikely(require_extra_req)) {
 793                         extra_id = blkif_ring_get_request(rinfo, req,
 794                                                           &final_extra_ring_req);
 795                         extra_ring_req = &rinfo->shadow[extra_id].req;
 796
 797                         /*
 798                          * Only the first request contains the scatter-gather
 799                          * list.
 800                          */
 801                         rinfo->shadow[extra_id].num_sg = 0;
 802
 803                         blkif_setup_extra_req(ring_req, extra_ring_req);
 804
 805                         /* Link the 2 requests together */
 806                         rinfo->shadow[extra_id].associated_id = id;
 807                         rinfo->shadow[id].associated_id = extra_id;
 808                 }
 809         }
 810
 811         setup.ring_req = ring_req;
 812         setup.id = id;
 813
 814         setup.require_extra_req = require_extra_req;
 815         if (unlikely(require_extra_req))
 816                 setup.extra_ring_req = extra_ring_req;
 817
 818         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 819                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 820
 821                 if (setup.need_copy) {
 822                         setup.bvec_off = sg->offset;
 823                         setup.bvec_data = kmap_atomic(sg_page(sg));
 824                 }
 825
 826                 gnttab_foreach_grant_in_range(sg_page(sg),
 827                                               sg->offset,
 828                                               sg->length,
 829                                               blkif_setup_rw_req_grant,
 830                                               &setup);
 831
 832                 if (setup.need_copy)
 833                         kunmap_atomic(setup.bvec_data);
 834         }
 835         if (setup.segments)
 836                 kunmap_atomic(setup.segments);
 837
 838         /* Copy request(s) to the ring page. */
 839         *final_ring_req = *ring_req;
 840         rinfo->shadow[id].status = REQ_WAITING;
 841         if (unlikely(require_extra_req)) {
 842                 *final_extra_ring_req = *extra_ring_req;
 843                 rinfo->shadow[extra_id].status = REQ_WAITING;
 844         }
 845
 846         if (new_persistent_gnts)
 847                 gnttab_free_grant_references(setup.gref_head);
 848
 849         return 0;
 850 }
 851
 852 /*
 853  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 854  * and writes are handled as expected.
 855  *
 856  * @req: a request struct
 857  */
 858 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 859 {
 860         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 861                 return 1;
 862
 863         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 864                      req_op(req) == REQ_OP_SECURE_ERASE))
 865                 return blkif_queue_discard_req(req, rinfo);
 866         else
 867                 return blkif_queue_rw_req(req, rinfo);
 868 }
 869
 870 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 871 {
 872         int notify;
 873
 874         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 875
 876         if (notify)
 877                 notify_remote_via_irq(rinfo->irq);
 878 }
 879
 880 static inline bool blkif_request_flush_invalid(struct request *req,
 881                                                struct blkfront_info *info)
 882 {
 883         return (blk_rq_is_passthrough(req) ||
 884                 ((req_op(req) == REQ_OP_FLUSH) &&
 885                  !info->feature_flush) ||
 886                 ((req->cmd_flags & REQ_FUA) &&
 887                  !info->feature_fua));
 888 }
 889
 890 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 891                           const struct blk_mq_queue_data *qd)
 892 {
 893         unsigned long flags;
 894         int qid = hctx->queue_num;
 895         struct blkfront_info *info = hctx->queue->queuedata;
 896         struct blkfront_ring_info *rinfo = NULL;
 897
 898         rinfo = get_rinfo(info, qid);
 899         blk_mq_start_request(qd->rq);
 900         spin_lock_irqsave(&rinfo->ring_lock, flags);
 901         if (RING_FULL(&rinfo->ring))
 902                 goto out_busy;
 903
 904         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 905                 goto out_err;
 906
 907         if (blkif_queue_request(qd->rq, rinfo))
 908                 goto out_busy;
 909
 910         flush_requests(rinfo);
 911         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 912         return BLK_STS_OK;
 913
 914 out_err:
 915         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 916         return BLK_STS_IOERR;
 917
 918 out_busy:
 919         blk_mq_stop_hw_queue(hctx);
 920         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 921         return BLK_STS_DEV_RESOURCE;
 922 }
 923
 924 static void blkif_complete_rq(struct request *rq)
 925 {
 926         blk_mq_end_request(rq, blkif_req(rq)->error);
 927 }
 928
 929 static const struct blk_mq_ops blkfront_mq_ops = {
 930         .queue_rq = blkif_queue_rq,
 931         .complete = blkif_complete_rq,
 932 };
 933
 934 static void blkif_set_queue_limits(struct blkfront_info *info)
 935 {
 936         struct request_queue *rq = info->rq;
 937         struct gendisk *gd = info->gd;
 938         unsigned int segments = info->max_indirect_segments ? :
 939                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 940
 941         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 942
 943         if (info->feature_discard) {
 944                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
 945                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 946                 rq->limits.discard_granularity = info->discard_granularity ?:
 947                                                  info->physical_sector_size;
 948                 rq->limits.discard_alignment = info->discard_alignment;
 949                 if (info->feature_secdiscard)
 950                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
 951         }
 952
 953         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 954         blk_queue_logical_block_size(rq, info->sector_size);
 955         blk_queue_physical_block_size(rq, info->physical_sector_size);
 956         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 957
 958         /* Each segment in a request is up to an aligned page in size. */
 959         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 960         blk_queue_max_segment_size(rq, PAGE_SIZE);
 961
 962         /* Ensure a merged request will fit in a single I/O ring slot. */
 963         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 964
 965         /* Make sure buffer addresses are sector-aligned. */
 966         blk_queue_dma_alignment(rq, 511);
 967 }
 968
 969 static const char *flush_info(struct blkfront_info *info)
 970 {
 971         if (info->feature_flush && info->feature_fua)
 972                 return "barrier: enabled;";
 973         else if (info->feature_flush)
 974                 return "flush diskcache: enabled;";
 975         else
 976                 return "barrier or flush: disabled;";
 977 }
 978
 979 static void xlvbd_flush(struct blkfront_info *info)
 980 {
 981         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
 982                               info->feature_fua ? true : false);
 983         pr_info("blkfront: %s: %s %s %s %s %s\n",
 984                 info->gd->disk_name, flush_info(info),
 985                 "persistent grants:", info->feature_persistent ?
 986                 "enabled;" : "disabled;", "indirect descriptors:",
 987                 info->max_indirect_segments ? "enabled;" : "disabled;");
 988 }
 989
 990 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 991 {
 992         int major;
 993         major = BLKIF_MAJOR(vdevice);
 994         *minor = BLKIF_MINOR(vdevice);
 995         switch (major) {
 996                 case XEN_IDE0_MAJOR:
 997                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 998                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 999                                 EMULATED_HD_DISK_MINOR_OFFSET;
1000                         break;
1001                 case XEN_IDE1_MAJOR:
1002                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1003                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1004                                 EMULATED_HD_DISK_MINOR_OFFSET;
1005                         break;
1006                 case XEN_SCSI_DISK0_MAJOR:
1007                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1008                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1009                         break;
1010                 case XEN_SCSI_DISK1_MAJOR:
1011                 case XEN_SCSI_DISK2_MAJOR:
1012                 case XEN_SCSI_DISK3_MAJOR:
1013                 case XEN_SCSI_DISK4_MAJOR:
1014                 case XEN_SCSI_DISK5_MAJOR:
1015                 case XEN_SCSI_DISK6_MAJOR:
1016                 case XEN_SCSI_DISK7_MAJOR:
1017                         *offset = (*minor / PARTS_PER_DISK) +
1018                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1019                                 EMULATED_SD_DISK_NAME_OFFSET;
1020                         *minor = *minor +
1021                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1022                                 EMULATED_SD_DISK_MINOR_OFFSET;
1023                         break;
1024                 case XEN_SCSI_DISK8_MAJOR:
1025                 case XEN_SCSI_DISK9_MAJOR:
1026                 case XEN_SCSI_DISK10_MAJOR:
1027                 case XEN_SCSI_DISK11_MAJOR:
1028                 case XEN_SCSI_DISK12_MAJOR:
1029                 case XEN_SCSI_DISK13_MAJOR:
1030                 case XEN_SCSI_DISK14_MAJOR:
1031                 case XEN_SCSI_DISK15_MAJOR:
1032                         *offset = (*minor / PARTS_PER_DISK) +
1033                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1034                                 EMULATED_SD_DISK_NAME_OFFSET;
1035                         *minor = *minor +
1036                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1037                                 EMULATED_SD_DISK_MINOR_OFFSET;
1038                         break;
1039                 case XENVBD_MAJOR:
1040                         *offset = *minor / PARTS_PER_DISK;
1041                         break;
1042                 default:
1043                         printk(KERN_WARNING "blkfront: your disk configuration is "
1044                                         "incorrect, please use an xvd device instead\n");
1045                         return -ENODEV;
1046         }
1047         return 0;
1048 }
1049
1050 static char *encode_disk_name(char *ptr, unsigned int n)
1051 {
1052         if (n >= 26)
1053                 ptr = encode_disk_name(ptr, n / 26 - 1);
1054         *ptr = 'a' + n % 26;
1055         return ptr + 1;
1056 }
1057
1058 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1059                                struct blkfront_info *info,
1060                                u16 vdisk_info, u16 sector_size,
1061                                unsigned int physical_sector_size)
1062 {
1063         struct gendisk *gd;
1064         int nr_minors = 1;
1065         int err;
1066         unsigned int offset;
1067         int minor;
1068         int nr_parts;
1069         char *ptr;
1070
1071         BUG_ON(info->gd != NULL);
1072         BUG_ON(info->rq != NULL);
1073
1074         if ((info->vdevice>>EXT_SHIFT) > 1) {
1075                 /* this is above the extended range; something is wrong */
1076                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1077                 return -ENODEV;
1078         }
1079
1080         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1081                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1082                 if (err)
1083                         return err;
1084                 nr_parts = PARTS_PER_DISK;
1085         } else {
1086                 minor = BLKIF_MINOR_EXT(info->vdevice);
1087                 nr_parts = PARTS_PER_EXT_DISK;
1088                 offset = minor / nr_parts;
1089                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1090                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1091                                         "emulated IDE disks,\n\t choose an xvd device name"
1092                                         "from xvde on\n", info->vdevice);
1093         }
1094         if (minor >> MINORBITS) {
1095                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1096                         info->vdevice, minor);
1097                 return -ENODEV;
1098         }
1099
1100         if ((minor % nr_parts) == 0)
1101                 nr_minors = nr_parts;
1102
1103         err = xlbd_reserve_minors(minor, nr_minors);
1104         if (err)
1105                 return err;
1106
1107         memset(&info->tag_set, 0, sizeof(info->tag_set));
1108         info->tag_set.ops = &blkfront_mq_ops;
1109         info->tag_set.nr_hw_queues = info->nr_rings;
1110         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1111                 /*
1112                  * When indirect descriptior is not supported, the I/O request
1113                  * will be split between multiple request in the ring.
1114                  * To avoid problems when sending the request, divide by
1115                  * 2 the depth of the queue.
1116                  */
1117                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
1118         } else
1119                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1120         info->tag_set.numa_node = NUMA_NO_NODE;
1121         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1122         info->tag_set.cmd_size = sizeof(struct blkif_req);
1123         info->tag_set.driver_data = info;
1124
1125         err = blk_mq_alloc_tag_set(&info->tag_set);
1126         if (err)
1127                 goto out_release_minors;
1128
1129         gd = blk_mq_alloc_disk(&info->tag_set, info);
1130         if (IS_ERR(gd)) {
1131                 err = PTR_ERR(gd);
1132                 goto out_free_tag_set;
1133         }
1134
1135         strcpy(gd->disk_name, DEV_NAME);
1136         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1137         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1138         if (nr_minors > 1)
1139                 *ptr = 0;
1140         else
1141                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1142                          "%d", minor & (nr_parts - 1));
1143
1144         gd->major = XENVBD_MAJOR;
1145         gd->first_minor = minor;
1146         gd->minors = nr_minors;
1147         gd->fops = &xlvbd_block_fops;
1148         gd->private_data = info;
1149         set_capacity(gd, capacity);
1150
1151         info->rq = gd->queue;
1152         info->gd = gd;
1153         info->sector_size = sector_size;
1154         info->physical_sector_size = physical_sector_size;
1155         blkif_set_queue_limits(info);
1156
1157         xlvbd_flush(info);
1158
1159         if (vdisk_info & VDISK_READONLY)
1160                 set_disk_ro(gd, 1);
1161
1162         if (vdisk_info & VDISK_REMOVABLE)
1163                 gd->flags |= GENHD_FL_REMOVABLE;
1164
1165         if (vdisk_info & VDISK_CDROM)
1166                 gd->flags |= GENHD_FL_CD;
1167
1168         return 0;
1169
1170 out_free_tag_set:
1171         blk_mq_free_tag_set(&info->tag_set);
1172 out_release_minors:
1173         xlbd_release_minors(minor, nr_minors);
1174         return err;
1175 }
1176
1177 /* Already hold rinfo->ring_lock. */
1178 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1179 {
1180         if (!RING_FULL(&rinfo->ring))
1181                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1182 }
1183
1184 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1185 {
1186         unsigned long flags;
1187
1188         spin_lock_irqsave(&rinfo->ring_lock, flags);
1189         kick_pending_request_queues_locked(rinfo);
1190         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1191 }
1192
1193 static void blkif_restart_queue(struct work_struct *work)
1194 {
1195         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1196
1197         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1198                 kick_pending_request_queues(rinfo);
1199 }
1200
1201 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1202 {
1203         struct grant *persistent_gnt, *n;
1204         struct blkfront_info *info = rinfo->dev_info;
1205         int i, j, segs;
1206
1207         /*
1208          * Remove indirect pages, this only happens when using indirect
1209          * descriptors but not persistent grants
1210          */
1211         if (!list_empty(&rinfo->indirect_pages)) {
1212                 struct page *indirect_page, *n;
1213
1214                 BUG_ON(info->feature_persistent);
1215                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1216                         list_del(&indirect_page->lru);
1217                         __free_page(indirect_page);
1218                 }
1219         }
1220
1221         /* Remove all persistent grants. */
1222         if (!list_empty(&rinfo->grants)) {
1223                 list_for_each_entry_safe(persistent_gnt, n,
1224                                          &rinfo->grants, node) {
1225                         list_del(&persistent_gnt->node);
1226                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1227                                 gnttab_end_foreign_access(persistent_gnt->gref,
1228                                                           0, 0UL);
1229                                 rinfo->persistent_gnts_c--;
1230                         }
1231                         if (info->feature_persistent)
1232                                 __free_page(persistent_gnt->page);
1233                         kfree(persistent_gnt);
1234                 }
1235         }
1236         BUG_ON(rinfo->persistent_gnts_c != 0);
1237
1238         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1239                 /*
1240                  * Clear persistent grants present in requests already
1241                  * on the shared ring
1242                  */
1243                 if (!rinfo->shadow[i].request)
1244                         goto free_shadow;
1245
1246                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1247                        rinfo->shadow[i].req.u.indirect.nr_segments :
1248                        rinfo->shadow[i].req.u.rw.nr_segments;
1249                 for (j = 0; j < segs; j++) {
1250                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1251                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1252                         if (info->feature_persistent)
1253                                 __free_page(persistent_gnt->page);
1254                         kfree(persistent_gnt);
1255                 }
1256
1257                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1258                         /*
1259                          * If this is not an indirect operation don't try to
1260                          * free indirect segments
1261                          */
1262                         goto free_shadow;
1263
1264                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1265                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1266                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1267                         __free_page(persistent_gnt->page);
1268                         kfree(persistent_gnt);
1269                 }
1270
1271 free_shadow:
1272                 kvfree(rinfo->shadow[i].grants_used);
1273                 rinfo->shadow[i].grants_used = NULL;
1274                 kvfree(rinfo->shadow[i].indirect_grants);
1275                 rinfo->shadow[i].indirect_grants = NULL;
1276                 kvfree(rinfo->shadow[i].sg);
1277                 rinfo->shadow[i].sg = NULL;
1278         }
1279
1280         /* No more gnttab callback work. */
1281         gnttab_cancel_free_callback(&rinfo->callback);
1282
1283         /* Flush gnttab callback work. Must be done with no locks held. */
1284         flush_work(&rinfo->work);
1285
1286         /* Free resources associated with old device channel. */
1287         for (i = 0; i < info->nr_ring_pages; i++) {
1288                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1289                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1290                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1291                 }
1292         }
1293         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1294         rinfo->ring.sring = NULL;
1295
1296         if (rinfo->irq)
1297                 unbind_from_irqhandler(rinfo->irq, rinfo);
1298         rinfo->evtchn = rinfo->irq = 0;
1299 }
1300
1301 static void blkif_free(struct blkfront_info *info, int suspend)
1302 {
1303         unsigned int i;
1304         struct blkfront_ring_info *rinfo;
1305
1306         /* Prevent new requests being issued until we fix things up. */
1307         info->connected = suspend ?
1308                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1309         /* No more blkif_request(). */
1310         if (info->rq)
1311                 blk_mq_stop_hw_queues(info->rq);
1312
1313         for_each_rinfo(info, rinfo, i)
1314                 blkif_free_ring(rinfo);
1315
1316         kvfree(info->rinfo);
1317         info->rinfo = NULL;
1318         info->nr_rings = 0;
1319 }
1320
1321 struct copy_from_grant {
1322         const struct blk_shadow *s;
1323         unsigned int grant_idx;
1324         unsigned int bvec_offset;
1325         char *bvec_data;
1326 };
1327
1328 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1329                                   unsigned int len, void *data)
1330 {
1331         struct copy_from_grant *info = data;
1332         char *shared_data;
1333         /* Convenient aliases */
1334         const struct blk_shadow *s = info->s;
1335
1336         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1337
1338         memcpy(info->bvec_data + info->bvec_offset,
1339                shared_data + offset, len);
1340
1341         info->bvec_offset += len;
1342         info->grant_idx++;
1343
1344         kunmap_atomic(shared_data);
1345 }
1346
1347 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1348 {
1349         switch (rsp)
1350         {
1351         case BLKIF_RSP_OKAY:
1352                 return REQ_DONE;
1353         case BLKIF_RSP_EOPNOTSUPP:
1354                 return REQ_EOPNOTSUPP;
1355         case BLKIF_RSP_ERROR:
1356         default:
1357                 return REQ_ERROR;
1358         }
1359 }
1360
1361 /*
1362  * Get the final status of the block request based on two ring response
1363  */
1364 static int blkif_get_final_status(enum blk_req_status s1,
1365                                   enum blk_req_status s2)
1366 {
1367         BUG_ON(s1 < REQ_DONE);
1368         BUG_ON(s2 < REQ_DONE);
1369
1370         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1371                 return BLKIF_RSP_ERROR;
1372         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1373                 return BLKIF_RSP_EOPNOTSUPP;
1374         return BLKIF_RSP_OKAY;
1375 }
1376
1377 static bool blkif_completion(unsigned long *id,
1378                              struct blkfront_ring_info *rinfo,
1379                              struct blkif_response *bret)
1380 {
1381         int i = 0;
1382         struct scatterlist *sg;
1383         int num_sg, num_grant;
1384         struct blkfront_info *info = rinfo->dev_info;
1385         struct blk_shadow *s = &rinfo->shadow[*id];
1386         struct copy_from_grant data = {
1387                 .grant_idx = 0,
1388         };
1389
1390         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1391                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1392
1393         /* The I/O request may be split in two. */
1394         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1395                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1396
1397                 /* Keep the status of the current response in shadow. */
1398                 s->status = blkif_rsp_to_req_status(bret->status);
1399
1400                 /* Wait the second response if not yet here. */
1401                 if (s2->status < REQ_DONE)
1402                         return false;
1403
1404                 bret->status = blkif_get_final_status(s->status,
1405                                                       s2->status);
1406
1407                 /*
1408                  * All the grants is stored in the first shadow in order
1409                  * to make the completion code simpler.
1410                  */
1411                 num_grant += s2->req.u.rw.nr_segments;
1412
1413                 /*
1414                  * The two responses may not come in order. Only the
1415                  * first request will store the scatter-gather list.
1416                  */
1417                 if (s2->num_sg != 0) {
1418                         /* Update "id" with the ID of the first response. */
1419                         *id = s->associated_id;
1420                         s = s2;
1421                 }
1422
1423                 /*
1424                  * We don't need anymore the second request, so recycling
1425                  * it now.
1426                  */
1427                 if (add_id_to_freelist(rinfo, s->associated_id))
1428                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1429                              info->gd->disk_name, s->associated_id);
1430         }
1431
1432         data.s = s;
1433         num_sg = s->num_sg;
1434
1435         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1436                 for_each_sg(s->sg, sg, num_sg, i) {
1437                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1438
1439                         data.bvec_offset = sg->offset;
1440                         data.bvec_data = kmap_atomic(sg_page(sg));
1441
1442                         gnttab_foreach_grant_in_range(sg_page(sg),
1443                                                       sg->offset,
1444                                                       sg->length,
1445                                                       blkif_copy_from_grant,
1446                                                       &data);
1447
1448                         kunmap_atomic(data.bvec_data);
1449                 }
1450         }
1451         /* Add the persistent grant into the list of free grants */
1452         for (i = 0; i < num_grant; i++) {
1453                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1454                         /*
1455                          * If the grant is still mapped by the backend (the
1456                          * backend has chosen to make this grant persistent)
1457                          * we add it at the head of the list, so it will be
1458                          * reused first.
1459                          */
1460                         if (!info->feature_persistent)
1461                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1462                                                      s->grants_used[i]->gref);
1463                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1464                         rinfo->persistent_gnts_c++;
1465                 } else {
1466                         /*
1467                          * If the grant is not mapped by the backend we end the
1468                          * foreign access and add it to the tail of the list,
1469                          * so it will not be picked again unless we run out of
1470                          * persistent grants.
1471                          */
1472                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1473                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1474                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1475                 }
1476         }
1477         if (s->req.operation == BLKIF_OP_INDIRECT) {
1478                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1479                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1480                                 if (!info->feature_persistent)
1481                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1482                                                              s->indirect_grants[i]->gref);
1483                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1484                                 rinfo->persistent_gnts_c++;
1485                         } else {
1486                                 struct page *indirect_page;
1487
1488                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1489                                 /*
1490                                  * Add the used indirect page back to the list of
1491                                  * available pages for indirect grefs.
1492                                  */
1493                                 if (!info->feature_persistent) {
1494                                         indirect_page = s->indirect_grants[i]->page;
1495                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1496                                 }
1497                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1498                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1499                         }
1500                 }
1501         }
1502
1503         return true;
1504 }
1505
1506 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1507 {
1508         struct request *req;
1509         struct blkif_response bret;
1510         RING_IDX i, rp;
1511         unsigned long flags;
1512         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1513         struct blkfront_info *info = rinfo->dev_info;
1514
1515         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1516                 return IRQ_HANDLED;
1517
1518         spin_lock_irqsave(&rinfo->ring_lock, flags);
1519  again:
1520         rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1521         virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1522         if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1523                 pr_alert("%s: illegal number of responses %u\n",
1524                          info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1525                 goto err;
1526         }
1527
1528         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1529                 unsigned long id;
1530                 unsigned int op;
1531
1532                 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1533                 id = bret.id;
1534
1535                 /*
1536                  * The backend has messed up and given us an id that we would
1537                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1538                  * look in get_id_from_freelist.
1539                  */
1540                 if (id >= BLK_RING_SIZE(info)) {
1541                         pr_alert("%s: response has incorrect id (%ld)\n",
1542                                  info->gd->disk_name, id);
1543                         goto err;
1544                 }
1545                 if (rinfo->shadow[id].status != REQ_WAITING) {
1546                         pr_alert("%s: response references no pending request\n",
1547                                  info->gd->disk_name);
1548                         goto err;
1549                 }
1550
1551                 rinfo->shadow[id].status = REQ_PROCESSING;
1552                 req  = rinfo->shadow[id].request;
1553
1554                 op = rinfo->shadow[id].req.operation;
1555                 if (op == BLKIF_OP_INDIRECT)
1556                         op = rinfo->shadow[id].req.u.indirect.indirect_op;
1557                 if (bret.operation != op) {
1558                         pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1559                                  info->gd->disk_name, bret.operation, op);
1560                         goto err;
1561                 }
1562
1563                 if (bret.operation != BLKIF_OP_DISCARD) {
1564                         /*
1565                          * We may need to wait for an extra response if the
1566                          * I/O request is split in 2
1567                          */
1568                         if (!blkif_completion(&id, rinfo, &bret))
1569                                 continue;
1570                 }
1571
1572                 if (add_id_to_freelist(rinfo, id)) {
1573                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1574                              info->gd->disk_name, op_name(bret.operation), id);
1575                         continue;
1576                 }
1577
1578                 if (bret.status == BLKIF_RSP_OKAY)
1579                         blkif_req(req)->error = BLK_STS_OK;
1580                 else
1581                         blkif_req(req)->error = BLK_STS_IOERR;
1582
1583                 switch (bret.operation) {
1584                 case BLKIF_OP_DISCARD:
1585                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1586                                 struct request_queue *rq = info->rq;
1587
1588                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1589                                            info->gd->disk_name, op_name(bret.operation));
1590                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1591                                 info->feature_discard = 0;
1592                                 info->feature_secdiscard = 0;
1593                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1594                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1595                         }
1596                         break;
1597                 case BLKIF_OP_FLUSH_DISKCACHE:
1598                 case BLKIF_OP_WRITE_BARRIER:
1599                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1600                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1601                                        info->gd->disk_name, op_name(bret.operation));
1602                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1603                         }
1604                         if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1605                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1606                                 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1607                                        info->gd->disk_name, op_name(bret.operation));
1608                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1609                         }
1610                         if (unlikely(blkif_req(req)->error)) {
1611                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1612                                         blkif_req(req)->error = BLK_STS_OK;
1613                                 info->feature_fua = 0;
1614                                 info->feature_flush = 0;
1615                                 xlvbd_flush(info);
1616                         }
1617                         fallthrough;
1618                 case BLKIF_OP_READ:
1619                 case BLKIF_OP_WRITE:
1620                         if (unlikely(bret.status != BLKIF_RSP_OKAY))
1621                                 dev_dbg_ratelimited(&info->xbdev->dev,
1622                                         "Bad return from blkdev data request: %#x\n",
1623                                         bret.status);
1624
1625                         break;
1626                 default:
1627                         BUG();
1628                 }
1629
1630                 if (likely(!blk_should_fake_timeout(req->q)))
1631                         blk_mq_complete_request(req);
1632         }
1633
1634         rinfo->ring.rsp_cons = i;
1635
1636         if (i != rinfo->ring.req_prod_pvt) {
1637                 int more_to_do;
1638                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1639                 if (more_to_do)
1640                         goto again;
1641         } else
1642                 rinfo->ring.sring->rsp_event = i + 1;
1643
1644         kick_pending_request_queues_locked(rinfo);
1645
1646         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1647
1648         return IRQ_HANDLED;
1649
1650  err:
1651         info->connected = BLKIF_STATE_ERROR;
1652
1653         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1654
1655         pr_alert("%s disabled for further use\n", info->gd->disk_name);
1656         return IRQ_HANDLED;
1657 }
1658
1659
1660 static int setup_blkring(struct xenbus_device *dev,
1661                          struct blkfront_ring_info *rinfo)
1662 {
1663         struct blkif_sring *sring;
1664         int err, i;
1665         struct blkfront_info *info = rinfo->dev_info;
1666         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1667         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1668
1669         for (i = 0; i < info->nr_ring_pages; i++)
1670                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1671
1672         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1673                                                        get_order(ring_size));
1674         if (!sring) {
1675                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1676                 return -ENOMEM;
1677         }
1678         SHARED_RING_INIT(sring);
1679         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1680
1681         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1682         if (err < 0) {
1683                 free_pages((unsigned long)sring, get_order(ring_size));
1684                 rinfo->ring.sring = NULL;
1685                 goto fail;
1686         }
1687         for (i = 0; i < info->nr_ring_pages; i++)
1688                 rinfo->ring_ref[i] = gref[i];
1689
1690         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1691         if (err)
1692                 goto fail;
1693
1694         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1695                                         "blkif", rinfo);
1696         if (err <= 0) {
1697                 xenbus_dev_fatal(dev, err,
1698                                  "bind_evtchn_to_irqhandler failed");
1699                 goto fail;
1700         }
1701         rinfo->irq = err;
1702
1703         return 0;
1704 fail:
1705         blkif_free(info, 0);
1706         return err;
1707 }
1708
1709 /*
1710  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1711  * ring buffer may have multi pages depending on ->nr_ring_pages.
1712  */
1713 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1714                                 struct blkfront_ring_info *rinfo, const char *dir)
1715 {
1716         int err;
1717         unsigned int i;
1718         const char *message = NULL;
1719         struct blkfront_info *info = rinfo->dev_info;
1720
1721         if (info->nr_ring_pages == 1) {
1722                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1723                 if (err) {
1724                         message = "writing ring-ref";
1725                         goto abort_transaction;
1726                 }
1727         } else {
1728                 for (i = 0; i < info->nr_ring_pages; i++) {
1729                         char ring_ref_name[RINGREF_NAME_LEN];
1730
1731                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1732                         err = xenbus_printf(xbt, dir, ring_ref_name,
1733                                             "%u", rinfo->ring_ref[i]);
1734                         if (err) {
1735                                 message = "writing ring-ref";
1736                                 goto abort_transaction;
1737                         }
1738                 }
1739         }
1740
1741         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1742         if (err) {
1743                 message = "writing event-channel";
1744                 goto abort_transaction;
1745         }
1746
1747         return 0;
1748
1749 abort_transaction:
1750         xenbus_transaction_end(xbt, 1);
1751         if (message)
1752                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1753
1754         return err;
1755 }
1756
1757 /* Common code used when first setting up, and when resuming. */
1758 static int talk_to_blkback(struct xenbus_device *dev,
1759                            struct blkfront_info *info)
1760 {
1761         const char *message = NULL;
1762         struct xenbus_transaction xbt;
1763         int err;
1764         unsigned int i, max_page_order;
1765         unsigned int ring_page_order;
1766         struct blkfront_ring_info *rinfo;
1767
1768         if (!info)
1769                 return -ENODEV;
1770
1771         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1772                                               "max-ring-page-order", 0);
1773         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1774         info->nr_ring_pages = 1 << ring_page_order;
1775
1776         err = negotiate_mq(info);
1777         if (err)
1778                 goto destroy_blkring;
1779
1780         for_each_rinfo(info, rinfo, i) {
1781                 /* Create shared ring, alloc event channel. */
1782                 err = setup_blkring(dev, rinfo);
1783                 if (err)
1784                         goto destroy_blkring;
1785         }
1786
1787 again:
1788         err = xenbus_transaction_start(&xbt);
1789         if (err) {
1790                 xenbus_dev_fatal(dev, err, "starting transaction");
1791                 goto destroy_blkring;
1792         }
1793
1794         if (info->nr_ring_pages > 1) {
1795                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1796                                     ring_page_order);
1797                 if (err) {
1798                         message = "writing ring-page-order";
1799                         goto abort_transaction;
1800                 }
1801         }
1802
1803         /* We already got the number of queues/rings in _probe */
1804         if (info->nr_rings == 1) {
1805                 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1806                 if (err)
1807                         goto destroy_blkring;
1808         } else {
1809                 char *path;
1810                 size_t pathsize;
1811
1812                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1813                                     info->nr_rings);
1814                 if (err) {
1815                         message = "writing multi-queue-num-queues";
1816                         goto abort_transaction;
1817                 }
1818
1819                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1820                 path = kmalloc(pathsize, GFP_KERNEL);
1821                 if (!path) {
1822                         err = -ENOMEM;
1823                         message = "ENOMEM while writing ring references";
1824                         goto abort_transaction;
1825                 }
1826
1827                 for_each_rinfo(info, rinfo, i) {
1828                         memset(path, 0, pathsize);
1829                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1830                         err = write_per_ring_nodes(xbt, rinfo, path);
1831                         if (err) {
1832                                 kfree(path);
1833                                 goto destroy_blkring;
1834                         }
1835                 }
1836                 kfree(path);
1837         }
1838         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1839                             XEN_IO_PROTO_ABI_NATIVE);
1840         if (err) {
1841                 message = "writing protocol";
1842                 goto abort_transaction;
1843         }
1844         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1845                         info->feature_persistent);
1846         if (err)
1847                 dev_warn(&dev->dev,
1848                          "writing persistent grants feature to xenbus");
1849
1850         err = xenbus_transaction_end(xbt, 0);
1851         if (err) {
1852                 if (err == -EAGAIN)
1853                         goto again;
1854                 xenbus_dev_fatal(dev, err, "completing transaction");
1855                 goto destroy_blkring;
1856         }
1857
1858         for_each_rinfo(info, rinfo, i) {
1859                 unsigned int j;
1860
1861                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1862                         rinfo->shadow[j].req.u.rw.id = j + 1;
1863                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1864         }
1865         xenbus_switch_state(dev, XenbusStateInitialised);
1866
1867         return 0;
1868
1869  abort_transaction:
1870         xenbus_transaction_end(xbt, 1);
1871         if (message)
1872                 xenbus_dev_fatal(dev, err, "%s", message);
1873  destroy_blkring:
1874         blkif_free(info, 0);
1875         return err;
1876 }
1877
1878 static int negotiate_mq(struct blkfront_info *info)
1879 {
1880         unsigned int backend_max_queues;
1881         unsigned int i;
1882         struct blkfront_ring_info *rinfo;
1883
1884         BUG_ON(info->nr_rings);
1885
1886         /* Check if backend supports multiple queues. */
1887         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1888                                                   "multi-queue-max-queues", 1);
1889         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1890         /* We need at least one ring. */
1891         if (!info->nr_rings)
1892                 info->nr_rings = 1;
1893
1894         info->rinfo_size = struct_size(info->rinfo, shadow,
1895                                        BLK_RING_SIZE(info));
1896         info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1897         if (!info->rinfo) {
1898                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1899                 info->nr_rings = 0;
1900                 return -ENOMEM;
1901         }
1902
1903         for_each_rinfo(info, rinfo, i) {
1904                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1905                 INIT_LIST_HEAD(&rinfo->grants);
1906                 rinfo->dev_info = info;
1907                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1908                 spin_lock_init(&rinfo->ring_lock);
1909         }
1910         return 0;
1911 }
1912
1913 /* Enable the persistent grants feature. */
1914 static bool feature_persistent = true;
1915 module_param(feature_persistent, bool, 0644);
1916 MODULE_PARM_DESC(feature_persistent,
1917                 "Enables the persistent grants feature");
1918
1919 /*
1920  * Entry point to this code when a new device is created.  Allocate the basic
1921  * structures and the ring buffer for communication with the backend, and
1922  * inform the backend of the appropriate details for those.  Switch to
1923  * Initialised state.
1924  */
1925 static int blkfront_probe(struct xenbus_device *dev,
1926                           const struct xenbus_device_id *id)
1927 {
1928         int err, vdevice;
1929         struct blkfront_info *info;
1930
1931         /* FIXME: Use dynamic device id if this is not set. */
1932         err = xenbus_scanf(XBT_NIL, dev->nodename,
1933                            "virtual-device", "%i", &vdevice);
1934         if (err != 1) {
1935                 /* go looking in the extended area instead */
1936                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1937                                    "%i", &vdevice);
1938                 if (err != 1) {
1939                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1940                         return err;
1941                 }
1942         }
1943
1944         if (xen_hvm_domain()) {
1945                 char *type;
1946                 int len;
1947                 /* no unplug has been done: do not hook devices != xen vbds */
1948                 if (xen_has_pv_and_legacy_disk_devices()) {
1949                         int major;
1950
1951                         if (!VDEV_IS_EXTENDED(vdevice))
1952                                 major = BLKIF_MAJOR(vdevice);
1953                         else
1954                                 major = XENVBD_MAJOR;
1955
1956                         if (major != XENVBD_MAJOR) {
1957                                 printk(KERN_INFO
1958                                                 "%s: HVM does not support vbd %d as xen block device\n",
1959                                                 __func__, vdevice);
1960                                 return -ENODEV;
1961                         }
1962                 }
1963                 /* do not create a PV cdrom device if we are an HVM guest */
1964                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1965                 if (IS_ERR(type))
1966                         return -ENODEV;
1967                 if (strncmp(type, "cdrom", 5) == 0) {
1968                         kfree(type);
1969                         return -ENODEV;
1970                 }
1971                 kfree(type);
1972         }
1973         info = kzalloc(sizeof(*info), GFP_KERNEL);
1974         if (!info) {
1975                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1976                 return -ENOMEM;
1977         }
1978
1979         info->xbdev = dev;
1980
1981         mutex_init(&info->mutex);
1982         info->vdevice = vdevice;
1983         info->connected = BLKIF_STATE_DISCONNECTED;
1984
1985         info->feature_persistent = feature_persistent;
1986
1987         /* Front end dir is a number, which is used as the id. */
1988         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1989         dev_set_drvdata(&dev->dev, info);
1990
1991         mutex_lock(&blkfront_mutex);
1992         list_add(&info->info_list, &info_list);
1993         mutex_unlock(&blkfront_mutex);
1994
1995         return 0;
1996 }
1997
1998 static int blkif_recover(struct blkfront_info *info)
1999 {
2000         unsigned int r_index;
2001         struct request *req, *n;
2002         int rc;
2003         struct bio *bio;
2004         unsigned int segs;
2005         struct blkfront_ring_info *rinfo;
2006
2007         blkfront_gather_backend_features(info);
2008         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2009         blkif_set_queue_limits(info);
2010         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2011         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2012
2013         for_each_rinfo(info, rinfo, r_index) {
2014                 rc = blkfront_setup_indirect(rinfo);
2015                 if (rc)
2016                         return rc;
2017         }
2018         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2019
2020         /* Now safe for us to use the shared ring */
2021         info->connected = BLKIF_STATE_CONNECTED;
2022
2023         for_each_rinfo(info, rinfo, r_index) {
2024                 /* Kick any other new requests queued since we resumed */
2025                 kick_pending_request_queues(rinfo);
2026         }
2027
2028         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2029                 /* Requeue pending requests (flush or discard) */
2030                 list_del_init(&req->queuelist);
2031                 BUG_ON(req->nr_phys_segments > segs);
2032                 blk_mq_requeue_request(req, false);
2033         }
2034         blk_mq_start_stopped_hw_queues(info->rq, true);
2035         blk_mq_kick_requeue_list(info->rq);
2036
2037         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2038                 /* Traverse the list of pending bios and re-queue them */
2039                 submit_bio(bio);
2040         }
2041
2042         return 0;
2043 }
2044
2045 /*
2046  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2047  * driver restart.  We tear down our blkif structure and recreate it, but
2048  * leave the device-layer structures intact so that this is transparent to the
2049  * rest of the kernel.
2050  */
2051 static int blkfront_resume(struct xenbus_device *dev)
2052 {
2053         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2054         int err = 0;
2055         unsigned int i, j;
2056         struct blkfront_ring_info *rinfo;
2057
2058         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2059
2060         bio_list_init(&info->bio_list);
2061         INIT_LIST_HEAD(&info->requests);
2062         for_each_rinfo(info, rinfo, i) {
2063                 struct bio_list merge_bio;
2064                 struct blk_shadow *shadow = rinfo->shadow;
2065
2066                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2067                         /* Not in use? */
2068                         if (!shadow[j].request)
2069                                 continue;
2070
2071                         /*
2072                          * Get the bios in the request so we can re-queue them.
2073                          */
2074                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2075                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2076                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2077                             shadow[j].request->cmd_flags & REQ_FUA) {
2078                                 /*
2079                                  * Flush operations don't contain bios, so
2080                                  * we need to requeue the whole request
2081                                  *
2082                                  * XXX: but this doesn't make any sense for a
2083                                  * write with the FUA flag set..
2084                                  */
2085                                 list_add(&shadow[j].request->queuelist, &info->requests);
2086                                 continue;
2087                         }
2088                         merge_bio.head = shadow[j].request->bio;
2089                         merge_bio.tail = shadow[j].request->biotail;
2090                         bio_list_merge(&info->bio_list, &merge_bio);
2091                         shadow[j].request->bio = NULL;
2092                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2093                 }
2094         }
2095
2096         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2097
2098         err = talk_to_blkback(dev, info);
2099         if (!err)
2100                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2101
2102         /*
2103          * We have to wait for the backend to switch to
2104          * connected state, since we want to read which
2105          * features it supports.
2106          */
2107
2108         return err;
2109 }
2110
2111 static void blkfront_closing(struct blkfront_info *info)
2112 {
2113         struct xenbus_device *xbdev = info->xbdev;
2114         struct blkfront_ring_info *rinfo;
2115         unsigned int i;
2116
2117         if (xbdev->state == XenbusStateClosing)
2118                 return;
2119
2120         /* No more blkif_request(). */
2121         blk_mq_stop_hw_queues(info->rq);
2122         blk_set_queue_dying(info->rq);
2123         set_capacity(info->gd, 0);
2124
2125         for_each_rinfo(info, rinfo, i) {
2126                 /* No more gnttab callback work. */
2127                 gnttab_cancel_free_callback(&rinfo->callback);
2128
2129                 /* Flush gnttab callback work. Must be done with no locks held. */
2130                 flush_work(&rinfo->work);
2131         }
2132
2133         xenbus_frontend_closed(xbdev);
2134 }
2135
2136 static void blkfront_setup_discard(struct blkfront_info *info)
2137 {
2138         info->feature_discard = 1;
2139         info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2140                                                          "discard-granularity",
2141                                                          0);
2142         info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2143                                                        "discard-alignment", 0);
2144         info->feature_secdiscard =
2145                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2146                                        0);
2147 }
2148
2149 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2150 {
2151         unsigned int psegs, grants, memflags;
2152         int err, i;
2153         struct blkfront_info *info = rinfo->dev_info;
2154
2155         memflags = memalloc_noio_save();
2156
2157         if (info->max_indirect_segments == 0) {
2158                 if (!HAS_EXTRA_REQ)
2159                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2160                 else {
2161                         /*
2162                          * When an extra req is required, the maximum
2163                          * grants supported is related to the size of the
2164                          * Linux block segment.
2165                          */
2166                         grants = GRANTS_PER_PSEG;
2167                 }
2168         }
2169         else
2170                 grants = info->max_indirect_segments;
2171         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2172
2173         err = fill_grant_buffer(rinfo,
2174                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2175         if (err)
2176                 goto out_of_memory;
2177
2178         if (!info->feature_persistent && info->max_indirect_segments) {
2179                 /*
2180                  * We are using indirect descriptors but not persistent
2181                  * grants, we need to allocate a set of pages that can be
2182                  * used for mapping indirect grefs
2183                  */
2184                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2185
2186                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2187                 for (i = 0; i < num; i++) {
2188                         struct page *indirect_page = alloc_page(GFP_KERNEL);
2189                         if (!indirect_page)
2190                                 goto out_of_memory;
2191                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2192                 }
2193         }
2194
2195         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2196                 rinfo->shadow[i].grants_used =
2197                         kvcalloc(grants,
2198                                  sizeof(rinfo->shadow[i].grants_used[0]),
2199                                  GFP_KERNEL);
2200                 rinfo->shadow[i].sg = kvcalloc(psegs,
2201                                                sizeof(rinfo->shadow[i].sg[0]),
2202                                                GFP_KERNEL);
2203                 if (info->max_indirect_segments)
2204                         rinfo->shadow[i].indirect_grants =
2205                                 kvcalloc(INDIRECT_GREFS(grants),
2206                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2207                                          GFP_KERNEL);
2208                 if ((rinfo->shadow[i].grants_used == NULL) ||
2209                         (rinfo->shadow[i].sg == NULL) ||
2210                      (info->max_indirect_segments &&
2211                      (rinfo->shadow[i].indirect_grants == NULL)))
2212                         goto out_of_memory;
2213                 sg_init_table(rinfo->shadow[i].sg, psegs);
2214         }
2215
2216         memalloc_noio_restore(memflags);
2217
2218         return 0;
2219
2220 out_of_memory:
2221         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2222                 kvfree(rinfo->shadow[i].grants_used);
2223                 rinfo->shadow[i].grants_used = NULL;
2224                 kvfree(rinfo->shadow[i].sg);
2225                 rinfo->shadow[i].sg = NULL;
2226                 kvfree(rinfo->shadow[i].indirect_grants);
2227                 rinfo->shadow[i].indirect_grants = NULL;
2228         }
2229         if (!list_empty(&rinfo->indirect_pages)) {
2230                 struct page *indirect_page, *n;
2231                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2232                         list_del(&indirect_page->lru);
2233                         __free_page(indirect_page);
2234                 }
2235         }
2236
2237         memalloc_noio_restore(memflags);
2238
2239         return -ENOMEM;
2240 }
2241
2242 /*
2243  * Gather all backend feature-*
2244  */
2245 static void blkfront_gather_backend_features(struct blkfront_info *info)
2246 {
2247         unsigned int indirect_segments;
2248
2249         info->feature_flush = 0;
2250         info->feature_fua = 0;
2251
2252         /*
2253          * If there's no "feature-barrier" defined, then it means
2254          * we're dealing with a very old backend which writes
2255          * synchronously; nothing to do.
2256          *
2257          * If there are barriers, then we use flush.
2258          */
2259         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2260                 info->feature_flush = 1;
2261                 info->feature_fua = 1;
2262         }
2263
2264         /*
2265          * And if there is "feature-flush-cache" use that above
2266          * barriers.
2267          */
2268         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2269                                  0)) {
2270                 info->feature_flush = 1;
2271                 info->feature_fua = 0;
2272         }
2273
2274         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2275                 blkfront_setup_discard(info);
2276
2277         if (info->feature_persistent)
2278                 info->feature_persistent =
2279                         !!xenbus_read_unsigned(info->xbdev->otherend,
2280                                                "feature-persistent", 0);
2281
2282         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2283                                         "feature-max-indirect-segments", 0);
2284         if (indirect_segments > xen_blkif_max_segments)
2285                 indirect_segments = xen_blkif_max_segments;
2286         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2287                 indirect_segments = 0;
2288         info->max_indirect_segments = indirect_segments;
2289
2290         if (info->feature_persistent) {
2291                 mutex_lock(&blkfront_mutex);
2292                 schedule_delayed_work(&blkfront_work, HZ * 10);
2293                 mutex_unlock(&blkfront_mutex);
2294         }
2295 }
2296
2297 /*
2298  * Invoked when the backend is finally 'ready' (and has told produced
2299  * the details about the physical device - #sectors, size, etc).
2300  */
2301 static void blkfront_connect(struct blkfront_info *info)
2302 {
2303         unsigned long long sectors;
2304         unsigned long sector_size;
2305         unsigned int physical_sector_size;
2306         unsigned int binfo;
2307         int err, i;
2308         struct blkfront_ring_info *rinfo;
2309
2310         switch (info->connected) {
2311         case BLKIF_STATE_CONNECTED:
2312                 /*
2313                  * Potentially, the back-end may be signalling
2314                  * a capacity change; update the capacity.
2315                  */
2316                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2317                                    "sectors", "%Lu", &sectors);
2318                 if (XENBUS_EXIST_ERR(err))
2319                         return;
2320                 printk(KERN_INFO "Setting capacity to %Lu\n",
2321                        sectors);
2322                 set_capacity_and_notify(info->gd, sectors);
2323
2324                 return;
2325         case BLKIF_STATE_SUSPENDED:
2326                 /*
2327                  * If we are recovering from suspension, we need to wait
2328                  * for the backend to announce it's features before
2329                  * reconnecting, at least we need to know if the backend
2330                  * supports indirect descriptors, and how many.
2331                  */
2332                 blkif_recover(info);
2333                 return;
2334
2335         default:
2336                 break;
2337         }
2338
2339         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2340                 __func__, info->xbdev->otherend);
2341
2342         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2343                             "sectors", "%llu", &sectors,
2344                             "info", "%u", &binfo,
2345                             "sector-size", "%lu", &sector_size,
2346                             NULL);
2347         if (err) {
2348                 xenbus_dev_fatal(info->xbdev, err,
2349                                  "reading backend fields at %s",
2350                                  info->xbdev->otherend);
2351                 return;
2352         }
2353
2354         /*
2355          * physical-sector-size is a newer field, so old backends may not
2356          * provide this. Assume physical sector size to be the same as
2357          * sector_size in that case.
2358          */
2359         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2360                                                     "physical-sector-size",
2361                                                     sector_size);
2362         blkfront_gather_backend_features(info);
2363         for_each_rinfo(info, rinfo, i) {
2364                 err = blkfront_setup_indirect(rinfo);
2365                 if (err) {
2366                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2367                                          info->xbdev->otherend);
2368                         blkif_free(info, 0);
2369                         break;
2370                 }
2371         }
2372
2373         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2374                                   physical_sector_size);
2375         if (err) {
2376                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2377                                  info->xbdev->otherend);
2378                 goto fail;
2379         }
2380
2381         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2382
2383         /* Kick pending requests. */
2384         info->connected = BLKIF_STATE_CONNECTED;
2385         for_each_rinfo(info, rinfo, i)
2386                 kick_pending_request_queues(rinfo);
2387
2388         device_add_disk(&info->xbdev->dev, info->gd, NULL);
2389
2390         info->is_ready = 1;
2391         return;
2392
2393 fail:
2394         blkif_free(info, 0);
2395         return;
2396 }
2397
2398 /*
2399  * Callback received when the backend's state changes.
2400  */
2401 static void blkback_changed(struct xenbus_device *dev,
2402                             enum xenbus_state backend_state)
2403 {
2404         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2405
2406         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2407
2408         switch (backend_state) {
2409         case XenbusStateInitWait:
2410                 if (dev->state != XenbusStateInitialising)
2411                         break;
2412                 if (talk_to_blkback(dev, info))
2413                         break;
2414                 break;
2415         case XenbusStateInitialising:
2416         case XenbusStateInitialised:
2417         case XenbusStateReconfiguring:
2418         case XenbusStateReconfigured:
2419         case XenbusStateUnknown:
2420                 break;
2421
2422         case XenbusStateConnected:
2423                 /*
2424                  * talk_to_blkback sets state to XenbusStateInitialised
2425                  * and blkfront_connect sets it to XenbusStateConnected
2426                  * (if connection went OK).
2427                  *
2428                  * If the backend (or toolstack) decides to poke at backend
2429                  * state (and re-trigger the watch by setting the state repeatedly
2430                  * to XenbusStateConnected (4)) we need to deal with this.
2431                  * This is allowed as this is used to communicate to the guest
2432                  * that the size of disk has changed!
2433                  */
2434                 if ((dev->state != XenbusStateInitialised) &&
2435                     (dev->state != XenbusStateConnected)) {
2436                         if (talk_to_blkback(dev, info))
2437                                 break;
2438                 }
2439
2440                 blkfront_connect(info);
2441                 break;
2442
2443         case XenbusStateClosed:
2444                 if (dev->state == XenbusStateClosed)
2445                         break;
2446                 fallthrough;
2447         case XenbusStateClosing:
2448                 blkfront_closing(info);
2449                 break;
2450         }
2451 }
2452
2453 static int blkfront_remove(struct xenbus_device *xbdev)
2454 {
2455         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2456
2457         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2458
2459         del_gendisk(info->gd);
2460
2461         mutex_lock(&blkfront_mutex);
2462         list_del(&info->info_list);
2463         mutex_unlock(&blkfront_mutex);
2464
2465         blkif_free(info, 0);
2466         xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2467         blk_cleanup_disk(info->gd);
2468         blk_mq_free_tag_set(&info->tag_set);
2469
2470         kfree(info);
2471         return 0;
2472 }
2473
2474 static int blkfront_is_ready(struct xenbus_device *dev)
2475 {
2476         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2477
2478         return info->is_ready && info->xbdev;
2479 }
2480
2481 static const struct block_device_operations xlvbd_block_fops =
2482 {
2483         .owner = THIS_MODULE,
2484         .getgeo = blkif_getgeo,
2485         .ioctl = blkif_ioctl,
2486         .compat_ioctl = blkdev_compat_ptr_ioctl,
2487 };
2488
2489
2490 static const struct xenbus_device_id blkfront_ids[] = {
2491         { "vbd" },
2492         { "" }
2493 };
2494
2495 static struct xenbus_driver blkfront_driver = {
2496         .ids  = blkfront_ids,
2497         .probe = blkfront_probe,
2498         .remove = blkfront_remove,
2499         .resume = blkfront_resume,
2500         .otherend_changed = blkback_changed,
2501         .is_ready = blkfront_is_ready,
2502 };
2503
2504 static void purge_persistent_grants(struct blkfront_info *info)
2505 {
2506         unsigned int i;
2507         unsigned long flags;
2508         struct blkfront_ring_info *rinfo;
2509
2510         for_each_rinfo(info, rinfo, i) {
2511                 struct grant *gnt_list_entry, *tmp;
2512
2513                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2514
2515                 if (rinfo->persistent_gnts_c == 0) {
2516                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2517                         continue;
2518                 }
2519
2520                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2521                                          node) {
2522                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2523                             gnttab_query_foreign_access(gnt_list_entry->gref))
2524                                 continue;
2525
2526                         list_del(&gnt_list_entry->node);
2527                         gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2528                         rinfo->persistent_gnts_c--;
2529                         gnt_list_entry->gref = GRANT_INVALID_REF;
2530                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2531                 }
2532
2533                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2534         }
2535 }
2536
2537 static void blkfront_delay_work(struct work_struct *work)
2538 {
2539         struct blkfront_info *info;
2540         bool need_schedule_work = false;
2541
2542         mutex_lock(&blkfront_mutex);
2543
2544         list_for_each_entry(info, &info_list, info_list) {
2545                 if (info->feature_persistent) {
2546                         need_schedule_work = true;
2547                         mutex_lock(&info->mutex);
2548                         purge_persistent_grants(info);
2549                         mutex_unlock(&info->mutex);
2550                 }
2551         }
2552
2553         if (need_schedule_work)
2554                 schedule_delayed_work(&blkfront_work, HZ * 10);
2555
2556         mutex_unlock(&blkfront_mutex);
2557 }
2558
2559 static int __init xlblk_init(void)
2560 {
2561         int ret;
2562         int nr_cpus = num_online_cpus();
2563
2564         if (!xen_domain())
2565                 return -ENODEV;
2566
2567         if (!xen_has_pv_disk_devices())
2568                 return -ENODEV;
2569
2570         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2571                 pr_warn("xen_blk: can't get major %d with name %s\n",
2572                         XENVBD_MAJOR, DEV_NAME);
2573                 return -ENODEV;
2574         }
2575
2576         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2577                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2578
2579         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2580                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2581                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2582                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2583         }
2584
2585         if (xen_blkif_max_queues > nr_cpus) {
2586                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2587                         xen_blkif_max_queues, nr_cpus);
2588                 xen_blkif_max_queues = nr_cpus;
2589         }
2590
2591         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2592
2593         ret = xenbus_register_frontend(&blkfront_driver);
2594         if (ret) {
2595                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2596                 return ret;
2597         }
2598
2599         return 0;
2600 }
2601 module_init(xlblk_init);
2602
2603
2604 static void __exit xlblk_exit(void)
2605 {
2606         cancel_delayed_work_sync(&blkfront_work);
2607
2608         xenbus_unregister_driver(&blkfront_driver);
2609         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2610         kfree(minors);
2611 }
2612 module_exit(xlblk_exit);
2613
2614 MODULE_DESCRIPTION("Xen virtual block device frontend");
2615 MODULE_LICENSE("GPL");
2616 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2617 MODULE_ALIAS("xen:vbd");
2618 MODULE_ALIAS("xenblk");