drivers/block/xen-blkfront.c

   1 /*
   2  * blkfront.c
   3  *
   4  * XenLinux virtual block device driver.
   5  *
   6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8  * Copyright (c) 2004, Christian Limpach
   9  * Copyright (c) 2004, Andrew Warfield
  10  * Copyright (c) 2005, Christopher Clark
  11  * Copyright (c) 2005, XenSource Ltd
  12  *
  13  * This program is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU General Public License version 2
  15  * as published by the Free Software Foundation; or, when distributed
  16  * separately from the Linux kernel or incorporated into other
  17  * software packages, subject to the following license:
  18  *
  19  * Permission is hereby granted, free of charge, to any person obtaining a copy
  20  * of this source file (the "Software"), to deal in the Software without
  21  * restriction, including without limitation the rights to use, copy, modify,
  22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23  * and to permit persons to whom the Software is furnished to do so, subject to
  24  * the following conditions:
  25  *
  26  * The above copyright notice and this permission notice shall be included in
  27  * all copies or substantial portions of the Software.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35  * IN THE SOFTWARE.
  36  */
  37
  38 #include <linux/interrupt.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/blk-mq.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/cdrom.h>
  43 #include <linux/module.h>
  44 #include <linux/slab.h>
  45 #include <linux/major.h>
  46 #include <linux/mutex.h>
  47 #include <linux/scatterlist.h>
  48 #include <linux/bitmap.h>
  49 #include <linux/list.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/sched/mm.h>
  52
  53 #include <xen/xen.h>
  54 #include <xen/xenbus.h>
  55 #include <xen/grant_table.h>
  56 #include <xen/events.h>
  57 #include <xen/page.h>
  58 #include <xen/platform_pci.h>
  59
  60 #include <xen/interface/grant_table.h>
  61 #include <xen/interface/io/blkif.h>
  62 #include <xen/interface/io/protocols.h>
  63
  64 #include <asm/xen/hypervisor.h>
  65
  66 /*
  67  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  68  * When Linux is using a different page size than Xen, it may not be possible
  69  * to put all the data in a single segment.
  70  * This can happen when the backend doesn't support indirect descriptor and
  71  * therefore the maximum amount of data that a request can carry is
  72  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  73  *
  74  * Note that we only support one extra request. So the Linux page size
  75  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  76  * 88KB.
  77  */
  78 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  79
  80 enum blkif_state {
  81         BLKIF_STATE_DISCONNECTED,
  82         BLKIF_STATE_CONNECTED,
  83         BLKIF_STATE_SUSPENDED,
  84         BLKIF_STATE_ERROR,
  85 };
  86
  87 struct grant {
  88         grant_ref_t gref;
  89         struct page *page;
  90         struct list_head node;
  91 };
  92
  93 enum blk_req_status {
  94         REQ_PROCESSING,
  95         REQ_WAITING,
  96         REQ_DONE,
  97         REQ_ERROR,
  98         REQ_EOPNOTSUPP,
  99 };
 100
 101 struct blk_shadow {
 102         struct blkif_request req;
 103         struct request *request;
 104         struct grant **grants_used;
 105         struct grant **indirect_grants;
 106         struct scatterlist *sg;
 107         unsigned int num_sg;
 108         enum blk_req_status status;
 109
 110         #define NO_ASSOCIATED_ID ~0UL
 111         /*
 112          * Id of the sibling if we ever need 2 requests when handling a
 113          * block I/O request
 114          */
 115         unsigned long associated_id;
 116 };
 117
 118 struct blkif_req {
 119         blk_status_t    error;
 120 };
 121
 122 static inline struct blkif_req *blkif_req(struct request *rq)
 123 {
 124         return blk_mq_rq_to_pdu(rq);
 125 }
 126
 127 static DEFINE_MUTEX(blkfront_mutex);
 128 static const struct block_device_operations xlvbd_block_fops;
 129 static struct delayed_work blkfront_work;
 130 static LIST_HEAD(info_list);
 131
 132 /*
 133  * Maximum number of segments in indirect requests, the actual value used by
 134  * the frontend driver is the minimum of this value and the value provided
 135  * by the backend driver.
 136  */
 137
 138 static unsigned int xen_blkif_max_segments = 32;
 139 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 140 MODULE_PARM_DESC(max_indirect_segments,
 141                  "Maximum amount of segments in indirect requests (default is 32)");
 142
 143 static unsigned int xen_blkif_max_queues = 4;
 144 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 145 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 146
 147 /*
 148  * Maximum order of pages to be used for the shared ring between front and
 149  * backend, 4KB page granularity is used.
 150  */
 151 static unsigned int xen_blkif_max_ring_order;
 152 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 153 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 154
 155 #define BLK_RING_SIZE(info)     \
 156         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 157
 158 /*
 159  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 160  * characters are enough. Define to 20 to keep consistent with backend.
 161  */
 162 #define RINGREF_NAME_LEN (20)
 163 /*
 164  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 165  */
 166 #define QUEUE_NAME_LEN (17)
 167
 168 /*
 169  *  Per-ring info.
 170  *  Every blkfront device can associate with one or more blkfront_ring_info,
 171  *  depending on how many hardware queues/rings to be used.
 172  */
 173 struct blkfront_ring_info {
 174         /* Lock to protect data in every ring buffer. */
 175         spinlock_t ring_lock;
 176         struct blkif_front_ring ring;
 177         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 178         unsigned int evtchn, irq;
 179         struct work_struct work;
 180         struct gnttab_free_callback callback;
 181         struct list_head indirect_pages;
 182         struct list_head grants;
 183         unsigned int persistent_gnts_c;
 184         unsigned long shadow_free;
 185         struct blkfront_info *dev_info;
 186         struct blk_shadow shadow[];
 187 };
 188
 189 /*
 190  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 191  * hang in private_data off the gendisk structure. We may end up
 192  * putting all kinds of interesting stuff here :-)
 193  */
 194 struct blkfront_info
 195 {
 196         struct mutex mutex;
 197         struct xenbus_device *xbdev;
 198         struct gendisk *gd;
 199         u16 sector_size;
 200         unsigned int physical_sector_size;
 201         unsigned long vdisk_info;
 202         int vdevice;
 203         blkif_vdev_t handle;
 204         enum blkif_state connected;
 205         /* Number of pages per ring buffer. */
 206         unsigned int nr_ring_pages;
 207         struct request_queue *rq;
 208         unsigned int feature_flush:1;
 209         unsigned int feature_fua:1;
 210         unsigned int feature_discard:1;
 211         unsigned int feature_secdiscard:1;
 212         unsigned int feature_persistent:1;
 213         unsigned int discard_granularity;
 214         unsigned int discard_alignment;
 215         /* Number of 4KB segments handled */
 216         unsigned int max_indirect_segments;
 217         int is_ready;
 218         struct blk_mq_tag_set tag_set;
 219         struct blkfront_ring_info *rinfo;
 220         unsigned int nr_rings;
 221         unsigned int rinfo_size;
 222         /* Save uncomplete reqs and bios for migration. */
 223         struct list_head requests;
 224         struct bio_list bio_list;
 225         struct list_head info_list;
 226 };
 227
 228 static unsigned int nr_minors;
 229 static unsigned long *minors;
 230 static DEFINE_SPINLOCK(minor_lock);
 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 = INVALID_GRANT_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 != INVALID_GRANT_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 != INVALID_GRANT_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 != INVALID_GRANT_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         struct blkfront_info *info = bdev->bd_disk->private_data;
 508         int i;
 509
 510         switch (command) {
 511         case CDROMMULTISESSION:
 512                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 513                         if (put_user(0, (char __user *)(argument + i)))
 514                                 return -EFAULT;
 515                 return 0;
 516         case CDROM_GET_CAPABILITY:
 517                 if (!(info->vdisk_info & VDISK_CDROM))
 518                         return -EINVAL;
 519                 return 0;
 520         default:
 521                 return -EINVAL;
 522         }
 523 }
 524
 525 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 526                                             struct request *req,
 527                                             struct blkif_request **ring_req)
 528 {
 529         unsigned long id;
 530
 531         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 532         rinfo->ring.req_prod_pvt++;
 533
 534         id = get_id_from_freelist(rinfo);
 535         rinfo->shadow[id].request = req;
 536         rinfo->shadow[id].status = REQ_PROCESSING;
 537         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 538
 539         rinfo->shadow[id].req.u.rw.id = id;
 540
 541         return id;
 542 }
 543
 544 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 545 {
 546         struct blkfront_info *info = rinfo->dev_info;
 547         struct blkif_request *ring_req, *final_ring_req;
 548         unsigned long id;
 549
 550         /* Fill out a communications ring structure. */
 551         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 552         ring_req = &rinfo->shadow[id].req;
 553
 554         ring_req->operation = BLKIF_OP_DISCARD;
 555         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 556         ring_req->u.discard.id = id;
 557         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 558         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 559                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 560         else
 561                 ring_req->u.discard.flag = 0;
 562
 563         /* Copy the request to the ring page. */
 564         *final_ring_req = *ring_req;
 565         rinfo->shadow[id].status = REQ_WAITING;
 566
 567         return 0;
 568 }
 569
 570 struct setup_rw_req {
 571         unsigned int grant_idx;
 572         struct blkif_request_segment *segments;
 573         struct blkfront_ring_info *rinfo;
 574         struct blkif_request *ring_req;
 575         grant_ref_t gref_head;
 576         unsigned int id;
 577         /* Only used when persistent grant is used and it's a write request */
 578         bool need_copy;
 579         unsigned int bvec_off;
 580         char *bvec_data;
 581
 582         bool require_extra_req;
 583         struct blkif_request *extra_ring_req;
 584 };
 585
 586 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 587                                      unsigned int len, void *data)
 588 {
 589         struct setup_rw_req *setup = data;
 590         int n, ref;
 591         struct grant *gnt_list_entry;
 592         unsigned int fsect, lsect;
 593         /* Convenient aliases */
 594         unsigned int grant_idx = setup->grant_idx;
 595         struct blkif_request *ring_req = setup->ring_req;
 596         struct blkfront_ring_info *rinfo = setup->rinfo;
 597         /*
 598          * We always use the shadow of the first request to store the list
 599          * of grant associated to the block I/O request. This made the
 600          * completion more easy to handle even if the block I/O request is
 601          * split.
 602          */
 603         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 604
 605         if (unlikely(setup->require_extra_req &&
 606                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 607                 /*
 608                  * We are using the second request, setup grant_idx
 609                  * to be the index of the segment array.
 610                  */
 611                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 612                 ring_req = setup->extra_ring_req;
 613         }
 614
 615         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 616             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 617                 if (setup->segments)
 618                         kunmap_atomic(setup->segments);
 619
 620                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 621                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 622                 shadow->indirect_grants[n] = gnt_list_entry;
 623                 setup->segments = kmap_atomic(gnt_list_entry->page);
 624                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 625         }
 626
 627         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 628         ref = gnt_list_entry->gref;
 629         /*
 630          * All the grants are stored in the shadow of the first
 631          * request. Therefore we have to use the global index.
 632          */
 633         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 634
 635         if (setup->need_copy) {
 636                 void *shared_data;
 637
 638                 shared_data = kmap_atomic(gnt_list_entry->page);
 639                 /*
 640                  * this does not wipe data stored outside the
 641                  * range sg->offset..sg->offset+sg->length.
 642                  * Therefore, blkback *could* see data from
 643                  * previous requests. This is OK as long as
 644                  * persistent grants are shared with just one
 645                  * domain. It may need refactoring if this
 646                  * changes
 647                  */
 648                 memcpy(shared_data + offset,
 649                        setup->bvec_data + setup->bvec_off,
 650                        len);
 651
 652                 kunmap_atomic(shared_data);
 653                 setup->bvec_off += len;
 654         }
 655
 656         fsect = offset >> 9;
 657         lsect = fsect + (len >> 9) - 1;
 658         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 659                 ring_req->u.rw.seg[grant_idx] =
 660                         (struct blkif_request_segment) {
 661                                 .gref       = ref,
 662                                 .first_sect = fsect,
 663                                 .last_sect  = lsect };
 664         } else {
 665                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 666                         (struct blkif_request_segment) {
 667                                 .gref       = ref,
 668                                 .first_sect = fsect,
 669                                 .last_sect  = lsect };
 670         }
 671
 672         (setup->grant_idx)++;
 673 }
 674
 675 static void blkif_setup_extra_req(struct blkif_request *first,
 676                                   struct blkif_request *second)
 677 {
 678         uint16_t nr_segments = first->u.rw.nr_segments;
 679
 680         /*
 681          * The second request is only present when the first request uses
 682          * all its segments. It's always the continuity of the first one.
 683          */
 684         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 685
 686         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 687         second->u.rw.sector_number = first->u.rw.sector_number +
 688                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 689
 690         second->u.rw.handle = first->u.rw.handle;
 691         second->operation = first->operation;
 692 }
 693
 694 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 695 {
 696         struct blkfront_info *info = rinfo->dev_info;
 697         struct blkif_request *ring_req, *extra_ring_req = NULL;
 698         struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
 699         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 700         bool require_extra_req = false;
 701         int i;
 702         struct setup_rw_req setup = {
 703                 .grant_idx = 0,
 704                 .segments = NULL,
 705                 .rinfo = rinfo,
 706                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 707         };
 708
 709         /*
 710          * Used to store if we are able to queue the request by just using
 711          * existing persistent grants, or if we have to get new grants,
 712          * as there are not sufficiently many free.
 713          */
 714         bool new_persistent_gnts = false;
 715         struct scatterlist *sg;
 716         int num_sg, max_grefs, num_grant;
 717
 718         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 719         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 720                 /*
 721                  * If we are using indirect segments we need to account
 722                  * for the indirect grefs used in the request.
 723                  */
 724                 max_grefs += INDIRECT_GREFS(max_grefs);
 725
 726         /* Check if we have enough persistent grants to allocate a requests */
 727         if (rinfo->persistent_gnts_c < max_grefs) {
 728                 new_persistent_gnts = true;
 729
 730                 if (gnttab_alloc_grant_references(
 731                     max_grefs - rinfo->persistent_gnts_c,
 732                     &setup.gref_head) < 0) {
 733                         gnttab_request_free_callback(
 734                                 &rinfo->callback,
 735                                 blkif_restart_queue_callback,
 736                                 rinfo,
 737                                 max_grefs - rinfo->persistent_gnts_c);
 738                         return 1;
 739                 }
 740         }
 741
 742         /* Fill out a communications ring structure. */
 743         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 744         ring_req = &rinfo->shadow[id].req;
 745
 746         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 747         num_grant = 0;
 748         /* Calculate the number of grant used */
 749         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 750                num_grant += gnttab_count_grant(sg->offset, sg->length);
 751
 752         require_extra_req = info->max_indirect_segments == 0 &&
 753                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 754         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 755
 756         rinfo->shadow[id].num_sg = num_sg;
 757         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 758             likely(!require_extra_req)) {
 759                 /*
 760                  * The indirect operation can only be a BLKIF_OP_READ or
 761                  * BLKIF_OP_WRITE
 762                  */
 763                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 764                 ring_req->operation = BLKIF_OP_INDIRECT;
 765                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 766                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 767                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 768                 ring_req->u.indirect.handle = info->handle;
 769                 ring_req->u.indirect.nr_segments = num_grant;
 770         } else {
 771                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 772                 ring_req->u.rw.handle = info->handle;
 773                 ring_req->operation = rq_data_dir(req) ?
 774                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 775                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 776                         /*
 777                          * Ideally we can do an unordered flush-to-disk.
 778                          * In case the backend onlysupports barriers, use that.
 779                          * A barrier request a superset of FUA, so we can
 780                          * implement it the same way.  (It's also a FLUSH+FUA,
 781                          * since it is guaranteed ordered WRT previous writes.)
 782                          */
 783                         if (info->feature_flush && info->feature_fua)
 784                                 ring_req->operation =
 785                                         BLKIF_OP_WRITE_BARRIER;
 786                         else if (info->feature_flush)
 787                                 ring_req->operation =
 788                                         BLKIF_OP_FLUSH_DISKCACHE;
 789                         else
 790                                 ring_req->operation = 0;
 791                 }
 792                 ring_req->u.rw.nr_segments = num_grant;
 793                 if (unlikely(require_extra_req)) {
 794                         extra_id = blkif_ring_get_request(rinfo, req,
 795                                                           &final_extra_ring_req);
 796                         extra_ring_req = &rinfo->shadow[extra_id].req;
 797
 798                         /*
 799                          * Only the first request contains the scatter-gather
 800                          * list.
 801                          */
 802                         rinfo->shadow[extra_id].num_sg = 0;
 803
 804                         blkif_setup_extra_req(ring_req, extra_ring_req);
 805
 806                         /* Link the 2 requests together */
 807                         rinfo->shadow[extra_id].associated_id = id;
 808                         rinfo->shadow[id].associated_id = extra_id;
 809                 }
 810         }
 811
 812         setup.ring_req = ring_req;
 813         setup.id = id;
 814
 815         setup.require_extra_req = require_extra_req;
 816         if (unlikely(require_extra_req))
 817                 setup.extra_ring_req = extra_ring_req;
 818
 819         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 820                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 821
 822                 if (setup.need_copy) {
 823                         setup.bvec_off = sg->offset;
 824                         setup.bvec_data = kmap_atomic(sg_page(sg));
 825                 }
 826
 827                 gnttab_foreach_grant_in_range(sg_page(sg),
 828                                               sg->offset,
 829                                               sg->length,
 830                                               blkif_setup_rw_req_grant,
 831                                               &setup);
 832
 833                 if (setup.need_copy)
 834                         kunmap_atomic(setup.bvec_data);
 835         }
 836         if (setup.segments)
 837                 kunmap_atomic(setup.segments);
 838
 839         /* Copy request(s) to the ring page. */
 840         *final_ring_req = *ring_req;
 841         rinfo->shadow[id].status = REQ_WAITING;
 842         if (unlikely(require_extra_req)) {
 843                 *final_extra_ring_req = *extra_ring_req;
 844                 rinfo->shadow[extra_id].status = REQ_WAITING;
 845         }
 846
 847         if (new_persistent_gnts)
 848                 gnttab_free_grant_references(setup.gref_head);
 849
 850         return 0;
 851 }
 852
 853 /*
 854  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 855  * and writes are handled as expected.
 856  *
 857  * @req: a request struct
 858  */
 859 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 860 {
 861         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 862                 return 1;
 863
 864         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 865                      req_op(req) == REQ_OP_SECURE_ERASE))
 866                 return blkif_queue_discard_req(req, rinfo);
 867         else
 868                 return blkif_queue_rw_req(req, rinfo);
 869 }
 870
 871 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 872 {
 873         int notify;
 874
 875         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 876
 877         if (notify)
 878                 notify_remote_via_irq(rinfo->irq);
 879 }
 880
 881 static inline bool blkif_request_flush_invalid(struct request *req,
 882                                                struct blkfront_info *info)
 883 {
 884         return (blk_rq_is_passthrough(req) ||
 885                 ((req_op(req) == REQ_OP_FLUSH) &&
 886                  !info->feature_flush) ||
 887                 ((req->cmd_flags & REQ_FUA) &&
 888                  !info->feature_fua));
 889 }
 890
 891 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 892                           const struct blk_mq_queue_data *qd)
 893 {
 894         unsigned long flags;
 895         int qid = hctx->queue_num;
 896         struct blkfront_info *info = hctx->queue->queuedata;
 897         struct blkfront_ring_info *rinfo = NULL;
 898
 899         rinfo = get_rinfo(info, qid);
 900         blk_mq_start_request(qd->rq);
 901         spin_lock_irqsave(&rinfo->ring_lock, flags);
 902         if (RING_FULL(&rinfo->ring))
 903                 goto out_busy;
 904
 905         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 906                 goto out_err;
 907
 908         if (blkif_queue_request(qd->rq, rinfo))
 909                 goto out_busy;
 910
 911         flush_requests(rinfo);
 912         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 913         return BLK_STS_OK;
 914
 915 out_err:
 916         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 917         return BLK_STS_IOERR;
 918
 919 out_busy:
 920         blk_mq_stop_hw_queue(hctx);
 921         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 922         return BLK_STS_DEV_RESOURCE;
 923 }
 924
 925 static void blkif_complete_rq(struct request *rq)
 926 {
 927         blk_mq_end_request(rq, blkif_req(rq)->error);
 928 }
 929
 930 static const struct blk_mq_ops blkfront_mq_ops = {
 931         .queue_rq = blkif_queue_rq,
 932         .complete = blkif_complete_rq,
 933 };
 934
 935 static void blkif_set_queue_limits(struct blkfront_info *info)
 936 {
 937         struct request_queue *rq = info->rq;
 938         struct gendisk *gd = info->gd;
 939         unsigned int segments = info->max_indirect_segments ? :
 940                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 941
 942         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 943
 944         if (info->feature_discard) {
 945                 blk_queue_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_max_secure_erase_sectors(rq,
 951                                                            get_capacity(gd));
 952         }
 953
 954         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 955         blk_queue_logical_block_size(rq, info->sector_size);
 956         blk_queue_physical_block_size(rq, info->physical_sector_size);
 957         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 958
 959         /* Each segment in a request is up to an aligned page in size. */
 960         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 961         blk_queue_max_segment_size(rq, PAGE_SIZE);
 962
 963         /* Ensure a merged request will fit in a single I/O ring slot. */
 964         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 965
 966         /* Make sure buffer addresses are sector-aligned. */
 967         blk_queue_dma_alignment(rq, 511);
 968 }
 969
 970 static const char *flush_info(struct blkfront_info *info)
 971 {
 972         if (info->feature_flush && info->feature_fua)
 973                 return "barrier: enabled;";
 974         else if (info->feature_flush)
 975                 return "flush diskcache: enabled;";
 976         else
 977                 return "barrier or flush: disabled;";
 978 }
 979
 980 static void xlvbd_flush(struct blkfront_info *info)
 981 {
 982         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
 983                               info->feature_fua ? true : false);
 984         pr_info("blkfront: %s: %s %s %s %s %s\n",
 985                 info->gd->disk_name, flush_info(info),
 986                 "persistent grants:", info->feature_persistent ?
 987                 "enabled;" : "disabled;", "indirect descriptors:",
 988                 info->max_indirect_segments ? "enabled;" : "disabled;");
 989 }
 990
 991 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 992 {
 993         int major;
 994         major = BLKIF_MAJOR(vdevice);
 995         *minor = BLKIF_MINOR(vdevice);
 996         switch (major) {
 997                 case XEN_IDE0_MAJOR:
 998                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 999                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1000                                 EMULATED_HD_DISK_MINOR_OFFSET;
1001                         break;
1002                 case XEN_IDE1_MAJOR:
1003                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1004                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1005                                 EMULATED_HD_DISK_MINOR_OFFSET;
1006                         break;
1007                 case XEN_SCSI_DISK0_MAJOR:
1008                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1009                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1010                         break;
1011                 case XEN_SCSI_DISK1_MAJOR:
1012                 case XEN_SCSI_DISK2_MAJOR:
1013                 case XEN_SCSI_DISK3_MAJOR:
1014                 case XEN_SCSI_DISK4_MAJOR:
1015                 case XEN_SCSI_DISK5_MAJOR:
1016                 case XEN_SCSI_DISK6_MAJOR:
1017                 case XEN_SCSI_DISK7_MAJOR:
1018                         *offset = (*minor / PARTS_PER_DISK) +
1019                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1020                                 EMULATED_SD_DISK_NAME_OFFSET;
1021                         *minor = *minor +
1022                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1023                                 EMULATED_SD_DISK_MINOR_OFFSET;
1024                         break;
1025                 case XEN_SCSI_DISK8_MAJOR:
1026                 case XEN_SCSI_DISK9_MAJOR:
1027                 case XEN_SCSI_DISK10_MAJOR:
1028                 case XEN_SCSI_DISK11_MAJOR:
1029                 case XEN_SCSI_DISK12_MAJOR:
1030                 case XEN_SCSI_DISK13_MAJOR:
1031                 case XEN_SCSI_DISK14_MAJOR:
1032                 case XEN_SCSI_DISK15_MAJOR:
1033                         *offset = (*minor / PARTS_PER_DISK) +
1034                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1035                                 EMULATED_SD_DISK_NAME_OFFSET;
1036                         *minor = *minor +
1037                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1038                                 EMULATED_SD_DISK_MINOR_OFFSET;
1039                         break;
1040                 case XENVBD_MAJOR:
1041                         *offset = *minor / PARTS_PER_DISK;
1042                         break;
1043                 default:
1044                         printk(KERN_WARNING "blkfront: your disk configuration is "
1045                                         "incorrect, please use an xvd device instead\n");
1046                         return -ENODEV;
1047         }
1048         return 0;
1049 }
1050
1051 static char *encode_disk_name(char *ptr, unsigned int n)
1052 {
1053         if (n >= 26)
1054                 ptr = encode_disk_name(ptr, n / 26 - 1);
1055         *ptr = 'a' + n % 26;
1056         return ptr + 1;
1057 }
1058
1059 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1060                 struct blkfront_info *info, 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 (info->vdisk_info & VDISK_READONLY)
1160                 set_disk_ro(gd, 1);
1161         if (info->vdisk_info & VDISK_REMOVABLE)
1162                 gd->flags |= GENHD_FL_REMOVABLE;
1163
1164         return 0;
1165
1166 out_free_tag_set:
1167         blk_mq_free_tag_set(&info->tag_set);
1168 out_release_minors:
1169         xlbd_release_minors(minor, nr_minors);
1170         return err;
1171 }
1172
1173 /* Already hold rinfo->ring_lock. */
1174 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1175 {
1176         if (!RING_FULL(&rinfo->ring))
1177                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1178 }
1179
1180 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1181 {
1182         unsigned long flags;
1183
1184         spin_lock_irqsave(&rinfo->ring_lock, flags);
1185         kick_pending_request_queues_locked(rinfo);
1186         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1187 }
1188
1189 static void blkif_restart_queue(struct work_struct *work)
1190 {
1191         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1192
1193         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1194                 kick_pending_request_queues(rinfo);
1195 }
1196
1197 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1198 {
1199         struct grant *persistent_gnt, *n;
1200         struct blkfront_info *info = rinfo->dev_info;
1201         int i, j, segs;
1202
1203         /*
1204          * Remove indirect pages, this only happens when using indirect
1205          * descriptors but not persistent grants
1206          */
1207         if (!list_empty(&rinfo->indirect_pages)) {
1208                 struct page *indirect_page, *n;
1209
1210                 BUG_ON(info->feature_persistent);
1211                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1212                         list_del(&indirect_page->lru);
1213                         __free_page(indirect_page);
1214                 }
1215         }
1216
1217         /* Remove all persistent grants. */
1218         if (!list_empty(&rinfo->grants)) {
1219                 list_for_each_entry_safe(persistent_gnt, n,
1220                                          &rinfo->grants, node) {
1221                         list_del(&persistent_gnt->node);
1222                         if (persistent_gnt->gref != INVALID_GRANT_REF) {
1223                                 gnttab_end_foreign_access(persistent_gnt->gref,
1224                                                           NULL);
1225                                 rinfo->persistent_gnts_c--;
1226                         }
1227                         if (info->feature_persistent)
1228                                 __free_page(persistent_gnt->page);
1229                         kfree(persistent_gnt);
1230                 }
1231         }
1232         BUG_ON(rinfo->persistent_gnts_c != 0);
1233
1234         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1235                 /*
1236                  * Clear persistent grants present in requests already
1237                  * on the shared ring
1238                  */
1239                 if (!rinfo->shadow[i].request)
1240                         goto free_shadow;
1241
1242                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1243                        rinfo->shadow[i].req.u.indirect.nr_segments :
1244                        rinfo->shadow[i].req.u.rw.nr_segments;
1245                 for (j = 0; j < segs; j++) {
1246                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1247                         gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1248                         if (info->feature_persistent)
1249                                 __free_page(persistent_gnt->page);
1250                         kfree(persistent_gnt);
1251                 }
1252
1253                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1254                         /*
1255                          * If this is not an indirect operation don't try to
1256                          * free indirect segments
1257                          */
1258                         goto free_shadow;
1259
1260                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1261                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1262                         gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1263                         __free_page(persistent_gnt->page);
1264                         kfree(persistent_gnt);
1265                 }
1266
1267 free_shadow:
1268                 kvfree(rinfo->shadow[i].grants_used);
1269                 rinfo->shadow[i].grants_used = NULL;
1270                 kvfree(rinfo->shadow[i].indirect_grants);
1271                 rinfo->shadow[i].indirect_grants = NULL;
1272                 kvfree(rinfo->shadow[i].sg);
1273                 rinfo->shadow[i].sg = NULL;
1274         }
1275
1276         /* No more gnttab callback work. */
1277         gnttab_cancel_free_callback(&rinfo->callback);
1278
1279         /* Flush gnttab callback work. Must be done with no locks held. */
1280         flush_work(&rinfo->work);
1281
1282         /* Free resources associated with old device channel. */
1283         xenbus_teardown_ring((void **)&rinfo->ring.sring, info->nr_ring_pages,
1284                              rinfo->ring_ref);
1285
1286         if (rinfo->irq)
1287                 unbind_from_irqhandler(rinfo->irq, rinfo);
1288         rinfo->evtchn = rinfo->irq = 0;
1289 }
1290
1291 static void blkif_free(struct blkfront_info *info, int suspend)
1292 {
1293         unsigned int i;
1294         struct blkfront_ring_info *rinfo;
1295
1296         /* Prevent new requests being issued until we fix things up. */
1297         info->connected = suspend ?
1298                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1299         /* No more blkif_request(). */
1300         if (info->rq)
1301                 blk_mq_stop_hw_queues(info->rq);
1302
1303         for_each_rinfo(info, rinfo, i)
1304                 blkif_free_ring(rinfo);
1305
1306         kvfree(info->rinfo);
1307         info->rinfo = NULL;
1308         info->nr_rings = 0;
1309 }
1310
1311 struct copy_from_grant {
1312         const struct blk_shadow *s;
1313         unsigned int grant_idx;
1314         unsigned int bvec_offset;
1315         char *bvec_data;
1316 };
1317
1318 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1319                                   unsigned int len, void *data)
1320 {
1321         struct copy_from_grant *info = data;
1322         char *shared_data;
1323         /* Convenient aliases */
1324         const struct blk_shadow *s = info->s;
1325
1326         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1327
1328         memcpy(info->bvec_data + info->bvec_offset,
1329                shared_data + offset, len);
1330
1331         info->bvec_offset += len;
1332         info->grant_idx++;
1333
1334         kunmap_atomic(shared_data);
1335 }
1336
1337 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1338 {
1339         switch (rsp)
1340         {
1341         case BLKIF_RSP_OKAY:
1342                 return REQ_DONE;
1343         case BLKIF_RSP_EOPNOTSUPP:
1344                 return REQ_EOPNOTSUPP;
1345         case BLKIF_RSP_ERROR:
1346         default:
1347                 return REQ_ERROR;
1348         }
1349 }
1350
1351 /*
1352  * Get the final status of the block request based on two ring response
1353  */
1354 static int blkif_get_final_status(enum blk_req_status s1,
1355                                   enum blk_req_status s2)
1356 {
1357         BUG_ON(s1 < REQ_DONE);
1358         BUG_ON(s2 < REQ_DONE);
1359
1360         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1361                 return BLKIF_RSP_ERROR;
1362         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1363                 return BLKIF_RSP_EOPNOTSUPP;
1364         return BLKIF_RSP_OKAY;
1365 }
1366
1367 /*
1368  * Return values:
1369  *  1 response processed.
1370  *  0 missing further responses.
1371  * -1 error while processing.
1372  */
1373 static int blkif_completion(unsigned long *id,
1374                             struct blkfront_ring_info *rinfo,
1375                             struct blkif_response *bret)
1376 {
1377         int i = 0;
1378         struct scatterlist *sg;
1379         int num_sg, num_grant;
1380         struct blkfront_info *info = rinfo->dev_info;
1381         struct blk_shadow *s = &rinfo->shadow[*id];
1382         struct copy_from_grant data = {
1383                 .grant_idx = 0,
1384         };
1385
1386         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1387                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1388
1389         /* The I/O request may be split in two. */
1390         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1391                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1392
1393                 /* Keep the status of the current response in shadow. */
1394                 s->status = blkif_rsp_to_req_status(bret->status);
1395
1396                 /* Wait the second response if not yet here. */
1397                 if (s2->status < REQ_DONE)
1398                         return 0;
1399
1400                 bret->status = blkif_get_final_status(s->status,
1401                                                       s2->status);
1402
1403                 /*
1404                  * All the grants is stored in the first shadow in order
1405                  * to make the completion code simpler.
1406                  */
1407                 num_grant += s2->req.u.rw.nr_segments;
1408
1409                 /*
1410                  * The two responses may not come in order. Only the
1411                  * first request will store the scatter-gather list.
1412                  */
1413                 if (s2->num_sg != 0) {
1414                         /* Update "id" with the ID of the first response. */
1415                         *id = s->associated_id;
1416                         s = s2;
1417                 }
1418
1419                 /*
1420                  * We don't need anymore the second request, so recycling
1421                  * it now.
1422                  */
1423                 if (add_id_to_freelist(rinfo, s->associated_id))
1424                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1425                              info->gd->disk_name, s->associated_id);
1426         }
1427
1428         data.s = s;
1429         num_sg = s->num_sg;
1430
1431         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1432                 for_each_sg(s->sg, sg, num_sg, i) {
1433                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1434
1435                         data.bvec_offset = sg->offset;
1436                         data.bvec_data = kmap_atomic(sg_page(sg));
1437
1438                         gnttab_foreach_grant_in_range(sg_page(sg),
1439                                                       sg->offset,
1440                                                       sg->length,
1441                                                       blkif_copy_from_grant,
1442                                                       &data);
1443
1444                         kunmap_atomic(data.bvec_data);
1445                 }
1446         }
1447         /* Add the persistent grant into the list of free grants */
1448         for (i = 0; i < num_grant; i++) {
1449                 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1450                         /*
1451                          * If the grant is still mapped by the backend (the
1452                          * backend has chosen to make this grant persistent)
1453                          * we add it at the head of the list, so it will be
1454                          * reused first.
1455                          */
1456                         if (!info->feature_persistent) {
1457                                 pr_alert("backed has not unmapped grant: %u\n",
1458                                          s->grants_used[i]->gref);
1459                                 return -1;
1460                         }
1461                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1462                         rinfo->persistent_gnts_c++;
1463                 } else {
1464                         /*
1465                          * If the grant is not mapped by the backend we add it
1466                          * to the tail of the list, so it will not be picked
1467                          * again unless we run out of persistent grants.
1468                          */
1469                         s->grants_used[i]->gref = INVALID_GRANT_REF;
1470                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1471                 }
1472         }
1473         if (s->req.operation == BLKIF_OP_INDIRECT) {
1474                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1475                         if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1476                                 if (!info->feature_persistent) {
1477                                         pr_alert("backed has not unmapped grant: %u\n",
1478                                                  s->indirect_grants[i]->gref);
1479                                         return -1;
1480                                 }
1481                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1482                                 rinfo->persistent_gnts_c++;
1483                         } else {
1484                                 struct page *indirect_page;
1485
1486                                 /*
1487                                  * Add the used indirect page back to the list of
1488                                  * available pages for indirect grefs.
1489                                  */
1490                                 if (!info->feature_persistent) {
1491                                         indirect_page = s->indirect_grants[i]->page;
1492                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1493                                 }
1494                                 s->indirect_grants[i]->gref = INVALID_GRANT_REF;
1495                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1496                         }
1497                 }
1498         }
1499
1500         return 1;
1501 }
1502
1503 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1504 {
1505         struct request *req;
1506         struct blkif_response bret;
1507         RING_IDX i, rp;
1508         unsigned long flags;
1509         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1510         struct blkfront_info *info = rinfo->dev_info;
1511         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1512
1513         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1514                 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1515                 return IRQ_HANDLED;
1516         }
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                 eoiflag = 0;
1533
1534                 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1535                 id = bret.id;
1536
1537                 /*
1538                  * The backend has messed up and given us an id that we would
1539                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1540                  * look in get_id_from_freelist.
1541                  */
1542                 if (id >= BLK_RING_SIZE(info)) {
1543                         pr_alert("%s: response has incorrect id (%ld)\n",
1544                                  info->gd->disk_name, id);
1545                         goto err;
1546                 }
1547                 if (rinfo->shadow[id].status != REQ_WAITING) {
1548                         pr_alert("%s: response references no pending request\n",
1549                                  info->gd->disk_name);
1550                         goto err;
1551                 }
1552
1553                 rinfo->shadow[id].status = REQ_PROCESSING;
1554                 req  = rinfo->shadow[id].request;
1555
1556                 op = rinfo->shadow[id].req.operation;
1557                 if (op == BLKIF_OP_INDIRECT)
1558                         op = rinfo->shadow[id].req.u.indirect.indirect_op;
1559                 if (bret.operation != op) {
1560                         pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1561                                  info->gd->disk_name, bret.operation, op);
1562                         goto err;
1563                 }
1564
1565                 if (bret.operation != BLKIF_OP_DISCARD) {
1566                         int ret;
1567
1568                         /*
1569                          * We may need to wait for an extra response if the
1570                          * I/O request is split in 2
1571                          */
1572                         ret = blkif_completion(&id, rinfo, &bret);
1573                         if (!ret)
1574                                 continue;
1575                         if (unlikely(ret < 0))
1576                                 goto err;
1577                 }
1578
1579                 if (add_id_to_freelist(rinfo, id)) {
1580                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1581                              info->gd->disk_name, op_name(bret.operation), id);
1582                         continue;
1583                 }
1584
1585                 if (bret.status == BLKIF_RSP_OKAY)
1586                         blkif_req(req)->error = BLK_STS_OK;
1587                 else
1588                         blkif_req(req)->error = BLK_STS_IOERR;
1589
1590                 switch (bret.operation) {
1591                 case BLKIF_OP_DISCARD:
1592                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1593                                 struct request_queue *rq = info->rq;
1594
1595                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1596                                            info->gd->disk_name, op_name(bret.operation));
1597                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1598                                 info->feature_discard = 0;
1599                                 info->feature_secdiscard = 0;
1600                                 blk_queue_max_discard_sectors(rq, 0);
1601                                 blk_queue_max_secure_erase_sectors(rq, 0);
1602                         }
1603                         break;
1604                 case BLKIF_OP_FLUSH_DISKCACHE:
1605                 case BLKIF_OP_WRITE_BARRIER:
1606                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1607                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1608                                        info->gd->disk_name, op_name(bret.operation));
1609                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1610                         }
1611                         if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1612                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1613                                 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1614                                        info->gd->disk_name, op_name(bret.operation));
1615                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1616                         }
1617                         if (unlikely(blkif_req(req)->error)) {
1618                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1619                                         blkif_req(req)->error = BLK_STS_OK;
1620                                 info->feature_fua = 0;
1621                                 info->feature_flush = 0;
1622                                 xlvbd_flush(info);
1623                         }
1624                         fallthrough;
1625                 case BLKIF_OP_READ:
1626                 case BLKIF_OP_WRITE:
1627                         if (unlikely(bret.status != BLKIF_RSP_OKAY))
1628                                 dev_dbg_ratelimited(&info->xbdev->dev,
1629                                         "Bad return from blkdev data request: %#x\n",
1630                                         bret.status);
1631
1632                         break;
1633                 default:
1634                         BUG();
1635                 }
1636
1637                 if (likely(!blk_should_fake_timeout(req->q)))
1638                         blk_mq_complete_request(req);
1639         }
1640
1641         rinfo->ring.rsp_cons = i;
1642
1643         if (i != rinfo->ring.req_prod_pvt) {
1644                 int more_to_do;
1645                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1646                 if (more_to_do)
1647                         goto again;
1648         } else
1649                 rinfo->ring.sring->rsp_event = i + 1;
1650
1651         kick_pending_request_queues_locked(rinfo);
1652
1653         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1654
1655         xen_irq_lateeoi(irq, eoiflag);
1656
1657         return IRQ_HANDLED;
1658
1659  err:
1660         info->connected = BLKIF_STATE_ERROR;
1661
1662         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1663
1664         /* No EOI in order to avoid further interrupts. */
1665
1666         pr_alert("%s disabled for further use\n", info->gd->disk_name);
1667         return IRQ_HANDLED;
1668 }
1669
1670
1671 static int setup_blkring(struct xenbus_device *dev,
1672                          struct blkfront_ring_info *rinfo)
1673 {
1674         struct blkif_sring *sring;
1675         int err;
1676         struct blkfront_info *info = rinfo->dev_info;
1677         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1678
1679         err = xenbus_setup_ring(dev, GFP_NOIO, (void **)&sring,
1680                                 info->nr_ring_pages, rinfo->ring_ref);
1681         if (err)
1682                 goto fail;
1683
1684         XEN_FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1685
1686         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1687         if (err)
1688                 goto fail;
1689
1690         err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1691                                                 0, "blkif", rinfo);
1692         if (err <= 0) {
1693                 xenbus_dev_fatal(dev, err,
1694                                  "bind_evtchn_to_irqhandler failed");
1695                 goto fail;
1696         }
1697         rinfo->irq = err;
1698
1699         return 0;
1700 fail:
1701         blkif_free(info, 0);
1702         return err;
1703 }
1704
1705 /*
1706  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1707  * ring buffer may have multi pages depending on ->nr_ring_pages.
1708  */
1709 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1710                                 struct blkfront_ring_info *rinfo, const char *dir)
1711 {
1712         int err;
1713         unsigned int i;
1714         const char *message = NULL;
1715         struct blkfront_info *info = rinfo->dev_info;
1716
1717         if (info->nr_ring_pages == 1) {
1718                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1719                 if (err) {
1720                         message = "writing ring-ref";
1721                         goto abort_transaction;
1722                 }
1723         } else {
1724                 for (i = 0; i < info->nr_ring_pages; i++) {
1725                         char ring_ref_name[RINGREF_NAME_LEN];
1726
1727                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1728                         err = xenbus_printf(xbt, dir, ring_ref_name,
1729                                             "%u", rinfo->ring_ref[i]);
1730                         if (err) {
1731                                 message = "writing ring-ref";
1732                                 goto abort_transaction;
1733                         }
1734                 }
1735         }
1736
1737         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1738         if (err) {
1739                 message = "writing event-channel";
1740                 goto abort_transaction;
1741         }
1742
1743         return 0;
1744
1745 abort_transaction:
1746         xenbus_transaction_end(xbt, 1);
1747         if (message)
1748                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1749
1750         return err;
1751 }
1752
1753 /* Common code used when first setting up, and when resuming. */
1754 static int talk_to_blkback(struct xenbus_device *dev,
1755                            struct blkfront_info *info)
1756 {
1757         const char *message = NULL;
1758         struct xenbus_transaction xbt;
1759         int err;
1760         unsigned int i, max_page_order;
1761         unsigned int ring_page_order;
1762         struct blkfront_ring_info *rinfo;
1763
1764         if (!info)
1765                 return -ENODEV;
1766
1767         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1768                                               "max-ring-page-order", 0);
1769         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1770         info->nr_ring_pages = 1 << ring_page_order;
1771
1772         err = negotiate_mq(info);
1773         if (err)
1774                 goto destroy_blkring;
1775
1776         for_each_rinfo(info, rinfo, i) {
1777                 /* Create shared ring, alloc event channel. */
1778                 err = setup_blkring(dev, rinfo);
1779                 if (err)
1780                         goto destroy_blkring;
1781         }
1782
1783 again:
1784         err = xenbus_transaction_start(&xbt);
1785         if (err) {
1786                 xenbus_dev_fatal(dev, err, "starting transaction");
1787                 goto destroy_blkring;
1788         }
1789
1790         if (info->nr_ring_pages > 1) {
1791                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1792                                     ring_page_order);
1793                 if (err) {
1794                         message = "writing ring-page-order";
1795                         goto abort_transaction;
1796                 }
1797         }
1798
1799         /* We already got the number of queues/rings in _probe */
1800         if (info->nr_rings == 1) {
1801                 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1802                 if (err)
1803                         goto destroy_blkring;
1804         } else {
1805                 char *path;
1806                 size_t pathsize;
1807
1808                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1809                                     info->nr_rings);
1810                 if (err) {
1811                         message = "writing multi-queue-num-queues";
1812                         goto abort_transaction;
1813                 }
1814
1815                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1816                 path = kmalloc(pathsize, GFP_KERNEL);
1817                 if (!path) {
1818                         err = -ENOMEM;
1819                         message = "ENOMEM while writing ring references";
1820                         goto abort_transaction;
1821                 }
1822
1823                 for_each_rinfo(info, rinfo, i) {
1824                         memset(path, 0, pathsize);
1825                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1826                         err = write_per_ring_nodes(xbt, rinfo, path);
1827                         if (err) {
1828                                 kfree(path);
1829                                 goto destroy_blkring;
1830                         }
1831                 }
1832                 kfree(path);
1833         }
1834         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1835                             XEN_IO_PROTO_ABI_NATIVE);
1836         if (err) {
1837                 message = "writing protocol";
1838                 goto abort_transaction;
1839         }
1840         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1841                         info->feature_persistent);
1842         if (err)
1843                 dev_warn(&dev->dev,
1844                          "writing persistent grants feature to xenbus");
1845
1846         err = xenbus_transaction_end(xbt, 0);
1847         if (err) {
1848                 if (err == -EAGAIN)
1849                         goto again;
1850                 xenbus_dev_fatal(dev, err, "completing transaction");
1851                 goto destroy_blkring;
1852         }
1853
1854         for_each_rinfo(info, rinfo, i) {
1855                 unsigned int j;
1856
1857                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1858                         rinfo->shadow[j].req.u.rw.id = j + 1;
1859                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1860         }
1861         xenbus_switch_state(dev, XenbusStateInitialised);
1862
1863         return 0;
1864
1865  abort_transaction:
1866         xenbus_transaction_end(xbt, 1);
1867         if (message)
1868                 xenbus_dev_fatal(dev, err, "%s", message);
1869  destroy_blkring:
1870         blkif_free(info, 0);
1871         return err;
1872 }
1873
1874 static int negotiate_mq(struct blkfront_info *info)
1875 {
1876         unsigned int backend_max_queues;
1877         unsigned int i;
1878         struct blkfront_ring_info *rinfo;
1879
1880         BUG_ON(info->nr_rings);
1881
1882         /* Check if backend supports multiple queues. */
1883         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1884                                                   "multi-queue-max-queues", 1);
1885         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1886         /* We need at least one ring. */
1887         if (!info->nr_rings)
1888                 info->nr_rings = 1;
1889
1890         info->rinfo_size = struct_size(info->rinfo, shadow,
1891                                        BLK_RING_SIZE(info));
1892         info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1893         if (!info->rinfo) {
1894                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1895                 info->nr_rings = 0;
1896                 return -ENOMEM;
1897         }
1898
1899         for_each_rinfo(info, rinfo, i) {
1900                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1901                 INIT_LIST_HEAD(&rinfo->grants);
1902                 rinfo->dev_info = info;
1903                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1904                 spin_lock_init(&rinfo->ring_lock);
1905         }
1906         return 0;
1907 }
1908
1909 /* Enable the persistent grants feature. */
1910 static bool feature_persistent = true;
1911 module_param(feature_persistent, bool, 0644);
1912 MODULE_PARM_DESC(feature_persistent,
1913                 "Enables the persistent grants feature");
1914
1915 /*
1916  * Entry point to this code when a new device is created.  Allocate the basic
1917  * structures and the ring buffer for communication with the backend, and
1918  * inform the backend of the appropriate details for those.  Switch to
1919  * Initialised state.
1920  */
1921 static int blkfront_probe(struct xenbus_device *dev,
1922                           const struct xenbus_device_id *id)
1923 {
1924         int err, vdevice;
1925         struct blkfront_info *info;
1926
1927         /* FIXME: Use dynamic device id if this is not set. */
1928         err = xenbus_scanf(XBT_NIL, dev->nodename,
1929                            "virtual-device", "%i", &vdevice);
1930         if (err != 1) {
1931                 /* go looking in the extended area instead */
1932                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1933                                    "%i", &vdevice);
1934                 if (err != 1) {
1935                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1936                         return err;
1937                 }
1938         }
1939
1940         if (xen_hvm_domain()) {
1941                 char *type;
1942                 int len;
1943                 /* no unplug has been done: do not hook devices != xen vbds */
1944                 if (xen_has_pv_and_legacy_disk_devices()) {
1945                         int major;
1946
1947                         if (!VDEV_IS_EXTENDED(vdevice))
1948                                 major = BLKIF_MAJOR(vdevice);
1949                         else
1950                                 major = XENVBD_MAJOR;
1951
1952                         if (major != XENVBD_MAJOR) {
1953                                 printk(KERN_INFO
1954                                                 "%s: HVM does not support vbd %d as xen block device\n",
1955                                                 __func__, vdevice);
1956                                 return -ENODEV;
1957                         }
1958                 }
1959                 /* do not create a PV cdrom device if we are an HVM guest */
1960                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1961                 if (IS_ERR(type))
1962                         return -ENODEV;
1963                 if (strncmp(type, "cdrom", 5) == 0) {
1964                         kfree(type);
1965                         return -ENODEV;
1966                 }
1967                 kfree(type);
1968         }
1969         info = kzalloc(sizeof(*info), GFP_KERNEL);
1970         if (!info) {
1971                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1972                 return -ENOMEM;
1973         }
1974
1975         info->xbdev = dev;
1976
1977         mutex_init(&info->mutex);
1978         info->vdevice = vdevice;
1979         info->connected = BLKIF_STATE_DISCONNECTED;
1980
1981         info->feature_persistent = feature_persistent;
1982
1983         /* Front end dir is a number, which is used as the id. */
1984         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1985         dev_set_drvdata(&dev->dev, info);
1986
1987         mutex_lock(&blkfront_mutex);
1988         list_add(&info->info_list, &info_list);
1989         mutex_unlock(&blkfront_mutex);
1990
1991         return 0;
1992 }
1993
1994 static int blkif_recover(struct blkfront_info *info)
1995 {
1996         unsigned int r_index;
1997         struct request *req, *n;
1998         int rc;
1999         struct bio *bio;
2000         unsigned int segs;
2001         struct blkfront_ring_info *rinfo;
2002
2003         blkfront_gather_backend_features(info);
2004         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2005         blkif_set_queue_limits(info);
2006         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2007         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2008
2009         for_each_rinfo(info, rinfo, r_index) {
2010                 rc = blkfront_setup_indirect(rinfo);
2011                 if (rc)
2012                         return rc;
2013         }
2014         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2015
2016         /* Now safe for us to use the shared ring */
2017         info->connected = BLKIF_STATE_CONNECTED;
2018
2019         for_each_rinfo(info, rinfo, r_index) {
2020                 /* Kick any other new requests queued since we resumed */
2021                 kick_pending_request_queues(rinfo);
2022         }
2023
2024         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2025                 /* Requeue pending requests (flush or discard) */
2026                 list_del_init(&req->queuelist);
2027                 BUG_ON(req->nr_phys_segments > segs);
2028                 blk_mq_requeue_request(req, false);
2029         }
2030         blk_mq_start_stopped_hw_queues(info->rq, true);
2031         blk_mq_kick_requeue_list(info->rq);
2032
2033         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2034                 /* Traverse the list of pending bios and re-queue them */
2035                 submit_bio(bio);
2036         }
2037
2038         return 0;
2039 }
2040
2041 /*
2042  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2043  * driver restart.  We tear down our blkif structure and recreate it, but
2044  * leave the device-layer structures intact so that this is transparent to the
2045  * rest of the kernel.
2046  */
2047 static int blkfront_resume(struct xenbus_device *dev)
2048 {
2049         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2050         int err = 0;
2051         unsigned int i, j;
2052         struct blkfront_ring_info *rinfo;
2053
2054         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2055
2056         bio_list_init(&info->bio_list);
2057         INIT_LIST_HEAD(&info->requests);
2058         for_each_rinfo(info, rinfo, i) {
2059                 struct bio_list merge_bio;
2060                 struct blk_shadow *shadow = rinfo->shadow;
2061
2062                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2063                         /* Not in use? */
2064                         if (!shadow[j].request)
2065                                 continue;
2066
2067                         /*
2068                          * Get the bios in the request so we can re-queue them.
2069                          */
2070                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2071                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2072                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2073                             shadow[j].request->cmd_flags & REQ_FUA) {
2074                                 /*
2075                                  * Flush operations don't contain bios, so
2076                                  * we need to requeue the whole request
2077                                  *
2078                                  * XXX: but this doesn't make any sense for a
2079                                  * write with the FUA flag set..
2080                                  */
2081                                 list_add(&shadow[j].request->queuelist, &info->requests);
2082                                 continue;
2083                         }
2084                         merge_bio.head = shadow[j].request->bio;
2085                         merge_bio.tail = shadow[j].request->biotail;
2086                         bio_list_merge(&info->bio_list, &merge_bio);
2087                         shadow[j].request->bio = NULL;
2088                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2089                 }
2090         }
2091
2092         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2093
2094         err = talk_to_blkback(dev, info);
2095         if (!err)
2096                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2097
2098         /*
2099          * We have to wait for the backend to switch to
2100          * connected state, since we want to read which
2101          * features it supports.
2102          */
2103
2104         return err;
2105 }
2106
2107 static void blkfront_closing(struct blkfront_info *info)
2108 {
2109         struct xenbus_device *xbdev = info->xbdev;
2110         struct blkfront_ring_info *rinfo;
2111         unsigned int i;
2112
2113         if (xbdev->state == XenbusStateClosing)
2114                 return;
2115
2116         /* No more blkif_request(). */
2117         if (info->rq && info->gd) {
2118                 blk_mq_stop_hw_queues(info->rq);
2119                 blk_mark_disk_dead(info->gd);
2120                 set_capacity(info->gd, 0);
2121         }
2122
2123         for_each_rinfo(info, rinfo, i) {
2124                 /* No more gnttab callback work. */
2125                 gnttab_cancel_free_callback(&rinfo->callback);
2126
2127                 /* Flush gnttab callback work. Must be done with no locks held. */
2128                 flush_work(&rinfo->work);
2129         }
2130
2131         xenbus_frontend_closed(xbdev);
2132 }
2133
2134 static void blkfront_setup_discard(struct blkfront_info *info)
2135 {
2136         info->feature_discard = 1;
2137         info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2138                                                          "discard-granularity",
2139                                                          0);
2140         info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2141                                                        "discard-alignment", 0);
2142         info->feature_secdiscard =
2143                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2144                                        0);
2145 }
2146
2147 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2148 {
2149         unsigned int psegs, grants, memflags;
2150         int err, i;
2151         struct blkfront_info *info = rinfo->dev_info;
2152
2153         memflags = memalloc_noio_save();
2154
2155         if (info->max_indirect_segments == 0) {
2156                 if (!HAS_EXTRA_REQ)
2157                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2158                 else {
2159                         /*
2160                          * When an extra req is required, the maximum
2161                          * grants supported is related to the size of the
2162                          * Linux block segment.
2163                          */
2164                         grants = GRANTS_PER_PSEG;
2165                 }
2166         }
2167         else
2168                 grants = info->max_indirect_segments;
2169         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2170
2171         err = fill_grant_buffer(rinfo,
2172                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2173         if (err)
2174                 goto out_of_memory;
2175
2176         if (!info->feature_persistent && info->max_indirect_segments) {
2177                 /*
2178                  * We are using indirect descriptors but not persistent
2179                  * grants, we need to allocate a set of pages that can be
2180                  * used for mapping indirect grefs
2181                  */
2182                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2183
2184                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2185                 for (i = 0; i < num; i++) {
2186                         struct page *indirect_page = alloc_page(GFP_KERNEL);
2187                         if (!indirect_page)
2188                                 goto out_of_memory;
2189                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2190                 }
2191         }
2192
2193         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2194                 rinfo->shadow[i].grants_used =
2195                         kvcalloc(grants,
2196                                  sizeof(rinfo->shadow[i].grants_used[0]),
2197                                  GFP_KERNEL);
2198                 rinfo->shadow[i].sg = kvcalloc(psegs,
2199                                                sizeof(rinfo->shadow[i].sg[0]),
2200                                                GFP_KERNEL);
2201                 if (info->max_indirect_segments)
2202                         rinfo->shadow[i].indirect_grants =
2203                                 kvcalloc(INDIRECT_GREFS(grants),
2204                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2205                                          GFP_KERNEL);
2206                 if ((rinfo->shadow[i].grants_used == NULL) ||
2207                         (rinfo->shadow[i].sg == NULL) ||
2208                      (info->max_indirect_segments &&
2209                      (rinfo->shadow[i].indirect_grants == NULL)))
2210                         goto out_of_memory;
2211                 sg_init_table(rinfo->shadow[i].sg, psegs);
2212         }
2213
2214         memalloc_noio_restore(memflags);
2215
2216         return 0;
2217
2218 out_of_memory:
2219         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2220                 kvfree(rinfo->shadow[i].grants_used);
2221                 rinfo->shadow[i].grants_used = NULL;
2222                 kvfree(rinfo->shadow[i].sg);
2223                 rinfo->shadow[i].sg = NULL;
2224                 kvfree(rinfo->shadow[i].indirect_grants);
2225                 rinfo->shadow[i].indirect_grants = NULL;
2226         }
2227         if (!list_empty(&rinfo->indirect_pages)) {
2228                 struct page *indirect_page, *n;
2229                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2230                         list_del(&indirect_page->lru);
2231                         __free_page(indirect_page);
2232                 }
2233         }
2234
2235         memalloc_noio_restore(memflags);
2236
2237         return -ENOMEM;
2238 }
2239
2240 /*
2241  * Gather all backend feature-*
2242  */
2243 static void blkfront_gather_backend_features(struct blkfront_info *info)
2244 {
2245         unsigned int indirect_segments;
2246
2247         info->feature_flush = 0;
2248         info->feature_fua = 0;
2249
2250         /*
2251          * If there's no "feature-barrier" defined, then it means
2252          * we're dealing with a very old backend which writes
2253          * synchronously; nothing to do.
2254          *
2255          * If there are barriers, then we use flush.
2256          */
2257         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2258                 info->feature_flush = 1;
2259                 info->feature_fua = 1;
2260         }
2261
2262         /*
2263          * And if there is "feature-flush-cache" use that above
2264          * barriers.
2265          */
2266         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2267                                  0)) {
2268                 info->feature_flush = 1;
2269                 info->feature_fua = 0;
2270         }
2271
2272         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2273                 blkfront_setup_discard(info);
2274
2275         if (info->feature_persistent)
2276                 info->feature_persistent =
2277                         !!xenbus_read_unsigned(info->xbdev->otherend,
2278                                                "feature-persistent", 0);
2279
2280         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2281                                         "feature-max-indirect-segments", 0);
2282         if (indirect_segments > xen_blkif_max_segments)
2283                 indirect_segments = xen_blkif_max_segments;
2284         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2285                 indirect_segments = 0;
2286         info->max_indirect_segments = indirect_segments;
2287
2288         if (info->feature_persistent) {
2289                 mutex_lock(&blkfront_mutex);
2290                 schedule_delayed_work(&blkfront_work, HZ * 10);
2291                 mutex_unlock(&blkfront_mutex);
2292         }
2293 }
2294
2295 /*
2296  * Invoked when the backend is finally 'ready' (and has told produced
2297  * the details about the physical device - #sectors, size, etc).
2298  */
2299 static void blkfront_connect(struct blkfront_info *info)
2300 {
2301         unsigned long long sectors;
2302         unsigned long sector_size;
2303         unsigned int physical_sector_size;
2304         int err, i;
2305         struct blkfront_ring_info *rinfo;
2306
2307         switch (info->connected) {
2308         case BLKIF_STATE_CONNECTED:
2309                 /*
2310                  * Potentially, the back-end may be signalling
2311                  * a capacity change; update the capacity.
2312                  */
2313                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2314                                    "sectors", "%Lu", &sectors);
2315                 if (XENBUS_EXIST_ERR(err))
2316                         return;
2317                 printk(KERN_INFO "Setting capacity to %Lu\n",
2318                        sectors);
2319                 set_capacity_and_notify(info->gd, sectors);
2320
2321                 return;
2322         case BLKIF_STATE_SUSPENDED:
2323                 /*
2324                  * If we are recovering from suspension, we need to wait
2325                  * for the backend to announce it's features before
2326                  * reconnecting, at least we need to know if the backend
2327                  * supports indirect descriptors, and how many.
2328                  */
2329                 blkif_recover(info);
2330                 return;
2331
2332         default:
2333                 break;
2334         }
2335
2336         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2337                 __func__, info->xbdev->otherend);
2338
2339         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2340                             "sectors", "%llu", &sectors,
2341                             "info", "%u", &info->vdisk_info,
2342                             "sector-size", "%lu", &sector_size,
2343                             NULL);
2344         if (err) {
2345                 xenbus_dev_fatal(info->xbdev, err,
2346                                  "reading backend fields at %s",
2347                                  info->xbdev->otherend);
2348                 return;
2349         }
2350
2351         /*
2352          * physical-sector-size is a newer field, so old backends may not
2353          * provide this. Assume physical sector size to be the same as
2354          * sector_size in that case.
2355          */
2356         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2357                                                     "physical-sector-size",
2358                                                     sector_size);
2359         blkfront_gather_backend_features(info);
2360         for_each_rinfo(info, rinfo, i) {
2361                 err = blkfront_setup_indirect(rinfo);
2362                 if (err) {
2363                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2364                                          info->xbdev->otherend);
2365                         blkif_free(info, 0);
2366                         break;
2367                 }
2368         }
2369
2370         err = xlvbd_alloc_gendisk(sectors, info, sector_size,
2371                                   physical_sector_size);
2372         if (err) {
2373                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2374                                  info->xbdev->otherend);
2375                 goto fail;
2376         }
2377
2378         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2379
2380         /* Kick pending requests. */
2381         info->connected = BLKIF_STATE_CONNECTED;
2382         for_each_rinfo(info, rinfo, i)
2383                 kick_pending_request_queues(rinfo);
2384
2385         err = device_add_disk(&info->xbdev->dev, info->gd, NULL);
2386         if (err) {
2387                 blk_cleanup_disk(info->gd);
2388                 blk_mq_free_tag_set(&info->tag_set);
2389                 info->rq = NULL;
2390                 goto fail;
2391         }
2392
2393         info->is_ready = 1;
2394         return;
2395
2396 fail:
2397         blkif_free(info, 0);
2398         return;
2399 }
2400
2401 /*
2402  * Callback received when the backend's state changes.
2403  */
2404 static void blkback_changed(struct xenbus_device *dev,
2405                             enum xenbus_state backend_state)
2406 {
2407         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2408
2409         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2410
2411         switch (backend_state) {
2412         case XenbusStateInitWait:
2413                 if (dev->state != XenbusStateInitialising)
2414                         break;
2415                 if (talk_to_blkback(dev, info))
2416                         break;
2417                 break;
2418         case XenbusStateInitialising:
2419         case XenbusStateInitialised:
2420         case XenbusStateReconfiguring:
2421         case XenbusStateReconfigured:
2422         case XenbusStateUnknown:
2423                 break;
2424
2425         case XenbusStateConnected:
2426                 /*
2427                  * talk_to_blkback sets state to XenbusStateInitialised
2428                  * and blkfront_connect sets it to XenbusStateConnected
2429                  * (if connection went OK).
2430                  *
2431                  * If the backend (or toolstack) decides to poke at backend
2432                  * state (and re-trigger the watch by setting the state repeatedly
2433                  * to XenbusStateConnected (4)) we need to deal with this.
2434                  * This is allowed as this is used to communicate to the guest
2435                  * that the size of disk has changed!
2436                  */
2437                 if ((dev->state != XenbusStateInitialised) &&
2438                     (dev->state != XenbusStateConnected)) {
2439                         if (talk_to_blkback(dev, info))
2440                                 break;
2441                 }
2442
2443                 blkfront_connect(info);
2444                 break;
2445
2446         case XenbusStateClosed:
2447                 if (dev->state == XenbusStateClosed)
2448                         break;
2449                 fallthrough;
2450         case XenbusStateClosing:
2451                 blkfront_closing(info);
2452                 break;
2453         }
2454 }
2455
2456 static int blkfront_remove(struct xenbus_device *xbdev)
2457 {
2458         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2459
2460         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2461
2462         if (info->gd)
2463                 del_gendisk(info->gd);
2464
2465         mutex_lock(&blkfront_mutex);
2466         list_del(&info->info_list);
2467         mutex_unlock(&blkfront_mutex);
2468
2469         blkif_free(info, 0);
2470         if (info->gd) {
2471                 xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2472                 blk_cleanup_disk(info->gd);
2473                 blk_mq_free_tag_set(&info->tag_set);
2474         }
2475
2476         kfree(info);
2477         return 0;
2478 }
2479
2480 static int blkfront_is_ready(struct xenbus_device *dev)
2481 {
2482         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2483
2484         return info->is_ready && info->xbdev;
2485 }
2486
2487 static const struct block_device_operations xlvbd_block_fops =
2488 {
2489         .owner = THIS_MODULE,
2490         .getgeo = blkif_getgeo,
2491         .ioctl = blkif_ioctl,
2492         .compat_ioctl = blkdev_compat_ptr_ioctl,
2493 };
2494
2495
2496 static const struct xenbus_device_id blkfront_ids[] = {
2497         { "vbd" },
2498         { "" }
2499 };
2500
2501 static struct xenbus_driver blkfront_driver = {
2502         .ids  = blkfront_ids,
2503         .probe = blkfront_probe,
2504         .remove = blkfront_remove,
2505         .resume = blkfront_resume,
2506         .otherend_changed = blkback_changed,
2507         .is_ready = blkfront_is_ready,
2508 };
2509
2510 static void purge_persistent_grants(struct blkfront_info *info)
2511 {
2512         unsigned int i;
2513         unsigned long flags;
2514         struct blkfront_ring_info *rinfo;
2515
2516         for_each_rinfo(info, rinfo, i) {
2517                 struct grant *gnt_list_entry, *tmp;
2518                 LIST_HEAD(grants);
2519
2520                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2521
2522                 if (rinfo->persistent_gnts_c == 0) {
2523                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2524                         continue;
2525                 }
2526
2527                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2528                                          node) {
2529                         if (gnt_list_entry->gref == INVALID_GRANT_REF ||
2530                             !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2531                                 continue;
2532
2533                         list_del(&gnt_list_entry->node);
2534                         rinfo->persistent_gnts_c--;
2535                         gnt_list_entry->gref = INVALID_GRANT_REF;
2536                         list_add_tail(&gnt_list_entry->node, &grants);
2537                 }
2538
2539                 list_splice_tail(&grants, &rinfo->grants);
2540
2541                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2542         }
2543 }
2544
2545 static void blkfront_delay_work(struct work_struct *work)
2546 {
2547         struct blkfront_info *info;
2548         bool need_schedule_work = false;
2549
2550         mutex_lock(&blkfront_mutex);
2551
2552         list_for_each_entry(info, &info_list, info_list) {
2553                 if (info->feature_persistent) {
2554                         need_schedule_work = true;
2555                         mutex_lock(&info->mutex);
2556                         purge_persistent_grants(info);
2557                         mutex_unlock(&info->mutex);
2558                 }
2559         }
2560
2561         if (need_schedule_work)
2562                 schedule_delayed_work(&blkfront_work, HZ * 10);
2563
2564         mutex_unlock(&blkfront_mutex);
2565 }
2566
2567 static int __init xlblk_init(void)
2568 {
2569         int ret;
2570         int nr_cpus = num_online_cpus();
2571
2572         if (!xen_domain())
2573                 return -ENODEV;
2574
2575         if (!xen_has_pv_disk_devices())
2576                 return -ENODEV;
2577
2578         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2579                 pr_warn("xen_blk: can't get major %d with name %s\n",
2580                         XENVBD_MAJOR, DEV_NAME);
2581                 return -ENODEV;
2582         }
2583
2584         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2585                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2586
2587         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2588                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2589                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2590                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2591         }
2592
2593         if (xen_blkif_max_queues > nr_cpus) {
2594                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2595                         xen_blkif_max_queues, nr_cpus);
2596                 xen_blkif_max_queues = nr_cpus;
2597         }
2598
2599         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2600
2601         ret = xenbus_register_frontend(&blkfront_driver);
2602         if (ret) {
2603                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2604                 return ret;
2605         }
2606
2607         return 0;
2608 }
2609 module_init(xlblk_init);
2610
2611
2612 static void __exit xlblk_exit(void)
2613 {
2614         cancel_delayed_work_sync(&blkfront_work);
2615
2616         xenbus_unregister_driver(&blkfront_driver);
2617         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2618         kfree(minors);
2619 }
2620 module_exit(xlblk_exit);
2621
2622 MODULE_DESCRIPTION("Xen virtual block device frontend");
2623 MODULE_LICENSE("GPL");
2624 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2625 MODULE_ALIAS("xen:vbd");
2626 MODULE_ALIAS("xenblk");