drivers/block/xen-blkfront.c

   1 /*
   2  * blkfront.c
   3  *
   4  * XenLinux virtual block device driver.
   5  *
   6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8  * Copyright (c) 2004, Christian Limpach
   9  * Copyright (c) 2004, Andrew Warfield
  10  * Copyright (c) 2005, Christopher Clark
  11  * Copyright (c) 2005, XenSource Ltd
  12  *
  13  * This program is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU General Public License version 2
  15  * as published by the Free Software Foundation; or, when distributed
  16  * separately from the Linux kernel or incorporated into other
  17  * software packages, subject to the following license:
  18  *
  19  * Permission is hereby granted, free of charge, to any person obtaining a copy
  20  * of this source file (the "Software"), to deal in the Software without
  21  * restriction, including without limitation the rights to use, copy, modify,
  22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23  * and to permit persons to whom the Software is furnished to do so, subject to
  24  * the following conditions:
  25  *
  26  * The above copyright notice and this permission notice shall be included in
  27  * all copies or substantial portions of the Software.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35  * IN THE SOFTWARE.
  36  */
  37
  38 #include <linux/interrupt.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/blk-mq.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/cdrom.h>
  43 #include <linux/module.h>
  44 #include <linux/slab.h>
  45 #include <linux/mutex.h>
  46 #include <linux/scatterlist.h>
  47 #include <linux/bitmap.h>
  48 #include <linux/list.h>
  49
  50 #include <xen/xen.h>
  51 #include <xen/xenbus.h>
  52 #include <xen/grant_table.h>
  53 #include <xen/events.h>
  54 #include <xen/page.h>
  55 #include <xen/platform_pci.h>
  56
  57 #include <xen/interface/grant_table.h>
  58 #include <xen/interface/io/blkif.h>
  59 #include <xen/interface/io/protocols.h>
  60
  61 #include <asm/xen/hypervisor.h>
  62
  63 /*
  64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  65  * When Linux is using a different page size than Xen, it may not be possible
  66  * to put all the data in a single segment.
  67  * This can happen when the backend doesn't support indirect descriptor and
  68  * therefore the maximum amount of data that a request can carry is
  69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  70  *
  71  * Note that we only support one extra request. So the Linux page size
  72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  73  * 88KB.
  74  */
  75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  76
  77 enum blkif_state {
  78         BLKIF_STATE_DISCONNECTED,
  79         BLKIF_STATE_CONNECTED,
  80         BLKIF_STATE_SUSPENDED,
  81         BLKIF_STATE_ERROR,
  82 };
  83
  84 struct grant {
  85         grant_ref_t gref;
  86         struct page *page;
  87         struct list_head node;
  88 };
  89
  90 enum blk_req_status {
  91         REQ_PROCESSING,
  92         REQ_WAITING,
  93         REQ_DONE,
  94         REQ_ERROR,
  95         REQ_EOPNOTSUPP,
  96 };
  97
  98 struct blk_shadow {
  99         struct blkif_request req;
 100         struct request *request;
 101         struct grant **grants_used;
 102         struct grant **indirect_grants;
 103         struct scatterlist *sg;
 104         unsigned int num_sg;
 105         enum blk_req_status status;
 106
 107         #define NO_ASSOCIATED_ID ~0UL
 108         /*
 109          * Id of the sibling if we ever need 2 requests when handling a
 110          * block I/O request
 111          */
 112         unsigned long associated_id;
 113 };
 114
 115 struct blkif_req {
 116         blk_status_t    error;
 117 };
 118
 119 static inline struct blkif_req *blkif_req(struct request *rq)
 120 {
 121         return blk_mq_rq_to_pdu(rq);
 122 }
 123
 124 static DEFINE_MUTEX(blkfront_mutex);
 125 static const struct block_device_operations xlvbd_block_fops;
 126
 127 /*
 128  * Maximum number of segments in indirect requests, the actual value used by
 129  * the frontend driver is the minimum of this value and the value provided
 130  * by the backend driver.
 131  */
 132
 133 static unsigned int xen_blkif_max_segments = 32;
 134 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
 135                    S_IRUGO);
 136 MODULE_PARM_DESC(max_indirect_segments,
 137                  "Maximum amount of segments in indirect requests (default is 32)");
 138
 139 static unsigned int xen_blkif_max_queues = 4;
 140 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
 141 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 142
 143 /*
 144  * Maximum order of pages to be used for the shared ring between front and
 145  * backend, 4KB page granularity is used.
 146  */
 147 static unsigned int xen_blkif_max_ring_order;
 148 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
 149 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 150
 151 #define BLK_RING_SIZE(info)     \
 152         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 153
 154 #define BLK_MAX_RING_SIZE       \
 155         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
 156
 157 /*
 158  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 159  * characters are enough. Define to 20 to keep consistent with backend.
 160  */
 161 #define RINGREF_NAME_LEN (20)
 162 /*
 163  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 164  */
 165 #define QUEUE_NAME_LEN (17)
 166
 167 /*
 168  *  Per-ring info.
 169  *  Every blkfront device can associate with one or more blkfront_ring_info,
 170  *  depending on how many hardware queues/rings to be used.
 171  */
 172 struct blkfront_ring_info {
 173         /* Lock to protect data in every ring buffer. */
 174         spinlock_t ring_lock;
 175         struct blkif_front_ring ring;
 176         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 177         unsigned int evtchn, irq;
 178         struct work_struct work;
 179         struct gnttab_free_callback callback;
 180         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 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 };
 187
 188 /*
 189  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 190  * hang in private_data off the gendisk structure. We may end up
 191  * putting all kinds of interesting stuff here :-)
 192  */
 193 struct blkfront_info
 194 {
 195         struct mutex mutex;
 196         struct xenbus_device *xbdev;
 197         struct gendisk *gd;
 198         u16 sector_size;
 199         unsigned int physical_sector_size;
 200         int vdevice;
 201         blkif_vdev_t handle;
 202         enum blkif_state connected;
 203         /* Number of pages per ring buffer. */
 204         unsigned int nr_ring_pages;
 205         struct request_queue *rq;
 206         unsigned int feature_flush:1;
 207         unsigned int feature_fua:1;
 208         unsigned int feature_discard:1;
 209         unsigned int feature_secdiscard:1;
 210         unsigned int feature_persistent:1;
 211         unsigned int discard_granularity;
 212         unsigned int discard_alignment;
 213         /* Number of 4KB segments handled */
 214         unsigned int max_indirect_segments;
 215         int is_ready;
 216         struct blk_mq_tag_set tag_set;
 217         struct blkfront_ring_info *rinfo;
 218         unsigned int nr_rings;
 219         /* Save uncomplete reqs and bios for migration. */
 220         struct list_head requests;
 221         struct bio_list bio_list;
 222 };
 223
 224 static unsigned int nr_minors;
 225 static unsigned long *minors;
 226 static DEFINE_SPINLOCK(minor_lock);
 227
 228 #define GRANT_INVALID_REF       0
 229
 230 #define PARTS_PER_DISK          16
 231 #define PARTS_PER_EXT_DISK      256
 232
 233 #define BLKIF_MAJOR(dev) ((dev)>>8)
 234 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 235
 236 #define EXT_SHIFT 28
 237 #define EXTENDED (1<<EXT_SHIFT)
 238 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 239 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 240 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 241 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 242 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 243 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 244
 245 #define DEV_NAME        "xvd"   /* name in /dev */
 246
 247 /*
 248  * Grants are always the same size as a Xen page (i.e 4KB).
 249  * A physical segment is always the same size as a Linux page.
 250  * Number of grants per physical segment
 251  */
 252 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 253
 254 #define GRANTS_PER_INDIRECT_FRAME \
 255         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 256
 257 #define PSEGS_PER_INDIRECT_FRAME        \
 258         (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
 259
 260 #define INDIRECT_GREFS(_grants)         \
 261         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 262
 263 #define GREFS(_psegs)   ((_psegs) * GRANTS_PER_PSEG)
 264
 265 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 266 static void blkfront_gather_backend_features(struct blkfront_info *info);
 267 static int negotiate_mq(struct blkfront_info *info);
 268
 269 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 270 {
 271         unsigned long free = rinfo->shadow_free;
 272
 273         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 274         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 275         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 276         return free;
 277 }
 278
 279 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 280                               unsigned long id)
 281 {
 282         if (rinfo->shadow[id].req.u.rw.id != id)
 283                 return -EINVAL;
 284         if (rinfo->shadow[id].request == NULL)
 285                 return -EINVAL;
 286         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 287         rinfo->shadow[id].request = NULL;
 288         rinfo->shadow_free = id;
 289         return 0;
 290 }
 291
 292 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 293 {
 294         struct blkfront_info *info = rinfo->dev_info;
 295         struct page *granted_page;
 296         struct grant *gnt_list_entry, *n;
 297         int i = 0;
 298
 299         while (i < num) {
 300                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 301                 if (!gnt_list_entry)
 302                         goto out_of_memory;
 303
 304                 if (info->feature_persistent) {
 305                         granted_page = alloc_page(GFP_NOIO);
 306                         if (!granted_page) {
 307                                 kfree(gnt_list_entry);
 308                                 goto out_of_memory;
 309                         }
 310                         gnt_list_entry->page = granted_page;
 311                 }
 312
 313                 gnt_list_entry->gref = GRANT_INVALID_REF;
 314                 list_add(&gnt_list_entry->node, &rinfo->grants);
 315                 i++;
 316         }
 317
 318         return 0;
 319
 320 out_of_memory:
 321         list_for_each_entry_safe(gnt_list_entry, n,
 322                                  &rinfo->grants, node) {
 323                 list_del(&gnt_list_entry->node);
 324                 if (info->feature_persistent)
 325                         __free_page(gnt_list_entry->page);
 326                 kfree(gnt_list_entry);
 327                 i--;
 328         }
 329         BUG_ON(i != 0);
 330         return -ENOMEM;
 331 }
 332
 333 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 334 {
 335         struct grant *gnt_list_entry;
 336
 337         BUG_ON(list_empty(&rinfo->grants));
 338         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 339                                           node);
 340         list_del(&gnt_list_entry->node);
 341
 342         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 343                 rinfo->persistent_gnts_c--;
 344
 345         return gnt_list_entry;
 346 }
 347
 348 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 349                                         const struct blkfront_info *info)
 350 {
 351         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 352                                                  info->xbdev->otherend_id,
 353                                                  gnt_list_entry->page,
 354                                                  0);
 355 }
 356
 357 static struct grant *get_grant(grant_ref_t *gref_head,
 358                                unsigned long gfn,
 359                                struct blkfront_ring_info *rinfo)
 360 {
 361         struct grant *gnt_list_entry = get_free_grant(rinfo);
 362         struct blkfront_info *info = rinfo->dev_info;
 363
 364         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 365                 return gnt_list_entry;
 366
 367         /* Assign a gref to this page */
 368         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 369         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 370         if (info->feature_persistent)
 371                 grant_foreign_access(gnt_list_entry, info);
 372         else {
 373                 /* Grant access to the GFN passed by the caller */
 374                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 375                                                 info->xbdev->otherend_id,
 376                                                 gfn, 0);
 377         }
 378
 379         return gnt_list_entry;
 380 }
 381
 382 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 383                                         struct blkfront_ring_info *rinfo)
 384 {
 385         struct grant *gnt_list_entry = get_free_grant(rinfo);
 386         struct blkfront_info *info = rinfo->dev_info;
 387
 388         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 389                 return gnt_list_entry;
 390
 391         /* Assign a gref to this page */
 392         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 393         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 394         if (!info->feature_persistent) {
 395                 struct page *indirect_page;
 396
 397                 /* Fetch a pre-allocated page to use for indirect grefs */
 398                 BUG_ON(list_empty(&rinfo->indirect_pages));
 399                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 400                                                  struct page, lru);
 401                 list_del(&indirect_page->lru);
 402                 gnt_list_entry->page = indirect_page;
 403         }
 404         grant_foreign_access(gnt_list_entry, info);
 405
 406         return gnt_list_entry;
 407 }
 408
 409 static const char *op_name(int op)
 410 {
 411         static const char *const names[] = {
 412                 [BLKIF_OP_READ] = "read",
 413                 [BLKIF_OP_WRITE] = "write",
 414                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 415                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 416                 [BLKIF_OP_DISCARD] = "discard" };
 417
 418         if (op < 0 || op >= ARRAY_SIZE(names))
 419                 return "unknown";
 420
 421         if (!names[op])
 422                 return "reserved";
 423
 424         return names[op];
 425 }
 426 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 427 {
 428         unsigned int end = minor + nr;
 429         int rc;
 430
 431         if (end > nr_minors) {
 432                 unsigned long *bitmap, *old;
 433
 434                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 435                                  GFP_KERNEL);
 436                 if (bitmap == NULL)
 437                         return -ENOMEM;
 438
 439                 spin_lock(&minor_lock);
 440                 if (end > nr_minors) {
 441                         old = minors;
 442                         memcpy(bitmap, minors,
 443                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 444                         minors = bitmap;
 445                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 446                 } else
 447                         old = bitmap;
 448                 spin_unlock(&minor_lock);
 449                 kfree(old);
 450         }
 451
 452         spin_lock(&minor_lock);
 453         if (find_next_bit(minors, end, minor) >= end) {
 454                 bitmap_set(minors, minor, nr);
 455                 rc = 0;
 456         } else
 457                 rc = -EBUSY;
 458         spin_unlock(&minor_lock);
 459
 460         return rc;
 461 }
 462
 463 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 464 {
 465         unsigned int end = minor + nr;
 466
 467         BUG_ON(end > nr_minors);
 468         spin_lock(&minor_lock);
 469         bitmap_clear(minors,  minor, nr);
 470         spin_unlock(&minor_lock);
 471 }
 472
 473 static void blkif_restart_queue_callback(void *arg)
 474 {
 475         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 476         schedule_work(&rinfo->work);
 477 }
 478
 479 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 480 {
 481         /* We don't have real geometry info, but let's at least return
 482            values consistent with the size of the device */
 483         sector_t nsect = get_capacity(bd->bd_disk);
 484         sector_t cylinders = nsect;
 485
 486         hg->heads = 0xff;
 487         hg->sectors = 0x3f;
 488         sector_div(cylinders, hg->heads * hg->sectors);
 489         hg->cylinders = cylinders;
 490         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 491                 hg->cylinders = 0xffff;
 492         return 0;
 493 }
 494
 495 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 496                        unsigned command, unsigned long argument)
 497 {
 498         struct blkfront_info *info = bdev->bd_disk->private_data;
 499         int i;
 500
 501         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 502                 command, (long)argument);
 503
 504         switch (command) {
 505         case CDROMMULTISESSION:
 506                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 507                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 508                         if (put_user(0, (char __user *)(argument + i)))
 509                                 return -EFAULT;
 510                 return 0;
 511
 512         case CDROM_GET_CAPABILITY: {
 513                 struct gendisk *gd = info->gd;
 514                 if (gd->flags & GENHD_FL_CD)
 515                         return 0;
 516                 return -EINVAL;
 517         }
 518
 519         default:
 520                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 521                   command);*/
 522                 return -EINVAL; /* same return as native Linux */
 523         }
 524
 525         return 0;
 526 }
 527
 528 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 529                                             struct request *req,
 530                                             struct blkif_request **ring_req)
 531 {
 532         unsigned long id;
 533
 534         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 535         rinfo->ring.req_prod_pvt++;
 536
 537         id = get_id_from_freelist(rinfo);
 538         rinfo->shadow[id].request = req;
 539         rinfo->shadow[id].status = REQ_PROCESSING;
 540         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 541
 542         rinfo->shadow[id].req.u.rw.id = id;
 543
 544         return id;
 545 }
 546
 547 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 548 {
 549         struct blkfront_info *info = rinfo->dev_info;
 550         struct blkif_request *ring_req, *final_ring_req;
 551         unsigned long id;
 552
 553         /* Fill out a communications ring structure. */
 554         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 555         ring_req = &rinfo->shadow[id].req;
 556
 557         ring_req->operation = BLKIF_OP_DISCARD;
 558         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 559         ring_req->u.discard.id = id;
 560         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 561         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 562                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 563         else
 564                 ring_req->u.discard.flag = 0;
 565
 566         /* Copy the request to the ring page. */
 567         *final_ring_req = *ring_req;
 568         rinfo->shadow[id].status = REQ_WAITING;
 569
 570         return 0;
 571 }
 572
 573 struct setup_rw_req {
 574         unsigned int grant_idx;
 575         struct blkif_request_segment *segments;
 576         struct blkfront_ring_info *rinfo;
 577         struct blkif_request *ring_req;
 578         grant_ref_t gref_head;
 579         unsigned int id;
 580         /* Only used when persistent grant is used and it's a read request */
 581         bool need_copy;
 582         unsigned int bvec_off;
 583         char *bvec_data;
 584
 585         bool require_extra_req;
 586         struct blkif_request *extra_ring_req;
 587 };
 588
 589 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 590                                      unsigned int len, void *data)
 591 {
 592         struct setup_rw_req *setup = data;
 593         int n, ref;
 594         struct grant *gnt_list_entry;
 595         unsigned int fsect, lsect;
 596         /* Convenient aliases */
 597         unsigned int grant_idx = setup->grant_idx;
 598         struct blkif_request *ring_req = setup->ring_req;
 599         struct blkfront_ring_info *rinfo = setup->rinfo;
 600         /*
 601          * We always use the shadow of the first request to store the list
 602          * of grant associated to the block I/O request. This made the
 603          * completion more easy to handle even if the block I/O request is
 604          * split.
 605          */
 606         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 607
 608         if (unlikely(setup->require_extra_req &&
 609                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 610                 /*
 611                  * We are using the second request, setup grant_idx
 612                  * to be the index of the segment array.
 613                  */
 614                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 615                 ring_req = setup->extra_ring_req;
 616         }
 617
 618         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 619             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 620                 if (setup->segments)
 621                         kunmap_atomic(setup->segments);
 622
 623                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 624                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 625                 shadow->indirect_grants[n] = gnt_list_entry;
 626                 setup->segments = kmap_atomic(gnt_list_entry->page);
 627                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 628         }
 629
 630         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 631         ref = gnt_list_entry->gref;
 632         /*
 633          * All the grants are stored in the shadow of the first
 634          * request. Therefore we have to use the global index.
 635          */
 636         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 637
 638         if (setup->need_copy) {
 639                 void *shared_data;
 640
 641                 shared_data = kmap_atomic(gnt_list_entry->page);
 642                 /*
 643                  * this does not wipe data stored outside the
 644                  * range sg->offset..sg->offset+sg->length.
 645                  * Therefore, blkback *could* see data from
 646                  * previous requests. This is OK as long as
 647                  * persistent grants are shared with just one
 648                  * domain. It may need refactoring if this
 649                  * changes
 650                  */
 651                 memcpy(shared_data + offset,
 652                        setup->bvec_data + setup->bvec_off,
 653                        len);
 654
 655                 kunmap_atomic(shared_data);
 656                 setup->bvec_off += len;
 657         }
 658
 659         fsect = offset >> 9;
 660         lsect = fsect + (len >> 9) - 1;
 661         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 662                 ring_req->u.rw.seg[grant_idx] =
 663                         (struct blkif_request_segment) {
 664                                 .gref       = ref,
 665                                 .first_sect = fsect,
 666                                 .last_sect  = lsect };
 667         } else {
 668                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 669                         (struct blkif_request_segment) {
 670                                 .gref       = ref,
 671                                 .first_sect = fsect,
 672                                 .last_sect  = lsect };
 673         }
 674
 675         (setup->grant_idx)++;
 676 }
 677
 678 static void blkif_setup_extra_req(struct blkif_request *first,
 679                                   struct blkif_request *second)
 680 {
 681         uint16_t nr_segments = first->u.rw.nr_segments;
 682
 683         /*
 684          * The second request is only present when the first request uses
 685          * all its segments. It's always the continuity of the first one.
 686          */
 687         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 688
 689         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 690         second->u.rw.sector_number = first->u.rw.sector_number +
 691                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 692
 693         second->u.rw.handle = first->u.rw.handle;
 694         second->operation = first->operation;
 695 }
 696
 697 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 698 {
 699         struct blkfront_info *info = rinfo->dev_info;
 700         struct blkif_request *ring_req, *extra_ring_req = NULL;
 701         struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
 702         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 703         bool require_extra_req = false;
 704         int i;
 705         struct setup_rw_req setup = {
 706                 .grant_idx = 0,
 707                 .segments = NULL,
 708                 .rinfo = rinfo,
 709                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 710         };
 711
 712         /*
 713          * Used to store if we are able to queue the request by just using
 714          * existing persistent grants, or if we have to get new grants,
 715          * as there are not sufficiently many free.
 716          */
 717         bool new_persistent_gnts = false;
 718         struct scatterlist *sg;
 719         int num_sg, max_grefs, num_grant;
 720
 721         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 722         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 723                 /*
 724                  * If we are using indirect segments we need to account
 725                  * for the indirect grefs used in the request.
 726                  */
 727                 max_grefs += INDIRECT_GREFS(max_grefs);
 728
 729         /* Check if we have enough persistent grants to allocate a requests */
 730         if (rinfo->persistent_gnts_c < max_grefs) {
 731                 new_persistent_gnts = true;
 732
 733                 if (gnttab_alloc_grant_references(
 734                     max_grefs - rinfo->persistent_gnts_c,
 735                     &setup.gref_head) < 0) {
 736                         gnttab_request_free_callback(
 737                                 &rinfo->callback,
 738                                 blkif_restart_queue_callback,
 739                                 rinfo,
 740                                 max_grefs - rinfo->persistent_gnts_c);
 741                         return 1;
 742                 }
 743         }
 744
 745         /* Fill out a communications ring structure. */
 746         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 747         ring_req = &rinfo->shadow[id].req;
 748
 749         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 750         num_grant = 0;
 751         /* Calculate the number of grant used */
 752         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 753                num_grant += gnttab_count_grant(sg->offset, sg->length);
 754
 755         require_extra_req = info->max_indirect_segments == 0 &&
 756                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 757         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 758
 759         rinfo->shadow[id].num_sg = num_sg;
 760         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 761             likely(!require_extra_req)) {
 762                 /*
 763                  * The indirect operation can only be a BLKIF_OP_READ or
 764                  * BLKIF_OP_WRITE
 765                  */
 766                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 767                 ring_req->operation = BLKIF_OP_INDIRECT;
 768                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 769                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 770                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 771                 ring_req->u.indirect.handle = info->handle;
 772                 ring_req->u.indirect.nr_segments = num_grant;
 773         } else {
 774                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 775                 ring_req->u.rw.handle = info->handle;
 776                 ring_req->operation = rq_data_dir(req) ?
 777                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 778                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 779                         /*
 780                          * Ideally we can do an unordered flush-to-disk.
 781                          * In case the backend onlysupports barriers, use that.
 782                          * A barrier request a superset of FUA, so we can
 783                          * implement it the same way.  (It's also a FLUSH+FUA,
 784                          * since it is guaranteed ordered WRT previous writes.)
 785                          */
 786                         if (info->feature_flush && info->feature_fua)
 787                                 ring_req->operation =
 788                                         BLKIF_OP_WRITE_BARRIER;
 789                         else if (info->feature_flush)
 790                                 ring_req->operation =
 791                                         BLKIF_OP_FLUSH_DISKCACHE;
 792                         else
 793                                 ring_req->operation = 0;
 794                 }
 795                 ring_req->u.rw.nr_segments = num_grant;
 796                 if (unlikely(require_extra_req)) {
 797                         extra_id = blkif_ring_get_request(rinfo, req,
 798                                                           &final_extra_ring_req);
 799                         extra_ring_req = &rinfo->shadow[extra_id].req;
 800
 801                         /*
 802                          * Only the first request contains the scatter-gather
 803                          * list.
 804                          */
 805                         rinfo->shadow[extra_id].num_sg = 0;
 806
 807                         blkif_setup_extra_req(ring_req, extra_ring_req);
 808
 809                         /* Link the 2 requests together */
 810                         rinfo->shadow[extra_id].associated_id = id;
 811                         rinfo->shadow[id].associated_id = extra_id;
 812                 }
 813         }
 814
 815         setup.ring_req = ring_req;
 816         setup.id = id;
 817
 818         setup.require_extra_req = require_extra_req;
 819         if (unlikely(require_extra_req))
 820                 setup.extra_ring_req = extra_ring_req;
 821
 822         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 823                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 824
 825                 if (setup.need_copy) {
 826                         setup.bvec_off = sg->offset;
 827                         setup.bvec_data = kmap_atomic(sg_page(sg));
 828                 }
 829
 830                 gnttab_foreach_grant_in_range(sg_page(sg),
 831                                               sg->offset,
 832                                               sg->length,
 833                                               blkif_setup_rw_req_grant,
 834                                               &setup);
 835
 836                 if (setup.need_copy)
 837                         kunmap_atomic(setup.bvec_data);
 838         }
 839         if (setup.segments)
 840                 kunmap_atomic(setup.segments);
 841
 842         /* Copy request(s) to the ring page. */
 843         *final_ring_req = *ring_req;
 844         rinfo->shadow[id].status = REQ_WAITING;
 845         if (unlikely(require_extra_req)) {
 846                 *final_extra_ring_req = *extra_ring_req;
 847                 rinfo->shadow[extra_id].status = REQ_WAITING;
 848         }
 849
 850         if (new_persistent_gnts)
 851                 gnttab_free_grant_references(setup.gref_head);
 852
 853         return 0;
 854 }
 855
 856 /*
 857  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 858  * and writes are handled as expected.
 859  *
 860  * @req: a request struct
 861  */
 862 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 863 {
 864         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 865                 return 1;
 866
 867         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 868                      req_op(req) == REQ_OP_SECURE_ERASE))
 869                 return blkif_queue_discard_req(req, rinfo);
 870         else
 871                 return blkif_queue_rw_req(req, rinfo);
 872 }
 873
 874 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 875 {
 876         int notify;
 877
 878         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 879
 880         if (notify)
 881                 notify_remote_via_irq(rinfo->irq);
 882 }
 883
 884 static inline bool blkif_request_flush_invalid(struct request *req,
 885                                                struct blkfront_info *info)
 886 {
 887         return (blk_rq_is_passthrough(req) ||
 888                 ((req_op(req) == REQ_OP_FLUSH) &&
 889                  !info->feature_flush) ||
 890                 ((req->cmd_flags & REQ_FUA) &&
 891                  !info->feature_fua));
 892 }
 893
 894 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 895                           const struct blk_mq_queue_data *qd)
 896 {
 897         unsigned long flags;
 898         int qid = hctx->queue_num;
 899         struct blkfront_info *info = hctx->queue->queuedata;
 900         struct blkfront_ring_info *rinfo = NULL;
 901
 902         BUG_ON(info->nr_rings <= qid);
 903         rinfo = &info->rinfo[qid];
 904         blk_mq_start_request(qd->rq);
 905         spin_lock_irqsave(&rinfo->ring_lock, flags);
 906         if (RING_FULL(&rinfo->ring))
 907                 goto out_busy;
 908
 909         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 910                 goto out_err;
 911
 912         if (blkif_queue_request(qd->rq, rinfo))
 913                 goto out_busy;
 914
 915         flush_requests(rinfo);
 916         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 917         return BLK_STS_OK;
 918
 919 out_err:
 920         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 921         return BLK_STS_IOERR;
 922
 923 out_busy:
 924         blk_mq_stop_hw_queue(hctx);
 925         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 926         return BLK_STS_RESOURCE;
 927 }
 928
 929 static void blkif_complete_rq(struct request *rq)
 930 {
 931         blk_mq_end_request(rq, blkif_req(rq)->error);
 932 }
 933
 934 static const struct blk_mq_ops blkfront_mq_ops = {
 935         .queue_rq = blkif_queue_rq,
 936         .complete = blkif_complete_rq,
 937 };
 938
 939 static void blkif_set_queue_limits(struct blkfront_info *info)
 940 {
 941         struct request_queue *rq = info->rq;
 942         struct gendisk *gd = info->gd;
 943         unsigned int segments = info->max_indirect_segments ? :
 944                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 945
 946         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 947
 948         if (info->feature_discard) {
 949                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 950                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 951                 rq->limits.discard_granularity = info->discard_granularity ?:
 952                                                  info->physical_sector_size;
 953                 rq->limits.discard_alignment = info->discard_alignment;
 954                 if (info->feature_secdiscard)
 955                         queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
 956         }
 957
 958         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 959         blk_queue_logical_block_size(rq, info->sector_size);
 960         blk_queue_physical_block_size(rq, info->physical_sector_size);
 961         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 962
 963         /* Each segment in a request is up to an aligned page in size. */
 964         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 965         blk_queue_max_segment_size(rq, PAGE_SIZE);
 966
 967         /* Ensure a merged request will fit in a single I/O ring slot. */
 968         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 969
 970         /* Make sure buffer addresses are sector-aligned. */
 971         blk_queue_dma_alignment(rq, 511);
 972 }
 973
 974 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 975                                 unsigned int physical_sector_size)
 976 {
 977         struct request_queue *rq;
 978         struct blkfront_info *info = gd->private_data;
 979
 980         memset(&info->tag_set, 0, sizeof(info->tag_set));
 981         info->tag_set.ops = &blkfront_mq_ops;
 982         info->tag_set.nr_hw_queues = info->nr_rings;
 983         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
 984                 /*
 985                  * When indirect descriptior is not supported, the I/O request
 986                  * will be split between multiple request in the ring.
 987                  * To avoid problems when sending the request, divide by
 988                  * 2 the depth of the queue.
 989                  */
 990                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
 991         } else
 992                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
 993         info->tag_set.numa_node = NUMA_NO_NODE;
 994         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
 995         info->tag_set.cmd_size = sizeof(struct blkif_req);
 996         info->tag_set.driver_data = info;
 997
 998         if (blk_mq_alloc_tag_set(&info->tag_set))
 999                 return -EINVAL;
1000         rq = blk_mq_init_queue(&info->tag_set);
1001         if (IS_ERR(rq)) {
1002                 blk_mq_free_tag_set(&info->tag_set);
1003                 return PTR_ERR(rq);
1004         }
1005
1006         rq->queuedata = info;
1007         info->rq = gd->queue = rq;
1008         info->gd = gd;
1009         info->sector_size = sector_size;
1010         info->physical_sector_size = physical_sector_size;
1011         blkif_set_queue_limits(info);
1012
1013         return 0;
1014 }
1015
1016 static const char *flush_info(struct blkfront_info *info)
1017 {
1018         if (info->feature_flush && info->feature_fua)
1019                 return "barrier: enabled;";
1020         else if (info->feature_flush)
1021                 return "flush diskcache: enabled;";
1022         else
1023                 return "barrier or flush: disabled;";
1024 }
1025
1026 static void xlvbd_flush(struct blkfront_info *info)
1027 {
1028         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1029                               info->feature_fua ? true : false);
1030         pr_info("blkfront: %s: %s %s %s %s %s\n",
1031                 info->gd->disk_name, flush_info(info),
1032                 "persistent grants:", info->feature_persistent ?
1033                 "enabled;" : "disabled;", "indirect descriptors:",
1034                 info->max_indirect_segments ? "enabled;" : "disabled;");
1035 }
1036
1037 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1038 {
1039         int major;
1040         major = BLKIF_MAJOR(vdevice);
1041         *minor = BLKIF_MINOR(vdevice);
1042         switch (major) {
1043                 case XEN_IDE0_MAJOR:
1044                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1045                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1046                                 EMULATED_HD_DISK_MINOR_OFFSET;
1047                         break;
1048                 case XEN_IDE1_MAJOR:
1049                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1050                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1051                                 EMULATED_HD_DISK_MINOR_OFFSET;
1052                         break;
1053                 case XEN_SCSI_DISK0_MAJOR:
1054                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1055                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1056                         break;
1057                 case XEN_SCSI_DISK1_MAJOR:
1058                 case XEN_SCSI_DISK2_MAJOR:
1059                 case XEN_SCSI_DISK3_MAJOR:
1060                 case XEN_SCSI_DISK4_MAJOR:
1061                 case XEN_SCSI_DISK5_MAJOR:
1062                 case XEN_SCSI_DISK6_MAJOR:
1063                 case XEN_SCSI_DISK7_MAJOR:
1064                         *offset = (*minor / PARTS_PER_DISK) +
1065                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1066                                 EMULATED_SD_DISK_NAME_OFFSET;
1067                         *minor = *minor +
1068                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1069                                 EMULATED_SD_DISK_MINOR_OFFSET;
1070                         break;
1071                 case XEN_SCSI_DISK8_MAJOR:
1072                 case XEN_SCSI_DISK9_MAJOR:
1073                 case XEN_SCSI_DISK10_MAJOR:
1074                 case XEN_SCSI_DISK11_MAJOR:
1075                 case XEN_SCSI_DISK12_MAJOR:
1076                 case XEN_SCSI_DISK13_MAJOR:
1077                 case XEN_SCSI_DISK14_MAJOR:
1078                 case XEN_SCSI_DISK15_MAJOR:
1079                         *offset = (*minor / PARTS_PER_DISK) +
1080                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1081                                 EMULATED_SD_DISK_NAME_OFFSET;
1082                         *minor = *minor +
1083                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1084                                 EMULATED_SD_DISK_MINOR_OFFSET;
1085                         break;
1086                 case XENVBD_MAJOR:
1087                         *offset = *minor / PARTS_PER_DISK;
1088                         break;
1089                 default:
1090                         printk(KERN_WARNING "blkfront: your disk configuration is "
1091                                         "incorrect, please use an xvd device instead\n");
1092                         return -ENODEV;
1093         }
1094         return 0;
1095 }
1096
1097 static char *encode_disk_name(char *ptr, unsigned int n)
1098 {
1099         if (n >= 26)
1100                 ptr = encode_disk_name(ptr, n / 26 - 1);
1101         *ptr = 'a' + n % 26;
1102         return ptr + 1;
1103 }
1104
1105 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1106                                struct blkfront_info *info,
1107                                u16 vdisk_info, u16 sector_size,
1108                                unsigned int physical_sector_size)
1109 {
1110         struct gendisk *gd;
1111         int nr_minors = 1;
1112         int err;
1113         unsigned int offset;
1114         int minor;
1115         int nr_parts;
1116         char *ptr;
1117
1118         BUG_ON(info->gd != NULL);
1119         BUG_ON(info->rq != NULL);
1120
1121         if ((info->vdevice>>EXT_SHIFT) > 1) {
1122                 /* this is above the extended range; something is wrong */
1123                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1124                 return -ENODEV;
1125         }
1126
1127         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1128                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1129                 if (err)
1130                         return err;
1131                 nr_parts = PARTS_PER_DISK;
1132         } else {
1133                 minor = BLKIF_MINOR_EXT(info->vdevice);
1134                 nr_parts = PARTS_PER_EXT_DISK;
1135                 offset = minor / nr_parts;
1136                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1137                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1138                                         "emulated IDE disks,\n\t choose an xvd device name"
1139                                         "from xvde on\n", info->vdevice);
1140         }
1141         if (minor >> MINORBITS) {
1142                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1143                         info->vdevice, minor);
1144                 return -ENODEV;
1145         }
1146
1147         if ((minor % nr_parts) == 0)
1148                 nr_minors = nr_parts;
1149
1150         err = xlbd_reserve_minors(minor, nr_minors);
1151         if (err)
1152                 goto out;
1153         err = -ENODEV;
1154
1155         gd = alloc_disk(nr_minors);
1156         if (gd == NULL)
1157                 goto release;
1158
1159         strcpy(gd->disk_name, DEV_NAME);
1160         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1161         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1162         if (nr_minors > 1)
1163                 *ptr = 0;
1164         else
1165                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1166                          "%d", minor & (nr_parts - 1));
1167
1168         gd->major = XENVBD_MAJOR;
1169         gd->first_minor = minor;
1170         gd->fops = &xlvbd_block_fops;
1171         gd->private_data = info;
1172         set_capacity(gd, capacity);
1173
1174         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1175                 del_gendisk(gd);
1176                 goto release;
1177         }
1178
1179         xlvbd_flush(info);
1180
1181         if (vdisk_info & VDISK_READONLY)
1182                 set_disk_ro(gd, 1);
1183
1184         if (vdisk_info & VDISK_REMOVABLE)
1185                 gd->flags |= GENHD_FL_REMOVABLE;
1186
1187         if (vdisk_info & VDISK_CDROM)
1188                 gd->flags |= GENHD_FL_CD;
1189
1190         return 0;
1191
1192  release:
1193         xlbd_release_minors(minor, nr_minors);
1194  out:
1195         return err;
1196 }
1197
1198 static void xlvbd_release_gendisk(struct blkfront_info *info)
1199 {
1200         unsigned int minor, nr_minors, i;
1201
1202         if (info->rq == NULL)
1203                 return;
1204
1205         /* No more blkif_request(). */
1206         blk_mq_stop_hw_queues(info->rq);
1207
1208         for (i = 0; i < info->nr_rings; i++) {
1209                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1210
1211                 /* No more gnttab callback work. */
1212                 gnttab_cancel_free_callback(&rinfo->callback);
1213
1214                 /* Flush gnttab callback work. Must be done with no locks held. */
1215                 flush_work(&rinfo->work);
1216         }
1217
1218         del_gendisk(info->gd);
1219
1220         minor = info->gd->first_minor;
1221         nr_minors = info->gd->minors;
1222         xlbd_release_minors(minor, nr_minors);
1223
1224         blk_cleanup_queue(info->rq);
1225         blk_mq_free_tag_set(&info->tag_set);
1226         info->rq = NULL;
1227
1228         put_disk(info->gd);
1229         info->gd = NULL;
1230 }
1231
1232 /* Already hold rinfo->ring_lock. */
1233 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1234 {
1235         if (!RING_FULL(&rinfo->ring))
1236                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1237 }
1238
1239 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1240 {
1241         unsigned long flags;
1242
1243         spin_lock_irqsave(&rinfo->ring_lock, flags);
1244         kick_pending_request_queues_locked(rinfo);
1245         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1246 }
1247
1248 static void blkif_restart_queue(struct work_struct *work)
1249 {
1250         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1251
1252         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1253                 kick_pending_request_queues(rinfo);
1254 }
1255
1256 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1257 {
1258         struct grant *persistent_gnt, *n;
1259         struct blkfront_info *info = rinfo->dev_info;
1260         int i, j, segs;
1261
1262         /*
1263          * Remove indirect pages, this only happens when using indirect
1264          * descriptors but not persistent grants
1265          */
1266         if (!list_empty(&rinfo->indirect_pages)) {
1267                 struct page *indirect_page, *n;
1268
1269                 BUG_ON(info->feature_persistent);
1270                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1271                         list_del(&indirect_page->lru);
1272                         __free_page(indirect_page);
1273                 }
1274         }
1275
1276         /* Remove all persistent grants. */
1277         if (!list_empty(&rinfo->grants)) {
1278                 list_for_each_entry_safe(persistent_gnt, n,
1279                                          &rinfo->grants, node) {
1280                         list_del(&persistent_gnt->node);
1281                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1282                                 gnttab_end_foreign_access(persistent_gnt->gref,
1283                                                           0, 0UL);
1284                                 rinfo->persistent_gnts_c--;
1285                         }
1286                         if (info->feature_persistent)
1287                                 __free_page(persistent_gnt->page);
1288                         kfree(persistent_gnt);
1289                 }
1290         }
1291         BUG_ON(rinfo->persistent_gnts_c != 0);
1292
1293         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1294                 /*
1295                  * Clear persistent grants present in requests already
1296                  * on the shared ring
1297                  */
1298                 if (!rinfo->shadow[i].request)
1299                         goto free_shadow;
1300
1301                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1302                        rinfo->shadow[i].req.u.indirect.nr_segments :
1303                        rinfo->shadow[i].req.u.rw.nr_segments;
1304                 for (j = 0; j < segs; j++) {
1305                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1306                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1307                         if (info->feature_persistent)
1308                                 __free_page(persistent_gnt->page);
1309                         kfree(persistent_gnt);
1310                 }
1311
1312                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1313                         /*
1314                          * If this is not an indirect operation don't try to
1315                          * free indirect segments
1316                          */
1317                         goto free_shadow;
1318
1319                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1320                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1321                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1322                         __free_page(persistent_gnt->page);
1323                         kfree(persistent_gnt);
1324                 }
1325
1326 free_shadow:
1327                 kfree(rinfo->shadow[i].grants_used);
1328                 rinfo->shadow[i].grants_used = NULL;
1329                 kfree(rinfo->shadow[i].indirect_grants);
1330                 rinfo->shadow[i].indirect_grants = NULL;
1331                 kfree(rinfo->shadow[i].sg);
1332                 rinfo->shadow[i].sg = NULL;
1333         }
1334
1335         /* No more gnttab callback work. */
1336         gnttab_cancel_free_callback(&rinfo->callback);
1337
1338         /* Flush gnttab callback work. Must be done with no locks held. */
1339         flush_work(&rinfo->work);
1340
1341         /* Free resources associated with old device channel. */
1342         for (i = 0; i < info->nr_ring_pages; i++) {
1343                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1344                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1345                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1346                 }
1347         }
1348         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1349         rinfo->ring.sring = NULL;
1350
1351         if (rinfo->irq)
1352                 unbind_from_irqhandler(rinfo->irq, rinfo);
1353         rinfo->evtchn = rinfo->irq = 0;
1354 }
1355
1356 static void blkif_free(struct blkfront_info *info, int suspend)
1357 {
1358         unsigned int i;
1359
1360         /* Prevent new requests being issued until we fix things up. */
1361         info->connected = suspend ?
1362                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1363         /* No more blkif_request(). */
1364         if (info->rq)
1365                 blk_mq_stop_hw_queues(info->rq);
1366
1367         for (i = 0; i < info->nr_rings; i++)
1368                 blkif_free_ring(&info->rinfo[i]);
1369
1370         kfree(info->rinfo);
1371         info->rinfo = NULL;
1372         info->nr_rings = 0;
1373 }
1374
1375 struct copy_from_grant {
1376         const struct blk_shadow *s;
1377         unsigned int grant_idx;
1378         unsigned int bvec_offset;
1379         char *bvec_data;
1380 };
1381
1382 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1383                                   unsigned int len, void *data)
1384 {
1385         struct copy_from_grant *info = data;
1386         char *shared_data;
1387         /* Convenient aliases */
1388         const struct blk_shadow *s = info->s;
1389
1390         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1391
1392         memcpy(info->bvec_data + info->bvec_offset,
1393                shared_data + offset, len);
1394
1395         info->bvec_offset += len;
1396         info->grant_idx++;
1397
1398         kunmap_atomic(shared_data);
1399 }
1400
1401 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1402 {
1403         switch (rsp)
1404         {
1405         case BLKIF_RSP_OKAY:
1406                 return REQ_DONE;
1407         case BLKIF_RSP_EOPNOTSUPP:
1408                 return REQ_EOPNOTSUPP;
1409         case BLKIF_RSP_ERROR:
1410                 /* Fallthrough. */
1411         default:
1412                 return REQ_ERROR;
1413         }
1414 }
1415
1416 /*
1417  * Get the final status of the block request based on two ring response
1418  */
1419 static int blkif_get_final_status(enum blk_req_status s1,
1420                                   enum blk_req_status s2)
1421 {
1422         BUG_ON(s1 < REQ_DONE);
1423         BUG_ON(s2 < REQ_DONE);
1424
1425         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1426                 return BLKIF_RSP_ERROR;
1427         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1428                 return BLKIF_RSP_EOPNOTSUPP;
1429         return BLKIF_RSP_OKAY;
1430 }
1431
1432 static bool blkif_completion(unsigned long *id,
1433                              struct blkfront_ring_info *rinfo,
1434                              struct blkif_response *bret)
1435 {
1436         int i = 0;
1437         struct scatterlist *sg;
1438         int num_sg, num_grant;
1439         struct blkfront_info *info = rinfo->dev_info;
1440         struct blk_shadow *s = &rinfo->shadow[*id];
1441         struct copy_from_grant data = {
1442                 .grant_idx = 0,
1443         };
1444
1445         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1446                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1447
1448         /* The I/O request may be split in two. */
1449         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1450                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1451
1452                 /* Keep the status of the current response in shadow. */
1453                 s->status = blkif_rsp_to_req_status(bret->status);
1454
1455                 /* Wait the second response if not yet here. */
1456                 if (s2->status < REQ_DONE)
1457                         return 0;
1458
1459                 bret->status = blkif_get_final_status(s->status,
1460                                                       s2->status);
1461
1462                 /*
1463                  * All the grants is stored in the first shadow in order
1464                  * to make the completion code simpler.
1465                  */
1466                 num_grant += s2->req.u.rw.nr_segments;
1467
1468                 /*
1469                  * The two responses may not come in order. Only the
1470                  * first request will store the scatter-gather list.
1471                  */
1472                 if (s2->num_sg != 0) {
1473                         /* Update "id" with the ID of the first response. */
1474                         *id = s->associated_id;
1475                         s = s2;
1476                 }
1477
1478                 /*
1479                  * We don't need anymore the second request, so recycling
1480                  * it now.
1481                  */
1482                 if (add_id_to_freelist(rinfo, s->associated_id))
1483                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1484                              info->gd->disk_name, s->associated_id);
1485         }
1486
1487         data.s = s;
1488         num_sg = s->num_sg;
1489
1490         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1491                 for_each_sg(s->sg, sg, num_sg, i) {
1492                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1493
1494                         data.bvec_offset = sg->offset;
1495                         data.bvec_data = kmap_atomic(sg_page(sg));
1496
1497                         gnttab_foreach_grant_in_range(sg_page(sg),
1498                                                       sg->offset,
1499                                                       sg->length,
1500                                                       blkif_copy_from_grant,
1501                                                       &data);
1502
1503                         kunmap_atomic(data.bvec_data);
1504                 }
1505         }
1506         /* Add the persistent grant into the list of free grants */
1507         for (i = 0; i < num_grant; i++) {
1508                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1509                         /*
1510                          * If the grant is still mapped by the backend (the
1511                          * backend has chosen to make this grant persistent)
1512                          * we add it at the head of the list, so it will be
1513                          * reused first.
1514                          */
1515                         if (!info->feature_persistent)
1516                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1517                                                      s->grants_used[i]->gref);
1518                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1519                         rinfo->persistent_gnts_c++;
1520                 } else {
1521                         /*
1522                          * If the grant is not mapped by the backend we end the
1523                          * foreign access and add it to the tail of the list,
1524                          * so it will not be picked again unless we run out of
1525                          * persistent grants.
1526                          */
1527                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1528                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1529                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1530                 }
1531         }
1532         if (s->req.operation == BLKIF_OP_INDIRECT) {
1533                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1534                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1535                                 if (!info->feature_persistent)
1536                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1537                                                              s->indirect_grants[i]->gref);
1538                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1539                                 rinfo->persistent_gnts_c++;
1540                         } else {
1541                                 struct page *indirect_page;
1542
1543                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1544                                 /*
1545                                  * Add the used indirect page back to the list of
1546                                  * available pages for indirect grefs.
1547                                  */
1548                                 if (!info->feature_persistent) {
1549                                         indirect_page = s->indirect_grants[i]->page;
1550                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1551                                 }
1552                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1553                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1554                         }
1555                 }
1556         }
1557
1558         return 1;
1559 }
1560
1561 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1562 {
1563         struct request *req;
1564         struct blkif_response bret;
1565         RING_IDX i, rp;
1566         unsigned long flags;
1567         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1568         struct blkfront_info *info = rinfo->dev_info;
1569         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1570
1571         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1572                 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1573                 return IRQ_HANDLED;
1574         }
1575
1576         spin_lock_irqsave(&rinfo->ring_lock, flags);
1577  again:
1578         rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1579         virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1580         if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1581                 pr_alert("%s: illegal number of responses %u\n",
1582                          info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1583                 goto err;
1584         }
1585
1586         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1587                 unsigned long id;
1588                 unsigned int op;
1589
1590                 eoiflag = 0;
1591
1592                 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1593                 id = bret.id;
1594
1595                 /*
1596                  * The backend has messed up and given us an id that we would
1597                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1598                  * look in get_id_from_freelist.
1599                  */
1600                 if (id >= BLK_RING_SIZE(info)) {
1601                         pr_alert("%s: response has incorrect id (%ld)\n",
1602                                  info->gd->disk_name, id);
1603                         goto err;
1604                 }
1605                 if (rinfo->shadow[id].status != REQ_WAITING) {
1606                         pr_alert("%s: response references no pending request\n",
1607                                  info->gd->disk_name);
1608                         goto err;
1609                 }
1610
1611                 rinfo->shadow[id].status = REQ_PROCESSING;
1612                 req  = rinfo->shadow[id].request;
1613
1614                 op = rinfo->shadow[id].req.operation;
1615                 if (op == BLKIF_OP_INDIRECT)
1616                         op = rinfo->shadow[id].req.u.indirect.indirect_op;
1617                 if (bret.operation != op) {
1618                         pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1619                                  info->gd->disk_name, bret.operation, op);
1620                         goto err;
1621                 }
1622
1623                 if (bret.operation != BLKIF_OP_DISCARD) {
1624                         /*
1625                          * We may need to wait for an extra response if the
1626                          * I/O request is split in 2
1627                          */
1628                         if (!blkif_completion(&id, rinfo, &bret))
1629                                 continue;
1630                 }
1631
1632                 if (add_id_to_freelist(rinfo, id)) {
1633                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1634                              info->gd->disk_name, op_name(bret.operation), id);
1635                         continue;
1636                 }
1637
1638                 if (bret.status == BLKIF_RSP_OKAY)
1639                         blkif_req(req)->error = BLK_STS_OK;
1640                 else
1641                         blkif_req(req)->error = BLK_STS_IOERR;
1642
1643                 switch (bret.operation) {
1644                 case BLKIF_OP_DISCARD:
1645                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1646                                 struct request_queue *rq = info->rq;
1647
1648                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1649                                            info->gd->disk_name, op_name(bret.operation));
1650                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1651                                 info->feature_discard = 0;
1652                                 info->feature_secdiscard = 0;
1653                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1654                                 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1655                         }
1656                         break;
1657                 case BLKIF_OP_FLUSH_DISKCACHE:
1658                 case BLKIF_OP_WRITE_BARRIER:
1659                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1660                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1661                                        info->gd->disk_name, op_name(bret.operation));
1662                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1663                         }
1664                         if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1665                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1666                                 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1667                                        info->gd->disk_name, op_name(bret.operation));
1668                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1669                         }
1670                         if (unlikely(blkif_req(req)->error)) {
1671                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1672                                         blkif_req(req)->error = BLK_STS_OK;
1673                                 info->feature_fua = 0;
1674                                 info->feature_flush = 0;
1675                                 xlvbd_flush(info);
1676                         }
1677                         /* fall through */
1678                 case BLKIF_OP_READ:
1679                 case BLKIF_OP_WRITE:
1680                         if (unlikely(bret.status != BLKIF_RSP_OKAY))
1681                                 dev_dbg_ratelimited(&info->xbdev->dev,
1682                                         "Bad return from blkdev data request: %#x\n",
1683                                         bret.status);
1684
1685                         break;
1686                 default:
1687                         BUG();
1688                 }
1689
1690                 blk_mq_complete_request(req);
1691         }
1692
1693         rinfo->ring.rsp_cons = i;
1694
1695         if (i != rinfo->ring.req_prod_pvt) {
1696                 int more_to_do;
1697                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1698                 if (more_to_do)
1699                         goto again;
1700         } else
1701                 rinfo->ring.sring->rsp_event = i + 1;
1702
1703         kick_pending_request_queues_locked(rinfo);
1704
1705         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1706
1707         xen_irq_lateeoi(irq, eoiflag);
1708
1709         return IRQ_HANDLED;
1710
1711  err:
1712         info->connected = BLKIF_STATE_ERROR;
1713
1714         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1715
1716         /* No EOI in order to avoid further interrupts. */
1717
1718         pr_alert("%s disabled for further use\n", info->gd->disk_name);
1719         return IRQ_HANDLED;
1720 }
1721
1722
1723 static int setup_blkring(struct xenbus_device *dev,
1724                          struct blkfront_ring_info *rinfo)
1725 {
1726         struct blkif_sring *sring;
1727         int err, i;
1728         struct blkfront_info *info = rinfo->dev_info;
1729         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1730         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1731
1732         for (i = 0; i < info->nr_ring_pages; i++)
1733                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1734
1735         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1736                                                        get_order(ring_size));
1737         if (!sring) {
1738                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1739                 return -ENOMEM;
1740         }
1741         SHARED_RING_INIT(sring);
1742         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1743
1744         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1745         if (err < 0) {
1746                 free_pages((unsigned long)sring, get_order(ring_size));
1747                 rinfo->ring.sring = NULL;
1748                 goto fail;
1749         }
1750         for (i = 0; i < info->nr_ring_pages; i++)
1751                 rinfo->ring_ref[i] = gref[i];
1752
1753         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1754         if (err)
1755                 goto fail;
1756
1757         err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1758                                                 0, "blkif", rinfo);
1759         if (err <= 0) {
1760                 xenbus_dev_fatal(dev, err,
1761                                  "bind_evtchn_to_irqhandler failed");
1762                 goto fail;
1763         }
1764         rinfo->irq = err;
1765
1766         return 0;
1767 fail:
1768         blkif_free(info, 0);
1769         return err;
1770 }
1771
1772 /*
1773  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1774  * ring buffer may have multi pages depending on ->nr_ring_pages.
1775  */
1776 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1777                                 struct blkfront_ring_info *rinfo, const char *dir)
1778 {
1779         int err;
1780         unsigned int i;
1781         const char *message = NULL;
1782         struct blkfront_info *info = rinfo->dev_info;
1783
1784         if (info->nr_ring_pages == 1) {
1785                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1786                 if (err) {
1787                         message = "writing ring-ref";
1788                         goto abort_transaction;
1789                 }
1790         } else {
1791                 for (i = 0; i < info->nr_ring_pages; i++) {
1792                         char ring_ref_name[RINGREF_NAME_LEN];
1793
1794                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1795                         err = xenbus_printf(xbt, dir, ring_ref_name,
1796                                             "%u", rinfo->ring_ref[i]);
1797                         if (err) {
1798                                 message = "writing ring-ref";
1799                                 goto abort_transaction;
1800                         }
1801                 }
1802         }
1803
1804         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1805         if (err) {
1806                 message = "writing event-channel";
1807                 goto abort_transaction;
1808         }
1809
1810         return 0;
1811
1812 abort_transaction:
1813         xenbus_transaction_end(xbt, 1);
1814         if (message)
1815                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1816
1817         return err;
1818 }
1819
1820 /* Common code used when first setting up, and when resuming. */
1821 static int talk_to_blkback(struct xenbus_device *dev,
1822                            struct blkfront_info *info)
1823 {
1824         const char *message = NULL;
1825         struct xenbus_transaction xbt;
1826         int err;
1827         unsigned int i, max_page_order;
1828         unsigned int ring_page_order;
1829
1830         if (!info)
1831                 return -ENODEV;
1832
1833         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1834                                               "max-ring-page-order", 0);
1835         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1836         info->nr_ring_pages = 1 << ring_page_order;
1837
1838         err = negotiate_mq(info);
1839         if (err)
1840                 goto destroy_blkring;
1841
1842         for (i = 0; i < info->nr_rings; i++) {
1843                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1844
1845                 /* Create shared ring, alloc event channel. */
1846                 err = setup_blkring(dev, rinfo);
1847                 if (err)
1848                         goto destroy_blkring;
1849         }
1850
1851 again:
1852         err = xenbus_transaction_start(&xbt);
1853         if (err) {
1854                 xenbus_dev_fatal(dev, err, "starting transaction");
1855                 goto destroy_blkring;
1856         }
1857
1858         if (info->nr_ring_pages > 1) {
1859                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1860                                     ring_page_order);
1861                 if (err) {
1862                         message = "writing ring-page-order";
1863                         goto abort_transaction;
1864                 }
1865         }
1866
1867         /* We already got the number of queues/rings in _probe */
1868         if (info->nr_rings == 1) {
1869                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1870                 if (err)
1871                         goto destroy_blkring;
1872         } else {
1873                 char *path;
1874                 size_t pathsize;
1875
1876                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1877                                     info->nr_rings);
1878                 if (err) {
1879                         message = "writing multi-queue-num-queues";
1880                         goto abort_transaction;
1881                 }
1882
1883                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1884                 path = kmalloc(pathsize, GFP_KERNEL);
1885                 if (!path) {
1886                         err = -ENOMEM;
1887                         message = "ENOMEM while writing ring references";
1888                         goto abort_transaction;
1889                 }
1890
1891                 for (i = 0; i < info->nr_rings; i++) {
1892                         memset(path, 0, pathsize);
1893                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1894                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1895                         if (err) {
1896                                 kfree(path);
1897                                 goto destroy_blkring;
1898                         }
1899                 }
1900                 kfree(path);
1901         }
1902         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1903                             XEN_IO_PROTO_ABI_NATIVE);
1904         if (err) {
1905                 message = "writing protocol";
1906                 goto abort_transaction;
1907         }
1908         err = xenbus_printf(xbt, dev->nodename,
1909                             "feature-persistent", "%u", 1);
1910         if (err)
1911                 dev_warn(&dev->dev,
1912                          "writing persistent grants feature to xenbus");
1913
1914         err = xenbus_transaction_end(xbt, 0);
1915         if (err) {
1916                 if (err == -EAGAIN)
1917                         goto again;
1918                 xenbus_dev_fatal(dev, err, "completing transaction");
1919                 goto destroy_blkring;
1920         }
1921
1922         for (i = 0; i < info->nr_rings; i++) {
1923                 unsigned int j;
1924                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1925
1926                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1927                         rinfo->shadow[j].req.u.rw.id = j + 1;
1928                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1929         }
1930         xenbus_switch_state(dev, XenbusStateInitialised);
1931
1932         return 0;
1933
1934  abort_transaction:
1935         xenbus_transaction_end(xbt, 1);
1936         if (message)
1937                 xenbus_dev_fatal(dev, err, "%s", message);
1938  destroy_blkring:
1939         blkif_free(info, 0);
1940
1941         kfree(info);
1942         dev_set_drvdata(&dev->dev, NULL);
1943
1944         return err;
1945 }
1946
1947 static int negotiate_mq(struct blkfront_info *info)
1948 {
1949         unsigned int backend_max_queues;
1950         unsigned int i;
1951
1952         BUG_ON(info->nr_rings);
1953
1954         /* Check if backend supports multiple queues. */
1955         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1956                                                   "multi-queue-max-queues", 1);
1957         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1958         /* We need at least one ring. */
1959         if (!info->nr_rings)
1960                 info->nr_rings = 1;
1961
1962         info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1963         if (!info->rinfo) {
1964                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1965                 info->nr_rings = 0;
1966                 return -ENOMEM;
1967         }
1968
1969         for (i = 0; i < info->nr_rings; i++) {
1970                 struct blkfront_ring_info *rinfo;
1971
1972                 rinfo = &info->rinfo[i];
1973                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1974                 INIT_LIST_HEAD(&rinfo->grants);
1975                 rinfo->dev_info = info;
1976                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1977                 spin_lock_init(&rinfo->ring_lock);
1978         }
1979         return 0;
1980 }
1981 /**
1982  * Entry point to this code when a new device is created.  Allocate the basic
1983  * structures and the ring buffer for communication with the backend, and
1984  * inform the backend of the appropriate details for those.  Switch to
1985  * Initialised state.
1986  */
1987 static int blkfront_probe(struct xenbus_device *dev,
1988                           const struct xenbus_device_id *id)
1989 {
1990         int err, vdevice;
1991         struct blkfront_info *info;
1992
1993         /* FIXME: Use dynamic device id if this is not set. */
1994         err = xenbus_scanf(XBT_NIL, dev->nodename,
1995                            "virtual-device", "%i", &vdevice);
1996         if (err != 1) {
1997                 /* go looking in the extended area instead */
1998                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1999                                    "%i", &vdevice);
2000                 if (err != 1) {
2001                         xenbus_dev_fatal(dev, err, "reading virtual-device");
2002                         return err;
2003                 }
2004         }
2005
2006         if (xen_hvm_domain()) {
2007                 char *type;
2008                 int len;
2009                 /* no unplug has been done: do not hook devices != xen vbds */
2010                 if (xen_has_pv_and_legacy_disk_devices()) {
2011                         int major;
2012
2013                         if (!VDEV_IS_EXTENDED(vdevice))
2014                                 major = BLKIF_MAJOR(vdevice);
2015                         else
2016                                 major = XENVBD_MAJOR;
2017
2018                         if (major != XENVBD_MAJOR) {
2019                                 printk(KERN_INFO
2020                                                 "%s: HVM does not support vbd %d as xen block device\n",
2021                                                 __func__, vdevice);
2022                                 return -ENODEV;
2023                         }
2024                 }
2025                 /* do not create a PV cdrom device if we are an HVM guest */
2026                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
2027                 if (IS_ERR(type))
2028                         return -ENODEV;
2029                 if (strncmp(type, "cdrom", 5) == 0) {
2030                         kfree(type);
2031                         return -ENODEV;
2032                 }
2033                 kfree(type);
2034         }
2035         info = kzalloc(sizeof(*info), GFP_KERNEL);
2036         if (!info) {
2037                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2038                 return -ENOMEM;
2039         }
2040
2041         info->xbdev = dev;
2042
2043         mutex_init(&info->mutex);
2044         info->vdevice = vdevice;
2045         info->connected = BLKIF_STATE_DISCONNECTED;
2046
2047         /* Front end dir is a number, which is used as the id. */
2048         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2049         dev_set_drvdata(&dev->dev, info);
2050
2051         return 0;
2052 }
2053
2054 static int blkif_recover(struct blkfront_info *info)
2055 {
2056         unsigned int r_index;
2057         struct request *req, *n;
2058         int rc;
2059         struct bio *bio;
2060         unsigned int segs;
2061
2062         blkfront_gather_backend_features(info);
2063         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2064         blkif_set_queue_limits(info);
2065         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2066         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2067
2068         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2069                 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2070
2071                 rc = blkfront_setup_indirect(rinfo);
2072                 if (rc)
2073                         return rc;
2074         }
2075         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2076
2077         /* Now safe for us to use the shared ring */
2078         info->connected = BLKIF_STATE_CONNECTED;
2079
2080         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2081                 struct blkfront_ring_info *rinfo;
2082
2083                 rinfo = &info->rinfo[r_index];
2084                 /* Kick any other new requests queued since we resumed */
2085                 kick_pending_request_queues(rinfo);
2086         }
2087
2088         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2089                 /* Requeue pending requests (flush or discard) */
2090                 list_del_init(&req->queuelist);
2091                 BUG_ON(req->nr_phys_segments > segs);
2092                 blk_mq_requeue_request(req, false);
2093         }
2094         blk_mq_start_stopped_hw_queues(info->rq, true);
2095         blk_mq_kick_requeue_list(info->rq);
2096
2097         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2098                 /* Traverse the list of pending bios and re-queue them */
2099                 submit_bio(bio);
2100         }
2101
2102         return 0;
2103 }
2104
2105 /**
2106  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2107  * driver restart.  We tear down our blkif structure and recreate it, but
2108  * leave the device-layer structures intact so that this is transparent to the
2109  * rest of the kernel.
2110  */
2111 static int blkfront_resume(struct xenbus_device *dev)
2112 {
2113         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2114         int err = 0;
2115         unsigned int i, j;
2116
2117         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2118
2119         bio_list_init(&info->bio_list);
2120         INIT_LIST_HEAD(&info->requests);
2121         for (i = 0; i < info->nr_rings; i++) {
2122                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2123                 struct bio_list merge_bio;
2124                 struct blk_shadow *shadow = rinfo->shadow;
2125
2126                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2127                         /* Not in use? */
2128                         if (!shadow[j].request)
2129                                 continue;
2130
2131                         /*
2132                          * Get the bios in the request so we can re-queue them.
2133                          */
2134                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2135                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2136                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2137                             shadow[j].request->cmd_flags & REQ_FUA) {
2138                                 /*
2139                                  * Flush operations don't contain bios, so
2140                                  * we need to requeue the whole request
2141                                  *
2142                                  * XXX: but this doesn't make any sense for a
2143                                  * write with the FUA flag set..
2144                                  */
2145                                 list_add(&shadow[j].request->queuelist, &info->requests);
2146                                 continue;
2147                         }
2148                         merge_bio.head = shadow[j].request->bio;
2149                         merge_bio.tail = shadow[j].request->biotail;
2150                         bio_list_merge(&info->bio_list, &merge_bio);
2151                         shadow[j].request->bio = NULL;
2152                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2153                 }
2154         }
2155
2156         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2157
2158         err = talk_to_blkback(dev, info);
2159         if (!err)
2160                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2161
2162         /*
2163          * We have to wait for the backend to switch to
2164          * connected state, since we want to read which
2165          * features it supports.
2166          */
2167
2168         return err;
2169 }
2170
2171 static void blkfront_closing(struct blkfront_info *info)
2172 {
2173         struct xenbus_device *xbdev = info->xbdev;
2174         struct block_device *bdev = NULL;
2175
2176         mutex_lock(&info->mutex);
2177
2178         if (xbdev->state == XenbusStateClosing) {
2179                 mutex_unlock(&info->mutex);
2180                 return;
2181         }
2182
2183         if (info->gd)
2184                 bdev = bdget_disk(info->gd, 0);
2185
2186         mutex_unlock(&info->mutex);
2187
2188         if (!bdev) {
2189                 xenbus_frontend_closed(xbdev);
2190                 return;
2191         }
2192
2193         mutex_lock(&bdev->bd_mutex);
2194
2195         if (bdev->bd_openers) {
2196                 xenbus_dev_error(xbdev, -EBUSY,
2197                                  "Device in use; refusing to close");
2198                 xenbus_switch_state(xbdev, XenbusStateClosing);
2199         } else {
2200                 xlvbd_release_gendisk(info);
2201                 xenbus_frontend_closed(xbdev);
2202         }
2203
2204         mutex_unlock(&bdev->bd_mutex);
2205         bdput(bdev);
2206 }
2207
2208 static void blkfront_setup_discard(struct blkfront_info *info)
2209 {
2210         info->feature_discard = 1;
2211         info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2212                                                          "discard-granularity",
2213                                                          0);
2214         info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2215                                                        "discard-alignment", 0);
2216         info->feature_secdiscard =
2217                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2218                                        0);
2219 }
2220
2221 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2222 {
2223         unsigned int psegs, grants;
2224         int err, i;
2225         struct blkfront_info *info = rinfo->dev_info;
2226
2227         if (info->max_indirect_segments == 0) {
2228                 if (!HAS_EXTRA_REQ)
2229                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2230                 else {
2231                         /*
2232                          * When an extra req is required, the maximum
2233                          * grants supported is related to the size of the
2234                          * Linux block segment.
2235                          */
2236                         grants = GRANTS_PER_PSEG;
2237                 }
2238         }
2239         else
2240                 grants = info->max_indirect_segments;
2241         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2242
2243         err = fill_grant_buffer(rinfo,
2244                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2245         if (err)
2246                 goto out_of_memory;
2247
2248         if (!info->feature_persistent && info->max_indirect_segments) {
2249                 /*
2250                  * We are using indirect descriptors but not persistent
2251                  * grants, we need to allocate a set of pages that can be
2252                  * used for mapping indirect grefs
2253                  */
2254                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2255
2256                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2257                 for (i = 0; i < num; i++) {
2258                         struct page *indirect_page = alloc_page(GFP_NOIO);
2259                         if (!indirect_page)
2260                                 goto out_of_memory;
2261                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2262                 }
2263         }
2264
2265         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2266                 rinfo->shadow[i].grants_used = kzalloc(
2267                         sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2268                         GFP_NOIO);
2269                 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2270                 if (info->max_indirect_segments)
2271                         rinfo->shadow[i].indirect_grants = kzalloc(
2272                                 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2273                                 INDIRECT_GREFS(grants),
2274                                 GFP_NOIO);
2275                 if ((rinfo->shadow[i].grants_used == NULL) ||
2276                         (rinfo->shadow[i].sg == NULL) ||
2277                      (info->max_indirect_segments &&
2278                      (rinfo->shadow[i].indirect_grants == NULL)))
2279                         goto out_of_memory;
2280                 sg_init_table(rinfo->shadow[i].sg, psegs);
2281         }
2282
2283
2284         return 0;
2285
2286 out_of_memory:
2287         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2288                 kfree(rinfo->shadow[i].grants_used);
2289                 rinfo->shadow[i].grants_used = NULL;
2290                 kfree(rinfo->shadow[i].sg);
2291                 rinfo->shadow[i].sg = NULL;
2292                 kfree(rinfo->shadow[i].indirect_grants);
2293                 rinfo->shadow[i].indirect_grants = NULL;
2294         }
2295         if (!list_empty(&rinfo->indirect_pages)) {
2296                 struct page *indirect_page, *n;
2297                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2298                         list_del(&indirect_page->lru);
2299                         __free_page(indirect_page);
2300                 }
2301         }
2302         return -ENOMEM;
2303 }
2304
2305 /*
2306  * Gather all backend feature-*
2307  */
2308 static void blkfront_gather_backend_features(struct blkfront_info *info)
2309 {
2310         unsigned int indirect_segments;
2311
2312         info->feature_flush = 0;
2313         info->feature_fua = 0;
2314
2315         /*
2316          * If there's no "feature-barrier" defined, then it means
2317          * we're dealing with a very old backend which writes
2318          * synchronously; nothing to do.
2319          *
2320          * If there are barriers, then we use flush.
2321          */
2322         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2323                 info->feature_flush = 1;
2324                 info->feature_fua = 1;
2325         }
2326
2327         /*
2328          * And if there is "feature-flush-cache" use that above
2329          * barriers.
2330          */
2331         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2332                                  0)) {
2333                 info->feature_flush = 1;
2334                 info->feature_fua = 0;
2335         }
2336
2337         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2338                 blkfront_setup_discard(info);
2339
2340         info->feature_persistent =
2341                 !!xenbus_read_unsigned(info->xbdev->otherend,
2342                                        "feature-persistent", 0);
2343
2344         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2345                                         "feature-max-indirect-segments", 0);
2346         if (indirect_segments > xen_blkif_max_segments)
2347                 indirect_segments = xen_blkif_max_segments;
2348         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2349                 indirect_segments = 0;
2350         info->max_indirect_segments = indirect_segments;
2351 }
2352
2353 /*
2354  * Invoked when the backend is finally 'ready' (and has told produced
2355  * the details about the physical device - #sectors, size, etc).
2356  */
2357 static void blkfront_connect(struct blkfront_info *info)
2358 {
2359         unsigned long long sectors;
2360         unsigned long sector_size;
2361         unsigned int physical_sector_size;
2362         unsigned int binfo;
2363         char *envp[] = { "RESIZE=1", NULL };
2364         int err, i;
2365
2366         switch (info->connected) {
2367         case BLKIF_STATE_CONNECTED:
2368                 /*
2369                  * Potentially, the back-end may be signalling
2370                  * a capacity change; update the capacity.
2371                  */
2372                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2373                                    "sectors", "%Lu", &sectors);
2374                 if (XENBUS_EXIST_ERR(err))
2375                         return;
2376                 printk(KERN_INFO "Setting capacity to %Lu\n",
2377                        sectors);
2378                 set_capacity(info->gd, sectors);
2379                 revalidate_disk(info->gd);
2380                 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2381                                    KOBJ_CHANGE, envp);
2382
2383                 return;
2384         case BLKIF_STATE_SUSPENDED:
2385                 /*
2386                  * If we are recovering from suspension, we need to wait
2387                  * for the backend to announce it's features before
2388                  * reconnecting, at least we need to know if the backend
2389                  * supports indirect descriptors, and how many.
2390                  */
2391                 blkif_recover(info);
2392                 return;
2393
2394         default:
2395                 break;
2396         }
2397
2398         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2399                 __func__, info->xbdev->otherend);
2400
2401         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2402                             "sectors", "%llu", &sectors,
2403                             "info", "%u", &binfo,
2404                             "sector-size", "%lu", &sector_size,
2405                             NULL);
2406         if (err) {
2407                 xenbus_dev_fatal(info->xbdev, err,
2408                                  "reading backend fields at %s",
2409                                  info->xbdev->otherend);
2410                 return;
2411         }
2412
2413         /*
2414          * physcial-sector-size is a newer field, so old backends may not
2415          * provide this. Assume physical sector size to be the same as
2416          * sector_size in that case.
2417          */
2418         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2419                                                     "physical-sector-size",
2420                                                     sector_size);
2421         blkfront_gather_backend_features(info);
2422         for (i = 0; i < info->nr_rings; i++) {
2423                 err = blkfront_setup_indirect(&info->rinfo[i]);
2424                 if (err) {
2425                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2426                                          info->xbdev->otherend);
2427                         blkif_free(info, 0);
2428                         break;
2429                 }
2430         }
2431
2432         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2433                                   physical_sector_size);
2434         if (err) {
2435                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2436                                  info->xbdev->otherend);
2437                 goto fail;
2438         }
2439
2440         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2441
2442         /* Kick pending requests. */
2443         info->connected = BLKIF_STATE_CONNECTED;
2444         for (i = 0; i < info->nr_rings; i++)
2445                 kick_pending_request_queues(&info->rinfo[i]);
2446
2447         device_add_disk(&info->xbdev->dev, info->gd);
2448
2449         info->is_ready = 1;
2450         return;
2451
2452 fail:
2453         blkif_free(info, 0);
2454         return;
2455 }
2456
2457 /**
2458  * Callback received when the backend's state changes.
2459  */
2460 static void blkback_changed(struct xenbus_device *dev,
2461                             enum xenbus_state backend_state)
2462 {
2463         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2464
2465         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2466
2467         switch (backend_state) {
2468         case XenbusStateInitWait:
2469                 if (dev->state != XenbusStateInitialising)
2470                         break;
2471                 if (talk_to_blkback(dev, info))
2472                         break;
2473         case XenbusStateInitialising:
2474         case XenbusStateInitialised:
2475         case XenbusStateReconfiguring:
2476         case XenbusStateReconfigured:
2477         case XenbusStateUnknown:
2478                 break;
2479
2480         case XenbusStateConnected:
2481                 /*
2482                  * talk_to_blkback sets state to XenbusStateInitialised
2483                  * and blkfront_connect sets it to XenbusStateConnected
2484                  * (if connection went OK).
2485                  *
2486                  * If the backend (or toolstack) decides to poke at backend
2487                  * state (and re-trigger the watch by setting the state repeatedly
2488                  * to XenbusStateConnected (4)) we need to deal with this.
2489                  * This is allowed as this is used to communicate to the guest
2490                  * that the size of disk has changed!
2491                  */
2492                 if ((dev->state != XenbusStateInitialised) &&
2493                     (dev->state != XenbusStateConnected)) {
2494                         if (talk_to_blkback(dev, info))
2495                                 break;
2496                 }
2497
2498                 blkfront_connect(info);
2499                 break;
2500
2501         case XenbusStateClosed:
2502                 if (dev->state == XenbusStateClosed)
2503                         break;
2504                 /* fall through */
2505         case XenbusStateClosing:
2506                 if (info)
2507                         blkfront_closing(info);
2508                 break;
2509         }
2510 }
2511
2512 static int blkfront_remove(struct xenbus_device *xbdev)
2513 {
2514         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2515         struct block_device *bdev = NULL;
2516         struct gendisk *disk;
2517
2518         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2519
2520         if (!info)
2521                 return 0;
2522
2523         blkif_free(info, 0);
2524
2525         mutex_lock(&info->mutex);
2526
2527         disk = info->gd;
2528         if (disk)
2529                 bdev = bdget_disk(disk, 0);
2530
2531         info->xbdev = NULL;
2532         mutex_unlock(&info->mutex);
2533
2534         if (!bdev) {
2535                 kfree(info);
2536                 return 0;
2537         }
2538
2539         /*
2540          * The xbdev was removed before we reached the Closed
2541          * state. See if it's safe to remove the disk. If the bdev
2542          * isn't closed yet, we let release take care of it.
2543          */
2544
2545         mutex_lock(&bdev->bd_mutex);
2546         info = disk->private_data;
2547
2548         dev_warn(disk_to_dev(disk),
2549                  "%s was hot-unplugged, %d stale handles\n",
2550                  xbdev->nodename, bdev->bd_openers);
2551
2552         if (info && !bdev->bd_openers) {
2553                 xlvbd_release_gendisk(info);
2554                 disk->private_data = NULL;
2555                 kfree(info);
2556         }
2557
2558         mutex_unlock(&bdev->bd_mutex);
2559         bdput(bdev);
2560
2561         return 0;
2562 }
2563
2564 static int blkfront_is_ready(struct xenbus_device *dev)
2565 {
2566         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2567
2568         return info->is_ready && info->xbdev;
2569 }
2570
2571 static int blkif_open(struct block_device *bdev, fmode_t mode)
2572 {
2573         struct gendisk *disk = bdev->bd_disk;
2574         struct blkfront_info *info;
2575         int err = 0;
2576
2577         mutex_lock(&blkfront_mutex);
2578
2579         info = disk->private_data;
2580         if (!info) {
2581                 /* xbdev gone */
2582                 err = -ERESTARTSYS;
2583                 goto out;
2584         }
2585
2586         mutex_lock(&info->mutex);
2587
2588         if (!info->gd)
2589                 /* xbdev is closed */
2590                 err = -ERESTARTSYS;
2591
2592         mutex_unlock(&info->mutex);
2593
2594 out:
2595         mutex_unlock(&blkfront_mutex);
2596         return err;
2597 }
2598
2599 static void blkif_release(struct gendisk *disk, fmode_t mode)
2600 {
2601         struct blkfront_info *info = disk->private_data;
2602         struct block_device *bdev;
2603         struct xenbus_device *xbdev;
2604
2605         mutex_lock(&blkfront_mutex);
2606
2607         bdev = bdget_disk(disk, 0);
2608
2609         if (!bdev) {
2610                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2611                 goto out_mutex;
2612         }
2613         if (bdev->bd_openers)
2614                 goto out;
2615
2616         /*
2617          * Check if we have been instructed to close. We will have
2618          * deferred this request, because the bdev was still open.
2619          */
2620
2621         mutex_lock(&info->mutex);
2622         xbdev = info->xbdev;
2623
2624         if (xbdev && xbdev->state == XenbusStateClosing) {
2625                 /* pending switch to state closed */
2626                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2627                 xlvbd_release_gendisk(info);
2628                 xenbus_frontend_closed(info->xbdev);
2629         }
2630
2631         mutex_unlock(&info->mutex);
2632
2633         if (!xbdev) {
2634                 /* sudden device removal */
2635                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2636                 xlvbd_release_gendisk(info);
2637                 disk->private_data = NULL;
2638                 kfree(info);
2639         }
2640
2641 out:
2642         bdput(bdev);
2643 out_mutex:
2644         mutex_unlock(&blkfront_mutex);
2645 }
2646
2647 static const struct block_device_operations xlvbd_block_fops =
2648 {
2649         .owner = THIS_MODULE,
2650         .open = blkif_open,
2651         .release = blkif_release,
2652         .getgeo = blkif_getgeo,
2653         .ioctl = blkif_ioctl,
2654 };
2655
2656
2657 static const struct xenbus_device_id blkfront_ids[] = {
2658         { "vbd" },
2659         { "" }
2660 };
2661
2662 static struct xenbus_driver blkfront_driver = {
2663         .ids  = blkfront_ids,
2664         .probe = blkfront_probe,
2665         .remove = blkfront_remove,
2666         .resume = blkfront_resume,
2667         .otherend_changed = blkback_changed,
2668         .is_ready = blkfront_is_ready,
2669 };
2670
2671 static int __init xlblk_init(void)
2672 {
2673         int ret;
2674         int nr_cpus = num_online_cpus();
2675
2676         if (!xen_domain())
2677                 return -ENODEV;
2678
2679         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2680                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2681
2682         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2683                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2684                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2685                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2686         }
2687
2688         if (xen_blkif_max_queues > nr_cpus) {
2689                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2690                         xen_blkif_max_queues, nr_cpus);
2691                 xen_blkif_max_queues = nr_cpus;
2692         }
2693
2694         if (!xen_has_pv_disk_devices())
2695                 return -ENODEV;
2696
2697         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2698                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2699                        XENVBD_MAJOR, DEV_NAME);
2700                 return -ENODEV;
2701         }
2702
2703         ret = xenbus_register_frontend(&blkfront_driver);
2704         if (ret) {
2705                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2706                 return ret;
2707         }
2708
2709         return 0;
2710 }
2711 module_init(xlblk_init);
2712
2713
2714 static void __exit xlblk_exit(void)
2715 {
2716         xenbus_unregister_driver(&blkfront_driver);
2717         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2718         kfree(minors);
2719 }
2720 module_exit(xlblk_exit);
2721
2722 MODULE_DESCRIPTION("Xen virtual block device frontend");
2723 MODULE_LICENSE("GPL");
2724 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2725 MODULE_ALIAS("xen:vbd");
2726 MODULE_ALIAS("xenblk");