drivers/block/virtio_blk.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 //#define DEBUG
   3 #include <linux/spinlock.h>
   4 #include <linux/slab.h>
   5 #include <linux/blkdev.h>
   6 #include <linux/hdreg.h>
   7 #include <linux/module.h>
   8 #include <linux/mutex.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/virtio.h>
  11 #include <linux/virtio_blk.h>
  12 #include <linux/scatterlist.h>
  13 #include <linux/string_helpers.h>
  14 #include <linux/idr.h>
  15 #include <linux/blk-mq.h>
  16 #include <linux/blk-mq-virtio.h>
  17 #include <linux/numa.h>
  18 #include <linux/vmalloc.h>
  19 #include <uapi/linux/virtio_ring.h>
  20
  21 #define PART_BITS 4
  22 #define VQ_NAME_LEN 16
  23 #define MAX_DISCARD_SEGMENTS 256u
  24
  25 /* The maximum number of sg elements that fit into a virtqueue */
  26 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
  27
  28 #ifdef CONFIG_ARCH_NO_SG_CHAIN
  29 #define VIRTIO_BLK_INLINE_SG_CNT        0
  30 #else
  31 #define VIRTIO_BLK_INLINE_SG_CNT        2
  32 #endif
  33
  34 static unsigned int num_request_queues;
  35 module_param(num_request_queues, uint, 0644);
  36 MODULE_PARM_DESC(num_request_queues,
  37                  "Limit the number of request queues to use for blk device. "
  38                  "0 for no limit. "
  39                  "Values > nr_cpu_ids truncated to nr_cpu_ids.");
  40
  41 static unsigned int poll_queues;
  42 module_param(poll_queues, uint, 0644);
  43 MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/O");
  44
  45 static int major;
  46 static DEFINE_IDA(vd_index_ida);
  47
  48 static struct workqueue_struct *virtblk_wq;
  49
  50 struct virtio_blk_vq {
  51         struct virtqueue *vq;
  52         spinlock_t lock;
  53         char name[VQ_NAME_LEN];
  54 } ____cacheline_aligned_in_smp;
  55
  56 struct virtio_blk {
  57         /*
  58          * This mutex must be held by anything that may run after
  59          * virtblk_remove() sets vblk->vdev to NULL.
  60          *
  61          * blk-mq, virtqueue processing, and sysfs attribute code paths are
  62          * shut down before vblk->vdev is set to NULL and therefore do not need
  63          * to hold this mutex.
  64          */
  65         struct mutex vdev_mutex;
  66         struct virtio_device *vdev;
  67
  68         /* The disk structure for the kernel. */
  69         struct gendisk *disk;
  70
  71         /* Block layer tags. */
  72         struct blk_mq_tag_set tag_set;
  73
  74         /* Process context for config space updates */
  75         struct work_struct config_work;
  76
  77         /* Ida index - used to track minor number allocations. */
  78         int index;
  79
  80         /* num of vqs */
  81         int num_vqs;
  82         int io_queues[HCTX_MAX_TYPES];
  83         struct virtio_blk_vq *vqs;
  84
  85         /* For zoned device */
  86         unsigned int zone_sectors;
  87 };
  88
  89 struct virtblk_req {
  90         /* Out header */
  91         struct virtio_blk_outhdr out_hdr;
  92
  93         /* In header */
  94         union {
  95                 u8 status;
  96
  97                 /*
  98                  * The zone append command has an extended in header.
  99                  * The status field in zone_append_in_hdr must always
 100                  * be the last byte.
 101                  */
 102                 struct {
 103                         __virtio64 sector;
 104                         u8 status;
 105                 } zone_append;
 106         } in_hdr;
 107
 108         size_t in_hdr_len;
 109
 110         struct sg_table sg_table;
 111         struct scatterlist sg[];
 112 };
 113
 114 static inline blk_status_t virtblk_result(u8 status)
 115 {
 116         switch (status) {
 117         case VIRTIO_BLK_S_OK:
 118                 return BLK_STS_OK;
 119         case VIRTIO_BLK_S_UNSUPP:
 120                 return BLK_STS_NOTSUPP;
 121         case VIRTIO_BLK_S_ZONE_OPEN_RESOURCE:
 122                 return BLK_STS_ZONE_OPEN_RESOURCE;
 123         case VIRTIO_BLK_S_ZONE_ACTIVE_RESOURCE:
 124                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 125         case VIRTIO_BLK_S_IOERR:
 126         case VIRTIO_BLK_S_ZONE_UNALIGNED_WP:
 127         default:
 128                 return BLK_STS_IOERR;
 129         }
 130 }
 131
 132 static inline struct virtio_blk_vq *get_virtio_blk_vq(struct blk_mq_hw_ctx *hctx)
 133 {
 134         struct virtio_blk *vblk = hctx->queue->queuedata;
 135         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 136
 137         return vq;
 138 }
 139
 140 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr)
 141 {
 142         struct scatterlist out_hdr, in_hdr, *sgs[3];
 143         unsigned int num_out = 0, num_in = 0;
 144
 145         sg_init_one(&out_hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
 146         sgs[num_out++] = &out_hdr;
 147
 148         if (vbr->sg_table.nents) {
 149                 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
 150                         sgs[num_out++] = vbr->sg_table.sgl;
 151                 else
 152                         sgs[num_out + num_in++] = vbr->sg_table.sgl;
 153         }
 154
 155         sg_init_one(&in_hdr, &vbr->in_hdr.status, vbr->in_hdr_len);
 156         sgs[num_out + num_in++] = &in_hdr;
 157
 158         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 159 }
 160
 161 static int virtblk_setup_discard_write_zeroes_erase(struct request *req, bool unmap)
 162 {
 163         unsigned short segments = blk_rq_nr_discard_segments(req);
 164         unsigned short n = 0;
 165         struct virtio_blk_discard_write_zeroes *range;
 166         struct bio *bio;
 167         u32 flags = 0;
 168
 169         if (unmap)
 170                 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
 171
 172         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 173         if (!range)
 174                 return -ENOMEM;
 175
 176         /*
 177          * Single max discard segment means multi-range discard isn't
 178          * supported, and block layer only runs contiguity merge like
 179          * normal RW request. So we can't reply on bio for retrieving
 180          * each range info.
 181          */
 182         if (queue_max_discard_segments(req->q) == 1) {
 183                 range[0].flags = cpu_to_le32(flags);
 184                 range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
 185                 range[0].sector = cpu_to_le64(blk_rq_pos(req));
 186                 n = 1;
 187         } else {
 188                 __rq_for_each_bio(bio, req) {
 189                         u64 sector = bio->bi_iter.bi_sector;
 190                         u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
 191
 192                         range[n].flags = cpu_to_le32(flags);
 193                         range[n].num_sectors = cpu_to_le32(num_sectors);
 194                         range[n].sector = cpu_to_le64(sector);
 195                         n++;
 196                 }
 197         }
 198
 199         WARN_ON_ONCE(n != segments);
 200
 201         bvec_set_virt(&req->special_vec, range, sizeof(*range) * segments);
 202         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 203
 204         return 0;
 205 }
 206
 207 static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
 208 {
 209         if (blk_rq_nr_phys_segments(req))
 210                 sg_free_table_chained(&vbr->sg_table,
 211                                       VIRTIO_BLK_INLINE_SG_CNT);
 212 }
 213
 214 static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
 215                 struct virtblk_req *vbr)
 216 {
 217         int err;
 218
 219         if (!blk_rq_nr_phys_segments(req))
 220                 return 0;
 221
 222         vbr->sg_table.sgl = vbr->sg;
 223         err = sg_alloc_table_chained(&vbr->sg_table,
 224                                      blk_rq_nr_phys_segments(req),
 225                                      vbr->sg_table.sgl,
 226                                      VIRTIO_BLK_INLINE_SG_CNT);
 227         if (unlikely(err))
 228                 return -ENOMEM;
 229
 230         return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl);
 231 }
 232
 233 static void virtblk_cleanup_cmd(struct request *req)
 234 {
 235         if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
 236                 kfree(bvec_virt(&req->special_vec));
 237 }
 238
 239 static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 240                                       struct request *req,
 241                                       struct virtblk_req *vbr)
 242 {
 243         size_t in_hdr_len = sizeof(vbr->in_hdr.status);
 244         bool unmap = false;
 245         u32 type;
 246         u64 sector = 0;
 247
 248         if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && op_is_zone_mgmt(req_op(req)))
 249                 return BLK_STS_NOTSUPP;
 250
 251         /* Set fields for all request types */
 252         vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
 253
 254         switch (req_op(req)) {
 255         case REQ_OP_READ:
 256                 type = VIRTIO_BLK_T_IN;
 257                 sector = blk_rq_pos(req);
 258                 break;
 259         case REQ_OP_WRITE:
 260                 type = VIRTIO_BLK_T_OUT;
 261                 sector = blk_rq_pos(req);
 262                 break;
 263         case REQ_OP_FLUSH:
 264                 type = VIRTIO_BLK_T_FLUSH;
 265                 break;
 266         case REQ_OP_DISCARD:
 267                 type = VIRTIO_BLK_T_DISCARD;
 268                 break;
 269         case REQ_OP_WRITE_ZEROES:
 270                 type = VIRTIO_BLK_T_WRITE_ZEROES;
 271                 unmap = !(req->cmd_flags & REQ_NOUNMAP);
 272                 break;
 273         case REQ_OP_SECURE_ERASE:
 274                 type = VIRTIO_BLK_T_SECURE_ERASE;
 275                 break;
 276         case REQ_OP_ZONE_OPEN:
 277                 type = VIRTIO_BLK_T_ZONE_OPEN;
 278                 sector = blk_rq_pos(req);
 279                 break;
 280         case REQ_OP_ZONE_CLOSE:
 281                 type = VIRTIO_BLK_T_ZONE_CLOSE;
 282                 sector = blk_rq_pos(req);
 283                 break;
 284         case REQ_OP_ZONE_FINISH:
 285                 type = VIRTIO_BLK_T_ZONE_FINISH;
 286                 sector = blk_rq_pos(req);
 287                 break;
 288         case REQ_OP_ZONE_APPEND:
 289                 type = VIRTIO_BLK_T_ZONE_APPEND;
 290                 sector = blk_rq_pos(req);
 291                 in_hdr_len = sizeof(vbr->in_hdr.zone_append);
 292                 break;
 293         case REQ_OP_ZONE_RESET:
 294                 type = VIRTIO_BLK_T_ZONE_RESET;
 295                 sector = blk_rq_pos(req);
 296                 break;
 297         case REQ_OP_ZONE_RESET_ALL:
 298                 type = VIRTIO_BLK_T_ZONE_RESET_ALL;
 299                 break;
 300         case REQ_OP_DRV_IN:
 301                 /*
 302                  * Out header has already been prepared by the caller (virtblk_get_id()
 303                  * or virtblk_submit_zone_report()), nothing to do here.
 304                  */
 305                 return 0;
 306         default:
 307                 WARN_ON_ONCE(1);
 308                 return BLK_STS_IOERR;
 309         }
 310
 311         /* Set fields for non-REQ_OP_DRV_IN request types */
 312         vbr->in_hdr_len = in_hdr_len;
 313         vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
 314         vbr->out_hdr.sector = cpu_to_virtio64(vdev, sector);
 315
 316         if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
 317             type == VIRTIO_BLK_T_SECURE_ERASE) {
 318                 if (virtblk_setup_discard_write_zeroes_erase(req, unmap))
 319                         return BLK_STS_RESOURCE;
 320         }
 321
 322         return 0;
 323 }
 324
 325 /*
 326  * The status byte is always the last byte of the virtblk request
 327  * in-header. This helper fetches its value for all in-header formats
 328  * that are currently defined.
 329  */
 330 static inline u8 virtblk_vbr_status(struct virtblk_req *vbr)
 331 {
 332         return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1);
 333 }
 334
 335 static inline void virtblk_request_done(struct request *req)
 336 {
 337         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 338         blk_status_t status = virtblk_result(virtblk_vbr_status(vbr));
 339         struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
 340
 341         virtblk_unmap_data(req, vbr);
 342         virtblk_cleanup_cmd(req);
 343
 344         if (req_op(req) == REQ_OP_ZONE_APPEND)
 345                 req->__sector = virtio64_to_cpu(vblk->vdev,
 346                                                 vbr->in_hdr.zone_append.sector);
 347
 348         blk_mq_end_request(req, status);
 349 }
 350
 351 static void virtblk_done(struct virtqueue *vq)
 352 {
 353         struct virtio_blk *vblk = vq->vdev->priv;
 354         bool req_done = false;
 355         int qid = vq->index;
 356         struct virtblk_req *vbr;
 357         unsigned long flags;
 358         unsigned int len;
 359
 360         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 361         do {
 362                 virtqueue_disable_cb(vq);
 363                 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
 364                         struct request *req = blk_mq_rq_from_pdu(vbr);
 365
 366                         if (likely(!blk_should_fake_timeout(req->q)))
 367                                 blk_mq_complete_request(req);
 368                         req_done = true;
 369                 }
 370                 if (unlikely(virtqueue_is_broken(vq)))
 371                         break;
 372         } while (!virtqueue_enable_cb(vq));
 373
 374         /* In case queue is stopped waiting for more buffers. */
 375         if (req_done)
 376                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 377         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 378 }
 379
 380 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 381 {
 382         struct virtio_blk *vblk = hctx->queue->queuedata;
 383         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 384         bool kick;
 385
 386         spin_lock_irq(&vq->lock);
 387         kick = virtqueue_kick_prepare(vq->vq);
 388         spin_unlock_irq(&vq->lock);
 389
 390         if (kick)
 391                 virtqueue_notify(vq->vq);
 392 }
 393
 394 static blk_status_t virtblk_fail_to_queue(struct request *req, int rc)
 395 {
 396         virtblk_cleanup_cmd(req);
 397         switch (rc) {
 398         case -ENOSPC:
 399                 return BLK_STS_DEV_RESOURCE;
 400         case -ENOMEM:
 401                 return BLK_STS_RESOURCE;
 402         default:
 403                 return BLK_STS_IOERR;
 404         }
 405 }
 406
 407 static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
 408                                         struct virtio_blk *vblk,
 409                                         struct request *req,
 410                                         struct virtblk_req *vbr)
 411 {
 412         blk_status_t status;
 413         int num;
 414
 415         status = virtblk_setup_cmd(vblk->vdev, req, vbr);
 416         if (unlikely(status))
 417                 return status;
 418
 419         num = virtblk_map_data(hctx, req, vbr);
 420         if (unlikely(num < 0))
 421                 return virtblk_fail_to_queue(req, -ENOMEM);
 422         vbr->sg_table.nents = num;
 423
 424         blk_mq_start_request(req);
 425
 426         return BLK_STS_OK;
 427 }
 428
 429 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 430                            const struct blk_mq_queue_data *bd)
 431 {
 432         struct virtio_blk *vblk = hctx->queue->queuedata;
 433         struct request *req = bd->rq;
 434         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 435         unsigned long flags;
 436         int qid = hctx->queue_num;
 437         bool notify = false;
 438         blk_status_t status;
 439         int err;
 440
 441         status = virtblk_prep_rq(hctx, vblk, req, vbr);
 442         if (unlikely(status))
 443                 return status;
 444
 445         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 446         err = virtblk_add_req(vblk->vqs[qid].vq, vbr);
 447         if (err) {
 448                 virtqueue_kick(vblk->vqs[qid].vq);
 449                 /* Don't stop the queue if -ENOMEM: we may have failed to
 450                  * bounce the buffer due to global resource outage.
 451                  */
 452                 if (err == -ENOSPC)
 453                         blk_mq_stop_hw_queue(hctx);
 454                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 455                 virtblk_unmap_data(req, vbr);
 456                 return virtblk_fail_to_queue(req, err);
 457         }
 458
 459         if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 460                 notify = true;
 461         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 462
 463         if (notify)
 464                 virtqueue_notify(vblk->vqs[qid].vq);
 465         return BLK_STS_OK;
 466 }
 467
 468 static bool virtblk_prep_rq_batch(struct request *req)
 469 {
 470         struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
 471         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 472
 473         return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
 474 }
 475
 476 static bool virtblk_add_req_batch(struct virtio_blk_vq *vq,
 477                                         struct request **rqlist)
 478 {
 479         unsigned long flags;
 480         int err;
 481         bool kick;
 482
 483         spin_lock_irqsave(&vq->lock, flags);
 484
 485         while (!rq_list_empty(*rqlist)) {
 486                 struct request *req = rq_list_pop(rqlist);
 487                 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 488
 489                 err = virtblk_add_req(vq->vq, vbr);
 490                 if (err) {
 491                         virtblk_unmap_data(req, vbr);
 492                         virtblk_cleanup_cmd(req);
 493                         blk_mq_requeue_request(req, true);
 494                 }
 495         }
 496
 497         kick = virtqueue_kick_prepare(vq->vq);
 498         spin_unlock_irqrestore(&vq->lock, flags);
 499
 500         return kick;
 501 }
 502
 503 static void virtio_queue_rqs(struct request **rqlist)
 504 {
 505         struct request *req, *next, *prev = NULL;
 506         struct request *requeue_list = NULL;
 507
 508         rq_list_for_each_safe(rqlist, req, next) {
 509                 struct virtio_blk_vq *vq = get_virtio_blk_vq(req->mq_hctx);
 510                 bool kick;
 511
 512                 if (!virtblk_prep_rq_batch(req)) {
 513                         rq_list_move(rqlist, &requeue_list, req, prev);
 514                         req = prev;
 515                         if (!req)
 516                                 continue;
 517                 }
 518
 519                 if (!next || req->mq_hctx != next->mq_hctx) {
 520                         req->rq_next = NULL;
 521                         kick = virtblk_add_req_batch(vq, rqlist);
 522                         if (kick)
 523                                 virtqueue_notify(vq->vq);
 524
 525                         *rqlist = next;
 526                         prev = NULL;
 527                 } else
 528                         prev = req;
 529         }
 530
 531         *rqlist = requeue_list;
 532 }
 533
 534 #ifdef CONFIG_BLK_DEV_ZONED
 535 static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk,
 536                                           unsigned int nr_zones,
 537                                           size_t *buflen)
 538 {
 539         struct request_queue *q = vblk->disk->queue;
 540         size_t bufsize;
 541         void *buf;
 542
 543         nr_zones = min_t(unsigned int, nr_zones,
 544                          get_capacity(vblk->disk) >> ilog2(vblk->zone_sectors));
 545
 546         bufsize = sizeof(struct virtio_blk_zone_report) +
 547                 nr_zones * sizeof(struct virtio_blk_zone_descriptor);
 548         bufsize = min_t(size_t, bufsize,
 549                         queue_max_hw_sectors(q) << SECTOR_SHIFT);
 550         bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
 551
 552         while (bufsize >= sizeof(struct virtio_blk_zone_report)) {
 553                 buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
 554                 if (buf) {
 555                         *buflen = bufsize;
 556                         return buf;
 557                 }
 558                 bufsize >>= 1;
 559         }
 560
 561         return NULL;
 562 }
 563
 564 static int virtblk_submit_zone_report(struct virtio_blk *vblk,
 565                                        char *report_buf, size_t report_len,
 566                                        sector_t sector)
 567 {
 568         struct request_queue *q = vblk->disk->queue;
 569         struct request *req;
 570         struct virtblk_req *vbr;
 571         int err;
 572
 573         req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
 574         if (IS_ERR(req))
 575                 return PTR_ERR(req);
 576
 577         vbr = blk_mq_rq_to_pdu(req);
 578         vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
 579         vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT);
 580         vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector);
 581
 582         err = blk_rq_map_kern(q, req, report_buf, report_len, GFP_KERNEL);
 583         if (err)
 584                 goto out;
 585
 586         blk_execute_rq(req, false);
 587         err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
 588 out:
 589         blk_mq_free_request(req);
 590         return err;
 591 }
 592
 593 static int virtblk_parse_zone(struct virtio_blk *vblk,
 594                                struct virtio_blk_zone_descriptor *entry,
 595                                unsigned int idx, report_zones_cb cb, void *data)
 596 {
 597         struct blk_zone zone = { };
 598
 599         zone.start = virtio64_to_cpu(vblk->vdev, entry->z_start);
 600         if (zone.start + vblk->zone_sectors <= get_capacity(vblk->disk))
 601                 zone.len = vblk->zone_sectors;
 602         else
 603                 zone.len = get_capacity(vblk->disk) - zone.start;
 604         zone.capacity = virtio64_to_cpu(vblk->vdev, entry->z_cap);
 605         zone.wp = virtio64_to_cpu(vblk->vdev, entry->z_wp);
 606
 607         switch (entry->z_type) {
 608         case VIRTIO_BLK_ZT_SWR:
 609                 zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
 610                 break;
 611         case VIRTIO_BLK_ZT_SWP:
 612                 zone.type = BLK_ZONE_TYPE_SEQWRITE_PREF;
 613                 break;
 614         case VIRTIO_BLK_ZT_CONV:
 615                 zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
 616                 break;
 617         default:
 618                 dev_err(&vblk->vdev->dev, "zone %llu: invalid type %#x\n",
 619                         zone.start, entry->z_type);
 620                 return -EIO;
 621         }
 622
 623         switch (entry->z_state) {
 624         case VIRTIO_BLK_ZS_EMPTY:
 625                 zone.cond = BLK_ZONE_COND_EMPTY;
 626                 break;
 627         case VIRTIO_BLK_ZS_CLOSED:
 628                 zone.cond = BLK_ZONE_COND_CLOSED;
 629                 break;
 630         case VIRTIO_BLK_ZS_FULL:
 631                 zone.cond = BLK_ZONE_COND_FULL;
 632                 zone.wp = zone.start + zone.len;
 633                 break;
 634         case VIRTIO_BLK_ZS_EOPEN:
 635                 zone.cond = BLK_ZONE_COND_EXP_OPEN;
 636                 break;
 637         case VIRTIO_BLK_ZS_IOPEN:
 638                 zone.cond = BLK_ZONE_COND_IMP_OPEN;
 639                 break;
 640         case VIRTIO_BLK_ZS_NOT_WP:
 641                 zone.cond = BLK_ZONE_COND_NOT_WP;
 642                 break;
 643         case VIRTIO_BLK_ZS_RDONLY:
 644                 zone.cond = BLK_ZONE_COND_READONLY;
 645                 zone.wp = ULONG_MAX;
 646                 break;
 647         case VIRTIO_BLK_ZS_OFFLINE:
 648                 zone.cond = BLK_ZONE_COND_OFFLINE;
 649                 zone.wp = ULONG_MAX;
 650                 break;
 651         default:
 652                 dev_err(&vblk->vdev->dev, "zone %llu: invalid condition %#x\n",
 653                         zone.start, entry->z_state);
 654                 return -EIO;
 655         }
 656
 657         /*
 658          * The callback below checks the validity of the reported
 659          * entry data, no need to further validate it here.
 660          */
 661         return cb(&zone, idx, data);
 662 }
 663
 664 static int virtblk_report_zones(struct gendisk *disk, sector_t sector,
 665                                  unsigned int nr_zones, report_zones_cb cb,
 666                                  void *data)
 667 {
 668         struct virtio_blk *vblk = disk->private_data;
 669         struct virtio_blk_zone_report *report;
 670         unsigned long long nz, i;
 671         size_t buflen;
 672         unsigned int zone_idx = 0;
 673         int ret;
 674
 675         if (WARN_ON_ONCE(!vblk->zone_sectors))
 676                 return -EOPNOTSUPP;
 677
 678         report = virtblk_alloc_report_buffer(vblk, nr_zones, &buflen);
 679         if (!report)
 680                 return -ENOMEM;
 681
 682         mutex_lock(&vblk->vdev_mutex);
 683
 684         if (!vblk->vdev) {
 685                 ret = -ENXIO;
 686                 goto fail_report;
 687         }
 688
 689         while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) {
 690                 memset(report, 0, buflen);
 691
 692                 ret = virtblk_submit_zone_report(vblk, (char *)report,
 693                                                  buflen, sector);
 694                 if (ret)
 695                         goto fail_report;
 696
 697                 nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones),
 698                            nr_zones);
 699                 if (!nz)
 700                         break;
 701
 702                 for (i = 0; i < nz && zone_idx < nr_zones; i++) {
 703                         ret = virtblk_parse_zone(vblk, &report->zones[i],
 704                                                  zone_idx, cb, data);
 705                         if (ret)
 706                                 goto fail_report;
 707
 708                         sector = virtio64_to_cpu(vblk->vdev,
 709                                                  report->zones[i].z_start) +
 710                                  vblk->zone_sectors;
 711                         zone_idx++;
 712                 }
 713         }
 714
 715         if (zone_idx > 0)
 716                 ret = zone_idx;
 717         else
 718                 ret = -EINVAL;
 719 fail_report:
 720         mutex_unlock(&vblk->vdev_mutex);
 721         kvfree(report);
 722         return ret;
 723 }
 724
 725 static int virtblk_probe_zoned_device(struct virtio_device *vdev,
 726                                        struct virtio_blk *vblk,
 727                                        struct request_queue *q)
 728 {
 729         u32 v, wg;
 730
 731         dev_dbg(&vdev->dev, "probing host-managed zoned device\n");
 732
 733         disk_set_zoned(vblk->disk);
 734         blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
 735
 736         virtio_cread(vdev, struct virtio_blk_config,
 737                      zoned.max_open_zones, &v);
 738         disk_set_max_open_zones(vblk->disk, v);
 739         dev_dbg(&vdev->dev, "max open zones = %u\n", v);
 740
 741         virtio_cread(vdev, struct virtio_blk_config,
 742                      zoned.max_active_zones, &v);
 743         disk_set_max_active_zones(vblk->disk, v);
 744         dev_dbg(&vdev->dev, "max active zones = %u\n", v);
 745
 746         virtio_cread(vdev, struct virtio_blk_config,
 747                      zoned.write_granularity, &wg);
 748         if (!wg) {
 749                 dev_warn(&vdev->dev, "zero write granularity reported\n");
 750                 return -ENODEV;
 751         }
 752         blk_queue_physical_block_size(q, wg);
 753         blk_queue_io_min(q, wg);
 754
 755         dev_dbg(&vdev->dev, "write granularity = %u\n", wg);
 756
 757         /*
 758          * virtio ZBD specification doesn't require zones to be a power of
 759          * two sectors in size, but the code in this driver expects that.
 760          */
 761         virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors,
 762                      &vblk->zone_sectors);
 763         if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) {
 764                 dev_err(&vdev->dev,
 765                         "zoned device with non power of two zone size %u\n",
 766                         vblk->zone_sectors);
 767                 return -ENODEV;
 768         }
 769         blk_queue_chunk_sectors(q, vblk->zone_sectors);
 770         dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors);
 771
 772         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
 773                 dev_warn(&vblk->vdev->dev,
 774                          "ignoring negotiated F_DISCARD for zoned device\n");
 775                 blk_queue_max_discard_sectors(q, 0);
 776         }
 777
 778         virtio_cread(vdev, struct virtio_blk_config,
 779                      zoned.max_append_sectors, &v);
 780         if (!v) {
 781                 dev_warn(&vdev->dev, "zero max_append_sectors reported\n");
 782                 return -ENODEV;
 783         }
 784         if ((v << SECTOR_SHIFT) < wg) {
 785                 dev_err(&vdev->dev,
 786                         "write granularity %u exceeds max_append_sectors %u limit\n",
 787                         wg, v);
 788                 return -ENODEV;
 789         }
 790         blk_queue_max_zone_append_sectors(q, v);
 791         dev_dbg(&vdev->dev, "max append sectors = %u\n", v);
 792
 793         return blk_revalidate_disk_zones(vblk->disk, NULL);
 794 }
 795
 796 #else
 797
 798 /*
 799  * Zoned block device support is not configured in this kernel.
 800  * Host-managed zoned devices can't be supported, but others are
 801  * good to go as regular block devices.
 802  */
 803 #define virtblk_report_zones       NULL
 804
 805 static inline int virtblk_probe_zoned_device(struct virtio_device *vdev,
 806                         struct virtio_blk *vblk, struct request_queue *q)
 807 {
 808         dev_err(&vdev->dev,
 809                 "virtio_blk: zoned devices are not supported");
 810         return -EOPNOTSUPP;
 811 }
 812 #endif /* CONFIG_BLK_DEV_ZONED */
 813
 814 /* return id (s/n) string for *disk to *id_str
 815  */
 816 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 817 {
 818         struct virtio_blk *vblk = disk->private_data;
 819         struct request_queue *q = vblk->disk->queue;
 820         struct request *req;
 821         struct virtblk_req *vbr;
 822         int err;
 823
 824         req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
 825         if (IS_ERR(req))
 826                 return PTR_ERR(req);
 827
 828         vbr = blk_mq_rq_to_pdu(req);
 829         vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
 830         vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
 831         vbr->out_hdr.sector = 0;
 832
 833         err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 834         if (err)
 835                 goto out;
 836
 837         blk_execute_rq(req, false);
 838         err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
 839 out:
 840         blk_mq_free_request(req);
 841         return err;
 842 }
 843
 844 /* We provide getgeo only to please some old bootloader/partitioning tools */
 845 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 846 {
 847         struct virtio_blk *vblk = bd->bd_disk->private_data;
 848         int ret = 0;
 849
 850         mutex_lock(&vblk->vdev_mutex);
 851
 852         if (!vblk->vdev) {
 853                 ret = -ENXIO;
 854                 goto out;
 855         }
 856
 857         /* see if the host passed in geometry config */
 858         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 859                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 860                              geometry.cylinders, &geo->cylinders);
 861                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 862                              geometry.heads, &geo->heads);
 863                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 864                              geometry.sectors, &geo->sectors);
 865         } else {
 866                 /* some standard values, similar to sd */
 867                 geo->heads = 1 << 6;
 868                 geo->sectors = 1 << 5;
 869                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 870         }
 871 out:
 872         mutex_unlock(&vblk->vdev_mutex);
 873         return ret;
 874 }
 875
 876 static void virtblk_free_disk(struct gendisk *disk)
 877 {
 878         struct virtio_blk *vblk = disk->private_data;
 879
 880         ida_free(&vd_index_ida, vblk->index);
 881         mutex_destroy(&vblk->vdev_mutex);
 882         kfree(vblk);
 883 }
 884
 885 static const struct block_device_operations virtblk_fops = {
 886         .owner          = THIS_MODULE,
 887         .getgeo         = virtblk_getgeo,
 888         .free_disk      = virtblk_free_disk,
 889         .report_zones   = virtblk_report_zones,
 890 };
 891
 892 static int index_to_minor(int index)
 893 {
 894         return index << PART_BITS;
 895 }
 896
 897 static int minor_to_index(int minor)
 898 {
 899         return minor >> PART_BITS;
 900 }
 901
 902 static ssize_t serial_show(struct device *dev,
 903                            struct device_attribute *attr, char *buf)
 904 {
 905         struct gendisk *disk = dev_to_disk(dev);
 906         int err;
 907
 908         /* sysfs gives us a PAGE_SIZE buffer */
 909         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 910
 911         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 912         err = virtblk_get_id(disk, buf);
 913         if (!err)
 914                 return strlen(buf);
 915
 916         if (err == -EIO) /* Unsupported? Make it empty. */
 917                 return 0;
 918
 919         return err;
 920 }
 921
 922 static DEVICE_ATTR_RO(serial);
 923
 924 /* The queue's logical block size must be set before calling this */
 925 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 926 {
 927         struct virtio_device *vdev = vblk->vdev;
 928         struct request_queue *q = vblk->disk->queue;
 929         char cap_str_2[10], cap_str_10[10];
 930         unsigned long long nblocks;
 931         u64 capacity;
 932
 933         /* Host must always specify the capacity. */
 934         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 935
 936         nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
 937
 938         string_get_size(nblocks, queue_logical_block_size(q),
 939                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 940         string_get_size(nblocks, queue_logical_block_size(q),
 941                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 942
 943         dev_notice(&vdev->dev,
 944                    "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
 945                    vblk->disk->disk_name,
 946                    resize ? "new size: " : "",
 947                    nblocks,
 948                    queue_logical_block_size(q),
 949                    cap_str_10,
 950                    cap_str_2);
 951
 952         set_capacity_and_notify(vblk->disk, capacity);
 953 }
 954
 955 static void virtblk_config_changed_work(struct work_struct *work)
 956 {
 957         struct virtio_blk *vblk =
 958                 container_of(work, struct virtio_blk, config_work);
 959
 960         virtblk_update_capacity(vblk, true);
 961 }
 962
 963 static void virtblk_config_changed(struct virtio_device *vdev)
 964 {
 965         struct virtio_blk *vblk = vdev->priv;
 966
 967         queue_work(virtblk_wq, &vblk->config_work);
 968 }
 969
 970 static int init_vq(struct virtio_blk *vblk)
 971 {
 972         int err;
 973         unsigned short i;
 974         vq_callback_t **callbacks;
 975         const char **names;
 976         struct virtqueue **vqs;
 977         unsigned short num_vqs;
 978         unsigned short num_poll_vqs;
 979         struct virtio_device *vdev = vblk->vdev;
 980         struct irq_affinity desc = { 0, };
 981
 982         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 983                                    struct virtio_blk_config, num_queues,
 984                                    &num_vqs);
 985         if (err)
 986                 num_vqs = 1;
 987
 988         if (!err && !num_vqs) {
 989                 dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
 990                 return -EINVAL;
 991         }
 992
 993         num_vqs = min_t(unsigned int,
 994                         min_not_zero(num_request_queues, nr_cpu_ids),
 995                         num_vqs);
 996
 997         num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
 998
 999         vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
1000         vblk->io_queues[HCTX_TYPE_READ] = 0;
1001         vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
1002
1003         dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
1004                                 vblk->io_queues[HCTX_TYPE_DEFAULT],
1005                                 vblk->io_queues[HCTX_TYPE_READ],
1006                                 vblk->io_queues[HCTX_TYPE_POLL]);
1007
1008         vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
1009         if (!vblk->vqs)
1010                 return -ENOMEM;
1011
1012         names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
1013         callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
1014         vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
1015         if (!names || !callbacks || !vqs) {
1016                 err = -ENOMEM;
1017                 goto out;
1018         }
1019
1020         for (i = 0; i < num_vqs - num_poll_vqs; i++) {
1021                 callbacks[i] = virtblk_done;
1022                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%u", i);
1023                 names[i] = vblk->vqs[i].name;
1024         }
1025
1026         for (; i < num_vqs; i++) {
1027                 callbacks[i] = NULL;
1028                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%u", i);
1029                 names[i] = vblk->vqs[i].name;
1030         }
1031
1032         /* Discover virtqueues and write information to configuration.  */
1033         err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
1034         if (err)
1035                 goto out;
1036
1037         for (i = 0; i < num_vqs; i++) {
1038                 spin_lock_init(&vblk->vqs[i].lock);
1039                 vblk->vqs[i].vq = vqs[i];
1040         }
1041         vblk->num_vqs = num_vqs;
1042
1043 out:
1044         kfree(vqs);
1045         kfree(callbacks);
1046         kfree(names);
1047         if (err)
1048                 kfree(vblk->vqs);
1049         return err;
1050 }
1051
1052 /*
1053  * Legacy naming scheme used for virtio devices.  We are stuck with it for
1054  * virtio blk but don't ever use it for any new driver.
1055  */
1056 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
1057 {
1058         const int base = 'z' - 'a' + 1;
1059         char *begin = buf + strlen(prefix);
1060         char *end = buf + buflen;
1061         char *p;
1062         int unit;
1063
1064         p = end - 1;
1065         *p = '\0';
1066         unit = base;
1067         do {
1068                 if (p == begin)
1069                         return -EINVAL;
1070                 *--p = 'a' + (index % unit);
1071                 index = (index / unit) - 1;
1072         } while (index >= 0);
1073
1074         memmove(begin, p, end - p);
1075         memcpy(buf, prefix, strlen(prefix));
1076
1077         return 0;
1078 }
1079
1080 static int virtblk_get_cache_mode(struct virtio_device *vdev)
1081 {
1082         u8 writeback;
1083         int err;
1084
1085         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
1086                                    struct virtio_blk_config, wce,
1087                                    &writeback);
1088
1089         /*
1090          * If WCE is not configurable and flush is not available,
1091          * assume no writeback cache is in use.
1092          */
1093         if (err)
1094                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
1095
1096         return writeback;
1097 }
1098
1099 static void virtblk_update_cache_mode(struct virtio_device *vdev)
1100 {
1101         u8 writeback = virtblk_get_cache_mode(vdev);
1102         struct virtio_blk *vblk = vdev->priv;
1103
1104         blk_queue_write_cache(vblk->disk->queue, writeback, false);
1105 }
1106
1107 static const char *const virtblk_cache_types[] = {
1108         "write through", "write back"
1109 };
1110
1111 static ssize_t
1112 cache_type_store(struct device *dev, struct device_attribute *attr,
1113                  const char *buf, size_t count)
1114 {
1115         struct gendisk *disk = dev_to_disk(dev);
1116         struct virtio_blk *vblk = disk->private_data;
1117         struct virtio_device *vdev = vblk->vdev;
1118         int i;
1119
1120         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
1121         i = sysfs_match_string(virtblk_cache_types, buf);
1122         if (i < 0)
1123                 return i;
1124
1125         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
1126         virtblk_update_cache_mode(vdev);
1127         return count;
1128 }
1129
1130 static ssize_t
1131 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
1132 {
1133         struct gendisk *disk = dev_to_disk(dev);
1134         struct virtio_blk *vblk = disk->private_data;
1135         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
1136
1137         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
1138         return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
1139 }
1140
1141 static DEVICE_ATTR_RW(cache_type);
1142
1143 static struct attribute *virtblk_attrs[] = {
1144         &dev_attr_serial.attr,
1145         &dev_attr_cache_type.attr,
1146         NULL,
1147 };
1148
1149 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
1150                 struct attribute *a, int n)
1151 {
1152         struct device *dev = kobj_to_dev(kobj);
1153         struct gendisk *disk = dev_to_disk(dev);
1154         struct virtio_blk *vblk = disk->private_data;
1155         struct virtio_device *vdev = vblk->vdev;
1156
1157         if (a == &dev_attr_cache_type.attr &&
1158             !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
1159                 return S_IRUGO;
1160
1161         return a->mode;
1162 }
1163
1164 static const struct attribute_group virtblk_attr_group = {
1165         .attrs = virtblk_attrs,
1166         .is_visible = virtblk_attrs_are_visible,
1167 };
1168
1169 static const struct attribute_group *virtblk_attr_groups[] = {
1170         &virtblk_attr_group,
1171         NULL,
1172 };
1173
1174 static void virtblk_map_queues(struct blk_mq_tag_set *set)
1175 {
1176         struct virtio_blk *vblk = set->driver_data;
1177         int i, qoff;
1178
1179         for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1180                 struct blk_mq_queue_map *map = &set->map[i];
1181
1182                 map->nr_queues = vblk->io_queues[i];
1183                 map->queue_offset = qoff;
1184                 qoff += map->nr_queues;
1185
1186                 if (map->nr_queues == 0)
1187                         continue;
1188
1189                 /*
1190                  * Regular queues have interrupts and hence CPU affinity is
1191                  * defined by the core virtio code, but polling queues have
1192                  * no interrupts so we let the block layer assign CPU affinity.
1193                  */
1194                 if (i == HCTX_TYPE_POLL)
1195                         blk_mq_map_queues(&set->map[i]);
1196                 else
1197                         blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
1198         }
1199 }
1200
1201 static void virtblk_complete_batch(struct io_comp_batch *iob)
1202 {
1203         struct request *req;
1204
1205         rq_list_for_each(&iob->req_list, req) {
1206                 virtblk_unmap_data(req, blk_mq_rq_to_pdu(req));
1207                 virtblk_cleanup_cmd(req);
1208         }
1209         blk_mq_end_request_batch(iob);
1210 }
1211
1212 static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1213 {
1214         struct virtio_blk *vblk = hctx->queue->queuedata;
1215         struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
1216         struct virtblk_req *vbr;
1217         unsigned long flags;
1218         unsigned int len;
1219         int found = 0;
1220
1221         spin_lock_irqsave(&vq->lock, flags);
1222
1223         while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
1224                 struct request *req = blk_mq_rq_from_pdu(vbr);
1225
1226                 found++;
1227                 if (!blk_mq_complete_request_remote(req) &&
1228                     !blk_mq_add_to_batch(req, iob, virtblk_vbr_status(vbr),
1229                                                 virtblk_complete_batch))
1230                         virtblk_request_done(req);
1231         }
1232
1233         if (found)
1234                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
1235
1236         spin_unlock_irqrestore(&vq->lock, flags);
1237
1238         return found;
1239 }
1240
1241 static const struct blk_mq_ops virtio_mq_ops = {
1242         .queue_rq       = virtio_queue_rq,
1243         .queue_rqs      = virtio_queue_rqs,
1244         .commit_rqs     = virtio_commit_rqs,
1245         .complete       = virtblk_request_done,
1246         .map_queues     = virtblk_map_queues,
1247         .poll           = virtblk_poll,
1248 };
1249
1250 static unsigned int virtblk_queue_depth;
1251 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
1252
1253 static int virtblk_probe(struct virtio_device *vdev)
1254 {
1255         struct virtio_blk *vblk;
1256         struct request_queue *q;
1257         int err, index;
1258
1259         u32 v, blk_size, max_size, sg_elems, opt_io_size;
1260         u32 max_discard_segs = 0;
1261         u32 discard_granularity = 0;
1262         u16 min_io_size;
1263         u8 physical_block_exp, alignment_offset;
1264         unsigned int queue_depth;
1265         size_t max_dma_size;
1266
1267         if (!vdev->config->get) {
1268                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1269                         __func__);
1270                 return -EINVAL;
1271         }
1272
1273         err = ida_alloc_range(&vd_index_ida, 0,
1274                               minor_to_index(1 << MINORBITS) - 1, GFP_KERNEL);
1275         if (err < 0)
1276                 goto out;
1277         index = err;
1278
1279         /* We need to know how many segments before we allocate. */
1280         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
1281                                    struct virtio_blk_config, seg_max,
1282                                    &sg_elems);
1283
1284         /* We need at least one SG element, whatever they say. */
1285         if (err || !sg_elems)
1286                 sg_elems = 1;
1287
1288         /* Prevent integer overflows and honor max vq size */
1289         sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
1290
1291         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
1292         if (!vblk) {
1293                 err = -ENOMEM;
1294                 goto out_free_index;
1295         }
1296
1297         mutex_init(&vblk->vdev_mutex);
1298
1299         vblk->vdev = vdev;
1300
1301         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
1302
1303         err = init_vq(vblk);
1304         if (err)
1305                 goto out_free_vblk;
1306
1307         /* Default queue sizing is to fill the ring. */
1308         if (!virtblk_queue_depth) {
1309                 queue_depth = vblk->vqs[0].vq->num_free;
1310                 /* ... but without indirect descs, we use 2 descs per req */
1311                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
1312                         queue_depth /= 2;
1313         } else {
1314                 queue_depth = virtblk_queue_depth;
1315         }
1316
1317         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
1318         vblk->tag_set.ops = &virtio_mq_ops;
1319         vblk->tag_set.queue_depth = queue_depth;
1320         vblk->tag_set.numa_node = NUMA_NO_NODE;
1321         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1322         vblk->tag_set.cmd_size =
1323                 sizeof(struct virtblk_req) +
1324                 sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
1325         vblk->tag_set.driver_data = vblk;
1326         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
1327         vblk->tag_set.nr_maps = 1;
1328         if (vblk->io_queues[HCTX_TYPE_POLL])
1329                 vblk->tag_set.nr_maps = 3;
1330
1331         err = blk_mq_alloc_tag_set(&vblk->tag_set);
1332         if (err)
1333                 goto out_free_vq;
1334
1335         vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
1336         if (IS_ERR(vblk->disk)) {
1337                 err = PTR_ERR(vblk->disk);
1338                 goto out_free_tags;
1339         }
1340         q = vblk->disk->queue;
1341
1342         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
1343
1344         vblk->disk->major = major;
1345         vblk->disk->first_minor = index_to_minor(index);
1346         vblk->disk->minors = 1 << PART_BITS;
1347         vblk->disk->private_data = vblk;
1348         vblk->disk->fops = &virtblk_fops;
1349         vblk->index = index;
1350
1351         /* configure queue flush support */
1352         virtblk_update_cache_mode(vdev);
1353
1354         /* If disk is read-only in the host, the guest should obey */
1355         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
1356                 set_disk_ro(vblk->disk, 1);
1357
1358         /* We can handle whatever the host told us to handle. */
1359         blk_queue_max_segments(q, sg_elems);
1360
1361         /* No real sector limit. */
1362         blk_queue_max_hw_sectors(q, UINT_MAX);
1363
1364         max_dma_size = virtio_max_dma_size(vdev);
1365         max_size = max_dma_size > U32_MAX ? U32_MAX : max_dma_size;
1366
1367         /* Host can optionally specify maximum segment size and number of
1368          * segments. */
1369         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1370                                    struct virtio_blk_config, size_max, &v);
1371         if (!err)
1372                 max_size = min(max_size, v);
1373
1374         blk_queue_max_segment_size(q, max_size);
1375
1376         /* Host can optionally specify the block size of the device */
1377         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1378                                    struct virtio_blk_config, blk_size,
1379                                    &blk_size);
1380         if (!err) {
1381                 err = blk_validate_block_size(blk_size);
1382                 if (err) {
1383                         dev_err(&vdev->dev,
1384                                 "virtio_blk: invalid block size: 0x%x\n",
1385                                 blk_size);
1386                         goto out_cleanup_disk;
1387                 }
1388
1389                 blk_queue_logical_block_size(q, blk_size);
1390         } else
1391                 blk_size = queue_logical_block_size(q);
1392
1393         /* Use topology information if available */
1394         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1395                                    struct virtio_blk_config, physical_block_exp,
1396                                    &physical_block_exp);
1397         if (!err && physical_block_exp)
1398                 blk_queue_physical_block_size(q,
1399                                 blk_size * (1 << physical_block_exp));
1400
1401         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1402                                    struct virtio_blk_config, alignment_offset,
1403                                    &alignment_offset);
1404         if (!err && alignment_offset)
1405                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
1406
1407         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1408                                    struct virtio_blk_config, min_io_size,
1409                                    &min_io_size);
1410         if (!err && min_io_size)
1411                 blk_queue_io_min(q, blk_size * min_io_size);
1412
1413         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1414                                    struct virtio_blk_config, opt_io_size,
1415                                    &opt_io_size);
1416         if (!err && opt_io_size)
1417                 blk_queue_io_opt(q, blk_size * opt_io_size);
1418
1419         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1420                 virtio_cread(vdev, struct virtio_blk_config,
1421                              discard_sector_alignment, &discard_granularity);
1422
1423                 virtio_cread(vdev, struct virtio_blk_config,
1424                              max_discard_sectors, &v);
1425                 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
1426
1427                 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1428                              &max_discard_segs);
1429         }
1430
1431         if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1432                 virtio_cread(vdev, struct virtio_blk_config,
1433                              max_write_zeroes_sectors, &v);
1434                 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
1435         }
1436
1437         /* The discard and secure erase limits are combined since the Linux
1438          * block layer uses the same limit for both commands.
1439          *
1440          * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1441          * are negotiated, we will use the minimum between the limits.
1442          *
1443          * discard sector alignment is set to the minimum between discard_sector_alignment
1444          * and secure_erase_sector_alignment.
1445          *
1446          * max discard sectors is set to the minimum between max_discard_seg and
1447          * max_secure_erase_seg.
1448          */
1449         if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1450
1451                 virtio_cread(vdev, struct virtio_blk_config,
1452                              secure_erase_sector_alignment, &v);
1453
1454                 /* secure_erase_sector_alignment should not be zero, the device should set a
1455                  * valid number of sectors.
1456                  */
1457                 if (!v) {
1458                         dev_err(&vdev->dev,
1459                                 "virtio_blk: secure_erase_sector_alignment can't be 0\n");
1460                         err = -EINVAL;
1461                         goto out_cleanup_disk;
1462                 }
1463
1464                 discard_granularity = min_not_zero(discard_granularity, v);
1465
1466                 virtio_cread(vdev, struct virtio_blk_config,
1467                              max_secure_erase_sectors, &v);
1468
1469                 /* max_secure_erase_sectors should not be zero, the device should set a
1470                  * valid number of sectors.
1471                  */
1472                 if (!v) {
1473                         dev_err(&vdev->dev,
1474                                 "virtio_blk: max_secure_erase_sectors can't be 0\n");
1475                         err = -EINVAL;
1476                         goto out_cleanup_disk;
1477                 }
1478
1479                 blk_queue_max_secure_erase_sectors(q, v);
1480
1481                 virtio_cread(vdev, struct virtio_blk_config,
1482                              max_secure_erase_seg, &v);
1483
1484                 /* max_secure_erase_seg should not be zero, the device should set a
1485                  * valid number of segments
1486                  */
1487                 if (!v) {
1488                         dev_err(&vdev->dev,
1489                                 "virtio_blk: max_secure_erase_seg can't be 0\n");
1490                         err = -EINVAL;
1491                         goto out_cleanup_disk;
1492                 }
1493
1494                 max_discard_segs = min_not_zero(max_discard_segs, v);
1495         }
1496
1497         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1498             virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1499                 /* max_discard_seg and discard_granularity will be 0 only
1500                  * if max_discard_seg and discard_sector_alignment fields in the virtio
1501                  * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1502                  * In this case, we use default values.
1503                  */
1504                 if (!max_discard_segs)
1505                         max_discard_segs = sg_elems;
1506
1507                 blk_queue_max_discard_segments(q,
1508                                                min(max_discard_segs, MAX_DISCARD_SEGMENTS));
1509
1510                 if (discard_granularity)
1511                         q->limits.discard_granularity = discard_granularity << SECTOR_SHIFT;
1512                 else
1513                         q->limits.discard_granularity = blk_size;
1514         }
1515
1516         virtblk_update_capacity(vblk, false);
1517         virtio_device_ready(vdev);
1518
1519         /*
1520          * All steps that follow use the VQs therefore they need to be
1521          * placed after the virtio_device_ready() call above.
1522          */
1523         if (virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED)) {
1524                 u8 model;
1525
1526                 virtio_cread(vdev, struct virtio_blk_config, zoned.model,
1527                                 &model);
1528                 switch (model) {
1529                 case VIRTIO_BLK_Z_NONE:
1530                 case VIRTIO_BLK_Z_HA:
1531                         /* Present the host-aware device as non-zoned */
1532                         break;
1533                 case VIRTIO_BLK_Z_HM:
1534                         err = virtblk_probe_zoned_device(vdev, vblk, q);
1535                         if (err)
1536                                 goto out_cleanup_disk;
1537                         break;
1538                 default:
1539                         dev_err(&vdev->dev, "unsupported zone model %d\n",
1540                                 model);
1541                         err = -EINVAL;
1542                         goto out_cleanup_disk;
1543                 }
1544         }
1545
1546         err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
1547         if (err)
1548                 goto out_cleanup_disk;
1549
1550         return 0;
1551
1552 out_cleanup_disk:
1553         put_disk(vblk->disk);
1554 out_free_tags:
1555         blk_mq_free_tag_set(&vblk->tag_set);
1556 out_free_vq:
1557         vdev->config->del_vqs(vdev);
1558         kfree(vblk->vqs);
1559 out_free_vblk:
1560         kfree(vblk);
1561 out_free_index:
1562         ida_free(&vd_index_ida, index);
1563 out:
1564         return err;
1565 }
1566
1567 static void virtblk_remove(struct virtio_device *vdev)
1568 {
1569         struct virtio_blk *vblk = vdev->priv;
1570
1571         /* Make sure no work handler is accessing the device. */
1572         flush_work(&vblk->config_work);
1573
1574         del_gendisk(vblk->disk);
1575         blk_mq_free_tag_set(&vblk->tag_set);
1576
1577         mutex_lock(&vblk->vdev_mutex);
1578
1579         /* Stop all the virtqueues. */
1580         virtio_reset_device(vdev);
1581
1582         /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1583         vblk->vdev = NULL;
1584
1585         vdev->config->del_vqs(vdev);
1586         kfree(vblk->vqs);
1587
1588         mutex_unlock(&vblk->vdev_mutex);
1589
1590         put_disk(vblk->disk);
1591 }
1592
1593 #ifdef CONFIG_PM_SLEEP
1594 static int virtblk_freeze(struct virtio_device *vdev)
1595 {
1596         struct virtio_blk *vblk = vdev->priv;
1597
1598         /* Ensure we don't receive any more interrupts */
1599         virtio_reset_device(vdev);
1600
1601         /* Make sure no work handler is accessing the device. */
1602         flush_work(&vblk->config_work);
1603
1604         blk_mq_quiesce_queue(vblk->disk->queue);
1605
1606         vdev->config->del_vqs(vdev);
1607         kfree(vblk->vqs);
1608
1609         return 0;
1610 }
1611
1612 static int virtblk_restore(struct virtio_device *vdev)
1613 {
1614         struct virtio_blk *vblk = vdev->priv;
1615         int ret;
1616
1617         ret = init_vq(vdev->priv);
1618         if (ret)
1619                 return ret;
1620
1621         virtio_device_ready(vdev);
1622
1623         blk_mq_unquiesce_queue(vblk->disk->queue);
1624         return 0;
1625 }
1626 #endif
1627
1628 static const struct virtio_device_id id_table[] = {
1629         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1630         { 0 },
1631 };
1632
1633 static unsigned int features_legacy[] = {
1634         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1635         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1636         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1637         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1638         VIRTIO_BLK_F_SECURE_ERASE,
1639 }
1640 ;
1641 static unsigned int features[] = {
1642         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1643         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1644         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1645         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1646         VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED,
1647 };
1648
1649 static struct virtio_driver virtio_blk = {
1650         .feature_table                  = features,
1651         .feature_table_size             = ARRAY_SIZE(features),
1652         .feature_table_legacy           = features_legacy,
1653         .feature_table_size_legacy      = ARRAY_SIZE(features_legacy),
1654         .driver.name                    = KBUILD_MODNAME,
1655         .driver.owner                   = THIS_MODULE,
1656         .id_table                       = id_table,
1657         .probe                          = virtblk_probe,
1658         .remove                         = virtblk_remove,
1659         .config_changed                 = virtblk_config_changed,
1660 #ifdef CONFIG_PM_SLEEP
1661         .freeze                         = virtblk_freeze,
1662         .restore                        = virtblk_restore,
1663 #endif
1664 };
1665
1666 static int __init virtio_blk_init(void)
1667 {
1668         int error;
1669
1670         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1671         if (!virtblk_wq)
1672                 return -ENOMEM;
1673
1674         major = register_blkdev(0, "virtblk");
1675         if (major < 0) {
1676                 error = major;
1677                 goto out_destroy_workqueue;
1678         }
1679
1680         error = register_virtio_driver(&virtio_blk);
1681         if (error)
1682                 goto out_unregister_blkdev;
1683         return 0;
1684
1685 out_unregister_blkdev:
1686         unregister_blkdev(major, "virtblk");
1687 out_destroy_workqueue:
1688         destroy_workqueue(virtblk_wq);
1689         return error;
1690 }
1691
1692 static void __exit virtio_blk_fini(void)
1693 {
1694         unregister_virtio_driver(&virtio_blk);
1695         unregister_blkdev(major, "virtblk");
1696         destroy_workqueue(virtblk_wq);
1697 }
1698 module_init(virtio_blk_init);
1699 module_exit(virtio_blk_fini);
1700
1701 MODULE_DEVICE_TABLE(virtio, id_table);
1702 MODULE_DESCRIPTION("Virtio block driver");
1703 MODULE_LICENSE("GPL");