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_complete_batch(struct io_comp_batch *iob)
 352 {
 353         struct request *req;
 354
 355         rq_list_for_each(&iob->req_list, req) {
 356                 virtblk_unmap_data(req, blk_mq_rq_to_pdu(req));
 357                 virtblk_cleanup_cmd(req);
 358         }
 359         blk_mq_end_request_batch(iob);
 360 }
 361
 362 static int virtblk_handle_req(struct virtio_blk_vq *vq,
 363                               struct io_comp_batch *iob)
 364 {
 365         struct virtblk_req *vbr;
 366         int req_done = 0;
 367         unsigned int len;
 368
 369         while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
 370                 struct request *req = blk_mq_rq_from_pdu(vbr);
 371
 372                 if (likely(!blk_should_fake_timeout(req->q)) &&
 373                     !blk_mq_complete_request_remote(req) &&
 374                     !blk_mq_add_to_batch(req, iob, virtblk_vbr_status(vbr),
 375                                          virtblk_complete_batch))
 376                         virtblk_request_done(req);
 377                 req_done++;
 378         }
 379
 380         return req_done;
 381 }
 382
 383 static void virtblk_done(struct virtqueue *vq)
 384 {
 385         struct virtio_blk *vblk = vq->vdev->priv;
 386         struct virtio_blk_vq *vblk_vq = &vblk->vqs[vq->index];
 387         int req_done = 0;
 388         unsigned long flags;
 389         DEFINE_IO_COMP_BATCH(iob);
 390
 391         spin_lock_irqsave(&vblk_vq->lock, flags);
 392         do {
 393                 virtqueue_disable_cb(vq);
 394                 req_done += virtblk_handle_req(vblk_vq, &iob);
 395
 396                 if (unlikely(virtqueue_is_broken(vq)))
 397                         break;
 398         } while (!virtqueue_enable_cb(vq));
 399
 400         if (req_done) {
 401                 if (!rq_list_empty(iob.req_list))
 402                         iob.complete(&iob);
 403
 404                 /* In case queue is stopped waiting for more buffers. */
 405                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 406         }
 407         spin_unlock_irqrestore(&vblk_vq->lock, flags);
 408 }
 409
 410 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 411 {
 412         struct virtio_blk *vblk = hctx->queue->queuedata;
 413         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 414         bool kick;
 415
 416         spin_lock_irq(&vq->lock);
 417         kick = virtqueue_kick_prepare(vq->vq);
 418         spin_unlock_irq(&vq->lock);
 419
 420         if (kick)
 421                 virtqueue_notify(vq->vq);
 422 }
 423
 424 static blk_status_t virtblk_fail_to_queue(struct request *req, int rc)
 425 {
 426         virtblk_cleanup_cmd(req);
 427         switch (rc) {
 428         case -ENOSPC:
 429                 return BLK_STS_DEV_RESOURCE;
 430         case -ENOMEM:
 431                 return BLK_STS_RESOURCE;
 432         default:
 433                 return BLK_STS_IOERR;
 434         }
 435 }
 436
 437 static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
 438                                         struct virtio_blk *vblk,
 439                                         struct request *req,
 440                                         struct virtblk_req *vbr)
 441 {
 442         blk_status_t status;
 443         int num;
 444
 445         status = virtblk_setup_cmd(vblk->vdev, req, vbr);
 446         if (unlikely(status))
 447                 return status;
 448
 449         num = virtblk_map_data(hctx, req, vbr);
 450         if (unlikely(num < 0))
 451                 return virtblk_fail_to_queue(req, -ENOMEM);
 452         vbr->sg_table.nents = num;
 453
 454         blk_mq_start_request(req);
 455
 456         return BLK_STS_OK;
 457 }
 458
 459 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 460                            const struct blk_mq_queue_data *bd)
 461 {
 462         struct virtio_blk *vblk = hctx->queue->queuedata;
 463         struct request *req = bd->rq;
 464         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 465         unsigned long flags;
 466         int qid = hctx->queue_num;
 467         bool notify = false;
 468         blk_status_t status;
 469         int err;
 470
 471         status = virtblk_prep_rq(hctx, vblk, req, vbr);
 472         if (unlikely(status))
 473                 return status;
 474
 475         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 476         err = virtblk_add_req(vblk->vqs[qid].vq, vbr);
 477         if (err) {
 478                 virtqueue_kick(vblk->vqs[qid].vq);
 479                 /* Don't stop the queue if -ENOMEM: we may have failed to
 480                  * bounce the buffer due to global resource outage.
 481                  */
 482                 if (err == -ENOSPC)
 483                         blk_mq_stop_hw_queue(hctx);
 484                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 485                 virtblk_unmap_data(req, vbr);
 486                 return virtblk_fail_to_queue(req, err);
 487         }
 488
 489         if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 490                 notify = true;
 491         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 492
 493         if (notify)
 494                 virtqueue_notify(vblk->vqs[qid].vq);
 495         return BLK_STS_OK;
 496 }
 497
 498 static bool virtblk_prep_rq_batch(struct request *req)
 499 {
 500         struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
 501         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 502
 503         req->mq_hctx->tags->rqs[req->tag] = req;
 504
 505         return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
 506 }
 507
 508 static bool virtblk_add_req_batch(struct virtio_blk_vq *vq,
 509                                         struct request **rqlist)
 510 {
 511         unsigned long flags;
 512         int err;
 513         bool kick;
 514
 515         spin_lock_irqsave(&vq->lock, flags);
 516
 517         while (!rq_list_empty(*rqlist)) {
 518                 struct request *req = rq_list_pop(rqlist);
 519                 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 520
 521                 err = virtblk_add_req(vq->vq, vbr);
 522                 if (err) {
 523                         virtblk_unmap_data(req, vbr);
 524                         virtblk_cleanup_cmd(req);
 525                         blk_mq_requeue_request(req, true);
 526                 }
 527         }
 528
 529         kick = virtqueue_kick_prepare(vq->vq);
 530         spin_unlock_irqrestore(&vq->lock, flags);
 531
 532         return kick;
 533 }
 534
 535 static void virtio_queue_rqs(struct request **rqlist)
 536 {
 537         struct request *req, *next, *prev = NULL;
 538         struct request *requeue_list = NULL;
 539
 540         rq_list_for_each_safe(rqlist, req, next) {
 541                 struct virtio_blk_vq *vq = get_virtio_blk_vq(req->mq_hctx);
 542                 bool kick;
 543
 544                 if (!virtblk_prep_rq_batch(req)) {
 545                         rq_list_move(rqlist, &requeue_list, req, prev);
 546                         req = prev;
 547                         if (!req)
 548                                 continue;
 549                 }
 550
 551                 if (!next || req->mq_hctx != next->mq_hctx) {
 552                         req->rq_next = NULL;
 553                         kick = virtblk_add_req_batch(vq, rqlist);
 554                         if (kick)
 555                                 virtqueue_notify(vq->vq);
 556
 557                         *rqlist = next;
 558                         prev = NULL;
 559                 } else
 560                         prev = req;
 561         }
 562
 563         *rqlist = requeue_list;
 564 }
 565
 566 #ifdef CONFIG_BLK_DEV_ZONED
 567 static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk,
 568                                           unsigned int nr_zones,
 569                                           size_t *buflen)
 570 {
 571         struct request_queue *q = vblk->disk->queue;
 572         size_t bufsize;
 573         void *buf;
 574
 575         nr_zones = min_t(unsigned int, nr_zones,
 576                          get_capacity(vblk->disk) >> ilog2(vblk->zone_sectors));
 577
 578         bufsize = sizeof(struct virtio_blk_zone_report) +
 579                 nr_zones * sizeof(struct virtio_blk_zone_descriptor);
 580         bufsize = min_t(size_t, bufsize,
 581                         queue_max_hw_sectors(q) << SECTOR_SHIFT);
 582         bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
 583
 584         while (bufsize >= sizeof(struct virtio_blk_zone_report)) {
 585                 buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
 586                 if (buf) {
 587                         *buflen = bufsize;
 588                         return buf;
 589                 }
 590                 bufsize >>= 1;
 591         }
 592
 593         return NULL;
 594 }
 595
 596 static int virtblk_submit_zone_report(struct virtio_blk *vblk,
 597                                        char *report_buf, size_t report_len,
 598                                        sector_t sector)
 599 {
 600         struct request_queue *q = vblk->disk->queue;
 601         struct request *req;
 602         struct virtblk_req *vbr;
 603         int err;
 604
 605         req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
 606         if (IS_ERR(req))
 607                 return PTR_ERR(req);
 608
 609         vbr = blk_mq_rq_to_pdu(req);
 610         vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
 611         vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT);
 612         vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector);
 613
 614         err = blk_rq_map_kern(q, req, report_buf, report_len, GFP_KERNEL);
 615         if (err)
 616                 goto out;
 617
 618         blk_execute_rq(req, false);
 619         err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
 620 out:
 621         blk_mq_free_request(req);
 622         return err;
 623 }
 624
 625 static int virtblk_parse_zone(struct virtio_blk *vblk,
 626                                struct virtio_blk_zone_descriptor *entry,
 627                                unsigned int idx, report_zones_cb cb, void *data)
 628 {
 629         struct blk_zone zone = { };
 630
 631         zone.start = virtio64_to_cpu(vblk->vdev, entry->z_start);
 632         if (zone.start + vblk->zone_sectors <= get_capacity(vblk->disk))
 633                 zone.len = vblk->zone_sectors;
 634         else
 635                 zone.len = get_capacity(vblk->disk) - zone.start;
 636         zone.capacity = virtio64_to_cpu(vblk->vdev, entry->z_cap);
 637         zone.wp = virtio64_to_cpu(vblk->vdev, entry->z_wp);
 638
 639         switch (entry->z_type) {
 640         case VIRTIO_BLK_ZT_SWR:
 641                 zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
 642                 break;
 643         case VIRTIO_BLK_ZT_SWP:
 644                 zone.type = BLK_ZONE_TYPE_SEQWRITE_PREF;
 645                 break;
 646         case VIRTIO_BLK_ZT_CONV:
 647                 zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
 648                 break;
 649         default:
 650                 dev_err(&vblk->vdev->dev, "zone %llu: invalid type %#x\n",
 651                         zone.start, entry->z_type);
 652                 return -EIO;
 653         }
 654
 655         switch (entry->z_state) {
 656         case VIRTIO_BLK_ZS_EMPTY:
 657                 zone.cond = BLK_ZONE_COND_EMPTY;
 658                 break;
 659         case VIRTIO_BLK_ZS_CLOSED:
 660                 zone.cond = BLK_ZONE_COND_CLOSED;
 661                 break;
 662         case VIRTIO_BLK_ZS_FULL:
 663                 zone.cond = BLK_ZONE_COND_FULL;
 664                 zone.wp = zone.start + zone.len;
 665                 break;
 666         case VIRTIO_BLK_ZS_EOPEN:
 667                 zone.cond = BLK_ZONE_COND_EXP_OPEN;
 668                 break;
 669         case VIRTIO_BLK_ZS_IOPEN:
 670                 zone.cond = BLK_ZONE_COND_IMP_OPEN;
 671                 break;
 672         case VIRTIO_BLK_ZS_NOT_WP:
 673                 zone.cond = BLK_ZONE_COND_NOT_WP;
 674                 break;
 675         case VIRTIO_BLK_ZS_RDONLY:
 676                 zone.cond = BLK_ZONE_COND_READONLY;
 677                 zone.wp = ULONG_MAX;
 678                 break;
 679         case VIRTIO_BLK_ZS_OFFLINE:
 680                 zone.cond = BLK_ZONE_COND_OFFLINE;
 681                 zone.wp = ULONG_MAX;
 682                 break;
 683         default:
 684                 dev_err(&vblk->vdev->dev, "zone %llu: invalid condition %#x\n",
 685                         zone.start, entry->z_state);
 686                 return -EIO;
 687         }
 688
 689         /*
 690          * The callback below checks the validity of the reported
 691          * entry data, no need to further validate it here.
 692          */
 693         return cb(&zone, idx, data);
 694 }
 695
 696 static int virtblk_report_zones(struct gendisk *disk, sector_t sector,
 697                                  unsigned int nr_zones, report_zones_cb cb,
 698                                  void *data)
 699 {
 700         struct virtio_blk *vblk = disk->private_data;
 701         struct virtio_blk_zone_report *report;
 702         unsigned long long nz, i;
 703         size_t buflen;
 704         unsigned int zone_idx = 0;
 705         int ret;
 706
 707         if (WARN_ON_ONCE(!vblk->zone_sectors))
 708                 return -EOPNOTSUPP;
 709
 710         report = virtblk_alloc_report_buffer(vblk, nr_zones, &buflen);
 711         if (!report)
 712                 return -ENOMEM;
 713
 714         mutex_lock(&vblk->vdev_mutex);
 715
 716         if (!vblk->vdev) {
 717                 ret = -ENXIO;
 718                 goto fail_report;
 719         }
 720
 721         while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) {
 722                 memset(report, 0, buflen);
 723
 724                 ret = virtblk_submit_zone_report(vblk, (char *)report,
 725                                                  buflen, sector);
 726                 if (ret)
 727                         goto fail_report;
 728
 729                 nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones),
 730                            nr_zones);
 731                 if (!nz)
 732                         break;
 733
 734                 for (i = 0; i < nz && zone_idx < nr_zones; i++) {
 735                         ret = virtblk_parse_zone(vblk, &report->zones[i],
 736                                                  zone_idx, cb, data);
 737                         if (ret)
 738                                 goto fail_report;
 739
 740                         sector = virtio64_to_cpu(vblk->vdev,
 741                                                  report->zones[i].z_start) +
 742                                  vblk->zone_sectors;
 743                         zone_idx++;
 744                 }
 745         }
 746
 747         if (zone_idx > 0)
 748                 ret = zone_idx;
 749         else
 750                 ret = -EINVAL;
 751 fail_report:
 752         mutex_unlock(&vblk->vdev_mutex);
 753         kvfree(report);
 754         return ret;
 755 }
 756
 757 static void virtblk_revalidate_zones(struct virtio_blk *vblk)
 758 {
 759         u8 model;
 760
 761         virtio_cread(vblk->vdev, struct virtio_blk_config,
 762                      zoned.model, &model);
 763         switch (model) {
 764         default:
 765                 dev_err(&vblk->vdev->dev, "unknown zone model %d\n", model);
 766                 fallthrough;
 767         case VIRTIO_BLK_Z_NONE:
 768         case VIRTIO_BLK_Z_HA:
 769                 disk_set_zoned(vblk->disk, BLK_ZONED_NONE);
 770                 return;
 771         case VIRTIO_BLK_Z_HM:
 772                 WARN_ON_ONCE(!vblk->zone_sectors);
 773                 if (!blk_revalidate_disk_zones(vblk->disk, NULL))
 774                         set_capacity_and_notify(vblk->disk, 0);
 775         }
 776 }
 777
 778 static int virtblk_probe_zoned_device(struct virtio_device *vdev,
 779                                        struct virtio_blk *vblk,
 780                                        struct request_queue *q)
 781 {
 782         u32 v, wg;
 783         u8 model;
 784         int ret;
 785
 786         virtio_cread(vdev, struct virtio_blk_config,
 787                      zoned.model, &model);
 788
 789         switch (model) {
 790         case VIRTIO_BLK_Z_NONE:
 791         case VIRTIO_BLK_Z_HA:
 792                 /* Present the host-aware device as non-zoned */
 793                 return 0;
 794         case VIRTIO_BLK_Z_HM:
 795                 break;
 796         default:
 797                 dev_err(&vdev->dev, "unsupported zone model %d\n", model);
 798                 return -EINVAL;
 799         }
 800
 801         dev_dbg(&vdev->dev, "probing host-managed zoned device\n");
 802
 803         disk_set_zoned(vblk->disk, BLK_ZONED_HM);
 804         blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
 805
 806         virtio_cread(vdev, struct virtio_blk_config,
 807                      zoned.max_open_zones, &v);
 808         disk_set_max_open_zones(vblk->disk, v);
 809         dev_dbg(&vdev->dev, "max open zones = %u\n", v);
 810
 811         virtio_cread(vdev, struct virtio_blk_config,
 812                      zoned.max_active_zones, &v);
 813         disk_set_max_active_zones(vblk->disk, v);
 814         dev_dbg(&vdev->dev, "max active zones = %u\n", v);
 815
 816         virtio_cread(vdev, struct virtio_blk_config,
 817                      zoned.write_granularity, &wg);
 818         if (!wg) {
 819                 dev_warn(&vdev->dev, "zero write granularity reported\n");
 820                 return -ENODEV;
 821         }
 822         blk_queue_physical_block_size(q, wg);
 823         blk_queue_io_min(q, wg);
 824
 825         dev_dbg(&vdev->dev, "write granularity = %u\n", wg);
 826
 827         /*
 828          * virtio ZBD specification doesn't require zones to be a power of
 829          * two sectors in size, but the code in this driver expects that.
 830          */
 831         virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors,
 832                      &vblk->zone_sectors);
 833         if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) {
 834                 dev_err(&vdev->dev,
 835                         "zoned device with non power of two zone size %u\n",
 836                         vblk->zone_sectors);
 837                 return -ENODEV;
 838         }
 839         dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors);
 840
 841         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
 842                 dev_warn(&vblk->vdev->dev,
 843                          "ignoring negotiated F_DISCARD for zoned device\n");
 844                 blk_queue_max_discard_sectors(q, 0);
 845         }
 846
 847         ret = blk_revalidate_disk_zones(vblk->disk, NULL);
 848         if (!ret) {
 849                 virtio_cread(vdev, struct virtio_blk_config,
 850                              zoned.max_append_sectors, &v);
 851                 if (!v) {
 852                         dev_warn(&vdev->dev, "zero max_append_sectors reported\n");
 853                         return -ENODEV;
 854                 }
 855                 if ((v << SECTOR_SHIFT) < wg) {
 856                         dev_err(&vdev->dev,
 857                                 "write granularity %u exceeds max_append_sectors %u limit\n",
 858                                 wg, v);
 859                         return -ENODEV;
 860                 }
 861
 862                 blk_queue_max_zone_append_sectors(q, v);
 863                 dev_dbg(&vdev->dev, "max append sectors = %u\n", v);
 864         }
 865
 866         return ret;
 867 }
 868
 869 #else
 870
 871 /*
 872  * Zoned block device support is not configured in this kernel.
 873  * Host-managed zoned devices can't be supported, but others are
 874  * good to go as regular block devices.
 875  */
 876 #define virtblk_report_zones       NULL
 877
 878 static inline void virtblk_revalidate_zones(struct virtio_blk *vblk)
 879 {
 880 }
 881
 882 static inline int virtblk_probe_zoned_device(struct virtio_device *vdev,
 883                         struct virtio_blk *vblk, struct request_queue *q)
 884 {
 885         u8 model;
 886
 887         virtio_cread(vdev, struct virtio_blk_config, zoned.model, &model);
 888         if (model == VIRTIO_BLK_Z_HM) {
 889                 dev_err(&vdev->dev,
 890                         "virtio_blk: zoned devices are not supported");
 891                 return -EOPNOTSUPP;
 892         }
 893
 894         return 0;
 895 }
 896 #endif /* CONFIG_BLK_DEV_ZONED */
 897
 898 /* return id (s/n) string for *disk to *id_str
 899  */
 900 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 901 {
 902         struct virtio_blk *vblk = disk->private_data;
 903         struct request_queue *q = vblk->disk->queue;
 904         struct request *req;
 905         struct virtblk_req *vbr;
 906         int err;
 907
 908         req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
 909         if (IS_ERR(req))
 910                 return PTR_ERR(req);
 911
 912         vbr = blk_mq_rq_to_pdu(req);
 913         vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
 914         vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
 915         vbr->out_hdr.sector = 0;
 916
 917         err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 918         if (err)
 919                 goto out;
 920
 921         blk_execute_rq(req, false);
 922         err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
 923 out:
 924         blk_mq_free_request(req);
 925         return err;
 926 }
 927
 928 /* We provide getgeo only to please some old bootloader/partitioning tools */
 929 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 930 {
 931         struct virtio_blk *vblk = bd->bd_disk->private_data;
 932         int ret = 0;
 933
 934         mutex_lock(&vblk->vdev_mutex);
 935
 936         if (!vblk->vdev) {
 937                 ret = -ENXIO;
 938                 goto out;
 939         }
 940
 941         /* see if the host passed in geometry config */
 942         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 943                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 944                              geometry.cylinders, &geo->cylinders);
 945                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 946                              geometry.heads, &geo->heads);
 947                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 948                              geometry.sectors, &geo->sectors);
 949         } else {
 950                 /* some standard values, similar to sd */
 951                 geo->heads = 1 << 6;
 952                 geo->sectors = 1 << 5;
 953                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 954         }
 955 out:
 956         mutex_unlock(&vblk->vdev_mutex);
 957         return ret;
 958 }
 959
 960 static void virtblk_free_disk(struct gendisk *disk)
 961 {
 962         struct virtio_blk *vblk = disk->private_data;
 963
 964         ida_free(&vd_index_ida, vblk->index);
 965         mutex_destroy(&vblk->vdev_mutex);
 966         kfree(vblk);
 967 }
 968
 969 static const struct block_device_operations virtblk_fops = {
 970         .owner          = THIS_MODULE,
 971         .getgeo         = virtblk_getgeo,
 972         .free_disk      = virtblk_free_disk,
 973         .report_zones   = virtblk_report_zones,
 974 };
 975
 976 static int index_to_minor(int index)
 977 {
 978         return index << PART_BITS;
 979 }
 980
 981 static int minor_to_index(int minor)
 982 {
 983         return minor >> PART_BITS;
 984 }
 985
 986 static ssize_t serial_show(struct device *dev,
 987                            struct device_attribute *attr, char *buf)
 988 {
 989         struct gendisk *disk = dev_to_disk(dev);
 990         int err;
 991
 992         /* sysfs gives us a PAGE_SIZE buffer */
 993         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 994
 995         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 996         err = virtblk_get_id(disk, buf);
 997         if (!err)
 998                 return strlen(buf);
 999
1000         if (err == -EIO) /* Unsupported? Make it empty. */
1001                 return 0;
1002
1003         return err;
1004 }
1005
1006 static DEVICE_ATTR_RO(serial);
1007
1008 /* The queue's logical block size must be set before calling this */
1009 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
1010 {
1011         struct virtio_device *vdev = vblk->vdev;
1012         struct request_queue *q = vblk->disk->queue;
1013         char cap_str_2[10], cap_str_10[10];
1014         unsigned long long nblocks;
1015         u64 capacity;
1016
1017         /* Host must always specify the capacity. */
1018         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
1019
1020         nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
1021
1022         string_get_size(nblocks, queue_logical_block_size(q),
1023                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
1024         string_get_size(nblocks, queue_logical_block_size(q),
1025                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
1026
1027         dev_notice(&vdev->dev,
1028                    "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
1029                    vblk->disk->disk_name,
1030                    resize ? "new size: " : "",
1031                    nblocks,
1032                    queue_logical_block_size(q),
1033                    cap_str_10,
1034                    cap_str_2);
1035
1036         set_capacity_and_notify(vblk->disk, capacity);
1037 }
1038
1039 static void virtblk_config_changed_work(struct work_struct *work)
1040 {
1041         struct virtio_blk *vblk =
1042                 container_of(work, struct virtio_blk, config_work);
1043
1044         virtblk_revalidate_zones(vblk);
1045         virtblk_update_capacity(vblk, true);
1046 }
1047
1048 static void virtblk_config_changed(struct virtio_device *vdev)
1049 {
1050         struct virtio_blk *vblk = vdev->priv;
1051
1052         queue_work(virtblk_wq, &vblk->config_work);
1053 }
1054
1055 static int init_vq(struct virtio_blk *vblk)
1056 {
1057         int err;
1058         int i;
1059         vq_callback_t **callbacks;
1060         const char **names;
1061         struct virtqueue **vqs;
1062         unsigned short num_vqs;
1063         unsigned int num_poll_vqs;
1064         struct virtio_device *vdev = vblk->vdev;
1065         struct irq_affinity desc = { 0, };
1066
1067         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
1068                                    struct virtio_blk_config, num_queues,
1069                                    &num_vqs);
1070         if (err)
1071                 num_vqs = 1;
1072
1073         if (!err && !num_vqs) {
1074                 dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
1075                 return -EINVAL;
1076         }
1077
1078         num_vqs = min_t(unsigned int,
1079                         min_not_zero(num_request_queues, nr_cpu_ids),
1080                         num_vqs);
1081
1082         num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
1083
1084         vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
1085         vblk->io_queues[HCTX_TYPE_READ] = 0;
1086         vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
1087
1088         dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
1089                                 vblk->io_queues[HCTX_TYPE_DEFAULT],
1090                                 vblk->io_queues[HCTX_TYPE_READ],
1091                                 vblk->io_queues[HCTX_TYPE_POLL]);
1092
1093         vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
1094         if (!vblk->vqs)
1095                 return -ENOMEM;
1096
1097         names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
1098         callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
1099         vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
1100         if (!names || !callbacks || !vqs) {
1101                 err = -ENOMEM;
1102                 goto out;
1103         }
1104
1105         for (i = 0; i < num_vqs - num_poll_vqs; i++) {
1106                 callbacks[i] = virtblk_done;
1107                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
1108                 names[i] = vblk->vqs[i].name;
1109         }
1110
1111         for (; i < num_vqs; i++) {
1112                 callbacks[i] = NULL;
1113                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i);
1114                 names[i] = vblk->vqs[i].name;
1115         }
1116
1117         /* Discover virtqueues and write information to configuration.  */
1118         err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
1119         if (err)
1120                 goto out;
1121
1122         for (i = 0; i < num_vqs; i++) {
1123                 spin_lock_init(&vblk->vqs[i].lock);
1124                 vblk->vqs[i].vq = vqs[i];
1125         }
1126         vblk->num_vqs = num_vqs;
1127
1128 out:
1129         kfree(vqs);
1130         kfree(callbacks);
1131         kfree(names);
1132         if (err)
1133                 kfree(vblk->vqs);
1134         return err;
1135 }
1136
1137 /*
1138  * Legacy naming scheme used for virtio devices.  We are stuck with it for
1139  * virtio blk but don't ever use it for any new driver.
1140  */
1141 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
1142 {
1143         const int base = 'z' - 'a' + 1;
1144         char *begin = buf + strlen(prefix);
1145         char *end = buf + buflen;
1146         char *p;
1147         int unit;
1148
1149         p = end - 1;
1150         *p = '\0';
1151         unit = base;
1152         do {
1153                 if (p == begin)
1154                         return -EINVAL;
1155                 *--p = 'a' + (index % unit);
1156                 index = (index / unit) - 1;
1157         } while (index >= 0);
1158
1159         memmove(begin, p, end - p);
1160         memcpy(buf, prefix, strlen(prefix));
1161
1162         return 0;
1163 }
1164
1165 static int virtblk_get_cache_mode(struct virtio_device *vdev)
1166 {
1167         u8 writeback;
1168         int err;
1169
1170         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
1171                                    struct virtio_blk_config, wce,
1172                                    &writeback);
1173
1174         /*
1175          * If WCE is not configurable and flush is not available,
1176          * assume no writeback cache is in use.
1177          */
1178         if (err)
1179                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
1180
1181         return writeback;
1182 }
1183
1184 static void virtblk_update_cache_mode(struct virtio_device *vdev)
1185 {
1186         u8 writeback = virtblk_get_cache_mode(vdev);
1187         struct virtio_blk *vblk = vdev->priv;
1188
1189         blk_queue_write_cache(vblk->disk->queue, writeback, false);
1190 }
1191
1192 static const char *const virtblk_cache_types[] = {
1193         "write through", "write back"
1194 };
1195
1196 static ssize_t
1197 cache_type_store(struct device *dev, struct device_attribute *attr,
1198                  const char *buf, size_t count)
1199 {
1200         struct gendisk *disk = dev_to_disk(dev);
1201         struct virtio_blk *vblk = disk->private_data;
1202         struct virtio_device *vdev = vblk->vdev;
1203         int i;
1204
1205         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
1206         i = sysfs_match_string(virtblk_cache_types, buf);
1207         if (i < 0)
1208                 return i;
1209
1210         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
1211         virtblk_update_cache_mode(vdev);
1212         return count;
1213 }
1214
1215 static ssize_t
1216 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
1217 {
1218         struct gendisk *disk = dev_to_disk(dev);
1219         struct virtio_blk *vblk = disk->private_data;
1220         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
1221
1222         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
1223         return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
1224 }
1225
1226 static DEVICE_ATTR_RW(cache_type);
1227
1228 static struct attribute *virtblk_attrs[] = {
1229         &dev_attr_serial.attr,
1230         &dev_attr_cache_type.attr,
1231         NULL,
1232 };
1233
1234 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
1235                 struct attribute *a, int n)
1236 {
1237         struct device *dev = kobj_to_dev(kobj);
1238         struct gendisk *disk = dev_to_disk(dev);
1239         struct virtio_blk *vblk = disk->private_data;
1240         struct virtio_device *vdev = vblk->vdev;
1241
1242         if (a == &dev_attr_cache_type.attr &&
1243             !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
1244                 return S_IRUGO;
1245
1246         return a->mode;
1247 }
1248
1249 static const struct attribute_group virtblk_attr_group = {
1250         .attrs = virtblk_attrs,
1251         .is_visible = virtblk_attrs_are_visible,
1252 };
1253
1254 static const struct attribute_group *virtblk_attr_groups[] = {
1255         &virtblk_attr_group,
1256         NULL,
1257 };
1258
1259 static void virtblk_map_queues(struct blk_mq_tag_set *set)
1260 {
1261         struct virtio_blk *vblk = set->driver_data;
1262         int i, qoff;
1263
1264         for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1265                 struct blk_mq_queue_map *map = &set->map[i];
1266
1267                 map->nr_queues = vblk->io_queues[i];
1268                 map->queue_offset = qoff;
1269                 qoff += map->nr_queues;
1270
1271                 if (map->nr_queues == 0)
1272                         continue;
1273
1274                 /*
1275                  * Regular queues have interrupts and hence CPU affinity is
1276                  * defined by the core virtio code, but polling queues have
1277                  * no interrupts so we let the block layer assign CPU affinity.
1278                  */
1279                 if (i == HCTX_TYPE_POLL)
1280                         blk_mq_map_queues(&set->map[i]);
1281                 else
1282                         blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
1283         }
1284 }
1285
1286 static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1287 {
1288         struct virtio_blk *vblk = hctx->queue->queuedata;
1289         struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
1290         unsigned long flags;
1291         int found = 0;
1292
1293         spin_lock_irqsave(&vq->lock, flags);
1294         found = virtblk_handle_req(vq, iob);
1295
1296         if (found)
1297                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
1298
1299         spin_unlock_irqrestore(&vq->lock, flags);
1300
1301         return found;
1302 }
1303
1304 static const struct blk_mq_ops virtio_mq_ops = {
1305         .queue_rq       = virtio_queue_rq,
1306         .queue_rqs      = virtio_queue_rqs,
1307         .commit_rqs     = virtio_commit_rqs,
1308         .complete       = virtblk_request_done,
1309         .map_queues     = virtblk_map_queues,
1310         .poll           = virtblk_poll,
1311 };
1312
1313 static unsigned int virtblk_queue_depth;
1314 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
1315
1316 static int virtblk_probe(struct virtio_device *vdev)
1317 {
1318         struct virtio_blk *vblk;
1319         struct request_queue *q;
1320         int err, index;
1321
1322         u32 v, blk_size, max_size, sg_elems, opt_io_size;
1323         u32 max_discard_segs = 0;
1324         u32 discard_granularity = 0;
1325         u16 min_io_size;
1326         u8 physical_block_exp, alignment_offset;
1327         unsigned int queue_depth;
1328
1329         if (!vdev->config->get) {
1330                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1331                         __func__);
1332                 return -EINVAL;
1333         }
1334
1335         err = ida_alloc_range(&vd_index_ida, 0,
1336                               minor_to_index(1 << MINORBITS) - 1, GFP_KERNEL);
1337         if (err < 0)
1338                 goto out;
1339         index = err;
1340
1341         /* We need to know how many segments before we allocate. */
1342         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
1343                                    struct virtio_blk_config, seg_max,
1344                                    &sg_elems);
1345
1346         /* We need at least one SG element, whatever they say. */
1347         if (err || !sg_elems)
1348                 sg_elems = 1;
1349
1350         /* Prevent integer overflows and honor max vq size */
1351         sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
1352
1353         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
1354         if (!vblk) {
1355                 err = -ENOMEM;
1356                 goto out_free_index;
1357         }
1358
1359         mutex_init(&vblk->vdev_mutex);
1360
1361         vblk->vdev = vdev;
1362
1363         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
1364
1365         err = init_vq(vblk);
1366         if (err)
1367                 goto out_free_vblk;
1368
1369         /* Default queue sizing is to fill the ring. */
1370         if (!virtblk_queue_depth) {
1371                 queue_depth = vblk->vqs[0].vq->num_free;
1372                 /* ... but without indirect descs, we use 2 descs per req */
1373                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
1374                         queue_depth /= 2;
1375         } else {
1376                 queue_depth = virtblk_queue_depth;
1377         }
1378
1379         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
1380         vblk->tag_set.ops = &virtio_mq_ops;
1381         vblk->tag_set.queue_depth = queue_depth;
1382         vblk->tag_set.numa_node = NUMA_NO_NODE;
1383         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1384         vblk->tag_set.cmd_size =
1385                 sizeof(struct virtblk_req) +
1386                 sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
1387         vblk->tag_set.driver_data = vblk;
1388         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
1389         vblk->tag_set.nr_maps = 1;
1390         if (vblk->io_queues[HCTX_TYPE_POLL])
1391                 vblk->tag_set.nr_maps = 3;
1392
1393         err = blk_mq_alloc_tag_set(&vblk->tag_set);
1394         if (err)
1395                 goto out_free_vq;
1396
1397         vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
1398         if (IS_ERR(vblk->disk)) {
1399                 err = PTR_ERR(vblk->disk);
1400                 goto out_free_tags;
1401         }
1402         q = vblk->disk->queue;
1403
1404         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
1405
1406         vblk->disk->major = major;
1407         vblk->disk->first_minor = index_to_minor(index);
1408         vblk->disk->minors = 1 << PART_BITS;
1409         vblk->disk->private_data = vblk;
1410         vblk->disk->fops = &virtblk_fops;
1411         vblk->index = index;
1412
1413         /* configure queue flush support */
1414         virtblk_update_cache_mode(vdev);
1415
1416         /* If disk is read-only in the host, the guest should obey */
1417         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
1418                 set_disk_ro(vblk->disk, 1);
1419
1420         /* We can handle whatever the host told us to handle. */
1421         blk_queue_max_segments(q, sg_elems);
1422
1423         /* No real sector limit. */
1424         blk_queue_max_hw_sectors(q, UINT_MAX);
1425
1426         max_size = virtio_max_dma_size(vdev);
1427
1428         /* Host can optionally specify maximum segment size and number of
1429          * segments. */
1430         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1431                                    struct virtio_blk_config, size_max, &v);
1432         if (!err)
1433                 max_size = min(max_size, v);
1434
1435         blk_queue_max_segment_size(q, max_size);
1436
1437         /* Host can optionally specify the block size of the device */
1438         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1439                                    struct virtio_blk_config, blk_size,
1440                                    &blk_size);
1441         if (!err) {
1442                 err = blk_validate_block_size(blk_size);
1443                 if (err) {
1444                         dev_err(&vdev->dev,
1445                                 "virtio_blk: invalid block size: 0x%x\n",
1446                                 blk_size);
1447                         goto out_cleanup_disk;
1448                 }
1449
1450                 blk_queue_logical_block_size(q, blk_size);
1451         } else
1452                 blk_size = queue_logical_block_size(q);
1453
1454         /* Use topology information if available */
1455         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1456                                    struct virtio_blk_config, physical_block_exp,
1457                                    &physical_block_exp);
1458         if (!err && physical_block_exp)
1459                 blk_queue_physical_block_size(q,
1460                                 blk_size * (1 << physical_block_exp));
1461
1462         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1463                                    struct virtio_blk_config, alignment_offset,
1464                                    &alignment_offset);
1465         if (!err && alignment_offset)
1466                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
1467
1468         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1469                                    struct virtio_blk_config, min_io_size,
1470                                    &min_io_size);
1471         if (!err && min_io_size)
1472                 blk_queue_io_min(q, blk_size * min_io_size);
1473
1474         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1475                                    struct virtio_blk_config, opt_io_size,
1476                                    &opt_io_size);
1477         if (!err && opt_io_size)
1478                 blk_queue_io_opt(q, blk_size * opt_io_size);
1479
1480         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1481                 virtio_cread(vdev, struct virtio_blk_config,
1482                              discard_sector_alignment, &discard_granularity);
1483
1484                 virtio_cread(vdev, struct virtio_blk_config,
1485                              max_discard_sectors, &v);
1486                 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
1487
1488                 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1489                              &max_discard_segs);
1490         }
1491
1492         if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1493                 virtio_cread(vdev, struct virtio_blk_config,
1494                              max_write_zeroes_sectors, &v);
1495                 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
1496         }
1497
1498         /* The discard and secure erase limits are combined since the Linux
1499          * block layer uses the same limit for both commands.
1500          *
1501          * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1502          * are negotiated, we will use the minimum between the limits.
1503          *
1504          * discard sector alignment is set to the minimum between discard_sector_alignment
1505          * and secure_erase_sector_alignment.
1506          *
1507          * max discard sectors is set to the minimum between max_discard_seg and
1508          * max_secure_erase_seg.
1509          */
1510         if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1511
1512                 virtio_cread(vdev, struct virtio_blk_config,
1513                              secure_erase_sector_alignment, &v);
1514
1515                 /* secure_erase_sector_alignment should not be zero, the device should set a
1516                  * valid number of sectors.
1517                  */
1518                 if (!v) {
1519                         dev_err(&vdev->dev,
1520                                 "virtio_blk: secure_erase_sector_alignment can't be 0\n");
1521                         err = -EINVAL;
1522                         goto out_cleanup_disk;
1523                 }
1524
1525                 discard_granularity = min_not_zero(discard_granularity, v);
1526
1527                 virtio_cread(vdev, struct virtio_blk_config,
1528                              max_secure_erase_sectors, &v);
1529
1530                 /* max_secure_erase_sectors should not be zero, the device should set a
1531                  * valid number of sectors.
1532                  */
1533                 if (!v) {
1534                         dev_err(&vdev->dev,
1535                                 "virtio_blk: max_secure_erase_sectors can't be 0\n");
1536                         err = -EINVAL;
1537                         goto out_cleanup_disk;
1538                 }
1539
1540                 blk_queue_max_secure_erase_sectors(q, v);
1541
1542                 virtio_cread(vdev, struct virtio_blk_config,
1543                              max_secure_erase_seg, &v);
1544
1545                 /* max_secure_erase_seg should not be zero, the device should set a
1546                  * valid number of segments
1547                  */
1548                 if (!v) {
1549                         dev_err(&vdev->dev,
1550                                 "virtio_blk: max_secure_erase_seg can't be 0\n");
1551                         err = -EINVAL;
1552                         goto out_cleanup_disk;
1553                 }
1554
1555                 max_discard_segs = min_not_zero(max_discard_segs, v);
1556         }
1557
1558         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1559             virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1560                 /* max_discard_seg and discard_granularity will be 0 only
1561                  * if max_discard_seg and discard_sector_alignment fields in the virtio
1562                  * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1563                  * In this case, we use default values.
1564                  */
1565                 if (!max_discard_segs)
1566                         max_discard_segs = sg_elems;
1567
1568                 blk_queue_max_discard_segments(q,
1569                                                min(max_discard_segs, MAX_DISCARD_SEGMENTS));
1570
1571                 if (discard_granularity)
1572                         q->limits.discard_granularity = discard_granularity << SECTOR_SHIFT;
1573                 else
1574                         q->limits.discard_granularity = blk_size;
1575         }
1576
1577         virtblk_update_capacity(vblk, false);
1578         virtio_device_ready(vdev);
1579
1580         /*
1581          * All steps that follow use the VQs therefore they need to be
1582          * placed after the virtio_device_ready() call above.
1583          */
1584         if (virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED)) {
1585                 err = virtblk_probe_zoned_device(vdev, vblk, q);
1586                 if (err)
1587                         goto out_cleanup_disk;
1588         }
1589
1590         err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
1591         if (err)
1592                 goto out_cleanup_disk;
1593
1594         return 0;
1595
1596 out_cleanup_disk:
1597         put_disk(vblk->disk);
1598 out_free_tags:
1599         blk_mq_free_tag_set(&vblk->tag_set);
1600 out_free_vq:
1601         vdev->config->del_vqs(vdev);
1602         kfree(vblk->vqs);
1603 out_free_vblk:
1604         kfree(vblk);
1605 out_free_index:
1606         ida_free(&vd_index_ida, index);
1607 out:
1608         return err;
1609 }
1610
1611 static void virtblk_remove(struct virtio_device *vdev)
1612 {
1613         struct virtio_blk *vblk = vdev->priv;
1614
1615         /* Make sure no work handler is accessing the device. */
1616         flush_work(&vblk->config_work);
1617
1618         del_gendisk(vblk->disk);
1619         blk_mq_free_tag_set(&vblk->tag_set);
1620
1621         mutex_lock(&vblk->vdev_mutex);
1622
1623         /* Stop all the virtqueues. */
1624         virtio_reset_device(vdev);
1625
1626         /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1627         vblk->vdev = NULL;
1628
1629         vdev->config->del_vqs(vdev);
1630         kfree(vblk->vqs);
1631
1632         mutex_unlock(&vblk->vdev_mutex);
1633
1634         put_disk(vblk->disk);
1635 }
1636
1637 #ifdef CONFIG_PM_SLEEP
1638 static int virtblk_freeze(struct virtio_device *vdev)
1639 {
1640         struct virtio_blk *vblk = vdev->priv;
1641
1642         /* Ensure we don't receive any more interrupts */
1643         virtio_reset_device(vdev);
1644
1645         /* Make sure no work handler is accessing the device. */
1646         flush_work(&vblk->config_work);
1647
1648         blk_mq_quiesce_queue(vblk->disk->queue);
1649
1650         vdev->config->del_vqs(vdev);
1651         kfree(vblk->vqs);
1652
1653         return 0;
1654 }
1655
1656 static int virtblk_restore(struct virtio_device *vdev)
1657 {
1658         struct virtio_blk *vblk = vdev->priv;
1659         int ret;
1660
1661         ret = init_vq(vdev->priv);
1662         if (ret)
1663                 return ret;
1664
1665         virtio_device_ready(vdev);
1666
1667         blk_mq_unquiesce_queue(vblk->disk->queue);
1668         return 0;
1669 }
1670 #endif
1671
1672 static const struct virtio_device_id id_table[] = {
1673         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1674         { 0 },
1675 };
1676
1677 static unsigned int features_legacy[] = {
1678         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1679         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1680         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1681         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1682         VIRTIO_BLK_F_SECURE_ERASE,
1683 }
1684 ;
1685 static unsigned int features[] = {
1686         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1687         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1688         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1689         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1690         VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED,
1691 };
1692
1693 static struct virtio_driver virtio_blk = {
1694         .feature_table                  = features,
1695         .feature_table_size             = ARRAY_SIZE(features),
1696         .feature_table_legacy           = features_legacy,
1697         .feature_table_size_legacy      = ARRAY_SIZE(features_legacy),
1698         .driver.name                    = KBUILD_MODNAME,
1699         .driver.owner                   = THIS_MODULE,
1700         .id_table                       = id_table,
1701         .probe                          = virtblk_probe,
1702         .remove                         = virtblk_remove,
1703         .config_changed                 = virtblk_config_changed,
1704 #ifdef CONFIG_PM_SLEEP
1705         .freeze                         = virtblk_freeze,
1706         .restore                        = virtblk_restore,
1707 #endif
1708 };
1709
1710 static int __init virtio_blk_init(void)
1711 {
1712         int error;
1713
1714         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1715         if (!virtblk_wq)
1716                 return -ENOMEM;
1717
1718         major = register_blkdev(0, "virtblk");
1719         if (major < 0) {
1720                 error = major;
1721                 goto out_destroy_workqueue;
1722         }
1723
1724         error = register_virtio_driver(&virtio_blk);
1725         if (error)
1726                 goto out_unregister_blkdev;
1727         return 0;
1728
1729 out_unregister_blkdev:
1730         unregister_blkdev(major, "virtblk");
1731 out_destroy_workqueue:
1732         destroy_workqueue(virtblk_wq);
1733         return error;
1734 }
1735
1736 static void __exit virtio_blk_fini(void)
1737 {
1738         unregister_virtio_driver(&virtio_blk);
1739         unregister_blkdev(major, "virtblk");
1740         destroy_workqueue(virtblk_wq);
1741 }
1742 module_init(virtio_blk_init);
1743 module_exit(virtio_blk_fini);
1744
1745 MODULE_DEVICE_TABLE(virtio, id_table);
1746 MODULE_DESCRIPTION("Virtio block driver");
1747 MODULE_LICENSE("GPL");