drivers/block/nbd.c

   1 /*
   2  * Network block device - make block devices work over TCP
   3  *
   4  * Note that you can not swap over this thing, yet. Seems to work but
   5  * deadlocks sometimes - you can not swap over TCP in general.
   6  *
   7  * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
   8  * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
   9  *
  10  * This file is released under GPLv2 or later.
  11  *
  12  * (part of code stolen from loop.c)
  13  */
  14
  15 #include <linux/major.h>
  16
  17 #include <linux/blkdev.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/sched.h>
  21 #include <linux/fs.h>
  22 #include <linux/bio.h>
  23 #include <linux/stat.h>
  24 #include <linux/errno.h>
  25 #include <linux/file.h>
  26 #include <linux/ioctl.h>
  27 #include <linux/mutex.h>
  28 #include <linux/compiler.h>
  29 #include <linux/err.h>
  30 #include <linux/kernel.h>
  31 #include <linux/slab.h>
  32 #include <net/sock.h>
  33 #include <linux/net.h>
  34 #include <linux/kthread.h>
  35 #include <linux/types.h>
  36 #include <linux/debugfs.h>
  37 #include <linux/blk-mq.h>
  38
  39 #include <asm/uaccess.h>
  40 #include <asm/types.h>
  41
  42 #include <linux/nbd.h>
  43
  44 #define NBD_TIMEDOUT                    0
  45 #define NBD_DISCONNECT_REQUESTED        1
  46
  47 struct nbd_device {
  48         u32 flags;
  49         unsigned long runtime_flags;
  50         struct socket * sock;   /* If == NULL, device is not ready, yet */
  51         int magic;
  52
  53         struct blk_mq_tag_set tag_set;
  54
  55         struct mutex tx_lock;
  56         struct gendisk *disk;
  57         int blksize;
  58         loff_t bytesize;
  59
  60         /* protects initialization and shutdown of the socket */
  61         spinlock_t sock_lock;
  62         struct task_struct *task_recv;
  63         struct task_struct *task_send;
  64
  65 #if IS_ENABLED(CONFIG_DEBUG_FS)
  66         struct dentry *dbg_dir;
  67 #endif
  68 };
  69
  70 struct nbd_cmd {
  71         struct nbd_device *nbd;
  72         struct list_head list;
  73 };
  74
  75 #if IS_ENABLED(CONFIG_DEBUG_FS)
  76 static struct dentry *nbd_dbg_dir;
  77 #endif
  78
  79 #define nbd_name(nbd) ((nbd)->disk->disk_name)
  80
  81 #define NBD_MAGIC 0x68797548
  82
  83 static unsigned int nbds_max = 16;
  84 static struct nbd_device *nbd_dev;
  85 static int max_part;
  86
  87 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
  88 {
  89         return disk_to_dev(nbd->disk);
  90 }
  91
  92 static bool nbd_is_connected(struct nbd_device *nbd)
  93 {
  94         return !!nbd->task_recv;
  95 }
  96
  97 static const char *nbdcmd_to_ascii(int cmd)
  98 {
  99         switch (cmd) {
 100         case  NBD_CMD_READ: return "read";
 101         case NBD_CMD_WRITE: return "write";
 102         case  NBD_CMD_DISC: return "disconnect";
 103         case NBD_CMD_FLUSH: return "flush";
 104         case  NBD_CMD_TRIM: return "trim/discard";
 105         }
 106         return "invalid";
 107 }
 108
 109 static int nbd_size_clear(struct nbd_device *nbd, struct block_device *bdev)
 110 {
 111         bdev->bd_inode->i_size = 0;
 112         set_capacity(nbd->disk, 0);
 113         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 114
 115         return 0;
 116 }
 117
 118 static void nbd_size_update(struct nbd_device *nbd, struct block_device *bdev)
 119 {
 120         if (!nbd_is_connected(nbd))
 121                 return;
 122
 123         bdev->bd_inode->i_size = nbd->bytesize;
 124         set_capacity(nbd->disk, nbd->bytesize >> 9);
 125         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 126 }
 127
 128 static int nbd_size_set(struct nbd_device *nbd, struct block_device *bdev,
 129                         int blocksize, int nr_blocks)
 130 {
 131         int ret;
 132
 133         ret = set_blocksize(bdev, blocksize);
 134         if (ret)
 135                 return ret;
 136
 137         nbd->blksize = blocksize;
 138         nbd->bytesize = (loff_t)blocksize * (loff_t)nr_blocks;
 139
 140         nbd_size_update(nbd, bdev);
 141
 142         return 0;
 143 }
 144
 145 static void nbd_end_request(struct nbd_cmd *cmd)
 146 {
 147         struct nbd_device *nbd = cmd->nbd;
 148         struct request *req = blk_mq_rq_from_pdu(cmd);
 149         int error = req->errors ? -EIO : 0;
 150
 151         dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", cmd,
 152                 error ? "failed" : "done");
 153
 154         blk_mq_complete_request(req, error);
 155 }
 156
 157 /*
 158  * Forcibly shutdown the socket causing all listeners to error
 159  */
 160 static void sock_shutdown(struct nbd_device *nbd)
 161 {
 162         struct socket *sock;
 163
 164         spin_lock(&nbd->sock_lock);
 165
 166         if (!nbd->sock) {
 167                 spin_unlock_irq(&nbd->sock_lock);
 168                 return;
 169         }
 170
 171         sock = nbd->sock;
 172         dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n");
 173         nbd->sock = NULL;
 174         spin_unlock(&nbd->sock_lock);
 175
 176         kernel_sock_shutdown(sock, SHUT_RDWR);
 177         sockfd_put(sock);
 178 }
 179
 180 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 181                                                  bool reserved)
 182 {
 183         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 184         struct nbd_device *nbd = cmd->nbd;
 185         struct socket *sock = NULL;
 186
 187         spin_lock(&nbd->sock_lock);
 188
 189         set_bit(NBD_TIMEDOUT, &nbd->runtime_flags);
 190
 191         if (nbd->sock) {
 192                 sock = nbd->sock;
 193                 get_file(sock->file);
 194         }
 195
 196         spin_unlock(&nbd->sock_lock);
 197         if (sock) {
 198                 kernel_sock_shutdown(sock, SHUT_RDWR);
 199                 sockfd_put(sock);
 200         }
 201
 202         req->errors++;
 203         dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n");
 204         return BLK_EH_HANDLED;
 205 }
 206
 207 /*
 208  *  Send or receive packet.
 209  */
 210 static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size,
 211                 int msg_flags)
 212 {
 213         struct socket *sock = nbd->sock;
 214         int result;
 215         struct msghdr msg;
 216         struct kvec iov;
 217         unsigned long pflags = current->flags;
 218
 219         if (unlikely(!sock)) {
 220                 dev_err(disk_to_dev(nbd->disk),
 221                         "Attempted %s on closed socket in sock_xmit\n",
 222                         (send ? "send" : "recv"));
 223                 return -EINVAL;
 224         }
 225
 226         current->flags |= PF_MEMALLOC;
 227         do {
 228                 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
 229                 iov.iov_base = buf;
 230                 iov.iov_len = size;
 231                 msg.msg_name = NULL;
 232                 msg.msg_namelen = 0;
 233                 msg.msg_control = NULL;
 234                 msg.msg_controllen = 0;
 235                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 236
 237                 if (send)
 238                         result = kernel_sendmsg(sock, &msg, &iov, 1, size);
 239                 else
 240                         result = kernel_recvmsg(sock, &msg, &iov, 1, size,
 241                                                 msg.msg_flags);
 242
 243                 if (result <= 0) {
 244                         if (result == 0)
 245                                 result = -EPIPE; /* short read */
 246                         break;
 247                 }
 248                 size -= result;
 249                 buf += result;
 250         } while (size > 0);
 251
 252         tsk_restore_flags(current, pflags, PF_MEMALLOC);
 253
 254         return result;
 255 }
 256
 257 static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec,
 258                 int flags)
 259 {
 260         int result;
 261         void *kaddr = kmap(bvec->bv_page);
 262         result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset,
 263                            bvec->bv_len, flags);
 264         kunmap(bvec->bv_page);
 265         return result;
 266 }
 267
 268 /* always call with the tx_lock held */
 269 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd)
 270 {
 271         struct request *req = blk_mq_rq_from_pdu(cmd);
 272         int result, flags;
 273         struct nbd_request request;
 274         unsigned long size = blk_rq_bytes(req);
 275         u32 type;
 276
 277         if (req->cmd_type == REQ_TYPE_DRV_PRIV)
 278                 type = NBD_CMD_DISC;
 279         else if (req_op(req) == REQ_OP_DISCARD)
 280                 type = NBD_CMD_TRIM;
 281         else if (req_op(req) == REQ_OP_FLUSH)
 282                 type = NBD_CMD_FLUSH;
 283         else if (rq_data_dir(req) == WRITE)
 284                 type = NBD_CMD_WRITE;
 285         else
 286                 type = NBD_CMD_READ;
 287
 288         memset(&request, 0, sizeof(request));
 289         request.magic = htonl(NBD_REQUEST_MAGIC);
 290         request.type = htonl(type);
 291         if (type != NBD_CMD_FLUSH && type != NBD_CMD_DISC) {
 292                 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
 293                 request.len = htonl(size);
 294         }
 295         memcpy(request.handle, &req->tag, sizeof(req->tag));
 296
 297         dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
 298                 cmd, nbdcmd_to_ascii(type),
 299                 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
 300         result = sock_xmit(nbd, 1, &request, sizeof(request),
 301                         (type == NBD_CMD_WRITE) ? MSG_MORE : 0);
 302         if (result <= 0) {
 303                 dev_err(disk_to_dev(nbd->disk),
 304                         "Send control failed (result %d)\n", result);
 305                 return -EIO;
 306         }
 307
 308         if (type == NBD_CMD_WRITE) {
 309                 struct req_iterator iter;
 310                 struct bio_vec bvec;
 311                 /*
 312                  * we are really probing at internals to determine
 313                  * whether to set MSG_MORE or not...
 314                  */
 315                 rq_for_each_segment(bvec, req, iter) {
 316                         flags = 0;
 317                         if (!rq_iter_last(bvec, iter))
 318                                 flags = MSG_MORE;
 319                         dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
 320                                 cmd, bvec.bv_len);
 321                         result = sock_send_bvec(nbd, &bvec, flags);
 322                         if (result <= 0) {
 323                                 dev_err(disk_to_dev(nbd->disk),
 324                                         "Send data failed (result %d)\n",
 325                                         result);
 326                                 return -EIO;
 327                         }
 328                 }
 329         }
 330         return 0;
 331 }
 332
 333 static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec)
 334 {
 335         int result;
 336         void *kaddr = kmap(bvec->bv_page);
 337         result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len,
 338                         MSG_WAITALL);
 339         kunmap(bvec->bv_page);
 340         return result;
 341 }
 342
 343 /* NULL returned = something went wrong, inform userspace */
 344 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd)
 345 {
 346         int result;
 347         struct nbd_reply reply;
 348         struct nbd_cmd *cmd;
 349         struct request *req = NULL;
 350         u16 hwq;
 351         int tag;
 352
 353         reply.magic = 0;
 354         result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL);
 355         if (result <= 0) {
 356                 dev_err(disk_to_dev(nbd->disk),
 357                         "Receive control failed (result %d)\n", result);
 358                 return ERR_PTR(result);
 359         }
 360
 361         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 362                 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
 363                                 (unsigned long)ntohl(reply.magic));
 364                 return ERR_PTR(-EPROTO);
 365         }
 366
 367         memcpy(&tag, reply.handle, sizeof(int));
 368
 369         hwq = blk_mq_unique_tag_to_hwq(tag);
 370         if (hwq < nbd->tag_set.nr_hw_queues)
 371                 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
 372                                        blk_mq_unique_tag_to_tag(tag));
 373         if (!req || !blk_mq_request_started(req)) {
 374                 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
 375                         tag, req);
 376                 return ERR_PTR(-ENOENT);
 377         }
 378         cmd = blk_mq_rq_to_pdu(req);
 379
 380         if (ntohl(reply.error)) {
 381                 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
 382                         ntohl(reply.error));
 383                 req->errors++;
 384                 return cmd;
 385         }
 386
 387         dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
 388         if (rq_data_dir(req) != WRITE) {
 389                 struct req_iterator iter;
 390                 struct bio_vec bvec;
 391
 392                 rq_for_each_segment(bvec, req, iter) {
 393                         result = sock_recv_bvec(nbd, &bvec);
 394                         if (result <= 0) {
 395                                 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 396                                         result);
 397                                 req->errors++;
 398                                 return cmd;
 399                         }
 400                         dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 401                                 cmd, bvec.bv_len);
 402                 }
 403         }
 404         return cmd;
 405 }
 406
 407 static ssize_t pid_show(struct device *dev,
 408                         struct device_attribute *attr, char *buf)
 409 {
 410         struct gendisk *disk = dev_to_disk(dev);
 411         struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
 412
 413         return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
 414 }
 415
 416 static struct device_attribute pid_attr = {
 417         .attr = { .name = "pid", .mode = S_IRUGO},
 418         .show = pid_show,
 419 };
 420
 421 static int nbd_thread_recv(struct nbd_device *nbd, struct block_device *bdev)
 422 {
 423         struct nbd_cmd *cmd;
 424         int ret;
 425
 426         BUG_ON(nbd->magic != NBD_MAGIC);
 427
 428         sk_set_memalloc(nbd->sock->sk);
 429
 430         ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
 431         if (ret) {
 432                 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
 433                 return ret;
 434         }
 435
 436         nbd_size_update(nbd, bdev);
 437
 438         while (1) {
 439                 cmd = nbd_read_stat(nbd);
 440                 if (IS_ERR(cmd)) {
 441                         ret = PTR_ERR(cmd);
 442                         break;
 443                 }
 444
 445                 nbd_end_request(cmd);
 446         }
 447
 448         nbd_size_clear(nbd, bdev);
 449
 450         device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
 451         return ret;
 452 }
 453
 454 static void nbd_clear_req(struct request *req, void *data, bool reserved)
 455 {
 456         struct nbd_cmd *cmd;
 457
 458         if (!blk_mq_request_started(req))
 459                 return;
 460         cmd = blk_mq_rq_to_pdu(req);
 461         req->errors++;
 462         nbd_end_request(cmd);
 463 }
 464
 465 static void nbd_clear_que(struct nbd_device *nbd)
 466 {
 467         BUG_ON(nbd->magic != NBD_MAGIC);
 468
 469         /*
 470          * Because we have set nbd->sock to NULL under the tx_lock, all
 471          * modifications to the list must have completed by now.
 472          */
 473         BUG_ON(nbd->sock);
 474
 475         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 476         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 477 }
 478
 479
 480 static void nbd_handle_cmd(struct nbd_cmd *cmd)
 481 {
 482         struct request *req = blk_mq_rq_from_pdu(cmd);
 483         struct nbd_device *nbd = cmd->nbd;
 484
 485         if (req->cmd_type != REQ_TYPE_FS)
 486                 goto error_out;
 487
 488         if (rq_data_dir(req) == WRITE &&
 489             (nbd->flags & NBD_FLAG_READ_ONLY)) {
 490                 dev_err(disk_to_dev(nbd->disk),
 491                         "Write on read-only\n");
 492                 goto error_out;
 493         }
 494
 495         req->errors = 0;
 496
 497         mutex_lock(&nbd->tx_lock);
 498         nbd->task_send = current;
 499         if (unlikely(!nbd->sock)) {
 500                 mutex_unlock(&nbd->tx_lock);
 501                 dev_err(disk_to_dev(nbd->disk),
 502                         "Attempted send on closed socket\n");
 503                 goto error_out;
 504         }
 505
 506         if (nbd_send_cmd(nbd, cmd) != 0) {
 507                 dev_err(disk_to_dev(nbd->disk), "Request send failed\n");
 508                 req->errors++;
 509                 nbd_end_request(cmd);
 510         }
 511
 512         nbd->task_send = NULL;
 513         mutex_unlock(&nbd->tx_lock);
 514
 515         return;
 516
 517 error_out:
 518         req->errors++;
 519         nbd_end_request(cmd);
 520 }
 521
 522 static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 523                         const struct blk_mq_queue_data *bd)
 524 {
 525         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 526
 527         blk_mq_start_request(bd->rq);
 528         nbd_handle_cmd(cmd);
 529         return BLK_MQ_RQ_QUEUE_OK;
 530 }
 531
 532 static int nbd_set_socket(struct nbd_device *nbd, struct socket *sock)
 533 {
 534         int ret = 0;
 535
 536         spin_lock_irq(&nbd->sock_lock);
 537
 538         if (nbd->sock) {
 539                 ret = -EBUSY;
 540                 goto out;
 541         }
 542
 543         nbd->sock = sock;
 544
 545 out:
 546         spin_unlock_irq(&nbd->sock_lock);
 547
 548         return ret;
 549 }
 550
 551 /* Reset all properties of an NBD device */
 552 static void nbd_reset(struct nbd_device *nbd)
 553 {
 554         nbd->runtime_flags = 0;
 555         nbd->blksize = 1024;
 556         nbd->bytesize = 0;
 557         set_capacity(nbd->disk, 0);
 558         nbd->flags = 0;
 559         nbd->tag_set.timeout = 0;
 560         queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 561 }
 562
 563 static void nbd_bdev_reset(struct block_device *bdev)
 564 {
 565         set_device_ro(bdev, false);
 566         bdev->bd_inode->i_size = 0;
 567         if (max_part > 0) {
 568                 blkdev_reread_part(bdev);
 569                 bdev->bd_invalidated = 1;
 570         }
 571 }
 572
 573 static void nbd_parse_flags(struct nbd_device *nbd, struct block_device *bdev)
 574 {
 575         if (nbd->flags & NBD_FLAG_READ_ONLY)
 576                 set_device_ro(bdev, true);
 577         if (nbd->flags & NBD_FLAG_SEND_TRIM)
 578                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 579         if (nbd->flags & NBD_FLAG_SEND_FLUSH)
 580                 blk_queue_write_cache(nbd->disk->queue, true, false);
 581         else
 582                 blk_queue_write_cache(nbd->disk->queue, false, false);
 583 }
 584
 585 static int nbd_dev_dbg_init(struct nbd_device *nbd);
 586 static void nbd_dev_dbg_close(struct nbd_device *nbd);
 587
 588 /* Must be called with tx_lock held */
 589
 590 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
 591                        unsigned int cmd, unsigned long arg)
 592 {
 593         switch (cmd) {
 594         case NBD_DISCONNECT: {
 595                 struct request *sreq;
 596
 597                 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
 598                 if (!nbd->sock)
 599                         return -EINVAL;
 600
 601                 sreq = blk_mq_alloc_request(bdev_get_queue(bdev), WRITE, 0);
 602                 if (!sreq)
 603                         return -ENOMEM;
 604
 605                 mutex_unlock(&nbd->tx_lock);
 606                 fsync_bdev(bdev);
 607                 mutex_lock(&nbd->tx_lock);
 608                 sreq->cmd_type = REQ_TYPE_DRV_PRIV;
 609
 610                 /* Check again after getting mutex back.  */
 611                 if (!nbd->sock) {
 612                         blk_mq_free_request(sreq);
 613                         return -EINVAL;
 614                 }
 615
 616                 set_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags);
 617
 618                 nbd_send_cmd(nbd, blk_mq_rq_to_pdu(sreq));
 619                 blk_mq_free_request(sreq);
 620                 return 0;
 621         }
 622
 623         case NBD_CLEAR_SOCK:
 624                 sock_shutdown(nbd);
 625                 nbd_clear_que(nbd);
 626                 kill_bdev(bdev);
 627                 return 0;
 628
 629         case NBD_SET_SOCK: {
 630                 int err;
 631                 struct socket *sock = sockfd_lookup(arg, &err);
 632
 633                 if (!sock)
 634                         return err;
 635
 636                 err = nbd_set_socket(nbd, sock);
 637                 if (!err && max_part)
 638                         bdev->bd_invalidated = 1;
 639
 640                 return err;
 641         }
 642
 643         case NBD_SET_BLKSIZE: {
 644                 loff_t bsize = div_s64(nbd->bytesize, arg);
 645
 646                 return nbd_size_set(nbd, bdev, arg, bsize);
 647         }
 648
 649         case NBD_SET_SIZE:
 650                 return nbd_size_set(nbd, bdev, nbd->blksize,
 651                                     arg / nbd->blksize);
 652
 653         case NBD_SET_SIZE_BLOCKS:
 654                 return nbd_size_set(nbd, bdev, nbd->blksize, arg);
 655
 656         case NBD_SET_TIMEOUT:
 657                 nbd->tag_set.timeout = arg * HZ;
 658                 return 0;
 659
 660         case NBD_SET_FLAGS:
 661                 nbd->flags = arg;
 662                 return 0;
 663
 664         case NBD_DO_IT: {
 665                 int error;
 666
 667                 if (nbd->task_recv)
 668                         return -EBUSY;
 669                 if (!nbd->sock)
 670                         return -EINVAL;
 671
 672                 /* We have to claim the device under the lock */
 673                 nbd->task_recv = current;
 674                 mutex_unlock(&nbd->tx_lock);
 675
 676                 nbd_parse_flags(nbd, bdev);
 677
 678                 nbd_dev_dbg_init(nbd);
 679                 error = nbd_thread_recv(nbd, bdev);
 680                 nbd_dev_dbg_close(nbd);
 681
 682                 mutex_lock(&nbd->tx_lock);
 683                 nbd->task_recv = NULL;
 684
 685                 sock_shutdown(nbd);
 686                 nbd_clear_que(nbd);
 687                 kill_bdev(bdev);
 688                 nbd_bdev_reset(bdev);
 689
 690                 /* user requested, ignore socket errors */
 691                 if (test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
 692                         error = 0;
 693                 if (test_bit(NBD_TIMEDOUT, &nbd->runtime_flags))
 694                         error = -ETIMEDOUT;
 695
 696                 nbd_reset(nbd);
 697
 698                 return error;
 699         }
 700
 701         case NBD_CLEAR_QUE:
 702                 /*
 703                  * This is for compatibility only.  The queue is always cleared
 704                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
 705                  */
 706                 return 0;
 707
 708         case NBD_PRINT_DEBUG:
 709                 /*
 710                  * For compatibility only, we no longer keep a list of
 711                  * outstanding requests.
 712                  */
 713                 return 0;
 714         }
 715         return -ENOTTY;
 716 }
 717
 718 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
 719                      unsigned int cmd, unsigned long arg)
 720 {
 721         struct nbd_device *nbd = bdev->bd_disk->private_data;
 722         int error;
 723
 724         if (!capable(CAP_SYS_ADMIN))
 725                 return -EPERM;
 726
 727         BUG_ON(nbd->magic != NBD_MAGIC);
 728
 729         mutex_lock(&nbd->tx_lock);
 730         error = __nbd_ioctl(bdev, nbd, cmd, arg);
 731         mutex_unlock(&nbd->tx_lock);
 732
 733         return error;
 734 }
 735
 736 static const struct block_device_operations nbd_fops =
 737 {
 738         .owner =        THIS_MODULE,
 739         .ioctl =        nbd_ioctl,
 740         .compat_ioctl = nbd_ioctl,
 741 };
 742
 743 #if IS_ENABLED(CONFIG_DEBUG_FS)
 744
 745 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
 746 {
 747         struct nbd_device *nbd = s->private;
 748
 749         if (nbd->task_recv)
 750                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
 751         if (nbd->task_send)
 752                 seq_printf(s, "send: %d\n", task_pid_nr(nbd->task_send));
 753
 754         return 0;
 755 }
 756
 757 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
 758 {
 759         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
 760 }
 761
 762 static const struct file_operations nbd_dbg_tasks_ops = {
 763         .open = nbd_dbg_tasks_open,
 764         .read = seq_read,
 765         .llseek = seq_lseek,
 766         .release = single_release,
 767 };
 768
 769 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
 770 {
 771         struct nbd_device *nbd = s->private;
 772         u32 flags = nbd->flags;
 773
 774         seq_printf(s, "Hex: 0x%08x\n\n", flags);
 775
 776         seq_puts(s, "Known flags:\n");
 777
 778         if (flags & NBD_FLAG_HAS_FLAGS)
 779                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
 780         if (flags & NBD_FLAG_READ_ONLY)
 781                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
 782         if (flags & NBD_FLAG_SEND_FLUSH)
 783                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
 784         if (flags & NBD_FLAG_SEND_TRIM)
 785                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
 786
 787         return 0;
 788 }
 789
 790 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
 791 {
 792         return single_open(file, nbd_dbg_flags_show, inode->i_private);
 793 }
 794
 795 static const struct file_operations nbd_dbg_flags_ops = {
 796         .open = nbd_dbg_flags_open,
 797         .read = seq_read,
 798         .llseek = seq_lseek,
 799         .release = single_release,
 800 };
 801
 802 static int nbd_dev_dbg_init(struct nbd_device *nbd)
 803 {
 804         struct dentry *dir;
 805
 806         if (!nbd_dbg_dir)
 807                 return -EIO;
 808
 809         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
 810         if (!dir) {
 811                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
 812                         nbd_name(nbd));
 813                 return -EIO;
 814         }
 815         nbd->dbg_dir = dir;
 816
 817         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
 818         debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize);
 819         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
 820         debugfs_create_u32("blocksize", 0444, dir, &nbd->blksize);
 821         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
 822
 823         return 0;
 824 }
 825
 826 static void nbd_dev_dbg_close(struct nbd_device *nbd)
 827 {
 828         debugfs_remove_recursive(nbd->dbg_dir);
 829 }
 830
 831 static int nbd_dbg_init(void)
 832 {
 833         struct dentry *dbg_dir;
 834
 835         dbg_dir = debugfs_create_dir("nbd", NULL);
 836         if (!dbg_dir)
 837                 return -EIO;
 838
 839         nbd_dbg_dir = dbg_dir;
 840
 841         return 0;
 842 }
 843
 844 static void nbd_dbg_close(void)
 845 {
 846         debugfs_remove_recursive(nbd_dbg_dir);
 847 }
 848
 849 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
 850
 851 static int nbd_dev_dbg_init(struct nbd_device *nbd)
 852 {
 853         return 0;
 854 }
 855
 856 static void nbd_dev_dbg_close(struct nbd_device *nbd)
 857 {
 858 }
 859
 860 static int nbd_dbg_init(void)
 861 {
 862         return 0;
 863 }
 864
 865 static void nbd_dbg_close(void)
 866 {
 867 }
 868
 869 #endif
 870
 871 static int nbd_init_request(void *data, struct request *rq,
 872                             unsigned int hctx_idx, unsigned int request_idx,
 873                             unsigned int numa_node)
 874 {
 875         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
 876
 877         cmd->nbd = data;
 878         INIT_LIST_HEAD(&cmd->list);
 879         return 0;
 880 }
 881
 882 static struct blk_mq_ops nbd_mq_ops = {
 883         .queue_rq       = nbd_queue_rq,
 884         .init_request   = nbd_init_request,
 885         .timeout        = nbd_xmit_timeout,
 886 };
 887
 888 /*
 889  * And here should be modules and kernel interface
 890  *  (Just smiley confuses emacs :-)
 891  */
 892
 893 static int __init nbd_init(void)
 894 {
 895         int err = -ENOMEM;
 896         int i;
 897         int part_shift;
 898
 899         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
 900
 901         if (max_part < 0) {
 902                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
 903                 return -EINVAL;
 904         }
 905
 906         part_shift = 0;
 907         if (max_part > 0) {
 908                 part_shift = fls(max_part);
 909
 910                 /*
 911                  * Adjust max_part according to part_shift as it is exported
 912                  * to user space so that user can know the max number of
 913                  * partition kernel should be able to manage.
 914                  *
 915                  * Note that -1 is required because partition 0 is reserved
 916                  * for the whole disk.
 917                  */
 918                 max_part = (1UL << part_shift) - 1;
 919         }
 920
 921         if ((1UL << part_shift) > DISK_MAX_PARTS)
 922                 return -EINVAL;
 923
 924         if (nbds_max > 1UL << (MINORBITS - part_shift))
 925                 return -EINVAL;
 926
 927         nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
 928         if (!nbd_dev)
 929                 return -ENOMEM;
 930
 931         for (i = 0; i < nbds_max; i++) {
 932                 struct gendisk *disk = alloc_disk(1 << part_shift);
 933                 if (!disk)
 934                         goto out;
 935                 nbd_dev[i].disk = disk;
 936
 937                 nbd_dev[i].tag_set.ops = &nbd_mq_ops;
 938                 nbd_dev[i].tag_set.nr_hw_queues = 1;
 939                 nbd_dev[i].tag_set.queue_depth = 128;
 940                 nbd_dev[i].tag_set.numa_node = NUMA_NO_NODE;
 941                 nbd_dev[i].tag_set.cmd_size = sizeof(struct nbd_cmd);
 942                 nbd_dev[i].tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
 943                         BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
 944                 nbd_dev[i].tag_set.driver_data = &nbd_dev[i];
 945
 946                 err = blk_mq_alloc_tag_set(&nbd_dev[i].tag_set);
 947                 if (err) {
 948                         put_disk(disk);
 949                         goto out;
 950                 }
 951
 952                 /*
 953                  * The new linux 2.5 block layer implementation requires
 954                  * every gendisk to have its very own request_queue struct.
 955                  * These structs are big so we dynamically allocate them.
 956                  */
 957                 disk->queue = blk_mq_init_queue(&nbd_dev[i].tag_set);
 958                 if (!disk->queue) {
 959                         blk_mq_free_tag_set(&nbd_dev[i].tag_set);
 960                         put_disk(disk);
 961                         goto out;
 962                 }
 963
 964                 /*
 965                  * Tell the block layer that we are not a rotational device
 966                  */
 967                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
 968                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
 969                 disk->queue->limits.discard_granularity = 512;
 970                 blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
 971                 disk->queue->limits.discard_zeroes_data = 0;
 972                 blk_queue_max_hw_sectors(disk->queue, 65536);
 973                 disk->queue->limits.max_sectors = 256;
 974         }
 975
 976         if (register_blkdev(NBD_MAJOR, "nbd")) {
 977                 err = -EIO;
 978                 goto out;
 979         }
 980
 981         printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
 982
 983         nbd_dbg_init();
 984
 985         for (i = 0; i < nbds_max; i++) {
 986                 struct gendisk *disk = nbd_dev[i].disk;
 987                 nbd_dev[i].magic = NBD_MAGIC;
 988                 spin_lock_init(&nbd_dev[i].sock_lock);
 989                 mutex_init(&nbd_dev[i].tx_lock);
 990                 disk->major = NBD_MAJOR;
 991                 disk->first_minor = i << part_shift;
 992                 disk->fops = &nbd_fops;
 993                 disk->private_data = &nbd_dev[i];
 994                 sprintf(disk->disk_name, "nbd%d", i);
 995                 nbd_reset(&nbd_dev[i]);
 996                 add_disk(disk);
 997         }
 998
 999         return 0;
1000 out:
1001         while (i--) {
1002                 blk_mq_free_tag_set(&nbd_dev[i].tag_set);
1003                 blk_cleanup_queue(nbd_dev[i].disk->queue);
1004                 put_disk(nbd_dev[i].disk);
1005         }
1006         kfree(nbd_dev);
1007         return err;
1008 }
1009
1010 static void __exit nbd_cleanup(void)
1011 {
1012         int i;
1013
1014         nbd_dbg_close();
1015
1016         for (i = 0; i < nbds_max; i++) {
1017                 struct gendisk *disk = nbd_dev[i].disk;
1018                 nbd_dev[i].magic = 0;
1019                 if (disk) {
1020                         del_gendisk(disk);
1021                         blk_cleanup_queue(disk->queue);
1022                         blk_mq_free_tag_set(&nbd_dev[i].tag_set);
1023                         put_disk(disk);
1024                 }
1025         }
1026         unregister_blkdev(NBD_MAJOR, "nbd");
1027         kfree(nbd_dev);
1028         printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
1029 }
1030
1031 module_init(nbd_init);
1032 module_exit(nbd_cleanup);
1033
1034 MODULE_DESCRIPTION("Network Block Device");
1035 MODULE_LICENSE("GPL");
1036
1037 module_param(nbds_max, int, 0444);
1038 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
1039 module_param(max_part, int, 0444);
1040 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");