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 <linux/uaccess.h>
  40 #include <asm/types.h>
  41
  42 #include <linux/nbd.h>
  43
  44 static DEFINE_IDR(nbd_index_idr);
  45 static DEFINE_MUTEX(nbd_index_mutex);
  46
  47 struct nbd_sock {
  48         struct socket *sock;
  49         struct mutex tx_lock;
  50 };
  51
  52 #define NBD_TIMEDOUT                    0
  53 #define NBD_DISCONNECT_REQUESTED        1
  54 #define NBD_DISCONNECTED                2
  55 #define NBD_RUNNING                     3
  56
  57 struct nbd_device {
  58         u32 flags;
  59         unsigned long runtime_flags;
  60         struct nbd_sock **socks;
  61         int magic;
  62
  63         struct blk_mq_tag_set tag_set;
  64
  65         struct mutex config_lock;
  66         struct gendisk *disk;
  67         int num_connections;
  68         atomic_t recv_threads;
  69         wait_queue_head_t recv_wq;
  70         loff_t blksize;
  71         loff_t bytesize;
  72
  73         struct task_struct *task_recv;
  74         struct task_struct *task_setup;
  75
  76 #if IS_ENABLED(CONFIG_DEBUG_FS)
  77         struct dentry *dbg_dir;
  78 #endif
  79 };
  80
  81 struct nbd_cmd {
  82         struct nbd_device *nbd;
  83         struct completion send_complete;
  84 };
  85
  86 #if IS_ENABLED(CONFIG_DEBUG_FS)
  87 static struct dentry *nbd_dbg_dir;
  88 #endif
  89
  90 #define nbd_name(nbd) ((nbd)->disk->disk_name)
  91
  92 #define NBD_MAGIC 0x68797548
  93
  94 static unsigned int nbds_max = 16;
  95 static int max_part;
  96 static struct workqueue_struct *recv_workqueue;
  97 static int part_shift;
  98
  99 static int nbd_dev_dbg_init(struct nbd_device *nbd);
 100 static void nbd_dev_dbg_close(struct nbd_device *nbd);
 101
 102
 103 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
 104 {
 105         return disk_to_dev(nbd->disk);
 106 }
 107
 108 static bool nbd_is_connected(struct nbd_device *nbd)
 109 {
 110         return !!nbd->task_recv;
 111 }
 112
 113 static const char *nbdcmd_to_ascii(int cmd)
 114 {
 115         switch (cmd) {
 116         case  NBD_CMD_READ: return "read";
 117         case NBD_CMD_WRITE: return "write";
 118         case  NBD_CMD_DISC: return "disconnect";
 119         case NBD_CMD_FLUSH: return "flush";
 120         case  NBD_CMD_TRIM: return "trim/discard";
 121         }
 122         return "invalid";
 123 }
 124
 125 static int nbd_size_clear(struct nbd_device *nbd, struct block_device *bdev)
 126 {
 127         bd_set_size(bdev, 0);
 128         set_capacity(nbd->disk, 0);
 129         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 130
 131         return 0;
 132 }
 133
 134 static void nbd_size_update(struct nbd_device *nbd, struct block_device *bdev)
 135 {
 136         blk_queue_logical_block_size(nbd->disk->queue, nbd->blksize);
 137         blk_queue_physical_block_size(nbd->disk->queue, nbd->blksize);
 138         bd_set_size(bdev, nbd->bytesize);
 139         set_capacity(nbd->disk, nbd->bytesize >> 9);
 140         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 141 }
 142
 143 static void nbd_size_set(struct nbd_device *nbd, struct block_device *bdev,
 144                         loff_t blocksize, loff_t nr_blocks)
 145 {
 146         nbd->blksize = blocksize;
 147         nbd->bytesize = blocksize * nr_blocks;
 148         if (nbd_is_connected(nbd))
 149                 nbd_size_update(nbd, bdev);
 150 }
 151
 152 static void nbd_end_request(struct nbd_cmd *cmd)
 153 {
 154         struct nbd_device *nbd = cmd->nbd;
 155         struct request *req = blk_mq_rq_from_pdu(cmd);
 156         int error = req->errors ? -EIO : 0;
 157
 158         dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", cmd,
 159                 error ? "failed" : "done");
 160
 161         blk_mq_complete_request(req, error);
 162 }
 163
 164 /*
 165  * Forcibly shutdown the socket causing all listeners to error
 166  */
 167 static void sock_shutdown(struct nbd_device *nbd)
 168 {
 169         int i;
 170
 171         if (nbd->num_connections == 0)
 172                 return;
 173         if (test_and_set_bit(NBD_DISCONNECTED, &nbd->runtime_flags))
 174                 return;
 175
 176         for (i = 0; i < nbd->num_connections; i++) {
 177                 struct nbd_sock *nsock = nbd->socks[i];
 178                 mutex_lock(&nsock->tx_lock);
 179                 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
 180                 mutex_unlock(&nsock->tx_lock);
 181         }
 182         dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
 183 }
 184
 185 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 186                                                  bool reserved)
 187 {
 188         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 189         struct nbd_device *nbd = cmd->nbd;
 190
 191         dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n");
 192         set_bit(NBD_TIMEDOUT, &nbd->runtime_flags);
 193         req->errors++;
 194
 195         mutex_lock(&nbd->config_lock);
 196         sock_shutdown(nbd);
 197         mutex_unlock(&nbd->config_lock);
 198         return BLK_EH_HANDLED;
 199 }
 200
 201 /*
 202  *  Send or receive packet.
 203  */
 204 static int sock_xmit(struct nbd_device *nbd, int index, int send,
 205                      struct iov_iter *iter, int msg_flags)
 206 {
 207         struct socket *sock = nbd->socks[index]->sock;
 208         int result;
 209         struct msghdr msg;
 210         unsigned long pflags = current->flags;
 211
 212         if (unlikely(!sock)) {
 213                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 214                         "Attempted %s on closed socket in sock_xmit\n",
 215                         (send ? "send" : "recv"));
 216                 return -EINVAL;
 217         }
 218
 219         msg.msg_iter = *iter;
 220
 221         current->flags |= PF_MEMALLOC;
 222         do {
 223                 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
 224                 msg.msg_name = NULL;
 225                 msg.msg_namelen = 0;
 226                 msg.msg_control = NULL;
 227                 msg.msg_controllen = 0;
 228                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 229
 230                 if (send)
 231                         result = sock_sendmsg(sock, &msg);
 232                 else
 233                         result = sock_recvmsg(sock, &msg, msg.msg_flags);
 234
 235                 if (result <= 0) {
 236                         if (result == 0)
 237                                 result = -EPIPE; /* short read */
 238                         break;
 239                 }
 240         } while (msg_data_left(&msg));
 241
 242         tsk_restore_flags(current, pflags, PF_MEMALLOC);
 243
 244         return result;
 245 }
 246
 247 /* always call with the tx_lock held */
 248 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 249 {
 250         struct request *req = blk_mq_rq_from_pdu(cmd);
 251         int result;
 252         struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
 253         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 254         struct iov_iter from;
 255         unsigned long size = blk_rq_bytes(req);
 256         struct bio *bio;
 257         u32 type;
 258         u32 tag = blk_mq_unique_tag(req);
 259
 260         iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
 261
 262         switch (req_op(req)) {
 263         case REQ_OP_DISCARD:
 264                 type = NBD_CMD_TRIM;
 265                 break;
 266         case REQ_OP_FLUSH:
 267                 type = NBD_CMD_FLUSH;
 268                 break;
 269         case REQ_OP_WRITE:
 270                 type = NBD_CMD_WRITE;
 271                 break;
 272         case REQ_OP_READ:
 273                 type = NBD_CMD_READ;
 274                 break;
 275         default:
 276                 return -EIO;
 277         }
 278
 279         if (rq_data_dir(req) == WRITE &&
 280             (nbd->flags & NBD_FLAG_READ_ONLY)) {
 281                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 282                                     "Write on read-only\n");
 283                 return -EIO;
 284         }
 285
 286         request.type = htonl(type);
 287         if (type != NBD_CMD_FLUSH) {
 288                 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
 289                 request.len = htonl(size);
 290         }
 291         memcpy(request.handle, &tag, sizeof(tag));
 292
 293         dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
 294                 cmd, nbdcmd_to_ascii(type),
 295                 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
 296         result = sock_xmit(nbd, index, 1, &from,
 297                         (type == NBD_CMD_WRITE) ? MSG_MORE : 0);
 298         if (result <= 0) {
 299                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 300                         "Send control failed (result %d)\n", result);
 301                 return -EIO;
 302         }
 303
 304         if (type != NBD_CMD_WRITE)
 305                 return 0;
 306
 307         bio = req->bio;
 308         while (bio) {
 309                 struct bio *next = bio->bi_next;
 310                 struct bvec_iter iter;
 311                 struct bio_vec bvec;
 312
 313                 bio_for_each_segment(bvec, bio, iter) {
 314                         bool is_last = !next && bio_iter_last(bvec, iter);
 315                         int flags = is_last ? 0 : MSG_MORE;
 316
 317                         dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
 318                                 cmd, bvec.bv_len);
 319                         iov_iter_bvec(&from, ITER_BVEC | WRITE,
 320                                       &bvec, 1, bvec.bv_len);
 321                         result = sock_xmit(nbd, index, 1, &from, 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                          * The completion might already have come in,
 330                          * so break for the last one instead of letting
 331                          * the iterator do it. This prevents use-after-free
 332                          * of the bio.
 333                          */
 334                         if (is_last)
 335                                 break;
 336                 }
 337                 bio = next;
 338         }
 339         return 0;
 340 }
 341
 342 /* NULL returned = something went wrong, inform userspace */
 343 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 344 {
 345         int result;
 346         struct nbd_reply reply;
 347         struct nbd_cmd *cmd;
 348         struct request *req = NULL;
 349         u16 hwq;
 350         u32 tag;
 351         struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
 352         struct iov_iter to;
 353
 354         reply.magic = 0;
 355         iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply));
 356         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL);
 357         if (result <= 0) {
 358                 if (!test_bit(NBD_DISCONNECTED, &nbd->runtime_flags) &&
 359                     !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
 360                         dev_err(disk_to_dev(nbd->disk),
 361                                 "Receive control failed (result %d)\n", result);
 362                 return ERR_PTR(result);
 363         }
 364
 365         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 366                 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
 367                                 (unsigned long)ntohl(reply.magic));
 368                 return ERR_PTR(-EPROTO);
 369         }
 370
 371         memcpy(&tag, reply.handle, sizeof(u32));
 372
 373         hwq = blk_mq_unique_tag_to_hwq(tag);
 374         if (hwq < nbd->tag_set.nr_hw_queues)
 375                 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
 376                                        blk_mq_unique_tag_to_tag(tag));
 377         if (!req || !blk_mq_request_started(req)) {
 378                 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
 379                         tag, req);
 380                 return ERR_PTR(-ENOENT);
 381         }
 382         cmd = blk_mq_rq_to_pdu(req);
 383         if (ntohl(reply.error)) {
 384                 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
 385                         ntohl(reply.error));
 386                 req->errors++;
 387                 return cmd;
 388         }
 389
 390         dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
 391         if (rq_data_dir(req) != WRITE) {
 392                 struct req_iterator iter;
 393                 struct bio_vec bvec;
 394
 395                 rq_for_each_segment(bvec, req, iter) {
 396                         iov_iter_bvec(&to, ITER_BVEC | READ,
 397                                       &bvec, 1, bvec.bv_len);
 398                         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL);
 399                         if (result <= 0) {
 400                                 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 401                                         result);
 402                                 req->errors++;
 403                                 return cmd;
 404                         }
 405                         dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 406                                 cmd, bvec.bv_len);
 407                 }
 408         } else {
 409                 /* See the comment in nbd_queue_rq. */
 410                 wait_for_completion(&cmd->send_complete);
 411         }
 412         return cmd;
 413 }
 414
 415 static ssize_t pid_show(struct device *dev,
 416                         struct device_attribute *attr, char *buf)
 417 {
 418         struct gendisk *disk = dev_to_disk(dev);
 419         struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
 420
 421         return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
 422 }
 423
 424 static struct device_attribute pid_attr = {
 425         .attr = { .name = "pid", .mode = S_IRUGO},
 426         .show = pid_show,
 427 };
 428
 429 struct recv_thread_args {
 430         struct work_struct work;
 431         struct nbd_device *nbd;
 432         int index;
 433 };
 434
 435 static void recv_work(struct work_struct *work)
 436 {
 437         struct recv_thread_args *args = container_of(work,
 438                                                      struct recv_thread_args,
 439                                                      work);
 440         struct nbd_device *nbd = args->nbd;
 441         struct nbd_cmd *cmd;
 442         int ret = 0;
 443
 444         BUG_ON(nbd->magic != NBD_MAGIC);
 445         while (1) {
 446                 cmd = nbd_read_stat(nbd, args->index);
 447                 if (IS_ERR(cmd)) {
 448                         ret = PTR_ERR(cmd);
 449                         break;
 450                 }
 451
 452                 nbd_end_request(cmd);
 453         }
 454
 455         /*
 456          * We got an error, shut everybody down if this wasn't the result of a
 457          * disconnect request.
 458          */
 459         if (ret && !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
 460                 sock_shutdown(nbd);
 461         atomic_dec(&nbd->recv_threads);
 462         wake_up(&nbd->recv_wq);
 463 }
 464
 465 static void nbd_clear_req(struct request *req, void *data, bool reserved)
 466 {
 467         struct nbd_cmd *cmd;
 468
 469         if (!blk_mq_request_started(req))
 470                 return;
 471         cmd = blk_mq_rq_to_pdu(req);
 472         req->errors++;
 473         nbd_end_request(cmd);
 474 }
 475
 476 static void nbd_clear_que(struct nbd_device *nbd)
 477 {
 478         BUG_ON(nbd->magic != NBD_MAGIC);
 479
 480         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 481         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 482 }
 483
 484
 485 static void nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 486 {
 487         struct request *req = blk_mq_rq_from_pdu(cmd);
 488         struct nbd_device *nbd = cmd->nbd;
 489         struct nbd_sock *nsock;
 490
 491         if (index >= nbd->num_connections) {
 492                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 493                                     "Attempted send on invalid socket\n");
 494                 goto error_out;
 495         }
 496
 497         if (test_bit(NBD_DISCONNECTED, &nbd->runtime_flags)) {
 498                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 499                                     "Attempted send on closed socket\n");
 500                 goto error_out;
 501         }
 502
 503         req->errors = 0;
 504
 505         nsock = nbd->socks[index];
 506         mutex_lock(&nsock->tx_lock);
 507         if (unlikely(!nsock->sock)) {
 508                 mutex_unlock(&nsock->tx_lock);
 509                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 510                                     "Attempted send on closed socket\n");
 511                 goto error_out;
 512         }
 513
 514         if (nbd_send_cmd(nbd, cmd, index) != 0) {
 515                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 516                                     "Request send failed\n");
 517                 req->errors++;
 518                 nbd_end_request(cmd);
 519         }
 520
 521         mutex_unlock(&nsock->tx_lock);
 522
 523         return;
 524
 525 error_out:
 526         req->errors++;
 527         nbd_end_request(cmd);
 528 }
 529
 530 static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 531                         const struct blk_mq_queue_data *bd)
 532 {
 533         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 534
 535         /*
 536          * Since we look at the bio's to send the request over the network we
 537          * need to make sure the completion work doesn't mark this request done
 538          * before we are done doing our send.  This keeps us from dereferencing
 539          * freed data if we have particularly fast completions (ie we get the
 540          * completion before we exit sock_xmit on the last bvec) or in the case
 541          * that the server is misbehaving (or there was an error) before we're
 542          * done sending everything over the wire.
 543          */
 544         init_completion(&cmd->send_complete);
 545         blk_mq_start_request(bd->rq);
 546         nbd_handle_cmd(cmd, hctx->queue_num);
 547         complete(&cmd->send_complete);
 548
 549         return BLK_MQ_RQ_QUEUE_OK;
 550 }
 551
 552 static int nbd_add_socket(struct nbd_device *nbd, struct block_device *bdev,
 553                           unsigned long arg)
 554 {
 555         struct socket *sock;
 556         struct nbd_sock **socks;
 557         struct nbd_sock *nsock;
 558         int err;
 559
 560         sock = sockfd_lookup(arg, &err);
 561         if (!sock)
 562                 return err;
 563
 564         if (!nbd->task_setup)
 565                 nbd->task_setup = current;
 566         if (nbd->task_setup != current) {
 567                 dev_err(disk_to_dev(nbd->disk),
 568                         "Device being setup by another task");
 569                 return -EINVAL;
 570         }
 571
 572         socks = krealloc(nbd->socks, (nbd->num_connections + 1) *
 573                          sizeof(struct nbd_sock *), GFP_KERNEL);
 574         if (!socks)
 575                 return -ENOMEM;
 576         nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
 577         if (!nsock)
 578                 return -ENOMEM;
 579
 580         nbd->socks = socks;
 581
 582         mutex_init(&nsock->tx_lock);
 583         nsock->sock = sock;
 584         socks[nbd->num_connections++] = nsock;
 585
 586         if (max_part)
 587                 bdev->bd_invalidated = 1;
 588         return 0;
 589 }
 590
 591 /* Reset all properties of an NBD device */
 592 static void nbd_reset(struct nbd_device *nbd)
 593 {
 594         nbd->runtime_flags = 0;
 595         nbd->blksize = 1024;
 596         nbd->bytesize = 0;
 597         set_capacity(nbd->disk, 0);
 598         nbd->flags = 0;
 599         nbd->tag_set.timeout = 0;
 600         queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 601 }
 602
 603 static void nbd_bdev_reset(struct block_device *bdev)
 604 {
 605         set_device_ro(bdev, false);
 606         bdev->bd_inode->i_size = 0;
 607         if (max_part > 0) {
 608                 blkdev_reread_part(bdev);
 609                 bdev->bd_invalidated = 1;
 610         }
 611 }
 612
 613 static void nbd_parse_flags(struct nbd_device *nbd, struct block_device *bdev)
 614 {
 615         if (nbd->flags & NBD_FLAG_READ_ONLY)
 616                 set_device_ro(bdev, true);
 617         if (nbd->flags & NBD_FLAG_SEND_TRIM)
 618                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 619         if (nbd->flags & NBD_FLAG_SEND_FLUSH)
 620                 blk_queue_write_cache(nbd->disk->queue, true, false);
 621         else
 622                 blk_queue_write_cache(nbd->disk->queue, false, false);
 623 }
 624
 625 static void send_disconnects(struct nbd_device *nbd)
 626 {
 627         struct nbd_request request = {
 628                 .magic = htonl(NBD_REQUEST_MAGIC),
 629                 .type = htonl(NBD_CMD_DISC),
 630         };
 631         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 632         struct iov_iter from;
 633         int i, ret;
 634
 635         for (i = 0; i < nbd->num_connections; i++) {
 636                 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
 637                 ret = sock_xmit(nbd, i, 1, &from, 0);
 638                 if (ret <= 0)
 639                         dev_err(disk_to_dev(nbd->disk),
 640                                 "Send disconnect failed %d\n", ret);
 641         }
 642 }
 643
 644 static int nbd_disconnect(struct nbd_device *nbd, struct block_device *bdev)
 645 {
 646         dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
 647         if (!nbd->socks)
 648                 return -EINVAL;
 649
 650         mutex_unlock(&nbd->config_lock);
 651         fsync_bdev(bdev);
 652         mutex_lock(&nbd->config_lock);
 653
 654         /* Check again after getting mutex back.  */
 655         if (!nbd->socks)
 656                 return -EINVAL;
 657
 658         if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED,
 659                               &nbd->runtime_flags))
 660                 send_disconnects(nbd);
 661         return 0;
 662 }
 663
 664 static int nbd_clear_sock(struct nbd_device *nbd, struct block_device *bdev)
 665 {
 666         sock_shutdown(nbd);
 667         nbd_clear_que(nbd);
 668         kill_bdev(bdev);
 669         nbd_bdev_reset(bdev);
 670         /*
 671          * We want to give the run thread a chance to wait for everybody
 672          * to clean up and then do it's own cleanup.
 673          */
 674         if (!test_bit(NBD_RUNNING, &nbd->runtime_flags) &&
 675             nbd->num_connections) {
 676                 int i;
 677
 678                 for (i = 0; i < nbd->num_connections; i++) {
 679                         sockfd_put(nbd->socks[i]->sock);
 680                         kfree(nbd->socks[i]);
 681                 }
 682                 kfree(nbd->socks);
 683                 nbd->socks = NULL;
 684                 nbd->num_connections = 0;
 685         }
 686         nbd->task_setup = NULL;
 687
 688         return 0;
 689 }
 690
 691 static int nbd_start_device(struct nbd_device *nbd, struct block_device *bdev)
 692 {
 693         struct recv_thread_args *args;
 694         int num_connections = nbd->num_connections;
 695         int error = 0, i;
 696
 697         if (nbd->task_recv)
 698                 return -EBUSY;
 699         if (!nbd->socks)
 700                 return -EINVAL;
 701         if (num_connections > 1 &&
 702             !(nbd->flags & NBD_FLAG_CAN_MULTI_CONN)) {
 703                 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
 704                 error = -EINVAL;
 705                 goto out_err;
 706         }
 707
 708         set_bit(NBD_RUNNING, &nbd->runtime_flags);
 709         blk_mq_update_nr_hw_queues(&nbd->tag_set, nbd->num_connections);
 710         args = kcalloc(num_connections, sizeof(*args), GFP_KERNEL);
 711         if (!args) {
 712                 error = -ENOMEM;
 713                 goto out_err;
 714         }
 715         nbd->task_recv = current;
 716         mutex_unlock(&nbd->config_lock);
 717
 718         nbd_parse_flags(nbd, bdev);
 719
 720         error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
 721         if (error) {
 722                 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
 723                 goto out_recv;
 724         }
 725
 726         nbd_size_update(nbd, bdev);
 727
 728         nbd_dev_dbg_init(nbd);
 729         for (i = 0; i < num_connections; i++) {
 730                 sk_set_memalloc(nbd->socks[i]->sock->sk);
 731                 atomic_inc(&nbd->recv_threads);
 732                 INIT_WORK(&args[i].work, recv_work);
 733                 args[i].nbd = nbd;
 734                 args[i].index = i;
 735                 queue_work(recv_workqueue, &args[i].work);
 736         }
 737         wait_event_interruptible(nbd->recv_wq,
 738                                  atomic_read(&nbd->recv_threads) == 0);
 739         for (i = 0; i < num_connections; i++)
 740                 flush_work(&args[i].work);
 741         nbd_dev_dbg_close(nbd);
 742         nbd_size_clear(nbd, bdev);
 743         device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
 744 out_recv:
 745         mutex_lock(&nbd->config_lock);
 746         nbd->task_recv = NULL;
 747 out_err:
 748         clear_bit(NBD_RUNNING, &nbd->runtime_flags);
 749         nbd_clear_sock(nbd, bdev);
 750
 751         /* user requested, ignore socket errors */
 752         if (test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
 753                 error = 0;
 754         if (test_bit(NBD_TIMEDOUT, &nbd->runtime_flags))
 755                 error = -ETIMEDOUT;
 756
 757         nbd_reset(nbd);
 758         return error;
 759 }
 760
 761 /* Must be called with config_lock held */
 762 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
 763                        unsigned int cmd, unsigned long arg)
 764 {
 765         switch (cmd) {
 766         case NBD_DISCONNECT:
 767                 return nbd_disconnect(nbd, bdev);
 768         case NBD_CLEAR_SOCK:
 769                 return nbd_clear_sock(nbd, bdev);
 770         case NBD_SET_SOCK:
 771                 return nbd_add_socket(nbd, bdev, arg);
 772         case NBD_SET_BLKSIZE:
 773                 nbd_size_set(nbd, bdev, arg,
 774                              div_s64(nbd->bytesize, arg));
 775                 return 0;
 776         case NBD_SET_SIZE:
 777                 nbd_size_set(nbd, bdev, nbd->blksize,
 778                              div_s64(arg, nbd->blksize));
 779                 return 0;
 780         case NBD_SET_SIZE_BLOCKS:
 781                 nbd_size_set(nbd, bdev, nbd->blksize, arg);
 782                 return 0;
 783         case NBD_SET_TIMEOUT:
 784                 nbd->tag_set.timeout = arg * HZ;
 785                 return 0;
 786
 787         case NBD_SET_FLAGS:
 788                 nbd->flags = arg;
 789                 return 0;
 790         case NBD_DO_IT:
 791                 return nbd_start_device(nbd, bdev);
 792         case NBD_CLEAR_QUE:
 793                 /*
 794                  * This is for compatibility only.  The queue is always cleared
 795                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
 796                  */
 797                 return 0;
 798         case NBD_PRINT_DEBUG:
 799                 /*
 800                  * For compatibility only, we no longer keep a list of
 801                  * outstanding requests.
 802                  */
 803                 return 0;
 804         }
 805         return -ENOTTY;
 806 }
 807
 808 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
 809                      unsigned int cmd, unsigned long arg)
 810 {
 811         struct nbd_device *nbd = bdev->bd_disk->private_data;
 812         int error;
 813
 814         if (!capable(CAP_SYS_ADMIN))
 815                 return -EPERM;
 816
 817         BUG_ON(nbd->magic != NBD_MAGIC);
 818
 819         mutex_lock(&nbd->config_lock);
 820         error = __nbd_ioctl(bdev, nbd, cmd, arg);
 821         mutex_unlock(&nbd->config_lock);
 822
 823         return error;
 824 }
 825
 826 static const struct block_device_operations nbd_fops =
 827 {
 828         .owner =        THIS_MODULE,
 829         .ioctl =        nbd_ioctl,
 830         .compat_ioctl = nbd_ioctl,
 831 };
 832
 833 #if IS_ENABLED(CONFIG_DEBUG_FS)
 834
 835 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
 836 {
 837         struct nbd_device *nbd = s->private;
 838
 839         if (nbd->task_recv)
 840                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
 841
 842         return 0;
 843 }
 844
 845 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
 846 {
 847         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
 848 }
 849
 850 static const struct file_operations nbd_dbg_tasks_ops = {
 851         .open = nbd_dbg_tasks_open,
 852         .read = seq_read,
 853         .llseek = seq_lseek,
 854         .release = single_release,
 855 };
 856
 857 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
 858 {
 859         struct nbd_device *nbd = s->private;
 860         u32 flags = nbd->flags;
 861
 862         seq_printf(s, "Hex: 0x%08x\n\n", flags);
 863
 864         seq_puts(s, "Known flags:\n");
 865
 866         if (flags & NBD_FLAG_HAS_FLAGS)
 867                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
 868         if (flags & NBD_FLAG_READ_ONLY)
 869                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
 870         if (flags & NBD_FLAG_SEND_FLUSH)
 871                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
 872         if (flags & NBD_FLAG_SEND_TRIM)
 873                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
 874
 875         return 0;
 876 }
 877
 878 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
 879 {
 880         return single_open(file, nbd_dbg_flags_show, inode->i_private);
 881 }
 882
 883 static const struct file_operations nbd_dbg_flags_ops = {
 884         .open = nbd_dbg_flags_open,
 885         .read = seq_read,
 886         .llseek = seq_lseek,
 887         .release = single_release,
 888 };
 889
 890 static int nbd_dev_dbg_init(struct nbd_device *nbd)
 891 {
 892         struct dentry *dir;
 893
 894         if (!nbd_dbg_dir)
 895                 return -EIO;
 896
 897         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
 898         if (!dir) {
 899                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
 900                         nbd_name(nbd));
 901                 return -EIO;
 902         }
 903         nbd->dbg_dir = dir;
 904
 905         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
 906         debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize);
 907         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
 908         debugfs_create_u64("blocksize", 0444, dir, &nbd->blksize);
 909         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
 910
 911         return 0;
 912 }
 913
 914 static void nbd_dev_dbg_close(struct nbd_device *nbd)
 915 {
 916         debugfs_remove_recursive(nbd->dbg_dir);
 917 }
 918
 919 static int nbd_dbg_init(void)
 920 {
 921         struct dentry *dbg_dir;
 922
 923         dbg_dir = debugfs_create_dir("nbd", NULL);
 924         if (!dbg_dir)
 925                 return -EIO;
 926
 927         nbd_dbg_dir = dbg_dir;
 928
 929         return 0;
 930 }
 931
 932 static void nbd_dbg_close(void)
 933 {
 934         debugfs_remove_recursive(nbd_dbg_dir);
 935 }
 936
 937 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
 938
 939 static int nbd_dev_dbg_init(struct nbd_device *nbd)
 940 {
 941         return 0;
 942 }
 943
 944 static void nbd_dev_dbg_close(struct nbd_device *nbd)
 945 {
 946 }
 947
 948 static int nbd_dbg_init(void)
 949 {
 950         return 0;
 951 }
 952
 953 static void nbd_dbg_close(void)
 954 {
 955 }
 956
 957 #endif
 958
 959 static int nbd_init_request(void *data, struct request *rq,
 960                             unsigned int hctx_idx, unsigned int request_idx,
 961                             unsigned int numa_node)
 962 {
 963         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
 964         cmd->nbd = data;
 965         return 0;
 966 }
 967
 968 static struct blk_mq_ops nbd_mq_ops = {
 969         .queue_rq       = nbd_queue_rq,
 970         .init_request   = nbd_init_request,
 971         .timeout        = nbd_xmit_timeout,
 972 };
 973
 974 static void nbd_dev_remove(struct nbd_device *nbd)
 975 {
 976         struct gendisk *disk = nbd->disk;
 977         nbd->magic = 0;
 978         if (disk) {
 979                 del_gendisk(disk);
 980                 blk_cleanup_queue(disk->queue);
 981                 blk_mq_free_tag_set(&nbd->tag_set);
 982                 put_disk(disk);
 983         }
 984         kfree(nbd);
 985 }
 986
 987 static int nbd_dev_add(int index)
 988 {
 989         struct nbd_device *nbd;
 990         struct gendisk *disk;
 991         struct request_queue *q;
 992         int err = -ENOMEM;
 993
 994         nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
 995         if (!nbd)
 996                 goto out;
 997
 998         disk = alloc_disk(1 << part_shift);
 999         if (!disk)
1000                 goto out_free_nbd;
1001
1002         if (index >= 0) {
1003                 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1004                                 GFP_KERNEL);
1005                 if (err == -ENOSPC)
1006                         err = -EEXIST;
1007         } else {
1008                 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1009                 if (err >= 0)
1010                         index = err;
1011         }
1012         if (err < 0)
1013                 goto out_free_disk;
1014
1015         nbd->disk = disk;
1016         nbd->tag_set.ops = &nbd_mq_ops;
1017         nbd->tag_set.nr_hw_queues = 1;
1018         nbd->tag_set.queue_depth = 128;
1019         nbd->tag_set.numa_node = NUMA_NO_NODE;
1020         nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1021         nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1022                 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1023         nbd->tag_set.driver_data = nbd;
1024
1025         err = blk_mq_alloc_tag_set(&nbd->tag_set);
1026         if (err)
1027                 goto out_free_idr;
1028
1029         q = blk_mq_init_queue(&nbd->tag_set);
1030         if (IS_ERR(q)) {
1031                 err = PTR_ERR(q);
1032                 goto out_free_tags;
1033         }
1034         disk->queue = q;
1035
1036         /*
1037          * Tell the block layer that we are not a rotational device
1038          */
1039         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1040         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1041         disk->queue->limits.discard_granularity = 512;
1042         blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1043         disk->queue->limits.discard_zeroes_data = 0;
1044         blk_queue_max_hw_sectors(disk->queue, 65536);
1045         disk->queue->limits.max_sectors = 256;
1046
1047         nbd->magic = NBD_MAGIC;
1048         mutex_init(&nbd->config_lock);
1049         disk->major = NBD_MAJOR;
1050         disk->first_minor = index << part_shift;
1051         disk->fops = &nbd_fops;
1052         disk->private_data = nbd;
1053         sprintf(disk->disk_name, "nbd%d", index);
1054         init_waitqueue_head(&nbd->recv_wq);
1055         nbd_reset(nbd);
1056         add_disk(disk);
1057         return index;
1058
1059 out_free_tags:
1060         blk_mq_free_tag_set(&nbd->tag_set);
1061 out_free_idr:
1062         idr_remove(&nbd_index_idr, index);
1063 out_free_disk:
1064         put_disk(disk);
1065 out_free_nbd:
1066         kfree(nbd);
1067 out:
1068         return err;
1069 }
1070
1071 /*
1072  * And here should be modules and kernel interface
1073  *  (Just smiley confuses emacs :-)
1074  */
1075
1076 static int __init nbd_init(void)
1077 {
1078         int i;
1079
1080         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
1081
1082         if (max_part < 0) {
1083                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
1084                 return -EINVAL;
1085         }
1086
1087         part_shift = 0;
1088         if (max_part > 0) {
1089                 part_shift = fls(max_part);
1090
1091                 /*
1092                  * Adjust max_part according to part_shift as it is exported
1093                  * to user space so that user can know the max number of
1094                  * partition kernel should be able to manage.
1095                  *
1096                  * Note that -1 is required because partition 0 is reserved
1097                  * for the whole disk.
1098                  */
1099                 max_part = (1UL << part_shift) - 1;
1100         }
1101
1102         if ((1UL << part_shift) > DISK_MAX_PARTS)
1103                 return -EINVAL;
1104
1105         if (nbds_max > 1UL << (MINORBITS - part_shift))
1106                 return -EINVAL;
1107         recv_workqueue = alloc_workqueue("knbd-recv",
1108                                          WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1109         if (!recv_workqueue)
1110                 return -ENOMEM;
1111
1112         if (register_blkdev(NBD_MAJOR, "nbd")) {
1113                 destroy_workqueue(recv_workqueue);
1114                 return -EIO;
1115         }
1116
1117         nbd_dbg_init();
1118
1119         mutex_lock(&nbd_index_mutex);
1120         for (i = 0; i < nbds_max; i++)
1121                 nbd_dev_add(i);
1122         mutex_unlock(&nbd_index_mutex);
1123         return 0;
1124 }
1125
1126 static int nbd_exit_cb(int id, void *ptr, void *data)
1127 {
1128         struct nbd_device *nbd = ptr;
1129         nbd_dev_remove(nbd);
1130         return 0;
1131 }
1132
1133 static void __exit nbd_cleanup(void)
1134 {
1135         nbd_dbg_close();
1136
1137         idr_for_each(&nbd_index_idr, &nbd_exit_cb, NULL);
1138         idr_destroy(&nbd_index_idr);
1139         destroy_workqueue(recv_workqueue);
1140         unregister_blkdev(NBD_MAJOR, "nbd");
1141 }
1142
1143 module_init(nbd_init);
1144 module_exit(nbd_cleanup);
1145
1146 MODULE_DESCRIPTION("Network Block Device");
1147 MODULE_LICENSE("GPL");
1148
1149 module_param(nbds_max, int, 0444);
1150 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
1151 module_param(max_part, int, 0444);
1152 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");