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 inline struct device *nbd_to_dev(struct nbd_device *nbd)
 100 {
 101         return disk_to_dev(nbd->disk);
 102 }
 103
 104 static bool nbd_is_connected(struct nbd_device *nbd)
 105 {
 106         return !!nbd->task_recv;
 107 }
 108
 109 static const char *nbdcmd_to_ascii(int cmd)
 110 {
 111         switch (cmd) {
 112         case  NBD_CMD_READ: return "read";
 113         case NBD_CMD_WRITE: return "write";
 114         case  NBD_CMD_DISC: return "disconnect";
 115         case NBD_CMD_FLUSH: return "flush";
 116         case  NBD_CMD_TRIM: return "trim/discard";
 117         }
 118         return "invalid";
 119 }
 120
 121 static int nbd_size_clear(struct nbd_device *nbd, struct block_device *bdev)
 122 {
 123         bdev->bd_inode->i_size = 0;
 124         set_capacity(nbd->disk, 0);
 125         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 126
 127         return 0;
 128 }
 129
 130 static void nbd_size_update(struct nbd_device *nbd, struct block_device *bdev)
 131 {
 132         if (!nbd_is_connected(nbd))
 133                 return;
 134
 135         bdev->bd_inode->i_size = nbd->bytesize;
 136         set_capacity(nbd->disk, nbd->bytesize >> 9);
 137         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 138 }
 139
 140 static int nbd_size_set(struct nbd_device *nbd, struct block_device *bdev,
 141                         loff_t blocksize, loff_t nr_blocks)
 142 {
 143         int ret;
 144
 145         ret = set_blocksize(bdev, blocksize);
 146         if (ret)
 147                 return ret;
 148
 149         nbd->blksize = blocksize;
 150         nbd->bytesize = blocksize * nr_blocks;
 151
 152         nbd_size_update(nbd, bdev);
 153
 154         return 0;
 155 }
 156
 157 static void nbd_end_request(struct nbd_cmd *cmd)
 158 {
 159         struct nbd_device *nbd = cmd->nbd;
 160         struct request *req = blk_mq_rq_from_pdu(cmd);
 161         int error = req->errors ? -EIO : 0;
 162
 163         dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", cmd,
 164                 error ? "failed" : "done");
 165
 166         blk_mq_complete_request(req, error);
 167 }
 168
 169 /*
 170  * Forcibly shutdown the socket causing all listeners to error
 171  */
 172 static void sock_shutdown(struct nbd_device *nbd)
 173 {
 174         int i;
 175
 176         if (nbd->num_connections == 0)
 177                 return;
 178         if (test_and_set_bit(NBD_DISCONNECTED, &nbd->runtime_flags))
 179                 return;
 180
 181         for (i = 0; i < nbd->num_connections; i++) {
 182                 struct nbd_sock *nsock = nbd->socks[i];
 183                 mutex_lock(&nsock->tx_lock);
 184                 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
 185                 mutex_unlock(&nsock->tx_lock);
 186         }
 187         dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
 188 }
 189
 190 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 191                                                  bool reserved)
 192 {
 193         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 194         struct nbd_device *nbd = cmd->nbd;
 195
 196         dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n");
 197         set_bit(NBD_TIMEDOUT, &nbd->runtime_flags);
 198         req->errors++;
 199
 200         mutex_lock(&nbd->config_lock);
 201         sock_shutdown(nbd);
 202         mutex_unlock(&nbd->config_lock);
 203         return BLK_EH_HANDLED;
 204 }
 205
 206 /*
 207  *  Send or receive packet.
 208  */
 209 static int sock_xmit(struct nbd_device *nbd, int index, int send, void *buf,
 210                      int size, int msg_flags)
 211 {
 212         struct socket *sock = nbd->socks[index]->sock;
 213         int result;
 214         struct msghdr msg;
 215         struct kvec iov;
 216         unsigned long pflags = current->flags;
 217
 218         if (unlikely(!sock)) {
 219                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 220                         "Attempted %s on closed socket in sock_xmit\n",
 221                         (send ? "send" : "recv"));
 222                 return -EINVAL;
 223         }
 224
 225         current->flags |= PF_MEMALLOC;
 226         do {
 227                 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
 228                 iov.iov_base = buf;
 229                 iov.iov_len = size;
 230                 msg.msg_name = NULL;
 231                 msg.msg_namelen = 0;
 232                 msg.msg_control = NULL;
 233                 msg.msg_controllen = 0;
 234                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 235
 236                 if (send)
 237                         result = kernel_sendmsg(sock, &msg, &iov, 1, size);
 238                 else
 239                         result = kernel_recvmsg(sock, &msg, &iov, 1, size,
 240                                                 msg.msg_flags);
 241
 242                 if (result <= 0) {
 243                         if (result == 0)
 244                                 result = -EPIPE; /* short read */
 245                         break;
 246                 }
 247                 size -= result;
 248                 buf += result;
 249         } while (size > 0);
 250
 251         tsk_restore_flags(current, pflags, PF_MEMALLOC);
 252
 253         return result;
 254 }
 255
 256 static inline int sock_send_bvec(struct nbd_device *nbd, int index,
 257                                  struct bio_vec *bvec, int flags)
 258 {
 259         int result;
 260         void *kaddr = kmap(bvec->bv_page);
 261         result = sock_xmit(nbd, index, 1, kaddr + bvec->bv_offset,
 262                            bvec->bv_len, flags);
 263         kunmap(bvec->bv_page);
 264         return result;
 265 }
 266
 267 /* always call with the tx_lock held */
 268 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 269 {
 270         struct request *req = blk_mq_rq_from_pdu(cmd);
 271         int result;
 272         struct nbd_request request;
 273         unsigned long size = blk_rq_bytes(req);
 274         struct bio *bio;
 275         u32 type;
 276         u32 tag = blk_mq_unique_tag(req);
 277
 278         switch (req_op(req)) {
 279         case REQ_OP_DISCARD:
 280                 type = NBD_CMD_TRIM;
 281                 break;
 282         case REQ_OP_FLUSH:
 283                 type = NBD_CMD_FLUSH;
 284                 break;
 285         case REQ_OP_WRITE:
 286                 type = NBD_CMD_WRITE;
 287                 break;
 288         case REQ_OP_READ:
 289                 type = NBD_CMD_READ;
 290                 break;
 291         default:
 292                 return -EIO;
 293         }
 294
 295         if (rq_data_dir(req) == WRITE &&
 296             (nbd->flags & NBD_FLAG_READ_ONLY)) {
 297                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 298                                     "Write on read-only\n");
 299                 return -EIO;
 300         }
 301
 302         memset(&request, 0, sizeof(request));
 303         request.magic = htonl(NBD_REQUEST_MAGIC);
 304         request.type = htonl(type);
 305         if (type != NBD_CMD_FLUSH) {
 306                 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
 307                 request.len = htonl(size);
 308         }
 309         memcpy(request.handle, &tag, sizeof(tag));
 310
 311         dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
 312                 cmd, nbdcmd_to_ascii(type),
 313                 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
 314         result = sock_xmit(nbd, index, 1, &request, sizeof(request),
 315                         (type == NBD_CMD_WRITE) ? MSG_MORE : 0);
 316         if (result <= 0) {
 317                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 318                         "Send control failed (result %d)\n", result);
 319                 return -EIO;
 320         }
 321
 322         if (type != NBD_CMD_WRITE)
 323                 return 0;
 324
 325         bio = req->bio;
 326         while (bio) {
 327                 struct bio *next = bio->bi_next;
 328                 struct bvec_iter iter;
 329                 struct bio_vec bvec;
 330
 331                 bio_for_each_segment(bvec, bio, iter) {
 332                         bool is_last = !next && bio_iter_last(bvec, iter);
 333                         int flags = is_last ? 0 : MSG_MORE;
 334
 335                         dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
 336                                 cmd, bvec.bv_len);
 337                         result = sock_send_bvec(nbd, index, &bvec, flags);
 338                         if (result <= 0) {
 339                                 dev_err(disk_to_dev(nbd->disk),
 340                                         "Send data failed (result %d)\n",
 341                                         result);
 342                                 return -EIO;
 343                         }
 344                         /*
 345                          * The completion might already have come in,
 346                          * so break for the last one instead of letting
 347                          * the iterator do it. This prevents use-after-free
 348                          * of the bio.
 349                          */
 350                         if (is_last)
 351                                 break;
 352                 }
 353                 bio = next;
 354         }
 355         return 0;
 356 }
 357
 358 static inline int sock_recv_bvec(struct nbd_device *nbd, int index,
 359                                  struct bio_vec *bvec)
 360 {
 361         int result;
 362         void *kaddr = kmap(bvec->bv_page);
 363         result = sock_xmit(nbd, index, 0, kaddr + bvec->bv_offset,
 364                            bvec->bv_len, MSG_WAITALL);
 365         kunmap(bvec->bv_page);
 366         return result;
 367 }
 368
 369 /* NULL returned = something went wrong, inform userspace */
 370 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 371 {
 372         int result;
 373         struct nbd_reply reply;
 374         struct nbd_cmd *cmd;
 375         struct request *req = NULL;
 376         u16 hwq;
 377         u32 tag;
 378
 379         reply.magic = 0;
 380         result = sock_xmit(nbd, index, 0, &reply, sizeof(reply), MSG_WAITALL);
 381         if (result <= 0) {
 382                 if (!test_bit(NBD_DISCONNECTED, &nbd->runtime_flags) &&
 383                     !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
 384                         dev_err(disk_to_dev(nbd->disk),
 385                                 "Receive control failed (result %d)\n", result);
 386                 return ERR_PTR(result);
 387         }
 388
 389         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 390                 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
 391                                 (unsigned long)ntohl(reply.magic));
 392                 return ERR_PTR(-EPROTO);
 393         }
 394
 395         memcpy(&tag, reply.handle, sizeof(u32));
 396
 397         hwq = blk_mq_unique_tag_to_hwq(tag);
 398         if (hwq < nbd->tag_set.nr_hw_queues)
 399                 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
 400                                        blk_mq_unique_tag_to_tag(tag));
 401         if (!req || !blk_mq_request_started(req)) {
 402                 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
 403                         tag, req);
 404                 return ERR_PTR(-ENOENT);
 405         }
 406         cmd = blk_mq_rq_to_pdu(req);
 407         if (ntohl(reply.error)) {
 408                 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
 409                         ntohl(reply.error));
 410                 req->errors++;
 411                 return cmd;
 412         }
 413
 414         dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
 415         if (rq_data_dir(req) != WRITE) {
 416                 struct req_iterator iter;
 417                 struct bio_vec bvec;
 418
 419                 rq_for_each_segment(bvec, req, iter) {
 420                         result = sock_recv_bvec(nbd, index, &bvec);
 421                         if (result <= 0) {
 422                                 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 423                                         result);
 424                                 req->errors++;
 425                                 return cmd;
 426                         }
 427                         dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 428                                 cmd, bvec.bv_len);
 429                 }
 430         } else {
 431                 /* See the comment in nbd_queue_rq. */
 432                 wait_for_completion(&cmd->send_complete);
 433         }
 434         return cmd;
 435 }
 436
 437 static ssize_t pid_show(struct device *dev,
 438                         struct device_attribute *attr, char *buf)
 439 {
 440         struct gendisk *disk = dev_to_disk(dev);
 441         struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
 442
 443         return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
 444 }
 445
 446 static struct device_attribute pid_attr = {
 447         .attr = { .name = "pid", .mode = S_IRUGO},
 448         .show = pid_show,
 449 };
 450
 451 struct recv_thread_args {
 452         struct work_struct work;
 453         struct nbd_device *nbd;
 454         int index;
 455 };
 456
 457 static void recv_work(struct work_struct *work)
 458 {
 459         struct recv_thread_args *args = container_of(work,
 460                                                      struct recv_thread_args,
 461                                                      work);
 462         struct nbd_device *nbd = args->nbd;
 463         struct nbd_cmd *cmd;
 464         int ret = 0;
 465
 466         BUG_ON(nbd->magic != NBD_MAGIC);
 467         while (1) {
 468                 cmd = nbd_read_stat(nbd, args->index);
 469                 if (IS_ERR(cmd)) {
 470                         ret = PTR_ERR(cmd);
 471                         break;
 472                 }
 473
 474                 nbd_end_request(cmd);
 475         }
 476
 477         /*
 478          * We got an error, shut everybody down if this wasn't the result of a
 479          * disconnect request.
 480          */
 481         if (ret && !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
 482                 sock_shutdown(nbd);
 483         atomic_dec(&nbd->recv_threads);
 484         wake_up(&nbd->recv_wq);
 485 }
 486
 487 static void nbd_clear_req(struct request *req, void *data, bool reserved)
 488 {
 489         struct nbd_cmd *cmd;
 490
 491         if (!blk_mq_request_started(req))
 492                 return;
 493         cmd = blk_mq_rq_to_pdu(req);
 494         req->errors++;
 495         nbd_end_request(cmd);
 496 }
 497
 498 static void nbd_clear_que(struct nbd_device *nbd)
 499 {
 500         BUG_ON(nbd->magic != NBD_MAGIC);
 501
 502         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 503         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 504 }
 505
 506
 507 static void nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 508 {
 509         struct request *req = blk_mq_rq_from_pdu(cmd);
 510         struct nbd_device *nbd = cmd->nbd;
 511         struct nbd_sock *nsock;
 512
 513         if (index >= nbd->num_connections) {
 514                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 515                                     "Attempted send on invalid socket\n");
 516                 goto error_out;
 517         }
 518
 519         if (test_bit(NBD_DISCONNECTED, &nbd->runtime_flags)) {
 520                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 521                                     "Attempted send on closed socket\n");
 522                 goto error_out;
 523         }
 524
 525         req->errors = 0;
 526
 527         nsock = nbd->socks[index];
 528         mutex_lock(&nsock->tx_lock);
 529         if (unlikely(!nsock->sock)) {
 530                 mutex_unlock(&nsock->tx_lock);
 531                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 532                                     "Attempted send on closed socket\n");
 533                 goto error_out;
 534         }
 535
 536         if (nbd_send_cmd(nbd, cmd, index) != 0) {
 537                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 538                                     "Request send failed\n");
 539                 req->errors++;
 540                 nbd_end_request(cmd);
 541         }
 542
 543         mutex_unlock(&nsock->tx_lock);
 544
 545         return;
 546
 547 error_out:
 548         req->errors++;
 549         nbd_end_request(cmd);
 550 }
 551
 552 static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 553                         const struct blk_mq_queue_data *bd)
 554 {
 555         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 556
 557         /*
 558          * Since we look at the bio's to send the request over the network we
 559          * need to make sure the completion work doesn't mark this request done
 560          * before we are done doing our send.  This keeps us from dereferencing
 561          * freed data if we have particularly fast completions (ie we get the
 562          * completion before we exit sock_xmit on the last bvec) or in the case
 563          * that the server is misbehaving (or there was an error) before we're
 564          * done sending everything over the wire.
 565          */
 566         init_completion(&cmd->send_complete);
 567         blk_mq_start_request(bd->rq);
 568         nbd_handle_cmd(cmd, hctx->queue_num);
 569         complete(&cmd->send_complete);
 570
 571         return BLK_MQ_RQ_QUEUE_OK;
 572 }
 573
 574 static int nbd_add_socket(struct nbd_device *nbd, struct socket *sock)
 575 {
 576         struct nbd_sock **socks;
 577         struct nbd_sock *nsock;
 578
 579         if (!nbd->task_setup)
 580                 nbd->task_setup = current;
 581         if (nbd->task_setup != current) {
 582                 dev_err(disk_to_dev(nbd->disk),
 583                         "Device being setup by another task");
 584                 return -EINVAL;
 585         }
 586
 587         socks = krealloc(nbd->socks, (nbd->num_connections + 1) *
 588                          sizeof(struct nbd_sock *), GFP_KERNEL);
 589         if (!socks)
 590                 return -ENOMEM;
 591         nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
 592         if (!nsock)
 593                 return -ENOMEM;
 594
 595         nbd->socks = socks;
 596
 597         mutex_init(&nsock->tx_lock);
 598         nsock->sock = sock;
 599         socks[nbd->num_connections++] = nsock;
 600
 601         return 0;
 602 }
 603
 604 /* Reset all properties of an NBD device */
 605 static void nbd_reset(struct nbd_device *nbd)
 606 {
 607         int i;
 608
 609         for (i = 0; i < nbd->num_connections; i++)
 610                 kfree(nbd->socks[i]);
 611         kfree(nbd->socks);
 612         nbd->socks = NULL;
 613         nbd->runtime_flags = 0;
 614         nbd->blksize = 1024;
 615         nbd->bytesize = 0;
 616         set_capacity(nbd->disk, 0);
 617         nbd->flags = 0;
 618         nbd->tag_set.timeout = 0;
 619         nbd->num_connections = 0;
 620         nbd->task_setup = NULL;
 621         queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 622 }
 623
 624 static void nbd_bdev_reset(struct block_device *bdev)
 625 {
 626         set_device_ro(bdev, false);
 627         bdev->bd_inode->i_size = 0;
 628         if (max_part > 0) {
 629                 blkdev_reread_part(bdev);
 630                 bdev->bd_invalidated = 1;
 631         }
 632 }
 633
 634 static void nbd_parse_flags(struct nbd_device *nbd, struct block_device *bdev)
 635 {
 636         if (nbd->flags & NBD_FLAG_READ_ONLY)
 637                 set_device_ro(bdev, true);
 638         if (nbd->flags & NBD_FLAG_SEND_TRIM)
 639                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 640         if (nbd->flags & NBD_FLAG_SEND_FLUSH)
 641                 blk_queue_write_cache(nbd->disk->queue, true, false);
 642         else
 643                 blk_queue_write_cache(nbd->disk->queue, false, false);
 644 }
 645
 646 static void send_disconnects(struct nbd_device *nbd)
 647 {
 648         struct nbd_request request = {};
 649         int i, ret;
 650
 651         request.magic = htonl(NBD_REQUEST_MAGIC);
 652         request.type = htonl(NBD_CMD_DISC);
 653
 654         for (i = 0; i < nbd->num_connections; i++) {
 655                 ret = sock_xmit(nbd, i, 1, &request, sizeof(request), 0);
 656                 if (ret <= 0)
 657                         dev_err(disk_to_dev(nbd->disk),
 658                                 "Send disconnect failed %d\n", ret);
 659         }
 660 }
 661
 662 static int nbd_dev_dbg_init(struct nbd_device *nbd);
 663 static void nbd_dev_dbg_close(struct nbd_device *nbd);
 664
 665 /* Must be called with config_lock held */
 666 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
 667                        unsigned int cmd, unsigned long arg)
 668 {
 669         switch (cmd) {
 670         case NBD_DISCONNECT: {
 671                 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
 672                 if (!nbd->socks)
 673                         return -EINVAL;
 674
 675                 mutex_unlock(&nbd->config_lock);
 676                 fsync_bdev(bdev);
 677                 mutex_lock(&nbd->config_lock);
 678
 679                 /* Check again after getting mutex back.  */
 680                 if (!nbd->socks)
 681                         return -EINVAL;
 682
 683                 if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED,
 684                                       &nbd->runtime_flags))
 685                         send_disconnects(nbd);
 686                 return 0;
 687         }
 688
 689         case NBD_CLEAR_SOCK:
 690                 sock_shutdown(nbd);
 691                 nbd_clear_que(nbd);
 692                 kill_bdev(bdev);
 693                 nbd_bdev_reset(bdev);
 694                 /*
 695                  * We want to give the run thread a chance to wait for everybody
 696                  * to clean up and then do it's own cleanup.
 697                  */
 698                 if (!test_bit(NBD_RUNNING, &nbd->runtime_flags)) {
 699                         int i;
 700
 701                         for (i = 0; i < nbd->num_connections; i++)
 702                                 kfree(nbd->socks[i]);
 703                         kfree(nbd->socks);
 704                         nbd->socks = NULL;
 705                         nbd->num_connections = 0;
 706                         nbd->task_setup = NULL;
 707                 }
 708                 return 0;
 709
 710         case NBD_SET_SOCK: {
 711                 int err;
 712                 struct socket *sock = sockfd_lookup(arg, &err);
 713
 714                 if (!sock)
 715                         return err;
 716
 717                 err = nbd_add_socket(nbd, sock);
 718                 if (!err && max_part)
 719                         bdev->bd_invalidated = 1;
 720
 721                 return err;
 722         }
 723
 724         case NBD_SET_BLKSIZE: {
 725                 loff_t bsize = div_s64(nbd->bytesize, arg);
 726
 727                 return nbd_size_set(nbd, bdev, arg, bsize);
 728         }
 729
 730         case NBD_SET_SIZE:
 731                 return nbd_size_set(nbd, bdev, nbd->blksize,
 732                                         div_s64(arg, nbd->blksize));
 733
 734         case NBD_SET_SIZE_BLOCKS:
 735                 return nbd_size_set(nbd, bdev, nbd->blksize, arg);
 736
 737         case NBD_SET_TIMEOUT:
 738                 nbd->tag_set.timeout = arg * HZ;
 739                 return 0;
 740
 741         case NBD_SET_FLAGS:
 742                 nbd->flags = arg;
 743                 return 0;
 744
 745         case NBD_DO_IT: {
 746                 struct recv_thread_args *args;
 747                 int num_connections = nbd->num_connections;
 748                 int error = 0, i;
 749
 750                 if (nbd->task_recv)
 751                         return -EBUSY;
 752                 if (!nbd->socks)
 753                         return -EINVAL;
 754                 if (num_connections > 1 &&
 755                     !(nbd->flags & NBD_FLAG_CAN_MULTI_CONN)) {
 756                         dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
 757                         error = -EINVAL;
 758                         goto out_err;
 759                 }
 760
 761                 set_bit(NBD_RUNNING, &nbd->runtime_flags);
 762                 blk_mq_update_nr_hw_queues(&nbd->tag_set, nbd->num_connections);
 763                 args = kcalloc(num_connections, sizeof(*args), GFP_KERNEL);
 764                 if (!args) {
 765                         error = -ENOMEM;
 766                         goto out_err;
 767                 }
 768                 nbd->task_recv = current;
 769                 mutex_unlock(&nbd->config_lock);
 770
 771                 nbd_parse_flags(nbd, bdev);
 772
 773                 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
 774                 if (error) {
 775                         dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
 776                         goto out_recv;
 777                 }
 778
 779                 nbd_size_update(nbd, bdev);
 780
 781                 nbd_dev_dbg_init(nbd);
 782                 for (i = 0; i < num_connections; i++) {
 783                         sk_set_memalloc(nbd->socks[i]->sock->sk);
 784                         atomic_inc(&nbd->recv_threads);
 785                         INIT_WORK(&args[i].work, recv_work);
 786                         args[i].nbd = nbd;
 787                         args[i].index = i;
 788                         queue_work(recv_workqueue, &args[i].work);
 789                 }
 790                 wait_event_interruptible(nbd->recv_wq,
 791                                          atomic_read(&nbd->recv_threads) == 0);
 792                 for (i = 0; i < num_connections; i++)
 793                         flush_work(&args[i].work);
 794                 nbd_dev_dbg_close(nbd);
 795                 nbd_size_clear(nbd, bdev);
 796                 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
 797 out_recv:
 798                 mutex_lock(&nbd->config_lock);
 799                 nbd->task_recv = NULL;
 800 out_err:
 801                 sock_shutdown(nbd);
 802                 nbd_clear_que(nbd);
 803                 kill_bdev(bdev);
 804                 nbd_bdev_reset(bdev);
 805
 806                 /* user requested, ignore socket errors */
 807                 if (test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
 808                         error = 0;
 809                 if (test_bit(NBD_TIMEDOUT, &nbd->runtime_flags))
 810                         error = -ETIMEDOUT;
 811
 812                 nbd_reset(nbd);
 813                 return error;
 814         }
 815
 816         case NBD_CLEAR_QUE:
 817                 /*
 818                  * This is for compatibility only.  The queue is always cleared
 819                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
 820                  */
 821                 return 0;
 822
 823         case NBD_PRINT_DEBUG:
 824                 /*
 825                  * For compatibility only, we no longer keep a list of
 826                  * outstanding requests.
 827                  */
 828                 return 0;
 829         }
 830         return -ENOTTY;
 831 }
 832
 833 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
 834                      unsigned int cmd, unsigned long arg)
 835 {
 836         struct nbd_device *nbd = bdev->bd_disk->private_data;
 837         int error;
 838
 839         if (!capable(CAP_SYS_ADMIN))
 840                 return -EPERM;
 841
 842         BUG_ON(nbd->magic != NBD_MAGIC);
 843
 844         mutex_lock(&nbd->config_lock);
 845         error = __nbd_ioctl(bdev, nbd, cmd, arg);
 846         mutex_unlock(&nbd->config_lock);
 847
 848         return error;
 849 }
 850
 851 static const struct block_device_operations nbd_fops =
 852 {
 853         .owner =        THIS_MODULE,
 854         .ioctl =        nbd_ioctl,
 855         .compat_ioctl = nbd_ioctl,
 856 };
 857
 858 #if IS_ENABLED(CONFIG_DEBUG_FS)
 859
 860 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
 861 {
 862         struct nbd_device *nbd = s->private;
 863
 864         if (nbd->task_recv)
 865                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
 866
 867         return 0;
 868 }
 869
 870 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
 871 {
 872         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
 873 }
 874
 875 static const struct file_operations nbd_dbg_tasks_ops = {
 876         .open = nbd_dbg_tasks_open,
 877         .read = seq_read,
 878         .llseek = seq_lseek,
 879         .release = single_release,
 880 };
 881
 882 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
 883 {
 884         struct nbd_device *nbd = s->private;
 885         u32 flags = nbd->flags;
 886
 887         seq_printf(s, "Hex: 0x%08x\n\n", flags);
 888
 889         seq_puts(s, "Known flags:\n");
 890
 891         if (flags & NBD_FLAG_HAS_FLAGS)
 892                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
 893         if (flags & NBD_FLAG_READ_ONLY)
 894                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
 895         if (flags & NBD_FLAG_SEND_FLUSH)
 896                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
 897         if (flags & NBD_FLAG_SEND_TRIM)
 898                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
 899
 900         return 0;
 901 }
 902
 903 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
 904 {
 905         return single_open(file, nbd_dbg_flags_show, inode->i_private);
 906 }
 907
 908 static const struct file_operations nbd_dbg_flags_ops = {
 909         .open = nbd_dbg_flags_open,
 910         .read = seq_read,
 911         .llseek = seq_lseek,
 912         .release = single_release,
 913 };
 914
 915 static int nbd_dev_dbg_init(struct nbd_device *nbd)
 916 {
 917         struct dentry *dir;
 918
 919         if (!nbd_dbg_dir)
 920                 return -EIO;
 921
 922         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
 923         if (!dir) {
 924                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
 925                         nbd_name(nbd));
 926                 return -EIO;
 927         }
 928         nbd->dbg_dir = dir;
 929
 930         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
 931         debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize);
 932         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
 933         debugfs_create_u64("blocksize", 0444, dir, &nbd->blksize);
 934         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
 935
 936         return 0;
 937 }
 938
 939 static void nbd_dev_dbg_close(struct nbd_device *nbd)
 940 {
 941         debugfs_remove_recursive(nbd->dbg_dir);
 942 }
 943
 944 static int nbd_dbg_init(void)
 945 {
 946         struct dentry *dbg_dir;
 947
 948         dbg_dir = debugfs_create_dir("nbd", NULL);
 949         if (!dbg_dir)
 950                 return -EIO;
 951
 952         nbd_dbg_dir = dbg_dir;
 953
 954         return 0;
 955 }
 956
 957 static void nbd_dbg_close(void)
 958 {
 959         debugfs_remove_recursive(nbd_dbg_dir);
 960 }
 961
 962 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
 963
 964 static int nbd_dev_dbg_init(struct nbd_device *nbd)
 965 {
 966         return 0;
 967 }
 968
 969 static void nbd_dev_dbg_close(struct nbd_device *nbd)
 970 {
 971 }
 972
 973 static int nbd_dbg_init(void)
 974 {
 975         return 0;
 976 }
 977
 978 static void nbd_dbg_close(void)
 979 {
 980 }
 981
 982 #endif
 983
 984 static int nbd_init_request(void *data, struct request *rq,
 985                             unsigned int hctx_idx, unsigned int request_idx,
 986                             unsigned int numa_node)
 987 {
 988         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
 989         cmd->nbd = data;
 990         return 0;
 991 }
 992
 993 static struct blk_mq_ops nbd_mq_ops = {
 994         .queue_rq       = nbd_queue_rq,
 995         .init_request   = nbd_init_request,
 996         .timeout        = nbd_xmit_timeout,
 997 };
 998
 999 static void nbd_dev_remove(struct nbd_device *nbd)
1000 {
1001         struct gendisk *disk = nbd->disk;
1002         nbd->magic = 0;
1003         if (disk) {
1004                 del_gendisk(disk);
1005                 blk_cleanup_queue(disk->queue);
1006                 blk_mq_free_tag_set(&nbd->tag_set);
1007                 put_disk(disk);
1008         }
1009         kfree(nbd);
1010 }
1011
1012 static int nbd_dev_add(int index)
1013 {
1014         struct nbd_device *nbd;
1015         struct gendisk *disk;
1016         struct request_queue *q;
1017         int err = -ENOMEM;
1018
1019         nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1020         if (!nbd)
1021                 goto out;
1022
1023         disk = alloc_disk(1 << part_shift);
1024         if (!disk)
1025                 goto out_free_nbd;
1026
1027         if (index >= 0) {
1028                 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1029                                 GFP_KERNEL);
1030                 if (err == -ENOSPC)
1031                         err = -EEXIST;
1032         } else {
1033                 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1034                 if (err >= 0)
1035                         index = err;
1036         }
1037         if (err < 0)
1038                 goto out_free_disk;
1039
1040         nbd->disk = disk;
1041         nbd->tag_set.ops = &nbd_mq_ops;
1042         nbd->tag_set.nr_hw_queues = 1;
1043         nbd->tag_set.queue_depth = 128;
1044         nbd->tag_set.numa_node = NUMA_NO_NODE;
1045         nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1046         nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1047                 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1048         nbd->tag_set.driver_data = nbd;
1049
1050         err = blk_mq_alloc_tag_set(&nbd->tag_set);
1051         if (err)
1052                 goto out_free_idr;
1053
1054         q = blk_mq_init_queue(&nbd->tag_set);
1055         if (IS_ERR(q)) {
1056                 err = PTR_ERR(q);
1057                 goto out_free_tags;
1058         }
1059         disk->queue = q;
1060
1061         /*
1062          * Tell the block layer that we are not a rotational device
1063          */
1064         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1065         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1066         disk->queue->limits.discard_granularity = 512;
1067         blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1068         disk->queue->limits.discard_zeroes_data = 0;
1069         blk_queue_max_hw_sectors(disk->queue, 65536);
1070         disk->queue->limits.max_sectors = 256;
1071
1072         nbd->magic = NBD_MAGIC;
1073         mutex_init(&nbd->config_lock);
1074         disk->major = NBD_MAJOR;
1075         disk->first_minor = index << part_shift;
1076         disk->fops = &nbd_fops;
1077         disk->private_data = nbd;
1078         sprintf(disk->disk_name, "nbd%d", index);
1079         init_waitqueue_head(&nbd->recv_wq);
1080         nbd_reset(nbd);
1081         add_disk(disk);
1082         return index;
1083
1084 out_free_tags:
1085         blk_mq_free_tag_set(&nbd->tag_set);
1086 out_free_idr:
1087         idr_remove(&nbd_index_idr, index);
1088 out_free_disk:
1089         put_disk(disk);
1090 out_free_nbd:
1091         kfree(nbd);
1092 out:
1093         return err;
1094 }
1095
1096 /*
1097  * And here should be modules and kernel interface
1098  *  (Just smiley confuses emacs :-)
1099  */
1100
1101 static int __init nbd_init(void)
1102 {
1103         int i;
1104
1105         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
1106
1107         if (max_part < 0) {
1108                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
1109                 return -EINVAL;
1110         }
1111
1112         part_shift = 0;
1113         if (max_part > 0) {
1114                 part_shift = fls(max_part);
1115
1116                 /*
1117                  * Adjust max_part according to part_shift as it is exported
1118                  * to user space so that user can know the max number of
1119                  * partition kernel should be able to manage.
1120                  *
1121                  * Note that -1 is required because partition 0 is reserved
1122                  * for the whole disk.
1123                  */
1124                 max_part = (1UL << part_shift) - 1;
1125         }
1126
1127         if ((1UL << part_shift) > DISK_MAX_PARTS)
1128                 return -EINVAL;
1129
1130         if (nbds_max > 1UL << (MINORBITS - part_shift))
1131                 return -EINVAL;
1132         recv_workqueue = alloc_workqueue("knbd-recv",
1133                                          WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1134         if (!recv_workqueue)
1135                 return -ENOMEM;
1136
1137         if (register_blkdev(NBD_MAJOR, "nbd"))
1138                 return -EIO;
1139
1140         nbd_dbg_init();
1141
1142         mutex_lock(&nbd_index_mutex);
1143         for (i = 0; i < nbds_max; i++)
1144                 nbd_dev_add(i);
1145         mutex_unlock(&nbd_index_mutex);
1146         return 0;
1147 }
1148
1149 static int nbd_exit_cb(int id, void *ptr, void *data)
1150 {
1151         struct nbd_device *nbd = ptr;
1152         nbd_dev_remove(nbd);
1153         return 0;
1154 }
1155
1156 static void __exit nbd_cleanup(void)
1157 {
1158         nbd_dbg_close();
1159
1160         idr_for_each(&nbd_index_idr, &nbd_exit_cb, NULL);
1161         idr_destroy(&nbd_index_idr);
1162         destroy_workqueue(recv_workqueue);
1163         unregister_blkdev(NBD_MAJOR, "nbd");
1164 }
1165
1166 module_init(nbd_init);
1167 module_exit(nbd_cleanup);
1168
1169 MODULE_DESCRIPTION("Network Block Device");
1170 MODULE_LICENSE("GPL");
1171
1172 module_param(nbds_max, int, 0444);
1173 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
1174 module_param(max_part, int, 0444);
1175 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");