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/sched/mm.h>
  22 #include <linux/fs.h>
  23 #include <linux/bio.h>
  24 #include <linux/stat.h>
  25 #include <linux/errno.h>
  26 #include <linux/file.h>
  27 #include <linux/ioctl.h>
  28 #include <linux/mutex.h>
  29 #include <linux/compiler.h>
  30 #include <linux/err.h>
  31 #include <linux/kernel.h>
  32 #include <linux/slab.h>
  33 #include <net/sock.h>
  34 #include <linux/net.h>
  35 #include <linux/kthread.h>
  36 #include <linux/types.h>
  37 #include <linux/debugfs.h>
  38 #include <linux/blk-mq.h>
  39
  40 #include <linux/uaccess.h>
  41 #include <asm/types.h>
  42
  43 #include <linux/nbd.h>
  44 #include <linux/nbd-netlink.h>
  45 #include <net/genetlink.h>
  46
  47 #define CREATE_TRACE_POINTS
  48 #include <trace/events/nbd.h>
  49
  50 static DEFINE_IDR(nbd_index_idr);
  51 static DEFINE_MUTEX(nbd_index_mutex);
  52 static int nbd_total_devices = 0;
  53
  54 struct nbd_sock {
  55         struct socket *sock;
  56         struct mutex tx_lock;
  57         struct request *pending;
  58         int sent;
  59         bool dead;
  60         int fallback_index;
  61         int cookie;
  62 };
  63
  64 struct recv_thread_args {
  65         struct work_struct work;
  66         struct nbd_device *nbd;
  67         int index;
  68 };
  69
  70 struct link_dead_args {
  71         struct work_struct work;
  72         int index;
  73 };
  74
  75 #define NBD_TIMEDOUT                    0
  76 #define NBD_DISCONNECT_REQUESTED        1
  77 #define NBD_DISCONNECTED                2
  78 #define NBD_HAS_PID_FILE                3
  79 #define NBD_HAS_CONFIG_REF              4
  80 #define NBD_BOUND                       5
  81 #define NBD_DESTROY_ON_DISCONNECT       6
  82 #define NBD_DISCONNECT_ON_CLOSE         7
  83
  84 struct nbd_config {
  85         u32 flags;
  86         unsigned long runtime_flags;
  87         u64 dead_conn_timeout;
  88
  89         struct nbd_sock **socks;
  90         int num_connections;
  91         atomic_t live_connections;
  92         wait_queue_head_t conn_wait;
  93
  94         atomic_t recv_threads;
  95         wait_queue_head_t recv_wq;
  96         loff_t blksize;
  97         loff_t bytesize;
  98 #if IS_ENABLED(CONFIG_DEBUG_FS)
  99         struct dentry *dbg_dir;
 100 #endif
 101 };
 102
 103 struct nbd_device {
 104         struct blk_mq_tag_set tag_set;
 105
 106         int index;
 107         refcount_t config_refs;
 108         refcount_t refs;
 109         struct nbd_config *config;
 110         struct mutex config_lock;
 111         struct gendisk *disk;
 112
 113         struct list_head list;
 114         struct task_struct *task_recv;
 115         struct task_struct *task_setup;
 116 };
 117
 118 #define NBD_CMD_REQUEUED        1
 119
 120 struct nbd_cmd {
 121         struct nbd_device *nbd;
 122         struct mutex lock;
 123         int index;
 124         int cookie;
 125         blk_status_t status;
 126         unsigned long flags;
 127         u32 cmd_cookie;
 128 };
 129
 130 #if IS_ENABLED(CONFIG_DEBUG_FS)
 131 static struct dentry *nbd_dbg_dir;
 132 #endif
 133
 134 #define nbd_name(nbd) ((nbd)->disk->disk_name)
 135
 136 #define NBD_MAGIC 0x68797548
 137
 138 static unsigned int nbds_max = 16;
 139 static int max_part = 16;
 140 static struct workqueue_struct *recv_workqueue;
 141 static int part_shift;
 142
 143 static int nbd_dev_dbg_init(struct nbd_device *nbd);
 144 static void nbd_dev_dbg_close(struct nbd_device *nbd);
 145 static void nbd_config_put(struct nbd_device *nbd);
 146 static void nbd_connect_reply(struct genl_info *info, int index);
 147 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
 148 static void nbd_dead_link_work(struct work_struct *work);
 149 static void nbd_disconnect_and_put(struct nbd_device *nbd);
 150
 151 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
 152 {
 153         return disk_to_dev(nbd->disk);
 154 }
 155
 156 static void nbd_requeue_cmd(struct nbd_cmd *cmd)
 157 {
 158         struct request *req = blk_mq_rq_from_pdu(cmd);
 159
 160         if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
 161                 blk_mq_requeue_request(req, true);
 162 }
 163
 164 #define NBD_COOKIE_BITS 32
 165
 166 static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
 167 {
 168         struct request *req = blk_mq_rq_from_pdu(cmd);
 169         u32 tag = blk_mq_unique_tag(req);
 170         u64 cookie = cmd->cmd_cookie;
 171
 172         return (cookie << NBD_COOKIE_BITS) | tag;
 173 }
 174
 175 static u32 nbd_handle_to_tag(u64 handle)
 176 {
 177         return (u32)handle;
 178 }
 179
 180 static u32 nbd_handle_to_cookie(u64 handle)
 181 {
 182         return (u32)(handle >> NBD_COOKIE_BITS);
 183 }
 184
 185 static const char *nbdcmd_to_ascii(int cmd)
 186 {
 187         switch (cmd) {
 188         case  NBD_CMD_READ: return "read";
 189         case NBD_CMD_WRITE: return "write";
 190         case  NBD_CMD_DISC: return "disconnect";
 191         case NBD_CMD_FLUSH: return "flush";
 192         case  NBD_CMD_TRIM: return "trim/discard";
 193         }
 194         return "invalid";
 195 }
 196
 197 static ssize_t pid_show(struct device *dev,
 198                         struct device_attribute *attr, char *buf)
 199 {
 200         struct gendisk *disk = dev_to_disk(dev);
 201         struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
 202
 203         return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
 204 }
 205
 206 static const struct device_attribute pid_attr = {
 207         .attr = { .name = "pid", .mode = 0444},
 208         .show = pid_show,
 209 };
 210
 211 static void nbd_dev_remove(struct nbd_device *nbd)
 212 {
 213         struct gendisk *disk = nbd->disk;
 214         struct request_queue *q;
 215
 216         if (disk) {
 217                 q = disk->queue;
 218                 del_gendisk(disk);
 219                 blk_cleanup_queue(q);
 220                 blk_mq_free_tag_set(&nbd->tag_set);
 221                 disk->private_data = NULL;
 222                 put_disk(disk);
 223         }
 224         kfree(nbd);
 225 }
 226
 227 static void nbd_put(struct nbd_device *nbd)
 228 {
 229         if (refcount_dec_and_mutex_lock(&nbd->refs,
 230                                         &nbd_index_mutex)) {
 231                 idr_remove(&nbd_index_idr, nbd->index);
 232                 mutex_unlock(&nbd_index_mutex);
 233                 nbd_dev_remove(nbd);
 234         }
 235 }
 236
 237 static int nbd_disconnected(struct nbd_config *config)
 238 {
 239         return test_bit(NBD_DISCONNECTED, &config->runtime_flags) ||
 240                 test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
 241 }
 242
 243 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
 244                                 int notify)
 245 {
 246         if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
 247                 struct link_dead_args *args;
 248                 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
 249                 if (args) {
 250                         INIT_WORK(&args->work, nbd_dead_link_work);
 251                         args->index = nbd->index;
 252                         queue_work(system_wq, &args->work);
 253                 }
 254         }
 255         if (!nsock->dead) {
 256                 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
 257                 if (atomic_dec_return(&nbd->config->live_connections) == 0) {
 258                         if (test_and_clear_bit(NBD_DISCONNECT_REQUESTED,
 259                                                &nbd->config->runtime_flags)) {
 260                                 set_bit(NBD_DISCONNECTED,
 261                                         &nbd->config->runtime_flags);
 262                                 dev_info(nbd_to_dev(nbd),
 263                                         "Disconnected due to user request.\n");
 264                         }
 265                 }
 266         }
 267         nsock->dead = true;
 268         nsock->pending = NULL;
 269         nsock->sent = 0;
 270 }
 271
 272 static void nbd_size_clear(struct nbd_device *nbd)
 273 {
 274         if (nbd->config->bytesize) {
 275                 set_capacity(nbd->disk, 0);
 276                 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 277         }
 278 }
 279
 280 static void nbd_size_update(struct nbd_device *nbd)
 281 {
 282         struct nbd_config *config = nbd->config;
 283         struct block_device *bdev = bdget_disk(nbd->disk, 0);
 284
 285         if (config->flags & NBD_FLAG_SEND_TRIM) {
 286                 nbd->disk->queue->limits.discard_granularity = config->blksize;
 287                 nbd->disk->queue->limits.discard_alignment = config->blksize;
 288                 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
 289         }
 290         blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
 291         blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
 292         set_capacity(nbd->disk, config->bytesize >> 9);
 293         if (bdev) {
 294                 if (bdev->bd_disk) {
 295                         bd_set_size(bdev, config->bytesize);
 296                         set_blocksize(bdev, config->blksize);
 297                 } else
 298                         bdev->bd_invalidated = 1;
 299                 bdput(bdev);
 300         }
 301         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 302 }
 303
 304 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
 305                          loff_t nr_blocks)
 306 {
 307         struct nbd_config *config = nbd->config;
 308         config->blksize = blocksize;
 309         config->bytesize = blocksize * nr_blocks;
 310         if (nbd->task_recv != NULL)
 311                 nbd_size_update(nbd);
 312 }
 313
 314 static void nbd_complete_rq(struct request *req)
 315 {
 316         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 317
 318         dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
 319                 cmd->status ? "failed" : "done");
 320
 321         blk_mq_end_request(req, cmd->status);
 322 }
 323
 324 /*
 325  * Forcibly shutdown the socket causing all listeners to error
 326  */
 327 static void sock_shutdown(struct nbd_device *nbd)
 328 {
 329         struct nbd_config *config = nbd->config;
 330         int i;
 331
 332         if (config->num_connections == 0)
 333                 return;
 334         if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
 335                 return;
 336
 337         for (i = 0; i < config->num_connections; i++) {
 338                 struct nbd_sock *nsock = config->socks[i];
 339                 mutex_lock(&nsock->tx_lock);
 340                 nbd_mark_nsock_dead(nbd, nsock, 0);
 341                 mutex_unlock(&nsock->tx_lock);
 342         }
 343         dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
 344 }
 345
 346 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 347                                                  bool reserved)
 348 {
 349         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 350         struct nbd_device *nbd = cmd->nbd;
 351         struct nbd_config *config;
 352
 353         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 354                 cmd->status = BLK_STS_TIMEOUT;
 355                 goto done;
 356         }
 357         config = nbd->config;
 358
 359         if (!mutex_trylock(&cmd->lock))
 360                 return BLK_EH_RESET_TIMER;
 361
 362         if (config->num_connections > 1) {
 363                 dev_err_ratelimited(nbd_to_dev(nbd),
 364                                     "Connection timed out, retrying (%d/%d alive)\n",
 365                                     atomic_read(&config->live_connections),
 366                                     config->num_connections);
 367                 /*
 368                  * Hooray we have more connections, requeue this IO, the submit
 369                  * path will put it on a real connection.
 370                  */
 371                 if (config->socks && config->num_connections > 1) {
 372                         if (cmd->index < config->num_connections) {
 373                                 struct nbd_sock *nsock =
 374                                         config->socks[cmd->index];
 375                                 mutex_lock(&nsock->tx_lock);
 376                                 /* We can have multiple outstanding requests, so
 377                                  * we don't want to mark the nsock dead if we've
 378                                  * already reconnected with a new socket, so
 379                                  * only mark it dead if its the same socket we
 380                                  * were sent out on.
 381                                  */
 382                                 if (cmd->cookie == nsock->cookie)
 383                                         nbd_mark_nsock_dead(nbd, nsock, 1);
 384                                 mutex_unlock(&nsock->tx_lock);
 385                         }
 386                         mutex_unlock(&cmd->lock);
 387                         nbd_requeue_cmd(cmd);
 388                         nbd_config_put(nbd);
 389                         return BLK_EH_DONE;
 390                 }
 391         } else {
 392                 dev_err_ratelimited(nbd_to_dev(nbd),
 393                                     "Connection timed out\n");
 394         }
 395         set_bit(NBD_TIMEDOUT, &config->runtime_flags);
 396         cmd->status = BLK_STS_IOERR;
 397         mutex_unlock(&cmd->lock);
 398         sock_shutdown(nbd);
 399         nbd_config_put(nbd);
 400 done:
 401         blk_mq_complete_request(req);
 402         return BLK_EH_DONE;
 403 }
 404
 405 /*
 406  *  Send or receive packet.
 407  */
 408 static int sock_xmit(struct nbd_device *nbd, int index, int send,
 409                      struct iov_iter *iter, int msg_flags, int *sent)
 410 {
 411         struct nbd_config *config = nbd->config;
 412         struct socket *sock = config->socks[index]->sock;
 413         int result;
 414         struct msghdr msg;
 415         unsigned int noreclaim_flag;
 416
 417         if (unlikely(!sock)) {
 418                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 419                         "Attempted %s on closed socket in sock_xmit\n",
 420                         (send ? "send" : "recv"));
 421                 return -EINVAL;
 422         }
 423
 424         msg.msg_iter = *iter;
 425
 426         noreclaim_flag = memalloc_noreclaim_save();
 427         do {
 428                 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
 429                 msg.msg_name = NULL;
 430                 msg.msg_namelen = 0;
 431                 msg.msg_control = NULL;
 432                 msg.msg_controllen = 0;
 433                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 434
 435                 if (send)
 436                         result = sock_sendmsg(sock, &msg);
 437                 else
 438                         result = sock_recvmsg(sock, &msg, msg.msg_flags);
 439
 440                 if (result <= 0) {
 441                         if (result == 0)
 442                                 result = -EPIPE; /* short read */
 443                         break;
 444                 }
 445                 if (sent)
 446                         *sent += result;
 447         } while (msg_data_left(&msg));
 448
 449         memalloc_noreclaim_restore(noreclaim_flag);
 450
 451         return result;
 452 }
 453
 454 /*
 455  * Different settings for sk->sk_sndtimeo can result in different return values
 456  * if there is a signal pending when we enter sendmsg, because reasons?
 457  */
 458 static inline int was_interrupted(int result)
 459 {
 460         return result == -ERESTARTSYS || result == -EINTR;
 461 }
 462
 463 /* always call with the tx_lock held */
 464 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 465 {
 466         struct request *req = blk_mq_rq_from_pdu(cmd);
 467         struct nbd_config *config = nbd->config;
 468         struct nbd_sock *nsock = config->socks[index];
 469         int result;
 470         struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
 471         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 472         struct iov_iter from;
 473         unsigned long size = blk_rq_bytes(req);
 474         struct bio *bio;
 475         u64 handle;
 476         u32 type;
 477         u32 nbd_cmd_flags = 0;
 478         int sent = nsock->sent, skip = 0;
 479
 480         iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
 481
 482         switch (req_op(req)) {
 483         case REQ_OP_DISCARD:
 484                 type = NBD_CMD_TRIM;
 485                 break;
 486         case REQ_OP_FLUSH:
 487                 type = NBD_CMD_FLUSH;
 488                 break;
 489         case REQ_OP_WRITE:
 490                 type = NBD_CMD_WRITE;
 491                 break;
 492         case REQ_OP_READ:
 493                 type = NBD_CMD_READ;
 494                 break;
 495         default:
 496                 return -EIO;
 497         }
 498
 499         if (rq_data_dir(req) == WRITE &&
 500             (config->flags & NBD_FLAG_READ_ONLY)) {
 501                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 502                                     "Write on read-only\n");
 503                 return -EIO;
 504         }
 505
 506         if (req->cmd_flags & REQ_FUA)
 507                 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
 508
 509         /* We did a partial send previously, and we at least sent the whole
 510          * request struct, so just go and send the rest of the pages in the
 511          * request.
 512          */
 513         if (sent) {
 514                 if (sent >= sizeof(request)) {
 515                         skip = sent - sizeof(request);
 516
 517                         /* initialize handle for tracing purposes */
 518                         handle = nbd_cmd_handle(cmd);
 519
 520                         goto send_pages;
 521                 }
 522                 iov_iter_advance(&from, sent);
 523         } else {
 524                 cmd->cmd_cookie++;
 525         }
 526         cmd->index = index;
 527         cmd->cookie = nsock->cookie;
 528         request.type = htonl(type | nbd_cmd_flags);
 529         if (type != NBD_CMD_FLUSH) {
 530                 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
 531                 request.len = htonl(size);
 532         }
 533         handle = nbd_cmd_handle(cmd);
 534         memcpy(request.handle, &handle, sizeof(handle));
 535
 536         trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
 537
 538         dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
 539                 req, nbdcmd_to_ascii(type),
 540                 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
 541         result = sock_xmit(nbd, index, 1, &from,
 542                         (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
 543         trace_nbd_header_sent(req, handle);
 544         if (result <= 0) {
 545                 if (was_interrupted(result)) {
 546                         /* If we havne't sent anything we can just return BUSY,
 547                          * however if we have sent something we need to make
 548                          * sure we only allow this req to be sent until we are
 549                          * completely done.
 550                          */
 551                         if (sent) {
 552                                 nsock->pending = req;
 553                                 nsock->sent = sent;
 554                         }
 555                         set_bit(NBD_CMD_REQUEUED, &cmd->flags);
 556                         return BLK_STS_RESOURCE;
 557                 }
 558                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 559                         "Send control failed (result %d)\n", result);
 560                 return -EAGAIN;
 561         }
 562 send_pages:
 563         if (type != NBD_CMD_WRITE)
 564                 goto out;
 565
 566         bio = req->bio;
 567         while (bio) {
 568                 struct bio *next = bio->bi_next;
 569                 struct bvec_iter iter;
 570                 struct bio_vec bvec;
 571
 572                 bio_for_each_segment(bvec, bio, iter) {
 573                         bool is_last = !next && bio_iter_last(bvec, iter);
 574                         int flags = is_last ? 0 : MSG_MORE;
 575
 576                         dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
 577                                 req, bvec.bv_len);
 578                         iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
 579                         if (skip) {
 580                                 if (skip >= iov_iter_count(&from)) {
 581                                         skip -= iov_iter_count(&from);
 582                                         continue;
 583                                 }
 584                                 iov_iter_advance(&from, skip);
 585                                 skip = 0;
 586                         }
 587                         result = sock_xmit(nbd, index, 1, &from, flags, &sent);
 588                         if (result <= 0) {
 589                                 if (was_interrupted(result)) {
 590                                         /* We've already sent the header, we
 591                                          * have no choice but to set pending and
 592                                          * return BUSY.
 593                                          */
 594                                         nsock->pending = req;
 595                                         nsock->sent = sent;
 596                                         set_bit(NBD_CMD_REQUEUED, &cmd->flags);
 597                                         return BLK_STS_RESOURCE;
 598                                 }
 599                                 dev_err(disk_to_dev(nbd->disk),
 600                                         "Send data failed (result %d)\n",
 601                                         result);
 602                                 return -EAGAIN;
 603                         }
 604                         /*
 605                          * The completion might already have come in,
 606                          * so break for the last one instead of letting
 607                          * the iterator do it. This prevents use-after-free
 608                          * of the bio.
 609                          */
 610                         if (is_last)
 611                                 break;
 612                 }
 613                 bio = next;
 614         }
 615 out:
 616         trace_nbd_payload_sent(req, handle);
 617         nsock->pending = NULL;
 618         nsock->sent = 0;
 619         return 0;
 620 }
 621
 622 /* NULL returned = something went wrong, inform userspace */
 623 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 624 {
 625         struct nbd_config *config = nbd->config;
 626         int result;
 627         struct nbd_reply reply;
 628         struct nbd_cmd *cmd;
 629         struct request *req = NULL;
 630         u64 handle;
 631         u16 hwq;
 632         u32 tag;
 633         struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
 634         struct iov_iter to;
 635         int ret = 0;
 636
 637         reply.magic = 0;
 638         iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
 639         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 640         if (result <= 0) {
 641                 if (!nbd_disconnected(config))
 642                         dev_err(disk_to_dev(nbd->disk),
 643                                 "Receive control failed (result %d)\n", result);
 644                 return ERR_PTR(result);
 645         }
 646
 647         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 648                 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
 649                                 (unsigned long)ntohl(reply.magic));
 650                 return ERR_PTR(-EPROTO);
 651         }
 652
 653         memcpy(&handle, reply.handle, sizeof(handle));
 654         tag = nbd_handle_to_tag(handle);
 655         hwq = blk_mq_unique_tag_to_hwq(tag);
 656         if (hwq < nbd->tag_set.nr_hw_queues)
 657                 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
 658                                        blk_mq_unique_tag_to_tag(tag));
 659         if (!req || !blk_mq_request_started(req)) {
 660                 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
 661                         tag, req);
 662                 return ERR_PTR(-ENOENT);
 663         }
 664         trace_nbd_header_received(req, handle);
 665         cmd = blk_mq_rq_to_pdu(req);
 666
 667         mutex_lock(&cmd->lock);
 668         if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
 669                 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
 670                         req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
 671                 ret = -ENOENT;
 672                 goto out;
 673         }
 674         if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
 675                 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
 676                         req);
 677                 ret = -ENOENT;
 678                 goto out;
 679         }
 680         if (ntohl(reply.error)) {
 681                 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
 682                         ntohl(reply.error));
 683                 cmd->status = BLK_STS_IOERR;
 684                 goto out;
 685         }
 686
 687         dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
 688         if (rq_data_dir(req) != WRITE) {
 689                 struct req_iterator iter;
 690                 struct bio_vec bvec;
 691
 692                 rq_for_each_segment(bvec, req, iter) {
 693                         iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
 694                         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 695                         if (result <= 0) {
 696                                 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 697                                         result);
 698                                 /*
 699                                  * If we've disconnected or we only have 1
 700                                  * connection then we need to make sure we
 701                                  * complete this request, otherwise error out
 702                                  * and let the timeout stuff handle resubmitting
 703                                  * this request onto another connection.
 704                                  */
 705                                 if (nbd_disconnected(config) ||
 706                                     config->num_connections <= 1) {
 707                                         cmd->status = BLK_STS_IOERR;
 708                                         goto out;
 709                                 }
 710                                 ret = -EIO;
 711                                 goto out;
 712                         }
 713                         dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 714                                 req, bvec.bv_len);
 715                 }
 716         }
 717 out:
 718         trace_nbd_payload_received(req, handle);
 719         mutex_unlock(&cmd->lock);
 720         return ret ? ERR_PTR(ret) : cmd;
 721 }
 722
 723 static void recv_work(struct work_struct *work)
 724 {
 725         struct recv_thread_args *args = container_of(work,
 726                                                      struct recv_thread_args,
 727                                                      work);
 728         struct nbd_device *nbd = args->nbd;
 729         struct nbd_config *config = nbd->config;
 730         struct nbd_cmd *cmd;
 731
 732         while (1) {
 733                 cmd = nbd_read_stat(nbd, args->index);
 734                 if (IS_ERR(cmd)) {
 735                         struct nbd_sock *nsock = config->socks[args->index];
 736
 737                         mutex_lock(&nsock->tx_lock);
 738                         nbd_mark_nsock_dead(nbd, nsock, 1);
 739                         mutex_unlock(&nsock->tx_lock);
 740                         break;
 741                 }
 742
 743                 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
 744         }
 745         atomic_dec(&config->recv_threads);
 746         wake_up(&config->recv_wq);
 747         nbd_config_put(nbd);
 748         kfree(args);
 749 }
 750
 751 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
 752 {
 753         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 754
 755         cmd->status = BLK_STS_IOERR;
 756         blk_mq_complete_request(req);
 757         return true;
 758 }
 759
 760 static void nbd_clear_que(struct nbd_device *nbd)
 761 {
 762         blk_mq_quiesce_queue(nbd->disk->queue);
 763         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 764         blk_mq_unquiesce_queue(nbd->disk->queue);
 765         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 766 }
 767
 768 static int find_fallback(struct nbd_device *nbd, int index)
 769 {
 770         struct nbd_config *config = nbd->config;
 771         int new_index = -1;
 772         struct nbd_sock *nsock = config->socks[index];
 773         int fallback = nsock->fallback_index;
 774
 775         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 776                 return new_index;
 777
 778         if (config->num_connections <= 1) {
 779                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 780                                     "Attempted send on invalid socket\n");
 781                 return new_index;
 782         }
 783
 784         if (fallback >= 0 && fallback < config->num_connections &&
 785             !config->socks[fallback]->dead)
 786                 return fallback;
 787
 788         if (nsock->fallback_index < 0 ||
 789             nsock->fallback_index >= config->num_connections ||
 790             config->socks[nsock->fallback_index]->dead) {
 791                 int i;
 792                 for (i = 0; i < config->num_connections; i++) {
 793                         if (i == index)
 794                                 continue;
 795                         if (!config->socks[i]->dead) {
 796                                 new_index = i;
 797                                 break;
 798                         }
 799                 }
 800                 nsock->fallback_index = new_index;
 801                 if (new_index < 0) {
 802                         dev_err_ratelimited(disk_to_dev(nbd->disk),
 803                                             "Dead connection, failed to find a fallback\n");
 804                         return new_index;
 805                 }
 806         }
 807         new_index = nsock->fallback_index;
 808         return new_index;
 809 }
 810
 811 static int wait_for_reconnect(struct nbd_device *nbd)
 812 {
 813         struct nbd_config *config = nbd->config;
 814         if (!config->dead_conn_timeout)
 815                 return 0;
 816         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 817                 return 0;
 818         return wait_event_timeout(config->conn_wait,
 819                                   atomic_read(&config->live_connections) > 0,
 820                                   config->dead_conn_timeout) > 0;
 821 }
 822
 823 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 824 {
 825         struct request *req = blk_mq_rq_from_pdu(cmd);
 826         struct nbd_device *nbd = cmd->nbd;
 827         struct nbd_config *config;
 828         struct nbd_sock *nsock;
 829         int ret;
 830
 831         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 832                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 833                                     "Socks array is empty\n");
 834                 blk_mq_start_request(req);
 835                 return -EINVAL;
 836         }
 837         config = nbd->config;
 838
 839         if (index >= config->num_connections) {
 840                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 841                                     "Attempted send on invalid socket\n");
 842                 nbd_config_put(nbd);
 843                 blk_mq_start_request(req);
 844                 return -EINVAL;
 845         }
 846         cmd->status = BLK_STS_OK;
 847 again:
 848         nsock = config->socks[index];
 849         mutex_lock(&nsock->tx_lock);
 850         if (nsock->dead) {
 851                 int old_index = index;
 852                 index = find_fallback(nbd, index);
 853                 mutex_unlock(&nsock->tx_lock);
 854                 if (index < 0) {
 855                         if (wait_for_reconnect(nbd)) {
 856                                 index = old_index;
 857                                 goto again;
 858                         }
 859                         /* All the sockets should already be down at this point,
 860                          * we just want to make sure that DISCONNECTED is set so
 861                          * any requests that come in that were queue'ed waiting
 862                          * for the reconnect timer don't trigger the timer again
 863                          * and instead just error out.
 864                          */
 865                         sock_shutdown(nbd);
 866                         nbd_config_put(nbd);
 867                         blk_mq_start_request(req);
 868                         return -EIO;
 869                 }
 870                 goto again;
 871         }
 872
 873         /* Handle the case that we have a pending request that was partially
 874          * transmitted that _has_ to be serviced first.  We need to call requeue
 875          * here so that it gets put _after_ the request that is already on the
 876          * dispatch list.
 877          */
 878         blk_mq_start_request(req);
 879         if (unlikely(nsock->pending && nsock->pending != req)) {
 880                 nbd_requeue_cmd(cmd);
 881                 ret = 0;
 882                 goto out;
 883         }
 884         /*
 885          * Some failures are related to the link going down, so anything that
 886          * returns EAGAIN can be retried on a different socket.
 887          */
 888         ret = nbd_send_cmd(nbd, cmd, index);
 889         if (ret == -EAGAIN) {
 890                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 891                                     "Request send failed, requeueing\n");
 892                 nbd_mark_nsock_dead(nbd, nsock, 1);
 893                 nbd_requeue_cmd(cmd);
 894                 ret = 0;
 895         }
 896 out:
 897         mutex_unlock(&nsock->tx_lock);
 898         nbd_config_put(nbd);
 899         return ret;
 900 }
 901
 902 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 903                         const struct blk_mq_queue_data *bd)
 904 {
 905         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 906         int ret;
 907
 908         /*
 909          * Since we look at the bio's to send the request over the network we
 910          * need to make sure the completion work doesn't mark this request done
 911          * before we are done doing our send.  This keeps us from dereferencing
 912          * freed data if we have particularly fast completions (ie we get the
 913          * completion before we exit sock_xmit on the last bvec) or in the case
 914          * that the server is misbehaving (or there was an error) before we're
 915          * done sending everything over the wire.
 916          */
 917         mutex_lock(&cmd->lock);
 918         clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
 919
 920         /* We can be called directly from the user space process, which means we
 921          * could possibly have signals pending so our sendmsg will fail.  In
 922          * this case we need to return that we are busy, otherwise error out as
 923          * appropriate.
 924          */
 925         ret = nbd_handle_cmd(cmd, hctx->queue_num);
 926         if (ret < 0)
 927                 ret = BLK_STS_IOERR;
 928         else if (!ret)
 929                 ret = BLK_STS_OK;
 930         mutex_unlock(&cmd->lock);
 931
 932         return ret;
 933 }
 934
 935 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
 936                           bool netlink)
 937 {
 938         struct nbd_config *config = nbd->config;
 939         struct socket *sock;
 940         struct nbd_sock **socks;
 941         struct nbd_sock *nsock;
 942         int err;
 943
 944         sock = sockfd_lookup(arg, &err);
 945         if (!sock)
 946                 return err;
 947
 948         if (!netlink && !nbd->task_setup &&
 949             !test_bit(NBD_BOUND, &config->runtime_flags))
 950                 nbd->task_setup = current;
 951
 952         if (!netlink &&
 953             (nbd->task_setup != current ||
 954              test_bit(NBD_BOUND, &config->runtime_flags))) {
 955                 dev_err(disk_to_dev(nbd->disk),
 956                         "Device being setup by another task");
 957                 sockfd_put(sock);
 958                 return -EBUSY;
 959         }
 960
 961         socks = krealloc(config->socks, (config->num_connections + 1) *
 962                          sizeof(struct nbd_sock *), GFP_KERNEL);
 963         if (!socks) {
 964                 sockfd_put(sock);
 965                 return -ENOMEM;
 966         }
 967         nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
 968         if (!nsock) {
 969                 sockfd_put(sock);
 970                 return -ENOMEM;
 971         }
 972
 973         config->socks = socks;
 974
 975         nsock->fallback_index = -1;
 976         nsock->dead = false;
 977         mutex_init(&nsock->tx_lock);
 978         nsock->sock = sock;
 979         nsock->pending = NULL;
 980         nsock->sent = 0;
 981         nsock->cookie = 0;
 982         socks[config->num_connections++] = nsock;
 983         atomic_inc(&config->live_connections);
 984
 985         return 0;
 986 }
 987
 988 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
 989 {
 990         struct nbd_config *config = nbd->config;
 991         struct socket *sock, *old;
 992         struct recv_thread_args *args;
 993         int i;
 994         int err;
 995
 996         sock = sockfd_lookup(arg, &err);
 997         if (!sock)
 998                 return err;
 999
1000         args = kzalloc(sizeof(*args), GFP_KERNEL);
1001         if (!args) {
1002                 sockfd_put(sock);
1003                 return -ENOMEM;
1004         }
1005
1006         for (i = 0; i < config->num_connections; i++) {
1007                 struct nbd_sock *nsock = config->socks[i];
1008
1009                 if (!nsock->dead)
1010                         continue;
1011
1012                 mutex_lock(&nsock->tx_lock);
1013                 if (!nsock->dead) {
1014                         mutex_unlock(&nsock->tx_lock);
1015                         continue;
1016                 }
1017                 sk_set_memalloc(sock->sk);
1018                 if (nbd->tag_set.timeout)
1019                         sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1020                 atomic_inc(&config->recv_threads);
1021                 refcount_inc(&nbd->config_refs);
1022                 old = nsock->sock;
1023                 nsock->fallback_index = -1;
1024                 nsock->sock = sock;
1025                 nsock->dead = false;
1026                 INIT_WORK(&args->work, recv_work);
1027                 args->index = i;
1028                 args->nbd = nbd;
1029                 nsock->cookie++;
1030                 mutex_unlock(&nsock->tx_lock);
1031                 sockfd_put(old);
1032
1033                 clear_bit(NBD_DISCONNECTED, &config->runtime_flags);
1034
1035                 /* We take the tx_mutex in an error path in the recv_work, so we
1036                  * need to queue_work outside of the tx_mutex.
1037                  */
1038                 queue_work(recv_workqueue, &args->work);
1039
1040                 atomic_inc(&config->live_connections);
1041                 wake_up(&config->conn_wait);
1042                 return 0;
1043         }
1044         sockfd_put(sock);
1045         kfree(args);
1046         return -ENOSPC;
1047 }
1048
1049 static void nbd_bdev_reset(struct block_device *bdev)
1050 {
1051         if (bdev->bd_openers > 1)
1052                 return;
1053         bd_set_size(bdev, 0);
1054 }
1055
1056 static void nbd_parse_flags(struct nbd_device *nbd)
1057 {
1058         struct nbd_config *config = nbd->config;
1059         if (config->flags & NBD_FLAG_READ_ONLY)
1060                 set_disk_ro(nbd->disk, true);
1061         else
1062                 set_disk_ro(nbd->disk, false);
1063         if (config->flags & NBD_FLAG_SEND_TRIM)
1064                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1065         if (config->flags & NBD_FLAG_SEND_FLUSH) {
1066                 if (config->flags & NBD_FLAG_SEND_FUA)
1067                         blk_queue_write_cache(nbd->disk->queue, true, true);
1068                 else
1069                         blk_queue_write_cache(nbd->disk->queue, true, false);
1070         }
1071         else
1072                 blk_queue_write_cache(nbd->disk->queue, false, false);
1073 }
1074
1075 static void send_disconnects(struct nbd_device *nbd)
1076 {
1077         struct nbd_config *config = nbd->config;
1078         struct nbd_request request = {
1079                 .magic = htonl(NBD_REQUEST_MAGIC),
1080                 .type = htonl(NBD_CMD_DISC),
1081         };
1082         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1083         struct iov_iter from;
1084         int i, ret;
1085
1086         for (i = 0; i < config->num_connections; i++) {
1087                 struct nbd_sock *nsock = config->socks[i];
1088
1089                 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1090                 mutex_lock(&nsock->tx_lock);
1091                 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1092                 if (ret <= 0)
1093                         dev_err(disk_to_dev(nbd->disk),
1094                                 "Send disconnect failed %d\n", ret);
1095                 mutex_unlock(&nsock->tx_lock);
1096         }
1097 }
1098
1099 static int nbd_disconnect(struct nbd_device *nbd)
1100 {
1101         struct nbd_config *config = nbd->config;
1102
1103         dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1104         set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
1105         send_disconnects(nbd);
1106         return 0;
1107 }
1108
1109 static void nbd_clear_sock(struct nbd_device *nbd)
1110 {
1111         sock_shutdown(nbd);
1112         nbd_clear_que(nbd);
1113         nbd->task_setup = NULL;
1114 }
1115
1116 static void nbd_config_put(struct nbd_device *nbd)
1117 {
1118         if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1119                                         &nbd->config_lock)) {
1120                 struct nbd_config *config = nbd->config;
1121                 nbd_dev_dbg_close(nbd);
1122                 nbd_size_clear(nbd);
1123                 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1124                                        &config->runtime_flags))
1125                         device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1126                 nbd->task_recv = NULL;
1127                 nbd_clear_sock(nbd);
1128                 if (config->num_connections) {
1129                         int i;
1130                         for (i = 0; i < config->num_connections; i++) {
1131                                 sockfd_put(config->socks[i]->sock);
1132                                 kfree(config->socks[i]);
1133                         }
1134                         kfree(config->socks);
1135                 }
1136                 kfree(nbd->config);
1137                 nbd->config = NULL;
1138
1139                 nbd->tag_set.timeout = 0;
1140                 nbd->disk->queue->limits.discard_granularity = 0;
1141                 nbd->disk->queue->limits.discard_alignment = 0;
1142                 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1143                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1144
1145                 mutex_unlock(&nbd->config_lock);
1146                 nbd_put(nbd);
1147                 module_put(THIS_MODULE);
1148         }
1149 }
1150
1151 static int nbd_start_device(struct nbd_device *nbd)
1152 {
1153         struct nbd_config *config = nbd->config;
1154         int num_connections = config->num_connections;
1155         int error = 0, i;
1156
1157         if (nbd->task_recv)
1158                 return -EBUSY;
1159         if (!config->socks)
1160                 return -EINVAL;
1161         if (num_connections > 1 &&
1162             !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1163                 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1164                 return -EINVAL;
1165         }
1166
1167         blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1168         nbd->task_recv = current;
1169
1170         nbd_parse_flags(nbd);
1171
1172         error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1173         if (error) {
1174                 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1175                 return error;
1176         }
1177         set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1178
1179         nbd_dev_dbg_init(nbd);
1180         for (i = 0; i < num_connections; i++) {
1181                 struct recv_thread_args *args;
1182
1183                 args = kzalloc(sizeof(*args), GFP_KERNEL);
1184                 if (!args) {
1185                         sock_shutdown(nbd);
1186                         return -ENOMEM;
1187                 }
1188                 sk_set_memalloc(config->socks[i]->sock->sk);
1189                 if (nbd->tag_set.timeout)
1190                         config->socks[i]->sock->sk->sk_sndtimeo =
1191                                 nbd->tag_set.timeout;
1192                 atomic_inc(&config->recv_threads);
1193                 refcount_inc(&nbd->config_refs);
1194                 INIT_WORK(&args->work, recv_work);
1195                 args->nbd = nbd;
1196                 args->index = i;
1197                 queue_work(recv_workqueue, &args->work);
1198         }
1199         nbd_size_update(nbd);
1200         return error;
1201 }
1202
1203 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1204 {
1205         struct nbd_config *config = nbd->config;
1206         int ret;
1207
1208         ret = nbd_start_device(nbd);
1209         if (ret)
1210                 return ret;
1211
1212         if (max_part)
1213                 bdev->bd_invalidated = 1;
1214         mutex_unlock(&nbd->config_lock);
1215         ret = wait_event_interruptible(config->recv_wq,
1216                                          atomic_read(&config->recv_threads) == 0);
1217         if (ret)
1218                 sock_shutdown(nbd);
1219         mutex_lock(&nbd->config_lock);
1220         nbd_bdev_reset(bdev);
1221         /* user requested, ignore socket errors */
1222         if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1223                 ret = 0;
1224         if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1225                 ret = -ETIMEDOUT;
1226         return ret;
1227 }
1228
1229 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1230                                  struct block_device *bdev)
1231 {
1232         sock_shutdown(nbd);
1233         kill_bdev(bdev);
1234         nbd_bdev_reset(bdev);
1235         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1236                                &nbd->config->runtime_flags))
1237                 nbd_config_put(nbd);
1238 }
1239
1240 /* Must be called with config_lock held */
1241 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1242                        unsigned int cmd, unsigned long arg)
1243 {
1244         struct nbd_config *config = nbd->config;
1245
1246         switch (cmd) {
1247         case NBD_DISCONNECT:
1248                 return nbd_disconnect(nbd);
1249         case NBD_CLEAR_SOCK:
1250                 nbd_clear_sock_ioctl(nbd, bdev);
1251                 return 0;
1252         case NBD_SET_SOCK:
1253                 return nbd_add_socket(nbd, arg, false);
1254         case NBD_SET_BLKSIZE:
1255                 if (!arg || !is_power_of_2(arg) || arg < 512 ||
1256                     arg > PAGE_SIZE)
1257                         return -EINVAL;
1258                 nbd_size_set(nbd, arg,
1259                              div_s64(config->bytesize, arg));
1260                 return 0;
1261         case NBD_SET_SIZE:
1262                 nbd_size_set(nbd, config->blksize,
1263                              div_s64(arg, config->blksize));
1264                 return 0;
1265         case NBD_SET_SIZE_BLOCKS:
1266                 nbd_size_set(nbd, config->blksize, arg);
1267                 return 0;
1268         case NBD_SET_TIMEOUT:
1269                 if (arg) {
1270                         nbd->tag_set.timeout = arg * HZ;
1271                         blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1272                 }
1273                 return 0;
1274
1275         case NBD_SET_FLAGS:
1276                 config->flags = arg;
1277                 return 0;
1278         case NBD_DO_IT:
1279                 return nbd_start_device_ioctl(nbd, bdev);
1280         case NBD_CLEAR_QUE:
1281                 /*
1282                  * This is for compatibility only.  The queue is always cleared
1283                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
1284                  */
1285                 return 0;
1286         case NBD_PRINT_DEBUG:
1287                 /*
1288                  * For compatibility only, we no longer keep a list of
1289                  * outstanding requests.
1290                  */
1291                 return 0;
1292         }
1293         return -ENOTTY;
1294 }
1295
1296 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1297                      unsigned int cmd, unsigned long arg)
1298 {
1299         struct nbd_device *nbd = bdev->bd_disk->private_data;
1300         struct nbd_config *config = nbd->config;
1301         int error = -EINVAL;
1302
1303         if (!capable(CAP_SYS_ADMIN))
1304                 return -EPERM;
1305
1306         /* The block layer will pass back some non-nbd ioctls in case we have
1307          * special handling for them, but we don't so just return an error.
1308          */
1309         if (_IOC_TYPE(cmd) != 0xab)
1310                 return -EINVAL;
1311
1312         mutex_lock(&nbd->config_lock);
1313
1314         /* Don't allow ioctl operations on a nbd device that was created with
1315          * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1316          */
1317         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1318             (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1319                 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1320         else
1321                 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1322         mutex_unlock(&nbd->config_lock);
1323         return error;
1324 }
1325
1326 static struct nbd_config *nbd_alloc_config(void)
1327 {
1328         struct nbd_config *config;
1329
1330         config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1331         if (!config)
1332                 return NULL;
1333         atomic_set(&config->recv_threads, 0);
1334         init_waitqueue_head(&config->recv_wq);
1335         init_waitqueue_head(&config->conn_wait);
1336         config->blksize = 1024;
1337         atomic_set(&config->live_connections, 0);
1338         try_module_get(THIS_MODULE);
1339         return config;
1340 }
1341
1342 static int nbd_open(struct block_device *bdev, fmode_t mode)
1343 {
1344         struct nbd_device *nbd;
1345         int ret = 0;
1346
1347         mutex_lock(&nbd_index_mutex);
1348         nbd = bdev->bd_disk->private_data;
1349         if (!nbd) {
1350                 ret = -ENXIO;
1351                 goto out;
1352         }
1353         if (!refcount_inc_not_zero(&nbd->refs)) {
1354                 ret = -ENXIO;
1355                 goto out;
1356         }
1357         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1358                 struct nbd_config *config;
1359
1360                 mutex_lock(&nbd->config_lock);
1361                 if (refcount_inc_not_zero(&nbd->config_refs)) {
1362                         mutex_unlock(&nbd->config_lock);
1363                         goto out;
1364                 }
1365                 config = nbd->config = nbd_alloc_config();
1366                 if (!config) {
1367                         ret = -ENOMEM;
1368                         mutex_unlock(&nbd->config_lock);
1369                         goto out;
1370                 }
1371                 refcount_set(&nbd->config_refs, 1);
1372                 refcount_inc(&nbd->refs);
1373                 mutex_unlock(&nbd->config_lock);
1374                 bdev->bd_invalidated = 1;
1375         } else if (nbd_disconnected(nbd->config)) {
1376                 bdev->bd_invalidated = 1;
1377         }
1378 out:
1379         mutex_unlock(&nbd_index_mutex);
1380         return ret;
1381 }
1382
1383 static void nbd_release(struct gendisk *disk, fmode_t mode)
1384 {
1385         struct nbd_device *nbd = disk->private_data;
1386         struct block_device *bdev = bdget_disk(disk, 0);
1387
1388         if (test_bit(NBD_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1389                         bdev->bd_openers == 0)
1390                 nbd_disconnect_and_put(nbd);
1391
1392         nbd_config_put(nbd);
1393         nbd_put(nbd);
1394 }
1395
1396 static const struct block_device_operations nbd_fops =
1397 {
1398         .owner =        THIS_MODULE,
1399         .open =         nbd_open,
1400         .release =      nbd_release,
1401         .ioctl =        nbd_ioctl,
1402         .compat_ioctl = nbd_ioctl,
1403 };
1404
1405 #if IS_ENABLED(CONFIG_DEBUG_FS)
1406
1407 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1408 {
1409         struct nbd_device *nbd = s->private;
1410
1411         if (nbd->task_recv)
1412                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1413
1414         return 0;
1415 }
1416
1417 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1418 {
1419         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1420 }
1421
1422 static const struct file_operations nbd_dbg_tasks_ops = {
1423         .open = nbd_dbg_tasks_open,
1424         .read = seq_read,
1425         .llseek = seq_lseek,
1426         .release = single_release,
1427 };
1428
1429 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1430 {
1431         struct nbd_device *nbd = s->private;
1432         u32 flags = nbd->config->flags;
1433
1434         seq_printf(s, "Hex: 0x%08x\n\n", flags);
1435
1436         seq_puts(s, "Known flags:\n");
1437
1438         if (flags & NBD_FLAG_HAS_FLAGS)
1439                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1440         if (flags & NBD_FLAG_READ_ONLY)
1441                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1442         if (flags & NBD_FLAG_SEND_FLUSH)
1443                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1444         if (flags & NBD_FLAG_SEND_FUA)
1445                 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1446         if (flags & NBD_FLAG_SEND_TRIM)
1447                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1448
1449         return 0;
1450 }
1451
1452 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1453 {
1454         return single_open(file, nbd_dbg_flags_show, inode->i_private);
1455 }
1456
1457 static const struct file_operations nbd_dbg_flags_ops = {
1458         .open = nbd_dbg_flags_open,
1459         .read = seq_read,
1460         .llseek = seq_lseek,
1461         .release = single_release,
1462 };
1463
1464 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1465 {
1466         struct dentry *dir;
1467         struct nbd_config *config = nbd->config;
1468
1469         if (!nbd_dbg_dir)
1470                 return -EIO;
1471
1472         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1473         if (!dir) {
1474                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1475                         nbd_name(nbd));
1476                 return -EIO;
1477         }
1478         config->dbg_dir = dir;
1479
1480         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1481         debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1482         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1483         debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1484         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1485
1486         return 0;
1487 }
1488
1489 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1490 {
1491         debugfs_remove_recursive(nbd->config->dbg_dir);
1492 }
1493
1494 static int nbd_dbg_init(void)
1495 {
1496         struct dentry *dbg_dir;
1497
1498         dbg_dir = debugfs_create_dir("nbd", NULL);
1499         if (!dbg_dir)
1500                 return -EIO;
1501
1502         nbd_dbg_dir = dbg_dir;
1503
1504         return 0;
1505 }
1506
1507 static void nbd_dbg_close(void)
1508 {
1509         debugfs_remove_recursive(nbd_dbg_dir);
1510 }
1511
1512 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1513
1514 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1515 {
1516         return 0;
1517 }
1518
1519 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1520 {
1521 }
1522
1523 static int nbd_dbg_init(void)
1524 {
1525         return 0;
1526 }
1527
1528 static void nbd_dbg_close(void)
1529 {
1530 }
1531
1532 #endif
1533
1534 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1535                             unsigned int hctx_idx, unsigned int numa_node)
1536 {
1537         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1538         cmd->nbd = set->driver_data;
1539         cmd->flags = 0;
1540         mutex_init(&cmd->lock);
1541         return 0;
1542 }
1543
1544 static const struct blk_mq_ops nbd_mq_ops = {
1545         .queue_rq       = nbd_queue_rq,
1546         .complete       = nbd_complete_rq,
1547         .init_request   = nbd_init_request,
1548         .timeout        = nbd_xmit_timeout,
1549 };
1550
1551 static int nbd_dev_add(int index)
1552 {
1553         struct nbd_device *nbd;
1554         struct gendisk *disk;
1555         struct request_queue *q;
1556         int err = -ENOMEM;
1557
1558         nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1559         if (!nbd)
1560                 goto out;
1561
1562         disk = alloc_disk(1 << part_shift);
1563         if (!disk)
1564                 goto out_free_nbd;
1565
1566         if (index >= 0) {
1567                 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1568                                 GFP_KERNEL);
1569                 if (err == -ENOSPC)
1570                         err = -EEXIST;
1571         } else {
1572                 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1573                 if (err >= 0)
1574                         index = err;
1575         }
1576         if (err < 0)
1577                 goto out_free_disk;
1578
1579         nbd->index = index;
1580         nbd->disk = disk;
1581         nbd->tag_set.ops = &nbd_mq_ops;
1582         nbd->tag_set.nr_hw_queues = 1;
1583         nbd->tag_set.queue_depth = 128;
1584         nbd->tag_set.numa_node = NUMA_NO_NODE;
1585         nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1586         nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1587                 BLK_MQ_F_BLOCKING;
1588         nbd->tag_set.driver_data = nbd;
1589
1590         err = blk_mq_alloc_tag_set(&nbd->tag_set);
1591         if (err)
1592                 goto out_free_idr;
1593
1594         q = blk_mq_init_queue(&nbd->tag_set);
1595         if (IS_ERR(q)) {
1596                 err = PTR_ERR(q);
1597                 goto out_free_tags;
1598         }
1599         disk->queue = q;
1600
1601         /*
1602          * Tell the block layer that we are not a rotational device
1603          */
1604         blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1605         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1606         disk->queue->limits.discard_granularity = 0;
1607         disk->queue->limits.discard_alignment = 0;
1608         blk_queue_max_discard_sectors(disk->queue, 0);
1609         blk_queue_max_segment_size(disk->queue, UINT_MAX);
1610         blk_queue_max_segments(disk->queue, USHRT_MAX);
1611         blk_queue_max_hw_sectors(disk->queue, 65536);
1612         disk->queue->limits.max_sectors = 256;
1613
1614         mutex_init(&nbd->config_lock);
1615         refcount_set(&nbd->config_refs, 0);
1616         refcount_set(&nbd->refs, 1);
1617         INIT_LIST_HEAD(&nbd->list);
1618         disk->major = NBD_MAJOR;
1619         disk->first_minor = index << part_shift;
1620         disk->fops = &nbd_fops;
1621         disk->private_data = nbd;
1622         sprintf(disk->disk_name, "nbd%d", index);
1623         add_disk(disk);
1624         nbd_total_devices++;
1625         return index;
1626
1627 out_free_tags:
1628         blk_mq_free_tag_set(&nbd->tag_set);
1629 out_free_idr:
1630         idr_remove(&nbd_index_idr, index);
1631 out_free_disk:
1632         put_disk(disk);
1633 out_free_nbd:
1634         kfree(nbd);
1635 out:
1636         return err;
1637 }
1638
1639 static int find_free_cb(int id, void *ptr, void *data)
1640 {
1641         struct nbd_device *nbd = ptr;
1642         struct nbd_device **found = data;
1643
1644         if (!refcount_read(&nbd->config_refs)) {
1645                 *found = nbd;
1646                 return 1;
1647         }
1648         return 0;
1649 }
1650
1651 /* Netlink interface. */
1652 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1653         [NBD_ATTR_INDEX]                =       { .type = NLA_U32 },
1654         [NBD_ATTR_SIZE_BYTES]           =       { .type = NLA_U64 },
1655         [NBD_ATTR_BLOCK_SIZE_BYTES]     =       { .type = NLA_U64 },
1656         [NBD_ATTR_TIMEOUT]              =       { .type = NLA_U64 },
1657         [NBD_ATTR_SERVER_FLAGS]         =       { .type = NLA_U64 },
1658         [NBD_ATTR_CLIENT_FLAGS]         =       { .type = NLA_U64 },
1659         [NBD_ATTR_SOCKETS]              =       { .type = NLA_NESTED},
1660         [NBD_ATTR_DEAD_CONN_TIMEOUT]    =       { .type = NLA_U64 },
1661         [NBD_ATTR_DEVICE_LIST]          =       { .type = NLA_NESTED},
1662 };
1663
1664 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1665         [NBD_SOCK_FD]                   =       { .type = NLA_U32 },
1666 };
1667
1668 /* We don't use this right now since we don't parse the incoming list, but we
1669  * still want it here so userspace knows what to expect.
1670  */
1671 static const struct nla_policy __attribute__((unused))
1672 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1673         [NBD_DEVICE_INDEX]              =       { .type = NLA_U32 },
1674         [NBD_DEVICE_CONNECTED]          =       { .type = NLA_U8 },
1675 };
1676
1677 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1678 {
1679         struct nbd_device *nbd = NULL;
1680         struct nbd_config *config;
1681         int index = -1;
1682         int ret;
1683         bool put_dev = false;
1684
1685         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1686                 return -EPERM;
1687
1688         if (info->attrs[NBD_ATTR_INDEX])
1689                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1690         if (!info->attrs[NBD_ATTR_SOCKETS]) {
1691                 printk(KERN_ERR "nbd: must specify at least one socket\n");
1692                 return -EINVAL;
1693         }
1694         if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1695                 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1696                 return -EINVAL;
1697         }
1698 again:
1699         mutex_lock(&nbd_index_mutex);
1700         if (index == -1) {
1701                 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1702                 if (ret == 0) {
1703                         int new_index;
1704                         new_index = nbd_dev_add(-1);
1705                         if (new_index < 0) {
1706                                 mutex_unlock(&nbd_index_mutex);
1707                                 printk(KERN_ERR "nbd: failed to add new device\n");
1708                                 return new_index;
1709                         }
1710                         nbd = idr_find(&nbd_index_idr, new_index);
1711                 }
1712         } else {
1713                 nbd = idr_find(&nbd_index_idr, index);
1714                 if (!nbd) {
1715                         ret = nbd_dev_add(index);
1716                         if (ret < 0) {
1717                                 mutex_unlock(&nbd_index_mutex);
1718                                 printk(KERN_ERR "nbd: failed to add new device\n");
1719                                 return ret;
1720                         }
1721                         nbd = idr_find(&nbd_index_idr, index);
1722                 }
1723         }
1724         if (!nbd) {
1725                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1726                        index);
1727                 mutex_unlock(&nbd_index_mutex);
1728                 return -EINVAL;
1729         }
1730         if (!refcount_inc_not_zero(&nbd->refs)) {
1731                 mutex_unlock(&nbd_index_mutex);
1732                 if (index == -1)
1733                         goto again;
1734                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1735                        index);
1736                 return -EINVAL;
1737         }
1738         mutex_unlock(&nbd_index_mutex);
1739
1740         mutex_lock(&nbd->config_lock);
1741         if (refcount_read(&nbd->config_refs)) {
1742                 mutex_unlock(&nbd->config_lock);
1743                 nbd_put(nbd);
1744                 if (index == -1)
1745                         goto again;
1746                 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1747                 return -EBUSY;
1748         }
1749         if (WARN_ON(nbd->config)) {
1750                 mutex_unlock(&nbd->config_lock);
1751                 nbd_put(nbd);
1752                 return -EINVAL;
1753         }
1754         config = nbd->config = nbd_alloc_config();
1755         if (!nbd->config) {
1756                 mutex_unlock(&nbd->config_lock);
1757                 nbd_put(nbd);
1758                 printk(KERN_ERR "nbd: couldn't allocate config\n");
1759                 return -ENOMEM;
1760         }
1761         refcount_set(&nbd->config_refs, 1);
1762         set_bit(NBD_BOUND, &config->runtime_flags);
1763
1764         if (info->attrs[NBD_ATTR_SIZE_BYTES]) {
1765                 u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1766                 nbd_size_set(nbd, config->blksize,
1767                              div64_u64(bytes, config->blksize));
1768         }
1769         if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1770                 u64 bsize =
1771                         nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1772                 nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize));
1773         }
1774         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1775                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1776                 nbd->tag_set.timeout = timeout * HZ;
1777                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1778         }
1779         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1780                 config->dead_conn_timeout =
1781                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1782                 config->dead_conn_timeout *= HZ;
1783         }
1784         if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1785                 config->flags =
1786                         nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1787         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1788                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1789                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1790                         set_bit(NBD_DESTROY_ON_DISCONNECT,
1791                                 &config->runtime_flags);
1792                         put_dev = true;
1793                 }
1794                 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1795                         set_bit(NBD_DISCONNECT_ON_CLOSE,
1796                                 &config->runtime_flags);
1797                 }
1798         }
1799
1800         if (info->attrs[NBD_ATTR_SOCKETS]) {
1801                 struct nlattr *attr;
1802                 int rem, fd;
1803
1804                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1805                                     rem) {
1806                         struct nlattr *socks[NBD_SOCK_MAX+1];
1807
1808                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1809                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1810                                 ret = -EINVAL;
1811                                 goto out;
1812                         }
1813                         ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1814                                                           attr,
1815                                                           nbd_sock_policy,
1816                                                           info->extack);
1817                         if (ret != 0) {
1818                                 printk(KERN_ERR "nbd: error processing sock list\n");
1819                                 ret = -EINVAL;
1820                                 goto out;
1821                         }
1822                         if (!socks[NBD_SOCK_FD])
1823                                 continue;
1824                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1825                         ret = nbd_add_socket(nbd, fd, true);
1826                         if (ret)
1827                                 goto out;
1828                 }
1829         }
1830         ret = nbd_start_device(nbd);
1831 out:
1832         mutex_unlock(&nbd->config_lock);
1833         if (!ret) {
1834                 set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1835                 refcount_inc(&nbd->config_refs);
1836                 nbd_connect_reply(info, nbd->index);
1837         }
1838         nbd_config_put(nbd);
1839         if (put_dev)
1840                 nbd_put(nbd);
1841         return ret;
1842 }
1843
1844 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1845 {
1846         mutex_lock(&nbd->config_lock);
1847         nbd_disconnect(nbd);
1848         nbd_clear_sock(nbd);
1849         mutex_unlock(&nbd->config_lock);
1850         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1851                                &nbd->config->runtime_flags))
1852                 nbd_config_put(nbd);
1853 }
1854
1855 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1856 {
1857         struct nbd_device *nbd;
1858         int index;
1859
1860         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1861                 return -EPERM;
1862
1863         if (!info->attrs[NBD_ATTR_INDEX]) {
1864                 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1865                 return -EINVAL;
1866         }
1867         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1868         mutex_lock(&nbd_index_mutex);
1869         nbd = idr_find(&nbd_index_idr, index);
1870         if (!nbd) {
1871                 mutex_unlock(&nbd_index_mutex);
1872                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1873                        index);
1874                 return -EINVAL;
1875         }
1876         if (!refcount_inc_not_zero(&nbd->refs)) {
1877                 mutex_unlock(&nbd_index_mutex);
1878                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1879                        index);
1880                 return -EINVAL;
1881         }
1882         mutex_unlock(&nbd_index_mutex);
1883         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1884                 nbd_put(nbd);
1885                 return 0;
1886         }
1887         nbd_disconnect_and_put(nbd);
1888         nbd_config_put(nbd);
1889         nbd_put(nbd);
1890         return 0;
1891 }
1892
1893 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1894 {
1895         struct nbd_device *nbd = NULL;
1896         struct nbd_config *config;
1897         int index;
1898         int ret = 0;
1899         bool put_dev = false;
1900
1901         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1902                 return -EPERM;
1903
1904         if (!info->attrs[NBD_ATTR_INDEX]) {
1905                 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1906                 return -EINVAL;
1907         }
1908         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1909         mutex_lock(&nbd_index_mutex);
1910         nbd = idr_find(&nbd_index_idr, index);
1911         if (!nbd) {
1912                 mutex_unlock(&nbd_index_mutex);
1913                 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1914                        index);
1915                 return -EINVAL;
1916         }
1917         if (!refcount_inc_not_zero(&nbd->refs)) {
1918                 mutex_unlock(&nbd_index_mutex);
1919                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1920                        index);
1921                 return -EINVAL;
1922         }
1923         mutex_unlock(&nbd_index_mutex);
1924
1925         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1926                 dev_err(nbd_to_dev(nbd),
1927                         "not configured, cannot reconfigure\n");
1928                 nbd_put(nbd);
1929                 return -EINVAL;
1930         }
1931
1932         mutex_lock(&nbd->config_lock);
1933         config = nbd->config;
1934         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1935             !nbd->task_recv) {
1936                 dev_err(nbd_to_dev(nbd),
1937                         "not configured, cannot reconfigure\n");
1938                 ret = -EINVAL;
1939                 goto out;
1940         }
1941
1942         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1943                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1944                 nbd->tag_set.timeout = timeout * HZ;
1945                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1946         }
1947         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1948                 config->dead_conn_timeout =
1949                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1950                 config->dead_conn_timeout *= HZ;
1951         }
1952         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1953                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1954                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1955                         if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1956                                               &config->runtime_flags))
1957                                 put_dev = true;
1958                 } else {
1959                         if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1960                                                &config->runtime_flags))
1961                                 refcount_inc(&nbd->refs);
1962                 }
1963
1964                 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1965                         set_bit(NBD_DISCONNECT_ON_CLOSE,
1966                                         &config->runtime_flags);
1967                 } else {
1968                         clear_bit(NBD_DISCONNECT_ON_CLOSE,
1969                                         &config->runtime_flags);
1970                 }
1971         }
1972
1973         if (info->attrs[NBD_ATTR_SOCKETS]) {
1974                 struct nlattr *attr;
1975                 int rem, fd;
1976
1977                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1978                                     rem) {
1979                         struct nlattr *socks[NBD_SOCK_MAX+1];
1980
1981                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1982                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1983                                 ret = -EINVAL;
1984                                 goto out;
1985                         }
1986                         ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1987                                                           attr,
1988                                                           nbd_sock_policy,
1989                                                           info->extack);
1990                         if (ret != 0) {
1991                                 printk(KERN_ERR "nbd: error processing sock list\n");
1992                                 ret = -EINVAL;
1993                                 goto out;
1994                         }
1995                         if (!socks[NBD_SOCK_FD])
1996                                 continue;
1997                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1998                         ret = nbd_reconnect_socket(nbd, fd);
1999                         if (ret) {
2000                                 if (ret == -ENOSPC)
2001                                         ret = 0;
2002                                 goto out;
2003                         }
2004                         dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2005                 }
2006         }
2007 out:
2008         mutex_unlock(&nbd->config_lock);
2009         nbd_config_put(nbd);
2010         nbd_put(nbd);
2011         if (put_dev)
2012                 nbd_put(nbd);
2013         return ret;
2014 }
2015
2016 static const struct genl_ops nbd_connect_genl_ops[] = {
2017         {
2018                 .cmd    = NBD_CMD_CONNECT,
2019                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2020                 .doit   = nbd_genl_connect,
2021         },
2022         {
2023                 .cmd    = NBD_CMD_DISCONNECT,
2024                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2025                 .doit   = nbd_genl_disconnect,
2026         },
2027         {
2028                 .cmd    = NBD_CMD_RECONFIGURE,
2029                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2030                 .doit   = nbd_genl_reconfigure,
2031         },
2032         {
2033                 .cmd    = NBD_CMD_STATUS,
2034                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2035                 .doit   = nbd_genl_status,
2036         },
2037 };
2038
2039 static const struct genl_multicast_group nbd_mcast_grps[] = {
2040         { .name = NBD_GENL_MCAST_GROUP_NAME, },
2041 };
2042
2043 static struct genl_family nbd_genl_family __ro_after_init = {
2044         .hdrsize        = 0,
2045         .name           = NBD_GENL_FAMILY_NAME,
2046         .version        = NBD_GENL_VERSION,
2047         .module         = THIS_MODULE,
2048         .ops            = nbd_connect_genl_ops,
2049         .n_ops          = ARRAY_SIZE(nbd_connect_genl_ops),
2050         .maxattr        = NBD_ATTR_MAX,
2051         .policy = nbd_attr_policy,
2052         .mcgrps         = nbd_mcast_grps,
2053         .n_mcgrps       = ARRAY_SIZE(nbd_mcast_grps),
2054 };
2055
2056 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2057 {
2058         struct nlattr *dev_opt;
2059         u8 connected = 0;
2060         int ret;
2061
2062         /* This is a little racey, but for status it's ok.  The
2063          * reason we don't take a ref here is because we can't
2064          * take a ref in the index == -1 case as we would need
2065          * to put under the nbd_index_mutex, which could
2066          * deadlock if we are configured to remove ourselves
2067          * once we're disconnected.
2068          */
2069         if (refcount_read(&nbd->config_refs))
2070                 connected = 1;
2071         dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2072         if (!dev_opt)
2073                 return -EMSGSIZE;
2074         ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2075         if (ret)
2076                 return -EMSGSIZE;
2077         ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2078                          connected);
2079         if (ret)
2080                 return -EMSGSIZE;
2081         nla_nest_end(reply, dev_opt);
2082         return 0;
2083 }
2084
2085 static int status_cb(int id, void *ptr, void *data)
2086 {
2087         struct nbd_device *nbd = ptr;
2088         return populate_nbd_status(nbd, (struct sk_buff *)data);
2089 }
2090
2091 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2092 {
2093         struct nlattr *dev_list;
2094         struct sk_buff *reply;
2095         void *reply_head;
2096         size_t msg_size;
2097         int index = -1;
2098         int ret = -ENOMEM;
2099
2100         if (info->attrs[NBD_ATTR_INDEX])
2101                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2102
2103         mutex_lock(&nbd_index_mutex);
2104
2105         msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2106                                   nla_attr_size(sizeof(u8)));
2107         msg_size *= (index == -1) ? nbd_total_devices : 1;
2108
2109         reply = genlmsg_new(msg_size, GFP_KERNEL);
2110         if (!reply)
2111                 goto out;
2112         reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2113                                        NBD_CMD_STATUS);
2114         if (!reply_head) {
2115                 nlmsg_free(reply);
2116                 goto out;
2117         }
2118
2119         dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2120         if (index == -1) {
2121                 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2122                 if (ret) {
2123                         nlmsg_free(reply);
2124                         goto out;
2125                 }
2126         } else {
2127                 struct nbd_device *nbd;
2128                 nbd = idr_find(&nbd_index_idr, index);
2129                 if (nbd) {
2130                         ret = populate_nbd_status(nbd, reply);
2131                         if (ret) {
2132                                 nlmsg_free(reply);
2133                                 goto out;
2134                         }
2135                 }
2136         }
2137         nla_nest_end(reply, dev_list);
2138         genlmsg_end(reply, reply_head);
2139         ret = genlmsg_reply(reply, info);
2140 out:
2141         mutex_unlock(&nbd_index_mutex);
2142         return ret;
2143 }
2144
2145 static void nbd_connect_reply(struct genl_info *info, int index)
2146 {
2147         struct sk_buff *skb;
2148         void *msg_head;
2149         int ret;
2150
2151         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2152         if (!skb)
2153                 return;
2154         msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2155                                      NBD_CMD_CONNECT);
2156         if (!msg_head) {
2157                 nlmsg_free(skb);
2158                 return;
2159         }
2160         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2161         if (ret) {
2162                 nlmsg_free(skb);
2163                 return;
2164         }
2165         genlmsg_end(skb, msg_head);
2166         genlmsg_reply(skb, info);
2167 }
2168
2169 static void nbd_mcast_index(int index)
2170 {
2171         struct sk_buff *skb;
2172         void *msg_head;
2173         int ret;
2174
2175         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2176         if (!skb)
2177                 return;
2178         msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2179                                      NBD_CMD_LINK_DEAD);
2180         if (!msg_head) {
2181                 nlmsg_free(skb);
2182                 return;
2183         }
2184         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2185         if (ret) {
2186                 nlmsg_free(skb);
2187                 return;
2188         }
2189         genlmsg_end(skb, msg_head);
2190         genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2191 }
2192
2193 static void nbd_dead_link_work(struct work_struct *work)
2194 {
2195         struct link_dead_args *args = container_of(work, struct link_dead_args,
2196                                                    work);
2197         nbd_mcast_index(args->index);
2198         kfree(args);
2199 }
2200
2201 static int __init nbd_init(void)
2202 {
2203         int i;
2204
2205         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2206
2207         if (max_part < 0) {
2208                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2209                 return -EINVAL;
2210         }
2211
2212         part_shift = 0;
2213         if (max_part > 0) {
2214                 part_shift = fls(max_part);
2215
2216                 /*
2217                  * Adjust max_part according to part_shift as it is exported
2218                  * to user space so that user can know the max number of
2219                  * partition kernel should be able to manage.
2220                  *
2221                  * Note that -1 is required because partition 0 is reserved
2222                  * for the whole disk.
2223                  */
2224                 max_part = (1UL << part_shift) - 1;
2225         }
2226
2227         if ((1UL << part_shift) > DISK_MAX_PARTS)
2228                 return -EINVAL;
2229
2230         if (nbds_max > 1UL << (MINORBITS - part_shift))
2231                 return -EINVAL;
2232         recv_workqueue = alloc_workqueue("knbd-recv",
2233                                          WQ_MEM_RECLAIM | WQ_HIGHPRI |
2234                                          WQ_UNBOUND, 0);
2235         if (!recv_workqueue)
2236                 return -ENOMEM;
2237
2238         if (register_blkdev(NBD_MAJOR, "nbd")) {
2239                 destroy_workqueue(recv_workqueue);
2240                 return -EIO;
2241         }
2242
2243         if (genl_register_family(&nbd_genl_family)) {
2244                 unregister_blkdev(NBD_MAJOR, "nbd");
2245                 destroy_workqueue(recv_workqueue);
2246                 return -EINVAL;
2247         }
2248         nbd_dbg_init();
2249
2250         mutex_lock(&nbd_index_mutex);
2251         for (i = 0; i < nbds_max; i++)
2252                 nbd_dev_add(i);
2253         mutex_unlock(&nbd_index_mutex);
2254         return 0;
2255 }
2256
2257 static int nbd_exit_cb(int id, void *ptr, void *data)
2258 {
2259         struct list_head *list = (struct list_head *)data;
2260         struct nbd_device *nbd = ptr;
2261
2262         list_add_tail(&nbd->list, list);
2263         return 0;
2264 }
2265
2266 static void __exit nbd_cleanup(void)
2267 {
2268         struct nbd_device *nbd;
2269         LIST_HEAD(del_list);
2270
2271         nbd_dbg_close();
2272
2273         mutex_lock(&nbd_index_mutex);
2274         idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2275         mutex_unlock(&nbd_index_mutex);
2276
2277         while (!list_empty(&del_list)) {
2278                 nbd = list_first_entry(&del_list, struct nbd_device, list);
2279                 list_del_init(&nbd->list);
2280                 if (refcount_read(&nbd->refs) != 1)
2281                         printk(KERN_ERR "nbd: possibly leaking a device\n");
2282                 nbd_put(nbd);
2283         }
2284
2285         idr_destroy(&nbd_index_idr);
2286         genl_unregister_family(&nbd_genl_family);
2287         destroy_workqueue(recv_workqueue);
2288         unregister_blkdev(NBD_MAJOR, "nbd");
2289 }
2290
2291 module_init(nbd_init);
2292 module_exit(nbd_cleanup);
2293
2294 MODULE_DESCRIPTION("Network Block Device");
2295 MODULE_LICENSE("GPL");
2296
2297 module_param(nbds_max, int, 0444);
2298 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2299 module_param(max_part, int, 0444);
2300 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");