drivers/block/nbd.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Network block device - make block devices work over TCP
   4  *
   5  * Note that you can not swap over this thing, yet. Seems to work but
   6  * deadlocks sometimes - you can not swap over TCP in general.
   7  *
   8  * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
   9  * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
  10  *
  11  * (part of code stolen from loop.c)
  12  */
  13
  14 #include <linux/major.h>
  15
  16 #include <linux/blkdev.h>
  17 #include <linux/module.h>
  18 #include <linux/init.h>
  19 #include <linux/sched.h>
  20 #include <linux/sched/mm.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 #include <linux/nbd-netlink.h>
  44 #include <net/genetlink.h>
  45
  46 #define CREATE_TRACE_POINTS
  47 #include <trace/events/nbd.h>
  48
  49 static DEFINE_IDR(nbd_index_idr);
  50 static DEFINE_MUTEX(nbd_index_mutex);
  51 static int nbd_total_devices = 0;
  52
  53 struct nbd_sock {
  54         struct socket *sock;
  55         struct mutex tx_lock;
  56         struct request *pending;
  57         int sent;
  58         bool dead;
  59         int fallback_index;
  60         int cookie;
  61 };
  62
  63 struct recv_thread_args {
  64         struct work_struct work;
  65         struct nbd_device *nbd;
  66         int index;
  67 };
  68
  69 struct link_dead_args {
  70         struct work_struct work;
  71         int index;
  72 };
  73
  74 #define NBD_TIMEDOUT                    0
  75 #define NBD_DISCONNECT_REQUESTED        1
  76 #define NBD_DISCONNECTED                2
  77 #define NBD_HAS_PID_FILE                3
  78 #define NBD_HAS_CONFIG_REF              4
  79 #define NBD_BOUND                       5
  80 #define NBD_DESTROY_ON_DISCONNECT       6
  81 #define NBD_DISCONNECT_ON_CLOSE         7
  82
  83 struct nbd_config {
  84         u32 flags;
  85         unsigned long runtime_flags;
  86         u64 dead_conn_timeout;
  87
  88         struct nbd_sock **socks;
  89         int num_connections;
  90         atomic_t live_connections;
  91         wait_queue_head_t conn_wait;
  92
  93         atomic_t recv_threads;
  94         wait_queue_head_t recv_wq;
  95         loff_t blksize;
  96         loff_t bytesize;
  97 #if IS_ENABLED(CONFIG_DEBUG_FS)
  98         struct dentry *dbg_dir;
  99 #endif
 100 };
 101
 102 struct nbd_device {
 103         struct blk_mq_tag_set tag_set;
 104
 105         int index;
 106         refcount_t config_refs;
 107         refcount_t refs;
 108         struct nbd_config *config;
 109         struct mutex config_lock;
 110         struct gendisk *disk;
 111
 112         struct list_head list;
 113         struct task_struct *task_recv;
 114         struct task_struct *task_setup;
 115 };
 116
 117 #define NBD_CMD_REQUEUED        1
 118
 119 struct nbd_cmd {
 120         struct nbd_device *nbd;
 121         struct mutex lock;
 122         int index;
 123         int cookie;
 124         blk_status_t status;
 125         unsigned long flags;
 126         u32 cmd_cookie;
 127 };
 128
 129 #if IS_ENABLED(CONFIG_DEBUG_FS)
 130 static struct dentry *nbd_dbg_dir;
 131 #endif
 132
 133 #define nbd_name(nbd) ((nbd)->disk->disk_name)
 134
 135 #define NBD_MAGIC 0x68797548
 136
 137 static unsigned int nbds_max = 16;
 138 static int max_part = 16;
 139 static struct workqueue_struct *recv_workqueue;
 140 static int part_shift;
 141
 142 static int nbd_dev_dbg_init(struct nbd_device *nbd);
 143 static void nbd_dev_dbg_close(struct nbd_device *nbd);
 144 static void nbd_config_put(struct nbd_device *nbd);
 145 static void nbd_connect_reply(struct genl_info *info, int index);
 146 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
 147 static void nbd_dead_link_work(struct work_struct *work);
 148 static void nbd_disconnect_and_put(struct nbd_device *nbd);
 149
 150 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
 151 {
 152         return disk_to_dev(nbd->disk);
 153 }
 154
 155 static void nbd_requeue_cmd(struct nbd_cmd *cmd)
 156 {
 157         struct request *req = blk_mq_rq_from_pdu(cmd);
 158
 159         if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
 160                 blk_mq_requeue_request(req, true);
 161 }
 162
 163 #define NBD_COOKIE_BITS 32
 164
 165 static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
 166 {
 167         struct request *req = blk_mq_rq_from_pdu(cmd);
 168         u32 tag = blk_mq_unique_tag(req);
 169         u64 cookie = cmd->cmd_cookie;
 170
 171         return (cookie << NBD_COOKIE_BITS) | tag;
 172 }
 173
 174 static u32 nbd_handle_to_tag(u64 handle)
 175 {
 176         return (u32)handle;
 177 }
 178
 179 static u32 nbd_handle_to_cookie(u64 handle)
 180 {
 181         return (u32)(handle >> NBD_COOKIE_BITS);
 182 }
 183
 184 static const char *nbdcmd_to_ascii(int cmd)
 185 {
 186         switch (cmd) {
 187         case  NBD_CMD_READ: return "read";
 188         case NBD_CMD_WRITE: return "write";
 189         case  NBD_CMD_DISC: return "disconnect";
 190         case NBD_CMD_FLUSH: return "flush";
 191         case  NBD_CMD_TRIM: return "trim/discard";
 192         }
 193         return "invalid";
 194 }
 195
 196 static ssize_t pid_show(struct device *dev,
 197                         struct device_attribute *attr, char *buf)
 198 {
 199         struct gendisk *disk = dev_to_disk(dev);
 200         struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
 201
 202         return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
 203 }
 204
 205 static const struct device_attribute pid_attr = {
 206         .attr = { .name = "pid", .mode = 0444},
 207         .show = pid_show,
 208 };
 209
 210 static void nbd_dev_remove(struct nbd_device *nbd)
 211 {
 212         struct gendisk *disk = nbd->disk;
 213         struct request_queue *q;
 214
 215         if (disk) {
 216                 q = disk->queue;
 217                 del_gendisk(disk);
 218                 blk_cleanup_queue(q);
 219                 blk_mq_free_tag_set(&nbd->tag_set);
 220                 disk->private_data = NULL;
 221                 put_disk(disk);
 222         }
 223         kfree(nbd);
 224 }
 225
 226 static void nbd_put(struct nbd_device *nbd)
 227 {
 228         if (refcount_dec_and_mutex_lock(&nbd->refs,
 229                                         &nbd_index_mutex)) {
 230                 idr_remove(&nbd_index_idr, nbd->index);
 231                 mutex_unlock(&nbd_index_mutex);
 232                 nbd_dev_remove(nbd);
 233         }
 234 }
 235
 236 static int nbd_disconnected(struct nbd_config *config)
 237 {
 238         return test_bit(NBD_DISCONNECTED, &config->runtime_flags) ||
 239                 test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
 240 }
 241
 242 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
 243                                 int notify)
 244 {
 245         if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
 246                 struct link_dead_args *args;
 247                 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
 248                 if (args) {
 249                         INIT_WORK(&args->work, nbd_dead_link_work);
 250                         args->index = nbd->index;
 251                         queue_work(system_wq, &args->work);
 252                 }
 253         }
 254         if (!nsock->dead) {
 255                 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
 256                 if (atomic_dec_return(&nbd->config->live_connections) == 0) {
 257                         if (test_and_clear_bit(NBD_DISCONNECT_REQUESTED,
 258                                                &nbd->config->runtime_flags)) {
 259                                 set_bit(NBD_DISCONNECTED,
 260                                         &nbd->config->runtime_flags);
 261                                 dev_info(nbd_to_dev(nbd),
 262                                         "Disconnected due to user request.\n");
 263                         }
 264                 }
 265         }
 266         nsock->dead = true;
 267         nsock->pending = NULL;
 268         nsock->sent = 0;
 269 }
 270
 271 static void nbd_size_clear(struct nbd_device *nbd)
 272 {
 273         if (nbd->config->bytesize) {
 274                 set_capacity(nbd->disk, 0);
 275                 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 276         }
 277 }
 278
 279 static void nbd_size_update(struct nbd_device *nbd)
 280 {
 281         struct nbd_config *config = nbd->config;
 282         struct block_device *bdev = bdget_disk(nbd->disk, 0);
 283
 284         if (config->flags & NBD_FLAG_SEND_TRIM) {
 285                 nbd->disk->queue->limits.discard_granularity = config->blksize;
 286                 nbd->disk->queue->limits.discard_alignment = config->blksize;
 287                 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
 288         }
 289         blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
 290         blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
 291         set_capacity(nbd->disk, config->bytesize >> 9);
 292         if (bdev) {
 293                 if (bdev->bd_disk) {
 294                         bd_set_size(bdev, config->bytesize);
 295                         set_blocksize(bdev, config->blksize);
 296                 } else
 297                         bdev->bd_invalidated = 1;
 298                 bdput(bdev);
 299         }
 300         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 301 }
 302
 303 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
 304                          loff_t nr_blocks)
 305 {
 306         struct nbd_config *config = nbd->config;
 307         config->blksize = blocksize;
 308         config->bytesize = blocksize * nr_blocks;
 309         if (nbd->task_recv != NULL)
 310                 nbd_size_update(nbd);
 311 }
 312
 313 static void nbd_complete_rq(struct request *req)
 314 {
 315         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 316
 317         dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
 318                 cmd->status ? "failed" : "done");
 319
 320         blk_mq_end_request(req, cmd->status);
 321 }
 322
 323 /*
 324  * Forcibly shutdown the socket causing all listeners to error
 325  */
 326 static void sock_shutdown(struct nbd_device *nbd)
 327 {
 328         struct nbd_config *config = nbd->config;
 329         int i;
 330
 331         if (config->num_connections == 0)
 332                 return;
 333         if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
 334                 return;
 335
 336         for (i = 0; i < config->num_connections; i++) {
 337                 struct nbd_sock *nsock = config->socks[i];
 338                 mutex_lock(&nsock->tx_lock);
 339                 nbd_mark_nsock_dead(nbd, nsock, 0);
 340                 mutex_unlock(&nsock->tx_lock);
 341         }
 342         dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
 343 }
 344
 345 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 346                                                  bool reserved)
 347 {
 348         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 349         struct nbd_device *nbd = cmd->nbd;
 350         struct nbd_config *config;
 351
 352         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 353                 cmd->status = BLK_STS_TIMEOUT;
 354                 goto done;
 355         }
 356         config = nbd->config;
 357
 358         if (!mutex_trylock(&cmd->lock))
 359                 return BLK_EH_RESET_TIMER;
 360
 361         if (config->num_connections > 1) {
 362                 dev_err_ratelimited(nbd_to_dev(nbd),
 363                                     "Connection timed out, retrying (%d/%d alive)\n",
 364                                     atomic_read(&config->live_connections),
 365                                     config->num_connections);
 366                 /*
 367                  * Hooray we have more connections, requeue this IO, the submit
 368                  * path will put it on a real connection.
 369                  */
 370                 if (config->socks && config->num_connections > 1) {
 371                         if (cmd->index < config->num_connections) {
 372                                 struct nbd_sock *nsock =
 373                                         config->socks[cmd->index];
 374                                 mutex_lock(&nsock->tx_lock);
 375                                 /* We can have multiple outstanding requests, so
 376                                  * we don't want to mark the nsock dead if we've
 377                                  * already reconnected with a new socket, so
 378                                  * only mark it dead if its the same socket we
 379                                  * were sent out on.
 380                                  */
 381                                 if (cmd->cookie == nsock->cookie)
 382                                         nbd_mark_nsock_dead(nbd, nsock, 1);
 383                                 mutex_unlock(&nsock->tx_lock);
 384                         }
 385                         mutex_unlock(&cmd->lock);
 386                         nbd_requeue_cmd(cmd);
 387                         nbd_config_put(nbd);
 388                         return BLK_EH_DONE;
 389                 }
 390         } else {
 391                 dev_err_ratelimited(nbd_to_dev(nbd),
 392                                     "Connection timed out\n");
 393         }
 394         set_bit(NBD_TIMEDOUT, &config->runtime_flags);
 395         cmd->status = BLK_STS_IOERR;
 396         mutex_unlock(&cmd->lock);
 397         sock_shutdown(nbd);
 398         nbd_config_put(nbd);
 399 done:
 400         blk_mq_complete_request(req);
 401         return BLK_EH_DONE;
 402 }
 403
 404 /*
 405  *  Send or receive packet.
 406  */
 407 static int sock_xmit(struct nbd_device *nbd, int index, int send,
 408                      struct iov_iter *iter, int msg_flags, int *sent)
 409 {
 410         struct nbd_config *config = nbd->config;
 411         struct socket *sock = config->socks[index]->sock;
 412         int result;
 413         struct msghdr msg;
 414         unsigned int noreclaim_flag;
 415
 416         if (unlikely(!sock)) {
 417                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 418                         "Attempted %s on closed socket in sock_xmit\n",
 419                         (send ? "send" : "recv"));
 420                 return -EINVAL;
 421         }
 422
 423         msg.msg_iter = *iter;
 424
 425         noreclaim_flag = memalloc_noreclaim_save();
 426         do {
 427                 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
 428                 msg.msg_name = NULL;
 429                 msg.msg_namelen = 0;
 430                 msg.msg_control = NULL;
 431                 msg.msg_controllen = 0;
 432                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 433
 434                 if (send)
 435                         result = sock_sendmsg(sock, &msg);
 436                 else
 437                         result = sock_recvmsg(sock, &msg, msg.msg_flags);
 438
 439                 if (result <= 0) {
 440                         if (result == 0)
 441                                 result = -EPIPE; /* short read */
 442                         break;
 443                 }
 444                 if (sent)
 445                         *sent += result;
 446         } while (msg_data_left(&msg));
 447
 448         memalloc_noreclaim_restore(noreclaim_flag);
 449
 450         return result;
 451 }
 452
 453 /*
 454  * Different settings for sk->sk_sndtimeo can result in different return values
 455  * if there is a signal pending when we enter sendmsg, because reasons?
 456  */
 457 static inline int was_interrupted(int result)
 458 {
 459         return result == -ERESTARTSYS || result == -EINTR;
 460 }
 461
 462 /* always call with the tx_lock held */
 463 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 464 {
 465         struct request *req = blk_mq_rq_from_pdu(cmd);
 466         struct nbd_config *config = nbd->config;
 467         struct nbd_sock *nsock = config->socks[index];
 468         int result;
 469         struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
 470         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 471         struct iov_iter from;
 472         unsigned long size = blk_rq_bytes(req);
 473         struct bio *bio;
 474         u64 handle;
 475         u32 type;
 476         u32 nbd_cmd_flags = 0;
 477         int sent = nsock->sent, skip = 0;
 478
 479         iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
 480
 481         switch (req_op(req)) {
 482         case REQ_OP_DISCARD:
 483                 type = NBD_CMD_TRIM;
 484                 break;
 485         case REQ_OP_FLUSH:
 486                 type = NBD_CMD_FLUSH;
 487                 break;
 488         case REQ_OP_WRITE:
 489                 type = NBD_CMD_WRITE;
 490                 break;
 491         case REQ_OP_READ:
 492                 type = NBD_CMD_READ;
 493                 break;
 494         default:
 495                 return -EIO;
 496         }
 497
 498         if (rq_data_dir(req) == WRITE &&
 499             (config->flags & NBD_FLAG_READ_ONLY)) {
 500                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 501                                     "Write on read-only\n");
 502                 return -EIO;
 503         }
 504
 505         if (req->cmd_flags & REQ_FUA)
 506                 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
 507
 508         /* We did a partial send previously, and we at least sent the whole
 509          * request struct, so just go and send the rest of the pages in the
 510          * request.
 511          */
 512         if (sent) {
 513                 if (sent >= sizeof(request)) {
 514                         skip = sent - sizeof(request);
 515
 516                         /* initialize handle for tracing purposes */
 517                         handle = nbd_cmd_handle(cmd);
 518
 519                         goto send_pages;
 520                 }
 521                 iov_iter_advance(&from, sent);
 522         } else {
 523                 cmd->cmd_cookie++;
 524         }
 525         cmd->index = index;
 526         cmd->cookie = nsock->cookie;
 527         request.type = htonl(type | nbd_cmd_flags);
 528         if (type != NBD_CMD_FLUSH) {
 529                 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
 530                 request.len = htonl(size);
 531         }
 532         handle = nbd_cmd_handle(cmd);
 533         memcpy(request.handle, &handle, sizeof(handle));
 534
 535         trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
 536
 537         dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
 538                 req, nbdcmd_to_ascii(type),
 539                 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
 540         result = sock_xmit(nbd, index, 1, &from,
 541                         (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
 542         trace_nbd_header_sent(req, handle);
 543         if (result <= 0) {
 544                 if (was_interrupted(result)) {
 545                         /* If we havne't sent anything we can just return BUSY,
 546                          * however if we have sent something we need to make
 547                          * sure we only allow this req to be sent until we are
 548                          * completely done.
 549                          */
 550                         if (sent) {
 551                                 nsock->pending = req;
 552                                 nsock->sent = sent;
 553                         }
 554                         set_bit(NBD_CMD_REQUEUED, &cmd->flags);
 555                         return BLK_STS_RESOURCE;
 556                 }
 557                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 558                         "Send control failed (result %d)\n", result);
 559                 return -EAGAIN;
 560         }
 561 send_pages:
 562         if (type != NBD_CMD_WRITE)
 563                 goto out;
 564
 565         bio = req->bio;
 566         while (bio) {
 567                 struct bio *next = bio->bi_next;
 568                 struct bvec_iter iter;
 569                 struct bio_vec bvec;
 570
 571                 bio_for_each_segment(bvec, bio, iter) {
 572                         bool is_last = !next && bio_iter_last(bvec, iter);
 573                         int flags = is_last ? 0 : MSG_MORE;
 574
 575                         dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
 576                                 req, bvec.bv_len);
 577                         iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
 578                         if (skip) {
 579                                 if (skip >= iov_iter_count(&from)) {
 580                                         skip -= iov_iter_count(&from);
 581                                         continue;
 582                                 }
 583                                 iov_iter_advance(&from, skip);
 584                                 skip = 0;
 585                         }
 586                         result = sock_xmit(nbd, index, 1, &from, flags, &sent);
 587                         if (result <= 0) {
 588                                 if (was_interrupted(result)) {
 589                                         /* We've already sent the header, we
 590                                          * have no choice but to set pending and
 591                                          * return BUSY.
 592                                          */
 593                                         nsock->pending = req;
 594                                         nsock->sent = sent;
 595                                         set_bit(NBD_CMD_REQUEUED, &cmd->flags);
 596                                         return BLK_STS_RESOURCE;
 597                                 }
 598                                 dev_err(disk_to_dev(nbd->disk),
 599                                         "Send data failed (result %d)\n",
 600                                         result);
 601                                 return -EAGAIN;
 602                         }
 603                         /*
 604                          * The completion might already have come in,
 605                          * so break for the last one instead of letting
 606                          * the iterator do it. This prevents use-after-free
 607                          * of the bio.
 608                          */
 609                         if (is_last)
 610                                 break;
 611                 }
 612                 bio = next;
 613         }
 614 out:
 615         trace_nbd_payload_sent(req, handle);
 616         nsock->pending = NULL;
 617         nsock->sent = 0;
 618         return 0;
 619 }
 620
 621 /* NULL returned = something went wrong, inform userspace */
 622 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 623 {
 624         struct nbd_config *config = nbd->config;
 625         int result;
 626         struct nbd_reply reply;
 627         struct nbd_cmd *cmd;
 628         struct request *req = NULL;
 629         u64 handle;
 630         u16 hwq;
 631         u32 tag;
 632         struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
 633         struct iov_iter to;
 634         int ret = 0;
 635
 636         reply.magic = 0;
 637         iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
 638         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 639         if (result <= 0) {
 640                 if (!nbd_disconnected(config))
 641                         dev_err(disk_to_dev(nbd->disk),
 642                                 "Receive control failed (result %d)\n", result);
 643                 return ERR_PTR(result);
 644         }
 645
 646         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 647                 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
 648                                 (unsigned long)ntohl(reply.magic));
 649                 return ERR_PTR(-EPROTO);
 650         }
 651
 652         memcpy(&handle, reply.handle, sizeof(handle));
 653         tag = nbd_handle_to_tag(handle);
 654         hwq = blk_mq_unique_tag_to_hwq(tag);
 655         if (hwq < nbd->tag_set.nr_hw_queues)
 656                 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
 657                                        blk_mq_unique_tag_to_tag(tag));
 658         if (!req || !blk_mq_request_started(req)) {
 659                 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
 660                         tag, req);
 661                 return ERR_PTR(-ENOENT);
 662         }
 663         trace_nbd_header_received(req, handle);
 664         cmd = blk_mq_rq_to_pdu(req);
 665
 666         mutex_lock(&cmd->lock);
 667         if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
 668                 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
 669                         req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
 670                 ret = -ENOENT;
 671                 goto out;
 672         }
 673         if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
 674                 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
 675                         req);
 676                 ret = -ENOENT;
 677                 goto out;
 678         }
 679         if (ntohl(reply.error)) {
 680                 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
 681                         ntohl(reply.error));
 682                 cmd->status = BLK_STS_IOERR;
 683                 goto out;
 684         }
 685
 686         dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
 687         if (rq_data_dir(req) != WRITE) {
 688                 struct req_iterator iter;
 689                 struct bio_vec bvec;
 690
 691                 rq_for_each_segment(bvec, req, iter) {
 692                         iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
 693                         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 694                         if (result <= 0) {
 695                                 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 696                                         result);
 697                                 /*
 698                                  * If we've disconnected or we only have 1
 699                                  * connection then we need to make sure we
 700                                  * complete this request, otherwise error out
 701                                  * and let the timeout stuff handle resubmitting
 702                                  * this request onto another connection.
 703                                  */
 704                                 if (nbd_disconnected(config) ||
 705                                     config->num_connections <= 1) {
 706                                         cmd->status = BLK_STS_IOERR;
 707                                         goto out;
 708                                 }
 709                                 ret = -EIO;
 710                                 goto out;
 711                         }
 712                         dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 713                                 req, bvec.bv_len);
 714                 }
 715         }
 716 out:
 717         trace_nbd_payload_received(req, handle);
 718         mutex_unlock(&cmd->lock);
 719         return ret ? ERR_PTR(ret) : cmd;
 720 }
 721
 722 static void recv_work(struct work_struct *work)
 723 {
 724         struct recv_thread_args *args = container_of(work,
 725                                                      struct recv_thread_args,
 726                                                      work);
 727         struct nbd_device *nbd = args->nbd;
 728         struct nbd_config *config = nbd->config;
 729         struct nbd_cmd *cmd;
 730
 731         while (1) {
 732                 cmd = nbd_read_stat(nbd, args->index);
 733                 if (IS_ERR(cmd)) {
 734                         struct nbd_sock *nsock = config->socks[args->index];
 735
 736                         mutex_lock(&nsock->tx_lock);
 737                         nbd_mark_nsock_dead(nbd, nsock, 1);
 738                         mutex_unlock(&nsock->tx_lock);
 739                         break;
 740                 }
 741
 742                 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
 743         }
 744         atomic_dec(&config->recv_threads);
 745         wake_up(&config->recv_wq);
 746         nbd_config_put(nbd);
 747         kfree(args);
 748 }
 749
 750 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
 751 {
 752         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 753
 754         cmd->status = BLK_STS_IOERR;
 755         blk_mq_complete_request(req);
 756         return true;
 757 }
 758
 759 static void nbd_clear_que(struct nbd_device *nbd)
 760 {
 761         blk_mq_quiesce_queue(nbd->disk->queue);
 762         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 763         blk_mq_unquiesce_queue(nbd->disk->queue);
 764         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 765 }
 766
 767 static int find_fallback(struct nbd_device *nbd, int index)
 768 {
 769         struct nbd_config *config = nbd->config;
 770         int new_index = -1;
 771         struct nbd_sock *nsock = config->socks[index];
 772         int fallback = nsock->fallback_index;
 773
 774         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 775                 return new_index;
 776
 777         if (config->num_connections <= 1) {
 778                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 779                                     "Attempted send on invalid socket\n");
 780                 return new_index;
 781         }
 782
 783         if (fallback >= 0 && fallback < config->num_connections &&
 784             !config->socks[fallback]->dead)
 785                 return fallback;
 786
 787         if (nsock->fallback_index < 0 ||
 788             nsock->fallback_index >= config->num_connections ||
 789             config->socks[nsock->fallback_index]->dead) {
 790                 int i;
 791                 for (i = 0; i < config->num_connections; i++) {
 792                         if (i == index)
 793                                 continue;
 794                         if (!config->socks[i]->dead) {
 795                                 new_index = i;
 796                                 break;
 797                         }
 798                 }
 799                 nsock->fallback_index = new_index;
 800                 if (new_index < 0) {
 801                         dev_err_ratelimited(disk_to_dev(nbd->disk),
 802                                             "Dead connection, failed to find a fallback\n");
 803                         return new_index;
 804                 }
 805         }
 806         new_index = nsock->fallback_index;
 807         return new_index;
 808 }
 809
 810 static int wait_for_reconnect(struct nbd_device *nbd)
 811 {
 812         struct nbd_config *config = nbd->config;
 813         if (!config->dead_conn_timeout)
 814                 return 0;
 815         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 816                 return 0;
 817         return wait_event_timeout(config->conn_wait,
 818                                   atomic_read(&config->live_connections) > 0,
 819                                   config->dead_conn_timeout) > 0;
 820 }
 821
 822 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 823 {
 824         struct request *req = blk_mq_rq_from_pdu(cmd);
 825         struct nbd_device *nbd = cmd->nbd;
 826         struct nbd_config *config;
 827         struct nbd_sock *nsock;
 828         int ret;
 829
 830         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 831                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 832                                     "Socks array is empty\n");
 833                 blk_mq_start_request(req);
 834                 return -EINVAL;
 835         }
 836         config = nbd->config;
 837
 838         if (index >= config->num_connections) {
 839                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 840                                     "Attempted send on invalid socket\n");
 841                 nbd_config_put(nbd);
 842                 blk_mq_start_request(req);
 843                 return -EINVAL;
 844         }
 845         cmd->status = BLK_STS_OK;
 846 again:
 847         nsock = config->socks[index];
 848         mutex_lock(&nsock->tx_lock);
 849         if (nsock->dead) {
 850                 int old_index = index;
 851                 index = find_fallback(nbd, index);
 852                 mutex_unlock(&nsock->tx_lock);
 853                 if (index < 0) {
 854                         if (wait_for_reconnect(nbd)) {
 855                                 index = old_index;
 856                                 goto again;
 857                         }
 858                         /* All the sockets should already be down at this point,
 859                          * we just want to make sure that DISCONNECTED is set so
 860                          * any requests that come in that were queue'ed waiting
 861                          * for the reconnect timer don't trigger the timer again
 862                          * and instead just error out.
 863                          */
 864                         sock_shutdown(nbd);
 865                         nbd_config_put(nbd);
 866                         blk_mq_start_request(req);
 867                         return -EIO;
 868                 }
 869                 goto again;
 870         }
 871
 872         /* Handle the case that we have a pending request that was partially
 873          * transmitted that _has_ to be serviced first.  We need to call requeue
 874          * here so that it gets put _after_ the request that is already on the
 875          * dispatch list.
 876          */
 877         blk_mq_start_request(req);
 878         if (unlikely(nsock->pending && nsock->pending != req)) {
 879                 nbd_requeue_cmd(cmd);
 880                 ret = 0;
 881                 goto out;
 882         }
 883         /*
 884          * Some failures are related to the link going down, so anything that
 885          * returns EAGAIN can be retried on a different socket.
 886          */
 887         ret = nbd_send_cmd(nbd, cmd, index);
 888         if (ret == -EAGAIN) {
 889                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 890                                     "Request send failed, requeueing\n");
 891                 nbd_mark_nsock_dead(nbd, nsock, 1);
 892                 nbd_requeue_cmd(cmd);
 893                 ret = 0;
 894         }
 895 out:
 896         mutex_unlock(&nsock->tx_lock);
 897         nbd_config_put(nbd);
 898         return ret;
 899 }
 900
 901 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 902                         const struct blk_mq_queue_data *bd)
 903 {
 904         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 905         int ret;
 906
 907         /*
 908          * Since we look at the bio's to send the request over the network we
 909          * need to make sure the completion work doesn't mark this request done
 910          * before we are done doing our send.  This keeps us from dereferencing
 911          * freed data if we have particularly fast completions (ie we get the
 912          * completion before we exit sock_xmit on the last bvec) or in the case
 913          * that the server is misbehaving (or there was an error) before we're
 914          * done sending everything over the wire.
 915          */
 916         mutex_lock(&cmd->lock);
 917         clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
 918
 919         /* We can be called directly from the user space process, which means we
 920          * could possibly have signals pending so our sendmsg will fail.  In
 921          * this case we need to return that we are busy, otherwise error out as
 922          * appropriate.
 923          */
 924         ret = nbd_handle_cmd(cmd, hctx->queue_num);
 925         if (ret < 0)
 926                 ret = BLK_STS_IOERR;
 927         else if (!ret)
 928                 ret = BLK_STS_OK;
 929         mutex_unlock(&cmd->lock);
 930
 931         return ret;
 932 }
 933
 934 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
 935                           bool netlink)
 936 {
 937         struct nbd_config *config = nbd->config;
 938         struct socket *sock;
 939         struct nbd_sock **socks;
 940         struct nbd_sock *nsock;
 941         int err;
 942
 943         sock = sockfd_lookup(arg, &err);
 944         if (!sock)
 945                 return err;
 946
 947         if (!netlink && !nbd->task_setup &&
 948             !test_bit(NBD_BOUND, &config->runtime_flags))
 949                 nbd->task_setup = current;
 950
 951         if (!netlink &&
 952             (nbd->task_setup != current ||
 953              test_bit(NBD_BOUND, &config->runtime_flags))) {
 954                 dev_err(disk_to_dev(nbd->disk),
 955                         "Device being setup by another task");
 956                 sockfd_put(sock);
 957                 return -EBUSY;
 958         }
 959
 960         socks = krealloc(config->socks, (config->num_connections + 1) *
 961                          sizeof(struct nbd_sock *), GFP_KERNEL);
 962         if (!socks) {
 963                 sockfd_put(sock);
 964                 return -ENOMEM;
 965         }
 966         nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
 967         if (!nsock) {
 968                 sockfd_put(sock);
 969                 return -ENOMEM;
 970         }
 971
 972         config->socks = socks;
 973
 974         nsock->fallback_index = -1;
 975         nsock->dead = false;
 976         mutex_init(&nsock->tx_lock);
 977         nsock->sock = sock;
 978         nsock->pending = NULL;
 979         nsock->sent = 0;
 980         nsock->cookie = 0;
 981         socks[config->num_connections++] = nsock;
 982         atomic_inc(&config->live_connections);
 983
 984         return 0;
 985 }
 986
 987 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
 988 {
 989         struct nbd_config *config = nbd->config;
 990         struct socket *sock, *old;
 991         struct recv_thread_args *args;
 992         int i;
 993         int err;
 994
 995         sock = sockfd_lookup(arg, &err);
 996         if (!sock)
 997                 return err;
 998
 999         args = kzalloc(sizeof(*args), GFP_KERNEL);
1000         if (!args) {
1001                 sockfd_put(sock);
1002                 return -ENOMEM;
1003         }
1004
1005         for (i = 0; i < config->num_connections; i++) {
1006                 struct nbd_sock *nsock = config->socks[i];
1007
1008                 if (!nsock->dead)
1009                         continue;
1010
1011                 mutex_lock(&nsock->tx_lock);
1012                 if (!nsock->dead) {
1013                         mutex_unlock(&nsock->tx_lock);
1014                         continue;
1015                 }
1016                 sk_set_memalloc(sock->sk);
1017                 if (nbd->tag_set.timeout)
1018                         sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1019                 atomic_inc(&config->recv_threads);
1020                 refcount_inc(&nbd->config_refs);
1021                 old = nsock->sock;
1022                 nsock->fallback_index = -1;
1023                 nsock->sock = sock;
1024                 nsock->dead = false;
1025                 INIT_WORK(&args->work, recv_work);
1026                 args->index = i;
1027                 args->nbd = nbd;
1028                 nsock->cookie++;
1029                 mutex_unlock(&nsock->tx_lock);
1030                 sockfd_put(old);
1031
1032                 clear_bit(NBD_DISCONNECTED, &config->runtime_flags);
1033
1034                 /* We take the tx_mutex in an error path in the recv_work, so we
1035                  * need to queue_work outside of the tx_mutex.
1036                  */
1037                 queue_work(recv_workqueue, &args->work);
1038
1039                 atomic_inc(&config->live_connections);
1040                 wake_up(&config->conn_wait);
1041                 return 0;
1042         }
1043         sockfd_put(sock);
1044         kfree(args);
1045         return -ENOSPC;
1046 }
1047
1048 static void nbd_bdev_reset(struct block_device *bdev)
1049 {
1050         if (bdev->bd_openers > 1)
1051                 return;
1052         bd_set_size(bdev, 0);
1053 }
1054
1055 static void nbd_parse_flags(struct nbd_device *nbd)
1056 {
1057         struct nbd_config *config = nbd->config;
1058         if (config->flags & NBD_FLAG_READ_ONLY)
1059                 set_disk_ro(nbd->disk, true);
1060         else
1061                 set_disk_ro(nbd->disk, false);
1062         if (config->flags & NBD_FLAG_SEND_TRIM)
1063                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1064         if (config->flags & NBD_FLAG_SEND_FLUSH) {
1065                 if (config->flags & NBD_FLAG_SEND_FUA)
1066                         blk_queue_write_cache(nbd->disk->queue, true, true);
1067                 else
1068                         blk_queue_write_cache(nbd->disk->queue, true, false);
1069         }
1070         else
1071                 blk_queue_write_cache(nbd->disk->queue, false, false);
1072 }
1073
1074 static void send_disconnects(struct nbd_device *nbd)
1075 {
1076         struct nbd_config *config = nbd->config;
1077         struct nbd_request request = {
1078                 .magic = htonl(NBD_REQUEST_MAGIC),
1079                 .type = htonl(NBD_CMD_DISC),
1080         };
1081         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1082         struct iov_iter from;
1083         int i, ret;
1084
1085         for (i = 0; i < config->num_connections; i++) {
1086                 struct nbd_sock *nsock = config->socks[i];
1087
1088                 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1089                 mutex_lock(&nsock->tx_lock);
1090                 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1091                 if (ret <= 0)
1092                         dev_err(disk_to_dev(nbd->disk),
1093                                 "Send disconnect failed %d\n", ret);
1094                 mutex_unlock(&nsock->tx_lock);
1095         }
1096 }
1097
1098 static int nbd_disconnect(struct nbd_device *nbd)
1099 {
1100         struct nbd_config *config = nbd->config;
1101
1102         dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1103         set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
1104         send_disconnects(nbd);
1105         return 0;
1106 }
1107
1108 static void nbd_clear_sock(struct nbd_device *nbd)
1109 {
1110         sock_shutdown(nbd);
1111         nbd_clear_que(nbd);
1112         nbd->task_setup = NULL;
1113 }
1114
1115 static void nbd_config_put(struct nbd_device *nbd)
1116 {
1117         if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1118                                         &nbd->config_lock)) {
1119                 struct nbd_config *config = nbd->config;
1120                 nbd_dev_dbg_close(nbd);
1121                 nbd_size_clear(nbd);
1122                 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1123                                        &config->runtime_flags))
1124                         device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1125                 nbd->task_recv = NULL;
1126                 nbd_clear_sock(nbd);
1127                 if (config->num_connections) {
1128                         int i;
1129                         for (i = 0; i < config->num_connections; i++) {
1130                                 sockfd_put(config->socks[i]->sock);
1131                                 kfree(config->socks[i]);
1132                         }
1133                         kfree(config->socks);
1134                 }
1135                 kfree(nbd->config);
1136                 nbd->config = NULL;
1137
1138                 nbd->tag_set.timeout = 0;
1139                 nbd->disk->queue->limits.discard_granularity = 0;
1140                 nbd->disk->queue->limits.discard_alignment = 0;
1141                 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1142                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1143
1144                 mutex_unlock(&nbd->config_lock);
1145                 nbd_put(nbd);
1146                 module_put(THIS_MODULE);
1147         }
1148 }
1149
1150 static int nbd_start_device(struct nbd_device *nbd)
1151 {
1152         struct nbd_config *config = nbd->config;
1153         int num_connections = config->num_connections;
1154         int error = 0, i;
1155
1156         if (nbd->task_recv)
1157                 return -EBUSY;
1158         if (!config->socks)
1159                 return -EINVAL;
1160         if (num_connections > 1 &&
1161             !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1162                 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1163                 return -EINVAL;
1164         }
1165
1166         blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1167         nbd->task_recv = current;
1168
1169         nbd_parse_flags(nbd);
1170
1171         error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1172         if (error) {
1173                 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1174                 return error;
1175         }
1176         set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1177
1178         nbd_dev_dbg_init(nbd);
1179         for (i = 0; i < num_connections; i++) {
1180                 struct recv_thread_args *args;
1181
1182                 args = kzalloc(sizeof(*args), GFP_KERNEL);
1183                 if (!args) {
1184                         sock_shutdown(nbd);
1185                         return -ENOMEM;
1186                 }
1187                 sk_set_memalloc(config->socks[i]->sock->sk);
1188                 if (nbd->tag_set.timeout)
1189                         config->socks[i]->sock->sk->sk_sndtimeo =
1190                                 nbd->tag_set.timeout;
1191                 atomic_inc(&config->recv_threads);
1192                 refcount_inc(&nbd->config_refs);
1193                 INIT_WORK(&args->work, recv_work);
1194                 args->nbd = nbd;
1195                 args->index = i;
1196                 queue_work(recv_workqueue, &args->work);
1197         }
1198         nbd_size_update(nbd);
1199         return error;
1200 }
1201
1202 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1203 {
1204         struct nbd_config *config = nbd->config;
1205         int ret;
1206
1207         ret = nbd_start_device(nbd);
1208         if (ret)
1209                 return ret;
1210
1211         if (max_part)
1212                 bdev->bd_invalidated = 1;
1213         mutex_unlock(&nbd->config_lock);
1214         ret = wait_event_interruptible(config->recv_wq,
1215                                          atomic_read(&config->recv_threads) == 0);
1216         if (ret)
1217                 sock_shutdown(nbd);
1218         mutex_lock(&nbd->config_lock);
1219         nbd_bdev_reset(bdev);
1220         /* user requested, ignore socket errors */
1221         if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1222                 ret = 0;
1223         if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1224                 ret = -ETIMEDOUT;
1225         return ret;
1226 }
1227
1228 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1229                                  struct block_device *bdev)
1230 {
1231         sock_shutdown(nbd);
1232         kill_bdev(bdev);
1233         nbd_bdev_reset(bdev);
1234         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1235                                &nbd->config->runtime_flags))
1236                 nbd_config_put(nbd);
1237 }
1238
1239 /* Must be called with config_lock held */
1240 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1241                        unsigned int cmd, unsigned long arg)
1242 {
1243         struct nbd_config *config = nbd->config;
1244
1245         switch (cmd) {
1246         case NBD_DISCONNECT:
1247                 return nbd_disconnect(nbd);
1248         case NBD_CLEAR_SOCK:
1249                 nbd_clear_sock_ioctl(nbd, bdev);
1250                 return 0;
1251         case NBD_SET_SOCK:
1252                 return nbd_add_socket(nbd, arg, false);
1253         case NBD_SET_BLKSIZE:
1254                 if (!arg || !is_power_of_2(arg) || arg < 512 ||
1255                     arg > PAGE_SIZE)
1256                         return -EINVAL;
1257                 nbd_size_set(nbd, arg,
1258                              div_s64(config->bytesize, arg));
1259                 return 0;
1260         case NBD_SET_SIZE:
1261                 nbd_size_set(nbd, config->blksize,
1262                              div_s64(arg, config->blksize));
1263                 return 0;
1264         case NBD_SET_SIZE_BLOCKS:
1265                 nbd_size_set(nbd, config->blksize, arg);
1266                 return 0;
1267         case NBD_SET_TIMEOUT:
1268                 if (arg) {
1269                         nbd->tag_set.timeout = arg * HZ;
1270                         blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1271                 }
1272                 return 0;
1273
1274         case NBD_SET_FLAGS:
1275                 config->flags = arg;
1276                 return 0;
1277         case NBD_DO_IT:
1278                 return nbd_start_device_ioctl(nbd, bdev);
1279         case NBD_CLEAR_QUE:
1280                 /*
1281                  * This is for compatibility only.  The queue is always cleared
1282                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
1283                  */
1284                 return 0;
1285         case NBD_PRINT_DEBUG:
1286                 /*
1287                  * For compatibility only, we no longer keep a list of
1288                  * outstanding requests.
1289                  */
1290                 return 0;
1291         }
1292         return -ENOTTY;
1293 }
1294
1295 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1296                      unsigned int cmd, unsigned long arg)
1297 {
1298         struct nbd_device *nbd = bdev->bd_disk->private_data;
1299         struct nbd_config *config = nbd->config;
1300         int error = -EINVAL;
1301
1302         if (!capable(CAP_SYS_ADMIN))
1303                 return -EPERM;
1304
1305         /* The block layer will pass back some non-nbd ioctls in case we have
1306          * special handling for them, but we don't so just return an error.
1307          */
1308         if (_IOC_TYPE(cmd) != 0xab)
1309                 return -EINVAL;
1310
1311         mutex_lock(&nbd->config_lock);
1312
1313         /* Don't allow ioctl operations on a nbd device that was created with
1314          * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1315          */
1316         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1317             (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1318                 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1319         else
1320                 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1321         mutex_unlock(&nbd->config_lock);
1322         return error;
1323 }
1324
1325 static struct nbd_config *nbd_alloc_config(void)
1326 {
1327         struct nbd_config *config;
1328
1329         config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1330         if (!config)
1331                 return NULL;
1332         atomic_set(&config->recv_threads, 0);
1333         init_waitqueue_head(&config->recv_wq);
1334         init_waitqueue_head(&config->conn_wait);
1335         config->blksize = 1024;
1336         atomic_set(&config->live_connections, 0);
1337         try_module_get(THIS_MODULE);
1338         return config;
1339 }
1340
1341 static int nbd_open(struct block_device *bdev, fmode_t mode)
1342 {
1343         struct nbd_device *nbd;
1344         int ret = 0;
1345
1346         mutex_lock(&nbd_index_mutex);
1347         nbd = bdev->bd_disk->private_data;
1348         if (!nbd) {
1349                 ret = -ENXIO;
1350                 goto out;
1351         }
1352         if (!refcount_inc_not_zero(&nbd->refs)) {
1353                 ret = -ENXIO;
1354                 goto out;
1355         }
1356         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1357                 struct nbd_config *config;
1358
1359                 mutex_lock(&nbd->config_lock);
1360                 if (refcount_inc_not_zero(&nbd->config_refs)) {
1361                         mutex_unlock(&nbd->config_lock);
1362                         goto out;
1363                 }
1364                 config = nbd->config = nbd_alloc_config();
1365                 if (!config) {
1366                         ret = -ENOMEM;
1367                         mutex_unlock(&nbd->config_lock);
1368                         goto out;
1369                 }
1370                 refcount_set(&nbd->config_refs, 1);
1371                 refcount_inc(&nbd->refs);
1372                 mutex_unlock(&nbd->config_lock);
1373                 bdev->bd_invalidated = 1;
1374         } else if (nbd_disconnected(nbd->config)) {
1375                 bdev->bd_invalidated = 1;
1376         }
1377 out:
1378         mutex_unlock(&nbd_index_mutex);
1379         return ret;
1380 }
1381
1382 static void nbd_release(struct gendisk *disk, fmode_t mode)
1383 {
1384         struct nbd_device *nbd = disk->private_data;
1385         struct block_device *bdev = bdget_disk(disk, 0);
1386
1387         if (test_bit(NBD_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1388                         bdev->bd_openers == 0)
1389                 nbd_disconnect_and_put(nbd);
1390
1391         nbd_config_put(nbd);
1392         nbd_put(nbd);
1393 }
1394
1395 static const struct block_device_operations nbd_fops =
1396 {
1397         .owner =        THIS_MODULE,
1398         .open =         nbd_open,
1399         .release =      nbd_release,
1400         .ioctl =        nbd_ioctl,
1401         .compat_ioctl = nbd_ioctl,
1402 };
1403
1404 #if IS_ENABLED(CONFIG_DEBUG_FS)
1405
1406 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1407 {
1408         struct nbd_device *nbd = s->private;
1409
1410         if (nbd->task_recv)
1411                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1412
1413         return 0;
1414 }
1415
1416 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1417 {
1418         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1419 }
1420
1421 static const struct file_operations nbd_dbg_tasks_ops = {
1422         .open = nbd_dbg_tasks_open,
1423         .read = seq_read,
1424         .llseek = seq_lseek,
1425         .release = single_release,
1426 };
1427
1428 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1429 {
1430         struct nbd_device *nbd = s->private;
1431         u32 flags = nbd->config->flags;
1432
1433         seq_printf(s, "Hex: 0x%08x\n\n", flags);
1434
1435         seq_puts(s, "Known flags:\n");
1436
1437         if (flags & NBD_FLAG_HAS_FLAGS)
1438                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1439         if (flags & NBD_FLAG_READ_ONLY)
1440                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1441         if (flags & NBD_FLAG_SEND_FLUSH)
1442                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1443         if (flags & NBD_FLAG_SEND_FUA)
1444                 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1445         if (flags & NBD_FLAG_SEND_TRIM)
1446                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1447
1448         return 0;
1449 }
1450
1451 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1452 {
1453         return single_open(file, nbd_dbg_flags_show, inode->i_private);
1454 }
1455
1456 static const struct file_operations nbd_dbg_flags_ops = {
1457         .open = nbd_dbg_flags_open,
1458         .read = seq_read,
1459         .llseek = seq_lseek,
1460         .release = single_release,
1461 };
1462
1463 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1464 {
1465         struct dentry *dir;
1466         struct nbd_config *config = nbd->config;
1467
1468         if (!nbd_dbg_dir)
1469                 return -EIO;
1470
1471         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1472         if (!dir) {
1473                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1474                         nbd_name(nbd));
1475                 return -EIO;
1476         }
1477         config->dbg_dir = dir;
1478
1479         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1480         debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1481         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1482         debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1483         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1484
1485         return 0;
1486 }
1487
1488 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1489 {
1490         debugfs_remove_recursive(nbd->config->dbg_dir);
1491 }
1492
1493 static int nbd_dbg_init(void)
1494 {
1495         struct dentry *dbg_dir;
1496
1497         dbg_dir = debugfs_create_dir("nbd", NULL);
1498         if (!dbg_dir)
1499                 return -EIO;
1500
1501         nbd_dbg_dir = dbg_dir;
1502
1503         return 0;
1504 }
1505
1506 static void nbd_dbg_close(void)
1507 {
1508         debugfs_remove_recursive(nbd_dbg_dir);
1509 }
1510
1511 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1512
1513 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1514 {
1515         return 0;
1516 }
1517
1518 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1519 {
1520 }
1521
1522 static int nbd_dbg_init(void)
1523 {
1524         return 0;
1525 }
1526
1527 static void nbd_dbg_close(void)
1528 {
1529 }
1530
1531 #endif
1532
1533 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1534                             unsigned int hctx_idx, unsigned int numa_node)
1535 {
1536         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1537         cmd->nbd = set->driver_data;
1538         cmd->flags = 0;
1539         mutex_init(&cmd->lock);
1540         return 0;
1541 }
1542
1543 static const struct blk_mq_ops nbd_mq_ops = {
1544         .queue_rq       = nbd_queue_rq,
1545         .complete       = nbd_complete_rq,
1546         .init_request   = nbd_init_request,
1547         .timeout        = nbd_xmit_timeout,
1548 };
1549
1550 static int nbd_dev_add(int index)
1551 {
1552         struct nbd_device *nbd;
1553         struct gendisk *disk;
1554         struct request_queue *q;
1555         int err = -ENOMEM;
1556
1557         nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1558         if (!nbd)
1559                 goto out;
1560
1561         disk = alloc_disk(1 << part_shift);
1562         if (!disk)
1563                 goto out_free_nbd;
1564
1565         if (index >= 0) {
1566                 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1567                                 GFP_KERNEL);
1568                 if (err == -ENOSPC)
1569                         err = -EEXIST;
1570         } else {
1571                 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1572                 if (err >= 0)
1573                         index = err;
1574         }
1575         if (err < 0)
1576                 goto out_free_disk;
1577
1578         nbd->index = index;
1579         nbd->disk = disk;
1580         nbd->tag_set.ops = &nbd_mq_ops;
1581         nbd->tag_set.nr_hw_queues = 1;
1582         nbd->tag_set.queue_depth = 128;
1583         nbd->tag_set.numa_node = NUMA_NO_NODE;
1584         nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1585         nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1586                 BLK_MQ_F_BLOCKING;
1587         nbd->tag_set.driver_data = nbd;
1588
1589         err = blk_mq_alloc_tag_set(&nbd->tag_set);
1590         if (err)
1591                 goto out_free_idr;
1592
1593         q = blk_mq_init_queue(&nbd->tag_set);
1594         if (IS_ERR(q)) {
1595                 err = PTR_ERR(q);
1596                 goto out_free_tags;
1597         }
1598         disk->queue = q;
1599
1600         /*
1601          * Tell the block layer that we are not a rotational device
1602          */
1603         blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1604         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1605         disk->queue->limits.discard_granularity = 0;
1606         disk->queue->limits.discard_alignment = 0;
1607         blk_queue_max_discard_sectors(disk->queue, 0);
1608         blk_queue_max_segment_size(disk->queue, UINT_MAX);
1609         blk_queue_max_segments(disk->queue, USHRT_MAX);
1610         blk_queue_max_hw_sectors(disk->queue, 65536);
1611         disk->queue->limits.max_sectors = 256;
1612
1613         mutex_init(&nbd->config_lock);
1614         refcount_set(&nbd->config_refs, 0);
1615         refcount_set(&nbd->refs, 1);
1616         INIT_LIST_HEAD(&nbd->list);
1617         disk->major = NBD_MAJOR;
1618         disk->first_minor = index << part_shift;
1619         disk->fops = &nbd_fops;
1620         disk->private_data = nbd;
1621         sprintf(disk->disk_name, "nbd%d", index);
1622         add_disk(disk);
1623         nbd_total_devices++;
1624         return index;
1625
1626 out_free_tags:
1627         blk_mq_free_tag_set(&nbd->tag_set);
1628 out_free_idr:
1629         idr_remove(&nbd_index_idr, index);
1630 out_free_disk:
1631         put_disk(disk);
1632 out_free_nbd:
1633         kfree(nbd);
1634 out:
1635         return err;
1636 }
1637
1638 static int find_free_cb(int id, void *ptr, void *data)
1639 {
1640         struct nbd_device *nbd = ptr;
1641         struct nbd_device **found = data;
1642
1643         if (!refcount_read(&nbd->config_refs)) {
1644                 *found = nbd;
1645                 return 1;
1646         }
1647         return 0;
1648 }
1649
1650 /* Netlink interface. */
1651 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1652         [NBD_ATTR_INDEX]                =       { .type = NLA_U32 },
1653         [NBD_ATTR_SIZE_BYTES]           =       { .type = NLA_U64 },
1654         [NBD_ATTR_BLOCK_SIZE_BYTES]     =       { .type = NLA_U64 },
1655         [NBD_ATTR_TIMEOUT]              =       { .type = NLA_U64 },
1656         [NBD_ATTR_SERVER_FLAGS]         =       { .type = NLA_U64 },
1657         [NBD_ATTR_CLIENT_FLAGS]         =       { .type = NLA_U64 },
1658         [NBD_ATTR_SOCKETS]              =       { .type = NLA_NESTED},
1659         [NBD_ATTR_DEAD_CONN_TIMEOUT]    =       { .type = NLA_U64 },
1660         [NBD_ATTR_DEVICE_LIST]          =       { .type = NLA_NESTED},
1661 };
1662
1663 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1664         [NBD_SOCK_FD]                   =       { .type = NLA_U32 },
1665 };
1666
1667 /* We don't use this right now since we don't parse the incoming list, but we
1668  * still want it here so userspace knows what to expect.
1669  */
1670 static const struct nla_policy __attribute__((unused))
1671 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1672         [NBD_DEVICE_INDEX]              =       { .type = NLA_U32 },
1673         [NBD_DEVICE_CONNECTED]          =       { .type = NLA_U8 },
1674 };
1675
1676 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1677 {
1678         struct nbd_device *nbd = NULL;
1679         struct nbd_config *config;
1680         int index = -1;
1681         int ret;
1682         bool put_dev = false;
1683
1684         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1685                 return -EPERM;
1686
1687         if (info->attrs[NBD_ATTR_INDEX])
1688                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1689         if (!info->attrs[NBD_ATTR_SOCKETS]) {
1690                 printk(KERN_ERR "nbd: must specify at least one socket\n");
1691                 return -EINVAL;
1692         }
1693         if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1694                 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1695                 return -EINVAL;
1696         }
1697 again:
1698         mutex_lock(&nbd_index_mutex);
1699         if (index == -1) {
1700                 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1701                 if (ret == 0) {
1702                         int new_index;
1703                         new_index = nbd_dev_add(-1);
1704                         if (new_index < 0) {
1705                                 mutex_unlock(&nbd_index_mutex);
1706                                 printk(KERN_ERR "nbd: failed to add new device\n");
1707                                 return new_index;
1708                         }
1709                         nbd = idr_find(&nbd_index_idr, new_index);
1710                 }
1711         } else {
1712                 nbd = idr_find(&nbd_index_idr, index);
1713                 if (!nbd) {
1714                         ret = nbd_dev_add(index);
1715                         if (ret < 0) {
1716                                 mutex_unlock(&nbd_index_mutex);
1717                                 printk(KERN_ERR "nbd: failed to add new device\n");
1718                                 return ret;
1719                         }
1720                         nbd = idr_find(&nbd_index_idr, index);
1721                 }
1722         }
1723         if (!nbd) {
1724                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1725                        index);
1726                 mutex_unlock(&nbd_index_mutex);
1727                 return -EINVAL;
1728         }
1729         if (!refcount_inc_not_zero(&nbd->refs)) {
1730                 mutex_unlock(&nbd_index_mutex);
1731                 if (index == -1)
1732                         goto again;
1733                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1734                        index);
1735                 return -EINVAL;
1736         }
1737         mutex_unlock(&nbd_index_mutex);
1738
1739         mutex_lock(&nbd->config_lock);
1740         if (refcount_read(&nbd->config_refs)) {
1741                 mutex_unlock(&nbd->config_lock);
1742                 nbd_put(nbd);
1743                 if (index == -1)
1744                         goto again;
1745                 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1746                 return -EBUSY;
1747         }
1748         if (WARN_ON(nbd->config)) {
1749                 mutex_unlock(&nbd->config_lock);
1750                 nbd_put(nbd);
1751                 return -EINVAL;
1752         }
1753         config = nbd->config = nbd_alloc_config();
1754         if (!nbd->config) {
1755                 mutex_unlock(&nbd->config_lock);
1756                 nbd_put(nbd);
1757                 printk(KERN_ERR "nbd: couldn't allocate config\n");
1758                 return -ENOMEM;
1759         }
1760         refcount_set(&nbd->config_refs, 1);
1761         set_bit(NBD_BOUND, &config->runtime_flags);
1762
1763         if (info->attrs[NBD_ATTR_SIZE_BYTES]) {
1764                 u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1765                 nbd_size_set(nbd, config->blksize,
1766                              div64_u64(bytes, config->blksize));
1767         }
1768         if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1769                 u64 bsize =
1770                         nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1771                 nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize));
1772         }
1773         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1774                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1775                 nbd->tag_set.timeout = timeout * HZ;
1776                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1777         }
1778         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1779                 config->dead_conn_timeout =
1780                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1781                 config->dead_conn_timeout *= HZ;
1782         }
1783         if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1784                 config->flags =
1785                         nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1786         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1787                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1788                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1789                         set_bit(NBD_DESTROY_ON_DISCONNECT,
1790                                 &config->runtime_flags);
1791                         put_dev = true;
1792                 }
1793                 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1794                         set_bit(NBD_DISCONNECT_ON_CLOSE,
1795                                 &config->runtime_flags);
1796                 }
1797         }
1798
1799         if (info->attrs[NBD_ATTR_SOCKETS]) {
1800                 struct nlattr *attr;
1801                 int rem, fd;
1802
1803                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1804                                     rem) {
1805                         struct nlattr *socks[NBD_SOCK_MAX+1];
1806
1807                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1808                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1809                                 ret = -EINVAL;
1810                                 goto out;
1811                         }
1812                         ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1813                                                           attr,
1814                                                           nbd_sock_policy,
1815                                                           info->extack);
1816                         if (ret != 0) {
1817                                 printk(KERN_ERR "nbd: error processing sock list\n");
1818                                 ret = -EINVAL;
1819                                 goto out;
1820                         }
1821                         if (!socks[NBD_SOCK_FD])
1822                                 continue;
1823                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1824                         ret = nbd_add_socket(nbd, fd, true);
1825                         if (ret)
1826                                 goto out;
1827                 }
1828         }
1829         ret = nbd_start_device(nbd);
1830 out:
1831         mutex_unlock(&nbd->config_lock);
1832         if (!ret) {
1833                 set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1834                 refcount_inc(&nbd->config_refs);
1835                 nbd_connect_reply(info, nbd->index);
1836         }
1837         nbd_config_put(nbd);
1838         if (put_dev)
1839                 nbd_put(nbd);
1840         return ret;
1841 }
1842
1843 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1844 {
1845         mutex_lock(&nbd->config_lock);
1846         nbd_disconnect(nbd);
1847         nbd_clear_sock(nbd);
1848         mutex_unlock(&nbd->config_lock);
1849         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1850                                &nbd->config->runtime_flags))
1851                 nbd_config_put(nbd);
1852 }
1853
1854 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1855 {
1856         struct nbd_device *nbd;
1857         int index;
1858
1859         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1860                 return -EPERM;
1861
1862         if (!info->attrs[NBD_ATTR_INDEX]) {
1863                 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1864                 return -EINVAL;
1865         }
1866         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1867         mutex_lock(&nbd_index_mutex);
1868         nbd = idr_find(&nbd_index_idr, index);
1869         if (!nbd) {
1870                 mutex_unlock(&nbd_index_mutex);
1871                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1872                        index);
1873                 return -EINVAL;
1874         }
1875         if (!refcount_inc_not_zero(&nbd->refs)) {
1876                 mutex_unlock(&nbd_index_mutex);
1877                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1878                        index);
1879                 return -EINVAL;
1880         }
1881         mutex_unlock(&nbd_index_mutex);
1882         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1883                 nbd_put(nbd);
1884                 return 0;
1885         }
1886         nbd_disconnect_and_put(nbd);
1887         nbd_config_put(nbd);
1888         nbd_put(nbd);
1889         return 0;
1890 }
1891
1892 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1893 {
1894         struct nbd_device *nbd = NULL;
1895         struct nbd_config *config;
1896         int index;
1897         int ret = 0;
1898         bool put_dev = false;
1899
1900         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1901                 return -EPERM;
1902
1903         if (!info->attrs[NBD_ATTR_INDEX]) {
1904                 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1905                 return -EINVAL;
1906         }
1907         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1908         mutex_lock(&nbd_index_mutex);
1909         nbd = idr_find(&nbd_index_idr, index);
1910         if (!nbd) {
1911                 mutex_unlock(&nbd_index_mutex);
1912                 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1913                        index);
1914                 return -EINVAL;
1915         }
1916         if (!refcount_inc_not_zero(&nbd->refs)) {
1917                 mutex_unlock(&nbd_index_mutex);
1918                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1919                        index);
1920                 return -EINVAL;
1921         }
1922         mutex_unlock(&nbd_index_mutex);
1923
1924         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1925                 dev_err(nbd_to_dev(nbd),
1926                         "not configured, cannot reconfigure\n");
1927                 nbd_put(nbd);
1928                 return -EINVAL;
1929         }
1930
1931         mutex_lock(&nbd->config_lock);
1932         config = nbd->config;
1933         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1934             !nbd->task_recv) {
1935                 dev_err(nbd_to_dev(nbd),
1936                         "not configured, cannot reconfigure\n");
1937                 ret = -EINVAL;
1938                 goto out;
1939         }
1940
1941         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1942                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1943                 nbd->tag_set.timeout = timeout * HZ;
1944                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1945         }
1946         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1947                 config->dead_conn_timeout =
1948                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1949                 config->dead_conn_timeout *= HZ;
1950         }
1951         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1952                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1953                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1954                         if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1955                                               &config->runtime_flags))
1956                                 put_dev = true;
1957                 } else {
1958                         if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1959                                                &config->runtime_flags))
1960                                 refcount_inc(&nbd->refs);
1961                 }
1962
1963                 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1964                         set_bit(NBD_DISCONNECT_ON_CLOSE,
1965                                         &config->runtime_flags);
1966                 } else {
1967                         clear_bit(NBD_DISCONNECT_ON_CLOSE,
1968                                         &config->runtime_flags);
1969                 }
1970         }
1971
1972         if (info->attrs[NBD_ATTR_SOCKETS]) {
1973                 struct nlattr *attr;
1974                 int rem, fd;
1975
1976                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1977                                     rem) {
1978                         struct nlattr *socks[NBD_SOCK_MAX+1];
1979
1980                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1981                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1982                                 ret = -EINVAL;
1983                                 goto out;
1984                         }
1985                         ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1986                                                           attr,
1987                                                           nbd_sock_policy,
1988                                                           info->extack);
1989                         if (ret != 0) {
1990                                 printk(KERN_ERR "nbd: error processing sock list\n");
1991                                 ret = -EINVAL;
1992                                 goto out;
1993                         }
1994                         if (!socks[NBD_SOCK_FD])
1995                                 continue;
1996                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1997                         ret = nbd_reconnect_socket(nbd, fd);
1998                         if (ret) {
1999                                 if (ret == -ENOSPC)
2000                                         ret = 0;
2001                                 goto out;
2002                         }
2003                         dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2004                 }
2005         }
2006 out:
2007         mutex_unlock(&nbd->config_lock);
2008         nbd_config_put(nbd);
2009         nbd_put(nbd);
2010         if (put_dev)
2011                 nbd_put(nbd);
2012         return ret;
2013 }
2014
2015 static const struct genl_ops nbd_connect_genl_ops[] = {
2016         {
2017                 .cmd    = NBD_CMD_CONNECT,
2018                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2019                 .doit   = nbd_genl_connect,
2020         },
2021         {
2022                 .cmd    = NBD_CMD_DISCONNECT,
2023                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2024                 .doit   = nbd_genl_disconnect,
2025         },
2026         {
2027                 .cmd    = NBD_CMD_RECONFIGURE,
2028                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2029                 .doit   = nbd_genl_reconfigure,
2030         },
2031         {
2032                 .cmd    = NBD_CMD_STATUS,
2033                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2034                 .doit   = nbd_genl_status,
2035         },
2036 };
2037
2038 static const struct genl_multicast_group nbd_mcast_grps[] = {
2039         { .name = NBD_GENL_MCAST_GROUP_NAME, },
2040 };
2041
2042 static struct genl_family nbd_genl_family __ro_after_init = {
2043         .hdrsize        = 0,
2044         .name           = NBD_GENL_FAMILY_NAME,
2045         .version        = NBD_GENL_VERSION,
2046         .module         = THIS_MODULE,
2047         .ops            = nbd_connect_genl_ops,
2048         .n_ops          = ARRAY_SIZE(nbd_connect_genl_ops),
2049         .maxattr        = NBD_ATTR_MAX,
2050         .policy = nbd_attr_policy,
2051         .mcgrps         = nbd_mcast_grps,
2052         .n_mcgrps       = ARRAY_SIZE(nbd_mcast_grps),
2053 };
2054
2055 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2056 {
2057         struct nlattr *dev_opt;
2058         u8 connected = 0;
2059         int ret;
2060
2061         /* This is a little racey, but for status it's ok.  The
2062          * reason we don't take a ref here is because we can't
2063          * take a ref in the index == -1 case as we would need
2064          * to put under the nbd_index_mutex, which could
2065          * deadlock if we are configured to remove ourselves
2066          * once we're disconnected.
2067          */
2068         if (refcount_read(&nbd->config_refs))
2069                 connected = 1;
2070         dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2071         if (!dev_opt)
2072                 return -EMSGSIZE;
2073         ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2074         if (ret)
2075                 return -EMSGSIZE;
2076         ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2077                          connected);
2078         if (ret)
2079                 return -EMSGSIZE;
2080         nla_nest_end(reply, dev_opt);
2081         return 0;
2082 }
2083
2084 static int status_cb(int id, void *ptr, void *data)
2085 {
2086         struct nbd_device *nbd = ptr;
2087         return populate_nbd_status(nbd, (struct sk_buff *)data);
2088 }
2089
2090 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2091 {
2092         struct nlattr *dev_list;
2093         struct sk_buff *reply;
2094         void *reply_head;
2095         size_t msg_size;
2096         int index = -1;
2097         int ret = -ENOMEM;
2098
2099         if (info->attrs[NBD_ATTR_INDEX])
2100                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2101
2102         mutex_lock(&nbd_index_mutex);
2103
2104         msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2105                                   nla_attr_size(sizeof(u8)));
2106         msg_size *= (index == -1) ? nbd_total_devices : 1;
2107
2108         reply = genlmsg_new(msg_size, GFP_KERNEL);
2109         if (!reply)
2110                 goto out;
2111         reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2112                                        NBD_CMD_STATUS);
2113         if (!reply_head) {
2114                 nlmsg_free(reply);
2115                 goto out;
2116         }
2117
2118         dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2119         if (index == -1) {
2120                 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2121                 if (ret) {
2122                         nlmsg_free(reply);
2123                         goto out;
2124                 }
2125         } else {
2126                 struct nbd_device *nbd;
2127                 nbd = idr_find(&nbd_index_idr, index);
2128                 if (nbd) {
2129                         ret = populate_nbd_status(nbd, reply);
2130                         if (ret) {
2131                                 nlmsg_free(reply);
2132                                 goto out;
2133                         }
2134                 }
2135         }
2136         nla_nest_end(reply, dev_list);
2137         genlmsg_end(reply, reply_head);
2138         ret = genlmsg_reply(reply, info);
2139 out:
2140         mutex_unlock(&nbd_index_mutex);
2141         return ret;
2142 }
2143
2144 static void nbd_connect_reply(struct genl_info *info, int index)
2145 {
2146         struct sk_buff *skb;
2147         void *msg_head;
2148         int ret;
2149
2150         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2151         if (!skb)
2152                 return;
2153         msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2154                                      NBD_CMD_CONNECT);
2155         if (!msg_head) {
2156                 nlmsg_free(skb);
2157                 return;
2158         }
2159         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2160         if (ret) {
2161                 nlmsg_free(skb);
2162                 return;
2163         }
2164         genlmsg_end(skb, msg_head);
2165         genlmsg_reply(skb, info);
2166 }
2167
2168 static void nbd_mcast_index(int index)
2169 {
2170         struct sk_buff *skb;
2171         void *msg_head;
2172         int ret;
2173
2174         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2175         if (!skb)
2176                 return;
2177         msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2178                                      NBD_CMD_LINK_DEAD);
2179         if (!msg_head) {
2180                 nlmsg_free(skb);
2181                 return;
2182         }
2183         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2184         if (ret) {
2185                 nlmsg_free(skb);
2186                 return;
2187         }
2188         genlmsg_end(skb, msg_head);
2189         genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2190 }
2191
2192 static void nbd_dead_link_work(struct work_struct *work)
2193 {
2194         struct link_dead_args *args = container_of(work, struct link_dead_args,
2195                                                    work);
2196         nbd_mcast_index(args->index);
2197         kfree(args);
2198 }
2199
2200 static int __init nbd_init(void)
2201 {
2202         int i;
2203
2204         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2205
2206         if (max_part < 0) {
2207                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2208                 return -EINVAL;
2209         }
2210
2211         part_shift = 0;
2212         if (max_part > 0) {
2213                 part_shift = fls(max_part);
2214
2215                 /*
2216                  * Adjust max_part according to part_shift as it is exported
2217                  * to user space so that user can know the max number of
2218                  * partition kernel should be able to manage.
2219                  *
2220                  * Note that -1 is required because partition 0 is reserved
2221                  * for the whole disk.
2222                  */
2223                 max_part = (1UL << part_shift) - 1;
2224         }
2225
2226         if ((1UL << part_shift) > DISK_MAX_PARTS)
2227                 return -EINVAL;
2228
2229         if (nbds_max > 1UL << (MINORBITS - part_shift))
2230                 return -EINVAL;
2231         recv_workqueue = alloc_workqueue("knbd-recv",
2232                                          WQ_MEM_RECLAIM | WQ_HIGHPRI |
2233                                          WQ_UNBOUND, 0);
2234         if (!recv_workqueue)
2235                 return -ENOMEM;
2236
2237         if (register_blkdev(NBD_MAJOR, "nbd")) {
2238                 destroy_workqueue(recv_workqueue);
2239                 return -EIO;
2240         }
2241
2242         if (genl_register_family(&nbd_genl_family)) {
2243                 unregister_blkdev(NBD_MAJOR, "nbd");
2244                 destroy_workqueue(recv_workqueue);
2245                 return -EINVAL;
2246         }
2247         nbd_dbg_init();
2248
2249         mutex_lock(&nbd_index_mutex);
2250         for (i = 0; i < nbds_max; i++)
2251                 nbd_dev_add(i);
2252         mutex_unlock(&nbd_index_mutex);
2253         return 0;
2254 }
2255
2256 static int nbd_exit_cb(int id, void *ptr, void *data)
2257 {
2258         struct list_head *list = (struct list_head *)data;
2259         struct nbd_device *nbd = ptr;
2260
2261         list_add_tail(&nbd->list, list);
2262         return 0;
2263 }
2264
2265 static void __exit nbd_cleanup(void)
2266 {
2267         struct nbd_device *nbd;
2268         LIST_HEAD(del_list);
2269
2270         nbd_dbg_close();
2271
2272         mutex_lock(&nbd_index_mutex);
2273         idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2274         mutex_unlock(&nbd_index_mutex);
2275
2276         while (!list_empty(&del_list)) {
2277                 nbd = list_first_entry(&del_list, struct nbd_device, list);
2278                 list_del_init(&nbd->list);
2279                 if (refcount_read(&nbd->refs) != 1)
2280                         printk(KERN_ERR "nbd: possibly leaking a device\n");
2281                 nbd_put(nbd);
2282         }
2283
2284         idr_destroy(&nbd_index_idr);
2285         genl_unregister_family(&nbd_genl_family);
2286         destroy_workqueue(recv_workqueue);
2287         unregister_blkdev(NBD_MAJOR, "nbd");
2288 }
2289
2290 module_init(nbd_init);
2291 module_exit(nbd_cleanup);
2292
2293 MODULE_DESCRIPTION("Network Block Device");
2294 MODULE_LICENSE("GPL");
2295
2296 module_param(nbds_max, int, 0444);
2297 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2298 module_param(max_part, int, 0444);
2299 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");