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 #define NBD_DEF_BLKSIZE 1024
 138
 139 static unsigned int nbds_max = 16;
 140 static int max_part = 16;
 141 static struct workqueue_struct *recv_workqueue;
 142 static int part_shift;
 143
 144 static int nbd_dev_dbg_init(struct nbd_device *nbd);
 145 static void nbd_dev_dbg_close(struct nbd_device *nbd);
 146 static void nbd_config_put(struct nbd_device *nbd);
 147 static void nbd_connect_reply(struct genl_info *info, int index);
 148 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
 149 static void nbd_dead_link_work(struct work_struct *work);
 150 static void nbd_disconnect_and_put(struct nbd_device *nbd);
 151
 152 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
 153 {
 154         return disk_to_dev(nbd->disk);
 155 }
 156
 157 static void nbd_requeue_cmd(struct nbd_cmd *cmd)
 158 {
 159         struct request *req = blk_mq_rq_from_pdu(cmd);
 160
 161         if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
 162                 blk_mq_requeue_request(req, true);
 163 }
 164
 165 #define NBD_COOKIE_BITS 32
 166
 167 static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
 168 {
 169         struct request *req = blk_mq_rq_from_pdu(cmd);
 170         u32 tag = blk_mq_unique_tag(req);
 171         u64 cookie = cmd->cmd_cookie;
 172
 173         return (cookie << NBD_COOKIE_BITS) | tag;
 174 }
 175
 176 static u32 nbd_handle_to_tag(u64 handle)
 177 {
 178         return (u32)handle;
 179 }
 180
 181 static u32 nbd_handle_to_cookie(u64 handle)
 182 {
 183         return (u32)(handle >> NBD_COOKIE_BITS);
 184 }
 185
 186 static const char *nbdcmd_to_ascii(int cmd)
 187 {
 188         switch (cmd) {
 189         case  NBD_CMD_READ: return "read";
 190         case NBD_CMD_WRITE: return "write";
 191         case  NBD_CMD_DISC: return "disconnect";
 192         case NBD_CMD_FLUSH: return "flush";
 193         case  NBD_CMD_TRIM: return "trim/discard";
 194         }
 195         return "invalid";
 196 }
 197
 198 static ssize_t pid_show(struct device *dev,
 199                         struct device_attribute *attr, char *buf)
 200 {
 201         struct gendisk *disk = dev_to_disk(dev);
 202         struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
 203
 204         return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
 205 }
 206
 207 static const struct device_attribute pid_attr = {
 208         .attr = { .name = "pid", .mode = 0444},
 209         .show = pid_show,
 210 };
 211
 212 static void nbd_dev_remove(struct nbd_device *nbd)
 213 {
 214         struct gendisk *disk = nbd->disk;
 215         struct request_queue *q;
 216
 217         if (disk) {
 218                 q = disk->queue;
 219                 del_gendisk(disk);
 220                 blk_cleanup_queue(q);
 221                 blk_mq_free_tag_set(&nbd->tag_set);
 222                 disk->private_data = NULL;
 223                 put_disk(disk);
 224         }
 225         kfree(nbd);
 226 }
 227
 228 static void nbd_put(struct nbd_device *nbd)
 229 {
 230         if (refcount_dec_and_mutex_lock(&nbd->refs,
 231                                         &nbd_index_mutex)) {
 232                 idr_remove(&nbd_index_idr, nbd->index);
 233                 mutex_unlock(&nbd_index_mutex);
 234                 nbd_dev_remove(nbd);
 235         }
 236 }
 237
 238 static int nbd_disconnected(struct nbd_config *config)
 239 {
 240         return test_bit(NBD_DISCONNECTED, &config->runtime_flags) ||
 241                 test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
 242 }
 243
 244 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
 245                                 int notify)
 246 {
 247         if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
 248                 struct link_dead_args *args;
 249                 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
 250                 if (args) {
 251                         INIT_WORK(&args->work, nbd_dead_link_work);
 252                         args->index = nbd->index;
 253                         queue_work(system_wq, &args->work);
 254                 }
 255         }
 256         if (!nsock->dead) {
 257                 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
 258                 if (atomic_dec_return(&nbd->config->live_connections) == 0) {
 259                         if (test_and_clear_bit(NBD_DISCONNECT_REQUESTED,
 260                                                &nbd->config->runtime_flags)) {
 261                                 set_bit(NBD_DISCONNECTED,
 262                                         &nbd->config->runtime_flags);
 263                                 dev_info(nbd_to_dev(nbd),
 264                                         "Disconnected due to user request.\n");
 265                         }
 266                 }
 267         }
 268         nsock->dead = true;
 269         nsock->pending = NULL;
 270         nsock->sent = 0;
 271 }
 272
 273 static void nbd_size_clear(struct nbd_device *nbd)
 274 {
 275         if (nbd->config->bytesize) {
 276                 set_capacity(nbd->disk, 0);
 277                 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 278         }
 279 }
 280
 281 static void nbd_size_update(struct nbd_device *nbd)
 282 {
 283         struct nbd_config *config = nbd->config;
 284         struct block_device *bdev = bdget_disk(nbd->disk, 0);
 285
 286         if (config->flags & NBD_FLAG_SEND_TRIM) {
 287                 nbd->disk->queue->limits.discard_granularity = config->blksize;
 288                 nbd->disk->queue->limits.discard_alignment = config->blksize;
 289                 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
 290         }
 291         blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
 292         blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
 293         set_capacity(nbd->disk, config->bytesize >> 9);
 294         if (bdev) {
 295                 if (bdev->bd_disk) {
 296                         bd_set_size(bdev, config->bytesize);
 297                         set_blocksize(bdev, config->blksize);
 298                 } else
 299                         bdev->bd_invalidated = 1;
 300                 bdput(bdev);
 301         }
 302         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 303 }
 304
 305 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
 306                          loff_t nr_blocks)
 307 {
 308         struct nbd_config *config = nbd->config;
 309         config->blksize = blocksize;
 310         config->bytesize = blocksize * nr_blocks;
 311         if (nbd->task_recv != NULL)
 312                 nbd_size_update(nbd);
 313 }
 314
 315 static void nbd_complete_rq(struct request *req)
 316 {
 317         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 318
 319         dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
 320                 cmd->status ? "failed" : "done");
 321
 322         blk_mq_end_request(req, cmd->status);
 323 }
 324
 325 /*
 326  * Forcibly shutdown the socket causing all listeners to error
 327  */
 328 static void sock_shutdown(struct nbd_device *nbd)
 329 {
 330         struct nbd_config *config = nbd->config;
 331         int i;
 332
 333         if (config->num_connections == 0)
 334                 return;
 335         if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
 336                 return;
 337
 338         for (i = 0; i < config->num_connections; i++) {
 339                 struct nbd_sock *nsock = config->socks[i];
 340                 mutex_lock(&nsock->tx_lock);
 341                 nbd_mark_nsock_dead(nbd, nsock, 0);
 342                 mutex_unlock(&nsock->tx_lock);
 343         }
 344         dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
 345 }
 346
 347 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 348                                                  bool reserved)
 349 {
 350         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 351         struct nbd_device *nbd = cmd->nbd;
 352         struct nbd_config *config;
 353
 354         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 355                 cmd->status = BLK_STS_TIMEOUT;
 356                 goto done;
 357         }
 358         config = nbd->config;
 359
 360         if (!mutex_trylock(&cmd->lock))
 361                 return BLK_EH_RESET_TIMER;
 362
 363         if (config->num_connections > 1) {
 364                 dev_err_ratelimited(nbd_to_dev(nbd),
 365                                     "Connection timed out, retrying (%d/%d alive)\n",
 366                                     atomic_read(&config->live_connections),
 367                                     config->num_connections);
 368                 /*
 369                  * Hooray we have more connections, requeue this IO, the submit
 370                  * path will put it on a real connection.
 371                  */
 372                 if (config->socks && config->num_connections > 1) {
 373                         if (cmd->index < config->num_connections) {
 374                                 struct nbd_sock *nsock =
 375                                         config->socks[cmd->index];
 376                                 mutex_lock(&nsock->tx_lock);
 377                                 /* We can have multiple outstanding requests, so
 378                                  * we don't want to mark the nsock dead if we've
 379                                  * already reconnected with a new socket, so
 380                                  * only mark it dead if its the same socket we
 381                                  * were sent out on.
 382                                  */
 383                                 if (cmd->cookie == nsock->cookie)
 384                                         nbd_mark_nsock_dead(nbd, nsock, 1);
 385                                 mutex_unlock(&nsock->tx_lock);
 386                         }
 387                         mutex_unlock(&cmd->lock);
 388                         nbd_requeue_cmd(cmd);
 389                         nbd_config_put(nbd);
 390                         return BLK_EH_DONE;
 391                 }
 392         } else {
 393                 dev_err_ratelimited(nbd_to_dev(nbd),
 394                                     "Connection timed out\n");
 395         }
 396         set_bit(NBD_TIMEDOUT, &config->runtime_flags);
 397         cmd->status = BLK_STS_IOERR;
 398         mutex_unlock(&cmd->lock);
 399         sock_shutdown(nbd);
 400         nbd_config_put(nbd);
 401 done:
 402         blk_mq_complete_request(req);
 403         return BLK_EH_DONE;
 404 }
 405
 406 /*
 407  *  Send or receive packet.
 408  */
 409 static int sock_xmit(struct nbd_device *nbd, int index, int send,
 410                      struct iov_iter *iter, int msg_flags, int *sent)
 411 {
 412         struct nbd_config *config = nbd->config;
 413         struct socket *sock = config->socks[index]->sock;
 414         int result;
 415         struct msghdr msg;
 416         unsigned int noreclaim_flag;
 417
 418         if (unlikely(!sock)) {
 419                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 420                         "Attempted %s on closed socket in sock_xmit\n",
 421                         (send ? "send" : "recv"));
 422                 return -EINVAL;
 423         }
 424
 425         msg.msg_iter = *iter;
 426
 427         noreclaim_flag = memalloc_noreclaim_save();
 428         do {
 429                 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
 430                 msg.msg_name = NULL;
 431                 msg.msg_namelen = 0;
 432                 msg.msg_control = NULL;
 433                 msg.msg_controllen = 0;
 434                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 435
 436                 if (send)
 437                         result = sock_sendmsg(sock, &msg);
 438                 else
 439                         result = sock_recvmsg(sock, &msg, msg.msg_flags);
 440
 441                 if (result <= 0) {
 442                         if (result == 0)
 443                                 result = -EPIPE; /* short read */
 444                         break;
 445                 }
 446                 if (sent)
 447                         *sent += result;
 448         } while (msg_data_left(&msg));
 449
 450         memalloc_noreclaim_restore(noreclaim_flag);
 451
 452         return result;
 453 }
 454
 455 /*
 456  * Different settings for sk->sk_sndtimeo can result in different return values
 457  * if there is a signal pending when we enter sendmsg, because reasons?
 458  */
 459 static inline int was_interrupted(int result)
 460 {
 461         return result == -ERESTARTSYS || result == -EINTR;
 462 }
 463
 464 /* always call with the tx_lock held */
 465 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 466 {
 467         struct request *req = blk_mq_rq_from_pdu(cmd);
 468         struct nbd_config *config = nbd->config;
 469         struct nbd_sock *nsock = config->socks[index];
 470         int result;
 471         struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
 472         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 473         struct iov_iter from;
 474         unsigned long size = blk_rq_bytes(req);
 475         struct bio *bio;
 476         u64 handle;
 477         u32 type;
 478         u32 nbd_cmd_flags = 0;
 479         int sent = nsock->sent, skip = 0;
 480
 481         iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
 482
 483         switch (req_op(req)) {
 484         case REQ_OP_DISCARD:
 485                 type = NBD_CMD_TRIM;
 486                 break;
 487         case REQ_OP_FLUSH:
 488                 type = NBD_CMD_FLUSH;
 489                 break;
 490         case REQ_OP_WRITE:
 491                 type = NBD_CMD_WRITE;
 492                 break;
 493         case REQ_OP_READ:
 494                 type = NBD_CMD_READ;
 495                 break;
 496         default:
 497                 return -EIO;
 498         }
 499
 500         if (rq_data_dir(req) == WRITE &&
 501             (config->flags & NBD_FLAG_READ_ONLY)) {
 502                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 503                                     "Write on read-only\n");
 504                 return -EIO;
 505         }
 506
 507         if (req->cmd_flags & REQ_FUA)
 508                 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
 509
 510         /* We did a partial send previously, and we at least sent the whole
 511          * request struct, so just go and send the rest of the pages in the
 512          * request.
 513          */
 514         if (sent) {
 515                 if (sent >= sizeof(request)) {
 516                         skip = sent - sizeof(request);
 517
 518                         /* initialize handle for tracing purposes */
 519                         handle = nbd_cmd_handle(cmd);
 520
 521                         goto send_pages;
 522                 }
 523                 iov_iter_advance(&from, sent);
 524         } else {
 525                 cmd->cmd_cookie++;
 526         }
 527         cmd->index = index;
 528         cmd->cookie = nsock->cookie;
 529         request.type = htonl(type | nbd_cmd_flags);
 530         if (type != NBD_CMD_FLUSH) {
 531                 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
 532                 request.len = htonl(size);
 533         }
 534         handle = nbd_cmd_handle(cmd);
 535         memcpy(request.handle, &handle, sizeof(handle));
 536
 537         trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
 538
 539         dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
 540                 req, nbdcmd_to_ascii(type),
 541                 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
 542         result = sock_xmit(nbd, index, 1, &from,
 543                         (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
 544         trace_nbd_header_sent(req, handle);
 545         if (result <= 0) {
 546                 if (was_interrupted(result)) {
 547                         /* If we havne't sent anything we can just return BUSY,
 548                          * however if we have sent something we need to make
 549                          * sure we only allow this req to be sent until we are
 550                          * completely done.
 551                          */
 552                         if (sent) {
 553                                 nsock->pending = req;
 554                                 nsock->sent = sent;
 555                         }
 556                         set_bit(NBD_CMD_REQUEUED, &cmd->flags);
 557                         return BLK_STS_RESOURCE;
 558                 }
 559                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 560                         "Send control failed (result %d)\n", result);
 561                 return -EAGAIN;
 562         }
 563 send_pages:
 564         if (type != NBD_CMD_WRITE)
 565                 goto out;
 566
 567         bio = req->bio;
 568         while (bio) {
 569                 struct bio *next = bio->bi_next;
 570                 struct bvec_iter iter;
 571                 struct bio_vec bvec;
 572
 573                 bio_for_each_segment(bvec, bio, iter) {
 574                         bool is_last = !next && bio_iter_last(bvec, iter);
 575                         int flags = is_last ? 0 : MSG_MORE;
 576
 577                         dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
 578                                 req, bvec.bv_len);
 579                         iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
 580                         if (skip) {
 581                                 if (skip >= iov_iter_count(&from)) {
 582                                         skip -= iov_iter_count(&from);
 583                                         continue;
 584                                 }
 585                                 iov_iter_advance(&from, skip);
 586                                 skip = 0;
 587                         }
 588                         result = sock_xmit(nbd, index, 1, &from, flags, &sent);
 589                         if (result <= 0) {
 590                                 if (was_interrupted(result)) {
 591                                         /* We've already sent the header, we
 592                                          * have no choice but to set pending and
 593                                          * return BUSY.
 594                                          */
 595                                         nsock->pending = req;
 596                                         nsock->sent = sent;
 597                                         set_bit(NBD_CMD_REQUEUED, &cmd->flags);
 598                                         return BLK_STS_RESOURCE;
 599                                 }
 600                                 dev_err(disk_to_dev(nbd->disk),
 601                                         "Send data failed (result %d)\n",
 602                                         result);
 603                                 return -EAGAIN;
 604                         }
 605                         /*
 606                          * The completion might already have come in,
 607                          * so break for the last one instead of letting
 608                          * the iterator do it. This prevents use-after-free
 609                          * of the bio.
 610                          */
 611                         if (is_last)
 612                                 break;
 613                 }
 614                 bio = next;
 615         }
 616 out:
 617         trace_nbd_payload_sent(req, handle);
 618         nsock->pending = NULL;
 619         nsock->sent = 0;
 620         return 0;
 621 }
 622
 623 /* NULL returned = something went wrong, inform userspace */
 624 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 625 {
 626         struct nbd_config *config = nbd->config;
 627         int result;
 628         struct nbd_reply reply;
 629         struct nbd_cmd *cmd;
 630         struct request *req = NULL;
 631         u64 handle;
 632         u16 hwq;
 633         u32 tag;
 634         struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
 635         struct iov_iter to;
 636         int ret = 0;
 637
 638         reply.magic = 0;
 639         iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
 640         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 641         if (result <= 0) {
 642                 if (!nbd_disconnected(config))
 643                         dev_err(disk_to_dev(nbd->disk),
 644                                 "Receive control failed (result %d)\n", result);
 645                 return ERR_PTR(result);
 646         }
 647
 648         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 649                 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
 650                                 (unsigned long)ntohl(reply.magic));
 651                 return ERR_PTR(-EPROTO);
 652         }
 653
 654         memcpy(&handle, reply.handle, sizeof(handle));
 655         tag = nbd_handle_to_tag(handle);
 656         hwq = blk_mq_unique_tag_to_hwq(tag);
 657         if (hwq < nbd->tag_set.nr_hw_queues)
 658                 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
 659                                        blk_mq_unique_tag_to_tag(tag));
 660         if (!req || !blk_mq_request_started(req)) {
 661                 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
 662                         tag, req);
 663                 return ERR_PTR(-ENOENT);
 664         }
 665         trace_nbd_header_received(req, handle);
 666         cmd = blk_mq_rq_to_pdu(req);
 667
 668         mutex_lock(&cmd->lock);
 669         if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
 670                 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
 671                         req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
 672                 ret = -ENOENT;
 673                 goto out;
 674         }
 675         if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
 676                 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
 677                         req);
 678                 ret = -ENOENT;
 679                 goto out;
 680         }
 681         if (ntohl(reply.error)) {
 682                 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
 683                         ntohl(reply.error));
 684                 cmd->status = BLK_STS_IOERR;
 685                 goto out;
 686         }
 687
 688         dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
 689         if (rq_data_dir(req) != WRITE) {
 690                 struct req_iterator iter;
 691                 struct bio_vec bvec;
 692
 693                 rq_for_each_segment(bvec, req, iter) {
 694                         iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
 695                         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 696                         if (result <= 0) {
 697                                 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 698                                         result);
 699                                 /*
 700                                  * If we've disconnected or we only have 1
 701                                  * connection then we need to make sure we
 702                                  * complete this request, otherwise error out
 703                                  * and let the timeout stuff handle resubmitting
 704                                  * this request onto another connection.
 705                                  */
 706                                 if (nbd_disconnected(config) ||
 707                                     config->num_connections <= 1) {
 708                                         cmd->status = BLK_STS_IOERR;
 709                                         goto out;
 710                                 }
 711                                 ret = -EIO;
 712                                 goto out;
 713                         }
 714                         dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 715                                 req, bvec.bv_len);
 716                 }
 717         }
 718 out:
 719         trace_nbd_payload_received(req, handle);
 720         mutex_unlock(&cmd->lock);
 721         return ret ? ERR_PTR(ret) : cmd;
 722 }
 723
 724 static void recv_work(struct work_struct *work)
 725 {
 726         struct recv_thread_args *args = container_of(work,
 727                                                      struct recv_thread_args,
 728                                                      work);
 729         struct nbd_device *nbd = args->nbd;
 730         struct nbd_config *config = nbd->config;
 731         struct nbd_cmd *cmd;
 732
 733         while (1) {
 734                 cmd = nbd_read_stat(nbd, args->index);
 735                 if (IS_ERR(cmd)) {
 736                         struct nbd_sock *nsock = config->socks[args->index];
 737
 738                         mutex_lock(&nsock->tx_lock);
 739                         nbd_mark_nsock_dead(nbd, nsock, 1);
 740                         mutex_unlock(&nsock->tx_lock);
 741                         break;
 742                 }
 743
 744                 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
 745         }
 746         atomic_dec(&config->recv_threads);
 747         wake_up(&config->recv_wq);
 748         nbd_config_put(nbd);
 749         kfree(args);
 750 }
 751
 752 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
 753 {
 754         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 755
 756         cmd->status = BLK_STS_IOERR;
 757         blk_mq_complete_request(req);
 758         return true;
 759 }
 760
 761 static void nbd_clear_que(struct nbd_device *nbd)
 762 {
 763         blk_mq_quiesce_queue(nbd->disk->queue);
 764         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 765         blk_mq_unquiesce_queue(nbd->disk->queue);
 766         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 767 }
 768
 769 static int find_fallback(struct nbd_device *nbd, int index)
 770 {
 771         struct nbd_config *config = nbd->config;
 772         int new_index = -1;
 773         struct nbd_sock *nsock = config->socks[index];
 774         int fallback = nsock->fallback_index;
 775
 776         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 777                 return new_index;
 778
 779         if (config->num_connections <= 1) {
 780                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 781                                     "Attempted send on invalid socket\n");
 782                 return new_index;
 783         }
 784
 785         if (fallback >= 0 && fallback < config->num_connections &&
 786             !config->socks[fallback]->dead)
 787                 return fallback;
 788
 789         if (nsock->fallback_index < 0 ||
 790             nsock->fallback_index >= config->num_connections ||
 791             config->socks[nsock->fallback_index]->dead) {
 792                 int i;
 793                 for (i = 0; i < config->num_connections; i++) {
 794                         if (i == index)
 795                                 continue;
 796                         if (!config->socks[i]->dead) {
 797                                 new_index = i;
 798                                 break;
 799                         }
 800                 }
 801                 nsock->fallback_index = new_index;
 802                 if (new_index < 0) {
 803                         dev_err_ratelimited(disk_to_dev(nbd->disk),
 804                                             "Dead connection, failed to find a fallback\n");
 805                         return new_index;
 806                 }
 807         }
 808         new_index = nsock->fallback_index;
 809         return new_index;
 810 }
 811
 812 static int wait_for_reconnect(struct nbd_device *nbd)
 813 {
 814         struct nbd_config *config = nbd->config;
 815         if (!config->dead_conn_timeout)
 816                 return 0;
 817         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 818                 return 0;
 819         return wait_event_timeout(config->conn_wait,
 820                                   atomic_read(&config->live_connections) > 0,
 821                                   config->dead_conn_timeout) > 0;
 822 }
 823
 824 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 825 {
 826         struct request *req = blk_mq_rq_from_pdu(cmd);
 827         struct nbd_device *nbd = cmd->nbd;
 828         struct nbd_config *config;
 829         struct nbd_sock *nsock;
 830         int ret;
 831
 832         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 833                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 834                                     "Socks array is empty\n");
 835                 blk_mq_start_request(req);
 836                 return -EINVAL;
 837         }
 838         config = nbd->config;
 839
 840         if (index >= config->num_connections) {
 841                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 842                                     "Attempted send on invalid socket\n");
 843                 nbd_config_put(nbd);
 844                 blk_mq_start_request(req);
 845                 return -EINVAL;
 846         }
 847         cmd->status = BLK_STS_OK;
 848 again:
 849         nsock = config->socks[index];
 850         mutex_lock(&nsock->tx_lock);
 851         if (nsock->dead) {
 852                 int old_index = index;
 853                 index = find_fallback(nbd, index);
 854                 mutex_unlock(&nsock->tx_lock);
 855                 if (index < 0) {
 856                         if (wait_for_reconnect(nbd)) {
 857                                 index = old_index;
 858                                 goto again;
 859                         }
 860                         /* All the sockets should already be down at this point,
 861                          * we just want to make sure that DISCONNECTED is set so
 862                          * any requests that come in that were queue'ed waiting
 863                          * for the reconnect timer don't trigger the timer again
 864                          * and instead just error out.
 865                          */
 866                         sock_shutdown(nbd);
 867                         nbd_config_put(nbd);
 868                         blk_mq_start_request(req);
 869                         return -EIO;
 870                 }
 871                 goto again;
 872         }
 873
 874         /* Handle the case that we have a pending request that was partially
 875          * transmitted that _has_ to be serviced first.  We need to call requeue
 876          * here so that it gets put _after_ the request that is already on the
 877          * dispatch list.
 878          */
 879         blk_mq_start_request(req);
 880         if (unlikely(nsock->pending && nsock->pending != req)) {
 881                 nbd_requeue_cmd(cmd);
 882                 ret = 0;
 883                 goto out;
 884         }
 885         /*
 886          * Some failures are related to the link going down, so anything that
 887          * returns EAGAIN can be retried on a different socket.
 888          */
 889         ret = nbd_send_cmd(nbd, cmd, index);
 890         if (ret == -EAGAIN) {
 891                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 892                                     "Request send failed, requeueing\n");
 893                 nbd_mark_nsock_dead(nbd, nsock, 1);
 894                 nbd_requeue_cmd(cmd);
 895                 ret = 0;
 896         }
 897 out:
 898         mutex_unlock(&nsock->tx_lock);
 899         nbd_config_put(nbd);
 900         return ret;
 901 }
 902
 903 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 904                         const struct blk_mq_queue_data *bd)
 905 {
 906         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 907         int ret;
 908
 909         /*
 910          * Since we look at the bio's to send the request over the network we
 911          * need to make sure the completion work doesn't mark this request done
 912          * before we are done doing our send.  This keeps us from dereferencing
 913          * freed data if we have particularly fast completions (ie we get the
 914          * completion before we exit sock_xmit on the last bvec) or in the case
 915          * that the server is misbehaving (or there was an error) before we're
 916          * done sending everything over the wire.
 917          */
 918         mutex_lock(&cmd->lock);
 919         clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
 920
 921         /* We can be called directly from the user space process, which means we
 922          * could possibly have signals pending so our sendmsg will fail.  In
 923          * this case we need to return that we are busy, otherwise error out as
 924          * appropriate.
 925          */
 926         ret = nbd_handle_cmd(cmd, hctx->queue_num);
 927         if (ret < 0)
 928                 ret = BLK_STS_IOERR;
 929         else if (!ret)
 930                 ret = BLK_STS_OK;
 931         mutex_unlock(&cmd->lock);
 932
 933         return ret;
 934 }
 935
 936 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
 937                           bool netlink)
 938 {
 939         struct nbd_config *config = nbd->config;
 940         struct socket *sock;
 941         struct nbd_sock **socks;
 942         struct nbd_sock *nsock;
 943         int err;
 944
 945         sock = sockfd_lookup(arg, &err);
 946         if (!sock)
 947                 return err;
 948
 949         if (!netlink && !nbd->task_setup &&
 950             !test_bit(NBD_BOUND, &config->runtime_flags))
 951                 nbd->task_setup = current;
 952
 953         if (!netlink &&
 954             (nbd->task_setup != current ||
 955              test_bit(NBD_BOUND, &config->runtime_flags))) {
 956                 dev_err(disk_to_dev(nbd->disk),
 957                         "Device being setup by another task");
 958                 sockfd_put(sock);
 959                 return -EBUSY;
 960         }
 961
 962         socks = krealloc(config->socks, (config->num_connections + 1) *
 963                          sizeof(struct nbd_sock *), GFP_KERNEL);
 964         if (!socks) {
 965                 sockfd_put(sock);
 966                 return -ENOMEM;
 967         }
 968         nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
 969         if (!nsock) {
 970                 sockfd_put(sock);
 971                 return -ENOMEM;
 972         }
 973
 974         config->socks = socks;
 975
 976         nsock->fallback_index = -1;
 977         nsock->dead = false;
 978         mutex_init(&nsock->tx_lock);
 979         nsock->sock = sock;
 980         nsock->pending = NULL;
 981         nsock->sent = 0;
 982         nsock->cookie = 0;
 983         socks[config->num_connections++] = nsock;
 984         atomic_inc(&config->live_connections);
 985
 986         return 0;
 987 }
 988
 989 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
 990 {
 991         struct nbd_config *config = nbd->config;
 992         struct socket *sock, *old;
 993         struct recv_thread_args *args;
 994         int i;
 995         int err;
 996
 997         sock = sockfd_lookup(arg, &err);
 998         if (!sock)
 999                 return err;
1000
1001         args = kzalloc(sizeof(*args), GFP_KERNEL);
1002         if (!args) {
1003                 sockfd_put(sock);
1004                 return -ENOMEM;
1005         }
1006
1007         for (i = 0; i < config->num_connections; i++) {
1008                 struct nbd_sock *nsock = config->socks[i];
1009
1010                 if (!nsock->dead)
1011                         continue;
1012
1013                 mutex_lock(&nsock->tx_lock);
1014                 if (!nsock->dead) {
1015                         mutex_unlock(&nsock->tx_lock);
1016                         continue;
1017                 }
1018                 sk_set_memalloc(sock->sk);
1019                 if (nbd->tag_set.timeout)
1020                         sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1021                 atomic_inc(&config->recv_threads);
1022                 refcount_inc(&nbd->config_refs);
1023                 old = nsock->sock;
1024                 nsock->fallback_index = -1;
1025                 nsock->sock = sock;
1026                 nsock->dead = false;
1027                 INIT_WORK(&args->work, recv_work);
1028                 args->index = i;
1029                 args->nbd = nbd;
1030                 nsock->cookie++;
1031                 mutex_unlock(&nsock->tx_lock);
1032                 sockfd_put(old);
1033
1034                 clear_bit(NBD_DISCONNECTED, &config->runtime_flags);
1035
1036                 /* We take the tx_mutex in an error path in the recv_work, so we
1037                  * need to queue_work outside of the tx_mutex.
1038                  */
1039                 queue_work(recv_workqueue, &args->work);
1040
1041                 atomic_inc(&config->live_connections);
1042                 wake_up(&config->conn_wait);
1043                 return 0;
1044         }
1045         sockfd_put(sock);
1046         kfree(args);
1047         return -ENOSPC;
1048 }
1049
1050 static void nbd_bdev_reset(struct block_device *bdev)
1051 {
1052         if (bdev->bd_openers > 1)
1053                 return;
1054         bd_set_size(bdev, 0);
1055 }
1056
1057 static void nbd_parse_flags(struct nbd_device *nbd)
1058 {
1059         struct nbd_config *config = nbd->config;
1060         if (config->flags & NBD_FLAG_READ_ONLY)
1061                 set_disk_ro(nbd->disk, true);
1062         else
1063                 set_disk_ro(nbd->disk, false);
1064         if (config->flags & NBD_FLAG_SEND_TRIM)
1065                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1066         if (config->flags & NBD_FLAG_SEND_FLUSH) {
1067                 if (config->flags & NBD_FLAG_SEND_FUA)
1068                         blk_queue_write_cache(nbd->disk->queue, true, true);
1069                 else
1070                         blk_queue_write_cache(nbd->disk->queue, true, false);
1071         }
1072         else
1073                 blk_queue_write_cache(nbd->disk->queue, false, false);
1074 }
1075
1076 static void send_disconnects(struct nbd_device *nbd)
1077 {
1078         struct nbd_config *config = nbd->config;
1079         struct nbd_request request = {
1080                 .magic = htonl(NBD_REQUEST_MAGIC),
1081                 .type = htonl(NBD_CMD_DISC),
1082         };
1083         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1084         struct iov_iter from;
1085         int i, ret;
1086
1087         for (i = 0; i < config->num_connections; i++) {
1088                 struct nbd_sock *nsock = config->socks[i];
1089
1090                 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1091                 mutex_lock(&nsock->tx_lock);
1092                 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1093                 if (ret <= 0)
1094                         dev_err(disk_to_dev(nbd->disk),
1095                                 "Send disconnect failed %d\n", ret);
1096                 mutex_unlock(&nsock->tx_lock);
1097         }
1098 }
1099
1100 static int nbd_disconnect(struct nbd_device *nbd)
1101 {
1102         struct nbd_config *config = nbd->config;
1103
1104         dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1105         set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
1106         send_disconnects(nbd);
1107         return 0;
1108 }
1109
1110 static void nbd_clear_sock(struct nbd_device *nbd)
1111 {
1112         sock_shutdown(nbd);
1113         nbd_clear_que(nbd);
1114         nbd->task_setup = NULL;
1115 }
1116
1117 static void nbd_config_put(struct nbd_device *nbd)
1118 {
1119         if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1120                                         &nbd->config_lock)) {
1121                 struct nbd_config *config = nbd->config;
1122                 nbd_dev_dbg_close(nbd);
1123                 nbd_size_clear(nbd);
1124                 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1125                                        &config->runtime_flags))
1126                         device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1127                 nbd->task_recv = NULL;
1128                 nbd_clear_sock(nbd);
1129                 if (config->num_connections) {
1130                         int i;
1131                         for (i = 0; i < config->num_connections; i++) {
1132                                 sockfd_put(config->socks[i]->sock);
1133                                 kfree(config->socks[i]);
1134                         }
1135                         kfree(config->socks);
1136                 }
1137                 kfree(nbd->config);
1138                 nbd->config = NULL;
1139
1140                 nbd->tag_set.timeout = 0;
1141                 nbd->disk->queue->limits.discard_granularity = 0;
1142                 nbd->disk->queue->limits.discard_alignment = 0;
1143                 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1144                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1145
1146                 mutex_unlock(&nbd->config_lock);
1147                 nbd_put(nbd);
1148                 module_put(THIS_MODULE);
1149         }
1150 }
1151
1152 static int nbd_start_device(struct nbd_device *nbd)
1153 {
1154         struct nbd_config *config = nbd->config;
1155         int num_connections = config->num_connections;
1156         int error = 0, i;
1157
1158         if (nbd->task_recv)
1159                 return -EBUSY;
1160         if (!config->socks)
1161                 return -EINVAL;
1162         if (num_connections > 1 &&
1163             !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1164                 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1165                 return -EINVAL;
1166         }
1167
1168         blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1169         nbd->task_recv = current;
1170
1171         nbd_parse_flags(nbd);
1172
1173         error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1174         if (error) {
1175                 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1176                 return error;
1177         }
1178         set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1179
1180         nbd_dev_dbg_init(nbd);
1181         for (i = 0; i < num_connections; i++) {
1182                 struct recv_thread_args *args;
1183
1184                 args = kzalloc(sizeof(*args), GFP_KERNEL);
1185                 if (!args) {
1186                         sock_shutdown(nbd);
1187                         return -ENOMEM;
1188                 }
1189                 sk_set_memalloc(config->socks[i]->sock->sk);
1190                 if (nbd->tag_set.timeout)
1191                         config->socks[i]->sock->sk->sk_sndtimeo =
1192                                 nbd->tag_set.timeout;
1193                 atomic_inc(&config->recv_threads);
1194                 refcount_inc(&nbd->config_refs);
1195                 INIT_WORK(&args->work, recv_work);
1196                 args->nbd = nbd;
1197                 args->index = i;
1198                 queue_work(recv_workqueue, &args->work);
1199         }
1200         nbd_size_update(nbd);
1201         return error;
1202 }
1203
1204 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1205 {
1206         struct nbd_config *config = nbd->config;
1207         int ret;
1208
1209         ret = nbd_start_device(nbd);
1210         if (ret)
1211                 return ret;
1212
1213         if (max_part)
1214                 bdev->bd_invalidated = 1;
1215         mutex_unlock(&nbd->config_lock);
1216         ret = wait_event_interruptible(config->recv_wq,
1217                                          atomic_read(&config->recv_threads) == 0);
1218         if (ret)
1219                 sock_shutdown(nbd);
1220         mutex_lock(&nbd->config_lock);
1221         nbd_bdev_reset(bdev);
1222         /* user requested, ignore socket errors */
1223         if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1224                 ret = 0;
1225         if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1226                 ret = -ETIMEDOUT;
1227         return ret;
1228 }
1229
1230 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1231                                  struct block_device *bdev)
1232 {
1233         sock_shutdown(nbd);
1234         __invalidate_device(bdev, true);
1235         nbd_bdev_reset(bdev);
1236         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1237                                &nbd->config->runtime_flags))
1238                 nbd_config_put(nbd);
1239 }
1240
1241 static bool nbd_is_valid_blksize(unsigned long blksize)
1242 {
1243         if (!blksize || !is_power_of_2(blksize) || blksize < 512 ||
1244             blksize > PAGE_SIZE)
1245                 return false;
1246         return true;
1247 }
1248
1249 /* Must be called with config_lock held */
1250 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1251                        unsigned int cmd, unsigned long arg)
1252 {
1253         struct nbd_config *config = nbd->config;
1254
1255         switch (cmd) {
1256         case NBD_DISCONNECT:
1257                 return nbd_disconnect(nbd);
1258         case NBD_CLEAR_SOCK:
1259                 nbd_clear_sock_ioctl(nbd, bdev);
1260                 return 0;
1261         case NBD_SET_SOCK:
1262                 return nbd_add_socket(nbd, arg, false);
1263         case NBD_SET_BLKSIZE:
1264                 if (!arg)
1265                         arg = NBD_DEF_BLKSIZE;
1266                 if (!nbd_is_valid_blksize(arg))
1267                         return -EINVAL;
1268                 nbd_size_set(nbd, arg,
1269                              div_s64(config->bytesize, arg));
1270                 return 0;
1271         case NBD_SET_SIZE:
1272                 nbd_size_set(nbd, config->blksize,
1273                              div_s64(arg, config->blksize));
1274                 return 0;
1275         case NBD_SET_SIZE_BLOCKS:
1276                 nbd_size_set(nbd, config->blksize, arg);
1277                 return 0;
1278         case NBD_SET_TIMEOUT:
1279                 if (arg) {
1280                         nbd->tag_set.timeout = arg * HZ;
1281                         blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1282                 }
1283                 return 0;
1284
1285         case NBD_SET_FLAGS:
1286                 config->flags = arg;
1287                 return 0;
1288         case NBD_DO_IT:
1289                 return nbd_start_device_ioctl(nbd, bdev);
1290         case NBD_CLEAR_QUE:
1291                 /*
1292                  * This is for compatibility only.  The queue is always cleared
1293                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
1294                  */
1295                 return 0;
1296         case NBD_PRINT_DEBUG:
1297                 /*
1298                  * For compatibility only, we no longer keep a list of
1299                  * outstanding requests.
1300                  */
1301                 return 0;
1302         }
1303         return -ENOTTY;
1304 }
1305
1306 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1307                      unsigned int cmd, unsigned long arg)
1308 {
1309         struct nbd_device *nbd = bdev->bd_disk->private_data;
1310         struct nbd_config *config = nbd->config;
1311         int error = -EINVAL;
1312
1313         if (!capable(CAP_SYS_ADMIN))
1314                 return -EPERM;
1315
1316         /* The block layer will pass back some non-nbd ioctls in case we have
1317          * special handling for them, but we don't so just return an error.
1318          */
1319         if (_IOC_TYPE(cmd) != 0xab)
1320                 return -EINVAL;
1321
1322         mutex_lock(&nbd->config_lock);
1323
1324         /* Don't allow ioctl operations on a nbd device that was created with
1325          * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1326          */
1327         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1328             (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1329                 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1330         else
1331                 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1332         mutex_unlock(&nbd->config_lock);
1333         return error;
1334 }
1335
1336 static struct nbd_config *nbd_alloc_config(void)
1337 {
1338         struct nbd_config *config;
1339
1340         config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1341         if (!config)
1342                 return NULL;
1343         atomic_set(&config->recv_threads, 0);
1344         init_waitqueue_head(&config->recv_wq);
1345         init_waitqueue_head(&config->conn_wait);
1346         config->blksize = NBD_DEF_BLKSIZE;
1347         atomic_set(&config->live_connections, 0);
1348         try_module_get(THIS_MODULE);
1349         return config;
1350 }
1351
1352 static int nbd_open(struct block_device *bdev, fmode_t mode)
1353 {
1354         struct nbd_device *nbd;
1355         int ret = 0;
1356
1357         mutex_lock(&nbd_index_mutex);
1358         nbd = bdev->bd_disk->private_data;
1359         if (!nbd) {
1360                 ret = -ENXIO;
1361                 goto out;
1362         }
1363         if (!refcount_inc_not_zero(&nbd->refs)) {
1364                 ret = -ENXIO;
1365                 goto out;
1366         }
1367         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1368                 struct nbd_config *config;
1369
1370                 mutex_lock(&nbd->config_lock);
1371                 if (refcount_inc_not_zero(&nbd->config_refs)) {
1372                         mutex_unlock(&nbd->config_lock);
1373                         goto out;
1374                 }
1375                 config = nbd->config = nbd_alloc_config();
1376                 if (!config) {
1377                         ret = -ENOMEM;
1378                         mutex_unlock(&nbd->config_lock);
1379                         goto out;
1380                 }
1381                 refcount_set(&nbd->config_refs, 1);
1382                 refcount_inc(&nbd->refs);
1383                 mutex_unlock(&nbd->config_lock);
1384                 bdev->bd_invalidated = 1;
1385         } else if (nbd_disconnected(nbd->config)) {
1386                 bdev->bd_invalidated = 1;
1387         }
1388 out:
1389         mutex_unlock(&nbd_index_mutex);
1390         return ret;
1391 }
1392
1393 static void nbd_release(struct gendisk *disk, fmode_t mode)
1394 {
1395         struct nbd_device *nbd = disk->private_data;
1396         struct block_device *bdev = bdget_disk(disk, 0);
1397
1398         if (test_bit(NBD_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1399                         bdev->bd_openers == 0)
1400                 nbd_disconnect_and_put(nbd);
1401
1402         nbd_config_put(nbd);
1403         nbd_put(nbd);
1404 }
1405
1406 static const struct block_device_operations nbd_fops =
1407 {
1408         .owner =        THIS_MODULE,
1409         .open =         nbd_open,
1410         .release =      nbd_release,
1411         .ioctl =        nbd_ioctl,
1412         .compat_ioctl = nbd_ioctl,
1413 };
1414
1415 #if IS_ENABLED(CONFIG_DEBUG_FS)
1416
1417 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1418 {
1419         struct nbd_device *nbd = s->private;
1420
1421         if (nbd->task_recv)
1422                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1423
1424         return 0;
1425 }
1426
1427 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1428 {
1429         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1430 }
1431
1432 static const struct file_operations nbd_dbg_tasks_ops = {
1433         .open = nbd_dbg_tasks_open,
1434         .read = seq_read,
1435         .llseek = seq_lseek,
1436         .release = single_release,
1437 };
1438
1439 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1440 {
1441         struct nbd_device *nbd = s->private;
1442         u32 flags = nbd->config->flags;
1443
1444         seq_printf(s, "Hex: 0x%08x\n\n", flags);
1445
1446         seq_puts(s, "Known flags:\n");
1447
1448         if (flags & NBD_FLAG_HAS_FLAGS)
1449                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1450         if (flags & NBD_FLAG_READ_ONLY)
1451                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1452         if (flags & NBD_FLAG_SEND_FLUSH)
1453                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1454         if (flags & NBD_FLAG_SEND_FUA)
1455                 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1456         if (flags & NBD_FLAG_SEND_TRIM)
1457                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1458
1459         return 0;
1460 }
1461
1462 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1463 {
1464         return single_open(file, nbd_dbg_flags_show, inode->i_private);
1465 }
1466
1467 static const struct file_operations nbd_dbg_flags_ops = {
1468         .open = nbd_dbg_flags_open,
1469         .read = seq_read,
1470         .llseek = seq_lseek,
1471         .release = single_release,
1472 };
1473
1474 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1475 {
1476         struct dentry *dir;
1477         struct nbd_config *config = nbd->config;
1478
1479         if (!nbd_dbg_dir)
1480                 return -EIO;
1481
1482         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1483         if (!dir) {
1484                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1485                         nbd_name(nbd));
1486                 return -EIO;
1487         }
1488         config->dbg_dir = dir;
1489
1490         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1491         debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1492         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1493         debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1494         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1495
1496         return 0;
1497 }
1498
1499 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1500 {
1501         debugfs_remove_recursive(nbd->config->dbg_dir);
1502 }
1503
1504 static int nbd_dbg_init(void)
1505 {
1506         struct dentry *dbg_dir;
1507
1508         dbg_dir = debugfs_create_dir("nbd", NULL);
1509         if (!dbg_dir)
1510                 return -EIO;
1511
1512         nbd_dbg_dir = dbg_dir;
1513
1514         return 0;
1515 }
1516
1517 static void nbd_dbg_close(void)
1518 {
1519         debugfs_remove_recursive(nbd_dbg_dir);
1520 }
1521
1522 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1523
1524 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1525 {
1526         return 0;
1527 }
1528
1529 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1530 {
1531 }
1532
1533 static int nbd_dbg_init(void)
1534 {
1535         return 0;
1536 }
1537
1538 static void nbd_dbg_close(void)
1539 {
1540 }
1541
1542 #endif
1543
1544 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1545                             unsigned int hctx_idx, unsigned int numa_node)
1546 {
1547         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1548         cmd->nbd = set->driver_data;
1549         cmd->flags = 0;
1550         mutex_init(&cmd->lock);
1551         return 0;
1552 }
1553
1554 static const struct blk_mq_ops nbd_mq_ops = {
1555         .queue_rq       = nbd_queue_rq,
1556         .complete       = nbd_complete_rq,
1557         .init_request   = nbd_init_request,
1558         .timeout        = nbd_xmit_timeout,
1559 };
1560
1561 static int nbd_dev_add(int index)
1562 {
1563         struct nbd_device *nbd;
1564         struct gendisk *disk;
1565         struct request_queue *q;
1566         int err = -ENOMEM;
1567
1568         nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1569         if (!nbd)
1570                 goto out;
1571
1572         disk = alloc_disk(1 << part_shift);
1573         if (!disk)
1574                 goto out_free_nbd;
1575
1576         if (index >= 0) {
1577                 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1578                                 GFP_KERNEL);
1579                 if (err == -ENOSPC)
1580                         err = -EEXIST;
1581         } else {
1582                 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1583                 if (err >= 0)
1584                         index = err;
1585         }
1586         if (err < 0)
1587                 goto out_free_disk;
1588
1589         nbd->index = index;
1590         nbd->disk = disk;
1591         nbd->tag_set.ops = &nbd_mq_ops;
1592         nbd->tag_set.nr_hw_queues = 1;
1593         nbd->tag_set.queue_depth = 128;
1594         nbd->tag_set.numa_node = NUMA_NO_NODE;
1595         nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1596         nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1597                 BLK_MQ_F_BLOCKING;
1598         nbd->tag_set.driver_data = nbd;
1599
1600         err = blk_mq_alloc_tag_set(&nbd->tag_set);
1601         if (err)
1602                 goto out_free_idr;
1603
1604         q = blk_mq_init_queue(&nbd->tag_set);
1605         if (IS_ERR(q)) {
1606                 err = PTR_ERR(q);
1607                 goto out_free_tags;
1608         }
1609         disk->queue = q;
1610
1611         /*
1612          * Tell the block layer that we are not a rotational device
1613          */
1614         blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1615         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1616         disk->queue->limits.discard_granularity = 0;
1617         disk->queue->limits.discard_alignment = 0;
1618         blk_queue_max_discard_sectors(disk->queue, 0);
1619         blk_queue_max_segment_size(disk->queue, UINT_MAX);
1620         blk_queue_max_segments(disk->queue, USHRT_MAX);
1621         blk_queue_max_hw_sectors(disk->queue, 65536);
1622         disk->queue->limits.max_sectors = 256;
1623
1624         mutex_init(&nbd->config_lock);
1625         refcount_set(&nbd->config_refs, 0);
1626         refcount_set(&nbd->refs, 1);
1627         INIT_LIST_HEAD(&nbd->list);
1628         disk->major = NBD_MAJOR;
1629         disk->first_minor = index << part_shift;
1630         disk->fops = &nbd_fops;
1631         disk->private_data = nbd;
1632         sprintf(disk->disk_name, "nbd%d", index);
1633         add_disk(disk);
1634         nbd_total_devices++;
1635         return index;
1636
1637 out_free_tags:
1638         blk_mq_free_tag_set(&nbd->tag_set);
1639 out_free_idr:
1640         idr_remove(&nbd_index_idr, index);
1641 out_free_disk:
1642         put_disk(disk);
1643 out_free_nbd:
1644         kfree(nbd);
1645 out:
1646         return err;
1647 }
1648
1649 static int find_free_cb(int id, void *ptr, void *data)
1650 {
1651         struct nbd_device *nbd = ptr;
1652         struct nbd_device **found = data;
1653
1654         if (!refcount_read(&nbd->config_refs)) {
1655                 *found = nbd;
1656                 return 1;
1657         }
1658         return 0;
1659 }
1660
1661 /* Netlink interface. */
1662 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1663         [NBD_ATTR_INDEX]                =       { .type = NLA_U32 },
1664         [NBD_ATTR_SIZE_BYTES]           =       { .type = NLA_U64 },
1665         [NBD_ATTR_BLOCK_SIZE_BYTES]     =       { .type = NLA_U64 },
1666         [NBD_ATTR_TIMEOUT]              =       { .type = NLA_U64 },
1667         [NBD_ATTR_SERVER_FLAGS]         =       { .type = NLA_U64 },
1668         [NBD_ATTR_CLIENT_FLAGS]         =       { .type = NLA_U64 },
1669         [NBD_ATTR_SOCKETS]              =       { .type = NLA_NESTED},
1670         [NBD_ATTR_DEAD_CONN_TIMEOUT]    =       { .type = NLA_U64 },
1671         [NBD_ATTR_DEVICE_LIST]          =       { .type = NLA_NESTED},
1672 };
1673
1674 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1675         [NBD_SOCK_FD]                   =       { .type = NLA_U32 },
1676 };
1677
1678 /* We don't use this right now since we don't parse the incoming list, but we
1679  * still want it here so userspace knows what to expect.
1680  */
1681 static const struct nla_policy __attribute__((unused))
1682 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1683         [NBD_DEVICE_INDEX]              =       { .type = NLA_U32 },
1684         [NBD_DEVICE_CONNECTED]          =       { .type = NLA_U8 },
1685 };
1686
1687 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1688 {
1689         struct nbd_config *config = nbd->config;
1690         u64 bsize = config->blksize;
1691         u64 bytes = config->bytesize;
1692
1693         if (info->attrs[NBD_ATTR_SIZE_BYTES])
1694                 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1695
1696         if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1697                 bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1698                 if (!bsize)
1699                         bsize = NBD_DEF_BLKSIZE;
1700                 if (!nbd_is_valid_blksize(bsize)) {
1701                         printk(KERN_ERR "Invalid block size %llu\n", bsize);
1702                         return -EINVAL;
1703                 }
1704         }
1705
1706         if (bytes != config->bytesize || bsize != config->blksize)
1707                 nbd_size_set(nbd, bsize, div64_u64(bytes, bsize));
1708         return 0;
1709 }
1710
1711 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1712 {
1713         struct nbd_device *nbd = NULL;
1714         struct nbd_config *config;
1715         int index = -1;
1716         int ret;
1717         bool put_dev = false;
1718
1719         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1720                 return -EPERM;
1721
1722         if (info->attrs[NBD_ATTR_INDEX])
1723                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1724         if (!info->attrs[NBD_ATTR_SOCKETS]) {
1725                 printk(KERN_ERR "nbd: must specify at least one socket\n");
1726                 return -EINVAL;
1727         }
1728         if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1729                 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1730                 return -EINVAL;
1731         }
1732 again:
1733         mutex_lock(&nbd_index_mutex);
1734         if (index == -1) {
1735                 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1736                 if (ret == 0) {
1737                         int new_index;
1738                         new_index = nbd_dev_add(-1);
1739                         if (new_index < 0) {
1740                                 mutex_unlock(&nbd_index_mutex);
1741                                 printk(KERN_ERR "nbd: failed to add new device\n");
1742                                 return new_index;
1743                         }
1744                         nbd = idr_find(&nbd_index_idr, new_index);
1745                 }
1746         } else {
1747                 nbd = idr_find(&nbd_index_idr, index);
1748                 if (!nbd) {
1749                         ret = nbd_dev_add(index);
1750                         if (ret < 0) {
1751                                 mutex_unlock(&nbd_index_mutex);
1752                                 printk(KERN_ERR "nbd: failed to add new device\n");
1753                                 return ret;
1754                         }
1755                         nbd = idr_find(&nbd_index_idr, index);
1756                 }
1757         }
1758         if (!nbd) {
1759                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1760                        index);
1761                 mutex_unlock(&nbd_index_mutex);
1762                 return -EINVAL;
1763         }
1764         if (!refcount_inc_not_zero(&nbd->refs)) {
1765                 mutex_unlock(&nbd_index_mutex);
1766                 if (index == -1)
1767                         goto again;
1768                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1769                        index);
1770                 return -EINVAL;
1771         }
1772         mutex_unlock(&nbd_index_mutex);
1773
1774         mutex_lock(&nbd->config_lock);
1775         if (refcount_read(&nbd->config_refs)) {
1776                 mutex_unlock(&nbd->config_lock);
1777                 nbd_put(nbd);
1778                 if (index == -1)
1779                         goto again;
1780                 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1781                 return -EBUSY;
1782         }
1783         if (WARN_ON(nbd->config)) {
1784                 mutex_unlock(&nbd->config_lock);
1785                 nbd_put(nbd);
1786                 return -EINVAL;
1787         }
1788         config = nbd->config = nbd_alloc_config();
1789         if (!nbd->config) {
1790                 mutex_unlock(&nbd->config_lock);
1791                 nbd_put(nbd);
1792                 printk(KERN_ERR "nbd: couldn't allocate config\n");
1793                 return -ENOMEM;
1794         }
1795         refcount_set(&nbd->config_refs, 1);
1796         set_bit(NBD_BOUND, &config->runtime_flags);
1797
1798         ret = nbd_genl_size_set(info, nbd);
1799         if (ret)
1800                 goto out;
1801
1802         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1803                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1804                 nbd->tag_set.timeout = timeout * HZ;
1805                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1806         }
1807         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1808                 config->dead_conn_timeout =
1809                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1810                 config->dead_conn_timeout *= HZ;
1811         }
1812         if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1813                 config->flags =
1814                         nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1815         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1816                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1817                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1818                         set_bit(NBD_DESTROY_ON_DISCONNECT,
1819                                 &config->runtime_flags);
1820                         put_dev = true;
1821                 }
1822                 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1823                         set_bit(NBD_DISCONNECT_ON_CLOSE,
1824                                 &config->runtime_flags);
1825                 }
1826         }
1827
1828         if (info->attrs[NBD_ATTR_SOCKETS]) {
1829                 struct nlattr *attr;
1830                 int rem, fd;
1831
1832                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1833                                     rem) {
1834                         struct nlattr *socks[NBD_SOCK_MAX+1];
1835
1836                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1837                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1838                                 ret = -EINVAL;
1839                                 goto out;
1840                         }
1841                         ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1842                                                           attr,
1843                                                           nbd_sock_policy,
1844                                                           info->extack);
1845                         if (ret != 0) {
1846                                 printk(KERN_ERR "nbd: error processing sock list\n");
1847                                 ret = -EINVAL;
1848                                 goto out;
1849                         }
1850                         if (!socks[NBD_SOCK_FD])
1851                                 continue;
1852                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1853                         ret = nbd_add_socket(nbd, fd, true);
1854                         if (ret)
1855                                 goto out;
1856                 }
1857         }
1858         ret = nbd_start_device(nbd);
1859 out:
1860         mutex_unlock(&nbd->config_lock);
1861         if (!ret) {
1862                 set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1863                 refcount_inc(&nbd->config_refs);
1864                 nbd_connect_reply(info, nbd->index);
1865         }
1866         nbd_config_put(nbd);
1867         if (put_dev)
1868                 nbd_put(nbd);
1869         return ret;
1870 }
1871
1872 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1873 {
1874         mutex_lock(&nbd->config_lock);
1875         nbd_disconnect(nbd);
1876         nbd_clear_sock(nbd);
1877         mutex_unlock(&nbd->config_lock);
1878         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1879                                &nbd->config->runtime_flags))
1880                 nbd_config_put(nbd);
1881 }
1882
1883 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1884 {
1885         struct nbd_device *nbd;
1886         int index;
1887
1888         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1889                 return -EPERM;
1890
1891         if (!info->attrs[NBD_ATTR_INDEX]) {
1892                 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1893                 return -EINVAL;
1894         }
1895         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1896         mutex_lock(&nbd_index_mutex);
1897         nbd = idr_find(&nbd_index_idr, index);
1898         if (!nbd) {
1899                 mutex_unlock(&nbd_index_mutex);
1900                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1901                        index);
1902                 return -EINVAL;
1903         }
1904         if (!refcount_inc_not_zero(&nbd->refs)) {
1905                 mutex_unlock(&nbd_index_mutex);
1906                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1907                        index);
1908                 return -EINVAL;
1909         }
1910         mutex_unlock(&nbd_index_mutex);
1911         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1912                 nbd_put(nbd);
1913                 return 0;
1914         }
1915         nbd_disconnect_and_put(nbd);
1916         nbd_config_put(nbd);
1917         nbd_put(nbd);
1918         return 0;
1919 }
1920
1921 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1922 {
1923         struct nbd_device *nbd = NULL;
1924         struct nbd_config *config;
1925         int index;
1926         int ret = 0;
1927         bool put_dev = false;
1928
1929         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1930                 return -EPERM;
1931
1932         if (!info->attrs[NBD_ATTR_INDEX]) {
1933                 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1934                 return -EINVAL;
1935         }
1936         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1937         mutex_lock(&nbd_index_mutex);
1938         nbd = idr_find(&nbd_index_idr, index);
1939         if (!nbd) {
1940                 mutex_unlock(&nbd_index_mutex);
1941                 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1942                        index);
1943                 return -EINVAL;
1944         }
1945         if (!refcount_inc_not_zero(&nbd->refs)) {
1946                 mutex_unlock(&nbd_index_mutex);
1947                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1948                        index);
1949                 return -EINVAL;
1950         }
1951         mutex_unlock(&nbd_index_mutex);
1952
1953         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1954                 dev_err(nbd_to_dev(nbd),
1955                         "not configured, cannot reconfigure\n");
1956                 nbd_put(nbd);
1957                 return -EINVAL;
1958         }
1959
1960         mutex_lock(&nbd->config_lock);
1961         config = nbd->config;
1962         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1963             !nbd->task_recv) {
1964                 dev_err(nbd_to_dev(nbd),
1965                         "not configured, cannot reconfigure\n");
1966                 ret = -EINVAL;
1967                 goto out;
1968         }
1969
1970         ret = nbd_genl_size_set(info, nbd);
1971         if (ret)
1972                 goto out;
1973
1974         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1975                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1976                 nbd->tag_set.timeout = timeout * HZ;
1977                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1978         }
1979         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1980                 config->dead_conn_timeout =
1981                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1982                 config->dead_conn_timeout *= HZ;
1983         }
1984         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1985                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1986                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1987                         if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1988                                               &config->runtime_flags))
1989                                 put_dev = true;
1990                 } else {
1991                         if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1992                                                &config->runtime_flags))
1993                                 refcount_inc(&nbd->refs);
1994                 }
1995
1996                 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1997                         set_bit(NBD_DISCONNECT_ON_CLOSE,
1998                                         &config->runtime_flags);
1999                 } else {
2000                         clear_bit(NBD_DISCONNECT_ON_CLOSE,
2001                                         &config->runtime_flags);
2002                 }
2003         }
2004
2005         if (info->attrs[NBD_ATTR_SOCKETS]) {
2006                 struct nlattr *attr;
2007                 int rem, fd;
2008
2009                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2010                                     rem) {
2011                         struct nlattr *socks[NBD_SOCK_MAX+1];
2012
2013                         if (nla_type(attr) != NBD_SOCK_ITEM) {
2014                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2015                                 ret = -EINVAL;
2016                                 goto out;
2017                         }
2018                         ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2019                                                           attr,
2020                                                           nbd_sock_policy,
2021                                                           info->extack);
2022                         if (ret != 0) {
2023                                 printk(KERN_ERR "nbd: error processing sock list\n");
2024                                 ret = -EINVAL;
2025                                 goto out;
2026                         }
2027                         if (!socks[NBD_SOCK_FD])
2028                                 continue;
2029                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2030                         ret = nbd_reconnect_socket(nbd, fd);
2031                         if (ret) {
2032                                 if (ret == -ENOSPC)
2033                                         ret = 0;
2034                                 goto out;
2035                         }
2036                         dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2037                 }
2038         }
2039 out:
2040         mutex_unlock(&nbd->config_lock);
2041         nbd_config_put(nbd);
2042         nbd_put(nbd);
2043         if (put_dev)
2044                 nbd_put(nbd);
2045         return ret;
2046 }
2047
2048 static const struct genl_ops nbd_connect_genl_ops[] = {
2049         {
2050                 .cmd    = NBD_CMD_CONNECT,
2051                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2052                 .doit   = nbd_genl_connect,
2053         },
2054         {
2055                 .cmd    = NBD_CMD_DISCONNECT,
2056                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2057                 .doit   = nbd_genl_disconnect,
2058         },
2059         {
2060                 .cmd    = NBD_CMD_RECONFIGURE,
2061                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2062                 .doit   = nbd_genl_reconfigure,
2063         },
2064         {
2065                 .cmd    = NBD_CMD_STATUS,
2066                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2067                 .doit   = nbd_genl_status,
2068         },
2069 };
2070
2071 static const struct genl_multicast_group nbd_mcast_grps[] = {
2072         { .name = NBD_GENL_MCAST_GROUP_NAME, },
2073 };
2074
2075 static struct genl_family nbd_genl_family __ro_after_init = {
2076         .hdrsize        = 0,
2077         .name           = NBD_GENL_FAMILY_NAME,
2078         .version        = NBD_GENL_VERSION,
2079         .module         = THIS_MODULE,
2080         .ops            = nbd_connect_genl_ops,
2081         .n_ops          = ARRAY_SIZE(nbd_connect_genl_ops),
2082         .maxattr        = NBD_ATTR_MAX,
2083         .policy = nbd_attr_policy,
2084         .mcgrps         = nbd_mcast_grps,
2085         .n_mcgrps       = ARRAY_SIZE(nbd_mcast_grps),
2086 };
2087
2088 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2089 {
2090         struct nlattr *dev_opt;
2091         u8 connected = 0;
2092         int ret;
2093
2094         /* This is a little racey, but for status it's ok.  The
2095          * reason we don't take a ref here is because we can't
2096          * take a ref in the index == -1 case as we would need
2097          * to put under the nbd_index_mutex, which could
2098          * deadlock if we are configured to remove ourselves
2099          * once we're disconnected.
2100          */
2101         if (refcount_read(&nbd->config_refs))
2102                 connected = 1;
2103         dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2104         if (!dev_opt)
2105                 return -EMSGSIZE;
2106         ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2107         if (ret)
2108                 return -EMSGSIZE;
2109         ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2110                          connected);
2111         if (ret)
2112                 return -EMSGSIZE;
2113         nla_nest_end(reply, dev_opt);
2114         return 0;
2115 }
2116
2117 static int status_cb(int id, void *ptr, void *data)
2118 {
2119         struct nbd_device *nbd = ptr;
2120         return populate_nbd_status(nbd, (struct sk_buff *)data);
2121 }
2122
2123 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2124 {
2125         struct nlattr *dev_list;
2126         struct sk_buff *reply;
2127         void *reply_head;
2128         size_t msg_size;
2129         int index = -1;
2130         int ret = -ENOMEM;
2131
2132         if (info->attrs[NBD_ATTR_INDEX])
2133                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2134
2135         mutex_lock(&nbd_index_mutex);
2136
2137         msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2138                                   nla_attr_size(sizeof(u8)));
2139         msg_size *= (index == -1) ? nbd_total_devices : 1;
2140
2141         reply = genlmsg_new(msg_size, GFP_KERNEL);
2142         if (!reply)
2143                 goto out;
2144         reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2145                                        NBD_CMD_STATUS);
2146         if (!reply_head) {
2147                 nlmsg_free(reply);
2148                 goto out;
2149         }
2150
2151         dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2152         if (index == -1) {
2153                 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2154                 if (ret) {
2155                         nlmsg_free(reply);
2156                         goto out;
2157                 }
2158         } else {
2159                 struct nbd_device *nbd;
2160                 nbd = idr_find(&nbd_index_idr, index);
2161                 if (nbd) {
2162                         ret = populate_nbd_status(nbd, reply);
2163                         if (ret) {
2164                                 nlmsg_free(reply);
2165                                 goto out;
2166                         }
2167                 }
2168         }
2169         nla_nest_end(reply, dev_list);
2170         genlmsg_end(reply, reply_head);
2171         ret = genlmsg_reply(reply, info);
2172 out:
2173         mutex_unlock(&nbd_index_mutex);
2174         return ret;
2175 }
2176
2177 static void nbd_connect_reply(struct genl_info *info, int index)
2178 {
2179         struct sk_buff *skb;
2180         void *msg_head;
2181         int ret;
2182
2183         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2184         if (!skb)
2185                 return;
2186         msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2187                                      NBD_CMD_CONNECT);
2188         if (!msg_head) {
2189                 nlmsg_free(skb);
2190                 return;
2191         }
2192         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2193         if (ret) {
2194                 nlmsg_free(skb);
2195                 return;
2196         }
2197         genlmsg_end(skb, msg_head);
2198         genlmsg_reply(skb, info);
2199 }
2200
2201 static void nbd_mcast_index(int index)
2202 {
2203         struct sk_buff *skb;
2204         void *msg_head;
2205         int ret;
2206
2207         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2208         if (!skb)
2209                 return;
2210         msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2211                                      NBD_CMD_LINK_DEAD);
2212         if (!msg_head) {
2213                 nlmsg_free(skb);
2214                 return;
2215         }
2216         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2217         if (ret) {
2218                 nlmsg_free(skb);
2219                 return;
2220         }
2221         genlmsg_end(skb, msg_head);
2222         genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2223 }
2224
2225 static void nbd_dead_link_work(struct work_struct *work)
2226 {
2227         struct link_dead_args *args = container_of(work, struct link_dead_args,
2228                                                    work);
2229         nbd_mcast_index(args->index);
2230         kfree(args);
2231 }
2232
2233 static int __init nbd_init(void)
2234 {
2235         int i;
2236
2237         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2238
2239         if (max_part < 0) {
2240                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2241                 return -EINVAL;
2242         }
2243
2244         part_shift = 0;
2245         if (max_part > 0) {
2246                 part_shift = fls(max_part);
2247
2248                 /*
2249                  * Adjust max_part according to part_shift as it is exported
2250                  * to user space so that user can know the max number of
2251                  * partition kernel should be able to manage.
2252                  *
2253                  * Note that -1 is required because partition 0 is reserved
2254                  * for the whole disk.
2255                  */
2256                 max_part = (1UL << part_shift) - 1;
2257         }
2258
2259         if ((1UL << part_shift) > DISK_MAX_PARTS)
2260                 return -EINVAL;
2261
2262         if (nbds_max > 1UL << (MINORBITS - part_shift))
2263                 return -EINVAL;
2264         recv_workqueue = alloc_workqueue("knbd-recv",
2265                                          WQ_MEM_RECLAIM | WQ_HIGHPRI |
2266                                          WQ_UNBOUND, 0);
2267         if (!recv_workqueue)
2268                 return -ENOMEM;
2269
2270         if (register_blkdev(NBD_MAJOR, "nbd")) {
2271                 destroy_workqueue(recv_workqueue);
2272                 return -EIO;
2273         }
2274
2275         if (genl_register_family(&nbd_genl_family)) {
2276                 unregister_blkdev(NBD_MAJOR, "nbd");
2277                 destroy_workqueue(recv_workqueue);
2278                 return -EINVAL;
2279         }
2280         nbd_dbg_init();
2281
2282         mutex_lock(&nbd_index_mutex);
2283         for (i = 0; i < nbds_max; i++)
2284                 nbd_dev_add(i);
2285         mutex_unlock(&nbd_index_mutex);
2286         return 0;
2287 }
2288
2289 static int nbd_exit_cb(int id, void *ptr, void *data)
2290 {
2291         struct list_head *list = (struct list_head *)data;
2292         struct nbd_device *nbd = ptr;
2293
2294         list_add_tail(&nbd->list, list);
2295         return 0;
2296 }
2297
2298 static void __exit nbd_cleanup(void)
2299 {
2300         struct nbd_device *nbd;
2301         LIST_HEAD(del_list);
2302
2303         nbd_dbg_close();
2304
2305         mutex_lock(&nbd_index_mutex);
2306         idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2307         mutex_unlock(&nbd_index_mutex);
2308
2309         while (!list_empty(&del_list)) {
2310                 nbd = list_first_entry(&del_list, struct nbd_device, list);
2311                 list_del_init(&nbd->list);
2312                 if (refcount_read(&nbd->refs) != 1)
2313                         printk(KERN_ERR "nbd: possibly leaking a device\n");
2314                 nbd_put(nbd);
2315         }
2316
2317         idr_destroy(&nbd_index_idr);
2318         genl_unregister_family(&nbd_genl_family);
2319         destroy_workqueue(recv_workqueue);
2320         unregister_blkdev(NBD_MAJOR, "nbd");
2321 }
2322
2323 module_init(nbd_init);
2324 module_exit(nbd_cleanup);
2325
2326 MODULE_DESCRIPTION("Network Block Device");
2327 MODULE_LICENSE("GPL");
2328
2329 module_param(nbds_max, int, 0444);
2330 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2331 module_param(max_part, int, 0444);
2332 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");