drivers/block/nbd.c

   1 /*
   2  * Network block device - make block devices work over TCP
   3  *
   4  * Note that you can not swap over this thing, yet. Seems to work but
   5  * deadlocks sometimes - you can not swap over TCP in general.
   6  *
   7  * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
   8  * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
   9  *
  10  * This file is released under GPLv2 or later.
  11  *
  12  * (part of code stolen from loop.c)
  13  */
  14
  15 #include <linux/major.h>
  16
  17 #include <linux/blkdev.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/sched.h>
  21 #include <linux/sched/mm.h>
  22 #include <linux/fs.h>
  23 #include <linux/bio.h>
  24 #include <linux/stat.h>
  25 #include <linux/errno.h>
  26 #include <linux/file.h>
  27 #include <linux/ioctl.h>
  28 #include <linux/mutex.h>
  29 #include <linux/compiler.h>
  30 #include <linux/err.h>
  31 #include <linux/kernel.h>
  32 #include <linux/slab.h>
  33 #include <net/sock.h>
  34 #include <linux/net.h>
  35 #include <linux/kthread.h>
  36 #include <linux/types.h>
  37 #include <linux/debugfs.h>
  38 #include <linux/blk-mq.h>
  39
  40 #include <linux/uaccess.h>
  41 #include <asm/types.h>
  42
  43 #include <linux/nbd.h>
  44 #include <linux/nbd-netlink.h>
  45 #include <net/genetlink.h>
  46
  47 static DEFINE_IDR(nbd_index_idr);
  48 static DEFINE_MUTEX(nbd_index_mutex);
  49 static int nbd_total_devices = 0;
  50
  51 struct nbd_sock {
  52         struct socket *sock;
  53         struct mutex tx_lock;
  54         struct request *pending;
  55         int sent;
  56         bool dead;
  57         int fallback_index;
  58         int cookie;
  59 };
  60
  61 struct recv_thread_args {
  62         struct work_struct work;
  63         struct nbd_device *nbd;
  64         int index;
  65 };
  66
  67 struct link_dead_args {
  68         struct work_struct work;
  69         int index;
  70 };
  71
  72 #define NBD_TIMEDOUT                    0
  73 #define NBD_DISCONNECT_REQUESTED        1
  74 #define NBD_DISCONNECTED                2
  75 #define NBD_HAS_PID_FILE                3
  76 #define NBD_HAS_CONFIG_REF              4
  77 #define NBD_BOUND                       5
  78 #define NBD_DESTROY_ON_DISCONNECT       6
  79
  80 struct nbd_config {
  81         u32 flags;
  82         unsigned long runtime_flags;
  83         u64 dead_conn_timeout;
  84
  85         struct nbd_sock **socks;
  86         int num_connections;
  87         atomic_t live_connections;
  88         wait_queue_head_t conn_wait;
  89
  90         atomic_t recv_threads;
  91         wait_queue_head_t recv_wq;
  92         loff_t blksize;
  93         loff_t bytesize;
  94 #if IS_ENABLED(CONFIG_DEBUG_FS)
  95         struct dentry *dbg_dir;
  96 #endif
  97 };
  98
  99 struct nbd_device {
 100         struct blk_mq_tag_set tag_set;
 101
 102         int index;
 103         refcount_t config_refs;
 104         refcount_t refs;
 105         struct nbd_config *config;
 106         struct mutex config_lock;
 107         struct gendisk *disk;
 108
 109         struct list_head list;
 110         struct task_struct *task_recv;
 111         struct task_struct *task_setup;
 112 };
 113
 114 struct nbd_cmd {
 115         struct nbd_device *nbd;
 116         int index;
 117         int cookie;
 118         struct completion send_complete;
 119         blk_status_t status;
 120 };
 121
 122 #if IS_ENABLED(CONFIG_DEBUG_FS)
 123 static struct dentry *nbd_dbg_dir;
 124 #endif
 125
 126 #define nbd_name(nbd) ((nbd)->disk->disk_name)
 127
 128 #define NBD_MAGIC 0x68797548
 129
 130 static unsigned int nbds_max = 16;
 131 static int max_part;
 132 static struct workqueue_struct *recv_workqueue;
 133 static int part_shift;
 134
 135 static int nbd_dev_dbg_init(struct nbd_device *nbd);
 136 static void nbd_dev_dbg_close(struct nbd_device *nbd);
 137 static void nbd_config_put(struct nbd_device *nbd);
 138 static void nbd_connect_reply(struct genl_info *info, int index);
 139 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
 140 static void nbd_dead_link_work(struct work_struct *work);
 141
 142 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
 143 {
 144         return disk_to_dev(nbd->disk);
 145 }
 146
 147 static const char *nbdcmd_to_ascii(int cmd)
 148 {
 149         switch (cmd) {
 150         case  NBD_CMD_READ: return "read";
 151         case NBD_CMD_WRITE: return "write";
 152         case  NBD_CMD_DISC: return "disconnect";
 153         case NBD_CMD_FLUSH: return "flush";
 154         case  NBD_CMD_TRIM: return "trim/discard";
 155         }
 156         return "invalid";
 157 }
 158
 159 static ssize_t pid_show(struct device *dev,
 160                         struct device_attribute *attr, char *buf)
 161 {
 162         struct gendisk *disk = dev_to_disk(dev);
 163         struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
 164
 165         return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
 166 }
 167
 168 static struct device_attribute pid_attr = {
 169         .attr = { .name = "pid", .mode = S_IRUGO},
 170         .show = pid_show,
 171 };
 172
 173 static void nbd_dev_remove(struct nbd_device *nbd)
 174 {
 175         struct gendisk *disk = nbd->disk;
 176         if (disk) {
 177                 del_gendisk(disk);
 178                 blk_cleanup_queue(disk->queue);
 179                 blk_mq_free_tag_set(&nbd->tag_set);
 180                 disk->private_data = NULL;
 181                 put_disk(disk);
 182         }
 183         kfree(nbd);
 184 }
 185
 186 static void nbd_put(struct nbd_device *nbd)
 187 {
 188         if (refcount_dec_and_mutex_lock(&nbd->refs,
 189                                         &nbd_index_mutex)) {
 190                 idr_remove(&nbd_index_idr, nbd->index);
 191                 mutex_unlock(&nbd_index_mutex);
 192                 nbd_dev_remove(nbd);
 193         }
 194 }
 195
 196 static int nbd_disconnected(struct nbd_config *config)
 197 {
 198         return test_bit(NBD_DISCONNECTED, &config->runtime_flags) ||
 199                 test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
 200 }
 201
 202 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
 203                                 int notify)
 204 {
 205         if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
 206                 struct link_dead_args *args;
 207                 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
 208                 if (args) {
 209                         INIT_WORK(&args->work, nbd_dead_link_work);
 210                         args->index = nbd->index;
 211                         queue_work(system_wq, &args->work);
 212                 }
 213         }
 214         if (!nsock->dead) {
 215                 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
 216                 atomic_dec(&nbd->config->live_connections);
 217         }
 218         nsock->dead = true;
 219         nsock->pending = NULL;
 220         nsock->sent = 0;
 221 }
 222
 223 static void nbd_size_clear(struct nbd_device *nbd)
 224 {
 225         if (nbd->config->bytesize) {
 226                 set_capacity(nbd->disk, 0);
 227                 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 228         }
 229 }
 230
 231 static void nbd_size_update(struct nbd_device *nbd)
 232 {
 233         struct nbd_config *config = nbd->config;
 234         blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
 235         blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
 236         set_capacity(nbd->disk, config->bytesize >> 9);
 237         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 238 }
 239
 240 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
 241                          loff_t nr_blocks)
 242 {
 243         struct nbd_config *config = nbd->config;
 244         config->blksize = blocksize;
 245         config->bytesize = blocksize * nr_blocks;
 246         nbd_size_update(nbd);
 247 }
 248
 249 static void nbd_complete_rq(struct request *req)
 250 {
 251         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 252
 253         dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", cmd,
 254                 cmd->status ? "failed" : "done");
 255
 256         blk_mq_end_request(req, cmd->status);
 257 }
 258
 259 /*
 260  * Forcibly shutdown the socket causing all listeners to error
 261  */
 262 static void sock_shutdown(struct nbd_device *nbd)
 263 {
 264         struct nbd_config *config = nbd->config;
 265         int i;
 266
 267         if (config->num_connections == 0)
 268                 return;
 269         if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
 270                 return;
 271
 272         for (i = 0; i < config->num_connections; i++) {
 273                 struct nbd_sock *nsock = config->socks[i];
 274                 mutex_lock(&nsock->tx_lock);
 275                 nbd_mark_nsock_dead(nbd, nsock, 0);
 276                 mutex_unlock(&nsock->tx_lock);
 277         }
 278         dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
 279 }
 280
 281 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 282                                                  bool reserved)
 283 {
 284         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 285         struct nbd_device *nbd = cmd->nbd;
 286         struct nbd_config *config;
 287
 288         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 289                 cmd->status = BLK_STS_TIMEOUT;
 290                 return BLK_EH_HANDLED;
 291         }
 292
 293         /* If we are waiting on our dead timer then we could get timeout
 294          * callbacks for our request.  For this we just want to reset the timer
 295          * and let the queue side take care of everything.
 296          */
 297         if (!completion_done(&cmd->send_complete)) {
 298                 nbd_config_put(nbd);
 299                 return BLK_EH_RESET_TIMER;
 300         }
 301         config = nbd->config;
 302
 303         if (config->num_connections > 1) {
 304                 dev_err_ratelimited(nbd_to_dev(nbd),
 305                                     "Connection timed out, retrying\n");
 306                 /*
 307                  * Hooray we have more connections, requeue this IO, the submit
 308                  * path will put it on a real connection.
 309                  */
 310                 if (config->socks && config->num_connections > 1) {
 311                         if (cmd->index < config->num_connections) {
 312                                 struct nbd_sock *nsock =
 313                                         config->socks[cmd->index];
 314                                 mutex_lock(&nsock->tx_lock);
 315                                 /* We can have multiple outstanding requests, so
 316                                  * we don't want to mark the nsock dead if we've
 317                                  * already reconnected with a new socket, so
 318                                  * only mark it dead if its the same socket we
 319                                  * were sent out on.
 320                                  */
 321                                 if (cmd->cookie == nsock->cookie)
 322                                         nbd_mark_nsock_dead(nbd, nsock, 1);
 323                                 mutex_unlock(&nsock->tx_lock);
 324                         }
 325                         blk_mq_requeue_request(req, true);
 326                         nbd_config_put(nbd);
 327                         return BLK_EH_NOT_HANDLED;
 328                 }
 329         } else {
 330                 dev_err_ratelimited(nbd_to_dev(nbd),
 331                                     "Connection timed out\n");
 332         }
 333         set_bit(NBD_TIMEDOUT, &config->runtime_flags);
 334         cmd->status = BLK_STS_IOERR;
 335         sock_shutdown(nbd);
 336         nbd_config_put(nbd);
 337
 338         return BLK_EH_HANDLED;
 339 }
 340
 341 /*
 342  *  Send or receive packet.
 343  */
 344 static int sock_xmit(struct nbd_device *nbd, int index, int send,
 345                      struct iov_iter *iter, int msg_flags, int *sent)
 346 {
 347         struct nbd_config *config = nbd->config;
 348         struct socket *sock = config->socks[index]->sock;
 349         int result;
 350         struct msghdr msg;
 351         unsigned int noreclaim_flag;
 352
 353         if (unlikely(!sock)) {
 354                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 355                         "Attempted %s on closed socket in sock_xmit\n",
 356                         (send ? "send" : "recv"));
 357                 return -EINVAL;
 358         }
 359
 360         msg.msg_iter = *iter;
 361
 362         noreclaim_flag = memalloc_noreclaim_save();
 363         do {
 364                 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
 365                 msg.msg_name = NULL;
 366                 msg.msg_namelen = 0;
 367                 msg.msg_control = NULL;
 368                 msg.msg_controllen = 0;
 369                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 370
 371                 if (send)
 372                         result = sock_sendmsg(sock, &msg);
 373                 else
 374                         result = sock_recvmsg(sock, &msg, msg.msg_flags);
 375
 376                 if (result <= 0) {
 377                         if (result == 0)
 378                                 result = -EPIPE; /* short read */
 379                         break;
 380                 }
 381                 if (sent)
 382                         *sent += result;
 383         } while (msg_data_left(&msg));
 384
 385         memalloc_noreclaim_restore(noreclaim_flag);
 386
 387         return result;
 388 }
 389
 390 /* always call with the tx_lock held */
 391 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 392 {
 393         struct request *req = blk_mq_rq_from_pdu(cmd);
 394         struct nbd_config *config = nbd->config;
 395         struct nbd_sock *nsock = config->socks[index];
 396         int result;
 397         struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
 398         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 399         struct iov_iter from;
 400         unsigned long size = blk_rq_bytes(req);
 401         struct bio *bio;
 402         u32 type;
 403         u32 nbd_cmd_flags = 0;
 404         u32 tag = blk_mq_unique_tag(req);
 405         int sent = nsock->sent, skip = 0;
 406
 407         iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
 408
 409         switch (req_op(req)) {
 410         case REQ_OP_DISCARD:
 411                 type = NBD_CMD_TRIM;
 412                 break;
 413         case REQ_OP_FLUSH:
 414                 type = NBD_CMD_FLUSH;
 415                 break;
 416         case REQ_OP_WRITE:
 417                 type = NBD_CMD_WRITE;
 418                 break;
 419         case REQ_OP_READ:
 420                 type = NBD_CMD_READ;
 421                 break;
 422         default:
 423                 return -EIO;
 424         }
 425
 426         if (rq_data_dir(req) == WRITE &&
 427             (config->flags & NBD_FLAG_READ_ONLY)) {
 428                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 429                                     "Write on read-only\n");
 430                 return -EIO;
 431         }
 432
 433         if (req->cmd_flags & REQ_FUA)
 434                 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
 435
 436         /* We did a partial send previously, and we at least sent the whole
 437          * request struct, so just go and send the rest of the pages in the
 438          * request.
 439          */
 440         if (sent) {
 441                 if (sent >= sizeof(request)) {
 442                         skip = sent - sizeof(request);
 443                         goto send_pages;
 444                 }
 445                 iov_iter_advance(&from, sent);
 446         }
 447         cmd->index = index;
 448         cmd->cookie = nsock->cookie;
 449         request.type = htonl(type | nbd_cmd_flags);
 450         if (type != NBD_CMD_FLUSH) {
 451                 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
 452                 request.len = htonl(size);
 453         }
 454         memcpy(request.handle, &tag, sizeof(tag));
 455
 456         dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
 457                 cmd, nbdcmd_to_ascii(type),
 458                 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
 459         result = sock_xmit(nbd, index, 1, &from,
 460                         (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
 461         if (result <= 0) {
 462                 if (result == -ERESTARTSYS) {
 463                         /* If we havne't sent anything we can just return BUSY,
 464                          * however if we have sent something we need to make
 465                          * sure we only allow this req to be sent until we are
 466                          * completely done.
 467                          */
 468                         if (sent) {
 469                                 nsock->pending = req;
 470                                 nsock->sent = sent;
 471                         }
 472                         return BLK_STS_RESOURCE;
 473                 }
 474                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 475                         "Send control failed (result %d)\n", result);
 476                 return -EAGAIN;
 477         }
 478 send_pages:
 479         if (type != NBD_CMD_WRITE)
 480                 goto out;
 481
 482         bio = req->bio;
 483         while (bio) {
 484                 struct bio *next = bio->bi_next;
 485                 struct bvec_iter iter;
 486                 struct bio_vec bvec;
 487
 488                 bio_for_each_segment(bvec, bio, iter) {
 489                         bool is_last = !next && bio_iter_last(bvec, iter);
 490                         int flags = is_last ? 0 : MSG_MORE;
 491
 492                         dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
 493                                 cmd, bvec.bv_len);
 494                         iov_iter_bvec(&from, ITER_BVEC | WRITE,
 495                                       &bvec, 1, bvec.bv_len);
 496                         if (skip) {
 497                                 if (skip >= iov_iter_count(&from)) {
 498                                         skip -= iov_iter_count(&from);
 499                                         continue;
 500                                 }
 501                                 iov_iter_advance(&from, skip);
 502                                 skip = 0;
 503                         }
 504                         result = sock_xmit(nbd, index, 1, &from, flags, &sent);
 505                         if (result <= 0) {
 506                                 if (result == -ERESTARTSYS) {
 507                                         /* We've already sent the header, we
 508                                          * have no choice but to set pending and
 509                                          * return BUSY.
 510                                          */
 511                                         nsock->pending = req;
 512                                         nsock->sent = sent;
 513                                         return BLK_STS_RESOURCE;
 514                                 }
 515                                 dev_err(disk_to_dev(nbd->disk),
 516                                         "Send data failed (result %d)\n",
 517                                         result);
 518                                 return -EAGAIN;
 519                         }
 520                         /*
 521                          * The completion might already have come in,
 522                          * so break for the last one instead of letting
 523                          * the iterator do it. This prevents use-after-free
 524                          * of the bio.
 525                          */
 526                         if (is_last)
 527                                 break;
 528                 }
 529                 bio = next;
 530         }
 531 out:
 532         nsock->pending = NULL;
 533         nsock->sent = 0;
 534         return 0;
 535 }
 536
 537 /* NULL returned = something went wrong, inform userspace */
 538 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 539 {
 540         struct nbd_config *config = nbd->config;
 541         int result;
 542         struct nbd_reply reply;
 543         struct nbd_cmd *cmd;
 544         struct request *req = NULL;
 545         u16 hwq;
 546         u32 tag;
 547         struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
 548         struct iov_iter to;
 549
 550         reply.magic = 0;
 551         iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply));
 552         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 553         if (result <= 0) {
 554                 if (!nbd_disconnected(config))
 555                         dev_err(disk_to_dev(nbd->disk),
 556                                 "Receive control failed (result %d)\n", result);
 557                 return ERR_PTR(result);
 558         }
 559
 560         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 561                 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
 562                                 (unsigned long)ntohl(reply.magic));
 563                 return ERR_PTR(-EPROTO);
 564         }
 565
 566         memcpy(&tag, reply.handle, sizeof(u32));
 567
 568         hwq = blk_mq_unique_tag_to_hwq(tag);
 569         if (hwq < nbd->tag_set.nr_hw_queues)
 570                 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
 571                                        blk_mq_unique_tag_to_tag(tag));
 572         if (!req || !blk_mq_request_started(req)) {
 573                 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
 574                         tag, req);
 575                 return ERR_PTR(-ENOENT);
 576         }
 577         cmd = blk_mq_rq_to_pdu(req);
 578         if (ntohl(reply.error)) {
 579                 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
 580                         ntohl(reply.error));
 581                 cmd->status = BLK_STS_IOERR;
 582                 return cmd;
 583         }
 584
 585         dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
 586         if (rq_data_dir(req) != WRITE) {
 587                 struct req_iterator iter;
 588                 struct bio_vec bvec;
 589
 590                 rq_for_each_segment(bvec, req, iter) {
 591                         iov_iter_bvec(&to, ITER_BVEC | READ,
 592                                       &bvec, 1, bvec.bv_len);
 593                         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 594                         if (result <= 0) {
 595                                 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 596                                         result);
 597                                 /*
 598                                  * If we've disconnected or we only have 1
 599                                  * connection then we need to make sure we
 600                                  * complete this request, otherwise error out
 601                                  * and let the timeout stuff handle resubmitting
 602                                  * this request onto another connection.
 603                                  */
 604                                 if (nbd_disconnected(config) ||
 605                                     config->num_connections <= 1) {
 606                                         cmd->status = BLK_STS_IOERR;
 607                                         return cmd;
 608                                 }
 609                                 return ERR_PTR(-EIO);
 610                         }
 611                         dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 612                                 cmd, bvec.bv_len);
 613                 }
 614         } else {
 615                 /* See the comment in nbd_queue_rq. */
 616                 wait_for_completion(&cmd->send_complete);
 617         }
 618         return cmd;
 619 }
 620
 621 static void recv_work(struct work_struct *work)
 622 {
 623         struct recv_thread_args *args = container_of(work,
 624                                                      struct recv_thread_args,
 625                                                      work);
 626         struct nbd_device *nbd = args->nbd;
 627         struct nbd_config *config = nbd->config;
 628         struct nbd_cmd *cmd;
 629         int ret = 0;
 630
 631         while (1) {
 632                 cmd = nbd_read_stat(nbd, args->index);
 633                 if (IS_ERR(cmd)) {
 634                         struct nbd_sock *nsock = config->socks[args->index];
 635
 636                         mutex_lock(&nsock->tx_lock);
 637                         nbd_mark_nsock_dead(nbd, nsock, 1);
 638                         mutex_unlock(&nsock->tx_lock);
 639                         ret = PTR_ERR(cmd);
 640                         break;
 641                 }
 642
 643                 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
 644         }
 645         atomic_dec(&config->recv_threads);
 646         wake_up(&config->recv_wq);
 647         nbd_config_put(nbd);
 648         kfree(args);
 649 }
 650
 651 static void nbd_clear_req(struct request *req, void *data, bool reserved)
 652 {
 653         struct nbd_cmd *cmd;
 654
 655         if (!blk_mq_request_started(req))
 656                 return;
 657         cmd = blk_mq_rq_to_pdu(req);
 658         cmd->status = BLK_STS_IOERR;
 659         blk_mq_complete_request(req);
 660 }
 661
 662 static void nbd_clear_que(struct nbd_device *nbd)
 663 {
 664         blk_mq_quiesce_queue(nbd->disk->queue);
 665         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 666         blk_mq_unquiesce_queue(nbd->disk->queue);
 667         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 668 }
 669
 670 static int find_fallback(struct nbd_device *nbd, int index)
 671 {
 672         struct nbd_config *config = nbd->config;
 673         int new_index = -1;
 674         struct nbd_sock *nsock = config->socks[index];
 675         int fallback = nsock->fallback_index;
 676
 677         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 678                 return new_index;
 679
 680         if (config->num_connections <= 1) {
 681                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 682                                     "Attempted send on invalid socket\n");
 683                 return new_index;
 684         }
 685
 686         if (fallback >= 0 && fallback < config->num_connections &&
 687             !config->socks[fallback]->dead)
 688                 return fallback;
 689
 690         if (nsock->fallback_index < 0 ||
 691             nsock->fallback_index >= config->num_connections ||
 692             config->socks[nsock->fallback_index]->dead) {
 693                 int i;
 694                 for (i = 0; i < config->num_connections; i++) {
 695                         if (i == index)
 696                                 continue;
 697                         if (!config->socks[i]->dead) {
 698                                 new_index = i;
 699                                 break;
 700                         }
 701                 }
 702                 nsock->fallback_index = new_index;
 703                 if (new_index < 0) {
 704                         dev_err_ratelimited(disk_to_dev(nbd->disk),
 705                                             "Dead connection, failed to find a fallback\n");
 706                         return new_index;
 707                 }
 708         }
 709         new_index = nsock->fallback_index;
 710         return new_index;
 711 }
 712
 713 static int wait_for_reconnect(struct nbd_device *nbd)
 714 {
 715         struct nbd_config *config = nbd->config;
 716         if (!config->dead_conn_timeout)
 717                 return 0;
 718         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 719                 return 0;
 720         wait_event_interruptible_timeout(config->conn_wait,
 721                                          atomic_read(&config->live_connections),
 722                                          config->dead_conn_timeout);
 723         return atomic_read(&config->live_connections);
 724 }
 725
 726 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 727 {
 728         struct request *req = blk_mq_rq_from_pdu(cmd);
 729         struct nbd_device *nbd = cmd->nbd;
 730         struct nbd_config *config;
 731         struct nbd_sock *nsock;
 732         int ret;
 733
 734         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 735                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 736                                     "Socks array is empty\n");
 737                 return -EINVAL;
 738         }
 739         config = nbd->config;
 740
 741         if (index >= config->num_connections) {
 742                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 743                                     "Attempted send on invalid socket\n");
 744                 nbd_config_put(nbd);
 745                 return -EINVAL;
 746         }
 747         cmd->status = BLK_STS_OK;
 748 again:
 749         nsock = config->socks[index];
 750         mutex_lock(&nsock->tx_lock);
 751         if (nsock->dead) {
 752                 int old_index = index;
 753                 index = find_fallback(nbd, index);
 754                 mutex_unlock(&nsock->tx_lock);
 755                 if (index < 0) {
 756                         if (wait_for_reconnect(nbd)) {
 757                                 index = old_index;
 758                                 goto again;
 759                         }
 760                         /* All the sockets should already be down at this point,
 761                          * we just want to make sure that DISCONNECTED is set so
 762                          * any requests that come in that were queue'ed waiting
 763                          * for the reconnect timer don't trigger the timer again
 764                          * and instead just error out.
 765                          */
 766                         sock_shutdown(nbd);
 767                         nbd_config_put(nbd);
 768                         return -EIO;
 769                 }
 770                 goto again;
 771         }
 772
 773         /* Handle the case that we have a pending request that was partially
 774          * transmitted that _has_ to be serviced first.  We need to call requeue
 775          * here so that it gets put _after_ the request that is already on the
 776          * dispatch list.
 777          */
 778         if (unlikely(nsock->pending && nsock->pending != req)) {
 779                 blk_mq_requeue_request(req, true);
 780                 ret = 0;
 781                 goto out;
 782         }
 783         /*
 784          * Some failures are related to the link going down, so anything that
 785          * returns EAGAIN can be retried on a different socket.
 786          */
 787         ret = nbd_send_cmd(nbd, cmd, index);
 788         if (ret == -EAGAIN) {
 789                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 790                                     "Request send failed trying another connection\n");
 791                 nbd_mark_nsock_dead(nbd, nsock, 1);
 792                 mutex_unlock(&nsock->tx_lock);
 793                 goto again;
 794         }
 795 out:
 796         mutex_unlock(&nsock->tx_lock);
 797         nbd_config_put(nbd);
 798         return ret;
 799 }
 800
 801 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 802                         const struct blk_mq_queue_data *bd)
 803 {
 804         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 805         int ret;
 806
 807         /*
 808          * Since we look at the bio's to send the request over the network we
 809          * need to make sure the completion work doesn't mark this request done
 810          * before we are done doing our send.  This keeps us from dereferencing
 811          * freed data if we have particularly fast completions (ie we get the
 812          * completion before we exit sock_xmit on the last bvec) or in the case
 813          * that the server is misbehaving (or there was an error) before we're
 814          * done sending everything over the wire.
 815          */
 816         init_completion(&cmd->send_complete);
 817         blk_mq_start_request(bd->rq);
 818
 819         /* We can be called directly from the user space process, which means we
 820          * could possibly have signals pending so our sendmsg will fail.  In
 821          * this case we need to return that we are busy, otherwise error out as
 822          * appropriate.
 823          */
 824         ret = nbd_handle_cmd(cmd, hctx->queue_num);
 825         complete(&cmd->send_complete);
 826
 827         return ret < 0 ? BLK_STS_IOERR : BLK_STS_OK;
 828 }
 829
 830 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
 831                           bool netlink)
 832 {
 833         struct nbd_config *config = nbd->config;
 834         struct socket *sock;
 835         struct nbd_sock **socks;
 836         struct nbd_sock *nsock;
 837         int err;
 838
 839         sock = sockfd_lookup(arg, &err);
 840         if (!sock)
 841                 return err;
 842
 843         if (!netlink && !nbd->task_setup &&
 844             !test_bit(NBD_BOUND, &config->runtime_flags))
 845                 nbd->task_setup = current;
 846
 847         if (!netlink &&
 848             (nbd->task_setup != current ||
 849              test_bit(NBD_BOUND, &config->runtime_flags))) {
 850                 dev_err(disk_to_dev(nbd->disk),
 851                         "Device being setup by another task");
 852                 sockfd_put(sock);
 853                 return -EBUSY;
 854         }
 855
 856         socks = krealloc(config->socks, (config->num_connections + 1) *
 857                          sizeof(struct nbd_sock *), GFP_KERNEL);
 858         if (!socks) {
 859                 sockfd_put(sock);
 860                 return -ENOMEM;
 861         }
 862         nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
 863         if (!nsock) {
 864                 sockfd_put(sock);
 865                 return -ENOMEM;
 866         }
 867
 868         config->socks = socks;
 869
 870         nsock->fallback_index = -1;
 871         nsock->dead = false;
 872         mutex_init(&nsock->tx_lock);
 873         nsock->sock = sock;
 874         nsock->pending = NULL;
 875         nsock->sent = 0;
 876         nsock->cookie = 0;
 877         socks[config->num_connections++] = nsock;
 878         atomic_inc(&config->live_connections);
 879
 880         return 0;
 881 }
 882
 883 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
 884 {
 885         struct nbd_config *config = nbd->config;
 886         struct socket *sock, *old;
 887         struct recv_thread_args *args;
 888         int i;
 889         int err;
 890
 891         sock = sockfd_lookup(arg, &err);
 892         if (!sock)
 893                 return err;
 894
 895         args = kzalloc(sizeof(*args), GFP_KERNEL);
 896         if (!args) {
 897                 sockfd_put(sock);
 898                 return -ENOMEM;
 899         }
 900
 901         for (i = 0; i < config->num_connections; i++) {
 902                 struct nbd_sock *nsock = config->socks[i];
 903
 904                 if (!nsock->dead)
 905                         continue;
 906
 907                 mutex_lock(&nsock->tx_lock);
 908                 if (!nsock->dead) {
 909                         mutex_unlock(&nsock->tx_lock);
 910                         continue;
 911                 }
 912                 sk_set_memalloc(sock->sk);
 913                 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
 914                 atomic_inc(&config->recv_threads);
 915                 refcount_inc(&nbd->config_refs);
 916                 old = nsock->sock;
 917                 nsock->fallback_index = -1;
 918                 nsock->sock = sock;
 919                 nsock->dead = false;
 920                 INIT_WORK(&args->work, recv_work);
 921                 args->index = i;
 922                 args->nbd = nbd;
 923                 nsock->cookie++;
 924                 mutex_unlock(&nsock->tx_lock);
 925                 sockfd_put(old);
 926
 927                 /* We take the tx_mutex in an error path in the recv_work, so we
 928                  * need to queue_work outside of the tx_mutex.
 929                  */
 930                 queue_work(recv_workqueue, &args->work);
 931
 932                 atomic_inc(&config->live_connections);
 933                 wake_up(&config->conn_wait);
 934                 return 0;
 935         }
 936         sockfd_put(sock);
 937         kfree(args);
 938         return -ENOSPC;
 939 }
 940
 941 static void nbd_bdev_reset(struct block_device *bdev)
 942 {
 943         if (bdev->bd_openers > 1)
 944                 return;
 945         bd_set_size(bdev, 0);
 946         if (max_part > 0) {
 947                 blkdev_reread_part(bdev);
 948                 bdev->bd_invalidated = 1;
 949         }
 950 }
 951
 952 static void nbd_parse_flags(struct nbd_device *nbd)
 953 {
 954         struct nbd_config *config = nbd->config;
 955         if (config->flags & NBD_FLAG_READ_ONLY)
 956                 set_disk_ro(nbd->disk, true);
 957         else
 958                 set_disk_ro(nbd->disk, false);
 959         if (config->flags & NBD_FLAG_SEND_TRIM)
 960                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 961         if (config->flags & NBD_FLAG_SEND_FLUSH) {
 962                 if (config->flags & NBD_FLAG_SEND_FUA)
 963                         blk_queue_write_cache(nbd->disk->queue, true, true);
 964                 else
 965                         blk_queue_write_cache(nbd->disk->queue, true, false);
 966         }
 967         else
 968                 blk_queue_write_cache(nbd->disk->queue, false, false);
 969 }
 970
 971 static void send_disconnects(struct nbd_device *nbd)
 972 {
 973         struct nbd_config *config = nbd->config;
 974         struct nbd_request request = {
 975                 .magic = htonl(NBD_REQUEST_MAGIC),
 976                 .type = htonl(NBD_CMD_DISC),
 977         };
 978         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 979         struct iov_iter from;
 980         int i, ret;
 981
 982         for (i = 0; i < config->num_connections; i++) {
 983                 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
 984                 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
 985                 if (ret <= 0)
 986                         dev_err(disk_to_dev(nbd->disk),
 987                                 "Send disconnect failed %d\n", ret);
 988         }
 989 }
 990
 991 static int nbd_disconnect(struct nbd_device *nbd)
 992 {
 993         struct nbd_config *config = nbd->config;
 994
 995         dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
 996         if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED,
 997                               &config->runtime_flags))
 998                 send_disconnects(nbd);
 999         return 0;
1000 }
1001
1002 static void nbd_clear_sock(struct nbd_device *nbd)
1003 {
1004         sock_shutdown(nbd);
1005         nbd_clear_que(nbd);
1006         nbd->task_setup = NULL;
1007 }
1008
1009 static void nbd_config_put(struct nbd_device *nbd)
1010 {
1011         if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1012                                         &nbd->config_lock)) {
1013                 struct nbd_config *config = nbd->config;
1014                 nbd_dev_dbg_close(nbd);
1015                 nbd_size_clear(nbd);
1016                 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1017                                        &config->runtime_flags))
1018                         device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1019                 nbd->task_recv = NULL;
1020                 nbd_clear_sock(nbd);
1021                 if (config->num_connections) {
1022                         int i;
1023                         for (i = 0; i < config->num_connections; i++) {
1024                                 sockfd_put(config->socks[i]->sock);
1025                                 kfree(config->socks[i]);
1026                         }
1027                         kfree(config->socks);
1028                 }
1029                 kfree(nbd->config);
1030                 nbd->config = NULL;
1031
1032                 nbd->tag_set.timeout = 0;
1033                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1034
1035                 mutex_unlock(&nbd->config_lock);
1036                 nbd_put(nbd);
1037                 module_put(THIS_MODULE);
1038         }
1039 }
1040
1041 static int nbd_start_device(struct nbd_device *nbd)
1042 {
1043         struct nbd_config *config = nbd->config;
1044         int num_connections = config->num_connections;
1045         int error = 0, i;
1046
1047         if (nbd->task_recv)
1048                 return -EBUSY;
1049         if (!config->socks)
1050                 return -EINVAL;
1051         if (num_connections > 1 &&
1052             !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1053                 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1054                 return -EINVAL;
1055         }
1056
1057         blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1058         nbd->task_recv = current;
1059
1060         nbd_parse_flags(nbd);
1061
1062         error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1063         if (error) {
1064                 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1065                 return error;
1066         }
1067         set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1068
1069         nbd_dev_dbg_init(nbd);
1070         for (i = 0; i < num_connections; i++) {
1071                 struct recv_thread_args *args;
1072
1073                 args = kzalloc(sizeof(*args), GFP_KERNEL);
1074                 if (!args) {
1075                         sock_shutdown(nbd);
1076                         return -ENOMEM;
1077                 }
1078                 sk_set_memalloc(config->socks[i]->sock->sk);
1079                 config->socks[i]->sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1080                 atomic_inc(&config->recv_threads);
1081                 refcount_inc(&nbd->config_refs);
1082                 INIT_WORK(&args->work, recv_work);
1083                 args->nbd = nbd;
1084                 args->index = i;
1085                 queue_work(recv_workqueue, &args->work);
1086         }
1087         return error;
1088 }
1089
1090 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1091 {
1092         struct nbd_config *config = nbd->config;
1093         int ret;
1094
1095         ret = nbd_start_device(nbd);
1096         if (ret)
1097                 return ret;
1098
1099         bd_set_size(bdev, config->bytesize);
1100         if (max_part)
1101                 bdev->bd_invalidated = 1;
1102         mutex_unlock(&nbd->config_lock);
1103         ret = wait_event_interruptible(config->recv_wq,
1104                                          atomic_read(&config->recv_threads) == 0);
1105         if (ret)
1106                 sock_shutdown(nbd);
1107         mutex_lock(&nbd->config_lock);
1108         bd_set_size(bdev, 0);
1109         /* user requested, ignore socket errors */
1110         if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1111                 ret = 0;
1112         if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1113                 ret = -ETIMEDOUT;
1114         return ret;
1115 }
1116
1117 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1118                                  struct block_device *bdev)
1119 {
1120         sock_shutdown(nbd);
1121         kill_bdev(bdev);
1122         nbd_bdev_reset(bdev);
1123         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1124                                &nbd->config->runtime_flags))
1125                 nbd_config_put(nbd);
1126 }
1127
1128 /* Must be called with config_lock held */
1129 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1130                        unsigned int cmd, unsigned long arg)
1131 {
1132         struct nbd_config *config = nbd->config;
1133
1134         switch (cmd) {
1135         case NBD_DISCONNECT:
1136                 return nbd_disconnect(nbd);
1137         case NBD_CLEAR_SOCK:
1138                 nbd_clear_sock_ioctl(nbd, bdev);
1139                 return 0;
1140         case NBD_SET_SOCK:
1141                 return nbd_add_socket(nbd, arg, false);
1142         case NBD_SET_BLKSIZE:
1143                 nbd_size_set(nbd, arg,
1144                              div_s64(config->bytesize, arg));
1145                 return 0;
1146         case NBD_SET_SIZE:
1147                 nbd_size_set(nbd, config->blksize,
1148                              div_s64(arg, config->blksize));
1149                 return 0;
1150         case NBD_SET_SIZE_BLOCKS:
1151                 nbd_size_set(nbd, config->blksize, arg);
1152                 return 0;
1153         case NBD_SET_TIMEOUT:
1154                 if (arg) {
1155                         nbd->tag_set.timeout = arg * HZ;
1156                         blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1157                 }
1158                 return 0;
1159
1160         case NBD_SET_FLAGS:
1161                 config->flags = arg;
1162                 return 0;
1163         case NBD_DO_IT:
1164                 return nbd_start_device_ioctl(nbd, bdev);
1165         case NBD_CLEAR_QUE:
1166                 /*
1167                  * This is for compatibility only.  The queue is always cleared
1168                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
1169                  */
1170                 return 0;
1171         case NBD_PRINT_DEBUG:
1172                 /*
1173                  * For compatibility only, we no longer keep a list of
1174                  * outstanding requests.
1175                  */
1176                 return 0;
1177         }
1178         return -ENOTTY;
1179 }
1180
1181 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1182                      unsigned int cmd, unsigned long arg)
1183 {
1184         struct nbd_device *nbd = bdev->bd_disk->private_data;
1185         struct nbd_config *config = nbd->config;
1186         int error = -EINVAL;
1187
1188         if (!capable(CAP_SYS_ADMIN))
1189                 return -EPERM;
1190
1191         mutex_lock(&nbd->config_lock);
1192
1193         /* Don't allow ioctl operations on a nbd device that was created with
1194          * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1195          */
1196         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1197             (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1198                 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1199         else
1200                 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1201         mutex_unlock(&nbd->config_lock);
1202         return error;
1203 }
1204
1205 static struct nbd_config *nbd_alloc_config(void)
1206 {
1207         struct nbd_config *config;
1208
1209         config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1210         if (!config)
1211                 return NULL;
1212         atomic_set(&config->recv_threads, 0);
1213         init_waitqueue_head(&config->recv_wq);
1214         init_waitqueue_head(&config->conn_wait);
1215         config->blksize = 1024;
1216         atomic_set(&config->live_connections, 0);
1217         try_module_get(THIS_MODULE);
1218         return config;
1219 }
1220
1221 static int nbd_open(struct block_device *bdev, fmode_t mode)
1222 {
1223         struct nbd_device *nbd;
1224         int ret = 0;
1225
1226         mutex_lock(&nbd_index_mutex);
1227         nbd = bdev->bd_disk->private_data;
1228         if (!nbd) {
1229                 ret = -ENXIO;
1230                 goto out;
1231         }
1232         if (!refcount_inc_not_zero(&nbd->refs)) {
1233                 ret = -ENXIO;
1234                 goto out;
1235         }
1236         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1237                 struct nbd_config *config;
1238
1239                 mutex_lock(&nbd->config_lock);
1240                 if (refcount_inc_not_zero(&nbd->config_refs)) {
1241                         mutex_unlock(&nbd->config_lock);
1242                         goto out;
1243                 }
1244                 config = nbd->config = nbd_alloc_config();
1245                 if (!config) {
1246                         ret = -ENOMEM;
1247                         mutex_unlock(&nbd->config_lock);
1248                         goto out;
1249                 }
1250                 refcount_set(&nbd->config_refs, 1);
1251                 refcount_inc(&nbd->refs);
1252                 mutex_unlock(&nbd->config_lock);
1253         }
1254 out:
1255         mutex_unlock(&nbd_index_mutex);
1256         return ret;
1257 }
1258
1259 static void nbd_release(struct gendisk *disk, fmode_t mode)
1260 {
1261         struct nbd_device *nbd = disk->private_data;
1262         nbd_config_put(nbd);
1263         nbd_put(nbd);
1264 }
1265
1266 static const struct block_device_operations nbd_fops =
1267 {
1268         .owner =        THIS_MODULE,
1269         .open =         nbd_open,
1270         .release =      nbd_release,
1271         .ioctl =        nbd_ioctl,
1272         .compat_ioctl = nbd_ioctl,
1273 };
1274
1275 #if IS_ENABLED(CONFIG_DEBUG_FS)
1276
1277 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1278 {
1279         struct nbd_device *nbd = s->private;
1280
1281         if (nbd->task_recv)
1282                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1283
1284         return 0;
1285 }
1286
1287 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1288 {
1289         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1290 }
1291
1292 static const struct file_operations nbd_dbg_tasks_ops = {
1293         .open = nbd_dbg_tasks_open,
1294         .read = seq_read,
1295         .llseek = seq_lseek,
1296         .release = single_release,
1297 };
1298
1299 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1300 {
1301         struct nbd_device *nbd = s->private;
1302         u32 flags = nbd->config->flags;
1303
1304         seq_printf(s, "Hex: 0x%08x\n\n", flags);
1305
1306         seq_puts(s, "Known flags:\n");
1307
1308         if (flags & NBD_FLAG_HAS_FLAGS)
1309                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1310         if (flags & NBD_FLAG_READ_ONLY)
1311                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1312         if (flags & NBD_FLAG_SEND_FLUSH)
1313                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1314         if (flags & NBD_FLAG_SEND_FUA)
1315                 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1316         if (flags & NBD_FLAG_SEND_TRIM)
1317                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1318
1319         return 0;
1320 }
1321
1322 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1323 {
1324         return single_open(file, nbd_dbg_flags_show, inode->i_private);
1325 }
1326
1327 static const struct file_operations nbd_dbg_flags_ops = {
1328         .open = nbd_dbg_flags_open,
1329         .read = seq_read,
1330         .llseek = seq_lseek,
1331         .release = single_release,
1332 };
1333
1334 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1335 {
1336         struct dentry *dir;
1337         struct nbd_config *config = nbd->config;
1338
1339         if (!nbd_dbg_dir)
1340                 return -EIO;
1341
1342         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1343         if (!dir) {
1344                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1345                         nbd_name(nbd));
1346                 return -EIO;
1347         }
1348         config->dbg_dir = dir;
1349
1350         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1351         debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1352         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1353         debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1354         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1355
1356         return 0;
1357 }
1358
1359 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1360 {
1361         debugfs_remove_recursive(nbd->config->dbg_dir);
1362 }
1363
1364 static int nbd_dbg_init(void)
1365 {
1366         struct dentry *dbg_dir;
1367
1368         dbg_dir = debugfs_create_dir("nbd", NULL);
1369         if (!dbg_dir)
1370                 return -EIO;
1371
1372         nbd_dbg_dir = dbg_dir;
1373
1374         return 0;
1375 }
1376
1377 static void nbd_dbg_close(void)
1378 {
1379         debugfs_remove_recursive(nbd_dbg_dir);
1380 }
1381
1382 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1383
1384 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1385 {
1386         return 0;
1387 }
1388
1389 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1390 {
1391 }
1392
1393 static int nbd_dbg_init(void)
1394 {
1395         return 0;
1396 }
1397
1398 static void nbd_dbg_close(void)
1399 {
1400 }
1401
1402 #endif
1403
1404 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1405                             unsigned int hctx_idx, unsigned int numa_node)
1406 {
1407         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1408         cmd->nbd = set->driver_data;
1409         return 0;
1410 }
1411
1412 static const struct blk_mq_ops nbd_mq_ops = {
1413         .queue_rq       = nbd_queue_rq,
1414         .complete       = nbd_complete_rq,
1415         .init_request   = nbd_init_request,
1416         .timeout        = nbd_xmit_timeout,
1417 };
1418
1419 static int nbd_dev_add(int index)
1420 {
1421         struct nbd_device *nbd;
1422         struct gendisk *disk;
1423         struct request_queue *q;
1424         int err = -ENOMEM;
1425
1426         nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1427         if (!nbd)
1428                 goto out;
1429
1430         disk = alloc_disk(1 << part_shift);
1431         if (!disk)
1432                 goto out_free_nbd;
1433
1434         if (index >= 0) {
1435                 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1436                                 GFP_KERNEL);
1437                 if (err == -ENOSPC)
1438                         err = -EEXIST;
1439         } else {
1440                 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1441                 if (err >= 0)
1442                         index = err;
1443         }
1444         if (err < 0)
1445                 goto out_free_disk;
1446
1447         nbd->index = index;
1448         nbd->disk = disk;
1449         nbd->tag_set.ops = &nbd_mq_ops;
1450         nbd->tag_set.nr_hw_queues = 1;
1451         nbd->tag_set.queue_depth = 128;
1452         nbd->tag_set.numa_node = NUMA_NO_NODE;
1453         nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1454         nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1455                 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1456         nbd->tag_set.driver_data = nbd;
1457
1458         err = blk_mq_alloc_tag_set(&nbd->tag_set);
1459         if (err)
1460                 goto out_free_idr;
1461
1462         q = blk_mq_init_queue(&nbd->tag_set);
1463         if (IS_ERR(q)) {
1464                 err = PTR_ERR(q);
1465                 goto out_free_tags;
1466         }
1467         disk->queue = q;
1468
1469         /*
1470          * Tell the block layer that we are not a rotational device
1471          */
1472         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1473         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1474         disk->queue->limits.discard_granularity = 512;
1475         blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1476         blk_queue_max_segment_size(disk->queue, UINT_MAX);
1477         blk_queue_max_segments(disk->queue, USHRT_MAX);
1478         blk_queue_max_hw_sectors(disk->queue, 65536);
1479         disk->queue->limits.max_sectors = 256;
1480
1481         mutex_init(&nbd->config_lock);
1482         refcount_set(&nbd->config_refs, 0);
1483         refcount_set(&nbd->refs, 1);
1484         INIT_LIST_HEAD(&nbd->list);
1485         disk->major = NBD_MAJOR;
1486         disk->first_minor = index << part_shift;
1487         disk->fops = &nbd_fops;
1488         disk->private_data = nbd;
1489         sprintf(disk->disk_name, "nbd%d", index);
1490         add_disk(disk);
1491         nbd_total_devices++;
1492         return index;
1493
1494 out_free_tags:
1495         blk_mq_free_tag_set(&nbd->tag_set);
1496 out_free_idr:
1497         idr_remove(&nbd_index_idr, index);
1498 out_free_disk:
1499         put_disk(disk);
1500 out_free_nbd:
1501         kfree(nbd);
1502 out:
1503         return err;
1504 }
1505
1506 static int find_free_cb(int id, void *ptr, void *data)
1507 {
1508         struct nbd_device *nbd = ptr;
1509         struct nbd_device **found = data;
1510
1511         if (!refcount_read(&nbd->config_refs)) {
1512                 *found = nbd;
1513                 return 1;
1514         }
1515         return 0;
1516 }
1517
1518 /* Netlink interface. */
1519 static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1520         [NBD_ATTR_INDEX]                =       { .type = NLA_U32 },
1521         [NBD_ATTR_SIZE_BYTES]           =       { .type = NLA_U64 },
1522         [NBD_ATTR_BLOCK_SIZE_BYTES]     =       { .type = NLA_U64 },
1523         [NBD_ATTR_TIMEOUT]              =       { .type = NLA_U64 },
1524         [NBD_ATTR_SERVER_FLAGS]         =       { .type = NLA_U64 },
1525         [NBD_ATTR_CLIENT_FLAGS]         =       { .type = NLA_U64 },
1526         [NBD_ATTR_SOCKETS]              =       { .type = NLA_NESTED},
1527         [NBD_ATTR_DEAD_CONN_TIMEOUT]    =       { .type = NLA_U64 },
1528         [NBD_ATTR_DEVICE_LIST]          =       { .type = NLA_NESTED},
1529 };
1530
1531 static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1532         [NBD_SOCK_FD]                   =       { .type = NLA_U32 },
1533 };
1534
1535 /* We don't use this right now since we don't parse the incoming list, but we
1536  * still want it here so userspace knows what to expect.
1537  */
1538 static struct nla_policy __attribute__((unused))
1539 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1540         [NBD_DEVICE_INDEX]              =       { .type = NLA_U32 },
1541         [NBD_DEVICE_CONNECTED]          =       { .type = NLA_U8 },
1542 };
1543
1544 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1545 {
1546         struct nbd_device *nbd = NULL;
1547         struct nbd_config *config;
1548         int index = -1;
1549         int ret;
1550         bool put_dev = false;
1551
1552         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1553                 return -EPERM;
1554
1555         if (info->attrs[NBD_ATTR_INDEX])
1556                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1557         if (!info->attrs[NBD_ATTR_SOCKETS]) {
1558                 printk(KERN_ERR "nbd: must specify at least one socket\n");
1559                 return -EINVAL;
1560         }
1561         if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1562                 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1563                 return -EINVAL;
1564         }
1565 again:
1566         mutex_lock(&nbd_index_mutex);
1567         if (index == -1) {
1568                 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1569                 if (ret == 0) {
1570                         int new_index;
1571                         new_index = nbd_dev_add(-1);
1572                         if (new_index < 0) {
1573                                 mutex_unlock(&nbd_index_mutex);
1574                                 printk(KERN_ERR "nbd: failed to add new device\n");
1575                                 return ret;
1576                         }
1577                         nbd = idr_find(&nbd_index_idr, new_index);
1578                 }
1579         } else {
1580                 nbd = idr_find(&nbd_index_idr, index);
1581         }
1582         if (!nbd) {
1583                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1584                        index);
1585                 mutex_unlock(&nbd_index_mutex);
1586                 return -EINVAL;
1587         }
1588         if (!refcount_inc_not_zero(&nbd->refs)) {
1589                 mutex_unlock(&nbd_index_mutex);
1590                 if (index == -1)
1591                         goto again;
1592                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1593                        index);
1594                 return -EINVAL;
1595         }
1596         mutex_unlock(&nbd_index_mutex);
1597
1598         mutex_lock(&nbd->config_lock);
1599         if (refcount_read(&nbd->config_refs)) {
1600                 mutex_unlock(&nbd->config_lock);
1601                 nbd_put(nbd);
1602                 if (index == -1)
1603                         goto again;
1604                 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1605                 return -EBUSY;
1606         }
1607         if (WARN_ON(nbd->config)) {
1608                 mutex_unlock(&nbd->config_lock);
1609                 nbd_put(nbd);
1610                 return -EINVAL;
1611         }
1612         config = nbd->config = nbd_alloc_config();
1613         if (!nbd->config) {
1614                 mutex_unlock(&nbd->config_lock);
1615                 nbd_put(nbd);
1616                 printk(KERN_ERR "nbd: couldn't allocate config\n");
1617                 return -ENOMEM;
1618         }
1619         refcount_set(&nbd->config_refs, 1);
1620         set_bit(NBD_BOUND, &config->runtime_flags);
1621
1622         if (info->attrs[NBD_ATTR_SIZE_BYTES]) {
1623                 u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1624                 nbd_size_set(nbd, config->blksize,
1625                              div64_u64(bytes, config->blksize));
1626         }
1627         if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1628                 u64 bsize =
1629                         nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1630                 nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize));
1631         }
1632         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1633                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1634                 nbd->tag_set.timeout = timeout * HZ;
1635                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1636         }
1637         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1638                 config->dead_conn_timeout =
1639                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1640                 config->dead_conn_timeout *= HZ;
1641         }
1642         if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1643                 config->flags =
1644                         nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1645         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1646                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1647                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1648                         set_bit(NBD_DESTROY_ON_DISCONNECT,
1649                                 &config->runtime_flags);
1650                         put_dev = true;
1651                 }
1652         }
1653
1654         if (info->attrs[NBD_ATTR_SOCKETS]) {
1655                 struct nlattr *attr;
1656                 int rem, fd;
1657
1658                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1659                                     rem) {
1660                         struct nlattr *socks[NBD_SOCK_MAX+1];
1661
1662                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1663                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1664                                 ret = -EINVAL;
1665                                 goto out;
1666                         }
1667                         ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1668                                                nbd_sock_policy, info->extack);
1669                         if (ret != 0) {
1670                                 printk(KERN_ERR "nbd: error processing sock list\n");
1671                                 ret = -EINVAL;
1672                                 goto out;
1673                         }
1674                         if (!socks[NBD_SOCK_FD])
1675                                 continue;
1676                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1677                         ret = nbd_add_socket(nbd, fd, true);
1678                         if (ret)
1679                                 goto out;
1680                 }
1681         }
1682         ret = nbd_start_device(nbd);
1683 out:
1684         mutex_unlock(&nbd->config_lock);
1685         if (!ret) {
1686                 set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1687                 refcount_inc(&nbd->config_refs);
1688                 nbd_connect_reply(info, nbd->index);
1689         }
1690         nbd_config_put(nbd);
1691         if (put_dev)
1692                 nbd_put(nbd);
1693         return ret;
1694 }
1695
1696 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1697 {
1698         struct nbd_device *nbd;
1699         int index;
1700
1701         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1702                 return -EPERM;
1703
1704         if (!info->attrs[NBD_ATTR_INDEX]) {
1705                 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1706                 return -EINVAL;
1707         }
1708         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1709         mutex_lock(&nbd_index_mutex);
1710         nbd = idr_find(&nbd_index_idr, index);
1711         if (!nbd) {
1712                 mutex_unlock(&nbd_index_mutex);
1713                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1714                        index);
1715                 return -EINVAL;
1716         }
1717         if (!refcount_inc_not_zero(&nbd->refs)) {
1718                 mutex_unlock(&nbd_index_mutex);
1719                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1720                        index);
1721                 return -EINVAL;
1722         }
1723         mutex_unlock(&nbd_index_mutex);
1724         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1725                 nbd_put(nbd);
1726                 return 0;
1727         }
1728         mutex_lock(&nbd->config_lock);
1729         nbd_disconnect(nbd);
1730         mutex_unlock(&nbd->config_lock);
1731         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1732                                &nbd->config->runtime_flags))
1733                 nbd_config_put(nbd);
1734         nbd_config_put(nbd);
1735         nbd_put(nbd);
1736         return 0;
1737 }
1738
1739 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1740 {
1741         struct nbd_device *nbd = NULL;
1742         struct nbd_config *config;
1743         int index;
1744         int ret = -EINVAL;
1745         bool put_dev = false;
1746
1747         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1748                 return -EPERM;
1749
1750         if (!info->attrs[NBD_ATTR_INDEX]) {
1751                 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1752                 return -EINVAL;
1753         }
1754         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1755         mutex_lock(&nbd_index_mutex);
1756         nbd = idr_find(&nbd_index_idr, index);
1757         if (!nbd) {
1758                 mutex_unlock(&nbd_index_mutex);
1759                 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1760                        index);
1761                 return -EINVAL;
1762         }
1763         if (!refcount_inc_not_zero(&nbd->refs)) {
1764                 mutex_unlock(&nbd_index_mutex);
1765                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1766                        index);
1767                 return -EINVAL;
1768         }
1769         mutex_unlock(&nbd_index_mutex);
1770
1771         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1772                 dev_err(nbd_to_dev(nbd),
1773                         "not configured, cannot reconfigure\n");
1774                 nbd_put(nbd);
1775                 return -EINVAL;
1776         }
1777
1778         mutex_lock(&nbd->config_lock);
1779         config = nbd->config;
1780         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1781             !nbd->task_recv) {
1782                 dev_err(nbd_to_dev(nbd),
1783                         "not configured, cannot reconfigure\n");
1784                 goto out;
1785         }
1786
1787         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1788                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1789                 nbd->tag_set.timeout = timeout * HZ;
1790                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1791         }
1792         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1793                 config->dead_conn_timeout =
1794                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1795                 config->dead_conn_timeout *= HZ;
1796         }
1797         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1798                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1799                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1800                         if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1801                                               &config->runtime_flags))
1802                                 put_dev = true;
1803                 } else {
1804                         if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1805                                                &config->runtime_flags))
1806                                 refcount_inc(&nbd->refs);
1807                 }
1808         }
1809
1810         if (info->attrs[NBD_ATTR_SOCKETS]) {
1811                 struct nlattr *attr;
1812                 int rem, fd;
1813
1814                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1815                                     rem) {
1816                         struct nlattr *socks[NBD_SOCK_MAX+1];
1817
1818                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1819                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1820                                 ret = -EINVAL;
1821                                 goto out;
1822                         }
1823                         ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1824                                                nbd_sock_policy, info->extack);
1825                         if (ret != 0) {
1826                                 printk(KERN_ERR "nbd: error processing sock list\n");
1827                                 ret = -EINVAL;
1828                                 goto out;
1829                         }
1830                         if (!socks[NBD_SOCK_FD])
1831                                 continue;
1832                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1833                         ret = nbd_reconnect_socket(nbd, fd);
1834                         if (ret) {
1835                                 if (ret == -ENOSPC)
1836                                         ret = 0;
1837                                 goto out;
1838                         }
1839                         dev_info(nbd_to_dev(nbd), "reconnected socket\n");
1840                 }
1841         }
1842 out:
1843         mutex_unlock(&nbd->config_lock);
1844         nbd_config_put(nbd);
1845         nbd_put(nbd);
1846         if (put_dev)
1847                 nbd_put(nbd);
1848         return ret;
1849 }
1850
1851 static const struct genl_ops nbd_connect_genl_ops[] = {
1852         {
1853                 .cmd    = NBD_CMD_CONNECT,
1854                 .policy = nbd_attr_policy,
1855                 .doit   = nbd_genl_connect,
1856         },
1857         {
1858                 .cmd    = NBD_CMD_DISCONNECT,
1859                 .policy = nbd_attr_policy,
1860                 .doit   = nbd_genl_disconnect,
1861         },
1862         {
1863                 .cmd    = NBD_CMD_RECONFIGURE,
1864                 .policy = nbd_attr_policy,
1865                 .doit   = nbd_genl_reconfigure,
1866         },
1867         {
1868                 .cmd    = NBD_CMD_STATUS,
1869                 .policy = nbd_attr_policy,
1870                 .doit   = nbd_genl_status,
1871         },
1872 };
1873
1874 static const struct genl_multicast_group nbd_mcast_grps[] = {
1875         { .name = NBD_GENL_MCAST_GROUP_NAME, },
1876 };
1877
1878 static struct genl_family nbd_genl_family __ro_after_init = {
1879         .hdrsize        = 0,
1880         .name           = NBD_GENL_FAMILY_NAME,
1881         .version        = NBD_GENL_VERSION,
1882         .module         = THIS_MODULE,
1883         .ops            = nbd_connect_genl_ops,
1884         .n_ops          = ARRAY_SIZE(nbd_connect_genl_ops),
1885         .maxattr        = NBD_ATTR_MAX,
1886         .mcgrps         = nbd_mcast_grps,
1887         .n_mcgrps       = ARRAY_SIZE(nbd_mcast_grps),
1888 };
1889
1890 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
1891 {
1892         struct nlattr *dev_opt;
1893         u8 connected = 0;
1894         int ret;
1895
1896         /* This is a little racey, but for status it's ok.  The
1897          * reason we don't take a ref here is because we can't
1898          * take a ref in the index == -1 case as we would need
1899          * to put under the nbd_index_mutex, which could
1900          * deadlock if we are configured to remove ourselves
1901          * once we're disconnected.
1902          */
1903         if (refcount_read(&nbd->config_refs))
1904                 connected = 1;
1905         dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM);
1906         if (!dev_opt)
1907                 return -EMSGSIZE;
1908         ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
1909         if (ret)
1910                 return -EMSGSIZE;
1911         ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
1912                          connected);
1913         if (ret)
1914                 return -EMSGSIZE;
1915         nla_nest_end(reply, dev_opt);
1916         return 0;
1917 }
1918
1919 static int status_cb(int id, void *ptr, void *data)
1920 {
1921         struct nbd_device *nbd = ptr;
1922         return populate_nbd_status(nbd, (struct sk_buff *)data);
1923 }
1924
1925 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
1926 {
1927         struct nlattr *dev_list;
1928         struct sk_buff *reply;
1929         void *reply_head;
1930         size_t msg_size;
1931         int index = -1;
1932         int ret = -ENOMEM;
1933
1934         if (info->attrs[NBD_ATTR_INDEX])
1935                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1936
1937         mutex_lock(&nbd_index_mutex);
1938
1939         msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
1940                                   nla_attr_size(sizeof(u8)));
1941         msg_size *= (index == -1) ? nbd_total_devices : 1;
1942
1943         reply = genlmsg_new(msg_size, GFP_KERNEL);
1944         if (!reply)
1945                 goto out;
1946         reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
1947                                        NBD_CMD_STATUS);
1948         if (!reply_head) {
1949                 nlmsg_free(reply);
1950                 goto out;
1951         }
1952
1953         dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST);
1954         if (index == -1) {
1955                 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
1956                 if (ret) {
1957                         nlmsg_free(reply);
1958                         goto out;
1959                 }
1960         } else {
1961                 struct nbd_device *nbd;
1962                 nbd = idr_find(&nbd_index_idr, index);
1963                 if (nbd) {
1964                         ret = populate_nbd_status(nbd, reply);
1965                         if (ret) {
1966                                 nlmsg_free(reply);
1967                                 goto out;
1968                         }
1969                 }
1970         }
1971         nla_nest_end(reply, dev_list);
1972         genlmsg_end(reply, reply_head);
1973         genlmsg_reply(reply, info);
1974         ret = 0;
1975 out:
1976         mutex_unlock(&nbd_index_mutex);
1977         return ret;
1978 }
1979
1980 static void nbd_connect_reply(struct genl_info *info, int index)
1981 {
1982         struct sk_buff *skb;
1983         void *msg_head;
1984         int ret;
1985
1986         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
1987         if (!skb)
1988                 return;
1989         msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
1990                                      NBD_CMD_CONNECT);
1991         if (!msg_head) {
1992                 nlmsg_free(skb);
1993                 return;
1994         }
1995         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
1996         if (ret) {
1997                 nlmsg_free(skb);
1998                 return;
1999         }
2000         genlmsg_end(skb, msg_head);
2001         genlmsg_reply(skb, info);
2002 }
2003
2004 static void nbd_mcast_index(int index)
2005 {
2006         struct sk_buff *skb;
2007         void *msg_head;
2008         int ret;
2009
2010         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2011         if (!skb)
2012                 return;
2013         msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2014                                      NBD_CMD_LINK_DEAD);
2015         if (!msg_head) {
2016                 nlmsg_free(skb);
2017                 return;
2018         }
2019         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2020         if (ret) {
2021                 nlmsg_free(skb);
2022                 return;
2023         }
2024         genlmsg_end(skb, msg_head);
2025         genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2026 }
2027
2028 static void nbd_dead_link_work(struct work_struct *work)
2029 {
2030         struct link_dead_args *args = container_of(work, struct link_dead_args,
2031                                                    work);
2032         nbd_mcast_index(args->index);
2033         kfree(args);
2034 }
2035
2036 static int __init nbd_init(void)
2037 {
2038         int i;
2039
2040         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2041
2042         if (max_part < 0) {
2043                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2044                 return -EINVAL;
2045         }
2046
2047         part_shift = 0;
2048         if (max_part > 0) {
2049                 part_shift = fls(max_part);
2050
2051                 /*
2052                  * Adjust max_part according to part_shift as it is exported
2053                  * to user space so that user can know the max number of
2054                  * partition kernel should be able to manage.
2055                  *
2056                  * Note that -1 is required because partition 0 is reserved
2057                  * for the whole disk.
2058                  */
2059                 max_part = (1UL << part_shift) - 1;
2060         }
2061
2062         if ((1UL << part_shift) > DISK_MAX_PARTS)
2063                 return -EINVAL;
2064
2065         if (nbds_max > 1UL << (MINORBITS - part_shift))
2066                 return -EINVAL;
2067         recv_workqueue = alloc_workqueue("knbd-recv",
2068                                          WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2069         if (!recv_workqueue)
2070                 return -ENOMEM;
2071
2072         if (register_blkdev(NBD_MAJOR, "nbd")) {
2073                 destroy_workqueue(recv_workqueue);
2074                 return -EIO;
2075         }
2076
2077         if (genl_register_family(&nbd_genl_family)) {
2078                 unregister_blkdev(NBD_MAJOR, "nbd");
2079                 destroy_workqueue(recv_workqueue);
2080                 return -EINVAL;
2081         }
2082         nbd_dbg_init();
2083
2084         mutex_lock(&nbd_index_mutex);
2085         for (i = 0; i < nbds_max; i++)
2086                 nbd_dev_add(i);
2087         mutex_unlock(&nbd_index_mutex);
2088         return 0;
2089 }
2090
2091 static int nbd_exit_cb(int id, void *ptr, void *data)
2092 {
2093         struct list_head *list = (struct list_head *)data;
2094         struct nbd_device *nbd = ptr;
2095
2096         list_add_tail(&nbd->list, list);
2097         return 0;
2098 }
2099
2100 static void __exit nbd_cleanup(void)
2101 {
2102         struct nbd_device *nbd;
2103         LIST_HEAD(del_list);
2104
2105         nbd_dbg_close();
2106
2107         mutex_lock(&nbd_index_mutex);
2108         idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2109         mutex_unlock(&nbd_index_mutex);
2110
2111         while (!list_empty(&del_list)) {
2112                 nbd = list_first_entry(&del_list, struct nbd_device, list);
2113                 list_del_init(&nbd->list);
2114                 if (refcount_read(&nbd->refs) != 1)
2115                         printk(KERN_ERR "nbd: possibly leaking a device\n");
2116                 nbd_put(nbd);
2117         }
2118
2119         idr_destroy(&nbd_index_idr);
2120         genl_unregister_family(&nbd_genl_family);
2121         destroy_workqueue(recv_workqueue);
2122         unregister_blkdev(NBD_MAJOR, "nbd");
2123 }
2124
2125 module_init(nbd_init);
2126 module_exit(nbd_cleanup);
2127
2128 MODULE_DESCRIPTION("Network Block Device");
2129 MODULE_LICENSE("GPL");
2130
2131 module_param(nbds_max, int, 0444);
2132 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2133 module_param(max_part, int, 0444);
2134 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");