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
 630         while (1) {
 631                 cmd = nbd_read_stat(nbd, args->index);
 632                 if (IS_ERR(cmd)) {
 633                         struct nbd_sock *nsock = config->socks[args->index];
 634
 635                         mutex_lock(&nsock->tx_lock);
 636                         nbd_mark_nsock_dead(nbd, nsock, 1);
 637                         mutex_unlock(&nsock->tx_lock);
 638                         break;
 639                 }
 640
 641                 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
 642         }
 643         atomic_dec(&config->recv_threads);
 644         wake_up(&config->recv_wq);
 645         nbd_config_put(nbd);
 646         kfree(args);
 647 }
 648
 649 static void nbd_clear_req(struct request *req, void *data, bool reserved)
 650 {
 651         struct nbd_cmd *cmd;
 652
 653         if (!blk_mq_request_started(req))
 654                 return;
 655         cmd = blk_mq_rq_to_pdu(req);
 656         cmd->status = BLK_STS_IOERR;
 657         blk_mq_complete_request(req);
 658 }
 659
 660 static void nbd_clear_que(struct nbd_device *nbd)
 661 {
 662         blk_mq_quiesce_queue(nbd->disk->queue);
 663         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 664         blk_mq_unquiesce_queue(nbd->disk->queue);
 665         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 666 }
 667
 668 static int find_fallback(struct nbd_device *nbd, int index)
 669 {
 670         struct nbd_config *config = nbd->config;
 671         int new_index = -1;
 672         struct nbd_sock *nsock = config->socks[index];
 673         int fallback = nsock->fallback_index;
 674
 675         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 676                 return new_index;
 677
 678         if (config->num_connections <= 1) {
 679                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 680                                     "Attempted send on invalid socket\n");
 681                 return new_index;
 682         }
 683
 684         if (fallback >= 0 && fallback < config->num_connections &&
 685             !config->socks[fallback]->dead)
 686                 return fallback;
 687
 688         if (nsock->fallback_index < 0 ||
 689             nsock->fallback_index >= config->num_connections ||
 690             config->socks[nsock->fallback_index]->dead) {
 691                 int i;
 692                 for (i = 0; i < config->num_connections; i++) {
 693                         if (i == index)
 694                                 continue;
 695                         if (!config->socks[i]->dead) {
 696                                 new_index = i;
 697                                 break;
 698                         }
 699                 }
 700                 nsock->fallback_index = new_index;
 701                 if (new_index < 0) {
 702                         dev_err_ratelimited(disk_to_dev(nbd->disk),
 703                                             "Dead connection, failed to find a fallback\n");
 704                         return new_index;
 705                 }
 706         }
 707         new_index = nsock->fallback_index;
 708         return new_index;
 709 }
 710
 711 static int wait_for_reconnect(struct nbd_device *nbd)
 712 {
 713         struct nbd_config *config = nbd->config;
 714         if (!config->dead_conn_timeout)
 715                 return 0;
 716         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 717                 return 0;
 718         wait_event_interruptible_timeout(config->conn_wait,
 719                                          atomic_read(&config->live_connections),
 720                                          config->dead_conn_timeout);
 721         return atomic_read(&config->live_connections);
 722 }
 723
 724 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 725 {
 726         struct request *req = blk_mq_rq_from_pdu(cmd);
 727         struct nbd_device *nbd = cmd->nbd;
 728         struct nbd_config *config;
 729         struct nbd_sock *nsock;
 730         int ret;
 731
 732         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 733                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 734                                     "Socks array is empty\n");
 735                 return -EINVAL;
 736         }
 737         config = nbd->config;
 738
 739         if (index >= config->num_connections) {
 740                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 741                                     "Attempted send on invalid socket\n");
 742                 nbd_config_put(nbd);
 743                 return -EINVAL;
 744         }
 745         cmd->status = BLK_STS_OK;
 746 again:
 747         nsock = config->socks[index];
 748         mutex_lock(&nsock->tx_lock);
 749         if (nsock->dead) {
 750                 int old_index = index;
 751                 index = find_fallback(nbd, index);
 752                 mutex_unlock(&nsock->tx_lock);
 753                 if (index < 0) {
 754                         if (wait_for_reconnect(nbd)) {
 755                                 index = old_index;
 756                                 goto again;
 757                         }
 758                         /* All the sockets should already be down at this point,
 759                          * we just want to make sure that DISCONNECTED is set so
 760                          * any requests that come in that were queue'ed waiting
 761                          * for the reconnect timer don't trigger the timer again
 762                          * and instead just error out.
 763                          */
 764                         sock_shutdown(nbd);
 765                         nbd_config_put(nbd);
 766                         return -EIO;
 767                 }
 768                 goto again;
 769         }
 770
 771         /* Handle the case that we have a pending request that was partially
 772          * transmitted that _has_ to be serviced first.  We need to call requeue
 773          * here so that it gets put _after_ the request that is already on the
 774          * dispatch list.
 775          */
 776         if (unlikely(nsock->pending && nsock->pending != req)) {
 777                 blk_mq_requeue_request(req, true);
 778                 ret = 0;
 779                 goto out;
 780         }
 781         /*
 782          * Some failures are related to the link going down, so anything that
 783          * returns EAGAIN can be retried on a different socket.
 784          */
 785         ret = nbd_send_cmd(nbd, cmd, index);
 786         if (ret == -EAGAIN) {
 787                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 788                                     "Request send failed trying another connection\n");
 789                 nbd_mark_nsock_dead(nbd, nsock, 1);
 790                 mutex_unlock(&nsock->tx_lock);
 791                 goto again;
 792         }
 793 out:
 794         mutex_unlock(&nsock->tx_lock);
 795         nbd_config_put(nbd);
 796         return ret;
 797 }
 798
 799 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 800                         const struct blk_mq_queue_data *bd)
 801 {
 802         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 803         int ret;
 804
 805         /*
 806          * Since we look at the bio's to send the request over the network we
 807          * need to make sure the completion work doesn't mark this request done
 808          * before we are done doing our send.  This keeps us from dereferencing
 809          * freed data if we have particularly fast completions (ie we get the
 810          * completion before we exit sock_xmit on the last bvec) or in the case
 811          * that the server is misbehaving (or there was an error) before we're
 812          * done sending everything over the wire.
 813          */
 814         init_completion(&cmd->send_complete);
 815         blk_mq_start_request(bd->rq);
 816
 817         /* We can be called directly from the user space process, which means we
 818          * could possibly have signals pending so our sendmsg will fail.  In
 819          * this case we need to return that we are busy, otherwise error out as
 820          * appropriate.
 821          */
 822         ret = nbd_handle_cmd(cmd, hctx->queue_num);
 823         complete(&cmd->send_complete);
 824
 825         return ret < 0 ? BLK_STS_IOERR : BLK_STS_OK;
 826 }
 827
 828 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
 829                           bool netlink)
 830 {
 831         struct nbd_config *config = nbd->config;
 832         struct socket *sock;
 833         struct nbd_sock **socks;
 834         struct nbd_sock *nsock;
 835         int err;
 836
 837         sock = sockfd_lookup(arg, &err);
 838         if (!sock)
 839                 return err;
 840
 841         if (!netlink && !nbd->task_setup &&
 842             !test_bit(NBD_BOUND, &config->runtime_flags))
 843                 nbd->task_setup = current;
 844
 845         if (!netlink &&
 846             (nbd->task_setup != current ||
 847              test_bit(NBD_BOUND, &config->runtime_flags))) {
 848                 dev_err(disk_to_dev(nbd->disk),
 849                         "Device being setup by another task");
 850                 sockfd_put(sock);
 851                 return -EBUSY;
 852         }
 853
 854         socks = krealloc(config->socks, (config->num_connections + 1) *
 855                          sizeof(struct nbd_sock *), GFP_KERNEL);
 856         if (!socks) {
 857                 sockfd_put(sock);
 858                 return -ENOMEM;
 859         }
 860         nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
 861         if (!nsock) {
 862                 sockfd_put(sock);
 863                 return -ENOMEM;
 864         }
 865
 866         config->socks = socks;
 867
 868         nsock->fallback_index = -1;
 869         nsock->dead = false;
 870         mutex_init(&nsock->tx_lock);
 871         nsock->sock = sock;
 872         nsock->pending = NULL;
 873         nsock->sent = 0;
 874         nsock->cookie = 0;
 875         socks[config->num_connections++] = nsock;
 876         atomic_inc(&config->live_connections);
 877
 878         return 0;
 879 }
 880
 881 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
 882 {
 883         struct nbd_config *config = nbd->config;
 884         struct socket *sock, *old;
 885         struct recv_thread_args *args;
 886         int i;
 887         int err;
 888
 889         sock = sockfd_lookup(arg, &err);
 890         if (!sock)
 891                 return err;
 892
 893         args = kzalloc(sizeof(*args), GFP_KERNEL);
 894         if (!args) {
 895                 sockfd_put(sock);
 896                 return -ENOMEM;
 897         }
 898
 899         for (i = 0; i < config->num_connections; i++) {
 900                 struct nbd_sock *nsock = config->socks[i];
 901
 902                 if (!nsock->dead)
 903                         continue;
 904
 905                 mutex_lock(&nsock->tx_lock);
 906                 if (!nsock->dead) {
 907                         mutex_unlock(&nsock->tx_lock);
 908                         continue;
 909                 }
 910                 sk_set_memalloc(sock->sk);
 911                 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
 912                 atomic_inc(&config->recv_threads);
 913                 refcount_inc(&nbd->config_refs);
 914                 old = nsock->sock;
 915                 nsock->fallback_index = -1;
 916                 nsock->sock = sock;
 917                 nsock->dead = false;
 918                 INIT_WORK(&args->work, recv_work);
 919                 args->index = i;
 920                 args->nbd = nbd;
 921                 nsock->cookie++;
 922                 mutex_unlock(&nsock->tx_lock);
 923                 sockfd_put(old);
 924
 925                 /* We take the tx_mutex in an error path in the recv_work, so we
 926                  * need to queue_work outside of the tx_mutex.
 927                  */
 928                 queue_work(recv_workqueue, &args->work);
 929
 930                 atomic_inc(&config->live_connections);
 931                 wake_up(&config->conn_wait);
 932                 return 0;
 933         }
 934         sockfd_put(sock);
 935         kfree(args);
 936         return -ENOSPC;
 937 }
 938
 939 static void nbd_bdev_reset(struct block_device *bdev)
 940 {
 941         if (bdev->bd_openers > 1)
 942                 return;
 943         bd_set_size(bdev, 0);
 944         if (max_part > 0) {
 945                 blkdev_reread_part(bdev);
 946                 bdev->bd_invalidated = 1;
 947         }
 948 }
 949
 950 static void nbd_parse_flags(struct nbd_device *nbd)
 951 {
 952         struct nbd_config *config = nbd->config;
 953         if (config->flags & NBD_FLAG_READ_ONLY)
 954                 set_disk_ro(nbd->disk, true);
 955         else
 956                 set_disk_ro(nbd->disk, false);
 957         if (config->flags & NBD_FLAG_SEND_TRIM)
 958                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 959         if (config->flags & NBD_FLAG_SEND_FLUSH) {
 960                 if (config->flags & NBD_FLAG_SEND_FUA)
 961                         blk_queue_write_cache(nbd->disk->queue, true, true);
 962                 else
 963                         blk_queue_write_cache(nbd->disk->queue, true, false);
 964         }
 965         else
 966                 blk_queue_write_cache(nbd->disk->queue, false, false);
 967 }
 968
 969 static void send_disconnects(struct nbd_device *nbd)
 970 {
 971         struct nbd_config *config = nbd->config;
 972         struct nbd_request request = {
 973                 .magic = htonl(NBD_REQUEST_MAGIC),
 974                 .type = htonl(NBD_CMD_DISC),
 975         };
 976         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 977         struct iov_iter from;
 978         int i, ret;
 979
 980         for (i = 0; i < config->num_connections; i++) {
 981                 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
 982                 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
 983                 if (ret <= 0)
 984                         dev_err(disk_to_dev(nbd->disk),
 985                                 "Send disconnect failed %d\n", ret);
 986         }
 987 }
 988
 989 static int nbd_disconnect(struct nbd_device *nbd)
 990 {
 991         struct nbd_config *config = nbd->config;
 992
 993         dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
 994         if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED,
 995                               &config->runtime_flags))
 996                 send_disconnects(nbd);
 997         return 0;
 998 }
 999
1000 static void nbd_clear_sock(struct nbd_device *nbd)
1001 {
1002         sock_shutdown(nbd);
1003         nbd_clear_que(nbd);
1004         nbd->task_setup = NULL;
1005 }
1006
1007 static void nbd_config_put(struct nbd_device *nbd)
1008 {
1009         if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1010                                         &nbd->config_lock)) {
1011                 struct nbd_config *config = nbd->config;
1012                 nbd_dev_dbg_close(nbd);
1013                 nbd_size_clear(nbd);
1014                 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1015                                        &config->runtime_flags))
1016                         device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1017                 nbd->task_recv = NULL;
1018                 nbd_clear_sock(nbd);
1019                 if (config->num_connections) {
1020                         int i;
1021                         for (i = 0; i < config->num_connections; i++) {
1022                                 sockfd_put(config->socks[i]->sock);
1023                                 kfree(config->socks[i]);
1024                         }
1025                         kfree(config->socks);
1026                 }
1027                 kfree(nbd->config);
1028                 nbd->config = NULL;
1029
1030                 nbd->tag_set.timeout = 0;
1031                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1032
1033                 mutex_unlock(&nbd->config_lock);
1034                 nbd_put(nbd);
1035                 module_put(THIS_MODULE);
1036         }
1037 }
1038
1039 static int nbd_start_device(struct nbd_device *nbd)
1040 {
1041         struct nbd_config *config = nbd->config;
1042         int num_connections = config->num_connections;
1043         int error = 0, i;
1044
1045         if (nbd->task_recv)
1046                 return -EBUSY;
1047         if (!config->socks)
1048                 return -EINVAL;
1049         if (num_connections > 1 &&
1050             !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1051                 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1052                 return -EINVAL;
1053         }
1054
1055         blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1056         nbd->task_recv = current;
1057
1058         nbd_parse_flags(nbd);
1059
1060         error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1061         if (error) {
1062                 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1063                 return error;
1064         }
1065         set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1066
1067         nbd_dev_dbg_init(nbd);
1068         for (i = 0; i < num_connections; i++) {
1069                 struct recv_thread_args *args;
1070
1071                 args = kzalloc(sizeof(*args), GFP_KERNEL);
1072                 if (!args) {
1073                         sock_shutdown(nbd);
1074                         return -ENOMEM;
1075                 }
1076                 sk_set_memalloc(config->socks[i]->sock->sk);
1077                 config->socks[i]->sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1078                 atomic_inc(&config->recv_threads);
1079                 refcount_inc(&nbd->config_refs);
1080                 INIT_WORK(&args->work, recv_work);
1081                 args->nbd = nbd;
1082                 args->index = i;
1083                 queue_work(recv_workqueue, &args->work);
1084         }
1085         return error;
1086 }
1087
1088 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1089 {
1090         struct nbd_config *config = nbd->config;
1091         int ret;
1092
1093         ret = nbd_start_device(nbd);
1094         if (ret)
1095                 return ret;
1096
1097         bd_set_size(bdev, config->bytesize);
1098         if (max_part)
1099                 bdev->bd_invalidated = 1;
1100         mutex_unlock(&nbd->config_lock);
1101         ret = wait_event_interruptible(config->recv_wq,
1102                                          atomic_read(&config->recv_threads) == 0);
1103         if (ret)
1104                 sock_shutdown(nbd);
1105         mutex_lock(&nbd->config_lock);
1106         bd_set_size(bdev, 0);
1107         /* user requested, ignore socket errors */
1108         if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1109                 ret = 0;
1110         if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1111                 ret = -ETIMEDOUT;
1112         return ret;
1113 }
1114
1115 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1116                                  struct block_device *bdev)
1117 {
1118         sock_shutdown(nbd);
1119         kill_bdev(bdev);
1120         nbd_bdev_reset(bdev);
1121         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1122                                &nbd->config->runtime_flags))
1123                 nbd_config_put(nbd);
1124 }
1125
1126 /* Must be called with config_lock held */
1127 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1128                        unsigned int cmd, unsigned long arg)
1129 {
1130         struct nbd_config *config = nbd->config;
1131
1132         switch (cmd) {
1133         case NBD_DISCONNECT:
1134                 return nbd_disconnect(nbd);
1135         case NBD_CLEAR_SOCK:
1136                 nbd_clear_sock_ioctl(nbd, bdev);
1137                 return 0;
1138         case NBD_SET_SOCK:
1139                 return nbd_add_socket(nbd, arg, false);
1140         case NBD_SET_BLKSIZE:
1141                 nbd_size_set(nbd, arg,
1142                              div_s64(config->bytesize, arg));
1143                 return 0;
1144         case NBD_SET_SIZE:
1145                 nbd_size_set(nbd, config->blksize,
1146                              div_s64(arg, config->blksize));
1147                 return 0;
1148         case NBD_SET_SIZE_BLOCKS:
1149                 nbd_size_set(nbd, config->blksize, arg);
1150                 return 0;
1151         case NBD_SET_TIMEOUT:
1152                 if (arg) {
1153                         nbd->tag_set.timeout = arg * HZ;
1154                         blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1155                 }
1156                 return 0;
1157
1158         case NBD_SET_FLAGS:
1159                 config->flags = arg;
1160                 return 0;
1161         case NBD_DO_IT:
1162                 return nbd_start_device_ioctl(nbd, bdev);
1163         case NBD_CLEAR_QUE:
1164                 /*
1165                  * This is for compatibility only.  The queue is always cleared
1166                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
1167                  */
1168                 return 0;
1169         case NBD_PRINT_DEBUG:
1170                 /*
1171                  * For compatibility only, we no longer keep a list of
1172                  * outstanding requests.
1173                  */
1174                 return 0;
1175         }
1176         return -ENOTTY;
1177 }
1178
1179 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1180                      unsigned int cmd, unsigned long arg)
1181 {
1182         struct nbd_device *nbd = bdev->bd_disk->private_data;
1183         struct nbd_config *config = nbd->config;
1184         int error = -EINVAL;
1185
1186         if (!capable(CAP_SYS_ADMIN))
1187                 return -EPERM;
1188
1189         mutex_lock(&nbd->config_lock);
1190
1191         /* Don't allow ioctl operations on a nbd device that was created with
1192          * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1193          */
1194         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1195             (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1196                 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1197         else
1198                 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1199         mutex_unlock(&nbd->config_lock);
1200         return error;
1201 }
1202
1203 static struct nbd_config *nbd_alloc_config(void)
1204 {
1205         struct nbd_config *config;
1206
1207         config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1208         if (!config)
1209                 return NULL;
1210         atomic_set(&config->recv_threads, 0);
1211         init_waitqueue_head(&config->recv_wq);
1212         init_waitqueue_head(&config->conn_wait);
1213         config->blksize = 1024;
1214         atomic_set(&config->live_connections, 0);
1215         try_module_get(THIS_MODULE);
1216         return config;
1217 }
1218
1219 static int nbd_open(struct block_device *bdev, fmode_t mode)
1220 {
1221         struct nbd_device *nbd;
1222         int ret = 0;
1223
1224         mutex_lock(&nbd_index_mutex);
1225         nbd = bdev->bd_disk->private_data;
1226         if (!nbd) {
1227                 ret = -ENXIO;
1228                 goto out;
1229         }
1230         if (!refcount_inc_not_zero(&nbd->refs)) {
1231                 ret = -ENXIO;
1232                 goto out;
1233         }
1234         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1235                 struct nbd_config *config;
1236
1237                 mutex_lock(&nbd->config_lock);
1238                 if (refcount_inc_not_zero(&nbd->config_refs)) {
1239                         mutex_unlock(&nbd->config_lock);
1240                         goto out;
1241                 }
1242                 config = nbd->config = nbd_alloc_config();
1243                 if (!config) {
1244                         ret = -ENOMEM;
1245                         mutex_unlock(&nbd->config_lock);
1246                         goto out;
1247                 }
1248                 refcount_set(&nbd->config_refs, 1);
1249                 refcount_inc(&nbd->refs);
1250                 mutex_unlock(&nbd->config_lock);
1251         }
1252 out:
1253         mutex_unlock(&nbd_index_mutex);
1254         return ret;
1255 }
1256
1257 static void nbd_release(struct gendisk *disk, fmode_t mode)
1258 {
1259         struct nbd_device *nbd = disk->private_data;
1260         nbd_config_put(nbd);
1261         nbd_put(nbd);
1262 }
1263
1264 static const struct block_device_operations nbd_fops =
1265 {
1266         .owner =        THIS_MODULE,
1267         .open =         nbd_open,
1268         .release =      nbd_release,
1269         .ioctl =        nbd_ioctl,
1270         .compat_ioctl = nbd_ioctl,
1271 };
1272
1273 #if IS_ENABLED(CONFIG_DEBUG_FS)
1274
1275 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1276 {
1277         struct nbd_device *nbd = s->private;
1278
1279         if (nbd->task_recv)
1280                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1281
1282         return 0;
1283 }
1284
1285 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1286 {
1287         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1288 }
1289
1290 static const struct file_operations nbd_dbg_tasks_ops = {
1291         .open = nbd_dbg_tasks_open,
1292         .read = seq_read,
1293         .llseek = seq_lseek,
1294         .release = single_release,
1295 };
1296
1297 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1298 {
1299         struct nbd_device *nbd = s->private;
1300         u32 flags = nbd->config->flags;
1301
1302         seq_printf(s, "Hex: 0x%08x\n\n", flags);
1303
1304         seq_puts(s, "Known flags:\n");
1305
1306         if (flags & NBD_FLAG_HAS_FLAGS)
1307                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1308         if (flags & NBD_FLAG_READ_ONLY)
1309                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1310         if (flags & NBD_FLAG_SEND_FLUSH)
1311                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1312         if (flags & NBD_FLAG_SEND_FUA)
1313                 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1314         if (flags & NBD_FLAG_SEND_TRIM)
1315                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1316
1317         return 0;
1318 }
1319
1320 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1321 {
1322         return single_open(file, nbd_dbg_flags_show, inode->i_private);
1323 }
1324
1325 static const struct file_operations nbd_dbg_flags_ops = {
1326         .open = nbd_dbg_flags_open,
1327         .read = seq_read,
1328         .llseek = seq_lseek,
1329         .release = single_release,
1330 };
1331
1332 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1333 {
1334         struct dentry *dir;
1335         struct nbd_config *config = nbd->config;
1336
1337         if (!nbd_dbg_dir)
1338                 return -EIO;
1339
1340         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1341         if (!dir) {
1342                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1343                         nbd_name(nbd));
1344                 return -EIO;
1345         }
1346         config->dbg_dir = dir;
1347
1348         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1349         debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1350         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1351         debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1352         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1353
1354         return 0;
1355 }
1356
1357 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1358 {
1359         debugfs_remove_recursive(nbd->config->dbg_dir);
1360 }
1361
1362 static int nbd_dbg_init(void)
1363 {
1364         struct dentry *dbg_dir;
1365
1366         dbg_dir = debugfs_create_dir("nbd", NULL);
1367         if (!dbg_dir)
1368                 return -EIO;
1369
1370         nbd_dbg_dir = dbg_dir;
1371
1372         return 0;
1373 }
1374
1375 static void nbd_dbg_close(void)
1376 {
1377         debugfs_remove_recursive(nbd_dbg_dir);
1378 }
1379
1380 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1381
1382 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1383 {
1384         return 0;
1385 }
1386
1387 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1388 {
1389 }
1390
1391 static int nbd_dbg_init(void)
1392 {
1393         return 0;
1394 }
1395
1396 static void nbd_dbg_close(void)
1397 {
1398 }
1399
1400 #endif
1401
1402 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1403                             unsigned int hctx_idx, unsigned int numa_node)
1404 {
1405         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1406         cmd->nbd = set->driver_data;
1407         return 0;
1408 }
1409
1410 static const struct blk_mq_ops nbd_mq_ops = {
1411         .queue_rq       = nbd_queue_rq,
1412         .complete       = nbd_complete_rq,
1413         .init_request   = nbd_init_request,
1414         .timeout        = nbd_xmit_timeout,
1415 };
1416
1417 static int nbd_dev_add(int index)
1418 {
1419         struct nbd_device *nbd;
1420         struct gendisk *disk;
1421         struct request_queue *q;
1422         int err = -ENOMEM;
1423
1424         nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1425         if (!nbd)
1426                 goto out;
1427
1428         disk = alloc_disk(1 << part_shift);
1429         if (!disk)
1430                 goto out_free_nbd;
1431
1432         if (index >= 0) {
1433                 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1434                                 GFP_KERNEL);
1435                 if (err == -ENOSPC)
1436                         err = -EEXIST;
1437         } else {
1438                 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1439                 if (err >= 0)
1440                         index = err;
1441         }
1442         if (err < 0)
1443                 goto out_free_disk;
1444
1445         nbd->index = index;
1446         nbd->disk = disk;
1447         nbd->tag_set.ops = &nbd_mq_ops;
1448         nbd->tag_set.nr_hw_queues = 1;
1449         nbd->tag_set.queue_depth = 128;
1450         nbd->tag_set.numa_node = NUMA_NO_NODE;
1451         nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1452         nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1453                 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1454         nbd->tag_set.driver_data = nbd;
1455
1456         err = blk_mq_alloc_tag_set(&nbd->tag_set);
1457         if (err)
1458                 goto out_free_idr;
1459
1460         q = blk_mq_init_queue(&nbd->tag_set);
1461         if (IS_ERR(q)) {
1462                 err = PTR_ERR(q);
1463                 goto out_free_tags;
1464         }
1465         disk->queue = q;
1466
1467         /*
1468          * Tell the block layer that we are not a rotational device
1469          */
1470         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1471         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1472         disk->queue->limits.discard_granularity = 512;
1473         blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1474         blk_queue_max_segment_size(disk->queue, UINT_MAX);
1475         blk_queue_max_segments(disk->queue, USHRT_MAX);
1476         blk_queue_max_hw_sectors(disk->queue, 65536);
1477         disk->queue->limits.max_sectors = 256;
1478
1479         mutex_init(&nbd->config_lock);
1480         refcount_set(&nbd->config_refs, 0);
1481         refcount_set(&nbd->refs, 1);
1482         INIT_LIST_HEAD(&nbd->list);
1483         disk->major = NBD_MAJOR;
1484         disk->first_minor = index << part_shift;
1485         disk->fops = &nbd_fops;
1486         disk->private_data = nbd;
1487         sprintf(disk->disk_name, "nbd%d", index);
1488         add_disk(disk);
1489         nbd_total_devices++;
1490         return index;
1491
1492 out_free_tags:
1493         blk_mq_free_tag_set(&nbd->tag_set);
1494 out_free_idr:
1495         idr_remove(&nbd_index_idr, index);
1496 out_free_disk:
1497         put_disk(disk);
1498 out_free_nbd:
1499         kfree(nbd);
1500 out:
1501         return err;
1502 }
1503
1504 static int find_free_cb(int id, void *ptr, void *data)
1505 {
1506         struct nbd_device *nbd = ptr;
1507         struct nbd_device **found = data;
1508
1509         if (!refcount_read(&nbd->config_refs)) {
1510                 *found = nbd;
1511                 return 1;
1512         }
1513         return 0;
1514 }
1515
1516 /* Netlink interface. */
1517 static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1518         [NBD_ATTR_INDEX]                =       { .type = NLA_U32 },
1519         [NBD_ATTR_SIZE_BYTES]           =       { .type = NLA_U64 },
1520         [NBD_ATTR_BLOCK_SIZE_BYTES]     =       { .type = NLA_U64 },
1521         [NBD_ATTR_TIMEOUT]              =       { .type = NLA_U64 },
1522         [NBD_ATTR_SERVER_FLAGS]         =       { .type = NLA_U64 },
1523         [NBD_ATTR_CLIENT_FLAGS]         =       { .type = NLA_U64 },
1524         [NBD_ATTR_SOCKETS]              =       { .type = NLA_NESTED},
1525         [NBD_ATTR_DEAD_CONN_TIMEOUT]    =       { .type = NLA_U64 },
1526         [NBD_ATTR_DEVICE_LIST]          =       { .type = NLA_NESTED},
1527 };
1528
1529 static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1530         [NBD_SOCK_FD]                   =       { .type = NLA_U32 },
1531 };
1532
1533 /* We don't use this right now since we don't parse the incoming list, but we
1534  * still want it here so userspace knows what to expect.
1535  */
1536 static struct nla_policy __attribute__((unused))
1537 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1538         [NBD_DEVICE_INDEX]              =       { .type = NLA_U32 },
1539         [NBD_DEVICE_CONNECTED]          =       { .type = NLA_U8 },
1540 };
1541
1542 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1543 {
1544         struct nbd_device *nbd = NULL;
1545         struct nbd_config *config;
1546         int index = -1;
1547         int ret;
1548         bool put_dev = false;
1549
1550         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1551                 return -EPERM;
1552
1553         if (info->attrs[NBD_ATTR_INDEX])
1554                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1555         if (!info->attrs[NBD_ATTR_SOCKETS]) {
1556                 printk(KERN_ERR "nbd: must specify at least one socket\n");
1557                 return -EINVAL;
1558         }
1559         if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1560                 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1561                 return -EINVAL;
1562         }
1563 again:
1564         mutex_lock(&nbd_index_mutex);
1565         if (index == -1) {
1566                 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1567                 if (ret == 0) {
1568                         int new_index;
1569                         new_index = nbd_dev_add(-1);
1570                         if (new_index < 0) {
1571                                 mutex_unlock(&nbd_index_mutex);
1572                                 printk(KERN_ERR "nbd: failed to add new device\n");
1573                                 return ret;
1574                         }
1575                         nbd = idr_find(&nbd_index_idr, new_index);
1576                 }
1577         } else {
1578                 nbd = idr_find(&nbd_index_idr, index);
1579         }
1580         if (!nbd) {
1581                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1582                        index);
1583                 mutex_unlock(&nbd_index_mutex);
1584                 return -EINVAL;
1585         }
1586         if (!refcount_inc_not_zero(&nbd->refs)) {
1587                 mutex_unlock(&nbd_index_mutex);
1588                 if (index == -1)
1589                         goto again;
1590                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1591                        index);
1592                 return -EINVAL;
1593         }
1594         mutex_unlock(&nbd_index_mutex);
1595
1596         mutex_lock(&nbd->config_lock);
1597         if (refcount_read(&nbd->config_refs)) {
1598                 mutex_unlock(&nbd->config_lock);
1599                 nbd_put(nbd);
1600                 if (index == -1)
1601                         goto again;
1602                 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1603                 return -EBUSY;
1604         }
1605         if (WARN_ON(nbd->config)) {
1606                 mutex_unlock(&nbd->config_lock);
1607                 nbd_put(nbd);
1608                 return -EINVAL;
1609         }
1610         config = nbd->config = nbd_alloc_config();
1611         if (!nbd->config) {
1612                 mutex_unlock(&nbd->config_lock);
1613                 nbd_put(nbd);
1614                 printk(KERN_ERR "nbd: couldn't allocate config\n");
1615                 return -ENOMEM;
1616         }
1617         refcount_set(&nbd->config_refs, 1);
1618         set_bit(NBD_BOUND, &config->runtime_flags);
1619
1620         if (info->attrs[NBD_ATTR_SIZE_BYTES]) {
1621                 u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1622                 nbd_size_set(nbd, config->blksize,
1623                              div64_u64(bytes, config->blksize));
1624         }
1625         if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1626                 u64 bsize =
1627                         nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1628                 nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize));
1629         }
1630         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1631                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1632                 nbd->tag_set.timeout = timeout * HZ;
1633                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1634         }
1635         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1636                 config->dead_conn_timeout =
1637                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1638                 config->dead_conn_timeout *= HZ;
1639         }
1640         if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1641                 config->flags =
1642                         nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1643         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1644                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1645                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1646                         set_bit(NBD_DESTROY_ON_DISCONNECT,
1647                                 &config->runtime_flags);
1648                         put_dev = true;
1649                 }
1650         }
1651
1652         if (info->attrs[NBD_ATTR_SOCKETS]) {
1653                 struct nlattr *attr;
1654                 int rem, fd;
1655
1656                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1657                                     rem) {
1658                         struct nlattr *socks[NBD_SOCK_MAX+1];
1659
1660                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1661                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1662                                 ret = -EINVAL;
1663                                 goto out;
1664                         }
1665                         ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1666                                                nbd_sock_policy, info->extack);
1667                         if (ret != 0) {
1668                                 printk(KERN_ERR "nbd: error processing sock list\n");
1669                                 ret = -EINVAL;
1670                                 goto out;
1671                         }
1672                         if (!socks[NBD_SOCK_FD])
1673                                 continue;
1674                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1675                         ret = nbd_add_socket(nbd, fd, true);
1676                         if (ret)
1677                                 goto out;
1678                 }
1679         }
1680         ret = nbd_start_device(nbd);
1681 out:
1682         mutex_unlock(&nbd->config_lock);
1683         if (!ret) {
1684                 set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1685                 refcount_inc(&nbd->config_refs);
1686                 nbd_connect_reply(info, nbd->index);
1687         }
1688         nbd_config_put(nbd);
1689         if (put_dev)
1690                 nbd_put(nbd);
1691         return ret;
1692 }
1693
1694 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1695 {
1696         struct nbd_device *nbd;
1697         int index;
1698
1699         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1700                 return -EPERM;
1701
1702         if (!info->attrs[NBD_ATTR_INDEX]) {
1703                 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1704                 return -EINVAL;
1705         }
1706         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1707         mutex_lock(&nbd_index_mutex);
1708         nbd = idr_find(&nbd_index_idr, index);
1709         if (!nbd) {
1710                 mutex_unlock(&nbd_index_mutex);
1711                 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1712                        index);
1713                 return -EINVAL;
1714         }
1715         if (!refcount_inc_not_zero(&nbd->refs)) {
1716                 mutex_unlock(&nbd_index_mutex);
1717                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1718                        index);
1719                 return -EINVAL;
1720         }
1721         mutex_unlock(&nbd_index_mutex);
1722         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1723                 nbd_put(nbd);
1724                 return 0;
1725         }
1726         mutex_lock(&nbd->config_lock);
1727         nbd_disconnect(nbd);
1728         mutex_unlock(&nbd->config_lock);
1729         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1730                                &nbd->config->runtime_flags))
1731                 nbd_config_put(nbd);
1732         nbd_config_put(nbd);
1733         nbd_put(nbd);
1734         return 0;
1735 }
1736
1737 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1738 {
1739         struct nbd_device *nbd = NULL;
1740         struct nbd_config *config;
1741         int index;
1742         int ret = -EINVAL;
1743         bool put_dev = false;
1744
1745         if (!netlink_capable(skb, CAP_SYS_ADMIN))
1746                 return -EPERM;
1747
1748         if (!info->attrs[NBD_ATTR_INDEX]) {
1749                 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1750                 return -EINVAL;
1751         }
1752         index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1753         mutex_lock(&nbd_index_mutex);
1754         nbd = idr_find(&nbd_index_idr, index);
1755         if (!nbd) {
1756                 mutex_unlock(&nbd_index_mutex);
1757                 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1758                        index);
1759                 return -EINVAL;
1760         }
1761         if (!refcount_inc_not_zero(&nbd->refs)) {
1762                 mutex_unlock(&nbd_index_mutex);
1763                 printk(KERN_ERR "nbd: device at index %d is going down\n",
1764                        index);
1765                 return -EINVAL;
1766         }
1767         mutex_unlock(&nbd_index_mutex);
1768
1769         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1770                 dev_err(nbd_to_dev(nbd),
1771                         "not configured, cannot reconfigure\n");
1772                 nbd_put(nbd);
1773                 return -EINVAL;
1774         }
1775
1776         mutex_lock(&nbd->config_lock);
1777         config = nbd->config;
1778         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1779             !nbd->task_recv) {
1780                 dev_err(nbd_to_dev(nbd),
1781                         "not configured, cannot reconfigure\n");
1782                 goto out;
1783         }
1784
1785         if (info->attrs[NBD_ATTR_TIMEOUT]) {
1786                 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1787                 nbd->tag_set.timeout = timeout * HZ;
1788                 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1789         }
1790         if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1791                 config->dead_conn_timeout =
1792                         nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1793                 config->dead_conn_timeout *= HZ;
1794         }
1795         if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1796                 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1797                 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1798                         if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1799                                               &config->runtime_flags))
1800                                 put_dev = true;
1801                 } else {
1802                         if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1803                                                &config->runtime_flags))
1804                                 refcount_inc(&nbd->refs);
1805                 }
1806         }
1807
1808         if (info->attrs[NBD_ATTR_SOCKETS]) {
1809                 struct nlattr *attr;
1810                 int rem, fd;
1811
1812                 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1813                                     rem) {
1814                         struct nlattr *socks[NBD_SOCK_MAX+1];
1815
1816                         if (nla_type(attr) != NBD_SOCK_ITEM) {
1817                                 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1818                                 ret = -EINVAL;
1819                                 goto out;
1820                         }
1821                         ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1822                                                nbd_sock_policy, info->extack);
1823                         if (ret != 0) {
1824                                 printk(KERN_ERR "nbd: error processing sock list\n");
1825                                 ret = -EINVAL;
1826                                 goto out;
1827                         }
1828                         if (!socks[NBD_SOCK_FD])
1829                                 continue;
1830                         fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1831                         ret = nbd_reconnect_socket(nbd, fd);
1832                         if (ret) {
1833                                 if (ret == -ENOSPC)
1834                                         ret = 0;
1835                                 goto out;
1836                         }
1837                         dev_info(nbd_to_dev(nbd), "reconnected socket\n");
1838                 }
1839         }
1840 out:
1841         mutex_unlock(&nbd->config_lock);
1842         nbd_config_put(nbd);
1843         nbd_put(nbd);
1844         if (put_dev)
1845                 nbd_put(nbd);
1846         return ret;
1847 }
1848
1849 static const struct genl_ops nbd_connect_genl_ops[] = {
1850         {
1851                 .cmd    = NBD_CMD_CONNECT,
1852                 .policy = nbd_attr_policy,
1853                 .doit   = nbd_genl_connect,
1854         },
1855         {
1856                 .cmd    = NBD_CMD_DISCONNECT,
1857                 .policy = nbd_attr_policy,
1858                 .doit   = nbd_genl_disconnect,
1859         },
1860         {
1861                 .cmd    = NBD_CMD_RECONFIGURE,
1862                 .policy = nbd_attr_policy,
1863                 .doit   = nbd_genl_reconfigure,
1864         },
1865         {
1866                 .cmd    = NBD_CMD_STATUS,
1867                 .policy = nbd_attr_policy,
1868                 .doit   = nbd_genl_status,
1869         },
1870 };
1871
1872 static const struct genl_multicast_group nbd_mcast_grps[] = {
1873         { .name = NBD_GENL_MCAST_GROUP_NAME, },
1874 };
1875
1876 static struct genl_family nbd_genl_family __ro_after_init = {
1877         .hdrsize        = 0,
1878         .name           = NBD_GENL_FAMILY_NAME,
1879         .version        = NBD_GENL_VERSION,
1880         .module         = THIS_MODULE,
1881         .ops            = nbd_connect_genl_ops,
1882         .n_ops          = ARRAY_SIZE(nbd_connect_genl_ops),
1883         .maxattr        = NBD_ATTR_MAX,
1884         .mcgrps         = nbd_mcast_grps,
1885         .n_mcgrps       = ARRAY_SIZE(nbd_mcast_grps),
1886 };
1887
1888 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
1889 {
1890         struct nlattr *dev_opt;
1891         u8 connected = 0;
1892         int ret;
1893
1894         /* This is a little racey, but for status it's ok.  The
1895          * reason we don't take a ref here is because we can't
1896          * take a ref in the index == -1 case as we would need
1897          * to put under the nbd_index_mutex, which could
1898          * deadlock if we are configured to remove ourselves
1899          * once we're disconnected.
1900          */
1901         if (refcount_read(&nbd->config_refs))
1902                 connected = 1;
1903         dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM);
1904         if (!dev_opt)
1905                 return -EMSGSIZE;
1906         ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
1907         if (ret)
1908                 return -EMSGSIZE;
1909         ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
1910                          connected);
1911         if (ret)
1912                 return -EMSGSIZE;
1913         nla_nest_end(reply, dev_opt);
1914         return 0;
1915 }
1916
1917 static int status_cb(int id, void *ptr, void *data)
1918 {
1919         struct nbd_device *nbd = ptr;
1920         return populate_nbd_status(nbd, (struct sk_buff *)data);
1921 }
1922
1923 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
1924 {
1925         struct nlattr *dev_list;
1926         struct sk_buff *reply;
1927         void *reply_head;
1928         size_t msg_size;
1929         int index = -1;
1930         int ret = -ENOMEM;
1931
1932         if (info->attrs[NBD_ATTR_INDEX])
1933                 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1934
1935         mutex_lock(&nbd_index_mutex);
1936
1937         msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
1938                                   nla_attr_size(sizeof(u8)));
1939         msg_size *= (index == -1) ? nbd_total_devices : 1;
1940
1941         reply = genlmsg_new(msg_size, GFP_KERNEL);
1942         if (!reply)
1943                 goto out;
1944         reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
1945                                        NBD_CMD_STATUS);
1946         if (!reply_head) {
1947                 nlmsg_free(reply);
1948                 goto out;
1949         }
1950
1951         dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST);
1952         if (index == -1) {
1953                 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
1954                 if (ret) {
1955                         nlmsg_free(reply);
1956                         goto out;
1957                 }
1958         } else {
1959                 struct nbd_device *nbd;
1960                 nbd = idr_find(&nbd_index_idr, index);
1961                 if (nbd) {
1962                         ret = populate_nbd_status(nbd, reply);
1963                         if (ret) {
1964                                 nlmsg_free(reply);
1965                                 goto out;
1966                         }
1967                 }
1968         }
1969         nla_nest_end(reply, dev_list);
1970         genlmsg_end(reply, reply_head);
1971         genlmsg_reply(reply, info);
1972         ret = 0;
1973 out:
1974         mutex_unlock(&nbd_index_mutex);
1975         return ret;
1976 }
1977
1978 static void nbd_connect_reply(struct genl_info *info, int index)
1979 {
1980         struct sk_buff *skb;
1981         void *msg_head;
1982         int ret;
1983
1984         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
1985         if (!skb)
1986                 return;
1987         msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
1988                                      NBD_CMD_CONNECT);
1989         if (!msg_head) {
1990                 nlmsg_free(skb);
1991                 return;
1992         }
1993         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
1994         if (ret) {
1995                 nlmsg_free(skb);
1996                 return;
1997         }
1998         genlmsg_end(skb, msg_head);
1999         genlmsg_reply(skb, info);
2000 }
2001
2002 static void nbd_mcast_index(int index)
2003 {
2004         struct sk_buff *skb;
2005         void *msg_head;
2006         int ret;
2007
2008         skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2009         if (!skb)
2010                 return;
2011         msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2012                                      NBD_CMD_LINK_DEAD);
2013         if (!msg_head) {
2014                 nlmsg_free(skb);
2015                 return;
2016         }
2017         ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2018         if (ret) {
2019                 nlmsg_free(skb);
2020                 return;
2021         }
2022         genlmsg_end(skb, msg_head);
2023         genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2024 }
2025
2026 static void nbd_dead_link_work(struct work_struct *work)
2027 {
2028         struct link_dead_args *args = container_of(work, struct link_dead_args,
2029                                                    work);
2030         nbd_mcast_index(args->index);
2031         kfree(args);
2032 }
2033
2034 static int __init nbd_init(void)
2035 {
2036         int i;
2037
2038         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2039
2040         if (max_part < 0) {
2041                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2042                 return -EINVAL;
2043         }
2044
2045         part_shift = 0;
2046         if (max_part > 0) {
2047                 part_shift = fls(max_part);
2048
2049                 /*
2050                  * Adjust max_part according to part_shift as it is exported
2051                  * to user space so that user can know the max number of
2052                  * partition kernel should be able to manage.
2053                  *
2054                  * Note that -1 is required because partition 0 is reserved
2055                  * for the whole disk.
2056                  */
2057                 max_part = (1UL << part_shift) - 1;
2058         }
2059
2060         if ((1UL << part_shift) > DISK_MAX_PARTS)
2061                 return -EINVAL;
2062
2063         if (nbds_max > 1UL << (MINORBITS - part_shift))
2064                 return -EINVAL;
2065         recv_workqueue = alloc_workqueue("knbd-recv",
2066                                          WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2067         if (!recv_workqueue)
2068                 return -ENOMEM;
2069
2070         if (register_blkdev(NBD_MAJOR, "nbd")) {
2071                 destroy_workqueue(recv_workqueue);
2072                 return -EIO;
2073         }
2074
2075         if (genl_register_family(&nbd_genl_family)) {
2076                 unregister_blkdev(NBD_MAJOR, "nbd");
2077                 destroy_workqueue(recv_workqueue);
2078                 return -EINVAL;
2079         }
2080         nbd_dbg_init();
2081
2082         mutex_lock(&nbd_index_mutex);
2083         for (i = 0; i < nbds_max; i++)
2084                 nbd_dev_add(i);
2085         mutex_unlock(&nbd_index_mutex);
2086         return 0;
2087 }
2088
2089 static int nbd_exit_cb(int id, void *ptr, void *data)
2090 {
2091         struct list_head *list = (struct list_head *)data;
2092         struct nbd_device *nbd = ptr;
2093
2094         list_add_tail(&nbd->list, list);
2095         return 0;
2096 }
2097
2098 static void __exit nbd_cleanup(void)
2099 {
2100         struct nbd_device *nbd;
2101         LIST_HEAD(del_list);
2102
2103         nbd_dbg_close();
2104
2105         mutex_lock(&nbd_index_mutex);
2106         idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2107         mutex_unlock(&nbd_index_mutex);
2108
2109         while (!list_empty(&del_list)) {
2110                 nbd = list_first_entry(&del_list, struct nbd_device, list);
2111                 list_del_init(&nbd->list);
2112                 if (refcount_read(&nbd->refs) != 1)
2113                         printk(KERN_ERR "nbd: possibly leaking a device\n");
2114                 nbd_put(nbd);
2115         }
2116
2117         idr_destroy(&nbd_index_idr);
2118         genl_unregister_family(&nbd_genl_family);
2119         destroy_workqueue(recv_workqueue);
2120         unregister_blkdev(NBD_MAJOR, "nbd");
2121 }
2122
2123 module_init(nbd_init);
2124 module_exit(nbd_cleanup);
2125
2126 MODULE_DESCRIPTION("Network Block Device");
2127 MODULE_LICENSE("GPL");
2128
2129 module_param(nbds_max, int, 0444);
2130 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2131 module_param(max_part, int, 0444);
2132 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");