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