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