drivers/block/nbd.c

   1 /*
   2  * Network block device - make block devices work over TCP
   3  *
   4  * Note that you can not swap over this thing, yet. Seems to work but
   5  * deadlocks sometimes - you can not swap over TCP in general.
   6  *
   7  * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
   8  * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
   9  *
  10  * This file is released under GPLv2 or later.
  11  *
  12  * (part of code stolen from loop.c)
  13  */
  14
  15 #include <linux/major.h>
  16
  17 #include <linux/blkdev.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/sched.h>
  21 #include <linux/sched/mm.h>
  22 #include <linux/fs.h>
  23 #include <linux/bio.h>
  24 #include <linux/stat.h>
  25 #include <linux/errno.h>
  26 #include <linux/file.h>
  27 #include <linux/ioctl.h>
  28 #include <linux/mutex.h>
  29 #include <linux/compiler.h>
  30 #include <linux/err.h>
  31 #include <linux/kernel.h>
  32 #include <linux/slab.h>
  33 #include <net/sock.h>
  34 #include <linux/net.h>
  35 #include <linux/kthread.h>
  36 #include <linux/types.h>
  37 #include <linux/debugfs.h>
  38 #include <linux/blk-mq.h>
  39
  40 #include <linux/uaccess.h>
  41 #include <asm/types.h>
  42
  43 #include <linux/nbd.h>
  44 #include <linux/nbd-netlink.h>
  45 #include <net/genetlink.h>
  46
  47 static DEFINE_IDR(nbd_index_idr);
  48 static DEFINE_MUTEX(nbd_index_mutex);
  49 static int nbd_total_devices = 0;
  50
  51 struct nbd_sock {
  52         struct socket *sock;
  53         struct mutex tx_lock;
  54         struct request *pending;
  55         int sent;
  56         bool dead;
  57         int fallback_index;
  58         int cookie;
  59 };
  60
  61 struct recv_thread_args {
  62         struct work_struct work;
  63         struct nbd_device *nbd;
  64         int index;
  65 };
  66
  67 struct link_dead_args {
  68         struct work_struct work;
  69         int index;
  70 };
  71
  72 #define NBD_TIMEDOUT                    0
  73 #define NBD_DISCONNECT_REQUESTED        1
  74 #define NBD_DISCONNECTED                2
  75 #define NBD_HAS_PID_FILE                3
  76 #define NBD_HAS_CONFIG_REF              4
  77 #define NBD_BOUND                       5
  78 #define NBD_DESTROY_ON_DISCONNECT       6
  79
  80 struct nbd_config {
  81         u32 flags;
  82         unsigned long runtime_flags;
  83         u64 dead_conn_timeout;
  84
  85         struct nbd_sock **socks;
  86         int num_connections;
  87         atomic_t live_connections;
  88         wait_queue_head_t conn_wait;
  89
  90         atomic_t recv_threads;
  91         wait_queue_head_t recv_wq;
  92         loff_t blksize;
  93         loff_t bytesize;
  94 #if IS_ENABLED(CONFIG_DEBUG_FS)
  95         struct dentry *dbg_dir;
  96 #endif
  97 };
  98
  99 struct nbd_device {
 100         struct blk_mq_tag_set tag_set;
 101
 102         int index;
 103         refcount_t config_refs;
 104         refcount_t refs;
 105         struct nbd_config *config;
 106         struct mutex config_lock;
 107         struct gendisk *disk;
 108
 109         struct list_head list;
 110         struct task_struct *task_recv;
 111         struct task_struct *task_setup;
 112 };
 113
 114 struct nbd_cmd {
 115         struct nbd_device *nbd;
 116         int index;
 117         int cookie;
 118         struct completion send_complete;
 119         int status;
 120 };
 121
 122 #if IS_ENABLED(CONFIG_DEBUG_FS)
 123 static struct dentry *nbd_dbg_dir;
 124 #endif
 125
 126 #define nbd_name(nbd) ((nbd)->disk->disk_name)
 127
 128 #define NBD_MAGIC 0x68797548
 129
 130 static unsigned int nbds_max = 16;
 131 static int max_part;
 132 static struct workqueue_struct *recv_workqueue;
 133 static int part_shift;
 134
 135 static int nbd_dev_dbg_init(struct nbd_device *nbd);
 136 static void nbd_dev_dbg_close(struct nbd_device *nbd);
 137 static void nbd_config_put(struct nbd_device *nbd);
 138 static void nbd_connect_reply(struct genl_info *info, int index);
 139 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
 140 static void nbd_dead_link_work(struct work_struct *work);
 141
 142 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
 143 {
 144         return disk_to_dev(nbd->disk);
 145 }
 146
 147 static const char *nbdcmd_to_ascii(int cmd)
 148 {
 149         switch (cmd) {
 150         case  NBD_CMD_READ: return "read";
 151         case NBD_CMD_WRITE: return "write";
 152         case  NBD_CMD_DISC: return "disconnect";
 153         case NBD_CMD_FLUSH: return "flush";
 154         case  NBD_CMD_TRIM: return "trim/discard";
 155         }
 156         return "invalid";
 157 }
 158
 159 static ssize_t pid_show(struct device *dev,
 160                         struct device_attribute *attr, char *buf)
 161 {
 162         struct gendisk *disk = dev_to_disk(dev);
 163         struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
 164
 165         return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
 166 }
 167
 168 static struct device_attribute pid_attr = {
 169         .attr = { .name = "pid", .mode = S_IRUGO},
 170         .show = pid_show,
 171 };
 172
 173 static void nbd_dev_remove(struct nbd_device *nbd)
 174 {
 175         struct gendisk *disk = nbd->disk;
 176         if (disk) {
 177                 del_gendisk(disk);
 178                 blk_cleanup_queue(disk->queue);
 179                 blk_mq_free_tag_set(&nbd->tag_set);
 180                 disk->private_data = NULL;
 181                 put_disk(disk);
 182         }
 183         kfree(nbd);
 184 }
 185
 186 static void nbd_put(struct nbd_device *nbd)
 187 {
 188         if (refcount_dec_and_mutex_lock(&nbd->refs,
 189                                         &nbd_index_mutex)) {
 190                 idr_remove(&nbd_index_idr, nbd->index);
 191                 mutex_unlock(&nbd_index_mutex);
 192                 nbd_dev_remove(nbd);
 193         }
 194 }
 195
 196 static int nbd_disconnected(struct nbd_config *config)
 197 {
 198         return test_bit(NBD_DISCONNECTED, &config->runtime_flags) ||
 199                 test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
 200 }
 201
 202 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
 203                                 int notify)
 204 {
 205         if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
 206                 struct link_dead_args *args;
 207                 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
 208                 if (args) {
 209                         INIT_WORK(&args->work, nbd_dead_link_work);
 210                         args->index = nbd->index;
 211                         queue_work(system_wq, &args->work);
 212                 }
 213         }
 214         if (!nsock->dead) {
 215                 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
 216                 atomic_dec(&nbd->config->live_connections);
 217         }
 218         nsock->dead = true;
 219         nsock->pending = NULL;
 220         nsock->sent = 0;
 221 }
 222
 223 static void nbd_size_clear(struct nbd_device *nbd)
 224 {
 225         if (nbd->config->bytesize) {
 226                 set_capacity(nbd->disk, 0);
 227                 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 228         }
 229 }
 230
 231 static void nbd_size_update(struct nbd_device *nbd)
 232 {
 233         struct nbd_config *config = nbd->config;
 234         blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
 235         blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
 236         set_capacity(nbd->disk, config->bytesize >> 9);
 237         kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
 238 }
 239
 240 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
 241                          loff_t nr_blocks)
 242 {
 243         struct nbd_config *config = nbd->config;
 244         config->blksize = blocksize;
 245         config->bytesize = blocksize * nr_blocks;
 246         nbd_size_update(nbd);
 247 }
 248
 249 static void nbd_complete_rq(struct request *req)
 250 {
 251         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 252
 253         dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", cmd,
 254                 cmd->status ? "failed" : "done");
 255
 256         blk_mq_end_request(req, cmd->status);
 257 }
 258
 259 /*
 260  * Forcibly shutdown the socket causing all listeners to error
 261  */
 262 static void sock_shutdown(struct nbd_device *nbd)
 263 {
 264         struct nbd_config *config = nbd->config;
 265         int i;
 266
 267         if (config->num_connections == 0)
 268                 return;
 269         if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
 270                 return;
 271
 272         for (i = 0; i < config->num_connections; i++) {
 273                 struct nbd_sock *nsock = config->socks[i];
 274                 mutex_lock(&nsock->tx_lock);
 275                 nbd_mark_nsock_dead(nbd, nsock, 0);
 276                 mutex_unlock(&nsock->tx_lock);
 277         }
 278         dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
 279 }
 280
 281 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 282                                                  bool reserved)
 283 {
 284         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 285         struct nbd_device *nbd = cmd->nbd;
 286         struct nbd_config *config;
 287
 288         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 289                 cmd->status = -EIO;
 290                 return BLK_EH_HANDLED;
 291         }
 292
 293         /* If we are waiting on our dead timer then we could get timeout
 294          * callbacks for our request.  For this we just want to reset the timer
 295          * and let the queue side take care of everything.
 296          */
 297         if (!completion_done(&cmd->send_complete)) {
 298                 nbd_config_put(nbd);
 299                 return BLK_EH_RESET_TIMER;
 300         }
 301         config = nbd->config;
 302
 303         if (config->num_connections > 1) {
 304                 dev_err_ratelimited(nbd_to_dev(nbd),
 305                                     "Connection timed out, retrying\n");
 306                 /*
 307                  * Hooray we have more connections, requeue this IO, the submit
 308                  * path will put it on a real connection.
 309                  */
 310                 if (config->socks && config->num_connections > 1) {
 311                         if (cmd->index < config->num_connections) {
 312                                 struct nbd_sock *nsock =
 313                                         config->socks[cmd->index];
 314                                 mutex_lock(&nsock->tx_lock);
 315                                 /* We can have multiple outstanding requests, so
 316                                  * we don't want to mark the nsock dead if we've
 317                                  * already reconnected with a new socket, so
 318                                  * only mark it dead if its the same socket we
 319                                  * were sent out on.
 320                                  */
 321                                 if (cmd->cookie == nsock->cookie)
 322                                         nbd_mark_nsock_dead(nbd, nsock, 1);
 323                                 mutex_unlock(&nsock->tx_lock);
 324                         }
 325                         blk_mq_requeue_request(req, true);
 326                         nbd_config_put(nbd);
 327                         return BLK_EH_NOT_HANDLED;
 328                 }
 329         } else {
 330                 dev_err_ratelimited(nbd_to_dev(nbd),
 331                                     "Connection timed out\n");
 332         }
 333         set_bit(NBD_TIMEDOUT, &config->runtime_flags);
 334         cmd->status = -EIO;
 335         sock_shutdown(nbd);
 336         nbd_config_put(nbd);
 337
 338         return BLK_EH_HANDLED;
 339 }
 340
 341 /*
 342  *  Send or receive packet.
 343  */
 344 static int sock_xmit(struct nbd_device *nbd, int index, int send,
 345                      struct iov_iter *iter, int msg_flags, int *sent)
 346 {
 347         struct nbd_config *config = nbd->config;
 348         struct socket *sock = config->socks[index]->sock;
 349         int result;
 350         struct msghdr msg;
 351         unsigned int noreclaim_flag;
 352
 353         if (unlikely(!sock)) {
 354                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 355                         "Attempted %s on closed socket in sock_xmit\n",
 356                         (send ? "send" : "recv"));
 357                 return -EINVAL;
 358         }
 359
 360         msg.msg_iter = *iter;
 361
 362         noreclaim_flag = memalloc_noreclaim_save();
 363         do {
 364                 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
 365                 msg.msg_name = NULL;
 366                 msg.msg_namelen = 0;
 367                 msg.msg_control = NULL;
 368                 msg.msg_controllen = 0;
 369                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 370
 371                 if (send)
 372                         result = sock_sendmsg(sock, &msg);
 373                 else
 374                         result = sock_recvmsg(sock, &msg, msg.msg_flags);
 375
 376                 if (result <= 0) {
 377                         if (result == 0)
 378                                 result = -EPIPE; /* short read */
 379                         break;
 380                 }
 381                 if (sent)
 382                         *sent += result;
 383         } while (msg_data_left(&msg));
 384
 385         memalloc_noreclaim_restore(noreclaim_flag);
 386
 387         return result;
 388 }
 389
 390 /* always call with the tx_lock held */
 391 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 392 {
 393         struct request *req = blk_mq_rq_from_pdu(cmd);
 394         struct nbd_config *config = nbd->config;
 395         struct nbd_sock *nsock = config->socks[index];
 396         int result;
 397         struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
 398         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 399         struct iov_iter from;
 400         unsigned long size = blk_rq_bytes(req);
 401         struct bio *bio;
 402         u32 type;
 403         u32 tag = blk_mq_unique_tag(req);
 404         int sent = nsock->sent, skip = 0;
 405
 406         iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
 407
 408         switch (req_op(req)) {
 409         case REQ_OP_DISCARD:
 410                 type = NBD_CMD_TRIM;
 411                 break;
 412         case REQ_OP_FLUSH:
 413                 type = NBD_CMD_FLUSH;
 414                 break;
 415         case REQ_OP_WRITE:
 416                 type = NBD_CMD_WRITE;
 417                 break;
 418         case REQ_OP_READ:
 419                 type = NBD_CMD_READ;
 420                 break;
 421         default:
 422                 return -EIO;
 423         }
 424
 425         if (rq_data_dir(req) == WRITE &&
 426             (config->flags & NBD_FLAG_READ_ONLY)) {
 427                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 428                                     "Write on read-only\n");
 429                 return -EIO;
 430         }
 431
 432         /* We did a partial send previously, and we at least sent the whole
 433          * request struct, so just go and send the rest of the pages in the
 434          * request.
 435          */
 436         if (sent) {
 437                 if (sent >= sizeof(request)) {
 438                         skip = sent - sizeof(request);
 439                         goto send_pages;
 440                 }
 441                 iov_iter_advance(&from, sent);
 442         }
 443         cmd->index = index;
 444         cmd->cookie = nsock->cookie;
 445         request.type = htonl(type);
 446         if (type != NBD_CMD_FLUSH) {
 447                 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
 448                 request.len = htonl(size);
 449         }
 450         memcpy(request.handle, &tag, sizeof(tag));
 451
 452         dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
 453                 cmd, nbdcmd_to_ascii(type),
 454                 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
 455         result = sock_xmit(nbd, index, 1, &from,
 456                         (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
 457         if (result <= 0) {
 458                 if (result == -ERESTARTSYS) {
 459                         /* If we havne't sent anything we can just return BUSY,
 460                          * however if we have sent something we need to make
 461                          * sure we only allow this req to be sent until we are
 462                          * completely done.
 463                          */
 464                         if (sent) {
 465                                 nsock->pending = req;
 466                                 nsock->sent = sent;
 467                         }
 468                         return BLK_MQ_RQ_QUEUE_BUSY;
 469                 }
 470                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 471                         "Send control failed (result %d)\n", result);
 472                 return -EAGAIN;
 473         }
 474 send_pages:
 475         if (type != NBD_CMD_WRITE)
 476                 goto out;
 477
 478         bio = req->bio;
 479         while (bio) {
 480                 struct bio *next = bio->bi_next;
 481                 struct bvec_iter iter;
 482                 struct bio_vec bvec;
 483
 484                 bio_for_each_segment(bvec, bio, iter) {
 485                         bool is_last = !next && bio_iter_last(bvec, iter);
 486                         int flags = is_last ? 0 : MSG_MORE;
 487
 488                         dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
 489                                 cmd, bvec.bv_len);
 490                         iov_iter_bvec(&from, ITER_BVEC | WRITE,
 491                                       &bvec, 1, bvec.bv_len);
 492                         if (skip) {
 493                                 if (skip >= iov_iter_count(&from)) {
 494                                         skip -= iov_iter_count(&from);
 495                                         continue;
 496                                 }
 497                                 iov_iter_advance(&from, skip);
 498                                 skip = 0;
 499                         }
 500                         result = sock_xmit(nbd, index, 1, &from, flags, &sent);
 501                         if (result <= 0) {
 502                                 if (result == -ERESTARTSYS) {
 503                                         /* We've already sent the header, we
 504                                          * have no choice but to set pending and
 505                                          * return BUSY.
 506                                          */
 507                                         nsock->pending = req;
 508                                         nsock->sent = sent;
 509                                         return BLK_MQ_RQ_QUEUE_BUSY;
 510                                 }
 511                                 dev_err(disk_to_dev(nbd->disk),
 512                                         "Send data failed (result %d)\n",
 513                                         result);
 514                                 return -EAGAIN;
 515                         }
 516                         /*
 517                          * The completion might already have come in,
 518                          * so break for the last one instead of letting
 519                          * the iterator do it. This prevents use-after-free
 520                          * of the bio.
 521                          */
 522                         if (is_last)
 523                                 break;
 524                 }
 525                 bio = next;
 526         }
 527 out:
 528         nsock->pending = NULL;
 529         nsock->sent = 0;
 530         return 0;
 531 }
 532
 533 /* NULL returned = something went wrong, inform userspace */
 534 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 535 {
 536         struct nbd_config *config = nbd->config;
 537         int result;
 538         struct nbd_reply reply;
 539         struct nbd_cmd *cmd;
 540         struct request *req = NULL;
 541         u16 hwq;
 542         u32 tag;
 543         struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
 544         struct iov_iter to;
 545
 546         reply.magic = 0;
 547         iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply));
 548         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 549         if (result <= 0) {
 550                 if (!nbd_disconnected(config))
 551                         dev_err(disk_to_dev(nbd->disk),
 552                                 "Receive control failed (result %d)\n", result);
 553                 return ERR_PTR(result);
 554         }
 555
 556         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 557                 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
 558                                 (unsigned long)ntohl(reply.magic));
 559                 return ERR_PTR(-EPROTO);
 560         }
 561
 562         memcpy(&tag, reply.handle, sizeof(u32));
 563
 564         hwq = blk_mq_unique_tag_to_hwq(tag);
 565         if (hwq < nbd->tag_set.nr_hw_queues)
 566                 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
 567                                        blk_mq_unique_tag_to_tag(tag));
 568         if (!req || !blk_mq_request_started(req)) {
 569                 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
 570                         tag, req);
 571                 return ERR_PTR(-ENOENT);
 572         }
 573         cmd = blk_mq_rq_to_pdu(req);
 574         if (ntohl(reply.error)) {
 575                 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
 576                         ntohl(reply.error));
 577                 cmd->status = -EIO;
 578                 return cmd;
 579         }
 580
 581         dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
 582         if (rq_data_dir(req) != WRITE) {
 583                 struct req_iterator iter;
 584                 struct bio_vec bvec;
 585
 586                 rq_for_each_segment(bvec, req, iter) {
 587                         iov_iter_bvec(&to, ITER_BVEC | READ,
 588                                       &bvec, 1, bvec.bv_len);
 589                         result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 590                         if (result <= 0) {
 591                                 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 592                                         result);
 593                                 /*
 594                                  * If we've disconnected or we only have 1
 595                                  * connection then we need to make sure we
 596                                  * complete this request, otherwise error out
 597                                  * and let the timeout stuff handle resubmitting
 598                                  * this request onto another connection.
 599                                  */
 600                                 if (nbd_disconnected(config) ||
 601                                     config->num_connections <= 1) {
 602                                         cmd->status = -EIO;
 603                                         return cmd;
 604                                 }
 605                                 return ERR_PTR(-EIO);
 606                         }
 607                         dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 608                                 cmd, bvec.bv_len);
 609                 }
 610         } else {
 611                 /* See the comment in nbd_queue_rq. */
 612                 wait_for_completion(&cmd->send_complete);
 613         }
 614         return cmd;
 615 }
 616
 617 static void recv_work(struct work_struct *work)
 618 {
 619         struct recv_thread_args *args = container_of(work,
 620                                                      struct recv_thread_args,
 621                                                      work);
 622         struct nbd_device *nbd = args->nbd;
 623         struct nbd_config *config = nbd->config;
 624         struct nbd_cmd *cmd;
 625         int ret = 0;
 626
 627         while (1) {
 628                 cmd = nbd_read_stat(nbd, args->index);
 629                 if (IS_ERR(cmd)) {
 630                         struct nbd_sock *nsock = config->socks[args->index];
 631
 632                         mutex_lock(&nsock->tx_lock);
 633                         nbd_mark_nsock_dead(nbd, nsock, 1);
 634                         mutex_unlock(&nsock->tx_lock);
 635                         ret = PTR_ERR(cmd);
 636                         break;
 637                 }
 638
 639                 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
 640         }
 641         atomic_dec(&config->recv_threads);
 642         wake_up(&config->recv_wq);
 643         nbd_config_put(nbd);
 644         kfree(args);
 645 }
 646
 647 static void nbd_clear_req(struct request *req, void *data, bool reserved)
 648 {
 649         struct nbd_cmd *cmd;
 650
 651         if (!blk_mq_request_started(req))
 652                 return;
 653         cmd = blk_mq_rq_to_pdu(req);
 654         cmd->status = -EIO;
 655         blk_mq_complete_request(req);
 656 }
 657
 658 static void nbd_clear_que(struct nbd_device *nbd)
 659 {
 660         blk_mq_stop_hw_queues(nbd->disk->queue);
 661         blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
 662         blk_mq_start_hw_queues(nbd->disk->queue);
 663         dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 664 }
 665
 666 static int find_fallback(struct nbd_device *nbd, int index)
 667 {
 668         struct nbd_config *config = nbd->config;
 669         int new_index = -1;
 670         struct nbd_sock *nsock = config->socks[index];
 671         int fallback = nsock->fallback_index;
 672
 673         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 674                 return new_index;
 675
 676         if (config->num_connections <= 1) {
 677                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 678                                     "Attempted send on invalid socket\n");
 679                 return new_index;
 680         }
 681
 682         if (fallback >= 0 && fallback < config->num_connections &&
 683             !config->socks[fallback]->dead)
 684                 return fallback;
 685
 686         if (nsock->fallback_index < 0 ||
 687             nsock->fallback_index >= config->num_connections ||
 688             config->socks[nsock->fallback_index]->dead) {
 689                 int i;
 690                 for (i = 0; i < config->num_connections; i++) {
 691                         if (i == index)
 692                                 continue;
 693                         if (!config->socks[i]->dead) {
 694                                 new_index = i;
 695                                 break;
 696                         }
 697                 }
 698                 nsock->fallback_index = new_index;
 699                 if (new_index < 0) {
 700                         dev_err_ratelimited(disk_to_dev(nbd->disk),
 701                                             "Dead connection, failed to find a fallback\n");
 702                         return new_index;
 703                 }
 704         }
 705         new_index = nsock->fallback_index;
 706         return new_index;
 707 }
 708
 709 static int wait_for_reconnect(struct nbd_device *nbd)
 710 {
 711         struct nbd_config *config = nbd->config;
 712         if (!config->dead_conn_timeout)
 713                 return 0;
 714         if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
 715                 return 0;
 716         wait_event_interruptible_timeout(config->conn_wait,
 717                                          atomic_read(&config->live_connections),
 718                                          config->dead_conn_timeout);
 719         return atomic_read(&config->live_connections);
 720 }
 721
 722 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 723 {
 724         struct request *req = blk_mq_rq_from_pdu(cmd);
 725         struct nbd_device *nbd = cmd->nbd;
 726         struct nbd_config *config;
 727         struct nbd_sock *nsock;
 728         int ret;
 729
 730         if (!refcount_inc_not_zero(&nbd->config_refs)) {
 731                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 732                                     "Socks array is empty\n");
 733                 return -EINVAL;
 734         }
 735         config = nbd->config;
 736
 737         if (index >= config->num_connections) {
 738                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 739                                     "Attempted send on invalid socket\n");
 740                 nbd_config_put(nbd);
 741                 return -EINVAL;
 742         }
 743         cmd->status = 0;
 744 again:
 745         nsock = config->socks[index];
 746         mutex_lock(&nsock->tx_lock);
 747         if (nsock->dead) {
 748                 int old_index = index;
 749                 index = find_fallback(nbd, index);
 750                 mutex_unlock(&nsock->tx_lock);
 751                 if (index < 0) {
 752                         if (wait_for_reconnect(nbd)) {
 753                                 index = old_index;
 754                                 goto again;
 755                         }
 756                         /* All the sockets should already be down at this point,
 757                          * we just want to make sure that DISCONNECTED is set so
 758                          * any requests that come in that were queue'ed waiting
 759                          * for the reconnect timer don't trigger the timer again
 760                          * and instead just error out.
 761                          */
 762                         sock_shutdown(nbd);
 763                         nbd_config_put(nbd);
 764                         return -EIO;
 765                 }
 766                 goto again;
 767         }
 768
 769         /* Handle the case that we have a pending request that was partially
 770          * transmitted that _has_ to be serviced first.  We need to call requeue
 771          * here so that it gets put _after_ the request that is already on the
 772          * dispatch list.
 773          */
 774         if (unlikely(nsock->pending && nsock->pending != req)) {
 775                 blk_mq_requeue_request(req, true);
 776                 ret = 0;
 777                 goto out;
 778         }
 779         /*
 780          * Some failures are related to the link going down, so anything that
 781          * returns EAGAIN can be retried on a different socket.
 782          */
 783         ret = nbd_send_cmd(nbd, cmd, index);
 784         if (ret == -EAGAIN) {
 785                 dev_err_ratelimited(disk_to_dev(nbd->disk),
 786                                     "Request send failed trying another connection\n");
 787                 nbd_mark_nsock_dead(nbd, nsock, 1);
 788                 mutex_unlock(&nsock->tx_lock);
 789                 goto again;
 790         }
 791 out:
 792         mutex_unlock(&nsock->tx_lock);
 793         nbd_config_put(nbd);
 794         return ret;
 795 }
 796
 797 static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
 798                         const struct blk_mq_queue_data *bd)
 799 {
 800         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 801         int ret;
 802
 803         /*
 804          * Since we look at the bio's to send the request over the network we
 805          * need to make sure the completion work doesn't mark this request done
 806          * before we are done doing our send.  This keeps us from dereferencing
 807          * freed data if we have particularly fast completions (ie we get the
 808          * completion before we exit sock_xmit on the last bvec) or in the case
 809          * that the server is misbehaving (or there was an error) before we're
 810          * done sending everything over the wire.
 811          */
 812         init_completion(&cmd->send_complete);
 813         blk_mq_start_request(bd->rq);
 814
 815         /* We can be called directly from the user space process, which means we
 816          * could possibly have signals pending so our sendmsg will fail.  In
 817          * this case we need to return that we are busy, otherwise error out as
 818          * appropriate.
 819          */
 820         ret = nbd_handle_cmd(cmd, hctx->queue_num);
 821         if (ret < 0)
 822                 ret = BLK_MQ_RQ_QUEUE_ERROR;
 823         if (!ret)
 824                 ret = BLK_MQ_RQ_QUEUE_OK;
 825         complete(&cmd->send_complete);
 826
 827         return ret;
 828 }
 829
 830 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
 831                           bool netlink)
 832 {
 833         struct nbd_config *config = nbd->config;
 834         struct socket *sock;
 835         struct nbd_sock **socks;
 836         struct nbd_sock *nsock;
 837         int err;
 838
 839         sock = sockfd_lookup(arg, &err);
 840         if (!sock)
 841                 return err;
 842
 843         if (!netlink && !nbd->task_setup &&
 844             !test_bit(NBD_BOUND, &config->runtime_flags))
 845                 nbd->task_setup = current;
 846
 847         if (!netlink &&
 848             (nbd->task_setup != current ||
 849              test_bit(NBD_BOUND, &config->runtime_flags))) {
 850                 dev_err(disk_to_dev(nbd->disk),
 851                         "Device being setup by another task");
 852                 sockfd_put(sock);
 853                 return -EBUSY;
 854         }
 855
 856         socks = krealloc(config->socks, (config->num_connections + 1) *
 857                          sizeof(struct nbd_sock *), GFP_KERNEL);
 858         if (!socks) {
 859                 sockfd_put(sock);
 860                 return -ENOMEM;
 861         }
 862         nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
 863         if (!nsock) {
 864                 sockfd_put(sock);
 865                 return -ENOMEM;
 866         }
 867
 868         config->socks = socks;
 869
 870         nsock->fallback_index = -1;
 871         nsock->dead = false;
 872         mutex_init(&nsock->tx_lock);
 873         nsock->sock = sock;
 874         nsock->pending = NULL;
 875         nsock->sent = 0;
 876         nsock->cookie = 0;
 877         socks[config->num_connections++] = nsock;
 878         atomic_inc(&config->live_connections);
 879
 880         return 0;
 881 }
 882
 883 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
 884 {
 885         struct nbd_config *config = nbd->config;
 886         struct socket *sock, *old;
 887         struct recv_thread_args *args;
 888         int i;
 889         int err;
 890
 891         sock = sockfd_lookup(arg, &err);
 892         if (!sock)
 893                 return err;
 894
 895         args = kzalloc(sizeof(*args), GFP_KERNEL);
 896         if (!args) {
 897                 sockfd_put(sock);
 898                 return -ENOMEM;
 899         }
 900
 901         for (i = 0; i < config->num_connections; i++) {
 902                 struct nbd_sock *nsock = config->socks[i];
 903
 904                 if (!nsock->dead)
 905                         continue;
 906
 907                 mutex_lock(&nsock->tx_lock);
 908                 if (!nsock->dead) {
 909                         mutex_unlock(&nsock->tx_lock);
 910                         continue;
 911                 }
 912                 sk_set_memalloc(sock->sk);
 913                 atomic_inc(&config->recv_threads);
 914                 refcount_inc(&nbd->config_refs);
 915                 old = nsock->sock;
 916                 nsock->fallback_index = -1;
 917                 nsock->sock = sock;
 918                 nsock->dead = false;
 919                 INIT_WORK(&args->work, recv_work);
 920                 args->index = i;
 921                 args->nbd = nbd;
 922                 nsock->cookie++;
 923                 mutex_unlock(&nsock->tx_lock);
 924                 sockfd_put(old);
 925
 926                 /* We take the tx_mutex in an error path in the recv_work, so we
 927                  * need to queue_work outside of the tx_mutex.
 928                  */
 929                 queue_work(recv_workqueue, &args->work);
 930
 931                 atomic_inc(&config->live_connections);
 932                 wake_up(&config->conn_wait);
 933                 return 0;
 934         }
 935         sockfd_put(sock);
 936         kfree(args);
 937         return -ENOSPC;
 938 }
 939
 940 /* Reset all properties of an NBD device */
 941 static void nbd_reset(struct nbd_device *nbd)
 942 {
 943         nbd->config = NULL;
 944         nbd->tag_set.timeout = 0;
 945         queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 946 }
 947
 948 static void nbd_bdev_reset(struct block_device *bdev)
 949 {
 950         if (bdev->bd_openers > 1)
 951                 return;
 952         bd_set_size(bdev, 0);
 953         if (max_part > 0) {
 954                 blkdev_reread_part(bdev);
 955                 bdev->bd_invalidated = 1;
 956         }
 957 }
 958
 959 static void nbd_parse_flags(struct nbd_device *nbd)
 960 {
 961         struct nbd_config *config = nbd->config;
 962         if (config->flags & NBD_FLAG_READ_ONLY)
 963                 set_disk_ro(nbd->disk, true);
 964         else
 965                 set_disk_ro(nbd->disk, false);
 966         if (config->flags & NBD_FLAG_SEND_TRIM)
 967                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
 968         if (config->flags & NBD_FLAG_SEND_FLUSH)
 969                 blk_queue_write_cache(nbd->disk->queue, true, false);
 970         else
 971                 blk_queue_write_cache(nbd->disk->queue, false, false);
 972 }
 973
 974 static void send_disconnects(struct nbd_device *nbd)
 975 {
 976         struct nbd_config *config = nbd->config;
 977         struct nbd_request request = {
 978                 .magic = htonl(NBD_REQUEST_MAGIC),
 979                 .type = htonl(NBD_CMD_DISC),
 980         };
 981         struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
 982         struct iov_iter from;
 983         int i, ret;
 984
 985         for (i = 0; i < config->num_connections; i++) {
 986                 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
 987                 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
 988                 if (ret <= 0)
 989                         dev_err(disk_to_dev(nbd->disk),
 990                                 "Send disconnect failed %d\n", ret);
 991         }
 992 }
 993
 994 static int nbd_disconnect(struct nbd_device *nbd)
 995 {
 996         struct nbd_config *config = nbd->config;
 997
 998         dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
 999         if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED,
1000                               &config->runtime_flags))
1001                 send_disconnects(nbd);
1002         return 0;
1003 }
1004
1005 static void nbd_clear_sock(struct nbd_device *nbd)
1006 {
1007         sock_shutdown(nbd);
1008         nbd_clear_que(nbd);
1009         nbd->task_setup = NULL;
1010 }
1011
1012 static void nbd_config_put(struct nbd_device *nbd)
1013 {
1014         if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1015                                         &nbd->config_lock)) {
1016                 struct nbd_config *config = nbd->config;
1017                 nbd_dev_dbg_close(nbd);
1018                 nbd_size_clear(nbd);
1019                 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1020                                        &config->runtime_flags))
1021                         device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1022                 nbd->task_recv = NULL;
1023                 nbd_clear_sock(nbd);
1024                 if (config->num_connections) {
1025                         int i;
1026                         for (i = 0; i < config->num_connections; i++) {
1027                                 sockfd_put(config->socks[i]->sock);
1028                                 kfree(config->socks[i]);
1029                         }
1030                         kfree(config->socks);
1031                 }
1032                 nbd_reset(nbd);
1033
1034                 mutex_unlock(&nbd->config_lock);
1035                 nbd_put(nbd);
1036                 module_put(THIS_MODULE);
1037         }
1038 }
1039
1040 static int nbd_start_device(struct nbd_device *nbd)
1041 {
1042         struct nbd_config *config = nbd->config;
1043         int num_connections = config->num_connections;
1044         int error = 0, i;
1045
1046         if (nbd->task_recv)
1047                 return -EBUSY;
1048         if (!config->socks)
1049                 return -EINVAL;
1050         if (num_connections > 1 &&
1051             !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1052                 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1053                 return -EINVAL;
1054         }
1055
1056         blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1057         nbd->task_recv = current;
1058
1059         nbd_parse_flags(nbd);
1060
1061         error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1062         if (error) {
1063                 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1064                 return error;
1065         }
1066         set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1067
1068         nbd_dev_dbg_init(nbd);
1069         for (i = 0; i < num_connections; i++) {
1070                 struct recv_thread_args *args;
1071
1072                 args = kzalloc(sizeof(*args), GFP_KERNEL);
1073                 if (!args) {
1074                         sock_shutdown(nbd);
1075                         return -ENOMEM;
1076                 }
1077                 sk_set_memalloc(config->socks[i]->sock->sk);
1078                 atomic_inc(&config->recv_threads);
1079                 refcount_inc(&nbd->config_refs);
1080                 INIT_WORK(&args->work, recv_work);
1081                 args->nbd = nbd;
1082                 args->index = i;
1083                 queue_work(recv_workqueue, &args->work);
1084         }
1085         return error;
1086 }
1087
1088 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1089 {
1090         struct nbd_config *config = nbd->config;
1091         int ret;
1092
1093         ret = nbd_start_device(nbd);
1094         if (ret)
1095                 return ret;
1096
1097         bd_set_size(bdev, config->bytesize);
1098         if (max_part)
1099                 bdev->bd_invalidated = 1;
1100         mutex_unlock(&nbd->config_lock);
1101         ret = wait_event_interruptible(config->recv_wq,
1102                                          atomic_read(&config->recv_threads) == 0);
1103         if (ret)
1104                 sock_shutdown(nbd);
1105         mutex_lock(&nbd->config_lock);
1106         bd_set_size(bdev, 0);
1107         /* user requested, ignore socket errors */
1108         if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1109                 ret = 0;
1110         if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1111                 ret = -ETIMEDOUT;
1112         return ret;
1113 }
1114
1115 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1116                                  struct block_device *bdev)
1117 {
1118         sock_shutdown(nbd);
1119         kill_bdev(bdev);
1120         nbd_bdev_reset(bdev);
1121         if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1122                                &nbd->config->runtime_flags))
1123                 nbd_config_put(nbd);
1124 }
1125
1126 /* Must be called with config_lock held */
1127 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1128                        unsigned int cmd, unsigned long arg)
1129 {
1130         struct nbd_config *config = nbd->config;
1131
1132         switch (cmd) {
1133         case NBD_DISCONNECT:
1134                 return nbd_disconnect(nbd);
1135         case NBD_CLEAR_SOCK:
1136                 nbd_clear_sock_ioctl(nbd, bdev);
1137                 return 0;
1138         case NBD_SET_SOCK:
1139                 return nbd_add_socket(nbd, arg, false);
1140         case NBD_SET_BLKSIZE:
1141                 nbd_size_set(nbd, arg,
1142                              div_s64(config->bytesize, arg));
1143                 return 0;
1144         case NBD_SET_SIZE:
1145                 nbd_size_set(nbd, config->blksize,
1146                              div_s64(arg, config->blksize));
1147                 return 0;
1148         case NBD_SET_SIZE_BLOCKS:
1149                 nbd_size_set(nbd, config->blksize, arg);
1150                 return 0;
1151         case NBD_SET_TIMEOUT:
1152                 if (arg) {
1153                         nbd->tag_set.timeout = arg * HZ;
1154                         blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1155                 }
1156                 return 0;
1157
1158         case NBD_SET_FLAGS:
1159                 config->flags = arg;
1160                 return 0;
1161         case NBD_DO_IT:
1162                 return nbd_start_device_ioctl(nbd, bdev);
1163         case NBD_CLEAR_QUE:
1164                 /*
1165                  * This is for compatibility only.  The queue is always cleared
1166                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
1167                  */
1168                 return 0;
1169         case NBD_PRINT_DEBUG:
1170                 /*
1171                  * For compatibility only, we no longer keep a list of
1172                  * outstanding requests.
1173                  */
1174                 return 0;
1175         }
1176         return -ENOTTY;
1177 }
1178
1179 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1180                      unsigned int cmd, unsigned long arg)
1181 {
1182         struct nbd_device *nbd = bdev->bd_disk->private_data;
1183         struct nbd_config *config = nbd->config;
1184         int error = -EINVAL;
1185
1186         if (!capable(CAP_SYS_ADMIN))
1187                 return -EPERM;
1188
1189         mutex_lock(&nbd->config_lock);
1190
1191         /* Don't allow ioctl operations on a nbd device that was created with
1192          * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1193          */
1194         if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1195             (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1196                 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1197         else
1198                 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1199         mutex_unlock(&nbd->config_lock);
1200         return error;
1201 }
1202
1203 static struct nbd_config *nbd_alloc_config(void)
1204 {
1205         struct nbd_config *config;
1206
1207         config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1208         if (!config)
1209                 return NULL;
1210         atomic_set(&config->recv_threads, 0);
1211         init_waitqueue_head(&config->recv_wq);
1212         init_waitqueue_head(&config->conn_wait);
1213         config->blksize = 1024;
1214         atomic_set(&config->live_connections, 0);
1215         try_module_get(THIS_MODULE);
1216         return config;
1217 }
1218
1219 static int nbd_open(struct block_device *bdev, fmode_t mode)
1220 {
1221         struct nbd_device *nbd;
1222         int ret = 0;
1223
1224         mutex_lock(&nbd_index_mutex);
1225         nbd = bdev->bd_disk->private_data;
1226         if (!nbd) {
1227                 ret = -ENXIO;
1228                 goto out;
1229         }
1230         if (!refcount_inc_not_zero(&nbd->refs)) {
1231                 ret = -ENXIO;
1232                 goto out;
1233         }
1234         if (!refcount_inc_not_zero(&nbd->config_refs)) {
1235                 struct nbd_config *config;
1236
1237                 mutex_lock(&nbd->config_lock);
1238                 if (refcount_inc_not_zero(&nbd->config_refs)) {
1239                         mutex_unlock(&nbd->config_lock);
1240                         goto out;
1241                 }
1242                 config = nbd->config = nbd_alloc_config();
1243                 if (!config) {
1244                         ret = -ENOMEM;
1245                         mutex_unlock(&nbd->config_lock);
1246                         goto out;
1247                 }
1248                 refcount_set(&nbd->config_refs, 1);
1249                 refcount_inc(&nbd->refs);
1250                 mutex_unlock(&nbd->config_lock);
1251         }
1252 out:
1253         mutex_unlock(&nbd_index_mutex);
1254         return ret;
1255 }
1256
1257 static void nbd_release(struct gendisk *disk, fmode_t mode)
1258 {
1259         struct nbd_device *nbd = disk->private_data;
1260         nbd_config_put(nbd);
1261         nbd_put(nbd);
1262 }
1263
1264 static const struct block_device_operations nbd_fops =
1265 {
1266         .owner =        THIS_MODULE,
1267         .open =         nbd_open,
1268         .release =      nbd_release,
1269         .ioctl =        nbd_ioctl,
1270         .compat_ioctl = nbd_ioctl,
1271 };
1272
1273 #if IS_ENABLED(CONFIG_DEBUG_FS)
1274
1275 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1276 {
1277         struct nbd_device *nbd = s->private;
1278
1279         if (nbd->task_recv)
1280                 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1281
1282         return 0;
1283 }
1284
1285 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1286 {
1287         return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1288 }
1289
1290 static const struct file_operations nbd_dbg_tasks_ops = {
1291         .open = nbd_dbg_tasks_open,
1292         .read = seq_read,
1293         .llseek = seq_lseek,
1294         .release = single_release,
1295 };
1296
1297 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1298 {
1299         struct nbd_device *nbd = s->private;
1300         u32 flags = nbd->config->flags;
1301
1302         seq_printf(s, "Hex: 0x%08x\n\n", flags);
1303
1304         seq_puts(s, "Known flags:\n");
1305
1306         if (flags & NBD_FLAG_HAS_FLAGS)
1307                 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1308         if (flags & NBD_FLAG_READ_ONLY)
1309                 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1310         if (flags & NBD_FLAG_SEND_FLUSH)
1311                 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1312         if (flags & NBD_FLAG_SEND_TRIM)
1313                 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1314
1315         return 0;
1316 }
1317
1318 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1319 {
1320         return single_open(file, nbd_dbg_flags_show, inode->i_private);
1321 }
1322
1323 static const struct file_operations nbd_dbg_flags_ops = {
1324         .open = nbd_dbg_flags_open,
1325         .read = seq_read,
1326         .llseek = seq_lseek,
1327         .release = single_release,
1328 };
1329
1330 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1331 {
1332         struct dentry *dir;
1333         struct nbd_config *config = nbd->config;
1334
1335         if (!nbd_dbg_dir)
1336                 return -EIO;
1337
1338         dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1339         if (!dir) {
1340                 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1341                         nbd_name(nbd));
1342                 return -EIO;
1343         }
1344         config->dbg_dir = dir;
1345
1346         debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1347         debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1348         debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1349         debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1350         debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1351
1352         return 0;
1353 }
1354
1355 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1356 {
1357         debugfs_remove_recursive(nbd->config->dbg_dir);
1358 }
1359
1360 static int nbd_dbg_init(void)
1361 {
1362         struct dentry *dbg_dir;
1363
1364         dbg_dir = debugfs_create_dir("nbd", NULL);
1365         if (!dbg_dir)
1366                 return -EIO;
1367
1368         nbd_dbg_dir = dbg_dir;
1369
1370         return 0;
1371 }
1372
1373 static void nbd_dbg_close(void)
1374 {
1375         debugfs_remove_recursive(nbd_dbg_dir);
1376 }
1377
1378 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1379
1380 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1381 {
1382         return 0;
1383 }
1384
1385 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1386 {
1387 }
1388
1389 static int nbd_dbg_init(void)
1390 {
1391         return 0;
1392 }
1393
1394 static void nbd_dbg_close(void)
1395 {
1396 }
1397
1398 #endif
1399
1400 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1401                             unsigned int hctx_idx, unsigned int numa_node)
1402 {
1403         struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1404         cmd->nbd = set->driver_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, info->extack);
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, info->extack);
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)");