drivers/block/rnbd/rnbd-srv.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * RDMA Network Block Driver
   4  *
   5  * Copyright (c) 2014 - 2018 ProfitBricks GmbH. All rights reserved.
   6  * Copyright (c) 2018 - 2019 1&1 IONOS Cloud GmbH. All rights reserved.
   7  * Copyright (c) 2019 - 2020 1&1 IONOS SE. All rights reserved.
   8  */
   9 #undef pr_fmt
  10 #define pr_fmt(fmt) KBUILD_MODNAME " L" __stringify(__LINE__) ": " fmt
  11
  12 #include <linux/module.h>
  13 #include <linux/blkdev.h>
  14
  15 #include "rnbd-srv.h"
  16 #include "rnbd-srv-trace.h"
  17
  18 MODULE_DESCRIPTION("RDMA Network Block Device Server");
  19 MODULE_LICENSE("GPL");
  20
  21 static u16 port_nr = RTRS_PORT;
  22
  23 module_param_named(port_nr, port_nr, ushort, 0444);
  24 MODULE_PARM_DESC(port_nr,
  25                  "The port number the server is listening on (default: "
  26                  __stringify(RTRS_PORT)")");
  27
  28 #define DEFAULT_DEV_SEARCH_PATH "/"
  29
  30 static char dev_search_path[PATH_MAX] = DEFAULT_DEV_SEARCH_PATH;
  31
  32 static int dev_search_path_set(const char *val, const struct kernel_param *kp)
  33 {
  34         const char *p = strrchr(val, '\n') ? : val + strlen(val);
  35
  36         if (strlen(val) >= sizeof(dev_search_path))
  37                 return -EINVAL;
  38
  39         snprintf(dev_search_path, sizeof(dev_search_path), "%.*s",
  40                  (int)(p - val), val);
  41
  42         pr_info("dev_search_path changed to '%s'\n", dev_search_path);
  43
  44         return 0;
  45 }
  46
  47 static struct kparam_string dev_search_path_kparam_str = {
  48         .maxlen = sizeof(dev_search_path),
  49         .string = dev_search_path
  50 };
  51
  52 static const struct kernel_param_ops dev_search_path_ops = {
  53         .set    = dev_search_path_set,
  54         .get    = param_get_string,
  55 };
  56
  57 module_param_cb(dev_search_path, &dev_search_path_ops,
  58                 &dev_search_path_kparam_str, 0444);
  59 MODULE_PARM_DESC(dev_search_path,
  60                  "Sets the dev_search_path. When a device is mapped this path is prepended to the device path from the map device operation.  If %SESSNAME% is specified in a path, then device will be searched in a session namespace. (default: "
  61                  DEFAULT_DEV_SEARCH_PATH ")");
  62
  63 static DEFINE_MUTEX(sess_lock);
  64 static DEFINE_SPINLOCK(dev_lock);
  65
  66 static LIST_HEAD(sess_list);
  67 static LIST_HEAD(dev_list);
  68
  69 struct rnbd_io_private {
  70         struct rtrs_srv_op              *id;
  71         struct rnbd_srv_sess_dev        *sess_dev;
  72 };
  73
  74 static void rnbd_sess_dev_release(struct kref *kref)
  75 {
  76         struct rnbd_srv_sess_dev *sess_dev;
  77
  78         sess_dev = container_of(kref, struct rnbd_srv_sess_dev, kref);
  79         complete(sess_dev->destroy_comp);
  80 }
  81
  82 static inline void rnbd_put_sess_dev(struct rnbd_srv_sess_dev *sess_dev)
  83 {
  84         kref_put(&sess_dev->kref, rnbd_sess_dev_release);
  85 }
  86
  87 static struct rnbd_srv_sess_dev *
  88 rnbd_get_sess_dev(int dev_id, struct rnbd_srv_session *srv_sess)
  89 {
  90         struct rnbd_srv_sess_dev *sess_dev;
  91         int ret = 0;
  92
  93         rcu_read_lock();
  94         sess_dev = xa_load(&srv_sess->index_idr, dev_id);
  95         if (sess_dev)
  96                 ret = kref_get_unless_zero(&sess_dev->kref);
  97         rcu_read_unlock();
  98
  99         if (!ret)
 100                 return ERR_PTR(-ENXIO);
 101
 102         return sess_dev;
 103 }
 104
 105 static void rnbd_dev_bi_end_io(struct bio *bio)
 106 {
 107         struct rnbd_io_private *rnbd_priv = bio->bi_private;
 108         struct rnbd_srv_sess_dev *sess_dev = rnbd_priv->sess_dev;
 109
 110         rnbd_put_sess_dev(sess_dev);
 111         rtrs_srv_resp_rdma(rnbd_priv->id, blk_status_to_errno(bio->bi_status));
 112
 113         kfree(rnbd_priv);
 114         bio_put(bio);
 115 }
 116
 117 static int process_rdma(struct rnbd_srv_session *srv_sess,
 118                         struct rtrs_srv_op *id, void *data, u32 datalen,
 119                         const void *usr, size_t usrlen)
 120 {
 121         const struct rnbd_msg_io *msg = usr;
 122         struct rnbd_io_private *priv;
 123         struct rnbd_srv_sess_dev *sess_dev;
 124         u32 dev_id;
 125         int err;
 126         struct bio *bio;
 127         short prio;
 128
 129         trace_process_rdma(srv_sess, msg, id, datalen, usrlen);
 130
 131         priv = kmalloc(sizeof(*priv), GFP_KERNEL);
 132         if (!priv)
 133                 return -ENOMEM;
 134
 135         dev_id = le32_to_cpu(msg->device_id);
 136
 137         sess_dev = rnbd_get_sess_dev(dev_id, srv_sess);
 138         if (IS_ERR(sess_dev)) {
 139                 pr_err_ratelimited("Got I/O request on session %s for unknown device id %d: %pe\n",
 140                                    srv_sess->sessname, dev_id, sess_dev);
 141                 err = -ENOTCONN;
 142                 goto err;
 143         }
 144
 145         priv->sess_dev = sess_dev;
 146         priv->id = id;
 147
 148         bio = bio_alloc(sess_dev->bdev_handle->bdev, 1,
 149                         rnbd_to_bio_flags(le32_to_cpu(msg->rw)), GFP_KERNEL);
 150         if (bio_add_page(bio, virt_to_page(data), datalen,
 151                         offset_in_page(data)) != datalen) {
 152                 rnbd_srv_err(sess_dev, "Failed to map data to bio\n");
 153                 err = -EINVAL;
 154                 goto bio_put;
 155         }
 156
 157         bio->bi_end_io = rnbd_dev_bi_end_io;
 158         bio->bi_private = priv;
 159         bio->bi_iter.bi_sector = le64_to_cpu(msg->sector);
 160         bio->bi_iter.bi_size = le32_to_cpu(msg->bi_size);
 161         prio = srv_sess->ver < RNBD_PROTO_VER_MAJOR ||
 162                usrlen < sizeof(*msg) ? 0 : le16_to_cpu(msg->prio);
 163         bio_set_prio(bio, prio);
 164
 165         submit_bio(bio);
 166
 167         return 0;
 168
 169 bio_put:
 170         bio_put(bio);
 171         rnbd_put_sess_dev(sess_dev);
 172 err:
 173         kfree(priv);
 174         return err;
 175 }
 176
 177 static void destroy_device(struct kref *kref)
 178 {
 179         struct rnbd_srv_dev *dev = container_of(kref, struct rnbd_srv_dev, kref);
 180
 181         WARN_ONCE(!list_empty(&dev->sess_dev_list),
 182                   "Device %s is being destroyed but still in use!\n",
 183                   dev->name);
 184
 185         spin_lock(&dev_lock);
 186         list_del(&dev->list);
 187         spin_unlock(&dev_lock);
 188
 189         mutex_destroy(&dev->lock);
 190         if (dev->dev_kobj.state_in_sysfs)
 191                 /*
 192                  * Destroy kobj only if it was really created.
 193                  */
 194                 rnbd_srv_destroy_dev_sysfs(dev);
 195         else
 196                 kfree(dev);
 197 }
 198
 199 static void rnbd_put_srv_dev(struct rnbd_srv_dev *dev)
 200 {
 201         kref_put(&dev->kref, destroy_device);
 202 }
 203
 204 void rnbd_destroy_sess_dev(struct rnbd_srv_sess_dev *sess_dev, bool keep_id)
 205 {
 206         DECLARE_COMPLETION_ONSTACK(dc);
 207
 208         if (keep_id)
 209                 /* free the resources for the id but don't  */
 210                 /* allow to re-use the id itself because it */
 211                 /* is still used by the client              */
 212                 xa_cmpxchg(&sess_dev->sess->index_idr, sess_dev->device_id,
 213                            sess_dev, NULL, 0);
 214         else
 215                 xa_erase(&sess_dev->sess->index_idr, sess_dev->device_id);
 216         synchronize_rcu();
 217
 218         sess_dev->destroy_comp = &dc;
 219         rnbd_put_sess_dev(sess_dev);
 220         wait_for_completion(&dc); /* wait for inflights to drop to zero */
 221
 222         bdev_release(sess_dev->bdev_handle);
 223         mutex_lock(&sess_dev->dev->lock);
 224         list_del(&sess_dev->dev_list);
 225         if (!sess_dev->readonly)
 226                 sess_dev->dev->open_write_cnt--;
 227         mutex_unlock(&sess_dev->dev->lock);
 228
 229         rnbd_put_srv_dev(sess_dev->dev);
 230
 231         rnbd_srv_info(sess_dev, "Device closed\n");
 232         kfree(sess_dev);
 233 }
 234
 235 static void destroy_sess(struct rnbd_srv_session *srv_sess)
 236 {
 237         struct rnbd_srv_sess_dev *sess_dev;
 238         unsigned long index;
 239
 240         if (xa_empty(&srv_sess->index_idr))
 241                 goto out;
 242
 243         trace_destroy_sess(srv_sess);
 244
 245         mutex_lock(&srv_sess->lock);
 246         xa_for_each(&srv_sess->index_idr, index, sess_dev)
 247                 rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 248         mutex_unlock(&srv_sess->lock);
 249
 250 out:
 251         xa_destroy(&srv_sess->index_idr);
 252
 253         pr_info("RTRS Session %s disconnected\n", srv_sess->sessname);
 254
 255         mutex_lock(&sess_lock);
 256         list_del(&srv_sess->list);
 257         mutex_unlock(&sess_lock);
 258
 259         mutex_destroy(&srv_sess->lock);
 260         kfree(srv_sess);
 261 }
 262
 263 static int create_sess(struct rtrs_srv_sess *rtrs)
 264 {
 265         struct rnbd_srv_session *srv_sess;
 266         char pathname[NAME_MAX];
 267         int err;
 268
 269         err = rtrs_srv_get_path_name(rtrs, pathname, sizeof(pathname));
 270         if (err) {
 271                 pr_err("rtrs_srv_get_path_name(%s): %d\n", pathname, err);
 272
 273                 return err;
 274         }
 275         srv_sess = kzalloc(sizeof(*srv_sess), GFP_KERNEL);
 276         if (!srv_sess)
 277                 return -ENOMEM;
 278
 279         srv_sess->queue_depth = rtrs_srv_get_queue_depth(rtrs);
 280         xa_init_flags(&srv_sess->index_idr, XA_FLAGS_ALLOC);
 281         mutex_init(&srv_sess->lock);
 282         mutex_lock(&sess_lock);
 283         list_add(&srv_sess->list, &sess_list);
 284         mutex_unlock(&sess_lock);
 285
 286         srv_sess->rtrs = rtrs;
 287         strscpy(srv_sess->sessname, pathname, sizeof(srv_sess->sessname));
 288
 289         rtrs_srv_set_sess_priv(rtrs, srv_sess);
 290
 291         trace_create_sess(srv_sess);
 292
 293         return 0;
 294 }
 295
 296 static int rnbd_srv_link_ev(struct rtrs_srv_sess *rtrs,
 297                              enum rtrs_srv_link_ev ev, void *priv)
 298 {
 299         struct rnbd_srv_session *srv_sess = priv;
 300
 301         switch (ev) {
 302         case RTRS_SRV_LINK_EV_CONNECTED:
 303                 return create_sess(rtrs);
 304
 305         case RTRS_SRV_LINK_EV_DISCONNECTED:
 306                 if (WARN_ON_ONCE(!srv_sess))
 307                         return -EINVAL;
 308
 309                 destroy_sess(srv_sess);
 310                 return 0;
 311
 312         default:
 313                 pr_warn("Received unknown RTRS session event %d from session %s\n",
 314                         ev, srv_sess->sessname);
 315                 return -EINVAL;
 316         }
 317 }
 318
 319 void rnbd_srv_sess_dev_force_close(struct rnbd_srv_sess_dev *sess_dev,
 320                                    struct kobj_attribute *attr)
 321 {
 322         struct rnbd_srv_session *sess = sess_dev->sess;
 323
 324         /* It is already started to close by client's close message. */
 325         if (!mutex_trylock(&sess->lock))
 326                 return;
 327
 328         sess_dev->keep_id = true;
 329         /* first remove sysfs itself to avoid deadlock */
 330         sysfs_remove_file_self(&sess_dev->kobj, &attr->attr);
 331         rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 332         mutex_unlock(&sess->lock);
 333 }
 334
 335 static void process_msg_close(struct rnbd_srv_session *srv_sess,
 336                              void *data, size_t datalen, const void *usr,
 337                              size_t usrlen)
 338 {
 339         const struct rnbd_msg_close *close_msg = usr;
 340         struct rnbd_srv_sess_dev *sess_dev;
 341
 342         trace_process_msg_close(srv_sess, close_msg);
 343
 344         sess_dev = rnbd_get_sess_dev(le32_to_cpu(close_msg->device_id),
 345                                       srv_sess);
 346         if (IS_ERR(sess_dev))
 347                 return;
 348
 349         rnbd_put_sess_dev(sess_dev);
 350         mutex_lock(&srv_sess->lock);
 351         rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 352         mutex_unlock(&srv_sess->lock);
 353 }
 354
 355 static int process_msg_open(struct rnbd_srv_session *srv_sess,
 356                             const void *msg, size_t len,
 357                             void *data, size_t datalen);
 358
 359 static void process_msg_sess_info(struct rnbd_srv_session *srv_sess,
 360                                  const void *msg, size_t len,
 361                                  void *data, size_t datalen);
 362
 363 static int rnbd_srv_rdma_ev(void *priv, struct rtrs_srv_op *id,
 364                             void *data, size_t datalen,
 365                             const void *usr, size_t usrlen)
 366 {
 367         struct rnbd_srv_session *srv_sess = priv;
 368         const struct rnbd_msg_hdr *hdr = usr;
 369         int ret = 0;
 370         u16 type;
 371
 372         if (WARN_ON_ONCE(!srv_sess))
 373                 return -ENODEV;
 374
 375         type = le16_to_cpu(hdr->type);
 376
 377         switch (type) {
 378         case RNBD_MSG_IO:
 379                 return process_rdma(srv_sess, id, data, datalen, usr, usrlen);
 380         case RNBD_MSG_CLOSE:
 381                 process_msg_close(srv_sess, data, datalen, usr, usrlen);
 382                 break;
 383         case RNBD_MSG_OPEN:
 384                 ret = process_msg_open(srv_sess, usr, usrlen, data, datalen);
 385                 break;
 386         case RNBD_MSG_SESS_INFO:
 387                 process_msg_sess_info(srv_sess, usr, usrlen, data, datalen);
 388                 break;
 389         default:
 390                 pr_warn("Received unexpected message type %d from session %s\n",
 391                         type, srv_sess->sessname);
 392                 return -EINVAL;
 393         }
 394
 395         /*
 396          * Since ret is passed to rtrs to handle the failure case, we
 397          * just return 0 at the end otherwise callers in rtrs would call
 398          * send_io_resp_imm again to print redundant err message.
 399          */
 400         rtrs_srv_resp_rdma(id, ret);
 401         return 0;
 402 }
 403
 404 static struct rnbd_srv_sess_dev
 405 *rnbd_sess_dev_alloc(struct rnbd_srv_session *srv_sess)
 406 {
 407         struct rnbd_srv_sess_dev *sess_dev;
 408         int error;
 409
 410         sess_dev = kzalloc(sizeof(*sess_dev), GFP_KERNEL);
 411         if (!sess_dev)
 412                 return ERR_PTR(-ENOMEM);
 413
 414         error = xa_alloc(&srv_sess->index_idr, &sess_dev->device_id, sess_dev,
 415                          xa_limit_32b, GFP_NOWAIT);
 416         if (error < 0) {
 417                 pr_warn("Allocating idr failed, err: %d\n", error);
 418                 kfree(sess_dev);
 419                 return ERR_PTR(error);
 420         }
 421
 422         return sess_dev;
 423 }
 424
 425 static struct rnbd_srv_dev *rnbd_srv_init_srv_dev(struct block_device *bdev)
 426 {
 427         struct rnbd_srv_dev *dev;
 428
 429         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 430         if (!dev)
 431                 return ERR_PTR(-ENOMEM);
 432
 433         snprintf(dev->name, sizeof(dev->name), "%pg", bdev);
 434         kref_init(&dev->kref);
 435         INIT_LIST_HEAD(&dev->sess_dev_list);
 436         mutex_init(&dev->lock);
 437
 438         return dev;
 439 }
 440
 441 static struct rnbd_srv_dev *
 442 rnbd_srv_find_or_add_srv_dev(struct rnbd_srv_dev *new_dev)
 443 {
 444         struct rnbd_srv_dev *dev;
 445
 446         spin_lock(&dev_lock);
 447         list_for_each_entry(dev, &dev_list, list) {
 448                 if (!strncmp(dev->name, new_dev->name, sizeof(dev->name))) {
 449                         if (!kref_get_unless_zero(&dev->kref))
 450                                 /*
 451                                  * We lost the race, device is almost dead.
 452                                  *  Continue traversing to find a valid one.
 453                                  */
 454                                 continue;
 455                         spin_unlock(&dev_lock);
 456                         return dev;
 457                 }
 458         }
 459         list_add(&new_dev->list, &dev_list);
 460         spin_unlock(&dev_lock);
 461
 462         return new_dev;
 463 }
 464
 465 static int rnbd_srv_check_update_open_perm(struct rnbd_srv_dev *srv_dev,
 466                                             struct rnbd_srv_session *srv_sess,
 467                                             enum rnbd_access_mode access_mode)
 468 {
 469         int ret = 0;
 470
 471         mutex_lock(&srv_dev->lock);
 472
 473         switch (access_mode) {
 474         case RNBD_ACCESS_RO:
 475                 break;
 476         case RNBD_ACCESS_RW:
 477                 if (srv_dev->open_write_cnt == 0)  {
 478                         srv_dev->open_write_cnt++;
 479                 } else {
 480                         pr_err("Mapping device '%s' for session %s with RW permissions failed. Device already opened as 'RW' by %d client(s), access mode %s.\n",
 481                                srv_dev->name, srv_sess->sessname,
 482                                srv_dev->open_write_cnt,
 483                                rnbd_access_modes[access_mode].str);
 484                         ret = -EPERM;
 485                 }
 486                 break;
 487         case RNBD_ACCESS_MIGRATION:
 488                 if (srv_dev->open_write_cnt < 2) {
 489                         srv_dev->open_write_cnt++;
 490                 } else {
 491                         pr_err("Mapping device '%s' for session %s with migration permissions failed. Device already opened as 'RW' by %d client(s), access mode %s.\n",
 492                                srv_dev->name, srv_sess->sessname,
 493                                srv_dev->open_write_cnt,
 494                                rnbd_access_modes[access_mode].str);
 495                         ret = -EPERM;
 496                 }
 497                 break;
 498         default:
 499                 pr_err("Received mapping request for device '%s' on session %s with invalid access mode: %d\n",
 500                        srv_dev->name, srv_sess->sessname, access_mode);
 501                 ret = -EINVAL;
 502         }
 503
 504         mutex_unlock(&srv_dev->lock);
 505
 506         return ret;
 507 }
 508
 509 static struct rnbd_srv_dev *
 510 rnbd_srv_get_or_create_srv_dev(struct block_device *bdev,
 511                                 struct rnbd_srv_session *srv_sess,
 512                                 enum rnbd_access_mode access_mode)
 513 {
 514         int ret;
 515         struct rnbd_srv_dev *new_dev, *dev;
 516
 517         new_dev = rnbd_srv_init_srv_dev(bdev);
 518         if (IS_ERR(new_dev))
 519                 return new_dev;
 520
 521         dev = rnbd_srv_find_or_add_srv_dev(new_dev);
 522         if (dev != new_dev)
 523                 kfree(new_dev);
 524
 525         ret = rnbd_srv_check_update_open_perm(dev, srv_sess, access_mode);
 526         if (ret) {
 527                 rnbd_put_srv_dev(dev);
 528                 return ERR_PTR(ret);
 529         }
 530
 531         return dev;
 532 }
 533
 534 static void rnbd_srv_fill_msg_open_rsp(struct rnbd_msg_open_rsp *rsp,
 535                                         struct rnbd_srv_sess_dev *sess_dev)
 536 {
 537         struct block_device *bdev = sess_dev->bdev_handle->bdev;
 538
 539         rsp->hdr.type = cpu_to_le16(RNBD_MSG_OPEN_RSP);
 540         rsp->device_id = cpu_to_le32(sess_dev->device_id);
 541         rsp->nsectors = cpu_to_le64(bdev_nr_sectors(bdev));
 542         rsp->logical_block_size = cpu_to_le16(bdev_logical_block_size(bdev));
 543         rsp->physical_block_size = cpu_to_le16(bdev_physical_block_size(bdev));
 544         rsp->max_segments = cpu_to_le16(bdev_max_segments(bdev));
 545         rsp->max_hw_sectors =
 546                 cpu_to_le32(queue_max_hw_sectors(bdev_get_queue(bdev)));
 547         rsp->max_write_zeroes_sectors =
 548                 cpu_to_le32(bdev_write_zeroes_sectors(bdev));
 549         rsp->max_discard_sectors = cpu_to_le32(bdev_max_discard_sectors(bdev));
 550         rsp->discard_granularity = cpu_to_le32(bdev_discard_granularity(bdev));
 551         rsp->discard_alignment = cpu_to_le32(bdev_discard_alignment(bdev));
 552         rsp->secure_discard = cpu_to_le16(bdev_max_secure_erase_sectors(bdev));
 553         rsp->cache_policy = 0;
 554         if (bdev_write_cache(bdev))
 555                 rsp->cache_policy |= RNBD_WRITEBACK;
 556         if (bdev_fua(bdev))
 557                 rsp->cache_policy |= RNBD_FUA;
 558 }
 559
 560 static struct rnbd_srv_sess_dev *
 561 rnbd_srv_create_set_sess_dev(struct rnbd_srv_session *srv_sess,
 562                               const struct rnbd_msg_open *open_msg,
 563                               struct bdev_handle *handle, bool readonly,
 564                               struct rnbd_srv_dev *srv_dev)
 565 {
 566         struct rnbd_srv_sess_dev *sdev = rnbd_sess_dev_alloc(srv_sess);
 567
 568         if (IS_ERR(sdev))
 569                 return sdev;
 570
 571         kref_init(&sdev->kref);
 572
 573         strscpy(sdev->pathname, open_msg->dev_name, sizeof(sdev->pathname));
 574
 575         sdev->bdev_handle       = handle;
 576         sdev->sess              = srv_sess;
 577         sdev->dev               = srv_dev;
 578         sdev->readonly          = readonly;
 579         sdev->access_mode       = open_msg->access_mode;
 580
 581         return sdev;
 582 }
 583
 584 static char *rnbd_srv_get_full_path(struct rnbd_srv_session *srv_sess,
 585                                      const char *dev_name)
 586 {
 587         char *full_path;
 588         char *a, *b;
 589         int len;
 590
 591         full_path = kmalloc(PATH_MAX, GFP_KERNEL);
 592         if (!full_path)
 593                 return ERR_PTR(-ENOMEM);
 594
 595         /*
 596          * Replace %SESSNAME% with a real session name in order to
 597          * create device namespace.
 598          */
 599         a = strnstr(dev_search_path, "%SESSNAME%", sizeof(dev_search_path));
 600         if (a) {
 601                 len = a - dev_search_path;
 602
 603                 len = snprintf(full_path, PATH_MAX, "%.*s/%s/%s", len,
 604                                dev_search_path, srv_sess->sessname, dev_name);
 605         } else {
 606                 len = snprintf(full_path, PATH_MAX, "%s/%s",
 607                                dev_search_path, dev_name);
 608         }
 609         if (len >= PATH_MAX) {
 610                 pr_err("Too long path: %s, %s, %s\n",
 611                        dev_search_path, srv_sess->sessname, dev_name);
 612                 kfree(full_path);
 613                 return ERR_PTR(-EINVAL);
 614         }
 615
 616         /* eliminitate duplicated slashes */
 617         a = strchr(full_path, '/');
 618         b = a;
 619         while (*b != '\0') {
 620                 if (*b == '/' && *a == '/') {
 621                         b++;
 622                 } else {
 623                         a++;
 624                         *a = *b;
 625                         b++;
 626                 }
 627         }
 628         a++;
 629         *a = '\0';
 630
 631         return full_path;
 632 }
 633
 634 static void process_msg_sess_info(struct rnbd_srv_session *srv_sess,
 635                                  const void *msg, size_t len,
 636                                  void *data, size_t datalen)
 637 {
 638         const struct rnbd_msg_sess_info *sess_info_msg = msg;
 639         struct rnbd_msg_sess_info_rsp *rsp = data;
 640
 641         srv_sess->ver = min_t(u8, sess_info_msg->ver, RNBD_PROTO_VER_MAJOR);
 642
 643         trace_process_msg_sess_info(srv_sess, sess_info_msg);
 644
 645         rsp->hdr.type = cpu_to_le16(RNBD_MSG_SESS_INFO_RSP);
 646         rsp->ver = srv_sess->ver;
 647 }
 648
 649 /**
 650  * find_srv_sess_dev() - a dev is already opened by this name
 651  * @srv_sess:   the session to search.
 652  * @dev_name:   string containing the name of the device.
 653  *
 654  * Return struct rnbd_srv_sess_dev if srv_sess already opened the dev_name
 655  * NULL if the session didn't open the device yet.
 656  */
 657 static struct rnbd_srv_sess_dev *
 658 find_srv_sess_dev(struct rnbd_srv_session *srv_sess, const char *dev_name)
 659 {
 660         struct rnbd_srv_sess_dev *sess_dev;
 661         unsigned long index;
 662
 663         if (xa_empty(&srv_sess->index_idr))
 664                 return NULL;
 665
 666         xa_for_each(&srv_sess->index_idr, index, sess_dev)
 667                 if (!strcmp(sess_dev->pathname, dev_name))
 668                         return sess_dev;
 669
 670         return NULL;
 671 }
 672
 673 static int process_msg_open(struct rnbd_srv_session *srv_sess,
 674                             const void *msg, size_t len,
 675                             void *data, size_t datalen)
 676 {
 677         int ret;
 678         struct rnbd_srv_dev *srv_dev;
 679         struct rnbd_srv_sess_dev *srv_sess_dev;
 680         const struct rnbd_msg_open *open_msg = msg;
 681         struct bdev_handle *bdev_handle;
 682         blk_mode_t open_flags = BLK_OPEN_READ;
 683         char *full_path;
 684         struct rnbd_msg_open_rsp *rsp = data;
 685
 686         trace_process_msg_open(srv_sess, open_msg);
 687
 688         if (open_msg->access_mode != RNBD_ACCESS_RO)
 689                 open_flags |= BLK_OPEN_WRITE;
 690
 691         mutex_lock(&srv_sess->lock);
 692
 693         srv_sess_dev = find_srv_sess_dev(srv_sess, open_msg->dev_name);
 694         if (srv_sess_dev)
 695                 goto fill_response;
 696
 697         if ((strlen(dev_search_path) + strlen(open_msg->dev_name))
 698             >= PATH_MAX) {
 699                 pr_err("Opening device for session %s failed, device path too long. '%s/%s' is longer than PATH_MAX (%d)\n",
 700                        srv_sess->sessname, dev_search_path, open_msg->dev_name,
 701                        PATH_MAX);
 702                 ret = -EINVAL;
 703                 goto reject;
 704         }
 705         if (strstr(open_msg->dev_name, "..")) {
 706                 pr_err("Opening device for session %s failed, device path %s contains relative path ..\n",
 707                        srv_sess->sessname, open_msg->dev_name);
 708                 ret = -EINVAL;
 709                 goto reject;
 710         }
 711         full_path = rnbd_srv_get_full_path(srv_sess, open_msg->dev_name);
 712         if (IS_ERR(full_path)) {
 713                 ret = PTR_ERR(full_path);
 714                 pr_err("Opening device '%s' for client %s failed, failed to get device full path, err: %pe\n",
 715                        open_msg->dev_name, srv_sess->sessname, full_path);
 716                 goto reject;
 717         }
 718
 719         bdev_handle = bdev_open_by_path(full_path, open_flags, NULL, NULL);
 720         if (IS_ERR(bdev_handle)) {
 721                 ret = PTR_ERR(bdev_handle);
 722                 pr_err("Opening device '%s' on session %s failed, failed to open the block device, err: %pe\n",
 723                        full_path, srv_sess->sessname, bdev_handle);
 724                 goto free_path;
 725         }
 726
 727         srv_dev = rnbd_srv_get_or_create_srv_dev(bdev_handle->bdev, srv_sess,
 728                                                   open_msg->access_mode);
 729         if (IS_ERR(srv_dev)) {
 730                 pr_err("Opening device '%s' on session %s failed, creating srv_dev failed, err: %pe\n",
 731                        full_path, srv_sess->sessname, srv_dev);
 732                 ret = PTR_ERR(srv_dev);
 733                 goto blkdev_put;
 734         }
 735
 736         srv_sess_dev = rnbd_srv_create_set_sess_dev(srv_sess, open_msg,
 737                                 bdev_handle,
 738                                 open_msg->access_mode == RNBD_ACCESS_RO,
 739                                 srv_dev);
 740         if (IS_ERR(srv_sess_dev)) {
 741                 pr_err("Opening device '%s' on session %s failed, creating sess_dev failed, err: %pe\n",
 742                        full_path, srv_sess->sessname, srv_sess_dev);
 743                 ret = PTR_ERR(srv_sess_dev);
 744                 goto srv_dev_put;
 745         }
 746
 747         /* Create the srv_dev sysfs files if they haven't been created yet. The
 748          * reason to delay the creation is not to create the sysfs files before
 749          * we are sure the device can be opened.
 750          */
 751         mutex_lock(&srv_dev->lock);
 752         if (!srv_dev->dev_kobj.state_in_sysfs) {
 753                 ret = rnbd_srv_create_dev_sysfs(srv_dev, bdev_handle->bdev);
 754                 if (ret) {
 755                         mutex_unlock(&srv_dev->lock);
 756                         rnbd_srv_err(srv_sess_dev,
 757                                       "Opening device failed, failed to create device sysfs files, err: %d\n",
 758                                       ret);
 759                         goto free_srv_sess_dev;
 760                 }
 761         }
 762
 763         ret = rnbd_srv_create_dev_session_sysfs(srv_sess_dev);
 764         if (ret) {
 765                 mutex_unlock(&srv_dev->lock);
 766                 rnbd_srv_err(srv_sess_dev,
 767                               "Opening device failed, failed to create dev client sysfs files, err: %d\n",
 768                               ret);
 769                 goto free_srv_sess_dev;
 770         }
 771
 772         list_add(&srv_sess_dev->dev_list, &srv_dev->sess_dev_list);
 773         mutex_unlock(&srv_dev->lock);
 774
 775         rnbd_srv_info(srv_sess_dev, "Opened device '%s'\n", srv_dev->name);
 776
 777         kfree(full_path);
 778
 779 fill_response:
 780         rnbd_srv_fill_msg_open_rsp(rsp, srv_sess_dev);
 781         mutex_unlock(&srv_sess->lock);
 782         return 0;
 783
 784 free_srv_sess_dev:
 785         xa_erase(&srv_sess->index_idr, srv_sess_dev->device_id);
 786         synchronize_rcu();
 787         kfree(srv_sess_dev);
 788 srv_dev_put:
 789         if (open_msg->access_mode != RNBD_ACCESS_RO) {
 790                 mutex_lock(&srv_dev->lock);
 791                 srv_dev->open_write_cnt--;
 792                 mutex_unlock(&srv_dev->lock);
 793         }
 794         rnbd_put_srv_dev(srv_dev);
 795 blkdev_put:
 796         bdev_release(bdev_handle);
 797 free_path:
 798         kfree(full_path);
 799 reject:
 800         mutex_unlock(&srv_sess->lock);
 801         return ret;
 802 }
 803
 804 static struct rtrs_srv_ctx *rtrs_ctx;
 805
 806 static struct rtrs_srv_ops rtrs_ops;
 807 static int __init rnbd_srv_init_module(void)
 808 {
 809         int err = 0;
 810
 811         BUILD_BUG_ON(sizeof(struct rnbd_msg_hdr) != 4);
 812         BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info) != 36);
 813         BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info_rsp) != 36);
 814         BUILD_BUG_ON(sizeof(struct rnbd_msg_open) != 264);
 815         BUILD_BUG_ON(sizeof(struct rnbd_msg_close) != 8);
 816         BUILD_BUG_ON(sizeof(struct rnbd_msg_open_rsp) != 56);
 817         rtrs_ops = (struct rtrs_srv_ops) {
 818                 .rdma_ev = rnbd_srv_rdma_ev,
 819                 .link_ev = rnbd_srv_link_ev,
 820         };
 821         rtrs_ctx = rtrs_srv_open(&rtrs_ops, port_nr);
 822         if (IS_ERR(rtrs_ctx)) {
 823                 pr_err("rtrs_srv_open(), err: %pe\n", rtrs_ctx);
 824                 return PTR_ERR(rtrs_ctx);
 825         }
 826
 827         err = rnbd_srv_create_sysfs_files();
 828         if (err) {
 829                 pr_err("rnbd_srv_create_sysfs_files(), err: %d\n", err);
 830                 rtrs_srv_close(rtrs_ctx);
 831         }
 832
 833         return err;
 834 }
 835
 836 static void __exit rnbd_srv_cleanup_module(void)
 837 {
 838         rtrs_srv_close(rtrs_ctx);
 839         WARN_ON(!list_empty(&sess_list));
 840         rnbd_srv_destroy_sysfs_files();
 841 }
 842
 843 module_init(rnbd_srv_init_module);
 844 module_exit(rnbd_srv_cleanup_module);