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-dev.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 void rnbd_endio(void *priv, int error)
  88 {
  89         struct rnbd_io_private *rnbd_priv = priv;
  90         struct rnbd_srv_sess_dev *sess_dev = rnbd_priv->sess_dev;
  91
  92         rnbd_put_sess_dev(sess_dev);
  93
  94         rtrs_srv_resp_rdma(rnbd_priv->id, error);
  95
  96         kfree(priv);
  97 }
  98
  99 static struct rnbd_srv_sess_dev *
 100 rnbd_get_sess_dev(int dev_id, struct rnbd_srv_session *srv_sess)
 101 {
 102         struct rnbd_srv_sess_dev *sess_dev;
 103         int ret = 0;
 104
 105         rcu_read_lock();
 106         sess_dev = xa_load(&srv_sess->index_idr, dev_id);
 107         if (likely(sess_dev))
 108                 ret = kref_get_unless_zero(&sess_dev->kref);
 109         rcu_read_unlock();
 110
 111         if (!sess_dev || !ret)
 112                 return ERR_PTR(-ENXIO);
 113
 114         return sess_dev;
 115 }
 116
 117 static int process_rdma(struct rtrs_srv *sess,
 118                         struct rnbd_srv_session *srv_sess,
 119                         struct rtrs_srv_op *id, void *data, u32 datalen,
 120                         const void *usr, size_t usrlen)
 121 {
 122         const struct rnbd_msg_io *msg = usr;
 123         struct rnbd_io_private *priv;
 124         struct rnbd_srv_sess_dev *sess_dev;
 125         u32 dev_id;
 126         int err;
 127         struct rnbd_dev_blk_io *io;
 128         struct bio *bio;
 129         short prio;
 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\n",
 140                                    srv_sess->sessname, dev_id);
 141                 err = -ENOTCONN;
 142                 goto err;
 143         }
 144
 145         priv->sess_dev = sess_dev;
 146         priv->id = id;
 147
 148         /* Generate bio with pages pointing to the rdma buffer */
 149         bio = rnbd_bio_map_kern(data, sess_dev->rnbd_dev->ibd_bio_set, datalen, GFP_KERNEL);
 150         if (IS_ERR(bio)) {
 151                 err = PTR_ERR(bio);
 152                 rnbd_srv_err(sess_dev, "Failed to generate bio, err: %d\n", err);
 153                 goto sess_dev_put;
 154         }
 155
 156         io = container_of(bio, struct rnbd_dev_blk_io, bio);
 157         io->dev = sess_dev->rnbd_dev;
 158         io->priv = priv;
 159
 160         bio->bi_end_io = rnbd_dev_bi_end_io;
 161         bio->bi_private = io;
 162         bio->bi_opf = rnbd_to_bio_flags(le32_to_cpu(msg->rw));
 163         bio->bi_iter.bi_sector = le64_to_cpu(msg->sector);
 164         bio->bi_iter.bi_size = le32_to_cpu(msg->bi_size);
 165         prio = srv_sess->ver < RNBD_PROTO_VER_MAJOR ||
 166                usrlen < sizeof(*msg) ? 0 : le16_to_cpu(msg->prio);
 167         bio_set_prio(bio, prio);
 168         bio_set_dev(bio, sess_dev->rnbd_dev->bdev);
 169
 170         submit_bio(bio);
 171
 172         return 0;
 173
 174 sess_dev_put:
 175         rnbd_put_sess_dev(sess_dev);
 176 err:
 177         kfree(priv);
 178         return err;
 179 }
 180
 181 static void destroy_device(struct rnbd_srv_dev *dev)
 182 {
 183         WARN_ONCE(!list_empty(&dev->sess_dev_list),
 184                   "Device %s is being destroyed but still in use!\n",
 185                   dev->id);
 186
 187         spin_lock(&dev_lock);
 188         list_del(&dev->list);
 189         spin_unlock(&dev_lock);
 190
 191         mutex_destroy(&dev->lock);
 192         if (dev->dev_kobj.state_in_sysfs)
 193                 /*
 194                  * Destroy kobj only if it was really created.
 195                  */
 196                 rnbd_srv_destroy_dev_sysfs(dev);
 197         else
 198                 kfree(dev);
 199 }
 200
 201 static void destroy_device_cb(struct kref *kref)
 202 {
 203         struct rnbd_srv_dev *dev;
 204
 205         dev = container_of(kref, struct rnbd_srv_dev, kref);
 206
 207         destroy_device(dev);
 208 }
 209
 210 static void rnbd_put_srv_dev(struct rnbd_srv_dev *dev)
 211 {
 212         kref_put(&dev->kref, destroy_device_cb);
 213 }
 214
 215 void rnbd_destroy_sess_dev(struct rnbd_srv_sess_dev *sess_dev, bool keep_id)
 216 {
 217         DECLARE_COMPLETION_ONSTACK(dc);
 218
 219         if (keep_id)
 220                 /* free the resources for the id but don't  */
 221                 /* allow to re-use the id itself because it */
 222                 /* is still used by the client              */
 223                 xa_cmpxchg(&sess_dev->sess->index_idr, sess_dev->device_id,
 224                            sess_dev, NULL, 0);
 225         else
 226                 xa_erase(&sess_dev->sess->index_idr, sess_dev->device_id);
 227         synchronize_rcu();
 228
 229         sess_dev->destroy_comp = &dc;
 230         rnbd_put_sess_dev(sess_dev);
 231         wait_for_completion(&dc); /* wait for inflights to drop to zero */
 232
 233         rnbd_dev_close(sess_dev->rnbd_dev);
 234         list_del(&sess_dev->sess_list);
 235         mutex_lock(&sess_dev->dev->lock);
 236         list_del(&sess_dev->dev_list);
 237         if (sess_dev->open_flags & FMODE_WRITE)
 238                 sess_dev->dev->open_write_cnt--;
 239         mutex_unlock(&sess_dev->dev->lock);
 240
 241         rnbd_put_srv_dev(sess_dev->dev);
 242
 243         rnbd_srv_info(sess_dev, "Device closed\n");
 244         kfree(sess_dev);
 245 }
 246
 247 static void destroy_sess(struct rnbd_srv_session *srv_sess)
 248 {
 249         struct rnbd_srv_sess_dev *sess_dev, *tmp;
 250
 251         if (list_empty(&srv_sess->sess_dev_list))
 252                 goto out;
 253
 254         mutex_lock(&srv_sess->lock);
 255         list_for_each_entry_safe(sess_dev, tmp, &srv_sess->sess_dev_list,
 256                                  sess_list)
 257                 rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 258         mutex_unlock(&srv_sess->lock);
 259
 260 out:
 261         xa_destroy(&srv_sess->index_idr);
 262         bioset_exit(&srv_sess->sess_bio_set);
 263
 264         pr_info("RTRS Session %s disconnected\n", srv_sess->sessname);
 265
 266         mutex_lock(&sess_lock);
 267         list_del(&srv_sess->list);
 268         mutex_unlock(&sess_lock);
 269
 270         mutex_destroy(&srv_sess->lock);
 271         kfree(srv_sess);
 272 }
 273
 274 static int create_sess(struct rtrs_srv *rtrs)
 275 {
 276         struct rnbd_srv_session *srv_sess;
 277         char sessname[NAME_MAX];
 278         int err;
 279
 280         err = rtrs_srv_get_sess_name(rtrs, sessname, sizeof(sessname));
 281         if (err) {
 282                 pr_err("rtrs_srv_get_sess_name(%s): %d\n", sessname, err);
 283
 284                 return err;
 285         }
 286         srv_sess = kzalloc(sizeof(*srv_sess), GFP_KERNEL);
 287         if (!srv_sess)
 288                 return -ENOMEM;
 289
 290         srv_sess->queue_depth = rtrs_srv_get_queue_depth(rtrs);
 291         err = bioset_init(&srv_sess->sess_bio_set, srv_sess->queue_depth,
 292                           offsetof(struct rnbd_dev_blk_io, bio),
 293                           BIOSET_NEED_BVECS);
 294         if (err) {
 295                 pr_err("Allocating srv_session for session %s failed\n",
 296                        sessname);
 297                 kfree(srv_sess);
 298                 return err;
 299         }
 300
 301         xa_init_flags(&srv_sess->index_idr, XA_FLAGS_ALLOC);
 302         INIT_LIST_HEAD(&srv_sess->sess_dev_list);
 303         mutex_init(&srv_sess->lock);
 304         mutex_lock(&sess_lock);
 305         list_add(&srv_sess->list, &sess_list);
 306         mutex_unlock(&sess_lock);
 307
 308         srv_sess->rtrs = rtrs;
 309         strlcpy(srv_sess->sessname, sessname, sizeof(srv_sess->sessname));
 310
 311         rtrs_srv_set_sess_priv(rtrs, srv_sess);
 312
 313         return 0;
 314 }
 315
 316 static int rnbd_srv_link_ev(struct rtrs_srv *rtrs,
 317                              enum rtrs_srv_link_ev ev, void *priv)
 318 {
 319         struct rnbd_srv_session *srv_sess = priv;
 320
 321         switch (ev) {
 322         case RTRS_SRV_LINK_EV_CONNECTED:
 323                 return create_sess(rtrs);
 324
 325         case RTRS_SRV_LINK_EV_DISCONNECTED:
 326                 if (WARN_ON_ONCE(!srv_sess))
 327                         return -EINVAL;
 328
 329                 destroy_sess(srv_sess);
 330                 return 0;
 331
 332         default:
 333                 pr_warn("Received unknown RTRS session event %d from session %s\n",
 334                         ev, srv_sess->sessname);
 335                 return -EINVAL;
 336         }
 337 }
 338
 339 void rnbd_srv_sess_dev_force_close(struct rnbd_srv_sess_dev *sess_dev)
 340 {
 341         struct rnbd_srv_session *sess = sess_dev->sess;
 342
 343         sess_dev->keep_id = true;
 344         mutex_lock(&sess->lock);
 345         rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 346         mutex_unlock(&sess->lock);
 347 }
 348
 349 static int process_msg_close(struct rtrs_srv *rtrs,
 350                              struct rnbd_srv_session *srv_sess,
 351                              void *data, size_t datalen, const void *usr,
 352                              size_t usrlen)
 353 {
 354         const struct rnbd_msg_close *close_msg = usr;
 355         struct rnbd_srv_sess_dev *sess_dev;
 356
 357         sess_dev = rnbd_get_sess_dev(le32_to_cpu(close_msg->device_id),
 358                                       srv_sess);
 359         if (IS_ERR(sess_dev))
 360                 return 0;
 361
 362         rnbd_put_sess_dev(sess_dev);
 363         mutex_lock(&srv_sess->lock);
 364         rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 365         mutex_unlock(&srv_sess->lock);
 366         return 0;
 367 }
 368
 369 static int process_msg_open(struct rtrs_srv *rtrs,
 370                             struct rnbd_srv_session *srv_sess,
 371                             const void *msg, size_t len,
 372                             void *data, size_t datalen);
 373
 374 static int process_msg_sess_info(struct rtrs_srv *rtrs,
 375                                  struct rnbd_srv_session *srv_sess,
 376                                  const void *msg, size_t len,
 377                                  void *data, size_t datalen);
 378
 379 static int rnbd_srv_rdma_ev(struct rtrs_srv *rtrs, void *priv,
 380                              struct rtrs_srv_op *id, int dir,
 381                              void *data, size_t datalen, const void *usr,
 382                              size_t usrlen)
 383 {
 384         struct rnbd_srv_session *srv_sess = priv;
 385         const struct rnbd_msg_hdr *hdr = usr;
 386         int ret = 0;
 387         u16 type;
 388
 389         if (WARN_ON_ONCE(!srv_sess))
 390                 return -ENODEV;
 391
 392         type = le16_to_cpu(hdr->type);
 393
 394         switch (type) {
 395         case RNBD_MSG_IO:
 396                 return process_rdma(rtrs, srv_sess, id, data, datalen, usr,
 397                                     usrlen);
 398         case RNBD_MSG_CLOSE:
 399                 ret = process_msg_close(rtrs, srv_sess, data, datalen,
 400                                         usr, usrlen);
 401                 break;
 402         case RNBD_MSG_OPEN:
 403                 ret = process_msg_open(rtrs, srv_sess, usr, usrlen,
 404                                        data, datalen);
 405                 break;
 406         case RNBD_MSG_SESS_INFO:
 407                 ret = process_msg_sess_info(rtrs, srv_sess, usr, usrlen,
 408                                             data, datalen);
 409                 break;
 410         default:
 411                 pr_warn("Received unexpected message type %d with dir %d from session %s\n",
 412                         type, dir, srv_sess->sessname);
 413                 return -EINVAL;
 414         }
 415
 416         rtrs_srv_resp_rdma(id, ret);
 417         return 0;
 418 }
 419
 420 static struct rnbd_srv_sess_dev
 421 *rnbd_sess_dev_alloc(struct rnbd_srv_session *srv_sess)
 422 {
 423         struct rnbd_srv_sess_dev *sess_dev;
 424         int error;
 425
 426         sess_dev = kzalloc(sizeof(*sess_dev), GFP_KERNEL);
 427         if (!sess_dev)
 428                 return ERR_PTR(-ENOMEM);
 429
 430         error = xa_alloc(&srv_sess->index_idr, &sess_dev->device_id, sess_dev,
 431                          xa_limit_32b, GFP_NOWAIT);
 432         if (error < 0) {
 433                 pr_warn("Allocating idr failed, err: %d\n", error);
 434                 kfree(sess_dev);
 435                 return ERR_PTR(error);
 436         }
 437
 438         return sess_dev;
 439 }
 440
 441 static struct rnbd_srv_dev *rnbd_srv_init_srv_dev(const char *id)
 442 {
 443         struct rnbd_srv_dev *dev;
 444
 445         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 446         if (!dev)
 447                 return ERR_PTR(-ENOMEM);
 448
 449         strlcpy(dev->id, id, sizeof(dev->id));
 450         kref_init(&dev->kref);
 451         INIT_LIST_HEAD(&dev->sess_dev_list);
 452         mutex_init(&dev->lock);
 453
 454         return dev;
 455 }
 456
 457 static struct rnbd_srv_dev *
 458 rnbd_srv_find_or_add_srv_dev(struct rnbd_srv_dev *new_dev)
 459 {
 460         struct rnbd_srv_dev *dev;
 461
 462         spin_lock(&dev_lock);
 463         list_for_each_entry(dev, &dev_list, list) {
 464                 if (!strncmp(dev->id, new_dev->id, sizeof(dev->id))) {
 465                         if (!kref_get_unless_zero(&dev->kref))
 466                                 /*
 467                                  * We lost the race, device is almost dead.
 468                                  *  Continue traversing to find a valid one.
 469                                  */
 470                                 continue;
 471                         spin_unlock(&dev_lock);
 472                         return dev;
 473                 }
 474         }
 475         list_add(&new_dev->list, &dev_list);
 476         spin_unlock(&dev_lock);
 477
 478         return new_dev;
 479 }
 480
 481 static int rnbd_srv_check_update_open_perm(struct rnbd_srv_dev *srv_dev,
 482                                             struct rnbd_srv_session *srv_sess,
 483                                             enum rnbd_access_mode access_mode)
 484 {
 485         int ret = -EPERM;
 486
 487         mutex_lock(&srv_dev->lock);
 488
 489         switch (access_mode) {
 490         case RNBD_ACCESS_RO:
 491                 ret = 0;
 492                 break;
 493         case RNBD_ACCESS_RW:
 494                 if (srv_dev->open_write_cnt == 0)  {
 495                         srv_dev->open_write_cnt++;
 496                         ret = 0;
 497                 } else {
 498                         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",
 499                                srv_dev->id, srv_sess->sessname,
 500                                srv_dev->open_write_cnt,
 501                                rnbd_access_mode_str(access_mode));
 502                 }
 503                 break;
 504         case RNBD_ACCESS_MIGRATION:
 505                 if (srv_dev->open_write_cnt < 2) {
 506                         srv_dev->open_write_cnt++;
 507                         ret = 0;
 508                 } else {
 509                         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",
 510                                srv_dev->id, srv_sess->sessname,
 511                                srv_dev->open_write_cnt,
 512                                rnbd_access_mode_str(access_mode));
 513                 }
 514                 break;
 515         default:
 516                 pr_err("Received mapping request for device '%s' on session %s with invalid access mode: %d\n",
 517                        srv_dev->id, srv_sess->sessname, access_mode);
 518                 ret = -EINVAL;
 519         }
 520
 521         mutex_unlock(&srv_dev->lock);
 522
 523         return ret;
 524 }
 525
 526 static struct rnbd_srv_dev *
 527 rnbd_srv_get_or_create_srv_dev(struct rnbd_dev *rnbd_dev,
 528                                 struct rnbd_srv_session *srv_sess,
 529                                 enum rnbd_access_mode access_mode)
 530 {
 531         int ret;
 532         struct rnbd_srv_dev *new_dev, *dev;
 533
 534         new_dev = rnbd_srv_init_srv_dev(rnbd_dev->name);
 535         if (IS_ERR(new_dev))
 536                 return new_dev;
 537
 538         dev = rnbd_srv_find_or_add_srv_dev(new_dev);
 539         if (dev != new_dev)
 540                 kfree(new_dev);
 541
 542         ret = rnbd_srv_check_update_open_perm(dev, srv_sess, access_mode);
 543         if (ret) {
 544                 rnbd_put_srv_dev(dev);
 545                 return ERR_PTR(ret);
 546         }
 547
 548         return dev;
 549 }
 550
 551 static void rnbd_srv_fill_msg_open_rsp(struct rnbd_msg_open_rsp *rsp,
 552                                         struct rnbd_srv_sess_dev *sess_dev)
 553 {
 554         struct rnbd_dev *rnbd_dev = sess_dev->rnbd_dev;
 555         struct request_queue *q = bdev_get_queue(rnbd_dev->bdev);
 556
 557         rsp->hdr.type = cpu_to_le16(RNBD_MSG_OPEN_RSP);
 558         rsp->device_id =
 559                 cpu_to_le32(sess_dev->device_id);
 560         rsp->nsectors =
 561                 cpu_to_le64(get_capacity(rnbd_dev->bdev->bd_disk));
 562         rsp->logical_block_size =
 563                 cpu_to_le16(bdev_logical_block_size(rnbd_dev->bdev));
 564         rsp->physical_block_size =
 565                 cpu_to_le16(bdev_physical_block_size(rnbd_dev->bdev));
 566         rsp->max_segments =
 567                 cpu_to_le16(rnbd_dev_get_max_segs(rnbd_dev));
 568         rsp->max_hw_sectors =
 569                 cpu_to_le32(rnbd_dev_get_max_hw_sects(rnbd_dev));
 570         rsp->max_write_same_sectors =
 571                 cpu_to_le32(bdev_write_same(rnbd_dev->bdev));
 572         rsp->max_discard_sectors =
 573                 cpu_to_le32(rnbd_dev_get_max_discard_sects(rnbd_dev));
 574         rsp->discard_granularity =
 575                 cpu_to_le32(rnbd_dev_get_discard_granularity(rnbd_dev));
 576         rsp->discard_alignment =
 577                 cpu_to_le32(rnbd_dev_get_discard_alignment(rnbd_dev));
 578         rsp->secure_discard =
 579                 cpu_to_le16(rnbd_dev_get_secure_discard(rnbd_dev));
 580         rsp->rotational = !blk_queue_nonrot(q);
 581         rsp->cache_policy = 0;
 582         if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 583                 rsp->cache_policy |= RNBD_WRITEBACK;
 584         if (blk_queue_fua(q))
 585                 rsp->cache_policy |= RNBD_FUA;
 586 }
 587
 588 static struct rnbd_srv_sess_dev *
 589 rnbd_srv_create_set_sess_dev(struct rnbd_srv_session *srv_sess,
 590                               const struct rnbd_msg_open *open_msg,
 591                               struct rnbd_dev *rnbd_dev, fmode_t open_flags,
 592                               struct rnbd_srv_dev *srv_dev)
 593 {
 594         struct rnbd_srv_sess_dev *sdev = rnbd_sess_dev_alloc(srv_sess);
 595
 596         if (IS_ERR(sdev))
 597                 return sdev;
 598
 599         kref_init(&sdev->kref);
 600
 601         strlcpy(sdev->pathname, open_msg->dev_name, sizeof(sdev->pathname));
 602
 603         sdev->rnbd_dev          = rnbd_dev;
 604         sdev->sess              = srv_sess;
 605         sdev->dev               = srv_dev;
 606         sdev->open_flags        = open_flags;
 607         sdev->access_mode       = open_msg->access_mode;
 608
 609         return sdev;
 610 }
 611
 612 static char *rnbd_srv_get_full_path(struct rnbd_srv_session *srv_sess,
 613                                      const char *dev_name)
 614 {
 615         char *full_path;
 616         char *a, *b;
 617
 618         full_path = kmalloc(PATH_MAX, GFP_KERNEL);
 619         if (!full_path)
 620                 return ERR_PTR(-ENOMEM);
 621
 622         /*
 623          * Replace %SESSNAME% with a real session name in order to
 624          * create device namespace.
 625          */
 626         a = strnstr(dev_search_path, "%SESSNAME%", sizeof(dev_search_path));
 627         if (a) {
 628                 int len = a - dev_search_path;
 629
 630                 len = snprintf(full_path, PATH_MAX, "%.*s/%s/%s", len,
 631                                dev_search_path, srv_sess->sessname, dev_name);
 632                 if (len >= PATH_MAX) {
 633                         pr_err("Too long path: %s, %s, %s\n",
 634                                dev_search_path, srv_sess->sessname, dev_name);
 635                         kfree(full_path);
 636                         return ERR_PTR(-EINVAL);
 637                 }
 638         } else {
 639                 snprintf(full_path, PATH_MAX, "%s/%s",
 640                          dev_search_path, dev_name);
 641         }
 642
 643         /* eliminitate duplicated slashes */
 644         a = strchr(full_path, '/');
 645         b = a;
 646         while (*b != '\0') {
 647                 if (*b == '/' && *a == '/') {
 648                         b++;
 649                 } else {
 650                         a++;
 651                         *a = *b;
 652                         b++;
 653                 }
 654         }
 655         a++;
 656         *a = '\0';
 657
 658         return full_path;
 659 }
 660
 661 static int process_msg_sess_info(struct rtrs_srv *rtrs,
 662                                  struct rnbd_srv_session *srv_sess,
 663                                  const void *msg, size_t len,
 664                                  void *data, size_t datalen)
 665 {
 666         const struct rnbd_msg_sess_info *sess_info_msg = msg;
 667         struct rnbd_msg_sess_info_rsp *rsp = data;
 668
 669         srv_sess->ver = min_t(u8, sess_info_msg->ver, RNBD_PROTO_VER_MAJOR);
 670         pr_debug("Session %s using protocol version %d (client version: %d, server version: %d)\n",
 671                  srv_sess->sessname, srv_sess->ver,
 672                  sess_info_msg->ver, RNBD_PROTO_VER_MAJOR);
 673
 674         rsp->hdr.type = cpu_to_le16(RNBD_MSG_SESS_INFO_RSP);
 675         rsp->ver = srv_sess->ver;
 676
 677         return 0;
 678 }
 679
 680 /**
 681  * find_srv_sess_dev() - a dev is already opened by this name
 682  * @srv_sess:   the session to search.
 683  * @dev_name:   string containing the name of the device.
 684  *
 685  * Return struct rnbd_srv_sess_dev if srv_sess already opened the dev_name
 686  * NULL if the session didn't open the device yet.
 687  */
 688 static struct rnbd_srv_sess_dev *
 689 find_srv_sess_dev(struct rnbd_srv_session *srv_sess, const char *dev_name)
 690 {
 691         struct rnbd_srv_sess_dev *sess_dev;
 692
 693         if (list_empty(&srv_sess->sess_dev_list))
 694                 return NULL;
 695
 696         list_for_each_entry(sess_dev, &srv_sess->sess_dev_list, sess_list)
 697                 if (!strcmp(sess_dev->pathname, dev_name))
 698                         return sess_dev;
 699
 700         return NULL;
 701 }
 702
 703 static int process_msg_open(struct rtrs_srv *rtrs,
 704                             struct rnbd_srv_session *srv_sess,
 705                             const void *msg, size_t len,
 706                             void *data, size_t datalen)
 707 {
 708         int ret;
 709         struct rnbd_srv_dev *srv_dev;
 710         struct rnbd_srv_sess_dev *srv_sess_dev;
 711         const struct rnbd_msg_open *open_msg = msg;
 712         fmode_t open_flags;
 713         char *full_path;
 714         struct rnbd_dev *rnbd_dev;
 715         struct rnbd_msg_open_rsp *rsp = data;
 716
 717         pr_debug("Open message received: session='%s' path='%s' access_mode=%d\n",
 718                  srv_sess->sessname, open_msg->dev_name,
 719                  open_msg->access_mode);
 720         open_flags = FMODE_READ;
 721         if (open_msg->access_mode != RNBD_ACCESS_RO)
 722                 open_flags |= FMODE_WRITE;
 723
 724         mutex_lock(&srv_sess->lock);
 725
 726         srv_sess_dev = find_srv_sess_dev(srv_sess, open_msg->dev_name);
 727         if (srv_sess_dev)
 728                 goto fill_response;
 729
 730         if ((strlen(dev_search_path) + strlen(open_msg->dev_name))
 731             >= PATH_MAX) {
 732                 pr_err("Opening device for session %s failed, device path too long. '%s/%s' is longer than PATH_MAX (%d)\n",
 733                        srv_sess->sessname, dev_search_path, open_msg->dev_name,
 734                        PATH_MAX);
 735                 ret = -EINVAL;
 736                 goto reject;
 737         }
 738         if (strstr(open_msg->dev_name, "..")) {
 739                 pr_err("Opening device for session %s failed, device path %s contains relative path ..\n",
 740                        srv_sess->sessname, open_msg->dev_name);
 741                 ret = -EINVAL;
 742                 goto reject;
 743         }
 744         full_path = rnbd_srv_get_full_path(srv_sess, open_msg->dev_name);
 745         if (IS_ERR(full_path)) {
 746                 ret = PTR_ERR(full_path);
 747                 pr_err("Opening device '%s' for client %s failed, failed to get device full path, err: %d\n",
 748                        open_msg->dev_name, srv_sess->sessname, ret);
 749                 goto reject;
 750         }
 751
 752         rnbd_dev = rnbd_dev_open(full_path, open_flags,
 753                                  &srv_sess->sess_bio_set);
 754         if (IS_ERR(rnbd_dev)) {
 755                 pr_err("Opening device '%s' on session %s failed, failed to open the block device, err: %ld\n",
 756                        full_path, srv_sess->sessname, PTR_ERR(rnbd_dev));
 757                 ret = PTR_ERR(rnbd_dev);
 758                 goto free_path;
 759         }
 760
 761         srv_dev = rnbd_srv_get_or_create_srv_dev(rnbd_dev, srv_sess,
 762                                                   open_msg->access_mode);
 763         if (IS_ERR(srv_dev)) {
 764                 pr_err("Opening device '%s' on session %s failed, creating srv_dev failed, err: %ld\n",
 765                        full_path, srv_sess->sessname, PTR_ERR(srv_dev));
 766                 ret = PTR_ERR(srv_dev);
 767                 goto rnbd_dev_close;
 768         }
 769
 770         srv_sess_dev = rnbd_srv_create_set_sess_dev(srv_sess, open_msg,
 771                                                      rnbd_dev, open_flags,
 772                                                      srv_dev);
 773         if (IS_ERR(srv_sess_dev)) {
 774                 pr_err("Opening device '%s' on session %s failed, creating sess_dev failed, err: %ld\n",
 775                        full_path, srv_sess->sessname, PTR_ERR(srv_sess_dev));
 776                 ret = PTR_ERR(srv_sess_dev);
 777                 goto srv_dev_put;
 778         }
 779
 780         /* Create the srv_dev sysfs files if they haven't been created yet. The
 781          * reason to delay the creation is not to create the sysfs files before
 782          * we are sure the device can be opened.
 783          */
 784         mutex_lock(&srv_dev->lock);
 785         if (!srv_dev->dev_kobj.state_in_sysfs) {
 786                 ret = rnbd_srv_create_dev_sysfs(srv_dev, rnbd_dev->bdev,
 787                                                  rnbd_dev->name);
 788                 if (ret) {
 789                         mutex_unlock(&srv_dev->lock);
 790                         rnbd_srv_err(srv_sess_dev,
 791                                       "Opening device failed, failed to create device sysfs files, err: %d\n",
 792                                       ret);
 793                         goto free_srv_sess_dev;
 794                 }
 795         }
 796
 797         ret = rnbd_srv_create_dev_session_sysfs(srv_sess_dev);
 798         if (ret) {
 799                 mutex_unlock(&srv_dev->lock);
 800                 rnbd_srv_err(srv_sess_dev,
 801                               "Opening device failed, failed to create dev client sysfs files, err: %d\n",
 802                               ret);
 803                 goto free_srv_sess_dev;
 804         }
 805
 806         list_add(&srv_sess_dev->dev_list, &srv_dev->sess_dev_list);
 807         mutex_unlock(&srv_dev->lock);
 808
 809         list_add(&srv_sess_dev->sess_list, &srv_sess->sess_dev_list);
 810
 811         rnbd_srv_info(srv_sess_dev, "Opened device '%s'\n", srv_dev->id);
 812
 813         kfree(full_path);
 814
 815 fill_response:
 816         rnbd_srv_fill_msg_open_rsp(rsp, srv_sess_dev);
 817         mutex_unlock(&srv_sess->lock);
 818         return 0;
 819
 820 free_srv_sess_dev:
 821         xa_erase(&srv_sess->index_idr, srv_sess_dev->device_id);
 822         synchronize_rcu();
 823         kfree(srv_sess_dev);
 824 srv_dev_put:
 825         if (open_msg->access_mode != RNBD_ACCESS_RO) {
 826                 mutex_lock(&srv_dev->lock);
 827                 srv_dev->open_write_cnt--;
 828                 mutex_unlock(&srv_dev->lock);
 829         }
 830         rnbd_put_srv_dev(srv_dev);
 831 rnbd_dev_close:
 832         rnbd_dev_close(rnbd_dev);
 833 free_path:
 834         kfree(full_path);
 835 reject:
 836         mutex_unlock(&srv_sess->lock);
 837         return ret;
 838 }
 839
 840 static struct rtrs_srv_ctx *rtrs_ctx;
 841
 842 static struct rtrs_srv_ops rtrs_ops;
 843 static int __init rnbd_srv_init_module(void)
 844 {
 845         int err;
 846
 847         BUILD_BUG_ON(sizeof(struct rnbd_msg_hdr) != 4);
 848         BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info) != 36);
 849         BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info_rsp) != 36);
 850         BUILD_BUG_ON(sizeof(struct rnbd_msg_open) != 264);
 851         BUILD_BUG_ON(sizeof(struct rnbd_msg_close) != 8);
 852         BUILD_BUG_ON(sizeof(struct rnbd_msg_open_rsp) != 56);
 853         rtrs_ops = (struct rtrs_srv_ops) {
 854                 .rdma_ev = rnbd_srv_rdma_ev,
 855                 .link_ev = rnbd_srv_link_ev,
 856         };
 857         rtrs_ctx = rtrs_srv_open(&rtrs_ops, port_nr);
 858         if (IS_ERR(rtrs_ctx)) {
 859                 err = PTR_ERR(rtrs_ctx);
 860                 pr_err("rtrs_srv_open(), err: %d\n", err);
 861                 return err;
 862         }
 863
 864         err = rnbd_srv_create_sysfs_files();
 865         if (err) {
 866                 pr_err("rnbd_srv_create_sysfs_files(), err: %d\n", err);
 867                 rtrs_srv_close(rtrs_ctx);
 868                 return err;
 869         }
 870
 871         return 0;
 872 }
 873
 874 static void __exit rnbd_srv_cleanup_module(void)
 875 {
 876         rtrs_srv_close(rtrs_ctx);
 877         WARN_ON(!list_empty(&sess_list));
 878         rnbd_srv_destroy_sysfs_files();
 879 }
 880
 881 module_init(rnbd_srv_init_module);
 882 module_exit(rnbd_srv_cleanup_module);