Quelle  rnbd-srv.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0-or-later
/*
 * RDMA Network Block Driver
 *
 * Copyright (c) 2014 - 2018 ProfitBricks GmbH. All rights reserved.
 * Copyright (c) 2018 - 2019 1&1 IONOS Cloud GmbH. All rights reserved.
 * Copyright (c) 2019 - 2020 1&1 IONOS SE. All rights reserved.
 */
#undef pr_fmt
#define pr_fmt(fmt) KBUILD_MODNAME " L" __stringify(__LINE__) ": " fmt

#include <linux/module.h>
#include <linux/blkdev.h>

#include "rnbd-srv.h"
#include "rnbd-srv-trace.h"

MODULE_DESCRIPTION("RDMA Network Block Device Server");
MODULE_LICENSE("GPL");

static u16 port_nr = RTRS_PORT;

module_param_named(port_nr, port_nr, ushort, 0444);
MODULE_PARM_DESC(port_nr,
   "The port number the server is listening on (default: "
   __stringify(RTRS_PORT)")");

#define DEFAULT_DEV_SEARCH_PATH "/"

static char dev_search_path[PATH_MAX] = DEFAULT_DEV_SEARCH_PATH;

static int dev_search_path_set(const char *val, const struct kernel_param *kp)
{
 const char *p = strrchr(val, '\n') ? : val + strlen(val);

 if (strlen(val) >= sizeof(dev_search_path))
  return -EINVAL;

 snprintf(dev_search_path, sizeof(dev_search_path), "%.*s",
   (int)(p - val), val);

 pr_info("dev_search_path changed to '%s'\n", dev_search_path);

 return 0;
}

static struct kparam_string dev_search_path_kparam_str = {
 .maxlen = sizeof(dev_search_path),
 .string = dev_search_path
};

static const struct kernel_param_ops dev_search_path_ops = {
 .set = dev_search_path_set,
 .get = param_get_string,
};

module_param_cb(dev_search_path, &dev_search_path_ops,
  &dev_search_path_kparam_str, 0444);
MODULE_PARM_DESC(dev_search_path,
   "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: "
   DEFAULT_DEV_SEARCH_PATH ")");

static DEFINE_MUTEX(sess_lock);
static DEFINE_SPINLOCK(dev_lock);

static LIST_HEAD(sess_list);
static LIST_HEAD(dev_list);

struct rnbd_io_private {
 struct rtrs_srv_op  *id;
 struct rnbd_srv_sess_dev *sess_dev;
};

static void rnbd_sess_dev_release(struct kref *kref)
{
 struct rnbd_srv_sess_dev *sess_dev;

 sess_dev = container_of(kref, struct rnbd_srv_sess_dev, kref);
 complete(sess_dev->destroy_comp);
}

static inline void rnbd_put_sess_dev(struct rnbd_srv_sess_dev *sess_dev)
{
 kref_put(&sess_dev->kref, rnbd_sess_dev_release);
}

static struct rnbd_srv_sess_dev *
rnbd_get_sess_dev(int dev_id, struct rnbd_srv_session *srv_sess)
{
 struct rnbd_srv_sess_dev *sess_dev;
 int ret = 0;

 rcu_read_lock();
 sess_dev = xa_load(&srv_sess->index_idr, dev_id);
 if (sess_dev)
  ret = kref_get_unless_zero(&sess_dev->kref);
 rcu_read_unlock();

 if (!ret)
  return ERR_PTR(-ENXIO);

 return sess_dev;
}

static void rnbd_dev_bi_end_io(struct bio *bio)
{
 struct rnbd_io_private *rnbd_priv = bio->bi_private;
 struct rnbd_srv_sess_dev *sess_dev = rnbd_priv->sess_dev;

 rnbd_put_sess_dev(sess_dev);
 rtrs_srv_resp_rdma(rnbd_priv->id, blk_status_to_errno(bio->bi_status));

 kfree(rnbd_priv);
 bio_put(bio);
}

static int process_rdma(struct rnbd_srv_session *srv_sess,
   struct rtrs_srv_op *id, void *data, u32 datalen,
   const void *usr, size_t usrlen)
{
 const struct rnbd_msg_io *msg = usr;
 struct rnbd_io_private *priv;
 struct rnbd_srv_sess_dev *sess_dev;
 u32 dev_id;
 int err;
 struct bio *bio;
 short prio;

 trace_process_rdma(srv_sess, msg, id, datalen, usrlen);

 priv = kmalloc(sizeof(*priv), GFP_KERNEL);
 if (!priv)
  return -ENOMEM;

 dev_id = le32_to_cpu(msg->device_id);

 sess_dev = rnbd_get_sess_dev(dev_id, srv_sess);
 if (IS_ERR(sess_dev)) {
  pr_err_ratelimited("Got I/O request on session %s for unknown device id %d: %pe\n",
       srv_sess->sessname, dev_id, sess_dev);
  err = -ENOTCONN;
  goto err;
 }

 priv->sess_dev = sess_dev;
 priv->id = id;

 bio = bio_alloc(file_bdev(sess_dev->bdev_file), 1,
   rnbd_to_bio_flags(le32_to_cpu(msg->rw)), GFP_KERNEL);
 bio_add_virt_nofail(bio, data, datalen);

 bio->bi_opf = rnbd_to_bio_flags(le32_to_cpu(msg->rw));
 if (bio_has_data(bio) &&
     bio->bi_iter.bi_size != le32_to_cpu(msg->bi_size)) {
  rnbd_srv_err_rl(sess_dev, "Datalen mismatch: bio bi_size (%u), bi_size (%u)\n",
    bio->bi_iter.bi_size, msg->bi_size);
  err = -EINVAL;
  goto bio_put;
 }
 bio->bi_end_io = rnbd_dev_bi_end_io;
 bio->bi_private = priv;
 bio->bi_iter.bi_sector = le64_to_cpu(msg->sector);
 prio = srv_sess->ver < RNBD_PROTO_VER_MAJOR ||
        usrlen < sizeof(*msg) ? 0 : le16_to_cpu(msg->prio);
 bio->bi_ioprio = prio;

 submit_bio(bio);

 return 0;

bio_put:
 bio_put(bio);
 rnbd_put_sess_dev(sess_dev);
err:
 kfree(priv);
 return err;
}

static void destroy_device(struct kref *kref)
{
 struct rnbd_srv_dev *dev = container_of(kref, struct rnbd_srv_dev, kref);

 WARN_ONCE(!list_empty(&dev->sess_dev_list),
    "Device %s is being destroyed but still in use!\n",
    dev->name);

 spin_lock(&dev_lock);
 list_del(&dev->list);
 spin_unlock(&dev_lock);

 mutex_destroy(&dev->lock);
 if (dev->dev_kobj.state_in_sysfs)
  /*
 * Destroy kobj only if it was really created.
 */
  rnbd_srv_destroy_dev_sysfs(dev);
 else
  kfree(dev);
}

static void rnbd_put_srv_dev(struct rnbd_srv_dev *dev)
{
 kref_put(&dev->kref, destroy_device);
}

void rnbd_destroy_sess_dev(struct rnbd_srv_sess_dev *sess_dev, bool keep_id)
{
 DECLARE_COMPLETION_ONSTACK(dc);

 if (keep_id)
  /* free the resources for the id but don't  */
  /* allow to re-use the id itself because it */
  /* is still used by the client              */
  xa_cmpxchg(&sess_dev->sess->index_idr, sess_dev->device_id,
      sess_dev, NULL, 0);
 else
  xa_erase(&sess_dev->sess->index_idr, sess_dev->device_id);
 synchronize_rcu();

 sess_dev->destroy_comp = &dc;
 rnbd_put_sess_dev(sess_dev);
 wait_for_completion(&dc); /* wait for inflights to drop to zero */

 fput(sess_dev->bdev_file);
 mutex_lock(&sess_dev->dev->lock);
 list_del(&sess_dev->dev_list);
 if (!sess_dev->readonly)
  sess_dev->dev->open_write_cnt--;
 mutex_unlock(&sess_dev->dev->lock);

 rnbd_put_srv_dev(sess_dev->dev);

 rnbd_srv_info(sess_dev, "Device closed\n");
 kfree(sess_dev);
}

static void destroy_sess(struct rnbd_srv_session *srv_sess)
{
 struct rnbd_srv_sess_dev *sess_dev;
 unsigned long index;

 if (xa_empty(&srv_sess->index_idr))
  goto out;

 trace_destroy_sess(srv_sess);

 mutex_lock(&srv_sess->lock);
 xa_for_each(&srv_sess->index_idr, index, sess_dev)
  rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 mutex_unlock(&srv_sess->lock);

out:
 xa_destroy(&srv_sess->index_idr);

 pr_info("RTRS Session %s disconnected\n", srv_sess->sessname);

 mutex_lock(&sess_lock);
 list_del(&srv_sess->list);
 mutex_unlock(&sess_lock);

 mutex_destroy(&srv_sess->lock);
 kfree(srv_sess);
}

static int create_sess(struct rtrs_srv_sess *rtrs)
{
 struct rnbd_srv_session *srv_sess;
 char pathname[NAME_MAX];
 int err;

 err = rtrs_srv_get_path_name(rtrs, pathname, sizeof(pathname));
 if (err) {
  pr_err("rtrs_srv_get_path_name(%s): %d\n", pathname, err);

  return err;
 }
 srv_sess = kzalloc(sizeof(*srv_sess), GFP_KERNEL);
 if (!srv_sess)
  return -ENOMEM;

 srv_sess->queue_depth = rtrs_srv_get_queue_depth(rtrs);
 xa_init_flags(&srv_sess->index_idr, XA_FLAGS_ALLOC);
 mutex_init(&srv_sess->lock);
 mutex_lock(&sess_lock);
 list_add(&srv_sess->list, &sess_list);
 mutex_unlock(&sess_lock);

 srv_sess->rtrs = rtrs;
 strscpy(srv_sess->sessname, pathname, sizeof(srv_sess->sessname));

 rtrs_srv_set_sess_priv(rtrs, srv_sess);

 trace_create_sess(srv_sess);

 return 0;
}

static int rnbd_srv_link_ev(struct rtrs_srv_sess *rtrs,
        enum rtrs_srv_link_ev ev, void *priv)
{
 struct rnbd_srv_session *srv_sess = priv;

 switch (ev) {
 case RTRS_SRV_LINK_EV_CONNECTED:
  return create_sess(rtrs);

 case RTRS_SRV_LINK_EV_DISCONNECTED:
  if (WARN_ON_ONCE(!srv_sess))
   return -EINVAL;

  destroy_sess(srv_sess);
  return 0;

 default:
  pr_warn("Received unknown RTRS session event %d from session %s\n",
   ev, srv_sess->sessname);
  return -EINVAL;
 }
}

void rnbd_srv_sess_dev_force_close(struct rnbd_srv_sess_dev *sess_dev,
       struct kobj_attribute *attr)
{
 struct rnbd_srv_session *sess = sess_dev->sess;

 /* It is already started to close by client's close message. */
 if (!mutex_trylock(&sess->lock))
  return;

 sess_dev->keep_id = true;
 /* first remove sysfs itself to avoid deadlock */
 sysfs_remove_file_self(&sess_dev->kobj, &attr->attr);
 rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 mutex_unlock(&sess->lock);
}

static void process_msg_close(struct rnbd_srv_session *srv_sess,
        void *data, size_t datalen, const void *usr,
        size_t usrlen)
{
 const struct rnbd_msg_close *close_msg = usr;
 struct rnbd_srv_sess_dev *sess_dev;

 trace_process_msg_close(srv_sess, close_msg);

 sess_dev = rnbd_get_sess_dev(le32_to_cpu(close_msg->device_id),
          srv_sess);
 if (IS_ERR(sess_dev))
  return;

 rnbd_put_sess_dev(sess_dev);
 mutex_lock(&srv_sess->lock);
 rnbd_srv_destroy_dev_session_sysfs(sess_dev);
 mutex_unlock(&srv_sess->lock);
}

static int process_msg_open(struct rnbd_srv_session *srv_sess,
       const void *msg, size_t len,
       void *data, size_t datalen);

static void process_msg_sess_info(struct rnbd_srv_session *srv_sess,
     const void *msg, size_t len,
     void *data, size_t datalen);

static int rnbd_srv_rdma_ev(void *priv, struct rtrs_srv_op *id,
       void *data, size_t datalen,
       const void *usr, size_t usrlen)
{
 struct rnbd_srv_session *srv_sess = priv;
 const struct rnbd_msg_hdr *hdr = usr;
 int ret = 0;
 u16 type;

 if (WARN_ON_ONCE(!srv_sess))
  return -ENODEV;

 type = le16_to_cpu(hdr->type);

 switch (type) {
 case RNBD_MSG_IO:
  return process_rdma(srv_sess, id, data, datalen, usr, usrlen);
 case RNBD_MSG_CLOSE:
  process_msg_close(srv_sess, data, datalen, usr, usrlen);
  break;
 case RNBD_MSG_OPEN:
  ret = process_msg_open(srv_sess, usr, usrlen, data, datalen);
  break;
 case RNBD_MSG_SESS_INFO:
  process_msg_sess_info(srv_sess, usr, usrlen, data, datalen);
  break;
 default:
  pr_warn("Received unexpected message type %d from session %s\n",
   type, srv_sess->sessname);
  return -EINVAL;
 }

 /*
 * Since ret is passed to rtrs to handle the failure case, we
 * just return 0 at the end otherwise callers in rtrs would call
 * send_io_resp_imm again to print redundant err message.
 */
 rtrs_srv_resp_rdma(id, ret);
 return 0;
}

static struct rnbd_srv_sess_dev
*rnbd_sess_dev_alloc(struct rnbd_srv_session *srv_sess)
{
 struct rnbd_srv_sess_dev *sess_dev;
 int error;

 sess_dev = kzalloc(sizeof(*sess_dev), GFP_KERNEL);
 if (!sess_dev)
  return ERR_PTR(-ENOMEM);

 error = xa_alloc(&srv_sess->index_idr, &sess_dev->device_id, sess_dev,
    xa_limit_32b, GFP_NOWAIT);
 if (error < 0) {
  pr_warn("Allocating idr failed, err: %d\n", error);
  kfree(sess_dev);
  return ERR_PTR(error);
 }

 return sess_dev;
}

static struct rnbd_srv_dev *rnbd_srv_init_srv_dev(struct block_device *bdev)
{
 struct rnbd_srv_dev *dev;

 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 if (!dev)
  return ERR_PTR(-ENOMEM);

 snprintf(dev->name, sizeof(dev->name), "%pg", bdev);
 kref_init(&dev->kref);
 INIT_LIST_HEAD(&dev->sess_dev_list);
 mutex_init(&dev->lock);

 return dev;
}

static struct rnbd_srv_dev *
rnbd_srv_find_or_add_srv_dev(struct rnbd_srv_dev *new_dev)
{
 struct rnbd_srv_dev *dev;

 spin_lock(&dev_lock);
 list_for_each_entry(dev, &dev_list, list) {
  if (!strncmp(dev->name, new_dev->name, sizeof(dev->name))) {
   if (!kref_get_unless_zero(&dev->kref))
    /*
 * We lost the race, device is almost dead.
 *  Continue traversing to find a valid one.
 */
    continue;
   spin_unlock(&dev_lock);
   return dev;
  }
 }
 list_add(&new_dev->list, &dev_list);
 spin_unlock(&dev_lock);

 return new_dev;
}

static int rnbd_srv_check_update_open_perm(struct rnbd_srv_dev *srv_dev,
         struct rnbd_srv_session *srv_sess,
         enum rnbd_access_mode access_mode)
{
 int ret = 0;

 mutex_lock(&srv_dev->lock);

 switch (access_mode) {
 case RNBD_ACCESS_RO:
  break;
 case RNBD_ACCESS_RW:
  if (srv_dev->open_write_cnt == 0)  {
   srv_dev->open_write_cnt++;
  } else {
   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",
          srv_dev->name, srv_sess->sessname,
          srv_dev->open_write_cnt,
          rnbd_access_modes[access_mode].str);
   ret = -EPERM;
  }
  break;
 case RNBD_ACCESS_MIGRATION:
  if (srv_dev->open_write_cnt < 2) {
   srv_dev->open_write_cnt++;
  } else {
   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",
          srv_dev->name, srv_sess->sessname,
          srv_dev->open_write_cnt,
          rnbd_access_modes[access_mode].str);
   ret = -EPERM;
  }
  break;
 default:
  pr_err("Received mapping request for device '%s' on session %s with invalid access mode: %d\n",
         srv_dev->name, srv_sess->sessname, access_mode);
  ret = -EINVAL;
 }

 mutex_unlock(&srv_dev->lock);

 return ret;
}

static struct rnbd_srv_dev *
rnbd_srv_get_or_create_srv_dev(struct block_device *bdev,
    struct rnbd_srv_session *srv_sess,
    enum rnbd_access_mode access_mode)
{
 int ret;
 struct rnbd_srv_dev *new_dev, *dev;

 new_dev = rnbd_srv_init_srv_dev(bdev);
 if (IS_ERR(new_dev))
  return new_dev;

 dev = rnbd_srv_find_or_add_srv_dev(new_dev);
 if (dev != new_dev)
  kfree(new_dev);

 ret = rnbd_srv_check_update_open_perm(dev, srv_sess, access_mode);
 if (ret) {
  rnbd_put_srv_dev(dev);
  return ERR_PTR(ret);
 }

 return dev;
}

static void rnbd_srv_fill_msg_open_rsp(struct rnbd_msg_open_rsp *rsp,
     struct rnbd_srv_sess_dev *sess_dev)
{
 struct block_device *bdev = file_bdev(sess_dev->bdev_file);

 rsp->hdr.type = cpu_to_le16(RNBD_MSG_OPEN_RSP);
 rsp->device_id = cpu_to_le32(sess_dev->device_id);
 rsp->nsectors = cpu_to_le64(bdev_nr_sectors(bdev));
 rsp->logical_block_size = cpu_to_le16(bdev_logical_block_size(bdev));
 rsp->physical_block_size = cpu_to_le16(bdev_physical_block_size(bdev));
 rsp->max_segments = cpu_to_le16(bdev_max_segments(bdev));
 rsp->max_hw_sectors =
  cpu_to_le32(queue_max_hw_sectors(bdev_get_queue(bdev)));
 rsp->max_write_zeroes_sectors =
  cpu_to_le32(bdev_write_zeroes_sectors(bdev));
 rsp->max_discard_sectors = cpu_to_le32(bdev_max_discard_sectors(bdev));
 rsp->discard_granularity = cpu_to_le32(bdev_discard_granularity(bdev));
 rsp->discard_alignment = cpu_to_le32(bdev_discard_alignment(bdev));
 rsp->secure_discard = cpu_to_le16(bdev_max_secure_erase_sectors(bdev));
 rsp->cache_policy = 0;
 if (bdev_write_cache(bdev))
  rsp->cache_policy |= RNBD_WRITEBACK;
 if (bdev_fua(bdev))
  rsp->cache_policy |= RNBD_FUA;
}

static struct rnbd_srv_sess_dev *
rnbd_srv_create_set_sess_dev(struct rnbd_srv_session *srv_sess,
         const struct rnbd_msg_open *open_msg,
         struct file *bdev_file, bool readonly,
         struct rnbd_srv_dev *srv_dev)
{
 struct rnbd_srv_sess_dev *sdev = rnbd_sess_dev_alloc(srv_sess);

 if (IS_ERR(sdev))
  return sdev;

 kref_init(&sdev->kref);

 strscpy(sdev->pathname, open_msg->dev_name, sizeof(sdev->pathname));

 sdev->bdev_file  = bdev_file;
 sdev->sess  = srv_sess;
 sdev->dev  = srv_dev;
 sdev->readonly  = readonly;
 sdev->access_mode = open_msg->access_mode;

 return sdev;
}

static char *rnbd_srv_get_full_path(struct rnbd_srv_session *srv_sess,
         const char *dev_name)
{
 char *full_path;
 char *a, *b;
 int len;

 full_path = kmalloc(PATH_MAX, GFP_KERNEL);
 if (!full_path)
  return ERR_PTR(-ENOMEM);

 /*
 * Replace %SESSNAME% with a real session name in order to
 * create device namespace.
 */
 a = strnstr(dev_search_path, "%SESSNAME%", sizeof(dev_search_path));
 if (a) {
  len = a - dev_search_path;

  len = snprintf(full_path, PATH_MAX, "%.*s/%s/%s", len,
          dev_search_path, srv_sess->sessname, dev_name);
 } else {
  len = snprintf(full_path, PATH_MAX, "%s/%s",
          dev_search_path, dev_name);
 }
 if (len >= PATH_MAX) {
  pr_err("Too long path: %s, %s, %s\n",
         dev_search_path, srv_sess->sessname, dev_name);
  kfree(full_path);
  return ERR_PTR(-EINVAL);
 }

 /* eliminitate duplicated slashes */
 a = strchr(full_path, '/');
 b = a;
 while (*b != '\0') {
  if (*b == '/' && *a == '/') {
   b++;
  } else {
   a++;
   *a = *b;
   b++;
  }
 }
 a++;
 *a = '\0';

 return full_path;
}

static void process_msg_sess_info(struct rnbd_srv_session *srv_sess,
     const void *msg, size_t len,
     void *data, size_t datalen)
{
 const struct rnbd_msg_sess_info *sess_info_msg = msg;
 struct rnbd_msg_sess_info_rsp *rsp = data;

 srv_sess->ver = min_t(u8, sess_info_msg->ver, RNBD_PROTO_VER_MAJOR);

 trace_process_msg_sess_info(srv_sess, sess_info_msg);

 rsp->hdr.type = cpu_to_le16(RNBD_MSG_SESS_INFO_RSP);
 rsp->ver = srv_sess->ver;
}

/**
 * find_srv_sess_dev() - a dev is already opened by this name
 * @srv_sess: the session to search.
 * @dev_name: string containing the name of the device.
 *
 * Return struct rnbd_srv_sess_dev if srv_sess already opened the dev_name
 * NULL if the session didn't open the device yet.
 */
static struct rnbd_srv_sess_dev *
find_srv_sess_dev(struct rnbd_srv_session *srv_sess, const char *dev_name)
{
 struct rnbd_srv_sess_dev *sess_dev;
 unsigned long index;

 if (xa_empty(&srv_sess->index_idr))
  return NULL;

 xa_for_each(&srv_sess->index_idr, index, sess_dev)
  if (!strcmp(sess_dev->pathname, dev_name))
   return sess_dev;

 return NULL;
}

static int process_msg_open(struct rnbd_srv_session *srv_sess,
       const void *msg, size_t len,
       void *data, size_t datalen)
{
 int ret;
 struct rnbd_srv_dev *srv_dev;
 struct rnbd_srv_sess_dev *srv_sess_dev;
 const struct rnbd_msg_open *open_msg = msg;
 struct file *bdev_file;
 blk_mode_t open_flags = BLK_OPEN_READ;
 char *full_path;
 struct rnbd_msg_open_rsp *rsp = data;

 trace_process_msg_open(srv_sess, open_msg);

 if (open_msg->access_mode != RNBD_ACCESS_RO)
  open_flags |= BLK_OPEN_WRITE;

 mutex_lock(&srv_sess->lock);

 srv_sess_dev = find_srv_sess_dev(srv_sess, open_msg->dev_name);
 if (srv_sess_dev)
  goto fill_response;

 if ((strlen(dev_search_path) + strlen(open_msg->dev_name))
     >= PATH_MAX) {
  pr_err("Opening device for session %s failed, device path too long. '%s/%s' is longer than PATH_MAX (%d)\n",
         srv_sess->sessname, dev_search_path, open_msg->dev_name,
         PATH_MAX);
  ret = -EINVAL;
  goto reject;
 }
 if (strstr(open_msg->dev_name, "..")) {
  pr_err("Opening device for session %s failed, device path %s contains relative path ..\n",
         srv_sess->sessname, open_msg->dev_name);
  ret = -EINVAL;
  goto reject;
 }
 full_path = rnbd_srv_get_full_path(srv_sess, open_msg->dev_name);
 if (IS_ERR(full_path)) {
  ret = PTR_ERR(full_path);
  pr_err("Opening device '%s' for client %s failed, failed to get device full path, err: %pe\n",
         open_msg->dev_name, srv_sess->sessname, full_path);
  goto reject;
 }

 bdev_file = bdev_file_open_by_path(full_path, open_flags, NULL, NULL);
 if (IS_ERR(bdev_file)) {
  ret = PTR_ERR(bdev_file);
  pr_err("Opening device '%s' on session %s failed, failed to open the block device, err: %pe\n",
         full_path, srv_sess->sessname, bdev_file);
  goto free_path;
 }

 srv_dev = rnbd_srv_get_or_create_srv_dev(file_bdev(bdev_file), srv_sess,
        open_msg->access_mode);
 if (IS_ERR(srv_dev)) {
  pr_err("Opening device '%s' on session %s failed, creating srv_dev failed, err: %pe\n",
         full_path, srv_sess->sessname, srv_dev);
  ret = PTR_ERR(srv_dev);
  goto blkdev_put;
 }

 srv_sess_dev = rnbd_srv_create_set_sess_dev(srv_sess, open_msg,
    bdev_file,
    open_msg->access_mode == RNBD_ACCESS_RO,
    srv_dev);
 if (IS_ERR(srv_sess_dev)) {
  pr_err("Opening device '%s' on session %s failed, creating sess_dev failed, err: %pe\n",
         full_path, srv_sess->sessname, srv_sess_dev);
  ret = PTR_ERR(srv_sess_dev);
  goto srv_dev_put;
 }

 /* Create the srv_dev sysfs files if they haven't been created yet. The
 * reason to delay the creation is not to create the sysfs files before
 * we are sure the device can be opened.
 */
 mutex_lock(&srv_dev->lock);
 if (!srv_dev->dev_kobj.state_in_sysfs) {
  ret = rnbd_srv_create_dev_sysfs(srv_dev, file_bdev(bdev_file));
  if (ret) {
   mutex_unlock(&srv_dev->lock);
   rnbd_srv_err(srv_sess_dev,
          "Opening device failed, failed to create device sysfs files, err: %d\n",
          ret);
   goto free_srv_sess_dev;
  }
 }

 ret = rnbd_srv_create_dev_session_sysfs(srv_sess_dev);
 if (ret) {
  mutex_unlock(&srv_dev->lock);
  rnbd_srv_err(srv_sess_dev,
         "Opening device failed, failed to create dev client sysfs files, err: %d\n",
         ret);
  goto free_srv_sess_dev;
 }

 list_add(&srv_sess_dev->dev_list, &srv_dev->sess_dev_list);
 mutex_unlock(&srv_dev->lock);

 rnbd_srv_info(srv_sess_dev, "Opened device '%s'\n", srv_dev->name);

 kfree(full_path);

fill_response:
 rnbd_srv_fill_msg_open_rsp(rsp, srv_sess_dev);
 mutex_unlock(&srv_sess->lock);
 return 0;

free_srv_sess_dev:
 xa_erase(&srv_sess->index_idr, srv_sess_dev->device_id);
 synchronize_rcu();
 kfree(srv_sess_dev);
srv_dev_put:
 if (open_msg->access_mode != RNBD_ACCESS_RO) {
  mutex_lock(&srv_dev->lock);
  srv_dev->open_write_cnt--;
  mutex_unlock(&srv_dev->lock);
 }
 rnbd_put_srv_dev(srv_dev);
blkdev_put:
 fput(bdev_file);
free_path:
 kfree(full_path);
reject:
 mutex_unlock(&srv_sess->lock);
 return ret;
}

static struct rtrs_srv_ctx *rtrs_ctx;

static struct rtrs_srv_ops rtrs_ops;
static int __init rnbd_srv_init_module(void)
{
 int err = 0;

 BUILD_BUG_ON(sizeof(struct rnbd_msg_hdr) != 4);
 BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info) != 36);
 BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info_rsp) != 36);
 BUILD_BUG_ON(sizeof(struct rnbd_msg_open) != 264);
 BUILD_BUG_ON(sizeof(struct rnbd_msg_close) != 8);
 BUILD_BUG_ON(sizeof(struct rnbd_msg_open_rsp) != 56);
 rtrs_ops = (struct rtrs_srv_ops) {
  .rdma_ev = rnbd_srv_rdma_ev,
  .link_ev = rnbd_srv_link_ev,
 };
 rtrs_ctx = rtrs_srv_open(&rtrs_ops, port_nr);
 if (IS_ERR(rtrs_ctx)) {
  pr_err("rtrs_srv_open(), err: %pe\n", rtrs_ctx);
  return PTR_ERR(rtrs_ctx);
 }

 err = rnbd_srv_create_sysfs_files();
 if (err) {
  pr_err("rnbd_srv_create_sysfs_files(), err: %d\n", err);
  rtrs_srv_close(rtrs_ctx);
 }

 return err;
}

static void __exit rnbd_srv_cleanup_module(void)
{
 rtrs_srv_close(rtrs_ctx);
 WARN_ON(!list_empty(&sess_list));
 rnbd_srv_destroy_sysfs_files();
}

module_init(rnbd_srv_init_module);
module_exit(rnbd_srv_cleanup_module);