gnu/packages/patches/ucx-tcp-iface-ioctl.patch

   1 Since /sys is unavailable in build environments, the list of available
   2 TCP network interfaces cannot be obtained via /sys/class/net.  This patch
   3 provides alternative code that uses the SIOCGIFCONF ioctl to get the
   4 names of the available TCP network interfaces.
   5
   6 diff --git a/src/uct/tcp/tcp_iface.c b/src/uct/tcp/tcp_iface.c
   7 index 81ad459..10024a6 100644
   8 --- a/src/uct/tcp/tcp_iface.c
   9 +++ b/src/uct/tcp/tcp_iface.c
  10 @@ -12,6 +12,8 @@
  11  #include <sys/poll.h>
  12  #include <netinet/tcp.h>
  13  #include <dirent.h>
  14 +#include <net/if.h>
  15 +#include <sys/ioctl.h>
  16
  17  static ucs_config_field_t uct_tcp_iface_config_table[] = {
  18    {"", "MAX_SHORT=8k", NULL,
  19 @@ -483,6 +485,70 @@ static UCS_CLASS_DEFINE_NEW_FUNC(uct_tcp_iface_t, uct_iface_t, uct_md_h,
  20                                   uct_worker_h, const uct_iface_params_t*,
  21                                   const uct_iface_config_t*);
  22
  23 +/* Fetch information about available network devices through an ioctl.  */
  24 +static ucs_status_t query_devices_ioctl(uct_md_h md,
  25 +                                       uct_tl_resource_desc_t **resource_p,
  26 +                                       unsigned *num_resources_p)
  27 +{
  28 +    int sock, err, i;
  29 +    uct_tl_resource_desc_t *resources, *tmp;
  30 +    unsigned num_resources;
  31 +    ucs_status_t status;
  32 +    struct ifconf conf;
  33 +    struct ifreq reqs[10];
  34 +
  35 +    conf.ifc_len = sizeof reqs;
  36 +    conf.ifc_req = reqs;
  37 +
  38 +    sock = socket(SOCK_STREAM, AF_INET, 0);
  39 +    if (sock < 0) {
  40 +       ucs_error("socket(2) failed: %m");
  41 +       status = UCS_ERR_IO_ERROR;
  42 +       goto out;
  43 +    }
  44 +
  45 +    err = ioctl(sock, SIOCGIFCONF, &conf);
  46 +    if (err < 0) {
  47 +       ucs_error("SIOCGIFCONF ioctl failed: %m");
  48 +       status = UCS_ERR_IO_ERROR;
  49 +       goto out;
  50 +    }
  51 +
  52 +    resources     = NULL;
  53 +    num_resources = 0;
  54 +    for (i = 0; i < conf.ifc_len / sizeof(struct ifreq); i++) {
  55 +       const char *name = reqs[i].ifr_name;
  56 +
  57 +        if (!ucs_netif_is_active(name)) {
  58 +            continue;
  59 +        }
  60 +
  61 +        tmp = ucs_realloc(resources, sizeof(*resources) * (num_resources + 1),
  62 +                          "tcp resources");
  63 +        if (tmp == NULL) {
  64 +            ucs_free(resources);
  65 +            status = UCS_ERR_NO_MEMORY;
  66 +            goto out;
  67 +        }
  68 +        resources = tmp;
  69 +
  70 +        ucs_snprintf_zero(resources[i].tl_name, sizeof(resources[i].tl_name),
  71 +                          "%s", UCT_TCP_NAME);
  72 +        ucs_snprintf_zero(resources[i].dev_name, sizeof(resources[i].dev_name),
  73 +                          "%s", name);
  74 +        resources[i].dev_type = UCT_DEVICE_TYPE_NET;
  75 +        ++num_resources;
  76 +    }
  77 +
  78 +    *num_resources_p = num_resources;
  79 +    *resource_p      = resources;
  80 +    status           = UCS_OK;
  81 +
  82 +out:
  83 +    if (sock >= 0) close(sock);
  84 +    return status;
  85 +}
  86 +
  87  static ucs_status_t uct_tcp_query_tl_resources(uct_md_h md,
  88                                                 uct_tl_resource_desc_t **resource_p,
  89                                                 unsigned *num_resources_p)
  90 @@ -496,9 +562,9 @@ static ucs_status_t uct_tcp_query_tl_resources(uct_md_h md,
  91
  92      dir = opendir(netdev_dir);
  93      if (dir == NULL) {
  94 -        ucs_error("opendir(%s) failed: %m", netdev_dir);
  95 -        status = UCS_ERR_IO_ERROR;
  96 -        goto out;
  97 +       /* When /sys is unavailable, as can be the case in a container,
  98 +        * resort to a good old 'ioctl'.  */
  99 +       return query_devices_ioctl(md, resource_p, num_resources_p);
 100      }
 101
 102      resources     = NULL;
 103 @@ -543,6 +609,5 @@ static ucs_status_t uct_tcp_query_tl_resources(uct_md_h md,
 104
 105  out_closedir:
 106      closedir(dir);
 107 -out:
 108      return status;
 109  }