ceph/src/seastar/dpdk/drivers/net/tap/tap_netlink.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright 2017 6WIND S.A.
   5  *   Copyright 2017 Mellanox.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of 6WIND S.A. nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <errno.h>
  35 #include <inttypes.h>
  36 #include <linux/netlink.h>
  37 #include <string.h>
  38 #include <sys/socket.h>
  39 #include <unistd.h>
  40
  41 #include <rte_malloc.h>
  42 #include <tap_netlink.h>
  43 #include <rte_random.h>
  44
  45 /* Must be quite large to support dumping a huge list of QDISC or filters. */
  46 #define BUF_SIZE (32 * 1024) /* Size of the buffer to receive kernel messages */
  47 #define SNDBUF_SIZE 32768 /* Send buffer size for the netlink socket */
  48 #define RCVBUF_SIZE 32768 /* Receive buffer size for the netlink socket */
  49
  50 struct nested_tail {
  51         struct rtattr *tail;
  52         struct nested_tail *prev;
  53 };
  54
  55 /**
  56  * Initialize a netlink socket for communicating with the kernel.
  57  *
  58  * @param nl_groups
  59  *   Set it to a netlink group value (e.g. RTMGRP_LINK) to receive messages for
  60  *   specific netlink multicast groups. Otherwise, no subscription will be made.
  61  *
  62  * @return
  63  *   netlink socket file descriptor on success, -1 otherwise.
  64  */
  65 int
  66 nl_init(uint32_t nl_groups)
  67 {
  68         int fd, sndbuf_size = SNDBUF_SIZE, rcvbuf_size = RCVBUF_SIZE;
  69         struct sockaddr_nl local = {
  70                 .nl_family = AF_NETLINK,
  71                 .nl_groups = nl_groups,
  72         };
  73
  74         fd = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, NETLINK_ROUTE);
  75         if (fd < 0) {
  76                 RTE_LOG(ERR, PMD, "Unable to create a netlink socket\n");
  77                 return -1;
  78         }
  79         if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &sndbuf_size, sizeof(int))) {
  80                 RTE_LOG(ERR, PMD, "Unable to set socket buffer send size\n");
  81                 return -1;
  82         }
  83         if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf_size, sizeof(int))) {
  84                 RTE_LOG(ERR, PMD, "Unable to set socket buffer receive size\n");
  85                 return -1;
  86         }
  87         if (bind(fd, (struct sockaddr *)&local, sizeof(local)) < 0) {
  88                 RTE_LOG(ERR, PMD, "Unable to bind to the netlink socket\n");
  89                 return -1;
  90         }
  91         return fd;
  92 }
  93
  94 /**
  95  * Clean up a netlink socket once all communicating with the kernel is finished.
  96  *
  97  * @param[in] nlsk_fd
  98  *   The netlink socket file descriptor used for communication.
  99  *
 100  * @return
 101  *   0 on success, -1 otherwise.
 102  */
 103 int
 104 nl_final(int nlsk_fd)
 105 {
 106         if (close(nlsk_fd)) {
 107                 RTE_LOG(ERR, PMD, "Failed to close netlink socket: %s (%d)\n",
 108                         strerror(errno), errno);
 109                 return -1;
 110         }
 111         return 0;
 112 }
 113
 114 /**
 115  * Send a message to the kernel on the netlink socket.
 116  *
 117  * @param[in] nlsk_fd
 118  *   The netlink socket file descriptor used for communication.
 119  * @param[in] nh
 120  *   The netlink message send to the kernel.
 121  *
 122  * @return
 123  *   the number of sent bytes on success, -1 otherwise.
 124  */
 125 int
 126 nl_send(int nlsk_fd, struct nlmsghdr *nh)
 127 {
 128         /* man 7 netlink EXAMPLE */
 129         struct sockaddr_nl sa = {
 130                 .nl_family = AF_NETLINK,
 131         };
 132         struct iovec iov = {
 133                 .iov_base = nh,
 134                 .iov_len = nh->nlmsg_len,
 135         };
 136         struct msghdr msg = {
 137                 .msg_name = &sa,
 138                 .msg_namelen = sizeof(sa),
 139                 .msg_iov = &iov,
 140                 .msg_iovlen = 1,
 141         };
 142         int send_bytes;
 143
 144         nh->nlmsg_pid = 0; /* communication with the kernel uses pid 0 */
 145         nh->nlmsg_seq = (uint32_t)rte_rand();
 146         send_bytes = sendmsg(nlsk_fd, &msg, 0);
 147         if (send_bytes < 0) {
 148                 RTE_LOG(ERR, PMD, "Failed to send netlink message: %s (%d)\n",
 149                         strerror(errno), errno);
 150                 return -1;
 151         }
 152         return send_bytes;
 153 }
 154
 155 /**
 156  * Check that the kernel sends an appropriate ACK in response to an nl_send().
 157  *
 158  * @param[in] nlsk_fd
 159  *   The netlink socket file descriptor used for communication.
 160  *
 161  * @return
 162  *   0 on success, -1 otherwise with errno set.
 163  */
 164 int
 165 nl_recv_ack(int nlsk_fd)
 166 {
 167         return nl_recv(nlsk_fd, NULL, NULL);
 168 }
 169
 170 /**
 171  * Receive a message from the kernel on the netlink socket, following an
 172  * nl_send().
 173  *
 174  * @param[in] nlsk_fd
 175  *   The netlink socket file descriptor used for communication.
 176  * @param[in] cb
 177  *   The callback function to call for each netlink message received.
 178  * @param[in, out] arg
 179  *   Custom arguments for the callback.
 180  *
 181  * @return
 182  *   0 on success, -1 otherwise with errno set.
 183  */
 184 int
 185 nl_recv(int nlsk_fd, int (*cb)(struct nlmsghdr *, void *arg), void *arg)
 186 {
 187         /* man 7 netlink EXAMPLE */
 188         struct sockaddr_nl sa;
 189         char buf[BUF_SIZE];
 190         struct iovec iov = {
 191                 .iov_base = buf,
 192                 .iov_len = sizeof(buf),
 193         };
 194         struct msghdr msg = {
 195                 .msg_name = &sa,
 196                 .msg_namelen = sizeof(sa),
 197                 .msg_iov = &iov,
 198                 /* One message at a time */
 199                 .msg_iovlen = 1,
 200         };
 201         int multipart = 0;
 202         int ret = 0;
 203
 204         do {
 205                 struct nlmsghdr *nh;
 206                 int recv_bytes = 0;
 207
 208                 recv_bytes = recvmsg(nlsk_fd, &msg, 0);
 209                 if (recv_bytes < 0)
 210                         return -1;
 211                 for (nh = (struct nlmsghdr *)buf;
 212                      NLMSG_OK(nh, (unsigned int)recv_bytes);
 213                      nh = NLMSG_NEXT(nh, recv_bytes)) {
 214                         if (nh->nlmsg_type == NLMSG_ERROR) {
 215                                 struct nlmsgerr *err_data = NLMSG_DATA(nh);
 216
 217                                 if (err_data->error < 0) {
 218                                         errno = -err_data->error;
 219                                         return -1;
 220                                 }
 221                                 /* Ack message. */
 222                                 return 0;
 223                         }
 224                         /* Multi-part msgs and their trailing DONE message. */
 225                         if (nh->nlmsg_flags & NLM_F_MULTI) {
 226                                 if (nh->nlmsg_type == NLMSG_DONE)
 227                                         return 0;
 228                                 multipart = 1;
 229                         }
 230                         if (cb)
 231                                 ret = cb(nh, arg);
 232                 }
 233         } while (multipart);
 234         return ret;
 235 }
 236
 237 /**
 238  * Append a netlink attribute to a message.
 239  *
 240  * @param[in, out] nh
 241  *   The netlink message to parse, received from the kernel.
 242  * @param[in] type
 243  *   The type of attribute to append.
 244  * @param[in] data_len
 245  *   The length of the data to append.
 246  * @param[in] data
 247  *   The data to append.
 248  */
 249 void
 250 nlattr_add(struct nlmsghdr *nh, unsigned short type,
 251            unsigned int data_len, const void *data)
 252 {
 253         /* see man 3 rtnetlink */
 254         struct rtattr *rta;
 255
 256         rta = (struct rtattr *)NLMSG_TAIL(nh);
 257         rta->rta_len = RTA_LENGTH(data_len);
 258         rta->rta_type = type;
 259         memcpy(RTA_DATA(rta), data, data_len);
 260         nh->nlmsg_len = NLMSG_ALIGN(nh->nlmsg_len) + RTA_ALIGN(rta->rta_len);
 261 }
 262
 263 /**
 264  * Append a uint8_t netlink attribute to a message.
 265  *
 266  * @param[in, out] nh
 267  *   The netlink message to parse, received from the kernel.
 268  * @param[in] type
 269  *   The type of attribute to append.
 270  * @param[in] data
 271  *   The data to append.
 272  */
 273 void
 274 nlattr_add8(struct nlmsghdr *nh, unsigned short type, uint8_t data)
 275 {
 276         nlattr_add(nh, type, sizeof(uint8_t), &data);
 277 }
 278
 279 /**
 280  * Append a uint16_t netlink attribute to a message.
 281  *
 282  * @param[in, out] nh
 283  *   The netlink message to parse, received from the kernel.
 284  * @param[in] type
 285  *   The type of attribute to append.
 286  * @param[in] data
 287  *   The data to append.
 288  */
 289 void
 290 nlattr_add16(struct nlmsghdr *nh, unsigned short type, uint16_t data)
 291 {
 292         nlattr_add(nh, type, sizeof(uint16_t), &data);
 293 }
 294
 295 /**
 296  * Append a uint16_t netlink attribute to a message.
 297  *
 298  * @param[in, out] nh
 299  *   The netlink message to parse, received from the kernel.
 300  * @param[in] type
 301  *   The type of attribute to append.
 302  * @param[in] data
 303  *   The data to append.
 304  */
 305 void
 306 nlattr_add32(struct nlmsghdr *nh, unsigned short type, uint32_t data)
 307 {
 308         nlattr_add(nh, type, sizeof(uint32_t), &data);
 309 }
 310
 311 /**
 312  * Start a nested netlink attribute.
 313  * It must be followed later by a call to nlattr_nested_finish().
 314  *
 315  * @param[in, out] msg
 316  *   The netlink message where to edit the nested_tails metadata.
 317  * @param[in] type
 318  *   The nested attribute type to append.
 319  *
 320  * @return
 321  *   -1 if adding a nested netlink attribute failed, 0 otherwise.
 322  */
 323 int
 324 nlattr_nested_start(struct nlmsg *msg, uint16_t type)
 325 {
 326         struct nested_tail *tail;
 327
 328         tail = rte_zmalloc(NULL, sizeof(struct nested_tail), 0);
 329         if (!tail) {
 330                 RTE_LOG(ERR, PMD,
 331                         "Couldn't allocate memory for nested netlink"
 332                         " attribute\n");
 333                 return -1;
 334         }
 335
 336         tail->tail = (struct rtattr *)NLMSG_TAIL(&msg->nh);
 337
 338         nlattr_add(&msg->nh, type, 0, NULL);
 339
 340         tail->prev = msg->nested_tails;
 341
 342         msg->nested_tails = tail;
 343
 344         return 0;
 345 }
 346
 347 /**
 348  * End a nested netlink attribute.
 349  * It follows a call to nlattr_nested_start().
 350  * In effect, it will modify the nested attribute length to include every bytes
 351  * from the nested attribute start, up to here.
 352  *
 353  * @param[in, out] msg
 354  *   The netlink message where to edit the nested_tails metadata.
 355  */
 356 void
 357 nlattr_nested_finish(struct nlmsg *msg)
 358 {
 359         struct nested_tail *tail = msg->nested_tails;
 360
 361         tail->tail->rta_len = (char *)NLMSG_TAIL(&msg->nh) - (char *)tail->tail;
 362
 363         if (tail->prev)
 364                 msg->nested_tails = tail->prev;
 365
 366         rte_free(tail);
 367 }