Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 1 | /* |
| 2 | * QEMU paravirtual RDMA - rdmacm-mux implementation |
| 3 | * |
| 4 | * Copyright (C) 2018 Oracle |
| 5 | * Copyright (C) 2018 Red Hat Inc |
| 6 | * |
| 7 | * Authors: |
| 8 | * Yuval Shaia <yuval.shaia@oracle.com> |
| 9 | * Marcel Apfelbaum <marcel@redhat.com> |
| 10 | * |
| 11 | * This work is licensed under the terms of the GNU GPL, version 2 or later. |
| 12 | * See the COPYING file in the top-level directory. |
| 13 | * |
| 14 | */ |
| 15 | |
| 16 | #include "qemu/osdep.h" |
Markus Armbruster | 1b6d3b5 | 2019-03-15 15:51:16 +0100 | [diff] [blame] | 17 | #include <sys/poll.h> |
| 18 | #include <sys/ioctl.h> |
| 19 | #include <pthread.h> |
| 20 | #include <syslog.h> |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 21 | |
Markus Armbruster | 1b6d3b5 | 2019-03-15 15:51:16 +0100 | [diff] [blame] | 22 | #include <infiniband/verbs.h> |
| 23 | #include <infiniband/umad.h> |
| 24 | #include <infiniband/umad_types.h> |
| 25 | #include <infiniband/umad_sa.h> |
| 26 | #include <infiniband/umad_cm.h> |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 27 | |
| 28 | #include "rdmacm-mux.h" |
| 29 | |
| 30 | #define SCALE_US 1000 |
| 31 | #define COMMID_TTL 2 /* How many SCALE_US a context of MAD session is saved */ |
| 32 | #define SLEEP_SECS 5 /* This is used both in poll() and thread */ |
| 33 | #define SERVER_LISTEN_BACKLOG 10 |
| 34 | #define MAX_CLIENTS 4096 |
| 35 | #define MAD_RMPP_VERSION 0 |
| 36 | #define MAD_METHOD_MASK0 0x8 |
| 37 | |
| 38 | #define IB_USER_MAD_LONGS_PER_METHOD_MASK (128 / (8 * sizeof(long))) |
| 39 | |
| 40 | #define CM_REQ_DGID_POS 80 |
| 41 | #define CM_SIDR_REQ_DGID_POS 44 |
| 42 | |
| 43 | /* The below can be override by command line parameter */ |
| 44 | #define UNIX_SOCKET_PATH "/var/run/rdmacm-mux" |
Marcel Apfelbaum | 4a3d55a | 2019-01-18 14:46:12 +0200 | [diff] [blame] | 45 | /* Has format %s-%s-%d" <path>-<rdma-dev--name>-<port> */ |
| 46 | #define SOCKET_PATH_MAX (PATH_MAX - NAME_MAX - sizeof(int) - 2) |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 47 | #define RDMA_PORT_NUM 1 |
| 48 | |
| 49 | typedef struct RdmaCmServerArgs { |
| 50 | char unix_socket_path[PATH_MAX]; |
| 51 | char rdma_dev_name[NAME_MAX]; |
| 52 | int rdma_port_num; |
| 53 | } RdmaCMServerArgs; |
| 54 | |
| 55 | typedef struct CommId2FdEntry { |
| 56 | int fd; |
| 57 | int ttl; /* Initialized to 2, decrement each timeout, entry delete when 0 */ |
| 58 | __be64 gid_ifid; |
| 59 | } CommId2FdEntry; |
| 60 | |
| 61 | typedef struct RdmaCmUMadAgent { |
| 62 | int port_id; |
| 63 | int agent_id; |
| 64 | GHashTable *gid2fd; /* Used to find fd of a given gid */ |
| 65 | GHashTable *commid2fd; /* Used to find fd on of a given comm_id */ |
| 66 | } RdmaCmUMadAgent; |
| 67 | |
| 68 | typedef struct RdmaCmServer { |
| 69 | bool run; |
| 70 | RdmaCMServerArgs args; |
| 71 | struct pollfd fds[MAX_CLIENTS]; |
| 72 | int nfds; |
| 73 | RdmaCmUMadAgent umad_agent; |
| 74 | pthread_t umad_recv_thread; |
| 75 | pthread_rwlock_t lock; |
| 76 | } RdmaCMServer; |
| 77 | |
| 78 | static RdmaCMServer server = {0}; |
| 79 | |
| 80 | static void usage(const char *progname) |
| 81 | { |
| 82 | printf("Usage: %s [OPTION]...\n" |
| 83 | "Start a RDMA-CM multiplexer\n" |
| 84 | "\n" |
| 85 | "\t-h Show this help\n" |
| 86 | "\t-d rdma-device-name Name of RDMA device to register with\n" |
| 87 | "\t-s unix-socket-path Path to unix socket to listen on (default %s)\n" |
| 88 | "\t-p rdma-device-port Port number of RDMA device to register with (default %d)\n", |
| 89 | progname, UNIX_SOCKET_PATH, RDMA_PORT_NUM); |
| 90 | } |
| 91 | |
| 92 | static void help(const char *progname) |
| 93 | { |
| 94 | fprintf(stderr, "Try '%s -h' for more information.\n", progname); |
| 95 | } |
| 96 | |
| 97 | static void parse_args(int argc, char *argv[]) |
| 98 | { |
| 99 | int c; |
Marcel Apfelbaum | 4a3d55a | 2019-01-18 14:46:12 +0200 | [diff] [blame] | 100 | char unix_socket_path[SOCKET_PATH_MAX]; |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 101 | |
| 102 | strcpy(server.args.rdma_dev_name, ""); |
| 103 | strcpy(unix_socket_path, UNIX_SOCKET_PATH); |
| 104 | server.args.rdma_port_num = RDMA_PORT_NUM; |
| 105 | |
| 106 | while ((c = getopt(argc, argv, "hs:d:p:")) != -1) { |
| 107 | switch (c) { |
| 108 | case 'h': |
| 109 | usage(argv[0]); |
| 110 | exit(0); |
| 111 | |
| 112 | case 'd': |
| 113 | strncpy(server.args.rdma_dev_name, optarg, NAME_MAX - 1); |
| 114 | break; |
| 115 | |
| 116 | case 's': |
| 117 | /* This is temporary, final name will build below */ |
Marc-André Lureau | 60ae0b9 | 2019-07-12 20:51:54 +0400 | [diff] [blame] | 118 | strncpy(unix_socket_path, optarg, SOCKET_PATH_MAX - 1); |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 119 | break; |
| 120 | |
| 121 | case 'p': |
| 122 | server.args.rdma_port_num = atoi(optarg); |
| 123 | break; |
| 124 | |
| 125 | default: |
| 126 | help(argv[0]); |
| 127 | exit(1); |
| 128 | } |
| 129 | } |
| 130 | |
| 131 | if (!strcmp(server.args.rdma_dev_name, "")) { |
| 132 | fprintf(stderr, "Missing RDMA device name\n"); |
| 133 | help(argv[0]); |
| 134 | exit(1); |
| 135 | } |
| 136 | |
| 137 | /* Build unique unix-socket file name */ |
| 138 | snprintf(server.args.unix_socket_path, PATH_MAX, "%s-%s-%d", |
| 139 | unix_socket_path, server.args.rdma_dev_name, |
| 140 | server.args.rdma_port_num); |
| 141 | |
| 142 | syslog(LOG_INFO, "unix_socket_path=%s", server.args.unix_socket_path); |
| 143 | syslog(LOG_INFO, "rdma-device-name=%s", server.args.rdma_dev_name); |
| 144 | syslog(LOG_INFO, "rdma-device-port=%d", server.args.rdma_port_num); |
| 145 | } |
| 146 | |
| 147 | static void hash_tbl_alloc(void) |
| 148 | { |
| 149 | |
| 150 | server.umad_agent.gid2fd = g_hash_table_new_full(g_int64_hash, |
| 151 | g_int64_equal, |
| 152 | g_free, g_free); |
| 153 | server.umad_agent.commid2fd = g_hash_table_new_full(g_int_hash, |
| 154 | g_int_equal, |
| 155 | g_free, g_free); |
| 156 | } |
| 157 | |
| 158 | static void hash_tbl_free(void) |
| 159 | { |
| 160 | if (server.umad_agent.commid2fd) { |
| 161 | g_hash_table_destroy(server.umad_agent.commid2fd); |
| 162 | } |
| 163 | if (server.umad_agent.gid2fd) { |
| 164 | g_hash_table_destroy(server.umad_agent.gid2fd); |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | |
| 169 | static int _hash_tbl_search_fd_by_ifid(__be64 *gid_ifid) |
| 170 | { |
| 171 | int *fd; |
| 172 | |
| 173 | fd = g_hash_table_lookup(server.umad_agent.gid2fd, gid_ifid); |
| 174 | if (!fd) { |
| 175 | /* Let's try IPv4 */ |
| 176 | *gid_ifid |= 0x00000000ffff0000; |
| 177 | fd = g_hash_table_lookup(server.umad_agent.gid2fd, gid_ifid); |
| 178 | } |
| 179 | |
| 180 | return fd ? *fd : 0; |
| 181 | } |
| 182 | |
| 183 | static int hash_tbl_search_fd_by_ifid(int *fd, __be64 *gid_ifid) |
| 184 | { |
| 185 | pthread_rwlock_rdlock(&server.lock); |
| 186 | *fd = _hash_tbl_search_fd_by_ifid(gid_ifid); |
| 187 | pthread_rwlock_unlock(&server.lock); |
| 188 | |
AlexChen | eed31bc | 2020-10-29 21:43:27 +0800 | [diff] [blame] | 189 | if (!*fd) { |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 190 | syslog(LOG_WARNING, "Can't find matching for ifid 0x%llx\n", *gid_ifid); |
| 191 | return -ENOENT; |
| 192 | } |
| 193 | |
| 194 | return 0; |
| 195 | } |
| 196 | |
| 197 | static int hash_tbl_search_fd_by_comm_id(uint32_t comm_id, int *fd, |
| 198 | __be64 *gid_idid) |
| 199 | { |
| 200 | CommId2FdEntry *fde; |
| 201 | |
| 202 | pthread_rwlock_rdlock(&server.lock); |
| 203 | fde = g_hash_table_lookup(server.umad_agent.commid2fd, &comm_id); |
| 204 | pthread_rwlock_unlock(&server.lock); |
| 205 | |
| 206 | if (!fde) { |
| 207 | syslog(LOG_WARNING, "Can't find matching for comm_id 0x%x\n", comm_id); |
| 208 | return -ENOENT; |
| 209 | } |
| 210 | |
| 211 | *fd = fde->fd; |
| 212 | *gid_idid = fde->gid_ifid; |
| 213 | |
| 214 | return 0; |
| 215 | } |
| 216 | |
| 217 | static RdmaCmMuxErrCode add_fd_ifid_pair(int fd, __be64 gid_ifid) |
| 218 | { |
| 219 | int fd1; |
| 220 | |
| 221 | pthread_rwlock_wrlock(&server.lock); |
| 222 | |
| 223 | fd1 = _hash_tbl_search_fd_by_ifid(&gid_ifid); |
| 224 | if (fd1) { /* record already exist - an error */ |
| 225 | pthread_rwlock_unlock(&server.lock); |
| 226 | return fd == fd1 ? RDMACM_MUX_ERR_CODE_EEXIST : |
| 227 | RDMACM_MUX_ERR_CODE_EACCES; |
| 228 | } |
| 229 | |
| 230 | g_hash_table_insert(server.umad_agent.gid2fd, g_memdup(&gid_ifid, |
| 231 | sizeof(gid_ifid)), g_memdup(&fd, sizeof(fd))); |
| 232 | |
| 233 | pthread_rwlock_unlock(&server.lock); |
| 234 | |
| 235 | syslog(LOG_INFO, "0x%lx registered on socket %d", |
| 236 | be64toh((uint64_t)gid_ifid), fd); |
| 237 | |
| 238 | return RDMACM_MUX_ERR_CODE_OK; |
| 239 | } |
| 240 | |
| 241 | static RdmaCmMuxErrCode delete_fd_ifid_pair(int fd, __be64 gid_ifid) |
| 242 | { |
| 243 | int fd1; |
| 244 | |
| 245 | pthread_rwlock_wrlock(&server.lock); |
| 246 | |
| 247 | fd1 = _hash_tbl_search_fd_by_ifid(&gid_ifid); |
| 248 | if (!fd1) { /* record not exist - an error */ |
| 249 | pthread_rwlock_unlock(&server.lock); |
| 250 | return RDMACM_MUX_ERR_CODE_ENOTFOUND; |
| 251 | } |
| 252 | |
| 253 | g_hash_table_remove(server.umad_agent.gid2fd, g_memdup(&gid_ifid, |
| 254 | sizeof(gid_ifid))); |
| 255 | pthread_rwlock_unlock(&server.lock); |
| 256 | |
| 257 | syslog(LOG_INFO, "0x%lx unregistered on socket %d", |
| 258 | be64toh((uint64_t)gid_ifid), fd); |
| 259 | |
| 260 | return RDMACM_MUX_ERR_CODE_OK; |
| 261 | } |
| 262 | |
| 263 | static void hash_tbl_save_fd_comm_id_pair(int fd, uint32_t comm_id, |
| 264 | uint64_t gid_ifid) |
| 265 | { |
| 266 | CommId2FdEntry fde = {fd, COMMID_TTL, gid_ifid}; |
| 267 | |
| 268 | pthread_rwlock_wrlock(&server.lock); |
| 269 | g_hash_table_insert(server.umad_agent.commid2fd, |
| 270 | g_memdup(&comm_id, sizeof(comm_id)), |
| 271 | g_memdup(&fde, sizeof(fde))); |
| 272 | pthread_rwlock_unlock(&server.lock); |
| 273 | } |
| 274 | |
| 275 | static gboolean remove_old_comm_ids(gpointer key, gpointer value, |
| 276 | gpointer user_data) |
| 277 | { |
| 278 | CommId2FdEntry *fde = (CommId2FdEntry *)value; |
| 279 | |
| 280 | return !fde->ttl--; |
| 281 | } |
| 282 | |
| 283 | static gboolean remove_entry_from_gid2fd(gpointer key, gpointer value, |
| 284 | gpointer user_data) |
| 285 | { |
| 286 | if (*(int *)value == *(int *)user_data) { |
| 287 | syslog(LOG_INFO, "0x%lx unregistered on socket %d", |
| 288 | be64toh(*(uint64_t *)key), *(int *)value); |
| 289 | return true; |
| 290 | } |
| 291 | |
| 292 | return false; |
| 293 | } |
| 294 | |
| 295 | static void hash_tbl_remove_fd_ifid_pair(int fd) |
| 296 | { |
| 297 | pthread_rwlock_wrlock(&server.lock); |
| 298 | g_hash_table_foreach_remove(server.umad_agent.gid2fd, |
| 299 | remove_entry_from_gid2fd, (gpointer)&fd); |
| 300 | pthread_rwlock_unlock(&server.lock); |
| 301 | } |
| 302 | |
Yuval Shaia | ade0075 | 2019-02-12 13:23:47 +0200 | [diff] [blame] | 303 | static int get_fd(const char *mad, int umad_len, int *fd, __be64 *gid_ifid) |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 304 | { |
| 305 | struct umad_hdr *hdr = (struct umad_hdr *)mad; |
| 306 | char *data = (char *)hdr + sizeof(*hdr); |
| 307 | int32_t comm_id = 0; |
| 308 | uint16_t attr_id = be16toh(hdr->attr_id); |
| 309 | int rc = 0; |
| 310 | |
Yuval Shaia | ade0075 | 2019-02-12 13:23:47 +0200 | [diff] [blame] | 311 | if (umad_len <= sizeof(*hdr)) { |
| 312 | rc = -EINVAL; |
| 313 | syslog(LOG_DEBUG, "Ignoring MAD packets with header only\n"); |
| 314 | goto out; |
| 315 | } |
| 316 | |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 317 | switch (attr_id) { |
| 318 | case UMAD_CM_ATTR_REQ: |
Yuval Shaia | ade0075 | 2019-02-12 13:23:47 +0200 | [diff] [blame] | 319 | if (unlikely(umad_len < sizeof(*hdr) + CM_REQ_DGID_POS + |
| 320 | sizeof(*gid_ifid))) { |
| 321 | rc = -EINVAL; |
| 322 | syslog(LOG_WARNING, |
| 323 | "Invalid MAD packet size (%d) for attr_id 0x%x\n", umad_len, |
| 324 | attr_id); |
| 325 | goto out; |
| 326 | } |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 327 | memcpy(gid_ifid, data + CM_REQ_DGID_POS, sizeof(*gid_ifid)); |
| 328 | rc = hash_tbl_search_fd_by_ifid(fd, gid_ifid); |
| 329 | break; |
| 330 | |
| 331 | case UMAD_CM_ATTR_SIDR_REQ: |
Yuval Shaia | ade0075 | 2019-02-12 13:23:47 +0200 | [diff] [blame] | 332 | if (unlikely(umad_len < sizeof(*hdr) + CM_SIDR_REQ_DGID_POS + |
| 333 | sizeof(*gid_ifid))) { |
| 334 | rc = -EINVAL; |
| 335 | syslog(LOG_WARNING, |
| 336 | "Invalid MAD packet size (%d) for attr_id 0x%x\n", umad_len, |
| 337 | attr_id); |
| 338 | goto out; |
| 339 | } |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 340 | memcpy(gid_ifid, data + CM_SIDR_REQ_DGID_POS, sizeof(*gid_ifid)); |
| 341 | rc = hash_tbl_search_fd_by_ifid(fd, gid_ifid); |
| 342 | break; |
| 343 | |
| 344 | case UMAD_CM_ATTR_REP: |
| 345 | /* Fall through */ |
| 346 | case UMAD_CM_ATTR_REJ: |
| 347 | /* Fall through */ |
| 348 | case UMAD_CM_ATTR_DREQ: |
| 349 | /* Fall through */ |
| 350 | case UMAD_CM_ATTR_DREP: |
| 351 | /* Fall through */ |
| 352 | case UMAD_CM_ATTR_RTU: |
| 353 | data += sizeof(comm_id); |
| 354 | /* Fall through */ |
| 355 | case UMAD_CM_ATTR_SIDR_REP: |
Yuval Shaia | ade0075 | 2019-02-12 13:23:47 +0200 | [diff] [blame] | 356 | if (unlikely(umad_len < sizeof(*hdr) + sizeof(comm_id))) { |
| 357 | rc = -EINVAL; |
| 358 | syslog(LOG_WARNING, |
| 359 | "Invalid MAD packet size (%d) for attr_id 0x%x\n", umad_len, |
| 360 | attr_id); |
| 361 | goto out; |
| 362 | } |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 363 | memcpy(&comm_id, data, sizeof(comm_id)); |
| 364 | if (comm_id) { |
| 365 | rc = hash_tbl_search_fd_by_comm_id(comm_id, fd, gid_ifid); |
| 366 | } |
| 367 | break; |
| 368 | |
| 369 | default: |
| 370 | rc = -EINVAL; |
| 371 | syslog(LOG_WARNING, "Unsupported attr_id 0x%x\n", attr_id); |
| 372 | } |
| 373 | |
| 374 | syslog(LOG_DEBUG, "mad_to_vm: %d 0x%x 0x%x\n", *fd, attr_id, comm_id); |
| 375 | |
Yuval Shaia | ade0075 | 2019-02-12 13:23:47 +0200 | [diff] [blame] | 376 | out: |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 377 | return rc; |
| 378 | } |
| 379 | |
| 380 | static void *umad_recv_thread_func(void *args) |
| 381 | { |
| 382 | int rc; |
Marcel Apfelbaum | 555b3d6 | 2019-01-18 14:46:13 +0200 | [diff] [blame] | 383 | RdmaCmMuxMsg msg = {}; |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 384 | int fd = -2; |
| 385 | |
| 386 | msg.hdr.msg_type = RDMACM_MUX_MSG_TYPE_REQ; |
| 387 | msg.hdr.op_code = RDMACM_MUX_OP_CODE_MAD; |
| 388 | |
| 389 | while (server.run) { |
| 390 | do { |
| 391 | msg.umad_len = sizeof(msg.umad.mad); |
| 392 | rc = umad_recv(server.umad_agent.port_id, &msg.umad, &msg.umad_len, |
| 393 | SLEEP_SECS * SCALE_US); |
| 394 | if ((rc == -EIO) || (rc == -EINVAL)) { |
| 395 | syslog(LOG_CRIT, "Fatal error while trying to read MAD"); |
| 396 | } |
| 397 | |
| 398 | if (rc == -ETIMEDOUT) { |
| 399 | g_hash_table_foreach_remove(server.umad_agent.commid2fd, |
| 400 | remove_old_comm_ids, NULL); |
| 401 | } |
| 402 | } while (rc && server.run); |
| 403 | |
| 404 | if (server.run) { |
Yuval Shaia | ade0075 | 2019-02-12 13:23:47 +0200 | [diff] [blame] | 405 | rc = get_fd(msg.umad.mad, msg.umad_len, &fd, |
| 406 | &msg.hdr.sgid.global.interface_id); |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 407 | if (rc) { |
| 408 | continue; |
| 409 | } |
| 410 | |
| 411 | send(fd, &msg, sizeof(msg), 0); |
| 412 | } |
| 413 | } |
| 414 | |
| 415 | return NULL; |
| 416 | } |
| 417 | |
| 418 | static int read_and_process(int fd) |
| 419 | { |
| 420 | int rc; |
Marcel Apfelbaum | 555b3d6 | 2019-01-18 14:46:13 +0200 | [diff] [blame] | 421 | RdmaCmMuxMsg msg = {}; |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 422 | struct umad_hdr *hdr; |
| 423 | uint32_t *comm_id = 0; |
| 424 | uint16_t attr_id; |
| 425 | |
| 426 | rc = recv(fd, &msg, sizeof(msg), 0); |
| 427 | syslog(LOG_DEBUG, "Socket %d, recv %d\n", fd, rc); |
| 428 | |
| 429 | if (rc < 0 && errno != EWOULDBLOCK) { |
| 430 | syslog(LOG_ERR, "Fail to read from socket %d\n", fd); |
| 431 | return -EIO; |
| 432 | } |
| 433 | |
| 434 | if (!rc) { |
| 435 | syslog(LOG_ERR, "Fail to read from socket %d\n", fd); |
| 436 | return -EPIPE; |
| 437 | } |
| 438 | |
| 439 | if (msg.hdr.msg_type != RDMACM_MUX_MSG_TYPE_REQ) { |
| 440 | syslog(LOG_WARNING, "Got non-request message (%d) from socket %d\n", |
| 441 | msg.hdr.msg_type, fd); |
| 442 | return -EPERM; |
| 443 | } |
| 444 | |
| 445 | switch (msg.hdr.op_code) { |
| 446 | case RDMACM_MUX_OP_CODE_REG: |
| 447 | rc = add_fd_ifid_pair(fd, msg.hdr.sgid.global.interface_id); |
| 448 | break; |
| 449 | |
| 450 | case RDMACM_MUX_OP_CODE_UNREG: |
| 451 | rc = delete_fd_ifid_pair(fd, msg.hdr.sgid.global.interface_id); |
| 452 | break; |
| 453 | |
| 454 | case RDMACM_MUX_OP_CODE_MAD: |
| 455 | /* If this is REQ or REP then store the pair comm_id,fd to be later |
| 456 | * used for other messages where gid is unknown */ |
| 457 | hdr = (struct umad_hdr *)msg.umad.mad; |
| 458 | attr_id = be16toh(hdr->attr_id); |
| 459 | if ((attr_id == UMAD_CM_ATTR_REQ) || (attr_id == UMAD_CM_ATTR_DREQ) || |
| 460 | (attr_id == UMAD_CM_ATTR_SIDR_REQ) || |
| 461 | (attr_id == UMAD_CM_ATTR_REP) || (attr_id == UMAD_CM_ATTR_DREP)) { |
| 462 | comm_id = (uint32_t *)(msg.umad.mad + sizeof(*hdr)); |
| 463 | hash_tbl_save_fd_comm_id_pair(fd, *comm_id, |
| 464 | msg.hdr.sgid.global.interface_id); |
| 465 | } |
| 466 | |
| 467 | syslog(LOG_DEBUG, "vm_to_mad: %d 0x%x 0x%x\n", fd, attr_id, |
| 468 | comm_id ? *comm_id : 0); |
| 469 | rc = umad_send(server.umad_agent.port_id, server.umad_agent.agent_id, |
| 470 | &msg.umad, msg.umad_len, 1, 0); |
| 471 | if (rc) { |
| 472 | syslog(LOG_ERR, |
| 473 | "Fail to send MAD message (0x%x) from socket %d, err=%d", |
| 474 | attr_id, fd, rc); |
| 475 | } |
| 476 | break; |
| 477 | |
| 478 | default: |
| 479 | syslog(LOG_ERR, "Got invalid op_code (%d) from socket %d", |
| 480 | msg.hdr.msg_type, fd); |
| 481 | rc = RDMACM_MUX_ERR_CODE_EINVAL; |
| 482 | } |
| 483 | |
| 484 | msg.hdr.msg_type = RDMACM_MUX_MSG_TYPE_RESP; |
| 485 | msg.hdr.err_code = rc; |
| 486 | rc = send(fd, &msg, sizeof(msg), 0); |
| 487 | |
| 488 | return rc == sizeof(msg) ? 0 : -EPIPE; |
| 489 | } |
| 490 | |
| 491 | static int accept_all(void) |
| 492 | { |
Philippe Mathieu-Daudé | 1e1f6ab | 2020-02-18 10:44:02 +0100 | [diff] [blame] | 493 | int fd, rc = 0; |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 494 | |
| 495 | pthread_rwlock_wrlock(&server.lock); |
| 496 | |
| 497 | do { |
| 498 | if ((server.nfds + 1) > MAX_CLIENTS) { |
| 499 | syslog(LOG_WARNING, "Too many clients (%d)", server.nfds); |
| 500 | rc = -EIO; |
| 501 | goto out; |
| 502 | } |
| 503 | |
| 504 | fd = accept(server.fds[0].fd, NULL, NULL); |
| 505 | if (fd < 0) { |
| 506 | if (errno != EWOULDBLOCK) { |
| 507 | syslog(LOG_WARNING, "accept() failed"); |
| 508 | rc = -EIO; |
| 509 | goto out; |
| 510 | } |
| 511 | break; |
| 512 | } |
| 513 | |
| 514 | syslog(LOG_INFO, "Client connected on socket %d\n", fd); |
| 515 | server.fds[server.nfds].fd = fd; |
| 516 | server.fds[server.nfds].events = POLLIN; |
| 517 | server.nfds++; |
| 518 | } while (fd != -1); |
| 519 | |
| 520 | out: |
| 521 | pthread_rwlock_unlock(&server.lock); |
| 522 | return rc; |
| 523 | } |
| 524 | |
| 525 | static void compress_fds(void) |
| 526 | { |
| 527 | int i, j; |
| 528 | int closed = 0; |
| 529 | |
| 530 | pthread_rwlock_wrlock(&server.lock); |
| 531 | |
| 532 | for (i = 1; i < server.nfds; i++) { |
| 533 | if (!server.fds[i].fd) { |
| 534 | closed++; |
| 535 | for (j = i; j < server.nfds - 1; j++) { |
| 536 | server.fds[j] = server.fds[j + 1]; |
| 537 | } |
| 538 | } |
| 539 | } |
| 540 | |
| 541 | server.nfds -= closed; |
| 542 | |
| 543 | pthread_rwlock_unlock(&server.lock); |
| 544 | } |
| 545 | |
| 546 | static void close_fd(int idx) |
| 547 | { |
| 548 | close(server.fds[idx].fd); |
| 549 | syslog(LOG_INFO, "Socket %d closed\n", server.fds[idx].fd); |
| 550 | hash_tbl_remove_fd_ifid_pair(server.fds[idx].fd); |
| 551 | server.fds[idx].fd = 0; |
| 552 | } |
| 553 | |
| 554 | static void run(void) |
| 555 | { |
| 556 | int rc, nfds, i; |
| 557 | bool compress = false; |
| 558 | |
| 559 | syslog(LOG_INFO, "Service started"); |
| 560 | |
| 561 | while (server.run) { |
| 562 | rc = poll(server.fds, server.nfds, SLEEP_SECS * SCALE_US); |
| 563 | if (rc < 0) { |
| 564 | if (errno != EINTR) { |
| 565 | syslog(LOG_WARNING, "poll() failed"); |
| 566 | } |
| 567 | continue; |
| 568 | } |
| 569 | |
| 570 | if (rc == 0) { |
| 571 | continue; |
| 572 | } |
| 573 | |
| 574 | nfds = server.nfds; |
| 575 | for (i = 0; i < nfds; i++) { |
| 576 | syslog(LOG_DEBUG, "pollfd[%d]: revents 0x%x, events 0x%x\n", i, |
| 577 | server.fds[i].revents, server.fds[i].events); |
| 578 | if (server.fds[i].revents == 0) { |
| 579 | continue; |
| 580 | } |
| 581 | |
| 582 | if (server.fds[i].revents != POLLIN) { |
| 583 | if (i == 0) { |
| 584 | syslog(LOG_NOTICE, "Unexpected poll() event (0x%x)\n", |
| 585 | server.fds[i].revents); |
| 586 | } else { |
| 587 | close_fd(i); |
| 588 | compress = true; |
| 589 | } |
| 590 | continue; |
| 591 | } |
| 592 | |
| 593 | if (i == 0) { |
| 594 | rc = accept_all(); |
| 595 | if (rc) { |
| 596 | continue; |
| 597 | } |
| 598 | } else { |
| 599 | rc = read_and_process(server.fds[i].fd); |
| 600 | if (rc) { |
| 601 | close_fd(i); |
| 602 | compress = true; |
| 603 | } |
| 604 | } |
| 605 | } |
| 606 | |
| 607 | if (compress) { |
| 608 | compress = false; |
| 609 | compress_fds(); |
| 610 | } |
| 611 | } |
| 612 | } |
| 613 | |
| 614 | static void fini_listener(void) |
| 615 | { |
| 616 | int i; |
| 617 | |
| 618 | if (server.fds[0].fd <= 0) { |
| 619 | return; |
| 620 | } |
| 621 | |
| 622 | for (i = server.nfds - 1; i >= 0; i--) { |
| 623 | if (server.fds[i].fd) { |
| 624 | close(server.fds[i].fd); |
| 625 | } |
| 626 | } |
| 627 | |
| 628 | unlink(server.args.unix_socket_path); |
| 629 | } |
| 630 | |
| 631 | static void fini_umad(void) |
| 632 | { |
| 633 | if (server.umad_agent.agent_id) { |
| 634 | umad_unregister(server.umad_agent.port_id, server.umad_agent.agent_id); |
| 635 | } |
| 636 | |
| 637 | if (server.umad_agent.port_id) { |
| 638 | umad_close_port(server.umad_agent.port_id); |
| 639 | } |
| 640 | |
| 641 | hash_tbl_free(); |
| 642 | } |
| 643 | |
| 644 | static void fini(void) |
| 645 | { |
| 646 | if (server.umad_recv_thread) { |
| 647 | pthread_join(server.umad_recv_thread, NULL); |
| 648 | server.umad_recv_thread = 0; |
| 649 | } |
| 650 | fini_umad(); |
| 651 | fini_listener(); |
| 652 | pthread_rwlock_destroy(&server.lock); |
| 653 | |
| 654 | syslog(LOG_INFO, "Service going down"); |
| 655 | } |
| 656 | |
| 657 | static int init_listener(void) |
| 658 | { |
| 659 | struct sockaddr_un sun; |
| 660 | int rc, on = 1; |
| 661 | |
| 662 | server.fds[0].fd = socket(AF_UNIX, SOCK_STREAM, 0); |
| 663 | if (server.fds[0].fd < 0) { |
| 664 | syslog(LOG_ALERT, "socket() failed"); |
| 665 | return -EIO; |
| 666 | } |
| 667 | |
| 668 | rc = setsockopt(server.fds[0].fd, SOL_SOCKET, SO_REUSEADDR, (char *)&on, |
| 669 | sizeof(on)); |
| 670 | if (rc < 0) { |
| 671 | syslog(LOG_ALERT, "setsockopt() failed"); |
| 672 | rc = -EIO; |
| 673 | goto err; |
| 674 | } |
| 675 | |
| 676 | rc = ioctl(server.fds[0].fd, FIONBIO, (char *)&on); |
| 677 | if (rc < 0) { |
| 678 | syslog(LOG_ALERT, "ioctl() failed"); |
| 679 | rc = -EIO; |
| 680 | goto err; |
| 681 | } |
| 682 | |
| 683 | if (strlen(server.args.unix_socket_path) >= sizeof(sun.sun_path)) { |
| 684 | syslog(LOG_ALERT, |
| 685 | "Invalid unix_socket_path, size must be less than %ld\n", |
| 686 | sizeof(sun.sun_path)); |
| 687 | rc = -EINVAL; |
| 688 | goto err; |
| 689 | } |
| 690 | |
| 691 | sun.sun_family = AF_UNIX; |
| 692 | rc = snprintf(sun.sun_path, sizeof(sun.sun_path), "%s", |
| 693 | server.args.unix_socket_path); |
| 694 | if (rc < 0 || rc >= sizeof(sun.sun_path)) { |
| 695 | syslog(LOG_ALERT, "Could not copy unix socket path\n"); |
| 696 | rc = -EINVAL; |
| 697 | goto err; |
| 698 | } |
| 699 | |
| 700 | rc = bind(server.fds[0].fd, (struct sockaddr *)&sun, sizeof(sun)); |
| 701 | if (rc < 0) { |
| 702 | syslog(LOG_ALERT, "bind() failed"); |
| 703 | rc = -EIO; |
| 704 | goto err; |
| 705 | } |
| 706 | |
| 707 | rc = listen(server.fds[0].fd, SERVER_LISTEN_BACKLOG); |
| 708 | if (rc < 0) { |
| 709 | syslog(LOG_ALERT, "listen() failed"); |
| 710 | rc = -EIO; |
| 711 | goto err; |
| 712 | } |
| 713 | |
| 714 | server.fds[0].events = POLLIN; |
| 715 | server.nfds = 1; |
| 716 | server.run = true; |
| 717 | |
| 718 | return 0; |
| 719 | |
| 720 | err: |
| 721 | close(server.fds[0].fd); |
| 722 | return rc; |
| 723 | } |
| 724 | |
| 725 | static int init_umad(void) |
| 726 | { |
| 727 | long method_mask[IB_USER_MAD_LONGS_PER_METHOD_MASK]; |
| 728 | |
| 729 | server.umad_agent.port_id = umad_open_port(server.args.rdma_dev_name, |
| 730 | server.args.rdma_port_num); |
| 731 | |
| 732 | if (server.umad_agent.port_id < 0) { |
| 733 | syslog(LOG_WARNING, "umad_open_port() failed"); |
| 734 | return -EIO; |
| 735 | } |
| 736 | |
| 737 | memset(&method_mask, 0, sizeof(method_mask)); |
| 738 | method_mask[0] = MAD_METHOD_MASK0; |
| 739 | server.umad_agent.agent_id = umad_register(server.umad_agent.port_id, |
| 740 | UMAD_CLASS_CM, |
| 741 | UMAD_SA_CLASS_VERSION, |
| 742 | MAD_RMPP_VERSION, method_mask); |
| 743 | if (server.umad_agent.agent_id < 0) { |
| 744 | syslog(LOG_WARNING, "umad_register() failed"); |
| 745 | return -EIO; |
| 746 | } |
| 747 | |
| 748 | hash_tbl_alloc(); |
| 749 | |
| 750 | return 0; |
| 751 | } |
| 752 | |
| 753 | static void signal_handler(int sig, siginfo_t *siginfo, void *context) |
| 754 | { |
| 755 | static bool warned; |
| 756 | |
| 757 | /* Prevent stop if clients are connected */ |
| 758 | if (server.nfds != 1) { |
| 759 | if (!warned) { |
| 760 | syslog(LOG_WARNING, |
| 761 | "Can't stop while active client exist, resend SIGINT to overid"); |
| 762 | warned = true; |
| 763 | return; |
| 764 | } |
| 765 | } |
| 766 | |
| 767 | if (sig == SIGINT) { |
| 768 | server.run = false; |
| 769 | fini(); |
| 770 | } |
| 771 | |
| 772 | exit(0); |
| 773 | } |
| 774 | |
| 775 | static int init(void) |
| 776 | { |
| 777 | int rc; |
Marcel Apfelbaum | 555b3d6 | 2019-01-18 14:46:13 +0200 | [diff] [blame] | 778 | struct sigaction sig = {}; |
Yuval Shaia | a5d2f6f | 2018-12-21 16:40:15 +0200 | [diff] [blame] | 779 | |
| 780 | rc = init_listener(); |
| 781 | if (rc) { |
| 782 | return rc; |
| 783 | } |
| 784 | |
| 785 | rc = init_umad(); |
| 786 | if (rc) { |
| 787 | return rc; |
| 788 | } |
| 789 | |
| 790 | pthread_rwlock_init(&server.lock, 0); |
| 791 | |
| 792 | rc = pthread_create(&server.umad_recv_thread, NULL, umad_recv_thread_func, |
| 793 | NULL); |
| 794 | if (rc) { |
| 795 | syslog(LOG_ERR, "Fail to create UMAD receiver thread (%d)\n", rc); |
| 796 | return rc; |
| 797 | } |
| 798 | |
| 799 | sig.sa_sigaction = &signal_handler; |
| 800 | sig.sa_flags = SA_SIGINFO; |
| 801 | rc = sigaction(SIGINT, &sig, NULL); |
| 802 | if (rc < 0) { |
| 803 | syslog(LOG_ERR, "Fail to install SIGINT handler (%d)\n", errno); |
| 804 | return rc; |
| 805 | } |
| 806 | |
| 807 | return 0; |
| 808 | } |
| 809 | |
| 810 | int main(int argc, char *argv[]) |
| 811 | { |
| 812 | int rc; |
| 813 | |
| 814 | memset(&server, 0, sizeof(server)); |
| 815 | |
| 816 | parse_args(argc, argv); |
| 817 | |
| 818 | rc = init(); |
| 819 | if (rc) { |
| 820 | syslog(LOG_ERR, "Fail to initialize server (%d)\n", rc); |
| 821 | rc = -EAGAIN; |
| 822 | goto out; |
| 823 | } |
| 824 | |
| 825 | run(); |
| 826 | |
| 827 | out: |
| 828 | fini(); |
| 829 | |
| 830 | return rc; |
| 831 | } |