drivers/staging/lustre/lnet/lnet/lib-socket.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * GPL HEADER START
   4  *
   5  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 only,
   9  * as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License version 2 for more details (a copy is included
  15  * in the LICENSE file that accompanied this code).
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * version 2 along with this program; If not, see
  19  * http://www.gnu.org/licenses/gpl-2.0.html
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  25  * Use is subject to license terms.
  26  *
  27  * Copyright (c) 2012, 2015, Intel Corporation.
  28  */
  29 /*
  30  * This file is part of Lustre, http://www.lustre.org/
  31  * Lustre is a trademark of Seagate, Inc.
  32  */
  33 #define DEBUG_SUBSYSTEM S_LNET
  34
  35 #include <linux/if.h>
  36 #include <linux/in.h>
  37 #include <linux/net.h>
  38 #include <linux/file.h>
  39 #include <linux/pagemap.h>
  40 /* For sys_open & sys_close */
  41 #include <linux/syscalls.h>
  42 #include <net/sock.h>
  43
  44 #include <linux/libcfs/libcfs.h>
  45 #include <linux/lnet/lib-lnet.h>
  46
  47 static int
  48 kernel_sock_unlocked_ioctl(struct file *filp, int cmd, unsigned long arg)
  49 {
  50         mm_segment_t oldfs = get_fs();
  51         int err;
  52
  53         set_fs(KERNEL_DS);
  54         err = filp->f_op->unlocked_ioctl(filp, cmd, arg);
  55         set_fs(oldfs);
  56
  57         return err;
  58 }
  59
  60 static int
  61 lnet_sock_ioctl(int cmd, unsigned long arg)
  62 {
  63         struct file *sock_filp;
  64         struct socket *sock;
  65         int rc;
  66
  67         rc = sock_create(PF_INET, SOCK_STREAM, 0, &sock);
  68         if (rc) {
  69                 CERROR("Can't create socket: %d\n", rc);
  70                 return rc;
  71         }
  72
  73         sock_filp = sock_alloc_file(sock, 0, NULL);
  74         if (IS_ERR(sock_filp))
  75                 return PTR_ERR(sock_filp);
  76
  77         rc = kernel_sock_unlocked_ioctl(sock_filp, cmd, arg);
  78
  79         fput(sock_filp);
  80         return rc;
  81 }
  82
  83 int
  84 lnet_ipif_query(char *name, int *up, __u32 *ip, __u32 *mask)
  85 {
  86         struct ifreq ifr;
  87         int nob;
  88         int rc;
  89         __be32 val;
  90
  91         nob = strnlen(name, IFNAMSIZ);
  92         if (nob == IFNAMSIZ) {
  93                 CERROR("Interface name %s too long\n", name);
  94                 return -EINVAL;
  95         }
  96
  97         BUILD_BUG_ON(sizeof(ifr.ifr_name) < IFNAMSIZ);
  98
  99         if (strlen(name) > sizeof(ifr.ifr_name) - 1)
 100                 return -E2BIG;
 101         strncpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
 102
 103         rc = lnet_sock_ioctl(SIOCGIFFLAGS, (unsigned long)&ifr);
 104         if (rc) {
 105                 CERROR("Can't get flags for interface %s\n", name);
 106                 return rc;
 107         }
 108
 109         if (!(ifr.ifr_flags & IFF_UP)) {
 110                 CDEBUG(D_NET, "Interface %s down\n", name);
 111                 *up = 0;
 112                 *ip = *mask = 0;
 113                 return 0;
 114         }
 115         *up = 1;
 116
 117         if (strlen(name) > sizeof(ifr.ifr_name) - 1)
 118                 return -E2BIG;
 119         strncpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
 120
 121         ifr.ifr_addr.sa_family = AF_INET;
 122         rc = lnet_sock_ioctl(SIOCGIFADDR, (unsigned long)&ifr);
 123         if (rc) {
 124                 CERROR("Can't get IP address for interface %s\n", name);
 125                 return rc;
 126         }
 127
 128         val = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr.s_addr;
 129         *ip = ntohl(val);
 130
 131         if (strlen(name) > sizeof(ifr.ifr_name) - 1)
 132                 return -E2BIG;
 133         strncpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
 134
 135         ifr.ifr_addr.sa_family = AF_INET;
 136         rc = lnet_sock_ioctl(SIOCGIFNETMASK, (unsigned long)&ifr);
 137         if (rc) {
 138                 CERROR("Can't get netmask for interface %s\n", name);
 139                 return rc;
 140         }
 141
 142         val = ((struct sockaddr_in *)&ifr.ifr_netmask)->sin_addr.s_addr;
 143         *mask = ntohl(val);
 144
 145         return 0;
 146 }
 147 EXPORT_SYMBOL(lnet_ipif_query);
 148
 149 int
 150 lnet_ipif_enumerate(char ***namesp)
 151 {
 152         /* Allocate and fill in 'names', returning # interfaces/error */
 153         char **names;
 154         int toobig;
 155         int nalloc;
 156         int nfound;
 157         struct ifreq *ifr;
 158         struct ifconf ifc;
 159         int rc;
 160         int nob;
 161         int i;
 162
 163         nalloc = 16;    /* first guess at max interfaces */
 164         toobig = 0;
 165         for (;;) {
 166                 if (nalloc * sizeof(*ifr) > PAGE_SIZE) {
 167                         toobig = 1;
 168                         nalloc = PAGE_SIZE / sizeof(*ifr);
 169                         CWARN("Too many interfaces: only enumerating first %d\n",
 170                               nalloc);
 171                 }
 172
 173                 LIBCFS_ALLOC(ifr, nalloc * sizeof(*ifr));
 174                 if (!ifr) {
 175                         CERROR("ENOMEM enumerating up to %d interfaces\n",
 176                                nalloc);
 177                         rc = -ENOMEM;
 178                         goto out0;
 179                 }
 180
 181                 ifc.ifc_buf = (char *)ifr;
 182                 ifc.ifc_len = nalloc * sizeof(*ifr);
 183
 184                 rc = lnet_sock_ioctl(SIOCGIFCONF, (unsigned long)&ifc);
 185                 if (rc < 0) {
 186                         CERROR("Error %d enumerating interfaces\n", rc);
 187                         goto out1;
 188                 }
 189
 190                 LASSERT(!rc);
 191
 192                 nfound = ifc.ifc_len / sizeof(*ifr);
 193                 LASSERT(nfound <= nalloc);
 194
 195                 if (nfound < nalloc || toobig)
 196                         break;
 197
 198                 LIBCFS_FREE(ifr, nalloc * sizeof(*ifr));
 199                 nalloc *= 2;
 200         }
 201
 202         if (!nfound)
 203                 goto out1;
 204
 205         LIBCFS_ALLOC(names, nfound * sizeof(*names));
 206         if (!names) {
 207                 rc = -ENOMEM;
 208                 goto out1;
 209         }
 210
 211         for (i = 0; i < nfound; i++) {
 212                 nob = strnlen(ifr[i].ifr_name, IFNAMSIZ);
 213                 if (nob == IFNAMSIZ) {
 214                         /* no space for terminating NULL */
 215                         CERROR("interface name %.*s too long (%d max)\n",
 216                                nob, ifr[i].ifr_name, IFNAMSIZ);
 217                         rc = -ENAMETOOLONG;
 218                         goto out2;
 219                 }
 220
 221                 LIBCFS_ALLOC(names[i], IFNAMSIZ);
 222                 if (!names[i]) {
 223                         rc = -ENOMEM;
 224                         goto out2;
 225                 }
 226
 227                 memcpy(names[i], ifr[i].ifr_name, nob);
 228                 names[i][nob] = 0;
 229         }
 230
 231         *namesp = names;
 232         rc = nfound;
 233
 234 out2:
 235         if (rc < 0)
 236                 lnet_ipif_free_enumeration(names, nfound);
 237 out1:
 238         LIBCFS_FREE(ifr, nalloc * sizeof(*ifr));
 239 out0:
 240         return rc;
 241 }
 242 EXPORT_SYMBOL(lnet_ipif_enumerate);
 243
 244 void
 245 lnet_ipif_free_enumeration(char **names, int n)
 246 {
 247         int i;
 248
 249         LASSERT(n > 0);
 250
 251         for (i = 0; i < n && names[i]; i++)
 252                 LIBCFS_FREE(names[i], IFNAMSIZ);
 253
 254         LIBCFS_FREE(names, n * sizeof(*names));
 255 }
 256 EXPORT_SYMBOL(lnet_ipif_free_enumeration);
 257
 258 int
 259 lnet_sock_write(struct socket *sock, void *buffer, int nob, int timeout)
 260 {
 261         int rc;
 262         long jiffies_left = timeout * msecs_to_jiffies(MSEC_PER_SEC);
 263         unsigned long then;
 264         struct timeval tv;
 265         struct kvec  iov = { .iov_base = buffer, .iov_len  = nob };
 266         struct msghdr msg = {NULL,};
 267
 268         LASSERT(nob > 0);
 269         /*
 270          * Caller may pass a zero timeout if she thinks the socket buffer is
 271          * empty enough to take the whole message immediately
 272          */
 273         iov_iter_kvec(&msg.msg_iter, WRITE | ITER_KVEC, &iov, 1, nob);
 274         for (;;) {
 275                 msg.msg_flags = !timeout ? MSG_DONTWAIT : 0;
 276                 if (timeout) {
 277                         /* Set send timeout to remaining time */
 278                         jiffies_to_timeval(jiffies_left, &tv);
 279                         rc = kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO,
 280                                                (char *)&tv, sizeof(tv));
 281                         if (rc) {
 282                                 CERROR("Can't set socket send timeout %ld.%06d: %d\n",
 283                                        (long)tv.tv_sec, (int)tv.tv_usec, rc);
 284                                 return rc;
 285                         }
 286                 }
 287
 288                 then = jiffies;
 289                 rc = kernel_sendmsg(sock, &msg, &iov, 1, nob);
 290                 jiffies_left -= jiffies - then;
 291
 292                 if (rc < 0)
 293                         return rc;
 294
 295                 if (!rc) {
 296                         CERROR("Unexpected zero rc\n");
 297                         return -ECONNABORTED;
 298                 }
 299
 300                 if (!msg_data_left(&msg))
 301                         break;
 302
 303                 if (jiffies_left <= 0)
 304                         return -EAGAIN;
 305         }
 306         return 0;
 307 }
 308 EXPORT_SYMBOL(lnet_sock_write);
 309
 310 int
 311 lnet_sock_read(struct socket *sock, void *buffer, int nob, int timeout)
 312 {
 313         int rc;
 314         long jiffies_left = timeout * msecs_to_jiffies(MSEC_PER_SEC);
 315         unsigned long then;
 316         struct timeval tv;
 317
 318         LASSERT(nob > 0);
 319         LASSERT(jiffies_left > 0);
 320
 321         for (;;) {
 322                 struct kvec  iov = {
 323                         .iov_base = buffer,
 324                         .iov_len  = nob
 325                 };
 326                 struct msghdr msg = {
 327                         .msg_flags = 0
 328                 };
 329
 330                 /* Set receive timeout to remaining time */
 331                 jiffies_to_timeval(jiffies_left, &tv);
 332                 rc = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO,
 333                                        (char *)&tv, sizeof(tv));
 334                 if (rc) {
 335                         CERROR("Can't set socket recv timeout %ld.%06d: %d\n",
 336                                (long)tv.tv_sec, (int)tv.tv_usec, rc);
 337                         return rc;
 338                 }
 339
 340                 then = jiffies;
 341                 rc = kernel_recvmsg(sock, &msg, &iov, 1, nob, 0);
 342                 jiffies_left -= jiffies - then;
 343
 344                 if (rc < 0)
 345                         return rc;
 346
 347                 if (!rc)
 348                         return -ECONNRESET;
 349
 350                 buffer = ((char *)buffer) + rc;
 351                 nob -= rc;
 352
 353                 if (!nob)
 354                         return 0;
 355
 356                 if (jiffies_left <= 0)
 357                         return -ETIMEDOUT;
 358         }
 359 }
 360 EXPORT_SYMBOL(lnet_sock_read);
 361
 362 static int
 363 lnet_sock_create(struct socket **sockp, int *fatal, __u32 local_ip,
 364                  int local_port)
 365 {
 366         struct sockaddr_in locaddr;
 367         struct socket *sock;
 368         int rc;
 369         int option;
 370
 371         /* All errors are fatal except bind failure if the port is in use */
 372         *fatal = 1;
 373
 374         rc = sock_create(PF_INET, SOCK_STREAM, 0, &sock);
 375         *sockp = sock;
 376         if (rc) {
 377                 CERROR("Can't create socket: %d\n", rc);
 378                 return rc;
 379         }
 380
 381         option = 1;
 382         rc = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
 383                                (char *)&option, sizeof(option));
 384         if (rc) {
 385                 CERROR("Can't set SO_REUSEADDR for socket: %d\n", rc);
 386                 goto failed;
 387         }
 388
 389         if (local_ip || local_port) {
 390                 memset(&locaddr, 0, sizeof(locaddr));
 391                 locaddr.sin_family = AF_INET;
 392                 locaddr.sin_port = htons(local_port);
 393                 if (!local_ip)
 394                         locaddr.sin_addr.s_addr = htonl(INADDR_ANY);
 395                 else
 396                         locaddr.sin_addr.s_addr = htonl(local_ip);
 397
 398                 rc = kernel_bind(sock, (struct sockaddr *)&locaddr,
 399                                  sizeof(locaddr));
 400                 if (rc == -EADDRINUSE) {
 401                         CDEBUG(D_NET, "Port %d already in use\n", local_port);
 402                         *fatal = 0;
 403                         goto failed;
 404                 }
 405                 if (rc) {
 406                         CERROR("Error trying to bind to port %d: %d\n",
 407                                local_port, rc);
 408                         goto failed;
 409                 }
 410         }
 411         return 0;
 412
 413 failed:
 414         sock_release(sock);
 415         return rc;
 416 }
 417
 418 int
 419 lnet_sock_setbuf(struct socket *sock, int txbufsize, int rxbufsize)
 420 {
 421         int option;
 422         int rc;
 423
 424         if (txbufsize) {
 425                 option = txbufsize;
 426                 rc = kernel_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
 427                                        (char *)&option, sizeof(option));
 428                 if (rc) {
 429                         CERROR("Can't set send buffer %d: %d\n",
 430                                option, rc);
 431                         return rc;
 432                 }
 433         }
 434
 435         if (rxbufsize) {
 436                 option = rxbufsize;
 437                 rc = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
 438                                        (char *)&option, sizeof(option));
 439                 if (rc) {
 440                         CERROR("Can't set receive buffer %d: %d\n",
 441                                option, rc);
 442                         return rc;
 443                 }
 444         }
 445         return 0;
 446 }
 447 EXPORT_SYMBOL(lnet_sock_setbuf);
 448
 449 int
 450 lnet_sock_getaddr(struct socket *sock, bool remote, __u32 *ip, int *port)
 451 {
 452         struct sockaddr_in sin;
 453         int len = sizeof(sin);
 454         int rc;
 455
 456         if (remote)
 457                 rc = kernel_getpeername(sock, (struct sockaddr *)&sin, &len);
 458         else
 459                 rc = kernel_getsockname(sock, (struct sockaddr *)&sin, &len);
 460         if (rc) {
 461                 CERROR("Error %d getting sock %s IP/port\n",
 462                        rc, remote ? "peer" : "local");
 463                 return rc;
 464         }
 465
 466         if (ip)
 467                 *ip = ntohl(sin.sin_addr.s_addr);
 468
 469         if (port)
 470                 *port = ntohs(sin.sin_port);
 471
 472         return 0;
 473 }
 474 EXPORT_SYMBOL(lnet_sock_getaddr);
 475
 476 int
 477 lnet_sock_getbuf(struct socket *sock, int *txbufsize, int *rxbufsize)
 478 {
 479         if (txbufsize)
 480                 *txbufsize = sock->sk->sk_sndbuf;
 481
 482         if (rxbufsize)
 483                 *rxbufsize = sock->sk->sk_rcvbuf;
 484
 485         return 0;
 486 }
 487 EXPORT_SYMBOL(lnet_sock_getbuf);
 488
 489 int
 490 lnet_sock_listen(struct socket **sockp, __u32 local_ip, int local_port,
 491                  int backlog)
 492 {
 493         int fatal;
 494         int rc;
 495
 496         rc = lnet_sock_create(sockp, &fatal, local_ip, local_port);
 497         if (rc) {
 498                 if (!fatal)
 499                         CERROR("Can't create socket: port %d already in use\n",
 500                                local_port);
 501                 return rc;
 502         }
 503
 504         rc = kernel_listen(*sockp, backlog);
 505         if (!rc)
 506                 return 0;
 507
 508         CERROR("Can't set listen backlog %d: %d\n", backlog, rc);
 509         sock_release(*sockp);
 510         return rc;
 511 }
 512
 513 int
 514 lnet_sock_accept(struct socket **newsockp, struct socket *sock)
 515 {
 516         wait_queue_entry_t wait;
 517         struct socket *newsock;
 518         int rc;
 519
 520         /*
 521          * XXX this should add a ref to sock->ops->owner, if
 522          * TCP could be a module
 523          */
 524         rc = sock_create_lite(PF_PACKET, sock->type, IPPROTO_TCP, &newsock);
 525         if (rc) {
 526                 CERROR("Can't allocate socket\n");
 527                 return rc;
 528         }
 529
 530         newsock->ops = sock->ops;
 531
 532         rc = sock->ops->accept(sock, newsock, O_NONBLOCK, false);
 533         if (rc == -EAGAIN) {
 534                 /* Nothing ready, so wait for activity */
 535                 init_waitqueue_entry(&wait, current);
 536                 add_wait_queue(sk_sleep(sock->sk), &wait);
 537                 set_current_state(TASK_INTERRUPTIBLE);
 538                 schedule();
 539                 remove_wait_queue(sk_sleep(sock->sk), &wait);
 540                 rc = sock->ops->accept(sock, newsock, O_NONBLOCK, false);
 541         }
 542
 543         if (rc)
 544                 goto failed;
 545
 546         *newsockp = newsock;
 547         return 0;
 548
 549 failed:
 550         sock_release(newsock);
 551         return rc;
 552 }
 553
 554 int
 555 lnet_sock_connect(struct socket **sockp, int *fatal, __u32 local_ip,
 556                   int local_port, __u32 peer_ip, int peer_port)
 557 {
 558         struct sockaddr_in srvaddr;
 559         int rc;
 560
 561         rc = lnet_sock_create(sockp, fatal, local_ip, local_port);
 562         if (rc)
 563                 return rc;
 564
 565         memset(&srvaddr, 0, sizeof(srvaddr));
 566         srvaddr.sin_family = AF_INET;
 567         srvaddr.sin_port = htons(peer_port);
 568         srvaddr.sin_addr.s_addr = htonl(peer_ip);
 569
 570         rc = kernel_connect(*sockp, (struct sockaddr *)&srvaddr,
 571                             sizeof(srvaddr), 0);
 572         if (!rc)
 573                 return 0;
 574
 575         /*
 576          * EADDRNOTAVAIL probably means we're already connected to the same
 577          * peer/port on the same local port on a differently typed
 578          * connection.  Let our caller retry with a different local
 579          * port...
 580          */
 581         *fatal = !(rc == -EADDRNOTAVAIL);
 582
 583         CDEBUG_LIMIT(*fatal ? D_NETERROR : D_NET,
 584                      "Error %d connecting %pI4h/%d -> %pI4h/%d\n", rc,
 585                      &local_ip, local_port, &peer_ip, peer_port);
 586
 587         sock_release(*sockp);
 588         return rc;
 589 }