1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/net/sunrpc/xprtsock.c 4 * 5 * Client-side transport implementation for sockets. 6 * 7 * TCP callback races fixes (C) 1998 Red Hat 8 * TCP send fixes (C) 1998 Red Hat 9 * TCP NFS related read + write fixes 10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 11 * 12 * Rewrite of larges part of the code in order to stabilize TCP stuff. 13 * Fix behaviour when socket buffer is full. 14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> 15 * 16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> 17 * 18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005. 19 * <gilles.quillard@bull.net> 20 */ 21 22 #include <linux/types.h> 23 #include <linux/string.h> 24 #include <linux/slab.h> 25 #include <linux/module.h> 26 #include <linux/capability.h> 27 #include <linux/pagemap.h> 28 #include <linux/errno.h> 29 #include <linux/socket.h> 30 #include <linux/in.h> 31 #include <linux/net.h> 32 #include <linux/mm.h> 33 #include <linux/un.h> 34 #include <linux/udp.h> 35 #include <linux/tcp.h> 36 #include <linux/sunrpc/clnt.h> 37 #include <linux/sunrpc/addr.h> 38 #include <linux/sunrpc/sched.h> 39 #include <linux/sunrpc/svcsock.h> 40 #include <linux/sunrpc/xprtsock.h> 41 #include <linux/file.h> 42 #ifdef CONFIG_SUNRPC_BACKCHANNEL 43 #include <linux/sunrpc/bc_xprt.h> 44 #endif 45 46 #include <net/sock.h> 47 #include <net/checksum.h> 48 #include <net/udp.h> 49 #include <net/tcp.h> 50 #include <net/tls.h> 51 #include <net/tls_prot.h> 52 #include <net/handshake.h> 53 54 #include <linux/bvec.h> 55 #include <linux/highmem.h> 56 #include <linux/uio.h> 57 #include <linux/sched/mm.h> 58 59 #include <trace/events/sock.h> 60 #include <trace/events/sunrpc.h> 61 62 #include "socklib.h" 63 #include "sunrpc.h" 64 65 static void xs_close(struct rpc_xprt *xprt); 66 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock); 67 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 68 struct socket *sock); 69 70 /* 71 * xprtsock tunables 72 */ 73 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 74 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 75 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 76 77 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 78 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 79 80 #define XS_TCP_LINGER_TO (15U * HZ) 81 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 82 83 /* 84 * We can register our own files under /proc/sys/sunrpc by 85 * calling register_sysctl() again. The files in that 86 * directory become the union of all files registered there. 87 * 88 * We simply need to make sure that we don't collide with 89 * someone else's file names! 90 */ 91 92 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 93 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 94 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 95 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 96 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 97 98 static struct ctl_table_header *sunrpc_table_header; 99 100 static struct xprt_class xs_local_transport; 101 static struct xprt_class xs_udp_transport; 102 static struct xprt_class xs_tcp_transport; 103 static struct xprt_class xs_tcp_tls_transport; 104 static struct xprt_class xs_bc_tcp_transport; 105 106 /* 107 * FIXME: changing the UDP slot table size should also resize the UDP 108 * socket buffers for existing UDP transports 109 */ 110 static struct ctl_table xs_tunables_table[] = { 111 { 112 .procname = "udp_slot_table_entries", 113 .data = &xprt_udp_slot_table_entries, 114 .maxlen = sizeof(unsigned int), 115 .mode = 0644, 116 .proc_handler = proc_dointvec_minmax, 117 .extra1 = &min_slot_table_size, 118 .extra2 = &max_slot_table_size 119 }, 120 { 121 .procname = "tcp_slot_table_entries", 122 .data = &xprt_tcp_slot_table_entries, 123 .maxlen = sizeof(unsigned int), 124 .mode = 0644, 125 .proc_handler = proc_dointvec_minmax, 126 .extra1 = &min_slot_table_size, 127 .extra2 = &max_slot_table_size 128 }, 129 { 130 .procname = "tcp_max_slot_table_entries", 131 .data = &xprt_max_tcp_slot_table_entries, 132 .maxlen = sizeof(unsigned int), 133 .mode = 0644, 134 .proc_handler = proc_dointvec_minmax, 135 .extra1 = &min_slot_table_size, 136 .extra2 = &max_tcp_slot_table_limit 137 }, 138 { 139 .procname = "min_resvport", 140 .data = &xprt_min_resvport, 141 .maxlen = sizeof(unsigned int), 142 .mode = 0644, 143 .proc_handler = proc_dointvec_minmax, 144 .extra1 = &xprt_min_resvport_limit, 145 .extra2 = &xprt_max_resvport_limit 146 }, 147 { 148 .procname = "max_resvport", 149 .data = &xprt_max_resvport, 150 .maxlen = sizeof(unsigned int), 151 .mode = 0644, 152 .proc_handler = proc_dointvec_minmax, 153 .extra1 = &xprt_min_resvport_limit, 154 .extra2 = &xprt_max_resvport_limit 155 }, 156 { 157 .procname = "tcp_fin_timeout", 158 .data = &xs_tcp_fin_timeout, 159 .maxlen = sizeof(xs_tcp_fin_timeout), 160 .mode = 0644, 161 .proc_handler = proc_dointvec_jiffies, 162 }, 163 { }, 164 }; 165 166 /* 167 * Wait duration for a reply from the RPC portmapper. 168 */ 169 #define XS_BIND_TO (60U * HZ) 170 171 /* 172 * Delay if a UDP socket connect error occurs. This is most likely some 173 * kind of resource problem on the local host. 174 */ 175 #define XS_UDP_REEST_TO (2U * HZ) 176 177 /* 178 * The reestablish timeout allows clients to delay for a bit before attempting 179 * to reconnect to a server that just dropped our connection. 180 * 181 * We implement an exponential backoff when trying to reestablish a TCP 182 * transport connection with the server. Some servers like to drop a TCP 183 * connection when they are overworked, so we start with a short timeout and 184 * increase over time if the server is down or not responding. 185 */ 186 #define XS_TCP_INIT_REEST_TO (3U * HZ) 187 188 /* 189 * TCP idle timeout; client drops the transport socket if it is idle 190 * for this long. Note that we also timeout UDP sockets to prevent 191 * holding port numbers when there is no RPC traffic. 192 */ 193 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 194 195 /* 196 * TLS handshake timeout. 197 */ 198 #define XS_TLS_HANDSHAKE_TO (10U * HZ) 199 200 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 201 # undef RPC_DEBUG_DATA 202 # define RPCDBG_FACILITY RPCDBG_TRANS 203 #endif 204 205 #ifdef RPC_DEBUG_DATA 206 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 207 { 208 u8 *buf = (u8 *) packet; 209 int j; 210 211 dprintk("RPC: %s\n", msg); 212 for (j = 0; j < count && j < 128; j += 4) { 213 if (!(j & 31)) { 214 if (j) 215 dprintk("\n"); 216 dprintk("0x%04x ", j); 217 } 218 dprintk("%02x%02x%02x%02x ", 219 buf[j], buf[j+1], buf[j+2], buf[j+3]); 220 } 221 dprintk("\n"); 222 } 223 #else 224 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 225 { 226 /* NOP */ 227 } 228 #endif 229 230 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 231 { 232 return (struct rpc_xprt *) sk->sk_user_data; 233 } 234 235 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 236 { 237 return (struct sockaddr *) &xprt->addr; 238 } 239 240 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 241 { 242 return (struct sockaddr_un *) &xprt->addr; 243 } 244 245 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 246 { 247 return (struct sockaddr_in *) &xprt->addr; 248 } 249 250 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 251 { 252 return (struct sockaddr_in6 *) &xprt->addr; 253 } 254 255 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 256 { 257 struct sockaddr *sap = xs_addr(xprt); 258 struct sockaddr_in6 *sin6; 259 struct sockaddr_in *sin; 260 struct sockaddr_un *sun; 261 char buf[128]; 262 263 switch (sap->sa_family) { 264 case AF_LOCAL: 265 sun = xs_addr_un(xprt); 266 if (sun->sun_path[0]) { 267 strscpy(buf, sun->sun_path, sizeof(buf)); 268 } else { 269 buf[0] = '@'; 270 strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1); 271 } 272 xprt->address_strings[RPC_DISPLAY_ADDR] = 273 kstrdup(buf, GFP_KERNEL); 274 break; 275 case AF_INET: 276 (void)rpc_ntop(sap, buf, sizeof(buf)); 277 xprt->address_strings[RPC_DISPLAY_ADDR] = 278 kstrdup(buf, GFP_KERNEL); 279 sin = xs_addr_in(xprt); 280 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 281 break; 282 case AF_INET6: 283 (void)rpc_ntop(sap, buf, sizeof(buf)); 284 xprt->address_strings[RPC_DISPLAY_ADDR] = 285 kstrdup(buf, GFP_KERNEL); 286 sin6 = xs_addr_in6(xprt); 287 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 288 break; 289 default: 290 BUG(); 291 } 292 293 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 294 } 295 296 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 297 { 298 struct sockaddr *sap = xs_addr(xprt); 299 char buf[128]; 300 301 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 302 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 303 304 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 305 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 306 } 307 308 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 309 const char *protocol, 310 const char *netid) 311 { 312 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 313 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 314 xs_format_common_peer_addresses(xprt); 315 xs_format_common_peer_ports(xprt); 316 } 317 318 static void xs_update_peer_port(struct rpc_xprt *xprt) 319 { 320 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 321 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 322 323 xs_format_common_peer_ports(xprt); 324 } 325 326 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 327 { 328 unsigned int i; 329 330 for (i = 0; i < RPC_DISPLAY_MAX; i++) 331 switch (i) { 332 case RPC_DISPLAY_PROTO: 333 case RPC_DISPLAY_NETID: 334 continue; 335 default: 336 kfree(xprt->address_strings[i]); 337 } 338 } 339 340 static size_t 341 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp) 342 { 343 size_t i,n; 344 345 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES)) 346 return want; 347 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT; 348 for (i = 0; i < n; i++) { 349 if (buf->pages[i]) 350 continue; 351 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp); 352 if (!buf->pages[i]) { 353 i *= PAGE_SIZE; 354 return i > buf->page_base ? i - buf->page_base : 0; 355 } 356 } 357 return want; 358 } 359 360 static int 361 xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg, 362 struct cmsghdr *cmsg, int ret) 363 { 364 if (cmsg->cmsg_level == SOL_TLS && 365 cmsg->cmsg_type == TLS_GET_RECORD_TYPE) { 366 u8 content_type = *((u8 *)CMSG_DATA(cmsg)); 367 368 switch (content_type) { 369 case TLS_RECORD_TYPE_DATA: 370 /* TLS sets EOR at the end of each application data 371 * record, even though there might be more frames 372 * waiting to be decrypted. 373 */ 374 msg->msg_flags &= ~MSG_EOR; 375 break; 376 case TLS_RECORD_TYPE_ALERT: 377 ret = -ENOTCONN; 378 break; 379 default: 380 ret = -EAGAIN; 381 } 382 } 383 return ret; 384 } 385 386 static int 387 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags) 388 { 389 union { 390 struct cmsghdr cmsg; 391 u8 buf[CMSG_SPACE(sizeof(u8))]; 392 } u; 393 int ret; 394 395 msg->msg_control = &u; 396 msg->msg_controllen = sizeof(u); 397 ret = sock_recvmsg(sock, msg, flags); 398 if (msg->msg_controllen != sizeof(u)) 399 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret); 400 return ret; 401 } 402 403 static ssize_t 404 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek) 405 { 406 ssize_t ret; 407 if (seek != 0) 408 iov_iter_advance(&msg->msg_iter, seek); 409 ret = xs_sock_recv_cmsg(sock, msg, flags); 410 return ret > 0 ? ret + seek : ret; 411 } 412 413 static ssize_t 414 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags, 415 struct kvec *kvec, size_t count, size_t seek) 416 { 417 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count); 418 return xs_sock_recvmsg(sock, msg, flags, seek); 419 } 420 421 static ssize_t 422 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags, 423 struct bio_vec *bvec, unsigned long nr, size_t count, 424 size_t seek) 425 { 426 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count); 427 return xs_sock_recvmsg(sock, msg, flags, seek); 428 } 429 430 static ssize_t 431 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags, 432 size_t count) 433 { 434 iov_iter_discard(&msg->msg_iter, ITER_DEST, count); 435 return xs_sock_recv_cmsg(sock, msg, flags); 436 } 437 438 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 439 static void 440 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 441 { 442 struct bvec_iter bi = { 443 .bi_size = count, 444 }; 445 struct bio_vec bv; 446 447 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); 448 for_each_bvec(bv, bvec, bi, bi) 449 flush_dcache_page(bv.bv_page); 450 } 451 #else 452 static inline void 453 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 454 { 455 } 456 #endif 457 458 static ssize_t 459 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags, 460 struct xdr_buf *buf, size_t count, size_t seek, size_t *read) 461 { 462 size_t want, seek_init = seek, offset = 0; 463 ssize_t ret; 464 465 want = min_t(size_t, count, buf->head[0].iov_len); 466 if (seek < want) { 467 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek); 468 if (ret <= 0) 469 goto sock_err; 470 offset += ret; 471 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 472 goto out; 473 if (ret != want) 474 goto out; 475 seek = 0; 476 } else { 477 seek -= want; 478 offset += want; 479 } 480 481 want = xs_alloc_sparse_pages( 482 buf, min_t(size_t, count - offset, buf->page_len), 483 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN); 484 if (seek < want) { 485 ret = xs_read_bvec(sock, msg, flags, buf->bvec, 486 xdr_buf_pagecount(buf), 487 want + buf->page_base, 488 seek + buf->page_base); 489 if (ret <= 0) 490 goto sock_err; 491 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base); 492 ret -= buf->page_base; 493 offset += ret; 494 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 495 goto out; 496 if (ret != want) 497 goto out; 498 seek = 0; 499 } else { 500 seek -= want; 501 offset += want; 502 } 503 504 want = min_t(size_t, count - offset, buf->tail[0].iov_len); 505 if (seek < want) { 506 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek); 507 if (ret <= 0) 508 goto sock_err; 509 offset += ret; 510 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 511 goto out; 512 if (ret != want) 513 goto out; 514 } else if (offset < seek_init) 515 offset = seek_init; 516 ret = -EMSGSIZE; 517 out: 518 *read = offset - seek_init; 519 return ret; 520 sock_err: 521 offset += seek; 522 goto out; 523 } 524 525 static void 526 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf) 527 { 528 if (!transport->recv.copied) { 529 if (buf->head[0].iov_len >= transport->recv.offset) 530 memcpy(buf->head[0].iov_base, 531 &transport->recv.xid, 532 transport->recv.offset); 533 transport->recv.copied = transport->recv.offset; 534 } 535 } 536 537 static bool 538 xs_read_stream_request_done(struct sock_xprt *transport) 539 { 540 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT); 541 } 542 543 static void 544 xs_read_stream_check_eor(struct sock_xprt *transport, 545 struct msghdr *msg) 546 { 547 if (xs_read_stream_request_done(transport)) 548 msg->msg_flags |= MSG_EOR; 549 } 550 551 static ssize_t 552 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg, 553 int flags, struct rpc_rqst *req) 554 { 555 struct xdr_buf *buf = &req->rq_private_buf; 556 size_t want, read; 557 ssize_t ret; 558 559 xs_read_header(transport, buf); 560 561 want = transport->recv.len - transport->recv.offset; 562 if (want != 0) { 563 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf, 564 transport->recv.copied + want, 565 transport->recv.copied, 566 &read); 567 transport->recv.offset += read; 568 transport->recv.copied += read; 569 } 570 571 if (transport->recv.offset == transport->recv.len) 572 xs_read_stream_check_eor(transport, msg); 573 574 if (want == 0) 575 return 0; 576 577 switch (ret) { 578 default: 579 break; 580 case -EFAULT: 581 case -EMSGSIZE: 582 msg->msg_flags |= MSG_TRUNC; 583 return read; 584 case 0: 585 return -ESHUTDOWN; 586 } 587 return ret < 0 ? ret : read; 588 } 589 590 static size_t 591 xs_read_stream_headersize(bool isfrag) 592 { 593 if (isfrag) 594 return sizeof(__be32); 595 return 3 * sizeof(__be32); 596 } 597 598 static ssize_t 599 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg, 600 int flags, size_t want, size_t seek) 601 { 602 struct kvec kvec = { 603 .iov_base = &transport->recv.fraghdr, 604 .iov_len = want, 605 }; 606 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek); 607 } 608 609 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 610 static ssize_t 611 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 612 { 613 struct rpc_xprt *xprt = &transport->xprt; 614 struct rpc_rqst *req; 615 ssize_t ret; 616 617 /* Is this transport associated with the backchannel? */ 618 if (!xprt->bc_serv) 619 return -ESHUTDOWN; 620 621 /* Look up and lock the request corresponding to the given XID */ 622 req = xprt_lookup_bc_request(xprt, transport->recv.xid); 623 if (!req) { 624 printk(KERN_WARNING "Callback slot table overflowed\n"); 625 return -ESHUTDOWN; 626 } 627 if (transport->recv.copied && !req->rq_private_buf.len) 628 return -ESHUTDOWN; 629 630 ret = xs_read_stream_request(transport, msg, flags, req); 631 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 632 xprt_complete_bc_request(req, transport->recv.copied); 633 else 634 req->rq_private_buf.len = transport->recv.copied; 635 636 return ret; 637 } 638 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 639 static ssize_t 640 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 641 { 642 return -ESHUTDOWN; 643 } 644 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 645 646 static ssize_t 647 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags) 648 { 649 struct rpc_xprt *xprt = &transport->xprt; 650 struct rpc_rqst *req; 651 ssize_t ret = 0; 652 653 /* Look up and lock the request corresponding to the given XID */ 654 spin_lock(&xprt->queue_lock); 655 req = xprt_lookup_rqst(xprt, transport->recv.xid); 656 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) { 657 msg->msg_flags |= MSG_TRUNC; 658 goto out; 659 } 660 xprt_pin_rqst(req); 661 spin_unlock(&xprt->queue_lock); 662 663 ret = xs_read_stream_request(transport, msg, flags, req); 664 665 spin_lock(&xprt->queue_lock); 666 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 667 xprt_complete_rqst(req->rq_task, transport->recv.copied); 668 else 669 req->rq_private_buf.len = transport->recv.copied; 670 xprt_unpin_rqst(req); 671 out: 672 spin_unlock(&xprt->queue_lock); 673 return ret; 674 } 675 676 static ssize_t 677 xs_read_stream(struct sock_xprt *transport, int flags) 678 { 679 struct msghdr msg = { 0 }; 680 size_t want, read = 0; 681 ssize_t ret = 0; 682 683 if (transport->recv.len == 0) { 684 want = xs_read_stream_headersize(transport->recv.copied != 0); 685 ret = xs_read_stream_header(transport, &msg, flags, want, 686 transport->recv.offset); 687 if (ret <= 0) 688 goto out_err; 689 transport->recv.offset = ret; 690 if (transport->recv.offset != want) 691 return transport->recv.offset; 692 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) & 693 RPC_FRAGMENT_SIZE_MASK; 694 transport->recv.offset -= sizeof(transport->recv.fraghdr); 695 read = ret; 696 } 697 698 switch (be32_to_cpu(transport->recv.calldir)) { 699 default: 700 msg.msg_flags |= MSG_TRUNC; 701 break; 702 case RPC_CALL: 703 ret = xs_read_stream_call(transport, &msg, flags); 704 break; 705 case RPC_REPLY: 706 ret = xs_read_stream_reply(transport, &msg, flags); 707 } 708 if (msg.msg_flags & MSG_TRUNC) { 709 transport->recv.calldir = cpu_to_be32(-1); 710 transport->recv.copied = -1; 711 } 712 if (ret < 0) 713 goto out_err; 714 read += ret; 715 if (transport->recv.offset < transport->recv.len) { 716 if (!(msg.msg_flags & MSG_TRUNC)) 717 return read; 718 msg.msg_flags = 0; 719 ret = xs_read_discard(transport->sock, &msg, flags, 720 transport->recv.len - transport->recv.offset); 721 if (ret <= 0) 722 goto out_err; 723 transport->recv.offset += ret; 724 read += ret; 725 if (transport->recv.offset != transport->recv.len) 726 return read; 727 } 728 if (xs_read_stream_request_done(transport)) { 729 trace_xs_stream_read_request(transport); 730 transport->recv.copied = 0; 731 } 732 transport->recv.offset = 0; 733 transport->recv.len = 0; 734 return read; 735 out_err: 736 return ret != 0 ? ret : -ESHUTDOWN; 737 } 738 739 static __poll_t xs_poll_socket(struct sock_xprt *transport) 740 { 741 return transport->sock->ops->poll(transport->file, transport->sock, 742 NULL); 743 } 744 745 static bool xs_poll_socket_readable(struct sock_xprt *transport) 746 { 747 __poll_t events = xs_poll_socket(transport); 748 749 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP); 750 } 751 752 static void xs_poll_check_readable(struct sock_xprt *transport) 753 { 754 755 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 756 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state)) 757 return; 758 if (!xs_poll_socket_readable(transport)) 759 return; 760 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 761 queue_work(xprtiod_workqueue, &transport->recv_worker); 762 } 763 764 static void xs_stream_data_receive(struct sock_xprt *transport) 765 { 766 size_t read = 0; 767 ssize_t ret = 0; 768 769 mutex_lock(&transport->recv_mutex); 770 if (transport->sock == NULL) 771 goto out; 772 for (;;) { 773 ret = xs_read_stream(transport, MSG_DONTWAIT); 774 if (ret < 0) 775 break; 776 read += ret; 777 cond_resched(); 778 } 779 if (ret == -ESHUTDOWN) 780 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 781 else 782 xs_poll_check_readable(transport); 783 out: 784 mutex_unlock(&transport->recv_mutex); 785 trace_xs_stream_read_data(&transport->xprt, ret, read); 786 } 787 788 static void xs_stream_data_receive_workfn(struct work_struct *work) 789 { 790 struct sock_xprt *transport = 791 container_of(work, struct sock_xprt, recv_worker); 792 unsigned int pflags = memalloc_nofs_save(); 793 794 xs_stream_data_receive(transport); 795 memalloc_nofs_restore(pflags); 796 } 797 798 static void 799 xs_stream_reset_connect(struct sock_xprt *transport) 800 { 801 transport->recv.offset = 0; 802 transport->recv.len = 0; 803 transport->recv.copied = 0; 804 transport->xmit.offset = 0; 805 } 806 807 static void 808 xs_stream_start_connect(struct sock_xprt *transport) 809 { 810 transport->xprt.stat.connect_count++; 811 transport->xprt.stat.connect_start = jiffies; 812 } 813 814 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 815 816 /** 817 * xs_nospace - handle transmit was incomplete 818 * @req: pointer to RPC request 819 * @transport: pointer to struct sock_xprt 820 * 821 */ 822 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport) 823 { 824 struct rpc_xprt *xprt = &transport->xprt; 825 struct sock *sk = transport->inet; 826 int ret = -EAGAIN; 827 828 trace_rpc_socket_nospace(req, transport); 829 830 /* Protect against races with write_space */ 831 spin_lock(&xprt->transport_lock); 832 833 /* Don't race with disconnect */ 834 if (xprt_connected(xprt)) { 835 /* wait for more buffer space */ 836 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state); 837 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 838 sk->sk_write_pending++; 839 xprt_wait_for_buffer_space(xprt); 840 } else 841 ret = -ENOTCONN; 842 843 spin_unlock(&xprt->transport_lock); 844 return ret; 845 } 846 847 static int xs_sock_nospace(struct rpc_rqst *req) 848 { 849 struct sock_xprt *transport = 850 container_of(req->rq_xprt, struct sock_xprt, xprt); 851 struct sock *sk = transport->inet; 852 int ret = -EAGAIN; 853 854 lock_sock(sk); 855 if (!sock_writeable(sk)) 856 ret = xs_nospace(req, transport); 857 release_sock(sk); 858 return ret; 859 } 860 861 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait) 862 { 863 struct sock_xprt *transport = 864 container_of(req->rq_xprt, struct sock_xprt, xprt); 865 struct sock *sk = transport->inet; 866 int ret = -EAGAIN; 867 868 if (vm_wait) 869 return -ENOBUFS; 870 lock_sock(sk); 871 if (!sk_stream_memory_free(sk)) 872 ret = xs_nospace(req, transport); 873 release_sock(sk); 874 return ret; 875 } 876 877 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf) 878 { 879 return xdr_alloc_bvec(buf, rpc_task_gfp_mask()); 880 } 881 882 /* 883 * Determine if the previous message in the stream was aborted before it 884 * could complete transmission. 885 */ 886 static bool 887 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req) 888 { 889 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0; 890 } 891 892 /* 893 * Return the stream record marker field for a record of length < 2^31-1 894 */ 895 static rpc_fraghdr 896 xs_stream_record_marker(struct xdr_buf *xdr) 897 { 898 if (!xdr->len) 899 return 0; 900 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len); 901 } 902 903 /** 904 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 905 * @req: pointer to RPC request 906 * 907 * Return values: 908 * 0: The request has been sent 909 * EAGAIN: The socket was blocked, please call again later to 910 * complete the request 911 * ENOTCONN: Caller needs to invoke connect logic then call again 912 * other: Some other error occurred, the request was not sent 913 */ 914 static int xs_local_send_request(struct rpc_rqst *req) 915 { 916 struct rpc_xprt *xprt = req->rq_xprt; 917 struct sock_xprt *transport = 918 container_of(xprt, struct sock_xprt, xprt); 919 struct xdr_buf *xdr = &req->rq_snd_buf; 920 rpc_fraghdr rm = xs_stream_record_marker(xdr); 921 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 922 struct msghdr msg = { 923 .msg_flags = XS_SENDMSG_FLAGS, 924 }; 925 bool vm_wait; 926 unsigned int sent; 927 int status; 928 929 /* Close the stream if the previous transmission was incomplete */ 930 if (xs_send_request_was_aborted(transport, req)) { 931 xprt_force_disconnect(xprt); 932 return -ENOTCONN; 933 } 934 935 xs_pktdump("packet data:", 936 req->rq_svec->iov_base, req->rq_svec->iov_len); 937 938 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false; 939 940 req->rq_xtime = ktime_get(); 941 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 942 transport->xmit.offset, rm, &sent); 943 dprintk("RPC: %s(%u) = %d\n", 944 __func__, xdr->len - transport->xmit.offset, status); 945 946 if (likely(sent > 0) || status == 0) { 947 transport->xmit.offset += sent; 948 req->rq_bytes_sent = transport->xmit.offset; 949 if (likely(req->rq_bytes_sent >= msglen)) { 950 req->rq_xmit_bytes_sent += transport->xmit.offset; 951 transport->xmit.offset = 0; 952 return 0; 953 } 954 status = -EAGAIN; 955 vm_wait = false; 956 } 957 958 switch (status) { 959 case -EAGAIN: 960 status = xs_stream_nospace(req, vm_wait); 961 break; 962 default: 963 dprintk("RPC: sendmsg returned unrecognized error %d\n", 964 -status); 965 fallthrough; 966 case -EPIPE: 967 xprt_force_disconnect(xprt); 968 status = -ENOTCONN; 969 } 970 971 return status; 972 } 973 974 /** 975 * xs_udp_send_request - write an RPC request to a UDP socket 976 * @req: pointer to RPC request 977 * 978 * Return values: 979 * 0: The request has been sent 980 * EAGAIN: The socket was blocked, please call again later to 981 * complete the request 982 * ENOTCONN: Caller needs to invoke connect logic then call again 983 * other: Some other error occurred, the request was not sent 984 */ 985 static int xs_udp_send_request(struct rpc_rqst *req) 986 { 987 struct rpc_xprt *xprt = req->rq_xprt; 988 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 989 struct xdr_buf *xdr = &req->rq_snd_buf; 990 struct msghdr msg = { 991 .msg_name = xs_addr(xprt), 992 .msg_namelen = xprt->addrlen, 993 .msg_flags = XS_SENDMSG_FLAGS, 994 }; 995 unsigned int sent; 996 int status; 997 998 xs_pktdump("packet data:", 999 req->rq_svec->iov_base, 1000 req->rq_svec->iov_len); 1001 1002 if (!xprt_bound(xprt)) 1003 return -ENOTCONN; 1004 1005 if (!xprt_request_get_cong(xprt, req)) 1006 return -EBADSLT; 1007 1008 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask()); 1009 if (status < 0) 1010 return status; 1011 req->rq_xtime = ktime_get(); 1012 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent); 1013 1014 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 1015 xdr->len, status); 1016 1017 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 1018 if (status == -EPERM) 1019 goto process_status; 1020 1021 if (status == -EAGAIN && sock_writeable(transport->inet)) 1022 status = -ENOBUFS; 1023 1024 if (sent > 0 || status == 0) { 1025 req->rq_xmit_bytes_sent += sent; 1026 if (sent >= req->rq_slen) 1027 return 0; 1028 /* Still some bytes left; set up for a retry later. */ 1029 status = -EAGAIN; 1030 } 1031 1032 process_status: 1033 switch (status) { 1034 case -ENOTSOCK: 1035 status = -ENOTCONN; 1036 /* Should we call xs_close() here? */ 1037 break; 1038 case -EAGAIN: 1039 status = xs_sock_nospace(req); 1040 break; 1041 case -ENETUNREACH: 1042 case -ENOBUFS: 1043 case -EPIPE: 1044 case -ECONNREFUSED: 1045 case -EPERM: 1046 /* When the server has died, an ICMP port unreachable message 1047 * prompts ECONNREFUSED. */ 1048 break; 1049 default: 1050 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1051 -status); 1052 } 1053 1054 return status; 1055 } 1056 1057 /** 1058 * xs_tcp_send_request - write an RPC request to a TCP socket 1059 * @req: pointer to RPC request 1060 * 1061 * Return values: 1062 * 0: The request has been sent 1063 * EAGAIN: The socket was blocked, please call again later to 1064 * complete the request 1065 * ENOTCONN: Caller needs to invoke connect logic then call again 1066 * other: Some other error occurred, the request was not sent 1067 * 1068 * XXX: In the case of soft timeouts, should we eventually give up 1069 * if sendmsg is not able to make progress? 1070 */ 1071 static int xs_tcp_send_request(struct rpc_rqst *req) 1072 { 1073 struct rpc_xprt *xprt = req->rq_xprt; 1074 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1075 struct xdr_buf *xdr = &req->rq_snd_buf; 1076 rpc_fraghdr rm = xs_stream_record_marker(xdr); 1077 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 1078 struct msghdr msg = { 1079 .msg_flags = XS_SENDMSG_FLAGS, 1080 }; 1081 bool vm_wait; 1082 unsigned int sent; 1083 int status; 1084 1085 /* Close the stream if the previous transmission was incomplete */ 1086 if (xs_send_request_was_aborted(transport, req)) { 1087 if (transport->sock != NULL) 1088 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 1089 return -ENOTCONN; 1090 } 1091 if (!transport->inet) 1092 return -ENOTCONN; 1093 1094 xs_pktdump("packet data:", 1095 req->rq_svec->iov_base, 1096 req->rq_svec->iov_len); 1097 1098 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state)) 1099 xs_tcp_set_socket_timeouts(xprt, transport->sock); 1100 1101 xs_set_srcport(transport, transport->sock); 1102 1103 /* Continue transmitting the packet/record. We must be careful 1104 * to cope with writespace callbacks arriving _after_ we have 1105 * called sendmsg(). */ 1106 req->rq_xtime = ktime_get(); 1107 tcp_sock_set_cork(transport->inet, true); 1108 1109 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false; 1110 1111 do { 1112 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 1113 transport->xmit.offset, rm, &sent); 1114 1115 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 1116 xdr->len - transport->xmit.offset, status); 1117 1118 /* If we've sent the entire packet, immediately 1119 * reset the count of bytes sent. */ 1120 transport->xmit.offset += sent; 1121 req->rq_bytes_sent = transport->xmit.offset; 1122 if (likely(req->rq_bytes_sent >= msglen)) { 1123 req->rq_xmit_bytes_sent += transport->xmit.offset; 1124 transport->xmit.offset = 0; 1125 if (atomic_long_read(&xprt->xmit_queuelen) == 1) 1126 tcp_sock_set_cork(transport->inet, false); 1127 return 0; 1128 } 1129 1130 WARN_ON_ONCE(sent == 0 && status == 0); 1131 1132 if (sent > 0) 1133 vm_wait = false; 1134 1135 } while (status == 0); 1136 1137 switch (status) { 1138 case -ENOTSOCK: 1139 status = -ENOTCONN; 1140 /* Should we call xs_close() here? */ 1141 break; 1142 case -EAGAIN: 1143 status = xs_stream_nospace(req, vm_wait); 1144 break; 1145 case -ECONNRESET: 1146 case -ECONNREFUSED: 1147 case -ENOTCONN: 1148 case -EADDRINUSE: 1149 case -ENOBUFS: 1150 case -EPIPE: 1151 break; 1152 default: 1153 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1154 -status); 1155 } 1156 1157 return status; 1158 } 1159 1160 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1161 { 1162 transport->old_data_ready = sk->sk_data_ready; 1163 transport->old_state_change = sk->sk_state_change; 1164 transport->old_write_space = sk->sk_write_space; 1165 transport->old_error_report = sk->sk_error_report; 1166 } 1167 1168 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1169 { 1170 sk->sk_data_ready = transport->old_data_ready; 1171 sk->sk_state_change = transport->old_state_change; 1172 sk->sk_write_space = transport->old_write_space; 1173 sk->sk_error_report = transport->old_error_report; 1174 } 1175 1176 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt) 1177 { 1178 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1179 1180 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 1181 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state); 1182 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state); 1183 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state); 1184 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state); 1185 } 1186 1187 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr) 1188 { 1189 set_bit(nr, &transport->sock_state); 1190 queue_work(xprtiod_workqueue, &transport->error_worker); 1191 } 1192 1193 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1194 { 1195 xprt->connect_cookie++; 1196 smp_mb__before_atomic(); 1197 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1198 clear_bit(XPRT_CLOSING, &xprt->state); 1199 xs_sock_reset_state_flags(xprt); 1200 smp_mb__after_atomic(); 1201 } 1202 1203 /** 1204 * xs_error_report - callback to handle TCP socket state errors 1205 * @sk: socket 1206 * 1207 * Note: we don't call sock_error() since there may be a rpc_task 1208 * using the socket, and so we don't want to clear sk->sk_err. 1209 */ 1210 static void xs_error_report(struct sock *sk) 1211 { 1212 struct sock_xprt *transport; 1213 struct rpc_xprt *xprt; 1214 1215 if (!(xprt = xprt_from_sock(sk))) 1216 return; 1217 1218 transport = container_of(xprt, struct sock_xprt, xprt); 1219 transport->xprt_err = -sk->sk_err; 1220 if (transport->xprt_err == 0) 1221 return; 1222 dprintk("RPC: xs_error_report client %p, error=%d...\n", 1223 xprt, -transport->xprt_err); 1224 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err); 1225 1226 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */ 1227 smp_mb__before_atomic(); 1228 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR); 1229 } 1230 1231 static void xs_reset_transport(struct sock_xprt *transport) 1232 { 1233 struct socket *sock = transport->sock; 1234 struct sock *sk = transport->inet; 1235 struct rpc_xprt *xprt = &transport->xprt; 1236 struct file *filp = transport->file; 1237 1238 if (sk == NULL) 1239 return; 1240 /* 1241 * Make sure we're calling this in a context from which it is safe 1242 * to call __fput_sync(). In practice that means rpciod and the 1243 * system workqueue. 1244 */ 1245 if (!(current->flags & PF_WQ_WORKER)) { 1246 WARN_ON_ONCE(1); 1247 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 1248 return; 1249 } 1250 1251 if (atomic_read(&transport->xprt.swapper)) 1252 sk_clear_memalloc(sk); 1253 1254 tls_handshake_cancel(sk); 1255 1256 kernel_sock_shutdown(sock, SHUT_RDWR); 1257 1258 mutex_lock(&transport->recv_mutex); 1259 lock_sock(sk); 1260 transport->inet = NULL; 1261 transport->sock = NULL; 1262 transport->file = NULL; 1263 1264 sk->sk_user_data = NULL; 1265 1266 xs_restore_old_callbacks(transport, sk); 1267 xprt_clear_connected(xprt); 1268 xs_sock_reset_connection_flags(xprt); 1269 /* Reset stream record info */ 1270 xs_stream_reset_connect(transport); 1271 release_sock(sk); 1272 mutex_unlock(&transport->recv_mutex); 1273 1274 trace_rpc_socket_close(xprt, sock); 1275 __fput_sync(filp); 1276 1277 xprt_disconnect_done(xprt); 1278 } 1279 1280 /** 1281 * xs_close - close a socket 1282 * @xprt: transport 1283 * 1284 * This is used when all requests are complete; ie, no DRC state remains 1285 * on the server we want to save. 1286 * 1287 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 1288 * xs_reset_transport() zeroing the socket from underneath a writer. 1289 */ 1290 static void xs_close(struct rpc_xprt *xprt) 1291 { 1292 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1293 1294 dprintk("RPC: xs_close xprt %p\n", xprt); 1295 1296 if (transport->sock) 1297 tls_handshake_close(transport->sock); 1298 xs_reset_transport(transport); 1299 xprt->reestablish_timeout = 0; 1300 } 1301 1302 static void xs_inject_disconnect(struct rpc_xprt *xprt) 1303 { 1304 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 1305 xprt); 1306 xprt_disconnect_done(xprt); 1307 } 1308 1309 static void xs_xprt_free(struct rpc_xprt *xprt) 1310 { 1311 xs_free_peer_addresses(xprt); 1312 xprt_free(xprt); 1313 } 1314 1315 /** 1316 * xs_destroy - prepare to shutdown a transport 1317 * @xprt: doomed transport 1318 * 1319 */ 1320 static void xs_destroy(struct rpc_xprt *xprt) 1321 { 1322 struct sock_xprt *transport = container_of(xprt, 1323 struct sock_xprt, xprt); 1324 dprintk("RPC: xs_destroy xprt %p\n", xprt); 1325 1326 cancel_delayed_work_sync(&transport->connect_worker); 1327 xs_close(xprt); 1328 cancel_work_sync(&transport->recv_worker); 1329 cancel_work_sync(&transport->error_worker); 1330 xs_xprt_free(xprt); 1331 module_put(THIS_MODULE); 1332 } 1333 1334 /** 1335 * xs_udp_data_read_skb - receive callback for UDP sockets 1336 * @xprt: transport 1337 * @sk: socket 1338 * @skb: skbuff 1339 * 1340 */ 1341 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1342 struct sock *sk, 1343 struct sk_buff *skb) 1344 { 1345 struct rpc_task *task; 1346 struct rpc_rqst *rovr; 1347 int repsize, copied; 1348 u32 _xid; 1349 __be32 *xp; 1350 1351 repsize = skb->len; 1352 if (repsize < 4) { 1353 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1354 return; 1355 } 1356 1357 /* Copy the XID from the skb... */ 1358 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid); 1359 if (xp == NULL) 1360 return; 1361 1362 /* Look up and lock the request corresponding to the given XID */ 1363 spin_lock(&xprt->queue_lock); 1364 rovr = xprt_lookup_rqst(xprt, *xp); 1365 if (!rovr) 1366 goto out_unlock; 1367 xprt_pin_rqst(rovr); 1368 xprt_update_rtt(rovr->rq_task); 1369 spin_unlock(&xprt->queue_lock); 1370 task = rovr->rq_task; 1371 1372 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1373 copied = repsize; 1374 1375 /* Suck it into the iovec, verify checksum if not done by hw. */ 1376 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1377 spin_lock(&xprt->queue_lock); 1378 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS); 1379 goto out_unpin; 1380 } 1381 1382 1383 spin_lock(&xprt->transport_lock); 1384 xprt_adjust_cwnd(xprt, task, copied); 1385 spin_unlock(&xprt->transport_lock); 1386 spin_lock(&xprt->queue_lock); 1387 xprt_complete_rqst(task, copied); 1388 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS); 1389 out_unpin: 1390 xprt_unpin_rqst(rovr); 1391 out_unlock: 1392 spin_unlock(&xprt->queue_lock); 1393 } 1394 1395 static void xs_udp_data_receive(struct sock_xprt *transport) 1396 { 1397 struct sk_buff *skb; 1398 struct sock *sk; 1399 int err; 1400 1401 mutex_lock(&transport->recv_mutex); 1402 sk = transport->inet; 1403 if (sk == NULL) 1404 goto out; 1405 for (;;) { 1406 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1407 if (skb == NULL) 1408 break; 1409 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1410 consume_skb(skb); 1411 cond_resched(); 1412 } 1413 xs_poll_check_readable(transport); 1414 out: 1415 mutex_unlock(&transport->recv_mutex); 1416 } 1417 1418 static void xs_udp_data_receive_workfn(struct work_struct *work) 1419 { 1420 struct sock_xprt *transport = 1421 container_of(work, struct sock_xprt, recv_worker); 1422 unsigned int pflags = memalloc_nofs_save(); 1423 1424 xs_udp_data_receive(transport); 1425 memalloc_nofs_restore(pflags); 1426 } 1427 1428 /** 1429 * xs_data_ready - "data ready" callback for sockets 1430 * @sk: socket with data to read 1431 * 1432 */ 1433 static void xs_data_ready(struct sock *sk) 1434 { 1435 struct rpc_xprt *xprt; 1436 1437 trace_sk_data_ready(sk); 1438 1439 xprt = xprt_from_sock(sk); 1440 if (xprt != NULL) { 1441 struct sock_xprt *transport = container_of(xprt, 1442 struct sock_xprt, xprt); 1443 1444 trace_xs_data_ready(xprt); 1445 1446 transport->old_data_ready(sk); 1447 1448 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state)) 1449 return; 1450 1451 /* Any data means we had a useful conversation, so 1452 * then we don't need to delay the next reconnect 1453 */ 1454 if (xprt->reestablish_timeout) 1455 xprt->reestablish_timeout = 0; 1456 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1457 queue_work(xprtiod_workqueue, &transport->recv_worker); 1458 } 1459 } 1460 1461 /* 1462 * Helper function to force a TCP close if the server is sending 1463 * junk and/or it has put us in CLOSE_WAIT 1464 */ 1465 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1466 { 1467 xprt_force_disconnect(xprt); 1468 } 1469 1470 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1471 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt) 1472 { 1473 return PAGE_SIZE; 1474 } 1475 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1476 1477 /** 1478 * xs_local_state_change - callback to handle AF_LOCAL socket state changes 1479 * @sk: socket whose state has changed 1480 * 1481 */ 1482 static void xs_local_state_change(struct sock *sk) 1483 { 1484 struct rpc_xprt *xprt; 1485 struct sock_xprt *transport; 1486 1487 if (!(xprt = xprt_from_sock(sk))) 1488 return; 1489 transport = container_of(xprt, struct sock_xprt, xprt); 1490 if (sk->sk_shutdown & SHUTDOWN_MASK) { 1491 clear_bit(XPRT_CONNECTED, &xprt->state); 1492 /* Trigger the socket release */ 1493 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1494 } 1495 } 1496 1497 /** 1498 * xs_tcp_state_change - callback to handle TCP socket state changes 1499 * @sk: socket whose state has changed 1500 * 1501 */ 1502 static void xs_tcp_state_change(struct sock *sk) 1503 { 1504 struct rpc_xprt *xprt; 1505 struct sock_xprt *transport; 1506 1507 if (!(xprt = xprt_from_sock(sk))) 1508 return; 1509 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1510 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1511 sk->sk_state, xprt_connected(xprt), 1512 sock_flag(sk, SOCK_DEAD), 1513 sock_flag(sk, SOCK_ZAPPED), 1514 sk->sk_shutdown); 1515 1516 transport = container_of(xprt, struct sock_xprt, xprt); 1517 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1518 switch (sk->sk_state) { 1519 case TCP_ESTABLISHED: 1520 if (!xprt_test_and_set_connected(xprt)) { 1521 xprt->connect_cookie++; 1522 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1523 xprt_clear_connecting(xprt); 1524 1525 xprt->stat.connect_count++; 1526 xprt->stat.connect_time += (long)jiffies - 1527 xprt->stat.connect_start; 1528 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING); 1529 } 1530 break; 1531 case TCP_FIN_WAIT1: 1532 /* The client initiated a shutdown of the socket */ 1533 xprt->connect_cookie++; 1534 xprt->reestablish_timeout = 0; 1535 set_bit(XPRT_CLOSING, &xprt->state); 1536 smp_mb__before_atomic(); 1537 clear_bit(XPRT_CONNECTED, &xprt->state); 1538 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1539 smp_mb__after_atomic(); 1540 break; 1541 case TCP_CLOSE_WAIT: 1542 /* The server initiated a shutdown of the socket */ 1543 xprt->connect_cookie++; 1544 clear_bit(XPRT_CONNECTED, &xprt->state); 1545 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1546 fallthrough; 1547 case TCP_CLOSING: 1548 /* 1549 * If the server closed down the connection, make sure that 1550 * we back off before reconnecting 1551 */ 1552 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1553 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1554 break; 1555 case TCP_LAST_ACK: 1556 set_bit(XPRT_CLOSING, &xprt->state); 1557 smp_mb__before_atomic(); 1558 clear_bit(XPRT_CONNECTED, &xprt->state); 1559 smp_mb__after_atomic(); 1560 break; 1561 case TCP_CLOSE: 1562 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1563 &transport->sock_state)) 1564 xprt_clear_connecting(xprt); 1565 clear_bit(XPRT_CLOSING, &xprt->state); 1566 /* Trigger the socket release */ 1567 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1568 } 1569 } 1570 1571 static void xs_write_space(struct sock *sk) 1572 { 1573 struct sock_xprt *transport; 1574 struct rpc_xprt *xprt; 1575 1576 if (!sk->sk_socket) 1577 return; 1578 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1579 1580 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1581 return; 1582 transport = container_of(xprt, struct sock_xprt, xprt); 1583 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state)) 1584 return; 1585 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE); 1586 sk->sk_write_pending--; 1587 } 1588 1589 /** 1590 * xs_udp_write_space - callback invoked when socket buffer space 1591 * becomes available 1592 * @sk: socket whose state has changed 1593 * 1594 * Called when more output buffer space is available for this socket. 1595 * We try not to wake our writers until they can make "significant" 1596 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1597 * with a bunch of small requests. 1598 */ 1599 static void xs_udp_write_space(struct sock *sk) 1600 { 1601 /* from net/core/sock.c:sock_def_write_space */ 1602 if (sock_writeable(sk)) 1603 xs_write_space(sk); 1604 } 1605 1606 /** 1607 * xs_tcp_write_space - callback invoked when socket buffer space 1608 * becomes available 1609 * @sk: socket whose state has changed 1610 * 1611 * Called when more output buffer space is available for this socket. 1612 * We try not to wake our writers until they can make "significant" 1613 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1614 * with a bunch of small requests. 1615 */ 1616 static void xs_tcp_write_space(struct sock *sk) 1617 { 1618 /* from net/core/stream.c:sk_stream_write_space */ 1619 if (sk_stream_is_writeable(sk)) 1620 xs_write_space(sk); 1621 } 1622 1623 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1624 { 1625 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1626 struct sock *sk = transport->inet; 1627 1628 if (transport->rcvsize) { 1629 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1630 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1631 } 1632 if (transport->sndsize) { 1633 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1634 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1635 sk->sk_write_space(sk); 1636 } 1637 } 1638 1639 /** 1640 * xs_udp_set_buffer_size - set send and receive limits 1641 * @xprt: generic transport 1642 * @sndsize: requested size of send buffer, in bytes 1643 * @rcvsize: requested size of receive buffer, in bytes 1644 * 1645 * Set socket send and receive buffer size limits. 1646 */ 1647 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1648 { 1649 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1650 1651 transport->sndsize = 0; 1652 if (sndsize) 1653 transport->sndsize = sndsize + 1024; 1654 transport->rcvsize = 0; 1655 if (rcvsize) 1656 transport->rcvsize = rcvsize + 1024; 1657 1658 xs_udp_do_set_buffer_size(xprt); 1659 } 1660 1661 /** 1662 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1663 * @xprt: controlling transport 1664 * @task: task that timed out 1665 * 1666 * Adjust the congestion window after a retransmit timeout has occurred. 1667 */ 1668 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1669 { 1670 spin_lock(&xprt->transport_lock); 1671 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1672 spin_unlock(&xprt->transport_lock); 1673 } 1674 1675 static int xs_get_random_port(void) 1676 { 1677 unsigned short min = xprt_min_resvport, max = xprt_max_resvport; 1678 unsigned short range; 1679 unsigned short rand; 1680 1681 if (max < min) 1682 return -EADDRINUSE; 1683 range = max - min + 1; 1684 rand = get_random_u32_below(range); 1685 return rand + min; 1686 } 1687 1688 static unsigned short xs_sock_getport(struct socket *sock) 1689 { 1690 struct sockaddr_storage buf; 1691 unsigned short port = 0; 1692 1693 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0) 1694 goto out; 1695 switch (buf.ss_family) { 1696 case AF_INET6: 1697 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1698 break; 1699 case AF_INET: 1700 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1701 } 1702 out: 1703 return port; 1704 } 1705 1706 /** 1707 * xs_set_port - reset the port number in the remote endpoint address 1708 * @xprt: generic transport 1709 * @port: new port number 1710 * 1711 */ 1712 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1713 { 1714 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1715 1716 rpc_set_port(xs_addr(xprt), port); 1717 xs_update_peer_port(xprt); 1718 } 1719 1720 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1721 { 1722 if (transport->srcport == 0 && transport->xprt.reuseport) 1723 transport->srcport = xs_sock_getport(sock); 1724 } 1725 1726 static int xs_get_srcport(struct sock_xprt *transport) 1727 { 1728 int port = transport->srcport; 1729 1730 if (port == 0 && transport->xprt.resvport) 1731 port = xs_get_random_port(); 1732 return port; 1733 } 1734 1735 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt) 1736 { 1737 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1738 unsigned short ret = 0; 1739 mutex_lock(&sock->recv_mutex); 1740 if (sock->sock) 1741 ret = xs_sock_getport(sock->sock); 1742 mutex_unlock(&sock->recv_mutex); 1743 return ret; 1744 } 1745 1746 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen) 1747 { 1748 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1749 union { 1750 struct sockaddr sa; 1751 struct sockaddr_storage st; 1752 } saddr; 1753 int ret = -ENOTCONN; 1754 1755 mutex_lock(&sock->recv_mutex); 1756 if (sock->sock) { 1757 ret = kernel_getsockname(sock->sock, &saddr.sa); 1758 if (ret >= 0) 1759 ret = snprintf(buf, buflen, "%pISc", &saddr.sa); 1760 } 1761 mutex_unlock(&sock->recv_mutex); 1762 return ret; 1763 } 1764 1765 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1766 { 1767 if (transport->srcport != 0) 1768 transport->srcport = 0; 1769 if (!transport->xprt.resvport) 1770 return 0; 1771 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1772 return xprt_max_resvport; 1773 return --port; 1774 } 1775 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1776 { 1777 struct sockaddr_storage myaddr; 1778 int err, nloop = 0; 1779 int port = xs_get_srcport(transport); 1780 unsigned short last; 1781 1782 /* 1783 * If we are asking for any ephemeral port (i.e. port == 0 && 1784 * transport->xprt.resvport == 0), don't bind. Let the local 1785 * port selection happen implicitly when the socket is used 1786 * (for example at connect time). 1787 * 1788 * This ensures that we can continue to establish TCP 1789 * connections even when all local ephemeral ports are already 1790 * a part of some TCP connection. This makes no difference 1791 * for UDP sockets, but also doesn't harm them. 1792 * 1793 * If we're asking for any reserved port (i.e. port == 0 && 1794 * transport->xprt.resvport == 1) xs_get_srcport above will 1795 * ensure that port is non-zero and we will bind as needed. 1796 */ 1797 if (port <= 0) 1798 return port; 1799 1800 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1801 do { 1802 rpc_set_port((struct sockaddr *)&myaddr, port); 1803 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1804 transport->xprt.addrlen); 1805 if (err == 0) { 1806 if (transport->xprt.reuseport) 1807 transport->srcport = port; 1808 break; 1809 } 1810 last = port; 1811 port = xs_next_srcport(transport, port); 1812 if (port > last) 1813 nloop++; 1814 } while (err == -EADDRINUSE && nloop != 2); 1815 1816 if (myaddr.ss_family == AF_INET) 1817 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1818 &((struct sockaddr_in *)&myaddr)->sin_addr, 1819 port, err ? "failed" : "ok", err); 1820 else 1821 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1822 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1823 port, err ? "failed" : "ok", err); 1824 return err; 1825 } 1826 1827 /* 1828 * We don't support autobind on AF_LOCAL sockets 1829 */ 1830 static void xs_local_rpcbind(struct rpc_task *task) 1831 { 1832 xprt_set_bound(task->tk_xprt); 1833 } 1834 1835 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1836 { 1837 } 1838 1839 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1840 static struct lock_class_key xs_key[3]; 1841 static struct lock_class_key xs_slock_key[3]; 1842 1843 static inline void xs_reclassify_socketu(struct socket *sock) 1844 { 1845 struct sock *sk = sock->sk; 1846 1847 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1848 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]); 1849 } 1850 1851 static inline void xs_reclassify_socket4(struct socket *sock) 1852 { 1853 struct sock *sk = sock->sk; 1854 1855 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1856 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]); 1857 } 1858 1859 static inline void xs_reclassify_socket6(struct socket *sock) 1860 { 1861 struct sock *sk = sock->sk; 1862 1863 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1864 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]); 1865 } 1866 1867 static inline void xs_reclassify_socket(int family, struct socket *sock) 1868 { 1869 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk))) 1870 return; 1871 1872 switch (family) { 1873 case AF_LOCAL: 1874 xs_reclassify_socketu(sock); 1875 break; 1876 case AF_INET: 1877 xs_reclassify_socket4(sock); 1878 break; 1879 case AF_INET6: 1880 xs_reclassify_socket6(sock); 1881 break; 1882 } 1883 } 1884 #else 1885 static inline void xs_reclassify_socket(int family, struct socket *sock) 1886 { 1887 } 1888 #endif 1889 1890 static void xs_dummy_setup_socket(struct work_struct *work) 1891 { 1892 } 1893 1894 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1895 struct sock_xprt *transport, int family, int type, 1896 int protocol, bool reuseport) 1897 { 1898 struct file *filp; 1899 struct socket *sock; 1900 int err; 1901 1902 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1903 if (err < 0) { 1904 dprintk("RPC: can't create %d transport socket (%d).\n", 1905 protocol, -err); 1906 goto out; 1907 } 1908 xs_reclassify_socket(family, sock); 1909 1910 if (reuseport) 1911 sock_set_reuseport(sock->sk); 1912 1913 err = xs_bind(transport, sock); 1914 if (err) { 1915 sock_release(sock); 1916 goto out; 1917 } 1918 1919 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1920 if (IS_ERR(filp)) 1921 return ERR_CAST(filp); 1922 transport->file = filp; 1923 1924 return sock; 1925 out: 1926 return ERR_PTR(err); 1927 } 1928 1929 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1930 struct socket *sock) 1931 { 1932 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1933 xprt); 1934 1935 if (!transport->inet) { 1936 struct sock *sk = sock->sk; 1937 1938 lock_sock(sk); 1939 1940 xs_save_old_callbacks(transport, sk); 1941 1942 sk->sk_user_data = xprt; 1943 sk->sk_data_ready = xs_data_ready; 1944 sk->sk_write_space = xs_udp_write_space; 1945 sk->sk_state_change = xs_local_state_change; 1946 sk->sk_error_report = xs_error_report; 1947 sk->sk_use_task_frag = false; 1948 1949 xprt_clear_connected(xprt); 1950 1951 /* Reset to new socket */ 1952 transport->sock = sock; 1953 transport->inet = sk; 1954 1955 release_sock(sk); 1956 } 1957 1958 xs_stream_start_connect(transport); 1959 1960 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1961 } 1962 1963 /** 1964 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1965 * @transport: socket transport to connect 1966 */ 1967 static int xs_local_setup_socket(struct sock_xprt *transport) 1968 { 1969 struct rpc_xprt *xprt = &transport->xprt; 1970 struct file *filp; 1971 struct socket *sock; 1972 int status; 1973 1974 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1975 SOCK_STREAM, 0, &sock, 1); 1976 if (status < 0) { 1977 dprintk("RPC: can't create AF_LOCAL " 1978 "transport socket (%d).\n", -status); 1979 goto out; 1980 } 1981 xs_reclassify_socket(AF_LOCAL, sock); 1982 1983 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1984 if (IS_ERR(filp)) { 1985 status = PTR_ERR(filp); 1986 goto out; 1987 } 1988 transport->file = filp; 1989 1990 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1991 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1992 1993 status = xs_local_finish_connecting(xprt, sock); 1994 trace_rpc_socket_connect(xprt, sock, status); 1995 switch (status) { 1996 case 0: 1997 dprintk("RPC: xprt %p connected to %s\n", 1998 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1999 xprt->stat.connect_count++; 2000 xprt->stat.connect_time += (long)jiffies - 2001 xprt->stat.connect_start; 2002 xprt_set_connected(xprt); 2003 break; 2004 case -ENOBUFS: 2005 break; 2006 case -ENOENT: 2007 dprintk("RPC: xprt %p: socket %s does not exist\n", 2008 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2009 break; 2010 case -ECONNREFUSED: 2011 dprintk("RPC: xprt %p: connection refused for %s\n", 2012 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2013 break; 2014 default: 2015 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 2016 __func__, -status, 2017 xprt->address_strings[RPC_DISPLAY_ADDR]); 2018 } 2019 2020 out: 2021 xprt_clear_connecting(xprt); 2022 xprt_wake_pending_tasks(xprt, status); 2023 return status; 2024 } 2025 2026 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2027 { 2028 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2029 int ret; 2030 2031 if (transport->file) 2032 goto force_disconnect; 2033 2034 if (RPC_IS_ASYNC(task)) { 2035 /* 2036 * We want the AF_LOCAL connect to be resolved in the 2037 * filesystem namespace of the process making the rpc 2038 * call. Thus we connect synchronously. 2039 * 2040 * If we want to support asynchronous AF_LOCAL calls, 2041 * we'll need to figure out how to pass a namespace to 2042 * connect. 2043 */ 2044 rpc_task_set_rpc_status(task, -ENOTCONN); 2045 goto out_wake; 2046 } 2047 ret = xs_local_setup_socket(transport); 2048 if (ret && !RPC_IS_SOFTCONN(task)) 2049 msleep_interruptible(15000); 2050 return; 2051 force_disconnect: 2052 xprt_force_disconnect(xprt); 2053 out_wake: 2054 xprt_clear_connecting(xprt); 2055 xprt_wake_pending_tasks(xprt, -ENOTCONN); 2056 } 2057 2058 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 2059 /* 2060 * Note that this should be called with XPRT_LOCKED held, or recv_mutex 2061 * held, or when we otherwise know that we have exclusive access to the 2062 * socket, to guard against races with xs_reset_transport. 2063 */ 2064 static void xs_set_memalloc(struct rpc_xprt *xprt) 2065 { 2066 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2067 xprt); 2068 2069 /* 2070 * If there's no sock, then we have nothing to set. The 2071 * reconnecting process will get it for us. 2072 */ 2073 if (!transport->inet) 2074 return; 2075 if (atomic_read(&xprt->swapper)) 2076 sk_set_memalloc(transport->inet); 2077 } 2078 2079 /** 2080 * xs_enable_swap - Tag this transport as being used for swap. 2081 * @xprt: transport to tag 2082 * 2083 * Take a reference to this transport on behalf of the rpc_clnt, and 2084 * optionally mark it for swapping if it wasn't already. 2085 */ 2086 static int 2087 xs_enable_swap(struct rpc_xprt *xprt) 2088 { 2089 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2090 2091 mutex_lock(&xs->recv_mutex); 2092 if (atomic_inc_return(&xprt->swapper) == 1 && 2093 xs->inet) 2094 sk_set_memalloc(xs->inet); 2095 mutex_unlock(&xs->recv_mutex); 2096 return 0; 2097 } 2098 2099 /** 2100 * xs_disable_swap - Untag this transport as being used for swap. 2101 * @xprt: transport to tag 2102 * 2103 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2104 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2105 */ 2106 static void 2107 xs_disable_swap(struct rpc_xprt *xprt) 2108 { 2109 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2110 2111 mutex_lock(&xs->recv_mutex); 2112 if (atomic_dec_and_test(&xprt->swapper) && 2113 xs->inet) 2114 sk_clear_memalloc(xs->inet); 2115 mutex_unlock(&xs->recv_mutex); 2116 } 2117 #else 2118 static void xs_set_memalloc(struct rpc_xprt *xprt) 2119 { 2120 } 2121 2122 static int 2123 xs_enable_swap(struct rpc_xprt *xprt) 2124 { 2125 return -EINVAL; 2126 } 2127 2128 static void 2129 xs_disable_swap(struct rpc_xprt *xprt) 2130 { 2131 } 2132 #endif 2133 2134 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2135 { 2136 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2137 2138 if (!transport->inet) { 2139 struct sock *sk = sock->sk; 2140 2141 lock_sock(sk); 2142 2143 xs_save_old_callbacks(transport, sk); 2144 2145 sk->sk_user_data = xprt; 2146 sk->sk_data_ready = xs_data_ready; 2147 sk->sk_write_space = xs_udp_write_space; 2148 sk->sk_use_task_frag = false; 2149 2150 xprt_set_connected(xprt); 2151 2152 /* Reset to new socket */ 2153 transport->sock = sock; 2154 transport->inet = sk; 2155 2156 xs_set_memalloc(xprt); 2157 2158 release_sock(sk); 2159 } 2160 xs_udp_do_set_buffer_size(xprt); 2161 2162 xprt->stat.connect_start = jiffies; 2163 } 2164 2165 static void xs_udp_setup_socket(struct work_struct *work) 2166 { 2167 struct sock_xprt *transport = 2168 container_of(work, struct sock_xprt, connect_worker.work); 2169 struct rpc_xprt *xprt = &transport->xprt; 2170 struct socket *sock; 2171 int status = -EIO; 2172 unsigned int pflags = current->flags; 2173 2174 if (atomic_read(&xprt->swapper)) 2175 current->flags |= PF_MEMALLOC; 2176 sock = xs_create_sock(xprt, transport, 2177 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2178 IPPROTO_UDP, false); 2179 if (IS_ERR(sock)) 2180 goto out; 2181 2182 dprintk("RPC: worker connecting xprt %p via %s to " 2183 "%s (port %s)\n", xprt, 2184 xprt->address_strings[RPC_DISPLAY_PROTO], 2185 xprt->address_strings[RPC_DISPLAY_ADDR], 2186 xprt->address_strings[RPC_DISPLAY_PORT]); 2187 2188 xs_udp_finish_connecting(xprt, sock); 2189 trace_rpc_socket_connect(xprt, sock, 0); 2190 status = 0; 2191 out: 2192 xprt_clear_connecting(xprt); 2193 xprt_unlock_connect(xprt, transport); 2194 xprt_wake_pending_tasks(xprt, status); 2195 current_restore_flags(pflags, PF_MEMALLOC); 2196 } 2197 2198 /** 2199 * xs_tcp_shutdown - gracefully shut down a TCP socket 2200 * @xprt: transport 2201 * 2202 * Initiates a graceful shutdown of the TCP socket by calling the 2203 * equivalent of shutdown(SHUT_RDWR); 2204 */ 2205 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2206 { 2207 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2208 struct socket *sock = transport->sock; 2209 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE; 2210 2211 if (sock == NULL) 2212 return; 2213 if (!xprt->reuseport) { 2214 xs_close(xprt); 2215 return; 2216 } 2217 switch (skst) { 2218 case TCP_FIN_WAIT1: 2219 case TCP_FIN_WAIT2: 2220 case TCP_LAST_ACK: 2221 break; 2222 case TCP_ESTABLISHED: 2223 case TCP_CLOSE_WAIT: 2224 kernel_sock_shutdown(sock, SHUT_RDWR); 2225 trace_rpc_socket_shutdown(xprt, sock); 2226 break; 2227 default: 2228 xs_reset_transport(transport); 2229 } 2230 } 2231 2232 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 2233 struct socket *sock) 2234 { 2235 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2236 unsigned int keepidle; 2237 unsigned int keepcnt; 2238 unsigned int timeo; 2239 2240 spin_lock(&xprt->transport_lock); 2241 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ); 2242 keepcnt = xprt->timeout->to_retries + 1; 2243 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2244 (xprt->timeout->to_retries + 1); 2245 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2246 spin_unlock(&xprt->transport_lock); 2247 2248 /* TCP Keepalive options */ 2249 sock_set_keepalive(sock->sk); 2250 tcp_sock_set_keepidle(sock->sk, keepidle); 2251 tcp_sock_set_keepintvl(sock->sk, keepidle); 2252 tcp_sock_set_keepcnt(sock->sk, keepcnt); 2253 2254 /* TCP user timeout (see RFC5482) */ 2255 tcp_sock_set_user_timeout(sock->sk, timeo); 2256 } 2257 2258 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt, 2259 unsigned long connect_timeout, 2260 unsigned long reconnect_timeout) 2261 { 2262 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2263 struct rpc_timeout to; 2264 unsigned long initval; 2265 2266 spin_lock(&xprt->transport_lock); 2267 if (reconnect_timeout < xprt->max_reconnect_timeout) 2268 xprt->max_reconnect_timeout = reconnect_timeout; 2269 if (connect_timeout < xprt->connect_timeout) { 2270 memcpy(&to, xprt->timeout, sizeof(to)); 2271 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1); 2272 /* Arbitrary lower limit */ 2273 if (initval < XS_TCP_INIT_REEST_TO << 1) 2274 initval = XS_TCP_INIT_REEST_TO << 1; 2275 to.to_initval = initval; 2276 to.to_maxval = initval; 2277 memcpy(&transport->tcp_timeout, &to, 2278 sizeof(transport->tcp_timeout)); 2279 xprt->timeout = &transport->tcp_timeout; 2280 xprt->connect_timeout = connect_timeout; 2281 } 2282 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2283 spin_unlock(&xprt->transport_lock); 2284 } 2285 2286 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2287 { 2288 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2289 2290 if (!transport->inet) { 2291 struct sock *sk = sock->sk; 2292 2293 /* Avoid temporary address, they are bad for long-lived 2294 * connections such as NFS mounts. 2295 * RFC4941, section 3.6 suggests that: 2296 * Individual applications, which have specific 2297 * knowledge about the normal duration of connections, 2298 * MAY override this as appropriate. 2299 */ 2300 if (xs_addr(xprt)->sa_family == PF_INET6) { 2301 ip6_sock_set_addr_preferences(sk, 2302 IPV6_PREFER_SRC_PUBLIC); 2303 } 2304 2305 xs_tcp_set_socket_timeouts(xprt, sock); 2306 tcp_sock_set_nodelay(sk); 2307 2308 lock_sock(sk); 2309 2310 xs_save_old_callbacks(transport, sk); 2311 2312 sk->sk_user_data = xprt; 2313 sk->sk_data_ready = xs_data_ready; 2314 sk->sk_state_change = xs_tcp_state_change; 2315 sk->sk_write_space = xs_tcp_write_space; 2316 sk->sk_error_report = xs_error_report; 2317 sk->sk_use_task_frag = false; 2318 2319 /* socket options */ 2320 sock_reset_flag(sk, SOCK_LINGER); 2321 2322 xprt_clear_connected(xprt); 2323 2324 /* Reset to new socket */ 2325 transport->sock = sock; 2326 transport->inet = sk; 2327 2328 release_sock(sk); 2329 } 2330 2331 if (!xprt_bound(xprt)) 2332 return -ENOTCONN; 2333 2334 xs_set_memalloc(xprt); 2335 2336 xs_stream_start_connect(transport); 2337 2338 /* Tell the socket layer to start connecting... */ 2339 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2340 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2341 } 2342 2343 /** 2344 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2345 * @work: queued work item 2346 * 2347 * Invoked by a work queue tasklet. 2348 */ 2349 static void xs_tcp_setup_socket(struct work_struct *work) 2350 { 2351 struct sock_xprt *transport = 2352 container_of(work, struct sock_xprt, connect_worker.work); 2353 struct socket *sock = transport->sock; 2354 struct rpc_xprt *xprt = &transport->xprt; 2355 int status; 2356 unsigned int pflags = current->flags; 2357 2358 if (atomic_read(&xprt->swapper)) 2359 current->flags |= PF_MEMALLOC; 2360 2361 if (xprt_connected(xprt)) 2362 goto out; 2363 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT, 2364 &transport->sock_state) || 2365 !sock) { 2366 xs_reset_transport(transport); 2367 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family, 2368 SOCK_STREAM, IPPROTO_TCP, true); 2369 if (IS_ERR(sock)) { 2370 xprt_wake_pending_tasks(xprt, PTR_ERR(sock)); 2371 goto out; 2372 } 2373 } 2374 2375 dprintk("RPC: worker connecting xprt %p via %s to " 2376 "%s (port %s)\n", xprt, 2377 xprt->address_strings[RPC_DISPLAY_PROTO], 2378 xprt->address_strings[RPC_DISPLAY_ADDR], 2379 xprt->address_strings[RPC_DISPLAY_PORT]); 2380 2381 status = xs_tcp_finish_connecting(xprt, sock); 2382 trace_rpc_socket_connect(xprt, sock, status); 2383 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2384 xprt, -status, xprt_connected(xprt), 2385 sock->sk->sk_state); 2386 switch (status) { 2387 case 0: 2388 case -EINPROGRESS: 2389 /* SYN_SENT! */ 2390 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state); 2391 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2392 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2393 fallthrough; 2394 case -EALREADY: 2395 goto out_unlock; 2396 case -EADDRNOTAVAIL: 2397 /* Source port number is unavailable. Try a new one! */ 2398 transport->srcport = 0; 2399 status = -EAGAIN; 2400 break; 2401 case -EINVAL: 2402 /* Happens, for instance, if the user specified a link 2403 * local IPv6 address without a scope-id. 2404 */ 2405 case -ECONNREFUSED: 2406 case -ECONNRESET: 2407 case -ENETDOWN: 2408 case -ENETUNREACH: 2409 case -EHOSTUNREACH: 2410 case -EADDRINUSE: 2411 case -ENOBUFS: 2412 break; 2413 default: 2414 printk("%s: connect returned unhandled error %d\n", 2415 __func__, status); 2416 status = -EAGAIN; 2417 } 2418 2419 /* xs_tcp_force_close() wakes tasks with a fixed error code. 2420 * We need to wake them first to ensure the correct error code. 2421 */ 2422 xprt_wake_pending_tasks(xprt, status); 2423 xs_tcp_force_close(xprt); 2424 out: 2425 xprt_clear_connecting(xprt); 2426 out_unlock: 2427 xprt_unlock_connect(xprt, transport); 2428 current_restore_flags(pflags, PF_MEMALLOC); 2429 } 2430 2431 /* 2432 * Transfer the connected socket to @upper_transport, then mark that 2433 * xprt CONNECTED. 2434 */ 2435 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt, 2436 struct sock_xprt *upper_transport) 2437 { 2438 struct sock_xprt *lower_transport = 2439 container_of(lower_xprt, struct sock_xprt, xprt); 2440 struct rpc_xprt *upper_xprt = &upper_transport->xprt; 2441 2442 if (!upper_transport->inet) { 2443 struct socket *sock = lower_transport->sock; 2444 struct sock *sk = sock->sk; 2445 2446 /* Avoid temporary address, they are bad for long-lived 2447 * connections such as NFS mounts. 2448 * RFC4941, section 3.6 suggests that: 2449 * Individual applications, which have specific 2450 * knowledge about the normal duration of connections, 2451 * MAY override this as appropriate. 2452 */ 2453 if (xs_addr(upper_xprt)->sa_family == PF_INET6) 2454 ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC); 2455 2456 xs_tcp_set_socket_timeouts(upper_xprt, sock); 2457 tcp_sock_set_nodelay(sk); 2458 2459 lock_sock(sk); 2460 2461 /* @sk is already connected, so it now has the RPC callbacks. 2462 * Reach into @lower_transport to save the original ones. 2463 */ 2464 upper_transport->old_data_ready = lower_transport->old_data_ready; 2465 upper_transport->old_state_change = lower_transport->old_state_change; 2466 upper_transport->old_write_space = lower_transport->old_write_space; 2467 upper_transport->old_error_report = lower_transport->old_error_report; 2468 sk->sk_user_data = upper_xprt; 2469 2470 /* socket options */ 2471 sock_reset_flag(sk, SOCK_LINGER); 2472 2473 xprt_clear_connected(upper_xprt); 2474 2475 upper_transport->sock = sock; 2476 upper_transport->inet = sk; 2477 upper_transport->file = lower_transport->file; 2478 2479 release_sock(sk); 2480 2481 /* Reset lower_transport before shutting down its clnt */ 2482 mutex_lock(&lower_transport->recv_mutex); 2483 lower_transport->inet = NULL; 2484 lower_transport->sock = NULL; 2485 lower_transport->file = NULL; 2486 2487 xprt_clear_connected(lower_xprt); 2488 xs_sock_reset_connection_flags(lower_xprt); 2489 xs_stream_reset_connect(lower_transport); 2490 mutex_unlock(&lower_transport->recv_mutex); 2491 } 2492 2493 if (!xprt_bound(upper_xprt)) 2494 return -ENOTCONN; 2495 2496 xs_set_memalloc(upper_xprt); 2497 2498 if (!xprt_test_and_set_connected(upper_xprt)) { 2499 upper_xprt->connect_cookie++; 2500 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state); 2501 xprt_clear_connecting(upper_xprt); 2502 2503 upper_xprt->stat.connect_count++; 2504 upper_xprt->stat.connect_time += (long)jiffies - 2505 upper_xprt->stat.connect_start; 2506 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING); 2507 } 2508 return 0; 2509 } 2510 2511 /** 2512 * xs_tls_handshake_done - TLS handshake completion handler 2513 * @data: address of xprt to wake 2514 * @status: status of handshake 2515 * @peerid: serial number of key containing the remote's identity 2516 * 2517 */ 2518 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid) 2519 { 2520 struct rpc_xprt *lower_xprt = data; 2521 struct sock_xprt *lower_transport = 2522 container_of(lower_xprt, struct sock_xprt, xprt); 2523 2524 lower_transport->xprt_err = status ? -EACCES : 0; 2525 complete(&lower_transport->handshake_done); 2526 xprt_put(lower_xprt); 2527 } 2528 2529 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec) 2530 { 2531 struct sock_xprt *lower_transport = 2532 container_of(lower_xprt, struct sock_xprt, xprt); 2533 struct tls_handshake_args args = { 2534 .ta_sock = lower_transport->sock, 2535 .ta_done = xs_tls_handshake_done, 2536 .ta_data = xprt_get(lower_xprt), 2537 .ta_peername = lower_xprt->servername, 2538 }; 2539 struct sock *sk = lower_transport->inet; 2540 int rc; 2541 2542 init_completion(&lower_transport->handshake_done); 2543 set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state); 2544 lower_transport->xprt_err = -ETIMEDOUT; 2545 switch (xprtsec->policy) { 2546 case RPC_XPRTSEC_TLS_ANON: 2547 rc = tls_client_hello_anon(&args, GFP_KERNEL); 2548 if (rc) 2549 goto out_put_xprt; 2550 break; 2551 case RPC_XPRTSEC_TLS_X509: 2552 args.ta_my_cert = xprtsec->cert_serial; 2553 args.ta_my_privkey = xprtsec->privkey_serial; 2554 rc = tls_client_hello_x509(&args, GFP_KERNEL); 2555 if (rc) 2556 goto out_put_xprt; 2557 break; 2558 default: 2559 rc = -EACCES; 2560 goto out_put_xprt; 2561 } 2562 2563 rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done, 2564 XS_TLS_HANDSHAKE_TO); 2565 if (rc <= 0) { 2566 if (!tls_handshake_cancel(sk)) { 2567 if (rc == 0) 2568 rc = -ETIMEDOUT; 2569 goto out_put_xprt; 2570 } 2571 } 2572 2573 rc = lower_transport->xprt_err; 2574 2575 out: 2576 xs_stream_reset_connect(lower_transport); 2577 clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state); 2578 return rc; 2579 2580 out_put_xprt: 2581 xprt_put(lower_xprt); 2582 goto out; 2583 } 2584 2585 /** 2586 * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket 2587 * @work: queued work item 2588 * 2589 * Invoked by a work queue tasklet. 2590 * 2591 * For RPC-with-TLS, there is a two-stage connection process. 2592 * 2593 * The "upper-layer xprt" is visible to the RPC consumer. Once it has 2594 * been marked connected, the consumer knows that a TCP connection and 2595 * a TLS session have been established. 2596 * 2597 * A "lower-layer xprt", created in this function, handles the mechanics 2598 * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and 2599 * then driving the TLS handshake. Once all that is complete, the upper 2600 * layer xprt is marked connected. 2601 */ 2602 static void xs_tcp_tls_setup_socket(struct work_struct *work) 2603 { 2604 struct sock_xprt *upper_transport = 2605 container_of(work, struct sock_xprt, connect_worker.work); 2606 struct rpc_clnt *upper_clnt = upper_transport->clnt; 2607 struct rpc_xprt *upper_xprt = &upper_transport->xprt; 2608 struct rpc_create_args args = { 2609 .net = upper_xprt->xprt_net, 2610 .protocol = upper_xprt->prot, 2611 .address = (struct sockaddr *)&upper_xprt->addr, 2612 .addrsize = upper_xprt->addrlen, 2613 .timeout = upper_clnt->cl_timeout, 2614 .servername = upper_xprt->servername, 2615 .program = upper_clnt->cl_program, 2616 .prognumber = upper_clnt->cl_prog, 2617 .version = upper_clnt->cl_vers, 2618 .authflavor = RPC_AUTH_TLS, 2619 .cred = upper_clnt->cl_cred, 2620 .xprtsec = { 2621 .policy = RPC_XPRTSEC_NONE, 2622 }, 2623 }; 2624 unsigned int pflags = current->flags; 2625 struct rpc_clnt *lower_clnt; 2626 struct rpc_xprt *lower_xprt; 2627 int status; 2628 2629 if (atomic_read(&upper_xprt->swapper)) 2630 current->flags |= PF_MEMALLOC; 2631 2632 xs_stream_start_connect(upper_transport); 2633 2634 /* This implicitly sends an RPC_AUTH_TLS probe */ 2635 lower_clnt = rpc_create(&args); 2636 if (IS_ERR(lower_clnt)) { 2637 trace_rpc_tls_unavailable(upper_clnt, upper_xprt); 2638 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state); 2639 xprt_clear_connecting(upper_xprt); 2640 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt)); 2641 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING); 2642 goto out_unlock; 2643 } 2644 2645 /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on 2646 * the lower xprt. 2647 */ 2648 rcu_read_lock(); 2649 lower_xprt = rcu_dereference(lower_clnt->cl_xprt); 2650 rcu_read_unlock(); 2651 status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec); 2652 if (status) { 2653 trace_rpc_tls_not_started(upper_clnt, upper_xprt); 2654 goto out_close; 2655 } 2656 2657 status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport); 2658 if (status) 2659 goto out_close; 2660 2661 trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0); 2662 if (!xprt_test_and_set_connected(upper_xprt)) { 2663 upper_xprt->connect_cookie++; 2664 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state); 2665 xprt_clear_connecting(upper_xprt); 2666 2667 upper_xprt->stat.connect_count++; 2668 upper_xprt->stat.connect_time += (long)jiffies - 2669 upper_xprt->stat.connect_start; 2670 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING); 2671 } 2672 rpc_shutdown_client(lower_clnt); 2673 2674 out_unlock: 2675 current_restore_flags(pflags, PF_MEMALLOC); 2676 upper_transport->clnt = NULL; 2677 xprt_unlock_connect(upper_xprt, upper_transport); 2678 return; 2679 2680 out_close: 2681 rpc_shutdown_client(lower_clnt); 2682 2683 /* xprt_force_disconnect() wakes tasks with a fixed tk_status code. 2684 * Wake them first here to ensure they get our tk_status code. 2685 */ 2686 xprt_wake_pending_tasks(upper_xprt, status); 2687 xs_tcp_force_close(upper_xprt); 2688 xprt_clear_connecting(upper_xprt); 2689 goto out_unlock; 2690 } 2691 2692 /** 2693 * xs_connect - connect a socket to a remote endpoint 2694 * @xprt: pointer to transport structure 2695 * @task: address of RPC task that manages state of connect request 2696 * 2697 * TCP: If the remote end dropped the connection, delay reconnecting. 2698 * 2699 * UDP socket connects are synchronous, but we use a work queue anyway 2700 * to guarantee that even unprivileged user processes can set up a 2701 * socket on a privileged port. 2702 * 2703 * If a UDP socket connect fails, the delay behavior here prevents 2704 * retry floods (hard mounts). 2705 */ 2706 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2707 { 2708 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2709 unsigned long delay = 0; 2710 2711 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2712 2713 if (transport->sock != NULL) { 2714 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2715 "seconds\n", xprt, xprt->reestablish_timeout / HZ); 2716 2717 delay = xprt_reconnect_delay(xprt); 2718 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO); 2719 2720 } else 2721 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2722 2723 transport->clnt = task->tk_client; 2724 queue_delayed_work(xprtiod_workqueue, 2725 &transport->connect_worker, 2726 delay); 2727 } 2728 2729 static void xs_wake_disconnect(struct sock_xprt *transport) 2730 { 2731 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state)) 2732 xs_tcp_force_close(&transport->xprt); 2733 } 2734 2735 static void xs_wake_write(struct sock_xprt *transport) 2736 { 2737 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state)) 2738 xprt_write_space(&transport->xprt); 2739 } 2740 2741 static void xs_wake_error(struct sock_xprt *transport) 2742 { 2743 int sockerr; 2744 2745 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2746 return; 2747 mutex_lock(&transport->recv_mutex); 2748 if (transport->sock == NULL) 2749 goto out; 2750 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2751 goto out; 2752 sockerr = xchg(&transport->xprt_err, 0); 2753 if (sockerr < 0) 2754 xprt_wake_pending_tasks(&transport->xprt, sockerr); 2755 out: 2756 mutex_unlock(&transport->recv_mutex); 2757 } 2758 2759 static void xs_wake_pending(struct sock_xprt *transport) 2760 { 2761 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state)) 2762 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN); 2763 } 2764 2765 static void xs_error_handle(struct work_struct *work) 2766 { 2767 struct sock_xprt *transport = container_of(work, 2768 struct sock_xprt, error_worker); 2769 2770 xs_wake_disconnect(transport); 2771 xs_wake_write(transport); 2772 xs_wake_error(transport); 2773 xs_wake_pending(transport); 2774 } 2775 2776 /** 2777 * xs_local_print_stats - display AF_LOCAL socket-specific stats 2778 * @xprt: rpc_xprt struct containing statistics 2779 * @seq: output file 2780 * 2781 */ 2782 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2783 { 2784 long idle_time = 0; 2785 2786 if (xprt_connected(xprt)) 2787 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2788 2789 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2790 "%llu %llu %lu %llu %llu\n", 2791 xprt->stat.bind_count, 2792 xprt->stat.connect_count, 2793 xprt->stat.connect_time / HZ, 2794 idle_time, 2795 xprt->stat.sends, 2796 xprt->stat.recvs, 2797 xprt->stat.bad_xids, 2798 xprt->stat.req_u, 2799 xprt->stat.bklog_u, 2800 xprt->stat.max_slots, 2801 xprt->stat.sending_u, 2802 xprt->stat.pending_u); 2803 } 2804 2805 /** 2806 * xs_udp_print_stats - display UDP socket-specific stats 2807 * @xprt: rpc_xprt struct containing statistics 2808 * @seq: output file 2809 * 2810 */ 2811 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2812 { 2813 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2814 2815 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2816 "%lu %llu %llu\n", 2817 transport->srcport, 2818 xprt->stat.bind_count, 2819 xprt->stat.sends, 2820 xprt->stat.recvs, 2821 xprt->stat.bad_xids, 2822 xprt->stat.req_u, 2823 xprt->stat.bklog_u, 2824 xprt->stat.max_slots, 2825 xprt->stat.sending_u, 2826 xprt->stat.pending_u); 2827 } 2828 2829 /** 2830 * xs_tcp_print_stats - display TCP socket-specific stats 2831 * @xprt: rpc_xprt struct containing statistics 2832 * @seq: output file 2833 * 2834 */ 2835 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2836 { 2837 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2838 long idle_time = 0; 2839 2840 if (xprt_connected(xprt)) 2841 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2842 2843 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2844 "%llu %llu %lu %llu %llu\n", 2845 transport->srcport, 2846 xprt->stat.bind_count, 2847 xprt->stat.connect_count, 2848 xprt->stat.connect_time / HZ, 2849 idle_time, 2850 xprt->stat.sends, 2851 xprt->stat.recvs, 2852 xprt->stat.bad_xids, 2853 xprt->stat.req_u, 2854 xprt->stat.bklog_u, 2855 xprt->stat.max_slots, 2856 xprt->stat.sending_u, 2857 xprt->stat.pending_u); 2858 } 2859 2860 /* 2861 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2862 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2863 * to use the server side send routines. 2864 */ 2865 static int bc_malloc(struct rpc_task *task) 2866 { 2867 struct rpc_rqst *rqst = task->tk_rqstp; 2868 size_t size = rqst->rq_callsize; 2869 struct page *page; 2870 struct rpc_buffer *buf; 2871 2872 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) { 2873 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n", 2874 size); 2875 return -EINVAL; 2876 } 2877 2878 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN); 2879 if (!page) 2880 return -ENOMEM; 2881 2882 buf = page_address(page); 2883 buf->len = PAGE_SIZE; 2884 2885 rqst->rq_buffer = buf->data; 2886 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; 2887 return 0; 2888 } 2889 2890 /* 2891 * Free the space allocated in the bc_alloc routine 2892 */ 2893 static void bc_free(struct rpc_task *task) 2894 { 2895 void *buffer = task->tk_rqstp->rq_buffer; 2896 struct rpc_buffer *buf; 2897 2898 buf = container_of(buffer, struct rpc_buffer, data); 2899 free_page((unsigned long)buf); 2900 } 2901 2902 static int bc_sendto(struct rpc_rqst *req) 2903 { 2904 struct xdr_buf *xdr = &req->rq_snd_buf; 2905 struct sock_xprt *transport = 2906 container_of(req->rq_xprt, struct sock_xprt, xprt); 2907 struct msghdr msg = { 2908 .msg_flags = 0, 2909 }; 2910 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 2911 (u32)xdr->len); 2912 unsigned int sent = 0; 2913 int err; 2914 2915 req->rq_xtime = ktime_get(); 2916 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask()); 2917 if (err < 0) 2918 return err; 2919 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent); 2920 xdr_free_bvec(xdr); 2921 if (err < 0 || sent != (xdr->len + sizeof(marker))) 2922 return -EAGAIN; 2923 return sent; 2924 } 2925 2926 /** 2927 * bc_send_request - Send a backchannel Call on a TCP socket 2928 * @req: rpc_rqst containing Call message to be sent 2929 * 2930 * xpt_mutex ensures @rqstp's whole message is written to the socket 2931 * without interruption. 2932 * 2933 * Return values: 2934 * %0 if the message was sent successfully 2935 * %ENOTCONN if the message was not sent 2936 */ 2937 static int bc_send_request(struct rpc_rqst *req) 2938 { 2939 struct svc_xprt *xprt; 2940 int len; 2941 2942 /* 2943 * Get the server socket associated with this callback xprt 2944 */ 2945 xprt = req->rq_xprt->bc_xprt; 2946 2947 /* 2948 * Grab the mutex to serialize data as the connection is shared 2949 * with the fore channel 2950 */ 2951 mutex_lock(&xprt->xpt_mutex); 2952 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2953 len = -ENOTCONN; 2954 else 2955 len = bc_sendto(req); 2956 mutex_unlock(&xprt->xpt_mutex); 2957 2958 if (len > 0) 2959 len = 0; 2960 2961 return len; 2962 } 2963 2964 /* 2965 * The close routine. Since this is client initiated, we do nothing 2966 */ 2967 2968 static void bc_close(struct rpc_xprt *xprt) 2969 { 2970 xprt_disconnect_done(xprt); 2971 } 2972 2973 /* 2974 * The xprt destroy routine. Again, because this connection is client 2975 * initiated, we do nothing 2976 */ 2977 2978 static void bc_destroy(struct rpc_xprt *xprt) 2979 { 2980 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2981 2982 xs_xprt_free(xprt); 2983 module_put(THIS_MODULE); 2984 } 2985 2986 static const struct rpc_xprt_ops xs_local_ops = { 2987 .reserve_xprt = xprt_reserve_xprt, 2988 .release_xprt = xprt_release_xprt, 2989 .alloc_slot = xprt_alloc_slot, 2990 .free_slot = xprt_free_slot, 2991 .rpcbind = xs_local_rpcbind, 2992 .set_port = xs_local_set_port, 2993 .connect = xs_local_connect, 2994 .buf_alloc = rpc_malloc, 2995 .buf_free = rpc_free, 2996 .prepare_request = xs_stream_prepare_request, 2997 .send_request = xs_local_send_request, 2998 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2999 .close = xs_close, 3000 .destroy = xs_destroy, 3001 .print_stats = xs_local_print_stats, 3002 .enable_swap = xs_enable_swap, 3003 .disable_swap = xs_disable_swap, 3004 }; 3005 3006 static const struct rpc_xprt_ops xs_udp_ops = { 3007 .set_buffer_size = xs_udp_set_buffer_size, 3008 .reserve_xprt = xprt_reserve_xprt_cong, 3009 .release_xprt = xprt_release_xprt_cong, 3010 .alloc_slot = xprt_alloc_slot, 3011 .free_slot = xprt_free_slot, 3012 .rpcbind = rpcb_getport_async, 3013 .set_port = xs_set_port, 3014 .connect = xs_connect, 3015 .get_srcaddr = xs_sock_srcaddr, 3016 .get_srcport = xs_sock_srcport, 3017 .buf_alloc = rpc_malloc, 3018 .buf_free = rpc_free, 3019 .send_request = xs_udp_send_request, 3020 .wait_for_reply_request = xprt_wait_for_reply_request_rtt, 3021 .timer = xs_udp_timer, 3022 .release_request = xprt_release_rqst_cong, 3023 .close = xs_close, 3024 .destroy = xs_destroy, 3025 .print_stats = xs_udp_print_stats, 3026 .enable_swap = xs_enable_swap, 3027 .disable_swap = xs_disable_swap, 3028 .inject_disconnect = xs_inject_disconnect, 3029 }; 3030 3031 static const struct rpc_xprt_ops xs_tcp_ops = { 3032 .reserve_xprt = xprt_reserve_xprt, 3033 .release_xprt = xprt_release_xprt, 3034 .alloc_slot = xprt_alloc_slot, 3035 .free_slot = xprt_free_slot, 3036 .rpcbind = rpcb_getport_async, 3037 .set_port = xs_set_port, 3038 .connect = xs_connect, 3039 .get_srcaddr = xs_sock_srcaddr, 3040 .get_srcport = xs_sock_srcport, 3041 .buf_alloc = rpc_malloc, 3042 .buf_free = rpc_free, 3043 .prepare_request = xs_stream_prepare_request, 3044 .send_request = xs_tcp_send_request, 3045 .wait_for_reply_request = xprt_wait_for_reply_request_def, 3046 .close = xs_tcp_shutdown, 3047 .destroy = xs_destroy, 3048 .set_connect_timeout = xs_tcp_set_connect_timeout, 3049 .print_stats = xs_tcp_print_stats, 3050 .enable_swap = xs_enable_swap, 3051 .disable_swap = xs_disable_swap, 3052 .inject_disconnect = xs_inject_disconnect, 3053 #ifdef CONFIG_SUNRPC_BACKCHANNEL 3054 .bc_setup = xprt_setup_bc, 3055 .bc_maxpayload = xs_tcp_bc_maxpayload, 3056 .bc_num_slots = xprt_bc_max_slots, 3057 .bc_free_rqst = xprt_free_bc_rqst, 3058 .bc_destroy = xprt_destroy_bc, 3059 #endif 3060 }; 3061 3062 /* 3063 * The rpc_xprt_ops for the server backchannel 3064 */ 3065 3066 static const struct rpc_xprt_ops bc_tcp_ops = { 3067 .reserve_xprt = xprt_reserve_xprt, 3068 .release_xprt = xprt_release_xprt, 3069 .alloc_slot = xprt_alloc_slot, 3070 .free_slot = xprt_free_slot, 3071 .buf_alloc = bc_malloc, 3072 .buf_free = bc_free, 3073 .send_request = bc_send_request, 3074 .wait_for_reply_request = xprt_wait_for_reply_request_def, 3075 .close = bc_close, 3076 .destroy = bc_destroy, 3077 .print_stats = xs_tcp_print_stats, 3078 .enable_swap = xs_enable_swap, 3079 .disable_swap = xs_disable_swap, 3080 .inject_disconnect = xs_inject_disconnect, 3081 }; 3082 3083 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 3084 { 3085 static const struct sockaddr_in sin = { 3086 .sin_family = AF_INET, 3087 .sin_addr.s_addr = htonl(INADDR_ANY), 3088 }; 3089 static const struct sockaddr_in6 sin6 = { 3090 .sin6_family = AF_INET6, 3091 .sin6_addr = IN6ADDR_ANY_INIT, 3092 }; 3093 3094 switch (family) { 3095 case AF_LOCAL: 3096 break; 3097 case AF_INET: 3098 memcpy(sap, &sin, sizeof(sin)); 3099 break; 3100 case AF_INET6: 3101 memcpy(sap, &sin6, sizeof(sin6)); 3102 break; 3103 default: 3104 dprintk("RPC: %s: Bad address family\n", __func__); 3105 return -EAFNOSUPPORT; 3106 } 3107 return 0; 3108 } 3109 3110 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 3111 unsigned int slot_table_size, 3112 unsigned int max_slot_table_size) 3113 { 3114 struct rpc_xprt *xprt; 3115 struct sock_xprt *new; 3116 3117 if (args->addrlen > sizeof(xprt->addr)) { 3118 dprintk("RPC: xs_setup_xprt: address too large\n"); 3119 return ERR_PTR(-EBADF); 3120 } 3121 3122 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 3123 max_slot_table_size); 3124 if (xprt == NULL) { 3125 dprintk("RPC: xs_setup_xprt: couldn't allocate " 3126 "rpc_xprt\n"); 3127 return ERR_PTR(-ENOMEM); 3128 } 3129 3130 new = container_of(xprt, struct sock_xprt, xprt); 3131 mutex_init(&new->recv_mutex); 3132 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 3133 xprt->addrlen = args->addrlen; 3134 if (args->srcaddr) 3135 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 3136 else { 3137 int err; 3138 err = xs_init_anyaddr(args->dstaddr->sa_family, 3139 (struct sockaddr *)&new->srcaddr); 3140 if (err != 0) { 3141 xprt_free(xprt); 3142 return ERR_PTR(err); 3143 } 3144 } 3145 3146 return xprt; 3147 } 3148 3149 static const struct rpc_timeout xs_local_default_timeout = { 3150 .to_initval = 10 * HZ, 3151 .to_maxval = 10 * HZ, 3152 .to_retries = 2, 3153 }; 3154 3155 /** 3156 * xs_setup_local - Set up transport to use an AF_LOCAL socket 3157 * @args: rpc transport creation arguments 3158 * 3159 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 3160 */ 3161 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 3162 { 3163 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 3164 struct sock_xprt *transport; 3165 struct rpc_xprt *xprt; 3166 struct rpc_xprt *ret; 3167 3168 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3169 xprt_max_tcp_slot_table_entries); 3170 if (IS_ERR(xprt)) 3171 return xprt; 3172 transport = container_of(xprt, struct sock_xprt, xprt); 3173 3174 xprt->prot = 0; 3175 xprt->xprt_class = &xs_local_transport; 3176 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3177 3178 xprt->bind_timeout = XS_BIND_TO; 3179 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 3180 xprt->idle_timeout = XS_IDLE_DISC_TO; 3181 3182 xprt->ops = &xs_local_ops; 3183 xprt->timeout = &xs_local_default_timeout; 3184 3185 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 3186 INIT_WORK(&transport->error_worker, xs_error_handle); 3187 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket); 3188 3189 switch (sun->sun_family) { 3190 case AF_LOCAL: 3191 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') { 3192 dprintk("RPC: bad AF_LOCAL address: %s\n", 3193 sun->sun_path); 3194 ret = ERR_PTR(-EINVAL); 3195 goto out_err; 3196 } 3197 xprt_set_bound(xprt); 3198 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 3199 break; 3200 default: 3201 ret = ERR_PTR(-EAFNOSUPPORT); 3202 goto out_err; 3203 } 3204 3205 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 3206 xprt->address_strings[RPC_DISPLAY_ADDR]); 3207 3208 if (try_module_get(THIS_MODULE)) 3209 return xprt; 3210 ret = ERR_PTR(-EINVAL); 3211 out_err: 3212 xs_xprt_free(xprt); 3213 return ret; 3214 } 3215 3216 static const struct rpc_timeout xs_udp_default_timeout = { 3217 .to_initval = 5 * HZ, 3218 .to_maxval = 30 * HZ, 3219 .to_increment = 5 * HZ, 3220 .to_retries = 5, 3221 }; 3222 3223 /** 3224 * xs_setup_udp - Set up transport to use a UDP socket 3225 * @args: rpc transport creation arguments 3226 * 3227 */ 3228 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 3229 { 3230 struct sockaddr *addr = args->dstaddr; 3231 struct rpc_xprt *xprt; 3232 struct sock_xprt *transport; 3233 struct rpc_xprt *ret; 3234 3235 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 3236 xprt_udp_slot_table_entries); 3237 if (IS_ERR(xprt)) 3238 return xprt; 3239 transport = container_of(xprt, struct sock_xprt, xprt); 3240 3241 xprt->prot = IPPROTO_UDP; 3242 xprt->xprt_class = &xs_udp_transport; 3243 /* XXX: header size can vary due to auth type, IPv6, etc. */ 3244 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 3245 3246 xprt->bind_timeout = XS_BIND_TO; 3247 xprt->reestablish_timeout = XS_UDP_REEST_TO; 3248 xprt->idle_timeout = XS_IDLE_DISC_TO; 3249 3250 xprt->ops = &xs_udp_ops; 3251 3252 xprt->timeout = &xs_udp_default_timeout; 3253 3254 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 3255 INIT_WORK(&transport->error_worker, xs_error_handle); 3256 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 3257 3258 switch (addr->sa_family) { 3259 case AF_INET: 3260 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3261 xprt_set_bound(xprt); 3262 3263 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 3264 break; 3265 case AF_INET6: 3266 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3267 xprt_set_bound(xprt); 3268 3269 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 3270 break; 3271 default: 3272 ret = ERR_PTR(-EAFNOSUPPORT); 3273 goto out_err; 3274 } 3275 3276 if (xprt_bound(xprt)) 3277 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3278 xprt->address_strings[RPC_DISPLAY_ADDR], 3279 xprt->address_strings[RPC_DISPLAY_PORT], 3280 xprt->address_strings[RPC_DISPLAY_PROTO]); 3281 else 3282 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3283 xprt->address_strings[RPC_DISPLAY_ADDR], 3284 xprt->address_strings[RPC_DISPLAY_PROTO]); 3285 3286 if (try_module_get(THIS_MODULE)) 3287 return xprt; 3288 ret = ERR_PTR(-EINVAL); 3289 out_err: 3290 xs_xprt_free(xprt); 3291 return ret; 3292 } 3293 3294 static const struct rpc_timeout xs_tcp_default_timeout = { 3295 .to_initval = 60 * HZ, 3296 .to_maxval = 60 * HZ, 3297 .to_retries = 2, 3298 }; 3299 3300 /** 3301 * xs_setup_tcp - Set up transport to use a TCP socket 3302 * @args: rpc transport creation arguments 3303 * 3304 */ 3305 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 3306 { 3307 struct sockaddr *addr = args->dstaddr; 3308 struct rpc_xprt *xprt; 3309 struct sock_xprt *transport; 3310 struct rpc_xprt *ret; 3311 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 3312 3313 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 3314 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 3315 3316 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3317 max_slot_table_size); 3318 if (IS_ERR(xprt)) 3319 return xprt; 3320 transport = container_of(xprt, struct sock_xprt, xprt); 3321 3322 xprt->prot = IPPROTO_TCP; 3323 xprt->xprt_class = &xs_tcp_transport; 3324 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3325 3326 xprt->bind_timeout = XS_BIND_TO; 3327 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 3328 xprt->idle_timeout = XS_IDLE_DISC_TO; 3329 3330 xprt->ops = &xs_tcp_ops; 3331 xprt->timeout = &xs_tcp_default_timeout; 3332 3333 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 3334 xprt->connect_timeout = xprt->timeout->to_initval * 3335 (xprt->timeout->to_retries + 1); 3336 3337 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 3338 INIT_WORK(&transport->error_worker, xs_error_handle); 3339 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 3340 3341 switch (addr->sa_family) { 3342 case AF_INET: 3343 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3344 xprt_set_bound(xprt); 3345 3346 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 3347 break; 3348 case AF_INET6: 3349 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3350 xprt_set_bound(xprt); 3351 3352 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 3353 break; 3354 default: 3355 ret = ERR_PTR(-EAFNOSUPPORT); 3356 goto out_err; 3357 } 3358 3359 if (xprt_bound(xprt)) 3360 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3361 xprt->address_strings[RPC_DISPLAY_ADDR], 3362 xprt->address_strings[RPC_DISPLAY_PORT], 3363 xprt->address_strings[RPC_DISPLAY_PROTO]); 3364 else 3365 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3366 xprt->address_strings[RPC_DISPLAY_ADDR], 3367 xprt->address_strings[RPC_DISPLAY_PROTO]); 3368 3369 if (try_module_get(THIS_MODULE)) 3370 return xprt; 3371 ret = ERR_PTR(-EINVAL); 3372 out_err: 3373 xs_xprt_free(xprt); 3374 return ret; 3375 } 3376 3377 /** 3378 * xs_setup_tcp_tls - Set up transport to use a TCP with TLS 3379 * @args: rpc transport creation arguments 3380 * 3381 */ 3382 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args) 3383 { 3384 struct sockaddr *addr = args->dstaddr; 3385 struct rpc_xprt *xprt; 3386 struct sock_xprt *transport; 3387 struct rpc_xprt *ret; 3388 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 3389 3390 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 3391 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 3392 3393 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3394 max_slot_table_size); 3395 if (IS_ERR(xprt)) 3396 return xprt; 3397 transport = container_of(xprt, struct sock_xprt, xprt); 3398 3399 xprt->prot = IPPROTO_TCP; 3400 xprt->xprt_class = &xs_tcp_transport; 3401 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3402 3403 xprt->bind_timeout = XS_BIND_TO; 3404 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 3405 xprt->idle_timeout = XS_IDLE_DISC_TO; 3406 3407 xprt->ops = &xs_tcp_ops; 3408 xprt->timeout = &xs_tcp_default_timeout; 3409 3410 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 3411 xprt->connect_timeout = xprt->timeout->to_initval * 3412 (xprt->timeout->to_retries + 1); 3413 3414 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 3415 INIT_WORK(&transport->error_worker, xs_error_handle); 3416 3417 switch (args->xprtsec.policy) { 3418 case RPC_XPRTSEC_TLS_ANON: 3419 case RPC_XPRTSEC_TLS_X509: 3420 xprt->xprtsec = args->xprtsec; 3421 INIT_DELAYED_WORK(&transport->connect_worker, 3422 xs_tcp_tls_setup_socket); 3423 break; 3424 default: 3425 ret = ERR_PTR(-EACCES); 3426 goto out_err; 3427 } 3428 3429 switch (addr->sa_family) { 3430 case AF_INET: 3431 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3432 xprt_set_bound(xprt); 3433 3434 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 3435 break; 3436 case AF_INET6: 3437 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3438 xprt_set_bound(xprt); 3439 3440 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 3441 break; 3442 default: 3443 ret = ERR_PTR(-EAFNOSUPPORT); 3444 goto out_err; 3445 } 3446 3447 if (xprt_bound(xprt)) 3448 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3449 xprt->address_strings[RPC_DISPLAY_ADDR], 3450 xprt->address_strings[RPC_DISPLAY_PORT], 3451 xprt->address_strings[RPC_DISPLAY_PROTO]); 3452 else 3453 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3454 xprt->address_strings[RPC_DISPLAY_ADDR], 3455 xprt->address_strings[RPC_DISPLAY_PROTO]); 3456 3457 if (try_module_get(THIS_MODULE)) 3458 return xprt; 3459 ret = ERR_PTR(-EINVAL); 3460 out_err: 3461 xs_xprt_free(xprt); 3462 return ret; 3463 } 3464 3465 /** 3466 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 3467 * @args: rpc transport creation arguments 3468 * 3469 */ 3470 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 3471 { 3472 struct sockaddr *addr = args->dstaddr; 3473 struct rpc_xprt *xprt; 3474 struct sock_xprt *transport; 3475 struct svc_sock *bc_sock; 3476 struct rpc_xprt *ret; 3477 3478 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3479 xprt_tcp_slot_table_entries); 3480 if (IS_ERR(xprt)) 3481 return xprt; 3482 transport = container_of(xprt, struct sock_xprt, xprt); 3483 3484 xprt->prot = IPPROTO_TCP; 3485 xprt->xprt_class = &xs_bc_tcp_transport; 3486 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3487 xprt->timeout = &xs_tcp_default_timeout; 3488 3489 /* backchannel */ 3490 xprt_set_bound(xprt); 3491 xprt->bind_timeout = 0; 3492 xprt->reestablish_timeout = 0; 3493 xprt->idle_timeout = 0; 3494 3495 xprt->ops = &bc_tcp_ops; 3496 3497 switch (addr->sa_family) { 3498 case AF_INET: 3499 xs_format_peer_addresses(xprt, "tcp", 3500 RPCBIND_NETID_TCP); 3501 break; 3502 case AF_INET6: 3503 xs_format_peer_addresses(xprt, "tcp", 3504 RPCBIND_NETID_TCP6); 3505 break; 3506 default: 3507 ret = ERR_PTR(-EAFNOSUPPORT); 3508 goto out_err; 3509 } 3510 3511 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3512 xprt->address_strings[RPC_DISPLAY_ADDR], 3513 xprt->address_strings[RPC_DISPLAY_PORT], 3514 xprt->address_strings[RPC_DISPLAY_PROTO]); 3515 3516 /* 3517 * Once we've associated a backchannel xprt with a connection, 3518 * we want to keep it around as long as the connection lasts, 3519 * in case we need to start using it for a backchannel again; 3520 * this reference won't be dropped until bc_xprt is destroyed. 3521 */ 3522 xprt_get(xprt); 3523 args->bc_xprt->xpt_bc_xprt = xprt; 3524 xprt->bc_xprt = args->bc_xprt; 3525 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3526 transport->sock = bc_sock->sk_sock; 3527 transport->inet = bc_sock->sk_sk; 3528 3529 /* 3530 * Since we don't want connections for the backchannel, we set 3531 * the xprt status to connected 3532 */ 3533 xprt_set_connected(xprt); 3534 3535 if (try_module_get(THIS_MODULE)) 3536 return xprt; 3537 3538 args->bc_xprt->xpt_bc_xprt = NULL; 3539 args->bc_xprt->xpt_bc_xps = NULL; 3540 xprt_put(xprt); 3541 ret = ERR_PTR(-EINVAL); 3542 out_err: 3543 xs_xprt_free(xprt); 3544 return ret; 3545 } 3546 3547 static struct xprt_class xs_local_transport = { 3548 .list = LIST_HEAD_INIT(xs_local_transport.list), 3549 .name = "named UNIX socket", 3550 .owner = THIS_MODULE, 3551 .ident = XPRT_TRANSPORT_LOCAL, 3552 .setup = xs_setup_local, 3553 .netid = { "" }, 3554 }; 3555 3556 static struct xprt_class xs_udp_transport = { 3557 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3558 .name = "udp", 3559 .owner = THIS_MODULE, 3560 .ident = XPRT_TRANSPORT_UDP, 3561 .setup = xs_setup_udp, 3562 .netid = { "udp", "udp6", "" }, 3563 }; 3564 3565 static struct xprt_class xs_tcp_transport = { 3566 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3567 .name = "tcp", 3568 .owner = THIS_MODULE, 3569 .ident = XPRT_TRANSPORT_TCP, 3570 .setup = xs_setup_tcp, 3571 .netid = { "tcp", "tcp6", "" }, 3572 }; 3573 3574 static struct xprt_class xs_tcp_tls_transport = { 3575 .list = LIST_HEAD_INIT(xs_tcp_tls_transport.list), 3576 .name = "tcp-with-tls", 3577 .owner = THIS_MODULE, 3578 .ident = XPRT_TRANSPORT_TCP_TLS, 3579 .setup = xs_setup_tcp_tls, 3580 .netid = { "tcp", "tcp6", "" }, 3581 }; 3582 3583 static struct xprt_class xs_bc_tcp_transport = { 3584 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3585 .name = "tcp NFSv4.1 backchannel", 3586 .owner = THIS_MODULE, 3587 .ident = XPRT_TRANSPORT_BC_TCP, 3588 .setup = xs_setup_bc_tcp, 3589 .netid = { "" }, 3590 }; 3591 3592 /** 3593 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3594 * 3595 */ 3596 int init_socket_xprt(void) 3597 { 3598 if (!sunrpc_table_header) 3599 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table); 3600 3601 xprt_register_transport(&xs_local_transport); 3602 xprt_register_transport(&xs_udp_transport); 3603 xprt_register_transport(&xs_tcp_transport); 3604 xprt_register_transport(&xs_tcp_tls_transport); 3605 xprt_register_transport(&xs_bc_tcp_transport); 3606 3607 return 0; 3608 } 3609 3610 /** 3611 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3612 * 3613 */ 3614 void cleanup_socket_xprt(void) 3615 { 3616 if (sunrpc_table_header) { 3617 unregister_sysctl_table(sunrpc_table_header); 3618 sunrpc_table_header = NULL; 3619 } 3620 3621 xprt_unregister_transport(&xs_local_transport); 3622 xprt_unregister_transport(&xs_udp_transport); 3623 xprt_unregister_transport(&xs_tcp_transport); 3624 xprt_unregister_transport(&xs_tcp_tls_transport); 3625 xprt_unregister_transport(&xs_bc_tcp_transport); 3626 } 3627 3628 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3629 { 3630 return param_set_uint_minmax(val, kp, 3631 RPC_MIN_RESVPORT, 3632 RPC_MAX_RESVPORT); 3633 } 3634 3635 static const struct kernel_param_ops param_ops_portnr = { 3636 .set = param_set_portnr, 3637 .get = param_get_uint, 3638 }; 3639 3640 #define param_check_portnr(name, p) \ 3641 __param_check(name, p, unsigned int); 3642 3643 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3644 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3645 3646 static int param_set_slot_table_size(const char *val, 3647 const struct kernel_param *kp) 3648 { 3649 return param_set_uint_minmax(val, kp, 3650 RPC_MIN_SLOT_TABLE, 3651 RPC_MAX_SLOT_TABLE); 3652 } 3653 3654 static const struct kernel_param_ops param_ops_slot_table_size = { 3655 .set = param_set_slot_table_size, 3656 .get = param_get_uint, 3657 }; 3658 3659 #define param_check_slot_table_size(name, p) \ 3660 __param_check(name, p, unsigned int); 3661 3662 static int param_set_max_slot_table_size(const char *val, 3663 const struct kernel_param *kp) 3664 { 3665 return param_set_uint_minmax(val, kp, 3666 RPC_MIN_SLOT_TABLE, 3667 RPC_MAX_SLOT_TABLE_LIMIT); 3668 } 3669 3670 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3671 .set = param_set_max_slot_table_size, 3672 .get = param_get_uint, 3673 }; 3674 3675 #define param_check_max_slot_table_size(name, p) \ 3676 __param_check(name, p, unsigned int); 3677 3678 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3679 slot_table_size, 0644); 3680 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3681 max_slot_table_size, 0644); 3682 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3683 slot_table_size, 0644); 3684