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_prot.h> 51 #include <net/handshake.h> 52 53 #include <linux/bvec.h> 54 #include <linux/highmem.h> 55 #include <linux/uio.h> 56 #include <linux/sched/mm.h> 57 58 #include <trace/events/sock.h> 59 #include <trace/events/sunrpc.h> 60 61 #include "socklib.h" 62 #include "sunrpc.h" 63 64 static void xs_close(struct rpc_xprt *xprt); 65 static void xs_reset_srcport(struct sock_xprt *transport); 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 * Wait duration for a reply from the RPC portmapper. 167 */ 168 #define XS_BIND_TO (60U * HZ) 169 170 /* 171 * Delay if a UDP socket connect error occurs. This is most likely some 172 * kind of resource problem on the local host. 173 */ 174 #define XS_UDP_REEST_TO (2U * HZ) 175 176 /* 177 * The reestablish timeout allows clients to delay for a bit before attempting 178 * to reconnect to a server that just dropped our connection. 179 * 180 * We implement an exponential backoff when trying to reestablish a TCP 181 * transport connection with the server. Some servers like to drop a TCP 182 * connection when they are overworked, so we start with a short timeout and 183 * increase over time if the server is down or not responding. 184 */ 185 #define XS_TCP_INIT_REEST_TO (3U * HZ) 186 187 /* 188 * TCP idle timeout; client drops the transport socket if it is idle 189 * for this long. Note that we also timeout UDP sockets to prevent 190 * holding port numbers when there is no RPC traffic. 191 */ 192 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 193 194 /* 195 * TLS handshake timeout. 196 */ 197 #define XS_TLS_HANDSHAKE_TO (10U * HZ) 198 199 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 200 # undef RPC_DEBUG_DATA 201 # define RPCDBG_FACILITY RPCDBG_TRANS 202 #endif 203 204 #ifdef RPC_DEBUG_DATA 205 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 206 { 207 u8 *buf = (u8 *) packet; 208 int j; 209 210 dprintk("RPC: %s\n", msg); 211 for (j = 0; j < count && j < 128; j += 4) { 212 if (!(j & 31)) { 213 if (j) 214 dprintk("\n"); 215 dprintk("0x%04x ", j); 216 } 217 dprintk("%02x%02x%02x%02x ", 218 buf[j], buf[j+1], buf[j+2], buf[j+3]); 219 } 220 dprintk("\n"); 221 } 222 #else 223 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 224 { 225 /* NOP */ 226 } 227 #endif 228 229 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 230 { 231 return (struct rpc_xprt *) sk->sk_user_data; 232 } 233 234 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 235 { 236 return (struct sockaddr *) &xprt->addr; 237 } 238 239 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 240 { 241 return (struct sockaddr_un *) &xprt->addr; 242 } 243 244 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 245 { 246 return (struct sockaddr_in *) &xprt->addr; 247 } 248 249 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 250 { 251 return (struct sockaddr_in6 *) &xprt->addr; 252 } 253 254 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 255 { 256 struct sockaddr *sap = xs_addr(xprt); 257 struct sockaddr_in6 *sin6; 258 struct sockaddr_in *sin; 259 struct sockaddr_un *sun; 260 char buf[128]; 261 262 switch (sap->sa_family) { 263 case AF_LOCAL: 264 sun = xs_addr_un(xprt); 265 if (sun->sun_path[0]) { 266 strscpy(buf, sun->sun_path, sizeof(buf)); 267 } else { 268 buf[0] = '@'; 269 strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1); 270 } 271 xprt->address_strings[RPC_DISPLAY_ADDR] = 272 kstrdup(buf, GFP_KERNEL); 273 break; 274 case AF_INET: 275 (void)rpc_ntop(sap, buf, sizeof(buf)); 276 xprt->address_strings[RPC_DISPLAY_ADDR] = 277 kstrdup(buf, GFP_KERNEL); 278 sin = xs_addr_in(xprt); 279 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 280 break; 281 case AF_INET6: 282 (void)rpc_ntop(sap, buf, sizeof(buf)); 283 xprt->address_strings[RPC_DISPLAY_ADDR] = 284 kstrdup(buf, GFP_KERNEL); 285 sin6 = xs_addr_in6(xprt); 286 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 287 break; 288 default: 289 BUG(); 290 } 291 292 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 293 } 294 295 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 296 { 297 struct sockaddr *sap = xs_addr(xprt); 298 char buf[128]; 299 300 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 301 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 302 303 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 304 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 305 } 306 307 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 308 const char *protocol, 309 const char *netid) 310 { 311 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 312 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 313 xs_format_common_peer_addresses(xprt); 314 xs_format_common_peer_ports(xprt); 315 } 316 317 static void xs_update_peer_port(struct rpc_xprt *xprt) 318 { 319 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 320 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 321 322 xs_format_common_peer_ports(xprt); 323 } 324 325 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 326 { 327 unsigned int i; 328 329 for (i = 0; i < RPC_DISPLAY_MAX; i++) 330 switch (i) { 331 case RPC_DISPLAY_PROTO: 332 case RPC_DISPLAY_NETID: 333 continue; 334 default: 335 kfree(xprt->address_strings[i]); 336 } 337 } 338 339 static size_t 340 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp) 341 { 342 size_t i,n; 343 344 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES)) 345 return want; 346 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT; 347 for (i = 0; i < n; i++) { 348 if (buf->pages[i]) 349 continue; 350 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp); 351 if (!buf->pages[i]) { 352 i *= PAGE_SIZE; 353 return i > buf->page_base ? i - buf->page_base : 0; 354 } 355 } 356 return want; 357 } 358 359 static int 360 xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg, 361 struct cmsghdr *cmsg, int ret) 362 { 363 u8 content_type = tls_get_record_type(sock->sk, cmsg); 364 u8 level, description; 365 366 switch (content_type) { 367 case 0: 368 break; 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 tls_alert_recv(sock->sk, msg, &level, &description); 378 ret = (level == TLS_ALERT_LEVEL_FATAL) ? 379 -EACCES : -EAGAIN; 380 break; 381 default: 382 /* discard this record type */ 383 ret = -EAGAIN; 384 } 385 return ret; 386 } 387 388 static int 389 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags) 390 { 391 union { 392 struct cmsghdr cmsg; 393 u8 buf[CMSG_SPACE(sizeof(u8))]; 394 } u; 395 int ret; 396 397 msg->msg_control = &u; 398 msg->msg_controllen = sizeof(u); 399 ret = sock_recvmsg(sock, msg, flags); 400 if (msg->msg_controllen != sizeof(u)) 401 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret); 402 return ret; 403 } 404 405 static ssize_t 406 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek) 407 { 408 ssize_t ret; 409 if (seek != 0) 410 iov_iter_advance(&msg->msg_iter, seek); 411 ret = xs_sock_recv_cmsg(sock, msg, flags); 412 return ret > 0 ? ret + seek : ret; 413 } 414 415 static ssize_t 416 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags, 417 struct kvec *kvec, size_t count, size_t seek) 418 { 419 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count); 420 return xs_sock_recvmsg(sock, msg, flags, seek); 421 } 422 423 static ssize_t 424 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags, 425 struct bio_vec *bvec, unsigned long nr, size_t count, 426 size_t seek) 427 { 428 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count); 429 return xs_sock_recvmsg(sock, msg, flags, seek); 430 } 431 432 static ssize_t 433 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags, 434 size_t count) 435 { 436 iov_iter_discard(&msg->msg_iter, ITER_DEST, count); 437 return xs_sock_recv_cmsg(sock, msg, flags); 438 } 439 440 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 441 static void 442 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 443 { 444 struct bvec_iter bi = { 445 .bi_size = count, 446 }; 447 struct bio_vec bv; 448 449 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); 450 for_each_bvec(bv, bvec, bi, bi) 451 flush_dcache_page(bv.bv_page); 452 } 453 #else 454 static inline void 455 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 456 { 457 } 458 #endif 459 460 static ssize_t 461 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags, 462 struct xdr_buf *buf, size_t count, size_t seek, size_t *read) 463 { 464 size_t want, seek_init = seek, offset = 0; 465 ssize_t ret; 466 467 want = min_t(size_t, count, buf->head[0].iov_len); 468 if (seek < want) { 469 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek); 470 if (ret <= 0) 471 goto sock_err; 472 offset += ret; 473 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 474 goto out; 475 if (ret != want) 476 goto out; 477 seek = 0; 478 } else { 479 seek -= want; 480 offset += want; 481 } 482 483 want = xs_alloc_sparse_pages( 484 buf, min_t(size_t, count - offset, buf->page_len), 485 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN); 486 if (seek < want) { 487 ret = xs_read_bvec(sock, msg, flags, buf->bvec, 488 xdr_buf_pagecount(buf), 489 want + buf->page_base, 490 seek + buf->page_base); 491 if (ret <= 0) 492 goto sock_err; 493 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base); 494 ret -= buf->page_base; 495 offset += ret; 496 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 497 goto out; 498 if (ret != want) 499 goto out; 500 seek = 0; 501 } else { 502 seek -= want; 503 offset += want; 504 } 505 506 want = min_t(size_t, count - offset, buf->tail[0].iov_len); 507 if (seek < want) { 508 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek); 509 if (ret <= 0) 510 goto sock_err; 511 offset += ret; 512 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 513 goto out; 514 if (ret != want) 515 goto out; 516 } else if (offset < seek_init) 517 offset = seek_init; 518 ret = -EMSGSIZE; 519 out: 520 *read = offset - seek_init; 521 return ret; 522 sock_err: 523 offset += seek; 524 goto out; 525 } 526 527 static void 528 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf) 529 { 530 if (!transport->recv.copied) { 531 if (buf->head[0].iov_len >= transport->recv.offset) 532 memcpy(buf->head[0].iov_base, 533 &transport->recv.xid, 534 transport->recv.offset); 535 transport->recv.copied = transport->recv.offset; 536 } 537 } 538 539 static bool 540 xs_read_stream_request_done(struct sock_xprt *transport) 541 { 542 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT); 543 } 544 545 static void 546 xs_read_stream_check_eor(struct sock_xprt *transport, 547 struct msghdr *msg) 548 { 549 if (xs_read_stream_request_done(transport)) 550 msg->msg_flags |= MSG_EOR; 551 } 552 553 static ssize_t 554 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg, 555 int flags, struct rpc_rqst *req) 556 { 557 struct xdr_buf *buf = &req->rq_private_buf; 558 size_t want, read; 559 ssize_t ret; 560 561 xs_read_header(transport, buf); 562 563 want = transport->recv.len - transport->recv.offset; 564 if (want != 0) { 565 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf, 566 transport->recv.copied + want, 567 transport->recv.copied, 568 &read); 569 transport->recv.offset += read; 570 transport->recv.copied += read; 571 } 572 573 if (transport->recv.offset == transport->recv.len) 574 xs_read_stream_check_eor(transport, msg); 575 576 if (want == 0) 577 return 0; 578 579 switch (ret) { 580 default: 581 break; 582 case -EFAULT: 583 case -EMSGSIZE: 584 msg->msg_flags |= MSG_TRUNC; 585 return read; 586 case 0: 587 return -ESHUTDOWN; 588 } 589 return ret < 0 ? ret : read; 590 } 591 592 static size_t 593 xs_read_stream_headersize(bool isfrag) 594 { 595 if (isfrag) 596 return sizeof(__be32); 597 return 3 * sizeof(__be32); 598 } 599 600 static ssize_t 601 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg, 602 int flags, size_t want, size_t seek) 603 { 604 struct kvec kvec = { 605 .iov_base = &transport->recv.fraghdr, 606 .iov_len = want, 607 }; 608 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek); 609 } 610 611 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 612 static ssize_t 613 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 614 { 615 struct rpc_xprt *xprt = &transport->xprt; 616 struct rpc_rqst *req; 617 ssize_t ret; 618 619 /* Is this transport associated with the backchannel? */ 620 if (!xprt->bc_serv) 621 return -ESHUTDOWN; 622 623 /* Look up and lock the request corresponding to the given XID */ 624 req = xprt_lookup_bc_request(xprt, transport->recv.xid); 625 if (!req) { 626 printk(KERN_WARNING "Callback slot table overflowed\n"); 627 return -ESHUTDOWN; 628 } 629 if (transport->recv.copied && !req->rq_private_buf.len) 630 return -ESHUTDOWN; 631 632 ret = xs_read_stream_request(transport, msg, flags, req); 633 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 634 xprt_complete_bc_request(req, transport->recv.copied); 635 else 636 req->rq_private_buf.len = transport->recv.copied; 637 638 return ret; 639 } 640 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 641 static ssize_t 642 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 643 { 644 return -ESHUTDOWN; 645 } 646 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 647 648 static ssize_t 649 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags) 650 { 651 struct rpc_xprt *xprt = &transport->xprt; 652 struct rpc_rqst *req; 653 ssize_t ret = 0; 654 655 /* Look up and lock the request corresponding to the given XID */ 656 spin_lock(&xprt->queue_lock); 657 req = xprt_lookup_rqst(xprt, transport->recv.xid); 658 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) { 659 msg->msg_flags |= MSG_TRUNC; 660 goto out; 661 } 662 xprt_pin_rqst(req); 663 spin_unlock(&xprt->queue_lock); 664 665 ret = xs_read_stream_request(transport, msg, flags, req); 666 667 spin_lock(&xprt->queue_lock); 668 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 669 xprt_complete_rqst(req->rq_task, transport->recv.copied); 670 else 671 req->rq_private_buf.len = transport->recv.copied; 672 xprt_unpin_rqst(req); 673 out: 674 spin_unlock(&xprt->queue_lock); 675 return ret; 676 } 677 678 static ssize_t 679 xs_read_stream(struct sock_xprt *transport, int flags) 680 { 681 struct msghdr msg = { 0 }; 682 size_t want, read = 0; 683 ssize_t ret = 0; 684 685 if (transport->recv.len == 0) { 686 want = xs_read_stream_headersize(transport->recv.copied != 0); 687 ret = xs_read_stream_header(transport, &msg, flags, want, 688 transport->recv.offset); 689 if (ret <= 0) 690 goto out_err; 691 transport->recv.offset = ret; 692 if (transport->recv.offset != want) 693 return transport->recv.offset; 694 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) & 695 RPC_FRAGMENT_SIZE_MASK; 696 transport->recv.offset -= sizeof(transport->recv.fraghdr); 697 read = ret; 698 } 699 700 switch (be32_to_cpu(transport->recv.calldir)) { 701 default: 702 msg.msg_flags |= MSG_TRUNC; 703 break; 704 case RPC_CALL: 705 ret = xs_read_stream_call(transport, &msg, flags); 706 break; 707 case RPC_REPLY: 708 ret = xs_read_stream_reply(transport, &msg, flags); 709 } 710 if (msg.msg_flags & MSG_TRUNC) { 711 transport->recv.calldir = cpu_to_be32(-1); 712 transport->recv.copied = -1; 713 } 714 if (ret < 0) 715 goto out_err; 716 read += ret; 717 if (transport->recv.offset < transport->recv.len) { 718 if (!(msg.msg_flags & MSG_TRUNC)) 719 return read; 720 msg.msg_flags = 0; 721 ret = xs_read_discard(transport->sock, &msg, flags, 722 transport->recv.len - transport->recv.offset); 723 if (ret <= 0) 724 goto out_err; 725 transport->recv.offset += ret; 726 read += ret; 727 if (transport->recv.offset != transport->recv.len) 728 return read; 729 } 730 if (xs_read_stream_request_done(transport)) { 731 trace_xs_stream_read_request(transport); 732 transport->recv.copied = 0; 733 } 734 transport->recv.offset = 0; 735 transport->recv.len = 0; 736 return read; 737 out_err: 738 return ret != 0 ? ret : -ESHUTDOWN; 739 } 740 741 static __poll_t xs_poll_socket(struct sock_xprt *transport) 742 { 743 return transport->sock->ops->poll(transport->file, transport->sock, 744 NULL); 745 } 746 747 static bool xs_poll_socket_readable(struct sock_xprt *transport) 748 { 749 __poll_t events = xs_poll_socket(transport); 750 751 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP); 752 } 753 754 static void xs_poll_check_readable(struct sock_xprt *transport) 755 { 756 757 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 758 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state)) 759 return; 760 if (!xs_poll_socket_readable(transport)) 761 return; 762 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 763 queue_work(xprtiod_workqueue, &transport->recv_worker); 764 } 765 766 static void xs_stream_data_receive(struct sock_xprt *transport) 767 { 768 size_t read = 0; 769 ssize_t ret = 0; 770 771 mutex_lock(&transport->recv_mutex); 772 if (transport->sock == NULL) 773 goto out; 774 for (;;) { 775 ret = xs_read_stream(transport, MSG_DONTWAIT); 776 if (ret < 0) 777 break; 778 read += ret; 779 cond_resched(); 780 } 781 if (ret == -ESHUTDOWN) 782 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 783 else if (ret == -EACCES) 784 xprt_wake_pending_tasks(&transport->xprt, -EACCES); 785 else 786 xs_poll_check_readable(transport); 787 out: 788 mutex_unlock(&transport->recv_mutex); 789 trace_xs_stream_read_data(&transport->xprt, ret, read); 790 } 791 792 static void xs_stream_data_receive_workfn(struct work_struct *work) 793 { 794 struct sock_xprt *transport = 795 container_of(work, struct sock_xprt, recv_worker); 796 unsigned int pflags = memalloc_nofs_save(); 797 798 xs_stream_data_receive(transport); 799 memalloc_nofs_restore(pflags); 800 } 801 802 static void 803 xs_stream_reset_connect(struct sock_xprt *transport) 804 { 805 transport->recv.offset = 0; 806 transport->recv.len = 0; 807 transport->recv.copied = 0; 808 transport->xmit.offset = 0; 809 } 810 811 static void 812 xs_stream_start_connect(struct sock_xprt *transport) 813 { 814 transport->xprt.stat.connect_count++; 815 transport->xprt.stat.connect_start = jiffies; 816 } 817 818 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 819 820 /** 821 * xs_nospace - handle transmit was incomplete 822 * @req: pointer to RPC request 823 * @transport: pointer to struct sock_xprt 824 * 825 */ 826 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport) 827 { 828 struct rpc_xprt *xprt = &transport->xprt; 829 struct sock *sk = transport->inet; 830 int ret = -EAGAIN; 831 832 trace_rpc_socket_nospace(req, transport); 833 834 /* Protect against races with write_space */ 835 spin_lock(&xprt->transport_lock); 836 837 /* Don't race with disconnect */ 838 if (xprt_connected(xprt)) { 839 /* wait for more buffer space */ 840 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state); 841 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 842 sk->sk_write_pending++; 843 xprt_wait_for_buffer_space(xprt); 844 } else 845 ret = -ENOTCONN; 846 847 spin_unlock(&xprt->transport_lock); 848 return ret; 849 } 850 851 static int xs_sock_nospace(struct rpc_rqst *req) 852 { 853 struct sock_xprt *transport = 854 container_of(req->rq_xprt, struct sock_xprt, xprt); 855 struct sock *sk = transport->inet; 856 int ret = -EAGAIN; 857 858 lock_sock(sk); 859 if (!sock_writeable(sk)) 860 ret = xs_nospace(req, transport); 861 release_sock(sk); 862 return ret; 863 } 864 865 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait) 866 { 867 struct sock_xprt *transport = 868 container_of(req->rq_xprt, struct sock_xprt, xprt); 869 struct sock *sk = transport->inet; 870 int ret = -EAGAIN; 871 872 if (vm_wait) 873 return -ENOBUFS; 874 lock_sock(sk); 875 if (!sk_stream_memory_free(sk)) 876 ret = xs_nospace(req, transport); 877 release_sock(sk); 878 return ret; 879 } 880 881 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf) 882 { 883 return xdr_alloc_bvec(buf, rpc_task_gfp_mask()); 884 } 885 886 static void xs_stream_abort_send_request(struct rpc_rqst *req) 887 { 888 struct rpc_xprt *xprt = req->rq_xprt; 889 struct sock_xprt *transport = 890 container_of(xprt, struct sock_xprt, xprt); 891 892 if (transport->xmit.offset != 0 && 893 !test_bit(XPRT_CLOSE_WAIT, &xprt->state)) 894 xprt_force_disconnect(xprt); 895 } 896 897 /* 898 * Determine if the previous message in the stream was aborted before it 899 * could complete transmission. 900 */ 901 static bool 902 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req) 903 { 904 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0; 905 } 906 907 /* 908 * Return the stream record marker field for a record of length < 2^31-1 909 */ 910 static rpc_fraghdr 911 xs_stream_record_marker(struct xdr_buf *xdr) 912 { 913 if (!xdr->len) 914 return 0; 915 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len); 916 } 917 918 /** 919 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 920 * @req: pointer to RPC request 921 * 922 * Return values: 923 * 0: The request has been sent 924 * EAGAIN: The socket was blocked, please call again later to 925 * complete the request 926 * ENOTCONN: Caller needs to invoke connect logic then call again 927 * other: Some other error occurred, the request was not sent 928 */ 929 static int xs_local_send_request(struct rpc_rqst *req) 930 { 931 struct rpc_xprt *xprt = req->rq_xprt; 932 struct sock_xprt *transport = 933 container_of(xprt, struct sock_xprt, xprt); 934 struct xdr_buf *xdr = &req->rq_snd_buf; 935 rpc_fraghdr rm = xs_stream_record_marker(xdr); 936 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 937 struct msghdr msg = { 938 .msg_flags = XS_SENDMSG_FLAGS, 939 }; 940 bool vm_wait; 941 unsigned int sent; 942 int status; 943 944 /* Close the stream if the previous transmission was incomplete */ 945 if (xs_send_request_was_aborted(transport, req)) { 946 xprt_force_disconnect(xprt); 947 return -ENOTCONN; 948 } 949 950 xs_pktdump("packet data:", 951 req->rq_svec->iov_base, req->rq_svec->iov_len); 952 953 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false; 954 955 req->rq_xtime = ktime_get(); 956 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 957 transport->xmit.offset, rm, &sent); 958 dprintk("RPC: %s(%u) = %d\n", 959 __func__, xdr->len - transport->xmit.offset, status); 960 961 if (likely(sent > 0) || status == 0) { 962 transport->xmit.offset += sent; 963 req->rq_bytes_sent = transport->xmit.offset; 964 if (likely(req->rq_bytes_sent >= msglen)) { 965 req->rq_xmit_bytes_sent += transport->xmit.offset; 966 transport->xmit.offset = 0; 967 return 0; 968 } 969 status = -EAGAIN; 970 vm_wait = false; 971 } 972 973 switch (status) { 974 case -EAGAIN: 975 status = xs_stream_nospace(req, vm_wait); 976 break; 977 default: 978 dprintk("RPC: sendmsg returned unrecognized error %d\n", 979 -status); 980 fallthrough; 981 case -EPIPE: 982 xprt_force_disconnect(xprt); 983 status = -ENOTCONN; 984 } 985 986 return status; 987 } 988 989 /** 990 * xs_udp_send_request - write an RPC request to a UDP socket 991 * @req: pointer to RPC request 992 * 993 * Return values: 994 * 0: The request has been sent 995 * EAGAIN: The socket was blocked, please call again later to 996 * complete the request 997 * ENOTCONN: Caller needs to invoke connect logic then call again 998 * other: Some other error occurred, the request was not sent 999 */ 1000 static int xs_udp_send_request(struct rpc_rqst *req) 1001 { 1002 struct rpc_xprt *xprt = req->rq_xprt; 1003 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1004 struct xdr_buf *xdr = &req->rq_snd_buf; 1005 struct msghdr msg = { 1006 .msg_name = xs_addr(xprt), 1007 .msg_namelen = xprt->addrlen, 1008 .msg_flags = XS_SENDMSG_FLAGS, 1009 }; 1010 unsigned int sent; 1011 int status; 1012 1013 xs_pktdump("packet data:", 1014 req->rq_svec->iov_base, 1015 req->rq_svec->iov_len); 1016 1017 if (!xprt_bound(xprt)) 1018 return -ENOTCONN; 1019 1020 if (!xprt_request_get_cong(xprt, req)) 1021 return -EBADSLT; 1022 1023 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask()); 1024 if (status < 0) 1025 return status; 1026 req->rq_xtime = ktime_get(); 1027 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent); 1028 1029 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 1030 xdr->len, status); 1031 1032 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 1033 if (status == -EPERM) 1034 goto process_status; 1035 1036 if (status == -EAGAIN && sock_writeable(transport->inet)) 1037 status = -ENOBUFS; 1038 1039 if (sent > 0 || status == 0) { 1040 req->rq_xmit_bytes_sent += sent; 1041 if (sent >= req->rq_slen) 1042 return 0; 1043 /* Still some bytes left; set up for a retry later. */ 1044 status = -EAGAIN; 1045 } 1046 1047 process_status: 1048 switch (status) { 1049 case -ENOTSOCK: 1050 status = -ENOTCONN; 1051 /* Should we call xs_close() here? */ 1052 break; 1053 case -EAGAIN: 1054 status = xs_sock_nospace(req); 1055 break; 1056 case -ENETUNREACH: 1057 case -ENOBUFS: 1058 case -EPIPE: 1059 case -ECONNREFUSED: 1060 case -EPERM: 1061 /* When the server has died, an ICMP port unreachable message 1062 * prompts ECONNREFUSED. */ 1063 break; 1064 default: 1065 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1066 -status); 1067 } 1068 1069 return status; 1070 } 1071 1072 /** 1073 * xs_tcp_send_request - write an RPC request to a TCP socket 1074 * @req: pointer to RPC request 1075 * 1076 * Return values: 1077 * 0: The request has been sent 1078 * EAGAIN: The socket was blocked, please call again later to 1079 * complete the request 1080 * ENOTCONN: Caller needs to invoke connect logic then call again 1081 * other: Some other error occurred, the request was not sent 1082 * 1083 * XXX: In the case of soft timeouts, should we eventually give up 1084 * if sendmsg is not able to make progress? 1085 */ 1086 static int xs_tcp_send_request(struct rpc_rqst *req) 1087 { 1088 struct rpc_xprt *xprt = req->rq_xprt; 1089 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1090 struct xdr_buf *xdr = &req->rq_snd_buf; 1091 rpc_fraghdr rm = xs_stream_record_marker(xdr); 1092 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 1093 struct msghdr msg = { 1094 .msg_flags = XS_SENDMSG_FLAGS, 1095 }; 1096 bool vm_wait; 1097 unsigned int sent; 1098 int status; 1099 1100 /* Close the stream if the previous transmission was incomplete */ 1101 if (xs_send_request_was_aborted(transport, req)) { 1102 if (transport->sock != NULL) 1103 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 1104 return -ENOTCONN; 1105 } 1106 if (!transport->inet) 1107 return -ENOTCONN; 1108 1109 xs_pktdump("packet data:", 1110 req->rq_svec->iov_base, 1111 req->rq_svec->iov_len); 1112 1113 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state)) 1114 xs_tcp_set_socket_timeouts(xprt, transport->sock); 1115 1116 xs_set_srcport(transport, transport->sock); 1117 1118 /* Continue transmitting the packet/record. We must be careful 1119 * to cope with writespace callbacks arriving _after_ we have 1120 * called sendmsg(). */ 1121 req->rq_xtime = ktime_get(); 1122 tcp_sock_set_cork(transport->inet, true); 1123 1124 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false; 1125 1126 do { 1127 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 1128 transport->xmit.offset, rm, &sent); 1129 1130 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 1131 xdr->len - transport->xmit.offset, status); 1132 1133 /* If we've sent the entire packet, immediately 1134 * reset the count of bytes sent. */ 1135 transport->xmit.offset += sent; 1136 req->rq_bytes_sent = transport->xmit.offset; 1137 if (likely(req->rq_bytes_sent >= msglen)) { 1138 req->rq_xmit_bytes_sent += transport->xmit.offset; 1139 transport->xmit.offset = 0; 1140 if (atomic_long_read(&xprt->xmit_queuelen) == 1) 1141 tcp_sock_set_cork(transport->inet, false); 1142 return 0; 1143 } 1144 1145 WARN_ON_ONCE(sent == 0 && status == 0); 1146 1147 if (sent > 0) 1148 vm_wait = false; 1149 1150 } while (status == 0); 1151 1152 switch (status) { 1153 case -ENOTSOCK: 1154 status = -ENOTCONN; 1155 /* Should we call xs_close() here? */ 1156 break; 1157 case -EAGAIN: 1158 status = xs_stream_nospace(req, vm_wait); 1159 break; 1160 case -ECONNRESET: 1161 case -ECONNREFUSED: 1162 case -ENOTCONN: 1163 case -EADDRINUSE: 1164 case -ENOBUFS: 1165 case -EPIPE: 1166 break; 1167 default: 1168 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1169 -status); 1170 } 1171 1172 return status; 1173 } 1174 1175 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1176 { 1177 transport->old_data_ready = sk->sk_data_ready; 1178 transport->old_state_change = sk->sk_state_change; 1179 transport->old_write_space = sk->sk_write_space; 1180 transport->old_error_report = sk->sk_error_report; 1181 } 1182 1183 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1184 { 1185 sk->sk_data_ready = transport->old_data_ready; 1186 sk->sk_state_change = transport->old_state_change; 1187 sk->sk_write_space = transport->old_write_space; 1188 sk->sk_error_report = transport->old_error_report; 1189 } 1190 1191 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt) 1192 { 1193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1194 1195 transport->xprt_err = 0; 1196 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 1197 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state); 1198 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state); 1199 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state); 1200 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state); 1201 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 1202 } 1203 1204 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr) 1205 { 1206 set_bit(nr, &transport->sock_state); 1207 queue_work(xprtiod_workqueue, &transport->error_worker); 1208 } 1209 1210 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1211 { 1212 xprt->connect_cookie++; 1213 smp_mb__before_atomic(); 1214 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1215 clear_bit(XPRT_CLOSING, &xprt->state); 1216 xs_sock_reset_state_flags(xprt); 1217 smp_mb__after_atomic(); 1218 } 1219 1220 /** 1221 * xs_error_report - callback to handle TCP socket state errors 1222 * @sk: socket 1223 * 1224 * Note: we don't call sock_error() since there may be a rpc_task 1225 * using the socket, and so we don't want to clear sk->sk_err. 1226 */ 1227 static void xs_error_report(struct sock *sk) 1228 { 1229 struct sock_xprt *transport; 1230 struct rpc_xprt *xprt; 1231 1232 if (!(xprt = xprt_from_sock(sk))) 1233 return; 1234 1235 transport = container_of(xprt, struct sock_xprt, xprt); 1236 transport->xprt_err = -sk->sk_err; 1237 if (transport->xprt_err == 0) 1238 return; 1239 dprintk("RPC: xs_error_report client %p, error=%d...\n", 1240 xprt, -transport->xprt_err); 1241 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err); 1242 1243 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */ 1244 smp_mb__before_atomic(); 1245 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR); 1246 } 1247 1248 static void xs_reset_transport(struct sock_xprt *transport) 1249 { 1250 struct socket *sock = transport->sock; 1251 struct sock *sk = transport->inet; 1252 struct rpc_xprt *xprt = &transport->xprt; 1253 struct file *filp = transport->file; 1254 1255 if (sk == NULL) 1256 return; 1257 /* 1258 * Make sure we're calling this in a context from which it is safe 1259 * to call __fput_sync(). In practice that means rpciod and the 1260 * system workqueue. 1261 */ 1262 if (!(current->flags & PF_WQ_WORKER)) { 1263 WARN_ON_ONCE(1); 1264 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 1265 return; 1266 } 1267 1268 if (atomic_read(&transport->xprt.swapper)) 1269 sk_clear_memalloc(sk); 1270 1271 tls_handshake_cancel(sk); 1272 1273 kernel_sock_shutdown(sock, SHUT_RDWR); 1274 1275 mutex_lock(&transport->recv_mutex); 1276 lock_sock(sk); 1277 transport->inet = NULL; 1278 transport->sock = NULL; 1279 transport->file = NULL; 1280 1281 sk->sk_user_data = NULL; 1282 sk->sk_sndtimeo = 0; 1283 1284 xs_restore_old_callbacks(transport, sk); 1285 xprt_clear_connected(xprt); 1286 xs_sock_reset_connection_flags(xprt); 1287 /* Reset stream record info */ 1288 xs_stream_reset_connect(transport); 1289 release_sock(sk); 1290 mutex_unlock(&transport->recv_mutex); 1291 1292 trace_rpc_socket_close(xprt, sock); 1293 __fput_sync(filp); 1294 1295 xprt_disconnect_done(xprt); 1296 } 1297 1298 /** 1299 * xs_close - close a socket 1300 * @xprt: transport 1301 * 1302 * This is used when all requests are complete; ie, no DRC state remains 1303 * on the server we want to save. 1304 * 1305 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 1306 * xs_reset_transport() zeroing the socket from underneath a writer. 1307 */ 1308 static void xs_close(struct rpc_xprt *xprt) 1309 { 1310 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1311 1312 dprintk("RPC: xs_close xprt %p\n", xprt); 1313 1314 if (transport->sock) 1315 tls_handshake_close(transport->sock); 1316 xs_reset_transport(transport); 1317 xprt->reestablish_timeout = 0; 1318 } 1319 1320 static void xs_inject_disconnect(struct rpc_xprt *xprt) 1321 { 1322 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 1323 xprt); 1324 xprt_disconnect_done(xprt); 1325 } 1326 1327 static void xs_xprt_free(struct rpc_xprt *xprt) 1328 { 1329 xs_free_peer_addresses(xprt); 1330 xprt_free(xprt); 1331 } 1332 1333 /** 1334 * xs_destroy - prepare to shutdown a transport 1335 * @xprt: doomed transport 1336 * 1337 */ 1338 static void xs_destroy(struct rpc_xprt *xprt) 1339 { 1340 struct sock_xprt *transport = container_of(xprt, 1341 struct sock_xprt, xprt); 1342 dprintk("RPC: xs_destroy xprt %p\n", xprt); 1343 1344 cancel_delayed_work_sync(&transport->connect_worker); 1345 xs_close(xprt); 1346 cancel_work_sync(&transport->recv_worker); 1347 cancel_work_sync(&transport->error_worker); 1348 xs_xprt_free(xprt); 1349 module_put(THIS_MODULE); 1350 } 1351 1352 /** 1353 * xs_udp_data_read_skb - receive callback for UDP sockets 1354 * @xprt: transport 1355 * @sk: socket 1356 * @skb: skbuff 1357 * 1358 */ 1359 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1360 struct sock *sk, 1361 struct sk_buff *skb) 1362 { 1363 struct rpc_task *task; 1364 struct rpc_rqst *rovr; 1365 int repsize, copied; 1366 u32 _xid; 1367 __be32 *xp; 1368 1369 repsize = skb->len; 1370 if (repsize < 4) { 1371 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1372 return; 1373 } 1374 1375 /* Copy the XID from the skb... */ 1376 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid); 1377 if (xp == NULL) 1378 return; 1379 1380 /* Look up and lock the request corresponding to the given XID */ 1381 spin_lock(&xprt->queue_lock); 1382 rovr = xprt_lookup_rqst(xprt, *xp); 1383 if (!rovr) 1384 goto out_unlock; 1385 xprt_pin_rqst(rovr); 1386 xprt_update_rtt(rovr->rq_task); 1387 spin_unlock(&xprt->queue_lock); 1388 task = rovr->rq_task; 1389 1390 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1391 copied = repsize; 1392 1393 /* Suck it into the iovec, verify checksum if not done by hw. */ 1394 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1395 spin_lock(&xprt->queue_lock); 1396 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS); 1397 goto out_unpin; 1398 } 1399 1400 1401 spin_lock(&xprt->transport_lock); 1402 xprt_adjust_cwnd(xprt, task, copied); 1403 spin_unlock(&xprt->transport_lock); 1404 spin_lock(&xprt->queue_lock); 1405 xprt_complete_rqst(task, copied); 1406 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS); 1407 out_unpin: 1408 xprt_unpin_rqst(rovr); 1409 out_unlock: 1410 spin_unlock(&xprt->queue_lock); 1411 } 1412 1413 static void xs_udp_data_receive(struct sock_xprt *transport) 1414 { 1415 struct sk_buff *skb; 1416 struct sock *sk; 1417 int err; 1418 1419 mutex_lock(&transport->recv_mutex); 1420 sk = transport->inet; 1421 if (sk == NULL) 1422 goto out; 1423 for (;;) { 1424 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1425 if (skb == NULL) 1426 break; 1427 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1428 consume_skb(skb); 1429 cond_resched(); 1430 } 1431 xs_poll_check_readable(transport); 1432 out: 1433 mutex_unlock(&transport->recv_mutex); 1434 } 1435 1436 static void xs_udp_data_receive_workfn(struct work_struct *work) 1437 { 1438 struct sock_xprt *transport = 1439 container_of(work, struct sock_xprt, recv_worker); 1440 unsigned int pflags = memalloc_nofs_save(); 1441 1442 xs_udp_data_receive(transport); 1443 memalloc_nofs_restore(pflags); 1444 } 1445 1446 /** 1447 * xs_data_ready - "data ready" callback for sockets 1448 * @sk: socket with data to read 1449 * 1450 */ 1451 static void xs_data_ready(struct sock *sk) 1452 { 1453 struct rpc_xprt *xprt; 1454 1455 trace_sk_data_ready(sk); 1456 1457 xprt = xprt_from_sock(sk); 1458 if (xprt != NULL) { 1459 struct sock_xprt *transport = container_of(xprt, 1460 struct sock_xprt, xprt); 1461 1462 trace_xs_data_ready(xprt); 1463 1464 transport->old_data_ready(sk); 1465 1466 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state)) 1467 return; 1468 1469 /* Any data means we had a useful conversation, so 1470 * then we don't need to delay the next reconnect 1471 */ 1472 if (xprt->reestablish_timeout) 1473 xprt->reestablish_timeout = 0; 1474 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1475 queue_work(xprtiod_workqueue, &transport->recv_worker); 1476 } 1477 } 1478 1479 /* 1480 * Helper function to force a TCP close if the server is sending 1481 * junk and/or it has put us in CLOSE_WAIT 1482 */ 1483 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1484 { 1485 xprt_force_disconnect(xprt); 1486 } 1487 1488 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1489 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt) 1490 { 1491 return PAGE_SIZE; 1492 } 1493 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1494 1495 /** 1496 * xs_local_state_change - callback to handle AF_LOCAL socket state changes 1497 * @sk: socket whose state has changed 1498 * 1499 */ 1500 static void xs_local_state_change(struct sock *sk) 1501 { 1502 struct rpc_xprt *xprt; 1503 struct sock_xprt *transport; 1504 1505 if (!(xprt = xprt_from_sock(sk))) 1506 return; 1507 transport = container_of(xprt, struct sock_xprt, xprt); 1508 if (sk->sk_shutdown & SHUTDOWN_MASK) { 1509 clear_bit(XPRT_CONNECTED, &xprt->state); 1510 /* Trigger the socket release */ 1511 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1512 } 1513 } 1514 1515 /** 1516 * xs_tcp_state_change - callback to handle TCP socket state changes 1517 * @sk: socket whose state has changed 1518 * 1519 */ 1520 static void xs_tcp_state_change(struct sock *sk) 1521 { 1522 struct rpc_xprt *xprt; 1523 struct sock_xprt *transport; 1524 1525 if (!(xprt = xprt_from_sock(sk))) 1526 return; 1527 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1528 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1529 sk->sk_state, xprt_connected(xprt), 1530 sock_flag(sk, SOCK_DEAD), 1531 sock_flag(sk, SOCK_ZAPPED), 1532 sk->sk_shutdown); 1533 1534 transport = container_of(xprt, struct sock_xprt, xprt); 1535 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1536 switch (sk->sk_state) { 1537 case TCP_ESTABLISHED: 1538 if (!xprt_test_and_set_connected(xprt)) { 1539 xprt->connect_cookie++; 1540 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1541 xprt_clear_connecting(xprt); 1542 1543 xprt->stat.connect_count++; 1544 xprt->stat.connect_time += (long)jiffies - 1545 xprt->stat.connect_start; 1546 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING); 1547 } 1548 break; 1549 case TCP_FIN_WAIT1: 1550 /* The client initiated a shutdown of the socket */ 1551 xprt->connect_cookie++; 1552 xprt->reestablish_timeout = 0; 1553 set_bit(XPRT_CLOSING, &xprt->state); 1554 smp_mb__before_atomic(); 1555 clear_bit(XPRT_CONNECTED, &xprt->state); 1556 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1557 smp_mb__after_atomic(); 1558 break; 1559 case TCP_CLOSE_WAIT: 1560 /* The server initiated a shutdown of the socket */ 1561 xprt->connect_cookie++; 1562 clear_bit(XPRT_CONNECTED, &xprt->state); 1563 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1564 fallthrough; 1565 case TCP_CLOSING: 1566 /* 1567 * If the server closed down the connection, make sure that 1568 * we back off before reconnecting 1569 */ 1570 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1571 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1572 break; 1573 case TCP_LAST_ACK: 1574 set_bit(XPRT_CLOSING, &xprt->state); 1575 smp_mb__before_atomic(); 1576 clear_bit(XPRT_CONNECTED, &xprt->state); 1577 smp_mb__after_atomic(); 1578 break; 1579 case TCP_CLOSE: 1580 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1581 &transport->sock_state)) { 1582 xs_reset_srcport(transport); 1583 xprt_clear_connecting(xprt); 1584 } 1585 clear_bit(XPRT_CLOSING, &xprt->state); 1586 /* Trigger the socket release */ 1587 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1588 } 1589 } 1590 1591 static void xs_write_space(struct sock *sk) 1592 { 1593 struct sock_xprt *transport; 1594 struct rpc_xprt *xprt; 1595 1596 if (!sk->sk_socket) 1597 return; 1598 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1599 1600 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1601 return; 1602 transport = container_of(xprt, struct sock_xprt, xprt); 1603 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state)) 1604 return; 1605 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE); 1606 sk->sk_write_pending--; 1607 } 1608 1609 /** 1610 * xs_udp_write_space - callback invoked when socket buffer space 1611 * becomes available 1612 * @sk: socket whose state has changed 1613 * 1614 * Called when more output buffer space is available for this socket. 1615 * We try not to wake our writers until they can make "significant" 1616 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1617 * with a bunch of small requests. 1618 */ 1619 static void xs_udp_write_space(struct sock *sk) 1620 { 1621 /* from net/core/sock.c:sock_def_write_space */ 1622 if (sock_writeable(sk)) 1623 xs_write_space(sk); 1624 } 1625 1626 /** 1627 * xs_tcp_write_space - callback invoked when socket buffer space 1628 * becomes available 1629 * @sk: socket whose state has changed 1630 * 1631 * Called when more output buffer space is available for this socket. 1632 * We try not to wake our writers until they can make "significant" 1633 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1634 * with a bunch of small requests. 1635 */ 1636 static void xs_tcp_write_space(struct sock *sk) 1637 { 1638 /* from net/core/stream.c:sk_stream_write_space */ 1639 if (sk_stream_is_writeable(sk)) 1640 xs_write_space(sk); 1641 } 1642 1643 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1644 { 1645 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1646 struct sock *sk = transport->inet; 1647 1648 if (transport->rcvsize) { 1649 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1650 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1651 } 1652 if (transport->sndsize) { 1653 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1654 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1655 sk->sk_write_space(sk); 1656 } 1657 } 1658 1659 /** 1660 * xs_udp_set_buffer_size - set send and receive limits 1661 * @xprt: generic transport 1662 * @sndsize: requested size of send buffer, in bytes 1663 * @rcvsize: requested size of receive buffer, in bytes 1664 * 1665 * Set socket send and receive buffer size limits. 1666 */ 1667 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1668 { 1669 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1670 1671 transport->sndsize = 0; 1672 if (sndsize) 1673 transport->sndsize = sndsize + 1024; 1674 transport->rcvsize = 0; 1675 if (rcvsize) 1676 transport->rcvsize = rcvsize + 1024; 1677 1678 xs_udp_do_set_buffer_size(xprt); 1679 } 1680 1681 /** 1682 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1683 * @xprt: controlling transport 1684 * @task: task that timed out 1685 * 1686 * Adjust the congestion window after a retransmit timeout has occurred. 1687 */ 1688 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1689 { 1690 spin_lock(&xprt->transport_lock); 1691 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1692 spin_unlock(&xprt->transport_lock); 1693 } 1694 1695 static int xs_get_random_port(void) 1696 { 1697 unsigned short min = xprt_min_resvport, max = xprt_max_resvport; 1698 unsigned short range; 1699 unsigned short rand; 1700 1701 if (max < min) 1702 return -EADDRINUSE; 1703 range = max - min + 1; 1704 rand = get_random_u32_below(range); 1705 return rand + min; 1706 } 1707 1708 static unsigned short xs_sock_getport(struct socket *sock) 1709 { 1710 struct sockaddr_storage buf; 1711 unsigned short port = 0; 1712 1713 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0) 1714 goto out; 1715 switch (buf.ss_family) { 1716 case AF_INET6: 1717 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1718 break; 1719 case AF_INET: 1720 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1721 } 1722 out: 1723 return port; 1724 } 1725 1726 /** 1727 * xs_set_port - reset the port number in the remote endpoint address 1728 * @xprt: generic transport 1729 * @port: new port number 1730 * 1731 */ 1732 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1733 { 1734 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1735 1736 rpc_set_port(xs_addr(xprt), port); 1737 xs_update_peer_port(xprt); 1738 } 1739 1740 static void xs_reset_srcport(struct sock_xprt *transport) 1741 { 1742 transport->srcport = 0; 1743 } 1744 1745 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1746 { 1747 if (transport->srcport == 0 && transport->xprt.reuseport) 1748 transport->srcport = xs_sock_getport(sock); 1749 } 1750 1751 static int xs_get_srcport(struct sock_xprt *transport) 1752 { 1753 int port = transport->srcport; 1754 1755 if (port == 0 && transport->xprt.resvport) 1756 port = xs_get_random_port(); 1757 return port; 1758 } 1759 1760 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt) 1761 { 1762 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1763 unsigned short ret = 0; 1764 mutex_lock(&sock->recv_mutex); 1765 if (sock->sock) 1766 ret = xs_sock_getport(sock->sock); 1767 mutex_unlock(&sock->recv_mutex); 1768 return ret; 1769 } 1770 1771 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen) 1772 { 1773 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1774 union { 1775 struct sockaddr sa; 1776 struct sockaddr_storage st; 1777 } saddr; 1778 int ret = -ENOTCONN; 1779 1780 mutex_lock(&sock->recv_mutex); 1781 if (sock->sock) { 1782 ret = kernel_getsockname(sock->sock, &saddr.sa); 1783 if (ret >= 0) 1784 ret = snprintf(buf, buflen, "%pISc", &saddr.sa); 1785 } 1786 mutex_unlock(&sock->recv_mutex); 1787 return ret; 1788 } 1789 1790 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1791 { 1792 if (transport->srcport != 0) 1793 transport->srcport = 0; 1794 if (!transport->xprt.resvport) 1795 return 0; 1796 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1797 return xprt_max_resvport; 1798 return --port; 1799 } 1800 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1801 { 1802 struct sockaddr_storage myaddr; 1803 int err, nloop = 0; 1804 int port = xs_get_srcport(transport); 1805 unsigned short last; 1806 1807 /* 1808 * If we are asking for any ephemeral port (i.e. port == 0 && 1809 * transport->xprt.resvport == 0), don't bind. Let the local 1810 * port selection happen implicitly when the socket is used 1811 * (for example at connect time). 1812 * 1813 * This ensures that we can continue to establish TCP 1814 * connections even when all local ephemeral ports are already 1815 * a part of some TCP connection. This makes no difference 1816 * for UDP sockets, but also doesn't harm them. 1817 * 1818 * If we're asking for any reserved port (i.e. port == 0 && 1819 * transport->xprt.resvport == 1) xs_get_srcport above will 1820 * ensure that port is non-zero and we will bind as needed. 1821 */ 1822 if (port <= 0) 1823 return port; 1824 1825 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1826 do { 1827 rpc_set_port((struct sockaddr *)&myaddr, port); 1828 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1829 transport->xprt.addrlen); 1830 if (err == 0) { 1831 if (transport->xprt.reuseport) 1832 transport->srcport = port; 1833 break; 1834 } 1835 last = port; 1836 port = xs_next_srcport(transport, port); 1837 if (port > last) 1838 nloop++; 1839 } while (err == -EADDRINUSE && nloop != 2); 1840 1841 if (myaddr.ss_family == AF_INET) 1842 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1843 &((struct sockaddr_in *)&myaddr)->sin_addr, 1844 port, err ? "failed" : "ok", err); 1845 else 1846 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1847 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1848 port, err ? "failed" : "ok", err); 1849 return err; 1850 } 1851 1852 /* 1853 * We don't support autobind on AF_LOCAL sockets 1854 */ 1855 static void xs_local_rpcbind(struct rpc_task *task) 1856 { 1857 xprt_set_bound(task->tk_xprt); 1858 } 1859 1860 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1861 { 1862 } 1863 1864 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1865 static struct lock_class_key xs_key[3]; 1866 static struct lock_class_key xs_slock_key[3]; 1867 1868 static inline void xs_reclassify_socketu(struct socket *sock) 1869 { 1870 struct sock *sk = sock->sk; 1871 1872 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1873 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]); 1874 } 1875 1876 static inline void xs_reclassify_socket4(struct socket *sock) 1877 { 1878 struct sock *sk = sock->sk; 1879 1880 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1881 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]); 1882 } 1883 1884 static inline void xs_reclassify_socket6(struct socket *sock) 1885 { 1886 struct sock *sk = sock->sk; 1887 1888 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1889 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]); 1890 } 1891 1892 static inline void xs_reclassify_socket(int family, struct socket *sock) 1893 { 1894 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk))) 1895 return; 1896 1897 switch (family) { 1898 case AF_LOCAL: 1899 xs_reclassify_socketu(sock); 1900 break; 1901 case AF_INET: 1902 xs_reclassify_socket4(sock); 1903 break; 1904 case AF_INET6: 1905 xs_reclassify_socket6(sock); 1906 break; 1907 } 1908 } 1909 #else 1910 static inline void xs_reclassify_socket(int family, struct socket *sock) 1911 { 1912 } 1913 #endif 1914 1915 static void xs_dummy_setup_socket(struct work_struct *work) 1916 { 1917 } 1918 1919 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1920 struct sock_xprt *transport, int family, int type, 1921 int protocol, bool reuseport) 1922 { 1923 struct file *filp; 1924 struct socket *sock; 1925 int err; 1926 1927 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1928 if (err < 0) { 1929 dprintk("RPC: can't create %d transport socket (%d).\n", 1930 protocol, -err); 1931 goto out; 1932 } 1933 xs_reclassify_socket(family, sock); 1934 1935 if (reuseport) 1936 sock_set_reuseport(sock->sk); 1937 1938 err = xs_bind(transport, sock); 1939 if (err) { 1940 sock_release(sock); 1941 goto out; 1942 } 1943 1944 if (protocol == IPPROTO_TCP) { 1945 __netns_tracker_free(xprt->xprt_net, &sock->sk->ns_tracker, false); 1946 sock->sk->sk_net_refcnt = 1; 1947 get_net_track(xprt->xprt_net, &sock->sk->ns_tracker, GFP_KERNEL); 1948 sock_inuse_add(xprt->xprt_net, 1); 1949 } 1950 1951 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1952 if (IS_ERR(filp)) 1953 return ERR_CAST(filp); 1954 transport->file = filp; 1955 1956 return sock; 1957 out: 1958 return ERR_PTR(err); 1959 } 1960 1961 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1962 struct socket *sock) 1963 { 1964 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1965 xprt); 1966 1967 if (!transport->inet) { 1968 struct sock *sk = sock->sk; 1969 1970 lock_sock(sk); 1971 1972 xs_save_old_callbacks(transport, sk); 1973 1974 sk->sk_user_data = xprt; 1975 sk->sk_data_ready = xs_data_ready; 1976 sk->sk_write_space = xs_udp_write_space; 1977 sk->sk_state_change = xs_local_state_change; 1978 sk->sk_error_report = xs_error_report; 1979 sk->sk_use_task_frag = false; 1980 1981 xprt_clear_connected(xprt); 1982 1983 /* Reset to new socket */ 1984 transport->sock = sock; 1985 transport->inet = sk; 1986 1987 release_sock(sk); 1988 } 1989 1990 xs_stream_start_connect(transport); 1991 1992 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1993 } 1994 1995 /** 1996 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1997 * @transport: socket transport to connect 1998 */ 1999 static int xs_local_setup_socket(struct sock_xprt *transport) 2000 { 2001 struct rpc_xprt *xprt = &transport->xprt; 2002 struct file *filp; 2003 struct socket *sock; 2004 int status; 2005 2006 status = __sock_create(xprt->xprt_net, AF_LOCAL, 2007 SOCK_STREAM, 0, &sock, 1); 2008 if (status < 0) { 2009 dprintk("RPC: can't create AF_LOCAL " 2010 "transport socket (%d).\n", -status); 2011 goto out; 2012 } 2013 xs_reclassify_socket(AF_LOCAL, sock); 2014 2015 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 2016 if (IS_ERR(filp)) { 2017 status = PTR_ERR(filp); 2018 goto out; 2019 } 2020 transport->file = filp; 2021 2022 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 2023 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2024 2025 status = xs_local_finish_connecting(xprt, sock); 2026 trace_rpc_socket_connect(xprt, sock, status); 2027 switch (status) { 2028 case 0: 2029 dprintk("RPC: xprt %p connected to %s\n", 2030 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2031 xprt->stat.connect_count++; 2032 xprt->stat.connect_time += (long)jiffies - 2033 xprt->stat.connect_start; 2034 xprt_set_connected(xprt); 2035 break; 2036 case -ENOBUFS: 2037 break; 2038 case -ENOENT: 2039 dprintk("RPC: xprt %p: socket %s does not exist\n", 2040 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2041 break; 2042 case -ECONNREFUSED: 2043 dprintk("RPC: xprt %p: connection refused for %s\n", 2044 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2045 break; 2046 default: 2047 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 2048 __func__, -status, 2049 xprt->address_strings[RPC_DISPLAY_ADDR]); 2050 } 2051 2052 out: 2053 xprt_clear_connecting(xprt); 2054 xprt_wake_pending_tasks(xprt, status); 2055 return status; 2056 } 2057 2058 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2059 { 2060 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2061 int ret; 2062 2063 if (transport->file) 2064 goto force_disconnect; 2065 2066 if (RPC_IS_ASYNC(task)) { 2067 /* 2068 * We want the AF_LOCAL connect to be resolved in the 2069 * filesystem namespace of the process making the rpc 2070 * call. Thus we connect synchronously. 2071 * 2072 * If we want to support asynchronous AF_LOCAL calls, 2073 * we'll need to figure out how to pass a namespace to 2074 * connect. 2075 */ 2076 rpc_task_set_rpc_status(task, -ENOTCONN); 2077 goto out_wake; 2078 } 2079 ret = xs_local_setup_socket(transport); 2080 if (ret && !RPC_IS_SOFTCONN(task)) 2081 msleep_interruptible(15000); 2082 return; 2083 force_disconnect: 2084 xprt_force_disconnect(xprt); 2085 out_wake: 2086 xprt_clear_connecting(xprt); 2087 xprt_wake_pending_tasks(xprt, -ENOTCONN); 2088 } 2089 2090 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 2091 /* 2092 * Note that this should be called with XPRT_LOCKED held, or recv_mutex 2093 * held, or when we otherwise know that we have exclusive access to the 2094 * socket, to guard against races with xs_reset_transport. 2095 */ 2096 static void xs_set_memalloc(struct rpc_xprt *xprt) 2097 { 2098 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2099 xprt); 2100 2101 /* 2102 * If there's no sock, then we have nothing to set. The 2103 * reconnecting process will get it for us. 2104 */ 2105 if (!transport->inet) 2106 return; 2107 if (atomic_read(&xprt->swapper)) 2108 sk_set_memalloc(transport->inet); 2109 } 2110 2111 /** 2112 * xs_enable_swap - Tag this transport as being used for swap. 2113 * @xprt: transport to tag 2114 * 2115 * Take a reference to this transport on behalf of the rpc_clnt, and 2116 * optionally mark it for swapping if it wasn't already. 2117 */ 2118 static int 2119 xs_enable_swap(struct rpc_xprt *xprt) 2120 { 2121 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2122 2123 mutex_lock(&xs->recv_mutex); 2124 if (atomic_inc_return(&xprt->swapper) == 1 && 2125 xs->inet) 2126 sk_set_memalloc(xs->inet); 2127 mutex_unlock(&xs->recv_mutex); 2128 return 0; 2129 } 2130 2131 /** 2132 * xs_disable_swap - Untag this transport as being used for swap. 2133 * @xprt: transport to tag 2134 * 2135 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2136 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2137 */ 2138 static void 2139 xs_disable_swap(struct rpc_xprt *xprt) 2140 { 2141 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2142 2143 mutex_lock(&xs->recv_mutex); 2144 if (atomic_dec_and_test(&xprt->swapper) && 2145 xs->inet) 2146 sk_clear_memalloc(xs->inet); 2147 mutex_unlock(&xs->recv_mutex); 2148 } 2149 #else 2150 static void xs_set_memalloc(struct rpc_xprt *xprt) 2151 { 2152 } 2153 2154 static int 2155 xs_enable_swap(struct rpc_xprt *xprt) 2156 { 2157 return -EINVAL; 2158 } 2159 2160 static void 2161 xs_disable_swap(struct rpc_xprt *xprt) 2162 { 2163 } 2164 #endif 2165 2166 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2167 { 2168 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2169 2170 if (!transport->inet) { 2171 struct sock *sk = sock->sk; 2172 2173 lock_sock(sk); 2174 2175 xs_save_old_callbacks(transport, sk); 2176 2177 sk->sk_user_data = xprt; 2178 sk->sk_data_ready = xs_data_ready; 2179 sk->sk_write_space = xs_udp_write_space; 2180 sk->sk_use_task_frag = false; 2181 2182 xprt_set_connected(xprt); 2183 2184 /* Reset to new socket */ 2185 transport->sock = sock; 2186 transport->inet = sk; 2187 2188 xs_set_memalloc(xprt); 2189 2190 release_sock(sk); 2191 } 2192 xs_udp_do_set_buffer_size(xprt); 2193 2194 xprt->stat.connect_start = jiffies; 2195 } 2196 2197 static void xs_udp_setup_socket(struct work_struct *work) 2198 { 2199 struct sock_xprt *transport = 2200 container_of(work, struct sock_xprt, connect_worker.work); 2201 struct rpc_xprt *xprt = &transport->xprt; 2202 struct socket *sock; 2203 int status = -EIO; 2204 unsigned int pflags = current->flags; 2205 2206 if (atomic_read(&xprt->swapper)) 2207 current->flags |= PF_MEMALLOC; 2208 sock = xs_create_sock(xprt, transport, 2209 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2210 IPPROTO_UDP, false); 2211 if (IS_ERR(sock)) 2212 goto out; 2213 2214 dprintk("RPC: worker connecting xprt %p via %s to " 2215 "%s (port %s)\n", xprt, 2216 xprt->address_strings[RPC_DISPLAY_PROTO], 2217 xprt->address_strings[RPC_DISPLAY_ADDR], 2218 xprt->address_strings[RPC_DISPLAY_PORT]); 2219 2220 xs_udp_finish_connecting(xprt, sock); 2221 trace_rpc_socket_connect(xprt, sock, 0); 2222 status = 0; 2223 out: 2224 xprt_clear_connecting(xprt); 2225 xprt_unlock_connect(xprt, transport); 2226 xprt_wake_pending_tasks(xprt, status); 2227 current_restore_flags(pflags, PF_MEMALLOC); 2228 } 2229 2230 /** 2231 * xs_tcp_shutdown - gracefully shut down a TCP socket 2232 * @xprt: transport 2233 * 2234 * Initiates a graceful shutdown of the TCP socket by calling the 2235 * equivalent of shutdown(SHUT_RDWR); 2236 */ 2237 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2238 { 2239 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2240 struct socket *sock = transport->sock; 2241 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE; 2242 2243 if (sock == NULL) 2244 return; 2245 if (!xprt->reuseport) { 2246 xs_close(xprt); 2247 return; 2248 } 2249 switch (skst) { 2250 case TCP_FIN_WAIT1: 2251 case TCP_FIN_WAIT2: 2252 case TCP_LAST_ACK: 2253 break; 2254 case TCP_ESTABLISHED: 2255 case TCP_CLOSE_WAIT: 2256 kernel_sock_shutdown(sock, SHUT_RDWR); 2257 trace_rpc_socket_shutdown(xprt, sock); 2258 break; 2259 default: 2260 xs_reset_transport(transport); 2261 } 2262 } 2263 2264 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 2265 struct socket *sock) 2266 { 2267 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2268 struct net *net = sock_net(sock->sk); 2269 unsigned long connect_timeout; 2270 unsigned long syn_retries; 2271 unsigned int keepidle; 2272 unsigned int keepcnt; 2273 unsigned int timeo; 2274 unsigned long t; 2275 2276 spin_lock(&xprt->transport_lock); 2277 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ); 2278 keepcnt = xprt->timeout->to_retries + 1; 2279 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2280 (xprt->timeout->to_retries + 1); 2281 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2282 spin_unlock(&xprt->transport_lock); 2283 2284 /* TCP Keepalive options */ 2285 sock_set_keepalive(sock->sk); 2286 tcp_sock_set_keepidle(sock->sk, keepidle); 2287 tcp_sock_set_keepintvl(sock->sk, keepidle); 2288 tcp_sock_set_keepcnt(sock->sk, keepcnt); 2289 2290 /* TCP user timeout (see RFC5482) */ 2291 tcp_sock_set_user_timeout(sock->sk, timeo); 2292 2293 /* Connect timeout */ 2294 connect_timeout = max_t(unsigned long, 2295 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1); 2296 syn_retries = max_t(unsigned long, 2297 READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1); 2298 for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++) 2299 ; 2300 if (t <= syn_retries) 2301 tcp_sock_set_syncnt(sock->sk, t - 1); 2302 } 2303 2304 static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt, 2305 unsigned long connect_timeout) 2306 { 2307 struct sock_xprt *transport = 2308 container_of(xprt, struct sock_xprt, xprt); 2309 struct rpc_timeout to; 2310 unsigned long initval; 2311 2312 memcpy(&to, xprt->timeout, sizeof(to)); 2313 /* Arbitrary lower limit */ 2314 initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO); 2315 to.to_initval = initval; 2316 to.to_maxval = initval; 2317 to.to_retries = 0; 2318 memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout)); 2319 xprt->timeout = &transport->tcp_timeout; 2320 xprt->connect_timeout = connect_timeout; 2321 } 2322 2323 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt, 2324 unsigned long connect_timeout, 2325 unsigned long reconnect_timeout) 2326 { 2327 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2328 2329 spin_lock(&xprt->transport_lock); 2330 if (reconnect_timeout < xprt->max_reconnect_timeout) 2331 xprt->max_reconnect_timeout = reconnect_timeout; 2332 if (connect_timeout < xprt->connect_timeout) 2333 xs_tcp_do_set_connect_timeout(xprt, connect_timeout); 2334 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2335 spin_unlock(&xprt->transport_lock); 2336 } 2337 2338 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2339 { 2340 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2341 2342 if (!transport->inet) { 2343 struct sock *sk = sock->sk; 2344 2345 /* Avoid temporary address, they are bad for long-lived 2346 * connections such as NFS mounts. 2347 * RFC4941, section 3.6 suggests that: 2348 * Individual applications, which have specific 2349 * knowledge about the normal duration of connections, 2350 * MAY override this as appropriate. 2351 */ 2352 if (xs_addr(xprt)->sa_family == PF_INET6) { 2353 ip6_sock_set_addr_preferences(sk, 2354 IPV6_PREFER_SRC_PUBLIC); 2355 } 2356 2357 xs_tcp_set_socket_timeouts(xprt, sock); 2358 tcp_sock_set_nodelay(sk); 2359 2360 lock_sock(sk); 2361 2362 xs_save_old_callbacks(transport, sk); 2363 2364 sk->sk_user_data = xprt; 2365 sk->sk_data_ready = xs_data_ready; 2366 sk->sk_state_change = xs_tcp_state_change; 2367 sk->sk_write_space = xs_tcp_write_space; 2368 sk->sk_error_report = xs_error_report; 2369 sk->sk_use_task_frag = false; 2370 2371 /* socket options */ 2372 sock_reset_flag(sk, SOCK_LINGER); 2373 2374 xprt_clear_connected(xprt); 2375 2376 /* Reset to new socket */ 2377 transport->sock = sock; 2378 transport->inet = sk; 2379 2380 release_sock(sk); 2381 } 2382 2383 if (!xprt_bound(xprt)) 2384 return -ENOTCONN; 2385 2386 xs_set_memalloc(xprt); 2387 2388 xs_stream_start_connect(transport); 2389 2390 /* Tell the socket layer to start connecting... */ 2391 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2392 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2393 } 2394 2395 /** 2396 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2397 * @work: queued work item 2398 * 2399 * Invoked by a work queue tasklet. 2400 */ 2401 static void xs_tcp_setup_socket(struct work_struct *work) 2402 { 2403 struct sock_xprt *transport = 2404 container_of(work, struct sock_xprt, connect_worker.work); 2405 struct socket *sock = transport->sock; 2406 struct rpc_xprt *xprt = &transport->xprt; 2407 int status; 2408 unsigned int pflags = current->flags; 2409 2410 if (atomic_read(&xprt->swapper)) 2411 current->flags |= PF_MEMALLOC; 2412 2413 if (xprt_connected(xprt)) 2414 goto out; 2415 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT, 2416 &transport->sock_state) || 2417 !sock) { 2418 xs_reset_transport(transport); 2419 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family, 2420 SOCK_STREAM, IPPROTO_TCP, true); 2421 if (IS_ERR(sock)) { 2422 xprt_wake_pending_tasks(xprt, PTR_ERR(sock)); 2423 goto out; 2424 } 2425 } 2426 2427 dprintk("RPC: worker connecting xprt %p via %s to " 2428 "%s (port %s)\n", xprt, 2429 xprt->address_strings[RPC_DISPLAY_PROTO], 2430 xprt->address_strings[RPC_DISPLAY_ADDR], 2431 xprt->address_strings[RPC_DISPLAY_PORT]); 2432 2433 status = xs_tcp_finish_connecting(xprt, sock); 2434 trace_rpc_socket_connect(xprt, sock, status); 2435 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2436 xprt, -status, xprt_connected(xprt), 2437 sock->sk->sk_state); 2438 switch (status) { 2439 case 0: 2440 case -EINPROGRESS: 2441 /* SYN_SENT! */ 2442 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state); 2443 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2444 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2445 fallthrough; 2446 case -EALREADY: 2447 goto out_unlock; 2448 case -EADDRNOTAVAIL: 2449 /* Source port number is unavailable. Try a new one! */ 2450 transport->srcport = 0; 2451 status = -EAGAIN; 2452 break; 2453 case -EPERM: 2454 /* Happens, for instance, if a BPF program is preventing 2455 * the connect. Remap the error so upper layers can better 2456 * deal with it. 2457 */ 2458 status = -ECONNREFUSED; 2459 fallthrough; 2460 case -EINVAL: 2461 /* Happens, for instance, if the user specified a link 2462 * local IPv6 address without a scope-id. 2463 */ 2464 case -ECONNREFUSED: 2465 case -ECONNRESET: 2466 case -ENETDOWN: 2467 case -ENETUNREACH: 2468 case -EHOSTUNREACH: 2469 case -EADDRINUSE: 2470 case -ENOBUFS: 2471 case -ENOTCONN: 2472 break; 2473 default: 2474 printk("%s: connect returned unhandled error %d\n", 2475 __func__, status); 2476 status = -EAGAIN; 2477 } 2478 2479 /* xs_tcp_force_close() wakes tasks with a fixed error code. 2480 * We need to wake them first to ensure the correct error code. 2481 */ 2482 xprt_wake_pending_tasks(xprt, status); 2483 xs_tcp_force_close(xprt); 2484 out: 2485 xprt_clear_connecting(xprt); 2486 out_unlock: 2487 xprt_unlock_connect(xprt, transport); 2488 current_restore_flags(pflags, PF_MEMALLOC); 2489 } 2490 2491 /* 2492 * Transfer the connected socket to @upper_transport, then mark that 2493 * xprt CONNECTED. 2494 */ 2495 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt, 2496 struct sock_xprt *upper_transport) 2497 { 2498 struct sock_xprt *lower_transport = 2499 container_of(lower_xprt, struct sock_xprt, xprt); 2500 struct rpc_xprt *upper_xprt = &upper_transport->xprt; 2501 2502 if (!upper_transport->inet) { 2503 struct socket *sock = lower_transport->sock; 2504 struct sock *sk = sock->sk; 2505 2506 /* Avoid temporary address, they are bad for long-lived 2507 * connections such as NFS mounts. 2508 * RFC4941, section 3.6 suggests that: 2509 * Individual applications, which have specific 2510 * knowledge about the normal duration of connections, 2511 * MAY override this as appropriate. 2512 */ 2513 if (xs_addr(upper_xprt)->sa_family == PF_INET6) 2514 ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC); 2515 2516 xs_tcp_set_socket_timeouts(upper_xprt, sock); 2517 tcp_sock_set_nodelay(sk); 2518 2519 lock_sock(sk); 2520 2521 /* @sk is already connected, so it now has the RPC callbacks. 2522 * Reach into @lower_transport to save the original ones. 2523 */ 2524 upper_transport->old_data_ready = lower_transport->old_data_ready; 2525 upper_transport->old_state_change = lower_transport->old_state_change; 2526 upper_transport->old_write_space = lower_transport->old_write_space; 2527 upper_transport->old_error_report = lower_transport->old_error_report; 2528 sk->sk_user_data = upper_xprt; 2529 2530 /* socket options */ 2531 sock_reset_flag(sk, SOCK_LINGER); 2532 2533 xprt_clear_connected(upper_xprt); 2534 2535 upper_transport->sock = sock; 2536 upper_transport->inet = sk; 2537 upper_transport->file = lower_transport->file; 2538 2539 release_sock(sk); 2540 2541 /* Reset lower_transport before shutting down its clnt */ 2542 mutex_lock(&lower_transport->recv_mutex); 2543 lower_transport->inet = NULL; 2544 lower_transport->sock = NULL; 2545 lower_transport->file = NULL; 2546 2547 xprt_clear_connected(lower_xprt); 2548 xs_sock_reset_connection_flags(lower_xprt); 2549 xs_stream_reset_connect(lower_transport); 2550 mutex_unlock(&lower_transport->recv_mutex); 2551 } 2552 2553 if (!xprt_bound(upper_xprt)) 2554 return -ENOTCONN; 2555 2556 xs_set_memalloc(upper_xprt); 2557 2558 if (!xprt_test_and_set_connected(upper_xprt)) { 2559 upper_xprt->connect_cookie++; 2560 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state); 2561 xprt_clear_connecting(upper_xprt); 2562 2563 upper_xprt->stat.connect_count++; 2564 upper_xprt->stat.connect_time += (long)jiffies - 2565 upper_xprt->stat.connect_start; 2566 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING); 2567 } 2568 return 0; 2569 } 2570 2571 /** 2572 * xs_tls_handshake_done - TLS handshake completion handler 2573 * @data: address of xprt to wake 2574 * @status: status of handshake 2575 * @peerid: serial number of key containing the remote's identity 2576 * 2577 */ 2578 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid) 2579 { 2580 struct rpc_xprt *lower_xprt = data; 2581 struct sock_xprt *lower_transport = 2582 container_of(lower_xprt, struct sock_xprt, xprt); 2583 2584 lower_transport->xprt_err = status ? -EACCES : 0; 2585 complete(&lower_transport->handshake_done); 2586 xprt_put(lower_xprt); 2587 } 2588 2589 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec) 2590 { 2591 struct sock_xprt *lower_transport = 2592 container_of(lower_xprt, struct sock_xprt, xprt); 2593 struct tls_handshake_args args = { 2594 .ta_sock = lower_transport->sock, 2595 .ta_done = xs_tls_handshake_done, 2596 .ta_data = xprt_get(lower_xprt), 2597 .ta_peername = lower_xprt->servername, 2598 }; 2599 struct sock *sk = lower_transport->inet; 2600 int rc; 2601 2602 init_completion(&lower_transport->handshake_done); 2603 set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state); 2604 lower_transport->xprt_err = -ETIMEDOUT; 2605 switch (xprtsec->policy) { 2606 case RPC_XPRTSEC_TLS_ANON: 2607 rc = tls_client_hello_anon(&args, GFP_KERNEL); 2608 if (rc) 2609 goto out_put_xprt; 2610 break; 2611 case RPC_XPRTSEC_TLS_X509: 2612 args.ta_my_cert = xprtsec->cert_serial; 2613 args.ta_my_privkey = xprtsec->privkey_serial; 2614 rc = tls_client_hello_x509(&args, GFP_KERNEL); 2615 if (rc) 2616 goto out_put_xprt; 2617 break; 2618 default: 2619 rc = -EACCES; 2620 goto out_put_xprt; 2621 } 2622 2623 rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done, 2624 XS_TLS_HANDSHAKE_TO); 2625 if (rc <= 0) { 2626 tls_handshake_cancel(sk); 2627 if (rc == 0) 2628 rc = -ETIMEDOUT; 2629 goto out_put_xprt; 2630 } 2631 2632 rc = lower_transport->xprt_err; 2633 2634 out: 2635 xs_stream_reset_connect(lower_transport); 2636 clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state); 2637 return rc; 2638 2639 out_put_xprt: 2640 xprt_put(lower_xprt); 2641 goto out; 2642 } 2643 2644 /** 2645 * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket 2646 * @work: queued work item 2647 * 2648 * Invoked by a work queue tasklet. 2649 * 2650 * For RPC-with-TLS, there is a two-stage connection process. 2651 * 2652 * The "upper-layer xprt" is visible to the RPC consumer. Once it has 2653 * been marked connected, the consumer knows that a TCP connection and 2654 * a TLS session have been established. 2655 * 2656 * A "lower-layer xprt", created in this function, handles the mechanics 2657 * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and 2658 * then driving the TLS handshake. Once all that is complete, the upper 2659 * layer xprt is marked connected. 2660 */ 2661 static void xs_tcp_tls_setup_socket(struct work_struct *work) 2662 { 2663 struct sock_xprt *upper_transport = 2664 container_of(work, struct sock_xprt, connect_worker.work); 2665 struct rpc_clnt *upper_clnt = upper_transport->clnt; 2666 struct rpc_xprt *upper_xprt = &upper_transport->xprt; 2667 struct rpc_create_args args = { 2668 .net = upper_xprt->xprt_net, 2669 .protocol = upper_xprt->prot, 2670 .address = (struct sockaddr *)&upper_xprt->addr, 2671 .addrsize = upper_xprt->addrlen, 2672 .timeout = upper_clnt->cl_timeout, 2673 .servername = upper_xprt->servername, 2674 .program = upper_clnt->cl_program, 2675 .prognumber = upper_clnt->cl_prog, 2676 .version = upper_clnt->cl_vers, 2677 .authflavor = RPC_AUTH_TLS, 2678 .cred = upper_clnt->cl_cred, 2679 .xprtsec = { 2680 .policy = RPC_XPRTSEC_NONE, 2681 }, 2682 .stats = upper_clnt->cl_stats, 2683 }; 2684 unsigned int pflags = current->flags; 2685 struct rpc_clnt *lower_clnt; 2686 struct rpc_xprt *lower_xprt; 2687 int status; 2688 2689 if (atomic_read(&upper_xprt->swapper)) 2690 current->flags |= PF_MEMALLOC; 2691 2692 xs_stream_start_connect(upper_transport); 2693 2694 /* This implicitly sends an RPC_AUTH_TLS probe */ 2695 lower_clnt = rpc_create(&args); 2696 if (IS_ERR(lower_clnt)) { 2697 trace_rpc_tls_unavailable(upper_clnt, upper_xprt); 2698 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state); 2699 xprt_clear_connecting(upper_xprt); 2700 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt)); 2701 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING); 2702 goto out_unlock; 2703 } 2704 2705 /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on 2706 * the lower xprt. 2707 */ 2708 rcu_read_lock(); 2709 lower_xprt = rcu_dereference(lower_clnt->cl_xprt); 2710 rcu_read_unlock(); 2711 2712 if (wait_on_bit_lock(&lower_xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 2713 goto out_unlock; 2714 2715 status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec); 2716 if (status) { 2717 trace_rpc_tls_not_started(upper_clnt, upper_xprt); 2718 goto out_close; 2719 } 2720 2721 status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport); 2722 if (status) 2723 goto out_close; 2724 xprt_release_write(lower_xprt, NULL); 2725 2726 trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0); 2727 if (!xprt_test_and_set_connected(upper_xprt)) { 2728 upper_xprt->connect_cookie++; 2729 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state); 2730 xprt_clear_connecting(upper_xprt); 2731 2732 upper_xprt->stat.connect_count++; 2733 upper_xprt->stat.connect_time += (long)jiffies - 2734 upper_xprt->stat.connect_start; 2735 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING); 2736 } 2737 rpc_shutdown_client(lower_clnt); 2738 2739 out_unlock: 2740 current_restore_flags(pflags, PF_MEMALLOC); 2741 upper_transport->clnt = NULL; 2742 xprt_unlock_connect(upper_xprt, upper_transport); 2743 return; 2744 2745 out_close: 2746 xprt_release_write(lower_xprt, NULL); 2747 rpc_shutdown_client(lower_clnt); 2748 2749 /* xprt_force_disconnect() wakes tasks with a fixed tk_status code. 2750 * Wake them first here to ensure they get our tk_status code. 2751 */ 2752 xprt_wake_pending_tasks(upper_xprt, status); 2753 xs_tcp_force_close(upper_xprt); 2754 xprt_clear_connecting(upper_xprt); 2755 goto out_unlock; 2756 } 2757 2758 /** 2759 * xs_connect - connect a socket to a remote endpoint 2760 * @xprt: pointer to transport structure 2761 * @task: address of RPC task that manages state of connect request 2762 * 2763 * TCP: If the remote end dropped the connection, delay reconnecting. 2764 * 2765 * UDP socket connects are synchronous, but we use a work queue anyway 2766 * to guarantee that even unprivileged user processes can set up a 2767 * socket on a privileged port. 2768 * 2769 * If a UDP socket connect fails, the delay behavior here prevents 2770 * retry floods (hard mounts). 2771 */ 2772 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2773 { 2774 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2775 unsigned long delay = 0; 2776 2777 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2778 2779 if (transport->sock != NULL) { 2780 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2781 "seconds\n", xprt, xprt->reestablish_timeout / HZ); 2782 2783 delay = xprt_reconnect_delay(xprt); 2784 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO); 2785 2786 } else 2787 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2788 2789 transport->clnt = task->tk_client; 2790 queue_delayed_work(xprtiod_workqueue, 2791 &transport->connect_worker, 2792 delay); 2793 } 2794 2795 static void xs_wake_disconnect(struct sock_xprt *transport) 2796 { 2797 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state)) 2798 xs_tcp_force_close(&transport->xprt); 2799 } 2800 2801 static void xs_wake_write(struct sock_xprt *transport) 2802 { 2803 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state)) 2804 xprt_write_space(&transport->xprt); 2805 } 2806 2807 static void xs_wake_error(struct sock_xprt *transport) 2808 { 2809 int sockerr; 2810 2811 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2812 return; 2813 sockerr = xchg(&transport->xprt_err, 0); 2814 if (sockerr < 0) { 2815 xprt_wake_pending_tasks(&transport->xprt, sockerr); 2816 xs_tcp_force_close(&transport->xprt); 2817 } 2818 } 2819 2820 static void xs_wake_pending(struct sock_xprt *transport) 2821 { 2822 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state)) 2823 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN); 2824 } 2825 2826 static void xs_error_handle(struct work_struct *work) 2827 { 2828 struct sock_xprt *transport = container_of(work, 2829 struct sock_xprt, error_worker); 2830 2831 xs_wake_disconnect(transport); 2832 xs_wake_write(transport); 2833 xs_wake_error(transport); 2834 xs_wake_pending(transport); 2835 } 2836 2837 /** 2838 * xs_local_print_stats - display AF_LOCAL socket-specific stats 2839 * @xprt: rpc_xprt struct containing statistics 2840 * @seq: output file 2841 * 2842 */ 2843 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2844 { 2845 long idle_time = 0; 2846 2847 if (xprt_connected(xprt)) 2848 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2849 2850 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2851 "%llu %llu %lu %llu %llu\n", 2852 xprt->stat.bind_count, 2853 xprt->stat.connect_count, 2854 xprt->stat.connect_time / HZ, 2855 idle_time, 2856 xprt->stat.sends, 2857 xprt->stat.recvs, 2858 xprt->stat.bad_xids, 2859 xprt->stat.req_u, 2860 xprt->stat.bklog_u, 2861 xprt->stat.max_slots, 2862 xprt->stat.sending_u, 2863 xprt->stat.pending_u); 2864 } 2865 2866 /** 2867 * xs_udp_print_stats - display UDP socket-specific stats 2868 * @xprt: rpc_xprt struct containing statistics 2869 * @seq: output file 2870 * 2871 */ 2872 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2873 { 2874 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2875 2876 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2877 "%lu %llu %llu\n", 2878 transport->srcport, 2879 xprt->stat.bind_count, 2880 xprt->stat.sends, 2881 xprt->stat.recvs, 2882 xprt->stat.bad_xids, 2883 xprt->stat.req_u, 2884 xprt->stat.bklog_u, 2885 xprt->stat.max_slots, 2886 xprt->stat.sending_u, 2887 xprt->stat.pending_u); 2888 } 2889 2890 /** 2891 * xs_tcp_print_stats - display TCP socket-specific stats 2892 * @xprt: rpc_xprt struct containing statistics 2893 * @seq: output file 2894 * 2895 */ 2896 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2897 { 2898 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2899 long idle_time = 0; 2900 2901 if (xprt_connected(xprt)) 2902 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2903 2904 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2905 "%llu %llu %lu %llu %llu\n", 2906 transport->srcport, 2907 xprt->stat.bind_count, 2908 xprt->stat.connect_count, 2909 xprt->stat.connect_time / HZ, 2910 idle_time, 2911 xprt->stat.sends, 2912 xprt->stat.recvs, 2913 xprt->stat.bad_xids, 2914 xprt->stat.req_u, 2915 xprt->stat.bklog_u, 2916 xprt->stat.max_slots, 2917 xprt->stat.sending_u, 2918 xprt->stat.pending_u); 2919 } 2920 2921 /* 2922 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2923 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2924 * to use the server side send routines. 2925 */ 2926 static int bc_malloc(struct rpc_task *task) 2927 { 2928 struct rpc_rqst *rqst = task->tk_rqstp; 2929 size_t size = rqst->rq_callsize; 2930 struct page *page; 2931 struct rpc_buffer *buf; 2932 2933 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) { 2934 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n", 2935 size); 2936 return -EINVAL; 2937 } 2938 2939 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN); 2940 if (!page) 2941 return -ENOMEM; 2942 2943 buf = page_address(page); 2944 buf->len = PAGE_SIZE; 2945 2946 rqst->rq_buffer = buf->data; 2947 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; 2948 return 0; 2949 } 2950 2951 /* 2952 * Free the space allocated in the bc_alloc routine 2953 */ 2954 static void bc_free(struct rpc_task *task) 2955 { 2956 void *buffer = task->tk_rqstp->rq_buffer; 2957 struct rpc_buffer *buf; 2958 2959 buf = container_of(buffer, struct rpc_buffer, data); 2960 free_page((unsigned long)buf); 2961 } 2962 2963 static int bc_sendto(struct rpc_rqst *req) 2964 { 2965 struct xdr_buf *xdr = &req->rq_snd_buf; 2966 struct sock_xprt *transport = 2967 container_of(req->rq_xprt, struct sock_xprt, xprt); 2968 struct msghdr msg = { 2969 .msg_flags = 0, 2970 }; 2971 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 2972 (u32)xdr->len); 2973 unsigned int sent = 0; 2974 int err; 2975 2976 req->rq_xtime = ktime_get(); 2977 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask()); 2978 if (err < 0) 2979 return err; 2980 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent); 2981 xdr_free_bvec(xdr); 2982 if (err < 0 || sent != (xdr->len + sizeof(marker))) 2983 return -EAGAIN; 2984 return sent; 2985 } 2986 2987 /** 2988 * bc_send_request - Send a backchannel Call on a TCP socket 2989 * @req: rpc_rqst containing Call message to be sent 2990 * 2991 * xpt_mutex ensures @rqstp's whole message is written to the socket 2992 * without interruption. 2993 * 2994 * Return values: 2995 * %0 if the message was sent successfully 2996 * %ENOTCONN if the message was not sent 2997 */ 2998 static int bc_send_request(struct rpc_rqst *req) 2999 { 3000 struct svc_xprt *xprt; 3001 int len; 3002 3003 /* 3004 * Get the server socket associated with this callback xprt 3005 */ 3006 xprt = req->rq_xprt->bc_xprt; 3007 3008 /* 3009 * Grab the mutex to serialize data as the connection is shared 3010 * with the fore channel 3011 */ 3012 mutex_lock(&xprt->xpt_mutex); 3013 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 3014 len = -ENOTCONN; 3015 else 3016 len = bc_sendto(req); 3017 mutex_unlock(&xprt->xpt_mutex); 3018 3019 if (len > 0) 3020 len = 0; 3021 3022 return len; 3023 } 3024 3025 static void bc_close(struct rpc_xprt *xprt) 3026 { 3027 xprt_disconnect_done(xprt); 3028 } 3029 3030 static void bc_destroy(struct rpc_xprt *xprt) 3031 { 3032 dprintk("RPC: bc_destroy xprt %p\n", xprt); 3033 3034 xs_xprt_free(xprt); 3035 module_put(THIS_MODULE); 3036 } 3037 3038 static const struct rpc_xprt_ops xs_local_ops = { 3039 .reserve_xprt = xprt_reserve_xprt, 3040 .release_xprt = xprt_release_xprt, 3041 .alloc_slot = xprt_alloc_slot, 3042 .free_slot = xprt_free_slot, 3043 .rpcbind = xs_local_rpcbind, 3044 .set_port = xs_local_set_port, 3045 .connect = xs_local_connect, 3046 .buf_alloc = rpc_malloc, 3047 .buf_free = rpc_free, 3048 .prepare_request = xs_stream_prepare_request, 3049 .send_request = xs_local_send_request, 3050 .abort_send_request = xs_stream_abort_send_request, 3051 .wait_for_reply_request = xprt_wait_for_reply_request_def, 3052 .close = xs_close, 3053 .destroy = xs_destroy, 3054 .print_stats = xs_local_print_stats, 3055 .enable_swap = xs_enable_swap, 3056 .disable_swap = xs_disable_swap, 3057 }; 3058 3059 static const struct rpc_xprt_ops xs_udp_ops = { 3060 .set_buffer_size = xs_udp_set_buffer_size, 3061 .reserve_xprt = xprt_reserve_xprt_cong, 3062 .release_xprt = xprt_release_xprt_cong, 3063 .alloc_slot = xprt_alloc_slot, 3064 .free_slot = xprt_free_slot, 3065 .rpcbind = rpcb_getport_async, 3066 .set_port = xs_set_port, 3067 .connect = xs_connect, 3068 .get_srcaddr = xs_sock_srcaddr, 3069 .get_srcport = xs_sock_srcport, 3070 .buf_alloc = rpc_malloc, 3071 .buf_free = rpc_free, 3072 .send_request = xs_udp_send_request, 3073 .wait_for_reply_request = xprt_wait_for_reply_request_rtt, 3074 .timer = xs_udp_timer, 3075 .release_request = xprt_release_rqst_cong, 3076 .close = xs_close, 3077 .destroy = xs_destroy, 3078 .print_stats = xs_udp_print_stats, 3079 .enable_swap = xs_enable_swap, 3080 .disable_swap = xs_disable_swap, 3081 .inject_disconnect = xs_inject_disconnect, 3082 }; 3083 3084 static const struct rpc_xprt_ops xs_tcp_ops = { 3085 .reserve_xprt = xprt_reserve_xprt, 3086 .release_xprt = xprt_release_xprt, 3087 .alloc_slot = xprt_alloc_slot, 3088 .free_slot = xprt_free_slot, 3089 .rpcbind = rpcb_getport_async, 3090 .set_port = xs_set_port, 3091 .connect = xs_connect, 3092 .get_srcaddr = xs_sock_srcaddr, 3093 .get_srcport = xs_sock_srcport, 3094 .buf_alloc = rpc_malloc, 3095 .buf_free = rpc_free, 3096 .prepare_request = xs_stream_prepare_request, 3097 .send_request = xs_tcp_send_request, 3098 .abort_send_request = xs_stream_abort_send_request, 3099 .wait_for_reply_request = xprt_wait_for_reply_request_def, 3100 .close = xs_tcp_shutdown, 3101 .destroy = xs_destroy, 3102 .set_connect_timeout = xs_tcp_set_connect_timeout, 3103 .print_stats = xs_tcp_print_stats, 3104 .enable_swap = xs_enable_swap, 3105 .disable_swap = xs_disable_swap, 3106 .inject_disconnect = xs_inject_disconnect, 3107 #ifdef CONFIG_SUNRPC_BACKCHANNEL 3108 .bc_setup = xprt_setup_bc, 3109 .bc_maxpayload = xs_tcp_bc_maxpayload, 3110 .bc_num_slots = xprt_bc_max_slots, 3111 .bc_free_rqst = xprt_free_bc_rqst, 3112 .bc_destroy = xprt_destroy_bc, 3113 #endif 3114 }; 3115 3116 /* 3117 * The rpc_xprt_ops for the server backchannel 3118 */ 3119 3120 static const struct rpc_xprt_ops bc_tcp_ops = { 3121 .reserve_xprt = xprt_reserve_xprt, 3122 .release_xprt = xprt_release_xprt, 3123 .alloc_slot = xprt_alloc_slot, 3124 .free_slot = xprt_free_slot, 3125 .buf_alloc = bc_malloc, 3126 .buf_free = bc_free, 3127 .send_request = bc_send_request, 3128 .wait_for_reply_request = xprt_wait_for_reply_request_def, 3129 .close = bc_close, 3130 .destroy = bc_destroy, 3131 .print_stats = xs_tcp_print_stats, 3132 .enable_swap = xs_enable_swap, 3133 .disable_swap = xs_disable_swap, 3134 .inject_disconnect = xs_inject_disconnect, 3135 }; 3136 3137 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 3138 { 3139 static const struct sockaddr_in sin = { 3140 .sin_family = AF_INET, 3141 .sin_addr.s_addr = htonl(INADDR_ANY), 3142 }; 3143 static const struct sockaddr_in6 sin6 = { 3144 .sin6_family = AF_INET6, 3145 .sin6_addr = IN6ADDR_ANY_INIT, 3146 }; 3147 3148 switch (family) { 3149 case AF_LOCAL: 3150 break; 3151 case AF_INET: 3152 memcpy(sap, &sin, sizeof(sin)); 3153 break; 3154 case AF_INET6: 3155 memcpy(sap, &sin6, sizeof(sin6)); 3156 break; 3157 default: 3158 dprintk("RPC: %s: Bad address family\n", __func__); 3159 return -EAFNOSUPPORT; 3160 } 3161 return 0; 3162 } 3163 3164 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 3165 unsigned int slot_table_size, 3166 unsigned int max_slot_table_size) 3167 { 3168 struct rpc_xprt *xprt; 3169 struct sock_xprt *new; 3170 3171 if (args->addrlen > sizeof(xprt->addr)) { 3172 dprintk("RPC: xs_setup_xprt: address too large\n"); 3173 return ERR_PTR(-EBADF); 3174 } 3175 3176 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 3177 max_slot_table_size); 3178 if (xprt == NULL) { 3179 dprintk("RPC: xs_setup_xprt: couldn't allocate " 3180 "rpc_xprt\n"); 3181 return ERR_PTR(-ENOMEM); 3182 } 3183 3184 new = container_of(xprt, struct sock_xprt, xprt); 3185 mutex_init(&new->recv_mutex); 3186 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 3187 xprt->addrlen = args->addrlen; 3188 if (args->srcaddr) 3189 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 3190 else { 3191 int err; 3192 err = xs_init_anyaddr(args->dstaddr->sa_family, 3193 (struct sockaddr *)&new->srcaddr); 3194 if (err != 0) { 3195 xprt_free(xprt); 3196 return ERR_PTR(err); 3197 } 3198 } 3199 3200 return xprt; 3201 } 3202 3203 static const struct rpc_timeout xs_local_default_timeout = { 3204 .to_initval = 10 * HZ, 3205 .to_maxval = 10 * HZ, 3206 .to_retries = 2, 3207 }; 3208 3209 /** 3210 * xs_setup_local - Set up transport to use an AF_LOCAL socket 3211 * @args: rpc transport creation arguments 3212 * 3213 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 3214 */ 3215 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 3216 { 3217 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 3218 struct sock_xprt *transport; 3219 struct rpc_xprt *xprt; 3220 struct rpc_xprt *ret; 3221 3222 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3223 xprt_max_tcp_slot_table_entries); 3224 if (IS_ERR(xprt)) 3225 return xprt; 3226 transport = container_of(xprt, struct sock_xprt, xprt); 3227 3228 xprt->prot = 0; 3229 xprt->xprt_class = &xs_local_transport; 3230 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3231 3232 xprt->bind_timeout = XS_BIND_TO; 3233 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 3234 xprt->idle_timeout = XS_IDLE_DISC_TO; 3235 3236 xprt->ops = &xs_local_ops; 3237 xprt->timeout = &xs_local_default_timeout; 3238 3239 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 3240 INIT_WORK(&transport->error_worker, xs_error_handle); 3241 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket); 3242 3243 switch (sun->sun_family) { 3244 case AF_LOCAL: 3245 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') { 3246 dprintk("RPC: bad AF_LOCAL address: %s\n", 3247 sun->sun_path); 3248 ret = ERR_PTR(-EINVAL); 3249 goto out_err; 3250 } 3251 xprt_set_bound(xprt); 3252 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 3253 break; 3254 default: 3255 ret = ERR_PTR(-EAFNOSUPPORT); 3256 goto out_err; 3257 } 3258 3259 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 3260 xprt->address_strings[RPC_DISPLAY_ADDR]); 3261 3262 if (try_module_get(THIS_MODULE)) 3263 return xprt; 3264 ret = ERR_PTR(-EINVAL); 3265 out_err: 3266 xs_xprt_free(xprt); 3267 return ret; 3268 } 3269 3270 static const struct rpc_timeout xs_udp_default_timeout = { 3271 .to_initval = 5 * HZ, 3272 .to_maxval = 30 * HZ, 3273 .to_increment = 5 * HZ, 3274 .to_retries = 5, 3275 }; 3276 3277 /** 3278 * xs_setup_udp - Set up transport to use a UDP socket 3279 * @args: rpc transport creation arguments 3280 * 3281 */ 3282 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 3283 { 3284 struct sockaddr *addr = args->dstaddr; 3285 struct rpc_xprt *xprt; 3286 struct sock_xprt *transport; 3287 struct rpc_xprt *ret; 3288 3289 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 3290 xprt_udp_slot_table_entries); 3291 if (IS_ERR(xprt)) 3292 return xprt; 3293 transport = container_of(xprt, struct sock_xprt, xprt); 3294 3295 xprt->prot = IPPROTO_UDP; 3296 xprt->xprt_class = &xs_udp_transport; 3297 /* XXX: header size can vary due to auth type, IPv6, etc. */ 3298 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 3299 3300 xprt->bind_timeout = XS_BIND_TO; 3301 xprt->reestablish_timeout = XS_UDP_REEST_TO; 3302 xprt->idle_timeout = XS_IDLE_DISC_TO; 3303 3304 xprt->ops = &xs_udp_ops; 3305 3306 xprt->timeout = &xs_udp_default_timeout; 3307 3308 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 3309 INIT_WORK(&transport->error_worker, xs_error_handle); 3310 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 3311 3312 switch (addr->sa_family) { 3313 case AF_INET: 3314 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3315 xprt_set_bound(xprt); 3316 3317 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 3318 break; 3319 case AF_INET6: 3320 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3321 xprt_set_bound(xprt); 3322 3323 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 3324 break; 3325 default: 3326 ret = ERR_PTR(-EAFNOSUPPORT); 3327 goto out_err; 3328 } 3329 3330 if (xprt_bound(xprt)) 3331 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3332 xprt->address_strings[RPC_DISPLAY_ADDR], 3333 xprt->address_strings[RPC_DISPLAY_PORT], 3334 xprt->address_strings[RPC_DISPLAY_PROTO]); 3335 else 3336 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3337 xprt->address_strings[RPC_DISPLAY_ADDR], 3338 xprt->address_strings[RPC_DISPLAY_PROTO]); 3339 3340 if (try_module_get(THIS_MODULE)) 3341 return xprt; 3342 ret = ERR_PTR(-EINVAL); 3343 out_err: 3344 xs_xprt_free(xprt); 3345 return ret; 3346 } 3347 3348 static const struct rpc_timeout xs_tcp_default_timeout = { 3349 .to_initval = 60 * HZ, 3350 .to_maxval = 60 * HZ, 3351 .to_retries = 2, 3352 }; 3353 3354 /** 3355 * xs_setup_tcp - Set up transport to use a TCP socket 3356 * @args: rpc transport creation arguments 3357 * 3358 */ 3359 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 3360 { 3361 struct sockaddr *addr = args->dstaddr; 3362 struct rpc_xprt *xprt; 3363 struct sock_xprt *transport; 3364 struct rpc_xprt *ret; 3365 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 3366 3367 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 3368 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 3369 3370 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3371 max_slot_table_size); 3372 if (IS_ERR(xprt)) 3373 return xprt; 3374 transport = container_of(xprt, struct sock_xprt, xprt); 3375 3376 xprt->prot = IPPROTO_TCP; 3377 xprt->xprt_class = &xs_tcp_transport; 3378 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3379 3380 xprt->bind_timeout = XS_BIND_TO; 3381 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 3382 xprt->idle_timeout = XS_IDLE_DISC_TO; 3383 3384 xprt->ops = &xs_tcp_ops; 3385 xprt->timeout = &xs_tcp_default_timeout; 3386 3387 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 3388 if (args->reconnect_timeout) 3389 xprt->max_reconnect_timeout = args->reconnect_timeout; 3390 3391 xprt->connect_timeout = xprt->timeout->to_initval * 3392 (xprt->timeout->to_retries + 1); 3393 if (args->connect_timeout) 3394 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout); 3395 3396 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 3397 INIT_WORK(&transport->error_worker, xs_error_handle); 3398 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 3399 3400 switch (addr->sa_family) { 3401 case AF_INET: 3402 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3403 xprt_set_bound(xprt); 3404 3405 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 3406 break; 3407 case AF_INET6: 3408 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3409 xprt_set_bound(xprt); 3410 3411 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 3412 break; 3413 default: 3414 ret = ERR_PTR(-EAFNOSUPPORT); 3415 goto out_err; 3416 } 3417 3418 if (xprt_bound(xprt)) 3419 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3420 xprt->address_strings[RPC_DISPLAY_ADDR], 3421 xprt->address_strings[RPC_DISPLAY_PORT], 3422 xprt->address_strings[RPC_DISPLAY_PROTO]); 3423 else 3424 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3425 xprt->address_strings[RPC_DISPLAY_ADDR], 3426 xprt->address_strings[RPC_DISPLAY_PROTO]); 3427 3428 if (try_module_get(THIS_MODULE)) 3429 return xprt; 3430 ret = ERR_PTR(-EINVAL); 3431 out_err: 3432 xs_xprt_free(xprt); 3433 return ret; 3434 } 3435 3436 /** 3437 * xs_setup_tcp_tls - Set up transport to use a TCP with TLS 3438 * @args: rpc transport creation arguments 3439 * 3440 */ 3441 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args) 3442 { 3443 struct sockaddr *addr = args->dstaddr; 3444 struct rpc_xprt *xprt; 3445 struct sock_xprt *transport; 3446 struct rpc_xprt *ret; 3447 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 3448 3449 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 3450 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 3451 3452 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3453 max_slot_table_size); 3454 if (IS_ERR(xprt)) 3455 return xprt; 3456 transport = container_of(xprt, struct sock_xprt, xprt); 3457 3458 xprt->prot = IPPROTO_TCP; 3459 xprt->xprt_class = &xs_tcp_transport; 3460 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3461 3462 xprt->bind_timeout = XS_BIND_TO; 3463 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 3464 xprt->idle_timeout = XS_IDLE_DISC_TO; 3465 3466 xprt->ops = &xs_tcp_ops; 3467 xprt->timeout = &xs_tcp_default_timeout; 3468 3469 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 3470 xprt->connect_timeout = xprt->timeout->to_initval * 3471 (xprt->timeout->to_retries + 1); 3472 3473 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 3474 INIT_WORK(&transport->error_worker, xs_error_handle); 3475 3476 switch (args->xprtsec.policy) { 3477 case RPC_XPRTSEC_TLS_ANON: 3478 case RPC_XPRTSEC_TLS_X509: 3479 xprt->xprtsec = args->xprtsec; 3480 INIT_DELAYED_WORK(&transport->connect_worker, 3481 xs_tcp_tls_setup_socket); 3482 break; 3483 default: 3484 ret = ERR_PTR(-EACCES); 3485 goto out_err; 3486 } 3487 3488 switch (addr->sa_family) { 3489 case AF_INET: 3490 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3491 xprt_set_bound(xprt); 3492 3493 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 3494 break; 3495 case AF_INET6: 3496 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3497 xprt_set_bound(xprt); 3498 3499 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 3500 break; 3501 default: 3502 ret = ERR_PTR(-EAFNOSUPPORT); 3503 goto out_err; 3504 } 3505 3506 if (xprt_bound(xprt)) 3507 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3508 xprt->address_strings[RPC_DISPLAY_ADDR], 3509 xprt->address_strings[RPC_DISPLAY_PORT], 3510 xprt->address_strings[RPC_DISPLAY_PROTO]); 3511 else 3512 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3513 xprt->address_strings[RPC_DISPLAY_ADDR], 3514 xprt->address_strings[RPC_DISPLAY_PROTO]); 3515 3516 if (try_module_get(THIS_MODULE)) 3517 return xprt; 3518 ret = ERR_PTR(-EINVAL); 3519 out_err: 3520 xs_xprt_free(xprt); 3521 return ret; 3522 } 3523 3524 /** 3525 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 3526 * @args: rpc transport creation arguments 3527 * 3528 */ 3529 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 3530 { 3531 struct sockaddr *addr = args->dstaddr; 3532 struct rpc_xprt *xprt; 3533 struct sock_xprt *transport; 3534 struct svc_sock *bc_sock; 3535 struct rpc_xprt *ret; 3536 3537 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3538 xprt_tcp_slot_table_entries); 3539 if (IS_ERR(xprt)) 3540 return xprt; 3541 transport = container_of(xprt, struct sock_xprt, xprt); 3542 3543 xprt->prot = IPPROTO_TCP; 3544 xprt->xprt_class = &xs_bc_tcp_transport; 3545 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3546 xprt->timeout = &xs_tcp_default_timeout; 3547 3548 /* backchannel */ 3549 xprt_set_bound(xprt); 3550 xprt->bind_timeout = 0; 3551 xprt->reestablish_timeout = 0; 3552 xprt->idle_timeout = 0; 3553 3554 xprt->ops = &bc_tcp_ops; 3555 3556 switch (addr->sa_family) { 3557 case AF_INET: 3558 xs_format_peer_addresses(xprt, "tcp", 3559 RPCBIND_NETID_TCP); 3560 break; 3561 case AF_INET6: 3562 xs_format_peer_addresses(xprt, "tcp", 3563 RPCBIND_NETID_TCP6); 3564 break; 3565 default: 3566 ret = ERR_PTR(-EAFNOSUPPORT); 3567 goto out_err; 3568 } 3569 3570 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3571 xprt->address_strings[RPC_DISPLAY_ADDR], 3572 xprt->address_strings[RPC_DISPLAY_PORT], 3573 xprt->address_strings[RPC_DISPLAY_PROTO]); 3574 3575 /* 3576 * Once we've associated a backchannel xprt with a connection, 3577 * we want to keep it around as long as the connection lasts, 3578 * in case we need to start using it for a backchannel again; 3579 * this reference won't be dropped until bc_xprt is destroyed. 3580 */ 3581 xprt_get(xprt); 3582 args->bc_xprt->xpt_bc_xprt = xprt; 3583 xprt->bc_xprt = args->bc_xprt; 3584 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3585 transport->sock = bc_sock->sk_sock; 3586 transport->inet = bc_sock->sk_sk; 3587 3588 /* 3589 * Since we don't want connections for the backchannel, we set 3590 * the xprt status to connected 3591 */ 3592 xprt_set_connected(xprt); 3593 3594 if (try_module_get(THIS_MODULE)) 3595 return xprt; 3596 3597 args->bc_xprt->xpt_bc_xprt = NULL; 3598 args->bc_xprt->xpt_bc_xps = NULL; 3599 xprt_put(xprt); 3600 ret = ERR_PTR(-EINVAL); 3601 out_err: 3602 xs_xprt_free(xprt); 3603 return ret; 3604 } 3605 3606 static struct xprt_class xs_local_transport = { 3607 .list = LIST_HEAD_INIT(xs_local_transport.list), 3608 .name = "named UNIX socket", 3609 .owner = THIS_MODULE, 3610 .ident = XPRT_TRANSPORT_LOCAL, 3611 .setup = xs_setup_local, 3612 .netid = { "" }, 3613 }; 3614 3615 static struct xprt_class xs_udp_transport = { 3616 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3617 .name = "udp", 3618 .owner = THIS_MODULE, 3619 .ident = XPRT_TRANSPORT_UDP, 3620 .setup = xs_setup_udp, 3621 .netid = { "udp", "udp6", "" }, 3622 }; 3623 3624 static struct xprt_class xs_tcp_transport = { 3625 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3626 .name = "tcp", 3627 .owner = THIS_MODULE, 3628 .ident = XPRT_TRANSPORT_TCP, 3629 .setup = xs_setup_tcp, 3630 .netid = { "tcp", "tcp6", "" }, 3631 }; 3632 3633 static struct xprt_class xs_tcp_tls_transport = { 3634 .list = LIST_HEAD_INIT(xs_tcp_tls_transport.list), 3635 .name = "tcp-with-tls", 3636 .owner = THIS_MODULE, 3637 .ident = XPRT_TRANSPORT_TCP_TLS, 3638 .setup = xs_setup_tcp_tls, 3639 .netid = { "tcp", "tcp6", "" }, 3640 }; 3641 3642 static struct xprt_class xs_bc_tcp_transport = { 3643 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3644 .name = "tcp NFSv4.1 backchannel", 3645 .owner = THIS_MODULE, 3646 .ident = XPRT_TRANSPORT_BC_TCP, 3647 .setup = xs_setup_bc_tcp, 3648 .netid = { "" }, 3649 }; 3650 3651 /** 3652 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3653 * 3654 */ 3655 int init_socket_xprt(void) 3656 { 3657 if (!sunrpc_table_header) 3658 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table); 3659 3660 xprt_register_transport(&xs_local_transport); 3661 xprt_register_transport(&xs_udp_transport); 3662 xprt_register_transport(&xs_tcp_transport); 3663 xprt_register_transport(&xs_tcp_tls_transport); 3664 xprt_register_transport(&xs_bc_tcp_transport); 3665 3666 return 0; 3667 } 3668 3669 /** 3670 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3671 * 3672 */ 3673 void cleanup_socket_xprt(void) 3674 { 3675 if (sunrpc_table_header) { 3676 unregister_sysctl_table(sunrpc_table_header); 3677 sunrpc_table_header = NULL; 3678 } 3679 3680 xprt_unregister_transport(&xs_local_transport); 3681 xprt_unregister_transport(&xs_udp_transport); 3682 xprt_unregister_transport(&xs_tcp_transport); 3683 xprt_unregister_transport(&xs_tcp_tls_transport); 3684 xprt_unregister_transport(&xs_bc_tcp_transport); 3685 } 3686 3687 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3688 { 3689 return param_set_uint_minmax(val, kp, 3690 RPC_MIN_RESVPORT, 3691 RPC_MAX_RESVPORT); 3692 } 3693 3694 static const struct kernel_param_ops param_ops_portnr = { 3695 .set = param_set_portnr, 3696 .get = param_get_uint, 3697 }; 3698 3699 #define param_check_portnr(name, p) \ 3700 __param_check(name, p, unsigned int); 3701 3702 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3703 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3704 3705 static int param_set_slot_table_size(const char *val, 3706 const struct kernel_param *kp) 3707 { 3708 return param_set_uint_minmax(val, kp, 3709 RPC_MIN_SLOT_TABLE, 3710 RPC_MAX_SLOT_TABLE); 3711 } 3712 3713 static const struct kernel_param_ops param_ops_slot_table_size = { 3714 .set = param_set_slot_table_size, 3715 .get = param_get_uint, 3716 }; 3717 3718 #define param_check_slot_table_size(name, p) \ 3719 __param_check(name, p, unsigned int); 3720 3721 static int param_set_max_slot_table_size(const char *val, 3722 const struct kernel_param *kp) 3723 { 3724 return param_set_uint_minmax(val, kp, 3725 RPC_MIN_SLOT_TABLE, 3726 RPC_MAX_SLOT_TABLE_LIMIT); 3727 } 3728 3729 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3730 .set = param_set_max_slot_table_size, 3731 .get = param_get_uint, 3732 }; 3733 3734 #define param_check_max_slot_table_size(name, p) \ 3735 __param_check(name, p, unsigned int); 3736 3737 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3738 slot_table_size, 0644); 3739 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3740 max_slot_table_size, 0644); 3741 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3742 slot_table_size, 0644); 3743