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 <linux/bvec.h> 51 #include <linux/highmem.h> 52 #include <linux/uio.h> 53 #include <linux/sched/mm.h> 54 55 #include <trace/events/sunrpc.h> 56 57 #include "sunrpc.h" 58 59 static void xs_close(struct rpc_xprt *xprt); 60 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 61 struct socket *sock); 62 63 /* 64 * xprtsock tunables 65 */ 66 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 67 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 68 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 69 70 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 71 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 72 73 #define XS_TCP_LINGER_TO (15U * HZ) 74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 75 76 /* 77 * We can register our own files under /proc/sys/sunrpc by 78 * calling register_sysctl_table() again. The files in that 79 * directory become the union of all files registered there. 80 * 81 * We simply need to make sure that we don't collide with 82 * someone else's file names! 83 */ 84 85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 90 91 static struct ctl_table_header *sunrpc_table_header; 92 93 /* 94 * FIXME: changing the UDP slot table size should also resize the UDP 95 * socket buffers for existing UDP transports 96 */ 97 static struct ctl_table xs_tunables_table[] = { 98 { 99 .procname = "udp_slot_table_entries", 100 .data = &xprt_udp_slot_table_entries, 101 .maxlen = sizeof(unsigned int), 102 .mode = 0644, 103 .proc_handler = proc_dointvec_minmax, 104 .extra1 = &min_slot_table_size, 105 .extra2 = &max_slot_table_size 106 }, 107 { 108 .procname = "tcp_slot_table_entries", 109 .data = &xprt_tcp_slot_table_entries, 110 .maxlen = sizeof(unsigned int), 111 .mode = 0644, 112 .proc_handler = proc_dointvec_minmax, 113 .extra1 = &min_slot_table_size, 114 .extra2 = &max_slot_table_size 115 }, 116 { 117 .procname = "tcp_max_slot_table_entries", 118 .data = &xprt_max_tcp_slot_table_entries, 119 .maxlen = sizeof(unsigned int), 120 .mode = 0644, 121 .proc_handler = proc_dointvec_minmax, 122 .extra1 = &min_slot_table_size, 123 .extra2 = &max_tcp_slot_table_limit 124 }, 125 { 126 .procname = "min_resvport", 127 .data = &xprt_min_resvport, 128 .maxlen = sizeof(unsigned int), 129 .mode = 0644, 130 .proc_handler = proc_dointvec_minmax, 131 .extra1 = &xprt_min_resvport_limit, 132 .extra2 = &xprt_max_resvport_limit 133 }, 134 { 135 .procname = "max_resvport", 136 .data = &xprt_max_resvport, 137 .maxlen = sizeof(unsigned int), 138 .mode = 0644, 139 .proc_handler = proc_dointvec_minmax, 140 .extra1 = &xprt_min_resvport_limit, 141 .extra2 = &xprt_max_resvport_limit 142 }, 143 { 144 .procname = "tcp_fin_timeout", 145 .data = &xs_tcp_fin_timeout, 146 .maxlen = sizeof(xs_tcp_fin_timeout), 147 .mode = 0644, 148 .proc_handler = proc_dointvec_jiffies, 149 }, 150 { }, 151 }; 152 153 static struct ctl_table sunrpc_table[] = { 154 { 155 .procname = "sunrpc", 156 .mode = 0555, 157 .child = xs_tunables_table 158 }, 159 { }, 160 }; 161 162 /* 163 * Wait duration for a reply from the RPC portmapper. 164 */ 165 #define XS_BIND_TO (60U * HZ) 166 167 /* 168 * Delay if a UDP socket connect error occurs. This is most likely some 169 * kind of resource problem on the local host. 170 */ 171 #define XS_UDP_REEST_TO (2U * HZ) 172 173 /* 174 * The reestablish timeout allows clients to delay for a bit before attempting 175 * to reconnect to a server that just dropped our connection. 176 * 177 * We implement an exponential backoff when trying to reestablish a TCP 178 * transport connection with the server. Some servers like to drop a TCP 179 * connection when they are overworked, so we start with a short timeout and 180 * increase over time if the server is down or not responding. 181 */ 182 #define XS_TCP_INIT_REEST_TO (3U * HZ) 183 184 /* 185 * TCP idle timeout; client drops the transport socket if it is idle 186 * for this long. Note that we also timeout UDP sockets to prevent 187 * holding port numbers when there is no RPC traffic. 188 */ 189 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 190 191 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 192 # undef RPC_DEBUG_DATA 193 # define RPCDBG_FACILITY RPCDBG_TRANS 194 #endif 195 196 #ifdef RPC_DEBUG_DATA 197 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 198 { 199 u8 *buf = (u8 *) packet; 200 int j; 201 202 dprintk("RPC: %s\n", msg); 203 for (j = 0; j < count && j < 128; j += 4) { 204 if (!(j & 31)) { 205 if (j) 206 dprintk("\n"); 207 dprintk("0x%04x ", j); 208 } 209 dprintk("%02x%02x%02x%02x ", 210 buf[j], buf[j+1], buf[j+2], buf[j+3]); 211 } 212 dprintk("\n"); 213 } 214 #else 215 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 216 { 217 /* NOP */ 218 } 219 #endif 220 221 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 222 { 223 return (struct rpc_xprt *) sk->sk_user_data; 224 } 225 226 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 227 { 228 return (struct sockaddr *) &xprt->addr; 229 } 230 231 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 232 { 233 return (struct sockaddr_un *) &xprt->addr; 234 } 235 236 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 237 { 238 return (struct sockaddr_in *) &xprt->addr; 239 } 240 241 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 242 { 243 return (struct sockaddr_in6 *) &xprt->addr; 244 } 245 246 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 247 { 248 struct sockaddr *sap = xs_addr(xprt); 249 struct sockaddr_in6 *sin6; 250 struct sockaddr_in *sin; 251 struct sockaddr_un *sun; 252 char buf[128]; 253 254 switch (sap->sa_family) { 255 case AF_LOCAL: 256 sun = xs_addr_un(xprt); 257 strlcpy(buf, sun->sun_path, sizeof(buf)); 258 xprt->address_strings[RPC_DISPLAY_ADDR] = 259 kstrdup(buf, GFP_KERNEL); 260 break; 261 case AF_INET: 262 (void)rpc_ntop(sap, buf, sizeof(buf)); 263 xprt->address_strings[RPC_DISPLAY_ADDR] = 264 kstrdup(buf, GFP_KERNEL); 265 sin = xs_addr_in(xprt); 266 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 267 break; 268 case AF_INET6: 269 (void)rpc_ntop(sap, buf, sizeof(buf)); 270 xprt->address_strings[RPC_DISPLAY_ADDR] = 271 kstrdup(buf, GFP_KERNEL); 272 sin6 = xs_addr_in6(xprt); 273 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 274 break; 275 default: 276 BUG(); 277 } 278 279 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 280 } 281 282 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 283 { 284 struct sockaddr *sap = xs_addr(xprt); 285 char buf[128]; 286 287 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 288 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 289 290 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 291 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 292 } 293 294 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 295 const char *protocol, 296 const char *netid) 297 { 298 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 299 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 300 xs_format_common_peer_addresses(xprt); 301 xs_format_common_peer_ports(xprt); 302 } 303 304 static void xs_update_peer_port(struct rpc_xprt *xprt) 305 { 306 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 307 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 308 309 xs_format_common_peer_ports(xprt); 310 } 311 312 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 313 { 314 unsigned int i; 315 316 for (i = 0; i < RPC_DISPLAY_MAX; i++) 317 switch (i) { 318 case RPC_DISPLAY_PROTO: 319 case RPC_DISPLAY_NETID: 320 continue; 321 default: 322 kfree(xprt->address_strings[i]); 323 } 324 } 325 326 static size_t 327 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp) 328 { 329 size_t i,n; 330 331 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES)) 332 return want; 333 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT; 334 for (i = 0; i < n; i++) { 335 if (buf->pages[i]) 336 continue; 337 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp); 338 if (!buf->pages[i]) { 339 i *= PAGE_SIZE; 340 return i > buf->page_base ? i - buf->page_base : 0; 341 } 342 } 343 return want; 344 } 345 346 static ssize_t 347 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek) 348 { 349 ssize_t ret; 350 if (seek != 0) 351 iov_iter_advance(&msg->msg_iter, seek); 352 ret = sock_recvmsg(sock, msg, flags); 353 return ret > 0 ? ret + seek : ret; 354 } 355 356 static ssize_t 357 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags, 358 struct kvec *kvec, size_t count, size_t seek) 359 { 360 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count); 361 return xs_sock_recvmsg(sock, msg, flags, seek); 362 } 363 364 static ssize_t 365 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags, 366 struct bio_vec *bvec, unsigned long nr, size_t count, 367 size_t seek) 368 { 369 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count); 370 return xs_sock_recvmsg(sock, msg, flags, seek); 371 } 372 373 static ssize_t 374 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags, 375 size_t count) 376 { 377 iov_iter_discard(&msg->msg_iter, READ, count); 378 return sock_recvmsg(sock, msg, flags); 379 } 380 381 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 382 static void 383 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 384 { 385 struct bvec_iter bi = { 386 .bi_size = count, 387 }; 388 struct bio_vec bv; 389 390 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); 391 for_each_bvec(bv, bvec, bi, bi) 392 flush_dcache_page(bv.bv_page); 393 } 394 #else 395 static inline void 396 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 397 { 398 } 399 #endif 400 401 static ssize_t 402 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags, 403 struct xdr_buf *buf, size_t count, size_t seek, size_t *read) 404 { 405 size_t want, seek_init = seek, offset = 0; 406 ssize_t ret; 407 408 want = min_t(size_t, count, buf->head[0].iov_len); 409 if (seek < want) { 410 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek); 411 if (ret <= 0) 412 goto sock_err; 413 offset += ret; 414 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 415 goto out; 416 if (ret != want) 417 goto out; 418 seek = 0; 419 } else { 420 seek -= want; 421 offset += want; 422 } 423 424 want = xs_alloc_sparse_pages(buf, 425 min_t(size_t, count - offset, buf->page_len), 426 GFP_NOWAIT); 427 if (seek < want) { 428 ret = xs_read_bvec(sock, msg, flags, buf->bvec, 429 xdr_buf_pagecount(buf), 430 want + buf->page_base, 431 seek + buf->page_base); 432 if (ret <= 0) 433 goto sock_err; 434 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base); 435 offset += ret - buf->page_base; 436 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 437 goto out; 438 if (ret != want) 439 goto out; 440 seek = 0; 441 } else { 442 seek -= want; 443 offset += want; 444 } 445 446 want = min_t(size_t, count - offset, buf->tail[0].iov_len); 447 if (seek < want) { 448 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek); 449 if (ret <= 0) 450 goto sock_err; 451 offset += ret; 452 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 453 goto out; 454 if (ret != want) 455 goto out; 456 } else 457 offset = seek_init; 458 ret = -EMSGSIZE; 459 out: 460 *read = offset - seek_init; 461 return ret; 462 sock_err: 463 offset += seek; 464 goto out; 465 } 466 467 static void 468 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf) 469 { 470 if (!transport->recv.copied) { 471 if (buf->head[0].iov_len >= transport->recv.offset) 472 memcpy(buf->head[0].iov_base, 473 &transport->recv.xid, 474 transport->recv.offset); 475 transport->recv.copied = transport->recv.offset; 476 } 477 } 478 479 static bool 480 xs_read_stream_request_done(struct sock_xprt *transport) 481 { 482 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT); 483 } 484 485 static void 486 xs_read_stream_check_eor(struct sock_xprt *transport, 487 struct msghdr *msg) 488 { 489 if (xs_read_stream_request_done(transport)) 490 msg->msg_flags |= MSG_EOR; 491 } 492 493 static ssize_t 494 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg, 495 int flags, struct rpc_rqst *req) 496 { 497 struct xdr_buf *buf = &req->rq_private_buf; 498 size_t want, read; 499 ssize_t ret; 500 501 xs_read_header(transport, buf); 502 503 want = transport->recv.len - transport->recv.offset; 504 if (want != 0) { 505 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf, 506 transport->recv.copied + want, 507 transport->recv.copied, 508 &read); 509 transport->recv.offset += read; 510 transport->recv.copied += read; 511 } 512 513 if (transport->recv.offset == transport->recv.len) 514 xs_read_stream_check_eor(transport, msg); 515 516 if (want == 0) 517 return 0; 518 519 switch (ret) { 520 default: 521 break; 522 case -EFAULT: 523 case -EMSGSIZE: 524 msg->msg_flags |= MSG_TRUNC; 525 return read; 526 case 0: 527 return -ESHUTDOWN; 528 } 529 return ret < 0 ? ret : read; 530 } 531 532 static size_t 533 xs_read_stream_headersize(bool isfrag) 534 { 535 if (isfrag) 536 return sizeof(__be32); 537 return 3 * sizeof(__be32); 538 } 539 540 static ssize_t 541 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg, 542 int flags, size_t want, size_t seek) 543 { 544 struct kvec kvec = { 545 .iov_base = &transport->recv.fraghdr, 546 .iov_len = want, 547 }; 548 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek); 549 } 550 551 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 552 static ssize_t 553 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 554 { 555 struct rpc_xprt *xprt = &transport->xprt; 556 struct rpc_rqst *req; 557 ssize_t ret; 558 559 /* Look up and lock the request corresponding to the given XID */ 560 req = xprt_lookup_bc_request(xprt, transport->recv.xid); 561 if (!req) { 562 printk(KERN_WARNING "Callback slot table overflowed\n"); 563 return -ESHUTDOWN; 564 } 565 566 ret = xs_read_stream_request(transport, msg, flags, req); 567 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 568 xprt_complete_bc_request(req, transport->recv.copied); 569 570 return ret; 571 } 572 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 573 static ssize_t 574 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 575 { 576 return -ESHUTDOWN; 577 } 578 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 579 580 static ssize_t 581 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags) 582 { 583 struct rpc_xprt *xprt = &transport->xprt; 584 struct rpc_rqst *req; 585 ssize_t ret = 0; 586 587 /* Look up and lock the request corresponding to the given XID */ 588 spin_lock(&xprt->queue_lock); 589 req = xprt_lookup_rqst(xprt, transport->recv.xid); 590 if (!req) { 591 msg->msg_flags |= MSG_TRUNC; 592 goto out; 593 } 594 xprt_pin_rqst(req); 595 spin_unlock(&xprt->queue_lock); 596 597 ret = xs_read_stream_request(transport, msg, flags, req); 598 599 spin_lock(&xprt->queue_lock); 600 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 601 xprt_complete_rqst(req->rq_task, transport->recv.copied); 602 xprt_unpin_rqst(req); 603 out: 604 spin_unlock(&xprt->queue_lock); 605 return ret; 606 } 607 608 static ssize_t 609 xs_read_stream(struct sock_xprt *transport, int flags) 610 { 611 struct msghdr msg = { 0 }; 612 size_t want, read = 0; 613 ssize_t ret = 0; 614 615 if (transport->recv.len == 0) { 616 want = xs_read_stream_headersize(transport->recv.copied != 0); 617 ret = xs_read_stream_header(transport, &msg, flags, want, 618 transport->recv.offset); 619 if (ret <= 0) 620 goto out_err; 621 transport->recv.offset = ret; 622 if (transport->recv.offset != want) 623 return transport->recv.offset; 624 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) & 625 RPC_FRAGMENT_SIZE_MASK; 626 transport->recv.offset -= sizeof(transport->recv.fraghdr); 627 read = ret; 628 } 629 630 switch (be32_to_cpu(transport->recv.calldir)) { 631 default: 632 msg.msg_flags |= MSG_TRUNC; 633 break; 634 case RPC_CALL: 635 ret = xs_read_stream_call(transport, &msg, flags); 636 break; 637 case RPC_REPLY: 638 ret = xs_read_stream_reply(transport, &msg, flags); 639 } 640 if (msg.msg_flags & MSG_TRUNC) { 641 transport->recv.calldir = cpu_to_be32(-1); 642 transport->recv.copied = -1; 643 } 644 if (ret < 0) 645 goto out_err; 646 read += ret; 647 if (transport->recv.offset < transport->recv.len) { 648 if (!(msg.msg_flags & MSG_TRUNC)) 649 return read; 650 msg.msg_flags = 0; 651 ret = xs_read_discard(transport->sock, &msg, flags, 652 transport->recv.len - transport->recv.offset); 653 if (ret <= 0) 654 goto out_err; 655 transport->recv.offset += ret; 656 read += ret; 657 if (transport->recv.offset != transport->recv.len) 658 return read; 659 } 660 if (xs_read_stream_request_done(transport)) { 661 trace_xs_stream_read_request(transport); 662 transport->recv.copied = 0; 663 } 664 transport->recv.offset = 0; 665 transport->recv.len = 0; 666 return read; 667 out_err: 668 return ret != 0 ? ret : -ESHUTDOWN; 669 } 670 671 static __poll_t xs_poll_socket(struct sock_xprt *transport) 672 { 673 return transport->sock->ops->poll(NULL, transport->sock, NULL); 674 } 675 676 static bool xs_poll_socket_readable(struct sock_xprt *transport) 677 { 678 __poll_t events = xs_poll_socket(transport); 679 680 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP); 681 } 682 683 static void xs_poll_check_readable(struct sock_xprt *transport) 684 { 685 686 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 687 if (!xs_poll_socket_readable(transport)) 688 return; 689 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 690 queue_work(xprtiod_workqueue, &transport->recv_worker); 691 } 692 693 static void xs_stream_data_receive(struct sock_xprt *transport) 694 { 695 size_t read = 0; 696 ssize_t ret = 0; 697 698 mutex_lock(&transport->recv_mutex); 699 if (transport->sock == NULL) 700 goto out; 701 for (;;) { 702 ret = xs_read_stream(transport, MSG_DONTWAIT); 703 if (ret < 0) 704 break; 705 read += ret; 706 cond_resched(); 707 } 708 if (ret == -ESHUTDOWN) 709 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 710 else 711 xs_poll_check_readable(transport); 712 out: 713 mutex_unlock(&transport->recv_mutex); 714 trace_xs_stream_read_data(&transport->xprt, ret, read); 715 } 716 717 static void xs_stream_data_receive_workfn(struct work_struct *work) 718 { 719 struct sock_xprt *transport = 720 container_of(work, struct sock_xprt, recv_worker); 721 unsigned int pflags = memalloc_nofs_save(); 722 723 xs_stream_data_receive(transport); 724 memalloc_nofs_restore(pflags); 725 } 726 727 static void 728 xs_stream_reset_connect(struct sock_xprt *transport) 729 { 730 transport->recv.offset = 0; 731 transport->recv.len = 0; 732 transport->recv.copied = 0; 733 transport->xmit.offset = 0; 734 } 735 736 static void 737 xs_stream_start_connect(struct sock_xprt *transport) 738 { 739 transport->xprt.stat.connect_count++; 740 transport->xprt.stat.connect_start = jiffies; 741 } 742 743 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 744 745 static int xs_sendmsg(struct socket *sock, struct msghdr *msg, size_t seek) 746 { 747 if (seek) 748 iov_iter_advance(&msg->msg_iter, seek); 749 return sock_sendmsg(sock, msg); 750 } 751 752 static int xs_send_kvec(struct socket *sock, struct msghdr *msg, struct kvec *vec, size_t seek) 753 { 754 iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len); 755 return xs_sendmsg(sock, msg, seek); 756 } 757 758 static int xs_send_pagedata(struct socket *sock, struct msghdr *msg, struct xdr_buf *xdr, size_t base) 759 { 760 int err; 761 762 err = xdr_alloc_bvec(xdr, GFP_KERNEL); 763 if (err < 0) 764 return err; 765 766 iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec, 767 xdr_buf_pagecount(xdr), 768 xdr->page_len + xdr->page_base); 769 return xs_sendmsg(sock, msg, base + xdr->page_base); 770 } 771 772 #define xs_record_marker_len() sizeof(rpc_fraghdr) 773 774 /* Common case: 775 * - stream transport 776 * - sending from byte 0 of the message 777 * - the message is wholly contained in @xdr's head iovec 778 */ 779 static int xs_send_rm_and_kvec(struct socket *sock, struct msghdr *msg, 780 rpc_fraghdr marker, struct kvec *vec, size_t base) 781 { 782 struct kvec iov[2] = { 783 [0] = { 784 .iov_base = &marker, 785 .iov_len = sizeof(marker) 786 }, 787 [1] = *vec, 788 }; 789 size_t len = iov[0].iov_len + iov[1].iov_len; 790 791 iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len); 792 return xs_sendmsg(sock, msg, base); 793 } 794 795 /** 796 * xs_sendpages - write pages directly to a socket 797 * @sock: socket to send on 798 * @addr: UDP only -- address of destination 799 * @addrlen: UDP only -- length of destination address 800 * @xdr: buffer containing this request 801 * @base: starting position in the buffer 802 * @rm: stream record marker field 803 * @sent_p: return the total number of bytes successfully queued for sending 804 * 805 */ 806 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, rpc_fraghdr rm, int *sent_p) 807 { 808 struct msghdr msg = { 809 .msg_name = addr, 810 .msg_namelen = addrlen, 811 .msg_flags = XS_SENDMSG_FLAGS | MSG_MORE, 812 }; 813 unsigned int rmsize = rm ? sizeof(rm) : 0; 814 unsigned int remainder = rmsize + xdr->len - base; 815 unsigned int want; 816 int err = 0; 817 818 if (unlikely(!sock)) 819 return -ENOTSOCK; 820 821 want = xdr->head[0].iov_len + rmsize; 822 if (base < want) { 823 unsigned int len = want - base; 824 remainder -= len; 825 if (remainder == 0) 826 msg.msg_flags &= ~MSG_MORE; 827 if (rmsize) 828 err = xs_send_rm_and_kvec(sock, &msg, rm, 829 &xdr->head[0], base); 830 else 831 err = xs_send_kvec(sock, &msg, &xdr->head[0], base); 832 if (remainder == 0 || err != len) 833 goto out; 834 *sent_p += err; 835 base = 0; 836 } else 837 base -= want; 838 839 if (base < xdr->page_len) { 840 unsigned int len = xdr->page_len - base; 841 remainder -= len; 842 if (remainder == 0) 843 msg.msg_flags &= ~MSG_MORE; 844 err = xs_send_pagedata(sock, &msg, xdr, base); 845 if (remainder == 0 || err != len) 846 goto out; 847 *sent_p += err; 848 base = 0; 849 } else 850 base -= xdr->page_len; 851 852 if (base >= xdr->tail[0].iov_len) 853 return 0; 854 msg.msg_flags &= ~MSG_MORE; 855 err = xs_send_kvec(sock, &msg, &xdr->tail[0], base); 856 out: 857 if (err > 0) { 858 *sent_p += err; 859 err = 0; 860 } 861 return err; 862 } 863 864 /** 865 * xs_nospace - handle transmit was incomplete 866 * @req: pointer to RPC request 867 * 868 */ 869 static int xs_nospace(struct rpc_rqst *req) 870 { 871 struct rpc_xprt *xprt = req->rq_xprt; 872 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 873 struct sock *sk = transport->inet; 874 int ret = -EAGAIN; 875 876 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", 877 req->rq_task->tk_pid, 878 req->rq_slen - transport->xmit.offset, 879 req->rq_slen); 880 881 /* Protect against races with write_space */ 882 spin_lock_bh(&xprt->transport_lock); 883 884 /* Don't race with disconnect */ 885 if (xprt_connected(xprt)) { 886 /* wait for more buffer space */ 887 sk->sk_write_pending++; 888 xprt_wait_for_buffer_space(xprt); 889 } else 890 ret = -ENOTCONN; 891 892 spin_unlock_bh(&xprt->transport_lock); 893 894 /* Race breaker in case memory is freed before above code is called */ 895 if (ret == -EAGAIN) { 896 struct socket_wq *wq; 897 898 rcu_read_lock(); 899 wq = rcu_dereference(sk->sk_wq); 900 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags); 901 rcu_read_unlock(); 902 903 sk->sk_write_space(sk); 904 } 905 return ret; 906 } 907 908 static void 909 xs_stream_prepare_request(struct rpc_rqst *req) 910 { 911 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_NOIO); 912 } 913 914 /* 915 * Determine if the previous message in the stream was aborted before it 916 * could complete transmission. 917 */ 918 static bool 919 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req) 920 { 921 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0; 922 } 923 924 /* 925 * Return the stream record marker field for a record of length < 2^31-1 926 */ 927 static rpc_fraghdr 928 xs_stream_record_marker(struct xdr_buf *xdr) 929 { 930 if (!xdr->len) 931 return 0; 932 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len); 933 } 934 935 /** 936 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 937 * @req: pointer to RPC request 938 * 939 * Return values: 940 * 0: The request has been sent 941 * EAGAIN: The socket was blocked, please call again later to 942 * complete the request 943 * ENOTCONN: Caller needs to invoke connect logic then call again 944 * other: Some other error occured, the request was not sent 945 */ 946 static int xs_local_send_request(struct rpc_rqst *req) 947 { 948 struct rpc_xprt *xprt = req->rq_xprt; 949 struct sock_xprt *transport = 950 container_of(xprt, struct sock_xprt, xprt); 951 struct xdr_buf *xdr = &req->rq_snd_buf; 952 int status; 953 int sent = 0; 954 955 /* Close the stream if the previous transmission was incomplete */ 956 if (xs_send_request_was_aborted(transport, req)) { 957 xs_close(xprt); 958 return -ENOTCONN; 959 } 960 961 xs_pktdump("packet data:", 962 req->rq_svec->iov_base, req->rq_svec->iov_len); 963 964 req->rq_xtime = ktime_get(); 965 status = xs_sendpages(transport->sock, NULL, 0, xdr, 966 transport->xmit.offset, 967 xs_stream_record_marker(xdr), 968 &sent); 969 dprintk("RPC: %s(%u) = %d\n", 970 __func__, xdr->len - transport->xmit.offset, status); 971 972 if (status == -EAGAIN && sock_writeable(transport->inet)) 973 status = -ENOBUFS; 974 975 if (likely(sent > 0) || status == 0) { 976 transport->xmit.offset += sent; 977 req->rq_bytes_sent = transport->xmit.offset; 978 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 979 req->rq_xmit_bytes_sent += transport->xmit.offset; 980 transport->xmit.offset = 0; 981 return 0; 982 } 983 status = -EAGAIN; 984 } 985 986 switch (status) { 987 case -ENOBUFS: 988 break; 989 case -EAGAIN: 990 status = xs_nospace(req); 991 break; 992 default: 993 dprintk("RPC: sendmsg returned unrecognized error %d\n", 994 -status); 995 /* fall through */ 996 case -EPIPE: 997 xs_close(xprt); 998 status = -ENOTCONN; 999 } 1000 1001 return status; 1002 } 1003 1004 /** 1005 * xs_udp_send_request - write an RPC request to a UDP socket 1006 * @req: pointer to RPC request 1007 * 1008 * Return values: 1009 * 0: The request has been sent 1010 * EAGAIN: The socket was blocked, please call again later to 1011 * complete the request 1012 * ENOTCONN: Caller needs to invoke connect logic then call again 1013 * other: Some other error occurred, the request was not sent 1014 */ 1015 static int xs_udp_send_request(struct rpc_rqst *req) 1016 { 1017 struct rpc_xprt *xprt = req->rq_xprt; 1018 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1019 struct xdr_buf *xdr = &req->rq_snd_buf; 1020 int sent = 0; 1021 int status; 1022 1023 xs_pktdump("packet data:", 1024 req->rq_svec->iov_base, 1025 req->rq_svec->iov_len); 1026 1027 if (!xprt_bound(xprt)) 1028 return -ENOTCONN; 1029 1030 if (!xprt_request_get_cong(xprt, req)) 1031 return -EBADSLT; 1032 1033 req->rq_xtime = ktime_get(); 1034 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen, 1035 xdr, 0, 0, &sent); 1036 1037 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 1038 xdr->len, status); 1039 1040 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 1041 if (status == -EPERM) 1042 goto process_status; 1043 1044 if (status == -EAGAIN && sock_writeable(transport->inet)) 1045 status = -ENOBUFS; 1046 1047 if (sent > 0 || status == 0) { 1048 req->rq_xmit_bytes_sent += sent; 1049 if (sent >= req->rq_slen) 1050 return 0; 1051 /* Still some bytes left; set up for a retry later. */ 1052 status = -EAGAIN; 1053 } 1054 1055 process_status: 1056 switch (status) { 1057 case -ENOTSOCK: 1058 status = -ENOTCONN; 1059 /* Should we call xs_close() here? */ 1060 break; 1061 case -EAGAIN: 1062 status = xs_nospace(req); 1063 break; 1064 case -ENETUNREACH: 1065 case -ENOBUFS: 1066 case -EPIPE: 1067 case -ECONNREFUSED: 1068 case -EPERM: 1069 /* When the server has died, an ICMP port unreachable message 1070 * prompts ECONNREFUSED. */ 1071 break; 1072 default: 1073 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1074 -status); 1075 } 1076 1077 return status; 1078 } 1079 1080 /** 1081 * xs_tcp_send_request - write an RPC request to a TCP socket 1082 * @req: pointer to RPC request 1083 * 1084 * Return values: 1085 * 0: The request has been sent 1086 * EAGAIN: The socket was blocked, please call again later to 1087 * complete the request 1088 * ENOTCONN: Caller needs to invoke connect logic then call again 1089 * other: Some other error occurred, the request was not sent 1090 * 1091 * XXX: In the case of soft timeouts, should we eventually give up 1092 * if sendmsg is not able to make progress? 1093 */ 1094 static int xs_tcp_send_request(struct rpc_rqst *req) 1095 { 1096 struct rpc_xprt *xprt = req->rq_xprt; 1097 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1098 struct xdr_buf *xdr = &req->rq_snd_buf; 1099 bool vm_wait = false; 1100 int status; 1101 int sent; 1102 1103 /* Close the stream if the previous transmission was incomplete */ 1104 if (xs_send_request_was_aborted(transport, req)) { 1105 if (transport->sock != NULL) 1106 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 1107 return -ENOTCONN; 1108 } 1109 1110 xs_pktdump("packet data:", 1111 req->rq_svec->iov_base, 1112 req->rq_svec->iov_len); 1113 1114 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state)) 1115 xs_tcp_set_socket_timeouts(xprt, transport->sock); 1116 1117 /* Continue transmitting the packet/record. We must be careful 1118 * to cope with writespace callbacks arriving _after_ we have 1119 * called sendmsg(). */ 1120 req->rq_xtime = ktime_get(); 1121 while (1) { 1122 sent = 0; 1123 status = xs_sendpages(transport->sock, NULL, 0, xdr, 1124 transport->xmit.offset, 1125 xs_stream_record_marker(xdr), 1126 &sent); 1127 1128 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 1129 xdr->len - transport->xmit.offset, status); 1130 1131 /* If we've sent the entire packet, immediately 1132 * reset the count of bytes sent. */ 1133 transport->xmit.offset += sent; 1134 req->rq_bytes_sent = transport->xmit.offset; 1135 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 1136 req->rq_xmit_bytes_sent += transport->xmit.offset; 1137 transport->xmit.offset = 0; 1138 return 0; 1139 } 1140 1141 WARN_ON_ONCE(sent == 0 && status == 0); 1142 1143 if (status == -EAGAIN ) { 1144 /* 1145 * Return EAGAIN if we're sure we're hitting the 1146 * socket send buffer limits. 1147 */ 1148 if (test_bit(SOCK_NOSPACE, &transport->sock->flags)) 1149 break; 1150 /* 1151 * Did we hit a memory allocation failure? 1152 */ 1153 if (sent == 0) { 1154 status = -ENOBUFS; 1155 if (vm_wait) 1156 break; 1157 /* Retry, knowing now that we're below the 1158 * socket send buffer limit 1159 */ 1160 vm_wait = true; 1161 } 1162 continue; 1163 } 1164 if (status < 0) 1165 break; 1166 vm_wait = false; 1167 } 1168 1169 switch (status) { 1170 case -ENOTSOCK: 1171 status = -ENOTCONN; 1172 /* Should we call xs_close() here? */ 1173 break; 1174 case -EAGAIN: 1175 status = xs_nospace(req); 1176 break; 1177 case -ECONNRESET: 1178 case -ECONNREFUSED: 1179 case -ENOTCONN: 1180 case -EADDRINUSE: 1181 case -ENOBUFS: 1182 case -EPIPE: 1183 break; 1184 default: 1185 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1186 -status); 1187 } 1188 1189 return status; 1190 } 1191 1192 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1193 { 1194 transport->old_data_ready = sk->sk_data_ready; 1195 transport->old_state_change = sk->sk_state_change; 1196 transport->old_write_space = sk->sk_write_space; 1197 transport->old_error_report = sk->sk_error_report; 1198 } 1199 1200 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1201 { 1202 sk->sk_data_ready = transport->old_data_ready; 1203 sk->sk_state_change = transport->old_state_change; 1204 sk->sk_write_space = transport->old_write_space; 1205 sk->sk_error_report = transport->old_error_report; 1206 } 1207 1208 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt) 1209 { 1210 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1211 1212 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 1213 } 1214 1215 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1216 { 1217 smp_mb__before_atomic(); 1218 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1219 clear_bit(XPRT_CLOSING, &xprt->state); 1220 xs_sock_reset_state_flags(xprt); 1221 smp_mb__after_atomic(); 1222 } 1223 1224 /** 1225 * xs_error_report - callback to handle TCP socket state errors 1226 * @sk: socket 1227 * 1228 * Note: we don't call sock_error() since there may be a rpc_task 1229 * using the socket, and so we don't want to clear sk->sk_err. 1230 */ 1231 static void xs_error_report(struct sock *sk) 1232 { 1233 struct rpc_xprt *xprt; 1234 int err; 1235 1236 read_lock_bh(&sk->sk_callback_lock); 1237 if (!(xprt = xprt_from_sock(sk))) 1238 goto out; 1239 1240 err = -sk->sk_err; 1241 if (err == 0) 1242 goto out; 1243 dprintk("RPC: xs_error_report client %p, error=%d...\n", 1244 xprt, -err); 1245 trace_rpc_socket_error(xprt, sk->sk_socket, err); 1246 xprt_wake_pending_tasks(xprt, err); 1247 out: 1248 read_unlock_bh(&sk->sk_callback_lock); 1249 } 1250 1251 static void xs_reset_transport(struct sock_xprt *transport) 1252 { 1253 struct socket *sock = transport->sock; 1254 struct sock *sk = transport->inet; 1255 struct rpc_xprt *xprt = &transport->xprt; 1256 1257 if (sk == NULL) 1258 return; 1259 1260 if (atomic_read(&transport->xprt.swapper)) 1261 sk_clear_memalloc(sk); 1262 1263 kernel_sock_shutdown(sock, SHUT_RDWR); 1264 1265 mutex_lock(&transport->recv_mutex); 1266 write_lock_bh(&sk->sk_callback_lock); 1267 transport->inet = NULL; 1268 transport->sock = NULL; 1269 1270 sk->sk_user_data = NULL; 1271 1272 xs_restore_old_callbacks(transport, sk); 1273 xprt_clear_connected(xprt); 1274 write_unlock_bh(&sk->sk_callback_lock); 1275 xs_sock_reset_connection_flags(xprt); 1276 /* Reset stream record info */ 1277 xs_stream_reset_connect(transport); 1278 mutex_unlock(&transport->recv_mutex); 1279 1280 trace_rpc_socket_close(xprt, sock); 1281 sock_release(sock); 1282 1283 xprt_disconnect_done(xprt); 1284 } 1285 1286 /** 1287 * xs_close - close a socket 1288 * @xprt: transport 1289 * 1290 * This is used when all requests are complete; ie, no DRC state remains 1291 * on the server we want to save. 1292 * 1293 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 1294 * xs_reset_transport() zeroing the socket from underneath a writer. 1295 */ 1296 static void xs_close(struct rpc_xprt *xprt) 1297 { 1298 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1299 1300 dprintk("RPC: xs_close xprt %p\n", xprt); 1301 1302 xs_reset_transport(transport); 1303 xprt->reestablish_timeout = 0; 1304 } 1305 1306 static void xs_inject_disconnect(struct rpc_xprt *xprt) 1307 { 1308 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 1309 xprt); 1310 xprt_disconnect_done(xprt); 1311 } 1312 1313 static void xs_xprt_free(struct rpc_xprt *xprt) 1314 { 1315 xs_free_peer_addresses(xprt); 1316 xprt_free(xprt); 1317 } 1318 1319 /** 1320 * xs_destroy - prepare to shutdown a transport 1321 * @xprt: doomed transport 1322 * 1323 */ 1324 static void xs_destroy(struct rpc_xprt *xprt) 1325 { 1326 struct sock_xprt *transport = container_of(xprt, 1327 struct sock_xprt, xprt); 1328 dprintk("RPC: xs_destroy xprt %p\n", xprt); 1329 1330 cancel_delayed_work_sync(&transport->connect_worker); 1331 xs_close(xprt); 1332 cancel_work_sync(&transport->recv_worker); 1333 xs_xprt_free(xprt); 1334 module_put(THIS_MODULE); 1335 } 1336 1337 /** 1338 * xs_udp_data_read_skb - receive callback for UDP sockets 1339 * @xprt: transport 1340 * @sk: socket 1341 * @skb: skbuff 1342 * 1343 */ 1344 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1345 struct sock *sk, 1346 struct sk_buff *skb) 1347 { 1348 struct rpc_task *task; 1349 struct rpc_rqst *rovr; 1350 int repsize, copied; 1351 u32 _xid; 1352 __be32 *xp; 1353 1354 repsize = skb->len; 1355 if (repsize < 4) { 1356 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1357 return; 1358 } 1359 1360 /* Copy the XID from the skb... */ 1361 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid); 1362 if (xp == NULL) 1363 return; 1364 1365 /* Look up and lock the request corresponding to the given XID */ 1366 spin_lock(&xprt->queue_lock); 1367 rovr = xprt_lookup_rqst(xprt, *xp); 1368 if (!rovr) 1369 goto out_unlock; 1370 xprt_pin_rqst(rovr); 1371 xprt_update_rtt(rovr->rq_task); 1372 spin_unlock(&xprt->queue_lock); 1373 task = rovr->rq_task; 1374 1375 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1376 copied = repsize; 1377 1378 /* Suck it into the iovec, verify checksum if not done by hw. */ 1379 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1380 spin_lock(&xprt->queue_lock); 1381 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS); 1382 goto out_unpin; 1383 } 1384 1385 1386 spin_lock_bh(&xprt->transport_lock); 1387 xprt_adjust_cwnd(xprt, task, copied); 1388 spin_unlock_bh(&xprt->transport_lock); 1389 spin_lock(&xprt->queue_lock); 1390 xprt_complete_rqst(task, copied); 1391 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS); 1392 out_unpin: 1393 xprt_unpin_rqst(rovr); 1394 out_unlock: 1395 spin_unlock(&xprt->queue_lock); 1396 } 1397 1398 static void xs_udp_data_receive(struct sock_xprt *transport) 1399 { 1400 struct sk_buff *skb; 1401 struct sock *sk; 1402 int err; 1403 1404 mutex_lock(&transport->recv_mutex); 1405 sk = transport->inet; 1406 if (sk == NULL) 1407 goto out; 1408 for (;;) { 1409 skb = skb_recv_udp(sk, 0, 1, &err); 1410 if (skb == NULL) 1411 break; 1412 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1413 consume_skb(skb); 1414 cond_resched(); 1415 } 1416 xs_poll_check_readable(transport); 1417 out: 1418 mutex_unlock(&transport->recv_mutex); 1419 } 1420 1421 static void xs_udp_data_receive_workfn(struct work_struct *work) 1422 { 1423 struct sock_xprt *transport = 1424 container_of(work, struct sock_xprt, recv_worker); 1425 unsigned int pflags = memalloc_nofs_save(); 1426 1427 xs_udp_data_receive(transport); 1428 memalloc_nofs_restore(pflags); 1429 } 1430 1431 /** 1432 * xs_data_ready - "data ready" callback for UDP sockets 1433 * @sk: socket with data to read 1434 * 1435 */ 1436 static void xs_data_ready(struct sock *sk) 1437 { 1438 struct rpc_xprt *xprt; 1439 1440 read_lock_bh(&sk->sk_callback_lock); 1441 dprintk("RPC: xs_data_ready...\n"); 1442 xprt = xprt_from_sock(sk); 1443 if (xprt != NULL) { 1444 struct sock_xprt *transport = container_of(xprt, 1445 struct sock_xprt, xprt); 1446 transport->old_data_ready(sk); 1447 /* Any data means we had a useful conversation, so 1448 * then we don't need to delay the next reconnect 1449 */ 1450 if (xprt->reestablish_timeout) 1451 xprt->reestablish_timeout = 0; 1452 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1453 queue_work(xprtiod_workqueue, &transport->recv_worker); 1454 } 1455 read_unlock_bh(&sk->sk_callback_lock); 1456 } 1457 1458 /* 1459 * Helper function to force a TCP close if the server is sending 1460 * junk and/or it has put us in CLOSE_WAIT 1461 */ 1462 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1463 { 1464 xprt_force_disconnect(xprt); 1465 } 1466 1467 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1468 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt) 1469 { 1470 return PAGE_SIZE; 1471 } 1472 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1473 1474 /** 1475 * xs_tcp_state_change - callback to handle TCP socket state changes 1476 * @sk: socket whose state has changed 1477 * 1478 */ 1479 static void xs_tcp_state_change(struct sock *sk) 1480 { 1481 struct rpc_xprt *xprt; 1482 struct sock_xprt *transport; 1483 1484 read_lock_bh(&sk->sk_callback_lock); 1485 if (!(xprt = xprt_from_sock(sk))) 1486 goto out; 1487 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1488 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1489 sk->sk_state, xprt_connected(xprt), 1490 sock_flag(sk, SOCK_DEAD), 1491 sock_flag(sk, SOCK_ZAPPED), 1492 sk->sk_shutdown); 1493 1494 transport = container_of(xprt, struct sock_xprt, xprt); 1495 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1496 switch (sk->sk_state) { 1497 case TCP_ESTABLISHED: 1498 spin_lock(&xprt->transport_lock); 1499 if (!xprt_test_and_set_connected(xprt)) { 1500 xprt->connect_cookie++; 1501 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1502 xprt_clear_connecting(xprt); 1503 1504 xprt->stat.connect_count++; 1505 xprt->stat.connect_time += (long)jiffies - 1506 xprt->stat.connect_start; 1507 xprt_wake_pending_tasks(xprt, -EAGAIN); 1508 } 1509 spin_unlock(&xprt->transport_lock); 1510 break; 1511 case TCP_FIN_WAIT1: 1512 /* The client initiated a shutdown of the socket */ 1513 xprt->connect_cookie++; 1514 xprt->reestablish_timeout = 0; 1515 set_bit(XPRT_CLOSING, &xprt->state); 1516 smp_mb__before_atomic(); 1517 clear_bit(XPRT_CONNECTED, &xprt->state); 1518 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1519 smp_mb__after_atomic(); 1520 break; 1521 case TCP_CLOSE_WAIT: 1522 /* The server initiated a shutdown of the socket */ 1523 xprt->connect_cookie++; 1524 clear_bit(XPRT_CONNECTED, &xprt->state); 1525 xs_tcp_force_close(xprt); 1526 /* fall through */ 1527 case TCP_CLOSING: 1528 /* 1529 * If the server closed down the connection, make sure that 1530 * we back off before reconnecting 1531 */ 1532 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1533 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1534 break; 1535 case TCP_LAST_ACK: 1536 set_bit(XPRT_CLOSING, &xprt->state); 1537 smp_mb__before_atomic(); 1538 clear_bit(XPRT_CONNECTED, &xprt->state); 1539 smp_mb__after_atomic(); 1540 break; 1541 case TCP_CLOSE: 1542 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1543 &transport->sock_state)) 1544 xprt_clear_connecting(xprt); 1545 clear_bit(XPRT_CLOSING, &xprt->state); 1546 /* Trigger the socket release */ 1547 xs_tcp_force_close(xprt); 1548 } 1549 out: 1550 read_unlock_bh(&sk->sk_callback_lock); 1551 } 1552 1553 static void xs_write_space(struct sock *sk) 1554 { 1555 struct socket_wq *wq; 1556 struct rpc_xprt *xprt; 1557 1558 if (!sk->sk_socket) 1559 return; 1560 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1561 1562 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1563 return; 1564 rcu_read_lock(); 1565 wq = rcu_dereference(sk->sk_wq); 1566 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0) 1567 goto out; 1568 1569 if (xprt_write_space(xprt)) 1570 sk->sk_write_pending--; 1571 out: 1572 rcu_read_unlock(); 1573 } 1574 1575 /** 1576 * xs_udp_write_space - callback invoked when socket buffer space 1577 * becomes available 1578 * @sk: socket whose state has changed 1579 * 1580 * Called when more output buffer space is available for this socket. 1581 * We try not to wake our writers until they can make "significant" 1582 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1583 * with a bunch of small requests. 1584 */ 1585 static void xs_udp_write_space(struct sock *sk) 1586 { 1587 read_lock_bh(&sk->sk_callback_lock); 1588 1589 /* from net/core/sock.c:sock_def_write_space */ 1590 if (sock_writeable(sk)) 1591 xs_write_space(sk); 1592 1593 read_unlock_bh(&sk->sk_callback_lock); 1594 } 1595 1596 /** 1597 * xs_tcp_write_space - callback invoked when socket buffer space 1598 * becomes available 1599 * @sk: socket whose state has changed 1600 * 1601 * Called when more output buffer space is available for this socket. 1602 * We try not to wake our writers until they can make "significant" 1603 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1604 * with a bunch of small requests. 1605 */ 1606 static void xs_tcp_write_space(struct sock *sk) 1607 { 1608 read_lock_bh(&sk->sk_callback_lock); 1609 1610 /* from net/core/stream.c:sk_stream_write_space */ 1611 if (sk_stream_is_writeable(sk)) 1612 xs_write_space(sk); 1613 1614 read_unlock_bh(&sk->sk_callback_lock); 1615 } 1616 1617 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1618 { 1619 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1620 struct sock *sk = transport->inet; 1621 1622 if (transport->rcvsize) { 1623 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1624 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1625 } 1626 if (transport->sndsize) { 1627 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1628 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1629 sk->sk_write_space(sk); 1630 } 1631 } 1632 1633 /** 1634 * xs_udp_set_buffer_size - set send and receive limits 1635 * @xprt: generic transport 1636 * @sndsize: requested size of send buffer, in bytes 1637 * @rcvsize: requested size of receive buffer, in bytes 1638 * 1639 * Set socket send and receive buffer size limits. 1640 */ 1641 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1642 { 1643 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1644 1645 transport->sndsize = 0; 1646 if (sndsize) 1647 transport->sndsize = sndsize + 1024; 1648 transport->rcvsize = 0; 1649 if (rcvsize) 1650 transport->rcvsize = rcvsize + 1024; 1651 1652 xs_udp_do_set_buffer_size(xprt); 1653 } 1654 1655 /** 1656 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1657 * @xprt: controlling transport 1658 * @task: task that timed out 1659 * 1660 * Adjust the congestion window after a retransmit timeout has occurred. 1661 */ 1662 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1663 { 1664 spin_lock_bh(&xprt->transport_lock); 1665 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1666 spin_unlock_bh(&xprt->transport_lock); 1667 } 1668 1669 static int xs_get_random_port(void) 1670 { 1671 unsigned short min = xprt_min_resvport, max = xprt_max_resvport; 1672 unsigned short range; 1673 unsigned short rand; 1674 1675 if (max < min) 1676 return -EADDRINUSE; 1677 range = max - min + 1; 1678 rand = (unsigned short) prandom_u32() % range; 1679 return rand + min; 1680 } 1681 1682 /** 1683 * xs_set_reuseaddr_port - set the socket's port and address reuse options 1684 * @sock: socket 1685 * 1686 * Note that this function has to be called on all sockets that share the 1687 * same port, and it must be called before binding. 1688 */ 1689 static void xs_sock_set_reuseport(struct socket *sock) 1690 { 1691 int opt = 1; 1692 1693 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, 1694 (char *)&opt, sizeof(opt)); 1695 } 1696 1697 static unsigned short xs_sock_getport(struct socket *sock) 1698 { 1699 struct sockaddr_storage buf; 1700 unsigned short port = 0; 1701 1702 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0) 1703 goto out; 1704 switch (buf.ss_family) { 1705 case AF_INET6: 1706 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1707 break; 1708 case AF_INET: 1709 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1710 } 1711 out: 1712 return port; 1713 } 1714 1715 /** 1716 * xs_set_port - reset the port number in the remote endpoint address 1717 * @xprt: generic transport 1718 * @port: new port number 1719 * 1720 */ 1721 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1722 { 1723 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1724 1725 rpc_set_port(xs_addr(xprt), port); 1726 xs_update_peer_port(xprt); 1727 } 1728 1729 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1730 { 1731 if (transport->srcport == 0) 1732 transport->srcport = xs_sock_getport(sock); 1733 } 1734 1735 static int xs_get_srcport(struct sock_xprt *transport) 1736 { 1737 int port = transport->srcport; 1738 1739 if (port == 0 && transport->xprt.resvport) 1740 port = xs_get_random_port(); 1741 return port; 1742 } 1743 1744 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1745 { 1746 if (transport->srcport != 0) 1747 transport->srcport = 0; 1748 if (!transport->xprt.resvport) 1749 return 0; 1750 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1751 return xprt_max_resvport; 1752 return --port; 1753 } 1754 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1755 { 1756 struct sockaddr_storage myaddr; 1757 int err, nloop = 0; 1758 int port = xs_get_srcport(transport); 1759 unsigned short last; 1760 1761 /* 1762 * If we are asking for any ephemeral port (i.e. port == 0 && 1763 * transport->xprt.resvport == 0), don't bind. Let the local 1764 * port selection happen implicitly when the socket is used 1765 * (for example at connect time). 1766 * 1767 * This ensures that we can continue to establish TCP 1768 * connections even when all local ephemeral ports are already 1769 * a part of some TCP connection. This makes no difference 1770 * for UDP sockets, but also doens't harm them. 1771 * 1772 * If we're asking for any reserved port (i.e. port == 0 && 1773 * transport->xprt.resvport == 1) xs_get_srcport above will 1774 * ensure that port is non-zero and we will bind as needed. 1775 */ 1776 if (port <= 0) 1777 return port; 1778 1779 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1780 do { 1781 rpc_set_port((struct sockaddr *)&myaddr, port); 1782 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1783 transport->xprt.addrlen); 1784 if (err == 0) { 1785 transport->srcport = port; 1786 break; 1787 } 1788 last = port; 1789 port = xs_next_srcport(transport, port); 1790 if (port > last) 1791 nloop++; 1792 } while (err == -EADDRINUSE && nloop != 2); 1793 1794 if (myaddr.ss_family == AF_INET) 1795 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1796 &((struct sockaddr_in *)&myaddr)->sin_addr, 1797 port, err ? "failed" : "ok", err); 1798 else 1799 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1800 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1801 port, err ? "failed" : "ok", err); 1802 return err; 1803 } 1804 1805 /* 1806 * We don't support autobind on AF_LOCAL sockets 1807 */ 1808 static void xs_local_rpcbind(struct rpc_task *task) 1809 { 1810 xprt_set_bound(task->tk_xprt); 1811 } 1812 1813 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1814 { 1815 } 1816 1817 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1818 static struct lock_class_key xs_key[2]; 1819 static struct lock_class_key xs_slock_key[2]; 1820 1821 static inline void xs_reclassify_socketu(struct socket *sock) 1822 { 1823 struct sock *sk = sock->sk; 1824 1825 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1826 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); 1827 } 1828 1829 static inline void xs_reclassify_socket4(struct socket *sock) 1830 { 1831 struct sock *sk = sock->sk; 1832 1833 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1834 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1835 } 1836 1837 static inline void xs_reclassify_socket6(struct socket *sock) 1838 { 1839 struct sock *sk = sock->sk; 1840 1841 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1842 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1843 } 1844 1845 static inline void xs_reclassify_socket(int family, struct socket *sock) 1846 { 1847 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk))) 1848 return; 1849 1850 switch (family) { 1851 case AF_LOCAL: 1852 xs_reclassify_socketu(sock); 1853 break; 1854 case AF_INET: 1855 xs_reclassify_socket4(sock); 1856 break; 1857 case AF_INET6: 1858 xs_reclassify_socket6(sock); 1859 break; 1860 } 1861 } 1862 #else 1863 static inline void xs_reclassify_socket(int family, struct socket *sock) 1864 { 1865 } 1866 #endif 1867 1868 static void xs_dummy_setup_socket(struct work_struct *work) 1869 { 1870 } 1871 1872 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1873 struct sock_xprt *transport, int family, int type, 1874 int protocol, bool reuseport) 1875 { 1876 struct socket *sock; 1877 int err; 1878 1879 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1880 if (err < 0) { 1881 dprintk("RPC: can't create %d transport socket (%d).\n", 1882 protocol, -err); 1883 goto out; 1884 } 1885 xs_reclassify_socket(family, sock); 1886 1887 if (reuseport) 1888 xs_sock_set_reuseport(sock); 1889 1890 err = xs_bind(transport, sock); 1891 if (err) { 1892 sock_release(sock); 1893 goto out; 1894 } 1895 1896 return sock; 1897 out: 1898 return ERR_PTR(err); 1899 } 1900 1901 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1902 struct socket *sock) 1903 { 1904 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1905 xprt); 1906 1907 if (!transport->inet) { 1908 struct sock *sk = sock->sk; 1909 1910 write_lock_bh(&sk->sk_callback_lock); 1911 1912 xs_save_old_callbacks(transport, sk); 1913 1914 sk->sk_user_data = xprt; 1915 sk->sk_data_ready = xs_data_ready; 1916 sk->sk_write_space = xs_udp_write_space; 1917 sock_set_flag(sk, SOCK_FASYNC); 1918 sk->sk_error_report = xs_error_report; 1919 1920 xprt_clear_connected(xprt); 1921 1922 /* Reset to new socket */ 1923 transport->sock = sock; 1924 transport->inet = sk; 1925 1926 write_unlock_bh(&sk->sk_callback_lock); 1927 } 1928 1929 xs_stream_start_connect(transport); 1930 1931 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1932 } 1933 1934 /** 1935 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1936 * @transport: socket transport to connect 1937 */ 1938 static int xs_local_setup_socket(struct sock_xprt *transport) 1939 { 1940 struct rpc_xprt *xprt = &transport->xprt; 1941 struct socket *sock; 1942 int status = -EIO; 1943 1944 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1945 SOCK_STREAM, 0, &sock, 1); 1946 if (status < 0) { 1947 dprintk("RPC: can't create AF_LOCAL " 1948 "transport socket (%d).\n", -status); 1949 goto out; 1950 } 1951 xs_reclassify_socket(AF_LOCAL, sock); 1952 1953 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1954 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1955 1956 status = xs_local_finish_connecting(xprt, sock); 1957 trace_rpc_socket_connect(xprt, sock, status); 1958 switch (status) { 1959 case 0: 1960 dprintk("RPC: xprt %p connected to %s\n", 1961 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1962 xprt->stat.connect_count++; 1963 xprt->stat.connect_time += (long)jiffies - 1964 xprt->stat.connect_start; 1965 xprt_set_connected(xprt); 1966 case -ENOBUFS: 1967 break; 1968 case -ENOENT: 1969 dprintk("RPC: xprt %p: socket %s does not exist\n", 1970 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1971 break; 1972 case -ECONNREFUSED: 1973 dprintk("RPC: xprt %p: connection refused for %s\n", 1974 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1975 break; 1976 default: 1977 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 1978 __func__, -status, 1979 xprt->address_strings[RPC_DISPLAY_ADDR]); 1980 } 1981 1982 out: 1983 xprt_clear_connecting(xprt); 1984 xprt_wake_pending_tasks(xprt, status); 1985 return status; 1986 } 1987 1988 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 1989 { 1990 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1991 int ret; 1992 1993 if (RPC_IS_ASYNC(task)) { 1994 /* 1995 * We want the AF_LOCAL connect to be resolved in the 1996 * filesystem namespace of the process making the rpc 1997 * call. Thus we connect synchronously. 1998 * 1999 * If we want to support asynchronous AF_LOCAL calls, 2000 * we'll need to figure out how to pass a namespace to 2001 * connect. 2002 */ 2003 rpc_exit(task, -ENOTCONN); 2004 return; 2005 } 2006 ret = xs_local_setup_socket(transport); 2007 if (ret && !RPC_IS_SOFTCONN(task)) 2008 msleep_interruptible(15000); 2009 } 2010 2011 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 2012 /* 2013 * Note that this should be called with XPRT_LOCKED held (or when we otherwise 2014 * know that we have exclusive access to the socket), to guard against 2015 * races with xs_reset_transport. 2016 */ 2017 static void xs_set_memalloc(struct rpc_xprt *xprt) 2018 { 2019 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2020 xprt); 2021 2022 /* 2023 * If there's no sock, then we have nothing to set. The 2024 * reconnecting process will get it for us. 2025 */ 2026 if (!transport->inet) 2027 return; 2028 if (atomic_read(&xprt->swapper)) 2029 sk_set_memalloc(transport->inet); 2030 } 2031 2032 /** 2033 * xs_enable_swap - Tag this transport as being used for swap. 2034 * @xprt: transport to tag 2035 * 2036 * Take a reference to this transport on behalf of the rpc_clnt, and 2037 * optionally mark it for swapping if it wasn't already. 2038 */ 2039 static int 2040 xs_enable_swap(struct rpc_xprt *xprt) 2041 { 2042 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2043 2044 if (atomic_inc_return(&xprt->swapper) != 1) 2045 return 0; 2046 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 2047 return -ERESTARTSYS; 2048 if (xs->inet) 2049 sk_set_memalloc(xs->inet); 2050 xprt_release_xprt(xprt, NULL); 2051 return 0; 2052 } 2053 2054 /** 2055 * xs_disable_swap - Untag this transport as being used for swap. 2056 * @xprt: transport to tag 2057 * 2058 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2059 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2060 */ 2061 static void 2062 xs_disable_swap(struct rpc_xprt *xprt) 2063 { 2064 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2065 2066 if (!atomic_dec_and_test(&xprt->swapper)) 2067 return; 2068 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 2069 return; 2070 if (xs->inet) 2071 sk_clear_memalloc(xs->inet); 2072 xprt_release_xprt(xprt, NULL); 2073 } 2074 #else 2075 static void xs_set_memalloc(struct rpc_xprt *xprt) 2076 { 2077 } 2078 2079 static int 2080 xs_enable_swap(struct rpc_xprt *xprt) 2081 { 2082 return -EINVAL; 2083 } 2084 2085 static void 2086 xs_disable_swap(struct rpc_xprt *xprt) 2087 { 2088 } 2089 #endif 2090 2091 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2092 { 2093 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2094 2095 if (!transport->inet) { 2096 struct sock *sk = sock->sk; 2097 2098 write_lock_bh(&sk->sk_callback_lock); 2099 2100 xs_save_old_callbacks(transport, sk); 2101 2102 sk->sk_user_data = xprt; 2103 sk->sk_data_ready = xs_data_ready; 2104 sk->sk_write_space = xs_udp_write_space; 2105 sock_set_flag(sk, SOCK_FASYNC); 2106 2107 xprt_set_connected(xprt); 2108 2109 /* Reset to new socket */ 2110 transport->sock = sock; 2111 transport->inet = sk; 2112 2113 xs_set_memalloc(xprt); 2114 2115 write_unlock_bh(&sk->sk_callback_lock); 2116 } 2117 xs_udp_do_set_buffer_size(xprt); 2118 2119 xprt->stat.connect_start = jiffies; 2120 } 2121 2122 static void xs_udp_setup_socket(struct work_struct *work) 2123 { 2124 struct sock_xprt *transport = 2125 container_of(work, struct sock_xprt, connect_worker.work); 2126 struct rpc_xprt *xprt = &transport->xprt; 2127 struct socket *sock; 2128 int status = -EIO; 2129 2130 sock = xs_create_sock(xprt, transport, 2131 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2132 IPPROTO_UDP, false); 2133 if (IS_ERR(sock)) 2134 goto out; 2135 2136 dprintk("RPC: worker connecting xprt %p via %s to " 2137 "%s (port %s)\n", xprt, 2138 xprt->address_strings[RPC_DISPLAY_PROTO], 2139 xprt->address_strings[RPC_DISPLAY_ADDR], 2140 xprt->address_strings[RPC_DISPLAY_PORT]); 2141 2142 xs_udp_finish_connecting(xprt, sock); 2143 trace_rpc_socket_connect(xprt, sock, 0); 2144 status = 0; 2145 out: 2146 xprt_clear_connecting(xprt); 2147 xprt_unlock_connect(xprt, transport); 2148 xprt_wake_pending_tasks(xprt, status); 2149 } 2150 2151 /** 2152 * xs_tcp_shutdown - gracefully shut down a TCP socket 2153 * @xprt: transport 2154 * 2155 * Initiates a graceful shutdown of the TCP socket by calling the 2156 * equivalent of shutdown(SHUT_RDWR); 2157 */ 2158 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2159 { 2160 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2161 struct socket *sock = transport->sock; 2162 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE; 2163 2164 if (sock == NULL) 2165 return; 2166 switch (skst) { 2167 default: 2168 kernel_sock_shutdown(sock, SHUT_RDWR); 2169 trace_rpc_socket_shutdown(xprt, sock); 2170 break; 2171 case TCP_CLOSE: 2172 case TCP_TIME_WAIT: 2173 xs_reset_transport(transport); 2174 } 2175 } 2176 2177 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 2178 struct socket *sock) 2179 { 2180 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2181 unsigned int keepidle; 2182 unsigned int keepcnt; 2183 unsigned int opt_on = 1; 2184 unsigned int timeo; 2185 2186 spin_lock_bh(&xprt->transport_lock); 2187 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ); 2188 keepcnt = xprt->timeout->to_retries + 1; 2189 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2190 (xprt->timeout->to_retries + 1); 2191 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2192 spin_unlock_bh(&xprt->transport_lock); 2193 2194 /* TCP Keepalive options */ 2195 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 2196 (char *)&opt_on, sizeof(opt_on)); 2197 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, 2198 (char *)&keepidle, sizeof(keepidle)); 2199 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, 2200 (char *)&keepidle, sizeof(keepidle)); 2201 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, 2202 (char *)&keepcnt, sizeof(keepcnt)); 2203 2204 /* TCP user timeout (see RFC5482) */ 2205 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT, 2206 (char *)&timeo, sizeof(timeo)); 2207 } 2208 2209 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt, 2210 unsigned long connect_timeout, 2211 unsigned long reconnect_timeout) 2212 { 2213 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2214 struct rpc_timeout to; 2215 unsigned long initval; 2216 2217 spin_lock_bh(&xprt->transport_lock); 2218 if (reconnect_timeout < xprt->max_reconnect_timeout) 2219 xprt->max_reconnect_timeout = reconnect_timeout; 2220 if (connect_timeout < xprt->connect_timeout) { 2221 memcpy(&to, xprt->timeout, sizeof(to)); 2222 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1); 2223 /* Arbitrary lower limit */ 2224 if (initval < XS_TCP_INIT_REEST_TO << 1) 2225 initval = XS_TCP_INIT_REEST_TO << 1; 2226 to.to_initval = initval; 2227 to.to_maxval = initval; 2228 memcpy(&transport->tcp_timeout, &to, 2229 sizeof(transport->tcp_timeout)); 2230 xprt->timeout = &transport->tcp_timeout; 2231 xprt->connect_timeout = connect_timeout; 2232 } 2233 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2234 spin_unlock_bh(&xprt->transport_lock); 2235 } 2236 2237 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2238 { 2239 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2240 int ret = -ENOTCONN; 2241 2242 if (!transport->inet) { 2243 struct sock *sk = sock->sk; 2244 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC; 2245 2246 /* Avoid temporary address, they are bad for long-lived 2247 * connections such as NFS mounts. 2248 * RFC4941, section 3.6 suggests that: 2249 * Individual applications, which have specific 2250 * knowledge about the normal duration of connections, 2251 * MAY override this as appropriate. 2252 */ 2253 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES, 2254 (char *)&addr_pref, sizeof(addr_pref)); 2255 2256 xs_tcp_set_socket_timeouts(xprt, sock); 2257 2258 write_lock_bh(&sk->sk_callback_lock); 2259 2260 xs_save_old_callbacks(transport, sk); 2261 2262 sk->sk_user_data = xprt; 2263 sk->sk_data_ready = xs_data_ready; 2264 sk->sk_state_change = xs_tcp_state_change; 2265 sk->sk_write_space = xs_tcp_write_space; 2266 sock_set_flag(sk, SOCK_FASYNC); 2267 sk->sk_error_report = xs_error_report; 2268 2269 /* socket options */ 2270 sock_reset_flag(sk, SOCK_LINGER); 2271 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 2272 2273 xprt_clear_connected(xprt); 2274 2275 /* Reset to new socket */ 2276 transport->sock = sock; 2277 transport->inet = sk; 2278 2279 write_unlock_bh(&sk->sk_callback_lock); 2280 } 2281 2282 if (!xprt_bound(xprt)) 2283 goto out; 2284 2285 xs_set_memalloc(xprt); 2286 2287 xs_stream_start_connect(transport); 2288 2289 /* Tell the socket layer to start connecting... */ 2290 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2291 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2292 switch (ret) { 2293 case 0: 2294 xs_set_srcport(transport, sock); 2295 /* fall through */ 2296 case -EINPROGRESS: 2297 /* SYN_SENT! */ 2298 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2299 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2300 break; 2301 case -EADDRNOTAVAIL: 2302 /* Source port number is unavailable. Try a new one! */ 2303 transport->srcport = 0; 2304 } 2305 out: 2306 return ret; 2307 } 2308 2309 /** 2310 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2311 * @work: queued work item 2312 * 2313 * Invoked by a work queue tasklet. 2314 */ 2315 static void xs_tcp_setup_socket(struct work_struct *work) 2316 { 2317 struct sock_xprt *transport = 2318 container_of(work, struct sock_xprt, connect_worker.work); 2319 struct socket *sock = transport->sock; 2320 struct rpc_xprt *xprt = &transport->xprt; 2321 int status = -EIO; 2322 2323 if (!sock) { 2324 sock = xs_create_sock(xprt, transport, 2325 xs_addr(xprt)->sa_family, SOCK_STREAM, 2326 IPPROTO_TCP, true); 2327 if (IS_ERR(sock)) { 2328 status = PTR_ERR(sock); 2329 goto out; 2330 } 2331 } 2332 2333 dprintk("RPC: worker connecting xprt %p via %s to " 2334 "%s (port %s)\n", xprt, 2335 xprt->address_strings[RPC_DISPLAY_PROTO], 2336 xprt->address_strings[RPC_DISPLAY_ADDR], 2337 xprt->address_strings[RPC_DISPLAY_PORT]); 2338 2339 status = xs_tcp_finish_connecting(xprt, sock); 2340 trace_rpc_socket_connect(xprt, sock, status); 2341 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2342 xprt, -status, xprt_connected(xprt), 2343 sock->sk->sk_state); 2344 switch (status) { 2345 default: 2346 printk("%s: connect returned unhandled error %d\n", 2347 __func__, status); 2348 /* fall through */ 2349 case -EADDRNOTAVAIL: 2350 /* We're probably in TIME_WAIT. Get rid of existing socket, 2351 * and retry 2352 */ 2353 xs_tcp_force_close(xprt); 2354 break; 2355 case 0: 2356 case -EINPROGRESS: 2357 case -EALREADY: 2358 xprt_unlock_connect(xprt, transport); 2359 return; 2360 case -EINVAL: 2361 /* Happens, for instance, if the user specified a link 2362 * local IPv6 address without a scope-id. 2363 */ 2364 case -ECONNREFUSED: 2365 case -ECONNRESET: 2366 case -ENETDOWN: 2367 case -ENETUNREACH: 2368 case -EHOSTUNREACH: 2369 case -EADDRINUSE: 2370 case -ENOBUFS: 2371 /* 2372 * xs_tcp_force_close() wakes tasks with -EIO. 2373 * We need to wake them first to ensure the 2374 * correct error code. 2375 */ 2376 xprt_wake_pending_tasks(xprt, status); 2377 xs_tcp_force_close(xprt); 2378 goto out; 2379 } 2380 status = -EAGAIN; 2381 out: 2382 xprt_clear_connecting(xprt); 2383 xprt_unlock_connect(xprt, transport); 2384 xprt_wake_pending_tasks(xprt, status); 2385 } 2386 2387 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt) 2388 { 2389 unsigned long start, now = jiffies; 2390 2391 start = xprt->stat.connect_start + xprt->reestablish_timeout; 2392 if (time_after(start, now)) 2393 return start - now; 2394 return 0; 2395 } 2396 2397 static void xs_reconnect_backoff(struct rpc_xprt *xprt) 2398 { 2399 xprt->reestablish_timeout <<= 1; 2400 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout) 2401 xprt->reestablish_timeout = xprt->max_reconnect_timeout; 2402 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2403 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2404 } 2405 2406 /** 2407 * xs_connect - connect a socket to a remote endpoint 2408 * @xprt: pointer to transport structure 2409 * @task: address of RPC task that manages state of connect request 2410 * 2411 * TCP: If the remote end dropped the connection, delay reconnecting. 2412 * 2413 * UDP socket connects are synchronous, but we use a work queue anyway 2414 * to guarantee that even unprivileged user processes can set up a 2415 * socket on a privileged port. 2416 * 2417 * If a UDP socket connect fails, the delay behavior here prevents 2418 * retry floods (hard mounts). 2419 */ 2420 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2421 { 2422 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2423 unsigned long delay = 0; 2424 2425 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2426 2427 if (transport->sock != NULL) { 2428 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2429 "seconds\n", 2430 xprt, xprt->reestablish_timeout / HZ); 2431 2432 /* Start by resetting any existing state */ 2433 xs_reset_transport(transport); 2434 2435 delay = xs_reconnect_delay(xprt); 2436 xs_reconnect_backoff(xprt); 2437 2438 } else 2439 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2440 2441 queue_delayed_work(xprtiod_workqueue, 2442 &transport->connect_worker, 2443 delay); 2444 } 2445 2446 /** 2447 * xs_local_print_stats - display AF_LOCAL socket-specifc stats 2448 * @xprt: rpc_xprt struct containing statistics 2449 * @seq: output file 2450 * 2451 */ 2452 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2453 { 2454 long idle_time = 0; 2455 2456 if (xprt_connected(xprt)) 2457 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2458 2459 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2460 "%llu %llu %lu %llu %llu\n", 2461 xprt->stat.bind_count, 2462 xprt->stat.connect_count, 2463 xprt->stat.connect_time / HZ, 2464 idle_time, 2465 xprt->stat.sends, 2466 xprt->stat.recvs, 2467 xprt->stat.bad_xids, 2468 xprt->stat.req_u, 2469 xprt->stat.bklog_u, 2470 xprt->stat.max_slots, 2471 xprt->stat.sending_u, 2472 xprt->stat.pending_u); 2473 } 2474 2475 /** 2476 * xs_udp_print_stats - display UDP socket-specifc stats 2477 * @xprt: rpc_xprt struct containing statistics 2478 * @seq: output file 2479 * 2480 */ 2481 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2482 { 2483 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2484 2485 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2486 "%lu %llu %llu\n", 2487 transport->srcport, 2488 xprt->stat.bind_count, 2489 xprt->stat.sends, 2490 xprt->stat.recvs, 2491 xprt->stat.bad_xids, 2492 xprt->stat.req_u, 2493 xprt->stat.bklog_u, 2494 xprt->stat.max_slots, 2495 xprt->stat.sending_u, 2496 xprt->stat.pending_u); 2497 } 2498 2499 /** 2500 * xs_tcp_print_stats - display TCP socket-specifc stats 2501 * @xprt: rpc_xprt struct containing statistics 2502 * @seq: output file 2503 * 2504 */ 2505 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2506 { 2507 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2508 long idle_time = 0; 2509 2510 if (xprt_connected(xprt)) 2511 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2512 2513 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2514 "%llu %llu %lu %llu %llu\n", 2515 transport->srcport, 2516 xprt->stat.bind_count, 2517 xprt->stat.connect_count, 2518 xprt->stat.connect_time / HZ, 2519 idle_time, 2520 xprt->stat.sends, 2521 xprt->stat.recvs, 2522 xprt->stat.bad_xids, 2523 xprt->stat.req_u, 2524 xprt->stat.bklog_u, 2525 xprt->stat.max_slots, 2526 xprt->stat.sending_u, 2527 xprt->stat.pending_u); 2528 } 2529 2530 /* 2531 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2532 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2533 * to use the server side send routines. 2534 */ 2535 static int bc_malloc(struct rpc_task *task) 2536 { 2537 struct rpc_rqst *rqst = task->tk_rqstp; 2538 size_t size = rqst->rq_callsize; 2539 struct page *page; 2540 struct rpc_buffer *buf; 2541 2542 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) { 2543 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n", 2544 size); 2545 return -EINVAL; 2546 } 2547 2548 page = alloc_page(GFP_KERNEL); 2549 if (!page) 2550 return -ENOMEM; 2551 2552 buf = page_address(page); 2553 buf->len = PAGE_SIZE; 2554 2555 rqst->rq_buffer = buf->data; 2556 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; 2557 return 0; 2558 } 2559 2560 /* 2561 * Free the space allocated in the bc_alloc routine 2562 */ 2563 static void bc_free(struct rpc_task *task) 2564 { 2565 void *buffer = task->tk_rqstp->rq_buffer; 2566 struct rpc_buffer *buf; 2567 2568 buf = container_of(buffer, struct rpc_buffer, data); 2569 free_page((unsigned long)buf); 2570 } 2571 2572 /* 2573 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex 2574 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request. 2575 */ 2576 static int bc_sendto(struct rpc_rqst *req) 2577 { 2578 int len; 2579 struct xdr_buf *xbufp = &req->rq_snd_buf; 2580 struct sock_xprt *transport = 2581 container_of(req->rq_xprt, struct sock_xprt, xprt); 2582 unsigned long headoff; 2583 unsigned long tailoff; 2584 struct page *tailpage; 2585 struct msghdr msg = { 2586 .msg_flags = MSG_MORE 2587 }; 2588 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 2589 (u32)xbufp->len); 2590 struct kvec iov = { 2591 .iov_base = &marker, 2592 .iov_len = sizeof(marker), 2593 }; 2594 2595 len = kernel_sendmsg(transport->sock, &msg, &iov, 1, iov.iov_len); 2596 if (len != iov.iov_len) 2597 return -EAGAIN; 2598 2599 tailpage = NULL; 2600 if (xbufp->tail[0].iov_len) 2601 tailpage = virt_to_page(xbufp->tail[0].iov_base); 2602 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK; 2603 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK; 2604 len = svc_send_common(transport->sock, xbufp, 2605 virt_to_page(xbufp->head[0].iov_base), headoff, 2606 tailpage, tailoff); 2607 if (len != xbufp->len) 2608 return -EAGAIN; 2609 return len; 2610 } 2611 2612 /* 2613 * The send routine. Borrows from svc_send 2614 */ 2615 static int bc_send_request(struct rpc_rqst *req) 2616 { 2617 struct svc_xprt *xprt; 2618 int len; 2619 2620 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid)); 2621 /* 2622 * Get the server socket associated with this callback xprt 2623 */ 2624 xprt = req->rq_xprt->bc_xprt; 2625 2626 /* 2627 * Grab the mutex to serialize data as the connection is shared 2628 * with the fore channel 2629 */ 2630 mutex_lock(&xprt->xpt_mutex); 2631 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2632 len = -ENOTCONN; 2633 else 2634 len = bc_sendto(req); 2635 mutex_unlock(&xprt->xpt_mutex); 2636 2637 if (len > 0) 2638 len = 0; 2639 2640 return len; 2641 } 2642 2643 /* 2644 * The close routine. Since this is client initiated, we do nothing 2645 */ 2646 2647 static void bc_close(struct rpc_xprt *xprt) 2648 { 2649 } 2650 2651 /* 2652 * The xprt destroy routine. Again, because this connection is client 2653 * initiated, we do nothing 2654 */ 2655 2656 static void bc_destroy(struct rpc_xprt *xprt) 2657 { 2658 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2659 2660 xs_xprt_free(xprt); 2661 module_put(THIS_MODULE); 2662 } 2663 2664 static const struct rpc_xprt_ops xs_local_ops = { 2665 .reserve_xprt = xprt_reserve_xprt, 2666 .release_xprt = xprt_release_xprt, 2667 .alloc_slot = xprt_alloc_slot, 2668 .free_slot = xprt_free_slot, 2669 .rpcbind = xs_local_rpcbind, 2670 .set_port = xs_local_set_port, 2671 .connect = xs_local_connect, 2672 .buf_alloc = rpc_malloc, 2673 .buf_free = rpc_free, 2674 .prepare_request = xs_stream_prepare_request, 2675 .send_request = xs_local_send_request, 2676 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2677 .close = xs_close, 2678 .destroy = xs_destroy, 2679 .print_stats = xs_local_print_stats, 2680 .enable_swap = xs_enable_swap, 2681 .disable_swap = xs_disable_swap, 2682 }; 2683 2684 static const struct rpc_xprt_ops xs_udp_ops = { 2685 .set_buffer_size = xs_udp_set_buffer_size, 2686 .reserve_xprt = xprt_reserve_xprt_cong, 2687 .release_xprt = xprt_release_xprt_cong, 2688 .alloc_slot = xprt_alloc_slot, 2689 .free_slot = xprt_free_slot, 2690 .rpcbind = rpcb_getport_async, 2691 .set_port = xs_set_port, 2692 .connect = xs_connect, 2693 .buf_alloc = rpc_malloc, 2694 .buf_free = rpc_free, 2695 .send_request = xs_udp_send_request, 2696 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 2697 .timer = xs_udp_timer, 2698 .release_request = xprt_release_rqst_cong, 2699 .close = xs_close, 2700 .destroy = xs_destroy, 2701 .print_stats = xs_udp_print_stats, 2702 .enable_swap = xs_enable_swap, 2703 .disable_swap = xs_disable_swap, 2704 .inject_disconnect = xs_inject_disconnect, 2705 }; 2706 2707 static const struct rpc_xprt_ops xs_tcp_ops = { 2708 .reserve_xprt = xprt_reserve_xprt, 2709 .release_xprt = xprt_release_xprt, 2710 .alloc_slot = xprt_alloc_slot, 2711 .free_slot = xprt_free_slot, 2712 .rpcbind = rpcb_getport_async, 2713 .set_port = xs_set_port, 2714 .connect = xs_connect, 2715 .buf_alloc = rpc_malloc, 2716 .buf_free = rpc_free, 2717 .prepare_request = xs_stream_prepare_request, 2718 .send_request = xs_tcp_send_request, 2719 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2720 .close = xs_tcp_shutdown, 2721 .destroy = xs_destroy, 2722 .set_connect_timeout = xs_tcp_set_connect_timeout, 2723 .print_stats = xs_tcp_print_stats, 2724 .enable_swap = xs_enable_swap, 2725 .disable_swap = xs_disable_swap, 2726 .inject_disconnect = xs_inject_disconnect, 2727 #ifdef CONFIG_SUNRPC_BACKCHANNEL 2728 .bc_setup = xprt_setup_bc, 2729 .bc_maxpayload = xs_tcp_bc_maxpayload, 2730 .bc_free_rqst = xprt_free_bc_rqst, 2731 .bc_destroy = xprt_destroy_bc, 2732 #endif 2733 }; 2734 2735 /* 2736 * The rpc_xprt_ops for the server backchannel 2737 */ 2738 2739 static const struct rpc_xprt_ops bc_tcp_ops = { 2740 .reserve_xprt = xprt_reserve_xprt, 2741 .release_xprt = xprt_release_xprt, 2742 .alloc_slot = xprt_alloc_slot, 2743 .free_slot = xprt_free_slot, 2744 .buf_alloc = bc_malloc, 2745 .buf_free = bc_free, 2746 .send_request = bc_send_request, 2747 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2748 .close = bc_close, 2749 .destroy = bc_destroy, 2750 .print_stats = xs_tcp_print_stats, 2751 .enable_swap = xs_enable_swap, 2752 .disable_swap = xs_disable_swap, 2753 .inject_disconnect = xs_inject_disconnect, 2754 }; 2755 2756 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2757 { 2758 static const struct sockaddr_in sin = { 2759 .sin_family = AF_INET, 2760 .sin_addr.s_addr = htonl(INADDR_ANY), 2761 }; 2762 static const struct sockaddr_in6 sin6 = { 2763 .sin6_family = AF_INET6, 2764 .sin6_addr = IN6ADDR_ANY_INIT, 2765 }; 2766 2767 switch (family) { 2768 case AF_LOCAL: 2769 break; 2770 case AF_INET: 2771 memcpy(sap, &sin, sizeof(sin)); 2772 break; 2773 case AF_INET6: 2774 memcpy(sap, &sin6, sizeof(sin6)); 2775 break; 2776 default: 2777 dprintk("RPC: %s: Bad address family\n", __func__); 2778 return -EAFNOSUPPORT; 2779 } 2780 return 0; 2781 } 2782 2783 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2784 unsigned int slot_table_size, 2785 unsigned int max_slot_table_size) 2786 { 2787 struct rpc_xprt *xprt; 2788 struct sock_xprt *new; 2789 2790 if (args->addrlen > sizeof(xprt->addr)) { 2791 dprintk("RPC: xs_setup_xprt: address too large\n"); 2792 return ERR_PTR(-EBADF); 2793 } 2794 2795 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2796 max_slot_table_size); 2797 if (xprt == NULL) { 2798 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2799 "rpc_xprt\n"); 2800 return ERR_PTR(-ENOMEM); 2801 } 2802 2803 new = container_of(xprt, struct sock_xprt, xprt); 2804 mutex_init(&new->recv_mutex); 2805 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2806 xprt->addrlen = args->addrlen; 2807 if (args->srcaddr) 2808 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2809 else { 2810 int err; 2811 err = xs_init_anyaddr(args->dstaddr->sa_family, 2812 (struct sockaddr *)&new->srcaddr); 2813 if (err != 0) { 2814 xprt_free(xprt); 2815 return ERR_PTR(err); 2816 } 2817 } 2818 2819 return xprt; 2820 } 2821 2822 static const struct rpc_timeout xs_local_default_timeout = { 2823 .to_initval = 10 * HZ, 2824 .to_maxval = 10 * HZ, 2825 .to_retries = 2, 2826 }; 2827 2828 /** 2829 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2830 * @args: rpc transport creation arguments 2831 * 2832 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2833 */ 2834 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2835 { 2836 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2837 struct sock_xprt *transport; 2838 struct rpc_xprt *xprt; 2839 struct rpc_xprt *ret; 2840 2841 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2842 xprt_max_tcp_slot_table_entries); 2843 if (IS_ERR(xprt)) 2844 return xprt; 2845 transport = container_of(xprt, struct sock_xprt, xprt); 2846 2847 xprt->prot = 0; 2848 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2849 2850 xprt->bind_timeout = XS_BIND_TO; 2851 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2852 xprt->idle_timeout = XS_IDLE_DISC_TO; 2853 2854 xprt->ops = &xs_local_ops; 2855 xprt->timeout = &xs_local_default_timeout; 2856 2857 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 2858 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket); 2859 2860 switch (sun->sun_family) { 2861 case AF_LOCAL: 2862 if (sun->sun_path[0] != '/') { 2863 dprintk("RPC: bad AF_LOCAL address: %s\n", 2864 sun->sun_path); 2865 ret = ERR_PTR(-EINVAL); 2866 goto out_err; 2867 } 2868 xprt_set_bound(xprt); 2869 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2870 ret = ERR_PTR(xs_local_setup_socket(transport)); 2871 if (ret) 2872 goto out_err; 2873 break; 2874 default: 2875 ret = ERR_PTR(-EAFNOSUPPORT); 2876 goto out_err; 2877 } 2878 2879 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2880 xprt->address_strings[RPC_DISPLAY_ADDR]); 2881 2882 if (try_module_get(THIS_MODULE)) 2883 return xprt; 2884 ret = ERR_PTR(-EINVAL); 2885 out_err: 2886 xs_xprt_free(xprt); 2887 return ret; 2888 } 2889 2890 static const struct rpc_timeout xs_udp_default_timeout = { 2891 .to_initval = 5 * HZ, 2892 .to_maxval = 30 * HZ, 2893 .to_increment = 5 * HZ, 2894 .to_retries = 5, 2895 }; 2896 2897 /** 2898 * xs_setup_udp - Set up transport to use a UDP socket 2899 * @args: rpc transport creation arguments 2900 * 2901 */ 2902 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2903 { 2904 struct sockaddr *addr = args->dstaddr; 2905 struct rpc_xprt *xprt; 2906 struct sock_xprt *transport; 2907 struct rpc_xprt *ret; 2908 2909 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2910 xprt_udp_slot_table_entries); 2911 if (IS_ERR(xprt)) 2912 return xprt; 2913 transport = container_of(xprt, struct sock_xprt, xprt); 2914 2915 xprt->prot = IPPROTO_UDP; 2916 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2917 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2918 2919 xprt->bind_timeout = XS_BIND_TO; 2920 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2921 xprt->idle_timeout = XS_IDLE_DISC_TO; 2922 2923 xprt->ops = &xs_udp_ops; 2924 2925 xprt->timeout = &xs_udp_default_timeout; 2926 2927 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 2928 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 2929 2930 switch (addr->sa_family) { 2931 case AF_INET: 2932 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2933 xprt_set_bound(xprt); 2934 2935 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2936 break; 2937 case AF_INET6: 2938 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2939 xprt_set_bound(xprt); 2940 2941 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2942 break; 2943 default: 2944 ret = ERR_PTR(-EAFNOSUPPORT); 2945 goto out_err; 2946 } 2947 2948 if (xprt_bound(xprt)) 2949 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2950 xprt->address_strings[RPC_DISPLAY_ADDR], 2951 xprt->address_strings[RPC_DISPLAY_PORT], 2952 xprt->address_strings[RPC_DISPLAY_PROTO]); 2953 else 2954 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2955 xprt->address_strings[RPC_DISPLAY_ADDR], 2956 xprt->address_strings[RPC_DISPLAY_PROTO]); 2957 2958 if (try_module_get(THIS_MODULE)) 2959 return xprt; 2960 ret = ERR_PTR(-EINVAL); 2961 out_err: 2962 xs_xprt_free(xprt); 2963 return ret; 2964 } 2965 2966 static const struct rpc_timeout xs_tcp_default_timeout = { 2967 .to_initval = 60 * HZ, 2968 .to_maxval = 60 * HZ, 2969 .to_retries = 2, 2970 }; 2971 2972 /** 2973 * xs_setup_tcp - Set up transport to use a TCP socket 2974 * @args: rpc transport creation arguments 2975 * 2976 */ 2977 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2978 { 2979 struct sockaddr *addr = args->dstaddr; 2980 struct rpc_xprt *xprt; 2981 struct sock_xprt *transport; 2982 struct rpc_xprt *ret; 2983 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2984 2985 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2986 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2987 2988 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2989 max_slot_table_size); 2990 if (IS_ERR(xprt)) 2991 return xprt; 2992 transport = container_of(xprt, struct sock_xprt, xprt); 2993 2994 xprt->prot = IPPROTO_TCP; 2995 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2996 2997 xprt->bind_timeout = XS_BIND_TO; 2998 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2999 xprt->idle_timeout = XS_IDLE_DISC_TO; 3000 3001 xprt->ops = &xs_tcp_ops; 3002 xprt->timeout = &xs_tcp_default_timeout; 3003 3004 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 3005 xprt->connect_timeout = xprt->timeout->to_initval * 3006 (xprt->timeout->to_retries + 1); 3007 3008 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 3009 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 3010 3011 switch (addr->sa_family) { 3012 case AF_INET: 3013 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3014 xprt_set_bound(xprt); 3015 3016 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 3017 break; 3018 case AF_INET6: 3019 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3020 xprt_set_bound(xprt); 3021 3022 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 3023 break; 3024 default: 3025 ret = ERR_PTR(-EAFNOSUPPORT); 3026 goto out_err; 3027 } 3028 3029 if (xprt_bound(xprt)) 3030 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3031 xprt->address_strings[RPC_DISPLAY_ADDR], 3032 xprt->address_strings[RPC_DISPLAY_PORT], 3033 xprt->address_strings[RPC_DISPLAY_PROTO]); 3034 else 3035 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3036 xprt->address_strings[RPC_DISPLAY_ADDR], 3037 xprt->address_strings[RPC_DISPLAY_PROTO]); 3038 3039 if (try_module_get(THIS_MODULE)) 3040 return xprt; 3041 ret = ERR_PTR(-EINVAL); 3042 out_err: 3043 xs_xprt_free(xprt); 3044 return ret; 3045 } 3046 3047 /** 3048 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 3049 * @args: rpc transport creation arguments 3050 * 3051 */ 3052 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 3053 { 3054 struct sockaddr *addr = args->dstaddr; 3055 struct rpc_xprt *xprt; 3056 struct sock_xprt *transport; 3057 struct svc_sock *bc_sock; 3058 struct rpc_xprt *ret; 3059 3060 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3061 xprt_tcp_slot_table_entries); 3062 if (IS_ERR(xprt)) 3063 return xprt; 3064 transport = container_of(xprt, struct sock_xprt, xprt); 3065 3066 xprt->prot = IPPROTO_TCP; 3067 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3068 xprt->timeout = &xs_tcp_default_timeout; 3069 3070 /* backchannel */ 3071 xprt_set_bound(xprt); 3072 xprt->bind_timeout = 0; 3073 xprt->reestablish_timeout = 0; 3074 xprt->idle_timeout = 0; 3075 3076 xprt->ops = &bc_tcp_ops; 3077 3078 switch (addr->sa_family) { 3079 case AF_INET: 3080 xs_format_peer_addresses(xprt, "tcp", 3081 RPCBIND_NETID_TCP); 3082 break; 3083 case AF_INET6: 3084 xs_format_peer_addresses(xprt, "tcp", 3085 RPCBIND_NETID_TCP6); 3086 break; 3087 default: 3088 ret = ERR_PTR(-EAFNOSUPPORT); 3089 goto out_err; 3090 } 3091 3092 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3093 xprt->address_strings[RPC_DISPLAY_ADDR], 3094 xprt->address_strings[RPC_DISPLAY_PORT], 3095 xprt->address_strings[RPC_DISPLAY_PROTO]); 3096 3097 /* 3098 * Once we've associated a backchannel xprt with a connection, 3099 * we want to keep it around as long as the connection lasts, 3100 * in case we need to start using it for a backchannel again; 3101 * this reference won't be dropped until bc_xprt is destroyed. 3102 */ 3103 xprt_get(xprt); 3104 args->bc_xprt->xpt_bc_xprt = xprt; 3105 xprt->bc_xprt = args->bc_xprt; 3106 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3107 transport->sock = bc_sock->sk_sock; 3108 transport->inet = bc_sock->sk_sk; 3109 3110 /* 3111 * Since we don't want connections for the backchannel, we set 3112 * the xprt status to connected 3113 */ 3114 xprt_set_connected(xprt); 3115 3116 if (try_module_get(THIS_MODULE)) 3117 return xprt; 3118 3119 args->bc_xprt->xpt_bc_xprt = NULL; 3120 args->bc_xprt->xpt_bc_xps = NULL; 3121 xprt_put(xprt); 3122 ret = ERR_PTR(-EINVAL); 3123 out_err: 3124 xs_xprt_free(xprt); 3125 return ret; 3126 } 3127 3128 static struct xprt_class xs_local_transport = { 3129 .list = LIST_HEAD_INIT(xs_local_transport.list), 3130 .name = "named UNIX socket", 3131 .owner = THIS_MODULE, 3132 .ident = XPRT_TRANSPORT_LOCAL, 3133 .setup = xs_setup_local, 3134 }; 3135 3136 static struct xprt_class xs_udp_transport = { 3137 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3138 .name = "udp", 3139 .owner = THIS_MODULE, 3140 .ident = XPRT_TRANSPORT_UDP, 3141 .setup = xs_setup_udp, 3142 }; 3143 3144 static struct xprt_class xs_tcp_transport = { 3145 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3146 .name = "tcp", 3147 .owner = THIS_MODULE, 3148 .ident = XPRT_TRANSPORT_TCP, 3149 .setup = xs_setup_tcp, 3150 }; 3151 3152 static struct xprt_class xs_bc_tcp_transport = { 3153 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3154 .name = "tcp NFSv4.1 backchannel", 3155 .owner = THIS_MODULE, 3156 .ident = XPRT_TRANSPORT_BC_TCP, 3157 .setup = xs_setup_bc_tcp, 3158 }; 3159 3160 /** 3161 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3162 * 3163 */ 3164 int init_socket_xprt(void) 3165 { 3166 if (!sunrpc_table_header) 3167 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3168 3169 xprt_register_transport(&xs_local_transport); 3170 xprt_register_transport(&xs_udp_transport); 3171 xprt_register_transport(&xs_tcp_transport); 3172 xprt_register_transport(&xs_bc_tcp_transport); 3173 3174 return 0; 3175 } 3176 3177 /** 3178 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3179 * 3180 */ 3181 void cleanup_socket_xprt(void) 3182 { 3183 if (sunrpc_table_header) { 3184 unregister_sysctl_table(sunrpc_table_header); 3185 sunrpc_table_header = NULL; 3186 } 3187 3188 xprt_unregister_transport(&xs_local_transport); 3189 xprt_unregister_transport(&xs_udp_transport); 3190 xprt_unregister_transport(&xs_tcp_transport); 3191 xprt_unregister_transport(&xs_bc_tcp_transport); 3192 } 3193 3194 static int param_set_uint_minmax(const char *val, 3195 const struct kernel_param *kp, 3196 unsigned int min, unsigned int max) 3197 { 3198 unsigned int num; 3199 int ret; 3200 3201 if (!val) 3202 return -EINVAL; 3203 ret = kstrtouint(val, 0, &num); 3204 if (ret) 3205 return ret; 3206 if (num < min || num > max) 3207 return -EINVAL; 3208 *((unsigned int *)kp->arg) = num; 3209 return 0; 3210 } 3211 3212 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3213 { 3214 return param_set_uint_minmax(val, kp, 3215 RPC_MIN_RESVPORT, 3216 RPC_MAX_RESVPORT); 3217 } 3218 3219 static const struct kernel_param_ops param_ops_portnr = { 3220 .set = param_set_portnr, 3221 .get = param_get_uint, 3222 }; 3223 3224 #define param_check_portnr(name, p) \ 3225 __param_check(name, p, unsigned int); 3226 3227 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3228 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3229 3230 static int param_set_slot_table_size(const char *val, 3231 const struct kernel_param *kp) 3232 { 3233 return param_set_uint_minmax(val, kp, 3234 RPC_MIN_SLOT_TABLE, 3235 RPC_MAX_SLOT_TABLE); 3236 } 3237 3238 static const struct kernel_param_ops param_ops_slot_table_size = { 3239 .set = param_set_slot_table_size, 3240 .get = param_get_uint, 3241 }; 3242 3243 #define param_check_slot_table_size(name, p) \ 3244 __param_check(name, p, unsigned int); 3245 3246 static int param_set_max_slot_table_size(const char *val, 3247 const struct kernel_param *kp) 3248 { 3249 return param_set_uint_minmax(val, kp, 3250 RPC_MIN_SLOT_TABLE, 3251 RPC_MAX_SLOT_TABLE_LIMIT); 3252 } 3253 3254 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3255 .set = param_set_max_slot_table_size, 3256 .get = param_get_uint, 3257 }; 3258 3259 #define param_check_max_slot_table_size(name, p) \ 3260 __param_check(name, p, unsigned int); 3261 3262 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3263 slot_table_size, 0644); 3264 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3265 max_slot_table_size, 0644); 3266 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3267 slot_table_size, 0644); 3268