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