1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/net/sunrpc/xprtsock.c 4 * 5 * Client-side transport implementation for sockets. 6 * 7 * TCP callback races fixes (C) 1998 Red Hat 8 * TCP send fixes (C) 1998 Red Hat 9 * TCP NFS related read + write fixes 10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 11 * 12 * Rewrite of larges part of the code in order to stabilize TCP stuff. 13 * Fix behaviour when socket buffer is full. 14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> 15 * 16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> 17 * 18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005. 19 * <gilles.quillard@bull.net> 20 */ 21 22 #include <linux/types.h> 23 #include <linux/string.h> 24 #include <linux/slab.h> 25 #include <linux/module.h> 26 #include <linux/capability.h> 27 #include <linux/pagemap.h> 28 #include <linux/errno.h> 29 #include <linux/socket.h> 30 #include <linux/in.h> 31 #include <linux/net.h> 32 #include <linux/mm.h> 33 #include <linux/un.h> 34 #include <linux/udp.h> 35 #include <linux/tcp.h> 36 #include <linux/sunrpc/clnt.h> 37 #include <linux/sunrpc/addr.h> 38 #include <linux/sunrpc/sched.h> 39 #include <linux/sunrpc/svcsock.h> 40 #include <linux/sunrpc/xprtsock.h> 41 #include <linux/file.h> 42 #ifdef CONFIG_SUNRPC_BACKCHANNEL 43 #include <linux/sunrpc/bc_xprt.h> 44 #endif 45 46 #include <net/sock.h> 47 #include <net/checksum.h> 48 #include <net/udp.h> 49 #include <net/tcp.h> 50 #include <linux/bvec.h> 51 #include <linux/highmem.h> 52 #include <linux/uio.h> 53 #include <linux/sched/mm.h> 54 55 #include <trace/events/sunrpc.h> 56 57 #include "socklib.h" 58 #include "sunrpc.h" 59 60 static void xs_close(struct rpc_xprt *xprt); 61 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 62 struct socket *sock); 63 64 /* 65 * xprtsock tunables 66 */ 67 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 68 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 69 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 70 71 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 72 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 73 74 #define XS_TCP_LINGER_TO (15U * HZ) 75 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 76 77 /* 78 * We can register our own files under /proc/sys/sunrpc by 79 * calling register_sysctl_table() again. The files in that 80 * directory become the union of all files registered there. 81 * 82 * We simply need to make sure that we don't collide with 83 * someone else's file names! 84 */ 85 86 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 87 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 88 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 89 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 90 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 91 92 static struct ctl_table_header *sunrpc_table_header; 93 94 /* 95 * FIXME: changing the UDP slot table size should also resize the UDP 96 * socket buffers for existing UDP transports 97 */ 98 static struct ctl_table xs_tunables_table[] = { 99 { 100 .procname = "udp_slot_table_entries", 101 .data = &xprt_udp_slot_table_entries, 102 .maxlen = sizeof(unsigned int), 103 .mode = 0644, 104 .proc_handler = proc_dointvec_minmax, 105 .extra1 = &min_slot_table_size, 106 .extra2 = &max_slot_table_size 107 }, 108 { 109 .procname = "tcp_slot_table_entries", 110 .data = &xprt_tcp_slot_table_entries, 111 .maxlen = sizeof(unsigned int), 112 .mode = 0644, 113 .proc_handler = proc_dointvec_minmax, 114 .extra1 = &min_slot_table_size, 115 .extra2 = &max_slot_table_size 116 }, 117 { 118 .procname = "tcp_max_slot_table_entries", 119 .data = &xprt_max_tcp_slot_table_entries, 120 .maxlen = sizeof(unsigned int), 121 .mode = 0644, 122 .proc_handler = proc_dointvec_minmax, 123 .extra1 = &min_slot_table_size, 124 .extra2 = &max_tcp_slot_table_limit 125 }, 126 { 127 .procname = "min_resvport", 128 .data = &xprt_min_resvport, 129 .maxlen = sizeof(unsigned int), 130 .mode = 0644, 131 .proc_handler = proc_dointvec_minmax, 132 .extra1 = &xprt_min_resvport_limit, 133 .extra2 = &xprt_max_resvport_limit 134 }, 135 { 136 .procname = "max_resvport", 137 .data = &xprt_max_resvport, 138 .maxlen = sizeof(unsigned int), 139 .mode = 0644, 140 .proc_handler = proc_dointvec_minmax, 141 .extra1 = &xprt_min_resvport_limit, 142 .extra2 = &xprt_max_resvport_limit 143 }, 144 { 145 .procname = "tcp_fin_timeout", 146 .data = &xs_tcp_fin_timeout, 147 .maxlen = sizeof(xs_tcp_fin_timeout), 148 .mode = 0644, 149 .proc_handler = proc_dointvec_jiffies, 150 }, 151 { }, 152 }; 153 154 static struct ctl_table sunrpc_table[] = { 155 { 156 .procname = "sunrpc", 157 .mode = 0555, 158 .child = xs_tunables_table 159 }, 160 { }, 161 }; 162 163 /* 164 * Wait duration for a reply from the RPC portmapper. 165 */ 166 #define XS_BIND_TO (60U * HZ) 167 168 /* 169 * Delay if a UDP socket connect error occurs. This is most likely some 170 * kind of resource problem on the local host. 171 */ 172 #define XS_UDP_REEST_TO (2U * HZ) 173 174 /* 175 * The reestablish timeout allows clients to delay for a bit before attempting 176 * to reconnect to a server that just dropped our connection. 177 * 178 * We implement an exponential backoff when trying to reestablish a TCP 179 * transport connection with the server. Some servers like to drop a TCP 180 * connection when they are overworked, so we start with a short timeout and 181 * increase over time if the server is down or not responding. 182 */ 183 #define XS_TCP_INIT_REEST_TO (3U * HZ) 184 185 /* 186 * TCP idle timeout; client drops the transport socket if it is idle 187 * for this long. Note that we also timeout UDP sockets to prevent 188 * holding port numbers when there is no RPC traffic. 189 */ 190 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 191 192 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 193 # undef RPC_DEBUG_DATA 194 # define RPCDBG_FACILITY RPCDBG_TRANS 195 #endif 196 197 #ifdef RPC_DEBUG_DATA 198 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 199 { 200 u8 *buf = (u8 *) packet; 201 int j; 202 203 dprintk("RPC: %s\n", msg); 204 for (j = 0; j < count && j < 128; j += 4) { 205 if (!(j & 31)) { 206 if (j) 207 dprintk("\n"); 208 dprintk("0x%04x ", j); 209 } 210 dprintk("%02x%02x%02x%02x ", 211 buf[j], buf[j+1], buf[j+2], buf[j+3]); 212 } 213 dprintk("\n"); 214 } 215 #else 216 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 217 { 218 /* NOP */ 219 } 220 #endif 221 222 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 223 { 224 return (struct rpc_xprt *) sk->sk_user_data; 225 } 226 227 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 228 { 229 return (struct sockaddr *) &xprt->addr; 230 } 231 232 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 233 { 234 return (struct sockaddr_un *) &xprt->addr; 235 } 236 237 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 238 { 239 return (struct sockaddr_in *) &xprt->addr; 240 } 241 242 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 243 { 244 return (struct sockaddr_in6 *) &xprt->addr; 245 } 246 247 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 248 { 249 struct sockaddr *sap = xs_addr(xprt); 250 struct sockaddr_in6 *sin6; 251 struct sockaddr_in *sin; 252 struct sockaddr_un *sun; 253 char buf[128]; 254 255 switch (sap->sa_family) { 256 case AF_LOCAL: 257 sun = xs_addr_un(xprt); 258 strlcpy(buf, sun->sun_path, sizeof(buf)); 259 xprt->address_strings[RPC_DISPLAY_ADDR] = 260 kstrdup(buf, GFP_KERNEL); 261 break; 262 case AF_INET: 263 (void)rpc_ntop(sap, buf, sizeof(buf)); 264 xprt->address_strings[RPC_DISPLAY_ADDR] = 265 kstrdup(buf, GFP_KERNEL); 266 sin = xs_addr_in(xprt); 267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 268 break; 269 case AF_INET6: 270 (void)rpc_ntop(sap, buf, sizeof(buf)); 271 xprt->address_strings[RPC_DISPLAY_ADDR] = 272 kstrdup(buf, GFP_KERNEL); 273 sin6 = xs_addr_in6(xprt); 274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 275 break; 276 default: 277 BUG(); 278 } 279 280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 281 } 282 283 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 284 { 285 struct sockaddr *sap = xs_addr(xprt); 286 char buf[128]; 287 288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 290 291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 293 } 294 295 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 296 const char *protocol, 297 const char *netid) 298 { 299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 300 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 301 xs_format_common_peer_addresses(xprt); 302 xs_format_common_peer_ports(xprt); 303 } 304 305 static void xs_update_peer_port(struct rpc_xprt *xprt) 306 { 307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 309 310 xs_format_common_peer_ports(xprt); 311 } 312 313 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 314 { 315 unsigned int i; 316 317 for (i = 0; i < RPC_DISPLAY_MAX; i++) 318 switch (i) { 319 case RPC_DISPLAY_PROTO: 320 case RPC_DISPLAY_NETID: 321 continue; 322 default: 323 kfree(xprt->address_strings[i]); 324 } 325 } 326 327 static size_t 328 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp) 329 { 330 size_t i,n; 331 332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES)) 333 return want; 334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT; 335 for (i = 0; i < n; i++) { 336 if (buf->pages[i]) 337 continue; 338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp); 339 if (!buf->pages[i]) { 340 i *= PAGE_SIZE; 341 return i > buf->page_base ? i - buf->page_base : 0; 342 } 343 } 344 return want; 345 } 346 347 static ssize_t 348 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek) 349 { 350 ssize_t ret; 351 if (seek != 0) 352 iov_iter_advance(&msg->msg_iter, seek); 353 ret = sock_recvmsg(sock, msg, flags); 354 return ret > 0 ? ret + seek : ret; 355 } 356 357 static ssize_t 358 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags, 359 struct kvec *kvec, size_t count, size_t seek) 360 { 361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count); 362 return xs_sock_recvmsg(sock, msg, flags, seek); 363 } 364 365 static ssize_t 366 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags, 367 struct bio_vec *bvec, unsigned long nr, size_t count, 368 size_t seek) 369 { 370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count); 371 return xs_sock_recvmsg(sock, msg, flags, seek); 372 } 373 374 static ssize_t 375 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags, 376 size_t count) 377 { 378 iov_iter_discard(&msg->msg_iter, READ, count); 379 return sock_recvmsg(sock, msg, flags); 380 } 381 382 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 383 static void 384 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 385 { 386 struct bvec_iter bi = { 387 .bi_size = count, 388 }; 389 struct bio_vec bv; 390 391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); 392 for_each_bvec(bv, bvec, bi, bi) 393 flush_dcache_page(bv.bv_page); 394 } 395 #else 396 static inline void 397 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 398 { 399 } 400 #endif 401 402 static ssize_t 403 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags, 404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read) 405 { 406 size_t want, seek_init = seek, offset = 0; 407 ssize_t ret; 408 409 want = min_t(size_t, count, buf->head[0].iov_len); 410 if (seek < want) { 411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek); 412 if (ret <= 0) 413 goto sock_err; 414 offset += ret; 415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 416 goto out; 417 if (ret != want) 418 goto out; 419 seek = 0; 420 } else { 421 seek -= want; 422 offset += want; 423 } 424 425 want = xs_alloc_sparse_pages(buf, 426 min_t(size_t, count - offset, buf->page_len), 427 GFP_KERNEL); 428 if (seek < want) { 429 ret = xs_read_bvec(sock, msg, flags, buf->bvec, 430 xdr_buf_pagecount(buf), 431 want + buf->page_base, 432 seek + buf->page_base); 433 if (ret <= 0) 434 goto sock_err; 435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base); 436 offset += ret - buf->page_base; 437 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 438 goto out; 439 if (ret != want) 440 goto out; 441 seek = 0; 442 } else { 443 seek -= want; 444 offset += want; 445 } 446 447 want = min_t(size_t, count - offset, buf->tail[0].iov_len); 448 if (seek < want) { 449 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek); 450 if (ret <= 0) 451 goto sock_err; 452 offset += ret; 453 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 454 goto out; 455 if (ret != want) 456 goto out; 457 } else if (offset < seek_init) 458 offset = seek_init; 459 ret = -EMSGSIZE; 460 out: 461 *read = offset - seek_init; 462 return ret; 463 sock_err: 464 offset += seek; 465 goto out; 466 } 467 468 static void 469 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf) 470 { 471 if (!transport->recv.copied) { 472 if (buf->head[0].iov_len >= transport->recv.offset) 473 memcpy(buf->head[0].iov_base, 474 &transport->recv.xid, 475 transport->recv.offset); 476 transport->recv.copied = transport->recv.offset; 477 } 478 } 479 480 static bool 481 xs_read_stream_request_done(struct sock_xprt *transport) 482 { 483 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT); 484 } 485 486 static void 487 xs_read_stream_check_eor(struct sock_xprt *transport, 488 struct msghdr *msg) 489 { 490 if (xs_read_stream_request_done(transport)) 491 msg->msg_flags |= MSG_EOR; 492 } 493 494 static ssize_t 495 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg, 496 int flags, struct rpc_rqst *req) 497 { 498 struct xdr_buf *buf = &req->rq_private_buf; 499 size_t want, uninitialized_var(read); 500 ssize_t uninitialized_var(ret); 501 502 xs_read_header(transport, buf); 503 504 want = transport->recv.len - transport->recv.offset; 505 if (want != 0) { 506 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf, 507 transport->recv.copied + want, 508 transport->recv.copied, 509 &read); 510 transport->recv.offset += read; 511 transport->recv.copied += read; 512 } 513 514 if (transport->recv.offset == transport->recv.len) 515 xs_read_stream_check_eor(transport, msg); 516 517 if (want == 0) 518 return 0; 519 520 switch (ret) { 521 default: 522 break; 523 case -EFAULT: 524 case -EMSGSIZE: 525 msg->msg_flags |= MSG_TRUNC; 526 return read; 527 case 0: 528 return -ESHUTDOWN; 529 } 530 return ret < 0 ? ret : read; 531 } 532 533 static size_t 534 xs_read_stream_headersize(bool isfrag) 535 { 536 if (isfrag) 537 return sizeof(__be32); 538 return 3 * sizeof(__be32); 539 } 540 541 static ssize_t 542 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg, 543 int flags, size_t want, size_t seek) 544 { 545 struct kvec kvec = { 546 .iov_base = &transport->recv.fraghdr, 547 .iov_len = want, 548 }; 549 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek); 550 } 551 552 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 553 static ssize_t 554 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 555 { 556 struct rpc_xprt *xprt = &transport->xprt; 557 struct rpc_rqst *req; 558 ssize_t ret; 559 560 /* Look up and lock the request corresponding to the given XID */ 561 req = xprt_lookup_bc_request(xprt, transport->recv.xid); 562 if (!req) { 563 printk(KERN_WARNING "Callback slot table overflowed\n"); 564 return -ESHUTDOWN; 565 } 566 if (transport->recv.copied && !req->rq_private_buf.len) 567 return -ESHUTDOWN; 568 569 ret = xs_read_stream_request(transport, msg, flags, req); 570 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 571 xprt_complete_bc_request(req, transport->recv.copied); 572 else 573 req->rq_private_buf.len = transport->recv.copied; 574 575 return ret; 576 } 577 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 578 static ssize_t 579 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 580 { 581 return -ESHUTDOWN; 582 } 583 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 584 585 static ssize_t 586 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags) 587 { 588 struct rpc_xprt *xprt = &transport->xprt; 589 struct rpc_rqst *req; 590 ssize_t ret = 0; 591 592 /* Look up and lock the request corresponding to the given XID */ 593 spin_lock(&xprt->queue_lock); 594 req = xprt_lookup_rqst(xprt, transport->recv.xid); 595 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) { 596 msg->msg_flags |= MSG_TRUNC; 597 goto out; 598 } 599 xprt_pin_rqst(req); 600 spin_unlock(&xprt->queue_lock); 601 602 ret = xs_read_stream_request(transport, msg, flags, req); 603 604 spin_lock(&xprt->queue_lock); 605 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 606 xprt_complete_rqst(req->rq_task, transport->recv.copied); 607 else 608 req->rq_private_buf.len = transport->recv.copied; 609 xprt_unpin_rqst(req); 610 out: 611 spin_unlock(&xprt->queue_lock); 612 return ret; 613 } 614 615 static ssize_t 616 xs_read_stream(struct sock_xprt *transport, int flags) 617 { 618 struct msghdr msg = { 0 }; 619 size_t want, read = 0; 620 ssize_t ret = 0; 621 622 if (transport->recv.len == 0) { 623 want = xs_read_stream_headersize(transport->recv.copied != 0); 624 ret = xs_read_stream_header(transport, &msg, flags, want, 625 transport->recv.offset); 626 if (ret <= 0) 627 goto out_err; 628 transport->recv.offset = ret; 629 if (transport->recv.offset != want) 630 return transport->recv.offset; 631 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) & 632 RPC_FRAGMENT_SIZE_MASK; 633 transport->recv.offset -= sizeof(transport->recv.fraghdr); 634 read = ret; 635 } 636 637 switch (be32_to_cpu(transport->recv.calldir)) { 638 default: 639 msg.msg_flags |= MSG_TRUNC; 640 break; 641 case RPC_CALL: 642 ret = xs_read_stream_call(transport, &msg, flags); 643 break; 644 case RPC_REPLY: 645 ret = xs_read_stream_reply(transport, &msg, flags); 646 } 647 if (msg.msg_flags & MSG_TRUNC) { 648 transport->recv.calldir = cpu_to_be32(-1); 649 transport->recv.copied = -1; 650 } 651 if (ret < 0) 652 goto out_err; 653 read += ret; 654 if (transport->recv.offset < transport->recv.len) { 655 if (!(msg.msg_flags & MSG_TRUNC)) 656 return read; 657 msg.msg_flags = 0; 658 ret = xs_read_discard(transport->sock, &msg, flags, 659 transport->recv.len - transport->recv.offset); 660 if (ret <= 0) 661 goto out_err; 662 transport->recv.offset += ret; 663 read += ret; 664 if (transport->recv.offset != transport->recv.len) 665 return read; 666 } 667 if (xs_read_stream_request_done(transport)) { 668 trace_xs_stream_read_request(transport); 669 transport->recv.copied = 0; 670 } 671 transport->recv.offset = 0; 672 transport->recv.len = 0; 673 return read; 674 out_err: 675 return ret != 0 ? ret : -ESHUTDOWN; 676 } 677 678 static __poll_t xs_poll_socket(struct sock_xprt *transport) 679 { 680 return transport->sock->ops->poll(transport->file, transport->sock, 681 NULL); 682 } 683 684 static bool xs_poll_socket_readable(struct sock_xprt *transport) 685 { 686 __poll_t events = xs_poll_socket(transport); 687 688 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP); 689 } 690 691 static void xs_poll_check_readable(struct sock_xprt *transport) 692 { 693 694 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 695 if (!xs_poll_socket_readable(transport)) 696 return; 697 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 698 queue_work(xprtiod_workqueue, &transport->recv_worker); 699 } 700 701 static void xs_stream_data_receive(struct sock_xprt *transport) 702 { 703 size_t read = 0; 704 ssize_t ret = 0; 705 706 mutex_lock(&transport->recv_mutex); 707 if (transport->sock == NULL) 708 goto out; 709 for (;;) { 710 ret = xs_read_stream(transport, MSG_DONTWAIT); 711 if (ret < 0) 712 break; 713 read += ret; 714 cond_resched(); 715 } 716 if (ret == -ESHUTDOWN) 717 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 718 else 719 xs_poll_check_readable(transport); 720 out: 721 mutex_unlock(&transport->recv_mutex); 722 trace_xs_stream_read_data(&transport->xprt, ret, read); 723 } 724 725 static void xs_stream_data_receive_workfn(struct work_struct *work) 726 { 727 struct sock_xprt *transport = 728 container_of(work, struct sock_xprt, recv_worker); 729 unsigned int pflags = memalloc_nofs_save(); 730 731 xs_stream_data_receive(transport); 732 memalloc_nofs_restore(pflags); 733 } 734 735 static void 736 xs_stream_reset_connect(struct sock_xprt *transport) 737 { 738 transport->recv.offset = 0; 739 transport->recv.len = 0; 740 transport->recv.copied = 0; 741 transport->xmit.offset = 0; 742 } 743 744 static void 745 xs_stream_start_connect(struct sock_xprt *transport) 746 { 747 transport->xprt.stat.connect_count++; 748 transport->xprt.stat.connect_start = jiffies; 749 } 750 751 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 752 753 /** 754 * xs_nospace - handle transmit was incomplete 755 * @req: pointer to RPC request 756 * 757 */ 758 static int xs_nospace(struct rpc_rqst *req) 759 { 760 struct rpc_xprt *xprt = req->rq_xprt; 761 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 762 struct sock *sk = transport->inet; 763 int ret = -EAGAIN; 764 765 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", 766 req->rq_task->tk_pid, 767 req->rq_slen - transport->xmit.offset, 768 req->rq_slen); 769 770 /* Protect against races with write_space */ 771 spin_lock(&xprt->transport_lock); 772 773 /* Don't race with disconnect */ 774 if (xprt_connected(xprt)) { 775 /* wait for more buffer space */ 776 sk->sk_write_pending++; 777 xprt_wait_for_buffer_space(xprt); 778 } else 779 ret = -ENOTCONN; 780 781 spin_unlock(&xprt->transport_lock); 782 783 /* Race breaker in case memory is freed before above code is called */ 784 if (ret == -EAGAIN) { 785 struct socket_wq *wq; 786 787 rcu_read_lock(); 788 wq = rcu_dereference(sk->sk_wq); 789 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags); 790 rcu_read_unlock(); 791 792 sk->sk_write_space(sk); 793 } 794 return ret; 795 } 796 797 static void 798 xs_stream_prepare_request(struct rpc_rqst *req) 799 { 800 xdr_free_bvec(&req->rq_rcv_buf); 801 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL); 802 } 803 804 /* 805 * Determine if the previous message in the stream was aborted before it 806 * could complete transmission. 807 */ 808 static bool 809 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req) 810 { 811 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0; 812 } 813 814 /* 815 * Return the stream record marker field for a record of length < 2^31-1 816 */ 817 static rpc_fraghdr 818 xs_stream_record_marker(struct xdr_buf *xdr) 819 { 820 if (!xdr->len) 821 return 0; 822 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len); 823 } 824 825 /** 826 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 827 * @req: pointer to RPC request 828 * 829 * Return values: 830 * 0: The request has been sent 831 * EAGAIN: The socket was blocked, please call again later to 832 * complete the request 833 * ENOTCONN: Caller needs to invoke connect logic then call again 834 * other: Some other error occured, the request was not sent 835 */ 836 static int xs_local_send_request(struct rpc_rqst *req) 837 { 838 struct rpc_xprt *xprt = req->rq_xprt; 839 struct sock_xprt *transport = 840 container_of(xprt, struct sock_xprt, xprt); 841 struct xdr_buf *xdr = &req->rq_snd_buf; 842 rpc_fraghdr rm = xs_stream_record_marker(xdr); 843 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 844 struct msghdr msg = { 845 .msg_flags = XS_SENDMSG_FLAGS, 846 }; 847 unsigned int uninitialized_var(sent); 848 int status; 849 850 /* Close the stream if the previous transmission was incomplete */ 851 if (xs_send_request_was_aborted(transport, req)) { 852 xs_close(xprt); 853 return -ENOTCONN; 854 } 855 856 xs_pktdump("packet data:", 857 req->rq_svec->iov_base, req->rq_svec->iov_len); 858 859 req->rq_xtime = ktime_get(); 860 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 861 transport->xmit.offset, rm, &sent); 862 dprintk("RPC: %s(%u) = %d\n", 863 __func__, xdr->len - transport->xmit.offset, status); 864 865 if (status == -EAGAIN && sock_writeable(transport->inet)) 866 status = -ENOBUFS; 867 868 if (likely(sent > 0) || status == 0) { 869 transport->xmit.offset += sent; 870 req->rq_bytes_sent = transport->xmit.offset; 871 if (likely(req->rq_bytes_sent >= msglen)) { 872 req->rq_xmit_bytes_sent += transport->xmit.offset; 873 transport->xmit.offset = 0; 874 return 0; 875 } 876 status = -EAGAIN; 877 } 878 879 switch (status) { 880 case -ENOBUFS: 881 break; 882 case -EAGAIN: 883 status = xs_nospace(req); 884 break; 885 default: 886 dprintk("RPC: sendmsg returned unrecognized error %d\n", 887 -status); 888 /* fall through */ 889 case -EPIPE: 890 xs_close(xprt); 891 status = -ENOTCONN; 892 } 893 894 return status; 895 } 896 897 /** 898 * xs_udp_send_request - write an RPC request to a UDP socket 899 * @req: pointer to RPC request 900 * 901 * Return values: 902 * 0: The request has been sent 903 * EAGAIN: The socket was blocked, please call again later to 904 * complete the request 905 * ENOTCONN: Caller needs to invoke connect logic then call again 906 * other: Some other error occurred, the request was not sent 907 */ 908 static int xs_udp_send_request(struct rpc_rqst *req) 909 { 910 struct rpc_xprt *xprt = req->rq_xprt; 911 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 912 struct xdr_buf *xdr = &req->rq_snd_buf; 913 struct msghdr msg = { 914 .msg_name = xs_addr(xprt), 915 .msg_namelen = xprt->addrlen, 916 .msg_flags = XS_SENDMSG_FLAGS, 917 }; 918 unsigned int uninitialized_var(sent); 919 int status; 920 921 xs_pktdump("packet data:", 922 req->rq_svec->iov_base, 923 req->rq_svec->iov_len); 924 925 if (!xprt_bound(xprt)) 926 return -ENOTCONN; 927 928 if (!xprt_request_get_cong(xprt, req)) 929 return -EBADSLT; 930 931 req->rq_xtime = ktime_get(); 932 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent); 933 934 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 935 xdr->len, status); 936 937 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 938 if (status == -EPERM) 939 goto process_status; 940 941 if (status == -EAGAIN && sock_writeable(transport->inet)) 942 status = -ENOBUFS; 943 944 if (sent > 0 || status == 0) { 945 req->rq_xmit_bytes_sent += sent; 946 if (sent >= req->rq_slen) 947 return 0; 948 /* Still some bytes left; set up for a retry later. */ 949 status = -EAGAIN; 950 } 951 952 process_status: 953 switch (status) { 954 case -ENOTSOCK: 955 status = -ENOTCONN; 956 /* Should we call xs_close() here? */ 957 break; 958 case -EAGAIN: 959 status = xs_nospace(req); 960 break; 961 case -ENETUNREACH: 962 case -ENOBUFS: 963 case -EPIPE: 964 case -ECONNREFUSED: 965 case -EPERM: 966 /* When the server has died, an ICMP port unreachable message 967 * prompts ECONNREFUSED. */ 968 break; 969 default: 970 dprintk("RPC: sendmsg returned unrecognized error %d\n", 971 -status); 972 } 973 974 return status; 975 } 976 977 /** 978 * xs_tcp_send_request - write an RPC request to a TCP socket 979 * @req: pointer to RPC request 980 * 981 * Return values: 982 * 0: The request has been sent 983 * EAGAIN: The socket was blocked, please call again later to 984 * complete the request 985 * ENOTCONN: Caller needs to invoke connect logic then call again 986 * other: Some other error occurred, the request was not sent 987 * 988 * XXX: In the case of soft timeouts, should we eventually give up 989 * if sendmsg is not able to make progress? 990 */ 991 static int xs_tcp_send_request(struct rpc_rqst *req) 992 { 993 struct rpc_xprt *xprt = req->rq_xprt; 994 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 995 struct xdr_buf *xdr = &req->rq_snd_buf; 996 rpc_fraghdr rm = xs_stream_record_marker(xdr); 997 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 998 struct msghdr msg = { 999 .msg_flags = XS_SENDMSG_FLAGS, 1000 }; 1001 bool vm_wait = false; 1002 unsigned int uninitialized_var(sent); 1003 int status; 1004 1005 /* Close the stream if the previous transmission was incomplete */ 1006 if (xs_send_request_was_aborted(transport, req)) { 1007 if (transport->sock != NULL) 1008 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 1009 return -ENOTCONN; 1010 } 1011 1012 xs_pktdump("packet data:", 1013 req->rq_svec->iov_base, 1014 req->rq_svec->iov_len); 1015 1016 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state)) 1017 xs_tcp_set_socket_timeouts(xprt, transport->sock); 1018 1019 /* Continue transmitting the packet/record. We must be careful 1020 * to cope with writespace callbacks arriving _after_ we have 1021 * called sendmsg(). */ 1022 req->rq_xtime = ktime_get(); 1023 while (1) { 1024 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 1025 transport->xmit.offset, rm, &sent); 1026 1027 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 1028 xdr->len - transport->xmit.offset, status); 1029 1030 /* If we've sent the entire packet, immediately 1031 * reset the count of bytes sent. */ 1032 transport->xmit.offset += sent; 1033 req->rq_bytes_sent = transport->xmit.offset; 1034 if (likely(req->rq_bytes_sent >= msglen)) { 1035 req->rq_xmit_bytes_sent += transport->xmit.offset; 1036 transport->xmit.offset = 0; 1037 return 0; 1038 } 1039 1040 WARN_ON_ONCE(sent == 0 && status == 0); 1041 1042 if (status == -EAGAIN ) { 1043 /* 1044 * Return EAGAIN if we're sure we're hitting the 1045 * socket send buffer limits. 1046 */ 1047 if (test_bit(SOCK_NOSPACE, &transport->sock->flags)) 1048 break; 1049 /* 1050 * Did we hit a memory allocation failure? 1051 */ 1052 if (sent == 0) { 1053 status = -ENOBUFS; 1054 if (vm_wait) 1055 break; 1056 /* Retry, knowing now that we're below the 1057 * socket send buffer limit 1058 */ 1059 vm_wait = true; 1060 } 1061 continue; 1062 } 1063 if (status < 0) 1064 break; 1065 vm_wait = false; 1066 } 1067 1068 switch (status) { 1069 case -ENOTSOCK: 1070 status = -ENOTCONN; 1071 /* Should we call xs_close() here? */ 1072 break; 1073 case -EAGAIN: 1074 status = xs_nospace(req); 1075 break; 1076 case -ECONNRESET: 1077 case -ECONNREFUSED: 1078 case -ENOTCONN: 1079 case -EADDRINUSE: 1080 case -ENOBUFS: 1081 case -EPIPE: 1082 break; 1083 default: 1084 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1085 -status); 1086 } 1087 1088 return status; 1089 } 1090 1091 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1092 { 1093 transport->old_data_ready = sk->sk_data_ready; 1094 transport->old_state_change = sk->sk_state_change; 1095 transport->old_write_space = sk->sk_write_space; 1096 transport->old_error_report = sk->sk_error_report; 1097 } 1098 1099 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1100 { 1101 sk->sk_data_ready = transport->old_data_ready; 1102 sk->sk_state_change = transport->old_state_change; 1103 sk->sk_write_space = transport->old_write_space; 1104 sk->sk_error_report = transport->old_error_report; 1105 } 1106 1107 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt) 1108 { 1109 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1110 1111 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 1112 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state); 1113 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state); 1114 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state); 1115 } 1116 1117 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr) 1118 { 1119 set_bit(nr, &transport->sock_state); 1120 queue_work(xprtiod_workqueue, &transport->error_worker); 1121 } 1122 1123 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1124 { 1125 smp_mb__before_atomic(); 1126 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1127 clear_bit(XPRT_CLOSING, &xprt->state); 1128 xs_sock_reset_state_flags(xprt); 1129 smp_mb__after_atomic(); 1130 } 1131 1132 /** 1133 * xs_error_report - callback to handle TCP socket state errors 1134 * @sk: socket 1135 * 1136 * Note: we don't call sock_error() since there may be a rpc_task 1137 * using the socket, and so we don't want to clear sk->sk_err. 1138 */ 1139 static void xs_error_report(struct sock *sk) 1140 { 1141 struct sock_xprt *transport; 1142 struct rpc_xprt *xprt; 1143 1144 read_lock_bh(&sk->sk_callback_lock); 1145 if (!(xprt = xprt_from_sock(sk))) 1146 goto out; 1147 1148 transport = container_of(xprt, struct sock_xprt, xprt); 1149 transport->xprt_err = -sk->sk_err; 1150 if (transport->xprt_err == 0) 1151 goto out; 1152 dprintk("RPC: xs_error_report client %p, error=%d...\n", 1153 xprt, -transport->xprt_err); 1154 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err); 1155 1156 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */ 1157 smp_mb__before_atomic(); 1158 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR); 1159 out: 1160 read_unlock_bh(&sk->sk_callback_lock); 1161 } 1162 1163 static void xs_reset_transport(struct sock_xprt *transport) 1164 { 1165 struct socket *sock = transport->sock; 1166 struct sock *sk = transport->inet; 1167 struct rpc_xprt *xprt = &transport->xprt; 1168 struct file *filp = transport->file; 1169 1170 if (sk == NULL) 1171 return; 1172 1173 if (atomic_read(&transport->xprt.swapper)) 1174 sk_clear_memalloc(sk); 1175 1176 kernel_sock_shutdown(sock, SHUT_RDWR); 1177 1178 mutex_lock(&transport->recv_mutex); 1179 write_lock_bh(&sk->sk_callback_lock); 1180 transport->inet = NULL; 1181 transport->sock = NULL; 1182 transport->file = NULL; 1183 1184 sk->sk_user_data = NULL; 1185 1186 xs_restore_old_callbacks(transport, sk); 1187 xprt_clear_connected(xprt); 1188 write_unlock_bh(&sk->sk_callback_lock); 1189 xs_sock_reset_connection_flags(xprt); 1190 /* Reset stream record info */ 1191 xs_stream_reset_connect(transport); 1192 mutex_unlock(&transport->recv_mutex); 1193 1194 trace_rpc_socket_close(xprt, sock); 1195 fput(filp); 1196 1197 xprt_disconnect_done(xprt); 1198 } 1199 1200 /** 1201 * xs_close - close a socket 1202 * @xprt: transport 1203 * 1204 * This is used when all requests are complete; ie, no DRC state remains 1205 * on the server we want to save. 1206 * 1207 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 1208 * xs_reset_transport() zeroing the socket from underneath a writer. 1209 */ 1210 static void xs_close(struct rpc_xprt *xprt) 1211 { 1212 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1213 1214 dprintk("RPC: xs_close xprt %p\n", xprt); 1215 1216 xs_reset_transport(transport); 1217 xprt->reestablish_timeout = 0; 1218 } 1219 1220 static void xs_inject_disconnect(struct rpc_xprt *xprt) 1221 { 1222 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 1223 xprt); 1224 xprt_disconnect_done(xprt); 1225 } 1226 1227 static void xs_xprt_free(struct rpc_xprt *xprt) 1228 { 1229 xs_free_peer_addresses(xprt); 1230 xprt_free(xprt); 1231 } 1232 1233 /** 1234 * xs_destroy - prepare to shutdown a transport 1235 * @xprt: doomed transport 1236 * 1237 */ 1238 static void xs_destroy(struct rpc_xprt *xprt) 1239 { 1240 struct sock_xprt *transport = container_of(xprt, 1241 struct sock_xprt, xprt); 1242 dprintk("RPC: xs_destroy xprt %p\n", xprt); 1243 1244 cancel_delayed_work_sync(&transport->connect_worker); 1245 xs_close(xprt); 1246 cancel_work_sync(&transport->recv_worker); 1247 cancel_work_sync(&transport->error_worker); 1248 xs_xprt_free(xprt); 1249 module_put(THIS_MODULE); 1250 } 1251 1252 /** 1253 * xs_udp_data_read_skb - receive callback for UDP sockets 1254 * @xprt: transport 1255 * @sk: socket 1256 * @skb: skbuff 1257 * 1258 */ 1259 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1260 struct sock *sk, 1261 struct sk_buff *skb) 1262 { 1263 struct rpc_task *task; 1264 struct rpc_rqst *rovr; 1265 int repsize, copied; 1266 u32 _xid; 1267 __be32 *xp; 1268 1269 repsize = skb->len; 1270 if (repsize < 4) { 1271 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1272 return; 1273 } 1274 1275 /* Copy the XID from the skb... */ 1276 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid); 1277 if (xp == NULL) 1278 return; 1279 1280 /* Look up and lock the request corresponding to the given XID */ 1281 spin_lock(&xprt->queue_lock); 1282 rovr = xprt_lookup_rqst(xprt, *xp); 1283 if (!rovr) 1284 goto out_unlock; 1285 xprt_pin_rqst(rovr); 1286 xprt_update_rtt(rovr->rq_task); 1287 spin_unlock(&xprt->queue_lock); 1288 task = rovr->rq_task; 1289 1290 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1291 copied = repsize; 1292 1293 /* Suck it into the iovec, verify checksum if not done by hw. */ 1294 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1295 spin_lock(&xprt->queue_lock); 1296 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS); 1297 goto out_unpin; 1298 } 1299 1300 1301 spin_lock(&xprt->transport_lock); 1302 xprt_adjust_cwnd(xprt, task, copied); 1303 spin_unlock(&xprt->transport_lock); 1304 spin_lock(&xprt->queue_lock); 1305 xprt_complete_rqst(task, copied); 1306 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS); 1307 out_unpin: 1308 xprt_unpin_rqst(rovr); 1309 out_unlock: 1310 spin_unlock(&xprt->queue_lock); 1311 } 1312 1313 static void xs_udp_data_receive(struct sock_xprt *transport) 1314 { 1315 struct sk_buff *skb; 1316 struct sock *sk; 1317 int err; 1318 1319 mutex_lock(&transport->recv_mutex); 1320 sk = transport->inet; 1321 if (sk == NULL) 1322 goto out; 1323 for (;;) { 1324 skb = skb_recv_udp(sk, 0, 1, &err); 1325 if (skb == NULL) 1326 break; 1327 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1328 consume_skb(skb); 1329 cond_resched(); 1330 } 1331 xs_poll_check_readable(transport); 1332 out: 1333 mutex_unlock(&transport->recv_mutex); 1334 } 1335 1336 static void xs_udp_data_receive_workfn(struct work_struct *work) 1337 { 1338 struct sock_xprt *transport = 1339 container_of(work, struct sock_xprt, recv_worker); 1340 unsigned int pflags = memalloc_nofs_save(); 1341 1342 xs_udp_data_receive(transport); 1343 memalloc_nofs_restore(pflags); 1344 } 1345 1346 /** 1347 * xs_data_ready - "data ready" callback for UDP sockets 1348 * @sk: socket with data to read 1349 * 1350 */ 1351 static void xs_data_ready(struct sock *sk) 1352 { 1353 struct rpc_xprt *xprt; 1354 1355 read_lock_bh(&sk->sk_callback_lock); 1356 dprintk("RPC: xs_data_ready...\n"); 1357 xprt = xprt_from_sock(sk); 1358 if (xprt != NULL) { 1359 struct sock_xprt *transport = container_of(xprt, 1360 struct sock_xprt, xprt); 1361 transport->old_data_ready(sk); 1362 /* Any data means we had a useful conversation, so 1363 * then we don't need to delay the next reconnect 1364 */ 1365 if (xprt->reestablish_timeout) 1366 xprt->reestablish_timeout = 0; 1367 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1368 queue_work(xprtiod_workqueue, &transport->recv_worker); 1369 } 1370 read_unlock_bh(&sk->sk_callback_lock); 1371 } 1372 1373 /* 1374 * Helper function to force a TCP close if the server is sending 1375 * junk and/or it has put us in CLOSE_WAIT 1376 */ 1377 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1378 { 1379 xprt_force_disconnect(xprt); 1380 } 1381 1382 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1383 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt) 1384 { 1385 return PAGE_SIZE; 1386 } 1387 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1388 1389 /** 1390 * xs_tcp_state_change - callback to handle TCP socket state changes 1391 * @sk: socket whose state has changed 1392 * 1393 */ 1394 static void xs_tcp_state_change(struct sock *sk) 1395 { 1396 struct rpc_xprt *xprt; 1397 struct sock_xprt *transport; 1398 1399 read_lock_bh(&sk->sk_callback_lock); 1400 if (!(xprt = xprt_from_sock(sk))) 1401 goto out; 1402 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1403 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1404 sk->sk_state, xprt_connected(xprt), 1405 sock_flag(sk, SOCK_DEAD), 1406 sock_flag(sk, SOCK_ZAPPED), 1407 sk->sk_shutdown); 1408 1409 transport = container_of(xprt, struct sock_xprt, xprt); 1410 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1411 switch (sk->sk_state) { 1412 case TCP_ESTABLISHED: 1413 if (!xprt_test_and_set_connected(xprt)) { 1414 xprt->connect_cookie++; 1415 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1416 xprt_clear_connecting(xprt); 1417 1418 xprt->stat.connect_count++; 1419 xprt->stat.connect_time += (long)jiffies - 1420 xprt->stat.connect_start; 1421 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING); 1422 } 1423 break; 1424 case TCP_FIN_WAIT1: 1425 /* The client initiated a shutdown of the socket */ 1426 xprt->connect_cookie++; 1427 xprt->reestablish_timeout = 0; 1428 set_bit(XPRT_CLOSING, &xprt->state); 1429 smp_mb__before_atomic(); 1430 clear_bit(XPRT_CONNECTED, &xprt->state); 1431 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1432 smp_mb__after_atomic(); 1433 break; 1434 case TCP_CLOSE_WAIT: 1435 /* The server initiated a shutdown of the socket */ 1436 xprt->connect_cookie++; 1437 clear_bit(XPRT_CONNECTED, &xprt->state); 1438 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1439 /* fall through */ 1440 case TCP_CLOSING: 1441 /* 1442 * If the server closed down the connection, make sure that 1443 * we back off before reconnecting 1444 */ 1445 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1446 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1447 break; 1448 case TCP_LAST_ACK: 1449 set_bit(XPRT_CLOSING, &xprt->state); 1450 smp_mb__before_atomic(); 1451 clear_bit(XPRT_CONNECTED, &xprt->state); 1452 smp_mb__after_atomic(); 1453 break; 1454 case TCP_CLOSE: 1455 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1456 &transport->sock_state)) 1457 xprt_clear_connecting(xprt); 1458 clear_bit(XPRT_CLOSING, &xprt->state); 1459 /* Trigger the socket release */ 1460 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1461 } 1462 out: 1463 read_unlock_bh(&sk->sk_callback_lock); 1464 } 1465 1466 static void xs_write_space(struct sock *sk) 1467 { 1468 struct socket_wq *wq; 1469 struct sock_xprt *transport; 1470 struct rpc_xprt *xprt; 1471 1472 if (!sk->sk_socket) 1473 return; 1474 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1475 1476 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1477 return; 1478 transport = container_of(xprt, struct sock_xprt, xprt); 1479 rcu_read_lock(); 1480 wq = rcu_dereference(sk->sk_wq); 1481 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0) 1482 goto out; 1483 1484 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE); 1485 sk->sk_write_pending--; 1486 out: 1487 rcu_read_unlock(); 1488 } 1489 1490 /** 1491 * xs_udp_write_space - callback invoked when socket buffer space 1492 * becomes available 1493 * @sk: socket whose state has changed 1494 * 1495 * Called when more output buffer space is available for this socket. 1496 * We try not to wake our writers until they can make "significant" 1497 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1498 * with a bunch of small requests. 1499 */ 1500 static void xs_udp_write_space(struct sock *sk) 1501 { 1502 read_lock_bh(&sk->sk_callback_lock); 1503 1504 /* from net/core/sock.c:sock_def_write_space */ 1505 if (sock_writeable(sk)) 1506 xs_write_space(sk); 1507 1508 read_unlock_bh(&sk->sk_callback_lock); 1509 } 1510 1511 /** 1512 * xs_tcp_write_space - callback invoked when socket buffer space 1513 * becomes available 1514 * @sk: socket whose state has changed 1515 * 1516 * Called when more output buffer space is available for this socket. 1517 * We try not to wake our writers until they can make "significant" 1518 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1519 * with a bunch of small requests. 1520 */ 1521 static void xs_tcp_write_space(struct sock *sk) 1522 { 1523 read_lock_bh(&sk->sk_callback_lock); 1524 1525 /* from net/core/stream.c:sk_stream_write_space */ 1526 if (sk_stream_is_writeable(sk)) 1527 xs_write_space(sk); 1528 1529 read_unlock_bh(&sk->sk_callback_lock); 1530 } 1531 1532 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1533 { 1534 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1535 struct sock *sk = transport->inet; 1536 1537 if (transport->rcvsize) { 1538 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1539 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1540 } 1541 if (transport->sndsize) { 1542 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1543 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1544 sk->sk_write_space(sk); 1545 } 1546 } 1547 1548 /** 1549 * xs_udp_set_buffer_size - set send and receive limits 1550 * @xprt: generic transport 1551 * @sndsize: requested size of send buffer, in bytes 1552 * @rcvsize: requested size of receive buffer, in bytes 1553 * 1554 * Set socket send and receive buffer size limits. 1555 */ 1556 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1557 { 1558 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1559 1560 transport->sndsize = 0; 1561 if (sndsize) 1562 transport->sndsize = sndsize + 1024; 1563 transport->rcvsize = 0; 1564 if (rcvsize) 1565 transport->rcvsize = rcvsize + 1024; 1566 1567 xs_udp_do_set_buffer_size(xprt); 1568 } 1569 1570 /** 1571 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1572 * @xprt: controlling transport 1573 * @task: task that timed out 1574 * 1575 * Adjust the congestion window after a retransmit timeout has occurred. 1576 */ 1577 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1578 { 1579 spin_lock(&xprt->transport_lock); 1580 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1581 spin_unlock(&xprt->transport_lock); 1582 } 1583 1584 static int xs_get_random_port(void) 1585 { 1586 unsigned short min = xprt_min_resvport, max = xprt_max_resvport; 1587 unsigned short range; 1588 unsigned short rand; 1589 1590 if (max < min) 1591 return -EADDRINUSE; 1592 range = max - min + 1; 1593 rand = (unsigned short) prandom_u32() % range; 1594 return rand + min; 1595 } 1596 1597 /** 1598 * xs_set_reuseaddr_port - set the socket's port and address reuse options 1599 * @sock: socket 1600 * 1601 * Note that this function has to be called on all sockets that share the 1602 * same port, and it must be called before binding. 1603 */ 1604 static void xs_sock_set_reuseport(struct socket *sock) 1605 { 1606 int opt = 1; 1607 1608 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, 1609 (char *)&opt, sizeof(opt)); 1610 } 1611 1612 static unsigned short xs_sock_getport(struct socket *sock) 1613 { 1614 struct sockaddr_storage buf; 1615 unsigned short port = 0; 1616 1617 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0) 1618 goto out; 1619 switch (buf.ss_family) { 1620 case AF_INET6: 1621 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1622 break; 1623 case AF_INET: 1624 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1625 } 1626 out: 1627 return port; 1628 } 1629 1630 /** 1631 * xs_set_port - reset the port number in the remote endpoint address 1632 * @xprt: generic transport 1633 * @port: new port number 1634 * 1635 */ 1636 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1637 { 1638 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1639 1640 rpc_set_port(xs_addr(xprt), port); 1641 xs_update_peer_port(xprt); 1642 } 1643 1644 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1645 { 1646 if (transport->srcport == 0 && transport->xprt.reuseport) 1647 transport->srcport = xs_sock_getport(sock); 1648 } 1649 1650 static int xs_get_srcport(struct sock_xprt *transport) 1651 { 1652 int port = transport->srcport; 1653 1654 if (port == 0 && transport->xprt.resvport) 1655 port = xs_get_random_port(); 1656 return port; 1657 } 1658 1659 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1660 { 1661 if (transport->srcport != 0) 1662 transport->srcport = 0; 1663 if (!transport->xprt.resvport) 1664 return 0; 1665 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1666 return xprt_max_resvport; 1667 return --port; 1668 } 1669 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1670 { 1671 struct sockaddr_storage myaddr; 1672 int err, nloop = 0; 1673 int port = xs_get_srcport(transport); 1674 unsigned short last; 1675 1676 /* 1677 * If we are asking for any ephemeral port (i.e. port == 0 && 1678 * transport->xprt.resvport == 0), don't bind. Let the local 1679 * port selection happen implicitly when the socket is used 1680 * (for example at connect time). 1681 * 1682 * This ensures that we can continue to establish TCP 1683 * connections even when all local ephemeral ports are already 1684 * a part of some TCP connection. This makes no difference 1685 * for UDP sockets, but also doens't harm them. 1686 * 1687 * If we're asking for any reserved port (i.e. port == 0 && 1688 * transport->xprt.resvport == 1) xs_get_srcport above will 1689 * ensure that port is non-zero and we will bind as needed. 1690 */ 1691 if (port <= 0) 1692 return port; 1693 1694 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1695 do { 1696 rpc_set_port((struct sockaddr *)&myaddr, port); 1697 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1698 transport->xprt.addrlen); 1699 if (err == 0) { 1700 transport->srcport = port; 1701 break; 1702 } 1703 last = port; 1704 port = xs_next_srcport(transport, port); 1705 if (port > last) 1706 nloop++; 1707 } while (err == -EADDRINUSE && nloop != 2); 1708 1709 if (myaddr.ss_family == AF_INET) 1710 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1711 &((struct sockaddr_in *)&myaddr)->sin_addr, 1712 port, err ? "failed" : "ok", err); 1713 else 1714 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1715 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1716 port, err ? "failed" : "ok", err); 1717 return err; 1718 } 1719 1720 /* 1721 * We don't support autobind on AF_LOCAL sockets 1722 */ 1723 static void xs_local_rpcbind(struct rpc_task *task) 1724 { 1725 xprt_set_bound(task->tk_xprt); 1726 } 1727 1728 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1729 { 1730 } 1731 1732 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1733 static struct lock_class_key xs_key[2]; 1734 static struct lock_class_key xs_slock_key[2]; 1735 1736 static inline void xs_reclassify_socketu(struct socket *sock) 1737 { 1738 struct sock *sk = sock->sk; 1739 1740 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1741 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); 1742 } 1743 1744 static inline void xs_reclassify_socket4(struct socket *sock) 1745 { 1746 struct sock *sk = sock->sk; 1747 1748 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1749 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1750 } 1751 1752 static inline void xs_reclassify_socket6(struct socket *sock) 1753 { 1754 struct sock *sk = sock->sk; 1755 1756 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1757 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1758 } 1759 1760 static inline void xs_reclassify_socket(int family, struct socket *sock) 1761 { 1762 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk))) 1763 return; 1764 1765 switch (family) { 1766 case AF_LOCAL: 1767 xs_reclassify_socketu(sock); 1768 break; 1769 case AF_INET: 1770 xs_reclassify_socket4(sock); 1771 break; 1772 case AF_INET6: 1773 xs_reclassify_socket6(sock); 1774 break; 1775 } 1776 } 1777 #else 1778 static inline void xs_reclassify_socket(int family, struct socket *sock) 1779 { 1780 } 1781 #endif 1782 1783 static void xs_dummy_setup_socket(struct work_struct *work) 1784 { 1785 } 1786 1787 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1788 struct sock_xprt *transport, int family, int type, 1789 int protocol, bool reuseport) 1790 { 1791 struct file *filp; 1792 struct socket *sock; 1793 int err; 1794 1795 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1796 if (err < 0) { 1797 dprintk("RPC: can't create %d transport socket (%d).\n", 1798 protocol, -err); 1799 goto out; 1800 } 1801 xs_reclassify_socket(family, sock); 1802 1803 if (reuseport) 1804 xs_sock_set_reuseport(sock); 1805 1806 err = xs_bind(transport, sock); 1807 if (err) { 1808 sock_release(sock); 1809 goto out; 1810 } 1811 1812 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1813 if (IS_ERR(filp)) 1814 return ERR_CAST(filp); 1815 transport->file = filp; 1816 1817 return sock; 1818 out: 1819 return ERR_PTR(err); 1820 } 1821 1822 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1823 struct socket *sock) 1824 { 1825 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1826 xprt); 1827 1828 if (!transport->inet) { 1829 struct sock *sk = sock->sk; 1830 1831 write_lock_bh(&sk->sk_callback_lock); 1832 1833 xs_save_old_callbacks(transport, sk); 1834 1835 sk->sk_user_data = xprt; 1836 sk->sk_data_ready = xs_data_ready; 1837 sk->sk_write_space = xs_udp_write_space; 1838 sock_set_flag(sk, SOCK_FASYNC); 1839 sk->sk_error_report = xs_error_report; 1840 1841 xprt_clear_connected(xprt); 1842 1843 /* Reset to new socket */ 1844 transport->sock = sock; 1845 transport->inet = sk; 1846 1847 write_unlock_bh(&sk->sk_callback_lock); 1848 } 1849 1850 xs_stream_start_connect(transport); 1851 1852 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1853 } 1854 1855 /** 1856 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1857 * @transport: socket transport to connect 1858 */ 1859 static int xs_local_setup_socket(struct sock_xprt *transport) 1860 { 1861 struct rpc_xprt *xprt = &transport->xprt; 1862 struct file *filp; 1863 struct socket *sock; 1864 int status; 1865 1866 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1867 SOCK_STREAM, 0, &sock, 1); 1868 if (status < 0) { 1869 dprintk("RPC: can't create AF_LOCAL " 1870 "transport socket (%d).\n", -status); 1871 goto out; 1872 } 1873 xs_reclassify_socket(AF_LOCAL, sock); 1874 1875 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1876 if (IS_ERR(filp)) { 1877 status = PTR_ERR(filp); 1878 goto out; 1879 } 1880 transport->file = filp; 1881 1882 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1883 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1884 1885 status = xs_local_finish_connecting(xprt, sock); 1886 trace_rpc_socket_connect(xprt, sock, status); 1887 switch (status) { 1888 case 0: 1889 dprintk("RPC: xprt %p connected to %s\n", 1890 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1891 xprt->stat.connect_count++; 1892 xprt->stat.connect_time += (long)jiffies - 1893 xprt->stat.connect_start; 1894 xprt_set_connected(xprt); 1895 case -ENOBUFS: 1896 break; 1897 case -ENOENT: 1898 dprintk("RPC: xprt %p: socket %s does not exist\n", 1899 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1900 break; 1901 case -ECONNREFUSED: 1902 dprintk("RPC: xprt %p: connection refused for %s\n", 1903 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1904 break; 1905 default: 1906 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 1907 __func__, -status, 1908 xprt->address_strings[RPC_DISPLAY_ADDR]); 1909 } 1910 1911 out: 1912 xprt_clear_connecting(xprt); 1913 xprt_wake_pending_tasks(xprt, status); 1914 return status; 1915 } 1916 1917 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 1918 { 1919 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1920 int ret; 1921 1922 if (RPC_IS_ASYNC(task)) { 1923 /* 1924 * We want the AF_LOCAL connect to be resolved in the 1925 * filesystem namespace of the process making the rpc 1926 * call. Thus we connect synchronously. 1927 * 1928 * If we want to support asynchronous AF_LOCAL calls, 1929 * we'll need to figure out how to pass a namespace to 1930 * connect. 1931 */ 1932 task->tk_rpc_status = -ENOTCONN; 1933 rpc_exit(task, -ENOTCONN); 1934 return; 1935 } 1936 ret = xs_local_setup_socket(transport); 1937 if (ret && !RPC_IS_SOFTCONN(task)) 1938 msleep_interruptible(15000); 1939 } 1940 1941 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 1942 /* 1943 * Note that this should be called with XPRT_LOCKED held (or when we otherwise 1944 * know that we have exclusive access to the socket), to guard against 1945 * races with xs_reset_transport. 1946 */ 1947 static void xs_set_memalloc(struct rpc_xprt *xprt) 1948 { 1949 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1950 xprt); 1951 1952 /* 1953 * If there's no sock, then we have nothing to set. The 1954 * reconnecting process will get it for us. 1955 */ 1956 if (!transport->inet) 1957 return; 1958 if (atomic_read(&xprt->swapper)) 1959 sk_set_memalloc(transport->inet); 1960 } 1961 1962 /** 1963 * xs_enable_swap - Tag this transport as being used for swap. 1964 * @xprt: transport to tag 1965 * 1966 * Take a reference to this transport on behalf of the rpc_clnt, and 1967 * optionally mark it for swapping if it wasn't already. 1968 */ 1969 static int 1970 xs_enable_swap(struct rpc_xprt *xprt) 1971 { 1972 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 1973 1974 if (atomic_inc_return(&xprt->swapper) != 1) 1975 return 0; 1976 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 1977 return -ERESTARTSYS; 1978 if (xs->inet) 1979 sk_set_memalloc(xs->inet); 1980 xprt_release_xprt(xprt, NULL); 1981 return 0; 1982 } 1983 1984 /** 1985 * xs_disable_swap - Untag this transport as being used for swap. 1986 * @xprt: transport to tag 1987 * 1988 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 1989 * swapper refcount goes to 0, untag the socket as a memalloc socket. 1990 */ 1991 static void 1992 xs_disable_swap(struct rpc_xprt *xprt) 1993 { 1994 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 1995 1996 if (!atomic_dec_and_test(&xprt->swapper)) 1997 return; 1998 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 1999 return; 2000 if (xs->inet) 2001 sk_clear_memalloc(xs->inet); 2002 xprt_release_xprt(xprt, NULL); 2003 } 2004 #else 2005 static void xs_set_memalloc(struct rpc_xprt *xprt) 2006 { 2007 } 2008 2009 static int 2010 xs_enable_swap(struct rpc_xprt *xprt) 2011 { 2012 return -EINVAL; 2013 } 2014 2015 static void 2016 xs_disable_swap(struct rpc_xprt *xprt) 2017 { 2018 } 2019 #endif 2020 2021 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2022 { 2023 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2024 2025 if (!transport->inet) { 2026 struct sock *sk = sock->sk; 2027 2028 write_lock_bh(&sk->sk_callback_lock); 2029 2030 xs_save_old_callbacks(transport, sk); 2031 2032 sk->sk_user_data = xprt; 2033 sk->sk_data_ready = xs_data_ready; 2034 sk->sk_write_space = xs_udp_write_space; 2035 sock_set_flag(sk, SOCK_FASYNC); 2036 2037 xprt_set_connected(xprt); 2038 2039 /* Reset to new socket */ 2040 transport->sock = sock; 2041 transport->inet = sk; 2042 2043 xs_set_memalloc(xprt); 2044 2045 write_unlock_bh(&sk->sk_callback_lock); 2046 } 2047 xs_udp_do_set_buffer_size(xprt); 2048 2049 xprt->stat.connect_start = jiffies; 2050 } 2051 2052 static void xs_udp_setup_socket(struct work_struct *work) 2053 { 2054 struct sock_xprt *transport = 2055 container_of(work, struct sock_xprt, connect_worker.work); 2056 struct rpc_xprt *xprt = &transport->xprt; 2057 struct socket *sock; 2058 int status = -EIO; 2059 2060 sock = xs_create_sock(xprt, transport, 2061 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2062 IPPROTO_UDP, false); 2063 if (IS_ERR(sock)) 2064 goto out; 2065 2066 dprintk("RPC: worker connecting xprt %p via %s to " 2067 "%s (port %s)\n", xprt, 2068 xprt->address_strings[RPC_DISPLAY_PROTO], 2069 xprt->address_strings[RPC_DISPLAY_ADDR], 2070 xprt->address_strings[RPC_DISPLAY_PORT]); 2071 2072 xs_udp_finish_connecting(xprt, sock); 2073 trace_rpc_socket_connect(xprt, sock, 0); 2074 status = 0; 2075 out: 2076 xprt_clear_connecting(xprt); 2077 xprt_unlock_connect(xprt, transport); 2078 xprt_wake_pending_tasks(xprt, status); 2079 } 2080 2081 /** 2082 * xs_tcp_shutdown - gracefully shut down a TCP socket 2083 * @xprt: transport 2084 * 2085 * Initiates a graceful shutdown of the TCP socket by calling the 2086 * equivalent of shutdown(SHUT_RDWR); 2087 */ 2088 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2089 { 2090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2091 struct socket *sock = transport->sock; 2092 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE; 2093 2094 if (sock == NULL) 2095 return; 2096 switch (skst) { 2097 default: 2098 kernel_sock_shutdown(sock, SHUT_RDWR); 2099 trace_rpc_socket_shutdown(xprt, sock); 2100 break; 2101 case TCP_CLOSE: 2102 case TCP_TIME_WAIT: 2103 xs_reset_transport(transport); 2104 } 2105 } 2106 2107 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 2108 struct socket *sock) 2109 { 2110 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2111 unsigned int keepidle; 2112 unsigned int keepcnt; 2113 unsigned int opt_on = 1; 2114 unsigned int timeo; 2115 2116 spin_lock(&xprt->transport_lock); 2117 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ); 2118 keepcnt = xprt->timeout->to_retries + 1; 2119 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2120 (xprt->timeout->to_retries + 1); 2121 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2122 spin_unlock(&xprt->transport_lock); 2123 2124 /* TCP Keepalive options */ 2125 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 2126 (char *)&opt_on, sizeof(opt_on)); 2127 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, 2128 (char *)&keepidle, sizeof(keepidle)); 2129 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, 2130 (char *)&keepidle, sizeof(keepidle)); 2131 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, 2132 (char *)&keepcnt, sizeof(keepcnt)); 2133 2134 /* TCP user timeout (see RFC5482) */ 2135 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT, 2136 (char *)&timeo, sizeof(timeo)); 2137 } 2138 2139 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt, 2140 unsigned long connect_timeout, 2141 unsigned long reconnect_timeout) 2142 { 2143 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2144 struct rpc_timeout to; 2145 unsigned long initval; 2146 2147 spin_lock(&xprt->transport_lock); 2148 if (reconnect_timeout < xprt->max_reconnect_timeout) 2149 xprt->max_reconnect_timeout = reconnect_timeout; 2150 if (connect_timeout < xprt->connect_timeout) { 2151 memcpy(&to, xprt->timeout, sizeof(to)); 2152 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1); 2153 /* Arbitrary lower limit */ 2154 if (initval < XS_TCP_INIT_REEST_TO << 1) 2155 initval = XS_TCP_INIT_REEST_TO << 1; 2156 to.to_initval = initval; 2157 to.to_maxval = initval; 2158 memcpy(&transport->tcp_timeout, &to, 2159 sizeof(transport->tcp_timeout)); 2160 xprt->timeout = &transport->tcp_timeout; 2161 xprt->connect_timeout = connect_timeout; 2162 } 2163 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2164 spin_unlock(&xprt->transport_lock); 2165 } 2166 2167 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2168 { 2169 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2170 int ret = -ENOTCONN; 2171 2172 if (!transport->inet) { 2173 struct sock *sk = sock->sk; 2174 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC; 2175 2176 /* Avoid temporary address, they are bad for long-lived 2177 * connections such as NFS mounts. 2178 * RFC4941, section 3.6 suggests that: 2179 * Individual applications, which have specific 2180 * knowledge about the normal duration of connections, 2181 * MAY override this as appropriate. 2182 */ 2183 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES, 2184 (char *)&addr_pref, sizeof(addr_pref)); 2185 2186 xs_tcp_set_socket_timeouts(xprt, sock); 2187 2188 write_lock_bh(&sk->sk_callback_lock); 2189 2190 xs_save_old_callbacks(transport, sk); 2191 2192 sk->sk_user_data = xprt; 2193 sk->sk_data_ready = xs_data_ready; 2194 sk->sk_state_change = xs_tcp_state_change; 2195 sk->sk_write_space = xs_tcp_write_space; 2196 sock_set_flag(sk, SOCK_FASYNC); 2197 sk->sk_error_report = xs_error_report; 2198 2199 /* socket options */ 2200 sock_reset_flag(sk, SOCK_LINGER); 2201 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 2202 2203 xprt_clear_connected(xprt); 2204 2205 /* Reset to new socket */ 2206 transport->sock = sock; 2207 transport->inet = sk; 2208 2209 write_unlock_bh(&sk->sk_callback_lock); 2210 } 2211 2212 if (!xprt_bound(xprt)) 2213 goto out; 2214 2215 xs_set_memalloc(xprt); 2216 2217 xs_stream_start_connect(transport); 2218 2219 /* Tell the socket layer to start connecting... */ 2220 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2221 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2222 switch (ret) { 2223 case 0: 2224 xs_set_srcport(transport, sock); 2225 /* fall through */ 2226 case -EINPROGRESS: 2227 /* SYN_SENT! */ 2228 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2229 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2230 break; 2231 case -EADDRNOTAVAIL: 2232 /* Source port number is unavailable. Try a new one! */ 2233 transport->srcport = 0; 2234 } 2235 out: 2236 return ret; 2237 } 2238 2239 /** 2240 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2241 * @work: queued work item 2242 * 2243 * Invoked by a work queue tasklet. 2244 */ 2245 static void xs_tcp_setup_socket(struct work_struct *work) 2246 { 2247 struct sock_xprt *transport = 2248 container_of(work, struct sock_xprt, connect_worker.work); 2249 struct socket *sock = transport->sock; 2250 struct rpc_xprt *xprt = &transport->xprt; 2251 int status = -EIO; 2252 2253 if (!sock) { 2254 sock = xs_create_sock(xprt, transport, 2255 xs_addr(xprt)->sa_family, SOCK_STREAM, 2256 IPPROTO_TCP, true); 2257 if (IS_ERR(sock)) { 2258 status = PTR_ERR(sock); 2259 goto out; 2260 } 2261 } 2262 2263 dprintk("RPC: worker connecting xprt %p via %s to " 2264 "%s (port %s)\n", xprt, 2265 xprt->address_strings[RPC_DISPLAY_PROTO], 2266 xprt->address_strings[RPC_DISPLAY_ADDR], 2267 xprt->address_strings[RPC_DISPLAY_PORT]); 2268 2269 status = xs_tcp_finish_connecting(xprt, sock); 2270 trace_rpc_socket_connect(xprt, sock, status); 2271 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2272 xprt, -status, xprt_connected(xprt), 2273 sock->sk->sk_state); 2274 switch (status) { 2275 default: 2276 printk("%s: connect returned unhandled error %d\n", 2277 __func__, status); 2278 /* fall through */ 2279 case -EADDRNOTAVAIL: 2280 /* We're probably in TIME_WAIT. Get rid of existing socket, 2281 * and retry 2282 */ 2283 xs_tcp_force_close(xprt); 2284 break; 2285 case 0: 2286 case -EINPROGRESS: 2287 case -EALREADY: 2288 xprt_unlock_connect(xprt, transport); 2289 return; 2290 case -EINVAL: 2291 /* Happens, for instance, if the user specified a link 2292 * local IPv6 address without a scope-id. 2293 */ 2294 case -ECONNREFUSED: 2295 case -ECONNRESET: 2296 case -ENETDOWN: 2297 case -ENETUNREACH: 2298 case -EHOSTUNREACH: 2299 case -EADDRINUSE: 2300 case -ENOBUFS: 2301 /* 2302 * xs_tcp_force_close() wakes tasks with -EIO. 2303 * We need to wake them first to ensure the 2304 * correct error code. 2305 */ 2306 xprt_wake_pending_tasks(xprt, status); 2307 xs_tcp_force_close(xprt); 2308 goto out; 2309 } 2310 status = -EAGAIN; 2311 out: 2312 xprt_clear_connecting(xprt); 2313 xprt_unlock_connect(xprt, transport); 2314 xprt_wake_pending_tasks(xprt, status); 2315 } 2316 2317 /** 2318 * xs_connect - connect a socket to a remote endpoint 2319 * @xprt: pointer to transport structure 2320 * @task: address of RPC task that manages state of connect request 2321 * 2322 * TCP: If the remote end dropped the connection, delay reconnecting. 2323 * 2324 * UDP socket connects are synchronous, but we use a work queue anyway 2325 * to guarantee that even unprivileged user processes can set up a 2326 * socket on a privileged port. 2327 * 2328 * If a UDP socket connect fails, the delay behavior here prevents 2329 * retry floods (hard mounts). 2330 */ 2331 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2332 { 2333 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2334 unsigned long delay = 0; 2335 2336 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2337 2338 if (transport->sock != NULL) { 2339 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2340 "seconds\n", 2341 xprt, xprt->reestablish_timeout / HZ); 2342 2343 /* Start by resetting any existing state */ 2344 xs_reset_transport(transport); 2345 2346 delay = xprt_reconnect_delay(xprt); 2347 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO); 2348 2349 } else 2350 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2351 2352 queue_delayed_work(xprtiod_workqueue, 2353 &transport->connect_worker, 2354 delay); 2355 } 2356 2357 static void xs_wake_disconnect(struct sock_xprt *transport) 2358 { 2359 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state)) 2360 xs_tcp_force_close(&transport->xprt); 2361 } 2362 2363 static void xs_wake_write(struct sock_xprt *transport) 2364 { 2365 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state)) 2366 xprt_write_space(&transport->xprt); 2367 } 2368 2369 static void xs_wake_error(struct sock_xprt *transport) 2370 { 2371 int sockerr; 2372 2373 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2374 return; 2375 mutex_lock(&transport->recv_mutex); 2376 if (transport->sock == NULL) 2377 goto out; 2378 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2379 goto out; 2380 sockerr = xchg(&transport->xprt_err, 0); 2381 if (sockerr < 0) 2382 xprt_wake_pending_tasks(&transport->xprt, sockerr); 2383 out: 2384 mutex_unlock(&transport->recv_mutex); 2385 } 2386 2387 static void xs_wake_pending(struct sock_xprt *transport) 2388 { 2389 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state)) 2390 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN); 2391 } 2392 2393 static void xs_error_handle(struct work_struct *work) 2394 { 2395 struct sock_xprt *transport = container_of(work, 2396 struct sock_xprt, error_worker); 2397 2398 xs_wake_disconnect(transport); 2399 xs_wake_write(transport); 2400 xs_wake_error(transport); 2401 xs_wake_pending(transport); 2402 } 2403 2404 /** 2405 * xs_local_print_stats - display AF_LOCAL socket-specifc stats 2406 * @xprt: rpc_xprt struct containing statistics 2407 * @seq: output file 2408 * 2409 */ 2410 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2411 { 2412 long idle_time = 0; 2413 2414 if (xprt_connected(xprt)) 2415 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2416 2417 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2418 "%llu %llu %lu %llu %llu\n", 2419 xprt->stat.bind_count, 2420 xprt->stat.connect_count, 2421 xprt->stat.connect_time / HZ, 2422 idle_time, 2423 xprt->stat.sends, 2424 xprt->stat.recvs, 2425 xprt->stat.bad_xids, 2426 xprt->stat.req_u, 2427 xprt->stat.bklog_u, 2428 xprt->stat.max_slots, 2429 xprt->stat.sending_u, 2430 xprt->stat.pending_u); 2431 } 2432 2433 /** 2434 * xs_udp_print_stats - display UDP socket-specifc stats 2435 * @xprt: rpc_xprt struct containing statistics 2436 * @seq: output file 2437 * 2438 */ 2439 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2440 { 2441 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2442 2443 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2444 "%lu %llu %llu\n", 2445 transport->srcport, 2446 xprt->stat.bind_count, 2447 xprt->stat.sends, 2448 xprt->stat.recvs, 2449 xprt->stat.bad_xids, 2450 xprt->stat.req_u, 2451 xprt->stat.bklog_u, 2452 xprt->stat.max_slots, 2453 xprt->stat.sending_u, 2454 xprt->stat.pending_u); 2455 } 2456 2457 /** 2458 * xs_tcp_print_stats - display TCP socket-specifc stats 2459 * @xprt: rpc_xprt struct containing statistics 2460 * @seq: output file 2461 * 2462 */ 2463 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2464 { 2465 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2466 long idle_time = 0; 2467 2468 if (xprt_connected(xprt)) 2469 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2470 2471 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2472 "%llu %llu %lu %llu %llu\n", 2473 transport->srcport, 2474 xprt->stat.bind_count, 2475 xprt->stat.connect_count, 2476 xprt->stat.connect_time / HZ, 2477 idle_time, 2478 xprt->stat.sends, 2479 xprt->stat.recvs, 2480 xprt->stat.bad_xids, 2481 xprt->stat.req_u, 2482 xprt->stat.bklog_u, 2483 xprt->stat.max_slots, 2484 xprt->stat.sending_u, 2485 xprt->stat.pending_u); 2486 } 2487 2488 /* 2489 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2490 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2491 * to use the server side send routines. 2492 */ 2493 static int bc_malloc(struct rpc_task *task) 2494 { 2495 struct rpc_rqst *rqst = task->tk_rqstp; 2496 size_t size = rqst->rq_callsize; 2497 struct page *page; 2498 struct rpc_buffer *buf; 2499 2500 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) { 2501 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n", 2502 size); 2503 return -EINVAL; 2504 } 2505 2506 page = alloc_page(GFP_KERNEL); 2507 if (!page) 2508 return -ENOMEM; 2509 2510 buf = page_address(page); 2511 buf->len = PAGE_SIZE; 2512 2513 rqst->rq_buffer = buf->data; 2514 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; 2515 return 0; 2516 } 2517 2518 /* 2519 * Free the space allocated in the bc_alloc routine 2520 */ 2521 static void bc_free(struct rpc_task *task) 2522 { 2523 void *buffer = task->tk_rqstp->rq_buffer; 2524 struct rpc_buffer *buf; 2525 2526 buf = container_of(buffer, struct rpc_buffer, data); 2527 free_page((unsigned long)buf); 2528 } 2529 2530 static int bc_sendto(struct rpc_rqst *req) 2531 { 2532 struct xdr_buf *xdr = &req->rq_snd_buf; 2533 struct sock_xprt *transport = 2534 container_of(req->rq_xprt, struct sock_xprt, xprt); 2535 struct msghdr msg = { 2536 .msg_flags = 0, 2537 }; 2538 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 2539 (u32)xdr->len); 2540 unsigned int sent = 0; 2541 int err; 2542 2543 req->rq_xtime = ktime_get(); 2544 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent); 2545 xdr_free_bvec(xdr); 2546 if (err < 0 || sent != (xdr->len + sizeof(marker))) 2547 return -EAGAIN; 2548 return sent; 2549 } 2550 2551 /* 2552 * The send routine. Borrows from svc_send 2553 */ 2554 static int bc_send_request(struct rpc_rqst *req) 2555 { 2556 struct svc_xprt *xprt; 2557 int len; 2558 2559 /* 2560 * Get the server socket associated with this callback xprt 2561 */ 2562 xprt = req->rq_xprt->bc_xprt; 2563 2564 /* 2565 * Grab the mutex to serialize data as the connection is shared 2566 * with the fore channel 2567 */ 2568 mutex_lock(&xprt->xpt_mutex); 2569 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2570 len = -ENOTCONN; 2571 else 2572 len = bc_sendto(req); 2573 mutex_unlock(&xprt->xpt_mutex); 2574 2575 if (len > 0) 2576 len = 0; 2577 2578 return len; 2579 } 2580 2581 /* 2582 * The close routine. Since this is client initiated, we do nothing 2583 */ 2584 2585 static void bc_close(struct rpc_xprt *xprt) 2586 { 2587 xprt_disconnect_done(xprt); 2588 } 2589 2590 /* 2591 * The xprt destroy routine. Again, because this connection is client 2592 * initiated, we do nothing 2593 */ 2594 2595 static void bc_destroy(struct rpc_xprt *xprt) 2596 { 2597 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2598 2599 xs_xprt_free(xprt); 2600 module_put(THIS_MODULE); 2601 } 2602 2603 static const struct rpc_xprt_ops xs_local_ops = { 2604 .reserve_xprt = xprt_reserve_xprt, 2605 .release_xprt = xprt_release_xprt, 2606 .alloc_slot = xprt_alloc_slot, 2607 .free_slot = xprt_free_slot, 2608 .rpcbind = xs_local_rpcbind, 2609 .set_port = xs_local_set_port, 2610 .connect = xs_local_connect, 2611 .buf_alloc = rpc_malloc, 2612 .buf_free = rpc_free, 2613 .prepare_request = xs_stream_prepare_request, 2614 .send_request = xs_local_send_request, 2615 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2616 .close = xs_close, 2617 .destroy = xs_destroy, 2618 .print_stats = xs_local_print_stats, 2619 .enable_swap = xs_enable_swap, 2620 .disable_swap = xs_disable_swap, 2621 }; 2622 2623 static const struct rpc_xprt_ops xs_udp_ops = { 2624 .set_buffer_size = xs_udp_set_buffer_size, 2625 .reserve_xprt = xprt_reserve_xprt_cong, 2626 .release_xprt = xprt_release_xprt_cong, 2627 .alloc_slot = xprt_alloc_slot, 2628 .free_slot = xprt_free_slot, 2629 .rpcbind = rpcb_getport_async, 2630 .set_port = xs_set_port, 2631 .connect = xs_connect, 2632 .buf_alloc = rpc_malloc, 2633 .buf_free = rpc_free, 2634 .send_request = xs_udp_send_request, 2635 .wait_for_reply_request = xprt_wait_for_reply_request_rtt, 2636 .timer = xs_udp_timer, 2637 .release_request = xprt_release_rqst_cong, 2638 .close = xs_close, 2639 .destroy = xs_destroy, 2640 .print_stats = xs_udp_print_stats, 2641 .enable_swap = xs_enable_swap, 2642 .disable_swap = xs_disable_swap, 2643 .inject_disconnect = xs_inject_disconnect, 2644 }; 2645 2646 static const struct rpc_xprt_ops xs_tcp_ops = { 2647 .reserve_xprt = xprt_reserve_xprt, 2648 .release_xprt = xprt_release_xprt, 2649 .alloc_slot = xprt_alloc_slot, 2650 .free_slot = xprt_free_slot, 2651 .rpcbind = rpcb_getport_async, 2652 .set_port = xs_set_port, 2653 .connect = xs_connect, 2654 .buf_alloc = rpc_malloc, 2655 .buf_free = rpc_free, 2656 .prepare_request = xs_stream_prepare_request, 2657 .send_request = xs_tcp_send_request, 2658 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2659 .close = xs_tcp_shutdown, 2660 .destroy = xs_destroy, 2661 .set_connect_timeout = xs_tcp_set_connect_timeout, 2662 .print_stats = xs_tcp_print_stats, 2663 .enable_swap = xs_enable_swap, 2664 .disable_swap = xs_disable_swap, 2665 .inject_disconnect = xs_inject_disconnect, 2666 #ifdef CONFIG_SUNRPC_BACKCHANNEL 2667 .bc_setup = xprt_setup_bc, 2668 .bc_maxpayload = xs_tcp_bc_maxpayload, 2669 .bc_num_slots = xprt_bc_max_slots, 2670 .bc_free_rqst = xprt_free_bc_rqst, 2671 .bc_destroy = xprt_destroy_bc, 2672 #endif 2673 }; 2674 2675 /* 2676 * The rpc_xprt_ops for the server backchannel 2677 */ 2678 2679 static const struct rpc_xprt_ops bc_tcp_ops = { 2680 .reserve_xprt = xprt_reserve_xprt, 2681 .release_xprt = xprt_release_xprt, 2682 .alloc_slot = xprt_alloc_slot, 2683 .free_slot = xprt_free_slot, 2684 .buf_alloc = bc_malloc, 2685 .buf_free = bc_free, 2686 .send_request = bc_send_request, 2687 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2688 .close = bc_close, 2689 .destroy = bc_destroy, 2690 .print_stats = xs_tcp_print_stats, 2691 .enable_swap = xs_enable_swap, 2692 .disable_swap = xs_disable_swap, 2693 .inject_disconnect = xs_inject_disconnect, 2694 }; 2695 2696 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2697 { 2698 static const struct sockaddr_in sin = { 2699 .sin_family = AF_INET, 2700 .sin_addr.s_addr = htonl(INADDR_ANY), 2701 }; 2702 static const struct sockaddr_in6 sin6 = { 2703 .sin6_family = AF_INET6, 2704 .sin6_addr = IN6ADDR_ANY_INIT, 2705 }; 2706 2707 switch (family) { 2708 case AF_LOCAL: 2709 break; 2710 case AF_INET: 2711 memcpy(sap, &sin, sizeof(sin)); 2712 break; 2713 case AF_INET6: 2714 memcpy(sap, &sin6, sizeof(sin6)); 2715 break; 2716 default: 2717 dprintk("RPC: %s: Bad address family\n", __func__); 2718 return -EAFNOSUPPORT; 2719 } 2720 return 0; 2721 } 2722 2723 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2724 unsigned int slot_table_size, 2725 unsigned int max_slot_table_size) 2726 { 2727 struct rpc_xprt *xprt; 2728 struct sock_xprt *new; 2729 2730 if (args->addrlen > sizeof(xprt->addr)) { 2731 dprintk("RPC: xs_setup_xprt: address too large\n"); 2732 return ERR_PTR(-EBADF); 2733 } 2734 2735 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2736 max_slot_table_size); 2737 if (xprt == NULL) { 2738 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2739 "rpc_xprt\n"); 2740 return ERR_PTR(-ENOMEM); 2741 } 2742 2743 new = container_of(xprt, struct sock_xprt, xprt); 2744 mutex_init(&new->recv_mutex); 2745 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2746 xprt->addrlen = args->addrlen; 2747 if (args->srcaddr) 2748 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2749 else { 2750 int err; 2751 err = xs_init_anyaddr(args->dstaddr->sa_family, 2752 (struct sockaddr *)&new->srcaddr); 2753 if (err != 0) { 2754 xprt_free(xprt); 2755 return ERR_PTR(err); 2756 } 2757 } 2758 2759 return xprt; 2760 } 2761 2762 static const struct rpc_timeout xs_local_default_timeout = { 2763 .to_initval = 10 * HZ, 2764 .to_maxval = 10 * HZ, 2765 .to_retries = 2, 2766 }; 2767 2768 /** 2769 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2770 * @args: rpc transport creation arguments 2771 * 2772 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2773 */ 2774 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2775 { 2776 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2777 struct sock_xprt *transport; 2778 struct rpc_xprt *xprt; 2779 struct rpc_xprt *ret; 2780 2781 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2782 xprt_max_tcp_slot_table_entries); 2783 if (IS_ERR(xprt)) 2784 return xprt; 2785 transport = container_of(xprt, struct sock_xprt, xprt); 2786 2787 xprt->prot = 0; 2788 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2789 2790 xprt->bind_timeout = XS_BIND_TO; 2791 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2792 xprt->idle_timeout = XS_IDLE_DISC_TO; 2793 2794 xprt->ops = &xs_local_ops; 2795 xprt->timeout = &xs_local_default_timeout; 2796 2797 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 2798 INIT_WORK(&transport->error_worker, xs_error_handle); 2799 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket); 2800 2801 switch (sun->sun_family) { 2802 case AF_LOCAL: 2803 if (sun->sun_path[0] != '/') { 2804 dprintk("RPC: bad AF_LOCAL address: %s\n", 2805 sun->sun_path); 2806 ret = ERR_PTR(-EINVAL); 2807 goto out_err; 2808 } 2809 xprt_set_bound(xprt); 2810 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2811 ret = ERR_PTR(xs_local_setup_socket(transport)); 2812 if (ret) 2813 goto out_err; 2814 break; 2815 default: 2816 ret = ERR_PTR(-EAFNOSUPPORT); 2817 goto out_err; 2818 } 2819 2820 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2821 xprt->address_strings[RPC_DISPLAY_ADDR]); 2822 2823 if (try_module_get(THIS_MODULE)) 2824 return xprt; 2825 ret = ERR_PTR(-EINVAL); 2826 out_err: 2827 xs_xprt_free(xprt); 2828 return ret; 2829 } 2830 2831 static const struct rpc_timeout xs_udp_default_timeout = { 2832 .to_initval = 5 * HZ, 2833 .to_maxval = 30 * HZ, 2834 .to_increment = 5 * HZ, 2835 .to_retries = 5, 2836 }; 2837 2838 /** 2839 * xs_setup_udp - Set up transport to use a UDP socket 2840 * @args: rpc transport creation arguments 2841 * 2842 */ 2843 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2844 { 2845 struct sockaddr *addr = args->dstaddr; 2846 struct rpc_xprt *xprt; 2847 struct sock_xprt *transport; 2848 struct rpc_xprt *ret; 2849 2850 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2851 xprt_udp_slot_table_entries); 2852 if (IS_ERR(xprt)) 2853 return xprt; 2854 transport = container_of(xprt, struct sock_xprt, xprt); 2855 2856 xprt->prot = IPPROTO_UDP; 2857 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2858 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2859 2860 xprt->bind_timeout = XS_BIND_TO; 2861 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2862 xprt->idle_timeout = XS_IDLE_DISC_TO; 2863 2864 xprt->ops = &xs_udp_ops; 2865 2866 xprt->timeout = &xs_udp_default_timeout; 2867 2868 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 2869 INIT_WORK(&transport->error_worker, xs_error_handle); 2870 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 2871 2872 switch (addr->sa_family) { 2873 case AF_INET: 2874 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2875 xprt_set_bound(xprt); 2876 2877 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2878 break; 2879 case AF_INET6: 2880 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2881 xprt_set_bound(xprt); 2882 2883 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2884 break; 2885 default: 2886 ret = ERR_PTR(-EAFNOSUPPORT); 2887 goto out_err; 2888 } 2889 2890 if (xprt_bound(xprt)) 2891 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2892 xprt->address_strings[RPC_DISPLAY_ADDR], 2893 xprt->address_strings[RPC_DISPLAY_PORT], 2894 xprt->address_strings[RPC_DISPLAY_PROTO]); 2895 else 2896 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2897 xprt->address_strings[RPC_DISPLAY_ADDR], 2898 xprt->address_strings[RPC_DISPLAY_PROTO]); 2899 2900 if (try_module_get(THIS_MODULE)) 2901 return xprt; 2902 ret = ERR_PTR(-EINVAL); 2903 out_err: 2904 xs_xprt_free(xprt); 2905 return ret; 2906 } 2907 2908 static const struct rpc_timeout xs_tcp_default_timeout = { 2909 .to_initval = 60 * HZ, 2910 .to_maxval = 60 * HZ, 2911 .to_retries = 2, 2912 }; 2913 2914 /** 2915 * xs_setup_tcp - Set up transport to use a TCP socket 2916 * @args: rpc transport creation arguments 2917 * 2918 */ 2919 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2920 { 2921 struct sockaddr *addr = args->dstaddr; 2922 struct rpc_xprt *xprt; 2923 struct sock_xprt *transport; 2924 struct rpc_xprt *ret; 2925 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2926 2927 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2928 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2929 2930 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2931 max_slot_table_size); 2932 if (IS_ERR(xprt)) 2933 return xprt; 2934 transport = container_of(xprt, struct sock_xprt, xprt); 2935 2936 xprt->prot = IPPROTO_TCP; 2937 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2938 2939 xprt->bind_timeout = XS_BIND_TO; 2940 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2941 xprt->idle_timeout = XS_IDLE_DISC_TO; 2942 2943 xprt->ops = &xs_tcp_ops; 2944 xprt->timeout = &xs_tcp_default_timeout; 2945 2946 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 2947 xprt->connect_timeout = xprt->timeout->to_initval * 2948 (xprt->timeout->to_retries + 1); 2949 2950 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 2951 INIT_WORK(&transport->error_worker, xs_error_handle); 2952 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 2953 2954 switch (addr->sa_family) { 2955 case AF_INET: 2956 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2957 xprt_set_bound(xprt); 2958 2959 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2960 break; 2961 case AF_INET6: 2962 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2963 xprt_set_bound(xprt); 2964 2965 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 2966 break; 2967 default: 2968 ret = ERR_PTR(-EAFNOSUPPORT); 2969 goto out_err; 2970 } 2971 2972 if (xprt_bound(xprt)) 2973 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2974 xprt->address_strings[RPC_DISPLAY_ADDR], 2975 xprt->address_strings[RPC_DISPLAY_PORT], 2976 xprt->address_strings[RPC_DISPLAY_PROTO]); 2977 else 2978 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2979 xprt->address_strings[RPC_DISPLAY_ADDR], 2980 xprt->address_strings[RPC_DISPLAY_PROTO]); 2981 2982 if (try_module_get(THIS_MODULE)) 2983 return xprt; 2984 ret = ERR_PTR(-EINVAL); 2985 out_err: 2986 xs_xprt_free(xprt); 2987 return ret; 2988 } 2989 2990 /** 2991 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 2992 * @args: rpc transport creation arguments 2993 * 2994 */ 2995 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 2996 { 2997 struct sockaddr *addr = args->dstaddr; 2998 struct rpc_xprt *xprt; 2999 struct sock_xprt *transport; 3000 struct svc_sock *bc_sock; 3001 struct rpc_xprt *ret; 3002 3003 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3004 xprt_tcp_slot_table_entries); 3005 if (IS_ERR(xprt)) 3006 return xprt; 3007 transport = container_of(xprt, struct sock_xprt, xprt); 3008 3009 xprt->prot = IPPROTO_TCP; 3010 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3011 xprt->timeout = &xs_tcp_default_timeout; 3012 3013 /* backchannel */ 3014 xprt_set_bound(xprt); 3015 xprt->bind_timeout = 0; 3016 xprt->reestablish_timeout = 0; 3017 xprt->idle_timeout = 0; 3018 3019 xprt->ops = &bc_tcp_ops; 3020 3021 switch (addr->sa_family) { 3022 case AF_INET: 3023 xs_format_peer_addresses(xprt, "tcp", 3024 RPCBIND_NETID_TCP); 3025 break; 3026 case AF_INET6: 3027 xs_format_peer_addresses(xprt, "tcp", 3028 RPCBIND_NETID_TCP6); 3029 break; 3030 default: 3031 ret = ERR_PTR(-EAFNOSUPPORT); 3032 goto out_err; 3033 } 3034 3035 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3036 xprt->address_strings[RPC_DISPLAY_ADDR], 3037 xprt->address_strings[RPC_DISPLAY_PORT], 3038 xprt->address_strings[RPC_DISPLAY_PROTO]); 3039 3040 /* 3041 * Once we've associated a backchannel xprt with a connection, 3042 * we want to keep it around as long as the connection lasts, 3043 * in case we need to start using it for a backchannel again; 3044 * this reference won't be dropped until bc_xprt is destroyed. 3045 */ 3046 xprt_get(xprt); 3047 args->bc_xprt->xpt_bc_xprt = xprt; 3048 xprt->bc_xprt = args->bc_xprt; 3049 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3050 transport->sock = bc_sock->sk_sock; 3051 transport->inet = bc_sock->sk_sk; 3052 3053 /* 3054 * Since we don't want connections for the backchannel, we set 3055 * the xprt status to connected 3056 */ 3057 xprt_set_connected(xprt); 3058 3059 if (try_module_get(THIS_MODULE)) 3060 return xprt; 3061 3062 args->bc_xprt->xpt_bc_xprt = NULL; 3063 args->bc_xprt->xpt_bc_xps = NULL; 3064 xprt_put(xprt); 3065 ret = ERR_PTR(-EINVAL); 3066 out_err: 3067 xs_xprt_free(xprt); 3068 return ret; 3069 } 3070 3071 static struct xprt_class xs_local_transport = { 3072 .list = LIST_HEAD_INIT(xs_local_transport.list), 3073 .name = "named UNIX socket", 3074 .owner = THIS_MODULE, 3075 .ident = XPRT_TRANSPORT_LOCAL, 3076 .setup = xs_setup_local, 3077 }; 3078 3079 static struct xprt_class xs_udp_transport = { 3080 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3081 .name = "udp", 3082 .owner = THIS_MODULE, 3083 .ident = XPRT_TRANSPORT_UDP, 3084 .setup = xs_setup_udp, 3085 }; 3086 3087 static struct xprt_class xs_tcp_transport = { 3088 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3089 .name = "tcp", 3090 .owner = THIS_MODULE, 3091 .ident = XPRT_TRANSPORT_TCP, 3092 .setup = xs_setup_tcp, 3093 }; 3094 3095 static struct xprt_class xs_bc_tcp_transport = { 3096 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3097 .name = "tcp NFSv4.1 backchannel", 3098 .owner = THIS_MODULE, 3099 .ident = XPRT_TRANSPORT_BC_TCP, 3100 .setup = xs_setup_bc_tcp, 3101 }; 3102 3103 /** 3104 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3105 * 3106 */ 3107 int init_socket_xprt(void) 3108 { 3109 if (!sunrpc_table_header) 3110 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3111 3112 xprt_register_transport(&xs_local_transport); 3113 xprt_register_transport(&xs_udp_transport); 3114 xprt_register_transport(&xs_tcp_transport); 3115 xprt_register_transport(&xs_bc_tcp_transport); 3116 3117 return 0; 3118 } 3119 3120 /** 3121 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3122 * 3123 */ 3124 void cleanup_socket_xprt(void) 3125 { 3126 if (sunrpc_table_header) { 3127 unregister_sysctl_table(sunrpc_table_header); 3128 sunrpc_table_header = NULL; 3129 } 3130 3131 xprt_unregister_transport(&xs_local_transport); 3132 xprt_unregister_transport(&xs_udp_transport); 3133 xprt_unregister_transport(&xs_tcp_transport); 3134 xprt_unregister_transport(&xs_bc_tcp_transport); 3135 } 3136 3137 static int param_set_uint_minmax(const char *val, 3138 const struct kernel_param *kp, 3139 unsigned int min, unsigned int max) 3140 { 3141 unsigned int num; 3142 int ret; 3143 3144 if (!val) 3145 return -EINVAL; 3146 ret = kstrtouint(val, 0, &num); 3147 if (ret) 3148 return ret; 3149 if (num < min || num > max) 3150 return -EINVAL; 3151 *((unsigned int *)kp->arg) = num; 3152 return 0; 3153 } 3154 3155 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3156 { 3157 return param_set_uint_minmax(val, kp, 3158 RPC_MIN_RESVPORT, 3159 RPC_MAX_RESVPORT); 3160 } 3161 3162 static const struct kernel_param_ops param_ops_portnr = { 3163 .set = param_set_portnr, 3164 .get = param_get_uint, 3165 }; 3166 3167 #define param_check_portnr(name, p) \ 3168 __param_check(name, p, unsigned int); 3169 3170 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3171 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3172 3173 static int param_set_slot_table_size(const char *val, 3174 const struct kernel_param *kp) 3175 { 3176 return param_set_uint_minmax(val, kp, 3177 RPC_MIN_SLOT_TABLE, 3178 RPC_MAX_SLOT_TABLE); 3179 } 3180 3181 static const struct kernel_param_ops param_ops_slot_table_size = { 3182 .set = param_set_slot_table_size, 3183 .get = param_get_uint, 3184 }; 3185 3186 #define param_check_slot_table_size(name, p) \ 3187 __param_check(name, p, unsigned int); 3188 3189 static int param_set_max_slot_table_size(const char *val, 3190 const struct kernel_param *kp) 3191 { 3192 return param_set_uint_minmax(val, kp, 3193 RPC_MIN_SLOT_TABLE, 3194 RPC_MAX_SLOT_TABLE_LIMIT); 3195 } 3196 3197 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3198 .set = param_set_max_slot_table_size, 3199 .get = param_get_uint, 3200 }; 3201 3202 #define param_check_max_slot_table_size(name, p) \ 3203 __param_check(name, p, unsigned int); 3204 3205 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3206 slot_table_size, 0644); 3207 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3208 max_slot_table_size, 0644); 3209 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3210 slot_table_size, 0644); 3211