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_set_srcport(struct sock_xprt *transport, struct socket *sock); 62 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 63 struct socket *sock); 64 65 /* 66 * xprtsock tunables 67 */ 68 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 69 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 70 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 71 72 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 73 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 74 75 #define XS_TCP_LINGER_TO (15U * HZ) 76 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 77 78 /* 79 * We can register our own files under /proc/sys/sunrpc by 80 * calling register_sysctl_table() again. The files in that 81 * directory become the union of all files registered there. 82 * 83 * We simply need to make sure that we don't collide with 84 * someone else's file names! 85 */ 86 87 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 88 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 89 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 90 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 91 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 92 93 static struct ctl_table_header *sunrpc_table_header; 94 95 static struct xprt_class xs_local_transport; 96 static struct xprt_class xs_udp_transport; 97 static struct xprt_class xs_tcp_transport; 98 static struct xprt_class xs_bc_tcp_transport; 99 100 /* 101 * FIXME: changing the UDP slot table size should also resize the UDP 102 * socket buffers for existing UDP transports 103 */ 104 static struct ctl_table xs_tunables_table[] = { 105 { 106 .procname = "udp_slot_table_entries", 107 .data = &xprt_udp_slot_table_entries, 108 .maxlen = sizeof(unsigned int), 109 .mode = 0644, 110 .proc_handler = proc_dointvec_minmax, 111 .extra1 = &min_slot_table_size, 112 .extra2 = &max_slot_table_size 113 }, 114 { 115 .procname = "tcp_slot_table_entries", 116 .data = &xprt_tcp_slot_table_entries, 117 .maxlen = sizeof(unsigned int), 118 .mode = 0644, 119 .proc_handler = proc_dointvec_minmax, 120 .extra1 = &min_slot_table_size, 121 .extra2 = &max_slot_table_size 122 }, 123 { 124 .procname = "tcp_max_slot_table_entries", 125 .data = &xprt_max_tcp_slot_table_entries, 126 .maxlen = sizeof(unsigned int), 127 .mode = 0644, 128 .proc_handler = proc_dointvec_minmax, 129 .extra1 = &min_slot_table_size, 130 .extra2 = &max_tcp_slot_table_limit 131 }, 132 { 133 .procname = "min_resvport", 134 .data = &xprt_min_resvport, 135 .maxlen = sizeof(unsigned int), 136 .mode = 0644, 137 .proc_handler = proc_dointvec_minmax, 138 .extra1 = &xprt_min_resvport_limit, 139 .extra2 = &xprt_max_resvport_limit 140 }, 141 { 142 .procname = "max_resvport", 143 .data = &xprt_max_resvport, 144 .maxlen = sizeof(unsigned int), 145 .mode = 0644, 146 .proc_handler = proc_dointvec_minmax, 147 .extra1 = &xprt_min_resvport_limit, 148 .extra2 = &xprt_max_resvport_limit 149 }, 150 { 151 .procname = "tcp_fin_timeout", 152 .data = &xs_tcp_fin_timeout, 153 .maxlen = sizeof(xs_tcp_fin_timeout), 154 .mode = 0644, 155 .proc_handler = proc_dointvec_jiffies, 156 }, 157 { }, 158 }; 159 160 static struct ctl_table sunrpc_table[] = { 161 { 162 .procname = "sunrpc", 163 .mode = 0555, 164 .child = xs_tunables_table 165 }, 166 { }, 167 }; 168 169 /* 170 * Wait duration for a reply from the RPC portmapper. 171 */ 172 #define XS_BIND_TO (60U * HZ) 173 174 /* 175 * Delay if a UDP socket connect error occurs. This is most likely some 176 * kind of resource problem on the local host. 177 */ 178 #define XS_UDP_REEST_TO (2U * HZ) 179 180 /* 181 * The reestablish timeout allows clients to delay for a bit before attempting 182 * to reconnect to a server that just dropped our connection. 183 * 184 * We implement an exponential backoff when trying to reestablish a TCP 185 * transport connection with the server. Some servers like to drop a TCP 186 * connection when they are overworked, so we start with a short timeout and 187 * increase over time if the server is down or not responding. 188 */ 189 #define XS_TCP_INIT_REEST_TO (3U * HZ) 190 191 /* 192 * TCP idle timeout; client drops the transport socket if it is idle 193 * for this long. Note that we also timeout UDP sockets to prevent 194 * holding port numbers when there is no RPC traffic. 195 */ 196 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 197 198 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 199 # undef RPC_DEBUG_DATA 200 # define RPCDBG_FACILITY RPCDBG_TRANS 201 #endif 202 203 #ifdef RPC_DEBUG_DATA 204 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 205 { 206 u8 *buf = (u8 *) packet; 207 int j; 208 209 dprintk("RPC: %s\n", msg); 210 for (j = 0; j < count && j < 128; j += 4) { 211 if (!(j & 31)) { 212 if (j) 213 dprintk("\n"); 214 dprintk("0x%04x ", j); 215 } 216 dprintk("%02x%02x%02x%02x ", 217 buf[j], buf[j+1], buf[j+2], buf[j+3]); 218 } 219 dprintk("\n"); 220 } 221 #else 222 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 223 { 224 /* NOP */ 225 } 226 #endif 227 228 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 229 { 230 return (struct rpc_xprt *) sk->sk_user_data; 231 } 232 233 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 234 { 235 return (struct sockaddr *) &xprt->addr; 236 } 237 238 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 239 { 240 return (struct sockaddr_un *) &xprt->addr; 241 } 242 243 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 244 { 245 return (struct sockaddr_in *) &xprt->addr; 246 } 247 248 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 249 { 250 return (struct sockaddr_in6 *) &xprt->addr; 251 } 252 253 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 254 { 255 struct sockaddr *sap = xs_addr(xprt); 256 struct sockaddr_in6 *sin6; 257 struct sockaddr_in *sin; 258 struct sockaddr_un *sun; 259 char buf[128]; 260 261 switch (sap->sa_family) { 262 case AF_LOCAL: 263 sun = xs_addr_un(xprt); 264 strlcpy(buf, sun->sun_path, sizeof(buf)); 265 xprt->address_strings[RPC_DISPLAY_ADDR] = 266 kstrdup(buf, GFP_KERNEL); 267 break; 268 case AF_INET: 269 (void)rpc_ntop(sap, buf, sizeof(buf)); 270 xprt->address_strings[RPC_DISPLAY_ADDR] = 271 kstrdup(buf, GFP_KERNEL); 272 sin = xs_addr_in(xprt); 273 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 274 break; 275 case AF_INET6: 276 (void)rpc_ntop(sap, buf, sizeof(buf)); 277 xprt->address_strings[RPC_DISPLAY_ADDR] = 278 kstrdup(buf, GFP_KERNEL); 279 sin6 = xs_addr_in6(xprt); 280 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 281 break; 282 default: 283 BUG(); 284 } 285 286 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 287 } 288 289 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 290 { 291 struct sockaddr *sap = xs_addr(xprt); 292 char buf[128]; 293 294 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 295 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 296 297 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 298 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 299 } 300 301 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 302 const char *protocol, 303 const char *netid) 304 { 305 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 306 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 307 xs_format_common_peer_addresses(xprt); 308 xs_format_common_peer_ports(xprt); 309 } 310 311 static void xs_update_peer_port(struct rpc_xprt *xprt) 312 { 313 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 314 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 315 316 xs_format_common_peer_ports(xprt); 317 } 318 319 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 320 { 321 unsigned int i; 322 323 for (i = 0; i < RPC_DISPLAY_MAX; i++) 324 switch (i) { 325 case RPC_DISPLAY_PROTO: 326 case RPC_DISPLAY_NETID: 327 continue; 328 default: 329 kfree(xprt->address_strings[i]); 330 } 331 } 332 333 static size_t 334 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp) 335 { 336 size_t i,n; 337 338 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES)) 339 return want; 340 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT; 341 for (i = 0; i < n; i++) { 342 if (buf->pages[i]) 343 continue; 344 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp); 345 if (!buf->pages[i]) { 346 i *= PAGE_SIZE; 347 return i > buf->page_base ? i - buf->page_base : 0; 348 } 349 } 350 return want; 351 } 352 353 static ssize_t 354 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek) 355 { 356 ssize_t ret; 357 if (seek != 0) 358 iov_iter_advance(&msg->msg_iter, seek); 359 ret = sock_recvmsg(sock, msg, flags); 360 return ret > 0 ? ret + seek : ret; 361 } 362 363 static ssize_t 364 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags, 365 struct kvec *kvec, size_t count, size_t seek) 366 { 367 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count); 368 return xs_sock_recvmsg(sock, msg, flags, seek); 369 } 370 371 static ssize_t 372 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags, 373 struct bio_vec *bvec, unsigned long nr, size_t count, 374 size_t seek) 375 { 376 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count); 377 return xs_sock_recvmsg(sock, msg, flags, seek); 378 } 379 380 static ssize_t 381 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags, 382 size_t count) 383 { 384 iov_iter_discard(&msg->msg_iter, READ, count); 385 return sock_recvmsg(sock, msg, flags); 386 } 387 388 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 389 static void 390 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 391 { 392 struct bvec_iter bi = { 393 .bi_size = count, 394 }; 395 struct bio_vec bv; 396 397 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); 398 for_each_bvec(bv, bvec, bi, bi) 399 flush_dcache_page(bv.bv_page); 400 } 401 #else 402 static inline void 403 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 404 { 405 } 406 #endif 407 408 static ssize_t 409 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags, 410 struct xdr_buf *buf, size_t count, size_t seek, size_t *read) 411 { 412 size_t want, seek_init = seek, offset = 0; 413 ssize_t ret; 414 415 want = min_t(size_t, count, buf->head[0].iov_len); 416 if (seek < want) { 417 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek); 418 if (ret <= 0) 419 goto sock_err; 420 offset += ret; 421 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 422 goto out; 423 if (ret != want) 424 goto out; 425 seek = 0; 426 } else { 427 seek -= want; 428 offset += want; 429 } 430 431 want = xs_alloc_sparse_pages(buf, 432 min_t(size_t, count - offset, buf->page_len), 433 GFP_KERNEL); 434 if (seek < want) { 435 ret = xs_read_bvec(sock, msg, flags, buf->bvec, 436 xdr_buf_pagecount(buf), 437 want + buf->page_base, 438 seek + buf->page_base); 439 if (ret <= 0) 440 goto sock_err; 441 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base); 442 ret -= buf->page_base; 443 offset += ret; 444 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 445 goto out; 446 if (ret != want) 447 goto out; 448 seek = 0; 449 } else { 450 seek -= want; 451 offset += want; 452 } 453 454 want = min_t(size_t, count - offset, buf->tail[0].iov_len); 455 if (seek < want) { 456 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek); 457 if (ret <= 0) 458 goto sock_err; 459 offset += ret; 460 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 461 goto out; 462 if (ret != want) 463 goto out; 464 } else if (offset < seek_init) 465 offset = seek_init; 466 ret = -EMSGSIZE; 467 out: 468 *read = offset - seek_init; 469 return ret; 470 sock_err: 471 offset += seek; 472 goto out; 473 } 474 475 static void 476 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf) 477 { 478 if (!transport->recv.copied) { 479 if (buf->head[0].iov_len >= transport->recv.offset) 480 memcpy(buf->head[0].iov_base, 481 &transport->recv.xid, 482 transport->recv.offset); 483 transport->recv.copied = transport->recv.offset; 484 } 485 } 486 487 static bool 488 xs_read_stream_request_done(struct sock_xprt *transport) 489 { 490 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT); 491 } 492 493 static void 494 xs_read_stream_check_eor(struct sock_xprt *transport, 495 struct msghdr *msg) 496 { 497 if (xs_read_stream_request_done(transport)) 498 msg->msg_flags |= MSG_EOR; 499 } 500 501 static ssize_t 502 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg, 503 int flags, struct rpc_rqst *req) 504 { 505 struct xdr_buf *buf = &req->rq_private_buf; 506 size_t want, read; 507 ssize_t ret; 508 509 xs_read_header(transport, buf); 510 511 want = transport->recv.len - transport->recv.offset; 512 if (want != 0) { 513 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf, 514 transport->recv.copied + want, 515 transport->recv.copied, 516 &read); 517 transport->recv.offset += read; 518 transport->recv.copied += read; 519 } 520 521 if (transport->recv.offset == transport->recv.len) 522 xs_read_stream_check_eor(transport, msg); 523 524 if (want == 0) 525 return 0; 526 527 switch (ret) { 528 default: 529 break; 530 case -EFAULT: 531 case -EMSGSIZE: 532 msg->msg_flags |= MSG_TRUNC; 533 return read; 534 case 0: 535 return -ESHUTDOWN; 536 } 537 return ret < 0 ? ret : read; 538 } 539 540 static size_t 541 xs_read_stream_headersize(bool isfrag) 542 { 543 if (isfrag) 544 return sizeof(__be32); 545 return 3 * sizeof(__be32); 546 } 547 548 static ssize_t 549 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg, 550 int flags, size_t want, size_t seek) 551 { 552 struct kvec kvec = { 553 .iov_base = &transport->recv.fraghdr, 554 .iov_len = want, 555 }; 556 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek); 557 } 558 559 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 560 static ssize_t 561 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 562 { 563 struct rpc_xprt *xprt = &transport->xprt; 564 struct rpc_rqst *req; 565 ssize_t ret; 566 567 /* Is this transport associated with the backchannel? */ 568 if (!xprt->bc_serv) 569 return -ESHUTDOWN; 570 571 /* Look up and lock the request corresponding to the given XID */ 572 req = xprt_lookup_bc_request(xprt, transport->recv.xid); 573 if (!req) { 574 printk(KERN_WARNING "Callback slot table overflowed\n"); 575 return -ESHUTDOWN; 576 } 577 if (transport->recv.copied && !req->rq_private_buf.len) 578 return -ESHUTDOWN; 579 580 ret = xs_read_stream_request(transport, msg, flags, req); 581 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 582 xprt_complete_bc_request(req, transport->recv.copied); 583 else 584 req->rq_private_buf.len = transport->recv.copied; 585 586 return ret; 587 } 588 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 589 static ssize_t 590 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 591 { 592 return -ESHUTDOWN; 593 } 594 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 595 596 static ssize_t 597 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags) 598 { 599 struct rpc_xprt *xprt = &transport->xprt; 600 struct rpc_rqst *req; 601 ssize_t ret = 0; 602 603 /* Look up and lock the request corresponding to the given XID */ 604 spin_lock(&xprt->queue_lock); 605 req = xprt_lookup_rqst(xprt, transport->recv.xid); 606 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) { 607 msg->msg_flags |= MSG_TRUNC; 608 goto out; 609 } 610 xprt_pin_rqst(req); 611 spin_unlock(&xprt->queue_lock); 612 613 ret = xs_read_stream_request(transport, msg, flags, req); 614 615 spin_lock(&xprt->queue_lock); 616 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 617 xprt_complete_rqst(req->rq_task, transport->recv.copied); 618 else 619 req->rq_private_buf.len = transport->recv.copied; 620 xprt_unpin_rqst(req); 621 out: 622 spin_unlock(&xprt->queue_lock); 623 return ret; 624 } 625 626 static ssize_t 627 xs_read_stream(struct sock_xprt *transport, int flags) 628 { 629 struct msghdr msg = { 0 }; 630 size_t want, read = 0; 631 ssize_t ret = 0; 632 633 if (transport->recv.len == 0) { 634 want = xs_read_stream_headersize(transport->recv.copied != 0); 635 ret = xs_read_stream_header(transport, &msg, flags, want, 636 transport->recv.offset); 637 if (ret <= 0) 638 goto out_err; 639 transport->recv.offset = ret; 640 if (transport->recv.offset != want) 641 return transport->recv.offset; 642 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) & 643 RPC_FRAGMENT_SIZE_MASK; 644 transport->recv.offset -= sizeof(transport->recv.fraghdr); 645 read = ret; 646 } 647 648 switch (be32_to_cpu(transport->recv.calldir)) { 649 default: 650 msg.msg_flags |= MSG_TRUNC; 651 break; 652 case RPC_CALL: 653 ret = xs_read_stream_call(transport, &msg, flags); 654 break; 655 case RPC_REPLY: 656 ret = xs_read_stream_reply(transport, &msg, flags); 657 } 658 if (msg.msg_flags & MSG_TRUNC) { 659 transport->recv.calldir = cpu_to_be32(-1); 660 transport->recv.copied = -1; 661 } 662 if (ret < 0) 663 goto out_err; 664 read += ret; 665 if (transport->recv.offset < transport->recv.len) { 666 if (!(msg.msg_flags & MSG_TRUNC)) 667 return read; 668 msg.msg_flags = 0; 669 ret = xs_read_discard(transport->sock, &msg, flags, 670 transport->recv.len - transport->recv.offset); 671 if (ret <= 0) 672 goto out_err; 673 transport->recv.offset += ret; 674 read += ret; 675 if (transport->recv.offset != transport->recv.len) 676 return read; 677 } 678 if (xs_read_stream_request_done(transport)) { 679 trace_xs_stream_read_request(transport); 680 transport->recv.copied = 0; 681 } 682 transport->recv.offset = 0; 683 transport->recv.len = 0; 684 return read; 685 out_err: 686 return ret != 0 ? ret : -ESHUTDOWN; 687 } 688 689 static __poll_t xs_poll_socket(struct sock_xprt *transport) 690 { 691 return transport->sock->ops->poll(transport->file, transport->sock, 692 NULL); 693 } 694 695 static bool xs_poll_socket_readable(struct sock_xprt *transport) 696 { 697 __poll_t events = xs_poll_socket(transport); 698 699 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP); 700 } 701 702 static void xs_poll_check_readable(struct sock_xprt *transport) 703 { 704 705 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 706 if (!xs_poll_socket_readable(transport)) 707 return; 708 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 709 queue_work(xprtiod_workqueue, &transport->recv_worker); 710 } 711 712 static void xs_stream_data_receive(struct sock_xprt *transport) 713 { 714 size_t read = 0; 715 ssize_t ret = 0; 716 717 mutex_lock(&transport->recv_mutex); 718 if (transport->sock == NULL) 719 goto out; 720 for (;;) { 721 ret = xs_read_stream(transport, MSG_DONTWAIT); 722 if (ret < 0) 723 break; 724 read += ret; 725 cond_resched(); 726 } 727 if (ret == -ESHUTDOWN) 728 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 729 else 730 xs_poll_check_readable(transport); 731 out: 732 mutex_unlock(&transport->recv_mutex); 733 trace_xs_stream_read_data(&transport->xprt, ret, read); 734 } 735 736 static void xs_stream_data_receive_workfn(struct work_struct *work) 737 { 738 struct sock_xprt *transport = 739 container_of(work, struct sock_xprt, recv_worker); 740 unsigned int pflags = memalloc_nofs_save(); 741 742 xs_stream_data_receive(transport); 743 memalloc_nofs_restore(pflags); 744 } 745 746 static void 747 xs_stream_reset_connect(struct sock_xprt *transport) 748 { 749 transport->recv.offset = 0; 750 transport->recv.len = 0; 751 transport->recv.copied = 0; 752 transport->xmit.offset = 0; 753 } 754 755 static void 756 xs_stream_start_connect(struct sock_xprt *transport) 757 { 758 transport->xprt.stat.connect_count++; 759 transport->xprt.stat.connect_start = jiffies; 760 } 761 762 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 763 764 /** 765 * xs_nospace - handle transmit was incomplete 766 * @req: pointer to RPC request 767 * @transport: pointer to struct sock_xprt 768 * 769 */ 770 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport) 771 { 772 struct rpc_xprt *xprt = &transport->xprt; 773 struct sock *sk = transport->inet; 774 int ret = -EAGAIN; 775 776 trace_rpc_socket_nospace(req, transport); 777 778 /* Protect against races with write_space */ 779 spin_lock(&xprt->transport_lock); 780 781 /* Don't race with disconnect */ 782 if (xprt_connected(xprt)) { 783 struct socket_wq *wq; 784 785 rcu_read_lock(); 786 wq = rcu_dereference(sk->sk_wq); 787 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags); 788 rcu_read_unlock(); 789 790 /* wait for more buffer space */ 791 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 792 sk->sk_write_pending++; 793 xprt_wait_for_buffer_space(xprt); 794 } else 795 ret = -ENOTCONN; 796 797 spin_unlock(&xprt->transport_lock); 798 return ret; 799 } 800 801 static int xs_sock_nospace(struct rpc_rqst *req) 802 { 803 struct sock_xprt *transport = 804 container_of(req->rq_xprt, struct sock_xprt, xprt); 805 struct sock *sk = transport->inet; 806 int ret = -EAGAIN; 807 808 lock_sock(sk); 809 if (!sock_writeable(sk)) 810 ret = xs_nospace(req, transport); 811 release_sock(sk); 812 return ret; 813 } 814 815 static int xs_stream_nospace(struct rpc_rqst *req) 816 { 817 struct sock_xprt *transport = 818 container_of(req->rq_xprt, struct sock_xprt, xprt); 819 struct sock *sk = transport->inet; 820 int ret = -EAGAIN; 821 822 lock_sock(sk); 823 if (!sk_stream_memory_free(sk)) 824 ret = xs_nospace(req, transport); 825 release_sock(sk); 826 return ret; 827 } 828 829 static void 830 xs_stream_prepare_request(struct rpc_rqst *req) 831 { 832 xdr_free_bvec(&req->rq_rcv_buf); 833 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL); 834 } 835 836 /* 837 * Determine if the previous message in the stream was aborted before it 838 * could complete transmission. 839 */ 840 static bool 841 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req) 842 { 843 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0; 844 } 845 846 /* 847 * Return the stream record marker field for a record of length < 2^31-1 848 */ 849 static rpc_fraghdr 850 xs_stream_record_marker(struct xdr_buf *xdr) 851 { 852 if (!xdr->len) 853 return 0; 854 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len); 855 } 856 857 /** 858 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 859 * @req: pointer to RPC request 860 * 861 * Return values: 862 * 0: The request has been sent 863 * EAGAIN: The socket was blocked, please call again later to 864 * complete the request 865 * ENOTCONN: Caller needs to invoke connect logic then call again 866 * other: Some other error occurred, the request was not sent 867 */ 868 static int xs_local_send_request(struct rpc_rqst *req) 869 { 870 struct rpc_xprt *xprt = req->rq_xprt; 871 struct sock_xprt *transport = 872 container_of(xprt, struct sock_xprt, xprt); 873 struct xdr_buf *xdr = &req->rq_snd_buf; 874 rpc_fraghdr rm = xs_stream_record_marker(xdr); 875 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 876 struct msghdr msg = { 877 .msg_flags = XS_SENDMSG_FLAGS, 878 }; 879 unsigned int sent; 880 int status; 881 882 /* Close the stream if the previous transmission was incomplete */ 883 if (xs_send_request_was_aborted(transport, req)) { 884 xs_close(xprt); 885 return -ENOTCONN; 886 } 887 888 xs_pktdump("packet data:", 889 req->rq_svec->iov_base, req->rq_svec->iov_len); 890 891 req->rq_xtime = ktime_get(); 892 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 893 transport->xmit.offset, rm, &sent); 894 dprintk("RPC: %s(%u) = %d\n", 895 __func__, xdr->len - transport->xmit.offset, status); 896 897 if (status == -EAGAIN && sock_writeable(transport->inet)) 898 status = -ENOBUFS; 899 900 if (likely(sent > 0) || status == 0) { 901 transport->xmit.offset += sent; 902 req->rq_bytes_sent = transport->xmit.offset; 903 if (likely(req->rq_bytes_sent >= msglen)) { 904 req->rq_xmit_bytes_sent += transport->xmit.offset; 905 transport->xmit.offset = 0; 906 return 0; 907 } 908 status = -EAGAIN; 909 } 910 911 switch (status) { 912 case -ENOBUFS: 913 break; 914 case -EAGAIN: 915 status = xs_stream_nospace(req); 916 break; 917 default: 918 dprintk("RPC: sendmsg returned unrecognized error %d\n", 919 -status); 920 fallthrough; 921 case -EPIPE: 922 xs_close(xprt); 923 status = -ENOTCONN; 924 } 925 926 return status; 927 } 928 929 /** 930 * xs_udp_send_request - write an RPC request to a UDP socket 931 * @req: pointer to RPC request 932 * 933 * Return values: 934 * 0: The request has been sent 935 * EAGAIN: The socket was blocked, please call again later to 936 * complete the request 937 * ENOTCONN: Caller needs to invoke connect logic then call again 938 * other: Some other error occurred, the request was not sent 939 */ 940 static int xs_udp_send_request(struct rpc_rqst *req) 941 { 942 struct rpc_xprt *xprt = req->rq_xprt; 943 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 944 struct xdr_buf *xdr = &req->rq_snd_buf; 945 struct msghdr msg = { 946 .msg_name = xs_addr(xprt), 947 .msg_namelen = xprt->addrlen, 948 .msg_flags = XS_SENDMSG_FLAGS, 949 }; 950 unsigned int sent; 951 int status; 952 953 xs_pktdump("packet data:", 954 req->rq_svec->iov_base, 955 req->rq_svec->iov_len); 956 957 if (!xprt_bound(xprt)) 958 return -ENOTCONN; 959 960 if (!xprt_request_get_cong(xprt, req)) 961 return -EBADSLT; 962 963 req->rq_xtime = ktime_get(); 964 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent); 965 966 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 967 xdr->len, status); 968 969 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 970 if (status == -EPERM) 971 goto process_status; 972 973 if (status == -EAGAIN && sock_writeable(transport->inet)) 974 status = -ENOBUFS; 975 976 if (sent > 0 || status == 0) { 977 req->rq_xmit_bytes_sent += sent; 978 if (sent >= req->rq_slen) 979 return 0; 980 /* Still some bytes left; set up for a retry later. */ 981 status = -EAGAIN; 982 } 983 984 process_status: 985 switch (status) { 986 case -ENOTSOCK: 987 status = -ENOTCONN; 988 /* Should we call xs_close() here? */ 989 break; 990 case -EAGAIN: 991 status = xs_sock_nospace(req); 992 break; 993 case -ENETUNREACH: 994 case -ENOBUFS: 995 case -EPIPE: 996 case -ECONNREFUSED: 997 case -EPERM: 998 /* When the server has died, an ICMP port unreachable message 999 * prompts ECONNREFUSED. */ 1000 break; 1001 default: 1002 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1003 -status); 1004 } 1005 1006 return status; 1007 } 1008 1009 /** 1010 * xs_tcp_send_request - write an RPC request to a TCP socket 1011 * @req: pointer to RPC request 1012 * 1013 * Return values: 1014 * 0: The request has been sent 1015 * EAGAIN: The socket was blocked, please call again later to 1016 * complete the request 1017 * ENOTCONN: Caller needs to invoke connect logic then call again 1018 * other: Some other error occurred, the request was not sent 1019 * 1020 * XXX: In the case of soft timeouts, should we eventually give up 1021 * if sendmsg is not able to make progress? 1022 */ 1023 static int xs_tcp_send_request(struct rpc_rqst *req) 1024 { 1025 struct rpc_xprt *xprt = req->rq_xprt; 1026 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1027 struct xdr_buf *xdr = &req->rq_snd_buf; 1028 rpc_fraghdr rm = xs_stream_record_marker(xdr); 1029 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 1030 struct msghdr msg = { 1031 .msg_flags = XS_SENDMSG_FLAGS, 1032 }; 1033 bool vm_wait = false; 1034 unsigned int sent; 1035 int status; 1036 1037 /* Close the stream if the previous transmission was incomplete */ 1038 if (xs_send_request_was_aborted(transport, req)) { 1039 if (transport->sock != NULL) 1040 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 1041 return -ENOTCONN; 1042 } 1043 if (!transport->inet) 1044 return -ENOTCONN; 1045 1046 xs_pktdump("packet data:", 1047 req->rq_svec->iov_base, 1048 req->rq_svec->iov_len); 1049 1050 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state)) 1051 xs_tcp_set_socket_timeouts(xprt, transport->sock); 1052 1053 xs_set_srcport(transport, transport->sock); 1054 1055 /* Continue transmitting the packet/record. We must be careful 1056 * to cope with writespace callbacks arriving _after_ we have 1057 * called sendmsg(). */ 1058 req->rq_xtime = ktime_get(); 1059 tcp_sock_set_cork(transport->inet, true); 1060 while (1) { 1061 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 1062 transport->xmit.offset, rm, &sent); 1063 1064 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 1065 xdr->len - transport->xmit.offset, status); 1066 1067 /* If we've sent the entire packet, immediately 1068 * reset the count of bytes sent. */ 1069 transport->xmit.offset += sent; 1070 req->rq_bytes_sent = transport->xmit.offset; 1071 if (likely(req->rq_bytes_sent >= msglen)) { 1072 req->rq_xmit_bytes_sent += transport->xmit.offset; 1073 transport->xmit.offset = 0; 1074 if (atomic_long_read(&xprt->xmit_queuelen) == 1) 1075 tcp_sock_set_cork(transport->inet, false); 1076 return 0; 1077 } 1078 1079 WARN_ON_ONCE(sent == 0 && status == 0); 1080 1081 if (status == -EAGAIN ) { 1082 /* 1083 * Return EAGAIN if we're sure we're hitting the 1084 * socket send buffer limits. 1085 */ 1086 if (test_bit(SOCK_NOSPACE, &transport->sock->flags)) 1087 break; 1088 /* 1089 * Did we hit a memory allocation failure? 1090 */ 1091 if (sent == 0) { 1092 status = -ENOBUFS; 1093 if (vm_wait) 1094 break; 1095 /* Retry, knowing now that we're below the 1096 * socket send buffer limit 1097 */ 1098 vm_wait = true; 1099 } 1100 continue; 1101 } 1102 if (status < 0) 1103 break; 1104 vm_wait = false; 1105 } 1106 1107 switch (status) { 1108 case -ENOTSOCK: 1109 status = -ENOTCONN; 1110 /* Should we call xs_close() here? */ 1111 break; 1112 case -EAGAIN: 1113 status = xs_stream_nospace(req); 1114 break; 1115 case -ECONNRESET: 1116 case -ECONNREFUSED: 1117 case -ENOTCONN: 1118 case -EADDRINUSE: 1119 case -ENOBUFS: 1120 case -EPIPE: 1121 break; 1122 default: 1123 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1124 -status); 1125 } 1126 1127 return status; 1128 } 1129 1130 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1131 { 1132 transport->old_data_ready = sk->sk_data_ready; 1133 transport->old_state_change = sk->sk_state_change; 1134 transport->old_write_space = sk->sk_write_space; 1135 transport->old_error_report = sk->sk_error_report; 1136 } 1137 1138 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1139 { 1140 sk->sk_data_ready = transport->old_data_ready; 1141 sk->sk_state_change = transport->old_state_change; 1142 sk->sk_write_space = transport->old_write_space; 1143 sk->sk_error_report = transport->old_error_report; 1144 } 1145 1146 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt) 1147 { 1148 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1149 1150 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 1151 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state); 1152 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state); 1153 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state); 1154 } 1155 1156 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr) 1157 { 1158 set_bit(nr, &transport->sock_state); 1159 queue_work(xprtiod_workqueue, &transport->error_worker); 1160 } 1161 1162 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1163 { 1164 xprt->connect_cookie++; 1165 smp_mb__before_atomic(); 1166 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1167 clear_bit(XPRT_CLOSING, &xprt->state); 1168 xs_sock_reset_state_flags(xprt); 1169 smp_mb__after_atomic(); 1170 } 1171 1172 /** 1173 * xs_error_report - callback to handle TCP socket state errors 1174 * @sk: socket 1175 * 1176 * Note: we don't call sock_error() since there may be a rpc_task 1177 * using the socket, and so we don't want to clear sk->sk_err. 1178 */ 1179 static void xs_error_report(struct sock *sk) 1180 { 1181 struct sock_xprt *transport; 1182 struct rpc_xprt *xprt; 1183 1184 if (!(xprt = xprt_from_sock(sk))) 1185 return; 1186 1187 transport = container_of(xprt, struct sock_xprt, xprt); 1188 transport->xprt_err = -sk->sk_err; 1189 if (transport->xprt_err == 0) 1190 return; 1191 dprintk("RPC: xs_error_report client %p, error=%d...\n", 1192 xprt, -transport->xprt_err); 1193 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err); 1194 1195 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */ 1196 smp_mb__before_atomic(); 1197 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR); 1198 } 1199 1200 static void xs_reset_transport(struct sock_xprt *transport) 1201 { 1202 struct socket *sock = transport->sock; 1203 struct sock *sk = transport->inet; 1204 struct rpc_xprt *xprt = &transport->xprt; 1205 struct file *filp = transport->file; 1206 1207 if (sk == NULL) 1208 return; 1209 1210 if (atomic_read(&transport->xprt.swapper)) 1211 sk_clear_memalloc(sk); 1212 1213 kernel_sock_shutdown(sock, SHUT_RDWR); 1214 1215 mutex_lock(&transport->recv_mutex); 1216 lock_sock(sk); 1217 transport->inet = NULL; 1218 transport->sock = NULL; 1219 transport->file = NULL; 1220 1221 sk->sk_user_data = NULL; 1222 1223 xs_restore_old_callbacks(transport, sk); 1224 xprt_clear_connected(xprt); 1225 xs_sock_reset_connection_flags(xprt); 1226 /* Reset stream record info */ 1227 xs_stream_reset_connect(transport); 1228 release_sock(sk); 1229 mutex_unlock(&transport->recv_mutex); 1230 1231 trace_rpc_socket_close(xprt, sock); 1232 fput(filp); 1233 1234 xprt_disconnect_done(xprt); 1235 } 1236 1237 /** 1238 * xs_close - close a socket 1239 * @xprt: transport 1240 * 1241 * This is used when all requests are complete; ie, no DRC state remains 1242 * on the server we want to save. 1243 * 1244 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 1245 * xs_reset_transport() zeroing the socket from underneath a writer. 1246 */ 1247 static void xs_close(struct rpc_xprt *xprt) 1248 { 1249 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1250 1251 dprintk("RPC: xs_close xprt %p\n", xprt); 1252 1253 xs_reset_transport(transport); 1254 xprt->reestablish_timeout = 0; 1255 } 1256 1257 static void xs_inject_disconnect(struct rpc_xprt *xprt) 1258 { 1259 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 1260 xprt); 1261 xprt_disconnect_done(xprt); 1262 } 1263 1264 static void xs_xprt_free(struct rpc_xprt *xprt) 1265 { 1266 xs_free_peer_addresses(xprt); 1267 xprt_free(xprt); 1268 } 1269 1270 /** 1271 * xs_destroy - prepare to shutdown a transport 1272 * @xprt: doomed transport 1273 * 1274 */ 1275 static void xs_destroy(struct rpc_xprt *xprt) 1276 { 1277 struct sock_xprt *transport = container_of(xprt, 1278 struct sock_xprt, xprt); 1279 dprintk("RPC: xs_destroy xprt %p\n", xprt); 1280 1281 cancel_delayed_work_sync(&transport->connect_worker); 1282 xs_close(xprt); 1283 cancel_work_sync(&transport->recv_worker); 1284 cancel_work_sync(&transport->error_worker); 1285 xs_xprt_free(xprt); 1286 module_put(THIS_MODULE); 1287 } 1288 1289 /** 1290 * xs_udp_data_read_skb - receive callback for UDP sockets 1291 * @xprt: transport 1292 * @sk: socket 1293 * @skb: skbuff 1294 * 1295 */ 1296 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1297 struct sock *sk, 1298 struct sk_buff *skb) 1299 { 1300 struct rpc_task *task; 1301 struct rpc_rqst *rovr; 1302 int repsize, copied; 1303 u32 _xid; 1304 __be32 *xp; 1305 1306 repsize = skb->len; 1307 if (repsize < 4) { 1308 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1309 return; 1310 } 1311 1312 /* Copy the XID from the skb... */ 1313 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid); 1314 if (xp == NULL) 1315 return; 1316 1317 /* Look up and lock the request corresponding to the given XID */ 1318 spin_lock(&xprt->queue_lock); 1319 rovr = xprt_lookup_rqst(xprt, *xp); 1320 if (!rovr) 1321 goto out_unlock; 1322 xprt_pin_rqst(rovr); 1323 xprt_update_rtt(rovr->rq_task); 1324 spin_unlock(&xprt->queue_lock); 1325 task = rovr->rq_task; 1326 1327 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1328 copied = repsize; 1329 1330 /* Suck it into the iovec, verify checksum if not done by hw. */ 1331 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1332 spin_lock(&xprt->queue_lock); 1333 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS); 1334 goto out_unpin; 1335 } 1336 1337 1338 spin_lock(&xprt->transport_lock); 1339 xprt_adjust_cwnd(xprt, task, copied); 1340 spin_unlock(&xprt->transport_lock); 1341 spin_lock(&xprt->queue_lock); 1342 xprt_complete_rqst(task, copied); 1343 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS); 1344 out_unpin: 1345 xprt_unpin_rqst(rovr); 1346 out_unlock: 1347 spin_unlock(&xprt->queue_lock); 1348 } 1349 1350 static void xs_udp_data_receive(struct sock_xprt *transport) 1351 { 1352 struct sk_buff *skb; 1353 struct sock *sk; 1354 int err; 1355 1356 mutex_lock(&transport->recv_mutex); 1357 sk = transport->inet; 1358 if (sk == NULL) 1359 goto out; 1360 for (;;) { 1361 skb = skb_recv_udp(sk, 0, 1, &err); 1362 if (skb == NULL) 1363 break; 1364 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1365 consume_skb(skb); 1366 cond_resched(); 1367 } 1368 xs_poll_check_readable(transport); 1369 out: 1370 mutex_unlock(&transport->recv_mutex); 1371 } 1372 1373 static void xs_udp_data_receive_workfn(struct work_struct *work) 1374 { 1375 struct sock_xprt *transport = 1376 container_of(work, struct sock_xprt, recv_worker); 1377 unsigned int pflags = memalloc_nofs_save(); 1378 1379 xs_udp_data_receive(transport); 1380 memalloc_nofs_restore(pflags); 1381 } 1382 1383 /** 1384 * xs_data_ready - "data ready" callback for UDP sockets 1385 * @sk: socket with data to read 1386 * 1387 */ 1388 static void xs_data_ready(struct sock *sk) 1389 { 1390 struct rpc_xprt *xprt; 1391 1392 dprintk("RPC: xs_data_ready...\n"); 1393 xprt = xprt_from_sock(sk); 1394 if (xprt != NULL) { 1395 struct sock_xprt *transport = container_of(xprt, 1396 struct sock_xprt, xprt); 1397 transport->old_data_ready(sk); 1398 /* Any data means we had a useful conversation, so 1399 * then we don't need to delay the next reconnect 1400 */ 1401 if (xprt->reestablish_timeout) 1402 xprt->reestablish_timeout = 0; 1403 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1404 queue_work(xprtiod_workqueue, &transport->recv_worker); 1405 } 1406 } 1407 1408 /* 1409 * Helper function to force a TCP close if the server is sending 1410 * junk and/or it has put us in CLOSE_WAIT 1411 */ 1412 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1413 { 1414 xprt_force_disconnect(xprt); 1415 } 1416 1417 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1418 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt) 1419 { 1420 return PAGE_SIZE; 1421 } 1422 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1423 1424 /** 1425 * xs_tcp_state_change - callback to handle TCP socket state changes 1426 * @sk: socket whose state has changed 1427 * 1428 */ 1429 static void xs_tcp_state_change(struct sock *sk) 1430 { 1431 struct rpc_xprt *xprt; 1432 struct sock_xprt *transport; 1433 1434 if (!(xprt = xprt_from_sock(sk))) 1435 return; 1436 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1437 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1438 sk->sk_state, xprt_connected(xprt), 1439 sock_flag(sk, SOCK_DEAD), 1440 sock_flag(sk, SOCK_ZAPPED), 1441 sk->sk_shutdown); 1442 1443 transport = container_of(xprt, struct sock_xprt, xprt); 1444 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1445 switch (sk->sk_state) { 1446 case TCP_ESTABLISHED: 1447 if (!xprt_test_and_set_connected(xprt)) { 1448 xprt->connect_cookie++; 1449 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1450 xprt_clear_connecting(xprt); 1451 1452 xprt->stat.connect_count++; 1453 xprt->stat.connect_time += (long)jiffies - 1454 xprt->stat.connect_start; 1455 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING); 1456 } 1457 break; 1458 case TCP_FIN_WAIT1: 1459 /* The client initiated a shutdown of the socket */ 1460 xprt->connect_cookie++; 1461 xprt->reestablish_timeout = 0; 1462 set_bit(XPRT_CLOSING, &xprt->state); 1463 smp_mb__before_atomic(); 1464 clear_bit(XPRT_CONNECTED, &xprt->state); 1465 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1466 smp_mb__after_atomic(); 1467 break; 1468 case TCP_CLOSE_WAIT: 1469 /* The server initiated a shutdown of the socket */ 1470 xprt->connect_cookie++; 1471 clear_bit(XPRT_CONNECTED, &xprt->state); 1472 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1473 fallthrough; 1474 case TCP_CLOSING: 1475 /* 1476 * If the server closed down the connection, make sure that 1477 * we back off before reconnecting 1478 */ 1479 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1480 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1481 break; 1482 case TCP_LAST_ACK: 1483 set_bit(XPRT_CLOSING, &xprt->state); 1484 smp_mb__before_atomic(); 1485 clear_bit(XPRT_CONNECTED, &xprt->state); 1486 smp_mb__after_atomic(); 1487 break; 1488 case TCP_CLOSE: 1489 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1490 &transport->sock_state)) 1491 xprt_clear_connecting(xprt); 1492 clear_bit(XPRT_CLOSING, &xprt->state); 1493 /* Trigger the socket release */ 1494 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1495 } 1496 } 1497 1498 static void xs_write_space(struct sock *sk) 1499 { 1500 struct socket_wq *wq; 1501 struct sock_xprt *transport; 1502 struct rpc_xprt *xprt; 1503 1504 if (!sk->sk_socket) 1505 return; 1506 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1507 1508 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1509 return; 1510 transport = container_of(xprt, struct sock_xprt, xprt); 1511 rcu_read_lock(); 1512 wq = rcu_dereference(sk->sk_wq); 1513 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0) 1514 goto out; 1515 1516 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE); 1517 sk->sk_write_pending--; 1518 out: 1519 rcu_read_unlock(); 1520 } 1521 1522 /** 1523 * xs_udp_write_space - callback invoked when socket buffer space 1524 * becomes available 1525 * @sk: socket whose state has changed 1526 * 1527 * Called when more output buffer space is available for this socket. 1528 * We try not to wake our writers until they can make "significant" 1529 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1530 * with a bunch of small requests. 1531 */ 1532 static void xs_udp_write_space(struct sock *sk) 1533 { 1534 /* from net/core/sock.c:sock_def_write_space */ 1535 if (sock_writeable(sk)) 1536 xs_write_space(sk); 1537 } 1538 1539 /** 1540 * xs_tcp_write_space - callback invoked when socket buffer space 1541 * becomes available 1542 * @sk: socket whose state has changed 1543 * 1544 * Called when more output buffer space is available for this socket. 1545 * We try not to wake our writers until they can make "significant" 1546 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1547 * with a bunch of small requests. 1548 */ 1549 static void xs_tcp_write_space(struct sock *sk) 1550 { 1551 /* from net/core/stream.c:sk_stream_write_space */ 1552 if (sk_stream_is_writeable(sk)) 1553 xs_write_space(sk); 1554 } 1555 1556 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1557 { 1558 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1559 struct sock *sk = transport->inet; 1560 1561 if (transport->rcvsize) { 1562 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1563 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1564 } 1565 if (transport->sndsize) { 1566 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1567 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1568 sk->sk_write_space(sk); 1569 } 1570 } 1571 1572 /** 1573 * xs_udp_set_buffer_size - set send and receive limits 1574 * @xprt: generic transport 1575 * @sndsize: requested size of send buffer, in bytes 1576 * @rcvsize: requested size of receive buffer, in bytes 1577 * 1578 * Set socket send and receive buffer size limits. 1579 */ 1580 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1581 { 1582 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1583 1584 transport->sndsize = 0; 1585 if (sndsize) 1586 transport->sndsize = sndsize + 1024; 1587 transport->rcvsize = 0; 1588 if (rcvsize) 1589 transport->rcvsize = rcvsize + 1024; 1590 1591 xs_udp_do_set_buffer_size(xprt); 1592 } 1593 1594 /** 1595 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1596 * @xprt: controlling transport 1597 * @task: task that timed out 1598 * 1599 * Adjust the congestion window after a retransmit timeout has occurred. 1600 */ 1601 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1602 { 1603 spin_lock(&xprt->transport_lock); 1604 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1605 spin_unlock(&xprt->transport_lock); 1606 } 1607 1608 static int xs_get_random_port(void) 1609 { 1610 unsigned short min = xprt_min_resvport, max = xprt_max_resvport; 1611 unsigned short range; 1612 unsigned short rand; 1613 1614 if (max < min) 1615 return -EADDRINUSE; 1616 range = max - min + 1; 1617 rand = (unsigned short) prandom_u32() % range; 1618 return rand + min; 1619 } 1620 1621 static unsigned short xs_sock_getport(struct socket *sock) 1622 { 1623 struct sockaddr_storage buf; 1624 unsigned short port = 0; 1625 1626 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0) 1627 goto out; 1628 switch (buf.ss_family) { 1629 case AF_INET6: 1630 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1631 break; 1632 case AF_INET: 1633 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1634 } 1635 out: 1636 return port; 1637 } 1638 1639 /** 1640 * xs_set_port - reset the port number in the remote endpoint address 1641 * @xprt: generic transport 1642 * @port: new port number 1643 * 1644 */ 1645 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1646 { 1647 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1648 1649 rpc_set_port(xs_addr(xprt), port); 1650 xs_update_peer_port(xprt); 1651 } 1652 1653 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1654 { 1655 if (transport->srcport == 0 && transport->xprt.reuseport) 1656 transport->srcport = xs_sock_getport(sock); 1657 } 1658 1659 static int xs_get_srcport(struct sock_xprt *transport) 1660 { 1661 int port = transport->srcport; 1662 1663 if (port == 0 && transport->xprt.resvport) 1664 port = xs_get_random_port(); 1665 return port; 1666 } 1667 1668 unsigned short get_srcport(struct rpc_xprt *xprt) 1669 { 1670 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1671 unsigned short ret = 0; 1672 mutex_lock(&sock->recv_mutex); 1673 if (sock->sock) 1674 ret = xs_sock_getport(sock->sock); 1675 mutex_unlock(&sock->recv_mutex); 1676 return ret; 1677 } 1678 EXPORT_SYMBOL(get_srcport); 1679 1680 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1681 { 1682 if (transport->srcport != 0) 1683 transport->srcport = 0; 1684 if (!transport->xprt.resvport) 1685 return 0; 1686 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1687 return xprt_max_resvport; 1688 return --port; 1689 } 1690 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1691 { 1692 struct sockaddr_storage myaddr; 1693 int err, nloop = 0; 1694 int port = xs_get_srcport(transport); 1695 unsigned short last; 1696 1697 /* 1698 * If we are asking for any ephemeral port (i.e. port == 0 && 1699 * transport->xprt.resvport == 0), don't bind. Let the local 1700 * port selection happen implicitly when the socket is used 1701 * (for example at connect time). 1702 * 1703 * This ensures that we can continue to establish TCP 1704 * connections even when all local ephemeral ports are already 1705 * a part of some TCP connection. This makes no difference 1706 * for UDP sockets, but also doesn't harm them. 1707 * 1708 * If we're asking for any reserved port (i.e. port == 0 && 1709 * transport->xprt.resvport == 1) xs_get_srcport above will 1710 * ensure that port is non-zero and we will bind as needed. 1711 */ 1712 if (port <= 0) 1713 return port; 1714 1715 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1716 do { 1717 rpc_set_port((struct sockaddr *)&myaddr, port); 1718 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1719 transport->xprt.addrlen); 1720 if (err == 0) { 1721 if (transport->xprt.reuseport) 1722 transport->srcport = port; 1723 break; 1724 } 1725 last = port; 1726 port = xs_next_srcport(transport, port); 1727 if (port > last) 1728 nloop++; 1729 } while (err == -EADDRINUSE && nloop != 2); 1730 1731 if (myaddr.ss_family == AF_INET) 1732 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1733 &((struct sockaddr_in *)&myaddr)->sin_addr, 1734 port, err ? "failed" : "ok", err); 1735 else 1736 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1737 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1738 port, err ? "failed" : "ok", err); 1739 return err; 1740 } 1741 1742 /* 1743 * We don't support autobind on AF_LOCAL sockets 1744 */ 1745 static void xs_local_rpcbind(struct rpc_task *task) 1746 { 1747 xprt_set_bound(task->tk_xprt); 1748 } 1749 1750 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1751 { 1752 } 1753 1754 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1755 static struct lock_class_key xs_key[3]; 1756 static struct lock_class_key xs_slock_key[3]; 1757 1758 static inline void xs_reclassify_socketu(struct socket *sock) 1759 { 1760 struct sock *sk = sock->sk; 1761 1762 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1763 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]); 1764 } 1765 1766 static inline void xs_reclassify_socket4(struct socket *sock) 1767 { 1768 struct sock *sk = sock->sk; 1769 1770 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1771 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]); 1772 } 1773 1774 static inline void xs_reclassify_socket6(struct socket *sock) 1775 { 1776 struct sock *sk = sock->sk; 1777 1778 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1779 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]); 1780 } 1781 1782 static inline void xs_reclassify_socket(int family, struct socket *sock) 1783 { 1784 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk))) 1785 return; 1786 1787 switch (family) { 1788 case AF_LOCAL: 1789 xs_reclassify_socketu(sock); 1790 break; 1791 case AF_INET: 1792 xs_reclassify_socket4(sock); 1793 break; 1794 case AF_INET6: 1795 xs_reclassify_socket6(sock); 1796 break; 1797 } 1798 } 1799 #else 1800 static inline void xs_reclassify_socket(int family, struct socket *sock) 1801 { 1802 } 1803 #endif 1804 1805 static void xs_dummy_setup_socket(struct work_struct *work) 1806 { 1807 } 1808 1809 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1810 struct sock_xprt *transport, int family, int type, 1811 int protocol, bool reuseport) 1812 { 1813 struct file *filp; 1814 struct socket *sock; 1815 int err; 1816 1817 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1818 if (err < 0) { 1819 dprintk("RPC: can't create %d transport socket (%d).\n", 1820 protocol, -err); 1821 goto out; 1822 } 1823 xs_reclassify_socket(family, sock); 1824 1825 if (reuseport) 1826 sock_set_reuseport(sock->sk); 1827 1828 err = xs_bind(transport, sock); 1829 if (err) { 1830 sock_release(sock); 1831 goto out; 1832 } 1833 1834 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1835 if (IS_ERR(filp)) 1836 return ERR_CAST(filp); 1837 transport->file = filp; 1838 1839 return sock; 1840 out: 1841 return ERR_PTR(err); 1842 } 1843 1844 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1845 struct socket *sock) 1846 { 1847 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1848 xprt); 1849 1850 if (!transport->inet) { 1851 struct sock *sk = sock->sk; 1852 1853 lock_sock(sk); 1854 1855 xs_save_old_callbacks(transport, sk); 1856 1857 sk->sk_user_data = xprt; 1858 sk->sk_data_ready = xs_data_ready; 1859 sk->sk_write_space = xs_udp_write_space; 1860 sock_set_flag(sk, SOCK_FASYNC); 1861 sk->sk_error_report = xs_error_report; 1862 1863 xprt_clear_connected(xprt); 1864 1865 /* Reset to new socket */ 1866 transport->sock = sock; 1867 transport->inet = sk; 1868 1869 release_sock(sk); 1870 } 1871 1872 xs_stream_start_connect(transport); 1873 1874 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1875 } 1876 1877 /** 1878 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1879 * @transport: socket transport to connect 1880 */ 1881 static int xs_local_setup_socket(struct sock_xprt *transport) 1882 { 1883 struct rpc_xprt *xprt = &transport->xprt; 1884 struct file *filp; 1885 struct socket *sock; 1886 int status; 1887 1888 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1889 SOCK_STREAM, 0, &sock, 1); 1890 if (status < 0) { 1891 dprintk("RPC: can't create AF_LOCAL " 1892 "transport socket (%d).\n", -status); 1893 goto out; 1894 } 1895 xs_reclassify_socket(AF_LOCAL, sock); 1896 1897 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1898 if (IS_ERR(filp)) { 1899 status = PTR_ERR(filp); 1900 goto out; 1901 } 1902 transport->file = filp; 1903 1904 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1905 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1906 1907 status = xs_local_finish_connecting(xprt, sock); 1908 trace_rpc_socket_connect(xprt, sock, status); 1909 switch (status) { 1910 case 0: 1911 dprintk("RPC: xprt %p connected to %s\n", 1912 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1913 xprt->stat.connect_count++; 1914 xprt->stat.connect_time += (long)jiffies - 1915 xprt->stat.connect_start; 1916 xprt_set_connected(xprt); 1917 break; 1918 case -ENOBUFS: 1919 break; 1920 case -ENOENT: 1921 dprintk("RPC: xprt %p: socket %s does not exist\n", 1922 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1923 break; 1924 case -ECONNREFUSED: 1925 dprintk("RPC: xprt %p: connection refused for %s\n", 1926 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1927 break; 1928 default: 1929 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 1930 __func__, -status, 1931 xprt->address_strings[RPC_DISPLAY_ADDR]); 1932 } 1933 1934 out: 1935 xprt_clear_connecting(xprt); 1936 xprt_wake_pending_tasks(xprt, status); 1937 return status; 1938 } 1939 1940 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 1941 { 1942 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1943 int ret; 1944 1945 if (RPC_IS_ASYNC(task)) { 1946 /* 1947 * We want the AF_LOCAL connect to be resolved in the 1948 * filesystem namespace of the process making the rpc 1949 * call. Thus we connect synchronously. 1950 * 1951 * If we want to support asynchronous AF_LOCAL calls, 1952 * we'll need to figure out how to pass a namespace to 1953 * connect. 1954 */ 1955 task->tk_rpc_status = -ENOTCONN; 1956 rpc_exit(task, -ENOTCONN); 1957 return; 1958 } 1959 ret = xs_local_setup_socket(transport); 1960 if (ret && !RPC_IS_SOFTCONN(task)) 1961 msleep_interruptible(15000); 1962 } 1963 1964 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 1965 /* 1966 * Note that this should be called with XPRT_LOCKED held, or recv_mutex 1967 * held, or when we otherwise know that we have exclusive access to the 1968 * socket, to guard against races with xs_reset_transport. 1969 */ 1970 static void xs_set_memalloc(struct rpc_xprt *xprt) 1971 { 1972 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1973 xprt); 1974 1975 /* 1976 * If there's no sock, then we have nothing to set. The 1977 * reconnecting process will get it for us. 1978 */ 1979 if (!transport->inet) 1980 return; 1981 if (atomic_read(&xprt->swapper)) 1982 sk_set_memalloc(transport->inet); 1983 } 1984 1985 /** 1986 * xs_enable_swap - Tag this transport as being used for swap. 1987 * @xprt: transport to tag 1988 * 1989 * Take a reference to this transport on behalf of the rpc_clnt, and 1990 * optionally mark it for swapping if it wasn't already. 1991 */ 1992 static int 1993 xs_enable_swap(struct rpc_xprt *xprt) 1994 { 1995 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 1996 1997 mutex_lock(&xs->recv_mutex); 1998 if (atomic_inc_return(&xprt->swapper) == 1 && 1999 xs->inet) 2000 sk_set_memalloc(xs->inet); 2001 mutex_unlock(&xs->recv_mutex); 2002 return 0; 2003 } 2004 2005 /** 2006 * xs_disable_swap - Untag this transport as being used for swap. 2007 * @xprt: transport to tag 2008 * 2009 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2010 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2011 */ 2012 static void 2013 xs_disable_swap(struct rpc_xprt *xprt) 2014 { 2015 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2016 2017 mutex_lock(&xs->recv_mutex); 2018 if (atomic_dec_and_test(&xprt->swapper) && 2019 xs->inet) 2020 sk_clear_memalloc(xs->inet); 2021 mutex_unlock(&xs->recv_mutex); 2022 } 2023 #else 2024 static void xs_set_memalloc(struct rpc_xprt *xprt) 2025 { 2026 } 2027 2028 static int 2029 xs_enable_swap(struct rpc_xprt *xprt) 2030 { 2031 return -EINVAL; 2032 } 2033 2034 static void 2035 xs_disable_swap(struct rpc_xprt *xprt) 2036 { 2037 } 2038 #endif 2039 2040 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2041 { 2042 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2043 2044 if (!transport->inet) { 2045 struct sock *sk = sock->sk; 2046 2047 lock_sock(sk); 2048 2049 xs_save_old_callbacks(transport, sk); 2050 2051 sk->sk_user_data = xprt; 2052 sk->sk_data_ready = xs_data_ready; 2053 sk->sk_write_space = xs_udp_write_space; 2054 sock_set_flag(sk, SOCK_FASYNC); 2055 2056 xprt_set_connected(xprt); 2057 2058 /* Reset to new socket */ 2059 transport->sock = sock; 2060 transport->inet = sk; 2061 2062 xs_set_memalloc(xprt); 2063 2064 release_sock(sk); 2065 } 2066 xs_udp_do_set_buffer_size(xprt); 2067 2068 xprt->stat.connect_start = jiffies; 2069 } 2070 2071 static void xs_udp_setup_socket(struct work_struct *work) 2072 { 2073 struct sock_xprt *transport = 2074 container_of(work, struct sock_xprt, connect_worker.work); 2075 struct rpc_xprt *xprt = &transport->xprt; 2076 struct socket *sock; 2077 int status = -EIO; 2078 unsigned int pflags = current->flags; 2079 2080 if (atomic_read(&xprt->swapper)) 2081 current->flags |= PF_MEMALLOC; 2082 sock = xs_create_sock(xprt, transport, 2083 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2084 IPPROTO_UDP, false); 2085 if (IS_ERR(sock)) 2086 goto out; 2087 2088 dprintk("RPC: worker connecting xprt %p via %s to " 2089 "%s (port %s)\n", xprt, 2090 xprt->address_strings[RPC_DISPLAY_PROTO], 2091 xprt->address_strings[RPC_DISPLAY_ADDR], 2092 xprt->address_strings[RPC_DISPLAY_PORT]); 2093 2094 xs_udp_finish_connecting(xprt, sock); 2095 trace_rpc_socket_connect(xprt, sock, 0); 2096 status = 0; 2097 out: 2098 xprt_clear_connecting(xprt); 2099 xprt_unlock_connect(xprt, transport); 2100 xprt_wake_pending_tasks(xprt, status); 2101 current_restore_flags(pflags, PF_MEMALLOC); 2102 } 2103 2104 /** 2105 * xs_tcp_shutdown - gracefully shut down a TCP socket 2106 * @xprt: transport 2107 * 2108 * Initiates a graceful shutdown of the TCP socket by calling the 2109 * equivalent of shutdown(SHUT_RDWR); 2110 */ 2111 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2112 { 2113 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2114 struct socket *sock = transport->sock; 2115 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE; 2116 2117 if (sock == NULL) 2118 return; 2119 if (!xprt->reuseport) { 2120 xs_close(xprt); 2121 return; 2122 } 2123 switch (skst) { 2124 case TCP_FIN_WAIT1: 2125 case TCP_FIN_WAIT2: 2126 break; 2127 case TCP_ESTABLISHED: 2128 case TCP_CLOSE_WAIT: 2129 kernel_sock_shutdown(sock, SHUT_RDWR); 2130 trace_rpc_socket_shutdown(xprt, sock); 2131 break; 2132 default: 2133 xs_reset_transport(transport); 2134 } 2135 } 2136 2137 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 2138 struct socket *sock) 2139 { 2140 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2141 unsigned int keepidle; 2142 unsigned int keepcnt; 2143 unsigned int timeo; 2144 2145 spin_lock(&xprt->transport_lock); 2146 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ); 2147 keepcnt = xprt->timeout->to_retries + 1; 2148 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2149 (xprt->timeout->to_retries + 1); 2150 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2151 spin_unlock(&xprt->transport_lock); 2152 2153 /* TCP Keepalive options */ 2154 sock_set_keepalive(sock->sk); 2155 tcp_sock_set_keepidle(sock->sk, keepidle); 2156 tcp_sock_set_keepintvl(sock->sk, keepidle); 2157 tcp_sock_set_keepcnt(sock->sk, keepcnt); 2158 2159 /* TCP user timeout (see RFC5482) */ 2160 tcp_sock_set_user_timeout(sock->sk, timeo); 2161 } 2162 2163 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt, 2164 unsigned long connect_timeout, 2165 unsigned long reconnect_timeout) 2166 { 2167 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2168 struct rpc_timeout to; 2169 unsigned long initval; 2170 2171 spin_lock(&xprt->transport_lock); 2172 if (reconnect_timeout < xprt->max_reconnect_timeout) 2173 xprt->max_reconnect_timeout = reconnect_timeout; 2174 if (connect_timeout < xprt->connect_timeout) { 2175 memcpy(&to, xprt->timeout, sizeof(to)); 2176 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1); 2177 /* Arbitrary lower limit */ 2178 if (initval < XS_TCP_INIT_REEST_TO << 1) 2179 initval = XS_TCP_INIT_REEST_TO << 1; 2180 to.to_initval = initval; 2181 to.to_maxval = initval; 2182 memcpy(&transport->tcp_timeout, &to, 2183 sizeof(transport->tcp_timeout)); 2184 xprt->timeout = &transport->tcp_timeout; 2185 xprt->connect_timeout = connect_timeout; 2186 } 2187 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2188 spin_unlock(&xprt->transport_lock); 2189 } 2190 2191 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2192 { 2193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2194 2195 if (!transport->inet) { 2196 struct sock *sk = sock->sk; 2197 2198 /* Avoid temporary address, they are bad for long-lived 2199 * connections such as NFS mounts. 2200 * RFC4941, section 3.6 suggests that: 2201 * Individual applications, which have specific 2202 * knowledge about the normal duration of connections, 2203 * MAY override this as appropriate. 2204 */ 2205 if (xs_addr(xprt)->sa_family == PF_INET6) { 2206 ip6_sock_set_addr_preferences(sk, 2207 IPV6_PREFER_SRC_PUBLIC); 2208 } 2209 2210 xs_tcp_set_socket_timeouts(xprt, sock); 2211 tcp_sock_set_nodelay(sk); 2212 2213 lock_sock(sk); 2214 2215 xs_save_old_callbacks(transport, sk); 2216 2217 sk->sk_user_data = xprt; 2218 sk->sk_data_ready = xs_data_ready; 2219 sk->sk_state_change = xs_tcp_state_change; 2220 sk->sk_write_space = xs_tcp_write_space; 2221 sock_set_flag(sk, SOCK_FASYNC); 2222 sk->sk_error_report = xs_error_report; 2223 2224 /* socket options */ 2225 sock_reset_flag(sk, SOCK_LINGER); 2226 2227 xprt_clear_connected(xprt); 2228 2229 /* Reset to new socket */ 2230 transport->sock = sock; 2231 transport->inet = sk; 2232 2233 release_sock(sk); 2234 } 2235 2236 if (!xprt_bound(xprt)) 2237 return -ENOTCONN; 2238 2239 xs_set_memalloc(xprt); 2240 2241 xs_stream_start_connect(transport); 2242 2243 /* Tell the socket layer to start connecting... */ 2244 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2245 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2246 } 2247 2248 /** 2249 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2250 * @work: queued work item 2251 * 2252 * Invoked by a work queue tasklet. 2253 */ 2254 static void xs_tcp_setup_socket(struct work_struct *work) 2255 { 2256 struct sock_xprt *transport = 2257 container_of(work, struct sock_xprt, connect_worker.work); 2258 struct socket *sock = transport->sock; 2259 struct rpc_xprt *xprt = &transport->xprt; 2260 int status; 2261 unsigned int pflags = current->flags; 2262 2263 if (atomic_read(&xprt->swapper)) 2264 current->flags |= PF_MEMALLOC; 2265 2266 if (xprt_connected(xprt)) 2267 goto out; 2268 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT, 2269 &transport->sock_state) || 2270 !sock) { 2271 xs_reset_transport(transport); 2272 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family, 2273 SOCK_STREAM, IPPROTO_TCP, true); 2274 if (IS_ERR(sock)) { 2275 xprt_wake_pending_tasks(xprt, PTR_ERR(sock)); 2276 goto out; 2277 } 2278 } 2279 2280 dprintk("RPC: worker connecting xprt %p via %s to " 2281 "%s (port %s)\n", xprt, 2282 xprt->address_strings[RPC_DISPLAY_PROTO], 2283 xprt->address_strings[RPC_DISPLAY_ADDR], 2284 xprt->address_strings[RPC_DISPLAY_PORT]); 2285 2286 status = xs_tcp_finish_connecting(xprt, sock); 2287 trace_rpc_socket_connect(xprt, sock, status); 2288 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2289 xprt, -status, xprt_connected(xprt), 2290 sock->sk->sk_state); 2291 switch (status) { 2292 case 0: 2293 case -EINPROGRESS: 2294 /* SYN_SENT! */ 2295 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state); 2296 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2297 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2298 fallthrough; 2299 case -EALREADY: 2300 goto out_unlock; 2301 case -EADDRNOTAVAIL: 2302 /* Source port number is unavailable. Try a new one! */ 2303 transport->srcport = 0; 2304 status = -EAGAIN; 2305 break; 2306 case -EINVAL: 2307 /* Happens, for instance, if the user specified a link 2308 * local IPv6 address without a scope-id. 2309 */ 2310 case -ECONNREFUSED: 2311 case -ECONNRESET: 2312 case -ENETDOWN: 2313 case -ENETUNREACH: 2314 case -EHOSTUNREACH: 2315 case -EADDRINUSE: 2316 case -ENOBUFS: 2317 break; 2318 default: 2319 printk("%s: connect returned unhandled error %d\n", 2320 __func__, status); 2321 status = -EAGAIN; 2322 } 2323 2324 /* xs_tcp_force_close() wakes tasks with a fixed error code. 2325 * We need to wake them first to ensure the correct error code. 2326 */ 2327 xprt_wake_pending_tasks(xprt, status); 2328 xs_tcp_force_close(xprt); 2329 out: 2330 xprt_clear_connecting(xprt); 2331 out_unlock: 2332 xprt_unlock_connect(xprt, transport); 2333 current_restore_flags(pflags, PF_MEMALLOC); 2334 } 2335 2336 /** 2337 * xs_connect - connect a socket to a remote endpoint 2338 * @xprt: pointer to transport structure 2339 * @task: address of RPC task that manages state of connect request 2340 * 2341 * TCP: If the remote end dropped the connection, delay reconnecting. 2342 * 2343 * UDP socket connects are synchronous, but we use a work queue anyway 2344 * to guarantee that even unprivileged user processes can set up a 2345 * socket on a privileged port. 2346 * 2347 * If a UDP socket connect fails, the delay behavior here prevents 2348 * retry floods (hard mounts). 2349 */ 2350 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2351 { 2352 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2353 unsigned long delay = 0; 2354 2355 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2356 2357 if (transport->sock != NULL) { 2358 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2359 "seconds\n", xprt, xprt->reestablish_timeout / HZ); 2360 2361 delay = xprt_reconnect_delay(xprt); 2362 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO); 2363 2364 } else 2365 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2366 2367 queue_delayed_work(xprtiod_workqueue, 2368 &transport->connect_worker, 2369 delay); 2370 } 2371 2372 static void xs_wake_disconnect(struct sock_xprt *transport) 2373 { 2374 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state)) 2375 xs_tcp_force_close(&transport->xprt); 2376 } 2377 2378 static void xs_wake_write(struct sock_xprt *transport) 2379 { 2380 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state)) 2381 xprt_write_space(&transport->xprt); 2382 } 2383 2384 static void xs_wake_error(struct sock_xprt *transport) 2385 { 2386 int sockerr; 2387 2388 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2389 return; 2390 mutex_lock(&transport->recv_mutex); 2391 if (transport->sock == NULL) 2392 goto out; 2393 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2394 goto out; 2395 sockerr = xchg(&transport->xprt_err, 0); 2396 if (sockerr < 0) 2397 xprt_wake_pending_tasks(&transport->xprt, sockerr); 2398 out: 2399 mutex_unlock(&transport->recv_mutex); 2400 } 2401 2402 static void xs_wake_pending(struct sock_xprt *transport) 2403 { 2404 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state)) 2405 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN); 2406 } 2407 2408 static void xs_error_handle(struct work_struct *work) 2409 { 2410 struct sock_xprt *transport = container_of(work, 2411 struct sock_xprt, error_worker); 2412 2413 xs_wake_disconnect(transport); 2414 xs_wake_write(transport); 2415 xs_wake_error(transport); 2416 xs_wake_pending(transport); 2417 } 2418 2419 /** 2420 * xs_local_print_stats - display AF_LOCAL socket-specific stats 2421 * @xprt: rpc_xprt struct containing statistics 2422 * @seq: output file 2423 * 2424 */ 2425 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2426 { 2427 long idle_time = 0; 2428 2429 if (xprt_connected(xprt)) 2430 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2431 2432 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2433 "%llu %llu %lu %llu %llu\n", 2434 xprt->stat.bind_count, 2435 xprt->stat.connect_count, 2436 xprt->stat.connect_time / HZ, 2437 idle_time, 2438 xprt->stat.sends, 2439 xprt->stat.recvs, 2440 xprt->stat.bad_xids, 2441 xprt->stat.req_u, 2442 xprt->stat.bklog_u, 2443 xprt->stat.max_slots, 2444 xprt->stat.sending_u, 2445 xprt->stat.pending_u); 2446 } 2447 2448 /** 2449 * xs_udp_print_stats - display UDP socket-specific stats 2450 * @xprt: rpc_xprt struct containing statistics 2451 * @seq: output file 2452 * 2453 */ 2454 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2455 { 2456 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2457 2458 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2459 "%lu %llu %llu\n", 2460 transport->srcport, 2461 xprt->stat.bind_count, 2462 xprt->stat.sends, 2463 xprt->stat.recvs, 2464 xprt->stat.bad_xids, 2465 xprt->stat.req_u, 2466 xprt->stat.bklog_u, 2467 xprt->stat.max_slots, 2468 xprt->stat.sending_u, 2469 xprt->stat.pending_u); 2470 } 2471 2472 /** 2473 * xs_tcp_print_stats - display TCP socket-specific stats 2474 * @xprt: rpc_xprt struct containing statistics 2475 * @seq: output file 2476 * 2477 */ 2478 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2479 { 2480 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2481 long idle_time = 0; 2482 2483 if (xprt_connected(xprt)) 2484 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2485 2486 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2487 "%llu %llu %lu %llu %llu\n", 2488 transport->srcport, 2489 xprt->stat.bind_count, 2490 xprt->stat.connect_count, 2491 xprt->stat.connect_time / HZ, 2492 idle_time, 2493 xprt->stat.sends, 2494 xprt->stat.recvs, 2495 xprt->stat.bad_xids, 2496 xprt->stat.req_u, 2497 xprt->stat.bklog_u, 2498 xprt->stat.max_slots, 2499 xprt->stat.sending_u, 2500 xprt->stat.pending_u); 2501 } 2502 2503 /* 2504 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2505 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2506 * to use the server side send routines. 2507 */ 2508 static int bc_malloc(struct rpc_task *task) 2509 { 2510 struct rpc_rqst *rqst = task->tk_rqstp; 2511 size_t size = rqst->rq_callsize; 2512 struct page *page; 2513 struct rpc_buffer *buf; 2514 2515 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) { 2516 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n", 2517 size); 2518 return -EINVAL; 2519 } 2520 2521 page = alloc_page(GFP_KERNEL); 2522 if (!page) 2523 return -ENOMEM; 2524 2525 buf = page_address(page); 2526 buf->len = PAGE_SIZE; 2527 2528 rqst->rq_buffer = buf->data; 2529 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; 2530 return 0; 2531 } 2532 2533 /* 2534 * Free the space allocated in the bc_alloc routine 2535 */ 2536 static void bc_free(struct rpc_task *task) 2537 { 2538 void *buffer = task->tk_rqstp->rq_buffer; 2539 struct rpc_buffer *buf; 2540 2541 buf = container_of(buffer, struct rpc_buffer, data); 2542 free_page((unsigned long)buf); 2543 } 2544 2545 static int bc_sendto(struct rpc_rqst *req) 2546 { 2547 struct xdr_buf *xdr = &req->rq_snd_buf; 2548 struct sock_xprt *transport = 2549 container_of(req->rq_xprt, struct sock_xprt, xprt); 2550 struct msghdr msg = { 2551 .msg_flags = 0, 2552 }; 2553 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 2554 (u32)xdr->len); 2555 unsigned int sent = 0; 2556 int err; 2557 2558 req->rq_xtime = ktime_get(); 2559 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent); 2560 xdr_free_bvec(xdr); 2561 if (err < 0 || sent != (xdr->len + sizeof(marker))) 2562 return -EAGAIN; 2563 return sent; 2564 } 2565 2566 /** 2567 * bc_send_request - Send a backchannel Call on a TCP socket 2568 * @req: rpc_rqst containing Call message to be sent 2569 * 2570 * xpt_mutex ensures @rqstp's whole message is written to the socket 2571 * without interruption. 2572 * 2573 * Return values: 2574 * %0 if the message was sent successfully 2575 * %ENOTCONN if the message was not sent 2576 */ 2577 static int bc_send_request(struct rpc_rqst *req) 2578 { 2579 struct svc_xprt *xprt; 2580 int len; 2581 2582 /* 2583 * Get the server socket associated with this callback xprt 2584 */ 2585 xprt = req->rq_xprt->bc_xprt; 2586 2587 /* 2588 * Grab the mutex to serialize data as the connection is shared 2589 * with the fore channel 2590 */ 2591 mutex_lock(&xprt->xpt_mutex); 2592 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2593 len = -ENOTCONN; 2594 else 2595 len = bc_sendto(req); 2596 mutex_unlock(&xprt->xpt_mutex); 2597 2598 if (len > 0) 2599 len = 0; 2600 2601 return len; 2602 } 2603 2604 /* 2605 * The close routine. Since this is client initiated, we do nothing 2606 */ 2607 2608 static void bc_close(struct rpc_xprt *xprt) 2609 { 2610 xprt_disconnect_done(xprt); 2611 } 2612 2613 /* 2614 * The xprt destroy routine. Again, because this connection is client 2615 * initiated, we do nothing 2616 */ 2617 2618 static void bc_destroy(struct rpc_xprt *xprt) 2619 { 2620 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2621 2622 xs_xprt_free(xprt); 2623 module_put(THIS_MODULE); 2624 } 2625 2626 static const struct rpc_xprt_ops xs_local_ops = { 2627 .reserve_xprt = xprt_reserve_xprt, 2628 .release_xprt = xprt_release_xprt, 2629 .alloc_slot = xprt_alloc_slot, 2630 .free_slot = xprt_free_slot, 2631 .rpcbind = xs_local_rpcbind, 2632 .set_port = xs_local_set_port, 2633 .connect = xs_local_connect, 2634 .buf_alloc = rpc_malloc, 2635 .buf_free = rpc_free, 2636 .prepare_request = xs_stream_prepare_request, 2637 .send_request = xs_local_send_request, 2638 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2639 .close = xs_close, 2640 .destroy = xs_destroy, 2641 .print_stats = xs_local_print_stats, 2642 .enable_swap = xs_enable_swap, 2643 .disable_swap = xs_disable_swap, 2644 }; 2645 2646 static const struct rpc_xprt_ops xs_udp_ops = { 2647 .set_buffer_size = xs_udp_set_buffer_size, 2648 .reserve_xprt = xprt_reserve_xprt_cong, 2649 .release_xprt = xprt_release_xprt_cong, 2650 .alloc_slot = xprt_alloc_slot, 2651 .free_slot = xprt_free_slot, 2652 .rpcbind = rpcb_getport_async, 2653 .set_port = xs_set_port, 2654 .connect = xs_connect, 2655 .buf_alloc = rpc_malloc, 2656 .buf_free = rpc_free, 2657 .send_request = xs_udp_send_request, 2658 .wait_for_reply_request = xprt_wait_for_reply_request_rtt, 2659 .timer = xs_udp_timer, 2660 .release_request = xprt_release_rqst_cong, 2661 .close = xs_close, 2662 .destroy = xs_destroy, 2663 .print_stats = xs_udp_print_stats, 2664 .enable_swap = xs_enable_swap, 2665 .disable_swap = xs_disable_swap, 2666 .inject_disconnect = xs_inject_disconnect, 2667 }; 2668 2669 static const struct rpc_xprt_ops xs_tcp_ops = { 2670 .reserve_xprt = xprt_reserve_xprt, 2671 .release_xprt = xprt_release_xprt, 2672 .alloc_slot = xprt_alloc_slot, 2673 .free_slot = xprt_free_slot, 2674 .rpcbind = rpcb_getport_async, 2675 .set_port = xs_set_port, 2676 .connect = xs_connect, 2677 .buf_alloc = rpc_malloc, 2678 .buf_free = rpc_free, 2679 .prepare_request = xs_stream_prepare_request, 2680 .send_request = xs_tcp_send_request, 2681 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2682 .close = xs_tcp_shutdown, 2683 .destroy = xs_destroy, 2684 .set_connect_timeout = xs_tcp_set_connect_timeout, 2685 .print_stats = xs_tcp_print_stats, 2686 .enable_swap = xs_enable_swap, 2687 .disable_swap = xs_disable_swap, 2688 .inject_disconnect = xs_inject_disconnect, 2689 #ifdef CONFIG_SUNRPC_BACKCHANNEL 2690 .bc_setup = xprt_setup_bc, 2691 .bc_maxpayload = xs_tcp_bc_maxpayload, 2692 .bc_num_slots = xprt_bc_max_slots, 2693 .bc_free_rqst = xprt_free_bc_rqst, 2694 .bc_destroy = xprt_destroy_bc, 2695 #endif 2696 }; 2697 2698 /* 2699 * The rpc_xprt_ops for the server backchannel 2700 */ 2701 2702 static const struct rpc_xprt_ops bc_tcp_ops = { 2703 .reserve_xprt = xprt_reserve_xprt, 2704 .release_xprt = xprt_release_xprt, 2705 .alloc_slot = xprt_alloc_slot, 2706 .free_slot = xprt_free_slot, 2707 .buf_alloc = bc_malloc, 2708 .buf_free = bc_free, 2709 .send_request = bc_send_request, 2710 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2711 .close = bc_close, 2712 .destroy = bc_destroy, 2713 .print_stats = xs_tcp_print_stats, 2714 .enable_swap = xs_enable_swap, 2715 .disable_swap = xs_disable_swap, 2716 .inject_disconnect = xs_inject_disconnect, 2717 }; 2718 2719 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2720 { 2721 static const struct sockaddr_in sin = { 2722 .sin_family = AF_INET, 2723 .sin_addr.s_addr = htonl(INADDR_ANY), 2724 }; 2725 static const struct sockaddr_in6 sin6 = { 2726 .sin6_family = AF_INET6, 2727 .sin6_addr = IN6ADDR_ANY_INIT, 2728 }; 2729 2730 switch (family) { 2731 case AF_LOCAL: 2732 break; 2733 case AF_INET: 2734 memcpy(sap, &sin, sizeof(sin)); 2735 break; 2736 case AF_INET6: 2737 memcpy(sap, &sin6, sizeof(sin6)); 2738 break; 2739 default: 2740 dprintk("RPC: %s: Bad address family\n", __func__); 2741 return -EAFNOSUPPORT; 2742 } 2743 return 0; 2744 } 2745 2746 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2747 unsigned int slot_table_size, 2748 unsigned int max_slot_table_size) 2749 { 2750 struct rpc_xprt *xprt; 2751 struct sock_xprt *new; 2752 2753 if (args->addrlen > sizeof(xprt->addr)) { 2754 dprintk("RPC: xs_setup_xprt: address too large\n"); 2755 return ERR_PTR(-EBADF); 2756 } 2757 2758 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2759 max_slot_table_size); 2760 if (xprt == NULL) { 2761 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2762 "rpc_xprt\n"); 2763 return ERR_PTR(-ENOMEM); 2764 } 2765 2766 new = container_of(xprt, struct sock_xprt, xprt); 2767 mutex_init(&new->recv_mutex); 2768 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2769 xprt->addrlen = args->addrlen; 2770 if (args->srcaddr) 2771 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2772 else { 2773 int err; 2774 err = xs_init_anyaddr(args->dstaddr->sa_family, 2775 (struct sockaddr *)&new->srcaddr); 2776 if (err != 0) { 2777 xprt_free(xprt); 2778 return ERR_PTR(err); 2779 } 2780 } 2781 2782 return xprt; 2783 } 2784 2785 static const struct rpc_timeout xs_local_default_timeout = { 2786 .to_initval = 10 * HZ, 2787 .to_maxval = 10 * HZ, 2788 .to_retries = 2, 2789 }; 2790 2791 /** 2792 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2793 * @args: rpc transport creation arguments 2794 * 2795 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2796 */ 2797 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2798 { 2799 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2800 struct sock_xprt *transport; 2801 struct rpc_xprt *xprt; 2802 struct rpc_xprt *ret; 2803 2804 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2805 xprt_max_tcp_slot_table_entries); 2806 if (IS_ERR(xprt)) 2807 return xprt; 2808 transport = container_of(xprt, struct sock_xprt, xprt); 2809 2810 xprt->prot = 0; 2811 xprt->xprt_class = &xs_local_transport; 2812 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2813 2814 xprt->bind_timeout = XS_BIND_TO; 2815 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2816 xprt->idle_timeout = XS_IDLE_DISC_TO; 2817 2818 xprt->ops = &xs_local_ops; 2819 xprt->timeout = &xs_local_default_timeout; 2820 2821 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 2822 INIT_WORK(&transport->error_worker, xs_error_handle); 2823 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket); 2824 2825 switch (sun->sun_family) { 2826 case AF_LOCAL: 2827 if (sun->sun_path[0] != '/') { 2828 dprintk("RPC: bad AF_LOCAL address: %s\n", 2829 sun->sun_path); 2830 ret = ERR_PTR(-EINVAL); 2831 goto out_err; 2832 } 2833 xprt_set_bound(xprt); 2834 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2835 ret = ERR_PTR(xs_local_setup_socket(transport)); 2836 if (ret) 2837 goto out_err; 2838 break; 2839 default: 2840 ret = ERR_PTR(-EAFNOSUPPORT); 2841 goto out_err; 2842 } 2843 2844 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2845 xprt->address_strings[RPC_DISPLAY_ADDR]); 2846 2847 if (try_module_get(THIS_MODULE)) 2848 return xprt; 2849 ret = ERR_PTR(-EINVAL); 2850 out_err: 2851 xs_xprt_free(xprt); 2852 return ret; 2853 } 2854 2855 static const struct rpc_timeout xs_udp_default_timeout = { 2856 .to_initval = 5 * HZ, 2857 .to_maxval = 30 * HZ, 2858 .to_increment = 5 * HZ, 2859 .to_retries = 5, 2860 }; 2861 2862 /** 2863 * xs_setup_udp - Set up transport to use a UDP socket 2864 * @args: rpc transport creation arguments 2865 * 2866 */ 2867 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2868 { 2869 struct sockaddr *addr = args->dstaddr; 2870 struct rpc_xprt *xprt; 2871 struct sock_xprt *transport; 2872 struct rpc_xprt *ret; 2873 2874 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2875 xprt_udp_slot_table_entries); 2876 if (IS_ERR(xprt)) 2877 return xprt; 2878 transport = container_of(xprt, struct sock_xprt, xprt); 2879 2880 xprt->prot = IPPROTO_UDP; 2881 xprt->xprt_class = &xs_udp_transport; 2882 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2883 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2884 2885 xprt->bind_timeout = XS_BIND_TO; 2886 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2887 xprt->idle_timeout = XS_IDLE_DISC_TO; 2888 2889 xprt->ops = &xs_udp_ops; 2890 2891 xprt->timeout = &xs_udp_default_timeout; 2892 2893 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 2894 INIT_WORK(&transport->error_worker, xs_error_handle); 2895 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 2896 2897 switch (addr->sa_family) { 2898 case AF_INET: 2899 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2900 xprt_set_bound(xprt); 2901 2902 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2903 break; 2904 case AF_INET6: 2905 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2906 xprt_set_bound(xprt); 2907 2908 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2909 break; 2910 default: 2911 ret = ERR_PTR(-EAFNOSUPPORT); 2912 goto out_err; 2913 } 2914 2915 if (xprt_bound(xprt)) 2916 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2917 xprt->address_strings[RPC_DISPLAY_ADDR], 2918 xprt->address_strings[RPC_DISPLAY_PORT], 2919 xprt->address_strings[RPC_DISPLAY_PROTO]); 2920 else 2921 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2922 xprt->address_strings[RPC_DISPLAY_ADDR], 2923 xprt->address_strings[RPC_DISPLAY_PROTO]); 2924 2925 if (try_module_get(THIS_MODULE)) 2926 return xprt; 2927 ret = ERR_PTR(-EINVAL); 2928 out_err: 2929 xs_xprt_free(xprt); 2930 return ret; 2931 } 2932 2933 static const struct rpc_timeout xs_tcp_default_timeout = { 2934 .to_initval = 60 * HZ, 2935 .to_maxval = 60 * HZ, 2936 .to_retries = 2, 2937 }; 2938 2939 /** 2940 * xs_setup_tcp - Set up transport to use a TCP socket 2941 * @args: rpc transport creation arguments 2942 * 2943 */ 2944 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2945 { 2946 struct sockaddr *addr = args->dstaddr; 2947 struct rpc_xprt *xprt; 2948 struct sock_xprt *transport; 2949 struct rpc_xprt *ret; 2950 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2951 2952 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2953 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2954 2955 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2956 max_slot_table_size); 2957 if (IS_ERR(xprt)) 2958 return xprt; 2959 transport = container_of(xprt, struct sock_xprt, xprt); 2960 2961 xprt->prot = IPPROTO_TCP; 2962 xprt->xprt_class = &xs_tcp_transport; 2963 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2964 2965 xprt->bind_timeout = XS_BIND_TO; 2966 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2967 xprt->idle_timeout = XS_IDLE_DISC_TO; 2968 2969 xprt->ops = &xs_tcp_ops; 2970 xprt->timeout = &xs_tcp_default_timeout; 2971 2972 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 2973 xprt->connect_timeout = xprt->timeout->to_initval * 2974 (xprt->timeout->to_retries + 1); 2975 2976 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 2977 INIT_WORK(&transport->error_worker, xs_error_handle); 2978 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 2979 2980 switch (addr->sa_family) { 2981 case AF_INET: 2982 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2983 xprt_set_bound(xprt); 2984 2985 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2986 break; 2987 case AF_INET6: 2988 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2989 xprt_set_bound(xprt); 2990 2991 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 2992 break; 2993 default: 2994 ret = ERR_PTR(-EAFNOSUPPORT); 2995 goto out_err; 2996 } 2997 2998 if (xprt_bound(xprt)) 2999 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3000 xprt->address_strings[RPC_DISPLAY_ADDR], 3001 xprt->address_strings[RPC_DISPLAY_PORT], 3002 xprt->address_strings[RPC_DISPLAY_PROTO]); 3003 else 3004 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3005 xprt->address_strings[RPC_DISPLAY_ADDR], 3006 xprt->address_strings[RPC_DISPLAY_PROTO]); 3007 3008 if (try_module_get(THIS_MODULE)) 3009 return xprt; 3010 ret = ERR_PTR(-EINVAL); 3011 out_err: 3012 xs_xprt_free(xprt); 3013 return ret; 3014 } 3015 3016 /** 3017 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 3018 * @args: rpc transport creation arguments 3019 * 3020 */ 3021 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 3022 { 3023 struct sockaddr *addr = args->dstaddr; 3024 struct rpc_xprt *xprt; 3025 struct sock_xprt *transport; 3026 struct svc_sock *bc_sock; 3027 struct rpc_xprt *ret; 3028 3029 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3030 xprt_tcp_slot_table_entries); 3031 if (IS_ERR(xprt)) 3032 return xprt; 3033 transport = container_of(xprt, struct sock_xprt, xprt); 3034 3035 xprt->prot = IPPROTO_TCP; 3036 xprt->xprt_class = &xs_bc_tcp_transport; 3037 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3038 xprt->timeout = &xs_tcp_default_timeout; 3039 3040 /* backchannel */ 3041 xprt_set_bound(xprt); 3042 xprt->bind_timeout = 0; 3043 xprt->reestablish_timeout = 0; 3044 xprt->idle_timeout = 0; 3045 3046 xprt->ops = &bc_tcp_ops; 3047 3048 switch (addr->sa_family) { 3049 case AF_INET: 3050 xs_format_peer_addresses(xprt, "tcp", 3051 RPCBIND_NETID_TCP); 3052 break; 3053 case AF_INET6: 3054 xs_format_peer_addresses(xprt, "tcp", 3055 RPCBIND_NETID_TCP6); 3056 break; 3057 default: 3058 ret = ERR_PTR(-EAFNOSUPPORT); 3059 goto out_err; 3060 } 3061 3062 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3063 xprt->address_strings[RPC_DISPLAY_ADDR], 3064 xprt->address_strings[RPC_DISPLAY_PORT], 3065 xprt->address_strings[RPC_DISPLAY_PROTO]); 3066 3067 /* 3068 * Once we've associated a backchannel xprt with a connection, 3069 * we want to keep it around as long as the connection lasts, 3070 * in case we need to start using it for a backchannel again; 3071 * this reference won't be dropped until bc_xprt is destroyed. 3072 */ 3073 xprt_get(xprt); 3074 args->bc_xprt->xpt_bc_xprt = xprt; 3075 xprt->bc_xprt = args->bc_xprt; 3076 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3077 transport->sock = bc_sock->sk_sock; 3078 transport->inet = bc_sock->sk_sk; 3079 3080 /* 3081 * Since we don't want connections for the backchannel, we set 3082 * the xprt status to connected 3083 */ 3084 xprt_set_connected(xprt); 3085 3086 if (try_module_get(THIS_MODULE)) 3087 return xprt; 3088 3089 args->bc_xprt->xpt_bc_xprt = NULL; 3090 args->bc_xprt->xpt_bc_xps = NULL; 3091 xprt_put(xprt); 3092 ret = ERR_PTR(-EINVAL); 3093 out_err: 3094 xs_xprt_free(xprt); 3095 return ret; 3096 } 3097 3098 static struct xprt_class xs_local_transport = { 3099 .list = LIST_HEAD_INIT(xs_local_transport.list), 3100 .name = "named UNIX socket", 3101 .owner = THIS_MODULE, 3102 .ident = XPRT_TRANSPORT_LOCAL, 3103 .setup = xs_setup_local, 3104 .netid = { "" }, 3105 }; 3106 3107 static struct xprt_class xs_udp_transport = { 3108 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3109 .name = "udp", 3110 .owner = THIS_MODULE, 3111 .ident = XPRT_TRANSPORT_UDP, 3112 .setup = xs_setup_udp, 3113 .netid = { "udp", "udp6", "" }, 3114 }; 3115 3116 static struct xprt_class xs_tcp_transport = { 3117 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3118 .name = "tcp", 3119 .owner = THIS_MODULE, 3120 .ident = XPRT_TRANSPORT_TCP, 3121 .setup = xs_setup_tcp, 3122 .netid = { "tcp", "tcp6", "" }, 3123 }; 3124 3125 static struct xprt_class xs_bc_tcp_transport = { 3126 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3127 .name = "tcp NFSv4.1 backchannel", 3128 .owner = THIS_MODULE, 3129 .ident = XPRT_TRANSPORT_BC_TCP, 3130 .setup = xs_setup_bc_tcp, 3131 .netid = { "" }, 3132 }; 3133 3134 /** 3135 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3136 * 3137 */ 3138 int init_socket_xprt(void) 3139 { 3140 if (!sunrpc_table_header) 3141 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3142 3143 xprt_register_transport(&xs_local_transport); 3144 xprt_register_transport(&xs_udp_transport); 3145 xprt_register_transport(&xs_tcp_transport); 3146 xprt_register_transport(&xs_bc_tcp_transport); 3147 3148 return 0; 3149 } 3150 3151 /** 3152 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3153 * 3154 */ 3155 void cleanup_socket_xprt(void) 3156 { 3157 if (sunrpc_table_header) { 3158 unregister_sysctl_table(sunrpc_table_header); 3159 sunrpc_table_header = NULL; 3160 } 3161 3162 xprt_unregister_transport(&xs_local_transport); 3163 xprt_unregister_transport(&xs_udp_transport); 3164 xprt_unregister_transport(&xs_tcp_transport); 3165 xprt_unregister_transport(&xs_bc_tcp_transport); 3166 } 3167 3168 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3169 { 3170 return param_set_uint_minmax(val, kp, 3171 RPC_MIN_RESVPORT, 3172 RPC_MAX_RESVPORT); 3173 } 3174 3175 static const struct kernel_param_ops param_ops_portnr = { 3176 .set = param_set_portnr, 3177 .get = param_get_uint, 3178 }; 3179 3180 #define param_check_portnr(name, p) \ 3181 __param_check(name, p, unsigned int); 3182 3183 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3184 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3185 3186 static int param_set_slot_table_size(const char *val, 3187 const struct kernel_param *kp) 3188 { 3189 return param_set_uint_minmax(val, kp, 3190 RPC_MIN_SLOT_TABLE, 3191 RPC_MAX_SLOT_TABLE); 3192 } 3193 3194 static const struct kernel_param_ops param_ops_slot_table_size = { 3195 .set = param_set_slot_table_size, 3196 .get = param_get_uint, 3197 }; 3198 3199 #define param_check_slot_table_size(name, p) \ 3200 __param_check(name, p, unsigned int); 3201 3202 static int param_set_max_slot_table_size(const char *val, 3203 const struct kernel_param *kp) 3204 { 3205 return param_set_uint_minmax(val, kp, 3206 RPC_MIN_SLOT_TABLE, 3207 RPC_MAX_SLOT_TABLE_LIMIT); 3208 } 3209 3210 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3211 .set = param_set_max_slot_table_size, 3212 .get = param_get_uint, 3213 }; 3214 3215 #define param_check_max_slot_table_size(name, p) \ 3216 __param_check(name, p, unsigned int); 3217 3218 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3219 slot_table_size, 0644); 3220 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3221 max_slot_table_size, 0644); 3222 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3223 slot_table_size, 0644); 3224