1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/net/sunrpc/xprtsock.c 4 * 5 * Client-side transport implementation for sockets. 6 * 7 * TCP callback races fixes (C) 1998 Red Hat 8 * TCP send fixes (C) 1998 Red Hat 9 * TCP NFS related read + write fixes 10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 11 * 12 * Rewrite of larges part of the code in order to stabilize TCP stuff. 13 * Fix behaviour when socket buffer is full. 14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> 15 * 16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> 17 * 18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005. 19 * <gilles.quillard@bull.net> 20 */ 21 22 #include <linux/types.h> 23 #include <linux/string.h> 24 #include <linux/slab.h> 25 #include <linux/module.h> 26 #include <linux/capability.h> 27 #include <linux/pagemap.h> 28 #include <linux/errno.h> 29 #include <linux/socket.h> 30 #include <linux/in.h> 31 #include <linux/net.h> 32 #include <linux/mm.h> 33 #include <linux/un.h> 34 #include <linux/udp.h> 35 #include <linux/tcp.h> 36 #include <linux/sunrpc/clnt.h> 37 #include <linux/sunrpc/addr.h> 38 #include <linux/sunrpc/sched.h> 39 #include <linux/sunrpc/svcsock.h> 40 #include <linux/sunrpc/xprtsock.h> 41 #include <linux/file.h> 42 #ifdef CONFIG_SUNRPC_BACKCHANNEL 43 #include <linux/sunrpc/bc_xprt.h> 44 #endif 45 46 #include <net/sock.h> 47 #include <net/checksum.h> 48 #include <net/udp.h> 49 #include <net/tcp.h> 50 #include <net/tls.h> 51 52 #include <linux/bvec.h> 53 #include <linux/highmem.h> 54 #include <linux/uio.h> 55 #include <linux/sched/mm.h> 56 57 #include <trace/events/sock.h> 58 #include <trace/events/sunrpc.h> 59 60 #include "socklib.h" 61 #include "sunrpc.h" 62 63 static void xs_close(struct rpc_xprt *xprt); 64 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock); 65 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 66 struct socket *sock); 67 68 /* 69 * xprtsock tunables 70 */ 71 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 72 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 73 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 74 75 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 76 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 77 78 #define XS_TCP_LINGER_TO (15U * HZ) 79 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 80 81 /* 82 * We can register our own files under /proc/sys/sunrpc by 83 * calling register_sysctl() again. The files in that 84 * directory become the union of all files registered there. 85 * 86 * We simply need to make sure that we don't collide with 87 * someone else's file names! 88 */ 89 90 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 91 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 92 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 93 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 94 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 95 96 static struct ctl_table_header *sunrpc_table_header; 97 98 static struct xprt_class xs_local_transport; 99 static struct xprt_class xs_udp_transport; 100 static struct xprt_class xs_tcp_transport; 101 static struct xprt_class xs_bc_tcp_transport; 102 103 /* 104 * FIXME: changing the UDP slot table size should also resize the UDP 105 * socket buffers for existing UDP transports 106 */ 107 static struct ctl_table xs_tunables_table[] = { 108 { 109 .procname = "udp_slot_table_entries", 110 .data = &xprt_udp_slot_table_entries, 111 .maxlen = sizeof(unsigned int), 112 .mode = 0644, 113 .proc_handler = proc_dointvec_minmax, 114 .extra1 = &min_slot_table_size, 115 .extra2 = &max_slot_table_size 116 }, 117 { 118 .procname = "tcp_slot_table_entries", 119 .data = &xprt_tcp_slot_table_entries, 120 .maxlen = sizeof(unsigned int), 121 .mode = 0644, 122 .proc_handler = proc_dointvec_minmax, 123 .extra1 = &min_slot_table_size, 124 .extra2 = &max_slot_table_size 125 }, 126 { 127 .procname = "tcp_max_slot_table_entries", 128 .data = &xprt_max_tcp_slot_table_entries, 129 .maxlen = sizeof(unsigned int), 130 .mode = 0644, 131 .proc_handler = proc_dointvec_minmax, 132 .extra1 = &min_slot_table_size, 133 .extra2 = &max_tcp_slot_table_limit 134 }, 135 { 136 .procname = "min_resvport", 137 .data = &xprt_min_resvport, 138 .maxlen = sizeof(unsigned int), 139 .mode = 0644, 140 .proc_handler = proc_dointvec_minmax, 141 .extra1 = &xprt_min_resvport_limit, 142 .extra2 = &xprt_max_resvport_limit 143 }, 144 { 145 .procname = "max_resvport", 146 .data = &xprt_max_resvport, 147 .maxlen = sizeof(unsigned int), 148 .mode = 0644, 149 .proc_handler = proc_dointvec_minmax, 150 .extra1 = &xprt_min_resvport_limit, 151 .extra2 = &xprt_max_resvport_limit 152 }, 153 { 154 .procname = "tcp_fin_timeout", 155 .data = &xs_tcp_fin_timeout, 156 .maxlen = sizeof(xs_tcp_fin_timeout), 157 .mode = 0644, 158 .proc_handler = proc_dointvec_jiffies, 159 }, 160 { }, 161 }; 162 163 /* 164 * Wait duration for a reply from the RPC portmapper. 165 */ 166 #define XS_BIND_TO (60U * HZ) 167 168 /* 169 * Delay if a UDP socket connect error occurs. This is most likely some 170 * kind of resource problem on the local host. 171 */ 172 #define XS_UDP_REEST_TO (2U * HZ) 173 174 /* 175 * The reestablish timeout allows clients to delay for a bit before attempting 176 * to reconnect to a server that just dropped our connection. 177 * 178 * We implement an exponential backoff when trying to reestablish a TCP 179 * transport connection with the server. Some servers like to drop a TCP 180 * connection when they are overworked, so we start with a short timeout and 181 * increase over time if the server is down or not responding. 182 */ 183 #define XS_TCP_INIT_REEST_TO (3U * HZ) 184 185 /* 186 * TCP idle timeout; client drops the transport socket if it is idle 187 * for this long. Note that we also timeout UDP sockets to prevent 188 * holding port numbers when there is no RPC traffic. 189 */ 190 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 191 192 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 193 # undef RPC_DEBUG_DATA 194 # define RPCDBG_FACILITY RPCDBG_TRANS 195 #endif 196 197 #ifdef RPC_DEBUG_DATA 198 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 199 { 200 u8 *buf = (u8 *) packet; 201 int j; 202 203 dprintk("RPC: %s\n", msg); 204 for (j = 0; j < count && j < 128; j += 4) { 205 if (!(j & 31)) { 206 if (j) 207 dprintk("\n"); 208 dprintk("0x%04x ", j); 209 } 210 dprintk("%02x%02x%02x%02x ", 211 buf[j], buf[j+1], buf[j+2], buf[j+3]); 212 } 213 dprintk("\n"); 214 } 215 #else 216 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 217 { 218 /* NOP */ 219 } 220 #endif 221 222 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 223 { 224 return (struct rpc_xprt *) sk->sk_user_data; 225 } 226 227 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 228 { 229 return (struct sockaddr *) &xprt->addr; 230 } 231 232 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 233 { 234 return (struct sockaddr_un *) &xprt->addr; 235 } 236 237 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 238 { 239 return (struct sockaddr_in *) &xprt->addr; 240 } 241 242 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 243 { 244 return (struct sockaddr_in6 *) &xprt->addr; 245 } 246 247 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 248 { 249 struct sockaddr *sap = xs_addr(xprt); 250 struct sockaddr_in6 *sin6; 251 struct sockaddr_in *sin; 252 struct sockaddr_un *sun; 253 char buf[128]; 254 255 switch (sap->sa_family) { 256 case AF_LOCAL: 257 sun = xs_addr_un(xprt); 258 if (sun->sun_path[0]) { 259 strscpy(buf, sun->sun_path, sizeof(buf)); 260 } else { 261 buf[0] = '@'; 262 strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1); 263 } 264 xprt->address_strings[RPC_DISPLAY_ADDR] = 265 kstrdup(buf, GFP_KERNEL); 266 break; 267 case AF_INET: 268 (void)rpc_ntop(sap, buf, sizeof(buf)); 269 xprt->address_strings[RPC_DISPLAY_ADDR] = 270 kstrdup(buf, GFP_KERNEL); 271 sin = xs_addr_in(xprt); 272 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 273 break; 274 case AF_INET6: 275 (void)rpc_ntop(sap, buf, sizeof(buf)); 276 xprt->address_strings[RPC_DISPLAY_ADDR] = 277 kstrdup(buf, GFP_KERNEL); 278 sin6 = xs_addr_in6(xprt); 279 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 280 break; 281 default: 282 BUG(); 283 } 284 285 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 286 } 287 288 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 289 { 290 struct sockaddr *sap = xs_addr(xprt); 291 char buf[128]; 292 293 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 294 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 295 296 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 297 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 298 } 299 300 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 301 const char *protocol, 302 const char *netid) 303 { 304 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 305 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 306 xs_format_common_peer_addresses(xprt); 307 xs_format_common_peer_ports(xprt); 308 } 309 310 static void xs_update_peer_port(struct rpc_xprt *xprt) 311 { 312 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 313 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 314 315 xs_format_common_peer_ports(xprt); 316 } 317 318 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 319 { 320 unsigned int i; 321 322 for (i = 0; i < RPC_DISPLAY_MAX; i++) 323 switch (i) { 324 case RPC_DISPLAY_PROTO: 325 case RPC_DISPLAY_NETID: 326 continue; 327 default: 328 kfree(xprt->address_strings[i]); 329 } 330 } 331 332 static size_t 333 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp) 334 { 335 size_t i,n; 336 337 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES)) 338 return want; 339 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT; 340 for (i = 0; i < n; i++) { 341 if (buf->pages[i]) 342 continue; 343 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp); 344 if (!buf->pages[i]) { 345 i *= PAGE_SIZE; 346 return i > buf->page_base ? i - buf->page_base : 0; 347 } 348 } 349 return want; 350 } 351 352 static int 353 xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg, 354 struct cmsghdr *cmsg, int ret) 355 { 356 if (cmsg->cmsg_level == SOL_TLS && 357 cmsg->cmsg_type == TLS_GET_RECORD_TYPE) { 358 u8 content_type = *((u8 *)CMSG_DATA(cmsg)); 359 360 switch (content_type) { 361 case TLS_RECORD_TYPE_DATA: 362 /* TLS sets EOR at the end of each application data 363 * record, even though there might be more frames 364 * waiting to be decrypted. 365 */ 366 msg->msg_flags &= ~MSG_EOR; 367 break; 368 case TLS_RECORD_TYPE_ALERT: 369 ret = -ENOTCONN; 370 break; 371 default: 372 ret = -EAGAIN; 373 } 374 } 375 return ret; 376 } 377 378 static int 379 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags) 380 { 381 union { 382 struct cmsghdr cmsg; 383 u8 buf[CMSG_SPACE(sizeof(u8))]; 384 } u; 385 int ret; 386 387 msg->msg_control = &u; 388 msg->msg_controllen = sizeof(u); 389 ret = sock_recvmsg(sock, msg, flags); 390 if (msg->msg_controllen != sizeof(u)) 391 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret); 392 return ret; 393 } 394 395 static ssize_t 396 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek) 397 { 398 ssize_t ret; 399 if (seek != 0) 400 iov_iter_advance(&msg->msg_iter, seek); 401 ret = xs_sock_recv_cmsg(sock, msg, flags); 402 return ret > 0 ? ret + seek : ret; 403 } 404 405 static ssize_t 406 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags, 407 struct kvec *kvec, size_t count, size_t seek) 408 { 409 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count); 410 return xs_sock_recvmsg(sock, msg, flags, seek); 411 } 412 413 static ssize_t 414 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags, 415 struct bio_vec *bvec, unsigned long nr, size_t count, 416 size_t seek) 417 { 418 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count); 419 return xs_sock_recvmsg(sock, msg, flags, seek); 420 } 421 422 static ssize_t 423 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags, 424 size_t count) 425 { 426 iov_iter_discard(&msg->msg_iter, ITER_DEST, count); 427 return xs_sock_recv_cmsg(sock, msg, flags); 428 } 429 430 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 431 static void 432 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 433 { 434 struct bvec_iter bi = { 435 .bi_size = count, 436 }; 437 struct bio_vec bv; 438 439 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); 440 for_each_bvec(bv, bvec, bi, bi) 441 flush_dcache_page(bv.bv_page); 442 } 443 #else 444 static inline void 445 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 446 { 447 } 448 #endif 449 450 static ssize_t 451 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags, 452 struct xdr_buf *buf, size_t count, size_t seek, size_t *read) 453 { 454 size_t want, seek_init = seek, offset = 0; 455 ssize_t ret; 456 457 want = min_t(size_t, count, buf->head[0].iov_len); 458 if (seek < want) { 459 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek); 460 if (ret <= 0) 461 goto sock_err; 462 offset += ret; 463 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 464 goto out; 465 if (ret != want) 466 goto out; 467 seek = 0; 468 } else { 469 seek -= want; 470 offset += want; 471 } 472 473 want = xs_alloc_sparse_pages( 474 buf, min_t(size_t, count - offset, buf->page_len), 475 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN); 476 if (seek < want) { 477 ret = xs_read_bvec(sock, msg, flags, buf->bvec, 478 xdr_buf_pagecount(buf), 479 want + buf->page_base, 480 seek + buf->page_base); 481 if (ret <= 0) 482 goto sock_err; 483 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base); 484 ret -= buf->page_base; 485 offset += ret; 486 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 487 goto out; 488 if (ret != want) 489 goto out; 490 seek = 0; 491 } else { 492 seek -= want; 493 offset += want; 494 } 495 496 want = min_t(size_t, count - offset, buf->tail[0].iov_len); 497 if (seek < want) { 498 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek); 499 if (ret <= 0) 500 goto sock_err; 501 offset += ret; 502 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 503 goto out; 504 if (ret != want) 505 goto out; 506 } else if (offset < seek_init) 507 offset = seek_init; 508 ret = -EMSGSIZE; 509 out: 510 *read = offset - seek_init; 511 return ret; 512 sock_err: 513 offset += seek; 514 goto out; 515 } 516 517 static void 518 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf) 519 { 520 if (!transport->recv.copied) { 521 if (buf->head[0].iov_len >= transport->recv.offset) 522 memcpy(buf->head[0].iov_base, 523 &transport->recv.xid, 524 transport->recv.offset); 525 transport->recv.copied = transport->recv.offset; 526 } 527 } 528 529 static bool 530 xs_read_stream_request_done(struct sock_xprt *transport) 531 { 532 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT); 533 } 534 535 static void 536 xs_read_stream_check_eor(struct sock_xprt *transport, 537 struct msghdr *msg) 538 { 539 if (xs_read_stream_request_done(transport)) 540 msg->msg_flags |= MSG_EOR; 541 } 542 543 static ssize_t 544 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg, 545 int flags, struct rpc_rqst *req) 546 { 547 struct xdr_buf *buf = &req->rq_private_buf; 548 size_t want, read; 549 ssize_t ret; 550 551 xs_read_header(transport, buf); 552 553 want = transport->recv.len - transport->recv.offset; 554 if (want != 0) { 555 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf, 556 transport->recv.copied + want, 557 transport->recv.copied, 558 &read); 559 transport->recv.offset += read; 560 transport->recv.copied += read; 561 } 562 563 if (transport->recv.offset == transport->recv.len) 564 xs_read_stream_check_eor(transport, msg); 565 566 if (want == 0) 567 return 0; 568 569 switch (ret) { 570 default: 571 break; 572 case -EFAULT: 573 case -EMSGSIZE: 574 msg->msg_flags |= MSG_TRUNC; 575 return read; 576 case 0: 577 return -ESHUTDOWN; 578 } 579 return ret < 0 ? ret : read; 580 } 581 582 static size_t 583 xs_read_stream_headersize(bool isfrag) 584 { 585 if (isfrag) 586 return sizeof(__be32); 587 return 3 * sizeof(__be32); 588 } 589 590 static ssize_t 591 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg, 592 int flags, size_t want, size_t seek) 593 { 594 struct kvec kvec = { 595 .iov_base = &transport->recv.fraghdr, 596 .iov_len = want, 597 }; 598 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek); 599 } 600 601 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 602 static ssize_t 603 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 604 { 605 struct rpc_xprt *xprt = &transport->xprt; 606 struct rpc_rqst *req; 607 ssize_t ret; 608 609 /* Is this transport associated with the backchannel? */ 610 if (!xprt->bc_serv) 611 return -ESHUTDOWN; 612 613 /* Look up and lock the request corresponding to the given XID */ 614 req = xprt_lookup_bc_request(xprt, transport->recv.xid); 615 if (!req) { 616 printk(KERN_WARNING "Callback slot table overflowed\n"); 617 return -ESHUTDOWN; 618 } 619 if (transport->recv.copied && !req->rq_private_buf.len) 620 return -ESHUTDOWN; 621 622 ret = xs_read_stream_request(transport, msg, flags, req); 623 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 624 xprt_complete_bc_request(req, transport->recv.copied); 625 else 626 req->rq_private_buf.len = transport->recv.copied; 627 628 return ret; 629 } 630 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 631 static ssize_t 632 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 633 { 634 return -ESHUTDOWN; 635 } 636 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 637 638 static ssize_t 639 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags) 640 { 641 struct rpc_xprt *xprt = &transport->xprt; 642 struct rpc_rqst *req; 643 ssize_t ret = 0; 644 645 /* Look up and lock the request corresponding to the given XID */ 646 spin_lock(&xprt->queue_lock); 647 req = xprt_lookup_rqst(xprt, transport->recv.xid); 648 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) { 649 msg->msg_flags |= MSG_TRUNC; 650 goto out; 651 } 652 xprt_pin_rqst(req); 653 spin_unlock(&xprt->queue_lock); 654 655 ret = xs_read_stream_request(transport, msg, flags, req); 656 657 spin_lock(&xprt->queue_lock); 658 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 659 xprt_complete_rqst(req->rq_task, transport->recv.copied); 660 else 661 req->rq_private_buf.len = transport->recv.copied; 662 xprt_unpin_rqst(req); 663 out: 664 spin_unlock(&xprt->queue_lock); 665 return ret; 666 } 667 668 static ssize_t 669 xs_read_stream(struct sock_xprt *transport, int flags) 670 { 671 struct msghdr msg = { 0 }; 672 size_t want, read = 0; 673 ssize_t ret = 0; 674 675 if (transport->recv.len == 0) { 676 want = xs_read_stream_headersize(transport->recv.copied != 0); 677 ret = xs_read_stream_header(transport, &msg, flags, want, 678 transport->recv.offset); 679 if (ret <= 0) 680 goto out_err; 681 transport->recv.offset = ret; 682 if (transport->recv.offset != want) 683 return transport->recv.offset; 684 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) & 685 RPC_FRAGMENT_SIZE_MASK; 686 transport->recv.offset -= sizeof(transport->recv.fraghdr); 687 read = ret; 688 } 689 690 switch (be32_to_cpu(transport->recv.calldir)) { 691 default: 692 msg.msg_flags |= MSG_TRUNC; 693 break; 694 case RPC_CALL: 695 ret = xs_read_stream_call(transport, &msg, flags); 696 break; 697 case RPC_REPLY: 698 ret = xs_read_stream_reply(transport, &msg, flags); 699 } 700 if (msg.msg_flags & MSG_TRUNC) { 701 transport->recv.calldir = cpu_to_be32(-1); 702 transport->recv.copied = -1; 703 } 704 if (ret < 0) 705 goto out_err; 706 read += ret; 707 if (transport->recv.offset < transport->recv.len) { 708 if (!(msg.msg_flags & MSG_TRUNC)) 709 return read; 710 msg.msg_flags = 0; 711 ret = xs_read_discard(transport->sock, &msg, flags, 712 transport->recv.len - transport->recv.offset); 713 if (ret <= 0) 714 goto out_err; 715 transport->recv.offset += ret; 716 read += ret; 717 if (transport->recv.offset != transport->recv.len) 718 return read; 719 } 720 if (xs_read_stream_request_done(transport)) { 721 trace_xs_stream_read_request(transport); 722 transport->recv.copied = 0; 723 } 724 transport->recv.offset = 0; 725 transport->recv.len = 0; 726 return read; 727 out_err: 728 return ret != 0 ? ret : -ESHUTDOWN; 729 } 730 731 static __poll_t xs_poll_socket(struct sock_xprt *transport) 732 { 733 return transport->sock->ops->poll(transport->file, transport->sock, 734 NULL); 735 } 736 737 static bool xs_poll_socket_readable(struct sock_xprt *transport) 738 { 739 __poll_t events = xs_poll_socket(transport); 740 741 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP); 742 } 743 744 static void xs_poll_check_readable(struct sock_xprt *transport) 745 { 746 747 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 748 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state)) 749 return; 750 if (!xs_poll_socket_readable(transport)) 751 return; 752 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 753 queue_work(xprtiod_workqueue, &transport->recv_worker); 754 } 755 756 static void xs_stream_data_receive(struct sock_xprt *transport) 757 { 758 size_t read = 0; 759 ssize_t ret = 0; 760 761 mutex_lock(&transport->recv_mutex); 762 if (transport->sock == NULL) 763 goto out; 764 for (;;) { 765 ret = xs_read_stream(transport, MSG_DONTWAIT); 766 if (ret < 0) 767 break; 768 read += ret; 769 cond_resched(); 770 } 771 if (ret == -ESHUTDOWN) 772 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 773 else 774 xs_poll_check_readable(transport); 775 out: 776 mutex_unlock(&transport->recv_mutex); 777 trace_xs_stream_read_data(&transport->xprt, ret, read); 778 } 779 780 static void xs_stream_data_receive_workfn(struct work_struct *work) 781 { 782 struct sock_xprt *transport = 783 container_of(work, struct sock_xprt, recv_worker); 784 unsigned int pflags = memalloc_nofs_save(); 785 786 xs_stream_data_receive(transport); 787 memalloc_nofs_restore(pflags); 788 } 789 790 static void 791 xs_stream_reset_connect(struct sock_xprt *transport) 792 { 793 transport->recv.offset = 0; 794 transport->recv.len = 0; 795 transport->recv.copied = 0; 796 transport->xmit.offset = 0; 797 } 798 799 static void 800 xs_stream_start_connect(struct sock_xprt *transport) 801 { 802 transport->xprt.stat.connect_count++; 803 transport->xprt.stat.connect_start = jiffies; 804 } 805 806 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 807 808 /** 809 * xs_nospace - handle transmit was incomplete 810 * @req: pointer to RPC request 811 * @transport: pointer to struct sock_xprt 812 * 813 */ 814 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport) 815 { 816 struct rpc_xprt *xprt = &transport->xprt; 817 struct sock *sk = transport->inet; 818 int ret = -EAGAIN; 819 820 trace_rpc_socket_nospace(req, transport); 821 822 /* Protect against races with write_space */ 823 spin_lock(&xprt->transport_lock); 824 825 /* Don't race with disconnect */ 826 if (xprt_connected(xprt)) { 827 /* wait for more buffer space */ 828 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state); 829 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 830 sk->sk_write_pending++; 831 xprt_wait_for_buffer_space(xprt); 832 } else 833 ret = -ENOTCONN; 834 835 spin_unlock(&xprt->transport_lock); 836 return ret; 837 } 838 839 static int xs_sock_nospace(struct rpc_rqst *req) 840 { 841 struct sock_xprt *transport = 842 container_of(req->rq_xprt, struct sock_xprt, xprt); 843 struct sock *sk = transport->inet; 844 int ret = -EAGAIN; 845 846 lock_sock(sk); 847 if (!sock_writeable(sk)) 848 ret = xs_nospace(req, transport); 849 release_sock(sk); 850 return ret; 851 } 852 853 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait) 854 { 855 struct sock_xprt *transport = 856 container_of(req->rq_xprt, struct sock_xprt, xprt); 857 struct sock *sk = transport->inet; 858 int ret = -EAGAIN; 859 860 if (vm_wait) 861 return -ENOBUFS; 862 lock_sock(sk); 863 if (!sk_stream_memory_free(sk)) 864 ret = xs_nospace(req, transport); 865 release_sock(sk); 866 return ret; 867 } 868 869 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf) 870 { 871 return xdr_alloc_bvec(buf, rpc_task_gfp_mask()); 872 } 873 874 /* 875 * Determine if the previous message in the stream was aborted before it 876 * could complete transmission. 877 */ 878 static bool 879 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req) 880 { 881 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0; 882 } 883 884 /* 885 * Return the stream record marker field for a record of length < 2^31-1 886 */ 887 static rpc_fraghdr 888 xs_stream_record_marker(struct xdr_buf *xdr) 889 { 890 if (!xdr->len) 891 return 0; 892 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len); 893 } 894 895 /** 896 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 897 * @req: pointer to RPC request 898 * 899 * Return values: 900 * 0: The request has been sent 901 * EAGAIN: The socket was blocked, please call again later to 902 * complete the request 903 * ENOTCONN: Caller needs to invoke connect logic then call again 904 * other: Some other error occurred, the request was not sent 905 */ 906 static int xs_local_send_request(struct rpc_rqst *req) 907 { 908 struct rpc_xprt *xprt = req->rq_xprt; 909 struct sock_xprt *transport = 910 container_of(xprt, struct sock_xprt, xprt); 911 struct xdr_buf *xdr = &req->rq_snd_buf; 912 rpc_fraghdr rm = xs_stream_record_marker(xdr); 913 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 914 struct msghdr msg = { 915 .msg_flags = XS_SENDMSG_FLAGS, 916 }; 917 bool vm_wait; 918 unsigned int sent; 919 int status; 920 921 /* Close the stream if the previous transmission was incomplete */ 922 if (xs_send_request_was_aborted(transport, req)) { 923 xprt_force_disconnect(xprt); 924 return -ENOTCONN; 925 } 926 927 xs_pktdump("packet data:", 928 req->rq_svec->iov_base, req->rq_svec->iov_len); 929 930 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false; 931 932 req->rq_xtime = ktime_get(); 933 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 934 transport->xmit.offset, rm, &sent); 935 dprintk("RPC: %s(%u) = %d\n", 936 __func__, xdr->len - transport->xmit.offset, status); 937 938 if (likely(sent > 0) || status == 0) { 939 transport->xmit.offset += sent; 940 req->rq_bytes_sent = transport->xmit.offset; 941 if (likely(req->rq_bytes_sent >= msglen)) { 942 req->rq_xmit_bytes_sent += transport->xmit.offset; 943 transport->xmit.offset = 0; 944 return 0; 945 } 946 status = -EAGAIN; 947 vm_wait = false; 948 } 949 950 switch (status) { 951 case -EAGAIN: 952 status = xs_stream_nospace(req, vm_wait); 953 break; 954 default: 955 dprintk("RPC: sendmsg returned unrecognized error %d\n", 956 -status); 957 fallthrough; 958 case -EPIPE: 959 xprt_force_disconnect(xprt); 960 status = -ENOTCONN; 961 } 962 963 return status; 964 } 965 966 /** 967 * xs_udp_send_request - write an RPC request to a UDP socket 968 * @req: pointer to RPC request 969 * 970 * Return values: 971 * 0: The request has been sent 972 * EAGAIN: The socket was blocked, please call again later to 973 * complete the request 974 * ENOTCONN: Caller needs to invoke connect logic then call again 975 * other: Some other error occurred, the request was not sent 976 */ 977 static int xs_udp_send_request(struct rpc_rqst *req) 978 { 979 struct rpc_xprt *xprt = req->rq_xprt; 980 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 981 struct xdr_buf *xdr = &req->rq_snd_buf; 982 struct msghdr msg = { 983 .msg_name = xs_addr(xprt), 984 .msg_namelen = xprt->addrlen, 985 .msg_flags = XS_SENDMSG_FLAGS, 986 }; 987 unsigned int sent; 988 int status; 989 990 xs_pktdump("packet data:", 991 req->rq_svec->iov_base, 992 req->rq_svec->iov_len); 993 994 if (!xprt_bound(xprt)) 995 return -ENOTCONN; 996 997 if (!xprt_request_get_cong(xprt, req)) 998 return -EBADSLT; 999 1000 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask()); 1001 if (status < 0) 1002 return status; 1003 req->rq_xtime = ktime_get(); 1004 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent); 1005 1006 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 1007 xdr->len, status); 1008 1009 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 1010 if (status == -EPERM) 1011 goto process_status; 1012 1013 if (status == -EAGAIN && sock_writeable(transport->inet)) 1014 status = -ENOBUFS; 1015 1016 if (sent > 0 || status == 0) { 1017 req->rq_xmit_bytes_sent += sent; 1018 if (sent >= req->rq_slen) 1019 return 0; 1020 /* Still some bytes left; set up for a retry later. */ 1021 status = -EAGAIN; 1022 } 1023 1024 process_status: 1025 switch (status) { 1026 case -ENOTSOCK: 1027 status = -ENOTCONN; 1028 /* Should we call xs_close() here? */ 1029 break; 1030 case -EAGAIN: 1031 status = xs_sock_nospace(req); 1032 break; 1033 case -ENETUNREACH: 1034 case -ENOBUFS: 1035 case -EPIPE: 1036 case -ECONNREFUSED: 1037 case -EPERM: 1038 /* When the server has died, an ICMP port unreachable message 1039 * prompts ECONNREFUSED. */ 1040 break; 1041 default: 1042 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1043 -status); 1044 } 1045 1046 return status; 1047 } 1048 1049 /** 1050 * xs_tcp_send_request - write an RPC request to a TCP socket 1051 * @req: pointer to RPC request 1052 * 1053 * Return values: 1054 * 0: The request has been sent 1055 * EAGAIN: The socket was blocked, please call again later to 1056 * complete the request 1057 * ENOTCONN: Caller needs to invoke connect logic then call again 1058 * other: Some other error occurred, the request was not sent 1059 * 1060 * XXX: In the case of soft timeouts, should we eventually give up 1061 * if sendmsg is not able to make progress? 1062 */ 1063 static int xs_tcp_send_request(struct rpc_rqst *req) 1064 { 1065 struct rpc_xprt *xprt = req->rq_xprt; 1066 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1067 struct xdr_buf *xdr = &req->rq_snd_buf; 1068 rpc_fraghdr rm = xs_stream_record_marker(xdr); 1069 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 1070 struct msghdr msg = { 1071 .msg_flags = XS_SENDMSG_FLAGS, 1072 }; 1073 bool vm_wait; 1074 unsigned int sent; 1075 int status; 1076 1077 /* Close the stream if the previous transmission was incomplete */ 1078 if (xs_send_request_was_aborted(transport, req)) { 1079 if (transport->sock != NULL) 1080 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 1081 return -ENOTCONN; 1082 } 1083 if (!transport->inet) 1084 return -ENOTCONN; 1085 1086 xs_pktdump("packet data:", 1087 req->rq_svec->iov_base, 1088 req->rq_svec->iov_len); 1089 1090 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state)) 1091 xs_tcp_set_socket_timeouts(xprt, transport->sock); 1092 1093 xs_set_srcport(transport, transport->sock); 1094 1095 /* Continue transmitting the packet/record. We must be careful 1096 * to cope with writespace callbacks arriving _after_ we have 1097 * called sendmsg(). */ 1098 req->rq_xtime = ktime_get(); 1099 tcp_sock_set_cork(transport->inet, true); 1100 1101 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false; 1102 1103 do { 1104 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 1105 transport->xmit.offset, rm, &sent); 1106 1107 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 1108 xdr->len - transport->xmit.offset, status); 1109 1110 /* If we've sent the entire packet, immediately 1111 * reset the count of bytes sent. */ 1112 transport->xmit.offset += sent; 1113 req->rq_bytes_sent = transport->xmit.offset; 1114 if (likely(req->rq_bytes_sent >= msglen)) { 1115 req->rq_xmit_bytes_sent += transport->xmit.offset; 1116 transport->xmit.offset = 0; 1117 if (atomic_long_read(&xprt->xmit_queuelen) == 1) 1118 tcp_sock_set_cork(transport->inet, false); 1119 return 0; 1120 } 1121 1122 WARN_ON_ONCE(sent == 0 && status == 0); 1123 1124 if (sent > 0) 1125 vm_wait = false; 1126 1127 } while (status == 0); 1128 1129 switch (status) { 1130 case -ENOTSOCK: 1131 status = -ENOTCONN; 1132 /* Should we call xs_close() here? */ 1133 break; 1134 case -EAGAIN: 1135 status = xs_stream_nospace(req, vm_wait); 1136 break; 1137 case -ECONNRESET: 1138 case -ECONNREFUSED: 1139 case -ENOTCONN: 1140 case -EADDRINUSE: 1141 case -ENOBUFS: 1142 case -EPIPE: 1143 break; 1144 default: 1145 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1146 -status); 1147 } 1148 1149 return status; 1150 } 1151 1152 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1153 { 1154 transport->old_data_ready = sk->sk_data_ready; 1155 transport->old_state_change = sk->sk_state_change; 1156 transport->old_write_space = sk->sk_write_space; 1157 transport->old_error_report = sk->sk_error_report; 1158 } 1159 1160 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1161 { 1162 sk->sk_data_ready = transport->old_data_ready; 1163 sk->sk_state_change = transport->old_state_change; 1164 sk->sk_write_space = transport->old_write_space; 1165 sk->sk_error_report = transport->old_error_report; 1166 } 1167 1168 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt) 1169 { 1170 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1171 1172 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 1173 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state); 1174 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state); 1175 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state); 1176 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state); 1177 } 1178 1179 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr) 1180 { 1181 set_bit(nr, &transport->sock_state); 1182 queue_work(xprtiod_workqueue, &transport->error_worker); 1183 } 1184 1185 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1186 { 1187 xprt->connect_cookie++; 1188 smp_mb__before_atomic(); 1189 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1190 clear_bit(XPRT_CLOSING, &xprt->state); 1191 xs_sock_reset_state_flags(xprt); 1192 smp_mb__after_atomic(); 1193 } 1194 1195 /** 1196 * xs_error_report - callback to handle TCP socket state errors 1197 * @sk: socket 1198 * 1199 * Note: we don't call sock_error() since there may be a rpc_task 1200 * using the socket, and so we don't want to clear sk->sk_err. 1201 */ 1202 static void xs_error_report(struct sock *sk) 1203 { 1204 struct sock_xprt *transport; 1205 struct rpc_xprt *xprt; 1206 1207 if (!(xprt = xprt_from_sock(sk))) 1208 return; 1209 1210 transport = container_of(xprt, struct sock_xprt, xprt); 1211 transport->xprt_err = -sk->sk_err; 1212 if (transport->xprt_err == 0) 1213 return; 1214 dprintk("RPC: xs_error_report client %p, error=%d...\n", 1215 xprt, -transport->xprt_err); 1216 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err); 1217 1218 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */ 1219 smp_mb__before_atomic(); 1220 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR); 1221 } 1222 1223 static void xs_reset_transport(struct sock_xprt *transport) 1224 { 1225 struct socket *sock = transport->sock; 1226 struct sock *sk = transport->inet; 1227 struct rpc_xprt *xprt = &transport->xprt; 1228 struct file *filp = transport->file; 1229 1230 if (sk == NULL) 1231 return; 1232 /* 1233 * Make sure we're calling this in a context from which it is safe 1234 * to call __fput_sync(). In practice that means rpciod and the 1235 * system workqueue. 1236 */ 1237 if (!(current->flags & PF_WQ_WORKER)) { 1238 WARN_ON_ONCE(1); 1239 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 1240 return; 1241 } 1242 1243 if (atomic_read(&transport->xprt.swapper)) 1244 sk_clear_memalloc(sk); 1245 1246 kernel_sock_shutdown(sock, SHUT_RDWR); 1247 1248 mutex_lock(&transport->recv_mutex); 1249 lock_sock(sk); 1250 transport->inet = NULL; 1251 transport->sock = NULL; 1252 transport->file = NULL; 1253 1254 sk->sk_user_data = NULL; 1255 1256 xs_restore_old_callbacks(transport, sk); 1257 xprt_clear_connected(xprt); 1258 xs_sock_reset_connection_flags(xprt); 1259 /* Reset stream record info */ 1260 xs_stream_reset_connect(transport); 1261 release_sock(sk); 1262 mutex_unlock(&transport->recv_mutex); 1263 1264 trace_rpc_socket_close(xprt, sock); 1265 __fput_sync(filp); 1266 1267 xprt_disconnect_done(xprt); 1268 } 1269 1270 /** 1271 * xs_close - close a socket 1272 * @xprt: transport 1273 * 1274 * This is used when all requests are complete; ie, no DRC state remains 1275 * on the server we want to save. 1276 * 1277 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 1278 * xs_reset_transport() zeroing the socket from underneath a writer. 1279 */ 1280 static void xs_close(struct rpc_xprt *xprt) 1281 { 1282 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1283 1284 dprintk("RPC: xs_close xprt %p\n", xprt); 1285 1286 xs_reset_transport(transport); 1287 xprt->reestablish_timeout = 0; 1288 } 1289 1290 static void xs_inject_disconnect(struct rpc_xprt *xprt) 1291 { 1292 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 1293 xprt); 1294 xprt_disconnect_done(xprt); 1295 } 1296 1297 static void xs_xprt_free(struct rpc_xprt *xprt) 1298 { 1299 xs_free_peer_addresses(xprt); 1300 xprt_free(xprt); 1301 } 1302 1303 /** 1304 * xs_destroy - prepare to shutdown a transport 1305 * @xprt: doomed transport 1306 * 1307 */ 1308 static void xs_destroy(struct rpc_xprt *xprt) 1309 { 1310 struct sock_xprt *transport = container_of(xprt, 1311 struct sock_xprt, xprt); 1312 dprintk("RPC: xs_destroy xprt %p\n", xprt); 1313 1314 cancel_delayed_work_sync(&transport->connect_worker); 1315 xs_close(xprt); 1316 cancel_work_sync(&transport->recv_worker); 1317 cancel_work_sync(&transport->error_worker); 1318 xs_xprt_free(xprt); 1319 module_put(THIS_MODULE); 1320 } 1321 1322 /** 1323 * xs_udp_data_read_skb - receive callback for UDP sockets 1324 * @xprt: transport 1325 * @sk: socket 1326 * @skb: skbuff 1327 * 1328 */ 1329 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1330 struct sock *sk, 1331 struct sk_buff *skb) 1332 { 1333 struct rpc_task *task; 1334 struct rpc_rqst *rovr; 1335 int repsize, copied; 1336 u32 _xid; 1337 __be32 *xp; 1338 1339 repsize = skb->len; 1340 if (repsize < 4) { 1341 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1342 return; 1343 } 1344 1345 /* Copy the XID from the skb... */ 1346 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid); 1347 if (xp == NULL) 1348 return; 1349 1350 /* Look up and lock the request corresponding to the given XID */ 1351 spin_lock(&xprt->queue_lock); 1352 rovr = xprt_lookup_rqst(xprt, *xp); 1353 if (!rovr) 1354 goto out_unlock; 1355 xprt_pin_rqst(rovr); 1356 xprt_update_rtt(rovr->rq_task); 1357 spin_unlock(&xprt->queue_lock); 1358 task = rovr->rq_task; 1359 1360 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1361 copied = repsize; 1362 1363 /* Suck it into the iovec, verify checksum if not done by hw. */ 1364 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1365 spin_lock(&xprt->queue_lock); 1366 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS); 1367 goto out_unpin; 1368 } 1369 1370 1371 spin_lock(&xprt->transport_lock); 1372 xprt_adjust_cwnd(xprt, task, copied); 1373 spin_unlock(&xprt->transport_lock); 1374 spin_lock(&xprt->queue_lock); 1375 xprt_complete_rqst(task, copied); 1376 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS); 1377 out_unpin: 1378 xprt_unpin_rqst(rovr); 1379 out_unlock: 1380 spin_unlock(&xprt->queue_lock); 1381 } 1382 1383 static void xs_udp_data_receive(struct sock_xprt *transport) 1384 { 1385 struct sk_buff *skb; 1386 struct sock *sk; 1387 int err; 1388 1389 mutex_lock(&transport->recv_mutex); 1390 sk = transport->inet; 1391 if (sk == NULL) 1392 goto out; 1393 for (;;) { 1394 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1395 if (skb == NULL) 1396 break; 1397 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1398 consume_skb(skb); 1399 cond_resched(); 1400 } 1401 xs_poll_check_readable(transport); 1402 out: 1403 mutex_unlock(&transport->recv_mutex); 1404 } 1405 1406 static void xs_udp_data_receive_workfn(struct work_struct *work) 1407 { 1408 struct sock_xprt *transport = 1409 container_of(work, struct sock_xprt, recv_worker); 1410 unsigned int pflags = memalloc_nofs_save(); 1411 1412 xs_udp_data_receive(transport); 1413 memalloc_nofs_restore(pflags); 1414 } 1415 1416 /** 1417 * xs_data_ready - "data ready" callback for sockets 1418 * @sk: socket with data to read 1419 * 1420 */ 1421 static void xs_data_ready(struct sock *sk) 1422 { 1423 struct rpc_xprt *xprt; 1424 1425 trace_sk_data_ready(sk); 1426 1427 xprt = xprt_from_sock(sk); 1428 if (xprt != NULL) { 1429 struct sock_xprt *transport = container_of(xprt, 1430 struct sock_xprt, xprt); 1431 1432 trace_xs_data_ready(xprt); 1433 1434 transport->old_data_ready(sk); 1435 1436 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state)) 1437 return; 1438 1439 /* Any data means we had a useful conversation, so 1440 * then we don't need to delay the next reconnect 1441 */ 1442 if (xprt->reestablish_timeout) 1443 xprt->reestablish_timeout = 0; 1444 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1445 queue_work(xprtiod_workqueue, &transport->recv_worker); 1446 } 1447 } 1448 1449 /* 1450 * Helper function to force a TCP close if the server is sending 1451 * junk and/or it has put us in CLOSE_WAIT 1452 */ 1453 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1454 { 1455 xprt_force_disconnect(xprt); 1456 } 1457 1458 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1459 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt) 1460 { 1461 return PAGE_SIZE; 1462 } 1463 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1464 1465 /** 1466 * xs_local_state_change - callback to handle AF_LOCAL socket state changes 1467 * @sk: socket whose state has changed 1468 * 1469 */ 1470 static void xs_local_state_change(struct sock *sk) 1471 { 1472 struct rpc_xprt *xprt; 1473 struct sock_xprt *transport; 1474 1475 if (!(xprt = xprt_from_sock(sk))) 1476 return; 1477 transport = container_of(xprt, struct sock_xprt, xprt); 1478 if (sk->sk_shutdown & SHUTDOWN_MASK) { 1479 clear_bit(XPRT_CONNECTED, &xprt->state); 1480 /* Trigger the socket release */ 1481 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1482 } 1483 } 1484 1485 /** 1486 * xs_tcp_state_change - callback to handle TCP socket state changes 1487 * @sk: socket whose state has changed 1488 * 1489 */ 1490 static void xs_tcp_state_change(struct sock *sk) 1491 { 1492 struct rpc_xprt *xprt; 1493 struct sock_xprt *transport; 1494 1495 if (!(xprt = xprt_from_sock(sk))) 1496 return; 1497 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1498 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1499 sk->sk_state, xprt_connected(xprt), 1500 sock_flag(sk, SOCK_DEAD), 1501 sock_flag(sk, SOCK_ZAPPED), 1502 sk->sk_shutdown); 1503 1504 transport = container_of(xprt, struct sock_xprt, xprt); 1505 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1506 switch (sk->sk_state) { 1507 case TCP_ESTABLISHED: 1508 if (!xprt_test_and_set_connected(xprt)) { 1509 xprt->connect_cookie++; 1510 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1511 xprt_clear_connecting(xprt); 1512 1513 xprt->stat.connect_count++; 1514 xprt->stat.connect_time += (long)jiffies - 1515 xprt->stat.connect_start; 1516 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING); 1517 } 1518 break; 1519 case TCP_FIN_WAIT1: 1520 /* The client initiated a shutdown of the socket */ 1521 xprt->connect_cookie++; 1522 xprt->reestablish_timeout = 0; 1523 set_bit(XPRT_CLOSING, &xprt->state); 1524 smp_mb__before_atomic(); 1525 clear_bit(XPRT_CONNECTED, &xprt->state); 1526 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1527 smp_mb__after_atomic(); 1528 break; 1529 case TCP_CLOSE_WAIT: 1530 /* The server initiated a shutdown of the socket */ 1531 xprt->connect_cookie++; 1532 clear_bit(XPRT_CONNECTED, &xprt->state); 1533 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1534 fallthrough; 1535 case TCP_CLOSING: 1536 /* 1537 * If the server closed down the connection, make sure that 1538 * we back off before reconnecting 1539 */ 1540 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1541 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1542 break; 1543 case TCP_LAST_ACK: 1544 set_bit(XPRT_CLOSING, &xprt->state); 1545 smp_mb__before_atomic(); 1546 clear_bit(XPRT_CONNECTED, &xprt->state); 1547 smp_mb__after_atomic(); 1548 break; 1549 case TCP_CLOSE: 1550 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1551 &transport->sock_state)) 1552 xprt_clear_connecting(xprt); 1553 clear_bit(XPRT_CLOSING, &xprt->state); 1554 /* Trigger the socket release */ 1555 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1556 } 1557 } 1558 1559 static void xs_write_space(struct sock *sk) 1560 { 1561 struct sock_xprt *transport; 1562 struct rpc_xprt *xprt; 1563 1564 if (!sk->sk_socket) 1565 return; 1566 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1567 1568 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1569 return; 1570 transport = container_of(xprt, struct sock_xprt, xprt); 1571 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state)) 1572 return; 1573 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE); 1574 sk->sk_write_pending--; 1575 } 1576 1577 /** 1578 * xs_udp_write_space - callback invoked when socket buffer space 1579 * becomes available 1580 * @sk: socket whose state has changed 1581 * 1582 * Called when more output buffer space is available for this socket. 1583 * We try not to wake our writers until they can make "significant" 1584 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1585 * with a bunch of small requests. 1586 */ 1587 static void xs_udp_write_space(struct sock *sk) 1588 { 1589 /* from net/core/sock.c:sock_def_write_space */ 1590 if (sock_writeable(sk)) 1591 xs_write_space(sk); 1592 } 1593 1594 /** 1595 * xs_tcp_write_space - callback invoked when socket buffer space 1596 * becomes available 1597 * @sk: socket whose state has changed 1598 * 1599 * Called when more output buffer space is available for this socket. 1600 * We try not to wake our writers until they can make "significant" 1601 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1602 * with a bunch of small requests. 1603 */ 1604 static void xs_tcp_write_space(struct sock *sk) 1605 { 1606 /* from net/core/stream.c:sk_stream_write_space */ 1607 if (sk_stream_is_writeable(sk)) 1608 xs_write_space(sk); 1609 } 1610 1611 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1612 { 1613 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1614 struct sock *sk = transport->inet; 1615 1616 if (transport->rcvsize) { 1617 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1618 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1619 } 1620 if (transport->sndsize) { 1621 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1622 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1623 sk->sk_write_space(sk); 1624 } 1625 } 1626 1627 /** 1628 * xs_udp_set_buffer_size - set send and receive limits 1629 * @xprt: generic transport 1630 * @sndsize: requested size of send buffer, in bytes 1631 * @rcvsize: requested size of receive buffer, in bytes 1632 * 1633 * Set socket send and receive buffer size limits. 1634 */ 1635 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1636 { 1637 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1638 1639 transport->sndsize = 0; 1640 if (sndsize) 1641 transport->sndsize = sndsize + 1024; 1642 transport->rcvsize = 0; 1643 if (rcvsize) 1644 transport->rcvsize = rcvsize + 1024; 1645 1646 xs_udp_do_set_buffer_size(xprt); 1647 } 1648 1649 /** 1650 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1651 * @xprt: controlling transport 1652 * @task: task that timed out 1653 * 1654 * Adjust the congestion window after a retransmit timeout has occurred. 1655 */ 1656 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1657 { 1658 spin_lock(&xprt->transport_lock); 1659 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1660 spin_unlock(&xprt->transport_lock); 1661 } 1662 1663 static int xs_get_random_port(void) 1664 { 1665 unsigned short min = xprt_min_resvport, max = xprt_max_resvport; 1666 unsigned short range; 1667 unsigned short rand; 1668 1669 if (max < min) 1670 return -EADDRINUSE; 1671 range = max - min + 1; 1672 rand = get_random_u32_below(range); 1673 return rand + min; 1674 } 1675 1676 static unsigned short xs_sock_getport(struct socket *sock) 1677 { 1678 struct sockaddr_storage buf; 1679 unsigned short port = 0; 1680 1681 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0) 1682 goto out; 1683 switch (buf.ss_family) { 1684 case AF_INET6: 1685 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1686 break; 1687 case AF_INET: 1688 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1689 } 1690 out: 1691 return port; 1692 } 1693 1694 /** 1695 * xs_set_port - reset the port number in the remote endpoint address 1696 * @xprt: generic transport 1697 * @port: new port number 1698 * 1699 */ 1700 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1701 { 1702 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1703 1704 rpc_set_port(xs_addr(xprt), port); 1705 xs_update_peer_port(xprt); 1706 } 1707 1708 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1709 { 1710 if (transport->srcport == 0 && transport->xprt.reuseport) 1711 transport->srcport = xs_sock_getport(sock); 1712 } 1713 1714 static int xs_get_srcport(struct sock_xprt *transport) 1715 { 1716 int port = transport->srcport; 1717 1718 if (port == 0 && transport->xprt.resvport) 1719 port = xs_get_random_port(); 1720 return port; 1721 } 1722 1723 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt) 1724 { 1725 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1726 unsigned short ret = 0; 1727 mutex_lock(&sock->recv_mutex); 1728 if (sock->sock) 1729 ret = xs_sock_getport(sock->sock); 1730 mutex_unlock(&sock->recv_mutex); 1731 return ret; 1732 } 1733 1734 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen) 1735 { 1736 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1737 union { 1738 struct sockaddr sa; 1739 struct sockaddr_storage st; 1740 } saddr; 1741 int ret = -ENOTCONN; 1742 1743 mutex_lock(&sock->recv_mutex); 1744 if (sock->sock) { 1745 ret = kernel_getsockname(sock->sock, &saddr.sa); 1746 if (ret >= 0) 1747 ret = snprintf(buf, buflen, "%pISc", &saddr.sa); 1748 } 1749 mutex_unlock(&sock->recv_mutex); 1750 return ret; 1751 } 1752 1753 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1754 { 1755 if (transport->srcport != 0) 1756 transport->srcport = 0; 1757 if (!transport->xprt.resvport) 1758 return 0; 1759 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1760 return xprt_max_resvport; 1761 return --port; 1762 } 1763 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1764 { 1765 struct sockaddr_storage myaddr; 1766 int err, nloop = 0; 1767 int port = xs_get_srcport(transport); 1768 unsigned short last; 1769 1770 /* 1771 * If we are asking for any ephemeral port (i.e. port == 0 && 1772 * transport->xprt.resvport == 0), don't bind. Let the local 1773 * port selection happen implicitly when the socket is used 1774 * (for example at connect time). 1775 * 1776 * This ensures that we can continue to establish TCP 1777 * connections even when all local ephemeral ports are already 1778 * a part of some TCP connection. This makes no difference 1779 * for UDP sockets, but also doesn't harm them. 1780 * 1781 * If we're asking for any reserved port (i.e. port == 0 && 1782 * transport->xprt.resvport == 1) xs_get_srcport above will 1783 * ensure that port is non-zero and we will bind as needed. 1784 */ 1785 if (port <= 0) 1786 return port; 1787 1788 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1789 do { 1790 rpc_set_port((struct sockaddr *)&myaddr, port); 1791 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1792 transport->xprt.addrlen); 1793 if (err == 0) { 1794 if (transport->xprt.reuseport) 1795 transport->srcport = port; 1796 break; 1797 } 1798 last = port; 1799 port = xs_next_srcport(transport, port); 1800 if (port > last) 1801 nloop++; 1802 } while (err == -EADDRINUSE && nloop != 2); 1803 1804 if (myaddr.ss_family == AF_INET) 1805 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1806 &((struct sockaddr_in *)&myaddr)->sin_addr, 1807 port, err ? "failed" : "ok", err); 1808 else 1809 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1810 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1811 port, err ? "failed" : "ok", err); 1812 return err; 1813 } 1814 1815 /* 1816 * We don't support autobind on AF_LOCAL sockets 1817 */ 1818 static void xs_local_rpcbind(struct rpc_task *task) 1819 { 1820 xprt_set_bound(task->tk_xprt); 1821 } 1822 1823 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1824 { 1825 } 1826 1827 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1828 static struct lock_class_key xs_key[3]; 1829 static struct lock_class_key xs_slock_key[3]; 1830 1831 static inline void xs_reclassify_socketu(struct socket *sock) 1832 { 1833 struct sock *sk = sock->sk; 1834 1835 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1836 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]); 1837 } 1838 1839 static inline void xs_reclassify_socket4(struct socket *sock) 1840 { 1841 struct sock *sk = sock->sk; 1842 1843 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1844 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]); 1845 } 1846 1847 static inline void xs_reclassify_socket6(struct socket *sock) 1848 { 1849 struct sock *sk = sock->sk; 1850 1851 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1852 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]); 1853 } 1854 1855 static inline void xs_reclassify_socket(int family, struct socket *sock) 1856 { 1857 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk))) 1858 return; 1859 1860 switch (family) { 1861 case AF_LOCAL: 1862 xs_reclassify_socketu(sock); 1863 break; 1864 case AF_INET: 1865 xs_reclassify_socket4(sock); 1866 break; 1867 case AF_INET6: 1868 xs_reclassify_socket6(sock); 1869 break; 1870 } 1871 } 1872 #else 1873 static inline void xs_reclassify_socket(int family, struct socket *sock) 1874 { 1875 } 1876 #endif 1877 1878 static void xs_dummy_setup_socket(struct work_struct *work) 1879 { 1880 } 1881 1882 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1883 struct sock_xprt *transport, int family, int type, 1884 int protocol, bool reuseport) 1885 { 1886 struct file *filp; 1887 struct socket *sock; 1888 int err; 1889 1890 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1891 if (err < 0) { 1892 dprintk("RPC: can't create %d transport socket (%d).\n", 1893 protocol, -err); 1894 goto out; 1895 } 1896 xs_reclassify_socket(family, sock); 1897 1898 if (reuseport) 1899 sock_set_reuseport(sock->sk); 1900 1901 err = xs_bind(transport, sock); 1902 if (err) { 1903 sock_release(sock); 1904 goto out; 1905 } 1906 1907 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1908 if (IS_ERR(filp)) 1909 return ERR_CAST(filp); 1910 transport->file = filp; 1911 1912 return sock; 1913 out: 1914 return ERR_PTR(err); 1915 } 1916 1917 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1918 struct socket *sock) 1919 { 1920 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1921 xprt); 1922 1923 if (!transport->inet) { 1924 struct sock *sk = sock->sk; 1925 1926 lock_sock(sk); 1927 1928 xs_save_old_callbacks(transport, sk); 1929 1930 sk->sk_user_data = xprt; 1931 sk->sk_data_ready = xs_data_ready; 1932 sk->sk_write_space = xs_udp_write_space; 1933 sk->sk_state_change = xs_local_state_change; 1934 sk->sk_error_report = xs_error_report; 1935 sk->sk_use_task_frag = false; 1936 1937 xprt_clear_connected(xprt); 1938 1939 /* Reset to new socket */ 1940 transport->sock = sock; 1941 transport->inet = sk; 1942 1943 release_sock(sk); 1944 } 1945 1946 xs_stream_start_connect(transport); 1947 1948 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1949 } 1950 1951 /** 1952 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1953 * @transport: socket transport to connect 1954 */ 1955 static int xs_local_setup_socket(struct sock_xprt *transport) 1956 { 1957 struct rpc_xprt *xprt = &transport->xprt; 1958 struct file *filp; 1959 struct socket *sock; 1960 int status; 1961 1962 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1963 SOCK_STREAM, 0, &sock, 1); 1964 if (status < 0) { 1965 dprintk("RPC: can't create AF_LOCAL " 1966 "transport socket (%d).\n", -status); 1967 goto out; 1968 } 1969 xs_reclassify_socket(AF_LOCAL, sock); 1970 1971 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1972 if (IS_ERR(filp)) { 1973 status = PTR_ERR(filp); 1974 goto out; 1975 } 1976 transport->file = filp; 1977 1978 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1979 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1980 1981 status = xs_local_finish_connecting(xprt, sock); 1982 trace_rpc_socket_connect(xprt, sock, status); 1983 switch (status) { 1984 case 0: 1985 dprintk("RPC: xprt %p connected to %s\n", 1986 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1987 xprt->stat.connect_count++; 1988 xprt->stat.connect_time += (long)jiffies - 1989 xprt->stat.connect_start; 1990 xprt_set_connected(xprt); 1991 break; 1992 case -ENOBUFS: 1993 break; 1994 case -ENOENT: 1995 dprintk("RPC: xprt %p: socket %s does not exist\n", 1996 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1997 break; 1998 case -ECONNREFUSED: 1999 dprintk("RPC: xprt %p: connection refused for %s\n", 2000 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2001 break; 2002 default: 2003 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 2004 __func__, -status, 2005 xprt->address_strings[RPC_DISPLAY_ADDR]); 2006 } 2007 2008 out: 2009 xprt_clear_connecting(xprt); 2010 xprt_wake_pending_tasks(xprt, status); 2011 return status; 2012 } 2013 2014 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2015 { 2016 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2017 int ret; 2018 2019 if (transport->file) 2020 goto force_disconnect; 2021 2022 if (RPC_IS_ASYNC(task)) { 2023 /* 2024 * We want the AF_LOCAL connect to be resolved in the 2025 * filesystem namespace of the process making the rpc 2026 * call. Thus we connect synchronously. 2027 * 2028 * If we want to support asynchronous AF_LOCAL calls, 2029 * we'll need to figure out how to pass a namespace to 2030 * connect. 2031 */ 2032 rpc_task_set_rpc_status(task, -ENOTCONN); 2033 goto out_wake; 2034 } 2035 ret = xs_local_setup_socket(transport); 2036 if (ret && !RPC_IS_SOFTCONN(task)) 2037 msleep_interruptible(15000); 2038 return; 2039 force_disconnect: 2040 xprt_force_disconnect(xprt); 2041 out_wake: 2042 xprt_clear_connecting(xprt); 2043 xprt_wake_pending_tasks(xprt, -ENOTCONN); 2044 } 2045 2046 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 2047 /* 2048 * Note that this should be called with XPRT_LOCKED held, or recv_mutex 2049 * held, or when we otherwise know that we have exclusive access to the 2050 * socket, to guard against races with xs_reset_transport. 2051 */ 2052 static void xs_set_memalloc(struct rpc_xprt *xprt) 2053 { 2054 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2055 xprt); 2056 2057 /* 2058 * If there's no sock, then we have nothing to set. The 2059 * reconnecting process will get it for us. 2060 */ 2061 if (!transport->inet) 2062 return; 2063 if (atomic_read(&xprt->swapper)) 2064 sk_set_memalloc(transport->inet); 2065 } 2066 2067 /** 2068 * xs_enable_swap - Tag this transport as being used for swap. 2069 * @xprt: transport to tag 2070 * 2071 * Take a reference to this transport on behalf of the rpc_clnt, and 2072 * optionally mark it for swapping if it wasn't already. 2073 */ 2074 static int 2075 xs_enable_swap(struct rpc_xprt *xprt) 2076 { 2077 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2078 2079 mutex_lock(&xs->recv_mutex); 2080 if (atomic_inc_return(&xprt->swapper) == 1 && 2081 xs->inet) 2082 sk_set_memalloc(xs->inet); 2083 mutex_unlock(&xs->recv_mutex); 2084 return 0; 2085 } 2086 2087 /** 2088 * xs_disable_swap - Untag this transport as being used for swap. 2089 * @xprt: transport to tag 2090 * 2091 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2092 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2093 */ 2094 static void 2095 xs_disable_swap(struct rpc_xprt *xprt) 2096 { 2097 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2098 2099 mutex_lock(&xs->recv_mutex); 2100 if (atomic_dec_and_test(&xprt->swapper) && 2101 xs->inet) 2102 sk_clear_memalloc(xs->inet); 2103 mutex_unlock(&xs->recv_mutex); 2104 } 2105 #else 2106 static void xs_set_memalloc(struct rpc_xprt *xprt) 2107 { 2108 } 2109 2110 static int 2111 xs_enable_swap(struct rpc_xprt *xprt) 2112 { 2113 return -EINVAL; 2114 } 2115 2116 static void 2117 xs_disable_swap(struct rpc_xprt *xprt) 2118 { 2119 } 2120 #endif 2121 2122 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2123 { 2124 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2125 2126 if (!transport->inet) { 2127 struct sock *sk = sock->sk; 2128 2129 lock_sock(sk); 2130 2131 xs_save_old_callbacks(transport, sk); 2132 2133 sk->sk_user_data = xprt; 2134 sk->sk_data_ready = xs_data_ready; 2135 sk->sk_write_space = xs_udp_write_space; 2136 sk->sk_use_task_frag = false; 2137 2138 xprt_set_connected(xprt); 2139 2140 /* Reset to new socket */ 2141 transport->sock = sock; 2142 transport->inet = sk; 2143 2144 xs_set_memalloc(xprt); 2145 2146 release_sock(sk); 2147 } 2148 xs_udp_do_set_buffer_size(xprt); 2149 2150 xprt->stat.connect_start = jiffies; 2151 } 2152 2153 static void xs_udp_setup_socket(struct work_struct *work) 2154 { 2155 struct sock_xprt *transport = 2156 container_of(work, struct sock_xprt, connect_worker.work); 2157 struct rpc_xprt *xprt = &transport->xprt; 2158 struct socket *sock; 2159 int status = -EIO; 2160 unsigned int pflags = current->flags; 2161 2162 if (atomic_read(&xprt->swapper)) 2163 current->flags |= PF_MEMALLOC; 2164 sock = xs_create_sock(xprt, transport, 2165 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2166 IPPROTO_UDP, false); 2167 if (IS_ERR(sock)) 2168 goto out; 2169 2170 dprintk("RPC: worker connecting xprt %p via %s to " 2171 "%s (port %s)\n", xprt, 2172 xprt->address_strings[RPC_DISPLAY_PROTO], 2173 xprt->address_strings[RPC_DISPLAY_ADDR], 2174 xprt->address_strings[RPC_DISPLAY_PORT]); 2175 2176 xs_udp_finish_connecting(xprt, sock); 2177 trace_rpc_socket_connect(xprt, sock, 0); 2178 status = 0; 2179 out: 2180 xprt_clear_connecting(xprt); 2181 xprt_unlock_connect(xprt, transport); 2182 xprt_wake_pending_tasks(xprt, status); 2183 current_restore_flags(pflags, PF_MEMALLOC); 2184 } 2185 2186 /** 2187 * xs_tcp_shutdown - gracefully shut down a TCP socket 2188 * @xprt: transport 2189 * 2190 * Initiates a graceful shutdown of the TCP socket by calling the 2191 * equivalent of shutdown(SHUT_RDWR); 2192 */ 2193 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2194 { 2195 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2196 struct socket *sock = transport->sock; 2197 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE; 2198 2199 if (sock == NULL) 2200 return; 2201 if (!xprt->reuseport) { 2202 xs_close(xprt); 2203 return; 2204 } 2205 switch (skst) { 2206 case TCP_FIN_WAIT1: 2207 case TCP_FIN_WAIT2: 2208 case TCP_LAST_ACK: 2209 break; 2210 case TCP_ESTABLISHED: 2211 case TCP_CLOSE_WAIT: 2212 kernel_sock_shutdown(sock, SHUT_RDWR); 2213 trace_rpc_socket_shutdown(xprt, sock); 2214 break; 2215 default: 2216 xs_reset_transport(transport); 2217 } 2218 } 2219 2220 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 2221 struct socket *sock) 2222 { 2223 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2224 unsigned int keepidle; 2225 unsigned int keepcnt; 2226 unsigned int timeo; 2227 2228 spin_lock(&xprt->transport_lock); 2229 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ); 2230 keepcnt = xprt->timeout->to_retries + 1; 2231 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2232 (xprt->timeout->to_retries + 1); 2233 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2234 spin_unlock(&xprt->transport_lock); 2235 2236 /* TCP Keepalive options */ 2237 sock_set_keepalive(sock->sk); 2238 tcp_sock_set_keepidle(sock->sk, keepidle); 2239 tcp_sock_set_keepintvl(sock->sk, keepidle); 2240 tcp_sock_set_keepcnt(sock->sk, keepcnt); 2241 2242 /* TCP user timeout (see RFC5482) */ 2243 tcp_sock_set_user_timeout(sock->sk, timeo); 2244 } 2245 2246 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt, 2247 unsigned long connect_timeout, 2248 unsigned long reconnect_timeout) 2249 { 2250 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2251 struct rpc_timeout to; 2252 unsigned long initval; 2253 2254 spin_lock(&xprt->transport_lock); 2255 if (reconnect_timeout < xprt->max_reconnect_timeout) 2256 xprt->max_reconnect_timeout = reconnect_timeout; 2257 if (connect_timeout < xprt->connect_timeout) { 2258 memcpy(&to, xprt->timeout, sizeof(to)); 2259 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1); 2260 /* Arbitrary lower limit */ 2261 if (initval < XS_TCP_INIT_REEST_TO << 1) 2262 initval = XS_TCP_INIT_REEST_TO << 1; 2263 to.to_initval = initval; 2264 to.to_maxval = initval; 2265 memcpy(&transport->tcp_timeout, &to, 2266 sizeof(transport->tcp_timeout)); 2267 xprt->timeout = &transport->tcp_timeout; 2268 xprt->connect_timeout = connect_timeout; 2269 } 2270 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2271 spin_unlock(&xprt->transport_lock); 2272 } 2273 2274 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2275 { 2276 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2277 2278 if (!transport->inet) { 2279 struct sock *sk = sock->sk; 2280 2281 /* Avoid temporary address, they are bad for long-lived 2282 * connections such as NFS mounts. 2283 * RFC4941, section 3.6 suggests that: 2284 * Individual applications, which have specific 2285 * knowledge about the normal duration of connections, 2286 * MAY override this as appropriate. 2287 */ 2288 if (xs_addr(xprt)->sa_family == PF_INET6) { 2289 ip6_sock_set_addr_preferences(sk, 2290 IPV6_PREFER_SRC_PUBLIC); 2291 } 2292 2293 xs_tcp_set_socket_timeouts(xprt, sock); 2294 tcp_sock_set_nodelay(sk); 2295 2296 lock_sock(sk); 2297 2298 xs_save_old_callbacks(transport, sk); 2299 2300 sk->sk_user_data = xprt; 2301 sk->sk_data_ready = xs_data_ready; 2302 sk->sk_state_change = xs_tcp_state_change; 2303 sk->sk_write_space = xs_tcp_write_space; 2304 sk->sk_error_report = xs_error_report; 2305 sk->sk_use_task_frag = false; 2306 2307 /* socket options */ 2308 sock_reset_flag(sk, SOCK_LINGER); 2309 2310 xprt_clear_connected(xprt); 2311 2312 /* Reset to new socket */ 2313 transport->sock = sock; 2314 transport->inet = sk; 2315 2316 release_sock(sk); 2317 } 2318 2319 if (!xprt_bound(xprt)) 2320 return -ENOTCONN; 2321 2322 xs_set_memalloc(xprt); 2323 2324 xs_stream_start_connect(transport); 2325 2326 /* Tell the socket layer to start connecting... */ 2327 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2328 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2329 } 2330 2331 /** 2332 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2333 * @work: queued work item 2334 * 2335 * Invoked by a work queue tasklet. 2336 */ 2337 static void xs_tcp_setup_socket(struct work_struct *work) 2338 { 2339 struct sock_xprt *transport = 2340 container_of(work, struct sock_xprt, connect_worker.work); 2341 struct socket *sock = transport->sock; 2342 struct rpc_xprt *xprt = &transport->xprt; 2343 int status; 2344 unsigned int pflags = current->flags; 2345 2346 if (atomic_read(&xprt->swapper)) 2347 current->flags |= PF_MEMALLOC; 2348 2349 if (xprt_connected(xprt)) 2350 goto out; 2351 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT, 2352 &transport->sock_state) || 2353 !sock) { 2354 xs_reset_transport(transport); 2355 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family, 2356 SOCK_STREAM, IPPROTO_TCP, true); 2357 if (IS_ERR(sock)) { 2358 xprt_wake_pending_tasks(xprt, PTR_ERR(sock)); 2359 goto out; 2360 } 2361 } 2362 2363 dprintk("RPC: worker connecting xprt %p via %s to " 2364 "%s (port %s)\n", xprt, 2365 xprt->address_strings[RPC_DISPLAY_PROTO], 2366 xprt->address_strings[RPC_DISPLAY_ADDR], 2367 xprt->address_strings[RPC_DISPLAY_PORT]); 2368 2369 status = xs_tcp_finish_connecting(xprt, sock); 2370 trace_rpc_socket_connect(xprt, sock, status); 2371 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2372 xprt, -status, xprt_connected(xprt), 2373 sock->sk->sk_state); 2374 switch (status) { 2375 case 0: 2376 case -EINPROGRESS: 2377 /* SYN_SENT! */ 2378 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state); 2379 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2380 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2381 fallthrough; 2382 case -EALREADY: 2383 goto out_unlock; 2384 case -EADDRNOTAVAIL: 2385 /* Source port number is unavailable. Try a new one! */ 2386 transport->srcport = 0; 2387 status = -EAGAIN; 2388 break; 2389 case -EINVAL: 2390 /* Happens, for instance, if the user specified a link 2391 * local IPv6 address without a scope-id. 2392 */ 2393 case -ECONNREFUSED: 2394 case -ECONNRESET: 2395 case -ENETDOWN: 2396 case -ENETUNREACH: 2397 case -EHOSTUNREACH: 2398 case -EADDRINUSE: 2399 case -ENOBUFS: 2400 break; 2401 default: 2402 printk("%s: connect returned unhandled error %d\n", 2403 __func__, status); 2404 status = -EAGAIN; 2405 } 2406 2407 /* xs_tcp_force_close() wakes tasks with a fixed error code. 2408 * We need to wake them first to ensure the correct error code. 2409 */ 2410 xprt_wake_pending_tasks(xprt, status); 2411 xs_tcp_force_close(xprt); 2412 out: 2413 xprt_clear_connecting(xprt); 2414 out_unlock: 2415 xprt_unlock_connect(xprt, transport); 2416 current_restore_flags(pflags, PF_MEMALLOC); 2417 } 2418 2419 /** 2420 * xs_connect - connect a socket to a remote endpoint 2421 * @xprt: pointer to transport structure 2422 * @task: address of RPC task that manages state of connect request 2423 * 2424 * TCP: If the remote end dropped the connection, delay reconnecting. 2425 * 2426 * UDP socket connects are synchronous, but we use a work queue anyway 2427 * to guarantee that even unprivileged user processes can set up a 2428 * socket on a privileged port. 2429 * 2430 * If a UDP socket connect fails, the delay behavior here prevents 2431 * retry floods (hard mounts). 2432 */ 2433 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2434 { 2435 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2436 unsigned long delay = 0; 2437 2438 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2439 2440 if (transport->sock != NULL) { 2441 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2442 "seconds\n", xprt, xprt->reestablish_timeout / HZ); 2443 2444 delay = xprt_reconnect_delay(xprt); 2445 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO); 2446 2447 } else 2448 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2449 2450 queue_delayed_work(xprtiod_workqueue, 2451 &transport->connect_worker, 2452 delay); 2453 } 2454 2455 static void xs_wake_disconnect(struct sock_xprt *transport) 2456 { 2457 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state)) 2458 xs_tcp_force_close(&transport->xprt); 2459 } 2460 2461 static void xs_wake_write(struct sock_xprt *transport) 2462 { 2463 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state)) 2464 xprt_write_space(&transport->xprt); 2465 } 2466 2467 static void xs_wake_error(struct sock_xprt *transport) 2468 { 2469 int sockerr; 2470 2471 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2472 return; 2473 mutex_lock(&transport->recv_mutex); 2474 if (transport->sock == NULL) 2475 goto out; 2476 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2477 goto out; 2478 sockerr = xchg(&transport->xprt_err, 0); 2479 if (sockerr < 0) 2480 xprt_wake_pending_tasks(&transport->xprt, sockerr); 2481 out: 2482 mutex_unlock(&transport->recv_mutex); 2483 } 2484 2485 static void xs_wake_pending(struct sock_xprt *transport) 2486 { 2487 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state)) 2488 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN); 2489 } 2490 2491 static void xs_error_handle(struct work_struct *work) 2492 { 2493 struct sock_xprt *transport = container_of(work, 2494 struct sock_xprt, error_worker); 2495 2496 xs_wake_disconnect(transport); 2497 xs_wake_write(transport); 2498 xs_wake_error(transport); 2499 xs_wake_pending(transport); 2500 } 2501 2502 /** 2503 * xs_local_print_stats - display AF_LOCAL socket-specific stats 2504 * @xprt: rpc_xprt struct containing statistics 2505 * @seq: output file 2506 * 2507 */ 2508 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2509 { 2510 long idle_time = 0; 2511 2512 if (xprt_connected(xprt)) 2513 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2514 2515 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2516 "%llu %llu %lu %llu %llu\n", 2517 xprt->stat.bind_count, 2518 xprt->stat.connect_count, 2519 xprt->stat.connect_time / HZ, 2520 idle_time, 2521 xprt->stat.sends, 2522 xprt->stat.recvs, 2523 xprt->stat.bad_xids, 2524 xprt->stat.req_u, 2525 xprt->stat.bklog_u, 2526 xprt->stat.max_slots, 2527 xprt->stat.sending_u, 2528 xprt->stat.pending_u); 2529 } 2530 2531 /** 2532 * xs_udp_print_stats - display UDP socket-specific stats 2533 * @xprt: rpc_xprt struct containing statistics 2534 * @seq: output file 2535 * 2536 */ 2537 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2538 { 2539 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2540 2541 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2542 "%lu %llu %llu\n", 2543 transport->srcport, 2544 xprt->stat.bind_count, 2545 xprt->stat.sends, 2546 xprt->stat.recvs, 2547 xprt->stat.bad_xids, 2548 xprt->stat.req_u, 2549 xprt->stat.bklog_u, 2550 xprt->stat.max_slots, 2551 xprt->stat.sending_u, 2552 xprt->stat.pending_u); 2553 } 2554 2555 /** 2556 * xs_tcp_print_stats - display TCP socket-specific stats 2557 * @xprt: rpc_xprt struct containing statistics 2558 * @seq: output file 2559 * 2560 */ 2561 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2562 { 2563 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2564 long idle_time = 0; 2565 2566 if (xprt_connected(xprt)) 2567 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2568 2569 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2570 "%llu %llu %lu %llu %llu\n", 2571 transport->srcport, 2572 xprt->stat.bind_count, 2573 xprt->stat.connect_count, 2574 xprt->stat.connect_time / HZ, 2575 idle_time, 2576 xprt->stat.sends, 2577 xprt->stat.recvs, 2578 xprt->stat.bad_xids, 2579 xprt->stat.req_u, 2580 xprt->stat.bklog_u, 2581 xprt->stat.max_slots, 2582 xprt->stat.sending_u, 2583 xprt->stat.pending_u); 2584 } 2585 2586 /* 2587 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2588 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2589 * to use the server side send routines. 2590 */ 2591 static int bc_malloc(struct rpc_task *task) 2592 { 2593 struct rpc_rqst *rqst = task->tk_rqstp; 2594 size_t size = rqst->rq_callsize; 2595 struct page *page; 2596 struct rpc_buffer *buf; 2597 2598 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) { 2599 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n", 2600 size); 2601 return -EINVAL; 2602 } 2603 2604 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN); 2605 if (!page) 2606 return -ENOMEM; 2607 2608 buf = page_address(page); 2609 buf->len = PAGE_SIZE; 2610 2611 rqst->rq_buffer = buf->data; 2612 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; 2613 return 0; 2614 } 2615 2616 /* 2617 * Free the space allocated in the bc_alloc routine 2618 */ 2619 static void bc_free(struct rpc_task *task) 2620 { 2621 void *buffer = task->tk_rqstp->rq_buffer; 2622 struct rpc_buffer *buf; 2623 2624 buf = container_of(buffer, struct rpc_buffer, data); 2625 free_page((unsigned long)buf); 2626 } 2627 2628 static int bc_sendto(struct rpc_rqst *req) 2629 { 2630 struct xdr_buf *xdr = &req->rq_snd_buf; 2631 struct sock_xprt *transport = 2632 container_of(req->rq_xprt, struct sock_xprt, xprt); 2633 struct msghdr msg = { 2634 .msg_flags = 0, 2635 }; 2636 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 2637 (u32)xdr->len); 2638 unsigned int sent = 0; 2639 int err; 2640 2641 req->rq_xtime = ktime_get(); 2642 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask()); 2643 if (err < 0) 2644 return err; 2645 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent); 2646 xdr_free_bvec(xdr); 2647 if (err < 0 || sent != (xdr->len + sizeof(marker))) 2648 return -EAGAIN; 2649 return sent; 2650 } 2651 2652 /** 2653 * bc_send_request - Send a backchannel Call on a TCP socket 2654 * @req: rpc_rqst containing Call message to be sent 2655 * 2656 * xpt_mutex ensures @rqstp's whole message is written to the socket 2657 * without interruption. 2658 * 2659 * Return values: 2660 * %0 if the message was sent successfully 2661 * %ENOTCONN if the message was not sent 2662 */ 2663 static int bc_send_request(struct rpc_rqst *req) 2664 { 2665 struct svc_xprt *xprt; 2666 int len; 2667 2668 /* 2669 * Get the server socket associated with this callback xprt 2670 */ 2671 xprt = req->rq_xprt->bc_xprt; 2672 2673 /* 2674 * Grab the mutex to serialize data as the connection is shared 2675 * with the fore channel 2676 */ 2677 mutex_lock(&xprt->xpt_mutex); 2678 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2679 len = -ENOTCONN; 2680 else 2681 len = bc_sendto(req); 2682 mutex_unlock(&xprt->xpt_mutex); 2683 2684 if (len > 0) 2685 len = 0; 2686 2687 return len; 2688 } 2689 2690 /* 2691 * The close routine. Since this is client initiated, we do nothing 2692 */ 2693 2694 static void bc_close(struct rpc_xprt *xprt) 2695 { 2696 xprt_disconnect_done(xprt); 2697 } 2698 2699 /* 2700 * The xprt destroy routine. Again, because this connection is client 2701 * initiated, we do nothing 2702 */ 2703 2704 static void bc_destroy(struct rpc_xprt *xprt) 2705 { 2706 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2707 2708 xs_xprt_free(xprt); 2709 module_put(THIS_MODULE); 2710 } 2711 2712 static const struct rpc_xprt_ops xs_local_ops = { 2713 .reserve_xprt = xprt_reserve_xprt, 2714 .release_xprt = xprt_release_xprt, 2715 .alloc_slot = xprt_alloc_slot, 2716 .free_slot = xprt_free_slot, 2717 .rpcbind = xs_local_rpcbind, 2718 .set_port = xs_local_set_port, 2719 .connect = xs_local_connect, 2720 .buf_alloc = rpc_malloc, 2721 .buf_free = rpc_free, 2722 .prepare_request = xs_stream_prepare_request, 2723 .send_request = xs_local_send_request, 2724 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2725 .close = xs_close, 2726 .destroy = xs_destroy, 2727 .print_stats = xs_local_print_stats, 2728 .enable_swap = xs_enable_swap, 2729 .disable_swap = xs_disable_swap, 2730 }; 2731 2732 static const struct rpc_xprt_ops xs_udp_ops = { 2733 .set_buffer_size = xs_udp_set_buffer_size, 2734 .reserve_xprt = xprt_reserve_xprt_cong, 2735 .release_xprt = xprt_release_xprt_cong, 2736 .alloc_slot = xprt_alloc_slot, 2737 .free_slot = xprt_free_slot, 2738 .rpcbind = rpcb_getport_async, 2739 .set_port = xs_set_port, 2740 .connect = xs_connect, 2741 .get_srcaddr = xs_sock_srcaddr, 2742 .get_srcport = xs_sock_srcport, 2743 .buf_alloc = rpc_malloc, 2744 .buf_free = rpc_free, 2745 .send_request = xs_udp_send_request, 2746 .wait_for_reply_request = xprt_wait_for_reply_request_rtt, 2747 .timer = xs_udp_timer, 2748 .release_request = xprt_release_rqst_cong, 2749 .close = xs_close, 2750 .destroy = xs_destroy, 2751 .print_stats = xs_udp_print_stats, 2752 .enable_swap = xs_enable_swap, 2753 .disable_swap = xs_disable_swap, 2754 .inject_disconnect = xs_inject_disconnect, 2755 }; 2756 2757 static const struct rpc_xprt_ops xs_tcp_ops = { 2758 .reserve_xprt = xprt_reserve_xprt, 2759 .release_xprt = xprt_release_xprt, 2760 .alloc_slot = xprt_alloc_slot, 2761 .free_slot = xprt_free_slot, 2762 .rpcbind = rpcb_getport_async, 2763 .set_port = xs_set_port, 2764 .connect = xs_connect, 2765 .get_srcaddr = xs_sock_srcaddr, 2766 .get_srcport = xs_sock_srcport, 2767 .buf_alloc = rpc_malloc, 2768 .buf_free = rpc_free, 2769 .prepare_request = xs_stream_prepare_request, 2770 .send_request = xs_tcp_send_request, 2771 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2772 .close = xs_tcp_shutdown, 2773 .destroy = xs_destroy, 2774 .set_connect_timeout = xs_tcp_set_connect_timeout, 2775 .print_stats = xs_tcp_print_stats, 2776 .enable_swap = xs_enable_swap, 2777 .disable_swap = xs_disable_swap, 2778 .inject_disconnect = xs_inject_disconnect, 2779 #ifdef CONFIG_SUNRPC_BACKCHANNEL 2780 .bc_setup = xprt_setup_bc, 2781 .bc_maxpayload = xs_tcp_bc_maxpayload, 2782 .bc_num_slots = xprt_bc_max_slots, 2783 .bc_free_rqst = xprt_free_bc_rqst, 2784 .bc_destroy = xprt_destroy_bc, 2785 #endif 2786 }; 2787 2788 /* 2789 * The rpc_xprt_ops for the server backchannel 2790 */ 2791 2792 static const struct rpc_xprt_ops bc_tcp_ops = { 2793 .reserve_xprt = xprt_reserve_xprt, 2794 .release_xprt = xprt_release_xprt, 2795 .alloc_slot = xprt_alloc_slot, 2796 .free_slot = xprt_free_slot, 2797 .buf_alloc = bc_malloc, 2798 .buf_free = bc_free, 2799 .send_request = bc_send_request, 2800 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2801 .close = bc_close, 2802 .destroy = bc_destroy, 2803 .print_stats = xs_tcp_print_stats, 2804 .enable_swap = xs_enable_swap, 2805 .disable_swap = xs_disable_swap, 2806 .inject_disconnect = xs_inject_disconnect, 2807 }; 2808 2809 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2810 { 2811 static const struct sockaddr_in sin = { 2812 .sin_family = AF_INET, 2813 .sin_addr.s_addr = htonl(INADDR_ANY), 2814 }; 2815 static const struct sockaddr_in6 sin6 = { 2816 .sin6_family = AF_INET6, 2817 .sin6_addr = IN6ADDR_ANY_INIT, 2818 }; 2819 2820 switch (family) { 2821 case AF_LOCAL: 2822 break; 2823 case AF_INET: 2824 memcpy(sap, &sin, sizeof(sin)); 2825 break; 2826 case AF_INET6: 2827 memcpy(sap, &sin6, sizeof(sin6)); 2828 break; 2829 default: 2830 dprintk("RPC: %s: Bad address family\n", __func__); 2831 return -EAFNOSUPPORT; 2832 } 2833 return 0; 2834 } 2835 2836 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2837 unsigned int slot_table_size, 2838 unsigned int max_slot_table_size) 2839 { 2840 struct rpc_xprt *xprt; 2841 struct sock_xprt *new; 2842 2843 if (args->addrlen > sizeof(xprt->addr)) { 2844 dprintk("RPC: xs_setup_xprt: address too large\n"); 2845 return ERR_PTR(-EBADF); 2846 } 2847 2848 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2849 max_slot_table_size); 2850 if (xprt == NULL) { 2851 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2852 "rpc_xprt\n"); 2853 return ERR_PTR(-ENOMEM); 2854 } 2855 2856 new = container_of(xprt, struct sock_xprt, xprt); 2857 mutex_init(&new->recv_mutex); 2858 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2859 xprt->addrlen = args->addrlen; 2860 if (args->srcaddr) 2861 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2862 else { 2863 int err; 2864 err = xs_init_anyaddr(args->dstaddr->sa_family, 2865 (struct sockaddr *)&new->srcaddr); 2866 if (err != 0) { 2867 xprt_free(xprt); 2868 return ERR_PTR(err); 2869 } 2870 } 2871 2872 return xprt; 2873 } 2874 2875 static const struct rpc_timeout xs_local_default_timeout = { 2876 .to_initval = 10 * HZ, 2877 .to_maxval = 10 * HZ, 2878 .to_retries = 2, 2879 }; 2880 2881 /** 2882 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2883 * @args: rpc transport creation arguments 2884 * 2885 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2886 */ 2887 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2888 { 2889 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2890 struct sock_xprt *transport; 2891 struct rpc_xprt *xprt; 2892 struct rpc_xprt *ret; 2893 2894 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2895 xprt_max_tcp_slot_table_entries); 2896 if (IS_ERR(xprt)) 2897 return xprt; 2898 transport = container_of(xprt, struct sock_xprt, xprt); 2899 2900 xprt->prot = 0; 2901 xprt->xprt_class = &xs_local_transport; 2902 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2903 2904 xprt->bind_timeout = XS_BIND_TO; 2905 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2906 xprt->idle_timeout = XS_IDLE_DISC_TO; 2907 2908 xprt->ops = &xs_local_ops; 2909 xprt->timeout = &xs_local_default_timeout; 2910 2911 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 2912 INIT_WORK(&transport->error_worker, xs_error_handle); 2913 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket); 2914 2915 switch (sun->sun_family) { 2916 case AF_LOCAL: 2917 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') { 2918 dprintk("RPC: bad AF_LOCAL address: %s\n", 2919 sun->sun_path); 2920 ret = ERR_PTR(-EINVAL); 2921 goto out_err; 2922 } 2923 xprt_set_bound(xprt); 2924 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2925 break; 2926 default: 2927 ret = ERR_PTR(-EAFNOSUPPORT); 2928 goto out_err; 2929 } 2930 2931 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2932 xprt->address_strings[RPC_DISPLAY_ADDR]); 2933 2934 if (try_module_get(THIS_MODULE)) 2935 return xprt; 2936 ret = ERR_PTR(-EINVAL); 2937 out_err: 2938 xs_xprt_free(xprt); 2939 return ret; 2940 } 2941 2942 static const struct rpc_timeout xs_udp_default_timeout = { 2943 .to_initval = 5 * HZ, 2944 .to_maxval = 30 * HZ, 2945 .to_increment = 5 * HZ, 2946 .to_retries = 5, 2947 }; 2948 2949 /** 2950 * xs_setup_udp - Set up transport to use a UDP socket 2951 * @args: rpc transport creation arguments 2952 * 2953 */ 2954 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2955 { 2956 struct sockaddr *addr = args->dstaddr; 2957 struct rpc_xprt *xprt; 2958 struct sock_xprt *transport; 2959 struct rpc_xprt *ret; 2960 2961 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2962 xprt_udp_slot_table_entries); 2963 if (IS_ERR(xprt)) 2964 return xprt; 2965 transport = container_of(xprt, struct sock_xprt, xprt); 2966 2967 xprt->prot = IPPROTO_UDP; 2968 xprt->xprt_class = &xs_udp_transport; 2969 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2970 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2971 2972 xprt->bind_timeout = XS_BIND_TO; 2973 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2974 xprt->idle_timeout = XS_IDLE_DISC_TO; 2975 2976 xprt->ops = &xs_udp_ops; 2977 2978 xprt->timeout = &xs_udp_default_timeout; 2979 2980 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 2981 INIT_WORK(&transport->error_worker, xs_error_handle); 2982 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 2983 2984 switch (addr->sa_family) { 2985 case AF_INET: 2986 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2987 xprt_set_bound(xprt); 2988 2989 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2990 break; 2991 case AF_INET6: 2992 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2993 xprt_set_bound(xprt); 2994 2995 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2996 break; 2997 default: 2998 ret = ERR_PTR(-EAFNOSUPPORT); 2999 goto out_err; 3000 } 3001 3002 if (xprt_bound(xprt)) 3003 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3004 xprt->address_strings[RPC_DISPLAY_ADDR], 3005 xprt->address_strings[RPC_DISPLAY_PORT], 3006 xprt->address_strings[RPC_DISPLAY_PROTO]); 3007 else 3008 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3009 xprt->address_strings[RPC_DISPLAY_ADDR], 3010 xprt->address_strings[RPC_DISPLAY_PROTO]); 3011 3012 if (try_module_get(THIS_MODULE)) 3013 return xprt; 3014 ret = ERR_PTR(-EINVAL); 3015 out_err: 3016 xs_xprt_free(xprt); 3017 return ret; 3018 } 3019 3020 static const struct rpc_timeout xs_tcp_default_timeout = { 3021 .to_initval = 60 * HZ, 3022 .to_maxval = 60 * HZ, 3023 .to_retries = 2, 3024 }; 3025 3026 /** 3027 * xs_setup_tcp - Set up transport to use a TCP socket 3028 * @args: rpc transport creation arguments 3029 * 3030 */ 3031 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 3032 { 3033 struct sockaddr *addr = args->dstaddr; 3034 struct rpc_xprt *xprt; 3035 struct sock_xprt *transport; 3036 struct rpc_xprt *ret; 3037 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 3038 3039 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 3040 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 3041 3042 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3043 max_slot_table_size); 3044 if (IS_ERR(xprt)) 3045 return xprt; 3046 transport = container_of(xprt, struct sock_xprt, xprt); 3047 3048 xprt->prot = IPPROTO_TCP; 3049 xprt->xprt_class = &xs_tcp_transport; 3050 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3051 3052 xprt->bind_timeout = XS_BIND_TO; 3053 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 3054 xprt->idle_timeout = XS_IDLE_DISC_TO; 3055 3056 xprt->ops = &xs_tcp_ops; 3057 xprt->timeout = &xs_tcp_default_timeout; 3058 3059 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 3060 xprt->connect_timeout = xprt->timeout->to_initval * 3061 (xprt->timeout->to_retries + 1); 3062 3063 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 3064 INIT_WORK(&transport->error_worker, xs_error_handle); 3065 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 3066 3067 switch (addr->sa_family) { 3068 case AF_INET: 3069 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3070 xprt_set_bound(xprt); 3071 3072 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 3073 break; 3074 case AF_INET6: 3075 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3076 xprt_set_bound(xprt); 3077 3078 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 3079 break; 3080 default: 3081 ret = ERR_PTR(-EAFNOSUPPORT); 3082 goto out_err; 3083 } 3084 3085 if (xprt_bound(xprt)) 3086 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3087 xprt->address_strings[RPC_DISPLAY_ADDR], 3088 xprt->address_strings[RPC_DISPLAY_PORT], 3089 xprt->address_strings[RPC_DISPLAY_PROTO]); 3090 else 3091 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3092 xprt->address_strings[RPC_DISPLAY_ADDR], 3093 xprt->address_strings[RPC_DISPLAY_PROTO]); 3094 3095 if (try_module_get(THIS_MODULE)) 3096 return xprt; 3097 ret = ERR_PTR(-EINVAL); 3098 out_err: 3099 xs_xprt_free(xprt); 3100 return ret; 3101 } 3102 3103 /** 3104 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 3105 * @args: rpc transport creation arguments 3106 * 3107 */ 3108 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 3109 { 3110 struct sockaddr *addr = args->dstaddr; 3111 struct rpc_xprt *xprt; 3112 struct sock_xprt *transport; 3113 struct svc_sock *bc_sock; 3114 struct rpc_xprt *ret; 3115 3116 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3117 xprt_tcp_slot_table_entries); 3118 if (IS_ERR(xprt)) 3119 return xprt; 3120 transport = container_of(xprt, struct sock_xprt, xprt); 3121 3122 xprt->prot = IPPROTO_TCP; 3123 xprt->xprt_class = &xs_bc_tcp_transport; 3124 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3125 xprt->timeout = &xs_tcp_default_timeout; 3126 3127 /* backchannel */ 3128 xprt_set_bound(xprt); 3129 xprt->bind_timeout = 0; 3130 xprt->reestablish_timeout = 0; 3131 xprt->idle_timeout = 0; 3132 3133 xprt->ops = &bc_tcp_ops; 3134 3135 switch (addr->sa_family) { 3136 case AF_INET: 3137 xs_format_peer_addresses(xprt, "tcp", 3138 RPCBIND_NETID_TCP); 3139 break; 3140 case AF_INET6: 3141 xs_format_peer_addresses(xprt, "tcp", 3142 RPCBIND_NETID_TCP6); 3143 break; 3144 default: 3145 ret = ERR_PTR(-EAFNOSUPPORT); 3146 goto out_err; 3147 } 3148 3149 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3150 xprt->address_strings[RPC_DISPLAY_ADDR], 3151 xprt->address_strings[RPC_DISPLAY_PORT], 3152 xprt->address_strings[RPC_DISPLAY_PROTO]); 3153 3154 /* 3155 * Once we've associated a backchannel xprt with a connection, 3156 * we want to keep it around as long as the connection lasts, 3157 * in case we need to start using it for a backchannel again; 3158 * this reference won't be dropped until bc_xprt is destroyed. 3159 */ 3160 xprt_get(xprt); 3161 args->bc_xprt->xpt_bc_xprt = xprt; 3162 xprt->bc_xprt = args->bc_xprt; 3163 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3164 transport->sock = bc_sock->sk_sock; 3165 transport->inet = bc_sock->sk_sk; 3166 3167 /* 3168 * Since we don't want connections for the backchannel, we set 3169 * the xprt status to connected 3170 */ 3171 xprt_set_connected(xprt); 3172 3173 if (try_module_get(THIS_MODULE)) 3174 return xprt; 3175 3176 args->bc_xprt->xpt_bc_xprt = NULL; 3177 args->bc_xprt->xpt_bc_xps = NULL; 3178 xprt_put(xprt); 3179 ret = ERR_PTR(-EINVAL); 3180 out_err: 3181 xs_xprt_free(xprt); 3182 return ret; 3183 } 3184 3185 static struct xprt_class xs_local_transport = { 3186 .list = LIST_HEAD_INIT(xs_local_transport.list), 3187 .name = "named UNIX socket", 3188 .owner = THIS_MODULE, 3189 .ident = XPRT_TRANSPORT_LOCAL, 3190 .setup = xs_setup_local, 3191 .netid = { "" }, 3192 }; 3193 3194 static struct xprt_class xs_udp_transport = { 3195 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3196 .name = "udp", 3197 .owner = THIS_MODULE, 3198 .ident = XPRT_TRANSPORT_UDP, 3199 .setup = xs_setup_udp, 3200 .netid = { "udp", "udp6", "" }, 3201 }; 3202 3203 static struct xprt_class xs_tcp_transport = { 3204 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3205 .name = "tcp", 3206 .owner = THIS_MODULE, 3207 .ident = XPRT_TRANSPORT_TCP, 3208 .setup = xs_setup_tcp, 3209 .netid = { "tcp", "tcp6", "" }, 3210 }; 3211 3212 static struct xprt_class xs_bc_tcp_transport = { 3213 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3214 .name = "tcp NFSv4.1 backchannel", 3215 .owner = THIS_MODULE, 3216 .ident = XPRT_TRANSPORT_BC_TCP, 3217 .setup = xs_setup_bc_tcp, 3218 .netid = { "" }, 3219 }; 3220 3221 /** 3222 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3223 * 3224 */ 3225 int init_socket_xprt(void) 3226 { 3227 if (!sunrpc_table_header) 3228 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table); 3229 3230 xprt_register_transport(&xs_local_transport); 3231 xprt_register_transport(&xs_udp_transport); 3232 xprt_register_transport(&xs_tcp_transport); 3233 xprt_register_transport(&xs_bc_tcp_transport); 3234 3235 return 0; 3236 } 3237 3238 /** 3239 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3240 * 3241 */ 3242 void cleanup_socket_xprt(void) 3243 { 3244 if (sunrpc_table_header) { 3245 unregister_sysctl_table(sunrpc_table_header); 3246 sunrpc_table_header = NULL; 3247 } 3248 3249 xprt_unregister_transport(&xs_local_transport); 3250 xprt_unregister_transport(&xs_udp_transport); 3251 xprt_unregister_transport(&xs_tcp_transport); 3252 xprt_unregister_transport(&xs_bc_tcp_transport); 3253 } 3254 3255 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3256 { 3257 return param_set_uint_minmax(val, kp, 3258 RPC_MIN_RESVPORT, 3259 RPC_MAX_RESVPORT); 3260 } 3261 3262 static const struct kernel_param_ops param_ops_portnr = { 3263 .set = param_set_portnr, 3264 .get = param_get_uint, 3265 }; 3266 3267 #define param_check_portnr(name, p) \ 3268 __param_check(name, p, unsigned int); 3269 3270 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3271 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3272 3273 static int param_set_slot_table_size(const char *val, 3274 const struct kernel_param *kp) 3275 { 3276 return param_set_uint_minmax(val, kp, 3277 RPC_MIN_SLOT_TABLE, 3278 RPC_MAX_SLOT_TABLE); 3279 } 3280 3281 static const struct kernel_param_ops param_ops_slot_table_size = { 3282 .set = param_set_slot_table_size, 3283 .get = param_get_uint, 3284 }; 3285 3286 #define param_check_slot_table_size(name, p) \ 3287 __param_check(name, p, unsigned int); 3288 3289 static int param_set_max_slot_table_size(const char *val, 3290 const struct kernel_param *kp) 3291 { 3292 return param_set_uint_minmax(val, kp, 3293 RPC_MIN_SLOT_TABLE, 3294 RPC_MAX_SLOT_TABLE_LIMIT); 3295 } 3296 3297 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3298 .set = param_set_max_slot_table_size, 3299 .get = param_get_uint, 3300 }; 3301 3302 #define param_check_max_slot_table_size(name, p) \ 3303 __param_check(name, p, unsigned int); 3304 3305 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3306 slot_table_size, 0644); 3307 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3308 max_slot_table_size, 0644); 3309 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3310 slot_table_size, 0644); 3311