1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/net/sunrpc/xprtsock.c 4 * 5 * Client-side transport implementation for sockets. 6 * 7 * TCP callback races fixes (C) 1998 Red Hat 8 * TCP send fixes (C) 1998 Red Hat 9 * TCP NFS related read + write fixes 10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 11 * 12 * Rewrite of larges part of the code in order to stabilize TCP stuff. 13 * Fix behaviour when socket buffer is full. 14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> 15 * 16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> 17 * 18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005. 19 * <gilles.quillard@bull.net> 20 */ 21 22 #include <linux/types.h> 23 #include <linux/string.h> 24 #include <linux/slab.h> 25 #include <linux/module.h> 26 #include <linux/capability.h> 27 #include <linux/pagemap.h> 28 #include <linux/errno.h> 29 #include <linux/socket.h> 30 #include <linux/in.h> 31 #include <linux/net.h> 32 #include <linux/mm.h> 33 #include <linux/un.h> 34 #include <linux/udp.h> 35 #include <linux/tcp.h> 36 #include <linux/sunrpc/clnt.h> 37 #include <linux/sunrpc/addr.h> 38 #include <linux/sunrpc/sched.h> 39 #include <linux/sunrpc/svcsock.h> 40 #include <linux/sunrpc/xprtsock.h> 41 #include <linux/file.h> 42 #ifdef CONFIG_SUNRPC_BACKCHANNEL 43 #include <linux/sunrpc/bc_xprt.h> 44 #endif 45 46 #include <net/sock.h> 47 #include <net/checksum.h> 48 #include <net/udp.h> 49 #include <net/tcp.h> 50 #include <linux/bvec.h> 51 #include <linux/highmem.h> 52 #include <linux/uio.h> 53 #include <linux/sched/mm.h> 54 55 #include <trace/events/sunrpc.h> 56 57 #include "socklib.h" 58 #include "sunrpc.h" 59 60 static void xs_close(struct rpc_xprt *xprt); 61 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock); 62 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 63 struct socket *sock); 64 65 /* 66 * xprtsock tunables 67 */ 68 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 69 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 70 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 71 72 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 73 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 74 75 #define XS_TCP_LINGER_TO (15U * HZ) 76 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 77 78 /* 79 * We can register our own files under /proc/sys/sunrpc by 80 * calling register_sysctl_table() again. The files in that 81 * directory become the union of all files registered there. 82 * 83 * We simply need to make sure that we don't collide with 84 * someone else's file names! 85 */ 86 87 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 88 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 89 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 90 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 91 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 92 93 static struct ctl_table_header *sunrpc_table_header; 94 95 static struct xprt_class xs_local_transport; 96 static struct xprt_class xs_udp_transport; 97 static struct xprt_class xs_tcp_transport; 98 static struct xprt_class xs_bc_tcp_transport; 99 100 /* 101 * FIXME: changing the UDP slot table size should also resize the UDP 102 * socket buffers for existing UDP transports 103 */ 104 static struct ctl_table xs_tunables_table[] = { 105 { 106 .procname = "udp_slot_table_entries", 107 .data = &xprt_udp_slot_table_entries, 108 .maxlen = sizeof(unsigned int), 109 .mode = 0644, 110 .proc_handler = proc_dointvec_minmax, 111 .extra1 = &min_slot_table_size, 112 .extra2 = &max_slot_table_size 113 }, 114 { 115 .procname = "tcp_slot_table_entries", 116 .data = &xprt_tcp_slot_table_entries, 117 .maxlen = sizeof(unsigned int), 118 .mode = 0644, 119 .proc_handler = proc_dointvec_minmax, 120 .extra1 = &min_slot_table_size, 121 .extra2 = &max_slot_table_size 122 }, 123 { 124 .procname = "tcp_max_slot_table_entries", 125 .data = &xprt_max_tcp_slot_table_entries, 126 .maxlen = sizeof(unsigned int), 127 .mode = 0644, 128 .proc_handler = proc_dointvec_minmax, 129 .extra1 = &min_slot_table_size, 130 .extra2 = &max_tcp_slot_table_limit 131 }, 132 { 133 .procname = "min_resvport", 134 .data = &xprt_min_resvport, 135 .maxlen = sizeof(unsigned int), 136 .mode = 0644, 137 .proc_handler = proc_dointvec_minmax, 138 .extra1 = &xprt_min_resvport_limit, 139 .extra2 = &xprt_max_resvport_limit 140 }, 141 { 142 .procname = "max_resvport", 143 .data = &xprt_max_resvport, 144 .maxlen = sizeof(unsigned int), 145 .mode = 0644, 146 .proc_handler = proc_dointvec_minmax, 147 .extra1 = &xprt_min_resvport_limit, 148 .extra2 = &xprt_max_resvport_limit 149 }, 150 { 151 .procname = "tcp_fin_timeout", 152 .data = &xs_tcp_fin_timeout, 153 .maxlen = sizeof(xs_tcp_fin_timeout), 154 .mode = 0644, 155 .proc_handler = proc_dointvec_jiffies, 156 }, 157 { }, 158 }; 159 160 static struct ctl_table sunrpc_table[] = { 161 { 162 .procname = "sunrpc", 163 .mode = 0555, 164 .child = xs_tunables_table 165 }, 166 { }, 167 }; 168 169 /* 170 * Wait duration for a reply from the RPC portmapper. 171 */ 172 #define XS_BIND_TO (60U * HZ) 173 174 /* 175 * Delay if a UDP socket connect error occurs. This is most likely some 176 * kind of resource problem on the local host. 177 */ 178 #define XS_UDP_REEST_TO (2U * HZ) 179 180 /* 181 * The reestablish timeout allows clients to delay for a bit before attempting 182 * to reconnect to a server that just dropped our connection. 183 * 184 * We implement an exponential backoff when trying to reestablish a TCP 185 * transport connection with the server. Some servers like to drop a TCP 186 * connection when they are overworked, so we start with a short timeout and 187 * increase over time if the server is down or not responding. 188 */ 189 #define XS_TCP_INIT_REEST_TO (3U * HZ) 190 191 /* 192 * TCP idle timeout; client drops the transport socket if it is idle 193 * for this long. Note that we also timeout UDP sockets to prevent 194 * holding port numbers when there is no RPC traffic. 195 */ 196 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 197 198 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 199 # undef RPC_DEBUG_DATA 200 # define RPCDBG_FACILITY RPCDBG_TRANS 201 #endif 202 203 #ifdef RPC_DEBUG_DATA 204 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 205 { 206 u8 *buf = (u8 *) packet; 207 int j; 208 209 dprintk("RPC: %s\n", msg); 210 for (j = 0; j < count && j < 128; j += 4) { 211 if (!(j & 31)) { 212 if (j) 213 dprintk("\n"); 214 dprintk("0x%04x ", j); 215 } 216 dprintk("%02x%02x%02x%02x ", 217 buf[j], buf[j+1], buf[j+2], buf[j+3]); 218 } 219 dprintk("\n"); 220 } 221 #else 222 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 223 { 224 /* NOP */ 225 } 226 #endif 227 228 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 229 { 230 return (struct rpc_xprt *) sk->sk_user_data; 231 } 232 233 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 234 { 235 return (struct sockaddr *) &xprt->addr; 236 } 237 238 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 239 { 240 return (struct sockaddr_un *) &xprt->addr; 241 } 242 243 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 244 { 245 return (struct sockaddr_in *) &xprt->addr; 246 } 247 248 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 249 { 250 return (struct sockaddr_in6 *) &xprt->addr; 251 } 252 253 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 254 { 255 struct sockaddr *sap = xs_addr(xprt); 256 struct sockaddr_in6 *sin6; 257 struct sockaddr_in *sin; 258 struct sockaddr_un *sun; 259 char buf[128]; 260 261 switch (sap->sa_family) { 262 case AF_LOCAL: 263 sun = xs_addr_un(xprt); 264 strlcpy(buf, sun->sun_path, sizeof(buf)); 265 xprt->address_strings[RPC_DISPLAY_ADDR] = 266 kstrdup(buf, GFP_KERNEL); 267 break; 268 case AF_INET: 269 (void)rpc_ntop(sap, buf, sizeof(buf)); 270 xprt->address_strings[RPC_DISPLAY_ADDR] = 271 kstrdup(buf, GFP_KERNEL); 272 sin = xs_addr_in(xprt); 273 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 274 break; 275 case AF_INET6: 276 (void)rpc_ntop(sap, buf, sizeof(buf)); 277 xprt->address_strings[RPC_DISPLAY_ADDR] = 278 kstrdup(buf, GFP_KERNEL); 279 sin6 = xs_addr_in6(xprt); 280 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 281 break; 282 default: 283 BUG(); 284 } 285 286 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 287 } 288 289 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 290 { 291 struct sockaddr *sap = xs_addr(xprt); 292 char buf[128]; 293 294 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 295 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 296 297 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 298 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 299 } 300 301 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 302 const char *protocol, 303 const char *netid) 304 { 305 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 306 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 307 xs_format_common_peer_addresses(xprt); 308 xs_format_common_peer_ports(xprt); 309 } 310 311 static void xs_update_peer_port(struct rpc_xprt *xprt) 312 { 313 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 314 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 315 316 xs_format_common_peer_ports(xprt); 317 } 318 319 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 320 { 321 unsigned int i; 322 323 for (i = 0; i < RPC_DISPLAY_MAX; i++) 324 switch (i) { 325 case RPC_DISPLAY_PROTO: 326 case RPC_DISPLAY_NETID: 327 continue; 328 default: 329 kfree(xprt->address_strings[i]); 330 } 331 } 332 333 static size_t 334 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp) 335 { 336 size_t i,n; 337 338 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES)) 339 return want; 340 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT; 341 for (i = 0; i < n; i++) { 342 if (buf->pages[i]) 343 continue; 344 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp); 345 if (!buf->pages[i]) { 346 i *= PAGE_SIZE; 347 return i > buf->page_base ? i - buf->page_base : 0; 348 } 349 } 350 return want; 351 } 352 353 static ssize_t 354 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek) 355 { 356 ssize_t ret; 357 if (seek != 0) 358 iov_iter_advance(&msg->msg_iter, seek); 359 ret = sock_recvmsg(sock, msg, flags); 360 return ret > 0 ? ret + seek : ret; 361 } 362 363 static ssize_t 364 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags, 365 struct kvec *kvec, size_t count, size_t seek) 366 { 367 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count); 368 return xs_sock_recvmsg(sock, msg, flags, seek); 369 } 370 371 static ssize_t 372 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags, 373 struct bio_vec *bvec, unsigned long nr, size_t count, 374 size_t seek) 375 { 376 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count); 377 return xs_sock_recvmsg(sock, msg, flags, seek); 378 } 379 380 static ssize_t 381 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags, 382 size_t count) 383 { 384 iov_iter_discard(&msg->msg_iter, READ, count); 385 return sock_recvmsg(sock, msg, flags); 386 } 387 388 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 389 static void 390 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 391 { 392 struct bvec_iter bi = { 393 .bi_size = count, 394 }; 395 struct bio_vec bv; 396 397 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); 398 for_each_bvec(bv, bvec, bi, bi) 399 flush_dcache_page(bv.bv_page); 400 } 401 #else 402 static inline void 403 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek) 404 { 405 } 406 #endif 407 408 static ssize_t 409 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags, 410 struct xdr_buf *buf, size_t count, size_t seek, size_t *read) 411 { 412 size_t want, seek_init = seek, offset = 0; 413 ssize_t ret; 414 415 want = min_t(size_t, count, buf->head[0].iov_len); 416 if (seek < want) { 417 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek); 418 if (ret <= 0) 419 goto sock_err; 420 offset += ret; 421 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 422 goto out; 423 if (ret != want) 424 goto out; 425 seek = 0; 426 } else { 427 seek -= want; 428 offset += want; 429 } 430 431 want = xs_alloc_sparse_pages( 432 buf, min_t(size_t, count - offset, buf->page_len), 433 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN); 434 if (seek < want) { 435 ret = xs_read_bvec(sock, msg, flags, buf->bvec, 436 xdr_buf_pagecount(buf), 437 want + buf->page_base, 438 seek + buf->page_base); 439 if (ret <= 0) 440 goto sock_err; 441 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base); 442 ret -= buf->page_base; 443 offset += ret; 444 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 445 goto out; 446 if (ret != want) 447 goto out; 448 seek = 0; 449 } else { 450 seek -= want; 451 offset += want; 452 } 453 454 want = min_t(size_t, count - offset, buf->tail[0].iov_len); 455 if (seek < want) { 456 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek); 457 if (ret <= 0) 458 goto sock_err; 459 offset += ret; 460 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 461 goto out; 462 if (ret != want) 463 goto out; 464 } else if (offset < seek_init) 465 offset = seek_init; 466 ret = -EMSGSIZE; 467 out: 468 *read = offset - seek_init; 469 return ret; 470 sock_err: 471 offset += seek; 472 goto out; 473 } 474 475 static void 476 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf) 477 { 478 if (!transport->recv.copied) { 479 if (buf->head[0].iov_len >= transport->recv.offset) 480 memcpy(buf->head[0].iov_base, 481 &transport->recv.xid, 482 transport->recv.offset); 483 transport->recv.copied = transport->recv.offset; 484 } 485 } 486 487 static bool 488 xs_read_stream_request_done(struct sock_xprt *transport) 489 { 490 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT); 491 } 492 493 static void 494 xs_read_stream_check_eor(struct sock_xprt *transport, 495 struct msghdr *msg) 496 { 497 if (xs_read_stream_request_done(transport)) 498 msg->msg_flags |= MSG_EOR; 499 } 500 501 static ssize_t 502 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg, 503 int flags, struct rpc_rqst *req) 504 { 505 struct xdr_buf *buf = &req->rq_private_buf; 506 size_t want, read; 507 ssize_t ret; 508 509 xs_read_header(transport, buf); 510 511 want = transport->recv.len - transport->recv.offset; 512 if (want != 0) { 513 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf, 514 transport->recv.copied + want, 515 transport->recv.copied, 516 &read); 517 transport->recv.offset += read; 518 transport->recv.copied += read; 519 } 520 521 if (transport->recv.offset == transport->recv.len) 522 xs_read_stream_check_eor(transport, msg); 523 524 if (want == 0) 525 return 0; 526 527 switch (ret) { 528 default: 529 break; 530 case -EFAULT: 531 case -EMSGSIZE: 532 msg->msg_flags |= MSG_TRUNC; 533 return read; 534 case 0: 535 return -ESHUTDOWN; 536 } 537 return ret < 0 ? ret : read; 538 } 539 540 static size_t 541 xs_read_stream_headersize(bool isfrag) 542 { 543 if (isfrag) 544 return sizeof(__be32); 545 return 3 * sizeof(__be32); 546 } 547 548 static ssize_t 549 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg, 550 int flags, size_t want, size_t seek) 551 { 552 struct kvec kvec = { 553 .iov_base = &transport->recv.fraghdr, 554 .iov_len = want, 555 }; 556 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek); 557 } 558 559 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 560 static ssize_t 561 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 562 { 563 struct rpc_xprt *xprt = &transport->xprt; 564 struct rpc_rqst *req; 565 ssize_t ret; 566 567 /* Is this transport associated with the backchannel? */ 568 if (!xprt->bc_serv) 569 return -ESHUTDOWN; 570 571 /* Look up and lock the request corresponding to the given XID */ 572 req = xprt_lookup_bc_request(xprt, transport->recv.xid); 573 if (!req) { 574 printk(KERN_WARNING "Callback slot table overflowed\n"); 575 return -ESHUTDOWN; 576 } 577 if (transport->recv.copied && !req->rq_private_buf.len) 578 return -ESHUTDOWN; 579 580 ret = xs_read_stream_request(transport, msg, flags, req); 581 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 582 xprt_complete_bc_request(req, transport->recv.copied); 583 else 584 req->rq_private_buf.len = transport->recv.copied; 585 586 return ret; 587 } 588 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 589 static ssize_t 590 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags) 591 { 592 return -ESHUTDOWN; 593 } 594 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 595 596 static ssize_t 597 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags) 598 { 599 struct rpc_xprt *xprt = &transport->xprt; 600 struct rpc_rqst *req; 601 ssize_t ret = 0; 602 603 /* Look up and lock the request corresponding to the given XID */ 604 spin_lock(&xprt->queue_lock); 605 req = xprt_lookup_rqst(xprt, transport->recv.xid); 606 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) { 607 msg->msg_flags |= MSG_TRUNC; 608 goto out; 609 } 610 xprt_pin_rqst(req); 611 spin_unlock(&xprt->queue_lock); 612 613 ret = xs_read_stream_request(transport, msg, flags, req); 614 615 spin_lock(&xprt->queue_lock); 616 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC)) 617 xprt_complete_rqst(req->rq_task, transport->recv.copied); 618 else 619 req->rq_private_buf.len = transport->recv.copied; 620 xprt_unpin_rqst(req); 621 out: 622 spin_unlock(&xprt->queue_lock); 623 return ret; 624 } 625 626 static ssize_t 627 xs_read_stream(struct sock_xprt *transport, int flags) 628 { 629 struct msghdr msg = { 0 }; 630 size_t want, read = 0; 631 ssize_t ret = 0; 632 633 if (transport->recv.len == 0) { 634 want = xs_read_stream_headersize(transport->recv.copied != 0); 635 ret = xs_read_stream_header(transport, &msg, flags, want, 636 transport->recv.offset); 637 if (ret <= 0) 638 goto out_err; 639 transport->recv.offset = ret; 640 if (transport->recv.offset != want) 641 return transport->recv.offset; 642 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) & 643 RPC_FRAGMENT_SIZE_MASK; 644 transport->recv.offset -= sizeof(transport->recv.fraghdr); 645 read = ret; 646 } 647 648 switch (be32_to_cpu(transport->recv.calldir)) { 649 default: 650 msg.msg_flags |= MSG_TRUNC; 651 break; 652 case RPC_CALL: 653 ret = xs_read_stream_call(transport, &msg, flags); 654 break; 655 case RPC_REPLY: 656 ret = xs_read_stream_reply(transport, &msg, flags); 657 } 658 if (msg.msg_flags & MSG_TRUNC) { 659 transport->recv.calldir = cpu_to_be32(-1); 660 transport->recv.copied = -1; 661 } 662 if (ret < 0) 663 goto out_err; 664 read += ret; 665 if (transport->recv.offset < transport->recv.len) { 666 if (!(msg.msg_flags & MSG_TRUNC)) 667 return read; 668 msg.msg_flags = 0; 669 ret = xs_read_discard(transport->sock, &msg, flags, 670 transport->recv.len - transport->recv.offset); 671 if (ret <= 0) 672 goto out_err; 673 transport->recv.offset += ret; 674 read += ret; 675 if (transport->recv.offset != transport->recv.len) 676 return read; 677 } 678 if (xs_read_stream_request_done(transport)) { 679 trace_xs_stream_read_request(transport); 680 transport->recv.copied = 0; 681 } 682 transport->recv.offset = 0; 683 transport->recv.len = 0; 684 return read; 685 out_err: 686 return ret != 0 ? ret : -ESHUTDOWN; 687 } 688 689 static __poll_t xs_poll_socket(struct sock_xprt *transport) 690 { 691 return transport->sock->ops->poll(transport->file, transport->sock, 692 NULL); 693 } 694 695 static bool xs_poll_socket_readable(struct sock_xprt *transport) 696 { 697 __poll_t events = xs_poll_socket(transport); 698 699 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP); 700 } 701 702 static void xs_poll_check_readable(struct sock_xprt *transport) 703 { 704 705 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 706 if (!xs_poll_socket_readable(transport)) 707 return; 708 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 709 queue_work(xprtiod_workqueue, &transport->recv_worker); 710 } 711 712 static void xs_stream_data_receive(struct sock_xprt *transport) 713 { 714 size_t read = 0; 715 ssize_t ret = 0; 716 717 mutex_lock(&transport->recv_mutex); 718 if (transport->sock == NULL) 719 goto out; 720 for (;;) { 721 ret = xs_read_stream(transport, MSG_DONTWAIT); 722 if (ret < 0) 723 break; 724 read += ret; 725 cond_resched(); 726 } 727 if (ret == -ESHUTDOWN) 728 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 729 else 730 xs_poll_check_readable(transport); 731 out: 732 mutex_unlock(&transport->recv_mutex); 733 trace_xs_stream_read_data(&transport->xprt, ret, read); 734 } 735 736 static void xs_stream_data_receive_workfn(struct work_struct *work) 737 { 738 struct sock_xprt *transport = 739 container_of(work, struct sock_xprt, recv_worker); 740 unsigned int pflags = memalloc_nofs_save(); 741 742 xs_stream_data_receive(transport); 743 memalloc_nofs_restore(pflags); 744 } 745 746 static void 747 xs_stream_reset_connect(struct sock_xprt *transport) 748 { 749 transport->recv.offset = 0; 750 transport->recv.len = 0; 751 transport->recv.copied = 0; 752 transport->xmit.offset = 0; 753 } 754 755 static void 756 xs_stream_start_connect(struct sock_xprt *transport) 757 { 758 transport->xprt.stat.connect_count++; 759 transport->xprt.stat.connect_start = jiffies; 760 } 761 762 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 763 764 /** 765 * xs_nospace - handle transmit was incomplete 766 * @req: pointer to RPC request 767 * @transport: pointer to struct sock_xprt 768 * 769 */ 770 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport) 771 { 772 struct rpc_xprt *xprt = &transport->xprt; 773 struct sock *sk = transport->inet; 774 int ret = -EAGAIN; 775 776 trace_rpc_socket_nospace(req, transport); 777 778 /* Protect against races with write_space */ 779 spin_lock(&xprt->transport_lock); 780 781 /* Don't race with disconnect */ 782 if (xprt_connected(xprt)) { 783 /* wait for more buffer space */ 784 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state); 785 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 786 sk->sk_write_pending++; 787 xprt_wait_for_buffer_space(xprt); 788 } else 789 ret = -ENOTCONN; 790 791 spin_unlock(&xprt->transport_lock); 792 return ret; 793 } 794 795 static int xs_sock_nospace(struct rpc_rqst *req) 796 { 797 struct sock_xprt *transport = 798 container_of(req->rq_xprt, struct sock_xprt, xprt); 799 struct sock *sk = transport->inet; 800 int ret = -EAGAIN; 801 802 lock_sock(sk); 803 if (!sock_writeable(sk)) 804 ret = xs_nospace(req, transport); 805 release_sock(sk); 806 return ret; 807 } 808 809 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait) 810 { 811 struct sock_xprt *transport = 812 container_of(req->rq_xprt, struct sock_xprt, xprt); 813 struct sock *sk = transport->inet; 814 int ret = -EAGAIN; 815 816 if (vm_wait) 817 return -ENOBUFS; 818 lock_sock(sk); 819 if (!sk_stream_memory_free(sk)) 820 ret = xs_nospace(req, transport); 821 release_sock(sk); 822 return ret; 823 } 824 825 static int 826 xs_stream_prepare_request(struct rpc_rqst *req) 827 { 828 gfp_t gfp = rpc_task_gfp_mask(); 829 int ret; 830 831 ret = xdr_alloc_bvec(&req->rq_snd_buf, gfp); 832 if (ret < 0) 833 return ret; 834 xdr_free_bvec(&req->rq_rcv_buf); 835 return xdr_alloc_bvec(&req->rq_rcv_buf, gfp); 836 } 837 838 /* 839 * Determine if the previous message in the stream was aborted before it 840 * could complete transmission. 841 */ 842 static bool 843 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req) 844 { 845 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0; 846 } 847 848 /* 849 * Return the stream record marker field for a record of length < 2^31-1 850 */ 851 static rpc_fraghdr 852 xs_stream_record_marker(struct xdr_buf *xdr) 853 { 854 if (!xdr->len) 855 return 0; 856 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len); 857 } 858 859 /** 860 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 861 * @req: pointer to RPC request 862 * 863 * Return values: 864 * 0: The request has been sent 865 * EAGAIN: The socket was blocked, please call again later to 866 * complete the request 867 * ENOTCONN: Caller needs to invoke connect logic then call again 868 * other: Some other error occurred, the request was not sent 869 */ 870 static int xs_local_send_request(struct rpc_rqst *req) 871 { 872 struct rpc_xprt *xprt = req->rq_xprt; 873 struct sock_xprt *transport = 874 container_of(xprt, struct sock_xprt, xprt); 875 struct xdr_buf *xdr = &req->rq_snd_buf; 876 rpc_fraghdr rm = xs_stream_record_marker(xdr); 877 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 878 struct msghdr msg = { 879 .msg_flags = XS_SENDMSG_FLAGS, 880 }; 881 bool vm_wait; 882 unsigned int sent; 883 int status; 884 885 /* Close the stream if the previous transmission was incomplete */ 886 if (xs_send_request_was_aborted(transport, req)) { 887 xprt_force_disconnect(xprt); 888 return -ENOTCONN; 889 } 890 891 xs_pktdump("packet data:", 892 req->rq_svec->iov_base, req->rq_svec->iov_len); 893 894 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false; 895 896 req->rq_xtime = ktime_get(); 897 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 898 transport->xmit.offset, rm, &sent); 899 dprintk("RPC: %s(%u) = %d\n", 900 __func__, xdr->len - transport->xmit.offset, status); 901 902 if (likely(sent > 0) || status == 0) { 903 transport->xmit.offset += sent; 904 req->rq_bytes_sent = transport->xmit.offset; 905 if (likely(req->rq_bytes_sent >= msglen)) { 906 req->rq_xmit_bytes_sent += transport->xmit.offset; 907 transport->xmit.offset = 0; 908 return 0; 909 } 910 status = -EAGAIN; 911 vm_wait = false; 912 } 913 914 switch (status) { 915 case -EAGAIN: 916 status = xs_stream_nospace(req, vm_wait); 917 break; 918 default: 919 dprintk("RPC: sendmsg returned unrecognized error %d\n", 920 -status); 921 fallthrough; 922 case -EPIPE: 923 xprt_force_disconnect(xprt); 924 status = -ENOTCONN; 925 } 926 927 return status; 928 } 929 930 /** 931 * xs_udp_send_request - write an RPC request to a UDP socket 932 * @req: pointer to RPC request 933 * 934 * Return values: 935 * 0: The request has been sent 936 * EAGAIN: The socket was blocked, please call again later to 937 * complete the request 938 * ENOTCONN: Caller needs to invoke connect logic then call again 939 * other: Some other error occurred, the request was not sent 940 */ 941 static int xs_udp_send_request(struct rpc_rqst *req) 942 { 943 struct rpc_xprt *xprt = req->rq_xprt; 944 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 945 struct xdr_buf *xdr = &req->rq_snd_buf; 946 struct msghdr msg = { 947 .msg_name = xs_addr(xprt), 948 .msg_namelen = xprt->addrlen, 949 .msg_flags = XS_SENDMSG_FLAGS, 950 }; 951 unsigned int sent; 952 int status; 953 954 xs_pktdump("packet data:", 955 req->rq_svec->iov_base, 956 req->rq_svec->iov_len); 957 958 if (!xprt_bound(xprt)) 959 return -ENOTCONN; 960 961 if (!xprt_request_get_cong(xprt, req)) 962 return -EBADSLT; 963 964 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask()); 965 if (status < 0) 966 return status; 967 req->rq_xtime = ktime_get(); 968 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent); 969 970 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 971 xdr->len, status); 972 973 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 974 if (status == -EPERM) 975 goto process_status; 976 977 if (status == -EAGAIN && sock_writeable(transport->inet)) 978 status = -ENOBUFS; 979 980 if (sent > 0 || status == 0) { 981 req->rq_xmit_bytes_sent += sent; 982 if (sent >= req->rq_slen) 983 return 0; 984 /* Still some bytes left; set up for a retry later. */ 985 status = -EAGAIN; 986 } 987 988 process_status: 989 switch (status) { 990 case -ENOTSOCK: 991 status = -ENOTCONN; 992 /* Should we call xs_close() here? */ 993 break; 994 case -EAGAIN: 995 status = xs_sock_nospace(req); 996 break; 997 case -ENETUNREACH: 998 case -ENOBUFS: 999 case -EPIPE: 1000 case -ECONNREFUSED: 1001 case -EPERM: 1002 /* When the server has died, an ICMP port unreachable message 1003 * prompts ECONNREFUSED. */ 1004 break; 1005 default: 1006 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1007 -status); 1008 } 1009 1010 return status; 1011 } 1012 1013 /** 1014 * xs_tcp_send_request - write an RPC request to a TCP socket 1015 * @req: pointer to RPC request 1016 * 1017 * Return values: 1018 * 0: The request has been sent 1019 * EAGAIN: The socket was blocked, please call again later to 1020 * complete the request 1021 * ENOTCONN: Caller needs to invoke connect logic then call again 1022 * other: Some other error occurred, the request was not sent 1023 * 1024 * XXX: In the case of soft timeouts, should we eventually give up 1025 * if sendmsg is not able to make progress? 1026 */ 1027 static int xs_tcp_send_request(struct rpc_rqst *req) 1028 { 1029 struct rpc_xprt *xprt = req->rq_xprt; 1030 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1031 struct xdr_buf *xdr = &req->rq_snd_buf; 1032 rpc_fraghdr rm = xs_stream_record_marker(xdr); 1033 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen; 1034 struct msghdr msg = { 1035 .msg_flags = XS_SENDMSG_FLAGS, 1036 }; 1037 bool vm_wait; 1038 unsigned int sent; 1039 int status; 1040 1041 /* Close the stream if the previous transmission was incomplete */ 1042 if (xs_send_request_was_aborted(transport, req)) { 1043 if (transport->sock != NULL) 1044 kernel_sock_shutdown(transport->sock, SHUT_RDWR); 1045 return -ENOTCONN; 1046 } 1047 if (!transport->inet) 1048 return -ENOTCONN; 1049 1050 xs_pktdump("packet data:", 1051 req->rq_svec->iov_base, 1052 req->rq_svec->iov_len); 1053 1054 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state)) 1055 xs_tcp_set_socket_timeouts(xprt, transport->sock); 1056 1057 xs_set_srcport(transport, transport->sock); 1058 1059 /* Continue transmitting the packet/record. We must be careful 1060 * to cope with writespace callbacks arriving _after_ we have 1061 * called sendmsg(). */ 1062 req->rq_xtime = ktime_get(); 1063 tcp_sock_set_cork(transport->inet, true); 1064 1065 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false; 1066 1067 do { 1068 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 1069 transport->xmit.offset, rm, &sent); 1070 1071 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 1072 xdr->len - transport->xmit.offset, status); 1073 1074 /* If we've sent the entire packet, immediately 1075 * reset the count of bytes sent. */ 1076 transport->xmit.offset += sent; 1077 req->rq_bytes_sent = transport->xmit.offset; 1078 if (likely(req->rq_bytes_sent >= msglen)) { 1079 req->rq_xmit_bytes_sent += transport->xmit.offset; 1080 transport->xmit.offset = 0; 1081 if (atomic_long_read(&xprt->xmit_queuelen) == 1) 1082 tcp_sock_set_cork(transport->inet, false); 1083 return 0; 1084 } 1085 1086 WARN_ON_ONCE(sent == 0 && status == 0); 1087 1088 if (sent > 0) 1089 vm_wait = false; 1090 1091 } while (status == 0); 1092 1093 switch (status) { 1094 case -ENOTSOCK: 1095 status = -ENOTCONN; 1096 /* Should we call xs_close() here? */ 1097 break; 1098 case -EAGAIN: 1099 status = xs_stream_nospace(req, vm_wait); 1100 break; 1101 case -ECONNRESET: 1102 case -ECONNREFUSED: 1103 case -ENOTCONN: 1104 case -EADDRINUSE: 1105 case -ENOBUFS: 1106 case -EPIPE: 1107 break; 1108 default: 1109 dprintk("RPC: sendmsg returned unrecognized error %d\n", 1110 -status); 1111 } 1112 1113 return status; 1114 } 1115 1116 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1117 { 1118 transport->old_data_ready = sk->sk_data_ready; 1119 transport->old_state_change = sk->sk_state_change; 1120 transport->old_write_space = sk->sk_write_space; 1121 transport->old_error_report = sk->sk_error_report; 1122 } 1123 1124 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 1125 { 1126 sk->sk_data_ready = transport->old_data_ready; 1127 sk->sk_state_change = transport->old_state_change; 1128 sk->sk_write_space = transport->old_write_space; 1129 sk->sk_error_report = transport->old_error_report; 1130 } 1131 1132 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt) 1133 { 1134 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1135 1136 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 1137 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state); 1138 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state); 1139 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state); 1140 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state); 1141 } 1142 1143 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr) 1144 { 1145 set_bit(nr, &transport->sock_state); 1146 queue_work(xprtiod_workqueue, &transport->error_worker); 1147 } 1148 1149 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1150 { 1151 xprt->connect_cookie++; 1152 smp_mb__before_atomic(); 1153 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1154 clear_bit(XPRT_CLOSING, &xprt->state); 1155 xs_sock_reset_state_flags(xprt); 1156 smp_mb__after_atomic(); 1157 } 1158 1159 /** 1160 * xs_error_report - callback to handle TCP socket state errors 1161 * @sk: socket 1162 * 1163 * Note: we don't call sock_error() since there may be a rpc_task 1164 * using the socket, and so we don't want to clear sk->sk_err. 1165 */ 1166 static void xs_error_report(struct sock *sk) 1167 { 1168 struct sock_xprt *transport; 1169 struct rpc_xprt *xprt; 1170 1171 if (!(xprt = xprt_from_sock(sk))) 1172 return; 1173 1174 transport = container_of(xprt, struct sock_xprt, xprt); 1175 transport->xprt_err = -sk->sk_err; 1176 if (transport->xprt_err == 0) 1177 return; 1178 dprintk("RPC: xs_error_report client %p, error=%d...\n", 1179 xprt, -transport->xprt_err); 1180 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err); 1181 1182 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */ 1183 smp_mb__before_atomic(); 1184 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR); 1185 } 1186 1187 static void xs_reset_transport(struct sock_xprt *transport) 1188 { 1189 struct socket *sock = transport->sock; 1190 struct sock *sk = transport->inet; 1191 struct rpc_xprt *xprt = &transport->xprt; 1192 struct file *filp = transport->file; 1193 1194 if (sk == NULL) 1195 return; 1196 /* 1197 * Make sure we're calling this in a context from which it is safe 1198 * to call __fput_sync(). In practice that means rpciod and the 1199 * system workqueue. 1200 */ 1201 if (!(current->flags & PF_WQ_WORKER)) { 1202 WARN_ON_ONCE(1); 1203 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 1204 return; 1205 } 1206 1207 if (atomic_read(&transport->xprt.swapper)) 1208 sk_clear_memalloc(sk); 1209 1210 kernel_sock_shutdown(sock, SHUT_RDWR); 1211 1212 mutex_lock(&transport->recv_mutex); 1213 lock_sock(sk); 1214 transport->inet = NULL; 1215 transport->sock = NULL; 1216 transport->file = NULL; 1217 1218 sk->sk_user_data = NULL; 1219 1220 xs_restore_old_callbacks(transport, sk); 1221 xprt_clear_connected(xprt); 1222 xs_sock_reset_connection_flags(xprt); 1223 /* Reset stream record info */ 1224 xs_stream_reset_connect(transport); 1225 release_sock(sk); 1226 mutex_unlock(&transport->recv_mutex); 1227 1228 trace_rpc_socket_close(xprt, sock); 1229 __fput_sync(filp); 1230 1231 xprt_disconnect_done(xprt); 1232 } 1233 1234 /** 1235 * xs_close - close a socket 1236 * @xprt: transport 1237 * 1238 * This is used when all requests are complete; ie, no DRC state remains 1239 * on the server we want to save. 1240 * 1241 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 1242 * xs_reset_transport() zeroing the socket from underneath a writer. 1243 */ 1244 static void xs_close(struct rpc_xprt *xprt) 1245 { 1246 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1247 1248 dprintk("RPC: xs_close xprt %p\n", xprt); 1249 1250 xs_reset_transport(transport); 1251 xprt->reestablish_timeout = 0; 1252 } 1253 1254 static void xs_inject_disconnect(struct rpc_xprt *xprt) 1255 { 1256 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 1257 xprt); 1258 xprt_disconnect_done(xprt); 1259 } 1260 1261 static void xs_xprt_free(struct rpc_xprt *xprt) 1262 { 1263 xs_free_peer_addresses(xprt); 1264 xprt_free(xprt); 1265 } 1266 1267 /** 1268 * xs_destroy - prepare to shutdown a transport 1269 * @xprt: doomed transport 1270 * 1271 */ 1272 static void xs_destroy(struct rpc_xprt *xprt) 1273 { 1274 struct sock_xprt *transport = container_of(xprt, 1275 struct sock_xprt, xprt); 1276 dprintk("RPC: xs_destroy xprt %p\n", xprt); 1277 1278 cancel_delayed_work_sync(&transport->connect_worker); 1279 xs_close(xprt); 1280 cancel_work_sync(&transport->recv_worker); 1281 cancel_work_sync(&transport->error_worker); 1282 xs_xprt_free(xprt); 1283 module_put(THIS_MODULE); 1284 } 1285 1286 /** 1287 * xs_udp_data_read_skb - receive callback for UDP sockets 1288 * @xprt: transport 1289 * @sk: socket 1290 * @skb: skbuff 1291 * 1292 */ 1293 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1294 struct sock *sk, 1295 struct sk_buff *skb) 1296 { 1297 struct rpc_task *task; 1298 struct rpc_rqst *rovr; 1299 int repsize, copied; 1300 u32 _xid; 1301 __be32 *xp; 1302 1303 repsize = skb->len; 1304 if (repsize < 4) { 1305 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1306 return; 1307 } 1308 1309 /* Copy the XID from the skb... */ 1310 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid); 1311 if (xp == NULL) 1312 return; 1313 1314 /* Look up and lock the request corresponding to the given XID */ 1315 spin_lock(&xprt->queue_lock); 1316 rovr = xprt_lookup_rqst(xprt, *xp); 1317 if (!rovr) 1318 goto out_unlock; 1319 xprt_pin_rqst(rovr); 1320 xprt_update_rtt(rovr->rq_task); 1321 spin_unlock(&xprt->queue_lock); 1322 task = rovr->rq_task; 1323 1324 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1325 copied = repsize; 1326 1327 /* Suck it into the iovec, verify checksum if not done by hw. */ 1328 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1329 spin_lock(&xprt->queue_lock); 1330 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS); 1331 goto out_unpin; 1332 } 1333 1334 1335 spin_lock(&xprt->transport_lock); 1336 xprt_adjust_cwnd(xprt, task, copied); 1337 spin_unlock(&xprt->transport_lock); 1338 spin_lock(&xprt->queue_lock); 1339 xprt_complete_rqst(task, copied); 1340 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS); 1341 out_unpin: 1342 xprt_unpin_rqst(rovr); 1343 out_unlock: 1344 spin_unlock(&xprt->queue_lock); 1345 } 1346 1347 static void xs_udp_data_receive(struct sock_xprt *transport) 1348 { 1349 struct sk_buff *skb; 1350 struct sock *sk; 1351 int err; 1352 1353 mutex_lock(&transport->recv_mutex); 1354 sk = transport->inet; 1355 if (sk == NULL) 1356 goto out; 1357 for (;;) { 1358 skb = skb_recv_udp(sk, 0, 1, &err); 1359 if (skb == NULL) 1360 break; 1361 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1362 consume_skb(skb); 1363 cond_resched(); 1364 } 1365 xs_poll_check_readable(transport); 1366 out: 1367 mutex_unlock(&transport->recv_mutex); 1368 } 1369 1370 static void xs_udp_data_receive_workfn(struct work_struct *work) 1371 { 1372 struct sock_xprt *transport = 1373 container_of(work, struct sock_xprt, recv_worker); 1374 unsigned int pflags = memalloc_nofs_save(); 1375 1376 xs_udp_data_receive(transport); 1377 memalloc_nofs_restore(pflags); 1378 } 1379 1380 /** 1381 * xs_data_ready - "data ready" callback for UDP sockets 1382 * @sk: socket with data to read 1383 * 1384 */ 1385 static void xs_data_ready(struct sock *sk) 1386 { 1387 struct rpc_xprt *xprt; 1388 1389 dprintk("RPC: xs_data_ready...\n"); 1390 xprt = xprt_from_sock(sk); 1391 if (xprt != NULL) { 1392 struct sock_xprt *transport = container_of(xprt, 1393 struct sock_xprt, xprt); 1394 transport->old_data_ready(sk); 1395 /* Any data means we had a useful conversation, so 1396 * then we don't need to delay the next reconnect 1397 */ 1398 if (xprt->reestablish_timeout) 1399 xprt->reestablish_timeout = 0; 1400 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1401 queue_work(xprtiod_workqueue, &transport->recv_worker); 1402 } 1403 } 1404 1405 /* 1406 * Helper function to force a TCP close if the server is sending 1407 * junk and/or it has put us in CLOSE_WAIT 1408 */ 1409 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1410 { 1411 xprt_force_disconnect(xprt); 1412 } 1413 1414 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1415 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt) 1416 { 1417 return PAGE_SIZE; 1418 } 1419 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1420 1421 /** 1422 * xs_tcp_state_change - callback to handle TCP socket state changes 1423 * @sk: socket whose state has changed 1424 * 1425 */ 1426 static void xs_tcp_state_change(struct sock *sk) 1427 { 1428 struct rpc_xprt *xprt; 1429 struct sock_xprt *transport; 1430 1431 if (!(xprt = xprt_from_sock(sk))) 1432 return; 1433 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1434 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1435 sk->sk_state, xprt_connected(xprt), 1436 sock_flag(sk, SOCK_DEAD), 1437 sock_flag(sk, SOCK_ZAPPED), 1438 sk->sk_shutdown); 1439 1440 transport = container_of(xprt, struct sock_xprt, xprt); 1441 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1442 switch (sk->sk_state) { 1443 case TCP_ESTABLISHED: 1444 if (!xprt_test_and_set_connected(xprt)) { 1445 xprt->connect_cookie++; 1446 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1447 xprt_clear_connecting(xprt); 1448 1449 xprt->stat.connect_count++; 1450 xprt->stat.connect_time += (long)jiffies - 1451 xprt->stat.connect_start; 1452 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING); 1453 } 1454 break; 1455 case TCP_FIN_WAIT1: 1456 /* The client initiated a shutdown of the socket */ 1457 xprt->connect_cookie++; 1458 xprt->reestablish_timeout = 0; 1459 set_bit(XPRT_CLOSING, &xprt->state); 1460 smp_mb__before_atomic(); 1461 clear_bit(XPRT_CONNECTED, &xprt->state); 1462 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1463 smp_mb__after_atomic(); 1464 break; 1465 case TCP_CLOSE_WAIT: 1466 /* The server initiated a shutdown of the socket */ 1467 xprt->connect_cookie++; 1468 clear_bit(XPRT_CONNECTED, &xprt->state); 1469 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1470 fallthrough; 1471 case TCP_CLOSING: 1472 /* 1473 * If the server closed down the connection, make sure that 1474 * we back off before reconnecting 1475 */ 1476 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1477 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1478 break; 1479 case TCP_LAST_ACK: 1480 set_bit(XPRT_CLOSING, &xprt->state); 1481 smp_mb__before_atomic(); 1482 clear_bit(XPRT_CONNECTED, &xprt->state); 1483 smp_mb__after_atomic(); 1484 break; 1485 case TCP_CLOSE: 1486 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1487 &transport->sock_state)) 1488 xprt_clear_connecting(xprt); 1489 clear_bit(XPRT_CLOSING, &xprt->state); 1490 /* Trigger the socket release */ 1491 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT); 1492 } 1493 } 1494 1495 static void xs_write_space(struct sock *sk) 1496 { 1497 struct sock_xprt *transport; 1498 struct rpc_xprt *xprt; 1499 1500 if (!sk->sk_socket) 1501 return; 1502 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1503 1504 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1505 return; 1506 transport = container_of(xprt, struct sock_xprt, xprt); 1507 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state)) 1508 return; 1509 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE); 1510 sk->sk_write_pending--; 1511 } 1512 1513 /** 1514 * xs_udp_write_space - callback invoked when socket buffer space 1515 * becomes available 1516 * @sk: socket whose state has changed 1517 * 1518 * Called when more output buffer space is available for this socket. 1519 * We try not to wake our writers until they can make "significant" 1520 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1521 * with a bunch of small requests. 1522 */ 1523 static void xs_udp_write_space(struct sock *sk) 1524 { 1525 /* from net/core/sock.c:sock_def_write_space */ 1526 if (sock_writeable(sk)) 1527 xs_write_space(sk); 1528 } 1529 1530 /** 1531 * xs_tcp_write_space - callback invoked when socket buffer space 1532 * becomes available 1533 * @sk: socket whose state has changed 1534 * 1535 * Called when more output buffer space is available for this socket. 1536 * We try not to wake our writers until they can make "significant" 1537 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1538 * with a bunch of small requests. 1539 */ 1540 static void xs_tcp_write_space(struct sock *sk) 1541 { 1542 /* from net/core/stream.c:sk_stream_write_space */ 1543 if (sk_stream_is_writeable(sk)) 1544 xs_write_space(sk); 1545 } 1546 1547 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1548 { 1549 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1550 struct sock *sk = transport->inet; 1551 1552 if (transport->rcvsize) { 1553 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1554 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1555 } 1556 if (transport->sndsize) { 1557 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1558 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1559 sk->sk_write_space(sk); 1560 } 1561 } 1562 1563 /** 1564 * xs_udp_set_buffer_size - set send and receive limits 1565 * @xprt: generic transport 1566 * @sndsize: requested size of send buffer, in bytes 1567 * @rcvsize: requested size of receive buffer, in bytes 1568 * 1569 * Set socket send and receive buffer size limits. 1570 */ 1571 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1572 { 1573 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1574 1575 transport->sndsize = 0; 1576 if (sndsize) 1577 transport->sndsize = sndsize + 1024; 1578 transport->rcvsize = 0; 1579 if (rcvsize) 1580 transport->rcvsize = rcvsize + 1024; 1581 1582 xs_udp_do_set_buffer_size(xprt); 1583 } 1584 1585 /** 1586 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1587 * @xprt: controlling transport 1588 * @task: task that timed out 1589 * 1590 * Adjust the congestion window after a retransmit timeout has occurred. 1591 */ 1592 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1593 { 1594 spin_lock(&xprt->transport_lock); 1595 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1596 spin_unlock(&xprt->transport_lock); 1597 } 1598 1599 static int xs_get_random_port(void) 1600 { 1601 unsigned short min = xprt_min_resvport, max = xprt_max_resvport; 1602 unsigned short range; 1603 unsigned short rand; 1604 1605 if (max < min) 1606 return -EADDRINUSE; 1607 range = max - min + 1; 1608 rand = (unsigned short) prandom_u32() % range; 1609 return rand + min; 1610 } 1611 1612 static unsigned short xs_sock_getport(struct socket *sock) 1613 { 1614 struct sockaddr_storage buf; 1615 unsigned short port = 0; 1616 1617 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0) 1618 goto out; 1619 switch (buf.ss_family) { 1620 case AF_INET6: 1621 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1622 break; 1623 case AF_INET: 1624 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1625 } 1626 out: 1627 return port; 1628 } 1629 1630 /** 1631 * xs_set_port - reset the port number in the remote endpoint address 1632 * @xprt: generic transport 1633 * @port: new port number 1634 * 1635 */ 1636 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1637 { 1638 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1639 1640 rpc_set_port(xs_addr(xprt), port); 1641 xs_update_peer_port(xprt); 1642 } 1643 1644 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1645 { 1646 if (transport->srcport == 0 && transport->xprt.reuseport) 1647 transport->srcport = xs_sock_getport(sock); 1648 } 1649 1650 static int xs_get_srcport(struct sock_xprt *transport) 1651 { 1652 int port = transport->srcport; 1653 1654 if (port == 0 && transport->xprt.resvport) 1655 port = xs_get_random_port(); 1656 return port; 1657 } 1658 1659 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt) 1660 { 1661 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1662 unsigned short ret = 0; 1663 mutex_lock(&sock->recv_mutex); 1664 if (sock->sock) 1665 ret = xs_sock_getport(sock->sock); 1666 mutex_unlock(&sock->recv_mutex); 1667 return ret; 1668 } 1669 1670 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen) 1671 { 1672 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt); 1673 union { 1674 struct sockaddr sa; 1675 struct sockaddr_storage st; 1676 } saddr; 1677 int ret = -ENOTCONN; 1678 1679 mutex_lock(&sock->recv_mutex); 1680 if (sock->sock) { 1681 ret = kernel_getsockname(sock->sock, &saddr.sa); 1682 if (ret >= 0) 1683 ret = snprintf(buf, buflen, "%pISc", &saddr.sa); 1684 } 1685 mutex_unlock(&sock->recv_mutex); 1686 return ret; 1687 } 1688 1689 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1690 { 1691 if (transport->srcport != 0) 1692 transport->srcport = 0; 1693 if (!transport->xprt.resvport) 1694 return 0; 1695 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1696 return xprt_max_resvport; 1697 return --port; 1698 } 1699 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1700 { 1701 struct sockaddr_storage myaddr; 1702 int err, nloop = 0; 1703 int port = xs_get_srcport(transport); 1704 unsigned short last; 1705 1706 /* 1707 * If we are asking for any ephemeral port (i.e. port == 0 && 1708 * transport->xprt.resvport == 0), don't bind. Let the local 1709 * port selection happen implicitly when the socket is used 1710 * (for example at connect time). 1711 * 1712 * This ensures that we can continue to establish TCP 1713 * connections even when all local ephemeral ports are already 1714 * a part of some TCP connection. This makes no difference 1715 * for UDP sockets, but also doesn't harm them. 1716 * 1717 * If we're asking for any reserved port (i.e. port == 0 && 1718 * transport->xprt.resvport == 1) xs_get_srcport above will 1719 * ensure that port is non-zero and we will bind as needed. 1720 */ 1721 if (port <= 0) 1722 return port; 1723 1724 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1725 do { 1726 rpc_set_port((struct sockaddr *)&myaddr, port); 1727 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1728 transport->xprt.addrlen); 1729 if (err == 0) { 1730 if (transport->xprt.reuseport) 1731 transport->srcport = port; 1732 break; 1733 } 1734 last = port; 1735 port = xs_next_srcport(transport, port); 1736 if (port > last) 1737 nloop++; 1738 } while (err == -EADDRINUSE && nloop != 2); 1739 1740 if (myaddr.ss_family == AF_INET) 1741 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1742 &((struct sockaddr_in *)&myaddr)->sin_addr, 1743 port, err ? "failed" : "ok", err); 1744 else 1745 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1746 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1747 port, err ? "failed" : "ok", err); 1748 return err; 1749 } 1750 1751 /* 1752 * We don't support autobind on AF_LOCAL sockets 1753 */ 1754 static void xs_local_rpcbind(struct rpc_task *task) 1755 { 1756 xprt_set_bound(task->tk_xprt); 1757 } 1758 1759 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1760 { 1761 } 1762 1763 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1764 static struct lock_class_key xs_key[3]; 1765 static struct lock_class_key xs_slock_key[3]; 1766 1767 static inline void xs_reclassify_socketu(struct socket *sock) 1768 { 1769 struct sock *sk = sock->sk; 1770 1771 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1772 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]); 1773 } 1774 1775 static inline void xs_reclassify_socket4(struct socket *sock) 1776 { 1777 struct sock *sk = sock->sk; 1778 1779 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1780 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]); 1781 } 1782 1783 static inline void xs_reclassify_socket6(struct socket *sock) 1784 { 1785 struct sock *sk = sock->sk; 1786 1787 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1788 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]); 1789 } 1790 1791 static inline void xs_reclassify_socket(int family, struct socket *sock) 1792 { 1793 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk))) 1794 return; 1795 1796 switch (family) { 1797 case AF_LOCAL: 1798 xs_reclassify_socketu(sock); 1799 break; 1800 case AF_INET: 1801 xs_reclassify_socket4(sock); 1802 break; 1803 case AF_INET6: 1804 xs_reclassify_socket6(sock); 1805 break; 1806 } 1807 } 1808 #else 1809 static inline void xs_reclassify_socket(int family, struct socket *sock) 1810 { 1811 } 1812 #endif 1813 1814 static void xs_dummy_setup_socket(struct work_struct *work) 1815 { 1816 } 1817 1818 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1819 struct sock_xprt *transport, int family, int type, 1820 int protocol, bool reuseport) 1821 { 1822 struct file *filp; 1823 struct socket *sock; 1824 int err; 1825 1826 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1827 if (err < 0) { 1828 dprintk("RPC: can't create %d transport socket (%d).\n", 1829 protocol, -err); 1830 goto out; 1831 } 1832 xs_reclassify_socket(family, sock); 1833 1834 if (reuseport) 1835 sock_set_reuseport(sock->sk); 1836 1837 err = xs_bind(transport, sock); 1838 if (err) { 1839 sock_release(sock); 1840 goto out; 1841 } 1842 1843 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1844 if (IS_ERR(filp)) 1845 return ERR_CAST(filp); 1846 transport->file = filp; 1847 1848 return sock; 1849 out: 1850 return ERR_PTR(err); 1851 } 1852 1853 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1854 struct socket *sock) 1855 { 1856 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1857 xprt); 1858 1859 if (!transport->inet) { 1860 struct sock *sk = sock->sk; 1861 1862 lock_sock(sk); 1863 1864 xs_save_old_callbacks(transport, sk); 1865 1866 sk->sk_user_data = xprt; 1867 sk->sk_data_ready = xs_data_ready; 1868 sk->sk_write_space = xs_udp_write_space; 1869 sk->sk_error_report = xs_error_report; 1870 1871 xprt_clear_connected(xprt); 1872 1873 /* Reset to new socket */ 1874 transport->sock = sock; 1875 transport->inet = sk; 1876 1877 release_sock(sk); 1878 } 1879 1880 xs_stream_start_connect(transport); 1881 1882 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1883 } 1884 1885 /** 1886 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1887 * @transport: socket transport to connect 1888 */ 1889 static int xs_local_setup_socket(struct sock_xprt *transport) 1890 { 1891 struct rpc_xprt *xprt = &transport->xprt; 1892 struct file *filp; 1893 struct socket *sock; 1894 int status; 1895 1896 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1897 SOCK_STREAM, 0, &sock, 1); 1898 if (status < 0) { 1899 dprintk("RPC: can't create AF_LOCAL " 1900 "transport socket (%d).\n", -status); 1901 goto out; 1902 } 1903 xs_reclassify_socket(AF_LOCAL, sock); 1904 1905 filp = sock_alloc_file(sock, O_NONBLOCK, NULL); 1906 if (IS_ERR(filp)) { 1907 status = PTR_ERR(filp); 1908 goto out; 1909 } 1910 transport->file = filp; 1911 1912 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1913 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1914 1915 status = xs_local_finish_connecting(xprt, sock); 1916 trace_rpc_socket_connect(xprt, sock, status); 1917 switch (status) { 1918 case 0: 1919 dprintk("RPC: xprt %p connected to %s\n", 1920 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1921 xprt->stat.connect_count++; 1922 xprt->stat.connect_time += (long)jiffies - 1923 xprt->stat.connect_start; 1924 xprt_set_connected(xprt); 1925 break; 1926 case -ENOBUFS: 1927 break; 1928 case -ENOENT: 1929 dprintk("RPC: xprt %p: socket %s does not exist\n", 1930 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1931 break; 1932 case -ECONNREFUSED: 1933 dprintk("RPC: xprt %p: connection refused for %s\n", 1934 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1935 break; 1936 default: 1937 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 1938 __func__, -status, 1939 xprt->address_strings[RPC_DISPLAY_ADDR]); 1940 } 1941 1942 out: 1943 xprt_clear_connecting(xprt); 1944 xprt_wake_pending_tasks(xprt, status); 1945 return status; 1946 } 1947 1948 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 1949 { 1950 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1951 int ret; 1952 1953 if (RPC_IS_ASYNC(task)) { 1954 /* 1955 * We want the AF_LOCAL connect to be resolved in the 1956 * filesystem namespace of the process making the rpc 1957 * call. Thus we connect synchronously. 1958 * 1959 * If we want to support asynchronous AF_LOCAL calls, 1960 * we'll need to figure out how to pass a namespace to 1961 * connect. 1962 */ 1963 task->tk_rpc_status = -ENOTCONN; 1964 rpc_exit(task, -ENOTCONN); 1965 return; 1966 } 1967 ret = xs_local_setup_socket(transport); 1968 if (ret && !RPC_IS_SOFTCONN(task)) 1969 msleep_interruptible(15000); 1970 } 1971 1972 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 1973 /* 1974 * Note that this should be called with XPRT_LOCKED held, or recv_mutex 1975 * held, or when we otherwise know that we have exclusive access to the 1976 * socket, to guard against races with xs_reset_transport. 1977 */ 1978 static void xs_set_memalloc(struct rpc_xprt *xprt) 1979 { 1980 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1981 xprt); 1982 1983 /* 1984 * If there's no sock, then we have nothing to set. The 1985 * reconnecting process will get it for us. 1986 */ 1987 if (!transport->inet) 1988 return; 1989 if (atomic_read(&xprt->swapper)) 1990 sk_set_memalloc(transport->inet); 1991 } 1992 1993 /** 1994 * xs_enable_swap - Tag this transport as being used for swap. 1995 * @xprt: transport to tag 1996 * 1997 * Take a reference to this transport on behalf of the rpc_clnt, and 1998 * optionally mark it for swapping if it wasn't already. 1999 */ 2000 static int 2001 xs_enable_swap(struct rpc_xprt *xprt) 2002 { 2003 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2004 2005 mutex_lock(&xs->recv_mutex); 2006 if (atomic_inc_return(&xprt->swapper) == 1 && 2007 xs->inet) 2008 sk_set_memalloc(xs->inet); 2009 mutex_unlock(&xs->recv_mutex); 2010 return 0; 2011 } 2012 2013 /** 2014 * xs_disable_swap - Untag this transport as being used for swap. 2015 * @xprt: transport to tag 2016 * 2017 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2018 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2019 */ 2020 static void 2021 xs_disable_swap(struct rpc_xprt *xprt) 2022 { 2023 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2024 2025 mutex_lock(&xs->recv_mutex); 2026 if (atomic_dec_and_test(&xprt->swapper) && 2027 xs->inet) 2028 sk_clear_memalloc(xs->inet); 2029 mutex_unlock(&xs->recv_mutex); 2030 } 2031 #else 2032 static void xs_set_memalloc(struct rpc_xprt *xprt) 2033 { 2034 } 2035 2036 static int 2037 xs_enable_swap(struct rpc_xprt *xprt) 2038 { 2039 return -EINVAL; 2040 } 2041 2042 static void 2043 xs_disable_swap(struct rpc_xprt *xprt) 2044 { 2045 } 2046 #endif 2047 2048 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2049 { 2050 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2051 2052 if (!transport->inet) { 2053 struct sock *sk = sock->sk; 2054 2055 lock_sock(sk); 2056 2057 xs_save_old_callbacks(transport, sk); 2058 2059 sk->sk_user_data = xprt; 2060 sk->sk_data_ready = xs_data_ready; 2061 sk->sk_write_space = xs_udp_write_space; 2062 2063 xprt_set_connected(xprt); 2064 2065 /* Reset to new socket */ 2066 transport->sock = sock; 2067 transport->inet = sk; 2068 2069 xs_set_memalloc(xprt); 2070 2071 release_sock(sk); 2072 } 2073 xs_udp_do_set_buffer_size(xprt); 2074 2075 xprt->stat.connect_start = jiffies; 2076 } 2077 2078 static void xs_udp_setup_socket(struct work_struct *work) 2079 { 2080 struct sock_xprt *transport = 2081 container_of(work, struct sock_xprt, connect_worker.work); 2082 struct rpc_xprt *xprt = &transport->xprt; 2083 struct socket *sock; 2084 int status = -EIO; 2085 unsigned int pflags = current->flags; 2086 2087 if (atomic_read(&xprt->swapper)) 2088 current->flags |= PF_MEMALLOC; 2089 sock = xs_create_sock(xprt, transport, 2090 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2091 IPPROTO_UDP, false); 2092 if (IS_ERR(sock)) 2093 goto out; 2094 2095 dprintk("RPC: worker connecting xprt %p via %s to " 2096 "%s (port %s)\n", xprt, 2097 xprt->address_strings[RPC_DISPLAY_PROTO], 2098 xprt->address_strings[RPC_DISPLAY_ADDR], 2099 xprt->address_strings[RPC_DISPLAY_PORT]); 2100 2101 xs_udp_finish_connecting(xprt, sock); 2102 trace_rpc_socket_connect(xprt, sock, 0); 2103 status = 0; 2104 out: 2105 xprt_clear_connecting(xprt); 2106 xprt_unlock_connect(xprt, transport); 2107 xprt_wake_pending_tasks(xprt, status); 2108 current_restore_flags(pflags, PF_MEMALLOC); 2109 } 2110 2111 /** 2112 * xs_tcp_shutdown - gracefully shut down a TCP socket 2113 * @xprt: transport 2114 * 2115 * Initiates a graceful shutdown of the TCP socket by calling the 2116 * equivalent of shutdown(SHUT_RDWR); 2117 */ 2118 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2119 { 2120 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2121 struct socket *sock = transport->sock; 2122 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE; 2123 2124 if (sock == NULL) 2125 return; 2126 if (!xprt->reuseport) { 2127 xs_close(xprt); 2128 return; 2129 } 2130 switch (skst) { 2131 case TCP_FIN_WAIT1: 2132 case TCP_FIN_WAIT2: 2133 break; 2134 case TCP_ESTABLISHED: 2135 case TCP_CLOSE_WAIT: 2136 kernel_sock_shutdown(sock, SHUT_RDWR); 2137 trace_rpc_socket_shutdown(xprt, sock); 2138 break; 2139 default: 2140 xs_reset_transport(transport); 2141 } 2142 } 2143 2144 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 2145 struct socket *sock) 2146 { 2147 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2148 unsigned int keepidle; 2149 unsigned int keepcnt; 2150 unsigned int timeo; 2151 2152 spin_lock(&xprt->transport_lock); 2153 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ); 2154 keepcnt = xprt->timeout->to_retries + 1; 2155 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2156 (xprt->timeout->to_retries + 1); 2157 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2158 spin_unlock(&xprt->transport_lock); 2159 2160 /* TCP Keepalive options */ 2161 sock_set_keepalive(sock->sk); 2162 tcp_sock_set_keepidle(sock->sk, keepidle); 2163 tcp_sock_set_keepintvl(sock->sk, keepidle); 2164 tcp_sock_set_keepcnt(sock->sk, keepcnt); 2165 2166 /* TCP user timeout (see RFC5482) */ 2167 tcp_sock_set_user_timeout(sock->sk, timeo); 2168 } 2169 2170 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt, 2171 unsigned long connect_timeout, 2172 unsigned long reconnect_timeout) 2173 { 2174 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2175 struct rpc_timeout to; 2176 unsigned long initval; 2177 2178 spin_lock(&xprt->transport_lock); 2179 if (reconnect_timeout < xprt->max_reconnect_timeout) 2180 xprt->max_reconnect_timeout = reconnect_timeout; 2181 if (connect_timeout < xprt->connect_timeout) { 2182 memcpy(&to, xprt->timeout, sizeof(to)); 2183 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1); 2184 /* Arbitrary lower limit */ 2185 if (initval < XS_TCP_INIT_REEST_TO << 1) 2186 initval = XS_TCP_INIT_REEST_TO << 1; 2187 to.to_initval = initval; 2188 to.to_maxval = initval; 2189 memcpy(&transport->tcp_timeout, &to, 2190 sizeof(transport->tcp_timeout)); 2191 xprt->timeout = &transport->tcp_timeout; 2192 xprt->connect_timeout = connect_timeout; 2193 } 2194 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2195 spin_unlock(&xprt->transport_lock); 2196 } 2197 2198 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2199 { 2200 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2201 2202 if (!transport->inet) { 2203 struct sock *sk = sock->sk; 2204 2205 /* Avoid temporary address, they are bad for long-lived 2206 * connections such as NFS mounts. 2207 * RFC4941, section 3.6 suggests that: 2208 * Individual applications, which have specific 2209 * knowledge about the normal duration of connections, 2210 * MAY override this as appropriate. 2211 */ 2212 if (xs_addr(xprt)->sa_family == PF_INET6) { 2213 ip6_sock_set_addr_preferences(sk, 2214 IPV6_PREFER_SRC_PUBLIC); 2215 } 2216 2217 xs_tcp_set_socket_timeouts(xprt, sock); 2218 tcp_sock_set_nodelay(sk); 2219 2220 lock_sock(sk); 2221 2222 xs_save_old_callbacks(transport, sk); 2223 2224 sk->sk_user_data = xprt; 2225 sk->sk_data_ready = xs_data_ready; 2226 sk->sk_state_change = xs_tcp_state_change; 2227 sk->sk_write_space = xs_tcp_write_space; 2228 sk->sk_error_report = xs_error_report; 2229 2230 /* socket options */ 2231 sock_reset_flag(sk, SOCK_LINGER); 2232 2233 xprt_clear_connected(xprt); 2234 2235 /* Reset to new socket */ 2236 transport->sock = sock; 2237 transport->inet = sk; 2238 2239 release_sock(sk); 2240 } 2241 2242 if (!xprt_bound(xprt)) 2243 return -ENOTCONN; 2244 2245 xs_set_memalloc(xprt); 2246 2247 xs_stream_start_connect(transport); 2248 2249 /* Tell the socket layer to start connecting... */ 2250 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2251 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2252 } 2253 2254 /** 2255 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2256 * @work: queued work item 2257 * 2258 * Invoked by a work queue tasklet. 2259 */ 2260 static void xs_tcp_setup_socket(struct work_struct *work) 2261 { 2262 struct sock_xprt *transport = 2263 container_of(work, struct sock_xprt, connect_worker.work); 2264 struct socket *sock = transport->sock; 2265 struct rpc_xprt *xprt = &transport->xprt; 2266 int status; 2267 unsigned int pflags = current->flags; 2268 2269 if (atomic_read(&xprt->swapper)) 2270 current->flags |= PF_MEMALLOC; 2271 2272 if (xprt_connected(xprt)) 2273 goto out; 2274 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT, 2275 &transport->sock_state) || 2276 !sock) { 2277 xs_reset_transport(transport); 2278 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family, 2279 SOCK_STREAM, IPPROTO_TCP, true); 2280 if (IS_ERR(sock)) { 2281 xprt_wake_pending_tasks(xprt, PTR_ERR(sock)); 2282 goto out; 2283 } 2284 } 2285 2286 dprintk("RPC: worker connecting xprt %p via %s to " 2287 "%s (port %s)\n", xprt, 2288 xprt->address_strings[RPC_DISPLAY_PROTO], 2289 xprt->address_strings[RPC_DISPLAY_ADDR], 2290 xprt->address_strings[RPC_DISPLAY_PORT]); 2291 2292 status = xs_tcp_finish_connecting(xprt, sock); 2293 trace_rpc_socket_connect(xprt, sock, status); 2294 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2295 xprt, -status, xprt_connected(xprt), 2296 sock->sk->sk_state); 2297 switch (status) { 2298 case 0: 2299 case -EINPROGRESS: 2300 /* SYN_SENT! */ 2301 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state); 2302 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2303 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2304 fallthrough; 2305 case -EALREADY: 2306 goto out_unlock; 2307 case -EADDRNOTAVAIL: 2308 /* Source port number is unavailable. Try a new one! */ 2309 transport->srcport = 0; 2310 status = -EAGAIN; 2311 break; 2312 case -EINVAL: 2313 /* Happens, for instance, if the user specified a link 2314 * local IPv6 address without a scope-id. 2315 */ 2316 case -ECONNREFUSED: 2317 case -ECONNRESET: 2318 case -ENETDOWN: 2319 case -ENETUNREACH: 2320 case -EHOSTUNREACH: 2321 case -EADDRINUSE: 2322 case -ENOBUFS: 2323 break; 2324 default: 2325 printk("%s: connect returned unhandled error %d\n", 2326 __func__, status); 2327 status = -EAGAIN; 2328 } 2329 2330 /* xs_tcp_force_close() wakes tasks with a fixed error code. 2331 * We need to wake them first to ensure the correct error code. 2332 */ 2333 xprt_wake_pending_tasks(xprt, status); 2334 xs_tcp_force_close(xprt); 2335 out: 2336 xprt_clear_connecting(xprt); 2337 out_unlock: 2338 xprt_unlock_connect(xprt, transport); 2339 current_restore_flags(pflags, PF_MEMALLOC); 2340 } 2341 2342 /** 2343 * xs_connect - connect a socket to a remote endpoint 2344 * @xprt: pointer to transport structure 2345 * @task: address of RPC task that manages state of connect request 2346 * 2347 * TCP: If the remote end dropped the connection, delay reconnecting. 2348 * 2349 * UDP socket connects are synchronous, but we use a work queue anyway 2350 * to guarantee that even unprivileged user processes can set up a 2351 * socket on a privileged port. 2352 * 2353 * If a UDP socket connect fails, the delay behavior here prevents 2354 * retry floods (hard mounts). 2355 */ 2356 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2357 { 2358 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2359 unsigned long delay = 0; 2360 2361 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2362 2363 if (transport->sock != NULL) { 2364 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2365 "seconds\n", xprt, xprt->reestablish_timeout / HZ); 2366 2367 delay = xprt_reconnect_delay(xprt); 2368 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO); 2369 2370 } else 2371 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2372 2373 queue_delayed_work(xprtiod_workqueue, 2374 &transport->connect_worker, 2375 delay); 2376 } 2377 2378 static void xs_wake_disconnect(struct sock_xprt *transport) 2379 { 2380 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state)) 2381 xs_tcp_force_close(&transport->xprt); 2382 } 2383 2384 static void xs_wake_write(struct sock_xprt *transport) 2385 { 2386 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state)) 2387 xprt_write_space(&transport->xprt); 2388 } 2389 2390 static void xs_wake_error(struct sock_xprt *transport) 2391 { 2392 int sockerr; 2393 2394 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2395 return; 2396 mutex_lock(&transport->recv_mutex); 2397 if (transport->sock == NULL) 2398 goto out; 2399 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state)) 2400 goto out; 2401 sockerr = xchg(&transport->xprt_err, 0); 2402 if (sockerr < 0) 2403 xprt_wake_pending_tasks(&transport->xprt, sockerr); 2404 out: 2405 mutex_unlock(&transport->recv_mutex); 2406 } 2407 2408 static void xs_wake_pending(struct sock_xprt *transport) 2409 { 2410 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state)) 2411 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN); 2412 } 2413 2414 static void xs_error_handle(struct work_struct *work) 2415 { 2416 struct sock_xprt *transport = container_of(work, 2417 struct sock_xprt, error_worker); 2418 2419 xs_wake_disconnect(transport); 2420 xs_wake_write(transport); 2421 xs_wake_error(transport); 2422 xs_wake_pending(transport); 2423 } 2424 2425 /** 2426 * xs_local_print_stats - display AF_LOCAL socket-specific stats 2427 * @xprt: rpc_xprt struct containing statistics 2428 * @seq: output file 2429 * 2430 */ 2431 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2432 { 2433 long idle_time = 0; 2434 2435 if (xprt_connected(xprt)) 2436 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2437 2438 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2439 "%llu %llu %lu %llu %llu\n", 2440 xprt->stat.bind_count, 2441 xprt->stat.connect_count, 2442 xprt->stat.connect_time / HZ, 2443 idle_time, 2444 xprt->stat.sends, 2445 xprt->stat.recvs, 2446 xprt->stat.bad_xids, 2447 xprt->stat.req_u, 2448 xprt->stat.bklog_u, 2449 xprt->stat.max_slots, 2450 xprt->stat.sending_u, 2451 xprt->stat.pending_u); 2452 } 2453 2454 /** 2455 * xs_udp_print_stats - display UDP socket-specific stats 2456 * @xprt: rpc_xprt struct containing statistics 2457 * @seq: output file 2458 * 2459 */ 2460 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2461 { 2462 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2463 2464 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2465 "%lu %llu %llu\n", 2466 transport->srcport, 2467 xprt->stat.bind_count, 2468 xprt->stat.sends, 2469 xprt->stat.recvs, 2470 xprt->stat.bad_xids, 2471 xprt->stat.req_u, 2472 xprt->stat.bklog_u, 2473 xprt->stat.max_slots, 2474 xprt->stat.sending_u, 2475 xprt->stat.pending_u); 2476 } 2477 2478 /** 2479 * xs_tcp_print_stats - display TCP socket-specific stats 2480 * @xprt: rpc_xprt struct containing statistics 2481 * @seq: output file 2482 * 2483 */ 2484 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2485 { 2486 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2487 long idle_time = 0; 2488 2489 if (xprt_connected(xprt)) 2490 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2491 2492 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2493 "%llu %llu %lu %llu %llu\n", 2494 transport->srcport, 2495 xprt->stat.bind_count, 2496 xprt->stat.connect_count, 2497 xprt->stat.connect_time / HZ, 2498 idle_time, 2499 xprt->stat.sends, 2500 xprt->stat.recvs, 2501 xprt->stat.bad_xids, 2502 xprt->stat.req_u, 2503 xprt->stat.bklog_u, 2504 xprt->stat.max_slots, 2505 xprt->stat.sending_u, 2506 xprt->stat.pending_u); 2507 } 2508 2509 /* 2510 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2511 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2512 * to use the server side send routines. 2513 */ 2514 static int bc_malloc(struct rpc_task *task) 2515 { 2516 struct rpc_rqst *rqst = task->tk_rqstp; 2517 size_t size = rqst->rq_callsize; 2518 struct page *page; 2519 struct rpc_buffer *buf; 2520 2521 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) { 2522 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n", 2523 size); 2524 return -EINVAL; 2525 } 2526 2527 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN); 2528 if (!page) 2529 return -ENOMEM; 2530 2531 buf = page_address(page); 2532 buf->len = PAGE_SIZE; 2533 2534 rqst->rq_buffer = buf->data; 2535 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; 2536 return 0; 2537 } 2538 2539 /* 2540 * Free the space allocated in the bc_alloc routine 2541 */ 2542 static void bc_free(struct rpc_task *task) 2543 { 2544 void *buffer = task->tk_rqstp->rq_buffer; 2545 struct rpc_buffer *buf; 2546 2547 buf = container_of(buffer, struct rpc_buffer, data); 2548 free_page((unsigned long)buf); 2549 } 2550 2551 static int bc_sendto(struct rpc_rqst *req) 2552 { 2553 struct xdr_buf *xdr = &req->rq_snd_buf; 2554 struct sock_xprt *transport = 2555 container_of(req->rq_xprt, struct sock_xprt, xprt); 2556 struct msghdr msg = { 2557 .msg_flags = 0, 2558 }; 2559 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 2560 (u32)xdr->len); 2561 unsigned int sent = 0; 2562 int err; 2563 2564 req->rq_xtime = ktime_get(); 2565 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask()); 2566 if (err < 0) 2567 return err; 2568 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent); 2569 xdr_free_bvec(xdr); 2570 if (err < 0 || sent != (xdr->len + sizeof(marker))) 2571 return -EAGAIN; 2572 return sent; 2573 } 2574 2575 /** 2576 * bc_send_request - Send a backchannel Call on a TCP socket 2577 * @req: rpc_rqst containing Call message to be sent 2578 * 2579 * xpt_mutex ensures @rqstp's whole message is written to the socket 2580 * without interruption. 2581 * 2582 * Return values: 2583 * %0 if the message was sent successfully 2584 * %ENOTCONN if the message was not sent 2585 */ 2586 static int bc_send_request(struct rpc_rqst *req) 2587 { 2588 struct svc_xprt *xprt; 2589 int len; 2590 2591 /* 2592 * Get the server socket associated with this callback xprt 2593 */ 2594 xprt = req->rq_xprt->bc_xprt; 2595 2596 /* 2597 * Grab the mutex to serialize data as the connection is shared 2598 * with the fore channel 2599 */ 2600 mutex_lock(&xprt->xpt_mutex); 2601 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2602 len = -ENOTCONN; 2603 else 2604 len = bc_sendto(req); 2605 mutex_unlock(&xprt->xpt_mutex); 2606 2607 if (len > 0) 2608 len = 0; 2609 2610 return len; 2611 } 2612 2613 /* 2614 * The close routine. Since this is client initiated, we do nothing 2615 */ 2616 2617 static void bc_close(struct rpc_xprt *xprt) 2618 { 2619 xprt_disconnect_done(xprt); 2620 } 2621 2622 /* 2623 * The xprt destroy routine. Again, because this connection is client 2624 * initiated, we do nothing 2625 */ 2626 2627 static void bc_destroy(struct rpc_xprt *xprt) 2628 { 2629 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2630 2631 xs_xprt_free(xprt); 2632 module_put(THIS_MODULE); 2633 } 2634 2635 static const struct rpc_xprt_ops xs_local_ops = { 2636 .reserve_xprt = xprt_reserve_xprt, 2637 .release_xprt = xprt_release_xprt, 2638 .alloc_slot = xprt_alloc_slot, 2639 .free_slot = xprt_free_slot, 2640 .rpcbind = xs_local_rpcbind, 2641 .set_port = xs_local_set_port, 2642 .connect = xs_local_connect, 2643 .buf_alloc = rpc_malloc, 2644 .buf_free = rpc_free, 2645 .prepare_request = xs_stream_prepare_request, 2646 .send_request = xs_local_send_request, 2647 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2648 .close = xs_close, 2649 .destroy = xs_destroy, 2650 .print_stats = xs_local_print_stats, 2651 .enable_swap = xs_enable_swap, 2652 .disable_swap = xs_disable_swap, 2653 }; 2654 2655 static const struct rpc_xprt_ops xs_udp_ops = { 2656 .set_buffer_size = xs_udp_set_buffer_size, 2657 .reserve_xprt = xprt_reserve_xprt_cong, 2658 .release_xprt = xprt_release_xprt_cong, 2659 .alloc_slot = xprt_alloc_slot, 2660 .free_slot = xprt_free_slot, 2661 .rpcbind = rpcb_getport_async, 2662 .set_port = xs_set_port, 2663 .connect = xs_connect, 2664 .get_srcaddr = xs_sock_srcaddr, 2665 .get_srcport = xs_sock_srcport, 2666 .buf_alloc = rpc_malloc, 2667 .buf_free = rpc_free, 2668 .send_request = xs_udp_send_request, 2669 .wait_for_reply_request = xprt_wait_for_reply_request_rtt, 2670 .timer = xs_udp_timer, 2671 .release_request = xprt_release_rqst_cong, 2672 .close = xs_close, 2673 .destroy = xs_destroy, 2674 .print_stats = xs_udp_print_stats, 2675 .enable_swap = xs_enable_swap, 2676 .disable_swap = xs_disable_swap, 2677 .inject_disconnect = xs_inject_disconnect, 2678 }; 2679 2680 static const struct rpc_xprt_ops xs_tcp_ops = { 2681 .reserve_xprt = xprt_reserve_xprt, 2682 .release_xprt = xprt_release_xprt, 2683 .alloc_slot = xprt_alloc_slot, 2684 .free_slot = xprt_free_slot, 2685 .rpcbind = rpcb_getport_async, 2686 .set_port = xs_set_port, 2687 .connect = xs_connect, 2688 .get_srcaddr = xs_sock_srcaddr, 2689 .get_srcport = xs_sock_srcport, 2690 .buf_alloc = rpc_malloc, 2691 .buf_free = rpc_free, 2692 .prepare_request = xs_stream_prepare_request, 2693 .send_request = xs_tcp_send_request, 2694 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2695 .close = xs_tcp_shutdown, 2696 .destroy = xs_destroy, 2697 .set_connect_timeout = xs_tcp_set_connect_timeout, 2698 .print_stats = xs_tcp_print_stats, 2699 .enable_swap = xs_enable_swap, 2700 .disable_swap = xs_disable_swap, 2701 .inject_disconnect = xs_inject_disconnect, 2702 #ifdef CONFIG_SUNRPC_BACKCHANNEL 2703 .bc_setup = xprt_setup_bc, 2704 .bc_maxpayload = xs_tcp_bc_maxpayload, 2705 .bc_num_slots = xprt_bc_max_slots, 2706 .bc_free_rqst = xprt_free_bc_rqst, 2707 .bc_destroy = xprt_destroy_bc, 2708 #endif 2709 }; 2710 2711 /* 2712 * The rpc_xprt_ops for the server backchannel 2713 */ 2714 2715 static const struct rpc_xprt_ops bc_tcp_ops = { 2716 .reserve_xprt = xprt_reserve_xprt, 2717 .release_xprt = xprt_release_xprt, 2718 .alloc_slot = xprt_alloc_slot, 2719 .free_slot = xprt_free_slot, 2720 .buf_alloc = bc_malloc, 2721 .buf_free = bc_free, 2722 .send_request = bc_send_request, 2723 .wait_for_reply_request = xprt_wait_for_reply_request_def, 2724 .close = bc_close, 2725 .destroy = bc_destroy, 2726 .print_stats = xs_tcp_print_stats, 2727 .enable_swap = xs_enable_swap, 2728 .disable_swap = xs_disable_swap, 2729 .inject_disconnect = xs_inject_disconnect, 2730 }; 2731 2732 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2733 { 2734 static const struct sockaddr_in sin = { 2735 .sin_family = AF_INET, 2736 .sin_addr.s_addr = htonl(INADDR_ANY), 2737 }; 2738 static const struct sockaddr_in6 sin6 = { 2739 .sin6_family = AF_INET6, 2740 .sin6_addr = IN6ADDR_ANY_INIT, 2741 }; 2742 2743 switch (family) { 2744 case AF_LOCAL: 2745 break; 2746 case AF_INET: 2747 memcpy(sap, &sin, sizeof(sin)); 2748 break; 2749 case AF_INET6: 2750 memcpy(sap, &sin6, sizeof(sin6)); 2751 break; 2752 default: 2753 dprintk("RPC: %s: Bad address family\n", __func__); 2754 return -EAFNOSUPPORT; 2755 } 2756 return 0; 2757 } 2758 2759 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2760 unsigned int slot_table_size, 2761 unsigned int max_slot_table_size) 2762 { 2763 struct rpc_xprt *xprt; 2764 struct sock_xprt *new; 2765 2766 if (args->addrlen > sizeof(xprt->addr)) { 2767 dprintk("RPC: xs_setup_xprt: address too large\n"); 2768 return ERR_PTR(-EBADF); 2769 } 2770 2771 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2772 max_slot_table_size); 2773 if (xprt == NULL) { 2774 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2775 "rpc_xprt\n"); 2776 return ERR_PTR(-ENOMEM); 2777 } 2778 2779 new = container_of(xprt, struct sock_xprt, xprt); 2780 mutex_init(&new->recv_mutex); 2781 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2782 xprt->addrlen = args->addrlen; 2783 if (args->srcaddr) 2784 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2785 else { 2786 int err; 2787 err = xs_init_anyaddr(args->dstaddr->sa_family, 2788 (struct sockaddr *)&new->srcaddr); 2789 if (err != 0) { 2790 xprt_free(xprt); 2791 return ERR_PTR(err); 2792 } 2793 } 2794 2795 return xprt; 2796 } 2797 2798 static const struct rpc_timeout xs_local_default_timeout = { 2799 .to_initval = 10 * HZ, 2800 .to_maxval = 10 * HZ, 2801 .to_retries = 2, 2802 }; 2803 2804 /** 2805 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2806 * @args: rpc transport creation arguments 2807 * 2808 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2809 */ 2810 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2811 { 2812 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2813 struct sock_xprt *transport; 2814 struct rpc_xprt *xprt; 2815 struct rpc_xprt *ret; 2816 2817 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2818 xprt_max_tcp_slot_table_entries); 2819 if (IS_ERR(xprt)) 2820 return xprt; 2821 transport = container_of(xprt, struct sock_xprt, xprt); 2822 2823 xprt->prot = 0; 2824 xprt->xprt_class = &xs_local_transport; 2825 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2826 2827 xprt->bind_timeout = XS_BIND_TO; 2828 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2829 xprt->idle_timeout = XS_IDLE_DISC_TO; 2830 2831 xprt->ops = &xs_local_ops; 2832 xprt->timeout = &xs_local_default_timeout; 2833 2834 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 2835 INIT_WORK(&transport->error_worker, xs_error_handle); 2836 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket); 2837 2838 switch (sun->sun_family) { 2839 case AF_LOCAL: 2840 if (sun->sun_path[0] != '/') { 2841 dprintk("RPC: bad AF_LOCAL address: %s\n", 2842 sun->sun_path); 2843 ret = ERR_PTR(-EINVAL); 2844 goto out_err; 2845 } 2846 xprt_set_bound(xprt); 2847 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2848 ret = ERR_PTR(xs_local_setup_socket(transport)); 2849 if (ret) 2850 goto out_err; 2851 break; 2852 default: 2853 ret = ERR_PTR(-EAFNOSUPPORT); 2854 goto out_err; 2855 } 2856 2857 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2858 xprt->address_strings[RPC_DISPLAY_ADDR]); 2859 2860 if (try_module_get(THIS_MODULE)) 2861 return xprt; 2862 ret = ERR_PTR(-EINVAL); 2863 out_err: 2864 xs_xprt_free(xprt); 2865 return ret; 2866 } 2867 2868 static const struct rpc_timeout xs_udp_default_timeout = { 2869 .to_initval = 5 * HZ, 2870 .to_maxval = 30 * HZ, 2871 .to_increment = 5 * HZ, 2872 .to_retries = 5, 2873 }; 2874 2875 /** 2876 * xs_setup_udp - Set up transport to use a UDP socket 2877 * @args: rpc transport creation arguments 2878 * 2879 */ 2880 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2881 { 2882 struct sockaddr *addr = args->dstaddr; 2883 struct rpc_xprt *xprt; 2884 struct sock_xprt *transport; 2885 struct rpc_xprt *ret; 2886 2887 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2888 xprt_udp_slot_table_entries); 2889 if (IS_ERR(xprt)) 2890 return xprt; 2891 transport = container_of(xprt, struct sock_xprt, xprt); 2892 2893 xprt->prot = IPPROTO_UDP; 2894 xprt->xprt_class = &xs_udp_transport; 2895 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2896 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2897 2898 xprt->bind_timeout = XS_BIND_TO; 2899 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2900 xprt->idle_timeout = XS_IDLE_DISC_TO; 2901 2902 xprt->ops = &xs_udp_ops; 2903 2904 xprt->timeout = &xs_udp_default_timeout; 2905 2906 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 2907 INIT_WORK(&transport->error_worker, xs_error_handle); 2908 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 2909 2910 switch (addr->sa_family) { 2911 case AF_INET: 2912 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2913 xprt_set_bound(xprt); 2914 2915 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2916 break; 2917 case AF_INET6: 2918 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2919 xprt_set_bound(xprt); 2920 2921 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2922 break; 2923 default: 2924 ret = ERR_PTR(-EAFNOSUPPORT); 2925 goto out_err; 2926 } 2927 2928 if (xprt_bound(xprt)) 2929 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2930 xprt->address_strings[RPC_DISPLAY_ADDR], 2931 xprt->address_strings[RPC_DISPLAY_PORT], 2932 xprt->address_strings[RPC_DISPLAY_PROTO]); 2933 else 2934 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2935 xprt->address_strings[RPC_DISPLAY_ADDR], 2936 xprt->address_strings[RPC_DISPLAY_PROTO]); 2937 2938 if (try_module_get(THIS_MODULE)) 2939 return xprt; 2940 ret = ERR_PTR(-EINVAL); 2941 out_err: 2942 xs_xprt_free(xprt); 2943 return ret; 2944 } 2945 2946 static const struct rpc_timeout xs_tcp_default_timeout = { 2947 .to_initval = 60 * HZ, 2948 .to_maxval = 60 * HZ, 2949 .to_retries = 2, 2950 }; 2951 2952 /** 2953 * xs_setup_tcp - Set up transport to use a TCP socket 2954 * @args: rpc transport creation arguments 2955 * 2956 */ 2957 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2958 { 2959 struct sockaddr *addr = args->dstaddr; 2960 struct rpc_xprt *xprt; 2961 struct sock_xprt *transport; 2962 struct rpc_xprt *ret; 2963 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2964 2965 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2966 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2967 2968 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2969 max_slot_table_size); 2970 if (IS_ERR(xprt)) 2971 return xprt; 2972 transport = container_of(xprt, struct sock_xprt, xprt); 2973 2974 xprt->prot = IPPROTO_TCP; 2975 xprt->xprt_class = &xs_tcp_transport; 2976 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2977 2978 xprt->bind_timeout = XS_BIND_TO; 2979 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2980 xprt->idle_timeout = XS_IDLE_DISC_TO; 2981 2982 xprt->ops = &xs_tcp_ops; 2983 xprt->timeout = &xs_tcp_default_timeout; 2984 2985 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 2986 xprt->connect_timeout = xprt->timeout->to_initval * 2987 (xprt->timeout->to_retries + 1); 2988 2989 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn); 2990 INIT_WORK(&transport->error_worker, xs_error_handle); 2991 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 2992 2993 switch (addr->sa_family) { 2994 case AF_INET: 2995 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2996 xprt_set_bound(xprt); 2997 2998 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2999 break; 3000 case AF_INET6: 3001 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3002 xprt_set_bound(xprt); 3003 3004 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 3005 break; 3006 default: 3007 ret = ERR_PTR(-EAFNOSUPPORT); 3008 goto out_err; 3009 } 3010 3011 if (xprt_bound(xprt)) 3012 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3013 xprt->address_strings[RPC_DISPLAY_ADDR], 3014 xprt->address_strings[RPC_DISPLAY_PORT], 3015 xprt->address_strings[RPC_DISPLAY_PROTO]); 3016 else 3017 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3018 xprt->address_strings[RPC_DISPLAY_ADDR], 3019 xprt->address_strings[RPC_DISPLAY_PROTO]); 3020 3021 if (try_module_get(THIS_MODULE)) 3022 return xprt; 3023 ret = ERR_PTR(-EINVAL); 3024 out_err: 3025 xs_xprt_free(xprt); 3026 return ret; 3027 } 3028 3029 /** 3030 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 3031 * @args: rpc transport creation arguments 3032 * 3033 */ 3034 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 3035 { 3036 struct sockaddr *addr = args->dstaddr; 3037 struct rpc_xprt *xprt; 3038 struct sock_xprt *transport; 3039 struct svc_sock *bc_sock; 3040 struct rpc_xprt *ret; 3041 3042 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3043 xprt_tcp_slot_table_entries); 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_bc_tcp_transport; 3050 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3051 xprt->timeout = &xs_tcp_default_timeout; 3052 3053 /* backchannel */ 3054 xprt_set_bound(xprt); 3055 xprt->bind_timeout = 0; 3056 xprt->reestablish_timeout = 0; 3057 xprt->idle_timeout = 0; 3058 3059 xprt->ops = &bc_tcp_ops; 3060 3061 switch (addr->sa_family) { 3062 case AF_INET: 3063 xs_format_peer_addresses(xprt, "tcp", 3064 RPCBIND_NETID_TCP); 3065 break; 3066 case AF_INET6: 3067 xs_format_peer_addresses(xprt, "tcp", 3068 RPCBIND_NETID_TCP6); 3069 break; 3070 default: 3071 ret = ERR_PTR(-EAFNOSUPPORT); 3072 goto out_err; 3073 } 3074 3075 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3076 xprt->address_strings[RPC_DISPLAY_ADDR], 3077 xprt->address_strings[RPC_DISPLAY_PORT], 3078 xprt->address_strings[RPC_DISPLAY_PROTO]); 3079 3080 /* 3081 * Once we've associated a backchannel xprt with a connection, 3082 * we want to keep it around as long as the connection lasts, 3083 * in case we need to start using it for a backchannel again; 3084 * this reference won't be dropped until bc_xprt is destroyed. 3085 */ 3086 xprt_get(xprt); 3087 args->bc_xprt->xpt_bc_xprt = xprt; 3088 xprt->bc_xprt = args->bc_xprt; 3089 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3090 transport->sock = bc_sock->sk_sock; 3091 transport->inet = bc_sock->sk_sk; 3092 3093 /* 3094 * Since we don't want connections for the backchannel, we set 3095 * the xprt status to connected 3096 */ 3097 xprt_set_connected(xprt); 3098 3099 if (try_module_get(THIS_MODULE)) 3100 return xprt; 3101 3102 args->bc_xprt->xpt_bc_xprt = NULL; 3103 args->bc_xprt->xpt_bc_xps = NULL; 3104 xprt_put(xprt); 3105 ret = ERR_PTR(-EINVAL); 3106 out_err: 3107 xs_xprt_free(xprt); 3108 return ret; 3109 } 3110 3111 static struct xprt_class xs_local_transport = { 3112 .list = LIST_HEAD_INIT(xs_local_transport.list), 3113 .name = "named UNIX socket", 3114 .owner = THIS_MODULE, 3115 .ident = XPRT_TRANSPORT_LOCAL, 3116 .setup = xs_setup_local, 3117 .netid = { "" }, 3118 }; 3119 3120 static struct xprt_class xs_udp_transport = { 3121 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3122 .name = "udp", 3123 .owner = THIS_MODULE, 3124 .ident = XPRT_TRANSPORT_UDP, 3125 .setup = xs_setup_udp, 3126 .netid = { "udp", "udp6", "" }, 3127 }; 3128 3129 static struct xprt_class xs_tcp_transport = { 3130 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3131 .name = "tcp", 3132 .owner = THIS_MODULE, 3133 .ident = XPRT_TRANSPORT_TCP, 3134 .setup = xs_setup_tcp, 3135 .netid = { "tcp", "tcp6", "" }, 3136 }; 3137 3138 static struct xprt_class xs_bc_tcp_transport = { 3139 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3140 .name = "tcp NFSv4.1 backchannel", 3141 .owner = THIS_MODULE, 3142 .ident = XPRT_TRANSPORT_BC_TCP, 3143 .setup = xs_setup_bc_tcp, 3144 .netid = { "" }, 3145 }; 3146 3147 /** 3148 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3149 * 3150 */ 3151 int init_socket_xprt(void) 3152 { 3153 if (!sunrpc_table_header) 3154 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3155 3156 xprt_register_transport(&xs_local_transport); 3157 xprt_register_transport(&xs_udp_transport); 3158 xprt_register_transport(&xs_tcp_transport); 3159 xprt_register_transport(&xs_bc_tcp_transport); 3160 3161 return 0; 3162 } 3163 3164 /** 3165 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3166 * 3167 */ 3168 void cleanup_socket_xprt(void) 3169 { 3170 if (sunrpc_table_header) { 3171 unregister_sysctl_table(sunrpc_table_header); 3172 sunrpc_table_header = NULL; 3173 } 3174 3175 xprt_unregister_transport(&xs_local_transport); 3176 xprt_unregister_transport(&xs_udp_transport); 3177 xprt_unregister_transport(&xs_tcp_transport); 3178 xprt_unregister_transport(&xs_bc_tcp_transport); 3179 } 3180 3181 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3182 { 3183 return param_set_uint_minmax(val, kp, 3184 RPC_MIN_RESVPORT, 3185 RPC_MAX_RESVPORT); 3186 } 3187 3188 static const struct kernel_param_ops param_ops_portnr = { 3189 .set = param_set_portnr, 3190 .get = param_get_uint, 3191 }; 3192 3193 #define param_check_portnr(name, p) \ 3194 __param_check(name, p, unsigned int); 3195 3196 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3197 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3198 3199 static int param_set_slot_table_size(const char *val, 3200 const struct kernel_param *kp) 3201 { 3202 return param_set_uint_minmax(val, kp, 3203 RPC_MIN_SLOT_TABLE, 3204 RPC_MAX_SLOT_TABLE); 3205 } 3206 3207 static const struct kernel_param_ops param_ops_slot_table_size = { 3208 .set = param_set_slot_table_size, 3209 .get = param_get_uint, 3210 }; 3211 3212 #define param_check_slot_table_size(name, p) \ 3213 __param_check(name, p, unsigned int); 3214 3215 static int param_set_max_slot_table_size(const char *val, 3216 const struct kernel_param *kp) 3217 { 3218 return param_set_uint_minmax(val, kp, 3219 RPC_MIN_SLOT_TABLE, 3220 RPC_MAX_SLOT_TABLE_LIMIT); 3221 } 3222 3223 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3224 .set = param_set_max_slot_table_size, 3225 .get = param_get_uint, 3226 }; 3227 3228 #define param_check_max_slot_table_size(name, p) \ 3229 __param_check(name, p, unsigned int); 3230 3231 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3232 slot_table_size, 0644); 3233 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3234 max_slot_table_size, 0644); 3235 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3236 slot_table_size, 0644); 3237