1 /* 2 * linux/net/sunrpc/xprtsock.c 3 * 4 * Client-side transport implementation for sockets. 5 * 6 * TCP callback races fixes (C) 1998 Red Hat 7 * TCP send fixes (C) 1998 Red Hat 8 * TCP NFS related read + write fixes 9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 10 * 11 * Rewrite of larges part of the code in order to stabilize TCP stuff. 12 * Fix behaviour when socket buffer is full. 13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> 14 * 15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> 16 * 17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005. 18 * <gilles.quillard@bull.net> 19 */ 20 21 #include <linux/types.h> 22 #include <linux/string.h> 23 #include <linux/slab.h> 24 #include <linux/module.h> 25 #include <linux/capability.h> 26 #include <linux/pagemap.h> 27 #include <linux/errno.h> 28 #include <linux/socket.h> 29 #include <linux/in.h> 30 #include <linux/net.h> 31 #include <linux/mm.h> 32 #include <linux/un.h> 33 #include <linux/udp.h> 34 #include <linux/tcp.h> 35 #include <linux/sunrpc/clnt.h> 36 #include <linux/sunrpc/addr.h> 37 #include <linux/sunrpc/sched.h> 38 #include <linux/sunrpc/svcsock.h> 39 #include <linux/sunrpc/xprtsock.h> 40 #include <linux/file.h> 41 #ifdef CONFIG_SUNRPC_BACKCHANNEL 42 #include <linux/sunrpc/bc_xprt.h> 43 #endif 44 45 #include <net/sock.h> 46 #include <net/checksum.h> 47 #include <net/udp.h> 48 #include <net/tcp.h> 49 50 #include <trace/events/sunrpc.h> 51 52 #include "sunrpc.h" 53 54 static void xs_close(struct rpc_xprt *xprt); 55 56 /* 57 * xprtsock tunables 58 */ 59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 62 63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 65 66 #define XS_TCP_LINGER_TO (15U * HZ) 67 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 68 69 /* 70 * We can register our own files under /proc/sys/sunrpc by 71 * calling register_sysctl_table() again. The files in that 72 * directory become the union of all files registered there. 73 * 74 * We simply need to make sure that we don't collide with 75 * someone else's file names! 76 */ 77 78 #ifdef RPC_DEBUG 79 80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 85 86 static struct ctl_table_header *sunrpc_table_header; 87 88 /* 89 * FIXME: changing the UDP slot table size should also resize the UDP 90 * socket buffers for existing UDP transports 91 */ 92 static struct ctl_table xs_tunables_table[] = { 93 { 94 .procname = "udp_slot_table_entries", 95 .data = &xprt_udp_slot_table_entries, 96 .maxlen = sizeof(unsigned int), 97 .mode = 0644, 98 .proc_handler = proc_dointvec_minmax, 99 .extra1 = &min_slot_table_size, 100 .extra2 = &max_slot_table_size 101 }, 102 { 103 .procname = "tcp_slot_table_entries", 104 .data = &xprt_tcp_slot_table_entries, 105 .maxlen = sizeof(unsigned int), 106 .mode = 0644, 107 .proc_handler = proc_dointvec_minmax, 108 .extra1 = &min_slot_table_size, 109 .extra2 = &max_slot_table_size 110 }, 111 { 112 .procname = "tcp_max_slot_table_entries", 113 .data = &xprt_max_tcp_slot_table_entries, 114 .maxlen = sizeof(unsigned int), 115 .mode = 0644, 116 .proc_handler = proc_dointvec_minmax, 117 .extra1 = &min_slot_table_size, 118 .extra2 = &max_tcp_slot_table_limit 119 }, 120 { 121 .procname = "min_resvport", 122 .data = &xprt_min_resvport, 123 .maxlen = sizeof(unsigned int), 124 .mode = 0644, 125 .proc_handler = proc_dointvec_minmax, 126 .extra1 = &xprt_min_resvport_limit, 127 .extra2 = &xprt_max_resvport_limit 128 }, 129 { 130 .procname = "max_resvport", 131 .data = &xprt_max_resvport, 132 .maxlen = sizeof(unsigned int), 133 .mode = 0644, 134 .proc_handler = proc_dointvec_minmax, 135 .extra1 = &xprt_min_resvport_limit, 136 .extra2 = &xprt_max_resvport_limit 137 }, 138 { 139 .procname = "tcp_fin_timeout", 140 .data = &xs_tcp_fin_timeout, 141 .maxlen = sizeof(xs_tcp_fin_timeout), 142 .mode = 0644, 143 .proc_handler = proc_dointvec_jiffies, 144 }, 145 { }, 146 }; 147 148 static struct ctl_table sunrpc_table[] = { 149 { 150 .procname = "sunrpc", 151 .mode = 0555, 152 .child = xs_tunables_table 153 }, 154 { }, 155 }; 156 157 #endif 158 159 /* 160 * Wait duration for a reply from the RPC portmapper. 161 */ 162 #define XS_BIND_TO (60U * HZ) 163 164 /* 165 * Delay if a UDP socket connect error occurs. This is most likely some 166 * kind of resource problem on the local host. 167 */ 168 #define XS_UDP_REEST_TO (2U * HZ) 169 170 /* 171 * The reestablish timeout allows clients to delay for a bit before attempting 172 * to reconnect to a server that just dropped our connection. 173 * 174 * We implement an exponential backoff when trying to reestablish a TCP 175 * transport connection with the server. Some servers like to drop a TCP 176 * connection when they are overworked, so we start with a short timeout and 177 * increase over time if the server is down or not responding. 178 */ 179 #define XS_TCP_INIT_REEST_TO (3U * HZ) 180 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ) 181 182 /* 183 * TCP idle timeout; client drops the transport socket if it is idle 184 * for this long. Note that we also timeout UDP sockets to prevent 185 * holding port numbers when there is no RPC traffic. 186 */ 187 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 188 189 #ifdef RPC_DEBUG 190 # undef RPC_DEBUG_DATA 191 # define RPCDBG_FACILITY RPCDBG_TRANS 192 #endif 193 194 #ifdef RPC_DEBUG_DATA 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 196 { 197 u8 *buf = (u8 *) packet; 198 int j; 199 200 dprintk("RPC: %s\n", msg); 201 for (j = 0; j < count && j < 128; j += 4) { 202 if (!(j & 31)) { 203 if (j) 204 dprintk("\n"); 205 dprintk("0x%04x ", j); 206 } 207 dprintk("%02x%02x%02x%02x ", 208 buf[j], buf[j+1], buf[j+2], buf[j+3]); 209 } 210 dprintk("\n"); 211 } 212 #else 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 214 { 215 /* NOP */ 216 } 217 #endif 218 219 struct sock_xprt { 220 struct rpc_xprt xprt; 221 222 /* 223 * Network layer 224 */ 225 struct socket * sock; 226 struct sock * inet; 227 228 /* 229 * State of TCP reply receive 230 */ 231 __be32 tcp_fraghdr, 232 tcp_xid, 233 tcp_calldir; 234 235 u32 tcp_offset, 236 tcp_reclen; 237 238 unsigned long tcp_copied, 239 tcp_flags; 240 241 /* 242 * Connection of transports 243 */ 244 struct delayed_work connect_worker; 245 struct sockaddr_storage srcaddr; 246 unsigned short srcport; 247 248 /* 249 * UDP socket buffer size parameters 250 */ 251 size_t rcvsize, 252 sndsize; 253 254 /* 255 * Saved socket callback addresses 256 */ 257 void (*old_data_ready)(struct sock *); 258 void (*old_state_change)(struct sock *); 259 void (*old_write_space)(struct sock *); 260 void (*old_error_report)(struct sock *); 261 }; 262 263 /* 264 * TCP receive state flags 265 */ 266 #define TCP_RCV_LAST_FRAG (1UL << 0) 267 #define TCP_RCV_COPY_FRAGHDR (1UL << 1) 268 #define TCP_RCV_COPY_XID (1UL << 2) 269 #define TCP_RCV_COPY_DATA (1UL << 3) 270 #define TCP_RCV_READ_CALLDIR (1UL << 4) 271 #define TCP_RCV_COPY_CALLDIR (1UL << 5) 272 273 /* 274 * TCP RPC flags 275 */ 276 #define TCP_RPC_REPLY (1UL << 6) 277 278 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 279 { 280 return (struct rpc_xprt *) sk->sk_user_data; 281 } 282 283 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 284 { 285 return (struct sockaddr *) &xprt->addr; 286 } 287 288 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 289 { 290 return (struct sockaddr_un *) &xprt->addr; 291 } 292 293 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 294 { 295 return (struct sockaddr_in *) &xprt->addr; 296 } 297 298 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 299 { 300 return (struct sockaddr_in6 *) &xprt->addr; 301 } 302 303 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 304 { 305 struct sockaddr *sap = xs_addr(xprt); 306 struct sockaddr_in6 *sin6; 307 struct sockaddr_in *sin; 308 struct sockaddr_un *sun; 309 char buf[128]; 310 311 switch (sap->sa_family) { 312 case AF_LOCAL: 313 sun = xs_addr_un(xprt); 314 strlcpy(buf, sun->sun_path, sizeof(buf)); 315 xprt->address_strings[RPC_DISPLAY_ADDR] = 316 kstrdup(buf, GFP_KERNEL); 317 break; 318 case AF_INET: 319 (void)rpc_ntop(sap, buf, sizeof(buf)); 320 xprt->address_strings[RPC_DISPLAY_ADDR] = 321 kstrdup(buf, GFP_KERNEL); 322 sin = xs_addr_in(xprt); 323 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 324 break; 325 case AF_INET6: 326 (void)rpc_ntop(sap, buf, sizeof(buf)); 327 xprt->address_strings[RPC_DISPLAY_ADDR] = 328 kstrdup(buf, GFP_KERNEL); 329 sin6 = xs_addr_in6(xprt); 330 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 331 break; 332 default: 333 BUG(); 334 } 335 336 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 337 } 338 339 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 340 { 341 struct sockaddr *sap = xs_addr(xprt); 342 char buf[128]; 343 344 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 345 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 346 347 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 348 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 349 } 350 351 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 352 const char *protocol, 353 const char *netid) 354 { 355 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 356 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 357 xs_format_common_peer_addresses(xprt); 358 xs_format_common_peer_ports(xprt); 359 } 360 361 static void xs_update_peer_port(struct rpc_xprt *xprt) 362 { 363 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 364 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 365 366 xs_format_common_peer_ports(xprt); 367 } 368 369 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 370 { 371 unsigned int i; 372 373 for (i = 0; i < RPC_DISPLAY_MAX; i++) 374 switch (i) { 375 case RPC_DISPLAY_PROTO: 376 case RPC_DISPLAY_NETID: 377 continue; 378 default: 379 kfree(xprt->address_strings[i]); 380 } 381 } 382 383 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 384 385 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more) 386 { 387 struct msghdr msg = { 388 .msg_name = addr, 389 .msg_namelen = addrlen, 390 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0), 391 }; 392 struct kvec iov = { 393 .iov_base = vec->iov_base + base, 394 .iov_len = vec->iov_len - base, 395 }; 396 397 if (iov.iov_len != 0) 398 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); 399 return kernel_sendmsg(sock, &msg, NULL, 0, 0); 400 } 401 402 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy) 403 { 404 ssize_t (*do_sendpage)(struct socket *sock, struct page *page, 405 int offset, size_t size, int flags); 406 struct page **ppage; 407 unsigned int remainder; 408 int err, sent = 0; 409 410 remainder = xdr->page_len - base; 411 base += xdr->page_base; 412 ppage = xdr->pages + (base >> PAGE_SHIFT); 413 base &= ~PAGE_MASK; 414 do_sendpage = sock->ops->sendpage; 415 if (!zerocopy) 416 do_sendpage = sock_no_sendpage; 417 for(;;) { 418 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); 419 int flags = XS_SENDMSG_FLAGS; 420 421 remainder -= len; 422 if (remainder != 0 || more) 423 flags |= MSG_MORE; 424 err = do_sendpage(sock, *ppage, base, len, flags); 425 if (remainder == 0 || err != len) 426 break; 427 sent += err; 428 ppage++; 429 base = 0; 430 } 431 if (sent == 0) 432 return err; 433 if (err > 0) 434 sent += err; 435 return sent; 436 } 437 438 /** 439 * xs_sendpages - write pages directly to a socket 440 * @sock: socket to send on 441 * @addr: UDP only -- address of destination 442 * @addrlen: UDP only -- length of destination address 443 * @xdr: buffer containing this request 444 * @base: starting position in the buffer 445 * @zerocopy: true if it is safe to use sendpage() 446 * 447 */ 448 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy) 449 { 450 unsigned int remainder = xdr->len - base; 451 int err, sent = 0; 452 453 if (unlikely(!sock)) 454 return -ENOTSOCK; 455 456 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); 457 if (base != 0) { 458 addr = NULL; 459 addrlen = 0; 460 } 461 462 if (base < xdr->head[0].iov_len || addr != NULL) { 463 unsigned int len = xdr->head[0].iov_len - base; 464 remainder -= len; 465 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0); 466 if (remainder == 0 || err != len) 467 goto out; 468 sent += err; 469 base = 0; 470 } else 471 base -= xdr->head[0].iov_len; 472 473 if (base < xdr->page_len) { 474 unsigned int len = xdr->page_len - base; 475 remainder -= len; 476 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy); 477 if (remainder == 0 || err != len) 478 goto out; 479 sent += err; 480 base = 0; 481 } else 482 base -= xdr->page_len; 483 484 if (base >= xdr->tail[0].iov_len) 485 return sent; 486 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0); 487 out: 488 if (sent == 0) 489 return err; 490 if (err > 0) 491 sent += err; 492 return sent; 493 } 494 495 static void xs_nospace_callback(struct rpc_task *task) 496 { 497 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt); 498 499 transport->inet->sk_write_pending--; 500 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 501 } 502 503 /** 504 * xs_nospace - place task on wait queue if transmit was incomplete 505 * @task: task to put to sleep 506 * 507 */ 508 static int xs_nospace(struct rpc_task *task) 509 { 510 struct rpc_rqst *req = task->tk_rqstp; 511 struct rpc_xprt *xprt = req->rq_xprt; 512 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 513 struct sock *sk = transport->inet; 514 int ret = -EAGAIN; 515 516 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", 517 task->tk_pid, req->rq_slen - req->rq_bytes_sent, 518 req->rq_slen); 519 520 /* Protect against races with write_space */ 521 spin_lock_bh(&xprt->transport_lock); 522 523 /* Don't race with disconnect */ 524 if (xprt_connected(xprt)) { 525 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) { 526 /* 527 * Notify TCP that we're limited by the application 528 * window size 529 */ 530 set_bit(SOCK_NOSPACE, &transport->sock->flags); 531 sk->sk_write_pending++; 532 /* ...and wait for more buffer space */ 533 xprt_wait_for_buffer_space(task, xs_nospace_callback); 534 } 535 } else { 536 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 537 ret = -ENOTCONN; 538 } 539 540 spin_unlock_bh(&xprt->transport_lock); 541 542 /* Race breaker in case memory is freed before above code is called */ 543 sk->sk_write_space(sk); 544 return ret; 545 } 546 547 /* 548 * Construct a stream transport record marker in @buf. 549 */ 550 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf) 551 { 552 u32 reclen = buf->len - sizeof(rpc_fraghdr); 553 rpc_fraghdr *base = buf->head[0].iov_base; 554 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen); 555 } 556 557 /** 558 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 559 * @task: RPC task that manages the state of an RPC request 560 * 561 * Return values: 562 * 0: The request has been sent 563 * EAGAIN: The socket was blocked, please call again later to 564 * complete the request 565 * ENOTCONN: Caller needs to invoke connect logic then call again 566 * other: Some other error occured, the request was not sent 567 */ 568 static int xs_local_send_request(struct rpc_task *task) 569 { 570 struct rpc_rqst *req = task->tk_rqstp; 571 struct rpc_xprt *xprt = req->rq_xprt; 572 struct sock_xprt *transport = 573 container_of(xprt, struct sock_xprt, xprt); 574 struct xdr_buf *xdr = &req->rq_snd_buf; 575 int status; 576 577 xs_encode_stream_record_marker(&req->rq_snd_buf); 578 579 xs_pktdump("packet data:", 580 req->rq_svec->iov_base, req->rq_svec->iov_len); 581 582 status = xs_sendpages(transport->sock, NULL, 0, 583 xdr, req->rq_bytes_sent, true); 584 dprintk("RPC: %s(%u) = %d\n", 585 __func__, xdr->len - req->rq_bytes_sent, status); 586 if (likely(status >= 0)) { 587 req->rq_bytes_sent += status; 588 req->rq_xmit_bytes_sent += status; 589 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 590 req->rq_bytes_sent = 0; 591 return 0; 592 } 593 status = -EAGAIN; 594 } 595 596 switch (status) { 597 case -EAGAIN: 598 status = xs_nospace(task); 599 break; 600 default: 601 dprintk("RPC: sendmsg returned unrecognized error %d\n", 602 -status); 603 case -EPIPE: 604 xs_close(xprt); 605 status = -ENOTCONN; 606 } 607 608 return status; 609 } 610 611 /** 612 * xs_udp_send_request - write an RPC request to a UDP socket 613 * @task: address of RPC task that manages the state of an RPC request 614 * 615 * Return values: 616 * 0: The request has been sent 617 * EAGAIN: The socket was blocked, please call again later to 618 * complete the request 619 * ENOTCONN: Caller needs to invoke connect logic then call again 620 * other: Some other error occurred, the request was not sent 621 */ 622 static int xs_udp_send_request(struct rpc_task *task) 623 { 624 struct rpc_rqst *req = task->tk_rqstp; 625 struct rpc_xprt *xprt = req->rq_xprt; 626 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 627 struct xdr_buf *xdr = &req->rq_snd_buf; 628 int status; 629 630 xs_pktdump("packet data:", 631 req->rq_svec->iov_base, 632 req->rq_svec->iov_len); 633 634 if (!xprt_bound(xprt)) 635 return -ENOTCONN; 636 status = xs_sendpages(transport->sock, 637 xs_addr(xprt), 638 xprt->addrlen, xdr, 639 req->rq_bytes_sent, true); 640 641 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 642 xdr->len - req->rq_bytes_sent, status); 643 644 if (status >= 0) { 645 req->rq_xmit_bytes_sent += status; 646 if (status >= req->rq_slen) 647 return 0; 648 /* Still some bytes left; set up for a retry later. */ 649 status = -EAGAIN; 650 } 651 652 switch (status) { 653 case -ENOTSOCK: 654 status = -ENOTCONN; 655 /* Should we call xs_close() here? */ 656 break; 657 case -EAGAIN: 658 status = xs_nospace(task); 659 break; 660 default: 661 dprintk("RPC: sendmsg returned unrecognized error %d\n", 662 -status); 663 case -ENETUNREACH: 664 case -EPIPE: 665 case -ECONNREFUSED: 666 /* When the server has died, an ICMP port unreachable message 667 * prompts ECONNREFUSED. */ 668 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 669 } 670 671 return status; 672 } 673 674 /** 675 * xs_tcp_shutdown - gracefully shut down a TCP socket 676 * @xprt: transport 677 * 678 * Initiates a graceful shutdown of the TCP socket by calling the 679 * equivalent of shutdown(SHUT_WR); 680 */ 681 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 682 { 683 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 684 struct socket *sock = transport->sock; 685 686 if (sock != NULL) { 687 kernel_sock_shutdown(sock, SHUT_WR); 688 trace_rpc_socket_shutdown(xprt, sock); 689 } 690 } 691 692 /** 693 * xs_tcp_send_request - write an RPC request to a TCP socket 694 * @task: address of RPC task that manages the state of an RPC request 695 * 696 * Return values: 697 * 0: The request has been sent 698 * EAGAIN: The socket was blocked, please call again later to 699 * complete the request 700 * ENOTCONN: Caller needs to invoke connect logic then call again 701 * other: Some other error occurred, the request was not sent 702 * 703 * XXX: In the case of soft timeouts, should we eventually give up 704 * if sendmsg is not able to make progress? 705 */ 706 static int xs_tcp_send_request(struct rpc_task *task) 707 { 708 struct rpc_rqst *req = task->tk_rqstp; 709 struct rpc_xprt *xprt = req->rq_xprt; 710 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 711 struct xdr_buf *xdr = &req->rq_snd_buf; 712 bool zerocopy = true; 713 int status; 714 715 xs_encode_stream_record_marker(&req->rq_snd_buf); 716 717 xs_pktdump("packet data:", 718 req->rq_svec->iov_base, 719 req->rq_svec->iov_len); 720 /* Don't use zero copy if this is a resend. If the RPC call 721 * completes while the socket holds a reference to the pages, 722 * then we may end up resending corrupted data. 723 */ 724 if (task->tk_flags & RPC_TASK_SENT) 725 zerocopy = false; 726 727 /* Continue transmitting the packet/record. We must be careful 728 * to cope with writespace callbacks arriving _after_ we have 729 * called sendmsg(). */ 730 while (1) { 731 status = xs_sendpages(transport->sock, 732 NULL, 0, xdr, req->rq_bytes_sent, 733 zerocopy); 734 735 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 736 xdr->len - req->rq_bytes_sent, status); 737 738 if (unlikely(status < 0)) 739 break; 740 741 /* If we've sent the entire packet, immediately 742 * reset the count of bytes sent. */ 743 req->rq_bytes_sent += status; 744 req->rq_xmit_bytes_sent += status; 745 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 746 req->rq_bytes_sent = 0; 747 return 0; 748 } 749 750 if (status != 0) 751 continue; 752 status = -EAGAIN; 753 break; 754 } 755 756 switch (status) { 757 case -ENOTSOCK: 758 status = -ENOTCONN; 759 /* Should we call xs_close() here? */ 760 break; 761 case -EAGAIN: 762 status = xs_nospace(task); 763 break; 764 default: 765 dprintk("RPC: sendmsg returned unrecognized error %d\n", 766 -status); 767 case -ECONNRESET: 768 xs_tcp_shutdown(xprt); 769 case -ECONNREFUSED: 770 case -ENOTCONN: 771 case -EPIPE: 772 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 773 } 774 775 return status; 776 } 777 778 /** 779 * xs_tcp_release_xprt - clean up after a tcp transmission 780 * @xprt: transport 781 * @task: rpc task 782 * 783 * This cleans up if an error causes us to abort the transmission of a request. 784 * In this case, the socket may need to be reset in order to avoid confusing 785 * the server. 786 */ 787 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 788 { 789 struct rpc_rqst *req; 790 791 if (task != xprt->snd_task) 792 return; 793 if (task == NULL) 794 goto out_release; 795 req = task->tk_rqstp; 796 if (req == NULL) 797 goto out_release; 798 if (req->rq_bytes_sent == 0) 799 goto out_release; 800 if (req->rq_bytes_sent == req->rq_snd_buf.len) 801 goto out_release; 802 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 803 out_release: 804 xprt_release_xprt(xprt, task); 805 } 806 807 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 808 { 809 transport->old_data_ready = sk->sk_data_ready; 810 transport->old_state_change = sk->sk_state_change; 811 transport->old_write_space = sk->sk_write_space; 812 transport->old_error_report = sk->sk_error_report; 813 } 814 815 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 816 { 817 sk->sk_data_ready = transport->old_data_ready; 818 sk->sk_state_change = transport->old_state_change; 819 sk->sk_write_space = transport->old_write_space; 820 sk->sk_error_report = transport->old_error_report; 821 } 822 823 /** 824 * xs_error_report - callback to handle TCP socket state errors 825 * @sk: socket 826 * 827 * Note: we don't call sock_error() since there may be a rpc_task 828 * using the socket, and so we don't want to clear sk->sk_err. 829 */ 830 static void xs_error_report(struct sock *sk) 831 { 832 struct rpc_xprt *xprt; 833 int err; 834 835 read_lock_bh(&sk->sk_callback_lock); 836 if (!(xprt = xprt_from_sock(sk))) 837 goto out; 838 839 err = -sk->sk_err; 840 if (err == 0) 841 goto out; 842 dprintk("RPC: xs_error_report client %p, error=%d...\n", 843 xprt, -err); 844 trace_rpc_socket_error(xprt, sk->sk_socket, err); 845 xprt_wake_pending_tasks(xprt, err); 846 out: 847 read_unlock_bh(&sk->sk_callback_lock); 848 } 849 850 static void xs_reset_transport(struct sock_xprt *transport) 851 { 852 struct socket *sock = transport->sock; 853 struct sock *sk = transport->inet; 854 855 if (sk == NULL) 856 return; 857 858 transport->srcport = 0; 859 860 write_lock_bh(&sk->sk_callback_lock); 861 transport->inet = NULL; 862 transport->sock = NULL; 863 864 sk->sk_user_data = NULL; 865 866 xs_restore_old_callbacks(transport, sk); 867 write_unlock_bh(&sk->sk_callback_lock); 868 869 trace_rpc_socket_close(&transport->xprt, sock); 870 sock_release(sock); 871 } 872 873 /** 874 * xs_close - close a socket 875 * @xprt: transport 876 * 877 * This is used when all requests are complete; ie, no DRC state remains 878 * on the server we want to save. 879 * 880 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 881 * xs_reset_transport() zeroing the socket from underneath a writer. 882 */ 883 static void xs_close(struct rpc_xprt *xprt) 884 { 885 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 886 887 dprintk("RPC: xs_close xprt %p\n", xprt); 888 889 cancel_delayed_work_sync(&transport->connect_worker); 890 891 xs_reset_transport(transport); 892 xprt->reestablish_timeout = 0; 893 894 smp_mb__before_atomic(); 895 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 896 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 897 clear_bit(XPRT_CLOSING, &xprt->state); 898 smp_mb__after_atomic(); 899 xprt_disconnect_done(xprt); 900 } 901 902 static void xs_tcp_close(struct rpc_xprt *xprt) 903 { 904 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state)) 905 xs_close(xprt); 906 else 907 xs_tcp_shutdown(xprt); 908 } 909 910 static void xs_xprt_free(struct rpc_xprt *xprt) 911 { 912 xs_free_peer_addresses(xprt); 913 xprt_free(xprt); 914 } 915 916 /** 917 * xs_destroy - prepare to shutdown a transport 918 * @xprt: doomed transport 919 * 920 */ 921 static void xs_destroy(struct rpc_xprt *xprt) 922 { 923 dprintk("RPC: xs_destroy xprt %p\n", xprt); 924 925 xs_close(xprt); 926 xs_xprt_free(xprt); 927 module_put(THIS_MODULE); 928 } 929 930 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb) 931 { 932 struct xdr_skb_reader desc = { 933 .skb = skb, 934 .offset = sizeof(rpc_fraghdr), 935 .count = skb->len - sizeof(rpc_fraghdr), 936 }; 937 938 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0) 939 return -1; 940 if (desc.count) 941 return -1; 942 return 0; 943 } 944 945 /** 946 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets 947 * @sk: socket with data to read 948 * @len: how much data to read 949 * 950 * Currently this assumes we can read the whole reply in a single gulp. 951 */ 952 static void xs_local_data_ready(struct sock *sk) 953 { 954 struct rpc_task *task; 955 struct rpc_xprt *xprt; 956 struct rpc_rqst *rovr; 957 struct sk_buff *skb; 958 int err, repsize, copied; 959 u32 _xid; 960 __be32 *xp; 961 962 read_lock_bh(&sk->sk_callback_lock); 963 dprintk("RPC: %s...\n", __func__); 964 xprt = xprt_from_sock(sk); 965 if (xprt == NULL) 966 goto out; 967 968 skb = skb_recv_datagram(sk, 0, 1, &err); 969 if (skb == NULL) 970 goto out; 971 972 repsize = skb->len - sizeof(rpc_fraghdr); 973 if (repsize < 4) { 974 dprintk("RPC: impossible RPC reply size %d\n", repsize); 975 goto dropit; 976 } 977 978 /* Copy the XID from the skb... */ 979 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid); 980 if (xp == NULL) 981 goto dropit; 982 983 /* Look up and lock the request corresponding to the given XID */ 984 spin_lock(&xprt->transport_lock); 985 rovr = xprt_lookup_rqst(xprt, *xp); 986 if (!rovr) 987 goto out_unlock; 988 task = rovr->rq_task; 989 990 copied = rovr->rq_private_buf.buflen; 991 if (copied > repsize) 992 copied = repsize; 993 994 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) { 995 dprintk("RPC: sk_buff copy failed\n"); 996 goto out_unlock; 997 } 998 999 xprt_complete_rqst(task, copied); 1000 1001 out_unlock: 1002 spin_unlock(&xprt->transport_lock); 1003 dropit: 1004 skb_free_datagram(sk, skb); 1005 out: 1006 read_unlock_bh(&sk->sk_callback_lock); 1007 } 1008 1009 /** 1010 * xs_udp_data_ready - "data ready" callback for UDP sockets 1011 * @sk: socket with data to read 1012 * @len: how much data to read 1013 * 1014 */ 1015 static void xs_udp_data_ready(struct sock *sk) 1016 { 1017 struct rpc_task *task; 1018 struct rpc_xprt *xprt; 1019 struct rpc_rqst *rovr; 1020 struct sk_buff *skb; 1021 int err, repsize, copied; 1022 u32 _xid; 1023 __be32 *xp; 1024 1025 read_lock_bh(&sk->sk_callback_lock); 1026 dprintk("RPC: xs_udp_data_ready...\n"); 1027 if (!(xprt = xprt_from_sock(sk))) 1028 goto out; 1029 1030 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) 1031 goto out; 1032 1033 repsize = skb->len - sizeof(struct udphdr); 1034 if (repsize < 4) { 1035 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1036 goto dropit; 1037 } 1038 1039 /* Copy the XID from the skb... */ 1040 xp = skb_header_pointer(skb, sizeof(struct udphdr), 1041 sizeof(_xid), &_xid); 1042 if (xp == NULL) 1043 goto dropit; 1044 1045 /* Look up and lock the request corresponding to the given XID */ 1046 spin_lock(&xprt->transport_lock); 1047 rovr = xprt_lookup_rqst(xprt, *xp); 1048 if (!rovr) 1049 goto out_unlock; 1050 task = rovr->rq_task; 1051 1052 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1053 copied = repsize; 1054 1055 /* Suck it into the iovec, verify checksum if not done by hw. */ 1056 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1057 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS); 1058 goto out_unlock; 1059 } 1060 1061 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS); 1062 1063 xprt_adjust_cwnd(xprt, task, copied); 1064 xprt_complete_rqst(task, copied); 1065 1066 out_unlock: 1067 spin_unlock(&xprt->transport_lock); 1068 dropit: 1069 skb_free_datagram(sk, skb); 1070 out: 1071 read_unlock_bh(&sk->sk_callback_lock); 1072 } 1073 1074 /* 1075 * Helper function to force a TCP close if the server is sending 1076 * junk and/or it has put us in CLOSE_WAIT 1077 */ 1078 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1079 { 1080 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); 1081 xprt_force_disconnect(xprt); 1082 } 1083 1084 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 1085 { 1086 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1087 size_t len, used; 1088 char *p; 1089 1090 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; 1091 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; 1092 used = xdr_skb_read_bits(desc, p, len); 1093 transport->tcp_offset += used; 1094 if (used != len) 1095 return; 1096 1097 transport->tcp_reclen = ntohl(transport->tcp_fraghdr); 1098 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) 1099 transport->tcp_flags |= TCP_RCV_LAST_FRAG; 1100 else 1101 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; 1102 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; 1103 1104 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; 1105 transport->tcp_offset = 0; 1106 1107 /* Sanity check of the record length */ 1108 if (unlikely(transport->tcp_reclen < 8)) { 1109 dprintk("RPC: invalid TCP record fragment length\n"); 1110 xs_tcp_force_close(xprt); 1111 return; 1112 } 1113 dprintk("RPC: reading TCP record fragment of length %d\n", 1114 transport->tcp_reclen); 1115 } 1116 1117 static void xs_tcp_check_fraghdr(struct sock_xprt *transport) 1118 { 1119 if (transport->tcp_offset == transport->tcp_reclen) { 1120 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; 1121 transport->tcp_offset = 0; 1122 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { 1123 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1124 transport->tcp_flags |= TCP_RCV_COPY_XID; 1125 transport->tcp_copied = 0; 1126 } 1127 } 1128 } 1129 1130 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1131 { 1132 size_t len, used; 1133 char *p; 1134 1135 len = sizeof(transport->tcp_xid) - transport->tcp_offset; 1136 dprintk("RPC: reading XID (%Zu bytes)\n", len); 1137 p = ((char *) &transport->tcp_xid) + transport->tcp_offset; 1138 used = xdr_skb_read_bits(desc, p, len); 1139 transport->tcp_offset += used; 1140 if (used != len) 1141 return; 1142 transport->tcp_flags &= ~TCP_RCV_COPY_XID; 1143 transport->tcp_flags |= TCP_RCV_READ_CALLDIR; 1144 transport->tcp_copied = 4; 1145 dprintk("RPC: reading %s XID %08x\n", 1146 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for" 1147 : "request with", 1148 ntohl(transport->tcp_xid)); 1149 xs_tcp_check_fraghdr(transport); 1150 } 1151 1152 static inline void xs_tcp_read_calldir(struct sock_xprt *transport, 1153 struct xdr_skb_reader *desc) 1154 { 1155 size_t len, used; 1156 u32 offset; 1157 char *p; 1158 1159 /* 1160 * We want transport->tcp_offset to be 8 at the end of this routine 1161 * (4 bytes for the xid and 4 bytes for the call/reply flag). 1162 * When this function is called for the first time, 1163 * transport->tcp_offset is 4 (after having already read the xid). 1164 */ 1165 offset = transport->tcp_offset - sizeof(transport->tcp_xid); 1166 len = sizeof(transport->tcp_calldir) - offset; 1167 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len); 1168 p = ((char *) &transport->tcp_calldir) + offset; 1169 used = xdr_skb_read_bits(desc, p, len); 1170 transport->tcp_offset += used; 1171 if (used != len) 1172 return; 1173 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR; 1174 /* 1175 * We don't yet have the XDR buffer, so we will write the calldir 1176 * out after we get the buffer from the 'struct rpc_rqst' 1177 */ 1178 switch (ntohl(transport->tcp_calldir)) { 1179 case RPC_REPLY: 1180 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1181 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1182 transport->tcp_flags |= TCP_RPC_REPLY; 1183 break; 1184 case RPC_CALL: 1185 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1186 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1187 transport->tcp_flags &= ~TCP_RPC_REPLY; 1188 break; 1189 default: 1190 dprintk("RPC: invalid request message type\n"); 1191 xs_tcp_force_close(&transport->xprt); 1192 } 1193 xs_tcp_check_fraghdr(transport); 1194 } 1195 1196 static inline void xs_tcp_read_common(struct rpc_xprt *xprt, 1197 struct xdr_skb_reader *desc, 1198 struct rpc_rqst *req) 1199 { 1200 struct sock_xprt *transport = 1201 container_of(xprt, struct sock_xprt, xprt); 1202 struct xdr_buf *rcvbuf; 1203 size_t len; 1204 ssize_t r; 1205 1206 rcvbuf = &req->rq_private_buf; 1207 1208 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) { 1209 /* 1210 * Save the RPC direction in the XDR buffer 1211 */ 1212 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied, 1213 &transport->tcp_calldir, 1214 sizeof(transport->tcp_calldir)); 1215 transport->tcp_copied += sizeof(transport->tcp_calldir); 1216 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR; 1217 } 1218 1219 len = desc->count; 1220 if (len > transport->tcp_reclen - transport->tcp_offset) { 1221 struct xdr_skb_reader my_desc; 1222 1223 len = transport->tcp_reclen - transport->tcp_offset; 1224 memcpy(&my_desc, desc, sizeof(my_desc)); 1225 my_desc.count = len; 1226 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1227 &my_desc, xdr_skb_read_bits); 1228 desc->count -= r; 1229 desc->offset += r; 1230 } else 1231 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1232 desc, xdr_skb_read_bits); 1233 1234 if (r > 0) { 1235 transport->tcp_copied += r; 1236 transport->tcp_offset += r; 1237 } 1238 if (r != len) { 1239 /* Error when copying to the receive buffer, 1240 * usually because we weren't able to allocate 1241 * additional buffer pages. All we can do now 1242 * is turn off TCP_RCV_COPY_DATA, so the request 1243 * will not receive any additional updates, 1244 * and time out. 1245 * Any remaining data from this record will 1246 * be discarded. 1247 */ 1248 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1249 dprintk("RPC: XID %08x truncated request\n", 1250 ntohl(transport->tcp_xid)); 1251 dprintk("RPC: xprt = %p, tcp_copied = %lu, " 1252 "tcp_offset = %u, tcp_reclen = %u\n", 1253 xprt, transport->tcp_copied, 1254 transport->tcp_offset, transport->tcp_reclen); 1255 return; 1256 } 1257 1258 dprintk("RPC: XID %08x read %Zd bytes\n", 1259 ntohl(transport->tcp_xid), r); 1260 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " 1261 "tcp_reclen = %u\n", xprt, transport->tcp_copied, 1262 transport->tcp_offset, transport->tcp_reclen); 1263 1264 if (transport->tcp_copied == req->rq_private_buf.buflen) 1265 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1266 else if (transport->tcp_offset == transport->tcp_reclen) { 1267 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) 1268 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1269 } 1270 } 1271 1272 /* 1273 * Finds the request corresponding to the RPC xid and invokes the common 1274 * tcp read code to read the data. 1275 */ 1276 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt, 1277 struct xdr_skb_reader *desc) 1278 { 1279 struct sock_xprt *transport = 1280 container_of(xprt, struct sock_xprt, xprt); 1281 struct rpc_rqst *req; 1282 1283 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid)); 1284 1285 /* Find and lock the request corresponding to this xid */ 1286 spin_lock(&xprt->transport_lock); 1287 req = xprt_lookup_rqst(xprt, transport->tcp_xid); 1288 if (!req) { 1289 dprintk("RPC: XID %08x request not found!\n", 1290 ntohl(transport->tcp_xid)); 1291 spin_unlock(&xprt->transport_lock); 1292 return -1; 1293 } 1294 1295 xs_tcp_read_common(xprt, desc, req); 1296 1297 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1298 xprt_complete_rqst(req->rq_task, transport->tcp_copied); 1299 1300 spin_unlock(&xprt->transport_lock); 1301 return 0; 1302 } 1303 1304 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1305 /* 1306 * Obtains an rpc_rqst previously allocated and invokes the common 1307 * tcp read code to read the data. The result is placed in the callback 1308 * queue. 1309 * If we're unable to obtain the rpc_rqst we schedule the closing of the 1310 * connection and return -1. 1311 */ 1312 static int xs_tcp_read_callback(struct rpc_xprt *xprt, 1313 struct xdr_skb_reader *desc) 1314 { 1315 struct sock_xprt *transport = 1316 container_of(xprt, struct sock_xprt, xprt); 1317 struct rpc_rqst *req; 1318 1319 /* Look up and lock the request corresponding to the given XID */ 1320 spin_lock(&xprt->transport_lock); 1321 req = xprt_lookup_bc_request(xprt, transport->tcp_xid); 1322 if (req == NULL) { 1323 spin_unlock(&xprt->transport_lock); 1324 printk(KERN_WARNING "Callback slot table overflowed\n"); 1325 xprt_force_disconnect(xprt); 1326 return -1; 1327 } 1328 1329 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid)); 1330 xs_tcp_read_common(xprt, desc, req); 1331 1332 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1333 xprt_complete_bc_request(req, transport->tcp_copied); 1334 spin_unlock(&xprt->transport_lock); 1335 1336 return 0; 1337 } 1338 1339 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1340 struct xdr_skb_reader *desc) 1341 { 1342 struct sock_xprt *transport = 1343 container_of(xprt, struct sock_xprt, xprt); 1344 1345 return (transport->tcp_flags & TCP_RPC_REPLY) ? 1346 xs_tcp_read_reply(xprt, desc) : 1347 xs_tcp_read_callback(xprt, desc); 1348 } 1349 #else 1350 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1351 struct xdr_skb_reader *desc) 1352 { 1353 return xs_tcp_read_reply(xprt, desc); 1354 } 1355 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1356 1357 /* 1358 * Read data off the transport. This can be either an RPC_CALL or an 1359 * RPC_REPLY. Relay the processing to helper functions. 1360 */ 1361 static void xs_tcp_read_data(struct rpc_xprt *xprt, 1362 struct xdr_skb_reader *desc) 1363 { 1364 struct sock_xprt *transport = 1365 container_of(xprt, struct sock_xprt, xprt); 1366 1367 if (_xs_tcp_read_data(xprt, desc) == 0) 1368 xs_tcp_check_fraghdr(transport); 1369 else { 1370 /* 1371 * The transport_lock protects the request handling. 1372 * There's no need to hold it to update the tcp_flags. 1373 */ 1374 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1375 } 1376 } 1377 1378 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1379 { 1380 size_t len; 1381 1382 len = transport->tcp_reclen - transport->tcp_offset; 1383 if (len > desc->count) 1384 len = desc->count; 1385 desc->count -= len; 1386 desc->offset += len; 1387 transport->tcp_offset += len; 1388 dprintk("RPC: discarded %Zu bytes\n", len); 1389 xs_tcp_check_fraghdr(transport); 1390 } 1391 1392 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) 1393 { 1394 struct rpc_xprt *xprt = rd_desc->arg.data; 1395 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1396 struct xdr_skb_reader desc = { 1397 .skb = skb, 1398 .offset = offset, 1399 .count = len, 1400 }; 1401 1402 dprintk("RPC: xs_tcp_data_recv started\n"); 1403 do { 1404 /* Read in a new fragment marker if necessary */ 1405 /* Can we ever really expect to get completely empty fragments? */ 1406 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { 1407 xs_tcp_read_fraghdr(xprt, &desc); 1408 continue; 1409 } 1410 /* Read in the xid if necessary */ 1411 if (transport->tcp_flags & TCP_RCV_COPY_XID) { 1412 xs_tcp_read_xid(transport, &desc); 1413 continue; 1414 } 1415 /* Read in the call/reply flag */ 1416 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) { 1417 xs_tcp_read_calldir(transport, &desc); 1418 continue; 1419 } 1420 /* Read in the request data */ 1421 if (transport->tcp_flags & TCP_RCV_COPY_DATA) { 1422 xs_tcp_read_data(xprt, &desc); 1423 continue; 1424 } 1425 /* Skip over any trailing bytes on short reads */ 1426 xs_tcp_read_discard(transport, &desc); 1427 } while (desc.count); 1428 dprintk("RPC: xs_tcp_data_recv done\n"); 1429 return len - desc.count; 1430 } 1431 1432 /** 1433 * xs_tcp_data_ready - "data ready" callback for TCP sockets 1434 * @sk: socket with data to read 1435 * @bytes: how much data to read 1436 * 1437 */ 1438 static void xs_tcp_data_ready(struct sock *sk) 1439 { 1440 struct rpc_xprt *xprt; 1441 read_descriptor_t rd_desc; 1442 int read; 1443 1444 dprintk("RPC: xs_tcp_data_ready...\n"); 1445 1446 read_lock_bh(&sk->sk_callback_lock); 1447 if (!(xprt = xprt_from_sock(sk))) 1448 goto out; 1449 /* Any data means we had a useful conversation, so 1450 * the we don't need to delay the next reconnect 1451 */ 1452 if (xprt->reestablish_timeout) 1453 xprt->reestablish_timeout = 0; 1454 1455 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ 1456 rd_desc.arg.data = xprt; 1457 do { 1458 rd_desc.count = 65536; 1459 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); 1460 } while (read > 0); 1461 out: 1462 read_unlock_bh(&sk->sk_callback_lock); 1463 } 1464 1465 /* 1466 * Do the equivalent of linger/linger2 handling for dealing with 1467 * broken servers that don't close the socket in a timely 1468 * fashion 1469 */ 1470 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt, 1471 unsigned long timeout) 1472 { 1473 struct sock_xprt *transport; 1474 1475 if (xprt_test_and_set_connecting(xprt)) 1476 return; 1477 set_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1478 transport = container_of(xprt, struct sock_xprt, xprt); 1479 queue_delayed_work(rpciod_workqueue, &transport->connect_worker, 1480 timeout); 1481 } 1482 1483 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt) 1484 { 1485 struct sock_xprt *transport; 1486 1487 transport = container_of(xprt, struct sock_xprt, xprt); 1488 1489 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) || 1490 !cancel_delayed_work(&transport->connect_worker)) 1491 return; 1492 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1493 xprt_clear_connecting(xprt); 1494 } 1495 1496 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1497 { 1498 smp_mb__before_atomic(); 1499 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1500 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state); 1501 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1502 clear_bit(XPRT_CLOSING, &xprt->state); 1503 smp_mb__after_atomic(); 1504 } 1505 1506 static void xs_sock_mark_closed(struct rpc_xprt *xprt) 1507 { 1508 xs_sock_reset_connection_flags(xprt); 1509 /* Mark transport as closed and wake up all pending tasks */ 1510 xprt_disconnect_done(xprt); 1511 } 1512 1513 /** 1514 * xs_tcp_state_change - callback to handle TCP socket state changes 1515 * @sk: socket whose state has changed 1516 * 1517 */ 1518 static void xs_tcp_state_change(struct sock *sk) 1519 { 1520 struct rpc_xprt *xprt; 1521 1522 read_lock_bh(&sk->sk_callback_lock); 1523 if (!(xprt = xprt_from_sock(sk))) 1524 goto out; 1525 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1526 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1527 sk->sk_state, xprt_connected(xprt), 1528 sock_flag(sk, SOCK_DEAD), 1529 sock_flag(sk, SOCK_ZAPPED), 1530 sk->sk_shutdown); 1531 1532 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1533 switch (sk->sk_state) { 1534 case TCP_ESTABLISHED: 1535 spin_lock(&xprt->transport_lock); 1536 if (!xprt_test_and_set_connected(xprt)) { 1537 struct sock_xprt *transport = container_of(xprt, 1538 struct sock_xprt, xprt); 1539 1540 /* Reset TCP record info */ 1541 transport->tcp_offset = 0; 1542 transport->tcp_reclen = 0; 1543 transport->tcp_copied = 0; 1544 transport->tcp_flags = 1545 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; 1546 xprt->connect_cookie++; 1547 1548 xprt_wake_pending_tasks(xprt, -EAGAIN); 1549 } 1550 spin_unlock(&xprt->transport_lock); 1551 break; 1552 case TCP_FIN_WAIT1: 1553 /* The client initiated a shutdown of the socket */ 1554 xprt->connect_cookie++; 1555 xprt->reestablish_timeout = 0; 1556 set_bit(XPRT_CLOSING, &xprt->state); 1557 smp_mb__before_atomic(); 1558 clear_bit(XPRT_CONNECTED, &xprt->state); 1559 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1560 smp_mb__after_atomic(); 1561 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); 1562 break; 1563 case TCP_CLOSE_WAIT: 1564 /* The server initiated a shutdown of the socket */ 1565 xprt->connect_cookie++; 1566 clear_bit(XPRT_CONNECTED, &xprt->state); 1567 xs_tcp_force_close(xprt); 1568 case TCP_CLOSING: 1569 /* 1570 * If the server closed down the connection, make sure that 1571 * we back off before reconnecting 1572 */ 1573 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1574 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1575 break; 1576 case TCP_LAST_ACK: 1577 set_bit(XPRT_CLOSING, &xprt->state); 1578 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); 1579 smp_mb__before_atomic(); 1580 clear_bit(XPRT_CONNECTED, &xprt->state); 1581 smp_mb__after_atomic(); 1582 break; 1583 case TCP_CLOSE: 1584 xs_tcp_cancel_linger_timeout(xprt); 1585 xs_sock_mark_closed(xprt); 1586 } 1587 out: 1588 read_unlock_bh(&sk->sk_callback_lock); 1589 } 1590 1591 static void xs_write_space(struct sock *sk) 1592 { 1593 struct socket *sock; 1594 struct rpc_xprt *xprt; 1595 1596 if (unlikely(!(sock = sk->sk_socket))) 1597 return; 1598 clear_bit(SOCK_NOSPACE, &sock->flags); 1599 1600 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1601 return; 1602 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) 1603 return; 1604 1605 xprt_write_space(xprt); 1606 } 1607 1608 /** 1609 * xs_udp_write_space - callback invoked when socket buffer space 1610 * becomes available 1611 * @sk: socket whose state has changed 1612 * 1613 * Called when more output buffer space is available for this socket. 1614 * We try not to wake our writers until they can make "significant" 1615 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1616 * with a bunch of small requests. 1617 */ 1618 static void xs_udp_write_space(struct sock *sk) 1619 { 1620 read_lock_bh(&sk->sk_callback_lock); 1621 1622 /* from net/core/sock.c:sock_def_write_space */ 1623 if (sock_writeable(sk)) 1624 xs_write_space(sk); 1625 1626 read_unlock_bh(&sk->sk_callback_lock); 1627 } 1628 1629 /** 1630 * xs_tcp_write_space - callback invoked when socket buffer space 1631 * becomes available 1632 * @sk: socket whose state has changed 1633 * 1634 * Called when more output buffer space is available for this socket. 1635 * We try not to wake our writers until they can make "significant" 1636 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1637 * with a bunch of small requests. 1638 */ 1639 static void xs_tcp_write_space(struct sock *sk) 1640 { 1641 read_lock_bh(&sk->sk_callback_lock); 1642 1643 /* from net/core/stream.c:sk_stream_write_space */ 1644 if (sk_stream_is_writeable(sk)) 1645 xs_write_space(sk); 1646 1647 read_unlock_bh(&sk->sk_callback_lock); 1648 } 1649 1650 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1651 { 1652 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1653 struct sock *sk = transport->inet; 1654 1655 if (transport->rcvsize) { 1656 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1657 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1658 } 1659 if (transport->sndsize) { 1660 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1661 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1662 sk->sk_write_space(sk); 1663 } 1664 } 1665 1666 /** 1667 * xs_udp_set_buffer_size - set send and receive limits 1668 * @xprt: generic transport 1669 * @sndsize: requested size of send buffer, in bytes 1670 * @rcvsize: requested size of receive buffer, in bytes 1671 * 1672 * Set socket send and receive buffer size limits. 1673 */ 1674 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1675 { 1676 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1677 1678 transport->sndsize = 0; 1679 if (sndsize) 1680 transport->sndsize = sndsize + 1024; 1681 transport->rcvsize = 0; 1682 if (rcvsize) 1683 transport->rcvsize = rcvsize + 1024; 1684 1685 xs_udp_do_set_buffer_size(xprt); 1686 } 1687 1688 /** 1689 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1690 * @task: task that timed out 1691 * 1692 * Adjust the congestion window after a retransmit timeout has occurred. 1693 */ 1694 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1695 { 1696 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1697 } 1698 1699 static unsigned short xs_get_random_port(void) 1700 { 1701 unsigned short range = xprt_max_resvport - xprt_min_resvport; 1702 unsigned short rand = (unsigned short) prandom_u32() % range; 1703 return rand + xprt_min_resvport; 1704 } 1705 1706 /** 1707 * xs_set_port - reset the port number in the remote endpoint address 1708 * @xprt: generic transport 1709 * @port: new port number 1710 * 1711 */ 1712 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1713 { 1714 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1715 1716 rpc_set_port(xs_addr(xprt), port); 1717 xs_update_peer_port(xprt); 1718 } 1719 1720 static unsigned short xs_get_srcport(struct sock_xprt *transport) 1721 { 1722 unsigned short port = transport->srcport; 1723 1724 if (port == 0 && transport->xprt.resvport) 1725 port = xs_get_random_port(); 1726 return port; 1727 } 1728 1729 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1730 { 1731 if (transport->srcport != 0) 1732 transport->srcport = 0; 1733 if (!transport->xprt.resvport) 1734 return 0; 1735 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1736 return xprt_max_resvport; 1737 return --port; 1738 } 1739 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1740 { 1741 struct sockaddr_storage myaddr; 1742 int err, nloop = 0; 1743 unsigned short port = xs_get_srcport(transport); 1744 unsigned short last; 1745 1746 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1747 do { 1748 rpc_set_port((struct sockaddr *)&myaddr, port); 1749 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1750 transport->xprt.addrlen); 1751 if (port == 0) 1752 break; 1753 if (err == 0) { 1754 transport->srcport = port; 1755 break; 1756 } 1757 last = port; 1758 port = xs_next_srcport(transport, port); 1759 if (port > last) 1760 nloop++; 1761 } while (err == -EADDRINUSE && nloop != 2); 1762 1763 if (myaddr.ss_family == AF_INET) 1764 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1765 &((struct sockaddr_in *)&myaddr)->sin_addr, 1766 port, err ? "failed" : "ok", err); 1767 else 1768 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1769 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1770 port, err ? "failed" : "ok", err); 1771 return err; 1772 } 1773 1774 /* 1775 * We don't support autobind on AF_LOCAL sockets 1776 */ 1777 static void xs_local_rpcbind(struct rpc_task *task) 1778 { 1779 rcu_read_lock(); 1780 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt)); 1781 rcu_read_unlock(); 1782 } 1783 1784 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1785 { 1786 } 1787 1788 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1789 static struct lock_class_key xs_key[2]; 1790 static struct lock_class_key xs_slock_key[2]; 1791 1792 static inline void xs_reclassify_socketu(struct socket *sock) 1793 { 1794 struct sock *sk = sock->sk; 1795 1796 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1797 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); 1798 } 1799 1800 static inline void xs_reclassify_socket4(struct socket *sock) 1801 { 1802 struct sock *sk = sock->sk; 1803 1804 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1805 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1806 } 1807 1808 static inline void xs_reclassify_socket6(struct socket *sock) 1809 { 1810 struct sock *sk = sock->sk; 1811 1812 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1813 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1814 } 1815 1816 static inline void xs_reclassify_socket(int family, struct socket *sock) 1817 { 1818 WARN_ON_ONCE(sock_owned_by_user(sock->sk)); 1819 if (sock_owned_by_user(sock->sk)) 1820 return; 1821 1822 switch (family) { 1823 case AF_LOCAL: 1824 xs_reclassify_socketu(sock); 1825 break; 1826 case AF_INET: 1827 xs_reclassify_socket4(sock); 1828 break; 1829 case AF_INET6: 1830 xs_reclassify_socket6(sock); 1831 break; 1832 } 1833 } 1834 #else 1835 static inline void xs_reclassify_socketu(struct socket *sock) 1836 { 1837 } 1838 1839 static inline void xs_reclassify_socket4(struct socket *sock) 1840 { 1841 } 1842 1843 static inline void xs_reclassify_socket6(struct socket *sock) 1844 { 1845 } 1846 1847 static inline void xs_reclassify_socket(int family, struct socket *sock) 1848 { 1849 } 1850 #endif 1851 1852 static void xs_dummy_setup_socket(struct work_struct *work) 1853 { 1854 } 1855 1856 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1857 struct sock_xprt *transport, int family, int type, int protocol) 1858 { 1859 struct socket *sock; 1860 int err; 1861 1862 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1863 if (err < 0) { 1864 dprintk("RPC: can't create %d transport socket (%d).\n", 1865 protocol, -err); 1866 goto out; 1867 } 1868 xs_reclassify_socket(family, sock); 1869 1870 err = xs_bind(transport, sock); 1871 if (err) { 1872 sock_release(sock); 1873 goto out; 1874 } 1875 1876 return sock; 1877 out: 1878 return ERR_PTR(err); 1879 } 1880 1881 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1882 struct socket *sock) 1883 { 1884 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1885 xprt); 1886 1887 if (!transport->inet) { 1888 struct sock *sk = sock->sk; 1889 1890 write_lock_bh(&sk->sk_callback_lock); 1891 1892 xs_save_old_callbacks(transport, sk); 1893 1894 sk->sk_user_data = xprt; 1895 sk->sk_data_ready = xs_local_data_ready; 1896 sk->sk_write_space = xs_udp_write_space; 1897 sk->sk_error_report = xs_error_report; 1898 sk->sk_allocation = GFP_ATOMIC; 1899 1900 xprt_clear_connected(xprt); 1901 1902 /* Reset to new socket */ 1903 transport->sock = sock; 1904 transport->inet = sk; 1905 1906 write_unlock_bh(&sk->sk_callback_lock); 1907 } 1908 1909 /* Tell the socket layer to start connecting... */ 1910 xprt->stat.connect_count++; 1911 xprt->stat.connect_start = jiffies; 1912 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1913 } 1914 1915 /** 1916 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1917 * @xprt: RPC transport to connect 1918 * @transport: socket transport to connect 1919 * @create_sock: function to create a socket of the correct type 1920 */ 1921 static int xs_local_setup_socket(struct sock_xprt *transport) 1922 { 1923 struct rpc_xprt *xprt = &transport->xprt; 1924 struct socket *sock; 1925 int status = -EIO; 1926 1927 current->flags |= PF_FSTRANS; 1928 1929 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1930 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1931 SOCK_STREAM, 0, &sock, 1); 1932 if (status < 0) { 1933 dprintk("RPC: can't create AF_LOCAL " 1934 "transport socket (%d).\n", -status); 1935 goto out; 1936 } 1937 xs_reclassify_socketu(sock); 1938 1939 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1940 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1941 1942 status = xs_local_finish_connecting(xprt, sock); 1943 trace_rpc_socket_connect(xprt, sock, status); 1944 switch (status) { 1945 case 0: 1946 dprintk("RPC: xprt %p connected to %s\n", 1947 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1948 xprt_set_connected(xprt); 1949 break; 1950 case -ENOENT: 1951 dprintk("RPC: xprt %p: socket %s does not exist\n", 1952 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1953 break; 1954 case -ECONNREFUSED: 1955 dprintk("RPC: xprt %p: connection refused for %s\n", 1956 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1957 break; 1958 default: 1959 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 1960 __func__, -status, 1961 xprt->address_strings[RPC_DISPLAY_ADDR]); 1962 } 1963 1964 out: 1965 xprt_clear_connecting(xprt); 1966 xprt_wake_pending_tasks(xprt, status); 1967 current->flags &= ~PF_FSTRANS; 1968 return status; 1969 } 1970 1971 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 1972 { 1973 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1974 int ret; 1975 1976 if (RPC_IS_ASYNC(task)) { 1977 /* 1978 * We want the AF_LOCAL connect to be resolved in the 1979 * filesystem namespace of the process making the rpc 1980 * call. Thus we connect synchronously. 1981 * 1982 * If we want to support asynchronous AF_LOCAL calls, 1983 * we'll need to figure out how to pass a namespace to 1984 * connect. 1985 */ 1986 rpc_exit(task, -ENOTCONN); 1987 return; 1988 } 1989 ret = xs_local_setup_socket(transport); 1990 if (ret && !RPC_IS_SOFTCONN(task)) 1991 msleep_interruptible(15000); 1992 } 1993 1994 #ifdef CONFIG_SUNRPC_SWAP 1995 static void xs_set_memalloc(struct rpc_xprt *xprt) 1996 { 1997 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1998 xprt); 1999 2000 if (xprt->swapper) 2001 sk_set_memalloc(transport->inet); 2002 } 2003 2004 /** 2005 * xs_swapper - Tag this transport as being used for swap. 2006 * @xprt: transport to tag 2007 * @enable: enable/disable 2008 * 2009 */ 2010 int xs_swapper(struct rpc_xprt *xprt, int enable) 2011 { 2012 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2013 xprt); 2014 int err = 0; 2015 2016 if (enable) { 2017 xprt->swapper++; 2018 xs_set_memalloc(xprt); 2019 } else if (xprt->swapper) { 2020 xprt->swapper--; 2021 sk_clear_memalloc(transport->inet); 2022 } 2023 2024 return err; 2025 } 2026 EXPORT_SYMBOL_GPL(xs_swapper); 2027 #else 2028 static void xs_set_memalloc(struct rpc_xprt *xprt) 2029 { 2030 } 2031 #endif 2032 2033 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2034 { 2035 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2036 2037 if (!transport->inet) { 2038 struct sock *sk = sock->sk; 2039 2040 write_lock_bh(&sk->sk_callback_lock); 2041 2042 xs_save_old_callbacks(transport, sk); 2043 2044 sk->sk_user_data = xprt; 2045 sk->sk_data_ready = xs_udp_data_ready; 2046 sk->sk_write_space = xs_udp_write_space; 2047 sk->sk_allocation = GFP_ATOMIC; 2048 2049 xprt_set_connected(xprt); 2050 2051 /* Reset to new socket */ 2052 transport->sock = sock; 2053 transport->inet = sk; 2054 2055 xs_set_memalloc(xprt); 2056 2057 write_unlock_bh(&sk->sk_callback_lock); 2058 } 2059 xs_udp_do_set_buffer_size(xprt); 2060 } 2061 2062 static void xs_udp_setup_socket(struct work_struct *work) 2063 { 2064 struct sock_xprt *transport = 2065 container_of(work, struct sock_xprt, connect_worker.work); 2066 struct rpc_xprt *xprt = &transport->xprt; 2067 struct socket *sock = transport->sock; 2068 int status = -EIO; 2069 2070 current->flags |= PF_FSTRANS; 2071 2072 /* Start by resetting any existing state */ 2073 xs_reset_transport(transport); 2074 sock = xs_create_sock(xprt, transport, 2075 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP); 2076 if (IS_ERR(sock)) 2077 goto out; 2078 2079 dprintk("RPC: worker connecting xprt %p via %s to " 2080 "%s (port %s)\n", xprt, 2081 xprt->address_strings[RPC_DISPLAY_PROTO], 2082 xprt->address_strings[RPC_DISPLAY_ADDR], 2083 xprt->address_strings[RPC_DISPLAY_PORT]); 2084 2085 xs_udp_finish_connecting(xprt, sock); 2086 trace_rpc_socket_connect(xprt, sock, 0); 2087 status = 0; 2088 out: 2089 xprt_clear_connecting(xprt); 2090 xprt_wake_pending_tasks(xprt, status); 2091 current->flags &= ~PF_FSTRANS; 2092 } 2093 2094 /* 2095 * We need to preserve the port number so the reply cache on the server can 2096 * find our cached RPC replies when we get around to reconnecting. 2097 */ 2098 static void xs_abort_connection(struct sock_xprt *transport) 2099 { 2100 int result; 2101 struct sockaddr any; 2102 2103 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport); 2104 2105 /* 2106 * Disconnect the transport socket by doing a connect operation 2107 * with AF_UNSPEC. This should return immediately... 2108 */ 2109 memset(&any, 0, sizeof(any)); 2110 any.sa_family = AF_UNSPEC; 2111 result = kernel_connect(transport->sock, &any, sizeof(any), 0); 2112 trace_rpc_socket_reset_connection(&transport->xprt, 2113 transport->sock, result); 2114 if (!result) 2115 xs_sock_reset_connection_flags(&transport->xprt); 2116 dprintk("RPC: AF_UNSPEC connect return code %d\n", result); 2117 } 2118 2119 static void xs_tcp_reuse_connection(struct sock_xprt *transport) 2120 { 2121 unsigned int state = transport->inet->sk_state; 2122 2123 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) { 2124 /* we don't need to abort the connection if the socket 2125 * hasn't undergone a shutdown 2126 */ 2127 if (transport->inet->sk_shutdown == 0) 2128 return; 2129 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n", 2130 __func__, transport->inet->sk_shutdown); 2131 } 2132 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) { 2133 /* we don't need to abort the connection if the socket 2134 * hasn't undergone a shutdown 2135 */ 2136 if (transport->inet->sk_shutdown == 0) 2137 return; 2138 dprintk("RPC: %s: ESTABLISHED/SYN_SENT " 2139 "sk_shutdown set to %d\n", 2140 __func__, transport->inet->sk_shutdown); 2141 } 2142 xs_abort_connection(transport); 2143 } 2144 2145 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2146 { 2147 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2148 int ret = -ENOTCONN; 2149 2150 if (!transport->inet) { 2151 struct sock *sk = sock->sk; 2152 unsigned int keepidle = xprt->timeout->to_initval / HZ; 2153 unsigned int keepcnt = xprt->timeout->to_retries + 1; 2154 unsigned int opt_on = 1; 2155 2156 /* TCP Keepalive options */ 2157 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 2158 (char *)&opt_on, sizeof(opt_on)); 2159 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, 2160 (char *)&keepidle, sizeof(keepidle)); 2161 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, 2162 (char *)&keepidle, sizeof(keepidle)); 2163 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, 2164 (char *)&keepcnt, sizeof(keepcnt)); 2165 2166 write_lock_bh(&sk->sk_callback_lock); 2167 2168 xs_save_old_callbacks(transport, sk); 2169 2170 sk->sk_user_data = xprt; 2171 sk->sk_data_ready = xs_tcp_data_ready; 2172 sk->sk_state_change = xs_tcp_state_change; 2173 sk->sk_write_space = xs_tcp_write_space; 2174 sk->sk_error_report = xs_error_report; 2175 sk->sk_allocation = GFP_ATOMIC; 2176 2177 /* socket options */ 2178 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 2179 sock_reset_flag(sk, SOCK_LINGER); 2180 tcp_sk(sk)->linger2 = 0; 2181 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 2182 2183 xprt_clear_connected(xprt); 2184 2185 /* Reset to new socket */ 2186 transport->sock = sock; 2187 transport->inet = sk; 2188 2189 write_unlock_bh(&sk->sk_callback_lock); 2190 } 2191 2192 if (!xprt_bound(xprt)) 2193 goto out; 2194 2195 xs_set_memalloc(xprt); 2196 2197 /* Tell the socket layer to start connecting... */ 2198 xprt->stat.connect_count++; 2199 xprt->stat.connect_start = jiffies; 2200 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2201 switch (ret) { 2202 case 0: 2203 case -EINPROGRESS: 2204 /* SYN_SENT! */ 2205 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2206 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2207 } 2208 out: 2209 return ret; 2210 } 2211 2212 /** 2213 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2214 * @xprt: RPC transport to connect 2215 * @transport: socket transport to connect 2216 * @create_sock: function to create a socket of the correct type 2217 * 2218 * Invoked by a work queue tasklet. 2219 */ 2220 static void xs_tcp_setup_socket(struct work_struct *work) 2221 { 2222 struct sock_xprt *transport = 2223 container_of(work, struct sock_xprt, connect_worker.work); 2224 struct socket *sock = transport->sock; 2225 struct rpc_xprt *xprt = &transport->xprt; 2226 int status = -EIO; 2227 2228 current->flags |= PF_FSTRANS; 2229 2230 if (!sock) { 2231 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 2232 sock = xs_create_sock(xprt, transport, 2233 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP); 2234 if (IS_ERR(sock)) { 2235 status = PTR_ERR(sock); 2236 goto out; 2237 } 2238 } else { 2239 int abort_and_exit; 2240 2241 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT, 2242 &xprt->state); 2243 /* "close" the socket, preserving the local port */ 2244 xs_tcp_reuse_connection(transport); 2245 2246 if (abort_and_exit) 2247 goto out_eagain; 2248 } 2249 2250 dprintk("RPC: worker connecting xprt %p via %s to " 2251 "%s (port %s)\n", xprt, 2252 xprt->address_strings[RPC_DISPLAY_PROTO], 2253 xprt->address_strings[RPC_DISPLAY_ADDR], 2254 xprt->address_strings[RPC_DISPLAY_PORT]); 2255 2256 status = xs_tcp_finish_connecting(xprt, sock); 2257 trace_rpc_socket_connect(xprt, sock, status); 2258 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2259 xprt, -status, xprt_connected(xprt), 2260 sock->sk->sk_state); 2261 switch (status) { 2262 default: 2263 printk("%s: connect returned unhandled error %d\n", 2264 __func__, status); 2265 case -EADDRNOTAVAIL: 2266 /* We're probably in TIME_WAIT. Get rid of existing socket, 2267 * and retry 2268 */ 2269 xs_tcp_force_close(xprt); 2270 break; 2271 case 0: 2272 case -EINPROGRESS: 2273 case -EALREADY: 2274 xprt_clear_connecting(xprt); 2275 current->flags &= ~PF_FSTRANS; 2276 return; 2277 case -EINVAL: 2278 /* Happens, for instance, if the user specified a link 2279 * local IPv6 address without a scope-id. 2280 */ 2281 case -ECONNREFUSED: 2282 case -ECONNRESET: 2283 case -ENETUNREACH: 2284 /* retry with existing socket, after a delay */ 2285 goto out; 2286 } 2287 out_eagain: 2288 status = -EAGAIN; 2289 out: 2290 xprt_clear_connecting(xprt); 2291 xprt_wake_pending_tasks(xprt, status); 2292 current->flags &= ~PF_FSTRANS; 2293 } 2294 2295 /** 2296 * xs_connect - connect a socket to a remote endpoint 2297 * @xprt: pointer to transport structure 2298 * @task: address of RPC task that manages state of connect request 2299 * 2300 * TCP: If the remote end dropped the connection, delay reconnecting. 2301 * 2302 * UDP socket connects are synchronous, but we use a work queue anyway 2303 * to guarantee that even unprivileged user processes can set up a 2304 * socket on a privileged port. 2305 * 2306 * If a UDP socket connect fails, the delay behavior here prevents 2307 * retry floods (hard mounts). 2308 */ 2309 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2310 { 2311 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2312 2313 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) { 2314 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2315 "seconds\n", 2316 xprt, xprt->reestablish_timeout / HZ); 2317 queue_delayed_work(rpciod_workqueue, 2318 &transport->connect_worker, 2319 xprt->reestablish_timeout); 2320 xprt->reestablish_timeout <<= 1; 2321 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2322 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2323 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) 2324 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; 2325 } else { 2326 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2327 queue_delayed_work(rpciod_workqueue, 2328 &transport->connect_worker, 0); 2329 } 2330 } 2331 2332 /** 2333 * xs_local_print_stats - display AF_LOCAL socket-specifc stats 2334 * @xprt: rpc_xprt struct containing statistics 2335 * @seq: output file 2336 * 2337 */ 2338 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2339 { 2340 long idle_time = 0; 2341 2342 if (xprt_connected(xprt)) 2343 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2344 2345 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2346 "%llu %llu %lu %llu %llu\n", 2347 xprt->stat.bind_count, 2348 xprt->stat.connect_count, 2349 xprt->stat.connect_time, 2350 idle_time, 2351 xprt->stat.sends, 2352 xprt->stat.recvs, 2353 xprt->stat.bad_xids, 2354 xprt->stat.req_u, 2355 xprt->stat.bklog_u, 2356 xprt->stat.max_slots, 2357 xprt->stat.sending_u, 2358 xprt->stat.pending_u); 2359 } 2360 2361 /** 2362 * xs_udp_print_stats - display UDP socket-specifc stats 2363 * @xprt: rpc_xprt struct containing statistics 2364 * @seq: output file 2365 * 2366 */ 2367 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2368 { 2369 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2370 2371 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2372 "%lu %llu %llu\n", 2373 transport->srcport, 2374 xprt->stat.bind_count, 2375 xprt->stat.sends, 2376 xprt->stat.recvs, 2377 xprt->stat.bad_xids, 2378 xprt->stat.req_u, 2379 xprt->stat.bklog_u, 2380 xprt->stat.max_slots, 2381 xprt->stat.sending_u, 2382 xprt->stat.pending_u); 2383 } 2384 2385 /** 2386 * xs_tcp_print_stats - display TCP socket-specifc stats 2387 * @xprt: rpc_xprt struct containing statistics 2388 * @seq: output file 2389 * 2390 */ 2391 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2392 { 2393 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2394 long idle_time = 0; 2395 2396 if (xprt_connected(xprt)) 2397 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2398 2399 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2400 "%llu %llu %lu %llu %llu\n", 2401 transport->srcport, 2402 xprt->stat.bind_count, 2403 xprt->stat.connect_count, 2404 xprt->stat.connect_time, 2405 idle_time, 2406 xprt->stat.sends, 2407 xprt->stat.recvs, 2408 xprt->stat.bad_xids, 2409 xprt->stat.req_u, 2410 xprt->stat.bklog_u, 2411 xprt->stat.max_slots, 2412 xprt->stat.sending_u, 2413 xprt->stat.pending_u); 2414 } 2415 2416 /* 2417 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2418 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2419 * to use the server side send routines. 2420 */ 2421 static void *bc_malloc(struct rpc_task *task, size_t size) 2422 { 2423 struct page *page; 2424 struct rpc_buffer *buf; 2425 2426 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer)); 2427 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) 2428 return NULL; 2429 2430 page = alloc_page(GFP_KERNEL); 2431 if (!page) 2432 return NULL; 2433 2434 buf = page_address(page); 2435 buf->len = PAGE_SIZE; 2436 2437 return buf->data; 2438 } 2439 2440 /* 2441 * Free the space allocated in the bc_alloc routine 2442 */ 2443 static void bc_free(void *buffer) 2444 { 2445 struct rpc_buffer *buf; 2446 2447 if (!buffer) 2448 return; 2449 2450 buf = container_of(buffer, struct rpc_buffer, data); 2451 free_page((unsigned long)buf); 2452 } 2453 2454 /* 2455 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex 2456 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request. 2457 */ 2458 static int bc_sendto(struct rpc_rqst *req) 2459 { 2460 int len; 2461 struct xdr_buf *xbufp = &req->rq_snd_buf; 2462 struct rpc_xprt *xprt = req->rq_xprt; 2463 struct sock_xprt *transport = 2464 container_of(xprt, struct sock_xprt, xprt); 2465 struct socket *sock = transport->sock; 2466 unsigned long headoff; 2467 unsigned long tailoff; 2468 2469 xs_encode_stream_record_marker(xbufp); 2470 2471 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK; 2472 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK; 2473 len = svc_send_common(sock, xbufp, 2474 virt_to_page(xbufp->head[0].iov_base), headoff, 2475 xbufp->tail[0].iov_base, tailoff); 2476 2477 if (len != xbufp->len) { 2478 printk(KERN_NOTICE "Error sending entire callback!\n"); 2479 len = -EAGAIN; 2480 } 2481 2482 return len; 2483 } 2484 2485 /* 2486 * The send routine. Borrows from svc_send 2487 */ 2488 static int bc_send_request(struct rpc_task *task) 2489 { 2490 struct rpc_rqst *req = task->tk_rqstp; 2491 struct svc_xprt *xprt; 2492 u32 len; 2493 2494 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid)); 2495 /* 2496 * Get the server socket associated with this callback xprt 2497 */ 2498 xprt = req->rq_xprt->bc_xprt; 2499 2500 /* 2501 * Grab the mutex to serialize data as the connection is shared 2502 * with the fore channel 2503 */ 2504 if (!mutex_trylock(&xprt->xpt_mutex)) { 2505 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL); 2506 if (!mutex_trylock(&xprt->xpt_mutex)) 2507 return -EAGAIN; 2508 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task); 2509 } 2510 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2511 len = -ENOTCONN; 2512 else 2513 len = bc_sendto(req); 2514 mutex_unlock(&xprt->xpt_mutex); 2515 2516 if (len > 0) 2517 len = 0; 2518 2519 return len; 2520 } 2521 2522 /* 2523 * The close routine. Since this is client initiated, we do nothing 2524 */ 2525 2526 static void bc_close(struct rpc_xprt *xprt) 2527 { 2528 } 2529 2530 /* 2531 * The xprt destroy routine. Again, because this connection is client 2532 * initiated, we do nothing 2533 */ 2534 2535 static void bc_destroy(struct rpc_xprt *xprt) 2536 { 2537 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2538 2539 xs_xprt_free(xprt); 2540 module_put(THIS_MODULE); 2541 } 2542 2543 static struct rpc_xprt_ops xs_local_ops = { 2544 .reserve_xprt = xprt_reserve_xprt, 2545 .release_xprt = xs_tcp_release_xprt, 2546 .alloc_slot = xprt_alloc_slot, 2547 .rpcbind = xs_local_rpcbind, 2548 .set_port = xs_local_set_port, 2549 .connect = xs_local_connect, 2550 .buf_alloc = rpc_malloc, 2551 .buf_free = rpc_free, 2552 .send_request = xs_local_send_request, 2553 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2554 .close = xs_close, 2555 .destroy = xs_destroy, 2556 .print_stats = xs_local_print_stats, 2557 }; 2558 2559 static struct rpc_xprt_ops xs_udp_ops = { 2560 .set_buffer_size = xs_udp_set_buffer_size, 2561 .reserve_xprt = xprt_reserve_xprt_cong, 2562 .release_xprt = xprt_release_xprt_cong, 2563 .alloc_slot = xprt_alloc_slot, 2564 .rpcbind = rpcb_getport_async, 2565 .set_port = xs_set_port, 2566 .connect = xs_connect, 2567 .buf_alloc = rpc_malloc, 2568 .buf_free = rpc_free, 2569 .send_request = xs_udp_send_request, 2570 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 2571 .timer = xs_udp_timer, 2572 .release_request = xprt_release_rqst_cong, 2573 .close = xs_close, 2574 .destroy = xs_destroy, 2575 .print_stats = xs_udp_print_stats, 2576 }; 2577 2578 static struct rpc_xprt_ops xs_tcp_ops = { 2579 .reserve_xprt = xprt_reserve_xprt, 2580 .release_xprt = xs_tcp_release_xprt, 2581 .alloc_slot = xprt_lock_and_alloc_slot, 2582 .rpcbind = rpcb_getport_async, 2583 .set_port = xs_set_port, 2584 .connect = xs_connect, 2585 .buf_alloc = rpc_malloc, 2586 .buf_free = rpc_free, 2587 .send_request = xs_tcp_send_request, 2588 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2589 .close = xs_tcp_close, 2590 .destroy = xs_destroy, 2591 .print_stats = xs_tcp_print_stats, 2592 }; 2593 2594 /* 2595 * The rpc_xprt_ops for the server backchannel 2596 */ 2597 2598 static struct rpc_xprt_ops bc_tcp_ops = { 2599 .reserve_xprt = xprt_reserve_xprt, 2600 .release_xprt = xprt_release_xprt, 2601 .alloc_slot = xprt_alloc_slot, 2602 .buf_alloc = bc_malloc, 2603 .buf_free = bc_free, 2604 .send_request = bc_send_request, 2605 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2606 .close = bc_close, 2607 .destroy = bc_destroy, 2608 .print_stats = xs_tcp_print_stats, 2609 }; 2610 2611 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2612 { 2613 static const struct sockaddr_in sin = { 2614 .sin_family = AF_INET, 2615 .sin_addr.s_addr = htonl(INADDR_ANY), 2616 }; 2617 static const struct sockaddr_in6 sin6 = { 2618 .sin6_family = AF_INET6, 2619 .sin6_addr = IN6ADDR_ANY_INIT, 2620 }; 2621 2622 switch (family) { 2623 case AF_LOCAL: 2624 break; 2625 case AF_INET: 2626 memcpy(sap, &sin, sizeof(sin)); 2627 break; 2628 case AF_INET6: 2629 memcpy(sap, &sin6, sizeof(sin6)); 2630 break; 2631 default: 2632 dprintk("RPC: %s: Bad address family\n", __func__); 2633 return -EAFNOSUPPORT; 2634 } 2635 return 0; 2636 } 2637 2638 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2639 unsigned int slot_table_size, 2640 unsigned int max_slot_table_size) 2641 { 2642 struct rpc_xprt *xprt; 2643 struct sock_xprt *new; 2644 2645 if (args->addrlen > sizeof(xprt->addr)) { 2646 dprintk("RPC: xs_setup_xprt: address too large\n"); 2647 return ERR_PTR(-EBADF); 2648 } 2649 2650 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2651 max_slot_table_size); 2652 if (xprt == NULL) { 2653 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2654 "rpc_xprt\n"); 2655 return ERR_PTR(-ENOMEM); 2656 } 2657 2658 new = container_of(xprt, struct sock_xprt, xprt); 2659 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2660 xprt->addrlen = args->addrlen; 2661 if (args->srcaddr) 2662 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2663 else { 2664 int err; 2665 err = xs_init_anyaddr(args->dstaddr->sa_family, 2666 (struct sockaddr *)&new->srcaddr); 2667 if (err != 0) { 2668 xprt_free(xprt); 2669 return ERR_PTR(err); 2670 } 2671 } 2672 2673 return xprt; 2674 } 2675 2676 static const struct rpc_timeout xs_local_default_timeout = { 2677 .to_initval = 10 * HZ, 2678 .to_maxval = 10 * HZ, 2679 .to_retries = 2, 2680 }; 2681 2682 /** 2683 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2684 * @args: rpc transport creation arguments 2685 * 2686 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2687 */ 2688 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2689 { 2690 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2691 struct sock_xprt *transport; 2692 struct rpc_xprt *xprt; 2693 struct rpc_xprt *ret; 2694 2695 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2696 xprt_max_tcp_slot_table_entries); 2697 if (IS_ERR(xprt)) 2698 return xprt; 2699 transport = container_of(xprt, struct sock_xprt, xprt); 2700 2701 xprt->prot = 0; 2702 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2703 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2704 2705 xprt->bind_timeout = XS_BIND_TO; 2706 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2707 xprt->idle_timeout = XS_IDLE_DISC_TO; 2708 2709 xprt->ops = &xs_local_ops; 2710 xprt->timeout = &xs_local_default_timeout; 2711 2712 INIT_DELAYED_WORK(&transport->connect_worker, 2713 xs_dummy_setup_socket); 2714 2715 switch (sun->sun_family) { 2716 case AF_LOCAL: 2717 if (sun->sun_path[0] != '/') { 2718 dprintk("RPC: bad AF_LOCAL address: %s\n", 2719 sun->sun_path); 2720 ret = ERR_PTR(-EINVAL); 2721 goto out_err; 2722 } 2723 xprt_set_bound(xprt); 2724 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2725 ret = ERR_PTR(xs_local_setup_socket(transport)); 2726 if (ret) 2727 goto out_err; 2728 break; 2729 default: 2730 ret = ERR_PTR(-EAFNOSUPPORT); 2731 goto out_err; 2732 } 2733 2734 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2735 xprt->address_strings[RPC_DISPLAY_ADDR]); 2736 2737 if (try_module_get(THIS_MODULE)) 2738 return xprt; 2739 ret = ERR_PTR(-EINVAL); 2740 out_err: 2741 xs_xprt_free(xprt); 2742 return ret; 2743 } 2744 2745 static const struct rpc_timeout xs_udp_default_timeout = { 2746 .to_initval = 5 * HZ, 2747 .to_maxval = 30 * HZ, 2748 .to_increment = 5 * HZ, 2749 .to_retries = 5, 2750 }; 2751 2752 /** 2753 * xs_setup_udp - Set up transport to use a UDP socket 2754 * @args: rpc transport creation arguments 2755 * 2756 */ 2757 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2758 { 2759 struct sockaddr *addr = args->dstaddr; 2760 struct rpc_xprt *xprt; 2761 struct sock_xprt *transport; 2762 struct rpc_xprt *ret; 2763 2764 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2765 xprt_udp_slot_table_entries); 2766 if (IS_ERR(xprt)) 2767 return xprt; 2768 transport = container_of(xprt, struct sock_xprt, xprt); 2769 2770 xprt->prot = IPPROTO_UDP; 2771 xprt->tsh_size = 0; 2772 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2773 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2774 2775 xprt->bind_timeout = XS_BIND_TO; 2776 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2777 xprt->idle_timeout = XS_IDLE_DISC_TO; 2778 2779 xprt->ops = &xs_udp_ops; 2780 2781 xprt->timeout = &xs_udp_default_timeout; 2782 2783 switch (addr->sa_family) { 2784 case AF_INET: 2785 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2786 xprt_set_bound(xprt); 2787 2788 INIT_DELAYED_WORK(&transport->connect_worker, 2789 xs_udp_setup_socket); 2790 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2791 break; 2792 case AF_INET6: 2793 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2794 xprt_set_bound(xprt); 2795 2796 INIT_DELAYED_WORK(&transport->connect_worker, 2797 xs_udp_setup_socket); 2798 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2799 break; 2800 default: 2801 ret = ERR_PTR(-EAFNOSUPPORT); 2802 goto out_err; 2803 } 2804 2805 if (xprt_bound(xprt)) 2806 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2807 xprt->address_strings[RPC_DISPLAY_ADDR], 2808 xprt->address_strings[RPC_DISPLAY_PORT], 2809 xprt->address_strings[RPC_DISPLAY_PROTO]); 2810 else 2811 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2812 xprt->address_strings[RPC_DISPLAY_ADDR], 2813 xprt->address_strings[RPC_DISPLAY_PROTO]); 2814 2815 if (try_module_get(THIS_MODULE)) 2816 return xprt; 2817 ret = ERR_PTR(-EINVAL); 2818 out_err: 2819 xs_xprt_free(xprt); 2820 return ret; 2821 } 2822 2823 static const struct rpc_timeout xs_tcp_default_timeout = { 2824 .to_initval = 60 * HZ, 2825 .to_maxval = 60 * HZ, 2826 .to_retries = 2, 2827 }; 2828 2829 /** 2830 * xs_setup_tcp - Set up transport to use a TCP socket 2831 * @args: rpc transport creation arguments 2832 * 2833 */ 2834 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2835 { 2836 struct sockaddr *addr = args->dstaddr; 2837 struct rpc_xprt *xprt; 2838 struct sock_xprt *transport; 2839 struct rpc_xprt *ret; 2840 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2841 2842 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2843 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2844 2845 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2846 max_slot_table_size); 2847 if (IS_ERR(xprt)) 2848 return xprt; 2849 transport = container_of(xprt, struct sock_xprt, xprt); 2850 2851 xprt->prot = IPPROTO_TCP; 2852 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2853 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2854 2855 xprt->bind_timeout = XS_BIND_TO; 2856 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2857 xprt->idle_timeout = XS_IDLE_DISC_TO; 2858 2859 xprt->ops = &xs_tcp_ops; 2860 xprt->timeout = &xs_tcp_default_timeout; 2861 2862 switch (addr->sa_family) { 2863 case AF_INET: 2864 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2865 xprt_set_bound(xprt); 2866 2867 INIT_DELAYED_WORK(&transport->connect_worker, 2868 xs_tcp_setup_socket); 2869 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2870 break; 2871 case AF_INET6: 2872 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2873 xprt_set_bound(xprt); 2874 2875 INIT_DELAYED_WORK(&transport->connect_worker, 2876 xs_tcp_setup_socket); 2877 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 2878 break; 2879 default: 2880 ret = ERR_PTR(-EAFNOSUPPORT); 2881 goto out_err; 2882 } 2883 2884 if (xprt_bound(xprt)) 2885 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2886 xprt->address_strings[RPC_DISPLAY_ADDR], 2887 xprt->address_strings[RPC_DISPLAY_PORT], 2888 xprt->address_strings[RPC_DISPLAY_PROTO]); 2889 else 2890 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2891 xprt->address_strings[RPC_DISPLAY_ADDR], 2892 xprt->address_strings[RPC_DISPLAY_PROTO]); 2893 2894 if (try_module_get(THIS_MODULE)) 2895 return xprt; 2896 ret = ERR_PTR(-EINVAL); 2897 out_err: 2898 xs_xprt_free(xprt); 2899 return ret; 2900 } 2901 2902 /** 2903 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 2904 * @args: rpc transport creation arguments 2905 * 2906 */ 2907 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 2908 { 2909 struct sockaddr *addr = args->dstaddr; 2910 struct rpc_xprt *xprt; 2911 struct sock_xprt *transport; 2912 struct svc_sock *bc_sock; 2913 struct rpc_xprt *ret; 2914 2915 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2916 xprt_tcp_slot_table_entries); 2917 if (IS_ERR(xprt)) 2918 return xprt; 2919 transport = container_of(xprt, struct sock_xprt, xprt); 2920 2921 xprt->prot = IPPROTO_TCP; 2922 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2923 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2924 xprt->timeout = &xs_tcp_default_timeout; 2925 2926 /* backchannel */ 2927 xprt_set_bound(xprt); 2928 xprt->bind_timeout = 0; 2929 xprt->reestablish_timeout = 0; 2930 xprt->idle_timeout = 0; 2931 2932 xprt->ops = &bc_tcp_ops; 2933 2934 switch (addr->sa_family) { 2935 case AF_INET: 2936 xs_format_peer_addresses(xprt, "tcp", 2937 RPCBIND_NETID_TCP); 2938 break; 2939 case AF_INET6: 2940 xs_format_peer_addresses(xprt, "tcp", 2941 RPCBIND_NETID_TCP6); 2942 break; 2943 default: 2944 ret = ERR_PTR(-EAFNOSUPPORT); 2945 goto out_err; 2946 } 2947 2948 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2949 xprt->address_strings[RPC_DISPLAY_ADDR], 2950 xprt->address_strings[RPC_DISPLAY_PORT], 2951 xprt->address_strings[RPC_DISPLAY_PROTO]); 2952 2953 /* 2954 * Once we've associated a backchannel xprt with a connection, 2955 * we want to keep it around as long as the connection lasts, 2956 * in case we need to start using it for a backchannel again; 2957 * this reference won't be dropped until bc_xprt is destroyed. 2958 */ 2959 xprt_get(xprt); 2960 args->bc_xprt->xpt_bc_xprt = xprt; 2961 xprt->bc_xprt = args->bc_xprt; 2962 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 2963 transport->sock = bc_sock->sk_sock; 2964 transport->inet = bc_sock->sk_sk; 2965 2966 /* 2967 * Since we don't want connections for the backchannel, we set 2968 * the xprt status to connected 2969 */ 2970 xprt_set_connected(xprt); 2971 2972 if (try_module_get(THIS_MODULE)) 2973 return xprt; 2974 2975 args->bc_xprt->xpt_bc_xprt = NULL; 2976 xprt_put(xprt); 2977 ret = ERR_PTR(-EINVAL); 2978 out_err: 2979 xs_xprt_free(xprt); 2980 return ret; 2981 } 2982 2983 static struct xprt_class xs_local_transport = { 2984 .list = LIST_HEAD_INIT(xs_local_transport.list), 2985 .name = "named UNIX socket", 2986 .owner = THIS_MODULE, 2987 .ident = XPRT_TRANSPORT_LOCAL, 2988 .setup = xs_setup_local, 2989 }; 2990 2991 static struct xprt_class xs_udp_transport = { 2992 .list = LIST_HEAD_INIT(xs_udp_transport.list), 2993 .name = "udp", 2994 .owner = THIS_MODULE, 2995 .ident = XPRT_TRANSPORT_UDP, 2996 .setup = xs_setup_udp, 2997 }; 2998 2999 static struct xprt_class xs_tcp_transport = { 3000 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3001 .name = "tcp", 3002 .owner = THIS_MODULE, 3003 .ident = XPRT_TRANSPORT_TCP, 3004 .setup = xs_setup_tcp, 3005 }; 3006 3007 static struct xprt_class xs_bc_tcp_transport = { 3008 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3009 .name = "tcp NFSv4.1 backchannel", 3010 .owner = THIS_MODULE, 3011 .ident = XPRT_TRANSPORT_BC_TCP, 3012 .setup = xs_setup_bc_tcp, 3013 }; 3014 3015 /** 3016 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3017 * 3018 */ 3019 int init_socket_xprt(void) 3020 { 3021 #ifdef RPC_DEBUG 3022 if (!sunrpc_table_header) 3023 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3024 #endif 3025 3026 xprt_register_transport(&xs_local_transport); 3027 xprt_register_transport(&xs_udp_transport); 3028 xprt_register_transport(&xs_tcp_transport); 3029 xprt_register_transport(&xs_bc_tcp_transport); 3030 3031 return 0; 3032 } 3033 3034 /** 3035 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3036 * 3037 */ 3038 void cleanup_socket_xprt(void) 3039 { 3040 #ifdef RPC_DEBUG 3041 if (sunrpc_table_header) { 3042 unregister_sysctl_table(sunrpc_table_header); 3043 sunrpc_table_header = NULL; 3044 } 3045 #endif 3046 3047 xprt_unregister_transport(&xs_local_transport); 3048 xprt_unregister_transport(&xs_udp_transport); 3049 xprt_unregister_transport(&xs_tcp_transport); 3050 xprt_unregister_transport(&xs_bc_tcp_transport); 3051 } 3052 3053 static int param_set_uint_minmax(const char *val, 3054 const struct kernel_param *kp, 3055 unsigned int min, unsigned int max) 3056 { 3057 unsigned long num; 3058 int ret; 3059 3060 if (!val) 3061 return -EINVAL; 3062 ret = strict_strtoul(val, 0, &num); 3063 if (ret == -EINVAL || num < min || num > max) 3064 return -EINVAL; 3065 *((unsigned int *)kp->arg) = num; 3066 return 0; 3067 } 3068 3069 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3070 { 3071 return param_set_uint_minmax(val, kp, 3072 RPC_MIN_RESVPORT, 3073 RPC_MAX_RESVPORT); 3074 } 3075 3076 static struct kernel_param_ops param_ops_portnr = { 3077 .set = param_set_portnr, 3078 .get = param_get_uint, 3079 }; 3080 3081 #define param_check_portnr(name, p) \ 3082 __param_check(name, p, unsigned int); 3083 3084 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3085 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3086 3087 static int param_set_slot_table_size(const char *val, 3088 const struct kernel_param *kp) 3089 { 3090 return param_set_uint_minmax(val, kp, 3091 RPC_MIN_SLOT_TABLE, 3092 RPC_MAX_SLOT_TABLE); 3093 } 3094 3095 static struct kernel_param_ops param_ops_slot_table_size = { 3096 .set = param_set_slot_table_size, 3097 .get = param_get_uint, 3098 }; 3099 3100 #define param_check_slot_table_size(name, p) \ 3101 __param_check(name, p, unsigned int); 3102 3103 static int param_set_max_slot_table_size(const char *val, 3104 const struct kernel_param *kp) 3105 { 3106 return param_set_uint_minmax(val, kp, 3107 RPC_MIN_SLOT_TABLE, 3108 RPC_MAX_SLOT_TABLE_LIMIT); 3109 } 3110 3111 static struct kernel_param_ops param_ops_max_slot_table_size = { 3112 .set = param_set_max_slot_table_size, 3113 .get = param_get_uint, 3114 }; 3115 3116 #define param_check_max_slot_table_size(name, p) \ 3117 __param_check(name, p, unsigned int); 3118 3119 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3120 slot_table_size, 0644); 3121 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3122 max_slot_table_size, 0644); 3123 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3124 slot_table_size, 0644); 3125 3126