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 sk->sk_no_check = 0; 870 871 trace_rpc_socket_close(&transport->xprt, sock); 872 sock_release(sock); 873 } 874 875 /** 876 * xs_close - close a socket 877 * @xprt: transport 878 * 879 * This is used when all requests are complete; ie, no DRC state remains 880 * on the server we want to save. 881 * 882 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 883 * xs_reset_transport() zeroing the socket from underneath a writer. 884 */ 885 static void xs_close(struct rpc_xprt *xprt) 886 { 887 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 888 889 dprintk("RPC: xs_close xprt %p\n", xprt); 890 891 cancel_delayed_work_sync(&transport->connect_worker); 892 893 xs_reset_transport(transport); 894 xprt->reestablish_timeout = 0; 895 896 smp_mb__before_clear_bit(); 897 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 898 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 899 clear_bit(XPRT_CLOSING, &xprt->state); 900 smp_mb__after_clear_bit(); 901 xprt_disconnect_done(xprt); 902 } 903 904 static void xs_tcp_close(struct rpc_xprt *xprt) 905 { 906 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state)) 907 xs_close(xprt); 908 else 909 xs_tcp_shutdown(xprt); 910 } 911 912 static void xs_xprt_free(struct rpc_xprt *xprt) 913 { 914 xs_free_peer_addresses(xprt); 915 xprt_free(xprt); 916 } 917 918 /** 919 * xs_destroy - prepare to shutdown a transport 920 * @xprt: doomed transport 921 * 922 */ 923 static void xs_destroy(struct rpc_xprt *xprt) 924 { 925 dprintk("RPC: xs_destroy xprt %p\n", xprt); 926 927 xs_close(xprt); 928 xs_xprt_free(xprt); 929 module_put(THIS_MODULE); 930 } 931 932 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb) 933 { 934 struct xdr_skb_reader desc = { 935 .skb = skb, 936 .offset = sizeof(rpc_fraghdr), 937 .count = skb->len - sizeof(rpc_fraghdr), 938 }; 939 940 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0) 941 return -1; 942 if (desc.count) 943 return -1; 944 return 0; 945 } 946 947 /** 948 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets 949 * @sk: socket with data to read 950 * @len: how much data to read 951 * 952 * Currently this assumes we can read the whole reply in a single gulp. 953 */ 954 static void xs_local_data_ready(struct sock *sk) 955 { 956 struct rpc_task *task; 957 struct rpc_xprt *xprt; 958 struct rpc_rqst *rovr; 959 struct sk_buff *skb; 960 int err, repsize, copied; 961 u32 _xid; 962 __be32 *xp; 963 964 read_lock_bh(&sk->sk_callback_lock); 965 dprintk("RPC: %s...\n", __func__); 966 xprt = xprt_from_sock(sk); 967 if (xprt == NULL) 968 goto out; 969 970 skb = skb_recv_datagram(sk, 0, 1, &err); 971 if (skb == NULL) 972 goto out; 973 974 repsize = skb->len - sizeof(rpc_fraghdr); 975 if (repsize < 4) { 976 dprintk("RPC: impossible RPC reply size %d\n", repsize); 977 goto dropit; 978 } 979 980 /* Copy the XID from the skb... */ 981 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid); 982 if (xp == NULL) 983 goto dropit; 984 985 /* Look up and lock the request corresponding to the given XID */ 986 spin_lock(&xprt->transport_lock); 987 rovr = xprt_lookup_rqst(xprt, *xp); 988 if (!rovr) 989 goto out_unlock; 990 task = rovr->rq_task; 991 992 copied = rovr->rq_private_buf.buflen; 993 if (copied > repsize) 994 copied = repsize; 995 996 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) { 997 dprintk("RPC: sk_buff copy failed\n"); 998 goto out_unlock; 999 } 1000 1001 xprt_complete_rqst(task, copied); 1002 1003 out_unlock: 1004 spin_unlock(&xprt->transport_lock); 1005 dropit: 1006 skb_free_datagram(sk, skb); 1007 out: 1008 read_unlock_bh(&sk->sk_callback_lock); 1009 } 1010 1011 /** 1012 * xs_udp_data_ready - "data ready" callback for UDP sockets 1013 * @sk: socket with data to read 1014 * @len: how much data to read 1015 * 1016 */ 1017 static void xs_udp_data_ready(struct sock *sk) 1018 { 1019 struct rpc_task *task; 1020 struct rpc_xprt *xprt; 1021 struct rpc_rqst *rovr; 1022 struct sk_buff *skb; 1023 int err, repsize, copied; 1024 u32 _xid; 1025 __be32 *xp; 1026 1027 read_lock_bh(&sk->sk_callback_lock); 1028 dprintk("RPC: xs_udp_data_ready...\n"); 1029 if (!(xprt = xprt_from_sock(sk))) 1030 goto out; 1031 1032 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) 1033 goto out; 1034 1035 repsize = skb->len - sizeof(struct udphdr); 1036 if (repsize < 4) { 1037 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1038 goto dropit; 1039 } 1040 1041 /* Copy the XID from the skb... */ 1042 xp = skb_header_pointer(skb, sizeof(struct udphdr), 1043 sizeof(_xid), &_xid); 1044 if (xp == NULL) 1045 goto dropit; 1046 1047 /* Look up and lock the request corresponding to the given XID */ 1048 spin_lock(&xprt->transport_lock); 1049 rovr = xprt_lookup_rqst(xprt, *xp); 1050 if (!rovr) 1051 goto out_unlock; 1052 task = rovr->rq_task; 1053 1054 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1055 copied = repsize; 1056 1057 /* Suck it into the iovec, verify checksum if not done by hw. */ 1058 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1059 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS); 1060 goto out_unlock; 1061 } 1062 1063 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS); 1064 1065 xprt_adjust_cwnd(xprt, task, copied); 1066 xprt_complete_rqst(task, copied); 1067 1068 out_unlock: 1069 spin_unlock(&xprt->transport_lock); 1070 dropit: 1071 skb_free_datagram(sk, skb); 1072 out: 1073 read_unlock_bh(&sk->sk_callback_lock); 1074 } 1075 1076 /* 1077 * Helper function to force a TCP close if the server is sending 1078 * junk and/or it has put us in CLOSE_WAIT 1079 */ 1080 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1081 { 1082 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); 1083 xprt_force_disconnect(xprt); 1084 } 1085 1086 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 1087 { 1088 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1089 size_t len, used; 1090 char *p; 1091 1092 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; 1093 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; 1094 used = xdr_skb_read_bits(desc, p, len); 1095 transport->tcp_offset += used; 1096 if (used != len) 1097 return; 1098 1099 transport->tcp_reclen = ntohl(transport->tcp_fraghdr); 1100 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) 1101 transport->tcp_flags |= TCP_RCV_LAST_FRAG; 1102 else 1103 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; 1104 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; 1105 1106 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; 1107 transport->tcp_offset = 0; 1108 1109 /* Sanity check of the record length */ 1110 if (unlikely(transport->tcp_reclen < 8)) { 1111 dprintk("RPC: invalid TCP record fragment length\n"); 1112 xs_tcp_force_close(xprt); 1113 return; 1114 } 1115 dprintk("RPC: reading TCP record fragment of length %d\n", 1116 transport->tcp_reclen); 1117 } 1118 1119 static void xs_tcp_check_fraghdr(struct sock_xprt *transport) 1120 { 1121 if (transport->tcp_offset == transport->tcp_reclen) { 1122 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; 1123 transport->tcp_offset = 0; 1124 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { 1125 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1126 transport->tcp_flags |= TCP_RCV_COPY_XID; 1127 transport->tcp_copied = 0; 1128 } 1129 } 1130 } 1131 1132 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1133 { 1134 size_t len, used; 1135 char *p; 1136 1137 len = sizeof(transport->tcp_xid) - transport->tcp_offset; 1138 dprintk("RPC: reading XID (%Zu bytes)\n", len); 1139 p = ((char *) &transport->tcp_xid) + transport->tcp_offset; 1140 used = xdr_skb_read_bits(desc, p, len); 1141 transport->tcp_offset += used; 1142 if (used != len) 1143 return; 1144 transport->tcp_flags &= ~TCP_RCV_COPY_XID; 1145 transport->tcp_flags |= TCP_RCV_READ_CALLDIR; 1146 transport->tcp_copied = 4; 1147 dprintk("RPC: reading %s XID %08x\n", 1148 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for" 1149 : "request with", 1150 ntohl(transport->tcp_xid)); 1151 xs_tcp_check_fraghdr(transport); 1152 } 1153 1154 static inline void xs_tcp_read_calldir(struct sock_xprt *transport, 1155 struct xdr_skb_reader *desc) 1156 { 1157 size_t len, used; 1158 u32 offset; 1159 char *p; 1160 1161 /* 1162 * We want transport->tcp_offset to be 8 at the end of this routine 1163 * (4 bytes for the xid and 4 bytes for the call/reply flag). 1164 * When this function is called for the first time, 1165 * transport->tcp_offset is 4 (after having already read the xid). 1166 */ 1167 offset = transport->tcp_offset - sizeof(transport->tcp_xid); 1168 len = sizeof(transport->tcp_calldir) - offset; 1169 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len); 1170 p = ((char *) &transport->tcp_calldir) + offset; 1171 used = xdr_skb_read_bits(desc, p, len); 1172 transport->tcp_offset += used; 1173 if (used != len) 1174 return; 1175 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR; 1176 /* 1177 * We don't yet have the XDR buffer, so we will write the calldir 1178 * out after we get the buffer from the 'struct rpc_rqst' 1179 */ 1180 switch (ntohl(transport->tcp_calldir)) { 1181 case RPC_REPLY: 1182 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1183 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1184 transport->tcp_flags |= TCP_RPC_REPLY; 1185 break; 1186 case RPC_CALL: 1187 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1188 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1189 transport->tcp_flags &= ~TCP_RPC_REPLY; 1190 break; 1191 default: 1192 dprintk("RPC: invalid request message type\n"); 1193 xs_tcp_force_close(&transport->xprt); 1194 } 1195 xs_tcp_check_fraghdr(transport); 1196 } 1197 1198 static inline void xs_tcp_read_common(struct rpc_xprt *xprt, 1199 struct xdr_skb_reader *desc, 1200 struct rpc_rqst *req) 1201 { 1202 struct sock_xprt *transport = 1203 container_of(xprt, struct sock_xprt, xprt); 1204 struct xdr_buf *rcvbuf; 1205 size_t len; 1206 ssize_t r; 1207 1208 rcvbuf = &req->rq_private_buf; 1209 1210 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) { 1211 /* 1212 * Save the RPC direction in the XDR buffer 1213 */ 1214 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied, 1215 &transport->tcp_calldir, 1216 sizeof(transport->tcp_calldir)); 1217 transport->tcp_copied += sizeof(transport->tcp_calldir); 1218 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR; 1219 } 1220 1221 len = desc->count; 1222 if (len > transport->tcp_reclen - transport->tcp_offset) { 1223 struct xdr_skb_reader my_desc; 1224 1225 len = transport->tcp_reclen - transport->tcp_offset; 1226 memcpy(&my_desc, desc, sizeof(my_desc)); 1227 my_desc.count = len; 1228 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1229 &my_desc, xdr_skb_read_bits); 1230 desc->count -= r; 1231 desc->offset += r; 1232 } else 1233 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1234 desc, xdr_skb_read_bits); 1235 1236 if (r > 0) { 1237 transport->tcp_copied += r; 1238 transport->tcp_offset += r; 1239 } 1240 if (r != len) { 1241 /* Error when copying to the receive buffer, 1242 * usually because we weren't able to allocate 1243 * additional buffer pages. All we can do now 1244 * is turn off TCP_RCV_COPY_DATA, so the request 1245 * will not receive any additional updates, 1246 * and time out. 1247 * Any remaining data from this record will 1248 * be discarded. 1249 */ 1250 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1251 dprintk("RPC: XID %08x truncated request\n", 1252 ntohl(transport->tcp_xid)); 1253 dprintk("RPC: xprt = %p, tcp_copied = %lu, " 1254 "tcp_offset = %u, tcp_reclen = %u\n", 1255 xprt, transport->tcp_copied, 1256 transport->tcp_offset, transport->tcp_reclen); 1257 return; 1258 } 1259 1260 dprintk("RPC: XID %08x read %Zd bytes\n", 1261 ntohl(transport->tcp_xid), r); 1262 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " 1263 "tcp_reclen = %u\n", xprt, transport->tcp_copied, 1264 transport->tcp_offset, transport->tcp_reclen); 1265 1266 if (transport->tcp_copied == req->rq_private_buf.buflen) 1267 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1268 else if (transport->tcp_offset == transport->tcp_reclen) { 1269 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) 1270 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1271 } 1272 } 1273 1274 /* 1275 * Finds the request corresponding to the RPC xid and invokes the common 1276 * tcp read code to read the data. 1277 */ 1278 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt, 1279 struct xdr_skb_reader *desc) 1280 { 1281 struct sock_xprt *transport = 1282 container_of(xprt, struct sock_xprt, xprt); 1283 struct rpc_rqst *req; 1284 1285 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid)); 1286 1287 /* Find and lock the request corresponding to this xid */ 1288 spin_lock(&xprt->transport_lock); 1289 req = xprt_lookup_rqst(xprt, transport->tcp_xid); 1290 if (!req) { 1291 dprintk("RPC: XID %08x request not found!\n", 1292 ntohl(transport->tcp_xid)); 1293 spin_unlock(&xprt->transport_lock); 1294 return -1; 1295 } 1296 1297 xs_tcp_read_common(xprt, desc, req); 1298 1299 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1300 xprt_complete_rqst(req->rq_task, transport->tcp_copied); 1301 1302 spin_unlock(&xprt->transport_lock); 1303 return 0; 1304 } 1305 1306 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1307 /* 1308 * Obtains an rpc_rqst previously allocated and invokes the common 1309 * tcp read code to read the data. The result is placed in the callback 1310 * queue. 1311 * If we're unable to obtain the rpc_rqst we schedule the closing of the 1312 * connection and return -1. 1313 */ 1314 static int xs_tcp_read_callback(struct rpc_xprt *xprt, 1315 struct xdr_skb_reader *desc) 1316 { 1317 struct sock_xprt *transport = 1318 container_of(xprt, struct sock_xprt, xprt); 1319 struct rpc_rqst *req; 1320 1321 /* Look up and lock the request corresponding to the given XID */ 1322 spin_lock(&xprt->transport_lock); 1323 req = xprt_lookup_bc_request(xprt, transport->tcp_xid); 1324 if (req == NULL) { 1325 spin_unlock(&xprt->transport_lock); 1326 printk(KERN_WARNING "Callback slot table overflowed\n"); 1327 xprt_force_disconnect(xprt); 1328 return -1; 1329 } 1330 1331 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid)); 1332 xs_tcp_read_common(xprt, desc, req); 1333 1334 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1335 xprt_complete_bc_request(req, transport->tcp_copied); 1336 spin_unlock(&xprt->transport_lock); 1337 1338 return 0; 1339 } 1340 1341 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1342 struct xdr_skb_reader *desc) 1343 { 1344 struct sock_xprt *transport = 1345 container_of(xprt, struct sock_xprt, xprt); 1346 1347 return (transport->tcp_flags & TCP_RPC_REPLY) ? 1348 xs_tcp_read_reply(xprt, desc) : 1349 xs_tcp_read_callback(xprt, desc); 1350 } 1351 #else 1352 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1353 struct xdr_skb_reader *desc) 1354 { 1355 return xs_tcp_read_reply(xprt, desc); 1356 } 1357 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1358 1359 /* 1360 * Read data off the transport. This can be either an RPC_CALL or an 1361 * RPC_REPLY. Relay the processing to helper functions. 1362 */ 1363 static void xs_tcp_read_data(struct rpc_xprt *xprt, 1364 struct xdr_skb_reader *desc) 1365 { 1366 struct sock_xprt *transport = 1367 container_of(xprt, struct sock_xprt, xprt); 1368 1369 if (_xs_tcp_read_data(xprt, desc) == 0) 1370 xs_tcp_check_fraghdr(transport); 1371 else { 1372 /* 1373 * The transport_lock protects the request handling. 1374 * There's no need to hold it to update the tcp_flags. 1375 */ 1376 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1377 } 1378 } 1379 1380 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1381 { 1382 size_t len; 1383 1384 len = transport->tcp_reclen - transport->tcp_offset; 1385 if (len > desc->count) 1386 len = desc->count; 1387 desc->count -= len; 1388 desc->offset += len; 1389 transport->tcp_offset += len; 1390 dprintk("RPC: discarded %Zu bytes\n", len); 1391 xs_tcp_check_fraghdr(transport); 1392 } 1393 1394 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) 1395 { 1396 struct rpc_xprt *xprt = rd_desc->arg.data; 1397 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1398 struct xdr_skb_reader desc = { 1399 .skb = skb, 1400 .offset = offset, 1401 .count = len, 1402 }; 1403 1404 dprintk("RPC: xs_tcp_data_recv started\n"); 1405 do { 1406 /* Read in a new fragment marker if necessary */ 1407 /* Can we ever really expect to get completely empty fragments? */ 1408 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { 1409 xs_tcp_read_fraghdr(xprt, &desc); 1410 continue; 1411 } 1412 /* Read in the xid if necessary */ 1413 if (transport->tcp_flags & TCP_RCV_COPY_XID) { 1414 xs_tcp_read_xid(transport, &desc); 1415 continue; 1416 } 1417 /* Read in the call/reply flag */ 1418 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) { 1419 xs_tcp_read_calldir(transport, &desc); 1420 continue; 1421 } 1422 /* Read in the request data */ 1423 if (transport->tcp_flags & TCP_RCV_COPY_DATA) { 1424 xs_tcp_read_data(xprt, &desc); 1425 continue; 1426 } 1427 /* Skip over any trailing bytes on short reads */ 1428 xs_tcp_read_discard(transport, &desc); 1429 } while (desc.count); 1430 dprintk("RPC: xs_tcp_data_recv done\n"); 1431 return len - desc.count; 1432 } 1433 1434 /** 1435 * xs_tcp_data_ready - "data ready" callback for TCP sockets 1436 * @sk: socket with data to read 1437 * @bytes: how much data to read 1438 * 1439 */ 1440 static void xs_tcp_data_ready(struct sock *sk) 1441 { 1442 struct rpc_xprt *xprt; 1443 read_descriptor_t rd_desc; 1444 int read; 1445 1446 dprintk("RPC: xs_tcp_data_ready...\n"); 1447 1448 read_lock_bh(&sk->sk_callback_lock); 1449 if (!(xprt = xprt_from_sock(sk))) 1450 goto out; 1451 /* Any data means we had a useful conversation, so 1452 * the we don't need to delay the next reconnect 1453 */ 1454 if (xprt->reestablish_timeout) 1455 xprt->reestablish_timeout = 0; 1456 1457 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ 1458 rd_desc.arg.data = xprt; 1459 do { 1460 rd_desc.count = 65536; 1461 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); 1462 } while (read > 0); 1463 out: 1464 read_unlock_bh(&sk->sk_callback_lock); 1465 } 1466 1467 /* 1468 * Do the equivalent of linger/linger2 handling for dealing with 1469 * broken servers that don't close the socket in a timely 1470 * fashion 1471 */ 1472 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt, 1473 unsigned long timeout) 1474 { 1475 struct sock_xprt *transport; 1476 1477 if (xprt_test_and_set_connecting(xprt)) 1478 return; 1479 set_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1480 transport = container_of(xprt, struct sock_xprt, xprt); 1481 queue_delayed_work(rpciod_workqueue, &transport->connect_worker, 1482 timeout); 1483 } 1484 1485 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt) 1486 { 1487 struct sock_xprt *transport; 1488 1489 transport = container_of(xprt, struct sock_xprt, xprt); 1490 1491 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) || 1492 !cancel_delayed_work(&transport->connect_worker)) 1493 return; 1494 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1495 xprt_clear_connecting(xprt); 1496 } 1497 1498 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1499 { 1500 smp_mb__before_clear_bit(); 1501 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1502 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state); 1503 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1504 clear_bit(XPRT_CLOSING, &xprt->state); 1505 smp_mb__after_clear_bit(); 1506 } 1507 1508 static void xs_sock_mark_closed(struct rpc_xprt *xprt) 1509 { 1510 xs_sock_reset_connection_flags(xprt); 1511 /* Mark transport as closed and wake up all pending tasks */ 1512 xprt_disconnect_done(xprt); 1513 } 1514 1515 /** 1516 * xs_tcp_state_change - callback to handle TCP socket state changes 1517 * @sk: socket whose state has changed 1518 * 1519 */ 1520 static void xs_tcp_state_change(struct sock *sk) 1521 { 1522 struct rpc_xprt *xprt; 1523 1524 read_lock_bh(&sk->sk_callback_lock); 1525 if (!(xprt = xprt_from_sock(sk))) 1526 goto out; 1527 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1528 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1529 sk->sk_state, xprt_connected(xprt), 1530 sock_flag(sk, SOCK_DEAD), 1531 sock_flag(sk, SOCK_ZAPPED), 1532 sk->sk_shutdown); 1533 1534 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1535 switch (sk->sk_state) { 1536 case TCP_ESTABLISHED: 1537 spin_lock(&xprt->transport_lock); 1538 if (!xprt_test_and_set_connected(xprt)) { 1539 struct sock_xprt *transport = container_of(xprt, 1540 struct sock_xprt, xprt); 1541 1542 /* Reset TCP record info */ 1543 transport->tcp_offset = 0; 1544 transport->tcp_reclen = 0; 1545 transport->tcp_copied = 0; 1546 transport->tcp_flags = 1547 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; 1548 xprt->connect_cookie++; 1549 1550 xprt_wake_pending_tasks(xprt, -EAGAIN); 1551 } 1552 spin_unlock(&xprt->transport_lock); 1553 break; 1554 case TCP_FIN_WAIT1: 1555 /* The client initiated a shutdown of the socket */ 1556 xprt->connect_cookie++; 1557 xprt->reestablish_timeout = 0; 1558 set_bit(XPRT_CLOSING, &xprt->state); 1559 smp_mb__before_clear_bit(); 1560 clear_bit(XPRT_CONNECTED, &xprt->state); 1561 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1562 smp_mb__after_clear_bit(); 1563 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); 1564 break; 1565 case TCP_CLOSE_WAIT: 1566 /* The server initiated a shutdown of the socket */ 1567 xprt->connect_cookie++; 1568 clear_bit(XPRT_CONNECTED, &xprt->state); 1569 xs_tcp_force_close(xprt); 1570 case TCP_CLOSING: 1571 /* 1572 * If the server closed down the connection, make sure that 1573 * we back off before reconnecting 1574 */ 1575 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1576 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1577 break; 1578 case TCP_LAST_ACK: 1579 set_bit(XPRT_CLOSING, &xprt->state); 1580 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); 1581 smp_mb__before_clear_bit(); 1582 clear_bit(XPRT_CONNECTED, &xprt->state); 1583 smp_mb__after_clear_bit(); 1584 break; 1585 case TCP_CLOSE: 1586 xs_tcp_cancel_linger_timeout(xprt); 1587 xs_sock_mark_closed(xprt); 1588 } 1589 out: 1590 read_unlock_bh(&sk->sk_callback_lock); 1591 } 1592 1593 static void xs_write_space(struct sock *sk) 1594 { 1595 struct socket *sock; 1596 struct rpc_xprt *xprt; 1597 1598 if (unlikely(!(sock = sk->sk_socket))) 1599 return; 1600 clear_bit(SOCK_NOSPACE, &sock->flags); 1601 1602 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1603 return; 1604 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) 1605 return; 1606 1607 xprt_write_space(xprt); 1608 } 1609 1610 /** 1611 * xs_udp_write_space - callback invoked when socket buffer space 1612 * becomes available 1613 * @sk: socket whose state has changed 1614 * 1615 * Called when more output buffer space is available for this socket. 1616 * We try not to wake our writers until they can make "significant" 1617 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1618 * with a bunch of small requests. 1619 */ 1620 static void xs_udp_write_space(struct sock *sk) 1621 { 1622 read_lock_bh(&sk->sk_callback_lock); 1623 1624 /* from net/core/sock.c:sock_def_write_space */ 1625 if (sock_writeable(sk)) 1626 xs_write_space(sk); 1627 1628 read_unlock_bh(&sk->sk_callback_lock); 1629 } 1630 1631 /** 1632 * xs_tcp_write_space - callback invoked when socket buffer space 1633 * becomes available 1634 * @sk: socket whose state has changed 1635 * 1636 * Called when more output buffer space is available for this socket. 1637 * We try not to wake our writers until they can make "significant" 1638 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1639 * with a bunch of small requests. 1640 */ 1641 static void xs_tcp_write_space(struct sock *sk) 1642 { 1643 read_lock_bh(&sk->sk_callback_lock); 1644 1645 /* from net/core/stream.c:sk_stream_write_space */ 1646 if (sk_stream_is_writeable(sk)) 1647 xs_write_space(sk); 1648 1649 read_unlock_bh(&sk->sk_callback_lock); 1650 } 1651 1652 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1653 { 1654 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1655 struct sock *sk = transport->inet; 1656 1657 if (transport->rcvsize) { 1658 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1659 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1660 } 1661 if (transport->sndsize) { 1662 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1663 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1664 sk->sk_write_space(sk); 1665 } 1666 } 1667 1668 /** 1669 * xs_udp_set_buffer_size - set send and receive limits 1670 * @xprt: generic transport 1671 * @sndsize: requested size of send buffer, in bytes 1672 * @rcvsize: requested size of receive buffer, in bytes 1673 * 1674 * Set socket send and receive buffer size limits. 1675 */ 1676 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1677 { 1678 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1679 1680 transport->sndsize = 0; 1681 if (sndsize) 1682 transport->sndsize = sndsize + 1024; 1683 transport->rcvsize = 0; 1684 if (rcvsize) 1685 transport->rcvsize = rcvsize + 1024; 1686 1687 xs_udp_do_set_buffer_size(xprt); 1688 } 1689 1690 /** 1691 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1692 * @task: task that timed out 1693 * 1694 * Adjust the congestion window after a retransmit timeout has occurred. 1695 */ 1696 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1697 { 1698 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1699 } 1700 1701 static unsigned short xs_get_random_port(void) 1702 { 1703 unsigned short range = xprt_max_resvport - xprt_min_resvport; 1704 unsigned short rand = (unsigned short) prandom_u32() % range; 1705 return rand + xprt_min_resvport; 1706 } 1707 1708 /** 1709 * xs_set_port - reset the port number in the remote endpoint address 1710 * @xprt: generic transport 1711 * @port: new port number 1712 * 1713 */ 1714 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1715 { 1716 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1717 1718 rpc_set_port(xs_addr(xprt), port); 1719 xs_update_peer_port(xprt); 1720 } 1721 1722 static unsigned short xs_get_srcport(struct sock_xprt *transport) 1723 { 1724 unsigned short port = transport->srcport; 1725 1726 if (port == 0 && transport->xprt.resvport) 1727 port = xs_get_random_port(); 1728 return port; 1729 } 1730 1731 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1732 { 1733 if (transport->srcport != 0) 1734 transport->srcport = 0; 1735 if (!transport->xprt.resvport) 1736 return 0; 1737 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1738 return xprt_max_resvport; 1739 return --port; 1740 } 1741 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1742 { 1743 struct sockaddr_storage myaddr; 1744 int err, nloop = 0; 1745 unsigned short port = xs_get_srcport(transport); 1746 unsigned short last; 1747 1748 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1749 do { 1750 rpc_set_port((struct sockaddr *)&myaddr, port); 1751 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1752 transport->xprt.addrlen); 1753 if (port == 0) 1754 break; 1755 if (err == 0) { 1756 transport->srcport = port; 1757 break; 1758 } 1759 last = port; 1760 port = xs_next_srcport(transport, port); 1761 if (port > last) 1762 nloop++; 1763 } while (err == -EADDRINUSE && nloop != 2); 1764 1765 if (myaddr.ss_family == AF_INET) 1766 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1767 &((struct sockaddr_in *)&myaddr)->sin_addr, 1768 port, err ? "failed" : "ok", err); 1769 else 1770 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1771 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1772 port, err ? "failed" : "ok", err); 1773 return err; 1774 } 1775 1776 /* 1777 * We don't support autobind on AF_LOCAL sockets 1778 */ 1779 static void xs_local_rpcbind(struct rpc_task *task) 1780 { 1781 rcu_read_lock(); 1782 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt)); 1783 rcu_read_unlock(); 1784 } 1785 1786 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1787 { 1788 } 1789 1790 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1791 static struct lock_class_key xs_key[2]; 1792 static struct lock_class_key xs_slock_key[2]; 1793 1794 static inline void xs_reclassify_socketu(struct socket *sock) 1795 { 1796 struct sock *sk = sock->sk; 1797 1798 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1799 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); 1800 } 1801 1802 static inline void xs_reclassify_socket4(struct socket *sock) 1803 { 1804 struct sock *sk = sock->sk; 1805 1806 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1807 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1808 } 1809 1810 static inline void xs_reclassify_socket6(struct socket *sock) 1811 { 1812 struct sock *sk = sock->sk; 1813 1814 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1815 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1816 } 1817 1818 static inline void xs_reclassify_socket(int family, struct socket *sock) 1819 { 1820 WARN_ON_ONCE(sock_owned_by_user(sock->sk)); 1821 if (sock_owned_by_user(sock->sk)) 1822 return; 1823 1824 switch (family) { 1825 case AF_LOCAL: 1826 xs_reclassify_socketu(sock); 1827 break; 1828 case AF_INET: 1829 xs_reclassify_socket4(sock); 1830 break; 1831 case AF_INET6: 1832 xs_reclassify_socket6(sock); 1833 break; 1834 } 1835 } 1836 #else 1837 static inline void xs_reclassify_socketu(struct socket *sock) 1838 { 1839 } 1840 1841 static inline void xs_reclassify_socket4(struct socket *sock) 1842 { 1843 } 1844 1845 static inline void xs_reclassify_socket6(struct socket *sock) 1846 { 1847 } 1848 1849 static inline void xs_reclassify_socket(int family, struct socket *sock) 1850 { 1851 } 1852 #endif 1853 1854 static void xs_dummy_setup_socket(struct work_struct *work) 1855 { 1856 } 1857 1858 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1859 struct sock_xprt *transport, int family, int type, int protocol) 1860 { 1861 struct socket *sock; 1862 int err; 1863 1864 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1865 if (err < 0) { 1866 dprintk("RPC: can't create %d transport socket (%d).\n", 1867 protocol, -err); 1868 goto out; 1869 } 1870 xs_reclassify_socket(family, sock); 1871 1872 err = xs_bind(transport, sock); 1873 if (err) { 1874 sock_release(sock); 1875 goto out; 1876 } 1877 1878 return sock; 1879 out: 1880 return ERR_PTR(err); 1881 } 1882 1883 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1884 struct socket *sock) 1885 { 1886 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1887 xprt); 1888 1889 if (!transport->inet) { 1890 struct sock *sk = sock->sk; 1891 1892 write_lock_bh(&sk->sk_callback_lock); 1893 1894 xs_save_old_callbacks(transport, sk); 1895 1896 sk->sk_user_data = xprt; 1897 sk->sk_data_ready = xs_local_data_ready; 1898 sk->sk_write_space = xs_udp_write_space; 1899 sk->sk_error_report = xs_error_report; 1900 sk->sk_allocation = GFP_ATOMIC; 1901 1902 xprt_clear_connected(xprt); 1903 1904 /* Reset to new socket */ 1905 transport->sock = sock; 1906 transport->inet = sk; 1907 1908 write_unlock_bh(&sk->sk_callback_lock); 1909 } 1910 1911 /* Tell the socket layer to start connecting... */ 1912 xprt->stat.connect_count++; 1913 xprt->stat.connect_start = jiffies; 1914 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1915 } 1916 1917 /** 1918 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1919 * @xprt: RPC transport to connect 1920 * @transport: socket transport to connect 1921 * @create_sock: function to create a socket of the correct type 1922 */ 1923 static int xs_local_setup_socket(struct sock_xprt *transport) 1924 { 1925 struct rpc_xprt *xprt = &transport->xprt; 1926 struct socket *sock; 1927 int status = -EIO; 1928 1929 current->flags |= PF_FSTRANS; 1930 1931 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1932 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1933 SOCK_STREAM, 0, &sock, 1); 1934 if (status < 0) { 1935 dprintk("RPC: can't create AF_LOCAL " 1936 "transport socket (%d).\n", -status); 1937 goto out; 1938 } 1939 xs_reclassify_socketu(sock); 1940 1941 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1942 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1943 1944 status = xs_local_finish_connecting(xprt, sock); 1945 trace_rpc_socket_connect(xprt, sock, status); 1946 switch (status) { 1947 case 0: 1948 dprintk("RPC: xprt %p connected to %s\n", 1949 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1950 xprt_set_connected(xprt); 1951 break; 1952 case -ENOENT: 1953 dprintk("RPC: xprt %p: socket %s does not exist\n", 1954 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1955 break; 1956 case -ECONNREFUSED: 1957 dprintk("RPC: xprt %p: connection refused for %s\n", 1958 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1959 break; 1960 default: 1961 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 1962 __func__, -status, 1963 xprt->address_strings[RPC_DISPLAY_ADDR]); 1964 } 1965 1966 out: 1967 xprt_clear_connecting(xprt); 1968 xprt_wake_pending_tasks(xprt, status); 1969 current->flags &= ~PF_FSTRANS; 1970 return status; 1971 } 1972 1973 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 1974 { 1975 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1976 int ret; 1977 1978 if (RPC_IS_ASYNC(task)) { 1979 /* 1980 * We want the AF_LOCAL connect to be resolved in the 1981 * filesystem namespace of the process making the rpc 1982 * call. Thus we connect synchronously. 1983 * 1984 * If we want to support asynchronous AF_LOCAL calls, 1985 * we'll need to figure out how to pass a namespace to 1986 * connect. 1987 */ 1988 rpc_exit(task, -ENOTCONN); 1989 return; 1990 } 1991 ret = xs_local_setup_socket(transport); 1992 if (ret && !RPC_IS_SOFTCONN(task)) 1993 msleep_interruptible(15000); 1994 } 1995 1996 #ifdef CONFIG_SUNRPC_SWAP 1997 static void xs_set_memalloc(struct rpc_xprt *xprt) 1998 { 1999 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2000 xprt); 2001 2002 if (xprt->swapper) 2003 sk_set_memalloc(transport->inet); 2004 } 2005 2006 /** 2007 * xs_swapper - Tag this transport as being used for swap. 2008 * @xprt: transport to tag 2009 * @enable: enable/disable 2010 * 2011 */ 2012 int xs_swapper(struct rpc_xprt *xprt, int enable) 2013 { 2014 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2015 xprt); 2016 int err = 0; 2017 2018 if (enable) { 2019 xprt->swapper++; 2020 xs_set_memalloc(xprt); 2021 } else if (xprt->swapper) { 2022 xprt->swapper--; 2023 sk_clear_memalloc(transport->inet); 2024 } 2025 2026 return err; 2027 } 2028 EXPORT_SYMBOL_GPL(xs_swapper); 2029 #else 2030 static void xs_set_memalloc(struct rpc_xprt *xprt) 2031 { 2032 } 2033 #endif 2034 2035 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2036 { 2037 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2038 2039 if (!transport->inet) { 2040 struct sock *sk = sock->sk; 2041 2042 write_lock_bh(&sk->sk_callback_lock); 2043 2044 xs_save_old_callbacks(transport, sk); 2045 2046 sk->sk_user_data = xprt; 2047 sk->sk_data_ready = xs_udp_data_ready; 2048 sk->sk_write_space = xs_udp_write_space; 2049 sk->sk_no_check = UDP_CSUM_NORCV; 2050 sk->sk_allocation = GFP_ATOMIC; 2051 2052 xprt_set_connected(xprt); 2053 2054 /* Reset to new socket */ 2055 transport->sock = sock; 2056 transport->inet = sk; 2057 2058 xs_set_memalloc(xprt); 2059 2060 write_unlock_bh(&sk->sk_callback_lock); 2061 } 2062 xs_udp_do_set_buffer_size(xprt); 2063 } 2064 2065 static void xs_udp_setup_socket(struct work_struct *work) 2066 { 2067 struct sock_xprt *transport = 2068 container_of(work, struct sock_xprt, connect_worker.work); 2069 struct rpc_xprt *xprt = &transport->xprt; 2070 struct socket *sock = transport->sock; 2071 int status = -EIO; 2072 2073 current->flags |= PF_FSTRANS; 2074 2075 /* Start by resetting any existing state */ 2076 xs_reset_transport(transport); 2077 sock = xs_create_sock(xprt, transport, 2078 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP); 2079 if (IS_ERR(sock)) 2080 goto out; 2081 2082 dprintk("RPC: worker connecting xprt %p via %s to " 2083 "%s (port %s)\n", xprt, 2084 xprt->address_strings[RPC_DISPLAY_PROTO], 2085 xprt->address_strings[RPC_DISPLAY_ADDR], 2086 xprt->address_strings[RPC_DISPLAY_PORT]); 2087 2088 xs_udp_finish_connecting(xprt, sock); 2089 trace_rpc_socket_connect(xprt, sock, 0); 2090 status = 0; 2091 out: 2092 xprt_clear_connecting(xprt); 2093 xprt_wake_pending_tasks(xprt, status); 2094 current->flags &= ~PF_FSTRANS; 2095 } 2096 2097 /* 2098 * We need to preserve the port number so the reply cache on the server can 2099 * find our cached RPC replies when we get around to reconnecting. 2100 */ 2101 static void xs_abort_connection(struct sock_xprt *transport) 2102 { 2103 int result; 2104 struct sockaddr any; 2105 2106 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport); 2107 2108 /* 2109 * Disconnect the transport socket by doing a connect operation 2110 * with AF_UNSPEC. This should return immediately... 2111 */ 2112 memset(&any, 0, sizeof(any)); 2113 any.sa_family = AF_UNSPEC; 2114 result = kernel_connect(transport->sock, &any, sizeof(any), 0); 2115 trace_rpc_socket_reset_connection(&transport->xprt, 2116 transport->sock, result); 2117 if (!result) 2118 xs_sock_reset_connection_flags(&transport->xprt); 2119 dprintk("RPC: AF_UNSPEC connect return code %d\n", result); 2120 } 2121 2122 static void xs_tcp_reuse_connection(struct sock_xprt *transport) 2123 { 2124 unsigned int state = transport->inet->sk_state; 2125 2126 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) { 2127 /* we don't need to abort the connection if the socket 2128 * hasn't undergone a shutdown 2129 */ 2130 if (transport->inet->sk_shutdown == 0) 2131 return; 2132 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n", 2133 __func__, transport->inet->sk_shutdown); 2134 } 2135 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) { 2136 /* we don't need to abort the connection if the socket 2137 * hasn't undergone a shutdown 2138 */ 2139 if (transport->inet->sk_shutdown == 0) 2140 return; 2141 dprintk("RPC: %s: ESTABLISHED/SYN_SENT " 2142 "sk_shutdown set to %d\n", 2143 __func__, transport->inet->sk_shutdown); 2144 } 2145 xs_abort_connection(transport); 2146 } 2147 2148 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2149 { 2150 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2151 int ret = -ENOTCONN; 2152 2153 if (!transport->inet) { 2154 struct sock *sk = sock->sk; 2155 unsigned int keepidle = xprt->timeout->to_initval / HZ; 2156 unsigned int keepcnt = xprt->timeout->to_retries + 1; 2157 unsigned int opt_on = 1; 2158 2159 /* TCP Keepalive options */ 2160 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 2161 (char *)&opt_on, sizeof(opt_on)); 2162 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, 2163 (char *)&keepidle, sizeof(keepidle)); 2164 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, 2165 (char *)&keepidle, sizeof(keepidle)); 2166 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, 2167 (char *)&keepcnt, sizeof(keepcnt)); 2168 2169 write_lock_bh(&sk->sk_callback_lock); 2170 2171 xs_save_old_callbacks(transport, sk); 2172 2173 sk->sk_user_data = xprt; 2174 sk->sk_data_ready = xs_tcp_data_ready; 2175 sk->sk_state_change = xs_tcp_state_change; 2176 sk->sk_write_space = xs_tcp_write_space; 2177 sk->sk_error_report = xs_error_report; 2178 sk->sk_allocation = GFP_ATOMIC; 2179 2180 /* socket options */ 2181 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 2182 sock_reset_flag(sk, SOCK_LINGER); 2183 tcp_sk(sk)->linger2 = 0; 2184 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 2185 2186 xprt_clear_connected(xprt); 2187 2188 /* Reset to new socket */ 2189 transport->sock = sock; 2190 transport->inet = sk; 2191 2192 write_unlock_bh(&sk->sk_callback_lock); 2193 } 2194 2195 if (!xprt_bound(xprt)) 2196 goto out; 2197 2198 xs_set_memalloc(xprt); 2199 2200 /* Tell the socket layer to start connecting... */ 2201 xprt->stat.connect_count++; 2202 xprt->stat.connect_start = jiffies; 2203 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2204 switch (ret) { 2205 case 0: 2206 case -EINPROGRESS: 2207 /* SYN_SENT! */ 2208 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2209 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2210 } 2211 out: 2212 return ret; 2213 } 2214 2215 /** 2216 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2217 * @xprt: RPC transport to connect 2218 * @transport: socket transport to connect 2219 * @create_sock: function to create a socket of the correct type 2220 * 2221 * Invoked by a work queue tasklet. 2222 */ 2223 static void xs_tcp_setup_socket(struct work_struct *work) 2224 { 2225 struct sock_xprt *transport = 2226 container_of(work, struct sock_xprt, connect_worker.work); 2227 struct socket *sock = transport->sock; 2228 struct rpc_xprt *xprt = &transport->xprt; 2229 int status = -EIO; 2230 2231 current->flags |= PF_FSTRANS; 2232 2233 if (!sock) { 2234 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 2235 sock = xs_create_sock(xprt, transport, 2236 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP); 2237 if (IS_ERR(sock)) { 2238 status = PTR_ERR(sock); 2239 goto out; 2240 } 2241 } else { 2242 int abort_and_exit; 2243 2244 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT, 2245 &xprt->state); 2246 /* "close" the socket, preserving the local port */ 2247 xs_tcp_reuse_connection(transport); 2248 2249 if (abort_and_exit) 2250 goto out_eagain; 2251 } 2252 2253 dprintk("RPC: worker connecting xprt %p via %s to " 2254 "%s (port %s)\n", xprt, 2255 xprt->address_strings[RPC_DISPLAY_PROTO], 2256 xprt->address_strings[RPC_DISPLAY_ADDR], 2257 xprt->address_strings[RPC_DISPLAY_PORT]); 2258 2259 status = xs_tcp_finish_connecting(xprt, sock); 2260 trace_rpc_socket_connect(xprt, sock, status); 2261 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2262 xprt, -status, xprt_connected(xprt), 2263 sock->sk->sk_state); 2264 switch (status) { 2265 default: 2266 printk("%s: connect returned unhandled error %d\n", 2267 __func__, status); 2268 case -EADDRNOTAVAIL: 2269 /* We're probably in TIME_WAIT. Get rid of existing socket, 2270 * and retry 2271 */ 2272 xs_tcp_force_close(xprt); 2273 break; 2274 case 0: 2275 case -EINPROGRESS: 2276 case -EALREADY: 2277 xprt_clear_connecting(xprt); 2278 current->flags &= ~PF_FSTRANS; 2279 return; 2280 case -EINVAL: 2281 /* Happens, for instance, if the user specified a link 2282 * local IPv6 address without a scope-id. 2283 */ 2284 case -ECONNREFUSED: 2285 case -ECONNRESET: 2286 case -ENETUNREACH: 2287 /* retry with existing socket, after a delay */ 2288 goto out; 2289 } 2290 out_eagain: 2291 status = -EAGAIN; 2292 out: 2293 xprt_clear_connecting(xprt); 2294 xprt_wake_pending_tasks(xprt, status); 2295 current->flags &= ~PF_FSTRANS; 2296 } 2297 2298 /** 2299 * xs_connect - connect a socket to a remote endpoint 2300 * @xprt: pointer to transport structure 2301 * @task: address of RPC task that manages state of connect request 2302 * 2303 * TCP: If the remote end dropped the connection, delay reconnecting. 2304 * 2305 * UDP socket connects are synchronous, but we use a work queue anyway 2306 * to guarantee that even unprivileged user processes can set up a 2307 * socket on a privileged port. 2308 * 2309 * If a UDP socket connect fails, the delay behavior here prevents 2310 * retry floods (hard mounts). 2311 */ 2312 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2313 { 2314 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2315 2316 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) { 2317 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2318 "seconds\n", 2319 xprt, xprt->reestablish_timeout / HZ); 2320 queue_delayed_work(rpciod_workqueue, 2321 &transport->connect_worker, 2322 xprt->reestablish_timeout); 2323 xprt->reestablish_timeout <<= 1; 2324 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2325 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2326 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) 2327 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; 2328 } else { 2329 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2330 queue_delayed_work(rpciod_workqueue, 2331 &transport->connect_worker, 0); 2332 } 2333 } 2334 2335 /** 2336 * xs_local_print_stats - display AF_LOCAL socket-specifc stats 2337 * @xprt: rpc_xprt struct containing statistics 2338 * @seq: output file 2339 * 2340 */ 2341 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2342 { 2343 long idle_time = 0; 2344 2345 if (xprt_connected(xprt)) 2346 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2347 2348 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2349 "%llu %llu %lu %llu %llu\n", 2350 xprt->stat.bind_count, 2351 xprt->stat.connect_count, 2352 xprt->stat.connect_time, 2353 idle_time, 2354 xprt->stat.sends, 2355 xprt->stat.recvs, 2356 xprt->stat.bad_xids, 2357 xprt->stat.req_u, 2358 xprt->stat.bklog_u, 2359 xprt->stat.max_slots, 2360 xprt->stat.sending_u, 2361 xprt->stat.pending_u); 2362 } 2363 2364 /** 2365 * xs_udp_print_stats - display UDP socket-specifc stats 2366 * @xprt: rpc_xprt struct containing statistics 2367 * @seq: output file 2368 * 2369 */ 2370 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2371 { 2372 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2373 2374 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2375 "%lu %llu %llu\n", 2376 transport->srcport, 2377 xprt->stat.bind_count, 2378 xprt->stat.sends, 2379 xprt->stat.recvs, 2380 xprt->stat.bad_xids, 2381 xprt->stat.req_u, 2382 xprt->stat.bklog_u, 2383 xprt->stat.max_slots, 2384 xprt->stat.sending_u, 2385 xprt->stat.pending_u); 2386 } 2387 2388 /** 2389 * xs_tcp_print_stats - display TCP socket-specifc stats 2390 * @xprt: rpc_xprt struct containing statistics 2391 * @seq: output file 2392 * 2393 */ 2394 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2395 { 2396 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2397 long idle_time = 0; 2398 2399 if (xprt_connected(xprt)) 2400 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2401 2402 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2403 "%llu %llu %lu %llu %llu\n", 2404 transport->srcport, 2405 xprt->stat.bind_count, 2406 xprt->stat.connect_count, 2407 xprt->stat.connect_time, 2408 idle_time, 2409 xprt->stat.sends, 2410 xprt->stat.recvs, 2411 xprt->stat.bad_xids, 2412 xprt->stat.req_u, 2413 xprt->stat.bklog_u, 2414 xprt->stat.max_slots, 2415 xprt->stat.sending_u, 2416 xprt->stat.pending_u); 2417 } 2418 2419 /* 2420 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2421 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2422 * to use the server side send routines. 2423 */ 2424 static void *bc_malloc(struct rpc_task *task, size_t size) 2425 { 2426 struct page *page; 2427 struct rpc_buffer *buf; 2428 2429 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer)); 2430 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) 2431 return NULL; 2432 2433 page = alloc_page(GFP_KERNEL); 2434 if (!page) 2435 return NULL; 2436 2437 buf = page_address(page); 2438 buf->len = PAGE_SIZE; 2439 2440 return buf->data; 2441 } 2442 2443 /* 2444 * Free the space allocated in the bc_alloc routine 2445 */ 2446 static void bc_free(void *buffer) 2447 { 2448 struct rpc_buffer *buf; 2449 2450 if (!buffer) 2451 return; 2452 2453 buf = container_of(buffer, struct rpc_buffer, data); 2454 free_page((unsigned long)buf); 2455 } 2456 2457 /* 2458 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex 2459 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request. 2460 */ 2461 static int bc_sendto(struct rpc_rqst *req) 2462 { 2463 int len; 2464 struct xdr_buf *xbufp = &req->rq_snd_buf; 2465 struct rpc_xprt *xprt = req->rq_xprt; 2466 struct sock_xprt *transport = 2467 container_of(xprt, struct sock_xprt, xprt); 2468 struct socket *sock = transport->sock; 2469 unsigned long headoff; 2470 unsigned long tailoff; 2471 2472 xs_encode_stream_record_marker(xbufp); 2473 2474 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK; 2475 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK; 2476 len = svc_send_common(sock, xbufp, 2477 virt_to_page(xbufp->head[0].iov_base), headoff, 2478 xbufp->tail[0].iov_base, tailoff); 2479 2480 if (len != xbufp->len) { 2481 printk(KERN_NOTICE "Error sending entire callback!\n"); 2482 len = -EAGAIN; 2483 } 2484 2485 return len; 2486 } 2487 2488 /* 2489 * The send routine. Borrows from svc_send 2490 */ 2491 static int bc_send_request(struct rpc_task *task) 2492 { 2493 struct rpc_rqst *req = task->tk_rqstp; 2494 struct svc_xprt *xprt; 2495 u32 len; 2496 2497 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid)); 2498 /* 2499 * Get the server socket associated with this callback xprt 2500 */ 2501 xprt = req->rq_xprt->bc_xprt; 2502 2503 /* 2504 * Grab the mutex to serialize data as the connection is shared 2505 * with the fore channel 2506 */ 2507 if (!mutex_trylock(&xprt->xpt_mutex)) { 2508 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL); 2509 if (!mutex_trylock(&xprt->xpt_mutex)) 2510 return -EAGAIN; 2511 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task); 2512 } 2513 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2514 len = -ENOTCONN; 2515 else 2516 len = bc_sendto(req); 2517 mutex_unlock(&xprt->xpt_mutex); 2518 2519 if (len > 0) 2520 len = 0; 2521 2522 return len; 2523 } 2524 2525 /* 2526 * The close routine. Since this is client initiated, we do nothing 2527 */ 2528 2529 static void bc_close(struct rpc_xprt *xprt) 2530 { 2531 } 2532 2533 /* 2534 * The xprt destroy routine. Again, because this connection is client 2535 * initiated, we do nothing 2536 */ 2537 2538 static void bc_destroy(struct rpc_xprt *xprt) 2539 { 2540 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2541 2542 xs_xprt_free(xprt); 2543 module_put(THIS_MODULE); 2544 } 2545 2546 static struct rpc_xprt_ops xs_local_ops = { 2547 .reserve_xprt = xprt_reserve_xprt, 2548 .release_xprt = xs_tcp_release_xprt, 2549 .alloc_slot = xprt_alloc_slot, 2550 .rpcbind = xs_local_rpcbind, 2551 .set_port = xs_local_set_port, 2552 .connect = xs_local_connect, 2553 .buf_alloc = rpc_malloc, 2554 .buf_free = rpc_free, 2555 .send_request = xs_local_send_request, 2556 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2557 .close = xs_close, 2558 .destroy = xs_destroy, 2559 .print_stats = xs_local_print_stats, 2560 }; 2561 2562 static struct rpc_xprt_ops xs_udp_ops = { 2563 .set_buffer_size = xs_udp_set_buffer_size, 2564 .reserve_xprt = xprt_reserve_xprt_cong, 2565 .release_xprt = xprt_release_xprt_cong, 2566 .alloc_slot = xprt_alloc_slot, 2567 .rpcbind = rpcb_getport_async, 2568 .set_port = xs_set_port, 2569 .connect = xs_connect, 2570 .buf_alloc = rpc_malloc, 2571 .buf_free = rpc_free, 2572 .send_request = xs_udp_send_request, 2573 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 2574 .timer = xs_udp_timer, 2575 .release_request = xprt_release_rqst_cong, 2576 .close = xs_close, 2577 .destroy = xs_destroy, 2578 .print_stats = xs_udp_print_stats, 2579 }; 2580 2581 static struct rpc_xprt_ops xs_tcp_ops = { 2582 .reserve_xprt = xprt_reserve_xprt, 2583 .release_xprt = xs_tcp_release_xprt, 2584 .alloc_slot = xprt_lock_and_alloc_slot, 2585 .rpcbind = rpcb_getport_async, 2586 .set_port = xs_set_port, 2587 .connect = xs_connect, 2588 .buf_alloc = rpc_malloc, 2589 .buf_free = rpc_free, 2590 .send_request = xs_tcp_send_request, 2591 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2592 .close = xs_tcp_close, 2593 .destroy = xs_destroy, 2594 .print_stats = xs_tcp_print_stats, 2595 }; 2596 2597 /* 2598 * The rpc_xprt_ops for the server backchannel 2599 */ 2600 2601 static struct rpc_xprt_ops bc_tcp_ops = { 2602 .reserve_xprt = xprt_reserve_xprt, 2603 .release_xprt = xprt_release_xprt, 2604 .alloc_slot = xprt_alloc_slot, 2605 .buf_alloc = bc_malloc, 2606 .buf_free = bc_free, 2607 .send_request = bc_send_request, 2608 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2609 .close = bc_close, 2610 .destroy = bc_destroy, 2611 .print_stats = xs_tcp_print_stats, 2612 }; 2613 2614 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2615 { 2616 static const struct sockaddr_in sin = { 2617 .sin_family = AF_INET, 2618 .sin_addr.s_addr = htonl(INADDR_ANY), 2619 }; 2620 static const struct sockaddr_in6 sin6 = { 2621 .sin6_family = AF_INET6, 2622 .sin6_addr = IN6ADDR_ANY_INIT, 2623 }; 2624 2625 switch (family) { 2626 case AF_LOCAL: 2627 break; 2628 case AF_INET: 2629 memcpy(sap, &sin, sizeof(sin)); 2630 break; 2631 case AF_INET6: 2632 memcpy(sap, &sin6, sizeof(sin6)); 2633 break; 2634 default: 2635 dprintk("RPC: %s: Bad address family\n", __func__); 2636 return -EAFNOSUPPORT; 2637 } 2638 return 0; 2639 } 2640 2641 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2642 unsigned int slot_table_size, 2643 unsigned int max_slot_table_size) 2644 { 2645 struct rpc_xprt *xprt; 2646 struct sock_xprt *new; 2647 2648 if (args->addrlen > sizeof(xprt->addr)) { 2649 dprintk("RPC: xs_setup_xprt: address too large\n"); 2650 return ERR_PTR(-EBADF); 2651 } 2652 2653 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2654 max_slot_table_size); 2655 if (xprt == NULL) { 2656 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2657 "rpc_xprt\n"); 2658 return ERR_PTR(-ENOMEM); 2659 } 2660 2661 new = container_of(xprt, struct sock_xprt, xprt); 2662 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2663 xprt->addrlen = args->addrlen; 2664 if (args->srcaddr) 2665 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2666 else { 2667 int err; 2668 err = xs_init_anyaddr(args->dstaddr->sa_family, 2669 (struct sockaddr *)&new->srcaddr); 2670 if (err != 0) { 2671 xprt_free(xprt); 2672 return ERR_PTR(err); 2673 } 2674 } 2675 2676 return xprt; 2677 } 2678 2679 static const struct rpc_timeout xs_local_default_timeout = { 2680 .to_initval = 10 * HZ, 2681 .to_maxval = 10 * HZ, 2682 .to_retries = 2, 2683 }; 2684 2685 /** 2686 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2687 * @args: rpc transport creation arguments 2688 * 2689 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2690 */ 2691 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2692 { 2693 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2694 struct sock_xprt *transport; 2695 struct rpc_xprt *xprt; 2696 struct rpc_xprt *ret; 2697 2698 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2699 xprt_max_tcp_slot_table_entries); 2700 if (IS_ERR(xprt)) 2701 return xprt; 2702 transport = container_of(xprt, struct sock_xprt, xprt); 2703 2704 xprt->prot = 0; 2705 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2706 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2707 2708 xprt->bind_timeout = XS_BIND_TO; 2709 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2710 xprt->idle_timeout = XS_IDLE_DISC_TO; 2711 2712 xprt->ops = &xs_local_ops; 2713 xprt->timeout = &xs_local_default_timeout; 2714 2715 INIT_DELAYED_WORK(&transport->connect_worker, 2716 xs_dummy_setup_socket); 2717 2718 switch (sun->sun_family) { 2719 case AF_LOCAL: 2720 if (sun->sun_path[0] != '/') { 2721 dprintk("RPC: bad AF_LOCAL address: %s\n", 2722 sun->sun_path); 2723 ret = ERR_PTR(-EINVAL); 2724 goto out_err; 2725 } 2726 xprt_set_bound(xprt); 2727 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2728 ret = ERR_PTR(xs_local_setup_socket(transport)); 2729 if (ret) 2730 goto out_err; 2731 break; 2732 default: 2733 ret = ERR_PTR(-EAFNOSUPPORT); 2734 goto out_err; 2735 } 2736 2737 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2738 xprt->address_strings[RPC_DISPLAY_ADDR]); 2739 2740 if (try_module_get(THIS_MODULE)) 2741 return xprt; 2742 ret = ERR_PTR(-EINVAL); 2743 out_err: 2744 xs_xprt_free(xprt); 2745 return ret; 2746 } 2747 2748 static const struct rpc_timeout xs_udp_default_timeout = { 2749 .to_initval = 5 * HZ, 2750 .to_maxval = 30 * HZ, 2751 .to_increment = 5 * HZ, 2752 .to_retries = 5, 2753 }; 2754 2755 /** 2756 * xs_setup_udp - Set up transport to use a UDP socket 2757 * @args: rpc transport creation arguments 2758 * 2759 */ 2760 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2761 { 2762 struct sockaddr *addr = args->dstaddr; 2763 struct rpc_xprt *xprt; 2764 struct sock_xprt *transport; 2765 struct rpc_xprt *ret; 2766 2767 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2768 xprt_udp_slot_table_entries); 2769 if (IS_ERR(xprt)) 2770 return xprt; 2771 transport = container_of(xprt, struct sock_xprt, xprt); 2772 2773 xprt->prot = IPPROTO_UDP; 2774 xprt->tsh_size = 0; 2775 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2776 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2777 2778 xprt->bind_timeout = XS_BIND_TO; 2779 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2780 xprt->idle_timeout = XS_IDLE_DISC_TO; 2781 2782 xprt->ops = &xs_udp_ops; 2783 2784 xprt->timeout = &xs_udp_default_timeout; 2785 2786 switch (addr->sa_family) { 2787 case AF_INET: 2788 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2789 xprt_set_bound(xprt); 2790 2791 INIT_DELAYED_WORK(&transport->connect_worker, 2792 xs_udp_setup_socket); 2793 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2794 break; 2795 case AF_INET6: 2796 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2797 xprt_set_bound(xprt); 2798 2799 INIT_DELAYED_WORK(&transport->connect_worker, 2800 xs_udp_setup_socket); 2801 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2802 break; 2803 default: 2804 ret = ERR_PTR(-EAFNOSUPPORT); 2805 goto out_err; 2806 } 2807 2808 if (xprt_bound(xprt)) 2809 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2810 xprt->address_strings[RPC_DISPLAY_ADDR], 2811 xprt->address_strings[RPC_DISPLAY_PORT], 2812 xprt->address_strings[RPC_DISPLAY_PROTO]); 2813 else 2814 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2815 xprt->address_strings[RPC_DISPLAY_ADDR], 2816 xprt->address_strings[RPC_DISPLAY_PROTO]); 2817 2818 if (try_module_get(THIS_MODULE)) 2819 return xprt; 2820 ret = ERR_PTR(-EINVAL); 2821 out_err: 2822 xs_xprt_free(xprt); 2823 return ret; 2824 } 2825 2826 static const struct rpc_timeout xs_tcp_default_timeout = { 2827 .to_initval = 60 * HZ, 2828 .to_maxval = 60 * HZ, 2829 .to_retries = 2, 2830 }; 2831 2832 /** 2833 * xs_setup_tcp - Set up transport to use a TCP socket 2834 * @args: rpc transport creation arguments 2835 * 2836 */ 2837 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2838 { 2839 struct sockaddr *addr = args->dstaddr; 2840 struct rpc_xprt *xprt; 2841 struct sock_xprt *transport; 2842 struct rpc_xprt *ret; 2843 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2844 2845 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2846 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2847 2848 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2849 max_slot_table_size); 2850 if (IS_ERR(xprt)) 2851 return xprt; 2852 transport = container_of(xprt, struct sock_xprt, xprt); 2853 2854 xprt->prot = IPPROTO_TCP; 2855 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2856 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2857 2858 xprt->bind_timeout = XS_BIND_TO; 2859 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2860 xprt->idle_timeout = XS_IDLE_DISC_TO; 2861 2862 xprt->ops = &xs_tcp_ops; 2863 xprt->timeout = &xs_tcp_default_timeout; 2864 2865 switch (addr->sa_family) { 2866 case AF_INET: 2867 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2868 xprt_set_bound(xprt); 2869 2870 INIT_DELAYED_WORK(&transport->connect_worker, 2871 xs_tcp_setup_socket); 2872 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2873 break; 2874 case AF_INET6: 2875 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2876 xprt_set_bound(xprt); 2877 2878 INIT_DELAYED_WORK(&transport->connect_worker, 2879 xs_tcp_setup_socket); 2880 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 2881 break; 2882 default: 2883 ret = ERR_PTR(-EAFNOSUPPORT); 2884 goto out_err; 2885 } 2886 2887 if (xprt_bound(xprt)) 2888 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2889 xprt->address_strings[RPC_DISPLAY_ADDR], 2890 xprt->address_strings[RPC_DISPLAY_PORT], 2891 xprt->address_strings[RPC_DISPLAY_PROTO]); 2892 else 2893 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2894 xprt->address_strings[RPC_DISPLAY_ADDR], 2895 xprt->address_strings[RPC_DISPLAY_PROTO]); 2896 2897 if (try_module_get(THIS_MODULE)) 2898 return xprt; 2899 ret = ERR_PTR(-EINVAL); 2900 out_err: 2901 xs_xprt_free(xprt); 2902 return ret; 2903 } 2904 2905 /** 2906 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 2907 * @args: rpc transport creation arguments 2908 * 2909 */ 2910 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 2911 { 2912 struct sockaddr *addr = args->dstaddr; 2913 struct rpc_xprt *xprt; 2914 struct sock_xprt *transport; 2915 struct svc_sock *bc_sock; 2916 struct rpc_xprt *ret; 2917 2918 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2919 xprt_tcp_slot_table_entries); 2920 if (IS_ERR(xprt)) 2921 return xprt; 2922 transport = container_of(xprt, struct sock_xprt, xprt); 2923 2924 xprt->prot = IPPROTO_TCP; 2925 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2926 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2927 xprt->timeout = &xs_tcp_default_timeout; 2928 2929 /* backchannel */ 2930 xprt_set_bound(xprt); 2931 xprt->bind_timeout = 0; 2932 xprt->reestablish_timeout = 0; 2933 xprt->idle_timeout = 0; 2934 2935 xprt->ops = &bc_tcp_ops; 2936 2937 switch (addr->sa_family) { 2938 case AF_INET: 2939 xs_format_peer_addresses(xprt, "tcp", 2940 RPCBIND_NETID_TCP); 2941 break; 2942 case AF_INET6: 2943 xs_format_peer_addresses(xprt, "tcp", 2944 RPCBIND_NETID_TCP6); 2945 break; 2946 default: 2947 ret = ERR_PTR(-EAFNOSUPPORT); 2948 goto out_err; 2949 } 2950 2951 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2952 xprt->address_strings[RPC_DISPLAY_ADDR], 2953 xprt->address_strings[RPC_DISPLAY_PORT], 2954 xprt->address_strings[RPC_DISPLAY_PROTO]); 2955 2956 /* 2957 * Once we've associated a backchannel xprt with a connection, 2958 * we want to keep it around as long as the connection lasts, 2959 * in case we need to start using it for a backchannel again; 2960 * this reference won't be dropped until bc_xprt is destroyed. 2961 */ 2962 xprt_get(xprt); 2963 args->bc_xprt->xpt_bc_xprt = xprt; 2964 xprt->bc_xprt = args->bc_xprt; 2965 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 2966 transport->sock = bc_sock->sk_sock; 2967 transport->inet = bc_sock->sk_sk; 2968 2969 /* 2970 * Since we don't want connections for the backchannel, we set 2971 * the xprt status to connected 2972 */ 2973 xprt_set_connected(xprt); 2974 2975 if (try_module_get(THIS_MODULE)) 2976 return xprt; 2977 2978 args->bc_xprt->xpt_bc_xprt = NULL; 2979 xprt_put(xprt); 2980 ret = ERR_PTR(-EINVAL); 2981 out_err: 2982 xs_xprt_free(xprt); 2983 return ret; 2984 } 2985 2986 static struct xprt_class xs_local_transport = { 2987 .list = LIST_HEAD_INIT(xs_local_transport.list), 2988 .name = "named UNIX socket", 2989 .owner = THIS_MODULE, 2990 .ident = XPRT_TRANSPORT_LOCAL, 2991 .setup = xs_setup_local, 2992 }; 2993 2994 static struct xprt_class xs_udp_transport = { 2995 .list = LIST_HEAD_INIT(xs_udp_transport.list), 2996 .name = "udp", 2997 .owner = THIS_MODULE, 2998 .ident = XPRT_TRANSPORT_UDP, 2999 .setup = xs_setup_udp, 3000 }; 3001 3002 static struct xprt_class xs_tcp_transport = { 3003 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3004 .name = "tcp", 3005 .owner = THIS_MODULE, 3006 .ident = XPRT_TRANSPORT_TCP, 3007 .setup = xs_setup_tcp, 3008 }; 3009 3010 static struct xprt_class xs_bc_tcp_transport = { 3011 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3012 .name = "tcp NFSv4.1 backchannel", 3013 .owner = THIS_MODULE, 3014 .ident = XPRT_TRANSPORT_BC_TCP, 3015 .setup = xs_setup_bc_tcp, 3016 }; 3017 3018 /** 3019 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3020 * 3021 */ 3022 int init_socket_xprt(void) 3023 { 3024 #ifdef RPC_DEBUG 3025 if (!sunrpc_table_header) 3026 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3027 #endif 3028 3029 xprt_register_transport(&xs_local_transport); 3030 xprt_register_transport(&xs_udp_transport); 3031 xprt_register_transport(&xs_tcp_transport); 3032 xprt_register_transport(&xs_bc_tcp_transport); 3033 3034 return 0; 3035 } 3036 3037 /** 3038 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3039 * 3040 */ 3041 void cleanup_socket_xprt(void) 3042 { 3043 #ifdef RPC_DEBUG 3044 if (sunrpc_table_header) { 3045 unregister_sysctl_table(sunrpc_table_header); 3046 sunrpc_table_header = NULL; 3047 } 3048 #endif 3049 3050 xprt_unregister_transport(&xs_local_transport); 3051 xprt_unregister_transport(&xs_udp_transport); 3052 xprt_unregister_transport(&xs_tcp_transport); 3053 xprt_unregister_transport(&xs_bc_tcp_transport); 3054 } 3055 3056 static int param_set_uint_minmax(const char *val, 3057 const struct kernel_param *kp, 3058 unsigned int min, unsigned int max) 3059 { 3060 unsigned long num; 3061 int ret; 3062 3063 if (!val) 3064 return -EINVAL; 3065 ret = strict_strtoul(val, 0, &num); 3066 if (ret == -EINVAL || num < min || num > max) 3067 return -EINVAL; 3068 *((unsigned int *)kp->arg) = num; 3069 return 0; 3070 } 3071 3072 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3073 { 3074 return param_set_uint_minmax(val, kp, 3075 RPC_MIN_RESVPORT, 3076 RPC_MAX_RESVPORT); 3077 } 3078 3079 static struct kernel_param_ops param_ops_portnr = { 3080 .set = param_set_portnr, 3081 .get = param_get_uint, 3082 }; 3083 3084 #define param_check_portnr(name, p) \ 3085 __param_check(name, p, unsigned int); 3086 3087 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3088 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3089 3090 static int param_set_slot_table_size(const char *val, 3091 const struct kernel_param *kp) 3092 { 3093 return param_set_uint_minmax(val, kp, 3094 RPC_MIN_SLOT_TABLE, 3095 RPC_MAX_SLOT_TABLE); 3096 } 3097 3098 static struct kernel_param_ops param_ops_slot_table_size = { 3099 .set = param_set_slot_table_size, 3100 .get = param_get_uint, 3101 }; 3102 3103 #define param_check_slot_table_size(name, p) \ 3104 __param_check(name, p, unsigned int); 3105 3106 static int param_set_max_slot_table_size(const char *val, 3107 const struct kernel_param *kp) 3108 { 3109 return param_set_uint_minmax(val, kp, 3110 RPC_MIN_SLOT_TABLE, 3111 RPC_MAX_SLOT_TABLE_LIMIT); 3112 } 3113 3114 static struct kernel_param_ops param_ops_max_slot_table_size = { 3115 .set = param_set_max_slot_table_size, 3116 .get = param_get_uint, 3117 }; 3118 3119 #define param_check_max_slot_table_size(name, p) \ 3120 __param_check(name, p, unsigned int); 3121 3122 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3123 slot_table_size, 0644); 3124 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3125 max_slot_table_size, 0644); 3126 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3127 slot_table_size, 0644); 3128 3129