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/slab.h> 23 #include <linux/module.h> 24 #include <linux/capability.h> 25 #include <linux/pagemap.h> 26 #include <linux/errno.h> 27 #include <linux/socket.h> 28 #include <linux/in.h> 29 #include <linux/net.h> 30 #include <linux/mm.h> 31 #include <linux/udp.h> 32 #include <linux/tcp.h> 33 #include <linux/sunrpc/clnt.h> 34 #include <linux/sunrpc/sched.h> 35 #include <linux/sunrpc/xprtsock.h> 36 #include <linux/file.h> 37 38 #include <net/sock.h> 39 #include <net/checksum.h> 40 #include <net/udp.h> 41 #include <net/tcp.h> 42 43 /* 44 * xprtsock tunables 45 */ 46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE; 48 49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 51 52 #define XS_TCP_LINGER_TO (15U * HZ) 53 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 54 55 /* 56 * We can register our own files under /proc/sys/sunrpc by 57 * calling register_sysctl_table() again. The files in that 58 * directory become the union of all files registered there. 59 * 60 * We simply need to make sure that we don't collide with 61 * someone else's file names! 62 */ 63 64 #ifdef RPC_DEBUG 65 66 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 67 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 68 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 69 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 70 71 static struct ctl_table_header *sunrpc_table_header; 72 73 /* 74 * FIXME: changing the UDP slot table size should also resize the UDP 75 * socket buffers for existing UDP transports 76 */ 77 static ctl_table xs_tunables_table[] = { 78 { 79 .ctl_name = CTL_SLOTTABLE_UDP, 80 .procname = "udp_slot_table_entries", 81 .data = &xprt_udp_slot_table_entries, 82 .maxlen = sizeof(unsigned int), 83 .mode = 0644, 84 .proc_handler = &proc_dointvec_minmax, 85 .strategy = &sysctl_intvec, 86 .extra1 = &min_slot_table_size, 87 .extra2 = &max_slot_table_size 88 }, 89 { 90 .ctl_name = CTL_SLOTTABLE_TCP, 91 .procname = "tcp_slot_table_entries", 92 .data = &xprt_tcp_slot_table_entries, 93 .maxlen = sizeof(unsigned int), 94 .mode = 0644, 95 .proc_handler = &proc_dointvec_minmax, 96 .strategy = &sysctl_intvec, 97 .extra1 = &min_slot_table_size, 98 .extra2 = &max_slot_table_size 99 }, 100 { 101 .ctl_name = CTL_MIN_RESVPORT, 102 .procname = "min_resvport", 103 .data = &xprt_min_resvport, 104 .maxlen = sizeof(unsigned int), 105 .mode = 0644, 106 .proc_handler = &proc_dointvec_minmax, 107 .strategy = &sysctl_intvec, 108 .extra1 = &xprt_min_resvport_limit, 109 .extra2 = &xprt_max_resvport_limit 110 }, 111 { 112 .ctl_name = CTL_MAX_RESVPORT, 113 .procname = "max_resvport", 114 .data = &xprt_max_resvport, 115 .maxlen = sizeof(unsigned int), 116 .mode = 0644, 117 .proc_handler = &proc_dointvec_minmax, 118 .strategy = &sysctl_intvec, 119 .extra1 = &xprt_min_resvport_limit, 120 .extra2 = &xprt_max_resvport_limit 121 }, 122 { 123 .procname = "tcp_fin_timeout", 124 .data = &xs_tcp_fin_timeout, 125 .maxlen = sizeof(xs_tcp_fin_timeout), 126 .mode = 0644, 127 .proc_handler = &proc_dointvec_jiffies, 128 .strategy = sysctl_jiffies 129 }, 130 { 131 .ctl_name = 0, 132 }, 133 }; 134 135 static ctl_table sunrpc_table[] = { 136 { 137 .ctl_name = CTL_SUNRPC, 138 .procname = "sunrpc", 139 .mode = 0555, 140 .child = xs_tunables_table 141 }, 142 { 143 .ctl_name = 0, 144 }, 145 }; 146 147 #endif 148 149 /* 150 * Time out for an RPC UDP socket connect. UDP socket connects are 151 * synchronous, but we set a timeout anyway in case of resource 152 * exhaustion on the local host. 153 */ 154 #define XS_UDP_CONN_TO (5U * HZ) 155 156 /* 157 * Wait duration for an RPC TCP connection to be established. Solaris 158 * NFS over TCP uses 60 seconds, for example, which is in line with how 159 * long a server takes to reboot. 160 */ 161 #define XS_TCP_CONN_TO (60U * HZ) 162 163 /* 164 * Wait duration for a reply from the RPC portmapper. 165 */ 166 #define XS_BIND_TO (60U * HZ) 167 168 /* 169 * Delay if a UDP socket connect error occurs. This is most likely some 170 * kind of resource problem on the local host. 171 */ 172 #define XS_UDP_REEST_TO (2U * HZ) 173 174 /* 175 * The reestablish timeout allows clients to delay for a bit before attempting 176 * to reconnect to a server that just dropped our connection. 177 * 178 * We implement an exponential backoff when trying to reestablish a TCP 179 * transport connection with the server. Some servers like to drop a TCP 180 * connection when they are overworked, so we start with a short timeout and 181 * increase over time if the server is down or not responding. 182 */ 183 #define XS_TCP_INIT_REEST_TO (3U * HZ) 184 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ) 185 186 /* 187 * TCP idle timeout; client drops the transport socket if it is idle 188 * for this long. Note that we also timeout UDP sockets to prevent 189 * holding port numbers when there is no RPC traffic. 190 */ 191 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 192 193 #ifdef RPC_DEBUG 194 # undef RPC_DEBUG_DATA 195 # define RPCDBG_FACILITY RPCDBG_TRANS 196 #endif 197 198 #ifdef RPC_DEBUG_DATA 199 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 200 { 201 u8 *buf = (u8 *) packet; 202 int j; 203 204 dprintk("RPC: %s\n", msg); 205 for (j = 0; j < count && j < 128; j += 4) { 206 if (!(j & 31)) { 207 if (j) 208 dprintk("\n"); 209 dprintk("0x%04x ", j); 210 } 211 dprintk("%02x%02x%02x%02x ", 212 buf[j], buf[j+1], buf[j+2], buf[j+3]); 213 } 214 dprintk("\n"); 215 } 216 #else 217 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 218 { 219 /* NOP */ 220 } 221 #endif 222 223 struct sock_xprt { 224 struct rpc_xprt xprt; 225 226 /* 227 * Network layer 228 */ 229 struct socket * sock; 230 struct sock * inet; 231 232 /* 233 * State of TCP reply receive 234 */ 235 __be32 tcp_fraghdr, 236 tcp_xid; 237 238 u32 tcp_offset, 239 tcp_reclen; 240 241 unsigned long tcp_copied, 242 tcp_flags; 243 244 /* 245 * Connection of transports 246 */ 247 struct delayed_work connect_worker; 248 struct sockaddr_storage addr; 249 unsigned short port; 250 251 /* 252 * UDP socket buffer size parameters 253 */ 254 size_t rcvsize, 255 sndsize; 256 257 /* 258 * Saved socket callback addresses 259 */ 260 void (*old_data_ready)(struct sock *, int); 261 void (*old_state_change)(struct sock *); 262 void (*old_write_space)(struct sock *); 263 void (*old_error_report)(struct sock *); 264 }; 265 266 /* 267 * TCP receive state flags 268 */ 269 #define TCP_RCV_LAST_FRAG (1UL << 0) 270 #define TCP_RCV_COPY_FRAGHDR (1UL << 1) 271 #define TCP_RCV_COPY_XID (1UL << 2) 272 #define TCP_RCV_COPY_DATA (1UL << 3) 273 #define TCP_RCV_READ_CALLDIR (1UL << 4) 274 #define TCP_RCV_COPY_CALLDIR (1UL << 5) 275 276 /* 277 * TCP RPC flags 278 */ 279 #define TCP_RPC_REPLY (1UL << 6) 280 281 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 282 { 283 return (struct sockaddr *) &xprt->addr; 284 } 285 286 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 287 { 288 return (struct sockaddr_in *) &xprt->addr; 289 } 290 291 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 292 { 293 return (struct sockaddr_in6 *) &xprt->addr; 294 } 295 296 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt, 297 const char *protocol, 298 const char *netid) 299 { 300 struct sockaddr_in *addr = xs_addr_in(xprt); 301 char *buf; 302 303 buf = kzalloc(20, GFP_KERNEL); 304 if (buf) { 305 snprintf(buf, 20, "%pI4", &addr->sin_addr.s_addr); 306 } 307 xprt->address_strings[RPC_DISPLAY_ADDR] = buf; 308 309 buf = kzalloc(8, GFP_KERNEL); 310 if (buf) { 311 snprintf(buf, 8, "%u", 312 ntohs(addr->sin_port)); 313 } 314 xprt->address_strings[RPC_DISPLAY_PORT] = buf; 315 316 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 317 318 buf = kzalloc(48, GFP_KERNEL); 319 if (buf) { 320 snprintf(buf, 48, "addr=%pI4 port=%u proto=%s", 321 &addr->sin_addr.s_addr, 322 ntohs(addr->sin_port), 323 protocol); 324 } 325 xprt->address_strings[RPC_DISPLAY_ALL] = buf; 326 327 buf = kzalloc(10, GFP_KERNEL); 328 if (buf) { 329 snprintf(buf, 10, "%02x%02x%02x%02x", 330 NIPQUAD(addr->sin_addr.s_addr)); 331 } 332 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf; 333 334 buf = kzalloc(8, GFP_KERNEL); 335 if (buf) { 336 snprintf(buf, 8, "%4hx", 337 ntohs(addr->sin_port)); 338 } 339 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf; 340 341 buf = kzalloc(30, GFP_KERNEL); 342 if (buf) { 343 snprintf(buf, 30, "%pI4.%u.%u", 344 &addr->sin_addr.s_addr, 345 ntohs(addr->sin_port) >> 8, 346 ntohs(addr->sin_port) & 0xff); 347 } 348 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf; 349 350 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 351 } 352 353 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt, 354 const char *protocol, 355 const char *netid) 356 { 357 struct sockaddr_in6 *addr = xs_addr_in6(xprt); 358 char *buf; 359 360 buf = kzalloc(40, GFP_KERNEL); 361 if (buf) { 362 snprintf(buf, 40, "%pI6",&addr->sin6_addr); 363 } 364 xprt->address_strings[RPC_DISPLAY_ADDR] = buf; 365 366 buf = kzalloc(8, GFP_KERNEL); 367 if (buf) { 368 snprintf(buf, 8, "%u", 369 ntohs(addr->sin6_port)); 370 } 371 xprt->address_strings[RPC_DISPLAY_PORT] = buf; 372 373 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 374 375 buf = kzalloc(64, GFP_KERNEL); 376 if (buf) { 377 snprintf(buf, 64, "addr=%pI6 port=%u proto=%s", 378 &addr->sin6_addr, 379 ntohs(addr->sin6_port), 380 protocol); 381 } 382 xprt->address_strings[RPC_DISPLAY_ALL] = buf; 383 384 buf = kzalloc(36, GFP_KERNEL); 385 if (buf) 386 snprintf(buf, 36, "%pi6", &addr->sin6_addr); 387 388 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf; 389 390 buf = kzalloc(8, GFP_KERNEL); 391 if (buf) { 392 snprintf(buf, 8, "%4hx", 393 ntohs(addr->sin6_port)); 394 } 395 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf; 396 397 buf = kzalloc(50, GFP_KERNEL); 398 if (buf) { 399 snprintf(buf, 50, "%pI6.%u.%u", 400 &addr->sin6_addr, 401 ntohs(addr->sin6_port) >> 8, 402 ntohs(addr->sin6_port) & 0xff); 403 } 404 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf; 405 406 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 407 } 408 409 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 410 { 411 unsigned int i; 412 413 for (i = 0; i < RPC_DISPLAY_MAX; i++) 414 switch (i) { 415 case RPC_DISPLAY_PROTO: 416 case RPC_DISPLAY_NETID: 417 continue; 418 default: 419 kfree(xprt->address_strings[i]); 420 } 421 } 422 423 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 424 425 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more) 426 { 427 struct msghdr msg = { 428 .msg_name = addr, 429 .msg_namelen = addrlen, 430 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0), 431 }; 432 struct kvec iov = { 433 .iov_base = vec->iov_base + base, 434 .iov_len = vec->iov_len - base, 435 }; 436 437 if (iov.iov_len != 0) 438 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); 439 return kernel_sendmsg(sock, &msg, NULL, 0, 0); 440 } 441 442 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more) 443 { 444 struct page **ppage; 445 unsigned int remainder; 446 int err, sent = 0; 447 448 remainder = xdr->page_len - base; 449 base += xdr->page_base; 450 ppage = xdr->pages + (base >> PAGE_SHIFT); 451 base &= ~PAGE_MASK; 452 for(;;) { 453 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); 454 int flags = XS_SENDMSG_FLAGS; 455 456 remainder -= len; 457 if (remainder != 0 || more) 458 flags |= MSG_MORE; 459 err = sock->ops->sendpage(sock, *ppage, base, len, flags); 460 if (remainder == 0 || err != len) 461 break; 462 sent += err; 463 ppage++; 464 base = 0; 465 } 466 if (sent == 0) 467 return err; 468 if (err > 0) 469 sent += err; 470 return sent; 471 } 472 473 /** 474 * xs_sendpages - write pages directly to a socket 475 * @sock: socket to send on 476 * @addr: UDP only -- address of destination 477 * @addrlen: UDP only -- length of destination address 478 * @xdr: buffer containing this request 479 * @base: starting position in the buffer 480 * 481 */ 482 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base) 483 { 484 unsigned int remainder = xdr->len - base; 485 int err, sent = 0; 486 487 if (unlikely(!sock)) 488 return -ENOTSOCK; 489 490 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); 491 if (base != 0) { 492 addr = NULL; 493 addrlen = 0; 494 } 495 496 if (base < xdr->head[0].iov_len || addr != NULL) { 497 unsigned int len = xdr->head[0].iov_len - base; 498 remainder -= len; 499 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0); 500 if (remainder == 0 || err != len) 501 goto out; 502 sent += err; 503 base = 0; 504 } else 505 base -= xdr->head[0].iov_len; 506 507 if (base < xdr->page_len) { 508 unsigned int len = xdr->page_len - base; 509 remainder -= len; 510 err = xs_send_pagedata(sock, xdr, base, remainder != 0); 511 if (remainder == 0 || err != len) 512 goto out; 513 sent += err; 514 base = 0; 515 } else 516 base -= xdr->page_len; 517 518 if (base >= xdr->tail[0].iov_len) 519 return sent; 520 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0); 521 out: 522 if (sent == 0) 523 return err; 524 if (err > 0) 525 sent += err; 526 return sent; 527 } 528 529 static void xs_nospace_callback(struct rpc_task *task) 530 { 531 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt); 532 533 transport->inet->sk_write_pending--; 534 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 535 } 536 537 /** 538 * xs_nospace - place task on wait queue if transmit was incomplete 539 * @task: task to put to sleep 540 * 541 */ 542 static int xs_nospace(struct rpc_task *task) 543 { 544 struct rpc_rqst *req = task->tk_rqstp; 545 struct rpc_xprt *xprt = req->rq_xprt; 546 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 547 int ret = 0; 548 549 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", 550 task->tk_pid, req->rq_slen - req->rq_bytes_sent, 551 req->rq_slen); 552 553 /* Protect against races with write_space */ 554 spin_lock_bh(&xprt->transport_lock); 555 556 /* Don't race with disconnect */ 557 if (xprt_connected(xprt)) { 558 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) { 559 ret = -EAGAIN; 560 /* 561 * Notify TCP that we're limited by the application 562 * window size 563 */ 564 set_bit(SOCK_NOSPACE, &transport->sock->flags); 565 transport->inet->sk_write_pending++; 566 /* ...and wait for more buffer space */ 567 xprt_wait_for_buffer_space(task, xs_nospace_callback); 568 } 569 } else { 570 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 571 ret = -ENOTCONN; 572 } 573 574 spin_unlock_bh(&xprt->transport_lock); 575 return ret; 576 } 577 578 /** 579 * xs_udp_send_request - write an RPC request to a UDP socket 580 * @task: address of RPC task that manages the state of an RPC request 581 * 582 * Return values: 583 * 0: The request has been sent 584 * EAGAIN: The socket was blocked, please call again later to 585 * complete the request 586 * ENOTCONN: Caller needs to invoke connect logic then call again 587 * other: Some other error occured, the request was not sent 588 */ 589 static int xs_udp_send_request(struct rpc_task *task) 590 { 591 struct rpc_rqst *req = task->tk_rqstp; 592 struct rpc_xprt *xprt = req->rq_xprt; 593 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 594 struct xdr_buf *xdr = &req->rq_snd_buf; 595 int status; 596 597 xs_pktdump("packet data:", 598 req->rq_svec->iov_base, 599 req->rq_svec->iov_len); 600 601 if (!xprt_bound(xprt)) 602 return -ENOTCONN; 603 status = xs_sendpages(transport->sock, 604 xs_addr(xprt), 605 xprt->addrlen, xdr, 606 req->rq_bytes_sent); 607 608 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 609 xdr->len - req->rq_bytes_sent, status); 610 611 if (status >= 0) { 612 task->tk_bytes_sent += status; 613 if (status >= req->rq_slen) 614 return 0; 615 /* Still some bytes left; set up for a retry later. */ 616 status = -EAGAIN; 617 } 618 if (!transport->sock) 619 goto out; 620 621 switch (status) { 622 case -ENOTSOCK: 623 status = -ENOTCONN; 624 /* Should we call xs_close() here? */ 625 break; 626 case -EAGAIN: 627 status = xs_nospace(task); 628 break; 629 default: 630 dprintk("RPC: sendmsg returned unrecognized error %d\n", 631 -status); 632 case -ENETUNREACH: 633 case -EPIPE: 634 case -ECONNREFUSED: 635 /* When the server has died, an ICMP port unreachable message 636 * prompts ECONNREFUSED. */ 637 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 638 } 639 out: 640 return status; 641 } 642 643 /** 644 * xs_tcp_shutdown - gracefully shut down a TCP socket 645 * @xprt: transport 646 * 647 * Initiates a graceful shutdown of the TCP socket by calling the 648 * equivalent of shutdown(SHUT_WR); 649 */ 650 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 651 { 652 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 653 struct socket *sock = transport->sock; 654 655 if (sock != NULL) 656 kernel_sock_shutdown(sock, SHUT_WR); 657 } 658 659 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf) 660 { 661 u32 reclen = buf->len - sizeof(rpc_fraghdr); 662 rpc_fraghdr *base = buf->head[0].iov_base; 663 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen); 664 } 665 666 /** 667 * xs_tcp_send_request - write an RPC request to a TCP socket 668 * @task: address of RPC task that manages the state of an RPC request 669 * 670 * Return values: 671 * 0: The request has been sent 672 * EAGAIN: The socket was blocked, please call again later to 673 * complete the request 674 * ENOTCONN: Caller needs to invoke connect logic then call again 675 * other: Some other error occured, the request was not sent 676 * 677 * XXX: In the case of soft timeouts, should we eventually give up 678 * if sendmsg is not able to make progress? 679 */ 680 static int xs_tcp_send_request(struct rpc_task *task) 681 { 682 struct rpc_rqst *req = task->tk_rqstp; 683 struct rpc_xprt *xprt = req->rq_xprt; 684 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 685 struct xdr_buf *xdr = &req->rq_snd_buf; 686 int status; 687 688 xs_encode_tcp_record_marker(&req->rq_snd_buf); 689 690 xs_pktdump("packet data:", 691 req->rq_svec->iov_base, 692 req->rq_svec->iov_len); 693 694 /* Continue transmitting the packet/record. We must be careful 695 * to cope with writespace callbacks arriving _after_ we have 696 * called sendmsg(). */ 697 while (1) { 698 status = xs_sendpages(transport->sock, 699 NULL, 0, xdr, req->rq_bytes_sent); 700 701 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 702 xdr->len - req->rq_bytes_sent, status); 703 704 if (unlikely(status < 0)) 705 break; 706 707 /* If we've sent the entire packet, immediately 708 * reset the count of bytes sent. */ 709 req->rq_bytes_sent += status; 710 task->tk_bytes_sent += status; 711 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 712 req->rq_bytes_sent = 0; 713 return 0; 714 } 715 716 if (status != 0) 717 continue; 718 status = -EAGAIN; 719 break; 720 } 721 if (!transport->sock) 722 goto out; 723 724 switch (status) { 725 case -ENOTSOCK: 726 status = -ENOTCONN; 727 /* Should we call xs_close() here? */ 728 break; 729 case -EAGAIN: 730 status = xs_nospace(task); 731 break; 732 default: 733 dprintk("RPC: sendmsg returned unrecognized error %d\n", 734 -status); 735 case -ECONNRESET: 736 case -EPIPE: 737 xs_tcp_shutdown(xprt); 738 case -ECONNREFUSED: 739 case -ENOTCONN: 740 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 741 } 742 out: 743 return status; 744 } 745 746 /** 747 * xs_tcp_release_xprt - clean up after a tcp transmission 748 * @xprt: transport 749 * @task: rpc task 750 * 751 * This cleans up if an error causes us to abort the transmission of a request. 752 * In this case, the socket may need to be reset in order to avoid confusing 753 * the server. 754 */ 755 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 756 { 757 struct rpc_rqst *req; 758 759 if (task != xprt->snd_task) 760 return; 761 if (task == NULL) 762 goto out_release; 763 req = task->tk_rqstp; 764 if (req->rq_bytes_sent == 0) 765 goto out_release; 766 if (req->rq_bytes_sent == req->rq_snd_buf.len) 767 goto out_release; 768 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state); 769 out_release: 770 xprt_release_xprt(xprt, task); 771 } 772 773 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 774 { 775 transport->old_data_ready = sk->sk_data_ready; 776 transport->old_state_change = sk->sk_state_change; 777 transport->old_write_space = sk->sk_write_space; 778 transport->old_error_report = sk->sk_error_report; 779 } 780 781 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 782 { 783 sk->sk_data_ready = transport->old_data_ready; 784 sk->sk_state_change = transport->old_state_change; 785 sk->sk_write_space = transport->old_write_space; 786 sk->sk_error_report = transport->old_error_report; 787 } 788 789 static void xs_reset_transport(struct sock_xprt *transport) 790 { 791 struct socket *sock = transport->sock; 792 struct sock *sk = transport->inet; 793 794 if (sk == NULL) 795 return; 796 797 write_lock_bh(&sk->sk_callback_lock); 798 transport->inet = NULL; 799 transport->sock = NULL; 800 801 sk->sk_user_data = NULL; 802 803 xs_restore_old_callbacks(transport, sk); 804 write_unlock_bh(&sk->sk_callback_lock); 805 806 sk->sk_no_check = 0; 807 808 sock_release(sock); 809 } 810 811 /** 812 * xs_close - close a socket 813 * @xprt: transport 814 * 815 * This is used when all requests are complete; ie, no DRC state remains 816 * on the server we want to save. 817 * 818 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 819 * xs_reset_transport() zeroing the socket from underneath a writer. 820 */ 821 static void xs_close(struct rpc_xprt *xprt) 822 { 823 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 824 825 dprintk("RPC: xs_close xprt %p\n", xprt); 826 827 xs_reset_transport(transport); 828 829 smp_mb__before_clear_bit(); 830 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 831 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 832 clear_bit(XPRT_CLOSING, &xprt->state); 833 smp_mb__after_clear_bit(); 834 xprt_disconnect_done(xprt); 835 } 836 837 static void xs_tcp_close(struct rpc_xprt *xprt) 838 { 839 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state)) 840 xs_close(xprt); 841 else 842 xs_tcp_shutdown(xprt); 843 } 844 845 /** 846 * xs_destroy - prepare to shutdown a transport 847 * @xprt: doomed transport 848 * 849 */ 850 static void xs_destroy(struct rpc_xprt *xprt) 851 { 852 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 853 854 dprintk("RPC: xs_destroy xprt %p\n", xprt); 855 856 cancel_rearming_delayed_work(&transport->connect_worker); 857 858 xs_close(xprt); 859 xs_free_peer_addresses(xprt); 860 kfree(xprt->slot); 861 kfree(xprt); 862 module_put(THIS_MODULE); 863 } 864 865 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 866 { 867 return (struct rpc_xprt *) sk->sk_user_data; 868 } 869 870 /** 871 * xs_udp_data_ready - "data ready" callback for UDP sockets 872 * @sk: socket with data to read 873 * @len: how much data to read 874 * 875 */ 876 static void xs_udp_data_ready(struct sock *sk, int len) 877 { 878 struct rpc_task *task; 879 struct rpc_xprt *xprt; 880 struct rpc_rqst *rovr; 881 struct sk_buff *skb; 882 int err, repsize, copied; 883 u32 _xid; 884 __be32 *xp; 885 886 read_lock(&sk->sk_callback_lock); 887 dprintk("RPC: xs_udp_data_ready...\n"); 888 if (!(xprt = xprt_from_sock(sk))) 889 goto out; 890 891 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) 892 goto out; 893 894 if (xprt->shutdown) 895 goto dropit; 896 897 repsize = skb->len - sizeof(struct udphdr); 898 if (repsize < 4) { 899 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 900 goto dropit; 901 } 902 903 /* Copy the XID from the skb... */ 904 xp = skb_header_pointer(skb, sizeof(struct udphdr), 905 sizeof(_xid), &_xid); 906 if (xp == NULL) 907 goto dropit; 908 909 /* Look up and lock the request corresponding to the given XID */ 910 spin_lock(&xprt->transport_lock); 911 rovr = xprt_lookup_rqst(xprt, *xp); 912 if (!rovr) 913 goto out_unlock; 914 task = rovr->rq_task; 915 916 if ((copied = rovr->rq_private_buf.buflen) > repsize) 917 copied = repsize; 918 919 /* Suck it into the iovec, verify checksum if not done by hw. */ 920 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 921 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS); 922 goto out_unlock; 923 } 924 925 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS); 926 927 /* Something worked... */ 928 dst_confirm(skb->dst); 929 930 xprt_adjust_cwnd(task, copied); 931 xprt_update_rtt(task); 932 xprt_complete_rqst(task, copied); 933 934 out_unlock: 935 spin_unlock(&xprt->transport_lock); 936 dropit: 937 skb_free_datagram(sk, skb); 938 out: 939 read_unlock(&sk->sk_callback_lock); 940 } 941 942 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 943 { 944 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 945 size_t len, used; 946 char *p; 947 948 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; 949 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; 950 used = xdr_skb_read_bits(desc, p, len); 951 transport->tcp_offset += used; 952 if (used != len) 953 return; 954 955 transport->tcp_reclen = ntohl(transport->tcp_fraghdr); 956 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) 957 transport->tcp_flags |= TCP_RCV_LAST_FRAG; 958 else 959 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; 960 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; 961 962 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; 963 transport->tcp_offset = 0; 964 965 /* Sanity check of the record length */ 966 if (unlikely(transport->tcp_reclen < 8)) { 967 dprintk("RPC: invalid TCP record fragment length\n"); 968 xprt_force_disconnect(xprt); 969 return; 970 } 971 dprintk("RPC: reading TCP record fragment of length %d\n", 972 transport->tcp_reclen); 973 } 974 975 static void xs_tcp_check_fraghdr(struct sock_xprt *transport) 976 { 977 if (transport->tcp_offset == transport->tcp_reclen) { 978 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; 979 transport->tcp_offset = 0; 980 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { 981 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 982 transport->tcp_flags |= TCP_RCV_COPY_XID; 983 transport->tcp_copied = 0; 984 } 985 } 986 } 987 988 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) 989 { 990 size_t len, used; 991 char *p; 992 993 len = sizeof(transport->tcp_xid) - transport->tcp_offset; 994 dprintk("RPC: reading XID (%Zu bytes)\n", len); 995 p = ((char *) &transport->tcp_xid) + transport->tcp_offset; 996 used = xdr_skb_read_bits(desc, p, len); 997 transport->tcp_offset += used; 998 if (used != len) 999 return; 1000 transport->tcp_flags &= ~TCP_RCV_COPY_XID; 1001 transport->tcp_flags |= TCP_RCV_READ_CALLDIR; 1002 transport->tcp_copied = 4; 1003 dprintk("RPC: reading %s XID %08x\n", 1004 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for" 1005 : "request with", 1006 ntohl(transport->tcp_xid)); 1007 xs_tcp_check_fraghdr(transport); 1008 } 1009 1010 static inline void xs_tcp_read_calldir(struct sock_xprt *transport, 1011 struct xdr_skb_reader *desc) 1012 { 1013 size_t len, used; 1014 u32 offset; 1015 __be32 calldir; 1016 1017 /* 1018 * We want transport->tcp_offset to be 8 at the end of this routine 1019 * (4 bytes for the xid and 4 bytes for the call/reply flag). 1020 * When this function is called for the first time, 1021 * transport->tcp_offset is 4 (after having already read the xid). 1022 */ 1023 offset = transport->tcp_offset - sizeof(transport->tcp_xid); 1024 len = sizeof(calldir) - offset; 1025 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len); 1026 used = xdr_skb_read_bits(desc, &calldir, len); 1027 transport->tcp_offset += used; 1028 if (used != len) 1029 return; 1030 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR; 1031 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1032 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1033 /* 1034 * We don't yet have the XDR buffer, so we will write the calldir 1035 * out after we get the buffer from the 'struct rpc_rqst' 1036 */ 1037 if (ntohl(calldir) == RPC_REPLY) 1038 transport->tcp_flags |= TCP_RPC_REPLY; 1039 else 1040 transport->tcp_flags &= ~TCP_RPC_REPLY; 1041 dprintk("RPC: reading %s CALL/REPLY flag %08x\n", 1042 (transport->tcp_flags & TCP_RPC_REPLY) ? 1043 "reply for" : "request with", calldir); 1044 xs_tcp_check_fraghdr(transport); 1045 } 1046 1047 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 1048 { 1049 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1050 struct rpc_rqst *req; 1051 struct xdr_buf *rcvbuf; 1052 size_t len; 1053 ssize_t r; 1054 1055 /* Find and lock the request corresponding to this xid */ 1056 spin_lock(&xprt->transport_lock); 1057 req = xprt_lookup_rqst(xprt, transport->tcp_xid); 1058 if (!req) { 1059 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1060 dprintk("RPC: XID %08x request not found!\n", 1061 ntohl(transport->tcp_xid)); 1062 spin_unlock(&xprt->transport_lock); 1063 return; 1064 } 1065 1066 rcvbuf = &req->rq_private_buf; 1067 1068 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) { 1069 /* 1070 * Save the RPC direction in the XDR buffer 1071 */ 1072 __be32 calldir = transport->tcp_flags & TCP_RPC_REPLY ? 1073 htonl(RPC_REPLY) : 0; 1074 1075 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied, 1076 &calldir, sizeof(calldir)); 1077 transport->tcp_copied += sizeof(calldir); 1078 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR; 1079 } 1080 1081 len = desc->count; 1082 if (len > transport->tcp_reclen - transport->tcp_offset) { 1083 struct xdr_skb_reader my_desc; 1084 1085 len = transport->tcp_reclen - transport->tcp_offset; 1086 memcpy(&my_desc, desc, sizeof(my_desc)); 1087 my_desc.count = len; 1088 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1089 &my_desc, xdr_skb_read_bits); 1090 desc->count -= r; 1091 desc->offset += r; 1092 } else 1093 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1094 desc, xdr_skb_read_bits); 1095 1096 if (r > 0) { 1097 transport->tcp_copied += r; 1098 transport->tcp_offset += r; 1099 } 1100 if (r != len) { 1101 /* Error when copying to the receive buffer, 1102 * usually because we weren't able to allocate 1103 * additional buffer pages. All we can do now 1104 * is turn off TCP_RCV_COPY_DATA, so the request 1105 * will not receive any additional updates, 1106 * and time out. 1107 * Any remaining data from this record will 1108 * be discarded. 1109 */ 1110 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1111 dprintk("RPC: XID %08x truncated request\n", 1112 ntohl(transport->tcp_xid)); 1113 dprintk("RPC: xprt = %p, tcp_copied = %lu, " 1114 "tcp_offset = %u, tcp_reclen = %u\n", 1115 xprt, transport->tcp_copied, 1116 transport->tcp_offset, transport->tcp_reclen); 1117 goto out; 1118 } 1119 1120 dprintk("RPC: XID %08x read %Zd bytes\n", 1121 ntohl(transport->tcp_xid), r); 1122 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " 1123 "tcp_reclen = %u\n", xprt, transport->tcp_copied, 1124 transport->tcp_offset, transport->tcp_reclen); 1125 1126 if (transport->tcp_copied == req->rq_private_buf.buflen) 1127 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1128 else if (transport->tcp_offset == transport->tcp_reclen) { 1129 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) 1130 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1131 } 1132 1133 out: 1134 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1135 xprt_complete_rqst(req->rq_task, transport->tcp_copied); 1136 spin_unlock(&xprt->transport_lock); 1137 xs_tcp_check_fraghdr(transport); 1138 } 1139 1140 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1141 { 1142 size_t len; 1143 1144 len = transport->tcp_reclen - transport->tcp_offset; 1145 if (len > desc->count) 1146 len = desc->count; 1147 desc->count -= len; 1148 desc->offset += len; 1149 transport->tcp_offset += len; 1150 dprintk("RPC: discarded %Zu bytes\n", len); 1151 xs_tcp_check_fraghdr(transport); 1152 } 1153 1154 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) 1155 { 1156 struct rpc_xprt *xprt = rd_desc->arg.data; 1157 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1158 struct xdr_skb_reader desc = { 1159 .skb = skb, 1160 .offset = offset, 1161 .count = len, 1162 }; 1163 1164 dprintk("RPC: xs_tcp_data_recv started\n"); 1165 do { 1166 /* Read in a new fragment marker if necessary */ 1167 /* Can we ever really expect to get completely empty fragments? */ 1168 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { 1169 xs_tcp_read_fraghdr(xprt, &desc); 1170 continue; 1171 } 1172 /* Read in the xid if necessary */ 1173 if (transport->tcp_flags & TCP_RCV_COPY_XID) { 1174 xs_tcp_read_xid(transport, &desc); 1175 continue; 1176 } 1177 /* Read in the call/reply flag */ 1178 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) { 1179 xs_tcp_read_calldir(transport, &desc); 1180 continue; 1181 } 1182 /* Read in the request data */ 1183 if (transport->tcp_flags & TCP_RCV_COPY_DATA) { 1184 xs_tcp_read_request(xprt, &desc); 1185 continue; 1186 } 1187 /* Skip over any trailing bytes on short reads */ 1188 xs_tcp_read_discard(transport, &desc); 1189 } while (desc.count); 1190 dprintk("RPC: xs_tcp_data_recv done\n"); 1191 return len - desc.count; 1192 } 1193 1194 /** 1195 * xs_tcp_data_ready - "data ready" callback for TCP sockets 1196 * @sk: socket with data to read 1197 * @bytes: how much data to read 1198 * 1199 */ 1200 static void xs_tcp_data_ready(struct sock *sk, int bytes) 1201 { 1202 struct rpc_xprt *xprt; 1203 read_descriptor_t rd_desc; 1204 int read; 1205 1206 dprintk("RPC: xs_tcp_data_ready...\n"); 1207 1208 read_lock(&sk->sk_callback_lock); 1209 if (!(xprt = xprt_from_sock(sk))) 1210 goto out; 1211 if (xprt->shutdown) 1212 goto out; 1213 1214 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ 1215 rd_desc.arg.data = xprt; 1216 do { 1217 rd_desc.count = 65536; 1218 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); 1219 } while (read > 0); 1220 out: 1221 read_unlock(&sk->sk_callback_lock); 1222 } 1223 1224 /* 1225 * Do the equivalent of linger/linger2 handling for dealing with 1226 * broken servers that don't close the socket in a timely 1227 * fashion 1228 */ 1229 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt, 1230 unsigned long timeout) 1231 { 1232 struct sock_xprt *transport; 1233 1234 if (xprt_test_and_set_connecting(xprt)) 1235 return; 1236 set_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1237 transport = container_of(xprt, struct sock_xprt, xprt); 1238 queue_delayed_work(rpciod_workqueue, &transport->connect_worker, 1239 timeout); 1240 } 1241 1242 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt) 1243 { 1244 struct sock_xprt *transport; 1245 1246 transport = container_of(xprt, struct sock_xprt, xprt); 1247 1248 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) || 1249 !cancel_delayed_work(&transport->connect_worker)) 1250 return; 1251 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1252 xprt_clear_connecting(xprt); 1253 } 1254 1255 static void xs_sock_mark_closed(struct rpc_xprt *xprt) 1256 { 1257 smp_mb__before_clear_bit(); 1258 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1259 clear_bit(XPRT_CLOSING, &xprt->state); 1260 smp_mb__after_clear_bit(); 1261 /* Mark transport as closed and wake up all pending tasks */ 1262 xprt_disconnect_done(xprt); 1263 } 1264 1265 /** 1266 * xs_tcp_state_change - callback to handle TCP socket state changes 1267 * @sk: socket whose state has changed 1268 * 1269 */ 1270 static void xs_tcp_state_change(struct sock *sk) 1271 { 1272 struct rpc_xprt *xprt; 1273 1274 read_lock(&sk->sk_callback_lock); 1275 if (!(xprt = xprt_from_sock(sk))) 1276 goto out; 1277 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1278 dprintk("RPC: state %x conn %d dead %d zapped %d\n", 1279 sk->sk_state, xprt_connected(xprt), 1280 sock_flag(sk, SOCK_DEAD), 1281 sock_flag(sk, SOCK_ZAPPED)); 1282 1283 switch (sk->sk_state) { 1284 case TCP_ESTABLISHED: 1285 spin_lock_bh(&xprt->transport_lock); 1286 if (!xprt_test_and_set_connected(xprt)) { 1287 struct sock_xprt *transport = container_of(xprt, 1288 struct sock_xprt, xprt); 1289 1290 /* Reset TCP record info */ 1291 transport->tcp_offset = 0; 1292 transport->tcp_reclen = 0; 1293 transport->tcp_copied = 0; 1294 transport->tcp_flags = 1295 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; 1296 1297 xprt_wake_pending_tasks(xprt, -EAGAIN); 1298 } 1299 spin_unlock_bh(&xprt->transport_lock); 1300 break; 1301 case TCP_FIN_WAIT1: 1302 /* The client initiated a shutdown of the socket */ 1303 xprt->connect_cookie++; 1304 xprt->reestablish_timeout = 0; 1305 set_bit(XPRT_CLOSING, &xprt->state); 1306 smp_mb__before_clear_bit(); 1307 clear_bit(XPRT_CONNECTED, &xprt->state); 1308 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1309 smp_mb__after_clear_bit(); 1310 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); 1311 break; 1312 case TCP_CLOSE_WAIT: 1313 /* The server initiated a shutdown of the socket */ 1314 xprt_force_disconnect(xprt); 1315 case TCP_SYN_SENT: 1316 xprt->connect_cookie++; 1317 case TCP_CLOSING: 1318 /* 1319 * If the server closed down the connection, make sure that 1320 * we back off before reconnecting 1321 */ 1322 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1323 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1324 break; 1325 case TCP_LAST_ACK: 1326 set_bit(XPRT_CLOSING, &xprt->state); 1327 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); 1328 smp_mb__before_clear_bit(); 1329 clear_bit(XPRT_CONNECTED, &xprt->state); 1330 smp_mb__after_clear_bit(); 1331 break; 1332 case TCP_CLOSE: 1333 xs_tcp_cancel_linger_timeout(xprt); 1334 xs_sock_mark_closed(xprt); 1335 } 1336 out: 1337 read_unlock(&sk->sk_callback_lock); 1338 } 1339 1340 /** 1341 * xs_error_report - callback mainly for catching socket errors 1342 * @sk: socket 1343 */ 1344 static void xs_error_report(struct sock *sk) 1345 { 1346 struct rpc_xprt *xprt; 1347 1348 read_lock(&sk->sk_callback_lock); 1349 if (!(xprt = xprt_from_sock(sk))) 1350 goto out; 1351 dprintk("RPC: %s client %p...\n" 1352 "RPC: error %d\n", 1353 __func__, xprt, sk->sk_err); 1354 xprt_wake_pending_tasks(xprt, -EAGAIN); 1355 out: 1356 read_unlock(&sk->sk_callback_lock); 1357 } 1358 1359 static void xs_write_space(struct sock *sk) 1360 { 1361 struct socket *sock; 1362 struct rpc_xprt *xprt; 1363 1364 if (unlikely(!(sock = sk->sk_socket))) 1365 return; 1366 clear_bit(SOCK_NOSPACE, &sock->flags); 1367 1368 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1369 return; 1370 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) 1371 return; 1372 1373 xprt_write_space(xprt); 1374 } 1375 1376 /** 1377 * xs_udp_write_space - callback invoked when socket buffer space 1378 * becomes available 1379 * @sk: socket whose state has changed 1380 * 1381 * Called when more output buffer space is available for this socket. 1382 * We try not to wake our writers until they can make "significant" 1383 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1384 * with a bunch of small requests. 1385 */ 1386 static void xs_udp_write_space(struct sock *sk) 1387 { 1388 read_lock(&sk->sk_callback_lock); 1389 1390 /* from net/core/sock.c:sock_def_write_space */ 1391 if (sock_writeable(sk)) 1392 xs_write_space(sk); 1393 1394 read_unlock(&sk->sk_callback_lock); 1395 } 1396 1397 /** 1398 * xs_tcp_write_space - callback invoked when socket buffer space 1399 * becomes available 1400 * @sk: socket whose state has changed 1401 * 1402 * Called when more output buffer space is available for this socket. 1403 * We try not to wake our writers until they can make "significant" 1404 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1405 * with a bunch of small requests. 1406 */ 1407 static void xs_tcp_write_space(struct sock *sk) 1408 { 1409 read_lock(&sk->sk_callback_lock); 1410 1411 /* from net/core/stream.c:sk_stream_write_space */ 1412 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) 1413 xs_write_space(sk); 1414 1415 read_unlock(&sk->sk_callback_lock); 1416 } 1417 1418 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1419 { 1420 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1421 struct sock *sk = transport->inet; 1422 1423 if (transport->rcvsize) { 1424 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1425 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1426 } 1427 if (transport->sndsize) { 1428 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1429 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1430 sk->sk_write_space(sk); 1431 } 1432 } 1433 1434 /** 1435 * xs_udp_set_buffer_size - set send and receive limits 1436 * @xprt: generic transport 1437 * @sndsize: requested size of send buffer, in bytes 1438 * @rcvsize: requested size of receive buffer, in bytes 1439 * 1440 * Set socket send and receive buffer size limits. 1441 */ 1442 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1443 { 1444 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1445 1446 transport->sndsize = 0; 1447 if (sndsize) 1448 transport->sndsize = sndsize + 1024; 1449 transport->rcvsize = 0; 1450 if (rcvsize) 1451 transport->rcvsize = rcvsize + 1024; 1452 1453 xs_udp_do_set_buffer_size(xprt); 1454 } 1455 1456 /** 1457 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1458 * @task: task that timed out 1459 * 1460 * Adjust the congestion window after a retransmit timeout has occurred. 1461 */ 1462 static void xs_udp_timer(struct rpc_task *task) 1463 { 1464 xprt_adjust_cwnd(task, -ETIMEDOUT); 1465 } 1466 1467 static unsigned short xs_get_random_port(void) 1468 { 1469 unsigned short range = xprt_max_resvport - xprt_min_resvport; 1470 unsigned short rand = (unsigned short) net_random() % range; 1471 return rand + xprt_min_resvport; 1472 } 1473 1474 /** 1475 * xs_set_port - reset the port number in the remote endpoint address 1476 * @xprt: generic transport 1477 * @port: new port number 1478 * 1479 */ 1480 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1481 { 1482 struct sockaddr *addr = xs_addr(xprt); 1483 1484 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1485 1486 switch (addr->sa_family) { 1487 case AF_INET: 1488 ((struct sockaddr_in *)addr)->sin_port = htons(port); 1489 break; 1490 case AF_INET6: 1491 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port); 1492 break; 1493 default: 1494 BUG(); 1495 } 1496 } 1497 1498 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock) 1499 { 1500 unsigned short port = transport->port; 1501 1502 if (port == 0 && transport->xprt.resvport) 1503 port = xs_get_random_port(); 1504 return port; 1505 } 1506 1507 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port) 1508 { 1509 if (transport->port != 0) 1510 transport->port = 0; 1511 if (!transport->xprt.resvport) 1512 return 0; 1513 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1514 return xprt_max_resvport; 1515 return --port; 1516 } 1517 1518 static int xs_bind4(struct sock_xprt *transport, struct socket *sock) 1519 { 1520 struct sockaddr_in myaddr = { 1521 .sin_family = AF_INET, 1522 }; 1523 struct sockaddr_in *sa; 1524 int err, nloop = 0; 1525 unsigned short port = xs_get_srcport(transport, sock); 1526 unsigned short last; 1527 1528 sa = (struct sockaddr_in *)&transport->addr; 1529 myaddr.sin_addr = sa->sin_addr; 1530 do { 1531 myaddr.sin_port = htons(port); 1532 err = kernel_bind(sock, (struct sockaddr *) &myaddr, 1533 sizeof(myaddr)); 1534 if (port == 0) 1535 break; 1536 if (err == 0) { 1537 transport->port = port; 1538 break; 1539 } 1540 last = port; 1541 port = xs_next_srcport(transport, sock, port); 1542 if (port > last) 1543 nloop++; 1544 } while (err == -EADDRINUSE && nloop != 2); 1545 dprintk("RPC: %s %pI4:%u: %s (%d)\n", 1546 __func__, &myaddr.sin_addr, 1547 port, err ? "failed" : "ok", err); 1548 return err; 1549 } 1550 1551 static int xs_bind6(struct sock_xprt *transport, struct socket *sock) 1552 { 1553 struct sockaddr_in6 myaddr = { 1554 .sin6_family = AF_INET6, 1555 }; 1556 struct sockaddr_in6 *sa; 1557 int err, nloop = 0; 1558 unsigned short port = xs_get_srcport(transport, sock); 1559 unsigned short last; 1560 1561 sa = (struct sockaddr_in6 *)&transport->addr; 1562 myaddr.sin6_addr = sa->sin6_addr; 1563 do { 1564 myaddr.sin6_port = htons(port); 1565 err = kernel_bind(sock, (struct sockaddr *) &myaddr, 1566 sizeof(myaddr)); 1567 if (port == 0) 1568 break; 1569 if (err == 0) { 1570 transport->port = port; 1571 break; 1572 } 1573 last = port; 1574 port = xs_next_srcport(transport, sock, port); 1575 if (port > last) 1576 nloop++; 1577 } while (err == -EADDRINUSE && nloop != 2); 1578 dprintk("RPC: xs_bind6 %pI6:%u: %s (%d)\n", 1579 &myaddr.sin6_addr, port, err ? "failed" : "ok", err); 1580 return err; 1581 } 1582 1583 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1584 static struct lock_class_key xs_key[2]; 1585 static struct lock_class_key xs_slock_key[2]; 1586 1587 static inline void xs_reclassify_socket4(struct socket *sock) 1588 { 1589 struct sock *sk = sock->sk; 1590 1591 BUG_ON(sock_owned_by_user(sk)); 1592 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1593 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1594 } 1595 1596 static inline void xs_reclassify_socket6(struct socket *sock) 1597 { 1598 struct sock *sk = sock->sk; 1599 1600 BUG_ON(sock_owned_by_user(sk)); 1601 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1602 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1603 } 1604 #else 1605 static inline void xs_reclassify_socket4(struct socket *sock) 1606 { 1607 } 1608 1609 static inline void xs_reclassify_socket6(struct socket *sock) 1610 { 1611 } 1612 #endif 1613 1614 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 1615 { 1616 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1617 1618 if (!transport->inet) { 1619 struct sock *sk = sock->sk; 1620 1621 write_lock_bh(&sk->sk_callback_lock); 1622 1623 xs_save_old_callbacks(transport, sk); 1624 1625 sk->sk_user_data = xprt; 1626 sk->sk_data_ready = xs_udp_data_ready; 1627 sk->sk_write_space = xs_udp_write_space; 1628 sk->sk_error_report = xs_error_report; 1629 sk->sk_no_check = UDP_CSUM_NORCV; 1630 sk->sk_allocation = GFP_ATOMIC; 1631 1632 xprt_set_connected(xprt); 1633 1634 /* Reset to new socket */ 1635 transport->sock = sock; 1636 transport->inet = sk; 1637 1638 write_unlock_bh(&sk->sk_callback_lock); 1639 } 1640 xs_udp_do_set_buffer_size(xprt); 1641 } 1642 1643 /** 1644 * xs_udp_connect_worker4 - set up a UDP socket 1645 * @work: RPC transport to connect 1646 * 1647 * Invoked by a work queue tasklet. 1648 */ 1649 static void xs_udp_connect_worker4(struct work_struct *work) 1650 { 1651 struct sock_xprt *transport = 1652 container_of(work, struct sock_xprt, connect_worker.work); 1653 struct rpc_xprt *xprt = &transport->xprt; 1654 struct socket *sock = transport->sock; 1655 int err, status = -EIO; 1656 1657 if (xprt->shutdown) 1658 goto out; 1659 1660 /* Start by resetting any existing state */ 1661 xs_reset_transport(transport); 1662 1663 err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); 1664 if (err < 0) { 1665 dprintk("RPC: can't create UDP transport socket (%d).\n", -err); 1666 goto out; 1667 } 1668 xs_reclassify_socket4(sock); 1669 1670 if (xs_bind4(transport, sock)) { 1671 sock_release(sock); 1672 goto out; 1673 } 1674 1675 dprintk("RPC: worker connecting xprt %p to address: %s\n", 1676 xprt, xprt->address_strings[RPC_DISPLAY_ALL]); 1677 1678 xs_udp_finish_connecting(xprt, sock); 1679 status = 0; 1680 out: 1681 xprt_clear_connecting(xprt); 1682 xprt_wake_pending_tasks(xprt, status); 1683 } 1684 1685 /** 1686 * xs_udp_connect_worker6 - set up a UDP socket 1687 * @work: RPC transport to connect 1688 * 1689 * Invoked by a work queue tasklet. 1690 */ 1691 static void xs_udp_connect_worker6(struct work_struct *work) 1692 { 1693 struct sock_xprt *transport = 1694 container_of(work, struct sock_xprt, connect_worker.work); 1695 struct rpc_xprt *xprt = &transport->xprt; 1696 struct socket *sock = transport->sock; 1697 int err, status = -EIO; 1698 1699 if (xprt->shutdown) 1700 goto out; 1701 1702 /* Start by resetting any existing state */ 1703 xs_reset_transport(transport); 1704 1705 err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock); 1706 if (err < 0) { 1707 dprintk("RPC: can't create UDP transport socket (%d).\n", -err); 1708 goto out; 1709 } 1710 xs_reclassify_socket6(sock); 1711 1712 if (xs_bind6(transport, sock) < 0) { 1713 sock_release(sock); 1714 goto out; 1715 } 1716 1717 dprintk("RPC: worker connecting xprt %p to address: %s\n", 1718 xprt, xprt->address_strings[RPC_DISPLAY_ALL]); 1719 1720 xs_udp_finish_connecting(xprt, sock); 1721 status = 0; 1722 out: 1723 xprt_clear_connecting(xprt); 1724 xprt_wake_pending_tasks(xprt, status); 1725 } 1726 1727 /* 1728 * We need to preserve the port number so the reply cache on the server can 1729 * find our cached RPC replies when we get around to reconnecting. 1730 */ 1731 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport) 1732 { 1733 int result; 1734 struct sockaddr any; 1735 1736 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt); 1737 1738 /* 1739 * Disconnect the transport socket by doing a connect operation 1740 * with AF_UNSPEC. This should return immediately... 1741 */ 1742 memset(&any, 0, sizeof(any)); 1743 any.sa_family = AF_UNSPEC; 1744 result = kernel_connect(transport->sock, &any, sizeof(any), 0); 1745 if (!result) 1746 xs_sock_mark_closed(xprt); 1747 else 1748 dprintk("RPC: AF_UNSPEC connect return code %d\n", 1749 result); 1750 } 1751 1752 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport) 1753 { 1754 unsigned int state = transport->inet->sk_state; 1755 1756 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) 1757 return; 1758 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) 1759 return; 1760 xs_abort_connection(xprt, transport); 1761 } 1762 1763 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 1764 { 1765 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1766 1767 if (!transport->inet) { 1768 struct sock *sk = sock->sk; 1769 1770 write_lock_bh(&sk->sk_callback_lock); 1771 1772 xs_save_old_callbacks(transport, sk); 1773 1774 sk->sk_user_data = xprt; 1775 sk->sk_data_ready = xs_tcp_data_ready; 1776 sk->sk_state_change = xs_tcp_state_change; 1777 sk->sk_write_space = xs_tcp_write_space; 1778 sk->sk_error_report = xs_error_report; 1779 sk->sk_allocation = GFP_ATOMIC; 1780 1781 /* socket options */ 1782 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 1783 sock_reset_flag(sk, SOCK_LINGER); 1784 tcp_sk(sk)->linger2 = 0; 1785 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1786 1787 xprt_clear_connected(xprt); 1788 1789 /* Reset to new socket */ 1790 transport->sock = sock; 1791 transport->inet = sk; 1792 1793 write_unlock_bh(&sk->sk_callback_lock); 1794 } 1795 1796 if (!xprt_bound(xprt)) 1797 return -ENOTCONN; 1798 1799 /* Tell the socket layer to start connecting... */ 1800 xprt->stat.connect_count++; 1801 xprt->stat.connect_start = jiffies; 1802 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 1803 } 1804 1805 /** 1806 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 1807 * @xprt: RPC transport to connect 1808 * @transport: socket transport to connect 1809 * @create_sock: function to create a socket of the correct type 1810 * 1811 * Invoked by a work queue tasklet. 1812 */ 1813 static void xs_tcp_setup_socket(struct rpc_xprt *xprt, 1814 struct sock_xprt *transport, 1815 struct socket *(*create_sock)(struct rpc_xprt *, 1816 struct sock_xprt *)) 1817 { 1818 struct socket *sock = transport->sock; 1819 int status = -EIO; 1820 1821 if (xprt->shutdown) 1822 goto out; 1823 1824 if (!sock) { 1825 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1826 sock = create_sock(xprt, transport); 1827 if (IS_ERR(sock)) { 1828 status = PTR_ERR(sock); 1829 goto out; 1830 } 1831 } else { 1832 int abort_and_exit; 1833 1834 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT, 1835 &xprt->state); 1836 /* "close" the socket, preserving the local port */ 1837 xs_tcp_reuse_connection(xprt, transport); 1838 1839 if (abort_and_exit) 1840 goto out_eagain; 1841 } 1842 1843 dprintk("RPC: worker connecting xprt %p to address: %s\n", 1844 xprt, xprt->address_strings[RPC_DISPLAY_ALL]); 1845 1846 status = xs_tcp_finish_connecting(xprt, sock); 1847 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 1848 xprt, -status, xprt_connected(xprt), 1849 sock->sk->sk_state); 1850 switch (status) { 1851 default: 1852 printk("%s: connect returned unhandled error %d\n", 1853 __func__, status); 1854 case -EADDRNOTAVAIL: 1855 /* We're probably in TIME_WAIT. Get rid of existing socket, 1856 * and retry 1857 */ 1858 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); 1859 xprt_force_disconnect(xprt); 1860 case -ECONNREFUSED: 1861 case -ECONNRESET: 1862 case -ENETUNREACH: 1863 /* retry with existing socket, after a delay */ 1864 case 0: 1865 case -EINPROGRESS: 1866 case -EALREADY: 1867 xprt_clear_connecting(xprt); 1868 return; 1869 } 1870 out_eagain: 1871 status = -EAGAIN; 1872 out: 1873 xprt_clear_connecting(xprt); 1874 xprt_wake_pending_tasks(xprt, status); 1875 } 1876 1877 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt, 1878 struct sock_xprt *transport) 1879 { 1880 struct socket *sock; 1881 int err; 1882 1883 /* start from scratch */ 1884 err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock); 1885 if (err < 0) { 1886 dprintk("RPC: can't create TCP transport socket (%d).\n", 1887 -err); 1888 goto out_err; 1889 } 1890 xs_reclassify_socket4(sock); 1891 1892 if (xs_bind4(transport, sock) < 0) { 1893 sock_release(sock); 1894 goto out_err; 1895 } 1896 return sock; 1897 out_err: 1898 return ERR_PTR(-EIO); 1899 } 1900 1901 /** 1902 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint 1903 * @work: RPC transport to connect 1904 * 1905 * Invoked by a work queue tasklet. 1906 */ 1907 static void xs_tcp_connect_worker4(struct work_struct *work) 1908 { 1909 struct sock_xprt *transport = 1910 container_of(work, struct sock_xprt, connect_worker.work); 1911 struct rpc_xprt *xprt = &transport->xprt; 1912 1913 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4); 1914 } 1915 1916 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt, 1917 struct sock_xprt *transport) 1918 { 1919 struct socket *sock; 1920 int err; 1921 1922 /* start from scratch */ 1923 err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock); 1924 if (err < 0) { 1925 dprintk("RPC: can't create TCP transport socket (%d).\n", 1926 -err); 1927 goto out_err; 1928 } 1929 xs_reclassify_socket6(sock); 1930 1931 if (xs_bind6(transport, sock) < 0) { 1932 sock_release(sock); 1933 goto out_err; 1934 } 1935 return sock; 1936 out_err: 1937 return ERR_PTR(-EIO); 1938 } 1939 1940 /** 1941 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint 1942 * @work: RPC transport to connect 1943 * 1944 * Invoked by a work queue tasklet. 1945 */ 1946 static void xs_tcp_connect_worker6(struct work_struct *work) 1947 { 1948 struct sock_xprt *transport = 1949 container_of(work, struct sock_xprt, connect_worker.work); 1950 struct rpc_xprt *xprt = &transport->xprt; 1951 1952 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6); 1953 } 1954 1955 /** 1956 * xs_connect - connect a socket to a remote endpoint 1957 * @task: address of RPC task that manages state of connect request 1958 * 1959 * TCP: If the remote end dropped the connection, delay reconnecting. 1960 * 1961 * UDP socket connects are synchronous, but we use a work queue anyway 1962 * to guarantee that even unprivileged user processes can set up a 1963 * socket on a privileged port. 1964 * 1965 * If a UDP socket connect fails, the delay behavior here prevents 1966 * retry floods (hard mounts). 1967 */ 1968 static void xs_connect(struct rpc_task *task) 1969 { 1970 struct rpc_xprt *xprt = task->tk_xprt; 1971 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1972 1973 if (xprt_test_and_set_connecting(xprt)) 1974 return; 1975 1976 if (transport->sock != NULL) { 1977 dprintk("RPC: xs_connect delayed xprt %p for %lu " 1978 "seconds\n", 1979 xprt, xprt->reestablish_timeout / HZ); 1980 queue_delayed_work(rpciod_workqueue, 1981 &transport->connect_worker, 1982 xprt->reestablish_timeout); 1983 xprt->reestablish_timeout <<= 1; 1984 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) 1985 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; 1986 } else { 1987 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 1988 queue_delayed_work(rpciod_workqueue, 1989 &transport->connect_worker, 0); 1990 } 1991 } 1992 1993 static void xs_tcp_connect(struct rpc_task *task) 1994 { 1995 struct rpc_xprt *xprt = task->tk_xprt; 1996 1997 /* Exit if we need to wait for socket shutdown to complete */ 1998 if (test_bit(XPRT_CLOSING, &xprt->state)) 1999 return; 2000 xs_connect(task); 2001 } 2002 2003 /** 2004 * xs_udp_print_stats - display UDP socket-specifc stats 2005 * @xprt: rpc_xprt struct containing statistics 2006 * @seq: output file 2007 * 2008 */ 2009 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2010 { 2011 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2012 2013 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n", 2014 transport->port, 2015 xprt->stat.bind_count, 2016 xprt->stat.sends, 2017 xprt->stat.recvs, 2018 xprt->stat.bad_xids, 2019 xprt->stat.req_u, 2020 xprt->stat.bklog_u); 2021 } 2022 2023 /** 2024 * xs_tcp_print_stats - display TCP socket-specifc stats 2025 * @xprt: rpc_xprt struct containing statistics 2026 * @seq: output file 2027 * 2028 */ 2029 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2030 { 2031 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2032 long idle_time = 0; 2033 2034 if (xprt_connected(xprt)) 2035 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2036 2037 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n", 2038 transport->port, 2039 xprt->stat.bind_count, 2040 xprt->stat.connect_count, 2041 xprt->stat.connect_time, 2042 idle_time, 2043 xprt->stat.sends, 2044 xprt->stat.recvs, 2045 xprt->stat.bad_xids, 2046 xprt->stat.req_u, 2047 xprt->stat.bklog_u); 2048 } 2049 2050 static struct rpc_xprt_ops xs_udp_ops = { 2051 .set_buffer_size = xs_udp_set_buffer_size, 2052 .reserve_xprt = xprt_reserve_xprt_cong, 2053 .release_xprt = xprt_release_xprt_cong, 2054 .rpcbind = rpcb_getport_async, 2055 .set_port = xs_set_port, 2056 .connect = xs_connect, 2057 .buf_alloc = rpc_malloc, 2058 .buf_free = rpc_free, 2059 .send_request = xs_udp_send_request, 2060 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 2061 .timer = xs_udp_timer, 2062 .release_request = xprt_release_rqst_cong, 2063 .close = xs_close, 2064 .destroy = xs_destroy, 2065 .print_stats = xs_udp_print_stats, 2066 }; 2067 2068 static struct rpc_xprt_ops xs_tcp_ops = { 2069 .reserve_xprt = xprt_reserve_xprt, 2070 .release_xprt = xs_tcp_release_xprt, 2071 .rpcbind = rpcb_getport_async, 2072 .set_port = xs_set_port, 2073 .connect = xs_tcp_connect, 2074 .buf_alloc = rpc_malloc, 2075 .buf_free = rpc_free, 2076 .send_request = xs_tcp_send_request, 2077 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2078 .close = xs_tcp_close, 2079 .destroy = xs_destroy, 2080 .print_stats = xs_tcp_print_stats, 2081 }; 2082 2083 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2084 unsigned int slot_table_size) 2085 { 2086 struct rpc_xprt *xprt; 2087 struct sock_xprt *new; 2088 2089 if (args->addrlen > sizeof(xprt->addr)) { 2090 dprintk("RPC: xs_setup_xprt: address too large\n"); 2091 return ERR_PTR(-EBADF); 2092 } 2093 2094 new = kzalloc(sizeof(*new), GFP_KERNEL); 2095 if (new == NULL) { 2096 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2097 "rpc_xprt\n"); 2098 return ERR_PTR(-ENOMEM); 2099 } 2100 xprt = &new->xprt; 2101 2102 xprt->max_reqs = slot_table_size; 2103 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL); 2104 if (xprt->slot == NULL) { 2105 kfree(xprt); 2106 dprintk("RPC: xs_setup_xprt: couldn't allocate slot " 2107 "table\n"); 2108 return ERR_PTR(-ENOMEM); 2109 } 2110 2111 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2112 xprt->addrlen = args->addrlen; 2113 if (args->srcaddr) 2114 memcpy(&new->addr, args->srcaddr, args->addrlen); 2115 2116 return xprt; 2117 } 2118 2119 static const struct rpc_timeout xs_udp_default_timeout = { 2120 .to_initval = 5 * HZ, 2121 .to_maxval = 30 * HZ, 2122 .to_increment = 5 * HZ, 2123 .to_retries = 5, 2124 }; 2125 2126 /** 2127 * xs_setup_udp - Set up transport to use a UDP socket 2128 * @args: rpc transport creation arguments 2129 * 2130 */ 2131 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2132 { 2133 struct sockaddr *addr = args->dstaddr; 2134 struct rpc_xprt *xprt; 2135 struct sock_xprt *transport; 2136 2137 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries); 2138 if (IS_ERR(xprt)) 2139 return xprt; 2140 transport = container_of(xprt, struct sock_xprt, xprt); 2141 2142 xprt->prot = IPPROTO_UDP; 2143 xprt->tsh_size = 0; 2144 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2145 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2146 2147 xprt->bind_timeout = XS_BIND_TO; 2148 xprt->connect_timeout = XS_UDP_CONN_TO; 2149 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2150 xprt->idle_timeout = XS_IDLE_DISC_TO; 2151 2152 xprt->ops = &xs_udp_ops; 2153 2154 xprt->timeout = &xs_udp_default_timeout; 2155 2156 switch (addr->sa_family) { 2157 case AF_INET: 2158 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2159 xprt_set_bound(xprt); 2160 2161 INIT_DELAYED_WORK(&transport->connect_worker, 2162 xs_udp_connect_worker4); 2163 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2164 break; 2165 case AF_INET6: 2166 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2167 xprt_set_bound(xprt); 2168 2169 INIT_DELAYED_WORK(&transport->connect_worker, 2170 xs_udp_connect_worker6); 2171 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2172 break; 2173 default: 2174 kfree(xprt); 2175 return ERR_PTR(-EAFNOSUPPORT); 2176 } 2177 2178 dprintk("RPC: set up transport to address %s\n", 2179 xprt->address_strings[RPC_DISPLAY_ALL]); 2180 2181 if (try_module_get(THIS_MODULE)) 2182 return xprt; 2183 2184 kfree(xprt->slot); 2185 kfree(xprt); 2186 return ERR_PTR(-EINVAL); 2187 } 2188 2189 static const struct rpc_timeout xs_tcp_default_timeout = { 2190 .to_initval = 60 * HZ, 2191 .to_maxval = 60 * HZ, 2192 .to_retries = 2, 2193 }; 2194 2195 /** 2196 * xs_setup_tcp - Set up transport to use a TCP socket 2197 * @args: rpc transport creation arguments 2198 * 2199 */ 2200 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2201 { 2202 struct sockaddr *addr = args->dstaddr; 2203 struct rpc_xprt *xprt; 2204 struct sock_xprt *transport; 2205 2206 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries); 2207 if (IS_ERR(xprt)) 2208 return xprt; 2209 transport = container_of(xprt, struct sock_xprt, xprt); 2210 2211 xprt->prot = IPPROTO_TCP; 2212 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2213 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2214 2215 xprt->bind_timeout = XS_BIND_TO; 2216 xprt->connect_timeout = XS_TCP_CONN_TO; 2217 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2218 xprt->idle_timeout = XS_IDLE_DISC_TO; 2219 2220 xprt->ops = &xs_tcp_ops; 2221 xprt->timeout = &xs_tcp_default_timeout; 2222 2223 switch (addr->sa_family) { 2224 case AF_INET: 2225 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2226 xprt_set_bound(xprt); 2227 2228 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4); 2229 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2230 break; 2231 case AF_INET6: 2232 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2233 xprt_set_bound(xprt); 2234 2235 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6); 2236 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 2237 break; 2238 default: 2239 kfree(xprt); 2240 return ERR_PTR(-EAFNOSUPPORT); 2241 } 2242 2243 dprintk("RPC: set up transport to address %s\n", 2244 xprt->address_strings[RPC_DISPLAY_ALL]); 2245 2246 if (try_module_get(THIS_MODULE)) 2247 return xprt; 2248 2249 kfree(xprt->slot); 2250 kfree(xprt); 2251 return ERR_PTR(-EINVAL); 2252 } 2253 2254 static struct xprt_class xs_udp_transport = { 2255 .list = LIST_HEAD_INIT(xs_udp_transport.list), 2256 .name = "udp", 2257 .owner = THIS_MODULE, 2258 .ident = IPPROTO_UDP, 2259 .setup = xs_setup_udp, 2260 }; 2261 2262 static struct xprt_class xs_tcp_transport = { 2263 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 2264 .name = "tcp", 2265 .owner = THIS_MODULE, 2266 .ident = IPPROTO_TCP, 2267 .setup = xs_setup_tcp, 2268 }; 2269 2270 /** 2271 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 2272 * 2273 */ 2274 int init_socket_xprt(void) 2275 { 2276 #ifdef RPC_DEBUG 2277 if (!sunrpc_table_header) 2278 sunrpc_table_header = register_sysctl_table(sunrpc_table); 2279 #endif 2280 2281 xprt_register_transport(&xs_udp_transport); 2282 xprt_register_transport(&xs_tcp_transport); 2283 2284 return 0; 2285 } 2286 2287 /** 2288 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 2289 * 2290 */ 2291 void cleanup_socket_xprt(void) 2292 { 2293 #ifdef RPC_DEBUG 2294 if (sunrpc_table_header) { 2295 unregister_sysctl_table(sunrpc_table_header); 2296 sunrpc_table_header = NULL; 2297 } 2298 #endif 2299 2300 xprt_unregister_transport(&xs_udp_transport); 2301 xprt_unregister_transport(&xs_tcp_transport); 2302 } 2303