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 -ENOBUFS: 598 case -EAGAIN: 599 status = xs_nospace(task); 600 break; 601 default: 602 dprintk("RPC: sendmsg returned unrecognized error %d\n", 603 -status); 604 case -EPIPE: 605 xs_close(xprt); 606 status = -ENOTCONN; 607 } 608 609 return status; 610 } 611 612 /** 613 * xs_udp_send_request - write an RPC request to a UDP socket 614 * @task: address of RPC task that manages the state of an RPC request 615 * 616 * Return values: 617 * 0: The request has been sent 618 * EAGAIN: The socket was blocked, please call again later to 619 * complete the request 620 * ENOTCONN: Caller needs to invoke connect logic then call again 621 * other: Some other error occurred, the request was not sent 622 */ 623 static int xs_udp_send_request(struct rpc_task *task) 624 { 625 struct rpc_rqst *req = task->tk_rqstp; 626 struct rpc_xprt *xprt = req->rq_xprt; 627 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 628 struct xdr_buf *xdr = &req->rq_snd_buf; 629 int status; 630 631 xs_pktdump("packet data:", 632 req->rq_svec->iov_base, 633 req->rq_svec->iov_len); 634 635 if (!xprt_bound(xprt)) 636 return -ENOTCONN; 637 status = xs_sendpages(transport->sock, 638 xs_addr(xprt), 639 xprt->addrlen, xdr, 640 req->rq_bytes_sent, true); 641 642 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 643 xdr->len - req->rq_bytes_sent, status); 644 645 if (status >= 0) { 646 req->rq_xmit_bytes_sent += status; 647 if (status >= req->rq_slen) 648 return 0; 649 /* Still some bytes left; set up for a retry later. */ 650 status = -EAGAIN; 651 } 652 653 switch (status) { 654 case -ENOTSOCK: 655 status = -ENOTCONN; 656 /* Should we call xs_close() here? */ 657 break; 658 case -EAGAIN: 659 status = xs_nospace(task); 660 break; 661 default: 662 dprintk("RPC: sendmsg returned unrecognized error %d\n", 663 -status); 664 case -ENETUNREACH: 665 case -ENOBUFS: 666 case -EPIPE: 667 case -ECONNREFUSED: 668 /* When the server has died, an ICMP port unreachable message 669 * prompts ECONNREFUSED. */ 670 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 671 } 672 673 return status; 674 } 675 676 /** 677 * xs_tcp_shutdown - gracefully shut down a TCP socket 678 * @xprt: transport 679 * 680 * Initiates a graceful shutdown of the TCP socket by calling the 681 * equivalent of shutdown(SHUT_WR); 682 */ 683 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 684 { 685 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 686 struct socket *sock = transport->sock; 687 688 if (sock != NULL) { 689 kernel_sock_shutdown(sock, SHUT_WR); 690 trace_rpc_socket_shutdown(xprt, sock); 691 } 692 } 693 694 /** 695 * xs_tcp_send_request - write an RPC request to a TCP socket 696 * @task: address of RPC task that manages the state of an RPC request 697 * 698 * Return values: 699 * 0: The request has been sent 700 * EAGAIN: The socket was blocked, please call again later to 701 * complete the request 702 * ENOTCONN: Caller needs to invoke connect logic then call again 703 * other: Some other error occurred, the request was not sent 704 * 705 * XXX: In the case of soft timeouts, should we eventually give up 706 * if sendmsg is not able to make progress? 707 */ 708 static int xs_tcp_send_request(struct rpc_task *task) 709 { 710 struct rpc_rqst *req = task->tk_rqstp; 711 struct rpc_xprt *xprt = req->rq_xprt; 712 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 713 struct xdr_buf *xdr = &req->rq_snd_buf; 714 bool zerocopy = true; 715 int status; 716 717 xs_encode_stream_record_marker(&req->rq_snd_buf); 718 719 xs_pktdump("packet data:", 720 req->rq_svec->iov_base, 721 req->rq_svec->iov_len); 722 /* Don't use zero copy if this is a resend. If the RPC call 723 * completes while the socket holds a reference to the pages, 724 * then we may end up resending corrupted data. 725 */ 726 if (task->tk_flags & RPC_TASK_SENT) 727 zerocopy = false; 728 729 /* Continue transmitting the packet/record. We must be careful 730 * to cope with writespace callbacks arriving _after_ we have 731 * called sendmsg(). */ 732 while (1) { 733 status = xs_sendpages(transport->sock, 734 NULL, 0, xdr, req->rq_bytes_sent, 735 zerocopy); 736 737 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 738 xdr->len - req->rq_bytes_sent, status); 739 740 if (unlikely(status < 0)) 741 break; 742 743 /* If we've sent the entire packet, immediately 744 * reset the count of bytes sent. */ 745 req->rq_bytes_sent += status; 746 req->rq_xmit_bytes_sent += status; 747 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 748 req->rq_bytes_sent = 0; 749 return 0; 750 } 751 752 if (status != 0) 753 continue; 754 status = -EAGAIN; 755 break; 756 } 757 758 switch (status) { 759 case -ENOTSOCK: 760 status = -ENOTCONN; 761 /* Should we call xs_close() here? */ 762 break; 763 case -ENOBUFS: 764 case -EAGAIN: 765 status = xs_nospace(task); 766 break; 767 default: 768 dprintk("RPC: sendmsg returned unrecognized error %d\n", 769 -status); 770 case -ECONNRESET: 771 xs_tcp_shutdown(xprt); 772 case -ECONNREFUSED: 773 case -ENOTCONN: 774 case -EPIPE: 775 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 776 } 777 778 return status; 779 } 780 781 /** 782 * xs_tcp_release_xprt - clean up after a tcp transmission 783 * @xprt: transport 784 * @task: rpc task 785 * 786 * This cleans up if an error causes us to abort the transmission of a request. 787 * In this case, the socket may need to be reset in order to avoid confusing 788 * the server. 789 */ 790 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 791 { 792 struct rpc_rqst *req; 793 794 if (task != xprt->snd_task) 795 return; 796 if (task == NULL) 797 goto out_release; 798 req = task->tk_rqstp; 799 if (req == NULL) 800 goto out_release; 801 if (req->rq_bytes_sent == 0) 802 goto out_release; 803 if (req->rq_bytes_sent == req->rq_snd_buf.len) 804 goto out_release; 805 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 806 out_release: 807 xprt_release_xprt(xprt, task); 808 } 809 810 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 811 { 812 transport->old_data_ready = sk->sk_data_ready; 813 transport->old_state_change = sk->sk_state_change; 814 transport->old_write_space = sk->sk_write_space; 815 transport->old_error_report = sk->sk_error_report; 816 } 817 818 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 819 { 820 sk->sk_data_ready = transport->old_data_ready; 821 sk->sk_state_change = transport->old_state_change; 822 sk->sk_write_space = transport->old_write_space; 823 sk->sk_error_report = transport->old_error_report; 824 } 825 826 /** 827 * xs_error_report - callback to handle TCP socket state errors 828 * @sk: socket 829 * 830 * Note: we don't call sock_error() since there may be a rpc_task 831 * using the socket, and so we don't want to clear sk->sk_err. 832 */ 833 static void xs_error_report(struct sock *sk) 834 { 835 struct rpc_xprt *xprt; 836 int err; 837 838 read_lock_bh(&sk->sk_callback_lock); 839 if (!(xprt = xprt_from_sock(sk))) 840 goto out; 841 842 err = -sk->sk_err; 843 if (err == 0) 844 goto out; 845 dprintk("RPC: xs_error_report client %p, error=%d...\n", 846 xprt, -err); 847 trace_rpc_socket_error(xprt, sk->sk_socket, err); 848 xprt_wake_pending_tasks(xprt, err); 849 out: 850 read_unlock_bh(&sk->sk_callback_lock); 851 } 852 853 static void xs_reset_transport(struct sock_xprt *transport) 854 { 855 struct socket *sock = transport->sock; 856 struct sock *sk = transport->inet; 857 858 if (sk == NULL) 859 return; 860 861 transport->srcport = 0; 862 863 write_lock_bh(&sk->sk_callback_lock); 864 transport->inet = NULL; 865 transport->sock = NULL; 866 867 sk->sk_user_data = NULL; 868 869 xs_restore_old_callbacks(transport, sk); 870 write_unlock_bh(&sk->sk_callback_lock); 871 872 trace_rpc_socket_close(&transport->xprt, sock); 873 sock_release(sock); 874 } 875 876 /** 877 * xs_close - close a socket 878 * @xprt: transport 879 * 880 * This is used when all requests are complete; ie, no DRC state remains 881 * on the server we want to save. 882 * 883 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 884 * xs_reset_transport() zeroing the socket from underneath a writer. 885 */ 886 static void xs_close(struct rpc_xprt *xprt) 887 { 888 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 889 890 dprintk("RPC: xs_close xprt %p\n", xprt); 891 892 cancel_delayed_work_sync(&transport->connect_worker); 893 894 xs_reset_transport(transport); 895 xprt->reestablish_timeout = 0; 896 897 smp_mb__before_atomic(); 898 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 899 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 900 clear_bit(XPRT_CLOSING, &xprt->state); 901 smp_mb__after_atomic(); 902 xprt_disconnect_done(xprt); 903 } 904 905 static void xs_tcp_close(struct rpc_xprt *xprt) 906 { 907 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state)) 908 xs_close(xprt); 909 else 910 xs_tcp_shutdown(xprt); 911 } 912 913 static void xs_xprt_free(struct rpc_xprt *xprt) 914 { 915 xs_free_peer_addresses(xprt); 916 xprt_free(xprt); 917 } 918 919 /** 920 * xs_destroy - prepare to shutdown a transport 921 * @xprt: doomed transport 922 * 923 */ 924 static void xs_destroy(struct rpc_xprt *xprt) 925 { 926 dprintk("RPC: xs_destroy xprt %p\n", xprt); 927 928 xs_close(xprt); 929 xs_xprt_free(xprt); 930 module_put(THIS_MODULE); 931 } 932 933 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb) 934 { 935 struct xdr_skb_reader desc = { 936 .skb = skb, 937 .offset = sizeof(rpc_fraghdr), 938 .count = skb->len - sizeof(rpc_fraghdr), 939 }; 940 941 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0) 942 return -1; 943 if (desc.count) 944 return -1; 945 return 0; 946 } 947 948 /** 949 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets 950 * @sk: socket with data to read 951 * @len: how much data to read 952 * 953 * Currently this assumes we can read the whole reply in a single gulp. 954 */ 955 static void xs_local_data_ready(struct sock *sk) 956 { 957 struct rpc_task *task; 958 struct rpc_xprt *xprt; 959 struct rpc_rqst *rovr; 960 struct sk_buff *skb; 961 int err, repsize, copied; 962 u32 _xid; 963 __be32 *xp; 964 965 read_lock_bh(&sk->sk_callback_lock); 966 dprintk("RPC: %s...\n", __func__); 967 xprt = xprt_from_sock(sk); 968 if (xprt == NULL) 969 goto out; 970 971 skb = skb_recv_datagram(sk, 0, 1, &err); 972 if (skb == NULL) 973 goto out; 974 975 repsize = skb->len - sizeof(rpc_fraghdr); 976 if (repsize < 4) { 977 dprintk("RPC: impossible RPC reply size %d\n", repsize); 978 goto dropit; 979 } 980 981 /* Copy the XID from the skb... */ 982 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid); 983 if (xp == NULL) 984 goto dropit; 985 986 /* Look up and lock the request corresponding to the given XID */ 987 spin_lock(&xprt->transport_lock); 988 rovr = xprt_lookup_rqst(xprt, *xp); 989 if (!rovr) 990 goto out_unlock; 991 task = rovr->rq_task; 992 993 copied = rovr->rq_private_buf.buflen; 994 if (copied > repsize) 995 copied = repsize; 996 997 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) { 998 dprintk("RPC: sk_buff copy failed\n"); 999 goto out_unlock; 1000 } 1001 1002 xprt_complete_rqst(task, copied); 1003 1004 out_unlock: 1005 spin_unlock(&xprt->transport_lock); 1006 dropit: 1007 skb_free_datagram(sk, skb); 1008 out: 1009 read_unlock_bh(&sk->sk_callback_lock); 1010 } 1011 1012 /** 1013 * xs_udp_data_ready - "data ready" callback for UDP sockets 1014 * @sk: socket with data to read 1015 * @len: how much data to read 1016 * 1017 */ 1018 static void xs_udp_data_ready(struct sock *sk) 1019 { 1020 struct rpc_task *task; 1021 struct rpc_xprt *xprt; 1022 struct rpc_rqst *rovr; 1023 struct sk_buff *skb; 1024 int err, repsize, copied; 1025 u32 _xid; 1026 __be32 *xp; 1027 1028 read_lock_bh(&sk->sk_callback_lock); 1029 dprintk("RPC: xs_udp_data_ready...\n"); 1030 if (!(xprt = xprt_from_sock(sk))) 1031 goto out; 1032 1033 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) 1034 goto out; 1035 1036 repsize = skb->len - sizeof(struct udphdr); 1037 if (repsize < 4) { 1038 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1039 goto dropit; 1040 } 1041 1042 /* Copy the XID from the skb... */ 1043 xp = skb_header_pointer(skb, sizeof(struct udphdr), 1044 sizeof(_xid), &_xid); 1045 if (xp == NULL) 1046 goto dropit; 1047 1048 /* Look up and lock the request corresponding to the given XID */ 1049 spin_lock(&xprt->transport_lock); 1050 rovr = xprt_lookup_rqst(xprt, *xp); 1051 if (!rovr) 1052 goto out_unlock; 1053 task = rovr->rq_task; 1054 1055 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1056 copied = repsize; 1057 1058 /* Suck it into the iovec, verify checksum if not done by hw. */ 1059 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1060 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS); 1061 goto out_unlock; 1062 } 1063 1064 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS); 1065 1066 xprt_adjust_cwnd(xprt, task, copied); 1067 xprt_complete_rqst(task, copied); 1068 1069 out_unlock: 1070 spin_unlock(&xprt->transport_lock); 1071 dropit: 1072 skb_free_datagram(sk, skb); 1073 out: 1074 read_unlock_bh(&sk->sk_callback_lock); 1075 } 1076 1077 /* 1078 * Helper function to force a TCP close if the server is sending 1079 * junk and/or it has put us in CLOSE_WAIT 1080 */ 1081 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1082 { 1083 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); 1084 xprt_force_disconnect(xprt); 1085 } 1086 1087 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 1088 { 1089 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1090 size_t len, used; 1091 char *p; 1092 1093 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; 1094 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; 1095 used = xdr_skb_read_bits(desc, p, len); 1096 transport->tcp_offset += used; 1097 if (used != len) 1098 return; 1099 1100 transport->tcp_reclen = ntohl(transport->tcp_fraghdr); 1101 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) 1102 transport->tcp_flags |= TCP_RCV_LAST_FRAG; 1103 else 1104 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; 1105 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; 1106 1107 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; 1108 transport->tcp_offset = 0; 1109 1110 /* Sanity check of the record length */ 1111 if (unlikely(transport->tcp_reclen < 8)) { 1112 dprintk("RPC: invalid TCP record fragment length\n"); 1113 xs_tcp_force_close(xprt); 1114 return; 1115 } 1116 dprintk("RPC: reading TCP record fragment of length %d\n", 1117 transport->tcp_reclen); 1118 } 1119 1120 static void xs_tcp_check_fraghdr(struct sock_xprt *transport) 1121 { 1122 if (transport->tcp_offset == transport->tcp_reclen) { 1123 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; 1124 transport->tcp_offset = 0; 1125 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { 1126 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1127 transport->tcp_flags |= TCP_RCV_COPY_XID; 1128 transport->tcp_copied = 0; 1129 } 1130 } 1131 } 1132 1133 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1134 { 1135 size_t len, used; 1136 char *p; 1137 1138 len = sizeof(transport->tcp_xid) - transport->tcp_offset; 1139 dprintk("RPC: reading XID (%Zu bytes)\n", len); 1140 p = ((char *) &transport->tcp_xid) + transport->tcp_offset; 1141 used = xdr_skb_read_bits(desc, p, len); 1142 transport->tcp_offset += used; 1143 if (used != len) 1144 return; 1145 transport->tcp_flags &= ~TCP_RCV_COPY_XID; 1146 transport->tcp_flags |= TCP_RCV_READ_CALLDIR; 1147 transport->tcp_copied = 4; 1148 dprintk("RPC: reading %s XID %08x\n", 1149 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for" 1150 : "request with", 1151 ntohl(transport->tcp_xid)); 1152 xs_tcp_check_fraghdr(transport); 1153 } 1154 1155 static inline void xs_tcp_read_calldir(struct sock_xprt *transport, 1156 struct xdr_skb_reader *desc) 1157 { 1158 size_t len, used; 1159 u32 offset; 1160 char *p; 1161 1162 /* 1163 * We want transport->tcp_offset to be 8 at the end of this routine 1164 * (4 bytes for the xid and 4 bytes for the call/reply flag). 1165 * When this function is called for the first time, 1166 * transport->tcp_offset is 4 (after having already read the xid). 1167 */ 1168 offset = transport->tcp_offset - sizeof(transport->tcp_xid); 1169 len = sizeof(transport->tcp_calldir) - offset; 1170 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len); 1171 p = ((char *) &transport->tcp_calldir) + offset; 1172 used = xdr_skb_read_bits(desc, p, len); 1173 transport->tcp_offset += used; 1174 if (used != len) 1175 return; 1176 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR; 1177 /* 1178 * We don't yet have the XDR buffer, so we will write the calldir 1179 * out after we get the buffer from the 'struct rpc_rqst' 1180 */ 1181 switch (ntohl(transport->tcp_calldir)) { 1182 case RPC_REPLY: 1183 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1184 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1185 transport->tcp_flags |= TCP_RPC_REPLY; 1186 break; 1187 case RPC_CALL: 1188 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1189 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1190 transport->tcp_flags &= ~TCP_RPC_REPLY; 1191 break; 1192 default: 1193 dprintk("RPC: invalid request message type\n"); 1194 xs_tcp_force_close(&transport->xprt); 1195 } 1196 xs_tcp_check_fraghdr(transport); 1197 } 1198 1199 static inline void xs_tcp_read_common(struct rpc_xprt *xprt, 1200 struct xdr_skb_reader *desc, 1201 struct rpc_rqst *req) 1202 { 1203 struct sock_xprt *transport = 1204 container_of(xprt, struct sock_xprt, xprt); 1205 struct xdr_buf *rcvbuf; 1206 size_t len; 1207 ssize_t r; 1208 1209 rcvbuf = &req->rq_private_buf; 1210 1211 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) { 1212 /* 1213 * Save the RPC direction in the XDR buffer 1214 */ 1215 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied, 1216 &transport->tcp_calldir, 1217 sizeof(transport->tcp_calldir)); 1218 transport->tcp_copied += sizeof(transport->tcp_calldir); 1219 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR; 1220 } 1221 1222 len = desc->count; 1223 if (len > transport->tcp_reclen - transport->tcp_offset) { 1224 struct xdr_skb_reader my_desc; 1225 1226 len = transport->tcp_reclen - transport->tcp_offset; 1227 memcpy(&my_desc, desc, sizeof(my_desc)); 1228 my_desc.count = len; 1229 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1230 &my_desc, xdr_skb_read_bits); 1231 desc->count -= r; 1232 desc->offset += r; 1233 } else 1234 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1235 desc, xdr_skb_read_bits); 1236 1237 if (r > 0) { 1238 transport->tcp_copied += r; 1239 transport->tcp_offset += r; 1240 } 1241 if (r != len) { 1242 /* Error when copying to the receive buffer, 1243 * usually because we weren't able to allocate 1244 * additional buffer pages. All we can do now 1245 * is turn off TCP_RCV_COPY_DATA, so the request 1246 * will not receive any additional updates, 1247 * and time out. 1248 * Any remaining data from this record will 1249 * be discarded. 1250 */ 1251 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1252 dprintk("RPC: XID %08x truncated request\n", 1253 ntohl(transport->tcp_xid)); 1254 dprintk("RPC: xprt = %p, tcp_copied = %lu, " 1255 "tcp_offset = %u, tcp_reclen = %u\n", 1256 xprt, transport->tcp_copied, 1257 transport->tcp_offset, transport->tcp_reclen); 1258 return; 1259 } 1260 1261 dprintk("RPC: XID %08x read %Zd bytes\n", 1262 ntohl(transport->tcp_xid), r); 1263 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " 1264 "tcp_reclen = %u\n", xprt, transport->tcp_copied, 1265 transport->tcp_offset, transport->tcp_reclen); 1266 1267 if (transport->tcp_copied == req->rq_private_buf.buflen) 1268 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1269 else if (transport->tcp_offset == transport->tcp_reclen) { 1270 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) 1271 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1272 } 1273 } 1274 1275 /* 1276 * Finds the request corresponding to the RPC xid and invokes the common 1277 * tcp read code to read the data. 1278 */ 1279 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt, 1280 struct xdr_skb_reader *desc) 1281 { 1282 struct sock_xprt *transport = 1283 container_of(xprt, struct sock_xprt, xprt); 1284 struct rpc_rqst *req; 1285 1286 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid)); 1287 1288 /* Find and lock the request corresponding to this xid */ 1289 spin_lock(&xprt->transport_lock); 1290 req = xprt_lookup_rqst(xprt, transport->tcp_xid); 1291 if (!req) { 1292 dprintk("RPC: XID %08x request not found!\n", 1293 ntohl(transport->tcp_xid)); 1294 spin_unlock(&xprt->transport_lock); 1295 return -1; 1296 } 1297 1298 xs_tcp_read_common(xprt, desc, req); 1299 1300 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1301 xprt_complete_rqst(req->rq_task, transport->tcp_copied); 1302 1303 spin_unlock(&xprt->transport_lock); 1304 return 0; 1305 } 1306 1307 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1308 /* 1309 * Obtains an rpc_rqst previously allocated and invokes the common 1310 * tcp read code to read the data. The result is placed in the callback 1311 * queue. 1312 * If we're unable to obtain the rpc_rqst we schedule the closing of the 1313 * connection and return -1. 1314 */ 1315 static int xs_tcp_read_callback(struct rpc_xprt *xprt, 1316 struct xdr_skb_reader *desc) 1317 { 1318 struct sock_xprt *transport = 1319 container_of(xprt, struct sock_xprt, xprt); 1320 struct rpc_rqst *req; 1321 1322 /* Look up and lock the request corresponding to the given XID */ 1323 spin_lock(&xprt->transport_lock); 1324 req = xprt_lookup_bc_request(xprt, transport->tcp_xid); 1325 if (req == NULL) { 1326 spin_unlock(&xprt->transport_lock); 1327 printk(KERN_WARNING "Callback slot table overflowed\n"); 1328 xprt_force_disconnect(xprt); 1329 return -1; 1330 } 1331 1332 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid)); 1333 xs_tcp_read_common(xprt, desc, req); 1334 1335 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1336 xprt_complete_bc_request(req, transport->tcp_copied); 1337 spin_unlock(&xprt->transport_lock); 1338 1339 return 0; 1340 } 1341 1342 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1343 struct xdr_skb_reader *desc) 1344 { 1345 struct sock_xprt *transport = 1346 container_of(xprt, struct sock_xprt, xprt); 1347 1348 return (transport->tcp_flags & TCP_RPC_REPLY) ? 1349 xs_tcp_read_reply(xprt, desc) : 1350 xs_tcp_read_callback(xprt, desc); 1351 } 1352 #else 1353 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1354 struct xdr_skb_reader *desc) 1355 { 1356 return xs_tcp_read_reply(xprt, desc); 1357 } 1358 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1359 1360 /* 1361 * Read data off the transport. This can be either an RPC_CALL or an 1362 * RPC_REPLY. Relay the processing to helper functions. 1363 */ 1364 static void xs_tcp_read_data(struct rpc_xprt *xprt, 1365 struct xdr_skb_reader *desc) 1366 { 1367 struct sock_xprt *transport = 1368 container_of(xprt, struct sock_xprt, xprt); 1369 1370 if (_xs_tcp_read_data(xprt, desc) == 0) 1371 xs_tcp_check_fraghdr(transport); 1372 else { 1373 /* 1374 * The transport_lock protects the request handling. 1375 * There's no need to hold it to update the tcp_flags. 1376 */ 1377 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1378 } 1379 } 1380 1381 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1382 { 1383 size_t len; 1384 1385 len = transport->tcp_reclen - transport->tcp_offset; 1386 if (len > desc->count) 1387 len = desc->count; 1388 desc->count -= len; 1389 desc->offset += len; 1390 transport->tcp_offset += len; 1391 dprintk("RPC: discarded %Zu bytes\n", len); 1392 xs_tcp_check_fraghdr(transport); 1393 } 1394 1395 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) 1396 { 1397 struct rpc_xprt *xprt = rd_desc->arg.data; 1398 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1399 struct xdr_skb_reader desc = { 1400 .skb = skb, 1401 .offset = offset, 1402 .count = len, 1403 }; 1404 1405 dprintk("RPC: xs_tcp_data_recv started\n"); 1406 do { 1407 /* Read in a new fragment marker if necessary */ 1408 /* Can we ever really expect to get completely empty fragments? */ 1409 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { 1410 xs_tcp_read_fraghdr(xprt, &desc); 1411 continue; 1412 } 1413 /* Read in the xid if necessary */ 1414 if (transport->tcp_flags & TCP_RCV_COPY_XID) { 1415 xs_tcp_read_xid(transport, &desc); 1416 continue; 1417 } 1418 /* Read in the call/reply flag */ 1419 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) { 1420 xs_tcp_read_calldir(transport, &desc); 1421 continue; 1422 } 1423 /* Read in the request data */ 1424 if (transport->tcp_flags & TCP_RCV_COPY_DATA) { 1425 xs_tcp_read_data(xprt, &desc); 1426 continue; 1427 } 1428 /* Skip over any trailing bytes on short reads */ 1429 xs_tcp_read_discard(transport, &desc); 1430 } while (desc.count); 1431 dprintk("RPC: xs_tcp_data_recv done\n"); 1432 return len - desc.count; 1433 } 1434 1435 /** 1436 * xs_tcp_data_ready - "data ready" callback for TCP sockets 1437 * @sk: socket with data to read 1438 * @bytes: how much data to read 1439 * 1440 */ 1441 static void xs_tcp_data_ready(struct sock *sk) 1442 { 1443 struct rpc_xprt *xprt; 1444 read_descriptor_t rd_desc; 1445 int read; 1446 1447 dprintk("RPC: xs_tcp_data_ready...\n"); 1448 1449 read_lock_bh(&sk->sk_callback_lock); 1450 if (!(xprt = xprt_from_sock(sk))) 1451 goto out; 1452 /* Any data means we had a useful conversation, so 1453 * the we don't need to delay the next reconnect 1454 */ 1455 if (xprt->reestablish_timeout) 1456 xprt->reestablish_timeout = 0; 1457 1458 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ 1459 rd_desc.arg.data = xprt; 1460 do { 1461 rd_desc.count = 65536; 1462 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); 1463 } while (read > 0); 1464 out: 1465 read_unlock_bh(&sk->sk_callback_lock); 1466 } 1467 1468 /* 1469 * Do the equivalent of linger/linger2 handling for dealing with 1470 * broken servers that don't close the socket in a timely 1471 * fashion 1472 */ 1473 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt, 1474 unsigned long timeout) 1475 { 1476 struct sock_xprt *transport; 1477 1478 if (xprt_test_and_set_connecting(xprt)) 1479 return; 1480 set_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1481 transport = container_of(xprt, struct sock_xprt, xprt); 1482 queue_delayed_work(rpciod_workqueue, &transport->connect_worker, 1483 timeout); 1484 } 1485 1486 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt) 1487 { 1488 struct sock_xprt *transport; 1489 1490 transport = container_of(xprt, struct sock_xprt, xprt); 1491 1492 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) || 1493 !cancel_delayed_work(&transport->connect_worker)) 1494 return; 1495 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1496 xprt_clear_connecting(xprt); 1497 } 1498 1499 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 1500 { 1501 smp_mb__before_atomic(); 1502 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1503 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state); 1504 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1505 clear_bit(XPRT_CLOSING, &xprt->state); 1506 smp_mb__after_atomic(); 1507 } 1508 1509 static void xs_sock_mark_closed(struct rpc_xprt *xprt) 1510 { 1511 xs_sock_reset_connection_flags(xprt); 1512 /* Mark transport as closed and wake up all pending tasks */ 1513 xprt_disconnect_done(xprt); 1514 } 1515 1516 /** 1517 * xs_tcp_state_change - callback to handle TCP socket state changes 1518 * @sk: socket whose state has changed 1519 * 1520 */ 1521 static void xs_tcp_state_change(struct sock *sk) 1522 { 1523 struct rpc_xprt *xprt; 1524 1525 read_lock_bh(&sk->sk_callback_lock); 1526 if (!(xprt = xprt_from_sock(sk))) 1527 goto out; 1528 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1529 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1530 sk->sk_state, xprt_connected(xprt), 1531 sock_flag(sk, SOCK_DEAD), 1532 sock_flag(sk, SOCK_ZAPPED), 1533 sk->sk_shutdown); 1534 1535 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1536 switch (sk->sk_state) { 1537 case TCP_ESTABLISHED: 1538 spin_lock(&xprt->transport_lock); 1539 if (!xprt_test_and_set_connected(xprt)) { 1540 struct sock_xprt *transport = container_of(xprt, 1541 struct sock_xprt, xprt); 1542 1543 /* Reset TCP record info */ 1544 transport->tcp_offset = 0; 1545 transport->tcp_reclen = 0; 1546 transport->tcp_copied = 0; 1547 transport->tcp_flags = 1548 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; 1549 xprt->connect_cookie++; 1550 1551 xprt_wake_pending_tasks(xprt, -EAGAIN); 1552 } 1553 spin_unlock(&xprt->transport_lock); 1554 break; 1555 case TCP_FIN_WAIT1: 1556 /* The client initiated a shutdown of the socket */ 1557 xprt->connect_cookie++; 1558 xprt->reestablish_timeout = 0; 1559 set_bit(XPRT_CLOSING, &xprt->state); 1560 smp_mb__before_atomic(); 1561 clear_bit(XPRT_CONNECTED, &xprt->state); 1562 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1563 smp_mb__after_atomic(); 1564 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); 1565 break; 1566 case TCP_CLOSE_WAIT: 1567 /* The server initiated a shutdown of the socket */ 1568 xprt->connect_cookie++; 1569 clear_bit(XPRT_CONNECTED, &xprt->state); 1570 xs_tcp_force_close(xprt); 1571 case TCP_CLOSING: 1572 /* 1573 * If the server closed down the connection, make sure that 1574 * we back off before reconnecting 1575 */ 1576 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1577 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1578 break; 1579 case TCP_LAST_ACK: 1580 set_bit(XPRT_CLOSING, &xprt->state); 1581 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout); 1582 smp_mb__before_atomic(); 1583 clear_bit(XPRT_CONNECTED, &xprt->state); 1584 smp_mb__after_atomic(); 1585 break; 1586 case TCP_CLOSE: 1587 xs_tcp_cancel_linger_timeout(xprt); 1588 xs_sock_mark_closed(xprt); 1589 } 1590 out: 1591 read_unlock_bh(&sk->sk_callback_lock); 1592 } 1593 1594 static void xs_write_space(struct sock *sk) 1595 { 1596 struct socket *sock; 1597 struct rpc_xprt *xprt; 1598 1599 if (unlikely(!(sock = sk->sk_socket))) 1600 return; 1601 clear_bit(SOCK_NOSPACE, &sock->flags); 1602 1603 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1604 return; 1605 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) 1606 return; 1607 1608 xprt_write_space(xprt); 1609 } 1610 1611 /** 1612 * xs_udp_write_space - callback invoked when socket buffer space 1613 * becomes available 1614 * @sk: socket whose state has changed 1615 * 1616 * Called when more output buffer space is available for this socket. 1617 * We try not to wake our writers until they can make "significant" 1618 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1619 * with a bunch of small requests. 1620 */ 1621 static void xs_udp_write_space(struct sock *sk) 1622 { 1623 read_lock_bh(&sk->sk_callback_lock); 1624 1625 /* from net/core/sock.c:sock_def_write_space */ 1626 if (sock_writeable(sk)) 1627 xs_write_space(sk); 1628 1629 read_unlock_bh(&sk->sk_callback_lock); 1630 } 1631 1632 /** 1633 * xs_tcp_write_space - callback invoked when socket buffer space 1634 * becomes available 1635 * @sk: socket whose state has changed 1636 * 1637 * Called when more output buffer space is available for this socket. 1638 * We try not to wake our writers until they can make "significant" 1639 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1640 * with a bunch of small requests. 1641 */ 1642 static void xs_tcp_write_space(struct sock *sk) 1643 { 1644 read_lock_bh(&sk->sk_callback_lock); 1645 1646 /* from net/core/stream.c:sk_stream_write_space */ 1647 if (sk_stream_is_writeable(sk)) 1648 xs_write_space(sk); 1649 1650 read_unlock_bh(&sk->sk_callback_lock); 1651 } 1652 1653 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1654 { 1655 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1656 struct sock *sk = transport->inet; 1657 1658 if (transport->rcvsize) { 1659 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1660 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1661 } 1662 if (transport->sndsize) { 1663 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1664 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1665 sk->sk_write_space(sk); 1666 } 1667 } 1668 1669 /** 1670 * xs_udp_set_buffer_size - set send and receive limits 1671 * @xprt: generic transport 1672 * @sndsize: requested size of send buffer, in bytes 1673 * @rcvsize: requested size of receive buffer, in bytes 1674 * 1675 * Set socket send and receive buffer size limits. 1676 */ 1677 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1678 { 1679 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1680 1681 transport->sndsize = 0; 1682 if (sndsize) 1683 transport->sndsize = sndsize + 1024; 1684 transport->rcvsize = 0; 1685 if (rcvsize) 1686 transport->rcvsize = rcvsize + 1024; 1687 1688 xs_udp_do_set_buffer_size(xprt); 1689 } 1690 1691 /** 1692 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1693 * @task: task that timed out 1694 * 1695 * Adjust the congestion window after a retransmit timeout has occurred. 1696 */ 1697 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1698 { 1699 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1700 } 1701 1702 static unsigned short xs_get_random_port(void) 1703 { 1704 unsigned short range = xprt_max_resvport - xprt_min_resvport; 1705 unsigned short rand = (unsigned short) prandom_u32() % range; 1706 return rand + xprt_min_resvport; 1707 } 1708 1709 /** 1710 * xs_set_port - reset the port number in the remote endpoint address 1711 * @xprt: generic transport 1712 * @port: new port number 1713 * 1714 */ 1715 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1716 { 1717 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1718 1719 rpc_set_port(xs_addr(xprt), port); 1720 xs_update_peer_port(xprt); 1721 } 1722 1723 static unsigned short xs_get_srcport(struct sock_xprt *transport) 1724 { 1725 unsigned short port = transport->srcport; 1726 1727 if (port == 0 && transport->xprt.resvport) 1728 port = xs_get_random_port(); 1729 return port; 1730 } 1731 1732 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1733 { 1734 if (transport->srcport != 0) 1735 transport->srcport = 0; 1736 if (!transport->xprt.resvport) 1737 return 0; 1738 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1739 return xprt_max_resvport; 1740 return --port; 1741 } 1742 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1743 { 1744 struct sockaddr_storage myaddr; 1745 int err, nloop = 0; 1746 unsigned short port = xs_get_srcport(transport); 1747 unsigned short last; 1748 1749 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1750 do { 1751 rpc_set_port((struct sockaddr *)&myaddr, port); 1752 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1753 transport->xprt.addrlen); 1754 if (port == 0) 1755 break; 1756 if (err == 0) { 1757 transport->srcport = port; 1758 break; 1759 } 1760 last = port; 1761 port = xs_next_srcport(transport, port); 1762 if (port > last) 1763 nloop++; 1764 } while (err == -EADDRINUSE && nloop != 2); 1765 1766 if (myaddr.ss_family == AF_INET) 1767 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1768 &((struct sockaddr_in *)&myaddr)->sin_addr, 1769 port, err ? "failed" : "ok", err); 1770 else 1771 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1772 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1773 port, err ? "failed" : "ok", err); 1774 return err; 1775 } 1776 1777 /* 1778 * We don't support autobind on AF_LOCAL sockets 1779 */ 1780 static void xs_local_rpcbind(struct rpc_task *task) 1781 { 1782 rcu_read_lock(); 1783 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt)); 1784 rcu_read_unlock(); 1785 } 1786 1787 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1788 { 1789 } 1790 1791 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1792 static struct lock_class_key xs_key[2]; 1793 static struct lock_class_key xs_slock_key[2]; 1794 1795 static inline void xs_reclassify_socketu(struct socket *sock) 1796 { 1797 struct sock *sk = sock->sk; 1798 1799 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1800 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); 1801 } 1802 1803 static inline void xs_reclassify_socket4(struct socket *sock) 1804 { 1805 struct sock *sk = sock->sk; 1806 1807 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1808 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1809 } 1810 1811 static inline void xs_reclassify_socket6(struct socket *sock) 1812 { 1813 struct sock *sk = sock->sk; 1814 1815 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1816 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1817 } 1818 1819 static inline void xs_reclassify_socket(int family, struct socket *sock) 1820 { 1821 WARN_ON_ONCE(sock_owned_by_user(sock->sk)); 1822 if (sock_owned_by_user(sock->sk)) 1823 return; 1824 1825 switch (family) { 1826 case AF_LOCAL: 1827 xs_reclassify_socketu(sock); 1828 break; 1829 case AF_INET: 1830 xs_reclassify_socket4(sock); 1831 break; 1832 case AF_INET6: 1833 xs_reclassify_socket6(sock); 1834 break; 1835 } 1836 } 1837 #else 1838 static inline void xs_reclassify_socketu(struct socket *sock) 1839 { 1840 } 1841 1842 static inline void xs_reclassify_socket4(struct socket *sock) 1843 { 1844 } 1845 1846 static inline void xs_reclassify_socket6(struct socket *sock) 1847 { 1848 } 1849 1850 static inline void xs_reclassify_socket(int family, struct socket *sock) 1851 { 1852 } 1853 #endif 1854 1855 static void xs_dummy_setup_socket(struct work_struct *work) 1856 { 1857 } 1858 1859 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1860 struct sock_xprt *transport, int family, int type, int protocol) 1861 { 1862 struct socket *sock; 1863 int err; 1864 1865 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1866 if (err < 0) { 1867 dprintk("RPC: can't create %d transport socket (%d).\n", 1868 protocol, -err); 1869 goto out; 1870 } 1871 xs_reclassify_socket(family, sock); 1872 1873 err = xs_bind(transport, sock); 1874 if (err) { 1875 sock_release(sock); 1876 goto out; 1877 } 1878 1879 return sock; 1880 out: 1881 return ERR_PTR(err); 1882 } 1883 1884 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1885 struct socket *sock) 1886 { 1887 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1888 xprt); 1889 1890 if (!transport->inet) { 1891 struct sock *sk = sock->sk; 1892 1893 write_lock_bh(&sk->sk_callback_lock); 1894 1895 xs_save_old_callbacks(transport, sk); 1896 1897 sk->sk_user_data = xprt; 1898 sk->sk_data_ready = xs_local_data_ready; 1899 sk->sk_write_space = xs_udp_write_space; 1900 sk->sk_error_report = xs_error_report; 1901 sk->sk_allocation = GFP_ATOMIC; 1902 1903 xprt_clear_connected(xprt); 1904 1905 /* Reset to new socket */ 1906 transport->sock = sock; 1907 transport->inet = sk; 1908 1909 write_unlock_bh(&sk->sk_callback_lock); 1910 } 1911 1912 /* Tell the socket layer to start connecting... */ 1913 xprt->stat.connect_count++; 1914 xprt->stat.connect_start = jiffies; 1915 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1916 } 1917 1918 /** 1919 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1920 * @xprt: RPC transport to connect 1921 * @transport: socket transport to connect 1922 * @create_sock: function to create a socket of the correct type 1923 */ 1924 static int xs_local_setup_socket(struct sock_xprt *transport) 1925 { 1926 struct rpc_xprt *xprt = &transport->xprt; 1927 struct socket *sock; 1928 int status = -EIO; 1929 1930 current->flags |= PF_FSTRANS; 1931 1932 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 1933 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1934 SOCK_STREAM, 0, &sock, 1); 1935 if (status < 0) { 1936 dprintk("RPC: can't create AF_LOCAL " 1937 "transport socket (%d).\n", -status); 1938 goto out; 1939 } 1940 xs_reclassify_socketu(sock); 1941 1942 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1943 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1944 1945 status = xs_local_finish_connecting(xprt, sock); 1946 trace_rpc_socket_connect(xprt, sock, status); 1947 switch (status) { 1948 case 0: 1949 dprintk("RPC: xprt %p connected to %s\n", 1950 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1951 xprt_set_connected(xprt); 1952 case -ENOBUFS: 1953 break; 1954 case -ENOENT: 1955 dprintk("RPC: xprt %p: socket %s does not exist\n", 1956 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1957 break; 1958 case -ECONNREFUSED: 1959 dprintk("RPC: xprt %p: connection refused for %s\n", 1960 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1961 break; 1962 default: 1963 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 1964 __func__, -status, 1965 xprt->address_strings[RPC_DISPLAY_ADDR]); 1966 } 1967 1968 out: 1969 xprt_clear_connecting(xprt); 1970 xprt_wake_pending_tasks(xprt, status); 1971 current->flags &= ~PF_FSTRANS; 1972 return status; 1973 } 1974 1975 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 1976 { 1977 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1978 int ret; 1979 1980 if (RPC_IS_ASYNC(task)) { 1981 /* 1982 * We want the AF_LOCAL connect to be resolved in the 1983 * filesystem namespace of the process making the rpc 1984 * call. Thus we connect synchronously. 1985 * 1986 * If we want to support asynchronous AF_LOCAL calls, 1987 * we'll need to figure out how to pass a namespace to 1988 * connect. 1989 */ 1990 rpc_exit(task, -ENOTCONN); 1991 return; 1992 } 1993 ret = xs_local_setup_socket(transport); 1994 if (ret && !RPC_IS_SOFTCONN(task)) 1995 msleep_interruptible(15000); 1996 } 1997 1998 #ifdef CONFIG_SUNRPC_SWAP 1999 static void xs_set_memalloc(struct rpc_xprt *xprt) 2000 { 2001 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2002 xprt); 2003 2004 if (xprt->swapper) 2005 sk_set_memalloc(transport->inet); 2006 } 2007 2008 /** 2009 * xs_swapper - Tag this transport as being used for swap. 2010 * @xprt: transport to tag 2011 * @enable: enable/disable 2012 * 2013 */ 2014 int xs_swapper(struct rpc_xprt *xprt, int enable) 2015 { 2016 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2017 xprt); 2018 int err = 0; 2019 2020 if (enable) { 2021 xprt->swapper++; 2022 xs_set_memalloc(xprt); 2023 } else if (xprt->swapper) { 2024 xprt->swapper--; 2025 sk_clear_memalloc(transport->inet); 2026 } 2027 2028 return err; 2029 } 2030 EXPORT_SYMBOL_GPL(xs_swapper); 2031 #else 2032 static void xs_set_memalloc(struct rpc_xprt *xprt) 2033 { 2034 } 2035 #endif 2036 2037 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2038 { 2039 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2040 2041 if (!transport->inet) { 2042 struct sock *sk = sock->sk; 2043 2044 write_lock_bh(&sk->sk_callback_lock); 2045 2046 xs_save_old_callbacks(transport, sk); 2047 2048 sk->sk_user_data = xprt; 2049 sk->sk_data_ready = xs_udp_data_ready; 2050 sk->sk_write_space = xs_udp_write_space; 2051 sk->sk_allocation = GFP_ATOMIC; 2052 2053 xprt_set_connected(xprt); 2054 2055 /* Reset to new socket */ 2056 transport->sock = sock; 2057 transport->inet = sk; 2058 2059 xs_set_memalloc(xprt); 2060 2061 write_unlock_bh(&sk->sk_callback_lock); 2062 } 2063 xs_udp_do_set_buffer_size(xprt); 2064 } 2065 2066 static void xs_udp_setup_socket(struct work_struct *work) 2067 { 2068 struct sock_xprt *transport = 2069 container_of(work, struct sock_xprt, connect_worker.work); 2070 struct rpc_xprt *xprt = &transport->xprt; 2071 struct socket *sock = transport->sock; 2072 int status = -EIO; 2073 2074 current->flags |= PF_FSTRANS; 2075 2076 /* Start by resetting any existing state */ 2077 xs_reset_transport(transport); 2078 sock = xs_create_sock(xprt, transport, 2079 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP); 2080 if (IS_ERR(sock)) 2081 goto out; 2082 2083 dprintk("RPC: worker connecting xprt %p via %s to " 2084 "%s (port %s)\n", xprt, 2085 xprt->address_strings[RPC_DISPLAY_PROTO], 2086 xprt->address_strings[RPC_DISPLAY_ADDR], 2087 xprt->address_strings[RPC_DISPLAY_PORT]); 2088 2089 xs_udp_finish_connecting(xprt, sock); 2090 trace_rpc_socket_connect(xprt, sock, 0); 2091 status = 0; 2092 out: 2093 xprt_clear_connecting(xprt); 2094 xprt_wake_pending_tasks(xprt, status); 2095 current->flags &= ~PF_FSTRANS; 2096 } 2097 2098 /* 2099 * We need to preserve the port number so the reply cache on the server can 2100 * find our cached RPC replies when we get around to reconnecting. 2101 */ 2102 static void xs_abort_connection(struct sock_xprt *transport) 2103 { 2104 int result; 2105 struct sockaddr any; 2106 2107 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport); 2108 2109 /* 2110 * Disconnect the transport socket by doing a connect operation 2111 * with AF_UNSPEC. This should return immediately... 2112 */ 2113 memset(&any, 0, sizeof(any)); 2114 any.sa_family = AF_UNSPEC; 2115 result = kernel_connect(transport->sock, &any, sizeof(any), 0); 2116 trace_rpc_socket_reset_connection(&transport->xprt, 2117 transport->sock, result); 2118 if (!result) 2119 xs_sock_reset_connection_flags(&transport->xprt); 2120 dprintk("RPC: AF_UNSPEC connect return code %d\n", result); 2121 } 2122 2123 static void xs_tcp_reuse_connection(struct sock_xprt *transport) 2124 { 2125 unsigned int state = transport->inet->sk_state; 2126 2127 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) { 2128 /* we don't need to abort the connection if the socket 2129 * hasn't undergone a shutdown 2130 */ 2131 if (transport->inet->sk_shutdown == 0) 2132 return; 2133 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n", 2134 __func__, transport->inet->sk_shutdown); 2135 } 2136 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) { 2137 /* we don't need to abort the connection if the socket 2138 * hasn't undergone a shutdown 2139 */ 2140 if (transport->inet->sk_shutdown == 0) 2141 return; 2142 dprintk("RPC: %s: ESTABLISHED/SYN_SENT " 2143 "sk_shutdown set to %d\n", 2144 __func__, transport->inet->sk_shutdown); 2145 } 2146 xs_abort_connection(transport); 2147 } 2148 2149 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2150 { 2151 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2152 int ret = -ENOTCONN; 2153 2154 if (!transport->inet) { 2155 struct sock *sk = sock->sk; 2156 unsigned int keepidle = xprt->timeout->to_initval / HZ; 2157 unsigned int keepcnt = xprt->timeout->to_retries + 1; 2158 unsigned int opt_on = 1; 2159 2160 /* TCP Keepalive options */ 2161 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 2162 (char *)&opt_on, sizeof(opt_on)); 2163 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, 2164 (char *)&keepidle, sizeof(keepidle)); 2165 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, 2166 (char *)&keepidle, sizeof(keepidle)); 2167 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, 2168 (char *)&keepcnt, sizeof(keepcnt)); 2169 2170 write_lock_bh(&sk->sk_callback_lock); 2171 2172 xs_save_old_callbacks(transport, sk); 2173 2174 sk->sk_user_data = xprt; 2175 sk->sk_data_ready = xs_tcp_data_ready; 2176 sk->sk_state_change = xs_tcp_state_change; 2177 sk->sk_write_space = xs_tcp_write_space; 2178 sk->sk_error_report = xs_error_report; 2179 sk->sk_allocation = GFP_ATOMIC; 2180 2181 /* socket options */ 2182 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 2183 sock_reset_flag(sk, SOCK_LINGER); 2184 tcp_sk(sk)->linger2 = 0; 2185 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 2186 2187 xprt_clear_connected(xprt); 2188 2189 /* Reset to new socket */ 2190 transport->sock = sock; 2191 transport->inet = sk; 2192 2193 write_unlock_bh(&sk->sk_callback_lock); 2194 } 2195 2196 if (!xprt_bound(xprt)) 2197 goto out; 2198 2199 xs_set_memalloc(xprt); 2200 2201 /* Tell the socket layer to start connecting... */ 2202 xprt->stat.connect_count++; 2203 xprt->stat.connect_start = jiffies; 2204 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2205 switch (ret) { 2206 case 0: 2207 case -EINPROGRESS: 2208 /* SYN_SENT! */ 2209 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2210 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2211 } 2212 out: 2213 return ret; 2214 } 2215 2216 /** 2217 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2218 * @xprt: RPC transport to connect 2219 * @transport: socket transport to connect 2220 * @create_sock: function to create a socket of the correct type 2221 * 2222 * Invoked by a work queue tasklet. 2223 */ 2224 static void xs_tcp_setup_socket(struct work_struct *work) 2225 { 2226 struct sock_xprt *transport = 2227 container_of(work, struct sock_xprt, connect_worker.work); 2228 struct socket *sock = transport->sock; 2229 struct rpc_xprt *xprt = &transport->xprt; 2230 int status = -EIO; 2231 2232 current->flags |= PF_FSTRANS; 2233 2234 if (!sock) { 2235 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state); 2236 sock = xs_create_sock(xprt, transport, 2237 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP); 2238 if (IS_ERR(sock)) { 2239 status = PTR_ERR(sock); 2240 goto out; 2241 } 2242 } else { 2243 int abort_and_exit; 2244 2245 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT, 2246 &xprt->state); 2247 /* "close" the socket, preserving the local port */ 2248 xs_tcp_reuse_connection(transport); 2249 2250 if (abort_and_exit) 2251 goto out_eagain; 2252 } 2253 2254 dprintk("RPC: worker connecting xprt %p via %s to " 2255 "%s (port %s)\n", xprt, 2256 xprt->address_strings[RPC_DISPLAY_PROTO], 2257 xprt->address_strings[RPC_DISPLAY_ADDR], 2258 xprt->address_strings[RPC_DISPLAY_PORT]); 2259 2260 status = xs_tcp_finish_connecting(xprt, sock); 2261 trace_rpc_socket_connect(xprt, sock, status); 2262 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2263 xprt, -status, xprt_connected(xprt), 2264 sock->sk->sk_state); 2265 switch (status) { 2266 default: 2267 printk("%s: connect returned unhandled error %d\n", 2268 __func__, status); 2269 case -EADDRNOTAVAIL: 2270 /* We're probably in TIME_WAIT. Get rid of existing socket, 2271 * and retry 2272 */ 2273 xs_tcp_force_close(xprt); 2274 break; 2275 case 0: 2276 case -EINPROGRESS: 2277 case -EALREADY: 2278 xprt_clear_connecting(xprt); 2279 current->flags &= ~PF_FSTRANS; 2280 return; 2281 case -EINVAL: 2282 /* Happens, for instance, if the user specified a link 2283 * local IPv6 address without a scope-id. 2284 */ 2285 case -ECONNREFUSED: 2286 case -ECONNRESET: 2287 case -ENETUNREACH: 2288 case -ENOBUFS: 2289 /* retry with existing socket, after a delay */ 2290 goto out; 2291 } 2292 out_eagain: 2293 status = -EAGAIN; 2294 out: 2295 xprt_clear_connecting(xprt); 2296 xprt_wake_pending_tasks(xprt, status); 2297 current->flags &= ~PF_FSTRANS; 2298 } 2299 2300 /** 2301 * xs_connect - connect a socket to a remote endpoint 2302 * @xprt: pointer to transport structure 2303 * @task: address of RPC task that manages state of connect request 2304 * 2305 * TCP: If the remote end dropped the connection, delay reconnecting. 2306 * 2307 * UDP socket connects are synchronous, but we use a work queue anyway 2308 * to guarantee that even unprivileged user processes can set up a 2309 * socket on a privileged port. 2310 * 2311 * If a UDP socket connect fails, the delay behavior here prevents 2312 * retry floods (hard mounts). 2313 */ 2314 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2315 { 2316 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2317 2318 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) { 2319 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2320 "seconds\n", 2321 xprt, xprt->reestablish_timeout / HZ); 2322 queue_delayed_work(rpciod_workqueue, 2323 &transport->connect_worker, 2324 xprt->reestablish_timeout); 2325 xprt->reestablish_timeout <<= 1; 2326 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2327 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2328 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) 2329 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; 2330 } else { 2331 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2332 queue_delayed_work(rpciod_workqueue, 2333 &transport->connect_worker, 0); 2334 } 2335 } 2336 2337 /** 2338 * xs_local_print_stats - display AF_LOCAL socket-specifc stats 2339 * @xprt: rpc_xprt struct containing statistics 2340 * @seq: output file 2341 * 2342 */ 2343 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2344 { 2345 long idle_time = 0; 2346 2347 if (xprt_connected(xprt)) 2348 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2349 2350 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2351 "%llu %llu %lu %llu %llu\n", 2352 xprt->stat.bind_count, 2353 xprt->stat.connect_count, 2354 xprt->stat.connect_time, 2355 idle_time, 2356 xprt->stat.sends, 2357 xprt->stat.recvs, 2358 xprt->stat.bad_xids, 2359 xprt->stat.req_u, 2360 xprt->stat.bklog_u, 2361 xprt->stat.max_slots, 2362 xprt->stat.sending_u, 2363 xprt->stat.pending_u); 2364 } 2365 2366 /** 2367 * xs_udp_print_stats - display UDP socket-specifc stats 2368 * @xprt: rpc_xprt struct containing statistics 2369 * @seq: output file 2370 * 2371 */ 2372 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2373 { 2374 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2375 2376 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2377 "%lu %llu %llu\n", 2378 transport->srcport, 2379 xprt->stat.bind_count, 2380 xprt->stat.sends, 2381 xprt->stat.recvs, 2382 xprt->stat.bad_xids, 2383 xprt->stat.req_u, 2384 xprt->stat.bklog_u, 2385 xprt->stat.max_slots, 2386 xprt->stat.sending_u, 2387 xprt->stat.pending_u); 2388 } 2389 2390 /** 2391 * xs_tcp_print_stats - display TCP socket-specifc stats 2392 * @xprt: rpc_xprt struct containing statistics 2393 * @seq: output file 2394 * 2395 */ 2396 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2397 { 2398 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2399 long idle_time = 0; 2400 2401 if (xprt_connected(xprt)) 2402 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2403 2404 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2405 "%llu %llu %lu %llu %llu\n", 2406 transport->srcport, 2407 xprt->stat.bind_count, 2408 xprt->stat.connect_count, 2409 xprt->stat.connect_time, 2410 idle_time, 2411 xprt->stat.sends, 2412 xprt->stat.recvs, 2413 xprt->stat.bad_xids, 2414 xprt->stat.req_u, 2415 xprt->stat.bklog_u, 2416 xprt->stat.max_slots, 2417 xprt->stat.sending_u, 2418 xprt->stat.pending_u); 2419 } 2420 2421 /* 2422 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2423 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2424 * to use the server side send routines. 2425 */ 2426 static void *bc_malloc(struct rpc_task *task, size_t size) 2427 { 2428 struct page *page; 2429 struct rpc_buffer *buf; 2430 2431 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer)); 2432 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) 2433 return NULL; 2434 2435 page = alloc_page(GFP_KERNEL); 2436 if (!page) 2437 return NULL; 2438 2439 buf = page_address(page); 2440 buf->len = PAGE_SIZE; 2441 2442 return buf->data; 2443 } 2444 2445 /* 2446 * Free the space allocated in the bc_alloc routine 2447 */ 2448 static void bc_free(void *buffer) 2449 { 2450 struct rpc_buffer *buf; 2451 2452 if (!buffer) 2453 return; 2454 2455 buf = container_of(buffer, struct rpc_buffer, data); 2456 free_page((unsigned long)buf); 2457 } 2458 2459 /* 2460 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex 2461 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request. 2462 */ 2463 static int bc_sendto(struct rpc_rqst *req) 2464 { 2465 int len; 2466 struct xdr_buf *xbufp = &req->rq_snd_buf; 2467 struct rpc_xprt *xprt = req->rq_xprt; 2468 struct sock_xprt *transport = 2469 container_of(xprt, struct sock_xprt, xprt); 2470 struct socket *sock = transport->sock; 2471 unsigned long headoff; 2472 unsigned long tailoff; 2473 2474 xs_encode_stream_record_marker(xbufp); 2475 2476 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK; 2477 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK; 2478 len = svc_send_common(sock, xbufp, 2479 virt_to_page(xbufp->head[0].iov_base), headoff, 2480 xbufp->tail[0].iov_base, tailoff); 2481 2482 if (len != xbufp->len) { 2483 printk(KERN_NOTICE "Error sending entire callback!\n"); 2484 len = -EAGAIN; 2485 } 2486 2487 return len; 2488 } 2489 2490 /* 2491 * The send routine. Borrows from svc_send 2492 */ 2493 static int bc_send_request(struct rpc_task *task) 2494 { 2495 struct rpc_rqst *req = task->tk_rqstp; 2496 struct svc_xprt *xprt; 2497 u32 len; 2498 2499 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid)); 2500 /* 2501 * Get the server socket associated with this callback xprt 2502 */ 2503 xprt = req->rq_xprt->bc_xprt; 2504 2505 /* 2506 * Grab the mutex to serialize data as the connection is shared 2507 * with the fore channel 2508 */ 2509 if (!mutex_trylock(&xprt->xpt_mutex)) { 2510 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL); 2511 if (!mutex_trylock(&xprt->xpt_mutex)) 2512 return -EAGAIN; 2513 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task); 2514 } 2515 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2516 len = -ENOTCONN; 2517 else 2518 len = bc_sendto(req); 2519 mutex_unlock(&xprt->xpt_mutex); 2520 2521 if (len > 0) 2522 len = 0; 2523 2524 return len; 2525 } 2526 2527 /* 2528 * The close routine. Since this is client initiated, we do nothing 2529 */ 2530 2531 static void bc_close(struct rpc_xprt *xprt) 2532 { 2533 } 2534 2535 /* 2536 * The xprt destroy routine. Again, because this connection is client 2537 * initiated, we do nothing 2538 */ 2539 2540 static void bc_destroy(struct rpc_xprt *xprt) 2541 { 2542 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2543 2544 xs_xprt_free(xprt); 2545 module_put(THIS_MODULE); 2546 } 2547 2548 static struct rpc_xprt_ops xs_local_ops = { 2549 .reserve_xprt = xprt_reserve_xprt, 2550 .release_xprt = xs_tcp_release_xprt, 2551 .alloc_slot = xprt_alloc_slot, 2552 .rpcbind = xs_local_rpcbind, 2553 .set_port = xs_local_set_port, 2554 .connect = xs_local_connect, 2555 .buf_alloc = rpc_malloc, 2556 .buf_free = rpc_free, 2557 .send_request = xs_local_send_request, 2558 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2559 .close = xs_close, 2560 .destroy = xs_destroy, 2561 .print_stats = xs_local_print_stats, 2562 }; 2563 2564 static struct rpc_xprt_ops xs_udp_ops = { 2565 .set_buffer_size = xs_udp_set_buffer_size, 2566 .reserve_xprt = xprt_reserve_xprt_cong, 2567 .release_xprt = xprt_release_xprt_cong, 2568 .alloc_slot = xprt_alloc_slot, 2569 .rpcbind = rpcb_getport_async, 2570 .set_port = xs_set_port, 2571 .connect = xs_connect, 2572 .buf_alloc = rpc_malloc, 2573 .buf_free = rpc_free, 2574 .send_request = xs_udp_send_request, 2575 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 2576 .timer = xs_udp_timer, 2577 .release_request = xprt_release_rqst_cong, 2578 .close = xs_close, 2579 .destroy = xs_destroy, 2580 .print_stats = xs_udp_print_stats, 2581 }; 2582 2583 static struct rpc_xprt_ops xs_tcp_ops = { 2584 .reserve_xprt = xprt_reserve_xprt, 2585 .release_xprt = xs_tcp_release_xprt, 2586 .alloc_slot = xprt_lock_and_alloc_slot, 2587 .rpcbind = rpcb_getport_async, 2588 .set_port = xs_set_port, 2589 .connect = xs_connect, 2590 .buf_alloc = rpc_malloc, 2591 .buf_free = rpc_free, 2592 .send_request = xs_tcp_send_request, 2593 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2594 .close = xs_tcp_close, 2595 .destroy = xs_destroy, 2596 .print_stats = xs_tcp_print_stats, 2597 }; 2598 2599 /* 2600 * The rpc_xprt_ops for the server backchannel 2601 */ 2602 2603 static struct rpc_xprt_ops bc_tcp_ops = { 2604 .reserve_xprt = xprt_reserve_xprt, 2605 .release_xprt = xprt_release_xprt, 2606 .alloc_slot = xprt_alloc_slot, 2607 .buf_alloc = bc_malloc, 2608 .buf_free = bc_free, 2609 .send_request = bc_send_request, 2610 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2611 .close = bc_close, 2612 .destroy = bc_destroy, 2613 .print_stats = xs_tcp_print_stats, 2614 }; 2615 2616 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2617 { 2618 static const struct sockaddr_in sin = { 2619 .sin_family = AF_INET, 2620 .sin_addr.s_addr = htonl(INADDR_ANY), 2621 }; 2622 static const struct sockaddr_in6 sin6 = { 2623 .sin6_family = AF_INET6, 2624 .sin6_addr = IN6ADDR_ANY_INIT, 2625 }; 2626 2627 switch (family) { 2628 case AF_LOCAL: 2629 break; 2630 case AF_INET: 2631 memcpy(sap, &sin, sizeof(sin)); 2632 break; 2633 case AF_INET6: 2634 memcpy(sap, &sin6, sizeof(sin6)); 2635 break; 2636 default: 2637 dprintk("RPC: %s: Bad address family\n", __func__); 2638 return -EAFNOSUPPORT; 2639 } 2640 return 0; 2641 } 2642 2643 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2644 unsigned int slot_table_size, 2645 unsigned int max_slot_table_size) 2646 { 2647 struct rpc_xprt *xprt; 2648 struct sock_xprt *new; 2649 2650 if (args->addrlen > sizeof(xprt->addr)) { 2651 dprintk("RPC: xs_setup_xprt: address too large\n"); 2652 return ERR_PTR(-EBADF); 2653 } 2654 2655 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2656 max_slot_table_size); 2657 if (xprt == NULL) { 2658 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2659 "rpc_xprt\n"); 2660 return ERR_PTR(-ENOMEM); 2661 } 2662 2663 new = container_of(xprt, struct sock_xprt, xprt); 2664 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2665 xprt->addrlen = args->addrlen; 2666 if (args->srcaddr) 2667 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2668 else { 2669 int err; 2670 err = xs_init_anyaddr(args->dstaddr->sa_family, 2671 (struct sockaddr *)&new->srcaddr); 2672 if (err != 0) { 2673 xprt_free(xprt); 2674 return ERR_PTR(err); 2675 } 2676 } 2677 2678 return xprt; 2679 } 2680 2681 static const struct rpc_timeout xs_local_default_timeout = { 2682 .to_initval = 10 * HZ, 2683 .to_maxval = 10 * HZ, 2684 .to_retries = 2, 2685 }; 2686 2687 /** 2688 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2689 * @args: rpc transport creation arguments 2690 * 2691 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2692 */ 2693 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2694 { 2695 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2696 struct sock_xprt *transport; 2697 struct rpc_xprt *xprt; 2698 struct rpc_xprt *ret; 2699 2700 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2701 xprt_max_tcp_slot_table_entries); 2702 if (IS_ERR(xprt)) 2703 return xprt; 2704 transport = container_of(xprt, struct sock_xprt, xprt); 2705 2706 xprt->prot = 0; 2707 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2708 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2709 2710 xprt->bind_timeout = XS_BIND_TO; 2711 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2712 xprt->idle_timeout = XS_IDLE_DISC_TO; 2713 2714 xprt->ops = &xs_local_ops; 2715 xprt->timeout = &xs_local_default_timeout; 2716 2717 INIT_DELAYED_WORK(&transport->connect_worker, 2718 xs_dummy_setup_socket); 2719 2720 switch (sun->sun_family) { 2721 case AF_LOCAL: 2722 if (sun->sun_path[0] != '/') { 2723 dprintk("RPC: bad AF_LOCAL address: %s\n", 2724 sun->sun_path); 2725 ret = ERR_PTR(-EINVAL); 2726 goto out_err; 2727 } 2728 xprt_set_bound(xprt); 2729 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2730 ret = ERR_PTR(xs_local_setup_socket(transport)); 2731 if (ret) 2732 goto out_err; 2733 break; 2734 default: 2735 ret = ERR_PTR(-EAFNOSUPPORT); 2736 goto out_err; 2737 } 2738 2739 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2740 xprt->address_strings[RPC_DISPLAY_ADDR]); 2741 2742 if (try_module_get(THIS_MODULE)) 2743 return xprt; 2744 ret = ERR_PTR(-EINVAL); 2745 out_err: 2746 xs_xprt_free(xprt); 2747 return ret; 2748 } 2749 2750 static const struct rpc_timeout xs_udp_default_timeout = { 2751 .to_initval = 5 * HZ, 2752 .to_maxval = 30 * HZ, 2753 .to_increment = 5 * HZ, 2754 .to_retries = 5, 2755 }; 2756 2757 /** 2758 * xs_setup_udp - Set up transport to use a UDP socket 2759 * @args: rpc transport creation arguments 2760 * 2761 */ 2762 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2763 { 2764 struct sockaddr *addr = args->dstaddr; 2765 struct rpc_xprt *xprt; 2766 struct sock_xprt *transport; 2767 struct rpc_xprt *ret; 2768 2769 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2770 xprt_udp_slot_table_entries); 2771 if (IS_ERR(xprt)) 2772 return xprt; 2773 transport = container_of(xprt, struct sock_xprt, xprt); 2774 2775 xprt->prot = IPPROTO_UDP; 2776 xprt->tsh_size = 0; 2777 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2778 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2779 2780 xprt->bind_timeout = XS_BIND_TO; 2781 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2782 xprt->idle_timeout = XS_IDLE_DISC_TO; 2783 2784 xprt->ops = &xs_udp_ops; 2785 2786 xprt->timeout = &xs_udp_default_timeout; 2787 2788 switch (addr->sa_family) { 2789 case AF_INET: 2790 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2791 xprt_set_bound(xprt); 2792 2793 INIT_DELAYED_WORK(&transport->connect_worker, 2794 xs_udp_setup_socket); 2795 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2796 break; 2797 case AF_INET6: 2798 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2799 xprt_set_bound(xprt); 2800 2801 INIT_DELAYED_WORK(&transport->connect_worker, 2802 xs_udp_setup_socket); 2803 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2804 break; 2805 default: 2806 ret = ERR_PTR(-EAFNOSUPPORT); 2807 goto out_err; 2808 } 2809 2810 if (xprt_bound(xprt)) 2811 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2812 xprt->address_strings[RPC_DISPLAY_ADDR], 2813 xprt->address_strings[RPC_DISPLAY_PORT], 2814 xprt->address_strings[RPC_DISPLAY_PROTO]); 2815 else 2816 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2817 xprt->address_strings[RPC_DISPLAY_ADDR], 2818 xprt->address_strings[RPC_DISPLAY_PROTO]); 2819 2820 if (try_module_get(THIS_MODULE)) 2821 return xprt; 2822 ret = ERR_PTR(-EINVAL); 2823 out_err: 2824 xs_xprt_free(xprt); 2825 return ret; 2826 } 2827 2828 static const struct rpc_timeout xs_tcp_default_timeout = { 2829 .to_initval = 60 * HZ, 2830 .to_maxval = 60 * HZ, 2831 .to_retries = 2, 2832 }; 2833 2834 /** 2835 * xs_setup_tcp - Set up transport to use a TCP socket 2836 * @args: rpc transport creation arguments 2837 * 2838 */ 2839 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2840 { 2841 struct sockaddr *addr = args->dstaddr; 2842 struct rpc_xprt *xprt; 2843 struct sock_xprt *transport; 2844 struct rpc_xprt *ret; 2845 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2846 2847 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2848 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2849 2850 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2851 max_slot_table_size); 2852 if (IS_ERR(xprt)) 2853 return xprt; 2854 transport = container_of(xprt, struct sock_xprt, xprt); 2855 2856 xprt->prot = IPPROTO_TCP; 2857 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2858 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2859 2860 xprt->bind_timeout = XS_BIND_TO; 2861 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2862 xprt->idle_timeout = XS_IDLE_DISC_TO; 2863 2864 xprt->ops = &xs_tcp_ops; 2865 xprt->timeout = &xs_tcp_default_timeout; 2866 2867 switch (addr->sa_family) { 2868 case AF_INET: 2869 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2870 xprt_set_bound(xprt); 2871 2872 INIT_DELAYED_WORK(&transport->connect_worker, 2873 xs_tcp_setup_socket); 2874 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2875 break; 2876 case AF_INET6: 2877 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2878 xprt_set_bound(xprt); 2879 2880 INIT_DELAYED_WORK(&transport->connect_worker, 2881 xs_tcp_setup_socket); 2882 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 2883 break; 2884 default: 2885 ret = ERR_PTR(-EAFNOSUPPORT); 2886 goto out_err; 2887 } 2888 2889 if (xprt_bound(xprt)) 2890 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2891 xprt->address_strings[RPC_DISPLAY_ADDR], 2892 xprt->address_strings[RPC_DISPLAY_PORT], 2893 xprt->address_strings[RPC_DISPLAY_PROTO]); 2894 else 2895 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2896 xprt->address_strings[RPC_DISPLAY_ADDR], 2897 xprt->address_strings[RPC_DISPLAY_PROTO]); 2898 2899 if (try_module_get(THIS_MODULE)) 2900 return xprt; 2901 ret = ERR_PTR(-EINVAL); 2902 out_err: 2903 xs_xprt_free(xprt); 2904 return ret; 2905 } 2906 2907 /** 2908 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 2909 * @args: rpc transport creation arguments 2910 * 2911 */ 2912 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 2913 { 2914 struct sockaddr *addr = args->dstaddr; 2915 struct rpc_xprt *xprt; 2916 struct sock_xprt *transport; 2917 struct svc_sock *bc_sock; 2918 struct rpc_xprt *ret; 2919 2920 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2921 xprt_tcp_slot_table_entries); 2922 if (IS_ERR(xprt)) 2923 return xprt; 2924 transport = container_of(xprt, struct sock_xprt, xprt); 2925 2926 xprt->prot = IPPROTO_TCP; 2927 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2928 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2929 xprt->timeout = &xs_tcp_default_timeout; 2930 2931 /* backchannel */ 2932 xprt_set_bound(xprt); 2933 xprt->bind_timeout = 0; 2934 xprt->reestablish_timeout = 0; 2935 xprt->idle_timeout = 0; 2936 2937 xprt->ops = &bc_tcp_ops; 2938 2939 switch (addr->sa_family) { 2940 case AF_INET: 2941 xs_format_peer_addresses(xprt, "tcp", 2942 RPCBIND_NETID_TCP); 2943 break; 2944 case AF_INET6: 2945 xs_format_peer_addresses(xprt, "tcp", 2946 RPCBIND_NETID_TCP6); 2947 break; 2948 default: 2949 ret = ERR_PTR(-EAFNOSUPPORT); 2950 goto out_err; 2951 } 2952 2953 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2954 xprt->address_strings[RPC_DISPLAY_ADDR], 2955 xprt->address_strings[RPC_DISPLAY_PORT], 2956 xprt->address_strings[RPC_DISPLAY_PROTO]); 2957 2958 /* 2959 * Once we've associated a backchannel xprt with a connection, 2960 * we want to keep it around as long as the connection lasts, 2961 * in case we need to start using it for a backchannel again; 2962 * this reference won't be dropped until bc_xprt is destroyed. 2963 */ 2964 xprt_get(xprt); 2965 args->bc_xprt->xpt_bc_xprt = xprt; 2966 xprt->bc_xprt = args->bc_xprt; 2967 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 2968 transport->sock = bc_sock->sk_sock; 2969 transport->inet = bc_sock->sk_sk; 2970 2971 /* 2972 * Since we don't want connections for the backchannel, we set 2973 * the xprt status to connected 2974 */ 2975 xprt_set_connected(xprt); 2976 2977 if (try_module_get(THIS_MODULE)) 2978 return xprt; 2979 2980 args->bc_xprt->xpt_bc_xprt = NULL; 2981 xprt_put(xprt); 2982 ret = ERR_PTR(-EINVAL); 2983 out_err: 2984 xs_xprt_free(xprt); 2985 return ret; 2986 } 2987 2988 static struct xprt_class xs_local_transport = { 2989 .list = LIST_HEAD_INIT(xs_local_transport.list), 2990 .name = "named UNIX socket", 2991 .owner = THIS_MODULE, 2992 .ident = XPRT_TRANSPORT_LOCAL, 2993 .setup = xs_setup_local, 2994 }; 2995 2996 static struct xprt_class xs_udp_transport = { 2997 .list = LIST_HEAD_INIT(xs_udp_transport.list), 2998 .name = "udp", 2999 .owner = THIS_MODULE, 3000 .ident = XPRT_TRANSPORT_UDP, 3001 .setup = xs_setup_udp, 3002 }; 3003 3004 static struct xprt_class xs_tcp_transport = { 3005 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3006 .name = "tcp", 3007 .owner = THIS_MODULE, 3008 .ident = XPRT_TRANSPORT_TCP, 3009 .setup = xs_setup_tcp, 3010 }; 3011 3012 static struct xprt_class xs_bc_tcp_transport = { 3013 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3014 .name = "tcp NFSv4.1 backchannel", 3015 .owner = THIS_MODULE, 3016 .ident = XPRT_TRANSPORT_BC_TCP, 3017 .setup = xs_setup_bc_tcp, 3018 }; 3019 3020 /** 3021 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3022 * 3023 */ 3024 int init_socket_xprt(void) 3025 { 3026 #ifdef RPC_DEBUG 3027 if (!sunrpc_table_header) 3028 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3029 #endif 3030 3031 xprt_register_transport(&xs_local_transport); 3032 xprt_register_transport(&xs_udp_transport); 3033 xprt_register_transport(&xs_tcp_transport); 3034 xprt_register_transport(&xs_bc_tcp_transport); 3035 3036 return 0; 3037 } 3038 3039 /** 3040 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3041 * 3042 */ 3043 void cleanup_socket_xprt(void) 3044 { 3045 #ifdef RPC_DEBUG 3046 if (sunrpc_table_header) { 3047 unregister_sysctl_table(sunrpc_table_header); 3048 sunrpc_table_header = NULL; 3049 } 3050 #endif 3051 3052 xprt_unregister_transport(&xs_local_transport); 3053 xprt_unregister_transport(&xs_udp_transport); 3054 xprt_unregister_transport(&xs_tcp_transport); 3055 xprt_unregister_transport(&xs_bc_tcp_transport); 3056 } 3057 3058 static int param_set_uint_minmax(const char *val, 3059 const struct kernel_param *kp, 3060 unsigned int min, unsigned int max) 3061 { 3062 unsigned int num; 3063 int ret; 3064 3065 if (!val) 3066 return -EINVAL; 3067 ret = kstrtouint(val, 0, &num); 3068 if (ret == -EINVAL || num < min || num > max) 3069 return -EINVAL; 3070 *((unsigned int *)kp->arg) = num; 3071 return 0; 3072 } 3073 3074 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3075 { 3076 return param_set_uint_minmax(val, kp, 3077 RPC_MIN_RESVPORT, 3078 RPC_MAX_RESVPORT); 3079 } 3080 3081 static struct kernel_param_ops param_ops_portnr = { 3082 .set = param_set_portnr, 3083 .get = param_get_uint, 3084 }; 3085 3086 #define param_check_portnr(name, p) \ 3087 __param_check(name, p, unsigned int); 3088 3089 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3090 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3091 3092 static int param_set_slot_table_size(const char *val, 3093 const struct kernel_param *kp) 3094 { 3095 return param_set_uint_minmax(val, kp, 3096 RPC_MIN_SLOT_TABLE, 3097 RPC_MAX_SLOT_TABLE); 3098 } 3099 3100 static struct kernel_param_ops param_ops_slot_table_size = { 3101 .set = param_set_slot_table_size, 3102 .get = param_get_uint, 3103 }; 3104 3105 #define param_check_slot_table_size(name, p) \ 3106 __param_check(name, p, unsigned int); 3107 3108 static int param_set_max_slot_table_size(const char *val, 3109 const struct kernel_param *kp) 3110 { 3111 return param_set_uint_minmax(val, kp, 3112 RPC_MIN_SLOT_TABLE, 3113 RPC_MAX_SLOT_TABLE_LIMIT); 3114 } 3115 3116 static struct kernel_param_ops param_ops_max_slot_table_size = { 3117 .set = param_set_max_slot_table_size, 3118 .get = param_get_uint, 3119 }; 3120 3121 #define param_check_max_slot_table_size(name, p) \ 3122 __param_check(name, p, unsigned int); 3123 3124 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3125 slot_table_size, 0644); 3126 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3127 max_slot_table_size, 0644); 3128 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3129 slot_table_size, 0644); 3130 3131