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 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 67 68 #define XS_TCP_LINGER_TO (15U * HZ) 69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 70 71 /* 72 * We can register our own files under /proc/sys/sunrpc by 73 * calling register_sysctl_table() again. The files in that 74 * directory become the union of all files registered there. 75 * 76 * We simply need to make sure that we don't collide with 77 * someone else's file names! 78 */ 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 #if IS_ENABLED(CONFIG_SUNRPC_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 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 220 { 221 return (struct rpc_xprt *) sk->sk_user_data; 222 } 223 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 225 { 226 return (struct sockaddr *) &xprt->addr; 227 } 228 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 230 { 231 return (struct sockaddr_un *) &xprt->addr; 232 } 233 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 235 { 236 return (struct sockaddr_in *) &xprt->addr; 237 } 238 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 240 { 241 return (struct sockaddr_in6 *) &xprt->addr; 242 } 243 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 245 { 246 struct sockaddr *sap = xs_addr(xprt); 247 struct sockaddr_in6 *sin6; 248 struct sockaddr_in *sin; 249 struct sockaddr_un *sun; 250 char buf[128]; 251 252 switch (sap->sa_family) { 253 case AF_LOCAL: 254 sun = xs_addr_un(xprt); 255 strlcpy(buf, sun->sun_path, sizeof(buf)); 256 xprt->address_strings[RPC_DISPLAY_ADDR] = 257 kstrdup(buf, GFP_KERNEL); 258 break; 259 case AF_INET: 260 (void)rpc_ntop(sap, buf, sizeof(buf)); 261 xprt->address_strings[RPC_DISPLAY_ADDR] = 262 kstrdup(buf, GFP_KERNEL); 263 sin = xs_addr_in(xprt); 264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 265 break; 266 case AF_INET6: 267 (void)rpc_ntop(sap, buf, sizeof(buf)); 268 xprt->address_strings[RPC_DISPLAY_ADDR] = 269 kstrdup(buf, GFP_KERNEL); 270 sin6 = xs_addr_in6(xprt); 271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 272 break; 273 default: 274 BUG(); 275 } 276 277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 278 } 279 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 281 { 282 struct sockaddr *sap = xs_addr(xprt); 283 char buf[128]; 284 285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 287 288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 290 } 291 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 293 const char *protocol, 294 const char *netid) 295 { 296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 297 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 298 xs_format_common_peer_addresses(xprt); 299 xs_format_common_peer_ports(xprt); 300 } 301 302 static void xs_update_peer_port(struct rpc_xprt *xprt) 303 { 304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 306 307 xs_format_common_peer_ports(xprt); 308 } 309 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 311 { 312 unsigned int i; 313 314 for (i = 0; i < RPC_DISPLAY_MAX; i++) 315 switch (i) { 316 case RPC_DISPLAY_PROTO: 317 case RPC_DISPLAY_NETID: 318 continue; 319 default: 320 kfree(xprt->address_strings[i]); 321 } 322 } 323 324 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 325 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more) 327 { 328 struct msghdr msg = { 329 .msg_name = addr, 330 .msg_namelen = addrlen, 331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0), 332 }; 333 struct kvec iov = { 334 .iov_base = vec->iov_base + base, 335 .iov_len = vec->iov_len - base, 336 }; 337 338 if (iov.iov_len != 0) 339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); 340 return kernel_sendmsg(sock, &msg, NULL, 0, 0); 341 } 342 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p) 344 { 345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page, 346 int offset, size_t size, int flags); 347 struct page **ppage; 348 unsigned int remainder; 349 int err; 350 351 remainder = xdr->page_len - base; 352 base += xdr->page_base; 353 ppage = xdr->pages + (base >> PAGE_SHIFT); 354 base &= ~PAGE_MASK; 355 do_sendpage = sock->ops->sendpage; 356 if (!zerocopy) 357 do_sendpage = sock_no_sendpage; 358 for(;;) { 359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); 360 int flags = XS_SENDMSG_FLAGS; 361 362 remainder -= len; 363 if (remainder != 0 || more) 364 flags |= MSG_MORE; 365 err = do_sendpage(sock, *ppage, base, len, flags); 366 if (remainder == 0 || err != len) 367 break; 368 *sent_p += err; 369 ppage++; 370 base = 0; 371 } 372 if (err > 0) { 373 *sent_p += err; 374 err = 0; 375 } 376 return err; 377 } 378 379 /** 380 * xs_sendpages - write pages directly to a socket 381 * @sock: socket to send on 382 * @addr: UDP only -- address of destination 383 * @addrlen: UDP only -- length of destination address 384 * @xdr: buffer containing this request 385 * @base: starting position in the buffer 386 * @zerocopy: true if it is safe to use sendpage() 387 * @sent_p: return the total number of bytes successfully queued for sending 388 * 389 */ 390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p) 391 { 392 unsigned int remainder = xdr->len - base; 393 int err = 0; 394 int sent = 0; 395 396 if (unlikely(!sock)) 397 return -ENOTSOCK; 398 399 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); 400 if (base != 0) { 401 addr = NULL; 402 addrlen = 0; 403 } 404 405 if (base < xdr->head[0].iov_len || addr != NULL) { 406 unsigned int len = xdr->head[0].iov_len - base; 407 remainder -= len; 408 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0); 409 if (remainder == 0 || err != len) 410 goto out; 411 *sent_p += err; 412 base = 0; 413 } else 414 base -= xdr->head[0].iov_len; 415 416 if (base < xdr->page_len) { 417 unsigned int len = xdr->page_len - base; 418 remainder -= len; 419 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent); 420 *sent_p += sent; 421 if (remainder == 0 || sent != len) 422 goto out; 423 base = 0; 424 } else 425 base -= xdr->page_len; 426 427 if (base >= xdr->tail[0].iov_len) 428 return 0; 429 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0); 430 out: 431 if (err > 0) { 432 *sent_p += err; 433 err = 0; 434 } 435 return err; 436 } 437 438 static void xs_nospace_callback(struct rpc_task *task) 439 { 440 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt); 441 442 transport->inet->sk_write_pending--; 443 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 444 } 445 446 /** 447 * xs_nospace - place task on wait queue if transmit was incomplete 448 * @task: task to put to sleep 449 * 450 */ 451 static int xs_nospace(struct rpc_task *task) 452 { 453 struct rpc_rqst *req = task->tk_rqstp; 454 struct rpc_xprt *xprt = req->rq_xprt; 455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 456 struct sock *sk = transport->inet; 457 int ret = -EAGAIN; 458 459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", 460 task->tk_pid, req->rq_slen - req->rq_bytes_sent, 461 req->rq_slen); 462 463 /* Protect against races with write_space */ 464 spin_lock_bh(&xprt->transport_lock); 465 466 /* Don't race with disconnect */ 467 if (xprt_connected(xprt)) { 468 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) { 469 /* 470 * Notify TCP that we're limited by the application 471 * window size 472 */ 473 set_bit(SOCK_NOSPACE, &transport->sock->flags); 474 sk->sk_write_pending++; 475 /* ...and wait for more buffer space */ 476 xprt_wait_for_buffer_space(task, xs_nospace_callback); 477 } 478 } else { 479 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 480 ret = -ENOTCONN; 481 } 482 483 spin_unlock_bh(&xprt->transport_lock); 484 485 /* Race breaker in case memory is freed before above code is called */ 486 sk->sk_write_space(sk); 487 return ret; 488 } 489 490 /* 491 * Construct a stream transport record marker in @buf. 492 */ 493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf) 494 { 495 u32 reclen = buf->len - sizeof(rpc_fraghdr); 496 rpc_fraghdr *base = buf->head[0].iov_base; 497 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen); 498 } 499 500 /** 501 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 502 * @task: RPC task that manages the state of an RPC request 503 * 504 * Return values: 505 * 0: The request has been sent 506 * EAGAIN: The socket was blocked, please call again later to 507 * complete the request 508 * ENOTCONN: Caller needs to invoke connect logic then call again 509 * other: Some other error occured, the request was not sent 510 */ 511 static int xs_local_send_request(struct rpc_task *task) 512 { 513 struct rpc_rqst *req = task->tk_rqstp; 514 struct rpc_xprt *xprt = req->rq_xprt; 515 struct sock_xprt *transport = 516 container_of(xprt, struct sock_xprt, xprt); 517 struct xdr_buf *xdr = &req->rq_snd_buf; 518 int status; 519 int sent = 0; 520 521 xs_encode_stream_record_marker(&req->rq_snd_buf); 522 523 xs_pktdump("packet data:", 524 req->rq_svec->iov_base, req->rq_svec->iov_len); 525 526 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent, 527 true, &sent); 528 dprintk("RPC: %s(%u) = %d\n", 529 __func__, xdr->len - req->rq_bytes_sent, status); 530 531 if (status == -EAGAIN && sock_writeable(transport->inet)) 532 status = -ENOBUFS; 533 534 if (likely(sent > 0) || status == 0) { 535 req->rq_bytes_sent += sent; 536 req->rq_xmit_bytes_sent += sent; 537 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 538 req->rq_bytes_sent = 0; 539 return 0; 540 } 541 status = -EAGAIN; 542 } 543 544 switch (status) { 545 case -ENOBUFS: 546 break; 547 case -EAGAIN: 548 status = xs_nospace(task); 549 break; 550 default: 551 dprintk("RPC: sendmsg returned unrecognized error %d\n", 552 -status); 553 case -EPIPE: 554 xs_close(xprt); 555 status = -ENOTCONN; 556 } 557 558 return status; 559 } 560 561 /** 562 * xs_udp_send_request - write an RPC request to a UDP socket 563 * @task: address of RPC task that manages the state of an RPC request 564 * 565 * Return values: 566 * 0: The request has been sent 567 * EAGAIN: The socket was blocked, please call again later to 568 * complete the request 569 * ENOTCONN: Caller needs to invoke connect logic then call again 570 * other: Some other error occurred, the request was not sent 571 */ 572 static int xs_udp_send_request(struct rpc_task *task) 573 { 574 struct rpc_rqst *req = task->tk_rqstp; 575 struct rpc_xprt *xprt = req->rq_xprt; 576 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 577 struct xdr_buf *xdr = &req->rq_snd_buf; 578 int sent = 0; 579 int status; 580 581 xs_pktdump("packet data:", 582 req->rq_svec->iov_base, 583 req->rq_svec->iov_len); 584 585 if (!xprt_bound(xprt)) 586 return -ENOTCONN; 587 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen, 588 xdr, req->rq_bytes_sent, true, &sent); 589 590 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 591 xdr->len - req->rq_bytes_sent, status); 592 593 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 594 if (status == -EPERM) 595 goto process_status; 596 597 if (status == -EAGAIN && sock_writeable(transport->inet)) 598 status = -ENOBUFS; 599 600 if (sent > 0 || status == 0) { 601 req->rq_xmit_bytes_sent += sent; 602 if (sent >= req->rq_slen) 603 return 0; 604 /* Still some bytes left; set up for a retry later. */ 605 status = -EAGAIN; 606 } 607 608 process_status: 609 switch (status) { 610 case -ENOTSOCK: 611 status = -ENOTCONN; 612 /* Should we call xs_close() here? */ 613 break; 614 case -EAGAIN: 615 status = xs_nospace(task); 616 break; 617 default: 618 dprintk("RPC: sendmsg returned unrecognized error %d\n", 619 -status); 620 case -ENETUNREACH: 621 case -ENOBUFS: 622 case -EPIPE: 623 case -ECONNREFUSED: 624 case -EPERM: 625 /* When the server has died, an ICMP port unreachable message 626 * prompts ECONNREFUSED. */ 627 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 628 } 629 630 return status; 631 } 632 633 /** 634 * xs_tcp_send_request - write an RPC request to a TCP socket 635 * @task: address of RPC task that manages the state of an RPC request 636 * 637 * Return values: 638 * 0: The request has been sent 639 * EAGAIN: The socket was blocked, please call again later to 640 * complete the request 641 * ENOTCONN: Caller needs to invoke connect logic then call again 642 * other: Some other error occurred, the request was not sent 643 * 644 * XXX: In the case of soft timeouts, should we eventually give up 645 * if sendmsg is not able to make progress? 646 */ 647 static int xs_tcp_send_request(struct rpc_task *task) 648 { 649 struct rpc_rqst *req = task->tk_rqstp; 650 struct rpc_xprt *xprt = req->rq_xprt; 651 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 652 struct xdr_buf *xdr = &req->rq_snd_buf; 653 bool zerocopy = true; 654 int status; 655 int sent; 656 657 xs_encode_stream_record_marker(&req->rq_snd_buf); 658 659 xs_pktdump("packet data:", 660 req->rq_svec->iov_base, 661 req->rq_svec->iov_len); 662 /* Don't use zero copy if this is a resend. If the RPC call 663 * completes while the socket holds a reference to the pages, 664 * then we may end up resending corrupted data. 665 */ 666 if (task->tk_flags & RPC_TASK_SENT) 667 zerocopy = false; 668 669 /* Continue transmitting the packet/record. We must be careful 670 * to cope with writespace callbacks arriving _after_ we have 671 * called sendmsg(). */ 672 while (1) { 673 sent = 0; 674 status = xs_sendpages(transport->sock, NULL, 0, xdr, 675 req->rq_bytes_sent, zerocopy, &sent); 676 677 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 678 xdr->len - req->rq_bytes_sent, status); 679 680 /* If we've sent the entire packet, immediately 681 * reset the count of bytes sent. */ 682 req->rq_bytes_sent += sent; 683 req->rq_xmit_bytes_sent += sent; 684 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 685 req->rq_bytes_sent = 0; 686 return 0; 687 } 688 689 if (status < 0) 690 break; 691 if (sent == 0) { 692 status = -EAGAIN; 693 break; 694 } 695 } 696 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet)) 697 status = -ENOBUFS; 698 699 switch (status) { 700 case -ENOTSOCK: 701 status = -ENOTCONN; 702 /* Should we call xs_close() here? */ 703 break; 704 case -EAGAIN: 705 status = xs_nospace(task); 706 break; 707 default: 708 dprintk("RPC: sendmsg returned unrecognized error %d\n", 709 -status); 710 case -ECONNRESET: 711 case -ECONNREFUSED: 712 case -ENOTCONN: 713 case -EADDRINUSE: 714 case -ENOBUFS: 715 case -EPIPE: 716 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 717 } 718 719 return status; 720 } 721 722 /** 723 * xs_tcp_release_xprt - clean up after a tcp transmission 724 * @xprt: transport 725 * @task: rpc task 726 * 727 * This cleans up if an error causes us to abort the transmission of a request. 728 * In this case, the socket may need to be reset in order to avoid confusing 729 * the server. 730 */ 731 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 732 { 733 struct rpc_rqst *req; 734 735 if (task != xprt->snd_task) 736 return; 737 if (task == NULL) 738 goto out_release; 739 req = task->tk_rqstp; 740 if (req == NULL) 741 goto out_release; 742 if (req->rq_bytes_sent == 0) 743 goto out_release; 744 if (req->rq_bytes_sent == req->rq_snd_buf.len) 745 goto out_release; 746 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 747 out_release: 748 xprt_release_xprt(xprt, task); 749 } 750 751 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 752 { 753 transport->old_data_ready = sk->sk_data_ready; 754 transport->old_state_change = sk->sk_state_change; 755 transport->old_write_space = sk->sk_write_space; 756 transport->old_error_report = sk->sk_error_report; 757 } 758 759 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 760 { 761 sk->sk_data_ready = transport->old_data_ready; 762 sk->sk_state_change = transport->old_state_change; 763 sk->sk_write_space = transport->old_write_space; 764 sk->sk_error_report = transport->old_error_report; 765 } 766 767 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 768 { 769 smp_mb__before_atomic(); 770 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 771 clear_bit(XPRT_CLOSING, &xprt->state); 772 smp_mb__after_atomic(); 773 } 774 775 static void xs_sock_mark_closed(struct rpc_xprt *xprt) 776 { 777 xs_sock_reset_connection_flags(xprt); 778 /* Mark transport as closed and wake up all pending tasks */ 779 xprt_disconnect_done(xprt); 780 xprt_force_disconnect(xprt); 781 } 782 783 /** 784 * xs_error_report - callback to handle TCP socket state errors 785 * @sk: socket 786 * 787 * Note: we don't call sock_error() since there may be a rpc_task 788 * using the socket, and so we don't want to clear sk->sk_err. 789 */ 790 static void xs_error_report(struct sock *sk) 791 { 792 struct rpc_xprt *xprt; 793 int err; 794 795 read_lock_bh(&sk->sk_callback_lock); 796 if (!(xprt = xprt_from_sock(sk))) 797 goto out; 798 799 err = -sk->sk_err; 800 if (err == 0) 801 goto out; 802 /* Is this a reset event? */ 803 if (sk->sk_state == TCP_CLOSE) 804 xs_sock_mark_closed(xprt); 805 dprintk("RPC: xs_error_report client %p, error=%d...\n", 806 xprt, -err); 807 trace_rpc_socket_error(xprt, sk->sk_socket, err); 808 xprt_wake_pending_tasks(xprt, err); 809 out: 810 read_unlock_bh(&sk->sk_callback_lock); 811 } 812 813 static void xs_reset_transport(struct sock_xprt *transport) 814 { 815 struct socket *sock = transport->sock; 816 struct sock *sk = transport->inet; 817 struct rpc_xprt *xprt = &transport->xprt; 818 819 if (sk == NULL) 820 return; 821 822 if (atomic_read(&transport->xprt.swapper)) 823 sk_clear_memalloc(sk); 824 825 kernel_sock_shutdown(sock, SHUT_RDWR); 826 827 write_lock_bh(&sk->sk_callback_lock); 828 transport->inet = NULL; 829 transport->sock = NULL; 830 831 sk->sk_user_data = NULL; 832 833 xs_restore_old_callbacks(transport, sk); 834 xprt_clear_connected(xprt); 835 write_unlock_bh(&sk->sk_callback_lock); 836 xs_sock_reset_connection_flags(xprt); 837 838 trace_rpc_socket_close(xprt, sock); 839 sock_release(sock); 840 } 841 842 /** 843 * xs_close - close a socket 844 * @xprt: transport 845 * 846 * This is used when all requests are complete; ie, no DRC state remains 847 * on the server we want to save. 848 * 849 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 850 * xs_reset_transport() zeroing the socket from underneath a writer. 851 */ 852 static void xs_close(struct rpc_xprt *xprt) 853 { 854 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 855 856 dprintk("RPC: xs_close xprt %p\n", xprt); 857 858 xs_reset_transport(transport); 859 xprt->reestablish_timeout = 0; 860 861 xprt_disconnect_done(xprt); 862 } 863 864 static void xs_inject_disconnect(struct rpc_xprt *xprt) 865 { 866 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 867 xprt); 868 xprt_disconnect_done(xprt); 869 } 870 871 static void xs_xprt_free(struct rpc_xprt *xprt) 872 { 873 xs_free_peer_addresses(xprt); 874 xprt_free(xprt); 875 } 876 877 /** 878 * xs_destroy - prepare to shutdown a transport 879 * @xprt: doomed transport 880 * 881 */ 882 static void xs_destroy(struct rpc_xprt *xprt) 883 { 884 dprintk("RPC: xs_destroy xprt %p\n", xprt); 885 886 xs_close(xprt); 887 xs_xprt_free(xprt); 888 module_put(THIS_MODULE); 889 } 890 891 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb) 892 { 893 struct xdr_skb_reader desc = { 894 .skb = skb, 895 .offset = sizeof(rpc_fraghdr), 896 .count = skb->len - sizeof(rpc_fraghdr), 897 }; 898 899 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0) 900 return -1; 901 if (desc.count) 902 return -1; 903 return 0; 904 } 905 906 /** 907 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets 908 * @sk: socket with data to read 909 * 910 * Currently this assumes we can read the whole reply in a single gulp. 911 */ 912 static void xs_local_data_ready(struct sock *sk) 913 { 914 struct rpc_task *task; 915 struct rpc_xprt *xprt; 916 struct rpc_rqst *rovr; 917 struct sk_buff *skb; 918 int err, repsize, copied; 919 u32 _xid; 920 __be32 *xp; 921 922 read_lock_bh(&sk->sk_callback_lock); 923 dprintk("RPC: %s...\n", __func__); 924 xprt = xprt_from_sock(sk); 925 if (xprt == NULL) 926 goto out; 927 928 skb = skb_recv_datagram(sk, 0, 1, &err); 929 if (skb == NULL) 930 goto out; 931 932 repsize = skb->len - sizeof(rpc_fraghdr); 933 if (repsize < 4) { 934 dprintk("RPC: impossible RPC reply size %d\n", repsize); 935 goto dropit; 936 } 937 938 /* Copy the XID from the skb... */ 939 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid); 940 if (xp == NULL) 941 goto dropit; 942 943 /* Look up and lock the request corresponding to the given XID */ 944 spin_lock(&xprt->transport_lock); 945 rovr = xprt_lookup_rqst(xprt, *xp); 946 if (!rovr) 947 goto out_unlock; 948 task = rovr->rq_task; 949 950 copied = rovr->rq_private_buf.buflen; 951 if (copied > repsize) 952 copied = repsize; 953 954 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) { 955 dprintk("RPC: sk_buff copy failed\n"); 956 goto out_unlock; 957 } 958 959 xprt_complete_rqst(task, copied); 960 961 out_unlock: 962 spin_unlock(&xprt->transport_lock); 963 dropit: 964 skb_free_datagram(sk, skb); 965 out: 966 read_unlock_bh(&sk->sk_callback_lock); 967 } 968 969 /** 970 * xs_udp_data_ready - "data ready" callback for UDP sockets 971 * @sk: socket with data to read 972 * 973 */ 974 static void xs_udp_data_ready(struct sock *sk) 975 { 976 struct rpc_task *task; 977 struct rpc_xprt *xprt; 978 struct rpc_rqst *rovr; 979 struct sk_buff *skb; 980 int err, repsize, copied; 981 u32 _xid; 982 __be32 *xp; 983 984 read_lock_bh(&sk->sk_callback_lock); 985 dprintk("RPC: xs_udp_data_ready...\n"); 986 if (!(xprt = xprt_from_sock(sk))) 987 goto out; 988 989 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) 990 goto out; 991 992 repsize = skb->len - sizeof(struct udphdr); 993 if (repsize < 4) { 994 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 995 goto dropit; 996 } 997 998 /* Copy the XID from the skb... */ 999 xp = skb_header_pointer(skb, sizeof(struct udphdr), 1000 sizeof(_xid), &_xid); 1001 if (xp == NULL) 1002 goto dropit; 1003 1004 /* Look up and lock the request corresponding to the given XID */ 1005 spin_lock(&xprt->transport_lock); 1006 rovr = xprt_lookup_rqst(xprt, *xp); 1007 if (!rovr) 1008 goto out_unlock; 1009 task = rovr->rq_task; 1010 1011 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1012 copied = repsize; 1013 1014 /* Suck it into the iovec, verify checksum if not done by hw. */ 1015 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1016 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS); 1017 goto out_unlock; 1018 } 1019 1020 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS); 1021 1022 xprt_adjust_cwnd(xprt, task, copied); 1023 xprt_complete_rqst(task, copied); 1024 1025 out_unlock: 1026 spin_unlock(&xprt->transport_lock); 1027 dropit: 1028 skb_free_datagram(sk, skb); 1029 out: 1030 read_unlock_bh(&sk->sk_callback_lock); 1031 } 1032 1033 /* 1034 * Helper function to force a TCP close if the server is sending 1035 * junk and/or it has put us in CLOSE_WAIT 1036 */ 1037 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1038 { 1039 xprt_force_disconnect(xprt); 1040 } 1041 1042 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 1043 { 1044 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1045 size_t len, used; 1046 char *p; 1047 1048 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; 1049 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; 1050 used = xdr_skb_read_bits(desc, p, len); 1051 transport->tcp_offset += used; 1052 if (used != len) 1053 return; 1054 1055 transport->tcp_reclen = ntohl(transport->tcp_fraghdr); 1056 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) 1057 transport->tcp_flags |= TCP_RCV_LAST_FRAG; 1058 else 1059 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; 1060 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; 1061 1062 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; 1063 transport->tcp_offset = 0; 1064 1065 /* Sanity check of the record length */ 1066 if (unlikely(transport->tcp_reclen < 8)) { 1067 dprintk("RPC: invalid TCP record fragment length\n"); 1068 xs_tcp_force_close(xprt); 1069 return; 1070 } 1071 dprintk("RPC: reading TCP record fragment of length %d\n", 1072 transport->tcp_reclen); 1073 } 1074 1075 static void xs_tcp_check_fraghdr(struct sock_xprt *transport) 1076 { 1077 if (transport->tcp_offset == transport->tcp_reclen) { 1078 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; 1079 transport->tcp_offset = 0; 1080 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { 1081 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1082 transport->tcp_flags |= TCP_RCV_COPY_XID; 1083 transport->tcp_copied = 0; 1084 } 1085 } 1086 } 1087 1088 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1089 { 1090 size_t len, used; 1091 char *p; 1092 1093 len = sizeof(transport->tcp_xid) - transport->tcp_offset; 1094 dprintk("RPC: reading XID (%Zu bytes)\n", len); 1095 p = ((char *) &transport->tcp_xid) + transport->tcp_offset; 1096 used = xdr_skb_read_bits(desc, p, len); 1097 transport->tcp_offset += used; 1098 if (used != len) 1099 return; 1100 transport->tcp_flags &= ~TCP_RCV_COPY_XID; 1101 transport->tcp_flags |= TCP_RCV_READ_CALLDIR; 1102 transport->tcp_copied = 4; 1103 dprintk("RPC: reading %s XID %08x\n", 1104 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for" 1105 : "request with", 1106 ntohl(transport->tcp_xid)); 1107 xs_tcp_check_fraghdr(transport); 1108 } 1109 1110 static inline void xs_tcp_read_calldir(struct sock_xprt *transport, 1111 struct xdr_skb_reader *desc) 1112 { 1113 size_t len, used; 1114 u32 offset; 1115 char *p; 1116 1117 /* 1118 * We want transport->tcp_offset to be 8 at the end of this routine 1119 * (4 bytes for the xid and 4 bytes for the call/reply flag). 1120 * When this function is called for the first time, 1121 * transport->tcp_offset is 4 (after having already read the xid). 1122 */ 1123 offset = transport->tcp_offset - sizeof(transport->tcp_xid); 1124 len = sizeof(transport->tcp_calldir) - offset; 1125 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len); 1126 p = ((char *) &transport->tcp_calldir) + offset; 1127 used = xdr_skb_read_bits(desc, p, len); 1128 transport->tcp_offset += used; 1129 if (used != len) 1130 return; 1131 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR; 1132 /* 1133 * We don't yet have the XDR buffer, so we will write the calldir 1134 * out after we get the buffer from the 'struct rpc_rqst' 1135 */ 1136 switch (ntohl(transport->tcp_calldir)) { 1137 case RPC_REPLY: 1138 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1139 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1140 transport->tcp_flags |= TCP_RPC_REPLY; 1141 break; 1142 case RPC_CALL: 1143 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1144 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1145 transport->tcp_flags &= ~TCP_RPC_REPLY; 1146 break; 1147 default: 1148 dprintk("RPC: invalid request message type\n"); 1149 xs_tcp_force_close(&transport->xprt); 1150 } 1151 xs_tcp_check_fraghdr(transport); 1152 } 1153 1154 static inline void xs_tcp_read_common(struct rpc_xprt *xprt, 1155 struct xdr_skb_reader *desc, 1156 struct rpc_rqst *req) 1157 { 1158 struct sock_xprt *transport = 1159 container_of(xprt, struct sock_xprt, xprt); 1160 struct xdr_buf *rcvbuf; 1161 size_t len; 1162 ssize_t r; 1163 1164 rcvbuf = &req->rq_private_buf; 1165 1166 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) { 1167 /* 1168 * Save the RPC direction in the XDR buffer 1169 */ 1170 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied, 1171 &transport->tcp_calldir, 1172 sizeof(transport->tcp_calldir)); 1173 transport->tcp_copied += sizeof(transport->tcp_calldir); 1174 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR; 1175 } 1176 1177 len = desc->count; 1178 if (len > transport->tcp_reclen - transport->tcp_offset) { 1179 struct xdr_skb_reader my_desc; 1180 1181 len = transport->tcp_reclen - transport->tcp_offset; 1182 memcpy(&my_desc, desc, sizeof(my_desc)); 1183 my_desc.count = len; 1184 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1185 &my_desc, xdr_skb_read_bits); 1186 desc->count -= r; 1187 desc->offset += r; 1188 } else 1189 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1190 desc, xdr_skb_read_bits); 1191 1192 if (r > 0) { 1193 transport->tcp_copied += r; 1194 transport->tcp_offset += r; 1195 } 1196 if (r != len) { 1197 /* Error when copying to the receive buffer, 1198 * usually because we weren't able to allocate 1199 * additional buffer pages. All we can do now 1200 * is turn off TCP_RCV_COPY_DATA, so the request 1201 * will not receive any additional updates, 1202 * and time out. 1203 * Any remaining data from this record will 1204 * be discarded. 1205 */ 1206 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1207 dprintk("RPC: XID %08x truncated request\n", 1208 ntohl(transport->tcp_xid)); 1209 dprintk("RPC: xprt = %p, tcp_copied = %lu, " 1210 "tcp_offset = %u, tcp_reclen = %u\n", 1211 xprt, transport->tcp_copied, 1212 transport->tcp_offset, transport->tcp_reclen); 1213 return; 1214 } 1215 1216 dprintk("RPC: XID %08x read %Zd bytes\n", 1217 ntohl(transport->tcp_xid), r); 1218 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " 1219 "tcp_reclen = %u\n", xprt, transport->tcp_copied, 1220 transport->tcp_offset, transport->tcp_reclen); 1221 1222 if (transport->tcp_copied == req->rq_private_buf.buflen) 1223 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1224 else if (transport->tcp_offset == transport->tcp_reclen) { 1225 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) 1226 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1227 } 1228 } 1229 1230 /* 1231 * Finds the request corresponding to the RPC xid and invokes the common 1232 * tcp read code to read the data. 1233 */ 1234 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt, 1235 struct xdr_skb_reader *desc) 1236 { 1237 struct sock_xprt *transport = 1238 container_of(xprt, struct sock_xprt, xprt); 1239 struct rpc_rqst *req; 1240 1241 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid)); 1242 1243 /* Find and lock the request corresponding to this xid */ 1244 spin_lock(&xprt->transport_lock); 1245 req = xprt_lookup_rqst(xprt, transport->tcp_xid); 1246 if (!req) { 1247 dprintk("RPC: XID %08x request not found!\n", 1248 ntohl(transport->tcp_xid)); 1249 spin_unlock(&xprt->transport_lock); 1250 return -1; 1251 } 1252 1253 xs_tcp_read_common(xprt, desc, req); 1254 1255 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1256 xprt_complete_rqst(req->rq_task, transport->tcp_copied); 1257 1258 spin_unlock(&xprt->transport_lock); 1259 return 0; 1260 } 1261 1262 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1263 /* 1264 * Obtains an rpc_rqst previously allocated and invokes the common 1265 * tcp read code to read the data. The result is placed in the callback 1266 * queue. 1267 * If we're unable to obtain the rpc_rqst we schedule the closing of the 1268 * connection and return -1. 1269 */ 1270 static int xs_tcp_read_callback(struct rpc_xprt *xprt, 1271 struct xdr_skb_reader *desc) 1272 { 1273 struct sock_xprt *transport = 1274 container_of(xprt, struct sock_xprt, xprt); 1275 struct rpc_rqst *req; 1276 1277 /* Look up and lock the request corresponding to the given XID */ 1278 spin_lock(&xprt->transport_lock); 1279 req = xprt_lookup_bc_request(xprt, transport->tcp_xid); 1280 if (req == NULL) { 1281 spin_unlock(&xprt->transport_lock); 1282 printk(KERN_WARNING "Callback slot table overflowed\n"); 1283 xprt_force_disconnect(xprt); 1284 return -1; 1285 } 1286 1287 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid)); 1288 xs_tcp_read_common(xprt, desc, req); 1289 1290 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1291 xprt_complete_bc_request(req, transport->tcp_copied); 1292 spin_unlock(&xprt->transport_lock); 1293 1294 return 0; 1295 } 1296 1297 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1298 struct xdr_skb_reader *desc) 1299 { 1300 struct sock_xprt *transport = 1301 container_of(xprt, struct sock_xprt, xprt); 1302 1303 return (transport->tcp_flags & TCP_RPC_REPLY) ? 1304 xs_tcp_read_reply(xprt, desc) : 1305 xs_tcp_read_callback(xprt, desc); 1306 } 1307 #else 1308 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1309 struct xdr_skb_reader *desc) 1310 { 1311 return xs_tcp_read_reply(xprt, desc); 1312 } 1313 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1314 1315 /* 1316 * Read data off the transport. This can be either an RPC_CALL or an 1317 * RPC_REPLY. Relay the processing to helper functions. 1318 */ 1319 static void xs_tcp_read_data(struct rpc_xprt *xprt, 1320 struct xdr_skb_reader *desc) 1321 { 1322 struct sock_xprt *transport = 1323 container_of(xprt, struct sock_xprt, xprt); 1324 1325 if (_xs_tcp_read_data(xprt, desc) == 0) 1326 xs_tcp_check_fraghdr(transport); 1327 else { 1328 /* 1329 * The transport_lock protects the request handling. 1330 * There's no need to hold it to update the tcp_flags. 1331 */ 1332 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1333 } 1334 } 1335 1336 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1337 { 1338 size_t len; 1339 1340 len = transport->tcp_reclen - transport->tcp_offset; 1341 if (len > desc->count) 1342 len = desc->count; 1343 desc->count -= len; 1344 desc->offset += len; 1345 transport->tcp_offset += len; 1346 dprintk("RPC: discarded %Zu bytes\n", len); 1347 xs_tcp_check_fraghdr(transport); 1348 } 1349 1350 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) 1351 { 1352 struct rpc_xprt *xprt = rd_desc->arg.data; 1353 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1354 struct xdr_skb_reader desc = { 1355 .skb = skb, 1356 .offset = offset, 1357 .count = len, 1358 }; 1359 1360 dprintk("RPC: xs_tcp_data_recv started\n"); 1361 do { 1362 trace_xs_tcp_data_recv(transport); 1363 /* Read in a new fragment marker if necessary */ 1364 /* Can we ever really expect to get completely empty fragments? */ 1365 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { 1366 xs_tcp_read_fraghdr(xprt, &desc); 1367 continue; 1368 } 1369 /* Read in the xid if necessary */ 1370 if (transport->tcp_flags & TCP_RCV_COPY_XID) { 1371 xs_tcp_read_xid(transport, &desc); 1372 continue; 1373 } 1374 /* Read in the call/reply flag */ 1375 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) { 1376 xs_tcp_read_calldir(transport, &desc); 1377 continue; 1378 } 1379 /* Read in the request data */ 1380 if (transport->tcp_flags & TCP_RCV_COPY_DATA) { 1381 xs_tcp_read_data(xprt, &desc); 1382 continue; 1383 } 1384 /* Skip over any trailing bytes on short reads */ 1385 xs_tcp_read_discard(transport, &desc); 1386 } while (desc.count); 1387 trace_xs_tcp_data_recv(transport); 1388 dprintk("RPC: xs_tcp_data_recv done\n"); 1389 return len - desc.count; 1390 } 1391 1392 /** 1393 * xs_tcp_data_ready - "data ready" callback for TCP sockets 1394 * @sk: socket with data to read 1395 * 1396 */ 1397 static void xs_tcp_data_ready(struct sock *sk) 1398 { 1399 struct rpc_xprt *xprt; 1400 read_descriptor_t rd_desc; 1401 int read; 1402 unsigned long total = 0; 1403 1404 dprintk("RPC: xs_tcp_data_ready...\n"); 1405 1406 read_lock_bh(&sk->sk_callback_lock); 1407 if (!(xprt = xprt_from_sock(sk))) { 1408 read = 0; 1409 goto out; 1410 } 1411 /* Any data means we had a useful conversation, so 1412 * the we don't need to delay the next reconnect 1413 */ 1414 if (xprt->reestablish_timeout) 1415 xprt->reestablish_timeout = 0; 1416 1417 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ 1418 rd_desc.arg.data = xprt; 1419 do { 1420 rd_desc.count = 65536; 1421 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); 1422 if (read > 0) 1423 total += read; 1424 } while (read > 0); 1425 out: 1426 trace_xs_tcp_data_ready(xprt, read, total); 1427 read_unlock_bh(&sk->sk_callback_lock); 1428 } 1429 1430 /** 1431 * xs_tcp_state_change - callback to handle TCP socket state changes 1432 * @sk: socket whose state has changed 1433 * 1434 */ 1435 static void xs_tcp_state_change(struct sock *sk) 1436 { 1437 struct rpc_xprt *xprt; 1438 1439 read_lock_bh(&sk->sk_callback_lock); 1440 if (!(xprt = xprt_from_sock(sk))) 1441 goto out; 1442 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1443 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1444 sk->sk_state, xprt_connected(xprt), 1445 sock_flag(sk, SOCK_DEAD), 1446 sock_flag(sk, SOCK_ZAPPED), 1447 sk->sk_shutdown); 1448 1449 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1450 switch (sk->sk_state) { 1451 case TCP_ESTABLISHED: 1452 spin_lock(&xprt->transport_lock); 1453 if (!xprt_test_and_set_connected(xprt)) { 1454 struct sock_xprt *transport = container_of(xprt, 1455 struct sock_xprt, xprt); 1456 1457 /* Reset TCP record info */ 1458 transport->tcp_offset = 0; 1459 transport->tcp_reclen = 0; 1460 transport->tcp_copied = 0; 1461 transport->tcp_flags = 1462 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; 1463 xprt->connect_cookie++; 1464 1465 xprt_wake_pending_tasks(xprt, -EAGAIN); 1466 } 1467 spin_unlock(&xprt->transport_lock); 1468 break; 1469 case TCP_FIN_WAIT1: 1470 /* The client initiated a shutdown of the socket */ 1471 xprt->connect_cookie++; 1472 xprt->reestablish_timeout = 0; 1473 set_bit(XPRT_CLOSING, &xprt->state); 1474 smp_mb__before_atomic(); 1475 clear_bit(XPRT_CONNECTED, &xprt->state); 1476 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1477 smp_mb__after_atomic(); 1478 break; 1479 case TCP_CLOSE_WAIT: 1480 /* The server initiated a shutdown of the socket */ 1481 xprt->connect_cookie++; 1482 clear_bit(XPRT_CONNECTED, &xprt->state); 1483 xs_tcp_force_close(xprt); 1484 case TCP_CLOSING: 1485 /* 1486 * If the server closed down the connection, make sure that 1487 * we back off before reconnecting 1488 */ 1489 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1490 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1491 break; 1492 case TCP_LAST_ACK: 1493 set_bit(XPRT_CLOSING, &xprt->state); 1494 smp_mb__before_atomic(); 1495 clear_bit(XPRT_CONNECTED, &xprt->state); 1496 smp_mb__after_atomic(); 1497 break; 1498 case TCP_CLOSE: 1499 xs_sock_mark_closed(xprt); 1500 } 1501 out: 1502 read_unlock_bh(&sk->sk_callback_lock); 1503 } 1504 1505 static void xs_write_space(struct sock *sk) 1506 { 1507 struct socket *sock; 1508 struct rpc_xprt *xprt; 1509 1510 if (unlikely(!(sock = sk->sk_socket))) 1511 return; 1512 clear_bit(SOCK_NOSPACE, &sock->flags); 1513 1514 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1515 return; 1516 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) 1517 return; 1518 1519 xprt_write_space(xprt); 1520 } 1521 1522 /** 1523 * xs_udp_write_space - callback invoked when socket buffer space 1524 * becomes available 1525 * @sk: socket whose state has changed 1526 * 1527 * Called when more output buffer space is available for this socket. 1528 * We try not to wake our writers until they can make "significant" 1529 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1530 * with a bunch of small requests. 1531 */ 1532 static void xs_udp_write_space(struct sock *sk) 1533 { 1534 read_lock_bh(&sk->sk_callback_lock); 1535 1536 /* from net/core/sock.c:sock_def_write_space */ 1537 if (sock_writeable(sk)) 1538 xs_write_space(sk); 1539 1540 read_unlock_bh(&sk->sk_callback_lock); 1541 } 1542 1543 /** 1544 * xs_tcp_write_space - callback invoked when socket buffer space 1545 * becomes available 1546 * @sk: socket whose state has changed 1547 * 1548 * Called when more output buffer space is available for this socket. 1549 * We try not to wake our writers until they can make "significant" 1550 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1551 * with a bunch of small requests. 1552 */ 1553 static void xs_tcp_write_space(struct sock *sk) 1554 { 1555 read_lock_bh(&sk->sk_callback_lock); 1556 1557 /* from net/core/stream.c:sk_stream_write_space */ 1558 if (sk_stream_is_writeable(sk)) 1559 xs_write_space(sk); 1560 1561 read_unlock_bh(&sk->sk_callback_lock); 1562 } 1563 1564 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1565 { 1566 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1567 struct sock *sk = transport->inet; 1568 1569 if (transport->rcvsize) { 1570 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1571 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1572 } 1573 if (transport->sndsize) { 1574 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1575 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1576 sk->sk_write_space(sk); 1577 } 1578 } 1579 1580 /** 1581 * xs_udp_set_buffer_size - set send and receive limits 1582 * @xprt: generic transport 1583 * @sndsize: requested size of send buffer, in bytes 1584 * @rcvsize: requested size of receive buffer, in bytes 1585 * 1586 * Set socket send and receive buffer size limits. 1587 */ 1588 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1589 { 1590 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1591 1592 transport->sndsize = 0; 1593 if (sndsize) 1594 transport->sndsize = sndsize + 1024; 1595 transport->rcvsize = 0; 1596 if (rcvsize) 1597 transport->rcvsize = rcvsize + 1024; 1598 1599 xs_udp_do_set_buffer_size(xprt); 1600 } 1601 1602 /** 1603 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1604 * @task: task that timed out 1605 * 1606 * Adjust the congestion window after a retransmit timeout has occurred. 1607 */ 1608 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1609 { 1610 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1611 } 1612 1613 static unsigned short xs_get_random_port(void) 1614 { 1615 unsigned short range = xprt_max_resvport - xprt_min_resvport; 1616 unsigned short rand = (unsigned short) prandom_u32() % range; 1617 return rand + xprt_min_resvport; 1618 } 1619 1620 /** 1621 * xs_set_reuseaddr_port - set the socket's port and address reuse options 1622 * @sock: socket 1623 * 1624 * Note that this function has to be called on all sockets that share the 1625 * same port, and it must be called before binding. 1626 */ 1627 static void xs_sock_set_reuseport(struct socket *sock) 1628 { 1629 int opt = 1; 1630 1631 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, 1632 (char *)&opt, sizeof(opt)); 1633 } 1634 1635 static unsigned short xs_sock_getport(struct socket *sock) 1636 { 1637 struct sockaddr_storage buf; 1638 int buflen; 1639 unsigned short port = 0; 1640 1641 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0) 1642 goto out; 1643 switch (buf.ss_family) { 1644 case AF_INET6: 1645 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1646 break; 1647 case AF_INET: 1648 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1649 } 1650 out: 1651 return port; 1652 } 1653 1654 /** 1655 * xs_set_port - reset the port number in the remote endpoint address 1656 * @xprt: generic transport 1657 * @port: new port number 1658 * 1659 */ 1660 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1661 { 1662 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1663 1664 rpc_set_port(xs_addr(xprt), port); 1665 xs_update_peer_port(xprt); 1666 } 1667 1668 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1669 { 1670 if (transport->srcport == 0) 1671 transport->srcport = xs_sock_getport(sock); 1672 } 1673 1674 static unsigned short xs_get_srcport(struct sock_xprt *transport) 1675 { 1676 unsigned short port = transport->srcport; 1677 1678 if (port == 0 && transport->xprt.resvport) 1679 port = xs_get_random_port(); 1680 return port; 1681 } 1682 1683 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1684 { 1685 if (transport->srcport != 0) 1686 transport->srcport = 0; 1687 if (!transport->xprt.resvport) 1688 return 0; 1689 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1690 return xprt_max_resvport; 1691 return --port; 1692 } 1693 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1694 { 1695 struct sockaddr_storage myaddr; 1696 int err, nloop = 0; 1697 unsigned short port = xs_get_srcport(transport); 1698 unsigned short last; 1699 1700 /* 1701 * If we are asking for any ephemeral port (i.e. port == 0 && 1702 * transport->xprt.resvport == 0), don't bind. Let the local 1703 * port selection happen implicitly when the socket is used 1704 * (for example at connect time). 1705 * 1706 * This ensures that we can continue to establish TCP 1707 * connections even when all local ephemeral ports are already 1708 * a part of some TCP connection. This makes no difference 1709 * for UDP sockets, but also doens't harm them. 1710 * 1711 * If we're asking for any reserved port (i.e. port == 0 && 1712 * transport->xprt.resvport == 1) xs_get_srcport above will 1713 * ensure that port is non-zero and we will bind as needed. 1714 */ 1715 if (port == 0) 1716 return 0; 1717 1718 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1719 do { 1720 rpc_set_port((struct sockaddr *)&myaddr, port); 1721 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1722 transport->xprt.addrlen); 1723 if (err == 0) { 1724 transport->srcport = port; 1725 break; 1726 } 1727 last = port; 1728 port = xs_next_srcport(transport, port); 1729 if (port > last) 1730 nloop++; 1731 } while (err == -EADDRINUSE && nloop != 2); 1732 1733 if (myaddr.ss_family == AF_INET) 1734 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1735 &((struct sockaddr_in *)&myaddr)->sin_addr, 1736 port, err ? "failed" : "ok", err); 1737 else 1738 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1739 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1740 port, err ? "failed" : "ok", err); 1741 return err; 1742 } 1743 1744 /* 1745 * We don't support autobind on AF_LOCAL sockets 1746 */ 1747 static void xs_local_rpcbind(struct rpc_task *task) 1748 { 1749 rcu_read_lock(); 1750 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt)); 1751 rcu_read_unlock(); 1752 } 1753 1754 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1755 { 1756 } 1757 1758 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1759 static struct lock_class_key xs_key[2]; 1760 static struct lock_class_key xs_slock_key[2]; 1761 1762 static inline void xs_reclassify_socketu(struct socket *sock) 1763 { 1764 struct sock *sk = sock->sk; 1765 1766 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1767 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); 1768 } 1769 1770 static inline void xs_reclassify_socket4(struct socket *sock) 1771 { 1772 struct sock *sk = sock->sk; 1773 1774 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1775 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1776 } 1777 1778 static inline void xs_reclassify_socket6(struct socket *sock) 1779 { 1780 struct sock *sk = sock->sk; 1781 1782 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1783 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1784 } 1785 1786 static inline void xs_reclassify_socket(int family, struct socket *sock) 1787 { 1788 WARN_ON_ONCE(sock_owned_by_user(sock->sk)); 1789 if (sock_owned_by_user(sock->sk)) 1790 return; 1791 1792 switch (family) { 1793 case AF_LOCAL: 1794 xs_reclassify_socketu(sock); 1795 break; 1796 case AF_INET: 1797 xs_reclassify_socket4(sock); 1798 break; 1799 case AF_INET6: 1800 xs_reclassify_socket6(sock); 1801 break; 1802 } 1803 } 1804 #else 1805 static inline void xs_reclassify_socketu(struct socket *sock) 1806 { 1807 } 1808 1809 static inline void xs_reclassify_socket4(struct socket *sock) 1810 { 1811 } 1812 1813 static inline void xs_reclassify_socket6(struct socket *sock) 1814 { 1815 } 1816 1817 static inline void xs_reclassify_socket(int family, struct socket *sock) 1818 { 1819 } 1820 #endif 1821 1822 static void xs_dummy_setup_socket(struct work_struct *work) 1823 { 1824 } 1825 1826 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1827 struct sock_xprt *transport, int family, int type, 1828 int protocol, bool reuseport) 1829 { 1830 struct socket *sock; 1831 int err; 1832 1833 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1834 if (err < 0) { 1835 dprintk("RPC: can't create %d transport socket (%d).\n", 1836 protocol, -err); 1837 goto out; 1838 } 1839 xs_reclassify_socket(family, sock); 1840 1841 if (reuseport) 1842 xs_sock_set_reuseport(sock); 1843 1844 err = xs_bind(transport, sock); 1845 if (err) { 1846 sock_release(sock); 1847 goto out; 1848 } 1849 1850 return sock; 1851 out: 1852 return ERR_PTR(err); 1853 } 1854 1855 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1856 struct socket *sock) 1857 { 1858 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1859 xprt); 1860 1861 if (!transport->inet) { 1862 struct sock *sk = sock->sk; 1863 1864 write_lock_bh(&sk->sk_callback_lock); 1865 1866 xs_save_old_callbacks(transport, sk); 1867 1868 sk->sk_user_data = xprt; 1869 sk->sk_data_ready = xs_local_data_ready; 1870 sk->sk_write_space = xs_udp_write_space; 1871 sk->sk_error_report = xs_error_report; 1872 sk->sk_allocation = GFP_NOIO; 1873 1874 xprt_clear_connected(xprt); 1875 1876 /* Reset to new socket */ 1877 transport->sock = sock; 1878 transport->inet = sk; 1879 1880 write_unlock_bh(&sk->sk_callback_lock); 1881 } 1882 1883 /* Tell the socket layer to start connecting... */ 1884 xprt->stat.connect_count++; 1885 xprt->stat.connect_start = jiffies; 1886 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1887 } 1888 1889 /** 1890 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1891 * @transport: socket transport to connect 1892 */ 1893 static int xs_local_setup_socket(struct sock_xprt *transport) 1894 { 1895 struct rpc_xprt *xprt = &transport->xprt; 1896 struct socket *sock; 1897 int status = -EIO; 1898 1899 status = __sock_create(xprt->xprt_net, AF_LOCAL, 1900 SOCK_STREAM, 0, &sock, 1); 1901 if (status < 0) { 1902 dprintk("RPC: can't create AF_LOCAL " 1903 "transport socket (%d).\n", -status); 1904 goto out; 1905 } 1906 xs_reclassify_socketu(sock); 1907 1908 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 1909 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1910 1911 status = xs_local_finish_connecting(xprt, sock); 1912 trace_rpc_socket_connect(xprt, sock, status); 1913 switch (status) { 1914 case 0: 1915 dprintk("RPC: xprt %p connected to %s\n", 1916 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1917 xprt_set_connected(xprt); 1918 case -ENOBUFS: 1919 break; 1920 case -ENOENT: 1921 dprintk("RPC: xprt %p: socket %s does not exist\n", 1922 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1923 break; 1924 case -ECONNREFUSED: 1925 dprintk("RPC: xprt %p: connection refused for %s\n", 1926 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 1927 break; 1928 default: 1929 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 1930 __func__, -status, 1931 xprt->address_strings[RPC_DISPLAY_ADDR]); 1932 } 1933 1934 out: 1935 xprt_clear_connecting(xprt); 1936 xprt_wake_pending_tasks(xprt, status); 1937 return status; 1938 } 1939 1940 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 1941 { 1942 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1943 int ret; 1944 1945 if (RPC_IS_ASYNC(task)) { 1946 /* 1947 * We want the AF_LOCAL connect to be resolved in the 1948 * filesystem namespace of the process making the rpc 1949 * call. Thus we connect synchronously. 1950 * 1951 * If we want to support asynchronous AF_LOCAL calls, 1952 * we'll need to figure out how to pass a namespace to 1953 * connect. 1954 */ 1955 rpc_exit(task, -ENOTCONN); 1956 return; 1957 } 1958 ret = xs_local_setup_socket(transport); 1959 if (ret && !RPC_IS_SOFTCONN(task)) 1960 msleep_interruptible(15000); 1961 } 1962 1963 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 1964 /* 1965 * Note that this should be called with XPRT_LOCKED held (or when we otherwise 1966 * know that we have exclusive access to the socket), to guard against 1967 * races with xs_reset_transport. 1968 */ 1969 static void xs_set_memalloc(struct rpc_xprt *xprt) 1970 { 1971 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1972 xprt); 1973 1974 /* 1975 * If there's no sock, then we have nothing to set. The 1976 * reconnecting process will get it for us. 1977 */ 1978 if (!transport->inet) 1979 return; 1980 if (atomic_read(&xprt->swapper)) 1981 sk_set_memalloc(transport->inet); 1982 } 1983 1984 /** 1985 * xs_enable_swap - Tag this transport as being used for swap. 1986 * @xprt: transport to tag 1987 * 1988 * Take a reference to this transport on behalf of the rpc_clnt, and 1989 * optionally mark it for swapping if it wasn't already. 1990 */ 1991 static int 1992 xs_enable_swap(struct rpc_xprt *xprt) 1993 { 1994 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 1995 1996 if (atomic_inc_return(&xprt->swapper) != 1) 1997 return 0; 1998 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 1999 return -ERESTARTSYS; 2000 if (xs->inet) 2001 sk_set_memalloc(xs->inet); 2002 xprt_release_xprt(xprt, NULL); 2003 return 0; 2004 } 2005 2006 /** 2007 * xs_disable_swap - Untag this transport as being used for swap. 2008 * @xprt: transport to tag 2009 * 2010 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2011 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2012 */ 2013 static void 2014 xs_disable_swap(struct rpc_xprt *xprt) 2015 { 2016 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2017 2018 if (!atomic_dec_and_test(&xprt->swapper)) 2019 return; 2020 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 2021 return; 2022 if (xs->inet) 2023 sk_clear_memalloc(xs->inet); 2024 xprt_release_xprt(xprt, NULL); 2025 } 2026 #else 2027 static void xs_set_memalloc(struct rpc_xprt *xprt) 2028 { 2029 } 2030 2031 static int 2032 xs_enable_swap(struct rpc_xprt *xprt) 2033 { 2034 return -EINVAL; 2035 } 2036 2037 static void 2038 xs_disable_swap(struct rpc_xprt *xprt) 2039 { 2040 } 2041 #endif 2042 2043 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2044 { 2045 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2046 2047 if (!transport->inet) { 2048 struct sock *sk = sock->sk; 2049 2050 write_lock_bh(&sk->sk_callback_lock); 2051 2052 xs_save_old_callbacks(transport, sk); 2053 2054 sk->sk_user_data = xprt; 2055 sk->sk_data_ready = xs_udp_data_ready; 2056 sk->sk_write_space = xs_udp_write_space; 2057 sk->sk_allocation = GFP_NOIO; 2058 2059 xprt_set_connected(xprt); 2060 2061 /* Reset to new socket */ 2062 transport->sock = sock; 2063 transport->inet = sk; 2064 2065 xs_set_memalloc(xprt); 2066 2067 write_unlock_bh(&sk->sk_callback_lock); 2068 } 2069 xs_udp_do_set_buffer_size(xprt); 2070 } 2071 2072 static void xs_udp_setup_socket(struct work_struct *work) 2073 { 2074 struct sock_xprt *transport = 2075 container_of(work, struct sock_xprt, connect_worker.work); 2076 struct rpc_xprt *xprt = &transport->xprt; 2077 struct socket *sock = transport->sock; 2078 int status = -EIO; 2079 2080 sock = xs_create_sock(xprt, transport, 2081 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2082 IPPROTO_UDP, false); 2083 if (IS_ERR(sock)) 2084 goto out; 2085 2086 dprintk("RPC: worker connecting xprt %p via %s to " 2087 "%s (port %s)\n", xprt, 2088 xprt->address_strings[RPC_DISPLAY_PROTO], 2089 xprt->address_strings[RPC_DISPLAY_ADDR], 2090 xprt->address_strings[RPC_DISPLAY_PORT]); 2091 2092 xs_udp_finish_connecting(xprt, sock); 2093 trace_rpc_socket_connect(xprt, sock, 0); 2094 status = 0; 2095 out: 2096 xprt_unlock_connect(xprt, transport); 2097 xprt_clear_connecting(xprt); 2098 xprt_wake_pending_tasks(xprt, status); 2099 } 2100 2101 /** 2102 * xs_tcp_shutdown - gracefully shut down a TCP socket 2103 * @xprt: transport 2104 * 2105 * Initiates a graceful shutdown of the TCP socket by calling the 2106 * equivalent of shutdown(SHUT_RDWR); 2107 */ 2108 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2109 { 2110 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2111 struct socket *sock = transport->sock; 2112 2113 if (sock == NULL) 2114 return; 2115 if (xprt_connected(xprt)) { 2116 kernel_sock_shutdown(sock, SHUT_RDWR); 2117 trace_rpc_socket_shutdown(xprt, sock); 2118 } else 2119 xs_reset_transport(transport); 2120 } 2121 2122 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2123 { 2124 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2125 int ret = -ENOTCONN; 2126 2127 if (!transport->inet) { 2128 struct sock *sk = sock->sk; 2129 unsigned int keepidle = xprt->timeout->to_initval / HZ; 2130 unsigned int keepcnt = xprt->timeout->to_retries + 1; 2131 unsigned int opt_on = 1; 2132 unsigned int timeo; 2133 2134 /* TCP Keepalive options */ 2135 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 2136 (char *)&opt_on, sizeof(opt_on)); 2137 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, 2138 (char *)&keepidle, sizeof(keepidle)); 2139 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, 2140 (char *)&keepidle, sizeof(keepidle)); 2141 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, 2142 (char *)&keepcnt, sizeof(keepcnt)); 2143 2144 /* TCP user timeout (see RFC5482) */ 2145 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2146 (xprt->timeout->to_retries + 1); 2147 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT, 2148 (char *)&timeo, sizeof(timeo)); 2149 2150 write_lock_bh(&sk->sk_callback_lock); 2151 2152 xs_save_old_callbacks(transport, sk); 2153 2154 sk->sk_user_data = xprt; 2155 sk->sk_data_ready = xs_tcp_data_ready; 2156 sk->sk_state_change = xs_tcp_state_change; 2157 sk->sk_write_space = xs_tcp_write_space; 2158 sk->sk_error_report = xs_error_report; 2159 sk->sk_allocation = GFP_NOIO; 2160 2161 /* socket options */ 2162 sock_reset_flag(sk, SOCK_LINGER); 2163 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 2164 2165 xprt_clear_connected(xprt); 2166 2167 /* Reset to new socket */ 2168 transport->sock = sock; 2169 transport->inet = sk; 2170 2171 write_unlock_bh(&sk->sk_callback_lock); 2172 } 2173 2174 if (!xprt_bound(xprt)) 2175 goto out; 2176 2177 xs_set_memalloc(xprt); 2178 2179 /* Tell the socket layer to start connecting... */ 2180 xprt->stat.connect_count++; 2181 xprt->stat.connect_start = jiffies; 2182 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2183 switch (ret) { 2184 case 0: 2185 xs_set_srcport(transport, sock); 2186 case -EINPROGRESS: 2187 /* SYN_SENT! */ 2188 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2189 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2190 } 2191 out: 2192 return ret; 2193 } 2194 2195 /** 2196 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2197 * 2198 * Invoked by a work queue tasklet. 2199 */ 2200 static void xs_tcp_setup_socket(struct work_struct *work) 2201 { 2202 struct sock_xprt *transport = 2203 container_of(work, struct sock_xprt, connect_worker.work); 2204 struct socket *sock = transport->sock; 2205 struct rpc_xprt *xprt = &transport->xprt; 2206 int status = -EIO; 2207 2208 if (!sock) { 2209 sock = xs_create_sock(xprt, transport, 2210 xs_addr(xprt)->sa_family, SOCK_STREAM, 2211 IPPROTO_TCP, true); 2212 if (IS_ERR(sock)) { 2213 status = PTR_ERR(sock); 2214 goto out; 2215 } 2216 } 2217 2218 dprintk("RPC: worker connecting xprt %p via %s to " 2219 "%s (port %s)\n", xprt, 2220 xprt->address_strings[RPC_DISPLAY_PROTO], 2221 xprt->address_strings[RPC_DISPLAY_ADDR], 2222 xprt->address_strings[RPC_DISPLAY_PORT]); 2223 2224 status = xs_tcp_finish_connecting(xprt, sock); 2225 trace_rpc_socket_connect(xprt, sock, status); 2226 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2227 xprt, -status, xprt_connected(xprt), 2228 sock->sk->sk_state); 2229 switch (status) { 2230 default: 2231 printk("%s: connect returned unhandled error %d\n", 2232 __func__, status); 2233 case -EADDRNOTAVAIL: 2234 /* We're probably in TIME_WAIT. Get rid of existing socket, 2235 * and retry 2236 */ 2237 xs_tcp_force_close(xprt); 2238 break; 2239 case 0: 2240 case -EINPROGRESS: 2241 case -EALREADY: 2242 xprt_unlock_connect(xprt, transport); 2243 xprt_clear_connecting(xprt); 2244 return; 2245 case -EINVAL: 2246 /* Happens, for instance, if the user specified a link 2247 * local IPv6 address without a scope-id. 2248 */ 2249 case -ECONNREFUSED: 2250 case -ECONNRESET: 2251 case -ENETUNREACH: 2252 case -EADDRINUSE: 2253 case -ENOBUFS: 2254 /* retry with existing socket, after a delay */ 2255 xs_tcp_force_close(xprt); 2256 goto out; 2257 } 2258 status = -EAGAIN; 2259 out: 2260 xprt_unlock_connect(xprt, transport); 2261 xprt_clear_connecting(xprt); 2262 xprt_wake_pending_tasks(xprt, status); 2263 } 2264 2265 /** 2266 * xs_connect - connect a socket to a remote endpoint 2267 * @xprt: pointer to transport structure 2268 * @task: address of RPC task that manages state of connect request 2269 * 2270 * TCP: If the remote end dropped the connection, delay reconnecting. 2271 * 2272 * UDP socket connects are synchronous, but we use a work queue anyway 2273 * to guarantee that even unprivileged user processes can set up a 2274 * socket on a privileged port. 2275 * 2276 * If a UDP socket connect fails, the delay behavior here prevents 2277 * retry floods (hard mounts). 2278 */ 2279 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2280 { 2281 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2282 2283 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2284 2285 if (transport->sock != NULL) { 2286 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2287 "seconds\n", 2288 xprt, xprt->reestablish_timeout / HZ); 2289 2290 /* Start by resetting any existing state */ 2291 xs_reset_transport(transport); 2292 2293 queue_delayed_work(rpciod_workqueue, 2294 &transport->connect_worker, 2295 xprt->reestablish_timeout); 2296 xprt->reestablish_timeout <<= 1; 2297 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2298 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2299 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) 2300 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; 2301 } else { 2302 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2303 queue_delayed_work(rpciod_workqueue, 2304 &transport->connect_worker, 0); 2305 } 2306 } 2307 2308 /** 2309 * xs_local_print_stats - display AF_LOCAL socket-specifc stats 2310 * @xprt: rpc_xprt struct containing statistics 2311 * @seq: output file 2312 * 2313 */ 2314 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2315 { 2316 long idle_time = 0; 2317 2318 if (xprt_connected(xprt)) 2319 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2320 2321 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2322 "%llu %llu %lu %llu %llu\n", 2323 xprt->stat.bind_count, 2324 xprt->stat.connect_count, 2325 xprt->stat.connect_time, 2326 idle_time, 2327 xprt->stat.sends, 2328 xprt->stat.recvs, 2329 xprt->stat.bad_xids, 2330 xprt->stat.req_u, 2331 xprt->stat.bklog_u, 2332 xprt->stat.max_slots, 2333 xprt->stat.sending_u, 2334 xprt->stat.pending_u); 2335 } 2336 2337 /** 2338 * xs_udp_print_stats - display UDP socket-specifc stats 2339 * @xprt: rpc_xprt struct containing statistics 2340 * @seq: output file 2341 * 2342 */ 2343 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2344 { 2345 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2346 2347 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2348 "%lu %llu %llu\n", 2349 transport->srcport, 2350 xprt->stat.bind_count, 2351 xprt->stat.sends, 2352 xprt->stat.recvs, 2353 xprt->stat.bad_xids, 2354 xprt->stat.req_u, 2355 xprt->stat.bklog_u, 2356 xprt->stat.max_slots, 2357 xprt->stat.sending_u, 2358 xprt->stat.pending_u); 2359 } 2360 2361 /** 2362 * xs_tcp_print_stats - display TCP socket-specifc stats 2363 * @xprt: rpc_xprt struct containing statistics 2364 * @seq: output file 2365 * 2366 */ 2367 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2368 { 2369 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2370 long idle_time = 0; 2371 2372 if (xprt_connected(xprt)) 2373 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2374 2375 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2376 "%llu %llu %lu %llu %llu\n", 2377 transport->srcport, 2378 xprt->stat.bind_count, 2379 xprt->stat.connect_count, 2380 xprt->stat.connect_time, 2381 idle_time, 2382 xprt->stat.sends, 2383 xprt->stat.recvs, 2384 xprt->stat.bad_xids, 2385 xprt->stat.req_u, 2386 xprt->stat.bklog_u, 2387 xprt->stat.max_slots, 2388 xprt->stat.sending_u, 2389 xprt->stat.pending_u); 2390 } 2391 2392 /* 2393 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2394 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2395 * to use the server side send routines. 2396 */ 2397 static void *bc_malloc(struct rpc_task *task, size_t size) 2398 { 2399 struct page *page; 2400 struct rpc_buffer *buf; 2401 2402 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer)); 2403 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) 2404 return NULL; 2405 2406 page = alloc_page(GFP_KERNEL); 2407 if (!page) 2408 return NULL; 2409 2410 buf = page_address(page); 2411 buf->len = PAGE_SIZE; 2412 2413 return buf->data; 2414 } 2415 2416 /* 2417 * Free the space allocated in the bc_alloc routine 2418 */ 2419 static void bc_free(void *buffer) 2420 { 2421 struct rpc_buffer *buf; 2422 2423 if (!buffer) 2424 return; 2425 2426 buf = container_of(buffer, struct rpc_buffer, data); 2427 free_page((unsigned long)buf); 2428 } 2429 2430 /* 2431 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex 2432 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request. 2433 */ 2434 static int bc_sendto(struct rpc_rqst *req) 2435 { 2436 int len; 2437 struct xdr_buf *xbufp = &req->rq_snd_buf; 2438 struct rpc_xprt *xprt = req->rq_xprt; 2439 struct sock_xprt *transport = 2440 container_of(xprt, struct sock_xprt, xprt); 2441 struct socket *sock = transport->sock; 2442 unsigned long headoff; 2443 unsigned long tailoff; 2444 2445 xs_encode_stream_record_marker(xbufp); 2446 2447 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK; 2448 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK; 2449 len = svc_send_common(sock, xbufp, 2450 virt_to_page(xbufp->head[0].iov_base), headoff, 2451 xbufp->tail[0].iov_base, tailoff); 2452 2453 if (len != xbufp->len) { 2454 printk(KERN_NOTICE "Error sending entire callback!\n"); 2455 len = -EAGAIN; 2456 } 2457 2458 return len; 2459 } 2460 2461 /* 2462 * The send routine. Borrows from svc_send 2463 */ 2464 static int bc_send_request(struct rpc_task *task) 2465 { 2466 struct rpc_rqst *req = task->tk_rqstp; 2467 struct svc_xprt *xprt; 2468 u32 len; 2469 2470 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid)); 2471 /* 2472 * Get the server socket associated with this callback xprt 2473 */ 2474 xprt = req->rq_xprt->bc_xprt; 2475 2476 /* 2477 * Grab the mutex to serialize data as the connection is shared 2478 * with the fore channel 2479 */ 2480 if (!mutex_trylock(&xprt->xpt_mutex)) { 2481 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL); 2482 if (!mutex_trylock(&xprt->xpt_mutex)) 2483 return -EAGAIN; 2484 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task); 2485 } 2486 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2487 len = -ENOTCONN; 2488 else 2489 len = bc_sendto(req); 2490 mutex_unlock(&xprt->xpt_mutex); 2491 2492 if (len > 0) 2493 len = 0; 2494 2495 return len; 2496 } 2497 2498 /* 2499 * The close routine. Since this is client initiated, we do nothing 2500 */ 2501 2502 static void bc_close(struct rpc_xprt *xprt) 2503 { 2504 } 2505 2506 /* 2507 * The xprt destroy routine. Again, because this connection is client 2508 * initiated, we do nothing 2509 */ 2510 2511 static void bc_destroy(struct rpc_xprt *xprt) 2512 { 2513 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2514 2515 xs_xprt_free(xprt); 2516 module_put(THIS_MODULE); 2517 } 2518 2519 static struct rpc_xprt_ops xs_local_ops = { 2520 .reserve_xprt = xprt_reserve_xprt, 2521 .release_xprt = xs_tcp_release_xprt, 2522 .alloc_slot = xprt_alloc_slot, 2523 .rpcbind = xs_local_rpcbind, 2524 .set_port = xs_local_set_port, 2525 .connect = xs_local_connect, 2526 .buf_alloc = rpc_malloc, 2527 .buf_free = rpc_free, 2528 .send_request = xs_local_send_request, 2529 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2530 .close = xs_close, 2531 .destroy = xs_destroy, 2532 .print_stats = xs_local_print_stats, 2533 .enable_swap = xs_enable_swap, 2534 .disable_swap = xs_disable_swap, 2535 }; 2536 2537 static struct rpc_xprt_ops xs_udp_ops = { 2538 .set_buffer_size = xs_udp_set_buffer_size, 2539 .reserve_xprt = xprt_reserve_xprt_cong, 2540 .release_xprt = xprt_release_xprt_cong, 2541 .alloc_slot = xprt_alloc_slot, 2542 .rpcbind = rpcb_getport_async, 2543 .set_port = xs_set_port, 2544 .connect = xs_connect, 2545 .buf_alloc = rpc_malloc, 2546 .buf_free = rpc_free, 2547 .send_request = xs_udp_send_request, 2548 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 2549 .timer = xs_udp_timer, 2550 .release_request = xprt_release_rqst_cong, 2551 .close = xs_close, 2552 .destroy = xs_destroy, 2553 .print_stats = xs_udp_print_stats, 2554 .enable_swap = xs_enable_swap, 2555 .disable_swap = xs_disable_swap, 2556 .inject_disconnect = xs_inject_disconnect, 2557 }; 2558 2559 static struct rpc_xprt_ops xs_tcp_ops = { 2560 .reserve_xprt = xprt_reserve_xprt, 2561 .release_xprt = xs_tcp_release_xprt, 2562 .alloc_slot = xprt_lock_and_alloc_slot, 2563 .rpcbind = rpcb_getport_async, 2564 .set_port = xs_set_port, 2565 .connect = xs_connect, 2566 .buf_alloc = rpc_malloc, 2567 .buf_free = rpc_free, 2568 .send_request = xs_tcp_send_request, 2569 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2570 .close = xs_tcp_shutdown, 2571 .destroy = xs_destroy, 2572 .print_stats = xs_tcp_print_stats, 2573 .enable_swap = xs_enable_swap, 2574 .disable_swap = xs_disable_swap, 2575 .inject_disconnect = xs_inject_disconnect, 2576 }; 2577 2578 /* 2579 * The rpc_xprt_ops for the server backchannel 2580 */ 2581 2582 static struct rpc_xprt_ops bc_tcp_ops = { 2583 .reserve_xprt = xprt_reserve_xprt, 2584 .release_xprt = xprt_release_xprt, 2585 .alloc_slot = xprt_alloc_slot, 2586 .buf_alloc = bc_malloc, 2587 .buf_free = bc_free, 2588 .send_request = bc_send_request, 2589 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2590 .close = bc_close, 2591 .destroy = bc_destroy, 2592 .print_stats = xs_tcp_print_stats, 2593 .enable_swap = xs_enable_swap, 2594 .disable_swap = xs_disable_swap, 2595 .inject_disconnect = xs_inject_disconnect, 2596 }; 2597 2598 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2599 { 2600 static const struct sockaddr_in sin = { 2601 .sin_family = AF_INET, 2602 .sin_addr.s_addr = htonl(INADDR_ANY), 2603 }; 2604 static const struct sockaddr_in6 sin6 = { 2605 .sin6_family = AF_INET6, 2606 .sin6_addr = IN6ADDR_ANY_INIT, 2607 }; 2608 2609 switch (family) { 2610 case AF_LOCAL: 2611 break; 2612 case AF_INET: 2613 memcpy(sap, &sin, sizeof(sin)); 2614 break; 2615 case AF_INET6: 2616 memcpy(sap, &sin6, sizeof(sin6)); 2617 break; 2618 default: 2619 dprintk("RPC: %s: Bad address family\n", __func__); 2620 return -EAFNOSUPPORT; 2621 } 2622 return 0; 2623 } 2624 2625 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2626 unsigned int slot_table_size, 2627 unsigned int max_slot_table_size) 2628 { 2629 struct rpc_xprt *xprt; 2630 struct sock_xprt *new; 2631 2632 if (args->addrlen > sizeof(xprt->addr)) { 2633 dprintk("RPC: xs_setup_xprt: address too large\n"); 2634 return ERR_PTR(-EBADF); 2635 } 2636 2637 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2638 max_slot_table_size); 2639 if (xprt == NULL) { 2640 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2641 "rpc_xprt\n"); 2642 return ERR_PTR(-ENOMEM); 2643 } 2644 2645 new = container_of(xprt, struct sock_xprt, xprt); 2646 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2647 xprt->addrlen = args->addrlen; 2648 if (args->srcaddr) 2649 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2650 else { 2651 int err; 2652 err = xs_init_anyaddr(args->dstaddr->sa_family, 2653 (struct sockaddr *)&new->srcaddr); 2654 if (err != 0) { 2655 xprt_free(xprt); 2656 return ERR_PTR(err); 2657 } 2658 } 2659 2660 return xprt; 2661 } 2662 2663 static const struct rpc_timeout xs_local_default_timeout = { 2664 .to_initval = 10 * HZ, 2665 .to_maxval = 10 * HZ, 2666 .to_retries = 2, 2667 }; 2668 2669 /** 2670 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2671 * @args: rpc transport creation arguments 2672 * 2673 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2674 */ 2675 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2676 { 2677 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2678 struct sock_xprt *transport; 2679 struct rpc_xprt *xprt; 2680 struct rpc_xprt *ret; 2681 2682 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2683 xprt_max_tcp_slot_table_entries); 2684 if (IS_ERR(xprt)) 2685 return xprt; 2686 transport = container_of(xprt, struct sock_xprt, xprt); 2687 2688 xprt->prot = 0; 2689 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2690 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2691 2692 xprt->bind_timeout = XS_BIND_TO; 2693 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2694 xprt->idle_timeout = XS_IDLE_DISC_TO; 2695 2696 xprt->ops = &xs_local_ops; 2697 xprt->timeout = &xs_local_default_timeout; 2698 2699 INIT_DELAYED_WORK(&transport->connect_worker, 2700 xs_dummy_setup_socket); 2701 2702 switch (sun->sun_family) { 2703 case AF_LOCAL: 2704 if (sun->sun_path[0] != '/') { 2705 dprintk("RPC: bad AF_LOCAL address: %s\n", 2706 sun->sun_path); 2707 ret = ERR_PTR(-EINVAL); 2708 goto out_err; 2709 } 2710 xprt_set_bound(xprt); 2711 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2712 ret = ERR_PTR(xs_local_setup_socket(transport)); 2713 if (ret) 2714 goto out_err; 2715 break; 2716 default: 2717 ret = ERR_PTR(-EAFNOSUPPORT); 2718 goto out_err; 2719 } 2720 2721 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2722 xprt->address_strings[RPC_DISPLAY_ADDR]); 2723 2724 if (try_module_get(THIS_MODULE)) 2725 return xprt; 2726 ret = ERR_PTR(-EINVAL); 2727 out_err: 2728 xs_xprt_free(xprt); 2729 return ret; 2730 } 2731 2732 static const struct rpc_timeout xs_udp_default_timeout = { 2733 .to_initval = 5 * HZ, 2734 .to_maxval = 30 * HZ, 2735 .to_increment = 5 * HZ, 2736 .to_retries = 5, 2737 }; 2738 2739 /** 2740 * xs_setup_udp - Set up transport to use a UDP socket 2741 * @args: rpc transport creation arguments 2742 * 2743 */ 2744 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2745 { 2746 struct sockaddr *addr = args->dstaddr; 2747 struct rpc_xprt *xprt; 2748 struct sock_xprt *transport; 2749 struct rpc_xprt *ret; 2750 2751 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2752 xprt_udp_slot_table_entries); 2753 if (IS_ERR(xprt)) 2754 return xprt; 2755 transport = container_of(xprt, struct sock_xprt, xprt); 2756 2757 xprt->prot = IPPROTO_UDP; 2758 xprt->tsh_size = 0; 2759 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2760 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2761 2762 xprt->bind_timeout = XS_BIND_TO; 2763 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2764 xprt->idle_timeout = XS_IDLE_DISC_TO; 2765 2766 xprt->ops = &xs_udp_ops; 2767 2768 xprt->timeout = &xs_udp_default_timeout; 2769 2770 switch (addr->sa_family) { 2771 case AF_INET: 2772 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2773 xprt_set_bound(xprt); 2774 2775 INIT_DELAYED_WORK(&transport->connect_worker, 2776 xs_udp_setup_socket); 2777 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2778 break; 2779 case AF_INET6: 2780 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2781 xprt_set_bound(xprt); 2782 2783 INIT_DELAYED_WORK(&transport->connect_worker, 2784 xs_udp_setup_socket); 2785 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2786 break; 2787 default: 2788 ret = ERR_PTR(-EAFNOSUPPORT); 2789 goto out_err; 2790 } 2791 2792 if (xprt_bound(xprt)) 2793 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2794 xprt->address_strings[RPC_DISPLAY_ADDR], 2795 xprt->address_strings[RPC_DISPLAY_PORT], 2796 xprt->address_strings[RPC_DISPLAY_PROTO]); 2797 else 2798 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2799 xprt->address_strings[RPC_DISPLAY_ADDR], 2800 xprt->address_strings[RPC_DISPLAY_PROTO]); 2801 2802 if (try_module_get(THIS_MODULE)) 2803 return xprt; 2804 ret = ERR_PTR(-EINVAL); 2805 out_err: 2806 xs_xprt_free(xprt); 2807 return ret; 2808 } 2809 2810 static const struct rpc_timeout xs_tcp_default_timeout = { 2811 .to_initval = 60 * HZ, 2812 .to_maxval = 60 * HZ, 2813 .to_retries = 2, 2814 }; 2815 2816 /** 2817 * xs_setup_tcp - Set up transport to use a TCP socket 2818 * @args: rpc transport creation arguments 2819 * 2820 */ 2821 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2822 { 2823 struct sockaddr *addr = args->dstaddr; 2824 struct rpc_xprt *xprt; 2825 struct sock_xprt *transport; 2826 struct rpc_xprt *ret; 2827 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2828 2829 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2830 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2831 2832 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2833 max_slot_table_size); 2834 if (IS_ERR(xprt)) 2835 return xprt; 2836 transport = container_of(xprt, struct sock_xprt, xprt); 2837 2838 xprt->prot = IPPROTO_TCP; 2839 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2840 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2841 2842 xprt->bind_timeout = XS_BIND_TO; 2843 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2844 xprt->idle_timeout = XS_IDLE_DISC_TO; 2845 2846 xprt->ops = &xs_tcp_ops; 2847 xprt->timeout = &xs_tcp_default_timeout; 2848 2849 switch (addr->sa_family) { 2850 case AF_INET: 2851 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2852 xprt_set_bound(xprt); 2853 2854 INIT_DELAYED_WORK(&transport->connect_worker, 2855 xs_tcp_setup_socket); 2856 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2857 break; 2858 case AF_INET6: 2859 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2860 xprt_set_bound(xprt); 2861 2862 INIT_DELAYED_WORK(&transport->connect_worker, 2863 xs_tcp_setup_socket); 2864 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 2865 break; 2866 default: 2867 ret = ERR_PTR(-EAFNOSUPPORT); 2868 goto out_err; 2869 } 2870 2871 if (xprt_bound(xprt)) 2872 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2873 xprt->address_strings[RPC_DISPLAY_ADDR], 2874 xprt->address_strings[RPC_DISPLAY_PORT], 2875 xprt->address_strings[RPC_DISPLAY_PROTO]); 2876 else 2877 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2878 xprt->address_strings[RPC_DISPLAY_ADDR], 2879 xprt->address_strings[RPC_DISPLAY_PROTO]); 2880 2881 if (try_module_get(THIS_MODULE)) 2882 return xprt; 2883 ret = ERR_PTR(-EINVAL); 2884 out_err: 2885 xs_xprt_free(xprt); 2886 return ret; 2887 } 2888 2889 /** 2890 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 2891 * @args: rpc transport creation arguments 2892 * 2893 */ 2894 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 2895 { 2896 struct sockaddr *addr = args->dstaddr; 2897 struct rpc_xprt *xprt; 2898 struct sock_xprt *transport; 2899 struct svc_sock *bc_sock; 2900 struct rpc_xprt *ret; 2901 2902 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2903 xprt_tcp_slot_table_entries); 2904 if (IS_ERR(xprt)) 2905 return xprt; 2906 transport = container_of(xprt, struct sock_xprt, xprt); 2907 2908 xprt->prot = IPPROTO_TCP; 2909 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2910 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2911 xprt->timeout = &xs_tcp_default_timeout; 2912 2913 /* backchannel */ 2914 xprt_set_bound(xprt); 2915 xprt->bind_timeout = 0; 2916 xprt->reestablish_timeout = 0; 2917 xprt->idle_timeout = 0; 2918 2919 xprt->ops = &bc_tcp_ops; 2920 2921 switch (addr->sa_family) { 2922 case AF_INET: 2923 xs_format_peer_addresses(xprt, "tcp", 2924 RPCBIND_NETID_TCP); 2925 break; 2926 case AF_INET6: 2927 xs_format_peer_addresses(xprt, "tcp", 2928 RPCBIND_NETID_TCP6); 2929 break; 2930 default: 2931 ret = ERR_PTR(-EAFNOSUPPORT); 2932 goto out_err; 2933 } 2934 2935 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2936 xprt->address_strings[RPC_DISPLAY_ADDR], 2937 xprt->address_strings[RPC_DISPLAY_PORT], 2938 xprt->address_strings[RPC_DISPLAY_PROTO]); 2939 2940 /* 2941 * Once we've associated a backchannel xprt with a connection, 2942 * we want to keep it around as long as the connection lasts, 2943 * in case we need to start using it for a backchannel again; 2944 * this reference won't be dropped until bc_xprt is destroyed. 2945 */ 2946 xprt_get(xprt); 2947 args->bc_xprt->xpt_bc_xprt = xprt; 2948 xprt->bc_xprt = args->bc_xprt; 2949 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 2950 transport->sock = bc_sock->sk_sock; 2951 transport->inet = bc_sock->sk_sk; 2952 2953 /* 2954 * Since we don't want connections for the backchannel, we set 2955 * the xprt status to connected 2956 */ 2957 xprt_set_connected(xprt); 2958 2959 if (try_module_get(THIS_MODULE)) 2960 return xprt; 2961 2962 args->bc_xprt->xpt_bc_xprt = NULL; 2963 xprt_put(xprt); 2964 ret = ERR_PTR(-EINVAL); 2965 out_err: 2966 xs_xprt_free(xprt); 2967 return ret; 2968 } 2969 2970 static struct xprt_class xs_local_transport = { 2971 .list = LIST_HEAD_INIT(xs_local_transport.list), 2972 .name = "named UNIX socket", 2973 .owner = THIS_MODULE, 2974 .ident = XPRT_TRANSPORT_LOCAL, 2975 .setup = xs_setup_local, 2976 }; 2977 2978 static struct xprt_class xs_udp_transport = { 2979 .list = LIST_HEAD_INIT(xs_udp_transport.list), 2980 .name = "udp", 2981 .owner = THIS_MODULE, 2982 .ident = XPRT_TRANSPORT_UDP, 2983 .setup = xs_setup_udp, 2984 }; 2985 2986 static struct xprt_class xs_tcp_transport = { 2987 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 2988 .name = "tcp", 2989 .owner = THIS_MODULE, 2990 .ident = XPRT_TRANSPORT_TCP, 2991 .setup = xs_setup_tcp, 2992 }; 2993 2994 static struct xprt_class xs_bc_tcp_transport = { 2995 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 2996 .name = "tcp NFSv4.1 backchannel", 2997 .owner = THIS_MODULE, 2998 .ident = XPRT_TRANSPORT_BC_TCP, 2999 .setup = xs_setup_bc_tcp, 3000 }; 3001 3002 /** 3003 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3004 * 3005 */ 3006 int init_socket_xprt(void) 3007 { 3008 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 3009 if (!sunrpc_table_header) 3010 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3011 #endif 3012 3013 xprt_register_transport(&xs_local_transport); 3014 xprt_register_transport(&xs_udp_transport); 3015 xprt_register_transport(&xs_tcp_transport); 3016 xprt_register_transport(&xs_bc_tcp_transport); 3017 3018 return 0; 3019 } 3020 3021 /** 3022 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3023 * 3024 */ 3025 void cleanup_socket_xprt(void) 3026 { 3027 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 3028 if (sunrpc_table_header) { 3029 unregister_sysctl_table(sunrpc_table_header); 3030 sunrpc_table_header = NULL; 3031 } 3032 #endif 3033 3034 xprt_unregister_transport(&xs_local_transport); 3035 xprt_unregister_transport(&xs_udp_transport); 3036 xprt_unregister_transport(&xs_tcp_transport); 3037 xprt_unregister_transport(&xs_bc_tcp_transport); 3038 } 3039 3040 static int param_set_uint_minmax(const char *val, 3041 const struct kernel_param *kp, 3042 unsigned int min, unsigned int max) 3043 { 3044 unsigned int num; 3045 int ret; 3046 3047 if (!val) 3048 return -EINVAL; 3049 ret = kstrtouint(val, 0, &num); 3050 if (ret == -EINVAL || num < min || num > max) 3051 return -EINVAL; 3052 *((unsigned int *)kp->arg) = num; 3053 return 0; 3054 } 3055 3056 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3057 { 3058 return param_set_uint_minmax(val, kp, 3059 RPC_MIN_RESVPORT, 3060 RPC_MAX_RESVPORT); 3061 } 3062 3063 static const struct kernel_param_ops param_ops_portnr = { 3064 .set = param_set_portnr, 3065 .get = param_get_uint, 3066 }; 3067 3068 #define param_check_portnr(name, p) \ 3069 __param_check(name, p, unsigned int); 3070 3071 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3072 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3073 3074 static int param_set_slot_table_size(const char *val, 3075 const struct kernel_param *kp) 3076 { 3077 return param_set_uint_minmax(val, kp, 3078 RPC_MIN_SLOT_TABLE, 3079 RPC_MAX_SLOT_TABLE); 3080 } 3081 3082 static const struct kernel_param_ops param_ops_slot_table_size = { 3083 .set = param_set_slot_table_size, 3084 .get = param_get_uint, 3085 }; 3086 3087 #define param_check_slot_table_size(name, p) \ 3088 __param_check(name, p, unsigned int); 3089 3090 static int param_set_max_slot_table_size(const char *val, 3091 const struct kernel_param *kp) 3092 { 3093 return param_set_uint_minmax(val, kp, 3094 RPC_MIN_SLOT_TABLE, 3095 RPC_MAX_SLOT_TABLE_LIMIT); 3096 } 3097 3098 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3099 .set = param_set_max_slot_table_size, 3100 .get = param_get_uint, 3101 }; 3102 3103 #define param_check_max_slot_table_size(name, p) \ 3104 __param_check(name, p, unsigned int); 3105 3106 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3107 slot_table_size, 0644); 3108 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3109 max_slot_table_size, 0644); 3110 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3111 slot_table_size, 0644); 3112 3113