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