1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/net/sunrpc/xprtsock.c 4 * 5 * Client-side transport implementation for sockets. 6 * 7 * TCP callback races fixes (C) 1998 Red Hat 8 * TCP send fixes (C) 1998 Red Hat 9 * TCP NFS related read + write fixes 10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 11 * 12 * Rewrite of larges part of the code in order to stabilize TCP stuff. 13 * Fix behaviour when socket buffer is full. 14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> 15 * 16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> 17 * 18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005. 19 * <gilles.quillard@bull.net> 20 */ 21 22 #include <linux/types.h> 23 #include <linux/string.h> 24 #include <linux/slab.h> 25 #include <linux/module.h> 26 #include <linux/capability.h> 27 #include <linux/pagemap.h> 28 #include <linux/errno.h> 29 #include <linux/socket.h> 30 #include <linux/in.h> 31 #include <linux/net.h> 32 #include <linux/mm.h> 33 #include <linux/un.h> 34 #include <linux/udp.h> 35 #include <linux/tcp.h> 36 #include <linux/sunrpc/clnt.h> 37 #include <linux/sunrpc/addr.h> 38 #include <linux/sunrpc/sched.h> 39 #include <linux/sunrpc/svcsock.h> 40 #include <linux/sunrpc/xprtsock.h> 41 #include <linux/file.h> 42 #ifdef CONFIG_SUNRPC_BACKCHANNEL 43 #include <linux/sunrpc/bc_xprt.h> 44 #endif 45 46 #include <net/sock.h> 47 #include <net/checksum.h> 48 #include <net/udp.h> 49 #include <net/tcp.h> 50 51 #include <trace/events/sunrpc.h> 52 53 #include "sunrpc.h" 54 55 #define RPC_TCP_READ_CHUNK_SZ (3*512*1024) 56 57 static void xs_close(struct rpc_xprt *xprt); 58 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 59 struct socket *sock); 60 61 /* 62 * xprtsock tunables 63 */ 64 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 65 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 66 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 67 68 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 69 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 70 71 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 72 73 #define XS_TCP_LINGER_TO (15U * HZ) 74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 75 76 /* 77 * We can register our own files under /proc/sys/sunrpc by 78 * calling register_sysctl_table() again. The files in that 79 * directory become the union of all files registered there. 80 * 81 * We simply need to make sure that we don't collide with 82 * someone else's file names! 83 */ 84 85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 90 91 static struct ctl_table_header *sunrpc_table_header; 92 93 /* 94 * FIXME: changing the UDP slot table size should also resize the UDP 95 * socket buffers for existing UDP transports 96 */ 97 static struct ctl_table xs_tunables_table[] = { 98 { 99 .procname = "udp_slot_table_entries", 100 .data = &xprt_udp_slot_table_entries, 101 .maxlen = sizeof(unsigned int), 102 .mode = 0644, 103 .proc_handler = proc_dointvec_minmax, 104 .extra1 = &min_slot_table_size, 105 .extra2 = &max_slot_table_size 106 }, 107 { 108 .procname = "tcp_slot_table_entries", 109 .data = &xprt_tcp_slot_table_entries, 110 .maxlen = sizeof(unsigned int), 111 .mode = 0644, 112 .proc_handler = proc_dointvec_minmax, 113 .extra1 = &min_slot_table_size, 114 .extra2 = &max_slot_table_size 115 }, 116 { 117 .procname = "tcp_max_slot_table_entries", 118 .data = &xprt_max_tcp_slot_table_entries, 119 .maxlen = sizeof(unsigned int), 120 .mode = 0644, 121 .proc_handler = proc_dointvec_minmax, 122 .extra1 = &min_slot_table_size, 123 .extra2 = &max_tcp_slot_table_limit 124 }, 125 { 126 .procname = "min_resvport", 127 .data = &xprt_min_resvport, 128 .maxlen = sizeof(unsigned int), 129 .mode = 0644, 130 .proc_handler = proc_dointvec_minmax, 131 .extra1 = &xprt_min_resvport_limit, 132 .extra2 = &xprt_max_resvport 133 }, 134 { 135 .procname = "max_resvport", 136 .data = &xprt_max_resvport, 137 .maxlen = sizeof(unsigned int), 138 .mode = 0644, 139 .proc_handler = proc_dointvec_minmax, 140 .extra1 = &xprt_min_resvport, 141 .extra2 = &xprt_max_resvport_limit 142 }, 143 { 144 .procname = "tcp_fin_timeout", 145 .data = &xs_tcp_fin_timeout, 146 .maxlen = sizeof(xs_tcp_fin_timeout), 147 .mode = 0644, 148 .proc_handler = proc_dointvec_jiffies, 149 }, 150 { }, 151 }; 152 153 static struct ctl_table sunrpc_table[] = { 154 { 155 .procname = "sunrpc", 156 .mode = 0555, 157 .child = xs_tunables_table 158 }, 159 { }, 160 }; 161 162 #endif 163 164 /* 165 * Wait duration for a reply from the RPC portmapper. 166 */ 167 #define XS_BIND_TO (60U * HZ) 168 169 /* 170 * Delay if a UDP socket connect error occurs. This is most likely some 171 * kind of resource problem on the local host. 172 */ 173 #define XS_UDP_REEST_TO (2U * HZ) 174 175 /* 176 * The reestablish timeout allows clients to delay for a bit before attempting 177 * to reconnect to a server that just dropped our connection. 178 * 179 * We implement an exponential backoff when trying to reestablish a TCP 180 * transport connection with the server. Some servers like to drop a TCP 181 * connection when they are overworked, so we start with a short timeout and 182 * increase over time if the server is down or not responding. 183 */ 184 #define XS_TCP_INIT_REEST_TO (3U * HZ) 185 186 /* 187 * TCP idle timeout; client drops the transport socket if it is idle 188 * for this long. Note that we also timeout UDP sockets to prevent 189 * holding port numbers when there is no RPC traffic. 190 */ 191 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 192 193 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 194 # undef RPC_DEBUG_DATA 195 # define RPCDBG_FACILITY RPCDBG_TRANS 196 #endif 197 198 #ifdef RPC_DEBUG_DATA 199 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 200 { 201 u8 *buf = (u8 *) packet; 202 int j; 203 204 dprintk("RPC: %s\n", msg); 205 for (j = 0; j < count && j < 128; j += 4) { 206 if (!(j & 31)) { 207 if (j) 208 dprintk("\n"); 209 dprintk("0x%04x ", j); 210 } 211 dprintk("%02x%02x%02x%02x ", 212 buf[j], buf[j+1], buf[j+2], buf[j+3]); 213 } 214 dprintk("\n"); 215 } 216 #else 217 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 218 { 219 /* NOP */ 220 } 221 #endif 222 223 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 224 { 225 return (struct rpc_xprt *) sk->sk_user_data; 226 } 227 228 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 229 { 230 return (struct sockaddr *) &xprt->addr; 231 } 232 233 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 234 { 235 return (struct sockaddr_un *) &xprt->addr; 236 } 237 238 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 239 { 240 return (struct sockaddr_in *) &xprt->addr; 241 } 242 243 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 244 { 245 return (struct sockaddr_in6 *) &xprt->addr; 246 } 247 248 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 249 { 250 struct sockaddr *sap = xs_addr(xprt); 251 struct sockaddr_in6 *sin6; 252 struct sockaddr_in *sin; 253 struct sockaddr_un *sun; 254 char buf[128]; 255 256 switch (sap->sa_family) { 257 case AF_LOCAL: 258 sun = xs_addr_un(xprt); 259 strlcpy(buf, sun->sun_path, sizeof(buf)); 260 xprt->address_strings[RPC_DISPLAY_ADDR] = 261 kstrdup(buf, GFP_KERNEL); 262 break; 263 case AF_INET: 264 (void)rpc_ntop(sap, buf, sizeof(buf)); 265 xprt->address_strings[RPC_DISPLAY_ADDR] = 266 kstrdup(buf, GFP_KERNEL); 267 sin = xs_addr_in(xprt); 268 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 269 break; 270 case AF_INET6: 271 (void)rpc_ntop(sap, buf, sizeof(buf)); 272 xprt->address_strings[RPC_DISPLAY_ADDR] = 273 kstrdup(buf, GFP_KERNEL); 274 sin6 = xs_addr_in6(xprt); 275 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 276 break; 277 default: 278 BUG(); 279 } 280 281 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 282 } 283 284 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 285 { 286 struct sockaddr *sap = xs_addr(xprt); 287 char buf[128]; 288 289 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 290 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 291 292 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 293 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 294 } 295 296 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 297 const char *protocol, 298 const char *netid) 299 { 300 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 301 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 302 xs_format_common_peer_addresses(xprt); 303 xs_format_common_peer_ports(xprt); 304 } 305 306 static void xs_update_peer_port(struct rpc_xprt *xprt) 307 { 308 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 309 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 310 311 xs_format_common_peer_ports(xprt); 312 } 313 314 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 315 { 316 unsigned int i; 317 318 for (i = 0; i < RPC_DISPLAY_MAX; i++) 319 switch (i) { 320 case RPC_DISPLAY_PROTO: 321 case RPC_DISPLAY_NETID: 322 continue; 323 default: 324 kfree(xprt->address_strings[i]); 325 } 326 } 327 328 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 329 330 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more) 331 { 332 struct msghdr msg = { 333 .msg_name = addr, 334 .msg_namelen = addrlen, 335 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0), 336 }; 337 struct kvec iov = { 338 .iov_base = vec->iov_base + base, 339 .iov_len = vec->iov_len - base, 340 }; 341 342 if (iov.iov_len != 0) 343 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); 344 return kernel_sendmsg(sock, &msg, NULL, 0, 0); 345 } 346 347 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p) 348 { 349 ssize_t (*do_sendpage)(struct socket *sock, struct page *page, 350 int offset, size_t size, int flags); 351 struct page **ppage; 352 unsigned int remainder; 353 int err; 354 355 remainder = xdr->page_len - base; 356 base += xdr->page_base; 357 ppage = xdr->pages + (base >> PAGE_SHIFT); 358 base &= ~PAGE_MASK; 359 do_sendpage = sock->ops->sendpage; 360 if (!zerocopy) 361 do_sendpage = sock_no_sendpage; 362 for(;;) { 363 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); 364 int flags = XS_SENDMSG_FLAGS; 365 366 remainder -= len; 367 if (more) 368 flags |= MSG_MORE; 369 if (remainder != 0) 370 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE; 371 err = do_sendpage(sock, *ppage, base, len, flags); 372 if (remainder == 0 || err != len) 373 break; 374 *sent_p += err; 375 ppage++; 376 base = 0; 377 } 378 if (err > 0) { 379 *sent_p += err; 380 err = 0; 381 } 382 return err; 383 } 384 385 /** 386 * xs_sendpages - write pages directly to a socket 387 * @sock: socket to send on 388 * @addr: UDP only -- address of destination 389 * @addrlen: UDP only -- length of destination address 390 * @xdr: buffer containing this request 391 * @base: starting position in the buffer 392 * @zerocopy: true if it is safe to use sendpage() 393 * @sent_p: return the total number of bytes successfully queued for sending 394 * 395 */ 396 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p) 397 { 398 unsigned int remainder = xdr->len - base; 399 int err = 0; 400 int sent = 0; 401 402 if (unlikely(!sock)) 403 return -ENOTSOCK; 404 405 if (base != 0) { 406 addr = NULL; 407 addrlen = 0; 408 } 409 410 if (base < xdr->head[0].iov_len || addr != NULL) { 411 unsigned int len = xdr->head[0].iov_len - base; 412 remainder -= len; 413 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0); 414 if (remainder == 0 || err != len) 415 goto out; 416 *sent_p += err; 417 base = 0; 418 } else 419 base -= xdr->head[0].iov_len; 420 421 if (base < xdr->page_len) { 422 unsigned int len = xdr->page_len - base; 423 remainder -= len; 424 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent); 425 *sent_p += sent; 426 if (remainder == 0 || sent != len) 427 goto out; 428 base = 0; 429 } else 430 base -= xdr->page_len; 431 432 if (base >= xdr->tail[0].iov_len) 433 return 0; 434 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0); 435 out: 436 if (err > 0) { 437 *sent_p += err; 438 err = 0; 439 } 440 return err; 441 } 442 443 static void xs_nospace_callback(struct rpc_task *task) 444 { 445 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt); 446 447 transport->inet->sk_write_pending--; 448 } 449 450 /** 451 * xs_nospace - place task on wait queue if transmit was incomplete 452 * @task: task to put to sleep 453 * 454 */ 455 static int xs_nospace(struct rpc_task *task) 456 { 457 struct rpc_rqst *req = task->tk_rqstp; 458 struct rpc_xprt *xprt = req->rq_xprt; 459 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 460 struct sock *sk = transport->inet; 461 int ret = -EAGAIN; 462 463 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", 464 task->tk_pid, req->rq_slen - req->rq_bytes_sent, 465 req->rq_slen); 466 467 /* Protect against races with write_space */ 468 spin_lock_bh(&xprt->transport_lock); 469 470 /* Don't race with disconnect */ 471 if (xprt_connected(xprt)) { 472 /* wait for more buffer space */ 473 sk->sk_write_pending++; 474 xprt_wait_for_buffer_space(task, xs_nospace_callback); 475 } else 476 ret = -ENOTCONN; 477 478 spin_unlock_bh(&xprt->transport_lock); 479 480 /* Race breaker in case memory is freed before above code is called */ 481 if (ret == -EAGAIN) { 482 struct socket_wq *wq; 483 484 rcu_read_lock(); 485 wq = rcu_dereference(sk->sk_wq); 486 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags); 487 rcu_read_unlock(); 488 489 sk->sk_write_space(sk); 490 } 491 return ret; 492 } 493 494 /* 495 * Construct a stream transport record marker in @buf. 496 */ 497 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf) 498 { 499 u32 reclen = buf->len - sizeof(rpc_fraghdr); 500 rpc_fraghdr *base = buf->head[0].iov_base; 501 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen); 502 } 503 504 /** 505 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 506 * @task: RPC task that manages the state of an RPC request 507 * 508 * Return values: 509 * 0: The request has been sent 510 * EAGAIN: The socket was blocked, please call again later to 511 * complete the request 512 * ENOTCONN: Caller needs to invoke connect logic then call again 513 * other: Some other error occured, the request was not sent 514 */ 515 static int xs_local_send_request(struct rpc_task *task) 516 { 517 struct rpc_rqst *req = task->tk_rqstp; 518 struct rpc_xprt *xprt = req->rq_xprt; 519 struct sock_xprt *transport = 520 container_of(xprt, struct sock_xprt, xprt); 521 struct xdr_buf *xdr = &req->rq_snd_buf; 522 int status; 523 int sent = 0; 524 525 xs_encode_stream_record_marker(&req->rq_snd_buf); 526 527 xs_pktdump("packet data:", 528 req->rq_svec->iov_base, req->rq_svec->iov_len); 529 530 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent, 531 true, &sent); 532 dprintk("RPC: %s(%u) = %d\n", 533 __func__, xdr->len - req->rq_bytes_sent, status); 534 535 if (status == -EAGAIN && sock_writeable(transport->inet)) 536 status = -ENOBUFS; 537 538 if (likely(sent > 0) || status == 0) { 539 req->rq_bytes_sent += sent; 540 req->rq_xmit_bytes_sent += sent; 541 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 542 req->rq_bytes_sent = 0; 543 return 0; 544 } 545 status = -EAGAIN; 546 } 547 548 switch (status) { 549 case -ENOBUFS: 550 break; 551 case -EAGAIN: 552 status = xs_nospace(task); 553 break; 554 default: 555 dprintk("RPC: sendmsg returned unrecognized error %d\n", 556 -status); 557 /* fall through */ 558 case -EPIPE: 559 xs_close(xprt); 560 status = -ENOTCONN; 561 } 562 563 return status; 564 } 565 566 /** 567 * xs_udp_send_request - write an RPC request to a UDP socket 568 * @task: address of RPC task that manages the state of an RPC request 569 * 570 * Return values: 571 * 0: The request has been sent 572 * EAGAIN: The socket was blocked, please call again later to 573 * complete the request 574 * ENOTCONN: Caller needs to invoke connect logic then call again 575 * other: Some other error occurred, the request was not sent 576 */ 577 static int xs_udp_send_request(struct rpc_task *task) 578 { 579 struct rpc_rqst *req = task->tk_rqstp; 580 struct rpc_xprt *xprt = req->rq_xprt; 581 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 582 struct xdr_buf *xdr = &req->rq_snd_buf; 583 int sent = 0; 584 int status; 585 586 xs_pktdump("packet data:", 587 req->rq_svec->iov_base, 588 req->rq_svec->iov_len); 589 590 if (!xprt_bound(xprt)) 591 return -ENOTCONN; 592 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen, 593 xdr, req->rq_bytes_sent, true, &sent); 594 595 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 596 xdr->len - req->rq_bytes_sent, status); 597 598 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 599 if (status == -EPERM) 600 goto process_status; 601 602 if (status == -EAGAIN && sock_writeable(transport->inet)) 603 status = -ENOBUFS; 604 605 if (sent > 0 || status == 0) { 606 req->rq_xmit_bytes_sent += sent; 607 if (sent >= req->rq_slen) 608 return 0; 609 /* Still some bytes left; set up for a retry later. */ 610 status = -EAGAIN; 611 } 612 613 process_status: 614 switch (status) { 615 case -ENOTSOCK: 616 status = -ENOTCONN; 617 /* Should we call xs_close() here? */ 618 break; 619 case -EAGAIN: 620 status = xs_nospace(task); 621 break; 622 case -ENETUNREACH: 623 case -ENOBUFS: 624 case -EPIPE: 625 case -ECONNREFUSED: 626 case -EPERM: 627 /* When the server has died, an ICMP port unreachable message 628 * prompts ECONNREFUSED. */ 629 break; 630 default: 631 dprintk("RPC: sendmsg returned unrecognized error %d\n", 632 -status); 633 } 634 635 return status; 636 } 637 638 /** 639 * xs_tcp_send_request - write an RPC request to a TCP socket 640 * @task: address of RPC task that manages the state of an RPC request 641 * 642 * Return values: 643 * 0: The request has been sent 644 * EAGAIN: The socket was blocked, please call again later to 645 * complete the request 646 * ENOTCONN: Caller needs to invoke connect logic then call again 647 * other: Some other error occurred, the request was not sent 648 * 649 * XXX: In the case of soft timeouts, should we eventually give up 650 * if sendmsg is not able to make progress? 651 */ 652 static int xs_tcp_send_request(struct rpc_task *task) 653 { 654 struct rpc_rqst *req = task->tk_rqstp; 655 struct rpc_xprt *xprt = req->rq_xprt; 656 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 657 struct xdr_buf *xdr = &req->rq_snd_buf; 658 bool zerocopy = true; 659 bool vm_wait = false; 660 int status; 661 int sent; 662 663 xs_encode_stream_record_marker(&req->rq_snd_buf); 664 665 xs_pktdump("packet data:", 666 req->rq_svec->iov_base, 667 req->rq_svec->iov_len); 668 /* Don't use zero copy if this is a resend. If the RPC call 669 * completes while the socket holds a reference to the pages, 670 * then we may end up resending corrupted data. 671 */ 672 if (task->tk_flags & RPC_TASK_SENT) 673 zerocopy = false; 674 675 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state)) 676 xs_tcp_set_socket_timeouts(xprt, transport->sock); 677 678 /* Continue transmitting the packet/record. We must be careful 679 * to cope with writespace callbacks arriving _after_ we have 680 * called sendmsg(). */ 681 while (1) { 682 sent = 0; 683 status = xs_sendpages(transport->sock, NULL, 0, xdr, 684 req->rq_bytes_sent, zerocopy, &sent); 685 686 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 687 xdr->len - req->rq_bytes_sent, status); 688 689 /* If we've sent the entire packet, immediately 690 * reset the count of bytes sent. */ 691 req->rq_bytes_sent += sent; 692 req->rq_xmit_bytes_sent += sent; 693 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 694 req->rq_bytes_sent = 0; 695 return 0; 696 } 697 698 WARN_ON_ONCE(sent == 0 && status == 0); 699 700 if (status == -EAGAIN ) { 701 /* 702 * Return EAGAIN if we're sure we're hitting the 703 * socket send buffer limits. 704 */ 705 if (test_bit(SOCK_NOSPACE, &transport->sock->flags)) 706 break; 707 /* 708 * Did we hit a memory allocation failure? 709 */ 710 if (sent == 0) { 711 status = -ENOBUFS; 712 if (vm_wait) 713 break; 714 /* Retry, knowing now that we're below the 715 * socket send buffer limit 716 */ 717 vm_wait = true; 718 } 719 continue; 720 } 721 if (status < 0) 722 break; 723 vm_wait = false; 724 } 725 726 switch (status) { 727 case -ENOTSOCK: 728 status = -ENOTCONN; 729 /* Should we call xs_close() here? */ 730 break; 731 case -EAGAIN: 732 status = xs_nospace(task); 733 break; 734 case -ECONNRESET: 735 case -ECONNREFUSED: 736 case -ENOTCONN: 737 case -EADDRINUSE: 738 case -ENOBUFS: 739 case -EPIPE: 740 break; 741 default: 742 dprintk("RPC: sendmsg returned unrecognized error %d\n", 743 -status); 744 } 745 746 return status; 747 } 748 749 /** 750 * xs_tcp_release_xprt - clean up after a tcp transmission 751 * @xprt: transport 752 * @task: rpc task 753 * 754 * This cleans up if an error causes us to abort the transmission of a request. 755 * In this case, the socket may need to be reset in order to avoid confusing 756 * the server. 757 */ 758 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 759 { 760 struct rpc_rqst *req; 761 762 if (task != xprt->snd_task) 763 return; 764 if (task == NULL) 765 goto out_release; 766 req = task->tk_rqstp; 767 if (req == NULL) 768 goto out_release; 769 if (req->rq_bytes_sent == 0) 770 goto out_release; 771 if (req->rq_bytes_sent == req->rq_snd_buf.len) 772 goto out_release; 773 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 774 out_release: 775 xprt_release_xprt(xprt, task); 776 } 777 778 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 779 { 780 transport->old_data_ready = sk->sk_data_ready; 781 transport->old_state_change = sk->sk_state_change; 782 transport->old_write_space = sk->sk_write_space; 783 transport->old_error_report = sk->sk_error_report; 784 } 785 786 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 787 { 788 sk->sk_data_ready = transport->old_data_ready; 789 sk->sk_state_change = transport->old_state_change; 790 sk->sk_write_space = transport->old_write_space; 791 sk->sk_error_report = transport->old_error_report; 792 } 793 794 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt) 795 { 796 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 797 798 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 799 } 800 801 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 802 { 803 smp_mb__before_atomic(); 804 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 805 clear_bit(XPRT_CLOSING, &xprt->state); 806 xs_sock_reset_state_flags(xprt); 807 smp_mb__after_atomic(); 808 } 809 810 static void xs_sock_mark_closed(struct rpc_xprt *xprt) 811 { 812 xs_sock_reset_connection_flags(xprt); 813 /* Mark transport as closed and wake up all pending tasks */ 814 xprt_disconnect_done(xprt); 815 } 816 817 /** 818 * xs_error_report - callback to handle TCP socket state errors 819 * @sk: socket 820 * 821 * Note: we don't call sock_error() since there may be a rpc_task 822 * using the socket, and so we don't want to clear sk->sk_err. 823 */ 824 static void xs_error_report(struct sock *sk) 825 { 826 struct rpc_xprt *xprt; 827 int err; 828 829 read_lock_bh(&sk->sk_callback_lock); 830 if (!(xprt = xprt_from_sock(sk))) 831 goto out; 832 833 err = -sk->sk_err; 834 if (err == 0) 835 goto out; 836 /* Is this a reset event? */ 837 if (sk->sk_state == TCP_CLOSE) 838 xs_sock_mark_closed(xprt); 839 dprintk("RPC: xs_error_report client %p, error=%d...\n", 840 xprt, -err); 841 trace_rpc_socket_error(xprt, sk->sk_socket, err); 842 xprt_wake_pending_tasks(xprt, err); 843 out: 844 read_unlock_bh(&sk->sk_callback_lock); 845 } 846 847 static void xs_reset_transport(struct sock_xprt *transport) 848 { 849 struct socket *sock = transport->sock; 850 struct sock *sk = transport->inet; 851 struct rpc_xprt *xprt = &transport->xprt; 852 853 if (sk == NULL) 854 return; 855 856 if (atomic_read(&transport->xprt.swapper)) 857 sk_clear_memalloc(sk); 858 859 kernel_sock_shutdown(sock, SHUT_RDWR); 860 861 mutex_lock(&transport->recv_mutex); 862 write_lock_bh(&sk->sk_callback_lock); 863 transport->inet = NULL; 864 transport->sock = NULL; 865 866 sk->sk_user_data = NULL; 867 868 xs_restore_old_callbacks(transport, sk); 869 xprt_clear_connected(xprt); 870 write_unlock_bh(&sk->sk_callback_lock); 871 xs_sock_reset_connection_flags(xprt); 872 mutex_unlock(&transport->recv_mutex); 873 874 trace_rpc_socket_close(xprt, sock); 875 sock_release(sock); 876 } 877 878 /** 879 * xs_close - close a socket 880 * @xprt: transport 881 * 882 * This is used when all requests are complete; ie, no DRC state remains 883 * on the server we want to save. 884 * 885 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 886 * xs_reset_transport() zeroing the socket from underneath a writer. 887 */ 888 static void xs_close(struct rpc_xprt *xprt) 889 { 890 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 891 892 dprintk("RPC: xs_close xprt %p\n", xprt); 893 894 xs_reset_transport(transport); 895 xprt->reestablish_timeout = 0; 896 897 xprt_disconnect_done(xprt); 898 } 899 900 static void xs_inject_disconnect(struct rpc_xprt *xprt) 901 { 902 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 903 xprt); 904 xprt_disconnect_done(xprt); 905 } 906 907 static void xs_xprt_free(struct rpc_xprt *xprt) 908 { 909 xs_free_peer_addresses(xprt); 910 xprt_free(xprt); 911 } 912 913 /** 914 * xs_destroy - prepare to shutdown a transport 915 * @xprt: doomed transport 916 * 917 */ 918 static void xs_destroy(struct rpc_xprt *xprt) 919 { 920 struct sock_xprt *transport = container_of(xprt, 921 struct sock_xprt, xprt); 922 dprintk("RPC: xs_destroy xprt %p\n", xprt); 923 924 cancel_delayed_work_sync(&transport->connect_worker); 925 xs_close(xprt); 926 cancel_work_sync(&transport->recv_worker); 927 xs_xprt_free(xprt); 928 module_put(THIS_MODULE); 929 } 930 931 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb) 932 { 933 struct xdr_skb_reader desc = { 934 .skb = skb, 935 .offset = sizeof(rpc_fraghdr), 936 .count = skb->len - sizeof(rpc_fraghdr), 937 }; 938 939 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0) 940 return -1; 941 if (desc.count) 942 return -1; 943 return 0; 944 } 945 946 /** 947 * xs_local_data_read_skb 948 * @xprt: transport 949 * @sk: socket 950 * @skb: skbuff 951 * 952 * Currently this assumes we can read the whole reply in a single gulp. 953 */ 954 static void xs_local_data_read_skb(struct rpc_xprt *xprt, 955 struct sock *sk, 956 struct sk_buff *skb) 957 { 958 struct rpc_task *task; 959 struct rpc_rqst *rovr; 960 int repsize, copied; 961 u32 _xid; 962 __be32 *xp; 963 964 repsize = skb->len - sizeof(rpc_fraghdr); 965 if (repsize < 4) { 966 dprintk("RPC: impossible RPC reply size %d\n", repsize); 967 return; 968 } 969 970 /* Copy the XID from the skb... */ 971 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid); 972 if (xp == NULL) 973 return; 974 975 /* Look up and lock the request corresponding to the given XID */ 976 spin_lock(&xprt->recv_lock); 977 rovr = xprt_lookup_rqst(xprt, *xp); 978 if (!rovr) 979 goto out_unlock; 980 xprt_pin_rqst(rovr); 981 spin_unlock(&xprt->recv_lock); 982 task = rovr->rq_task; 983 984 copied = rovr->rq_private_buf.buflen; 985 if (copied > repsize) 986 copied = repsize; 987 988 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) { 989 dprintk("RPC: sk_buff copy failed\n"); 990 spin_lock(&xprt->recv_lock); 991 goto out_unpin; 992 } 993 994 spin_lock(&xprt->recv_lock); 995 xprt_complete_rqst(task, copied); 996 out_unpin: 997 xprt_unpin_rqst(rovr); 998 out_unlock: 999 spin_unlock(&xprt->recv_lock); 1000 } 1001 1002 static void xs_local_data_receive(struct sock_xprt *transport) 1003 { 1004 struct sk_buff *skb; 1005 struct sock *sk; 1006 int err; 1007 1008 restart: 1009 mutex_lock(&transport->recv_mutex); 1010 sk = transport->inet; 1011 if (sk == NULL) 1012 goto out; 1013 for (;;) { 1014 skb = skb_recv_datagram(sk, 0, 1, &err); 1015 if (skb != NULL) { 1016 xs_local_data_read_skb(&transport->xprt, sk, skb); 1017 skb_free_datagram(sk, skb); 1018 continue; 1019 } 1020 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1021 break; 1022 if (need_resched()) { 1023 mutex_unlock(&transport->recv_mutex); 1024 cond_resched(); 1025 goto restart; 1026 } 1027 } 1028 out: 1029 mutex_unlock(&transport->recv_mutex); 1030 } 1031 1032 static void xs_local_data_receive_workfn(struct work_struct *work) 1033 { 1034 struct sock_xprt *transport = 1035 container_of(work, struct sock_xprt, recv_worker); 1036 xs_local_data_receive(transport); 1037 } 1038 1039 /** 1040 * xs_udp_data_read_skb - receive callback for UDP sockets 1041 * @xprt: transport 1042 * @sk: socket 1043 * @skb: skbuff 1044 * 1045 */ 1046 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1047 struct sock *sk, 1048 struct sk_buff *skb) 1049 { 1050 struct rpc_task *task; 1051 struct rpc_rqst *rovr; 1052 int repsize, copied; 1053 u32 _xid; 1054 __be32 *xp; 1055 1056 repsize = skb->len; 1057 if (repsize < 4) { 1058 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1059 return; 1060 } 1061 1062 /* Copy the XID from the skb... */ 1063 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid); 1064 if (xp == NULL) 1065 return; 1066 1067 /* Look up and lock the request corresponding to the given XID */ 1068 spin_lock(&xprt->recv_lock); 1069 rovr = xprt_lookup_rqst(xprt, *xp); 1070 if (!rovr) 1071 goto out_unlock; 1072 xprt_pin_rqst(rovr); 1073 spin_unlock(&xprt->recv_lock); 1074 task = rovr->rq_task; 1075 1076 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1077 copied = repsize; 1078 1079 /* Suck it into the iovec, verify checksum if not done by hw. */ 1080 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1081 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS); 1082 spin_lock(&xprt->recv_lock); 1083 goto out_unpin; 1084 } 1085 1086 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS); 1087 1088 spin_lock_bh(&xprt->transport_lock); 1089 xprt_adjust_cwnd(xprt, task, copied); 1090 spin_unlock_bh(&xprt->transport_lock); 1091 spin_lock(&xprt->recv_lock); 1092 xprt_complete_rqst(task, copied); 1093 out_unpin: 1094 xprt_unpin_rqst(rovr); 1095 out_unlock: 1096 spin_unlock(&xprt->recv_lock); 1097 } 1098 1099 static void xs_udp_data_receive(struct sock_xprt *transport) 1100 { 1101 struct sk_buff *skb; 1102 struct sock *sk; 1103 int err; 1104 1105 restart: 1106 mutex_lock(&transport->recv_mutex); 1107 sk = transport->inet; 1108 if (sk == NULL) 1109 goto out; 1110 for (;;) { 1111 skb = skb_recv_udp(sk, 0, 1, &err); 1112 if (skb != NULL) { 1113 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1114 consume_skb(skb); 1115 continue; 1116 } 1117 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1118 break; 1119 if (need_resched()) { 1120 mutex_unlock(&transport->recv_mutex); 1121 cond_resched(); 1122 goto restart; 1123 } 1124 } 1125 out: 1126 mutex_unlock(&transport->recv_mutex); 1127 } 1128 1129 static void xs_udp_data_receive_workfn(struct work_struct *work) 1130 { 1131 struct sock_xprt *transport = 1132 container_of(work, struct sock_xprt, recv_worker); 1133 xs_udp_data_receive(transport); 1134 } 1135 1136 /** 1137 * xs_data_ready - "data ready" callback for UDP sockets 1138 * @sk: socket with data to read 1139 * 1140 */ 1141 static void xs_data_ready(struct sock *sk) 1142 { 1143 struct rpc_xprt *xprt; 1144 1145 read_lock_bh(&sk->sk_callback_lock); 1146 dprintk("RPC: xs_data_ready...\n"); 1147 xprt = xprt_from_sock(sk); 1148 if (xprt != NULL) { 1149 struct sock_xprt *transport = container_of(xprt, 1150 struct sock_xprt, xprt); 1151 transport->old_data_ready(sk); 1152 /* Any data means we had a useful conversation, so 1153 * then we don't need to delay the next reconnect 1154 */ 1155 if (xprt->reestablish_timeout) 1156 xprt->reestablish_timeout = 0; 1157 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1158 queue_work(xprtiod_workqueue, &transport->recv_worker); 1159 } 1160 read_unlock_bh(&sk->sk_callback_lock); 1161 } 1162 1163 /* 1164 * Helper function to force a TCP close if the server is sending 1165 * junk and/or it has put us in CLOSE_WAIT 1166 */ 1167 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1168 { 1169 xprt_force_disconnect(xprt); 1170 } 1171 1172 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 1173 { 1174 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1175 size_t len, used; 1176 char *p; 1177 1178 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; 1179 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; 1180 used = xdr_skb_read_bits(desc, p, len); 1181 transport->tcp_offset += used; 1182 if (used != len) 1183 return; 1184 1185 transport->tcp_reclen = ntohl(transport->tcp_fraghdr); 1186 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) 1187 transport->tcp_flags |= TCP_RCV_LAST_FRAG; 1188 else 1189 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; 1190 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; 1191 1192 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; 1193 transport->tcp_offset = 0; 1194 1195 /* Sanity check of the record length */ 1196 if (unlikely(transport->tcp_reclen < 8)) { 1197 dprintk("RPC: invalid TCP record fragment length\n"); 1198 xs_tcp_force_close(xprt); 1199 return; 1200 } 1201 dprintk("RPC: reading TCP record fragment of length %d\n", 1202 transport->tcp_reclen); 1203 } 1204 1205 static void xs_tcp_check_fraghdr(struct sock_xprt *transport) 1206 { 1207 if (transport->tcp_offset == transport->tcp_reclen) { 1208 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; 1209 transport->tcp_offset = 0; 1210 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { 1211 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1212 transport->tcp_flags |= TCP_RCV_COPY_XID; 1213 transport->tcp_copied = 0; 1214 } 1215 } 1216 } 1217 1218 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1219 { 1220 size_t len, used; 1221 char *p; 1222 1223 len = sizeof(transport->tcp_xid) - transport->tcp_offset; 1224 dprintk("RPC: reading XID (%zu bytes)\n", len); 1225 p = ((char *) &transport->tcp_xid) + transport->tcp_offset; 1226 used = xdr_skb_read_bits(desc, p, len); 1227 transport->tcp_offset += used; 1228 if (used != len) 1229 return; 1230 transport->tcp_flags &= ~TCP_RCV_COPY_XID; 1231 transport->tcp_flags |= TCP_RCV_READ_CALLDIR; 1232 transport->tcp_copied = 4; 1233 dprintk("RPC: reading %s XID %08x\n", 1234 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for" 1235 : "request with", 1236 ntohl(transport->tcp_xid)); 1237 xs_tcp_check_fraghdr(transport); 1238 } 1239 1240 static inline void xs_tcp_read_calldir(struct sock_xprt *transport, 1241 struct xdr_skb_reader *desc) 1242 { 1243 size_t len, used; 1244 u32 offset; 1245 char *p; 1246 1247 /* 1248 * We want transport->tcp_offset to be 8 at the end of this routine 1249 * (4 bytes for the xid and 4 bytes for the call/reply flag). 1250 * When this function is called for the first time, 1251 * transport->tcp_offset is 4 (after having already read the xid). 1252 */ 1253 offset = transport->tcp_offset - sizeof(transport->tcp_xid); 1254 len = sizeof(transport->tcp_calldir) - offset; 1255 dprintk("RPC: reading CALL/REPLY flag (%zu bytes)\n", len); 1256 p = ((char *) &transport->tcp_calldir) + offset; 1257 used = xdr_skb_read_bits(desc, p, len); 1258 transport->tcp_offset += used; 1259 if (used != len) 1260 return; 1261 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR; 1262 /* 1263 * We don't yet have the XDR buffer, so we will write the calldir 1264 * out after we get the buffer from the 'struct rpc_rqst' 1265 */ 1266 switch (ntohl(transport->tcp_calldir)) { 1267 case RPC_REPLY: 1268 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1269 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1270 transport->tcp_flags |= TCP_RPC_REPLY; 1271 break; 1272 case RPC_CALL: 1273 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1274 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1275 transport->tcp_flags &= ~TCP_RPC_REPLY; 1276 break; 1277 default: 1278 dprintk("RPC: invalid request message type\n"); 1279 xs_tcp_force_close(&transport->xprt); 1280 } 1281 xs_tcp_check_fraghdr(transport); 1282 } 1283 1284 static inline void xs_tcp_read_common(struct rpc_xprt *xprt, 1285 struct xdr_skb_reader *desc, 1286 struct rpc_rqst *req) 1287 { 1288 struct sock_xprt *transport = 1289 container_of(xprt, struct sock_xprt, xprt); 1290 struct xdr_buf *rcvbuf; 1291 size_t len; 1292 ssize_t r; 1293 1294 rcvbuf = &req->rq_private_buf; 1295 1296 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) { 1297 /* 1298 * Save the RPC direction in the XDR buffer 1299 */ 1300 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied, 1301 &transport->tcp_calldir, 1302 sizeof(transport->tcp_calldir)); 1303 transport->tcp_copied += sizeof(transport->tcp_calldir); 1304 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR; 1305 } 1306 1307 len = desc->count; 1308 if (len > transport->tcp_reclen - transport->tcp_offset) 1309 desc->count = transport->tcp_reclen - transport->tcp_offset; 1310 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1311 desc, xdr_skb_read_bits); 1312 1313 if (desc->count) { 1314 /* Error when copying to the receive buffer, 1315 * usually because we weren't able to allocate 1316 * additional buffer pages. All we can do now 1317 * is turn off TCP_RCV_COPY_DATA, so the request 1318 * will not receive any additional updates, 1319 * and time out. 1320 * Any remaining data from this record will 1321 * be discarded. 1322 */ 1323 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1324 dprintk("RPC: XID %08x truncated request\n", 1325 ntohl(transport->tcp_xid)); 1326 dprintk("RPC: xprt = %p, tcp_copied = %lu, " 1327 "tcp_offset = %u, tcp_reclen = %u\n", 1328 xprt, transport->tcp_copied, 1329 transport->tcp_offset, transport->tcp_reclen); 1330 return; 1331 } 1332 1333 transport->tcp_copied += r; 1334 transport->tcp_offset += r; 1335 desc->count = len - r; 1336 1337 dprintk("RPC: XID %08x read %zd bytes\n", 1338 ntohl(transport->tcp_xid), r); 1339 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " 1340 "tcp_reclen = %u\n", xprt, transport->tcp_copied, 1341 transport->tcp_offset, transport->tcp_reclen); 1342 1343 if (transport->tcp_copied == req->rq_private_buf.buflen) 1344 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1345 else if (transport->tcp_offset == transport->tcp_reclen) { 1346 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) 1347 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1348 } 1349 } 1350 1351 /* 1352 * Finds the request corresponding to the RPC xid and invokes the common 1353 * tcp read code to read the data. 1354 */ 1355 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt, 1356 struct xdr_skb_reader *desc) 1357 { 1358 struct sock_xprt *transport = 1359 container_of(xprt, struct sock_xprt, xprt); 1360 struct rpc_rqst *req; 1361 1362 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid)); 1363 1364 /* Find and lock the request corresponding to this xid */ 1365 spin_lock(&xprt->recv_lock); 1366 req = xprt_lookup_rqst(xprt, transport->tcp_xid); 1367 if (!req) { 1368 dprintk("RPC: XID %08x request not found!\n", 1369 ntohl(transport->tcp_xid)); 1370 spin_unlock(&xprt->recv_lock); 1371 return -1; 1372 } 1373 xprt_pin_rqst(req); 1374 spin_unlock(&xprt->recv_lock); 1375 1376 xs_tcp_read_common(xprt, desc, req); 1377 1378 spin_lock(&xprt->recv_lock); 1379 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1380 xprt_complete_rqst(req->rq_task, transport->tcp_copied); 1381 xprt_unpin_rqst(req); 1382 spin_unlock(&xprt->recv_lock); 1383 return 0; 1384 } 1385 1386 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1387 /* 1388 * Obtains an rpc_rqst previously allocated and invokes the common 1389 * tcp read code to read the data. The result is placed in the callback 1390 * queue. 1391 * If we're unable to obtain the rpc_rqst we schedule the closing of the 1392 * connection and return -1. 1393 */ 1394 static int xs_tcp_read_callback(struct rpc_xprt *xprt, 1395 struct xdr_skb_reader *desc) 1396 { 1397 struct sock_xprt *transport = 1398 container_of(xprt, struct sock_xprt, xprt); 1399 struct rpc_rqst *req; 1400 1401 /* Look up the request corresponding to the given XID */ 1402 req = xprt_lookup_bc_request(xprt, transport->tcp_xid); 1403 if (req == NULL) { 1404 printk(KERN_WARNING "Callback slot table overflowed\n"); 1405 xprt_force_disconnect(xprt); 1406 return -1; 1407 } 1408 1409 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid)); 1410 xs_tcp_read_common(xprt, desc, req); 1411 1412 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1413 xprt_complete_bc_request(req, transport->tcp_copied); 1414 1415 return 0; 1416 } 1417 1418 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1419 struct xdr_skb_reader *desc) 1420 { 1421 struct sock_xprt *transport = 1422 container_of(xprt, struct sock_xprt, xprt); 1423 1424 return (transport->tcp_flags & TCP_RPC_REPLY) ? 1425 xs_tcp_read_reply(xprt, desc) : 1426 xs_tcp_read_callback(xprt, desc); 1427 } 1428 1429 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net) 1430 { 1431 int ret; 1432 1433 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0, 1434 SVC_SOCK_ANONYMOUS); 1435 if (ret < 0) 1436 return ret; 1437 return 0; 1438 } 1439 1440 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt) 1441 { 1442 return PAGE_SIZE; 1443 } 1444 #else 1445 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1446 struct xdr_skb_reader *desc) 1447 { 1448 return xs_tcp_read_reply(xprt, desc); 1449 } 1450 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1451 1452 /* 1453 * Read data off the transport. This can be either an RPC_CALL or an 1454 * RPC_REPLY. Relay the processing to helper functions. 1455 */ 1456 static void xs_tcp_read_data(struct rpc_xprt *xprt, 1457 struct xdr_skb_reader *desc) 1458 { 1459 struct sock_xprt *transport = 1460 container_of(xprt, struct sock_xprt, xprt); 1461 1462 if (_xs_tcp_read_data(xprt, desc) == 0) 1463 xs_tcp_check_fraghdr(transport); 1464 else { 1465 /* 1466 * The transport_lock protects the request handling. 1467 * There's no need to hold it to update the tcp_flags. 1468 */ 1469 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1470 } 1471 } 1472 1473 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1474 { 1475 size_t len; 1476 1477 len = transport->tcp_reclen - transport->tcp_offset; 1478 if (len > desc->count) 1479 len = desc->count; 1480 desc->count -= len; 1481 desc->offset += len; 1482 transport->tcp_offset += len; 1483 dprintk("RPC: discarded %zu bytes\n", len); 1484 xs_tcp_check_fraghdr(transport); 1485 } 1486 1487 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) 1488 { 1489 struct rpc_xprt *xprt = rd_desc->arg.data; 1490 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1491 struct xdr_skb_reader desc = { 1492 .skb = skb, 1493 .offset = offset, 1494 .count = len, 1495 }; 1496 size_t ret; 1497 1498 dprintk("RPC: xs_tcp_data_recv started\n"); 1499 do { 1500 trace_xs_tcp_data_recv(transport); 1501 /* Read in a new fragment marker if necessary */ 1502 /* Can we ever really expect to get completely empty fragments? */ 1503 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { 1504 xs_tcp_read_fraghdr(xprt, &desc); 1505 continue; 1506 } 1507 /* Read in the xid if necessary */ 1508 if (transport->tcp_flags & TCP_RCV_COPY_XID) { 1509 xs_tcp_read_xid(transport, &desc); 1510 continue; 1511 } 1512 /* Read in the call/reply flag */ 1513 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) { 1514 xs_tcp_read_calldir(transport, &desc); 1515 continue; 1516 } 1517 /* Read in the request data */ 1518 if (transport->tcp_flags & TCP_RCV_COPY_DATA) { 1519 xs_tcp_read_data(xprt, &desc); 1520 continue; 1521 } 1522 /* Skip over any trailing bytes on short reads */ 1523 xs_tcp_read_discard(transport, &desc); 1524 } while (desc.count); 1525 ret = len - desc.count; 1526 if (ret < rd_desc->count) 1527 rd_desc->count -= ret; 1528 else 1529 rd_desc->count = 0; 1530 trace_xs_tcp_data_recv(transport); 1531 dprintk("RPC: xs_tcp_data_recv done\n"); 1532 return ret; 1533 } 1534 1535 static void xs_tcp_data_receive(struct sock_xprt *transport) 1536 { 1537 struct rpc_xprt *xprt = &transport->xprt; 1538 struct sock *sk; 1539 read_descriptor_t rd_desc = { 1540 .arg.data = xprt, 1541 }; 1542 unsigned long total = 0; 1543 int read = 0; 1544 1545 restart: 1546 mutex_lock(&transport->recv_mutex); 1547 sk = transport->inet; 1548 if (sk == NULL) 1549 goto out; 1550 1551 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ 1552 for (;;) { 1553 rd_desc.count = RPC_TCP_READ_CHUNK_SZ; 1554 lock_sock(sk); 1555 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); 1556 if (rd_desc.count != 0 || read < 0) { 1557 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state); 1558 release_sock(sk); 1559 break; 1560 } 1561 release_sock(sk); 1562 total += read; 1563 if (need_resched()) { 1564 mutex_unlock(&transport->recv_mutex); 1565 cond_resched(); 1566 goto restart; 1567 } 1568 } 1569 if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state)) 1570 queue_work(xprtiod_workqueue, &transport->recv_worker); 1571 out: 1572 mutex_unlock(&transport->recv_mutex); 1573 trace_xs_tcp_data_ready(xprt, read, total); 1574 } 1575 1576 static void xs_tcp_data_receive_workfn(struct work_struct *work) 1577 { 1578 struct sock_xprt *transport = 1579 container_of(work, struct sock_xprt, recv_worker); 1580 xs_tcp_data_receive(transport); 1581 } 1582 1583 /** 1584 * xs_tcp_state_change - callback to handle TCP socket state changes 1585 * @sk: socket whose state has changed 1586 * 1587 */ 1588 static void xs_tcp_state_change(struct sock *sk) 1589 { 1590 struct rpc_xprt *xprt; 1591 struct sock_xprt *transport; 1592 1593 read_lock_bh(&sk->sk_callback_lock); 1594 if (!(xprt = xprt_from_sock(sk))) 1595 goto out; 1596 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1597 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1598 sk->sk_state, xprt_connected(xprt), 1599 sock_flag(sk, SOCK_DEAD), 1600 sock_flag(sk, SOCK_ZAPPED), 1601 sk->sk_shutdown); 1602 1603 transport = container_of(xprt, struct sock_xprt, xprt); 1604 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1605 switch (sk->sk_state) { 1606 case TCP_ESTABLISHED: 1607 spin_lock(&xprt->transport_lock); 1608 if (!xprt_test_and_set_connected(xprt)) { 1609 1610 /* Reset TCP record info */ 1611 transport->tcp_offset = 0; 1612 transport->tcp_reclen = 0; 1613 transport->tcp_copied = 0; 1614 transport->tcp_flags = 1615 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; 1616 xprt->connect_cookie++; 1617 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1618 xprt_clear_connecting(xprt); 1619 1620 xprt_wake_pending_tasks(xprt, -EAGAIN); 1621 } 1622 spin_unlock(&xprt->transport_lock); 1623 break; 1624 case TCP_FIN_WAIT1: 1625 /* The client initiated a shutdown of the socket */ 1626 xprt->connect_cookie++; 1627 xprt->reestablish_timeout = 0; 1628 set_bit(XPRT_CLOSING, &xprt->state); 1629 smp_mb__before_atomic(); 1630 clear_bit(XPRT_CONNECTED, &xprt->state); 1631 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1632 smp_mb__after_atomic(); 1633 break; 1634 case TCP_CLOSE_WAIT: 1635 /* The server initiated a shutdown of the socket */ 1636 xprt->connect_cookie++; 1637 clear_bit(XPRT_CONNECTED, &xprt->state); 1638 xs_tcp_force_close(xprt); 1639 /* fall through */ 1640 case TCP_CLOSING: 1641 /* 1642 * If the server closed down the connection, make sure that 1643 * we back off before reconnecting 1644 */ 1645 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1646 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1647 break; 1648 case TCP_LAST_ACK: 1649 set_bit(XPRT_CLOSING, &xprt->state); 1650 smp_mb__before_atomic(); 1651 clear_bit(XPRT_CONNECTED, &xprt->state); 1652 smp_mb__after_atomic(); 1653 break; 1654 case TCP_CLOSE: 1655 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1656 &transport->sock_state)) 1657 xprt_clear_connecting(xprt); 1658 if (sk->sk_err) 1659 xprt_wake_pending_tasks(xprt, -sk->sk_err); 1660 xs_sock_mark_closed(xprt); 1661 } 1662 out: 1663 read_unlock_bh(&sk->sk_callback_lock); 1664 } 1665 1666 static void xs_write_space(struct sock *sk) 1667 { 1668 struct socket_wq *wq; 1669 struct rpc_xprt *xprt; 1670 1671 if (!sk->sk_socket) 1672 return; 1673 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1674 1675 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1676 return; 1677 rcu_read_lock(); 1678 wq = rcu_dereference(sk->sk_wq); 1679 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0) 1680 goto out; 1681 1682 xprt_write_space(xprt); 1683 out: 1684 rcu_read_unlock(); 1685 } 1686 1687 /** 1688 * xs_udp_write_space - callback invoked when socket buffer space 1689 * becomes available 1690 * @sk: socket whose state has changed 1691 * 1692 * Called when more output buffer space is available for this socket. 1693 * We try not to wake our writers until they can make "significant" 1694 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1695 * with a bunch of small requests. 1696 */ 1697 static void xs_udp_write_space(struct sock *sk) 1698 { 1699 read_lock_bh(&sk->sk_callback_lock); 1700 1701 /* from net/core/sock.c:sock_def_write_space */ 1702 if (sock_writeable(sk)) 1703 xs_write_space(sk); 1704 1705 read_unlock_bh(&sk->sk_callback_lock); 1706 } 1707 1708 /** 1709 * xs_tcp_write_space - callback invoked when socket buffer space 1710 * becomes available 1711 * @sk: socket whose state has changed 1712 * 1713 * Called when more output buffer space is available for this socket. 1714 * We try not to wake our writers until they can make "significant" 1715 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1716 * with a bunch of small requests. 1717 */ 1718 static void xs_tcp_write_space(struct sock *sk) 1719 { 1720 read_lock_bh(&sk->sk_callback_lock); 1721 1722 /* from net/core/stream.c:sk_stream_write_space */ 1723 if (sk_stream_is_writeable(sk)) 1724 xs_write_space(sk); 1725 1726 read_unlock_bh(&sk->sk_callback_lock); 1727 } 1728 1729 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1730 { 1731 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1732 struct sock *sk = transport->inet; 1733 1734 if (transport->rcvsize) { 1735 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1736 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1737 } 1738 if (transport->sndsize) { 1739 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1740 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1741 sk->sk_write_space(sk); 1742 } 1743 } 1744 1745 /** 1746 * xs_udp_set_buffer_size - set send and receive limits 1747 * @xprt: generic transport 1748 * @sndsize: requested size of send buffer, in bytes 1749 * @rcvsize: requested size of receive buffer, in bytes 1750 * 1751 * Set socket send and receive buffer size limits. 1752 */ 1753 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1754 { 1755 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1756 1757 transport->sndsize = 0; 1758 if (sndsize) 1759 transport->sndsize = sndsize + 1024; 1760 transport->rcvsize = 0; 1761 if (rcvsize) 1762 transport->rcvsize = rcvsize + 1024; 1763 1764 xs_udp_do_set_buffer_size(xprt); 1765 } 1766 1767 /** 1768 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1769 * @task: task that timed out 1770 * 1771 * Adjust the congestion window after a retransmit timeout has occurred. 1772 */ 1773 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1774 { 1775 spin_lock_bh(&xprt->transport_lock); 1776 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1777 spin_unlock_bh(&xprt->transport_lock); 1778 } 1779 1780 static unsigned short xs_get_random_port(void) 1781 { 1782 unsigned short range = xprt_max_resvport - xprt_min_resvport + 1; 1783 unsigned short rand = (unsigned short) prandom_u32() % range; 1784 return rand + xprt_min_resvport; 1785 } 1786 1787 /** 1788 * xs_set_reuseaddr_port - set the socket's port and address reuse options 1789 * @sock: socket 1790 * 1791 * Note that this function has to be called on all sockets that share the 1792 * same port, and it must be called before binding. 1793 */ 1794 static void xs_sock_set_reuseport(struct socket *sock) 1795 { 1796 int opt = 1; 1797 1798 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, 1799 (char *)&opt, sizeof(opt)); 1800 } 1801 1802 static unsigned short xs_sock_getport(struct socket *sock) 1803 { 1804 struct sockaddr_storage buf; 1805 int buflen; 1806 unsigned short port = 0; 1807 1808 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0) 1809 goto out; 1810 switch (buf.ss_family) { 1811 case AF_INET6: 1812 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1813 break; 1814 case AF_INET: 1815 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1816 } 1817 out: 1818 return port; 1819 } 1820 1821 /** 1822 * xs_set_port - reset the port number in the remote endpoint address 1823 * @xprt: generic transport 1824 * @port: new port number 1825 * 1826 */ 1827 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1828 { 1829 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1830 1831 rpc_set_port(xs_addr(xprt), port); 1832 xs_update_peer_port(xprt); 1833 } 1834 1835 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1836 { 1837 if (transport->srcport == 0) 1838 transport->srcport = xs_sock_getport(sock); 1839 } 1840 1841 static unsigned short xs_get_srcport(struct sock_xprt *transport) 1842 { 1843 unsigned short port = transport->srcport; 1844 1845 if (port == 0 && transport->xprt.resvport) 1846 port = xs_get_random_port(); 1847 return port; 1848 } 1849 1850 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1851 { 1852 if (transport->srcport != 0) 1853 transport->srcport = 0; 1854 if (!transport->xprt.resvport) 1855 return 0; 1856 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1857 return xprt_max_resvport; 1858 return --port; 1859 } 1860 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1861 { 1862 struct sockaddr_storage myaddr; 1863 int err, nloop = 0; 1864 unsigned short port = xs_get_srcport(transport); 1865 unsigned short last; 1866 1867 /* 1868 * If we are asking for any ephemeral port (i.e. port == 0 && 1869 * transport->xprt.resvport == 0), don't bind. Let the local 1870 * port selection happen implicitly when the socket is used 1871 * (for example at connect time). 1872 * 1873 * This ensures that we can continue to establish TCP 1874 * connections even when all local ephemeral ports are already 1875 * a part of some TCP connection. This makes no difference 1876 * for UDP sockets, but also doens't harm them. 1877 * 1878 * If we're asking for any reserved port (i.e. port == 0 && 1879 * transport->xprt.resvport == 1) xs_get_srcport above will 1880 * ensure that port is non-zero and we will bind as needed. 1881 */ 1882 if (port == 0) 1883 return 0; 1884 1885 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1886 do { 1887 rpc_set_port((struct sockaddr *)&myaddr, port); 1888 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1889 transport->xprt.addrlen); 1890 if (err == 0) { 1891 transport->srcport = port; 1892 break; 1893 } 1894 last = port; 1895 port = xs_next_srcport(transport, port); 1896 if (port > last) 1897 nloop++; 1898 } while (err == -EADDRINUSE && nloop != 2); 1899 1900 if (myaddr.ss_family == AF_INET) 1901 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1902 &((struct sockaddr_in *)&myaddr)->sin_addr, 1903 port, err ? "failed" : "ok", err); 1904 else 1905 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1906 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1907 port, err ? "failed" : "ok", err); 1908 return err; 1909 } 1910 1911 /* 1912 * We don't support autobind on AF_LOCAL sockets 1913 */ 1914 static void xs_local_rpcbind(struct rpc_task *task) 1915 { 1916 xprt_set_bound(task->tk_xprt); 1917 } 1918 1919 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1920 { 1921 } 1922 1923 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1924 static struct lock_class_key xs_key[2]; 1925 static struct lock_class_key xs_slock_key[2]; 1926 1927 static inline void xs_reclassify_socketu(struct socket *sock) 1928 { 1929 struct sock *sk = sock->sk; 1930 1931 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1932 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); 1933 } 1934 1935 static inline void xs_reclassify_socket4(struct socket *sock) 1936 { 1937 struct sock *sk = sock->sk; 1938 1939 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1940 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1941 } 1942 1943 static inline void xs_reclassify_socket6(struct socket *sock) 1944 { 1945 struct sock *sk = sock->sk; 1946 1947 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1948 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1949 } 1950 1951 static inline void xs_reclassify_socket(int family, struct socket *sock) 1952 { 1953 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk))) 1954 return; 1955 1956 switch (family) { 1957 case AF_LOCAL: 1958 xs_reclassify_socketu(sock); 1959 break; 1960 case AF_INET: 1961 xs_reclassify_socket4(sock); 1962 break; 1963 case AF_INET6: 1964 xs_reclassify_socket6(sock); 1965 break; 1966 } 1967 } 1968 #else 1969 static inline void xs_reclassify_socket(int family, struct socket *sock) 1970 { 1971 } 1972 #endif 1973 1974 static void xs_dummy_setup_socket(struct work_struct *work) 1975 { 1976 } 1977 1978 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1979 struct sock_xprt *transport, int family, int type, 1980 int protocol, bool reuseport) 1981 { 1982 struct socket *sock; 1983 int err; 1984 1985 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1986 if (err < 0) { 1987 dprintk("RPC: can't create %d transport socket (%d).\n", 1988 protocol, -err); 1989 goto out; 1990 } 1991 xs_reclassify_socket(family, sock); 1992 1993 if (reuseport) 1994 xs_sock_set_reuseport(sock); 1995 1996 err = xs_bind(transport, sock); 1997 if (err) { 1998 sock_release(sock); 1999 goto out; 2000 } 2001 2002 return sock; 2003 out: 2004 return ERR_PTR(err); 2005 } 2006 2007 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 2008 struct socket *sock) 2009 { 2010 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2011 xprt); 2012 2013 if (!transport->inet) { 2014 struct sock *sk = sock->sk; 2015 2016 write_lock_bh(&sk->sk_callback_lock); 2017 2018 xs_save_old_callbacks(transport, sk); 2019 2020 sk->sk_user_data = xprt; 2021 sk->sk_data_ready = xs_data_ready; 2022 sk->sk_write_space = xs_udp_write_space; 2023 sock_set_flag(sk, SOCK_FASYNC); 2024 sk->sk_error_report = xs_error_report; 2025 sk->sk_allocation = GFP_NOIO; 2026 2027 xprt_clear_connected(xprt); 2028 2029 /* Reset to new socket */ 2030 transport->sock = sock; 2031 transport->inet = sk; 2032 2033 write_unlock_bh(&sk->sk_callback_lock); 2034 } 2035 2036 /* Tell the socket layer to start connecting... */ 2037 xprt->stat.connect_count++; 2038 xprt->stat.connect_start = jiffies; 2039 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 2040 } 2041 2042 /** 2043 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 2044 * @transport: socket transport to connect 2045 */ 2046 static int xs_local_setup_socket(struct sock_xprt *transport) 2047 { 2048 struct rpc_xprt *xprt = &transport->xprt; 2049 struct socket *sock; 2050 int status = -EIO; 2051 2052 status = __sock_create(xprt->xprt_net, AF_LOCAL, 2053 SOCK_STREAM, 0, &sock, 1); 2054 if (status < 0) { 2055 dprintk("RPC: can't create AF_LOCAL " 2056 "transport socket (%d).\n", -status); 2057 goto out; 2058 } 2059 xs_reclassify_socket(AF_LOCAL, sock); 2060 2061 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 2062 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2063 2064 status = xs_local_finish_connecting(xprt, sock); 2065 trace_rpc_socket_connect(xprt, sock, status); 2066 switch (status) { 2067 case 0: 2068 dprintk("RPC: xprt %p connected to %s\n", 2069 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2070 xprt_set_connected(xprt); 2071 case -ENOBUFS: 2072 break; 2073 case -ENOENT: 2074 dprintk("RPC: xprt %p: socket %s does not exist\n", 2075 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2076 break; 2077 case -ECONNREFUSED: 2078 dprintk("RPC: xprt %p: connection refused for %s\n", 2079 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2080 break; 2081 default: 2082 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 2083 __func__, -status, 2084 xprt->address_strings[RPC_DISPLAY_ADDR]); 2085 } 2086 2087 out: 2088 xprt_clear_connecting(xprt); 2089 xprt_wake_pending_tasks(xprt, status); 2090 return status; 2091 } 2092 2093 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2094 { 2095 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2096 int ret; 2097 2098 if (RPC_IS_ASYNC(task)) { 2099 /* 2100 * We want the AF_LOCAL connect to be resolved in the 2101 * filesystem namespace of the process making the rpc 2102 * call. Thus we connect synchronously. 2103 * 2104 * If we want to support asynchronous AF_LOCAL calls, 2105 * we'll need to figure out how to pass a namespace to 2106 * connect. 2107 */ 2108 rpc_exit(task, -ENOTCONN); 2109 return; 2110 } 2111 ret = xs_local_setup_socket(transport); 2112 if (ret && !RPC_IS_SOFTCONN(task)) 2113 msleep_interruptible(15000); 2114 } 2115 2116 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 2117 /* 2118 * Note that this should be called with XPRT_LOCKED held (or when we otherwise 2119 * know that we have exclusive access to the socket), to guard against 2120 * races with xs_reset_transport. 2121 */ 2122 static void xs_set_memalloc(struct rpc_xprt *xprt) 2123 { 2124 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2125 xprt); 2126 2127 /* 2128 * If there's no sock, then we have nothing to set. The 2129 * reconnecting process will get it for us. 2130 */ 2131 if (!transport->inet) 2132 return; 2133 if (atomic_read(&xprt->swapper)) 2134 sk_set_memalloc(transport->inet); 2135 } 2136 2137 /** 2138 * xs_enable_swap - Tag this transport as being used for swap. 2139 * @xprt: transport to tag 2140 * 2141 * Take a reference to this transport on behalf of the rpc_clnt, and 2142 * optionally mark it for swapping if it wasn't already. 2143 */ 2144 static int 2145 xs_enable_swap(struct rpc_xprt *xprt) 2146 { 2147 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2148 2149 if (atomic_inc_return(&xprt->swapper) != 1) 2150 return 0; 2151 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 2152 return -ERESTARTSYS; 2153 if (xs->inet) 2154 sk_set_memalloc(xs->inet); 2155 xprt_release_xprt(xprt, NULL); 2156 return 0; 2157 } 2158 2159 /** 2160 * xs_disable_swap - Untag this transport as being used for swap. 2161 * @xprt: transport to tag 2162 * 2163 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2164 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2165 */ 2166 static void 2167 xs_disable_swap(struct rpc_xprt *xprt) 2168 { 2169 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2170 2171 if (!atomic_dec_and_test(&xprt->swapper)) 2172 return; 2173 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 2174 return; 2175 if (xs->inet) 2176 sk_clear_memalloc(xs->inet); 2177 xprt_release_xprt(xprt, NULL); 2178 } 2179 #else 2180 static void xs_set_memalloc(struct rpc_xprt *xprt) 2181 { 2182 } 2183 2184 static int 2185 xs_enable_swap(struct rpc_xprt *xprt) 2186 { 2187 return -EINVAL; 2188 } 2189 2190 static void 2191 xs_disable_swap(struct rpc_xprt *xprt) 2192 { 2193 } 2194 #endif 2195 2196 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2197 { 2198 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2199 2200 if (!transport->inet) { 2201 struct sock *sk = sock->sk; 2202 2203 write_lock_bh(&sk->sk_callback_lock); 2204 2205 xs_save_old_callbacks(transport, sk); 2206 2207 sk->sk_user_data = xprt; 2208 sk->sk_data_ready = xs_data_ready; 2209 sk->sk_write_space = xs_udp_write_space; 2210 sock_set_flag(sk, SOCK_FASYNC); 2211 sk->sk_allocation = GFP_NOIO; 2212 2213 xprt_set_connected(xprt); 2214 2215 /* Reset to new socket */ 2216 transport->sock = sock; 2217 transport->inet = sk; 2218 2219 xs_set_memalloc(xprt); 2220 2221 write_unlock_bh(&sk->sk_callback_lock); 2222 } 2223 xs_udp_do_set_buffer_size(xprt); 2224 2225 xprt->stat.connect_start = jiffies; 2226 } 2227 2228 static void xs_udp_setup_socket(struct work_struct *work) 2229 { 2230 struct sock_xprt *transport = 2231 container_of(work, struct sock_xprt, connect_worker.work); 2232 struct rpc_xprt *xprt = &transport->xprt; 2233 struct socket *sock; 2234 int status = -EIO; 2235 2236 sock = xs_create_sock(xprt, transport, 2237 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2238 IPPROTO_UDP, false); 2239 if (IS_ERR(sock)) 2240 goto out; 2241 2242 dprintk("RPC: worker connecting xprt %p via %s to " 2243 "%s (port %s)\n", xprt, 2244 xprt->address_strings[RPC_DISPLAY_PROTO], 2245 xprt->address_strings[RPC_DISPLAY_ADDR], 2246 xprt->address_strings[RPC_DISPLAY_PORT]); 2247 2248 xs_udp_finish_connecting(xprt, sock); 2249 trace_rpc_socket_connect(xprt, sock, 0); 2250 status = 0; 2251 out: 2252 xprt_unlock_connect(xprt, transport); 2253 xprt_clear_connecting(xprt); 2254 xprt_wake_pending_tasks(xprt, status); 2255 } 2256 2257 /** 2258 * xs_tcp_shutdown - gracefully shut down a TCP socket 2259 * @xprt: transport 2260 * 2261 * Initiates a graceful shutdown of the TCP socket by calling the 2262 * equivalent of shutdown(SHUT_RDWR); 2263 */ 2264 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2265 { 2266 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2267 struct socket *sock = transport->sock; 2268 2269 if (sock == NULL) 2270 return; 2271 if (xprt_connected(xprt)) { 2272 kernel_sock_shutdown(sock, SHUT_RDWR); 2273 trace_rpc_socket_shutdown(xprt, sock); 2274 } else 2275 xs_reset_transport(transport); 2276 } 2277 2278 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt, 2279 struct socket *sock) 2280 { 2281 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2282 unsigned int keepidle; 2283 unsigned int keepcnt; 2284 unsigned int opt_on = 1; 2285 unsigned int timeo; 2286 2287 spin_lock_bh(&xprt->transport_lock); 2288 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ); 2289 keepcnt = xprt->timeout->to_retries + 1; 2290 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2291 (xprt->timeout->to_retries + 1); 2292 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2293 spin_unlock_bh(&xprt->transport_lock); 2294 2295 /* TCP Keepalive options */ 2296 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 2297 (char *)&opt_on, sizeof(opt_on)); 2298 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, 2299 (char *)&keepidle, sizeof(keepidle)); 2300 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, 2301 (char *)&keepidle, sizeof(keepidle)); 2302 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, 2303 (char *)&keepcnt, sizeof(keepcnt)); 2304 2305 /* TCP user timeout (see RFC5482) */ 2306 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT, 2307 (char *)&timeo, sizeof(timeo)); 2308 } 2309 2310 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt, 2311 unsigned long connect_timeout, 2312 unsigned long reconnect_timeout) 2313 { 2314 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2315 struct rpc_timeout to; 2316 unsigned long initval; 2317 2318 spin_lock_bh(&xprt->transport_lock); 2319 if (reconnect_timeout < xprt->max_reconnect_timeout) 2320 xprt->max_reconnect_timeout = reconnect_timeout; 2321 if (connect_timeout < xprt->connect_timeout) { 2322 memcpy(&to, xprt->timeout, sizeof(to)); 2323 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1); 2324 /* Arbitrary lower limit */ 2325 if (initval < XS_TCP_INIT_REEST_TO << 1) 2326 initval = XS_TCP_INIT_REEST_TO << 1; 2327 to.to_initval = initval; 2328 to.to_maxval = initval; 2329 memcpy(&transport->tcp_timeout, &to, 2330 sizeof(transport->tcp_timeout)); 2331 xprt->timeout = &transport->tcp_timeout; 2332 xprt->connect_timeout = connect_timeout; 2333 } 2334 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state); 2335 spin_unlock_bh(&xprt->transport_lock); 2336 } 2337 2338 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2339 { 2340 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2341 int ret = -ENOTCONN; 2342 2343 if (!transport->inet) { 2344 struct sock *sk = sock->sk; 2345 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC; 2346 2347 /* Avoid temporary address, they are bad for long-lived 2348 * connections such as NFS mounts. 2349 * RFC4941, section 3.6 suggests that: 2350 * Individual applications, which have specific 2351 * knowledge about the normal duration of connections, 2352 * MAY override this as appropriate. 2353 */ 2354 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES, 2355 (char *)&addr_pref, sizeof(addr_pref)); 2356 2357 xs_tcp_set_socket_timeouts(xprt, sock); 2358 2359 write_lock_bh(&sk->sk_callback_lock); 2360 2361 xs_save_old_callbacks(transport, sk); 2362 2363 sk->sk_user_data = xprt; 2364 sk->sk_data_ready = xs_data_ready; 2365 sk->sk_state_change = xs_tcp_state_change; 2366 sk->sk_write_space = xs_tcp_write_space; 2367 sock_set_flag(sk, SOCK_FASYNC); 2368 sk->sk_error_report = xs_error_report; 2369 sk->sk_allocation = GFP_NOIO; 2370 2371 /* socket options */ 2372 sock_reset_flag(sk, SOCK_LINGER); 2373 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 2374 2375 xprt_clear_connected(xprt); 2376 2377 /* Reset to new socket */ 2378 transport->sock = sock; 2379 transport->inet = sk; 2380 2381 write_unlock_bh(&sk->sk_callback_lock); 2382 } 2383 2384 if (!xprt_bound(xprt)) 2385 goto out; 2386 2387 xs_set_memalloc(xprt); 2388 2389 /* Tell the socket layer to start connecting... */ 2390 xprt->stat.connect_count++; 2391 xprt->stat.connect_start = jiffies; 2392 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2393 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2394 switch (ret) { 2395 case 0: 2396 xs_set_srcport(transport, sock); 2397 /* fall through */ 2398 case -EINPROGRESS: 2399 /* SYN_SENT! */ 2400 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2401 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2402 break; 2403 case -EADDRNOTAVAIL: 2404 /* Source port number is unavailable. Try a new one! */ 2405 transport->srcport = 0; 2406 } 2407 out: 2408 return ret; 2409 } 2410 2411 /** 2412 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2413 * 2414 * Invoked by a work queue tasklet. 2415 */ 2416 static void xs_tcp_setup_socket(struct work_struct *work) 2417 { 2418 struct sock_xprt *transport = 2419 container_of(work, struct sock_xprt, connect_worker.work); 2420 struct socket *sock = transport->sock; 2421 struct rpc_xprt *xprt = &transport->xprt; 2422 int status = -EIO; 2423 2424 if (!sock) { 2425 sock = xs_create_sock(xprt, transport, 2426 xs_addr(xprt)->sa_family, SOCK_STREAM, 2427 IPPROTO_TCP, true); 2428 if (IS_ERR(sock)) { 2429 status = PTR_ERR(sock); 2430 goto out; 2431 } 2432 } 2433 2434 dprintk("RPC: worker connecting xprt %p via %s to " 2435 "%s (port %s)\n", xprt, 2436 xprt->address_strings[RPC_DISPLAY_PROTO], 2437 xprt->address_strings[RPC_DISPLAY_ADDR], 2438 xprt->address_strings[RPC_DISPLAY_PORT]); 2439 2440 status = xs_tcp_finish_connecting(xprt, sock); 2441 trace_rpc_socket_connect(xprt, sock, status); 2442 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2443 xprt, -status, xprt_connected(xprt), 2444 sock->sk->sk_state); 2445 switch (status) { 2446 default: 2447 printk("%s: connect returned unhandled error %d\n", 2448 __func__, status); 2449 /* fall through */ 2450 case -EADDRNOTAVAIL: 2451 /* We're probably in TIME_WAIT. Get rid of existing socket, 2452 * and retry 2453 */ 2454 xs_tcp_force_close(xprt); 2455 break; 2456 case 0: 2457 case -EINPROGRESS: 2458 case -EALREADY: 2459 xprt_unlock_connect(xprt, transport); 2460 return; 2461 case -EINVAL: 2462 /* Happens, for instance, if the user specified a link 2463 * local IPv6 address without a scope-id. 2464 */ 2465 case -ECONNREFUSED: 2466 case -ECONNRESET: 2467 case -ENETDOWN: 2468 case -ENETUNREACH: 2469 case -EHOSTUNREACH: 2470 case -EADDRINUSE: 2471 case -ENOBUFS: 2472 /* 2473 * xs_tcp_force_close() wakes tasks with -EIO. 2474 * We need to wake them first to ensure the 2475 * correct error code. 2476 */ 2477 xprt_wake_pending_tasks(xprt, status); 2478 xs_tcp_force_close(xprt); 2479 goto out; 2480 } 2481 status = -EAGAIN; 2482 out: 2483 xprt_unlock_connect(xprt, transport); 2484 xprt_clear_connecting(xprt); 2485 xprt_wake_pending_tasks(xprt, status); 2486 } 2487 2488 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt) 2489 { 2490 unsigned long start, now = jiffies; 2491 2492 start = xprt->stat.connect_start + xprt->reestablish_timeout; 2493 if (time_after(start, now)) 2494 return start - now; 2495 return 0; 2496 } 2497 2498 static void xs_reconnect_backoff(struct rpc_xprt *xprt) 2499 { 2500 xprt->reestablish_timeout <<= 1; 2501 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout) 2502 xprt->reestablish_timeout = xprt->max_reconnect_timeout; 2503 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2504 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2505 } 2506 2507 /** 2508 * xs_connect - connect a socket to a remote endpoint 2509 * @xprt: pointer to transport structure 2510 * @task: address of RPC task that manages state of connect request 2511 * 2512 * TCP: If the remote end dropped the connection, delay reconnecting. 2513 * 2514 * UDP socket connects are synchronous, but we use a work queue anyway 2515 * to guarantee that even unprivileged user processes can set up a 2516 * socket on a privileged port. 2517 * 2518 * If a UDP socket connect fails, the delay behavior here prevents 2519 * retry floods (hard mounts). 2520 */ 2521 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2522 { 2523 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2524 unsigned long delay = 0; 2525 2526 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2527 2528 if (transport->sock != NULL) { 2529 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2530 "seconds\n", 2531 xprt, xprt->reestablish_timeout / HZ); 2532 2533 /* Start by resetting any existing state */ 2534 xs_reset_transport(transport); 2535 2536 delay = xs_reconnect_delay(xprt); 2537 xs_reconnect_backoff(xprt); 2538 2539 } else 2540 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2541 2542 queue_delayed_work(xprtiod_workqueue, 2543 &transport->connect_worker, 2544 delay); 2545 } 2546 2547 /** 2548 * xs_local_print_stats - display AF_LOCAL socket-specifc stats 2549 * @xprt: rpc_xprt struct containing statistics 2550 * @seq: output file 2551 * 2552 */ 2553 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2554 { 2555 long idle_time = 0; 2556 2557 if (xprt_connected(xprt)) 2558 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2559 2560 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2561 "%llu %llu %lu %llu %llu\n", 2562 xprt->stat.bind_count, 2563 xprt->stat.connect_count, 2564 xprt->stat.connect_time, 2565 idle_time, 2566 xprt->stat.sends, 2567 xprt->stat.recvs, 2568 xprt->stat.bad_xids, 2569 xprt->stat.req_u, 2570 xprt->stat.bklog_u, 2571 xprt->stat.max_slots, 2572 xprt->stat.sending_u, 2573 xprt->stat.pending_u); 2574 } 2575 2576 /** 2577 * xs_udp_print_stats - display UDP socket-specifc stats 2578 * @xprt: rpc_xprt struct containing statistics 2579 * @seq: output file 2580 * 2581 */ 2582 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2583 { 2584 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2585 2586 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2587 "%lu %llu %llu\n", 2588 transport->srcport, 2589 xprt->stat.bind_count, 2590 xprt->stat.sends, 2591 xprt->stat.recvs, 2592 xprt->stat.bad_xids, 2593 xprt->stat.req_u, 2594 xprt->stat.bklog_u, 2595 xprt->stat.max_slots, 2596 xprt->stat.sending_u, 2597 xprt->stat.pending_u); 2598 } 2599 2600 /** 2601 * xs_tcp_print_stats - display TCP socket-specifc stats 2602 * @xprt: rpc_xprt struct containing statistics 2603 * @seq: output file 2604 * 2605 */ 2606 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2607 { 2608 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2609 long idle_time = 0; 2610 2611 if (xprt_connected(xprt)) 2612 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2613 2614 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2615 "%llu %llu %lu %llu %llu\n", 2616 transport->srcport, 2617 xprt->stat.bind_count, 2618 xprt->stat.connect_count, 2619 xprt->stat.connect_time, 2620 idle_time, 2621 xprt->stat.sends, 2622 xprt->stat.recvs, 2623 xprt->stat.bad_xids, 2624 xprt->stat.req_u, 2625 xprt->stat.bklog_u, 2626 xprt->stat.max_slots, 2627 xprt->stat.sending_u, 2628 xprt->stat.pending_u); 2629 } 2630 2631 /* 2632 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2633 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2634 * to use the server side send routines. 2635 */ 2636 static int bc_malloc(struct rpc_task *task) 2637 { 2638 struct rpc_rqst *rqst = task->tk_rqstp; 2639 size_t size = rqst->rq_callsize; 2640 struct page *page; 2641 struct rpc_buffer *buf; 2642 2643 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) { 2644 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n", 2645 size); 2646 return -EINVAL; 2647 } 2648 2649 page = alloc_page(GFP_KERNEL); 2650 if (!page) 2651 return -ENOMEM; 2652 2653 buf = page_address(page); 2654 buf->len = PAGE_SIZE; 2655 2656 rqst->rq_buffer = buf->data; 2657 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; 2658 return 0; 2659 } 2660 2661 /* 2662 * Free the space allocated in the bc_alloc routine 2663 */ 2664 static void bc_free(struct rpc_task *task) 2665 { 2666 void *buffer = task->tk_rqstp->rq_buffer; 2667 struct rpc_buffer *buf; 2668 2669 buf = container_of(buffer, struct rpc_buffer, data); 2670 free_page((unsigned long)buf); 2671 } 2672 2673 /* 2674 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex 2675 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request. 2676 */ 2677 static int bc_sendto(struct rpc_rqst *req) 2678 { 2679 int len; 2680 struct xdr_buf *xbufp = &req->rq_snd_buf; 2681 struct rpc_xprt *xprt = req->rq_xprt; 2682 struct sock_xprt *transport = 2683 container_of(xprt, struct sock_xprt, xprt); 2684 struct socket *sock = transport->sock; 2685 unsigned long headoff; 2686 unsigned long tailoff; 2687 2688 xs_encode_stream_record_marker(xbufp); 2689 2690 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK; 2691 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK; 2692 len = svc_send_common(sock, xbufp, 2693 virt_to_page(xbufp->head[0].iov_base), headoff, 2694 xbufp->tail[0].iov_base, tailoff); 2695 2696 if (len != xbufp->len) { 2697 printk(KERN_NOTICE "Error sending entire callback!\n"); 2698 len = -EAGAIN; 2699 } 2700 2701 return len; 2702 } 2703 2704 /* 2705 * The send routine. Borrows from svc_send 2706 */ 2707 static int bc_send_request(struct rpc_task *task) 2708 { 2709 struct rpc_rqst *req = task->tk_rqstp; 2710 struct svc_xprt *xprt; 2711 int len; 2712 2713 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid)); 2714 /* 2715 * Get the server socket associated with this callback xprt 2716 */ 2717 xprt = req->rq_xprt->bc_xprt; 2718 2719 /* 2720 * Grab the mutex to serialize data as the connection is shared 2721 * with the fore channel 2722 */ 2723 if (!mutex_trylock(&xprt->xpt_mutex)) { 2724 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL); 2725 if (!mutex_trylock(&xprt->xpt_mutex)) 2726 return -EAGAIN; 2727 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task); 2728 } 2729 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2730 len = -ENOTCONN; 2731 else 2732 len = bc_sendto(req); 2733 mutex_unlock(&xprt->xpt_mutex); 2734 2735 if (len > 0) 2736 len = 0; 2737 2738 return len; 2739 } 2740 2741 /* 2742 * The close routine. Since this is client initiated, we do nothing 2743 */ 2744 2745 static void bc_close(struct rpc_xprt *xprt) 2746 { 2747 } 2748 2749 /* 2750 * The xprt destroy routine. Again, because this connection is client 2751 * initiated, we do nothing 2752 */ 2753 2754 static void bc_destroy(struct rpc_xprt *xprt) 2755 { 2756 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2757 2758 xs_xprt_free(xprt); 2759 module_put(THIS_MODULE); 2760 } 2761 2762 static const struct rpc_xprt_ops xs_local_ops = { 2763 .reserve_xprt = xprt_reserve_xprt, 2764 .release_xprt = xs_tcp_release_xprt, 2765 .alloc_slot = xprt_alloc_slot, 2766 .rpcbind = xs_local_rpcbind, 2767 .set_port = xs_local_set_port, 2768 .connect = xs_local_connect, 2769 .buf_alloc = rpc_malloc, 2770 .buf_free = rpc_free, 2771 .send_request = xs_local_send_request, 2772 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2773 .close = xs_close, 2774 .destroy = xs_destroy, 2775 .print_stats = xs_local_print_stats, 2776 .enable_swap = xs_enable_swap, 2777 .disable_swap = xs_disable_swap, 2778 }; 2779 2780 static const struct rpc_xprt_ops xs_udp_ops = { 2781 .set_buffer_size = xs_udp_set_buffer_size, 2782 .reserve_xprt = xprt_reserve_xprt_cong, 2783 .release_xprt = xprt_release_xprt_cong, 2784 .alloc_slot = xprt_alloc_slot, 2785 .rpcbind = rpcb_getport_async, 2786 .set_port = xs_set_port, 2787 .connect = xs_connect, 2788 .buf_alloc = rpc_malloc, 2789 .buf_free = rpc_free, 2790 .send_request = xs_udp_send_request, 2791 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 2792 .timer = xs_udp_timer, 2793 .release_request = xprt_release_rqst_cong, 2794 .close = xs_close, 2795 .destroy = xs_destroy, 2796 .print_stats = xs_udp_print_stats, 2797 .enable_swap = xs_enable_swap, 2798 .disable_swap = xs_disable_swap, 2799 .inject_disconnect = xs_inject_disconnect, 2800 }; 2801 2802 static const struct rpc_xprt_ops xs_tcp_ops = { 2803 .reserve_xprt = xprt_reserve_xprt, 2804 .release_xprt = xs_tcp_release_xprt, 2805 .alloc_slot = xprt_lock_and_alloc_slot, 2806 .rpcbind = rpcb_getport_async, 2807 .set_port = xs_set_port, 2808 .connect = xs_connect, 2809 .buf_alloc = rpc_malloc, 2810 .buf_free = rpc_free, 2811 .send_request = xs_tcp_send_request, 2812 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2813 .close = xs_tcp_shutdown, 2814 .destroy = xs_destroy, 2815 .set_connect_timeout = xs_tcp_set_connect_timeout, 2816 .print_stats = xs_tcp_print_stats, 2817 .enable_swap = xs_enable_swap, 2818 .disable_swap = xs_disable_swap, 2819 .inject_disconnect = xs_inject_disconnect, 2820 #ifdef CONFIG_SUNRPC_BACKCHANNEL 2821 .bc_setup = xprt_setup_bc, 2822 .bc_up = xs_tcp_bc_up, 2823 .bc_maxpayload = xs_tcp_bc_maxpayload, 2824 .bc_free_rqst = xprt_free_bc_rqst, 2825 .bc_destroy = xprt_destroy_bc, 2826 #endif 2827 }; 2828 2829 /* 2830 * The rpc_xprt_ops for the server backchannel 2831 */ 2832 2833 static const struct rpc_xprt_ops bc_tcp_ops = { 2834 .reserve_xprt = xprt_reserve_xprt, 2835 .release_xprt = xprt_release_xprt, 2836 .alloc_slot = xprt_alloc_slot, 2837 .buf_alloc = bc_malloc, 2838 .buf_free = bc_free, 2839 .send_request = bc_send_request, 2840 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2841 .close = bc_close, 2842 .destroy = bc_destroy, 2843 .print_stats = xs_tcp_print_stats, 2844 .enable_swap = xs_enable_swap, 2845 .disable_swap = xs_disable_swap, 2846 .inject_disconnect = xs_inject_disconnect, 2847 }; 2848 2849 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2850 { 2851 static const struct sockaddr_in sin = { 2852 .sin_family = AF_INET, 2853 .sin_addr.s_addr = htonl(INADDR_ANY), 2854 }; 2855 static const struct sockaddr_in6 sin6 = { 2856 .sin6_family = AF_INET6, 2857 .sin6_addr = IN6ADDR_ANY_INIT, 2858 }; 2859 2860 switch (family) { 2861 case AF_LOCAL: 2862 break; 2863 case AF_INET: 2864 memcpy(sap, &sin, sizeof(sin)); 2865 break; 2866 case AF_INET6: 2867 memcpy(sap, &sin6, sizeof(sin6)); 2868 break; 2869 default: 2870 dprintk("RPC: %s: Bad address family\n", __func__); 2871 return -EAFNOSUPPORT; 2872 } 2873 return 0; 2874 } 2875 2876 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2877 unsigned int slot_table_size, 2878 unsigned int max_slot_table_size) 2879 { 2880 struct rpc_xprt *xprt; 2881 struct sock_xprt *new; 2882 2883 if (args->addrlen > sizeof(xprt->addr)) { 2884 dprintk("RPC: xs_setup_xprt: address too large\n"); 2885 return ERR_PTR(-EBADF); 2886 } 2887 2888 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2889 max_slot_table_size); 2890 if (xprt == NULL) { 2891 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2892 "rpc_xprt\n"); 2893 return ERR_PTR(-ENOMEM); 2894 } 2895 2896 new = container_of(xprt, struct sock_xprt, xprt); 2897 mutex_init(&new->recv_mutex); 2898 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2899 xprt->addrlen = args->addrlen; 2900 if (args->srcaddr) 2901 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2902 else { 2903 int err; 2904 err = xs_init_anyaddr(args->dstaddr->sa_family, 2905 (struct sockaddr *)&new->srcaddr); 2906 if (err != 0) { 2907 xprt_free(xprt); 2908 return ERR_PTR(err); 2909 } 2910 } 2911 2912 return xprt; 2913 } 2914 2915 static const struct rpc_timeout xs_local_default_timeout = { 2916 .to_initval = 10 * HZ, 2917 .to_maxval = 10 * HZ, 2918 .to_retries = 2, 2919 }; 2920 2921 /** 2922 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2923 * @args: rpc transport creation arguments 2924 * 2925 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2926 */ 2927 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2928 { 2929 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2930 struct sock_xprt *transport; 2931 struct rpc_xprt *xprt; 2932 struct rpc_xprt *ret; 2933 2934 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2935 xprt_max_tcp_slot_table_entries); 2936 if (IS_ERR(xprt)) 2937 return xprt; 2938 transport = container_of(xprt, struct sock_xprt, xprt); 2939 2940 xprt->prot = 0; 2941 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2942 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2943 2944 xprt->bind_timeout = XS_BIND_TO; 2945 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2946 xprt->idle_timeout = XS_IDLE_DISC_TO; 2947 2948 xprt->ops = &xs_local_ops; 2949 xprt->timeout = &xs_local_default_timeout; 2950 2951 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn); 2952 INIT_DELAYED_WORK(&transport->connect_worker, 2953 xs_dummy_setup_socket); 2954 2955 switch (sun->sun_family) { 2956 case AF_LOCAL: 2957 if (sun->sun_path[0] != '/') { 2958 dprintk("RPC: bad AF_LOCAL address: %s\n", 2959 sun->sun_path); 2960 ret = ERR_PTR(-EINVAL); 2961 goto out_err; 2962 } 2963 xprt_set_bound(xprt); 2964 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2965 ret = ERR_PTR(xs_local_setup_socket(transport)); 2966 if (ret) 2967 goto out_err; 2968 break; 2969 default: 2970 ret = ERR_PTR(-EAFNOSUPPORT); 2971 goto out_err; 2972 } 2973 2974 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2975 xprt->address_strings[RPC_DISPLAY_ADDR]); 2976 2977 if (try_module_get(THIS_MODULE)) 2978 return xprt; 2979 ret = ERR_PTR(-EINVAL); 2980 out_err: 2981 xs_xprt_free(xprt); 2982 return ret; 2983 } 2984 2985 static const struct rpc_timeout xs_udp_default_timeout = { 2986 .to_initval = 5 * HZ, 2987 .to_maxval = 30 * HZ, 2988 .to_increment = 5 * HZ, 2989 .to_retries = 5, 2990 }; 2991 2992 /** 2993 * xs_setup_udp - Set up transport to use a UDP socket 2994 * @args: rpc transport creation arguments 2995 * 2996 */ 2997 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2998 { 2999 struct sockaddr *addr = args->dstaddr; 3000 struct rpc_xprt *xprt; 3001 struct sock_xprt *transport; 3002 struct rpc_xprt *ret; 3003 3004 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 3005 xprt_udp_slot_table_entries); 3006 if (IS_ERR(xprt)) 3007 return xprt; 3008 transport = container_of(xprt, struct sock_xprt, xprt); 3009 3010 xprt->prot = IPPROTO_UDP; 3011 xprt->tsh_size = 0; 3012 /* XXX: header size can vary due to auth type, IPv6, etc. */ 3013 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 3014 3015 xprt->bind_timeout = XS_BIND_TO; 3016 xprt->reestablish_timeout = XS_UDP_REEST_TO; 3017 xprt->idle_timeout = XS_IDLE_DISC_TO; 3018 3019 xprt->ops = &xs_udp_ops; 3020 3021 xprt->timeout = &xs_udp_default_timeout; 3022 3023 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 3024 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 3025 3026 switch (addr->sa_family) { 3027 case AF_INET: 3028 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3029 xprt_set_bound(xprt); 3030 3031 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 3032 break; 3033 case AF_INET6: 3034 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3035 xprt_set_bound(xprt); 3036 3037 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 3038 break; 3039 default: 3040 ret = ERR_PTR(-EAFNOSUPPORT); 3041 goto out_err; 3042 } 3043 3044 if (xprt_bound(xprt)) 3045 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3046 xprt->address_strings[RPC_DISPLAY_ADDR], 3047 xprt->address_strings[RPC_DISPLAY_PORT], 3048 xprt->address_strings[RPC_DISPLAY_PROTO]); 3049 else 3050 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3051 xprt->address_strings[RPC_DISPLAY_ADDR], 3052 xprt->address_strings[RPC_DISPLAY_PROTO]); 3053 3054 if (try_module_get(THIS_MODULE)) 3055 return xprt; 3056 ret = ERR_PTR(-EINVAL); 3057 out_err: 3058 xs_xprt_free(xprt); 3059 return ret; 3060 } 3061 3062 static const struct rpc_timeout xs_tcp_default_timeout = { 3063 .to_initval = 60 * HZ, 3064 .to_maxval = 60 * HZ, 3065 .to_retries = 2, 3066 }; 3067 3068 /** 3069 * xs_setup_tcp - Set up transport to use a TCP socket 3070 * @args: rpc transport creation arguments 3071 * 3072 */ 3073 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 3074 { 3075 struct sockaddr *addr = args->dstaddr; 3076 struct rpc_xprt *xprt; 3077 struct sock_xprt *transport; 3078 struct rpc_xprt *ret; 3079 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 3080 3081 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 3082 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 3083 3084 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3085 max_slot_table_size); 3086 if (IS_ERR(xprt)) 3087 return xprt; 3088 transport = container_of(xprt, struct sock_xprt, xprt); 3089 3090 xprt->prot = IPPROTO_TCP; 3091 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 3092 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3093 3094 xprt->bind_timeout = XS_BIND_TO; 3095 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 3096 xprt->idle_timeout = XS_IDLE_DISC_TO; 3097 3098 xprt->ops = &xs_tcp_ops; 3099 xprt->timeout = &xs_tcp_default_timeout; 3100 3101 xprt->max_reconnect_timeout = xprt->timeout->to_maxval; 3102 xprt->connect_timeout = xprt->timeout->to_initval * 3103 (xprt->timeout->to_retries + 1); 3104 3105 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn); 3106 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 3107 3108 switch (addr->sa_family) { 3109 case AF_INET: 3110 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 3111 xprt_set_bound(xprt); 3112 3113 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 3114 break; 3115 case AF_INET6: 3116 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 3117 xprt_set_bound(xprt); 3118 3119 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 3120 break; 3121 default: 3122 ret = ERR_PTR(-EAFNOSUPPORT); 3123 goto out_err; 3124 } 3125 3126 if (xprt_bound(xprt)) 3127 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3128 xprt->address_strings[RPC_DISPLAY_ADDR], 3129 xprt->address_strings[RPC_DISPLAY_PORT], 3130 xprt->address_strings[RPC_DISPLAY_PROTO]); 3131 else 3132 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 3133 xprt->address_strings[RPC_DISPLAY_ADDR], 3134 xprt->address_strings[RPC_DISPLAY_PROTO]); 3135 3136 if (try_module_get(THIS_MODULE)) 3137 return xprt; 3138 ret = ERR_PTR(-EINVAL); 3139 out_err: 3140 xs_xprt_free(xprt); 3141 return ret; 3142 } 3143 3144 /** 3145 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 3146 * @args: rpc transport creation arguments 3147 * 3148 */ 3149 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 3150 { 3151 struct sockaddr *addr = args->dstaddr; 3152 struct rpc_xprt *xprt; 3153 struct sock_xprt *transport; 3154 struct svc_sock *bc_sock; 3155 struct rpc_xprt *ret; 3156 3157 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3158 xprt_tcp_slot_table_entries); 3159 if (IS_ERR(xprt)) 3160 return xprt; 3161 transport = container_of(xprt, struct sock_xprt, xprt); 3162 3163 xprt->prot = IPPROTO_TCP; 3164 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 3165 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3166 xprt->timeout = &xs_tcp_default_timeout; 3167 3168 /* backchannel */ 3169 xprt_set_bound(xprt); 3170 xprt->bind_timeout = 0; 3171 xprt->reestablish_timeout = 0; 3172 xprt->idle_timeout = 0; 3173 3174 xprt->ops = &bc_tcp_ops; 3175 3176 switch (addr->sa_family) { 3177 case AF_INET: 3178 xs_format_peer_addresses(xprt, "tcp", 3179 RPCBIND_NETID_TCP); 3180 break; 3181 case AF_INET6: 3182 xs_format_peer_addresses(xprt, "tcp", 3183 RPCBIND_NETID_TCP6); 3184 break; 3185 default: 3186 ret = ERR_PTR(-EAFNOSUPPORT); 3187 goto out_err; 3188 } 3189 3190 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3191 xprt->address_strings[RPC_DISPLAY_ADDR], 3192 xprt->address_strings[RPC_DISPLAY_PORT], 3193 xprt->address_strings[RPC_DISPLAY_PROTO]); 3194 3195 /* 3196 * Once we've associated a backchannel xprt with a connection, 3197 * we want to keep it around as long as the connection lasts, 3198 * in case we need to start using it for a backchannel again; 3199 * this reference won't be dropped until bc_xprt is destroyed. 3200 */ 3201 xprt_get(xprt); 3202 args->bc_xprt->xpt_bc_xprt = xprt; 3203 xprt->bc_xprt = args->bc_xprt; 3204 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3205 transport->sock = bc_sock->sk_sock; 3206 transport->inet = bc_sock->sk_sk; 3207 3208 /* 3209 * Since we don't want connections for the backchannel, we set 3210 * the xprt status to connected 3211 */ 3212 xprt_set_connected(xprt); 3213 3214 if (try_module_get(THIS_MODULE)) 3215 return xprt; 3216 3217 args->bc_xprt->xpt_bc_xprt = NULL; 3218 args->bc_xprt->xpt_bc_xps = NULL; 3219 xprt_put(xprt); 3220 ret = ERR_PTR(-EINVAL); 3221 out_err: 3222 xs_xprt_free(xprt); 3223 return ret; 3224 } 3225 3226 static struct xprt_class xs_local_transport = { 3227 .list = LIST_HEAD_INIT(xs_local_transport.list), 3228 .name = "named UNIX socket", 3229 .owner = THIS_MODULE, 3230 .ident = XPRT_TRANSPORT_LOCAL, 3231 .setup = xs_setup_local, 3232 }; 3233 3234 static struct xprt_class xs_udp_transport = { 3235 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3236 .name = "udp", 3237 .owner = THIS_MODULE, 3238 .ident = XPRT_TRANSPORT_UDP, 3239 .setup = xs_setup_udp, 3240 }; 3241 3242 static struct xprt_class xs_tcp_transport = { 3243 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3244 .name = "tcp", 3245 .owner = THIS_MODULE, 3246 .ident = XPRT_TRANSPORT_TCP, 3247 .setup = xs_setup_tcp, 3248 }; 3249 3250 static struct xprt_class xs_bc_tcp_transport = { 3251 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3252 .name = "tcp NFSv4.1 backchannel", 3253 .owner = THIS_MODULE, 3254 .ident = XPRT_TRANSPORT_BC_TCP, 3255 .setup = xs_setup_bc_tcp, 3256 }; 3257 3258 /** 3259 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3260 * 3261 */ 3262 int init_socket_xprt(void) 3263 { 3264 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 3265 if (!sunrpc_table_header) 3266 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3267 #endif 3268 3269 xprt_register_transport(&xs_local_transport); 3270 xprt_register_transport(&xs_udp_transport); 3271 xprt_register_transport(&xs_tcp_transport); 3272 xprt_register_transport(&xs_bc_tcp_transport); 3273 3274 return 0; 3275 } 3276 3277 /** 3278 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3279 * 3280 */ 3281 void cleanup_socket_xprt(void) 3282 { 3283 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 3284 if (sunrpc_table_header) { 3285 unregister_sysctl_table(sunrpc_table_header); 3286 sunrpc_table_header = NULL; 3287 } 3288 #endif 3289 3290 xprt_unregister_transport(&xs_local_transport); 3291 xprt_unregister_transport(&xs_udp_transport); 3292 xprt_unregister_transport(&xs_tcp_transport); 3293 xprt_unregister_transport(&xs_bc_tcp_transport); 3294 } 3295 3296 static int param_set_uint_minmax(const char *val, 3297 const struct kernel_param *kp, 3298 unsigned int min, unsigned int max) 3299 { 3300 unsigned int num; 3301 int ret; 3302 3303 if (!val) 3304 return -EINVAL; 3305 ret = kstrtouint(val, 0, &num); 3306 if (ret) 3307 return ret; 3308 if (num < min || num > max) 3309 return -EINVAL; 3310 *((unsigned int *)kp->arg) = num; 3311 return 0; 3312 } 3313 3314 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3315 { 3316 if (kp->arg == &xprt_min_resvport) 3317 return param_set_uint_minmax(val, kp, 3318 RPC_MIN_RESVPORT, 3319 xprt_max_resvport); 3320 return param_set_uint_minmax(val, kp, 3321 xprt_min_resvport, 3322 RPC_MAX_RESVPORT); 3323 } 3324 3325 static const struct kernel_param_ops param_ops_portnr = { 3326 .set = param_set_portnr, 3327 .get = param_get_uint, 3328 }; 3329 3330 #define param_check_portnr(name, p) \ 3331 __param_check(name, p, unsigned int); 3332 3333 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3334 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3335 3336 static int param_set_slot_table_size(const char *val, 3337 const struct kernel_param *kp) 3338 { 3339 return param_set_uint_minmax(val, kp, 3340 RPC_MIN_SLOT_TABLE, 3341 RPC_MAX_SLOT_TABLE); 3342 } 3343 3344 static const struct kernel_param_ops param_ops_slot_table_size = { 3345 .set = param_set_slot_table_size, 3346 .get = param_get_uint, 3347 }; 3348 3349 #define param_check_slot_table_size(name, p) \ 3350 __param_check(name, p, unsigned int); 3351 3352 static int param_set_max_slot_table_size(const char *val, 3353 const struct kernel_param *kp) 3354 { 3355 return param_set_uint_minmax(val, kp, 3356 RPC_MIN_SLOT_TABLE, 3357 RPC_MAX_SLOT_TABLE_LIMIT); 3358 } 3359 3360 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3361 .set = param_set_max_slot_table_size, 3362 .get = param_get_uint, 3363 }; 3364 3365 #define param_check_max_slot_table_size(name, p) \ 3366 __param_check(name, p, unsigned int); 3367 3368 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3369 slot_table_size, 0644); 3370 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3371 max_slot_table_size, 0644); 3372 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3373 slot_table_size, 0644); 3374 3375