1 /* 2 * linux/net/sunrpc/xprtsock.c 3 * 4 * Client-side transport implementation for sockets. 5 * 6 * TCP callback races fixes (C) 1998 Red Hat 7 * TCP send fixes (C) 1998 Red Hat 8 * TCP NFS related read + write fixes 9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 10 * 11 * Rewrite of larges part of the code in order to stabilize TCP stuff. 12 * Fix behaviour when socket buffer is full. 13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> 14 * 15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> 16 * 17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005. 18 * <gilles.quillard@bull.net> 19 */ 20 21 #include <linux/types.h> 22 #include <linux/string.h> 23 #include <linux/slab.h> 24 #include <linux/module.h> 25 #include <linux/capability.h> 26 #include <linux/pagemap.h> 27 #include <linux/errno.h> 28 #include <linux/socket.h> 29 #include <linux/in.h> 30 #include <linux/net.h> 31 #include <linux/mm.h> 32 #include <linux/un.h> 33 #include <linux/udp.h> 34 #include <linux/tcp.h> 35 #include <linux/sunrpc/clnt.h> 36 #include <linux/sunrpc/addr.h> 37 #include <linux/sunrpc/sched.h> 38 #include <linux/sunrpc/svcsock.h> 39 #include <linux/sunrpc/xprtsock.h> 40 #include <linux/file.h> 41 #ifdef CONFIG_SUNRPC_BACKCHANNEL 42 #include <linux/sunrpc/bc_xprt.h> 43 #endif 44 45 #include <net/sock.h> 46 #include <net/checksum.h> 47 #include <net/udp.h> 48 #include <net/tcp.h> 49 50 #include <trace/events/sunrpc.h> 51 52 #include "sunrpc.h" 53 54 static void xs_close(struct rpc_xprt *xprt); 55 56 /* 57 * xprtsock tunables 58 */ 59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE; 61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE; 62 63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 65 66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 67 68 #define XS_TCP_LINGER_TO (15U * HZ) 69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO; 70 71 /* 72 * We can register our own files under /proc/sys/sunrpc by 73 * calling register_sysctl_table() again. The files in that 74 * directory become the union of all files registered there. 75 * 76 * We simply need to make sure that we don't collide with 77 * someone else's file names! 78 */ 79 80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT; 83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 85 86 static struct ctl_table_header *sunrpc_table_header; 87 88 /* 89 * FIXME: changing the UDP slot table size should also resize the UDP 90 * socket buffers for existing UDP transports 91 */ 92 static struct ctl_table xs_tunables_table[] = { 93 { 94 .procname = "udp_slot_table_entries", 95 .data = &xprt_udp_slot_table_entries, 96 .maxlen = sizeof(unsigned int), 97 .mode = 0644, 98 .proc_handler = proc_dointvec_minmax, 99 .extra1 = &min_slot_table_size, 100 .extra2 = &max_slot_table_size 101 }, 102 { 103 .procname = "tcp_slot_table_entries", 104 .data = &xprt_tcp_slot_table_entries, 105 .maxlen = sizeof(unsigned int), 106 .mode = 0644, 107 .proc_handler = proc_dointvec_minmax, 108 .extra1 = &min_slot_table_size, 109 .extra2 = &max_slot_table_size 110 }, 111 { 112 .procname = "tcp_max_slot_table_entries", 113 .data = &xprt_max_tcp_slot_table_entries, 114 .maxlen = sizeof(unsigned int), 115 .mode = 0644, 116 .proc_handler = proc_dointvec_minmax, 117 .extra1 = &min_slot_table_size, 118 .extra2 = &max_tcp_slot_table_limit 119 }, 120 { 121 .procname = "min_resvport", 122 .data = &xprt_min_resvport, 123 .maxlen = sizeof(unsigned int), 124 .mode = 0644, 125 .proc_handler = proc_dointvec_minmax, 126 .extra1 = &xprt_min_resvport_limit, 127 .extra2 = &xprt_max_resvport_limit 128 }, 129 { 130 .procname = "max_resvport", 131 .data = &xprt_max_resvport, 132 .maxlen = sizeof(unsigned int), 133 .mode = 0644, 134 .proc_handler = proc_dointvec_minmax, 135 .extra1 = &xprt_min_resvport_limit, 136 .extra2 = &xprt_max_resvport_limit 137 }, 138 { 139 .procname = "tcp_fin_timeout", 140 .data = &xs_tcp_fin_timeout, 141 .maxlen = sizeof(xs_tcp_fin_timeout), 142 .mode = 0644, 143 .proc_handler = proc_dointvec_jiffies, 144 }, 145 { }, 146 }; 147 148 static struct ctl_table sunrpc_table[] = { 149 { 150 .procname = "sunrpc", 151 .mode = 0555, 152 .child = xs_tunables_table 153 }, 154 { }, 155 }; 156 157 #endif 158 159 /* 160 * Wait duration for a reply from the RPC portmapper. 161 */ 162 #define XS_BIND_TO (60U * HZ) 163 164 /* 165 * Delay if a UDP socket connect error occurs. This is most likely some 166 * kind of resource problem on the local host. 167 */ 168 #define XS_UDP_REEST_TO (2U * HZ) 169 170 /* 171 * The reestablish timeout allows clients to delay for a bit before attempting 172 * to reconnect to a server that just dropped our connection. 173 * 174 * We implement an exponential backoff when trying to reestablish a TCP 175 * transport connection with the server. Some servers like to drop a TCP 176 * connection when they are overworked, so we start with a short timeout and 177 * increase over time if the server is down or not responding. 178 */ 179 #define XS_TCP_INIT_REEST_TO (3U * HZ) 180 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ) 181 182 /* 183 * TCP idle timeout; client drops the transport socket if it is idle 184 * for this long. Note that we also timeout UDP sockets to prevent 185 * holding port numbers when there is no RPC traffic. 186 */ 187 #define XS_IDLE_DISC_TO (5U * 60 * HZ) 188 189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 190 # undef RPC_DEBUG_DATA 191 # define RPCDBG_FACILITY RPCDBG_TRANS 192 #endif 193 194 #ifdef RPC_DEBUG_DATA 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 196 { 197 u8 *buf = (u8 *) packet; 198 int j; 199 200 dprintk("RPC: %s\n", msg); 201 for (j = 0; j < count && j < 128; j += 4) { 202 if (!(j & 31)) { 203 if (j) 204 dprintk("\n"); 205 dprintk("0x%04x ", j); 206 } 207 dprintk("%02x%02x%02x%02x ", 208 buf[j], buf[j+1], buf[j+2], buf[j+3]); 209 } 210 dprintk("\n"); 211 } 212 #else 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 214 { 215 /* NOP */ 216 } 217 #endif 218 219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 220 { 221 return (struct rpc_xprt *) sk->sk_user_data; 222 } 223 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt) 225 { 226 return (struct sockaddr *) &xprt->addr; 227 } 228 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt) 230 { 231 return (struct sockaddr_un *) &xprt->addr; 232 } 233 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt) 235 { 236 return (struct sockaddr_in *) &xprt->addr; 237 } 238 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt) 240 { 241 return (struct sockaddr_in6 *) &xprt->addr; 242 } 243 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt) 245 { 246 struct sockaddr *sap = xs_addr(xprt); 247 struct sockaddr_in6 *sin6; 248 struct sockaddr_in *sin; 249 struct sockaddr_un *sun; 250 char buf[128]; 251 252 switch (sap->sa_family) { 253 case AF_LOCAL: 254 sun = xs_addr_un(xprt); 255 strlcpy(buf, sun->sun_path, sizeof(buf)); 256 xprt->address_strings[RPC_DISPLAY_ADDR] = 257 kstrdup(buf, GFP_KERNEL); 258 break; 259 case AF_INET: 260 (void)rpc_ntop(sap, buf, sizeof(buf)); 261 xprt->address_strings[RPC_DISPLAY_ADDR] = 262 kstrdup(buf, GFP_KERNEL); 263 sin = xs_addr_in(xprt); 264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); 265 break; 266 case AF_INET6: 267 (void)rpc_ntop(sap, buf, sizeof(buf)); 268 xprt->address_strings[RPC_DISPLAY_ADDR] = 269 kstrdup(buf, GFP_KERNEL); 270 sin6 = xs_addr_in6(xprt); 271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); 272 break; 273 default: 274 BUG(); 275 } 276 277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); 278 } 279 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt) 281 { 282 struct sockaddr *sap = xs_addr(xprt); 283 char buf[128]; 284 285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); 286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); 287 288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); 289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); 290 } 291 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt, 293 const char *protocol, 294 const char *netid) 295 { 296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol; 297 xprt->address_strings[RPC_DISPLAY_NETID] = netid; 298 xs_format_common_peer_addresses(xprt); 299 xs_format_common_peer_ports(xprt); 300 } 301 302 static void xs_update_peer_port(struct rpc_xprt *xprt) 303 { 304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); 305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 306 307 xs_format_common_peer_ports(xprt); 308 } 309 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt) 311 { 312 unsigned int i; 313 314 for (i = 0; i < RPC_DISPLAY_MAX; i++) 315 switch (i) { 316 case RPC_DISPLAY_PROTO: 317 case RPC_DISPLAY_NETID: 318 continue; 319 default: 320 kfree(xprt->address_strings[i]); 321 } 322 } 323 324 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 325 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more) 327 { 328 struct msghdr msg = { 329 .msg_name = addr, 330 .msg_namelen = addrlen, 331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0), 332 }; 333 struct kvec iov = { 334 .iov_base = vec->iov_base + base, 335 .iov_len = vec->iov_len - base, 336 }; 337 338 if (iov.iov_len != 0) 339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); 340 return kernel_sendmsg(sock, &msg, NULL, 0, 0); 341 } 342 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p) 344 { 345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page, 346 int offset, size_t size, int flags); 347 struct page **ppage; 348 unsigned int remainder; 349 int err; 350 351 remainder = xdr->page_len - base; 352 base += xdr->page_base; 353 ppage = xdr->pages + (base >> PAGE_SHIFT); 354 base &= ~PAGE_MASK; 355 do_sendpage = sock->ops->sendpage; 356 if (!zerocopy) 357 do_sendpage = sock_no_sendpage; 358 for(;;) { 359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); 360 int flags = XS_SENDMSG_FLAGS; 361 362 remainder -= len; 363 if (more) 364 flags |= MSG_MORE; 365 if (remainder != 0) 366 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE; 367 err = do_sendpage(sock, *ppage, base, len, flags); 368 if (remainder == 0 || err != len) 369 break; 370 *sent_p += err; 371 ppage++; 372 base = 0; 373 } 374 if (err > 0) { 375 *sent_p += err; 376 err = 0; 377 } 378 return err; 379 } 380 381 /** 382 * xs_sendpages - write pages directly to a socket 383 * @sock: socket to send on 384 * @addr: UDP only -- address of destination 385 * @addrlen: UDP only -- length of destination address 386 * @xdr: buffer containing this request 387 * @base: starting position in the buffer 388 * @zerocopy: true if it is safe to use sendpage() 389 * @sent_p: return the total number of bytes successfully queued for sending 390 * 391 */ 392 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p) 393 { 394 unsigned int remainder = xdr->len - base; 395 int err = 0; 396 int sent = 0; 397 398 if (unlikely(!sock)) 399 return -ENOTSOCK; 400 401 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); 402 if (base != 0) { 403 addr = NULL; 404 addrlen = 0; 405 } 406 407 if (base < xdr->head[0].iov_len || addr != NULL) { 408 unsigned int len = xdr->head[0].iov_len - base; 409 remainder -= len; 410 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0); 411 if (remainder == 0 || err != len) 412 goto out; 413 *sent_p += err; 414 base = 0; 415 } else 416 base -= xdr->head[0].iov_len; 417 418 if (base < xdr->page_len) { 419 unsigned int len = xdr->page_len - base; 420 remainder -= len; 421 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent); 422 *sent_p += sent; 423 if (remainder == 0 || sent != len) 424 goto out; 425 base = 0; 426 } else 427 base -= xdr->page_len; 428 429 if (base >= xdr->tail[0].iov_len) 430 return 0; 431 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0); 432 out: 433 if (err > 0) { 434 *sent_p += err; 435 err = 0; 436 } 437 return err; 438 } 439 440 static void xs_nospace_callback(struct rpc_task *task) 441 { 442 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt); 443 444 transport->inet->sk_write_pending--; 445 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 446 } 447 448 /** 449 * xs_nospace - place task on wait queue if transmit was incomplete 450 * @task: task to put to sleep 451 * 452 */ 453 static int xs_nospace(struct rpc_task *task) 454 { 455 struct rpc_rqst *req = task->tk_rqstp; 456 struct rpc_xprt *xprt = req->rq_xprt; 457 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 458 struct sock *sk = transport->inet; 459 int ret = -EAGAIN; 460 461 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", 462 task->tk_pid, req->rq_slen - req->rq_bytes_sent, 463 req->rq_slen); 464 465 /* Protect against races with write_space */ 466 spin_lock_bh(&xprt->transport_lock); 467 468 /* Don't race with disconnect */ 469 if (xprt_connected(xprt)) { 470 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) { 471 /* 472 * Notify TCP that we're limited by the application 473 * window size 474 */ 475 set_bit(SOCK_NOSPACE, &transport->sock->flags); 476 sk->sk_write_pending++; 477 /* ...and wait for more buffer space */ 478 xprt_wait_for_buffer_space(task, xs_nospace_callback); 479 } 480 } else { 481 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 482 ret = -ENOTCONN; 483 } 484 485 spin_unlock_bh(&xprt->transport_lock); 486 487 /* Race breaker in case memory is freed before above code is called */ 488 sk->sk_write_space(sk); 489 return ret; 490 } 491 492 /* 493 * Construct a stream transport record marker in @buf. 494 */ 495 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf) 496 { 497 u32 reclen = buf->len - sizeof(rpc_fraghdr); 498 rpc_fraghdr *base = buf->head[0].iov_base; 499 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen); 500 } 501 502 /** 503 * xs_local_send_request - write an RPC request to an AF_LOCAL socket 504 * @task: RPC task that manages the state of an RPC request 505 * 506 * Return values: 507 * 0: The request has been sent 508 * EAGAIN: The socket was blocked, please call again later to 509 * complete the request 510 * ENOTCONN: Caller needs to invoke connect logic then call again 511 * other: Some other error occured, the request was not sent 512 */ 513 static int xs_local_send_request(struct rpc_task *task) 514 { 515 struct rpc_rqst *req = task->tk_rqstp; 516 struct rpc_xprt *xprt = req->rq_xprt; 517 struct sock_xprt *transport = 518 container_of(xprt, struct sock_xprt, xprt); 519 struct xdr_buf *xdr = &req->rq_snd_buf; 520 int status; 521 int sent = 0; 522 523 xs_encode_stream_record_marker(&req->rq_snd_buf); 524 525 xs_pktdump("packet data:", 526 req->rq_svec->iov_base, req->rq_svec->iov_len); 527 528 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent, 529 true, &sent); 530 dprintk("RPC: %s(%u) = %d\n", 531 __func__, xdr->len - req->rq_bytes_sent, status); 532 533 if (status == -EAGAIN && sock_writeable(transport->inet)) 534 status = -ENOBUFS; 535 536 if (likely(sent > 0) || status == 0) { 537 req->rq_bytes_sent += sent; 538 req->rq_xmit_bytes_sent += sent; 539 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 540 req->rq_bytes_sent = 0; 541 return 0; 542 } 543 status = -EAGAIN; 544 } 545 546 switch (status) { 547 case -ENOBUFS: 548 break; 549 case -EAGAIN: 550 status = xs_nospace(task); 551 break; 552 default: 553 dprintk("RPC: sendmsg returned unrecognized error %d\n", 554 -status); 555 case -EPIPE: 556 xs_close(xprt); 557 status = -ENOTCONN; 558 } 559 560 return status; 561 } 562 563 /** 564 * xs_udp_send_request - write an RPC request to a UDP socket 565 * @task: address of RPC task that manages the state of an RPC request 566 * 567 * Return values: 568 * 0: The request has been sent 569 * EAGAIN: The socket was blocked, please call again later to 570 * complete the request 571 * ENOTCONN: Caller needs to invoke connect logic then call again 572 * other: Some other error occurred, the request was not sent 573 */ 574 static int xs_udp_send_request(struct rpc_task *task) 575 { 576 struct rpc_rqst *req = task->tk_rqstp; 577 struct rpc_xprt *xprt = req->rq_xprt; 578 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 579 struct xdr_buf *xdr = &req->rq_snd_buf; 580 int sent = 0; 581 int status; 582 583 xs_pktdump("packet data:", 584 req->rq_svec->iov_base, 585 req->rq_svec->iov_len); 586 587 if (!xprt_bound(xprt)) 588 return -ENOTCONN; 589 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen, 590 xdr, req->rq_bytes_sent, true, &sent); 591 592 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 593 xdr->len - req->rq_bytes_sent, status); 594 595 /* firewall is blocking us, don't return -EAGAIN or we end up looping */ 596 if (status == -EPERM) 597 goto process_status; 598 599 if (status == -EAGAIN && sock_writeable(transport->inet)) 600 status = -ENOBUFS; 601 602 if (sent > 0 || status == 0) { 603 req->rq_xmit_bytes_sent += sent; 604 if (sent >= req->rq_slen) 605 return 0; 606 /* Still some bytes left; set up for a retry later. */ 607 status = -EAGAIN; 608 } 609 610 process_status: 611 switch (status) { 612 case -ENOTSOCK: 613 status = -ENOTCONN; 614 /* Should we call xs_close() here? */ 615 break; 616 case -EAGAIN: 617 status = xs_nospace(task); 618 break; 619 default: 620 dprintk("RPC: sendmsg returned unrecognized error %d\n", 621 -status); 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 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 630 } 631 632 return status; 633 } 634 635 /** 636 * xs_tcp_send_request - write an RPC request to a TCP socket 637 * @task: address of RPC task that manages the state of an RPC request 638 * 639 * Return values: 640 * 0: The request has been sent 641 * EAGAIN: The socket was blocked, please call again later to 642 * complete the request 643 * ENOTCONN: Caller needs to invoke connect logic then call again 644 * other: Some other error occurred, the request was not sent 645 * 646 * XXX: In the case of soft timeouts, should we eventually give up 647 * if sendmsg is not able to make progress? 648 */ 649 static int xs_tcp_send_request(struct rpc_task *task) 650 { 651 struct rpc_rqst *req = task->tk_rqstp; 652 struct rpc_xprt *xprt = req->rq_xprt; 653 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 654 struct xdr_buf *xdr = &req->rq_snd_buf; 655 bool zerocopy = true; 656 int status; 657 int sent; 658 659 xs_encode_stream_record_marker(&req->rq_snd_buf); 660 661 xs_pktdump("packet data:", 662 req->rq_svec->iov_base, 663 req->rq_svec->iov_len); 664 /* Don't use zero copy if this is a resend. If the RPC call 665 * completes while the socket holds a reference to the pages, 666 * then we may end up resending corrupted data. 667 */ 668 if (task->tk_flags & RPC_TASK_SENT) 669 zerocopy = false; 670 671 /* Continue transmitting the packet/record. We must be careful 672 * to cope with writespace callbacks arriving _after_ we have 673 * called sendmsg(). */ 674 while (1) { 675 sent = 0; 676 status = xs_sendpages(transport->sock, NULL, 0, xdr, 677 req->rq_bytes_sent, zerocopy, &sent); 678 679 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 680 xdr->len - req->rq_bytes_sent, status); 681 682 /* If we've sent the entire packet, immediately 683 * reset the count of bytes sent. */ 684 req->rq_bytes_sent += sent; 685 req->rq_xmit_bytes_sent += sent; 686 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 687 req->rq_bytes_sent = 0; 688 return 0; 689 } 690 691 if (status < 0) 692 break; 693 if (sent == 0) { 694 status = -EAGAIN; 695 break; 696 } 697 } 698 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet)) 699 status = -ENOBUFS; 700 701 switch (status) { 702 case -ENOTSOCK: 703 status = -ENOTCONN; 704 /* Should we call xs_close() here? */ 705 break; 706 case -EAGAIN: 707 status = xs_nospace(task); 708 break; 709 default: 710 dprintk("RPC: sendmsg returned unrecognized error %d\n", 711 -status); 712 case -ECONNRESET: 713 case -ECONNREFUSED: 714 case -ENOTCONN: 715 case -EADDRINUSE: 716 case -ENOBUFS: 717 case -EPIPE: 718 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags); 719 } 720 721 return status; 722 } 723 724 /** 725 * xs_tcp_release_xprt - clean up after a tcp transmission 726 * @xprt: transport 727 * @task: rpc task 728 * 729 * This cleans up if an error causes us to abort the transmission of a request. 730 * In this case, the socket may need to be reset in order to avoid confusing 731 * the server. 732 */ 733 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 734 { 735 struct rpc_rqst *req; 736 737 if (task != xprt->snd_task) 738 return; 739 if (task == NULL) 740 goto out_release; 741 req = task->tk_rqstp; 742 if (req == NULL) 743 goto out_release; 744 if (req->rq_bytes_sent == 0) 745 goto out_release; 746 if (req->rq_bytes_sent == req->rq_snd_buf.len) 747 goto out_release; 748 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 749 out_release: 750 xprt_release_xprt(xprt, task); 751 } 752 753 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk) 754 { 755 transport->old_data_ready = sk->sk_data_ready; 756 transport->old_state_change = sk->sk_state_change; 757 transport->old_write_space = sk->sk_write_space; 758 transport->old_error_report = sk->sk_error_report; 759 } 760 761 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk) 762 { 763 sk->sk_data_ready = transport->old_data_ready; 764 sk->sk_state_change = transport->old_state_change; 765 sk->sk_write_space = transport->old_write_space; 766 sk->sk_error_report = transport->old_error_report; 767 } 768 769 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt) 770 { 771 smp_mb__before_atomic(); 772 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 773 clear_bit(XPRT_CLOSING, &xprt->state); 774 smp_mb__after_atomic(); 775 } 776 777 static void xs_sock_mark_closed(struct rpc_xprt *xprt) 778 { 779 xs_sock_reset_connection_flags(xprt); 780 /* Mark transport as closed and wake up all pending tasks */ 781 xprt_disconnect_done(xprt); 782 } 783 784 /** 785 * xs_error_report - callback to handle TCP socket state errors 786 * @sk: socket 787 * 788 * Note: we don't call sock_error() since there may be a rpc_task 789 * using the socket, and so we don't want to clear sk->sk_err. 790 */ 791 static void xs_error_report(struct sock *sk) 792 { 793 struct rpc_xprt *xprt; 794 int err; 795 796 read_lock_bh(&sk->sk_callback_lock); 797 if (!(xprt = xprt_from_sock(sk))) 798 goto out; 799 800 err = -sk->sk_err; 801 if (err == 0) 802 goto out; 803 /* Is this a reset event? */ 804 if (sk->sk_state == TCP_CLOSE) 805 xs_sock_mark_closed(xprt); 806 dprintk("RPC: xs_error_report client %p, error=%d...\n", 807 xprt, -err); 808 trace_rpc_socket_error(xprt, sk->sk_socket, err); 809 xprt_wake_pending_tasks(xprt, err); 810 out: 811 read_unlock_bh(&sk->sk_callback_lock); 812 } 813 814 static void xs_reset_transport(struct sock_xprt *transport) 815 { 816 struct socket *sock = transport->sock; 817 struct sock *sk = transport->inet; 818 struct rpc_xprt *xprt = &transport->xprt; 819 820 if (sk == NULL) 821 return; 822 823 if (atomic_read(&transport->xprt.swapper)) 824 sk_clear_memalloc(sk); 825 826 kernel_sock_shutdown(sock, SHUT_RDWR); 827 828 mutex_lock(&transport->recv_mutex); 829 write_lock_bh(&sk->sk_callback_lock); 830 transport->inet = NULL; 831 transport->sock = NULL; 832 833 sk->sk_user_data = NULL; 834 835 xs_restore_old_callbacks(transport, sk); 836 xprt_clear_connected(xprt); 837 write_unlock_bh(&sk->sk_callback_lock); 838 xs_sock_reset_connection_flags(xprt); 839 mutex_unlock(&transport->recv_mutex); 840 841 trace_rpc_socket_close(xprt, sock); 842 sock_release(sock); 843 } 844 845 /** 846 * xs_close - close a socket 847 * @xprt: transport 848 * 849 * This is used when all requests are complete; ie, no DRC state remains 850 * on the server we want to save. 851 * 852 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with 853 * xs_reset_transport() zeroing the socket from underneath a writer. 854 */ 855 static void xs_close(struct rpc_xprt *xprt) 856 { 857 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 858 859 dprintk("RPC: xs_close xprt %p\n", xprt); 860 861 xs_reset_transport(transport); 862 xprt->reestablish_timeout = 0; 863 864 xprt_disconnect_done(xprt); 865 } 866 867 static void xs_inject_disconnect(struct rpc_xprt *xprt) 868 { 869 dprintk("RPC: injecting transport disconnect on xprt=%p\n", 870 xprt); 871 xprt_disconnect_done(xprt); 872 } 873 874 static void xs_xprt_free(struct rpc_xprt *xprt) 875 { 876 xs_free_peer_addresses(xprt); 877 xprt_free(xprt); 878 } 879 880 /** 881 * xs_destroy - prepare to shutdown a transport 882 * @xprt: doomed transport 883 * 884 */ 885 static void xs_destroy(struct rpc_xprt *xprt) 886 { 887 struct sock_xprt *transport = container_of(xprt, 888 struct sock_xprt, xprt); 889 dprintk("RPC: xs_destroy xprt %p\n", xprt); 890 891 cancel_delayed_work_sync(&transport->connect_worker); 892 xs_close(xprt); 893 cancel_work_sync(&transport->recv_worker); 894 xs_xprt_free(xprt); 895 module_put(THIS_MODULE); 896 } 897 898 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb) 899 { 900 struct xdr_skb_reader desc = { 901 .skb = skb, 902 .offset = sizeof(rpc_fraghdr), 903 .count = skb->len - sizeof(rpc_fraghdr), 904 }; 905 906 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0) 907 return -1; 908 if (desc.count) 909 return -1; 910 return 0; 911 } 912 913 /** 914 * xs_local_data_read_skb 915 * @xprt: transport 916 * @sk: socket 917 * @skb: skbuff 918 * 919 * Currently this assumes we can read the whole reply in a single gulp. 920 */ 921 static void xs_local_data_read_skb(struct rpc_xprt *xprt, 922 struct sock *sk, 923 struct sk_buff *skb) 924 { 925 struct rpc_task *task; 926 struct rpc_rqst *rovr; 927 int repsize, copied; 928 u32 _xid; 929 __be32 *xp; 930 931 repsize = skb->len - sizeof(rpc_fraghdr); 932 if (repsize < 4) { 933 dprintk("RPC: impossible RPC reply size %d\n", repsize); 934 return; 935 } 936 937 /* Copy the XID from the skb... */ 938 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid); 939 if (xp == NULL) 940 return; 941 942 /* Look up and lock the request corresponding to the given XID */ 943 spin_lock_bh(&xprt->transport_lock); 944 rovr = xprt_lookup_rqst(xprt, *xp); 945 if (!rovr) 946 goto out_unlock; 947 task = rovr->rq_task; 948 949 copied = rovr->rq_private_buf.buflen; 950 if (copied > repsize) 951 copied = repsize; 952 953 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) { 954 dprintk("RPC: sk_buff copy failed\n"); 955 goto out_unlock; 956 } 957 958 xprt_complete_rqst(task, copied); 959 960 out_unlock: 961 spin_unlock_bh(&xprt->transport_lock); 962 } 963 964 static void xs_local_data_receive(struct sock_xprt *transport) 965 { 966 struct sk_buff *skb; 967 struct sock *sk; 968 int err; 969 970 mutex_lock(&transport->recv_mutex); 971 sk = transport->inet; 972 if (sk == NULL) 973 goto out; 974 for (;;) { 975 skb = skb_recv_datagram(sk, 0, 1, &err); 976 if (skb == NULL) 977 break; 978 xs_local_data_read_skb(&transport->xprt, sk, skb); 979 skb_free_datagram(sk, skb); 980 } 981 out: 982 mutex_unlock(&transport->recv_mutex); 983 } 984 985 static void xs_local_data_receive_workfn(struct work_struct *work) 986 { 987 struct sock_xprt *transport = 988 container_of(work, struct sock_xprt, recv_worker); 989 xs_local_data_receive(transport); 990 } 991 992 /** 993 * xs_udp_data_read_skb - receive callback for UDP sockets 994 * @xprt: transport 995 * @sk: socket 996 * @skb: skbuff 997 * 998 */ 999 static void xs_udp_data_read_skb(struct rpc_xprt *xprt, 1000 struct sock *sk, 1001 struct sk_buff *skb) 1002 { 1003 struct rpc_task *task; 1004 struct rpc_rqst *rovr; 1005 int repsize, copied; 1006 u32 _xid; 1007 __be32 *xp; 1008 1009 repsize = skb->len - sizeof(struct udphdr); 1010 if (repsize < 4) { 1011 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 1012 return; 1013 } 1014 1015 /* Copy the XID from the skb... */ 1016 xp = skb_header_pointer(skb, sizeof(struct udphdr), 1017 sizeof(_xid), &_xid); 1018 if (xp == NULL) 1019 return; 1020 1021 /* Look up and lock the request corresponding to the given XID */ 1022 spin_lock_bh(&xprt->transport_lock); 1023 rovr = xprt_lookup_rqst(xprt, *xp); 1024 if (!rovr) 1025 goto out_unlock; 1026 task = rovr->rq_task; 1027 1028 if ((copied = rovr->rq_private_buf.buflen) > repsize) 1029 copied = repsize; 1030 1031 /* Suck it into the iovec, verify checksum if not done by hw. */ 1032 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) { 1033 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS); 1034 goto out_unlock; 1035 } 1036 1037 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS); 1038 1039 xprt_adjust_cwnd(xprt, task, copied); 1040 xprt_complete_rqst(task, copied); 1041 1042 out_unlock: 1043 spin_unlock_bh(&xprt->transport_lock); 1044 } 1045 1046 static void xs_udp_data_receive(struct sock_xprt *transport) 1047 { 1048 struct sk_buff *skb; 1049 struct sock *sk; 1050 int err; 1051 1052 mutex_lock(&transport->recv_mutex); 1053 sk = transport->inet; 1054 if (sk == NULL) 1055 goto out; 1056 for (;;) { 1057 skb = skb_recv_datagram(sk, 0, 1, &err); 1058 if (skb == NULL) 1059 break; 1060 xs_udp_data_read_skb(&transport->xprt, sk, skb); 1061 skb_free_datagram(sk, skb); 1062 } 1063 out: 1064 mutex_unlock(&transport->recv_mutex); 1065 } 1066 1067 static void xs_udp_data_receive_workfn(struct work_struct *work) 1068 { 1069 struct sock_xprt *transport = 1070 container_of(work, struct sock_xprt, recv_worker); 1071 xs_udp_data_receive(transport); 1072 } 1073 1074 /** 1075 * xs_data_ready - "data ready" callback for UDP sockets 1076 * @sk: socket with data to read 1077 * 1078 */ 1079 static void xs_data_ready(struct sock *sk) 1080 { 1081 struct rpc_xprt *xprt; 1082 1083 read_lock_bh(&sk->sk_callback_lock); 1084 dprintk("RPC: xs_data_ready...\n"); 1085 xprt = xprt_from_sock(sk); 1086 if (xprt != NULL) { 1087 struct sock_xprt *transport = container_of(xprt, 1088 struct sock_xprt, xprt); 1089 queue_work(rpciod_workqueue, &transport->recv_worker); 1090 } 1091 read_unlock_bh(&sk->sk_callback_lock); 1092 } 1093 1094 /* 1095 * Helper function to force a TCP close if the server is sending 1096 * junk and/or it has put us in CLOSE_WAIT 1097 */ 1098 static void xs_tcp_force_close(struct rpc_xprt *xprt) 1099 { 1100 xprt_force_disconnect(xprt); 1101 } 1102 1103 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 1104 { 1105 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1106 size_t len, used; 1107 char *p; 1108 1109 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; 1110 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; 1111 used = xdr_skb_read_bits(desc, p, len); 1112 transport->tcp_offset += used; 1113 if (used != len) 1114 return; 1115 1116 transport->tcp_reclen = ntohl(transport->tcp_fraghdr); 1117 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) 1118 transport->tcp_flags |= TCP_RCV_LAST_FRAG; 1119 else 1120 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; 1121 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; 1122 1123 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; 1124 transport->tcp_offset = 0; 1125 1126 /* Sanity check of the record length */ 1127 if (unlikely(transport->tcp_reclen < 8)) { 1128 dprintk("RPC: invalid TCP record fragment length\n"); 1129 xs_tcp_force_close(xprt); 1130 return; 1131 } 1132 dprintk("RPC: reading TCP record fragment of length %d\n", 1133 transport->tcp_reclen); 1134 } 1135 1136 static void xs_tcp_check_fraghdr(struct sock_xprt *transport) 1137 { 1138 if (transport->tcp_offset == transport->tcp_reclen) { 1139 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; 1140 transport->tcp_offset = 0; 1141 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { 1142 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1143 transport->tcp_flags |= TCP_RCV_COPY_XID; 1144 transport->tcp_copied = 0; 1145 } 1146 } 1147 } 1148 1149 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1150 { 1151 size_t len, used; 1152 char *p; 1153 1154 len = sizeof(transport->tcp_xid) - transport->tcp_offset; 1155 dprintk("RPC: reading XID (%Zu bytes)\n", len); 1156 p = ((char *) &transport->tcp_xid) + transport->tcp_offset; 1157 used = xdr_skb_read_bits(desc, p, len); 1158 transport->tcp_offset += used; 1159 if (used != len) 1160 return; 1161 transport->tcp_flags &= ~TCP_RCV_COPY_XID; 1162 transport->tcp_flags |= TCP_RCV_READ_CALLDIR; 1163 transport->tcp_copied = 4; 1164 dprintk("RPC: reading %s XID %08x\n", 1165 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for" 1166 : "request with", 1167 ntohl(transport->tcp_xid)); 1168 xs_tcp_check_fraghdr(transport); 1169 } 1170 1171 static inline void xs_tcp_read_calldir(struct sock_xprt *transport, 1172 struct xdr_skb_reader *desc) 1173 { 1174 size_t len, used; 1175 u32 offset; 1176 char *p; 1177 1178 /* 1179 * We want transport->tcp_offset to be 8 at the end of this routine 1180 * (4 bytes for the xid and 4 bytes for the call/reply flag). 1181 * When this function is called for the first time, 1182 * transport->tcp_offset is 4 (after having already read the xid). 1183 */ 1184 offset = transport->tcp_offset - sizeof(transport->tcp_xid); 1185 len = sizeof(transport->tcp_calldir) - offset; 1186 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len); 1187 p = ((char *) &transport->tcp_calldir) + offset; 1188 used = xdr_skb_read_bits(desc, p, len); 1189 transport->tcp_offset += used; 1190 if (used != len) 1191 return; 1192 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR; 1193 /* 1194 * We don't yet have the XDR buffer, so we will write the calldir 1195 * out after we get the buffer from the 'struct rpc_rqst' 1196 */ 1197 switch (ntohl(transport->tcp_calldir)) { 1198 case RPC_REPLY: 1199 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1200 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1201 transport->tcp_flags |= TCP_RPC_REPLY; 1202 break; 1203 case RPC_CALL: 1204 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR; 1205 transport->tcp_flags |= TCP_RCV_COPY_DATA; 1206 transport->tcp_flags &= ~TCP_RPC_REPLY; 1207 break; 1208 default: 1209 dprintk("RPC: invalid request message type\n"); 1210 xs_tcp_force_close(&transport->xprt); 1211 } 1212 xs_tcp_check_fraghdr(transport); 1213 } 1214 1215 static inline void xs_tcp_read_common(struct rpc_xprt *xprt, 1216 struct xdr_skb_reader *desc, 1217 struct rpc_rqst *req) 1218 { 1219 struct sock_xprt *transport = 1220 container_of(xprt, struct sock_xprt, xprt); 1221 struct xdr_buf *rcvbuf; 1222 size_t len; 1223 ssize_t r; 1224 1225 rcvbuf = &req->rq_private_buf; 1226 1227 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) { 1228 /* 1229 * Save the RPC direction in the XDR buffer 1230 */ 1231 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied, 1232 &transport->tcp_calldir, 1233 sizeof(transport->tcp_calldir)); 1234 transport->tcp_copied += sizeof(transport->tcp_calldir); 1235 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR; 1236 } 1237 1238 len = desc->count; 1239 if (len > transport->tcp_reclen - transport->tcp_offset) { 1240 struct xdr_skb_reader my_desc; 1241 1242 len = transport->tcp_reclen - transport->tcp_offset; 1243 memcpy(&my_desc, desc, sizeof(my_desc)); 1244 my_desc.count = len; 1245 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1246 &my_desc, xdr_skb_read_bits); 1247 desc->count -= r; 1248 desc->offset += r; 1249 } else 1250 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 1251 desc, xdr_skb_read_bits); 1252 1253 if (r > 0) { 1254 transport->tcp_copied += r; 1255 transport->tcp_offset += r; 1256 } 1257 if (r != len) { 1258 /* Error when copying to the receive buffer, 1259 * usually because we weren't able to allocate 1260 * additional buffer pages. All we can do now 1261 * is turn off TCP_RCV_COPY_DATA, so the request 1262 * will not receive any additional updates, 1263 * and time out. 1264 * Any remaining data from this record will 1265 * be discarded. 1266 */ 1267 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1268 dprintk("RPC: XID %08x truncated request\n", 1269 ntohl(transport->tcp_xid)); 1270 dprintk("RPC: xprt = %p, tcp_copied = %lu, " 1271 "tcp_offset = %u, tcp_reclen = %u\n", 1272 xprt, transport->tcp_copied, 1273 transport->tcp_offset, transport->tcp_reclen); 1274 return; 1275 } 1276 1277 dprintk("RPC: XID %08x read %Zd bytes\n", 1278 ntohl(transport->tcp_xid), r); 1279 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " 1280 "tcp_reclen = %u\n", xprt, transport->tcp_copied, 1281 transport->tcp_offset, transport->tcp_reclen); 1282 1283 if (transport->tcp_copied == req->rq_private_buf.buflen) 1284 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1285 else if (transport->tcp_offset == transport->tcp_reclen) { 1286 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) 1287 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1288 } 1289 } 1290 1291 /* 1292 * Finds the request corresponding to the RPC xid and invokes the common 1293 * tcp read code to read the data. 1294 */ 1295 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt, 1296 struct xdr_skb_reader *desc) 1297 { 1298 struct sock_xprt *transport = 1299 container_of(xprt, struct sock_xprt, xprt); 1300 struct rpc_rqst *req; 1301 1302 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid)); 1303 1304 /* Find and lock the request corresponding to this xid */ 1305 spin_lock_bh(&xprt->transport_lock); 1306 req = xprt_lookup_rqst(xprt, transport->tcp_xid); 1307 if (!req) { 1308 dprintk("RPC: XID %08x request not found!\n", 1309 ntohl(transport->tcp_xid)); 1310 spin_unlock_bh(&xprt->transport_lock); 1311 return -1; 1312 } 1313 1314 xs_tcp_read_common(xprt, desc, req); 1315 1316 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1317 xprt_complete_rqst(req->rq_task, transport->tcp_copied); 1318 1319 spin_unlock_bh(&xprt->transport_lock); 1320 return 0; 1321 } 1322 1323 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1324 /* 1325 * Obtains an rpc_rqst previously allocated and invokes the common 1326 * tcp read code to read the data. The result is placed in the callback 1327 * queue. 1328 * If we're unable to obtain the rpc_rqst we schedule the closing of the 1329 * connection and return -1. 1330 */ 1331 static int xs_tcp_read_callback(struct rpc_xprt *xprt, 1332 struct xdr_skb_reader *desc) 1333 { 1334 struct sock_xprt *transport = 1335 container_of(xprt, struct sock_xprt, xprt); 1336 struct rpc_rqst *req; 1337 1338 /* Look up and lock the request corresponding to the given XID */ 1339 spin_lock_bh(&xprt->transport_lock); 1340 req = xprt_lookup_bc_request(xprt, transport->tcp_xid); 1341 if (req == NULL) { 1342 spin_unlock_bh(&xprt->transport_lock); 1343 printk(KERN_WARNING "Callback slot table overflowed\n"); 1344 xprt_force_disconnect(xprt); 1345 return -1; 1346 } 1347 1348 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid)); 1349 xs_tcp_read_common(xprt, desc, req); 1350 1351 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 1352 xprt_complete_bc_request(req, transport->tcp_copied); 1353 spin_unlock_bh(&xprt->transport_lock); 1354 1355 return 0; 1356 } 1357 1358 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1359 struct xdr_skb_reader *desc) 1360 { 1361 struct sock_xprt *transport = 1362 container_of(xprt, struct sock_xprt, xprt); 1363 1364 return (transport->tcp_flags & TCP_RPC_REPLY) ? 1365 xs_tcp_read_reply(xprt, desc) : 1366 xs_tcp_read_callback(xprt, desc); 1367 } 1368 1369 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net) 1370 { 1371 int ret; 1372 1373 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0, 1374 SVC_SOCK_ANONYMOUS); 1375 if (ret < 0) 1376 return ret; 1377 return 0; 1378 } 1379 #else 1380 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt, 1381 struct xdr_skb_reader *desc) 1382 { 1383 return xs_tcp_read_reply(xprt, desc); 1384 } 1385 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1386 1387 /* 1388 * Read data off the transport. This can be either an RPC_CALL or an 1389 * RPC_REPLY. Relay the processing to helper functions. 1390 */ 1391 static void xs_tcp_read_data(struct rpc_xprt *xprt, 1392 struct xdr_skb_reader *desc) 1393 { 1394 struct sock_xprt *transport = 1395 container_of(xprt, struct sock_xprt, xprt); 1396 1397 if (_xs_tcp_read_data(xprt, desc) == 0) 1398 xs_tcp_check_fraghdr(transport); 1399 else { 1400 /* 1401 * The transport_lock protects the request handling. 1402 * There's no need to hold it to update the tcp_flags. 1403 */ 1404 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 1405 } 1406 } 1407 1408 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) 1409 { 1410 size_t len; 1411 1412 len = transport->tcp_reclen - transport->tcp_offset; 1413 if (len > desc->count) 1414 len = desc->count; 1415 desc->count -= len; 1416 desc->offset += len; 1417 transport->tcp_offset += len; 1418 dprintk("RPC: discarded %Zu bytes\n", len); 1419 xs_tcp_check_fraghdr(transport); 1420 } 1421 1422 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) 1423 { 1424 struct rpc_xprt *xprt = rd_desc->arg.data; 1425 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1426 struct xdr_skb_reader desc = { 1427 .skb = skb, 1428 .offset = offset, 1429 .count = len, 1430 }; 1431 1432 dprintk("RPC: xs_tcp_data_recv started\n"); 1433 do { 1434 trace_xs_tcp_data_recv(transport); 1435 /* Read in a new fragment marker if necessary */ 1436 /* Can we ever really expect to get completely empty fragments? */ 1437 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { 1438 xs_tcp_read_fraghdr(xprt, &desc); 1439 continue; 1440 } 1441 /* Read in the xid if necessary */ 1442 if (transport->tcp_flags & TCP_RCV_COPY_XID) { 1443 xs_tcp_read_xid(transport, &desc); 1444 continue; 1445 } 1446 /* Read in the call/reply flag */ 1447 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) { 1448 xs_tcp_read_calldir(transport, &desc); 1449 continue; 1450 } 1451 /* Read in the request data */ 1452 if (transport->tcp_flags & TCP_RCV_COPY_DATA) { 1453 xs_tcp_read_data(xprt, &desc); 1454 continue; 1455 } 1456 /* Skip over any trailing bytes on short reads */ 1457 xs_tcp_read_discard(transport, &desc); 1458 } while (desc.count); 1459 trace_xs_tcp_data_recv(transport); 1460 dprintk("RPC: xs_tcp_data_recv done\n"); 1461 return len - desc.count; 1462 } 1463 1464 static void xs_tcp_data_receive(struct sock_xprt *transport) 1465 { 1466 struct rpc_xprt *xprt = &transport->xprt; 1467 struct sock *sk; 1468 read_descriptor_t rd_desc = { 1469 .count = 2*1024*1024, 1470 .arg.data = xprt, 1471 }; 1472 unsigned long total = 0; 1473 int read = 0; 1474 1475 mutex_lock(&transport->recv_mutex); 1476 sk = transport->inet; 1477 if (sk == NULL) 1478 goto out; 1479 1480 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ 1481 for (;;) { 1482 lock_sock(sk); 1483 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); 1484 release_sock(sk); 1485 if (read <= 0) 1486 break; 1487 total += read; 1488 rd_desc.count = 65536; 1489 } 1490 out: 1491 mutex_unlock(&transport->recv_mutex); 1492 trace_xs_tcp_data_ready(xprt, read, total); 1493 } 1494 1495 static void xs_tcp_data_receive_workfn(struct work_struct *work) 1496 { 1497 struct sock_xprt *transport = 1498 container_of(work, struct sock_xprt, recv_worker); 1499 xs_tcp_data_receive(transport); 1500 } 1501 1502 /** 1503 * xs_tcp_data_ready - "data ready" callback for TCP sockets 1504 * @sk: socket with data to read 1505 * 1506 */ 1507 static void xs_tcp_data_ready(struct sock *sk) 1508 { 1509 struct sock_xprt *transport; 1510 struct rpc_xprt *xprt; 1511 1512 dprintk("RPC: xs_tcp_data_ready...\n"); 1513 1514 read_lock_bh(&sk->sk_callback_lock); 1515 if (!(xprt = xprt_from_sock(sk))) 1516 goto out; 1517 transport = container_of(xprt, struct sock_xprt, xprt); 1518 1519 /* Any data means we had a useful conversation, so 1520 * the we don't need to delay the next reconnect 1521 */ 1522 if (xprt->reestablish_timeout) 1523 xprt->reestablish_timeout = 0; 1524 queue_work(rpciod_workqueue, &transport->recv_worker); 1525 1526 out: 1527 read_unlock_bh(&sk->sk_callback_lock); 1528 } 1529 1530 /** 1531 * xs_tcp_state_change - callback to handle TCP socket state changes 1532 * @sk: socket whose state has changed 1533 * 1534 */ 1535 static void xs_tcp_state_change(struct sock *sk) 1536 { 1537 struct rpc_xprt *xprt; 1538 struct sock_xprt *transport; 1539 1540 read_lock_bh(&sk->sk_callback_lock); 1541 if (!(xprt = xprt_from_sock(sk))) 1542 goto out; 1543 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 1544 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n", 1545 sk->sk_state, xprt_connected(xprt), 1546 sock_flag(sk, SOCK_DEAD), 1547 sock_flag(sk, SOCK_ZAPPED), 1548 sk->sk_shutdown); 1549 1550 transport = container_of(xprt, struct sock_xprt, xprt); 1551 trace_rpc_socket_state_change(xprt, sk->sk_socket); 1552 switch (sk->sk_state) { 1553 case TCP_ESTABLISHED: 1554 spin_lock(&xprt->transport_lock); 1555 if (!xprt_test_and_set_connected(xprt)) { 1556 1557 /* Reset TCP record info */ 1558 transport->tcp_offset = 0; 1559 transport->tcp_reclen = 0; 1560 transport->tcp_copied = 0; 1561 transport->tcp_flags = 1562 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; 1563 xprt->connect_cookie++; 1564 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 1565 xprt_clear_connecting(xprt); 1566 1567 xprt_wake_pending_tasks(xprt, -EAGAIN); 1568 } 1569 spin_unlock(&xprt->transport_lock); 1570 break; 1571 case TCP_FIN_WAIT1: 1572 /* The client initiated a shutdown of the socket */ 1573 xprt->connect_cookie++; 1574 xprt->reestablish_timeout = 0; 1575 set_bit(XPRT_CLOSING, &xprt->state); 1576 smp_mb__before_atomic(); 1577 clear_bit(XPRT_CONNECTED, &xprt->state); 1578 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 1579 smp_mb__after_atomic(); 1580 break; 1581 case TCP_CLOSE_WAIT: 1582 /* The server initiated a shutdown of the socket */ 1583 xprt->connect_cookie++; 1584 clear_bit(XPRT_CONNECTED, &xprt->state); 1585 xs_tcp_force_close(xprt); 1586 case TCP_CLOSING: 1587 /* 1588 * If the server closed down the connection, make sure that 1589 * we back off before reconnecting 1590 */ 1591 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 1592 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1593 break; 1594 case TCP_LAST_ACK: 1595 set_bit(XPRT_CLOSING, &xprt->state); 1596 smp_mb__before_atomic(); 1597 clear_bit(XPRT_CONNECTED, &xprt->state); 1598 smp_mb__after_atomic(); 1599 break; 1600 case TCP_CLOSE: 1601 if (test_and_clear_bit(XPRT_SOCK_CONNECTING, 1602 &transport->sock_state)) 1603 xprt_clear_connecting(xprt); 1604 xs_sock_mark_closed(xprt); 1605 } 1606 out: 1607 read_unlock_bh(&sk->sk_callback_lock); 1608 } 1609 1610 static void xs_write_space(struct sock *sk) 1611 { 1612 struct socket *sock; 1613 struct rpc_xprt *xprt; 1614 1615 if (unlikely(!(sock = sk->sk_socket))) 1616 return; 1617 clear_bit(SOCK_NOSPACE, &sock->flags); 1618 1619 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1620 return; 1621 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0) 1622 return; 1623 1624 xprt_write_space(xprt); 1625 } 1626 1627 /** 1628 * xs_udp_write_space - callback invoked when socket buffer space 1629 * becomes available 1630 * @sk: socket whose state has changed 1631 * 1632 * Called when more output buffer space is available for this socket. 1633 * We try not to wake our writers until they can make "significant" 1634 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1635 * with a bunch of small requests. 1636 */ 1637 static void xs_udp_write_space(struct sock *sk) 1638 { 1639 read_lock_bh(&sk->sk_callback_lock); 1640 1641 /* from net/core/sock.c:sock_def_write_space */ 1642 if (sock_writeable(sk)) 1643 xs_write_space(sk); 1644 1645 read_unlock_bh(&sk->sk_callback_lock); 1646 } 1647 1648 /** 1649 * xs_tcp_write_space - callback invoked when socket buffer space 1650 * becomes available 1651 * @sk: socket whose state has changed 1652 * 1653 * Called when more output buffer space is available for this socket. 1654 * We try not to wake our writers until they can make "significant" 1655 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1656 * with a bunch of small requests. 1657 */ 1658 static void xs_tcp_write_space(struct sock *sk) 1659 { 1660 read_lock_bh(&sk->sk_callback_lock); 1661 1662 /* from net/core/stream.c:sk_stream_write_space */ 1663 if (sk_stream_is_writeable(sk)) 1664 xs_write_space(sk); 1665 1666 read_unlock_bh(&sk->sk_callback_lock); 1667 } 1668 1669 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1670 { 1671 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1672 struct sock *sk = transport->inet; 1673 1674 if (transport->rcvsize) { 1675 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1676 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1677 } 1678 if (transport->sndsize) { 1679 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1680 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1681 sk->sk_write_space(sk); 1682 } 1683 } 1684 1685 /** 1686 * xs_udp_set_buffer_size - set send and receive limits 1687 * @xprt: generic transport 1688 * @sndsize: requested size of send buffer, in bytes 1689 * @rcvsize: requested size of receive buffer, in bytes 1690 * 1691 * Set socket send and receive buffer size limits. 1692 */ 1693 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1694 { 1695 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1696 1697 transport->sndsize = 0; 1698 if (sndsize) 1699 transport->sndsize = sndsize + 1024; 1700 transport->rcvsize = 0; 1701 if (rcvsize) 1702 transport->rcvsize = rcvsize + 1024; 1703 1704 xs_udp_do_set_buffer_size(xprt); 1705 } 1706 1707 /** 1708 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1709 * @task: task that timed out 1710 * 1711 * Adjust the congestion window after a retransmit timeout has occurred. 1712 */ 1713 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task) 1714 { 1715 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT); 1716 } 1717 1718 static unsigned short xs_get_random_port(void) 1719 { 1720 unsigned short range = xprt_max_resvport - xprt_min_resvport; 1721 unsigned short rand = (unsigned short) prandom_u32() % range; 1722 return rand + xprt_min_resvport; 1723 } 1724 1725 /** 1726 * xs_set_reuseaddr_port - set the socket's port and address reuse options 1727 * @sock: socket 1728 * 1729 * Note that this function has to be called on all sockets that share the 1730 * same port, and it must be called before binding. 1731 */ 1732 static void xs_sock_set_reuseport(struct socket *sock) 1733 { 1734 int opt = 1; 1735 1736 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, 1737 (char *)&opt, sizeof(opt)); 1738 } 1739 1740 static unsigned short xs_sock_getport(struct socket *sock) 1741 { 1742 struct sockaddr_storage buf; 1743 int buflen; 1744 unsigned short port = 0; 1745 1746 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0) 1747 goto out; 1748 switch (buf.ss_family) { 1749 case AF_INET6: 1750 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port); 1751 break; 1752 case AF_INET: 1753 port = ntohs(((struct sockaddr_in *)&buf)->sin_port); 1754 } 1755 out: 1756 return port; 1757 } 1758 1759 /** 1760 * xs_set_port - reset the port number in the remote endpoint address 1761 * @xprt: generic transport 1762 * @port: new port number 1763 * 1764 */ 1765 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1766 { 1767 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1768 1769 rpc_set_port(xs_addr(xprt), port); 1770 xs_update_peer_port(xprt); 1771 } 1772 1773 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock) 1774 { 1775 if (transport->srcport == 0) 1776 transport->srcport = xs_sock_getport(sock); 1777 } 1778 1779 static unsigned short xs_get_srcport(struct sock_xprt *transport) 1780 { 1781 unsigned short port = transport->srcport; 1782 1783 if (port == 0 && transport->xprt.resvport) 1784 port = xs_get_random_port(); 1785 return port; 1786 } 1787 1788 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port) 1789 { 1790 if (transport->srcport != 0) 1791 transport->srcport = 0; 1792 if (!transport->xprt.resvport) 1793 return 0; 1794 if (port <= xprt_min_resvport || port > xprt_max_resvport) 1795 return xprt_max_resvport; 1796 return --port; 1797 } 1798 static int xs_bind(struct sock_xprt *transport, struct socket *sock) 1799 { 1800 struct sockaddr_storage myaddr; 1801 int err, nloop = 0; 1802 unsigned short port = xs_get_srcport(transport); 1803 unsigned short last; 1804 1805 /* 1806 * If we are asking for any ephemeral port (i.e. port == 0 && 1807 * transport->xprt.resvport == 0), don't bind. Let the local 1808 * port selection happen implicitly when the socket is used 1809 * (for example at connect time). 1810 * 1811 * This ensures that we can continue to establish TCP 1812 * connections even when all local ephemeral ports are already 1813 * a part of some TCP connection. This makes no difference 1814 * for UDP sockets, but also doens't harm them. 1815 * 1816 * If we're asking for any reserved port (i.e. port == 0 && 1817 * transport->xprt.resvport == 1) xs_get_srcport above will 1818 * ensure that port is non-zero and we will bind as needed. 1819 */ 1820 if (port == 0) 1821 return 0; 1822 1823 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen); 1824 do { 1825 rpc_set_port((struct sockaddr *)&myaddr, port); 1826 err = kernel_bind(sock, (struct sockaddr *)&myaddr, 1827 transport->xprt.addrlen); 1828 if (err == 0) { 1829 transport->srcport = port; 1830 break; 1831 } 1832 last = port; 1833 port = xs_next_srcport(transport, port); 1834 if (port > last) 1835 nloop++; 1836 } while (err == -EADDRINUSE && nloop != 2); 1837 1838 if (myaddr.ss_family == AF_INET) 1839 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__, 1840 &((struct sockaddr_in *)&myaddr)->sin_addr, 1841 port, err ? "failed" : "ok", err); 1842 else 1843 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__, 1844 &((struct sockaddr_in6 *)&myaddr)->sin6_addr, 1845 port, err ? "failed" : "ok", err); 1846 return err; 1847 } 1848 1849 /* 1850 * We don't support autobind on AF_LOCAL sockets 1851 */ 1852 static void xs_local_rpcbind(struct rpc_task *task) 1853 { 1854 rcu_read_lock(); 1855 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt)); 1856 rcu_read_unlock(); 1857 } 1858 1859 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port) 1860 { 1861 } 1862 1863 #ifdef CONFIG_DEBUG_LOCK_ALLOC 1864 static struct lock_class_key xs_key[2]; 1865 static struct lock_class_key xs_slock_key[2]; 1866 1867 static inline void xs_reclassify_socketu(struct socket *sock) 1868 { 1869 struct sock *sk = sock->sk; 1870 1871 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC", 1872 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]); 1873 } 1874 1875 static inline void xs_reclassify_socket4(struct socket *sock) 1876 { 1877 struct sock *sk = sock->sk; 1878 1879 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC", 1880 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]); 1881 } 1882 1883 static inline void xs_reclassify_socket6(struct socket *sock) 1884 { 1885 struct sock *sk = sock->sk; 1886 1887 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC", 1888 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]); 1889 } 1890 1891 static inline void xs_reclassify_socket(int family, struct socket *sock) 1892 { 1893 WARN_ON_ONCE(sock_owned_by_user(sock->sk)); 1894 if (sock_owned_by_user(sock->sk)) 1895 return; 1896 1897 switch (family) { 1898 case AF_LOCAL: 1899 xs_reclassify_socketu(sock); 1900 break; 1901 case AF_INET: 1902 xs_reclassify_socket4(sock); 1903 break; 1904 case AF_INET6: 1905 xs_reclassify_socket6(sock); 1906 break; 1907 } 1908 } 1909 #else 1910 static inline void xs_reclassify_socketu(struct socket *sock) 1911 { 1912 } 1913 1914 static inline void xs_reclassify_socket4(struct socket *sock) 1915 { 1916 } 1917 1918 static inline void xs_reclassify_socket6(struct socket *sock) 1919 { 1920 } 1921 1922 static inline void xs_reclassify_socket(int family, struct socket *sock) 1923 { 1924 } 1925 #endif 1926 1927 static void xs_dummy_setup_socket(struct work_struct *work) 1928 { 1929 } 1930 1931 static struct socket *xs_create_sock(struct rpc_xprt *xprt, 1932 struct sock_xprt *transport, int family, int type, 1933 int protocol, bool reuseport) 1934 { 1935 struct socket *sock; 1936 int err; 1937 1938 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1); 1939 if (err < 0) { 1940 dprintk("RPC: can't create %d transport socket (%d).\n", 1941 protocol, -err); 1942 goto out; 1943 } 1944 xs_reclassify_socket(family, sock); 1945 1946 if (reuseport) 1947 xs_sock_set_reuseport(sock); 1948 1949 err = xs_bind(transport, sock); 1950 if (err) { 1951 sock_release(sock); 1952 goto out; 1953 } 1954 1955 return sock; 1956 out: 1957 return ERR_PTR(err); 1958 } 1959 1960 static int xs_local_finish_connecting(struct rpc_xprt *xprt, 1961 struct socket *sock) 1962 { 1963 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 1964 xprt); 1965 1966 if (!transport->inet) { 1967 struct sock *sk = sock->sk; 1968 1969 write_lock_bh(&sk->sk_callback_lock); 1970 1971 xs_save_old_callbacks(transport, sk); 1972 1973 sk->sk_user_data = xprt; 1974 sk->sk_data_ready = xs_data_ready; 1975 sk->sk_write_space = xs_udp_write_space; 1976 sk->sk_error_report = xs_error_report; 1977 sk->sk_allocation = GFP_NOIO; 1978 1979 xprt_clear_connected(xprt); 1980 1981 /* Reset to new socket */ 1982 transport->sock = sock; 1983 transport->inet = sk; 1984 1985 write_unlock_bh(&sk->sk_callback_lock); 1986 } 1987 1988 /* Tell the socket layer to start connecting... */ 1989 xprt->stat.connect_count++; 1990 xprt->stat.connect_start = jiffies; 1991 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0); 1992 } 1993 1994 /** 1995 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint 1996 * @transport: socket transport to connect 1997 */ 1998 static int xs_local_setup_socket(struct sock_xprt *transport) 1999 { 2000 struct rpc_xprt *xprt = &transport->xprt; 2001 struct socket *sock; 2002 int status = -EIO; 2003 2004 status = __sock_create(xprt->xprt_net, AF_LOCAL, 2005 SOCK_STREAM, 0, &sock, 1); 2006 if (status < 0) { 2007 dprintk("RPC: can't create AF_LOCAL " 2008 "transport socket (%d).\n", -status); 2009 goto out; 2010 } 2011 xs_reclassify_socketu(sock); 2012 2013 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n", 2014 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2015 2016 status = xs_local_finish_connecting(xprt, sock); 2017 trace_rpc_socket_connect(xprt, sock, status); 2018 switch (status) { 2019 case 0: 2020 dprintk("RPC: xprt %p connected to %s\n", 2021 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2022 xprt_set_connected(xprt); 2023 case -ENOBUFS: 2024 break; 2025 case -ENOENT: 2026 dprintk("RPC: xprt %p: socket %s does not exist\n", 2027 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2028 break; 2029 case -ECONNREFUSED: 2030 dprintk("RPC: xprt %p: connection refused for %s\n", 2031 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]); 2032 break; 2033 default: 2034 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n", 2035 __func__, -status, 2036 xprt->address_strings[RPC_DISPLAY_ADDR]); 2037 } 2038 2039 out: 2040 xprt_clear_connecting(xprt); 2041 xprt_wake_pending_tasks(xprt, status); 2042 return status; 2043 } 2044 2045 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2046 { 2047 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2048 int ret; 2049 2050 if (RPC_IS_ASYNC(task)) { 2051 /* 2052 * We want the AF_LOCAL connect to be resolved in the 2053 * filesystem namespace of the process making the rpc 2054 * call. Thus we connect synchronously. 2055 * 2056 * If we want to support asynchronous AF_LOCAL calls, 2057 * we'll need to figure out how to pass a namespace to 2058 * connect. 2059 */ 2060 rpc_exit(task, -ENOTCONN); 2061 return; 2062 } 2063 ret = xs_local_setup_socket(transport); 2064 if (ret && !RPC_IS_SOFTCONN(task)) 2065 msleep_interruptible(15000); 2066 } 2067 2068 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 2069 /* 2070 * Note that this should be called with XPRT_LOCKED held (or when we otherwise 2071 * know that we have exclusive access to the socket), to guard against 2072 * races with xs_reset_transport. 2073 */ 2074 static void xs_set_memalloc(struct rpc_xprt *xprt) 2075 { 2076 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, 2077 xprt); 2078 2079 /* 2080 * If there's no sock, then we have nothing to set. The 2081 * reconnecting process will get it for us. 2082 */ 2083 if (!transport->inet) 2084 return; 2085 if (atomic_read(&xprt->swapper)) 2086 sk_set_memalloc(transport->inet); 2087 } 2088 2089 /** 2090 * xs_enable_swap - Tag this transport as being used for swap. 2091 * @xprt: transport to tag 2092 * 2093 * Take a reference to this transport on behalf of the rpc_clnt, and 2094 * optionally mark it for swapping if it wasn't already. 2095 */ 2096 static int 2097 xs_enable_swap(struct rpc_xprt *xprt) 2098 { 2099 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2100 2101 if (atomic_inc_return(&xprt->swapper) != 1) 2102 return 0; 2103 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 2104 return -ERESTARTSYS; 2105 if (xs->inet) 2106 sk_set_memalloc(xs->inet); 2107 xprt_release_xprt(xprt, NULL); 2108 return 0; 2109 } 2110 2111 /** 2112 * xs_disable_swap - Untag this transport as being used for swap. 2113 * @xprt: transport to tag 2114 * 2115 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the 2116 * swapper refcount goes to 0, untag the socket as a memalloc socket. 2117 */ 2118 static void 2119 xs_disable_swap(struct rpc_xprt *xprt) 2120 { 2121 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt); 2122 2123 if (!atomic_dec_and_test(&xprt->swapper)) 2124 return; 2125 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) 2126 return; 2127 if (xs->inet) 2128 sk_clear_memalloc(xs->inet); 2129 xprt_release_xprt(xprt, NULL); 2130 } 2131 #else 2132 static void xs_set_memalloc(struct rpc_xprt *xprt) 2133 { 2134 } 2135 2136 static int 2137 xs_enable_swap(struct rpc_xprt *xprt) 2138 { 2139 return -EINVAL; 2140 } 2141 2142 static void 2143 xs_disable_swap(struct rpc_xprt *xprt) 2144 { 2145 } 2146 #endif 2147 2148 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2149 { 2150 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2151 2152 if (!transport->inet) { 2153 struct sock *sk = sock->sk; 2154 2155 write_lock_bh(&sk->sk_callback_lock); 2156 2157 xs_save_old_callbacks(transport, sk); 2158 2159 sk->sk_user_data = xprt; 2160 sk->sk_data_ready = xs_data_ready; 2161 sk->sk_write_space = xs_udp_write_space; 2162 sk->sk_allocation = GFP_NOIO; 2163 2164 xprt_set_connected(xprt); 2165 2166 /* Reset to new socket */ 2167 transport->sock = sock; 2168 transport->inet = sk; 2169 2170 xs_set_memalloc(xprt); 2171 2172 write_unlock_bh(&sk->sk_callback_lock); 2173 } 2174 xs_udp_do_set_buffer_size(xprt); 2175 } 2176 2177 static void xs_udp_setup_socket(struct work_struct *work) 2178 { 2179 struct sock_xprt *transport = 2180 container_of(work, struct sock_xprt, connect_worker.work); 2181 struct rpc_xprt *xprt = &transport->xprt; 2182 struct socket *sock = transport->sock; 2183 int status = -EIO; 2184 2185 sock = xs_create_sock(xprt, transport, 2186 xs_addr(xprt)->sa_family, SOCK_DGRAM, 2187 IPPROTO_UDP, false); 2188 if (IS_ERR(sock)) 2189 goto out; 2190 2191 dprintk("RPC: worker connecting xprt %p via %s to " 2192 "%s (port %s)\n", xprt, 2193 xprt->address_strings[RPC_DISPLAY_PROTO], 2194 xprt->address_strings[RPC_DISPLAY_ADDR], 2195 xprt->address_strings[RPC_DISPLAY_PORT]); 2196 2197 xs_udp_finish_connecting(xprt, sock); 2198 trace_rpc_socket_connect(xprt, sock, 0); 2199 status = 0; 2200 out: 2201 xprt_unlock_connect(xprt, transport); 2202 xprt_clear_connecting(xprt); 2203 xprt_wake_pending_tasks(xprt, status); 2204 } 2205 2206 /** 2207 * xs_tcp_shutdown - gracefully shut down a TCP socket 2208 * @xprt: transport 2209 * 2210 * Initiates a graceful shutdown of the TCP socket by calling the 2211 * equivalent of shutdown(SHUT_RDWR); 2212 */ 2213 static void xs_tcp_shutdown(struct rpc_xprt *xprt) 2214 { 2215 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2216 struct socket *sock = transport->sock; 2217 2218 if (sock == NULL) 2219 return; 2220 if (xprt_connected(xprt)) { 2221 kernel_sock_shutdown(sock, SHUT_RDWR); 2222 trace_rpc_socket_shutdown(xprt, sock); 2223 } else 2224 xs_reset_transport(transport); 2225 } 2226 2227 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock) 2228 { 2229 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2230 int ret = -ENOTCONN; 2231 2232 if (!transport->inet) { 2233 struct sock *sk = sock->sk; 2234 unsigned int keepidle = xprt->timeout->to_initval / HZ; 2235 unsigned int keepcnt = xprt->timeout->to_retries + 1; 2236 unsigned int opt_on = 1; 2237 unsigned int timeo; 2238 2239 /* TCP Keepalive options */ 2240 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 2241 (char *)&opt_on, sizeof(opt_on)); 2242 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE, 2243 (char *)&keepidle, sizeof(keepidle)); 2244 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL, 2245 (char *)&keepidle, sizeof(keepidle)); 2246 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT, 2247 (char *)&keepcnt, sizeof(keepcnt)); 2248 2249 /* TCP user timeout (see RFC5482) */ 2250 timeo = jiffies_to_msecs(xprt->timeout->to_initval) * 2251 (xprt->timeout->to_retries + 1); 2252 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT, 2253 (char *)&timeo, sizeof(timeo)); 2254 2255 write_lock_bh(&sk->sk_callback_lock); 2256 2257 xs_save_old_callbacks(transport, sk); 2258 2259 sk->sk_user_data = xprt; 2260 sk->sk_data_ready = xs_tcp_data_ready; 2261 sk->sk_state_change = xs_tcp_state_change; 2262 sk->sk_write_space = xs_tcp_write_space; 2263 sk->sk_error_report = xs_error_report; 2264 sk->sk_allocation = GFP_NOIO; 2265 2266 /* socket options */ 2267 sock_reset_flag(sk, SOCK_LINGER); 2268 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 2269 2270 xprt_clear_connected(xprt); 2271 2272 /* Reset to new socket */ 2273 transport->sock = sock; 2274 transport->inet = sk; 2275 2276 write_unlock_bh(&sk->sk_callback_lock); 2277 } 2278 2279 if (!xprt_bound(xprt)) 2280 goto out; 2281 2282 xs_set_memalloc(xprt); 2283 2284 /* Tell the socket layer to start connecting... */ 2285 xprt->stat.connect_count++; 2286 xprt->stat.connect_start = jiffies; 2287 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state); 2288 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK); 2289 switch (ret) { 2290 case 0: 2291 xs_set_srcport(transport, sock); 2292 case -EINPROGRESS: 2293 /* SYN_SENT! */ 2294 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2295 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2296 } 2297 out: 2298 return ret; 2299 } 2300 2301 /** 2302 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint 2303 * 2304 * Invoked by a work queue tasklet. 2305 */ 2306 static void xs_tcp_setup_socket(struct work_struct *work) 2307 { 2308 struct sock_xprt *transport = 2309 container_of(work, struct sock_xprt, connect_worker.work); 2310 struct socket *sock = transport->sock; 2311 struct rpc_xprt *xprt = &transport->xprt; 2312 int status = -EIO; 2313 2314 if (!sock) { 2315 sock = xs_create_sock(xprt, transport, 2316 xs_addr(xprt)->sa_family, SOCK_STREAM, 2317 IPPROTO_TCP, true); 2318 if (IS_ERR(sock)) { 2319 status = PTR_ERR(sock); 2320 goto out; 2321 } 2322 } 2323 2324 dprintk("RPC: worker connecting xprt %p via %s to " 2325 "%s (port %s)\n", xprt, 2326 xprt->address_strings[RPC_DISPLAY_PROTO], 2327 xprt->address_strings[RPC_DISPLAY_ADDR], 2328 xprt->address_strings[RPC_DISPLAY_PORT]); 2329 2330 status = xs_tcp_finish_connecting(xprt, sock); 2331 trace_rpc_socket_connect(xprt, sock, status); 2332 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 2333 xprt, -status, xprt_connected(xprt), 2334 sock->sk->sk_state); 2335 switch (status) { 2336 default: 2337 printk("%s: connect returned unhandled error %d\n", 2338 __func__, status); 2339 case -EADDRNOTAVAIL: 2340 /* We're probably in TIME_WAIT. Get rid of existing socket, 2341 * and retry 2342 */ 2343 xs_tcp_force_close(xprt); 2344 break; 2345 case 0: 2346 case -EINPROGRESS: 2347 case -EALREADY: 2348 xprt_unlock_connect(xprt, transport); 2349 return; 2350 case -EINVAL: 2351 /* Happens, for instance, if the user specified a link 2352 * local IPv6 address without a scope-id. 2353 */ 2354 case -ECONNREFUSED: 2355 case -ECONNRESET: 2356 case -ENETUNREACH: 2357 case -EADDRINUSE: 2358 case -ENOBUFS: 2359 /* retry with existing socket, after a delay */ 2360 xs_tcp_force_close(xprt); 2361 goto out; 2362 } 2363 status = -EAGAIN; 2364 out: 2365 xprt_unlock_connect(xprt, transport); 2366 xprt_clear_connecting(xprt); 2367 xprt_wake_pending_tasks(xprt, status); 2368 } 2369 2370 /** 2371 * xs_connect - connect a socket to a remote endpoint 2372 * @xprt: pointer to transport structure 2373 * @task: address of RPC task that manages state of connect request 2374 * 2375 * TCP: If the remote end dropped the connection, delay reconnecting. 2376 * 2377 * UDP socket connects are synchronous, but we use a work queue anyway 2378 * to guarantee that even unprivileged user processes can set up a 2379 * socket on a privileged port. 2380 * 2381 * If a UDP socket connect fails, the delay behavior here prevents 2382 * retry floods (hard mounts). 2383 */ 2384 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task) 2385 { 2386 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2387 2388 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport)); 2389 2390 if (transport->sock != NULL) { 2391 dprintk("RPC: xs_connect delayed xprt %p for %lu " 2392 "seconds\n", 2393 xprt, xprt->reestablish_timeout / HZ); 2394 2395 /* Start by resetting any existing state */ 2396 xs_reset_transport(transport); 2397 2398 queue_delayed_work(rpciod_workqueue, 2399 &transport->connect_worker, 2400 xprt->reestablish_timeout); 2401 xprt->reestablish_timeout <<= 1; 2402 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO) 2403 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2404 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) 2405 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; 2406 } else { 2407 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 2408 queue_delayed_work(rpciod_workqueue, 2409 &transport->connect_worker, 0); 2410 } 2411 } 2412 2413 /** 2414 * xs_local_print_stats - display AF_LOCAL socket-specifc stats 2415 * @xprt: rpc_xprt struct containing statistics 2416 * @seq: output file 2417 * 2418 */ 2419 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2420 { 2421 long idle_time = 0; 2422 2423 if (xprt_connected(xprt)) 2424 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2425 2426 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu " 2427 "%llu %llu %lu %llu %llu\n", 2428 xprt->stat.bind_count, 2429 xprt->stat.connect_count, 2430 xprt->stat.connect_time, 2431 idle_time, 2432 xprt->stat.sends, 2433 xprt->stat.recvs, 2434 xprt->stat.bad_xids, 2435 xprt->stat.req_u, 2436 xprt->stat.bklog_u, 2437 xprt->stat.max_slots, 2438 xprt->stat.sending_u, 2439 xprt->stat.pending_u); 2440 } 2441 2442 /** 2443 * xs_udp_print_stats - display UDP socket-specifc stats 2444 * @xprt: rpc_xprt struct containing statistics 2445 * @seq: output file 2446 * 2447 */ 2448 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2449 { 2450 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2451 2452 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu " 2453 "%lu %llu %llu\n", 2454 transport->srcport, 2455 xprt->stat.bind_count, 2456 xprt->stat.sends, 2457 xprt->stat.recvs, 2458 xprt->stat.bad_xids, 2459 xprt->stat.req_u, 2460 xprt->stat.bklog_u, 2461 xprt->stat.max_slots, 2462 xprt->stat.sending_u, 2463 xprt->stat.pending_u); 2464 } 2465 2466 /** 2467 * xs_tcp_print_stats - display TCP socket-specifc stats 2468 * @xprt: rpc_xprt struct containing statistics 2469 * @seq: output file 2470 * 2471 */ 2472 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 2473 { 2474 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 2475 long idle_time = 0; 2476 2477 if (xprt_connected(xprt)) 2478 idle_time = (long)(jiffies - xprt->last_used) / HZ; 2479 2480 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu " 2481 "%llu %llu %lu %llu %llu\n", 2482 transport->srcport, 2483 xprt->stat.bind_count, 2484 xprt->stat.connect_count, 2485 xprt->stat.connect_time, 2486 idle_time, 2487 xprt->stat.sends, 2488 xprt->stat.recvs, 2489 xprt->stat.bad_xids, 2490 xprt->stat.req_u, 2491 xprt->stat.bklog_u, 2492 xprt->stat.max_slots, 2493 xprt->stat.sending_u, 2494 xprt->stat.pending_u); 2495 } 2496 2497 /* 2498 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason 2499 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want 2500 * to use the server side send routines. 2501 */ 2502 static void *bc_malloc(struct rpc_task *task, size_t size) 2503 { 2504 struct page *page; 2505 struct rpc_buffer *buf; 2506 2507 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer)); 2508 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) 2509 return NULL; 2510 2511 page = alloc_page(GFP_KERNEL); 2512 if (!page) 2513 return NULL; 2514 2515 buf = page_address(page); 2516 buf->len = PAGE_SIZE; 2517 2518 return buf->data; 2519 } 2520 2521 /* 2522 * Free the space allocated in the bc_alloc routine 2523 */ 2524 static void bc_free(void *buffer) 2525 { 2526 struct rpc_buffer *buf; 2527 2528 if (!buffer) 2529 return; 2530 2531 buf = container_of(buffer, struct rpc_buffer, data); 2532 free_page((unsigned long)buf); 2533 } 2534 2535 /* 2536 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex 2537 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request. 2538 */ 2539 static int bc_sendto(struct rpc_rqst *req) 2540 { 2541 int len; 2542 struct xdr_buf *xbufp = &req->rq_snd_buf; 2543 struct rpc_xprt *xprt = req->rq_xprt; 2544 struct sock_xprt *transport = 2545 container_of(xprt, struct sock_xprt, xprt); 2546 struct socket *sock = transport->sock; 2547 unsigned long headoff; 2548 unsigned long tailoff; 2549 2550 xs_encode_stream_record_marker(xbufp); 2551 2552 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK; 2553 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK; 2554 len = svc_send_common(sock, xbufp, 2555 virt_to_page(xbufp->head[0].iov_base), headoff, 2556 xbufp->tail[0].iov_base, tailoff); 2557 2558 if (len != xbufp->len) { 2559 printk(KERN_NOTICE "Error sending entire callback!\n"); 2560 len = -EAGAIN; 2561 } 2562 2563 return len; 2564 } 2565 2566 /* 2567 * The send routine. Borrows from svc_send 2568 */ 2569 static int bc_send_request(struct rpc_task *task) 2570 { 2571 struct rpc_rqst *req = task->tk_rqstp; 2572 struct svc_xprt *xprt; 2573 int len; 2574 2575 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid)); 2576 /* 2577 * Get the server socket associated with this callback xprt 2578 */ 2579 xprt = req->rq_xprt->bc_xprt; 2580 2581 /* 2582 * Grab the mutex to serialize data as the connection is shared 2583 * with the fore channel 2584 */ 2585 if (!mutex_trylock(&xprt->xpt_mutex)) { 2586 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL); 2587 if (!mutex_trylock(&xprt->xpt_mutex)) 2588 return -EAGAIN; 2589 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task); 2590 } 2591 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) 2592 len = -ENOTCONN; 2593 else 2594 len = bc_sendto(req); 2595 mutex_unlock(&xprt->xpt_mutex); 2596 2597 if (len > 0) 2598 len = 0; 2599 2600 return len; 2601 } 2602 2603 /* 2604 * The close routine. Since this is client initiated, we do nothing 2605 */ 2606 2607 static void bc_close(struct rpc_xprt *xprt) 2608 { 2609 } 2610 2611 /* 2612 * The xprt destroy routine. Again, because this connection is client 2613 * initiated, we do nothing 2614 */ 2615 2616 static void bc_destroy(struct rpc_xprt *xprt) 2617 { 2618 dprintk("RPC: bc_destroy xprt %p\n", xprt); 2619 2620 xs_xprt_free(xprt); 2621 module_put(THIS_MODULE); 2622 } 2623 2624 static struct rpc_xprt_ops xs_local_ops = { 2625 .reserve_xprt = xprt_reserve_xprt, 2626 .release_xprt = xs_tcp_release_xprt, 2627 .alloc_slot = xprt_alloc_slot, 2628 .rpcbind = xs_local_rpcbind, 2629 .set_port = xs_local_set_port, 2630 .connect = xs_local_connect, 2631 .buf_alloc = rpc_malloc, 2632 .buf_free = rpc_free, 2633 .send_request = xs_local_send_request, 2634 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2635 .close = xs_close, 2636 .destroy = xs_destroy, 2637 .print_stats = xs_local_print_stats, 2638 .enable_swap = xs_enable_swap, 2639 .disable_swap = xs_disable_swap, 2640 }; 2641 2642 static struct rpc_xprt_ops xs_udp_ops = { 2643 .set_buffer_size = xs_udp_set_buffer_size, 2644 .reserve_xprt = xprt_reserve_xprt_cong, 2645 .release_xprt = xprt_release_xprt_cong, 2646 .alloc_slot = xprt_alloc_slot, 2647 .rpcbind = rpcb_getport_async, 2648 .set_port = xs_set_port, 2649 .connect = xs_connect, 2650 .buf_alloc = rpc_malloc, 2651 .buf_free = rpc_free, 2652 .send_request = xs_udp_send_request, 2653 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 2654 .timer = xs_udp_timer, 2655 .release_request = xprt_release_rqst_cong, 2656 .close = xs_close, 2657 .destroy = xs_destroy, 2658 .print_stats = xs_udp_print_stats, 2659 .enable_swap = xs_enable_swap, 2660 .disable_swap = xs_disable_swap, 2661 .inject_disconnect = xs_inject_disconnect, 2662 }; 2663 2664 static struct rpc_xprt_ops xs_tcp_ops = { 2665 .reserve_xprt = xprt_reserve_xprt, 2666 .release_xprt = xs_tcp_release_xprt, 2667 .alloc_slot = xprt_lock_and_alloc_slot, 2668 .rpcbind = rpcb_getport_async, 2669 .set_port = xs_set_port, 2670 .connect = xs_connect, 2671 .buf_alloc = rpc_malloc, 2672 .buf_free = rpc_free, 2673 .send_request = xs_tcp_send_request, 2674 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2675 .close = xs_tcp_shutdown, 2676 .destroy = xs_destroy, 2677 .print_stats = xs_tcp_print_stats, 2678 .enable_swap = xs_enable_swap, 2679 .disable_swap = xs_disable_swap, 2680 .inject_disconnect = xs_inject_disconnect, 2681 #ifdef CONFIG_SUNRPC_BACKCHANNEL 2682 .bc_setup = xprt_setup_bc, 2683 .bc_up = xs_tcp_bc_up, 2684 .bc_free_rqst = xprt_free_bc_rqst, 2685 .bc_destroy = xprt_destroy_bc, 2686 #endif 2687 }; 2688 2689 /* 2690 * The rpc_xprt_ops for the server backchannel 2691 */ 2692 2693 static struct rpc_xprt_ops bc_tcp_ops = { 2694 .reserve_xprt = xprt_reserve_xprt, 2695 .release_xprt = xprt_release_xprt, 2696 .alloc_slot = xprt_alloc_slot, 2697 .buf_alloc = bc_malloc, 2698 .buf_free = bc_free, 2699 .send_request = bc_send_request, 2700 .set_retrans_timeout = xprt_set_retrans_timeout_def, 2701 .close = bc_close, 2702 .destroy = bc_destroy, 2703 .print_stats = xs_tcp_print_stats, 2704 .enable_swap = xs_enable_swap, 2705 .disable_swap = xs_disable_swap, 2706 .inject_disconnect = xs_inject_disconnect, 2707 }; 2708 2709 static int xs_init_anyaddr(const int family, struct sockaddr *sap) 2710 { 2711 static const struct sockaddr_in sin = { 2712 .sin_family = AF_INET, 2713 .sin_addr.s_addr = htonl(INADDR_ANY), 2714 }; 2715 static const struct sockaddr_in6 sin6 = { 2716 .sin6_family = AF_INET6, 2717 .sin6_addr = IN6ADDR_ANY_INIT, 2718 }; 2719 2720 switch (family) { 2721 case AF_LOCAL: 2722 break; 2723 case AF_INET: 2724 memcpy(sap, &sin, sizeof(sin)); 2725 break; 2726 case AF_INET6: 2727 memcpy(sap, &sin6, sizeof(sin6)); 2728 break; 2729 default: 2730 dprintk("RPC: %s: Bad address family\n", __func__); 2731 return -EAFNOSUPPORT; 2732 } 2733 return 0; 2734 } 2735 2736 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args, 2737 unsigned int slot_table_size, 2738 unsigned int max_slot_table_size) 2739 { 2740 struct rpc_xprt *xprt; 2741 struct sock_xprt *new; 2742 2743 if (args->addrlen > sizeof(xprt->addr)) { 2744 dprintk("RPC: xs_setup_xprt: address too large\n"); 2745 return ERR_PTR(-EBADF); 2746 } 2747 2748 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size, 2749 max_slot_table_size); 2750 if (xprt == NULL) { 2751 dprintk("RPC: xs_setup_xprt: couldn't allocate " 2752 "rpc_xprt\n"); 2753 return ERR_PTR(-ENOMEM); 2754 } 2755 2756 new = container_of(xprt, struct sock_xprt, xprt); 2757 mutex_init(&new->recv_mutex); 2758 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 2759 xprt->addrlen = args->addrlen; 2760 if (args->srcaddr) 2761 memcpy(&new->srcaddr, args->srcaddr, args->addrlen); 2762 else { 2763 int err; 2764 err = xs_init_anyaddr(args->dstaddr->sa_family, 2765 (struct sockaddr *)&new->srcaddr); 2766 if (err != 0) { 2767 xprt_free(xprt); 2768 return ERR_PTR(err); 2769 } 2770 } 2771 2772 return xprt; 2773 } 2774 2775 static const struct rpc_timeout xs_local_default_timeout = { 2776 .to_initval = 10 * HZ, 2777 .to_maxval = 10 * HZ, 2778 .to_retries = 2, 2779 }; 2780 2781 /** 2782 * xs_setup_local - Set up transport to use an AF_LOCAL socket 2783 * @args: rpc transport creation arguments 2784 * 2785 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP 2786 */ 2787 static struct rpc_xprt *xs_setup_local(struct xprt_create *args) 2788 { 2789 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr; 2790 struct sock_xprt *transport; 2791 struct rpc_xprt *xprt; 2792 struct rpc_xprt *ret; 2793 2794 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2795 xprt_max_tcp_slot_table_entries); 2796 if (IS_ERR(xprt)) 2797 return xprt; 2798 transport = container_of(xprt, struct sock_xprt, xprt); 2799 2800 xprt->prot = 0; 2801 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2802 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2803 2804 xprt->bind_timeout = XS_BIND_TO; 2805 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2806 xprt->idle_timeout = XS_IDLE_DISC_TO; 2807 2808 xprt->ops = &xs_local_ops; 2809 xprt->timeout = &xs_local_default_timeout; 2810 2811 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn); 2812 INIT_DELAYED_WORK(&transport->connect_worker, 2813 xs_dummy_setup_socket); 2814 2815 switch (sun->sun_family) { 2816 case AF_LOCAL: 2817 if (sun->sun_path[0] != '/') { 2818 dprintk("RPC: bad AF_LOCAL address: %s\n", 2819 sun->sun_path); 2820 ret = ERR_PTR(-EINVAL); 2821 goto out_err; 2822 } 2823 xprt_set_bound(xprt); 2824 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL); 2825 ret = ERR_PTR(xs_local_setup_socket(transport)); 2826 if (ret) 2827 goto out_err; 2828 break; 2829 default: 2830 ret = ERR_PTR(-EAFNOSUPPORT); 2831 goto out_err; 2832 } 2833 2834 dprintk("RPC: set up xprt to %s via AF_LOCAL\n", 2835 xprt->address_strings[RPC_DISPLAY_ADDR]); 2836 2837 if (try_module_get(THIS_MODULE)) 2838 return xprt; 2839 ret = ERR_PTR(-EINVAL); 2840 out_err: 2841 xs_xprt_free(xprt); 2842 return ret; 2843 } 2844 2845 static const struct rpc_timeout xs_udp_default_timeout = { 2846 .to_initval = 5 * HZ, 2847 .to_maxval = 30 * HZ, 2848 .to_increment = 5 * HZ, 2849 .to_retries = 5, 2850 }; 2851 2852 /** 2853 * xs_setup_udp - Set up transport to use a UDP socket 2854 * @args: rpc transport creation arguments 2855 * 2856 */ 2857 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args) 2858 { 2859 struct sockaddr *addr = args->dstaddr; 2860 struct rpc_xprt *xprt; 2861 struct sock_xprt *transport; 2862 struct rpc_xprt *ret; 2863 2864 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries, 2865 xprt_udp_slot_table_entries); 2866 if (IS_ERR(xprt)) 2867 return xprt; 2868 transport = container_of(xprt, struct sock_xprt, xprt); 2869 2870 xprt->prot = IPPROTO_UDP; 2871 xprt->tsh_size = 0; 2872 /* XXX: header size can vary due to auth type, IPv6, etc. */ 2873 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 2874 2875 xprt->bind_timeout = XS_BIND_TO; 2876 xprt->reestablish_timeout = XS_UDP_REEST_TO; 2877 xprt->idle_timeout = XS_IDLE_DISC_TO; 2878 2879 xprt->ops = &xs_udp_ops; 2880 2881 xprt->timeout = &xs_udp_default_timeout; 2882 2883 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn); 2884 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket); 2885 2886 switch (addr->sa_family) { 2887 case AF_INET: 2888 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2889 xprt_set_bound(xprt); 2890 2891 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP); 2892 break; 2893 case AF_INET6: 2894 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2895 xprt_set_bound(xprt); 2896 2897 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6); 2898 break; 2899 default: 2900 ret = ERR_PTR(-EAFNOSUPPORT); 2901 goto out_err; 2902 } 2903 2904 if (xprt_bound(xprt)) 2905 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2906 xprt->address_strings[RPC_DISPLAY_ADDR], 2907 xprt->address_strings[RPC_DISPLAY_PORT], 2908 xprt->address_strings[RPC_DISPLAY_PROTO]); 2909 else 2910 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2911 xprt->address_strings[RPC_DISPLAY_ADDR], 2912 xprt->address_strings[RPC_DISPLAY_PROTO]); 2913 2914 if (try_module_get(THIS_MODULE)) 2915 return xprt; 2916 ret = ERR_PTR(-EINVAL); 2917 out_err: 2918 xs_xprt_free(xprt); 2919 return ret; 2920 } 2921 2922 static const struct rpc_timeout xs_tcp_default_timeout = { 2923 .to_initval = 60 * HZ, 2924 .to_maxval = 60 * HZ, 2925 .to_retries = 2, 2926 }; 2927 2928 /** 2929 * xs_setup_tcp - Set up transport to use a TCP socket 2930 * @args: rpc transport creation arguments 2931 * 2932 */ 2933 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args) 2934 { 2935 struct sockaddr *addr = args->dstaddr; 2936 struct rpc_xprt *xprt; 2937 struct sock_xprt *transport; 2938 struct rpc_xprt *ret; 2939 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries; 2940 2941 if (args->flags & XPRT_CREATE_INFINITE_SLOTS) 2942 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT; 2943 2944 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 2945 max_slot_table_size); 2946 if (IS_ERR(xprt)) 2947 return xprt; 2948 transport = container_of(xprt, struct sock_xprt, xprt); 2949 2950 xprt->prot = IPPROTO_TCP; 2951 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 2952 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 2953 2954 xprt->bind_timeout = XS_BIND_TO; 2955 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 2956 xprt->idle_timeout = XS_IDLE_DISC_TO; 2957 2958 xprt->ops = &xs_tcp_ops; 2959 xprt->timeout = &xs_tcp_default_timeout; 2960 2961 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn); 2962 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket); 2963 2964 switch (addr->sa_family) { 2965 case AF_INET: 2966 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 2967 xprt_set_bound(xprt); 2968 2969 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP); 2970 break; 2971 case AF_INET6: 2972 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 2973 xprt_set_bound(xprt); 2974 2975 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6); 2976 break; 2977 default: 2978 ret = ERR_PTR(-EAFNOSUPPORT); 2979 goto out_err; 2980 } 2981 2982 if (xprt_bound(xprt)) 2983 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 2984 xprt->address_strings[RPC_DISPLAY_ADDR], 2985 xprt->address_strings[RPC_DISPLAY_PORT], 2986 xprt->address_strings[RPC_DISPLAY_PROTO]); 2987 else 2988 dprintk("RPC: set up xprt to %s (autobind) via %s\n", 2989 xprt->address_strings[RPC_DISPLAY_ADDR], 2990 xprt->address_strings[RPC_DISPLAY_PROTO]); 2991 2992 if (try_module_get(THIS_MODULE)) 2993 return xprt; 2994 ret = ERR_PTR(-EINVAL); 2995 out_err: 2996 xs_xprt_free(xprt); 2997 return ret; 2998 } 2999 3000 /** 3001 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket 3002 * @args: rpc transport creation arguments 3003 * 3004 */ 3005 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args) 3006 { 3007 struct sockaddr *addr = args->dstaddr; 3008 struct rpc_xprt *xprt; 3009 struct sock_xprt *transport; 3010 struct svc_sock *bc_sock; 3011 struct rpc_xprt *ret; 3012 3013 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries, 3014 xprt_tcp_slot_table_entries); 3015 if (IS_ERR(xprt)) 3016 return xprt; 3017 transport = container_of(xprt, struct sock_xprt, xprt); 3018 3019 xprt->prot = IPPROTO_TCP; 3020 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 3021 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 3022 xprt->timeout = &xs_tcp_default_timeout; 3023 3024 /* backchannel */ 3025 xprt_set_bound(xprt); 3026 xprt->bind_timeout = 0; 3027 xprt->reestablish_timeout = 0; 3028 xprt->idle_timeout = 0; 3029 3030 xprt->ops = &bc_tcp_ops; 3031 3032 switch (addr->sa_family) { 3033 case AF_INET: 3034 xs_format_peer_addresses(xprt, "tcp", 3035 RPCBIND_NETID_TCP); 3036 break; 3037 case AF_INET6: 3038 xs_format_peer_addresses(xprt, "tcp", 3039 RPCBIND_NETID_TCP6); 3040 break; 3041 default: 3042 ret = ERR_PTR(-EAFNOSUPPORT); 3043 goto out_err; 3044 } 3045 3046 dprintk("RPC: set up xprt to %s (port %s) via %s\n", 3047 xprt->address_strings[RPC_DISPLAY_ADDR], 3048 xprt->address_strings[RPC_DISPLAY_PORT], 3049 xprt->address_strings[RPC_DISPLAY_PROTO]); 3050 3051 /* 3052 * Once we've associated a backchannel xprt with a connection, 3053 * we want to keep it around as long as the connection lasts, 3054 * in case we need to start using it for a backchannel again; 3055 * this reference won't be dropped until bc_xprt is destroyed. 3056 */ 3057 xprt_get(xprt); 3058 args->bc_xprt->xpt_bc_xprt = xprt; 3059 xprt->bc_xprt = args->bc_xprt; 3060 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt); 3061 transport->sock = bc_sock->sk_sock; 3062 transport->inet = bc_sock->sk_sk; 3063 3064 /* 3065 * Since we don't want connections for the backchannel, we set 3066 * the xprt status to connected 3067 */ 3068 xprt_set_connected(xprt); 3069 3070 if (try_module_get(THIS_MODULE)) 3071 return xprt; 3072 3073 args->bc_xprt->xpt_bc_xprt = NULL; 3074 xprt_put(xprt); 3075 ret = ERR_PTR(-EINVAL); 3076 out_err: 3077 xs_xprt_free(xprt); 3078 return ret; 3079 } 3080 3081 static struct xprt_class xs_local_transport = { 3082 .list = LIST_HEAD_INIT(xs_local_transport.list), 3083 .name = "named UNIX socket", 3084 .owner = THIS_MODULE, 3085 .ident = XPRT_TRANSPORT_LOCAL, 3086 .setup = xs_setup_local, 3087 }; 3088 3089 static struct xprt_class xs_udp_transport = { 3090 .list = LIST_HEAD_INIT(xs_udp_transport.list), 3091 .name = "udp", 3092 .owner = THIS_MODULE, 3093 .ident = XPRT_TRANSPORT_UDP, 3094 .setup = xs_setup_udp, 3095 }; 3096 3097 static struct xprt_class xs_tcp_transport = { 3098 .list = LIST_HEAD_INIT(xs_tcp_transport.list), 3099 .name = "tcp", 3100 .owner = THIS_MODULE, 3101 .ident = XPRT_TRANSPORT_TCP, 3102 .setup = xs_setup_tcp, 3103 }; 3104 3105 static struct xprt_class xs_bc_tcp_transport = { 3106 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list), 3107 .name = "tcp NFSv4.1 backchannel", 3108 .owner = THIS_MODULE, 3109 .ident = XPRT_TRANSPORT_BC_TCP, 3110 .setup = xs_setup_bc_tcp, 3111 }; 3112 3113 /** 3114 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client 3115 * 3116 */ 3117 int init_socket_xprt(void) 3118 { 3119 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 3120 if (!sunrpc_table_header) 3121 sunrpc_table_header = register_sysctl_table(sunrpc_table); 3122 #endif 3123 3124 xprt_register_transport(&xs_local_transport); 3125 xprt_register_transport(&xs_udp_transport); 3126 xprt_register_transport(&xs_tcp_transport); 3127 xprt_register_transport(&xs_bc_tcp_transport); 3128 3129 return 0; 3130 } 3131 3132 /** 3133 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister 3134 * 3135 */ 3136 void cleanup_socket_xprt(void) 3137 { 3138 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 3139 if (sunrpc_table_header) { 3140 unregister_sysctl_table(sunrpc_table_header); 3141 sunrpc_table_header = NULL; 3142 } 3143 #endif 3144 3145 xprt_unregister_transport(&xs_local_transport); 3146 xprt_unregister_transport(&xs_udp_transport); 3147 xprt_unregister_transport(&xs_tcp_transport); 3148 xprt_unregister_transport(&xs_bc_tcp_transport); 3149 } 3150 3151 static int param_set_uint_minmax(const char *val, 3152 const struct kernel_param *kp, 3153 unsigned int min, unsigned int max) 3154 { 3155 unsigned int num; 3156 int ret; 3157 3158 if (!val) 3159 return -EINVAL; 3160 ret = kstrtouint(val, 0, &num); 3161 if (ret == -EINVAL || num < min || num > max) 3162 return -EINVAL; 3163 *((unsigned int *)kp->arg) = num; 3164 return 0; 3165 } 3166 3167 static int param_set_portnr(const char *val, const struct kernel_param *kp) 3168 { 3169 return param_set_uint_minmax(val, kp, 3170 RPC_MIN_RESVPORT, 3171 RPC_MAX_RESVPORT); 3172 } 3173 3174 static const struct kernel_param_ops param_ops_portnr = { 3175 .set = param_set_portnr, 3176 .get = param_get_uint, 3177 }; 3178 3179 #define param_check_portnr(name, p) \ 3180 __param_check(name, p, unsigned int); 3181 3182 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644); 3183 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644); 3184 3185 static int param_set_slot_table_size(const char *val, 3186 const struct kernel_param *kp) 3187 { 3188 return param_set_uint_minmax(val, kp, 3189 RPC_MIN_SLOT_TABLE, 3190 RPC_MAX_SLOT_TABLE); 3191 } 3192 3193 static const struct kernel_param_ops param_ops_slot_table_size = { 3194 .set = param_set_slot_table_size, 3195 .get = param_get_uint, 3196 }; 3197 3198 #define param_check_slot_table_size(name, p) \ 3199 __param_check(name, p, unsigned int); 3200 3201 static int param_set_max_slot_table_size(const char *val, 3202 const struct kernel_param *kp) 3203 { 3204 return param_set_uint_minmax(val, kp, 3205 RPC_MIN_SLOT_TABLE, 3206 RPC_MAX_SLOT_TABLE_LIMIT); 3207 } 3208 3209 static const struct kernel_param_ops param_ops_max_slot_table_size = { 3210 .set = param_set_max_slot_table_size, 3211 .get = param_get_uint, 3212 }; 3213 3214 #define param_check_max_slot_table_size(name, p) \ 3215 __param_check(name, p, unsigned int); 3216 3217 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries, 3218 slot_table_size, 0644); 3219 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries, 3220 max_slot_table_size, 0644); 3221 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries, 3222 slot_table_size, 0644); 3223 3224