net/sunrpc/xprtsock.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/net/sunrpc/xprtsock.c
   4  *
   5  * Client-side transport implementation for sockets.
   6  *
   7  * TCP callback races fixes (C) 1998 Red Hat
   8  * TCP send fixes (C) 1998 Red Hat
   9  * TCP NFS related read + write fixes
  10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  11  *
  12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  13  * Fix behaviour when socket buffer is full.
  14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  15  *
  16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  17  *
  18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  19  *   <gilles.quillard@bull.net>
  20  */
  21
  22 #include <linux/types.h>
  23 #include <linux/string.h>
  24 #include <linux/slab.h>
  25 #include <linux/module.h>
  26 #include <linux/capability.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/errno.h>
  29 #include <linux/socket.h>
  30 #include <linux/in.h>
  31 #include <linux/net.h>
  32 #include <linux/mm.h>
  33 #include <linux/un.h>
  34 #include <linux/udp.h>
  35 #include <linux/tcp.h>
  36 #include <linux/sunrpc/clnt.h>
  37 #include <linux/sunrpc/addr.h>
  38 #include <linux/sunrpc/sched.h>
  39 #include <linux/sunrpc/svcsock.h>
  40 #include <linux/sunrpc/xprtsock.h>
  41 #include <linux/file.h>
  42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  43 #include <linux/sunrpc/bc_xprt.h>
  44 #endif
  45
  46 #include <net/sock.h>
  47 #include <net/checksum.h>
  48 #include <net/udp.h>
  49 #include <net/tcp.h>
  50
  51 #include <trace/events/sunrpc.h>
  52
  53 #include "sunrpc.h"
  54
  55 static void xs_close(struct rpc_xprt *xprt);
  56 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  57                 struct socket *sock);
  58
  59 /*
  60  * xprtsock tunables
  61  */
  62 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  63 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  64 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  65
  66 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  67 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  68
  69 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  70
  71 #define XS_TCP_LINGER_TO        (15U * HZ)
  72 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  73
  74 /*
  75  * We can register our own files under /proc/sys/sunrpc by
  76  * calling register_sysctl_table() again.  The files in that
  77  * directory become the union of all files registered there.
  78  *
  79  * We simply need to make sure that we don't collide with
  80  * someone else's file names!
  81  */
  82
  83 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  84 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  85 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  86 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  87 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  88
  89 static struct ctl_table_header *sunrpc_table_header;
  90
  91 /*
  92  * FIXME: changing the UDP slot table size should also resize the UDP
  93  *        socket buffers for existing UDP transports
  94  */
  95 static struct ctl_table xs_tunables_table[] = {
  96         {
  97                 .procname       = "udp_slot_table_entries",
  98                 .data           = &xprt_udp_slot_table_entries,
  99                 .maxlen         = sizeof(unsigned int),
 100                 .mode           = 0644,
 101                 .proc_handler   = proc_dointvec_minmax,
 102                 .extra1         = &min_slot_table_size,
 103                 .extra2         = &max_slot_table_size
 104         },
 105         {
 106                 .procname       = "tcp_slot_table_entries",
 107                 .data           = &xprt_tcp_slot_table_entries,
 108                 .maxlen         = sizeof(unsigned int),
 109                 .mode           = 0644,
 110                 .proc_handler   = proc_dointvec_minmax,
 111                 .extra1         = &min_slot_table_size,
 112                 .extra2         = &max_slot_table_size
 113         },
 114         {
 115                 .procname       = "tcp_max_slot_table_entries",
 116                 .data           = &xprt_max_tcp_slot_table_entries,
 117                 .maxlen         = sizeof(unsigned int),
 118                 .mode           = 0644,
 119                 .proc_handler   = proc_dointvec_minmax,
 120                 .extra1         = &min_slot_table_size,
 121                 .extra2         = &max_tcp_slot_table_limit
 122         },
 123         {
 124                 .procname       = "min_resvport",
 125                 .data           = &xprt_min_resvport,
 126                 .maxlen         = sizeof(unsigned int),
 127                 .mode           = 0644,
 128                 .proc_handler   = proc_dointvec_minmax,
 129                 .extra1         = &xprt_min_resvport_limit,
 130                 .extra2         = &xprt_max_resvport_limit
 131         },
 132         {
 133                 .procname       = "max_resvport",
 134                 .data           = &xprt_max_resvport,
 135                 .maxlen         = sizeof(unsigned int),
 136                 .mode           = 0644,
 137                 .proc_handler   = proc_dointvec_minmax,
 138                 .extra1         = &xprt_min_resvport_limit,
 139                 .extra2         = &xprt_max_resvport_limit
 140         },
 141         {
 142                 .procname       = "tcp_fin_timeout",
 143                 .data           = &xs_tcp_fin_timeout,
 144                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 145                 .mode           = 0644,
 146                 .proc_handler   = proc_dointvec_jiffies,
 147         },
 148         { },
 149 };
 150
 151 static struct ctl_table sunrpc_table[] = {
 152         {
 153                 .procname       = "sunrpc",
 154                 .mode           = 0555,
 155                 .child          = xs_tunables_table
 156         },
 157         { },
 158 };
 159
 160 #endif
 161
 162 /*
 163  * Wait duration for a reply from the RPC portmapper.
 164  */
 165 #define XS_BIND_TO              (60U * HZ)
 166
 167 /*
 168  * Delay if a UDP socket connect error occurs.  This is most likely some
 169  * kind of resource problem on the local host.
 170  */
 171 #define XS_UDP_REEST_TO         (2U * HZ)
 172
 173 /*
 174  * The reestablish timeout allows clients to delay for a bit before attempting
 175  * to reconnect to a server that just dropped our connection.
 176  *
 177  * We implement an exponential backoff when trying to reestablish a TCP
 178  * transport connection with the server.  Some servers like to drop a TCP
 179  * connection when they are overworked, so we start with a short timeout and
 180  * increase over time if the server is down or not responding.
 181  */
 182 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 183
 184 /*
 185  * TCP idle timeout; client drops the transport socket if it is idle
 186  * for this long.  Note that we also timeout UDP sockets to prevent
 187  * holding port numbers when there is no RPC traffic.
 188  */
 189 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 190
 191 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 192 # undef  RPC_DEBUG_DATA
 193 # define RPCDBG_FACILITY        RPCDBG_TRANS
 194 #endif
 195
 196 #ifdef RPC_DEBUG_DATA
 197 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 198 {
 199         u8 *buf = (u8 *) packet;
 200         int j;
 201
 202         dprintk("RPC:       %s\n", msg);
 203         for (j = 0; j < count && j < 128; j += 4) {
 204                 if (!(j & 31)) {
 205                         if (j)
 206                                 dprintk("\n");
 207                         dprintk("0x%04x ", j);
 208                 }
 209                 dprintk("%02x%02x%02x%02x ",
 210                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 211         }
 212         dprintk("\n");
 213 }
 214 #else
 215 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 216 {
 217         /* NOP */
 218 }
 219 #endif
 220
 221 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 222 {
 223         return (struct rpc_xprt *) sk->sk_user_data;
 224 }
 225
 226 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 227 {
 228         return (struct sockaddr *) &xprt->addr;
 229 }
 230
 231 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 232 {
 233         return (struct sockaddr_un *) &xprt->addr;
 234 }
 235
 236 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 237 {
 238         return (struct sockaddr_in *) &xprt->addr;
 239 }
 240
 241 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 242 {
 243         return (struct sockaddr_in6 *) &xprt->addr;
 244 }
 245
 246 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 247 {
 248         struct sockaddr *sap = xs_addr(xprt);
 249         struct sockaddr_in6 *sin6;
 250         struct sockaddr_in *sin;
 251         struct sockaddr_un *sun;
 252         char buf[128];
 253
 254         switch (sap->sa_family) {
 255         case AF_LOCAL:
 256                 sun = xs_addr_un(xprt);
 257                 strlcpy(buf, sun->sun_path, sizeof(buf));
 258                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 259                                                 kstrdup(buf, GFP_KERNEL);
 260                 break;
 261         case AF_INET:
 262                 (void)rpc_ntop(sap, buf, sizeof(buf));
 263                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 264                                                 kstrdup(buf, GFP_KERNEL);
 265                 sin = xs_addr_in(xprt);
 266                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 267                 break;
 268         case AF_INET6:
 269                 (void)rpc_ntop(sap, buf, sizeof(buf));
 270                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 271                                                 kstrdup(buf, GFP_KERNEL);
 272                 sin6 = xs_addr_in6(xprt);
 273                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 274                 break;
 275         default:
 276                 BUG();
 277         }
 278
 279         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 280 }
 281
 282 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 283 {
 284         struct sockaddr *sap = xs_addr(xprt);
 285         char buf[128];
 286
 287         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 288         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 289
 290         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 291         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 292 }
 293
 294 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 295                                      const char *protocol,
 296                                      const char *netid)
 297 {
 298         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 299         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 300         xs_format_common_peer_addresses(xprt);
 301         xs_format_common_peer_ports(xprt);
 302 }
 303
 304 static void xs_update_peer_port(struct rpc_xprt *xprt)
 305 {
 306         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 307         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 308
 309         xs_format_common_peer_ports(xprt);
 310 }
 311
 312 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 313 {
 314         unsigned int i;
 315
 316         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 317                 switch (i) {
 318                 case RPC_DISPLAY_PROTO:
 319                 case RPC_DISPLAY_NETID:
 320                         continue;
 321                 default:
 322                         kfree(xprt->address_strings[i]);
 323                 }
 324 }
 325
 326 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 327
 328 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 329 {
 330         struct msghdr msg = {
 331                 .msg_name       = addr,
 332                 .msg_namelen    = addrlen,
 333                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 334         };
 335         struct kvec iov = {
 336                 .iov_base       = vec->iov_base + base,
 337                 .iov_len        = vec->iov_len - base,
 338         };
 339
 340         if (iov.iov_len != 0)
 341                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 342         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 343 }
 344
 345 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 346 {
 347         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 348                         int offset, size_t size, int flags);
 349         struct page **ppage;
 350         unsigned int remainder;
 351         int err;
 352
 353         remainder = xdr->page_len - base;
 354         base += xdr->page_base;
 355         ppage = xdr->pages + (base >> PAGE_SHIFT);
 356         base &= ~PAGE_MASK;
 357         do_sendpage = sock->ops->sendpage;
 358         if (!zerocopy)
 359                 do_sendpage = sock_no_sendpage;
 360         for(;;) {
 361                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 362                 int flags = XS_SENDMSG_FLAGS;
 363
 364                 remainder -= len;
 365                 if (more)
 366                         flags |= MSG_MORE;
 367                 if (remainder != 0)
 368                         flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
 369                 err = do_sendpage(sock, *ppage, base, len, flags);
 370                 if (remainder == 0 || err != len)
 371                         break;
 372                 *sent_p += err;
 373                 ppage++;
 374                 base = 0;
 375         }
 376         if (err > 0) {
 377                 *sent_p += err;
 378                 err = 0;
 379         }
 380         return err;
 381 }
 382
 383 /**
 384  * xs_sendpages - write pages directly to a socket
 385  * @sock: socket to send on
 386  * @addr: UDP only -- address of destination
 387  * @addrlen: UDP only -- length of destination address
 388  * @xdr: buffer containing this request
 389  * @base: starting position in the buffer
 390  * @zerocopy: true if it is safe to use sendpage()
 391  * @sent_p: return the total number of bytes successfully queued for sending
 392  *
 393  */
 394 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 395 {
 396         unsigned int remainder = xdr->len - base;
 397         int err = 0;
 398         int sent = 0;
 399
 400         if (unlikely(!sock))
 401                 return -ENOTSOCK;
 402
 403         if (base != 0) {
 404                 addr = NULL;
 405                 addrlen = 0;
 406         }
 407
 408         if (base < xdr->head[0].iov_len || addr != NULL) {
 409                 unsigned int len = xdr->head[0].iov_len - base;
 410                 remainder -= len;
 411                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 412                 if (remainder == 0 || err != len)
 413                         goto out;
 414                 *sent_p += err;
 415                 base = 0;
 416         } else
 417                 base -= xdr->head[0].iov_len;
 418
 419         if (base < xdr->page_len) {
 420                 unsigned int len = xdr->page_len - base;
 421                 remainder -= len;
 422                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 423                 *sent_p += sent;
 424                 if (remainder == 0 || sent != len)
 425                         goto out;
 426                 base = 0;
 427         } else
 428                 base -= xdr->page_len;
 429
 430         if (base >= xdr->tail[0].iov_len)
 431                 return 0;
 432         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 433 out:
 434         if (err > 0) {
 435                 *sent_p += err;
 436                 err = 0;
 437         }
 438         return err;
 439 }
 440
 441 static void xs_nospace_callback(struct rpc_task *task)
 442 {
 443         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 444
 445         transport->inet->sk_write_pending--;
 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                 /* wait for more buffer space */
 471                 sk->sk_write_pending++;
 472                 xprt_wait_for_buffer_space(task, xs_nospace_callback);
 473         } else
 474                 ret = -ENOTCONN;
 475
 476         spin_unlock_bh(&xprt->transport_lock);
 477
 478         /* Race breaker in case memory is freed before above code is called */
 479         if (ret == -EAGAIN) {
 480                 struct socket_wq *wq;
 481
 482                 rcu_read_lock();
 483                 wq = rcu_dereference(sk->sk_wq);
 484                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 485                 rcu_read_unlock();
 486
 487                 sk->sk_write_space(sk);
 488         }
 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         case -ENETUNREACH:
 620         case -ENOBUFS:
 621         case -EPIPE:
 622         case -ECONNREFUSED:
 623         case -EPERM:
 624                 /* When the server has died, an ICMP port unreachable message
 625                  * prompts ECONNREFUSED. */
 626                 break;
 627         default:
 628                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 629                         -status);
 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         bool vm_wait = false;
 657         int status;
 658         int sent;
 659
 660         xs_encode_stream_record_marker(&req->rq_snd_buf);
 661
 662         xs_pktdump("packet data:",
 663                                 req->rq_svec->iov_base,
 664                                 req->rq_svec->iov_len);
 665         /* Don't use zero copy if this is a resend. If the RPC call
 666          * completes while the socket holds a reference to the pages,
 667          * then we may end up resending corrupted data.
 668          */
 669         if (task->tk_flags & RPC_TASK_SENT)
 670                 zerocopy = false;
 671
 672         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
 673                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
 674
 675         /* Continue transmitting the packet/record. We must be careful
 676          * to cope with writespace callbacks arriving _after_ we have
 677          * called sendmsg(). */
 678         while (1) {
 679                 sent = 0;
 680                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 681                                       req->rq_bytes_sent, zerocopy, &sent);
 682
 683                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 684                                 xdr->len - req->rq_bytes_sent, status);
 685
 686                 /* If we've sent the entire packet, immediately
 687                  * reset the count of bytes sent. */
 688                 req->rq_bytes_sent += sent;
 689                 req->rq_xmit_bytes_sent += sent;
 690                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 691                         req->rq_bytes_sent = 0;
 692                         return 0;
 693                 }
 694
 695                 WARN_ON_ONCE(sent == 0 && status == 0);
 696
 697                 if (status == -EAGAIN ) {
 698                         /*
 699                          * Return EAGAIN if we're sure we're hitting the
 700                          * socket send buffer limits.
 701                          */
 702                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 703                                 break;
 704                         /*
 705                          * Did we hit a memory allocation failure?
 706                          */
 707                         if (sent == 0) {
 708                                 status = -ENOBUFS;
 709                                 if (vm_wait)
 710                                         break;
 711                                 /* Retry, knowing now that we're below the
 712                                  * socket send buffer limit
 713                                  */
 714                                 vm_wait = true;
 715                         }
 716                         continue;
 717                 }
 718                 if (status < 0)
 719                         break;
 720                 vm_wait = false;
 721         }
 722
 723         switch (status) {
 724         case -ENOTSOCK:
 725                 status = -ENOTCONN;
 726                 /* Should we call xs_close() here? */
 727                 break;
 728         case -EAGAIN:
 729                 status = xs_nospace(task);
 730                 break;
 731         case -ECONNRESET:
 732         case -ECONNREFUSED:
 733         case -ENOTCONN:
 734         case -EADDRINUSE:
 735         case -ENOBUFS:
 736         case -EPIPE:
 737                 break;
 738         default:
 739                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 740                         -status);
 741         }
 742
 743         return status;
 744 }
 745
 746 /**
 747  * xs_tcp_release_xprt - clean up after a tcp transmission
 748  * @xprt: transport
 749  * @task: rpc task
 750  *
 751  * This cleans up if an error causes us to abort the transmission of a request.
 752  * In this case, the socket may need to be reset in order to avoid confusing
 753  * the server.
 754  */
 755 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 756 {
 757         struct rpc_rqst *req;
 758
 759         if (task != xprt->snd_task)
 760                 return;
 761         if (task == NULL)
 762                 goto out_release;
 763         req = task->tk_rqstp;
 764         if (req == NULL)
 765                 goto out_release;
 766         if (req->rq_bytes_sent == 0)
 767                 goto out_release;
 768         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 769                 goto out_release;
 770         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 771 out_release:
 772         xprt_release_xprt(xprt, task);
 773 }
 774
 775 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 776 {
 777         transport->old_data_ready = sk->sk_data_ready;
 778         transport->old_state_change = sk->sk_state_change;
 779         transport->old_write_space = sk->sk_write_space;
 780         transport->old_error_report = sk->sk_error_report;
 781 }
 782
 783 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 784 {
 785         sk->sk_data_ready = transport->old_data_ready;
 786         sk->sk_state_change = transport->old_state_change;
 787         sk->sk_write_space = transport->old_write_space;
 788         sk->sk_error_report = transport->old_error_report;
 789 }
 790
 791 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
 792 {
 793         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 794
 795         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 796 }
 797
 798 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 799 {
 800         smp_mb__before_atomic();
 801         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 802         clear_bit(XPRT_CLOSING, &xprt->state);
 803         xs_sock_reset_state_flags(xprt);
 804         smp_mb__after_atomic();
 805 }
 806
 807 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 808 {
 809         xs_sock_reset_connection_flags(xprt);
 810         /* Mark transport as closed and wake up all pending tasks */
 811         xprt_disconnect_done(xprt);
 812 }
 813
 814 /**
 815  * xs_error_report - callback to handle TCP socket state errors
 816  * @sk: socket
 817  *
 818  * Note: we don't call sock_error() since there may be a rpc_task
 819  * using the socket, and so we don't want to clear sk->sk_err.
 820  */
 821 static void xs_error_report(struct sock *sk)
 822 {
 823         struct rpc_xprt *xprt;
 824         int err;
 825
 826         read_lock_bh(&sk->sk_callback_lock);
 827         if (!(xprt = xprt_from_sock(sk)))
 828                 goto out;
 829
 830         err = -sk->sk_err;
 831         if (err == 0)
 832                 goto out;
 833         /* Is this a reset event? */
 834         if (sk->sk_state == TCP_CLOSE)
 835                 xs_sock_mark_closed(xprt);
 836         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 837                         xprt, -err);
 838         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 839         xprt_wake_pending_tasks(xprt, err);
 840  out:
 841         read_unlock_bh(&sk->sk_callback_lock);
 842 }
 843
 844 static void xs_reset_transport(struct sock_xprt *transport)
 845 {
 846         struct socket *sock = transport->sock;
 847         struct sock *sk = transport->inet;
 848         struct rpc_xprt *xprt = &transport->xprt;
 849
 850         if (sk == NULL)
 851                 return;
 852
 853         if (atomic_read(&transport->xprt.swapper))
 854                 sk_clear_memalloc(sk);
 855
 856         kernel_sock_shutdown(sock, SHUT_RDWR);
 857
 858         mutex_lock(&transport->recv_mutex);
 859         write_lock_bh(&sk->sk_callback_lock);
 860         transport->inet = NULL;
 861         transport->sock = NULL;
 862
 863         sk->sk_user_data = NULL;
 864
 865         xs_restore_old_callbacks(transport, sk);
 866         xprt_clear_connected(xprt);
 867         write_unlock_bh(&sk->sk_callback_lock);
 868         xs_sock_reset_connection_flags(xprt);
 869         mutex_unlock(&transport->recv_mutex);
 870
 871         trace_rpc_socket_close(xprt, sock);
 872         sock_release(sock);
 873 }
 874
 875 /**
 876  * xs_close - close a socket
 877  * @xprt: transport
 878  *
 879  * This is used when all requests are complete; ie, no DRC state remains
 880  * on the server we want to save.
 881  *
 882  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 883  * xs_reset_transport() zeroing the socket from underneath a writer.
 884  */
 885 static void xs_close(struct rpc_xprt *xprt)
 886 {
 887         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 888
 889         dprintk("RPC:       xs_close xprt %p\n", xprt);
 890
 891         xs_reset_transport(transport);
 892         xprt->reestablish_timeout = 0;
 893
 894         xprt_disconnect_done(xprt);
 895 }
 896
 897 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 898 {
 899         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 900                 xprt);
 901         xprt_disconnect_done(xprt);
 902 }
 903
 904 static void xs_xprt_free(struct rpc_xprt *xprt)
 905 {
 906         xs_free_peer_addresses(xprt);
 907         xprt_free(xprt);
 908 }
 909
 910 /**
 911  * xs_destroy - prepare to shutdown a transport
 912  * @xprt: doomed transport
 913  *
 914  */
 915 static void xs_destroy(struct rpc_xprt *xprt)
 916 {
 917         struct sock_xprt *transport = container_of(xprt,
 918                         struct sock_xprt, xprt);
 919         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 920
 921         cancel_delayed_work_sync(&transport->connect_worker);
 922         xs_close(xprt);
 923         cancel_work_sync(&transport->recv_worker);
 924         xs_xprt_free(xprt);
 925         module_put(THIS_MODULE);
 926 }
 927
 928 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 929 {
 930         struct xdr_skb_reader desc = {
 931                 .skb            = skb,
 932                 .offset         = sizeof(rpc_fraghdr),
 933                 .count          = skb->len - sizeof(rpc_fraghdr),
 934         };
 935
 936         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 937                 return -1;
 938         if (desc.count)
 939                 return -1;
 940         return 0;
 941 }
 942
 943 /**
 944  * xs_local_data_read_skb
 945  * @xprt: transport
 946  * @sk: socket
 947  * @skb: skbuff
 948  *
 949  * Currently this assumes we can read the whole reply in a single gulp.
 950  */
 951 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 952                 struct sock *sk,
 953                 struct sk_buff *skb)
 954 {
 955         struct rpc_task *task;
 956         struct rpc_rqst *rovr;
 957         int repsize, copied;
 958         u32 _xid;
 959         __be32 *xp;
 960
 961         repsize = skb->len - sizeof(rpc_fraghdr);
 962         if (repsize < 4) {
 963                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 964                 return;
 965         }
 966
 967         /* Copy the XID from the skb... */
 968         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 969         if (xp == NULL)
 970                 return;
 971
 972         /* Look up and lock the request corresponding to the given XID */
 973         spin_lock(&xprt->recv_lock);
 974         rovr = xprt_lookup_rqst(xprt, *xp);
 975         if (!rovr)
 976                 goto out_unlock;
 977         xprt_pin_rqst(rovr);
 978         spin_unlock(&xprt->recv_lock);
 979         task = rovr->rq_task;
 980
 981         copied = rovr->rq_private_buf.buflen;
 982         if (copied > repsize)
 983                 copied = repsize;
 984
 985         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 986                 dprintk("RPC:       sk_buff copy failed\n");
 987                 spin_lock(&xprt->recv_lock);
 988                 goto out_unpin;
 989         }
 990
 991         spin_lock(&xprt->recv_lock);
 992         xprt_complete_rqst(task, copied);
 993 out_unpin:
 994         xprt_unpin_rqst(rovr);
 995  out_unlock:
 996         spin_unlock(&xprt->recv_lock);
 997 }
 998
 999 static void xs_local_data_receive(struct sock_xprt *transport)
1000 {
1001         struct sk_buff *skb;
1002         struct sock *sk;
1003         int err;
1004
1005         mutex_lock(&transport->recv_mutex);
1006         sk = transport->inet;
1007         if (sk == NULL)
1008                 goto out;
1009         for (;;) {
1010                 skb = skb_recv_datagram(sk, 0, 1, &err);
1011                 if (skb != NULL) {
1012                         xs_local_data_read_skb(&transport->xprt, sk, skb);
1013                         skb_free_datagram(sk, skb);
1014                         continue;
1015                 }
1016                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1017                         break;
1018         }
1019 out:
1020         mutex_unlock(&transport->recv_mutex);
1021 }
1022
1023 static void xs_local_data_receive_workfn(struct work_struct *work)
1024 {
1025         struct sock_xprt *transport =
1026                 container_of(work, struct sock_xprt, recv_worker);
1027         xs_local_data_receive(transport);
1028 }
1029
1030 /**
1031  * xs_udp_data_read_skb - receive callback for UDP sockets
1032  * @xprt: transport
1033  * @sk: socket
1034  * @skb: skbuff
1035  *
1036  */
1037 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1038                 struct sock *sk,
1039                 struct sk_buff *skb)
1040 {
1041         struct rpc_task *task;
1042         struct rpc_rqst *rovr;
1043         int repsize, copied;
1044         u32 _xid;
1045         __be32 *xp;
1046
1047         repsize = skb->len;
1048         if (repsize < 4) {
1049                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1050                 return;
1051         }
1052
1053         /* Copy the XID from the skb... */
1054         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1055         if (xp == NULL)
1056                 return;
1057
1058         /* Look up and lock the request corresponding to the given XID */
1059         spin_lock(&xprt->recv_lock);
1060         rovr = xprt_lookup_rqst(xprt, *xp);
1061         if (!rovr)
1062                 goto out_unlock;
1063         xprt_pin_rqst(rovr);
1064         spin_unlock(&xprt->recv_lock);
1065         task = rovr->rq_task;
1066
1067         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1068                 copied = repsize;
1069
1070         /* Suck it into the iovec, verify checksum if not done by hw. */
1071         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1072                 spin_lock(&xprt->recv_lock);
1073                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1074                 goto out_unpin;
1075         }
1076
1077
1078         spin_lock_bh(&xprt->transport_lock);
1079         xprt_adjust_cwnd(xprt, task, copied);
1080         spin_unlock_bh(&xprt->transport_lock);
1081         spin_lock(&xprt->recv_lock);
1082         xprt_complete_rqst(task, copied);
1083         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1084 out_unpin:
1085         xprt_unpin_rqst(rovr);
1086  out_unlock:
1087         spin_unlock(&xprt->recv_lock);
1088 }
1089
1090 static void xs_udp_data_receive(struct sock_xprt *transport)
1091 {
1092         struct sk_buff *skb;
1093         struct sock *sk;
1094         int err;
1095
1096         mutex_lock(&transport->recv_mutex);
1097         sk = transport->inet;
1098         if (sk == NULL)
1099                 goto out;
1100         for (;;) {
1101                 skb = skb_recv_udp(sk, 0, 1, &err);
1102                 if (skb != NULL) {
1103                         xs_udp_data_read_skb(&transport->xprt, sk, skb);
1104                         consume_skb(skb);
1105                         continue;
1106                 }
1107                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1108                         break;
1109         }
1110 out:
1111         mutex_unlock(&transport->recv_mutex);
1112 }
1113
1114 static void xs_udp_data_receive_workfn(struct work_struct *work)
1115 {
1116         struct sock_xprt *transport =
1117                 container_of(work, struct sock_xprt, recv_worker);
1118         xs_udp_data_receive(transport);
1119 }
1120
1121 /**
1122  * xs_data_ready - "data ready" callback for UDP sockets
1123  * @sk: socket with data to read
1124  *
1125  */
1126 static void xs_data_ready(struct sock *sk)
1127 {
1128         struct rpc_xprt *xprt;
1129
1130         read_lock_bh(&sk->sk_callback_lock);
1131         dprintk("RPC:       xs_data_ready...\n");
1132         xprt = xprt_from_sock(sk);
1133         if (xprt != NULL) {
1134                 struct sock_xprt *transport = container_of(xprt,
1135                                 struct sock_xprt, xprt);
1136                 transport->old_data_ready(sk);
1137                 /* Any data means we had a useful conversation, so
1138                  * then we don't need to delay the next reconnect
1139                  */
1140                 if (xprt->reestablish_timeout)
1141                         xprt->reestablish_timeout = 0;
1142                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1143                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1144         }
1145         read_unlock_bh(&sk->sk_callback_lock);
1146 }
1147
1148 /*
1149  * Helper function to force a TCP close if the server is sending
1150  * junk and/or it has put us in CLOSE_WAIT
1151  */
1152 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1153 {
1154         xprt_force_disconnect(xprt);
1155 }
1156
1157 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1158 {
1159         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1160         size_t len, used;
1161         char *p;
1162
1163         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1164         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1165         used = xdr_skb_read_bits(desc, p, len);
1166         transport->tcp_offset += used;
1167         if (used != len)
1168                 return;
1169
1170         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1171         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1172                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1173         else
1174                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1175         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1176
1177         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1178         transport->tcp_offset = 0;
1179
1180         /* Sanity check of the record length */
1181         if (unlikely(transport->tcp_reclen < 8)) {
1182                 dprintk("RPC:       invalid TCP record fragment length\n");
1183                 xs_tcp_force_close(xprt);
1184                 return;
1185         }
1186         dprintk("RPC:       reading TCP record fragment of length %d\n",
1187                         transport->tcp_reclen);
1188 }
1189
1190 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1191 {
1192         if (transport->tcp_offset == transport->tcp_reclen) {
1193                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1194                 transport->tcp_offset = 0;
1195                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1196                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1197                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1198                         transport->tcp_copied = 0;
1199                 }
1200         }
1201 }
1202
1203 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1204 {
1205         size_t len, used;
1206         char *p;
1207
1208         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1209         dprintk("RPC:       reading XID (%zu bytes)\n", len);
1210         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1211         used = xdr_skb_read_bits(desc, p, len);
1212         transport->tcp_offset += used;
1213         if (used != len)
1214                 return;
1215         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1216         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1217         transport->tcp_copied = 4;
1218         dprintk("RPC:       reading %s XID %08x\n",
1219                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1220                                                               : "request with",
1221                         ntohl(transport->tcp_xid));
1222         xs_tcp_check_fraghdr(transport);
1223 }
1224
1225 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1226                                        struct xdr_skb_reader *desc)
1227 {
1228         size_t len, used;
1229         u32 offset;
1230         char *p;
1231
1232         /*
1233          * We want transport->tcp_offset to be 8 at the end of this routine
1234          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1235          * When this function is called for the first time,
1236          * transport->tcp_offset is 4 (after having already read the xid).
1237          */
1238         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1239         len = sizeof(transport->tcp_calldir) - offset;
1240         dprintk("RPC:       reading CALL/REPLY flag (%zu bytes)\n", len);
1241         p = ((char *) &transport->tcp_calldir) + offset;
1242         used = xdr_skb_read_bits(desc, p, len);
1243         transport->tcp_offset += used;
1244         if (used != len)
1245                 return;
1246         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1247         /*
1248          * We don't yet have the XDR buffer, so we will write the calldir
1249          * out after we get the buffer from the 'struct rpc_rqst'
1250          */
1251         switch (ntohl(transport->tcp_calldir)) {
1252         case RPC_REPLY:
1253                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1254                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1255                 transport->tcp_flags |= TCP_RPC_REPLY;
1256                 break;
1257         case RPC_CALL:
1258                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1259                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1260                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1261                 break;
1262         default:
1263                 dprintk("RPC:       invalid request message type\n");
1264                 xs_tcp_force_close(&transport->xprt);
1265         }
1266         xs_tcp_check_fraghdr(transport);
1267 }
1268
1269 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1270                                      struct xdr_skb_reader *desc,
1271                                      struct rpc_rqst *req)
1272 {
1273         struct sock_xprt *transport =
1274                                 container_of(xprt, struct sock_xprt, xprt);
1275         struct xdr_buf *rcvbuf;
1276         size_t len;
1277         ssize_t r;
1278
1279         rcvbuf = &req->rq_private_buf;
1280
1281         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1282                 /*
1283                  * Save the RPC direction in the XDR buffer
1284                  */
1285                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1286                         &transport->tcp_calldir,
1287                         sizeof(transport->tcp_calldir));
1288                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1289                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1290         }
1291
1292         len = desc->count;
1293         if (len > transport->tcp_reclen - transport->tcp_offset)
1294                 desc->count = transport->tcp_reclen - transport->tcp_offset;
1295         r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1296                                           desc, xdr_skb_read_bits);
1297
1298         if (desc->count) {
1299                 /* Error when copying to the receive buffer,
1300                  * usually because we weren't able to allocate
1301                  * additional buffer pages. All we can do now
1302                  * is turn off TCP_RCV_COPY_DATA, so the request
1303                  * will not receive any additional updates,
1304                  * and time out.
1305                  * Any remaining data from this record will
1306                  * be discarded.
1307                  */
1308                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1309                 dprintk("RPC:       XID %08x truncated request\n",
1310                                 ntohl(transport->tcp_xid));
1311                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1312                                 "tcp_offset = %u, tcp_reclen = %u\n",
1313                                 xprt, transport->tcp_copied,
1314                                 transport->tcp_offset, transport->tcp_reclen);
1315                 return;
1316         }
1317
1318         transport->tcp_copied += r;
1319         transport->tcp_offset += r;
1320         desc->count = len - r;
1321
1322         dprintk("RPC:       XID %08x read %zd bytes\n",
1323                         ntohl(transport->tcp_xid), r);
1324         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1325                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1326                         transport->tcp_offset, transport->tcp_reclen);
1327
1328         if (transport->tcp_copied == req->rq_private_buf.buflen)
1329                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1330         else if (transport->tcp_offset == transport->tcp_reclen) {
1331                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1332                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1333         }
1334 }
1335
1336 /*
1337  * Finds the request corresponding to the RPC xid and invokes the common
1338  * tcp read code to read the data.
1339  */
1340 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1341                                     struct xdr_skb_reader *desc)
1342 {
1343         struct sock_xprt *transport =
1344                                 container_of(xprt, struct sock_xprt, xprt);
1345         struct rpc_rqst *req;
1346
1347         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1348
1349         /* Find and lock the request corresponding to this xid */
1350         spin_lock(&xprt->recv_lock);
1351         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1352         if (!req) {
1353                 dprintk("RPC:       XID %08x request not found!\n",
1354                                 ntohl(transport->tcp_xid));
1355                 spin_unlock(&xprt->recv_lock);
1356                 return -1;
1357         }
1358         xprt_pin_rqst(req);
1359         spin_unlock(&xprt->recv_lock);
1360
1361         xs_tcp_read_common(xprt, desc, req);
1362
1363         spin_lock(&xprt->recv_lock);
1364         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1365                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1366         xprt_unpin_rqst(req);
1367         spin_unlock(&xprt->recv_lock);
1368         return 0;
1369 }
1370
1371 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1372 /*
1373  * Obtains an rpc_rqst previously allocated and invokes the common
1374  * tcp read code to read the data.  The result is placed in the callback
1375  * queue.
1376  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1377  * connection and return -1.
1378  */
1379 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1380                                        struct xdr_skb_reader *desc)
1381 {
1382         struct sock_xprt *transport =
1383                                 container_of(xprt, struct sock_xprt, xprt);
1384         struct rpc_rqst *req;
1385
1386         /* Look up the request corresponding to the given XID */
1387         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1388         if (req == NULL) {
1389                 printk(KERN_WARNING "Callback slot table overflowed\n");
1390                 xprt_force_disconnect(xprt);
1391                 return -1;
1392         }
1393
1394         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1395         xs_tcp_read_common(xprt, desc, req);
1396
1397         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1398                 xprt_complete_bc_request(req, transport->tcp_copied);
1399
1400         return 0;
1401 }
1402
1403 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1404                                         struct xdr_skb_reader *desc)
1405 {
1406         struct sock_xprt *transport =
1407                                 container_of(xprt, struct sock_xprt, xprt);
1408
1409         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1410                 xs_tcp_read_reply(xprt, desc) :
1411                 xs_tcp_read_callback(xprt, desc);
1412 }
1413
1414 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1415 {
1416         int ret;
1417
1418         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1419                               SVC_SOCK_ANONYMOUS);
1420         if (ret < 0)
1421                 return ret;
1422         return 0;
1423 }
1424
1425 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1426 {
1427         return PAGE_SIZE;
1428 }
1429 #else
1430 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1431                                         struct xdr_skb_reader *desc)
1432 {
1433         return xs_tcp_read_reply(xprt, desc);
1434 }
1435 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1436
1437 /*
1438  * Read data off the transport.  This can be either an RPC_CALL or an
1439  * RPC_REPLY.  Relay the processing to helper functions.
1440  */
1441 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1442                                     struct xdr_skb_reader *desc)
1443 {
1444         struct sock_xprt *transport =
1445                                 container_of(xprt, struct sock_xprt, xprt);
1446
1447         if (_xs_tcp_read_data(xprt, desc) == 0)
1448                 xs_tcp_check_fraghdr(transport);
1449         else {
1450                 /*
1451                  * The transport_lock protects the request handling.
1452                  * There's no need to hold it to update the tcp_flags.
1453                  */
1454                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1455         }
1456 }
1457
1458 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1459 {
1460         size_t len;
1461
1462         len = transport->tcp_reclen - transport->tcp_offset;
1463         if (len > desc->count)
1464                 len = desc->count;
1465         desc->count -= len;
1466         desc->offset += len;
1467         transport->tcp_offset += len;
1468         dprintk("RPC:       discarded %zu bytes\n", len);
1469         xs_tcp_check_fraghdr(transport);
1470 }
1471
1472 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1473 {
1474         struct rpc_xprt *xprt = rd_desc->arg.data;
1475         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1476         struct xdr_skb_reader desc = {
1477                 .skb    = skb,
1478                 .offset = offset,
1479                 .count  = len,
1480         };
1481
1482         dprintk("RPC:       xs_tcp_data_recv started\n");
1483         do {
1484                 trace_xs_tcp_data_recv(transport);
1485                 /* Read in a new fragment marker if necessary */
1486                 /* Can we ever really expect to get completely empty fragments? */
1487                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1488                         xs_tcp_read_fraghdr(xprt, &desc);
1489                         continue;
1490                 }
1491                 /* Read in the xid if necessary */
1492                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1493                         xs_tcp_read_xid(transport, &desc);
1494                         continue;
1495                 }
1496                 /* Read in the call/reply flag */
1497                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1498                         xs_tcp_read_calldir(transport, &desc);
1499                         continue;
1500                 }
1501                 /* Read in the request data */
1502                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1503                         xs_tcp_read_data(xprt, &desc);
1504                         continue;
1505                 }
1506                 /* Skip over any trailing bytes on short reads */
1507                 xs_tcp_read_discard(transport, &desc);
1508         } while (desc.count);
1509         trace_xs_tcp_data_recv(transport);
1510         dprintk("RPC:       xs_tcp_data_recv done\n");
1511         return len - desc.count;
1512 }
1513
1514 static void xs_tcp_data_receive(struct sock_xprt *transport)
1515 {
1516         struct rpc_xprt *xprt = &transport->xprt;
1517         struct sock *sk;
1518         read_descriptor_t rd_desc = {
1519                 .count = 2*1024*1024,
1520                 .arg.data = xprt,
1521         };
1522         unsigned long total = 0;
1523         int loop;
1524         int read = 0;
1525
1526         mutex_lock(&transport->recv_mutex);
1527         sk = transport->inet;
1528         if (sk == NULL)
1529                 goto out;
1530
1531         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1532         for (loop = 0; loop < 64; loop++) {
1533                 lock_sock(sk);
1534                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1535                 if (read <= 0) {
1536                         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1537                         release_sock(sk);
1538                         break;
1539                 }
1540                 release_sock(sk);
1541                 total += read;
1542                 rd_desc.count = 65536;
1543         }
1544         if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1545                 queue_work(xprtiod_workqueue, &transport->recv_worker);
1546 out:
1547         mutex_unlock(&transport->recv_mutex);
1548         trace_xs_tcp_data_ready(xprt, read, total);
1549 }
1550
1551 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1552 {
1553         struct sock_xprt *transport =
1554                 container_of(work, struct sock_xprt, recv_worker);
1555         xs_tcp_data_receive(transport);
1556 }
1557
1558 /**
1559  * xs_tcp_state_change - callback to handle TCP socket state changes
1560  * @sk: socket whose state has changed
1561  *
1562  */
1563 static void xs_tcp_state_change(struct sock *sk)
1564 {
1565         struct rpc_xprt *xprt;
1566         struct sock_xprt *transport;
1567
1568         read_lock_bh(&sk->sk_callback_lock);
1569         if (!(xprt = xprt_from_sock(sk)))
1570                 goto out;
1571         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1572         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1573                         sk->sk_state, xprt_connected(xprt),
1574                         sock_flag(sk, SOCK_DEAD),
1575                         sock_flag(sk, SOCK_ZAPPED),
1576                         sk->sk_shutdown);
1577
1578         transport = container_of(xprt, struct sock_xprt, xprt);
1579         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1580         switch (sk->sk_state) {
1581         case TCP_ESTABLISHED:
1582                 spin_lock(&xprt->transport_lock);
1583                 if (!xprt_test_and_set_connected(xprt)) {
1584
1585                         /* Reset TCP record info */
1586                         transport->tcp_offset = 0;
1587                         transport->tcp_reclen = 0;
1588                         transport->tcp_copied = 0;
1589                         transport->tcp_flags =
1590                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1591                         xprt->connect_cookie++;
1592                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1593                         xprt_clear_connecting(xprt);
1594
1595                         xprt->stat.connect_count++;
1596                         xprt->stat.connect_time += (long)jiffies -
1597                                                    xprt->stat.connect_start;
1598                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1599                 }
1600                 spin_unlock(&xprt->transport_lock);
1601                 break;
1602         case TCP_FIN_WAIT1:
1603                 /* The client initiated a shutdown of the socket */
1604                 xprt->connect_cookie++;
1605                 xprt->reestablish_timeout = 0;
1606                 set_bit(XPRT_CLOSING, &xprt->state);
1607                 smp_mb__before_atomic();
1608                 clear_bit(XPRT_CONNECTED, &xprt->state);
1609                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1610                 smp_mb__after_atomic();
1611                 break;
1612         case TCP_CLOSE_WAIT:
1613                 /* The server initiated a shutdown of the socket */
1614                 xprt->connect_cookie++;
1615                 clear_bit(XPRT_CONNECTED, &xprt->state);
1616                 xs_tcp_force_close(xprt);
1617         case TCP_CLOSING:
1618                 /*
1619                  * If the server closed down the connection, make sure that
1620                  * we back off before reconnecting
1621                  */
1622                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1623                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1624                 break;
1625         case TCP_LAST_ACK:
1626                 set_bit(XPRT_CLOSING, &xprt->state);
1627                 smp_mb__before_atomic();
1628                 clear_bit(XPRT_CONNECTED, &xprt->state);
1629                 smp_mb__after_atomic();
1630                 break;
1631         case TCP_CLOSE:
1632                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1633                                         &transport->sock_state))
1634                         xprt_clear_connecting(xprt);
1635                 if (sk->sk_err)
1636                         xprt_wake_pending_tasks(xprt, -sk->sk_err);
1637                 xs_sock_mark_closed(xprt);
1638         }
1639  out:
1640         read_unlock_bh(&sk->sk_callback_lock);
1641 }
1642
1643 static void xs_write_space(struct sock *sk)
1644 {
1645         struct socket_wq *wq;
1646         struct rpc_xprt *xprt;
1647
1648         if (!sk->sk_socket)
1649                 return;
1650         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1651
1652         if (unlikely(!(xprt = xprt_from_sock(sk))))
1653                 return;
1654         rcu_read_lock();
1655         wq = rcu_dereference(sk->sk_wq);
1656         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1657                 goto out;
1658
1659         xprt_write_space(xprt);
1660 out:
1661         rcu_read_unlock();
1662 }
1663
1664 /**
1665  * xs_udp_write_space - callback invoked when socket buffer space
1666  *                             becomes available
1667  * @sk: socket whose state has changed
1668  *
1669  * Called when more output buffer space is available for this socket.
1670  * We try not to wake our writers until they can make "significant"
1671  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1672  * with a bunch of small requests.
1673  */
1674 static void xs_udp_write_space(struct sock *sk)
1675 {
1676         read_lock_bh(&sk->sk_callback_lock);
1677
1678         /* from net/core/sock.c:sock_def_write_space */
1679         if (sock_writeable(sk))
1680                 xs_write_space(sk);
1681
1682         read_unlock_bh(&sk->sk_callback_lock);
1683 }
1684
1685 /**
1686  * xs_tcp_write_space - callback invoked when socket buffer space
1687  *                             becomes available
1688  * @sk: socket whose state has changed
1689  *
1690  * Called when more output buffer space is available for this socket.
1691  * We try not to wake our writers until they can make "significant"
1692  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1693  * with a bunch of small requests.
1694  */
1695 static void xs_tcp_write_space(struct sock *sk)
1696 {
1697         read_lock_bh(&sk->sk_callback_lock);
1698
1699         /* from net/core/stream.c:sk_stream_write_space */
1700         if (sk_stream_is_writeable(sk))
1701                 xs_write_space(sk);
1702
1703         read_unlock_bh(&sk->sk_callback_lock);
1704 }
1705
1706 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1707 {
1708         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1709         struct sock *sk = transport->inet;
1710
1711         if (transport->rcvsize) {
1712                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1713                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1714         }
1715         if (transport->sndsize) {
1716                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1717                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1718                 sk->sk_write_space(sk);
1719         }
1720 }
1721
1722 /**
1723  * xs_udp_set_buffer_size - set send and receive limits
1724  * @xprt: generic transport
1725  * @sndsize: requested size of send buffer, in bytes
1726  * @rcvsize: requested size of receive buffer, in bytes
1727  *
1728  * Set socket send and receive buffer size limits.
1729  */
1730 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1731 {
1732         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1733
1734         transport->sndsize = 0;
1735         if (sndsize)
1736                 transport->sndsize = sndsize + 1024;
1737         transport->rcvsize = 0;
1738         if (rcvsize)
1739                 transport->rcvsize = rcvsize + 1024;
1740
1741         xs_udp_do_set_buffer_size(xprt);
1742 }
1743
1744 /**
1745  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1746  * @task: task that timed out
1747  *
1748  * Adjust the congestion window after a retransmit timeout has occurred.
1749  */
1750 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1751 {
1752         spin_lock_bh(&xprt->transport_lock);
1753         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1754         spin_unlock_bh(&xprt->transport_lock);
1755 }
1756
1757 static int xs_get_random_port(void)
1758 {
1759         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1760         unsigned short range;
1761         unsigned short rand;
1762
1763         if (max < min)
1764                 return -EADDRINUSE;
1765         range = max - min + 1;
1766         rand = (unsigned short) prandom_u32() % range;
1767         return rand + min;
1768 }
1769
1770 /**
1771  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1772  * @sock: socket
1773  *
1774  * Note that this function has to be called on all sockets that share the
1775  * same port, and it must be called before binding.
1776  */
1777 static void xs_sock_set_reuseport(struct socket *sock)
1778 {
1779         int opt = 1;
1780
1781         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1782                         (char *)&opt, sizeof(opt));
1783 }
1784
1785 static unsigned short xs_sock_getport(struct socket *sock)
1786 {
1787         struct sockaddr_storage buf;
1788         int buflen;
1789         unsigned short port = 0;
1790
1791         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1792                 goto out;
1793         switch (buf.ss_family) {
1794         case AF_INET6:
1795                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1796                 break;
1797         case AF_INET:
1798                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1799         }
1800 out:
1801         return port;
1802 }
1803
1804 /**
1805  * xs_set_port - reset the port number in the remote endpoint address
1806  * @xprt: generic transport
1807  * @port: new port number
1808  *
1809  */
1810 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1811 {
1812         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1813
1814         rpc_set_port(xs_addr(xprt), port);
1815         xs_update_peer_port(xprt);
1816 }
1817
1818 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1819 {
1820         if (transport->srcport == 0)
1821                 transport->srcport = xs_sock_getport(sock);
1822 }
1823
1824 static int xs_get_srcport(struct sock_xprt *transport)
1825 {
1826         int port = transport->srcport;
1827
1828         if (port == 0 && transport->xprt.resvport)
1829                 port = xs_get_random_port();
1830         return port;
1831 }
1832
1833 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1834 {
1835         if (transport->srcport != 0)
1836                 transport->srcport = 0;
1837         if (!transport->xprt.resvport)
1838                 return 0;
1839         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1840                 return xprt_max_resvport;
1841         return --port;
1842 }
1843 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1844 {
1845         struct sockaddr_storage myaddr;
1846         int err, nloop = 0;
1847         int port = xs_get_srcport(transport);
1848         unsigned short last;
1849
1850         /*
1851          * If we are asking for any ephemeral port (i.e. port == 0 &&
1852          * transport->xprt.resvport == 0), don't bind.  Let the local
1853          * port selection happen implicitly when the socket is used
1854          * (for example at connect time).
1855          *
1856          * This ensures that we can continue to establish TCP
1857          * connections even when all local ephemeral ports are already
1858          * a part of some TCP connection.  This makes no difference
1859          * for UDP sockets, but also doens't harm them.
1860          *
1861          * If we're asking for any reserved port (i.e. port == 0 &&
1862          * transport->xprt.resvport == 1) xs_get_srcport above will
1863          * ensure that port is non-zero and we will bind as needed.
1864          */
1865         if (port <= 0)
1866                 return port;
1867
1868         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1869         do {
1870                 rpc_set_port((struct sockaddr *)&myaddr, port);
1871                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1872                                 transport->xprt.addrlen);
1873                 if (err == 0) {
1874                         transport->srcport = port;
1875                         break;
1876                 }
1877                 last = port;
1878                 port = xs_next_srcport(transport, port);
1879                 if (port > last)
1880                         nloop++;
1881         } while (err == -EADDRINUSE && nloop != 2);
1882
1883         if (myaddr.ss_family == AF_INET)
1884                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1885                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1886                                 port, err ? "failed" : "ok", err);
1887         else
1888                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1889                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1890                                 port, err ? "failed" : "ok", err);
1891         return err;
1892 }
1893
1894 /*
1895  * We don't support autobind on AF_LOCAL sockets
1896  */
1897 static void xs_local_rpcbind(struct rpc_task *task)
1898 {
1899         xprt_set_bound(task->tk_xprt);
1900 }
1901
1902 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1903 {
1904 }
1905
1906 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1907 static struct lock_class_key xs_key[2];
1908 static struct lock_class_key xs_slock_key[2];
1909
1910 static inline void xs_reclassify_socketu(struct socket *sock)
1911 {
1912         struct sock *sk = sock->sk;
1913
1914         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1915                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1916 }
1917
1918 static inline void xs_reclassify_socket4(struct socket *sock)
1919 {
1920         struct sock *sk = sock->sk;
1921
1922         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1923                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1924 }
1925
1926 static inline void xs_reclassify_socket6(struct socket *sock)
1927 {
1928         struct sock *sk = sock->sk;
1929
1930         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1931                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1932 }
1933
1934 static inline void xs_reclassify_socket(int family, struct socket *sock)
1935 {
1936         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1937                 return;
1938
1939         switch (family) {
1940         case AF_LOCAL:
1941                 xs_reclassify_socketu(sock);
1942                 break;
1943         case AF_INET:
1944                 xs_reclassify_socket4(sock);
1945                 break;
1946         case AF_INET6:
1947                 xs_reclassify_socket6(sock);
1948                 break;
1949         }
1950 }
1951 #else
1952 static inline void xs_reclassify_socket(int family, struct socket *sock)
1953 {
1954 }
1955 #endif
1956
1957 static void xs_dummy_setup_socket(struct work_struct *work)
1958 {
1959 }
1960
1961 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1962                 struct sock_xprt *transport, int family, int type,
1963                 int protocol, bool reuseport)
1964 {
1965         struct socket *sock;
1966         int err;
1967
1968         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1969         if (err < 0) {
1970                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1971                                 protocol, -err);
1972                 goto out;
1973         }
1974         xs_reclassify_socket(family, sock);
1975
1976         if (reuseport)
1977                 xs_sock_set_reuseport(sock);
1978
1979         err = xs_bind(transport, sock);
1980         if (err) {
1981                 sock_release(sock);
1982                 goto out;
1983         }
1984
1985         return sock;
1986 out:
1987         return ERR_PTR(err);
1988 }
1989
1990 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1991                                       struct socket *sock)
1992 {
1993         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1994                                                                         xprt);
1995
1996         if (!transport->inet) {
1997                 struct sock *sk = sock->sk;
1998
1999                 write_lock_bh(&sk->sk_callback_lock);
2000
2001                 xs_save_old_callbacks(transport, sk);
2002
2003                 sk->sk_user_data = xprt;
2004                 sk->sk_data_ready = xs_data_ready;
2005                 sk->sk_write_space = xs_udp_write_space;
2006                 sock_set_flag(sk, SOCK_FASYNC);
2007                 sk->sk_error_report = xs_error_report;
2008                 sk->sk_allocation = GFP_NOIO;
2009
2010                 xprt_clear_connected(xprt);
2011
2012                 /* Reset to new socket */
2013                 transport->sock = sock;
2014                 transport->inet = sk;
2015
2016                 write_unlock_bh(&sk->sk_callback_lock);
2017         }
2018
2019         /* Tell the socket layer to start connecting... */
2020         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2021 }
2022
2023 /**
2024  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2025  * @transport: socket transport to connect
2026  */
2027 static int xs_local_setup_socket(struct sock_xprt *transport)
2028 {
2029         struct rpc_xprt *xprt = &transport->xprt;
2030         struct socket *sock;
2031         int status = -EIO;
2032
2033         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2034                                         SOCK_STREAM, 0, &sock, 1);
2035         if (status < 0) {
2036                 dprintk("RPC:       can't create AF_LOCAL "
2037                         "transport socket (%d).\n", -status);
2038                 goto out;
2039         }
2040         xs_reclassify_socket(AF_LOCAL, sock);
2041
2042         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2043                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2044
2045         status = xs_local_finish_connecting(xprt, sock);
2046         trace_rpc_socket_connect(xprt, sock, status);
2047         switch (status) {
2048         case 0:
2049                 dprintk("RPC:       xprt %p connected to %s\n",
2050                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2051                 xprt->stat.connect_count++;
2052                 xprt->stat.connect_time += (long)jiffies -
2053                                            xprt->stat.connect_start;
2054                 xprt_set_connected(xprt);
2055         case -ENOBUFS:
2056                 break;
2057         case -ENOENT:
2058                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2059                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2060                 break;
2061         case -ECONNREFUSED:
2062                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2063                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2064                 break;
2065         default:
2066                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2067                                 __func__, -status,
2068                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2069         }
2070
2071 out:
2072         xprt_clear_connecting(xprt);
2073         xprt_wake_pending_tasks(xprt, status);
2074         return status;
2075 }
2076
2077 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2078 {
2079         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2080         int ret;
2081
2082          if (RPC_IS_ASYNC(task)) {
2083                 /*
2084                  * We want the AF_LOCAL connect to be resolved in the
2085                  * filesystem namespace of the process making the rpc
2086                  * call.  Thus we connect synchronously.
2087                  *
2088                  * If we want to support asynchronous AF_LOCAL calls,
2089                  * we'll need to figure out how to pass a namespace to
2090                  * connect.
2091                  */
2092                 rpc_exit(task, -ENOTCONN);
2093                 return;
2094         }
2095         ret = xs_local_setup_socket(transport);
2096         if (ret && !RPC_IS_SOFTCONN(task))
2097                 msleep_interruptible(15000);
2098 }
2099
2100 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2101 /*
2102  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2103  * know that we have exclusive access to the socket), to guard against
2104  * races with xs_reset_transport.
2105  */
2106 static void xs_set_memalloc(struct rpc_xprt *xprt)
2107 {
2108         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2109                         xprt);
2110
2111         /*
2112          * If there's no sock, then we have nothing to set. The
2113          * reconnecting process will get it for us.
2114          */
2115         if (!transport->inet)
2116                 return;
2117         if (atomic_read(&xprt->swapper))
2118                 sk_set_memalloc(transport->inet);
2119 }
2120
2121 /**
2122  * xs_enable_swap - Tag this transport as being used for swap.
2123  * @xprt: transport to tag
2124  *
2125  * Take a reference to this transport on behalf of the rpc_clnt, and
2126  * optionally mark it for swapping if it wasn't already.
2127  */
2128 static int
2129 xs_enable_swap(struct rpc_xprt *xprt)
2130 {
2131         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2132
2133         if (atomic_inc_return(&xprt->swapper) != 1)
2134                 return 0;
2135         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2136                 return -ERESTARTSYS;
2137         if (xs->inet)
2138                 sk_set_memalloc(xs->inet);
2139         xprt_release_xprt(xprt, NULL);
2140         return 0;
2141 }
2142
2143 /**
2144  * xs_disable_swap - Untag this transport as being used for swap.
2145  * @xprt: transport to tag
2146  *
2147  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2148  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2149  */
2150 static void
2151 xs_disable_swap(struct rpc_xprt *xprt)
2152 {
2153         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2154
2155         if (!atomic_dec_and_test(&xprt->swapper))
2156                 return;
2157         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2158                 return;
2159         if (xs->inet)
2160                 sk_clear_memalloc(xs->inet);
2161         xprt_release_xprt(xprt, NULL);
2162 }
2163 #else
2164 static void xs_set_memalloc(struct rpc_xprt *xprt)
2165 {
2166 }
2167
2168 static int
2169 xs_enable_swap(struct rpc_xprt *xprt)
2170 {
2171         return -EINVAL;
2172 }
2173
2174 static void
2175 xs_disable_swap(struct rpc_xprt *xprt)
2176 {
2177 }
2178 #endif
2179
2180 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2181 {
2182         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2183
2184         if (!transport->inet) {
2185                 struct sock *sk = sock->sk;
2186
2187                 write_lock_bh(&sk->sk_callback_lock);
2188
2189                 xs_save_old_callbacks(transport, sk);
2190
2191                 sk->sk_user_data = xprt;
2192                 sk->sk_data_ready = xs_data_ready;
2193                 sk->sk_write_space = xs_udp_write_space;
2194                 sock_set_flag(sk, SOCK_FASYNC);
2195                 sk->sk_allocation = GFP_NOIO;
2196
2197                 xprt_set_connected(xprt);
2198
2199                 /* Reset to new socket */
2200                 transport->sock = sock;
2201                 transport->inet = sk;
2202
2203                 xs_set_memalloc(xprt);
2204
2205                 write_unlock_bh(&sk->sk_callback_lock);
2206         }
2207         xs_udp_do_set_buffer_size(xprt);
2208
2209         xprt->stat.connect_start = jiffies;
2210 }
2211
2212 static void xs_udp_setup_socket(struct work_struct *work)
2213 {
2214         struct sock_xprt *transport =
2215                 container_of(work, struct sock_xprt, connect_worker.work);
2216         struct rpc_xprt *xprt = &transport->xprt;
2217         struct socket *sock;
2218         int status = -EIO;
2219
2220         sock = xs_create_sock(xprt, transport,
2221                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2222                         IPPROTO_UDP, false);
2223         if (IS_ERR(sock))
2224                 goto out;
2225
2226         dprintk("RPC:       worker connecting xprt %p via %s to "
2227                                 "%s (port %s)\n", xprt,
2228                         xprt->address_strings[RPC_DISPLAY_PROTO],
2229                         xprt->address_strings[RPC_DISPLAY_ADDR],
2230                         xprt->address_strings[RPC_DISPLAY_PORT]);
2231
2232         xs_udp_finish_connecting(xprt, sock);
2233         trace_rpc_socket_connect(xprt, sock, 0);
2234         status = 0;
2235 out:
2236         xprt_clear_connecting(xprt);
2237         xprt_unlock_connect(xprt, transport);
2238         xprt_wake_pending_tasks(xprt, status);
2239 }
2240
2241 /**
2242  * xs_tcp_shutdown - gracefully shut down a TCP socket
2243  * @xprt: transport
2244  *
2245  * Initiates a graceful shutdown of the TCP socket by calling the
2246  * equivalent of shutdown(SHUT_RDWR);
2247  */
2248 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2249 {
2250         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2251         struct socket *sock = transport->sock;
2252
2253         if (sock == NULL)
2254                 return;
2255         if (xprt_connected(xprt)) {
2256                 kernel_sock_shutdown(sock, SHUT_RDWR);
2257                 trace_rpc_socket_shutdown(xprt, sock);
2258         } else
2259                 xs_reset_transport(transport);
2260 }
2261
2262 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2263                 struct socket *sock)
2264 {
2265         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2266         unsigned int keepidle;
2267         unsigned int keepcnt;
2268         unsigned int opt_on = 1;
2269         unsigned int timeo;
2270
2271         spin_lock_bh(&xprt->transport_lock);
2272         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2273         keepcnt = xprt->timeout->to_retries + 1;
2274         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2275                 (xprt->timeout->to_retries + 1);
2276         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2277         spin_unlock_bh(&xprt->transport_lock);
2278
2279         /* TCP Keepalive options */
2280         kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2281                         (char *)&opt_on, sizeof(opt_on));
2282         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2283                         (char *)&keepidle, sizeof(keepidle));
2284         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2285                         (char *)&keepidle, sizeof(keepidle));
2286         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2287                         (char *)&keepcnt, sizeof(keepcnt));
2288
2289         /* TCP user timeout (see RFC5482) */
2290         kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2291                         (char *)&timeo, sizeof(timeo));
2292 }
2293
2294 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2295                 unsigned long connect_timeout,
2296                 unsigned long reconnect_timeout)
2297 {
2298         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2299         struct rpc_timeout to;
2300         unsigned long initval;
2301
2302         spin_lock_bh(&xprt->transport_lock);
2303         if (reconnect_timeout < xprt->max_reconnect_timeout)
2304                 xprt->max_reconnect_timeout = reconnect_timeout;
2305         if (connect_timeout < xprt->connect_timeout) {
2306                 memcpy(&to, xprt->timeout, sizeof(to));
2307                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2308                 /* Arbitrary lower limit */
2309                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2310                         initval = XS_TCP_INIT_REEST_TO << 1;
2311                 to.to_initval = initval;
2312                 to.to_maxval = initval;
2313                 memcpy(&transport->tcp_timeout, &to,
2314                                 sizeof(transport->tcp_timeout));
2315                 xprt->timeout = &transport->tcp_timeout;
2316                 xprt->connect_timeout = connect_timeout;
2317         }
2318         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2319         spin_unlock_bh(&xprt->transport_lock);
2320 }
2321
2322 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2323 {
2324         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2325         int ret = -ENOTCONN;
2326
2327         if (!transport->inet) {
2328                 struct sock *sk = sock->sk;
2329                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2330
2331                 /* Avoid temporary address, they are bad for long-lived
2332                  * connections such as NFS mounts.
2333                  * RFC4941, section 3.6 suggests that:
2334                  *    Individual applications, which have specific
2335                  *    knowledge about the normal duration of connections,
2336                  *    MAY override this as appropriate.
2337                  */
2338                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2339                                 (char *)&addr_pref, sizeof(addr_pref));
2340
2341                 xs_tcp_set_socket_timeouts(xprt, sock);
2342
2343                 write_lock_bh(&sk->sk_callback_lock);
2344
2345                 xs_save_old_callbacks(transport, sk);
2346
2347                 sk->sk_user_data = xprt;
2348                 sk->sk_data_ready = xs_data_ready;
2349                 sk->sk_state_change = xs_tcp_state_change;
2350                 sk->sk_write_space = xs_tcp_write_space;
2351                 sock_set_flag(sk, SOCK_FASYNC);
2352                 sk->sk_error_report = xs_error_report;
2353                 sk->sk_allocation = GFP_NOIO;
2354
2355                 /* socket options */
2356                 sock_reset_flag(sk, SOCK_LINGER);
2357                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2358
2359                 xprt_clear_connected(xprt);
2360
2361                 /* Reset to new socket */
2362                 transport->sock = sock;
2363                 transport->inet = sk;
2364
2365                 write_unlock_bh(&sk->sk_callback_lock);
2366         }
2367
2368         if (!xprt_bound(xprt))
2369                 goto out;
2370
2371         xs_set_memalloc(xprt);
2372
2373         /* Tell the socket layer to start connecting... */
2374         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2375         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2376         switch (ret) {
2377         case 0:
2378                 xs_set_srcport(transport, sock);
2379         case -EINPROGRESS:
2380                 /* SYN_SENT! */
2381                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2382                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2383                 break;
2384         case -EADDRNOTAVAIL:
2385                 /* Source port number is unavailable. Try a new one! */
2386                 transport->srcport = 0;
2387         }
2388 out:
2389         return ret;
2390 }
2391
2392 /**
2393  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2394  *
2395  * Invoked by a work queue tasklet.
2396  */
2397 static void xs_tcp_setup_socket(struct work_struct *work)
2398 {
2399         struct sock_xprt *transport =
2400                 container_of(work, struct sock_xprt, connect_worker.work);
2401         struct socket *sock = transport->sock;
2402         struct rpc_xprt *xprt = &transport->xprt;
2403         int status = -EIO;
2404
2405         if (!sock) {
2406                 sock = xs_create_sock(xprt, transport,
2407                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2408                                 IPPROTO_TCP, true);
2409                 if (IS_ERR(sock)) {
2410                         status = PTR_ERR(sock);
2411                         goto out;
2412                 }
2413         }
2414
2415         dprintk("RPC:       worker connecting xprt %p via %s to "
2416                                 "%s (port %s)\n", xprt,
2417                         xprt->address_strings[RPC_DISPLAY_PROTO],
2418                         xprt->address_strings[RPC_DISPLAY_ADDR],
2419                         xprt->address_strings[RPC_DISPLAY_PORT]);
2420
2421         status = xs_tcp_finish_connecting(xprt, sock);
2422         trace_rpc_socket_connect(xprt, sock, status);
2423         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2424                         xprt, -status, xprt_connected(xprt),
2425                         sock->sk->sk_state);
2426         switch (status) {
2427         default:
2428                 printk("%s: connect returned unhandled error %d\n",
2429                         __func__, status);
2430         case -EADDRNOTAVAIL:
2431                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2432                  * and retry
2433                  */
2434                 xs_tcp_force_close(xprt);
2435                 break;
2436         case 0:
2437         case -EINPROGRESS:
2438         case -EALREADY:
2439                 xprt_unlock_connect(xprt, transport);
2440                 return;
2441         case -EINVAL:
2442                 /* Happens, for instance, if the user specified a link
2443                  * local IPv6 address without a scope-id.
2444                  */
2445         case -ECONNREFUSED:
2446         case -ECONNRESET:
2447         case -ENETUNREACH:
2448         case -EHOSTUNREACH:
2449         case -EADDRINUSE:
2450         case -ENOBUFS:
2451                 /*
2452                  * xs_tcp_force_close() wakes tasks with -EIO.
2453                  * We need to wake them first to ensure the
2454                  * correct error code.
2455                  */
2456                 xprt_wake_pending_tasks(xprt, status);
2457                 xs_tcp_force_close(xprt);
2458                 goto out;
2459         }
2460         status = -EAGAIN;
2461 out:
2462         xprt_clear_connecting(xprt);
2463         xprt_unlock_connect(xprt, transport);
2464         xprt_wake_pending_tasks(xprt, status);
2465 }
2466
2467 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2468 {
2469         unsigned long start, now = jiffies;
2470
2471         start = xprt->stat.connect_start + xprt->reestablish_timeout;
2472         if (time_after(start, now))
2473                 return start - now;
2474         return 0;
2475 }
2476
2477 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2478 {
2479         xprt->reestablish_timeout <<= 1;
2480         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2481                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2482         if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2483                 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2484 }
2485
2486 /**
2487  * xs_connect - connect a socket to a remote endpoint
2488  * @xprt: pointer to transport structure
2489  * @task: address of RPC task that manages state of connect request
2490  *
2491  * TCP: If the remote end dropped the connection, delay reconnecting.
2492  *
2493  * UDP socket connects are synchronous, but we use a work queue anyway
2494  * to guarantee that even unprivileged user processes can set up a
2495  * socket on a privileged port.
2496  *
2497  * If a UDP socket connect fails, the delay behavior here prevents
2498  * retry floods (hard mounts).
2499  */
2500 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2501 {
2502         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2503         unsigned long delay = 0;
2504
2505         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2506
2507         if (transport->sock != NULL) {
2508                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2509                                 "seconds\n",
2510                                 xprt, xprt->reestablish_timeout / HZ);
2511
2512                 /* Start by resetting any existing state */
2513                 xs_reset_transport(transport);
2514
2515                 delay = xs_reconnect_delay(xprt);
2516                 xs_reconnect_backoff(xprt);
2517
2518         } else
2519                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2520
2521         queue_delayed_work(xprtiod_workqueue,
2522                         &transport->connect_worker,
2523                         delay);
2524 }
2525
2526 /**
2527  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2528  * @xprt: rpc_xprt struct containing statistics
2529  * @seq: output file
2530  *
2531  */
2532 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2533 {
2534         long idle_time = 0;
2535
2536         if (xprt_connected(xprt))
2537                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2538
2539         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2540                         "%llu %llu %lu %llu %llu\n",
2541                         xprt->stat.bind_count,
2542                         xprt->stat.connect_count,
2543                         xprt->stat.connect_time,
2544                         idle_time,
2545                         xprt->stat.sends,
2546                         xprt->stat.recvs,
2547                         xprt->stat.bad_xids,
2548                         xprt->stat.req_u,
2549                         xprt->stat.bklog_u,
2550                         xprt->stat.max_slots,
2551                         xprt->stat.sending_u,
2552                         xprt->stat.pending_u);
2553 }
2554
2555 /**
2556  * xs_udp_print_stats - display UDP socket-specifc stats
2557  * @xprt: rpc_xprt struct containing statistics
2558  * @seq: output file
2559  *
2560  */
2561 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2562 {
2563         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2564
2565         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2566                         "%lu %llu %llu\n",
2567                         transport->srcport,
2568                         xprt->stat.bind_count,
2569                         xprt->stat.sends,
2570                         xprt->stat.recvs,
2571                         xprt->stat.bad_xids,
2572                         xprt->stat.req_u,
2573                         xprt->stat.bklog_u,
2574                         xprt->stat.max_slots,
2575                         xprt->stat.sending_u,
2576                         xprt->stat.pending_u);
2577 }
2578
2579 /**
2580  * xs_tcp_print_stats - display TCP socket-specifc stats
2581  * @xprt: rpc_xprt struct containing statistics
2582  * @seq: output file
2583  *
2584  */
2585 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2586 {
2587         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2588         long idle_time = 0;
2589
2590         if (xprt_connected(xprt))
2591                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2592
2593         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2594                         "%llu %llu %lu %llu %llu\n",
2595                         transport->srcport,
2596                         xprt->stat.bind_count,
2597                         xprt->stat.connect_count,
2598                         xprt->stat.connect_time,
2599                         idle_time,
2600                         xprt->stat.sends,
2601                         xprt->stat.recvs,
2602                         xprt->stat.bad_xids,
2603                         xprt->stat.req_u,
2604                         xprt->stat.bklog_u,
2605                         xprt->stat.max_slots,
2606                         xprt->stat.sending_u,
2607                         xprt->stat.pending_u);
2608 }
2609
2610 /*
2611  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2612  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2613  * to use the server side send routines.
2614  */
2615 static int bc_malloc(struct rpc_task *task)
2616 {
2617         struct rpc_rqst *rqst = task->tk_rqstp;
2618         size_t size = rqst->rq_callsize;
2619         struct page *page;
2620         struct rpc_buffer *buf;
2621
2622         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2623                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2624                           size);
2625                 return -EINVAL;
2626         }
2627
2628         page = alloc_page(GFP_KERNEL);
2629         if (!page)
2630                 return -ENOMEM;
2631
2632         buf = page_address(page);
2633         buf->len = PAGE_SIZE;
2634
2635         rqst->rq_buffer = buf->data;
2636         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2637         return 0;
2638 }
2639
2640 /*
2641  * Free the space allocated in the bc_alloc routine
2642  */
2643 static void bc_free(struct rpc_task *task)
2644 {
2645         void *buffer = task->tk_rqstp->rq_buffer;
2646         struct rpc_buffer *buf;
2647
2648         buf = container_of(buffer, struct rpc_buffer, data);
2649         free_page((unsigned long)buf);
2650 }
2651
2652 /*
2653  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2654  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2655  */
2656 static int bc_sendto(struct rpc_rqst *req)
2657 {
2658         int len;
2659         struct xdr_buf *xbufp = &req->rq_snd_buf;
2660         struct rpc_xprt *xprt = req->rq_xprt;
2661         struct sock_xprt *transport =
2662                                 container_of(xprt, struct sock_xprt, xprt);
2663         struct socket *sock = transport->sock;
2664         unsigned long headoff;
2665         unsigned long tailoff;
2666
2667         xs_encode_stream_record_marker(xbufp);
2668
2669         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2670         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2671         len = svc_send_common(sock, xbufp,
2672                               virt_to_page(xbufp->head[0].iov_base), headoff,
2673                               xbufp->tail[0].iov_base, tailoff);
2674
2675         if (len != xbufp->len) {
2676                 printk(KERN_NOTICE "Error sending entire callback!\n");
2677                 len = -EAGAIN;
2678         }
2679
2680         return len;
2681 }
2682
2683 /*
2684  * The send routine. Borrows from svc_send
2685  */
2686 static int bc_send_request(struct rpc_task *task)
2687 {
2688         struct rpc_rqst *req = task->tk_rqstp;
2689         struct svc_xprt *xprt;
2690         int len;
2691
2692         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2693         /*
2694          * Get the server socket associated with this callback xprt
2695          */
2696         xprt = req->rq_xprt->bc_xprt;
2697
2698         /*
2699          * Grab the mutex to serialize data as the connection is shared
2700          * with the fore channel
2701          */
2702         if (!mutex_trylock(&xprt->xpt_mutex)) {
2703                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2704                 if (!mutex_trylock(&xprt->xpt_mutex))
2705                         return -EAGAIN;
2706                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2707         }
2708         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2709                 len = -ENOTCONN;
2710         else
2711                 len = bc_sendto(req);
2712         mutex_unlock(&xprt->xpt_mutex);
2713
2714         if (len > 0)
2715                 len = 0;
2716
2717         return len;
2718 }
2719
2720 /*
2721  * The close routine. Since this is client initiated, we do nothing
2722  */
2723
2724 static void bc_close(struct rpc_xprt *xprt)
2725 {
2726 }
2727
2728 /*
2729  * The xprt destroy routine. Again, because this connection is client
2730  * initiated, we do nothing
2731  */
2732
2733 static void bc_destroy(struct rpc_xprt *xprt)
2734 {
2735         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2736
2737         xs_xprt_free(xprt);
2738         module_put(THIS_MODULE);
2739 }
2740
2741 static const struct rpc_xprt_ops xs_local_ops = {
2742         .reserve_xprt           = xprt_reserve_xprt,
2743         .release_xprt           = xs_tcp_release_xprt,
2744         .alloc_slot             = xprt_alloc_slot,
2745         .rpcbind                = xs_local_rpcbind,
2746         .set_port               = xs_local_set_port,
2747         .connect                = xs_local_connect,
2748         .buf_alloc              = rpc_malloc,
2749         .buf_free               = rpc_free,
2750         .send_request           = xs_local_send_request,
2751         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2752         .close                  = xs_close,
2753         .destroy                = xs_destroy,
2754         .print_stats            = xs_local_print_stats,
2755         .enable_swap            = xs_enable_swap,
2756         .disable_swap           = xs_disable_swap,
2757 };
2758
2759 static const struct rpc_xprt_ops xs_udp_ops = {
2760         .set_buffer_size        = xs_udp_set_buffer_size,
2761         .reserve_xprt           = xprt_reserve_xprt_cong,
2762         .release_xprt           = xprt_release_xprt_cong,
2763         .alloc_slot             = xprt_alloc_slot,
2764         .rpcbind                = rpcb_getport_async,
2765         .set_port               = xs_set_port,
2766         .connect                = xs_connect,
2767         .buf_alloc              = rpc_malloc,
2768         .buf_free               = rpc_free,
2769         .send_request           = xs_udp_send_request,
2770         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2771         .timer                  = xs_udp_timer,
2772         .release_request        = xprt_release_rqst_cong,
2773         .close                  = xs_close,
2774         .destroy                = xs_destroy,
2775         .print_stats            = xs_udp_print_stats,
2776         .enable_swap            = xs_enable_swap,
2777         .disable_swap           = xs_disable_swap,
2778         .inject_disconnect      = xs_inject_disconnect,
2779 };
2780
2781 static const struct rpc_xprt_ops xs_tcp_ops = {
2782         .reserve_xprt           = xprt_reserve_xprt,
2783         .release_xprt           = xs_tcp_release_xprt,
2784         .alloc_slot             = xprt_lock_and_alloc_slot,
2785         .rpcbind                = rpcb_getport_async,
2786         .set_port               = xs_set_port,
2787         .connect                = xs_connect,
2788         .buf_alloc              = rpc_malloc,
2789         .buf_free               = rpc_free,
2790         .send_request           = xs_tcp_send_request,
2791         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2792         .close                  = xs_tcp_shutdown,
2793         .destroy                = xs_destroy,
2794         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2795         .print_stats            = xs_tcp_print_stats,
2796         .enable_swap            = xs_enable_swap,
2797         .disable_swap           = xs_disable_swap,
2798         .inject_disconnect      = xs_inject_disconnect,
2799 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2800         .bc_setup               = xprt_setup_bc,
2801         .bc_up                  = xs_tcp_bc_up,
2802         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2803         .bc_free_rqst           = xprt_free_bc_rqst,
2804         .bc_destroy             = xprt_destroy_bc,
2805 #endif
2806 };
2807
2808 /*
2809  * The rpc_xprt_ops for the server backchannel
2810  */
2811
2812 static const struct rpc_xprt_ops bc_tcp_ops = {
2813         .reserve_xprt           = xprt_reserve_xprt,
2814         .release_xprt           = xprt_release_xprt,
2815         .alloc_slot             = xprt_alloc_slot,
2816         .buf_alloc              = bc_malloc,
2817         .buf_free               = bc_free,
2818         .send_request           = bc_send_request,
2819         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2820         .close                  = bc_close,
2821         .destroy                = bc_destroy,
2822         .print_stats            = xs_tcp_print_stats,
2823         .enable_swap            = xs_enable_swap,
2824         .disable_swap           = xs_disable_swap,
2825         .inject_disconnect      = xs_inject_disconnect,
2826 };
2827
2828 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2829 {
2830         static const struct sockaddr_in sin = {
2831                 .sin_family             = AF_INET,
2832                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2833         };
2834         static const struct sockaddr_in6 sin6 = {
2835                 .sin6_family            = AF_INET6,
2836                 .sin6_addr              = IN6ADDR_ANY_INIT,
2837         };
2838
2839         switch (family) {
2840         case AF_LOCAL:
2841                 break;
2842         case AF_INET:
2843                 memcpy(sap, &sin, sizeof(sin));
2844                 break;
2845         case AF_INET6:
2846                 memcpy(sap, &sin6, sizeof(sin6));
2847                 break;
2848         default:
2849                 dprintk("RPC:       %s: Bad address family\n", __func__);
2850                 return -EAFNOSUPPORT;
2851         }
2852         return 0;
2853 }
2854
2855 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2856                                       unsigned int slot_table_size,
2857                                       unsigned int max_slot_table_size)
2858 {
2859         struct rpc_xprt *xprt;
2860         struct sock_xprt *new;
2861
2862         if (args->addrlen > sizeof(xprt->addr)) {
2863                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2864                 return ERR_PTR(-EBADF);
2865         }
2866
2867         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2868                         max_slot_table_size);
2869         if (xprt == NULL) {
2870                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2871                                 "rpc_xprt\n");
2872                 return ERR_PTR(-ENOMEM);
2873         }
2874
2875         new = container_of(xprt, struct sock_xprt, xprt);
2876         mutex_init(&new->recv_mutex);
2877         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2878         xprt->addrlen = args->addrlen;
2879         if (args->srcaddr)
2880                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2881         else {
2882                 int err;
2883                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2884                                         (struct sockaddr *)&new->srcaddr);
2885                 if (err != 0) {
2886                         xprt_free(xprt);
2887                         return ERR_PTR(err);
2888                 }
2889         }
2890
2891         return xprt;
2892 }
2893
2894 static const struct rpc_timeout xs_local_default_timeout = {
2895         .to_initval = 10 * HZ,
2896         .to_maxval = 10 * HZ,
2897         .to_retries = 2,
2898 };
2899
2900 /**
2901  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2902  * @args: rpc transport creation arguments
2903  *
2904  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2905  */
2906 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2907 {
2908         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2909         struct sock_xprt *transport;
2910         struct rpc_xprt *xprt;
2911         struct rpc_xprt *ret;
2912
2913         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2914                         xprt_max_tcp_slot_table_entries);
2915         if (IS_ERR(xprt))
2916                 return xprt;
2917         transport = container_of(xprt, struct sock_xprt, xprt);
2918
2919         xprt->prot = 0;
2920         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2921         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2922
2923         xprt->bind_timeout = XS_BIND_TO;
2924         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2925         xprt->idle_timeout = XS_IDLE_DISC_TO;
2926
2927         xprt->ops = &xs_local_ops;
2928         xprt->timeout = &xs_local_default_timeout;
2929
2930         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2931         INIT_DELAYED_WORK(&transport->connect_worker,
2932                         xs_dummy_setup_socket);
2933
2934         switch (sun->sun_family) {
2935         case AF_LOCAL:
2936                 if (sun->sun_path[0] != '/') {
2937                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2938                                         sun->sun_path);
2939                         ret = ERR_PTR(-EINVAL);
2940                         goto out_err;
2941                 }
2942                 xprt_set_bound(xprt);
2943                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2944                 break;
2945         default:
2946                 ret = ERR_PTR(-EAFNOSUPPORT);
2947                 goto out_err;
2948         }
2949
2950         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2951                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2952
2953         if (try_module_get(THIS_MODULE))
2954                 return xprt;
2955         ret = ERR_PTR(-EINVAL);
2956 out_err:
2957         xs_xprt_free(xprt);
2958         return ret;
2959 }
2960
2961 static const struct rpc_timeout xs_udp_default_timeout = {
2962         .to_initval = 5 * HZ,
2963         .to_maxval = 30 * HZ,
2964         .to_increment = 5 * HZ,
2965         .to_retries = 5,
2966 };
2967
2968 /**
2969  * xs_setup_udp - Set up transport to use a UDP socket
2970  * @args: rpc transport creation arguments
2971  *
2972  */
2973 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2974 {
2975         struct sockaddr *addr = args->dstaddr;
2976         struct rpc_xprt *xprt;
2977         struct sock_xprt *transport;
2978         struct rpc_xprt *ret;
2979
2980         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2981                         xprt_udp_slot_table_entries);
2982         if (IS_ERR(xprt))
2983                 return xprt;
2984         transport = container_of(xprt, struct sock_xprt, xprt);
2985
2986         xprt->prot = IPPROTO_UDP;
2987         xprt->tsh_size = 0;
2988         /* XXX: header size can vary due to auth type, IPv6, etc. */
2989         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2990
2991         xprt->bind_timeout = XS_BIND_TO;
2992         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2993         xprt->idle_timeout = XS_IDLE_DISC_TO;
2994
2995         xprt->ops = &xs_udp_ops;
2996
2997         xprt->timeout = &xs_udp_default_timeout;
2998
2999         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3000         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3001
3002         switch (addr->sa_family) {
3003         case AF_INET:
3004                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3005                         xprt_set_bound(xprt);
3006
3007                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3008                 break;
3009         case AF_INET6:
3010                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3011                         xprt_set_bound(xprt);
3012
3013                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3014                 break;
3015         default:
3016                 ret = ERR_PTR(-EAFNOSUPPORT);
3017                 goto out_err;
3018         }
3019
3020         if (xprt_bound(xprt))
3021                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3022                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3023                                 xprt->address_strings[RPC_DISPLAY_PORT],
3024                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3025         else
3026                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3027                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3028                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3029
3030         if (try_module_get(THIS_MODULE))
3031                 return xprt;
3032         ret = ERR_PTR(-EINVAL);
3033 out_err:
3034         xs_xprt_free(xprt);
3035         return ret;
3036 }
3037
3038 static const struct rpc_timeout xs_tcp_default_timeout = {
3039         .to_initval = 60 * HZ,
3040         .to_maxval = 60 * HZ,
3041         .to_retries = 2,
3042 };
3043
3044 /**
3045  * xs_setup_tcp - Set up transport to use a TCP socket
3046  * @args: rpc transport creation arguments
3047  *
3048  */
3049 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3050 {
3051         struct sockaddr *addr = args->dstaddr;
3052         struct rpc_xprt *xprt;
3053         struct sock_xprt *transport;
3054         struct rpc_xprt *ret;
3055         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3056
3057         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3058                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3059
3060         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3061                         max_slot_table_size);
3062         if (IS_ERR(xprt))
3063                 return xprt;
3064         transport = container_of(xprt, struct sock_xprt, xprt);
3065
3066         xprt->prot = IPPROTO_TCP;
3067         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3068         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3069
3070         xprt->bind_timeout = XS_BIND_TO;
3071         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3072         xprt->idle_timeout = XS_IDLE_DISC_TO;
3073
3074         xprt->ops = &xs_tcp_ops;
3075         xprt->timeout = &xs_tcp_default_timeout;
3076
3077         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3078         xprt->connect_timeout = xprt->timeout->to_initval *
3079                 (xprt->timeout->to_retries + 1);
3080
3081         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3082         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3083
3084         switch (addr->sa_family) {
3085         case AF_INET:
3086                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3087                         xprt_set_bound(xprt);
3088
3089                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3090                 break;
3091         case AF_INET6:
3092                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3093                         xprt_set_bound(xprt);
3094
3095                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3096                 break;
3097         default:
3098                 ret = ERR_PTR(-EAFNOSUPPORT);
3099                 goto out_err;
3100         }
3101
3102         if (xprt_bound(xprt))
3103                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3104                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3105                                 xprt->address_strings[RPC_DISPLAY_PORT],
3106                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3107         else
3108                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3109                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3110                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3111
3112         if (try_module_get(THIS_MODULE))
3113                 return xprt;
3114         ret = ERR_PTR(-EINVAL);
3115 out_err:
3116         xs_xprt_free(xprt);
3117         return ret;
3118 }
3119
3120 /**
3121  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3122  * @args: rpc transport creation arguments
3123  *
3124  */
3125 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3126 {
3127         struct sockaddr *addr = args->dstaddr;
3128         struct rpc_xprt *xprt;
3129         struct sock_xprt *transport;
3130         struct svc_sock *bc_sock;
3131         struct rpc_xprt *ret;
3132
3133         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3134                         xprt_tcp_slot_table_entries);
3135         if (IS_ERR(xprt))
3136                 return xprt;
3137         transport = container_of(xprt, struct sock_xprt, xprt);
3138
3139         xprt->prot = IPPROTO_TCP;
3140         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3141         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3142         xprt->timeout = &xs_tcp_default_timeout;
3143
3144         /* backchannel */
3145         xprt_set_bound(xprt);
3146         xprt->bind_timeout = 0;
3147         xprt->reestablish_timeout = 0;
3148         xprt->idle_timeout = 0;
3149
3150         xprt->ops = &bc_tcp_ops;
3151
3152         switch (addr->sa_family) {
3153         case AF_INET:
3154                 xs_format_peer_addresses(xprt, "tcp",
3155                                          RPCBIND_NETID_TCP);
3156                 break;
3157         case AF_INET6:
3158                 xs_format_peer_addresses(xprt, "tcp",
3159                                    RPCBIND_NETID_TCP6);
3160                 break;
3161         default:
3162                 ret = ERR_PTR(-EAFNOSUPPORT);
3163                 goto out_err;
3164         }
3165
3166         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3167                         xprt->address_strings[RPC_DISPLAY_ADDR],
3168                         xprt->address_strings[RPC_DISPLAY_PORT],
3169                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3170
3171         /*
3172          * Once we've associated a backchannel xprt with a connection,
3173          * we want to keep it around as long as the connection lasts,
3174          * in case we need to start using it for a backchannel again;
3175          * this reference won't be dropped until bc_xprt is destroyed.
3176          */
3177         xprt_get(xprt);
3178         args->bc_xprt->xpt_bc_xprt = xprt;
3179         xprt->bc_xprt = args->bc_xprt;
3180         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3181         transport->sock = bc_sock->sk_sock;
3182         transport->inet = bc_sock->sk_sk;
3183
3184         /*
3185          * Since we don't want connections for the backchannel, we set
3186          * the xprt status to connected
3187          */
3188         xprt_set_connected(xprt);
3189
3190         if (try_module_get(THIS_MODULE))
3191                 return xprt;
3192
3193         args->bc_xprt->xpt_bc_xprt = NULL;
3194         args->bc_xprt->xpt_bc_xps = NULL;
3195         xprt_put(xprt);
3196         ret = ERR_PTR(-EINVAL);
3197 out_err:
3198         xs_xprt_free(xprt);
3199         return ret;
3200 }
3201
3202 static struct xprt_class        xs_local_transport = {
3203         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3204         .name           = "named UNIX socket",
3205         .owner          = THIS_MODULE,
3206         .ident          = XPRT_TRANSPORT_LOCAL,
3207         .setup          = xs_setup_local,
3208         .netid          = { "" },
3209 };
3210
3211 static struct xprt_class        xs_udp_transport = {
3212         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3213         .name           = "udp",
3214         .owner          = THIS_MODULE,
3215         .ident          = XPRT_TRANSPORT_UDP,
3216         .setup          = xs_setup_udp,
3217         .netid          = { "udp", "udp6", "" },
3218 };
3219
3220 static struct xprt_class        xs_tcp_transport = {
3221         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3222         .name           = "tcp",
3223         .owner          = THIS_MODULE,
3224         .ident          = XPRT_TRANSPORT_TCP,
3225         .setup          = xs_setup_tcp,
3226         .netid          = { "tcp", "tcp6", "" },
3227 };
3228
3229 static struct xprt_class        xs_bc_tcp_transport = {
3230         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3231         .name           = "tcp NFSv4.1 backchannel",
3232         .owner          = THIS_MODULE,
3233         .ident          = XPRT_TRANSPORT_BC_TCP,
3234         .setup          = xs_setup_bc_tcp,
3235         .netid          = { "" },
3236 };
3237
3238 /**
3239  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3240  *
3241  */
3242 int init_socket_xprt(void)
3243 {
3244 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3245         if (!sunrpc_table_header)
3246                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3247 #endif
3248
3249         xprt_register_transport(&xs_local_transport);
3250         xprt_register_transport(&xs_udp_transport);
3251         xprt_register_transport(&xs_tcp_transport);
3252         xprt_register_transport(&xs_bc_tcp_transport);
3253
3254         return 0;
3255 }
3256
3257 /**
3258  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3259  *
3260  */
3261 void cleanup_socket_xprt(void)
3262 {
3263 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3264         if (sunrpc_table_header) {
3265                 unregister_sysctl_table(sunrpc_table_header);
3266                 sunrpc_table_header = NULL;
3267         }
3268 #endif
3269
3270         xprt_unregister_transport(&xs_local_transport);
3271         xprt_unregister_transport(&xs_udp_transport);
3272         xprt_unregister_transport(&xs_tcp_transport);
3273         xprt_unregister_transport(&xs_bc_tcp_transport);
3274 }
3275
3276 static int param_set_uint_minmax(const char *val,
3277                 const struct kernel_param *kp,
3278                 unsigned int min, unsigned int max)
3279 {
3280         unsigned int num;
3281         int ret;
3282
3283         if (!val)
3284                 return -EINVAL;
3285         ret = kstrtouint(val, 0, &num);
3286         if (ret)
3287                 return ret;
3288         if (num < min || num > max)
3289                 return -EINVAL;
3290         *((unsigned int *)kp->arg) = num;
3291         return 0;
3292 }
3293
3294 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3295 {
3296         return param_set_uint_minmax(val, kp,
3297                         RPC_MIN_RESVPORT,
3298                         RPC_MAX_RESVPORT);
3299 }
3300
3301 static const struct kernel_param_ops param_ops_portnr = {
3302         .set = param_set_portnr,
3303         .get = param_get_uint,
3304 };
3305
3306 #define param_check_portnr(name, p) \
3307         __param_check(name, p, unsigned int);
3308
3309 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3310 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3311
3312 static int param_set_slot_table_size(const char *val,
3313                                      const struct kernel_param *kp)
3314 {
3315         return param_set_uint_minmax(val, kp,
3316                         RPC_MIN_SLOT_TABLE,
3317                         RPC_MAX_SLOT_TABLE);
3318 }
3319
3320 static const struct kernel_param_ops param_ops_slot_table_size = {
3321         .set = param_set_slot_table_size,
3322         .get = param_get_uint,
3323 };
3324
3325 #define param_check_slot_table_size(name, p) \
3326         __param_check(name, p, unsigned int);
3327
3328 static int param_set_max_slot_table_size(const char *val,
3329                                      const struct kernel_param *kp)
3330 {
3331         return param_set_uint_minmax(val, kp,
3332                         RPC_MIN_SLOT_TABLE,
3333                         RPC_MAX_SLOT_TABLE_LIMIT);
3334 }
3335
3336 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3337         .set = param_set_max_slot_table_size,
3338         .get = param_get_uint,
3339 };
3340
3341 #define param_check_max_slot_table_size(name, p) \
3342         __param_check(name, p, unsigned int);
3343
3344 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3345                    slot_table_size, 0644);
3346 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3347                    max_slot_table_size, 0644);
3348 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3349                    slot_table_size, 0644);
3350