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