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_limit
 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_limit,
 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         req->rq_xtime = ktime_get();
 531         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 532                               true, &sent);
 533         dprintk("RPC:       %s(%u) = %d\n",
 534                         __func__, xdr->len - req->rq_bytes_sent, status);
 535
 536         if (status == -EAGAIN && sock_writeable(transport->inet))
 537                 status = -ENOBUFS;
 538
 539         if (likely(sent > 0) || status == 0) {
 540                 req->rq_bytes_sent += sent;
 541                 req->rq_xmit_bytes_sent += sent;
 542                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 543                         req->rq_bytes_sent = 0;
 544                         return 0;
 545                 }
 546                 status = -EAGAIN;
 547         }
 548
 549         switch (status) {
 550         case -ENOBUFS:
 551                 break;
 552         case -EAGAIN:
 553                 status = xs_nospace(task);
 554                 break;
 555         default:
 556                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 557                         -status);
 558                 /* fall through */
 559         case -EPIPE:
 560                 xs_close(xprt);
 561                 status = -ENOTCONN;
 562         }
 563
 564         return status;
 565 }
 566
 567 /**
 568  * xs_udp_send_request - write an RPC request to a UDP socket
 569  * @task: address of RPC task that manages the state of an RPC request
 570  *
 571  * Return values:
 572  *        0:    The request has been sent
 573  *   EAGAIN:    The socket was blocked, please call again later to
 574  *              complete the request
 575  * ENOTCONN:    Caller needs to invoke connect logic then call again
 576  *    other:    Some other error occurred, the request was not sent
 577  */
 578 static int xs_udp_send_request(struct rpc_task *task)
 579 {
 580         struct rpc_rqst *req = task->tk_rqstp;
 581         struct rpc_xprt *xprt = req->rq_xprt;
 582         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 583         struct xdr_buf *xdr = &req->rq_snd_buf;
 584         int sent = 0;
 585         int status;
 586
 587         xs_pktdump("packet data:",
 588                                 req->rq_svec->iov_base,
 589                                 req->rq_svec->iov_len);
 590
 591         if (!xprt_bound(xprt))
 592                 return -ENOTCONN;
 593         req->rq_xtime = ktime_get();
 594         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 595                               xdr, req->rq_bytes_sent, true, &sent);
 596
 597         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 598                         xdr->len - req->rq_bytes_sent, status);
 599
 600         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 601         if (status == -EPERM)
 602                 goto process_status;
 603
 604         if (status == -EAGAIN && sock_writeable(transport->inet))
 605                 status = -ENOBUFS;
 606
 607         if (sent > 0 || status == 0) {
 608                 req->rq_xmit_bytes_sent += sent;
 609                 if (sent >= req->rq_slen)
 610                         return 0;
 611                 /* Still some bytes left; set up for a retry later. */
 612                 status = -EAGAIN;
 613         }
 614
 615 process_status:
 616         switch (status) {
 617         case -ENOTSOCK:
 618                 status = -ENOTCONN;
 619                 /* Should we call xs_close() here? */
 620                 break;
 621         case -EAGAIN:
 622                 status = xs_nospace(task);
 623                 break;
 624         case -ENETUNREACH:
 625         case -ENOBUFS:
 626         case -EPIPE:
 627         case -ECONNREFUSED:
 628         case -EPERM:
 629                 /* When the server has died, an ICMP port unreachable message
 630                  * prompts ECONNREFUSED. */
 631                 break;
 632         default:
 633                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 634                         -status);
 635         }
 636
 637         return status;
 638 }
 639
 640 /**
 641  * xs_tcp_send_request - write an RPC request to a TCP socket
 642  * @task: address of RPC task that manages the state of an RPC request
 643  *
 644  * Return values:
 645  *        0:    The request has been sent
 646  *   EAGAIN:    The socket was blocked, please call again later to
 647  *              complete the request
 648  * ENOTCONN:    Caller needs to invoke connect logic then call again
 649  *    other:    Some other error occurred, the request was not sent
 650  *
 651  * XXX: In the case of soft timeouts, should we eventually give up
 652  *      if sendmsg is not able to make progress?
 653  */
 654 static int xs_tcp_send_request(struct rpc_task *task)
 655 {
 656         struct rpc_rqst *req = task->tk_rqstp;
 657         struct rpc_xprt *xprt = req->rq_xprt;
 658         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 659         struct xdr_buf *xdr = &req->rq_snd_buf;
 660         bool zerocopy = true;
 661         bool vm_wait = false;
 662         int status;
 663         int sent;
 664
 665         xs_encode_stream_record_marker(&req->rq_snd_buf);
 666
 667         xs_pktdump("packet data:",
 668                                 req->rq_svec->iov_base,
 669                                 req->rq_svec->iov_len);
 670         /* Don't use zero copy if this is a resend. If the RPC call
 671          * completes while the socket holds a reference to the pages,
 672          * then we may end up resending corrupted data.
 673          */
 674         if (task->tk_flags & RPC_TASK_SENT)
 675                 zerocopy = false;
 676
 677         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
 678                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
 679
 680         /* Continue transmitting the packet/record. We must be careful
 681          * to cope with writespace callbacks arriving _after_ we have
 682          * called sendmsg(). */
 683         req->rq_xtime = ktime_get();
 684         while (1) {
 685                 sent = 0;
 686                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 687                                       req->rq_bytes_sent, zerocopy, &sent);
 688
 689                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 690                                 xdr->len - req->rq_bytes_sent, status);
 691
 692                 /* If we've sent the entire packet, immediately
 693                  * reset the count of bytes sent. */
 694                 req->rq_bytes_sent += sent;
 695                 req->rq_xmit_bytes_sent += sent;
 696                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 697                         req->rq_bytes_sent = 0;
 698                         return 0;
 699                 }
 700
 701                 WARN_ON_ONCE(sent == 0 && status == 0);
 702
 703                 if (status == -EAGAIN ) {
 704                         /*
 705                          * Return EAGAIN if we're sure we're hitting the
 706                          * socket send buffer limits.
 707                          */
 708                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 709                                 break;
 710                         /*
 711                          * Did we hit a memory allocation failure?
 712                          */
 713                         if (sent == 0) {
 714                                 status = -ENOBUFS;
 715                                 if (vm_wait)
 716                                         break;
 717                                 /* Retry, knowing now that we're below the
 718                                  * socket send buffer limit
 719                                  */
 720                                 vm_wait = true;
 721                         }
 722                         continue;
 723                 }
 724                 if (status < 0)
 725                         break;
 726                 vm_wait = false;
 727         }
 728
 729         switch (status) {
 730         case -ENOTSOCK:
 731                 status = -ENOTCONN;
 732                 /* Should we call xs_close() here? */
 733                 break;
 734         case -EAGAIN:
 735                 status = xs_nospace(task);
 736                 break;
 737         case -ECONNRESET:
 738         case -ECONNREFUSED:
 739         case -ENOTCONN:
 740         case -EADDRINUSE:
 741         case -ENOBUFS:
 742         case -EPIPE:
 743                 break;
 744         default:
 745                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 746                         -status);
 747         }
 748
 749         return status;
 750 }
 751
 752 /**
 753  * xs_tcp_release_xprt - clean up after a tcp transmission
 754  * @xprt: transport
 755  * @task: rpc task
 756  *
 757  * This cleans up if an error causes us to abort the transmission of a request.
 758  * In this case, the socket may need to be reset in order to avoid confusing
 759  * the server.
 760  */
 761 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 762 {
 763         struct rpc_rqst *req;
 764
 765         if (task != xprt->snd_task)
 766                 return;
 767         if (task == NULL)
 768                 goto out_release;
 769         req = task->tk_rqstp;
 770         if (req == NULL)
 771                 goto out_release;
 772         if (req->rq_bytes_sent == 0)
 773                 goto out_release;
 774         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 775                 goto out_release;
 776         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 777 out_release:
 778         xprt_release_xprt(xprt, task);
 779 }
 780
 781 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 782 {
 783         transport->old_data_ready = sk->sk_data_ready;
 784         transport->old_state_change = sk->sk_state_change;
 785         transport->old_write_space = sk->sk_write_space;
 786         transport->old_error_report = sk->sk_error_report;
 787 }
 788
 789 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 790 {
 791         sk->sk_data_ready = transport->old_data_ready;
 792         sk->sk_state_change = transport->old_state_change;
 793         sk->sk_write_space = transport->old_write_space;
 794         sk->sk_error_report = transport->old_error_report;
 795 }
 796
 797 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
 798 {
 799         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 800
 801         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 802 }
 803
 804 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 805 {
 806         smp_mb__before_atomic();
 807         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 808         clear_bit(XPRT_CLOSING, &xprt->state);
 809         xs_sock_reset_state_flags(xprt);
 810         smp_mb__after_atomic();
 811 }
 812
 813 /**
 814  * xs_error_report - callback to handle TCP socket state errors
 815  * @sk: socket
 816  *
 817  * Note: we don't call sock_error() since there may be a rpc_task
 818  * using the socket, and so we don't want to clear sk->sk_err.
 819  */
 820 static void xs_error_report(struct sock *sk)
 821 {
 822         struct rpc_xprt *xprt;
 823         int err;
 824
 825         read_lock_bh(&sk->sk_callback_lock);
 826         if (!(xprt = xprt_from_sock(sk)))
 827                 goto out;
 828
 829         err = -sk->sk_err;
 830         if (err == 0)
 831                 goto out;
 832         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 833                         xprt, -err);
 834         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 835         xprt_wake_pending_tasks(xprt, err);
 836  out:
 837         read_unlock_bh(&sk->sk_callback_lock);
 838 }
 839
 840 static void xs_reset_transport(struct sock_xprt *transport)
 841 {
 842         struct socket *sock = transport->sock;
 843         struct sock *sk = transport->inet;
 844         struct rpc_xprt *xprt = &transport->xprt;
 845
 846         if (sk == NULL)
 847                 return;
 848
 849         if (atomic_read(&transport->xprt.swapper))
 850                 sk_clear_memalloc(sk);
 851
 852         kernel_sock_shutdown(sock, SHUT_RDWR);
 853
 854         mutex_lock(&transport->recv_mutex);
 855         write_lock_bh(&sk->sk_callback_lock);
 856         transport->inet = NULL;
 857         transport->sock = NULL;
 858
 859         sk->sk_user_data = NULL;
 860
 861         xs_restore_old_callbacks(transport, sk);
 862         xprt_clear_connected(xprt);
 863         write_unlock_bh(&sk->sk_callback_lock);
 864         xs_sock_reset_connection_flags(xprt);
 865         mutex_unlock(&transport->recv_mutex);
 866
 867         trace_rpc_socket_close(xprt, sock);
 868         sock_release(sock);
 869 }
 870
 871 /**
 872  * xs_close - close a socket
 873  * @xprt: transport
 874  *
 875  * This is used when all requests are complete; ie, no DRC state remains
 876  * on the server we want to save.
 877  *
 878  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 879  * xs_reset_transport() zeroing the socket from underneath a writer.
 880  */
 881 static void xs_close(struct rpc_xprt *xprt)
 882 {
 883         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 884
 885         dprintk("RPC:       xs_close xprt %p\n", xprt);
 886
 887         xs_reset_transport(transport);
 888         xprt->reestablish_timeout = 0;
 889
 890         xprt_disconnect_done(xprt);
 891 }
 892
 893 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 894 {
 895         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 896                 xprt);
 897         xprt_disconnect_done(xprt);
 898 }
 899
 900 static void xs_xprt_free(struct rpc_xprt *xprt)
 901 {
 902         xs_free_peer_addresses(xprt);
 903         xprt_free(xprt);
 904 }
 905
 906 /**
 907  * xs_destroy - prepare to shutdown a transport
 908  * @xprt: doomed transport
 909  *
 910  */
 911 static void xs_destroy(struct rpc_xprt *xprt)
 912 {
 913         struct sock_xprt *transport = container_of(xprt,
 914                         struct sock_xprt, xprt);
 915         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 916
 917         cancel_delayed_work_sync(&transport->connect_worker);
 918         xs_close(xprt);
 919         cancel_work_sync(&transport->recv_worker);
 920         xs_xprt_free(xprt);
 921         module_put(THIS_MODULE);
 922 }
 923
 924 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 925 {
 926         struct xdr_skb_reader desc = {
 927                 .skb            = skb,
 928                 .offset         = sizeof(rpc_fraghdr),
 929                 .count          = skb->len - sizeof(rpc_fraghdr),
 930         };
 931
 932         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 933                 return -1;
 934         if (desc.count)
 935                 return -1;
 936         return 0;
 937 }
 938
 939 /**
 940  * xs_local_data_read_skb
 941  * @xprt: transport
 942  * @sk: socket
 943  * @skb: skbuff
 944  *
 945  * Currently this assumes we can read the whole reply in a single gulp.
 946  */
 947 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 948                 struct sock *sk,
 949                 struct sk_buff *skb)
 950 {
 951         struct rpc_task *task;
 952         struct rpc_rqst *rovr;
 953         int repsize, copied;
 954         u32 _xid;
 955         __be32 *xp;
 956
 957         repsize = skb->len - sizeof(rpc_fraghdr);
 958         if (repsize < 4) {
 959                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 960                 return;
 961         }
 962
 963         /* Copy the XID from the skb... */
 964         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 965         if (xp == NULL)
 966                 return;
 967
 968         /* Look up and lock the request corresponding to the given XID */
 969         spin_lock(&xprt->recv_lock);
 970         rovr = xprt_lookup_rqst(xprt, *xp);
 971         if (!rovr)
 972                 goto out_unlock;
 973         xprt_pin_rqst(rovr);
 974         spin_unlock(&xprt->recv_lock);
 975         task = rovr->rq_task;
 976
 977         copied = rovr->rq_private_buf.buflen;
 978         if (copied > repsize)
 979                 copied = repsize;
 980
 981         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 982                 dprintk("RPC:       sk_buff copy failed\n");
 983                 spin_lock(&xprt->recv_lock);
 984                 goto out_unpin;
 985         }
 986
 987         spin_lock(&xprt->recv_lock);
 988         xprt_complete_rqst(task, copied);
 989 out_unpin:
 990         xprt_unpin_rqst(rovr);
 991  out_unlock:
 992         spin_unlock(&xprt->recv_lock);
 993 }
 994
 995 static void xs_local_data_receive(struct sock_xprt *transport)
 996 {
 997         struct sk_buff *skb;
 998         struct sock *sk;
 999         int err;
1000
1001 restart:
1002         mutex_lock(&transport->recv_mutex);
1003         sk = transport->inet;
1004         if (sk == NULL)
1005                 goto out;
1006         for (;;) {
1007                 skb = skb_recv_datagram(sk, 0, 1, &err);
1008                 if (skb != NULL) {
1009                         xs_local_data_read_skb(&transport->xprt, sk, skb);
1010                         skb_free_datagram(sk, skb);
1011                         continue;
1012                 }
1013                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1014                         break;
1015                 if (need_resched()) {
1016                         mutex_unlock(&transport->recv_mutex);
1017                         cond_resched();
1018                         goto restart;
1019                 }
1020         }
1021 out:
1022         mutex_unlock(&transport->recv_mutex);
1023 }
1024
1025 static void xs_local_data_receive_workfn(struct work_struct *work)
1026 {
1027         struct sock_xprt *transport =
1028                 container_of(work, struct sock_xprt, recv_worker);
1029         xs_local_data_receive(transport);
1030 }
1031
1032 /**
1033  * xs_udp_data_read_skb - receive callback for UDP sockets
1034  * @xprt: transport
1035  * @sk: socket
1036  * @skb: skbuff
1037  *
1038  */
1039 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1040                 struct sock *sk,
1041                 struct sk_buff *skb)
1042 {
1043         struct rpc_task *task;
1044         struct rpc_rqst *rovr;
1045         int repsize, copied;
1046         u32 _xid;
1047         __be32 *xp;
1048
1049         repsize = skb->len;
1050         if (repsize < 4) {
1051                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1052                 return;
1053         }
1054
1055         /* Copy the XID from the skb... */
1056         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1057         if (xp == NULL)
1058                 return;
1059
1060         /* Look up and lock the request corresponding to the given XID */
1061         spin_lock(&xprt->recv_lock);
1062         rovr = xprt_lookup_rqst(xprt, *xp);
1063         if (!rovr)
1064                 goto out_unlock;
1065         xprt_pin_rqst(rovr);
1066         xprt_update_rtt(rovr->rq_task);
1067         spin_unlock(&xprt->recv_lock);
1068         task = rovr->rq_task;
1069
1070         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1071                 copied = repsize;
1072
1073         /* Suck it into the iovec, verify checksum if not done by hw. */
1074         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1075                 spin_lock(&xprt->recv_lock);
1076                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1077                 goto out_unpin;
1078         }
1079
1080
1081         spin_lock_bh(&xprt->transport_lock);
1082         xprt_adjust_cwnd(xprt, task, copied);
1083         spin_unlock_bh(&xprt->transport_lock);
1084         spin_lock(&xprt->recv_lock);
1085         xprt_complete_rqst(task, copied);
1086         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1087 out_unpin:
1088         xprt_unpin_rqst(rovr);
1089  out_unlock:
1090         spin_unlock(&xprt->recv_lock);
1091 }
1092
1093 static void xs_udp_data_receive(struct sock_xprt *transport)
1094 {
1095         struct sk_buff *skb;
1096         struct sock *sk;
1097         int err;
1098
1099 restart:
1100         mutex_lock(&transport->recv_mutex);
1101         sk = transport->inet;
1102         if (sk == NULL)
1103                 goto out;
1104         for (;;) {
1105                 skb = skb_recv_udp(sk, 0, 1, &err);
1106                 if (skb != NULL) {
1107                         xs_udp_data_read_skb(&transport->xprt, sk, skb);
1108                         consume_skb(skb);
1109                         continue;
1110                 }
1111                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1112                         break;
1113                 if (need_resched()) {
1114                         mutex_unlock(&transport->recv_mutex);
1115                         cond_resched();
1116                         goto restart;
1117                 }
1118         }
1119 out:
1120         mutex_unlock(&transport->recv_mutex);
1121 }
1122
1123 static void xs_udp_data_receive_workfn(struct work_struct *work)
1124 {
1125         struct sock_xprt *transport =
1126                 container_of(work, struct sock_xprt, recv_worker);
1127         xs_udp_data_receive(transport);
1128 }
1129
1130 /**
1131  * xs_data_ready - "data ready" callback for UDP sockets
1132  * @sk: socket with data to read
1133  *
1134  */
1135 static void xs_data_ready(struct sock *sk)
1136 {
1137         struct rpc_xprt *xprt;
1138
1139         read_lock_bh(&sk->sk_callback_lock);
1140         dprintk("RPC:       xs_data_ready...\n");
1141         xprt = xprt_from_sock(sk);
1142         if (xprt != NULL) {
1143                 struct sock_xprt *transport = container_of(xprt,
1144                                 struct sock_xprt, xprt);
1145                 transport->old_data_ready(sk);
1146                 /* Any data means we had a useful conversation, so
1147                  * then we don't need to delay the next reconnect
1148                  */
1149                 if (xprt->reestablish_timeout)
1150                         xprt->reestablish_timeout = 0;
1151                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1152                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1153         }
1154         read_unlock_bh(&sk->sk_callback_lock);
1155 }
1156
1157 /*
1158  * Helper function to force a TCP close if the server is sending
1159  * junk and/or it has put us in CLOSE_WAIT
1160  */
1161 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1162 {
1163         xprt_force_disconnect(xprt);
1164 }
1165
1166 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1167 {
1168         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1169         size_t len, used;
1170         char *p;
1171
1172         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1173         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1174         used = xdr_skb_read_bits(desc, p, len);
1175         transport->tcp_offset += used;
1176         if (used != len)
1177                 return;
1178
1179         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1180         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1181                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1182         else
1183                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1184         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1185
1186         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1187         transport->tcp_offset = 0;
1188
1189         /* Sanity check of the record length */
1190         if (unlikely(transport->tcp_reclen < 8)) {
1191                 dprintk("RPC:       invalid TCP record fragment length\n");
1192                 xs_tcp_force_close(xprt);
1193                 return;
1194         }
1195         dprintk("RPC:       reading TCP record fragment of length %d\n",
1196                         transport->tcp_reclen);
1197 }
1198
1199 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1200 {
1201         if (transport->tcp_offset == transport->tcp_reclen) {
1202                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1203                 transport->tcp_offset = 0;
1204                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1205                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1206                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1207                         transport->tcp_copied = 0;
1208                 }
1209         }
1210 }
1211
1212 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1213 {
1214         size_t len, used;
1215         char *p;
1216
1217         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1218         dprintk("RPC:       reading XID (%zu bytes)\n", len);
1219         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1220         used = xdr_skb_read_bits(desc, p, len);
1221         transport->tcp_offset += used;
1222         if (used != len)
1223                 return;
1224         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1225         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1226         transport->tcp_copied = 4;
1227         dprintk("RPC:       reading %s XID %08x\n",
1228                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1229                                                               : "request with",
1230                         ntohl(transport->tcp_xid));
1231         xs_tcp_check_fraghdr(transport);
1232 }
1233
1234 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1235                                        struct xdr_skb_reader *desc)
1236 {
1237         size_t len, used;
1238         u32 offset;
1239         char *p;
1240
1241         /*
1242          * We want transport->tcp_offset to be 8 at the end of this routine
1243          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1244          * When this function is called for the first time,
1245          * transport->tcp_offset is 4 (after having already read the xid).
1246          */
1247         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1248         len = sizeof(transport->tcp_calldir) - offset;
1249         dprintk("RPC:       reading CALL/REPLY flag (%zu bytes)\n", len);
1250         p = ((char *) &transport->tcp_calldir) + offset;
1251         used = xdr_skb_read_bits(desc, p, len);
1252         transport->tcp_offset += used;
1253         if (used != len)
1254                 return;
1255         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1256         /*
1257          * We don't yet have the XDR buffer, so we will write the calldir
1258          * out after we get the buffer from the 'struct rpc_rqst'
1259          */
1260         switch (ntohl(transport->tcp_calldir)) {
1261         case RPC_REPLY:
1262                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1263                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1264                 transport->tcp_flags |= TCP_RPC_REPLY;
1265                 break;
1266         case RPC_CALL:
1267                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1268                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1269                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1270                 break;
1271         default:
1272                 dprintk("RPC:       invalid request message type\n");
1273                 xs_tcp_force_close(&transport->xprt);
1274         }
1275         xs_tcp_check_fraghdr(transport);
1276 }
1277
1278 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1279                                      struct xdr_skb_reader *desc,
1280                                      struct rpc_rqst *req)
1281 {
1282         struct sock_xprt *transport =
1283                                 container_of(xprt, struct sock_xprt, xprt);
1284         struct xdr_buf *rcvbuf;
1285         size_t len;
1286         ssize_t r;
1287
1288         rcvbuf = &req->rq_private_buf;
1289
1290         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1291                 /*
1292                  * Save the RPC direction in the XDR buffer
1293                  */
1294                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1295                         &transport->tcp_calldir,
1296                         sizeof(transport->tcp_calldir));
1297                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1298                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1299         }
1300
1301         len = desc->count;
1302         if (len > transport->tcp_reclen - transport->tcp_offset)
1303                 desc->count = transport->tcp_reclen - transport->tcp_offset;
1304         r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1305                                           desc, xdr_skb_read_bits);
1306
1307         if (desc->count) {
1308                 /* Error when copying to the receive buffer,
1309                  * usually because we weren't able to allocate
1310                  * additional buffer pages. All we can do now
1311                  * is turn off TCP_RCV_COPY_DATA, so the request
1312                  * will not receive any additional updates,
1313                  * and time out.
1314                  * Any remaining data from this record will
1315                  * be discarded.
1316                  */
1317                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1318                 dprintk("RPC:       XID %08x truncated request\n",
1319                                 ntohl(transport->tcp_xid));
1320                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1321                                 "tcp_offset = %u, tcp_reclen = %u\n",
1322                                 xprt, transport->tcp_copied,
1323                                 transport->tcp_offset, transport->tcp_reclen);
1324                 return;
1325         }
1326
1327         transport->tcp_copied += r;
1328         transport->tcp_offset += r;
1329         desc->count = len - r;
1330
1331         dprintk("RPC:       XID %08x read %zd bytes\n",
1332                         ntohl(transport->tcp_xid), r);
1333         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1334                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1335                         transport->tcp_offset, transport->tcp_reclen);
1336
1337         if (transport->tcp_copied == req->rq_private_buf.buflen)
1338                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1339         else if (transport->tcp_offset == transport->tcp_reclen) {
1340                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1341                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1342         }
1343 }
1344
1345 /*
1346  * Finds the request corresponding to the RPC xid and invokes the common
1347  * tcp read code to read the data.
1348  */
1349 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1350                                     struct xdr_skb_reader *desc)
1351 {
1352         struct sock_xprt *transport =
1353                                 container_of(xprt, struct sock_xprt, xprt);
1354         struct rpc_rqst *req;
1355
1356         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1357
1358         /* Find and lock the request corresponding to this xid */
1359         spin_lock(&xprt->recv_lock);
1360         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1361         if (!req) {
1362                 dprintk("RPC:       XID %08x request not found!\n",
1363                                 ntohl(transport->tcp_xid));
1364                 spin_unlock(&xprt->recv_lock);
1365                 return -1;
1366         }
1367         xprt_pin_rqst(req);
1368         spin_unlock(&xprt->recv_lock);
1369
1370         xs_tcp_read_common(xprt, desc, req);
1371
1372         spin_lock(&xprt->recv_lock);
1373         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1374                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1375         xprt_unpin_rqst(req);
1376         spin_unlock(&xprt->recv_lock);
1377         return 0;
1378 }
1379
1380 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1381 /*
1382  * Obtains an rpc_rqst previously allocated and invokes the common
1383  * tcp read code to read the data.  The result is placed in the callback
1384  * queue.
1385  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1386  * connection and return -1.
1387  */
1388 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1389                                        struct xdr_skb_reader *desc)
1390 {
1391         struct sock_xprt *transport =
1392                                 container_of(xprt, struct sock_xprt, xprt);
1393         struct rpc_rqst *req;
1394
1395         /* Look up the request corresponding to the given XID */
1396         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1397         if (req == NULL) {
1398                 printk(KERN_WARNING "Callback slot table overflowed\n");
1399                 xprt_force_disconnect(xprt);
1400                 return -1;
1401         }
1402
1403         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1404         xs_tcp_read_common(xprt, desc, req);
1405
1406         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1407                 xprt_complete_bc_request(req, transport->tcp_copied);
1408
1409         return 0;
1410 }
1411
1412 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1413                                         struct xdr_skb_reader *desc)
1414 {
1415         struct sock_xprt *transport =
1416                                 container_of(xprt, struct sock_xprt, xprt);
1417
1418         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1419                 xs_tcp_read_reply(xprt, desc) :
1420                 xs_tcp_read_callback(xprt, desc);
1421 }
1422
1423 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1424 {
1425         int ret;
1426
1427         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1428                               SVC_SOCK_ANONYMOUS);
1429         if (ret < 0)
1430                 return ret;
1431         return 0;
1432 }
1433
1434 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1435 {
1436         return PAGE_SIZE;
1437 }
1438 #else
1439 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1440                                         struct xdr_skb_reader *desc)
1441 {
1442         return xs_tcp_read_reply(xprt, desc);
1443 }
1444 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1445
1446 /*
1447  * Read data off the transport.  This can be either an RPC_CALL or an
1448  * RPC_REPLY.  Relay the processing to helper functions.
1449  */
1450 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1451                                     struct xdr_skb_reader *desc)
1452 {
1453         struct sock_xprt *transport =
1454                                 container_of(xprt, struct sock_xprt, xprt);
1455
1456         if (_xs_tcp_read_data(xprt, desc) == 0)
1457                 xs_tcp_check_fraghdr(transport);
1458         else {
1459                 /*
1460                  * The transport_lock protects the request handling.
1461                  * There's no need to hold it to update the tcp_flags.
1462                  */
1463                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1464         }
1465 }
1466
1467 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1468 {
1469         size_t len;
1470
1471         len = transport->tcp_reclen - transport->tcp_offset;
1472         if (len > desc->count)
1473                 len = desc->count;
1474         desc->count -= len;
1475         desc->offset += len;
1476         transport->tcp_offset += len;
1477         dprintk("RPC:       discarded %zu bytes\n", len);
1478         xs_tcp_check_fraghdr(transport);
1479 }
1480
1481 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1482 {
1483         struct rpc_xprt *xprt = rd_desc->arg.data;
1484         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1485         struct xdr_skb_reader desc = {
1486                 .skb    = skb,
1487                 .offset = offset,
1488                 .count  = len,
1489         };
1490         size_t ret;
1491
1492         dprintk("RPC:       xs_tcp_data_recv started\n");
1493         do {
1494                 trace_xs_tcp_data_recv(transport);
1495                 /* Read in a new fragment marker if necessary */
1496                 /* Can we ever really expect to get completely empty fragments? */
1497                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1498                         xs_tcp_read_fraghdr(xprt, &desc);
1499                         continue;
1500                 }
1501                 /* Read in the xid if necessary */
1502                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1503                         xs_tcp_read_xid(transport, &desc);
1504                         continue;
1505                 }
1506                 /* Read in the call/reply flag */
1507                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1508                         xs_tcp_read_calldir(transport, &desc);
1509                         continue;
1510                 }
1511                 /* Read in the request data */
1512                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1513                         xs_tcp_read_data(xprt, &desc);
1514                         continue;
1515                 }
1516                 /* Skip over any trailing bytes on short reads */
1517                 xs_tcp_read_discard(transport, &desc);
1518         } while (desc.count);
1519         ret = len - desc.count;
1520         if (ret < rd_desc->count)
1521                 rd_desc->count -= ret;
1522         else
1523                 rd_desc->count = 0;
1524         trace_xs_tcp_data_recv(transport);
1525         dprintk("RPC:       xs_tcp_data_recv done\n");
1526         return ret;
1527 }
1528
1529 static void xs_tcp_data_receive(struct sock_xprt *transport)
1530 {
1531         struct rpc_xprt *xprt = &transport->xprt;
1532         struct sock *sk;
1533         read_descriptor_t rd_desc = {
1534                 .arg.data = xprt,
1535         };
1536         unsigned long total = 0;
1537         int read = 0;
1538
1539 restart:
1540         mutex_lock(&transport->recv_mutex);
1541         sk = transport->inet;
1542         if (sk == NULL)
1543                 goto out;
1544
1545         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1546         for (;;) {
1547                 rd_desc.count = RPC_TCP_READ_CHUNK_SZ;
1548                 lock_sock(sk);
1549                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1550                 if (rd_desc.count != 0 || read < 0) {
1551                         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1552                         release_sock(sk);
1553                         break;
1554                 }
1555                 release_sock(sk);
1556                 total += read;
1557                 if (need_resched()) {
1558                         mutex_unlock(&transport->recv_mutex);
1559                         cond_resched();
1560                         goto restart;
1561                 }
1562         }
1563         if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1564                 queue_work(xprtiod_workqueue, &transport->recv_worker);
1565 out:
1566         mutex_unlock(&transport->recv_mutex);
1567         trace_xs_tcp_data_ready(xprt, read, total);
1568 }
1569
1570 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1571 {
1572         struct sock_xprt *transport =
1573                 container_of(work, struct sock_xprt, recv_worker);
1574         xs_tcp_data_receive(transport);
1575 }
1576
1577 /**
1578  * xs_tcp_state_change - callback to handle TCP socket state changes
1579  * @sk: socket whose state has changed
1580  *
1581  */
1582 static void xs_tcp_state_change(struct sock *sk)
1583 {
1584         struct rpc_xprt *xprt;
1585         struct sock_xprt *transport;
1586
1587         read_lock_bh(&sk->sk_callback_lock);
1588         if (!(xprt = xprt_from_sock(sk)))
1589                 goto out;
1590         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1591         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1592                         sk->sk_state, xprt_connected(xprt),
1593                         sock_flag(sk, SOCK_DEAD),
1594                         sock_flag(sk, SOCK_ZAPPED),
1595                         sk->sk_shutdown);
1596
1597         transport = container_of(xprt, struct sock_xprt, xprt);
1598         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1599         switch (sk->sk_state) {
1600         case TCP_ESTABLISHED:
1601                 spin_lock(&xprt->transport_lock);
1602                 if (!xprt_test_and_set_connected(xprt)) {
1603
1604                         /* Reset TCP record info */
1605                         transport->tcp_offset = 0;
1606                         transport->tcp_reclen = 0;
1607                         transport->tcp_copied = 0;
1608                         transport->tcp_flags =
1609                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1610                         xprt->connect_cookie++;
1611                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1612                         xprt_clear_connecting(xprt);
1613
1614                         xprt->stat.connect_count++;
1615                         xprt->stat.connect_time += (long)jiffies -
1616                                                    xprt->stat.connect_start;
1617                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1618                 }
1619                 spin_unlock(&xprt->transport_lock);
1620                 break;
1621         case TCP_FIN_WAIT1:
1622                 /* The client initiated a shutdown of the socket */
1623                 xprt->connect_cookie++;
1624                 xprt->reestablish_timeout = 0;
1625                 set_bit(XPRT_CLOSING, &xprt->state);
1626                 smp_mb__before_atomic();
1627                 clear_bit(XPRT_CONNECTED, &xprt->state);
1628                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1629                 smp_mb__after_atomic();
1630                 break;
1631         case TCP_CLOSE_WAIT:
1632                 /* The server initiated a shutdown of the socket */
1633                 xprt->connect_cookie++;
1634                 clear_bit(XPRT_CONNECTED, &xprt->state);
1635                 xs_tcp_force_close(xprt);
1636                 /* fall through */
1637         case TCP_CLOSING:
1638                 /*
1639                  * If the server closed down the connection, make sure that
1640                  * we back off before reconnecting
1641                  */
1642                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1643                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1644                 break;
1645         case TCP_LAST_ACK:
1646                 set_bit(XPRT_CLOSING, &xprt->state);
1647                 smp_mb__before_atomic();
1648                 clear_bit(XPRT_CONNECTED, &xprt->state);
1649                 smp_mb__after_atomic();
1650                 break;
1651         case TCP_CLOSE:
1652                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1653                                         &transport->sock_state))
1654                         xprt_clear_connecting(xprt);
1655                 clear_bit(XPRT_CLOSING, &xprt->state);
1656                 if (sk->sk_err)
1657                         xprt_wake_pending_tasks(xprt, -sk->sk_err);
1658                 /* Trigger the socket release */
1659                 xs_tcp_force_close(xprt);
1660         }
1661  out:
1662         read_unlock_bh(&sk->sk_callback_lock);
1663 }
1664
1665 static void xs_write_space(struct sock *sk)
1666 {
1667         struct socket_wq *wq;
1668         struct rpc_xprt *xprt;
1669
1670         if (!sk->sk_socket)
1671                 return;
1672         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1673
1674         if (unlikely(!(xprt = xprt_from_sock(sk))))
1675                 return;
1676         rcu_read_lock();
1677         wq = rcu_dereference(sk->sk_wq);
1678         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1679                 goto out;
1680
1681         xprt_write_space(xprt);
1682 out:
1683         rcu_read_unlock();
1684 }
1685
1686 /**
1687  * xs_udp_write_space - callback invoked when socket buffer space
1688  *                             becomes available
1689  * @sk: socket whose state has changed
1690  *
1691  * Called when more output buffer space is available for this socket.
1692  * We try not to wake our writers until they can make "significant"
1693  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1694  * with a bunch of small requests.
1695  */
1696 static void xs_udp_write_space(struct sock *sk)
1697 {
1698         read_lock_bh(&sk->sk_callback_lock);
1699
1700         /* from net/core/sock.c:sock_def_write_space */
1701         if (sock_writeable(sk))
1702                 xs_write_space(sk);
1703
1704         read_unlock_bh(&sk->sk_callback_lock);
1705 }
1706
1707 /**
1708  * xs_tcp_write_space - callback invoked when socket buffer space
1709  *                             becomes available
1710  * @sk: socket whose state has changed
1711  *
1712  * Called when more output buffer space is available for this socket.
1713  * We try not to wake our writers until they can make "significant"
1714  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1715  * with a bunch of small requests.
1716  */
1717 static void xs_tcp_write_space(struct sock *sk)
1718 {
1719         read_lock_bh(&sk->sk_callback_lock);
1720
1721         /* from net/core/stream.c:sk_stream_write_space */
1722         if (sk_stream_is_writeable(sk))
1723                 xs_write_space(sk);
1724
1725         read_unlock_bh(&sk->sk_callback_lock);
1726 }
1727
1728 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1729 {
1730         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1731         struct sock *sk = transport->inet;
1732
1733         if (transport->rcvsize) {
1734                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1735                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1736         }
1737         if (transport->sndsize) {
1738                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1739                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1740                 sk->sk_write_space(sk);
1741         }
1742 }
1743
1744 /**
1745  * xs_udp_set_buffer_size - set send and receive limits
1746  * @xprt: generic transport
1747  * @sndsize: requested size of send buffer, in bytes
1748  * @rcvsize: requested size of receive buffer, in bytes
1749  *
1750  * Set socket send and receive buffer size limits.
1751  */
1752 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1753 {
1754         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1755
1756         transport->sndsize = 0;
1757         if (sndsize)
1758                 transport->sndsize = sndsize + 1024;
1759         transport->rcvsize = 0;
1760         if (rcvsize)
1761                 transport->rcvsize = rcvsize + 1024;
1762
1763         xs_udp_do_set_buffer_size(xprt);
1764 }
1765
1766 /**
1767  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1768  * @task: task that timed out
1769  *
1770  * Adjust the congestion window after a retransmit timeout has occurred.
1771  */
1772 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1773 {
1774         spin_lock_bh(&xprt->transport_lock);
1775         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1776         spin_unlock_bh(&xprt->transport_lock);
1777 }
1778
1779 static int xs_get_random_port(void)
1780 {
1781         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1782         unsigned short range;
1783         unsigned short rand;
1784
1785         if (max < min)
1786                 return -EADDRINUSE;
1787         range = max - min + 1;
1788         rand = (unsigned short) prandom_u32() % range;
1789         return rand + min;
1790 }
1791
1792 /**
1793  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1794  * @sock: socket
1795  *
1796  * Note that this function has to be called on all sockets that share the
1797  * same port, and it must be called before binding.
1798  */
1799 static void xs_sock_set_reuseport(struct socket *sock)
1800 {
1801         int opt = 1;
1802
1803         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1804                         (char *)&opt, sizeof(opt));
1805 }
1806
1807 static unsigned short xs_sock_getport(struct socket *sock)
1808 {
1809         struct sockaddr_storage buf;
1810         unsigned short port = 0;
1811
1812         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1813                 goto out;
1814         switch (buf.ss_family) {
1815         case AF_INET6:
1816                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1817                 break;
1818         case AF_INET:
1819                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1820         }
1821 out:
1822         return port;
1823 }
1824
1825 /**
1826  * xs_set_port - reset the port number in the remote endpoint address
1827  * @xprt: generic transport
1828  * @port: new port number
1829  *
1830  */
1831 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1832 {
1833         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1834
1835         rpc_set_port(xs_addr(xprt), port);
1836         xs_update_peer_port(xprt);
1837 }
1838
1839 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1840 {
1841         if (transport->srcport == 0)
1842                 transport->srcport = xs_sock_getport(sock);
1843 }
1844
1845 static int xs_get_srcport(struct sock_xprt *transport)
1846 {
1847         int port = transport->srcport;
1848
1849         if (port == 0 && transport->xprt.resvport)
1850                 port = xs_get_random_port();
1851         return port;
1852 }
1853
1854 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1855 {
1856         if (transport->srcport != 0)
1857                 transport->srcport = 0;
1858         if (!transport->xprt.resvport)
1859                 return 0;
1860         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1861                 return xprt_max_resvport;
1862         return --port;
1863 }
1864 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1865 {
1866         struct sockaddr_storage myaddr;
1867         int err, nloop = 0;
1868         int port = xs_get_srcport(transport);
1869         unsigned short last;
1870
1871         /*
1872          * If we are asking for any ephemeral port (i.e. port == 0 &&
1873          * transport->xprt.resvport == 0), don't bind.  Let the local
1874          * port selection happen implicitly when the socket is used
1875          * (for example at connect time).
1876          *
1877          * This ensures that we can continue to establish TCP
1878          * connections even when all local ephemeral ports are already
1879          * a part of some TCP connection.  This makes no difference
1880          * for UDP sockets, but also doens't harm them.
1881          *
1882          * If we're asking for any reserved port (i.e. port == 0 &&
1883          * transport->xprt.resvport == 1) xs_get_srcport above will
1884          * ensure that port is non-zero and we will bind as needed.
1885          */
1886         if (port <= 0)
1887                 return port;
1888
1889         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1890         do {
1891                 rpc_set_port((struct sockaddr *)&myaddr, port);
1892                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1893                                 transport->xprt.addrlen);
1894                 if (err == 0) {
1895                         transport->srcport = port;
1896                         break;
1897                 }
1898                 last = port;
1899                 port = xs_next_srcport(transport, port);
1900                 if (port > last)
1901                         nloop++;
1902         } while (err == -EADDRINUSE && nloop != 2);
1903
1904         if (myaddr.ss_family == AF_INET)
1905                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1906                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1907                                 port, err ? "failed" : "ok", err);
1908         else
1909                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1910                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1911                                 port, err ? "failed" : "ok", err);
1912         return err;
1913 }
1914
1915 /*
1916  * We don't support autobind on AF_LOCAL sockets
1917  */
1918 static void xs_local_rpcbind(struct rpc_task *task)
1919 {
1920         xprt_set_bound(task->tk_xprt);
1921 }
1922
1923 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1924 {
1925 }
1926
1927 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1928 static struct lock_class_key xs_key[2];
1929 static struct lock_class_key xs_slock_key[2];
1930
1931 static inline void xs_reclassify_socketu(struct socket *sock)
1932 {
1933         struct sock *sk = sock->sk;
1934
1935         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1936                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1937 }
1938
1939 static inline void xs_reclassify_socket4(struct socket *sock)
1940 {
1941         struct sock *sk = sock->sk;
1942
1943         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1944                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1945 }
1946
1947 static inline void xs_reclassify_socket6(struct socket *sock)
1948 {
1949         struct sock *sk = sock->sk;
1950
1951         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1952                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1953 }
1954
1955 static inline void xs_reclassify_socket(int family, struct socket *sock)
1956 {
1957         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1958                 return;
1959
1960         switch (family) {
1961         case AF_LOCAL:
1962                 xs_reclassify_socketu(sock);
1963                 break;
1964         case AF_INET:
1965                 xs_reclassify_socket4(sock);
1966                 break;
1967         case AF_INET6:
1968                 xs_reclassify_socket6(sock);
1969                 break;
1970         }
1971 }
1972 #else
1973 static inline void xs_reclassify_socket(int family, struct socket *sock)
1974 {
1975 }
1976 #endif
1977
1978 static void xs_dummy_setup_socket(struct work_struct *work)
1979 {
1980 }
1981
1982 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1983                 struct sock_xprt *transport, int family, int type,
1984                 int protocol, bool reuseport)
1985 {
1986         struct socket *sock;
1987         int err;
1988
1989         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1990         if (err < 0) {
1991                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1992                                 protocol, -err);
1993                 goto out;
1994         }
1995         xs_reclassify_socket(family, sock);
1996
1997         if (reuseport)
1998                 xs_sock_set_reuseport(sock);
1999
2000         err = xs_bind(transport, sock);
2001         if (err) {
2002                 sock_release(sock);
2003                 goto out;
2004         }
2005
2006         return sock;
2007 out:
2008         return ERR_PTR(err);
2009 }
2010
2011 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
2012                                       struct socket *sock)
2013 {
2014         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2015                                                                         xprt);
2016
2017         if (!transport->inet) {
2018                 struct sock *sk = sock->sk;
2019
2020                 write_lock_bh(&sk->sk_callback_lock);
2021
2022                 xs_save_old_callbacks(transport, sk);
2023
2024                 sk->sk_user_data = xprt;
2025                 sk->sk_data_ready = xs_data_ready;
2026                 sk->sk_write_space = xs_udp_write_space;
2027                 sock_set_flag(sk, SOCK_FASYNC);
2028                 sk->sk_error_report = xs_error_report;
2029                 sk->sk_allocation = GFP_NOIO;
2030
2031                 xprt_clear_connected(xprt);
2032
2033                 /* Reset to new socket */
2034                 transport->sock = sock;
2035                 transport->inet = sk;
2036
2037                 write_unlock_bh(&sk->sk_callback_lock);
2038         }
2039
2040         /* Tell the socket layer to start connecting... */
2041         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2042 }
2043
2044 /**
2045  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2046  * @transport: socket transport to connect
2047  */
2048 static int xs_local_setup_socket(struct sock_xprt *transport)
2049 {
2050         struct rpc_xprt *xprt = &transport->xprt;
2051         struct socket *sock;
2052         int status = -EIO;
2053
2054         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2055                                         SOCK_STREAM, 0, &sock, 1);
2056         if (status < 0) {
2057                 dprintk("RPC:       can't create AF_LOCAL "
2058                         "transport socket (%d).\n", -status);
2059                 goto out;
2060         }
2061         xs_reclassify_socket(AF_LOCAL, sock);
2062
2063         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2064                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2065
2066         status = xs_local_finish_connecting(xprt, sock);
2067         trace_rpc_socket_connect(xprt, sock, status);
2068         switch (status) {
2069         case 0:
2070                 dprintk("RPC:       xprt %p connected to %s\n",
2071                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2072                 xprt->stat.connect_count++;
2073                 xprt->stat.connect_time += (long)jiffies -
2074                                            xprt->stat.connect_start;
2075                 xprt_set_connected(xprt);
2076         case -ENOBUFS:
2077                 break;
2078         case -ENOENT:
2079                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2080                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2081                 break;
2082         case -ECONNREFUSED:
2083                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2084                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2085                 break;
2086         default:
2087                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2088                                 __func__, -status,
2089                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2090         }
2091
2092 out:
2093         xprt_clear_connecting(xprt);
2094         xprt_wake_pending_tasks(xprt, status);
2095         return status;
2096 }
2097
2098 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2099 {
2100         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2101         int ret;
2102
2103          if (RPC_IS_ASYNC(task)) {
2104                 /*
2105                  * We want the AF_LOCAL connect to be resolved in the
2106                  * filesystem namespace of the process making the rpc
2107                  * call.  Thus we connect synchronously.
2108                  *
2109                  * If we want to support asynchronous AF_LOCAL calls,
2110                  * we'll need to figure out how to pass a namespace to
2111                  * connect.
2112                  */
2113                 rpc_exit(task, -ENOTCONN);
2114                 return;
2115         }
2116         ret = xs_local_setup_socket(transport);
2117         if (ret && !RPC_IS_SOFTCONN(task))
2118                 msleep_interruptible(15000);
2119 }
2120
2121 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2122 /*
2123  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2124  * know that we have exclusive access to the socket), to guard against
2125  * races with xs_reset_transport.
2126  */
2127 static void xs_set_memalloc(struct rpc_xprt *xprt)
2128 {
2129         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2130                         xprt);
2131
2132         /*
2133          * If there's no sock, then we have nothing to set. The
2134          * reconnecting process will get it for us.
2135          */
2136         if (!transport->inet)
2137                 return;
2138         if (atomic_read(&xprt->swapper))
2139                 sk_set_memalloc(transport->inet);
2140 }
2141
2142 /**
2143  * xs_enable_swap - Tag this transport as being used for swap.
2144  * @xprt: transport to tag
2145  *
2146  * Take a reference to this transport on behalf of the rpc_clnt, and
2147  * optionally mark it for swapping if it wasn't already.
2148  */
2149 static int
2150 xs_enable_swap(struct rpc_xprt *xprt)
2151 {
2152         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2153
2154         if (atomic_inc_return(&xprt->swapper) != 1)
2155                 return 0;
2156         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2157                 return -ERESTARTSYS;
2158         if (xs->inet)
2159                 sk_set_memalloc(xs->inet);
2160         xprt_release_xprt(xprt, NULL);
2161         return 0;
2162 }
2163
2164 /**
2165  * xs_disable_swap - Untag this transport as being used for swap.
2166  * @xprt: transport to tag
2167  *
2168  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2169  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2170  */
2171 static void
2172 xs_disable_swap(struct rpc_xprt *xprt)
2173 {
2174         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2175
2176         if (!atomic_dec_and_test(&xprt->swapper))
2177                 return;
2178         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2179                 return;
2180         if (xs->inet)
2181                 sk_clear_memalloc(xs->inet);
2182         xprt_release_xprt(xprt, NULL);
2183 }
2184 #else
2185 static void xs_set_memalloc(struct rpc_xprt *xprt)
2186 {
2187 }
2188
2189 static int
2190 xs_enable_swap(struct rpc_xprt *xprt)
2191 {
2192         return -EINVAL;
2193 }
2194
2195 static void
2196 xs_disable_swap(struct rpc_xprt *xprt)
2197 {
2198 }
2199 #endif
2200
2201 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2202 {
2203         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2204
2205         if (!transport->inet) {
2206                 struct sock *sk = sock->sk;
2207
2208                 write_lock_bh(&sk->sk_callback_lock);
2209
2210                 xs_save_old_callbacks(transport, sk);
2211
2212                 sk->sk_user_data = xprt;
2213                 sk->sk_data_ready = xs_data_ready;
2214                 sk->sk_write_space = xs_udp_write_space;
2215                 sock_set_flag(sk, SOCK_FASYNC);
2216                 sk->sk_allocation = GFP_NOIO;
2217
2218                 xprt_set_connected(xprt);
2219
2220                 /* Reset to new socket */
2221                 transport->sock = sock;
2222                 transport->inet = sk;
2223
2224                 xs_set_memalloc(xprt);
2225
2226                 write_unlock_bh(&sk->sk_callback_lock);
2227         }
2228         xs_udp_do_set_buffer_size(xprt);
2229
2230         xprt->stat.connect_start = jiffies;
2231 }
2232
2233 static void xs_udp_setup_socket(struct work_struct *work)
2234 {
2235         struct sock_xprt *transport =
2236                 container_of(work, struct sock_xprt, connect_worker.work);
2237         struct rpc_xprt *xprt = &transport->xprt;
2238         struct socket *sock;
2239         int status = -EIO;
2240
2241         sock = xs_create_sock(xprt, transport,
2242                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2243                         IPPROTO_UDP, false);
2244         if (IS_ERR(sock))
2245                 goto out;
2246
2247         dprintk("RPC:       worker connecting xprt %p via %s to "
2248                                 "%s (port %s)\n", xprt,
2249                         xprt->address_strings[RPC_DISPLAY_PROTO],
2250                         xprt->address_strings[RPC_DISPLAY_ADDR],
2251                         xprt->address_strings[RPC_DISPLAY_PORT]);
2252
2253         xs_udp_finish_connecting(xprt, sock);
2254         trace_rpc_socket_connect(xprt, sock, 0);
2255         status = 0;
2256 out:
2257         xprt_clear_connecting(xprt);
2258         xprt_unlock_connect(xprt, transport);
2259         xprt_wake_pending_tasks(xprt, status);
2260 }
2261
2262 /**
2263  * xs_tcp_shutdown - gracefully shut down a TCP socket
2264  * @xprt: transport
2265  *
2266  * Initiates a graceful shutdown of the TCP socket by calling the
2267  * equivalent of shutdown(SHUT_RDWR);
2268  */
2269 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2270 {
2271         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2272         struct socket *sock = transport->sock;
2273         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2274
2275         if (sock == NULL)
2276                 return;
2277         switch (skst) {
2278         default:
2279                 kernel_sock_shutdown(sock, SHUT_RDWR);
2280                 trace_rpc_socket_shutdown(xprt, sock);
2281                 break;
2282         case TCP_CLOSE:
2283         case TCP_TIME_WAIT:
2284                 xs_reset_transport(transport);
2285         }
2286 }
2287
2288 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2289                 struct socket *sock)
2290 {
2291         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2292         unsigned int keepidle;
2293         unsigned int keepcnt;
2294         unsigned int opt_on = 1;
2295         unsigned int timeo;
2296
2297         spin_lock_bh(&xprt->transport_lock);
2298         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2299         keepcnt = xprt->timeout->to_retries + 1;
2300         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2301                 (xprt->timeout->to_retries + 1);
2302         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2303         spin_unlock_bh(&xprt->transport_lock);
2304
2305         /* TCP Keepalive options */
2306         kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2307                         (char *)&opt_on, sizeof(opt_on));
2308         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2309                         (char *)&keepidle, sizeof(keepidle));
2310         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2311                         (char *)&keepidle, sizeof(keepidle));
2312         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2313                         (char *)&keepcnt, sizeof(keepcnt));
2314
2315         /* TCP user timeout (see RFC5482) */
2316         kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2317                         (char *)&timeo, sizeof(timeo));
2318 }
2319
2320 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2321                 unsigned long connect_timeout,
2322                 unsigned long reconnect_timeout)
2323 {
2324         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2325         struct rpc_timeout to;
2326         unsigned long initval;
2327
2328         spin_lock_bh(&xprt->transport_lock);
2329         if (reconnect_timeout < xprt->max_reconnect_timeout)
2330                 xprt->max_reconnect_timeout = reconnect_timeout;
2331         if (connect_timeout < xprt->connect_timeout) {
2332                 memcpy(&to, xprt->timeout, sizeof(to));
2333                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2334                 /* Arbitrary lower limit */
2335                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2336                         initval = XS_TCP_INIT_REEST_TO << 1;
2337                 to.to_initval = initval;
2338                 to.to_maxval = initval;
2339                 memcpy(&transport->tcp_timeout, &to,
2340                                 sizeof(transport->tcp_timeout));
2341                 xprt->timeout = &transport->tcp_timeout;
2342                 xprt->connect_timeout = connect_timeout;
2343         }
2344         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2345         spin_unlock_bh(&xprt->transport_lock);
2346 }
2347
2348 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2349 {
2350         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2351         int ret = -ENOTCONN;
2352
2353         if (!transport->inet) {
2354                 struct sock *sk = sock->sk;
2355                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2356
2357                 /* Avoid temporary address, they are bad for long-lived
2358                  * connections such as NFS mounts.
2359                  * RFC4941, section 3.6 suggests that:
2360                  *    Individual applications, which have specific
2361                  *    knowledge about the normal duration of connections,
2362                  *    MAY override this as appropriate.
2363                  */
2364                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2365                                 (char *)&addr_pref, sizeof(addr_pref));
2366
2367                 xs_tcp_set_socket_timeouts(xprt, sock);
2368
2369                 write_lock_bh(&sk->sk_callback_lock);
2370
2371                 xs_save_old_callbacks(transport, sk);
2372
2373                 sk->sk_user_data = xprt;
2374                 sk->sk_data_ready = xs_data_ready;
2375                 sk->sk_state_change = xs_tcp_state_change;
2376                 sk->sk_write_space = xs_tcp_write_space;
2377                 sock_set_flag(sk, SOCK_FASYNC);
2378                 sk->sk_error_report = xs_error_report;
2379                 sk->sk_allocation = GFP_NOIO;
2380
2381                 /* socket options */
2382                 sock_reset_flag(sk, SOCK_LINGER);
2383                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2384
2385                 xprt_clear_connected(xprt);
2386
2387                 /* Reset to new socket */
2388                 transport->sock = sock;
2389                 transport->inet = sk;
2390
2391                 write_unlock_bh(&sk->sk_callback_lock);
2392         }
2393
2394         if (!xprt_bound(xprt))
2395                 goto out;
2396
2397         xs_set_memalloc(xprt);
2398
2399         /* Tell the socket layer to start connecting... */
2400         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2401         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2402         switch (ret) {
2403         case 0:
2404                 xs_set_srcport(transport, sock);
2405                 /* fall through */
2406         case -EINPROGRESS:
2407                 /* SYN_SENT! */
2408                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2409                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2410                 break;
2411         case -EADDRNOTAVAIL:
2412                 /* Source port number is unavailable. Try a new one! */
2413                 transport->srcport = 0;
2414         }
2415 out:
2416         return ret;
2417 }
2418
2419 /**
2420  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2421  *
2422  * Invoked by a work queue tasklet.
2423  */
2424 static void xs_tcp_setup_socket(struct work_struct *work)
2425 {
2426         struct sock_xprt *transport =
2427                 container_of(work, struct sock_xprt, connect_worker.work);
2428         struct socket *sock = transport->sock;
2429         struct rpc_xprt *xprt = &transport->xprt;
2430         int status = -EIO;
2431
2432         if (!sock) {
2433                 sock = xs_create_sock(xprt, transport,
2434                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2435                                 IPPROTO_TCP, true);
2436                 if (IS_ERR(sock)) {
2437                         status = PTR_ERR(sock);
2438                         goto out;
2439                 }
2440         }
2441
2442         dprintk("RPC:       worker connecting xprt %p via %s to "
2443                                 "%s (port %s)\n", xprt,
2444                         xprt->address_strings[RPC_DISPLAY_PROTO],
2445                         xprt->address_strings[RPC_DISPLAY_ADDR],
2446                         xprt->address_strings[RPC_DISPLAY_PORT]);
2447
2448         status = xs_tcp_finish_connecting(xprt, sock);
2449         trace_rpc_socket_connect(xprt, sock, status);
2450         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2451                         xprt, -status, xprt_connected(xprt),
2452                         sock->sk->sk_state);
2453         switch (status) {
2454         default:
2455                 printk("%s: connect returned unhandled error %d\n",
2456                         __func__, status);
2457                 /* fall through */
2458         case -EADDRNOTAVAIL:
2459                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2460                  * and retry
2461                  */
2462                 xs_tcp_force_close(xprt);
2463                 break;
2464         case 0:
2465         case -EINPROGRESS:
2466         case -EALREADY:
2467                 xprt_unlock_connect(xprt, transport);
2468                 return;
2469         case -EINVAL:
2470                 /* Happens, for instance, if the user specified a link
2471                  * local IPv6 address without a scope-id.
2472                  */
2473         case -ECONNREFUSED:
2474         case -ECONNRESET:
2475         case -ENETDOWN:
2476         case -ENETUNREACH:
2477         case -EHOSTUNREACH:
2478         case -EADDRINUSE:
2479         case -ENOBUFS:
2480                 /*
2481                  * xs_tcp_force_close() wakes tasks with -EIO.
2482                  * We need to wake them first to ensure the
2483                  * correct error code.
2484                  */
2485                 xprt_wake_pending_tasks(xprt, status);
2486                 xs_tcp_force_close(xprt);
2487                 goto out;
2488         }
2489         status = -EAGAIN;
2490 out:
2491         xprt_clear_connecting(xprt);
2492         xprt_unlock_connect(xprt, transport);
2493         xprt_wake_pending_tasks(xprt, status);
2494 }
2495
2496 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2497 {
2498         unsigned long start, now = jiffies;
2499
2500         start = xprt->stat.connect_start + xprt->reestablish_timeout;
2501         if (time_after(start, now))
2502                 return start - now;
2503         return 0;
2504 }
2505
2506 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2507 {
2508         xprt->reestablish_timeout <<= 1;
2509         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2510                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2511         if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2512                 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2513 }
2514
2515 /**
2516  * xs_connect - connect a socket to a remote endpoint
2517  * @xprt: pointer to transport structure
2518  * @task: address of RPC task that manages state of connect request
2519  *
2520  * TCP: If the remote end dropped the connection, delay reconnecting.
2521  *
2522  * UDP socket connects are synchronous, but we use a work queue anyway
2523  * to guarantee that even unprivileged user processes can set up a
2524  * socket on a privileged port.
2525  *
2526  * If a UDP socket connect fails, the delay behavior here prevents
2527  * retry floods (hard mounts).
2528  */
2529 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2530 {
2531         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2532         unsigned long delay = 0;
2533
2534         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2535
2536         if (transport->sock != NULL) {
2537                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2538                                 "seconds\n",
2539                                 xprt, xprt->reestablish_timeout / HZ);
2540
2541                 /* Start by resetting any existing state */
2542                 xs_reset_transport(transport);
2543
2544                 delay = xs_reconnect_delay(xprt);
2545                 xs_reconnect_backoff(xprt);
2546
2547         } else
2548                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2549
2550         queue_delayed_work(xprtiod_workqueue,
2551                         &transport->connect_worker,
2552                         delay);
2553 }
2554
2555 /**
2556  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2557  * @xprt: rpc_xprt struct containing statistics
2558  * @seq: output file
2559  *
2560  */
2561 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2562 {
2563         long idle_time = 0;
2564
2565         if (xprt_connected(xprt))
2566                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2567
2568         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2569                         "%llu %llu %lu %llu %llu\n",
2570                         xprt->stat.bind_count,
2571                         xprt->stat.connect_count,
2572                         xprt->stat.connect_time,
2573                         idle_time,
2574                         xprt->stat.sends,
2575                         xprt->stat.recvs,
2576                         xprt->stat.bad_xids,
2577                         xprt->stat.req_u,
2578                         xprt->stat.bklog_u,
2579                         xprt->stat.max_slots,
2580                         xprt->stat.sending_u,
2581                         xprt->stat.pending_u);
2582 }
2583
2584 /**
2585  * xs_udp_print_stats - display UDP socket-specifc stats
2586  * @xprt: rpc_xprt struct containing statistics
2587  * @seq: output file
2588  *
2589  */
2590 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2591 {
2592         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2593
2594         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2595                         "%lu %llu %llu\n",
2596                         transport->srcport,
2597                         xprt->stat.bind_count,
2598                         xprt->stat.sends,
2599                         xprt->stat.recvs,
2600                         xprt->stat.bad_xids,
2601                         xprt->stat.req_u,
2602                         xprt->stat.bklog_u,
2603                         xprt->stat.max_slots,
2604                         xprt->stat.sending_u,
2605                         xprt->stat.pending_u);
2606 }
2607
2608 /**
2609  * xs_tcp_print_stats - display TCP socket-specifc stats
2610  * @xprt: rpc_xprt struct containing statistics
2611  * @seq: output file
2612  *
2613  */
2614 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2615 {
2616         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2617         long idle_time = 0;
2618
2619         if (xprt_connected(xprt))
2620                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2621
2622         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2623                         "%llu %llu %lu %llu %llu\n",
2624                         transport->srcport,
2625                         xprt->stat.bind_count,
2626                         xprt->stat.connect_count,
2627                         xprt->stat.connect_time,
2628                         idle_time,
2629                         xprt->stat.sends,
2630                         xprt->stat.recvs,
2631                         xprt->stat.bad_xids,
2632                         xprt->stat.req_u,
2633                         xprt->stat.bklog_u,
2634                         xprt->stat.max_slots,
2635                         xprt->stat.sending_u,
2636                         xprt->stat.pending_u);
2637 }
2638
2639 /*
2640  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2641  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2642  * to use the server side send routines.
2643  */
2644 static int bc_malloc(struct rpc_task *task)
2645 {
2646         struct rpc_rqst *rqst = task->tk_rqstp;
2647         size_t size = rqst->rq_callsize;
2648         struct page *page;
2649         struct rpc_buffer *buf;
2650
2651         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2652                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2653                           size);
2654                 return -EINVAL;
2655         }
2656
2657         page = alloc_page(GFP_KERNEL);
2658         if (!page)
2659                 return -ENOMEM;
2660
2661         buf = page_address(page);
2662         buf->len = PAGE_SIZE;
2663
2664         rqst->rq_buffer = buf->data;
2665         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2666         return 0;
2667 }
2668
2669 /*
2670  * Free the space allocated in the bc_alloc routine
2671  */
2672 static void bc_free(struct rpc_task *task)
2673 {
2674         void *buffer = task->tk_rqstp->rq_buffer;
2675         struct rpc_buffer *buf;
2676
2677         buf = container_of(buffer, struct rpc_buffer, data);
2678         free_page((unsigned long)buf);
2679 }
2680
2681 /*
2682  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2683  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2684  */
2685 static int bc_sendto(struct rpc_rqst *req)
2686 {
2687         int len;
2688         struct xdr_buf *xbufp = &req->rq_snd_buf;
2689         struct rpc_xprt *xprt = req->rq_xprt;
2690         struct sock_xprt *transport =
2691                                 container_of(xprt, struct sock_xprt, xprt);
2692         struct socket *sock = transport->sock;
2693         unsigned long headoff;
2694         unsigned long tailoff;
2695
2696         xs_encode_stream_record_marker(xbufp);
2697
2698         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2699         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2700         len = svc_send_common(sock, xbufp,
2701                               virt_to_page(xbufp->head[0].iov_base), headoff,
2702                               xbufp->tail[0].iov_base, tailoff);
2703
2704         if (len != xbufp->len) {
2705                 printk(KERN_NOTICE "Error sending entire callback!\n");
2706                 len = -EAGAIN;
2707         }
2708
2709         return len;
2710 }
2711
2712 /*
2713  * The send routine. Borrows from svc_send
2714  */
2715 static int bc_send_request(struct rpc_task *task)
2716 {
2717         struct rpc_rqst *req = task->tk_rqstp;
2718         struct svc_xprt *xprt;
2719         int len;
2720
2721         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2722         /*
2723          * Get the server socket associated with this callback xprt
2724          */
2725         xprt = req->rq_xprt->bc_xprt;
2726
2727         /*
2728          * Grab the mutex to serialize data as the connection is shared
2729          * with the fore channel
2730          */
2731         if (!mutex_trylock(&xprt->xpt_mutex)) {
2732                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2733                 if (!mutex_trylock(&xprt->xpt_mutex))
2734                         return -EAGAIN;
2735                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2736         }
2737         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2738                 len = -ENOTCONN;
2739         else
2740                 len = bc_sendto(req);
2741         mutex_unlock(&xprt->xpt_mutex);
2742
2743         if (len > 0)
2744                 len = 0;
2745
2746         return len;
2747 }
2748
2749 /*
2750  * The close routine. Since this is client initiated, we do nothing
2751  */
2752
2753 static void bc_close(struct rpc_xprt *xprt)
2754 {
2755 }
2756
2757 /*
2758  * The xprt destroy routine. Again, because this connection is client
2759  * initiated, we do nothing
2760  */
2761
2762 static void bc_destroy(struct rpc_xprt *xprt)
2763 {
2764         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2765
2766         xs_xprt_free(xprt);
2767         module_put(THIS_MODULE);
2768 }
2769
2770 static const struct rpc_xprt_ops xs_local_ops = {
2771         .reserve_xprt           = xprt_reserve_xprt,
2772         .release_xprt           = xs_tcp_release_xprt,
2773         .alloc_slot             = xprt_alloc_slot,
2774         .free_slot              = xprt_free_slot,
2775         .rpcbind                = xs_local_rpcbind,
2776         .set_port               = xs_local_set_port,
2777         .connect                = xs_local_connect,
2778         .buf_alloc              = rpc_malloc,
2779         .buf_free               = rpc_free,
2780         .send_request           = xs_local_send_request,
2781         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2782         .close                  = xs_close,
2783         .destroy                = xs_destroy,
2784         .print_stats            = xs_local_print_stats,
2785         .enable_swap            = xs_enable_swap,
2786         .disable_swap           = xs_disable_swap,
2787 };
2788
2789 static const struct rpc_xprt_ops xs_udp_ops = {
2790         .set_buffer_size        = xs_udp_set_buffer_size,
2791         .reserve_xprt           = xprt_reserve_xprt_cong,
2792         .release_xprt           = xprt_release_xprt_cong,
2793         .alloc_slot             = xprt_alloc_slot,
2794         .free_slot              = xprt_free_slot,
2795         .rpcbind                = rpcb_getport_async,
2796         .set_port               = xs_set_port,
2797         .connect                = xs_connect,
2798         .buf_alloc              = rpc_malloc,
2799         .buf_free               = rpc_free,
2800         .send_request           = xs_udp_send_request,
2801         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2802         .timer                  = xs_udp_timer,
2803         .release_request        = xprt_release_rqst_cong,
2804         .close                  = xs_close,
2805         .destroy                = xs_destroy,
2806         .print_stats            = xs_udp_print_stats,
2807         .enable_swap            = xs_enable_swap,
2808         .disable_swap           = xs_disable_swap,
2809         .inject_disconnect      = xs_inject_disconnect,
2810 };
2811
2812 static const struct rpc_xprt_ops xs_tcp_ops = {
2813         .reserve_xprt           = xprt_reserve_xprt,
2814         .release_xprt           = xs_tcp_release_xprt,
2815         .alloc_slot             = xprt_lock_and_alloc_slot,
2816         .free_slot              = xprt_free_slot,
2817         .rpcbind                = rpcb_getport_async,
2818         .set_port               = xs_set_port,
2819         .connect                = xs_connect,
2820         .buf_alloc              = rpc_malloc,
2821         .buf_free               = rpc_free,
2822         .send_request           = xs_tcp_send_request,
2823         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2824         .close                  = xs_tcp_shutdown,
2825         .destroy                = xs_destroy,
2826         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2827         .print_stats            = xs_tcp_print_stats,
2828         .enable_swap            = xs_enable_swap,
2829         .disable_swap           = xs_disable_swap,
2830         .inject_disconnect      = xs_inject_disconnect,
2831 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2832         .bc_setup               = xprt_setup_bc,
2833         .bc_up                  = xs_tcp_bc_up,
2834         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2835         .bc_free_rqst           = xprt_free_bc_rqst,
2836         .bc_destroy             = xprt_destroy_bc,
2837 #endif
2838 };
2839
2840 /*
2841  * The rpc_xprt_ops for the server backchannel
2842  */
2843
2844 static const struct rpc_xprt_ops bc_tcp_ops = {
2845         .reserve_xprt           = xprt_reserve_xprt,
2846         .release_xprt           = xprt_release_xprt,
2847         .alloc_slot             = xprt_alloc_slot,
2848         .free_slot              = xprt_free_slot,
2849         .buf_alloc              = bc_malloc,
2850         .buf_free               = bc_free,
2851         .send_request           = bc_send_request,
2852         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2853         .close                  = bc_close,
2854         .destroy                = bc_destroy,
2855         .print_stats            = xs_tcp_print_stats,
2856         .enable_swap            = xs_enable_swap,
2857         .disable_swap           = xs_disable_swap,
2858         .inject_disconnect      = xs_inject_disconnect,
2859 };
2860
2861 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2862 {
2863         static const struct sockaddr_in sin = {
2864                 .sin_family             = AF_INET,
2865                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2866         };
2867         static const struct sockaddr_in6 sin6 = {
2868                 .sin6_family            = AF_INET6,
2869                 .sin6_addr              = IN6ADDR_ANY_INIT,
2870         };
2871
2872         switch (family) {
2873         case AF_LOCAL:
2874                 break;
2875         case AF_INET:
2876                 memcpy(sap, &sin, sizeof(sin));
2877                 break;
2878         case AF_INET6:
2879                 memcpy(sap, &sin6, sizeof(sin6));
2880                 break;
2881         default:
2882                 dprintk("RPC:       %s: Bad address family\n", __func__);
2883                 return -EAFNOSUPPORT;
2884         }
2885         return 0;
2886 }
2887
2888 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2889                                       unsigned int slot_table_size,
2890                                       unsigned int max_slot_table_size)
2891 {
2892         struct rpc_xprt *xprt;
2893         struct sock_xprt *new;
2894
2895         if (args->addrlen > sizeof(xprt->addr)) {
2896                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2897                 return ERR_PTR(-EBADF);
2898         }
2899
2900         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2901                         max_slot_table_size);
2902         if (xprt == NULL) {
2903                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2904                                 "rpc_xprt\n");
2905                 return ERR_PTR(-ENOMEM);
2906         }
2907
2908         new = container_of(xprt, struct sock_xprt, xprt);
2909         mutex_init(&new->recv_mutex);
2910         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2911         xprt->addrlen = args->addrlen;
2912         if (args->srcaddr)
2913                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2914         else {
2915                 int err;
2916                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2917                                         (struct sockaddr *)&new->srcaddr);
2918                 if (err != 0) {
2919                         xprt_free(xprt);
2920                         return ERR_PTR(err);
2921                 }
2922         }
2923
2924         return xprt;
2925 }
2926
2927 static const struct rpc_timeout xs_local_default_timeout = {
2928         .to_initval = 10 * HZ,
2929         .to_maxval = 10 * HZ,
2930         .to_retries = 2,
2931 };
2932
2933 /**
2934  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2935  * @args: rpc transport creation arguments
2936  *
2937  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2938  */
2939 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2940 {
2941         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2942         struct sock_xprt *transport;
2943         struct rpc_xprt *xprt;
2944         struct rpc_xprt *ret;
2945
2946         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2947                         xprt_max_tcp_slot_table_entries);
2948         if (IS_ERR(xprt))
2949                 return xprt;
2950         transport = container_of(xprt, struct sock_xprt, xprt);
2951
2952         xprt->prot = 0;
2953         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2954         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2955
2956         xprt->bind_timeout = XS_BIND_TO;
2957         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2958         xprt->idle_timeout = XS_IDLE_DISC_TO;
2959
2960         xprt->ops = &xs_local_ops;
2961         xprt->timeout = &xs_local_default_timeout;
2962
2963         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2964         INIT_DELAYED_WORK(&transport->connect_worker,
2965                         xs_dummy_setup_socket);
2966
2967         switch (sun->sun_family) {
2968         case AF_LOCAL:
2969                 if (sun->sun_path[0] != '/') {
2970                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2971                                         sun->sun_path);
2972                         ret = ERR_PTR(-EINVAL);
2973                         goto out_err;
2974                 }
2975                 xprt_set_bound(xprt);
2976                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2977                 ret = ERR_PTR(xs_local_setup_socket(transport));
2978                 if (ret)
2979                         goto out_err;
2980                 break;
2981         default:
2982                 ret = ERR_PTR(-EAFNOSUPPORT);
2983                 goto out_err;
2984         }
2985
2986         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2987                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2988
2989         if (try_module_get(THIS_MODULE))
2990                 return xprt;
2991         ret = ERR_PTR(-EINVAL);
2992 out_err:
2993         xs_xprt_free(xprt);
2994         return ret;
2995 }
2996
2997 static const struct rpc_timeout xs_udp_default_timeout = {
2998         .to_initval = 5 * HZ,
2999         .to_maxval = 30 * HZ,
3000         .to_increment = 5 * HZ,
3001         .to_retries = 5,
3002 };
3003
3004 /**
3005  * xs_setup_udp - Set up transport to use a UDP socket
3006  * @args: rpc transport creation arguments
3007  *
3008  */
3009 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3010 {
3011         struct sockaddr *addr = args->dstaddr;
3012         struct rpc_xprt *xprt;
3013         struct sock_xprt *transport;
3014         struct rpc_xprt *ret;
3015
3016         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3017                         xprt_udp_slot_table_entries);
3018         if (IS_ERR(xprt))
3019                 return xprt;
3020         transport = container_of(xprt, struct sock_xprt, xprt);
3021
3022         xprt->prot = IPPROTO_UDP;
3023         xprt->tsh_size = 0;
3024         /* XXX: header size can vary due to auth type, IPv6, etc. */
3025         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3026
3027         xprt->bind_timeout = XS_BIND_TO;
3028         xprt->reestablish_timeout = XS_UDP_REEST_TO;
3029         xprt->idle_timeout = XS_IDLE_DISC_TO;
3030
3031         xprt->ops = &xs_udp_ops;
3032
3033         xprt->timeout = &xs_udp_default_timeout;
3034
3035         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3036         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3037
3038         switch (addr->sa_family) {
3039         case AF_INET:
3040                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3041                         xprt_set_bound(xprt);
3042
3043                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3044                 break;
3045         case AF_INET6:
3046                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3047                         xprt_set_bound(xprt);
3048
3049                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3050                 break;
3051         default:
3052                 ret = ERR_PTR(-EAFNOSUPPORT);
3053                 goto out_err;
3054         }
3055
3056         if (xprt_bound(xprt))
3057                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3058                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3059                                 xprt->address_strings[RPC_DISPLAY_PORT],
3060                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3061         else
3062                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3063                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3064                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3065
3066         if (try_module_get(THIS_MODULE))
3067                 return xprt;
3068         ret = ERR_PTR(-EINVAL);
3069 out_err:
3070         xs_xprt_free(xprt);
3071         return ret;
3072 }
3073
3074 static const struct rpc_timeout xs_tcp_default_timeout = {
3075         .to_initval = 60 * HZ,
3076         .to_maxval = 60 * HZ,
3077         .to_retries = 2,
3078 };
3079
3080 /**
3081  * xs_setup_tcp - Set up transport to use a TCP socket
3082  * @args: rpc transport creation arguments
3083  *
3084  */
3085 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3086 {
3087         struct sockaddr *addr = args->dstaddr;
3088         struct rpc_xprt *xprt;
3089         struct sock_xprt *transport;
3090         struct rpc_xprt *ret;
3091         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3092
3093         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3094                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3095
3096         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3097                         max_slot_table_size);
3098         if (IS_ERR(xprt))
3099                 return xprt;
3100         transport = container_of(xprt, struct sock_xprt, xprt);
3101
3102         xprt->prot = IPPROTO_TCP;
3103         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3104         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3105
3106         xprt->bind_timeout = XS_BIND_TO;
3107         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3108         xprt->idle_timeout = XS_IDLE_DISC_TO;
3109
3110         xprt->ops = &xs_tcp_ops;
3111         xprt->timeout = &xs_tcp_default_timeout;
3112
3113         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3114         xprt->connect_timeout = xprt->timeout->to_initval *
3115                 (xprt->timeout->to_retries + 1);
3116
3117         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3118         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3119
3120         switch (addr->sa_family) {
3121         case AF_INET:
3122                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3123                         xprt_set_bound(xprt);
3124
3125                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3126                 break;
3127         case AF_INET6:
3128                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3129                         xprt_set_bound(xprt);
3130
3131                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3132                 break;
3133         default:
3134                 ret = ERR_PTR(-EAFNOSUPPORT);
3135                 goto out_err;
3136         }
3137
3138         if (xprt_bound(xprt))
3139                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3140                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3141                                 xprt->address_strings[RPC_DISPLAY_PORT],
3142                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3143         else
3144                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3145                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3146                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3147
3148         if (try_module_get(THIS_MODULE))
3149                 return xprt;
3150         ret = ERR_PTR(-EINVAL);
3151 out_err:
3152         xs_xprt_free(xprt);
3153         return ret;
3154 }
3155
3156 /**
3157  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3158  * @args: rpc transport creation arguments
3159  *
3160  */
3161 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3162 {
3163         struct sockaddr *addr = args->dstaddr;
3164         struct rpc_xprt *xprt;
3165         struct sock_xprt *transport;
3166         struct svc_sock *bc_sock;
3167         struct rpc_xprt *ret;
3168
3169         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3170                         xprt_tcp_slot_table_entries);
3171         if (IS_ERR(xprt))
3172                 return xprt;
3173         transport = container_of(xprt, struct sock_xprt, xprt);
3174
3175         xprt->prot = IPPROTO_TCP;
3176         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3177         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3178         xprt->timeout = &xs_tcp_default_timeout;
3179
3180         /* backchannel */
3181         xprt_set_bound(xprt);
3182         xprt->bind_timeout = 0;
3183         xprt->reestablish_timeout = 0;
3184         xprt->idle_timeout = 0;
3185
3186         xprt->ops = &bc_tcp_ops;
3187
3188         switch (addr->sa_family) {
3189         case AF_INET:
3190                 xs_format_peer_addresses(xprt, "tcp",
3191                                          RPCBIND_NETID_TCP);
3192                 break;
3193         case AF_INET6:
3194                 xs_format_peer_addresses(xprt, "tcp",
3195                                    RPCBIND_NETID_TCP6);
3196                 break;
3197         default:
3198                 ret = ERR_PTR(-EAFNOSUPPORT);
3199                 goto out_err;
3200         }
3201
3202         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3203                         xprt->address_strings[RPC_DISPLAY_ADDR],
3204                         xprt->address_strings[RPC_DISPLAY_PORT],
3205                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3206
3207         /*
3208          * Once we've associated a backchannel xprt with a connection,
3209          * we want to keep it around as long as the connection lasts,
3210          * in case we need to start using it for a backchannel again;
3211          * this reference won't be dropped until bc_xprt is destroyed.
3212          */
3213         xprt_get(xprt);
3214         args->bc_xprt->xpt_bc_xprt = xprt;
3215         xprt->bc_xprt = args->bc_xprt;
3216         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3217         transport->sock = bc_sock->sk_sock;
3218         transport->inet = bc_sock->sk_sk;
3219
3220         /*
3221          * Since we don't want connections for the backchannel, we set
3222          * the xprt status to connected
3223          */
3224         xprt_set_connected(xprt);
3225
3226         if (try_module_get(THIS_MODULE))
3227                 return xprt;
3228
3229         args->bc_xprt->xpt_bc_xprt = NULL;
3230         args->bc_xprt->xpt_bc_xps = NULL;
3231         xprt_put(xprt);
3232         ret = ERR_PTR(-EINVAL);
3233 out_err:
3234         xs_xprt_free(xprt);
3235         return ret;
3236 }
3237
3238 static struct xprt_class        xs_local_transport = {
3239         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3240         .name           = "named UNIX socket",
3241         .owner          = THIS_MODULE,
3242         .ident          = XPRT_TRANSPORT_LOCAL,
3243         .setup          = xs_setup_local,
3244         .netid          = { "" },
3245 };
3246
3247 static struct xprt_class        xs_udp_transport = {
3248         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3249         .name           = "udp",
3250         .owner          = THIS_MODULE,
3251         .ident          = XPRT_TRANSPORT_UDP,
3252         .setup          = xs_setup_udp,
3253         .netid          = { "udp", "udp6", "" },
3254 };
3255
3256 static struct xprt_class        xs_tcp_transport = {
3257         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3258         .name           = "tcp",
3259         .owner          = THIS_MODULE,
3260         .ident          = XPRT_TRANSPORT_TCP,
3261         .setup          = xs_setup_tcp,
3262         .netid          = { "tcp", "tcp6", "" },
3263 };
3264
3265 static struct xprt_class        xs_bc_tcp_transport = {
3266         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3267         .name           = "tcp NFSv4.1 backchannel",
3268         .owner          = THIS_MODULE,
3269         .ident          = XPRT_TRANSPORT_BC_TCP,
3270         .setup          = xs_setup_bc_tcp,
3271         .netid          = { "" },
3272 };
3273
3274 /**
3275  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3276  *
3277  */
3278 int init_socket_xprt(void)
3279 {
3280 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3281         if (!sunrpc_table_header)
3282                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3283 #endif
3284
3285         xprt_register_transport(&xs_local_transport);
3286         xprt_register_transport(&xs_udp_transport);
3287         xprt_register_transport(&xs_tcp_transport);
3288         xprt_register_transport(&xs_bc_tcp_transport);
3289
3290         return 0;
3291 }
3292
3293 /**
3294  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3295  *
3296  */
3297 void cleanup_socket_xprt(void)
3298 {
3299 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3300         if (sunrpc_table_header) {
3301                 unregister_sysctl_table(sunrpc_table_header);
3302                 sunrpc_table_header = NULL;
3303         }
3304 #endif
3305
3306         xprt_unregister_transport(&xs_local_transport);
3307         xprt_unregister_transport(&xs_udp_transport);
3308         xprt_unregister_transport(&xs_tcp_transport);
3309         xprt_unregister_transport(&xs_bc_tcp_transport);
3310 }
3311
3312 static int param_set_uint_minmax(const char *val,
3313                 const struct kernel_param *kp,
3314                 unsigned int min, unsigned int max)
3315 {
3316         unsigned int num;
3317         int ret;
3318
3319         if (!val)
3320                 return -EINVAL;
3321         ret = kstrtouint(val, 0, &num);
3322         if (ret)
3323                 return ret;
3324         if (num < min || num > max)
3325                 return -EINVAL;
3326         *((unsigned int *)kp->arg) = num;
3327         return 0;
3328 }
3329
3330 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3331 {
3332         return param_set_uint_minmax(val, kp,
3333                         RPC_MIN_RESVPORT,
3334                         RPC_MAX_RESVPORT);
3335 }
3336
3337 static const struct kernel_param_ops param_ops_portnr = {
3338         .set = param_set_portnr,
3339         .get = param_get_uint,
3340 };
3341
3342 #define param_check_portnr(name, p) \
3343         __param_check(name, p, unsigned int);
3344
3345 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3346 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3347
3348 static int param_set_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);
3354 }
3355
3356 static const struct kernel_param_ops param_ops_slot_table_size = {
3357         .set = param_set_slot_table_size,
3358         .get = param_get_uint,
3359 };
3360
3361 #define param_check_slot_table_size(name, p) \
3362         __param_check(name, p, unsigned int);
3363
3364 static int param_set_max_slot_table_size(const char *val,
3365                                      const struct kernel_param *kp)
3366 {
3367         return param_set_uint_minmax(val, kp,
3368                         RPC_MIN_SLOT_TABLE,
3369                         RPC_MAX_SLOT_TABLE_LIMIT);
3370 }
3371
3372 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3373         .set = param_set_max_slot_table_size,
3374         .get = param_get_uint,
3375 };
3376
3377 #define param_check_max_slot_table_size(name, p) \
3378         __param_check(name, p, unsigned int);
3379
3380 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3381                    slot_table_size, 0644);
3382 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3383                    max_slot_table_size, 0644);
3384 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3385                    slot_table_size, 0644);