kernel/net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68 #define XS_TCP_LINGER_TO        (15U * HZ)
  69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71 /*
  72  * We can register our own files under /proc/sys/sunrpc by
  73  * calling register_sysctl_table() again.  The files in that
  74  * directory become the union of all files registered there.
  75  *
  76  * We simply need to make sure that we don't collide with
  77  * someone else's file names!
  78  */
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 181
 182 /*
 183  * TCP idle timeout; client drops the transport socket if it is idle
 184  * for this long.  Note that we also timeout UDP sockets to prevent
 185  * holding port numbers when there is no RPC traffic.
 186  */
 187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 188
 189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 190 # undef  RPC_DEBUG_DATA
 191 # define RPCDBG_FACILITY        RPCDBG_TRANS
 192 #endif
 193
 194 #ifdef RPC_DEBUG_DATA
 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196 {
 197         u8 *buf = (u8 *) packet;
 198         int j;
 199
 200         dprintk("RPC:       %s\n", msg);
 201         for (j = 0; j < count && j < 128; j += 4) {
 202                 if (!(j & 31)) {
 203                         if (j)
 204                                 dprintk("\n");
 205                         dprintk("0x%04x ", j);
 206                 }
 207                 dprintk("%02x%02x%02x%02x ",
 208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209         }
 210         dprintk("\n");
 211 }
 212 #else
 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214 {
 215         /* NOP */
 216 }
 217 #endif
 218
 219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 220 {
 221         return (struct rpc_xprt *) sk->sk_user_data;
 222 }
 223
 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 225 {
 226         return (struct sockaddr *) &xprt->addr;
 227 }
 228
 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 230 {
 231         return (struct sockaddr_un *) &xprt->addr;
 232 }
 233
 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr_in *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_in6 *) &xprt->addr;
 242 }
 243
 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 245 {
 246         struct sockaddr *sap = xs_addr(xprt);
 247         struct sockaddr_in6 *sin6;
 248         struct sockaddr_in *sin;
 249         struct sockaddr_un *sun;
 250         char buf[128];
 251
 252         switch (sap->sa_family) {
 253         case AF_LOCAL:
 254                 sun = xs_addr_un(xprt);
 255                 strlcpy(buf, sun->sun_path, sizeof(buf));
 256                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 257                                                 kstrdup(buf, GFP_KERNEL);
 258                 break;
 259         case AF_INET:
 260                 (void)rpc_ntop(sap, buf, sizeof(buf));
 261                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 262                                                 kstrdup(buf, GFP_KERNEL);
 263                 sin = xs_addr_in(xprt);
 264                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 265                 break;
 266         case AF_INET6:
 267                 (void)rpc_ntop(sap, buf, sizeof(buf));
 268                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 269                                                 kstrdup(buf, GFP_KERNEL);
 270                 sin6 = xs_addr_in6(xprt);
 271                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 272                 break;
 273         default:
 274                 BUG();
 275         }
 276
 277         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 278 }
 279
 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 281 {
 282         struct sockaddr *sap = xs_addr(xprt);
 283         char buf[128];
 284
 285         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 286         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 287
 288         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 289         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 290 }
 291
 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 293                                      const char *protocol,
 294                                      const char *netid)
 295 {
 296         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 297         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 298         xs_format_common_peer_addresses(xprt);
 299         xs_format_common_peer_ports(xprt);
 300 }
 301
 302 static void xs_update_peer_port(struct rpc_xprt *xprt)
 303 {
 304         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 305         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 306
 307         xs_format_common_peer_ports(xprt);
 308 }
 309
 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 311 {
 312         unsigned int i;
 313
 314         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 315                 switch (i) {
 316                 case RPC_DISPLAY_PROTO:
 317                 case RPC_DISPLAY_NETID:
 318                         continue;
 319                 default:
 320                         kfree(xprt->address_strings[i]);
 321                 }
 322 }
 323
 324 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 325
 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 327 {
 328         struct msghdr msg = {
 329                 .msg_name       = addr,
 330                 .msg_namelen    = addrlen,
 331                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 332         };
 333         struct kvec iov = {
 334                 .iov_base       = vec->iov_base + base,
 335                 .iov_len        = vec->iov_len - base,
 336         };
 337
 338         if (iov.iov_len != 0)
 339                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 340         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 341 }
 342
 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 344 {
 345         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 346                         int offset, size_t size, int flags);
 347         struct page **ppage;
 348         unsigned int remainder;
 349         int err;
 350
 351         remainder = xdr->page_len - base;
 352         base += xdr->page_base;
 353         ppage = xdr->pages + (base >> PAGE_SHIFT);
 354         base &= ~PAGE_MASK;
 355         do_sendpage = sock->ops->sendpage;
 356         if (!zerocopy)
 357                 do_sendpage = sock_no_sendpage;
 358         for(;;) {
 359                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 360                 int flags = XS_SENDMSG_FLAGS;
 361
 362                 remainder -= len;
 363                 if (remainder != 0 || more)
 364                         flags |= MSG_MORE;
 365                 err = do_sendpage(sock, *ppage, base, len, flags);
 366                 if (remainder == 0 || err != len)
 367                         break;
 368                 *sent_p += err;
 369                 ppage++;
 370                 base = 0;
 371         }
 372         if (err > 0) {
 373                 *sent_p += err;
 374                 err = 0;
 375         }
 376         return err;
 377 }
 378
 379 /**
 380  * xs_sendpages - write pages directly to a socket
 381  * @sock: socket to send on
 382  * @addr: UDP only -- address of destination
 383  * @addrlen: UDP only -- length of destination address
 384  * @xdr: buffer containing this request
 385  * @base: starting position in the buffer
 386  * @zerocopy: true if it is safe to use sendpage()
 387  * @sent_p: return the total number of bytes successfully queued for sending
 388  *
 389  */
 390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 391 {
 392         unsigned int remainder = xdr->len - base;
 393         int err = 0;
 394         int sent = 0;
 395
 396         if (unlikely(!sock))
 397                 return -ENOTSOCK;
 398
 399         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 400         if (base != 0) {
 401                 addr = NULL;
 402                 addrlen = 0;
 403         }
 404
 405         if (base < xdr->head[0].iov_len || addr != NULL) {
 406                 unsigned int len = xdr->head[0].iov_len - base;
 407                 remainder -= len;
 408                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 409                 if (remainder == 0 || err != len)
 410                         goto out;
 411                 *sent_p += err;
 412                 base = 0;
 413         } else
 414                 base -= xdr->head[0].iov_len;
 415
 416         if (base < xdr->page_len) {
 417                 unsigned int len = xdr->page_len - base;
 418                 remainder -= len;
 419                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 420                 *sent_p += sent;
 421                 if (remainder == 0 || sent != len)
 422                         goto out;
 423                 base = 0;
 424         } else
 425                 base -= xdr->page_len;
 426
 427         if (base >= xdr->tail[0].iov_len)
 428                 return 0;
 429         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 430 out:
 431         if (err > 0) {
 432                 *sent_p += err;
 433                 err = 0;
 434         }
 435         return err;
 436 }
 437
 438 static void xs_nospace_callback(struct rpc_task *task)
 439 {
 440         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 441
 442         transport->inet->sk_write_pending--;
 443         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 444 }
 445
 446 /**
 447  * xs_nospace - place task on wait queue if transmit was incomplete
 448  * @task: task to put to sleep
 449  *
 450  */
 451 static int xs_nospace(struct rpc_task *task)
 452 {
 453         struct rpc_rqst *req = task->tk_rqstp;
 454         struct rpc_xprt *xprt = req->rq_xprt;
 455         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 456         struct sock *sk = transport->inet;
 457         int ret = -EAGAIN;
 458
 459         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 460                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 461                         req->rq_slen);
 462
 463         /* Protect against races with write_space */
 464         spin_lock_bh(&xprt->transport_lock);
 465
 466         /* Don't race with disconnect */
 467         if (xprt_connected(xprt)) {
 468                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 469                         /*
 470                          * Notify TCP that we're limited by the application
 471                          * window size
 472                          */
 473                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 474                         sk->sk_write_pending++;
 475                         /* ...and wait for more buffer space */
 476                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 477                 }
 478         } else {
 479                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 480                 ret = -ENOTCONN;
 481         }
 482
 483         spin_unlock_bh(&xprt->transport_lock);
 484
 485         /* Race breaker in case memory is freed before above code is called */
 486         sk->sk_write_space(sk);
 487         return ret;
 488 }
 489
 490 /*
 491  * Construct a stream transport record marker in @buf.
 492  */
 493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 494 {
 495         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 496         rpc_fraghdr *base = buf->head[0].iov_base;
 497         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 498 }
 499
 500 /**
 501  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 502  * @task: RPC task that manages the state of an RPC request
 503  *
 504  * Return values:
 505  *        0:    The request has been sent
 506  *   EAGAIN:    The socket was blocked, please call again later to
 507  *              complete the request
 508  * ENOTCONN:    Caller needs to invoke connect logic then call again
 509  *    other:    Some other error occured, the request was not sent
 510  */
 511 static int xs_local_send_request(struct rpc_task *task)
 512 {
 513         struct rpc_rqst *req = task->tk_rqstp;
 514         struct rpc_xprt *xprt = req->rq_xprt;
 515         struct sock_xprt *transport =
 516                                 container_of(xprt, struct sock_xprt, xprt);
 517         struct xdr_buf *xdr = &req->rq_snd_buf;
 518         int status;
 519         int sent = 0;
 520
 521         xs_encode_stream_record_marker(&req->rq_snd_buf);
 522
 523         xs_pktdump("packet data:",
 524                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 525
 526         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 527                               true, &sent);
 528         dprintk("RPC:       %s(%u) = %d\n",
 529                         __func__, xdr->len - req->rq_bytes_sent, status);
 530         if (likely(sent > 0) || status == 0) {
 531                 req->rq_bytes_sent += sent;
 532                 req->rq_xmit_bytes_sent += sent;
 533                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 534                         req->rq_bytes_sent = 0;
 535                         return 0;
 536                 }
 537                 status = -EAGAIN;
 538         }
 539
 540         switch (status) {
 541         case -ENOBUFS:
 542         case -EAGAIN:
 543                 status = xs_nospace(task);
 544                 break;
 545         default:
 546                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 547                         -status);
 548         case -EPIPE:
 549                 xs_close(xprt);
 550                 status = -ENOTCONN;
 551         }
 552
 553         return status;
 554 }
 555
 556 /**
 557  * xs_udp_send_request - write an RPC request to a UDP socket
 558  * @task: address of RPC task that manages the state of an RPC request
 559  *
 560  * Return values:
 561  *        0:    The request has been sent
 562  *   EAGAIN:    The socket was blocked, please call again later to
 563  *              complete the request
 564  * ENOTCONN:    Caller needs to invoke connect logic then call again
 565  *    other:    Some other error occurred, the request was not sent
 566  */
 567 static int xs_udp_send_request(struct rpc_task *task)
 568 {
 569         struct rpc_rqst *req = task->tk_rqstp;
 570         struct rpc_xprt *xprt = req->rq_xprt;
 571         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 572         struct xdr_buf *xdr = &req->rq_snd_buf;
 573         int sent = 0;
 574         int status;
 575
 576         xs_pktdump("packet data:",
 577                                 req->rq_svec->iov_base,
 578                                 req->rq_svec->iov_len);
 579
 580         if (!xprt_bound(xprt))
 581                 return -ENOTCONN;
 582         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 583                               xdr, req->rq_bytes_sent, true, &sent);
 584
 585         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 586                         xdr->len - req->rq_bytes_sent, status);
 587
 588         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 589         if (status == -EPERM)
 590                 goto process_status;
 591
 592         if (sent > 0 || status == 0) {
 593                 req->rq_xmit_bytes_sent += sent;
 594                 if (sent >= req->rq_slen)
 595                         return 0;
 596                 /* Still some bytes left; set up for a retry later. */
 597                 status = -EAGAIN;
 598         }
 599
 600 process_status:
 601         switch (status) {
 602         case -ENOTSOCK:
 603                 status = -ENOTCONN;
 604                 /* Should we call xs_close() here? */
 605                 break;
 606         case -EAGAIN:
 607                 status = xs_nospace(task);
 608                 break;
 609         default:
 610                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 611                         -status);
 612         case -ENETUNREACH:
 613         case -ENOBUFS:
 614         case -EPIPE:
 615         case -ECONNREFUSED:
 616         case -EPERM:
 617                 /* When the server has died, an ICMP port unreachable message
 618                  * prompts ECONNREFUSED. */
 619                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 620         }
 621
 622         return status;
 623 }
 624
 625 /**
 626  * xs_tcp_shutdown - gracefully shut down a TCP socket
 627  * @xprt: transport
 628  *
 629  * Initiates a graceful shutdown of the TCP socket by calling the
 630  * equivalent of shutdown(SHUT_RDWR);
 631  */
 632 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 633 {
 634         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 635         struct socket *sock = transport->sock;
 636
 637         if (sock != NULL) {
 638                 kernel_sock_shutdown(sock, SHUT_RDWR);
 639                 trace_rpc_socket_shutdown(xprt, sock);
 640         }
 641 }
 642
 643 /**
 644  * xs_tcp_send_request - write an RPC request to a TCP socket
 645  * @task: address of RPC task that manages the state of an RPC request
 646  *
 647  * Return values:
 648  *        0:    The request has been sent
 649  *   EAGAIN:    The socket was blocked, please call again later to
 650  *              complete the request
 651  * ENOTCONN:    Caller needs to invoke connect logic then call again
 652  *    other:    Some other error occurred, the request was not sent
 653  *
 654  * XXX: In the case of soft timeouts, should we eventually give up
 655  *      if sendmsg is not able to make progress?
 656  */
 657 static int xs_tcp_send_request(struct rpc_task *task)
 658 {
 659         struct rpc_rqst *req = task->tk_rqstp;
 660         struct rpc_xprt *xprt = req->rq_xprt;
 661         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 662         struct xdr_buf *xdr = &req->rq_snd_buf;
 663         bool zerocopy = true;
 664         int status;
 665         int sent;
 666
 667         xs_encode_stream_record_marker(&req->rq_snd_buf);
 668
 669         xs_pktdump("packet data:",
 670                                 req->rq_svec->iov_base,
 671                                 req->rq_svec->iov_len);
 672         /* Don't use zero copy if this is a resend. If the RPC call
 673          * completes while the socket holds a reference to the pages,
 674          * then we may end up resending corrupted data.
 675          */
 676         if (task->tk_flags & RPC_TASK_SENT)
 677                 zerocopy = false;
 678
 679         /* Continue transmitting the packet/record. We must be careful
 680          * to cope with writespace callbacks arriving _after_ we have
 681          * called sendmsg(). */
 682         while (1) {
 683                 sent = 0;
 684                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 685                                       req->rq_bytes_sent, zerocopy, &sent);
 686
 687                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 688                                 xdr->len - req->rq_bytes_sent, status);
 689
 690                 if (unlikely(sent == 0 && status < 0))
 691                         break;
 692
 693                 /* If we've sent the entire packet, immediately
 694                  * reset the count of bytes sent. */
 695                 req->rq_bytes_sent += sent;
 696                 req->rq_xmit_bytes_sent += sent;
 697                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 698                         req->rq_bytes_sent = 0;
 699                         return 0;
 700                 }
 701
 702                 if (sent != 0)
 703                         continue;
 704                 status = -EAGAIN;
 705                 break;
 706         }
 707
 708         switch (status) {
 709         case -ENOTSOCK:
 710                 status = -ENOTCONN;
 711                 /* Should we call xs_close() here? */
 712                 break;
 713         case -ENOBUFS:
 714         case -EAGAIN:
 715                 status = xs_nospace(task);
 716                 break;
 717         default:
 718                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 719                         -status);
 720         case -ECONNRESET:
 721         case -ECONNREFUSED:
 722         case -ENOTCONN:
 723         case -EADDRINUSE:
 724         case -EPIPE:
 725                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 726         }
 727
 728         return status;
 729 }
 730
 731 /**
 732  * xs_tcp_release_xprt - clean up after a tcp transmission
 733  * @xprt: transport
 734  * @task: rpc task
 735  *
 736  * This cleans up if an error causes us to abort the transmission of a request.
 737  * In this case, the socket may need to be reset in order to avoid confusing
 738  * the server.
 739  */
 740 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 741 {
 742         struct rpc_rqst *req;
 743
 744         if (task != xprt->snd_task)
 745                 return;
 746         if (task == NULL)
 747                 goto out_release;
 748         req = task->tk_rqstp;
 749         if (req == NULL)
 750                 goto out_release;
 751         if (req->rq_bytes_sent == 0)
 752                 goto out_release;
 753         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 754                 goto out_release;
 755         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 756 out_release:
 757         xprt_release_xprt(xprt, task);
 758 }
 759
 760 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 761 {
 762         transport->old_data_ready = sk->sk_data_ready;
 763         transport->old_state_change = sk->sk_state_change;
 764         transport->old_write_space = sk->sk_write_space;
 765         transport->old_error_report = sk->sk_error_report;
 766 }
 767
 768 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 769 {
 770         sk->sk_data_ready = transport->old_data_ready;
 771         sk->sk_state_change = transport->old_state_change;
 772         sk->sk_write_space = transport->old_write_space;
 773         sk->sk_error_report = transport->old_error_report;
 774 }
 775
 776 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 777 {
 778         smp_mb__before_atomic();
 779         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 780         clear_bit(XPRT_CLOSING, &xprt->state);
 781         smp_mb__after_atomic();
 782 }
 783
 784 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 785 {
 786         xs_sock_reset_connection_flags(xprt);
 787         /* Mark transport as closed and wake up all pending tasks */
 788         xprt_disconnect_done(xprt);
 789 }
 790
 791 /**
 792  * xs_error_report - callback to handle TCP socket state errors
 793  * @sk: socket
 794  *
 795  * Note: we don't call sock_error() since there may be a rpc_task
 796  * using the socket, and so we don't want to clear sk->sk_err.
 797  */
 798 static void xs_error_report(struct sock *sk)
 799 {
 800         struct rpc_xprt *xprt;
 801         int err;
 802
 803         read_lock_bh(&sk->sk_callback_lock);
 804         if (!(xprt = xprt_from_sock(sk)))
 805                 goto out;
 806
 807         err = -sk->sk_err;
 808         if (err == 0)
 809                 goto out;
 810         /* Is this a reset event? */
 811         if (sk->sk_state == TCP_CLOSE)
 812                 xs_sock_mark_closed(xprt);
 813         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 814                         xprt, -err);
 815         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 816         xprt_wake_pending_tasks(xprt, err);
 817  out:
 818         read_unlock_bh(&sk->sk_callback_lock);
 819 }
 820
 821 static void xs_reset_transport(struct sock_xprt *transport)
 822 {
 823         struct socket *sock = transport->sock;
 824         struct sock *sk = transport->inet;
 825         struct rpc_xprt *xprt = &transport->xprt;
 826
 827         if (sk == NULL)
 828                 return;
 829
 830         write_lock_bh(&sk->sk_callback_lock);
 831         transport->inet = NULL;
 832         transport->sock = NULL;
 833
 834         sk->sk_user_data = NULL;
 835
 836         xs_restore_old_callbacks(transport, sk);
 837         write_unlock_bh(&sk->sk_callback_lock);
 838         xs_sock_reset_connection_flags(xprt);
 839
 840         trace_rpc_socket_close(xprt, sock);
 841         sock_release(sock);
 842 }
 843
 844 /**
 845  * xs_close - close a socket
 846  * @xprt: transport
 847  *
 848  * This is used when all requests are complete; ie, no DRC state remains
 849  * on the server we want to save.
 850  *
 851  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 852  * xs_reset_transport() zeroing the socket from underneath a writer.
 853  */
 854 static void xs_close(struct rpc_xprt *xprt)
 855 {
 856         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 857
 858         dprintk("RPC:       xs_close xprt %p\n", xprt);
 859
 860         xs_reset_transport(transport);
 861         xprt->reestablish_timeout = 0;
 862
 863         xprt_disconnect_done(xprt);
 864 }
 865
 866 static void xs_xprt_free(struct rpc_xprt *xprt)
 867 {
 868         xs_free_peer_addresses(xprt);
 869         xprt_free(xprt);
 870 }
 871
 872 /**
 873  * xs_destroy - prepare to shutdown a transport
 874  * @xprt: doomed transport
 875  *
 876  */
 877 static void xs_destroy(struct rpc_xprt *xprt)
 878 {
 879         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 880
 881         xs_close(xprt);
 882         xs_xprt_free(xprt);
 883         module_put(THIS_MODULE);
 884 }
 885
 886 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 887 {
 888         struct xdr_skb_reader desc = {
 889                 .skb            = skb,
 890                 .offset         = sizeof(rpc_fraghdr),
 891                 .count          = skb->len - sizeof(rpc_fraghdr),
 892         };
 893
 894         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 895                 return -1;
 896         if (desc.count)
 897                 return -1;
 898         return 0;
 899 }
 900
 901 /**
 902  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 903  * @sk: socket with data to read
 904  * @len: how much data to read
 905  *
 906  * Currently this assumes we can read the whole reply in a single gulp.
 907  */
 908 static void xs_local_data_ready(struct sock *sk)
 909 {
 910         struct rpc_task *task;
 911         struct rpc_xprt *xprt;
 912         struct rpc_rqst *rovr;
 913         struct sk_buff *skb;
 914         int err, repsize, copied;
 915         u32 _xid;
 916         __be32 *xp;
 917
 918         read_lock_bh(&sk->sk_callback_lock);
 919         dprintk("RPC:       %s...\n", __func__);
 920         xprt = xprt_from_sock(sk);
 921         if (xprt == NULL)
 922                 goto out;
 923
 924         skb = skb_recv_datagram(sk, 0, 1, &err);
 925         if (skb == NULL)
 926                 goto out;
 927
 928         repsize = skb->len - sizeof(rpc_fraghdr);
 929         if (repsize < 4) {
 930                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 931                 goto dropit;
 932         }
 933
 934         /* Copy the XID from the skb... */
 935         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 936         if (xp == NULL)
 937                 goto dropit;
 938
 939         /* Look up and lock the request corresponding to the given XID */
 940         spin_lock(&xprt->transport_lock);
 941         rovr = xprt_lookup_rqst(xprt, *xp);
 942         if (!rovr)
 943                 goto out_unlock;
 944         task = rovr->rq_task;
 945
 946         copied = rovr->rq_private_buf.buflen;
 947         if (copied > repsize)
 948                 copied = repsize;
 949
 950         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 951                 dprintk("RPC:       sk_buff copy failed\n");
 952                 goto out_unlock;
 953         }
 954
 955         xprt_complete_rqst(task, copied);
 956
 957  out_unlock:
 958         spin_unlock(&xprt->transport_lock);
 959  dropit:
 960         skb_free_datagram(sk, skb);
 961  out:
 962         read_unlock_bh(&sk->sk_callback_lock);
 963 }
 964
 965 /**
 966  * xs_udp_data_ready - "data ready" callback for UDP sockets
 967  * @sk: socket with data to read
 968  * @len: how much data to read
 969  *
 970  */
 971 static void xs_udp_data_ready(struct sock *sk)
 972 {
 973         struct rpc_task *task;
 974         struct rpc_xprt *xprt;
 975         struct rpc_rqst *rovr;
 976         struct sk_buff *skb;
 977         int err, repsize, copied;
 978         u32 _xid;
 979         __be32 *xp;
 980
 981         read_lock_bh(&sk->sk_callback_lock);
 982         dprintk("RPC:       xs_udp_data_ready...\n");
 983         if (!(xprt = xprt_from_sock(sk)))
 984                 goto out;
 985
 986         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 987                 goto out;
 988
 989         repsize = skb->len - sizeof(struct udphdr);
 990         if (repsize < 4) {
 991                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 992                 goto dropit;
 993         }
 994
 995         /* Copy the XID from the skb... */
 996         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 997                                 sizeof(_xid), &_xid);
 998         if (xp == NULL)
 999                 goto dropit;
1000
1001         /* Look up and lock the request corresponding to the given XID */
1002         spin_lock(&xprt->transport_lock);
1003         rovr = xprt_lookup_rqst(xprt, *xp);
1004         if (!rovr)
1005                 goto out_unlock;
1006         task = rovr->rq_task;
1007
1008         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1009                 copied = repsize;
1010
1011         /* Suck it into the iovec, verify checksum if not done by hw. */
1012         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1013                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1014                 goto out_unlock;
1015         }
1016
1017         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1018
1019         xprt_adjust_cwnd(xprt, task, copied);
1020         xprt_complete_rqst(task, copied);
1021
1022  out_unlock:
1023         spin_unlock(&xprt->transport_lock);
1024  dropit:
1025         skb_free_datagram(sk, skb);
1026  out:
1027         read_unlock_bh(&sk->sk_callback_lock);
1028 }
1029
1030 /*
1031  * Helper function to force a TCP close if the server is sending
1032  * junk and/or it has put us in CLOSE_WAIT
1033  */
1034 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1035 {
1036         xprt_force_disconnect(xprt);
1037 }
1038
1039 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1040 {
1041         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1042         size_t len, used;
1043         char *p;
1044
1045         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1046         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1047         used = xdr_skb_read_bits(desc, p, len);
1048         transport->tcp_offset += used;
1049         if (used != len)
1050                 return;
1051
1052         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1053         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1054                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1055         else
1056                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1057         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1058
1059         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1060         transport->tcp_offset = 0;
1061
1062         /* Sanity check of the record length */
1063         if (unlikely(transport->tcp_reclen < 8)) {
1064                 dprintk("RPC:       invalid TCP record fragment length\n");
1065                 xs_tcp_force_close(xprt);
1066                 return;
1067         }
1068         dprintk("RPC:       reading TCP record fragment of length %d\n",
1069                         transport->tcp_reclen);
1070 }
1071
1072 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1073 {
1074         if (transport->tcp_offset == transport->tcp_reclen) {
1075                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1076                 transport->tcp_offset = 0;
1077                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1078                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1079                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1080                         transport->tcp_copied = 0;
1081                 }
1082         }
1083 }
1084
1085 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1086 {
1087         size_t len, used;
1088         char *p;
1089
1090         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1091         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1092         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1093         used = xdr_skb_read_bits(desc, p, len);
1094         transport->tcp_offset += used;
1095         if (used != len)
1096                 return;
1097         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1098         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1099         transport->tcp_copied = 4;
1100         dprintk("RPC:       reading %s XID %08x\n",
1101                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1102                                                               : "request with",
1103                         ntohl(transport->tcp_xid));
1104         xs_tcp_check_fraghdr(transport);
1105 }
1106
1107 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1108                                        struct xdr_skb_reader *desc)
1109 {
1110         size_t len, used;
1111         u32 offset;
1112         char *p;
1113
1114         /*
1115          * We want transport->tcp_offset to be 8 at the end of this routine
1116          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1117          * When this function is called for the first time,
1118          * transport->tcp_offset is 4 (after having already read the xid).
1119          */
1120         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1121         len = sizeof(transport->tcp_calldir) - offset;
1122         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1123         p = ((char *) &transport->tcp_calldir) + offset;
1124         used = xdr_skb_read_bits(desc, p, len);
1125         transport->tcp_offset += used;
1126         if (used != len)
1127                 return;
1128         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1129         /*
1130          * We don't yet have the XDR buffer, so we will write the calldir
1131          * out after we get the buffer from the 'struct rpc_rqst'
1132          */
1133         switch (ntohl(transport->tcp_calldir)) {
1134         case RPC_REPLY:
1135                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1136                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1137                 transport->tcp_flags |= TCP_RPC_REPLY;
1138                 break;
1139         case RPC_CALL:
1140                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1141                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1142                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1143                 break;
1144         default:
1145                 dprintk("RPC:       invalid request message type\n");
1146                 xs_tcp_force_close(&transport->xprt);
1147         }
1148         xs_tcp_check_fraghdr(transport);
1149 }
1150
1151 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1152                                      struct xdr_skb_reader *desc,
1153                                      struct rpc_rqst *req)
1154 {
1155         struct sock_xprt *transport =
1156                                 container_of(xprt, struct sock_xprt, xprt);
1157         struct xdr_buf *rcvbuf;
1158         size_t len;
1159         ssize_t r;
1160
1161         rcvbuf = &req->rq_private_buf;
1162
1163         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1164                 /*
1165                  * Save the RPC direction in the XDR buffer
1166                  */
1167                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1168                         &transport->tcp_calldir,
1169                         sizeof(transport->tcp_calldir));
1170                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1171                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1172         }
1173
1174         len = desc->count;
1175         if (len > transport->tcp_reclen - transport->tcp_offset) {
1176                 struct xdr_skb_reader my_desc;
1177
1178                 len = transport->tcp_reclen - transport->tcp_offset;
1179                 memcpy(&my_desc, desc, sizeof(my_desc));
1180                 my_desc.count = len;
1181                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1182                                           &my_desc, xdr_skb_read_bits);
1183                 desc->count -= r;
1184                 desc->offset += r;
1185         } else
1186                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1187                                           desc, xdr_skb_read_bits);
1188
1189         if (r > 0) {
1190                 transport->tcp_copied += r;
1191                 transport->tcp_offset += r;
1192         }
1193         if (r != len) {
1194                 /* Error when copying to the receive buffer,
1195                  * usually because we weren't able to allocate
1196                  * additional buffer pages. All we can do now
1197                  * is turn off TCP_RCV_COPY_DATA, so the request
1198                  * will not receive any additional updates,
1199                  * and time out.
1200                  * Any remaining data from this record will
1201                  * be discarded.
1202                  */
1203                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1204                 dprintk("RPC:       XID %08x truncated request\n",
1205                                 ntohl(transport->tcp_xid));
1206                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1207                                 "tcp_offset = %u, tcp_reclen = %u\n",
1208                                 xprt, transport->tcp_copied,
1209                                 transport->tcp_offset, transport->tcp_reclen);
1210                 return;
1211         }
1212
1213         dprintk("RPC:       XID %08x read %Zd bytes\n",
1214                         ntohl(transport->tcp_xid), r);
1215         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1216                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1217                         transport->tcp_offset, transport->tcp_reclen);
1218
1219         if (transport->tcp_copied == req->rq_private_buf.buflen)
1220                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1221         else if (transport->tcp_offset == transport->tcp_reclen) {
1222                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1223                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1224         }
1225 }
1226
1227 /*
1228  * Finds the request corresponding to the RPC xid and invokes the common
1229  * tcp read code to read the data.
1230  */
1231 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1232                                     struct xdr_skb_reader *desc)
1233 {
1234         struct sock_xprt *transport =
1235                                 container_of(xprt, struct sock_xprt, xprt);
1236         struct rpc_rqst *req;
1237
1238         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1239
1240         /* Find and lock the request corresponding to this xid */
1241         spin_lock(&xprt->transport_lock);
1242         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1243         if (!req) {
1244                 dprintk("RPC:       XID %08x request not found!\n",
1245                                 ntohl(transport->tcp_xid));
1246                 spin_unlock(&xprt->transport_lock);
1247                 return -1;
1248         }
1249
1250         xs_tcp_read_common(xprt, desc, req);
1251
1252         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1253                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1254
1255         spin_unlock(&xprt->transport_lock);
1256         return 0;
1257 }
1258
1259 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1260 /*
1261  * Obtains an rpc_rqst previously allocated and invokes the common
1262  * tcp read code to read the data.  The result is placed in the callback
1263  * queue.
1264  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1265  * connection and return -1.
1266  */
1267 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1268                                        struct xdr_skb_reader *desc)
1269 {
1270         struct sock_xprt *transport =
1271                                 container_of(xprt, struct sock_xprt, xprt);
1272         struct rpc_rqst *req;
1273
1274         /* Look up and lock the request corresponding to the given XID */
1275         spin_lock(&xprt->transport_lock);
1276         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1277         if (req == NULL) {
1278                 spin_unlock(&xprt->transport_lock);
1279                 printk(KERN_WARNING "Callback slot table overflowed\n");
1280                 xprt_force_disconnect(xprt);
1281                 return -1;
1282         }
1283
1284         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1285         xs_tcp_read_common(xprt, desc, req);
1286
1287         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1288                 xprt_complete_bc_request(req, transport->tcp_copied);
1289         spin_unlock(&xprt->transport_lock);
1290
1291         return 0;
1292 }
1293
1294 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1295                                         struct xdr_skb_reader *desc)
1296 {
1297         struct sock_xprt *transport =
1298                                 container_of(xprt, struct sock_xprt, xprt);
1299
1300         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1301                 xs_tcp_read_reply(xprt, desc) :
1302                 xs_tcp_read_callback(xprt, desc);
1303 }
1304 #else
1305 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1306                                         struct xdr_skb_reader *desc)
1307 {
1308         return xs_tcp_read_reply(xprt, desc);
1309 }
1310 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1311
1312 /*
1313  * Read data off the transport.  This can be either an RPC_CALL or an
1314  * RPC_REPLY.  Relay the processing to helper functions.
1315  */
1316 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1317                                     struct xdr_skb_reader *desc)
1318 {
1319         struct sock_xprt *transport =
1320                                 container_of(xprt, struct sock_xprt, xprt);
1321
1322         if (_xs_tcp_read_data(xprt, desc) == 0)
1323                 xs_tcp_check_fraghdr(transport);
1324         else {
1325                 /*
1326                  * The transport_lock protects the request handling.
1327                  * There's no need to hold it to update the tcp_flags.
1328                  */
1329                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1330         }
1331 }
1332
1333 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1334 {
1335         size_t len;
1336
1337         len = transport->tcp_reclen - transport->tcp_offset;
1338         if (len > desc->count)
1339                 len = desc->count;
1340         desc->count -= len;
1341         desc->offset += len;
1342         transport->tcp_offset += len;
1343         dprintk("RPC:       discarded %Zu bytes\n", len);
1344         xs_tcp_check_fraghdr(transport);
1345 }
1346
1347 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1348 {
1349         struct rpc_xprt *xprt = rd_desc->arg.data;
1350         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1351         struct xdr_skb_reader desc = {
1352                 .skb    = skb,
1353                 .offset = offset,
1354                 .count  = len,
1355         };
1356
1357         dprintk("RPC:       xs_tcp_data_recv started\n");
1358         do {
1359                 trace_xs_tcp_data_recv(transport);
1360                 /* Read in a new fragment marker if necessary */
1361                 /* Can we ever really expect to get completely empty fragments? */
1362                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1363                         xs_tcp_read_fraghdr(xprt, &desc);
1364                         continue;
1365                 }
1366                 /* Read in the xid if necessary */
1367                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1368                         xs_tcp_read_xid(transport, &desc);
1369                         continue;
1370                 }
1371                 /* Read in the call/reply flag */
1372                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1373                         xs_tcp_read_calldir(transport, &desc);
1374                         continue;
1375                 }
1376                 /* Read in the request data */
1377                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1378                         xs_tcp_read_data(xprt, &desc);
1379                         continue;
1380                 }
1381                 /* Skip over any trailing bytes on short reads */
1382                 xs_tcp_read_discard(transport, &desc);
1383         } while (desc.count);
1384         trace_xs_tcp_data_recv(transport);
1385         dprintk("RPC:       xs_tcp_data_recv done\n");
1386         return len - desc.count;
1387 }
1388
1389 /**
1390  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1391  * @sk: socket with data to read
1392  * @bytes: how much data to read
1393  *
1394  */
1395 static void xs_tcp_data_ready(struct sock *sk)
1396 {
1397         struct rpc_xprt *xprt;
1398         read_descriptor_t rd_desc;
1399         int read;
1400         unsigned long total = 0;
1401
1402         dprintk("RPC:       xs_tcp_data_ready...\n");
1403
1404         read_lock_bh(&sk->sk_callback_lock);
1405         if (!(xprt = xprt_from_sock(sk))) {
1406                 read = 0;
1407                 goto out;
1408         }
1409         /* Any data means we had a useful conversation, so
1410          * the we don't need to delay the next reconnect
1411          */
1412         if (xprt->reestablish_timeout)
1413                 xprt->reestablish_timeout = 0;
1414
1415         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1416         rd_desc.arg.data = xprt;
1417         do {
1418                 rd_desc.count = 65536;
1419                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1420                 if (read > 0)
1421                         total += read;
1422         } while (read > 0);
1423 out:
1424         trace_xs_tcp_data_ready(xprt, read, total);
1425         read_unlock_bh(&sk->sk_callback_lock);
1426 }
1427
1428 /**
1429  * xs_tcp_state_change - callback to handle TCP socket state changes
1430  * @sk: socket whose state has changed
1431  *
1432  */
1433 static void xs_tcp_state_change(struct sock *sk)
1434 {
1435         struct rpc_xprt *xprt;
1436
1437         read_lock_bh(&sk->sk_callback_lock);
1438         if (!(xprt = xprt_from_sock(sk)))
1439                 goto out;
1440         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1441         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1442                         sk->sk_state, xprt_connected(xprt),
1443                         sock_flag(sk, SOCK_DEAD),
1444                         sock_flag(sk, SOCK_ZAPPED),
1445                         sk->sk_shutdown);
1446
1447         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1448         switch (sk->sk_state) {
1449         case TCP_ESTABLISHED:
1450                 spin_lock(&xprt->transport_lock);
1451                 if (!xprt_test_and_set_connected(xprt)) {
1452                         struct sock_xprt *transport = container_of(xprt,
1453                                         struct sock_xprt, xprt);
1454
1455                         /* Reset TCP record info */
1456                         transport->tcp_offset = 0;
1457                         transport->tcp_reclen = 0;
1458                         transport->tcp_copied = 0;
1459                         transport->tcp_flags =
1460                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1461                         xprt->connect_cookie++;
1462
1463                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1464                 }
1465                 spin_unlock(&xprt->transport_lock);
1466                 break;
1467         case TCP_FIN_WAIT1:
1468                 /* The client initiated a shutdown of the socket */
1469                 xprt->connect_cookie++;
1470                 xprt->reestablish_timeout = 0;
1471                 set_bit(XPRT_CLOSING, &xprt->state);
1472                 smp_mb__before_atomic();
1473                 clear_bit(XPRT_CONNECTED, &xprt->state);
1474                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1475                 smp_mb__after_atomic();
1476                 break;
1477         case TCP_CLOSE_WAIT:
1478                 /* The server initiated a shutdown of the socket */
1479                 xprt->connect_cookie++;
1480                 clear_bit(XPRT_CONNECTED, &xprt->state);
1481                 xs_tcp_force_close(xprt);
1482         case TCP_CLOSING:
1483                 /*
1484                  * If the server closed down the connection, make sure that
1485                  * we back off before reconnecting
1486                  */
1487                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1488                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1489                 break;
1490         case TCP_LAST_ACK:
1491                 set_bit(XPRT_CLOSING, &xprt->state);
1492                 smp_mb__before_atomic();
1493                 clear_bit(XPRT_CONNECTED, &xprt->state);
1494                 smp_mb__after_atomic();
1495                 break;
1496         case TCP_CLOSE:
1497                 xs_sock_mark_closed(xprt);
1498         }
1499  out:
1500         read_unlock_bh(&sk->sk_callback_lock);
1501 }
1502
1503 static void xs_write_space(struct sock *sk)
1504 {
1505         struct socket *sock;
1506         struct rpc_xprt *xprt;
1507
1508         if (unlikely(!(sock = sk->sk_socket)))
1509                 return;
1510         clear_bit(SOCK_NOSPACE, &sock->flags);
1511
1512         if (unlikely(!(xprt = xprt_from_sock(sk))))
1513                 return;
1514         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1515                 return;
1516
1517         xprt_write_space(xprt);
1518 }
1519
1520 /**
1521  * xs_udp_write_space - callback invoked when socket buffer space
1522  *                             becomes available
1523  * @sk: socket whose state has changed
1524  *
1525  * Called when more output buffer space is available for this socket.
1526  * We try not to wake our writers until they can make "significant"
1527  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1528  * with a bunch of small requests.
1529  */
1530 static void xs_udp_write_space(struct sock *sk)
1531 {
1532         read_lock_bh(&sk->sk_callback_lock);
1533
1534         /* from net/core/sock.c:sock_def_write_space */
1535         if (sock_writeable(sk))
1536                 xs_write_space(sk);
1537
1538         read_unlock_bh(&sk->sk_callback_lock);
1539 }
1540
1541 /**
1542  * xs_tcp_write_space - callback invoked when socket buffer space
1543  *                             becomes available
1544  * @sk: socket whose state has changed
1545  *
1546  * Called when more output buffer space is available for this socket.
1547  * We try not to wake our writers until they can make "significant"
1548  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1549  * with a bunch of small requests.
1550  */
1551 static void xs_tcp_write_space(struct sock *sk)
1552 {
1553         read_lock_bh(&sk->sk_callback_lock);
1554
1555         /* from net/core/stream.c:sk_stream_write_space */
1556         if (sk_stream_is_writeable(sk))
1557                 xs_write_space(sk);
1558
1559         read_unlock_bh(&sk->sk_callback_lock);
1560 }
1561
1562 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1563 {
1564         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1565         struct sock *sk = transport->inet;
1566
1567         if (transport->rcvsize) {
1568                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1569                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1570         }
1571         if (transport->sndsize) {
1572                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1573                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1574                 sk->sk_write_space(sk);
1575         }
1576 }
1577
1578 /**
1579  * xs_udp_set_buffer_size - set send and receive limits
1580  * @xprt: generic transport
1581  * @sndsize: requested size of send buffer, in bytes
1582  * @rcvsize: requested size of receive buffer, in bytes
1583  *
1584  * Set socket send and receive buffer size limits.
1585  */
1586 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1587 {
1588         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1589
1590         transport->sndsize = 0;
1591         if (sndsize)
1592                 transport->sndsize = sndsize + 1024;
1593         transport->rcvsize = 0;
1594         if (rcvsize)
1595                 transport->rcvsize = rcvsize + 1024;
1596
1597         xs_udp_do_set_buffer_size(xprt);
1598 }
1599
1600 /**
1601  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1602  * @task: task that timed out
1603  *
1604  * Adjust the congestion window after a retransmit timeout has occurred.
1605  */
1606 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1607 {
1608         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1609 }
1610
1611 static unsigned short xs_get_random_port(void)
1612 {
1613         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1614         unsigned short rand = (unsigned short) prandom_u32() % range;
1615         return rand + xprt_min_resvport;
1616 }
1617
1618 /**
1619  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1620  * @sock: socket
1621  *
1622  * Note that this function has to be called on all sockets that share the
1623  * same port, and it must be called before binding.
1624  */
1625 static void xs_sock_set_reuseport(struct socket *sock)
1626 {
1627         int opt = 1;
1628
1629         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1630                         (char *)&opt, sizeof(opt));
1631 }
1632
1633 static unsigned short xs_sock_getport(struct socket *sock)
1634 {
1635         struct sockaddr_storage buf;
1636         int buflen;
1637         unsigned short port = 0;
1638
1639         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1640                 goto out;
1641         switch (buf.ss_family) {
1642         case AF_INET6:
1643                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1644                 break;
1645         case AF_INET:
1646                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1647         }
1648 out:
1649         return port;
1650 }
1651
1652 /**
1653  * xs_set_port - reset the port number in the remote endpoint address
1654  * @xprt: generic transport
1655  * @port: new port number
1656  *
1657  */
1658 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1659 {
1660         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1661
1662         rpc_set_port(xs_addr(xprt), port);
1663         xs_update_peer_port(xprt);
1664 }
1665
1666 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1667 {
1668         if (transport->srcport == 0)
1669                 transport->srcport = xs_sock_getport(sock);
1670 }
1671
1672 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1673 {
1674         unsigned short port = transport->srcport;
1675
1676         if (port == 0 && transport->xprt.resvport)
1677                 port = xs_get_random_port();
1678         return port;
1679 }
1680
1681 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1682 {
1683         if (transport->srcport != 0)
1684                 transport->srcport = 0;
1685         if (!transport->xprt.resvport)
1686                 return 0;
1687         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1688                 return xprt_max_resvport;
1689         return --port;
1690 }
1691 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1692 {
1693         struct sockaddr_storage myaddr;
1694         int err, nloop = 0;
1695         unsigned short port = xs_get_srcport(transport);
1696         unsigned short last;
1697
1698         /*
1699          * If we are asking for any ephemeral port (i.e. port == 0 &&
1700          * transport->xprt.resvport == 0), don't bind.  Let the local
1701          * port selection happen implicitly when the socket is used
1702          * (for example at connect time).
1703          *
1704          * This ensures that we can continue to establish TCP
1705          * connections even when all local ephemeral ports are already
1706          * a part of some TCP connection.  This makes no difference
1707          * for UDP sockets, but also doens't harm them.
1708          *
1709          * If we're asking for any reserved port (i.e. port == 0 &&
1710          * transport->xprt.resvport == 1) xs_get_srcport above will
1711          * ensure that port is non-zero and we will bind as needed.
1712          */
1713         if (port == 0)
1714                 return 0;
1715
1716         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1717         do {
1718                 rpc_set_port((struct sockaddr *)&myaddr, port);
1719                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1720                                 transport->xprt.addrlen);
1721                 if (err == 0) {
1722                         transport->srcport = port;
1723                         break;
1724                 }
1725                 last = port;
1726                 port = xs_next_srcport(transport, port);
1727                 if (port > last)
1728                         nloop++;
1729         } while (err == -EADDRINUSE && nloop != 2);
1730
1731         if (myaddr.ss_family == AF_INET)
1732                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1733                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1734                                 port, err ? "failed" : "ok", err);
1735         else
1736                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1737                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1738                                 port, err ? "failed" : "ok", err);
1739         return err;
1740 }
1741
1742 /*
1743  * We don't support autobind on AF_LOCAL sockets
1744  */
1745 static void xs_local_rpcbind(struct rpc_task *task)
1746 {
1747         rcu_read_lock();
1748         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1749         rcu_read_unlock();
1750 }
1751
1752 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1753 {
1754 }
1755
1756 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1757 static struct lock_class_key xs_key[2];
1758 static struct lock_class_key xs_slock_key[2];
1759
1760 static inline void xs_reclassify_socketu(struct socket *sock)
1761 {
1762         struct sock *sk = sock->sk;
1763
1764         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1765                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1766 }
1767
1768 static inline void xs_reclassify_socket4(struct socket *sock)
1769 {
1770         struct sock *sk = sock->sk;
1771
1772         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1773                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1774 }
1775
1776 static inline void xs_reclassify_socket6(struct socket *sock)
1777 {
1778         struct sock *sk = sock->sk;
1779
1780         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1781                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1782 }
1783
1784 static inline void xs_reclassify_socket(int family, struct socket *sock)
1785 {
1786         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1787         if (sock_owned_by_user(sock->sk))
1788                 return;
1789
1790         switch (family) {
1791         case AF_LOCAL:
1792                 xs_reclassify_socketu(sock);
1793                 break;
1794         case AF_INET:
1795                 xs_reclassify_socket4(sock);
1796                 break;
1797         case AF_INET6:
1798                 xs_reclassify_socket6(sock);
1799                 break;
1800         }
1801 }
1802 #else
1803 static inline void xs_reclassify_socketu(struct socket *sock)
1804 {
1805 }
1806
1807 static inline void xs_reclassify_socket4(struct socket *sock)
1808 {
1809 }
1810
1811 static inline void xs_reclassify_socket6(struct socket *sock)
1812 {
1813 }
1814
1815 static inline void xs_reclassify_socket(int family, struct socket *sock)
1816 {
1817 }
1818 #endif
1819
1820 static void xs_dummy_setup_socket(struct work_struct *work)
1821 {
1822 }
1823
1824 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1825                 struct sock_xprt *transport, int family, int type,
1826                 int protocol, bool reuseport)
1827 {
1828         struct socket *sock;
1829         int err;
1830
1831         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1832         if (err < 0) {
1833                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1834                                 protocol, -err);
1835                 goto out;
1836         }
1837         xs_reclassify_socket(family, sock);
1838
1839         if (reuseport)
1840                 xs_sock_set_reuseport(sock);
1841
1842         err = xs_bind(transport, sock);
1843         if (err) {
1844                 sock_release(sock);
1845                 goto out;
1846         }
1847
1848         return sock;
1849 out:
1850         return ERR_PTR(err);
1851 }
1852
1853 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1854                                       struct socket *sock)
1855 {
1856         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1857                                                                         xprt);
1858
1859         if (!transport->inet) {
1860                 struct sock *sk = sock->sk;
1861
1862                 write_lock_bh(&sk->sk_callback_lock);
1863
1864                 xs_save_old_callbacks(transport, sk);
1865
1866                 sk->sk_user_data = xprt;
1867                 sk->sk_data_ready = xs_local_data_ready;
1868                 sk->sk_write_space = xs_udp_write_space;
1869                 sk->sk_error_report = xs_error_report;
1870                 sk->sk_allocation = GFP_ATOMIC;
1871
1872                 xprt_clear_connected(xprt);
1873
1874                 /* Reset to new socket */
1875                 transport->sock = sock;
1876                 transport->inet = sk;
1877
1878                 write_unlock_bh(&sk->sk_callback_lock);
1879         }
1880
1881         /* Tell the socket layer to start connecting... */
1882         xprt->stat.connect_count++;
1883         xprt->stat.connect_start = jiffies;
1884         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1885 }
1886
1887 /**
1888  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1889  * @xprt: RPC transport to connect
1890  * @transport: socket transport to connect
1891  * @create_sock: function to create a socket of the correct type
1892  */
1893 static int xs_local_setup_socket(struct sock_xprt *transport)
1894 {
1895         struct rpc_xprt *xprt = &transport->xprt;
1896         struct socket *sock;
1897         int status = -EIO;
1898
1899         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1900                                         SOCK_STREAM, 0, &sock, 1);
1901         if (status < 0) {
1902                 dprintk("RPC:       can't create AF_LOCAL "
1903                         "transport socket (%d).\n", -status);
1904                 goto out;
1905         }
1906         xs_reclassify_socketu(sock);
1907
1908         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1909                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1910
1911         status = xs_local_finish_connecting(xprt, sock);
1912         trace_rpc_socket_connect(xprt, sock, status);
1913         switch (status) {
1914         case 0:
1915                 dprintk("RPC:       xprt %p connected to %s\n",
1916                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1917                 xprt_set_connected(xprt);
1918         case -ENOBUFS:
1919                 break;
1920         case -ENOENT:
1921                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1922                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1923                 break;
1924         case -ECONNREFUSED:
1925                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1926                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1927                 break;
1928         default:
1929                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1930                                 __func__, -status,
1931                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1932         }
1933
1934 out:
1935         xprt_clear_connecting(xprt);
1936         xprt_wake_pending_tasks(xprt, status);
1937         return status;
1938 }
1939
1940 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1941 {
1942         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1943         int ret;
1944
1945          if (RPC_IS_ASYNC(task)) {
1946                 /*
1947                  * We want the AF_LOCAL connect to be resolved in the
1948                  * filesystem namespace of the process making the rpc
1949                  * call.  Thus we connect synchronously.
1950                  *
1951                  * If we want to support asynchronous AF_LOCAL calls,
1952                  * we'll need to figure out how to pass a namespace to
1953                  * connect.
1954                  */
1955                 rpc_exit(task, -ENOTCONN);
1956                 return;
1957         }
1958         ret = xs_local_setup_socket(transport);
1959         if (ret && !RPC_IS_SOFTCONN(task))
1960                 msleep_interruptible(15000);
1961 }
1962
1963 #ifdef CONFIG_SUNRPC_SWAP
1964 static void xs_set_memalloc(struct rpc_xprt *xprt)
1965 {
1966         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1967                         xprt);
1968
1969         if (xprt->swapper)
1970                 sk_set_memalloc(transport->inet);
1971 }
1972
1973 /**
1974  * xs_swapper - Tag this transport as being used for swap.
1975  * @xprt: transport to tag
1976  * @enable: enable/disable
1977  *
1978  */
1979 int xs_swapper(struct rpc_xprt *xprt, int enable)
1980 {
1981         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1982                         xprt);
1983         int err = 0;
1984
1985         if (enable) {
1986                 xprt->swapper++;
1987                 xs_set_memalloc(xprt);
1988         } else if (xprt->swapper) {
1989                 xprt->swapper--;
1990                 sk_clear_memalloc(transport->inet);
1991         }
1992
1993         return err;
1994 }
1995 EXPORT_SYMBOL_GPL(xs_swapper);
1996 #else
1997 static void xs_set_memalloc(struct rpc_xprt *xprt)
1998 {
1999 }
2000 #endif
2001
2002 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2003 {
2004         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2005
2006         if (!transport->inet) {
2007                 struct sock *sk = sock->sk;
2008
2009                 write_lock_bh(&sk->sk_callback_lock);
2010
2011                 xs_save_old_callbacks(transport, sk);
2012
2013                 sk->sk_user_data = xprt;
2014                 sk->sk_data_ready = xs_udp_data_ready;
2015                 sk->sk_write_space = xs_udp_write_space;
2016                 sk->sk_allocation = GFP_ATOMIC;
2017
2018                 xprt_set_connected(xprt);
2019
2020                 /* Reset to new socket */
2021                 transport->sock = sock;
2022                 transport->inet = sk;
2023
2024                 xs_set_memalloc(xprt);
2025
2026                 write_unlock_bh(&sk->sk_callback_lock);
2027         }
2028         xs_udp_do_set_buffer_size(xprt);
2029 }
2030
2031 static void xs_udp_setup_socket(struct work_struct *work)
2032 {
2033         struct sock_xprt *transport =
2034                 container_of(work, struct sock_xprt, connect_worker.work);
2035         struct rpc_xprt *xprt = &transport->xprt;
2036         struct socket *sock = transport->sock;
2037         int status = -EIO;
2038
2039         sock = xs_create_sock(xprt, transport,
2040                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2041                         IPPROTO_UDP, false);
2042         if (IS_ERR(sock))
2043                 goto out;
2044
2045         dprintk("RPC:       worker connecting xprt %p via %s to "
2046                                 "%s (port %s)\n", xprt,
2047                         xprt->address_strings[RPC_DISPLAY_PROTO],
2048                         xprt->address_strings[RPC_DISPLAY_ADDR],
2049                         xprt->address_strings[RPC_DISPLAY_PORT]);
2050
2051         xs_udp_finish_connecting(xprt, sock);
2052         trace_rpc_socket_connect(xprt, sock, 0);
2053         status = 0;
2054 out:
2055         xprt_unlock_connect(xprt, transport);
2056         xprt_clear_connecting(xprt);
2057         xprt_wake_pending_tasks(xprt, status);
2058 }
2059
2060 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2061 {
2062         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2063         int ret = -ENOTCONN;
2064
2065         if (!transport->inet) {
2066                 struct sock *sk = sock->sk;
2067                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2068                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2069                 unsigned int opt_on = 1;
2070
2071                 /* TCP Keepalive options */
2072                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2073                                 (char *)&opt_on, sizeof(opt_on));
2074                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2075                                 (char *)&keepidle, sizeof(keepidle));
2076                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2077                                 (char *)&keepidle, sizeof(keepidle));
2078                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2079                                 (char *)&keepcnt, sizeof(keepcnt));
2080
2081                 write_lock_bh(&sk->sk_callback_lock);
2082
2083                 xs_save_old_callbacks(transport, sk);
2084
2085                 sk->sk_user_data = xprt;
2086                 sk->sk_data_ready = xs_tcp_data_ready;
2087                 sk->sk_state_change = xs_tcp_state_change;
2088                 sk->sk_write_space = xs_tcp_write_space;
2089                 sk->sk_error_report = xs_error_report;
2090                 sk->sk_allocation = GFP_ATOMIC;
2091
2092                 /* socket options */
2093                 sock_reset_flag(sk, SOCK_LINGER);
2094                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2095
2096                 xprt_clear_connected(xprt);
2097
2098                 /* Reset to new socket */
2099                 transport->sock = sock;
2100                 transport->inet = sk;
2101
2102                 write_unlock_bh(&sk->sk_callback_lock);
2103         }
2104
2105         if (!xprt_bound(xprt))
2106                 goto out;
2107
2108         xs_set_memalloc(xprt);
2109
2110         /* Tell the socket layer to start connecting... */
2111         xprt->stat.connect_count++;
2112         xprt->stat.connect_start = jiffies;
2113         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2114         switch (ret) {
2115         case 0:
2116                 xs_set_srcport(transport, sock);
2117         case -EINPROGRESS:
2118                 /* SYN_SENT! */
2119                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2120                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2121         }
2122 out:
2123         return ret;
2124 }
2125
2126 /**
2127  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2128  * @xprt: RPC transport to connect
2129  * @transport: socket transport to connect
2130  * @create_sock: function to create a socket of the correct type
2131  *
2132  * Invoked by a work queue tasklet.
2133  */
2134 static void xs_tcp_setup_socket(struct work_struct *work)
2135 {
2136         struct sock_xprt *transport =
2137                 container_of(work, struct sock_xprt, connect_worker.work);
2138         struct socket *sock = transport->sock;
2139         struct rpc_xprt *xprt = &transport->xprt;
2140         int status = -EIO;
2141
2142         if (!sock) {
2143                 sock = xs_create_sock(xprt, transport,
2144                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2145                                 IPPROTO_TCP, true);
2146                 if (IS_ERR(sock)) {
2147                         status = PTR_ERR(sock);
2148                         goto out;
2149                 }
2150         }
2151
2152         dprintk("RPC:       worker connecting xprt %p via %s to "
2153                                 "%s (port %s)\n", xprt,
2154                         xprt->address_strings[RPC_DISPLAY_PROTO],
2155                         xprt->address_strings[RPC_DISPLAY_ADDR],
2156                         xprt->address_strings[RPC_DISPLAY_PORT]);
2157
2158         status = xs_tcp_finish_connecting(xprt, sock);
2159         trace_rpc_socket_connect(xprt, sock, status);
2160         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2161                         xprt, -status, xprt_connected(xprt),
2162                         sock->sk->sk_state);
2163         switch (status) {
2164         default:
2165                 printk("%s: connect returned unhandled error %d\n",
2166                         __func__, status);
2167         case -EADDRNOTAVAIL:
2168                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2169                  * and retry
2170                  */
2171                 xs_tcp_force_close(xprt);
2172                 break;
2173         case 0:
2174         case -EINPROGRESS:
2175         case -EALREADY:
2176                 xprt_unlock_connect(xprt, transport);
2177                 xprt_clear_connecting(xprt);
2178                 return;
2179         case -EINVAL:
2180                 /* Happens, for instance, if the user specified a link
2181                  * local IPv6 address without a scope-id.
2182                  */
2183         case -ECONNREFUSED:
2184         case -ECONNRESET:
2185         case -ENETUNREACH:
2186         case -EADDRINUSE:
2187         case -ENOBUFS:
2188                 /* retry with existing socket, after a delay */
2189                 xs_tcp_force_close(xprt);
2190                 goto out;
2191         }
2192         status = -EAGAIN;
2193 out:
2194         xprt_unlock_connect(xprt, transport);
2195         xprt_clear_connecting(xprt);
2196         xprt_wake_pending_tasks(xprt, status);
2197 }
2198
2199 /**
2200  * xs_connect - connect a socket to a remote endpoint
2201  * @xprt: pointer to transport structure
2202  * @task: address of RPC task that manages state of connect request
2203  *
2204  * TCP: If the remote end dropped the connection, delay reconnecting.
2205  *
2206  * UDP socket connects are synchronous, but we use a work queue anyway
2207  * to guarantee that even unprivileged user processes can set up a
2208  * socket on a privileged port.
2209  *
2210  * If a UDP socket connect fails, the delay behavior here prevents
2211  * retry floods (hard mounts).
2212  */
2213 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2214 {
2215         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2216
2217         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2218
2219         /* Start by resetting any existing state */
2220         xs_reset_transport(transport);
2221
2222         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2223                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2224                                 "seconds\n",
2225                                 xprt, xprt->reestablish_timeout / HZ);
2226                 queue_delayed_work(rpciod_workqueue,
2227                                    &transport->connect_worker,
2228                                    xprt->reestablish_timeout);
2229                 xprt->reestablish_timeout <<= 1;
2230                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2231                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2232                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2233                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2234         } else {
2235                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2236                 queue_delayed_work(rpciod_workqueue,
2237                                    &transport->connect_worker, 0);
2238         }
2239 }
2240
2241 /**
2242  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2243  * @xprt: rpc_xprt struct containing statistics
2244  * @seq: output file
2245  *
2246  */
2247 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2248 {
2249         long idle_time = 0;
2250
2251         if (xprt_connected(xprt))
2252                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2253
2254         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2255                         "%llu %llu %lu %llu %llu\n",
2256                         xprt->stat.bind_count,
2257                         xprt->stat.connect_count,
2258                         xprt->stat.connect_time,
2259                         idle_time,
2260                         xprt->stat.sends,
2261                         xprt->stat.recvs,
2262                         xprt->stat.bad_xids,
2263                         xprt->stat.req_u,
2264                         xprt->stat.bklog_u,
2265                         xprt->stat.max_slots,
2266                         xprt->stat.sending_u,
2267                         xprt->stat.pending_u);
2268 }
2269
2270 /**
2271  * xs_udp_print_stats - display UDP socket-specifc stats
2272  * @xprt: rpc_xprt struct containing statistics
2273  * @seq: output file
2274  *
2275  */
2276 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2277 {
2278         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2279
2280         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2281                         "%lu %llu %llu\n",
2282                         transport->srcport,
2283                         xprt->stat.bind_count,
2284                         xprt->stat.sends,
2285                         xprt->stat.recvs,
2286                         xprt->stat.bad_xids,
2287                         xprt->stat.req_u,
2288                         xprt->stat.bklog_u,
2289                         xprt->stat.max_slots,
2290                         xprt->stat.sending_u,
2291                         xprt->stat.pending_u);
2292 }
2293
2294 /**
2295  * xs_tcp_print_stats - display TCP socket-specifc stats
2296  * @xprt: rpc_xprt struct containing statistics
2297  * @seq: output file
2298  *
2299  */
2300 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2301 {
2302         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2303         long idle_time = 0;
2304
2305         if (xprt_connected(xprt))
2306                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2307
2308         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2309                         "%llu %llu %lu %llu %llu\n",
2310                         transport->srcport,
2311                         xprt->stat.bind_count,
2312                         xprt->stat.connect_count,
2313                         xprt->stat.connect_time,
2314                         idle_time,
2315                         xprt->stat.sends,
2316                         xprt->stat.recvs,
2317                         xprt->stat.bad_xids,
2318                         xprt->stat.req_u,
2319                         xprt->stat.bklog_u,
2320                         xprt->stat.max_slots,
2321                         xprt->stat.sending_u,
2322                         xprt->stat.pending_u);
2323 }
2324
2325 /*
2326  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2327  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2328  * to use the server side send routines.
2329  */
2330 static void *bc_malloc(struct rpc_task *task, size_t size)
2331 {
2332         struct page *page;
2333         struct rpc_buffer *buf;
2334
2335         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2336         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2337                 return NULL;
2338
2339         page = alloc_page(GFP_KERNEL);
2340         if (!page)
2341                 return NULL;
2342
2343         buf = page_address(page);
2344         buf->len = PAGE_SIZE;
2345
2346         return buf->data;
2347 }
2348
2349 /*
2350  * Free the space allocated in the bc_alloc routine
2351  */
2352 static void bc_free(void *buffer)
2353 {
2354         struct rpc_buffer *buf;
2355
2356         if (!buffer)
2357                 return;
2358
2359         buf = container_of(buffer, struct rpc_buffer, data);
2360         free_page((unsigned long)buf);
2361 }
2362
2363 /*
2364  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2365  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2366  */
2367 static int bc_sendto(struct rpc_rqst *req)
2368 {
2369         int len;
2370         struct xdr_buf *xbufp = &req->rq_snd_buf;
2371         struct rpc_xprt *xprt = req->rq_xprt;
2372         struct sock_xprt *transport =
2373                                 container_of(xprt, struct sock_xprt, xprt);
2374         struct socket *sock = transport->sock;
2375         unsigned long headoff;
2376         unsigned long tailoff;
2377
2378         xs_encode_stream_record_marker(xbufp);
2379
2380         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2381         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2382         len = svc_send_common(sock, xbufp,
2383                               virt_to_page(xbufp->head[0].iov_base), headoff,
2384                               xbufp->tail[0].iov_base, tailoff);
2385
2386         if (len != xbufp->len) {
2387                 printk(KERN_NOTICE "Error sending entire callback!\n");
2388                 len = -EAGAIN;
2389         }
2390
2391         return len;
2392 }
2393
2394 /*
2395  * The send routine. Borrows from svc_send
2396  */
2397 static int bc_send_request(struct rpc_task *task)
2398 {
2399         struct rpc_rqst *req = task->tk_rqstp;
2400         struct svc_xprt *xprt;
2401         u32                     len;
2402
2403         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2404         /*
2405          * Get the server socket associated with this callback xprt
2406          */
2407         xprt = req->rq_xprt->bc_xprt;
2408
2409         /*
2410          * Grab the mutex to serialize data as the connection is shared
2411          * with the fore channel
2412          */
2413         if (!mutex_trylock(&xprt->xpt_mutex)) {
2414                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2415                 if (!mutex_trylock(&xprt->xpt_mutex))
2416                         return -EAGAIN;
2417                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2418         }
2419         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2420                 len = -ENOTCONN;
2421         else
2422                 len = bc_sendto(req);
2423         mutex_unlock(&xprt->xpt_mutex);
2424
2425         if (len > 0)
2426                 len = 0;
2427
2428         return len;
2429 }
2430
2431 /*
2432  * The close routine. Since this is client initiated, we do nothing
2433  */
2434
2435 static void bc_close(struct rpc_xprt *xprt)
2436 {
2437 }
2438
2439 /*
2440  * The xprt destroy routine. Again, because this connection is client
2441  * initiated, we do nothing
2442  */
2443
2444 static void bc_destroy(struct rpc_xprt *xprt)
2445 {
2446         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2447
2448         xs_xprt_free(xprt);
2449         module_put(THIS_MODULE);
2450 }
2451
2452 static struct rpc_xprt_ops xs_local_ops = {
2453         .reserve_xprt           = xprt_reserve_xprt,
2454         .release_xprt           = xs_tcp_release_xprt,
2455         .alloc_slot             = xprt_alloc_slot,
2456         .rpcbind                = xs_local_rpcbind,
2457         .set_port               = xs_local_set_port,
2458         .connect                = xs_local_connect,
2459         .buf_alloc              = rpc_malloc,
2460         .buf_free               = rpc_free,
2461         .send_request           = xs_local_send_request,
2462         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2463         .close                  = xs_close,
2464         .destroy                = xs_destroy,
2465         .print_stats            = xs_local_print_stats,
2466 };
2467
2468 static struct rpc_xprt_ops xs_udp_ops = {
2469         .set_buffer_size        = xs_udp_set_buffer_size,
2470         .reserve_xprt           = xprt_reserve_xprt_cong,
2471         .release_xprt           = xprt_release_xprt_cong,
2472         .alloc_slot             = xprt_alloc_slot,
2473         .rpcbind                = rpcb_getport_async,
2474         .set_port               = xs_set_port,
2475         .connect                = xs_connect,
2476         .buf_alloc              = rpc_malloc,
2477         .buf_free               = rpc_free,
2478         .send_request           = xs_udp_send_request,
2479         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2480         .timer                  = xs_udp_timer,
2481         .release_request        = xprt_release_rqst_cong,
2482         .close                  = xs_close,
2483         .destroy                = xs_destroy,
2484         .print_stats            = xs_udp_print_stats,
2485 };
2486
2487 static struct rpc_xprt_ops xs_tcp_ops = {
2488         .reserve_xprt           = xprt_reserve_xprt,
2489         .release_xprt           = xs_tcp_release_xprt,
2490         .alloc_slot             = xprt_lock_and_alloc_slot,
2491         .rpcbind                = rpcb_getport_async,
2492         .set_port               = xs_set_port,
2493         .connect                = xs_connect,
2494         .buf_alloc              = rpc_malloc,
2495         .buf_free               = rpc_free,
2496         .send_request           = xs_tcp_send_request,
2497         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2498         .close                  = xs_tcp_shutdown,
2499         .destroy                = xs_destroy,
2500         .print_stats            = xs_tcp_print_stats,
2501 };
2502
2503 /*
2504  * The rpc_xprt_ops for the server backchannel
2505  */
2506
2507 static struct rpc_xprt_ops bc_tcp_ops = {
2508         .reserve_xprt           = xprt_reserve_xprt,
2509         .release_xprt           = xprt_release_xprt,
2510         .alloc_slot             = xprt_alloc_slot,
2511         .buf_alloc              = bc_malloc,
2512         .buf_free               = bc_free,
2513         .send_request           = bc_send_request,
2514         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2515         .close                  = bc_close,
2516         .destroy                = bc_destroy,
2517         .print_stats            = xs_tcp_print_stats,
2518 };
2519
2520 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2521 {
2522         static const struct sockaddr_in sin = {
2523                 .sin_family             = AF_INET,
2524                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2525         };
2526         static const struct sockaddr_in6 sin6 = {
2527                 .sin6_family            = AF_INET6,
2528                 .sin6_addr              = IN6ADDR_ANY_INIT,
2529         };
2530
2531         switch (family) {
2532         case AF_LOCAL:
2533                 break;
2534         case AF_INET:
2535                 memcpy(sap, &sin, sizeof(sin));
2536                 break;
2537         case AF_INET6:
2538                 memcpy(sap, &sin6, sizeof(sin6));
2539                 break;
2540         default:
2541                 dprintk("RPC:       %s: Bad address family\n", __func__);
2542                 return -EAFNOSUPPORT;
2543         }
2544         return 0;
2545 }
2546
2547 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2548                                       unsigned int slot_table_size,
2549                                       unsigned int max_slot_table_size)
2550 {
2551         struct rpc_xprt *xprt;
2552         struct sock_xprt *new;
2553
2554         if (args->addrlen > sizeof(xprt->addr)) {
2555                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2556                 return ERR_PTR(-EBADF);
2557         }
2558
2559         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2560                         max_slot_table_size);
2561         if (xprt == NULL) {
2562                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2563                                 "rpc_xprt\n");
2564                 return ERR_PTR(-ENOMEM);
2565         }
2566
2567         new = container_of(xprt, struct sock_xprt, xprt);
2568         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2569         xprt->addrlen = args->addrlen;
2570         if (args->srcaddr)
2571                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2572         else {
2573                 int err;
2574                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2575                                         (struct sockaddr *)&new->srcaddr);
2576                 if (err != 0) {
2577                         xprt_free(xprt);
2578                         return ERR_PTR(err);
2579                 }
2580         }
2581
2582         return xprt;
2583 }
2584
2585 static const struct rpc_timeout xs_local_default_timeout = {
2586         .to_initval = 10 * HZ,
2587         .to_maxval = 10 * HZ,
2588         .to_retries = 2,
2589 };
2590
2591 /**
2592  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2593  * @args: rpc transport creation arguments
2594  *
2595  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2596  */
2597 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2598 {
2599         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2600         struct sock_xprt *transport;
2601         struct rpc_xprt *xprt;
2602         struct rpc_xprt *ret;
2603
2604         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2605                         xprt_max_tcp_slot_table_entries);
2606         if (IS_ERR(xprt))
2607                 return xprt;
2608         transport = container_of(xprt, struct sock_xprt, xprt);
2609
2610         xprt->prot = 0;
2611         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2612         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2613
2614         xprt->bind_timeout = XS_BIND_TO;
2615         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2616         xprt->idle_timeout = XS_IDLE_DISC_TO;
2617
2618         xprt->ops = &xs_local_ops;
2619         xprt->timeout = &xs_local_default_timeout;
2620
2621         INIT_DELAYED_WORK(&transport->connect_worker,
2622                         xs_dummy_setup_socket);
2623
2624         switch (sun->sun_family) {
2625         case AF_LOCAL:
2626                 if (sun->sun_path[0] != '/') {
2627                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2628                                         sun->sun_path);
2629                         ret = ERR_PTR(-EINVAL);
2630                         goto out_err;
2631                 }
2632                 xprt_set_bound(xprt);
2633                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2634                 ret = ERR_PTR(xs_local_setup_socket(transport));
2635                 if (ret)
2636                         goto out_err;
2637                 break;
2638         default:
2639                 ret = ERR_PTR(-EAFNOSUPPORT);
2640                 goto out_err;
2641         }
2642
2643         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2644                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2645
2646         if (try_module_get(THIS_MODULE))
2647                 return xprt;
2648         ret = ERR_PTR(-EINVAL);
2649 out_err:
2650         xs_xprt_free(xprt);
2651         return ret;
2652 }
2653
2654 static const struct rpc_timeout xs_udp_default_timeout = {
2655         .to_initval = 5 * HZ,
2656         .to_maxval = 30 * HZ,
2657         .to_increment = 5 * HZ,
2658         .to_retries = 5,
2659 };
2660
2661 /**
2662  * xs_setup_udp - Set up transport to use a UDP socket
2663  * @args: rpc transport creation arguments
2664  *
2665  */
2666 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2667 {
2668         struct sockaddr *addr = args->dstaddr;
2669         struct rpc_xprt *xprt;
2670         struct sock_xprt *transport;
2671         struct rpc_xprt *ret;
2672
2673         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2674                         xprt_udp_slot_table_entries);
2675         if (IS_ERR(xprt))
2676                 return xprt;
2677         transport = container_of(xprt, struct sock_xprt, xprt);
2678
2679         xprt->prot = IPPROTO_UDP;
2680         xprt->tsh_size = 0;
2681         /* XXX: header size can vary due to auth type, IPv6, etc. */
2682         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2683
2684         xprt->bind_timeout = XS_BIND_TO;
2685         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2686         xprt->idle_timeout = XS_IDLE_DISC_TO;
2687
2688         xprt->ops = &xs_udp_ops;
2689
2690         xprt->timeout = &xs_udp_default_timeout;
2691
2692         switch (addr->sa_family) {
2693         case AF_INET:
2694                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2695                         xprt_set_bound(xprt);
2696
2697                 INIT_DELAYED_WORK(&transport->connect_worker,
2698                                         xs_udp_setup_socket);
2699                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2700                 break;
2701         case AF_INET6:
2702                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2703                         xprt_set_bound(xprt);
2704
2705                 INIT_DELAYED_WORK(&transport->connect_worker,
2706                                         xs_udp_setup_socket);
2707                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2708                 break;
2709         default:
2710                 ret = ERR_PTR(-EAFNOSUPPORT);
2711                 goto out_err;
2712         }
2713
2714         if (xprt_bound(xprt))
2715                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2716                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2717                                 xprt->address_strings[RPC_DISPLAY_PORT],
2718                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2719         else
2720                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2721                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2722                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2723
2724         if (try_module_get(THIS_MODULE))
2725                 return xprt;
2726         ret = ERR_PTR(-EINVAL);
2727 out_err:
2728         xs_xprt_free(xprt);
2729         return ret;
2730 }
2731
2732 static const struct rpc_timeout xs_tcp_default_timeout = {
2733         .to_initval = 60 * HZ,
2734         .to_maxval = 60 * HZ,
2735         .to_retries = 2,
2736 };
2737
2738 /**
2739  * xs_setup_tcp - Set up transport to use a TCP socket
2740  * @args: rpc transport creation arguments
2741  *
2742  */
2743 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2744 {
2745         struct sockaddr *addr = args->dstaddr;
2746         struct rpc_xprt *xprt;
2747         struct sock_xprt *transport;
2748         struct rpc_xprt *ret;
2749         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2750
2751         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2752                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2753
2754         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2755                         max_slot_table_size);
2756         if (IS_ERR(xprt))
2757                 return xprt;
2758         transport = container_of(xprt, struct sock_xprt, xprt);
2759
2760         xprt->prot = IPPROTO_TCP;
2761         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2762         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2763
2764         xprt->bind_timeout = XS_BIND_TO;
2765         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2766         xprt->idle_timeout = XS_IDLE_DISC_TO;
2767
2768         xprt->ops = &xs_tcp_ops;
2769         xprt->timeout = &xs_tcp_default_timeout;
2770
2771         switch (addr->sa_family) {
2772         case AF_INET:
2773                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2774                         xprt_set_bound(xprt);
2775
2776                 INIT_DELAYED_WORK(&transport->connect_worker,
2777                                         xs_tcp_setup_socket);
2778                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2779                 break;
2780         case AF_INET6:
2781                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2782                         xprt_set_bound(xprt);
2783
2784                 INIT_DELAYED_WORK(&transport->connect_worker,
2785                                         xs_tcp_setup_socket);
2786                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2787                 break;
2788         default:
2789                 ret = ERR_PTR(-EAFNOSUPPORT);
2790                 goto out_err;
2791         }
2792
2793         if (xprt_bound(xprt))
2794                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2795                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2796                                 xprt->address_strings[RPC_DISPLAY_PORT],
2797                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2798         else
2799                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2800                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2801                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2802
2803         if (try_module_get(THIS_MODULE))
2804                 return xprt;
2805         ret = ERR_PTR(-EINVAL);
2806 out_err:
2807         xs_xprt_free(xprt);
2808         return ret;
2809 }
2810
2811 /**
2812  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2813  * @args: rpc transport creation arguments
2814  *
2815  */
2816 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2817 {
2818         struct sockaddr *addr = args->dstaddr;
2819         struct rpc_xprt *xprt;
2820         struct sock_xprt *transport;
2821         struct svc_sock *bc_sock;
2822         struct rpc_xprt *ret;
2823
2824         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2825                         xprt_tcp_slot_table_entries);
2826         if (IS_ERR(xprt))
2827                 return xprt;
2828         transport = container_of(xprt, struct sock_xprt, xprt);
2829
2830         xprt->prot = IPPROTO_TCP;
2831         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2832         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2833         xprt->timeout = &xs_tcp_default_timeout;
2834
2835         /* backchannel */
2836         xprt_set_bound(xprt);
2837         xprt->bind_timeout = 0;
2838         xprt->reestablish_timeout = 0;
2839         xprt->idle_timeout = 0;
2840
2841         xprt->ops = &bc_tcp_ops;
2842
2843         switch (addr->sa_family) {
2844         case AF_INET:
2845                 xs_format_peer_addresses(xprt, "tcp",
2846                                          RPCBIND_NETID_TCP);
2847                 break;
2848         case AF_INET6:
2849                 xs_format_peer_addresses(xprt, "tcp",
2850                                    RPCBIND_NETID_TCP6);
2851                 break;
2852         default:
2853                 ret = ERR_PTR(-EAFNOSUPPORT);
2854                 goto out_err;
2855         }
2856
2857         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2858                         xprt->address_strings[RPC_DISPLAY_ADDR],
2859                         xprt->address_strings[RPC_DISPLAY_PORT],
2860                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2861
2862         /*
2863          * Once we've associated a backchannel xprt with a connection,
2864          * we want to keep it around as long as the connection lasts,
2865          * in case we need to start using it for a backchannel again;
2866          * this reference won't be dropped until bc_xprt is destroyed.
2867          */
2868         xprt_get(xprt);
2869         args->bc_xprt->xpt_bc_xprt = xprt;
2870         xprt->bc_xprt = args->bc_xprt;
2871         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2872         transport->sock = bc_sock->sk_sock;
2873         transport->inet = bc_sock->sk_sk;
2874
2875         /*
2876          * Since we don't want connections for the backchannel, we set
2877          * the xprt status to connected
2878          */
2879         xprt_set_connected(xprt);
2880
2881         if (try_module_get(THIS_MODULE))
2882                 return xprt;
2883
2884         args->bc_xprt->xpt_bc_xprt = NULL;
2885         xprt_put(xprt);
2886         ret = ERR_PTR(-EINVAL);
2887 out_err:
2888         xs_xprt_free(xprt);
2889         return ret;
2890 }
2891
2892 static struct xprt_class        xs_local_transport = {
2893         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2894         .name           = "named UNIX socket",
2895         .owner          = THIS_MODULE,
2896         .ident          = XPRT_TRANSPORT_LOCAL,
2897         .setup          = xs_setup_local,
2898 };
2899
2900 static struct xprt_class        xs_udp_transport = {
2901         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2902         .name           = "udp",
2903         .owner          = THIS_MODULE,
2904         .ident          = XPRT_TRANSPORT_UDP,
2905         .setup          = xs_setup_udp,
2906 };
2907
2908 static struct xprt_class        xs_tcp_transport = {
2909         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2910         .name           = "tcp",
2911         .owner          = THIS_MODULE,
2912         .ident          = XPRT_TRANSPORT_TCP,
2913         .setup          = xs_setup_tcp,
2914 };
2915
2916 static struct xprt_class        xs_bc_tcp_transport = {
2917         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2918         .name           = "tcp NFSv4.1 backchannel",
2919         .owner          = THIS_MODULE,
2920         .ident          = XPRT_TRANSPORT_BC_TCP,
2921         .setup          = xs_setup_bc_tcp,
2922 };
2923
2924 /**
2925  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2926  *
2927  */
2928 int init_socket_xprt(void)
2929 {
2930 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2931         if (!sunrpc_table_header)
2932                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2933 #endif
2934
2935         xprt_register_transport(&xs_local_transport);
2936         xprt_register_transport(&xs_udp_transport);
2937         xprt_register_transport(&xs_tcp_transport);
2938         xprt_register_transport(&xs_bc_tcp_transport);
2939
2940         return 0;
2941 }
2942
2943 /**
2944  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2945  *
2946  */
2947 void cleanup_socket_xprt(void)
2948 {
2949 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2950         if (sunrpc_table_header) {
2951                 unregister_sysctl_table(sunrpc_table_header);
2952                 sunrpc_table_header = NULL;
2953         }
2954 #endif
2955
2956         xprt_unregister_transport(&xs_local_transport);
2957         xprt_unregister_transport(&xs_udp_transport);
2958         xprt_unregister_transport(&xs_tcp_transport);
2959         xprt_unregister_transport(&xs_bc_tcp_transport);
2960 }
2961
2962 static int param_set_uint_minmax(const char *val,
2963                 const struct kernel_param *kp,
2964                 unsigned int min, unsigned int max)
2965 {
2966         unsigned int num;
2967         int ret;
2968
2969         if (!val)
2970                 return -EINVAL;
2971         ret = kstrtouint(val, 0, &num);
2972         if (ret == -EINVAL || num < min || num > max)
2973                 return -EINVAL;
2974         *((unsigned int *)kp->arg) = num;
2975         return 0;
2976 }
2977
2978 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2979 {
2980         return param_set_uint_minmax(val, kp,
2981                         RPC_MIN_RESVPORT,
2982                         RPC_MAX_RESVPORT);
2983 }
2984
2985 static struct kernel_param_ops param_ops_portnr = {
2986         .set = param_set_portnr,
2987         .get = param_get_uint,
2988 };
2989
2990 #define param_check_portnr(name, p) \
2991         __param_check(name, p, unsigned int);
2992
2993 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2994 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2995
2996 static int param_set_slot_table_size(const char *val,
2997                                      const struct kernel_param *kp)
2998 {
2999         return param_set_uint_minmax(val, kp,
3000                         RPC_MIN_SLOT_TABLE,
3001                         RPC_MAX_SLOT_TABLE);
3002 }
3003
3004 static struct kernel_param_ops param_ops_slot_table_size = {
3005         .set = param_set_slot_table_size,
3006         .get = param_get_uint,
3007 };
3008
3009 #define param_check_slot_table_size(name, p) \
3010         __param_check(name, p, unsigned int);
3011
3012 static int param_set_max_slot_table_size(const char *val,
3013                                      const struct kernel_param *kp)
3014 {
3015         return param_set_uint_minmax(val, kp,
3016                         RPC_MIN_SLOT_TABLE,
3017                         RPC_MAX_SLOT_TABLE_LIMIT);
3018 }
3019
3020 static struct kernel_param_ops param_ops_max_slot_table_size = {
3021         .set = param_set_max_slot_table_size,
3022         .get = param_get_uint,
3023 };
3024
3025 #define param_check_max_slot_table_size(name, p) \
3026         __param_check(name, p, unsigned int);
3027
3028 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3029                    slot_table_size, 0644);
3030 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3031                    max_slot_table_size, 0644);
3032 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3033                    slot_table_size, 0644);
3034