Pileus Git - ~andy/linux/blob - 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 #define XS_TCP_LINGER_TO        (15U * HZ)
  67 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  68
  69 /*
  70  * We can register our own files under /proc/sys/sunrpc by
  71  * calling register_sysctl_table() again.  The files in that
  72  * directory become the union of all files registered there.
  73  *
  74  * We simply need to make sure that we don't collide with
  75  * someone else's file names!
  76  */
  77
  78 #ifdef RPC_DEBUG
  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 #ifdef RPC_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 struct sock_xprt {
 220         struct rpc_xprt         xprt;
 221
 222         /*
 223          * Network layer
 224          */
 225         struct socket *         sock;
 226         struct sock *           inet;
 227
 228         /*
 229          * State of TCP reply receive
 230          */
 231         __be32                  tcp_fraghdr,
 232                                 tcp_xid,
 233                                 tcp_calldir;
 234
 235         u32                     tcp_offset,
 236                                 tcp_reclen;
 237
 238         unsigned long           tcp_copied,
 239                                 tcp_flags;
 240
 241         /*
 242          * Connection of transports
 243          */
 244         struct delayed_work     connect_worker;
 245         struct sockaddr_storage srcaddr;
 246         unsigned short          srcport;
 247
 248         /*
 249          * UDP socket buffer size parameters
 250          */
 251         size_t                  rcvsize,
 252                                 sndsize;
 253
 254         /*
 255          * Saved socket callback addresses
 256          */
 257         void                    (*old_data_ready)(struct sock *, int);
 258         void                    (*old_state_change)(struct sock *);
 259         void                    (*old_write_space)(struct sock *);
 260 };
 261
 262 /*
 263  * TCP receive state flags
 264  */
 265 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 266 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 267 #define TCP_RCV_COPY_XID        (1UL << 2)
 268 #define TCP_RCV_COPY_DATA       (1UL << 3)
 269 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 270 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 271
 272 /*
 273  * TCP RPC flags
 274  */
 275 #define TCP_RPC_REPLY           (1UL << 6)
 276
 277 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 278 {
 279         return (struct sockaddr *) &xprt->addr;
 280 }
 281
 282 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 283 {
 284         return (struct sockaddr_un *) &xprt->addr;
 285 }
 286
 287 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 288 {
 289         return (struct sockaddr_in *) &xprt->addr;
 290 }
 291
 292 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 293 {
 294         return (struct sockaddr_in6 *) &xprt->addr;
 295 }
 296
 297 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 298 {
 299         struct sockaddr *sap = xs_addr(xprt);
 300         struct sockaddr_in6 *sin6;
 301         struct sockaddr_in *sin;
 302         struct sockaddr_un *sun;
 303         char buf[128];
 304
 305         switch (sap->sa_family) {
 306         case AF_LOCAL:
 307                 sun = xs_addr_un(xprt);
 308                 strlcpy(buf, sun->sun_path, sizeof(buf));
 309                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 310                                                 kstrdup(buf, GFP_KERNEL);
 311                 break;
 312         case AF_INET:
 313                 (void)rpc_ntop(sap, buf, sizeof(buf));
 314                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 315                                                 kstrdup(buf, GFP_KERNEL);
 316                 sin = xs_addr_in(xprt);
 317                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 318                 break;
 319         case AF_INET6:
 320                 (void)rpc_ntop(sap, buf, sizeof(buf));
 321                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 322                                                 kstrdup(buf, GFP_KERNEL);
 323                 sin6 = xs_addr_in6(xprt);
 324                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 325                 break;
 326         default:
 327                 BUG();
 328         }
 329
 330         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 331 }
 332
 333 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 334 {
 335         struct sockaddr *sap = xs_addr(xprt);
 336         char buf[128];
 337
 338         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 339         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 340
 341         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 342         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 343 }
 344
 345 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 346                                      const char *protocol,
 347                                      const char *netid)
 348 {
 349         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 350         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 351         xs_format_common_peer_addresses(xprt);
 352         xs_format_common_peer_ports(xprt);
 353 }
 354
 355 static void xs_update_peer_port(struct rpc_xprt *xprt)
 356 {
 357         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 358         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 359
 360         xs_format_common_peer_ports(xprt);
 361 }
 362
 363 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 364 {
 365         unsigned int i;
 366
 367         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 368                 switch (i) {
 369                 case RPC_DISPLAY_PROTO:
 370                 case RPC_DISPLAY_NETID:
 371                         continue;
 372                 default:
 373                         kfree(xprt->address_strings[i]);
 374                 }
 375 }
 376
 377 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 378
 379 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 380 {
 381         struct msghdr msg = {
 382                 .msg_name       = addr,
 383                 .msg_namelen    = addrlen,
 384                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 385         };
 386         struct kvec iov = {
 387                 .iov_base       = vec->iov_base + base,
 388                 .iov_len        = vec->iov_len - base,
 389         };
 390
 391         if (iov.iov_len != 0)
 392                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 393         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 394 }
 395
 396 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
 397 {
 398         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 399                         int offset, size_t size, int flags);
 400         struct page **ppage;
 401         unsigned int remainder;
 402         int err, sent = 0;
 403
 404         remainder = xdr->page_len - base;
 405         base += xdr->page_base;
 406         ppage = xdr->pages + (base >> PAGE_SHIFT);
 407         base &= ~PAGE_MASK;
 408         do_sendpage = sock->ops->sendpage;
 409         if (!zerocopy)
 410                 do_sendpage = sock_no_sendpage;
 411         for(;;) {
 412                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 413                 int flags = XS_SENDMSG_FLAGS;
 414
 415                 remainder -= len;
 416                 if (remainder != 0 || more)
 417                         flags |= MSG_MORE;
 418                 err = do_sendpage(sock, *ppage, base, len, flags);
 419                 if (remainder == 0 || err != len)
 420                         break;
 421                 sent += err;
 422                 ppage++;
 423                 base = 0;
 424         }
 425         if (sent == 0)
 426                 return err;
 427         if (err > 0)
 428                 sent += err;
 429         return sent;
 430 }
 431
 432 /**
 433  * xs_sendpages - write pages directly to a socket
 434  * @sock: socket to send on
 435  * @addr: UDP only -- address of destination
 436  * @addrlen: UDP only -- length of destination address
 437  * @xdr: buffer containing this request
 438  * @base: starting position in the buffer
 439  * @zerocopy: true if it is safe to use sendpage()
 440  *
 441  */
 442 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
 443 {
 444         unsigned int remainder = xdr->len - base;
 445         int err, sent = 0;
 446
 447         if (unlikely(!sock))
 448                 return -ENOTSOCK;
 449
 450         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 451         if (base != 0) {
 452                 addr = NULL;
 453                 addrlen = 0;
 454         }
 455
 456         if (base < xdr->head[0].iov_len || addr != NULL) {
 457                 unsigned int len = xdr->head[0].iov_len - base;
 458                 remainder -= len;
 459                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 460                 if (remainder == 0 || err != len)
 461                         goto out;
 462                 sent += err;
 463                 base = 0;
 464         } else
 465                 base -= xdr->head[0].iov_len;
 466
 467         if (base < xdr->page_len) {
 468                 unsigned int len = xdr->page_len - base;
 469                 remainder -= len;
 470                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
 471                 if (remainder == 0 || err != len)
 472                         goto out;
 473                 sent += err;
 474                 base = 0;
 475         } else
 476                 base -= xdr->page_len;
 477
 478         if (base >= xdr->tail[0].iov_len)
 479                 return sent;
 480         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 481 out:
 482         if (sent == 0)
 483                 return err;
 484         if (err > 0)
 485                 sent += err;
 486         return sent;
 487 }
 488
 489 static void xs_nospace_callback(struct rpc_task *task)
 490 {
 491         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 492
 493         transport->inet->sk_write_pending--;
 494         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 495 }
 496
 497 /**
 498  * xs_nospace - place task on wait queue if transmit was incomplete
 499  * @task: task to put to sleep
 500  *
 501  */
 502 static int xs_nospace(struct rpc_task *task)
 503 {
 504         struct rpc_rqst *req = task->tk_rqstp;
 505         struct rpc_xprt *xprt = req->rq_xprt;
 506         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 507         int ret = -EAGAIN;
 508
 509         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 510                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 511                         req->rq_slen);
 512
 513         /* Protect against races with write_space */
 514         spin_lock_bh(&xprt->transport_lock);
 515
 516         /* Don't race with disconnect */
 517         if (xprt_connected(xprt)) {
 518                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 519                         /*
 520                          * Notify TCP that we're limited by the application
 521                          * window size
 522                          */
 523                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 524                         transport->inet->sk_write_pending++;
 525                         /* ...and wait for more buffer space */
 526                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 527                 }
 528         } else {
 529                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 530                 ret = -ENOTCONN;
 531         }
 532
 533         spin_unlock_bh(&xprt->transport_lock);
 534         return ret;
 535 }
 536
 537 /*
 538  * Construct a stream transport record marker in @buf.
 539  */
 540 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 541 {
 542         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 543         rpc_fraghdr *base = buf->head[0].iov_base;
 544         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 545 }
 546
 547 /**
 548  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 549  * @task: RPC task that manages the state of an RPC request
 550  *
 551  * Return values:
 552  *        0:    The request has been sent
 553  *   EAGAIN:    The socket was blocked, please call again later to
 554  *              complete the request
 555  * ENOTCONN:    Caller needs to invoke connect logic then call again
 556  *    other:    Some other error occured, the request was not sent
 557  */
 558 static int xs_local_send_request(struct rpc_task *task)
 559 {
 560         struct rpc_rqst *req = task->tk_rqstp;
 561         struct rpc_xprt *xprt = req->rq_xprt;
 562         struct sock_xprt *transport =
 563                                 container_of(xprt, struct sock_xprt, xprt);
 564         struct xdr_buf *xdr = &req->rq_snd_buf;
 565         int status;
 566
 567         xs_encode_stream_record_marker(&req->rq_snd_buf);
 568
 569         xs_pktdump("packet data:",
 570                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 571
 572         status = xs_sendpages(transport->sock, NULL, 0,
 573                                                 xdr, req->rq_bytes_sent, true);
 574         dprintk("RPC:       %s(%u) = %d\n",
 575                         __func__, xdr->len - req->rq_bytes_sent, status);
 576         if (likely(status >= 0)) {
 577                 req->rq_bytes_sent += status;
 578                 req->rq_xmit_bytes_sent += status;
 579                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 580                         req->rq_bytes_sent = 0;
 581                         return 0;
 582                 }
 583                 status = -EAGAIN;
 584         }
 585
 586         switch (status) {
 587         case -EAGAIN:
 588                 status = xs_nospace(task);
 589                 break;
 590         default:
 591                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 592                         -status);
 593         case -EPIPE:
 594                 xs_close(xprt);
 595                 status = -ENOTCONN;
 596         }
 597
 598         return status;
 599 }
 600
 601 /**
 602  * xs_udp_send_request - write an RPC request to a UDP socket
 603  * @task: address of RPC task that manages the state of an RPC request
 604  *
 605  * Return values:
 606  *        0:    The request has been sent
 607  *   EAGAIN:    The socket was blocked, please call again later to
 608  *              complete the request
 609  * ENOTCONN:    Caller needs to invoke connect logic then call again
 610  *    other:    Some other error occurred, the request was not sent
 611  */
 612 static int xs_udp_send_request(struct rpc_task *task)
 613 {
 614         struct rpc_rqst *req = task->tk_rqstp;
 615         struct rpc_xprt *xprt = req->rq_xprt;
 616         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 617         struct xdr_buf *xdr = &req->rq_snd_buf;
 618         int status;
 619
 620         xs_pktdump("packet data:",
 621                                 req->rq_svec->iov_base,
 622                                 req->rq_svec->iov_len);
 623
 624         if (!xprt_bound(xprt))
 625                 return -ENOTCONN;
 626         status = xs_sendpages(transport->sock,
 627                               xs_addr(xprt),
 628                               xprt->addrlen, xdr,
 629                               req->rq_bytes_sent, true);
 630
 631         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 632                         xdr->len - req->rq_bytes_sent, status);
 633
 634         if (status >= 0) {
 635                 req->rq_xmit_bytes_sent += status;
 636                 if (status >= req->rq_slen)
 637                         return 0;
 638                 /* Still some bytes left; set up for a retry later. */
 639                 status = -EAGAIN;
 640         }
 641
 642         switch (status) {
 643         case -ENOTSOCK:
 644                 status = -ENOTCONN;
 645                 /* Should we call xs_close() here? */
 646                 break;
 647         case -EAGAIN:
 648                 status = xs_nospace(task);
 649                 break;
 650         default:
 651                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 652                         -status);
 653         case -ENETUNREACH:
 654         case -EPIPE:
 655         case -ECONNREFUSED:
 656                 /* When the server has died, an ICMP port unreachable message
 657                  * prompts ECONNREFUSED. */
 658                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 659         }
 660
 661         return status;
 662 }
 663
 664 /**
 665  * xs_tcp_shutdown - gracefully shut down a TCP socket
 666  * @xprt: transport
 667  *
 668  * Initiates a graceful shutdown of the TCP socket by calling the
 669  * equivalent of shutdown(SHUT_WR);
 670  */
 671 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 672 {
 673         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 674         struct socket *sock = transport->sock;
 675
 676         if (sock != NULL) {
 677                 kernel_sock_shutdown(sock, SHUT_WR);
 678                 trace_rpc_socket_shutdown(xprt, sock);
 679         }
 680 }
 681
 682 /**
 683  * xs_tcp_send_request - write an RPC request to a TCP socket
 684  * @task: address of RPC task that manages the state of an RPC request
 685  *
 686  * Return values:
 687  *        0:    The request has been sent
 688  *   EAGAIN:    The socket was blocked, please call again later to
 689  *              complete the request
 690  * ENOTCONN:    Caller needs to invoke connect logic then call again
 691  *    other:    Some other error occurred, the request was not sent
 692  *
 693  * XXX: In the case of soft timeouts, should we eventually give up
 694  *      if sendmsg is not able to make progress?
 695  */
 696 static int xs_tcp_send_request(struct rpc_task *task)
 697 {
 698         struct rpc_rqst *req = task->tk_rqstp;
 699         struct rpc_xprt *xprt = req->rq_xprt;
 700         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 701         struct xdr_buf *xdr = &req->rq_snd_buf;
 702         bool zerocopy = true;
 703         int status;
 704
 705         xs_encode_stream_record_marker(&req->rq_snd_buf);
 706
 707         xs_pktdump("packet data:",
 708                                 req->rq_svec->iov_base,
 709                                 req->rq_svec->iov_len);
 710         /* Don't use zero copy if this is a resend. If the RPC call
 711          * completes while the socket holds a reference to the pages,
 712          * then we may end up resending corrupted data.
 713          */
 714         if (task->tk_flags & RPC_TASK_SENT)
 715                 zerocopy = false;
 716
 717         /* Continue transmitting the packet/record. We must be careful
 718          * to cope with writespace callbacks arriving _after_ we have
 719          * called sendmsg(). */
 720         while (1) {
 721                 status = xs_sendpages(transport->sock,
 722                                         NULL, 0, xdr, req->rq_bytes_sent,
 723                                         zerocopy);
 724
 725                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 726                                 xdr->len - req->rq_bytes_sent, status);
 727
 728                 if (unlikely(status < 0))
 729                         break;
 730
 731                 /* If we've sent the entire packet, immediately
 732                  * reset the count of bytes sent. */
 733                 req->rq_bytes_sent += status;
 734                 req->rq_xmit_bytes_sent += status;
 735                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 736                         req->rq_bytes_sent = 0;
 737                         return 0;
 738                 }
 739
 740                 if (status != 0)
 741                         continue;
 742                 status = -EAGAIN;
 743                 break;
 744         }
 745
 746         switch (status) {
 747         case -ENOTSOCK:
 748                 status = -ENOTCONN;
 749                 /* Should we call xs_close() here? */
 750                 break;
 751         case -EAGAIN:
 752                 status = xs_nospace(task);
 753                 break;
 754         default:
 755                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 756                         -status);
 757         case -ECONNRESET:
 758                 xs_tcp_shutdown(xprt);
 759         case -ECONNREFUSED:
 760         case -ENOTCONN:
 761         case -EPIPE:
 762                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 763         }
 764
 765         return status;
 766 }
 767
 768 /**
 769  * xs_tcp_release_xprt - clean up after a tcp transmission
 770  * @xprt: transport
 771  * @task: rpc task
 772  *
 773  * This cleans up if an error causes us to abort the transmission of a request.
 774  * In this case, the socket may need to be reset in order to avoid confusing
 775  * the server.
 776  */
 777 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 778 {
 779         struct rpc_rqst *req;
 780
 781         if (task != xprt->snd_task)
 782                 return;
 783         if (task == NULL)
 784                 goto out_release;
 785         req = task->tk_rqstp;
 786         if (req == NULL)
 787                 goto out_release;
 788         if (req->rq_bytes_sent == 0)
 789                 goto out_release;
 790         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 791                 goto out_release;
 792         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 793 out_release:
 794         xprt_release_xprt(xprt, task);
 795 }
 796
 797 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 798 {
 799         transport->old_data_ready = sk->sk_data_ready;
 800         transport->old_state_change = sk->sk_state_change;
 801         transport->old_write_space = sk->sk_write_space;
 802 }
 803
 804 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 805 {
 806         sk->sk_data_ready = transport->old_data_ready;
 807         sk->sk_state_change = transport->old_state_change;
 808         sk->sk_write_space = transport->old_write_space;
 809 }
 810
 811 static void xs_reset_transport(struct sock_xprt *transport)
 812 {
 813         struct socket *sock = transport->sock;
 814         struct sock *sk = transport->inet;
 815
 816         if (sk == NULL)
 817                 return;
 818
 819         transport->srcport = 0;
 820
 821         write_lock_bh(&sk->sk_callback_lock);
 822         transport->inet = NULL;
 823         transport->sock = NULL;
 824
 825         sk->sk_user_data = NULL;
 826
 827         xs_restore_old_callbacks(transport, sk);
 828         write_unlock_bh(&sk->sk_callback_lock);
 829
 830         sk->sk_no_check = 0;
 831
 832         trace_rpc_socket_close(&transport->xprt, sock);
 833         sock_release(sock);
 834 }
 835
 836 /**
 837  * xs_close - close a socket
 838  * @xprt: transport
 839  *
 840  * This is used when all requests are complete; ie, no DRC state remains
 841  * on the server we want to save.
 842  *
 843  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 844  * xs_reset_transport() zeroing the socket from underneath a writer.
 845  */
 846 static void xs_close(struct rpc_xprt *xprt)
 847 {
 848         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 849
 850         dprintk("RPC:       xs_close xprt %p\n", xprt);
 851
 852         cancel_delayed_work_sync(&transport->connect_worker);
 853
 854         xs_reset_transport(transport);
 855         xprt->reestablish_timeout = 0;
 856
 857         smp_mb__before_clear_bit();
 858         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 859         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 860         clear_bit(XPRT_CLOSING, &xprt->state);
 861         smp_mb__after_clear_bit();
 862         xprt_disconnect_done(xprt);
 863 }
 864
 865 static void xs_tcp_close(struct rpc_xprt *xprt)
 866 {
 867         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 868                 xs_close(xprt);
 869         else
 870                 xs_tcp_shutdown(xprt);
 871 }
 872
 873 /**
 874  * xs_destroy - prepare to shutdown a transport
 875  * @xprt: doomed transport
 876  *
 877  */
 878 static void xs_destroy(struct rpc_xprt *xprt)
 879 {
 880         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 881
 882         xs_close(xprt);
 883         xs_free_peer_addresses(xprt);
 884         xprt_free(xprt);
 885         module_put(THIS_MODULE);
 886 }
 887
 888 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 889 {
 890         return (struct rpc_xprt *) sk->sk_user_data;
 891 }
 892
 893 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 894 {
 895         struct xdr_skb_reader desc = {
 896                 .skb            = skb,
 897                 .offset         = sizeof(rpc_fraghdr),
 898                 .count          = skb->len - sizeof(rpc_fraghdr),
 899         };
 900
 901         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 902                 return -1;
 903         if (desc.count)
 904                 return -1;
 905         return 0;
 906 }
 907
 908 /**
 909  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 910  * @sk: socket with data to read
 911  * @len: how much data to read
 912  *
 913  * Currently this assumes we can read the whole reply in a single gulp.
 914  */
 915 static void xs_local_data_ready(struct sock *sk, int len)
 916 {
 917         struct rpc_task *task;
 918         struct rpc_xprt *xprt;
 919         struct rpc_rqst *rovr;
 920         struct sk_buff *skb;
 921         int err, repsize, copied;
 922         u32 _xid;
 923         __be32 *xp;
 924
 925         read_lock_bh(&sk->sk_callback_lock);
 926         dprintk("RPC:       %s...\n", __func__);
 927         xprt = xprt_from_sock(sk);
 928         if (xprt == NULL)
 929                 goto out;
 930
 931         skb = skb_recv_datagram(sk, 0, 1, &err);
 932         if (skb == NULL)
 933                 goto out;
 934
 935         repsize = skb->len - sizeof(rpc_fraghdr);
 936         if (repsize < 4) {
 937                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 938                 goto dropit;
 939         }
 940
 941         /* Copy the XID from the skb... */
 942         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 943         if (xp == NULL)
 944                 goto dropit;
 945
 946         /* Look up and lock the request corresponding to the given XID */
 947         spin_lock(&xprt->transport_lock);
 948         rovr = xprt_lookup_rqst(xprt, *xp);
 949         if (!rovr)
 950                 goto out_unlock;
 951         task = rovr->rq_task;
 952
 953         copied = rovr->rq_private_buf.buflen;
 954         if (copied > repsize)
 955                 copied = repsize;
 956
 957         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 958                 dprintk("RPC:       sk_buff copy failed\n");
 959                 goto out_unlock;
 960         }
 961
 962         xprt_complete_rqst(task, copied);
 963
 964  out_unlock:
 965         spin_unlock(&xprt->transport_lock);
 966  dropit:
 967         skb_free_datagram(sk, skb);
 968  out:
 969         read_unlock_bh(&sk->sk_callback_lock);
 970 }
 971
 972 /**
 973  * xs_udp_data_ready - "data ready" callback for UDP sockets
 974  * @sk: socket with data to read
 975  * @len: how much data to read
 976  *
 977  */
 978 static void xs_udp_data_ready(struct sock *sk, int len)
 979 {
 980         struct rpc_task *task;
 981         struct rpc_xprt *xprt;
 982         struct rpc_rqst *rovr;
 983         struct sk_buff *skb;
 984         int err, repsize, copied;
 985         u32 _xid;
 986         __be32 *xp;
 987
 988         read_lock_bh(&sk->sk_callback_lock);
 989         dprintk("RPC:       xs_udp_data_ready...\n");
 990         if (!(xprt = xprt_from_sock(sk)))
 991                 goto out;
 992
 993         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 994                 goto out;
 995
 996         repsize = skb->len - sizeof(struct udphdr);
 997         if (repsize < 4) {
 998                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 999                 goto dropit;
1000         }
1001
1002         /* Copy the XID from the skb... */
1003         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1004                                 sizeof(_xid), &_xid);
1005         if (xp == NULL)
1006                 goto dropit;
1007
1008         /* Look up and lock the request corresponding to the given XID */
1009         spin_lock(&xprt->transport_lock);
1010         rovr = xprt_lookup_rqst(xprt, *xp);
1011         if (!rovr)
1012                 goto out_unlock;
1013         task = rovr->rq_task;
1014
1015         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1016                 copied = repsize;
1017
1018         /* Suck it into the iovec, verify checksum if not done by hw. */
1019         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1020                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1021                 goto out_unlock;
1022         }
1023
1024         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1025
1026         xprt_adjust_cwnd(xprt, task, copied);
1027         xprt_complete_rqst(task, copied);
1028
1029  out_unlock:
1030         spin_unlock(&xprt->transport_lock);
1031  dropit:
1032         skb_free_datagram(sk, skb);
1033  out:
1034         read_unlock_bh(&sk->sk_callback_lock);
1035 }
1036
1037 /*
1038  * Helper function to force a TCP close if the server is sending
1039  * junk and/or it has put us in CLOSE_WAIT
1040  */
1041 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1042 {
1043         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1044         xprt_force_disconnect(xprt);
1045 }
1046
1047 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1048 {
1049         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1050         size_t len, used;
1051         char *p;
1052
1053         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1054         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1055         used = xdr_skb_read_bits(desc, p, len);
1056         transport->tcp_offset += used;
1057         if (used != len)
1058                 return;
1059
1060         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1061         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1062                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1063         else
1064                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1065         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1066
1067         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1068         transport->tcp_offset = 0;
1069
1070         /* Sanity check of the record length */
1071         if (unlikely(transport->tcp_reclen < 8)) {
1072                 dprintk("RPC:       invalid TCP record fragment length\n");
1073                 xs_tcp_force_close(xprt);
1074                 return;
1075         }
1076         dprintk("RPC:       reading TCP record fragment of length %d\n",
1077                         transport->tcp_reclen);
1078 }
1079
1080 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1081 {
1082         if (transport->tcp_offset == transport->tcp_reclen) {
1083                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1084                 transport->tcp_offset = 0;
1085                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1086                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1087                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1088                         transport->tcp_copied = 0;
1089                 }
1090         }
1091 }
1092
1093 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1094 {
1095         size_t len, used;
1096         char *p;
1097
1098         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1099         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1100         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1101         used = xdr_skb_read_bits(desc, p, len);
1102         transport->tcp_offset += used;
1103         if (used != len)
1104                 return;
1105         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1106         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1107         transport->tcp_copied = 4;
1108         dprintk("RPC:       reading %s XID %08x\n",
1109                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1110                                                               : "request with",
1111                         ntohl(transport->tcp_xid));
1112         xs_tcp_check_fraghdr(transport);
1113 }
1114
1115 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1116                                        struct xdr_skb_reader *desc)
1117 {
1118         size_t len, used;
1119         u32 offset;
1120         char *p;
1121
1122         /*
1123          * We want transport->tcp_offset to be 8 at the end of this routine
1124          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1125          * When this function is called for the first time,
1126          * transport->tcp_offset is 4 (after having already read the xid).
1127          */
1128         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1129         len = sizeof(transport->tcp_calldir) - offset;
1130         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1131         p = ((char *) &transport->tcp_calldir) + offset;
1132         used = xdr_skb_read_bits(desc, p, len);
1133         transport->tcp_offset += used;
1134         if (used != len)
1135                 return;
1136         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1137         /*
1138          * We don't yet have the XDR buffer, so we will write the calldir
1139          * out after we get the buffer from the 'struct rpc_rqst'
1140          */
1141         switch (ntohl(transport->tcp_calldir)) {
1142         case RPC_REPLY:
1143                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1144                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1145                 transport->tcp_flags |= TCP_RPC_REPLY;
1146                 break;
1147         case RPC_CALL:
1148                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1149                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1150                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1151                 break;
1152         default:
1153                 dprintk("RPC:       invalid request message type\n");
1154                 xs_tcp_force_close(&transport->xprt);
1155         }
1156         xs_tcp_check_fraghdr(transport);
1157 }
1158
1159 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1160                                      struct xdr_skb_reader *desc,
1161                                      struct rpc_rqst *req)
1162 {
1163         struct sock_xprt *transport =
1164                                 container_of(xprt, struct sock_xprt, xprt);
1165         struct xdr_buf *rcvbuf;
1166         size_t len;
1167         ssize_t r;
1168
1169         rcvbuf = &req->rq_private_buf;
1170
1171         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1172                 /*
1173                  * Save the RPC direction in the XDR buffer
1174                  */
1175                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1176                         &transport->tcp_calldir,
1177                         sizeof(transport->tcp_calldir));
1178                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1179                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1180         }
1181
1182         len = desc->count;
1183         if (len > transport->tcp_reclen - transport->tcp_offset) {
1184                 struct xdr_skb_reader my_desc;
1185
1186                 len = transport->tcp_reclen - transport->tcp_offset;
1187                 memcpy(&my_desc, desc, sizeof(my_desc));
1188                 my_desc.count = len;
1189                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1190                                           &my_desc, xdr_skb_read_bits);
1191                 desc->count -= r;
1192                 desc->offset += r;
1193         } else
1194                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1195                                           desc, xdr_skb_read_bits);
1196
1197         if (r > 0) {
1198                 transport->tcp_copied += r;
1199                 transport->tcp_offset += r;
1200         }
1201         if (r != len) {
1202                 /* Error when copying to the receive buffer,
1203                  * usually because we weren't able to allocate
1204                  * additional buffer pages. All we can do now
1205                  * is turn off TCP_RCV_COPY_DATA, so the request
1206                  * will not receive any additional updates,
1207                  * and time out.
1208                  * Any remaining data from this record will
1209                  * be discarded.
1210                  */
1211                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1212                 dprintk("RPC:       XID %08x truncated request\n",
1213                                 ntohl(transport->tcp_xid));
1214                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1215                                 "tcp_offset = %u, tcp_reclen = %u\n",
1216                                 xprt, transport->tcp_copied,
1217                                 transport->tcp_offset, transport->tcp_reclen);
1218                 return;
1219         }
1220
1221         dprintk("RPC:       XID %08x read %Zd bytes\n",
1222                         ntohl(transport->tcp_xid), r);
1223         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1224                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1225                         transport->tcp_offset, transport->tcp_reclen);
1226
1227         if (transport->tcp_copied == req->rq_private_buf.buflen)
1228                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1229         else if (transport->tcp_offset == transport->tcp_reclen) {
1230                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1231                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1232         }
1233 }
1234
1235 /*
1236  * Finds the request corresponding to the RPC xid and invokes the common
1237  * tcp read code to read the data.
1238  */
1239 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1240                                     struct xdr_skb_reader *desc)
1241 {
1242         struct sock_xprt *transport =
1243                                 container_of(xprt, struct sock_xprt, xprt);
1244         struct rpc_rqst *req;
1245
1246         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1247
1248         /* Find and lock the request corresponding to this xid */
1249         spin_lock(&xprt->transport_lock);
1250         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1251         if (!req) {
1252                 dprintk("RPC:       XID %08x request not found!\n",
1253                                 ntohl(transport->tcp_xid));
1254                 spin_unlock(&xprt->transport_lock);
1255                 return -1;
1256         }
1257
1258         xs_tcp_read_common(xprt, desc, req);
1259
1260         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1261                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1262
1263         spin_unlock(&xprt->transport_lock);
1264         return 0;
1265 }
1266
1267 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1268 /*
1269  * Obtains an rpc_rqst previously allocated and invokes the common
1270  * tcp read code to read the data.  The result is placed in the callback
1271  * queue.
1272  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1273  * connection and return -1.
1274  */
1275 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1276                                        struct xdr_skb_reader *desc)
1277 {
1278         struct sock_xprt *transport =
1279                                 container_of(xprt, struct sock_xprt, xprt);
1280         struct rpc_rqst *req;
1281
1282         req = xprt_alloc_bc_request(xprt);
1283         if (req == NULL) {
1284                 printk(KERN_WARNING "Callback slot table overflowed\n");
1285                 xprt_force_disconnect(xprt);
1286                 return -1;
1287         }
1288
1289         req->rq_xid = transport->tcp_xid;
1290         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1291         xs_tcp_read_common(xprt, desc, req);
1292
1293         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1294                 struct svc_serv *bc_serv = xprt->bc_serv;
1295
1296                 /*
1297                  * Add callback request to callback list.  The callback
1298                  * service sleeps on the sv_cb_waitq waiting for new
1299                  * requests.  Wake it up after adding enqueing the
1300                  * request.
1301                  */
1302                 dprintk("RPC:       add callback request to list\n");
1303                 spin_lock(&bc_serv->sv_cb_lock);
1304                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1305                 spin_unlock(&bc_serv->sv_cb_lock);
1306                 wake_up(&bc_serv->sv_cb_waitq);
1307         }
1308
1309         req->rq_private_buf.len = transport->tcp_copied;
1310
1311         return 0;
1312 }
1313
1314 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1315                                         struct xdr_skb_reader *desc)
1316 {
1317         struct sock_xprt *transport =
1318                                 container_of(xprt, struct sock_xprt, xprt);
1319
1320         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1321                 xs_tcp_read_reply(xprt, desc) :
1322                 xs_tcp_read_callback(xprt, desc);
1323 }
1324 #else
1325 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1326                                         struct xdr_skb_reader *desc)
1327 {
1328         return xs_tcp_read_reply(xprt, desc);
1329 }
1330 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1331
1332 /*
1333  * Read data off the transport.  This can be either an RPC_CALL or an
1334  * RPC_REPLY.  Relay the processing to helper functions.
1335  */
1336 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1337                                     struct xdr_skb_reader *desc)
1338 {
1339         struct sock_xprt *transport =
1340                                 container_of(xprt, struct sock_xprt, xprt);
1341
1342         if (_xs_tcp_read_data(xprt, desc) == 0)
1343                 xs_tcp_check_fraghdr(transport);
1344         else {
1345                 /*
1346                  * The transport_lock protects the request handling.
1347                  * There's no need to hold it to update the tcp_flags.
1348                  */
1349                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1350         }
1351 }
1352
1353 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1354 {
1355         size_t len;
1356
1357         len = transport->tcp_reclen - transport->tcp_offset;
1358         if (len > desc->count)
1359                 len = desc->count;
1360         desc->count -= len;
1361         desc->offset += len;
1362         transport->tcp_offset += len;
1363         dprintk("RPC:       discarded %Zu bytes\n", len);
1364         xs_tcp_check_fraghdr(transport);
1365 }
1366
1367 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1368 {
1369         struct rpc_xprt *xprt = rd_desc->arg.data;
1370         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1371         struct xdr_skb_reader desc = {
1372                 .skb    = skb,
1373                 .offset = offset,
1374                 .count  = len,
1375         };
1376
1377         dprintk("RPC:       xs_tcp_data_recv started\n");
1378         do {
1379                 /* Read in a new fragment marker if necessary */
1380                 /* Can we ever really expect to get completely empty fragments? */
1381                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1382                         xs_tcp_read_fraghdr(xprt, &desc);
1383                         continue;
1384                 }
1385                 /* Read in the xid if necessary */
1386                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1387                         xs_tcp_read_xid(transport, &desc);
1388                         continue;
1389                 }
1390                 /* Read in the call/reply flag */
1391                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1392                         xs_tcp_read_calldir(transport, &desc);
1393                         continue;
1394                 }
1395                 /* Read in the request data */
1396                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1397                         xs_tcp_read_data(xprt, &desc);
1398                         continue;
1399                 }
1400                 /* Skip over any trailing bytes on short reads */
1401                 xs_tcp_read_discard(transport, &desc);
1402         } while (desc.count);
1403         dprintk("RPC:       xs_tcp_data_recv done\n");
1404         return len - desc.count;
1405 }
1406
1407 /**
1408  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1409  * @sk: socket with data to read
1410  * @bytes: how much data to read
1411  *
1412  */
1413 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1414 {
1415         struct rpc_xprt *xprt;
1416         read_descriptor_t rd_desc;
1417         int read;
1418
1419         dprintk("RPC:       xs_tcp_data_ready...\n");
1420
1421         read_lock_bh(&sk->sk_callback_lock);
1422         if (!(xprt = xprt_from_sock(sk)))
1423                 goto out;
1424         /* Any data means we had a useful conversation, so
1425          * the we don't need to delay the next reconnect
1426          */
1427         if (xprt->reestablish_timeout)
1428                 xprt->reestablish_timeout = 0;
1429
1430         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1431         rd_desc.arg.data = xprt;
1432         do {
1433                 rd_desc.count = 65536;
1434                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1435         } while (read > 0);
1436 out:
1437         read_unlock_bh(&sk->sk_callback_lock);
1438 }
1439
1440 /*
1441  * Do the equivalent of linger/linger2 handling for dealing with
1442  * broken servers that don't close the socket in a timely
1443  * fashion
1444  */
1445 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1446                 unsigned long timeout)
1447 {
1448         struct sock_xprt *transport;
1449
1450         if (xprt_test_and_set_connecting(xprt))
1451                 return;
1452         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1453         transport = container_of(xprt, struct sock_xprt, xprt);
1454         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1455                            timeout);
1456 }
1457
1458 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1459 {
1460         struct sock_xprt *transport;
1461
1462         transport = container_of(xprt, struct sock_xprt, xprt);
1463
1464         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1465             !cancel_delayed_work(&transport->connect_worker))
1466                 return;
1467         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1468         xprt_clear_connecting(xprt);
1469 }
1470
1471 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1472 {
1473         smp_mb__before_clear_bit();
1474         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1475         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1476         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1477         clear_bit(XPRT_CLOSING, &xprt->state);
1478         smp_mb__after_clear_bit();
1479 }
1480
1481 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1482 {
1483         xs_sock_reset_connection_flags(xprt);
1484         /* Mark transport as closed and wake up all pending tasks */
1485         xprt_disconnect_done(xprt);
1486 }
1487
1488 /**
1489  * xs_tcp_state_change - callback to handle TCP socket state changes
1490  * @sk: socket whose state has changed
1491  *
1492  */
1493 static void xs_tcp_state_change(struct sock *sk)
1494 {
1495         struct rpc_xprt *xprt;
1496
1497         read_lock_bh(&sk->sk_callback_lock);
1498         if (!(xprt = xprt_from_sock(sk)))
1499                 goto out;
1500         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1501         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1502                         sk->sk_state, xprt_connected(xprt),
1503                         sock_flag(sk, SOCK_DEAD),
1504                         sock_flag(sk, SOCK_ZAPPED),
1505                         sk->sk_shutdown);
1506
1507         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1508         switch (sk->sk_state) {
1509         case TCP_ESTABLISHED:
1510                 spin_lock(&xprt->transport_lock);
1511                 if (!xprt_test_and_set_connected(xprt)) {
1512                         struct sock_xprt *transport = container_of(xprt,
1513                                         struct sock_xprt, xprt);
1514
1515                         /* Reset TCP record info */
1516                         transport->tcp_offset = 0;
1517                         transport->tcp_reclen = 0;
1518                         transport->tcp_copied = 0;
1519                         transport->tcp_flags =
1520                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1521                         xprt->connect_cookie++;
1522
1523                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1524                 }
1525                 spin_unlock(&xprt->transport_lock);
1526                 break;
1527         case TCP_FIN_WAIT1:
1528                 /* The client initiated a shutdown of the socket */
1529                 xprt->connect_cookie++;
1530                 xprt->reestablish_timeout = 0;
1531                 set_bit(XPRT_CLOSING, &xprt->state);
1532                 smp_mb__before_clear_bit();
1533                 clear_bit(XPRT_CONNECTED, &xprt->state);
1534                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1535                 smp_mb__after_clear_bit();
1536                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1537                 break;
1538         case TCP_CLOSE_WAIT:
1539                 /* The server initiated a shutdown of the socket */
1540                 xprt->connect_cookie++;
1541                 clear_bit(XPRT_CONNECTED, &xprt->state);
1542                 xs_tcp_force_close(xprt);
1543         case TCP_CLOSING:
1544                 /*
1545                  * If the server closed down the connection, make sure that
1546                  * we back off before reconnecting
1547                  */
1548                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1549                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1550                 break;
1551         case TCP_LAST_ACK:
1552                 set_bit(XPRT_CLOSING, &xprt->state);
1553                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1554                 smp_mb__before_clear_bit();
1555                 clear_bit(XPRT_CONNECTED, &xprt->state);
1556                 smp_mb__after_clear_bit();
1557                 break;
1558         case TCP_CLOSE:
1559                 xs_tcp_cancel_linger_timeout(xprt);
1560                 xs_sock_mark_closed(xprt);
1561         }
1562  out:
1563         read_unlock_bh(&sk->sk_callback_lock);
1564 }
1565
1566 static void xs_write_space(struct sock *sk)
1567 {
1568         struct socket *sock;
1569         struct rpc_xprt *xprt;
1570
1571         if (unlikely(!(sock = sk->sk_socket)))
1572                 return;
1573         clear_bit(SOCK_NOSPACE, &sock->flags);
1574
1575         if (unlikely(!(xprt = xprt_from_sock(sk))))
1576                 return;
1577         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1578                 return;
1579
1580         xprt_write_space(xprt);
1581 }
1582
1583 /**
1584  * xs_udp_write_space - callback invoked when socket buffer space
1585  *                             becomes available
1586  * @sk: socket whose state has changed
1587  *
1588  * Called when more output buffer space is available for this socket.
1589  * We try not to wake our writers until they can make "significant"
1590  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1591  * with a bunch of small requests.
1592  */
1593 static void xs_udp_write_space(struct sock *sk)
1594 {
1595         read_lock_bh(&sk->sk_callback_lock);
1596
1597         /* from net/core/sock.c:sock_def_write_space */
1598         if (sock_writeable(sk))
1599                 xs_write_space(sk);
1600
1601         read_unlock_bh(&sk->sk_callback_lock);
1602 }
1603
1604 /**
1605  * xs_tcp_write_space - callback invoked when socket buffer space
1606  *                             becomes available
1607  * @sk: socket whose state has changed
1608  *
1609  * Called when more output buffer space is available for this socket.
1610  * We try not to wake our writers until they can make "significant"
1611  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1612  * with a bunch of small requests.
1613  */
1614 static void xs_tcp_write_space(struct sock *sk)
1615 {
1616         read_lock_bh(&sk->sk_callback_lock);
1617
1618         /* from net/core/stream.c:sk_stream_write_space */
1619         if (sk_stream_is_writeable(sk))
1620                 xs_write_space(sk);
1621
1622         read_unlock_bh(&sk->sk_callback_lock);
1623 }
1624
1625 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1626 {
1627         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1628         struct sock *sk = transport->inet;
1629
1630         if (transport->rcvsize) {
1631                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1632                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1633         }
1634         if (transport->sndsize) {
1635                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1636                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1637                 sk->sk_write_space(sk);
1638         }
1639 }
1640
1641 /**
1642  * xs_udp_set_buffer_size - set send and receive limits
1643  * @xprt: generic transport
1644  * @sndsize: requested size of send buffer, in bytes
1645  * @rcvsize: requested size of receive buffer, in bytes
1646  *
1647  * Set socket send and receive buffer size limits.
1648  */
1649 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1650 {
1651         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1652
1653         transport->sndsize = 0;
1654         if (sndsize)
1655                 transport->sndsize = sndsize + 1024;
1656         transport->rcvsize = 0;
1657         if (rcvsize)
1658                 transport->rcvsize = rcvsize + 1024;
1659
1660         xs_udp_do_set_buffer_size(xprt);
1661 }
1662
1663 /**
1664  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1665  * @task: task that timed out
1666  *
1667  * Adjust the congestion window after a retransmit timeout has occurred.
1668  */
1669 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1670 {
1671         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1672 }
1673
1674 static unsigned short xs_get_random_port(void)
1675 {
1676         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1677         unsigned short rand = (unsigned short) net_random() % range;
1678         return rand + xprt_min_resvport;
1679 }
1680
1681 /**
1682  * xs_set_port - reset the port number in the remote endpoint address
1683  * @xprt: generic transport
1684  * @port: new port number
1685  *
1686  */
1687 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1688 {
1689         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1690
1691         rpc_set_port(xs_addr(xprt), port);
1692         xs_update_peer_port(xprt);
1693 }
1694
1695 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1696 {
1697         unsigned short port = transport->srcport;
1698
1699         if (port == 0 && transport->xprt.resvport)
1700                 port = xs_get_random_port();
1701         return port;
1702 }
1703
1704 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1705 {
1706         if (transport->srcport != 0)
1707                 transport->srcport = 0;
1708         if (!transport->xprt.resvport)
1709                 return 0;
1710         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1711                 return xprt_max_resvport;
1712         return --port;
1713 }
1714 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1715 {
1716         struct sockaddr_storage myaddr;
1717         int err, nloop = 0;
1718         unsigned short port = xs_get_srcport(transport);
1719         unsigned short last;
1720
1721         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1722         do {
1723                 rpc_set_port((struct sockaddr *)&myaddr, port);
1724                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1725                                 transport->xprt.addrlen);
1726                 if (port == 0)
1727                         break;
1728                 if (err == 0) {
1729                         transport->srcport = port;
1730                         break;
1731                 }
1732                 last = port;
1733                 port = xs_next_srcport(transport, port);
1734                 if (port > last)
1735                         nloop++;
1736         } while (err == -EADDRINUSE && nloop != 2);
1737
1738         if (myaddr.ss_family == AF_INET)
1739                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1740                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1741                                 port, err ? "failed" : "ok", err);
1742         else
1743                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1744                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1745                                 port, err ? "failed" : "ok", err);
1746         return err;
1747 }
1748
1749 /*
1750  * We don't support autobind on AF_LOCAL sockets
1751  */
1752 static void xs_local_rpcbind(struct rpc_task *task)
1753 {
1754         rcu_read_lock();
1755         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1756         rcu_read_unlock();
1757 }
1758
1759 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1760 {
1761 }
1762
1763 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1764 static struct lock_class_key xs_key[2];
1765 static struct lock_class_key xs_slock_key[2];
1766
1767 static inline void xs_reclassify_socketu(struct socket *sock)
1768 {
1769         struct sock *sk = sock->sk;
1770
1771         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1772                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1773 }
1774
1775 static inline void xs_reclassify_socket4(struct socket *sock)
1776 {
1777         struct sock *sk = sock->sk;
1778
1779         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1780                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1781 }
1782
1783 static inline void xs_reclassify_socket6(struct socket *sock)
1784 {
1785         struct sock *sk = sock->sk;
1786
1787         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1788                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1789 }
1790
1791 static inline void xs_reclassify_socket(int family, struct socket *sock)
1792 {
1793         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1794         if (sock_owned_by_user(sock->sk))
1795                 return;
1796
1797         switch (family) {
1798         case AF_LOCAL:
1799                 xs_reclassify_socketu(sock);
1800                 break;
1801         case AF_INET:
1802                 xs_reclassify_socket4(sock);
1803                 break;
1804         case AF_INET6:
1805                 xs_reclassify_socket6(sock);
1806                 break;
1807         }
1808 }
1809 #else
1810 static inline void xs_reclassify_socketu(struct socket *sock)
1811 {
1812 }
1813
1814 static inline void xs_reclassify_socket4(struct socket *sock)
1815 {
1816 }
1817
1818 static inline void xs_reclassify_socket6(struct socket *sock)
1819 {
1820 }
1821
1822 static inline void xs_reclassify_socket(int family, struct socket *sock)
1823 {
1824 }
1825 #endif
1826
1827 static void xs_dummy_setup_socket(struct work_struct *work)
1828 {
1829 }
1830
1831 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1832                 struct sock_xprt *transport, int family, int type, int protocol)
1833 {
1834         struct socket *sock;
1835         int err;
1836
1837         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1838         if (err < 0) {
1839                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1840                                 protocol, -err);
1841                 goto out;
1842         }
1843         xs_reclassify_socket(family, sock);
1844
1845         err = xs_bind(transport, sock);
1846         if (err) {
1847                 sock_release(sock);
1848                 goto out;
1849         }
1850
1851         return sock;
1852 out:
1853         return ERR_PTR(err);
1854 }
1855
1856 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1857                                       struct socket *sock)
1858 {
1859         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1860                                                                         xprt);
1861
1862         if (!transport->inet) {
1863                 struct sock *sk = sock->sk;
1864
1865                 write_lock_bh(&sk->sk_callback_lock);
1866
1867                 xs_save_old_callbacks(transport, sk);
1868
1869                 sk->sk_user_data = xprt;
1870                 sk->sk_data_ready = xs_local_data_ready;
1871                 sk->sk_write_space = xs_udp_write_space;
1872                 sk->sk_allocation = GFP_ATOMIC;
1873
1874                 xprt_clear_connected(xprt);
1875
1876                 /* Reset to new socket */
1877                 transport->sock = sock;
1878                 transport->inet = sk;
1879
1880                 write_unlock_bh(&sk->sk_callback_lock);
1881         }
1882
1883         /* Tell the socket layer to start connecting... */
1884         xprt->stat.connect_count++;
1885         xprt->stat.connect_start = jiffies;
1886         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1887 }
1888
1889 /**
1890  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1891  * @xprt: RPC transport to connect
1892  * @transport: socket transport to connect
1893  * @create_sock: function to create a socket of the correct type
1894  */
1895 static int xs_local_setup_socket(struct sock_xprt *transport)
1896 {
1897         struct rpc_xprt *xprt = &transport->xprt;
1898         struct socket *sock;
1899         int status = -EIO;
1900
1901         current->flags |= PF_FSTRANS;
1902
1903         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1904         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1905                                         SOCK_STREAM, 0, &sock, 1);
1906         if (status < 0) {
1907                 dprintk("RPC:       can't create AF_LOCAL "
1908                         "transport socket (%d).\n", -status);
1909                 goto out;
1910         }
1911         xs_reclassify_socketu(sock);
1912
1913         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1914                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915
1916         status = xs_local_finish_connecting(xprt, sock);
1917         trace_rpc_socket_connect(xprt, sock, status);
1918         switch (status) {
1919         case 0:
1920                 dprintk("RPC:       xprt %p connected to %s\n",
1921                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1922                 xprt_set_connected(xprt);
1923                 break;
1924         case -ENOENT:
1925                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1926                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1927                 break;
1928         case -ECONNREFUSED:
1929                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1930                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1931                 break;
1932         default:
1933                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1934                                 __func__, -status,
1935                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1936         }
1937
1938 out:
1939         xprt_clear_connecting(xprt);
1940         xprt_wake_pending_tasks(xprt, status);
1941         current->flags &= ~PF_FSTRANS;
1942         return status;
1943 }
1944
1945 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1946 {
1947         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1948         int ret;
1949
1950          if (RPC_IS_ASYNC(task)) {
1951                 /*
1952                  * We want the AF_LOCAL connect to be resolved in the
1953                  * filesystem namespace of the process making the rpc
1954                  * call.  Thus we connect synchronously.
1955                  *
1956                  * If we want to support asynchronous AF_LOCAL calls,
1957                  * we'll need to figure out how to pass a namespace to
1958                  * connect.
1959                  */
1960                 rpc_exit(task, -ENOTCONN);
1961                 return;
1962         }
1963         ret = xs_local_setup_socket(transport);
1964         if (ret && !RPC_IS_SOFTCONN(task))
1965                 msleep_interruptible(15000);
1966 }
1967
1968 #ifdef CONFIG_SUNRPC_SWAP
1969 static void xs_set_memalloc(struct rpc_xprt *xprt)
1970 {
1971         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1972                         xprt);
1973
1974         if (xprt->swapper)
1975                 sk_set_memalloc(transport->inet);
1976 }
1977
1978 /**
1979  * xs_swapper - Tag this transport as being used for swap.
1980  * @xprt: transport to tag
1981  * @enable: enable/disable
1982  *
1983  */
1984 int xs_swapper(struct rpc_xprt *xprt, int enable)
1985 {
1986         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1987                         xprt);
1988         int err = 0;
1989
1990         if (enable) {
1991                 xprt->swapper++;
1992                 xs_set_memalloc(xprt);
1993         } else if (xprt->swapper) {
1994                 xprt->swapper--;
1995                 sk_clear_memalloc(transport->inet);
1996         }
1997
1998         return err;
1999 }
2000 EXPORT_SYMBOL_GPL(xs_swapper);
2001 #else
2002 static void xs_set_memalloc(struct rpc_xprt *xprt)
2003 {
2004 }
2005 #endif
2006
2007 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2008 {
2009         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2010
2011         if (!transport->inet) {
2012                 struct sock *sk = sock->sk;
2013
2014                 write_lock_bh(&sk->sk_callback_lock);
2015
2016                 xs_save_old_callbacks(transport, sk);
2017
2018                 sk->sk_user_data = xprt;
2019                 sk->sk_data_ready = xs_udp_data_ready;
2020                 sk->sk_write_space = xs_udp_write_space;
2021                 sk->sk_no_check = UDP_CSUM_NORCV;
2022                 sk->sk_allocation = GFP_ATOMIC;
2023
2024                 xprt_set_connected(xprt);
2025
2026                 /* Reset to new socket */
2027                 transport->sock = sock;
2028                 transport->inet = sk;
2029
2030                 xs_set_memalloc(xprt);
2031
2032                 write_unlock_bh(&sk->sk_callback_lock);
2033         }
2034         xs_udp_do_set_buffer_size(xprt);
2035 }
2036
2037 static void xs_udp_setup_socket(struct work_struct *work)
2038 {
2039         struct sock_xprt *transport =
2040                 container_of(work, struct sock_xprt, connect_worker.work);
2041         struct rpc_xprt *xprt = &transport->xprt;
2042         struct socket *sock = transport->sock;
2043         int status = -EIO;
2044
2045         current->flags |= PF_FSTRANS;
2046
2047         /* Start by resetting any existing state */
2048         xs_reset_transport(transport);
2049         sock = xs_create_sock(xprt, transport,
2050                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2051         if (IS_ERR(sock))
2052                 goto out;
2053
2054         dprintk("RPC:       worker connecting xprt %p via %s to "
2055                                 "%s (port %s)\n", xprt,
2056                         xprt->address_strings[RPC_DISPLAY_PROTO],
2057                         xprt->address_strings[RPC_DISPLAY_ADDR],
2058                         xprt->address_strings[RPC_DISPLAY_PORT]);
2059
2060         xs_udp_finish_connecting(xprt, sock);
2061         trace_rpc_socket_connect(xprt, sock, 0);
2062         status = 0;
2063 out:
2064         xprt_clear_connecting(xprt);
2065         xprt_wake_pending_tasks(xprt, status);
2066         current->flags &= ~PF_FSTRANS;
2067 }
2068
2069 /*
2070  * We need to preserve the port number so the reply cache on the server can
2071  * find our cached RPC replies when we get around to reconnecting.
2072  */
2073 static void xs_abort_connection(struct sock_xprt *transport)
2074 {
2075         int result;
2076         struct sockaddr any;
2077
2078         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2079
2080         /*
2081          * Disconnect the transport socket by doing a connect operation
2082          * with AF_UNSPEC.  This should return immediately...
2083          */
2084         memset(&any, 0, sizeof(any));
2085         any.sa_family = AF_UNSPEC;
2086         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2087         trace_rpc_socket_reset_connection(&transport->xprt,
2088                         transport->sock, result);
2089         if (!result)
2090                 xs_sock_reset_connection_flags(&transport->xprt);
2091         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2092 }
2093
2094 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2095 {
2096         unsigned int state = transport->inet->sk_state;
2097
2098         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2099                 /* we don't need to abort the connection if the socket
2100                  * hasn't undergone a shutdown
2101                  */
2102                 if (transport->inet->sk_shutdown == 0)
2103                         return;
2104                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2105                                 __func__, transport->inet->sk_shutdown);
2106         }
2107         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2108                 /* we don't need to abort the connection if the socket
2109                  * hasn't undergone a shutdown
2110                  */
2111                 if (transport->inet->sk_shutdown == 0)
2112                         return;
2113                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2114                                 "sk_shutdown set to %d\n",
2115                                 __func__, transport->inet->sk_shutdown);
2116         }
2117         xs_abort_connection(transport);
2118 }
2119
2120 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2121 {
2122         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2123         int ret = -ENOTCONN;
2124
2125         if (!transport->inet) {
2126                 struct sock *sk = sock->sk;
2127                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2128                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2129                 unsigned int opt_on = 1;
2130
2131                 /* TCP Keepalive options */
2132                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2133                                 (char *)&opt_on, sizeof(opt_on));
2134                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2135                                 (char *)&keepidle, sizeof(keepidle));
2136                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2137                                 (char *)&keepidle, sizeof(keepidle));
2138                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2139                                 (char *)&keepcnt, sizeof(keepcnt));
2140
2141                 write_lock_bh(&sk->sk_callback_lock);
2142
2143                 xs_save_old_callbacks(transport, sk);
2144
2145                 sk->sk_user_data = xprt;
2146                 sk->sk_data_ready = xs_tcp_data_ready;
2147                 sk->sk_state_change = xs_tcp_state_change;
2148                 sk->sk_write_space = xs_tcp_write_space;
2149                 sk->sk_allocation = GFP_ATOMIC;
2150
2151                 /* socket options */
2152                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2153                 sock_reset_flag(sk, SOCK_LINGER);
2154                 tcp_sk(sk)->linger2 = 0;
2155                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2156
2157                 xprt_clear_connected(xprt);
2158
2159                 /* Reset to new socket */
2160                 transport->sock = sock;
2161                 transport->inet = sk;
2162
2163                 write_unlock_bh(&sk->sk_callback_lock);
2164         }
2165
2166         if (!xprt_bound(xprt))
2167                 goto out;
2168
2169         xs_set_memalloc(xprt);
2170
2171         /* Tell the socket layer to start connecting... */
2172         xprt->stat.connect_count++;
2173         xprt->stat.connect_start = jiffies;
2174         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2175         switch (ret) {
2176         case 0:
2177         case -EINPROGRESS:
2178                 /* SYN_SENT! */
2179                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2180                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2181         }
2182 out:
2183         return ret;
2184 }
2185
2186 /**
2187  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2188  * @xprt: RPC transport to connect
2189  * @transport: socket transport to connect
2190  * @create_sock: function to create a socket of the correct type
2191  *
2192  * Invoked by a work queue tasklet.
2193  */
2194 static void xs_tcp_setup_socket(struct work_struct *work)
2195 {
2196         struct sock_xprt *transport =
2197                 container_of(work, struct sock_xprt, connect_worker.work);
2198         struct socket *sock = transport->sock;
2199         struct rpc_xprt *xprt = &transport->xprt;
2200         int status = -EIO;
2201
2202         current->flags |= PF_FSTRANS;
2203
2204         if (!sock) {
2205                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2206                 sock = xs_create_sock(xprt, transport,
2207                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2208                 if (IS_ERR(sock)) {
2209                         status = PTR_ERR(sock);
2210                         goto out;
2211                 }
2212         } else {
2213                 int abort_and_exit;
2214
2215                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2216                                 &xprt->state);
2217                 /* "close" the socket, preserving the local port */
2218                 xs_tcp_reuse_connection(transport);
2219
2220                 if (abort_and_exit)
2221                         goto out_eagain;
2222         }
2223
2224         dprintk("RPC:       worker connecting xprt %p via %s to "
2225                                 "%s (port %s)\n", xprt,
2226                         xprt->address_strings[RPC_DISPLAY_PROTO],
2227                         xprt->address_strings[RPC_DISPLAY_ADDR],
2228                         xprt->address_strings[RPC_DISPLAY_PORT]);
2229
2230         status = xs_tcp_finish_connecting(xprt, sock);
2231         trace_rpc_socket_connect(xprt, sock, status);
2232         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2233                         xprt, -status, xprt_connected(xprt),
2234                         sock->sk->sk_state);
2235         switch (status) {
2236         default:
2237                 printk("%s: connect returned unhandled error %d\n",
2238                         __func__, status);
2239         case -EADDRNOTAVAIL:
2240                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2241                  * and retry
2242                  */
2243                 xs_tcp_force_close(xprt);
2244                 break;
2245         case 0:
2246         case -EINPROGRESS:
2247         case -EALREADY:
2248                 xprt_clear_connecting(xprt);
2249                 current->flags &= ~PF_FSTRANS;
2250                 return;
2251         case -EINVAL:
2252                 /* Happens, for instance, if the user specified a link
2253                  * local IPv6 address without a scope-id.
2254                  */
2255         case -ECONNREFUSED:
2256         case -ECONNRESET:
2257         case -ENETUNREACH:
2258                 /* retry with existing socket, after a delay */
2259                 goto out;
2260         }
2261 out_eagain:
2262         status = -EAGAIN;
2263 out:
2264         xprt_clear_connecting(xprt);
2265         xprt_wake_pending_tasks(xprt, status);
2266         current->flags &= ~PF_FSTRANS;
2267 }
2268
2269 /**
2270  * xs_connect - connect a socket to a remote endpoint
2271  * @xprt: pointer to transport structure
2272  * @task: address of RPC task that manages state of connect request
2273  *
2274  * TCP: If the remote end dropped the connection, delay reconnecting.
2275  *
2276  * UDP socket connects are synchronous, but we use a work queue anyway
2277  * to guarantee that even unprivileged user processes can set up a
2278  * socket on a privileged port.
2279  *
2280  * If a UDP socket connect fails, the delay behavior here prevents
2281  * retry floods (hard mounts).
2282  */
2283 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2284 {
2285         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2286
2287         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2288                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2289                                 "seconds\n",
2290                                 xprt, xprt->reestablish_timeout / HZ);
2291                 queue_delayed_work(rpciod_workqueue,
2292                                    &transport->connect_worker,
2293                                    xprt->reestablish_timeout);
2294                 xprt->reestablish_timeout <<= 1;
2295                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2296                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2297                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2298                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2299         } else {
2300                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2301                 queue_delayed_work(rpciod_workqueue,
2302                                    &transport->connect_worker, 0);
2303         }
2304 }
2305
2306 /**
2307  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2308  * @xprt: rpc_xprt struct containing statistics
2309  * @seq: output file
2310  *
2311  */
2312 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2313 {
2314         long idle_time = 0;
2315
2316         if (xprt_connected(xprt))
2317                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2318
2319         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2320                         "%llu %llu %lu %llu %llu\n",
2321                         xprt->stat.bind_count,
2322                         xprt->stat.connect_count,
2323                         xprt->stat.connect_time,
2324                         idle_time,
2325                         xprt->stat.sends,
2326                         xprt->stat.recvs,
2327                         xprt->stat.bad_xids,
2328                         xprt->stat.req_u,
2329                         xprt->stat.bklog_u,
2330                         xprt->stat.max_slots,
2331                         xprt->stat.sending_u,
2332                         xprt->stat.pending_u);
2333 }
2334
2335 /**
2336  * xs_udp_print_stats - display UDP socket-specifc stats
2337  * @xprt: rpc_xprt struct containing statistics
2338  * @seq: output file
2339  *
2340  */
2341 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2342 {
2343         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2344
2345         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2346                         "%lu %llu %llu\n",
2347                         transport->srcport,
2348                         xprt->stat.bind_count,
2349                         xprt->stat.sends,
2350                         xprt->stat.recvs,
2351                         xprt->stat.bad_xids,
2352                         xprt->stat.req_u,
2353                         xprt->stat.bklog_u,
2354                         xprt->stat.max_slots,
2355                         xprt->stat.sending_u,
2356                         xprt->stat.pending_u);
2357 }
2358
2359 /**
2360  * xs_tcp_print_stats - display TCP socket-specifc stats
2361  * @xprt: rpc_xprt struct containing statistics
2362  * @seq: output file
2363  *
2364  */
2365 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2366 {
2367         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2368         long idle_time = 0;
2369
2370         if (xprt_connected(xprt))
2371                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2372
2373         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2374                         "%llu %llu %lu %llu %llu\n",
2375                         transport->srcport,
2376                         xprt->stat.bind_count,
2377                         xprt->stat.connect_count,
2378                         xprt->stat.connect_time,
2379                         idle_time,
2380                         xprt->stat.sends,
2381                         xprt->stat.recvs,
2382                         xprt->stat.bad_xids,
2383                         xprt->stat.req_u,
2384                         xprt->stat.bklog_u,
2385                         xprt->stat.max_slots,
2386                         xprt->stat.sending_u,
2387                         xprt->stat.pending_u);
2388 }
2389
2390 /*
2391  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2392  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2393  * to use the server side send routines.
2394  */
2395 static void *bc_malloc(struct rpc_task *task, size_t size)
2396 {
2397         struct page *page;
2398         struct rpc_buffer *buf;
2399
2400         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2401         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2402                 return NULL;
2403
2404         page = alloc_page(GFP_KERNEL);
2405         if (!page)
2406                 return NULL;
2407
2408         buf = page_address(page);
2409         buf->len = PAGE_SIZE;
2410
2411         return buf->data;
2412 }
2413
2414 /*
2415  * Free the space allocated in the bc_alloc routine
2416  */
2417 static void bc_free(void *buffer)
2418 {
2419         struct rpc_buffer *buf;
2420
2421         if (!buffer)
2422                 return;
2423
2424         buf = container_of(buffer, struct rpc_buffer, data);
2425         free_page((unsigned long)buf);
2426 }
2427
2428 /*
2429  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2430  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2431  */
2432 static int bc_sendto(struct rpc_rqst *req)
2433 {
2434         int len;
2435         struct xdr_buf *xbufp = &req->rq_snd_buf;
2436         struct rpc_xprt *xprt = req->rq_xprt;
2437         struct sock_xprt *transport =
2438                                 container_of(xprt, struct sock_xprt, xprt);
2439         struct socket *sock = transport->sock;
2440         unsigned long headoff;
2441         unsigned long tailoff;
2442
2443         xs_encode_stream_record_marker(xbufp);
2444
2445         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2446         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2447         len = svc_send_common(sock, xbufp,
2448                               virt_to_page(xbufp->head[0].iov_base), headoff,
2449                               xbufp->tail[0].iov_base, tailoff);
2450
2451         if (len != xbufp->len) {
2452                 printk(KERN_NOTICE "Error sending entire callback!\n");
2453                 len = -EAGAIN;
2454         }
2455
2456         return len;
2457 }
2458
2459 /*
2460  * The send routine. Borrows from svc_send
2461  */
2462 static int bc_send_request(struct rpc_task *task)
2463 {
2464         struct rpc_rqst *req = task->tk_rqstp;
2465         struct svc_xprt *xprt;
2466         u32                     len;
2467
2468         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2469         /*
2470          * Get the server socket associated with this callback xprt
2471          */
2472         xprt = req->rq_xprt->bc_xprt;
2473
2474         /*
2475          * Grab the mutex to serialize data as the connection is shared
2476          * with the fore channel
2477          */
2478         if (!mutex_trylock(&xprt->xpt_mutex)) {
2479                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2480                 if (!mutex_trylock(&xprt->xpt_mutex))
2481                         return -EAGAIN;
2482                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2483         }
2484         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2485                 len = -ENOTCONN;
2486         else
2487                 len = bc_sendto(req);
2488         mutex_unlock(&xprt->xpt_mutex);
2489
2490         if (len > 0)
2491                 len = 0;
2492
2493         return len;
2494 }
2495
2496 /*
2497  * The close routine. Since this is client initiated, we do nothing
2498  */
2499
2500 static void bc_close(struct rpc_xprt *xprt)
2501 {
2502 }
2503
2504 /*
2505  * The xprt destroy routine. Again, because this connection is client
2506  * initiated, we do nothing
2507  */
2508
2509 static void bc_destroy(struct rpc_xprt *xprt)
2510 {
2511 }
2512
2513 static struct rpc_xprt_ops xs_local_ops = {
2514         .reserve_xprt           = xprt_reserve_xprt,
2515         .release_xprt           = xs_tcp_release_xprt,
2516         .alloc_slot             = xprt_alloc_slot,
2517         .rpcbind                = xs_local_rpcbind,
2518         .set_port               = xs_local_set_port,
2519         .connect                = xs_local_connect,
2520         .buf_alloc              = rpc_malloc,
2521         .buf_free               = rpc_free,
2522         .send_request           = xs_local_send_request,
2523         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2524         .close                  = xs_close,
2525         .destroy                = xs_destroy,
2526         .print_stats            = xs_local_print_stats,
2527 };
2528
2529 static struct rpc_xprt_ops xs_udp_ops = {
2530         .set_buffer_size        = xs_udp_set_buffer_size,
2531         .reserve_xprt           = xprt_reserve_xprt_cong,
2532         .release_xprt           = xprt_release_xprt_cong,
2533         .alloc_slot             = xprt_alloc_slot,
2534         .rpcbind                = rpcb_getport_async,
2535         .set_port               = xs_set_port,
2536         .connect                = xs_connect,
2537         .buf_alloc              = rpc_malloc,
2538         .buf_free               = rpc_free,
2539         .send_request           = xs_udp_send_request,
2540         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2541         .timer                  = xs_udp_timer,
2542         .release_request        = xprt_release_rqst_cong,
2543         .close                  = xs_close,
2544         .destroy                = xs_destroy,
2545         .print_stats            = xs_udp_print_stats,
2546 };
2547
2548 static struct rpc_xprt_ops xs_tcp_ops = {
2549         .reserve_xprt           = xprt_reserve_xprt,
2550         .release_xprt           = xs_tcp_release_xprt,
2551         .alloc_slot             = xprt_lock_and_alloc_slot,
2552         .rpcbind                = rpcb_getport_async,
2553         .set_port               = xs_set_port,
2554         .connect                = xs_connect,
2555         .buf_alloc              = rpc_malloc,
2556         .buf_free               = rpc_free,
2557         .send_request           = xs_tcp_send_request,
2558         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2559         .close                  = xs_tcp_close,
2560         .destroy                = xs_destroy,
2561         .print_stats            = xs_tcp_print_stats,
2562 };
2563
2564 /*
2565  * The rpc_xprt_ops for the server backchannel
2566  */
2567
2568 static struct rpc_xprt_ops bc_tcp_ops = {
2569         .reserve_xprt           = xprt_reserve_xprt,
2570         .release_xprt           = xprt_release_xprt,
2571         .alloc_slot             = xprt_alloc_slot,
2572         .buf_alloc              = bc_malloc,
2573         .buf_free               = bc_free,
2574         .send_request           = bc_send_request,
2575         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2576         .close                  = bc_close,
2577         .destroy                = bc_destroy,
2578         .print_stats            = xs_tcp_print_stats,
2579 };
2580
2581 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2582 {
2583         static const struct sockaddr_in sin = {
2584                 .sin_family             = AF_INET,
2585                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2586         };
2587         static const struct sockaddr_in6 sin6 = {
2588                 .sin6_family            = AF_INET6,
2589                 .sin6_addr              = IN6ADDR_ANY_INIT,
2590         };
2591
2592         switch (family) {
2593         case AF_LOCAL:
2594                 break;
2595         case AF_INET:
2596                 memcpy(sap, &sin, sizeof(sin));
2597                 break;
2598         case AF_INET6:
2599                 memcpy(sap, &sin6, sizeof(sin6));
2600                 break;
2601         default:
2602                 dprintk("RPC:       %s: Bad address family\n", __func__);
2603                 return -EAFNOSUPPORT;
2604         }
2605         return 0;
2606 }
2607
2608 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2609                                       unsigned int slot_table_size,
2610                                       unsigned int max_slot_table_size)
2611 {
2612         struct rpc_xprt *xprt;
2613         struct sock_xprt *new;
2614
2615         if (args->addrlen > sizeof(xprt->addr)) {
2616                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2617                 return ERR_PTR(-EBADF);
2618         }
2619
2620         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2621                         max_slot_table_size);
2622         if (xprt == NULL) {
2623                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2624                                 "rpc_xprt\n");
2625                 return ERR_PTR(-ENOMEM);
2626         }
2627
2628         new = container_of(xprt, struct sock_xprt, xprt);
2629         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2630         xprt->addrlen = args->addrlen;
2631         if (args->srcaddr)
2632                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2633         else {
2634                 int err;
2635                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2636                                         (struct sockaddr *)&new->srcaddr);
2637                 if (err != 0) {
2638                         xprt_free(xprt);
2639                         return ERR_PTR(err);
2640                 }
2641         }
2642
2643         return xprt;
2644 }
2645
2646 static const struct rpc_timeout xs_local_default_timeout = {
2647         .to_initval = 10 * HZ,
2648         .to_maxval = 10 * HZ,
2649         .to_retries = 2,
2650 };
2651
2652 /**
2653  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2654  * @args: rpc transport creation arguments
2655  *
2656  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2657  */
2658 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2659 {
2660         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2661         struct sock_xprt *transport;
2662         struct rpc_xprt *xprt;
2663         struct rpc_xprt *ret;
2664
2665         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2666                         xprt_max_tcp_slot_table_entries);
2667         if (IS_ERR(xprt))
2668                 return xprt;
2669         transport = container_of(xprt, struct sock_xprt, xprt);
2670
2671         xprt->prot = 0;
2672         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2673         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2674
2675         xprt->bind_timeout = XS_BIND_TO;
2676         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2677         xprt->idle_timeout = XS_IDLE_DISC_TO;
2678
2679         xprt->ops = &xs_local_ops;
2680         xprt->timeout = &xs_local_default_timeout;
2681
2682         INIT_DELAYED_WORK(&transport->connect_worker,
2683                         xs_dummy_setup_socket);
2684
2685         switch (sun->sun_family) {
2686         case AF_LOCAL:
2687                 if (sun->sun_path[0] != '/') {
2688                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2689                                         sun->sun_path);
2690                         ret = ERR_PTR(-EINVAL);
2691                         goto out_err;
2692                 }
2693                 xprt_set_bound(xprt);
2694                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2695                 ret = ERR_PTR(xs_local_setup_socket(transport));
2696                 if (ret)
2697                         goto out_err;
2698                 break;
2699         default:
2700                 ret = ERR_PTR(-EAFNOSUPPORT);
2701                 goto out_err;
2702         }
2703
2704         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2705                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2706
2707         if (try_module_get(THIS_MODULE))
2708                 return xprt;
2709         ret = ERR_PTR(-EINVAL);
2710 out_err:
2711         xprt_free(xprt);
2712         return ret;
2713 }
2714
2715 static const struct rpc_timeout xs_udp_default_timeout = {
2716         .to_initval = 5 * HZ,
2717         .to_maxval = 30 * HZ,
2718         .to_increment = 5 * HZ,
2719         .to_retries = 5,
2720 };
2721
2722 /**
2723  * xs_setup_udp - Set up transport to use a UDP socket
2724  * @args: rpc transport creation arguments
2725  *
2726  */
2727 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2728 {
2729         struct sockaddr *addr = args->dstaddr;
2730         struct rpc_xprt *xprt;
2731         struct sock_xprt *transport;
2732         struct rpc_xprt *ret;
2733
2734         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2735                         xprt_udp_slot_table_entries);
2736         if (IS_ERR(xprt))
2737                 return xprt;
2738         transport = container_of(xprt, struct sock_xprt, xprt);
2739
2740         xprt->prot = IPPROTO_UDP;
2741         xprt->tsh_size = 0;
2742         /* XXX: header size can vary due to auth type, IPv6, etc. */
2743         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2744
2745         xprt->bind_timeout = XS_BIND_TO;
2746         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2747         xprt->idle_timeout = XS_IDLE_DISC_TO;
2748
2749         xprt->ops = &xs_udp_ops;
2750
2751         xprt->timeout = &xs_udp_default_timeout;
2752
2753         switch (addr->sa_family) {
2754         case AF_INET:
2755                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2756                         xprt_set_bound(xprt);
2757
2758                 INIT_DELAYED_WORK(&transport->connect_worker,
2759                                         xs_udp_setup_socket);
2760                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2761                 break;
2762         case AF_INET6:
2763                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2764                         xprt_set_bound(xprt);
2765
2766                 INIT_DELAYED_WORK(&transport->connect_worker,
2767                                         xs_udp_setup_socket);
2768                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2769                 break;
2770         default:
2771                 ret = ERR_PTR(-EAFNOSUPPORT);
2772                 goto out_err;
2773         }
2774
2775         if (xprt_bound(xprt))
2776                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2777                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2778                                 xprt->address_strings[RPC_DISPLAY_PORT],
2779                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2780         else
2781                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2782                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2783                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2784
2785         if (try_module_get(THIS_MODULE))
2786                 return xprt;
2787         ret = ERR_PTR(-EINVAL);
2788 out_err:
2789         xprt_free(xprt);
2790         return ret;
2791 }
2792
2793 static const struct rpc_timeout xs_tcp_default_timeout = {
2794         .to_initval = 60 * HZ,
2795         .to_maxval = 60 * HZ,
2796         .to_retries = 2,
2797 };
2798
2799 /**
2800  * xs_setup_tcp - Set up transport to use a TCP socket
2801  * @args: rpc transport creation arguments
2802  *
2803  */
2804 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2805 {
2806         struct sockaddr *addr = args->dstaddr;
2807         struct rpc_xprt *xprt;
2808         struct sock_xprt *transport;
2809         struct rpc_xprt *ret;
2810         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2811
2812         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2813                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2814
2815         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2816                         max_slot_table_size);
2817         if (IS_ERR(xprt))
2818                 return xprt;
2819         transport = container_of(xprt, struct sock_xprt, xprt);
2820
2821         xprt->prot = IPPROTO_TCP;
2822         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2823         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2824
2825         xprt->bind_timeout = XS_BIND_TO;
2826         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2827         xprt->idle_timeout = XS_IDLE_DISC_TO;
2828
2829         xprt->ops = &xs_tcp_ops;
2830         xprt->timeout = &xs_tcp_default_timeout;
2831
2832         switch (addr->sa_family) {
2833         case AF_INET:
2834                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2835                         xprt_set_bound(xprt);
2836
2837                 INIT_DELAYED_WORK(&transport->connect_worker,
2838                                         xs_tcp_setup_socket);
2839                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2840                 break;
2841         case AF_INET6:
2842                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2843                         xprt_set_bound(xprt);
2844
2845                 INIT_DELAYED_WORK(&transport->connect_worker,
2846                                         xs_tcp_setup_socket);
2847                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2848                 break;
2849         default:
2850                 ret = ERR_PTR(-EAFNOSUPPORT);
2851                 goto out_err;
2852         }
2853
2854         if (xprt_bound(xprt))
2855                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2856                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2857                                 xprt->address_strings[RPC_DISPLAY_PORT],
2858                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2859         else
2860                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2861                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2862                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2863
2864
2865         if (try_module_get(THIS_MODULE))
2866                 return xprt;
2867         ret = ERR_PTR(-EINVAL);
2868 out_err:
2869         xprt_free(xprt);
2870         return ret;
2871 }
2872
2873 /**
2874  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2875  * @args: rpc transport creation arguments
2876  *
2877  */
2878 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2879 {
2880         struct sockaddr *addr = args->dstaddr;
2881         struct rpc_xprt *xprt;
2882         struct sock_xprt *transport;
2883         struct svc_sock *bc_sock;
2884         struct rpc_xprt *ret;
2885
2886         if (args->bc_xprt->xpt_bc_xprt) {
2887                 /*
2888                  * This server connection already has a backchannel
2889                  * transport; we can't create a new one, as we wouldn't
2890                  * be able to match replies based on xid any more.  So,
2891                  * reuse the already-existing one:
2892                  */
2893                  return args->bc_xprt->xpt_bc_xprt;
2894         }
2895         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2896                         xprt_tcp_slot_table_entries);
2897         if (IS_ERR(xprt))
2898                 return xprt;
2899         transport = container_of(xprt, struct sock_xprt, xprt);
2900
2901         xprt->prot = IPPROTO_TCP;
2902         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2903         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2904         xprt->timeout = &xs_tcp_default_timeout;
2905
2906         /* backchannel */
2907         xprt_set_bound(xprt);
2908         xprt->bind_timeout = 0;
2909         xprt->reestablish_timeout = 0;
2910         xprt->idle_timeout = 0;
2911
2912         xprt->ops = &bc_tcp_ops;
2913
2914         switch (addr->sa_family) {
2915         case AF_INET:
2916                 xs_format_peer_addresses(xprt, "tcp",
2917                                          RPCBIND_NETID_TCP);
2918                 break;
2919         case AF_INET6:
2920                 xs_format_peer_addresses(xprt, "tcp",
2921                                    RPCBIND_NETID_TCP6);
2922                 break;
2923         default:
2924                 ret = ERR_PTR(-EAFNOSUPPORT);
2925                 goto out_err;
2926         }
2927
2928         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2929                         xprt->address_strings[RPC_DISPLAY_ADDR],
2930                         xprt->address_strings[RPC_DISPLAY_PORT],
2931                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2932
2933         /*
2934          * Once we've associated a backchannel xprt with a connection,
2935          * we want to keep it around as long as long as the connection
2936          * lasts, in case we need to start using it for a backchannel
2937          * again; this reference won't be dropped until bc_xprt is
2938          * destroyed.
2939          */
2940         xprt_get(xprt);
2941         args->bc_xprt->xpt_bc_xprt = xprt;
2942         xprt->bc_xprt = args->bc_xprt;
2943         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2944         transport->sock = bc_sock->sk_sock;
2945         transport->inet = bc_sock->sk_sk;
2946
2947         /*
2948          * Since we don't want connections for the backchannel, we set
2949          * the xprt status to connected
2950          */
2951         xprt_set_connected(xprt);
2952
2953
2954         if (try_module_get(THIS_MODULE))
2955                 return xprt;
2956         xprt_put(xprt);
2957         ret = ERR_PTR(-EINVAL);
2958 out_err:
2959         xprt_free(xprt);
2960         return ret;
2961 }
2962
2963 static struct xprt_class        xs_local_transport = {
2964         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2965         .name           = "named UNIX socket",
2966         .owner          = THIS_MODULE,
2967         .ident          = XPRT_TRANSPORT_LOCAL,
2968         .setup          = xs_setup_local,
2969 };
2970
2971 static struct xprt_class        xs_udp_transport = {
2972         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2973         .name           = "udp",
2974         .owner          = THIS_MODULE,
2975         .ident          = XPRT_TRANSPORT_UDP,
2976         .setup          = xs_setup_udp,
2977 };
2978
2979 static struct xprt_class        xs_tcp_transport = {
2980         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2981         .name           = "tcp",
2982         .owner          = THIS_MODULE,
2983         .ident          = XPRT_TRANSPORT_TCP,
2984         .setup          = xs_setup_tcp,
2985 };
2986
2987 static struct xprt_class        xs_bc_tcp_transport = {
2988         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2989         .name           = "tcp NFSv4.1 backchannel",
2990         .owner          = THIS_MODULE,
2991         .ident          = XPRT_TRANSPORT_BC_TCP,
2992         .setup          = xs_setup_bc_tcp,
2993 };
2994
2995 /**
2996  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2997  *
2998  */
2999 int init_socket_xprt(void)
3000 {
3001 #ifdef RPC_DEBUG
3002         if (!sunrpc_table_header)
3003                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3004 #endif
3005
3006         xprt_register_transport(&xs_local_transport);
3007         xprt_register_transport(&xs_udp_transport);
3008         xprt_register_transport(&xs_tcp_transport);
3009         xprt_register_transport(&xs_bc_tcp_transport);
3010
3011         return 0;
3012 }
3013
3014 /**
3015  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3016  *
3017  */
3018 void cleanup_socket_xprt(void)
3019 {
3020 #ifdef RPC_DEBUG
3021         if (sunrpc_table_header) {
3022                 unregister_sysctl_table(sunrpc_table_header);
3023                 sunrpc_table_header = NULL;
3024         }
3025 #endif
3026
3027         xprt_unregister_transport(&xs_local_transport);
3028         xprt_unregister_transport(&xs_udp_transport);
3029         xprt_unregister_transport(&xs_tcp_transport);
3030         xprt_unregister_transport(&xs_bc_tcp_transport);
3031 }
3032
3033 static int param_set_uint_minmax(const char *val,
3034                 const struct kernel_param *kp,
3035                 unsigned int min, unsigned int max)
3036 {
3037         unsigned long num;
3038         int ret;
3039
3040         if (!val)
3041                 return -EINVAL;
3042         ret = strict_strtoul(val, 0, &num);
3043         if (ret == -EINVAL || num < min || num > max)
3044                 return -EINVAL;
3045         *((unsigned int *)kp->arg) = num;
3046         return 0;
3047 }
3048
3049 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3050 {
3051         return param_set_uint_minmax(val, kp,
3052                         RPC_MIN_RESVPORT,
3053                         RPC_MAX_RESVPORT);
3054 }
3055
3056 static struct kernel_param_ops param_ops_portnr = {
3057         .set = param_set_portnr,
3058         .get = param_get_uint,
3059 };
3060
3061 #define param_check_portnr(name, p) \
3062         __param_check(name, p, unsigned int);
3063
3064 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3065 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3066
3067 static int param_set_slot_table_size(const char *val,
3068                                      const struct kernel_param *kp)
3069 {
3070         return param_set_uint_minmax(val, kp,
3071                         RPC_MIN_SLOT_TABLE,
3072                         RPC_MAX_SLOT_TABLE);
3073 }
3074
3075 static struct kernel_param_ops param_ops_slot_table_size = {
3076         .set = param_set_slot_table_size,
3077         .get = param_get_uint,
3078 };
3079
3080 #define param_check_slot_table_size(name, p) \
3081         __param_check(name, p, unsigned int);
3082
3083 static int param_set_max_slot_table_size(const char *val,
3084                                      const struct kernel_param *kp)
3085 {
3086         return param_set_uint_minmax(val, kp,
3087                         RPC_MIN_SLOT_TABLE,
3088                         RPC_MAX_SLOT_TABLE_LIMIT);
3089 }
3090
3091 static struct kernel_param_ops param_ops_max_slot_table_size = {
3092         .set = param_set_max_slot_table_size,
3093         .get = param_get_uint,
3094 };
3095
3096 #define param_check_max_slot_table_size(name, p) \
3097         __param_check(name, p, unsigned int);
3098
3099 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3100                    slot_table_size, 0644);
3101 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3102                    max_slot_table_size, 0644);
3103 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3104                    slot_table_size, 0644);
3105