net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68 #define XS_TCP_LINGER_TO        (15U * HZ)
  69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71 /*
  72  * We can register our own files under /proc/sys/sunrpc by
  73  * calling register_sysctl_table() again.  The files in that
  74  * directory become the union of all files registered there.
  75  *
  76  * We simply need to make sure that we don't collide with
  77  * someone else's file names!
  78  */
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 181
 182 /*
 183  * TCP idle timeout; client drops the transport socket if it is idle
 184  * for this long.  Note that we also timeout UDP sockets to prevent
 185  * holding port numbers when there is no RPC traffic.
 186  */
 187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 188
 189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 190 # undef  RPC_DEBUG_DATA
 191 # define RPCDBG_FACILITY        RPCDBG_TRANS
 192 #endif
 193
 194 #ifdef RPC_DEBUG_DATA
 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196 {
 197         u8 *buf = (u8 *) packet;
 198         int j;
 199
 200         dprintk("RPC:       %s\n", msg);
 201         for (j = 0; j < count && j < 128; j += 4) {
 202                 if (!(j & 31)) {
 203                         if (j)
 204                                 dprintk("\n");
 205                         dprintk("0x%04x ", j);
 206                 }
 207                 dprintk("%02x%02x%02x%02x ",
 208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209         }
 210         dprintk("\n");
 211 }
 212 #else
 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214 {
 215         /* NOP */
 216 }
 217 #endif
 218
 219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 220 {
 221         return (struct rpc_xprt *) sk->sk_user_data;
 222 }
 223
 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 225 {
 226         return (struct sockaddr *) &xprt->addr;
 227 }
 228
 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 230 {
 231         return (struct sockaddr_un *) &xprt->addr;
 232 }
 233
 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr_in *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_in6 *) &xprt->addr;
 242 }
 243
 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 245 {
 246         struct sockaddr *sap = xs_addr(xprt);
 247         struct sockaddr_in6 *sin6;
 248         struct sockaddr_in *sin;
 249         struct sockaddr_un *sun;
 250         char buf[128];
 251
 252         switch (sap->sa_family) {
 253         case AF_LOCAL:
 254                 sun = xs_addr_un(xprt);
 255                 strlcpy(buf, sun->sun_path, sizeof(buf));
 256                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 257                                                 kstrdup(buf, GFP_KERNEL);
 258                 break;
 259         case AF_INET:
 260                 (void)rpc_ntop(sap, buf, sizeof(buf));
 261                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 262                                                 kstrdup(buf, GFP_KERNEL);
 263                 sin = xs_addr_in(xprt);
 264                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 265                 break;
 266         case AF_INET6:
 267                 (void)rpc_ntop(sap, buf, sizeof(buf));
 268                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 269                                                 kstrdup(buf, GFP_KERNEL);
 270                 sin6 = xs_addr_in6(xprt);
 271                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 272                 break;
 273         default:
 274                 BUG();
 275         }
 276
 277         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 278 }
 279
 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 281 {
 282         struct sockaddr *sap = xs_addr(xprt);
 283         char buf[128];
 284
 285         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 286         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 287
 288         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 289         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 290 }
 291
 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 293                                      const char *protocol,
 294                                      const char *netid)
 295 {
 296         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 297         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 298         xs_format_common_peer_addresses(xprt);
 299         xs_format_common_peer_ports(xprt);
 300 }
 301
 302 static void xs_update_peer_port(struct rpc_xprt *xprt)
 303 {
 304         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 305         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 306
 307         xs_format_common_peer_ports(xprt);
 308 }
 309
 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 311 {
 312         unsigned int i;
 313
 314         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 315                 switch (i) {
 316                 case RPC_DISPLAY_PROTO:
 317                 case RPC_DISPLAY_NETID:
 318                         continue;
 319                 default:
 320                         kfree(xprt->address_strings[i]);
 321                 }
 322 }
 323
 324 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 325
 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 327 {
 328         struct msghdr msg = {
 329                 .msg_name       = addr,
 330                 .msg_namelen    = addrlen,
 331                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 332         };
 333         struct kvec iov = {
 334                 .iov_base       = vec->iov_base + base,
 335                 .iov_len        = vec->iov_len - base,
 336         };
 337
 338         if (iov.iov_len != 0)
 339                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 340         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 341 }
 342
 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 344 {
 345         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 346                         int offset, size_t size, int flags);
 347         struct page **ppage;
 348         unsigned int remainder;
 349         int err;
 350
 351         remainder = xdr->page_len - base;
 352         base += xdr->page_base;
 353         ppage = xdr->pages + (base >> PAGE_SHIFT);
 354         base &= ~PAGE_MASK;
 355         do_sendpage = sock->ops->sendpage;
 356         if (!zerocopy)
 357                 do_sendpage = sock_no_sendpage;
 358         for(;;) {
 359                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 360                 int flags = XS_SENDMSG_FLAGS;
 361
 362                 remainder -= len;
 363                 if (remainder != 0 || more)
 364                         flags |= MSG_MORE;
 365                 err = do_sendpage(sock, *ppage, base, len, flags);
 366                 if (remainder == 0 || err != len)
 367                         break;
 368                 *sent_p += err;
 369                 ppage++;
 370                 base = 0;
 371         }
 372         if (err > 0) {
 373                 *sent_p += err;
 374                 err = 0;
 375         }
 376         return err;
 377 }
 378
 379 /**
 380  * xs_sendpages - write pages directly to a socket
 381  * @sock: socket to send on
 382  * @addr: UDP only -- address of destination
 383  * @addrlen: UDP only -- length of destination address
 384  * @xdr: buffer containing this request
 385  * @base: starting position in the buffer
 386  * @zerocopy: true if it is safe to use sendpage()
 387  * @sent_p: return the total number of bytes successfully queued for sending
 388  *
 389  */
 390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 391 {
 392         unsigned int remainder = xdr->len - base;
 393         int err = 0;
 394         int sent = 0;
 395
 396         if (unlikely(!sock))
 397                 return -ENOTSOCK;
 398
 399         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 400         if (base != 0) {
 401                 addr = NULL;
 402                 addrlen = 0;
 403         }
 404
 405         if (base < xdr->head[0].iov_len || addr != NULL) {
 406                 unsigned int len = xdr->head[0].iov_len - base;
 407                 remainder -= len;
 408                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 409                 if (remainder == 0 || err != len)
 410                         goto out;
 411                 *sent_p += err;
 412                 base = 0;
 413         } else
 414                 base -= xdr->head[0].iov_len;
 415
 416         if (base < xdr->page_len) {
 417                 unsigned int len = xdr->page_len - base;
 418                 remainder -= len;
 419                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 420                 *sent_p += sent;
 421                 if (remainder == 0 || sent != len)
 422                         goto out;
 423                 base = 0;
 424         } else
 425                 base -= xdr->page_len;
 426
 427         if (base >= xdr->tail[0].iov_len)
 428                 return 0;
 429         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 430 out:
 431         if (err > 0) {
 432                 *sent_p += err;
 433                 err = 0;
 434         }
 435         return err;
 436 }
 437
 438 static void xs_nospace_callback(struct rpc_task *task)
 439 {
 440         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 441
 442         transport->inet->sk_write_pending--;
 443         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 444 }
 445
 446 /**
 447  * xs_nospace - place task on wait queue if transmit was incomplete
 448  * @task: task to put to sleep
 449  *
 450  */
 451 static int xs_nospace(struct rpc_task *task)
 452 {
 453         struct rpc_rqst *req = task->tk_rqstp;
 454         struct rpc_xprt *xprt = req->rq_xprt;
 455         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 456         struct sock *sk = transport->inet;
 457         int ret = -EAGAIN;
 458
 459         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 460                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 461                         req->rq_slen);
 462
 463         /* Protect against races with write_space */
 464         spin_lock_bh(&xprt->transport_lock);
 465
 466         /* Don't race with disconnect */
 467         if (xprt_connected(xprt)) {
 468                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 469                         /*
 470                          * Notify TCP that we're limited by the application
 471                          * window size
 472                          */
 473                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 474                         sk->sk_write_pending++;
 475                         /* ...and wait for more buffer space */
 476                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 477                 }
 478         } else {
 479                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 480                 ret = -ENOTCONN;
 481         }
 482
 483         spin_unlock_bh(&xprt->transport_lock);
 484
 485         /* Race breaker in case memory is freed before above code is called */
 486         sk->sk_write_space(sk);
 487         return ret;
 488 }
 489
 490 /*
 491  * Construct a stream transport record marker in @buf.
 492  */
 493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 494 {
 495         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 496         rpc_fraghdr *base = buf->head[0].iov_base;
 497         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 498 }
 499
 500 /**
 501  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 502  * @task: RPC task that manages the state of an RPC request
 503  *
 504  * Return values:
 505  *        0:    The request has been sent
 506  *   EAGAIN:    The socket was blocked, please call again later to
 507  *              complete the request
 508  * ENOTCONN:    Caller needs to invoke connect logic then call again
 509  *    other:    Some other error occured, the request was not sent
 510  */
 511 static int xs_local_send_request(struct rpc_task *task)
 512 {
 513         struct rpc_rqst *req = task->tk_rqstp;
 514         struct rpc_xprt *xprt = req->rq_xprt;
 515         struct sock_xprt *transport =
 516                                 container_of(xprt, struct sock_xprt, xprt);
 517         struct xdr_buf *xdr = &req->rq_snd_buf;
 518         int status;
 519         int sent = 0;
 520
 521         xs_encode_stream_record_marker(&req->rq_snd_buf);
 522
 523         xs_pktdump("packet data:",
 524                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 525
 526         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 527                               true, &sent);
 528         dprintk("RPC:       %s(%u) = %d\n",
 529                         __func__, xdr->len - req->rq_bytes_sent, status);
 530         if (likely(sent > 0) || status == 0) {
 531                 req->rq_bytes_sent += sent;
 532                 req->rq_xmit_bytes_sent += sent;
 533                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 534                         req->rq_bytes_sent = 0;
 535                         return 0;
 536                 }
 537                 status = -EAGAIN;
 538         }
 539
 540         switch (status) {
 541         case -ENOBUFS:
 542         case -EAGAIN:
 543                 status = xs_nospace(task);
 544                 break;
 545         default:
 546                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 547                         -status);
 548         case -EPIPE:
 549                 xs_close(xprt);
 550                 status = -ENOTCONN;
 551         }
 552
 553         return status;
 554 }
 555
 556 /**
 557  * xs_udp_send_request - write an RPC request to a UDP socket
 558  * @task: address of RPC task that manages the state of an RPC request
 559  *
 560  * Return values:
 561  *        0:    The request has been sent
 562  *   EAGAIN:    The socket was blocked, please call again later to
 563  *              complete the request
 564  * ENOTCONN:    Caller needs to invoke connect logic then call again
 565  *    other:    Some other error occurred, the request was not sent
 566  */
 567 static int xs_udp_send_request(struct rpc_task *task)
 568 {
 569         struct rpc_rqst *req = task->tk_rqstp;
 570         struct rpc_xprt *xprt = req->rq_xprt;
 571         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 572         struct xdr_buf *xdr = &req->rq_snd_buf;
 573         int sent = 0;
 574         int status;
 575
 576         xs_pktdump("packet data:",
 577                                 req->rq_svec->iov_base,
 578                                 req->rq_svec->iov_len);
 579
 580         if (!xprt_bound(xprt))
 581                 return -ENOTCONN;
 582         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 583                               xdr, req->rq_bytes_sent, true, &sent);
 584
 585         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 586                         xdr->len - req->rq_bytes_sent, status);
 587
 588         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 589         if (status == -EPERM)
 590                 goto process_status;
 591
 592         if (sent > 0 || status == 0) {
 593                 req->rq_xmit_bytes_sent += sent;
 594                 if (sent >= req->rq_slen)
 595                         return 0;
 596                 /* Still some bytes left; set up for a retry later. */
 597                 status = -EAGAIN;
 598         }
 599
 600 process_status:
 601         switch (status) {
 602         case -ENOTSOCK:
 603                 status = -ENOTCONN;
 604                 /* Should we call xs_close() here? */
 605                 break;
 606         case -EAGAIN:
 607                 status = xs_nospace(task);
 608                 break;
 609         default:
 610                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 611                         -status);
 612         case -ENETUNREACH:
 613         case -ENOBUFS:
 614         case -EPIPE:
 615         case -ECONNREFUSED:
 616         case -EPERM:
 617                 /* When the server has died, an ICMP port unreachable message
 618                  * prompts ECONNREFUSED. */
 619                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 620         }
 621
 622         return status;
 623 }
 624
 625 /**
 626  * xs_tcp_send_request - write an RPC request to a TCP socket
 627  * @task: address of RPC task that manages the state of an RPC request
 628  *
 629  * Return values:
 630  *        0:    The request has been sent
 631  *   EAGAIN:    The socket was blocked, please call again later to
 632  *              complete the request
 633  * ENOTCONN:    Caller needs to invoke connect logic then call again
 634  *    other:    Some other error occurred, the request was not sent
 635  *
 636  * XXX: In the case of soft timeouts, should we eventually give up
 637  *      if sendmsg is not able to make progress?
 638  */
 639 static int xs_tcp_send_request(struct rpc_task *task)
 640 {
 641         struct rpc_rqst *req = task->tk_rqstp;
 642         struct rpc_xprt *xprt = req->rq_xprt;
 643         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 644         struct xdr_buf *xdr = &req->rq_snd_buf;
 645         bool zerocopy = true;
 646         int status;
 647         int sent;
 648
 649         xs_encode_stream_record_marker(&req->rq_snd_buf);
 650
 651         xs_pktdump("packet data:",
 652                                 req->rq_svec->iov_base,
 653                                 req->rq_svec->iov_len);
 654         /* Don't use zero copy if this is a resend. If the RPC call
 655          * completes while the socket holds a reference to the pages,
 656          * then we may end up resending corrupted data.
 657          */
 658         if (task->tk_flags & RPC_TASK_SENT)
 659                 zerocopy = false;
 660
 661         /* Continue transmitting the packet/record. We must be careful
 662          * to cope with writespace callbacks arriving _after_ we have
 663          * called sendmsg(). */
 664         while (1) {
 665                 sent = 0;
 666                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 667                                       req->rq_bytes_sent, zerocopy, &sent);
 668
 669                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 670                                 xdr->len - req->rq_bytes_sent, status);
 671
 672                 if (unlikely(sent == 0 && status < 0))
 673                         break;
 674
 675                 /* If we've sent the entire packet, immediately
 676                  * reset the count of bytes sent. */
 677                 req->rq_bytes_sent += sent;
 678                 req->rq_xmit_bytes_sent += sent;
 679                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 680                         req->rq_bytes_sent = 0;
 681                         return 0;
 682                 }
 683
 684                 if (sent != 0)
 685                         continue;
 686                 status = -EAGAIN;
 687                 break;
 688         }
 689
 690         switch (status) {
 691         case -ENOTSOCK:
 692                 status = -ENOTCONN;
 693                 /* Should we call xs_close() here? */
 694                 break;
 695         case -ENOBUFS:
 696         case -EAGAIN:
 697                 status = xs_nospace(task);
 698                 break;
 699         default:
 700                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 701                         -status);
 702         case -ECONNRESET:
 703         case -ECONNREFUSED:
 704         case -ENOTCONN:
 705         case -EADDRINUSE:
 706         case -EPIPE:
 707                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 708         }
 709
 710         return status;
 711 }
 712
 713 /**
 714  * xs_tcp_release_xprt - clean up after a tcp transmission
 715  * @xprt: transport
 716  * @task: rpc task
 717  *
 718  * This cleans up if an error causes us to abort the transmission of a request.
 719  * In this case, the socket may need to be reset in order to avoid confusing
 720  * the server.
 721  */
 722 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 723 {
 724         struct rpc_rqst *req;
 725
 726         if (task != xprt->snd_task)
 727                 return;
 728         if (task == NULL)
 729                 goto out_release;
 730         req = task->tk_rqstp;
 731         if (req == NULL)
 732                 goto out_release;
 733         if (req->rq_bytes_sent == 0)
 734                 goto out_release;
 735         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 736                 goto out_release;
 737         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 738 out_release:
 739         xprt_release_xprt(xprt, task);
 740 }
 741
 742 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 743 {
 744         transport->old_data_ready = sk->sk_data_ready;
 745         transport->old_state_change = sk->sk_state_change;
 746         transport->old_write_space = sk->sk_write_space;
 747         transport->old_error_report = sk->sk_error_report;
 748 }
 749
 750 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 751 {
 752         sk->sk_data_ready = transport->old_data_ready;
 753         sk->sk_state_change = transport->old_state_change;
 754         sk->sk_write_space = transport->old_write_space;
 755         sk->sk_error_report = transport->old_error_report;
 756 }
 757
 758 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 759 {
 760         smp_mb__before_atomic();
 761         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 762         clear_bit(XPRT_CLOSING, &xprt->state);
 763         smp_mb__after_atomic();
 764 }
 765
 766 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 767 {
 768         xs_sock_reset_connection_flags(xprt);
 769         /* Mark transport as closed and wake up all pending tasks */
 770         xprt_disconnect_done(xprt);
 771         xprt_force_disconnect(xprt);
 772 }
 773
 774 /**
 775  * xs_error_report - callback to handle TCP socket state errors
 776  * @sk: socket
 777  *
 778  * Note: we don't call sock_error() since there may be a rpc_task
 779  * using the socket, and so we don't want to clear sk->sk_err.
 780  */
 781 static void xs_error_report(struct sock *sk)
 782 {
 783         struct rpc_xprt *xprt;
 784         int err;
 785
 786         read_lock_bh(&sk->sk_callback_lock);
 787         if (!(xprt = xprt_from_sock(sk)))
 788                 goto out;
 789
 790         err = -sk->sk_err;
 791         if (err == 0)
 792                 goto out;
 793         /* Is this a reset event? */
 794         if (sk->sk_state == TCP_CLOSE)
 795                 xs_sock_mark_closed(xprt);
 796         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 797                         xprt, -err);
 798         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 799         xprt_wake_pending_tasks(xprt, err);
 800  out:
 801         read_unlock_bh(&sk->sk_callback_lock);
 802 }
 803
 804 static void xs_reset_transport(struct sock_xprt *transport)
 805 {
 806         struct socket *sock = transport->sock;
 807         struct sock *sk = transport->inet;
 808         struct rpc_xprt *xprt = &transport->xprt;
 809
 810         if (sk == NULL)
 811                 return;
 812
 813         if (atomic_read(&transport->xprt.swapper))
 814                 sk_clear_memalloc(sk);
 815
 816         write_lock_bh(&sk->sk_callback_lock);
 817         transport->inet = NULL;
 818         transport->sock = NULL;
 819
 820         sk->sk_user_data = NULL;
 821
 822         xs_restore_old_callbacks(transport, sk);
 823         write_unlock_bh(&sk->sk_callback_lock);
 824         xs_sock_reset_connection_flags(xprt);
 825
 826         trace_rpc_socket_close(xprt, sock);
 827         sock_release(sock);
 828 }
 829
 830 /**
 831  * xs_close - close a socket
 832  * @xprt: transport
 833  *
 834  * This is used when all requests are complete; ie, no DRC state remains
 835  * on the server we want to save.
 836  *
 837  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 838  * xs_reset_transport() zeroing the socket from underneath a writer.
 839  */
 840 static void xs_close(struct rpc_xprt *xprt)
 841 {
 842         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 843
 844         dprintk("RPC:       xs_close xprt %p\n", xprt);
 845
 846         xs_reset_transport(transport);
 847         xprt->reestablish_timeout = 0;
 848
 849         xprt_disconnect_done(xprt);
 850 }
 851
 852 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 853 {
 854         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 855                 xprt);
 856         xprt_disconnect_done(xprt);
 857 }
 858
 859 static void xs_xprt_free(struct rpc_xprt *xprt)
 860 {
 861         xs_free_peer_addresses(xprt);
 862         xprt_free(xprt);
 863 }
 864
 865 /**
 866  * xs_destroy - prepare to shutdown a transport
 867  * @xprt: doomed transport
 868  *
 869  */
 870 static void xs_destroy(struct rpc_xprt *xprt)
 871 {
 872         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 873
 874         xs_close(xprt);
 875         xs_xprt_free(xprt);
 876         module_put(THIS_MODULE);
 877 }
 878
 879 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 880 {
 881         struct xdr_skb_reader desc = {
 882                 .skb            = skb,
 883                 .offset         = sizeof(rpc_fraghdr),
 884                 .count          = skb->len - sizeof(rpc_fraghdr),
 885         };
 886
 887         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 888                 return -1;
 889         if (desc.count)
 890                 return -1;
 891         return 0;
 892 }
 893
 894 /**
 895  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 896  * @sk: socket with data to read
 897  *
 898  * Currently this assumes we can read the whole reply in a single gulp.
 899  */
 900 static void xs_local_data_ready(struct sock *sk)
 901 {
 902         struct rpc_task *task;
 903         struct rpc_xprt *xprt;
 904         struct rpc_rqst *rovr;
 905         struct sk_buff *skb;
 906         int err, repsize, copied;
 907         u32 _xid;
 908         __be32 *xp;
 909
 910         read_lock_bh(&sk->sk_callback_lock);
 911         dprintk("RPC:       %s...\n", __func__);
 912         xprt = xprt_from_sock(sk);
 913         if (xprt == NULL)
 914                 goto out;
 915
 916         skb = skb_recv_datagram(sk, 0, 1, &err);
 917         if (skb == NULL)
 918                 goto out;
 919
 920         repsize = skb->len - sizeof(rpc_fraghdr);
 921         if (repsize < 4) {
 922                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 923                 goto dropit;
 924         }
 925
 926         /* Copy the XID from the skb... */
 927         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 928         if (xp == NULL)
 929                 goto dropit;
 930
 931         /* Look up and lock the request corresponding to the given XID */
 932         spin_lock(&xprt->transport_lock);
 933         rovr = xprt_lookup_rqst(xprt, *xp);
 934         if (!rovr)
 935                 goto out_unlock;
 936         task = rovr->rq_task;
 937
 938         copied = rovr->rq_private_buf.buflen;
 939         if (copied > repsize)
 940                 copied = repsize;
 941
 942         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 943                 dprintk("RPC:       sk_buff copy failed\n");
 944                 goto out_unlock;
 945         }
 946
 947         xprt_complete_rqst(task, copied);
 948
 949  out_unlock:
 950         spin_unlock(&xprt->transport_lock);
 951  dropit:
 952         skb_free_datagram(sk, skb);
 953  out:
 954         read_unlock_bh(&sk->sk_callback_lock);
 955 }
 956
 957 /**
 958  * xs_udp_data_ready - "data ready" callback for UDP sockets
 959  * @sk: socket with data to read
 960  *
 961  */
 962 static void xs_udp_data_ready(struct sock *sk)
 963 {
 964         struct rpc_task *task;
 965         struct rpc_xprt *xprt;
 966         struct rpc_rqst *rovr;
 967         struct sk_buff *skb;
 968         int err, repsize, copied;
 969         u32 _xid;
 970         __be32 *xp;
 971
 972         read_lock_bh(&sk->sk_callback_lock);
 973         dprintk("RPC:       xs_udp_data_ready...\n");
 974         if (!(xprt = xprt_from_sock(sk)))
 975                 goto out;
 976
 977         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 978                 goto out;
 979
 980         repsize = skb->len - sizeof(struct udphdr);
 981         if (repsize < 4) {
 982                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 983                 goto dropit;
 984         }
 985
 986         /* Copy the XID from the skb... */
 987         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 988                                 sizeof(_xid), &_xid);
 989         if (xp == NULL)
 990                 goto dropit;
 991
 992         /* Look up and lock the request corresponding to the given XID */
 993         spin_lock(&xprt->transport_lock);
 994         rovr = xprt_lookup_rqst(xprt, *xp);
 995         if (!rovr)
 996                 goto out_unlock;
 997         task = rovr->rq_task;
 998
 999         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1000                 copied = repsize;
1001
1002         /* Suck it into the iovec, verify checksum if not done by hw. */
1003         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1004                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1005                 goto out_unlock;
1006         }
1007
1008         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1009
1010         xprt_adjust_cwnd(xprt, task, copied);
1011         xprt_complete_rqst(task, copied);
1012
1013  out_unlock:
1014         spin_unlock(&xprt->transport_lock);
1015  dropit:
1016         skb_free_datagram(sk, skb);
1017  out:
1018         read_unlock_bh(&sk->sk_callback_lock);
1019 }
1020
1021 /*
1022  * Helper function to force a TCP close if the server is sending
1023  * junk and/or it has put us in CLOSE_WAIT
1024  */
1025 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1026 {
1027         xprt_force_disconnect(xprt);
1028 }
1029
1030 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1031 {
1032         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1033         size_t len, used;
1034         char *p;
1035
1036         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1037         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1038         used = xdr_skb_read_bits(desc, p, len);
1039         transport->tcp_offset += used;
1040         if (used != len)
1041                 return;
1042
1043         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1044         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1045                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1046         else
1047                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1048         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1049
1050         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1051         transport->tcp_offset = 0;
1052
1053         /* Sanity check of the record length */
1054         if (unlikely(transport->tcp_reclen < 8)) {
1055                 dprintk("RPC:       invalid TCP record fragment length\n");
1056                 xs_tcp_force_close(xprt);
1057                 return;
1058         }
1059         dprintk("RPC:       reading TCP record fragment of length %d\n",
1060                         transport->tcp_reclen);
1061 }
1062
1063 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1064 {
1065         if (transport->tcp_offset == transport->tcp_reclen) {
1066                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1067                 transport->tcp_offset = 0;
1068                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1069                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1070                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1071                         transport->tcp_copied = 0;
1072                 }
1073         }
1074 }
1075
1076 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1077 {
1078         size_t len, used;
1079         char *p;
1080
1081         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1082         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1083         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1084         used = xdr_skb_read_bits(desc, p, len);
1085         transport->tcp_offset += used;
1086         if (used != len)
1087                 return;
1088         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1089         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1090         transport->tcp_copied = 4;
1091         dprintk("RPC:       reading %s XID %08x\n",
1092                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1093                                                               : "request with",
1094                         ntohl(transport->tcp_xid));
1095         xs_tcp_check_fraghdr(transport);
1096 }
1097
1098 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1099                                        struct xdr_skb_reader *desc)
1100 {
1101         size_t len, used;
1102         u32 offset;
1103         char *p;
1104
1105         /*
1106          * We want transport->tcp_offset to be 8 at the end of this routine
1107          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1108          * When this function is called for the first time,
1109          * transport->tcp_offset is 4 (after having already read the xid).
1110          */
1111         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1112         len = sizeof(transport->tcp_calldir) - offset;
1113         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1114         p = ((char *) &transport->tcp_calldir) + offset;
1115         used = xdr_skb_read_bits(desc, p, len);
1116         transport->tcp_offset += used;
1117         if (used != len)
1118                 return;
1119         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1120         /*
1121          * We don't yet have the XDR buffer, so we will write the calldir
1122          * out after we get the buffer from the 'struct rpc_rqst'
1123          */
1124         switch (ntohl(transport->tcp_calldir)) {
1125         case RPC_REPLY:
1126                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1127                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1128                 transport->tcp_flags |= TCP_RPC_REPLY;
1129                 break;
1130         case RPC_CALL:
1131                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1132                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1133                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1134                 break;
1135         default:
1136                 dprintk("RPC:       invalid request message type\n");
1137                 xs_tcp_force_close(&transport->xprt);
1138         }
1139         xs_tcp_check_fraghdr(transport);
1140 }
1141
1142 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1143                                      struct xdr_skb_reader *desc,
1144                                      struct rpc_rqst *req)
1145 {
1146         struct sock_xprt *transport =
1147                                 container_of(xprt, struct sock_xprt, xprt);
1148         struct xdr_buf *rcvbuf;
1149         size_t len;
1150         ssize_t r;
1151
1152         rcvbuf = &req->rq_private_buf;
1153
1154         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1155                 /*
1156                  * Save the RPC direction in the XDR buffer
1157                  */
1158                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1159                         &transport->tcp_calldir,
1160                         sizeof(transport->tcp_calldir));
1161                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1162                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1163         }
1164
1165         len = desc->count;
1166         if (len > transport->tcp_reclen - transport->tcp_offset) {
1167                 struct xdr_skb_reader my_desc;
1168
1169                 len = transport->tcp_reclen - transport->tcp_offset;
1170                 memcpy(&my_desc, desc, sizeof(my_desc));
1171                 my_desc.count = len;
1172                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1173                                           &my_desc, xdr_skb_read_bits);
1174                 desc->count -= r;
1175                 desc->offset += r;
1176         } else
1177                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1178                                           desc, xdr_skb_read_bits);
1179
1180         if (r > 0) {
1181                 transport->tcp_copied += r;
1182                 transport->tcp_offset += r;
1183         }
1184         if (r != len) {
1185                 /* Error when copying to the receive buffer,
1186                  * usually because we weren't able to allocate
1187                  * additional buffer pages. All we can do now
1188                  * is turn off TCP_RCV_COPY_DATA, so the request
1189                  * will not receive any additional updates,
1190                  * and time out.
1191                  * Any remaining data from this record will
1192                  * be discarded.
1193                  */
1194                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1195                 dprintk("RPC:       XID %08x truncated request\n",
1196                                 ntohl(transport->tcp_xid));
1197                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1198                                 "tcp_offset = %u, tcp_reclen = %u\n",
1199                                 xprt, transport->tcp_copied,
1200                                 transport->tcp_offset, transport->tcp_reclen);
1201                 return;
1202         }
1203
1204         dprintk("RPC:       XID %08x read %Zd bytes\n",
1205                         ntohl(transport->tcp_xid), r);
1206         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1207                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1208                         transport->tcp_offset, transport->tcp_reclen);
1209
1210         if (transport->tcp_copied == req->rq_private_buf.buflen)
1211                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1212         else if (transport->tcp_offset == transport->tcp_reclen) {
1213                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1214                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1215         }
1216 }
1217
1218 /*
1219  * Finds the request corresponding to the RPC xid and invokes the common
1220  * tcp read code to read the data.
1221  */
1222 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1223                                     struct xdr_skb_reader *desc)
1224 {
1225         struct sock_xprt *transport =
1226                                 container_of(xprt, struct sock_xprt, xprt);
1227         struct rpc_rqst *req;
1228
1229         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1230
1231         /* Find and lock the request corresponding to this xid */
1232         spin_lock(&xprt->transport_lock);
1233         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1234         if (!req) {
1235                 dprintk("RPC:       XID %08x request not found!\n",
1236                                 ntohl(transport->tcp_xid));
1237                 spin_unlock(&xprt->transport_lock);
1238                 return -1;
1239         }
1240
1241         xs_tcp_read_common(xprt, desc, req);
1242
1243         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1244                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1245
1246         spin_unlock(&xprt->transport_lock);
1247         return 0;
1248 }
1249
1250 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1251 /*
1252  * Obtains an rpc_rqst previously allocated and invokes the common
1253  * tcp read code to read the data.  The result is placed in the callback
1254  * queue.
1255  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1256  * connection and return -1.
1257  */
1258 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1259                                        struct xdr_skb_reader *desc)
1260 {
1261         struct sock_xprt *transport =
1262                                 container_of(xprt, struct sock_xprt, xprt);
1263         struct rpc_rqst *req;
1264
1265         /* Look up and lock the request corresponding to the given XID */
1266         spin_lock(&xprt->transport_lock);
1267         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1268         if (req == NULL) {
1269                 spin_unlock(&xprt->transport_lock);
1270                 printk(KERN_WARNING "Callback slot table overflowed\n");
1271                 xprt_force_disconnect(xprt);
1272                 return -1;
1273         }
1274
1275         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1276         xs_tcp_read_common(xprt, desc, req);
1277
1278         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1279                 xprt_complete_bc_request(req, transport->tcp_copied);
1280         spin_unlock(&xprt->transport_lock);
1281
1282         return 0;
1283 }
1284
1285 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1286                                         struct xdr_skb_reader *desc)
1287 {
1288         struct sock_xprt *transport =
1289                                 container_of(xprt, struct sock_xprt, xprt);
1290
1291         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1292                 xs_tcp_read_reply(xprt, desc) :
1293                 xs_tcp_read_callback(xprt, desc);
1294 }
1295 #else
1296 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1297                                         struct xdr_skb_reader *desc)
1298 {
1299         return xs_tcp_read_reply(xprt, desc);
1300 }
1301 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1302
1303 /*
1304  * Read data off the transport.  This can be either an RPC_CALL or an
1305  * RPC_REPLY.  Relay the processing to helper functions.
1306  */
1307 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1308                                     struct xdr_skb_reader *desc)
1309 {
1310         struct sock_xprt *transport =
1311                                 container_of(xprt, struct sock_xprt, xprt);
1312
1313         if (_xs_tcp_read_data(xprt, desc) == 0)
1314                 xs_tcp_check_fraghdr(transport);
1315         else {
1316                 /*
1317                  * The transport_lock protects the request handling.
1318                  * There's no need to hold it to update the tcp_flags.
1319                  */
1320                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1321         }
1322 }
1323
1324 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1325 {
1326         size_t len;
1327
1328         len = transport->tcp_reclen - transport->tcp_offset;
1329         if (len > desc->count)
1330                 len = desc->count;
1331         desc->count -= len;
1332         desc->offset += len;
1333         transport->tcp_offset += len;
1334         dprintk("RPC:       discarded %Zu bytes\n", len);
1335         xs_tcp_check_fraghdr(transport);
1336 }
1337
1338 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1339 {
1340         struct rpc_xprt *xprt = rd_desc->arg.data;
1341         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1342         struct xdr_skb_reader desc = {
1343                 .skb    = skb,
1344                 .offset = offset,
1345                 .count  = len,
1346         };
1347
1348         dprintk("RPC:       xs_tcp_data_recv started\n");
1349         do {
1350                 trace_xs_tcp_data_recv(transport);
1351                 /* Read in a new fragment marker if necessary */
1352                 /* Can we ever really expect to get completely empty fragments? */
1353                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1354                         xs_tcp_read_fraghdr(xprt, &desc);
1355                         continue;
1356                 }
1357                 /* Read in the xid if necessary */
1358                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1359                         xs_tcp_read_xid(transport, &desc);
1360                         continue;
1361                 }
1362                 /* Read in the call/reply flag */
1363                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1364                         xs_tcp_read_calldir(transport, &desc);
1365                         continue;
1366                 }
1367                 /* Read in the request data */
1368                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1369                         xs_tcp_read_data(xprt, &desc);
1370                         continue;
1371                 }
1372                 /* Skip over any trailing bytes on short reads */
1373                 xs_tcp_read_discard(transport, &desc);
1374         } while (desc.count);
1375         trace_xs_tcp_data_recv(transport);
1376         dprintk("RPC:       xs_tcp_data_recv done\n");
1377         return len - desc.count;
1378 }
1379
1380 /**
1381  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1382  * @sk: socket with data to read
1383  *
1384  */
1385 static void xs_tcp_data_ready(struct sock *sk)
1386 {
1387         struct rpc_xprt *xprt;
1388         read_descriptor_t rd_desc;
1389         int read;
1390         unsigned long total = 0;
1391
1392         dprintk("RPC:       xs_tcp_data_ready...\n");
1393
1394         read_lock_bh(&sk->sk_callback_lock);
1395         if (!(xprt = xprt_from_sock(sk))) {
1396                 read = 0;
1397                 goto out;
1398         }
1399         /* Any data means we had a useful conversation, so
1400          * the we don't need to delay the next reconnect
1401          */
1402         if (xprt->reestablish_timeout)
1403                 xprt->reestablish_timeout = 0;
1404
1405         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1406         rd_desc.arg.data = xprt;
1407         do {
1408                 rd_desc.count = 65536;
1409                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1410                 if (read > 0)
1411                         total += read;
1412         } while (read > 0);
1413 out:
1414         trace_xs_tcp_data_ready(xprt, read, total);
1415         read_unlock_bh(&sk->sk_callback_lock);
1416 }
1417
1418 /**
1419  * xs_tcp_state_change - callback to handle TCP socket state changes
1420  * @sk: socket whose state has changed
1421  *
1422  */
1423 static void xs_tcp_state_change(struct sock *sk)
1424 {
1425         struct rpc_xprt *xprt;
1426
1427         read_lock_bh(&sk->sk_callback_lock);
1428         if (!(xprt = xprt_from_sock(sk)))
1429                 goto out;
1430         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1431         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1432                         sk->sk_state, xprt_connected(xprt),
1433                         sock_flag(sk, SOCK_DEAD),
1434                         sock_flag(sk, SOCK_ZAPPED),
1435                         sk->sk_shutdown);
1436
1437         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1438         switch (sk->sk_state) {
1439         case TCP_ESTABLISHED:
1440                 spin_lock(&xprt->transport_lock);
1441                 if (!xprt_test_and_set_connected(xprt)) {
1442                         struct sock_xprt *transport = container_of(xprt,
1443                                         struct sock_xprt, xprt);
1444
1445                         /* Reset TCP record info */
1446                         transport->tcp_offset = 0;
1447                         transport->tcp_reclen = 0;
1448                         transport->tcp_copied = 0;
1449                         transport->tcp_flags =
1450                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1451                         xprt->connect_cookie++;
1452
1453                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1454                 }
1455                 spin_unlock(&xprt->transport_lock);
1456                 break;
1457         case TCP_FIN_WAIT1:
1458                 /* The client initiated a shutdown of the socket */
1459                 xprt->connect_cookie++;
1460                 xprt->reestablish_timeout = 0;
1461                 set_bit(XPRT_CLOSING, &xprt->state);
1462                 smp_mb__before_atomic();
1463                 clear_bit(XPRT_CONNECTED, &xprt->state);
1464                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1465                 smp_mb__after_atomic();
1466                 break;
1467         case TCP_CLOSE_WAIT:
1468                 /* The server initiated a shutdown of the socket */
1469                 xprt->connect_cookie++;
1470                 clear_bit(XPRT_CONNECTED, &xprt->state);
1471                 xs_tcp_force_close(xprt);
1472         case TCP_CLOSING:
1473                 /*
1474                  * If the server closed down the connection, make sure that
1475                  * we back off before reconnecting
1476                  */
1477                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1478                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1479                 break;
1480         case TCP_LAST_ACK:
1481                 set_bit(XPRT_CLOSING, &xprt->state);
1482                 smp_mb__before_atomic();
1483                 clear_bit(XPRT_CONNECTED, &xprt->state);
1484                 smp_mb__after_atomic();
1485                 break;
1486         case TCP_CLOSE:
1487                 xs_sock_mark_closed(xprt);
1488         }
1489  out:
1490         read_unlock_bh(&sk->sk_callback_lock);
1491 }
1492
1493 static void xs_write_space(struct sock *sk)
1494 {
1495         struct socket *sock;
1496         struct rpc_xprt *xprt;
1497
1498         if (unlikely(!(sock = sk->sk_socket)))
1499                 return;
1500         clear_bit(SOCK_NOSPACE, &sock->flags);
1501
1502         if (unlikely(!(xprt = xprt_from_sock(sk))))
1503                 return;
1504         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1505                 return;
1506
1507         xprt_write_space(xprt);
1508 }
1509
1510 /**
1511  * xs_udp_write_space - callback invoked when socket buffer space
1512  *                             becomes available
1513  * @sk: socket whose state has changed
1514  *
1515  * Called when more output buffer space is available for this socket.
1516  * We try not to wake our writers until they can make "significant"
1517  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1518  * with a bunch of small requests.
1519  */
1520 static void xs_udp_write_space(struct sock *sk)
1521 {
1522         read_lock_bh(&sk->sk_callback_lock);
1523
1524         /* from net/core/sock.c:sock_def_write_space */
1525         if (sock_writeable(sk))
1526                 xs_write_space(sk);
1527
1528         read_unlock_bh(&sk->sk_callback_lock);
1529 }
1530
1531 /**
1532  * xs_tcp_write_space - callback invoked when socket buffer space
1533  *                             becomes available
1534  * @sk: socket whose state has changed
1535  *
1536  * Called when more output buffer space is available for this socket.
1537  * We try not to wake our writers until they can make "significant"
1538  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1539  * with a bunch of small requests.
1540  */
1541 static void xs_tcp_write_space(struct sock *sk)
1542 {
1543         read_lock_bh(&sk->sk_callback_lock);
1544
1545         /* from net/core/stream.c:sk_stream_write_space */
1546         if (sk_stream_is_writeable(sk))
1547                 xs_write_space(sk);
1548
1549         read_unlock_bh(&sk->sk_callback_lock);
1550 }
1551
1552 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1553 {
1554         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1555         struct sock *sk = transport->inet;
1556
1557         if (transport->rcvsize) {
1558                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1559                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1560         }
1561         if (transport->sndsize) {
1562                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1563                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1564                 sk->sk_write_space(sk);
1565         }
1566 }
1567
1568 /**
1569  * xs_udp_set_buffer_size - set send and receive limits
1570  * @xprt: generic transport
1571  * @sndsize: requested size of send buffer, in bytes
1572  * @rcvsize: requested size of receive buffer, in bytes
1573  *
1574  * Set socket send and receive buffer size limits.
1575  */
1576 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1577 {
1578         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1579
1580         transport->sndsize = 0;
1581         if (sndsize)
1582                 transport->sndsize = sndsize + 1024;
1583         transport->rcvsize = 0;
1584         if (rcvsize)
1585                 transport->rcvsize = rcvsize + 1024;
1586
1587         xs_udp_do_set_buffer_size(xprt);
1588 }
1589
1590 /**
1591  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1592  * @task: task that timed out
1593  *
1594  * Adjust the congestion window after a retransmit timeout has occurred.
1595  */
1596 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1597 {
1598         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1599 }
1600
1601 static unsigned short xs_get_random_port(void)
1602 {
1603         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1604         unsigned short rand = (unsigned short) prandom_u32() % range;
1605         return rand + xprt_min_resvport;
1606 }
1607
1608 /**
1609  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1610  * @sock: socket
1611  *
1612  * Note that this function has to be called on all sockets that share the
1613  * same port, and it must be called before binding.
1614  */
1615 static void xs_sock_set_reuseport(struct socket *sock)
1616 {
1617         int opt = 1;
1618
1619         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1620                         (char *)&opt, sizeof(opt));
1621 }
1622
1623 static unsigned short xs_sock_getport(struct socket *sock)
1624 {
1625         struct sockaddr_storage buf;
1626         int buflen;
1627         unsigned short port = 0;
1628
1629         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1630                 goto out;
1631         switch (buf.ss_family) {
1632         case AF_INET6:
1633                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1634                 break;
1635         case AF_INET:
1636                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1637         }
1638 out:
1639         return port;
1640 }
1641
1642 /**
1643  * xs_set_port - reset the port number in the remote endpoint address
1644  * @xprt: generic transport
1645  * @port: new port number
1646  *
1647  */
1648 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1649 {
1650         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1651
1652         rpc_set_port(xs_addr(xprt), port);
1653         xs_update_peer_port(xprt);
1654 }
1655
1656 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1657 {
1658         if (transport->srcport == 0)
1659                 transport->srcport = xs_sock_getport(sock);
1660 }
1661
1662 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1663 {
1664         unsigned short port = transport->srcport;
1665
1666         if (port == 0 && transport->xprt.resvport)
1667                 port = xs_get_random_port();
1668         return port;
1669 }
1670
1671 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1672 {
1673         if (transport->srcport != 0)
1674                 transport->srcport = 0;
1675         if (!transport->xprt.resvport)
1676                 return 0;
1677         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1678                 return xprt_max_resvport;
1679         return --port;
1680 }
1681 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1682 {
1683         struct sockaddr_storage myaddr;
1684         int err, nloop = 0;
1685         unsigned short port = xs_get_srcport(transport);
1686         unsigned short last;
1687
1688         /*
1689          * If we are asking for any ephemeral port (i.e. port == 0 &&
1690          * transport->xprt.resvport == 0), don't bind.  Let the local
1691          * port selection happen implicitly when the socket is used
1692          * (for example at connect time).
1693          *
1694          * This ensures that we can continue to establish TCP
1695          * connections even when all local ephemeral ports are already
1696          * a part of some TCP connection.  This makes no difference
1697          * for UDP sockets, but also doens't harm them.
1698          *
1699          * If we're asking for any reserved port (i.e. port == 0 &&
1700          * transport->xprt.resvport == 1) xs_get_srcport above will
1701          * ensure that port is non-zero and we will bind as needed.
1702          */
1703         if (port == 0)
1704                 return 0;
1705
1706         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1707         do {
1708                 rpc_set_port((struct sockaddr *)&myaddr, port);
1709                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1710                                 transport->xprt.addrlen);
1711                 if (err == 0) {
1712                         transport->srcport = port;
1713                         break;
1714                 }
1715                 last = port;
1716                 port = xs_next_srcport(transport, port);
1717                 if (port > last)
1718                         nloop++;
1719         } while (err == -EADDRINUSE && nloop != 2);
1720
1721         if (myaddr.ss_family == AF_INET)
1722                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1723                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1724                                 port, err ? "failed" : "ok", err);
1725         else
1726                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1727                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1728                                 port, err ? "failed" : "ok", err);
1729         return err;
1730 }
1731
1732 /*
1733  * We don't support autobind on AF_LOCAL sockets
1734  */
1735 static void xs_local_rpcbind(struct rpc_task *task)
1736 {
1737         rcu_read_lock();
1738         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1739         rcu_read_unlock();
1740 }
1741
1742 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1743 {
1744 }
1745
1746 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1747 static struct lock_class_key xs_key[2];
1748 static struct lock_class_key xs_slock_key[2];
1749
1750 static inline void xs_reclassify_socketu(struct socket *sock)
1751 {
1752         struct sock *sk = sock->sk;
1753
1754         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1755                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1756 }
1757
1758 static inline void xs_reclassify_socket4(struct socket *sock)
1759 {
1760         struct sock *sk = sock->sk;
1761
1762         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1763                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1764 }
1765
1766 static inline void xs_reclassify_socket6(struct socket *sock)
1767 {
1768         struct sock *sk = sock->sk;
1769
1770         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1771                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1772 }
1773
1774 static inline void xs_reclassify_socket(int family, struct socket *sock)
1775 {
1776         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1777         if (sock_owned_by_user(sock->sk))
1778                 return;
1779
1780         switch (family) {
1781         case AF_LOCAL:
1782                 xs_reclassify_socketu(sock);
1783                 break;
1784         case AF_INET:
1785                 xs_reclassify_socket4(sock);
1786                 break;
1787         case AF_INET6:
1788                 xs_reclassify_socket6(sock);
1789                 break;
1790         }
1791 }
1792 #else
1793 static inline void xs_reclassify_socketu(struct socket *sock)
1794 {
1795 }
1796
1797 static inline void xs_reclassify_socket4(struct socket *sock)
1798 {
1799 }
1800
1801 static inline void xs_reclassify_socket6(struct socket *sock)
1802 {
1803 }
1804
1805 static inline void xs_reclassify_socket(int family, struct socket *sock)
1806 {
1807 }
1808 #endif
1809
1810 static void xs_dummy_setup_socket(struct work_struct *work)
1811 {
1812 }
1813
1814 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1815                 struct sock_xprt *transport, int family, int type,
1816                 int protocol, bool reuseport)
1817 {
1818         struct socket *sock;
1819         int err;
1820
1821         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1822         if (err < 0) {
1823                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1824                                 protocol, -err);
1825                 goto out;
1826         }
1827         xs_reclassify_socket(family, sock);
1828
1829         if (reuseport)
1830                 xs_sock_set_reuseport(sock);
1831
1832         err = xs_bind(transport, sock);
1833         if (err) {
1834                 sock_release(sock);
1835                 goto out;
1836         }
1837
1838         return sock;
1839 out:
1840         return ERR_PTR(err);
1841 }
1842
1843 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1844                                       struct socket *sock)
1845 {
1846         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1847                                                                         xprt);
1848
1849         if (!transport->inet) {
1850                 struct sock *sk = sock->sk;
1851
1852                 write_lock_bh(&sk->sk_callback_lock);
1853
1854                 xs_save_old_callbacks(transport, sk);
1855
1856                 sk->sk_user_data = xprt;
1857                 sk->sk_data_ready = xs_local_data_ready;
1858                 sk->sk_write_space = xs_udp_write_space;
1859                 sk->sk_error_report = xs_error_report;
1860                 sk->sk_allocation = GFP_ATOMIC;
1861
1862                 xprt_clear_connected(xprt);
1863
1864                 /* Reset to new socket */
1865                 transport->sock = sock;
1866                 transport->inet = sk;
1867
1868                 write_unlock_bh(&sk->sk_callback_lock);
1869         }
1870
1871         /* Tell the socket layer to start connecting... */
1872         xprt->stat.connect_count++;
1873         xprt->stat.connect_start = jiffies;
1874         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1875 }
1876
1877 /**
1878  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1879  * @transport: socket transport to connect
1880  */
1881 static int xs_local_setup_socket(struct sock_xprt *transport)
1882 {
1883         struct rpc_xprt *xprt = &transport->xprt;
1884         struct socket *sock;
1885         int status = -EIO;
1886
1887         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1888                                         SOCK_STREAM, 0, &sock, 1);
1889         if (status < 0) {
1890                 dprintk("RPC:       can't create AF_LOCAL "
1891                         "transport socket (%d).\n", -status);
1892                 goto out;
1893         }
1894         xs_reclassify_socketu(sock);
1895
1896         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1897                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1898
1899         status = xs_local_finish_connecting(xprt, sock);
1900         trace_rpc_socket_connect(xprt, sock, status);
1901         switch (status) {
1902         case 0:
1903                 dprintk("RPC:       xprt %p connected to %s\n",
1904                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1905                 xprt_set_connected(xprt);
1906         case -ENOBUFS:
1907                 break;
1908         case -ENOENT:
1909                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1910                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1911                 break;
1912         case -ECONNREFUSED:
1913                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1914                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915                 break;
1916         default:
1917                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1918                                 __func__, -status,
1919                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1920         }
1921
1922 out:
1923         xprt_clear_connecting(xprt);
1924         xprt_wake_pending_tasks(xprt, status);
1925         return status;
1926 }
1927
1928 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1929 {
1930         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1931         int ret;
1932
1933          if (RPC_IS_ASYNC(task)) {
1934                 /*
1935                  * We want the AF_LOCAL connect to be resolved in the
1936                  * filesystem namespace of the process making the rpc
1937                  * call.  Thus we connect synchronously.
1938                  *
1939                  * If we want to support asynchronous AF_LOCAL calls,
1940                  * we'll need to figure out how to pass a namespace to
1941                  * connect.
1942                  */
1943                 rpc_exit(task, -ENOTCONN);
1944                 return;
1945         }
1946         ret = xs_local_setup_socket(transport);
1947         if (ret && !RPC_IS_SOFTCONN(task))
1948                 msleep_interruptible(15000);
1949 }
1950
1951 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1952 /*
1953  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1954  * know that we have exclusive access to the socket), to guard against
1955  * races with xs_reset_transport.
1956  */
1957 static void xs_set_memalloc(struct rpc_xprt *xprt)
1958 {
1959         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1960                         xprt);
1961
1962         /*
1963          * If there's no sock, then we have nothing to set. The
1964          * reconnecting process will get it for us.
1965          */
1966         if (!transport->inet)
1967                 return;
1968         if (atomic_read(&xprt->swapper))
1969                 sk_set_memalloc(transport->inet);
1970 }
1971
1972 /**
1973  * xs_enable_swap - Tag this transport as being used for swap.
1974  * @xprt: transport to tag
1975  *
1976  * Take a reference to this transport on behalf of the rpc_clnt, and
1977  * optionally mark it for swapping if it wasn't already.
1978  */
1979 static int
1980 xs_enable_swap(struct rpc_xprt *xprt)
1981 {
1982         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1983
1984         if (atomic_inc_return(&xprt->swapper) != 1)
1985                 return 0;
1986         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1987                 return -ERESTARTSYS;
1988         if (xs->inet)
1989                 sk_set_memalloc(xs->inet);
1990         xprt_release_xprt(xprt, NULL);
1991         return 0;
1992 }
1993
1994 /**
1995  * xs_disable_swap - Untag this transport as being used for swap.
1996  * @xprt: transport to tag
1997  *
1998  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1999  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2000  */
2001 static void
2002 xs_disable_swap(struct rpc_xprt *xprt)
2003 {
2004         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2005
2006         if (!atomic_dec_and_test(&xprt->swapper))
2007                 return;
2008         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2009                 return;
2010         if (xs->inet)
2011                 sk_clear_memalloc(xs->inet);
2012         xprt_release_xprt(xprt, NULL);
2013 }
2014 #else
2015 static void xs_set_memalloc(struct rpc_xprt *xprt)
2016 {
2017 }
2018
2019 static int
2020 xs_enable_swap(struct rpc_xprt *xprt)
2021 {
2022         return -EINVAL;
2023 }
2024
2025 static void
2026 xs_disable_swap(struct rpc_xprt *xprt)
2027 {
2028 }
2029 #endif
2030
2031 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2032 {
2033         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2034
2035         if (!transport->inet) {
2036                 struct sock *sk = sock->sk;
2037
2038                 write_lock_bh(&sk->sk_callback_lock);
2039
2040                 xs_save_old_callbacks(transport, sk);
2041
2042                 sk->sk_user_data = xprt;
2043                 sk->sk_data_ready = xs_udp_data_ready;
2044                 sk->sk_write_space = xs_udp_write_space;
2045                 sk->sk_allocation = GFP_ATOMIC;
2046
2047                 xprt_set_connected(xprt);
2048
2049                 /* Reset to new socket */
2050                 transport->sock = sock;
2051                 transport->inet = sk;
2052
2053                 xs_set_memalloc(xprt);
2054
2055                 write_unlock_bh(&sk->sk_callback_lock);
2056         }
2057         xs_udp_do_set_buffer_size(xprt);
2058 }
2059
2060 static void xs_udp_setup_socket(struct work_struct *work)
2061 {
2062         struct sock_xprt *transport =
2063                 container_of(work, struct sock_xprt, connect_worker.work);
2064         struct rpc_xprt *xprt = &transport->xprt;
2065         struct socket *sock = transport->sock;
2066         int status = -EIO;
2067
2068         sock = xs_create_sock(xprt, transport,
2069                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2070                         IPPROTO_UDP, false);
2071         if (IS_ERR(sock))
2072                 goto out;
2073
2074         dprintk("RPC:       worker connecting xprt %p via %s to "
2075                                 "%s (port %s)\n", xprt,
2076                         xprt->address_strings[RPC_DISPLAY_PROTO],
2077                         xprt->address_strings[RPC_DISPLAY_ADDR],
2078                         xprt->address_strings[RPC_DISPLAY_PORT]);
2079
2080         xs_udp_finish_connecting(xprt, sock);
2081         trace_rpc_socket_connect(xprt, sock, 0);
2082         status = 0;
2083 out:
2084         xprt_unlock_connect(xprt, transport);
2085         xprt_clear_connecting(xprt);
2086         xprt_wake_pending_tasks(xprt, status);
2087 }
2088
2089 /**
2090  * xs_tcp_shutdown - gracefully shut down a TCP socket
2091  * @xprt: transport
2092  *
2093  * Initiates a graceful shutdown of the TCP socket by calling the
2094  * equivalent of shutdown(SHUT_RDWR);
2095  */
2096 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2097 {
2098         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2099         struct socket *sock = transport->sock;
2100
2101         if (sock == NULL)
2102                 return;
2103         if (xprt_connected(xprt)) {
2104                 kernel_sock_shutdown(sock, SHUT_RDWR);
2105                 trace_rpc_socket_shutdown(xprt, sock);
2106         } else
2107                 xs_reset_transport(transport);
2108 }
2109
2110 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2111 {
2112         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2113         int ret = -ENOTCONN;
2114
2115         if (!transport->inet) {
2116                 struct sock *sk = sock->sk;
2117                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2118                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2119                 unsigned int opt_on = 1;
2120                 unsigned int timeo;
2121
2122                 /* TCP Keepalive options */
2123                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2124                                 (char *)&opt_on, sizeof(opt_on));
2125                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2126                                 (char *)&keepidle, sizeof(keepidle));
2127                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2128                                 (char *)&keepidle, sizeof(keepidle));
2129                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2130                                 (char *)&keepcnt, sizeof(keepcnt));
2131
2132                 /* TCP user timeout (see RFC5482) */
2133                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2134                         (xprt->timeout->to_retries + 1);
2135                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2136                                 (char *)&timeo, sizeof(timeo));
2137
2138                 write_lock_bh(&sk->sk_callback_lock);
2139
2140                 xs_save_old_callbacks(transport, sk);
2141
2142                 sk->sk_user_data = xprt;
2143                 sk->sk_data_ready = xs_tcp_data_ready;
2144                 sk->sk_state_change = xs_tcp_state_change;
2145                 sk->sk_write_space = xs_tcp_write_space;
2146                 sk->sk_error_report = xs_error_report;
2147                 sk->sk_allocation = GFP_ATOMIC;
2148
2149                 /* socket options */
2150                 sock_reset_flag(sk, SOCK_LINGER);
2151                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2152
2153                 xprt_clear_connected(xprt);
2154
2155                 /* Reset to new socket */
2156                 transport->sock = sock;
2157                 transport->inet = sk;
2158
2159                 write_unlock_bh(&sk->sk_callback_lock);
2160         }
2161
2162         if (!xprt_bound(xprt))
2163                 goto out;
2164
2165         xs_set_memalloc(xprt);
2166
2167         /* Tell the socket layer to start connecting... */
2168         xprt->stat.connect_count++;
2169         xprt->stat.connect_start = jiffies;
2170         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2171         switch (ret) {
2172         case 0:
2173                 xs_set_srcport(transport, sock);
2174         case -EINPROGRESS:
2175                 /* SYN_SENT! */
2176                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2177                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2178         }
2179 out:
2180         return ret;
2181 }
2182
2183 /**
2184  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2185  *
2186  * Invoked by a work queue tasklet.
2187  */
2188 static void xs_tcp_setup_socket(struct work_struct *work)
2189 {
2190         struct sock_xprt *transport =
2191                 container_of(work, struct sock_xprt, connect_worker.work);
2192         struct socket *sock = transport->sock;
2193         struct rpc_xprt *xprt = &transport->xprt;
2194         int status = -EIO;
2195
2196         if (!sock) {
2197                 sock = xs_create_sock(xprt, transport,
2198                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2199                                 IPPROTO_TCP, true);
2200                 if (IS_ERR(sock)) {
2201                         status = PTR_ERR(sock);
2202                         goto out;
2203                 }
2204         }
2205
2206         dprintk("RPC:       worker connecting xprt %p via %s to "
2207                                 "%s (port %s)\n", xprt,
2208                         xprt->address_strings[RPC_DISPLAY_PROTO],
2209                         xprt->address_strings[RPC_DISPLAY_ADDR],
2210                         xprt->address_strings[RPC_DISPLAY_PORT]);
2211
2212         status = xs_tcp_finish_connecting(xprt, sock);
2213         trace_rpc_socket_connect(xprt, sock, status);
2214         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2215                         xprt, -status, xprt_connected(xprt),
2216                         sock->sk->sk_state);
2217         switch (status) {
2218         default:
2219                 printk("%s: connect returned unhandled error %d\n",
2220                         __func__, status);
2221         case -EADDRNOTAVAIL:
2222                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2223                  * and retry
2224                  */
2225                 xs_tcp_force_close(xprt);
2226                 break;
2227         case 0:
2228         case -EINPROGRESS:
2229         case -EALREADY:
2230                 xprt_unlock_connect(xprt, transport);
2231                 xprt_clear_connecting(xprt);
2232                 return;
2233         case -EINVAL:
2234                 /* Happens, for instance, if the user specified a link
2235                  * local IPv6 address without a scope-id.
2236                  */
2237         case -ECONNREFUSED:
2238         case -ECONNRESET:
2239         case -ENETUNREACH:
2240         case -EADDRINUSE:
2241         case -ENOBUFS:
2242                 /* retry with existing socket, after a delay */
2243                 xs_tcp_force_close(xprt);
2244                 goto out;
2245         }
2246         status = -EAGAIN;
2247 out:
2248         xprt_unlock_connect(xprt, transport);
2249         xprt_clear_connecting(xprt);
2250         xprt_wake_pending_tasks(xprt, status);
2251 }
2252
2253 /**
2254  * xs_connect - connect a socket to a remote endpoint
2255  * @xprt: pointer to transport structure
2256  * @task: address of RPC task that manages state of connect request
2257  *
2258  * TCP: If the remote end dropped the connection, delay reconnecting.
2259  *
2260  * UDP socket connects are synchronous, but we use a work queue anyway
2261  * to guarantee that even unprivileged user processes can set up a
2262  * socket on a privileged port.
2263  *
2264  * If a UDP socket connect fails, the delay behavior here prevents
2265  * retry floods (hard mounts).
2266  */
2267 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2268 {
2269         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2270
2271         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2272
2273         /* Start by resetting any existing state */
2274         xs_reset_transport(transport);
2275
2276         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2277                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2278                                 "seconds\n",
2279                                 xprt, xprt->reestablish_timeout / HZ);
2280                 queue_delayed_work(rpciod_workqueue,
2281                                    &transport->connect_worker,
2282                                    xprt->reestablish_timeout);
2283                 xprt->reestablish_timeout <<= 1;
2284                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2285                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2286                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2287                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2288         } else {
2289                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2290                 queue_delayed_work(rpciod_workqueue,
2291                                    &transport->connect_worker, 0);
2292         }
2293 }
2294
2295 /**
2296  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2297  * @xprt: rpc_xprt struct containing statistics
2298  * @seq: output file
2299  *
2300  */
2301 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2302 {
2303         long idle_time = 0;
2304
2305         if (xprt_connected(xprt))
2306                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2307
2308         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2309                         "%llu %llu %lu %llu %llu\n",
2310                         xprt->stat.bind_count,
2311                         xprt->stat.connect_count,
2312                         xprt->stat.connect_time,
2313                         idle_time,
2314                         xprt->stat.sends,
2315                         xprt->stat.recvs,
2316                         xprt->stat.bad_xids,
2317                         xprt->stat.req_u,
2318                         xprt->stat.bklog_u,
2319                         xprt->stat.max_slots,
2320                         xprt->stat.sending_u,
2321                         xprt->stat.pending_u);
2322 }
2323
2324 /**
2325  * xs_udp_print_stats - display UDP socket-specifc stats
2326  * @xprt: rpc_xprt struct containing statistics
2327  * @seq: output file
2328  *
2329  */
2330 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2331 {
2332         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2333
2334         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2335                         "%lu %llu %llu\n",
2336                         transport->srcport,
2337                         xprt->stat.bind_count,
2338                         xprt->stat.sends,
2339                         xprt->stat.recvs,
2340                         xprt->stat.bad_xids,
2341                         xprt->stat.req_u,
2342                         xprt->stat.bklog_u,
2343                         xprt->stat.max_slots,
2344                         xprt->stat.sending_u,
2345                         xprt->stat.pending_u);
2346 }
2347
2348 /**
2349  * xs_tcp_print_stats - display TCP socket-specifc stats
2350  * @xprt: rpc_xprt struct containing statistics
2351  * @seq: output file
2352  *
2353  */
2354 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2355 {
2356         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2357         long idle_time = 0;
2358
2359         if (xprt_connected(xprt))
2360                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2361
2362         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2363                         "%llu %llu %lu %llu %llu\n",
2364                         transport->srcport,
2365                         xprt->stat.bind_count,
2366                         xprt->stat.connect_count,
2367                         xprt->stat.connect_time,
2368                         idle_time,
2369                         xprt->stat.sends,
2370                         xprt->stat.recvs,
2371                         xprt->stat.bad_xids,
2372                         xprt->stat.req_u,
2373                         xprt->stat.bklog_u,
2374                         xprt->stat.max_slots,
2375                         xprt->stat.sending_u,
2376                         xprt->stat.pending_u);
2377 }
2378
2379 /*
2380  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2381  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2382  * to use the server side send routines.
2383  */
2384 static void *bc_malloc(struct rpc_task *task, size_t size)
2385 {
2386         struct page *page;
2387         struct rpc_buffer *buf;
2388
2389         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2390         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2391                 return NULL;
2392
2393         page = alloc_page(GFP_KERNEL);
2394         if (!page)
2395                 return NULL;
2396
2397         buf = page_address(page);
2398         buf->len = PAGE_SIZE;
2399
2400         return buf->data;
2401 }
2402
2403 /*
2404  * Free the space allocated in the bc_alloc routine
2405  */
2406 static void bc_free(void *buffer)
2407 {
2408         struct rpc_buffer *buf;
2409
2410         if (!buffer)
2411                 return;
2412
2413         buf = container_of(buffer, struct rpc_buffer, data);
2414         free_page((unsigned long)buf);
2415 }
2416
2417 /*
2418  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2419  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2420  */
2421 static int bc_sendto(struct rpc_rqst *req)
2422 {
2423         int len;
2424         struct xdr_buf *xbufp = &req->rq_snd_buf;
2425         struct rpc_xprt *xprt = req->rq_xprt;
2426         struct sock_xprt *transport =
2427                                 container_of(xprt, struct sock_xprt, xprt);
2428         struct socket *sock = transport->sock;
2429         unsigned long headoff;
2430         unsigned long tailoff;
2431
2432         xs_encode_stream_record_marker(xbufp);
2433
2434         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2435         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2436         len = svc_send_common(sock, xbufp,
2437                               virt_to_page(xbufp->head[0].iov_base), headoff,
2438                               xbufp->tail[0].iov_base, tailoff);
2439
2440         if (len != xbufp->len) {
2441                 printk(KERN_NOTICE "Error sending entire callback!\n");
2442                 len = -EAGAIN;
2443         }
2444
2445         return len;
2446 }
2447
2448 /*
2449  * The send routine. Borrows from svc_send
2450  */
2451 static int bc_send_request(struct rpc_task *task)
2452 {
2453         struct rpc_rqst *req = task->tk_rqstp;
2454         struct svc_xprt *xprt;
2455         u32                     len;
2456
2457         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2458         /*
2459          * Get the server socket associated with this callback xprt
2460          */
2461         xprt = req->rq_xprt->bc_xprt;
2462
2463         /*
2464          * Grab the mutex to serialize data as the connection is shared
2465          * with the fore channel
2466          */
2467         if (!mutex_trylock(&xprt->xpt_mutex)) {
2468                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2469                 if (!mutex_trylock(&xprt->xpt_mutex))
2470                         return -EAGAIN;
2471                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2472         }
2473         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2474                 len = -ENOTCONN;
2475         else
2476                 len = bc_sendto(req);
2477         mutex_unlock(&xprt->xpt_mutex);
2478
2479         if (len > 0)
2480                 len = 0;
2481
2482         return len;
2483 }
2484
2485 /*
2486  * The close routine. Since this is client initiated, we do nothing
2487  */
2488
2489 static void bc_close(struct rpc_xprt *xprt)
2490 {
2491 }
2492
2493 /*
2494  * The xprt destroy routine. Again, because this connection is client
2495  * initiated, we do nothing
2496  */
2497
2498 static void bc_destroy(struct rpc_xprt *xprt)
2499 {
2500         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2501
2502         xs_xprt_free(xprt);
2503         module_put(THIS_MODULE);
2504 }
2505
2506 static struct rpc_xprt_ops xs_local_ops = {
2507         .reserve_xprt           = xprt_reserve_xprt,
2508         .release_xprt           = xs_tcp_release_xprt,
2509         .alloc_slot             = xprt_alloc_slot,
2510         .rpcbind                = xs_local_rpcbind,
2511         .set_port               = xs_local_set_port,
2512         .connect                = xs_local_connect,
2513         .buf_alloc              = rpc_malloc,
2514         .buf_free               = rpc_free,
2515         .send_request           = xs_local_send_request,
2516         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2517         .close                  = xs_close,
2518         .destroy                = xs_destroy,
2519         .print_stats            = xs_local_print_stats,
2520         .enable_swap            = xs_enable_swap,
2521         .disable_swap           = xs_disable_swap,
2522 };
2523
2524 static struct rpc_xprt_ops xs_udp_ops = {
2525         .set_buffer_size        = xs_udp_set_buffer_size,
2526         .reserve_xprt           = xprt_reserve_xprt_cong,
2527         .release_xprt           = xprt_release_xprt_cong,
2528         .alloc_slot             = xprt_alloc_slot,
2529         .rpcbind                = rpcb_getport_async,
2530         .set_port               = xs_set_port,
2531         .connect                = xs_connect,
2532         .buf_alloc              = rpc_malloc,
2533         .buf_free               = rpc_free,
2534         .send_request           = xs_udp_send_request,
2535         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2536         .timer                  = xs_udp_timer,
2537         .release_request        = xprt_release_rqst_cong,
2538         .close                  = xs_close,
2539         .destroy                = xs_destroy,
2540         .print_stats            = xs_udp_print_stats,
2541         .enable_swap            = xs_enable_swap,
2542         .disable_swap           = xs_disable_swap,
2543         .inject_disconnect      = xs_inject_disconnect,
2544 };
2545
2546 static struct rpc_xprt_ops xs_tcp_ops = {
2547         .reserve_xprt           = xprt_reserve_xprt,
2548         .release_xprt           = xs_tcp_release_xprt,
2549         .alloc_slot             = xprt_lock_and_alloc_slot,
2550         .rpcbind                = rpcb_getport_async,
2551         .set_port               = xs_set_port,
2552         .connect                = xs_connect,
2553         .buf_alloc              = rpc_malloc,
2554         .buf_free               = rpc_free,
2555         .send_request           = xs_tcp_send_request,
2556         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2557         .close                  = xs_tcp_shutdown,
2558         .destroy                = xs_destroy,
2559         .print_stats            = xs_tcp_print_stats,
2560         .enable_swap            = xs_enable_swap,
2561         .disable_swap           = xs_disable_swap,
2562         .inject_disconnect      = xs_inject_disconnect,
2563 };
2564
2565 /*
2566  * The rpc_xprt_ops for the server backchannel
2567  */
2568
2569 static struct rpc_xprt_ops bc_tcp_ops = {
2570         .reserve_xprt           = xprt_reserve_xprt,
2571         .release_xprt           = xprt_release_xprt,
2572         .alloc_slot             = xprt_alloc_slot,
2573         .buf_alloc              = bc_malloc,
2574         .buf_free               = bc_free,
2575         .send_request           = bc_send_request,
2576         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2577         .close                  = bc_close,
2578         .destroy                = bc_destroy,
2579         .print_stats            = xs_tcp_print_stats,
2580         .enable_swap            = xs_enable_swap,
2581         .disable_swap           = xs_disable_swap,
2582         .inject_disconnect      = xs_inject_disconnect,
2583 };
2584
2585 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2586 {
2587         static const struct sockaddr_in sin = {
2588                 .sin_family             = AF_INET,
2589                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2590         };
2591         static const struct sockaddr_in6 sin6 = {
2592                 .sin6_family            = AF_INET6,
2593                 .sin6_addr              = IN6ADDR_ANY_INIT,
2594         };
2595
2596         switch (family) {
2597         case AF_LOCAL:
2598                 break;
2599         case AF_INET:
2600                 memcpy(sap, &sin, sizeof(sin));
2601                 break;
2602         case AF_INET6:
2603                 memcpy(sap, &sin6, sizeof(sin6));
2604                 break;
2605         default:
2606                 dprintk("RPC:       %s: Bad address family\n", __func__);
2607                 return -EAFNOSUPPORT;
2608         }
2609         return 0;
2610 }
2611
2612 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2613                                       unsigned int slot_table_size,
2614                                       unsigned int max_slot_table_size)
2615 {
2616         struct rpc_xprt *xprt;
2617         struct sock_xprt *new;
2618
2619         if (args->addrlen > sizeof(xprt->addr)) {
2620                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2621                 return ERR_PTR(-EBADF);
2622         }
2623
2624         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2625                         max_slot_table_size);
2626         if (xprt == NULL) {
2627                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2628                                 "rpc_xprt\n");
2629                 return ERR_PTR(-ENOMEM);
2630         }
2631
2632         new = container_of(xprt, struct sock_xprt, xprt);
2633         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2634         xprt->addrlen = args->addrlen;
2635         if (args->srcaddr)
2636                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2637         else {
2638                 int err;
2639                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2640                                         (struct sockaddr *)&new->srcaddr);
2641                 if (err != 0) {
2642                         xprt_free(xprt);
2643                         return ERR_PTR(err);
2644                 }
2645         }
2646
2647         return xprt;
2648 }
2649
2650 static const struct rpc_timeout xs_local_default_timeout = {
2651         .to_initval = 10 * HZ,
2652         .to_maxval = 10 * HZ,
2653         .to_retries = 2,
2654 };
2655
2656 /**
2657  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2658  * @args: rpc transport creation arguments
2659  *
2660  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2661  */
2662 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2663 {
2664         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2665         struct sock_xprt *transport;
2666         struct rpc_xprt *xprt;
2667         struct rpc_xprt *ret;
2668
2669         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2670                         xprt_max_tcp_slot_table_entries);
2671         if (IS_ERR(xprt))
2672                 return xprt;
2673         transport = container_of(xprt, struct sock_xprt, xprt);
2674
2675         xprt->prot = 0;
2676         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2677         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2678
2679         xprt->bind_timeout = XS_BIND_TO;
2680         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2681         xprt->idle_timeout = XS_IDLE_DISC_TO;
2682
2683         xprt->ops = &xs_local_ops;
2684         xprt->timeout = &xs_local_default_timeout;
2685
2686         INIT_DELAYED_WORK(&transport->connect_worker,
2687                         xs_dummy_setup_socket);
2688
2689         switch (sun->sun_family) {
2690         case AF_LOCAL:
2691                 if (sun->sun_path[0] != '/') {
2692                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2693                                         sun->sun_path);
2694                         ret = ERR_PTR(-EINVAL);
2695                         goto out_err;
2696                 }
2697                 xprt_set_bound(xprt);
2698                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2699                 ret = ERR_PTR(xs_local_setup_socket(transport));
2700                 if (ret)
2701                         goto out_err;
2702                 break;
2703         default:
2704                 ret = ERR_PTR(-EAFNOSUPPORT);
2705                 goto out_err;
2706         }
2707
2708         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2709                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2710
2711         if (try_module_get(THIS_MODULE))
2712                 return xprt;
2713         ret = ERR_PTR(-EINVAL);
2714 out_err:
2715         xs_xprt_free(xprt);
2716         return ret;
2717 }
2718
2719 static const struct rpc_timeout xs_udp_default_timeout = {
2720         .to_initval = 5 * HZ,
2721         .to_maxval = 30 * HZ,
2722         .to_increment = 5 * HZ,
2723         .to_retries = 5,
2724 };
2725
2726 /**
2727  * xs_setup_udp - Set up transport to use a UDP socket
2728  * @args: rpc transport creation arguments
2729  *
2730  */
2731 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2732 {
2733         struct sockaddr *addr = args->dstaddr;
2734         struct rpc_xprt *xprt;
2735         struct sock_xprt *transport;
2736         struct rpc_xprt *ret;
2737
2738         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2739                         xprt_udp_slot_table_entries);
2740         if (IS_ERR(xprt))
2741                 return xprt;
2742         transport = container_of(xprt, struct sock_xprt, xprt);
2743
2744         xprt->prot = IPPROTO_UDP;
2745         xprt->tsh_size = 0;
2746         /* XXX: header size can vary due to auth type, IPv6, etc. */
2747         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2748
2749         xprt->bind_timeout = XS_BIND_TO;
2750         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2751         xprt->idle_timeout = XS_IDLE_DISC_TO;
2752
2753         xprt->ops = &xs_udp_ops;
2754
2755         xprt->timeout = &xs_udp_default_timeout;
2756
2757         switch (addr->sa_family) {
2758         case AF_INET:
2759                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2760                         xprt_set_bound(xprt);
2761
2762                 INIT_DELAYED_WORK(&transport->connect_worker,
2763                                         xs_udp_setup_socket);
2764                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2765                 break;
2766         case AF_INET6:
2767                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2768                         xprt_set_bound(xprt);
2769
2770                 INIT_DELAYED_WORK(&transport->connect_worker,
2771                                         xs_udp_setup_socket);
2772                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2773                 break;
2774         default:
2775                 ret = ERR_PTR(-EAFNOSUPPORT);
2776                 goto out_err;
2777         }
2778
2779         if (xprt_bound(xprt))
2780                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2781                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2782                                 xprt->address_strings[RPC_DISPLAY_PORT],
2783                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2784         else
2785                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2786                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2787                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2788
2789         if (try_module_get(THIS_MODULE))
2790                 return xprt;
2791         ret = ERR_PTR(-EINVAL);
2792 out_err:
2793         xs_xprt_free(xprt);
2794         return ret;
2795 }
2796
2797 static const struct rpc_timeout xs_tcp_default_timeout = {
2798         .to_initval = 60 * HZ,
2799         .to_maxval = 60 * HZ,
2800         .to_retries = 2,
2801 };
2802
2803 /**
2804  * xs_setup_tcp - Set up transport to use a TCP socket
2805  * @args: rpc transport creation arguments
2806  *
2807  */
2808 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2809 {
2810         struct sockaddr *addr = args->dstaddr;
2811         struct rpc_xprt *xprt;
2812         struct sock_xprt *transport;
2813         struct rpc_xprt *ret;
2814         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2815
2816         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2817                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2818
2819         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2820                         max_slot_table_size);
2821         if (IS_ERR(xprt))
2822                 return xprt;
2823         transport = container_of(xprt, struct sock_xprt, xprt);
2824
2825         xprt->prot = IPPROTO_TCP;
2826         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2827         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2828
2829         xprt->bind_timeout = XS_BIND_TO;
2830         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2831         xprt->idle_timeout = XS_IDLE_DISC_TO;
2832
2833         xprt->ops = &xs_tcp_ops;
2834         xprt->timeout = &xs_tcp_default_timeout;
2835
2836         switch (addr->sa_family) {
2837         case AF_INET:
2838                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2839                         xprt_set_bound(xprt);
2840
2841                 INIT_DELAYED_WORK(&transport->connect_worker,
2842                                         xs_tcp_setup_socket);
2843                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2844                 break;
2845         case AF_INET6:
2846                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2847                         xprt_set_bound(xprt);
2848
2849                 INIT_DELAYED_WORK(&transport->connect_worker,
2850                                         xs_tcp_setup_socket);
2851                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2852                 break;
2853         default:
2854                 ret = ERR_PTR(-EAFNOSUPPORT);
2855                 goto out_err;
2856         }
2857
2858         if (xprt_bound(xprt))
2859                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2860                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2861                                 xprt->address_strings[RPC_DISPLAY_PORT],
2862                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2863         else
2864                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2865                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2866                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2867
2868         if (try_module_get(THIS_MODULE))
2869                 return xprt;
2870         ret = ERR_PTR(-EINVAL);
2871 out_err:
2872         xs_xprt_free(xprt);
2873         return ret;
2874 }
2875
2876 /**
2877  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2878  * @args: rpc transport creation arguments
2879  *
2880  */
2881 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2882 {
2883         struct sockaddr *addr = args->dstaddr;
2884         struct rpc_xprt *xprt;
2885         struct sock_xprt *transport;
2886         struct svc_sock *bc_sock;
2887         struct rpc_xprt *ret;
2888
2889         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2890                         xprt_tcp_slot_table_entries);
2891         if (IS_ERR(xprt))
2892                 return xprt;
2893         transport = container_of(xprt, struct sock_xprt, xprt);
2894
2895         xprt->prot = IPPROTO_TCP;
2896         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2897         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2898         xprt->timeout = &xs_tcp_default_timeout;
2899
2900         /* backchannel */
2901         xprt_set_bound(xprt);
2902         xprt->bind_timeout = 0;
2903         xprt->reestablish_timeout = 0;
2904         xprt->idle_timeout = 0;
2905
2906         xprt->ops = &bc_tcp_ops;
2907
2908         switch (addr->sa_family) {
2909         case AF_INET:
2910                 xs_format_peer_addresses(xprt, "tcp",
2911                                          RPCBIND_NETID_TCP);
2912                 break;
2913         case AF_INET6:
2914                 xs_format_peer_addresses(xprt, "tcp",
2915                                    RPCBIND_NETID_TCP6);
2916                 break;
2917         default:
2918                 ret = ERR_PTR(-EAFNOSUPPORT);
2919                 goto out_err;
2920         }
2921
2922         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2923                         xprt->address_strings[RPC_DISPLAY_ADDR],
2924                         xprt->address_strings[RPC_DISPLAY_PORT],
2925                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2926
2927         /*
2928          * Once we've associated a backchannel xprt with a connection,
2929          * we want to keep it around as long as the connection lasts,
2930          * in case we need to start using it for a backchannel again;
2931          * this reference won't be dropped until bc_xprt is destroyed.
2932          */
2933         xprt_get(xprt);
2934         args->bc_xprt->xpt_bc_xprt = xprt;
2935         xprt->bc_xprt = args->bc_xprt;
2936         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2937         transport->sock = bc_sock->sk_sock;
2938         transport->inet = bc_sock->sk_sk;
2939
2940         /*
2941          * Since we don't want connections for the backchannel, we set
2942          * the xprt status to connected
2943          */
2944         xprt_set_connected(xprt);
2945
2946         if (try_module_get(THIS_MODULE))
2947                 return xprt;
2948
2949         args->bc_xprt->xpt_bc_xprt = NULL;
2950         xprt_put(xprt);
2951         ret = ERR_PTR(-EINVAL);
2952 out_err:
2953         xs_xprt_free(xprt);
2954         return ret;
2955 }
2956
2957 static struct xprt_class        xs_local_transport = {
2958         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2959         .name           = "named UNIX socket",
2960         .owner          = THIS_MODULE,
2961         .ident          = XPRT_TRANSPORT_LOCAL,
2962         .setup          = xs_setup_local,
2963 };
2964
2965 static struct xprt_class        xs_udp_transport = {
2966         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2967         .name           = "udp",
2968         .owner          = THIS_MODULE,
2969         .ident          = XPRT_TRANSPORT_UDP,
2970         .setup          = xs_setup_udp,
2971 };
2972
2973 static struct xprt_class        xs_tcp_transport = {
2974         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2975         .name           = "tcp",
2976         .owner          = THIS_MODULE,
2977         .ident          = XPRT_TRANSPORT_TCP,
2978         .setup          = xs_setup_tcp,
2979 };
2980
2981 static struct xprt_class        xs_bc_tcp_transport = {
2982         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2983         .name           = "tcp NFSv4.1 backchannel",
2984         .owner          = THIS_MODULE,
2985         .ident          = XPRT_TRANSPORT_BC_TCP,
2986         .setup          = xs_setup_bc_tcp,
2987 };
2988
2989 /**
2990  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2991  *
2992  */
2993 int init_socket_xprt(void)
2994 {
2995 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2996         if (!sunrpc_table_header)
2997                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2998 #endif
2999
3000         xprt_register_transport(&xs_local_transport);
3001         xprt_register_transport(&xs_udp_transport);
3002         xprt_register_transport(&xs_tcp_transport);
3003         xprt_register_transport(&xs_bc_tcp_transport);
3004
3005         return 0;
3006 }
3007
3008 /**
3009  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3010  *
3011  */
3012 void cleanup_socket_xprt(void)
3013 {
3014 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3015         if (sunrpc_table_header) {
3016                 unregister_sysctl_table(sunrpc_table_header);
3017                 sunrpc_table_header = NULL;
3018         }
3019 #endif
3020
3021         xprt_unregister_transport(&xs_local_transport);
3022         xprt_unregister_transport(&xs_udp_transport);
3023         xprt_unregister_transport(&xs_tcp_transport);
3024         xprt_unregister_transport(&xs_bc_tcp_transport);
3025 }
3026
3027 static int param_set_uint_minmax(const char *val,
3028                 const struct kernel_param *kp,
3029                 unsigned int min, unsigned int max)
3030 {
3031         unsigned int num;
3032         int ret;
3033
3034         if (!val)
3035                 return -EINVAL;
3036         ret = kstrtouint(val, 0, &num);
3037         if (ret == -EINVAL || num < min || num > max)
3038                 return -EINVAL;
3039         *((unsigned int *)kp->arg) = num;
3040         return 0;
3041 }
3042
3043 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3044 {
3045         return param_set_uint_minmax(val, kp,
3046                         RPC_MIN_RESVPORT,
3047                         RPC_MAX_RESVPORT);
3048 }
3049
3050 static const struct kernel_param_ops param_ops_portnr = {
3051         .set = param_set_portnr,
3052         .get = param_get_uint,
3053 };
3054
3055 #define param_check_portnr(name, p) \
3056         __param_check(name, p, unsigned int);
3057
3058 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3059 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3060
3061 static int param_set_slot_table_size(const char *val,
3062                                      const struct kernel_param *kp)
3063 {
3064         return param_set_uint_minmax(val, kp,
3065                         RPC_MIN_SLOT_TABLE,
3066                         RPC_MAX_SLOT_TABLE);
3067 }
3068
3069 static const struct kernel_param_ops param_ops_slot_table_size = {
3070         .set = param_set_slot_table_size,
3071         .get = param_get_uint,
3072 };
3073
3074 #define param_check_slot_table_size(name, p) \
3075         __param_check(name, p, unsigned int);
3076
3077 static int param_set_max_slot_table_size(const char *val,
3078                                      const struct kernel_param *kp)
3079 {
3080         return param_set_uint_minmax(val, kp,
3081                         RPC_MIN_SLOT_TABLE,
3082                         RPC_MAX_SLOT_TABLE_LIMIT);
3083 }
3084
3085 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3086         .set = param_set_max_slot_table_size,
3087         .get = param_get_uint,
3088 };
3089
3090 #define param_check_max_slot_table_size(name, p) \
3091         __param_check(name, p, unsigned int);
3092
3093 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3094                    slot_table_size, 0644);
3095 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3096                    max_slot_table_size, 0644);
3097 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3098                    slot_table_size, 0644);
3099