kernel/net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68 #define XS_TCP_LINGER_TO        (15U * HZ)
  69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71 /*
  72  * We can register our own files under /proc/sys/sunrpc by
  73  * calling register_sysctl_table() again.  The files in that
  74  * directory become the union of all files registered there.
  75  *
  76  * We simply need to make sure that we don't collide with
  77  * someone else's file names!
  78  */
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 181
 182 /*
 183  * TCP idle timeout; client drops the transport socket if it is idle
 184  * for this long.  Note that we also timeout UDP sockets to prevent
 185  * holding port numbers when there is no RPC traffic.
 186  */
 187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 188
 189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 190 # undef  RPC_DEBUG_DATA
 191 # define RPCDBG_FACILITY        RPCDBG_TRANS
 192 #endif
 193
 194 #ifdef RPC_DEBUG_DATA
 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196 {
 197         u8 *buf = (u8 *) packet;
 198         int j;
 199
 200         dprintk("RPC:       %s\n", msg);
 201         for (j = 0; j < count && j < 128; j += 4) {
 202                 if (!(j & 31)) {
 203                         if (j)
 204                                 dprintk("\n");
 205                         dprintk("0x%04x ", j);
 206                 }
 207                 dprintk("%02x%02x%02x%02x ",
 208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209         }
 210         dprintk("\n");
 211 }
 212 #else
 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214 {
 215         /* NOP */
 216 }
 217 #endif
 218
 219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 220 {
 221         return (struct rpc_xprt *) sk->sk_user_data;
 222 }
 223
 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 225 {
 226         return (struct sockaddr *) &xprt->addr;
 227 }
 228
 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 230 {
 231         return (struct sockaddr_un *) &xprt->addr;
 232 }
 233
 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr_in *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_in6 *) &xprt->addr;
 242 }
 243
 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 245 {
 246         struct sockaddr *sap = xs_addr(xprt);
 247         struct sockaddr_in6 *sin6;
 248         struct sockaddr_in *sin;
 249         struct sockaddr_un *sun;
 250         char buf[128];
 251
 252         switch (sap->sa_family) {
 253         case AF_LOCAL:
 254                 sun = xs_addr_un(xprt);
 255                 strlcpy(buf, sun->sun_path, sizeof(buf));
 256                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 257                                                 kstrdup(buf, GFP_KERNEL);
 258                 break;
 259         case AF_INET:
 260                 (void)rpc_ntop(sap, buf, sizeof(buf));
 261                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 262                                                 kstrdup(buf, GFP_KERNEL);
 263                 sin = xs_addr_in(xprt);
 264                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 265                 break;
 266         case AF_INET6:
 267                 (void)rpc_ntop(sap, buf, sizeof(buf));
 268                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 269                                                 kstrdup(buf, GFP_KERNEL);
 270                 sin6 = xs_addr_in6(xprt);
 271                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 272                 break;
 273         default:
 274                 BUG();
 275         }
 276
 277         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 278 }
 279
 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 281 {
 282         struct sockaddr *sap = xs_addr(xprt);
 283         char buf[128];
 284
 285         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 286         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 287
 288         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 289         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 290 }
 291
 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 293                                      const char *protocol,
 294                                      const char *netid)
 295 {
 296         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 297         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 298         xs_format_common_peer_addresses(xprt);
 299         xs_format_common_peer_ports(xprt);
 300 }
 301
 302 static void xs_update_peer_port(struct rpc_xprt *xprt)
 303 {
 304         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 305         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 306
 307         xs_format_common_peer_ports(xprt);
 308 }
 309
 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 311 {
 312         unsigned int i;
 313
 314         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 315                 switch (i) {
 316                 case RPC_DISPLAY_PROTO:
 317                 case RPC_DISPLAY_NETID:
 318                         continue;
 319                 default:
 320                         kfree(xprt->address_strings[i]);
 321                 }
 322 }
 323
 324 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 325
 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 327 {
 328         struct msghdr msg = {
 329                 .msg_name       = addr,
 330                 .msg_namelen    = addrlen,
 331                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 332         };
 333         struct kvec iov = {
 334                 .iov_base       = vec->iov_base + base,
 335                 .iov_len        = vec->iov_len - base,
 336         };
 337
 338         if (iov.iov_len != 0)
 339                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 340         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 341 }
 342
 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 344 {
 345         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 346                         int offset, size_t size, int flags);
 347         struct page **ppage;
 348         unsigned int remainder;
 349         int err;
 350
 351         remainder = xdr->page_len - base;
 352         base += xdr->page_base;
 353         ppage = xdr->pages + (base >> PAGE_SHIFT);
 354         base &= ~PAGE_MASK;
 355         do_sendpage = sock->ops->sendpage;
 356         if (!zerocopy)
 357                 do_sendpage = sock_no_sendpage;
 358         for(;;) {
 359                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 360                 int flags = XS_SENDMSG_FLAGS;
 361
 362                 remainder -= len;
 363                 if (more)
 364                         flags |= MSG_MORE;
 365                 if (remainder != 0)
 366                         flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
 367                 err = do_sendpage(sock, *ppage, base, len, flags);
 368                 if (remainder == 0 || err != len)
 369                         break;
 370                 *sent_p += err;
 371                 ppage++;
 372                 base = 0;
 373         }
 374         if (err > 0) {
 375                 *sent_p += err;
 376                 err = 0;
 377         }
 378         return err;
 379 }
 380
 381 /**
 382  * xs_sendpages - write pages directly to a socket
 383  * @sock: socket to send on
 384  * @addr: UDP only -- address of destination
 385  * @addrlen: UDP only -- length of destination address
 386  * @xdr: buffer containing this request
 387  * @base: starting position in the buffer
 388  * @zerocopy: true if it is safe to use sendpage()
 389  * @sent_p: return the total number of bytes successfully queued for sending
 390  *
 391  */
 392 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 393 {
 394         unsigned int remainder = xdr->len - base;
 395         int err = 0;
 396         int sent = 0;
 397
 398         if (unlikely(!sock))
 399                 return -ENOTSOCK;
 400
 401         if (base != 0) {
 402                 addr = NULL;
 403                 addrlen = 0;
 404         }
 405
 406         if (base < xdr->head[0].iov_len || addr != NULL) {
 407                 unsigned int len = xdr->head[0].iov_len - base;
 408                 remainder -= len;
 409                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 410                 if (remainder == 0 || err != len)
 411                         goto out;
 412                 *sent_p += err;
 413                 base = 0;
 414         } else
 415                 base -= xdr->head[0].iov_len;
 416
 417         if (base < xdr->page_len) {
 418                 unsigned int len = xdr->page_len - base;
 419                 remainder -= len;
 420                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 421                 *sent_p += sent;
 422                 if (remainder == 0 || sent != len)
 423                         goto out;
 424                 base = 0;
 425         } else
 426                 base -= xdr->page_len;
 427
 428         if (base >= xdr->tail[0].iov_len)
 429                 return 0;
 430         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 431 out:
 432         if (err > 0) {
 433                 *sent_p += err;
 434                 err = 0;
 435         }
 436         return err;
 437 }
 438
 439 static void xs_nospace_callback(struct rpc_task *task)
 440 {
 441         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 442
 443         transport->inet->sk_write_pending--;
 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                 /* wait for more buffer space */
 469                 sk->sk_write_pending++;
 470                 xprt_wait_for_buffer_space(task, xs_nospace_callback);
 471         } else
 472                 ret = -ENOTCONN;
 473
 474         spin_unlock_bh(&xprt->transport_lock);
 475
 476         /* Race breaker in case memory is freed before above code is called */
 477         if (ret == -EAGAIN) {
 478                 struct socket_wq *wq;
 479
 480                 rcu_read_lock();
 481                 wq = rcu_dereference(sk->sk_wq);
 482                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 483                 rcu_read_unlock();
 484
 485                 sk->sk_write_space(sk);
 486         }
 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
 531         if (status == -EAGAIN && sock_writeable(transport->inet))
 532                 status = -ENOBUFS;
 533
 534         if (likely(sent > 0) || status == 0) {
 535                 req->rq_bytes_sent += sent;
 536                 req->rq_xmit_bytes_sent += sent;
 537                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 538                         req->rq_bytes_sent = 0;
 539                         return 0;
 540                 }
 541                 status = -EAGAIN;
 542         }
 543
 544         switch (status) {
 545         case -ENOBUFS:
 546                 break;
 547         case -EAGAIN:
 548                 status = xs_nospace(task);
 549                 break;
 550         default:
 551                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 552                         -status);
 553         case -EPIPE:
 554                 xs_close(xprt);
 555                 status = -ENOTCONN;
 556         }
 557
 558         return status;
 559 }
 560
 561 /**
 562  * xs_udp_send_request - write an RPC request to a UDP socket
 563  * @task: address of RPC task that manages the state of an RPC request
 564  *
 565  * Return values:
 566  *        0:    The request has been sent
 567  *   EAGAIN:    The socket was blocked, please call again later to
 568  *              complete the request
 569  * ENOTCONN:    Caller needs to invoke connect logic then call again
 570  *    other:    Some other error occurred, the request was not sent
 571  */
 572 static int xs_udp_send_request(struct rpc_task *task)
 573 {
 574         struct rpc_rqst *req = task->tk_rqstp;
 575         struct rpc_xprt *xprt = req->rq_xprt;
 576         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 577         struct xdr_buf *xdr = &req->rq_snd_buf;
 578         int sent = 0;
 579         int status;
 580
 581         xs_pktdump("packet data:",
 582                                 req->rq_svec->iov_base,
 583                                 req->rq_svec->iov_len);
 584
 585         if (!xprt_bound(xprt))
 586                 return -ENOTCONN;
 587         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 588                               xdr, req->rq_bytes_sent, true, &sent);
 589
 590         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 591                         xdr->len - req->rq_bytes_sent, status);
 592
 593         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 594         if (status == -EPERM)
 595                 goto process_status;
 596
 597         if (status == -EAGAIN && sock_writeable(transport->inet))
 598                 status = -ENOBUFS;
 599
 600         if (sent > 0 || status == 0) {
 601                 req->rq_xmit_bytes_sent += sent;
 602                 if (sent >= req->rq_slen)
 603                         return 0;
 604                 /* Still some bytes left; set up for a retry later. */
 605                 status = -EAGAIN;
 606         }
 607
 608 process_status:
 609         switch (status) {
 610         case -ENOTSOCK:
 611                 status = -ENOTCONN;
 612                 /* Should we call xs_close() here? */
 613                 break;
 614         case -EAGAIN:
 615                 status = xs_nospace(task);
 616                 break;
 617         case -ENETUNREACH:
 618         case -ENOBUFS:
 619         case -EPIPE:
 620         case -ECONNREFUSED:
 621         case -EPERM:
 622                 /* When the server has died, an ICMP port unreachable message
 623                  * prompts ECONNREFUSED. */
 624                 break;
 625         default:
 626                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 627                         -status);
 628         }
 629
 630         return status;
 631 }
 632
 633 /**
 634  * xs_tcp_send_request - write an RPC request to a TCP socket
 635  * @task: address of RPC task that manages the state of an RPC request
 636  *
 637  * Return values:
 638  *        0:    The request has been sent
 639  *   EAGAIN:    The socket was blocked, please call again later to
 640  *              complete the request
 641  * ENOTCONN:    Caller needs to invoke connect logic then call again
 642  *    other:    Some other error occurred, the request was not sent
 643  *
 644  * XXX: In the case of soft timeouts, should we eventually give up
 645  *      if sendmsg is not able to make progress?
 646  */
 647 static int xs_tcp_send_request(struct rpc_task *task)
 648 {
 649         struct rpc_rqst *req = task->tk_rqstp;
 650         struct rpc_xprt *xprt = req->rq_xprt;
 651         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 652         struct xdr_buf *xdr = &req->rq_snd_buf;
 653         bool zerocopy = true;
 654         int status;
 655         int sent;
 656
 657         xs_encode_stream_record_marker(&req->rq_snd_buf);
 658
 659         xs_pktdump("packet data:",
 660                                 req->rq_svec->iov_base,
 661                                 req->rq_svec->iov_len);
 662         /* Don't use zero copy if this is a resend. If the RPC call
 663          * completes while the socket holds a reference to the pages,
 664          * then we may end up resending corrupted data.
 665          */
 666         if (task->tk_flags & RPC_TASK_SENT)
 667                 zerocopy = false;
 668
 669         /* Continue transmitting the packet/record. We must be careful
 670          * to cope with writespace callbacks arriving _after_ we have
 671          * called sendmsg(). */
 672         while (1) {
 673                 sent = 0;
 674                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 675                                       req->rq_bytes_sent, zerocopy, &sent);
 676
 677                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 678                                 xdr->len - req->rq_bytes_sent, status);
 679
 680                 /* If we've sent the entire packet, immediately
 681                  * reset the count of bytes sent. */
 682                 req->rq_bytes_sent += sent;
 683                 req->rq_xmit_bytes_sent += sent;
 684                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 685                         req->rq_bytes_sent = 0;
 686                         return 0;
 687                 }
 688
 689                 if (status < 0)
 690                         break;
 691                 if (sent == 0) {
 692                         status = -EAGAIN;
 693                         break;
 694                 }
 695         }
 696         if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
 697                 status = -ENOBUFS;
 698
 699         switch (status) {
 700         case -ENOTSOCK:
 701                 status = -ENOTCONN;
 702                 /* Should we call xs_close() here? */
 703                 break;
 704         case -EAGAIN:
 705                 status = xs_nospace(task);
 706                 break;
 707         case -ECONNRESET:
 708         case -ECONNREFUSED:
 709         case -ENOTCONN:
 710         case -EADDRINUSE:
 711         case -ENOBUFS:
 712         case -EPIPE:
 713                 break;
 714         default:
 715                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 716                         -status);
 717         }
 718
 719         return status;
 720 }
 721
 722 /**
 723  * xs_tcp_release_xprt - clean up after a tcp transmission
 724  * @xprt: transport
 725  * @task: rpc task
 726  *
 727  * This cleans up if an error causes us to abort the transmission of a request.
 728  * In this case, the socket may need to be reset in order to avoid confusing
 729  * the server.
 730  */
 731 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 732 {
 733         struct rpc_rqst *req;
 734
 735         if (task != xprt->snd_task)
 736                 return;
 737         if (task == NULL)
 738                 goto out_release;
 739         req = task->tk_rqstp;
 740         if (req == NULL)
 741                 goto out_release;
 742         if (req->rq_bytes_sent == 0)
 743                 goto out_release;
 744         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 745                 goto out_release;
 746         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 747 out_release:
 748         xprt_release_xprt(xprt, task);
 749 }
 750
 751 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 752 {
 753         transport->old_data_ready = sk->sk_data_ready;
 754         transport->old_state_change = sk->sk_state_change;
 755         transport->old_write_space = sk->sk_write_space;
 756         transport->old_error_report = sk->sk_error_report;
 757 }
 758
 759 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 760 {
 761         sk->sk_data_ready = transport->old_data_ready;
 762         sk->sk_state_change = transport->old_state_change;
 763         sk->sk_write_space = transport->old_write_space;
 764         sk->sk_error_report = transport->old_error_report;
 765 }
 766
 767 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 768 {
 769         smp_mb__before_atomic();
 770         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 771         clear_bit(XPRT_CLOSING, &xprt->state);
 772         smp_mb__after_atomic();
 773 }
 774
 775 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 776 {
 777         xs_sock_reset_connection_flags(xprt);
 778         /* Mark transport as closed and wake up all pending tasks */
 779         xprt_disconnect_done(xprt);
 780 }
 781
 782 /**
 783  * xs_error_report - callback to handle TCP socket state errors
 784  * @sk: socket
 785  *
 786  * Note: we don't call sock_error() since there may be a rpc_task
 787  * using the socket, and so we don't want to clear sk->sk_err.
 788  */
 789 static void xs_error_report(struct sock *sk)
 790 {
 791         struct rpc_xprt *xprt;
 792         int err;
 793
 794         read_lock_bh(&sk->sk_callback_lock);
 795         if (!(xprt = xprt_from_sock(sk)))
 796                 goto out;
 797
 798         err = -sk->sk_err;
 799         if (err == 0)
 800                 goto out;
 801         /* Is this a reset event? */
 802         if (sk->sk_state == TCP_CLOSE)
 803                 xs_sock_mark_closed(xprt);
 804         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 805                         xprt, -err);
 806         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 807         xprt_wake_pending_tasks(xprt, err);
 808  out:
 809         read_unlock_bh(&sk->sk_callback_lock);
 810 }
 811
 812 static void xs_reset_transport(struct sock_xprt *transport)
 813 {
 814         struct socket *sock = transport->sock;
 815         struct sock *sk = transport->inet;
 816         struct rpc_xprt *xprt = &transport->xprt;
 817
 818         if (sk == NULL)
 819                 return;
 820
 821         if (atomic_read(&transport->xprt.swapper))
 822                 sk_clear_memalloc(sk);
 823
 824         kernel_sock_shutdown(sock, SHUT_RDWR);
 825
 826         mutex_lock(&transport->recv_mutex);
 827         write_lock_bh(&sk->sk_callback_lock);
 828         transport->inet = NULL;
 829         transport->sock = NULL;
 830
 831         sk->sk_user_data = NULL;
 832
 833         xs_restore_old_callbacks(transport, sk);
 834         xprt_clear_connected(xprt);
 835         write_unlock_bh(&sk->sk_callback_lock);
 836         xs_sock_reset_connection_flags(xprt);
 837         mutex_unlock(&transport->recv_mutex);
 838
 839         trace_rpc_socket_close(xprt, sock);
 840         sock_release(sock);
 841 }
 842
 843 /**
 844  * xs_close - close a socket
 845  * @xprt: transport
 846  *
 847  * This is used when all requests are complete; ie, no DRC state remains
 848  * on the server we want to save.
 849  *
 850  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 851  * xs_reset_transport() zeroing the socket from underneath a writer.
 852  */
 853 static void xs_close(struct rpc_xprt *xprt)
 854 {
 855         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 856
 857         dprintk("RPC:       xs_close xprt %p\n", xprt);
 858
 859         xs_reset_transport(transport);
 860         xprt->reestablish_timeout = 0;
 861
 862         xprt_disconnect_done(xprt);
 863 }
 864
 865 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 866 {
 867         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 868                 xprt);
 869         xprt_disconnect_done(xprt);
 870 }
 871
 872 static void xs_xprt_free(struct rpc_xprt *xprt)
 873 {
 874         xs_free_peer_addresses(xprt);
 875         xprt_free(xprt);
 876 }
 877
 878 /**
 879  * xs_destroy - prepare to shutdown a transport
 880  * @xprt: doomed transport
 881  *
 882  */
 883 static void xs_destroy(struct rpc_xprt *xprt)
 884 {
 885         struct sock_xprt *transport = container_of(xprt,
 886                         struct sock_xprt, xprt);
 887         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 888
 889         cancel_delayed_work_sync(&transport->connect_worker);
 890         xs_close(xprt);
 891         cancel_work_sync(&transport->recv_worker);
 892         xs_xprt_free(xprt);
 893         module_put(THIS_MODULE);
 894 }
 895
 896 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 897 {
 898         struct xdr_skb_reader desc = {
 899                 .skb            = skb,
 900                 .offset         = sizeof(rpc_fraghdr),
 901                 .count          = skb->len - sizeof(rpc_fraghdr),
 902         };
 903
 904         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 905                 return -1;
 906         if (desc.count)
 907                 return -1;
 908         return 0;
 909 }
 910
 911 /**
 912  * xs_local_data_read_skb
 913  * @xprt: transport
 914  * @sk: socket
 915  * @skb: skbuff
 916  *
 917  * Currently this assumes we can read the whole reply in a single gulp.
 918  */
 919 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 920                 struct sock *sk,
 921                 struct sk_buff *skb)
 922 {
 923         struct rpc_task *task;
 924         struct rpc_rqst *rovr;
 925         int repsize, copied;
 926         u32 _xid;
 927         __be32 *xp;
 928
 929         repsize = skb->len - sizeof(rpc_fraghdr);
 930         if (repsize < 4) {
 931                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 932                 return;
 933         }
 934
 935         /* Copy the XID from the skb... */
 936         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 937         if (xp == NULL)
 938                 return;
 939
 940         /* Look up and lock the request corresponding to the given XID */
 941         spin_lock_bh(&xprt->transport_lock);
 942         rovr = xprt_lookup_rqst(xprt, *xp);
 943         if (!rovr)
 944                 goto out_unlock;
 945         task = rovr->rq_task;
 946
 947         copied = rovr->rq_private_buf.buflen;
 948         if (copied > repsize)
 949                 copied = repsize;
 950
 951         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 952                 dprintk("RPC:       sk_buff copy failed\n");
 953                 goto out_unlock;
 954         }
 955
 956         xprt_complete_rqst(task, copied);
 957
 958  out_unlock:
 959         spin_unlock_bh(&xprt->transport_lock);
 960 }
 961
 962 static void xs_local_data_receive(struct sock_xprt *transport)
 963 {
 964         struct sk_buff *skb;
 965         struct sock *sk;
 966         int err;
 967
 968         mutex_lock(&transport->recv_mutex);
 969         sk = transport->inet;
 970         if (sk == NULL)
 971                 goto out;
 972         for (;;) {
 973                 skb = skb_recv_datagram(sk, 0, 1, &err);
 974                 if (skb == NULL)
 975                         break;
 976                 xs_local_data_read_skb(&transport->xprt, sk, skb);
 977                 skb_free_datagram(sk, skb);
 978         }
 979 out:
 980         mutex_unlock(&transport->recv_mutex);
 981 }
 982
 983 static void xs_local_data_receive_workfn(struct work_struct *work)
 984 {
 985         struct sock_xprt *transport =
 986                 container_of(work, struct sock_xprt, recv_worker);
 987         xs_local_data_receive(transport);
 988 }
 989
 990 /**
 991  * xs_udp_data_read_skb - receive callback for UDP sockets
 992  * @xprt: transport
 993  * @sk: socket
 994  * @skb: skbuff
 995  *
 996  */
 997 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
 998                 struct sock *sk,
 999                 struct sk_buff *skb)
1000 {
1001         struct rpc_task *task;
1002         struct rpc_rqst *rovr;
1003         int repsize, copied;
1004         u32 _xid;
1005         __be32 *xp;
1006
1007         repsize = skb->len - sizeof(struct udphdr);
1008         if (repsize < 4) {
1009                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1010                 return;
1011         }
1012
1013         /* Copy the XID from the skb... */
1014         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1015                                 sizeof(_xid), &_xid);
1016         if (xp == NULL)
1017                 return;
1018
1019         /* Look up and lock the request corresponding to the given XID */
1020         spin_lock_bh(&xprt->transport_lock);
1021         rovr = xprt_lookup_rqst(xprt, *xp);
1022         if (!rovr)
1023                 goto out_unlock;
1024         task = rovr->rq_task;
1025
1026         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1027                 copied = repsize;
1028
1029         /* Suck it into the iovec, verify checksum if not done by hw. */
1030         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1031                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1032                 goto out_unlock;
1033         }
1034
1035         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1036
1037         xprt_adjust_cwnd(xprt, task, copied);
1038         xprt_complete_rqst(task, copied);
1039
1040  out_unlock:
1041         spin_unlock_bh(&xprt->transport_lock);
1042 }
1043
1044 static void xs_udp_data_receive(struct sock_xprt *transport)
1045 {
1046         struct sk_buff *skb;
1047         struct sock *sk;
1048         int err;
1049
1050         mutex_lock(&transport->recv_mutex);
1051         sk = transport->inet;
1052         if (sk == NULL)
1053                 goto out;
1054         for (;;) {
1055                 skb = skb_recv_datagram(sk, 0, 1, &err);
1056                 if (skb == NULL)
1057                         break;
1058                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1059                 skb_free_datagram(sk, skb);
1060         }
1061 out:
1062         mutex_unlock(&transport->recv_mutex);
1063 }
1064
1065 static void xs_udp_data_receive_workfn(struct work_struct *work)
1066 {
1067         struct sock_xprt *transport =
1068                 container_of(work, struct sock_xprt, recv_worker);
1069         xs_udp_data_receive(transport);
1070 }
1071
1072 /**
1073  * xs_data_ready - "data ready" callback for UDP sockets
1074  * @sk: socket with data to read
1075  *
1076  */
1077 static void xs_data_ready(struct sock *sk)
1078 {
1079         struct rpc_xprt *xprt;
1080
1081         read_lock_bh(&sk->sk_callback_lock);
1082         dprintk("RPC:       xs_data_ready...\n");
1083         xprt = xprt_from_sock(sk);
1084         if (xprt != NULL) {
1085                 struct sock_xprt *transport = container_of(xprt,
1086                                 struct sock_xprt, xprt);
1087                 queue_work(rpciod_workqueue, &transport->recv_worker);
1088         }
1089         read_unlock_bh(&sk->sk_callback_lock);
1090 }
1091
1092 /*
1093  * Helper function to force a TCP close if the server is sending
1094  * junk and/or it has put us in CLOSE_WAIT
1095  */
1096 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1097 {
1098         xprt_force_disconnect(xprt);
1099 }
1100
1101 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1102 {
1103         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1104         size_t len, used;
1105         char *p;
1106
1107         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1108         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1109         used = xdr_skb_read_bits(desc, p, len);
1110         transport->tcp_offset += used;
1111         if (used != len)
1112                 return;
1113
1114         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1115         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1116                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1117         else
1118                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1119         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1120
1121         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1122         transport->tcp_offset = 0;
1123
1124         /* Sanity check of the record length */
1125         if (unlikely(transport->tcp_reclen < 8)) {
1126                 dprintk("RPC:       invalid TCP record fragment length\n");
1127                 xs_tcp_force_close(xprt);
1128                 return;
1129         }
1130         dprintk("RPC:       reading TCP record fragment of length %d\n",
1131                         transport->tcp_reclen);
1132 }
1133
1134 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1135 {
1136         if (transport->tcp_offset == transport->tcp_reclen) {
1137                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1138                 transport->tcp_offset = 0;
1139                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1140                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1141                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1142                         transport->tcp_copied = 0;
1143                 }
1144         }
1145 }
1146
1147 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1148 {
1149         size_t len, used;
1150         char *p;
1151
1152         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1153         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1154         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1155         used = xdr_skb_read_bits(desc, p, len);
1156         transport->tcp_offset += used;
1157         if (used != len)
1158                 return;
1159         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1160         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1161         transport->tcp_copied = 4;
1162         dprintk("RPC:       reading %s XID %08x\n",
1163                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1164                                                               : "request with",
1165                         ntohl(transport->tcp_xid));
1166         xs_tcp_check_fraghdr(transport);
1167 }
1168
1169 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1170                                        struct xdr_skb_reader *desc)
1171 {
1172         size_t len, used;
1173         u32 offset;
1174         char *p;
1175
1176         /*
1177          * We want transport->tcp_offset to be 8 at the end of this routine
1178          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1179          * When this function is called for the first time,
1180          * transport->tcp_offset is 4 (after having already read the xid).
1181          */
1182         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1183         len = sizeof(transport->tcp_calldir) - offset;
1184         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1185         p = ((char *) &transport->tcp_calldir) + offset;
1186         used = xdr_skb_read_bits(desc, p, len);
1187         transport->tcp_offset += used;
1188         if (used != len)
1189                 return;
1190         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1191         /*
1192          * We don't yet have the XDR buffer, so we will write the calldir
1193          * out after we get the buffer from the 'struct rpc_rqst'
1194          */
1195         switch (ntohl(transport->tcp_calldir)) {
1196         case RPC_REPLY:
1197                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1198                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1199                 transport->tcp_flags |= TCP_RPC_REPLY;
1200                 break;
1201         case RPC_CALL:
1202                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1203                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1204                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1205                 break;
1206         default:
1207                 dprintk("RPC:       invalid request message type\n");
1208                 xs_tcp_force_close(&transport->xprt);
1209         }
1210         xs_tcp_check_fraghdr(transport);
1211 }
1212
1213 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1214                                      struct xdr_skb_reader *desc,
1215                                      struct rpc_rqst *req)
1216 {
1217         struct sock_xprt *transport =
1218                                 container_of(xprt, struct sock_xprt, xprt);
1219         struct xdr_buf *rcvbuf;
1220         size_t len;
1221         ssize_t r;
1222
1223         rcvbuf = &req->rq_private_buf;
1224
1225         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1226                 /*
1227                  * Save the RPC direction in the XDR buffer
1228                  */
1229                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1230                         &transport->tcp_calldir,
1231                         sizeof(transport->tcp_calldir));
1232                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1233                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1234         }
1235
1236         len = desc->count;
1237         if (len > transport->tcp_reclen - transport->tcp_offset) {
1238                 struct xdr_skb_reader my_desc;
1239
1240                 len = transport->tcp_reclen - transport->tcp_offset;
1241                 memcpy(&my_desc, desc, sizeof(my_desc));
1242                 my_desc.count = len;
1243                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1244                                           &my_desc, xdr_skb_read_bits);
1245                 desc->count -= r;
1246                 desc->offset += r;
1247         } else
1248                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1249                                           desc, xdr_skb_read_bits);
1250
1251         if (r > 0) {
1252                 transport->tcp_copied += r;
1253                 transport->tcp_offset += r;
1254         }
1255         if (r != len) {
1256                 /* Error when copying to the receive buffer,
1257                  * usually because we weren't able to allocate
1258                  * additional buffer pages. All we can do now
1259                  * is turn off TCP_RCV_COPY_DATA, so the request
1260                  * will not receive any additional updates,
1261                  * and time out.
1262                  * Any remaining data from this record will
1263                  * be discarded.
1264                  */
1265                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1266                 dprintk("RPC:       XID %08x truncated request\n",
1267                                 ntohl(transport->tcp_xid));
1268                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1269                                 "tcp_offset = %u, tcp_reclen = %u\n",
1270                                 xprt, transport->tcp_copied,
1271                                 transport->tcp_offset, transport->tcp_reclen);
1272                 return;
1273         }
1274
1275         dprintk("RPC:       XID %08x read %Zd bytes\n",
1276                         ntohl(transport->tcp_xid), r);
1277         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1278                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1279                         transport->tcp_offset, transport->tcp_reclen);
1280
1281         if (transport->tcp_copied == req->rq_private_buf.buflen)
1282                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1283         else if (transport->tcp_offset == transport->tcp_reclen) {
1284                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1285                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1286         }
1287 }
1288
1289 /*
1290  * Finds the request corresponding to the RPC xid and invokes the common
1291  * tcp read code to read the data.
1292  */
1293 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1294                                     struct xdr_skb_reader *desc)
1295 {
1296         struct sock_xprt *transport =
1297                                 container_of(xprt, struct sock_xprt, xprt);
1298         struct rpc_rqst *req;
1299
1300         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1301
1302         /* Find and lock the request corresponding to this xid */
1303         spin_lock_bh(&xprt->transport_lock);
1304         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1305         if (!req) {
1306                 dprintk("RPC:       XID %08x request not found!\n",
1307                                 ntohl(transport->tcp_xid));
1308                 spin_unlock_bh(&xprt->transport_lock);
1309                 return -1;
1310         }
1311
1312         xs_tcp_read_common(xprt, desc, req);
1313
1314         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1315                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1316
1317         spin_unlock_bh(&xprt->transport_lock);
1318         return 0;
1319 }
1320
1321 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1322 /*
1323  * Obtains an rpc_rqst previously allocated and invokes the common
1324  * tcp read code to read the data.  The result is placed in the callback
1325  * queue.
1326  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1327  * connection and return -1.
1328  */
1329 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1330                                        struct xdr_skb_reader *desc)
1331 {
1332         struct sock_xprt *transport =
1333                                 container_of(xprt, struct sock_xprt, xprt);
1334         struct rpc_rqst *req;
1335
1336         /* Look up and lock the request corresponding to the given XID */
1337         spin_lock_bh(&xprt->transport_lock);
1338         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1339         if (req == NULL) {
1340                 spin_unlock_bh(&xprt->transport_lock);
1341                 printk(KERN_WARNING "Callback slot table overflowed\n");
1342                 xprt_force_disconnect(xprt);
1343                 return -1;
1344         }
1345
1346         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1347         xs_tcp_read_common(xprt, desc, req);
1348
1349         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1350                 xprt_complete_bc_request(req, transport->tcp_copied);
1351         spin_unlock_bh(&xprt->transport_lock);
1352
1353         return 0;
1354 }
1355
1356 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1357                                         struct xdr_skb_reader *desc)
1358 {
1359         struct sock_xprt *transport =
1360                                 container_of(xprt, struct sock_xprt, xprt);
1361
1362         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1363                 xs_tcp_read_reply(xprt, desc) :
1364                 xs_tcp_read_callback(xprt, desc);
1365 }
1366
1367 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1368 {
1369         int ret;
1370
1371         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1372                               SVC_SOCK_ANONYMOUS);
1373         if (ret < 0)
1374                 return ret;
1375         return 0;
1376 }
1377 #else
1378 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1379                                         struct xdr_skb_reader *desc)
1380 {
1381         return xs_tcp_read_reply(xprt, desc);
1382 }
1383 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1384
1385 /*
1386  * Read data off the transport.  This can be either an RPC_CALL or an
1387  * RPC_REPLY.  Relay the processing to helper functions.
1388  */
1389 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1390                                     struct xdr_skb_reader *desc)
1391 {
1392         struct sock_xprt *transport =
1393                                 container_of(xprt, struct sock_xprt, xprt);
1394
1395         if (_xs_tcp_read_data(xprt, desc) == 0)
1396                 xs_tcp_check_fraghdr(transport);
1397         else {
1398                 /*
1399                  * The transport_lock protects the request handling.
1400                  * There's no need to hold it to update the tcp_flags.
1401                  */
1402                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1403         }
1404 }
1405
1406 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1407 {
1408         size_t len;
1409
1410         len = transport->tcp_reclen - transport->tcp_offset;
1411         if (len > desc->count)
1412                 len = desc->count;
1413         desc->count -= len;
1414         desc->offset += len;
1415         transport->tcp_offset += len;
1416         dprintk("RPC:       discarded %Zu bytes\n", len);
1417         xs_tcp_check_fraghdr(transport);
1418 }
1419
1420 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1421 {
1422         struct rpc_xprt *xprt = rd_desc->arg.data;
1423         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1424         struct xdr_skb_reader desc = {
1425                 .skb    = skb,
1426                 .offset = offset,
1427                 .count  = len,
1428         };
1429
1430         dprintk("RPC:       xs_tcp_data_recv started\n");
1431         do {
1432                 trace_xs_tcp_data_recv(transport);
1433                 /* Read in a new fragment marker if necessary */
1434                 /* Can we ever really expect to get completely empty fragments? */
1435                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1436                         xs_tcp_read_fraghdr(xprt, &desc);
1437                         continue;
1438                 }
1439                 /* Read in the xid if necessary */
1440                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1441                         xs_tcp_read_xid(transport, &desc);
1442                         continue;
1443                 }
1444                 /* Read in the call/reply flag */
1445                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1446                         xs_tcp_read_calldir(transport, &desc);
1447                         continue;
1448                 }
1449                 /* Read in the request data */
1450                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1451                         xs_tcp_read_data(xprt, &desc);
1452                         continue;
1453                 }
1454                 /* Skip over any trailing bytes on short reads */
1455                 xs_tcp_read_discard(transport, &desc);
1456         } while (desc.count);
1457         trace_xs_tcp_data_recv(transport);
1458         dprintk("RPC:       xs_tcp_data_recv done\n");
1459         return len - desc.count;
1460 }
1461
1462 static void xs_tcp_data_receive(struct sock_xprt *transport)
1463 {
1464         struct rpc_xprt *xprt = &transport->xprt;
1465         struct sock *sk;
1466         read_descriptor_t rd_desc = {
1467                 .count = 2*1024*1024,
1468                 .arg.data = xprt,
1469         };
1470         unsigned long total = 0;
1471         int read = 0;
1472
1473         mutex_lock(&transport->recv_mutex);
1474         sk = transport->inet;
1475         if (sk == NULL)
1476                 goto out;
1477
1478         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1479         for (;;) {
1480                 lock_sock(sk);
1481                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1482                 release_sock(sk);
1483                 if (read <= 0)
1484                         break;
1485                 total += read;
1486                 rd_desc.count = 65536;
1487         }
1488 out:
1489         mutex_unlock(&transport->recv_mutex);
1490         trace_xs_tcp_data_ready(xprt, read, total);
1491 }
1492
1493 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1494 {
1495         struct sock_xprt *transport =
1496                 container_of(work, struct sock_xprt, recv_worker);
1497         xs_tcp_data_receive(transport);
1498 }
1499
1500 /**
1501  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1502  * @sk: socket with data to read
1503  *
1504  */
1505 static void xs_tcp_data_ready(struct sock *sk)
1506 {
1507         struct sock_xprt *transport;
1508         struct rpc_xprt *xprt;
1509
1510         dprintk("RPC:       xs_tcp_data_ready...\n");
1511
1512         read_lock_bh(&sk->sk_callback_lock);
1513         if (!(xprt = xprt_from_sock(sk)))
1514                 goto out;
1515         transport = container_of(xprt, struct sock_xprt, xprt);
1516
1517         /* Any data means we had a useful conversation, so
1518          * the we don't need to delay the next reconnect
1519          */
1520         if (xprt->reestablish_timeout)
1521                 xprt->reestablish_timeout = 0;
1522         queue_work(rpciod_workqueue, &transport->recv_worker);
1523
1524 out:
1525         read_unlock_bh(&sk->sk_callback_lock);
1526 }
1527
1528 /**
1529  * xs_tcp_state_change - callback to handle TCP socket state changes
1530  * @sk: socket whose state has changed
1531  *
1532  */
1533 static void xs_tcp_state_change(struct sock *sk)
1534 {
1535         struct rpc_xprt *xprt;
1536         struct sock_xprt *transport;
1537
1538         read_lock_bh(&sk->sk_callback_lock);
1539         if (!(xprt = xprt_from_sock(sk)))
1540                 goto out;
1541         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1542         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1543                         sk->sk_state, xprt_connected(xprt),
1544                         sock_flag(sk, SOCK_DEAD),
1545                         sock_flag(sk, SOCK_ZAPPED),
1546                         sk->sk_shutdown);
1547
1548         transport = container_of(xprt, struct sock_xprt, xprt);
1549         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1550         switch (sk->sk_state) {
1551         case TCP_ESTABLISHED:
1552                 spin_lock(&xprt->transport_lock);
1553                 if (!xprt_test_and_set_connected(xprt)) {
1554
1555                         /* Reset TCP record info */
1556                         transport->tcp_offset = 0;
1557                         transport->tcp_reclen = 0;
1558                         transport->tcp_copied = 0;
1559                         transport->tcp_flags =
1560                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1561                         xprt->connect_cookie++;
1562                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1563                         xprt_clear_connecting(xprt);
1564
1565                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1566                 }
1567                 spin_unlock(&xprt->transport_lock);
1568                 break;
1569         case TCP_FIN_WAIT1:
1570                 /* The client initiated a shutdown of the socket */
1571                 xprt->connect_cookie++;
1572                 xprt->reestablish_timeout = 0;
1573                 set_bit(XPRT_CLOSING, &xprt->state);
1574                 smp_mb__before_atomic();
1575                 clear_bit(XPRT_CONNECTED, &xprt->state);
1576                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1577                 smp_mb__after_atomic();
1578                 break;
1579         case TCP_CLOSE_WAIT:
1580                 /* The server initiated a shutdown of the socket */
1581                 xprt->connect_cookie++;
1582                 clear_bit(XPRT_CONNECTED, &xprt->state);
1583                 xs_tcp_force_close(xprt);
1584         case TCP_CLOSING:
1585                 /*
1586                  * If the server closed down the connection, make sure that
1587                  * we back off before reconnecting
1588                  */
1589                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1590                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1591                 break;
1592         case TCP_LAST_ACK:
1593                 set_bit(XPRT_CLOSING, &xprt->state);
1594                 smp_mb__before_atomic();
1595                 clear_bit(XPRT_CONNECTED, &xprt->state);
1596                 smp_mb__after_atomic();
1597                 break;
1598         case TCP_CLOSE:
1599                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1600                                         &transport->sock_state))
1601                         xprt_clear_connecting(xprt);
1602                 xs_sock_mark_closed(xprt);
1603         }
1604  out:
1605         read_unlock_bh(&sk->sk_callback_lock);
1606 }
1607
1608 static void xs_write_space(struct sock *sk)
1609 {
1610         struct socket_wq *wq;
1611         struct rpc_xprt *xprt;
1612
1613         if (!sk->sk_socket)
1614                 return;
1615         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1616
1617         if (unlikely(!(xprt = xprt_from_sock(sk))))
1618                 return;
1619         rcu_read_lock();
1620         wq = rcu_dereference(sk->sk_wq);
1621         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1622                 goto out;
1623
1624         xprt_write_space(xprt);
1625 out:
1626         rcu_read_unlock();
1627 }
1628
1629 /**
1630  * xs_udp_write_space - callback invoked when socket buffer space
1631  *                             becomes available
1632  * @sk: socket whose state has changed
1633  *
1634  * Called when more output buffer space is available for this socket.
1635  * We try not to wake our writers until they can make "significant"
1636  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1637  * with a bunch of small requests.
1638  */
1639 static void xs_udp_write_space(struct sock *sk)
1640 {
1641         read_lock_bh(&sk->sk_callback_lock);
1642
1643         /* from net/core/sock.c:sock_def_write_space */
1644         if (sock_writeable(sk))
1645                 xs_write_space(sk);
1646
1647         read_unlock_bh(&sk->sk_callback_lock);
1648 }
1649
1650 /**
1651  * xs_tcp_write_space - callback invoked when socket buffer space
1652  *                             becomes available
1653  * @sk: socket whose state has changed
1654  *
1655  * Called when more output buffer space is available for this socket.
1656  * We try not to wake our writers until they can make "significant"
1657  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1658  * with a bunch of small requests.
1659  */
1660 static void xs_tcp_write_space(struct sock *sk)
1661 {
1662         read_lock_bh(&sk->sk_callback_lock);
1663
1664         /* from net/core/stream.c:sk_stream_write_space */
1665         if (sk_stream_is_writeable(sk))
1666                 xs_write_space(sk);
1667
1668         read_unlock_bh(&sk->sk_callback_lock);
1669 }
1670
1671 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1672 {
1673         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1674         struct sock *sk = transport->inet;
1675
1676         if (transport->rcvsize) {
1677                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1678                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1679         }
1680         if (transport->sndsize) {
1681                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1682                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1683                 sk->sk_write_space(sk);
1684         }
1685 }
1686
1687 /**
1688  * xs_udp_set_buffer_size - set send and receive limits
1689  * @xprt: generic transport
1690  * @sndsize: requested size of send buffer, in bytes
1691  * @rcvsize: requested size of receive buffer, in bytes
1692  *
1693  * Set socket send and receive buffer size limits.
1694  */
1695 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1696 {
1697         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1698
1699         transport->sndsize = 0;
1700         if (sndsize)
1701                 transport->sndsize = sndsize + 1024;
1702         transport->rcvsize = 0;
1703         if (rcvsize)
1704                 transport->rcvsize = rcvsize + 1024;
1705
1706         xs_udp_do_set_buffer_size(xprt);
1707 }
1708
1709 /**
1710  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1711  * @task: task that timed out
1712  *
1713  * Adjust the congestion window after a retransmit timeout has occurred.
1714  */
1715 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1716 {
1717         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1718 }
1719
1720 static unsigned short xs_get_random_port(void)
1721 {
1722         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1723         unsigned short rand = (unsigned short) prandom_u32() % range;
1724         return rand + xprt_min_resvport;
1725 }
1726
1727 /**
1728  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1729  * @sock: socket
1730  *
1731  * Note that this function has to be called on all sockets that share the
1732  * same port, and it must be called before binding.
1733  */
1734 static void xs_sock_set_reuseport(struct socket *sock)
1735 {
1736         int opt = 1;
1737
1738         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1739                         (char *)&opt, sizeof(opt));
1740 }
1741
1742 static unsigned short xs_sock_getport(struct socket *sock)
1743 {
1744         struct sockaddr_storage buf;
1745         int buflen;
1746         unsigned short port = 0;
1747
1748         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1749                 goto out;
1750         switch (buf.ss_family) {
1751         case AF_INET6:
1752                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1753                 break;
1754         case AF_INET:
1755                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1756         }
1757 out:
1758         return port;
1759 }
1760
1761 /**
1762  * xs_set_port - reset the port number in the remote endpoint address
1763  * @xprt: generic transport
1764  * @port: new port number
1765  *
1766  */
1767 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1768 {
1769         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1770
1771         rpc_set_port(xs_addr(xprt), port);
1772         xs_update_peer_port(xprt);
1773 }
1774
1775 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1776 {
1777         if (transport->srcport == 0)
1778                 transport->srcport = xs_sock_getport(sock);
1779 }
1780
1781 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1782 {
1783         unsigned short port = transport->srcport;
1784
1785         if (port == 0 && transport->xprt.resvport)
1786                 port = xs_get_random_port();
1787         return port;
1788 }
1789
1790 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1791 {
1792         if (transport->srcport != 0)
1793                 transport->srcport = 0;
1794         if (!transport->xprt.resvport)
1795                 return 0;
1796         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1797                 return xprt_max_resvport;
1798         return --port;
1799 }
1800 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1801 {
1802         struct sockaddr_storage myaddr;
1803         int err, nloop = 0;
1804         unsigned short port = xs_get_srcport(transport);
1805         unsigned short last;
1806
1807         /*
1808          * If we are asking for any ephemeral port (i.e. port == 0 &&
1809          * transport->xprt.resvport == 0), don't bind.  Let the local
1810          * port selection happen implicitly when the socket is used
1811          * (for example at connect time).
1812          *
1813          * This ensures that we can continue to establish TCP
1814          * connections even when all local ephemeral ports are already
1815          * a part of some TCP connection.  This makes no difference
1816          * for UDP sockets, but also doens't harm them.
1817          *
1818          * If we're asking for any reserved port (i.e. port == 0 &&
1819          * transport->xprt.resvport == 1) xs_get_srcport above will
1820          * ensure that port is non-zero and we will bind as needed.
1821          */
1822         if (port == 0)
1823                 return 0;
1824
1825         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1826         do {
1827                 rpc_set_port((struct sockaddr *)&myaddr, port);
1828                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1829                                 transport->xprt.addrlen);
1830                 if (err == 0) {
1831                         transport->srcport = port;
1832                         break;
1833                 }
1834                 last = port;
1835                 port = xs_next_srcport(transport, port);
1836                 if (port > last)
1837                         nloop++;
1838         } while (err == -EADDRINUSE && nloop != 2);
1839
1840         if (myaddr.ss_family == AF_INET)
1841                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1842                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1843                                 port, err ? "failed" : "ok", err);
1844         else
1845                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1846                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1847                                 port, err ? "failed" : "ok", err);
1848         return err;
1849 }
1850
1851 /*
1852  * We don't support autobind on AF_LOCAL sockets
1853  */
1854 static void xs_local_rpcbind(struct rpc_task *task)
1855 {
1856         rcu_read_lock();
1857         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1858         rcu_read_unlock();
1859 }
1860
1861 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1862 {
1863 }
1864
1865 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1866 static struct lock_class_key xs_key[2];
1867 static struct lock_class_key xs_slock_key[2];
1868
1869 static inline void xs_reclassify_socketu(struct socket *sock)
1870 {
1871         struct sock *sk = sock->sk;
1872
1873         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1874                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1875 }
1876
1877 static inline void xs_reclassify_socket4(struct socket *sock)
1878 {
1879         struct sock *sk = sock->sk;
1880
1881         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1882                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1883 }
1884
1885 static inline void xs_reclassify_socket6(struct socket *sock)
1886 {
1887         struct sock *sk = sock->sk;
1888
1889         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1890                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1891 }
1892
1893 static inline void xs_reclassify_socket(int family, struct socket *sock)
1894 {
1895         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1896         if (sock_owned_by_user(sock->sk))
1897                 return;
1898
1899         switch (family) {
1900         case AF_LOCAL:
1901                 xs_reclassify_socketu(sock);
1902                 break;
1903         case AF_INET:
1904                 xs_reclassify_socket4(sock);
1905                 break;
1906         case AF_INET6:
1907                 xs_reclassify_socket6(sock);
1908                 break;
1909         }
1910 }
1911 #else
1912 static inline void xs_reclassify_socketu(struct socket *sock)
1913 {
1914 }
1915
1916 static inline void xs_reclassify_socket4(struct socket *sock)
1917 {
1918 }
1919
1920 static inline void xs_reclassify_socket6(struct socket *sock)
1921 {
1922 }
1923
1924 static inline void xs_reclassify_socket(int family, struct socket *sock)
1925 {
1926 }
1927 #endif
1928
1929 static void xs_dummy_setup_socket(struct work_struct *work)
1930 {
1931 }
1932
1933 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1934                 struct sock_xprt *transport, int family, int type,
1935                 int protocol, bool reuseport)
1936 {
1937         struct socket *sock;
1938         int err;
1939
1940         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1941         if (err < 0) {
1942                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1943                                 protocol, -err);
1944                 goto out;
1945         }
1946         xs_reclassify_socket(family, sock);
1947
1948         if (reuseport)
1949                 xs_sock_set_reuseport(sock);
1950
1951         err = xs_bind(transport, sock);
1952         if (err) {
1953                 sock_release(sock);
1954                 goto out;
1955         }
1956
1957         return sock;
1958 out:
1959         return ERR_PTR(err);
1960 }
1961
1962 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1963                                       struct socket *sock)
1964 {
1965         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1966                                                                         xprt);
1967
1968         if (!transport->inet) {
1969                 struct sock *sk = sock->sk;
1970
1971                 write_lock_bh(&sk->sk_callback_lock);
1972
1973                 xs_save_old_callbacks(transport, sk);
1974
1975                 sk->sk_user_data = xprt;
1976                 sk->sk_data_ready = xs_data_ready;
1977                 sk->sk_write_space = xs_udp_write_space;
1978                 sk->sk_error_report = xs_error_report;
1979                 sk->sk_allocation = GFP_NOIO;
1980
1981                 xprt_clear_connected(xprt);
1982
1983                 /* Reset to new socket */
1984                 transport->sock = sock;
1985                 transport->inet = sk;
1986
1987                 write_unlock_bh(&sk->sk_callback_lock);
1988         }
1989
1990         /* Tell the socket layer to start connecting... */
1991         xprt->stat.connect_count++;
1992         xprt->stat.connect_start = jiffies;
1993         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1994 }
1995
1996 /**
1997  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1998  * @transport: socket transport to connect
1999  */
2000 static int xs_local_setup_socket(struct sock_xprt *transport)
2001 {
2002         struct rpc_xprt *xprt = &transport->xprt;
2003         struct socket *sock;
2004         int status = -EIO;
2005
2006         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2007                                         SOCK_STREAM, 0, &sock, 1);
2008         if (status < 0) {
2009                 dprintk("RPC:       can't create AF_LOCAL "
2010                         "transport socket (%d).\n", -status);
2011                 goto out;
2012         }
2013         xs_reclassify_socketu(sock);
2014
2015         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2016                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2017
2018         status = xs_local_finish_connecting(xprt, sock);
2019         trace_rpc_socket_connect(xprt, sock, status);
2020         switch (status) {
2021         case 0:
2022                 dprintk("RPC:       xprt %p connected to %s\n",
2023                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2024                 xprt_set_connected(xprt);
2025         case -ENOBUFS:
2026                 break;
2027         case -ENOENT:
2028                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2029                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2030                 break;
2031         case -ECONNREFUSED:
2032                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2033                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2034                 break;
2035         default:
2036                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2037                                 __func__, -status,
2038                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2039         }
2040
2041 out:
2042         xprt_clear_connecting(xprt);
2043         xprt_wake_pending_tasks(xprt, status);
2044         return status;
2045 }
2046
2047 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2048 {
2049         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2050         int ret;
2051
2052          if (RPC_IS_ASYNC(task)) {
2053                 /*
2054                  * We want the AF_LOCAL connect to be resolved in the
2055                  * filesystem namespace of the process making the rpc
2056                  * call.  Thus we connect synchronously.
2057                  *
2058                  * If we want to support asynchronous AF_LOCAL calls,
2059                  * we'll need to figure out how to pass a namespace to
2060                  * connect.
2061                  */
2062                 rpc_exit(task, -ENOTCONN);
2063                 return;
2064         }
2065         ret = xs_local_setup_socket(transport);
2066         if (ret && !RPC_IS_SOFTCONN(task))
2067                 msleep_interruptible(15000);
2068 }
2069
2070 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2071 /*
2072  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2073  * know that we have exclusive access to the socket), to guard against
2074  * races with xs_reset_transport.
2075  */
2076 static void xs_set_memalloc(struct rpc_xprt *xprt)
2077 {
2078         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2079                         xprt);
2080
2081         /*
2082          * If there's no sock, then we have nothing to set. The
2083          * reconnecting process will get it for us.
2084          */
2085         if (!transport->inet)
2086                 return;
2087         if (atomic_read(&xprt->swapper))
2088                 sk_set_memalloc(transport->inet);
2089 }
2090
2091 /**
2092  * xs_enable_swap - Tag this transport as being used for swap.
2093  * @xprt: transport to tag
2094  *
2095  * Take a reference to this transport on behalf of the rpc_clnt, and
2096  * optionally mark it for swapping if it wasn't already.
2097  */
2098 static int
2099 xs_enable_swap(struct rpc_xprt *xprt)
2100 {
2101         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2102
2103         if (atomic_inc_return(&xprt->swapper) != 1)
2104                 return 0;
2105         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2106                 return -ERESTARTSYS;
2107         if (xs->inet)
2108                 sk_set_memalloc(xs->inet);
2109         xprt_release_xprt(xprt, NULL);
2110         return 0;
2111 }
2112
2113 /**
2114  * xs_disable_swap - Untag this transport as being used for swap.
2115  * @xprt: transport to tag
2116  *
2117  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2118  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2119  */
2120 static void
2121 xs_disable_swap(struct rpc_xprt *xprt)
2122 {
2123         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2124
2125         if (!atomic_dec_and_test(&xprt->swapper))
2126                 return;
2127         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2128                 return;
2129         if (xs->inet)
2130                 sk_clear_memalloc(xs->inet);
2131         xprt_release_xprt(xprt, NULL);
2132 }
2133 #else
2134 static void xs_set_memalloc(struct rpc_xprt *xprt)
2135 {
2136 }
2137
2138 static int
2139 xs_enable_swap(struct rpc_xprt *xprt)
2140 {
2141         return -EINVAL;
2142 }
2143
2144 static void
2145 xs_disable_swap(struct rpc_xprt *xprt)
2146 {
2147 }
2148 #endif
2149
2150 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2151 {
2152         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2153
2154         if (!transport->inet) {
2155                 struct sock *sk = sock->sk;
2156
2157                 write_lock_bh(&sk->sk_callback_lock);
2158
2159                 xs_save_old_callbacks(transport, sk);
2160
2161                 sk->sk_user_data = xprt;
2162                 sk->sk_data_ready = xs_data_ready;
2163                 sk->sk_write_space = xs_udp_write_space;
2164                 sk->sk_allocation = GFP_NOIO;
2165
2166                 xprt_set_connected(xprt);
2167
2168                 /* Reset to new socket */
2169                 transport->sock = sock;
2170                 transport->inet = sk;
2171
2172                 xs_set_memalloc(xprt);
2173
2174                 write_unlock_bh(&sk->sk_callback_lock);
2175         }
2176         xs_udp_do_set_buffer_size(xprt);
2177 }
2178
2179 static void xs_udp_setup_socket(struct work_struct *work)
2180 {
2181         struct sock_xprt *transport =
2182                 container_of(work, struct sock_xprt, connect_worker.work);
2183         struct rpc_xprt *xprt = &transport->xprt;
2184         struct socket *sock = transport->sock;
2185         int status = -EIO;
2186
2187         sock = xs_create_sock(xprt, transport,
2188                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2189                         IPPROTO_UDP, false);
2190         if (IS_ERR(sock))
2191                 goto out;
2192
2193         dprintk("RPC:       worker connecting xprt %p via %s to "
2194                                 "%s (port %s)\n", xprt,
2195                         xprt->address_strings[RPC_DISPLAY_PROTO],
2196                         xprt->address_strings[RPC_DISPLAY_ADDR],
2197                         xprt->address_strings[RPC_DISPLAY_PORT]);
2198
2199         xs_udp_finish_connecting(xprt, sock);
2200         trace_rpc_socket_connect(xprt, sock, 0);
2201         status = 0;
2202 out:
2203         xprt_unlock_connect(xprt, transport);
2204         xprt_clear_connecting(xprt);
2205         xprt_wake_pending_tasks(xprt, status);
2206 }
2207
2208 /**
2209  * xs_tcp_shutdown - gracefully shut down a TCP socket
2210  * @xprt: transport
2211  *
2212  * Initiates a graceful shutdown of the TCP socket by calling the
2213  * equivalent of shutdown(SHUT_RDWR);
2214  */
2215 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2216 {
2217         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2218         struct socket *sock = transport->sock;
2219
2220         if (sock == NULL)
2221                 return;
2222         if (xprt_connected(xprt)) {
2223                 kernel_sock_shutdown(sock, SHUT_RDWR);
2224                 trace_rpc_socket_shutdown(xprt, sock);
2225         } else
2226                 xs_reset_transport(transport);
2227 }
2228
2229 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2230 {
2231         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2232         int ret = -ENOTCONN;
2233
2234         if (!transport->inet) {
2235                 struct sock *sk = sock->sk;
2236                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2237                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2238                 unsigned int opt_on = 1;
2239                 unsigned int timeo;
2240
2241                 /* TCP Keepalive options */
2242                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2243                                 (char *)&opt_on, sizeof(opt_on));
2244                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2245                                 (char *)&keepidle, sizeof(keepidle));
2246                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2247                                 (char *)&keepidle, sizeof(keepidle));
2248                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2249                                 (char *)&keepcnt, sizeof(keepcnt));
2250
2251                 /* TCP user timeout (see RFC5482) */
2252                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2253                         (xprt->timeout->to_retries + 1);
2254                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2255                                 (char *)&timeo, sizeof(timeo));
2256
2257                 write_lock_bh(&sk->sk_callback_lock);
2258
2259                 xs_save_old_callbacks(transport, sk);
2260
2261                 sk->sk_user_data = xprt;
2262                 sk->sk_data_ready = xs_tcp_data_ready;
2263                 sk->sk_state_change = xs_tcp_state_change;
2264                 sk->sk_write_space = xs_tcp_write_space;
2265                 sk->sk_error_report = xs_error_report;
2266                 sk->sk_allocation = GFP_NOIO;
2267
2268                 /* socket options */
2269                 sock_reset_flag(sk, SOCK_LINGER);
2270                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2271
2272                 xprt_clear_connected(xprt);
2273
2274                 /* Reset to new socket */
2275                 transport->sock = sock;
2276                 transport->inet = sk;
2277
2278                 write_unlock_bh(&sk->sk_callback_lock);
2279         }
2280
2281         if (!xprt_bound(xprt))
2282                 goto out;
2283
2284         xs_set_memalloc(xprt);
2285
2286         /* Tell the socket layer to start connecting... */
2287         xprt->stat.connect_count++;
2288         xprt->stat.connect_start = jiffies;
2289         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2290         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2291         switch (ret) {
2292         case 0:
2293                 xs_set_srcport(transport, sock);
2294         case -EINPROGRESS:
2295                 /* SYN_SENT! */
2296                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2297                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2298                 break;
2299         case -EADDRNOTAVAIL:
2300                 /* Source port number is unavailable. Try a new one! */
2301                 transport->srcport = 0;
2302         }
2303 out:
2304         return ret;
2305 }
2306
2307 /**
2308  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2309  *
2310  * Invoked by a work queue tasklet.
2311  */
2312 static void xs_tcp_setup_socket(struct work_struct *work)
2313 {
2314         struct sock_xprt *transport =
2315                 container_of(work, struct sock_xprt, connect_worker.work);
2316         struct socket *sock = transport->sock;
2317         struct rpc_xprt *xprt = &transport->xprt;
2318         int status = -EIO;
2319
2320         if (!sock) {
2321                 sock = xs_create_sock(xprt, transport,
2322                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2323                                 IPPROTO_TCP, true);
2324                 if (IS_ERR(sock)) {
2325                         status = PTR_ERR(sock);
2326                         goto out;
2327                 }
2328         }
2329
2330         dprintk("RPC:       worker connecting xprt %p via %s to "
2331                                 "%s (port %s)\n", xprt,
2332                         xprt->address_strings[RPC_DISPLAY_PROTO],
2333                         xprt->address_strings[RPC_DISPLAY_ADDR],
2334                         xprt->address_strings[RPC_DISPLAY_PORT]);
2335
2336         status = xs_tcp_finish_connecting(xprt, sock);
2337         trace_rpc_socket_connect(xprt, sock, status);
2338         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2339                         xprt, -status, xprt_connected(xprt),
2340                         sock->sk->sk_state);
2341         switch (status) {
2342         default:
2343                 printk("%s: connect returned unhandled error %d\n",
2344                         __func__, status);
2345         case -EADDRNOTAVAIL:
2346                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2347                  * and retry
2348                  */
2349                 xs_tcp_force_close(xprt);
2350                 break;
2351         case 0:
2352         case -EINPROGRESS:
2353         case -EALREADY:
2354                 xprt_unlock_connect(xprt, transport);
2355                 return;
2356         case -EINVAL:
2357                 /* Happens, for instance, if the user specified a link
2358                  * local IPv6 address without a scope-id.
2359                  */
2360         case -ECONNREFUSED:
2361         case -ECONNRESET:
2362         case -ENETUNREACH:
2363         case -EADDRINUSE:
2364         case -ENOBUFS:
2365                 /* retry with existing socket, after a delay */
2366                 xs_tcp_force_close(xprt);
2367                 goto out;
2368         }
2369         status = -EAGAIN;
2370 out:
2371         xprt_unlock_connect(xprt, transport);
2372         xprt_clear_connecting(xprt);
2373         xprt_wake_pending_tasks(xprt, status);
2374 }
2375
2376 /**
2377  * xs_connect - connect a socket to a remote endpoint
2378  * @xprt: pointer to transport structure
2379  * @task: address of RPC task that manages state of connect request
2380  *
2381  * TCP: If the remote end dropped the connection, delay reconnecting.
2382  *
2383  * UDP socket connects are synchronous, but we use a work queue anyway
2384  * to guarantee that even unprivileged user processes can set up a
2385  * socket on a privileged port.
2386  *
2387  * If a UDP socket connect fails, the delay behavior here prevents
2388  * retry floods (hard mounts).
2389  */
2390 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2391 {
2392         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2393
2394         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2395
2396         if (transport->sock != NULL) {
2397                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2398                                 "seconds\n",
2399                                 xprt, xprt->reestablish_timeout / HZ);
2400
2401                 /* Start by resetting any existing state */
2402                 xs_reset_transport(transport);
2403
2404                 queue_delayed_work(rpciod_workqueue,
2405                                    &transport->connect_worker,
2406                                    xprt->reestablish_timeout);
2407                 xprt->reestablish_timeout <<= 1;
2408                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2409                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2410                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2411                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2412         } else {
2413                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2414                 queue_delayed_work(rpciod_workqueue,
2415                                    &transport->connect_worker, 0);
2416         }
2417 }
2418
2419 /**
2420  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2421  * @xprt: rpc_xprt struct containing statistics
2422  * @seq: output file
2423  *
2424  */
2425 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2426 {
2427         long idle_time = 0;
2428
2429         if (xprt_connected(xprt))
2430                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2431
2432         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2433                         "%llu %llu %lu %llu %llu\n",
2434                         xprt->stat.bind_count,
2435                         xprt->stat.connect_count,
2436                         xprt->stat.connect_time,
2437                         idle_time,
2438                         xprt->stat.sends,
2439                         xprt->stat.recvs,
2440                         xprt->stat.bad_xids,
2441                         xprt->stat.req_u,
2442                         xprt->stat.bklog_u,
2443                         xprt->stat.max_slots,
2444                         xprt->stat.sending_u,
2445                         xprt->stat.pending_u);
2446 }
2447
2448 /**
2449  * xs_udp_print_stats - display UDP socket-specifc stats
2450  * @xprt: rpc_xprt struct containing statistics
2451  * @seq: output file
2452  *
2453  */
2454 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2455 {
2456         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2457
2458         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2459                         "%lu %llu %llu\n",
2460                         transport->srcport,
2461                         xprt->stat.bind_count,
2462                         xprt->stat.sends,
2463                         xprt->stat.recvs,
2464                         xprt->stat.bad_xids,
2465                         xprt->stat.req_u,
2466                         xprt->stat.bklog_u,
2467                         xprt->stat.max_slots,
2468                         xprt->stat.sending_u,
2469                         xprt->stat.pending_u);
2470 }
2471
2472 /**
2473  * xs_tcp_print_stats - display TCP socket-specifc stats
2474  * @xprt: rpc_xprt struct containing statistics
2475  * @seq: output file
2476  *
2477  */
2478 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2479 {
2480         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2481         long idle_time = 0;
2482
2483         if (xprt_connected(xprt))
2484                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2485
2486         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2487                         "%llu %llu %lu %llu %llu\n",
2488                         transport->srcport,
2489                         xprt->stat.bind_count,
2490                         xprt->stat.connect_count,
2491                         xprt->stat.connect_time,
2492                         idle_time,
2493                         xprt->stat.sends,
2494                         xprt->stat.recvs,
2495                         xprt->stat.bad_xids,
2496                         xprt->stat.req_u,
2497                         xprt->stat.bklog_u,
2498                         xprt->stat.max_slots,
2499                         xprt->stat.sending_u,
2500                         xprt->stat.pending_u);
2501 }
2502
2503 /*
2504  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2505  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2506  * to use the server side send routines.
2507  */
2508 static void *bc_malloc(struct rpc_task *task, size_t size)
2509 {
2510         struct page *page;
2511         struct rpc_buffer *buf;
2512
2513         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2514         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2515                 return NULL;
2516
2517         page = alloc_page(GFP_KERNEL);
2518         if (!page)
2519                 return NULL;
2520
2521         buf = page_address(page);
2522         buf->len = PAGE_SIZE;
2523
2524         return buf->data;
2525 }
2526
2527 /*
2528  * Free the space allocated in the bc_alloc routine
2529  */
2530 static void bc_free(void *buffer)
2531 {
2532         struct rpc_buffer *buf;
2533
2534         if (!buffer)
2535                 return;
2536
2537         buf = container_of(buffer, struct rpc_buffer, data);
2538         free_page((unsigned long)buf);
2539 }
2540
2541 /*
2542  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2543  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2544  */
2545 static int bc_sendto(struct rpc_rqst *req)
2546 {
2547         int len;
2548         struct xdr_buf *xbufp = &req->rq_snd_buf;
2549         struct rpc_xprt *xprt = req->rq_xprt;
2550         struct sock_xprt *transport =
2551                                 container_of(xprt, struct sock_xprt, xprt);
2552         struct socket *sock = transport->sock;
2553         unsigned long headoff;
2554         unsigned long tailoff;
2555
2556         xs_encode_stream_record_marker(xbufp);
2557
2558         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2559         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2560         len = svc_send_common(sock, xbufp,
2561                               virt_to_page(xbufp->head[0].iov_base), headoff,
2562                               xbufp->tail[0].iov_base, tailoff);
2563
2564         if (len != xbufp->len) {
2565                 printk(KERN_NOTICE "Error sending entire callback!\n");
2566                 len = -EAGAIN;
2567         }
2568
2569         return len;
2570 }
2571
2572 /*
2573  * The send routine. Borrows from svc_send
2574  */
2575 static int bc_send_request(struct rpc_task *task)
2576 {
2577         struct rpc_rqst *req = task->tk_rqstp;
2578         struct svc_xprt *xprt;
2579         int len;
2580
2581         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2582         /*
2583          * Get the server socket associated with this callback xprt
2584          */
2585         xprt = req->rq_xprt->bc_xprt;
2586
2587         /*
2588          * Grab the mutex to serialize data as the connection is shared
2589          * with the fore channel
2590          */
2591         if (!mutex_trylock(&xprt->xpt_mutex)) {
2592                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2593                 if (!mutex_trylock(&xprt->xpt_mutex))
2594                         return -EAGAIN;
2595                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2596         }
2597         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2598                 len = -ENOTCONN;
2599         else
2600                 len = bc_sendto(req);
2601         mutex_unlock(&xprt->xpt_mutex);
2602
2603         if (len > 0)
2604                 len = 0;
2605
2606         return len;
2607 }
2608
2609 /*
2610  * The close routine. Since this is client initiated, we do nothing
2611  */
2612
2613 static void bc_close(struct rpc_xprt *xprt)
2614 {
2615 }
2616
2617 /*
2618  * The xprt destroy routine. Again, because this connection is client
2619  * initiated, we do nothing
2620  */
2621
2622 static void bc_destroy(struct rpc_xprt *xprt)
2623 {
2624         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2625
2626         xs_xprt_free(xprt);
2627         module_put(THIS_MODULE);
2628 }
2629
2630 static struct rpc_xprt_ops xs_local_ops = {
2631         .reserve_xprt           = xprt_reserve_xprt,
2632         .release_xprt           = xs_tcp_release_xprt,
2633         .alloc_slot             = xprt_alloc_slot,
2634         .rpcbind                = xs_local_rpcbind,
2635         .set_port               = xs_local_set_port,
2636         .connect                = xs_local_connect,
2637         .buf_alloc              = rpc_malloc,
2638         .buf_free               = rpc_free,
2639         .send_request           = xs_local_send_request,
2640         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2641         .close                  = xs_close,
2642         .destroy                = xs_destroy,
2643         .print_stats            = xs_local_print_stats,
2644         .enable_swap            = xs_enable_swap,
2645         .disable_swap           = xs_disable_swap,
2646 };
2647
2648 static struct rpc_xprt_ops xs_udp_ops = {
2649         .set_buffer_size        = xs_udp_set_buffer_size,
2650         .reserve_xprt           = xprt_reserve_xprt_cong,
2651         .release_xprt           = xprt_release_xprt_cong,
2652         .alloc_slot             = xprt_alloc_slot,
2653         .rpcbind                = rpcb_getport_async,
2654         .set_port               = xs_set_port,
2655         .connect                = xs_connect,
2656         .buf_alloc              = rpc_malloc,
2657         .buf_free               = rpc_free,
2658         .send_request           = xs_udp_send_request,
2659         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2660         .timer                  = xs_udp_timer,
2661         .release_request        = xprt_release_rqst_cong,
2662         .close                  = xs_close,
2663         .destroy                = xs_destroy,
2664         .print_stats            = xs_udp_print_stats,
2665         .enable_swap            = xs_enable_swap,
2666         .disable_swap           = xs_disable_swap,
2667         .inject_disconnect      = xs_inject_disconnect,
2668 };
2669
2670 static struct rpc_xprt_ops xs_tcp_ops = {
2671         .reserve_xprt           = xprt_reserve_xprt,
2672         .release_xprt           = xs_tcp_release_xprt,
2673         .alloc_slot             = xprt_lock_and_alloc_slot,
2674         .rpcbind                = rpcb_getport_async,
2675         .set_port               = xs_set_port,
2676         .connect                = xs_connect,
2677         .buf_alloc              = rpc_malloc,
2678         .buf_free               = rpc_free,
2679         .send_request           = xs_tcp_send_request,
2680         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2681         .close                  = xs_tcp_shutdown,
2682         .destroy                = xs_destroy,
2683         .print_stats            = xs_tcp_print_stats,
2684         .enable_swap            = xs_enable_swap,
2685         .disable_swap           = xs_disable_swap,
2686         .inject_disconnect      = xs_inject_disconnect,
2687 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2688         .bc_setup               = xprt_setup_bc,
2689         .bc_up                  = xs_tcp_bc_up,
2690         .bc_free_rqst           = xprt_free_bc_rqst,
2691         .bc_destroy             = xprt_destroy_bc,
2692 #endif
2693 };
2694
2695 /*
2696  * The rpc_xprt_ops for the server backchannel
2697  */
2698
2699 static struct rpc_xprt_ops bc_tcp_ops = {
2700         .reserve_xprt           = xprt_reserve_xprt,
2701         .release_xprt           = xprt_release_xprt,
2702         .alloc_slot             = xprt_alloc_slot,
2703         .buf_alloc              = bc_malloc,
2704         .buf_free               = bc_free,
2705         .send_request           = bc_send_request,
2706         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2707         .close                  = bc_close,
2708         .destroy                = bc_destroy,
2709         .print_stats            = xs_tcp_print_stats,
2710         .enable_swap            = xs_enable_swap,
2711         .disable_swap           = xs_disable_swap,
2712         .inject_disconnect      = xs_inject_disconnect,
2713 };
2714
2715 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2716 {
2717         static const struct sockaddr_in sin = {
2718                 .sin_family             = AF_INET,
2719                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2720         };
2721         static const struct sockaddr_in6 sin6 = {
2722                 .sin6_family            = AF_INET6,
2723                 .sin6_addr              = IN6ADDR_ANY_INIT,
2724         };
2725
2726         switch (family) {
2727         case AF_LOCAL:
2728                 break;
2729         case AF_INET:
2730                 memcpy(sap, &sin, sizeof(sin));
2731                 break;
2732         case AF_INET6:
2733                 memcpy(sap, &sin6, sizeof(sin6));
2734                 break;
2735         default:
2736                 dprintk("RPC:       %s: Bad address family\n", __func__);
2737                 return -EAFNOSUPPORT;
2738         }
2739         return 0;
2740 }
2741
2742 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2743                                       unsigned int slot_table_size,
2744                                       unsigned int max_slot_table_size)
2745 {
2746         struct rpc_xprt *xprt;
2747         struct sock_xprt *new;
2748
2749         if (args->addrlen > sizeof(xprt->addr)) {
2750                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2751                 return ERR_PTR(-EBADF);
2752         }
2753
2754         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2755                         max_slot_table_size);
2756         if (xprt == NULL) {
2757                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2758                                 "rpc_xprt\n");
2759                 return ERR_PTR(-ENOMEM);
2760         }
2761
2762         new = container_of(xprt, struct sock_xprt, xprt);
2763         mutex_init(&new->recv_mutex);
2764         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2765         xprt->addrlen = args->addrlen;
2766         if (args->srcaddr)
2767                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2768         else {
2769                 int err;
2770                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2771                                         (struct sockaddr *)&new->srcaddr);
2772                 if (err != 0) {
2773                         xprt_free(xprt);
2774                         return ERR_PTR(err);
2775                 }
2776         }
2777
2778         return xprt;
2779 }
2780
2781 static const struct rpc_timeout xs_local_default_timeout = {
2782         .to_initval = 10 * HZ,
2783         .to_maxval = 10 * HZ,
2784         .to_retries = 2,
2785 };
2786
2787 /**
2788  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2789  * @args: rpc transport creation arguments
2790  *
2791  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2792  */
2793 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2794 {
2795         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2796         struct sock_xprt *transport;
2797         struct rpc_xprt *xprt;
2798         struct rpc_xprt *ret;
2799
2800         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2801                         xprt_max_tcp_slot_table_entries);
2802         if (IS_ERR(xprt))
2803                 return xprt;
2804         transport = container_of(xprt, struct sock_xprt, xprt);
2805
2806         xprt->prot = 0;
2807         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2808         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2809
2810         xprt->bind_timeout = XS_BIND_TO;
2811         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2812         xprt->idle_timeout = XS_IDLE_DISC_TO;
2813
2814         xprt->ops = &xs_local_ops;
2815         xprt->timeout = &xs_local_default_timeout;
2816
2817         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2818         INIT_DELAYED_WORK(&transport->connect_worker,
2819                         xs_dummy_setup_socket);
2820
2821         switch (sun->sun_family) {
2822         case AF_LOCAL:
2823                 if (sun->sun_path[0] != '/') {
2824                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2825                                         sun->sun_path);
2826                         ret = ERR_PTR(-EINVAL);
2827                         goto out_err;
2828                 }
2829                 xprt_set_bound(xprt);
2830                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2831                 ret = ERR_PTR(xs_local_setup_socket(transport));
2832                 if (ret)
2833                         goto out_err;
2834                 break;
2835         default:
2836                 ret = ERR_PTR(-EAFNOSUPPORT);
2837                 goto out_err;
2838         }
2839
2840         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2841                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2842
2843         if (try_module_get(THIS_MODULE))
2844                 return xprt;
2845         ret = ERR_PTR(-EINVAL);
2846 out_err:
2847         xs_xprt_free(xprt);
2848         return ret;
2849 }
2850
2851 static const struct rpc_timeout xs_udp_default_timeout = {
2852         .to_initval = 5 * HZ,
2853         .to_maxval = 30 * HZ,
2854         .to_increment = 5 * HZ,
2855         .to_retries = 5,
2856 };
2857
2858 /**
2859  * xs_setup_udp - Set up transport to use a UDP socket
2860  * @args: rpc transport creation arguments
2861  *
2862  */
2863 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2864 {
2865         struct sockaddr *addr = args->dstaddr;
2866         struct rpc_xprt *xprt;
2867         struct sock_xprt *transport;
2868         struct rpc_xprt *ret;
2869
2870         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2871                         xprt_udp_slot_table_entries);
2872         if (IS_ERR(xprt))
2873                 return xprt;
2874         transport = container_of(xprt, struct sock_xprt, xprt);
2875
2876         xprt->prot = IPPROTO_UDP;
2877         xprt->tsh_size = 0;
2878         /* XXX: header size can vary due to auth type, IPv6, etc. */
2879         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2880
2881         xprt->bind_timeout = XS_BIND_TO;
2882         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2883         xprt->idle_timeout = XS_IDLE_DISC_TO;
2884
2885         xprt->ops = &xs_udp_ops;
2886
2887         xprt->timeout = &xs_udp_default_timeout;
2888
2889         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2890         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2891
2892         switch (addr->sa_family) {
2893         case AF_INET:
2894                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2895                         xprt_set_bound(xprt);
2896
2897                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2898                 break;
2899         case AF_INET6:
2900                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2901                         xprt_set_bound(xprt);
2902
2903                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2904                 break;
2905         default:
2906                 ret = ERR_PTR(-EAFNOSUPPORT);
2907                 goto out_err;
2908         }
2909
2910         if (xprt_bound(xprt))
2911                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2912                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2913                                 xprt->address_strings[RPC_DISPLAY_PORT],
2914                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2915         else
2916                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2917                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2918                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2919
2920         if (try_module_get(THIS_MODULE))
2921                 return xprt;
2922         ret = ERR_PTR(-EINVAL);
2923 out_err:
2924         xs_xprt_free(xprt);
2925         return ret;
2926 }
2927
2928 static const struct rpc_timeout xs_tcp_default_timeout = {
2929         .to_initval = 60 * HZ,
2930         .to_maxval = 60 * HZ,
2931         .to_retries = 2,
2932 };
2933
2934 /**
2935  * xs_setup_tcp - Set up transport to use a TCP socket
2936  * @args: rpc transport creation arguments
2937  *
2938  */
2939 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2940 {
2941         struct sockaddr *addr = args->dstaddr;
2942         struct rpc_xprt *xprt;
2943         struct sock_xprt *transport;
2944         struct rpc_xprt *ret;
2945         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2946
2947         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2948                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2949
2950         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2951                         max_slot_table_size);
2952         if (IS_ERR(xprt))
2953                 return xprt;
2954         transport = container_of(xprt, struct sock_xprt, xprt);
2955
2956         xprt->prot = IPPROTO_TCP;
2957         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2958         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2959
2960         xprt->bind_timeout = XS_BIND_TO;
2961         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2962         xprt->idle_timeout = XS_IDLE_DISC_TO;
2963
2964         xprt->ops = &xs_tcp_ops;
2965         xprt->timeout = &xs_tcp_default_timeout;
2966
2967         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2968         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2969
2970         switch (addr->sa_family) {
2971         case AF_INET:
2972                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2973                         xprt_set_bound(xprt);
2974
2975                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2976                 break;
2977         case AF_INET6:
2978                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2979                         xprt_set_bound(xprt);
2980
2981                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2982                 break;
2983         default:
2984                 ret = ERR_PTR(-EAFNOSUPPORT);
2985                 goto out_err;
2986         }
2987
2988         if (xprt_bound(xprt))
2989                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2990                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2991                                 xprt->address_strings[RPC_DISPLAY_PORT],
2992                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2993         else
2994                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2995                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2996                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2997
2998         if (try_module_get(THIS_MODULE))
2999                 return xprt;
3000         ret = ERR_PTR(-EINVAL);
3001 out_err:
3002         xs_xprt_free(xprt);
3003         return ret;
3004 }
3005
3006 /**
3007  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3008  * @args: rpc transport creation arguments
3009  *
3010  */
3011 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3012 {
3013         struct sockaddr *addr = args->dstaddr;
3014         struct rpc_xprt *xprt;
3015         struct sock_xprt *transport;
3016         struct svc_sock *bc_sock;
3017         struct rpc_xprt *ret;
3018
3019         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3020                         xprt_tcp_slot_table_entries);
3021         if (IS_ERR(xprt))
3022                 return xprt;
3023         transport = container_of(xprt, struct sock_xprt, xprt);
3024
3025         xprt->prot = IPPROTO_TCP;
3026         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3027         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3028         xprt->timeout = &xs_tcp_default_timeout;
3029
3030         /* backchannel */
3031         xprt_set_bound(xprt);
3032         xprt->bind_timeout = 0;
3033         xprt->reestablish_timeout = 0;
3034         xprt->idle_timeout = 0;
3035
3036         xprt->ops = &bc_tcp_ops;
3037
3038         switch (addr->sa_family) {
3039         case AF_INET:
3040                 xs_format_peer_addresses(xprt, "tcp",
3041                                          RPCBIND_NETID_TCP);
3042                 break;
3043         case AF_INET6:
3044                 xs_format_peer_addresses(xprt, "tcp",
3045                                    RPCBIND_NETID_TCP6);
3046                 break;
3047         default:
3048                 ret = ERR_PTR(-EAFNOSUPPORT);
3049                 goto out_err;
3050         }
3051
3052         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3053                         xprt->address_strings[RPC_DISPLAY_ADDR],
3054                         xprt->address_strings[RPC_DISPLAY_PORT],
3055                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3056
3057         /*
3058          * Once we've associated a backchannel xprt with a connection,
3059          * we want to keep it around as long as the connection lasts,
3060          * in case we need to start using it for a backchannel again;
3061          * this reference won't be dropped until bc_xprt is destroyed.
3062          */
3063         xprt_get(xprt);
3064         args->bc_xprt->xpt_bc_xprt = xprt;
3065         xprt->bc_xprt = args->bc_xprt;
3066         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3067         transport->sock = bc_sock->sk_sock;
3068         transport->inet = bc_sock->sk_sk;
3069
3070         /*
3071          * Since we don't want connections for the backchannel, we set
3072          * the xprt status to connected
3073          */
3074         xprt_set_connected(xprt);
3075
3076         if (try_module_get(THIS_MODULE))
3077                 return xprt;
3078
3079         args->bc_xprt->xpt_bc_xprt = NULL;
3080         xprt_put(xprt);
3081         ret = ERR_PTR(-EINVAL);
3082 out_err:
3083         xs_xprt_free(xprt);
3084         return ret;
3085 }
3086
3087 static struct xprt_class        xs_local_transport = {
3088         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3089         .name           = "named UNIX socket",
3090         .owner          = THIS_MODULE,
3091         .ident          = XPRT_TRANSPORT_LOCAL,
3092         .setup          = xs_setup_local,
3093 };
3094
3095 static struct xprt_class        xs_udp_transport = {
3096         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3097         .name           = "udp",
3098         .owner          = THIS_MODULE,
3099         .ident          = XPRT_TRANSPORT_UDP,
3100         .setup          = xs_setup_udp,
3101 };
3102
3103 static struct xprt_class        xs_tcp_transport = {
3104         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3105         .name           = "tcp",
3106         .owner          = THIS_MODULE,
3107         .ident          = XPRT_TRANSPORT_TCP,
3108         .setup          = xs_setup_tcp,
3109 };
3110
3111 static struct xprt_class        xs_bc_tcp_transport = {
3112         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3113         .name           = "tcp NFSv4.1 backchannel",
3114         .owner          = THIS_MODULE,
3115         .ident          = XPRT_TRANSPORT_BC_TCP,
3116         .setup          = xs_setup_bc_tcp,
3117 };
3118
3119 /**
3120  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3121  *
3122  */
3123 int init_socket_xprt(void)
3124 {
3125 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3126         if (!sunrpc_table_header)
3127                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3128 #endif
3129
3130         xprt_register_transport(&xs_local_transport);
3131         xprt_register_transport(&xs_udp_transport);
3132         xprt_register_transport(&xs_tcp_transport);
3133         xprt_register_transport(&xs_bc_tcp_transport);
3134
3135         return 0;
3136 }
3137
3138 /**
3139  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3140  *
3141  */
3142 void cleanup_socket_xprt(void)
3143 {
3144 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3145         if (sunrpc_table_header) {
3146                 unregister_sysctl_table(sunrpc_table_header);
3147                 sunrpc_table_header = NULL;
3148         }
3149 #endif
3150
3151         xprt_unregister_transport(&xs_local_transport);
3152         xprt_unregister_transport(&xs_udp_transport);
3153         xprt_unregister_transport(&xs_tcp_transport);
3154         xprt_unregister_transport(&xs_bc_tcp_transport);
3155 }
3156
3157 static int param_set_uint_minmax(const char *val,
3158                 const struct kernel_param *kp,
3159                 unsigned int min, unsigned int max)
3160 {
3161         unsigned int num;
3162         int ret;
3163
3164         if (!val)
3165                 return -EINVAL;
3166         ret = kstrtouint(val, 0, &num);
3167         if (ret == -EINVAL || num < min || num > max)
3168                 return -EINVAL;
3169         *((unsigned int *)kp->arg) = num;
3170         return 0;
3171 }
3172
3173 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3174 {
3175         return param_set_uint_minmax(val, kp,
3176                         RPC_MIN_RESVPORT,
3177                         RPC_MAX_RESVPORT);
3178 }
3179
3180 static const struct kernel_param_ops param_ops_portnr = {
3181         .set = param_set_portnr,
3182         .get = param_get_uint,
3183 };
3184
3185 #define param_check_portnr(name, p) \
3186         __param_check(name, p, unsigned int);
3187
3188 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3189 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3190
3191 static int param_set_slot_table_size(const char *val,
3192                                      const struct kernel_param *kp)
3193 {
3194         return param_set_uint_minmax(val, kp,
3195                         RPC_MIN_SLOT_TABLE,
3196                         RPC_MAX_SLOT_TABLE);
3197 }
3198
3199 static const struct kernel_param_ops param_ops_slot_table_size = {
3200         .set = param_set_slot_table_size,
3201         .get = param_get_uint,
3202 };
3203
3204 #define param_check_slot_table_size(name, p) \
3205         __param_check(name, p, unsigned int);
3206
3207 static int param_set_max_slot_table_size(const char *val,
3208                                      const struct kernel_param *kp)
3209 {
3210         return param_set_uint_minmax(val, kp,
3211                         RPC_MIN_SLOT_TABLE,
3212                         RPC_MAX_SLOT_TABLE_LIMIT);
3213 }
3214
3215 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3216         .set = param_set_max_slot_table_size,
3217         .get = param_get_uint,
3218 };
3219
3220 #define param_check_max_slot_table_size(name, p) \
3221         __param_check(name, p, unsigned int);
3222
3223 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3224                    slot_table_size, 0644);
3225 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3226                    max_slot_table_size, 0644);
3227 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3228                    slot_table_size, 0644);
3229