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
 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,
 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         sk->sk_write_space(sk);
 478         return ret;
 479 }
 480
 481 /*
 482  * Construct a stream transport record marker in @buf.
 483  */
 484 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 485 {
 486         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 487         rpc_fraghdr *base = buf->head[0].iov_base;
 488         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 489 }
 490
 491 /**
 492  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 493  * @task: RPC task that manages the state of an RPC request
 494  *
 495  * Return values:
 496  *        0:    The request has been sent
 497  *   EAGAIN:    The socket was blocked, please call again later to
 498  *              complete the request
 499  * ENOTCONN:    Caller needs to invoke connect logic then call again
 500  *    other:    Some other error occured, the request was not sent
 501  */
 502 static int xs_local_send_request(struct rpc_task *task)
 503 {
 504         struct rpc_rqst *req = task->tk_rqstp;
 505         struct rpc_xprt *xprt = req->rq_xprt;
 506         struct sock_xprt *transport =
 507                                 container_of(xprt, struct sock_xprt, xprt);
 508         struct xdr_buf *xdr = &req->rq_snd_buf;
 509         int status;
 510         int sent = 0;
 511
 512         xs_encode_stream_record_marker(&req->rq_snd_buf);
 513
 514         xs_pktdump("packet data:",
 515                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 516
 517         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 518                               true, &sent);
 519         dprintk("RPC:       %s(%u) = %d\n",
 520                         __func__, xdr->len - req->rq_bytes_sent, status);
 521
 522         if (status == -EAGAIN && sock_writeable(transport->inet))
 523                 status = -ENOBUFS;
 524
 525         if (likely(sent > 0) || status == 0) {
 526                 req->rq_bytes_sent += sent;
 527                 req->rq_xmit_bytes_sent += sent;
 528                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 529                         req->rq_bytes_sent = 0;
 530                         return 0;
 531                 }
 532                 status = -EAGAIN;
 533         }
 534
 535         switch (status) {
 536         case -ENOBUFS:
 537                 break;
 538         case -EAGAIN:
 539                 status = xs_nospace(task);
 540                 break;
 541         default:
 542                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 543                         -status);
 544         case -EPIPE:
 545                 xs_close(xprt);
 546                 status = -ENOTCONN;
 547         }
 548
 549         return status;
 550 }
 551
 552 /**
 553  * xs_udp_send_request - write an RPC request to a UDP socket
 554  * @task: address of RPC task that manages the state of an RPC request
 555  *
 556  * Return values:
 557  *        0:    The request has been sent
 558  *   EAGAIN:    The socket was blocked, please call again later to
 559  *              complete the request
 560  * ENOTCONN:    Caller needs to invoke connect logic then call again
 561  *    other:    Some other error occurred, the request was not sent
 562  */
 563 static int xs_udp_send_request(struct rpc_task *task)
 564 {
 565         struct rpc_rqst *req = task->tk_rqstp;
 566         struct rpc_xprt *xprt = req->rq_xprt;
 567         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 568         struct xdr_buf *xdr = &req->rq_snd_buf;
 569         int sent = 0;
 570         int status;
 571
 572         xs_pktdump("packet data:",
 573                                 req->rq_svec->iov_base,
 574                                 req->rq_svec->iov_len);
 575
 576         if (!xprt_bound(xprt))
 577                 return -ENOTCONN;
 578         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 579                               xdr, req->rq_bytes_sent, true, &sent);
 580
 581         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 582                         xdr->len - req->rq_bytes_sent, status);
 583
 584         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 585         if (status == -EPERM)
 586                 goto process_status;
 587
 588         if (status == -EAGAIN && sock_writeable(transport->inet))
 589                 status = -ENOBUFS;
 590
 591         if (sent > 0 || status == 0) {
 592                 req->rq_xmit_bytes_sent += sent;
 593                 if (sent >= req->rq_slen)
 594                         return 0;
 595                 /* Still some bytes left; set up for a retry later. */
 596                 status = -EAGAIN;
 597         }
 598
 599 process_status:
 600         switch (status) {
 601         case -ENOTSOCK:
 602                 status = -ENOTCONN;
 603                 /* Should we call xs_close() here? */
 604                 break;
 605         case -EAGAIN:
 606                 status = xs_nospace(task);
 607                 break;
 608         case -ENETUNREACH:
 609         case -ENOBUFS:
 610         case -EPIPE:
 611         case -ECONNREFUSED:
 612         case -EPERM:
 613                 /* When the server has died, an ICMP port unreachable message
 614                  * prompts ECONNREFUSED. */
 615                 break;
 616         default:
 617                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 618                         -status);
 619         }
 620
 621         return status;
 622 }
 623
 624 /**
 625  * xs_tcp_send_request - write an RPC request to a TCP socket
 626  * @task: address of RPC task that manages the state of an RPC request
 627  *
 628  * Return values:
 629  *        0:    The request has been sent
 630  *   EAGAIN:    The socket was blocked, please call again later to
 631  *              complete the request
 632  * ENOTCONN:    Caller needs to invoke connect logic then call again
 633  *    other:    Some other error occurred, the request was not sent
 634  *
 635  * XXX: In the case of soft timeouts, should we eventually give up
 636  *      if sendmsg is not able to make progress?
 637  */
 638 static int xs_tcp_send_request(struct rpc_task *task)
 639 {
 640         struct rpc_rqst *req = task->tk_rqstp;
 641         struct rpc_xprt *xprt = req->rq_xprt;
 642         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 643         struct xdr_buf *xdr = &req->rq_snd_buf;
 644         bool zerocopy = true;
 645         bool vm_wait = false;
 646         int status;
 647         int sent;
 648
 649         xs_encode_stream_record_marker(&req->rq_snd_buf);
 650
 651         xs_pktdump("packet data:",
 652                                 req->rq_svec->iov_base,
 653                                 req->rq_svec->iov_len);
 654         /* Don't use zero copy if this is a resend. If the RPC call
 655          * completes while the socket holds a reference to the pages,
 656          * then we may end up resending corrupted data.
 657          */
 658         if (task->tk_flags & RPC_TASK_SENT)
 659                 zerocopy = false;
 660
 661         /* Continue transmitting the packet/record. We must be careful
 662          * to cope with writespace callbacks arriving _after_ we have
 663          * called sendmsg(). */
 664         while (1) {
 665                 sent = 0;
 666                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 667                                       req->rq_bytes_sent, zerocopy, &sent);
 668
 669                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 670                                 xdr->len - req->rq_bytes_sent, status);
 671
 672                 /* If we've sent the entire packet, immediately
 673                  * reset the count of bytes sent. */
 674                 req->rq_bytes_sent += sent;
 675                 req->rq_xmit_bytes_sent += sent;
 676                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 677                         req->rq_bytes_sent = 0;
 678                         return 0;
 679                 }
 680
 681                 WARN_ON_ONCE(sent == 0 && status == 0);
 682
 683                 if (status == -EAGAIN ) {
 684                         /*
 685                          * Return EAGAIN if we're sure we're hitting the
 686                          * socket send buffer limits.
 687                          */
 688                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 689                                 break;
 690                         /*
 691                          * Did we hit a memory allocation failure?
 692                          */
 693                         if (sent == 0) {
 694                                 status = -ENOBUFS;
 695                                 if (vm_wait)
 696                                         break;
 697                                 /* Retry, knowing now that we're below the
 698                                  * socket send buffer limit
 699                                  */
 700                                 vm_wait = true;
 701                         }
 702                         continue;
 703                 }
 704                 if (status < 0)
 705                         break;
 706                 vm_wait = false;
 707         }
 708
 709         switch (status) {
 710         case -ENOTSOCK:
 711                 status = -ENOTCONN;
 712                 /* Should we call xs_close() here? */
 713                 break;
 714         case -EAGAIN:
 715                 status = xs_nospace(task);
 716                 break;
 717         case -ECONNRESET:
 718         case -ECONNREFUSED:
 719         case -ENOTCONN:
 720         case -EADDRINUSE:
 721         case -ENOBUFS:
 722         case -EPIPE:
 723                 break;
 724         default:
 725                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 726                         -status);
 727         }
 728
 729         return status;
 730 }
 731
 732 /**
 733  * xs_tcp_release_xprt - clean up after a tcp transmission
 734  * @xprt: transport
 735  * @task: rpc task
 736  *
 737  * This cleans up if an error causes us to abort the transmission of a request.
 738  * In this case, the socket may need to be reset in order to avoid confusing
 739  * the server.
 740  */
 741 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 742 {
 743         struct rpc_rqst *req;
 744
 745         if (task != xprt->snd_task)
 746                 return;
 747         if (task == NULL)
 748                 goto out_release;
 749         req = task->tk_rqstp;
 750         if (req == NULL)
 751                 goto out_release;
 752         if (req->rq_bytes_sent == 0)
 753                 goto out_release;
 754         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 755                 goto out_release;
 756         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 757 out_release:
 758         xprt_release_xprt(xprt, task);
 759 }
 760
 761 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 762 {
 763         transport->old_data_ready = sk->sk_data_ready;
 764         transport->old_state_change = sk->sk_state_change;
 765         transport->old_write_space = sk->sk_write_space;
 766         transport->old_error_report = sk->sk_error_report;
 767 }
 768
 769 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 770 {
 771         sk->sk_data_ready = transport->old_data_ready;
 772         sk->sk_state_change = transport->old_state_change;
 773         sk->sk_write_space = transport->old_write_space;
 774         sk->sk_error_report = transport->old_error_report;
 775 }
 776
 777 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
 778 {
 779         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 780
 781         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 782 }
 783
 784 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 785 {
 786         smp_mb__before_atomic();
 787         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 788         clear_bit(XPRT_CLOSING, &xprt->state);
 789         xs_sock_reset_state_flags(xprt);
 790         smp_mb__after_atomic();
 791 }
 792
 793 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 794 {
 795         xs_sock_reset_connection_flags(xprt);
 796         /* Mark transport as closed and wake up all pending tasks */
 797         xprt_disconnect_done(xprt);
 798 }
 799
 800 /**
 801  * xs_error_report - callback to handle TCP socket state errors
 802  * @sk: socket
 803  *
 804  * Note: we don't call sock_error() since there may be a rpc_task
 805  * using the socket, and so we don't want to clear sk->sk_err.
 806  */
 807 static void xs_error_report(struct sock *sk)
 808 {
 809         struct rpc_xprt *xprt;
 810         int err;
 811
 812         read_lock_bh(&sk->sk_callback_lock);
 813         if (!(xprt = xprt_from_sock(sk)))
 814                 goto out;
 815
 816         err = -sk->sk_err;
 817         if (err == 0)
 818                 goto out;
 819         /* Is this a reset event? */
 820         if (sk->sk_state == TCP_CLOSE)
 821                 xs_sock_mark_closed(xprt);
 822         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 823                         xprt, -err);
 824         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 825         xprt_wake_pending_tasks(xprt, err);
 826  out:
 827         read_unlock_bh(&sk->sk_callback_lock);
 828 }
 829
 830 static void xs_reset_transport(struct sock_xprt *transport)
 831 {
 832         struct socket *sock = transport->sock;
 833         struct sock *sk = transport->inet;
 834         struct rpc_xprt *xprt = &transport->xprt;
 835
 836         if (sk == NULL)
 837                 return;
 838
 839         if (atomic_read(&transport->xprt.swapper))
 840                 sk_clear_memalloc(sk);
 841
 842         kernel_sock_shutdown(sock, SHUT_RDWR);
 843
 844         mutex_lock(&transport->recv_mutex);
 845         write_lock_bh(&sk->sk_callback_lock);
 846         transport->inet = NULL;
 847         transport->sock = NULL;
 848
 849         sk->sk_user_data = NULL;
 850
 851         xs_restore_old_callbacks(transport, sk);
 852         xprt_clear_connected(xprt);
 853         write_unlock_bh(&sk->sk_callback_lock);
 854         xs_sock_reset_connection_flags(xprt);
 855         mutex_unlock(&transport->recv_mutex);
 856
 857         trace_rpc_socket_close(xprt, sock);
 858         sock_release(sock);
 859 }
 860
 861 /**
 862  * xs_close - close a socket
 863  * @xprt: transport
 864  *
 865  * This is used when all requests are complete; ie, no DRC state remains
 866  * on the server we want to save.
 867  *
 868  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 869  * xs_reset_transport() zeroing the socket from underneath a writer.
 870  */
 871 static void xs_close(struct rpc_xprt *xprt)
 872 {
 873         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 874
 875         dprintk("RPC:       xs_close xprt %p\n", xprt);
 876
 877         xs_reset_transport(transport);
 878         xprt->reestablish_timeout = 0;
 879
 880         xprt_disconnect_done(xprt);
 881 }
 882
 883 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 884 {
 885         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 886                 xprt);
 887         xprt_disconnect_done(xprt);
 888 }
 889
 890 static void xs_xprt_free(struct rpc_xprt *xprt)
 891 {
 892         xs_free_peer_addresses(xprt);
 893         xprt_free(xprt);
 894 }
 895
 896 /**
 897  * xs_destroy - prepare to shutdown a transport
 898  * @xprt: doomed transport
 899  *
 900  */
 901 static void xs_destroy(struct rpc_xprt *xprt)
 902 {
 903         struct sock_xprt *transport = container_of(xprt,
 904                         struct sock_xprt, xprt);
 905         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 906
 907         cancel_delayed_work_sync(&transport->connect_worker);
 908         xs_close(xprt);
 909         cancel_work_sync(&transport->recv_worker);
 910         xs_xprt_free(xprt);
 911         module_put(THIS_MODULE);
 912 }
 913
 914 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 915 {
 916         struct xdr_skb_reader desc = {
 917                 .skb            = skb,
 918                 .offset         = sizeof(rpc_fraghdr),
 919                 .count          = skb->len - sizeof(rpc_fraghdr),
 920         };
 921
 922         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 923                 return -1;
 924         if (desc.count)
 925                 return -1;
 926         return 0;
 927 }
 928
 929 /**
 930  * xs_local_data_read_skb
 931  * @xprt: transport
 932  * @sk: socket
 933  * @skb: skbuff
 934  *
 935  * Currently this assumes we can read the whole reply in a single gulp.
 936  */
 937 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 938                 struct sock *sk,
 939                 struct sk_buff *skb)
 940 {
 941         struct rpc_task *task;
 942         struct rpc_rqst *rovr;
 943         int repsize, copied;
 944         u32 _xid;
 945         __be32 *xp;
 946
 947         repsize = skb->len - sizeof(rpc_fraghdr);
 948         if (repsize < 4) {
 949                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 950                 return;
 951         }
 952
 953         /* Copy the XID from the skb... */
 954         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 955         if (xp == NULL)
 956                 return;
 957
 958         /* Look up and lock the request corresponding to the given XID */
 959         spin_lock_bh(&xprt->transport_lock);
 960         rovr = xprt_lookup_rqst(xprt, *xp);
 961         if (!rovr)
 962                 goto out_unlock;
 963         task = rovr->rq_task;
 964
 965         copied = rovr->rq_private_buf.buflen;
 966         if (copied > repsize)
 967                 copied = repsize;
 968
 969         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 970                 dprintk("RPC:       sk_buff copy failed\n");
 971                 goto out_unlock;
 972         }
 973
 974         xprt_complete_rqst(task, copied);
 975
 976  out_unlock:
 977         spin_unlock_bh(&xprt->transport_lock);
 978 }
 979
 980 static void xs_local_data_receive(struct sock_xprt *transport)
 981 {
 982         struct sk_buff *skb;
 983         struct sock *sk;
 984         int err;
 985
 986         mutex_lock(&transport->recv_mutex);
 987         sk = transport->inet;
 988         if (sk == NULL)
 989                 goto out;
 990         for (;;) {
 991                 skb = skb_recv_datagram(sk, 0, 1, &err);
 992                 if (skb != NULL) {
 993                         xs_local_data_read_skb(&transport->xprt, sk, skb);
 994                         skb_free_datagram(sk, skb);
 995                         continue;
 996                 }
 997                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 998                         break;
 999         }
1000 out:
1001         mutex_unlock(&transport->recv_mutex);
1002 }
1003
1004 static void xs_local_data_receive_workfn(struct work_struct *work)
1005 {
1006         struct sock_xprt *transport =
1007                 container_of(work, struct sock_xprt, recv_worker);
1008         xs_local_data_receive(transport);
1009 }
1010
1011 /**
1012  * xs_udp_data_read_skb - receive callback for UDP sockets
1013  * @xprt: transport
1014  * @sk: socket
1015  * @skb: skbuff
1016  *
1017  */
1018 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1019                 struct sock *sk,
1020                 struct sk_buff *skb)
1021 {
1022         struct rpc_task *task;
1023         struct rpc_rqst *rovr;
1024         int repsize, copied;
1025         u32 _xid;
1026         __be32 *xp;
1027
1028         repsize = skb->len;
1029         if (repsize < 4) {
1030                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1031                 return;
1032         }
1033
1034         /* Copy the XID from the skb... */
1035         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1036         if (xp == NULL)
1037                 return;
1038
1039         /* Look up and lock the request corresponding to the given XID */
1040         spin_lock_bh(&xprt->transport_lock);
1041         rovr = xprt_lookup_rqst(xprt, *xp);
1042         if (!rovr)
1043                 goto out_unlock;
1044         task = rovr->rq_task;
1045
1046         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1047                 copied = repsize;
1048
1049         /* Suck it into the iovec, verify checksum if not done by hw. */
1050         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1051                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1052                 goto out_unlock;
1053         }
1054
1055         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1056
1057         xprt_adjust_cwnd(xprt, task, copied);
1058         xprt_complete_rqst(task, copied);
1059
1060  out_unlock:
1061         spin_unlock_bh(&xprt->transport_lock);
1062 }
1063
1064 static void xs_udp_data_receive(struct sock_xprt *transport)
1065 {
1066         struct sk_buff *skb;
1067         struct sock *sk;
1068         int err;
1069
1070         mutex_lock(&transport->recv_mutex);
1071         sk = transport->inet;
1072         if (sk == NULL)
1073                 goto out;
1074         for (;;) {
1075                 skb = skb_recv_datagram(sk, 0, 1, &err);
1076                 if (skb != NULL) {
1077                         xs_udp_data_read_skb(&transport->xprt, sk, skb);
1078                         skb_free_datagram(sk, skb);
1079                         continue;
1080                 }
1081                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1082                         break;
1083         }
1084 out:
1085         mutex_unlock(&transport->recv_mutex);
1086 }
1087
1088 static void xs_udp_data_receive_workfn(struct work_struct *work)
1089 {
1090         struct sock_xprt *transport =
1091                 container_of(work, struct sock_xprt, recv_worker);
1092         xs_udp_data_receive(transport);
1093 }
1094
1095 /**
1096  * xs_data_ready - "data ready" callback for UDP sockets
1097  * @sk: socket with data to read
1098  *
1099  */
1100 static void xs_data_ready(struct sock *sk)
1101 {
1102         struct rpc_xprt *xprt;
1103
1104         read_lock_bh(&sk->sk_callback_lock);
1105         dprintk("RPC:       xs_data_ready...\n");
1106         xprt = xprt_from_sock(sk);
1107         if (xprt != NULL) {
1108                 struct sock_xprt *transport = container_of(xprt,
1109                                 struct sock_xprt, xprt);
1110                 transport->old_data_ready(sk);
1111                 /* Any data means we had a useful conversation, so
1112                  * then we don't need to delay the next reconnect
1113                  */
1114                 if (xprt->reestablish_timeout)
1115                         xprt->reestablish_timeout = 0;
1116                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1117                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1118         }
1119         read_unlock_bh(&sk->sk_callback_lock);
1120 }
1121
1122 /*
1123  * Helper function to force a TCP close if the server is sending
1124  * junk and/or it has put us in CLOSE_WAIT
1125  */
1126 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1127 {
1128         xprt_force_disconnect(xprt);
1129 }
1130
1131 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1132 {
1133         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1134         size_t len, used;
1135         char *p;
1136
1137         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1138         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1139         used = xdr_skb_read_bits(desc, p, len);
1140         transport->tcp_offset += used;
1141         if (used != len)
1142                 return;
1143
1144         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1145         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1146                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1147         else
1148                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1149         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1150
1151         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1152         transport->tcp_offset = 0;
1153
1154         /* Sanity check of the record length */
1155         if (unlikely(transport->tcp_reclen < 8)) {
1156                 dprintk("RPC:       invalid TCP record fragment length\n");
1157                 xs_tcp_force_close(xprt);
1158                 return;
1159         }
1160         dprintk("RPC:       reading TCP record fragment of length %d\n",
1161                         transport->tcp_reclen);
1162 }
1163
1164 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1165 {
1166         if (transport->tcp_offset == transport->tcp_reclen) {
1167                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1168                 transport->tcp_offset = 0;
1169                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1170                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1171                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1172                         transport->tcp_copied = 0;
1173                 }
1174         }
1175 }
1176
1177 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1178 {
1179         size_t len, used;
1180         char *p;
1181
1182         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1183         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1184         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1185         used = xdr_skb_read_bits(desc, p, len);
1186         transport->tcp_offset += used;
1187         if (used != len)
1188                 return;
1189         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1190         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1191         transport->tcp_copied = 4;
1192         dprintk("RPC:       reading %s XID %08x\n",
1193                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1194                                                               : "request with",
1195                         ntohl(transport->tcp_xid));
1196         xs_tcp_check_fraghdr(transport);
1197 }
1198
1199 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1200                                        struct xdr_skb_reader *desc)
1201 {
1202         size_t len, used;
1203         u32 offset;
1204         char *p;
1205
1206         /*
1207          * We want transport->tcp_offset to be 8 at the end of this routine
1208          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1209          * When this function is called for the first time,
1210          * transport->tcp_offset is 4 (after having already read the xid).
1211          */
1212         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1213         len = sizeof(transport->tcp_calldir) - offset;
1214         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1215         p = ((char *) &transport->tcp_calldir) + offset;
1216         used = xdr_skb_read_bits(desc, p, len);
1217         transport->tcp_offset += used;
1218         if (used != len)
1219                 return;
1220         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1221         /*
1222          * We don't yet have the XDR buffer, so we will write the calldir
1223          * out after we get the buffer from the 'struct rpc_rqst'
1224          */
1225         switch (ntohl(transport->tcp_calldir)) {
1226         case RPC_REPLY:
1227                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1228                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1229                 transport->tcp_flags |= TCP_RPC_REPLY;
1230                 break;
1231         case RPC_CALL:
1232                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1233                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1234                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1235                 break;
1236         default:
1237                 dprintk("RPC:       invalid request message type\n");
1238                 xs_tcp_force_close(&transport->xprt);
1239         }
1240         xs_tcp_check_fraghdr(transport);
1241 }
1242
1243 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1244                                      struct xdr_skb_reader *desc,
1245                                      struct rpc_rqst *req)
1246 {
1247         struct sock_xprt *transport =
1248                                 container_of(xprt, struct sock_xprt, xprt);
1249         struct xdr_buf *rcvbuf;
1250         size_t len;
1251         ssize_t r;
1252
1253         rcvbuf = &req->rq_private_buf;
1254
1255         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1256                 /*
1257                  * Save the RPC direction in the XDR buffer
1258                  */
1259                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1260                         &transport->tcp_calldir,
1261                         sizeof(transport->tcp_calldir));
1262                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1263                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1264         }
1265
1266         len = desc->count;
1267         if (len > transport->tcp_reclen - transport->tcp_offset) {
1268                 struct xdr_skb_reader my_desc;
1269
1270                 len = transport->tcp_reclen - transport->tcp_offset;
1271                 memcpy(&my_desc, desc, sizeof(my_desc));
1272                 my_desc.count = len;
1273                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1274                                           &my_desc, xdr_skb_read_bits);
1275                 desc->count -= r;
1276                 desc->offset += r;
1277         } else
1278                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1279                                           desc, xdr_skb_read_bits);
1280
1281         if (r > 0) {
1282                 transport->tcp_copied += r;
1283                 transport->tcp_offset += r;
1284         }
1285         if (r != len) {
1286                 /* Error when copying to the receive buffer,
1287                  * usually because we weren't able to allocate
1288                  * additional buffer pages. All we can do now
1289                  * is turn off TCP_RCV_COPY_DATA, so the request
1290                  * will not receive any additional updates,
1291                  * and time out.
1292                  * Any remaining data from this record will
1293                  * be discarded.
1294                  */
1295                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1296                 dprintk("RPC:       XID %08x truncated request\n",
1297                                 ntohl(transport->tcp_xid));
1298                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1299                                 "tcp_offset = %u, tcp_reclen = %u\n",
1300                                 xprt, transport->tcp_copied,
1301                                 transport->tcp_offset, transport->tcp_reclen);
1302                 return;
1303         }
1304
1305         dprintk("RPC:       XID %08x read %Zd bytes\n",
1306                         ntohl(transport->tcp_xid), r);
1307         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1308                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1309                         transport->tcp_offset, transport->tcp_reclen);
1310
1311         if (transport->tcp_copied == req->rq_private_buf.buflen)
1312                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1313         else if (transport->tcp_offset == transport->tcp_reclen) {
1314                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1315                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1316         }
1317 }
1318
1319 /*
1320  * Finds the request corresponding to the RPC xid and invokes the common
1321  * tcp read code to read the data.
1322  */
1323 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1324                                     struct xdr_skb_reader *desc)
1325 {
1326         struct sock_xprt *transport =
1327                                 container_of(xprt, struct sock_xprt, xprt);
1328         struct rpc_rqst *req;
1329
1330         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1331
1332         /* Find and lock the request corresponding to this xid */
1333         spin_lock_bh(&xprt->transport_lock);
1334         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1335         if (!req) {
1336                 dprintk("RPC:       XID %08x request not found!\n",
1337                                 ntohl(transport->tcp_xid));
1338                 spin_unlock_bh(&xprt->transport_lock);
1339                 return -1;
1340         }
1341
1342         xs_tcp_read_common(xprt, desc, req);
1343
1344         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1345                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1346
1347         spin_unlock_bh(&xprt->transport_lock);
1348         return 0;
1349 }
1350
1351 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1352 /*
1353  * Obtains an rpc_rqst previously allocated and invokes the common
1354  * tcp read code to read the data.  The result is placed in the callback
1355  * queue.
1356  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1357  * connection and return -1.
1358  */
1359 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1360                                        struct xdr_skb_reader *desc)
1361 {
1362         struct sock_xprt *transport =
1363                                 container_of(xprt, struct sock_xprt, xprt);
1364         struct rpc_rqst *req;
1365
1366         /* Look up and lock the request corresponding to the given XID */
1367         spin_lock_bh(&xprt->transport_lock);
1368         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1369         if (req == NULL) {
1370                 spin_unlock_bh(&xprt->transport_lock);
1371                 printk(KERN_WARNING "Callback slot table overflowed\n");
1372                 xprt_force_disconnect(xprt);
1373                 return -1;
1374         }
1375
1376         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1377         xs_tcp_read_common(xprt, desc, req);
1378
1379         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1380                 xprt_complete_bc_request(req, transport->tcp_copied);
1381         spin_unlock_bh(&xprt->transport_lock);
1382
1383         return 0;
1384 }
1385
1386 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1387                                         struct xdr_skb_reader *desc)
1388 {
1389         struct sock_xprt *transport =
1390                                 container_of(xprt, struct sock_xprt, xprt);
1391
1392         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1393                 xs_tcp_read_reply(xprt, desc) :
1394                 xs_tcp_read_callback(xprt, desc);
1395 }
1396
1397 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1398 {
1399         int ret;
1400
1401         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1402                               SVC_SOCK_ANONYMOUS);
1403         if (ret < 0)
1404                 return ret;
1405         return 0;
1406 }
1407
1408 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1409 {
1410         return PAGE_SIZE;
1411 }
1412 #else
1413 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1414                                         struct xdr_skb_reader *desc)
1415 {
1416         return xs_tcp_read_reply(xprt, desc);
1417 }
1418 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1419
1420 /*
1421  * Read data off the transport.  This can be either an RPC_CALL or an
1422  * RPC_REPLY.  Relay the processing to helper functions.
1423  */
1424 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1425                                     struct xdr_skb_reader *desc)
1426 {
1427         struct sock_xprt *transport =
1428                                 container_of(xprt, struct sock_xprt, xprt);
1429
1430         if (_xs_tcp_read_data(xprt, desc) == 0)
1431                 xs_tcp_check_fraghdr(transport);
1432         else {
1433                 /*
1434                  * The transport_lock protects the request handling.
1435                  * There's no need to hold it to update the tcp_flags.
1436                  */
1437                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1438         }
1439 }
1440
1441 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1442 {
1443         size_t len;
1444
1445         len = transport->tcp_reclen - transport->tcp_offset;
1446         if (len > desc->count)
1447                 len = desc->count;
1448         desc->count -= len;
1449         desc->offset += len;
1450         transport->tcp_offset += len;
1451         dprintk("RPC:       discarded %Zu bytes\n", len);
1452         xs_tcp_check_fraghdr(transport);
1453 }
1454
1455 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1456 {
1457         struct rpc_xprt *xprt = rd_desc->arg.data;
1458         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1459         struct xdr_skb_reader desc = {
1460                 .skb    = skb,
1461                 .offset = offset,
1462                 .count  = len,
1463         };
1464
1465         dprintk("RPC:       xs_tcp_data_recv started\n");
1466         do {
1467                 trace_xs_tcp_data_recv(transport);
1468                 /* Read in a new fragment marker if necessary */
1469                 /* Can we ever really expect to get completely empty fragments? */
1470                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1471                         xs_tcp_read_fraghdr(xprt, &desc);
1472                         continue;
1473                 }
1474                 /* Read in the xid if necessary */
1475                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1476                         xs_tcp_read_xid(transport, &desc);
1477                         continue;
1478                 }
1479                 /* Read in the call/reply flag */
1480                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1481                         xs_tcp_read_calldir(transport, &desc);
1482                         continue;
1483                 }
1484                 /* Read in the request data */
1485                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1486                         xs_tcp_read_data(xprt, &desc);
1487                         continue;
1488                 }
1489                 /* Skip over any trailing bytes on short reads */
1490                 xs_tcp_read_discard(transport, &desc);
1491         } while (desc.count);
1492         trace_xs_tcp_data_recv(transport);
1493         dprintk("RPC:       xs_tcp_data_recv done\n");
1494         return len - desc.count;
1495 }
1496
1497 static void xs_tcp_data_receive(struct sock_xprt *transport)
1498 {
1499         struct rpc_xprt *xprt = &transport->xprt;
1500         struct sock *sk;
1501         read_descriptor_t rd_desc = {
1502                 .count = 2*1024*1024,
1503                 .arg.data = xprt,
1504         };
1505         unsigned long total = 0;
1506         int read = 0;
1507
1508         mutex_lock(&transport->recv_mutex);
1509         sk = transport->inet;
1510         if (sk == NULL)
1511                 goto out;
1512
1513         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1514         for (;;) {
1515                 lock_sock(sk);
1516                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1517                 if (read <= 0) {
1518                         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1519                         release_sock(sk);
1520                         if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1521                                 break;
1522                 } else {
1523                         release_sock(sk);
1524                         total += read;
1525                 }
1526                 rd_desc.count = 65536;
1527         }
1528 out:
1529         mutex_unlock(&transport->recv_mutex);
1530         trace_xs_tcp_data_ready(xprt, read, total);
1531 }
1532
1533 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1534 {
1535         struct sock_xprt *transport =
1536                 container_of(work, struct sock_xprt, recv_worker);
1537         xs_tcp_data_receive(transport);
1538 }
1539
1540 /**
1541  * xs_tcp_state_change - callback to handle TCP socket state changes
1542  * @sk: socket whose state has changed
1543  *
1544  */
1545 static void xs_tcp_state_change(struct sock *sk)
1546 {
1547         struct rpc_xprt *xprt;
1548         struct sock_xprt *transport;
1549
1550         read_lock_bh(&sk->sk_callback_lock);
1551         if (!(xprt = xprt_from_sock(sk)))
1552                 goto out;
1553         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1554         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1555                         sk->sk_state, xprt_connected(xprt),
1556                         sock_flag(sk, SOCK_DEAD),
1557                         sock_flag(sk, SOCK_ZAPPED),
1558                         sk->sk_shutdown);
1559
1560         transport = container_of(xprt, struct sock_xprt, xprt);
1561         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1562         switch (sk->sk_state) {
1563         case TCP_ESTABLISHED:
1564                 spin_lock(&xprt->transport_lock);
1565                 if (!xprt_test_and_set_connected(xprt)) {
1566
1567                         /* Reset TCP record info */
1568                         transport->tcp_offset = 0;
1569                         transport->tcp_reclen = 0;
1570                         transport->tcp_copied = 0;
1571                         transport->tcp_flags =
1572                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1573                         xprt->connect_cookie++;
1574                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1575                         xprt_clear_connecting(xprt);
1576
1577                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1578                 }
1579                 spin_unlock(&xprt->transport_lock);
1580                 break;
1581         case TCP_FIN_WAIT1:
1582                 /* The client initiated a shutdown of the socket */
1583                 xprt->connect_cookie++;
1584                 xprt->reestablish_timeout = 0;
1585                 set_bit(XPRT_CLOSING, &xprt->state);
1586                 smp_mb__before_atomic();
1587                 clear_bit(XPRT_CONNECTED, &xprt->state);
1588                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1589                 smp_mb__after_atomic();
1590                 break;
1591         case TCP_CLOSE_WAIT:
1592                 /* The server initiated a shutdown of the socket */
1593                 xprt->connect_cookie++;
1594                 clear_bit(XPRT_CONNECTED, &xprt->state);
1595                 xs_tcp_force_close(xprt);
1596         case TCP_CLOSING:
1597                 /*
1598                  * If the server closed down the connection, make sure that
1599                  * we back off before reconnecting
1600                  */
1601                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1602                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1603                 break;
1604         case TCP_LAST_ACK:
1605                 set_bit(XPRT_CLOSING, &xprt->state);
1606                 smp_mb__before_atomic();
1607                 clear_bit(XPRT_CONNECTED, &xprt->state);
1608                 smp_mb__after_atomic();
1609                 break;
1610         case TCP_CLOSE:
1611                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1612                                         &transport->sock_state))
1613                         xprt_clear_connecting(xprt);
1614                 xs_sock_mark_closed(xprt);
1615         }
1616  out:
1617         read_unlock_bh(&sk->sk_callback_lock);
1618 }
1619
1620 static void xs_write_space(struct sock *sk)
1621 {
1622         struct socket_wq *wq;
1623         struct rpc_xprt *xprt;
1624
1625         if (!sk->sk_socket)
1626                 return;
1627         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1628
1629         if (unlikely(!(xprt = xprt_from_sock(sk))))
1630                 return;
1631         rcu_read_lock();
1632         wq = rcu_dereference(sk->sk_wq);
1633         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1634                 goto out;
1635
1636         xprt_write_space(xprt);
1637 out:
1638         rcu_read_unlock();
1639 }
1640
1641 /**
1642  * xs_udp_write_space - callback invoked when socket buffer space
1643  *                             becomes available
1644  * @sk: socket whose state has changed
1645  *
1646  * Called when more output buffer space is available for this socket.
1647  * We try not to wake our writers until they can make "significant"
1648  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1649  * with a bunch of small requests.
1650  */
1651 static void xs_udp_write_space(struct sock *sk)
1652 {
1653         read_lock_bh(&sk->sk_callback_lock);
1654
1655         /* from net/core/sock.c:sock_def_write_space */
1656         if (sock_writeable(sk))
1657                 xs_write_space(sk);
1658
1659         read_unlock_bh(&sk->sk_callback_lock);
1660 }
1661
1662 /**
1663  * xs_tcp_write_space - callback invoked when socket buffer space
1664  *                             becomes available
1665  * @sk: socket whose state has changed
1666  *
1667  * Called when more output buffer space is available for this socket.
1668  * We try not to wake our writers until they can make "significant"
1669  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1670  * with a bunch of small requests.
1671  */
1672 static void xs_tcp_write_space(struct sock *sk)
1673 {
1674         read_lock_bh(&sk->sk_callback_lock);
1675
1676         /* from net/core/stream.c:sk_stream_write_space */
1677         if (sk_stream_is_writeable(sk))
1678                 xs_write_space(sk);
1679
1680         read_unlock_bh(&sk->sk_callback_lock);
1681 }
1682
1683 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1684 {
1685         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1686         struct sock *sk = transport->inet;
1687
1688         if (transport->rcvsize) {
1689                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1690                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1691         }
1692         if (transport->sndsize) {
1693                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1694                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1695                 sk->sk_write_space(sk);
1696         }
1697 }
1698
1699 /**
1700  * xs_udp_set_buffer_size - set send and receive limits
1701  * @xprt: generic transport
1702  * @sndsize: requested size of send buffer, in bytes
1703  * @rcvsize: requested size of receive buffer, in bytes
1704  *
1705  * Set socket send and receive buffer size limits.
1706  */
1707 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1708 {
1709         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1710
1711         transport->sndsize = 0;
1712         if (sndsize)
1713                 transport->sndsize = sndsize + 1024;
1714         transport->rcvsize = 0;
1715         if (rcvsize)
1716                 transport->rcvsize = rcvsize + 1024;
1717
1718         xs_udp_do_set_buffer_size(xprt);
1719 }
1720
1721 /**
1722  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1723  * @task: task that timed out
1724  *
1725  * Adjust the congestion window after a retransmit timeout has occurred.
1726  */
1727 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1728 {
1729         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1730 }
1731
1732 static unsigned short xs_get_random_port(void)
1733 {
1734         unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1735         unsigned short rand = (unsigned short) prandom_u32() % range;
1736         return rand + xprt_min_resvport;
1737 }
1738
1739 /**
1740  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1741  * @sock: socket
1742  *
1743  * Note that this function has to be called on all sockets that share the
1744  * same port, and it must be called before binding.
1745  */
1746 static void xs_sock_set_reuseport(struct socket *sock)
1747 {
1748         int opt = 1;
1749
1750         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1751                         (char *)&opt, sizeof(opt));
1752 }
1753
1754 static unsigned short xs_sock_getport(struct socket *sock)
1755 {
1756         struct sockaddr_storage buf;
1757         int buflen;
1758         unsigned short port = 0;
1759
1760         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1761                 goto out;
1762         switch (buf.ss_family) {
1763         case AF_INET6:
1764                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1765                 break;
1766         case AF_INET:
1767                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1768         }
1769 out:
1770         return port;
1771 }
1772
1773 /**
1774  * xs_set_port - reset the port number in the remote endpoint address
1775  * @xprt: generic transport
1776  * @port: new port number
1777  *
1778  */
1779 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1780 {
1781         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1782
1783         rpc_set_port(xs_addr(xprt), port);
1784         xs_update_peer_port(xprt);
1785 }
1786
1787 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1788 {
1789         if (transport->srcport == 0)
1790                 transport->srcport = xs_sock_getport(sock);
1791 }
1792
1793 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1794 {
1795         unsigned short port = transport->srcport;
1796
1797         if (port == 0 && transport->xprt.resvport)
1798                 port = xs_get_random_port();
1799         return port;
1800 }
1801
1802 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1803 {
1804         if (transport->srcport != 0)
1805                 transport->srcport = 0;
1806         if (!transport->xprt.resvport)
1807                 return 0;
1808         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1809                 return xprt_max_resvport;
1810         return --port;
1811 }
1812 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1813 {
1814         struct sockaddr_storage myaddr;
1815         int err, nloop = 0;
1816         unsigned short port = xs_get_srcport(transport);
1817         unsigned short last;
1818
1819         /*
1820          * If we are asking for any ephemeral port (i.e. port == 0 &&
1821          * transport->xprt.resvport == 0), don't bind.  Let the local
1822          * port selection happen implicitly when the socket is used
1823          * (for example at connect time).
1824          *
1825          * This ensures that we can continue to establish TCP
1826          * connections even when all local ephemeral ports are already
1827          * a part of some TCP connection.  This makes no difference
1828          * for UDP sockets, but also doens't harm them.
1829          *
1830          * If we're asking for any reserved port (i.e. port == 0 &&
1831          * transport->xprt.resvport == 1) xs_get_srcport above will
1832          * ensure that port is non-zero and we will bind as needed.
1833          */
1834         if (port == 0)
1835                 return 0;
1836
1837         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1838         do {
1839                 rpc_set_port((struct sockaddr *)&myaddr, port);
1840                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1841                                 transport->xprt.addrlen);
1842                 if (err == 0) {
1843                         transport->srcport = port;
1844                         break;
1845                 }
1846                 last = port;
1847                 port = xs_next_srcport(transport, port);
1848                 if (port > last)
1849                         nloop++;
1850         } while (err == -EADDRINUSE && nloop != 2);
1851
1852         if (myaddr.ss_family == AF_INET)
1853                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1854                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1855                                 port, err ? "failed" : "ok", err);
1856         else
1857                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1858                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1859                                 port, err ? "failed" : "ok", err);
1860         return err;
1861 }
1862
1863 /*
1864  * We don't support autobind on AF_LOCAL sockets
1865  */
1866 static void xs_local_rpcbind(struct rpc_task *task)
1867 {
1868         xprt_set_bound(task->tk_xprt);
1869 }
1870
1871 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1872 {
1873 }
1874
1875 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1876 static struct lock_class_key xs_key[2];
1877 static struct lock_class_key xs_slock_key[2];
1878
1879 static inline void xs_reclassify_socketu(struct socket *sock)
1880 {
1881         struct sock *sk = sock->sk;
1882
1883         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1884                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1885 }
1886
1887 static inline void xs_reclassify_socket4(struct socket *sock)
1888 {
1889         struct sock *sk = sock->sk;
1890
1891         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1892                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1893 }
1894
1895 static inline void xs_reclassify_socket6(struct socket *sock)
1896 {
1897         struct sock *sk = sock->sk;
1898
1899         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1900                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1901 }
1902
1903 static inline void xs_reclassify_socket(int family, struct socket *sock)
1904 {
1905         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1906                 return;
1907
1908         switch (family) {
1909         case AF_LOCAL:
1910                 xs_reclassify_socketu(sock);
1911                 break;
1912         case AF_INET:
1913                 xs_reclassify_socket4(sock);
1914                 break;
1915         case AF_INET6:
1916                 xs_reclassify_socket6(sock);
1917                 break;
1918         }
1919 }
1920 #else
1921 static inline void xs_reclassify_socket(int family, struct socket *sock)
1922 {
1923 }
1924 #endif
1925
1926 static void xs_dummy_setup_socket(struct work_struct *work)
1927 {
1928 }
1929
1930 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1931                 struct sock_xprt *transport, int family, int type,
1932                 int protocol, bool reuseport)
1933 {
1934         struct socket *sock;
1935         int err;
1936
1937         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1938         if (err < 0) {
1939                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1940                                 protocol, -err);
1941                 goto out;
1942         }
1943         xs_reclassify_socket(family, sock);
1944
1945         if (reuseport)
1946                 xs_sock_set_reuseport(sock);
1947
1948         err = xs_bind(transport, sock);
1949         if (err) {
1950                 sock_release(sock);
1951                 goto out;
1952         }
1953
1954         return sock;
1955 out:
1956         return ERR_PTR(err);
1957 }
1958
1959 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1960                                       struct socket *sock)
1961 {
1962         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1963                                                                         xprt);
1964
1965         if (!transport->inet) {
1966                 struct sock *sk = sock->sk;
1967
1968                 write_lock_bh(&sk->sk_callback_lock);
1969
1970                 xs_save_old_callbacks(transport, sk);
1971
1972                 sk->sk_user_data = xprt;
1973                 sk->sk_data_ready = xs_data_ready;
1974                 sk->sk_write_space = xs_udp_write_space;
1975                 sock_set_flag(sk, SOCK_FASYNC);
1976                 sk->sk_error_report = xs_error_report;
1977                 sk->sk_allocation = GFP_NOIO;
1978
1979                 xprt_clear_connected(xprt);
1980
1981                 /* Reset to new socket */
1982                 transport->sock = sock;
1983                 transport->inet = sk;
1984
1985                 write_unlock_bh(&sk->sk_callback_lock);
1986         }
1987
1988         /* Tell the socket layer to start connecting... */
1989         xprt->stat.connect_count++;
1990         xprt->stat.connect_start = jiffies;
1991         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1992 }
1993
1994 /**
1995  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1996  * @transport: socket transport to connect
1997  */
1998 static int xs_local_setup_socket(struct sock_xprt *transport)
1999 {
2000         struct rpc_xprt *xprt = &transport->xprt;
2001         struct socket *sock;
2002         int status = -EIO;
2003
2004         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2005                                         SOCK_STREAM, 0, &sock, 1);
2006         if (status < 0) {
2007                 dprintk("RPC:       can't create AF_LOCAL "
2008                         "transport socket (%d).\n", -status);
2009                 goto out;
2010         }
2011         xs_reclassify_socket(AF_LOCAL, sock);
2012
2013         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2014                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2015
2016         status = xs_local_finish_connecting(xprt, sock);
2017         trace_rpc_socket_connect(xprt, sock, status);
2018         switch (status) {
2019         case 0:
2020                 dprintk("RPC:       xprt %p connected to %s\n",
2021                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2022                 xprt_set_connected(xprt);
2023         case -ENOBUFS:
2024                 break;
2025         case -ENOENT:
2026                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2027                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2028                 break;
2029         case -ECONNREFUSED:
2030                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2031                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2032                 break;
2033         default:
2034                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2035                                 __func__, -status,
2036                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2037         }
2038
2039 out:
2040         xprt_clear_connecting(xprt);
2041         xprt_wake_pending_tasks(xprt, status);
2042         return status;
2043 }
2044
2045 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2046 {
2047         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2048         int ret;
2049
2050          if (RPC_IS_ASYNC(task)) {
2051                 /*
2052                  * We want the AF_LOCAL connect to be resolved in the
2053                  * filesystem namespace of the process making the rpc
2054                  * call.  Thus we connect synchronously.
2055                  *
2056                  * If we want to support asynchronous AF_LOCAL calls,
2057                  * we'll need to figure out how to pass a namespace to
2058                  * connect.
2059                  */
2060                 rpc_exit(task, -ENOTCONN);
2061                 return;
2062         }
2063         ret = xs_local_setup_socket(transport);
2064         if (ret && !RPC_IS_SOFTCONN(task))
2065                 msleep_interruptible(15000);
2066 }
2067
2068 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2069 /*
2070  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2071  * know that we have exclusive access to the socket), to guard against
2072  * races with xs_reset_transport.
2073  */
2074 static void xs_set_memalloc(struct rpc_xprt *xprt)
2075 {
2076         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2077                         xprt);
2078
2079         /*
2080          * If there's no sock, then we have nothing to set. The
2081          * reconnecting process will get it for us.
2082          */
2083         if (!transport->inet)
2084                 return;
2085         if (atomic_read(&xprt->swapper))
2086                 sk_set_memalloc(transport->inet);
2087 }
2088
2089 /**
2090  * xs_enable_swap - Tag this transport as being used for swap.
2091  * @xprt: transport to tag
2092  *
2093  * Take a reference to this transport on behalf of the rpc_clnt, and
2094  * optionally mark it for swapping if it wasn't already.
2095  */
2096 static int
2097 xs_enable_swap(struct rpc_xprt *xprt)
2098 {
2099         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2100
2101         if (atomic_inc_return(&xprt->swapper) != 1)
2102                 return 0;
2103         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2104                 return -ERESTARTSYS;
2105         if (xs->inet)
2106                 sk_set_memalloc(xs->inet);
2107         xprt_release_xprt(xprt, NULL);
2108         return 0;
2109 }
2110
2111 /**
2112  * xs_disable_swap - Untag this transport as being used for swap.
2113  * @xprt: transport to tag
2114  *
2115  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2116  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2117  */
2118 static void
2119 xs_disable_swap(struct rpc_xprt *xprt)
2120 {
2121         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2122
2123         if (!atomic_dec_and_test(&xprt->swapper))
2124                 return;
2125         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2126                 return;
2127         if (xs->inet)
2128                 sk_clear_memalloc(xs->inet);
2129         xprt_release_xprt(xprt, NULL);
2130 }
2131 #else
2132 static void xs_set_memalloc(struct rpc_xprt *xprt)
2133 {
2134 }
2135
2136 static int
2137 xs_enable_swap(struct rpc_xprt *xprt)
2138 {
2139         return -EINVAL;
2140 }
2141
2142 static void
2143 xs_disable_swap(struct rpc_xprt *xprt)
2144 {
2145 }
2146 #endif
2147
2148 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2149 {
2150         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2151
2152         if (!transport->inet) {
2153                 struct sock *sk = sock->sk;
2154
2155                 write_lock_bh(&sk->sk_callback_lock);
2156
2157                 xs_save_old_callbacks(transport, sk);
2158
2159                 sk->sk_user_data = xprt;
2160                 sk->sk_data_ready = xs_data_ready;
2161                 sk->sk_write_space = xs_udp_write_space;
2162                 sock_set_flag(sk, SOCK_FASYNC);
2163                 sk->sk_allocation = GFP_NOIO;
2164
2165                 xprt_set_connected(xprt);
2166
2167                 /* Reset to new socket */
2168                 transport->sock = sock;
2169                 transport->inet = sk;
2170
2171                 xs_set_memalloc(xprt);
2172
2173                 write_unlock_bh(&sk->sk_callback_lock);
2174         }
2175         xs_udp_do_set_buffer_size(xprt);
2176
2177         xprt->stat.connect_start = jiffies;
2178 }
2179
2180 static void xs_udp_setup_socket(struct work_struct *work)
2181 {
2182         struct sock_xprt *transport =
2183                 container_of(work, struct sock_xprt, connect_worker.work);
2184         struct rpc_xprt *xprt = &transport->xprt;
2185         struct socket *sock = transport->sock;
2186         int status = -EIO;
2187
2188         sock = xs_create_sock(xprt, transport,
2189                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2190                         IPPROTO_UDP, false);
2191         if (IS_ERR(sock))
2192                 goto out;
2193
2194         dprintk("RPC:       worker connecting xprt %p via %s to "
2195                                 "%s (port %s)\n", xprt,
2196                         xprt->address_strings[RPC_DISPLAY_PROTO],
2197                         xprt->address_strings[RPC_DISPLAY_ADDR],
2198                         xprt->address_strings[RPC_DISPLAY_PORT]);
2199
2200         xs_udp_finish_connecting(xprt, sock);
2201         trace_rpc_socket_connect(xprt, sock, 0);
2202         status = 0;
2203 out:
2204         xprt_unlock_connect(xprt, transport);
2205         xprt_clear_connecting(xprt);
2206         xprt_wake_pending_tasks(xprt, status);
2207 }
2208
2209 /**
2210  * xs_tcp_shutdown - gracefully shut down a TCP socket
2211  * @xprt: transport
2212  *
2213  * Initiates a graceful shutdown of the TCP socket by calling the
2214  * equivalent of shutdown(SHUT_RDWR);
2215  */
2216 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2217 {
2218         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2219         struct socket *sock = transport->sock;
2220
2221         if (sock == NULL)
2222                 return;
2223         if (xprt_connected(xprt)) {
2224                 kernel_sock_shutdown(sock, SHUT_RDWR);
2225                 trace_rpc_socket_shutdown(xprt, sock);
2226         } else
2227                 xs_reset_transport(transport);
2228 }
2229
2230 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2231 {
2232         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2233         int ret = -ENOTCONN;
2234
2235         if (!transport->inet) {
2236                 struct sock *sk = sock->sk;
2237                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2238                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2239                 unsigned int opt_on = 1;
2240                 unsigned int timeo;
2241                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2242
2243                 /* TCP Keepalive options */
2244                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2245                                 (char *)&opt_on, sizeof(opt_on));
2246                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2247                                 (char *)&keepidle, sizeof(keepidle));
2248                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2249                                 (char *)&keepidle, sizeof(keepidle));
2250                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2251                                 (char *)&keepcnt, sizeof(keepcnt));
2252
2253                 /* Avoid temporary address, they are bad for long-lived
2254                  * connections such as NFS mounts.
2255                  * RFC4941, section 3.6 suggests that:
2256                  *    Individual applications, which have specific
2257                  *    knowledge about the normal duration of connections,
2258                  *    MAY override this as appropriate.
2259                  */
2260                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2261                                 (char *)&addr_pref, sizeof(addr_pref));
2262
2263                 /* TCP user timeout (see RFC5482) */
2264                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2265                         (xprt->timeout->to_retries + 1);
2266                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2267                                 (char *)&timeo, sizeof(timeo));
2268
2269                 write_lock_bh(&sk->sk_callback_lock);
2270
2271                 xs_save_old_callbacks(transport, sk);
2272
2273                 sk->sk_user_data = xprt;
2274                 sk->sk_data_ready = xs_data_ready;
2275                 sk->sk_state_change = xs_tcp_state_change;
2276                 sk->sk_write_space = xs_tcp_write_space;
2277                 sock_set_flag(sk, SOCK_FASYNC);
2278                 sk->sk_error_report = xs_error_report;
2279                 sk->sk_allocation = GFP_NOIO;
2280
2281                 /* socket options */
2282                 sock_reset_flag(sk, SOCK_LINGER);
2283                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2284
2285                 xprt_clear_connected(xprt);
2286
2287                 /* Reset to new socket */
2288                 transport->sock = sock;
2289                 transport->inet = sk;
2290
2291                 write_unlock_bh(&sk->sk_callback_lock);
2292         }
2293
2294         if (!xprt_bound(xprt))
2295                 goto out;
2296
2297         xs_set_memalloc(xprt);
2298
2299         /* Tell the socket layer to start connecting... */
2300         xprt->stat.connect_count++;
2301         xprt->stat.connect_start = jiffies;
2302         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2303         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2304         switch (ret) {
2305         case 0:
2306                 xs_set_srcport(transport, sock);
2307         case -EINPROGRESS:
2308                 /* SYN_SENT! */
2309                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2310                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2311                 break;
2312         case -EADDRNOTAVAIL:
2313                 /* Source port number is unavailable. Try a new one! */
2314                 transport->srcport = 0;
2315         }
2316 out:
2317         return ret;
2318 }
2319
2320 /**
2321  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2322  *
2323  * Invoked by a work queue tasklet.
2324  */
2325 static void xs_tcp_setup_socket(struct work_struct *work)
2326 {
2327         struct sock_xprt *transport =
2328                 container_of(work, struct sock_xprt, connect_worker.work);
2329         struct socket *sock = transport->sock;
2330         struct rpc_xprt *xprt = &transport->xprt;
2331         int status = -EIO;
2332
2333         if (!sock) {
2334                 sock = xs_create_sock(xprt, transport,
2335                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2336                                 IPPROTO_TCP, true);
2337                 if (IS_ERR(sock)) {
2338                         status = PTR_ERR(sock);
2339                         goto out;
2340                 }
2341         }
2342
2343         dprintk("RPC:       worker connecting xprt %p via %s to "
2344                                 "%s (port %s)\n", xprt,
2345                         xprt->address_strings[RPC_DISPLAY_PROTO],
2346                         xprt->address_strings[RPC_DISPLAY_ADDR],
2347                         xprt->address_strings[RPC_DISPLAY_PORT]);
2348
2349         status = xs_tcp_finish_connecting(xprt, sock);
2350         trace_rpc_socket_connect(xprt, sock, status);
2351         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2352                         xprt, -status, xprt_connected(xprt),
2353                         sock->sk->sk_state);
2354         switch (status) {
2355         default:
2356                 printk("%s: connect returned unhandled error %d\n",
2357                         __func__, status);
2358         case -EADDRNOTAVAIL:
2359                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2360                  * and retry
2361                  */
2362                 xs_tcp_force_close(xprt);
2363                 break;
2364         case 0:
2365         case -EINPROGRESS:
2366         case -EALREADY:
2367                 xprt_unlock_connect(xprt, transport);
2368                 return;
2369         case -EINVAL:
2370                 /* Happens, for instance, if the user specified a link
2371                  * local IPv6 address without a scope-id.
2372                  */
2373         case -ECONNREFUSED:
2374         case -ECONNRESET:
2375         case -ENETUNREACH:
2376         case -EADDRINUSE:
2377         case -ENOBUFS:
2378                 /* retry with existing socket, after a delay */
2379                 xs_tcp_force_close(xprt);
2380                 goto out;
2381         }
2382         status = -EAGAIN;
2383 out:
2384         xprt_unlock_connect(xprt, transport);
2385         xprt_clear_connecting(xprt);
2386         xprt_wake_pending_tasks(xprt, status);
2387 }
2388
2389 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2390 {
2391         unsigned long start, now = jiffies;
2392
2393         start = xprt->stat.connect_start + xprt->reestablish_timeout;
2394         if (time_after(start, now))
2395                 return start - now;
2396         return 0;
2397 }
2398
2399 /**
2400  * xs_connect - connect a socket to a remote endpoint
2401  * @xprt: pointer to transport structure
2402  * @task: address of RPC task that manages state of connect request
2403  *
2404  * TCP: If the remote end dropped the connection, delay reconnecting.
2405  *
2406  * UDP socket connects are synchronous, but we use a work queue anyway
2407  * to guarantee that even unprivileged user processes can set up a
2408  * socket on a privileged port.
2409  *
2410  * If a UDP socket connect fails, the delay behavior here prevents
2411  * retry floods (hard mounts).
2412  */
2413 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2414 {
2415         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2416         unsigned long delay = 0;
2417
2418         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2419
2420         if (transport->sock != NULL) {
2421                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2422                                 "seconds\n",
2423                                 xprt, xprt->reestablish_timeout / HZ);
2424
2425                 /* Start by resetting any existing state */
2426                 xs_reset_transport(transport);
2427
2428                 delay = xs_reconnect_delay(xprt);
2429
2430                 xprt->reestablish_timeout <<= 1;
2431                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2432                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2433                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2434                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2435         } else
2436                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2437
2438         queue_delayed_work(xprtiod_workqueue,
2439                         &transport->connect_worker,
2440                         delay);
2441 }
2442
2443 /**
2444  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2445  * @xprt: rpc_xprt struct containing statistics
2446  * @seq: output file
2447  *
2448  */
2449 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2450 {
2451         long idle_time = 0;
2452
2453         if (xprt_connected(xprt))
2454                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2455
2456         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2457                         "%llu %llu %lu %llu %llu\n",
2458                         xprt->stat.bind_count,
2459                         xprt->stat.connect_count,
2460                         xprt->stat.connect_time,
2461                         idle_time,
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_udp_print_stats - display UDP socket-specifc stats
2474  * @xprt: rpc_xprt struct containing statistics
2475  * @seq: output file
2476  *
2477  */
2478 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2479 {
2480         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2481
2482         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2483                         "%lu %llu %llu\n",
2484                         transport->srcport,
2485                         xprt->stat.bind_count,
2486                         xprt->stat.sends,
2487                         xprt->stat.recvs,
2488                         xprt->stat.bad_xids,
2489                         xprt->stat.req_u,
2490                         xprt->stat.bklog_u,
2491                         xprt->stat.max_slots,
2492                         xprt->stat.sending_u,
2493                         xprt->stat.pending_u);
2494 }
2495
2496 /**
2497  * xs_tcp_print_stats - display TCP socket-specifc stats
2498  * @xprt: rpc_xprt struct containing statistics
2499  * @seq: output file
2500  *
2501  */
2502 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2503 {
2504         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2505         long idle_time = 0;
2506
2507         if (xprt_connected(xprt))
2508                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2509
2510         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2511                         "%llu %llu %lu %llu %llu\n",
2512                         transport->srcport,
2513                         xprt->stat.bind_count,
2514                         xprt->stat.connect_count,
2515                         xprt->stat.connect_time,
2516                         idle_time,
2517                         xprt->stat.sends,
2518                         xprt->stat.recvs,
2519                         xprt->stat.bad_xids,
2520                         xprt->stat.req_u,
2521                         xprt->stat.bklog_u,
2522                         xprt->stat.max_slots,
2523                         xprt->stat.sending_u,
2524                         xprt->stat.pending_u);
2525 }
2526
2527 /*
2528  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2529  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2530  * to use the server side send routines.
2531  */
2532 static void *bc_malloc(struct rpc_task *task, size_t size)
2533 {
2534         struct page *page;
2535         struct rpc_buffer *buf;
2536
2537         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2538         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2539                 return NULL;
2540
2541         page = alloc_page(GFP_KERNEL);
2542         if (!page)
2543                 return NULL;
2544
2545         buf = page_address(page);
2546         buf->len = PAGE_SIZE;
2547
2548         return buf->data;
2549 }
2550
2551 /*
2552  * Free the space allocated in the bc_alloc routine
2553  */
2554 static void bc_free(void *buffer)
2555 {
2556         struct rpc_buffer *buf;
2557
2558         if (!buffer)
2559                 return;
2560
2561         buf = container_of(buffer, struct rpc_buffer, data);
2562         free_page((unsigned long)buf);
2563 }
2564
2565 /*
2566  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2567  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2568  */
2569 static int bc_sendto(struct rpc_rqst *req)
2570 {
2571         int len;
2572         struct xdr_buf *xbufp = &req->rq_snd_buf;
2573         struct rpc_xprt *xprt = req->rq_xprt;
2574         struct sock_xprt *transport =
2575                                 container_of(xprt, struct sock_xprt, xprt);
2576         struct socket *sock = transport->sock;
2577         unsigned long headoff;
2578         unsigned long tailoff;
2579
2580         xs_encode_stream_record_marker(xbufp);
2581
2582         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2583         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2584         len = svc_send_common(sock, xbufp,
2585                               virt_to_page(xbufp->head[0].iov_base), headoff,
2586                               xbufp->tail[0].iov_base, tailoff);
2587
2588         if (len != xbufp->len) {
2589                 printk(KERN_NOTICE "Error sending entire callback!\n");
2590                 len = -EAGAIN;
2591         }
2592
2593         return len;
2594 }
2595
2596 /*
2597  * The send routine. Borrows from svc_send
2598  */
2599 static int bc_send_request(struct rpc_task *task)
2600 {
2601         struct rpc_rqst *req = task->tk_rqstp;
2602         struct svc_xprt *xprt;
2603         int len;
2604
2605         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2606         /*
2607          * Get the server socket associated with this callback xprt
2608          */
2609         xprt = req->rq_xprt->bc_xprt;
2610
2611         /*
2612          * Grab the mutex to serialize data as the connection is shared
2613          * with the fore channel
2614          */
2615         if (!mutex_trylock(&xprt->xpt_mutex)) {
2616                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2617                 if (!mutex_trylock(&xprt->xpt_mutex))
2618                         return -EAGAIN;
2619                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2620         }
2621         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2622                 len = -ENOTCONN;
2623         else
2624                 len = bc_sendto(req);
2625         mutex_unlock(&xprt->xpt_mutex);
2626
2627         if (len > 0)
2628                 len = 0;
2629
2630         return len;
2631 }
2632
2633 /*
2634  * The close routine. Since this is client initiated, we do nothing
2635  */
2636
2637 static void bc_close(struct rpc_xprt *xprt)
2638 {
2639 }
2640
2641 /*
2642  * The xprt destroy routine. Again, because this connection is client
2643  * initiated, we do nothing
2644  */
2645
2646 static void bc_destroy(struct rpc_xprt *xprt)
2647 {
2648         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2649
2650         xs_xprt_free(xprt);
2651         module_put(THIS_MODULE);
2652 }
2653
2654 static struct rpc_xprt_ops xs_local_ops = {
2655         .reserve_xprt           = xprt_reserve_xprt,
2656         .release_xprt           = xs_tcp_release_xprt,
2657         .alloc_slot             = xprt_alloc_slot,
2658         .rpcbind                = xs_local_rpcbind,
2659         .set_port               = xs_local_set_port,
2660         .connect                = xs_local_connect,
2661         .buf_alloc              = rpc_malloc,
2662         .buf_free               = rpc_free,
2663         .send_request           = xs_local_send_request,
2664         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2665         .close                  = xs_close,
2666         .destroy                = xs_destroy,
2667         .print_stats            = xs_local_print_stats,
2668         .enable_swap            = xs_enable_swap,
2669         .disable_swap           = xs_disable_swap,
2670 };
2671
2672 static struct rpc_xprt_ops xs_udp_ops = {
2673         .set_buffer_size        = xs_udp_set_buffer_size,
2674         .reserve_xprt           = xprt_reserve_xprt_cong,
2675         .release_xprt           = xprt_release_xprt_cong,
2676         .alloc_slot             = xprt_alloc_slot,
2677         .rpcbind                = rpcb_getport_async,
2678         .set_port               = xs_set_port,
2679         .connect                = xs_connect,
2680         .buf_alloc              = rpc_malloc,
2681         .buf_free               = rpc_free,
2682         .send_request           = xs_udp_send_request,
2683         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2684         .timer                  = xs_udp_timer,
2685         .release_request        = xprt_release_rqst_cong,
2686         .close                  = xs_close,
2687         .destroy                = xs_destroy,
2688         .print_stats            = xs_udp_print_stats,
2689         .enable_swap            = xs_enable_swap,
2690         .disable_swap           = xs_disable_swap,
2691         .inject_disconnect      = xs_inject_disconnect,
2692 };
2693
2694 static struct rpc_xprt_ops xs_tcp_ops = {
2695         .reserve_xprt           = xprt_reserve_xprt,
2696         .release_xprt           = xs_tcp_release_xprt,
2697         .alloc_slot             = xprt_lock_and_alloc_slot,
2698         .rpcbind                = rpcb_getport_async,
2699         .set_port               = xs_set_port,
2700         .connect                = xs_connect,
2701         .buf_alloc              = rpc_malloc,
2702         .buf_free               = rpc_free,
2703         .send_request           = xs_tcp_send_request,
2704         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2705         .close                  = xs_tcp_shutdown,
2706         .destroy                = xs_destroy,
2707         .print_stats            = xs_tcp_print_stats,
2708         .enable_swap            = xs_enable_swap,
2709         .disable_swap           = xs_disable_swap,
2710         .inject_disconnect      = xs_inject_disconnect,
2711 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2712         .bc_setup               = xprt_setup_bc,
2713         .bc_up                  = xs_tcp_bc_up,
2714         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2715         .bc_free_rqst           = xprt_free_bc_rqst,
2716         .bc_destroy             = xprt_destroy_bc,
2717 #endif
2718 };
2719
2720 /*
2721  * The rpc_xprt_ops for the server backchannel
2722  */
2723
2724 static struct rpc_xprt_ops bc_tcp_ops = {
2725         .reserve_xprt           = xprt_reserve_xprt,
2726         .release_xprt           = xprt_release_xprt,
2727         .alloc_slot             = xprt_alloc_slot,
2728         .buf_alloc              = bc_malloc,
2729         .buf_free               = bc_free,
2730         .send_request           = bc_send_request,
2731         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2732         .close                  = bc_close,
2733         .destroy                = bc_destroy,
2734         .print_stats            = xs_tcp_print_stats,
2735         .enable_swap            = xs_enable_swap,
2736         .disable_swap           = xs_disable_swap,
2737         .inject_disconnect      = xs_inject_disconnect,
2738 };
2739
2740 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2741 {
2742         static const struct sockaddr_in sin = {
2743                 .sin_family             = AF_INET,
2744                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2745         };
2746         static const struct sockaddr_in6 sin6 = {
2747                 .sin6_family            = AF_INET6,
2748                 .sin6_addr              = IN6ADDR_ANY_INIT,
2749         };
2750
2751         switch (family) {
2752         case AF_LOCAL:
2753                 break;
2754         case AF_INET:
2755                 memcpy(sap, &sin, sizeof(sin));
2756                 break;
2757         case AF_INET6:
2758                 memcpy(sap, &sin6, sizeof(sin6));
2759                 break;
2760         default:
2761                 dprintk("RPC:       %s: Bad address family\n", __func__);
2762                 return -EAFNOSUPPORT;
2763         }
2764         return 0;
2765 }
2766
2767 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2768                                       unsigned int slot_table_size,
2769                                       unsigned int max_slot_table_size)
2770 {
2771         struct rpc_xprt *xprt;
2772         struct sock_xprt *new;
2773
2774         if (args->addrlen > sizeof(xprt->addr)) {
2775                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2776                 return ERR_PTR(-EBADF);
2777         }
2778
2779         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2780                         max_slot_table_size);
2781         if (xprt == NULL) {
2782                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2783                                 "rpc_xprt\n");
2784                 return ERR_PTR(-ENOMEM);
2785         }
2786
2787         new = container_of(xprt, struct sock_xprt, xprt);
2788         mutex_init(&new->recv_mutex);
2789         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2790         xprt->addrlen = args->addrlen;
2791         if (args->srcaddr)
2792                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2793         else {
2794                 int err;
2795                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2796                                         (struct sockaddr *)&new->srcaddr);
2797                 if (err != 0) {
2798                         xprt_free(xprt);
2799                         return ERR_PTR(err);
2800                 }
2801         }
2802
2803         return xprt;
2804 }
2805
2806 static const struct rpc_timeout xs_local_default_timeout = {
2807         .to_initval = 10 * HZ,
2808         .to_maxval = 10 * HZ,
2809         .to_retries = 2,
2810 };
2811
2812 /**
2813  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2814  * @args: rpc transport creation arguments
2815  *
2816  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2817  */
2818 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2819 {
2820         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2821         struct sock_xprt *transport;
2822         struct rpc_xprt *xprt;
2823         struct rpc_xprt *ret;
2824
2825         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2826                         xprt_max_tcp_slot_table_entries);
2827         if (IS_ERR(xprt))
2828                 return xprt;
2829         transport = container_of(xprt, struct sock_xprt, xprt);
2830
2831         xprt->prot = 0;
2832         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2833         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2834
2835         xprt->bind_timeout = XS_BIND_TO;
2836         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2837         xprt->idle_timeout = XS_IDLE_DISC_TO;
2838
2839         xprt->ops = &xs_local_ops;
2840         xprt->timeout = &xs_local_default_timeout;
2841
2842         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2843         INIT_DELAYED_WORK(&transport->connect_worker,
2844                         xs_dummy_setup_socket);
2845
2846         switch (sun->sun_family) {
2847         case AF_LOCAL:
2848                 if (sun->sun_path[0] != '/') {
2849                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2850                                         sun->sun_path);
2851                         ret = ERR_PTR(-EINVAL);
2852                         goto out_err;
2853                 }
2854                 xprt_set_bound(xprt);
2855                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2856                 ret = ERR_PTR(xs_local_setup_socket(transport));
2857                 if (ret)
2858                         goto out_err;
2859                 break;
2860         default:
2861                 ret = ERR_PTR(-EAFNOSUPPORT);
2862                 goto out_err;
2863         }
2864
2865         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2866                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2867
2868         if (try_module_get(THIS_MODULE))
2869                 return xprt;
2870         ret = ERR_PTR(-EINVAL);
2871 out_err:
2872         xs_xprt_free(xprt);
2873         return ret;
2874 }
2875
2876 static const struct rpc_timeout xs_udp_default_timeout = {
2877         .to_initval = 5 * HZ,
2878         .to_maxval = 30 * HZ,
2879         .to_increment = 5 * HZ,
2880         .to_retries = 5,
2881 };
2882
2883 /**
2884  * xs_setup_udp - Set up transport to use a UDP socket
2885  * @args: rpc transport creation arguments
2886  *
2887  */
2888 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2889 {
2890         struct sockaddr *addr = args->dstaddr;
2891         struct rpc_xprt *xprt;
2892         struct sock_xprt *transport;
2893         struct rpc_xprt *ret;
2894
2895         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2896                         xprt_udp_slot_table_entries);
2897         if (IS_ERR(xprt))
2898                 return xprt;
2899         transport = container_of(xprt, struct sock_xprt, xprt);
2900
2901         xprt->prot = IPPROTO_UDP;
2902         xprt->tsh_size = 0;
2903         /* XXX: header size can vary due to auth type, IPv6, etc. */
2904         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2905
2906         xprt->bind_timeout = XS_BIND_TO;
2907         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2908         xprt->idle_timeout = XS_IDLE_DISC_TO;
2909
2910         xprt->ops = &xs_udp_ops;
2911
2912         xprt->timeout = &xs_udp_default_timeout;
2913
2914         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2915         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2916
2917         switch (addr->sa_family) {
2918         case AF_INET:
2919                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2920                         xprt_set_bound(xprt);
2921
2922                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2923                 break;
2924         case AF_INET6:
2925                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2926                         xprt_set_bound(xprt);
2927
2928                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2929                 break;
2930         default:
2931                 ret = ERR_PTR(-EAFNOSUPPORT);
2932                 goto out_err;
2933         }
2934
2935         if (xprt_bound(xprt))
2936                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2937                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2938                                 xprt->address_strings[RPC_DISPLAY_PORT],
2939                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2940         else
2941                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2942                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2943                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2944
2945         if (try_module_get(THIS_MODULE))
2946                 return xprt;
2947         ret = ERR_PTR(-EINVAL);
2948 out_err:
2949         xs_xprt_free(xprt);
2950         return ret;
2951 }
2952
2953 static const struct rpc_timeout xs_tcp_default_timeout = {
2954         .to_initval = 60 * HZ,
2955         .to_maxval = 60 * HZ,
2956         .to_retries = 2,
2957 };
2958
2959 /**
2960  * xs_setup_tcp - Set up transport to use a TCP socket
2961  * @args: rpc transport creation arguments
2962  *
2963  */
2964 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2965 {
2966         struct sockaddr *addr = args->dstaddr;
2967         struct rpc_xprt *xprt;
2968         struct sock_xprt *transport;
2969         struct rpc_xprt *ret;
2970         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2971
2972         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2973                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2974
2975         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2976                         max_slot_table_size);
2977         if (IS_ERR(xprt))
2978                 return xprt;
2979         transport = container_of(xprt, struct sock_xprt, xprt);
2980
2981         xprt->prot = IPPROTO_TCP;
2982         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2983         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2984
2985         xprt->bind_timeout = XS_BIND_TO;
2986         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2987         xprt->idle_timeout = XS_IDLE_DISC_TO;
2988
2989         xprt->ops = &xs_tcp_ops;
2990         xprt->timeout = &xs_tcp_default_timeout;
2991
2992         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2993         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2994
2995         switch (addr->sa_family) {
2996         case AF_INET:
2997                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2998                         xprt_set_bound(xprt);
2999
3000                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3001                 break;
3002         case AF_INET6:
3003                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3004                         xprt_set_bound(xprt);
3005
3006                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3007                 break;
3008         default:
3009                 ret = ERR_PTR(-EAFNOSUPPORT);
3010                 goto out_err;
3011         }
3012
3013         if (xprt_bound(xprt))
3014                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3015                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3016                                 xprt->address_strings[RPC_DISPLAY_PORT],
3017                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3018         else
3019                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3020                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3021                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3022
3023         if (try_module_get(THIS_MODULE))
3024                 return xprt;
3025         ret = ERR_PTR(-EINVAL);
3026 out_err:
3027         xs_xprt_free(xprt);
3028         return ret;
3029 }
3030
3031 /**
3032  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3033  * @args: rpc transport creation arguments
3034  *
3035  */
3036 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3037 {
3038         struct sockaddr *addr = args->dstaddr;
3039         struct rpc_xprt *xprt;
3040         struct sock_xprt *transport;
3041         struct svc_sock *bc_sock;
3042         struct rpc_xprt *ret;
3043
3044         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3045                         xprt_tcp_slot_table_entries);
3046         if (IS_ERR(xprt))
3047                 return xprt;
3048         transport = container_of(xprt, struct sock_xprt, xprt);
3049
3050         xprt->prot = IPPROTO_TCP;
3051         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3052         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3053         xprt->timeout = &xs_tcp_default_timeout;
3054
3055         /* backchannel */
3056         xprt_set_bound(xprt);
3057         xprt->bind_timeout = 0;
3058         xprt->reestablish_timeout = 0;
3059         xprt->idle_timeout = 0;
3060
3061         xprt->ops = &bc_tcp_ops;
3062
3063         switch (addr->sa_family) {
3064         case AF_INET:
3065                 xs_format_peer_addresses(xprt, "tcp",
3066                                          RPCBIND_NETID_TCP);
3067                 break;
3068         case AF_INET6:
3069                 xs_format_peer_addresses(xprt, "tcp",
3070                                    RPCBIND_NETID_TCP6);
3071                 break;
3072         default:
3073                 ret = ERR_PTR(-EAFNOSUPPORT);
3074                 goto out_err;
3075         }
3076
3077         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3078                         xprt->address_strings[RPC_DISPLAY_ADDR],
3079                         xprt->address_strings[RPC_DISPLAY_PORT],
3080                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3081
3082         /*
3083          * Once we've associated a backchannel xprt with a connection,
3084          * we want to keep it around as long as the connection lasts,
3085          * in case we need to start using it for a backchannel again;
3086          * this reference won't be dropped until bc_xprt is destroyed.
3087          */
3088         xprt_get(xprt);
3089         args->bc_xprt->xpt_bc_xprt = xprt;
3090         xprt->bc_xprt = args->bc_xprt;
3091         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3092         transport->sock = bc_sock->sk_sock;
3093         transport->inet = bc_sock->sk_sk;
3094
3095         /*
3096          * Since we don't want connections for the backchannel, we set
3097          * the xprt status to connected
3098          */
3099         xprt_set_connected(xprt);
3100
3101         if (try_module_get(THIS_MODULE))
3102                 return xprt;
3103
3104         args->bc_xprt->xpt_bc_xprt = NULL;
3105         args->bc_xprt->xpt_bc_xps = NULL;
3106         xprt_put(xprt);
3107         ret = ERR_PTR(-EINVAL);
3108 out_err:
3109         xs_xprt_free(xprt);
3110         return ret;
3111 }
3112
3113 static struct xprt_class        xs_local_transport = {
3114         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3115         .name           = "named UNIX socket",
3116         .owner          = THIS_MODULE,
3117         .ident          = XPRT_TRANSPORT_LOCAL,
3118         .setup          = xs_setup_local,
3119 };
3120
3121 static struct xprt_class        xs_udp_transport = {
3122         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3123         .name           = "udp",
3124         .owner          = THIS_MODULE,
3125         .ident          = XPRT_TRANSPORT_UDP,
3126         .setup          = xs_setup_udp,
3127 };
3128
3129 static struct xprt_class        xs_tcp_transport = {
3130         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3131         .name           = "tcp",
3132         .owner          = THIS_MODULE,
3133         .ident          = XPRT_TRANSPORT_TCP,
3134         .setup          = xs_setup_tcp,
3135 };
3136
3137 static struct xprt_class        xs_bc_tcp_transport = {
3138         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3139         .name           = "tcp NFSv4.1 backchannel",
3140         .owner          = THIS_MODULE,
3141         .ident          = XPRT_TRANSPORT_BC_TCP,
3142         .setup          = xs_setup_bc_tcp,
3143 };
3144
3145 /**
3146  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3147  *
3148  */
3149 int init_socket_xprt(void)
3150 {
3151 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3152         if (!sunrpc_table_header)
3153                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3154 #endif
3155
3156         xprt_register_transport(&xs_local_transport);
3157         xprt_register_transport(&xs_udp_transport);
3158         xprt_register_transport(&xs_tcp_transport);
3159         xprt_register_transport(&xs_bc_tcp_transport);
3160
3161         return 0;
3162 }
3163
3164 /**
3165  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3166  *
3167  */
3168 void cleanup_socket_xprt(void)
3169 {
3170 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3171         if (sunrpc_table_header) {
3172                 unregister_sysctl_table(sunrpc_table_header);
3173                 sunrpc_table_header = NULL;
3174         }
3175 #endif
3176
3177         xprt_unregister_transport(&xs_local_transport);
3178         xprt_unregister_transport(&xs_udp_transport);
3179         xprt_unregister_transport(&xs_tcp_transport);
3180         xprt_unregister_transport(&xs_bc_tcp_transport);
3181 }
3182
3183 static int param_set_uint_minmax(const char *val,
3184                 const struct kernel_param *kp,
3185                 unsigned int min, unsigned int max)
3186 {
3187         unsigned int num;
3188         int ret;
3189
3190         if (!val)
3191                 return -EINVAL;
3192         ret = kstrtouint(val, 0, &num);
3193         if (ret == -EINVAL || num < min || num > max)
3194                 return -EINVAL;
3195         *((unsigned int *)kp->arg) = num;
3196         return 0;
3197 }
3198
3199 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3200 {
3201         if (kp->arg == &xprt_min_resvport)
3202                 return param_set_uint_minmax(val, kp,
3203                         RPC_MIN_RESVPORT,
3204                         xprt_max_resvport);
3205         return param_set_uint_minmax(val, kp,
3206                         xprt_min_resvport,
3207                         RPC_MAX_RESVPORT);
3208 }
3209
3210 static const struct kernel_param_ops param_ops_portnr = {
3211         .set = param_set_portnr,
3212         .get = param_get_uint,
3213 };
3214
3215 #define param_check_portnr(name, p) \
3216         __param_check(name, p, unsigned int);
3217
3218 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3219 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3220
3221 static int param_set_slot_table_size(const char *val,
3222                                      const struct kernel_param *kp)
3223 {
3224         return param_set_uint_minmax(val, kp,
3225                         RPC_MIN_SLOT_TABLE,
3226                         RPC_MAX_SLOT_TABLE);
3227 }
3228
3229 static const struct kernel_param_ops param_ops_slot_table_size = {
3230         .set = param_set_slot_table_size,
3231         .get = param_get_uint,
3232 };
3233
3234 #define param_check_slot_table_size(name, p) \
3235         __param_check(name, p, unsigned int);
3236
3237 static int param_set_max_slot_table_size(const char *val,
3238                                      const struct kernel_param *kp)
3239 {
3240         return param_set_uint_minmax(val, kp,
3241                         RPC_MIN_SLOT_TABLE,
3242                         RPC_MAX_SLOT_TABLE_LIMIT);
3243 }
3244
3245 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3246         .set = param_set_max_slot_table_size,
3247         .get = param_get_uint,
3248 };
3249
3250 #define param_check_max_slot_table_size(name, p) \
3251         __param_check(name, p, unsigned int);
3252
3253 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3254                    slot_table_size, 0644);
3255 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3256                    max_slot_table_size, 0644);
3257 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3258                    slot_table_size, 0644);
3259