net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68 #define XS_TCP_LINGER_TO        (15U * HZ)
  69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71 /*
  72  * We can register our own files under /proc/sys/sunrpc by
  73  * calling register_sysctl_table() again.  The files in that
  74  * directory become the union of all files registered there.
  75  *
  76  * We simply need to make sure that we don't collide with
  77  * someone else's file names!
  78  */
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 181
 182 /*
 183  * TCP idle timeout; client drops the transport socket if it is idle
 184  * for this long.  Note that we also timeout UDP sockets to prevent
 185  * holding port numbers when there is no RPC traffic.
 186  */
 187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 188
 189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 190 # undef  RPC_DEBUG_DATA
 191 # define RPCDBG_FACILITY        RPCDBG_TRANS
 192 #endif
 193
 194 #ifdef RPC_DEBUG_DATA
 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196 {
 197         u8 *buf = (u8 *) packet;
 198         int j;
 199
 200         dprintk("RPC:       %s\n", msg);
 201         for (j = 0; j < count && j < 128; j += 4) {
 202                 if (!(j & 31)) {
 203                         if (j)
 204                                 dprintk("\n");
 205                         dprintk("0x%04x ", j);
 206                 }
 207                 dprintk("%02x%02x%02x%02x ",
 208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209         }
 210         dprintk("\n");
 211 }
 212 #else
 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214 {
 215         /* NOP */
 216 }
 217 #endif
 218
 219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 220 {
 221         return (struct rpc_xprt *) sk->sk_user_data;
 222 }
 223
 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 225 {
 226         return (struct sockaddr *) &xprt->addr;
 227 }
 228
 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 230 {
 231         return (struct sockaddr_un *) &xprt->addr;
 232 }
 233
 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr_in *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_in6 *) &xprt->addr;
 242 }
 243
 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 245 {
 246         struct sockaddr *sap = xs_addr(xprt);
 247         struct sockaddr_in6 *sin6;
 248         struct sockaddr_in *sin;
 249         struct sockaddr_un *sun;
 250         char buf[128];
 251
 252         switch (sap->sa_family) {
 253         case AF_LOCAL:
 254                 sun = xs_addr_un(xprt);
 255                 strlcpy(buf, sun->sun_path, sizeof(buf));
 256                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 257                                                 kstrdup(buf, GFP_KERNEL);
 258                 break;
 259         case AF_INET:
 260                 (void)rpc_ntop(sap, buf, sizeof(buf));
 261                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 262                                                 kstrdup(buf, GFP_KERNEL);
 263                 sin = xs_addr_in(xprt);
 264                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 265                 break;
 266         case AF_INET6:
 267                 (void)rpc_ntop(sap, buf, sizeof(buf));
 268                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 269                                                 kstrdup(buf, GFP_KERNEL);
 270                 sin6 = xs_addr_in6(xprt);
 271                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 272                 break;
 273         default:
 274                 BUG();
 275         }
 276
 277         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 278 }
 279
 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 281 {
 282         struct sockaddr *sap = xs_addr(xprt);
 283         char buf[128];
 284
 285         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 286         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 287
 288         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 289         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 290 }
 291
 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 293                                      const char *protocol,
 294                                      const char *netid)
 295 {
 296         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 297         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 298         xs_format_common_peer_addresses(xprt);
 299         xs_format_common_peer_ports(xprt);
 300 }
 301
 302 static void xs_update_peer_port(struct rpc_xprt *xprt)
 303 {
 304         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 305         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 306
 307         xs_format_common_peer_ports(xprt);
 308 }
 309
 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 311 {
 312         unsigned int i;
 313
 314         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 315                 switch (i) {
 316                 case RPC_DISPLAY_PROTO:
 317                 case RPC_DISPLAY_NETID:
 318                         continue;
 319                 default:
 320                         kfree(xprt->address_strings[i]);
 321                 }
 322 }
 323
 324 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 325
 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 327 {
 328         struct msghdr msg = {
 329                 .msg_name       = addr,
 330                 .msg_namelen    = addrlen,
 331                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 332         };
 333         struct kvec iov = {
 334                 .iov_base       = vec->iov_base + base,
 335                 .iov_len        = vec->iov_len - base,
 336         };
 337
 338         if (iov.iov_len != 0)
 339                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 340         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 341 }
 342
 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 344 {
 345         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 346                         int offset, size_t size, int flags);
 347         struct page **ppage;
 348         unsigned int remainder;
 349         int err;
 350
 351         remainder = xdr->page_len - base;
 352         base += xdr->page_base;
 353         ppage = xdr->pages + (base >> PAGE_SHIFT);
 354         base &= ~PAGE_MASK;
 355         do_sendpage = sock->ops->sendpage;
 356         if (!zerocopy)
 357                 do_sendpage = sock_no_sendpage;
 358         for(;;) {
 359                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 360                 int flags = XS_SENDMSG_FLAGS;
 361
 362                 remainder -= len;
 363                 if (remainder != 0 || more)
 364                         flags |= MSG_MORE;
 365                 err = do_sendpage(sock, *ppage, base, len, flags);
 366                 if (remainder == 0 || err != len)
 367                         break;
 368                 *sent_p += err;
 369                 ppage++;
 370                 base = 0;
 371         }
 372         if (err > 0) {
 373                 *sent_p += err;
 374                 err = 0;
 375         }
 376         return err;
 377 }
 378
 379 /**
 380  * xs_sendpages - write pages directly to a socket
 381  * @sock: socket to send on
 382  * @addr: UDP only -- address of destination
 383  * @addrlen: UDP only -- length of destination address
 384  * @xdr: buffer containing this request
 385  * @base: starting position in the buffer
 386  * @zerocopy: true if it is safe to use sendpage()
 387  * @sent_p: return the total number of bytes successfully queued for sending
 388  *
 389  */
 390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 391 {
 392         unsigned int remainder = xdr->len - base;
 393         int err = 0;
 394         int sent = 0;
 395
 396         if (unlikely(!sock))
 397                 return -ENOTSOCK;
 398
 399         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 400         if (base != 0) {
 401                 addr = NULL;
 402                 addrlen = 0;
 403         }
 404
 405         if (base < xdr->head[0].iov_len || addr != NULL) {
 406                 unsigned int len = xdr->head[0].iov_len - base;
 407                 remainder -= len;
 408                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 409                 if (remainder == 0 || err != len)
 410                         goto out;
 411                 *sent_p += err;
 412                 base = 0;
 413         } else
 414                 base -= xdr->head[0].iov_len;
 415
 416         if (base < xdr->page_len) {
 417                 unsigned int len = xdr->page_len - base;
 418                 remainder -= len;
 419                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 420                 *sent_p += sent;
 421                 if (remainder == 0 || sent != len)
 422                         goto out;
 423                 base = 0;
 424         } else
 425                 base -= xdr->page_len;
 426
 427         if (base >= xdr->tail[0].iov_len)
 428                 return 0;
 429         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 430 out:
 431         if (err > 0) {
 432                 *sent_p += err;
 433                 err = 0;
 434         }
 435         return err;
 436 }
 437
 438 static void xs_nospace_callback(struct rpc_task *task)
 439 {
 440         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 441
 442         transport->inet->sk_write_pending--;
 443         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 444 }
 445
 446 /**
 447  * xs_nospace - place task on wait queue if transmit was incomplete
 448  * @task: task to put to sleep
 449  *
 450  */
 451 static int xs_nospace(struct rpc_task *task)
 452 {
 453         struct rpc_rqst *req = task->tk_rqstp;
 454         struct rpc_xprt *xprt = req->rq_xprt;
 455         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 456         struct sock *sk = transport->inet;
 457         int ret = -EAGAIN;
 458
 459         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 460                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 461                         req->rq_slen);
 462
 463         /* Protect against races with write_space */
 464         spin_lock_bh(&xprt->transport_lock);
 465
 466         /* Don't race with disconnect */
 467         if (xprt_connected(xprt)) {
 468                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 469                         /*
 470                          * Notify TCP that we're limited by the application
 471                          * window size
 472                          */
 473                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 474                         sk->sk_write_pending++;
 475                         /* ...and wait for more buffer space */
 476                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 477                 }
 478         } else {
 479                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 480                 ret = -ENOTCONN;
 481         }
 482
 483         spin_unlock_bh(&xprt->transport_lock);
 484
 485         /* Race breaker in case memory is freed before above code is called */
 486         sk->sk_write_space(sk);
 487         return ret;
 488 }
 489
 490 /*
 491  * Construct a stream transport record marker in @buf.
 492  */
 493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 494 {
 495         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 496         rpc_fraghdr *base = buf->head[0].iov_base;
 497         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 498 }
 499
 500 /**
 501  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 502  * @task: RPC task that manages the state of an RPC request
 503  *
 504  * Return values:
 505  *        0:    The request has been sent
 506  *   EAGAIN:    The socket was blocked, please call again later to
 507  *              complete the request
 508  * ENOTCONN:    Caller needs to invoke connect logic then call again
 509  *    other:    Some other error occured, the request was not sent
 510  */
 511 static int xs_local_send_request(struct rpc_task *task)
 512 {
 513         struct rpc_rqst *req = task->tk_rqstp;
 514         struct rpc_xprt *xprt = req->rq_xprt;
 515         struct sock_xprt *transport =
 516                                 container_of(xprt, struct sock_xprt, xprt);
 517         struct xdr_buf *xdr = &req->rq_snd_buf;
 518         int status;
 519         int sent = 0;
 520
 521         xs_encode_stream_record_marker(&req->rq_snd_buf);
 522
 523         xs_pktdump("packet data:",
 524                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 525
 526         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 527                               true, &sent);
 528         dprintk("RPC:       %s(%u) = %d\n",
 529                         __func__, xdr->len - req->rq_bytes_sent, status);
 530
 531         if (status == -EAGAIN && sock_writeable(transport->inet))
 532                 status = -ENOBUFS;
 533
 534         if (likely(sent > 0) || status == 0) {
 535                 req->rq_bytes_sent += sent;
 536                 req->rq_xmit_bytes_sent += sent;
 537                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 538                         req->rq_bytes_sent = 0;
 539                         return 0;
 540                 }
 541                 status = -EAGAIN;
 542         }
 543
 544         switch (status) {
 545         case -ENOBUFS:
 546                 break;
 547         case -EAGAIN:
 548                 status = xs_nospace(task);
 549                 break;
 550         default:
 551                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 552                         -status);
 553         case -EPIPE:
 554                 xs_close(xprt);
 555                 status = -ENOTCONN;
 556         }
 557
 558         return status;
 559 }
 560
 561 /**
 562  * xs_udp_send_request - write an RPC request to a UDP socket
 563  * @task: address of RPC task that manages the state of an RPC request
 564  *
 565  * Return values:
 566  *        0:    The request has been sent
 567  *   EAGAIN:    The socket was blocked, please call again later to
 568  *              complete the request
 569  * ENOTCONN:    Caller needs to invoke connect logic then call again
 570  *    other:    Some other error occurred, the request was not sent
 571  */
 572 static int xs_udp_send_request(struct rpc_task *task)
 573 {
 574         struct rpc_rqst *req = task->tk_rqstp;
 575         struct rpc_xprt *xprt = req->rq_xprt;
 576         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 577         struct xdr_buf *xdr = &req->rq_snd_buf;
 578         int sent = 0;
 579         int status;
 580
 581         xs_pktdump("packet data:",
 582                                 req->rq_svec->iov_base,
 583                                 req->rq_svec->iov_len);
 584
 585         if (!xprt_bound(xprt))
 586                 return -ENOTCONN;
 587         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 588                               xdr, req->rq_bytes_sent, true, &sent);
 589
 590         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 591                         xdr->len - req->rq_bytes_sent, status);
 592
 593         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 594         if (status == -EPERM)
 595                 goto process_status;
 596
 597         if (status == -EAGAIN && sock_writeable(transport->inet))
 598                 status = -ENOBUFS;
 599
 600         if (sent > 0 || status == 0) {
 601                 req->rq_xmit_bytes_sent += sent;
 602                 if (sent >= req->rq_slen)
 603                         return 0;
 604                 /* Still some bytes left; set up for a retry later. */
 605                 status = -EAGAIN;
 606         }
 607
 608 process_status:
 609         switch (status) {
 610         case -ENOTSOCK:
 611                 status = -ENOTCONN;
 612                 /* Should we call xs_close() here? */
 613                 break;
 614         case -EAGAIN:
 615                 status = xs_nospace(task);
 616                 break;
 617         default:
 618                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 619                         -status);
 620         case -ENETUNREACH:
 621         case -ENOBUFS:
 622         case -EPIPE:
 623         case -ECONNREFUSED:
 624         case -EPERM:
 625                 /* When the server has died, an ICMP port unreachable message
 626                  * prompts ECONNREFUSED. */
 627                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 628         }
 629
 630         return status;
 631 }
 632
 633 /**
 634  * xs_tcp_send_request - write an RPC request to a TCP socket
 635  * @task: address of RPC task that manages the state of an RPC request
 636  *
 637  * Return values:
 638  *        0:    The request has been sent
 639  *   EAGAIN:    The socket was blocked, please call again later to
 640  *              complete the request
 641  * ENOTCONN:    Caller needs to invoke connect logic then call again
 642  *    other:    Some other error occurred, the request was not sent
 643  *
 644  * XXX: In the case of soft timeouts, should we eventually give up
 645  *      if sendmsg is not able to make progress?
 646  */
 647 static int xs_tcp_send_request(struct rpc_task *task)
 648 {
 649         struct rpc_rqst *req = task->tk_rqstp;
 650         struct rpc_xprt *xprt = req->rq_xprt;
 651         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 652         struct xdr_buf *xdr = &req->rq_snd_buf;
 653         bool zerocopy = true;
 654         int status;
 655         int sent;
 656
 657         xs_encode_stream_record_marker(&req->rq_snd_buf);
 658
 659         xs_pktdump("packet data:",
 660                                 req->rq_svec->iov_base,
 661                                 req->rq_svec->iov_len);
 662         /* Don't use zero copy if this is a resend. If the RPC call
 663          * completes while the socket holds a reference to the pages,
 664          * then we may end up resending corrupted data.
 665          */
 666         if (task->tk_flags & RPC_TASK_SENT)
 667                 zerocopy = false;
 668
 669         /* Continue transmitting the packet/record. We must be careful
 670          * to cope with writespace callbacks arriving _after_ we have
 671          * called sendmsg(). */
 672         while (1) {
 673                 sent = 0;
 674                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 675                                       req->rq_bytes_sent, zerocopy, &sent);
 676
 677                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 678                                 xdr->len - req->rq_bytes_sent, status);
 679
 680                 /* If we've sent the entire packet, immediately
 681                  * reset the count of bytes sent. */
 682                 req->rq_bytes_sent += sent;
 683                 req->rq_xmit_bytes_sent += sent;
 684                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 685                         req->rq_bytes_sent = 0;
 686                         return 0;
 687                 }
 688
 689                 if (status < 0)
 690                         break;
 691                 if (sent == 0) {
 692                         status = -EAGAIN;
 693                         break;
 694                 }
 695         }
 696         if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
 697                 status = -ENOBUFS;
 698
 699         switch (status) {
 700         case -ENOTSOCK:
 701                 status = -ENOTCONN;
 702                 /* Should we call xs_close() here? */
 703                 break;
 704         case -EAGAIN:
 705                 status = xs_nospace(task);
 706                 break;
 707         default:
 708                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 709                         -status);
 710         case -ECONNRESET:
 711         case -ECONNREFUSED:
 712         case -ENOTCONN:
 713         case -EADDRINUSE:
 714         case -ENOBUFS:
 715         case -EPIPE:
 716                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 717         }
 718
 719         return status;
 720 }
 721
 722 /**
 723  * xs_tcp_release_xprt - clean up after a tcp transmission
 724  * @xprt: transport
 725  * @task: rpc task
 726  *
 727  * This cleans up if an error causes us to abort the transmission of a request.
 728  * In this case, the socket may need to be reset in order to avoid confusing
 729  * the server.
 730  */
 731 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 732 {
 733         struct rpc_rqst *req;
 734
 735         if (task != xprt->snd_task)
 736                 return;
 737         if (task == NULL)
 738                 goto out_release;
 739         req = task->tk_rqstp;
 740         if (req == NULL)
 741                 goto out_release;
 742         if (req->rq_bytes_sent == 0)
 743                 goto out_release;
 744         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 745                 goto out_release;
 746         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 747 out_release:
 748         xprt_release_xprt(xprt, task);
 749 }
 750
 751 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 752 {
 753         transport->old_data_ready = sk->sk_data_ready;
 754         transport->old_state_change = sk->sk_state_change;
 755         transport->old_write_space = sk->sk_write_space;
 756         transport->old_error_report = sk->sk_error_report;
 757 }
 758
 759 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 760 {
 761         sk->sk_data_ready = transport->old_data_ready;
 762         sk->sk_state_change = transport->old_state_change;
 763         sk->sk_write_space = transport->old_write_space;
 764         sk->sk_error_report = transport->old_error_report;
 765 }
 766
 767 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 768 {
 769         smp_mb__before_atomic();
 770         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 771         clear_bit(XPRT_CLOSING, &xprt->state);
 772         smp_mb__after_atomic();
 773 }
 774
 775 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 776 {
 777         xs_sock_reset_connection_flags(xprt);
 778         /* Mark transport as closed and wake up all pending tasks */
 779         xprt_disconnect_done(xprt);
 780 }
 781
 782 /**
 783  * xs_error_report - callback to handle TCP socket state errors
 784  * @sk: socket
 785  *
 786  * Note: we don't call sock_error() since there may be a rpc_task
 787  * using the socket, and so we don't want to clear sk->sk_err.
 788  */
 789 static void xs_error_report(struct sock *sk)
 790 {
 791         struct rpc_xprt *xprt;
 792         int err;
 793
 794         read_lock_bh(&sk->sk_callback_lock);
 795         if (!(xprt = xprt_from_sock(sk)))
 796                 goto out;
 797
 798         err = -sk->sk_err;
 799         if (err == 0)
 800                 goto out;
 801         /* Is this a reset event? */
 802         if (sk->sk_state == TCP_CLOSE)
 803                 xs_sock_mark_closed(xprt);
 804         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 805                         xprt, -err);
 806         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 807         xprt_wake_pending_tasks(xprt, err);
 808  out:
 809         read_unlock_bh(&sk->sk_callback_lock);
 810 }
 811
 812 static void xs_reset_transport(struct sock_xprt *transport)
 813 {
 814         struct socket *sock = transport->sock;
 815         struct sock *sk = transport->inet;
 816         struct rpc_xprt *xprt = &transport->xprt;
 817
 818         if (sk == NULL)
 819                 return;
 820
 821         if (atomic_read(&transport->xprt.swapper))
 822                 sk_clear_memalloc(sk);
 823
 824         kernel_sock_shutdown(sock, SHUT_RDWR);
 825
 826         mutex_lock(&transport->recv_mutex);
 827         write_lock_bh(&sk->sk_callback_lock);
 828         transport->inet = NULL;
 829         transport->sock = NULL;
 830
 831         sk->sk_user_data = NULL;
 832
 833         xs_restore_old_callbacks(transport, sk);
 834         xprt_clear_connected(xprt);
 835         write_unlock_bh(&sk->sk_callback_lock);
 836         xs_sock_reset_connection_flags(xprt);
 837         mutex_unlock(&transport->recv_mutex);
 838
 839         trace_rpc_socket_close(xprt, sock);
 840         sock_release(sock);
 841 }
 842
 843 /**
 844  * xs_close - close a socket
 845  * @xprt: transport
 846  *
 847  * This is used when all requests are complete; ie, no DRC state remains
 848  * on the server we want to save.
 849  *
 850  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 851  * xs_reset_transport() zeroing the socket from underneath a writer.
 852  */
 853 static void xs_close(struct rpc_xprt *xprt)
 854 {
 855         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 856
 857         dprintk("RPC:       xs_close xprt %p\n", xprt);
 858
 859         xs_reset_transport(transport);
 860         xprt->reestablish_timeout = 0;
 861
 862         xprt_disconnect_done(xprt);
 863 }
 864
 865 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 866 {
 867         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 868                 xprt);
 869         xprt_disconnect_done(xprt);
 870 }
 871
 872 static void xs_xprt_free(struct rpc_xprt *xprt)
 873 {
 874         xs_free_peer_addresses(xprt);
 875         xprt_free(xprt);
 876 }
 877
 878 /**
 879  * xs_destroy - prepare to shutdown a transport
 880  * @xprt: doomed transport
 881  *
 882  */
 883 static void xs_destroy(struct rpc_xprt *xprt)
 884 {
 885         struct sock_xprt *transport = container_of(xprt,
 886                         struct sock_xprt, xprt);
 887         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 888
 889         cancel_delayed_work_sync(&transport->connect_worker);
 890         xs_close(xprt);
 891         cancel_work_sync(&transport->recv_worker);
 892         xs_xprt_free(xprt);
 893         module_put(THIS_MODULE);
 894 }
 895
 896 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 897 {
 898         struct xdr_skb_reader desc = {
 899                 .skb            = skb,
 900                 .offset         = sizeof(rpc_fraghdr),
 901                 .count          = skb->len - sizeof(rpc_fraghdr),
 902         };
 903
 904         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 905                 return -1;
 906         if (desc.count)
 907                 return -1;
 908         return 0;
 909 }
 910
 911 /**
 912  * xs_local_data_read_skb
 913  * @xprt: transport
 914  * @sk: socket
 915  * @skb: skbuff
 916  *
 917  * Currently this assumes we can read the whole reply in a single gulp.
 918  */
 919 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 920                 struct sock *sk,
 921                 struct sk_buff *skb)
 922 {
 923         struct rpc_task *task;
 924         struct rpc_rqst *rovr;
 925         int repsize, copied;
 926         u32 _xid;
 927         __be32 *xp;
 928
 929         repsize = skb->len - sizeof(rpc_fraghdr);
 930         if (repsize < 4) {
 931                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 932                 return;
 933         }
 934
 935         /* Copy the XID from the skb... */
 936         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 937         if (xp == NULL)
 938                 return;
 939
 940         /* Look up and lock the request corresponding to the given XID */
 941         spin_lock_bh(&xprt->transport_lock);
 942         rovr = xprt_lookup_rqst(xprt, *xp);
 943         if (!rovr)
 944                 goto out_unlock;
 945         task = rovr->rq_task;
 946
 947         copied = rovr->rq_private_buf.buflen;
 948         if (copied > repsize)
 949                 copied = repsize;
 950
 951         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 952                 dprintk("RPC:       sk_buff copy failed\n");
 953                 goto out_unlock;
 954         }
 955
 956         xprt_complete_rqst(task, copied);
 957
 958  out_unlock:
 959         spin_unlock_bh(&xprt->transport_lock);
 960 }
 961
 962 static void xs_local_data_receive(struct sock_xprt *transport)
 963 {
 964         struct sk_buff *skb;
 965         struct sock *sk;
 966         int err;
 967
 968         mutex_lock(&transport->recv_mutex);
 969         sk = transport->inet;
 970         if (sk == NULL)
 971                 goto out;
 972         for (;;) {
 973                 skb = skb_recv_datagram(sk, 0, 1, &err);
 974                 if (skb == NULL)
 975                         break;
 976                 xs_local_data_read_skb(&transport->xprt, sk, skb);
 977                 skb_free_datagram(sk, skb);
 978         }
 979 out:
 980         mutex_unlock(&transport->recv_mutex);
 981 }
 982
 983 static void xs_local_data_receive_workfn(struct work_struct *work)
 984 {
 985         struct sock_xprt *transport =
 986                 container_of(work, struct sock_xprt, recv_worker);
 987         xs_local_data_receive(transport);
 988 }
 989
 990 /**
 991  * xs_udp_data_read_skb - receive callback for UDP sockets
 992  * @xprt: transport
 993  * @sk: socket
 994  * @skb: skbuff
 995  *
 996  */
 997 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
 998                 struct sock *sk,
 999                 struct sk_buff *skb)
1000 {
1001         struct rpc_task *task;
1002         struct rpc_rqst *rovr;
1003         int repsize, copied;
1004         u32 _xid;
1005         __be32 *xp;
1006
1007         repsize = skb->len - sizeof(struct udphdr);
1008         if (repsize < 4) {
1009                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1010                 return;
1011         }
1012
1013         /* Copy the XID from the skb... */
1014         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1015                                 sizeof(_xid), &_xid);
1016         if (xp == NULL)
1017                 return;
1018
1019         /* Look up and lock the request corresponding to the given XID */
1020         spin_lock_bh(&xprt->transport_lock);
1021         rovr = xprt_lookup_rqst(xprt, *xp);
1022         if (!rovr)
1023                 goto out_unlock;
1024         task = rovr->rq_task;
1025
1026         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1027                 copied = repsize;
1028
1029         /* Suck it into the iovec, verify checksum if not done by hw. */
1030         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1031                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1032                 goto out_unlock;
1033         }
1034
1035         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1036
1037         xprt_adjust_cwnd(xprt, task, copied);
1038         xprt_complete_rqst(task, copied);
1039
1040  out_unlock:
1041         spin_unlock_bh(&xprt->transport_lock);
1042 }
1043
1044 static void xs_udp_data_receive(struct sock_xprt *transport)
1045 {
1046         struct sk_buff *skb;
1047         struct sock *sk;
1048         int err;
1049
1050         mutex_lock(&transport->recv_mutex);
1051         sk = transport->inet;
1052         if (sk == NULL)
1053                 goto out;
1054         for (;;) {
1055                 skb = skb_recv_datagram(sk, 0, 1, &err);
1056                 if (skb == NULL)
1057                         break;
1058                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1059                 skb_free_datagram(sk, skb);
1060         }
1061 out:
1062         mutex_unlock(&transport->recv_mutex);
1063 }
1064
1065 static void xs_udp_data_receive_workfn(struct work_struct *work)
1066 {
1067         struct sock_xprt *transport =
1068                 container_of(work, struct sock_xprt, recv_worker);
1069         xs_udp_data_receive(transport);
1070 }
1071
1072 /**
1073  * xs_data_ready - "data ready" callback for UDP sockets
1074  * @sk: socket with data to read
1075  *
1076  */
1077 static void xs_data_ready(struct sock *sk)
1078 {
1079         struct rpc_xprt *xprt;
1080
1081         read_lock_bh(&sk->sk_callback_lock);
1082         dprintk("RPC:       xs_data_ready...\n");
1083         xprt = xprt_from_sock(sk);
1084         if (xprt != NULL) {
1085                 struct sock_xprt *transport = container_of(xprt,
1086                                 struct sock_xprt, xprt);
1087                 queue_work(rpciod_workqueue, &transport->recv_worker);
1088         }
1089         read_unlock_bh(&sk->sk_callback_lock);
1090 }
1091
1092 /*
1093  * Helper function to force a TCP close if the server is sending
1094  * junk and/or it has put us in CLOSE_WAIT
1095  */
1096 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1097 {
1098         xprt_force_disconnect(xprt);
1099 }
1100
1101 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1102 {
1103         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1104         size_t len, used;
1105         char *p;
1106
1107         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1108         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1109         used = xdr_skb_read_bits(desc, p, len);
1110         transport->tcp_offset += used;
1111         if (used != len)
1112                 return;
1113
1114         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1115         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1116                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1117         else
1118                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1119         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1120
1121         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1122         transport->tcp_offset = 0;
1123
1124         /* Sanity check of the record length */
1125         if (unlikely(transport->tcp_reclen < 8)) {
1126                 dprintk("RPC:       invalid TCP record fragment length\n");
1127                 xs_tcp_force_close(xprt);
1128                 return;
1129         }
1130         dprintk("RPC:       reading TCP record fragment of length %d\n",
1131                         transport->tcp_reclen);
1132 }
1133
1134 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1135 {
1136         if (transport->tcp_offset == transport->tcp_reclen) {
1137                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1138                 transport->tcp_offset = 0;
1139                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1140                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1141                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1142                         transport->tcp_copied = 0;
1143                 }
1144         }
1145 }
1146
1147 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1148 {
1149         size_t len, used;
1150         char *p;
1151
1152         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1153         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1154         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1155         used = xdr_skb_read_bits(desc, p, len);
1156         transport->tcp_offset += used;
1157         if (used != len)
1158                 return;
1159         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1160         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1161         transport->tcp_copied = 4;
1162         dprintk("RPC:       reading %s XID %08x\n",
1163                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1164                                                               : "request with",
1165                         ntohl(transport->tcp_xid));
1166         xs_tcp_check_fraghdr(transport);
1167 }
1168
1169 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1170                                        struct xdr_skb_reader *desc)
1171 {
1172         size_t len, used;
1173         u32 offset;
1174         char *p;
1175
1176         /*
1177          * We want transport->tcp_offset to be 8 at the end of this routine
1178          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1179          * When this function is called for the first time,
1180          * transport->tcp_offset is 4 (after having already read the xid).
1181          */
1182         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1183         len = sizeof(transport->tcp_calldir) - offset;
1184         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1185         p = ((char *) &transport->tcp_calldir) + offset;
1186         used = xdr_skb_read_bits(desc, p, len);
1187         transport->tcp_offset += used;
1188         if (used != len)
1189                 return;
1190         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1191         /*
1192          * We don't yet have the XDR buffer, so we will write the calldir
1193          * out after we get the buffer from the 'struct rpc_rqst'
1194          */
1195         switch (ntohl(transport->tcp_calldir)) {
1196         case RPC_REPLY:
1197                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1198                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1199                 transport->tcp_flags |= TCP_RPC_REPLY;
1200                 break;
1201         case RPC_CALL:
1202                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1203                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1204                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1205                 break;
1206         default:
1207                 dprintk("RPC:       invalid request message type\n");
1208                 xs_tcp_force_close(&transport->xprt);
1209         }
1210         xs_tcp_check_fraghdr(transport);
1211 }
1212
1213 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1214                                      struct xdr_skb_reader *desc,
1215                                      struct rpc_rqst *req)
1216 {
1217         struct sock_xprt *transport =
1218                                 container_of(xprt, struct sock_xprt, xprt);
1219         struct xdr_buf *rcvbuf;
1220         size_t len;
1221         ssize_t r;
1222
1223         rcvbuf = &req->rq_private_buf;
1224
1225         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1226                 /*
1227                  * Save the RPC direction in the XDR buffer
1228                  */
1229                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1230                         &transport->tcp_calldir,
1231                         sizeof(transport->tcp_calldir));
1232                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1233                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1234         }
1235
1236         len = desc->count;
1237         if (len > transport->tcp_reclen - transport->tcp_offset) {
1238                 struct xdr_skb_reader my_desc;
1239
1240                 len = transport->tcp_reclen - transport->tcp_offset;
1241                 memcpy(&my_desc, desc, sizeof(my_desc));
1242                 my_desc.count = len;
1243                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1244                                           &my_desc, xdr_skb_read_bits);
1245                 desc->count -= r;
1246                 desc->offset += r;
1247         } else
1248                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1249                                           desc, xdr_skb_read_bits);
1250
1251         if (r > 0) {
1252                 transport->tcp_copied += r;
1253                 transport->tcp_offset += r;
1254         }
1255         if (r != len) {
1256                 /* Error when copying to the receive buffer,
1257                  * usually because we weren't able to allocate
1258                  * additional buffer pages. All we can do now
1259                  * is turn off TCP_RCV_COPY_DATA, so the request
1260                  * will not receive any additional updates,
1261                  * and time out.
1262                  * Any remaining data from this record will
1263                  * be discarded.
1264                  */
1265                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1266                 dprintk("RPC:       XID %08x truncated request\n",
1267                                 ntohl(transport->tcp_xid));
1268                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1269                                 "tcp_offset = %u, tcp_reclen = %u\n",
1270                                 xprt, transport->tcp_copied,
1271                                 transport->tcp_offset, transport->tcp_reclen);
1272                 return;
1273         }
1274
1275         dprintk("RPC:       XID %08x read %Zd bytes\n",
1276                         ntohl(transport->tcp_xid), r);
1277         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1278                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1279                         transport->tcp_offset, transport->tcp_reclen);
1280
1281         if (transport->tcp_copied == req->rq_private_buf.buflen)
1282                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1283         else if (transport->tcp_offset == transport->tcp_reclen) {
1284                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1285                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1286         }
1287 }
1288
1289 /*
1290  * Finds the request corresponding to the RPC xid and invokes the common
1291  * tcp read code to read the data.
1292  */
1293 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1294                                     struct xdr_skb_reader *desc)
1295 {
1296         struct sock_xprt *transport =
1297                                 container_of(xprt, struct sock_xprt, xprt);
1298         struct rpc_rqst *req;
1299
1300         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1301
1302         /* Find and lock the request corresponding to this xid */
1303         spin_lock_bh(&xprt->transport_lock);
1304         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1305         if (!req) {
1306                 dprintk("RPC:       XID %08x request not found!\n",
1307                                 ntohl(transport->tcp_xid));
1308                 spin_unlock_bh(&xprt->transport_lock);
1309                 return -1;
1310         }
1311
1312         xs_tcp_read_common(xprt, desc, req);
1313
1314         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1315                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1316
1317         spin_unlock_bh(&xprt->transport_lock);
1318         return 0;
1319 }
1320
1321 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1322 /*
1323  * Obtains an rpc_rqst previously allocated and invokes the common
1324  * tcp read code to read the data.  The result is placed in the callback
1325  * queue.
1326  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1327  * connection and return -1.
1328  */
1329 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1330                                        struct xdr_skb_reader *desc)
1331 {
1332         struct sock_xprt *transport =
1333                                 container_of(xprt, struct sock_xprt, xprt);
1334         struct rpc_rqst *req;
1335
1336         /* Look up and lock the request corresponding to the given XID */
1337         spin_lock_bh(&xprt->transport_lock);
1338         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1339         if (req == NULL) {
1340                 spin_unlock_bh(&xprt->transport_lock);
1341                 printk(KERN_WARNING "Callback slot table overflowed\n");
1342                 xprt_force_disconnect(xprt);
1343                 return -1;
1344         }
1345
1346         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1347         xs_tcp_read_common(xprt, desc, req);
1348
1349         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1350                 xprt_complete_bc_request(req, transport->tcp_copied);
1351         spin_unlock_bh(&xprt->transport_lock);
1352
1353         return 0;
1354 }
1355
1356 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1357                                         struct xdr_skb_reader *desc)
1358 {
1359         struct sock_xprt *transport =
1360                                 container_of(xprt, struct sock_xprt, xprt);
1361
1362         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1363                 xs_tcp_read_reply(xprt, desc) :
1364                 xs_tcp_read_callback(xprt, desc);
1365 }
1366 #else
1367 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1368                                         struct xdr_skb_reader *desc)
1369 {
1370         return xs_tcp_read_reply(xprt, desc);
1371 }
1372 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1373
1374 /*
1375  * Read data off the transport.  This can be either an RPC_CALL or an
1376  * RPC_REPLY.  Relay the processing to helper functions.
1377  */
1378 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1379                                     struct xdr_skb_reader *desc)
1380 {
1381         struct sock_xprt *transport =
1382                                 container_of(xprt, struct sock_xprt, xprt);
1383
1384         if (_xs_tcp_read_data(xprt, desc) == 0)
1385                 xs_tcp_check_fraghdr(transport);
1386         else {
1387                 /*
1388                  * The transport_lock protects the request handling.
1389                  * There's no need to hold it to update the tcp_flags.
1390                  */
1391                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1392         }
1393 }
1394
1395 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1396 {
1397         size_t len;
1398
1399         len = transport->tcp_reclen - transport->tcp_offset;
1400         if (len > desc->count)
1401                 len = desc->count;
1402         desc->count -= len;
1403         desc->offset += len;
1404         transport->tcp_offset += len;
1405         dprintk("RPC:       discarded %Zu bytes\n", len);
1406         xs_tcp_check_fraghdr(transport);
1407 }
1408
1409 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1410 {
1411         struct rpc_xprt *xprt = rd_desc->arg.data;
1412         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1413         struct xdr_skb_reader desc = {
1414                 .skb    = skb,
1415                 .offset = offset,
1416                 .count  = len,
1417         };
1418
1419         dprintk("RPC:       xs_tcp_data_recv started\n");
1420         do {
1421                 trace_xs_tcp_data_recv(transport);
1422                 /* Read in a new fragment marker if necessary */
1423                 /* Can we ever really expect to get completely empty fragments? */
1424                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1425                         xs_tcp_read_fraghdr(xprt, &desc);
1426                         continue;
1427                 }
1428                 /* Read in the xid if necessary */
1429                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1430                         xs_tcp_read_xid(transport, &desc);
1431                         continue;
1432                 }
1433                 /* Read in the call/reply flag */
1434                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1435                         xs_tcp_read_calldir(transport, &desc);
1436                         continue;
1437                 }
1438                 /* Read in the request data */
1439                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1440                         xs_tcp_read_data(xprt, &desc);
1441                         continue;
1442                 }
1443                 /* Skip over any trailing bytes on short reads */
1444                 xs_tcp_read_discard(transport, &desc);
1445         } while (desc.count);
1446         trace_xs_tcp_data_recv(transport);
1447         dprintk("RPC:       xs_tcp_data_recv done\n");
1448         return len - desc.count;
1449 }
1450
1451 static void xs_tcp_data_receive(struct sock_xprt *transport)
1452 {
1453         struct rpc_xprt *xprt = &transport->xprt;
1454         struct sock *sk;
1455         read_descriptor_t rd_desc = {
1456                 .count = 2*1024*1024,
1457                 .arg.data = xprt,
1458         };
1459         unsigned long total = 0;
1460         int read = 0;
1461
1462         mutex_lock(&transport->recv_mutex);
1463         sk = transport->inet;
1464         if (sk == NULL)
1465                 goto out;
1466
1467         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1468         for (;;) {
1469                 lock_sock(sk);
1470                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1471                 release_sock(sk);
1472                 if (read <= 0)
1473                         break;
1474                 total += read;
1475                 rd_desc.count = 65536;
1476         }
1477 out:
1478         mutex_unlock(&transport->recv_mutex);
1479         trace_xs_tcp_data_ready(xprt, read, total);
1480 }
1481
1482 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1483 {
1484         struct sock_xprt *transport =
1485                 container_of(work, struct sock_xprt, recv_worker);
1486         xs_tcp_data_receive(transport);
1487 }
1488
1489 /**
1490  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1491  * @sk: socket with data to read
1492  *
1493  */
1494 static void xs_tcp_data_ready(struct sock *sk)
1495 {
1496         struct sock_xprt *transport;
1497         struct rpc_xprt *xprt;
1498
1499         dprintk("RPC:       xs_tcp_data_ready...\n");
1500
1501         read_lock_bh(&sk->sk_callback_lock);
1502         if (!(xprt = xprt_from_sock(sk)))
1503                 goto out;
1504         transport = container_of(xprt, struct sock_xprt, xprt);
1505
1506         /* Any data means we had a useful conversation, so
1507          * the we don't need to delay the next reconnect
1508          */
1509         if (xprt->reestablish_timeout)
1510                 xprt->reestablish_timeout = 0;
1511         queue_work(rpciod_workqueue, &transport->recv_worker);
1512
1513 out:
1514         read_unlock_bh(&sk->sk_callback_lock);
1515 }
1516
1517 /**
1518  * xs_tcp_state_change - callback to handle TCP socket state changes
1519  * @sk: socket whose state has changed
1520  *
1521  */
1522 static void xs_tcp_state_change(struct sock *sk)
1523 {
1524         struct rpc_xprt *xprt;
1525         struct sock_xprt *transport;
1526
1527         read_lock_bh(&sk->sk_callback_lock);
1528         if (!(xprt = xprt_from_sock(sk)))
1529                 goto out;
1530         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1531         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1532                         sk->sk_state, xprt_connected(xprt),
1533                         sock_flag(sk, SOCK_DEAD),
1534                         sock_flag(sk, SOCK_ZAPPED),
1535                         sk->sk_shutdown);
1536
1537         transport = container_of(xprt, struct sock_xprt, xprt);
1538         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1539         switch (sk->sk_state) {
1540         case TCP_ESTABLISHED:
1541                 spin_lock(&xprt->transport_lock);
1542                 if (!xprt_test_and_set_connected(xprt)) {
1543
1544                         /* Reset TCP record info */
1545                         transport->tcp_offset = 0;
1546                         transport->tcp_reclen = 0;
1547                         transport->tcp_copied = 0;
1548                         transport->tcp_flags =
1549                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1550                         xprt->connect_cookie++;
1551                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1552                         xprt_clear_connecting(xprt);
1553
1554                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1555                 }
1556                 spin_unlock(&xprt->transport_lock);
1557                 break;
1558         case TCP_FIN_WAIT1:
1559                 /* The client initiated a shutdown of the socket */
1560                 xprt->connect_cookie++;
1561                 xprt->reestablish_timeout = 0;
1562                 set_bit(XPRT_CLOSING, &xprt->state);
1563                 smp_mb__before_atomic();
1564                 clear_bit(XPRT_CONNECTED, &xprt->state);
1565                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1566                 smp_mb__after_atomic();
1567                 break;
1568         case TCP_CLOSE_WAIT:
1569                 /* The server initiated a shutdown of the socket */
1570                 xprt->connect_cookie++;
1571                 clear_bit(XPRT_CONNECTED, &xprt->state);
1572                 xs_tcp_force_close(xprt);
1573         case TCP_CLOSING:
1574                 /*
1575                  * If the server closed down the connection, make sure that
1576                  * we back off before reconnecting
1577                  */
1578                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1579                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1580                 break;
1581         case TCP_LAST_ACK:
1582                 set_bit(XPRT_CLOSING, &xprt->state);
1583                 smp_mb__before_atomic();
1584                 clear_bit(XPRT_CONNECTED, &xprt->state);
1585                 smp_mb__after_atomic();
1586                 break;
1587         case TCP_CLOSE:
1588                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1589                                         &transport->sock_state))
1590                         xprt_clear_connecting(xprt);
1591                 xs_sock_mark_closed(xprt);
1592         }
1593  out:
1594         read_unlock_bh(&sk->sk_callback_lock);
1595 }
1596
1597 static void xs_write_space(struct sock *sk)
1598 {
1599         struct socket *sock;
1600         struct rpc_xprt *xprt;
1601
1602         if (unlikely(!(sock = sk->sk_socket)))
1603                 return;
1604         clear_bit(SOCK_NOSPACE, &sock->flags);
1605
1606         if (unlikely(!(xprt = xprt_from_sock(sk))))
1607                 return;
1608         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1609                 return;
1610
1611         xprt_write_space(xprt);
1612 }
1613
1614 /**
1615  * xs_udp_write_space - callback invoked when socket buffer space
1616  *                             becomes available
1617  * @sk: socket whose state has changed
1618  *
1619  * Called when more output buffer space is available for this socket.
1620  * We try not to wake our writers until they can make "significant"
1621  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1622  * with a bunch of small requests.
1623  */
1624 static void xs_udp_write_space(struct sock *sk)
1625 {
1626         read_lock_bh(&sk->sk_callback_lock);
1627
1628         /* from net/core/sock.c:sock_def_write_space */
1629         if (sock_writeable(sk))
1630                 xs_write_space(sk);
1631
1632         read_unlock_bh(&sk->sk_callback_lock);
1633 }
1634
1635 /**
1636  * xs_tcp_write_space - callback invoked when socket buffer space
1637  *                             becomes available
1638  * @sk: socket whose state has changed
1639  *
1640  * Called when more output buffer space is available for this socket.
1641  * We try not to wake our writers until they can make "significant"
1642  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1643  * with a bunch of small requests.
1644  */
1645 static void xs_tcp_write_space(struct sock *sk)
1646 {
1647         read_lock_bh(&sk->sk_callback_lock);
1648
1649         /* from net/core/stream.c:sk_stream_write_space */
1650         if (sk_stream_is_writeable(sk))
1651                 xs_write_space(sk);
1652
1653         read_unlock_bh(&sk->sk_callback_lock);
1654 }
1655
1656 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1657 {
1658         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1659         struct sock *sk = transport->inet;
1660
1661         if (transport->rcvsize) {
1662                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1663                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1664         }
1665         if (transport->sndsize) {
1666                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1667                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1668                 sk->sk_write_space(sk);
1669         }
1670 }
1671
1672 /**
1673  * xs_udp_set_buffer_size - set send and receive limits
1674  * @xprt: generic transport
1675  * @sndsize: requested size of send buffer, in bytes
1676  * @rcvsize: requested size of receive buffer, in bytes
1677  *
1678  * Set socket send and receive buffer size limits.
1679  */
1680 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1681 {
1682         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1683
1684         transport->sndsize = 0;
1685         if (sndsize)
1686                 transport->sndsize = sndsize + 1024;
1687         transport->rcvsize = 0;
1688         if (rcvsize)
1689                 transport->rcvsize = rcvsize + 1024;
1690
1691         xs_udp_do_set_buffer_size(xprt);
1692 }
1693
1694 /**
1695  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1696  * @task: task that timed out
1697  *
1698  * Adjust the congestion window after a retransmit timeout has occurred.
1699  */
1700 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1701 {
1702         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1703 }
1704
1705 static unsigned short xs_get_random_port(void)
1706 {
1707         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1708         unsigned short rand = (unsigned short) prandom_u32() % range;
1709         return rand + xprt_min_resvport;
1710 }
1711
1712 /**
1713  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1714  * @sock: socket
1715  *
1716  * Note that this function has to be called on all sockets that share the
1717  * same port, and it must be called before binding.
1718  */
1719 static void xs_sock_set_reuseport(struct socket *sock)
1720 {
1721         int opt = 1;
1722
1723         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1724                         (char *)&opt, sizeof(opt));
1725 }
1726
1727 static unsigned short xs_sock_getport(struct socket *sock)
1728 {
1729         struct sockaddr_storage buf;
1730         int buflen;
1731         unsigned short port = 0;
1732
1733         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1734                 goto out;
1735         switch (buf.ss_family) {
1736         case AF_INET6:
1737                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1738                 break;
1739         case AF_INET:
1740                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1741         }
1742 out:
1743         return port;
1744 }
1745
1746 /**
1747  * xs_set_port - reset the port number in the remote endpoint address
1748  * @xprt: generic transport
1749  * @port: new port number
1750  *
1751  */
1752 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1753 {
1754         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1755
1756         rpc_set_port(xs_addr(xprt), port);
1757         xs_update_peer_port(xprt);
1758 }
1759
1760 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1761 {
1762         if (transport->srcport == 0)
1763                 transport->srcport = xs_sock_getport(sock);
1764 }
1765
1766 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1767 {
1768         unsigned short port = transport->srcport;
1769
1770         if (port == 0 && transport->xprt.resvport)
1771                 port = xs_get_random_port();
1772         return port;
1773 }
1774
1775 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1776 {
1777         if (transport->srcport != 0)
1778                 transport->srcport = 0;
1779         if (!transport->xprt.resvport)
1780                 return 0;
1781         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1782                 return xprt_max_resvport;
1783         return --port;
1784 }
1785 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1786 {
1787         struct sockaddr_storage myaddr;
1788         int err, nloop = 0;
1789         unsigned short port = xs_get_srcport(transport);
1790         unsigned short last;
1791
1792         /*
1793          * If we are asking for any ephemeral port (i.e. port == 0 &&
1794          * transport->xprt.resvport == 0), don't bind.  Let the local
1795          * port selection happen implicitly when the socket is used
1796          * (for example at connect time).
1797          *
1798          * This ensures that we can continue to establish TCP
1799          * connections even when all local ephemeral ports are already
1800          * a part of some TCP connection.  This makes no difference
1801          * for UDP sockets, but also doens't harm them.
1802          *
1803          * If we're asking for any reserved port (i.e. port == 0 &&
1804          * transport->xprt.resvport == 1) xs_get_srcport above will
1805          * ensure that port is non-zero and we will bind as needed.
1806          */
1807         if (port == 0)
1808                 return 0;
1809
1810         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1811         do {
1812                 rpc_set_port((struct sockaddr *)&myaddr, port);
1813                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1814                                 transport->xprt.addrlen);
1815                 if (err == 0) {
1816                         transport->srcport = port;
1817                         break;
1818                 }
1819                 last = port;
1820                 port = xs_next_srcport(transport, port);
1821                 if (port > last)
1822                         nloop++;
1823         } while (err == -EADDRINUSE && nloop != 2);
1824
1825         if (myaddr.ss_family == AF_INET)
1826                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1827                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1828                                 port, err ? "failed" : "ok", err);
1829         else
1830                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1831                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1832                                 port, err ? "failed" : "ok", err);
1833         return err;
1834 }
1835
1836 /*
1837  * We don't support autobind on AF_LOCAL sockets
1838  */
1839 static void xs_local_rpcbind(struct rpc_task *task)
1840 {
1841         rcu_read_lock();
1842         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1843         rcu_read_unlock();
1844 }
1845
1846 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1847 {
1848 }
1849
1850 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1851 static struct lock_class_key xs_key[2];
1852 static struct lock_class_key xs_slock_key[2];
1853
1854 static inline void xs_reclassify_socketu(struct socket *sock)
1855 {
1856         struct sock *sk = sock->sk;
1857
1858         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1859                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1860 }
1861
1862 static inline void xs_reclassify_socket4(struct socket *sock)
1863 {
1864         struct sock *sk = sock->sk;
1865
1866         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1867                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1868 }
1869
1870 static inline void xs_reclassify_socket6(struct socket *sock)
1871 {
1872         struct sock *sk = sock->sk;
1873
1874         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1875                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1876 }
1877
1878 static inline void xs_reclassify_socket(int family, struct socket *sock)
1879 {
1880         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1881         if (sock_owned_by_user(sock->sk))
1882                 return;
1883
1884         switch (family) {
1885         case AF_LOCAL:
1886                 xs_reclassify_socketu(sock);
1887                 break;
1888         case AF_INET:
1889                 xs_reclassify_socket4(sock);
1890                 break;
1891         case AF_INET6:
1892                 xs_reclassify_socket6(sock);
1893                 break;
1894         }
1895 }
1896 #else
1897 static inline void xs_reclassify_socketu(struct socket *sock)
1898 {
1899 }
1900
1901 static inline void xs_reclassify_socket4(struct socket *sock)
1902 {
1903 }
1904
1905 static inline void xs_reclassify_socket6(struct socket *sock)
1906 {
1907 }
1908
1909 static inline void xs_reclassify_socket(int family, struct socket *sock)
1910 {
1911 }
1912 #endif
1913
1914 static void xs_dummy_setup_socket(struct work_struct *work)
1915 {
1916 }
1917
1918 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1919                 struct sock_xprt *transport, int family, int type,
1920                 int protocol, bool reuseport)
1921 {
1922         struct socket *sock;
1923         int err;
1924
1925         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1926         if (err < 0) {
1927                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1928                                 protocol, -err);
1929                 goto out;
1930         }
1931         xs_reclassify_socket(family, sock);
1932
1933         if (reuseport)
1934                 xs_sock_set_reuseport(sock);
1935
1936         err = xs_bind(transport, sock);
1937         if (err) {
1938                 sock_release(sock);
1939                 goto out;
1940         }
1941
1942         return sock;
1943 out:
1944         return ERR_PTR(err);
1945 }
1946
1947 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1948                                       struct socket *sock)
1949 {
1950         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1951                                                                         xprt);
1952
1953         if (!transport->inet) {
1954                 struct sock *sk = sock->sk;
1955
1956                 write_lock_bh(&sk->sk_callback_lock);
1957
1958                 xs_save_old_callbacks(transport, sk);
1959
1960                 sk->sk_user_data = xprt;
1961                 sk->sk_data_ready = xs_data_ready;
1962                 sk->sk_write_space = xs_udp_write_space;
1963                 sk->sk_error_report = xs_error_report;
1964                 sk->sk_allocation = GFP_NOIO;
1965
1966                 xprt_clear_connected(xprt);
1967
1968                 /* Reset to new socket */
1969                 transport->sock = sock;
1970                 transport->inet = sk;
1971
1972                 write_unlock_bh(&sk->sk_callback_lock);
1973         }
1974
1975         /* Tell the socket layer to start connecting... */
1976         xprt->stat.connect_count++;
1977         xprt->stat.connect_start = jiffies;
1978         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1979 }
1980
1981 /**
1982  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1983  * @transport: socket transport to connect
1984  */
1985 static int xs_local_setup_socket(struct sock_xprt *transport)
1986 {
1987         struct rpc_xprt *xprt = &transport->xprt;
1988         struct socket *sock;
1989         int status = -EIO;
1990
1991         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1992                                         SOCK_STREAM, 0, &sock, 1);
1993         if (status < 0) {
1994                 dprintk("RPC:       can't create AF_LOCAL "
1995                         "transport socket (%d).\n", -status);
1996                 goto out;
1997         }
1998         xs_reclassify_socketu(sock);
1999
2000         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2001                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2002
2003         status = xs_local_finish_connecting(xprt, sock);
2004         trace_rpc_socket_connect(xprt, sock, status);
2005         switch (status) {
2006         case 0:
2007                 dprintk("RPC:       xprt %p connected to %s\n",
2008                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2009                 xprt_set_connected(xprt);
2010         case -ENOBUFS:
2011                 break;
2012         case -ENOENT:
2013                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2014                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2015                 break;
2016         case -ECONNREFUSED:
2017                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2018                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2019                 break;
2020         default:
2021                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2022                                 __func__, -status,
2023                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2024         }
2025
2026 out:
2027         xprt_clear_connecting(xprt);
2028         xprt_wake_pending_tasks(xprt, status);
2029         return status;
2030 }
2031
2032 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2033 {
2034         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2035         int ret;
2036
2037          if (RPC_IS_ASYNC(task)) {
2038                 /*
2039                  * We want the AF_LOCAL connect to be resolved in the
2040                  * filesystem namespace of the process making the rpc
2041                  * call.  Thus we connect synchronously.
2042                  *
2043                  * If we want to support asynchronous AF_LOCAL calls,
2044                  * we'll need to figure out how to pass a namespace to
2045                  * connect.
2046                  */
2047                 rpc_exit(task, -ENOTCONN);
2048                 return;
2049         }
2050         ret = xs_local_setup_socket(transport);
2051         if (ret && !RPC_IS_SOFTCONN(task))
2052                 msleep_interruptible(15000);
2053 }
2054
2055 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2056 /*
2057  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2058  * know that we have exclusive access to the socket), to guard against
2059  * races with xs_reset_transport.
2060  */
2061 static void xs_set_memalloc(struct rpc_xprt *xprt)
2062 {
2063         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2064                         xprt);
2065
2066         /*
2067          * If there's no sock, then we have nothing to set. The
2068          * reconnecting process will get it for us.
2069          */
2070         if (!transport->inet)
2071                 return;
2072         if (atomic_read(&xprt->swapper))
2073                 sk_set_memalloc(transport->inet);
2074 }
2075
2076 /**
2077  * xs_enable_swap - Tag this transport as being used for swap.
2078  * @xprt: transport to tag
2079  *
2080  * Take a reference to this transport on behalf of the rpc_clnt, and
2081  * optionally mark it for swapping if it wasn't already.
2082  */
2083 static int
2084 xs_enable_swap(struct rpc_xprt *xprt)
2085 {
2086         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2087
2088         if (atomic_inc_return(&xprt->swapper) != 1)
2089                 return 0;
2090         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2091                 return -ERESTARTSYS;
2092         if (xs->inet)
2093                 sk_set_memalloc(xs->inet);
2094         xprt_release_xprt(xprt, NULL);
2095         return 0;
2096 }
2097
2098 /**
2099  * xs_disable_swap - Untag this transport as being used for swap.
2100  * @xprt: transport to tag
2101  *
2102  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2103  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2104  */
2105 static void
2106 xs_disable_swap(struct rpc_xprt *xprt)
2107 {
2108         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2109
2110         if (!atomic_dec_and_test(&xprt->swapper))
2111                 return;
2112         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2113                 return;
2114         if (xs->inet)
2115                 sk_clear_memalloc(xs->inet);
2116         xprt_release_xprt(xprt, NULL);
2117 }
2118 #else
2119 static void xs_set_memalloc(struct rpc_xprt *xprt)
2120 {
2121 }
2122
2123 static int
2124 xs_enable_swap(struct rpc_xprt *xprt)
2125 {
2126         return -EINVAL;
2127 }
2128
2129 static void
2130 xs_disable_swap(struct rpc_xprt *xprt)
2131 {
2132 }
2133 #endif
2134
2135 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2136 {
2137         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2138
2139         if (!transport->inet) {
2140                 struct sock *sk = sock->sk;
2141
2142                 write_lock_bh(&sk->sk_callback_lock);
2143
2144                 xs_save_old_callbacks(transport, sk);
2145
2146                 sk->sk_user_data = xprt;
2147                 sk->sk_data_ready = xs_data_ready;
2148                 sk->sk_write_space = xs_udp_write_space;
2149                 sk->sk_allocation = GFP_NOIO;
2150
2151                 xprt_set_connected(xprt);
2152
2153                 /* Reset to new socket */
2154                 transport->sock = sock;
2155                 transport->inet = sk;
2156
2157                 xs_set_memalloc(xprt);
2158
2159                 write_unlock_bh(&sk->sk_callback_lock);
2160         }
2161         xs_udp_do_set_buffer_size(xprt);
2162 }
2163
2164 static void xs_udp_setup_socket(struct work_struct *work)
2165 {
2166         struct sock_xprt *transport =
2167                 container_of(work, struct sock_xprt, connect_worker.work);
2168         struct rpc_xprt *xprt = &transport->xprt;
2169         struct socket *sock = transport->sock;
2170         int status = -EIO;
2171
2172         sock = xs_create_sock(xprt, transport,
2173                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2174                         IPPROTO_UDP, false);
2175         if (IS_ERR(sock))
2176                 goto out;
2177
2178         dprintk("RPC:       worker connecting xprt %p via %s to "
2179                                 "%s (port %s)\n", xprt,
2180                         xprt->address_strings[RPC_DISPLAY_PROTO],
2181                         xprt->address_strings[RPC_DISPLAY_ADDR],
2182                         xprt->address_strings[RPC_DISPLAY_PORT]);
2183
2184         xs_udp_finish_connecting(xprt, sock);
2185         trace_rpc_socket_connect(xprt, sock, 0);
2186         status = 0;
2187 out:
2188         xprt_unlock_connect(xprt, transport);
2189         xprt_clear_connecting(xprt);
2190         xprt_wake_pending_tasks(xprt, status);
2191 }
2192
2193 /**
2194  * xs_tcp_shutdown - gracefully shut down a TCP socket
2195  * @xprt: transport
2196  *
2197  * Initiates a graceful shutdown of the TCP socket by calling the
2198  * equivalent of shutdown(SHUT_RDWR);
2199  */
2200 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2201 {
2202         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2203         struct socket *sock = transport->sock;
2204
2205         if (sock == NULL)
2206                 return;
2207         if (xprt_connected(xprt)) {
2208                 kernel_sock_shutdown(sock, SHUT_RDWR);
2209                 trace_rpc_socket_shutdown(xprt, sock);
2210         } else
2211                 xs_reset_transport(transport);
2212 }
2213
2214 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2215 {
2216         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2217         int ret = -ENOTCONN;
2218
2219         if (!transport->inet) {
2220                 struct sock *sk = sock->sk;
2221                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2222                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2223                 unsigned int opt_on = 1;
2224                 unsigned int timeo;
2225
2226                 /* TCP Keepalive options */
2227                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2228                                 (char *)&opt_on, sizeof(opt_on));
2229                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2230                                 (char *)&keepidle, sizeof(keepidle));
2231                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2232                                 (char *)&keepidle, sizeof(keepidle));
2233                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2234                                 (char *)&keepcnt, sizeof(keepcnt));
2235
2236                 /* TCP user timeout (see RFC5482) */
2237                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2238                         (xprt->timeout->to_retries + 1);
2239                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2240                                 (char *)&timeo, sizeof(timeo));
2241
2242                 write_lock_bh(&sk->sk_callback_lock);
2243
2244                 xs_save_old_callbacks(transport, sk);
2245
2246                 sk->sk_user_data = xprt;
2247                 sk->sk_data_ready = xs_tcp_data_ready;
2248                 sk->sk_state_change = xs_tcp_state_change;
2249                 sk->sk_write_space = xs_tcp_write_space;
2250                 sk->sk_error_report = xs_error_report;
2251                 sk->sk_allocation = GFP_NOIO;
2252
2253                 /* socket options */
2254                 sock_reset_flag(sk, SOCK_LINGER);
2255                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2256
2257                 xprt_clear_connected(xprt);
2258
2259                 /* Reset to new socket */
2260                 transport->sock = sock;
2261                 transport->inet = sk;
2262
2263                 write_unlock_bh(&sk->sk_callback_lock);
2264         }
2265
2266         if (!xprt_bound(xprt))
2267                 goto out;
2268
2269         xs_set_memalloc(xprt);
2270
2271         /* Tell the socket layer to start connecting... */
2272         xprt->stat.connect_count++;
2273         xprt->stat.connect_start = jiffies;
2274         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2275         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2276         switch (ret) {
2277         case 0:
2278                 xs_set_srcport(transport, sock);
2279         case -EINPROGRESS:
2280                 /* SYN_SENT! */
2281                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2282                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2283         }
2284 out:
2285         return ret;
2286 }
2287
2288 /**
2289  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2290  *
2291  * Invoked by a work queue tasklet.
2292  */
2293 static void xs_tcp_setup_socket(struct work_struct *work)
2294 {
2295         struct sock_xprt *transport =
2296                 container_of(work, struct sock_xprt, connect_worker.work);
2297         struct socket *sock = transport->sock;
2298         struct rpc_xprt *xprt = &transport->xprt;
2299         int status = -EIO;
2300
2301         if (!sock) {
2302                 sock = xs_create_sock(xprt, transport,
2303                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2304                                 IPPROTO_TCP, true);
2305                 if (IS_ERR(sock)) {
2306                         status = PTR_ERR(sock);
2307                         goto out;
2308                 }
2309         }
2310
2311         dprintk("RPC:       worker connecting xprt %p via %s to "
2312                                 "%s (port %s)\n", xprt,
2313                         xprt->address_strings[RPC_DISPLAY_PROTO],
2314                         xprt->address_strings[RPC_DISPLAY_ADDR],
2315                         xprt->address_strings[RPC_DISPLAY_PORT]);
2316
2317         status = xs_tcp_finish_connecting(xprt, sock);
2318         trace_rpc_socket_connect(xprt, sock, status);
2319         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2320                         xprt, -status, xprt_connected(xprt),
2321                         sock->sk->sk_state);
2322         switch (status) {
2323         default:
2324                 printk("%s: connect returned unhandled error %d\n",
2325                         __func__, status);
2326         case -EADDRNOTAVAIL:
2327                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2328                  * and retry
2329                  */
2330                 xs_tcp_force_close(xprt);
2331                 break;
2332         case 0:
2333         case -EINPROGRESS:
2334         case -EALREADY:
2335                 xprt_unlock_connect(xprt, transport);
2336                 return;
2337         case -EINVAL:
2338                 /* Happens, for instance, if the user specified a link
2339                  * local IPv6 address without a scope-id.
2340                  */
2341         case -ECONNREFUSED:
2342         case -ECONNRESET:
2343         case -ENETUNREACH:
2344         case -EADDRINUSE:
2345         case -ENOBUFS:
2346                 /* retry with existing socket, after a delay */
2347                 xs_tcp_force_close(xprt);
2348                 goto out;
2349         }
2350         status = -EAGAIN;
2351 out:
2352         xprt_unlock_connect(xprt, transport);
2353         xprt_clear_connecting(xprt);
2354         xprt_wake_pending_tasks(xprt, status);
2355 }
2356
2357 /**
2358  * xs_connect - connect a socket to a remote endpoint
2359  * @xprt: pointer to transport structure
2360  * @task: address of RPC task that manages state of connect request
2361  *
2362  * TCP: If the remote end dropped the connection, delay reconnecting.
2363  *
2364  * UDP socket connects are synchronous, but we use a work queue anyway
2365  * to guarantee that even unprivileged user processes can set up a
2366  * socket on a privileged port.
2367  *
2368  * If a UDP socket connect fails, the delay behavior here prevents
2369  * retry floods (hard mounts).
2370  */
2371 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2372 {
2373         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2374
2375         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2376
2377         if (transport->sock != NULL) {
2378                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2379                                 "seconds\n",
2380                                 xprt, xprt->reestablish_timeout / HZ);
2381
2382                 /* Start by resetting any existing state */
2383                 xs_reset_transport(transport);
2384
2385                 queue_delayed_work(rpciod_workqueue,
2386                                    &transport->connect_worker,
2387                                    xprt->reestablish_timeout);
2388                 xprt->reestablish_timeout <<= 1;
2389                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2390                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2391                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2392                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2393         } else {
2394                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2395                 queue_delayed_work(rpciod_workqueue,
2396                                    &transport->connect_worker, 0);
2397         }
2398 }
2399
2400 /**
2401  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2402  * @xprt: rpc_xprt struct containing statistics
2403  * @seq: output file
2404  *
2405  */
2406 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2407 {
2408         long idle_time = 0;
2409
2410         if (xprt_connected(xprt))
2411                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2412
2413         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2414                         "%llu %llu %lu %llu %llu\n",
2415                         xprt->stat.bind_count,
2416                         xprt->stat.connect_count,
2417                         xprt->stat.connect_time,
2418                         idle_time,
2419                         xprt->stat.sends,
2420                         xprt->stat.recvs,
2421                         xprt->stat.bad_xids,
2422                         xprt->stat.req_u,
2423                         xprt->stat.bklog_u,
2424                         xprt->stat.max_slots,
2425                         xprt->stat.sending_u,
2426                         xprt->stat.pending_u);
2427 }
2428
2429 /**
2430  * xs_udp_print_stats - display UDP socket-specifc stats
2431  * @xprt: rpc_xprt struct containing statistics
2432  * @seq: output file
2433  *
2434  */
2435 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2436 {
2437         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2438
2439         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2440                         "%lu %llu %llu\n",
2441                         transport->srcport,
2442                         xprt->stat.bind_count,
2443                         xprt->stat.sends,
2444                         xprt->stat.recvs,
2445                         xprt->stat.bad_xids,
2446                         xprt->stat.req_u,
2447                         xprt->stat.bklog_u,
2448                         xprt->stat.max_slots,
2449                         xprt->stat.sending_u,
2450                         xprt->stat.pending_u);
2451 }
2452
2453 /**
2454  * xs_tcp_print_stats - display TCP socket-specifc stats
2455  * @xprt: rpc_xprt struct containing statistics
2456  * @seq: output file
2457  *
2458  */
2459 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2460 {
2461         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2462         long idle_time = 0;
2463
2464         if (xprt_connected(xprt))
2465                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2466
2467         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2468                         "%llu %llu %lu %llu %llu\n",
2469                         transport->srcport,
2470                         xprt->stat.bind_count,
2471                         xprt->stat.connect_count,
2472                         xprt->stat.connect_time,
2473                         idle_time,
2474                         xprt->stat.sends,
2475                         xprt->stat.recvs,
2476                         xprt->stat.bad_xids,
2477                         xprt->stat.req_u,
2478                         xprt->stat.bklog_u,
2479                         xprt->stat.max_slots,
2480                         xprt->stat.sending_u,
2481                         xprt->stat.pending_u);
2482 }
2483
2484 /*
2485  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2486  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2487  * to use the server side send routines.
2488  */
2489 static void *bc_malloc(struct rpc_task *task, size_t size)
2490 {
2491         struct page *page;
2492         struct rpc_buffer *buf;
2493
2494         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2495         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2496                 return NULL;
2497
2498         page = alloc_page(GFP_KERNEL);
2499         if (!page)
2500                 return NULL;
2501
2502         buf = page_address(page);
2503         buf->len = PAGE_SIZE;
2504
2505         return buf->data;
2506 }
2507
2508 /*
2509  * Free the space allocated in the bc_alloc routine
2510  */
2511 static void bc_free(void *buffer)
2512 {
2513         struct rpc_buffer *buf;
2514
2515         if (!buffer)
2516                 return;
2517
2518         buf = container_of(buffer, struct rpc_buffer, data);
2519         free_page((unsigned long)buf);
2520 }
2521
2522 /*
2523  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2524  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2525  */
2526 static int bc_sendto(struct rpc_rqst *req)
2527 {
2528         int len;
2529         struct xdr_buf *xbufp = &req->rq_snd_buf;
2530         struct rpc_xprt *xprt = req->rq_xprt;
2531         struct sock_xprt *transport =
2532                                 container_of(xprt, struct sock_xprt, xprt);
2533         struct socket *sock = transport->sock;
2534         unsigned long headoff;
2535         unsigned long tailoff;
2536
2537         xs_encode_stream_record_marker(xbufp);
2538
2539         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2540         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2541         len = svc_send_common(sock, xbufp,
2542                               virt_to_page(xbufp->head[0].iov_base), headoff,
2543                               xbufp->tail[0].iov_base, tailoff);
2544
2545         if (len != xbufp->len) {
2546                 printk(KERN_NOTICE "Error sending entire callback!\n");
2547                 len = -EAGAIN;
2548         }
2549
2550         return len;
2551 }
2552
2553 /*
2554  * The send routine. Borrows from svc_send
2555  */
2556 static int bc_send_request(struct rpc_task *task)
2557 {
2558         struct rpc_rqst *req = task->tk_rqstp;
2559         struct svc_xprt *xprt;
2560         u32                     len;
2561
2562         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2563         /*
2564          * Get the server socket associated with this callback xprt
2565          */
2566         xprt = req->rq_xprt->bc_xprt;
2567
2568         /*
2569          * Grab the mutex to serialize data as the connection is shared
2570          * with the fore channel
2571          */
2572         if (!mutex_trylock(&xprt->xpt_mutex)) {
2573                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2574                 if (!mutex_trylock(&xprt->xpt_mutex))
2575                         return -EAGAIN;
2576                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2577         }
2578         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2579                 len = -ENOTCONN;
2580         else
2581                 len = bc_sendto(req);
2582         mutex_unlock(&xprt->xpt_mutex);
2583
2584         if (len > 0)
2585                 len = 0;
2586
2587         return len;
2588 }
2589
2590 /*
2591  * The close routine. Since this is client initiated, we do nothing
2592  */
2593
2594 static void bc_close(struct rpc_xprt *xprt)
2595 {
2596 }
2597
2598 /*
2599  * The xprt destroy routine. Again, because this connection is client
2600  * initiated, we do nothing
2601  */
2602
2603 static void bc_destroy(struct rpc_xprt *xprt)
2604 {
2605         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2606
2607         xs_xprt_free(xprt);
2608         module_put(THIS_MODULE);
2609 }
2610
2611 static struct rpc_xprt_ops xs_local_ops = {
2612         .reserve_xprt           = xprt_reserve_xprt,
2613         .release_xprt           = xs_tcp_release_xprt,
2614         .alloc_slot             = xprt_alloc_slot,
2615         .rpcbind                = xs_local_rpcbind,
2616         .set_port               = xs_local_set_port,
2617         .connect                = xs_local_connect,
2618         .buf_alloc              = rpc_malloc,
2619         .buf_free               = rpc_free,
2620         .send_request           = xs_local_send_request,
2621         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2622         .close                  = xs_close,
2623         .destroy                = xs_destroy,
2624         .print_stats            = xs_local_print_stats,
2625         .enable_swap            = xs_enable_swap,
2626         .disable_swap           = xs_disable_swap,
2627 };
2628
2629 static struct rpc_xprt_ops xs_udp_ops = {
2630         .set_buffer_size        = xs_udp_set_buffer_size,
2631         .reserve_xprt           = xprt_reserve_xprt_cong,
2632         .release_xprt           = xprt_release_xprt_cong,
2633         .alloc_slot             = xprt_alloc_slot,
2634         .rpcbind                = rpcb_getport_async,
2635         .set_port               = xs_set_port,
2636         .connect                = xs_connect,
2637         .buf_alloc              = rpc_malloc,
2638         .buf_free               = rpc_free,
2639         .send_request           = xs_udp_send_request,
2640         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2641         .timer                  = xs_udp_timer,
2642         .release_request        = xprt_release_rqst_cong,
2643         .close                  = xs_close,
2644         .destroy                = xs_destroy,
2645         .print_stats            = xs_udp_print_stats,
2646         .enable_swap            = xs_enable_swap,
2647         .disable_swap           = xs_disable_swap,
2648         .inject_disconnect      = xs_inject_disconnect,
2649 };
2650
2651 static struct rpc_xprt_ops xs_tcp_ops = {
2652         .reserve_xprt           = xprt_reserve_xprt,
2653         .release_xprt           = xs_tcp_release_xprt,
2654         .alloc_slot             = xprt_lock_and_alloc_slot,
2655         .rpcbind                = rpcb_getport_async,
2656         .set_port               = xs_set_port,
2657         .connect                = xs_connect,
2658         .buf_alloc              = rpc_malloc,
2659         .buf_free               = rpc_free,
2660         .send_request           = xs_tcp_send_request,
2661         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2662         .close                  = xs_tcp_shutdown,
2663         .destroy                = xs_destroy,
2664         .print_stats            = xs_tcp_print_stats,
2665         .enable_swap            = xs_enable_swap,
2666         .disable_swap           = xs_disable_swap,
2667         .inject_disconnect      = xs_inject_disconnect,
2668 };
2669
2670 /*
2671  * The rpc_xprt_ops for the server backchannel
2672  */
2673
2674 static struct rpc_xprt_ops bc_tcp_ops = {
2675         .reserve_xprt           = xprt_reserve_xprt,
2676         .release_xprt           = xprt_release_xprt,
2677         .alloc_slot             = xprt_alloc_slot,
2678         .buf_alloc              = bc_malloc,
2679         .buf_free               = bc_free,
2680         .send_request           = bc_send_request,
2681         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2682         .close                  = bc_close,
2683         .destroy                = bc_destroy,
2684         .print_stats            = xs_tcp_print_stats,
2685         .enable_swap            = xs_enable_swap,
2686         .disable_swap           = xs_disable_swap,
2687         .inject_disconnect      = xs_inject_disconnect,
2688 };
2689
2690 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2691 {
2692         static const struct sockaddr_in sin = {
2693                 .sin_family             = AF_INET,
2694                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2695         };
2696         static const struct sockaddr_in6 sin6 = {
2697                 .sin6_family            = AF_INET6,
2698                 .sin6_addr              = IN6ADDR_ANY_INIT,
2699         };
2700
2701         switch (family) {
2702         case AF_LOCAL:
2703                 break;
2704         case AF_INET:
2705                 memcpy(sap, &sin, sizeof(sin));
2706                 break;
2707         case AF_INET6:
2708                 memcpy(sap, &sin6, sizeof(sin6));
2709                 break;
2710         default:
2711                 dprintk("RPC:       %s: Bad address family\n", __func__);
2712                 return -EAFNOSUPPORT;
2713         }
2714         return 0;
2715 }
2716
2717 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2718                                       unsigned int slot_table_size,
2719                                       unsigned int max_slot_table_size)
2720 {
2721         struct rpc_xprt *xprt;
2722         struct sock_xprt *new;
2723
2724         if (args->addrlen > sizeof(xprt->addr)) {
2725                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2726                 return ERR_PTR(-EBADF);
2727         }
2728
2729         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2730                         max_slot_table_size);
2731         if (xprt == NULL) {
2732                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2733                                 "rpc_xprt\n");
2734                 return ERR_PTR(-ENOMEM);
2735         }
2736
2737         new = container_of(xprt, struct sock_xprt, xprt);
2738         mutex_init(&new->recv_mutex);
2739         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2740         xprt->addrlen = args->addrlen;
2741         if (args->srcaddr)
2742                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2743         else {
2744                 int err;
2745                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2746                                         (struct sockaddr *)&new->srcaddr);
2747                 if (err != 0) {
2748                         xprt_free(xprt);
2749                         return ERR_PTR(err);
2750                 }
2751         }
2752
2753         return xprt;
2754 }
2755
2756 static const struct rpc_timeout xs_local_default_timeout = {
2757         .to_initval = 10 * HZ,
2758         .to_maxval = 10 * HZ,
2759         .to_retries = 2,
2760 };
2761
2762 /**
2763  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2764  * @args: rpc transport creation arguments
2765  *
2766  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2767  */
2768 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2769 {
2770         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2771         struct sock_xprt *transport;
2772         struct rpc_xprt *xprt;
2773         struct rpc_xprt *ret;
2774
2775         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2776                         xprt_max_tcp_slot_table_entries);
2777         if (IS_ERR(xprt))
2778                 return xprt;
2779         transport = container_of(xprt, struct sock_xprt, xprt);
2780
2781         xprt->prot = 0;
2782         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2783         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2784
2785         xprt->bind_timeout = XS_BIND_TO;
2786         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2787         xprt->idle_timeout = XS_IDLE_DISC_TO;
2788
2789         xprt->ops = &xs_local_ops;
2790         xprt->timeout = &xs_local_default_timeout;
2791
2792         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2793         INIT_DELAYED_WORK(&transport->connect_worker,
2794                         xs_dummy_setup_socket);
2795
2796         switch (sun->sun_family) {
2797         case AF_LOCAL:
2798                 if (sun->sun_path[0] != '/') {
2799                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2800                                         sun->sun_path);
2801                         ret = ERR_PTR(-EINVAL);
2802                         goto out_err;
2803                 }
2804                 xprt_set_bound(xprt);
2805                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2806                 ret = ERR_PTR(xs_local_setup_socket(transport));
2807                 if (ret)
2808                         goto out_err;
2809                 break;
2810         default:
2811                 ret = ERR_PTR(-EAFNOSUPPORT);
2812                 goto out_err;
2813         }
2814
2815         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2816                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2817
2818         if (try_module_get(THIS_MODULE))
2819                 return xprt;
2820         ret = ERR_PTR(-EINVAL);
2821 out_err:
2822         xs_xprt_free(xprt);
2823         return ret;
2824 }
2825
2826 static const struct rpc_timeout xs_udp_default_timeout = {
2827         .to_initval = 5 * HZ,
2828         .to_maxval = 30 * HZ,
2829         .to_increment = 5 * HZ,
2830         .to_retries = 5,
2831 };
2832
2833 /**
2834  * xs_setup_udp - Set up transport to use a UDP socket
2835  * @args: rpc transport creation arguments
2836  *
2837  */
2838 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2839 {
2840         struct sockaddr *addr = args->dstaddr;
2841         struct rpc_xprt *xprt;
2842         struct sock_xprt *transport;
2843         struct rpc_xprt *ret;
2844
2845         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2846                         xprt_udp_slot_table_entries);
2847         if (IS_ERR(xprt))
2848                 return xprt;
2849         transport = container_of(xprt, struct sock_xprt, xprt);
2850
2851         xprt->prot = IPPROTO_UDP;
2852         xprt->tsh_size = 0;
2853         /* XXX: header size can vary due to auth type, IPv6, etc. */
2854         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2855
2856         xprt->bind_timeout = XS_BIND_TO;
2857         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2858         xprt->idle_timeout = XS_IDLE_DISC_TO;
2859
2860         xprt->ops = &xs_udp_ops;
2861
2862         xprt->timeout = &xs_udp_default_timeout;
2863
2864         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2865         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2866
2867         switch (addr->sa_family) {
2868         case AF_INET:
2869                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2870                         xprt_set_bound(xprt);
2871
2872                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2873                 break;
2874         case AF_INET6:
2875                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2876                         xprt_set_bound(xprt);
2877
2878                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2879                 break;
2880         default:
2881                 ret = ERR_PTR(-EAFNOSUPPORT);
2882                 goto out_err;
2883         }
2884
2885         if (xprt_bound(xprt))
2886                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2887                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2888                                 xprt->address_strings[RPC_DISPLAY_PORT],
2889                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2890         else
2891                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2892                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2893                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2894
2895         if (try_module_get(THIS_MODULE))
2896                 return xprt;
2897         ret = ERR_PTR(-EINVAL);
2898 out_err:
2899         xs_xprt_free(xprt);
2900         return ret;
2901 }
2902
2903 static const struct rpc_timeout xs_tcp_default_timeout = {
2904         .to_initval = 60 * HZ,
2905         .to_maxval = 60 * HZ,
2906         .to_retries = 2,
2907 };
2908
2909 /**
2910  * xs_setup_tcp - Set up transport to use a TCP socket
2911  * @args: rpc transport creation arguments
2912  *
2913  */
2914 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2915 {
2916         struct sockaddr *addr = args->dstaddr;
2917         struct rpc_xprt *xprt;
2918         struct sock_xprt *transport;
2919         struct rpc_xprt *ret;
2920         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2921
2922         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2923                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2924
2925         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2926                         max_slot_table_size);
2927         if (IS_ERR(xprt))
2928                 return xprt;
2929         transport = container_of(xprt, struct sock_xprt, xprt);
2930
2931         xprt->prot = IPPROTO_TCP;
2932         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2933         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2934
2935         xprt->bind_timeout = XS_BIND_TO;
2936         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2937         xprt->idle_timeout = XS_IDLE_DISC_TO;
2938
2939         xprt->ops = &xs_tcp_ops;
2940         xprt->timeout = &xs_tcp_default_timeout;
2941
2942         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2943         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2944
2945         switch (addr->sa_family) {
2946         case AF_INET:
2947                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2948                         xprt_set_bound(xprt);
2949
2950                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2951                 break;
2952         case AF_INET6:
2953                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2954                         xprt_set_bound(xprt);
2955
2956                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2957                 break;
2958         default:
2959                 ret = ERR_PTR(-EAFNOSUPPORT);
2960                 goto out_err;
2961         }
2962
2963         if (xprt_bound(xprt))
2964                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2965                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2966                                 xprt->address_strings[RPC_DISPLAY_PORT],
2967                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2968         else
2969                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2970                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2971                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2972
2973         if (try_module_get(THIS_MODULE))
2974                 return xprt;
2975         ret = ERR_PTR(-EINVAL);
2976 out_err:
2977         xs_xprt_free(xprt);
2978         return ret;
2979 }
2980
2981 /**
2982  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2983  * @args: rpc transport creation arguments
2984  *
2985  */
2986 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2987 {
2988         struct sockaddr *addr = args->dstaddr;
2989         struct rpc_xprt *xprt;
2990         struct sock_xprt *transport;
2991         struct svc_sock *bc_sock;
2992         struct rpc_xprt *ret;
2993
2994         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2995                         xprt_tcp_slot_table_entries);
2996         if (IS_ERR(xprt))
2997                 return xprt;
2998         transport = container_of(xprt, struct sock_xprt, xprt);
2999
3000         xprt->prot = IPPROTO_TCP;
3001         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3002         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3003         xprt->timeout = &xs_tcp_default_timeout;
3004
3005         /* backchannel */
3006         xprt_set_bound(xprt);
3007         xprt->bind_timeout = 0;
3008         xprt->reestablish_timeout = 0;
3009         xprt->idle_timeout = 0;
3010
3011         xprt->ops = &bc_tcp_ops;
3012
3013         switch (addr->sa_family) {
3014         case AF_INET:
3015                 xs_format_peer_addresses(xprt, "tcp",
3016                                          RPCBIND_NETID_TCP);
3017                 break;
3018         case AF_INET6:
3019                 xs_format_peer_addresses(xprt, "tcp",
3020                                    RPCBIND_NETID_TCP6);
3021                 break;
3022         default:
3023                 ret = ERR_PTR(-EAFNOSUPPORT);
3024                 goto out_err;
3025         }
3026
3027         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3028                         xprt->address_strings[RPC_DISPLAY_ADDR],
3029                         xprt->address_strings[RPC_DISPLAY_PORT],
3030                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3031
3032         /*
3033          * Once we've associated a backchannel xprt with a connection,
3034          * we want to keep it around as long as the connection lasts,
3035          * in case we need to start using it for a backchannel again;
3036          * this reference won't be dropped until bc_xprt is destroyed.
3037          */
3038         xprt_get(xprt);
3039         args->bc_xprt->xpt_bc_xprt = xprt;
3040         xprt->bc_xprt = args->bc_xprt;
3041         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3042         transport->sock = bc_sock->sk_sock;
3043         transport->inet = bc_sock->sk_sk;
3044
3045         /*
3046          * Since we don't want connections for the backchannel, we set
3047          * the xprt status to connected
3048          */
3049         xprt_set_connected(xprt);
3050
3051         if (try_module_get(THIS_MODULE))
3052                 return xprt;
3053
3054         args->bc_xprt->xpt_bc_xprt = NULL;
3055         xprt_put(xprt);
3056         ret = ERR_PTR(-EINVAL);
3057 out_err:
3058         xs_xprt_free(xprt);
3059         return ret;
3060 }
3061
3062 static struct xprt_class        xs_local_transport = {
3063         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3064         .name           = "named UNIX socket",
3065         .owner          = THIS_MODULE,
3066         .ident          = XPRT_TRANSPORT_LOCAL,
3067         .setup          = xs_setup_local,
3068 };
3069
3070 static struct xprt_class        xs_udp_transport = {
3071         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3072         .name           = "udp",
3073         .owner          = THIS_MODULE,
3074         .ident          = XPRT_TRANSPORT_UDP,
3075         .setup          = xs_setup_udp,
3076 };
3077
3078 static struct xprt_class        xs_tcp_transport = {
3079         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3080         .name           = "tcp",
3081         .owner          = THIS_MODULE,
3082         .ident          = XPRT_TRANSPORT_TCP,
3083         .setup          = xs_setup_tcp,
3084 };
3085
3086 static struct xprt_class        xs_bc_tcp_transport = {
3087         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3088         .name           = "tcp NFSv4.1 backchannel",
3089         .owner          = THIS_MODULE,
3090         .ident          = XPRT_TRANSPORT_BC_TCP,
3091         .setup          = xs_setup_bc_tcp,
3092 };
3093
3094 /**
3095  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3096  *
3097  */
3098 int init_socket_xprt(void)
3099 {
3100 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3101         if (!sunrpc_table_header)
3102                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3103 #endif
3104
3105         xprt_register_transport(&xs_local_transport);
3106         xprt_register_transport(&xs_udp_transport);
3107         xprt_register_transport(&xs_tcp_transport);
3108         xprt_register_transport(&xs_bc_tcp_transport);
3109
3110         return 0;
3111 }
3112
3113 /**
3114  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3115  *
3116  */
3117 void cleanup_socket_xprt(void)
3118 {
3119 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3120         if (sunrpc_table_header) {
3121                 unregister_sysctl_table(sunrpc_table_header);
3122                 sunrpc_table_header = NULL;
3123         }
3124 #endif
3125
3126         xprt_unregister_transport(&xs_local_transport);
3127         xprt_unregister_transport(&xs_udp_transport);
3128         xprt_unregister_transport(&xs_tcp_transport);
3129         xprt_unregister_transport(&xs_bc_tcp_transport);
3130 }
3131
3132 static int param_set_uint_minmax(const char *val,
3133                 const struct kernel_param *kp,
3134                 unsigned int min, unsigned int max)
3135 {
3136         unsigned int num;
3137         int ret;
3138
3139         if (!val)
3140                 return -EINVAL;
3141         ret = kstrtouint(val, 0, &num);
3142         if (ret == -EINVAL || num < min || num > max)
3143                 return -EINVAL;
3144         *((unsigned int *)kp->arg) = num;
3145         return 0;
3146 }
3147
3148 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3149 {
3150         return param_set_uint_minmax(val, kp,
3151                         RPC_MIN_RESVPORT,
3152                         RPC_MAX_RESVPORT);
3153 }
3154
3155 static const struct kernel_param_ops param_ops_portnr = {
3156         .set = param_set_portnr,
3157         .get = param_get_uint,
3158 };
3159
3160 #define param_check_portnr(name, p) \
3161         __param_check(name, p, unsigned int);
3162
3163 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3164 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3165
3166 static int param_set_slot_table_size(const char *val,
3167                                      const struct kernel_param *kp)
3168 {
3169         return param_set_uint_minmax(val, kp,
3170                         RPC_MIN_SLOT_TABLE,
3171                         RPC_MAX_SLOT_TABLE);
3172 }
3173
3174 static const struct kernel_param_ops param_ops_slot_table_size = {
3175         .set = param_set_slot_table_size,
3176         .get = param_get_uint,
3177 };
3178
3179 #define param_check_slot_table_size(name, p) \
3180         __param_check(name, p, unsigned int);
3181
3182 static int param_set_max_slot_table_size(const char *val,
3183                                      const struct kernel_param *kp)
3184 {
3185         return param_set_uint_minmax(val, kp,
3186                         RPC_MIN_SLOT_TABLE,
3187                         RPC_MAX_SLOT_TABLE_LIMIT);
3188 }
3189
3190 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3191         .set = param_set_max_slot_table_size,
3192         .get = param_get_uint,
3193 };
3194
3195 #define param_check_max_slot_table_size(name, p) \
3196         __param_check(name, p, unsigned int);
3197
3198 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3199                    slot_table_size, 0644);
3200 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3201                    max_slot_table_size, 0644);
3202 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3203                    slot_table_size, 0644);
3204