2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
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>
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
21 #include <linux/types.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/pagemap.h>
26 #include <linux/errno.h>
27 #include <linux/socket.h>
29 #include <linux/net.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/sunrpc/xprtsock.h>
36 #include <linux/file.h>
39 #include <net/checksum.h>
46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
53 * We can register our own files under /proc/sys/sunrpc by
54 * calling register_sysctl_table() again. The files in that
55 * directory become the union of all files registered there.
57 * We simply need to make sure that we don't collide with
58 * someone else's file names!
63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
68 static struct ctl_table_header *sunrpc_table_header;
71 * FIXME: changing the UDP slot table size should also resize the UDP
72 * socket buffers for existing UDP transports
74 static ctl_table xs_tunables_table[] = {
76 .ctl_name = CTL_SLOTTABLE_UDP,
77 .procname = "udp_slot_table_entries",
78 .data = &xprt_udp_slot_table_entries,
79 .maxlen = sizeof(unsigned int),
81 .proc_handler = &proc_dointvec_minmax,
82 .strategy = &sysctl_intvec,
83 .extra1 = &min_slot_table_size,
84 .extra2 = &max_slot_table_size
87 .ctl_name = CTL_SLOTTABLE_TCP,
88 .procname = "tcp_slot_table_entries",
89 .data = &xprt_tcp_slot_table_entries,
90 .maxlen = sizeof(unsigned int),
92 .proc_handler = &proc_dointvec_minmax,
93 .strategy = &sysctl_intvec,
94 .extra1 = &min_slot_table_size,
95 .extra2 = &max_slot_table_size
98 .ctl_name = CTL_MIN_RESVPORT,
99 .procname = "min_resvport",
100 .data = &xprt_min_resvport,
101 .maxlen = sizeof(unsigned int),
103 .proc_handler = &proc_dointvec_minmax,
104 .strategy = &sysctl_intvec,
105 .extra1 = &xprt_min_resvport_limit,
106 .extra2 = &xprt_max_resvport_limit
109 .ctl_name = CTL_MAX_RESVPORT,
110 .procname = "max_resvport",
111 .data = &xprt_max_resvport,
112 .maxlen = sizeof(unsigned int),
114 .proc_handler = &proc_dointvec_minmax,
115 .strategy = &sysctl_intvec,
116 .extra1 = &xprt_min_resvport_limit,
117 .extra2 = &xprt_max_resvport_limit
124 static ctl_table sunrpc_table[] = {
126 .ctl_name = CTL_SUNRPC,
127 .procname = "sunrpc",
129 .child = xs_tunables_table
139 * How many times to try sending a request on a socket before waiting
140 * for the socket buffer to clear.
142 #define XS_SENDMSG_RETRY (10U)
145 * Time out for an RPC UDP socket connect. UDP socket connects are
146 * synchronous, but we set a timeout anyway in case of resource
147 * exhaustion on the local host.
149 #define XS_UDP_CONN_TO (5U * HZ)
152 * Wait duration for an RPC TCP connection to be established. Solaris
153 * NFS over TCP uses 60 seconds, for example, which is in line with how
154 * long a server takes to reboot.
156 #define XS_TCP_CONN_TO (60U * HZ)
159 * Wait duration for a reply from the RPC portmapper.
161 #define XS_BIND_TO (60U * HZ)
164 * Delay if a UDP socket connect error occurs. This is most likely some
165 * kind of resource problem on the local host.
167 #define XS_UDP_REEST_TO (2U * HZ)
170 * The reestablish timeout allows clients to delay for a bit before attempting
171 * to reconnect to a server that just dropped our connection.
173 * We implement an exponential backoff when trying to reestablish a TCP
174 * transport connection with the server. Some servers like to drop a TCP
175 * connection when they are overworked, so we start with a short timeout and
176 * increase over time if the server is down or not responding.
178 #define XS_TCP_INIT_REEST_TO (3U * HZ)
179 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
186 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
189 # undef RPC_DEBUG_DATA
190 # define RPCDBG_FACILITY RPCDBG_TRANS
193 #ifdef RPC_DEBUG_DATA
194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196 u8 *buf = (u8 *) packet;
199 dprintk("RPC: %s\n", msg);
200 for (j = 0; j < count && j < 128; j += 4) {
204 dprintk("0x%04x ", j);
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
219 struct rpc_xprt xprt;
224 struct socket * sock;
228 * State of TCP reply receive
236 unsigned long tcp_copied,
240 * Connection of transports
242 struct delayed_work connect_worker;
243 struct sockaddr_storage addr;
247 * UDP socket buffer size parameters
253 * Saved socket callback addresses
255 void (*old_data_ready)(struct sock *, int);
256 void (*old_state_change)(struct sock *);
257 void (*old_write_space)(struct sock *);
261 * TCP receive state flags
263 #define TCP_RCV_LAST_FRAG (1UL << 0)
264 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
265 #define TCP_RCV_COPY_XID (1UL << 2)
266 #define TCP_RCV_COPY_DATA (1UL << 3)
268 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
270 return (struct sockaddr *) &xprt->addr;
273 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
275 return (struct sockaddr_in *) &xprt->addr;
278 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
280 return (struct sockaddr_in6 *) &xprt->addr;
283 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
284 const char *protocol,
287 struct sockaddr_in *addr = xs_addr_in(xprt);
290 buf = kzalloc(20, GFP_KERNEL);
292 snprintf(buf, 20, NIPQUAD_FMT,
293 NIPQUAD(addr->sin_addr.s_addr));
295 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
297 buf = kzalloc(8, GFP_KERNEL);
299 snprintf(buf, 8, "%u",
300 ntohs(addr->sin_port));
302 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
304 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
306 buf = kzalloc(48, GFP_KERNEL);
308 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
309 NIPQUAD(addr->sin_addr.s_addr),
310 ntohs(addr->sin_port),
313 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
315 buf = kzalloc(10, GFP_KERNEL);
317 snprintf(buf, 10, "%02x%02x%02x%02x",
318 NIPQUAD(addr->sin_addr.s_addr));
320 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
322 buf = kzalloc(8, GFP_KERNEL);
324 snprintf(buf, 8, "%4hx",
325 ntohs(addr->sin_port));
327 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
329 buf = kzalloc(30, GFP_KERNEL);
331 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
332 NIPQUAD(addr->sin_addr.s_addr),
333 ntohs(addr->sin_port) >> 8,
334 ntohs(addr->sin_port) & 0xff);
336 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
338 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
341 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
342 const char *protocol,
345 struct sockaddr_in6 *addr = xs_addr_in6(xprt);
348 buf = kzalloc(40, GFP_KERNEL);
350 snprintf(buf, 40, NIP6_FMT,
351 NIP6(addr->sin6_addr));
353 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
355 buf = kzalloc(8, GFP_KERNEL);
357 snprintf(buf, 8, "%u",
358 ntohs(addr->sin6_port));
360 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
362 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
364 buf = kzalloc(64, GFP_KERNEL);
366 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
367 NIP6(addr->sin6_addr),
368 ntohs(addr->sin6_port),
371 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
373 buf = kzalloc(36, GFP_KERNEL);
375 snprintf(buf, 36, NIP6_SEQFMT,
376 NIP6(addr->sin6_addr));
378 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
380 buf = kzalloc(8, GFP_KERNEL);
382 snprintf(buf, 8, "%4hx",
383 ntohs(addr->sin6_port));
385 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
387 buf = kzalloc(50, GFP_KERNEL);
389 snprintf(buf, 50, NIP6_FMT".%u.%u",
390 NIP6(addr->sin6_addr),
391 ntohs(addr->sin6_port) >> 8,
392 ntohs(addr->sin6_port) & 0xff);
394 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
396 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
399 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
403 for (i = 0; i < RPC_DISPLAY_MAX; i++)
405 case RPC_DISPLAY_PROTO:
406 case RPC_DISPLAY_NETID:
409 kfree(xprt->address_strings[i]);
413 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
415 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
417 struct msghdr msg = {
419 .msg_namelen = addrlen,
420 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
423 .iov_base = vec->iov_base + base,
424 .iov_len = vec->iov_len - base,
427 if (iov.iov_len != 0)
428 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
429 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
432 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
435 unsigned int remainder;
438 remainder = xdr->page_len - base;
439 base += xdr->page_base;
440 ppage = xdr->pages + (base >> PAGE_SHIFT);
443 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
444 int flags = XS_SENDMSG_FLAGS;
447 if (remainder != 0 || more)
449 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
450 if (remainder == 0 || err != len)
464 * xs_sendpages - write pages directly to a socket
465 * @sock: socket to send on
466 * @addr: UDP only -- address of destination
467 * @addrlen: UDP only -- length of destination address
468 * @xdr: buffer containing this request
469 * @base: starting position in the buffer
472 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
474 unsigned int remainder = xdr->len - base;
480 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
486 if (base < xdr->head[0].iov_len || addr != NULL) {
487 unsigned int len = xdr->head[0].iov_len - base;
489 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
490 if (remainder == 0 || err != len)
495 base -= xdr->head[0].iov_len;
497 if (base < xdr->page_len) {
498 unsigned int len = xdr->page_len - base;
500 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
501 if (remainder == 0 || err != len)
506 base -= xdr->page_len;
508 if (base >= xdr->tail[0].iov_len)
510 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
519 static void xs_nospace_callback(struct rpc_task *task)
521 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
523 transport->inet->sk_write_pending--;
524 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
528 * xs_nospace - place task on wait queue if transmit was incomplete
529 * @task: task to put to sleep
532 static void xs_nospace(struct rpc_task *task)
534 struct rpc_rqst *req = task->tk_rqstp;
535 struct rpc_xprt *xprt = req->rq_xprt;
536 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
538 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
539 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
542 /* Protect against races with write_space */
543 spin_lock_bh(&xprt->transport_lock);
545 /* Don't race with disconnect */
546 if (xprt_connected(xprt)) {
547 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
549 * Notify TCP that we're limited by the application
552 set_bit(SOCK_NOSPACE, &transport->sock->flags);
553 transport->inet->sk_write_pending++;
554 /* ...and wait for more buffer space */
555 xprt_wait_for_buffer_space(task, xs_nospace_callback);
558 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
559 task->tk_status = -ENOTCONN;
562 spin_unlock_bh(&xprt->transport_lock);
566 * xs_udp_send_request - write an RPC request to a UDP socket
567 * @task: address of RPC task that manages the state of an RPC request
570 * 0: The request has been sent
571 * EAGAIN: The socket was blocked, please call again later to
572 * complete the request
573 * ENOTCONN: Caller needs to invoke connect logic then call again
574 * other: Some other error occured, the request was not sent
576 static int xs_udp_send_request(struct rpc_task *task)
578 struct rpc_rqst *req = task->tk_rqstp;
579 struct rpc_xprt *xprt = req->rq_xprt;
580 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
581 struct xdr_buf *xdr = &req->rq_snd_buf;
584 xs_pktdump("packet data:",
585 req->rq_svec->iov_base,
586 req->rq_svec->iov_len);
588 req->rq_xtime = jiffies;
589 status = xs_sendpages(transport->sock,
594 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
595 xdr->len - req->rq_bytes_sent, status);
598 task->tk_bytes_sent += status;
599 if (status >= req->rq_slen)
601 /* Still some bytes left; set up for a retry later. */
612 /* When the server has died, an ICMP port unreachable message
613 * prompts ECONNREFUSED. */
614 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
617 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
618 dprintk("RPC: sendmsg returned unrecognized error %d\n",
626 * xs_tcp_shutdown - gracefully shut down a TCP socket
629 * Initiates a graceful shutdown of the TCP socket by calling the
630 * equivalent of shutdown(SHUT_WR);
632 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
634 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
635 struct socket *sock = transport->sock;
638 kernel_sock_shutdown(sock, SHUT_WR);
641 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
643 u32 reclen = buf->len - sizeof(rpc_fraghdr);
644 rpc_fraghdr *base = buf->head[0].iov_base;
645 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
649 * xs_tcp_send_request - write an RPC request to a TCP socket
650 * @task: address of RPC task that manages the state of an RPC request
653 * 0: The request has been sent
654 * EAGAIN: The socket was blocked, please call again later to
655 * complete the request
656 * ENOTCONN: Caller needs to invoke connect logic then call again
657 * other: Some other error occured, the request was not sent
659 * XXX: In the case of soft timeouts, should we eventually give up
660 * if sendmsg is not able to make progress?
662 static int xs_tcp_send_request(struct rpc_task *task)
664 struct rpc_rqst *req = task->tk_rqstp;
665 struct rpc_xprt *xprt = req->rq_xprt;
666 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
667 struct xdr_buf *xdr = &req->rq_snd_buf;
669 unsigned int retry = 0;
671 xs_encode_tcp_record_marker(&req->rq_snd_buf);
673 xs_pktdump("packet data:",
674 req->rq_svec->iov_base,
675 req->rq_svec->iov_len);
677 /* Continue transmitting the packet/record. We must be careful
678 * to cope with writespace callbacks arriving _after_ we have
679 * called sendmsg(). */
681 req->rq_xtime = jiffies;
682 status = xs_sendpages(transport->sock,
683 NULL, 0, xdr, req->rq_bytes_sent);
685 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
686 xdr->len - req->rq_bytes_sent, status);
688 if (unlikely(status < 0))
691 /* If we've sent the entire packet, immediately
692 * reset the count of bytes sent. */
693 req->rq_bytes_sent += status;
694 task->tk_bytes_sent += status;
695 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
696 req->rq_bytes_sent = 0;
701 if (retry++ > XS_SENDMSG_RETRY)
714 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
717 dprintk("RPC: sendmsg returned unrecognized error %d\n",
719 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
720 xs_tcp_shutdown(xprt);
727 * xs_tcp_release_xprt - clean up after a tcp transmission
731 * This cleans up if an error causes us to abort the transmission of a request.
732 * In this case, the socket may need to be reset in order to avoid confusing
735 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
737 struct rpc_rqst *req;
739 if (task != xprt->snd_task)
743 req = task->tk_rqstp;
744 if (req->rq_bytes_sent == 0)
746 if (req->rq_bytes_sent == req->rq_snd_buf.len)
748 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
750 xprt_release_xprt(xprt, task);
754 * xs_close - close a socket
757 * This is used when all requests are complete; ie, no DRC state remains
758 * on the server we want to save.
760 static void xs_close(struct rpc_xprt *xprt)
762 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
763 struct socket *sock = transport->sock;
764 struct sock *sk = transport->inet;
767 goto clear_close_wait;
769 dprintk("RPC: xs_close xprt %p\n", xprt);
771 write_lock_bh(&sk->sk_callback_lock);
772 transport->inet = NULL;
773 transport->sock = NULL;
775 sk->sk_user_data = NULL;
776 sk->sk_data_ready = transport->old_data_ready;
777 sk->sk_state_change = transport->old_state_change;
778 sk->sk_write_space = transport->old_write_space;
779 write_unlock_bh(&sk->sk_callback_lock);
785 smp_mb__before_clear_bit();
786 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
787 clear_bit(XPRT_CLOSING, &xprt->state);
788 smp_mb__after_clear_bit();
789 xprt_disconnect_done(xprt);
793 * xs_destroy - prepare to shutdown a transport
794 * @xprt: doomed transport
797 static void xs_destroy(struct rpc_xprt *xprt)
799 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
801 dprintk("RPC: xs_destroy xprt %p\n", xprt);
803 cancel_rearming_delayed_work(&transport->connect_worker);
806 xs_free_peer_addresses(xprt);
809 module_put(THIS_MODULE);
812 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
814 return (struct rpc_xprt *) sk->sk_user_data;
818 * xs_udp_data_ready - "data ready" callback for UDP sockets
819 * @sk: socket with data to read
820 * @len: how much data to read
823 static void xs_udp_data_ready(struct sock *sk, int len)
825 struct rpc_task *task;
826 struct rpc_xprt *xprt;
827 struct rpc_rqst *rovr;
829 int err, repsize, copied;
833 read_lock(&sk->sk_callback_lock);
834 dprintk("RPC: xs_udp_data_ready...\n");
835 if (!(xprt = xprt_from_sock(sk)))
838 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
844 repsize = skb->len - sizeof(struct udphdr);
846 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
850 /* Copy the XID from the skb... */
851 xp = skb_header_pointer(skb, sizeof(struct udphdr),
852 sizeof(_xid), &_xid);
856 /* Look up and lock the request corresponding to the given XID */
857 spin_lock(&xprt->transport_lock);
858 rovr = xprt_lookup_rqst(xprt, *xp);
861 task = rovr->rq_task;
863 if ((copied = rovr->rq_private_buf.buflen) > repsize)
866 /* Suck it into the iovec, verify checksum if not done by hw. */
867 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
868 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
872 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
874 /* Something worked... */
875 dst_confirm(skb->dst);
877 xprt_adjust_cwnd(task, copied);
878 xprt_update_rtt(task);
879 xprt_complete_rqst(task, copied);
882 spin_unlock(&xprt->transport_lock);
884 skb_free_datagram(sk, skb);
886 read_unlock(&sk->sk_callback_lock);
889 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
891 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
895 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
896 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
897 used = xdr_skb_read_bits(desc, p, len);
898 transport->tcp_offset += used;
902 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
903 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
904 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
906 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
907 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
909 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
910 transport->tcp_offset = 0;
912 /* Sanity check of the record length */
913 if (unlikely(transport->tcp_reclen < 4)) {
914 dprintk("RPC: invalid TCP record fragment length\n");
915 xprt_force_disconnect(xprt);
918 dprintk("RPC: reading TCP record fragment of length %d\n",
919 transport->tcp_reclen);
922 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
924 if (transport->tcp_offset == transport->tcp_reclen) {
925 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
926 transport->tcp_offset = 0;
927 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
928 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
929 transport->tcp_flags |= TCP_RCV_COPY_XID;
930 transport->tcp_copied = 0;
935 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
940 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
941 dprintk("RPC: reading XID (%Zu bytes)\n", len);
942 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
943 used = xdr_skb_read_bits(desc, p, len);
944 transport->tcp_offset += used;
947 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
948 transport->tcp_flags |= TCP_RCV_COPY_DATA;
949 transport->tcp_copied = 4;
950 dprintk("RPC: reading reply for XID %08x\n",
951 ntohl(transport->tcp_xid));
952 xs_tcp_check_fraghdr(transport);
955 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
957 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
958 struct rpc_rqst *req;
959 struct xdr_buf *rcvbuf;
963 /* Find and lock the request corresponding to this xid */
964 spin_lock(&xprt->transport_lock);
965 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
967 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
968 dprintk("RPC: XID %08x request not found!\n",
969 ntohl(transport->tcp_xid));
970 spin_unlock(&xprt->transport_lock);
974 rcvbuf = &req->rq_private_buf;
976 if (len > transport->tcp_reclen - transport->tcp_offset) {
977 struct xdr_skb_reader my_desc;
979 len = transport->tcp_reclen - transport->tcp_offset;
980 memcpy(&my_desc, desc, sizeof(my_desc));
982 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
983 &my_desc, xdr_skb_read_bits);
987 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
988 desc, xdr_skb_read_bits);
991 transport->tcp_copied += r;
992 transport->tcp_offset += r;
995 /* Error when copying to the receive buffer,
996 * usually because we weren't able to allocate
997 * additional buffer pages. All we can do now
998 * is turn off TCP_RCV_COPY_DATA, so the request
999 * will not receive any additional updates,
1001 * Any remaining data from this record will
1004 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1005 dprintk("RPC: XID %08x truncated request\n",
1006 ntohl(transport->tcp_xid));
1007 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1008 "tcp_offset = %u, tcp_reclen = %u\n",
1009 xprt, transport->tcp_copied,
1010 transport->tcp_offset, transport->tcp_reclen);
1014 dprintk("RPC: XID %08x read %Zd bytes\n",
1015 ntohl(transport->tcp_xid), r);
1016 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1017 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1018 transport->tcp_offset, transport->tcp_reclen);
1020 if (transport->tcp_copied == req->rq_private_buf.buflen)
1021 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1022 else if (transport->tcp_offset == transport->tcp_reclen) {
1023 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1024 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1028 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1029 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1030 spin_unlock(&xprt->transport_lock);
1031 xs_tcp_check_fraghdr(transport);
1034 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1038 len = transport->tcp_reclen - transport->tcp_offset;
1039 if (len > desc->count)
1042 desc->offset += len;
1043 transport->tcp_offset += len;
1044 dprintk("RPC: discarded %Zu bytes\n", len);
1045 xs_tcp_check_fraghdr(transport);
1048 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1050 struct rpc_xprt *xprt = rd_desc->arg.data;
1051 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1052 struct xdr_skb_reader desc = {
1058 dprintk("RPC: xs_tcp_data_recv started\n");
1060 /* Read in a new fragment marker if necessary */
1061 /* Can we ever really expect to get completely empty fragments? */
1062 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1063 xs_tcp_read_fraghdr(xprt, &desc);
1066 /* Read in the xid if necessary */
1067 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1068 xs_tcp_read_xid(transport, &desc);
1071 /* Read in the request data */
1072 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1073 xs_tcp_read_request(xprt, &desc);
1076 /* Skip over any trailing bytes on short reads */
1077 xs_tcp_read_discard(transport, &desc);
1078 } while (desc.count);
1079 dprintk("RPC: xs_tcp_data_recv done\n");
1080 return len - desc.count;
1084 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1085 * @sk: socket with data to read
1086 * @bytes: how much data to read
1089 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1091 struct rpc_xprt *xprt;
1092 read_descriptor_t rd_desc;
1095 dprintk("RPC: xs_tcp_data_ready...\n");
1097 read_lock(&sk->sk_callback_lock);
1098 if (!(xprt = xprt_from_sock(sk)))
1103 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1104 rd_desc.arg.data = xprt;
1106 rd_desc.count = 65536;
1107 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1110 read_unlock(&sk->sk_callback_lock);
1114 * xs_tcp_state_change - callback to handle TCP socket state changes
1115 * @sk: socket whose state has changed
1118 static void xs_tcp_state_change(struct sock *sk)
1120 struct rpc_xprt *xprt;
1122 read_lock(&sk->sk_callback_lock);
1123 if (!(xprt = xprt_from_sock(sk)))
1125 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1126 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1127 sk->sk_state, xprt_connected(xprt),
1128 sock_flag(sk, SOCK_DEAD),
1129 sock_flag(sk, SOCK_ZAPPED));
1131 switch (sk->sk_state) {
1132 case TCP_ESTABLISHED:
1133 spin_lock_bh(&xprt->transport_lock);
1134 if (!xprt_test_and_set_connected(xprt)) {
1135 struct sock_xprt *transport = container_of(xprt,
1136 struct sock_xprt, xprt);
1138 /* Reset TCP record info */
1139 transport->tcp_offset = 0;
1140 transport->tcp_reclen = 0;
1141 transport->tcp_copied = 0;
1142 transport->tcp_flags =
1143 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1145 xprt_wake_pending_tasks(xprt, 0);
1147 spin_unlock_bh(&xprt->transport_lock);
1150 /* The client initiated a shutdown of the socket */
1151 xprt->reestablish_timeout = 0;
1152 set_bit(XPRT_CLOSING, &xprt->state);
1153 smp_mb__before_clear_bit();
1154 clear_bit(XPRT_CONNECTED, &xprt->state);
1155 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1156 smp_mb__after_clear_bit();
1158 case TCP_CLOSE_WAIT:
1159 /* The server initiated a shutdown of the socket */
1160 set_bit(XPRT_CLOSING, &xprt->state);
1161 xprt_force_disconnect(xprt);
1165 * If the server closed down the connection, make sure that
1166 * we back off before reconnecting
1168 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1169 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1172 smp_mb__before_clear_bit();
1173 clear_bit(XPRT_CONNECTED, &xprt->state);
1174 smp_mb__after_clear_bit();
1177 smp_mb__before_clear_bit();
1178 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1179 clear_bit(XPRT_CLOSING, &xprt->state);
1180 smp_mb__after_clear_bit();
1181 /* Mark transport as closed and wake up all pending tasks */
1182 xprt_disconnect_done(xprt);
1185 read_unlock(&sk->sk_callback_lock);
1189 * xs_udp_write_space - callback invoked when socket buffer space
1191 * @sk: socket whose state has changed
1193 * Called when more output buffer space is available for this socket.
1194 * We try not to wake our writers until they can make "significant"
1195 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1196 * with a bunch of small requests.
1198 static void xs_udp_write_space(struct sock *sk)
1200 read_lock(&sk->sk_callback_lock);
1202 /* from net/core/sock.c:sock_def_write_space */
1203 if (sock_writeable(sk)) {
1204 struct socket *sock;
1205 struct rpc_xprt *xprt;
1207 if (unlikely(!(sock = sk->sk_socket)))
1209 clear_bit(SOCK_NOSPACE, &sock->flags);
1211 if (unlikely(!(xprt = xprt_from_sock(sk))))
1213 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1216 xprt_write_space(xprt);
1220 read_unlock(&sk->sk_callback_lock);
1224 * xs_tcp_write_space - callback invoked when socket buffer space
1226 * @sk: socket whose state has changed
1228 * Called when more output buffer space is available for this socket.
1229 * We try not to wake our writers until they can make "significant"
1230 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1231 * with a bunch of small requests.
1233 static void xs_tcp_write_space(struct sock *sk)
1235 read_lock(&sk->sk_callback_lock);
1237 /* from net/core/stream.c:sk_stream_write_space */
1238 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1239 struct socket *sock;
1240 struct rpc_xprt *xprt;
1242 if (unlikely(!(sock = sk->sk_socket)))
1244 clear_bit(SOCK_NOSPACE, &sock->flags);
1246 if (unlikely(!(xprt = xprt_from_sock(sk))))
1248 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1251 xprt_write_space(xprt);
1255 read_unlock(&sk->sk_callback_lock);
1258 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1260 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1261 struct sock *sk = transport->inet;
1263 if (transport->rcvsize) {
1264 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1265 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1267 if (transport->sndsize) {
1268 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1269 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1270 sk->sk_write_space(sk);
1275 * xs_udp_set_buffer_size - set send and receive limits
1276 * @xprt: generic transport
1277 * @sndsize: requested size of send buffer, in bytes
1278 * @rcvsize: requested size of receive buffer, in bytes
1280 * Set socket send and receive buffer size limits.
1282 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1284 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1286 transport->sndsize = 0;
1288 transport->sndsize = sndsize + 1024;
1289 transport->rcvsize = 0;
1291 transport->rcvsize = rcvsize + 1024;
1293 xs_udp_do_set_buffer_size(xprt);
1297 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1298 * @task: task that timed out
1300 * Adjust the congestion window after a retransmit timeout has occurred.
1302 static void xs_udp_timer(struct rpc_task *task)
1304 xprt_adjust_cwnd(task, -ETIMEDOUT);
1307 static unsigned short xs_get_random_port(void)
1309 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1310 unsigned short rand = (unsigned short) net_random() % range;
1311 return rand + xprt_min_resvport;
1315 * xs_set_port - reset the port number in the remote endpoint address
1316 * @xprt: generic transport
1317 * @port: new port number
1320 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1322 struct sockaddr *addr = xs_addr(xprt);
1324 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1326 switch (addr->sa_family) {
1328 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1331 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1338 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1340 unsigned short port = transport->port;
1342 if (port == 0 && transport->xprt.resvport)
1343 port = xs_get_random_port();
1347 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1349 if (transport->port != 0)
1350 transport->port = 0;
1351 if (!transport->xprt.resvport)
1353 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1354 return xprt_max_resvport;
1358 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1360 struct sockaddr_in myaddr = {
1361 .sin_family = AF_INET,
1363 struct sockaddr_in *sa;
1365 unsigned short port = xs_get_srcport(transport, sock);
1366 unsigned short last;
1368 sa = (struct sockaddr_in *)&transport->addr;
1369 myaddr.sin_addr = sa->sin_addr;
1371 myaddr.sin_port = htons(port);
1372 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1377 transport->port = port;
1381 port = xs_next_srcport(transport, sock, port);
1384 } while (err == -EADDRINUSE && nloop != 2);
1385 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1386 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1387 port, err ? "failed" : "ok", err);
1391 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1393 struct sockaddr_in6 myaddr = {
1394 .sin6_family = AF_INET6,
1396 struct sockaddr_in6 *sa;
1398 unsigned short port = xs_get_srcport(transport, sock);
1399 unsigned short last;
1401 sa = (struct sockaddr_in6 *)&transport->addr;
1402 myaddr.sin6_addr = sa->sin6_addr;
1404 myaddr.sin6_port = htons(port);
1405 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1410 transport->port = port;
1414 port = xs_next_srcport(transport, sock, port);
1417 } while (err == -EADDRINUSE && nloop != 2);
1418 dprintk("RPC: xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1419 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1423 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1424 static struct lock_class_key xs_key[2];
1425 static struct lock_class_key xs_slock_key[2];
1427 static inline void xs_reclassify_socket4(struct socket *sock)
1429 struct sock *sk = sock->sk;
1431 BUG_ON(sock_owned_by_user(sk));
1432 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1433 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1436 static inline void xs_reclassify_socket6(struct socket *sock)
1438 struct sock *sk = sock->sk;
1440 BUG_ON(sock_owned_by_user(sk));
1441 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1442 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1445 static inline void xs_reclassify_socket4(struct socket *sock)
1449 static inline void xs_reclassify_socket6(struct socket *sock)
1454 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1456 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1458 if (!transport->inet) {
1459 struct sock *sk = sock->sk;
1461 write_lock_bh(&sk->sk_callback_lock);
1463 sk->sk_user_data = xprt;
1464 transport->old_data_ready = sk->sk_data_ready;
1465 transport->old_state_change = sk->sk_state_change;
1466 transport->old_write_space = sk->sk_write_space;
1467 sk->sk_data_ready = xs_udp_data_ready;
1468 sk->sk_write_space = xs_udp_write_space;
1469 sk->sk_no_check = UDP_CSUM_NORCV;
1470 sk->sk_allocation = GFP_ATOMIC;
1472 xprt_set_connected(xprt);
1474 /* Reset to new socket */
1475 transport->sock = sock;
1476 transport->inet = sk;
1478 write_unlock_bh(&sk->sk_callback_lock);
1480 xs_udp_do_set_buffer_size(xprt);
1484 * xs_udp_connect_worker4 - set up a UDP socket
1485 * @work: RPC transport to connect
1487 * Invoked by a work queue tasklet.
1489 static void xs_udp_connect_worker4(struct work_struct *work)
1491 struct sock_xprt *transport =
1492 container_of(work, struct sock_xprt, connect_worker.work);
1493 struct rpc_xprt *xprt = &transport->xprt;
1494 struct socket *sock = transport->sock;
1495 int err, status = -EIO;
1497 if (xprt->shutdown || !xprt_bound(xprt))
1500 /* Start by resetting any existing state */
1503 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1504 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1507 xs_reclassify_socket4(sock);
1509 if (xs_bind4(transport, sock)) {
1514 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1515 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1517 xs_udp_finish_connecting(xprt, sock);
1520 xprt_wake_pending_tasks(xprt, status);
1521 xprt_clear_connecting(xprt);
1525 * xs_udp_connect_worker6 - set up a UDP socket
1526 * @work: RPC transport to connect
1528 * Invoked by a work queue tasklet.
1530 static void xs_udp_connect_worker6(struct work_struct *work)
1532 struct sock_xprt *transport =
1533 container_of(work, struct sock_xprt, connect_worker.work);
1534 struct rpc_xprt *xprt = &transport->xprt;
1535 struct socket *sock = transport->sock;
1536 int err, status = -EIO;
1538 if (xprt->shutdown || !xprt_bound(xprt))
1541 /* Start by resetting any existing state */
1544 if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1545 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1548 xs_reclassify_socket6(sock);
1550 if (xs_bind6(transport, sock) < 0) {
1555 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1556 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1558 xs_udp_finish_connecting(xprt, sock);
1561 xprt_wake_pending_tasks(xprt, status);
1562 xprt_clear_connecting(xprt);
1566 * We need to preserve the port number so the reply cache on the server can
1567 * find our cached RPC replies when we get around to reconnecting.
1569 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1572 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1573 struct sockaddr any;
1575 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1578 * Disconnect the transport socket by doing a connect operation
1579 * with AF_UNSPEC. This should return immediately...
1581 memset(&any, 0, sizeof(any));
1582 any.sa_family = AF_UNSPEC;
1583 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1585 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1589 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1591 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1593 if (!transport->inet) {
1594 struct sock *sk = sock->sk;
1596 write_lock_bh(&sk->sk_callback_lock);
1598 sk->sk_user_data = xprt;
1599 transport->old_data_ready = sk->sk_data_ready;
1600 transport->old_state_change = sk->sk_state_change;
1601 transport->old_write_space = sk->sk_write_space;
1602 sk->sk_data_ready = xs_tcp_data_ready;
1603 sk->sk_state_change = xs_tcp_state_change;
1604 sk->sk_write_space = xs_tcp_write_space;
1605 sk->sk_allocation = GFP_ATOMIC;
1607 /* socket options */
1608 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1609 sock_reset_flag(sk, SOCK_LINGER);
1610 tcp_sk(sk)->linger2 = 0;
1611 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1613 xprt_clear_connected(xprt);
1615 /* Reset to new socket */
1616 transport->sock = sock;
1617 transport->inet = sk;
1619 write_unlock_bh(&sk->sk_callback_lock);
1622 /* Tell the socket layer to start connecting... */
1623 xprt->stat.connect_count++;
1624 xprt->stat.connect_start = jiffies;
1625 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1629 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1630 * @work: RPC transport to connect
1632 * Invoked by a work queue tasklet.
1634 static void xs_tcp_connect_worker4(struct work_struct *work)
1636 struct sock_xprt *transport =
1637 container_of(work, struct sock_xprt, connect_worker.work);
1638 struct rpc_xprt *xprt = &transport->xprt;
1639 struct socket *sock = transport->sock;
1640 int err, status = -EIO;
1642 if (xprt->shutdown || !xprt_bound(xprt))
1646 /* start from scratch */
1647 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1648 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1651 xs_reclassify_socket4(sock);
1653 if (xs_bind4(transport, sock) < 0) {
1658 /* "close" the socket, preserving the local port */
1659 xs_tcp_reuse_connection(xprt);
1661 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1662 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1664 status = xs_tcp_finish_connecting(xprt, sock);
1665 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1666 xprt, -status, xprt_connected(xprt),
1667 sock->sk->sk_state);
1675 /* retry with existing socket, after a delay */
1678 /* get rid of existing socket, and retry */
1679 xs_tcp_shutdown(xprt);
1683 xprt_wake_pending_tasks(xprt, status);
1685 xprt_clear_connecting(xprt);
1689 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1690 * @work: RPC transport to connect
1692 * Invoked by a work queue tasklet.
1694 static void xs_tcp_connect_worker6(struct work_struct *work)
1696 struct sock_xprt *transport =
1697 container_of(work, struct sock_xprt, connect_worker.work);
1698 struct rpc_xprt *xprt = &transport->xprt;
1699 struct socket *sock = transport->sock;
1700 int err, status = -EIO;
1702 if (xprt->shutdown || !xprt_bound(xprt))
1706 /* start from scratch */
1707 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1708 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1711 xs_reclassify_socket6(sock);
1713 if (xs_bind6(transport, sock) < 0) {
1718 /* "close" the socket, preserving the local port */
1719 xs_tcp_reuse_connection(xprt);
1721 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1722 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1724 status = xs_tcp_finish_connecting(xprt, sock);
1725 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1726 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1734 /* retry with existing socket, after a delay */
1737 /* get rid of existing socket, and retry */
1738 xs_tcp_shutdown(xprt);
1742 xprt_wake_pending_tasks(xprt, status);
1744 xprt_clear_connecting(xprt);
1748 * xs_connect - connect a socket to a remote endpoint
1749 * @task: address of RPC task that manages state of connect request
1751 * TCP: If the remote end dropped the connection, delay reconnecting.
1753 * UDP socket connects are synchronous, but we use a work queue anyway
1754 * to guarantee that even unprivileged user processes can set up a
1755 * socket on a privileged port.
1757 * If a UDP socket connect fails, the delay behavior here prevents
1758 * retry floods (hard mounts).
1760 static void xs_connect(struct rpc_task *task)
1762 struct rpc_xprt *xprt = task->tk_xprt;
1763 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1765 if (xprt_test_and_set_connecting(xprt))
1768 if (transport->sock != NULL) {
1769 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1771 xprt, xprt->reestablish_timeout / HZ);
1772 queue_delayed_work(rpciod_workqueue,
1773 &transport->connect_worker,
1774 xprt->reestablish_timeout);
1775 xprt->reestablish_timeout <<= 1;
1776 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1777 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1779 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1780 queue_delayed_work(rpciod_workqueue,
1781 &transport->connect_worker, 0);
1785 static void xs_tcp_connect(struct rpc_task *task)
1787 struct rpc_xprt *xprt = task->tk_xprt;
1789 /* Initiate graceful shutdown of the socket if not already done */
1790 if (test_bit(XPRT_CONNECTED, &xprt->state))
1791 xs_tcp_shutdown(xprt);
1792 /* Exit if we need to wait for socket shutdown to complete */
1793 if (test_bit(XPRT_CLOSING, &xprt->state))
1799 * xs_udp_print_stats - display UDP socket-specifc stats
1800 * @xprt: rpc_xprt struct containing statistics
1804 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1806 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1808 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1810 xprt->stat.bind_count,
1813 xprt->stat.bad_xids,
1815 xprt->stat.bklog_u);
1819 * xs_tcp_print_stats - display TCP socket-specifc stats
1820 * @xprt: rpc_xprt struct containing statistics
1824 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1826 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1829 if (xprt_connected(xprt))
1830 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1832 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1834 xprt->stat.bind_count,
1835 xprt->stat.connect_count,
1836 xprt->stat.connect_time,
1840 xprt->stat.bad_xids,
1842 xprt->stat.bklog_u);
1845 static struct rpc_xprt_ops xs_udp_ops = {
1846 .set_buffer_size = xs_udp_set_buffer_size,
1847 .reserve_xprt = xprt_reserve_xprt_cong,
1848 .release_xprt = xprt_release_xprt_cong,
1849 .rpcbind = rpcb_getport_async,
1850 .set_port = xs_set_port,
1851 .connect = xs_connect,
1852 .buf_alloc = rpc_malloc,
1853 .buf_free = rpc_free,
1854 .send_request = xs_udp_send_request,
1855 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1856 .timer = xs_udp_timer,
1857 .release_request = xprt_release_rqst_cong,
1859 .destroy = xs_destroy,
1860 .print_stats = xs_udp_print_stats,
1863 static struct rpc_xprt_ops xs_tcp_ops = {
1864 .reserve_xprt = xprt_reserve_xprt,
1865 .release_xprt = xs_tcp_release_xprt,
1866 .rpcbind = rpcb_getport_async,
1867 .set_port = xs_set_port,
1868 .connect = xs_tcp_connect,
1869 .buf_alloc = rpc_malloc,
1870 .buf_free = rpc_free,
1871 .send_request = xs_tcp_send_request,
1872 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1873 .close = xs_tcp_shutdown,
1874 .destroy = xs_destroy,
1875 .print_stats = xs_tcp_print_stats,
1878 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1879 unsigned int slot_table_size)
1881 struct rpc_xprt *xprt;
1882 struct sock_xprt *new;
1884 if (args->addrlen > sizeof(xprt->addr)) {
1885 dprintk("RPC: xs_setup_xprt: address too large\n");
1886 return ERR_PTR(-EBADF);
1889 new = kzalloc(sizeof(*new), GFP_KERNEL);
1891 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1893 return ERR_PTR(-ENOMEM);
1897 xprt->max_reqs = slot_table_size;
1898 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1899 if (xprt->slot == NULL) {
1901 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1903 return ERR_PTR(-ENOMEM);
1906 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1907 xprt->addrlen = args->addrlen;
1909 memcpy(&new->addr, args->srcaddr, args->addrlen);
1914 static const struct rpc_timeout xs_udp_default_timeout = {
1915 .to_initval = 5 * HZ,
1916 .to_maxval = 30 * HZ,
1917 .to_increment = 5 * HZ,
1922 * xs_setup_udp - Set up transport to use a UDP socket
1923 * @args: rpc transport creation arguments
1926 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1928 struct sockaddr *addr = args->dstaddr;
1929 struct rpc_xprt *xprt;
1930 struct sock_xprt *transport;
1932 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1935 transport = container_of(xprt, struct sock_xprt, xprt);
1937 xprt->prot = IPPROTO_UDP;
1939 /* XXX: header size can vary due to auth type, IPv6, etc. */
1940 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1942 xprt->bind_timeout = XS_BIND_TO;
1943 xprt->connect_timeout = XS_UDP_CONN_TO;
1944 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1945 xprt->idle_timeout = XS_IDLE_DISC_TO;
1947 xprt->ops = &xs_udp_ops;
1949 xprt->timeout = &xs_udp_default_timeout;
1951 switch (addr->sa_family) {
1953 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1954 xprt_set_bound(xprt);
1956 INIT_DELAYED_WORK(&transport->connect_worker,
1957 xs_udp_connect_worker4);
1958 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1961 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1962 xprt_set_bound(xprt);
1964 INIT_DELAYED_WORK(&transport->connect_worker,
1965 xs_udp_connect_worker6);
1966 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1970 return ERR_PTR(-EAFNOSUPPORT);
1973 dprintk("RPC: set up transport to address %s\n",
1974 xprt->address_strings[RPC_DISPLAY_ALL]);
1976 if (try_module_get(THIS_MODULE))
1981 return ERR_PTR(-EINVAL);
1984 static const struct rpc_timeout xs_tcp_default_timeout = {
1985 .to_initval = 60 * HZ,
1986 .to_maxval = 60 * HZ,
1991 * xs_setup_tcp - Set up transport to use a TCP socket
1992 * @args: rpc transport creation arguments
1995 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1997 struct sockaddr *addr = args->dstaddr;
1998 struct rpc_xprt *xprt;
1999 struct sock_xprt *transport;
2001 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2004 transport = container_of(xprt, struct sock_xprt, xprt);
2006 xprt->prot = IPPROTO_TCP;
2007 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2008 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2010 xprt->bind_timeout = XS_BIND_TO;
2011 xprt->connect_timeout = XS_TCP_CONN_TO;
2012 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2013 xprt->idle_timeout = XS_IDLE_DISC_TO;
2015 xprt->ops = &xs_tcp_ops;
2016 xprt->timeout = &xs_tcp_default_timeout;
2018 switch (addr->sa_family) {
2020 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2021 xprt_set_bound(xprt);
2023 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2024 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2027 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2028 xprt_set_bound(xprt);
2030 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2031 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2035 return ERR_PTR(-EAFNOSUPPORT);
2038 dprintk("RPC: set up transport to address %s\n",
2039 xprt->address_strings[RPC_DISPLAY_ALL]);
2041 if (try_module_get(THIS_MODULE))
2046 return ERR_PTR(-EINVAL);
2049 static struct xprt_class xs_udp_transport = {
2050 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2052 .owner = THIS_MODULE,
2053 .ident = IPPROTO_UDP,
2054 .setup = xs_setup_udp,
2057 static struct xprt_class xs_tcp_transport = {
2058 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2060 .owner = THIS_MODULE,
2061 .ident = IPPROTO_TCP,
2062 .setup = xs_setup_tcp,
2066 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2069 int init_socket_xprt(void)
2072 if (!sunrpc_table_header)
2073 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2076 xprt_register_transport(&xs_udp_transport);
2077 xprt_register_transport(&xs_tcp_transport);
2083 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2086 void cleanup_socket_xprt(void)
2089 if (sunrpc_table_header) {
2090 unregister_sysctl_table(sunrpc_table_header);
2091 sunrpc_table_header = NULL;
2095 xprt_unregister_transport(&xs_udp_transport);
2096 xprt_unregister_transport(&xs_tcp_transport);