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>
18 #include <linux/types.h>
19 #include <linux/slab.h>
20 #include <linux/capability.h>
21 #include <linux/sched.h>
22 #include <linux/pagemap.h>
23 #include <linux/errno.h>
24 #include <linux/socket.h>
26 #include <linux/net.h>
28 #include <linux/udp.h>
29 #include <linux/tcp.h>
30 #include <linux/sunrpc/clnt.h>
31 #include <linux/sunrpc/sched.h>
32 #include <linux/file.h>
35 #include <net/checksum.h>
42 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
43 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
45 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
46 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
49 * How many times to try sending a request on a socket before waiting
50 * for the socket buffer to clear.
52 #define XS_SENDMSG_RETRY (10U)
55 * Time out for an RPC UDP socket connect. UDP socket connects are
56 * synchronous, but we set a timeout anyway in case of resource
57 * exhaustion on the local host.
59 #define XS_UDP_CONN_TO (5U * HZ)
62 * Wait duration for an RPC TCP connection to be established. Solaris
63 * NFS over TCP uses 60 seconds, for example, which is in line with how
64 * long a server takes to reboot.
66 #define XS_TCP_CONN_TO (60U * HZ)
69 * Wait duration for a reply from the RPC portmapper.
71 #define XS_BIND_TO (60U * HZ)
74 * Delay if a UDP socket connect error occurs. This is most likely some
75 * kind of resource problem on the local host.
77 #define XS_UDP_REEST_TO (2U * HZ)
80 * The reestablish timeout allows clients to delay for a bit before attempting
81 * to reconnect to a server that just dropped our connection.
83 * We implement an exponential backoff when trying to reestablish a TCP
84 * transport connection with the server. Some servers like to drop a TCP
85 * connection when they are overworked, so we start with a short timeout and
86 * increase over time if the server is down or not responding.
88 #define XS_TCP_INIT_REEST_TO (3U * HZ)
89 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
92 * TCP idle timeout; client drops the transport socket if it is idle
93 * for this long. Note that we also timeout UDP sockets to prevent
94 * holding port numbers when there is no RPC traffic.
96 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
99 # undef RPC_DEBUG_DATA
100 # define RPCDBG_FACILITY RPCDBG_TRANS
103 #ifdef RPC_DEBUG_DATA
104 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
106 u8 *buf = (u8 *) packet;
109 dprintk("RPC: %s\n", msg);
110 for (j = 0; j < count && j < 128; j += 4) {
114 dprintk("0x%04x ", j);
116 dprintk("%02x%02x%02x%02x ",
117 buf[j], buf[j+1], buf[j+2], buf[j+3]);
122 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
128 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
130 struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
133 buf = kzalloc(20, GFP_KERNEL);
135 snprintf(buf, 20, "%u.%u.%u.%u",
136 NIPQUAD(addr->sin_addr.s_addr));
138 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
140 buf = kzalloc(8, GFP_KERNEL);
142 snprintf(buf, 8, "%u",
143 ntohs(addr->sin_port));
145 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
147 if (xprt->prot == IPPROTO_UDP)
148 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
150 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
152 buf = kzalloc(48, GFP_KERNEL);
154 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
155 NIPQUAD(addr->sin_addr.s_addr),
156 ntohs(addr->sin_port),
157 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
159 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
162 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
164 kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
165 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
166 kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
169 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
171 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
173 struct msghdr msg = {
175 .msg_namelen = addrlen,
176 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
179 .iov_base = vec->iov_base + base,
180 .iov_len = vec->iov_len - base,
183 if (iov.iov_len != 0)
184 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
185 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
188 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
191 unsigned int remainder;
194 remainder = xdr->page_len - base;
195 base += xdr->page_base;
196 ppage = xdr->pages + (base >> PAGE_SHIFT);
199 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
200 int flags = XS_SENDMSG_FLAGS;
203 if (remainder != 0 || more)
205 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
206 if (remainder == 0 || err != len)
220 * xs_sendpages - write pages directly to a socket
221 * @sock: socket to send on
222 * @addr: UDP only -- address of destination
223 * @addrlen: UDP only -- length of destination address
224 * @xdr: buffer containing this request
225 * @base: starting position in the buffer
228 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
230 unsigned int remainder = xdr->len - base;
236 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
242 if (base < xdr->head[0].iov_len || addr != NULL) {
243 unsigned int len = xdr->head[0].iov_len - base;
245 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
246 if (remainder == 0 || err != len)
251 base -= xdr->head[0].iov_len;
253 if (base < xdr->page_len) {
254 unsigned int len = xdr->page_len - base;
256 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
257 if (remainder == 0 || err != len)
262 base -= xdr->page_len;
264 if (base >= xdr->tail[0].iov_len)
266 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
276 * xs_nospace - place task on wait queue if transmit was incomplete
277 * @task: task to put to sleep
280 static void xs_nospace(struct rpc_task *task)
282 struct rpc_rqst *req = task->tk_rqstp;
283 struct rpc_xprt *xprt = req->rq_xprt;
285 dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
286 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
289 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
290 /* Protect against races with write_space */
291 spin_lock_bh(&xprt->transport_lock);
293 /* Don't race with disconnect */
294 if (!xprt_connected(xprt))
295 task->tk_status = -ENOTCONN;
296 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
297 xprt_wait_for_buffer_space(task);
299 spin_unlock_bh(&xprt->transport_lock);
301 /* Keep holding the socket if it is blocked */
302 rpc_delay(task, HZ>>4);
306 * xs_udp_send_request - write an RPC request to a UDP socket
307 * @task: address of RPC task that manages the state of an RPC request
310 * 0: The request has been sent
311 * EAGAIN: The socket was blocked, please call again later to
312 * complete the request
313 * ENOTCONN: Caller needs to invoke connect logic then call again
314 * other: Some other error occured, the request was not sent
316 static int xs_udp_send_request(struct rpc_task *task)
318 struct rpc_rqst *req = task->tk_rqstp;
319 struct rpc_xprt *xprt = req->rq_xprt;
320 struct xdr_buf *xdr = &req->rq_snd_buf;
323 xs_pktdump("packet data:",
324 req->rq_svec->iov_base,
325 req->rq_svec->iov_len);
327 req->rq_xtime = jiffies;
328 status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
329 xprt->addrlen, xdr, req->rq_bytes_sent);
331 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
332 xdr->len - req->rq_bytes_sent, status);
334 if (likely(status >= (int) req->rq_slen))
337 /* Still some bytes left; set up for a retry later. */
345 /* When the server has died, an ICMP port unreachable message
346 * prompts ECONNREFUSED. */
352 dprintk("RPC: sendmsg returned unrecognized error %d\n",
360 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
362 u32 reclen = buf->len - sizeof(rpc_fraghdr);
363 rpc_fraghdr *base = buf->head[0].iov_base;
364 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
368 * xs_tcp_send_request - write an RPC request to a TCP socket
369 * @task: address of RPC task that manages the state of an RPC request
372 * 0: The request has been sent
373 * EAGAIN: The socket was blocked, please call again later to
374 * complete the request
375 * ENOTCONN: Caller needs to invoke connect logic then call again
376 * other: Some other error occured, the request was not sent
378 * XXX: In the case of soft timeouts, should we eventually give up
379 * if sendmsg is not able to make progress?
381 static int xs_tcp_send_request(struct rpc_task *task)
383 struct rpc_rqst *req = task->tk_rqstp;
384 struct rpc_xprt *xprt = req->rq_xprt;
385 struct xdr_buf *xdr = &req->rq_snd_buf;
386 int status, retry = 0;
388 xs_encode_tcp_record_marker(&req->rq_snd_buf);
390 xs_pktdump("packet data:",
391 req->rq_svec->iov_base,
392 req->rq_svec->iov_len);
394 /* Continue transmitting the packet/record. We must be careful
395 * to cope with writespace callbacks arriving _after_ we have
396 * called sendmsg(). */
398 req->rq_xtime = jiffies;
399 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
402 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
403 xdr->len - req->rq_bytes_sent, status);
405 if (unlikely(status < 0))
408 /* If we've sent the entire packet, immediately
409 * reset the count of bytes sent. */
410 req->rq_bytes_sent += status;
411 task->tk_bytes_sent += status;
412 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
413 req->rq_bytes_sent = 0;
418 if (retry++ > XS_SENDMSG_RETRY)
433 dprintk("RPC: sendmsg returned unrecognized error %d\n",
435 xprt_disconnect(xprt);
443 * xs_tcp_release_xprt - clean up after a tcp transmission
447 * This cleans up if an error causes us to abort the transmission of a request.
448 * In this case, the socket may need to be reset in order to avoid confusing
451 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
453 struct rpc_rqst *req;
455 if (task != xprt->snd_task)
459 req = task->tk_rqstp;
460 if (req->rq_bytes_sent == 0)
462 if (req->rq_bytes_sent == req->rq_snd_buf.len)
464 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
466 xprt_release_xprt(xprt, task);
470 * xs_close - close a socket
473 * This is used when all requests are complete; ie, no DRC state remains
474 * on the server we want to save.
476 static void xs_close(struct rpc_xprt *xprt)
478 struct socket *sock = xprt->sock;
479 struct sock *sk = xprt->inet;
482 goto clear_close_wait;
484 dprintk("RPC: xs_close xprt %p\n", xprt);
486 write_lock_bh(&sk->sk_callback_lock);
490 sk->sk_user_data = NULL;
491 sk->sk_data_ready = xprt->old_data_ready;
492 sk->sk_state_change = xprt->old_state_change;
493 sk->sk_write_space = xprt->old_write_space;
494 write_unlock_bh(&sk->sk_callback_lock);
500 smp_mb__before_clear_bit();
501 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
502 smp_mb__after_clear_bit();
506 * xs_destroy - prepare to shutdown a transport
507 * @xprt: doomed transport
510 static void xs_destroy(struct rpc_xprt *xprt)
512 dprintk("RPC: xs_destroy xprt %p\n", xprt);
514 cancel_delayed_work(&xprt->connect_worker);
515 flush_scheduled_work();
517 xprt_disconnect(xprt);
519 xs_free_peer_addresses(xprt);
524 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
526 return (struct rpc_xprt *) sk->sk_user_data;
530 * xs_udp_data_ready - "data ready" callback for UDP sockets
531 * @sk: socket with data to read
532 * @len: how much data to read
535 static void xs_udp_data_ready(struct sock *sk, int len)
537 struct rpc_task *task;
538 struct rpc_xprt *xprt;
539 struct rpc_rqst *rovr;
541 int err, repsize, copied;
545 read_lock(&sk->sk_callback_lock);
546 dprintk("RPC: xs_udp_data_ready...\n");
547 if (!(xprt = xprt_from_sock(sk)))
550 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
556 repsize = skb->len - sizeof(struct udphdr);
558 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
562 /* Copy the XID from the skb... */
563 xp = skb_header_pointer(skb, sizeof(struct udphdr),
564 sizeof(_xid), &_xid);
568 /* Look up and lock the request corresponding to the given XID */
569 spin_lock(&xprt->transport_lock);
570 rovr = xprt_lookup_rqst(xprt, *xp);
573 task = rovr->rq_task;
575 if ((copied = rovr->rq_private_buf.buflen) > repsize)
578 /* Suck it into the iovec, verify checksum if not done by hw. */
579 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
582 /* Something worked... */
583 dst_confirm(skb->dst);
585 xprt_adjust_cwnd(task, copied);
586 xprt_update_rtt(task);
587 xprt_complete_rqst(task, copied);
590 spin_unlock(&xprt->transport_lock);
592 skb_free_datagram(sk, skb);
594 read_unlock(&sk->sk_callback_lock);
597 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
599 if (len > desc->count)
601 if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
602 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
608 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
613 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
618 p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
619 len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
620 used = xs_tcp_copy_data(desc, p, len);
621 xprt->tcp_offset += used;
625 xprt->tcp_reclen = ntohl(xprt->tcp_recm);
626 if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
627 xprt->tcp_flags |= XPRT_LAST_FRAG;
629 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
630 xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
632 xprt->tcp_flags &= ~XPRT_COPY_RECM;
633 xprt->tcp_offset = 0;
635 /* Sanity check of the record length */
636 if (unlikely(xprt->tcp_reclen < 4)) {
637 dprintk("RPC: invalid TCP record fragment length\n");
638 xprt_disconnect(xprt);
641 dprintk("RPC: reading TCP record fragment of length %d\n",
645 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
647 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
648 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
649 if (xprt->tcp_offset == xprt->tcp_reclen) {
650 xprt->tcp_flags |= XPRT_COPY_RECM;
651 xprt->tcp_offset = 0;
652 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
653 xprt->tcp_flags &= ~XPRT_COPY_DATA;
654 xprt->tcp_flags |= XPRT_COPY_XID;
655 xprt->tcp_copied = 0;
660 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
665 len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
666 dprintk("RPC: reading XID (%Zu bytes)\n", len);
667 p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
668 used = xs_tcp_copy_data(desc, p, len);
669 xprt->tcp_offset += used;
672 xprt->tcp_flags &= ~XPRT_COPY_XID;
673 xprt->tcp_flags |= XPRT_COPY_DATA;
674 xprt->tcp_copied = 4;
675 dprintk("RPC: reading reply for XID %08x\n",
676 ntohl(xprt->tcp_xid));
677 xs_tcp_check_recm(xprt);
680 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
682 struct rpc_rqst *req;
683 struct xdr_buf *rcvbuf;
687 /* Find and lock the request corresponding to this xid */
688 spin_lock(&xprt->transport_lock);
689 req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
691 xprt->tcp_flags &= ~XPRT_COPY_DATA;
692 dprintk("RPC: XID %08x request not found!\n",
693 ntohl(xprt->tcp_xid));
694 spin_unlock(&xprt->transport_lock);
698 rcvbuf = &req->rq_private_buf;
700 if (len > xprt->tcp_reclen - xprt->tcp_offset) {
701 skb_reader_t my_desc;
703 len = xprt->tcp_reclen - xprt->tcp_offset;
704 memcpy(&my_desc, desc, sizeof(my_desc));
706 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
707 &my_desc, xs_tcp_copy_data);
711 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
712 desc, xs_tcp_copy_data);
715 xprt->tcp_copied += r;
716 xprt->tcp_offset += r;
719 /* Error when copying to the receive buffer,
720 * usually because we weren't able to allocate
721 * additional buffer pages. All we can do now
722 * is turn off XPRT_COPY_DATA, so the request
723 * will not receive any additional updates,
725 * Any remaining data from this record will
728 xprt->tcp_flags &= ~XPRT_COPY_DATA;
729 dprintk("RPC: XID %08x truncated request\n",
730 ntohl(xprt->tcp_xid));
731 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
732 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
736 dprintk("RPC: XID %08x read %Zd bytes\n",
737 ntohl(xprt->tcp_xid), r);
738 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
739 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
741 if (xprt->tcp_copied == req->rq_private_buf.buflen)
742 xprt->tcp_flags &= ~XPRT_COPY_DATA;
743 else if (xprt->tcp_offset == xprt->tcp_reclen) {
744 if (xprt->tcp_flags & XPRT_LAST_FRAG)
745 xprt->tcp_flags &= ~XPRT_COPY_DATA;
749 if (!(xprt->tcp_flags & XPRT_COPY_DATA))
750 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
751 spin_unlock(&xprt->transport_lock);
752 xs_tcp_check_recm(xprt);
755 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
759 len = xprt->tcp_reclen - xprt->tcp_offset;
760 if (len > desc->count)
764 xprt->tcp_offset += len;
765 dprintk("RPC: discarded %Zu bytes\n", len);
766 xs_tcp_check_recm(xprt);
769 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
771 struct rpc_xprt *xprt = rd_desc->arg.data;
772 skb_reader_t desc = {
779 dprintk("RPC: xs_tcp_data_recv started\n");
781 /* Read in a new fragment marker if necessary */
782 /* Can we ever really expect to get completely empty fragments? */
783 if (xprt->tcp_flags & XPRT_COPY_RECM) {
784 xs_tcp_read_fraghdr(xprt, &desc);
787 /* Read in the xid if necessary */
788 if (xprt->tcp_flags & XPRT_COPY_XID) {
789 xs_tcp_read_xid(xprt, &desc);
792 /* Read in the request data */
793 if (xprt->tcp_flags & XPRT_COPY_DATA) {
794 xs_tcp_read_request(xprt, &desc);
797 /* Skip over any trailing bytes on short reads */
798 xs_tcp_read_discard(xprt, &desc);
799 } while (desc.count);
800 dprintk("RPC: xs_tcp_data_recv done\n");
801 return len - desc.count;
805 * xs_tcp_data_ready - "data ready" callback for TCP sockets
806 * @sk: socket with data to read
807 * @bytes: how much data to read
810 static void xs_tcp_data_ready(struct sock *sk, int bytes)
812 struct rpc_xprt *xprt;
813 read_descriptor_t rd_desc;
815 read_lock(&sk->sk_callback_lock);
816 dprintk("RPC: xs_tcp_data_ready...\n");
817 if (!(xprt = xprt_from_sock(sk)))
822 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
823 rd_desc.arg.data = xprt;
824 rd_desc.count = 65536;
825 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
827 read_unlock(&sk->sk_callback_lock);
831 * xs_tcp_state_change - callback to handle TCP socket state changes
832 * @sk: socket whose state has changed
835 static void xs_tcp_state_change(struct sock *sk)
837 struct rpc_xprt *xprt;
839 read_lock(&sk->sk_callback_lock);
840 if (!(xprt = xprt_from_sock(sk)))
842 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
843 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
844 sk->sk_state, xprt_connected(xprt),
845 sock_flag(sk, SOCK_DEAD),
846 sock_flag(sk, SOCK_ZAPPED));
848 switch (sk->sk_state) {
849 case TCP_ESTABLISHED:
850 spin_lock_bh(&xprt->transport_lock);
851 if (!xprt_test_and_set_connected(xprt)) {
852 /* Reset TCP record info */
853 xprt->tcp_offset = 0;
854 xprt->tcp_reclen = 0;
855 xprt->tcp_copied = 0;
856 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
857 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
858 xprt_wake_pending_tasks(xprt, 0);
860 spin_unlock_bh(&xprt->transport_lock);
866 /* Try to schedule an autoclose RPC calls */
867 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
868 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
869 schedule_work(&xprt->task_cleanup);
871 xprt_disconnect(xprt);
874 read_unlock(&sk->sk_callback_lock);
878 * xs_udp_write_space - callback invoked when socket buffer space
880 * @sk: socket whose state has changed
882 * Called when more output buffer space is available for this socket.
883 * We try not to wake our writers until they can make "significant"
884 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
885 * with a bunch of small requests.
887 static void xs_udp_write_space(struct sock *sk)
889 read_lock(&sk->sk_callback_lock);
891 /* from net/core/sock.c:sock_def_write_space */
892 if (sock_writeable(sk)) {
894 struct rpc_xprt *xprt;
896 if (unlikely(!(sock = sk->sk_socket)))
898 if (unlikely(!(xprt = xprt_from_sock(sk))))
900 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
903 xprt_write_space(xprt);
907 read_unlock(&sk->sk_callback_lock);
911 * xs_tcp_write_space - callback invoked when socket buffer space
913 * @sk: socket whose state has changed
915 * Called when more output buffer space is available for this socket.
916 * We try not to wake our writers until they can make "significant"
917 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
918 * with a bunch of small requests.
920 static void xs_tcp_write_space(struct sock *sk)
922 read_lock(&sk->sk_callback_lock);
924 /* from net/core/stream.c:sk_stream_write_space */
925 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
927 struct rpc_xprt *xprt;
929 if (unlikely(!(sock = sk->sk_socket)))
931 if (unlikely(!(xprt = xprt_from_sock(sk))))
933 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
936 xprt_write_space(xprt);
940 read_unlock(&sk->sk_callback_lock);
943 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
945 struct sock *sk = xprt->inet;
948 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
949 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2;
952 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
953 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
954 sk->sk_write_space(sk);
959 * xs_udp_set_buffer_size - set send and receive limits
960 * @xprt: generic transport
961 * @sndsize: requested size of send buffer, in bytes
962 * @rcvsize: requested size of receive buffer, in bytes
964 * Set socket send and receive buffer size limits.
966 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
970 xprt->sndsize = sndsize + 1024;
973 xprt->rcvsize = rcvsize + 1024;
975 xs_udp_do_set_buffer_size(xprt);
979 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
980 * @task: task that timed out
982 * Adjust the congestion window after a retransmit timeout has occurred.
984 static void xs_udp_timer(struct rpc_task *task)
986 xprt_adjust_cwnd(task, -ETIMEDOUT);
989 static unsigned short xs_get_random_port(void)
991 unsigned short range = xprt_max_resvport - xprt_min_resvport;
992 unsigned short rand = (unsigned short) net_random() % range;
993 return rand + xprt_min_resvport;
997 * xs_print_peer_address - format an IPv4 address for printing
998 * @xprt: generic transport
999 * @format: flags field indicating which parts of the address to render
1001 static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format)
1003 if (xprt->address_strings[format] != NULL)
1004 return xprt->address_strings[format];
1006 return "unprintable";
1010 * xs_set_port - reset the port number in the remote endpoint address
1011 * @xprt: generic transport
1012 * @port: new port number
1015 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1017 struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1019 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1021 sap->sin_port = htons(port);
1024 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
1026 struct sockaddr_in myaddr = {
1027 .sin_family = AF_INET,
1030 unsigned short port = xprt->port;
1033 myaddr.sin_port = htons(port);
1034 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1038 dprintk("RPC: xs_bindresvport bound to port %u\n",
1042 if (port <= xprt_min_resvport)
1043 port = xprt_max_resvport;
1046 } while (err == -EADDRINUSE && port != xprt->port);
1048 dprintk("RPC: can't bind to reserved port (%d).\n", -err);
1053 * xs_udp_connect_worker - set up a UDP socket
1054 * @args: RPC transport to connect
1056 * Invoked by a work queue tasklet.
1058 static void xs_udp_connect_worker(void *args)
1060 struct rpc_xprt *xprt = (struct rpc_xprt *) args;
1061 struct socket *sock = xprt->sock;
1062 int err, status = -EIO;
1064 if (xprt->shutdown || !xprt_bound(xprt))
1067 /* Start by resetting any existing state */
1070 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1071 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1075 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1080 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1081 xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1084 struct sock *sk = sock->sk;
1086 write_lock_bh(&sk->sk_callback_lock);
1088 sk->sk_user_data = xprt;
1089 xprt->old_data_ready = sk->sk_data_ready;
1090 xprt->old_state_change = sk->sk_state_change;
1091 xprt->old_write_space = sk->sk_write_space;
1092 sk->sk_data_ready = xs_udp_data_ready;
1093 sk->sk_write_space = xs_udp_write_space;
1094 sk->sk_no_check = UDP_CSUM_NORCV;
1095 sk->sk_allocation = GFP_ATOMIC;
1097 xprt_set_connected(xprt);
1099 /* Reset to new socket */
1103 write_unlock_bh(&sk->sk_callback_lock);
1105 xs_udp_do_set_buffer_size(xprt);
1108 xprt_wake_pending_tasks(xprt, status);
1109 xprt_clear_connecting(xprt);
1113 * We need to preserve the port number so the reply cache on the server can
1114 * find our cached RPC replies when we get around to reconnecting.
1116 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1119 struct socket *sock = xprt->sock;
1120 struct sockaddr any;
1122 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1125 * Disconnect the transport socket by doing a connect operation
1126 * with AF_UNSPEC. This should return immediately...
1128 memset(&any, 0, sizeof(any));
1129 any.sa_family = AF_UNSPEC;
1130 result = kernel_connect(sock, &any, sizeof(any), 0);
1132 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1137 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1138 * @args: RPC transport to connect
1140 * Invoked by a work queue tasklet.
1142 static void xs_tcp_connect_worker(void *args)
1144 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1145 struct socket *sock = xprt->sock;
1146 int err, status = -EIO;
1148 if (xprt->shutdown || !xprt_bound(xprt))
1152 /* start from scratch */
1153 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1154 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1158 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1163 /* "close" the socket, preserving the local port */
1164 xs_tcp_reuse_connection(xprt);
1166 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1167 xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1170 struct sock *sk = sock->sk;
1172 write_lock_bh(&sk->sk_callback_lock);
1174 sk->sk_user_data = xprt;
1175 xprt->old_data_ready = sk->sk_data_ready;
1176 xprt->old_state_change = sk->sk_state_change;
1177 xprt->old_write_space = sk->sk_write_space;
1178 sk->sk_data_ready = xs_tcp_data_ready;
1179 sk->sk_state_change = xs_tcp_state_change;
1180 sk->sk_write_space = xs_tcp_write_space;
1181 sk->sk_allocation = GFP_ATOMIC;
1183 /* socket options */
1184 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1185 sock_reset_flag(sk, SOCK_LINGER);
1186 tcp_sk(sk)->linger2 = 0;
1187 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1189 xprt_clear_connected(xprt);
1191 /* Reset to new socket */
1195 write_unlock_bh(&sk->sk_callback_lock);
1198 /* Tell the socket layer to start connecting... */
1199 xprt->stat.connect_count++;
1200 xprt->stat.connect_start = jiffies;
1201 status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1202 xprt->addrlen, O_NONBLOCK);
1203 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1204 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1212 /* retry with existing socket, after a delay */
1215 /* get rid of existing socket, and retry */
1221 xprt_wake_pending_tasks(xprt, status);
1223 xprt_clear_connecting(xprt);
1227 * xs_connect - connect a socket to a remote endpoint
1228 * @task: address of RPC task that manages state of connect request
1230 * TCP: If the remote end dropped the connection, delay reconnecting.
1232 * UDP socket connects are synchronous, but we use a work queue anyway
1233 * to guarantee that even unprivileged user processes can set up a
1234 * socket on a privileged port.
1236 * If a UDP socket connect fails, the delay behavior here prevents
1237 * retry floods (hard mounts).
1239 static void xs_connect(struct rpc_task *task)
1241 struct rpc_xprt *xprt = task->tk_xprt;
1243 if (xprt_test_and_set_connecting(xprt))
1246 if (xprt->sock != NULL) {
1247 dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n",
1248 xprt, xprt->reestablish_timeout / HZ);
1249 schedule_delayed_work(&xprt->connect_worker,
1250 xprt->reestablish_timeout);
1251 xprt->reestablish_timeout <<= 1;
1252 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1253 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1255 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1256 schedule_work(&xprt->connect_worker);
1258 /* flush_scheduled_work can sleep... */
1259 if (!RPC_IS_ASYNC(task))
1260 flush_scheduled_work();
1265 * xs_udp_print_stats - display UDP socket-specifc stats
1266 * @xprt: rpc_xprt struct containing statistics
1270 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1272 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1274 xprt->stat.bind_count,
1277 xprt->stat.bad_xids,
1279 xprt->stat.bklog_u);
1283 * xs_tcp_print_stats - display TCP socket-specifc stats
1284 * @xprt: rpc_xprt struct containing statistics
1288 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1292 if (xprt_connected(xprt))
1293 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1295 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1297 xprt->stat.bind_count,
1298 xprt->stat.connect_count,
1299 xprt->stat.connect_time,
1303 xprt->stat.bad_xids,
1305 xprt->stat.bklog_u);
1308 static struct rpc_xprt_ops xs_udp_ops = {
1309 .set_buffer_size = xs_udp_set_buffer_size,
1310 .print_addr = xs_print_peer_address,
1311 .reserve_xprt = xprt_reserve_xprt_cong,
1312 .release_xprt = xprt_release_xprt_cong,
1313 .rpcbind = rpc_getport,
1314 .set_port = xs_set_port,
1315 .connect = xs_connect,
1316 .buf_alloc = rpc_malloc,
1317 .buf_free = rpc_free,
1318 .send_request = xs_udp_send_request,
1319 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1320 .timer = xs_udp_timer,
1321 .release_request = xprt_release_rqst_cong,
1323 .destroy = xs_destroy,
1324 .print_stats = xs_udp_print_stats,
1327 static struct rpc_xprt_ops xs_tcp_ops = {
1328 .print_addr = xs_print_peer_address,
1329 .reserve_xprt = xprt_reserve_xprt,
1330 .release_xprt = xs_tcp_release_xprt,
1331 .rpcbind = rpc_getport,
1332 .set_port = xs_set_port,
1333 .connect = xs_connect,
1334 .buf_alloc = rpc_malloc,
1335 .buf_free = rpc_free,
1336 .send_request = xs_tcp_send_request,
1337 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1339 .destroy = xs_destroy,
1340 .print_stats = xs_tcp_print_stats,
1343 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1345 struct rpc_xprt *xprt;
1347 if (addrlen > sizeof(xprt->addr)) {
1348 dprintk("RPC: xs_setup_xprt: address too large\n");
1349 return ERR_PTR(-EBADF);
1352 xprt = kzalloc(sizeof(struct rpc_xprt), GFP_KERNEL);
1354 dprintk("RPC: xs_setup_xprt: couldn't allocate rpc_xprt\n");
1355 return ERR_PTR(-ENOMEM);
1358 xprt->max_reqs = slot_table_size;
1359 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1360 if (xprt->slot == NULL) {
1362 dprintk("RPC: xs_setup_xprt: couldn't allocate slot table\n");
1363 return ERR_PTR(-ENOMEM);
1366 memcpy(&xprt->addr, addr, addrlen);
1367 xprt->addrlen = addrlen;
1368 xprt->port = xs_get_random_port();
1374 * xs_setup_udp - Set up transport to use a UDP socket
1375 * @addr: address of remote server
1376 * @addrlen: length of address in bytes
1377 * @to: timeout parameters
1380 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1382 struct rpc_xprt *xprt;
1384 xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1388 if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1389 xprt_set_bound(xprt);
1391 xprt->prot = IPPROTO_UDP;
1393 /* XXX: header size can vary due to auth type, IPv6, etc. */
1394 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1396 INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1397 xprt->bind_timeout = XS_BIND_TO;
1398 xprt->connect_timeout = XS_UDP_CONN_TO;
1399 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1400 xprt->idle_timeout = XS_IDLE_DISC_TO;
1402 xprt->ops = &xs_udp_ops;
1405 xprt->timeout = *to;
1407 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1409 xs_format_peer_addresses(xprt);
1410 dprintk("RPC: set up transport to address %s\n",
1411 xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1417 * xs_setup_tcp - Set up transport to use a TCP socket
1418 * @addr: address of remote server
1419 * @addrlen: length of address in bytes
1420 * @to: timeout parameters
1423 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1425 struct rpc_xprt *xprt;
1427 xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1431 if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1432 xprt_set_bound(xprt);
1434 xprt->prot = IPPROTO_TCP;
1435 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1436 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1438 INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1439 xprt->bind_timeout = XS_BIND_TO;
1440 xprt->connect_timeout = XS_TCP_CONN_TO;
1441 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1442 xprt->idle_timeout = XS_IDLE_DISC_TO;
1444 xprt->ops = &xs_tcp_ops;
1447 xprt->timeout = *to;
1449 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1451 xs_format_peer_addresses(xprt);
1452 dprintk("RPC: set up transport to address %s\n",
1453 xs_print_peer_address(xprt, RPC_DISPLAY_ALL));