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 inline int xs_send_head(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
, unsigned int len
)
174 .iov_base
= xdr
->head
[0].iov_base
+ base
,
175 .iov_len
= len
- base
,
177 struct msghdr msg
= {
179 .msg_namelen
= addrlen
,
180 .msg_flags
= XS_SENDMSG_FLAGS
,
184 msg
.msg_flags
|= MSG_MORE
;
186 if (likely(iov
.iov_len
))
187 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
188 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
191 static int xs_send_tail(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, unsigned int len
)
194 .iov_base
= xdr
->tail
[0].iov_base
+ base
,
195 .iov_len
= len
- base
,
197 struct msghdr msg
= {
198 .msg_flags
= XS_SENDMSG_FLAGS
,
201 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
205 * xs_sendpages - write pages directly to a socket
206 * @sock: socket to send on
207 * @addr: UDP only -- address of destination
208 * @addrlen: UDP only -- length of destination address
209 * @xdr: buffer containing this request
210 * @base: starting position in the buffer
213 static inline int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
)
215 struct page
**ppage
= xdr
->pages
;
216 unsigned int len
, pglen
= xdr
->page_len
;
222 clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
224 len
= xdr
->head
[0].iov_len
;
225 if (base
< len
|| (addr
!= NULL
&& base
== 0)) {
226 err
= xs_send_head(sock
, addr
, addrlen
, xdr
, base
, len
);
231 if (err
!= (len
- base
))
237 if (unlikely(pglen
== 0))
239 if (unlikely(base
>= pglen
)) {
243 if (base
|| xdr
->page_base
) {
245 base
+= xdr
->page_base
;
246 ppage
+= base
>> PAGE_CACHE_SHIFT
;
247 base
&= ~PAGE_CACHE_MASK
;
251 int flags
= XS_SENDMSG_FLAGS
;
253 len
= PAGE_CACHE_SIZE
;
259 if (pglen
!= len
|| xdr
->tail
[0].iov_len
!= 0)
262 err
= kernel_sendpage(sock
, *ppage
, base
, len
, flags
);
271 } while ((pglen
-= len
) != 0);
273 len
= xdr
->tail
[0].iov_len
;
275 err
= xs_send_tail(sock
, xdr
, base
, len
);
286 * xs_nospace - place task on wait queue if transmit was incomplete
287 * @task: task to put to sleep
290 static void xs_nospace(struct rpc_task
*task
)
292 struct rpc_rqst
*req
= task
->tk_rqstp
;
293 struct rpc_xprt
*xprt
= req
->rq_xprt
;
295 dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
296 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
299 if (test_bit(SOCK_ASYNC_NOSPACE
, &xprt
->sock
->flags
)) {
300 /* Protect against races with write_space */
301 spin_lock_bh(&xprt
->transport_lock
);
303 /* Don't race with disconnect */
304 if (!xprt_connected(xprt
))
305 task
->tk_status
= -ENOTCONN
;
306 else if (test_bit(SOCK_NOSPACE
, &xprt
->sock
->flags
))
307 xprt_wait_for_buffer_space(task
);
309 spin_unlock_bh(&xprt
->transport_lock
);
311 /* Keep holding the socket if it is blocked */
312 rpc_delay(task
, HZ
>>4);
316 * xs_udp_send_request - write an RPC request to a UDP socket
317 * @task: address of RPC task that manages the state of an RPC request
320 * 0: The request has been sent
321 * EAGAIN: The socket was blocked, please call again later to
322 * complete the request
323 * ENOTCONN: Caller needs to invoke connect logic then call again
324 * other: Some other error occured, the request was not sent
326 static int xs_udp_send_request(struct rpc_task
*task
)
328 struct rpc_rqst
*req
= task
->tk_rqstp
;
329 struct rpc_xprt
*xprt
= req
->rq_xprt
;
330 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
333 xs_pktdump("packet data:",
334 req
->rq_svec
->iov_base
,
335 req
->rq_svec
->iov_len
);
337 req
->rq_xtime
= jiffies
;
338 status
= xs_sendpages(xprt
->sock
, (struct sockaddr
*) &xprt
->addr
,
339 xprt
->addrlen
, xdr
, req
->rq_bytes_sent
);
341 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
342 xdr
->len
- req
->rq_bytes_sent
, status
);
344 if (likely(status
>= (int) req
->rq_slen
))
347 /* Still some bytes left; set up for a retry later. */
355 /* When the server has died, an ICMP port unreachable message
356 * prompts ECONNREFUSED. */
362 dprintk("RPC: sendmsg returned unrecognized error %d\n",
370 static inline void xs_encode_tcp_record_marker(struct xdr_buf
*buf
)
372 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
373 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
374 *base
= htonl(RPC_LAST_STREAM_FRAGMENT
| reclen
);
378 * xs_tcp_send_request - write an RPC request to a TCP socket
379 * @task: address of RPC task that manages the state of an RPC request
382 * 0: The request has been sent
383 * EAGAIN: The socket was blocked, please call again later to
384 * complete the request
385 * ENOTCONN: Caller needs to invoke connect logic then call again
386 * other: Some other error occured, the request was not sent
388 * XXX: In the case of soft timeouts, should we eventually give up
389 * if sendmsg is not able to make progress?
391 static int xs_tcp_send_request(struct rpc_task
*task
)
393 struct rpc_rqst
*req
= task
->tk_rqstp
;
394 struct rpc_xprt
*xprt
= req
->rq_xprt
;
395 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
396 int status
, retry
= 0;
398 xs_encode_tcp_record_marker(&req
->rq_snd_buf
);
400 xs_pktdump("packet data:",
401 req
->rq_svec
->iov_base
,
402 req
->rq_svec
->iov_len
);
404 /* Continue transmitting the packet/record. We must be careful
405 * to cope with writespace callbacks arriving _after_ we have
406 * called sendmsg(). */
408 req
->rq_xtime
= jiffies
;
409 status
= xs_sendpages(xprt
->sock
, NULL
, 0, xdr
,
412 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
413 xdr
->len
- req
->rq_bytes_sent
, status
);
415 if (unlikely(status
< 0))
418 /* If we've sent the entire packet, immediately
419 * reset the count of bytes sent. */
420 req
->rq_bytes_sent
+= status
;
421 task
->tk_bytes_sent
+= status
;
422 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
423 req
->rq_bytes_sent
= 0;
428 if (retry
++ > XS_SENDMSG_RETRY
)
443 dprintk("RPC: sendmsg returned unrecognized error %d\n",
445 xprt_disconnect(xprt
);
453 * xs_tcp_release_xprt - clean up after a tcp transmission
457 * This cleans up if an error causes us to abort the transmission of a request.
458 * In this case, the socket may need to be reset in order to avoid confusing
461 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
463 struct rpc_rqst
*req
;
465 if (task
!= xprt
->snd_task
)
469 req
= task
->tk_rqstp
;
470 if (req
->rq_bytes_sent
== 0)
472 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
474 set_bit(XPRT_CLOSE_WAIT
, &task
->tk_xprt
->state
);
476 xprt_release_xprt(xprt
, task
);
480 * xs_close - close a socket
483 * This is used when all requests are complete; ie, no DRC state remains
484 * on the server we want to save.
486 static void xs_close(struct rpc_xprt
*xprt
)
488 struct socket
*sock
= xprt
->sock
;
489 struct sock
*sk
= xprt
->inet
;
492 goto clear_close_wait
;
494 dprintk("RPC: xs_close xprt %p\n", xprt
);
496 write_lock_bh(&sk
->sk_callback_lock
);
500 sk
->sk_user_data
= NULL
;
501 sk
->sk_data_ready
= xprt
->old_data_ready
;
502 sk
->sk_state_change
= xprt
->old_state_change
;
503 sk
->sk_write_space
= xprt
->old_write_space
;
504 write_unlock_bh(&sk
->sk_callback_lock
);
510 smp_mb__before_clear_bit();
511 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
512 smp_mb__after_clear_bit();
516 * xs_destroy - prepare to shutdown a transport
517 * @xprt: doomed transport
520 static void xs_destroy(struct rpc_xprt
*xprt
)
522 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
524 cancel_delayed_work(&xprt
->connect_worker
);
525 flush_scheduled_work();
527 xprt_disconnect(xprt
);
529 xs_free_peer_addresses(xprt
);
533 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
535 return (struct rpc_xprt
*) sk
->sk_user_data
;
539 * xs_udp_data_ready - "data ready" callback for UDP sockets
540 * @sk: socket with data to read
541 * @len: how much data to read
544 static void xs_udp_data_ready(struct sock
*sk
, int len
)
546 struct rpc_task
*task
;
547 struct rpc_xprt
*xprt
;
548 struct rpc_rqst
*rovr
;
550 int err
, repsize
, copied
;
554 read_lock(&sk
->sk_callback_lock
);
555 dprintk("RPC: xs_udp_data_ready...\n");
556 if (!(xprt
= xprt_from_sock(sk
)))
559 if ((skb
= skb_recv_datagram(sk
, 0, 1, &err
)) == NULL
)
565 repsize
= skb
->len
- sizeof(struct udphdr
);
567 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
571 /* Copy the XID from the skb... */
572 xp
= skb_header_pointer(skb
, sizeof(struct udphdr
),
573 sizeof(_xid
), &_xid
);
577 /* Look up and lock the request corresponding to the given XID */
578 spin_lock(&xprt
->transport_lock
);
579 rovr
= xprt_lookup_rqst(xprt
, *xp
);
582 task
= rovr
->rq_task
;
584 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
587 /* Suck it into the iovec, verify checksum if not done by hw. */
588 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
))
591 /* Something worked... */
592 dst_confirm(skb
->dst
);
594 xprt_adjust_cwnd(task
, copied
);
595 xprt_update_rtt(task
);
596 xprt_complete_rqst(task
, copied
);
599 spin_unlock(&xprt
->transport_lock
);
601 skb_free_datagram(sk
, skb
);
603 read_unlock(&sk
->sk_callback_lock
);
606 static inline size_t xs_tcp_copy_data(skb_reader_t
*desc
, void *p
, size_t len
)
608 if (len
> desc
->count
)
610 if (skb_copy_bits(desc
->skb
, desc
->offset
, p
, len
)) {
611 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
617 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
622 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, skb_reader_t
*desc
)
627 p
= ((char *) &xprt
->tcp_recm
) + xprt
->tcp_offset
;
628 len
= sizeof(xprt
->tcp_recm
) - xprt
->tcp_offset
;
629 used
= xs_tcp_copy_data(desc
, p
, len
);
630 xprt
->tcp_offset
+= used
;
634 xprt
->tcp_reclen
= ntohl(xprt
->tcp_recm
);
635 if (xprt
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
636 xprt
->tcp_flags
|= XPRT_LAST_FRAG
;
638 xprt
->tcp_flags
&= ~XPRT_LAST_FRAG
;
639 xprt
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
641 xprt
->tcp_flags
&= ~XPRT_COPY_RECM
;
642 xprt
->tcp_offset
= 0;
644 /* Sanity check of the record length */
645 if (unlikely(xprt
->tcp_reclen
< 4)) {
646 dprintk("RPC: invalid TCP record fragment length\n");
647 xprt_disconnect(xprt
);
650 dprintk("RPC: reading TCP record fragment of length %d\n",
654 static void xs_tcp_check_recm(struct rpc_xprt
*xprt
)
656 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
657 xprt
, xprt
->tcp_copied
, xprt
->tcp_offset
, xprt
->tcp_reclen
, xprt
->tcp_flags
);
658 if (xprt
->tcp_offset
== xprt
->tcp_reclen
) {
659 xprt
->tcp_flags
|= XPRT_COPY_RECM
;
660 xprt
->tcp_offset
= 0;
661 if (xprt
->tcp_flags
& XPRT_LAST_FRAG
) {
662 xprt
->tcp_flags
&= ~XPRT_COPY_DATA
;
663 xprt
->tcp_flags
|= XPRT_COPY_XID
;
664 xprt
->tcp_copied
= 0;
669 static inline void xs_tcp_read_xid(struct rpc_xprt
*xprt
, skb_reader_t
*desc
)
674 len
= sizeof(xprt
->tcp_xid
) - xprt
->tcp_offset
;
675 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
676 p
= ((char *) &xprt
->tcp_xid
) + xprt
->tcp_offset
;
677 used
= xs_tcp_copy_data(desc
, p
, len
);
678 xprt
->tcp_offset
+= used
;
681 xprt
->tcp_flags
&= ~XPRT_COPY_XID
;
682 xprt
->tcp_flags
|= XPRT_COPY_DATA
;
683 xprt
->tcp_copied
= 4;
684 dprintk("RPC: reading reply for XID %08x\n",
685 ntohl(xprt
->tcp_xid
));
686 xs_tcp_check_recm(xprt
);
689 static inline void xs_tcp_read_request(struct rpc_xprt
*xprt
, skb_reader_t
*desc
)
691 struct rpc_rqst
*req
;
692 struct xdr_buf
*rcvbuf
;
696 /* Find and lock the request corresponding to this xid */
697 spin_lock(&xprt
->transport_lock
);
698 req
= xprt_lookup_rqst(xprt
, xprt
->tcp_xid
);
700 xprt
->tcp_flags
&= ~XPRT_COPY_DATA
;
701 dprintk("RPC: XID %08x request not found!\n",
702 ntohl(xprt
->tcp_xid
));
703 spin_unlock(&xprt
->transport_lock
);
707 rcvbuf
= &req
->rq_private_buf
;
709 if (len
> xprt
->tcp_reclen
- xprt
->tcp_offset
) {
710 skb_reader_t my_desc
;
712 len
= xprt
->tcp_reclen
- xprt
->tcp_offset
;
713 memcpy(&my_desc
, desc
, sizeof(my_desc
));
715 r
= xdr_partial_copy_from_skb(rcvbuf
, xprt
->tcp_copied
,
716 &my_desc
, xs_tcp_copy_data
);
720 r
= xdr_partial_copy_from_skb(rcvbuf
, xprt
->tcp_copied
,
721 desc
, xs_tcp_copy_data
);
724 xprt
->tcp_copied
+= r
;
725 xprt
->tcp_offset
+= r
;
728 /* Error when copying to the receive buffer,
729 * usually because we weren't able to allocate
730 * additional buffer pages. All we can do now
731 * is turn off XPRT_COPY_DATA, so the request
732 * will not receive any additional updates,
734 * Any remaining data from this record will
737 xprt
->tcp_flags
&= ~XPRT_COPY_DATA
;
738 dprintk("RPC: XID %08x truncated request\n",
739 ntohl(xprt
->tcp_xid
));
740 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
741 xprt
, xprt
->tcp_copied
, xprt
->tcp_offset
, xprt
->tcp_reclen
);
745 dprintk("RPC: XID %08x read %Zd bytes\n",
746 ntohl(xprt
->tcp_xid
), r
);
747 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
748 xprt
, xprt
->tcp_copied
, xprt
->tcp_offset
, xprt
->tcp_reclen
);
750 if (xprt
->tcp_copied
== req
->rq_private_buf
.buflen
)
751 xprt
->tcp_flags
&= ~XPRT_COPY_DATA
;
752 else if (xprt
->tcp_offset
== xprt
->tcp_reclen
) {
753 if (xprt
->tcp_flags
& XPRT_LAST_FRAG
)
754 xprt
->tcp_flags
&= ~XPRT_COPY_DATA
;
758 if (!(xprt
->tcp_flags
& XPRT_COPY_DATA
))
759 xprt_complete_rqst(req
->rq_task
, xprt
->tcp_copied
);
760 spin_unlock(&xprt
->transport_lock
);
761 xs_tcp_check_recm(xprt
);
764 static inline void xs_tcp_read_discard(struct rpc_xprt
*xprt
, skb_reader_t
*desc
)
768 len
= xprt
->tcp_reclen
- xprt
->tcp_offset
;
769 if (len
> desc
->count
)
773 xprt
->tcp_offset
+= len
;
774 dprintk("RPC: discarded %Zu bytes\n", len
);
775 xs_tcp_check_recm(xprt
);
778 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
780 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
781 skb_reader_t desc
= {
788 dprintk("RPC: xs_tcp_data_recv started\n");
790 /* Read in a new fragment marker if necessary */
791 /* Can we ever really expect to get completely empty fragments? */
792 if (xprt
->tcp_flags
& XPRT_COPY_RECM
) {
793 xs_tcp_read_fraghdr(xprt
, &desc
);
796 /* Read in the xid if necessary */
797 if (xprt
->tcp_flags
& XPRT_COPY_XID
) {
798 xs_tcp_read_xid(xprt
, &desc
);
801 /* Read in the request data */
802 if (xprt
->tcp_flags
& XPRT_COPY_DATA
) {
803 xs_tcp_read_request(xprt
, &desc
);
806 /* Skip over any trailing bytes on short reads */
807 xs_tcp_read_discard(xprt
, &desc
);
808 } while (desc
.count
);
809 dprintk("RPC: xs_tcp_data_recv done\n");
810 return len
- desc
.count
;
814 * xs_tcp_data_ready - "data ready" callback for TCP sockets
815 * @sk: socket with data to read
816 * @bytes: how much data to read
819 static void xs_tcp_data_ready(struct sock
*sk
, int bytes
)
821 struct rpc_xprt
*xprt
;
822 read_descriptor_t rd_desc
;
824 read_lock(&sk
->sk_callback_lock
);
825 dprintk("RPC: xs_tcp_data_ready...\n");
826 if (!(xprt
= xprt_from_sock(sk
)))
831 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
832 rd_desc
.arg
.data
= xprt
;
833 rd_desc
.count
= 65536;
834 tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
836 read_unlock(&sk
->sk_callback_lock
);
840 * xs_tcp_state_change - callback to handle TCP socket state changes
841 * @sk: socket whose state has changed
844 static void xs_tcp_state_change(struct sock
*sk
)
846 struct rpc_xprt
*xprt
;
848 read_lock(&sk
->sk_callback_lock
);
849 if (!(xprt
= xprt_from_sock(sk
)))
851 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
852 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
853 sk
->sk_state
, xprt_connected(xprt
),
854 sock_flag(sk
, SOCK_DEAD
),
855 sock_flag(sk
, SOCK_ZAPPED
));
857 switch (sk
->sk_state
) {
858 case TCP_ESTABLISHED
:
859 spin_lock_bh(&xprt
->transport_lock
);
860 if (!xprt_test_and_set_connected(xprt
)) {
861 /* Reset TCP record info */
862 xprt
->tcp_offset
= 0;
863 xprt
->tcp_reclen
= 0;
864 xprt
->tcp_copied
= 0;
865 xprt
->tcp_flags
= XPRT_COPY_RECM
| XPRT_COPY_XID
;
866 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
867 xprt_wake_pending_tasks(xprt
, 0);
869 spin_unlock_bh(&xprt
->transport_lock
);
875 /* Try to schedule an autoclose RPC calls */
876 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
877 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
878 schedule_work(&xprt
->task_cleanup
);
880 xprt_disconnect(xprt
);
883 read_unlock(&sk
->sk_callback_lock
);
887 * xs_udp_write_space - callback invoked when socket buffer space
889 * @sk: socket whose state has changed
891 * Called when more output buffer space is available for this socket.
892 * We try not to wake our writers until they can make "significant"
893 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
894 * with a bunch of small requests.
896 static void xs_udp_write_space(struct sock
*sk
)
898 read_lock(&sk
->sk_callback_lock
);
900 /* from net/core/sock.c:sock_def_write_space */
901 if (sock_writeable(sk
)) {
903 struct rpc_xprt
*xprt
;
905 if (unlikely(!(sock
= sk
->sk_socket
)))
907 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
909 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE
, &sock
->flags
)))
912 xprt_write_space(xprt
);
916 read_unlock(&sk
->sk_callback_lock
);
920 * xs_tcp_write_space - callback invoked when socket buffer space
922 * @sk: socket whose state has changed
924 * Called when more output buffer space is available for this socket.
925 * We try not to wake our writers until they can make "significant"
926 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
927 * with a bunch of small requests.
929 static void xs_tcp_write_space(struct sock
*sk
)
931 read_lock(&sk
->sk_callback_lock
);
933 /* from net/core/stream.c:sk_stream_write_space */
934 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
936 struct rpc_xprt
*xprt
;
938 if (unlikely(!(sock
= sk
->sk_socket
)))
940 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
942 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE
, &sock
->flags
)))
945 xprt_write_space(xprt
);
949 read_unlock(&sk
->sk_callback_lock
);
952 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
954 struct sock
*sk
= xprt
->inet
;
957 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
958 sk
->sk_rcvbuf
= xprt
->rcvsize
* xprt
->max_reqs
* 2;
961 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
962 sk
->sk_sndbuf
= xprt
->sndsize
* xprt
->max_reqs
* 2;
963 sk
->sk_write_space(sk
);
968 * xs_udp_set_buffer_size - set send and receive limits
969 * @xprt: generic transport
970 * @sndsize: requested size of send buffer, in bytes
971 * @rcvsize: requested size of receive buffer, in bytes
973 * Set socket send and receive buffer size limits.
975 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
979 xprt
->sndsize
= sndsize
+ 1024;
982 xprt
->rcvsize
= rcvsize
+ 1024;
984 xs_udp_do_set_buffer_size(xprt
);
988 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
989 * @task: task that timed out
991 * Adjust the congestion window after a retransmit timeout has occurred.
993 static void xs_udp_timer(struct rpc_task
*task
)
995 xprt_adjust_cwnd(task
, -ETIMEDOUT
);
998 static unsigned short xs_get_random_port(void)
1000 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
;
1001 unsigned short rand
= (unsigned short) net_random() % range
;
1002 return rand
+ xprt_min_resvport
;
1006 * xs_print_peer_address - format an IPv4 address for printing
1007 * @xprt: generic transport
1008 * @format: flags field indicating which parts of the address to render
1010 static char *xs_print_peer_address(struct rpc_xprt
*xprt
, enum rpc_display_format_t format
)
1012 if (xprt
->address_strings
[format
] != NULL
)
1013 return xprt
->address_strings
[format
];
1015 return "unprintable";
1019 * xs_set_port - reset the port number in the remote endpoint address
1020 * @xprt: generic transport
1021 * @port: new port number
1024 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1026 struct sockaddr_in
*sap
= (struct sockaddr_in
*) &xprt
->addr
;
1028 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1030 sap
->sin_port
= htons(port
);
1033 static int xs_bindresvport(struct rpc_xprt
*xprt
, struct socket
*sock
)
1035 struct sockaddr_in myaddr
= {
1036 .sin_family
= AF_INET
,
1039 unsigned short port
= xprt
->port
;
1042 myaddr
.sin_port
= htons(port
);
1043 err
= kernel_bind(sock
, (struct sockaddr
*) &myaddr
,
1047 dprintk("RPC: xs_bindresvport bound to port %u\n",
1051 if (port
<= xprt_min_resvport
)
1052 port
= xprt_max_resvport
;
1055 } while (err
== -EADDRINUSE
&& port
!= xprt
->port
);
1057 dprintk("RPC: can't bind to reserved port (%d).\n", -err
);
1062 * xs_udp_connect_worker - set up a UDP socket
1063 * @args: RPC transport to connect
1065 * Invoked by a work queue tasklet.
1067 static void xs_udp_connect_worker(void *args
)
1069 struct rpc_xprt
*xprt
= (struct rpc_xprt
*) args
;
1070 struct socket
*sock
= xprt
->sock
;
1071 int err
, status
= -EIO
;
1073 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1076 /* Start by resetting any existing state */
1079 if ((err
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
)) < 0) {
1080 dprintk("RPC: can't create UDP transport socket (%d).\n", -err
);
1084 if (xprt
->resvport
&& xs_bindresvport(xprt
, sock
) < 0) {
1089 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1090 xprt
, xs_print_peer_address(xprt
, RPC_DISPLAY_ALL
));
1093 struct sock
*sk
= sock
->sk
;
1095 write_lock_bh(&sk
->sk_callback_lock
);
1097 sk
->sk_user_data
= xprt
;
1098 xprt
->old_data_ready
= sk
->sk_data_ready
;
1099 xprt
->old_state_change
= sk
->sk_state_change
;
1100 xprt
->old_write_space
= sk
->sk_write_space
;
1101 sk
->sk_data_ready
= xs_udp_data_ready
;
1102 sk
->sk_write_space
= xs_udp_write_space
;
1103 sk
->sk_no_check
= UDP_CSUM_NORCV
;
1104 sk
->sk_allocation
= GFP_ATOMIC
;
1106 xprt_set_connected(xprt
);
1108 /* Reset to new socket */
1112 write_unlock_bh(&sk
->sk_callback_lock
);
1114 xs_udp_do_set_buffer_size(xprt
);
1117 xprt_wake_pending_tasks(xprt
, status
);
1118 xprt_clear_connecting(xprt
);
1122 * We need to preserve the port number so the reply cache on the server can
1123 * find our cached RPC replies when we get around to reconnecting.
1125 static void xs_tcp_reuse_connection(struct rpc_xprt
*xprt
)
1128 struct socket
*sock
= xprt
->sock
;
1129 struct sockaddr any
;
1131 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt
);
1134 * Disconnect the transport socket by doing a connect operation
1135 * with AF_UNSPEC. This should return immediately...
1137 memset(&any
, 0, sizeof(any
));
1138 any
.sa_family
= AF_UNSPEC
;
1139 result
= kernel_connect(sock
, &any
, sizeof(any
), 0);
1141 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1146 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1147 * @args: RPC transport to connect
1149 * Invoked by a work queue tasklet.
1151 static void xs_tcp_connect_worker(void *args
)
1153 struct rpc_xprt
*xprt
= (struct rpc_xprt
*)args
;
1154 struct socket
*sock
= xprt
->sock
;
1155 int err
, status
= -EIO
;
1157 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1161 /* start from scratch */
1162 if ((err
= sock_create_kern(PF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &sock
)) < 0) {
1163 dprintk("RPC: can't create TCP transport socket (%d).\n", -err
);
1167 if (xprt
->resvport
&& xs_bindresvport(xprt
, sock
) < 0) {
1172 /* "close" the socket, preserving the local port */
1173 xs_tcp_reuse_connection(xprt
);
1175 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1176 xprt
, xs_print_peer_address(xprt
, RPC_DISPLAY_ALL
));
1179 struct sock
*sk
= sock
->sk
;
1181 write_lock_bh(&sk
->sk_callback_lock
);
1183 sk
->sk_user_data
= xprt
;
1184 xprt
->old_data_ready
= sk
->sk_data_ready
;
1185 xprt
->old_state_change
= sk
->sk_state_change
;
1186 xprt
->old_write_space
= sk
->sk_write_space
;
1187 sk
->sk_data_ready
= xs_tcp_data_ready
;
1188 sk
->sk_state_change
= xs_tcp_state_change
;
1189 sk
->sk_write_space
= xs_tcp_write_space
;
1190 sk
->sk_allocation
= GFP_ATOMIC
;
1192 /* socket options */
1193 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
1194 sock_reset_flag(sk
, SOCK_LINGER
);
1195 tcp_sk(sk
)->linger2
= 0;
1196 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1198 xprt_clear_connected(xprt
);
1200 /* Reset to new socket */
1204 write_unlock_bh(&sk
->sk_callback_lock
);
1207 /* Tell the socket layer to start connecting... */
1208 xprt
->stat
.connect_count
++;
1209 xprt
->stat
.connect_start
= jiffies
;
1210 status
= kernel_connect(sock
, (struct sockaddr
*) &xprt
->addr
,
1211 xprt
->addrlen
, O_NONBLOCK
);
1212 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1213 xprt
, -status
, xprt_connected(xprt
), sock
->sk
->sk_state
);
1221 /* retry with existing socket, after a delay */
1224 /* get rid of existing socket, and retry */
1230 xprt_wake_pending_tasks(xprt
, status
);
1232 xprt_clear_connecting(xprt
);
1236 * xs_connect - connect a socket to a remote endpoint
1237 * @task: address of RPC task that manages state of connect request
1239 * TCP: If the remote end dropped the connection, delay reconnecting.
1241 * UDP socket connects are synchronous, but we use a work queue anyway
1242 * to guarantee that even unprivileged user processes can set up a
1243 * socket on a privileged port.
1245 * If a UDP socket connect fails, the delay behavior here prevents
1246 * retry floods (hard mounts).
1248 static void xs_connect(struct rpc_task
*task
)
1250 struct rpc_xprt
*xprt
= task
->tk_xprt
;
1252 if (xprt_test_and_set_connecting(xprt
))
1255 if (xprt
->sock
!= NULL
) {
1256 dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n",
1257 xprt
, xprt
->reestablish_timeout
/ HZ
);
1258 schedule_delayed_work(&xprt
->connect_worker
,
1259 xprt
->reestablish_timeout
);
1260 xprt
->reestablish_timeout
<<= 1;
1261 if (xprt
->reestablish_timeout
> XS_TCP_MAX_REEST_TO
)
1262 xprt
->reestablish_timeout
= XS_TCP_MAX_REEST_TO
;
1264 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
1265 schedule_work(&xprt
->connect_worker
);
1267 /* flush_scheduled_work can sleep... */
1268 if (!RPC_IS_ASYNC(task
))
1269 flush_scheduled_work();
1274 * xs_udp_print_stats - display UDP socket-specifc stats
1275 * @xprt: rpc_xprt struct containing statistics
1279 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1281 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1283 xprt
->stat
.bind_count
,
1286 xprt
->stat
.bad_xids
,
1288 xprt
->stat
.bklog_u
);
1292 * xs_tcp_print_stats - display TCP socket-specifc stats
1293 * @xprt: rpc_xprt struct containing statistics
1297 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1301 if (xprt_connected(xprt
))
1302 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
1304 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1306 xprt
->stat
.bind_count
,
1307 xprt
->stat
.connect_count
,
1308 xprt
->stat
.connect_time
,
1312 xprt
->stat
.bad_xids
,
1314 xprt
->stat
.bklog_u
);
1317 static struct rpc_xprt_ops xs_udp_ops
= {
1318 .set_buffer_size
= xs_udp_set_buffer_size
,
1319 .print_addr
= xs_print_peer_address
,
1320 .reserve_xprt
= xprt_reserve_xprt_cong
,
1321 .release_xprt
= xprt_release_xprt_cong
,
1322 .rpcbind
= rpc_getport
,
1323 .set_port
= xs_set_port
,
1324 .connect
= xs_connect
,
1325 .buf_alloc
= rpc_malloc
,
1326 .buf_free
= rpc_free
,
1327 .send_request
= xs_udp_send_request
,
1328 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
1329 .timer
= xs_udp_timer
,
1330 .release_request
= xprt_release_rqst_cong
,
1332 .destroy
= xs_destroy
,
1333 .print_stats
= xs_udp_print_stats
,
1336 static struct rpc_xprt_ops xs_tcp_ops
= {
1337 .print_addr
= xs_print_peer_address
,
1338 .reserve_xprt
= xprt_reserve_xprt
,
1339 .release_xprt
= xs_tcp_release_xprt
,
1340 .rpcbind
= rpc_getport
,
1341 .set_port
= xs_set_port
,
1342 .connect
= xs_connect
,
1343 .buf_alloc
= rpc_malloc
,
1344 .buf_free
= rpc_free
,
1345 .send_request
= xs_tcp_send_request
,
1346 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
1348 .destroy
= xs_destroy
,
1349 .print_stats
= xs_tcp_print_stats
,
1353 * xs_setup_udp - Set up transport to use a UDP socket
1354 * @xprt: transport to set up
1355 * @to: timeout parameters
1358 int xs_setup_udp(struct rpc_xprt
*xprt
, struct rpc_timeout
*to
)
1360 size_t slot_table_size
;
1361 struct sockaddr_in
*addr
= (struct sockaddr_in
*) &xprt
->addr
;
1363 xprt
->max_reqs
= xprt_udp_slot_table_entries
;
1364 slot_table_size
= xprt
->max_reqs
* sizeof(xprt
->slot
[0]);
1365 xprt
->slot
= kzalloc(slot_table_size
, GFP_KERNEL
);
1366 if (xprt
->slot
== NULL
)
1369 if (ntohs(addr
->sin_port
!= 0))
1370 xprt_set_bound(xprt
);
1371 xprt
->port
= xs_get_random_port();
1373 xprt
->prot
= IPPROTO_UDP
;
1375 /* XXX: header size can vary due to auth type, IPv6, etc. */
1376 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
1378 INIT_WORK(&xprt
->connect_worker
, xs_udp_connect_worker
, xprt
);
1379 xprt
->bind_timeout
= XS_BIND_TO
;
1380 xprt
->connect_timeout
= XS_UDP_CONN_TO
;
1381 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
1382 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
1384 xprt
->ops
= &xs_udp_ops
;
1387 xprt
->timeout
= *to
;
1389 xprt_set_timeout(&xprt
->timeout
, 5, 5 * HZ
);
1391 xs_format_peer_addresses(xprt
);
1392 dprintk("RPC: set up transport to address %s\n",
1393 xs_print_peer_address(xprt
, RPC_DISPLAY_ALL
));
1399 * xs_setup_tcp - Set up transport to use a TCP socket
1400 * @xprt: transport to set up
1401 * @to: timeout parameters
1404 int xs_setup_tcp(struct rpc_xprt
*xprt
, struct rpc_timeout
*to
)
1406 size_t slot_table_size
;
1407 struct sockaddr_in
*addr
= (struct sockaddr_in
*) &xprt
->addr
;
1409 xprt
->max_reqs
= xprt_tcp_slot_table_entries
;
1410 slot_table_size
= xprt
->max_reqs
* sizeof(xprt
->slot
[0]);
1411 xprt
->slot
= kzalloc(slot_table_size
, GFP_KERNEL
);
1412 if (xprt
->slot
== NULL
)
1415 if (ntohs(addr
->sin_port
) != 0)
1416 xprt_set_bound(xprt
);
1417 xprt
->port
= xs_get_random_port();
1419 xprt
->prot
= IPPROTO_TCP
;
1420 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
1421 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
1423 INIT_WORK(&xprt
->connect_worker
, xs_tcp_connect_worker
, xprt
);
1424 xprt
->bind_timeout
= XS_BIND_TO
;
1425 xprt
->connect_timeout
= XS_TCP_CONN_TO
;
1426 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1427 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
1429 xprt
->ops
= &xs_tcp_ops
;
1432 xprt
->timeout
= *to
;
1434 xprt_set_timeout(&xprt
->timeout
, 2, 60 * HZ
);
1436 xs_format_peer_addresses(xprt
);
1437 dprintk("RPC: set up transport to address %s\n",
1438 xs_print_peer_address(xprt
, RPC_DISPLAY_ALL
));