Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/net/sunrpc/svcsock.c |
| 3 | * |
| 4 | * These are the RPC server socket internals. |
| 5 | * |
| 6 | * The server scheduling algorithm does not always distribute the load |
| 7 | * evenly when servicing a single client. May need to modify the |
Tom Tucker | f6150c3 | 2007-12-30 21:07:57 -0600 | [diff] [blame] | 8 | * svc_xprt_enqueue procedure... |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | * |
| 10 | * TCP support is largely untested and may be a little slow. The problem |
| 11 | * is that we currently do two separate recvfrom's, one for the 4-byte |
| 12 | * record length, and the second for the actual record. This could possibly |
| 13 | * be improved by always reading a minimum size of around 100 bytes and |
| 14 | * tucking any superfluous bytes away in a temporary store. Still, that |
| 15 | * leaves write requests out in the rain. An alternative may be to peek at |
| 16 | * the first skb in the queue, and if it matches the next TCP sequence |
| 17 | * number, to extract the record marker. Yuck. |
| 18 | * |
| 19 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> |
| 20 | */ |
| 21 | |
Ilpo Järvinen | 172589c | 2007-08-28 15:50:33 -0700 | [diff] [blame] | 22 | #include <linux/kernel.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include <linux/sched.h> |
| 24 | #include <linux/errno.h> |
| 25 | #include <linux/fcntl.h> |
| 26 | #include <linux/net.h> |
| 27 | #include <linux/in.h> |
| 28 | #include <linux/inet.h> |
| 29 | #include <linux/udp.h> |
Andrew Morton | 91483c4 | 2005-08-09 20:20:07 -0700 | [diff] [blame] | 30 | #include <linux/tcp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | #include <linux/unistd.h> |
| 32 | #include <linux/slab.h> |
| 33 | #include <linux/netdevice.h> |
| 34 | #include <linux/skbuff.h> |
NeilBrown | b41b66d | 2006-10-02 02:17:48 -0700 | [diff] [blame] | 35 | #include <linux/file.h> |
Nigel Cunningham | 7dfb710 | 2006-12-06 20:34:23 -0800 | [diff] [blame] | 36 | #include <linux/freezer.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | #include <net/sock.h> |
| 38 | #include <net/checksum.h> |
| 39 | #include <net/ip.h> |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 40 | #include <net/ipv6.h> |
Chuck Lever | b7872fe | 2008-04-14 12:27:01 -0400 | [diff] [blame] | 41 | #include <net/tcp.h> |
Arnaldo Carvalho de Melo | c752f07 | 2005-08-09 20:08:28 -0700 | [diff] [blame] | 42 | #include <net/tcp_states.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | #include <asm/uaccess.h> |
| 44 | #include <asm/ioctls.h> |
| 45 | |
| 46 | #include <linux/sunrpc/types.h> |
Chuck Lever | ad06e4b | 2007-02-12 00:53:32 -0800 | [diff] [blame] | 47 | #include <linux/sunrpc/clnt.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | #include <linux/sunrpc/xdr.h> |
Chuck Lever | c0401ea | 2008-04-14 12:27:30 -0400 | [diff] [blame] | 49 | #include <linux/sunrpc/msg_prot.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | #include <linux/sunrpc/svcsock.h> |
| 51 | #include <linux/sunrpc/stats.h> |
| 52 | |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 53 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | |
| 55 | |
| 56 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, |
Chuck Lever | 6b17433 | 2007-02-12 00:53:28 -0800 | [diff] [blame] | 57 | int *errp, int flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | static void svc_udp_data_ready(struct sock *, int); |
| 59 | static int svc_udp_recvfrom(struct svc_rqst *); |
| 60 | static int svc_udp_sendto(struct svc_rqst *); |
Tom Tucker | 755ccea | 2007-12-30 21:07:27 -0600 | [diff] [blame] | 61 | static void svc_sock_detach(struct svc_xprt *); |
Trond Myklebust | 69b6ba3 | 2008-12-23 16:30:11 -0500 | [diff] [blame] | 62 | static void svc_tcp_sock_detach(struct svc_xprt *); |
Tom Tucker | 755ccea | 2007-12-30 21:07:27 -0600 | [diff] [blame] | 63 | static void svc_sock_free(struct svc_xprt *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 65 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
| 66 | struct sockaddr *, int, int); |
Peter Zijlstra | ed07536 | 2006-12-06 20:35:24 -0800 | [diff] [blame] | 67 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 68 | static struct lock_class_key svc_key[2]; |
| 69 | static struct lock_class_key svc_slock_key[2]; |
| 70 | |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 71 | static void svc_reclassify_socket(struct socket *sock) |
Peter Zijlstra | ed07536 | 2006-12-06 20:35:24 -0800 | [diff] [blame] | 72 | { |
| 73 | struct sock *sk = sock->sk; |
John Heffner | 02b3d34 | 2007-09-12 10:42:12 +0200 | [diff] [blame] | 74 | BUG_ON(sock_owned_by_user(sk)); |
Peter Zijlstra | ed07536 | 2006-12-06 20:35:24 -0800 | [diff] [blame] | 75 | switch (sk->sk_family) { |
| 76 | case AF_INET: |
| 77 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", |
Tom Tucker | def13d7 | 2007-12-30 21:08:08 -0600 | [diff] [blame] | 78 | &svc_slock_key[0], |
| 79 | "sk_xprt.xpt_lock-AF_INET-NFSD", |
| 80 | &svc_key[0]); |
Peter Zijlstra | ed07536 | 2006-12-06 20:35:24 -0800 | [diff] [blame] | 81 | break; |
| 82 | |
| 83 | case AF_INET6: |
| 84 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", |
Tom Tucker | def13d7 | 2007-12-30 21:08:08 -0600 | [diff] [blame] | 85 | &svc_slock_key[1], |
| 86 | "sk_xprt.xpt_lock-AF_INET6-NFSD", |
| 87 | &svc_key[1]); |
Peter Zijlstra | ed07536 | 2006-12-06 20:35:24 -0800 | [diff] [blame] | 88 | break; |
| 89 | |
| 90 | default: |
| 91 | BUG(); |
| 92 | } |
| 93 | } |
| 94 | #else |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 95 | static void svc_reclassify_socket(struct socket *sock) |
Peter Zijlstra | ed07536 | 2006-12-06 20:35:24 -0800 | [diff] [blame] | 96 | { |
| 97 | } |
| 98 | #endif |
| 99 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | /* |
| 101 | * Release an skbuff after use |
| 102 | */ |
Tom Tucker | 5148bf4 | 2007-12-30 21:07:25 -0600 | [diff] [blame] | 103 | static void svc_release_skb(struct svc_rqst *rqstp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | { |
Tom Tucker | 5148bf4 | 2007-12-30 21:07:25 -0600 | [diff] [blame] | 105 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | |
| 107 | if (skb) { |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 108 | struct svc_sock *svsk = |
| 109 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); |
Tom Tucker | 5148bf4 | 2007-12-30 21:07:25 -0600 | [diff] [blame] | 110 | rqstp->rq_xprt_ctxt = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | |
| 112 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 113 | skb_free_datagram(svsk->sk_sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | } |
| 116 | |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 117 | union svc_pktinfo_u { |
| 118 | struct in_pktinfo pkti; |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 119 | struct in6_pktinfo pkti6; |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 120 | }; |
David S. Miller | bc375ea | 2007-04-12 13:35:59 -0700 | [diff] [blame] | 121 | #define SVC_PKTINFO_SPACE \ |
| 122 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 123 | |
| 124 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) |
| 125 | { |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 126 | struct svc_sock *svsk = |
| 127 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); |
| 128 | switch (svsk->sk_sk->sk_family) { |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 129 | case AF_INET: { |
| 130 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
| 131 | |
| 132 | cmh->cmsg_level = SOL_IP; |
| 133 | cmh->cmsg_type = IP_PKTINFO; |
| 134 | pki->ipi_ifindex = 0; |
| 135 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; |
| 136 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
| 137 | } |
| 138 | break; |
NeilBrown | 5a05ed7 | 2007-03-06 01:42:22 -0800 | [diff] [blame] | 139 | |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 140 | case AF_INET6: { |
| 141 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
| 142 | |
| 143 | cmh->cmsg_level = SOL_IPV6; |
| 144 | cmh->cmsg_type = IPV6_PKTINFO; |
| 145 | pki->ipi6_ifindex = 0; |
| 146 | ipv6_addr_copy(&pki->ipi6_addr, |
| 147 | &rqstp->rq_daddr.addr6); |
| 148 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
| 149 | } |
| 150 | break; |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 151 | } |
| 152 | return; |
| 153 | } |
| 154 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | /* |
| 156 | * Generic sendto routine |
| 157 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 158 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | { |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 160 | struct svc_sock *svsk = |
| 161 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | struct socket *sock = svsk->sk_sock; |
| 163 | int slen; |
David S. Miller | bc375ea | 2007-04-12 13:35:59 -0700 | [diff] [blame] | 164 | union { |
| 165 | struct cmsghdr hdr; |
| 166 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; |
| 167 | } buffer; |
| 168 | struct cmsghdr *cmh = &buffer.hdr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 169 | int len = 0; |
| 170 | int result; |
| 171 | int size; |
| 172 | struct page **ppage = xdr->pages; |
| 173 | size_t base = xdr->page_base; |
| 174 | unsigned int pglen = xdr->page_len; |
| 175 | unsigned int flags = MSG_MORE; |
Pavel Emelyanov | 5216a8e | 2008-02-21 10:57:45 +0300 | [diff] [blame] | 176 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | |
| 178 | slen = xdr->len; |
| 179 | |
| 180 | if (rqstp->rq_prot == IPPROTO_UDP) { |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 181 | struct msghdr msg = { |
| 182 | .msg_name = &rqstp->rq_addr, |
| 183 | .msg_namelen = rqstp->rq_addrlen, |
| 184 | .msg_control = cmh, |
| 185 | .msg_controllen = sizeof(buffer), |
| 186 | .msg_flags = MSG_MORE, |
| 187 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | |
Chuck Lever | b92503b | 2007-02-12 00:53:36 -0800 | [diff] [blame] | 189 | svc_set_cmsg_data(rqstp, cmh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
| 191 | if (sock_sendmsg(sock, &msg, 0) < 0) |
| 192 | goto out; |
| 193 | } |
| 194 | |
| 195 | /* send head */ |
| 196 | if (slen == xdr->head[0].iov_len) |
| 197 | flags = 0; |
NeilBrown | 4452435 | 2006-10-04 02:15:46 -0700 | [diff] [blame] | 198 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
| 199 | xdr->head[0].iov_len, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | if (len != xdr->head[0].iov_len) |
| 201 | goto out; |
| 202 | slen -= xdr->head[0].iov_len; |
| 203 | if (slen == 0) |
| 204 | goto out; |
| 205 | |
| 206 | /* send page data */ |
| 207 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; |
| 208 | while (pglen > 0) { |
| 209 | if (slen == size) |
| 210 | flags = 0; |
Sridhar Samudrala | e6242e9 | 2006-08-07 20:58:01 -0700 | [diff] [blame] | 211 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | if (result > 0) |
| 213 | len += result; |
| 214 | if (result != size) |
| 215 | goto out; |
| 216 | slen -= size; |
| 217 | pglen -= size; |
| 218 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; |
| 219 | base = 0; |
| 220 | ppage++; |
| 221 | } |
| 222 | /* send tail */ |
| 223 | if (xdr->tail[0].iov_len) { |
NeilBrown | 4452435 | 2006-10-04 02:15:46 -0700 | [diff] [blame] | 224 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
| 225 | ((unsigned long)xdr->tail[0].iov_base) |
YOSHIFUJI Hideaki | cca5172 | 2007-02-09 15:38:13 -0800 | [diff] [blame] | 226 | & (PAGE_SIZE-1), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | xdr->tail[0].iov_len, 0); |
| 228 | |
| 229 | if (result > 0) |
| 230 | len += result; |
| 231 | } |
| 232 | out: |
Chuck Lever | ad06e4b | 2007-02-12 00:53:32 -0800 | [diff] [blame] | 233 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 234 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
Chuck Lever | ad06e4b | 2007-02-12 00:53:32 -0800 | [diff] [blame] | 235 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | |
| 237 | return len; |
| 238 | } |
| 239 | |
| 240 | /* |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 241 | * Report socket names for nfsdfs |
| 242 | */ |
Chuck Lever | e7942b9 | 2009-04-23 19:32:48 -0400 | [diff] [blame] | 243 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 244 | { |
| 245 | int len; |
| 246 | |
| 247 | switch(svsk->sk_sk->sk_family) { |
Chuck Lever | e7942b9 | 2009-04-23 19:32:48 -0400 | [diff] [blame] | 248 | case PF_INET: |
| 249 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", |
Harvey Harrison | 21454aa | 2008-10-31 00:54:56 -0700 | [diff] [blame] | 250 | svsk->sk_sk->sk_protocol == IPPROTO_UDP ? |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 251 | "udp" : "tcp", |
Harvey Harrison | 21454aa | 2008-10-31 00:54:56 -0700 | [diff] [blame] | 252 | &inet_sk(svsk->sk_sk)->rcv_saddr, |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 253 | inet_sk(svsk->sk_sk)->num); |
| 254 | break; |
Chuck Lever | 58de2f8 | 2009-04-23 19:32:55 -0400 | [diff] [blame^] | 255 | case PF_INET6: |
| 256 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", |
| 257 | svsk->sk_sk->sk_protocol == IPPROTO_UDP ? |
| 258 | "udp" : "tcp", |
| 259 | &inet6_sk(svsk->sk_sk)->rcv_saddr, |
| 260 | inet_sk(svsk->sk_sk)->num); |
| 261 | break; |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 262 | default: |
Chuck Lever | e7942b9 | 2009-04-23 19:32:48 -0400 | [diff] [blame] | 263 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 264 | svsk->sk_sk->sk_family); |
| 265 | } |
Chuck Lever | e7942b9 | 2009-04-23 19:32:48 -0400 | [diff] [blame] | 266 | |
| 267 | if (len >= remaining) { |
| 268 | *buf = '\0'; |
| 269 | return -ENAMETOOLONG; |
| 270 | } |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 271 | return len; |
| 272 | } |
| 273 | |
Chuck Lever | 8435d34 | 2009-04-23 19:32:40 -0400 | [diff] [blame] | 274 | /** |
| 275 | * svc_sock_names - construct a list of listener names in a string |
| 276 | * @serv: pointer to RPC service |
| 277 | * @buf: pointer to a buffer to fill in with socket names |
| 278 | * @buflen: size of the buffer to be filled |
| 279 | * @toclose: pointer to '\0'-terminated C string containing the name |
| 280 | * of a listener to be closed |
| 281 | * |
| 282 | * Fills in @buf with a '\n'-separated list of names of listener |
| 283 | * sockets. If @toclose is not NULL, the socket named by @toclose |
| 284 | * is closed, and is not included in the output list. |
| 285 | * |
| 286 | * Returns positive length of the socket name string, or a negative |
| 287 | * errno value on error. |
| 288 | */ |
| 289 | int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, |
| 290 | const char *toclose) |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 291 | { |
NeilBrown | b41b66d | 2006-10-02 02:17:48 -0700 | [diff] [blame] | 292 | struct svc_sock *svsk, *closesk = NULL; |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 293 | int len = 0; |
| 294 | |
| 295 | if (!serv) |
| 296 | return 0; |
Chuck Lever | e7942b9 | 2009-04-23 19:32:48 -0400 | [diff] [blame] | 297 | |
NeilBrown | aaf68cf | 2007-02-08 14:20:30 -0800 | [diff] [blame] | 298 | spin_lock_bh(&serv->sv_lock); |
Tom Tucker | 7a18208 | 2007-12-30 21:07:53 -0600 | [diff] [blame] | 299 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
Chuck Lever | e7942b9 | 2009-04-23 19:32:48 -0400 | [diff] [blame] | 300 | int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); |
| 301 | if (onelen < 0) { |
| 302 | len = onelen; |
| 303 | break; |
| 304 | } |
| 305 | if (toclose && strcmp(toclose, buf + len) == 0) |
NeilBrown | b41b66d | 2006-10-02 02:17:48 -0700 | [diff] [blame] | 306 | closesk = svsk; |
| 307 | else |
| 308 | len += onelen; |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 309 | } |
NeilBrown | aaf68cf | 2007-02-08 14:20:30 -0800 | [diff] [blame] | 310 | spin_unlock_bh(&serv->sv_lock); |
Chuck Lever | e7942b9 | 2009-04-23 19:32:48 -0400 | [diff] [blame] | 311 | |
NeilBrown | b41b66d | 2006-10-02 02:17:48 -0700 | [diff] [blame] | 312 | if (closesk) |
NeilBrown | 5680c44 | 2006-10-04 02:15:45 -0700 | [diff] [blame] | 313 | /* Should unregister with portmap, but you cannot |
| 314 | * unregister just one protocol... |
| 315 | */ |
Tom Tucker | 7a18208 | 2007-12-30 21:07:53 -0600 | [diff] [blame] | 316 | svc_close_xprt(&closesk->sk_xprt); |
NeilBrown | 37a0347 | 2006-10-04 02:15:44 -0700 | [diff] [blame] | 317 | else if (toclose) |
| 318 | return -ENOENT; |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 319 | return len; |
| 320 | } |
Trond Myklebust | 24c3767 | 2008-12-23 16:30:12 -0500 | [diff] [blame] | 321 | EXPORT_SYMBOL_GPL(svc_sock_names); |
NeilBrown | 80212d5 | 2006-10-02 02:17:47 -0700 | [diff] [blame] | 322 | |
| 323 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 324 | * Check input queue length |
| 325 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 326 | static int svc_recv_available(struct svc_sock *svsk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 327 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 328 | struct socket *sock = svsk->sk_sock; |
| 329 | int avail, err; |
| 330 | |
Sridhar Samudrala | e6242e9 | 2006-08-07 20:58:01 -0700 | [diff] [blame] | 331 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | |
| 333 | return (err >= 0)? avail : err; |
| 334 | } |
| 335 | |
| 336 | /* |
| 337 | * Generic recvfrom routine. |
| 338 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 339 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
| 340 | int buflen) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | { |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 342 | struct svc_sock *svsk = |
| 343 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); |
Chuck Lever | 1ba9510 | 2007-02-12 00:53:31 -0800 | [diff] [blame] | 344 | struct msghdr msg = { |
| 345 | .msg_flags = MSG_DONTWAIT, |
| 346 | }; |
| 347 | int len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | |
Tom Tucker | 260c1d1 | 2007-12-30 21:08:29 -0600 | [diff] [blame] | 349 | rqstp->rq_xprt_hlen = 0; |
| 350 | |
Chuck Lever | 1ba9510 | 2007-02-12 00:53:31 -0800 | [diff] [blame] | 351 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
| 352 | msg.msg_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 353 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
Chuck Lever | 1ba9510 | 2007-02-12 00:53:31 -0800 | [diff] [blame] | 355 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | return len; |
| 357 | } |
| 358 | |
| 359 | /* |
| 360 | * Set socket snd and rcv buffer lengths |
| 361 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 362 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
| 363 | unsigned int rcv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | { |
| 365 | #if 0 |
| 366 | mm_segment_t oldfs; |
| 367 | oldfs = get_fs(); set_fs(KERNEL_DS); |
| 368 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, |
| 369 | (char*)&snd, sizeof(snd)); |
| 370 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, |
| 371 | (char*)&rcv, sizeof(rcv)); |
| 372 | #else |
| 373 | /* sock_setsockopt limits use to sysctl_?mem_max, |
| 374 | * which isn't acceptable. Until that is made conditional |
| 375 | * on not having CAP_SYS_RESOURCE or similar, we go direct... |
| 376 | * DaveM said I could! |
| 377 | */ |
| 378 | lock_sock(sock->sk); |
| 379 | sock->sk->sk_sndbuf = snd * 2; |
| 380 | sock->sk->sk_rcvbuf = rcv * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 381 | release_sock(sock->sk); |
| 382 | #endif |
| 383 | } |
| 384 | /* |
| 385 | * INET callback when data has been received on the socket. |
| 386 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 387 | static void svc_udp_data_ready(struct sock *sk, int count) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | { |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 389 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 391 | if (svsk) { |
| 392 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 393 | svsk, sk, count, |
| 394 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
| 395 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
Tom Tucker | f6150c3 | 2007-12-30 21:07:57 -0600 | [diff] [blame] | 396 | svc_xprt_enqueue(&svsk->sk_xprt); |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 397 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
| 399 | wake_up_interruptible(sk->sk_sleep); |
| 400 | } |
| 401 | |
| 402 | /* |
| 403 | * INET callback when space is newly available on the socket. |
| 404 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 405 | static void svc_write_space(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | { |
| 407 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); |
| 408 | |
| 409 | if (svsk) { |
| 410 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 411 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
Tom Tucker | f6150c3 | 2007-12-30 21:07:57 -0600 | [diff] [blame] | 412 | svc_xprt_enqueue(&svsk->sk_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | } |
| 414 | |
| 415 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 416 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | svsk); |
| 418 | wake_up_interruptible(sk->sk_sleep); |
| 419 | } |
| 420 | } |
| 421 | |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 422 | /* |
| 423 | * Copy the UDP datagram's destination address to the rqstp structure. |
| 424 | * The 'destination' address in this case is the address to which the |
| 425 | * peer sent the datagram, i.e. our local address. For multihomed |
| 426 | * hosts, this can change from msg to msg. Note that only the IP |
| 427 | * address changes, the port number should remain the same. |
| 428 | */ |
| 429 | static void svc_udp_get_dest_address(struct svc_rqst *rqstp, |
| 430 | struct cmsghdr *cmh) |
Chuck Lever | 9575648 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 431 | { |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 432 | struct svc_sock *svsk = |
| 433 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); |
| 434 | switch (svsk->sk_sk->sk_family) { |
Chuck Lever | 9575648 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 435 | case AF_INET: { |
NeilBrown | 7a37f57 | 2007-03-06 01:42:21 -0800 | [diff] [blame] | 436 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
| 437 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; |
Chuck Lever | 9575648 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 438 | break; |
NeilBrown | 7a37f57 | 2007-03-06 01:42:21 -0800 | [diff] [blame] | 439 | } |
Chuck Lever | 9575648 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 440 | case AF_INET6: { |
NeilBrown | 7a37f57 | 2007-03-06 01:42:21 -0800 | [diff] [blame] | 441 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
| 442 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); |
Chuck Lever | 9575648 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 443 | break; |
NeilBrown | 7a37f57 | 2007-03-06 01:42:21 -0800 | [diff] [blame] | 444 | } |
Chuck Lever | 9575648 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 445 | } |
Chuck Lever | 9575648 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 446 | } |
| 447 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 448 | /* |
| 449 | * Receive a datagram from a UDP socket. |
| 450 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 451 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 452 | { |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 453 | struct svc_sock *svsk = |
| 454 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 455 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | struct sk_buff *skb; |
David S. Miller | bc375ea | 2007-04-12 13:35:59 -0700 | [diff] [blame] | 457 | union { |
| 458 | struct cmsghdr hdr; |
| 459 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; |
| 460 | } buffer; |
| 461 | struct cmsghdr *cmh = &buffer.hdr; |
NeilBrown | 7a37f57 | 2007-03-06 01:42:21 -0800 | [diff] [blame] | 462 | struct msghdr msg = { |
| 463 | .msg_name = svc_addr(rqstp), |
| 464 | .msg_control = cmh, |
| 465 | .msg_controllen = sizeof(buffer), |
| 466 | .msg_flags = MSG_DONTWAIT, |
| 467 | }; |
Chuck Lever | abc5c44 | 2009-04-23 19:31:25 -0400 | [diff] [blame] | 468 | size_t len; |
| 469 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 471 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | /* udp sockets need large rcvbuf as all pending |
| 473 | * requests are still in that buffer. sndbuf must |
| 474 | * also be large enough that there is enough space |
Greg Banks | 3262c81 | 2006-10-02 02:17:58 -0700 | [diff] [blame] | 475 | * for one reply per thread. We count all threads |
| 476 | * rather than threads in a particular pool, which |
| 477 | * provides an upper bound on the number of threads |
| 478 | * which will access the socket. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | */ |
| 480 | svc_sock_setbufsize(svsk->sk_sock, |
NeilBrown | c6b0a9f | 2006-10-06 00:44:05 -0700 | [diff] [blame] | 481 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
| 482 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 484 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
NeilBrown | 05ed690 | 2007-05-09 02:34:55 -0700 | [diff] [blame] | 485 | skb = NULL; |
| 486 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, |
| 487 | 0, 0, MSG_PEEK | MSG_DONTWAIT); |
| 488 | if (err >= 0) |
| 489 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); |
| 490 | |
| 491 | if (skb == NULL) { |
| 492 | if (err != -EAGAIN) { |
| 493 | /* possibly an icmp error */ |
| 494 | dprintk("svc: recvfrom returned error %d\n", -err); |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 495 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 496 | } |
Tom Tucker | a6046f7 | 2007-12-30 21:08:01 -0600 | [diff] [blame] | 497 | svc_xprt_received(&svsk->sk_xprt); |
NeilBrown | 05ed690 | 2007-05-09 02:34:55 -0700 | [diff] [blame] | 498 | return -EAGAIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 499 | } |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 500 | len = svc_addr_len(svc_addr(rqstp)); |
Chuck Lever | abc5c44 | 2009-04-23 19:31:25 -0400 | [diff] [blame] | 501 | if (len == 0) |
| 502 | return -EAFNOSUPPORT; |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 503 | rqstp->rq_addrlen = len; |
Eric Dumazet | b7aa0bf | 2007-04-19 16:16:32 -0700 | [diff] [blame] | 504 | if (skb->tstamp.tv64 == 0) { |
| 505 | skb->tstamp = ktime_get_real(); |
YOSHIFUJI Hideaki | cca5172 | 2007-02-09 15:38:13 -0800 | [diff] [blame] | 506 | /* Don't enable netstamp, sunrpc doesn't |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | need that much accuracy */ |
| 508 | } |
Eric Dumazet | b7aa0bf | 2007-04-19 16:16:32 -0700 | [diff] [blame] | 509 | svsk->sk_sk->sk_stamp = skb->tstamp; |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 510 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | |
| 512 | /* |
| 513 | * Maybe more packets - kick another thread ASAP. |
| 514 | */ |
Tom Tucker | a6046f7 | 2007-12-30 21:08:01 -0600 | [diff] [blame] | 515 | svc_xprt_received(&svsk->sk_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | |
| 517 | len = skb->len - sizeof(struct udphdr); |
| 518 | rqstp->rq_arg.len = len; |
| 519 | |
Chuck Lever | 9575648 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 520 | rqstp->rq_prot = IPPROTO_UDP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | |
NeilBrown | 7a37f57 | 2007-03-06 01:42:21 -0800 | [diff] [blame] | 522 | if (cmh->cmsg_level != IPPROTO_IP || |
| 523 | cmh->cmsg_type != IP_PKTINFO) { |
| 524 | if (net_ratelimit()) |
| 525 | printk("rpcsvc: received unknown control message:" |
| 526 | "%d/%d\n", |
| 527 | cmh->cmsg_level, cmh->cmsg_type); |
| 528 | skb_free_datagram(svsk->sk_sk, skb); |
| 529 | return 0; |
| 530 | } |
| 531 | svc_udp_get_dest_address(rqstp, cmh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | |
| 533 | if (skb_is_nonlinear(skb)) { |
| 534 | /* we have to copy */ |
| 535 | local_bh_disable(); |
| 536 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { |
| 537 | local_bh_enable(); |
| 538 | /* checksum error */ |
| 539 | skb_free_datagram(svsk->sk_sk, skb); |
| 540 | return 0; |
| 541 | } |
| 542 | local_bh_enable(); |
YOSHIFUJI Hideaki | cca5172 | 2007-02-09 15:38:13 -0800 | [diff] [blame] | 543 | skb_free_datagram(svsk->sk_sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | } else { |
| 545 | /* we can use it in-place */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 546 | rqstp->rq_arg.head[0].iov_base = skb->data + |
| 547 | sizeof(struct udphdr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | rqstp->rq_arg.head[0].iov_len = len; |
Herbert Xu | fb286bb | 2005-11-10 13:01:24 -0800 | [diff] [blame] | 549 | if (skb_checksum_complete(skb)) { |
| 550 | skb_free_datagram(svsk->sk_sk, skb); |
| 551 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | } |
Tom Tucker | 5148bf4 | 2007-12-30 21:07:25 -0600 | [diff] [blame] | 553 | rqstp->rq_xprt_ctxt = skb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | } |
| 555 | |
| 556 | rqstp->rq_arg.page_base = 0; |
| 557 | if (len <= rqstp->rq_arg.head[0].iov_len) { |
| 558 | rqstp->rq_arg.head[0].iov_len = len; |
| 559 | rqstp->rq_arg.page_len = 0; |
NeilBrown | 4452435 | 2006-10-04 02:15:46 -0700 | [diff] [blame] | 560 | rqstp->rq_respages = rqstp->rq_pages+1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | } else { |
| 562 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; |
NeilBrown | 4452435 | 2006-10-04 02:15:46 -0700 | [diff] [blame] | 563 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
Ilpo Järvinen | 172589c | 2007-08-28 15:50:33 -0700 | [diff] [blame] | 564 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 565 | } |
| 566 | |
| 567 | if (serv->sv_stats) |
| 568 | serv->sv_stats->netudpcnt++; |
| 569 | |
| 570 | return len; |
| 571 | } |
| 572 | |
| 573 | static int |
| 574 | svc_udp_sendto(struct svc_rqst *rqstp) |
| 575 | { |
| 576 | int error; |
| 577 | |
| 578 | error = svc_sendto(rqstp, &rqstp->rq_res); |
| 579 | if (error == -ECONNREFUSED) |
| 580 | /* ICMP error on earlier request. */ |
| 581 | error = svc_sendto(rqstp, &rqstp->rq_res); |
| 582 | |
| 583 | return error; |
| 584 | } |
| 585 | |
Tom Tucker | e831fe6 | 2007-12-30 21:07:29 -0600 | [diff] [blame] | 586 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
| 587 | { |
| 588 | } |
| 589 | |
Tom Tucker | 323bee3 | 2007-12-30 21:07:31 -0600 | [diff] [blame] | 590 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
| 591 | { |
| 592 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 593 | struct svc_serv *serv = xprt->xpt_server; |
Tom Tucker | 323bee3 | 2007-12-30 21:07:31 -0600 | [diff] [blame] | 594 | unsigned long required; |
| 595 | |
| 596 | /* |
| 597 | * Set the SOCK_NOSPACE flag before checking the available |
| 598 | * sock space. |
| 599 | */ |
| 600 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
Tom Tucker | 7a90e8c | 2007-12-30 21:07:55 -0600 | [diff] [blame] | 601 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
Tom Tucker | 323bee3 | 2007-12-30 21:07:31 -0600 | [diff] [blame] | 602 | if (required*2 > sock_wspace(svsk->sk_sk)) |
| 603 | return 0; |
| 604 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
| 605 | return 1; |
| 606 | } |
| 607 | |
Tom Tucker | 38a417c | 2007-12-30 21:07:36 -0600 | [diff] [blame] | 608 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
| 609 | { |
| 610 | BUG(); |
| 611 | return NULL; |
| 612 | } |
| 613 | |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 614 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
| 615 | struct sockaddr *sa, int salen, |
| 616 | int flags) |
| 617 | { |
| 618 | return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags); |
| 619 | } |
| 620 | |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 621 | static struct svc_xprt_ops svc_udp_ops = { |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 622 | .xpo_create = svc_udp_create, |
Tom Tucker | 5d13799 | 2007-12-30 21:07:23 -0600 | [diff] [blame] | 623 | .xpo_recvfrom = svc_udp_recvfrom, |
| 624 | .xpo_sendto = svc_udp_sendto, |
Tom Tucker | 5148bf4 | 2007-12-30 21:07:25 -0600 | [diff] [blame] | 625 | .xpo_release_rqst = svc_release_skb, |
Tom Tucker | 755ccea | 2007-12-30 21:07:27 -0600 | [diff] [blame] | 626 | .xpo_detach = svc_sock_detach, |
| 627 | .xpo_free = svc_sock_free, |
Tom Tucker | e831fe6 | 2007-12-30 21:07:29 -0600 | [diff] [blame] | 628 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
Tom Tucker | 323bee3 | 2007-12-30 21:07:31 -0600 | [diff] [blame] | 629 | .xpo_has_wspace = svc_udp_has_wspace, |
Tom Tucker | 38a417c | 2007-12-30 21:07:36 -0600 | [diff] [blame] | 630 | .xpo_accept = svc_udp_accept, |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 631 | }; |
| 632 | |
| 633 | static struct svc_xprt_class svc_udp_class = { |
| 634 | .xcl_name = "udp", |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 635 | .xcl_owner = THIS_MODULE, |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 636 | .xcl_ops = &svc_udp_ops, |
Tom Tucker | 4902315 | 2007-12-30 21:07:21 -0600 | [diff] [blame] | 637 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 638 | }; |
| 639 | |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 640 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | { |
NeilBrown | 7a37f57 | 2007-03-06 01:42:21 -0800 | [diff] [blame] | 642 | int one = 1; |
| 643 | mm_segment_t oldfs; |
| 644 | |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 645 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
Tom Tucker | def13d7 | 2007-12-30 21:08:08 -0600 | [diff] [blame] | 646 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
| 648 | svsk->sk_sk->sk_write_space = svc_write_space; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | |
| 650 | /* initialise setting must have enough space to |
YOSHIFUJI Hideaki | cca5172 | 2007-02-09 15:38:13 -0800 | [diff] [blame] | 651 | * receive and respond to one request. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | * svc_udp_recvfrom will re-adjust if necessary |
| 653 | */ |
| 654 | svc_sock_setbufsize(svsk->sk_sock, |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 655 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
| 656 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 658 | /* data might have come in before data_ready set up */ |
| 659 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 660 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
NeilBrown | 7a37f57 | 2007-03-06 01:42:21 -0800 | [diff] [blame] | 661 | |
| 662 | oldfs = get_fs(); |
| 663 | set_fs(KERNEL_DS); |
| 664 | /* make sure we get destination address info */ |
| 665 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, |
| 666 | (char __user *)&one, sizeof(one)); |
| 667 | set_fs(oldfs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | } |
| 669 | |
| 670 | /* |
| 671 | * A data_ready event on a listening socket means there's a connection |
| 672 | * pending. Do not use state_change as a substitute for it. |
| 673 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 674 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | { |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 676 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | |
| 678 | dprintk("svc: socket %p TCP (listen) state change %d\n", |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 679 | sk, sk->sk_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 681 | /* |
| 682 | * This callback may called twice when a new connection |
| 683 | * is established as a child socket inherits everything |
| 684 | * from a parent LISTEN socket. |
| 685 | * 1) data_ready method of the parent socket will be called |
| 686 | * when one of child sockets become ESTABLISHED. |
| 687 | * 2) data_ready method of the child socket may be called |
| 688 | * when it receives data before the socket is accepted. |
| 689 | * In case of 2, we should ignore it silently. |
| 690 | */ |
| 691 | if (sk->sk_state == TCP_LISTEN) { |
| 692 | if (svsk) { |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 693 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
Tom Tucker | f6150c3 | 2007-12-30 21:07:57 -0600 | [diff] [blame] | 694 | svc_xprt_enqueue(&svsk->sk_xprt); |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 695 | } else |
| 696 | printk("svc: socket %p: no user data\n", sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | } |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 698 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 699 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
| 700 | wake_up_interruptible_all(sk->sk_sleep); |
| 701 | } |
| 702 | |
| 703 | /* |
| 704 | * A state change on a connected socket means it's dying or dead. |
| 705 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 706 | static void svc_tcp_state_change(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 707 | { |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 708 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 709 | |
| 710 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 711 | sk, sk->sk_state, sk->sk_user_data); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 712 | |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 713 | if (!svsk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 714 | printk("svc: socket %p: no user data\n", sk); |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 715 | else { |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 716 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
Tom Tucker | f6150c3 | 2007-12-30 21:07:57 -0600 | [diff] [blame] | 717 | svc_xprt_enqueue(&svsk->sk_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 719 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
| 720 | wake_up_interruptible_all(sk->sk_sleep); |
| 721 | } |
| 722 | |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 723 | static void svc_tcp_data_ready(struct sock *sk, int count) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | { |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 725 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | |
| 727 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 728 | sk, sk->sk_user_data); |
| 729 | if (svsk) { |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 730 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
Tom Tucker | f6150c3 | 2007-12-30 21:07:57 -0600 | [diff] [blame] | 731 | svc_xprt_enqueue(&svsk->sk_xprt); |
Neil Brown | 939bb7e | 2005-09-13 01:25:39 -0700 | [diff] [blame] | 732 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 733 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
| 734 | wake_up_interruptible(sk->sk_sleep); |
| 735 | } |
| 736 | |
| 737 | /* |
| 738 | * Accept a TCP connection |
| 739 | */ |
Tom Tucker | 38a417c | 2007-12-30 21:07:36 -0600 | [diff] [blame] | 740 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 741 | { |
Tom Tucker | 38a417c | 2007-12-30 21:07:36 -0600 | [diff] [blame] | 742 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
akpm@linux-foundation.org | cdd88b9 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 743 | struct sockaddr_storage addr; |
| 744 | struct sockaddr *sin = (struct sockaddr *) &addr; |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 745 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 746 | struct socket *sock = svsk->sk_sock; |
| 747 | struct socket *newsock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 748 | struct svc_sock *newsvsk; |
| 749 | int err, slen; |
Pavel Emelyanov | 5216a8e | 2008-02-21 10:57:45 +0300 | [diff] [blame] | 750 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 751 | |
| 752 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); |
| 753 | if (!sock) |
Tom Tucker | 38a417c | 2007-12-30 21:07:36 -0600 | [diff] [blame] | 754 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 755 | |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 756 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
Sridhar Samudrala | e6242e9 | 2006-08-07 20:58:01 -0700 | [diff] [blame] | 757 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
| 758 | if (err < 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | if (err == -ENOMEM) |
| 760 | printk(KERN_WARNING "%s: no more sockets!\n", |
| 761 | serv->sv_name); |
Sridhar Samudrala | e6242e9 | 2006-08-07 20:58:01 -0700 | [diff] [blame] | 762 | else if (err != -EAGAIN && net_ratelimit()) |
| 763 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
| 764 | serv->sv_name, -err); |
Tom Tucker | 38a417c | 2007-12-30 21:07:36 -0600 | [diff] [blame] | 765 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 766 | } |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 767 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 768 | |
akpm@linux-foundation.org | cdd88b9 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 769 | err = kernel_getpeername(newsock, sin, &slen); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | if (err < 0) { |
| 771 | if (net_ratelimit()) |
| 772 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", |
| 773 | serv->sv_name, -err); |
| 774 | goto failed; /* aborted connection or whatever */ |
| 775 | } |
| 776 | |
| 777 | /* Ideally, we would want to reject connections from unauthorized |
Chuck Lever | ad06e4b | 2007-02-12 00:53:32 -0800 | [diff] [blame] | 778 | * hosts here, but when we get encryption, the IP of the host won't |
| 779 | * tell us anything. For now just warn about unpriv connections. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | */ |
akpm@linux-foundation.org | cdd88b9 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 781 | if (!svc_port_is_privileged(sin)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | dprintk(KERN_WARNING |
Chuck Lever | ad06e4b | 2007-02-12 00:53:32 -0800 | [diff] [blame] | 783 | "%s: connect from unprivileged port: %s\n", |
YOSHIFUJI Hideaki | cca5172 | 2007-02-09 15:38:13 -0800 | [diff] [blame] | 784 | serv->sv_name, |
akpm@linux-foundation.org | cdd88b9 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 785 | __svc_print_addr(sin, buf, sizeof(buf))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 786 | } |
Chuck Lever | ad06e4b | 2007-02-12 00:53:32 -0800 | [diff] [blame] | 787 | dprintk("%s: connect from %s\n", serv->sv_name, |
akpm@linux-foundation.org | cdd88b9 | 2007-02-12 00:53:38 -0800 | [diff] [blame] | 788 | __svc_print_addr(sin, buf, sizeof(buf))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | |
| 790 | /* make sure that a write doesn't block forever when |
| 791 | * low on memory |
| 792 | */ |
| 793 | newsock->sk->sk_sndtimeo = HZ*30; |
| 794 | |
Chuck Lever | 6b17433 | 2007-02-12 00:53:28 -0800 | [diff] [blame] | 795 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
| 796 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | goto failed; |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 798 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
Frank van Maarseveen | a974769 | 2007-07-09 22:21:39 +0200 | [diff] [blame] | 799 | err = kernel_getsockname(newsock, sin, &slen); |
| 800 | if (unlikely(err < 0)) { |
| 801 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); |
| 802 | slen = offsetof(struct sockaddr, sa_data); |
| 803 | } |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 804 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
Chuck Lever | 067d781 | 2007-02-12 00:53:30 -0800 | [diff] [blame] | 805 | |
Tom Tucker | f9f3cc4 | 2007-12-30 21:07:40 -0600 | [diff] [blame] | 806 | if (serv->sv_stats) |
| 807 | serv->sv_stats->nettcpconn++; |
| 808 | |
| 809 | return &newsvsk->sk_xprt; |
| 810 | |
| 811 | failed: |
| 812 | sock_release(newsock); |
| 813 | return NULL; |
| 814 | } |
| 815 | |
| 816 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 817 | * Receive data from a TCP socket. |
| 818 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 819 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 820 | { |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 821 | struct svc_sock *svsk = |
| 822 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 823 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 824 | int len; |
NeilBrown | 3cc03b1 | 2006-10-04 02:15:47 -0700 | [diff] [blame] | 825 | struct kvec *vec; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 826 | int pnum, vlen; |
| 827 | |
| 828 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 829 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), |
| 830 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), |
| 831 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 832 | |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 833 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | |
| 835 | /* Receive data. If we haven't got the record length yet, get |
| 836 | * the next four bytes. Otherwise try to gobble up as much as |
| 837 | * possible up to the complete record length. |
| 838 | */ |
Chuck Lever | c0401ea | 2008-04-14 12:27:30 -0400 | [diff] [blame] | 839 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
| 840 | int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 841 | struct kvec iov; |
| 842 | |
| 843 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; |
| 844 | iov.iov_len = want; |
| 845 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) |
| 846 | goto error; |
| 847 | svsk->sk_tcplen += len; |
| 848 | |
| 849 | if (len < want) { |
Chuck Lever | c0401ea | 2008-04-14 12:27:30 -0400 | [diff] [blame] | 850 | dprintk("svc: short recvfrom while reading record " |
| 851 | "length (%d of %d)\n", len, want); |
Tom Tucker | a6046f7 | 2007-12-30 21:08:01 -0600 | [diff] [blame] | 852 | svc_xprt_received(&svsk->sk_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 853 | return -EAGAIN; /* record header not complete */ |
| 854 | } |
| 855 | |
| 856 | svsk->sk_reclen = ntohl(svsk->sk_reclen); |
Chuck Lever | c0401ea | 2008-04-14 12:27:30 -0400 | [diff] [blame] | 857 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | /* FIXME: technically, a record can be fragmented, |
| 859 | * and non-terminal fragments will not have the top |
| 860 | * bit set in the fragment length header. |
| 861 | * But apparently no known nfs clients send fragmented |
| 862 | * records. */ |
NeilBrown | 34e9a63 | 2007-01-29 13:19:52 -0800 | [diff] [blame] | 863 | if (net_ratelimit()) |
Chuck Lever | c0401ea | 2008-04-14 12:27:30 -0400 | [diff] [blame] | 864 | printk(KERN_NOTICE "RPC: multiple fragments " |
| 865 | "per record not supported\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 866 | goto err_delete; |
| 867 | } |
Chuck Lever | c0401ea | 2008-04-14 12:27:30 -0400 | [diff] [blame] | 868 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 869 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
NeilBrown | c6b0a9f | 2006-10-06 00:44:05 -0700 | [diff] [blame] | 870 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
NeilBrown | 34e9a63 | 2007-01-29 13:19:52 -0800 | [diff] [blame] | 871 | if (net_ratelimit()) |
Chuck Lever | c0401ea | 2008-04-14 12:27:30 -0400 | [diff] [blame] | 872 | printk(KERN_NOTICE "RPC: " |
| 873 | "fragment too large: 0x%08lx\n", |
| 874 | (unsigned long)svsk->sk_reclen); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 875 | goto err_delete; |
| 876 | } |
| 877 | } |
| 878 | |
| 879 | /* Check whether enough data is available */ |
| 880 | len = svc_recv_available(svsk); |
| 881 | if (len < 0) |
| 882 | goto error; |
| 883 | |
| 884 | if (len < svsk->sk_reclen) { |
| 885 | dprintk("svc: incomplete TCP record (%d of %d)\n", |
| 886 | len, svsk->sk_reclen); |
Tom Tucker | a6046f7 | 2007-12-30 21:08:01 -0600 | [diff] [blame] | 887 | svc_xprt_received(&svsk->sk_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 888 | return -EAGAIN; /* record not complete */ |
| 889 | } |
| 890 | len = svsk->sk_reclen; |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 891 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 892 | |
NeilBrown | 3cc03b1 | 2006-10-04 02:15:47 -0700 | [diff] [blame] | 893 | vec = rqstp->rq_vec; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | vec[0] = rqstp->rq_arg.head[0]; |
| 895 | vlen = PAGE_SIZE; |
| 896 | pnum = 1; |
| 897 | while (vlen < len) { |
NeilBrown | 4452435 | 2006-10-04 02:15:46 -0700 | [diff] [blame] | 898 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 899 | vec[pnum].iov_len = PAGE_SIZE; |
| 900 | pnum++; |
| 901 | vlen += PAGE_SIZE; |
| 902 | } |
NeilBrown | 4452435 | 2006-10-04 02:15:46 -0700 | [diff] [blame] | 903 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 904 | |
| 905 | /* Now receive data */ |
| 906 | len = svc_recvfrom(rqstp, vec, pnum, len); |
| 907 | if (len < 0) |
| 908 | goto error; |
| 909 | |
| 910 | dprintk("svc: TCP complete record (%d bytes)\n", len); |
| 911 | rqstp->rq_arg.len = len; |
| 912 | rqstp->rq_arg.page_base = 0; |
| 913 | if (len <= rqstp->rq_arg.head[0].iov_len) { |
| 914 | rqstp->rq_arg.head[0].iov_len = len; |
| 915 | rqstp->rq_arg.page_len = 0; |
| 916 | } else { |
| 917 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; |
| 918 | } |
| 919 | |
Tom Tucker | 5148bf4 | 2007-12-30 21:07:25 -0600 | [diff] [blame] | 920 | rqstp->rq_xprt_ctxt = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | rqstp->rq_prot = IPPROTO_TCP; |
| 922 | |
| 923 | /* Reset TCP read info */ |
| 924 | svsk->sk_reclen = 0; |
| 925 | svsk->sk_tcplen = 0; |
| 926 | |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 927 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
Tom Tucker | a6046f7 | 2007-12-30 21:08:01 -0600 | [diff] [blame] | 928 | svc_xprt_received(&svsk->sk_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | if (serv->sv_stats) |
| 930 | serv->sv_stats->nettcpcnt++; |
| 931 | |
| 932 | return len; |
| 933 | |
| 934 | err_delete: |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 935 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 936 | return -EAGAIN; |
| 937 | |
| 938 | error: |
| 939 | if (len == -EAGAIN) { |
| 940 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); |
Tom Tucker | a6046f7 | 2007-12-30 21:08:01 -0600 | [diff] [blame] | 941 | svc_xprt_received(&svsk->sk_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 942 | } else { |
| 943 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 944 | svsk->sk_xprt.xpt_server->sv_name, -len); |
Olaf Kirch | 93fbf1a | 2006-01-06 00:19:56 -0800 | [diff] [blame] | 945 | goto err_delete; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | } |
| 947 | |
| 948 | return len; |
| 949 | } |
| 950 | |
| 951 | /* |
| 952 | * Send out data on TCP socket. |
| 953 | */ |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 954 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 955 | { |
| 956 | struct xdr_buf *xbufp = &rqstp->rq_res; |
| 957 | int sent; |
Alexey Dobriyan | d8ed029 | 2006-09-26 22:29:38 -0700 | [diff] [blame] | 958 | __be32 reclen; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | |
| 960 | /* Set up the first element of the reply kvec. |
| 961 | * Any other kvecs that may be in use have been taken |
| 962 | * care of by the server implementation itself. |
| 963 | */ |
| 964 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); |
| 965 | memcpy(xbufp->head[0].iov_base, &reclen, 4); |
| 966 | |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 967 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 968 | return -ENOTCONN; |
| 969 | |
| 970 | sent = svc_sendto(rqstp, &rqstp->rq_res); |
| 971 | if (sent != xbufp->len) { |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 972 | printk(KERN_NOTICE |
| 973 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " |
| 974 | "- shutting down socket\n", |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 975 | rqstp->rq_xprt->xpt_server->sv_name, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 976 | (sent<0)?"got error":"sent only", |
| 977 | sent, xbufp->len); |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 978 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
Tom Tucker | f6150c3 | 2007-12-30 21:07:57 -0600 | [diff] [blame] | 979 | svc_xprt_enqueue(rqstp->rq_xprt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | sent = -EAGAIN; |
| 981 | } |
| 982 | return sent; |
| 983 | } |
| 984 | |
Tom Tucker | e831fe6 | 2007-12-30 21:07:29 -0600 | [diff] [blame] | 985 | /* |
| 986 | * Setup response header. TCP has a 4B record length field. |
| 987 | */ |
| 988 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) |
| 989 | { |
| 990 | struct kvec *resv = &rqstp->rq_res.head[0]; |
| 991 | |
| 992 | /* tcp needs a space for the record length... */ |
| 993 | svc_putnl(resv, 0); |
| 994 | } |
| 995 | |
Tom Tucker | 323bee3 | 2007-12-30 21:07:31 -0600 | [diff] [blame] | 996 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
| 997 | { |
Tom Tucker | 57b1d3b | 2007-12-30 21:08:22 -0600 | [diff] [blame] | 998 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 999 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
Tom Tucker | 323bee3 | 2007-12-30 21:07:31 -0600 | [diff] [blame] | 1000 | int required; |
| 1001 | int wspace; |
| 1002 | |
| 1003 | /* |
| 1004 | * Set the SOCK_NOSPACE flag before checking the available |
| 1005 | * sock space. |
| 1006 | */ |
| 1007 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
Tom Tucker | 7a90e8c | 2007-12-30 21:07:55 -0600 | [diff] [blame] | 1008 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
Tom Tucker | 323bee3 | 2007-12-30 21:07:31 -0600 | [diff] [blame] | 1009 | wspace = sk_stream_wspace(svsk->sk_sk); |
| 1010 | |
| 1011 | if (wspace < sk_stream_min_wspace(svsk->sk_sk)) |
| 1012 | return 0; |
| 1013 | if (required * 2 > wspace) |
| 1014 | return 0; |
| 1015 | |
| 1016 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
| 1017 | return 1; |
| 1018 | } |
| 1019 | |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 1020 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
| 1021 | struct sockaddr *sa, int salen, |
| 1022 | int flags) |
| 1023 | { |
| 1024 | return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags); |
| 1025 | } |
| 1026 | |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 1027 | static struct svc_xprt_ops svc_tcp_ops = { |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 1028 | .xpo_create = svc_tcp_create, |
Tom Tucker | 5d13799 | 2007-12-30 21:07:23 -0600 | [diff] [blame] | 1029 | .xpo_recvfrom = svc_tcp_recvfrom, |
| 1030 | .xpo_sendto = svc_tcp_sendto, |
Tom Tucker | 5148bf4 | 2007-12-30 21:07:25 -0600 | [diff] [blame] | 1031 | .xpo_release_rqst = svc_release_skb, |
Trond Myklebust | 69b6ba3 | 2008-12-23 16:30:11 -0500 | [diff] [blame] | 1032 | .xpo_detach = svc_tcp_sock_detach, |
Tom Tucker | 755ccea | 2007-12-30 21:07:27 -0600 | [diff] [blame] | 1033 | .xpo_free = svc_sock_free, |
Tom Tucker | e831fe6 | 2007-12-30 21:07:29 -0600 | [diff] [blame] | 1034 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
Tom Tucker | 323bee3 | 2007-12-30 21:07:31 -0600 | [diff] [blame] | 1035 | .xpo_has_wspace = svc_tcp_has_wspace, |
Tom Tucker | 38a417c | 2007-12-30 21:07:36 -0600 | [diff] [blame] | 1036 | .xpo_accept = svc_tcp_accept, |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 1037 | }; |
| 1038 | |
| 1039 | static struct svc_xprt_class svc_tcp_class = { |
| 1040 | .xcl_name = "tcp", |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 1041 | .xcl_owner = THIS_MODULE, |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 1042 | .xcl_ops = &svc_tcp_ops, |
Tom Tucker | 4902315 | 2007-12-30 21:07:21 -0600 | [diff] [blame] | 1043 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
Tom Tucker | 360d8738 | 2007-12-30 21:07:17 -0600 | [diff] [blame] | 1044 | }; |
| 1045 | |
| 1046 | void svc_init_xprt_sock(void) |
| 1047 | { |
| 1048 | svc_reg_xprt_class(&svc_tcp_class); |
| 1049 | svc_reg_xprt_class(&svc_udp_class); |
| 1050 | } |
| 1051 | |
| 1052 | void svc_cleanup_xprt_sock(void) |
| 1053 | { |
| 1054 | svc_unreg_xprt_class(&svc_tcp_class); |
| 1055 | svc_unreg_xprt_class(&svc_udp_class); |
| 1056 | } |
| 1057 | |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 1058 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | { |
| 1060 | struct sock *sk = svsk->sk_sk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1061 | |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 1062 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
Tom Tucker | def13d7 | 2007-12-30 21:08:08 -0600 | [diff] [blame] | 1063 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | if (sk->sk_state == TCP_LISTEN) { |
| 1065 | dprintk("setting up TCP socket for listening\n"); |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 1066 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1067 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 1068 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | } else { |
| 1070 | dprintk("setting up TCP socket for reading\n"); |
| 1071 | sk->sk_state_change = svc_tcp_state_change; |
| 1072 | sk->sk_data_ready = svc_tcp_data_ready; |
| 1073 | sk->sk_write_space = svc_write_space; |
| 1074 | |
| 1075 | svsk->sk_reclen = 0; |
| 1076 | svsk->sk_tcplen = 0; |
| 1077 | |
Chuck Lever | b7872fe | 2008-04-14 12:27:01 -0400 | [diff] [blame] | 1078 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1079 | |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 1080 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
YOSHIFUJI Hideaki | cca5172 | 2007-02-09 15:38:13 -0800 | [diff] [blame] | 1081 | if (sk->sk_state != TCP_ESTABLISHED) |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 1082 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1083 | } |
| 1084 | } |
| 1085 | |
Tom Tucker | 0f0257e | 2007-12-30 21:08:27 -0600 | [diff] [blame] | 1086 | void svc_sock_update_bufs(struct svc_serv *serv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1087 | { |
| 1088 | /* |
| 1089 | * The number of server threads has changed. Update |
| 1090 | * rcvbuf and sndbuf accordingly on all sockets |
| 1091 | */ |
| 1092 | struct list_head *le; |
| 1093 | |
| 1094 | spin_lock_bh(&serv->sv_lock); |
| 1095 | list_for_each(le, &serv->sv_permsocks) { |
YOSHIFUJI Hideaki | cca5172 | 2007-02-09 15:38:13 -0800 | [diff] [blame] | 1096 | struct svc_sock *svsk = |
Tom Tucker | 7a18208 | 2007-12-30 21:07:53 -0600 | [diff] [blame] | 1097 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 1098 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1099 | } |
| 1100 | list_for_each(le, &serv->sv_tempsocks) { |
| 1101 | struct svc_sock *svsk = |
Tom Tucker | 7a18208 | 2007-12-30 21:07:53 -0600 | [diff] [blame] | 1102 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
Tom Tucker | 02fc6c3 | 2007-12-30 21:07:48 -0600 | [diff] [blame] | 1103 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1104 | } |
| 1105 | spin_unlock_bh(&serv->sv_lock); |
| 1106 | } |
Trond Myklebust | 24c3767 | 2008-12-23 16:30:12 -0500 | [diff] [blame] | 1107 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1108 | |
| 1109 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1110 | * Initialize socket for RPC use and create svc_sock struct |
| 1111 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. |
| 1112 | */ |
Chuck Lever | 6b17433 | 2007-02-12 00:53:28 -0800 | [diff] [blame] | 1113 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
| 1114 | struct socket *sock, |
| 1115 | int *errp, int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1116 | { |
| 1117 | struct svc_sock *svsk; |
| 1118 | struct sock *inet; |
Chuck Lever | 6b17433 | 2007-02-12 00:53:28 -0800 | [diff] [blame] | 1119 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | |
| 1121 | dprintk("svc: svc_setup_socket %p\n", sock); |
Panagiotis Issaris | 0da974f | 2006-07-21 14:51:30 -0700 | [diff] [blame] | 1122 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1123 | *errp = -ENOMEM; |
| 1124 | return NULL; |
| 1125 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1126 | |
| 1127 | inet = sock->sk; |
| 1128 | |
| 1129 | /* Register socket with portmapper */ |
| 1130 | if (*errp >= 0 && pmap_register) |
Chuck Lever | baf01ca | 2009-03-18 20:46:13 -0400 | [diff] [blame] | 1131 | *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1132 | ntohs(inet_sk(inet)->sport)); |
| 1133 | |
| 1134 | if (*errp < 0) { |
| 1135 | kfree(svsk); |
| 1136 | return NULL; |
| 1137 | } |
| 1138 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | inet->sk_user_data = svsk; |
| 1140 | svsk->sk_sock = sock; |
| 1141 | svsk->sk_sk = inet; |
| 1142 | svsk->sk_ostate = inet->sk_state_change; |
| 1143 | svsk->sk_odata = inet->sk_data_ready; |
| 1144 | svsk->sk_owspace = inet->sk_write_space; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1145 | |
| 1146 | /* Initialize the socket */ |
| 1147 | if (sock->type == SOCK_DGRAM) |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 1148 | svc_udp_init(svsk, serv); |
Olga Kornievskaia | 47a14ef | 2008-10-21 14:13:47 -0400 | [diff] [blame] | 1149 | else { |
| 1150 | /* initialise setting must have enough space to |
| 1151 | * receive and respond to one request. |
| 1152 | */ |
| 1153 | svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg, |
| 1154 | 4 * serv->sv_max_mesg); |
Tom Tucker | bb5cf16 | 2007-12-30 21:07:50 -0600 | [diff] [blame] | 1155 | svc_tcp_init(svsk, serv); |
Olga Kornievskaia | 47a14ef | 2008-10-21 14:13:47 -0400 | [diff] [blame] | 1156 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1157 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1158 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
| 1159 | svsk, svsk->sk_sk); |
| 1160 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1161 | return svsk; |
| 1162 | } |
| 1163 | |
Chuck Lever | bfba9ab | 2009-04-23 19:32:33 -0400 | [diff] [blame] | 1164 | /** |
| 1165 | * svc_addsock - add a listener socket to an RPC service |
| 1166 | * @serv: pointer to RPC service to which to add a new listener |
| 1167 | * @fd: file descriptor of the new listener |
| 1168 | * @name_return: pointer to buffer to fill in with name of listener |
| 1169 | * @len: size of the buffer |
| 1170 | * |
| 1171 | * Fills in socket name and returns positive length of name if successful. |
| 1172 | * Name is terminated with '\n'. On error, returns a negative errno |
| 1173 | * value. |
| 1174 | */ |
| 1175 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, |
| 1176 | const size_t len) |
NeilBrown | b41b66d | 2006-10-02 02:17:48 -0700 | [diff] [blame] | 1177 | { |
| 1178 | int err = 0; |
| 1179 | struct socket *so = sockfd_lookup(fd, &err); |
| 1180 | struct svc_sock *svsk = NULL; |
| 1181 | |
| 1182 | if (!so) |
| 1183 | return err; |
| 1184 | if (so->sk->sk_family != AF_INET) |
| 1185 | err = -EAFNOSUPPORT; |
| 1186 | else if (so->sk->sk_protocol != IPPROTO_TCP && |
| 1187 | so->sk->sk_protocol != IPPROTO_UDP) |
| 1188 | err = -EPROTONOSUPPORT; |
| 1189 | else if (so->state > SS_UNCONNECTED) |
| 1190 | err = -EISCONN; |
| 1191 | else { |
Tom Tucker | 2da2c21 | 2008-11-23 09:58:08 -0600 | [diff] [blame] | 1192 | if (!try_module_get(THIS_MODULE)) |
| 1193 | err = -ENOENT; |
| 1194 | else |
| 1195 | svsk = svc_setup_socket(serv, so, &err, |
| 1196 | SVC_SOCK_DEFAULTS); |
NeilBrown | e79eff1 | 2007-02-12 00:53:30 -0800 | [diff] [blame] | 1197 | if (svsk) { |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 1198 | struct sockaddr_storage addr; |
| 1199 | struct sockaddr *sin = (struct sockaddr *)&addr; |
| 1200 | int salen; |
| 1201 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) |
| 1202 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); |
Tom Tucker | 4e5caaa | 2007-12-30 21:08:20 -0600 | [diff] [blame] | 1203 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
| 1204 | spin_lock_bh(&serv->sv_lock); |
| 1205 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); |
| 1206 | spin_unlock_bh(&serv->sv_lock); |
Tom Tucker | a6046f7 | 2007-12-30 21:08:01 -0600 | [diff] [blame] | 1207 | svc_xprt_received(&svsk->sk_xprt); |
NeilBrown | b41b66d | 2006-10-02 02:17:48 -0700 | [diff] [blame] | 1208 | err = 0; |
Tom Tucker | 2da2c21 | 2008-11-23 09:58:08 -0600 | [diff] [blame] | 1209 | } else |
| 1210 | module_put(THIS_MODULE); |
NeilBrown | b41b66d | 2006-10-02 02:17:48 -0700 | [diff] [blame] | 1211 | } |
| 1212 | if (err) { |
| 1213 | sockfd_put(so); |
| 1214 | return err; |
| 1215 | } |
Chuck Lever | e7942b9 | 2009-04-23 19:32:48 -0400 | [diff] [blame] | 1216 | return svc_one_sock_name(svsk, name_return, len); |
NeilBrown | b41b66d | 2006-10-02 02:17:48 -0700 | [diff] [blame] | 1217 | } |
| 1218 | EXPORT_SYMBOL_GPL(svc_addsock); |
| 1219 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1220 | /* |
| 1221 | * Create socket for RPC service. |
| 1222 | */ |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 1223 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
| 1224 | int protocol, |
| 1225 | struct sockaddr *sin, int len, |
| 1226 | int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1227 | { |
| 1228 | struct svc_sock *svsk; |
| 1229 | struct socket *sock; |
| 1230 | int error; |
| 1231 | int type; |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 1232 | struct sockaddr_storage addr; |
| 1233 | struct sockaddr *newsin = (struct sockaddr *)&addr; |
| 1234 | int newlen; |
Trond Myklebust | c69da77 | 2009-03-30 18:59:17 -0400 | [diff] [blame] | 1235 | int family; |
| 1236 | int val; |
Pavel Emelyanov | 5216a8e | 2008-02-21 10:57:45 +0300 | [diff] [blame] | 1237 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1238 | |
Chuck Lever | ad06e4b | 2007-02-12 00:53:32 -0800 | [diff] [blame] | 1239 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
| 1240 | serv->sv_program->pg_name, protocol, |
Chuck Lever | 77f1f67 | 2007-02-12 00:53:39 -0800 | [diff] [blame] | 1241 | __svc_print_addr(sin, buf, sizeof(buf))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | |
| 1243 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { |
| 1244 | printk(KERN_WARNING "svc: only UDP and TCP " |
| 1245 | "sockets supported\n"); |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 1246 | return ERR_PTR(-EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1247 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1248 | |
Trond Myklebust | c69da77 | 2009-03-30 18:59:17 -0400 | [diff] [blame] | 1249 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
| 1250 | switch (sin->sa_family) { |
| 1251 | case AF_INET6: |
| 1252 | family = PF_INET6; |
| 1253 | break; |
| 1254 | case AF_INET: |
| 1255 | family = PF_INET; |
| 1256 | break; |
| 1257 | default: |
| 1258 | return ERR_PTR(-EINVAL); |
| 1259 | } |
| 1260 | |
| 1261 | error = sock_create_kern(family, type, protocol, &sock); |
Chuck Lever | 77f1f67 | 2007-02-12 00:53:39 -0800 | [diff] [blame] | 1262 | if (error < 0) |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 1263 | return ERR_PTR(error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1264 | |
Peter Zijlstra | ed07536 | 2006-12-06 20:35:24 -0800 | [diff] [blame] | 1265 | svc_reclassify_socket(sock); |
| 1266 | |
Trond Myklebust | c69da77 | 2009-03-30 18:59:17 -0400 | [diff] [blame] | 1267 | /* |
| 1268 | * If this is an PF_INET6 listener, we want to avoid |
| 1269 | * getting requests from IPv4 remotes. Those should |
| 1270 | * be shunted to a PF_INET listener via rpcbind. |
| 1271 | */ |
| 1272 | val = 1; |
| 1273 | if (family == PF_INET6) |
| 1274 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, |
| 1275 | (char *)&val, sizeof(val)); |
| 1276 | |
Eric Sesterhenn | 1811474 | 2006-09-28 14:37:07 -0700 | [diff] [blame] | 1277 | if (type == SOCK_STREAM) |
Chuck Lever | 77f1f67 | 2007-02-12 00:53:39 -0800 | [diff] [blame] | 1278 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
| 1279 | error = kernel_bind(sock, sin, len); |
Eric Sesterhenn | 1811474 | 2006-09-28 14:37:07 -0700 | [diff] [blame] | 1280 | if (error < 0) |
| 1281 | goto bummer; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1282 | |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 1283 | newlen = len; |
| 1284 | error = kernel_getsockname(sock, newsin, &newlen); |
| 1285 | if (error < 0) |
| 1286 | goto bummer; |
| 1287 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1288 | if (protocol == IPPROTO_TCP) { |
Sridhar Samudrala | e6242e9 | 2006-08-07 20:58:01 -0700 | [diff] [blame] | 1289 | if ((error = kernel_listen(sock, 64)) < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1290 | goto bummer; |
| 1291 | } |
| 1292 | |
NeilBrown | e79eff1 | 2007-02-12 00:53:30 -0800 | [diff] [blame] | 1293 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
Tom Tucker | 9dbc240 | 2007-12-30 21:08:12 -0600 | [diff] [blame] | 1294 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 1295 | return (struct svc_xprt *)svsk; |
NeilBrown | e79eff1 | 2007-02-12 00:53:30 -0800 | [diff] [blame] | 1296 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1297 | |
| 1298 | bummer: |
| 1299 | dprintk("svc: svc_create_socket error = %d\n", -error); |
| 1300 | sock_release(sock); |
Tom Tucker | b700cbb | 2007-12-30 21:07:42 -0600 | [diff] [blame] | 1301 | return ERR_PTR(error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1302 | } |
| 1303 | |
| 1304 | /* |
Tom Tucker | 755ccea | 2007-12-30 21:07:27 -0600 | [diff] [blame] | 1305 | * Detach the svc_sock from the socket so that no |
| 1306 | * more callbacks occur. |
| 1307 | */ |
| 1308 | static void svc_sock_detach(struct svc_xprt *xprt) |
| 1309 | { |
| 1310 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
| 1311 | struct sock *sk = svsk->sk_sk; |
| 1312 | |
| 1313 | dprintk("svc: svc_sock_detach(%p)\n", svsk); |
| 1314 | |
| 1315 | /* put back the old socket callbacks */ |
| 1316 | sk->sk_state_change = svsk->sk_ostate; |
| 1317 | sk->sk_data_ready = svsk->sk_odata; |
| 1318 | sk->sk_write_space = svsk->sk_owspace; |
Trond Myklebust | 69b6ba3 | 2008-12-23 16:30:11 -0500 | [diff] [blame] | 1319 | |
| 1320 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
| 1321 | wake_up_interruptible(sk->sk_sleep); |
| 1322 | } |
| 1323 | |
| 1324 | /* |
| 1325 | * Disconnect the socket, and reset the callbacks |
| 1326 | */ |
| 1327 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) |
| 1328 | { |
| 1329 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
| 1330 | |
| 1331 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); |
| 1332 | |
| 1333 | svc_sock_detach(xprt); |
| 1334 | |
| 1335 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) |
| 1336 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); |
Tom Tucker | 755ccea | 2007-12-30 21:07:27 -0600 | [diff] [blame] | 1337 | } |
| 1338 | |
| 1339 | /* |
| 1340 | * Free the svc_sock's socket resources and the svc_sock itself. |
| 1341 | */ |
| 1342 | static void svc_sock_free(struct svc_xprt *xprt) |
| 1343 | { |
| 1344 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
| 1345 | dprintk("svc: svc_sock_free(%p)\n", svsk); |
| 1346 | |
Tom Tucker | 755ccea | 2007-12-30 21:07:27 -0600 | [diff] [blame] | 1347 | if (svsk->sk_sock->file) |
| 1348 | sockfd_put(svsk->sk_sock); |
| 1349 | else |
| 1350 | sock_release(svsk->sk_sock); |
| 1351 | kfree(svsk); |
| 1352 | } |