2 * Copyright 2006-2008, Sine Nomine Associates and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
11 * localhost interprocess communication for servers
13 * currently handled by a localhost socket
14 * (yes, this needs to be replaced someday)
18 #define FD_SETSIZE 65536
21 #include <afsconfig.h>
22 #include <afs/param.h>
28 #include <afs/afsint.h>
29 #include <afs/errors.h>
30 #include <rx/rx_queue.h>
33 #include "daemon_com.h"
36 #include <afs/afssyscalls.h>
40 #include "partition.h"
42 #include <rx/rx_queue.h>
44 #ifdef USE_UNIX_SOCKETS
45 #include <afs/afsutil.h>
49 int (*V_BreakVolumeCallbacks) (VolumeId);
51 #define MAXHANDLERS 4 /* Up to 4 clients; must be at least 2, so that
52 * move = dump+restore can run on single server */
54 #define MAX_BIND_TRIES 5 /* Number of times to retry socket bind */
56 static int SYNC_ask_internal(SYNC_client_state * state, SYNC_command * com, SYNC_response * res);
60 * On AIX, connect() and bind() require use of SUN_LEN() macro;
61 * sizeof(struct sockaddr_un) will not suffice.
63 #if defined(AFS_AIX_ENV) && defined(USE_UNIX_SOCKETS)
64 #define AFS_SOCKADDR_LEN(sa) SUN_LEN(sa)
66 #define AFS_SOCKADDR_LEN(sa) sizeof(*sa)
70 /* daemon com SYNC general interfaces */
73 * fill in sockaddr structure.
75 * @param[in] endpoint pointer to sync endpoint object
76 * @param[out] addr pointer to sockaddr structure
78 * @post sockaddr structure populated using information from
82 SYNC_getAddr(SYNC_endpoint_t * endpoint, SYNC_sockaddr_t * addr)
84 #ifdef USE_UNIX_SOCKETS
85 char tbuffer[AFSDIR_PATH_MAX];
86 #endif /* USE_UNIX_SOCKETS */
88 memset(addr, 0, sizeof(*addr));
90 #ifdef USE_UNIX_SOCKETS
91 strcompose(tbuffer, AFSDIR_PATH_MAX, AFSDIR_SERVER_LOCAL_DIRPATH, "/",
92 endpoint->un, (char *)NULL);
93 addr->sun_family = AF_UNIX;
94 strncpy(addr->sun_path, tbuffer, (sizeof(struct sockaddr_un) - sizeof(short)));
95 #else /* !USE_UNIX_SOCKETS */
96 #ifdef STRUCT_SOCKADDR_HAS_SA_LEN
97 addr->sin_len = sizeof(struct sockaddr_in);
99 addr->sin_addr.s_addr = htonl(0x7f000001);
100 addr->sin_family = AF_INET; /* was localhost->h_addrtype */
101 addr->sin_port = htons(endpoint->in); /* XXXX htons not _really_ neccessary */
102 #endif /* !USE_UNIX_SOCKETS */
106 * get a socket descriptor of the appropriate domain.
108 * @param[in] endpoint pointer to sync endpoint object
110 * @return socket descriptor
112 * @post socket of domain specified in endpoint structure is created and
113 * returned to caller.
116 SYNC_getSock(SYNC_endpoint_t * endpoint)
119 opr_Verify((sd = socket(endpoint->domain, SOCK_STREAM, 0)) >= 0);
123 /* daemon com SYNC client interface */
126 * open a client connection to a sync server
128 * @param[in] state pointer to sync client handle
130 * @return operation status
133 * @note at present, this routine aborts rather than returning an error code
136 SYNC_connect(SYNC_client_state * state)
138 SYNC_sockaddr_t addr;
139 /* I can't believe the following is needed for localhost connections!! */
140 static time_t backoff[] =
141 { 3, 3, 3, 5, 5, 5, 7, 15, 16, 24, 32, 40, 48, 0 };
142 time_t *timeout = &backoff[0];
144 if (state->fd != OSI_NULLSOCKET) {
148 SYNC_getAddr(&state->endpoint, &addr);
151 state->fd = SYNC_getSock(&state->endpoint);
152 if (connect(state->fd, (struct sockaddr *)&addr, AFS_SOCKADDR_LEN(&addr)) >= 0)
157 Log("SYNC_connect: temporary failure on circuit '%s' (will retry)\n",
159 SYNC_disconnect(state);
162 perror("SYNC_connect failed (giving up!)");
167 * forcibly disconnect a sync client handle.
169 * @param[in] state pointer to sync client handle
171 * @retval operation status
175 SYNC_disconnect(SYNC_client_state * state)
177 rk_closesocket(state->fd);
178 state->fd = OSI_NULLSOCKET;
183 * gracefully disconnect a sync client handle.
185 * @param[in] state pointer to sync client handle
187 * @return operation status
188 * @retval SYNC_OK success
191 SYNC_closeChannel(SYNC_client_state * state)
195 SYNC_PROTO_BUF_DECL(ores);
197 if (state->fd == OSI_NULLSOCKET)
200 memset(&com, 0, sizeof(com));
201 memset(&res, 0, sizeof(res));
203 res.payload.len = SYNC_PROTO_MAX_LEN;
204 res.payload.buf = ores;
206 com.hdr.command = SYNC_COM_CHANNEL_CLOSE;
207 com.hdr.command_len = sizeof(SYNC_command_hdr);
208 com.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
210 /* in case the other end dropped, don't do any retries */
211 state->retry_limit = 0;
212 state->hard_timeout = 0;
214 SYNC_ask(state, &com, &res);
215 SYNC_disconnect(state);
221 * forcibly break a client connection, and then create a new connection.
223 * @param[in] state pointer to sync client handle
225 * @post old connection dropped; new connection established
227 * @return @see SYNC_connect()
230 SYNC_reconnect(SYNC_client_state * state)
232 SYNC_disconnect(state);
233 return SYNC_connect(state);
237 * send a command to a sync server and wait for a response.
239 * @param[in] state pointer to sync client handle
240 * @param[in] com command object
241 * @param[out] res response object
243 * @return operation status
244 * @retval SYNC_OK success
245 * @retval SYNC_COM_ERROR communications error
246 * @retval SYNC_BAD_COMMAND server did not recognize command code
248 * @note this routine merely handles error processing; SYNC_ask_internal()
249 * handles the low-level details of communicating with the SYNC server.
251 * @see SYNC_ask_internal
254 SYNC_ask(SYNC_client_state * state, SYNC_command * com, SYNC_response * res)
257 afs_uint32 now, timeout, code=SYNC_OK;
259 if (state->fd == OSI_NULLSOCKET) {
263 if (state->fd == OSI_NULLSOCKET) {
264 return SYNC_COM_ERROR;
267 #ifdef AFS_DEMAND_ATTACH_FS
268 com->hdr.flags |= SYNC_FLAG_DAFS_EXTENSIONS;
271 now = FT_ApproxTime();
272 timeout = now + state->hard_timeout;
274 (tries <= state->retry_limit) && (now <= timeout);
275 tries++, now = FT_ApproxTime()) {
276 code = SYNC_ask_internal(state, com, res);
277 if (code == SYNC_OK) {
279 } else if (code == SYNC_BAD_COMMAND) {
280 Log("SYNC_ask: protocol mismatch on circuit '%s'; make sure "
281 "fileserver, volserver, salvageserver and salvager are same "
282 "version\n", state->proto_name);
284 } else if ((code == SYNC_COM_ERROR) && (tries < state->retry_limit)) {
285 Log("SYNC_ask: protocol communications failure on circuit '%s'; "
286 "attempting reconnect to server\n", state->proto_name);
287 SYNC_reconnect(state);
291 * unknown (probably protocol-specific) response code, pass it up to
292 * the caller, and let them deal with it
298 if (code == SYNC_COM_ERROR) {
299 Log("SYNC_ask: too many / too latent fatal protocol errors on circuit "
300 "'%s'; giving up (tries %d timeout %d)\n",
301 state->proto_name, tries, timeout);
308 * send a command to a sync server and wait for a response.
310 * @param[in] state pointer to sync client handle
311 * @param[in] com command object
312 * @param[out] res response object
314 * @return operation status
315 * @retval SYNC_OK success
316 * @retval SYNC_COM_ERROR communications error
321 SYNC_ask_internal(SYNC_client_state * state, SYNC_command * com, SYNC_response * res)
324 SYNC_PROTO_BUF_DECL(buf);
330 if (state->fd == OSI_NULLSOCKET) {
331 Log("SYNC_ask: invalid sync file descriptor on circuit '%s'\n",
333 res->hdr.response = SYNC_COM_ERROR;
337 if (com->hdr.command_len > SYNC_PROTO_MAX_LEN) {
338 Log("SYNC_ask: internal SYNC buffer too small on circuit '%s'; "
339 "please file a bug\n", state->proto_name);
340 res->hdr.response = SYNC_COM_ERROR;
345 * fill in some common header fields
347 com->hdr.proto_version = state->proto_version;
348 com->hdr.pkt_seq = ++state->pkt_seq;
349 com->hdr.com_seq = ++state->com_seq;
354 com->hdr.pid = getpid();
355 #ifdef AFS_PTHREAD_ENV
356 com->hdr.tid = afs_pointer_to_int(pthread_self());
359 PROCESS handle = LWP_ThreadId();
360 com->hdr.tid = (handle) ? handle->index : 0;
362 #endif /* !AFS_PTHREAD_ENV */
363 #endif /* !AFS_NT40_ENV */
365 memcpy(buf, &com->hdr, sizeof(com->hdr));
366 if (com->payload.len) {
367 memcpy(buf + sizeof(com->hdr), com->payload.buf,
368 com->hdr.command_len - sizeof(com->hdr));
372 n = send(state->fd, buf, com->hdr.command_len, 0);
373 if (n != com->hdr.command_len) {
374 Log("SYNC_ask: write failed on circuit '%s'\n", state->proto_name);
375 res->hdr.response = SYNC_COM_ERROR;
379 if (com->hdr.command == SYNC_COM_CHANNEL_CLOSE) {
380 /* short circuit close channel requests */
381 res->hdr.response = SYNC_OK;
385 n = recv(state->fd, buf, SYNC_PROTO_MAX_LEN, 0);
386 if (n == 0 || (n < 0 && WSAEINTR != WSAGetLastError())) {
387 Log("SYNC_ask: No response on circuit '%s'\n", state->proto_name);
388 res->hdr.response = SYNC_COM_ERROR;
391 #else /* !AFS_NT40_ENV */
392 n = write(state->fd, buf, com->hdr.command_len);
393 if (com->hdr.command_len != n) {
394 Log("SYNC_ask: write failed on circuit '%s'\n", state->proto_name);
395 res->hdr.response = SYNC_COM_ERROR;
399 if (com->hdr.command == SYNC_COM_CHANNEL_CLOSE) {
400 /* short circuit close channel requests */
401 res->hdr.response = SYNC_OK;
405 /* receive the response */
406 iov[0].iov_base = (char *)&res->hdr;
407 iov[0].iov_len = sizeof(res->hdr);
408 if (res->payload.len) {
409 iov[1].iov_base = (char *)res->payload.buf;
410 iov[1].iov_len = res->payload.len;
415 n = readv(state->fd, iov, iovcnt);
416 if (n == 0 || (n < 0 && errno != EINTR)) {
417 Log("SYNC_ask: No response on circuit '%s'\n", state->proto_name);
418 res->hdr.response = SYNC_COM_ERROR;
421 #endif /* !AFS_NT40_ENV */
425 if (n < sizeof(res->hdr)) {
426 Log("SYNC_ask: response too short on circuit '%s'\n",
428 res->hdr.response = SYNC_COM_ERROR;
432 memcpy(&res->hdr, buf, sizeof(res->hdr));
435 if ((n - sizeof(res->hdr)) > res->payload.len) {
436 Log("SYNC_ask: response too long on circuit '%s'\n",
438 res->hdr.response = SYNC_COM_ERROR;
442 memcpy(res->payload.buf, buf + sizeof(res->hdr), n - sizeof(res->hdr));
445 if (res->hdr.response_len != n) {
446 Log("SYNC_ask: length field in response inconsistent "
447 "on circuit '%s'\n", state->proto_name);
448 res->hdr.response = SYNC_COM_ERROR;
451 if (res->hdr.response == SYNC_DENIED) {
452 Log("SYNC_ask: negative response on circuit '%s'\n", state->proto_name);
456 return res->hdr.response;
461 * daemon com SYNC server-side interfaces
465 * receive a command structure off a sync socket.
467 * @param[in] state pointer to server-side state object
468 * @param[in] fd file descriptor on which to perform i/o
469 * @param[out] com sync command object to be populated
471 * @return operation status
472 * @retval SYNC_OK command received
473 * @retval SYNC_COM_ERROR there was a socket communications error
476 SYNC_getCom(SYNC_server_state_t * state,
481 afs_int32 code = SYNC_OK;
483 SYNC_PROTO_BUF_DECL(buf);
490 n = recv(fd, buf, SYNC_PROTO_MAX_LEN, 0);
492 if (n == 0 || (n < 0 && WSAEINTR != WSAGetLastError())) {
493 Log("SYNC_getCom: error receiving command\n");
494 code = SYNC_COM_ERROR;
497 #else /* !AFS_NT40_ENV */
498 iov[0].iov_base = (char *)&com->hdr;
499 iov[0].iov_len = sizeof(com->hdr);
500 if (com->payload.len) {
501 iov[1].iov_base = (char *)com->payload.buf;
502 iov[1].iov_len = com->payload.len;
508 n = readv(fd, iov, iovcnt);
509 if (n == 0 || (n < 0 && errno != EINTR)) {
510 Log("SYNC_getCom: error receiving command\n");
511 code = SYNC_COM_ERROR;
514 #endif /* !AFS_NT40_ENV */
518 if (n < sizeof(com->hdr)) {
519 Log("SYNC_getCom: command too short\n");
520 code = SYNC_COM_ERROR;
524 memcpy(&com->hdr, buf, sizeof(com->hdr));
527 if ((n - sizeof(com->hdr)) > com->payload.len) {
528 Log("SYNC_getCom: command too long\n");
529 code = SYNC_COM_ERROR;
533 memcpy(com->payload.buf, buf + sizeof(com->hdr), n - sizeof(com->hdr));
541 * write a response structure to a sync socket.
543 * @param[in] state handle to server-side state object
544 * @param[in] fd file descriptor on which to perform i/o
545 * @param[in] res handle to response packet
547 * @return operation status
549 * @retval SYNC_COM_ERROR
552 SYNC_putRes(SYNC_server_state_t * state,
557 afs_int32 code = SYNC_OK;
558 SYNC_PROTO_BUF_DECL(buf);
560 if (res->hdr.response_len > (sizeof(res->hdr) + res->payload.len)) {
561 Log("SYNC_putRes: response_len field in response header inconsistent\n");
562 code = SYNC_COM_ERROR;
566 if (res->hdr.response_len > SYNC_PROTO_MAX_LEN) {
567 Log("SYNC_putRes: internal SYNC buffer too small; please file a bug\n");
568 code = SYNC_COM_ERROR;
572 #ifdef AFS_DEMAND_ATTACH_FS
573 res->hdr.flags |= SYNC_FLAG_DAFS_EXTENSIONS;
575 res->hdr.proto_version = state->proto_version;
576 res->hdr.pkt_seq = ++state->pkt_seq;
577 res->hdr.res_seq = ++state->res_seq;
579 memcpy(buf, &res->hdr, sizeof(res->hdr));
580 if (res->payload.len) {
581 memcpy(buf + sizeof(res->hdr), res->payload.buf,
582 res->hdr.response_len - sizeof(res->hdr));
586 n = send(fd, buf, res->hdr.response_len, 0);
587 #else /* !AFS_NT40_ENV */
588 n = write(fd, buf, res->hdr.response_len);
589 #endif /* !AFS_NT40_ENV */
591 if (res->hdr.response_len != n) {
592 Log("SYNC_putRes: write failed\n");
593 res->hdr.response = SYNC_COM_ERROR;
601 /* return 0 for legal (null-terminated) string,
602 * 1 for illegal (unterminated) string */
604 SYNC_verifyProtocolString(char * buf, size_t len)
608 s_len = strnlen(buf, len);
610 return (s_len == len) ? 1 : 0;
614 * clean up old sockets.
616 * @param[in] state server state object
618 * @post unix domain sockets are cleaned up
621 SYNC_cleanupSock(SYNC_server_state_t * state)
623 #ifdef USE_UNIX_SOCKETS
624 remove(state->addr.sun_path);
629 * bind socket and set it to listen state.
631 * @param[in] state server state object
633 * @return operation status
635 * @retval nonzero failure
637 * @post socket bound and set to listen state
640 SYNC_bindSock(SYNC_server_state_t * state)
646 /* Reuseaddr needed because system inexplicably leaves crud lying around */
648 setsockopt(state->fd, SOL_SOCKET, SO_REUSEADDR, (char *)&on,
651 Log("SYNC_bindSock: setsockopt failed with (%d)\n", errno);
653 for (numTries = 0; numTries < state->bind_retry_limit; numTries++) {
654 code = bind(state->fd,
655 (struct sockaddr *)&state->addr,
656 AFS_SOCKADDR_LEN(&state->addr));
659 Log("SYNC_bindSock: bind failed with (%d), will sleep and retry\n",
663 listen(state->fd, state->listen_depth);