2 * Copyright 2000, International Business Machines Corporation 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
13 Institution: The Information Technology Center, Carnegie-Mellon University
18 /* All this is going away in early 1989 */
19 int newVLDB; /* Compatibility flag */
22 static int newVLDB = 1;
24 #ifndef AFS_PTHREAD_ENV
25 #define USUAL_PRIORITY (LWP_MAX_PRIORITY - 2)
28 * stack size increased from 8K because the HP machine seemed to have trouble
29 * with the smaller stack
31 #define USUAL_STACK_SIZE (24 * 1024)
32 #endif /* !AFS_PTHREAD_ENV */
36 File server synchronization with external volume utilities.
38 #include <afsconfig.h>
39 #include <afs/param.h>
43 #include <sys/types.h>
49 #include <sys/param.h>
50 #include <sys/socket.h>
51 #include <netinet/in.h>
56 #ifdef AFS_PTHREAD_ENV
58 #else /* AFS_PTHREAD_ENV */
59 #include <afs/assert.h>
60 #endif /* AFS_PTHREAD_ENV */
73 #include <afs/afsint.h>
75 #include <afs/errors.h>
79 #include <afs/afssyscalls.h>
83 #include "partition.h"
85 extern int LogLevel; /* Vice loglevel */
86 int (*V_BreakVolumeCallbacks)();
88 #define MAXHANDLERS 4 /* Up to 4 clients; must be at least 2, so that
89 move = dump+restore can run on single server */
90 #define MAXOFFLINEVOLUMES 30 /* This needs to be as big as the maximum
91 number that would be offline for 1 operation.
92 Current winner is salvage, which needs all
93 cloned read-only copies offline when salvaging
94 a single read-write volume */
96 #define MAX_BIND_TRIES 5 /* Number of times to retry socket bind */
104 static struct offlineInfo OfflineVolumes[MAXHANDLERS][MAXOFFLINEVOLUMES];
106 static FS_sd = -1; /* Client socket for talking to file server */
107 static AcceptSd = -1; /* Socket used by server for accepting connections */
109 static int getport();
115 char partName[16]; /* partition name, e.g. /vicepa */
119 /* Forward declarations */
120 static int getport();
121 static void FSYNC_sync();
122 static void FSYNC_newconnection();
123 static void FSYNC_com();
124 static void FSYNC_Drop();
125 static void AcceptOn();
126 static void AcceptOff();
127 static void InitHandler();
128 static void CallHandler(fd_set *fdsetp);
129 static int AddHandler();
130 static int FindHandler();
131 static int FindHandler_r();
132 static int RemoveHandler();
133 static void GetHandler(fd_set *fdsetp, int *maxfdp);
136 * This lock controls access to the handler array. The overhead
137 * is minimal in non-preemptive environments.
139 struct Lock FSYNC_handler_lock;
141 int FSYNC_clientInit(void)
143 struct sockaddr_in addr;
144 /* I can't believe the following is needed for localhost connections!! */
145 static backoff[] = {3,3,3,5,5,5,7,15,16,24,32,40,48,0};
146 int *timeout = &backoff[0];
149 FS_sd = getport(&addr);
150 if (connect(FS_sd, (struct sockaddr *) &addr, sizeof(addr)) >= 0)
152 #if defined(AFS_SGI_ENV)
153 /* down with worthless error messages! */
155 perror("FSYNC_clientInit failed (after many retries)");
162 perror("FSYNC_clientInit temporary failure (will retry)");
167 perror("FSYNC_clientInit failed (giving up!)");
171 void FSYNC_clientFinis(void)
179 Lock_Destroy(&FSYNC_handler_lock);
182 int FSYNC_askfs(VolumeId volume, char *partName, int com, int reason)
185 struct command command;
187 command.volume = volume;
188 command.command = com;
189 command.reason = reason;
191 strcpy(command.partName, partName);
193 command.partName[0] = 0;
196 if (send(FS_sd, (char*)&command, sizeof(command), 0) != sizeof(command)) {
197 printf("FSYNC_askfs: write to file server failed\n");
200 while ((n = recv(FS_sd, &response, 1, 0)) != 1) {
201 if (n == 0 || WSAEINTR != WSAGetLastError()) {
202 printf("FSYNC_askfs: No response from file server\n");
207 if (write(FS_sd, &command, sizeof(command)) != sizeof(command)) {
208 printf("FSYNC_askfs: write to file server failed\n");
211 while ((n = read(FS_sd, &response, 1)) != 1) {
212 if (n == 0 || errno != EINTR) {
213 printf("FSYNC_askfs: No response from file server\n");
219 printf("FSYNC_askfs: negative response from file server; volume %u, command %d\n", command.volume, command.command);
225 void FSYNC_fsInit(void)
227 #ifdef AFS_PTHREAD_ENV
229 pthread_attr_t tattr;
230 assert(pthread_attr_init(&tattr) == 0);
231 assert(pthread_attr_setdetachstate(&tattr, PTHREAD_CREATE_DETACHED) == 0);
232 assert(pthread_create(&tid, &tattr, FSYNC_sync, NULL) == 0);
233 #else /* AFS_PTHREAD_ENV */
235 assert (LWP_CreateProcess(FSYNC_sync, USUAL_STACK_SIZE,
236 USUAL_PRIORITY, (void *) 0,
237 "FSYNC_sync", &pid) == LWP_SUCCESS);
238 #endif /* AFS_PTHREAD_ENV */
241 static int getport(addr)
242 struct sockaddr_in *addr;
246 memset(addr, 0, sizeof(*addr));
247 assert((sd = socket(AF_INET, SOCK_STREAM, 0)) >= 0);
248 #ifdef STRUCT_SOCKADDR_HAS_SA_LEN
249 addr->sin_len = sizeof(struct sockaddr_in);
251 addr->sin_addr.s_addr = htonl(0x7f000001);
252 addr->sin_family = AF_INET; /* was localhost->h_addrtype */
253 addr->sin_port = htons(2040); /* XXXX htons not _really_ neccessary */
258 static void FSYNC_sync() {
259 struct sockaddr_in addr;
264 #ifdef AFS_PTHREAD_ENV
269 signal(SIGPIPE, SIG_IGN);
272 #ifdef AFS_PTHREAD_ENV
273 /* set our 'thread-id' so that the host hold table works */
274 MUTEX_ENTER(&rx_stats_mutex); /* protects rxi_pthread_hinum */
275 tid=++rxi_pthread_hinum;
276 MUTEX_EXIT(&rx_stats_mutex);
277 pthread_setspecific(rx_thread_id_key, (void *)tid);
278 Log("Set thread id %d for FSYNC_sync\n", tid);
279 #endif /* AFS_PTHREAD_ENV */
282 /* Let somebody else run until level > 0. That doesn't mean that
283 * all volumes have been attached. */
284 #ifdef AFS_PTHREAD_ENV
286 #else /* AFS_PTHREAD_ENV */
287 LWP_DispatchProcess();
288 #endif /* AFS_PTHREAD_ENV */
290 AcceptSd = getport(&addr);
291 /* Reuseaddr needed because system inexplicably leaves crud lying around */
292 code = setsockopt(AcceptSd, SOL_SOCKET, SO_REUSEADDR, (char *) &on, sizeof(on));
293 if (code) Log("FSYNC_sync: setsockopt failed with (%d)\n",errno);
295 for (numTries=0; numTries < MAX_BIND_TRIES; numTries++) {
296 if ((code = bind(AcceptSd, (struct sockaddr *) &addr, sizeof(addr))) == 0) break;
297 Log("FSYNC_sync: bind failed with (%d), will sleep and retry\n",errno);
301 listen(AcceptSd,100);
307 GetHandler(&readfds, &maxfd);
308 /* Note: check for >= 1 below is essential since IOMGR_select
309 * doesn't have exactly same semantics as select.
311 #ifdef AFS_PTHREAD_ENV
312 if (select(maxfd+1, &readfds, NULL, NULL, NULL) >= 1)
313 #else /* AFS_PTHREAD_ENV */
314 if (IOMGR_Select(maxfd+1, &readfds, NULL, NULL, NULL) >= 1)
315 #endif /* AFS_PTHREAD_ENV */
316 CallHandler(&readfds);
320 static void FSYNC_newconnection(afd)
323 struct sockaddr_in other;
325 junk = sizeof(other);
326 fd = accept(afd, (struct sockaddr *) &other, &junk);
328 Log("FSYNC_newconnection: accept failed, errno==%d\n", errno);
331 else if (!AddHandler(fd, FSYNC_com)) {
333 assert(AddHandler(fd, FSYNC_com));
341 afs_int32 FS_cnt = 0;
342 static void FSYNC_com(fd)
348 struct command command;
350 register struct offlineInfo *volumes, *v;
352 char tvolName[VMAXPATHLEN];
356 n = recv(fd, &command, sizeof (command), 0);
358 n = read(fd, &command, sizeof (command));
364 if (n < sizeof(command)) {
365 Log("FSYNC_com: partial read (%d instead of %d); dropping connection (cnt=%d)\n", n, sizeof(command), FS_cnt);
371 volumes = OfflineVolumes[FindHandler(fd)];
372 for (v = 0, i = 0; i<MAXOFFLINEVOLUMES; i++) {
373 if (volumes[i].volumeID == command.volume
374 && strcmp(volumes[i].partName, command.partName)==0) {
379 switch (command.command) {
381 /* don't try to put online, this call is made only after deleting
382 a volume, in which case we want to remove the vol # from the
383 OfflineVolumes array only */
384 if (v) v->volumeID = 0;
389 This is where a detatched volume gets reattached. However in the
390 special case where the volume is merely busy, it is already
391 attatched and it is only necessary to clear the busy flag. See
392 defect #2080 for details.
395 /* is the volume already attatched? */
398 * XXX With the following enabled we had bizarre problems where the backup id would
399 * be reset to 0; that was due to the interaction between fileserver/volserver in that they
400 * both keep volumes in memory and the changes wouldn't be made to the fileserver. Some of
401 * the problems were due to refcnt changes as result of VGetVolume/VPutVolume which would call
402 * VOffline, etc. when we don't want to; someday the whole #2080 issue should be revisited to
405 vp=VGetVolume_r(&error,command.volume);
407 /* yep, is the BUSY flag set? */
408 if(vp->specialStatus==VBUSY) {
409 /* test harness for defect #2081 */
413 test #2081 by releasing TEST.2081,
414 so leave it alone here, zap it after
417 if(strcmp(vp->header->diskstuff.name,"TEST.2081")==0)
420 /* yep, clear BUSY flag */
423 /* make sure vol is online */
426 V_inUse(vp)=1; /* online */
435 /* so, we need to attach the volume */
440 sprintf(&tvolName[1], VFORMAT, command.volume);
442 vp = VAttachVolumeByName_r(&error, command.partName, tvolName, V_VOLUPD);
447 case FSYNC_NEEDVOLUME: {
449 /* not already offline, we need to find a slot for newly offline volume */
451 for (i = 0; i<MAXOFFLINEVOLUMES; i++) {
452 if (volumes[i].volumeID == 0) {
462 vp = VGetVolume_r(&error, command.volume);
464 if (command.partName[0] != 0
465 && strcmp(command.partName, vp->partition->name) != 0) {
466 /* volume on desired partition is not online, so we
467 * should treat this as an offline volume.
475 command.command==FSYNC_NEEDVOLUME
476 && (command.reason==V_READONLY
477 || (!VolumeWriteable(vp)
478 && (command.reason==V_CLONE || command.reason==V_DUMP))
482 if (command.command==FSYNC_NEEDVOLUME
483 && (command.reason==V_CLONE || command.reason==V_DUMP)) {
484 vp->specialStatus = VBUSY;
486 /* remember what volume we got, so we can keep track of how
487 * many volumes the volserver or whatever is using. Note that
488 * vp is valid since leaveonline is only set when vp is valid.
490 v->volumeID = command.volume;
491 strcpy(v->partName, vp->partition->name);
493 /* in this case, VOffline just returns sans decrementing
494 * ref count. We could try to fix it, but it has lots of
500 VOffline_r(vp, "A volume utility is running.");
505 VUpdateVolume_r(&error, vp); /* At least get volume stats right */
507 Log("FSYNC: Volume %u (%s) was left on line for an external %s request\n",
508 V_id(vp), V_name(vp),
509 command.reason == V_CLONE? "clone":
510 command.reason == V_READONLY? "readonly":
511 command.reason == V_DUMP? "dump" : "UNKNOWN");
520 case FSYNC_MOVEVOLUME:
521 /* Yuch: the "reason" for the move is the site it got moved to... */
522 /* still set specialStatus so we stop sending back VBUSY.
523 also should still break callbacks. Note that I don't know
524 how to tell if we should break all or not, so we just do it
525 since it doesn't matter much if we do an extra break
526 volume callbacks on a volume move within the same server */
527 vp = VGetVolume_r(&error, command.volume);
529 vp->specialStatus = VMOVED;
533 if (V_BreakVolumeCallbacks) {
534 Log("fssync: volume %u moved to %x; breaking all call backs\n",
535 command.volume, command.reason);
538 (*V_BreakVolumeCallbacks)(command.volume);
543 case FSYNC_RESTOREVOLUME:
544 /* if the volume is being restored, break all callbacks on it*/
545 if (V_BreakVolumeCallbacks) {
546 Log("fssync: volume %u restored; breaking all call backs\n",
550 (*V_BreakVolumeCallbacks)(command.volume);
568 static void FSYNC_Drop(fd)
571 struct offlineInfo *p;
574 char tvolName[VMAXPATHLEN];
578 p = OfflineVolumes[FindHandler(fd)];
579 for (i = 0; i<MAXOFFLINEVOLUMES; i++) {
584 sprintf(&tvolName[1], VFORMAT, p[i].volumeID);
585 vp = VAttachVolumeByName_r(&error, p[i].partName, tvolName, V_VOLUPD);
602 static int AcceptHandler = -1; /* handler id for accept, if turned on */
604 static void AcceptOn() {
605 if (AcceptHandler == -1) {
606 assert(AddHandler(AcceptSd, FSYNC_newconnection));
607 AcceptHandler = FindHandler(AcceptSd);
611 static void AcceptOff() {
612 if (AcceptHandler != -1) {
613 assert(RemoveHandler(AcceptSd));
618 /* The multiple FD handling code. */
620 static int HandlerFD[MAXHANDLERS];
621 static int (*HandlerProc[MAXHANDLERS])();
623 static void InitHandler ()
626 ObtainWriteLock(&FSYNC_handler_lock);
627 for(i=0;i<MAXHANDLERS;i++)
631 ReleaseWriteLock(&FSYNC_handler_lock);
634 static void CallHandler(fd_set *fdsetp)
637 ObtainReadLock(&FSYNC_handler_lock);
638 for(i=0;i<MAXHANDLERS;i++) {
639 if (HandlerFD[i] >= 0 && FD_ISSET(HandlerFD[i], fdsetp)) {
640 ReleaseReadLock(&FSYNC_handler_lock);
641 (*HandlerProc[i])(HandlerFD[i]);
642 ObtainReadLock(&FSYNC_handler_lock);
645 ReleaseReadLock(&FSYNC_handler_lock);
648 static int AddHandler (afd, aproc)
653 ObtainWriteLock(&FSYNC_handler_lock);
654 for(i=0;i<MAXHANDLERS;i++)
655 if (HandlerFD[i] == -1) break;
656 if (i>=MAXHANDLERS) {
657 ReleaseWriteLock(&FSYNC_handler_lock);
661 HandlerProc[i] = aproc;
662 ReleaseWriteLock(&FSYNC_handler_lock);
666 static int FindHandler (afd)
670 ObtainReadLock(&FSYNC_handler_lock);
671 for(i=0;i<MAXHANDLERS;i++)
672 if (HandlerFD[i] == afd) {
673 ReleaseReadLock(&FSYNC_handler_lock);
676 ReleaseReadLock(&FSYNC_handler_lock); /* just in case */
678 return -1; /* satisfy compiler */
681 static int FindHandler_r (afd)
685 for(i=0;i<MAXHANDLERS;i++)
686 if (HandlerFD[i] == afd) {
690 return -1; /* satisfy compiler */
693 static int RemoveHandler (afd)
696 ObtainWriteLock(&FSYNC_handler_lock);
697 HandlerFD[FindHandler_r(afd)] = -1;
698 ReleaseWriteLock(&FSYNC_handler_lock);
702 static void GetHandler (fd_set *fdsetp, int *maxfdp)
705 register int maxfd = -1;
707 ObtainReadLock(&FSYNC_handler_lock); /* just in case */
708 for(i=0;i<MAXHANDLERS;i++)
709 if (HandlerFD[i] != -1) {
710 FD_SET(HandlerFD[i], fdsetp);
711 if (maxfd < HandlerFD[i])
712 maxfd = HandlerFD[i];
715 ReleaseReadLock(&FSYNC_handler_lock); /* just in case */