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
13 * OpenAFS demand attach fileserver
14 * Salvage server synchronization with fileserver.
17 /* This controls the size of an fd_set; it must be defined early before
18 * the system headers define that type and the macros that operate on it.
19 * Its value should be as large as the maximum file descriptor limit we
20 * are likely to run into on any platform. Right now, that is 65536
21 * which is the default hard fd limit on Solaris 9 */
23 #define FD_SETSIZE 65536
26 #include <afsconfig.h>
27 #include <afs/param.h>
29 #include <afs/procmgmt.h>
36 #include <afs/afsint.h>
37 #include <rx/rx_queue.h>
40 #include <afs/errors.h>
43 #include <afs/afssyscalls.h>
47 #include "partition.h"
49 #include <rx/rx_queue.h>
51 #ifdef USE_UNIX_SOCKETS
52 #include <afs/afsutil.h>
57 #define WCOREDUMP(x) ((x) & 0200)
60 #define MAXHANDLERS 4 /* Up to 4 clients; must be at least 2, so that
61 * move = dump+restore can run on single server */
65 * This lock controls access to the handler array.
67 struct Lock SALVSYNC_handler_lock;
70 #ifdef AFS_DEMAND_ATTACH_FS
72 * SALVSYNC is a feature specific to the demand attach fileserver
75 /* Forward declarations */
76 static void * SALVSYNC_syncThread(void *);
77 static void SALVSYNC_newconnection(osi_socket fd);
78 static void SALVSYNC_com(osi_socket fd);
79 static void SALVSYNC_Drop(osi_socket fd);
80 static void AcceptOn(void);
81 static void AcceptOff(void);
82 static void InitHandler(void);
83 static void CallHandler(fd_set * fdsetp);
84 static int AddHandler(osi_socket afd, void (*aproc) (int));
85 static int FindHandler(osi_socket afd);
86 static int FindHandler_r(osi_socket afd);
87 static int RemoveHandler(osi_socket afd);
88 static void GetHandler(fd_set * fdsetp, int *maxfdp);
90 static int AllocNode(struct SalvageQueueNode ** node);
92 static int AddToSalvageQueue(struct SalvageQueueNode * node);
93 static void DeleteFromSalvageQueue(struct SalvageQueueNode * node);
94 static void AddToPendingQueue(struct SalvageQueueNode * node);
95 static void DeleteFromPendingQueue(struct SalvageQueueNode * node);
96 static struct SalvageQueueNode * LookupPendingCommandByPid(int pid);
97 static void UpdateCommandPrio(struct SalvageQueueNode * node);
98 static void HandlePrio(struct SalvageQueueNode * clone,
99 struct SalvageQueueNode * parent,
100 afs_uint32 new_prio);
102 static int LinkNode(struct SalvageQueueNode * parent,
103 struct SalvageQueueNode * clone);
105 static struct SalvageQueueNode * LookupNode(VolumeId vid, char * partName,
106 struct SalvageQueueNode ** parent);
107 static struct SalvageQueueNode * LookupNodeByCommand(SALVSYNC_command_hdr * qry,
108 struct SalvageQueueNode ** parent);
109 static void AddNodeToHash(struct SalvageQueueNode * node);
111 static afs_int32 SALVSYNC_com_Salvage(SALVSYNC_command * com, SALVSYNC_response * res);
112 static afs_int32 SALVSYNC_com_Cancel(SALVSYNC_command * com, SALVSYNC_response * res);
113 static afs_int32 SALVSYNC_com_Query(SALVSYNC_command * com, SALVSYNC_response * res);
114 static afs_int32 SALVSYNC_com_CancelAll(SALVSYNC_command * com, SALVSYNC_response * res);
115 static afs_int32 SALVSYNC_com_Link(SALVSYNC_command * com, SALVSYNC_response * res);
119 extern pthread_mutex_t vol_salvsync_mutex;
122 * salvsync server socket handle.
124 static SYNC_server_state_t salvsync_server_state =
125 { OSI_NULLSOCKET, /* file descriptor */
126 SALVSYNC_ENDPOINT_DECL, /* server endpoint */
127 SALVSYNC_PROTO_VERSION, /* protocol version */
128 5, /* bind() retry limit */
129 100, /* listen() queue depth */
130 "SALVSYNC", /* protocol name string */
135 * queue of all volumes waiting to be salvaged.
137 struct SalvageQueue {
138 volatile int total_len;
139 volatile afs_int32 last_insert; /**< id of last partition to have a salvage node inserted */
140 volatile int len[VOLMAXPARTS+1];
141 volatile struct rx_queue part[VOLMAXPARTS+1]; /**< per-partition queues of pending salvages */
144 static struct SalvageQueue salvageQueue; /* volumes waiting to be salvaged */
147 * queue of all volumes currently being salvaged.
150 volatile struct rx_queue q; /**< queue of salvages in progress */
151 volatile int len; /**< length of in-progress queue */
152 pthread_cond_t queue_change_cv;
154 static struct QueueHead pendingQueue; /* volumes being salvaged */
157 * whether a partition has a salvage in progress
159 * the salvager code only permits one salvage per partition at a time
161 * the following hack tries to keep salvaged parallelism high by
162 * only permitting one salvage dispatch per partition at a time
164 * unfortunately, the parallel salvager currently
165 * has a rather braindead routine that won't permit
166 * multiple salvages on the same "device". this
167 * function happens to break pretty badly on lvm, raid luns, etc.
169 * this hack isn't good enough to stop the device limiting code from
170 * crippling performance. someday that code needs to be rewritten
172 static int partition_salvaging[VOLMAXPARTS+1];
174 static int HandlerFD[MAXHANDLERS];
175 static void (*HandlerProc[MAXHANDLERS]) (int);
177 #define VSHASH_SIZE 64
178 #define VSHASH_MASK (VSHASH_SIZE-1)
179 #define VSHASH(vid) ((vid)&VSHASH_MASK)
181 static struct QueueHead SalvageHashTable[VSHASH_SIZE];
183 static struct SalvageQueueNode *
184 LookupNode(VolumeId vid, char * partName,
185 struct SalvageQueueNode ** parent)
187 struct rx_queue *qp, *nqp;
188 struct SalvageQueueNode *vsp = NULL;
189 int idx = VSHASH(vid);
191 for (queue_Scan(&SalvageHashTable[idx], qp, nqp, rx_queue)) {
192 vsp = (struct SalvageQueueNode *)((char *)qp - offsetof(struct SalvageQueueNode, hash_chain));
193 if ((vsp->command.sop.volume == vid) &&
194 !strncmp(vsp->command.sop.partName, partName, sizeof(vsp->command.sop.partName))) {
199 if (queue_IsEnd(&SalvageHashTable[idx], qp)) {
205 *parent = (vsp->type == SALVSYNC_VOLGROUP_CLONE) ?
206 vsp->volgroup.parent : vsp;
215 static struct SalvageQueueNode *
216 LookupNodeByCommand(SALVSYNC_command_hdr * qry,
217 struct SalvageQueueNode ** parent)
219 return LookupNode(qry->volume, qry->partName, parent);
223 AddNodeToHash(struct SalvageQueueNode * node)
225 int idx = VSHASH(node->command.sop.volume);
227 if (queue_IsOnQueue(&node->hash_chain)) {
231 queue_Append(&SalvageHashTable[idx], &node->hash_chain);
232 SalvageHashTable[idx].len++;
237 DeleteNodeFromHash(struct SalvageQueueNode * node)
239 int idx = VSHASH(node->command.sop.volume);
241 if (queue_IsNotOnQueue(&node->hash_chain)) {
245 queue_Remove(&node->hash_chain);
246 SalvageHashTable[idx].len--;
251 SALVSYNC_salvInit(void)
255 pthread_attr_t tattr;
257 /* initialize the queues */
258 Lock_Init(&SALVSYNC_handler_lock);
259 CV_INIT(&salvageQueue.cv, "sq", CV_DEFAULT, 0);
260 for (i = 0; i <= VOLMAXPARTS; i++) {
261 queue_Init(&salvageQueue.part[i]);
262 salvageQueue.len[i] = 0;
264 CV_INIT(&pendingQueue.queue_change_cv, "queuechange", CV_DEFAULT, 0);
265 queue_Init(&pendingQueue);
266 salvageQueue.total_len = pendingQueue.len = 0;
267 salvageQueue.last_insert = -1;
268 memset(partition_salvaging, 0, sizeof(partition_salvaging));
270 for (i = 0; i < VSHASH_SIZE; i++) {
271 CV_INIT(&SalvageHashTable[i].queue_change_cv, "queuechange", CV_DEFAULT, 0);
272 SalvageHashTable[i].len = 0;
273 queue_Init(&SalvageHashTable[i]);
276 /* start the salvsync thread */
277 opr_Verify(pthread_attr_init(&tattr) == 0);
278 opr_Verify(pthread_attr_setdetachstate(&tattr,
279 PTHREAD_CREATE_DETACHED) == 0);
280 opr_Verify(pthread_create(&tid, &tattr, SALVSYNC_syncThread, NULL) == 0);
287 for (i = 0; i < MAXHANDLERS; ++i) {
288 if (HandlerFD[i] >= 0) {
289 SALVSYNC_Drop(HandlerFD[i]);
293 /* just in case we were in AcceptOff mode, and thus this fd wouldn't
295 close(salvsync_server_state.fd);
296 salvsync_server_state.fd = OSI_NULLSOCKET;
299 static fd_set SALVSYNC_readfds;
302 SALVSYNC_syncThread(void * args)
305 SYNC_server_state_t * state = &salvsync_server_state;
307 /* when we fork, the child needs to close the salvsync server sockets,
308 * otherwise, it may get salvsync requests, instead of the parent
310 opr_Verify(pthread_atfork(NULL, NULL, CleanFDs) == 0);
312 SYNC_getAddr(&state->endpoint, &state->addr);
313 SYNC_cleanupSock(state);
316 (void)signal(SIGPIPE, SIG_IGN);
319 state->fd = SYNC_getSock(&state->endpoint);
320 code = SYNC_bindSock(state);
328 struct timeval s_timeout;
329 GetHandler(&SALVSYNC_readfds, &maxfd);
330 s_timeout.tv_sec = SYNC_SELECT_TIMEOUT;
331 s_timeout.tv_usec = 0;
332 /* Note: check for >= 1 below is essential since IOMGR_select
333 * doesn't have exactly same semantics as select.
335 if (select(maxfd + 1, &SALVSYNC_readfds, NULL, NULL, &s_timeout) >= 1)
336 CallHandler(&SALVSYNC_readfds);
343 SALVSYNC_newconnection(int afd)
345 #ifdef USE_UNIX_SOCKETS
346 struct sockaddr_un other;
347 #else /* USE_UNIX_SOCKETS */
348 struct sockaddr_in other;
353 junk = sizeof(other);
354 fd = accept(afd, (struct sockaddr *)&other, &junk);
355 if (fd == OSI_NULLSOCKET) {
356 osi_Panic("SALVSYNC_newconnection: accept failed, errno==%d\n", errno);
357 } else if (!AddHandler(fd, SALVSYNC_com)) {
359 opr_Verify(AddHandler(fd, SALVSYNC_com));
363 /* this function processes commands from an salvsync file descriptor (fd) */
364 static afs_int32 SALV_cnt = 0;
366 SALVSYNC_com(osi_socket fd)
370 SALVSYNC_response_hdr sres_hdr;
371 SALVSYNC_command scom;
372 SALVSYNC_response sres;
373 SYNC_PROTO_BUF_DECL(buf);
375 memset(&com, 0, sizeof(com));
376 memset(&res, 0, sizeof(res));
377 memset(&scom, 0, sizeof(scom));
378 memset(&sres, 0, sizeof(sres));
379 memset(&sres_hdr, 0, sizeof(sres_hdr));
381 com.payload.buf = (void *)buf;
382 com.payload.len = SYNC_PROTO_MAX_LEN;
383 res.payload.buf = (void *) &sres_hdr;
384 res.payload.len = sizeof(sres_hdr);
385 res.hdr.response_len = sizeof(res.hdr) + sizeof(sres_hdr);
386 res.hdr.proto_version = SALVSYNC_PROTO_VERSION;
389 scom.sop = (SALVSYNC_command_hdr *) buf;
392 sres.sop = &sres_hdr;
396 if (SYNC_getCom(&salvsync_server_state, fd, &com)) {
397 Log("SALVSYNC_com: read failed; dropping connection (cnt=%d)\n", SALV_cnt);
402 if (com.recv_len < sizeof(com.hdr)) {
403 Log("SALVSYNC_com: invalid protocol message length (%u)\n", com.recv_len);
404 res.hdr.response = SYNC_COM_ERROR;
405 res.hdr.reason = SYNC_REASON_MALFORMED_PACKET;
406 res.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
410 if (com.hdr.proto_version != SALVSYNC_PROTO_VERSION) {
411 Log("SALVSYNC_com: invalid protocol version (%u)\n", com.hdr.proto_version);
412 res.hdr.response = SYNC_COM_ERROR;
413 res.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
417 if (com.hdr.command == SYNC_COM_CHANNEL_CLOSE) {
418 res.hdr.response = SYNC_OK;
419 res.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
421 /* don't respond, just drop; senders of SYNC_COM_CHANNEL_CLOSE
422 * never wait for a response. */
426 if (com.recv_len != (sizeof(com.hdr) + sizeof(SALVSYNC_command_hdr))) {
427 Log("SALVSYNC_com: invalid protocol message length (%u)\n", com.recv_len);
428 res.hdr.response = SYNC_COM_ERROR;
429 res.hdr.reason = SYNC_REASON_MALFORMED_PACKET;
430 res.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
434 res.hdr.com_seq = com.hdr.com_seq;
437 switch (com.hdr.command) {
440 case SALVSYNC_SALVAGE:
441 case SALVSYNC_RAISEPRIO:
442 res.hdr.response = SALVSYNC_com_Salvage(&scom, &sres);
444 case SALVSYNC_CANCEL:
445 /* cancel a salvage */
446 res.hdr.response = SALVSYNC_com_Cancel(&scom, &sres);
448 case SALVSYNC_CANCELALL:
449 /* cancel all queued salvages */
450 res.hdr.response = SALVSYNC_com_CancelAll(&scom, &sres);
453 /* query whether a volume is done salvaging */
454 res.hdr.response = SALVSYNC_com_Query(&scom, &sres);
456 case SALVSYNC_OP_LINK:
457 /* link a clone to its parent in the scheduler */
458 res.hdr.response = SALVSYNC_com_Link(&scom, &sres);
461 res.hdr.response = SYNC_BAD_COMMAND;
465 sres_hdr.sq_len = salvageQueue.total_len;
466 sres_hdr.pq_len = pendingQueue.len;
470 SYNC_putRes(&salvsync_server_state, fd, &res);
473 if (res.hdr.flags & SYNC_FLAG_CHANNEL_SHUTDOWN) {
479 * request that a volume be salvaged.
481 * @param[in] com inbound command object
482 * @param[out] res outbound response object
484 * @return operation status
485 * @retval SYNC_OK success
486 * @retval SYNC_DENIED failed to enqueue request
487 * @retval SYNC_FAILED malformed command packet
489 * @note this is a SALVSYNC protocol rpc handler
493 * @post the volume is enqueued in the to-be-salvaged queue.
494 * if the volume was already in the salvage queue, its
495 * priority (and thus its location in the queue) are
499 SALVSYNC_com_Salvage(SALVSYNC_command * com, SALVSYNC_response * res)
501 afs_int32 code = SYNC_OK;
502 struct SalvageQueueNode * node, * clone;
505 if (SYNC_verifyProtocolString(com->sop->partName, sizeof(com->sop->partName))) {
507 res->hdr->reason = SYNC_REASON_MALFORMED_PACKET;
511 clone = LookupNodeByCommand(com->sop, &node);
514 if (AllocNode(&node)) {
516 res->hdr->reason = SYNC_REASON_NOMEM;
523 HandlePrio(clone, node, com->sop->prio);
525 switch (node->state) {
526 case SALVSYNC_STATE_QUEUED:
527 UpdateCommandPrio(node);
530 case SALVSYNC_STATE_ERROR:
531 case SALVSYNC_STATE_DONE:
532 case SALVSYNC_STATE_UNKNOWN:
533 memcpy(&clone->command.com, com->hdr, sizeof(SYNC_command_hdr));
534 memcpy(&clone->command.sop, com->sop, sizeof(SALVSYNC_command_hdr));
537 * make sure volgroup parent partition path is kept coherent
539 * If we ever want to support non-COW clones on a machine holding
540 * the RW site, please note that this code does not work under the
541 * conditions where someone zaps a COW clone on partition X, and
542 * subsequently creates a full clone on partition Y -- we'd need
543 * an inverse to SALVSYNC_com_Link.
544 * -- tkeiser 11/28/2007
546 strcpy(node->command.sop.partName, com->sop->partName);
548 if (AddToSalvageQueue(node)) {
561 res->hdr->flags |= SALVSYNC_FLAG_VOL_STATS_VALID;
562 res->sop->state = node->state;
563 res->sop->prio = node->command.sop.prio;
570 * cancel a pending salvage request.
572 * @param[in] com inbound command object
573 * @param[out] res outbound response object
575 * @return operation status
576 * @retval SYNC_OK success
577 * @retval SYNC_FAILED malformed command packet
579 * @note this is a SALVSYNC protocol rpc handler
584 SALVSYNC_com_Cancel(SALVSYNC_command * com, SALVSYNC_response * res)
586 afs_int32 code = SYNC_OK;
587 struct SalvageQueueNode * node;
589 if (SYNC_verifyProtocolString(com->sop->partName, sizeof(com->sop->partName))) {
591 res->hdr->reason = SYNC_REASON_MALFORMED_PACKET;
595 node = LookupNodeByCommand(com->sop, NULL);
598 res->sop->state = SALVSYNC_STATE_UNKNOWN;
601 res->hdr->flags |= SALVSYNC_FLAG_VOL_STATS_VALID;
602 res->sop->prio = node->command.sop.prio;
603 res->sop->state = node->state;
604 if ((node->type == SALVSYNC_VOLGROUP_PARENT) &&
605 (node->state == SALVSYNC_STATE_QUEUED)) {
606 DeleteFromSalvageQueue(node);
615 * cancel all pending salvage requests.
617 * @param[in] com incoming command object
618 * @param[out] res outbound response object
620 * @return operation status
621 * @retval SYNC_OK success
623 * @note this is a SALVSYNC protocol rpc handler
628 SALVSYNC_com_CancelAll(SALVSYNC_command * com, SALVSYNC_response * res)
630 struct SalvageQueueNode * np, *nnp;
631 struct DiskPartition64 * dp;
633 for (dp = DiskPartitionList ; dp ; dp = dp->next) {
634 for (queue_Scan(&salvageQueue.part[dp->index], np, nnp, SalvageQueueNode)) {
635 DeleteFromSalvageQueue(np);
643 * link a queue node for a clone to its parent volume.
645 * @param[in] com inbound command object
646 * @param[out] res outbound response object
648 * @return operation status
649 * @retval SYNC_OK success
650 * @retval SYNC_FAILED malformed command packet
651 * @retval SYNC_DENIED the request could not be completed
653 * @note this is a SALVSYNC protocol rpc handler
655 * @post the requested volume is marked as a child of another volume.
656 * thus, future salvage requests for this volume will result in the
657 * parent of the volume group being scheduled for salvage instead
663 SALVSYNC_com_Link(SALVSYNC_command * com, SALVSYNC_response * res)
665 afs_int32 code = SYNC_OK;
666 struct SalvageQueueNode * clone, * parent;
668 if (SYNC_verifyProtocolString(com->sop->partName, sizeof(com->sop->partName))) {
670 res->hdr->reason = SYNC_REASON_MALFORMED_PACKET;
674 /* lookup clone's salvage scheduling node */
675 clone = LookupNodeByCommand(com->sop, NULL);
678 res->hdr->reason = SALVSYNC_REASON_ERROR;
682 /* lookup parent's salvage scheduling node */
683 parent = LookupNode(com->sop->parent, com->sop->partName, NULL);
684 if (parent == NULL) {
685 if (AllocNode(&parent)) {
687 res->hdr->reason = SYNC_REASON_NOMEM;
690 memcpy(&parent->command.com, com->hdr, sizeof(SYNC_command_hdr));
691 memcpy(&parent->command.sop, com->sop, sizeof(SALVSYNC_command_hdr));
692 parent->command.sop.volume = parent->command.sop.parent = com->sop->parent;
693 AddNodeToHash(parent);
696 if (LinkNode(parent, clone)) {
706 * query the status of a volume salvage request.
708 * @param[in] com inbound command object
709 * @param[out] res outbound response object
711 * @return operation status
712 * @retval SYNC_OK success
713 * @retval SYNC_FAILED malformed command packet
715 * @note this is a SALVSYNC protocol rpc handler
720 SALVSYNC_com_Query(SALVSYNC_command * com, SALVSYNC_response * res)
722 afs_int32 code = SYNC_OK;
723 struct SalvageQueueNode * node;
725 if (SYNC_verifyProtocolString(com->sop->partName, sizeof(com->sop->partName))) {
727 res->hdr->reason = SYNC_REASON_MALFORMED_PACKET;
731 LookupNodeByCommand(com->sop, &node);
733 /* query whether a volume is done salvaging */
735 res->sop->state = SALVSYNC_STATE_UNKNOWN;
738 res->hdr->flags |= SALVSYNC_FLAG_VOL_STATS_VALID;
739 res->sop->state = node->state;
740 res->sop->prio = node->command.sop.prio;
748 SALVSYNC_Drop(osi_socket fd)
755 static int AcceptHandler = -1; /* handler id for accept, if turned on */
760 if (AcceptHandler == -1) {
761 opr_Verify(AddHandler(salvsync_server_state.fd,
762 SALVSYNC_newconnection));
763 AcceptHandler = FindHandler(salvsync_server_state.fd);
770 if (AcceptHandler != -1) {
771 opr_Verify(RemoveHandler(salvsync_server_state.fd));
776 /* The multiple FD handling code. */
782 ObtainWriteLock(&SALVSYNC_handler_lock);
783 for (i = 0; i < MAXHANDLERS; i++) {
784 HandlerFD[i] = OSI_NULLSOCKET;
785 HandlerProc[i] = NULL;
787 ReleaseWriteLock(&SALVSYNC_handler_lock);
791 CallHandler(fd_set * fdsetp)
794 ObtainReadLock(&SALVSYNC_handler_lock);
795 for (i = 0; i < MAXHANDLERS; i++) {
796 if (HandlerFD[i] >= 0 && FD_ISSET(HandlerFD[i], fdsetp)) {
797 ReleaseReadLock(&SALVSYNC_handler_lock);
798 (*HandlerProc[i]) (HandlerFD[i]);
799 ObtainReadLock(&SALVSYNC_handler_lock);
802 ReleaseReadLock(&SALVSYNC_handler_lock);
806 AddHandler(osi_socket afd, void (*aproc) (int))
809 ObtainWriteLock(&SALVSYNC_handler_lock);
810 for (i = 0; i < MAXHANDLERS; i++)
811 if (HandlerFD[i] == OSI_NULLSOCKET)
813 if (i >= MAXHANDLERS) {
814 ReleaseWriteLock(&SALVSYNC_handler_lock);
818 HandlerProc[i] = aproc;
819 ReleaseWriteLock(&SALVSYNC_handler_lock);
824 FindHandler(osi_socket afd)
827 ObtainReadLock(&SALVSYNC_handler_lock);
828 for (i = 0; i < MAXHANDLERS; i++)
829 if (HandlerFD[i] == afd) {
830 ReleaseReadLock(&SALVSYNC_handler_lock);
833 ReleaseReadLock(&SALVSYNC_handler_lock); /* just in case */
834 osi_Panic("Failed to find handler\n");
835 return -1; /* satisfy compiler */
839 FindHandler_r(osi_socket afd)
842 for (i = 0; i < MAXHANDLERS; i++)
843 if (HandlerFD[i] == afd) {
846 osi_Panic("Failed to find handler\n");
847 return -1; /* satisfy compiler */
851 RemoveHandler(osi_socket afd)
853 ObtainWriteLock(&SALVSYNC_handler_lock);
854 HandlerFD[FindHandler_r(afd)] = OSI_NULLSOCKET;
855 ReleaseWriteLock(&SALVSYNC_handler_lock);
860 GetHandler(fd_set * fdsetp, int *maxfdp)
865 ObtainReadLock(&SALVSYNC_handler_lock); /* just in case */
866 for (i = 0; i < MAXHANDLERS; i++)
867 if (HandlerFD[i] != OSI_NULLSOCKET) {
868 FD_SET(HandlerFD[i], fdsetp);
870 /* On Windows the nfds parameter to select() is ignored */
871 if (maxfd < HandlerFD[i] || maxfd == (int)-1)
872 maxfd = HandlerFD[i];
876 ReleaseReadLock(&SALVSYNC_handler_lock); /* just in case */
880 * allocate a salvage queue node.
882 * @param[out] node_out address in which to store new node pointer
884 * @return operation status
886 * @retval 1 failed to allocate node
891 AllocNode(struct SalvageQueueNode ** node_out)
894 struct SalvageQueueNode * node;
896 *node_out = node = calloc(1, sizeof(struct SalvageQueueNode));
902 node->type = SALVSYNC_VOLGROUP_PARENT;
903 node->state = SALVSYNC_STATE_UNKNOWN;
910 * link a salvage queue node to its parent.
912 * @param[in] parent pointer to queue node for parent of volume group
913 * @param[in] clone pointer to queue node for a clone
915 * @return operation status
922 LinkNode(struct SalvageQueueNode * parent,
923 struct SalvageQueueNode * clone)
928 /* check for attaching a clone to a clone */
929 if (parent->type != SALVSYNC_VOLGROUP_PARENT) {
934 /* check for pre-existing registration and openings */
935 for (idx = 0; idx < VOLMAXTYPES; idx++) {
936 if (parent->volgroup.children[idx] == clone) {
939 if (parent->volgroup.children[idx] == NULL) {
943 if (idx == VOLMAXTYPES) {
948 /* link parent and child */
949 parent->volgroup.children[idx] = clone;
950 clone->type = SALVSYNC_VOLGROUP_CLONE;
951 clone->volgroup.parent = parent;
955 switch (clone->state) {
956 case SALVSYNC_STATE_QUEUED:
957 DeleteFromSalvageQueue(clone);
959 case SALVSYNC_STATE_SALVAGING:
960 switch (parent->state) {
961 case SALVSYNC_STATE_UNKNOWN:
962 case SALVSYNC_STATE_ERROR:
963 case SALVSYNC_STATE_DONE:
964 parent->command.sop.prio = clone->command.sop.prio;
965 AddToSalvageQueue(parent);
968 case SALVSYNC_STATE_QUEUED:
969 if (clone->command.sop.prio) {
970 parent->command.sop.prio += clone->command.sop.prio;
971 UpdateCommandPrio(parent);
989 HandlePrio(struct SalvageQueueNode * clone,
990 struct SalvageQueueNode * node,
995 switch (node->state) {
996 case SALVSYNC_STATE_ERROR:
997 case SALVSYNC_STATE_DONE:
998 case SALVSYNC_STATE_UNKNOWN:
999 node->command.sop.prio = 0;
1005 if (new_prio < clone->command.sop.prio) {
1006 /* strange. let's just set our delta to 1 */
1009 delta = new_prio - clone->command.sop.prio;
1012 if (clone->type == SALVSYNC_VOLGROUP_CLONE) {
1013 clone->command.sop.prio = new_prio;
1016 node->command.sop.prio += delta;
1020 AddToSalvageQueue(struct SalvageQueueNode * node)
1023 struct SalvageQueueNode * last = NULL;
1025 id = volutil_GetPartitionID(node->command.sop.partName);
1026 if (id < 0 || id > VOLMAXPARTS) {
1029 if (!VGetPartitionById_r(id, 0)) {
1030 /* don't enqueue salvage requests for unmounted partitions */
1033 if (queue_IsOnQueue(node)) {
1037 if (queue_IsNotEmpty(&salvageQueue.part[id])) {
1038 last = queue_Last(&salvageQueue.part[id], SalvageQueueNode);
1040 queue_Append(&salvageQueue.part[id], node);
1041 salvageQueue.len[id]++;
1042 salvageQueue.total_len++;
1043 salvageQueue.last_insert = id;
1044 node->partition_id = id;
1045 node->state = SALVSYNC_STATE_QUEUED;
1047 /* reorder, if necessary */
1048 if (last && last->command.sop.prio < node->command.sop.prio) {
1049 UpdateCommandPrio(node);
1052 CV_BROADCAST(&salvageQueue.cv);
1057 DeleteFromSalvageQueue(struct SalvageQueueNode * node)
1059 if (queue_IsOnQueue(node)) {
1061 salvageQueue.len[node->partition_id]--;
1062 salvageQueue.total_len--;
1063 node->state = SALVSYNC_STATE_UNKNOWN;
1064 CV_BROADCAST(&salvageQueue.cv);
1069 AddToPendingQueue(struct SalvageQueueNode * node)
1071 queue_Append(&pendingQueue, node);
1073 node->state = SALVSYNC_STATE_SALVAGING;
1074 CV_BROADCAST(&pendingQueue.queue_change_cv);
1078 DeleteFromPendingQueue(struct SalvageQueueNode * node)
1080 if (queue_IsOnQueue(node)) {
1083 node->state = SALVSYNC_STATE_UNKNOWN;
1084 CV_BROADCAST(&pendingQueue.queue_change_cv);
1089 static struct SalvageQueueNode *
1090 LookupPendingCommand(SALVSYNC_command_hdr * qry)
1092 struct SalvageQueueNode * np, * nnp;
1094 for (queue_Scan(&pendingQueue, np, nnp, SalvageQueueNode)) {
1095 if ((np->command.sop.volume == qry->volume) &&
1096 !strncmp(np->command.sop.partName, qry->partName,
1097 sizeof(qry->partName)))
1101 if (queue_IsEnd(&pendingQueue, np))
1107 static struct SalvageQueueNode *
1108 LookupPendingCommandByPid(int pid)
1110 struct SalvageQueueNode * np, * nnp;
1112 for (queue_Scan(&pendingQueue, np, nnp, SalvageQueueNode)) {
1117 if (queue_IsEnd(&pendingQueue, np))
1123 /* raise the priority of a previously scheduled salvage */
1125 UpdateCommandPrio(struct SalvageQueueNode * node)
1127 struct SalvageQueueNode *np, *nnp;
1131 opr_Assert(queue_IsOnQueue(node));
1133 prio = node->command.sop.prio;
1134 id = node->partition_id;
1135 if (queue_First(&salvageQueue.part[id], SalvageQueueNode)->command.sop.prio < prio) {
1137 queue_Prepend(&salvageQueue.part[id], node);
1139 for (queue_ScanBackwardsFrom(&salvageQueue.part[id], node, np, nnp, SalvageQueueNode)) {
1140 if (np->command.sop.prio > prio)
1143 if (queue_IsEnd(&salvageQueue.part[id], np)) {
1145 queue_Prepend(&salvageQueue.part[id], node);
1146 } else if (node != np) {
1148 queue_InsertAfter(np, node);
1153 /* this will need to be rearchitected if we ever want more than one thread
1154 * to wait for new salvage nodes */
1155 struct SalvageQueueNode *
1156 SALVSYNC_getWork(void)
1159 struct DiskPartition64 * dp = NULL, * fdp;
1160 static afs_int32 next_part_sched = 0;
1161 struct SalvageQueueNode *node = NULL;
1166 * wait for work to be scheduled
1167 * if there are no disk partitions, just sit in this wait loop forever
1169 while (!salvageQueue.total_len || !DiskPartitionList) {
1170 VOL_CV_WAIT(&salvageQueue.cv);
1174 * short circuit for simple case where only one partition has
1175 * scheduled salvages
1177 if (salvageQueue.last_insert >= 0 && salvageQueue.last_insert <= VOLMAXPARTS &&
1178 (salvageQueue.total_len == salvageQueue.len[salvageQueue.last_insert])) {
1179 node = queue_First(&salvageQueue.part[salvageQueue.last_insert], SalvageQueueNode);
1185 * ok, more than one partition has scheduled salvages.
1186 * now search for partitions with scheduled salvages, but no pending salvages.
1188 dp = VGetPartitionById_r(next_part_sched, 0);
1190 dp = DiskPartitionList;
1196 dp = (dp->next) ? dp->next : DiskPartitionList, i++ ) {
1197 if (!partition_salvaging[dp->index] && salvageQueue.len[dp->index]) {
1198 node = queue_First(&salvageQueue.part[dp->index], SalvageQueueNode);
1205 * all partitions with scheduled salvages have at least one pending.
1206 * now do an exhaustive search for a scheduled salvage.
1212 dp = (dp->next) ? dp->next : DiskPartitionList, i++ ) {
1213 if (salvageQueue.len[dp->index]) {
1214 node = queue_First(&salvageQueue.part[dp->index], SalvageQueueNode);
1219 /* we should never reach this line */
1220 osi_Panic("Node not found\n");
1223 opr_Assert(node != NULL);
1225 partition_salvaging[node->partition_id]++;
1226 DeleteFromSalvageQueue(node);
1227 AddToPendingQueue(node);
1230 /* update next_part_sched field */
1232 next_part_sched = dp->next->index;
1233 } else if (DiskPartitionList) {
1234 next_part_sched = DiskPartitionList->index;
1236 next_part_sched = -1;
1245 * update internal scheduler state to reflect completion of a work unit.
1247 * @param[in] node salvage queue node object pointer
1248 * @param[in] result worker process result code
1250 * @post scheduler state is updated.
1255 SALVSYNC_doneWork_r(struct SalvageQueueNode * node, int result)
1260 DeleteFromPendingQueue(node);
1261 partid = node->partition_id;
1262 if (partid >=0 && partid <= VOLMAXPARTS) {
1263 partition_salvaging[partid]--;
1266 node->state = SALVSYNC_STATE_DONE;
1267 } else if (result != SALSRV_EXIT_VOLGROUP_LINK) {
1268 node->state = SALVSYNC_STATE_ERROR;
1271 if (node->type == SALVSYNC_VOLGROUP_PARENT) {
1272 for (idx = 0; idx < VOLMAXTYPES; idx++) {
1273 if (node->volgroup.children[idx]) {
1274 node->volgroup.children[idx]->state = node->state;
1281 * check whether worker child failed.
1283 * @param[in] status status bitfield return by wait()
1285 * @return boolean failure code
1286 * @retval 0 child succeeded
1287 * @retval 1 child failed
1292 ChildFailed(int status)
1294 return (WCOREDUMP(status) ||
1295 WIFSIGNALED(status) ||
1296 ((WEXITSTATUS(status) != 0) &&
1297 (WEXITSTATUS(status) != SALSRV_EXIT_VOLGROUP_LINK)));
1302 * notify salvsync scheduler of node completion, by child pid.
1304 * @param[in] pid pid of worker child
1305 * @param[in] status worker status bitfield from wait()
1307 * @post scheduler state is updated.
1308 * if status code is a failure, fileserver notification was attempted
1310 * @see SALVSYNC_doneWork_r
1313 SALVSYNC_doneWorkByPid(int pid, int status)
1315 struct SalvageQueueNode * node;
1317 VolumeId volids[VOLMAXTYPES+1];
1320 memset(volids, 0, sizeof(volids));
1323 node = LookupPendingCommandByPid(pid);
1325 SALVSYNC_doneWork_r(node, status);
1327 if (ChildFailed(status)) {
1328 /* populate volume id list for later processing outside the glock */
1329 volids[0] = node->command.sop.volume;
1330 strcpy(partName, node->command.sop.partName);
1331 if (node->type == SALVSYNC_VOLGROUP_PARENT) {
1332 for (idx = 0; idx < VOLMAXTYPES; idx++) {
1333 if (node->volgroup.children[idx]) {
1334 volids[idx+1] = node->volgroup.children[idx]->command.sop.volume;
1343 * if necessary, notify fileserver of
1344 * failure to salvage volume group
1345 * [we cannot guarantee that the child made the
1346 * appropriate notifications (e.g. SIGSEGV)]
1347 * -- tkeiser 11/28/2007
1349 if (ChildFailed(status)) {
1350 for (idx = 0; idx <= VOLMAXTYPES; idx++) {
1352 FSYNC_VolOp(volids[idx],
1354 FSYNC_VOL_FORCE_ERROR,
1362 #endif /* AFS_DEMAND_ATTACH_FS */