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>
30 #include <sys/types.h>
36 #include <sys/param.h>
37 #include <sys/socket.h>
38 #include <netinet/in.h>
49 #include <afs/afsint.h>
51 #include <afs/errors.h>
55 #include <afs/afssyscalls.h>
59 #include "partition.h"
61 #include <rx/rx_queue.h>
62 #include <afs/procmgmt.h>
64 #if !defined(offsetof)
68 #ifdef USE_UNIX_SOCKETS
69 #include <afs/afsutil.h>
74 #define WCOREDUMP(x) ((x) & 0200)
77 #define MAXHANDLERS 4 /* Up to 4 clients; must be at least 2, so that
78 * move = dump+restore can run on single server */
82 * This lock controls access to the handler array.
84 struct Lock SALVSYNC_handler_lock;
87 #ifdef AFS_DEMAND_ATTACH_FS
89 * SALVSYNC is a feature specific to the demand attach fileserver
92 /* Forward declarations */
93 static void * SALVSYNC_syncThread(void *);
94 static void SALVSYNC_newconnection(osi_socket fd);
95 static void SALVSYNC_com(osi_socket fd);
96 static void SALVSYNC_Drop(osi_socket fd);
97 static void AcceptOn(void);
98 static void AcceptOff(void);
99 static void InitHandler(void);
100 static void CallHandler(fd_set * fdsetp);
101 static int AddHandler(osi_socket afd, void (*aproc) (int));
102 static int FindHandler(osi_socket afd);
103 static int FindHandler_r(osi_socket afd);
104 static int RemoveHandler(osi_socket afd);
105 static void GetHandler(fd_set * fdsetp, int *maxfdp);
107 static int AllocNode(struct SalvageQueueNode ** node);
109 static int AddToSalvageQueue(struct SalvageQueueNode * node);
110 static void DeleteFromSalvageQueue(struct SalvageQueueNode * node);
111 static void AddToPendingQueue(struct SalvageQueueNode * node);
112 static void DeleteFromPendingQueue(struct SalvageQueueNode * node);
113 static struct SalvageQueueNode * LookupPendingCommandByPid(int pid);
114 static void UpdateCommandPrio(struct SalvageQueueNode * node);
115 static void HandlePrio(struct SalvageQueueNode * clone,
116 struct SalvageQueueNode * parent,
117 afs_uint32 new_prio);
119 static int LinkNode(struct SalvageQueueNode * parent,
120 struct SalvageQueueNode * clone);
122 static struct SalvageQueueNode * LookupNode(VolumeId vid, char * partName,
123 struct SalvageQueueNode ** parent);
124 static struct SalvageQueueNode * LookupNodeByCommand(SALVSYNC_command_hdr * qry,
125 struct SalvageQueueNode ** parent);
126 static void AddNodeToHash(struct SalvageQueueNode * node);
128 static afs_int32 SALVSYNC_com_Salvage(SALVSYNC_command * com, SALVSYNC_response * res);
129 static afs_int32 SALVSYNC_com_Cancel(SALVSYNC_command * com, SALVSYNC_response * res);
130 static afs_int32 SALVSYNC_com_Query(SALVSYNC_command * com, SALVSYNC_response * res);
131 static afs_int32 SALVSYNC_com_CancelAll(SALVSYNC_command * com, SALVSYNC_response * res);
132 static afs_int32 SALVSYNC_com_Link(SALVSYNC_command * com, SALVSYNC_response * res);
137 extern pthread_mutex_t vol_salvsync_mutex;
140 * salvsync server socket handle.
142 static SYNC_server_state_t salvsync_server_state =
143 { -1, /* file descriptor */
144 SALVSYNC_ENDPOINT_DECL, /* server endpoint */
145 SALVSYNC_PROTO_VERSION, /* protocol version */
146 5, /* bind() retry limit */
147 100, /* listen() queue depth */
148 "SALVSYNC", /* protocol name string */
153 * queue of all volumes waiting to be salvaged.
155 struct SalvageQueue {
156 volatile int total_len;
157 volatile afs_int32 last_insert; /**< id of last partition to have a salvage node inserted */
158 volatile int len[VOLMAXPARTS+1];
159 volatile struct rx_queue part[VOLMAXPARTS+1]; /**< per-partition queues of pending salvages */
162 static struct SalvageQueue salvageQueue; /* volumes waiting to be salvaged */
165 * queue of all volumes currently being salvaged.
168 volatile struct rx_queue q; /**< queue of salvages in progress */
169 volatile int len; /**< length of in-progress queue */
170 pthread_cond_t queue_change_cv;
172 static struct QueueHead pendingQueue; /* volumes being salvaged */
175 * whether a partition has a salvage in progress
177 * the salvager code only permits one salvage per partition at a time
179 * the following hack tries to keep salvaged parallelism high by
180 * only permitting one salvage dispatch per partition at a time
182 * unfortunately, the parallel salvager currently
183 * has a rather braindead routine that won't permit
184 * multiple salvages on the same "device". this
185 * function happens to break pretty badly on lvm, raid luns, etc.
187 * this hack isn't good enough to stop the device limiting code from
188 * crippling performance. someday that code needs to be rewritten
190 static int partition_salvaging[VOLMAXPARTS+1];
192 static int HandlerFD[MAXHANDLERS];
193 static void (*HandlerProc[MAXHANDLERS]) (int);
195 #define VSHASH_SIZE 64
196 #define VSHASH_MASK (VSHASH_SIZE-1)
197 #define VSHASH(vid) ((vid)&VSHASH_MASK)
199 static struct QueueHead SalvageHashTable[VSHASH_SIZE];
201 static struct SalvageQueueNode *
202 LookupNode(afs_uint32 vid, char * partName,
203 struct SalvageQueueNode ** parent)
205 struct rx_queue *qp, *nqp;
206 struct SalvageQueueNode *vsp;
207 int idx = VSHASH(vid);
209 for (queue_Scan(&SalvageHashTable[idx], qp, nqp, rx_queue)) {
210 vsp = (struct SalvageQueueNode *)((char *)qp - offsetof(struct SalvageQueueNode, hash_chain));
211 if ((vsp->command.sop.volume == vid) &&
212 !strncmp(vsp->command.sop.partName, partName, sizeof(vsp->command.sop.partName))) {
217 if (queue_IsEnd(&SalvageHashTable[idx], qp)) {
223 *parent = (vsp->type == SALVSYNC_VOLGROUP_CLONE) ?
224 vsp->volgroup.parent : vsp;
233 static struct SalvageQueueNode *
234 LookupNodeByCommand(SALVSYNC_command_hdr * qry,
235 struct SalvageQueueNode ** parent)
237 return LookupNode(qry->volume, qry->partName, parent);
241 AddNodeToHash(struct SalvageQueueNode * node)
243 int idx = VSHASH(node->command.sop.volume);
245 if (queue_IsOnQueue(&node->hash_chain)) {
249 queue_Append(&SalvageHashTable[idx], &node->hash_chain);
250 SalvageHashTable[idx].len++;
255 DeleteNodeFromHash(struct SalvageQueueNode * node)
257 int idx = VSHASH(node->command.sop.volume);
259 if (queue_IsNotOnQueue(&node->hash_chain)) {
263 queue_Remove(&node->hash_chain);
264 SalvageHashTable[idx].len--;
269 SALVSYNC_salvInit(void)
273 pthread_attr_t tattr;
275 /* initialize the queues */
276 Lock_Init(&SALVSYNC_handler_lock);
277 assert(pthread_cond_init(&salvageQueue.cv, NULL) == 0);
278 for (i = 0; i <= VOLMAXPARTS; i++) {
279 queue_Init(&salvageQueue.part[i]);
280 salvageQueue.len[i] = 0;
282 assert(pthread_cond_init(&pendingQueue.queue_change_cv, NULL) == 0);
283 queue_Init(&pendingQueue);
284 salvageQueue.total_len = pendingQueue.len = 0;
285 salvageQueue.last_insert = -1;
286 memset(partition_salvaging, 0, sizeof(partition_salvaging));
288 for (i = 0; i < VSHASH_SIZE; i++) {
289 assert(pthread_cond_init(&SalvageHashTable[i].queue_change_cv, NULL) == 0);
290 SalvageHashTable[i].len = 0;
291 queue_Init(&SalvageHashTable[i]);
294 /* start the salvsync thread */
295 assert(pthread_attr_init(&tattr) == 0);
296 assert(pthread_attr_setdetachstate(&tattr, PTHREAD_CREATE_DETACHED) == 0);
297 assert(pthread_create(&tid, &tattr, SALVSYNC_syncThread, NULL) == 0);
304 for (i = 0; i < MAXHANDLERS; ++i) {
305 if (HandlerFD[i] >= 0) {
306 SALVSYNC_Drop(HandlerFD[i]);
310 /* just in case we were in AcceptOff mode, and thus this fd wouldn't
312 close(salvsync_server_state.fd);
313 salvsync_server_state.fd = -1;
316 static fd_set SALVSYNC_readfds;
319 SALVSYNC_syncThread(void * args)
322 SYNC_server_state_t * state = &salvsync_server_state;
324 /* when we fork, the child needs to close the salvsync server sockets,
325 * otherwise, it may get salvsync requests, instead of the parent
327 assert(pthread_atfork(NULL, NULL, CleanFDs) == 0);
329 SYNC_getAddr(&state->endpoint, &state->addr);
330 SYNC_cleanupSock(state);
333 (void)signal(SIGPIPE, SIG_IGN);
336 state->fd = SYNC_getSock(&state->endpoint);
337 code = SYNC_bindSock(state);
345 GetHandler(&SALVSYNC_readfds, &maxfd);
346 /* Note: check for >= 1 below is essential since IOMGR_select
347 * doesn't have exactly same semantics as select.
349 if (select(maxfd + 1, &SALVSYNC_readfds, NULL, NULL, NULL) >= 1)
350 CallHandler(&SALVSYNC_readfds);
357 SALVSYNC_newconnection(int afd)
359 #ifdef USE_UNIX_SOCKETS
360 struct sockaddr_un other;
361 #else /* USE_UNIX_SOCKETS */
362 struct sockaddr_in other;
367 junk = sizeof(other);
368 fd = accept(afd, (struct sockaddr *)&other, &junk);
370 Log("SALVSYNC_newconnection: accept failed, errno==%d\n", errno);
372 } else if (!AddHandler(fd, SALVSYNC_com)) {
374 assert(AddHandler(fd, SALVSYNC_com));
378 /* this function processes commands from an salvsync file descriptor (fd) */
379 static afs_int32 SALV_cnt = 0;
381 SALVSYNC_com(osi_socket fd)
385 SALVSYNC_response_hdr sres_hdr;
386 SALVSYNC_command scom;
387 SALVSYNC_response sres;
388 SYNC_PROTO_BUF_DECL(buf);
390 memset(&com, 0, sizeof(com));
391 memset(&res, 0, sizeof(res));
392 memset(&scom, 0, sizeof(scom));
393 memset(&sres, 0, sizeof(sres));
394 memset(&sres_hdr, 0, sizeof(sres_hdr));
396 com.payload.buf = (void *)buf;
397 com.payload.len = SYNC_PROTO_MAX_LEN;
398 res.payload.buf = (void *) &sres_hdr;
399 res.payload.len = sizeof(sres_hdr);
400 res.hdr.response_len = sizeof(res.hdr) + sizeof(sres_hdr);
401 res.hdr.proto_version = SALVSYNC_PROTO_VERSION;
404 scom.sop = (SALVSYNC_command_hdr *) buf;
407 sres.sop = &sres_hdr;
411 if (SYNC_getCom(&salvsync_server_state, fd, &com)) {
412 Log("SALVSYNC_com: read failed; dropping connection (cnt=%d)\n", SALV_cnt);
417 if (com.recv_len < sizeof(com.hdr)) {
418 Log("SALVSYNC_com: invalid protocol message length (%u)\n", com.recv_len);
419 res.hdr.response = SYNC_COM_ERROR;
420 res.hdr.reason = SYNC_REASON_MALFORMED_PACKET;
421 res.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
425 if (com.hdr.proto_version != SALVSYNC_PROTO_VERSION) {
426 Log("SALVSYNC_com: invalid protocol version (%u)\n", com.hdr.proto_version);
427 res.hdr.response = SYNC_COM_ERROR;
428 res.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
432 if (com.hdr.command == SYNC_COM_CHANNEL_CLOSE) {
433 res.hdr.response = SYNC_OK;
434 res.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
436 /* don't respond, just drop; senders of SYNC_COM_CHANNEL_CLOSE
437 * never wait for a response. */
441 if (com.recv_len != (sizeof(com.hdr) + sizeof(SALVSYNC_command_hdr))) {
442 Log("SALVSYNC_com: invalid protocol message length (%u)\n", com.recv_len);
443 res.hdr.response = SYNC_COM_ERROR;
444 res.hdr.reason = SYNC_REASON_MALFORMED_PACKET;
445 res.hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
449 res.hdr.com_seq = com.hdr.com_seq;
452 switch (com.hdr.command) {
455 case SALVSYNC_SALVAGE:
456 case SALVSYNC_RAISEPRIO:
457 res.hdr.response = SALVSYNC_com_Salvage(&scom, &sres);
459 case SALVSYNC_CANCEL:
460 /* cancel a salvage */
461 res.hdr.response = SALVSYNC_com_Cancel(&scom, &sres);
463 case SALVSYNC_CANCELALL:
464 /* cancel all queued salvages */
465 res.hdr.response = SALVSYNC_com_CancelAll(&scom, &sres);
468 /* query whether a volume is done salvaging */
469 res.hdr.response = SALVSYNC_com_Query(&scom, &sres);
471 case SALVSYNC_OP_LINK:
472 /* link a clone to its parent in the scheduler */
473 res.hdr.response = SALVSYNC_com_Link(&scom, &sres);
476 res.hdr.response = SYNC_BAD_COMMAND;
480 sres_hdr.sq_len = salvageQueue.total_len;
481 sres_hdr.pq_len = pendingQueue.len;
485 SYNC_putRes(&salvsync_server_state, fd, &res);
488 if (res.hdr.flags & SYNC_FLAG_CHANNEL_SHUTDOWN) {
494 * request that a volume be salvaged.
496 * @param[in] com inbound command object
497 * @param[out] res outbound response object
499 * @return operation status
500 * @retval SYNC_OK success
501 * @retval SYNC_DENIED failed to enqueue request
502 * @retval SYNC_FAILED malformed command packet
504 * @note this is a SALVSYNC protocol rpc handler
508 * @post the volume is enqueued in the to-be-salvaged queue.
509 * if the volume was already in the salvage queue, its
510 * priority (and thus its location in the queue) are
514 SALVSYNC_com_Salvage(SALVSYNC_command * com, SALVSYNC_response * res)
516 afs_int32 code = SYNC_OK;
517 struct SalvageQueueNode * node, * clone;
520 if (SYNC_verifyProtocolString(com->sop->partName, sizeof(com->sop->partName))) {
522 res->hdr->reason = SYNC_REASON_MALFORMED_PACKET;
526 clone = LookupNodeByCommand(com->sop, &node);
529 if (AllocNode(&node)) {
531 res->hdr->reason = SYNC_REASON_NOMEM;
538 HandlePrio(clone, node, com->sop->prio);
540 switch (node->state) {
541 case SALVSYNC_STATE_QUEUED:
542 UpdateCommandPrio(node);
545 case SALVSYNC_STATE_ERROR:
546 case SALVSYNC_STATE_DONE:
547 case SALVSYNC_STATE_UNKNOWN:
548 memcpy(&clone->command.com, com->hdr, sizeof(SYNC_command_hdr));
549 memcpy(&clone->command.sop, com->sop, sizeof(SALVSYNC_command_hdr));
552 * make sure volgroup parent partition path is kept coherent
554 * If we ever want to support non-COW clones on a machine holding
555 * the RW site, please note that this code does not work under the
556 * conditions where someone zaps a COW clone on partition X, and
557 * subsequently creates a full clone on partition Y -- we'd need
558 * an inverse to SALVSYNC_com_Link.
559 * -- tkeiser 11/28/2007
561 strcpy(node->command.sop.partName, com->sop->partName);
563 if (AddToSalvageQueue(node)) {
576 res->hdr->flags |= SALVSYNC_FLAG_VOL_STATS_VALID;
577 res->sop->state = node->state;
578 res->sop->prio = node->command.sop.prio;
585 * cancel a pending salvage request.
587 * @param[in] com inbound command object
588 * @param[out] res outbound response object
590 * @return operation status
591 * @retval SYNC_OK success
592 * @retval SYNC_FAILED malformed command packet
594 * @note this is a SALVSYNC protocol rpc handler
599 SALVSYNC_com_Cancel(SALVSYNC_command * com, SALVSYNC_response * res)
601 afs_int32 code = SYNC_OK;
602 struct SalvageQueueNode * node;
604 if (SYNC_verifyProtocolString(com->sop->partName, sizeof(com->sop->partName))) {
606 res->hdr->reason = SYNC_REASON_MALFORMED_PACKET;
610 node = LookupNodeByCommand(com->sop, NULL);
613 res->sop->state = SALVSYNC_STATE_UNKNOWN;
616 res->hdr->flags |= SALVSYNC_FLAG_VOL_STATS_VALID;
617 res->sop->prio = node->command.sop.prio;
618 res->sop->state = node->state;
619 if ((node->type == SALVSYNC_VOLGROUP_PARENT) &&
620 (node->state == SALVSYNC_STATE_QUEUED)) {
621 DeleteFromSalvageQueue(node);
630 * cancel all pending salvage requests.
632 * @param[in] com incoming command object
633 * @param[out] res outbound response object
635 * @return operation status
636 * @retval SYNC_OK success
638 * @note this is a SALVSYNC protocol rpc handler
643 SALVSYNC_com_CancelAll(SALVSYNC_command * com, SALVSYNC_response * res)
645 struct SalvageQueueNode * np, *nnp;
646 struct DiskPartition64 * dp;
648 for (dp = DiskPartitionList ; dp ; dp = dp->next) {
649 for (queue_Scan(&salvageQueue.part[dp->index], np, nnp, SalvageQueueNode)) {
650 DeleteFromSalvageQueue(np);
658 * link a queue node for a clone to its parent volume.
660 * @param[in] com inbound command object
661 * @param[out] res outbound response object
663 * @return operation status
664 * @retval SYNC_OK success
665 * @retval SYNC_FAILED malformed command packet
666 * @retval SYNC_DENIED the request could not be completed
668 * @note this is a SALVSYNC protocol rpc handler
670 * @post the requested volume is marked as a child of another volume.
671 * thus, future salvage requests for this volume will result in the
672 * parent of the volume group being scheduled for salvage instead
678 SALVSYNC_com_Link(SALVSYNC_command * com, SALVSYNC_response * res)
680 afs_int32 code = SYNC_OK;
681 struct SalvageQueueNode * clone, * parent;
683 if (SYNC_verifyProtocolString(com->sop->partName, sizeof(com->sop->partName))) {
685 res->hdr->reason = SYNC_REASON_MALFORMED_PACKET;
689 /* lookup clone's salvage scheduling node */
690 clone = LookupNodeByCommand(com->sop, NULL);
693 res->hdr->reason = SALVSYNC_REASON_ERROR;
697 /* lookup parent's salvage scheduling node */
698 parent = LookupNode(com->sop->parent, com->sop->partName, NULL);
699 if (parent == NULL) {
700 if (AllocNode(&parent)) {
702 res->hdr->reason = SYNC_REASON_NOMEM;
705 memcpy(&parent->command.com, com->hdr, sizeof(SYNC_command_hdr));
706 memcpy(&parent->command.sop, com->sop, sizeof(SALVSYNC_command_hdr));
707 parent->command.sop.volume = parent->command.sop.parent = com->sop->parent;
708 AddNodeToHash(parent);
711 if (LinkNode(parent, clone)) {
721 * query the status of a volume salvage request.
723 * @param[in] com inbound command object
724 * @param[out] res outbound response object
726 * @return operation status
727 * @retval SYNC_OK success
728 * @retval SYNC_FAILED malformed command packet
730 * @note this is a SALVSYNC protocol rpc handler
735 SALVSYNC_com_Query(SALVSYNC_command * com, SALVSYNC_response * res)
737 afs_int32 code = SYNC_OK;
738 struct SalvageQueueNode * node;
740 if (SYNC_verifyProtocolString(com->sop->partName, sizeof(com->sop->partName))) {
742 res->hdr->reason = SYNC_REASON_MALFORMED_PACKET;
746 LookupNodeByCommand(com->sop, &node);
748 /* query whether a volume is done salvaging */
750 res->sop->state = SALVSYNC_STATE_UNKNOWN;
753 res->hdr->flags |= SALVSYNC_FLAG_VOL_STATS_VALID;
754 res->sop->state = node->state;
755 res->sop->prio = node->command.sop.prio;
763 SALVSYNC_Drop(osi_socket fd)
774 static int AcceptHandler = -1; /* handler id for accept, if turned on */
779 if (AcceptHandler == -1) {
780 assert(AddHandler(salvsync_server_state.fd, SALVSYNC_newconnection));
781 AcceptHandler = FindHandler(salvsync_server_state.fd);
788 if (AcceptHandler != -1) {
789 assert(RemoveHandler(salvsync_server_state.fd));
794 /* The multiple FD handling code. */
800 ObtainWriteLock(&SALVSYNC_handler_lock);
801 for (i = 0; i < MAXHANDLERS; i++) {
803 HandlerProc[i] = NULL;
805 ReleaseWriteLock(&SALVSYNC_handler_lock);
809 CallHandler(fd_set * fdsetp)
812 ObtainReadLock(&SALVSYNC_handler_lock);
813 for (i = 0; i < MAXHANDLERS; i++) {
814 if (HandlerFD[i] >= 0 && FD_ISSET(HandlerFD[i], fdsetp)) {
815 ReleaseReadLock(&SALVSYNC_handler_lock);
816 (*HandlerProc[i]) (HandlerFD[i]);
817 ObtainReadLock(&SALVSYNC_handler_lock);
820 ReleaseReadLock(&SALVSYNC_handler_lock);
824 AddHandler(osi_socket afd, void (*aproc) (int))
827 ObtainWriteLock(&SALVSYNC_handler_lock);
828 for (i = 0; i < MAXHANDLERS; i++)
829 if (HandlerFD[i] == -1)
831 if (i >= MAXHANDLERS) {
832 ReleaseWriteLock(&SALVSYNC_handler_lock);
836 HandlerProc[i] = aproc;
837 ReleaseWriteLock(&SALVSYNC_handler_lock);
842 FindHandler(osi_socket afd)
845 ObtainReadLock(&SALVSYNC_handler_lock);
846 for (i = 0; i < MAXHANDLERS; i++)
847 if (HandlerFD[i] == afd) {
848 ReleaseReadLock(&SALVSYNC_handler_lock);
851 ReleaseReadLock(&SALVSYNC_handler_lock); /* just in case */
853 return -1; /* satisfy compiler */
857 FindHandler_r(osi_socket afd)
860 for (i = 0; i < MAXHANDLERS; i++)
861 if (HandlerFD[i] == afd) {
865 return -1; /* satisfy compiler */
869 RemoveHandler(osi_socket afd)
871 ObtainWriteLock(&SALVSYNC_handler_lock);
872 HandlerFD[FindHandler_r(afd)] = -1;
873 ReleaseWriteLock(&SALVSYNC_handler_lock);
878 GetHandler(fd_set * fdsetp, int *maxfdp)
883 ObtainReadLock(&SALVSYNC_handler_lock); /* just in case */
884 for (i = 0; i < MAXHANDLERS; i++)
885 if (HandlerFD[i] != -1) {
886 FD_SET(HandlerFD[i], fdsetp);
887 if (maxfd < HandlerFD[i])
888 maxfd = HandlerFD[i];
891 ReleaseReadLock(&SALVSYNC_handler_lock); /* just in case */
895 * allocate a salvage queue node.
897 * @param[out] node_out address in which to store new node pointer
899 * @return operation status
901 * @retval 1 failed to allocate node
906 AllocNode(struct SalvageQueueNode ** node_out)
909 struct SalvageQueueNode * node;
911 *node_out = node = (struct SalvageQueueNode *)
912 malloc(sizeof(struct SalvageQueueNode));
918 memset(node, 0, sizeof(struct SalvageQueueNode));
919 node->type = SALVSYNC_VOLGROUP_PARENT;
920 node->state = SALVSYNC_STATE_UNKNOWN;
927 * link a salvage queue node to its parent.
929 * @param[in] parent pointer to queue node for parent of volume group
930 * @param[in] clone pointer to queue node for a clone
932 * @return operation status
939 LinkNode(struct SalvageQueueNode * parent,
940 struct SalvageQueueNode * clone)
945 /* check for attaching a clone to a clone */
946 if (parent->type != SALVSYNC_VOLGROUP_PARENT) {
951 /* check for pre-existing registration and openings */
952 for (idx = 0; idx < VOLMAXTYPES; idx++) {
953 if (parent->volgroup.children[idx] == clone) {
956 if (parent->volgroup.children[idx] == NULL) {
960 if (idx == VOLMAXTYPES) {
965 /* link parent and child */
966 parent->volgroup.children[idx] = clone;
967 clone->type = SALVSYNC_VOLGROUP_CLONE;
968 clone->volgroup.parent = parent;
972 switch (clone->state) {
973 case SALVSYNC_STATE_QUEUED:
974 DeleteFromSalvageQueue(clone);
976 case SALVSYNC_STATE_SALVAGING:
977 switch (parent->state) {
978 case SALVSYNC_STATE_UNKNOWN:
979 case SALVSYNC_STATE_ERROR:
980 case SALVSYNC_STATE_DONE:
981 parent->command.sop.prio = clone->command.sop.prio;
982 AddToSalvageQueue(parent);
985 case SALVSYNC_STATE_QUEUED:
986 if (clone->command.sop.prio) {
987 parent->command.sop.prio += clone->command.sop.prio;
988 UpdateCommandPrio(parent);
1006 HandlePrio(struct SalvageQueueNode * clone,
1007 struct SalvageQueueNode * node,
1008 afs_uint32 new_prio)
1012 switch (node->state) {
1013 case SALVSYNC_STATE_ERROR:
1014 case SALVSYNC_STATE_DONE:
1015 case SALVSYNC_STATE_UNKNOWN:
1016 node->command.sop.prio = 0;
1022 if (new_prio < clone->command.sop.prio) {
1023 /* strange. let's just set our delta to 1 */
1026 delta = new_prio - clone->command.sop.prio;
1029 if (clone->type == SALVSYNC_VOLGROUP_CLONE) {
1030 clone->command.sop.prio = new_prio;
1033 node->command.sop.prio += delta;
1037 AddToSalvageQueue(struct SalvageQueueNode * node)
1040 struct SalvageQueueNode * last = NULL;
1042 id = volutil_GetPartitionID(node->command.sop.partName);
1043 if (id < 0 || id > VOLMAXPARTS) {
1046 if (!VGetPartitionById_r(id, 0)) {
1047 /* don't enqueue salvage requests for unmounted partitions */
1050 if (queue_IsOnQueue(node)) {
1054 if (queue_IsNotEmpty(&salvageQueue.part[id])) {
1055 last = queue_Last(&salvageQueue.part[id], SalvageQueueNode);
1057 queue_Append(&salvageQueue.part[id], node);
1058 salvageQueue.len[id]++;
1059 salvageQueue.total_len++;
1060 salvageQueue.last_insert = id;
1061 node->partition_id = id;
1062 node->state = SALVSYNC_STATE_QUEUED;
1064 /* reorder, if necessary */
1065 if (last && last->command.sop.prio < node->command.sop.prio) {
1066 UpdateCommandPrio(node);
1069 assert(pthread_cond_broadcast(&salvageQueue.cv) == 0);
1074 DeleteFromSalvageQueue(struct SalvageQueueNode * node)
1076 if (queue_IsOnQueue(node)) {
1078 salvageQueue.len[node->partition_id]--;
1079 salvageQueue.total_len--;
1080 node->state = SALVSYNC_STATE_UNKNOWN;
1081 assert(pthread_cond_broadcast(&salvageQueue.cv) == 0);
1086 AddToPendingQueue(struct SalvageQueueNode * node)
1088 queue_Append(&pendingQueue, node);
1090 node->state = SALVSYNC_STATE_SALVAGING;
1091 assert(pthread_cond_broadcast(&pendingQueue.queue_change_cv) == 0);
1095 DeleteFromPendingQueue(struct SalvageQueueNode * node)
1097 if (queue_IsOnQueue(node)) {
1100 node->state = SALVSYNC_STATE_UNKNOWN;
1101 assert(pthread_cond_broadcast(&pendingQueue.queue_change_cv) == 0);
1106 static struct SalvageQueueNode *
1107 LookupPendingCommand(SALVSYNC_command_hdr * qry)
1109 struct SalvageQueueNode * np, * nnp;
1111 for (queue_Scan(&pendingQueue, np, nnp, SalvageQueueNode)) {
1112 if ((np->command.sop.volume == qry->volume) &&
1113 !strncmp(np->command.sop.partName, qry->partName,
1114 sizeof(qry->partName)))
1118 if (queue_IsEnd(&pendingQueue, np))
1124 static struct SalvageQueueNode *
1125 LookupPendingCommandByPid(int pid)
1127 struct SalvageQueueNode * np, * nnp;
1129 for (queue_Scan(&pendingQueue, np, nnp, SalvageQueueNode)) {
1134 if (queue_IsEnd(&pendingQueue, np))
1140 /* raise the priority of a previously scheduled salvage */
1142 UpdateCommandPrio(struct SalvageQueueNode * node)
1144 struct SalvageQueueNode *np, *nnp;
1148 assert(queue_IsOnQueue(node));
1150 prio = node->command.sop.prio;
1151 id = node->partition_id;
1152 if (queue_First(&salvageQueue.part[id], SalvageQueueNode)->command.sop.prio < prio) {
1154 queue_Prepend(&salvageQueue.part[id], node);
1156 for (queue_ScanBackwardsFrom(&salvageQueue.part[id], node, np, nnp, SalvageQueueNode)) {
1157 if (np->command.sop.prio > prio)
1160 if (queue_IsEnd(&salvageQueue.part[id], np)) {
1162 queue_Prepend(&salvageQueue.part[id], node);
1163 } else if (node != np) {
1165 queue_InsertAfter(np, node);
1170 /* this will need to be rearchitected if we ever want more than one thread
1171 * to wait for new salvage nodes */
1172 struct SalvageQueueNode *
1173 SALVSYNC_getWork(void)
1176 struct DiskPartition64 * dp = NULL, * fdp;
1177 static afs_int32 next_part_sched = 0;
1178 struct SalvageQueueNode *node = NULL;
1183 * wait for work to be scheduled
1184 * if there are no disk partitions, just sit in this wait loop forever
1186 while (!salvageQueue.total_len || !DiskPartitionList) {
1187 VOL_CV_WAIT(&salvageQueue.cv);
1191 * short circuit for simple case where only one partition has
1192 * scheduled salvages
1194 if (salvageQueue.last_insert >= 0 && salvageQueue.last_insert <= VOLMAXPARTS &&
1195 (salvageQueue.total_len == salvageQueue.len[salvageQueue.last_insert])) {
1196 node = queue_First(&salvageQueue.part[salvageQueue.last_insert], SalvageQueueNode);
1202 * ok, more than one partition has scheduled salvages.
1203 * now search for partitions with scheduled salvages, but no pending salvages.
1205 dp = VGetPartitionById_r(next_part_sched, 0);
1207 dp = DiskPartitionList;
1213 dp = (dp->next) ? dp->next : DiskPartitionList, i++ ) {
1214 if (!partition_salvaging[dp->index] && salvageQueue.len[dp->index]) {
1215 node = queue_First(&salvageQueue.part[dp->index], SalvageQueueNode);
1222 * all partitions with scheduled salvages have at least one pending.
1223 * now do an exhaustive search for a scheduled salvage.
1229 dp = (dp->next) ? dp->next : DiskPartitionList, i++ ) {
1230 if (salvageQueue.len[dp->index]) {
1231 node = queue_First(&salvageQueue.part[dp->index], SalvageQueueNode);
1236 /* we should never reach this line */
1240 assert(node != NULL);
1242 partition_salvaging[node->partition_id]++;
1243 DeleteFromSalvageQueue(node);
1244 AddToPendingQueue(node);
1247 /* update next_part_sched field */
1249 next_part_sched = dp->next->index;
1250 } else if (DiskPartitionList) {
1251 next_part_sched = DiskPartitionList->index;
1253 next_part_sched = -1;
1262 * update internal scheduler state to reflect completion of a work unit.
1264 * @param[in] node salvage queue node object pointer
1265 * @param[in] result worker process result code
1267 * @post scheduler state is updated.
1272 SALVSYNC_doneWork_r(struct SalvageQueueNode * node, int result)
1277 DeleteFromPendingQueue(node);
1278 partid = node->partition_id;
1279 if (partid >=0 && partid <= VOLMAXPARTS) {
1280 partition_salvaging[partid]--;
1283 node->state = SALVSYNC_STATE_DONE;
1284 } else if (result != SALSRV_EXIT_VOLGROUP_LINK) {
1285 node->state = SALVSYNC_STATE_ERROR;
1288 if (node->type == SALVSYNC_VOLGROUP_PARENT) {
1289 for (idx = 0; idx < VOLMAXTYPES; idx++) {
1290 if (node->volgroup.children[idx]) {
1291 node->volgroup.children[idx]->state = node->state;
1298 * check whether worker child failed.
1300 * @param[in] status status bitfield return by wait()
1302 * @return boolean failure code
1303 * @retval 0 child succeeded
1304 * @retval 1 child failed
1309 ChildFailed(int status)
1311 return (WCOREDUMP(status) ||
1312 WIFSIGNALED(status) ||
1313 ((WEXITSTATUS(status) != 0) &&
1314 (WEXITSTATUS(status) != SALSRV_EXIT_VOLGROUP_LINK)));
1319 * notify salvsync scheduler of node completion, by child pid.
1321 * @param[in] pid pid of worker child
1322 * @param[in] status worker status bitfield from wait()
1324 * @post scheduler state is updated.
1325 * if status code is a failure, fileserver notification was attempted
1327 * @see SALVSYNC_doneWork_r
1330 SALVSYNC_doneWorkByPid(int pid, int status)
1332 struct SalvageQueueNode * node;
1334 afs_uint32 volids[VOLMAXTYPES+1];
1337 memset(volids, 0, sizeof(volids));
1340 node = LookupPendingCommandByPid(pid);
1342 SALVSYNC_doneWork_r(node, status);
1344 if (ChildFailed(status)) {
1345 /* populate volume id list for later processing outside the glock */
1346 volids[0] = node->command.sop.volume;
1347 strcpy(partName, node->command.sop.partName);
1348 if (node->type == SALVSYNC_VOLGROUP_PARENT) {
1349 for (idx = 0; idx < VOLMAXTYPES; idx++) {
1350 if (node->volgroup.children[idx]) {
1351 volids[idx+1] = node->volgroup.children[idx]->command.sop.volume;
1360 * if necessary, notify fileserver of
1361 * failure to salvage volume group
1362 * [we cannot guarantee that the child made the
1363 * appropriate notifications (e.g. SIGSEGV)]
1364 * -- tkeiser 11/28/2007
1366 if (ChildFailed(status)) {
1367 for (idx = 0; idx <= VOLMAXTYPES; idx++) {
1369 FSYNC_VolOp(volids[idx],
1371 FSYNC_VOL_FORCE_ERROR,
1379 #endif /* AFS_DEMAND_ATTACH_FS */