2 * Copyright 2006-2007, 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
12 * online salvager daemon
15 /* Main program file. Define globals. */
18 #include <afsconfig.h>
19 #include <afs/param.h>
32 #include <WINNT/afsevent.h>
34 #include <sys/param.h>
38 #endif /* ITIMER_REAL */
41 #define WCOREDUMP(x) ((x) & 0200)
44 #include <afs/afsint.h>
45 #include <afs/afs_assert.h>
46 #if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
47 #if defined(AFS_VFSINCL_ENV)
48 #include <sys/vnode.h>
50 #include <sys/fs/ufs_inode.h>
52 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
53 #include <ufs/ufs/dinode.h>
54 #include <ufs/ffs/fs.h>
56 #include <ufs/inode.h>
59 #else /* AFS_VFSINCL_ENV */
61 #include <ufs/inode.h>
62 #else /* AFS_OSF_ENV */
63 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV)
64 #include <sys/inode.h>
67 #endif /* AFS_VFSINCL_ENV */
68 #endif /* AFS_SGI_ENV */
71 #include <sys/lockf.h>
75 #include <checklist.h>
77 #if defined(AFS_SGI_ENV)
82 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
85 #include <sys/mnttab.h>
86 #include <sys/mntent.h>
91 #endif /* AFS_SGI_ENV */
92 #endif /* AFS_HPUX_ENV */
97 #include <afs/osi_inode.h>
100 #include <afs/afsutil.h>
101 #include <afs/fileutil.h>
102 #include <afs/procmgmt.h> /* signal(), kill(), wait(), etc. */
111 #include <afs/afssyscalls.h>
115 #include "partition.h"
116 #include "daemon_com.h"
118 #include "salvsync.h"
119 #include "viceinode.h"
121 #include "vol-salvage.h"
128 #if !defined(AFS_DEMAND_ATTACH_FS)
129 #error "online salvager only supported for demand attach fileserver"
130 #endif /* AFS_DEMAND_ATTACH_FS */
132 #if defined(AFS_NT40_ENV)
133 #error "online salvager not supported on NT"
134 #endif /* AFS_NT40_ENV */
136 /*@+fcnmacros +macrofcndecl@*/
138 #define afs_fopen fopen64
139 #else /* !O_LARGEFILE */
140 #define afs_fopen fopen
141 #endif /* !O_LARGEFILE */
142 /*@=fcnmacros =macrofcndecl@*/
146 static volatile int current_workers = 0;
147 static volatile struct rx_queue pending_q;
148 static pthread_mutex_t worker_lock;
149 static pthread_cond_t worker_cv;
151 static void * SalvageChildReaperThread(void *);
152 static int DoSalvageVolume(struct SalvageQueueNode * node, int slot);
154 static void SalvageServer(int argc, char **argv);
155 static void SalvageClient(VolumeId vid, char * pname);
157 static int Reap_Child(char * prog, int * pid, int * status);
159 static void * SalvageLogCleanupThread(void *);
160 static int SalvageLogCleanup(int pid);
162 static void * SalvageLogScanningThread(void *);
163 static void ScanLogs(struct rx_queue *log_watch_queue);
165 struct cmdline_rock {
170 struct log_cleanup_node {
176 struct rx_queue queue_head;
177 pthread_cond_t queue_change_cv;
181 #define DEFAULT_PARALLELISM 4 /* allow 4 parallel salvage workers by default */
184 handleit(struct cmd_syndesc *as, void *arock)
187 char pname[100], *temp;
188 afs_int32 seenpart = 0, seenvol = 0, vid = 0;
189 struct cmdline_rock *rock = (struct cmdline_rock *)arock;
191 #ifdef AFS_SGI_VNODE_GLUE
192 if (afs_init_kernel_config(-1) < 0) {
194 ("Can't determine NUMA configuration, not starting salvager.\n");
199 if (as->parms[2].items) /* -debug */
201 if (as->parms[3].items) /* -nowrite */
203 if (as->parms[4].items) /* -inodes */
205 if (as->parms[5].items) /* -oktozap */
207 if (as->parms[6].items) /* -rootinodes */
209 if (as->parms[8].items) /* -ForceReads */
211 if ((ti = as->parms[9].items)) { /* -Parallel # */
213 if (strncmp(temp, "all", 3) == 0) {
217 if (strlen(temp) != 0) {
218 Parallel = atoi(temp);
221 if (Parallel > MAXPARALLEL) {
222 printf("Setting parallel salvages to maximum of %d \n",
224 Parallel = MAXPARALLEL;
228 Parallel = MIN(DEFAULT_PARALLELISM, MAXPARALLEL);
230 if ((ti = as->parms[10].items)) { /* -tmpdir */
234 dirp = opendir(tmpdir);
237 ("Can't open temporary placeholder dir %s; using current partition \n",
243 if ((ti = as->parms[11].items)) /* -showlog */
245 if ((ti = as->parms[12].items)) { /* -orphans */
247 orphans = ORPH_IGNORE;
248 else if (strcmp(ti->data, "remove") == 0
249 || strcmp(ti->data, "r") == 0)
250 orphans = ORPH_REMOVE;
251 else if (strcmp(ti->data, "attach") == 0
252 || strcmp(ti->data, "a") == 0)
253 orphans = ORPH_ATTACH;
255 #ifndef AFS_NT40_ENV /* ignore options on NT */
256 if ((ti = as->parms[13].items)) { /* -syslog */
260 if ((ti = as->parms[14].items)) { /* -syslogfacility */
261 useSyslogFacility = atoi(ti->data);
264 if ((ti = as->parms[15].items)) { /* -datelogs */
265 TimeStampLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
269 if ((ti = as->parms[16].items)) { /* -client */
270 if ((ti = as->parms[0].items)) { /* -partition */
272 strlcpy(pname, ti->data, sizeof(pname));
274 if ((ti = as->parms[1].items)) { /* -volumeid */
276 vid = atoi(ti->data);
280 printf("-showlog does not work with -client\n");
284 if (!seenpart || !seenvol) {
285 printf("You must specify '-partition' and '-volumeid' with the '-client' option\n");
289 SalvageClient(vid, pname);
291 } else { /* salvageserver mode */
292 SalvageServer(rock->argc, rock->argv);
299 #include "AFS_component_version_number.c"
303 char *save_args[MAX_ARGS];
305 pthread_t main_thread;
309 main(int argc, char **argv)
311 struct cmd_syndesc *ts;
313 struct cmdline_rock arock;
317 * The following signal action for AIX is necessary so that in case of a
318 * crash (i.e. core is generated) we can include the user's data section
319 * in the core dump. Unfortunately, by default, only a partial core is
320 * generated which, in many cases, isn't too useful.
322 struct sigaction nsa;
324 sigemptyset(&nsa.sa_mask);
325 nsa.sa_handler = SIG_DFL;
326 nsa.sa_flags = SA_FULLDUMP;
327 sigaction(SIGABRT, &nsa, NULL);
328 sigaction(SIGSEGV, &nsa, NULL);
331 /* Initialize directory paths */
332 if (!(initAFSDirPath() & AFSDIR_SERVER_PATHS_OK)) {
334 ReportErrorEventAlt(AFSEVT_SVR_NO_INSTALL_DIR, 0, argv[0], 0);
336 fprintf(stderr, "%s: Unable to obtain AFS server directory.\n",
341 main_thread = pthread_self();
342 if (spawnDatap && spawnDataLen) {
343 /* This is a child per partition salvager. Don't setup log or
344 * try to lock the salvager lock.
346 if (nt_SetupPartitionSalvage(spawnDatap, spawnDataLen) < 0)
352 if (geteuid() != 0) {
353 printf("Salvager must be run as root.\n");
359 /* bad for normal help flag processing, but can do nada */
368 ts = cmd_CreateSyntax("initcmd", handleit, &arock, "initialize the program");
369 cmd_AddParm(ts, "-partition", CMD_SINGLE, CMD_OPTIONAL,
370 "Name of partition to salvage");
371 cmd_AddParm(ts, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
372 "Volume Id to salvage");
373 cmd_AddParm(ts, "-debug", CMD_FLAG, CMD_OPTIONAL,
374 "Run in Debugging mode");
375 cmd_AddParm(ts, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
376 "Run readonly/test mode");
377 cmd_AddParm(ts, "-inodes", CMD_FLAG, CMD_OPTIONAL,
378 "Just list affected afs inodes - debugging flag");
379 cmd_AddParm(ts, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
380 "Give permission to destroy bogus inodes/volumes - debugging flag");
381 cmd_AddParm(ts, "-rootinodes", CMD_FLAG, CMD_OPTIONAL,
382 "Show inodes owned by root - debugging flag");
383 cmd_AddParm(ts, "-salvagedirs", CMD_FLAG, CMD_OPTIONAL,
384 "Force rebuild/salvage of all directories");
385 cmd_AddParm(ts, "-blockreads", CMD_FLAG, CMD_OPTIONAL,
386 "Read smaller blocks to handle IO/bad blocks");
387 cmd_AddParm(ts, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
388 "# of max parallel partition salvaging");
389 cmd_AddParm(ts, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
390 "Name of dir to place tmp files ");
391 cmd_AddParm(ts, "-showlog", CMD_FLAG, CMD_OPTIONAL,
392 "Show log file upon completion");
393 cmd_AddParm(ts, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
394 "ignore | remove | attach");
396 /* note - syslog isn't avail on NT, but if we make it conditional, have
397 * to deal with screwy offsets for cmd params */
398 cmd_AddParm(ts, "-syslog", CMD_FLAG, CMD_OPTIONAL,
399 "Write salvage log to syslogs");
400 cmd_AddParm(ts, "-syslogfacility", CMD_SINGLE, CMD_OPTIONAL,
401 "Syslog facility number to use");
402 cmd_AddParm(ts, "-datelogs", CMD_FLAG, CMD_OPTIONAL,
403 "Include timestamp in logfile filename");
405 cmd_AddParm(ts, "-client", CMD_FLAG, CMD_OPTIONAL,
406 "Use SALVSYNC to ask salvageserver to salvage a volume");
408 err = cmd_Dispatch(argc, argv);
410 return 0; /* not reached */
414 SalvageClient(VolumeId vid, char * pname)
419 SALVSYNC_response_hdr sres;
420 VolumePackageOptions opts;
422 VOptDefaults(volumeUtility, &opts);
423 if (VInitVolumePackage2(volumeUtility, &opts)) {
424 /* VInitVolumePackage2 can fail on e.g. partition attachment errors,
425 * but we don't really care, since all we're doing is trying to use
427 fprintf(stderr, "errors encountered initializing volume package, but "
428 "trying to continue anyway\n");
430 SALVSYNC_clientInit();
432 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_SALVAGE, SALVSYNC_OPERATOR, 0, NULL);
433 if (code != SYNC_OK) {
437 res.payload.buf = (void *) &sres;
438 res.payload.len = sizeof(sres);
442 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_QUERY, SALVSYNC_WHATEVER, 0, &res);
443 if (code != SYNC_OK) {
446 switch (sres.state) {
447 case SALVSYNC_STATE_ERROR:
448 printf("salvageserver reports salvage ended in an error; check log files for more details\n");
449 case SALVSYNC_STATE_DONE:
450 case SALVSYNC_STATE_UNKNOWN:
454 SALVSYNC_clientFinis();
458 if (code == SYNC_DENIED) {
459 printf("salvageserver refused to salvage volume %u on partition %s\n",
461 } else if (code == SYNC_BAD_COMMAND) {
462 printf("SALVSYNC protocol mismatch; please make sure fileserver, volserver, salvageserver and salvager are same version\n");
463 } else if (code == SYNC_COM_ERROR) {
464 printf("SALVSYNC communications error\n");
466 SALVSYNC_clientFinis();
470 static int * child_slot;
473 SalvageServer(int argc, char **argv)
476 struct SalvageQueueNode * node;
478 pthread_attr_t attrs;
480 VolumePackageOptions opts;
482 /* All entries to the log will be appended. Useful if there are
483 * multiple salvagers appending to the log.
486 CheckLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
488 #ifdef AFS_LINUX20_ENV
489 fcntl(fileno(logFile), F_SETFL, O_APPEND); /* Isn't this redundant? */
491 fcntl(fileno(logFile), F_SETFL, FAPPEND); /* Isn't this redundant? */
496 fprintf(logFile, "%s\n", cml_version_number);
497 LogCommandLine(argc, argv, "Online Salvage Server",
498 SalvageVersion, "Starting OpenAFS", Log);
499 /* Get and hold a lock for the duration of the salvage to make sure
500 * that no other salvage runs at the same time. The routine
501 * VInitVolumePackage2 (called below) makes sure that a file server or
502 * other volume utilities don't interfere with the salvage.
505 /* even demand attach online salvager
506 * still needs this because we don't want
507 * a stand-alone salvager to conflict with
508 * the salvager daemon */
509 ObtainSharedSalvageLock();
511 child_slot = (int *) malloc(Parallel * sizeof(int));
512 osi_Assert(child_slot != NULL);
513 memset(child_slot, 0, Parallel * sizeof(int));
515 /* initialize things */
516 VOptDefaults(salvageServer, &opts);
517 if (VInitVolumePackage2(salvageServer, &opts)) {
518 Log("Shutting down: errors encountered initializing volume package\n");
522 queue_Init(&pending_q);
523 queue_Init(&log_cleanup_queue);
524 MUTEX_INIT(&worker_lock, "worker", MUTEX_DEFAULT, 0);
525 CV_INIT(&worker_cv, "worker", CV_DEFAULT, 0);
526 CV_INIT(&log_cleanup_queue.queue_change_cv, "queuechange", CV_DEFAULT, 0);
527 osi_Assert(pthread_attr_init(&attrs) == 0);
529 /* start up the reaper and log cleaner threads */
530 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
531 osi_Assert(pthread_create(&tid,
533 &SalvageChildReaperThread,
535 osi_Assert(pthread_create(&tid,
537 &SalvageLogCleanupThread,
539 osi_Assert(pthread_create(&tid,
541 &SalvageLogScanningThread,
544 /* loop forever serving requests */
546 node = SALVSYNC_getWork();
547 osi_Assert(node != NULL);
549 Log("dispatching child to salvage volume %u...\n",
550 node->command.sop.parent);
554 for (slot = 0; slot < Parallel; slot++) {
555 if (!child_slot[slot])
558 osi_Assert (slot < Parallel);
564 ret = DoSalvageVolume(node, slot);
566 } else if (pid < 0) {
567 Log("failed to fork child worker process\n");
571 child_slot[slot] = pid;
575 MUTEX_ENTER(&worker_lock);
578 /* let the reaper thread know another worker was spawned */
579 CV_BROADCAST(&worker_cv);
581 /* if we're overquota, wait for the reaper */
582 while (current_workers >= Parallel) {
583 CV_WAIT(&worker_cv, &worker_lock);
585 MUTEX_EXIT(&worker_lock);
591 DoSalvageVolume(struct SalvageQueueNode * node, int slot)
593 char childLog[AFSDIR_PATH_MAX];
594 struct DiskPartition64 * partP;
596 /* do not allow further forking inside salvager */
599 /* do not attempt to close parent's logFile handle as
600 * another thread may have held the lock on the FILE
601 * structure when fork was called! */
603 afs_snprintf(childLog, sizeof(childLog), "%s.%d",
604 AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid());
606 logFile = afs_fopen(childLog, "a");
607 if (!logFile) { /* still nothing, use stdout */
612 if (node->command.sop.parent <= 0) {
613 Log("salvageServer: invalid volume id specified; salvage aborted\n");
617 partP = VGetPartition(node->command.sop.partName, 0);
619 Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
620 node->command.sop.partName);
624 /* obtain a shared salvage lock in the child worker, so if the
625 * salvageserver restarts (and we continue), we will still hold a lock and
626 * prevent standalone salvagers from interfering */
627 ObtainSharedSalvageLock();
629 /* Salvage individual volume; don't notify fs */
630 SalvageFileSys1(partP, node->command.sop.parent);
638 SalvageChildReaperThread(void * args)
640 int slot, pid, status;
641 struct log_cleanup_node * cleanup;
643 MUTEX_ENTER(&worker_lock);
645 /* loop reaping our children */
647 /* wait() won't block unless we have children, so
648 * block on the cond var if we're childless */
649 while (current_workers == 0) {
650 CV_WAIT(&worker_cv, &worker_lock);
653 MUTEX_EXIT(&worker_lock);
655 cleanup = (struct log_cleanup_node *) malloc(sizeof(struct log_cleanup_node));
657 while (Reap_Child("salvageserver", &pid, &status) < 0) {
658 /* try to prevent livelock if something goes wrong */
663 for (slot = 0; slot < Parallel; slot++) {
664 if (child_slot[slot] == pid)
667 osi_Assert(slot < Parallel);
668 child_slot[slot] = 0;
671 SALVSYNC_doneWorkByPid(pid, status);
673 MUTEX_ENTER(&worker_lock);
677 queue_Append(&log_cleanup_queue, cleanup);
678 CV_SIGNAL(&log_cleanup_queue.queue_change_cv);
681 /* ok, we've reaped a child */
683 CV_BROADCAST(&worker_cv);
690 Reap_Child(char *prog, int * pid, int * status)
697 if (WCOREDUMP(*status))
698 Log("\"%s\" core dumped!\n", prog);
699 if ((WIFSIGNALED(*status) != 0) ||
700 ((WEXITSTATUS(*status) != 0) &&
701 (WEXITSTATUS(*status) != SALSRV_EXIT_VOLGROUP_LINK)))
702 Log("\"%s\" (pid=%d) terminated abnormally!\n", prog, ret);
704 Log("wait returned -1\n");
710 * thread to combine salvager child logs
711 * back into the main salvageserver log
714 SalvageLogCleanupThread(void * arg)
716 struct log_cleanup_node * cleanup;
718 MUTEX_ENTER(&worker_lock);
721 while (queue_IsEmpty(&log_cleanup_queue)) {
722 CV_WAIT(&log_cleanup_queue.queue_change_cv, &worker_lock);
725 while (queue_IsNotEmpty(&log_cleanup_queue)) {
726 cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
727 queue_Remove(cleanup);
728 MUTEX_EXIT(&worker_lock);
729 SalvageLogCleanup(cleanup->pid);
731 MUTEX_ENTER(&worker_lock);
735 MUTEX_EXIT(&worker_lock);
739 #define LOG_XFER_BUF_SIZE 65536
741 SalvageLogCleanup(int pid)
744 char fn[AFSDIR_PATH_MAX];
745 static char buf[LOG_XFER_BUF_SIZE];
747 afs_snprintf(fn, sizeof(fn), "%s.%d",
748 AFSDIR_SERVER_SLVGLOG_FILEPATH, pid);
751 pidlog = open(fn, O_RDONLY);
756 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
758 fwrite(buf, len, 1, logFile);
759 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
767 /* wake up every five minutes to see if a non-child salvage has finished */
768 #define SALVAGE_SCAN_POLL_INTERVAL 300
771 * Thread to look for SalvageLog.$pid files that are not from our child
772 * worker salvagers, and notify SalvageLogCleanupThread to clean them
773 * up. This can happen if we restart during salvages, or the
774 * salvageserver crashes or something.
778 * @return always NULL
781 SalvageLogScanningThread(void * arg)
783 struct rx_queue log_watch_queue;
785 queue_Init(&log_watch_queue);
790 char prefix[AFSDIR_PATH_MAX];
793 afs_snprintf(prefix, sizeof(prefix), "%s.", AFSDIR_SLVGLOG_FILE);
794 prefix_len = strlen(prefix);
796 dp = opendir(AFSDIR_LOGS_DIR);
799 while ((dirp = readdir(dp)) != NULL) {
801 struct log_cleanup_node *cleanup;
804 if (strncmp(dirp->d_name, prefix, prefix_len) != 0) {
805 /* not a salvage logfile; skip */
810 pid = strtol(dirp->d_name + prefix_len, NULL, 10);
813 /* file is SalvageLog.<something> but <something> isn't
819 for (i = 0; i < Parallel; ++i) {
820 if (pid == child_slot[i]) {
826 /* this pid is one of our children, so the reaper thread
827 * will take care of it; skip */
832 (struct log_cleanup_node *) malloc(sizeof(struct log_cleanup_node));
835 queue_Append(&log_watch_queue, cleanup);
841 ScanLogs(&log_watch_queue);
843 while (queue_IsNotEmpty(&log_watch_queue)) {
844 sleep(SALVAGE_SCAN_POLL_INTERVAL);
845 ScanLogs(&log_watch_queue);
852 * look through log_watch_queue, and if any processes are not still
853 * running, hand them off to the SalvageLogCleanupThread
855 * @param log_watch_queue a queue of PIDs that we should clean up if
859 ScanLogs(struct rx_queue *log_watch_queue)
861 struct log_cleanup_node *cleanup, *next;
863 MUTEX_ENTER(&worker_lock);
865 for (queue_Scan(log_watch_queue, cleanup, next, log_cleanup_node)) {
866 /* if a process is still running, assume it's the salvage process
867 * still going, and keep waiting for it */
868 if (kill(cleanup->pid, 0) < 0 && errno == ESRCH) {
869 queue_Remove(cleanup);
870 queue_Append(&log_cleanup_queue, cleanup);
871 CV_SIGNAL(&log_cleanup_queue.queue_change_cv);
875 MUTEX_EXIT(&worker_lock);