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>
31 #include <WINNT/afsevent.h>
33 #include <sys/param.h>
37 #endif /* ITIMER_REAL */
40 #define WCOREDUMP(x) ((x) & 0200)
43 #include <afs/afsint.h>
44 #include <afs/assert.h>
45 #if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
46 #if defined(AFS_VFSINCL_ENV)
47 #include <sys/vnode.h>
49 #include <sys/fs/ufs_inode.h>
51 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
52 #include <ufs/ufs/dinode.h>
53 #include <ufs/ffs/fs.h>
55 #include <ufs/inode.h>
58 #else /* AFS_VFSINCL_ENV */
60 #include <ufs/inode.h>
61 #else /* AFS_OSF_ENV */
62 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV)
63 #include <sys/inode.h>
66 #endif /* AFS_VFSINCL_ENV */
67 #endif /* AFS_SGI_ENV */
70 #include <sys/lockf.h>
74 #include <checklist.h>
76 #if defined(AFS_SGI_ENV)
81 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
84 #include <sys/mnttab.h>
85 #include <sys/mntent.h>
90 #endif /* AFS_SGI_ENV */
91 #endif /* AFS_HPUX_ENV */
96 #include <afs/osi_inode.h>
99 #include <afs/afsutil.h>
100 #include <afs/fileutil.h>
101 #include <afs/procmgmt.h> /* signal(), kill(), wait(), etc. */
110 #include <afs/afssyscalls.h>
114 #include "partition.h"
115 #include "daemon_com.h"
117 #include "salvsync.h"
118 #include "viceinode.h"
120 #include "vol-salvage.h"
127 #if !defined(AFS_DEMAND_ATTACH_FS)
128 #error "online salvager only supported for demand attach fileserver"
129 #endif /* AFS_DEMAND_ATTACH_FS */
131 #if defined(AFS_NT40_ENV)
132 #error "online salvager not supported on NT"
133 #endif /* AFS_NT40_ENV */
135 /*@+fcnmacros +macrofcndecl@*/
137 #define afs_fopen fopen64
138 #else /* !O_LARGEFILE */
139 #define afs_fopen fopen
140 #endif /* !O_LARGEFILE */
141 /*@=fcnmacros =macrofcndecl@*/
145 static volatile int current_workers = 0;
146 static volatile struct rx_queue pending_q;
147 static pthread_mutex_t worker_lock;
148 static pthread_cond_t worker_cv;
150 static void * SalvageChildReaperThread(void *);
151 static int DoSalvageVolume(struct SalvageQueueNode * node, int slot);
153 static void SalvageServer(int argc, char **argv);
154 static void SalvageClient(VolumeId vid, char * pname);
156 static int Reap_Child(char * prog, int * pid, int * status);
158 static void * SalvageLogCleanupThread(void *);
159 static int SalvageLogCleanup(int pid);
161 static void * SalvageLogScanningThread(void *);
162 static void ScanLogs(struct rx_queue *log_watch_queue);
164 struct cmdline_rock {
169 struct log_cleanup_node {
175 struct rx_queue queue_head;
176 pthread_cond_t queue_change_cv;
180 #define DEFAULT_PARALLELISM 4 /* allow 4 parallel salvage workers by default */
183 handleit(struct cmd_syndesc *as, void *arock)
186 char pname[100], *temp;
187 afs_int32 seenpart = 0, seenvol = 0, vid = 0;
188 struct cmdline_rock *rock = (struct cmdline_rock *)arock;
190 #ifdef AFS_SGI_VNODE_GLUE
191 if (afs_init_kernel_config(-1) < 0) {
193 ("Can't determine NUMA configuration, not starting salvager.\n");
198 if (as->parms[2].items) /* -debug */
200 if (as->parms[3].items) /* -nowrite */
202 if (as->parms[4].items) /* -inodes */
204 if (as->parms[5].items) /* -oktozap */
206 if (as->parms[6].items) /* -rootinodes */
208 if (as->parms[8].items) /* -ForceReads */
210 if ((ti = as->parms[9].items)) { /* -Parallel # */
212 if (strncmp(temp, "all", 3) == 0) {
216 if (strlen(temp) != 0) {
217 Parallel = atoi(temp);
220 if (Parallel > MAXPARALLEL) {
221 printf("Setting parallel salvages to maximum of %d \n",
223 Parallel = MAXPARALLEL;
227 Parallel = MIN(DEFAULT_PARALLELISM, MAXPARALLEL);
229 if ((ti = as->parms[10].items)) { /* -tmpdir */
233 dirp = opendir(tmpdir);
236 ("Can't open temporary placeholder dir %s; using current partition \n",
242 if ((ti = as->parms[11].items)) /* -showlog */
244 if ((ti = as->parms[12].items)) { /* -orphans */
246 orphans = ORPH_IGNORE;
247 else if (strcmp(ti->data, "remove") == 0
248 || strcmp(ti->data, "r") == 0)
249 orphans = ORPH_REMOVE;
250 else if (strcmp(ti->data, "attach") == 0
251 || strcmp(ti->data, "a") == 0)
252 orphans = ORPH_ATTACH;
254 #ifndef AFS_NT40_ENV /* ignore options on NT */
255 if ((ti = as->parms[13].items)) { /* -syslog */
259 if ((ti = as->parms[14].items)) { /* -syslogfacility */
260 useSyslogFacility = atoi(ti->data);
263 if ((ti = as->parms[15].items)) { /* -datelogs */
264 TimeStampLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
268 if ((ti = as->parms[16].items)) { /* -client */
269 if ((ti = as->parms[0].items)) { /* -partition */
271 strlcpy(pname, ti->data, sizeof(pname));
273 if ((ti = as->parms[1].items)) { /* -volumeid */
275 vid = atoi(ti->data);
279 printf("-showlog does not work with -client\n");
283 if (!seenpart || !seenvol) {
284 printf("You must specify '-partition' and '-volumeid' with the '-client' option\n");
288 SalvageClient(vid, pname);
290 } else { /* salvageserver mode */
291 SalvageServer(rock->argc, rock->argv);
298 #include "AFS_component_version_number.c"
302 char *save_args[MAX_ARGS];
304 pthread_t main_thread;
308 main(int argc, char **argv)
310 struct cmd_syndesc *ts;
312 struct cmdline_rock arock;
316 * The following signal action for AIX is necessary so that in case of a
317 * crash (i.e. core is generated) we can include the user's data section
318 * in the core dump. Unfortunately, by default, only a partial core is
319 * generated which, in many cases, isn't too useful.
321 struct sigaction nsa;
323 sigemptyset(&nsa.sa_mask);
324 nsa.sa_handler = SIG_DFL;
325 nsa.sa_flags = SA_FULLDUMP;
326 sigaction(SIGABRT, &nsa, NULL);
327 sigaction(SIGSEGV, &nsa, NULL);
330 /* Initialize directory paths */
331 if (!(initAFSDirPath() & AFSDIR_SERVER_PATHS_OK)) {
333 ReportErrorEventAlt(AFSEVT_SVR_NO_INSTALL_DIR, 0, argv[0], 0);
335 fprintf(stderr, "%s: Unable to obtain AFS server directory.\n",
340 main_thread = pthread_self();
341 if (spawnDatap && spawnDataLen) {
342 /* This is a child per partition salvager. Don't setup log or
343 * try to lock the salvager lock.
345 if (nt_SetupPartitionSalvage(spawnDatap, spawnDataLen) < 0)
351 if (geteuid() != 0) {
352 printf("Salvager must be run as root.\n");
358 /* bad for normal help flag processing, but can do nada */
367 ts = cmd_CreateSyntax("initcmd", handleit, &arock, "initialize the program");
368 cmd_AddParm(ts, "-partition", CMD_SINGLE, CMD_OPTIONAL,
369 "Name of partition to salvage");
370 cmd_AddParm(ts, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
371 "Volume Id to salvage");
372 cmd_AddParm(ts, "-debug", CMD_FLAG, CMD_OPTIONAL,
373 "Run in Debugging mode");
374 cmd_AddParm(ts, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
375 "Run readonly/test mode");
376 cmd_AddParm(ts, "-inodes", CMD_FLAG, CMD_OPTIONAL,
377 "Just list affected afs inodes - debugging flag");
378 cmd_AddParm(ts, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
379 "Give permission to destroy bogus inodes/volumes - debugging flag");
380 cmd_AddParm(ts, "-rootinodes", CMD_FLAG, CMD_OPTIONAL,
381 "Show inodes owned by root - debugging flag");
382 cmd_AddParm(ts, "-salvagedirs", CMD_FLAG, CMD_OPTIONAL,
383 "Force rebuild/salvage of all directories");
384 cmd_AddParm(ts, "-blockreads", CMD_FLAG, CMD_OPTIONAL,
385 "Read smaller blocks to handle IO/bad blocks");
386 cmd_AddParm(ts, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
387 "# of max parallel partition salvaging");
388 cmd_AddParm(ts, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
389 "Name of dir to place tmp files ");
390 cmd_AddParm(ts, "-showlog", CMD_FLAG, CMD_OPTIONAL,
391 "Show log file upon completion");
392 cmd_AddParm(ts, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
393 "ignore | remove | attach");
395 /* note - syslog isn't avail on NT, but if we make it conditional, have
396 * to deal with screwy offsets for cmd params */
397 cmd_AddParm(ts, "-syslog", CMD_FLAG, CMD_OPTIONAL,
398 "Write salvage log to syslogs");
399 cmd_AddParm(ts, "-syslogfacility", CMD_SINGLE, CMD_OPTIONAL,
400 "Syslog facility number to use");
401 cmd_AddParm(ts, "-datelogs", CMD_FLAG, CMD_OPTIONAL,
402 "Include timestamp in logfile filename");
404 cmd_AddParm(ts, "-client", CMD_FLAG, CMD_OPTIONAL,
405 "Use SALVSYNC to ask salvageserver to salvage a volume");
407 err = cmd_Dispatch(argc, argv);
409 return 0; /* not reached */
413 SalvageClient(VolumeId vid, char * pname)
418 SALVSYNC_response_hdr sres;
419 VolumePackageOptions opts;
421 VOptDefaults(volumeUtility, &opts);
422 if (VInitVolumePackage2(volumeUtility, &opts)) {
423 /* VInitVolumePackage2 can fail on e.g. partition attachment errors,
424 * but we don't really care, since all we're doing is trying to use
426 fprintf(stderr, "errors encountered initializing volume package, but "
427 "trying to continue anyway\n");
429 SALVSYNC_clientInit();
431 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_SALVAGE, SALVSYNC_OPERATOR, 0, NULL);
432 if (code != SYNC_OK) {
436 res.payload.buf = (void *) &sres;
437 res.payload.len = sizeof(sres);
441 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_QUERY, SALVSYNC_WHATEVER, 0, &res);
442 if (code != SYNC_OK) {
445 switch (sres.state) {
446 case SALVSYNC_STATE_ERROR:
447 printf("salvageserver reports salvage ended in an error; check log files for more details\n");
448 case SALVSYNC_STATE_DONE:
449 case SALVSYNC_STATE_UNKNOWN:
453 SALVSYNC_clientFinis();
457 if (code == SYNC_DENIED) {
458 printf("salvageserver refused to salvage volume %u on partition %s\n",
460 } else if (code == SYNC_BAD_COMMAND) {
461 printf("SALVSYNC protocol mismatch; please make sure fileserver, volserver, salvageserver and salvager are same version\n");
462 } else if (code == SYNC_COM_ERROR) {
463 printf("SALVSYNC communications error\n");
465 SALVSYNC_clientFinis();
469 static int * child_slot;
472 SalvageServer(int argc, char **argv)
475 struct SalvageQueueNode * node;
477 pthread_attr_t attrs;
479 VolumePackageOptions opts;
481 /* All entries to the log will be appended. Useful if there are
482 * multiple salvagers appending to the log.
485 CheckLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
487 #ifdef AFS_LINUX20_ENV
488 fcntl(fileno(logFile), F_SETFL, O_APPEND); /* Isn't this redundant? */
490 fcntl(fileno(logFile), F_SETFL, FAPPEND); /* Isn't this redundant? */
495 fprintf(logFile, "%s\n", cml_version_number);
496 LogCommandLine(argc, argv, "Online Salvage Server",
497 SalvageVersion, "Starting OpenAFS", Log);
498 /* Get and hold a lock for the duration of the salvage to make sure
499 * that no other salvage runs at the same time. The routine
500 * VInitVolumePackage2 (called below) makes sure that a file server or
501 * other volume utilities don't interfere with the salvage.
504 /* even demand attach online salvager
505 * still needs this because we don't want
506 * a stand-alone salvager to conflict with
507 * the salvager daemon */
508 ObtainSharedSalvageLock();
510 child_slot = (int *) malloc(Parallel * sizeof(int));
511 assert(child_slot != NULL);
512 memset(child_slot, 0, Parallel * sizeof(int));
514 /* initialize things */
515 VOptDefaults(salvageServer, &opts);
516 if (VInitVolumePackage2(salvageServer, &opts)) {
517 Log("Shutting down: errors encountered initializing volume package\n");
521 queue_Init(&pending_q);
522 queue_Init(&log_cleanup_queue);
523 assert(pthread_mutex_init(&worker_lock, NULL) == 0);
524 assert(pthread_cond_init(&worker_cv, NULL) == 0);
525 assert(pthread_cond_init(&log_cleanup_queue.queue_change_cv, NULL) == 0);
526 assert(pthread_attr_init(&attrs) == 0);
528 /* start up the reaper and log cleaner threads */
529 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
530 assert(pthread_create(&tid,
532 &SalvageChildReaperThread,
534 assert(pthread_create(&tid,
536 &SalvageLogCleanupThread,
538 assert(pthread_create(&tid,
540 &SalvageLogScanningThread,
543 /* loop forever serving requests */
545 node = SALVSYNC_getWork();
546 assert(node != NULL);
548 Log("dispatching child to salvage volume %u...\n",
549 node->command.sop.parent);
553 for (slot = 0; slot < Parallel; slot++) {
554 if (!child_slot[slot])
557 assert (slot < Parallel);
563 ret = DoSalvageVolume(node, slot);
565 } else if (pid < 0) {
566 Log("failed to fork child worker process\n");
570 child_slot[slot] = pid;
574 assert(pthread_mutex_lock(&worker_lock) == 0);
577 /* let the reaper thread know another worker was spawned */
578 assert(pthread_cond_broadcast(&worker_cv) == 0);
580 /* if we're overquota, wait for the reaper */
581 while (current_workers >= Parallel) {
582 assert(pthread_cond_wait(&worker_cv, &worker_lock) == 0);
584 assert(pthread_mutex_unlock(&worker_lock) == 0);
590 DoSalvageVolume(struct SalvageQueueNode * node, int slot)
592 char childLog[AFSDIR_PATH_MAX];
593 struct DiskPartition64 * partP;
595 /* do not allow further forking inside salvager */
598 /* do not attempt to close parent's logFile handle as
599 * another thread may have held the lock on the FILE
600 * structure when fork was called! */
602 afs_snprintf(childLog, sizeof(childLog), "%s.%d",
603 AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid());
605 logFile = afs_fopen(childLog, "a");
606 if (!logFile) { /* still nothing, use stdout */
611 if (node->command.sop.parent <= 0) {
612 Log("salvageServer: invalid volume id specified; salvage aborted\n");
616 partP = VGetPartition(node->command.sop.partName, 0);
618 Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
619 node->command.sop.partName);
623 /* obtain a shared salvage lock in the child worker, so if the
624 * salvageserver restarts (and we continue), we will still hold a lock and
625 * prevent standalone salvagers from interfering */
626 ObtainSharedSalvageLock();
628 /* Salvage individual volume; don't notify fs */
629 SalvageFileSys1(partP, node->command.sop.parent);
637 SalvageChildReaperThread(void * args)
639 int slot, pid, status;
640 struct log_cleanup_node * cleanup;
642 assert(pthread_mutex_lock(&worker_lock) == 0);
644 /* loop reaping our children */
646 /* wait() won't block unless we have children, so
647 * block on the cond var if we're childless */
648 while (current_workers == 0) {
649 assert(pthread_cond_wait(&worker_cv, &worker_lock) == 0);
652 assert(pthread_mutex_unlock(&worker_lock) == 0);
654 cleanup = (struct log_cleanup_node *) malloc(sizeof(struct log_cleanup_node));
656 while (Reap_Child("salvageserver", &pid, &status) < 0) {
657 /* try to prevent livelock if something goes wrong */
662 for (slot = 0; slot < Parallel; slot++) {
663 if (child_slot[slot] == pid)
666 assert(slot < Parallel);
667 child_slot[slot] = 0;
670 SALVSYNC_doneWorkByPid(pid, status);
672 assert(pthread_mutex_lock(&worker_lock) == 0);
676 queue_Append(&log_cleanup_queue, cleanup);
677 assert(pthread_cond_signal(&log_cleanup_queue.queue_change_cv) == 0);
680 /* ok, we've reaped a child */
682 assert(pthread_cond_broadcast(&worker_cv) == 0);
689 Reap_Child(char *prog, int * pid, int * status)
696 if (WCOREDUMP(*status))
697 Log("\"%s\" core dumped!\n", prog);
698 if ((WIFSIGNALED(*status) != 0) ||
699 ((WEXITSTATUS(*status) != 0) &&
700 (WEXITSTATUS(*status) != SALSRV_EXIT_VOLGROUP_LINK)))
701 Log("\"%s\" (pid=%d) terminated abnormally!\n", prog, ret);
703 Log("wait returned -1\n");
709 * thread to combine salvager child logs
710 * back into the main salvageserver log
713 SalvageLogCleanupThread(void * arg)
715 struct log_cleanup_node * cleanup;
717 assert(pthread_mutex_lock(&worker_lock) == 0);
720 while (queue_IsEmpty(&log_cleanup_queue)) {
721 assert(pthread_cond_wait(&log_cleanup_queue.queue_change_cv, &worker_lock) == 0);
724 while (queue_IsNotEmpty(&log_cleanup_queue)) {
725 cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
726 queue_Remove(cleanup);
727 assert(pthread_mutex_unlock(&worker_lock) == 0);
728 SalvageLogCleanup(cleanup->pid);
730 assert(pthread_mutex_lock(&worker_lock) == 0);
734 assert(pthread_mutex_unlock(&worker_lock) == 0);
738 #define LOG_XFER_BUF_SIZE 65536
740 SalvageLogCleanup(int pid)
743 char fn[AFSDIR_PATH_MAX];
744 static char buf[LOG_XFER_BUF_SIZE];
746 afs_snprintf(fn, sizeof(fn), "%s.%d",
747 AFSDIR_SERVER_SLVGLOG_FILEPATH, pid);
750 pidlog = open(fn, O_RDONLY);
755 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
757 fwrite(buf, len, 1, logFile);
758 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
766 /* wake up every five minutes to see if a non-child salvage has finished */
767 #define SALVAGE_SCAN_POLL_INTERVAL 300
770 * Thread to look for SalvageLog.$pid files that are not from our child
771 * worker salvagers, and notify SalvageLogCleanupThread to clean them
772 * up. This can happen if we restart during salvages, or the
773 * salvageserver crashes or something.
777 * @return always NULL
780 SalvageLogScanningThread(void * arg)
782 struct rx_queue log_watch_queue;
784 queue_Init(&log_watch_queue);
789 char prefix[AFSDIR_PATH_MAX];
792 afs_snprintf(prefix, sizeof(prefix), "%s.", AFSDIR_SLVGLOG_FILE);
793 prefix_len = strlen(prefix);
795 dp = opendir(AFSDIR_LOGS_DIR);
798 while ((dirp = readdir(dp)) != NULL) {
800 struct log_cleanup_node *cleanup;
803 if (strncmp(dirp->d_name, prefix, prefix_len) != 0) {
804 /* not a salvage logfile; skip */
809 pid = strtol(dirp->d_name + prefix_len, NULL, 10);
812 /* file is SalvageLog.<something> but <something> isn't
818 for (i = 0; i < Parallel; ++i) {
819 if (pid == child_slot[i]) {
825 /* this pid is one of our children, so the reaper thread
826 * will take care of it; skip */
831 (struct log_cleanup_node *) malloc(sizeof(struct log_cleanup_node));
834 queue_Append(&log_watch_queue, cleanup);
840 ScanLogs(&log_watch_queue);
842 while (queue_IsNotEmpty(&log_watch_queue)) {
843 sleep(SALVAGE_SCAN_POLL_INTERVAL);
844 ScanLogs(&log_watch_queue);
851 * look through log_watch_queue, and if any processes are not still
852 * running, hand them off to the SalvageLogCleanupThread
854 * @param log_watch_queue a queue of PIDs that we should clean up if
858 ScanLogs(struct rx_queue *log_watch_queue)
860 struct log_cleanup_node *cleanup, *next;
862 assert(pthread_mutex_lock(&worker_lock) == 0);
864 for (queue_Scan(log_watch_queue, cleanup, next, log_cleanup_node)) {
865 /* if a process is still running, assume it's the salvage process
866 * still going, and keep waiting for it */
867 if (kill(cleanup->pid, 0) < 0 && errno == ESRCH) {
868 queue_Remove(cleanup);
869 queue_Append(&log_cleanup_queue, cleanup);
870 assert(pthread_cond_signal(&log_cleanup_queue.queue_change_cv) == 0);
874 assert(pthread_mutex_unlock(&worker_lock) == 0);