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>
23 #ifdef HAVE_SYS_FILE_H
28 #include <WINNT/afsevent.h>
32 #define WCOREDUMP(x) ((x) & 0200)
36 #include <afs/afsint.h>
38 #if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
39 #if defined(AFS_VFSINCL_ENV)
40 #include <sys/vnode.h>
42 #include <sys/fs/ufs_inode.h>
44 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
45 #include <ufs/ufs/dinode.h>
46 #include <ufs/ffs/fs.h>
48 #include <ufs/inode.h>
51 #else /* AFS_VFSINCL_ENV */
53 #include <ufs/inode.h>
54 #else /* AFS_OSF_ENV */
55 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
56 #include <sys/inode.h>
59 #endif /* AFS_VFSINCL_ENV */
60 #endif /* AFS_SGI_ENV */
63 #include <sys/lockf.h>
66 #include <checklist.h>
68 #if defined(AFS_SGI_ENV)
71 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
73 #include <sys/mnttab.h>
74 #include <sys/mntent.h>
79 #endif /* AFS_SGI_ENV */
80 #endif /* AFS_HPUX_ENV */
84 #include <afs/osi_inode.h>
87 #include <afs/afsutil.h>
88 #include <afs/fileutil.h>
89 #include <afs/procmgmt.h> /* signal(), kill(), wait(), etc. */
95 #include <afs/afssyscalls.h>
99 #include "partition.h"
100 #include "daemon_com.h"
102 #include "salvsync.h"
103 #include "viceinode.h"
105 #include "vol-salvage.h"
112 #if !defined(AFS_DEMAND_ATTACH_FS)
113 #error "online salvager only supported for demand attach fileserver"
114 #endif /* AFS_DEMAND_ATTACH_FS */
116 #if defined(AFS_NT40_ENV)
117 #error "online salvager not supported on NT"
118 #endif /* AFS_NT40_ENV */
120 /*@+fcnmacros +macrofcndecl@*/
122 #define afs_fopen fopen64
123 #else /* !O_LARGEFILE */
124 #define afs_fopen fopen
125 #endif /* !O_LARGEFILE */
126 /*@=fcnmacros =macrofcndecl@*/
130 static volatile int current_workers = 0;
131 static volatile struct rx_queue pending_q;
132 static pthread_mutex_t worker_lock;
133 static pthread_cond_t worker_cv;
135 static void * SalvageChildReaperThread(void *);
136 static int DoSalvageVolume(struct SalvageQueueNode * node, int slot);
138 static void SalvageServer(int argc, char **argv);
139 static void SalvageClient(VolumeId vid, char * pname);
141 static int Reap_Child(char * prog, int * pid, int * status);
143 static void * SalvageLogCleanupThread(void *);
144 static int SalvageLogCleanup(int pid);
146 static void * SalvageLogScanningThread(void *);
147 static void ScanLogs(struct rx_queue *log_watch_queue);
149 struct cmdline_rock {
154 struct log_cleanup_node {
160 struct rx_queue queue_head;
161 pthread_cond_t queue_change_cv;
165 #define DEFAULT_PARALLELISM 4 /* allow 4 parallel salvage workers by default */
168 handleit(struct cmd_syndesc *as, void *arock)
171 char pname[100], *temp;
172 afs_int32 seenpart = 0, seenvol = 0;
174 struct cmdline_rock *rock = (struct cmdline_rock *)arock;
176 #ifdef AFS_SGI_VNODE_GLUE
177 if (afs_init_kernel_config(-1) < 0) {
179 ("Can't determine NUMA configuration, not starting salvager.\n");
184 if (as->parms[2].items) /* -debug */
186 if (as->parms[3].items) /* -nowrite */
188 if (as->parms[4].items) /* -inodes */
190 if (as->parms[5].items) /* -oktozap */
192 if (as->parms[6].items) /* -rootinodes */
194 if (as->parms[8].items) /* -ForceReads */
196 if ((ti = as->parms[9].items)) { /* -Parallel # */
198 if (strncmp(temp, "all", 3) == 0) {
202 if (strlen(temp) != 0) {
203 Parallel = atoi(temp);
206 if (Parallel > MAXPARALLEL) {
207 printf("Setting parallel salvages to maximum of %d \n",
209 Parallel = MAXPARALLEL;
213 Parallel = MIN(DEFAULT_PARALLELISM, MAXPARALLEL);
215 if ((ti = as->parms[10].items)) { /* -tmpdir */
219 dirp = opendir(tmpdir);
222 ("Can't open temporary placeholder dir %s; using current partition \n",
228 if ((ti = as->parms[11].items)) /* -showlog */
230 if ((ti = as->parms[12].items)) { /* -orphans */
232 orphans = ORPH_IGNORE;
233 else if (strcmp(ti->data, "remove") == 0
234 || strcmp(ti->data, "r") == 0)
235 orphans = ORPH_REMOVE;
236 else if (strcmp(ti->data, "attach") == 0
237 || strcmp(ti->data, "a") == 0)
238 orphans = ORPH_ATTACH;
240 #ifndef AFS_NT40_ENV /* ignore options on NT */
241 if ((ti = as->parms[13].items)) { /* -syslog */
245 if ((ti = as->parms[14].items)) { /* -syslogfacility */
246 useSyslogFacility = atoi(ti->data);
249 if ((ti = as->parms[15].items)) { /* -datelogs */
250 TimeStampLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
254 if ((ti = as->parms[16].items)) { /* -client */
255 if ((ti = as->parms[0].items)) { /* -partition */
257 strlcpy(pname, ti->data, sizeof(pname));
259 if ((ti = as->parms[1].items)) { /* -volumeid */
263 vid_l = strtoul(ti->data, &end, 10);
264 if (vid_l >= MAX_AFS_UINT32 || vid_l == ULONG_MAX || *end != '\0') {
265 printf("Invalid volume id specified; salvage aborted\n");
268 vid = (VolumeId)vid_l;
272 printf("-showlog does not work with -client\n");
276 if (!seenpart || !seenvol) {
277 printf("You must specify '-partition' and '-volumeid' with the '-client' option\n");
281 SalvageClient(vid, pname);
283 } else { /* salvageserver mode */
284 SalvageServer(rock->argc, rock->argv);
291 #include "AFS_component_version_number.c"
295 char *save_args[MAX_ARGS];
297 pthread_t main_thread;
301 main(int argc, char **argv)
303 struct cmd_syndesc *ts;
305 struct cmdline_rock arock;
309 * The following signal action for AIX is necessary so that in case of a
310 * crash (i.e. core is generated) we can include the user's data section
311 * in the core dump. Unfortunately, by default, only a partial core is
312 * generated which, in many cases, isn't too useful.
314 struct sigaction nsa;
316 sigemptyset(&nsa.sa_mask);
317 nsa.sa_handler = SIG_DFL;
318 nsa.sa_flags = SA_FULLDUMP;
319 sigaction(SIGABRT, &nsa, NULL);
320 sigaction(SIGSEGV, &nsa, NULL);
323 /* Initialize directory paths */
324 if (!(initAFSDirPath() & AFSDIR_SERVER_PATHS_OK)) {
326 ReportErrorEventAlt(AFSEVT_SVR_NO_INSTALL_DIR, 0, argv[0], 0);
328 fprintf(stderr, "%s: Unable to obtain AFS server directory.\n",
333 /* Default to binary mode for fopen() */
334 _set_fmode(_O_BINARY);
336 main_thread = pthread_self();
337 if (spawnDatap && spawnDataLen) {
338 /* This is a child per partition salvager. Don't setup log or
339 * try to lock the salvager lock.
341 if (nt_SetupPartitionSalvage(spawnDatap, spawnDataLen) < 0)
347 if (geteuid() != 0) {
348 printf("Salvager must be run as root.\n");
354 /* bad for normal help flag processing, but can do nada */
363 ts = cmd_CreateSyntax("initcmd", handleit, &arock, "initialize the program");
364 cmd_AddParm(ts, "-partition", CMD_SINGLE, CMD_OPTIONAL,
365 "Name of partition to salvage");
366 cmd_AddParm(ts, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
367 "Volume Id to salvage");
368 cmd_AddParm(ts, "-debug", CMD_FLAG, CMD_OPTIONAL,
369 "Run in Debugging mode");
370 cmd_AddParm(ts, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
371 "Run readonly/test mode");
372 cmd_AddParm(ts, "-inodes", CMD_FLAG, CMD_OPTIONAL,
373 "Just list affected afs inodes - debugging flag");
374 cmd_AddParm(ts, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
375 "Give permission to destroy bogus inodes/volumes - debugging flag");
376 cmd_AddParm(ts, "-rootinodes", CMD_FLAG, CMD_OPTIONAL,
377 "Show inodes owned by root - debugging flag");
378 cmd_AddParm(ts, "-salvagedirs", CMD_FLAG, CMD_OPTIONAL,
379 "Force rebuild/salvage of all directories");
380 cmd_AddParm(ts, "-blockreads", CMD_FLAG, CMD_OPTIONAL,
381 "Read smaller blocks to handle IO/bad blocks");
382 cmd_AddParm(ts, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
383 "# of max parallel partition salvaging");
384 cmd_AddParm(ts, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
385 "Name of dir to place tmp files ");
386 cmd_AddParm(ts, "-showlog", CMD_FLAG, CMD_OPTIONAL,
387 "Show log file upon completion");
388 cmd_AddParm(ts, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
389 "ignore | remove | attach");
391 /* note - syslog isn't avail on NT, but if we make it conditional, have
392 * to deal with screwy offsets for cmd params */
393 cmd_AddParm(ts, "-syslog", CMD_FLAG, CMD_OPTIONAL,
394 "Write salvage log to syslogs");
395 cmd_AddParm(ts, "-syslogfacility", CMD_SINGLE, CMD_OPTIONAL,
396 "Syslog facility number to use");
397 cmd_AddParm(ts, "-datelogs", CMD_FLAG, CMD_OPTIONAL,
398 "Include timestamp in logfile filename");
400 cmd_AddParm(ts, "-client", CMD_FLAG, CMD_OPTIONAL,
401 "Use SALVSYNC to ask salvageserver to salvage a volume");
403 err = cmd_Dispatch(argc, argv);
405 return 0; /* not reached */
409 SalvageClient(VolumeId vid, char * pname)
414 SALVSYNC_response_hdr sres;
415 VolumePackageOptions opts;
417 VOptDefaults(volumeUtility, &opts);
418 if (VInitVolumePackage2(volumeUtility, &opts)) {
419 /* VInitVolumePackage2 can fail on e.g. partition attachment errors,
420 * but we don't really care, since all we're doing is trying to use
422 fprintf(stderr, "errors encountered initializing volume package, but "
423 "trying to continue anyway\n");
425 SALVSYNC_clientInit();
427 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_SALVAGE, SALVSYNC_OPERATOR, 0, NULL);
428 if (code != SYNC_OK) {
432 res.payload.buf = (void *) &sres;
433 res.payload.len = sizeof(sres);
437 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_QUERY, SALVSYNC_WHATEVER, 0, &res);
438 if (code != SYNC_OK) {
441 switch (sres.state) {
442 case SALVSYNC_STATE_ERROR:
443 printf("salvageserver reports salvage ended in an error; check log files for more details\n");
444 case SALVSYNC_STATE_DONE:
445 case SALVSYNC_STATE_UNKNOWN:
449 SALVSYNC_clientFinis();
453 if (code == SYNC_DENIED) {
454 printf("salvageserver refused to salvage volume %u on partition %s\n",
456 } else if (code == SYNC_BAD_COMMAND) {
457 printf("SALVSYNC protocol mismatch; please make sure fileserver, volserver, salvageserver and salvager are same version\n");
458 } else if (code == SYNC_COM_ERROR) {
459 printf("SALVSYNC communications error\n");
461 SALVSYNC_clientFinis();
465 static int * child_slot;
468 SalvageServer(int argc, char **argv)
471 struct SalvageQueueNode * node;
473 pthread_attr_t attrs;
475 VolumePackageOptions opts;
477 /* All entries to the log will be appended. Useful if there are
478 * multiple salvagers appending to the log.
481 CheckLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
483 #ifdef AFS_LINUX20_ENV
484 fcntl(fileno(logFile), F_SETFL, O_APPEND); /* Isn't this redundant? */
486 fcntl(fileno(logFile), F_SETFL, FAPPEND); /* Isn't this redundant? */
491 fprintf(logFile, "%s\n", cml_version_number);
492 LogCommandLine(argc, argv, "Online Salvage Server",
493 SalvageVersion, "Starting OpenAFS", Log);
494 /* Get and hold a lock for the duration of the salvage to make sure
495 * that no other salvage runs at the same time. The routine
496 * VInitVolumePackage2 (called below) makes sure that a file server or
497 * other volume utilities don't interfere with the salvage.
500 /* even demand attach online salvager
501 * still needs this because we don't want
502 * a stand-alone salvager to conflict with
503 * the salvager daemon */
504 ObtainSharedSalvageLock();
506 child_slot = (int *) malloc(Parallel * sizeof(int));
507 osi_Assert(child_slot != NULL);
508 memset(child_slot, 0, Parallel * sizeof(int));
510 /* initialize things */
511 VOptDefaults(salvageServer, &opts);
512 if (VInitVolumePackage2(salvageServer, &opts)) {
513 Log("Shutting down: errors encountered initializing volume package\n");
517 queue_Init(&pending_q);
518 queue_Init(&log_cleanup_queue);
519 MUTEX_INIT(&worker_lock, "worker", MUTEX_DEFAULT, 0);
520 CV_INIT(&worker_cv, "worker", CV_DEFAULT, 0);
521 CV_INIT(&log_cleanup_queue.queue_change_cv, "queuechange", CV_DEFAULT, 0);
522 osi_Assert(pthread_attr_init(&attrs) == 0);
524 /* start up the reaper and log cleaner threads */
525 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
526 osi_Assert(pthread_create(&tid,
528 &SalvageChildReaperThread,
530 osi_Assert(pthread_create(&tid,
532 &SalvageLogCleanupThread,
534 osi_Assert(pthread_create(&tid,
536 &SalvageLogScanningThread,
539 /* loop forever serving requests */
541 node = SALVSYNC_getWork();
542 osi_Assert(node != NULL);
544 Log("dispatching child to salvage volume %u...\n",
545 node->command.sop.parent);
549 for (slot = 0; slot < Parallel; slot++) {
550 if (!child_slot[slot])
553 osi_Assert (slot < Parallel);
559 ret = DoSalvageVolume(node, slot);
561 } else if (pid < 0) {
562 Log("failed to fork child worker process\n");
566 child_slot[slot] = pid;
570 MUTEX_ENTER(&worker_lock);
573 /* let the reaper thread know another worker was spawned */
574 CV_BROADCAST(&worker_cv);
576 /* if we're overquota, wait for the reaper */
577 while (current_workers >= Parallel) {
578 CV_WAIT(&worker_cv, &worker_lock);
580 MUTEX_EXIT(&worker_lock);
586 DoSalvageVolume(struct SalvageQueueNode * node, int slot)
588 char childLog[AFSDIR_PATH_MAX];
589 struct DiskPartition64 * partP;
591 /* do not allow further forking inside salvager */
594 /* do not attempt to close parent's logFile handle as
595 * another thread may have held the lock on the FILE
596 * structure when fork was called! */
598 snprintf(childLog, sizeof(childLog), "%s.%d",
599 AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid());
601 logFile = afs_fopen(childLog, "a");
602 if (!logFile) { /* still nothing, use stdout */
607 if (node->command.sop.parent <= 0) {
608 Log("salvageServer: invalid volume id specified; salvage aborted\n");
612 partP = VGetPartition(node->command.sop.partName, 0);
614 Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
615 node->command.sop.partName);
619 /* obtain a shared salvage lock in the child worker, so if the
620 * salvageserver restarts (and we continue), we will still hold a lock and
621 * prevent standalone salvagers from interfering */
622 ObtainSharedSalvageLock();
624 /* Salvage individual volume; don't notify fs */
625 SalvageFileSys1(partP, node->command.sop.parent);
633 SalvageChildReaperThread(void * args)
635 int slot, pid, status;
636 struct log_cleanup_node * cleanup;
638 MUTEX_ENTER(&worker_lock);
640 /* loop reaping our children */
642 /* wait() won't block unless we have children, so
643 * block on the cond var if we're childless */
644 while (current_workers == 0) {
645 CV_WAIT(&worker_cv, &worker_lock);
648 MUTEX_EXIT(&worker_lock);
650 cleanup = (struct log_cleanup_node *) malloc(sizeof(struct log_cleanup_node));
652 while (Reap_Child("salvageserver", &pid, &status) < 0) {
653 /* try to prevent livelock if something goes wrong */
658 for (slot = 0; slot < Parallel; slot++) {
659 if (child_slot[slot] == pid)
662 osi_Assert(slot < Parallel);
663 child_slot[slot] = 0;
666 SALVSYNC_doneWorkByPid(pid, status);
668 MUTEX_ENTER(&worker_lock);
672 queue_Append(&log_cleanup_queue, cleanup);
673 CV_SIGNAL(&log_cleanup_queue.queue_change_cv);
676 /* ok, we've reaped a child */
678 CV_BROADCAST(&worker_cv);
685 Reap_Child(char *prog, int * pid, int * status)
692 if (WCOREDUMP(*status))
693 Log("\"%s\" core dumped!\n", prog);
694 if ((WIFSIGNALED(*status) != 0) ||
695 ((WEXITSTATUS(*status) != 0) &&
696 (WEXITSTATUS(*status) != SALSRV_EXIT_VOLGROUP_LINK)))
697 Log("\"%s\" (pid=%d) terminated abnormally!\n", prog, ret);
699 Log("wait returned -1\n");
705 * thread to combine salvager child logs
706 * back into the main salvageserver log
709 SalvageLogCleanupThread(void * arg)
711 struct log_cleanup_node * cleanup;
713 MUTEX_ENTER(&worker_lock);
716 while (queue_IsEmpty(&log_cleanup_queue)) {
717 CV_WAIT(&log_cleanup_queue.queue_change_cv, &worker_lock);
720 while (queue_IsNotEmpty(&log_cleanup_queue)) {
721 cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
722 queue_Remove(cleanup);
723 MUTEX_EXIT(&worker_lock);
724 SalvageLogCleanup(cleanup->pid);
726 MUTEX_ENTER(&worker_lock);
730 MUTEX_EXIT(&worker_lock);
734 #define LOG_XFER_BUF_SIZE 65536
736 SalvageLogCleanup(int pid)
739 char fn[AFSDIR_PATH_MAX];
740 static char buf[LOG_XFER_BUF_SIZE];
742 snprintf(fn, sizeof(fn), "%s.%d",
743 AFSDIR_SERVER_SLVGLOG_FILEPATH, pid);
746 pidlog = open(fn, O_RDONLY);
751 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
753 fwrite(buf, len, 1, logFile);
754 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
762 /* wake up every five minutes to see if a non-child salvage has finished */
763 #define SALVAGE_SCAN_POLL_INTERVAL 300
766 * Thread to look for SalvageLog.$pid files that are not from our child
767 * worker salvagers, and notify SalvageLogCleanupThread to clean them
768 * up. This can happen if we restart during salvages, or the
769 * salvageserver crashes or something.
773 * @return always NULL
776 SalvageLogScanningThread(void * arg)
778 struct rx_queue log_watch_queue;
780 queue_Init(&log_watch_queue);
785 char prefix[AFSDIR_PATH_MAX];
788 snprintf(prefix, sizeof(prefix), "%s.", AFSDIR_SLVGLOG_FILE);
789 prefix_len = strlen(prefix);
791 dp = opendir(AFSDIR_LOGS_DIR);
794 while ((dirp = readdir(dp)) != NULL) {
796 struct log_cleanup_node *cleanup;
799 if (strncmp(dirp->d_name, prefix, prefix_len) != 0) {
800 /* not a salvage logfile; skip */
805 pid = strtol(dirp->d_name + prefix_len, NULL, 10);
808 /* file is SalvageLog.<something> but <something> isn't
814 for (i = 0; i < Parallel; ++i) {
815 if (pid == child_slot[i]) {
821 /* this pid is one of our children, so the reaper thread
822 * will take care of it; skip */
827 (struct log_cleanup_node *) malloc(sizeof(struct log_cleanup_node));
830 queue_Append(&log_watch_queue, cleanup);
836 ScanLogs(&log_watch_queue);
838 while (queue_IsNotEmpty(&log_watch_queue)) {
839 sleep(SALVAGE_SCAN_POLL_INTERVAL);
840 ScanLogs(&log_watch_queue);
847 * look through log_watch_queue, and if any processes are not still
848 * running, hand them off to the SalvageLogCleanupThread
850 * @param log_watch_queue a queue of PIDs that we should clean up if
854 ScanLogs(struct rx_queue *log_watch_queue)
856 struct log_cleanup_node *cleanup, *next;
858 MUTEX_ENTER(&worker_lock);
860 for (queue_Scan(log_watch_queue, cleanup, next, log_cleanup_node)) {
861 /* if a process is still running, assume it's the salvage process
862 * still going, and keep waiting for it */
863 if (kill(cleanup->pid, 0) < 0 && errno == ESRCH) {
864 queue_Remove(cleanup);
865 queue_Append(&log_cleanup_queue, cleanup);
866 CV_SIGNAL(&log_cleanup_queue.queue_change_cv);
870 MUTEX_EXIT(&worker_lock);