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)
37 #include <afs/afsint.h>
38 #include <rx/rx_queue.h>
40 #if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
41 #if defined(AFS_VFSINCL_ENV)
42 #include <sys/vnode.h>
44 #include <sys/fs/ufs_inode.h>
46 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
47 #include <ufs/ufs/dinode.h>
48 #include <ufs/ffs/fs.h>
50 #include <ufs/inode.h>
53 #else /* AFS_VFSINCL_ENV */
54 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
55 #include <sys/inode.h>
57 #endif /* AFS_VFSINCL_ENV */
58 #endif /* AFS_SGI_ENV */
61 #include <sys/lockf.h>
64 #include <checklist.h>
66 #if defined(AFS_SGI_ENV)
69 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
71 #include <sys/mnttab.h>
72 #include <sys/mntent.h>
77 #endif /* AFS_SGI_ENV */
78 #endif /* AFS_HPUX_ENV */
82 #include <afs/osi_inode.h>
85 #include <afs/afsutil.h>
86 #include <afs/fileutil.h>
87 #include <afs/procmgmt.h> /* signal(), kill(), wait(), etc. */
93 #include <afs/afssyscalls.h>
97 #include "partition.h"
98 #include "daemon_com.h"
100 #include "salvsync.h"
101 #include "viceinode.h"
103 #include "vol-salvage.h"
109 extern int ClientMode;
111 #if !defined(AFS_DEMAND_ATTACH_FS)
112 #error "online salvager only supported for demand attach fileserver"
113 #endif /* AFS_DEMAND_ATTACH_FS */
115 #if defined(AFS_NT40_ENV)
116 #error "online salvager not supported on NT"
117 #endif /* AFS_NT40_ENV */
119 /*@+fcnmacros +macrofcndecl@*/
121 #define afs_fopen fopen64
122 #else /* !O_LARGEFILE */
123 #define afs_fopen fopen
124 #endif /* !O_LARGEFILE */
125 /*@=fcnmacros =macrofcndecl@*/
129 static volatile int current_workers = 0;
130 static volatile struct rx_queue pending_q;
131 static pthread_mutex_t worker_lock;
132 static pthread_cond_t worker_cv;
134 static void * SalvageChildReaperThread(void *);
135 static int DoSalvageVolume(struct SalvageQueueNode * node, int slot);
137 static void SalvageServer(int argc, char **argv, struct logOptions *logopts);
138 static void SalvageClient(VolumeId vid, char * pname);
140 static int Reap_Child(char * prog, int * pid, int * status);
142 static void * SalvageLogCleanupThread(void *);
143 static void SalvageLogCleanup(int pid);
145 static void * SalvageLogScanningThread(void *);
146 static void ScanLogs(struct rx_queue *log_watch_queue);
148 struct cmdline_rock {
153 struct log_cleanup_node {
159 struct rx_queue queue_head;
160 pthread_cond_t queue_change_cv;
164 #define DEFAULT_PARALLELISM 4 /* allow 4 parallel salvage workers by default */
187 handleit(struct cmd_syndesc *opts, void *arock)
190 afs_int32 seenpart = 0, seenvol = 0;
192 struct cmdline_rock *rock = (struct cmdline_rock *)arock;
193 char *optstring = NULL;
194 struct logOptions logopts;
196 memset(&logopts, 0, sizeof(logopts));
198 #ifdef AFS_SGI_VNODE_GLUE
199 if (afs_init_kernel_config(-1) < 0) {
201 ("Can't determine NUMA configuration, not starting salvager.\n");
206 cmd_OptionAsFlag(opts, OPT_debug, &debug);
207 cmd_OptionAsFlag(opts, OPT_nowrite, &Testing);
208 cmd_OptionAsFlag(opts, OPT_inodes, &ListInodeOption);
209 cmd_OptionAsFlag(opts, OPT_oktozap, &OKToZap);
210 cmd_OptionAsFlag(opts, OPT_rootinodes, &ShowRootFiles);
211 cmd_OptionAsFlag(opts, OPT_salvagedirs, &RebuildDirs);
212 cmd_OptionAsFlag(opts, OPT_blockreads, &forceR);
213 if (cmd_OptionAsString(opts, OPT_parallel, &optstring) == 0) {
214 char *input = optstring;
215 if (strncmp(optstring, "all", 3) == 0) {
219 if (strlen(input) != 0) {
220 Parallel = atoi(input);
223 if (Parallel > MAXPARALLEL) {
224 printf("Setting parallel salvages to maximum of %d \n",
226 Parallel = MAXPARALLEL;
232 Parallel = min(DEFAULT_PARALLELISM, MAXPARALLEL);
234 if (cmd_OptionAsString(opts, OPT_tmpdir, &optstring) == 0) {
236 dirp = opendir(optstring);
239 ("Can't open temporary placeholder dir %s; using current partition \n",
247 if (cmd_OptionAsString(opts, OPT_orphans, &optstring) == 0) {
249 orphans = ORPH_IGNORE;
250 else if (strcmp(optstring, "remove") == 0
251 || strcmp(optstring, "r") == 0)
252 orphans = ORPH_REMOVE;
253 else if (strcmp(optstring, "attach") == 0
254 || strcmp(optstring, "a") == 0)
255 orphans = ORPH_ATTACH;
261 if (cmd_OptionPresent(opts, OPT_syslog)) {
262 if (cmd_OptionPresent(opts, OPT_logfile)) {
263 fprintf(stderr, "Invalid options: -syslog and -logfile are exclusive.\n");
266 if (cmd_OptionPresent(opts, OPT_transarc_logs)) {
267 fprintf(stderr, "Invalid options: -syslog and -transarc-logs are exclusive.\n");
270 logopts.lopt_dest = logDest_syslog;
271 logopts.lopt_facility = LOG_DAEMON;
272 logopts.lopt_tag = "salvageserver";
273 cmd_OptionAsInt(opts, OPT_syslogfacility, &logopts.lopt_facility);
277 logopts.lopt_dest = logDest_file;
278 if (cmd_OptionPresent(opts, OPT_transarc_logs)) {
279 logopts.lopt_rotateOnOpen = 1;
280 logopts.lopt_rotateStyle = logRotate_old;
282 if (cmd_OptionPresent(opts, OPT_logfile))
283 cmd_OptionAsString(opts, OPT_logfile, (char**)&logopts.lopt_filename);
285 logopts.lopt_filename = AFSDIR_SERVER_SALSRVLOG_FILEPATH;
288 if (cmd_OptionPresent(opts, OPT_client)) {
289 if (cmd_OptionAsString(opts, OPT_partition, &optstring) == 0) {
291 strlcpy(pname, optstring, sizeof(pname));
295 if (cmd_OptionAsString(opts, OPT_volumeid, &optstring) == 0) {
299 vid_l = strtoul(optstring, &end, 10);
300 if (vid_l >= MAX_AFS_UINT32 || vid_l == ULONG_MAX || *end != '\0') {
301 printf("Invalid volume id specified; salvage aborted\n");
304 vid = (VolumeId)vid_l;
307 if (!seenpart || !seenvol) {
308 printf("You must specify '-partition' and '-volumeid' with the '-client' option\n");
312 SalvageClient(vid, pname);
314 } else { /* salvageserver mode */
315 SalvageServer(rock->argc, rock->argv, &logopts);
322 #include "AFS_component_version_number.c"
326 char *save_args[MAX_ARGS];
328 pthread_t main_thread;
332 main(int argc, char **argv)
334 struct cmd_syndesc *ts;
336 struct cmdline_rock arock;
340 * The following signal action for AIX is necessary so that in case of a
341 * crash (i.e. core is generated) we can include the user's data section
342 * in the core dump. Unfortunately, by default, only a partial core is
343 * generated which, in many cases, isn't too useful.
345 struct sigaction nsa;
347 sigemptyset(&nsa.sa_mask);
348 nsa.sa_handler = SIG_DFL;
349 nsa.sa_flags = SA_FULLDUMP;
350 sigaction(SIGABRT, &nsa, NULL);
351 sigaction(SIGSEGV, &nsa, NULL);
354 /* Initialize directory paths */
355 if (!(initAFSDirPath() & AFSDIR_SERVER_PATHS_OK)) {
357 ReportErrorEventAlt(AFSEVT_SVR_NO_INSTALL_DIR, 0, argv[0], 0);
359 fprintf(stderr, "%s: Unable to obtain AFS server directory.\n",
364 /* Default to binary mode for fopen() */
365 _set_fmode(_O_BINARY);
367 main_thread = pthread_self();
368 if (spawnDatap && spawnDataLen) {
369 /* This is a child per partition salvager. Don't setup log or
370 * try to lock the salvager lock.
372 if (nt_SetupPartitionSalvage(spawnDatap, spawnDataLen) < 0)
378 if (geteuid() != 0) {
379 printf("Salvager must be run as root.\n");
385 /* bad for normal help flag processing, but can do nada */
395 ts = cmd_CreateSyntax("initcmd", handleit, &arock, 0, "initialize the program");
396 cmd_AddParmAtOffset(ts, OPT_partition, "-partition", CMD_SINGLE,
397 CMD_OPTIONAL, "Name of partition to salvage");
398 cmd_AddParmAtOffset(ts, OPT_volumeid, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
399 "Volume Id to salvage");
400 cmd_AddParmAtOffset(ts, OPT_debug, "-debug", CMD_FLAG, CMD_OPTIONAL,
401 "Run in Debugging mode");
402 cmd_AddParmAtOffset(ts, OPT_nowrite, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
403 "Run readonly/test mode");
404 cmd_AddParmAtOffset(ts, OPT_inodes, "-inodes", CMD_FLAG, CMD_OPTIONAL,
405 "Just list affected afs inodes - debugging flag");
406 cmd_AddParmAtOffset(ts, OPT_oktozap, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
407 "Give permission to destroy bogus inodes/volumes - debugging flag");
408 cmd_AddParmAtOffset(ts, OPT_rootinodes, "-rootinodes", CMD_FLAG,
409 CMD_OPTIONAL, "Show inodes owned by root - debugging flag");
410 cmd_AddParmAtOffset(ts, OPT_salvagedirs, "-salvagedirs", CMD_FLAG,
411 CMD_OPTIONAL, "Force rebuild/salvage of all directories");
412 cmd_AddParmAtOffset(ts, OPT_blockreads, "-blockreads", CMD_FLAG,
413 CMD_OPTIONAL, "Read smaller blocks to handle IO/bad blocks");
414 cmd_AddParmAtOffset(ts, OPT_parallel, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
415 "# of max parallel partition salvaging");
416 cmd_AddParmAtOffset(ts, OPT_tmpdir, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
417 "Name of dir to place tmp files ");
418 cmd_AddParmAtOffset(ts, OPT_orphans, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
419 "ignore | remove | attach");
422 cmd_AddParmAtOffset(ts, OPT_syslog, "-syslog", CMD_FLAG, CMD_OPTIONAL,
423 "Write salvage log to syslogs");
424 cmd_AddParmAtOffset(ts, OPT_syslogfacility, "-syslogfacility", CMD_SINGLE,
425 CMD_OPTIONAL, "Syslog facility number to use");
428 cmd_AddParmAtOffset(ts, OPT_client, "-client", CMD_FLAG, CMD_OPTIONAL,
429 "Use SALVSYNC to ask salvageserver to salvage a volume");
431 cmd_AddParmAtOffset(ts, OPT_logfile, "-logfile", CMD_SINGLE, CMD_OPTIONAL,
432 "Location of log file ");
434 cmd_AddParmAtOffset(ts, OPT_transarc_logs, "-transarc-logs", CMD_FLAG,
435 CMD_OPTIONAL, "enable Transarc style logging");
437 err = cmd_Dispatch(argc, argv);
439 AFS_UNREACHED(return 0);
443 SalvageClient(VolumeId vid, char * pname)
448 SALVSYNC_response_hdr sres;
449 VolumePackageOptions opts;
451 /* Send Log() messages to stderr in client mode. */
454 VOptDefaults(volumeUtility, &opts);
455 if (VInitVolumePackage2(volumeUtility, &opts)) {
456 /* VInitVolumePackage2 can fail on e.g. partition attachment errors,
457 * but we don't really care, since all we're doing is trying to use
459 fprintf(stderr, "errors encountered initializing volume package, but "
460 "trying to continue anyway\n");
462 SALVSYNC_clientInit();
464 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_SALVAGE, SALVSYNC_OPERATOR, 0, NULL);
465 if (code != SYNC_OK) {
469 res.payload.buf = (void *) &sres;
470 res.payload.len = sizeof(sres);
474 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_QUERY, SALVSYNC_WHATEVER, 0, &res);
475 if (code != SYNC_OK) {
478 switch (sres.state) {
479 case SALVSYNC_STATE_ERROR:
480 printf("salvageserver reports salvage ended in an error; check log files for more details\n");
482 case SALVSYNC_STATE_DONE:
483 case SALVSYNC_STATE_UNKNOWN:
487 SALVSYNC_clientFinis();
491 if (code == SYNC_DENIED) {
492 printf("salvageserver refused to salvage volume %u on partition %s\n",
494 } else if (code == SYNC_BAD_COMMAND) {
495 printf("SALVSYNC protocol mismatch; please make sure fileserver, volserver, salvageserver and salvager are same version\n");
496 } else if (code == SYNC_COM_ERROR) {
497 printf("SALVSYNC communications error\n");
499 SALVSYNC_clientFinis();
503 static int * child_slot;
506 SalvageServer(int argc, char **argv, struct logOptions *logopts)
509 struct SalvageQueueNode * node;
511 pthread_attr_t attrs;
513 VolumePackageOptions opts;
515 /* All entries to the log will be appended. Useful if there are
516 * multiple salvagers appending to the log.
521 Log("%s\n", cml_version_number);
522 LogCommandLine(argc, argv, "Online Salvage Server",
523 SalvageVersion, "Starting OpenAFS", Log);
524 /* Get and hold a lock for the duration of the salvage to make sure
525 * that no other salvage runs at the same time. The routine
526 * VInitVolumePackage2 (called below) makes sure that a file server or
527 * other volume utilities don't interfere with the salvage.
530 /* even demand attach online salvager
531 * still needs this because we don't want
532 * a stand-alone salvager to conflict with
533 * the salvager daemon */
534 ObtainSharedSalvageLock();
536 child_slot = calloc(Parallel, sizeof(int));
537 opr_Assert(child_slot != NULL);
539 /* initialize things */
540 VOptDefaults(salvageServer, &opts);
541 if (VInitVolumePackage2(salvageServer, &opts)) {
542 Log("Shutting down: errors encountered initializing volume package\n");
546 queue_Init(&pending_q);
547 queue_Init(&log_cleanup_queue);
548 opr_mutex_init(&worker_lock);
549 opr_cv_init(&worker_cv);
550 opr_cv_init(&log_cleanup_queue.queue_change_cv);
551 opr_Verify(pthread_attr_init(&attrs) == 0);
553 /* start up the reaper and log cleaner threads */
554 opr_Verify(pthread_attr_setdetachstate(&attrs,
555 PTHREAD_CREATE_DETACHED) == 0);
556 opr_Verify(pthread_create(&tid, &attrs,
557 &SalvageChildReaperThread, NULL) == 0);
558 opr_Verify(pthread_create(&tid, &attrs,
559 &SalvageLogCleanupThread, NULL) == 0);
560 opr_Verify(pthread_create(&tid, &attrs,
561 &SalvageLogScanningThread, NULL) == 0);
563 /* loop forever serving requests */
565 node = SALVSYNC_getWork();
566 opr_Assert(node != NULL);
568 Log("dispatching child to salvage volume %u...\n",
569 node->command.sop.parent);
573 for (slot = 0; slot < Parallel; slot++) {
574 if (!child_slot[slot])
577 opr_Assert (slot < Parallel);
583 ret = DoSalvageVolume(node, slot);
585 } else if (pid < 0) {
586 Log("failed to fork child worker process\n");
590 child_slot[slot] = pid;
594 opr_mutex_enter(&worker_lock);
597 /* let the reaper thread know another worker was spawned */
598 opr_cv_broadcast(&worker_cv);
600 /* if we're overquota, wait for the reaper */
601 while (current_workers >= Parallel) {
602 opr_cv_wait(&worker_cv, &worker_lock);
604 opr_mutex_exit(&worker_lock);
610 DoSalvageVolume(struct SalvageQueueNode * node, int slot)
612 char *filename = NULL;
613 struct logOptions logopts;
614 struct DiskPartition64 * partP;
616 /* do not allow further forking inside salvager */
620 * Do not attempt to close parent's log file handle as
621 * another thread may have held the lock when fork was
624 memset(&logopts, 0, sizeof(logopts));
625 logopts.lopt_dest = logDest_file;
626 logopts.lopt_rotateStyle = logRotate_none;
627 if (asprintf(&filename, "%s.%d",
628 AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid()) < 0) {
629 fprintf(stderr, "out of memory\n");
632 logopts.lopt_filename = filename;
636 if (node->command.sop.parent <= 0) {
637 Log("salvageServer: invalid volume id specified; salvage aborted\n");
641 partP = VGetPartition(node->command.sop.partName, 0);
643 Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
644 node->command.sop.partName);
648 /* obtain a shared salvage lock in the child worker, so if the
649 * salvageserver restarts (and we continue), we will still hold a lock and
650 * prevent standalone salvagers from interfering */
651 ObtainSharedSalvageLock();
653 /* Salvage individual volume; don't notify fs */
654 SalvageFileSys1(partP, node->command.sop.parent);
662 SalvageChildReaperThread(void * args)
664 int slot, pid, status;
665 struct log_cleanup_node * cleanup;
667 opr_mutex_enter(&worker_lock);
669 /* loop reaping our children */
671 /* wait() won't block unless we have children, so
672 * block on the cond var if we're childless */
673 while (current_workers == 0) {
674 opr_cv_wait(&worker_cv, &worker_lock);
677 opr_mutex_exit(&worker_lock);
679 cleanup = malloc(sizeof(struct log_cleanup_node));
681 while (Reap_Child("salvageserver", &pid, &status) < 0) {
682 /* try to prevent livelock if something goes wrong */
687 for (slot = 0; slot < Parallel; slot++) {
688 if (child_slot[slot] == pid)
691 opr_Assert(slot < Parallel);
692 child_slot[slot] = 0;
695 SALVSYNC_doneWorkByPid(pid, status);
697 opr_mutex_enter(&worker_lock);
701 queue_Append(&log_cleanup_queue, cleanup);
702 opr_cv_signal(&log_cleanup_queue.queue_change_cv);
705 /* ok, we've reaped a child */
707 opr_cv_broadcast(&worker_cv);
710 AFS_UNREACHED(return(NULL));
714 Reap_Child(char *prog, int * pid, int * status)
721 if (WCOREDUMP(*status))
722 Log("\"%s\" core dumped!\n", prog);
723 if ((WIFSIGNALED(*status) != 0) ||
724 ((WEXITSTATUS(*status) != 0) &&
725 (WEXITSTATUS(*status) != SALSRV_EXIT_VOLGROUP_LINK)))
726 Log("\"%s\" (pid=%d) terminated abnormally!\n", prog, ret);
728 Log("wait returned -1\n");
734 * thread to combine salvager child logs
735 * back into the main salvageserver log
738 SalvageLogCleanupThread(void * arg)
740 struct log_cleanup_node * cleanup;
742 opr_mutex_enter(&worker_lock);
745 while (queue_IsEmpty(&log_cleanup_queue)) {
746 opr_cv_wait(&log_cleanup_queue.queue_change_cv, &worker_lock);
749 while (queue_IsNotEmpty(&log_cleanup_queue)) {
750 cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
751 queue_Remove(cleanup);
752 opr_mutex_exit(&worker_lock);
753 SalvageLogCleanup(cleanup->pid);
755 opr_mutex_enter(&worker_lock);
759 AFS_UNREACHED(opr_mutex_exit(&worker_lock));
760 AFS_UNREACHED(return(NULL));
763 #define LOG_XFER_BUF_SIZE 65536
765 SalvageLogCleanup(int pid)
771 if (asprintf(&fn, "%s.%d", AFSDIR_SERVER_SLVGLOG_FILEPATH, pid) < 0) {
772 Log("Unable to write child log: out of memory\n");
776 buf = calloc(1, LOG_XFER_BUF_SIZE);
778 Log("Unable to write child log: out of memory\n");
782 pidlog = open(fn, O_RDONLY);
787 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
789 WriteLogBuffer(buf, len);
790 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
800 /* wake up every five minutes to see if a non-child salvage has finished */
801 #define SALVAGE_SCAN_POLL_INTERVAL 300
804 * Thread to look for SalvageLog.$pid files that are not from our child
805 * worker salvagers, and notify SalvageLogCleanupThread to clean them
806 * up. This can happen if we restart during salvages, or the
807 * salvageserver crashes or something.
811 * @return always NULL
814 SalvageLogScanningThread(void * arg)
816 struct rx_queue log_watch_queue;
820 queue_Init(&log_watch_queue);
822 prefix_len = asprintf(&prefix, "%s.", AFSDIR_SLVGLOG_FILE);
823 if (prefix_len >= 0) {
827 dp = opendir(AFSDIR_LOGS_DIR);
830 while ((dirp = readdir(dp)) != NULL) {
832 struct log_cleanup_node *cleanup;
835 if (strncmp(dirp->d_name, prefix, prefix_len) != 0) {
836 /* not a salvage logfile; skip */
841 pid = strtol(dirp->d_name + prefix_len, NULL, 10);
844 /* file is SalvageLog.<something> but <something> isn't
850 for (i = 0; i < Parallel; ++i) {
851 if (pid == child_slot[i]) {
857 /* this pid is one of our children, so the reaper thread
858 * will take care of it; skip */
862 cleanup = malloc(sizeof(struct log_cleanup_node));
865 queue_Append(&log_watch_queue, cleanup);
871 ScanLogs(&log_watch_queue);
873 while (queue_IsNotEmpty(&log_watch_queue)) {
874 sleep(SALVAGE_SCAN_POLL_INTERVAL);
875 ScanLogs(&log_watch_queue);
882 * look through log_watch_queue, and if any processes are not still
883 * running, hand them off to the SalvageLogCleanupThread
885 * @param log_watch_queue a queue of PIDs that we should clean up if
889 ScanLogs(struct rx_queue *log_watch_queue)
891 struct log_cleanup_node *cleanup, *next;
893 opr_mutex_enter(&worker_lock);
895 for (queue_Scan(log_watch_queue, cleanup, next, log_cleanup_node)) {
896 /* if a process is still running, assume it's the salvage process
897 * still going, and keep waiting for it */
898 if (kill(cleanup->pid, 0) < 0 && errno == ESRCH) {
899 queue_Remove(cleanup);
900 queue_Append(&log_cleanup_queue, cleanup);
901 opr_cv_signal(&log_cleanup_queue.queue_change_cv);
905 opr_mutex_exit(&worker_lock);