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 */
55 #include <ufs/inode.h>
56 #else /* AFS_OSF_ENV */
57 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
58 #include <sys/inode.h>
61 #endif /* AFS_VFSINCL_ENV */
62 #endif /* AFS_SGI_ENV */
65 #include <sys/lockf.h>
68 #include <checklist.h>
70 #if defined(AFS_SGI_ENV)
73 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
75 #include <sys/mnttab.h>
76 #include <sys/mntent.h>
81 #endif /* AFS_SGI_ENV */
82 #endif /* AFS_HPUX_ENV */
86 #include <afs/osi_inode.h>
89 #include <afs/afsutil.h>
90 #include <afs/fileutil.h>
91 #include <afs/procmgmt.h> /* signal(), kill(), wait(), etc. */
97 #include <afs/afssyscalls.h>
101 #include "partition.h"
102 #include "daemon_com.h"
104 #include "salvsync.h"
105 #include "viceinode.h"
107 #include "vol-salvage.h"
113 char *logFileName = NULL;
115 #if !defined(AFS_DEMAND_ATTACH_FS)
116 #error "online salvager only supported for demand attach fileserver"
117 #endif /* AFS_DEMAND_ATTACH_FS */
119 #if defined(AFS_NT40_ENV)
120 #error "online salvager not supported on NT"
121 #endif /* AFS_NT40_ENV */
123 /*@+fcnmacros +macrofcndecl@*/
125 #define afs_fopen fopen64
126 #else /* !O_LARGEFILE */
127 #define afs_fopen fopen
128 #endif /* !O_LARGEFILE */
129 /*@=fcnmacros =macrofcndecl@*/
133 static volatile int current_workers = 0;
134 static volatile struct rx_queue pending_q;
135 static pthread_mutex_t worker_lock;
136 static pthread_cond_t worker_cv;
138 static void * SalvageChildReaperThread(void *);
139 static int DoSalvageVolume(struct SalvageQueueNode * node, int slot);
141 static void SalvageServer(int argc, char **argv);
142 static void SalvageClient(VolumeId vid, char * pname);
144 static int Reap_Child(char * prog, int * pid, int * status);
146 static void * SalvageLogCleanupThread(void *);
147 static int SalvageLogCleanup(int pid);
149 static void * SalvageLogScanningThread(void *);
150 static void ScanLogs(struct rx_queue *log_watch_queue);
152 struct cmdline_rock {
157 struct log_cleanup_node {
163 struct rx_queue queue_head;
164 pthread_cond_t queue_change_cv;
168 #define DEFAULT_PARALLELISM 4 /* allow 4 parallel salvage workers by default */
192 handleit(struct cmd_syndesc *opts, void *arock)
195 afs_int32 seenpart = 0, seenvol = 0;
197 struct cmdline_rock *rock = (struct cmdline_rock *)arock;
198 char *optstring = NULL;
200 #ifdef AFS_SGI_VNODE_GLUE
201 if (afs_init_kernel_config(-1) < 0) {
203 ("Can't determine NUMA configuration, not starting salvager.\n");
208 cmd_OptionAsFlag(opts, OPT_debug, &debug);
209 cmd_OptionAsFlag(opts, OPT_nowrite, &Testing);
210 cmd_OptionAsFlag(opts, OPT_inodes, &ListInodeOption);
211 cmd_OptionAsFlag(opts, OPT_oktozap, &OKToZap);
212 cmd_OptionAsFlag(opts, OPT_rootinodes, &ShowRootFiles);
213 cmd_OptionAsFlag(opts, OPT_blockreads, &forceR);
214 if (cmd_OptionAsString(opts, OPT_parallel, &optstring) == 0) {
215 if (strncmp(optstring, "all", 3) == 0) {
218 if (strlen(optstring) != 0) {
219 Parallel = atoi(optstring);
222 if (Parallel > MAXPARALLEL) {
223 printf("Setting parallel salvages to maximum of %d \n",
225 Parallel = MAXPARALLEL;
231 Parallel = min(DEFAULT_PARALLELISM, MAXPARALLEL);
233 if (cmd_OptionAsString(opts, OPT_tmpdir, &optstring) == 0) {
235 dirp = opendir(optstring);
238 ("Can't open temporary placeholder dir %s; using current partition \n",
246 cmd_OptionAsFlag(opts, OPT_showlog, &ShowLog);
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;
259 #ifndef AFS_NT40_ENV /* ignore options on NT */
260 if (cmd_OptionPresent(opts, OPT_syslog)) {
264 cmd_OptionAsInt(opts, OPT_syslogfacility, &useSyslogFacility);
266 if (cmd_OptionPresent(opts, OPT_datelogs)) {
267 TimeStampLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
271 if (cmd_OptionPresent(opts, OPT_client)) {
272 if (cmd_OptionAsString(opts, OPT_partition, &optstring) == 0) {
274 strlcpy(pname, optstring, sizeof(pname));
278 if (cmd_OptionAsString(opts, OPT_volumeid, &optstring) == 0) {
282 vid_l = strtoul(optstring, &end, 10);
283 if (vid_l >= MAX_AFS_UINT32 || vid_l == ULONG_MAX || *end != '\0') {
284 printf("Invalid volume id specified; salvage aborted\n");
287 vid = (VolumeId)vid_l;
291 printf("-showlog does not work with -client\n");
295 if (!seenpart || !seenvol) {
296 printf("You must specify '-partition' and '-volumeid' with the '-client' option\n");
300 SalvageClient(vid, pname);
302 } else { /* salvageserver mode */
303 SalvageServer(rock->argc, rock->argv);
310 #include "AFS_component_version_number.c"
314 char *save_args[MAX_ARGS];
316 pthread_t main_thread;
320 main(int argc, char **argv)
322 struct cmd_syndesc *ts;
324 struct cmdline_rock arock;
328 * The following signal action for AIX is necessary so that in case of a
329 * crash (i.e. core is generated) we can include the user's data section
330 * in the core dump. Unfortunately, by default, only a partial core is
331 * generated which, in many cases, isn't too useful.
333 struct sigaction nsa;
335 sigemptyset(&nsa.sa_mask);
336 nsa.sa_handler = SIG_DFL;
337 nsa.sa_flags = SA_FULLDUMP;
338 sigaction(SIGABRT, &nsa, NULL);
339 sigaction(SIGSEGV, &nsa, NULL);
342 /* Initialize directory paths */
343 if (!(initAFSDirPath() & AFSDIR_SERVER_PATHS_OK)) {
345 ReportErrorEventAlt(AFSEVT_SVR_NO_INSTALL_DIR, 0, argv[0], 0);
347 fprintf(stderr, "%s: Unable to obtain AFS server directory.\n",
352 /* Default to binary mode for fopen() */
353 _set_fmode(_O_BINARY);
355 main_thread = pthread_self();
356 if (spawnDatap && spawnDataLen) {
357 /* This is a child per partition salvager. Don't setup log or
358 * try to lock the salvager lock.
360 if (nt_SetupPartitionSalvage(spawnDatap, spawnDataLen) < 0)
366 if (geteuid() != 0) {
367 printf("Salvager must be run as root.\n");
373 /* bad for normal help flag processing, but can do nada */
382 logFileName = strdup(AFSDIR_SERVER_SALSRVLOG_FILEPATH);
384 ts = cmd_CreateSyntax("initcmd", handleit, &arock, "initialize the program");
385 cmd_AddParmAtOffset(ts, OPT_partition, "-partition", CMD_SINGLE,
386 CMD_OPTIONAL, "Name of partition to salvage");
387 cmd_AddParmAtOffset(ts, OPT_volumeid, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
388 "Volume Id to salvage");
389 cmd_AddParmAtOffset(ts, OPT_debug, "-debug", CMD_FLAG, CMD_OPTIONAL,
390 "Run in Debugging mode");
391 cmd_AddParmAtOffset(ts, OPT_nowrite, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
392 "Run readonly/test mode");
393 cmd_AddParmAtOffset(ts, OPT_inodes, "-inodes", CMD_FLAG, CMD_OPTIONAL,
394 "Just list affected afs inodes - debugging flag");
395 cmd_AddParmAtOffset(ts, OPT_oktozap, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
396 "Give permission to destroy bogus inodes/volumes - debugging flag");
397 cmd_AddParmAtOffset(ts, OPT_rootinodes, "-rootinodes", CMD_FLAG,
398 CMD_OPTIONAL, "Show inodes owned by root - debugging flag");
399 cmd_AddParmAtOffset(ts, OPT_salvagedirs, "-salvagedirs", CMD_FLAG,
400 CMD_OPTIONAL, "Force rebuild/salvage of all directories");
401 cmd_AddParmAtOffset(ts, OPT_blockreads, "-blockreads", CMD_FLAG,
402 CMD_OPTIONAL, "Read smaller blocks to handle IO/bad blocks");
403 cmd_AddParmAtOffset(ts, OPT_parallel, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
404 "# of max parallel partition salvaging");
405 cmd_AddParmAtOffset(ts, OPT_tmpdir, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
406 "Name of dir to place tmp files ");
407 cmd_AddParmAtOffset(ts, OPT_showlog, "-showlog", CMD_FLAG, CMD_OPTIONAL,
408 "Show log file upon completion");
409 cmd_AddParmAtOffset(ts, OPT_orphans, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
410 "ignore | remove | attach");
412 #if !defined(AFS_NT40_ENV)
413 cmd_AddParmAtOffset(ts, OPT_syslog, "-syslog", CMD_FLAG, CMD_OPTIONAL,
414 "Write salvage log to syslogs");
415 cmd_AddParmAtOffset(ts, OPT_syslogfacility, "-syslogfacility", CMD_SINGLE,
416 CMD_OPTIONAL, "Syslog facility number to use");
417 cmd_AddParmAtOffset(ts, OPT_datelogs, "-datelogs", CMD_FLAG, CMD_OPTIONAL,
418 "Include timestamp in logfile filename");
421 cmd_AddParmAtOffset(ts, OPT_client, "-client", CMD_FLAG, CMD_OPTIONAL,
422 "Use SALVSYNC to ask salvageserver to salvage a volume");
424 cmd_AddParmAtOffset(ts, OPT_logfile, "-logfile", CMD_SINGLE, CMD_OPTIONAL,
425 "Location of log file ");
427 err = cmd_Dispatch(argc, argv);
429 return 0; /* not reached */
433 SalvageClient(VolumeId vid, char * pname)
438 SALVSYNC_response_hdr sres;
439 VolumePackageOptions opts;
441 VOptDefaults(volumeUtility, &opts);
442 if (VInitVolumePackage2(volumeUtility, &opts)) {
443 /* VInitVolumePackage2 can fail on e.g. partition attachment errors,
444 * but we don't really care, since all we're doing is trying to use
446 fprintf(stderr, "errors encountered initializing volume package, but "
447 "trying to continue anyway\n");
449 SALVSYNC_clientInit();
451 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_SALVAGE, SALVSYNC_OPERATOR, 0, NULL);
452 if (code != SYNC_OK) {
456 res.payload.buf = (void *) &sres;
457 res.payload.len = sizeof(sres);
461 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_QUERY, SALVSYNC_WHATEVER, 0, &res);
462 if (code != SYNC_OK) {
465 switch (sres.state) {
466 case SALVSYNC_STATE_ERROR:
467 printf("salvageserver reports salvage ended in an error; check log files for more details\n");
468 case SALVSYNC_STATE_DONE:
469 case SALVSYNC_STATE_UNKNOWN:
473 SALVSYNC_clientFinis();
477 if (code == SYNC_DENIED) {
478 printf("salvageserver refused to salvage volume %u on partition %s\n",
480 } else if (code == SYNC_BAD_COMMAND) {
481 printf("SALVSYNC protocol mismatch; please make sure fileserver, volserver, salvageserver and salvager are same version\n");
482 } else if (code == SYNC_COM_ERROR) {
483 printf("SALVSYNC communications error\n");
485 SALVSYNC_clientFinis();
489 static int * child_slot;
492 SalvageServer(int argc, char **argv)
495 struct SalvageQueueNode * node;
497 pthread_attr_t attrs;
499 VolumePackageOptions opts;
501 /* All entries to the log will be appended. Useful if there are
502 * multiple salvagers appending to the log.
505 CheckLogFile(logFileName);
507 #ifdef AFS_LINUX20_ENV
508 fcntl(fileno(logFile), F_SETFL, O_APPEND); /* Isn't this redundant? */
510 fcntl(fileno(logFile), F_SETFL, FAPPEND); /* Isn't this redundant? */
515 fprintf(logFile, "%s\n", cml_version_number);
516 LogCommandLine(argc, argv, "Online Salvage Server",
517 SalvageVersion, "Starting OpenAFS", Log);
518 /* Get and hold a lock for the duration of the salvage to make sure
519 * that no other salvage runs at the same time. The routine
520 * VInitVolumePackage2 (called below) makes sure that a file server or
521 * other volume utilities don't interfere with the salvage.
524 /* even demand attach online salvager
525 * still needs this because we don't want
526 * a stand-alone salvager to conflict with
527 * the salvager daemon */
528 ObtainSharedSalvageLock();
530 child_slot = calloc(Parallel, sizeof(int));
531 opr_Assert(child_slot != NULL);
533 /* initialize things */
534 VOptDefaults(salvageServer, &opts);
535 if (VInitVolumePackage2(salvageServer, &opts)) {
536 Log("Shutting down: errors encountered initializing volume package\n");
540 queue_Init(&pending_q);
541 queue_Init(&log_cleanup_queue);
542 opr_mutex_init(&worker_lock);
543 opr_cv_init(&worker_cv);
544 opr_cv_init(&log_cleanup_queue.queue_change_cv);
545 opr_Verify(pthread_attr_init(&attrs) == 0);
547 /* start up the reaper and log cleaner threads */
548 opr_Verify(pthread_attr_setdetachstate(&attrs,
549 PTHREAD_CREATE_DETACHED) == 0);
550 opr_Verify(pthread_create(&tid, &attrs,
551 &SalvageChildReaperThread, NULL) == 0);
552 opr_Verify(pthread_create(&tid, &attrs,
553 &SalvageLogCleanupThread, NULL) == 0);
554 opr_Verify(pthread_create(&tid, &attrs,
555 &SalvageLogScanningThread, NULL) == 0);
557 /* loop forever serving requests */
559 node = SALVSYNC_getWork();
560 opr_Assert(node != NULL);
562 Log("dispatching child to salvage volume %u...\n",
563 node->command.sop.parent);
567 for (slot = 0; slot < Parallel; slot++) {
568 if (!child_slot[slot])
571 opr_Assert (slot < Parallel);
577 ret = DoSalvageVolume(node, slot);
579 } else if (pid < 0) {
580 Log("failed to fork child worker process\n");
584 child_slot[slot] = pid;
588 opr_mutex_enter(&worker_lock);
591 /* let the reaper thread know another worker was spawned */
592 opr_cv_broadcast(&worker_cv);
594 /* if we're overquota, wait for the reaper */
595 while (current_workers >= Parallel) {
596 opr_cv_wait(&worker_cv, &worker_lock);
598 opr_mutex_exit(&worker_lock);
604 DoSalvageVolume(struct SalvageQueueNode * node, int slot)
606 char childLog[AFSDIR_PATH_MAX];
607 struct DiskPartition64 * partP;
609 /* do not allow further forking inside salvager */
612 /* do not attempt to close parent's logFile handle as
613 * another thread may have held the lock on the FILE
614 * structure when fork was called! */
616 snprintf(childLog, sizeof(childLog), "%s.%d",
617 AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid());
619 logFile = afs_fopen(childLog, "a");
620 if (!logFile) { /* still nothing, use stdout */
625 if (node->command.sop.parent <= 0) {
626 Log("salvageServer: invalid volume id specified; salvage aborted\n");
630 partP = VGetPartition(node->command.sop.partName, 0);
632 Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
633 node->command.sop.partName);
637 /* obtain a shared salvage lock in the child worker, so if the
638 * salvageserver restarts (and we continue), we will still hold a lock and
639 * prevent standalone salvagers from interfering */
640 ObtainSharedSalvageLock();
642 /* Salvage individual volume; don't notify fs */
643 SalvageFileSys1(partP, node->command.sop.parent);
651 SalvageChildReaperThread(void * args)
653 int slot, pid, status;
654 struct log_cleanup_node * cleanup;
656 opr_mutex_enter(&worker_lock);
658 /* loop reaping our children */
660 /* wait() won't block unless we have children, so
661 * block on the cond var if we're childless */
662 while (current_workers == 0) {
663 opr_cv_wait(&worker_cv, &worker_lock);
666 opr_mutex_exit(&worker_lock);
668 cleanup = malloc(sizeof(struct log_cleanup_node));
670 while (Reap_Child("salvageserver", &pid, &status) < 0) {
671 /* try to prevent livelock if something goes wrong */
676 for (slot = 0; slot < Parallel; slot++) {
677 if (child_slot[slot] == pid)
680 opr_Assert(slot < Parallel);
681 child_slot[slot] = 0;
684 SALVSYNC_doneWorkByPid(pid, status);
686 opr_mutex_enter(&worker_lock);
690 queue_Append(&log_cleanup_queue, cleanup);
691 opr_cv_signal(&log_cleanup_queue.queue_change_cv);
694 /* ok, we've reaped a child */
696 opr_cv_broadcast(&worker_cv);
703 Reap_Child(char *prog, int * pid, int * status)
710 if (WCOREDUMP(*status))
711 Log("\"%s\" core dumped!\n", prog);
712 if ((WIFSIGNALED(*status) != 0) ||
713 ((WEXITSTATUS(*status) != 0) &&
714 (WEXITSTATUS(*status) != SALSRV_EXIT_VOLGROUP_LINK)))
715 Log("\"%s\" (pid=%d) terminated abnormally!\n", prog, ret);
717 Log("wait returned -1\n");
723 * thread to combine salvager child logs
724 * back into the main salvageserver log
727 SalvageLogCleanupThread(void * arg)
729 struct log_cleanup_node * cleanup;
731 opr_mutex_enter(&worker_lock);
734 while (queue_IsEmpty(&log_cleanup_queue)) {
735 opr_cv_wait(&log_cleanup_queue.queue_change_cv, &worker_lock);
738 while (queue_IsNotEmpty(&log_cleanup_queue)) {
739 cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
740 queue_Remove(cleanup);
741 opr_mutex_exit(&worker_lock);
742 SalvageLogCleanup(cleanup->pid);
744 opr_mutex_enter(&worker_lock);
748 opr_mutex_exit(&worker_lock);
752 #define LOG_XFER_BUF_SIZE 65536
754 SalvageLogCleanup(int pid)
757 char fn[AFSDIR_PATH_MAX];
758 static char buf[LOG_XFER_BUF_SIZE];
760 snprintf(fn, sizeof(fn), "%s.%d",
761 AFSDIR_SERVER_SLVGLOG_FILEPATH, pid);
764 pidlog = open(fn, O_RDONLY);
769 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
771 fwrite(buf, len, 1, logFile);
772 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
780 /* wake up every five minutes to see if a non-child salvage has finished */
781 #define SALVAGE_SCAN_POLL_INTERVAL 300
784 * Thread to look for SalvageLog.$pid files that are not from our child
785 * worker salvagers, and notify SalvageLogCleanupThread to clean them
786 * up. This can happen if we restart during salvages, or the
787 * salvageserver crashes or something.
791 * @return always NULL
794 SalvageLogScanningThread(void * arg)
796 struct rx_queue log_watch_queue;
798 queue_Init(&log_watch_queue);
803 char prefix[AFSDIR_PATH_MAX];
806 snprintf(prefix, sizeof(prefix), "%s.", AFSDIR_SLVGLOG_FILE);
807 prefix_len = strlen(prefix);
809 dp = opendir(AFSDIR_LOGS_DIR);
812 while ((dirp = readdir(dp)) != NULL) {
814 struct log_cleanup_node *cleanup;
817 if (strncmp(dirp->d_name, prefix, prefix_len) != 0) {
818 /* not a salvage logfile; skip */
823 pid = strtol(dirp->d_name + prefix_len, NULL, 10);
826 /* file is SalvageLog.<something> but <something> isn't
832 for (i = 0; i < Parallel; ++i) {
833 if (pid == child_slot[i]) {
839 /* this pid is one of our children, so the reaper thread
840 * will take care of it; skip */
844 cleanup = malloc(sizeof(struct log_cleanup_node));
847 queue_Append(&log_watch_queue, cleanup);
853 ScanLogs(&log_watch_queue);
855 while (queue_IsNotEmpty(&log_watch_queue)) {
856 sleep(SALVAGE_SCAN_POLL_INTERVAL);
857 ScanLogs(&log_watch_queue);
864 * look through log_watch_queue, and if any processes are not still
865 * running, hand them off to the SalvageLogCleanupThread
867 * @param log_watch_queue a queue of PIDs that we should clean up if
871 ScanLogs(struct rx_queue *log_watch_queue)
873 struct log_cleanup_node *cleanup, *next;
875 opr_mutex_enter(&worker_lock);
877 for (queue_Scan(log_watch_queue, cleanup, next, log_cleanup_node)) {
878 /* if a process is still running, assume it's the salvage process
879 * still going, and keep waiting for it */
880 if (kill(cleanup->pid, 0) < 0 && errno == ESRCH) {
881 queue_Remove(cleanup);
882 queue_Append(&log_cleanup_queue, cleanup);
883 opr_cv_signal(&log_cleanup_queue.queue_change_cv);
887 opr_mutex_exit(&worker_lock);