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"
114 #if !defined(AFS_DEMAND_ATTACH_FS)
115 #error "online salvager only supported for demand attach fileserver"
116 #endif /* AFS_DEMAND_ATTACH_FS */
118 #if defined(AFS_NT40_ENV)
119 #error "online salvager not supported on NT"
120 #endif /* AFS_NT40_ENV */
122 /*@+fcnmacros +macrofcndecl@*/
124 #define afs_fopen fopen64
125 #else /* !O_LARGEFILE */
126 #define afs_fopen fopen
127 #endif /* !O_LARGEFILE */
128 /*@=fcnmacros =macrofcndecl@*/
132 static volatile int current_workers = 0;
133 static volatile struct rx_queue pending_q;
134 static pthread_mutex_t worker_lock;
135 static pthread_cond_t worker_cv;
137 static void * SalvageChildReaperThread(void *);
138 static int DoSalvageVolume(struct SalvageQueueNode * node, int slot);
140 static void SalvageServer(int argc, char **argv);
141 static void SalvageClient(VolumeId vid, char * pname);
143 static int Reap_Child(char * prog, int * pid, int * status);
145 static void * SalvageLogCleanupThread(void *);
146 static int SalvageLogCleanup(int pid);
148 static void * SalvageLogScanningThread(void *);
149 static void ScanLogs(struct rx_queue *log_watch_queue);
151 struct cmdline_rock {
156 struct log_cleanup_node {
162 struct rx_queue queue_head;
163 pthread_cond_t queue_change_cv;
167 #define DEFAULT_PARALLELISM 4 /* allow 4 parallel salvage workers by default */
170 handleit(struct cmd_syndesc *as, void *arock)
173 char pname[100], *temp;
174 afs_int32 seenpart = 0, seenvol = 0;
176 struct cmdline_rock *rock = (struct cmdline_rock *)arock;
178 #ifdef AFS_SGI_VNODE_GLUE
179 if (afs_init_kernel_config(-1) < 0) {
181 ("Can't determine NUMA configuration, not starting salvager.\n");
186 if (as->parms[2].items) /* -debug */
188 if (as->parms[3].items) /* -nowrite */
190 if (as->parms[4].items) /* -inodes */
192 if (as->parms[5].items) /* -oktozap */
194 if (as->parms[6].items) /* -rootinodes */
196 if (as->parms[8].items) /* -ForceReads */
198 if ((ti = as->parms[9].items)) { /* -Parallel # */
200 if (strncmp(temp, "all", 3) == 0) {
204 if (strlen(temp) != 0) {
205 Parallel = atoi(temp);
208 if (Parallel > MAXPARALLEL) {
209 printf("Setting parallel salvages to maximum of %d \n",
211 Parallel = MAXPARALLEL;
215 Parallel = min(DEFAULT_PARALLELISM, MAXPARALLEL);
217 if ((ti = as->parms[10].items)) { /* -tmpdir */
221 dirp = opendir(tmpdir);
224 ("Can't open temporary placeholder dir %s; using current partition \n",
230 if ((ti = as->parms[11].items)) /* -showlog */
232 if ((ti = as->parms[12].items)) { /* -orphans */
234 orphans = ORPH_IGNORE;
235 else if (strcmp(ti->data, "remove") == 0
236 || strcmp(ti->data, "r") == 0)
237 orphans = ORPH_REMOVE;
238 else if (strcmp(ti->data, "attach") == 0
239 || strcmp(ti->data, "a") == 0)
240 orphans = ORPH_ATTACH;
242 #ifndef AFS_NT40_ENV /* ignore options on NT */
243 if ((ti = as->parms[13].items)) { /* -syslog */
247 if ((ti = as->parms[14].items)) { /* -syslogfacility */
248 useSyslogFacility = atoi(ti->data);
251 if ((ti = as->parms[15].items)) { /* -datelogs */
252 TimeStampLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
256 if ((ti = as->parms[16].items)) { /* -client */
257 if ((ti = as->parms[0].items)) { /* -partition */
259 strlcpy(pname, ti->data, sizeof(pname));
261 if ((ti = as->parms[1].items)) { /* -volumeid */
265 vid_l = strtoul(ti->data, &end, 10);
266 if (vid_l >= MAX_AFS_UINT32 || vid_l == ULONG_MAX || *end != '\0') {
267 printf("Invalid volume id specified; salvage aborted\n");
270 vid = (VolumeId)vid_l;
274 printf("-showlog does not work with -client\n");
278 if (!seenpart || !seenvol) {
279 printf("You must specify '-partition' and '-volumeid' with the '-client' option\n");
283 SalvageClient(vid, pname);
285 } else { /* salvageserver mode */
286 SalvageServer(rock->argc, rock->argv);
293 #include "AFS_component_version_number.c"
297 char *save_args[MAX_ARGS];
299 pthread_t main_thread;
303 main(int argc, char **argv)
305 struct cmd_syndesc *ts;
307 struct cmdline_rock arock;
311 * The following signal action for AIX is necessary so that in case of a
312 * crash (i.e. core is generated) we can include the user's data section
313 * in the core dump. Unfortunately, by default, only a partial core is
314 * generated which, in many cases, isn't too useful.
316 struct sigaction nsa;
318 sigemptyset(&nsa.sa_mask);
319 nsa.sa_handler = SIG_DFL;
320 nsa.sa_flags = SA_FULLDUMP;
321 sigaction(SIGABRT, &nsa, NULL);
322 sigaction(SIGSEGV, &nsa, NULL);
325 /* Initialize directory paths */
326 if (!(initAFSDirPath() & AFSDIR_SERVER_PATHS_OK)) {
328 ReportErrorEventAlt(AFSEVT_SVR_NO_INSTALL_DIR, 0, argv[0], 0);
330 fprintf(stderr, "%s: Unable to obtain AFS server directory.\n",
335 /* Default to binary mode for fopen() */
336 _set_fmode(_O_BINARY);
338 main_thread = pthread_self();
339 if (spawnDatap && spawnDataLen) {
340 /* This is a child per partition salvager. Don't setup log or
341 * try to lock the salvager lock.
343 if (nt_SetupPartitionSalvage(spawnDatap, spawnDataLen) < 0)
349 if (geteuid() != 0) {
350 printf("Salvager must be run as root.\n");
356 /* bad for normal help flag processing, but can do nada */
365 ts = cmd_CreateSyntax("initcmd", handleit, &arock, "initialize the program");
366 cmd_AddParm(ts, "-partition", CMD_SINGLE, CMD_OPTIONAL,
367 "Name of partition to salvage");
368 cmd_AddParm(ts, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
369 "Volume Id to salvage");
370 cmd_AddParm(ts, "-debug", CMD_FLAG, CMD_OPTIONAL,
371 "Run in Debugging mode");
372 cmd_AddParm(ts, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
373 "Run readonly/test mode");
374 cmd_AddParm(ts, "-inodes", CMD_FLAG, CMD_OPTIONAL,
375 "Just list affected afs inodes - debugging flag");
376 cmd_AddParm(ts, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
377 "Give permission to destroy bogus inodes/volumes - debugging flag");
378 cmd_AddParm(ts, "-rootinodes", CMD_FLAG, CMD_OPTIONAL,
379 "Show inodes owned by root - debugging flag");
380 cmd_AddParm(ts, "-salvagedirs", CMD_FLAG, CMD_OPTIONAL,
381 "Force rebuild/salvage of all directories");
382 cmd_AddParm(ts, "-blockreads", CMD_FLAG, CMD_OPTIONAL,
383 "Read smaller blocks to handle IO/bad blocks");
384 cmd_AddParm(ts, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
385 "# of max parallel partition salvaging");
386 cmd_AddParm(ts, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
387 "Name of dir to place tmp files ");
388 cmd_AddParm(ts, "-showlog", CMD_FLAG, CMD_OPTIONAL,
389 "Show log file upon completion");
390 cmd_AddParm(ts, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
391 "ignore | remove | attach");
393 /* note - syslog isn't avail on NT, but if we make it conditional, have
394 * to deal with screwy offsets for cmd params */
395 cmd_AddParm(ts, "-syslog", CMD_FLAG, CMD_OPTIONAL,
396 "Write salvage log to syslogs");
397 cmd_AddParm(ts, "-syslogfacility", CMD_SINGLE, CMD_OPTIONAL,
398 "Syslog facility number to use");
399 cmd_AddParm(ts, "-datelogs", CMD_FLAG, CMD_OPTIONAL,
400 "Include timestamp in logfile filename");
402 cmd_AddParm(ts, "-client", CMD_FLAG, CMD_OPTIONAL,
403 "Use SALVSYNC to ask salvageserver to salvage a volume");
405 err = cmd_Dispatch(argc, argv);
407 return 0; /* not reached */
411 SalvageClient(VolumeId vid, char * pname)
416 SALVSYNC_response_hdr sres;
417 VolumePackageOptions opts;
419 VOptDefaults(volumeUtility, &opts);
420 if (VInitVolumePackage2(volumeUtility, &opts)) {
421 /* VInitVolumePackage2 can fail on e.g. partition attachment errors,
422 * but we don't really care, since all we're doing is trying to use
424 fprintf(stderr, "errors encountered initializing volume package, but "
425 "trying to continue anyway\n");
427 SALVSYNC_clientInit();
429 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_SALVAGE, SALVSYNC_OPERATOR, 0, NULL);
430 if (code != SYNC_OK) {
434 res.payload.buf = (void *) &sres;
435 res.payload.len = sizeof(sres);
439 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_QUERY, SALVSYNC_WHATEVER, 0, &res);
440 if (code != SYNC_OK) {
443 switch (sres.state) {
444 case SALVSYNC_STATE_ERROR:
445 printf("salvageserver reports salvage ended in an error; check log files for more details\n");
446 case SALVSYNC_STATE_DONE:
447 case SALVSYNC_STATE_UNKNOWN:
451 SALVSYNC_clientFinis();
455 if (code == SYNC_DENIED) {
456 printf("salvageserver refused to salvage volume %u on partition %s\n",
458 } else if (code == SYNC_BAD_COMMAND) {
459 printf("SALVSYNC protocol mismatch; please make sure fileserver, volserver, salvageserver and salvager are same version\n");
460 } else if (code == SYNC_COM_ERROR) {
461 printf("SALVSYNC communications error\n");
463 SALVSYNC_clientFinis();
467 static int * child_slot;
470 SalvageServer(int argc, char **argv)
473 struct SalvageQueueNode * node;
475 pthread_attr_t attrs;
477 VolumePackageOptions opts;
479 /* All entries to the log will be appended. Useful if there are
480 * multiple salvagers appending to the log.
483 CheckLogFile((char *)AFSDIR_SERVER_SALSRVLOG_FILEPATH);
485 #ifdef AFS_LINUX20_ENV
486 fcntl(fileno(logFile), F_SETFL, O_APPEND); /* Isn't this redundant? */
488 fcntl(fileno(logFile), F_SETFL, FAPPEND); /* Isn't this redundant? */
493 fprintf(logFile, "%s\n", cml_version_number);
494 LogCommandLine(argc, argv, "Online Salvage Server",
495 SalvageVersion, "Starting OpenAFS", Log);
496 /* Get and hold a lock for the duration of the salvage to make sure
497 * that no other salvage runs at the same time. The routine
498 * VInitVolumePackage2 (called below) makes sure that a file server or
499 * other volume utilities don't interfere with the salvage.
502 /* even demand attach online salvager
503 * still needs this because we don't want
504 * a stand-alone salvager to conflict with
505 * the salvager daemon */
506 ObtainSharedSalvageLock();
508 child_slot = calloc(Parallel, sizeof(int));
509 opr_Assert(child_slot != NULL);
511 /* initialize things */
512 VOptDefaults(salvageServer, &opts);
513 if (VInitVolumePackage2(salvageServer, &opts)) {
514 Log("Shutting down: errors encountered initializing volume package\n");
518 queue_Init(&pending_q);
519 queue_Init(&log_cleanup_queue);
520 opr_mutex_init(&worker_lock);
521 opr_cv_init(&worker_cv);
522 opr_cv_init(&log_cleanup_queue.queue_change_cv);
523 opr_Verify(pthread_attr_init(&attrs) == 0);
525 /* start up the reaper and log cleaner threads */
526 opr_Verify(pthread_attr_setdetachstate(&attrs,
527 PTHREAD_CREATE_DETACHED) == 0);
528 opr_Verify(pthread_create(&tid, &attrs,
529 &SalvageChildReaperThread, NULL) == 0);
530 opr_Verify(pthread_create(&tid, &attrs,
531 &SalvageLogCleanupThread, NULL) == 0);
532 opr_Verify(pthread_create(&tid, &attrs,
533 &SalvageLogScanningThread, NULL) == 0);
535 /* loop forever serving requests */
537 node = SALVSYNC_getWork();
538 opr_Assert(node != NULL);
540 Log("dispatching child to salvage volume %u...\n",
541 node->command.sop.parent);
545 for (slot = 0; slot < Parallel; slot++) {
546 if (!child_slot[slot])
549 opr_Assert (slot < Parallel);
555 ret = DoSalvageVolume(node, slot);
557 } else if (pid < 0) {
558 Log("failed to fork child worker process\n");
562 child_slot[slot] = pid;
566 opr_mutex_enter(&worker_lock);
569 /* let the reaper thread know another worker was spawned */
570 opr_cv_broadcast(&worker_cv);
572 /* if we're overquota, wait for the reaper */
573 while (current_workers >= Parallel) {
574 opr_cv_wait(&worker_cv, &worker_lock);
576 opr_mutex_exit(&worker_lock);
582 DoSalvageVolume(struct SalvageQueueNode * node, int slot)
584 char childLog[AFSDIR_PATH_MAX];
585 struct DiskPartition64 * partP;
587 /* do not allow further forking inside salvager */
590 /* do not attempt to close parent's logFile handle as
591 * another thread may have held the lock on the FILE
592 * structure when fork was called! */
594 snprintf(childLog, sizeof(childLog), "%s.%d",
595 AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid());
597 logFile = afs_fopen(childLog, "a");
598 if (!logFile) { /* still nothing, use stdout */
603 if (node->command.sop.parent <= 0) {
604 Log("salvageServer: invalid volume id specified; salvage aborted\n");
608 partP = VGetPartition(node->command.sop.partName, 0);
610 Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
611 node->command.sop.partName);
615 /* obtain a shared salvage lock in the child worker, so if the
616 * salvageserver restarts (and we continue), we will still hold a lock and
617 * prevent standalone salvagers from interfering */
618 ObtainSharedSalvageLock();
620 /* Salvage individual volume; don't notify fs */
621 SalvageFileSys1(partP, node->command.sop.parent);
629 SalvageChildReaperThread(void * args)
631 int slot, pid, status;
632 struct log_cleanup_node * cleanup;
634 opr_mutex_enter(&worker_lock);
636 /* loop reaping our children */
638 /* wait() won't block unless we have children, so
639 * block on the cond var if we're childless */
640 while (current_workers == 0) {
641 opr_cv_wait(&worker_cv, &worker_lock);
644 opr_mutex_exit(&worker_lock);
646 cleanup = malloc(sizeof(struct log_cleanup_node));
648 while (Reap_Child("salvageserver", &pid, &status) < 0) {
649 /* try to prevent livelock if something goes wrong */
654 for (slot = 0; slot < Parallel; slot++) {
655 if (child_slot[slot] == pid)
658 opr_Assert(slot < Parallel);
659 child_slot[slot] = 0;
662 SALVSYNC_doneWorkByPid(pid, status);
664 opr_mutex_enter(&worker_lock);
668 queue_Append(&log_cleanup_queue, cleanup);
669 opr_cv_signal(&log_cleanup_queue.queue_change_cv);
672 /* ok, we've reaped a child */
674 opr_cv_broadcast(&worker_cv);
681 Reap_Child(char *prog, int * pid, int * status)
688 if (WCOREDUMP(*status))
689 Log("\"%s\" core dumped!\n", prog);
690 if ((WIFSIGNALED(*status) != 0) ||
691 ((WEXITSTATUS(*status) != 0) &&
692 (WEXITSTATUS(*status) != SALSRV_EXIT_VOLGROUP_LINK)))
693 Log("\"%s\" (pid=%d) terminated abnormally!\n", prog, ret);
695 Log("wait returned -1\n");
701 * thread to combine salvager child logs
702 * back into the main salvageserver log
705 SalvageLogCleanupThread(void * arg)
707 struct log_cleanup_node * cleanup;
709 opr_mutex_enter(&worker_lock);
712 while (queue_IsEmpty(&log_cleanup_queue)) {
713 opr_cv_wait(&log_cleanup_queue.queue_change_cv, &worker_lock);
716 while (queue_IsNotEmpty(&log_cleanup_queue)) {
717 cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
718 queue_Remove(cleanup);
719 opr_mutex_exit(&worker_lock);
720 SalvageLogCleanup(cleanup->pid);
722 opr_mutex_enter(&worker_lock);
726 opr_mutex_exit(&worker_lock);
730 #define LOG_XFER_BUF_SIZE 65536
732 SalvageLogCleanup(int pid)
735 char fn[AFSDIR_PATH_MAX];
736 static char buf[LOG_XFER_BUF_SIZE];
738 snprintf(fn, sizeof(fn), "%s.%d",
739 AFSDIR_SERVER_SLVGLOG_FILEPATH, pid);
742 pidlog = open(fn, O_RDONLY);
747 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
749 fwrite(buf, len, 1, logFile);
750 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
758 /* wake up every five minutes to see if a non-child salvage has finished */
759 #define SALVAGE_SCAN_POLL_INTERVAL 300
762 * Thread to look for SalvageLog.$pid files that are not from our child
763 * worker salvagers, and notify SalvageLogCleanupThread to clean them
764 * up. This can happen if we restart during salvages, or the
765 * salvageserver crashes or something.
769 * @return always NULL
772 SalvageLogScanningThread(void * arg)
774 struct rx_queue log_watch_queue;
776 queue_Init(&log_watch_queue);
781 char prefix[AFSDIR_PATH_MAX];
784 snprintf(prefix, sizeof(prefix), "%s.", AFSDIR_SLVGLOG_FILE);
785 prefix_len = strlen(prefix);
787 dp = opendir(AFSDIR_LOGS_DIR);
790 while ((dirp = readdir(dp)) != NULL) {
792 struct log_cleanup_node *cleanup;
795 if (strncmp(dirp->d_name, prefix, prefix_len) != 0) {
796 /* not a salvage logfile; skip */
801 pid = strtol(dirp->d_name + prefix_len, NULL, 10);
804 /* file is SalvageLog.<something> but <something> isn't
810 for (i = 0; i < Parallel; ++i) {
811 if (pid == child_slot[i]) {
817 /* this pid is one of our children, so the reaper thread
818 * will take care of it; skip */
822 cleanup = malloc(sizeof(struct log_cleanup_node));
825 queue_Append(&log_watch_queue, cleanup);
831 ScanLogs(&log_watch_queue);
833 while (queue_IsNotEmpty(&log_watch_queue)) {
834 sleep(SALVAGE_SCAN_POLL_INTERVAL);
835 ScanLogs(&log_watch_queue);
842 * look through log_watch_queue, and if any processes are not still
843 * running, hand them off to the SalvageLogCleanupThread
845 * @param log_watch_queue a queue of PIDs that we should clean up if
849 ScanLogs(struct rx_queue *log_watch_queue)
851 struct log_cleanup_node *cleanup, *next;
853 opr_mutex_enter(&worker_lock);
855 for (queue_Scan(log_watch_queue, cleanup, next, log_cleanup_node)) {
856 /* if a process is still running, assume it's the salvage process
857 * still going, and keep waiting for it */
858 if (kill(cleanup->pid, 0) < 0 && errno == ESRCH) {
859 queue_Remove(cleanup);
860 queue_Append(&log_cleanup_queue, cleanup);
861 opr_cv_signal(&log_cleanup_queue.queue_change_cv);
865 opr_mutex_exit(&worker_lock);