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 */
190 handleit(struct cmd_syndesc *opts, void *arock)
193 afs_int32 seenpart = 0, seenvol = 0;
195 struct cmdline_rock *rock = (struct cmdline_rock *)arock;
196 char *optstring = NULL;
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_blockreads, &forceR);
212 if (cmd_OptionAsString(opts, OPT_parallel, &optstring) == 0) {
213 if (strncmp(optstring, "all", 3) == 0) {
216 if (strlen(optstring) != 0) {
217 Parallel = atoi(optstring);
220 if (Parallel > MAXPARALLEL) {
221 printf("Setting parallel salvages to maximum of %d \n",
223 Parallel = MAXPARALLEL;
229 Parallel = min(DEFAULT_PARALLELISM, MAXPARALLEL);
231 if (cmd_OptionAsString(opts, OPT_tmpdir, &optstring) == 0) {
233 dirp = opendir(optstring);
236 ("Can't open temporary placeholder dir %s; using current partition \n",
244 if (cmd_OptionAsString(opts, OPT_orphans, &optstring) == 0) {
246 orphans = ORPH_IGNORE;
247 else if (strcmp(optstring, "remove") == 0
248 || strcmp(optstring, "r") == 0)
249 orphans = ORPH_REMOVE;
250 else if (strcmp(optstring, "attach") == 0
251 || strcmp(optstring, "a") == 0)
252 orphans = ORPH_ATTACH;
256 #ifndef AFS_NT40_ENV /* ignore options on NT */
257 if (cmd_OptionPresent(opts, OPT_syslog)) {
260 cmd_OptionAsInt(opts, OPT_syslogfacility, &useSyslogFacility);
263 if (cmd_OptionPresent(opts, OPT_client)) {
264 if (cmd_OptionAsString(opts, OPT_partition, &optstring) == 0) {
266 strlcpy(pname, optstring, sizeof(pname));
270 if (cmd_OptionAsString(opts, OPT_volumeid, &optstring) == 0) {
274 vid_l = strtoul(optstring, &end, 10);
275 if (vid_l >= MAX_AFS_UINT32 || vid_l == ULONG_MAX || *end != '\0') {
276 printf("Invalid volume id specified; salvage aborted\n");
279 vid = (VolumeId)vid_l;
282 if (!seenpart || !seenvol) {
283 printf("You must specify '-partition' and '-volumeid' with the '-client' option\n");
287 SalvageClient(vid, pname);
289 } else { /* salvageserver mode */
290 SalvageServer(rock->argc, rock->argv);
297 #include "AFS_component_version_number.c"
301 char *save_args[MAX_ARGS];
303 pthread_t main_thread;
307 main(int argc, char **argv)
309 struct cmd_syndesc *ts;
311 struct cmdline_rock arock;
315 * The following signal action for AIX is necessary so that in case of a
316 * crash (i.e. core is generated) we can include the user's data section
317 * in the core dump. Unfortunately, by default, only a partial core is
318 * generated which, in many cases, isn't too useful.
320 struct sigaction nsa;
322 sigemptyset(&nsa.sa_mask);
323 nsa.sa_handler = SIG_DFL;
324 nsa.sa_flags = SA_FULLDUMP;
325 sigaction(SIGABRT, &nsa, NULL);
326 sigaction(SIGSEGV, &nsa, NULL);
329 /* Initialize directory paths */
330 if (!(initAFSDirPath() & AFSDIR_SERVER_PATHS_OK)) {
332 ReportErrorEventAlt(AFSEVT_SVR_NO_INSTALL_DIR, 0, argv[0], 0);
334 fprintf(stderr, "%s: Unable to obtain AFS server directory.\n",
339 /* Default to binary mode for fopen() */
340 _set_fmode(_O_BINARY);
342 main_thread = pthread_self();
343 if (spawnDatap && spawnDataLen) {
344 /* This is a child per partition salvager. Don't setup log or
345 * try to lock the salvager lock.
347 if (nt_SetupPartitionSalvage(spawnDatap, spawnDataLen) < 0)
353 if (geteuid() != 0) {
354 printf("Salvager must be run as root.\n");
360 /* bad for normal help flag processing, but can do nada */
369 logFileName = strdup(AFSDIR_SERVER_SALSRVLOG_FILEPATH);
371 ts = cmd_CreateSyntax("initcmd", handleit, &arock, 0, "initialize the program");
372 cmd_AddParmAtOffset(ts, OPT_partition, "-partition", CMD_SINGLE,
373 CMD_OPTIONAL, "Name of partition to salvage");
374 cmd_AddParmAtOffset(ts, OPT_volumeid, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
375 "Volume Id to salvage");
376 cmd_AddParmAtOffset(ts, OPT_debug, "-debug", CMD_FLAG, CMD_OPTIONAL,
377 "Run in Debugging mode");
378 cmd_AddParmAtOffset(ts, OPT_nowrite, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
379 "Run readonly/test mode");
380 cmd_AddParmAtOffset(ts, OPT_inodes, "-inodes", CMD_FLAG, CMD_OPTIONAL,
381 "Just list affected afs inodes - debugging flag");
382 cmd_AddParmAtOffset(ts, OPT_oktozap, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
383 "Give permission to destroy bogus inodes/volumes - debugging flag");
384 cmd_AddParmAtOffset(ts, OPT_rootinodes, "-rootinodes", CMD_FLAG,
385 CMD_OPTIONAL, "Show inodes owned by root - debugging flag");
386 cmd_AddParmAtOffset(ts, OPT_salvagedirs, "-salvagedirs", CMD_FLAG,
387 CMD_OPTIONAL, "Force rebuild/salvage of all directories");
388 cmd_AddParmAtOffset(ts, OPT_blockreads, "-blockreads", CMD_FLAG,
389 CMD_OPTIONAL, "Read smaller blocks to handle IO/bad blocks");
390 cmd_AddParmAtOffset(ts, OPT_parallel, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
391 "# of max parallel partition salvaging");
392 cmd_AddParmAtOffset(ts, OPT_tmpdir, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
393 "Name of dir to place tmp files ");
394 cmd_AddParmAtOffset(ts, OPT_orphans, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
395 "ignore | remove | attach");
397 #if !defined(AFS_NT40_ENV)
398 cmd_AddParmAtOffset(ts, OPT_syslog, "-syslog", CMD_FLAG, CMD_OPTIONAL,
399 "Write salvage log to syslogs");
400 cmd_AddParmAtOffset(ts, OPT_syslogfacility, "-syslogfacility", CMD_SINGLE,
401 CMD_OPTIONAL, "Syslog facility number to use");
404 cmd_AddParmAtOffset(ts, OPT_client, "-client", CMD_FLAG, CMD_OPTIONAL,
405 "Use SALVSYNC to ask salvageserver to salvage a volume");
407 cmd_AddParmAtOffset(ts, OPT_logfile, "-logfile", CMD_SINGLE, CMD_OPTIONAL,
408 "Location of log file ");
410 err = cmd_Dispatch(argc, argv);
412 return 0; /* not reached */
416 SalvageClient(VolumeId vid, char * pname)
421 SALVSYNC_response_hdr sres;
422 VolumePackageOptions opts;
424 /* Send Log() messages to stderr in client mode. */
427 VOptDefaults(volumeUtility, &opts);
428 if (VInitVolumePackage2(volumeUtility, &opts)) {
429 /* VInitVolumePackage2 can fail on e.g. partition attachment errors,
430 * but we don't really care, since all we're doing is trying to use
432 fprintf(stderr, "errors encountered initializing volume package, but "
433 "trying to continue anyway\n");
435 SALVSYNC_clientInit();
437 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_SALVAGE, SALVSYNC_OPERATOR, 0, NULL);
438 if (code != SYNC_OK) {
442 res.payload.buf = (void *) &sres;
443 res.payload.len = sizeof(sres);
447 code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_QUERY, SALVSYNC_WHATEVER, 0, &res);
448 if (code != SYNC_OK) {
451 switch (sres.state) {
452 case SALVSYNC_STATE_ERROR:
453 printf("salvageserver reports salvage ended in an error; check log files for more details\n");
454 case SALVSYNC_STATE_DONE:
455 case SALVSYNC_STATE_UNKNOWN:
459 SALVSYNC_clientFinis();
463 if (code == SYNC_DENIED) {
464 printf("salvageserver refused to salvage volume %u on partition %s\n",
466 } else if (code == SYNC_BAD_COMMAND) {
467 printf("SALVSYNC protocol mismatch; please make sure fileserver, volserver, salvageserver and salvager are same version\n");
468 } else if (code == SYNC_COM_ERROR) {
469 printf("SALVSYNC communications error\n");
471 SALVSYNC_clientFinis();
475 static int * child_slot;
478 SalvageServer(int argc, char **argv)
481 struct SalvageQueueNode * node;
483 pthread_attr_t attrs;
485 VolumePackageOptions opts;
487 /* All entries to the log will be appended. Useful if there are
488 * multiple salvagers appending to the log.
491 CheckLogFile(logFileName);
493 #ifdef AFS_LINUX20_ENV
494 fcntl(fileno(logFile), F_SETFL, O_APPEND); /* Isn't this redundant? */
496 fcntl(fileno(logFile), F_SETFL, FAPPEND); /* Isn't this redundant? */
501 fprintf(logFile, "%s\n", cml_version_number);
502 LogCommandLine(argc, argv, "Online Salvage Server",
503 SalvageVersion, "Starting OpenAFS", Log);
504 /* Get and hold a lock for the duration of the salvage to make sure
505 * that no other salvage runs at the same time. The routine
506 * VInitVolumePackage2 (called below) makes sure that a file server or
507 * other volume utilities don't interfere with the salvage.
510 /* even demand attach online salvager
511 * still needs this because we don't want
512 * a stand-alone salvager to conflict with
513 * the salvager daemon */
514 ObtainSharedSalvageLock();
516 child_slot = calloc(Parallel, sizeof(int));
517 opr_Assert(child_slot != NULL);
519 /* initialize things */
520 VOptDefaults(salvageServer, &opts);
521 if (VInitVolumePackage2(salvageServer, &opts)) {
522 Log("Shutting down: errors encountered initializing volume package\n");
526 queue_Init(&pending_q);
527 queue_Init(&log_cleanup_queue);
528 opr_mutex_init(&worker_lock);
529 opr_cv_init(&worker_cv);
530 opr_cv_init(&log_cleanup_queue.queue_change_cv);
531 opr_Verify(pthread_attr_init(&attrs) == 0);
533 /* start up the reaper and log cleaner threads */
534 opr_Verify(pthread_attr_setdetachstate(&attrs,
535 PTHREAD_CREATE_DETACHED) == 0);
536 opr_Verify(pthread_create(&tid, &attrs,
537 &SalvageChildReaperThread, NULL) == 0);
538 opr_Verify(pthread_create(&tid, &attrs,
539 &SalvageLogCleanupThread, NULL) == 0);
540 opr_Verify(pthread_create(&tid, &attrs,
541 &SalvageLogScanningThread, NULL) == 0);
543 /* loop forever serving requests */
545 node = SALVSYNC_getWork();
546 opr_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 opr_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 opr_mutex_enter(&worker_lock);
577 /* let the reaper thread know another worker was spawned */
578 opr_cv_broadcast(&worker_cv);
580 /* if we're overquota, wait for the reaper */
581 while (current_workers >= Parallel) {
582 opr_cv_wait(&worker_cv, &worker_lock);
584 opr_mutex_exit(&worker_lock);
590 DoSalvageVolume(struct SalvageQueueNode * node, int slot)
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 if (asprintf(&childLog, "%s.%d",
603 AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid()) < 0) {
606 logFile = afs_fopen(childLog, "a");
607 if (!logFile) { /* still nothing, use stdout */
613 if (node->command.sop.parent <= 0) {
614 Log("salvageServer: invalid volume id specified; salvage aborted\n");
618 partP = VGetPartition(node->command.sop.partName, 0);
620 Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
621 node->command.sop.partName);
625 /* obtain a shared salvage lock in the child worker, so if the
626 * salvageserver restarts (and we continue), we will still hold a lock and
627 * prevent standalone salvagers from interfering */
628 ObtainSharedSalvageLock();
630 /* Salvage individual volume; don't notify fs */
631 SalvageFileSys1(partP, node->command.sop.parent);
639 SalvageChildReaperThread(void * args)
641 int slot, pid, status;
642 struct log_cleanup_node * cleanup;
644 opr_mutex_enter(&worker_lock);
646 /* loop reaping our children */
648 /* wait() won't block unless we have children, so
649 * block on the cond var if we're childless */
650 while (current_workers == 0) {
651 opr_cv_wait(&worker_cv, &worker_lock);
654 opr_mutex_exit(&worker_lock);
656 cleanup = malloc(sizeof(struct log_cleanup_node));
658 while (Reap_Child("salvageserver", &pid, &status) < 0) {
659 /* try to prevent livelock if something goes wrong */
664 for (slot = 0; slot < Parallel; slot++) {
665 if (child_slot[slot] == pid)
668 opr_Assert(slot < Parallel);
669 child_slot[slot] = 0;
672 SALVSYNC_doneWorkByPid(pid, status);
674 opr_mutex_enter(&worker_lock);
678 queue_Append(&log_cleanup_queue, cleanup);
679 opr_cv_signal(&log_cleanup_queue.queue_change_cv);
682 /* ok, we've reaped a child */
684 opr_cv_broadcast(&worker_cv);
691 Reap_Child(char *prog, int * pid, int * status)
698 if (WCOREDUMP(*status))
699 Log("\"%s\" core dumped!\n", prog);
700 if ((WIFSIGNALED(*status) != 0) ||
701 ((WEXITSTATUS(*status) != 0) &&
702 (WEXITSTATUS(*status) != SALSRV_EXIT_VOLGROUP_LINK)))
703 Log("\"%s\" (pid=%d) terminated abnormally!\n", prog, ret);
705 Log("wait returned -1\n");
711 * thread to combine salvager child logs
712 * back into the main salvageserver log
715 SalvageLogCleanupThread(void * arg)
717 struct log_cleanup_node * cleanup;
719 opr_mutex_enter(&worker_lock);
722 while (queue_IsEmpty(&log_cleanup_queue)) {
723 opr_cv_wait(&log_cleanup_queue.queue_change_cv, &worker_lock);
726 while (queue_IsNotEmpty(&log_cleanup_queue)) {
727 cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
728 queue_Remove(cleanup);
729 opr_mutex_exit(&worker_lock);
730 SalvageLogCleanup(cleanup->pid);
732 opr_mutex_enter(&worker_lock);
736 opr_mutex_exit(&worker_lock);
740 #define LOG_XFER_BUF_SIZE 65536
742 SalvageLogCleanup(int pid)
746 static char buf[LOG_XFER_BUF_SIZE];
748 if (asprintf(&fn, "%s.%d", AFSDIR_SERVER_SLVGLOG_FILEPATH, pid) < 0)
751 pidlog = open(fn, O_RDONLY);
757 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
759 fwrite(buf, len, 1, logFile);
760 len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
768 /* wake up every five minutes to see if a non-child salvage has finished */
769 #define SALVAGE_SCAN_POLL_INTERVAL 300
772 * Thread to look for SalvageLog.$pid files that are not from our child
773 * worker salvagers, and notify SalvageLogCleanupThread to clean them
774 * up. This can happen if we restart during salvages, or the
775 * salvageserver crashes or something.
779 * @return always NULL
782 SalvageLogScanningThread(void * arg)
784 struct rx_queue log_watch_queue;
788 queue_Init(&log_watch_queue);
790 prefix_len = asprintf(&prefix, "%s.", AFSDIR_SLVGLOG_FILE);
791 if (prefix_len >= 0) {
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 */
830 cleanup = malloc(sizeof(struct log_cleanup_node));
833 queue_Append(&log_watch_queue, cleanup);
839 ScanLogs(&log_watch_queue);
841 while (queue_IsNotEmpty(&log_watch_queue)) {
842 sleep(SALVAGE_SCAN_POLL_INTERVAL);
843 ScanLogs(&log_watch_queue);
850 * look through log_watch_queue, and if any processes are not still
851 * running, hand them off to the SalvageLogCleanupThread
853 * @param log_watch_queue a queue of PIDs that we should clean up if
857 ScanLogs(struct rx_queue *log_watch_queue)
859 struct log_cleanup_node *cleanup, *next;
861 opr_mutex_enter(&worker_lock);
863 for (queue_Scan(log_watch_queue, cleanup, next, log_cleanup_node)) {
864 /* if a process is still running, assume it's the salvage process
865 * still going, and keep waiting for it */
866 if (kill(cleanup->pid, 0) < 0 && errno == ESRCH) {
867 queue_Remove(cleanup);
868 queue_Append(&log_cleanup_queue, cleanup);
869 opr_cv_signal(&log_cleanup_queue.queue_change_cv);
873 opr_mutex_exit(&worker_lock);