/*
* Copyright 2006-2007, Sine Nomine Associates and others.
* All Rights Reserved.
- *
+ *
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
* directory or online at http://www.openafs.org/dl/license10.html
*/
-/*
+/*
* demand attach fs
* online salvager daemon
*/
#include <afsconfig.h>
#include <afs/param.h>
-RCSID
- ("$Header$");
+#include <roken.h>
+
+#ifdef HAVE_SYS_FILE_H
+#include <sys/file.h>
+#endif
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include <dirent.h>
-#include <sys/stat.h>
-#include <time.h>
-#include <errno.h>
#ifdef AFS_NT40_ENV
-#include <io.h>
#include <WINNT/afsevent.h>
-#else
-#include <sys/param.h>
-#include <sys/file.h>
-#ifndef ITIMER_REAL
-#include <sys/time.h>
-#endif /* ITIMER_REAL */
#endif
-#if defined(AFS_AIX_ENV) || defined(AFS_SUN4_ENV)
-#define WCOREDUMP(x) (x & 0200)
+
+#ifndef WCOREDUMP
+#define WCOREDUMP(x) ((x) & 0200)
#endif
-#include <rx/xdr.h>
+
+#include <afs/opr.h>
+#include <opr/lock.h>
#include <afs/afsint.h>
-#include <afs/assert.h>
+#include <rx/rx_queue.h>
+
#if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
#if defined(AFS_VFSINCL_ENV)
#include <sys/vnode.h>
#endif
#endif
#else /* AFS_VFSINCL_ENV */
-#ifdef AFS_OSF_ENV
-#include <ufs/inode.h>
-#else /* AFS_OSF_ENV */
-#if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV)
+#if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
#include <sys/inode.h>
#endif
-#endif
#endif /* AFS_VFSINCL_ENV */
#endif /* AFS_SGI_ENV */
#ifdef AFS_AIX_ENV
#include <sys/lockf.h>
#else
#ifdef AFS_HPUX_ENV
-#include <unistd.h>
#include <checklist.h>
#else
#if defined(AFS_SGI_ENV)
-#include <unistd.h>
-#include <fcntl.h>
#include <mntent.h>
#else
#if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
#ifdef AFS_SUN5_ENV
-#include <unistd.h>
#include <sys/mnttab.h>
#include <sys/mntent.h>
#else
#endif /* AFS_HPUX_ENV */
#endif
#endif
-#include <fcntl.h>
#ifndef AFS_NT40_ENV
#include <afs/osi_inode.h>
#endif
#include <afs/afsutil.h>
#include <afs/fileutil.h>
#include <afs/procmgmt.h> /* signal(), kill(), wait(), etc. */
-#ifndef AFS_NT40_ENV
-#include <syslog.h>
-#endif
+#include <afs/dir.h>
#include "nfs.h"
#include "lwp.h"
#include "salvsync.h"
#include "viceinode.h"
#include "salvage.h"
-#include "volinodes.h" /* header magic number, etc. stuff */
#include "vol-salvage.h"
+#include "common.h"
#ifdef AFS_NT40_ENV
#include <pthread.h>
#endif
+extern int ClientMode;
#if !defined(AFS_DEMAND_ATTACH_FS)
#error "online salvager only supported for demand attach fileserver"
#error "online salvager not supported on NT"
#endif /* AFS_NT40_ENV */
-
-/* Forward declarations */
-/*@printflike@*/ void Log(const char *format, ...);
-/*@printflike@*/ void Abort(const char *format, ...);
-
-
/*@+fcnmacros +macrofcndecl@*/
#ifdef O_LARGEFILE
#define afs_fopen fopen64
static void * SalvageChildReaperThread(void *);
static int DoSalvageVolume(struct SalvageQueueNode * node, int slot);
-static void SalvageServer(void);
+static void SalvageServer(int argc, char **argv, struct logOptions *logopts);
static void SalvageClient(VolumeId vid, char * pname);
static int Reap_Child(char * prog, int * pid, int * status);
static void * SalvageLogCleanupThread(void *);
-static int SalvageLogCleanup(int pid);
+static void SalvageLogCleanup(int pid);
+
+static void * SalvageLogScanningThread(void *);
+static void ScanLogs(struct rx_queue *log_watch_queue);
+
+struct cmdline_rock {
+ int argc;
+ char **argv;
+};
struct log_cleanup_node {
struct rx_queue q;
#define DEFAULT_PARALLELISM 4 /* allow 4 parallel salvage workers by default */
+enum optionsList {
+ OPT_partition,
+ OPT_volumeid,
+ OPT_debug,
+ OPT_nowrite,
+ OPT_inodes,
+ OPT_oktozap,
+ OPT_rootinodes,
+ OPT_salvagedirs,
+ OPT_blockreads,
+ OPT_parallel,
+ OPT_tmpdir,
+ OPT_orphans,
+ OPT_syslog,
+ OPT_syslogfacility,
+ OPT_logfile,
+ OPT_client,
+ OPT_transarc_logs
+};
+
static int
-handleit(struct cmd_syndesc *as, void *arock)
+handleit(struct cmd_syndesc *opts, void *arock)
{
- register struct cmd_item *ti;
- char pname[100], *temp;
- afs_int32 seenpart = 0, seenvol = 0, vid = 0, seenany = 0;
- struct DiskPartition *partP;
+ char pname[100];
+ afs_int32 seenpart = 0, seenvol = 0;
+ VolumeId vid = 0;
+ struct cmdline_rock *rock = (struct cmdline_rock *)arock;
+ char *optstring = NULL;
+ struct logOptions logopts;
+ memset(&logopts, 0, sizeof(logopts));
#ifdef AFS_SGI_VNODE_GLUE
if (afs_init_kernel_config(-1) < 0) {
}
#endif
- if (as->parms[2].items) /* -debug */
- debug = 1;
- if (as->parms[3].items) /* -nowrite */
- Testing = 1;
- if (as->parms[4].items) /* -inodes */
- ListInodeOption = 1;
- if (as->parms[5].items) /* -oktozap */
- OKToZap = 1;
- if (as->parms[6].items) /* -rootinodes */
- ShowRootFiles = 1;
- if (as->parms[8].items) /* -ForceReads */
- forceR = 1;
- if ((ti = as->parms[9].items)) { /* -Parallel # */
- temp = ti->data;
- if (strncmp(temp, "all", 3) == 0) {
+ cmd_OptionAsFlag(opts, OPT_debug, &debug);
+ cmd_OptionAsFlag(opts, OPT_nowrite, &Testing);
+ cmd_OptionAsFlag(opts, OPT_inodes, &ListInodeOption);
+ cmd_OptionAsFlag(opts, OPT_oktozap, &OKToZap);
+ cmd_OptionAsFlag(opts, OPT_rootinodes, &ShowRootFiles);
+ cmd_OptionAsFlag(opts, OPT_salvagedirs, &RebuildDirs);
+ cmd_OptionAsFlag(opts, OPT_blockreads, &forceR);
+ if (cmd_OptionAsString(opts, OPT_parallel, &optstring) == 0) {
+ if (strncmp(optstring, "all", 3) == 0) {
PartsPerDisk = 1;
- temp += 3;
}
- if (strlen(temp) != 0) {
- Parallel = atoi(temp);
+ if (strlen(optstring) != 0) {
+ Parallel = atoi(optstring);
if (Parallel < 1)
Parallel = 1;
if (Parallel > MAXPARALLEL) {
Parallel = MAXPARALLEL;
}
}
+ free(optstring);
+ optstring = NULL;
} else {
- Parallel = MIN(DEFAULT_PARALLELISM, MAXPARALLEL);
+ Parallel = min(DEFAULT_PARALLELISM, MAXPARALLEL);
}
- if ((ti = as->parms[10].items)) { /* -tmpdir */
+ if (cmd_OptionAsString(opts, OPT_tmpdir, &optstring) == 0) {
DIR *dirp;
-
- tmpdir = ti->data;
- dirp = opendir(tmpdir);
+ dirp = opendir(optstring);
if (!dirp) {
printf
("Can't open temporary placeholder dir %s; using current partition \n",
- tmpdir);
+ optstring);
tmpdir = NULL;
} else
closedir(dirp);
+ free(optstring);
+ optstring = NULL;
}
- if ((ti = as->parms[11].items)) /* -showlog */
- ShowLog = 1;
- if ((ti = as->parms[12].items)) { /* -orphans */
+ if (cmd_OptionAsString(opts, OPT_orphans, &optstring) == 0) {
if (Testing)
orphans = ORPH_IGNORE;
- else if (strcmp(ti->data, "remove") == 0
- || strcmp(ti->data, "r") == 0)
+ else if (strcmp(optstring, "remove") == 0
+ || strcmp(optstring, "r") == 0)
orphans = ORPH_REMOVE;
- else if (strcmp(ti->data, "attach") == 0
- || strcmp(ti->data, "a") == 0)
+ else if (strcmp(optstring, "attach") == 0
+ || strcmp(optstring, "a") == 0)
orphans = ORPH_ATTACH;
- }
-#ifndef AFS_NT40_ENV /* ignore options on NT */
- if ((ti = as->parms[13].items)) { /* -syslog */
- useSyslog = 1;
- ShowLog = 0;
- }
- if ((ti = as->parms[14].items)) { /* -syslogfacility */
- useSyslogFacility = atoi(ti->data);
+ free(optstring);
+ optstring = NULL;
}
- if ((ti = as->parms[15].items)) { /* -datelogs */
- TimeStampLogFile(AFSDIR_SERVER_SALSRVLOG_FILEPATH);
- }
+#ifdef HAVE_SYSLOG
+ if (cmd_OptionPresent(opts, OPT_syslog)) {
+ if (cmd_OptionPresent(opts, OPT_logfile)) {
+ fprintf(stderr, "Invalid options: -syslog and -logfile are exclusive.\n");
+ return -1;
+ }
+ if (cmd_OptionPresent(opts, OPT_transarc_logs)) {
+ fprintf(stderr, "Invalid options: -syslog and -transarc-logs are exclusive.\n");
+ return -1;
+ }
+ logopts.lopt_dest = logDest_syslog;
+ logopts.lopt_facility = LOG_DAEMON;
+ logopts.lopt_tag = "salvageserver";
+ cmd_OptionAsInt(opts, OPT_syslogfacility, &logopts.lopt_facility);
+ } else
#endif
+ {
+ logopts.lopt_dest = logDest_file;
+ if (cmd_OptionPresent(opts, OPT_transarc_logs)) {
+ logopts.lopt_rotateOnOpen = 1;
+ logopts.lopt_rotateStyle = logRotate_old;
+ }
+ if (cmd_OptionPresent(opts, OPT_logfile))
+ cmd_OptionAsString(opts, OPT_logfile, (char**)&logopts.lopt_filename);
+ else
+ logopts.lopt_filename = AFSDIR_SERVER_SALSRVLOG_FILEPATH;
+ }
- if ((ti = as->parms[16].items)) { /* -client */
- if ((ti = as->parms[0].items)) { /* -partition */
+ if (cmd_OptionPresent(opts, OPT_client)) {
+ if (cmd_OptionAsString(opts, OPT_partition, &optstring) == 0) {
seenpart = 1;
- strlcpy(pname, ti->data, sizeof(pname));
+ strlcpy(pname, optstring, sizeof(pname));
+ free(optstring);
+ optstring = NULL;
}
- if ((ti = as->parms[1].items)) { /* -volumeid */
+ if (cmd_OptionAsString(opts, OPT_volumeid, &optstring) == 0) {
+ char *end;
+ unsigned long vid_l;
seenvol = 1;
- vid = atoi(ti->data);
+ vid_l = strtoul(optstring, &end, 10);
+ if (vid_l >= MAX_AFS_UINT32 || vid_l == ULONG_MAX || *end != '\0') {
+ printf("Invalid volume id specified; salvage aborted\n");
+ exit(-1);
+ }
+ vid = (VolumeId)vid_l;
}
if (!seenpart || !seenvol) {
SalvageClient(vid, pname);
} else { /* salvageserver mode */
- SalvageServer();
+ SalvageServer(rock->argc, rock->argv, &logopts);
}
return (0);
}
pthread_t main_thread;
#endif
-static char commandLine[150];
-
int
main(int argc, char **argv)
{
struct cmd_syndesc *ts;
int err = 0;
-
- int i;
- extern char cml_version_number[];
+ struct cmdline_rock arock;
#ifdef AFS_AIX32_ENV
/*
- * The following signal action for AIX is necessary so that in case of a
- * crash (i.e. core is generated) we can include the user's data section
+ * The following signal action for AIX is necessary so that in case of a
+ * crash (i.e. core is generated) we can include the user's data section
* in the core dump. Unfortunately, by default, only a partial core is
* generated which, in many cases, isn't too useful.
*/
exit(2);
}
#ifdef AFS_NT40_ENV
+ /* Default to binary mode for fopen() */
+ _set_fmode(_O_BINARY);
+
main_thread = pthread_self();
if (spawnDatap && spawnDataLen) {
/* This is a child per partition salvager. Don't setup log or
exit(3);
} else {
#endif
- for (commandLine[0] = '\0', i = 0; i < argc; i++) {
- if (i > 0)
- strlcat(commandLine, " ", sizeof(commandLine));
- strlcat(commandLine, argv[i], sizeof(commandLine));
- }
#ifndef AFS_NT40_ENV
if (geteuid() != 0) {
}
#endif
- ts = cmd_CreateSyntax("initcmd", handleit, NULL, "initialize the program");
- cmd_AddParm(ts, "-partition", CMD_SINGLE, CMD_OPTIONAL,
- "Name of partition to salvage");
- cmd_AddParm(ts, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
- "Volume Id to salvage");
- cmd_AddParm(ts, "-debug", CMD_FLAG, CMD_OPTIONAL,
- "Run in Debugging mode");
- cmd_AddParm(ts, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
- "Run readonly/test mode");
- cmd_AddParm(ts, "-inodes", CMD_FLAG, CMD_OPTIONAL,
- "Just list affected afs inodes - debugging flag");
- cmd_AddParm(ts, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
- "Give permission to destroy bogus inodes/volumes - debugging flag");
- cmd_AddParm(ts, "-rootinodes", CMD_FLAG, CMD_OPTIONAL,
- "Show inodes owned by root - debugging flag");
- cmd_AddParm(ts, "-salvagedirs", CMD_FLAG, CMD_OPTIONAL,
- "Force rebuild/salvage of all directories");
- cmd_AddParm(ts, "-blockreads", CMD_FLAG, CMD_OPTIONAL,
- "Read smaller blocks to handle IO/bad blocks");
- cmd_AddParm(ts, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
- "# of max parallel partition salvaging");
- cmd_AddParm(ts, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
- "Name of dir to place tmp files ");
- cmd_AddParm(ts, "-showlog", CMD_FLAG, CMD_OPTIONAL,
- "Show log file upon completion");
- cmd_AddParm(ts, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
- "ignore | remove | attach");
-
- /* note - syslog isn't avail on NT, but if we make it conditional, have
- * to deal with screwy offsets for cmd params */
- cmd_AddParm(ts, "-syslog", CMD_FLAG, CMD_OPTIONAL,
- "Write salvage log to syslogs");
- cmd_AddParm(ts, "-syslogfacility", CMD_SINGLE, CMD_OPTIONAL,
- "Syslog facility number to use");
- cmd_AddParm(ts, "-datelogs", CMD_FLAG, CMD_OPTIONAL,
- "Include timestamp in logfile filename");
-
- cmd_AddParm(ts, "-client", CMD_FLAG, CMD_OPTIONAL,
+ arock.argc = argc;
+ arock.argv = argv;
+
+
+ ts = cmd_CreateSyntax("initcmd", handleit, &arock, 0, "initialize the program");
+ cmd_AddParmAtOffset(ts, OPT_partition, "-partition", CMD_SINGLE,
+ CMD_OPTIONAL, "Name of partition to salvage");
+ cmd_AddParmAtOffset(ts, OPT_volumeid, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
+ "Volume Id to salvage");
+ cmd_AddParmAtOffset(ts, OPT_debug, "-debug", CMD_FLAG, CMD_OPTIONAL,
+ "Run in Debugging mode");
+ cmd_AddParmAtOffset(ts, OPT_nowrite, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
+ "Run readonly/test mode");
+ cmd_AddParmAtOffset(ts, OPT_inodes, "-inodes", CMD_FLAG, CMD_OPTIONAL,
+ "Just list affected afs inodes - debugging flag");
+ cmd_AddParmAtOffset(ts, OPT_oktozap, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
+ "Give permission to destroy bogus inodes/volumes - debugging flag");
+ cmd_AddParmAtOffset(ts, OPT_rootinodes, "-rootinodes", CMD_FLAG,
+ CMD_OPTIONAL, "Show inodes owned by root - debugging flag");
+ cmd_AddParmAtOffset(ts, OPT_salvagedirs, "-salvagedirs", CMD_FLAG,
+ CMD_OPTIONAL, "Force rebuild/salvage of all directories");
+ cmd_AddParmAtOffset(ts, OPT_blockreads, "-blockreads", CMD_FLAG,
+ CMD_OPTIONAL, "Read smaller blocks to handle IO/bad blocks");
+ cmd_AddParmAtOffset(ts, OPT_parallel, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
+ "# of max parallel partition salvaging");
+ cmd_AddParmAtOffset(ts, OPT_tmpdir, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
+ "Name of dir to place tmp files ");
+ cmd_AddParmAtOffset(ts, OPT_orphans, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
+ "ignore | remove | attach");
+
+#ifdef HAVE_SYSLOG
+ cmd_AddParmAtOffset(ts, OPT_syslog, "-syslog", CMD_FLAG, CMD_OPTIONAL,
+ "Write salvage log to syslogs");
+ cmd_AddParmAtOffset(ts, OPT_syslogfacility, "-syslogfacility", CMD_SINGLE,
+ CMD_OPTIONAL, "Syslog facility number to use");
+#endif
+
+ cmd_AddParmAtOffset(ts, OPT_client, "-client", CMD_FLAG, CMD_OPTIONAL,
"Use SALVSYNC to ask salvageserver to salvage a volume");
+ cmd_AddParmAtOffset(ts, OPT_logfile, "-logfile", CMD_SINGLE, CMD_OPTIONAL,
+ "Location of log file ");
+
+ cmd_AddParmAtOffset(ts, OPT_transarc_logs, "-transarc-logs", CMD_FLAG,
+ CMD_OPTIONAL, "enable Transarc style logging");
+
err = cmd_Dispatch(argc, argv);
Exit(err);
- return 0; /* not reached */
+ AFS_UNREACHED(return 0);
}
static void
afs_int32 code;
SYNC_response res;
SALVSYNC_response_hdr sres;
-
- VInitVolumePackage(volumeUtility, 5, 5, DONT_CONNECT_FS, 0);
+ VolumePackageOptions opts;
+
+ /* Send Log() messages to stderr in client mode. */
+ ClientMode = 1;
+
+ VOptDefaults(volumeUtility, &opts);
+ if (VInitVolumePackage2(volumeUtility, &opts)) {
+ /* VInitVolumePackage2 can fail on e.g. partition attachment errors,
+ * but we don't really care, since all we're doing is trying to use
+ * SALVSYNC */
+ fprintf(stderr, "errors encountered initializing volume package, but "
+ "trying to continue anyway\n");
+ }
SALVSYNC_clientInit();
-
+
code = SALVSYNC_SalvageVolume(vid, pname, SALVSYNC_SALVAGE, SALVSYNC_OPERATOR, 0, NULL);
if (code != SYNC_OK) {
goto sync_error;
static int * child_slot;
static void
-SalvageServer(void)
+SalvageServer(int argc, char **argv, struct logOptions *logopts)
{
int pid, ret;
struct SalvageQueueNode * node;
pthread_t tid;
pthread_attr_t attrs;
int slot;
+ VolumePackageOptions opts;
/* All entries to the log will be appended. Useful if there are
* multiple salvagers appending to the log.
*/
+ OpenLog(logopts);
+ SetupLogSignals();
- CheckLogFile(AFSDIR_SERVER_SALSRVLOG_FILEPATH);
-#ifndef AFS_NT40_ENV
-#ifdef AFS_LINUX20_ENV
- fcntl(fileno(logFile), F_SETFL, O_APPEND); /* Isn't this redundant? */
-#else
- fcntl(fileno(logFile), F_SETFL, FAPPEND); /* Isn't this redundant? */
-#endif
-#endif
- setlinebuf(logFile);
-
- fprintf(logFile, "%s\n", cml_version_number);
- Log("Starting OpenAFS Online Salvage Server %s (%s)\n", SalvageVersion, commandLine);
-
+ Log("%s\n", cml_version_number);
+ LogCommandLine(argc, argv, "Online Salvage Server",
+ SalvageVersion, "Starting OpenAFS", Log);
/* Get and hold a lock for the duration of the salvage to make sure
* that no other salvage runs at the same time. The routine
- * VInitVolumePackage (called below) makes sure that a file server or
+ * VInitVolumePackage2 (called below) makes sure that a file server or
* other volume utilities don't interfere with the salvage.
*/
-
+
/* even demand attach online salvager
* still needs this because we don't want
* a stand-alone salvager to conflict with
* the salvager daemon */
- ObtainSalvageLock();
+ ObtainSharedSalvageLock();
+
+ child_slot = calloc(Parallel, sizeof(int));
+ opr_Assert(child_slot != NULL);
- child_slot = (int *) malloc(Parallel * sizeof(int));
- assert(child_slot != NULL);
- memset(child_slot, 0, Parallel * sizeof(int));
-
/* initialize things */
- VInitVolumePackage(salvageServer, 5, 5,
- 1, 0);
+ VOptDefaults(salvageServer, &opts);
+ if (VInitVolumePackage2(salvageServer, &opts)) {
+ Log("Shutting down: errors encountered initializing volume package\n");
+ Exit(1);
+ }
DInit(10);
queue_Init(&pending_q);
queue_Init(&log_cleanup_queue);
- assert(pthread_mutex_init(&worker_lock, NULL) == 0);
- assert(pthread_cond_init(&worker_cv, NULL) == 0);
- assert(pthread_cond_init(&log_cleanup_queue.queue_change_cv, NULL) == 0);
- assert(pthread_attr_init(&attrs) == 0);
+ opr_mutex_init(&worker_lock);
+ opr_cv_init(&worker_cv);
+ opr_cv_init(&log_cleanup_queue.queue_change_cv);
+ opr_Verify(pthread_attr_init(&attrs) == 0);
/* start up the reaper and log cleaner threads */
- assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
- assert(pthread_create(&tid,
- &attrs,
- &SalvageChildReaperThread,
- NULL) == 0);
- assert(pthread_create(&tid,
- &attrs,
- &SalvageLogCleanupThread,
- NULL) == 0);
+ opr_Verify(pthread_attr_setdetachstate(&attrs,
+ PTHREAD_CREATE_DETACHED) == 0);
+ opr_Verify(pthread_create(&tid, &attrs,
+ &SalvageChildReaperThread, NULL) == 0);
+ opr_Verify(pthread_create(&tid, &attrs,
+ &SalvageLogCleanupThread, NULL) == 0);
+ opr_Verify(pthread_create(&tid, &attrs,
+ &SalvageLogScanningThread, NULL) == 0);
/* loop forever serving requests */
while (1) {
node = SALVSYNC_getWork();
- assert(node != NULL);
+ opr_Assert(node != NULL);
Log("dispatching child to salvage volume %u...\n",
node->command.sop.parent);
if (!child_slot[slot])
break;
}
- assert (slot < Parallel);
+ opr_Assert (slot < Parallel);
+ do_fork:
pid = Fork();
if (pid == 0) {
VOL_UNLOCK;
ret = DoSalvageVolume(node, slot);
Exit(ret);
} else if (pid < 0) {
- VOL_UNLOCK;
- SALVSYNC_doneWork(node, 1);
+ Log("failed to fork child worker process\n");
+ sleep(1);
+ goto do_fork;
} else {
child_slot[slot] = pid;
node->pid = pid;
VOL_UNLOCK;
-
- assert(pthread_mutex_lock(&worker_lock) == 0);
+
+ opr_mutex_enter(&worker_lock);
current_workers++;
-
+
/* let the reaper thread know another worker was spawned */
- assert(pthread_cond_broadcast(&worker_cv) == 0);
-
+ opr_cv_broadcast(&worker_cv);
+
/* if we're overquota, wait for the reaper */
while (current_workers >= Parallel) {
- assert(pthread_cond_wait(&worker_cv, &worker_lock) == 0);
+ opr_cv_wait(&worker_cv, &worker_lock);
}
- assert(pthread_mutex_unlock(&worker_lock) == 0);
+ opr_mutex_exit(&worker_lock);
}
}
}
static int
DoSalvageVolume(struct SalvageQueueNode * node, int slot)
{
- char childLog[AFSDIR_PATH_MAX];
- int ret;
- struct DiskPartition * partP;
+ char *filename = NULL;
+ struct logOptions logopts;
+ struct DiskPartition64 * partP;
/* do not allow further forking inside salvager */
canfork = 0;
- /* do not attempt to close parent's logFile handle as
- * another thread may have held the lock on the FILE
- * structure when fork was called! */
-
- afs_snprintf(childLog, sizeof(childLog), "%s.%d",
- AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid());
-
- logFile = afs_fopen(childLog, "a");
- if (!logFile) { /* still nothing, use stdout */
- logFile = stdout;
- ShowLog = 0;
+ /*
+ * Do not attempt to close parent's log file handle as
+ * another thread may have held the lock when fork was
+ * called!
+ */
+ memset(&logopts, 0, sizeof(logopts));
+ logopts.lopt_dest = logDest_file;
+ logopts.lopt_rotateStyle = logRotate_none;
+ if (asprintf(&filename, "%s.%d",
+ AFSDIR_SERVER_SLVGLOG_FILEPATH, getpid()) < 0) {
+ fprintf(stderr, "out of memory\n");
+ return ENOMEM;
}
+ logopts.lopt_filename = filename;
+ OpenLog(&logopts);
+ free(filename);
if (node->command.sop.parent <= 0) {
Log("salvageServer: invalid volume id specified; salvage aborted\n");
return 1;
}
-
+
partP = VGetPartition(node->command.sop.partName, 0);
if (!partP) {
- Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
+ Log("salvageServer: Unknown or unmounted partition %s; salvage aborted\n",
node->command.sop.partName);
return 1;
}
+ /* obtain a shared salvage lock in the child worker, so if the
+ * salvageserver restarts (and we continue), we will still hold a lock and
+ * prevent standalone salvagers from interfering */
+ ObtainSharedSalvageLock();
+
/* Salvage individual volume; don't notify fs */
SalvageFileSys1(partP, node->command.sop.parent);
- fclose(logFile);
+ CloseLog();
return 0;
}
static void *
SalvageChildReaperThread(void * args)
{
- int slot, pid, status, code, found;
- struct SalvageQueueNode *qp, *nqp;
+ int slot, pid, status;
struct log_cleanup_node * cleanup;
- assert(pthread_mutex_lock(&worker_lock) == 0);
+ opr_mutex_enter(&worker_lock);
/* loop reaping our children */
while (1) {
/* wait() won't block unless we have children, so
* block on the cond var if we're childless */
while (current_workers == 0) {
- assert(pthread_cond_wait(&worker_cv, &worker_lock) == 0);
+ opr_cv_wait(&worker_cv, &worker_lock);
}
- assert(pthread_mutex_unlock(&worker_lock) == 0);
+ opr_mutex_exit(&worker_lock);
- cleanup = (struct log_cleanup_node *) malloc(sizeof(struct log_cleanup_node));
+ cleanup = malloc(sizeof(struct log_cleanup_node));
while (Reap_Child("salvageserver", &pid, &status) < 0) {
/* try to prevent livelock if something goes wrong */
if (child_slot[slot] == pid)
break;
}
- assert(slot < Parallel);
+ opr_Assert(slot < Parallel);
child_slot[slot] = 0;
VOL_UNLOCK;
- assert(pthread_mutex_lock(&worker_lock) == 0);
+ SALVSYNC_doneWorkByPid(pid, status);
+
+ opr_mutex_enter(&worker_lock);
if (cleanup) {
cleanup->pid = pid;
queue_Append(&log_cleanup_queue, cleanup);
- assert(pthread_cond_signal(&log_cleanup_queue.queue_change_cv) == 0);
+ opr_cv_signal(&log_cleanup_queue.queue_change_cv);
}
/* ok, we've reaped a child */
current_workers--;
- SALVSYNC_doneWorkByPid(pid, WEXITSTATUS(status));
- assert(pthread_cond_broadcast(&worker_cv) == 0);
+ opr_cv_broadcast(&worker_cv);
}
- return NULL;
+ AFS_UNREACHED(return(NULL));
}
static int
{
struct log_cleanup_node * cleanup;
- assert(pthread_mutex_lock(&worker_lock) == 0);
+ opr_mutex_enter(&worker_lock);
while (1) {
while (queue_IsEmpty(&log_cleanup_queue)) {
- assert(pthread_cond_wait(&log_cleanup_queue.queue_change_cv, &worker_lock) == 0);
+ opr_cv_wait(&log_cleanup_queue.queue_change_cv, &worker_lock);
}
while (queue_IsNotEmpty(&log_cleanup_queue)) {
cleanup = queue_First(&log_cleanup_queue, log_cleanup_node);
queue_Remove(cleanup);
- assert(pthread_mutex_unlock(&worker_lock) == 0);
+ opr_mutex_exit(&worker_lock);
SalvageLogCleanup(cleanup->pid);
free(cleanup);
- assert(pthread_mutex_lock(&worker_lock) == 0);
- }
+ opr_mutex_enter(&worker_lock);
+ }
}
- assert(pthread_mutex_unlock(&worker_lock) == 0);
- return NULL;
+ AFS_UNREACHED(opr_mutex_exit(&worker_lock));
+ AFS_UNREACHED(return(NULL));
}
#define LOG_XFER_BUF_SIZE 65536
-static int
+static void
SalvageLogCleanup(int pid)
{
int pidlog, len;
- char fn[AFSDIR_PATH_MAX];
- static char buf[LOG_XFER_BUF_SIZE];
+ char *fn = NULL;
+ char *buf = NULL;
- afs_snprintf(fn, sizeof(fn), "%s.%d",
- AFSDIR_SERVER_SLVGLOG_FILEPATH, pid);
-
+ if (asprintf(&fn, "%s.%d", AFSDIR_SERVER_SLVGLOG_FILEPATH, pid) < 0) {
+ Log("Unable to write child log: out of memory\n");
+ goto done;
+ }
+
+ buf = calloc(1, LOG_XFER_BUF_SIZE);
+ if (buf == NULL) {
+ Log("Unable to write child log: out of memory\n");
+ goto done;
+ }
pidlog = open(fn, O_RDONLY);
unlink(fn);
if (pidlog < 0)
- return 1;
+ goto done;
len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
while (len) {
- fwrite(buf, len, 1, logFile);
+ WriteLogBuffer(buf, len);
len = read(pidlog, buf, LOG_XFER_BUF_SIZE);
}
close(pidlog);
- return 0;
+ done:
+ free(fn);
+ free(buf);
+}
+
+/* wake up every five minutes to see if a non-child salvage has finished */
+#define SALVAGE_SCAN_POLL_INTERVAL 300
+
+/**
+ * Thread to look for SalvageLog.$pid files that are not from our child
+ * worker salvagers, and notify SalvageLogCleanupThread to clean them
+ * up. This can happen if we restart during salvages, or the
+ * salvageserver crashes or something.
+ *
+ * @param arg unused
+ *
+ * @return always NULL
+ */
+static void *
+SalvageLogScanningThread(void * arg)
+{
+ struct rx_queue log_watch_queue;
+ char *prefix;
+ int prefix_len;
+
+ queue_Init(&log_watch_queue);
+
+ prefix_len = asprintf(&prefix, "%s.", AFSDIR_SLVGLOG_FILE);
+ if (prefix_len >= 0) {
+ DIR *dp;
+ struct dirent *dirp;
+
+ dp = opendir(AFSDIR_LOGS_DIR);
+ opr_Assert(dp);
+
+ while ((dirp = readdir(dp)) != NULL) {
+ pid_t pid;
+ struct log_cleanup_node *cleanup;
+ int i;
+
+ if (strncmp(dirp->d_name, prefix, prefix_len) != 0) {
+ /* not a salvage logfile; skip */
+ continue;
+ }
+
+ errno = 0;
+ pid = strtol(dirp->d_name + prefix_len, NULL, 10);
+
+ if (errno != 0) {
+ /* file is SalvageLog.<something> but <something> isn't
+ * a pid, so skip */
+ continue;
+ }
+
+ VOL_LOCK;
+ for (i = 0; i < Parallel; ++i) {
+ if (pid == child_slot[i]) {
+ break;
+ }
+ }
+ VOL_UNLOCK;
+ if (i < Parallel) {
+ /* this pid is one of our children, so the reaper thread
+ * will take care of it; skip */
+ continue;
+ }
+
+ cleanup = malloc(sizeof(struct log_cleanup_node));
+ cleanup->pid = pid;
+
+ queue_Append(&log_watch_queue, cleanup);
+ }
+ free(prefix);
+ closedir(dp);
+ }
+
+ ScanLogs(&log_watch_queue);
+
+ while (queue_IsNotEmpty(&log_watch_queue)) {
+ sleep(SALVAGE_SCAN_POLL_INTERVAL);
+ ScanLogs(&log_watch_queue);
+ }
+
+ return NULL;
+}
+
+/**
+ * look through log_watch_queue, and if any processes are not still
+ * running, hand them off to the SalvageLogCleanupThread
+ *
+ * @param log_watch_queue a queue of PIDs that we should clean up if
+ * that PID has died
+ */
+static void
+ScanLogs(struct rx_queue *log_watch_queue)
+{
+ struct log_cleanup_node *cleanup, *next;
+
+ opr_mutex_enter(&worker_lock);
+
+ for (queue_Scan(log_watch_queue, cleanup, next, log_cleanup_node)) {
+ /* if a process is still running, assume it's the salvage process
+ * still going, and keep waiting for it */
+ if (kill(cleanup->pid, 0) < 0 && errno == ESRCH) {
+ queue_Remove(cleanup);
+ queue_Append(&log_cleanup_queue, cleanup);
+ opr_cv_signal(&log_cleanup_queue.queue_change_cv);
+ }
+ }
+
+ opr_mutex_exit(&worker_lock);
}