/*
* Copyright 2000, International Business Machines Corporation 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
#include <afsconfig.h>
#include <afs/param.h>
-RCSID
- ("$Header$");
+#include <afs/procmgmt.h>
+#include <roken.h>
#include <stddef.h>
-#include <stdlib.h>
-#include <stdio.h>
-#include <errno.h>
-#include <sys/types.h>
-#ifdef AFS_NT40_ENV
-#include <io.h>
-#else
-#include <sys/file.h>
-#include <sys/time.h>
-#endif
-#ifdef BOZO_SAVE_CORES
-#include <time.h>
-#endif
-#include <sys/stat.h>
-#include <string.h>
-#include <afs/procmgmt.h> /* signal(), kill(), wait(), etc. */
#include <lwp.h>
+#include <rx/rx.h>
#include <afs/audit.h>
#include <afs/afsutil.h>
#include <afs/fileutil.h>
+
#include "bnode.h"
+#include "bosprototypes.h"
-#if defined(AFS_AIX_ENV) || defined(AFS_SUN4_ENV)
-/* All known versions of AIX lack WCOREDUMP but this works */
-#define WCOREDUMP(x) ((x) & 0x80)
+#ifndef WCOREDUMP
+#define WCOREDUMP(x) ((x) & 0200)
#endif
#define BNODE_LWP_STACKSIZE (16 * 1024)
+#define BNODE_ERROR_COUNT_MAX 16 /* maximum number of retries */
int bnode_waiting = 0;
static PROCESS bproc_pid; /* pid of waker-upper */
int weirdPids;
} bnode_stats;
+extern const char *DoCore;
+extern const char *DoPidFiles;
#ifndef AFS_NT40_ENV
extern char **environ; /* env structure */
#endif
+int hdl_notifier(struct bnode_proc *tp);
+
/* Remember the name of the process, if any, that failed last */
static void
-RememberProcName(register struct bnode_proc *ap)
+RememberProcName(struct bnode_proc *ap)
{
- register struct bnode *tbnodep;
+ struct bnode *tbnodep;
tbnodep = ap->bnode;
if (tbnodep->lastErrorName) {
free(tbnodep->lastErrorName);
tbnodep->lastErrorName = NULL;
}
- if (ap->coreName) {
- tbnodep->lastErrorName = (char *)malloc(strlen(ap->coreName) + 1);
- strcpy(tbnodep->lastErrorName, ap->coreName);
- }
+ if (ap->coreName)
+ tbnodep->lastErrorName = strdup(ap->coreName);
}
/* utility for use by BOP_HASCORE functions to determine where a core file might
* be stored.
*/
int
-bnode_CoreName(register struct bnode *abnode, char *acoreName, char *abuffer)
+bnode_CoreName(struct bnode *abnode, char *acoreName, char *abuffer)
{
- strcpy(abuffer, AFSDIR_SERVER_CORELOG_FILEPATH);
+ if (DoCore) {
+ strcpy(abuffer, DoCore);
+ strcat(abuffer, "/");
+ strcat(abuffer, AFSDIR_CORE_FILE);
+ } else
+ strcpy(abuffer, AFSDIR_SERVER_CORELOG_FILEPATH);
if (acoreName) {
strcat(abuffer, acoreName);
strcat(abuffer, ".");
/* save core file, if any */
static void
-SaveCore(register struct bnode *abnode, register struct bnode_proc
+SaveCore(struct bnode *abnode, struct bnode_proc
*aproc)
{
char tbuffer[256];
struct stat tstat;
- register afs_int32 code;
+ afs_int32 code = 0;
+ char *corefile = NULL;
#ifdef BOZO_SAVE_CORES
struct timeval Start;
struct tm *TimeFields;
char FileName[256];
#endif
- code = stat(AFSDIR_SERVER_CORELOG_FILEPATH, &tstat);
+ /* Linux always appends the PID to core dumps from threaded processes, so
+ * we have to scan the directory to find core files under another name. */
+ if (DoCore) {
+ strcpy(tbuffer, DoCore);
+ strcat(tbuffer, "/");
+ strcat(tbuffer, AFSDIR_CORE_FILE);
+ } else
+ code = stat(AFSDIR_SERVER_CORELOG_FILEPATH, &tstat);
+ if (code) {
+ DIR *logdir;
+ struct dirent *file;
+ unsigned long pid;
+ const char *coredir = AFSDIR_LOGS_DIR;
+
+ if (DoCore)
+ coredir = DoCore;
+
+ logdir = opendir(coredir);
+ if (logdir == NULL)
+ return;
+ while ((file = readdir(logdir)) != NULL) {
+ if (strncmp(file->d_name, "core.", 5) != 0)
+ continue;
+ pid = atol(file->d_name + 5);
+ if (pid == aproc->pid) {
+ asprintf(&corefile, "%s/%s", coredir, file->d_name);
+ if (corefile == NULL) {
+ closedir(logdir);
+ return;
+ }
+ code = 0;
+ break;
+ }
+ }
+ closedir(logdir);
+ } else {
+ corefile = strdup(tbuffer);
+ }
if (code)
return;
bnode_CoreName(abnode, aproc->coreName, tbuffer);
#ifdef BOZO_SAVE_CORES
- TM_GetTimeOfDay(&Start, 0);
+ FT_GetTimeOfDay(&Start, 0);
TimeFields = localtime(&Start.tv_sec);
sprintf(FileName, "%s.%d%02d%02d%02d%02d%02d", tbuffer,
- TimeFields->tm_year, TimeFields->tm_mon + 1, TimeFields->tm_mday,
+ TimeFields->tm_year + 1900, TimeFields->tm_mon + 1, TimeFields->tm_mday,
TimeFields->tm_hour, TimeFields->tm_min, TimeFields->tm_sec);
strcpy(tbuffer, FileName);
#endif
- code = renamefile(AFSDIR_SERVER_CORELOG_FILEPATH, tbuffer);
+ code = renamefile(corefile, tbuffer);
+ free(corefile);
}
int
-bnode_GetString(register struct bnode *abnode, register char *abuffer,
- register afs_int32 alen)
+bnode_GetString(struct bnode *abnode, char *abuffer,
+ afs_int32 alen)
{
return BOP_GETSTRING(abnode, abuffer, alen);
}
int
-bnode_GetParm(register struct bnode *abnode, register afs_int32 aindex,
- register char *abuffer, afs_int32 alen)
+bnode_GetParm(struct bnode *abnode, afs_int32 aindex,
+ char *abuffer, afs_int32 alen)
{
return BOP_GETPARM(abnode, aindex, abuffer, alen);
}
int
-bnode_GetStat(register struct bnode *abnode, register afs_int32 * astatus)
+bnode_GetStat(struct bnode *abnode, afs_int32 * astatus)
{
return BOP_GETSTAT(abnode, astatus);
}
int
-bnode_RestartP(register struct bnode *abnode)
+bnode_RestartP(struct bnode *abnode)
{
return BOP_RESTARTP(abnode);
}
static int
-bnode_Check(register struct bnode *abnode)
+bnode_Check(struct bnode *abnode)
{
if (abnode->flags & BNODE_WAIT) {
abnode->flags &= ~BNODE_WAIT;
/* tell if an instance has a core file */
int
-bnode_HasCore(register struct bnode *abnode)
+bnode_HasCore(struct bnode *abnode)
{
return BOP_HASCORE(abnode);
}
/* wait for all bnodes to stabilize */
int
-bnode_WaitAll()
+bnode_WaitAll(void)
{
- register struct bnode *tb;
- register afs_int32 code;
+ struct bnode *tb;
+ afs_int32 code;
afs_int32 stat;
retry:
/* wait until bnode status is correct */
int
-bnode_WaitStatus(register struct bnode *abnode, int astatus)
+bnode_WaitStatus(struct bnode *abnode, int astatus)
{
- register afs_int32 code;
+ afs_int32 code;
afs_int32 stat;
bnode_Hold(abnode);
}
int
-bnode_SetStat(register struct bnode *abnode, register int agoal)
+bnode_ResetErrorCount(struct bnode *abnode)
+{
+ abnode->errorStopCount = 0;
+ abnode->errorStopDelay = 0;
+ return 0;
+}
+
+int
+bnode_SetStat(struct bnode *abnode, int agoal)
{
abnode->goal = agoal;
bnode_Check(abnode);
}
int
-bnode_SetGoal(register struct bnode *abnode, register int agoal)
+bnode_SetGoal(struct bnode *abnode, int agoal)
{
abnode->goal = agoal;
bnode_Check(abnode);
}
int
-bnode_SetFileGoal(register struct bnode *abnode, register int agoal)
+bnode_SetFileGoal(struct bnode *abnode, int agoal)
{
if (abnode->fileGoal == agoal)
return 0; /* already done */
/* apply a function to all bnodes in the system */
int
-bnode_ApplyInstance(int (*aproc) (), char *arock)
+bnode_ApplyInstance(int (*aproc) (struct bnode *tb, void *), void *arock)
{
- register struct bnode *tb, *nb;
- register afs_int32 code;
+ struct bnode *tb, *nb;
+ afs_int32 code;
for (tb = allBnodes; tb; tb = nb) {
nb = tb->next;
}
struct bnode *
-bnode_FindInstance(register char *aname)
+bnode_FindInstance(char *aname)
{
- register struct bnode *tb;
+ struct bnode *tb;
for (tb = allBnodes; tb; tb = tb->next) {
if (!strcmp(tb->name, aname))
}
static struct bnode_type *
-FindType(register char *aname)
+FindType(char *aname)
{
- register struct bnode_type *tt;
+ struct bnode_type *tt;
for (tt = allTypes; tt; tt = tt->next) {
if (!strcmp(tt->name, aname))
int
bnode_Register(char *atype, struct bnode_ops *aprocs, int anparms)
{
- register struct bnode_type *tt;
+ struct bnode_type *tt;
for (tt = allTypes; tt; tt = tt->next) {
if (!strcmp(tt->name, atype))
*abp = tb;
tb->type = type;
- /* The fs_create above calls bnode_InitBnode() which always sets the
+ /* The fs_create above calls bnode_InitBnode() which always sets the
** fileGoal to BSTAT_NORMAL .... overwrite it with whatever is passed into
** this function as a parameter... */
tb->fileGoal = fileGoal;
int
bnode_DeleteName(char *ainstance)
{
- register struct bnode *tb;
+ struct bnode *tb;
tb = bnode_FindInstance(ainstance);
if (!tb)
}
int
-bnode_Hold(register struct bnode *abnode)
+bnode_Hold(struct bnode *abnode)
{
abnode->refCount++;
return 0;
}
int
-bnode_Release(register struct bnode *abnode)
+bnode_Release(struct bnode *abnode)
{
abnode->refCount--;
if (abnode->refCount == 0 && abnode->flags & BNODE_DELETE) {
}
int
-bnode_Delete(register struct bnode *abnode)
+bnode_Delete(struct bnode *abnode)
{
- register afs_int32 code;
- register struct bnode **lb, *ub;
+ afs_int32 code;
+ struct bnode **lb, *ub;
afs_int32 temp;
if (abnode->refCount != 0) {
/* function to tell if there's a timeout coming up */
int
-bnode_PendingTimeout(register struct bnode *abnode)
+bnode_PendingTimeout(struct bnode *abnode)
{
return (abnode->flags & BNODE_NEEDTIMEOUT);
}
/* function called to set / clear periodic bnode wakeup times */
int
-bnode_SetTimeout(register struct bnode *abnode, afs_int32 atimeout)
+bnode_SetTimeout(struct bnode *abnode, afs_int32 atimeout)
{
if (atimeout != 0) {
abnode->nextTimeout = FT_ApproxTime() + atimeout;
/* used by new bnode creation code to format bnode header */
int
-bnode_InitBnode(register struct bnode *abnode, struct bnode_ops *abnodeops,
+bnode_InitBnode(struct bnode *abnode, struct bnode_ops *abnodeops,
char *aname)
{
struct bnode **lb, *nb;
/* format the bnode properly */
memset(abnode, 0, sizeof(struct bnode));
abnode->ops = abnodeops;
- abnode->name = (char *)malloc(strlen(aname) + 1);
+ abnode->name = strdup(aname);
if (!abnode->name)
return ENOMEM;
- strcpy(abnode->name, aname);
abnode->flags = BNODE_ACTIVE;
abnode->fileGoal = BSTAT_NORMAL;
abnode->goal = BSTAT_SHUTDOWN;
}
static int
-DeleteProc(register struct bnode_proc *abproc)
+DeleteProc(struct bnode_proc *abproc)
{
- register struct bnode_proc **pb, *tb;
+ struct bnode_proc **pb, *tb;
struct bnode_proc *nb;
for (pb = &allProcs, tb = *pb; tb; pb = &tb->next, tb = nb) {
}
/* bnode lwp executes this code repeatedly */
-static int
-bproc()
+static void *
+bproc(void *unused)
{
- register afs_int32 code;
- register struct bnode *tb;
- register afs_int32 temp;
- register struct bnode_proc *tp;
+ afs_int32 code;
+ struct bnode *tb;
+ afs_int32 temp;
+ struct bnode_proc *tp;
struct bnode *nb;
int options; /* must not be register */
struct timeval tv;
tb = tp->bnode;
bnode_Hold(tb);
- /* count restarts in last 10 seconds */
+ /* count restarts in last 30 seconds */
if (temp > tb->rsTime + 30) {
- /* it's been 10 seconds we've been counting */
+ /* it's been 30 seconds we've been counting */
tb->rsTime = temp;
tb->rsCount = 0;
}
+
if (WIFSIGNALED(status) == 0) {
/* exited, not signalled */
tp->lastExit = WEXITSTATUS(status);
tb->notifier);
hdl_notifier(tp);
}
- BOP_PROCEXIT(tb, tp);
- bnode_Check(tb);
- if (tb->rsCount++ > 10) {
- /* 10 in 10 seconds */
+ if (tb->goal && tb->rsCount++ > 10) {
+ /* 10 in 30 seconds */
+ if (tb->errorStopCount >= BNODE_ERROR_COUNT_MAX) {
+ tb->errorStopDelay = 0; /* max reached, give up. */
+ } else {
+ tb->errorStopCount++;
+ if (!tb->errorStopDelay) {
+ tb->errorStopDelay = 1;
+ } else {
+ tb->errorStopDelay *= 2;
+ }
+ }
tb->flags |= BNODE_ERRORSTOP;
bnode_SetGoal(tb, BSTAT_SHUTDOWN);
bozo_Log
("BNODE '%s' repeatedly failed to start, perhaps missing executable.\n",
tb->name);
}
+ BOP_PROCEXIT(tb, tp);
+ bnode_Check(tb);
bnode_Release(tb); /* bnode delete can happen here */
DeleteProc(tp);
} else
}
}
}
+ return NULL;
}
static afs_int32
-SendNotifierData(register int fd, register struct bnode_proc *tp)
+SendNotifierData(int fd, struct bnode_proc *tp)
{
- register struct bnode *tb = tp->bnode;
+ struct bnode *tb = tp->bnode;
char buffer[1000], *bufp = buffer, *buf1;
- register int len;
+ int len;
/*
* First sent out the bnode_proc struct
buf1 = "(null)";
(void)sprintf(bufp, "coreName: %s\n", buf1);
bufp += strlen(bufp);
- (void)sprintf(bufp, "pid: %ld\n", tp->pid);
+ (void)sprintf(bufp, "pid: %ld\n", afs_printable_int32_ld(tp->pid));
bufp += strlen(bufp);
- (void)sprintf(bufp, "lastExit: %ld\n", tp->lastExit);
+ (void)sprintf(bufp, "lastExit: %ld\n", afs_printable_int32_ld(tp->lastExit));
bufp += strlen(bufp);
#ifdef notdef
- (void)sprintf(bufp, "lastSignal: %ld\n", tp->lastSignal);
+ (void)sprintf(bufp, "lastSignal: %ld\n", afs_printable_int32_ld(tp->lastSignal));
bufp += strlen(bufp);
#endif
- (void)sprintf(bufp, "flags: %ld\n", tp->flags);
+ (void)sprintf(bufp, "flags: %ld\n", afs_printable_int32_ld(tp->flags));
bufp += strlen(bufp);
(void)sprintf(bufp, "END bnode_proc\n");
bufp += strlen(bufp);
bufp += strlen(bufp);
(void)sprintf(bufp, "name: %s\n", tb->name);
bufp += strlen(bufp);
- (void)sprintf(bufp, "rsTime: %ld\n", tb->rsTime);
+ (void)sprintf(bufp, "rsTime: %ld\n", afs_printable_int32_ld(tb->rsTime));
bufp += strlen(bufp);
- (void)sprintf(bufp, "rsCount: %ld\n", tb->rsCount);
+ (void)sprintf(bufp, "rsCount: %ld\n", afs_printable_int32_ld(tb->rsCount));
bufp += strlen(bufp);
- (void)sprintf(bufp, "procStartTime: %ld\n", tb->procStartTime);
+ (void)sprintf(bufp, "procStartTime: %ld\n", afs_printable_int32_ld(tb->procStartTime));
bufp += strlen(bufp);
- (void)sprintf(bufp, "procStarts: %ld\n", tb->procStarts);
+ (void)sprintf(bufp, "procStarts: %ld\n", afs_printable_int32_ld(tb->procStarts));
bufp += strlen(bufp);
- (void)sprintf(bufp, "lastAnyExit: %ld\n", tb->lastAnyExit);
+ (void)sprintf(bufp, "lastAnyExit: %ld\n", afs_printable_int32_ld(tb->lastAnyExit));
bufp += strlen(bufp);
- (void)sprintf(bufp, "lastErrorExit: %ld\n", tb->lastErrorExit);
+ (void)sprintf(bufp, "lastErrorExit: %ld\n", afs_printable_int32_ld(tb->lastErrorExit));
bufp += strlen(bufp);
- (void)sprintf(bufp, "errorCode: %ld\n", tb->errorCode);
+ (void)sprintf(bufp, "errorCode: %ld\n", afs_printable_int32_ld(tb->errorCode));
bufp += strlen(bufp);
- (void)sprintf(bufp, "errorSignal: %ld\n", tb->errorSignal);
+ (void)sprintf(bufp, "errorSignal: %ld\n", afs_printable_int32_ld(tb->errorSignal));
bufp += strlen(bufp);
/*
(void) sprintf(bufp, "lastErrorName: %s\n", tb->lastErrorName);
if (write(fd, buffer, len) < 0) {
return -1;
}
+ return 0;
}
int
hdl_notifier(struct bnode_proc *tp)
{
#ifndef AFS_NT40_ENV /* NT notifier callout not yet implemented */
- int code, pid, status;
+ int pid;
struct stat tstat;
if (stat(tp->bnode->notifier, &tstat)) {
if ((pid = fork()) == 0) {
FILE *fout;
struct bnode *tb = tp->bnode;
- int ec;
#if defined(AFS_HPUX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_SGI51_ENV)
- ec = setsid();
+ setsid();
#elif defined(AFS_DARWIN90_ENV)
- ec = setpgid(0, 0);
-#elif defined(AFS_LINUX20_ENV) || defined(AFS_AIX_ENV)
- ec = setpgrp();
+ setpgid(0, 0);
+#elif defined(AFS_LINUX20_ENV) || defined(AFS_AIX_ENV)
+ setpgrp();
#else
- ec = setpgrp(0, 0);
+ setpgrp(0, 0);
#endif
fout = popen(tb->notifier, "w");
if (fout == NULL) {
perror(tb->notifier);
exit(1);
}
- code = SendNotifierData(fileno(fout), tp);
+ SendNotifierData(fileno(fout), tp);
pclose(fout);
exit(0);
} else if (pid < 0) {
/* Called by IOMGR at low priority on IOMGR's stack shortly after a SIGCHLD
* occurs. Wakes up bproc do redo things */
-int
-bnode_SoftInt(int asignal)
+void *
+bnode_SoftInt(void *param)
{
+ /* int asignal = (int) param; */
+
IOMGR_Cancel(bproc_pid);
return 0;
}
void
bnode_Int(int asignal)
{
- extern void bozo_ShutdownAndExit();
-
- if (asignal == SIGQUIT) {
- IOMGR_SoftSig(bozo_ShutdownAndExit, (char *)asignal);
+ if (asignal == SIGQUIT || asignal == SIGTERM) {
+ IOMGR_SoftSig(bozo_ShutdownAndExit, (void *)(intptr_t)asignal);
} else {
- IOMGR_SoftSig(bnode_SoftInt, (char *)asignal);
+ IOMGR_SoftSig(bnode_SoftInt, (void *)(intptr_t)asignal);
}
}
/* intialize the whole system */
int
-bnode_Init()
+bnode_Init(void)
{
PROCESS junk;
- register afs_int32 code;
+ afs_int32 code;
struct sigaction newaction;
- static initDone = 0;
+ static int initDone = 0;
if (initDone)
return 0;
"bnode-manager", &bproc_pid);
if (code)
return code;
- memset((char *)&newaction, 0, sizeof(newaction));
+ memset(&newaction, 0, sizeof(newaction));
newaction.sa_handler = bnode_Int;
code = sigaction(SIGCHLD, &newaction, NULL);
if (code)
code = sigaction(SIGQUIT, &newaction, NULL);
if (code)
return errno;
+ code = sigaction(SIGTERM, &newaction, NULL);
+ if (code)
+ return errno;
return code;
}
/* free token list returned by parseLine */
int
-bnode_FreeTokens(register struct bnode_token *alist)
+bnode_FreeTokens(struct bnode_token *alist)
{
- register struct bnode_token *nlist;
+ struct bnode_token *nlist;
for (; alist; alist = nlist) {
nlist = alist->next;
free(alist->key);
bnode_ParseLine(char *aline, struct bnode_token **alist)
{
char tbuffer[256];
- register char *tptr;
+ char *tptr = NULL;
int inToken;
struct bnode_token *first, *last;
- register struct bnode_token *ttok;
- register int tc;
+ struct bnode_token *ttok;
+ int tc;
inToken = 0; /* not copying token chars at start */
first = (struct bnode_token *)0;
ttok =
(struct bnode_token *)malloc(sizeof(struct bnode_token));
ttok->next = (struct bnode_token *)0;
- ttok->key = (char *)malloc(strlen(tbuffer) + 1);
- strcpy(ttok->key, tbuffer);
+ ttok->key = strdup(tbuffer);
if (last) {
last->next = ttok;
last = ttok;
free(tp);
return errno;
}
+ bozo_Log("%s started pid %ld: %s\n", abnode->name, cpid, aexecString);
bnode_FreeTokens(tlist);
allProcs = tp;
tp->pid = cpid;
tp->flags = BPROC_STARTED;
tp->flags &= ~BPROC_EXITED;
+ BOP_PROCSTARTED(abnode, tp);
bnode_Check(abnode);
return 0;
}
int
-bnode_StopProc(register struct bnode_proc *aproc, int asignal)
+bnode_StopProc(struct bnode_proc *aproc, int asignal)
{
- register int code;
+ int code;
if (!(aproc->flags & BPROC_STARTED) || (aproc->flags & BPROC_EXITED))
return BZNOTACTIVE;
}
int
-bnode_Deactivate(register struct bnode *abnode)
+bnode_Deactivate(struct bnode *abnode)
{
- register struct bnode **pb, *tb;
+ struct bnode **pb, *tb;
struct bnode *nb;
if (!(abnode->flags & BNODE_ACTIVE))
return BZNOTACTIVE;
git://git.openafs.org / openafs.git / blobdiff