/*
* 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 <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 <opr/queue.h>
+
#include "bnode.h"
+#include "bnode_internal.h"
+#include "bosprototypes.h"
-#ifdef AFS_AIX_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 */
+#define BNODE_ERROR_DELAY_MAX 60 /* maximum retry delay (seconds) */
-int bnode_waiting = 0;
-static PROCESS bproc_pid; /* pid of waker-upper */
-static struct bnode *allBnodes=0; /* list of all bnodes */
-static struct bnode_proc *allProcs=0; /* list of all processes for which we're waiting */
-static struct bnode_type *allTypes=0; /* list of registered type handlers */
+static PROCESS bproc_pid; /* pid of waker-upper */
+static struct opr_queue allBnodes; /**< List of all bnodes */
+static struct opr_queue allProcs; /**< List of all processes for which we're waiting */
+static struct opr_queue allTypes; /**< List of all registered type handlers */
static struct bnode_stats {
int weirdPids;
} bnode_stats;
-static afs_int32 SendNotifierData();
-static int DeleteProc();
-
+extern const char *DoCore;
+extern const char *DoPidFiles;
#ifndef AFS_NT40_ENV
-extern char **environ; /* env structure */
+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(ap)
-register struct bnode_proc *ap; {
- register struct bnode *tbnodep;
+static void
+RememberProcName(struct bnode_proc *ap)
+{
+ struct bnode *tbnodep;
tbnodep = ap->bnode;
if (tbnodep->lastErrorName) {
free(tbnodep->lastErrorName);
- tbnodep->lastErrorName = (char *) 0;
- }
- if (ap->coreName) {
- tbnodep->lastErrorName = (char *) malloc(strlen(ap->coreName)+1);
- strcpy(tbnodep->lastErrorName, ap->coreName);
+ tbnodep->lastErrorName = NULL;
}
+ if (ap->coreName)
+ tbnodep->lastErrorName = strdup(ap->coreName);
}
/* utility for use by BOP_HASCORE functions to determine where a core file might
* be stored.
*/
-bnode_CoreName(abnode, acoreName, abuffer)
-register struct bnode *abnode;
-char *acoreName;
-char *abuffer; {
- strcpy(abuffer, AFSDIR_SERVER_CORELOG_FILEPATH);
+int
+bnode_CoreName(struct bnode *abnode, char *acoreName, char *abuffer)
+{
+ 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(abnode, aproc)
-register struct bnode_proc *aproc;
-register struct bnode *abnode; {
+static void
+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 timeval Start;
struct tm *TimeFields;
char FileName[256];
#endif
- code = stat(AFSDIR_SERVER_CORELOG_FILEPATH, &tstat);
- if (code) return;
-
+ /* 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_hour, TimeFields->tm_min,
- TimeFields->tm_sec);
- strcpy(tbuffer,FileName);
+ sprintf(FileName, "%s.%d%02d%02d%02d%02d%02d", tbuffer,
+ 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);
+ rk_rename(corefile, tbuffer);
+ free(corefile);
}
-bnode_GetString(abnode, abuffer, alen)
-register struct bnode *abnode;
-register char *abuffer;
-register afs_int32 alen;{
+int
+bnode_GetString(struct bnode *abnode, char *abuffer,
+ afs_int32 alen)
+{
return BOP_GETSTRING(abnode, abuffer, alen);
}
-bnode_GetParm(abnode, aindex, abuffer, alen)
-register struct bnode *abnode;
-register afs_int32 aindex;
-register char *abuffer;
-afs_int32 alen; {
+int
+bnode_GetParm(struct bnode *abnode, afs_int32 aindex,
+ char *abuffer, afs_int32 alen)
+{
return BOP_GETPARM(abnode, aindex, abuffer, alen);
}
-bnode_GetStat(abnode, astatus)
-register struct bnode *abnode;
-register afs_int32 *astatus; {
+int
+bnode_GetStat(struct bnode *abnode, afs_int32 * astatus)
+{
return BOP_GETSTAT(abnode, astatus);
}
-bnode_RestartP(abnode)
-register struct bnode *abnode; {
+int
+bnode_RestartP(struct bnode *abnode)
+{
return BOP_RESTARTP(abnode);
}
-static bnode_Check(abnode)
-register struct bnode *abnode; {
+static int
+bnode_Check(struct bnode *abnode)
+{
if (abnode->flags & BNODE_WAIT) {
abnode->flags &= ~BNODE_WAIT;
LWP_NoYieldSignal(abnode);
}
/* tell if an instance has a core file */
-bnode_HasCore(abnode)
-register struct bnode *abnode; {
+int
+bnode_HasCore(struct bnode *abnode)
+{
return BOP_HASCORE(abnode);
}
/* wait for all bnodes to stabilize */
-bnode_WaitAll() {
- register struct bnode *tb;
- register afs_int32 code;
+int
+bnode_WaitAll(void)
+{
+ struct opr_queue *cursor;
+ afs_int32 code;
afs_int32 stat;
retry:
- for(tb = allBnodes; tb; tb=tb->next) {
+ for (opr_queue_Scan(&allBnodes, cursor)) {
+ struct bnode *tb = opr_queue_Entry(cursor, struct bnode, q);
+
bnode_Hold(tb);
code = BOP_GETSTAT(tb, &stat);
if (code) {
}
/* wait until bnode status is correct */
-bnode_WaitStatus(abnode, astatus)
-int astatus;
-register struct bnode *abnode; {
- register afs_int32 code;
+int
+bnode_WaitStatus(struct bnode *abnode, int astatus)
+{
+ afs_int32 code;
afs_int32 stat;
bnode_Hold(abnode);
while (1) {
/* get the status */
code = BOP_GETSTAT(abnode, &stat);
- if (code) return code;
+ if (code)
+ return code;
/* otherwise, check if we're done */
if (stat == astatus) {
bnode_Release(abnode);
- return 0; /* done */
+ return 0; /* done */
}
if (astatus != abnode->goal) {
bnode_Release(abnode);
- return -1; /* no longer our goal, don't keep waiting */
+ return -1; /* no longer our goal, don't keep waiting */
}
/* otherwise, block */
abnode->flags |= BNODE_WAIT;
}
}
-bnode_SetStat(abnode, agoal)
-register struct bnode *abnode;
-register int agoal; {
+int
+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);
BOP_SETSTAT(abnode, agoal);
return 0;
}
-bnode_SetGoal(abnode, agoal)
-register struct bnode *abnode;
-register int agoal; {
+int
+bnode_SetGoal(struct bnode *abnode, int agoal)
+{
abnode->goal = agoal;
bnode_Check(abnode);
return 0;
}
-bnode_SetFileGoal(abnode, agoal)
-register struct bnode *abnode;
-register int agoal; {
- if (abnode->fileGoal == agoal) return 0; /* already done */
+int
+bnode_SetFileGoal(struct bnode *abnode, int agoal)
+{
+ if (abnode->fileGoal == agoal)
+ return 0; /* already done */
abnode->fileGoal = agoal;
WriteBozoFile(0);
return 0;
}
/* apply a function to all bnodes in the system */
-int bnode_ApplyInstance(aproc, arock)
-int (*aproc)();
-char *arock; {
- register struct bnode *tb, *nb;
- register afs_int32 code;
-
- for(tb = allBnodes; tb; tb=nb) {
- nb = tb->next;
+int
+bnode_ApplyInstance(int (*aproc) (struct bnode *tb, void *), void *arock)
+{
+ struct opr_queue *cursor, *store;
+ afs_int32 code;
+
+ for (opr_queue_ScanSafe(&allBnodes, cursor, store)) {
+ struct bnode *tb = opr_queue_Entry(cursor, struct bnode, q);
code = (*aproc) (tb, arock);
- if (code) return code;
+ if (code)
+ return code;
}
return 0;
}
-struct bnode *bnode_FindInstance (aname)
-register char *aname; {
- register struct bnode *tb;
-
- for(tb=allBnodes;tb;tb=tb->next) {
- if (!strcmp(tb->name, aname)) return tb;
+struct bnode *
+bnode_FindInstance(char *aname)
+{
+ struct opr_queue *cursor;
+
+ for (opr_queue_Scan(&allBnodes, cursor)) {
+ struct bnode *tb = opr_queue_Entry(cursor, struct bnode, q);
+
+ if (!strcmp(tb->name, aname))
+ return tb;
}
- return (struct bnode *) 0;
+ return NULL;
}
-static struct bnode_type *FindType(aname)
-register char *aname; {
- register struct bnode_type *tt;
-
- for(tt=allTypes;tt;tt=tt->next) {
- if (!strcmp(tt->name, aname)) return tt;
+static struct bnode_type *
+FindType(char *aname)
+{
+ struct opr_queue *cursor;
+
+ for (opr_queue_Scan(&allTypes, cursor)) {
+ struct bnode_type *tt = opr_queue_Entry(cursor, struct bnode_type, q);
+
+ if (!strcmp(tt->name, aname))
+ return tt;
}
- return (struct bnode_type *) 0;
+ return NULL;
}
-bnode_Register(atype, aprocs, anparms)
-char *atype;
-int anparms; /* number of parms to create */
-struct bnode_ops *aprocs; {
- register struct bnode_type *tt;
+int
+bnode_Register(char *atype, struct bnode_ops *aprocs, int anparms)
+{
+ struct opr_queue *cursor;
+ struct bnode_type *tt = NULL;
- for(tt=allTypes;tt;tt=tt->next) {
- if (!strcmp(tt->name, atype)) break;
+ for (opr_queue_Scan(&allTypes, cursor), tt = NULL) {
+ tt = opr_queue_Entry(cursor, struct bnode_type, q);
+ if (!strcmp(tt->name, atype))
+ break;
}
if (!tt) {
- tt = (struct bnode_type *) malloc(sizeof(struct bnode_type));
- memset(tt, 0, sizeof(struct bnode_type));
- tt->next = allTypes;
- allTypes = tt;
+ tt = calloc(1, sizeof(struct bnode_type));
+ opr_queue_Init(&tt->q);
+ opr_queue_Prepend(&allTypes, &tt->q);
tt->name = atype;
}
tt->ops = aprocs;
return 0;
}
-afs_int32 bnode_Create(atype, ainstance, abp, ap1, ap2, ap3, ap4, ap5, notifier,fileGoal)
-char *atype;
-char *ainstance;
-struct bnode **abp;
-char *ap1, *ap2, *ap3, *ap4, *ap5, *notifier;
-int fileGoal;{
+afs_int32
+bnode_Create(char *atype, char *ainstance, struct bnode ** abp, char *ap1,
+ char *ap2, char *ap3, char *ap4, char *ap5, char *notifier,
+ int fileGoal, int rewritefile)
+{
struct bnode_type *type;
struct bnode *tb;
char *notifierpath = NULL;
struct stat tstat;
- if (bnode_FindInstance(ainstance)) return BZEXISTS;
+ if (bnode_FindInstance(ainstance))
+ return BZEXISTS;
type = FindType(atype);
- if (!type) return BZBADTYPE;
+ if (!type)
+ return BZBADTYPE;
if (notifier && strcmp(notifier, NONOTIFIER)) {
/* construct local path from canonical (wire-format) path */
if (ConstructLocalBinPath(notifier, ¬ifierpath)) {
- bozo_Log("BNODE-Create: Notifier program path invalid '%s'\n", notifier);
+ bozo_Log("BNODE-Create: Notifier program path invalid '%s'\n",
+ notifier);
return BZNOCREATE;
}
if (stat(notifierpath, &tstat)) {
- bozo_Log("BNODE-Create: Notifier program '%s' not found\n", notifierpath);
+ bozo_Log("BNODE-Create: Notifier program '%s' not found\n",
+ notifierpath);
free(notifierpath);
return BZNOCREATE;
}
}
- tb = (*type->ops->create)(ainstance, ap1, ap2, ap3, ap4, ap5);
+ tb = (*type->ops->create) (ainstance, ap1, ap2, ap3, ap4, ap5);
if (!tb) {
free(notifierpath);
return BZNOCREATE;
*abp = tb;
tb->type = type;
- /* 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... */
+ /* 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;
- bnode_SetStat(tb, tb->goal); /* nudge it once */
- WriteBozoFile(0);
+ bnode_SetStat(tb, tb->goal); /* nudge it once */
+
+ if (rewritefile != 0)
+ WriteBozoFile(0);
+
return 0;
}
-int bnode_DeleteName(ainstance)
-char *ainstance; {
- register struct bnode *tb;
+int
+bnode_DeleteName(char *ainstance)
+{
+ struct bnode *tb;
tb = bnode_FindInstance(ainstance);
- if (!tb) return BZNOENT;
-
+ if (!tb)
+ return BZNOENT;
+
return bnode_Delete(tb);
}
-bnode_Hold(abnode)
-register struct bnode *abnode; {
+int
+bnode_Hold(struct bnode *abnode)
+{
abnode->refCount++;
return 0;
}
-bnode_Release(abnode)
-register struct bnode *abnode; {
+int
+bnode_Release(struct bnode *abnode)
+{
abnode->refCount--;
if (abnode->refCount == 0 && abnode->flags & BNODE_DELETE) {
abnode->flags &= ~BNODE_DELETE; /* we're going for it */
return 0;
}
-int bnode_Delete(abnode)
-register struct bnode *abnode; {
- register afs_int32 code;
- register struct bnode **lb, *ub;
+int
+bnode_Delete(struct bnode *abnode)
+{
+ afs_int32 code;
afs_int32 temp;
-
+
if (abnode->refCount != 0) {
abnode->flags |= BNODE_DELETE;
return 0;
bnode_Hold(abnode);
code = BOP_GETSTAT(abnode, &temp);
bnode_Release(abnode);
- if (code) return code;
- if (temp != BSTAT_SHUTDOWN) return BZBUSY;
+ if (code)
+ return code;
+ if (temp != BSTAT_SHUTDOWN)
+ return BZBUSY;
/* all clear to zap */
- for(lb = &allBnodes, ub = *lb; ub; lb= &ub->next, ub = *lb) {
- if (ub == abnode) {
- /* unthread it from the list */
- *lb = ub->next;
- break;
- }
- }
- free(abnode->name); /* do this first, since bnode fields may be bad after BOP_DELETE */
+ opr_queue_Remove(&abnode->q);
+ free(abnode->name); /* do this first, since bnode fields may be bad after BOP_DELETE */
code = BOP_DELETE(abnode); /* don't play games like holding over this one */
WriteBozoFile(0);
return code;
}
/* function to tell if there's a timeout coming up */
-int bnode_PendingTimeout(abnode)
-register struct bnode *abnode; {
+int
+bnode_PendingTimeout(struct bnode *abnode)
+{
return (abnode->flags & BNODE_NEEDTIMEOUT);
}
/* function called to set / clear periodic bnode wakeup times */
-int bnode_SetTimeout(abnode, atimeout)
-register struct bnode *abnode;
-afs_int32 atimeout; {
+int
+bnode_SetTimeout(struct bnode *abnode, afs_int32 atimeout)
+{
if (atimeout != 0) {
abnode->nextTimeout = FT_ApproxTime() + atimeout;
abnode->flags |= BNODE_NEEDTIMEOUT;
abnode->period = atimeout;
IOMGR_Cancel(bproc_pid);
- }
- else {
+ } else {
abnode->flags &= ~BNODE_NEEDTIMEOUT;
}
return 0;
}
/* used by new bnode creation code to format bnode header */
-int bnode_InitBnode (abnode, abnodeops, aname)
-register struct bnode *abnode;
-char *aname;
-struct bnode_ops *abnodeops; {
- struct bnode **lb, *nb;
-
+int
+bnode_InitBnode(struct bnode *abnode, struct bnode_ops *abnodeops,
+ char *aname)
+{
/* format the bnode properly */
memset(abnode, 0, sizeof(struct bnode));
+ opr_queue_Init(&abnode->q);
abnode->ops = abnodeops;
- abnode->name = (char *) malloc(strlen(aname)+1);
- strcpy(abnode->name, aname);
+ abnode->name = strdup(aname);
+ if (!abnode->name)
+ return ENOMEM;
abnode->flags = BNODE_ACTIVE;
abnode->fileGoal = BSTAT_NORMAL;
abnode->goal = BSTAT_SHUTDOWN;
/* put the bnode at the end of the list so we write bnode file in same order */
- for(lb = &allBnodes, nb = *lb; nb; lb = &nb->next, nb = *lb);
- *lb = abnode;
+ opr_queue_Append(&allBnodes, &abnode->q);
return 0;
}
/* bnode lwp executes this code repeatedly */
-static int bproc() {
- register afs_int32 code;
- register struct bnode *tb;
- register afs_int32 temp;
- register struct bnode_proc *tp;
- struct bnode *nb;
- int options; /* must not be register */
+static void *
+bproc(void *unused)
+{
+ afs_int32 code;
+ struct bnode *tb;
+ afs_int32 temp;
+ struct opr_queue *cursor, *store;
+ struct bnode_proc *tp;
+ int options; /* must not be register */
struct timeval tv;
int setAny;
int status;
while (1) {
/* first figure out how long to sleep for */
- temp = 0x7fffffff; /* afs_int32 time; maxint doesn't work in select */
+ temp = 0x7fffffff; /* afs_int32 time; maxint doesn't work in select */
setAny = 0;
- for(tb = allBnodes; tb; tb=tb->next) {
+ for (opr_queue_Scan(&allBnodes, cursor)) {
+ tb = opr_queue_Entry(cursor, struct bnode, q);
if (tb->flags & BNODE_NEEDTIMEOUT) {
if (tb->nextTimeout < temp) {
setAny = 1;
/* now temp has the time at which we should wakeup next */
/* sleep */
- if (setAny) temp -= FT_ApproxTime(); /* how many seconds until next event */
- else temp = 999999;
+ if (setAny)
+ temp -= FT_ApproxTime(); /* how many seconds until next event */
+ else
+ temp = 999999;
if (temp > 0) {
tv.tv_sec = temp;
tv.tv_usec = 0;
code = IOMGR_Select(0, 0, 0, 0, &tv);
- }
- else code = 0; /* fake timeout code */
+ } else
+ code = 0; /* fake timeout code */
/* figure out why we woke up; child exit or timeouts */
- FT_GetTimeOfDay(&tv, 0); /* must do the real gettimeofday once and a while */
+ FT_GetTimeOfDay(&tv, 0); /* must do the real gettimeofday once and a while */
temp = tv.tv_sec;
/* check all bnodes to see which ones need timeout events */
- for(tb = allBnodes; tb; tb=nb) {
+ for (opr_queue_ScanSafe(&allBnodes, cursor, store)) {
+ tb = opr_queue_Entry(cursor, struct bnode, q);
if ((tb->flags & BNODE_NEEDTIMEOUT) && temp > tb->nextTimeout) {
bnode_Hold(tb);
BOP_TIMEOUT(tb);
bnode_Check(tb);
- if (tb->flags & BNODE_NEEDTIMEOUT) { /* check again, BOP_TIMEOUT could change */
+ if (tb->flags & BNODE_NEEDTIMEOUT) { /* check again, BOP_TIMEOUT could change */
tb->nextTimeout = FT_ApproxTime() + tb->period;
}
- nb = tb->next;
- bnode_Release(tb); /* delete may occur here */
+ bnode_Release(tb); /* delete may occur here */
}
- else nb=tb->next;
}
if (code < 0) {
/* signalled, probably by incoming signal */
while (1) {
options = WNOHANG;
- bnode_waiting = options | 0x800000;
- code = waitpid((pid_t)-1, &status, options);
- bnode_waiting = 0;
- if (code == 0 || code == -1) break; /* all done */
+ code = waitpid((pid_t) - 1, &status, options);
+ if (code == 0 || code == -1)
+ break; /* all done */
/* otherwise code has a process id, which we now search for */
- for(tp=allProcs; tp; tp=tp->next)
- if (tp->pid == code) break;
+ for (tp = NULL, opr_queue_Scan(&allProcs, cursor), tp = NULL) {
+ tp = opr_queue_Entry(cursor, struct bnode_proc, q);
+
+ if (tp->pid == code)
+ break;
+ }
if (tp) {
/* found the pid */
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->errorSignal = 0;
}
if (tp->coreName)
- bozo_Log("%s:%s exited with code %d",
- tb->name, tp->coreName, tp->lastExit);
+ bozo_Log("%s:%s exited with code %d\n", tb->name,
+ tp->coreName, tp->lastExit);
else
- bozo_Log("%s exited with code %d",
- tb->name, tp->lastExit);
- }
- else {
+ bozo_Log("%s exited with code %d\n", tb->name,
+ tp->lastExit);
+ } else {
/* Signal occurred, perhaps spurious due to shutdown request.
* If due to a shutdown request, don't overwrite last error
* information.
*/
tp->lastSignal = WTERMSIG(status);
tp->lastExit = 0;
- if (tp->lastSignal != SIGQUIT && tp->lastSignal != SIGTERM
+ if (tp->lastSignal != SIGQUIT
+ && tp->lastSignal != SIGTERM
&& tp->lastSignal != SIGKILL) {
tb->errorSignal = tp->lastSignal;
tb->lastErrorExit = FT_ApproxTime();
RememberProcName(tp);
}
if (tp->coreName)
- bozo_Log("%s:%s exited on signal %d%s",
- tb->name, tp->coreName, tp->lastSignal,
- WCOREDUMP(status) ? " (core dumped)" : "");
+ bozo_Log("%s:%s exited on signal %d%s\n",
+ tb->name, tp->coreName, tp->lastSignal,
+ WCOREDUMP(status) ? " (core dumped)" :
+ "");
else
- bozo_Log("%s exited on signal %d%s",
- tb->name, tp->lastSignal,
- WCOREDUMP(status) ? " (core dumped)" : "");
+ bozo_Log("%s exited on signal %d%s\n", tb->name,
+ tp->lastSignal,
+ WCOREDUMP(status) ? " (core dumped)" :
+ "");
SaveCore(tb, tp);
}
tb->lastAnyExit = FT_ApproxTime();
if (tb->notifier) {
- bozo_Log("BNODE: Notifier %s will be called\n", tb->notifier);
+ bozo_Log("BNODE: Notifier %s will be called\n",
+ 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; /* wait a second, then retry */
+ } else {
+ tb->errorStopDelay *= 2; /* ramp up the retry delays */
+ }
+ if (tb->errorStopDelay > BNODE_ERROR_DELAY_MAX) {
+ tb->errorStopDelay = BNODE_ERROR_DELAY_MAX; /* cap the delay */
+ }
+ }
tb->flags |= BNODE_ERRORSTOP;
bnode_SetGoal(tb, BSTAT_SHUTDOWN);
- bozo_Log("BNODE '%s' repeatedly failed to start, perhaps missing executable.\n",
- tb->name);
+ 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 bnode_stats.weirdPids++;
+ opr_queue_Remove(&tp->q);
+ free(tp);
+ } else
+ bnode_stats.weirdPids++;
}
}
}
+ return NULL;
}
-
-#ifdef notdef
-sigcatch()
-{
- signal(SIGPIPE, SIG_IGN);
- bozo_Log("Notifier aborted prematurely");
- exit(0);
-}
-#endif
-
-
-
-hdl_notifier(tp)
- struct bnode_proc *tp;
-{
-#ifndef AFS_NT40_ENV /* NT notifier callout not yet implemented */
- int code, pid, status;
- struct stat tstat;
-
- if (stat(tp->bnode->notifier, &tstat)) {
- bozo_Log("BNODE: Failed to find notifier '%s'; ignored\n", tp->bnode->notifier);
- return(1);
- }
- 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();
-#else
-#ifdef AFS_LINUX20_ENV
- ec = setpgrp();
-#else
- ec = setpgrp(0,0);
-#endif
-#endif
- fout = popen(tb->notifier, "w");
- if (fout == NULL) {
- bozo_Log("BNODE: Failed to find notifier '%s'; ignored\n", tb->notifier);
- perror(tb->notifier);
- exit(1);
- }
- code = SendNotifierData(fileno(fout), tp);
- pclose(fout);
- exit(0);
- } else if (pid < 0) {
- bozo_Log("Failed to fork creating process to handle notifier '%s'\n", tp->bnode->notifier);
- return -1;
- }
-#endif /* AFS_NT40_ENV */
- return(0);
-}
-
-
-static afs_int32 SendNotifierData(fd, tp)
- register int fd;
- register struct bnode_proc *tp;
+static afs_int32
+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
*/
- (void) sprintf(bufp, "BEGIN bnode_proc\n");
+ (void)sprintf(bufp, "BEGIN bnode_proc\n");
bufp += strlen(bufp);
- (void) sprintf(bufp, "comLine: %s\n", tp->comLine);
+ (void)sprintf(bufp, "comLine: %s\n", tp->comLine);
bufp += strlen(bufp);
if (!(buf1 = tp->coreName))
buf1 = "(null)";
- (void) sprintf(bufp, "coreName: %s\n", buf1);
+ (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");
+ (void)sprintf(bufp, "END bnode_proc\n");
bufp += strlen(bufp);
- len =(int)(bufp-buffer);
+ len = (int)(bufp - buffer);
if (write(fd, buffer, len) < 0) {
return -1;
}
/*
* Now sent out the bnode struct
- */
+ */
bufp = buffer;
- (void) sprintf(bufp, "BEGIN bnode\n");
+ (void)sprintf(bufp, "BEGIN bnode\n");
bufp += strlen(bufp);
- (void) sprintf(bufp, "name: %s\n", tb->name);
+ (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);
bufp += strlen(bufp);
*/
- (void) sprintf(bufp, "goal: %d\n", tb->goal);
+ (void)sprintf(bufp, "goal: %d\n", tb->goal);
bufp += strlen(bufp);
- (void) sprintf(bufp, "END bnode\n");
+ (void)sprintf(bufp, "END bnode\n");
bufp += strlen(bufp);
- len = (int)bufp-(int)buffer;
+ len = (int)(bufp - buffer);
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 pid;
+ struct stat tstat;
+
+ if (stat(tp->bnode->notifier, &tstat)) {
+ bozo_Log("BNODE: Failed to find notifier '%s'; ignored\n",
+ tp->bnode->notifier);
+ return (1);
+ }
+ if ((pid = fork()) == 0) {
+ FILE *fout;
+ struct bnode *tb = tp->bnode;
+
+#if defined(AFS_HPUX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_SGI51_ENV)
+ setsid();
+#elif defined(AFS_DARWIN90_ENV)
+ setpgid(0, 0);
+#elif defined(AFS_LINUX20_ENV) || defined(AFS_AIX_ENV)
+ setpgrp();
+#else
+ setpgrp(0, 0);
+#endif
+ fout = popen(tb->notifier, "w");
+ if (fout == NULL) {
+ bozo_Log("BNODE: Failed to find notifier '%s'; ignored\n",
+ tb->notifier);
+ perror(tb->notifier);
+ exit(1);
+ }
+ SendNotifierData(fileno(fout), tp);
+ pclose(fout);
+ exit(0);
+ } else if (pid < 0) {
+ bozo_Log("Failed to fork creating process to handle notifier '%s'\n",
+ tp->bnode->notifier);
+ return -1;
+ }
+#endif /* AFS_NT40_ENV */
+ return (0);
+}
-
/* Called by IOMGR at low priority on IOMGR's stack shortly after a SIGCHLD
* occurs. Wakes up bproc do redo things */
-bnode_SoftInt(int asignal)
+void *
+bnode_SoftInt(void *param)
{
+ /* int asignal = (int) param; */
+
IOMGR_Cancel(bproc_pid);
- return 0;
+ return NULL;
}
/* Called at signal interrupt level; queues function to be called
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() {
+int
+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;
+ if (initDone)
+ return 0;
initDone = 1;
+ opr_queue_Init(&allTypes);
+ opr_queue_Init(&allProcs);
+ opr_queue_Init(&allBnodes);
memset(&bnode_stats, 0, sizeof(bnode_stats));
- LWP_InitializeProcessSupport(1, &junk); /* just in case */
+ LWP_InitializeProcessSupport(1, &junk); /* just in case */
IOMGR_Initialize();
code = LWP_CreateProcess(bproc, BNODE_LWP_STACKSIZE,
- /* priority */ 1, /* parm */0, "bnode-manager", &bproc_pid);
- if (code) return code;
- memset((char *)&newaction, 0, sizeof(newaction));
+ /* priority */ 1, (void *) /* parm */ 0,
+ "bnode-manager", &bproc_pid);
+ if (code)
+ return code;
+ memset(&newaction, 0, sizeof(newaction));
newaction.sa_handler = bnode_Int;
code = sigaction(SIGCHLD, &newaction, NULL);
- if (code) return errno;
+ if (code)
+ return errno;
code = sigaction(SIGQUIT, &newaction, NULL);
- if (code) return errno;
+ if (code)
+ return errno;
+ code = sigaction(SIGTERM, &newaction, NULL);
+ if (code)
+ return errno;
return code;
}
-
+
/* free token list returned by parseLine */
-bnode_FreeTokens(alist)
- register struct bnode_token *alist; {
- register struct bnode_token *nlist;
- for(; alist; alist = nlist) {
+int
+bnode_FreeTokens(struct bnode_token *alist)
+{
+ struct bnode_token *nlist;
+ for (; alist; alist = nlist) {
nlist = alist->next;
free(alist->key);
free(alist);
return 0;
}
-static space(x)
-int x; {
- if (x == 0 || x == ' ' || x == '\t' || x== '\n') return 1;
- else return 0;
+static int
+space(int x)
+{
+ if (x == 0 || x == ' ' || x == '\t' || x == '\n')
+ return 1;
+ else
+ return 0;
}
-bnode_ParseLine(aline, alist)
- char *aline;
- struct bnode_token **alist; {
+int
+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;
-
- inToken = 0; /* not copying token chars at start */
- first = (struct bnode_token *) 0;
- last = (struct bnode_token *) 0;
+ struct bnode_token *ttok;
+ int tc;
+
+ inToken = 0; /* not copying token chars at start */
+ first = (struct bnode_token *)0;
+ last = (struct bnode_token *)0;
while (1) {
tc = *aline++;
- if (tc == 0 || space(tc)) { /* terminating null gets us in here, too */
+ if (tc == 0 || space(tc)) { /* terminating null gets us in here, too */
if (inToken) {
- inToken = 0; /* end of this token */
+ inToken = 0; /* end of this token */
*tptr++ = 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 = malloc(sizeof(struct bnode_token));
+ ttok->next = (struct bnode_token *)0;
+ ttok->key = strdup(tbuffer);
if (last) {
last->next = ttok;
last = ttok;
- }
- else last = ttok;
- if (!first) first = ttok;
+ } else
+ last = ttok;
+ if (!first)
+ first = ttok;
}
- }
- else {
+ } else {
/* an alpha character */
if (!inToken) {
tptr = tbuffer;
inToken = 1;
}
- if (tptr - tbuffer >= sizeof(tbuffer)) return -1; /* token too long */
+ if (tptr - tbuffer >= sizeof(tbuffer))
+ return -1; /* token too long */
*tptr++ = tc;
}
if (tc == 0) {
/* last token flushed 'cause space(0) --> true */
- if (last) last->next = (struct bnode_token *) 0;
+ if (last)
+ last->next = (struct bnode_token *)0;
*alist = first;
return 0;
}
}
#define MAXVARGS 128
-int bnode_NewProc(abnode, aexecString, coreName, aproc)
-struct bnode_proc **aproc;
-char *coreName;
-struct bnode *abnode;
-char *aexecString; {
+int
+bnode_NewProc(struct bnode *abnode, char *aexecString, char *coreName,
+ struct bnode_proc **aproc)
+{
struct bnode_token *tlist, *tt;
afs_int32 code;
struct bnode_proc *tp;
char *argv[MAXVARGS];
int i;
- code = bnode_ParseLine(aexecString, &tlist); /* try parsing first */
- if (code) return code;
- tp = (struct bnode_proc *) malloc(sizeof(struct bnode_proc));
- memset(tp, 0, sizeof(struct bnode_proc));
- tp->next = allProcs;
- allProcs = tp;
- *aproc = tp;
+ code = bnode_ParseLine(aexecString, &tlist); /* try parsing first */
+ if (code)
+ return code;
+ tp = calloc(1, sizeof(struct bnode_proc));
+ opr_queue_Init(&tp->q);
tp->bnode = abnode;
tp->comLine = aexecString;
tp->coreName = coreName; /* may be null */
for (tt = tlist, i = 0; i < (MAXVARGS - 1) && tt; tt = tt->next, i++) {
argv[i] = tt->key;
}
- argv[i] = (char *) 0; /* null-terminated */
+ argv[i] = NULL; /* null-terminated */
cpid = spawnprocve(argv[0], argv, environ, -1);
- osi_audit(BOSSpawnProcEvent, 0, AUD_STR, aexecString, AUD_END );
+ osi_audit(BOSSpawnProcEvent, 0, AUD_STR, aexecString, AUD_END);
- if (cpid == (pid_t)-1) {
+ if (cpid == (pid_t) - 1) {
bozo_Log("Failed to spawn process for bnode '%s'\n", abnode->name);
bnode_FreeTokens(tlist);
free(tp);
return errno;
}
+ bozo_Log("%s started pid %ld: %s\n", abnode->name, cpid, aexecString);
bnode_FreeTokens(tlist);
+ opr_queue_Prepend(&allProcs, &tp->q);
+ *aproc = tp;
tp->pid = cpid;
tp->flags = BPROC_STARTED;
tp->flags &= ~BPROC_EXITED;
+ BOP_PROCSTARTED(abnode, tp);
bnode_Check(abnode);
return 0;
}
-int bnode_StopProc(aproc, asignal)
-register struct bnode_proc *aproc;
-int asignal; {
- register int code;
+int
+bnode_StopProc(struct bnode_proc *aproc, int asignal)
+{
+ int code;
if (!(aproc->flags & BPROC_STARTED) || (aproc->flags & BPROC_EXITED))
return BZNOTACTIVE;
- osi_audit( BOSStopProcEvent, 0, AUD_STR, (aproc ? aproc->comLine : (char *)0), AUD_END );
+ osi_audit(BOSStopProcEvent, 0, AUD_STR, (aproc ? aproc->comLine : NULL),
+ AUD_END);
code = kill(aproc->pid, asignal);
bnode_Check(aproc->bnode);
return code;
}
-
-int bnode_Deactivate(abnode)
-register struct bnode *abnode; {
- register struct bnode **pb, *tb;
- struct bnode *nb;
- if (!(abnode->flags & BNODE_ACTIVE)) return BZNOTACTIVE;
- for(pb = &allBnodes,tb = *pb; tb; tb=nb) {
- nb = tb->next;
- if (tb == abnode) {
- *pb = nb;
- tb->flags &= ~BNODE_ACTIVE;
- return 0;
- }
- }
- return BZNOENT;
-}
-
-static int DeleteProc(abproc)
-register struct bnode_proc *abproc; {
- register struct bnode_proc **pb, *tb;
- struct bnode_proc *nb;
-
- for(pb = &allProcs,tb = *pb; tb; pb = &tb->next, tb=nb) {
- nb = tb->next;
- if (tb == abproc) {
- *pb = nb;
- free(tb);
- return 0;
- }
- }
- return BZNOENT;
-}
git://git.openafs.org / openafs.git / blobdiff