afs_int32
FindByID(struct ubik_trans *at, afs_int32 aid)
{
/* returns address of entry if found, 0 otherwise */
afs_int32 code;
afs_int32 i;
struct prentry tentry;
afs_int32 entry;
if ((aid == PRBADID) || (aid == 0))
return 0;
i = IDHash(aid);
entry = ntohl(cheader.idHash[i]);
if (entry == 0)
return entry;
memset(&tentry, 0, sizeof(tentry));
code = pr_ReadEntry(at, 0, entry, &tentry);
if (code != 0)
return 0;
if (aid == tentry.id)
return entry;
opr_Assert(entry != tentry.nextID);
entry = tentry.nextID;
while (entry != 0) {
memset(&tentry, 0, sizeof(tentry));
code = pr_ReadEntry(at, 0, entry, &tentry);
if (code != 0)
return 0;
if (aid == tentry.id)
return entry;
opr_Assert(entry != tentry.nextID);
entry = tentry.nextID;
}
return 0;
}
afs_int32
FindByName(struct ubik_trans *at, char aname[PR_MAXNAMELEN], struct prentry *tentryp)
{
/* ditto */
afs_int32 code;
afs_int32 i;
afs_int32 entry;
i = NameHash(aname);
entry = ntohl(cheader.nameHash[i]);
if (entry == 0)
return entry;
memset(tentryp, 0, sizeof(struct prentry));
code = pr_ReadEntry(at, 0, entry, tentryp);
if (code != 0)
return 0;
if ((strncmp(aname, tentryp->name, PR_MAXNAMELEN)) == 0)
return entry;
opr_Assert(entry != tentryp->nextName);
entry = tentryp->nextName;
while (entry != 0) {
memset(tentryp, 0, sizeof(struct prentry));
code = pr_ReadEntry(at, 0, entry, tentryp);
if (code != 0)
return 0;
if ((strncmp(aname, tentryp->name, PR_MAXNAMELEN)) == 0)
return entry;
opr_Assert(entry != tentryp->nextName);
entry = tentryp->nextName;
}
return 0;
}
/**
* lock a file on disk for the process.
*
* @param[in] lf the struct VLockFile representing the file to lock
* @param[in] offset the offset in the file to lock
* @param[in] locktype READ_LOCK or WRITE_LOCK
* @param[in] nonblock 0 to wait for conflicting locks to clear before
* obtaining the lock; 1 to fail immediately if a
* conflicting lock is held by someone else
*
* @return operation status
* @retval 0 success
* @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
* specified
* @retval EIO error acquiring file lock
*
* @note DAFS only
*
* @note do not try to lock/unlock the same offset in the same file from
* different threads; use VGetDiskLock to protect threads from each other in
* addition to other processes
*/
int
VLockFileLock(struct VLockFile *lf, afs_uint32 offset, int locktype, int nonblock)
{
int code;
opr_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
opr_mutex_enter(&lf->mutex);
if (lf->fd == INVALID_FD) {
lf->fd = _VOpenPath(lf->path);
if (lf->fd == INVALID_FD) {
opr_mutex_exit(&lf->mutex);
return EIO;
}
}
lf->refcount++;
opr_mutex_exit(&lf->mutex);
code = _VLockFd(lf->fd, offset, locktype, nonblock);
if (code) {
opr_mutex_enter(&lf->mutex);
if (--lf->refcount < 1) {
_VCloseFd(lf->fd);
lf->fd = INVALID_FD;
}
opr_mutex_exit(&lf->mutex);
}
return code;
}
/*!
* \brief send a Probe to all the network address of this server
*
* \return 0 if success, else return 1
*/
int
DoProbe(struct ubik_server *server)
{
struct rx_connection *conns[UBIK_MAX_INTERFACE_ADDR];
struct rx_connection *connSuccess = 0;
int i, j, success_i = -1;
afs_uint32 addr;
char buffer[32];
char hoststr[16];
UBIK_ADDR_LOCK;
for (i = 0; (addr = server->addr[i]) && (i < UBIK_MAX_INTERFACE_ADDR);
i++) {
conns[i] =
rx_NewConnection(addr, ubik_callPortal, DISK_SERVICE_ID,
addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
/* user requirement to use only the primary interface */
if (ubikPrimaryAddrOnly) {
i = 1;
break;
}
}
UBIK_ADDR_UNLOCK;
opr_Assert(i); /* at least one interface address for this server */
multi_Rx(conns, i) {
multi_DISK_Probe();
if (!multi_error) { /* first success */
success_i = multi_i;
multi_Abort;
}
} multi_End_Ignore;
/* strcompose - concatenate strings passed to it.
* Input:
* buf: storage for the composed string. Any data in it will be lost.
* len: length of the buffer.
* ...: variable number of string arguments. The last argument must be
* (char *)NULL.
* Returns buf or NULL if the buffer was not sufficiently large.
*/
char *
strcompose(char *buf, size_t len, ...)
{
va_list ap;
size_t spaceleft = len - 1;
char *str;
size_t slen;
opr_Assert(buf != NULL);
if (len <= 0)
return NULL;
*buf = '\0';
va_start(ap, len);
str = va_arg(ap, char *);
while (str) {
slen = strlen(str);
if (spaceleft < slen) /* not enough space left */
return NULL;
strcat(buf, str);
spaceleft -= slen;
str = va_arg(ap, char *);
}
va_end(ap);
return buf;
}
void
LogCommandLine(int argc, char **argv, const char *progname,
const char *version, const char *logstring,
void (*log) (const char *format, ...))
{
int i, l;
char *commandLine, *cx;
opr_Assert(argc != 0);
for (l = i = 0; i < argc; i++)
l += strlen(argv[i]) + 1;
if ((commandLine = malloc(l))) {
for (cx = commandLine, i = 0; i < argc; i++) {
strcpy(cx, argv[i]);
cx += strlen(cx);
*(cx++) = ' ';
}
commandLine[l-1] = '\0';
(*log)("%s %s %s%s(%s)\n", logstring, progname,
version, strlen(version)>0?" ":"", commandLine);
free(commandLine);
} else {
/* What, we're out of memory already!? */
(*log)("%s %s%s%s\n", logstring,
progname, strlen(version)>0?" ":"", version);
}
}
/*!
* Open the log file descriptor or a connection to the system log.
*
* This function should be called once during program initialization and
* must be called before calling FSLog() or WriteLogBuffer(). The
* fields of the given argument specify the logging destination and
* various optional features.
*
* The lopt_logLevel value specifies the initial logging level.
*
* The lopt_dest enum specifies the logging destination; either
* file based (logDest_file) or the system log (logDest_syslog).
*
* File Based Logging
* ------------------
*
* A file will be opened for log messages when the lopt_dest enum is set
* to logDest_file. The file specified by lopt_filename will be opened
* for appending log messages. A new file will be created if the log
* file does not exist.
*
* The lopt_rotateOnOpen flag specifies whether an existing log file is
* to be renamed and a new log file created during the call to OpenLog.
* The lopt_rotateOnOpen flag has no effect if the file given by
* lopt_filename is a named pipe (fifo).
*
* The lopt_rotateOnReset flag specifies whether the log file is renamed
* and then reopened when the reset signal (SIGHUP) is caught.
*
* The lopt_rotateStyle enum specifies how the new log file is renamed when
* lopt_rotateOnOpen or lopt_rotateOnReset are set. The lopt_rotateStyle
* may be the traditional Transarc style (logRotate_old) or the MR-AFS
* style (logRotate_timestamp).
*
* When lopt_rotateStyle is set to logRotate_old, the suffix ".old" is
* appended to the log file name. The existing ".old" log file is
* removed.
*
* When lopt_rotateStyle is set to logRotate_timestamp, a timestamp
* string is appended to the log file name and existing files are not
* removed.
*
* \note Messages written to stdout and stderr are redirected to the log
* file when file-based logging is in effect.
*
* System Logging
* --------------
*
* A connection to the system log (syslog) will be established for log
* messages when the lopt_dest enum is set to logDest_syslog.
*
* The lopt_facility specifies the system log facility to be used when
* writing messages to the system log.
*
* The lopt_tag string specifies the indentification string to be used
* when writing messages to the system log.
*
* \param opts logging options. A copy of the logging
* options will be made before returning to
* the caller.
*
* \returns 0 on success
*/
int
OpenLog(struct logOptions *opts)
{
int code;
#if defined(AFS_PTHREAD_ENV)
opr_Verify(pthread_once(&serverLogOnce, InitServerLogMutex) == 0);
#endif /* AFS_PTHREAD_ENV */
LogLevel = serverLogOpts.logLevel = opts->logLevel;
serverLogOpts.dest = opts->dest;
switch (serverLogOpts.dest) {
case logDest_file:
serverLogOpts.lopt_rotateOnOpen = opts->lopt_rotateOnOpen;
serverLogOpts.lopt_rotateOnReset = opts->lopt_rotateOnReset;
serverLogOpts.lopt_rotateStyle = opts->lopt_rotateStyle;
/* OpenLogFile() sets ourName; don't cache filename here. */
code = OpenLogFile(opts->lopt_filename);
break;
#ifdef HAVE_SYSLOG
case logDest_syslog:
serverLogOpts.lopt_rotateOnOpen = 0;
serverLogOpts.lopt_rotateOnReset = 0;
serverLogOpts.lopt_rotateStyle = logRotate_none;
openlog(opts->lopt_tag, LOG_PID, opts->lopt_facility);
code = 0;
break;
#endif
default:
opr_Assert(0);
}
return code;
} /*OpenLog */
/**
* initialize a struct VDiskLock.
*
* @param[in] dl struct VDiskLock to initialize
* @param[in] lf the struct VLockFile to associate with this disk lock
*/
void
VDiskLockInit(struct VDiskLock *dl, struct VLockFile *lf, afs_uint32 offset)
{
opr_Assert(lf);
memset(dl, 0, sizeof(*dl));
Lock_Init(&dl->rwlock);
opr_mutex_init(&dl->mutex);
opr_cv_init(&dl->cv);
dl->lockfile = lf;
dl->offset = offset;
}
/**
* initialize a struct VDiskLock.
*
* @param[in] dl struct VDiskLock to initialize
* @param[in] lf the struct VLockFile to associate with this disk lock
*/
void
VDiskLockInit(struct VDiskLock *dl, struct VLockFile *lf, afs_uint32 offset)
{
opr_Assert(lf);
memset(dl, 0, sizeof(*dl));
Lock_Init(&dl->rwlock);
MUTEX_INIT(&dl->mutex, "disklock", MUTEX_DEFAULT, 0);
CV_INIT(&dl->cv, "disklock cv", CV_DEFAULT, 0);
dl->lockfile = lf;
dl->offset = offset;
}
void
softsig_init(void)
{
int rc;
AFS_SIGSET_DECL;
AFS_SIGSET_CLEAR();
rc = pthread_create(&softsig_tid, NULL, &softsig_thread, NULL);
opr_Assert(0 == rc);
AFS_SIGSET_RESTORE();
signal (SIGUSR1, softsig_usr1);
}
/*!
* Open the log file.
*
* Open the log file using the options given in OpenLog().
*
* \returns 0 on success
*/
static int
OpenLogFile(const char *fileName)
{
/*
* This function should allow various libraries that inconsistently
* use stdout/stderr to all go to the same place
*/
int tempfd;
int flags = O_WRONLY | O_CREAT | O_APPEND;
opr_Assert(serverLogOpts.dest == logDest_file);
opr_Assert(fileName != NULL);
if (IsFIFO(fileName)) {
/* Support named pipes as logs by not rotating them. */
flags |= O_NONBLOCK;
} else if (serverLogOpts.lopt_rotateOnOpen) {
/* Old style logging always started a new log file. */
flags |= O_TRUNC;
RenameLogFile(fileName);
}
tempfd = open(fileName, flags, 0666);
if (tempfd < 0) {
printf("Unable to open log file %s\n", fileName);
return -1;
}
RedirectStdStreams(fileName);
/* Save our name for reopening. */
free(ourName);
ourName = strdup(fileName);
opr_Assert(ourName != NULL);
serverLogFD = tempfd;
return 0;
}
afs_int32
RemoveFromIDHash(struct ubik_trans *tt, afs_int32 aid, afs_int32 *loc) /* ??? in case ID hashed twice ??? */
{
/* remove entry designated by aid from id hash table */
afs_int32 code;
afs_int32 current, trail, i;
struct prentry tentry;
struct prentry bentry;
if ((aid == PRBADID) || (aid == 0))
return PRINCONSISTENT;
i = IDHash(aid);
current = ntohl(cheader.idHash[i]);
memset(&tentry, 0, sizeof(tentry));
memset(&bentry, 0, sizeof(bentry));
trail = 0;
if (current == 0)
return PRSUCCESS; /* already gone */
code = pr_ReadEntry(tt, 0, current, &tentry);
if (code)
return PRDBFAIL;
while (aid != tentry.id) {
opr_Assert(trail != current);
trail = current;
current = tentry.nextID;
if (current == 0)
break;
code = pr_ReadEntry(tt, 0, current, &tentry);
if (code)
return PRDBFAIL;
}
if (current == 0)
return PRSUCCESS; /* we didn't find him, so he's already gone */
if (trail == 0) {
/* it's the first entry! */
cheader.idHash[i] = htonl(tentry.nextID);
code =
pr_Write(tt, 0, 72 + HASHSIZE * 4 + i * 4,
(char *)&cheader.idHash[i], sizeof(cheader.idHash[i]));
if (code)
return PRDBFAIL;
} else {
code = pr_ReadEntry(tt, 0, trail, &bentry);
if (code)
return PRDBFAIL;
bentry.nextID = tentry.nextID;
code = pr_WriteEntry(tt, 0, trail, &bentry);
}
*loc = current;
return PRSUCCESS;
}
static void *
SalvageChildReaperThread(void * args)
{
int slot, pid, status;
struct log_cleanup_node * cleanup;
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) {
CV_WAIT(&worker_cv, &worker_lock);
}
MUTEX_EXIT(&worker_lock);
cleanup = malloc(sizeof(struct log_cleanup_node));
while (Reap_Child("salvageserver", &pid, &status) < 0) {
/* try to prevent livelock if something goes wrong */
sleep(1);
}
VOL_LOCK;
for (slot = 0; slot < Parallel; slot++) {
if (child_slot[slot] == pid)
break;
}
opr_Assert(slot < Parallel);
child_slot[slot] = 0;
VOL_UNLOCK;
SALVSYNC_doneWorkByPid(pid, status);
MUTEX_ENTER(&worker_lock);
if (cleanup) {
cleanup->pid = pid;
queue_Append(&log_cleanup_queue, cleanup);
CV_SIGNAL(&log_cleanup_queue.queue_change_cv);
}
/* ok, we've reaped a child */
current_workers--;
CV_BROADCAST(&worker_cv);
}
return NULL;
}
/**
* release a lock on a file on local disk.
*
* @param[in] dl the struct VDiskLock to release
* @param[in] locktype READ_LOCK if you are unlocking a read lock, or
* WRITE_LOCK if you are unlocking a write lock
*
* @return operation status
* @retval 0 success
*/
void
VReleaseDiskLock(struct VDiskLock *dl, int locktype)
{
opr_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
opr_mutex_enter(&dl->mutex);
opr_Assert(dl->lockers > 0);
if (--dl->lockers < 1) {
/* no threads are holding this lock anymore, so we can release the
* actual disk lock */
VLockFileUnlock(dl->lockfile, dl->offset);
dl->flags &= ~VDISKLOCK_ACQUIRED;
}
opr_mutex_exit(&dl->mutex);
if (locktype == READ_LOCK) {
ReleaseReadLock(&dl->rwlock);
} else {
ReleaseWriteLock(&dl->rwlock);
}
}
struct DiskPartition64 *
VGetPartitionById_r(afs_int32 id, int abortp)
{
struct DiskPartition64 *dp = NULL;
if ((id >= 0) && (id <= VOLMAXPARTS)) {
dp = DiskPartitionTable[id];
}
if (abortp) {
opr_Assert(dp != NULL);
}
return dp;
}
afs_int32
RemoveFromNameHash(struct ubik_trans *tt, char *aname, afs_int32 *loc)
{
/* remove from name hash */
afs_int32 code;
afs_int32 current, trail, i;
struct prentry tentry;
struct prentry bentry;
i = NameHash(aname);
current = ntohl(cheader.nameHash[i]);
memset(&tentry, 0, sizeof(tentry));
memset(&bentry, 0, sizeof(bentry));
trail = 0;
if (current == 0)
return PRSUCCESS; /* already gone */
code = pr_ReadEntry(tt, 0, current, &tentry);
if (code)
return PRDBFAIL;
while (strcmp(aname, tentry.name)) {
opr_Assert(trail != current);
trail = current;
current = tentry.nextName;
if (current == 0)
break;
code = pr_ReadEntry(tt, 0, current, &tentry);
if (code)
return PRDBFAIL;
}
if (current == 0)
return PRSUCCESS; /* we didn't find him, already gone */
if (trail == 0) {
/* it's the first entry! */
cheader.nameHash[i] = htonl(tentry.nextName);
code =
pr_Write(tt, 0, 72 + i * 4, (char *)&cheader.nameHash[i],
sizeof(cheader.nameHash[i]));
if (code)
return PRDBFAIL;
} else {
code = pr_ReadEntry(tt, 0, trail, &bentry);
if (code)
return PRDBFAIL;
bentry.nextName = tentry.nextName;
code = pr_WriteEntry(tt, 0, trail, &bentry);
}
*loc = current;
return PRSUCCESS;
}
void
VLockPartition_r(char *name)
{
struct DiskPartition64 *dp = VGetPartition_r(name, 0);
OVERLAPPED lap;
if (!dp)
return;
if (dp->lock_fd == INVALID_FD) {
char path[64];
int rc;
(void)sprintf(path, "%s\\%s", VPartitionPath(dp), LOCKFILE);
dp->lock_fd =
(FD_t)CreateFile(path, GENERIC_WRITE,
FILE_SHARE_READ | FILE_SHARE_WRITE, NULL,
CREATE_ALWAYS, FILE_ATTRIBUTE_HIDDEN, NULL);
opr_Assert(dp->lock_fd != INVALID_FD);
memset(&lap, 0, sizeof(lap));
rc = LockFileEx((HANDLE) dp->lock_fd, LOCKFILE_EXCLUSIVE_LOCK, 0, 1,
0, &lap);
opr_Assert(rc);
}
}
void
VLockFileUnlock(struct VLockFile *lf, afs_uint32 offset)
{
opr_mutex_enter(&lf->mutex);
opr_Assert(lf->fd != INVALID_FD);
if (--lf->refcount < 1) {
_VCloseFd(lf->fd);
lf->fd = INVALID_FD;
} else {
_VUnlockFd(lf->fd, offset);
}
opr_mutex_exit(&lf->mutex);
}
/* get partition structure, abortp tells us if we should abort on failure */
struct DiskPartition64 *
VGetPartition_r(char *name, int abortp)
{
struct DiskPartition64 *dp;
#ifdef AFS_DEMAND_ATTACH_FS
dp = VLookupPartition_r(name);
#else /* AFS_DEMAND_ATTACH_FS */
for (dp = DiskPartitionList; dp; dp = dp->next) {
if (strcmp(dp->name, name) == 0)
break;
}
#endif /* AFS_DEMAND_ATTACH_FS */
if (abortp)
opr_Assert(dp != NULL);
return dp;
}
void
VLockFileUnlock(struct VLockFile *lf, afs_uint32 offset)
{
MUTEX_ENTER(&lf->mutex);
opr_Assert(lf->fd != INVALID_FD);
if (--lf->refcount < 1) {
_VCloseFd(lf->fd);
lf->fd = INVALID_FD;
} else {
_VUnlockFd(lf->fd, offset);
}
MUTEX_EXIT(&lf->mutex);
}
/*
* Given two arrays of addresses, masks and mtus find the common ones
* and return them in the first buffer. Return number of common
* entries.
*/
static int
filterAddrs(afs_uint32 addr1[], afs_uint32 addr2[], afs_uint32 mask1[],
afs_uint32 mask2[], afs_uint32 mtu1[], afs_uint32 mtu2[], int n1,
int n2)
{
afs_uint32 taddr[MAXIPADDRS];
afs_uint32 tmask[MAXIPADDRS];
afs_uint32 tmtu[MAXIPADDRS];
int count = 0, i = 0, j = 0, found = 0;
opr_Assert(addr1);
opr_Assert(addr2);
opr_Assert(mask1);
opr_Assert(mask2);
opr_Assert(mtu1);
opr_Assert(mtu2);
for (i = 0; i < n1; i++) {
found = 0;
for (j = 0; j < n2; j++) {
if (addr1[i] == addr2[j]) {
found = 1;
break;
}
}
/* Always mask loopback address */
if (found && rx_IsLoopbackAddr(addr1[i]))
found = 0;
if (found) {
taddr[count] = addr1[i];
tmask[count] = mask1[i];
tmtu[count] = mtu1[i];
count++;
}
}
/* copy everything into addr1, mask1 and mtu1 */
for (i = 0; i < count; i++) {
addr1[i] = taddr[i];
if (mask1) {
mask1[i] = tmask[i];
mtu1[i] = tmtu[i];
}
}
/* and zero out the rest */
for (i = count; i < n1; i++) {
addr1[i] = 0;
if (mask1) {
mask1[i] = 0;
mtu1[i] = 0;
}
}
return count;
}
/*!
* Move the current log file out of the way so a new one can be started.
*
* The format of the new name depends on the logging style. The traditional
* Transarc style appends ".old" to the log file name. When MR-AFS style
* logging is in effect, a time stamp is appended to the log file name instead
* of ".old".
*
* \bug Unfortunately, no check is made to avoid overwriting
* old logs in the traditional Transarc mode.
*
* \param fileName fully qualified log file path
*/
static void
RenameLogFile(const char *fileName)
{
int code;
char *nextName = NULL;
int tries;
time_t t;
struct stat buf;
struct tm *timeFields;
switch (serverLogOpts.lopt_rotateStyle) {
case logRotate_none:
break;
case logRotate_old:
code = asprintf(&nextName, "%s.old", fileName);
if (code < 0) {
nextName = NULL;
}
break;
case logRotate_timestamp:
time(&t);
for (tries = 0; nextName == NULL && tries < 100; t++, tries++) {
timeFields = localtime(&t);
code = asprintf(&nextName, "%s.%d%02d%02d%02d%02d%02d",
fileName, timeFields->tm_year + 1900,
timeFields->tm_mon + 1, timeFields->tm_mday,
timeFields->tm_hour, timeFields->tm_min,
timeFields->tm_sec);
if (code < 0) {
nextName = NULL;
break;
}
if (lstat(nextName, &buf) == 0) {
/* Avoid clobbering a log. */
free(nextName);
nextName = NULL;
}
}
break;
default:
opr_Assert(0);
}
if (nextName != NULL) {
rk_rename(fileName, nextName); /* Don't check the error code. */
free(nextName);
}
}
char *
ucstring(char *d, const char *s, int n)
{
char *original_d = d;
char c;
opr_Assert(s != NULL && d != NULL);
while (n) {
c = *s++;
if (islower(c))
c = toupper(c);
*d++ = c;
if (c == 0)
break;
if (--n == 0)
*(d - 1) = 0; /* make sure null terminated */
}
return original_d;
}
static int
common_volop_prolog(struct cmd_syndesc * as, struct fssync_state * state)
{
struct cmd_item *ti;
state->vop = (struct volop_state *) calloc(1, sizeof(struct volop_state));
opr_Assert(state->vop != NULL);
if ((ti = as->parms[COMMON_VOLOP_PARMS_OFFSET].items)) { /* -volumeid */
state->vop->volume = atoi(ti->data);
} else {
fprintf(stderr, "required argument -volumeid not given\n");
}
if ((ti = as->parms[COMMON_VOLOP_PARMS_OFFSET+1].items)) { /* -partition */
strlcpy(state->vop->partName, ti->data, sizeof(state->vop->partName));
} else {
memset(state->vop->partName, 0, sizeof(state->vop->partName));
}
return 0;
}
/**
* acquire a lock on a file on local disk.
*
* @param[in] dl the VDiskLock structure corresponding to the file on disk
* @param[in] locktype READ_LOCK if you want a read lock, or WRITE_LOCK if
* you want a write lock
* @param[in] nonblock 0 to wait for conflicting locks to clear before
* obtaining the lock; 1 to fail immediately if a
* conflicting lock is held by someone else
*
* @return operation status
* @retval 0 success
* @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
* specified
* @retval EIO error acquiring file lock
*
* @note DAFS only
*
* @note while normal fcntl-y locks on Unix systems generally only work per-
* process, this interface also deals with locks between threads in the
* process in addition to different processes acquiring the lock
*/
int
VGetDiskLock(struct VDiskLock *dl, int locktype, int nonblock)
{
int code = 0;
opr_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
if (nonblock) {
if (locktype == READ_LOCK) {
ObtainReadLockNoBlock(&dl->rwlock, code);
} else {
ObtainWriteLockNoBlock(&dl->rwlock, code);
}
if (code) {
return EBUSY;
}
} else if (locktype == READ_LOCK) {
ObtainReadLock(&dl->rwlock);
} else {
ObtainWriteLock(&dl->rwlock);
}
opr_mutex_enter(&dl->mutex);
if ((dl->flags & VDISKLOCK_ACQUIRING)) {
/* Some other thread is waiting to acquire an fs lock. If nonblock=1,
* we can return immediately, since we know we'll need to wait to
* acquire. Otherwise, wait for the other thread to finish acquiring
* the fs lock */
if (nonblock) {
code = EBUSY;
} else {
while ((dl->flags & VDISKLOCK_ACQUIRING)) {
opr_cv_wait(&dl->cv, &dl->mutex);
}
}
}
if (code == 0 && !(dl->flags & VDISKLOCK_ACQUIRED)) {
/* no other thread holds the lock on the actual file; so grab one */
/* first try, don't block on the lock to see if we can get it without
* waiting */
code = VLockFileLock(dl->lockfile, dl->offset, locktype, 1);
if (code == EBUSY && !nonblock) {
/* mark that we are waiting on the fs lock */
dl->flags |= VDISKLOCK_ACQUIRING;
opr_mutex_exit(&dl->mutex);
code = VLockFileLock(dl->lockfile, dl->offset, locktype, nonblock);
opr_mutex_enter(&dl->mutex);
dl->flags &= ~VDISKLOCK_ACQUIRING;
if (code == 0) {
dl->flags |= VDISKLOCK_ACQUIRED;
}
opr_cv_broadcast(&dl->cv);
}
}
if (code) {
if (locktype == READ_LOCK) {
ReleaseReadLock(&dl->rwlock);
} else {
ReleaseWriteLock(&dl->rwlock);
}
} else {
/* successfully got the lock, so inc the number of unlocks we need
* to do before we can unlock the actual file */
++dl->lockers;
}
opr_mutex_exit(&dl->mutex);
return code;
}
/**
* Get a list of IP addresses for this host allowing only addresses found
* in the config file (fileName parameter): /usr/vice/etc/NetInfo for
* clients and /usr/afs/local/NetInfo for servers.
*
* All addresses should be in network byte order as returned by
* rx_getAllAddrMaskMtu() and parsed by extract_Addr().
*
* @param[out] outAddrs
* All the address that are found to be valid.
* @param[out] outMask
* Associated netmask for interface
* @param[out] outMtu
* Associated MTU for interface
* @param[in] max
* Length of the output above arrays
* @param[out] reason
* Reason for the parsing failure
* @param[in] fileName
* File to parse
* @param[in] fakeonly
* Only return addresses if they are marked as fake
*
* @return
* The number of valid address on success or < 0 on fatal failure.
*/
static int
ParseNetInfoFile_int(afs_uint32 outAddrs[], afs_uint32 outMask[], afs_uint32 outMtu[],
int max, char reason[], const char *fileName,
int fakeonly)
{
afs_uint32 existingAddr[MAXIPADDRS], existingMask[MAXIPADDRS],
existingMtu[MAXIPADDRS];
char line[MAX_NETFILE_LINE];
FILE *fp;
int i, existNu, count = 0;
afs_uint32 addr;
int lineNo = 0;
int l;
opr_Assert(fileName);
opr_Assert(outAddrs);
opr_Assert(outMask);
opr_Assert(outMtu);
opr_Assert(reason);
/* get all network interfaces from the kernel */
existNu =
rx_getAllAddrMaskMtu(existingAddr, existingMask, existingMtu,
MAXIPADDRS);
if (existNu < 0)
return existNu;
if ((fp = fopen(fileName, "r")) == 0) {
/* If file does not exist or is not readable, then
* use all interface addresses.
*/
sprintf(reason,
"Failed to open %s(%s)\nUsing all configured addresses\n",
fileName, strerror(errno));
for (i = 0; i < existNu; i++) {
outAddrs[i] = existingAddr[i];
outMask[i] = existingMask[i];
outMtu[i] = existingMtu[i];
}
return existNu;
}
/* For each line in the NetInfo file */
while (fgets(line, MAX_NETFILE_LINE, fp) != NULL) {
int fake = 0;
/* See if first char is an 'F' for fake */
/* Added to allow the fileserver to advertise fake IPS for use with
* the translation tables for NAT-like firewalls - defect 12462 */
for (fake = 0; ((fake < strlen(line)) && isspace(line[fake]));
fake++);
if ((fake < strlen(line))
&& ((line[fake] == 'f') || (line[fake] == 'F'))) {
fake++;
} else {
fake = 0;
}
lineNo++; /* input line number */
addr = extract_Addr(&line[fake], strlen(&line[fake]));
if (addr == AFS_IPINVALID) { /* syntactically invalid */
fprintf(stderr, "afs:%s : line %d : parse error\n", fileName,
lineNo);
continue;
}
if (addr == AFS_IPINVALIDIGNORE) { /* ignore error */
continue;
}
/* See if it is an address that really exists */
for (i = 0; i < existNu; i++) {
if (existingAddr[i] == addr)
break;
}
if ((i >= existNu) && (!fake))
continue; /* not found/fake - ignore */
/* Check if it is a duplicate address we alread have */
for (l = 0; l < count; l++) {
if (outAddrs[l] == addr)
break;
}
if (l < count) {
fprintf(stderr, "afs:%x specified twice in NetInfo file\n",
ntohl(addr));
continue; /* duplicate addr - ignore */
}
if (count > max) { /* no more space */
fprintf(stderr,
"afs:Too many interfaces. The current kernel configuration supports a maximum of %d interfaces\n",
max);
} else if (fake) {
if (!fake)
fprintf(stderr, "Client (2) also has address %s\n", line);
outAddrs[count] = addr;
outMask[count] = 0xffffffff;
outMtu[count] = htonl(1500);
count++;
} else if (!fakeonly) {
outAddrs[count] = existingAddr[i];
outMask[count] = existingMask[i];
outMtu[count] = existingMtu[i];
count++;
}
} /* while */
/* in case of any error, we use all the interfaces present */
if (count <= 0) {
sprintf(reason,
"Error in reading/parsing Interface file\nUsing all configured interface addresses \n");
for (i = 0; i < existNu; i++) {
outAddrs[i] = existingAddr[i];
outMask[i] = existingMask[i];
outMtu[i] = existingMtu[i];
}
return existNu;
}
return count;
}
/**
* Get a list of IP addresses for this host removing any address found
* in the config file (fileName parameter): /usr/vice/etc/NetRestrict
* for clients and /usr/afs/local/NetRestrict for servers.
*
* Returns the number of valid addresses in outAddrs[] and count in
* nAddrs. Returns 0 on success; or 1 if the config file was not
* there or empty (we still return the host's IP addresses). Returns
* -1 on fatal failure with reason in the reason argument (so the
* caller can choose to ignore the entire file but should write
* something to a log file).
*
* All addresses should be in network byte order as returned by
* rx_getAllAddrMaskMtu() and parsed by extract_Addr().
*
* @param[out] outAddrs
* All the address that are found to be valid.
* @param[out] outMask
* Optional associated netmask for address
* @param[out] outMtu
* Optional associated MTU for address
* @param[in] maxAddres
* Length of the above output arrays
* @param[out] nAddrs
* Count of valid addresses
* @param[out] reason
* Reason (if any) for the parsing failure
* @param[in] fileName
* Configuration file to parse
*
* @return
* 0 on success; 1 if the config file was not used; -1 on
* fatal failure.
*/
static int
parseNetRestrictFile_int(afs_uint32 outAddrs[], afs_uint32 outMask[],
afs_uint32 outMtu[], afs_uint32 maxAddrs,
afs_uint32 *nAddrs, char reason[],
const char *fileName, const char *fileName_ni)
{
FILE *fp;
char line[MAX_NETFILE_LINE];
int lineNo, usedfile = 0;
afs_uint32 i, neaddrs, nOutaddrs;
afs_uint32 addr, eAddrs[MAXIPADDRS], eMask[MAXIPADDRS], eMtu[MAXIPADDRS];
opr_Assert(outAddrs);
opr_Assert(reason);
opr_Assert(fileName);
opr_Assert(nAddrs);
if (outMask)
opr_Assert(outMtu);
/* Initialize */
*nAddrs = 0;
for (i = 0; i < maxAddrs; i++)
outAddrs[i] = 0;
strcpy(reason, "");
/* get all network interfaces from the kernel */
neaddrs = rx_getAllAddrMaskMtu(eAddrs, eMask, eMtu, MAXIPADDRS);
if (neaddrs <= 0) {
sprintf(reason, "No existing IP interfaces found");
return -1;
}
i = 0;
if ((neaddrs < MAXIPADDRS) && fileName_ni)
i = ParseNetInfoFile_int(&(eAddrs[neaddrs]), &(eMask[neaddrs]),
&(eMtu[neaddrs]), MAXIPADDRS-neaddrs, reason,
fileName_ni, 1);
if (i > 0)
neaddrs += i;
if ((fp = fopen(fileName, "r")) == 0) {
sprintf(reason, "Could not open file %s for reading:%s", fileName,
strerror(errno));
goto done;
}
/* For each line in the NetRestrict file */
lineNo = 0;
usedfile = 0;
while (fgets(line, MAX_NETFILE_LINE, fp) != NULL) {
lineNo++; /* input line number */
addr = extract_Addr(line, strlen(line));
if (addr == AFS_IPINVALID) { /* syntactically invalid */
fprintf(stderr, "%s : line %d : parse error - invalid IP\n",
fileName, lineNo);
continue;
}
if (addr == AFS_IPINVALIDIGNORE) { /* ignore error */
fprintf(stderr, "%s : line %d : invalid address ... ignoring\n",
fileName, lineNo);
continue;
}
usedfile = 1;
/* Check if we need to exclude this address */
for (i = 0; i < neaddrs; i++) {
if (eAddrs[i] && (eAddrs[i] == addr)) {
eAddrs[i] = 0; /* Yes - exclude it by zeroing it for now */
}
}
} /* while */
fclose(fp);
if (!usedfile) {
sprintf(reason, "No valid IP addresses in %s\n", fileName);
goto done;
}
done:
/* Collect the addresses we have left to return */
nOutaddrs = 0;
for (i = 0; i < neaddrs; i++) {
if (!eAddrs[i])
continue;
outAddrs[nOutaddrs] = eAddrs[i];
if (outMask) {
outMask[nOutaddrs] = eMask[i];
outMtu[nOutaddrs] = eMtu[i];
}
if (++nOutaddrs >= maxAddrs)
break;
}
if (nOutaddrs == 0) {
sprintf(reason, "No addresses to use after parsing %s", fileName);
return -1;
}
*nAddrs = nOutaddrs;
return (usedfile ? 0 : 1); /* 0=>used the file. 1=>didn't use file */
}
void
Afs_Lock_Obtain(struct Lock *lock, int how)
{
switch (how) {
case READ_LOCK:
lock->num_waiting++;
do {
lock->wait_states |= READ_LOCK;
#ifdef AFS_PTHREAD_ENV
opr_Verify(pthread_cond_wait(&lock->read_cv, &lock->mutex) == 0);
#else /* AFS_PTHREAD_ENV */
LWP_WaitProcess(&lock->readers_reading);
#endif /* AFS_PTHREAD_ENV */
} while (lock->excl_locked & WRITE_LOCK);
lock->num_waiting--;
lock->readers_reading++;
break;
case WRITE_LOCK:
lock->num_waiting++;
do {
lock->wait_states |= WRITE_LOCK;
#ifdef AFS_PTHREAD_ENV
opr_Verify(pthread_cond_wait(&lock->write_cv, &lock->mutex) == 0);
#else /* AFS_PTHREAD_ENV */
LWP_WaitProcess(&lock->excl_locked);
#endif /* AFS_PTHREAD_ENV */
} while (lock->excl_locked || lock->readers_reading);
lock->num_waiting--;
lock->excl_locked = WRITE_LOCK;
break;
case SHARED_LOCK:
lock->num_waiting++;
do {
lock->wait_states |= SHARED_LOCK;
#ifdef AFS_PTHREAD_ENV
opr_Verify(pthread_cond_wait(&lock->write_cv, &lock->mutex) == 0);
#else /* AFS_PTHREAD_ENV */
LWP_WaitProcess(&lock->excl_locked);
#endif /* AFS_PTHREAD_ENV */
} while (lock->excl_locked);
lock->num_waiting--;
lock->excl_locked = SHARED_LOCK;
break;
case BOOSTED_LOCK:
lock->num_waiting++;
do {
lock->wait_states |= WRITE_LOCK;
#ifdef AFS_PTHREAD_ENV
opr_Verify(pthread_cond_wait(&lock->write_cv, &lock->mutex) == 0);
#else /* AFS_PTHREAD_ENV */
LWP_WaitProcess(&lock->excl_locked);
#endif /* AFS_PTHREAD_ENV */
} while (lock->readers_reading);
lock->num_waiting--;
lock->excl_locked = WRITE_LOCK;
break;
default:
printf("Can't happen, bad LOCK type: %d\n", how);
opr_Assert(0);
}
}
void
opr_rbtree_remove(struct opr_rbtree *head, struct opr_rbtree_node *node)
{
struct opr_rbtree_node *child, *parent;
int red;
if (node->left == NULL && node->right == NULL) {
/* A node with no non-leaf children */
update_parent_ptr(head, node, NULL);
if (!node->red)
remove_recolour(head, node->parent, NULL);
return;
}
if (node->left != NULL && node->right != NULL) {
/* A node with two children.
*
* Move the next node in the tree (which will be a leaf node)
* onto our tree current position, then rebalance as required
*/
struct opr_rbtree_node *old, *left;
old = node;
/* Set node to the next node in the tree from the current
* position, where the next node is the left-most leaf node
* in our right child */
node = node->right;
while ((left = node->left) != NULL)
node = left;
/* Move 'node' into the position occupied by 'old', which is being
* removed */
update_parent_ptr(head, old, node);
child = node->right;
parent = node->parent;
red = node->red;
/* As we're logically just copying the value, must preserve the
* old node's colour */
node->red = old->red;
/* ... and the old node's linkage */
if (parent == old)
parent = node;
else {
if (child)
child->parent = parent;
parent->left = child;
node->right = old->right;
old->right->parent = node;
}
node->parent = old->parent;
node->left = old->left;
old->left->parent = node;
/* If the node being removed was black, then we must recolour the
* tree to maintain balance */
if (!red)
remove_recolour(head, parent, child);
return;
}
/* Only remaining option - node with a single child */
if (node->left == NULL)
child = node->right;
else {
opr_Assert(node->right == NULL);
child = node->left;
}
child->parent = node->parent;
update_parent_ptr(head, node, child);
if (!node->red)
remove_recolour(head, node->parent, child);
}
static int
usd_FileOpen(const char *path, int flags, int mode, usd_handle_t * usdP)
{
int fd;
int oflags;
usd_handle_t usd;
int code;
if (usdP)
*usdP = NULL;
oflags = (flags & USD_OPEN_RDWR) ? O_RDWR : O_RDONLY;
#ifdef O_SYNC /* AFS_DARWIN_ENV XXX */
if (flags & USD_OPEN_SYNC)
oflags |= O_SYNC;
#endif
if (flags & USD_OPEN_CREATE)
oflags |= O_CREAT;
#ifdef O_LARGEFILE
fd = open64(path, oflags | O_LARGEFILE, mode);
#else /* O_LARGEFILE */
fd = open(path, oflags, mode);
#endif /* O_LARGEFILE */
if (fd == -1)
return errno;
usd = calloc(1, sizeof(*usd));
usd->handle = (void *)(intptr_t)fd;
usd->read = usd_FileRead;
usd->write = usd_FileWrite;
usd->seek = usd_FileSeek;
usd->ioctl = usd_FileIoctl;
usd->close = usd_FileClose;
usd->fullPathName = strdup(path);
usd->openFlags = flags;
code = 0;
if (flags & (USD_OPEN_RLOCK | USD_OPEN_WLOCK)) {
#ifdef O_LARGEFILE
struct flock64 fl;
#else /* O_LARGEFILE */
struct flock fl;
#endif /* O_LARGEFILE */
/* make sure both lock bits aren't set */
opr_Assert(~flags & (USD_OPEN_RLOCK | USD_OPEN_WLOCK));
fl.l_type = ((flags & USD_OPEN_RLOCK) ? F_RDLCK : F_WRLCK);
fl.l_whence = SEEK_SET;
fl.l_start = (osi_lloff_t) 0;
fl.l_len = (osi_lloff_t) 0; /* whole file */
#ifdef O_LARGEFILE
code = fcntl(fd, F_SETLK64, &fl);
#else /* O_LARGEFILE */
code = fcntl(fd, F_SETLK, &fl);
#endif /* O_LARGEFILE */
if (code == -1)
code = errno;
/* If we're trying to obtain a write lock on a real disk, then the
* aggregate must not be attached by the kernel. If so, unlock it
* and fail.
* WARNING: The code to check for the above has been removed when this
* file was ported from DFS src. It should be put back if
* this library is used to access hard disks
*/
}
if (code == 0 && usdP)
*usdP = usd;
else
usd_FileClose(usd);
return code;
}