It could be triggered by delayed name resolving as it adds multiple new timeouts which can be called in the same dispatching if the DNS responses are slower than initial delay and sampling separation. Compare number of dispatched events also with current number of timeouts.
651 lines
16 KiB
C
651 lines
16 KiB
C
/*
|
|
chronyd/chronyc - Programs for keeping computer clocks accurate.
|
|
|
|
**********************************************************************
|
|
* Copyright (C) Richard P. Curnow 1997-2003
|
|
* Copyright (C) Miroslav Lichvar 2011
|
|
*
|
|
* This program is free software; you can redistribute it and/or modify
|
|
* it under the terms of version 2 of the GNU General Public License as
|
|
* published by the Free Software Foundation.
|
|
*
|
|
* This program is distributed in the hope that it will be useful, but
|
|
* WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
* General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License along
|
|
* with this program; if not, write to the Free Software Foundation, Inc.,
|
|
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
*
|
|
**********************************************************************
|
|
|
|
=======================================================================
|
|
|
|
This file contains the scheduling loop and the timeout queue.
|
|
|
|
*/
|
|
|
|
#include "config.h"
|
|
|
|
#include "sysincl.h"
|
|
|
|
#include "sched.h"
|
|
#include "memory.h"
|
|
#include "util.h"
|
|
#include "local.h"
|
|
#include "logging.h"
|
|
|
|
/* ================================================== */
|
|
|
|
/* Flag indicating that we are initialised */
|
|
static int initialised = 0;
|
|
|
|
/* ================================================== */
|
|
|
|
/* Variables to handle the capability to dispatch on particular file
|
|
handles becoming readable */
|
|
|
|
/* Each bit set in this fd set corresponds to a read descriptor that
|
|
we are watching and with which we have a handler associated in the
|
|
file_handlers array */
|
|
static fd_set read_fds;
|
|
|
|
/* This is the number of bits that we have set in read_fds */
|
|
static unsigned int n_read_fds;
|
|
|
|
/* One more than the highest file descriptor that is registered */
|
|
static unsigned int one_highest_fd;
|
|
|
|
/* This assumes that fd_set is implemented as a fixed size array of
|
|
bits, possibly embedded inside a record. It might therefore
|
|
somewhat non-portable. */
|
|
|
|
#define FD_SET_SIZE (sizeof(fd_set) * 8)
|
|
|
|
typedef struct {
|
|
SCH_FileHandler handler;
|
|
SCH_ArbitraryArgument arg;
|
|
} FileHandlerEntry;
|
|
|
|
static FileHandlerEntry file_handlers[FD_SET_SIZE];
|
|
|
|
/* Timestamp when last select() returned */
|
|
static struct timeval last_select_ts, last_select_ts_raw;
|
|
static double last_select_ts_err;
|
|
|
|
/* ================================================== */
|
|
|
|
/* Variables to handler the timer queue */
|
|
|
|
typedef struct _TimerQueueEntry
|
|
{
|
|
struct _TimerQueueEntry *next; /* Forward and back links in the list */
|
|
struct _TimerQueueEntry *prev;
|
|
struct timeval tv; /* Local system time at which the
|
|
timeout is to expire. Clearly this
|
|
must be in terms of what the
|
|
operating system thinks of as
|
|
system time, because it will be an
|
|
argument to select(). Therefore,
|
|
any fudges etc that our local time
|
|
driver module would apply to time
|
|
that we pass to clients etc doesn't
|
|
apply to this. */
|
|
SCH_TimeoutID id; /* ID to allow client to delete
|
|
timeout */
|
|
SCH_TimeoutClass class; /* The class that the epoch is in */
|
|
SCH_TimeoutHandler handler; /* The handler routine to use */
|
|
SCH_ArbitraryArgument arg; /* The argument to pass to the handler */
|
|
|
|
} TimerQueueEntry;
|
|
|
|
/* The timer queue. We only use the next and prev entries of this
|
|
record, these chain to the real entries. */
|
|
static TimerQueueEntry timer_queue;
|
|
static unsigned long n_timer_queue_entries;
|
|
static SCH_TimeoutID next_tqe_id;
|
|
|
|
/* Pointer to head of free list */
|
|
static TimerQueueEntry *tqe_free_list = NULL;
|
|
|
|
/* Timestamp when was last timeout dispatched for each class */
|
|
static struct timeval last_class_dispatch[SCH_NumberOfClasses];
|
|
|
|
/* ================================================== */
|
|
|
|
static int need_to_exit;
|
|
|
|
/* ================================================== */
|
|
|
|
static void
|
|
handle_slew(struct timeval *raw,
|
|
struct timeval *cooked,
|
|
double dfreq,
|
|
double doffset,
|
|
int is_step_change,
|
|
void *anything);
|
|
|
|
/* ================================================== */
|
|
|
|
void
|
|
SCH_Initialise(void)
|
|
{
|
|
struct timeval tv;
|
|
|
|
FD_ZERO(&read_fds);
|
|
n_read_fds = 0;
|
|
|
|
n_timer_queue_entries = 0;
|
|
next_tqe_id = 0;
|
|
|
|
timer_queue.next = &timer_queue;
|
|
timer_queue.prev = &timer_queue;
|
|
|
|
need_to_exit = 0;
|
|
|
|
LCL_AddParameterChangeHandler(handle_slew, NULL);
|
|
|
|
LCL_ReadRawTime(&tv);
|
|
srandom(tv.tv_sec * tv.tv_usec);
|
|
|
|
initialised = 1;
|
|
|
|
return;
|
|
}
|
|
|
|
|
|
/* ================================================== */
|
|
|
|
void
|
|
SCH_Finalise(void) {
|
|
initialised = 0;
|
|
return; /* Nothing to do for now */
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
void
|
|
SCH_AddInputFileHandler
|
|
(int fd, SCH_FileHandler handler, SCH_ArbitraryArgument arg)
|
|
{
|
|
|
|
assert(initialised);
|
|
|
|
/* Don't want to allow the same fd to register a handler more than
|
|
once without deleting a previous association - this suggests
|
|
a bug somewhere else in the program. */
|
|
assert(!FD_ISSET(fd, &read_fds));
|
|
|
|
++n_read_fds;
|
|
|
|
file_handlers[fd].handler = handler;
|
|
file_handlers[fd].arg = arg;
|
|
|
|
FD_SET(fd, &read_fds);
|
|
|
|
if ((fd + 1) > one_highest_fd) {
|
|
one_highest_fd = fd + 1;
|
|
}
|
|
|
|
return;
|
|
}
|
|
|
|
|
|
/* ================================================== */
|
|
|
|
void
|
|
SCH_RemoveInputFileHandler(int fd)
|
|
{
|
|
int fds_left, fd_to_check;
|
|
|
|
assert(initialised);
|
|
|
|
/* Check that a handler was registered for the fd in question */
|
|
assert(FD_ISSET(fd, &read_fds));
|
|
|
|
--n_read_fds;
|
|
|
|
FD_CLR(fd, &read_fds);
|
|
|
|
/* Find new highest file descriptor */
|
|
fds_left = n_read_fds;
|
|
fd_to_check = 0;
|
|
while (fds_left > 0) {
|
|
if (FD_ISSET(fd_to_check, &read_fds)) {
|
|
--fds_left;
|
|
}
|
|
++fd_to_check;
|
|
}
|
|
|
|
one_highest_fd = fd_to_check;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
void
|
|
SCH_GetFileReadyTime(struct timeval *tv, double *err)
|
|
{
|
|
*tv = last_select_ts;
|
|
if (err)
|
|
*err = last_select_ts_err;
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
#define TQE_ALLOC_QUANTUM 32
|
|
|
|
static TimerQueueEntry *
|
|
allocate_tqe(void)
|
|
{
|
|
TimerQueueEntry *new_block;
|
|
TimerQueueEntry *result;
|
|
int i;
|
|
if (tqe_free_list == NULL) {
|
|
new_block = MallocArray(TimerQueueEntry, TQE_ALLOC_QUANTUM);
|
|
for (i=1; i<TQE_ALLOC_QUANTUM; i++) {
|
|
new_block[i].next = &(new_block[i-1]);
|
|
}
|
|
new_block[0].next = NULL;
|
|
tqe_free_list = &(new_block[TQE_ALLOC_QUANTUM - 1]);
|
|
}
|
|
|
|
result = tqe_free_list;
|
|
tqe_free_list = tqe_free_list->next;
|
|
return result;
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
static void
|
|
release_tqe(TimerQueueEntry *node)
|
|
{
|
|
node->next = tqe_free_list;
|
|
tqe_free_list = node;
|
|
return;
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
SCH_TimeoutID
|
|
SCH_AddTimeout(struct timeval *tv, SCH_TimeoutHandler handler, SCH_ArbitraryArgument arg)
|
|
{
|
|
TimerQueueEntry *new_tqe;
|
|
TimerQueueEntry *ptr;
|
|
|
|
assert(initialised);
|
|
|
|
new_tqe = allocate_tqe();
|
|
|
|
new_tqe->id = next_tqe_id++;
|
|
new_tqe->handler = handler;
|
|
new_tqe->arg = arg;
|
|
new_tqe->tv = *tv;
|
|
new_tqe->class = SCH_ReservedTimeoutValue;
|
|
|
|
/* Now work out where to insert the new entry in the list */
|
|
for (ptr = timer_queue.next; ptr != &timer_queue; ptr = ptr->next) {
|
|
if (UTI_CompareTimevals(&new_tqe->tv, &ptr->tv) == -1) {
|
|
/* If the new entry comes before the current pointer location in
|
|
the list, we want to insert the new entry just before ptr. */
|
|
break;
|
|
}
|
|
}
|
|
|
|
/* At this stage, we want to insert the new entry immediately before
|
|
the entry identified by 'ptr' */
|
|
|
|
new_tqe->next = ptr;
|
|
new_tqe->prev = ptr->prev;
|
|
ptr->prev->next = new_tqe;
|
|
ptr->prev = new_tqe;
|
|
|
|
n_timer_queue_entries++;
|
|
|
|
return new_tqe->id;
|
|
}
|
|
|
|
/* ================================================== */
|
|
/* This queues a timeout to elapse at a given delta time relative to
|
|
the current (raw) time */
|
|
|
|
SCH_TimeoutID
|
|
SCH_AddTimeoutByDelay(double delay, SCH_TimeoutHandler handler, SCH_ArbitraryArgument arg)
|
|
{
|
|
struct timeval now, then;
|
|
|
|
assert(initialised);
|
|
assert(delay >= 0.0);
|
|
|
|
LCL_ReadRawTime(&now);
|
|
UTI_AddDoubleToTimeval(&now, delay, &then);
|
|
return SCH_AddTimeout(&then, handler, arg);
|
|
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
SCH_TimeoutID
|
|
SCH_AddTimeoutInClass(double min_delay, double separation, double randomness,
|
|
SCH_TimeoutClass class,
|
|
SCH_TimeoutHandler handler, SCH_ArbitraryArgument arg)
|
|
{
|
|
TimerQueueEntry *new_tqe;
|
|
TimerQueueEntry *ptr;
|
|
struct timeval now;
|
|
double diff, r;
|
|
double new_min_delay;
|
|
|
|
assert(initialised);
|
|
assert(min_delay >= 0.0);
|
|
assert(class < SCH_NumberOfClasses);
|
|
|
|
if (randomness > 0.0) {
|
|
r = random() % 0xffff / (0xffff - 1.0) * randomness + 1.0;
|
|
min_delay *= r;
|
|
separation *= r;
|
|
}
|
|
|
|
LCL_ReadRawTime(&now);
|
|
new_min_delay = min_delay;
|
|
|
|
/* Check the separation from the last dispatched timeout */
|
|
UTI_DiffTimevalsToDouble(&diff, &now, &last_class_dispatch[class]);
|
|
if (diff < separation && diff >= 0.0 && diff + new_min_delay < separation) {
|
|
new_min_delay = separation - diff;
|
|
}
|
|
|
|
/* Scan through list for entries in the same class and increase min_delay
|
|
if necessary to keep at least the separation away */
|
|
for (ptr = timer_queue.next; ptr != &timer_queue; ptr = ptr->next) {
|
|
if (ptr->class == class) {
|
|
UTI_DiffTimevalsToDouble(&diff, &ptr->tv, &now);
|
|
if (new_min_delay > diff) {
|
|
if (new_min_delay - diff < separation) {
|
|
new_min_delay = diff + separation;
|
|
}
|
|
} else {
|
|
if (diff - new_min_delay < separation) {
|
|
new_min_delay = diff + separation;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
for (ptr = timer_queue.next; ptr != &timer_queue; ptr = ptr->next) {
|
|
UTI_DiffTimevalsToDouble(&diff, &ptr->tv, &now);
|
|
if (diff > new_min_delay) {
|
|
break;
|
|
}
|
|
}
|
|
|
|
/* We have located the insertion point */
|
|
new_tqe = allocate_tqe();
|
|
|
|
new_tqe->id = next_tqe_id++;
|
|
new_tqe->handler = handler;
|
|
new_tqe->arg = arg;
|
|
UTI_AddDoubleToTimeval(&now, new_min_delay, &new_tqe->tv);
|
|
new_tqe->class = class;
|
|
|
|
new_tqe->next = ptr;
|
|
new_tqe->prev = ptr->prev;
|
|
ptr->prev->next = new_tqe;
|
|
ptr->prev = new_tqe;
|
|
n_timer_queue_entries++;
|
|
|
|
return new_tqe->id;
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
void
|
|
SCH_RemoveTimeout(SCH_TimeoutID id)
|
|
{
|
|
TimerQueueEntry *ptr;
|
|
|
|
assert(initialised);
|
|
|
|
for (ptr = timer_queue.next; ptr != &timer_queue; ptr = ptr->next) {
|
|
|
|
if (ptr->id == id) {
|
|
/* Found the required entry */
|
|
|
|
/* Unlink from the queue */
|
|
ptr->next->prev = ptr->prev;
|
|
ptr->prev->next = ptr->next;
|
|
|
|
/* Decrement entry count */
|
|
--n_timer_queue_entries;
|
|
|
|
/* Release memory back to the operating system */
|
|
release_tqe(ptr);
|
|
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
/* ================================================== */
|
|
/* Try to dispatch any timeouts that have already gone by, and
|
|
keep going until all are done. (The earlier ones may take so
|
|
long to do that the later ones come around by the time they are
|
|
completed). */
|
|
|
|
static void
|
|
dispatch_timeouts(struct timeval *now) {
|
|
TimerQueueEntry *ptr;
|
|
SCH_TimeoutHandler handler;
|
|
SCH_ArbitraryArgument arg;
|
|
int n_done = 0, n_entries_on_start = n_timer_queue_entries;
|
|
|
|
while (1) {
|
|
LCL_ReadRawTime(now);
|
|
|
|
if (!(n_timer_queue_entries > 0 &&
|
|
UTI_CompareTimevals(now, &(timer_queue.next->tv)) >= 0)) {
|
|
break;
|
|
}
|
|
|
|
ptr = timer_queue.next;
|
|
|
|
last_class_dispatch[ptr->class] = *now;
|
|
|
|
handler = ptr->handler;
|
|
arg = ptr->arg;
|
|
|
|
SCH_RemoveTimeout(ptr->id);
|
|
|
|
/* Dispatch the handler */
|
|
(handler)(arg);
|
|
|
|
/* Increment count of timeouts handled */
|
|
++n_done;
|
|
|
|
/* If more timeouts were handled than there were in the timer queue on
|
|
start and there are now, assume some code is scheduling timeouts with
|
|
negative delays and abort. Make the actual limit higher in case the
|
|
machine is temporarily overloaded and dispatching the handlers takes
|
|
more time than was delay of a scheduled timeout. */
|
|
if (n_done > n_timer_queue_entries * 4 &&
|
|
n_done > n_entries_on_start * 4) {
|
|
LOG_FATAL(LOGF_Scheduler, "Possible infinite loop in scheduling");
|
|
}
|
|
}
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
/* nfh is the number of bits set in fhs */
|
|
|
|
static void
|
|
dispatch_filehandlers(int nfh, fd_set *fhs)
|
|
{
|
|
int fh = 0;
|
|
|
|
while (nfh > 0) {
|
|
if (FD_ISSET(fh, fhs)) {
|
|
|
|
/* This descriptor can be read from, dispatch its handler */
|
|
(file_handlers[fh].handler)(file_handlers[fh].arg);
|
|
|
|
/* Decrement number of readable files still to find */
|
|
--nfh;
|
|
}
|
|
|
|
++fh;
|
|
}
|
|
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
static void
|
|
handle_slew(struct timeval *raw,
|
|
struct timeval *cooked,
|
|
double dfreq,
|
|
double doffset,
|
|
int is_step_change,
|
|
void *anything)
|
|
{
|
|
TimerQueueEntry *ptr;
|
|
int i;
|
|
|
|
if (is_step_change) {
|
|
/* We're not interested in anything else - it won't affect the
|
|
functionality of timer event dispatching. If a step change
|
|
occurs, just shift all the timeouts by the offset */
|
|
|
|
for (ptr = timer_queue.next; ptr != &timer_queue; ptr = ptr->next) {
|
|
UTI_AddDoubleToTimeval(&ptr->tv, -doffset, &ptr->tv);
|
|
}
|
|
|
|
for (i = 0; i < SCH_NumberOfClasses; i++) {
|
|
UTI_AddDoubleToTimeval(&last_class_dispatch[i], -doffset, &last_class_dispatch[i]);
|
|
}
|
|
|
|
UTI_AddDoubleToTimeval(&last_select_ts_raw, -doffset, &last_select_ts_raw);
|
|
UTI_AddDoubleToTimeval(&last_select_ts, -doffset, &last_select_ts);
|
|
}
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
/* Try to handle unexpected backward time jump */
|
|
|
|
static void
|
|
recover_backjump(struct timeval *raw, struct timeval *cooked, int timeout)
|
|
{
|
|
double diff, err;
|
|
|
|
UTI_DiffTimevalsToDouble(&diff, &last_select_ts_raw, raw);
|
|
|
|
if (n_timer_queue_entries > 0) {
|
|
UTI_DiffTimevalsToDouble(&err, &(timer_queue.next->tv), &last_select_ts_raw);
|
|
} else {
|
|
err = 0.0;
|
|
}
|
|
|
|
diff += err;
|
|
|
|
if (timeout) {
|
|
err = 1.0;
|
|
}
|
|
|
|
LOG(LOGS_WARN, LOGF_Scheduler, "Backward time jump detected! (correction %.1f +- %.1f seconds)", diff, err);
|
|
|
|
LCL_NotifyExternalTimeStep(raw, cooked, diff, err);
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
void
|
|
SCH_MainLoop(void)
|
|
{
|
|
fd_set rd;
|
|
int status;
|
|
struct timeval tv, *ptv;
|
|
struct timeval now, cooked;
|
|
double err;
|
|
|
|
assert(initialised);
|
|
|
|
while (!need_to_exit) {
|
|
|
|
/* Copy current set of read file descriptors */
|
|
memcpy((void *) &rd, (void *) &read_fds, sizeof(fd_set));
|
|
|
|
/* Dispatch timeouts and fill now with current raw time */
|
|
dispatch_timeouts(&now);
|
|
|
|
/* Check whether there is a timeout and set it up */
|
|
if (n_timer_queue_entries > 0) {
|
|
|
|
UTI_DiffTimevals(&tv, &(timer_queue.next->tv), &now);
|
|
ptv = &tv;
|
|
assert(tv.tv_sec > 0 || tv.tv_usec > 0);
|
|
|
|
} else {
|
|
ptv = NULL;
|
|
}
|
|
|
|
/* if there are no file descriptors being waited on and no
|
|
timeout set, this is clearly ridiculous, so stop the run */
|
|
|
|
if (!ptv && (n_read_fds == 0)) {
|
|
LOG_FATAL(LOGF_Scheduler, "No descriptors or timeout to wait for");
|
|
}
|
|
|
|
status = select(one_highest_fd, &rd, NULL, NULL, ptv);
|
|
|
|
LCL_ReadRawTime(&now);
|
|
LCL_CookTime(&now, &cooked, &err);
|
|
|
|
/* Check if time didn't jump backwards */
|
|
if (last_select_ts_raw.tv_sec > now.tv_sec + 1) {
|
|
recover_backjump(&now, &cooked, status == 0);
|
|
}
|
|
|
|
last_select_ts_raw = now;
|
|
last_select_ts = cooked;
|
|
last_select_ts_err = err;
|
|
|
|
if (status < 0) {
|
|
assert(need_to_exit);
|
|
} else if (status > 0) {
|
|
/* A file descriptor is ready to read */
|
|
|
|
dispatch_filehandlers(status, &rd);
|
|
|
|
} else {
|
|
assert(status == 0);
|
|
|
|
/* No descriptors readable, timeout must have elapsed.
|
|
Therefore, tv must be non-null */
|
|
assert(ptv);
|
|
|
|
/* There's nothing to do here, since the timeouts
|
|
will be dispatched at the top of the next loop
|
|
cycle */
|
|
|
|
}
|
|
}
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
/* ================================================== */
|
|
|
|
void
|
|
SCH_QuitProgram(void)
|
|
{
|
|
assert(initialised);
|
|
need_to_exit = 1;
|
|
}
|
|
|
|
/* ================================================== */
|
|
|