lib/timeval.c

   1 /*
   2  * Copyright (c) 2008, 2009 Nicira Networks.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18 #include "timeval.h"
  19 #include <assert.h>
  20 #include <errno.h>
  21 #include <poll.h>
  22 #include <signal.h>
  23 #include <string.h>
  24 #include <sys/time.h>
  25 #include <sys/resource.h>
  26 #include <unistd.h>
  27 #include "coverage.h"
  28 #include "fatal-signal.h"
  29 #include "util.h"
  30
  31 #include "vlog.h"
  32 #define THIS_MODULE VLM_timeval
  33
  34 /* Initialized? */
  35 static bool inited;
  36
  37 /* Has a timer tick occurred? */
  38 static volatile sig_atomic_t tick;
  39
  40 /* The current time, as of the last refresh. */
  41 static struct timeval now;
  42
  43 /* Time at which to die with SIGALRM (if not TIME_MIN). */
  44 static time_t deadline = TIME_MIN;
  45
  46 static void set_up_timer(void);
  47 static void set_up_signal(int flags);
  48 static void sigalrm_handler(int);
  49 static void refresh_if_ticked(void);
  50 static time_t time_add(time_t, time_t);
  51 static void block_sigalrm(sigset_t *);
  52 static void unblock_sigalrm(const sigset_t *);
  53 static void log_poll_interval(long long int last_wakeup,
  54                               const struct rusage *last_rusage);
  55
  56 /* Initializes the timetracking module. */
  57 void
  58 time_init(void)
  59 {
  60     if (inited) {
  61         return;
  62     }
  63
  64     coverage_init();
  65
  66     inited = true;
  67     gettimeofday(&now, NULL);
  68     tick = false;
  69
  70     set_up_signal(SA_RESTART);
  71     set_up_timer();
  72 }
  73
  74 static void
  75 set_up_signal(int flags)
  76 {
  77     struct sigaction sa;
  78
  79     memset(&sa, 0, sizeof sa);
  80     sa.sa_handler = sigalrm_handler;
  81     sigemptyset(&sa.sa_mask);
  82     sa.sa_flags = flags;
  83     if (sigaction(SIGALRM, &sa, NULL)) {
  84         ovs_fatal(errno, "sigaction(SIGALRM) failed");
  85     }
  86 }
  87
  88 /* Remove SA_RESTART from the flags for SIGALRM, so that any system call that
  89  * is interrupted by the periodic timer interrupt will return EINTR instead of
  90  * continuing after the signal handler returns.
  91  *
  92  * time_disable_restart() and time_enable_restart() may be usefully wrapped
  93  * around function calls that might otherwise block forever unless interrupted
  94  * by a signal, e.g.:
  95  *
  96  *   time_disable_restart();
  97  *   fcntl(fd, F_SETLKW, &lock);
  98  *   time_enable_restart();
  99  */
 100 void
 101 time_disable_restart(void)
 102 {
 103     set_up_signal(0);
 104 }
 105
 106 /* Add SA_RESTART to the flags for SIGALRM, so that any system call that
 107  * is interrupted by the periodic timer interrupt will continue after the
 108  * signal handler returns instead of returning EINTR. */
 109 void
 110 time_enable_restart(void)
 111 {
 112     set_up_signal(SA_RESTART);
 113 }
 114
 115 static void
 116 set_up_timer(void)
 117 {
 118     struct itimerval itimer;
 119
 120     itimer.it_interval.tv_sec = 0;
 121     itimer.it_interval.tv_usec = TIME_UPDATE_INTERVAL * 1000;
 122     itimer.it_value = itimer.it_interval;
 123     if (setitimer(ITIMER_REAL, &itimer, NULL)) {
 124         ovs_fatal(errno, "setitimer failed");
 125     }
 126 }
 127
 128 /* Set up the interval timer, to ensure that time advances even without calling
 129  * time_refresh().
 130  *
 131  * A child created with fork() does not inherit the parent's interval timer, so
 132  * this function needs to be called from the child after fork(). */
 133 void
 134 time_postfork(void)
 135 {
 136     set_up_timer();
 137 }
 138
 139 /* Forces a refresh of the current time from the kernel.  It is not usually
 140  * necessary to call this function, since the time will be refreshed
 141  * automatically at least every TIME_UPDATE_INTERVAL milliseconds. */
 142 void
 143 time_refresh(void)
 144 {
 145     gettimeofday(&now, NULL);
 146     tick = false;
 147 }
 148
 149 /* Returns the current time, in seconds. */
 150 time_t
 151 time_now(void)
 152 {
 153     refresh_if_ticked();
 154     return now.tv_sec;
 155 }
 156
 157 /* Returns the current time, in ms (within TIME_UPDATE_INTERVAL ms). */
 158 long long int
 159 time_msec(void)
 160 {
 161     refresh_if_ticked();
 162     return timeval_to_msec(&now);
 163 }
 164
 165 /* Stores the current time, accurate within TIME_UPDATE_INTERVAL ms, into
 166  * '*tv'. */
 167 void
 168 time_timeval(struct timeval *tv)
 169 {
 170     refresh_if_ticked();
 171     *tv = now;
 172 }
 173
 174 /* Configures the program to die with SIGALRM 'secs' seconds from now, if
 175  * 'secs' is nonzero, or disables the feature if 'secs' is zero. */
 176 void
 177 time_alarm(unsigned int secs)
 178 {
 179     sigset_t oldsigs;
 180
 181     time_init();
 182     block_sigalrm(&oldsigs);
 183     deadline = secs ? time_add(time_now(), secs) : TIME_MIN;
 184     unblock_sigalrm(&oldsigs);
 185 }
 186
 187 /* Like poll(), except:
 188  *
 189  *      - On error, returns a negative error code (instead of setting errno).
 190  *
 191  *      - If interrupted by a signal, retries automatically until the original
 192  *        'timeout' expires.  (Because of this property, this function will
 193  *        never return -EINTR.)
 194  *
 195  *      - As a side effect, refreshes the current time (like time_refresh()).
 196  */
 197 int
 198 time_poll(struct pollfd *pollfds, int n_pollfds, int timeout)
 199 {
 200     static long long int last_wakeup;
 201     static struct rusage last_rusage;
 202     long long int start;
 203     sigset_t oldsigs;
 204     bool blocked;
 205     int retval;
 206
 207     time_refresh();
 208     log_poll_interval(last_wakeup, &last_rusage);
 209     coverage_clear();
 210     start = time_msec();
 211     blocked = false;
 212     for (;;) {
 213         int time_left;
 214         if (timeout > 0) {
 215             long long int elapsed = time_msec() - start;
 216             time_left = timeout >= elapsed ? timeout - elapsed : 0;
 217         } else {
 218             time_left = timeout;
 219         }
 220
 221         retval = poll(pollfds, n_pollfds, time_left);
 222         if (retval < 0) {
 223             retval = -errno;
 224         }
 225         time_refresh();
 226         if (retval != -EINTR) {
 227             break;
 228         }
 229
 230         if (!blocked && deadline == TIME_MIN) {
 231             block_sigalrm(&oldsigs);
 232             blocked = true;
 233         }
 234     }
 235     if (blocked) {
 236         unblock_sigalrm(&oldsigs);
 237     }
 238     last_wakeup = time_msec();
 239     getrusage(RUSAGE_SELF, &last_rusage);
 240     return retval;
 241 }
 242
 243 /* Returns the sum of 'a' and 'b', with saturation on overflow or underflow. */
 244 static time_t
 245 time_add(time_t a, time_t b)
 246 {
 247     return (a >= 0
 248             ? (b > TIME_MAX - a ? TIME_MAX : a + b)
 249             : (b < TIME_MIN - a ? TIME_MIN : a + b));
 250 }
 251
 252 static void
 253 sigalrm_handler(int sig_nr)
 254 {
 255     tick = true;
 256     if (deadline != TIME_MIN && time(0) > deadline) {
 257         fatal_signal_handler(sig_nr);
 258     }
 259 }
 260
 261 static void
 262 refresh_if_ticked(void)
 263 {
 264     assert(inited);
 265     if (tick) {
 266         time_refresh();
 267     }
 268 }
 269
 270 static void
 271 block_sigalrm(sigset_t *oldsigs)
 272 {
 273     sigset_t sigalrm;
 274     sigemptyset(&sigalrm);
 275     sigaddset(&sigalrm, SIGALRM);
 276     if (sigprocmask(SIG_BLOCK, &sigalrm, oldsigs)) {
 277         ovs_fatal(errno, "sigprocmask");
 278     }
 279 }
 280
 281 static void
 282 unblock_sigalrm(const sigset_t *oldsigs)
 283 {
 284     if (sigprocmask(SIG_SETMASK, oldsigs, NULL)) {
 285         ovs_fatal(errno, "sigprocmask");
 286     }
 287 }
 288
 289 long long int
 290 timeval_to_msec(const struct timeval *tv)
 291 {
 292     return (long long int) tv->tv_sec * 1000 + tv->tv_usec / 1000;
 293 }
 294
 295 static long long int
 296 timeval_diff_msec(const struct timeval *a, const struct timeval *b)
 297 {
 298     return timeval_to_msec(a) - timeval_to_msec(b);
 299 }
 300
 301 static void
 302 log_poll_interval(long long int last_wakeup, const struct rusage *last_rusage)
 303 {
 304     static unsigned int mean_interval; /* In 16ths of a millisecond. */
 305     static unsigned int n_samples;
 306
 307     long long int now;
 308     unsigned int interval;      /* In 16ths of a millisecond. */
 309
 310     /* Compute interval from last wakeup to now in 16ths of a millisecond,
 311      * capped at 10 seconds (16000 in this unit). */
 312     now = time_msec();
 313     interval = MIN(10000, now - last_wakeup) << 4;
 314
 315     /* Warn if we took too much time between polls. */
 316     if (n_samples > 10 && interval > mean_interval * 8) {
 317         struct rusage rusage;
 318
 319         getrusage(RUSAGE_SELF, &rusage);
 320         VLOG_WARN("%u ms poll interval (%lld ms user, %lld ms system) "
 321                   "is over %u times the weighted mean interval %u ms "
 322                   "(%u samples)",
 323                   (interval + 8) / 16,
 324                   timeval_diff_msec(&rusage.ru_utime, &last_rusage->ru_utime),
 325                   timeval_diff_msec(&rusage.ru_stime, &last_rusage->ru_stime),
 326                   interval / mean_interval,
 327                   (mean_interval + 8) / 16, n_samples);
 328         if (rusage.ru_minflt > last_rusage->ru_minflt
 329             || rusage.ru_majflt > last_rusage->ru_majflt) {
 330             VLOG_WARN("faults: %ld minor, %ld major",
 331                       rusage.ru_minflt - last_rusage->ru_minflt,
 332                       rusage.ru_majflt - last_rusage->ru_majflt);
 333         }
 334         if (rusage.ru_inblock > last_rusage->ru_inblock
 335             || rusage.ru_oublock > last_rusage->ru_oublock) {
 336             VLOG_WARN("disk: %ld reads, %ld writes",
 337                       rusage.ru_inblock - last_rusage->ru_inblock,
 338                       rusage.ru_oublock - last_rusage->ru_oublock);
 339         }
 340         if (rusage.ru_nvcsw > last_rusage->ru_nvcsw
 341             || rusage.ru_nivcsw > last_rusage->ru_nivcsw) {
 342             VLOG_WARN("context switches: %ld voluntary, %ld involuntary",
 343                       rusage.ru_nvcsw - last_rusage->ru_nvcsw,
 344                       rusage.ru_nivcsw - last_rusage->ru_nivcsw);
 345         }
 346
 347         /* Care should be taken in the value chosen for logging.  Depending
 348          * on the configuration, syslog can write changes synchronously,
 349          * which can cause the coverage messages to take longer to log
 350          * than the processing delay that triggered it. */
 351         coverage_log(VLL_INFO, true);
 352     }
 353
 354     /* Update exponentially weighted moving average.  With these parameters, a
 355      * given value decays to 1% of its value in about 100 time steps.  */
 356     if (n_samples++) {
 357         mean_interval = (mean_interval * 122 + interval * 6 + 64) / 128;
 358     } else {
 359         mean_interval = interval;
 360     }
 361 }