src/bin/lttng-sessiond/health.c

   1 /*
   2  * Copyright (C) 2012 - David Goulet <dgoulet@efficios.com>
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License, version 2 only, as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 51
  15  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _GNU_SOURCE
  19 #include <assert.h>
  20 #include <inttypes.h>
  21 #include <stdio.h>
  22 #include <stdlib.h>
  23 #include <time.h>
  24
  25 #include <common/defaults.h>
  26 #include <common/error.h>
  27
  28 #include "health.h"
  29
  30 static const struct timespec time_delta = {
  31         .tv_sec = DEFAULT_HEALTH_CHECK_DELTA_S,
  32         .tv_nsec = DEFAULT_HEALTH_CHECK_DELTA_NS,
  33 };
  34
  35 /* Define TLS health state. */
  36 DEFINE_URCU_TLS(struct health_state, health_state);
  37
  38 /*
  39  * It ensures that TLS memory used for the node and its container structure
  40  * don't get reclaimed after the TLS owner thread exits until we have finished
  41  * using it.
  42  */
  43 static pthread_mutex_t health_mutex = PTHREAD_MUTEX_INITIALIZER;
  44
  45 static struct health_tls_state_list health_state_list = {
  46         .head = CDS_LIST_HEAD_INIT(health_state_list.head),
  47 };
  48
  49 /*
  50  * This keeps track of the error state for unregistered thread. A thread
  51  * reporting a health error, normally unregisters and quits. This makes the TLS
  52  * health state not available to the health_check_state() call so on unregister
  53  * we update this global error array so we can keep track of which thread was
  54  * on error if the TLS health state has been removed.
  55  */
  56 static enum health_flags global_error_state[HEALTH_NUM_TYPE];
  57
  58 /*
  59  * Lock health state global list mutex.
  60  */
  61 static void state_lock(void)
  62 {
  63         pthread_mutex_lock(&health_mutex);
  64 }
  65
  66 /*
  67  * Unlock health state global list mutex.
  68  */
  69 static void state_unlock(void)
  70 {
  71         pthread_mutex_unlock(&health_mutex);
  72 }
  73
  74 /*
  75  * Set time difference in res from time_a and time_b.
  76  */
  77 static void time_diff(const struct timespec *time_a,
  78                 const struct timespec *time_b, struct timespec *res)
  79 {
  80         if (time_a->tv_nsec - time_b->tv_nsec < 0) {
  81                 res->tv_sec = time_a->tv_sec - time_b->tv_sec - 1;
  82                 res->tv_nsec = 1000000000L + time_a->tv_sec - time_b->tv_sec;
  83         } else {
  84                 res->tv_sec = time_a->tv_sec - time_b->tv_sec;
  85                 res->tv_nsec = time_a->tv_nsec - time_b->tv_nsec;
  86         }
  87 }
  88
  89 /*
  90  * Return true if time_a - time_b > diff, else false.
  91  */
  92 static int time_diff_gt(const struct timespec *time_a,
  93                 const struct timespec *time_b, const struct timespec *diff)
  94 {
  95         struct timespec res;
  96
  97         time_diff(time_a, time_b, &res);
  98         time_diff(&res, diff, &res);
  99
 100         if (res.tv_sec > 0) {
 101                 return 1;
 102         } else if (res.tv_sec == 0 && res.tv_nsec > 0) {
 103                 return 1;
 104         }
 105
 106         return 0;
 107 }
 108
 109 /*
 110  * Validate health state. Checks for the error flag or health conditions.
 111  *
 112  * Return 0 if health is bad or else 1.
 113  */
 114 static int validate_state(struct health_state *state)
 115 {
 116         int retval = 1, ret;
 117         unsigned long current, last;
 118         struct timespec current_time;
 119
 120         assert(state);
 121
 122         last = state->last;
 123         current = uatomic_read(&state->current);
 124
 125         ret = clock_gettime(CLOCK_MONOTONIC, &current_time);
 126         if (ret < 0) {
 127                 PERROR("Error reading time\n");
 128                 /* error */
 129                 retval = 0;
 130                 goto end;
 131         }
 132
 133         /*
 134          * Thread is in bad health if flag HEALTH_ERROR is set. It is also in bad
 135          * health if, after the delta delay has passed, its the progress counter
 136          * has not moved and it has NOT been waiting for a poll() call.
 137          */
 138         if (uatomic_read(&state->flags) & HEALTH_ERROR) {
 139                 retval = 0;
 140                 goto end;
 141         }
 142
 143         /*
 144          * Initial condition need to update the last counter and sample time, but
 145          * should not check health in this initial case, because we don't know how
 146          * much time has passed.
 147          */
 148         if (state->last_time.tv_sec == 0 && state->last_time.tv_nsec == 0) {
 149                 /* update last counter and last sample time */
 150                 state->last = current;
 151                 memcpy(&state->last_time, &current_time, sizeof(current_time));
 152         } else {
 153                 if (time_diff_gt(&current_time, &state->last_time, &time_delta)) {
 154                         if (current == last && !HEALTH_IS_IN_POLL(current)) {
 155                                 /* error */
 156                                 retval = 0;
 157                         }
 158                         /* update last counter and last sample time */
 159                         state->last = current;
 160                         memcpy(&state->last_time, &current_time, sizeof(current_time));
 161
 162                         /* On error, stop right now and notify caller. */
 163                         if (retval == 0) {
 164                                 goto end;
 165                         }
 166                 }
 167         }
 168
 169 end:
 170         DBG("Health state current %lu, last %lu, ret %d",
 171                         current, last, ret);
 172         return retval;
 173 }
 174
 175 /*
 176  * Check health of a specific health type. Note that if a thread has not yet
 177  * initialize its health subsystem or has quit, it's considered in a good
 178  * state.
 179  *
 180  * Return 0 if health is bad or else 1.
 181  */
 182 int health_check_state(enum health_type type)
 183 {
 184         int retval = 1;
 185         struct health_state *state;
 186
 187         assert(type < HEALTH_NUM_TYPE);
 188
 189         state_lock();
 190
 191         cds_list_for_each_entry(state, &health_state_list.head, node) {
 192                 int ret;
 193
 194                 if (state->type != type) {
 195                         continue;
 196                 }
 197
 198                 ret = validate_state(state);
 199                 if (!ret) {
 200                         retval = 0;
 201                         goto end;
 202                 }
 203         }
 204
 205         /* Check the global state since some state might not be visible anymore. */
 206         if (global_error_state[type] & HEALTH_ERROR) {
 207                 retval = 0;
 208         }
 209
 210 end:
 211         state_unlock();
 212
 213         DBG("Health check for type %d is %s", (int) type,
 214                         (retval == 0) ? "BAD" : "GOOD");
 215         return retval;
 216 }
 217
 218 /*
 219  * Init health state.
 220  */
 221 void health_register(enum health_type type)
 222 {
 223         assert(type < HEALTH_NUM_TYPE);
 224
 225         /* Init TLS state. */
 226         uatomic_set(&URCU_TLS(health_state).last, 0);
 227         uatomic_set(&URCU_TLS(health_state).last_time.tv_sec, 0);
 228         uatomic_set(&URCU_TLS(health_state).last_time.tv_nsec, 0);
 229         uatomic_set(&URCU_TLS(health_state).current, 0);
 230         uatomic_set(&URCU_TLS(health_state).flags, 0);
 231         uatomic_set(&URCU_TLS(health_state).type, type);
 232
 233         /* Add it to the global TLS state list. */
 234         state_lock();
 235         cds_list_add(&URCU_TLS(health_state).node, &health_state_list.head);
 236         state_unlock();
 237 }
 238
 239 /*
 240  * Remove node from global list.
 241  */
 242 void health_unregister(void)
 243 {
 244         state_lock();
 245         /*
 246          * On error, set the global_error_state since we are about to remove
 247          * the node from the global list.
 248          */
 249         if (uatomic_read(&URCU_TLS(health_state).flags) & HEALTH_ERROR) {
 250                 uatomic_set(&global_error_state[URCU_TLS(health_state).type],
 251                                 HEALTH_ERROR);
 252         }
 253         cds_list_del(&URCU_TLS(health_state).node);
 254         state_unlock();
 255 }