chiark / gitweb /
journald: implement sophisticated rate limiting
[elogind.git] / src / journal / journal-rate-limit.c
diff --git a/src/journal/journal-rate-limit.c b/src/journal/journal-rate-limit.c
new file mode 100644 (file)
index 0000000..f69ab27
--- /dev/null
@@ -0,0 +1,273 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd 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 systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <string.h>
+#include <errno.h>
+
+#include "journal-rate-limit.h"
+#include "list.h"
+#include "util.h"
+#include "hashmap.h"
+
+#define POOLS_MAX 5
+#define BUCKETS_MAX 127
+#define GROUPS_MAX 2047
+
+static const int priority_map[] = {
+        [LOG_EMERG]   = 0,
+        [LOG_ALERT]   = 0,
+        [LOG_CRIT]    = 0,
+        [LOG_ERR]     = 1,
+        [LOG_WARNING] = 2,
+        [LOG_NOTICE]  = 3,
+        [LOG_INFO]    = 3,
+        [LOG_DEBUG]   = 4
+};
+
+typedef struct JournalRateLimitPool JournalRateLimitPool;
+typedef struct JournalRateLimitGroup JournalRateLimitGroup;
+
+struct JournalRateLimitPool {
+        usec_t begin;
+        unsigned num;
+        unsigned suppressed;
+};
+
+struct JournalRateLimitGroup {
+        JournalRateLimit *parent;
+
+        char *id;
+        JournalRateLimitPool pools[POOLS_MAX];
+        unsigned hash;
+
+        LIST_FIELDS(JournalRateLimitGroup, bucket);
+        LIST_FIELDS(JournalRateLimitGroup, lru);
+};
+
+struct JournalRateLimit {
+        usec_t interval;
+        unsigned burst;
+
+        JournalRateLimitGroup* buckets[BUCKETS_MAX];
+        JournalRateLimitGroup *lru, *lru_tail;
+
+        unsigned n_groups;
+};
+
+JournalRateLimit *journal_rate_limit_new(usec_t interval, unsigned burst) {
+        JournalRateLimit *r;
+
+        assert(interval > 0 || burst == 0);
+
+        r = new0(JournalRateLimit, 1);
+        if (!r)
+                return NULL;
+
+        r->interval = interval;
+        r->burst = burst;
+
+        return r;
+}
+
+static void journal_rate_limit_group_free(JournalRateLimitGroup *g) {
+        assert(g);
+
+        if (g->parent) {
+                assert(g->parent->n_groups > 0);
+
+                if (g->parent->lru_tail == g)
+                        g->parent->lru_tail = g->lru_prev;
+
+                LIST_REMOVE(JournalRateLimitGroup, lru, g->parent->lru, g);
+                LIST_REMOVE(JournalRateLimitGroup, bucket, g->parent->buckets[g->hash % BUCKETS_MAX], g);
+
+                g->parent->n_groups --;
+        }
+
+        free(g->id);
+        free(g);
+}
+
+void journal_rate_limit_free(JournalRateLimit *r) {
+        assert(r);
+
+        while (r->lru)
+                journal_rate_limit_group_free(r->lru);
+}
+
+static bool journal_rate_limit_group_expired(JournalRateLimitGroup *g, usec_t ts) {
+        unsigned i;
+
+        assert(g);
+
+        for (i = 0; i < POOLS_MAX; i++)
+                if (g->pools[i].begin + g->parent->interval >= ts)
+                        return false;
+
+        return true;
+}
+
+static void journal_rate_limit_vacuum(JournalRateLimit *r, usec_t ts) {
+        assert(r);
+
+        /* Makes room for at least one new item, but drop all
+         * expored items too. */
+
+        while (r->n_groups >= GROUPS_MAX ||
+               (r->lru_tail && journal_rate_limit_group_expired(r->lru_tail, ts)))
+                journal_rate_limit_group_free(r->lru_tail);
+}
+
+static JournalRateLimitGroup* journal_rate_limit_group_new(JournalRateLimit *r, const char *id, usec_t ts) {
+        JournalRateLimitGroup *g;
+
+        assert(r);
+        assert(id);
+
+        g = new0(JournalRateLimitGroup, 1);
+        if (!g)
+                return NULL;
+
+        g->id = strdup(id);
+        if (!g->id)
+                goto fail;
+
+        g->hash = string_hash_func(g->id);
+
+        journal_rate_limit_vacuum(r, ts);
+
+        LIST_PREPEND(JournalRateLimitGroup, bucket, r->buckets[g->hash % BUCKETS_MAX], g);
+        LIST_PREPEND(JournalRateLimitGroup, lru, r->lru, g);
+        if (!g->lru_next)
+                r->lru_tail = g;
+        r->n_groups ++;
+
+        g->parent = r;
+        return g;
+
+fail:
+        journal_rate_limit_group_free(g);
+        return NULL;
+}
+
+static uint64_t u64log2(uint64_t n) {
+        unsigned r;
+
+        if (n <= 1)
+                return 0;
+
+        r = 0;
+        for (;;) {
+                n = n >> 1;
+                if (!n)
+                        return r;
+                r++;
+        }
+}
+
+static unsigned burst_modulate(unsigned burst, uint64_t available) {
+        unsigned k;
+
+        /* Modulates the burst rate a bit with the amount of available
+         * disk space */
+
+        k = u64log2(available);
+
+        /* 1MB */
+        if (k <= 20)
+                return burst;
+
+        burst = (burst * (k-20)) / 4;
+
+        /*
+         * Example:
+         *
+         *      <= 1MB = rate * 1
+         *        16MB = rate * 2
+         *       256MB = rate * 3
+         *         4GB = rate * 4
+         *        64GB = rate * 5
+         *         1TB = rate * 6
+         */
+
+        return burst;
+}
+
+int journal_rate_limit_test(JournalRateLimit *r, const char *id, int priority, uint64_t available) {
+        unsigned h;
+        JournalRateLimitGroup *g;
+        JournalRateLimitPool *p;
+        unsigned burst;
+        usec_t ts;
+
+        assert(id);
+
+        if (!r)
+                return 1;
+
+        if (r->interval == 0 || r->burst == 0)
+                return 1;
+
+        burst = burst_modulate(r->burst, available);
+
+        ts = now(CLOCK_MONOTONIC);
+
+        h = string_hash_func(id);
+        g = r->buckets[h % BUCKETS_MAX];
+
+        LIST_FOREACH(bucket, g, g)
+                if (streq(g->id, id))
+                        break;
+
+        if (!g) {
+                g = journal_rate_limit_group_new(r, id, ts);
+                if (!g)
+                        return -ENOMEM;
+        }
+
+        p = &g->pools[priority_map[priority]];
+
+        if (p->begin <= 0) {
+                p->suppressed = 0;
+                p->num = 1;
+                p->begin = ts;
+                return 1;
+        }
+
+        if (p->begin + r->interval < ts) {
+                unsigned s;
+
+                s = p->suppressed;
+                p->suppressed = 0;
+                p->num = 1;
+                p->begin = ts;
+
+                return 1 + s;
+        }
+
+        if (p->num <= burst) {
+                p->num++;
+                return 1;
+        }
+
+        p->suppressed++;
+        return 0;
+}