systemd-oomd: unit testable helper functions

Helpers used by the systemd-oomd daemon to read/fill in various contexts and
make detection and kill decisions.

i.e. a simplified/less configurable version of
https://github.com/facebookincubator/oomd/tree/master/src/oomd.

4 files changed, 858 insertions, 0 deletions

diff --git a/src/oom/meson.build b/src/oom/meson.build
new file mode 100644
index 0000000000..3073809025
--- /dev/null
+++ b/src/oom/meson.build
@@ -0,0 +1,16 @@
+# SPDX-License-Identifier: LGPL-2.1+
+
+systemd_oomd_sources = files('''
+        oomd-util.c
+        oomd-util.h
+'''.split())
+
+if conf.get('ENABLE_OOMD') == 1
+        tests += [
+                [['src/oom/test-oomd-util.c',
+                  'src/oom/oomd-util.c',
+                  'src/oom/oomd-util.h'],
+                 [],
+                 []]
+        ]
+endif
diff --git a/src/oom/oomd-util.c b/src/oom/oomd-util.c
new file mode 100644
index 0000000000..c8786c745f
--- /dev/null
+++ b/src/oom/oomd-util.c
@@ -0,0 +1,386 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+
+#include <sys/xattr.h>
+#include <unistd.h>
+
+#include "fd-util.h"
+#include "format-util.h"
+#include "oomd-util.h"
+#include "parse-util.h"
+#include "path-util.h"
+#include "procfs-util.h"
+#include "signal-util.h"
+#include "sort-util.h"
+#include "stat-util.h"
+#include "stdio-util.h"
+
+DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(
+                oomd_cgroup_ctx_hash_ops,
+                char,
+                string_hash_func,
+                string_compare_func,
+                OomdCGroupContext,
+                oomd_cgroup_context_free);
+
+static int log_kill(pid_t pid, int sig, void *userdata) {
+        log_debug("oomd attempting to kill " PID_FMT " with %s", pid, signal_to_string(sig));
+        return 0;
+}
+
+static int increment_oomd_xattr(const char *path, const char *xattr, uint64_t num_procs_killed) {
+        _cleanup_free_ char *value = NULL;
+        char buf[DECIMAL_STR_MAX(uint64_t) + 1];
+        uint64_t curr_count = 0;
+        int r;
+
+        assert(path);
+        assert(xattr);
+
+        r = cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER, path, xattr, &value);
+        if (r < 0 && r != -ENODATA)
+                return r;
+
+        if (!isempty(value)) {
+                 r = safe_atou64(value, &curr_count);
+                 if (r < 0)
+                         return r;
+        }
+
+        if (curr_count > UINT64_MAX - num_procs_killed)
+                return -EOVERFLOW;
+
+        xsprintf(buf, "%"PRIu64, curr_count + num_procs_killed);
+        r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, path, xattr, buf, strlen(buf), 0);
+        if (r < 0)
+                return r;
+
+        return 0;
+}
+
+OomdCGroupContext *oomd_cgroup_context_free(OomdCGroupContext *ctx) {
+        if (!ctx)
+                return NULL;
+
+        free(ctx->path);
+        return mfree(ctx);
+}
+
+int oomd_pressure_above(Hashmap *h, usec_t duration, Set **ret) {
+        _cleanup_set_free_ Set *targets = NULL;
+        OomdCGroupContext *ctx;
+        char *key;
+        int r;
+
+        assert(h);
+        assert(ret);
+
+        targets = set_new(NULL);
+        if (!targets)
+                return -ENOMEM;
+
+        HASHMAP_FOREACH_KEY(ctx, key, h) {
+                if (ctx->memory_pressure.avg10 > ctx->mem_pressure_limit) {
+                        usec_t diff;
+
+                        if (ctx->last_hit_mem_pressure_limit == 0)
+                                ctx->last_hit_mem_pressure_limit = now(CLOCK_MONOTONIC);
+
+                        diff = now(CLOCK_MONOTONIC) - ctx->last_hit_mem_pressure_limit;
+                        if (diff >= duration) {
+                                r = set_put(targets, ctx);
+                                if (r < 0)
+                                        return -ENOMEM;
+                        }
+                } else
+                        ctx->last_hit_mem_pressure_limit = 0;
+        }
+
+        if (!set_isempty(targets)) {
+                *ret = TAKE_PTR(targets);
+                return 1;
+        }
+
+        *ret = NULL;
+        return 0;
+}
+
+bool oomd_memory_reclaim(Hashmap *h) {
+        uint64_t pgscan = 0, pgscan_of = 0, last_pgscan = 0, last_pgscan_of = 0;
+        OomdCGroupContext *ctx;
+
+        assert(h);
+
+        /* If sum of all the current pgscan values are greater than the sum of all the last_pgscan values,
+         * there was reclaim activity. Used along with pressure checks to decide whether to take action. */
+
+        HASHMAP_FOREACH(ctx, h) {
+                uint64_t sum;
+
+                sum = pgscan + ctx->pgscan;
+                if (sum < pgscan || sum < ctx->pgscan)
+                        pgscan_of++; /* count overflows */
+                pgscan = sum;
+
+                sum = last_pgscan + ctx->last_pgscan;
+                if (sum < last_pgscan || sum < ctx->last_pgscan)
+                        last_pgscan_of++; /* count overflows */
+                last_pgscan = sum;
+        }
+
+        /* overflow counts are the same, return sums comparison */
+        if (last_pgscan_of == pgscan_of)
+                return pgscan > last_pgscan;
+
+        return pgscan_of > last_pgscan_of;
+}
+
+bool oomd_swap_free_below(const OomdSystemContext *ctx, uint64_t threshold_percent) {
+        uint64_t swap_threshold;
+
+        assert(ctx);
+        assert(threshold_percent <= 100);
+
+        swap_threshold = ctx->swap_total * threshold_percent / ((uint64_t) 100);
+        return (ctx->swap_total - ctx->swap_used) < swap_threshold;
+}
+
+int oomd_sort_cgroup_contexts(Hashmap *h, oomd_compare_t compare_func, const char *prefix, OomdCGroupContext ***ret) {
+        _cleanup_free_ OomdCGroupContext **sorted = NULL;
+        OomdCGroupContext *item;
+        size_t k = 0;
+
+        assert(h);
+        assert(compare_func);
+        assert(ret);
+
+        sorted = new0(OomdCGroupContext*, hashmap_size(h));
+        if (!sorted)
+                return -ENOMEM;
+
+        HASHMAP_FOREACH(item, h) {
+                if (item->path && prefix && !path_startswith(item->path, prefix))
+                        continue;
+
+                sorted[k++] = item;
+        }
+
+        typesafe_qsort(sorted, k, compare_func);
+
+        *ret = TAKE_PTR(sorted);
+
+        assert(k <= INT_MAX);
+        return (int) k;
+}
+
+int oomd_cgroup_kill(const char *path, bool recurse, bool dry_run) {
+        _cleanup_set_free_ Set *pids_killed = NULL;
+        int r;
+
+        assert(path);
+
+        if (dry_run) {
+                _cleanup_free_ char *cg_path = NULL;
+
+                r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, path, NULL, &cg_path);
+                if (r < 0)
+                        return r;
+
+                log_debug("oomd dry-run: Would have tried to kill %s with recurse=%s", cg_path, true_false(recurse));
+                return 0;
+        }
+
+        pids_killed = set_new(NULL);
+        if (!pids_killed)
+                return -ENOMEM;
+
+        if (recurse)
+                r = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, path, SIGKILL, CGROUP_IGNORE_SELF, pids_killed, log_kill, NULL);
+        else
+                r = cg_kill(SYSTEMD_CGROUP_CONTROLLER, path, SIGKILL, CGROUP_IGNORE_SELF, pids_killed, log_kill, NULL);
+        if (r < 0)
+                return r;
+
+        r = increment_oomd_xattr(path, "user.systemd_oomd_kill", set_size(pids_killed));
+        if (r < 0)
+                log_debug_errno(r, "Failed to set user.systemd_oomd_kill on kill: %m");
+
+        return set_size(pids_killed) != 0;
+}
+
+int oomd_kill_by_pgscan(Hashmap *h, const char *prefix, bool dry_run) {
+        _cleanup_free_ OomdCGroupContext **sorted = NULL;
+        int r;
+
+        assert(h);
+
+        r = oomd_sort_cgroup_contexts(h, compare_pgscan, prefix, &sorted);
+        if (r < 0)
+                return r;
+
+        for (int i = 0; i < r; i++) {
+                if (sorted[i]->pgscan == 0)
+                        break;
+
+                r = oomd_cgroup_kill(sorted[i]->path, true, dry_run);
+                if (r > 0 || r == -ENOMEM)
+                        break;
+        }
+
+        return r;
+}
+
+int oomd_kill_by_swap_usage(Hashmap *h, bool dry_run) {
+        _cleanup_free_ OomdCGroupContext **sorted = NULL;
+        int r;
+
+        assert(h);
+
+        r = oomd_sort_cgroup_contexts(h, compare_swap_usage, NULL, &sorted);
+        if (r < 0)
+                return r;
+
+        /* Try to kill cgroups with non-zero swap usage until we either succeed in
+         * killing or we get to a cgroup with no swap usage. */
+        for (int i = 0; i < r; i++) {
+                if (sorted[i]->swap_usage == 0)
+                        break;
+
+                r = oomd_cgroup_kill(sorted[i]->path, true, dry_run);
+                if (r > 0 || r == -ENOMEM)
+                        break;
+        }
+
+        return r;
+}
+
+int oomd_cgroup_context_acquire(const char *path, OomdCGroupContext **ret) {
+        _cleanup_(oomd_cgroup_context_freep) OomdCGroupContext *ctx = NULL;
+        _cleanup_free_ char *p = NULL, *val = NULL;
+        bool is_root;
+        int r;
+
+        assert(path);
+        assert(ret);
+
+        ctx = new0(OomdCGroupContext, 1);
+        if (!ctx)
+                return -ENOMEM;
+
+        is_root = empty_or_root(path);
+
+        r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, path, "memory.pressure", &p);
+        if (r < 0)
+                return log_debug_errno(r, "Error getting cgroup memory pressure path from %s: %m", path);
+
+        r = read_resource_pressure(p, PRESSURE_TYPE_FULL, &ctx->memory_pressure);
+        if (r < 0)
+                return log_debug_errno(r, "Error parsing memory pressure from %s: %m", p);
+
+        if (is_root) {
+                r = procfs_memory_get_used(&ctx->current_memory_usage);
+                if (r < 0)
+                        return log_debug_errno(r, "Error getting memory used from procfs: %m");
+        } else {
+                r = cg_get_attribute_as_uint64(SYSTEMD_CGROUP_CONTROLLER, path, "memory.current", &ctx->current_memory_usage);
+                if (r < 0)
+                        return log_debug_errno(r, "Error getting memory.current from %s: %m", path);
+
+                r = cg_get_attribute_as_uint64(SYSTEMD_CGROUP_CONTROLLER, path, "memory.min", &ctx->memory_min);
+                if (r < 0)
+                        return log_debug_errno(r, "Error getting memory.min from %s: %m", path);
+
+                r = cg_get_attribute_as_uint64(SYSTEMD_CGROUP_CONTROLLER, path, "memory.low", &ctx->memory_low);
+                if (r < 0)
+                        return log_debug_errno(r, "Error getting memory.low from %s: %m", path);
+
+                r = cg_get_attribute_as_uint64(SYSTEMD_CGROUP_CONTROLLER, path, "memory.swap.current", &ctx->swap_usage);
+                if (r < 0)
+                        return log_debug_errno(r, "Error getting memory.swap.current from %s: %m", path);
+
+                r = cg_get_keyed_attribute(SYSTEMD_CGROUP_CONTROLLER, path, "memory.stat", STRV_MAKE("pgscan"), &val);
+                if (r < 0)
+                        return log_debug_errno(r, "Error getting pgscan from memory.stat under %s: %m", path);
+
+                r = safe_atou64(val, &ctx->pgscan);
+                if (r < 0)
+                        return log_debug_errno(r, "Error converting pgscan value to uint64_t: %m");
+        }
+
+        ctx->path = strdup(empty_to_root(path));
+        if (!ctx->path)
+                return -ENOMEM;
+
+        *ret = TAKE_PTR(ctx);
+        return 0;
+}
+
+int oomd_system_context_acquire(const char *proc_swaps_path, OomdSystemContext *ret) {
+        _cleanup_fclose_ FILE *f = NULL;
+        OomdSystemContext ctx = {};
+        int r;
+
+        assert(proc_swaps_path);
+        assert(ret);
+
+        f = fopen(proc_swaps_path, "re");
+        if (!f)
+                return -errno;
+
+        (void) fscanf(f, "%*s %*s %*s %*s %*s\n");
+
+        for (;;) {
+                uint64_t total, used;
+
+                r = fscanf(f,
+                           "%*s "          /* device/file */
+                           "%*s "          /* type of swap */
+                           "%" PRIu64 " "  /* swap size */
+                           "%" PRIu64 " "  /* used */
+                           "%*s\n",        /* priority */
+                           &total, &used);
+
+                if (r == EOF && feof(f))
+                         break;
+
+                if (r != 2) {
+                        if (ferror(f))
+                                return log_debug_errno(errno, "Error reading from %s: %m", proc_swaps_path);
+
+                        return log_debug_errno(SYNTHETIC_ERRNO(EIO),
+                                               "Failed to parse values from %s: %m", proc_swaps_path);
+                }
+
+                ctx.swap_total += total * 1024U;
+                ctx.swap_used += used * 1024U;
+        }
+
+        *ret = ctx;
+        return 0;
+}
+
+int oomd_insert_cgroup_context(Hashmap *old_h, Hashmap *new_h, const char *path) {
+        _cleanup_(oomd_cgroup_context_freep) OomdCGroupContext *curr_ctx = NULL;
+        OomdCGroupContext *old_ctx, *ctx;
+        int r;
+
+        assert(new_h);
+        assert(path);
+
+        r = oomd_cgroup_context_acquire(path, &curr_ctx);
+        if (r < 0)
+                return log_debug_errno(r, "Failed to get OomdCGroupContext for %s: %m", path);
+
+        old_ctx = hashmap_get(old_h, path);
+        if (old_ctx) {
+                curr_ctx->last_pgscan = old_ctx->pgscan;
+                curr_ctx->mem_pressure_limit = old_ctx->mem_pressure_limit;
+                curr_ctx->last_hit_mem_pressure_limit = old_ctx->last_hit_mem_pressure_limit;
+        }
+
+        ctx = TAKE_PTR(curr_ctx);
+        r = hashmap_put(new_h, ctx->path, ctx);
+        if (r < 0)
+                return r;
+
+        return 0;
+}
diff --git a/src/oom/oomd-util.h b/src/oom/oomd-util.h
new file mode 100644
index 0000000000..6d34d91cc2
--- /dev/null
+++ b/src/oom/oomd-util.h
@@ -0,0 +1,108 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+#pragma once
+
+#include <stdbool.h>
+
+#include "hashmap.h"
+#include "psi-util.h"
+
+#define GROWING_SIZE_PERCENTILE 80
+
+extern const struct hash_ops oomd_cgroup_ctx_hash_ops;
+
+typedef struct OomdCGroupContext OomdCGroupContext;
+typedef struct OomdSystemContext OomdSystemContext;
+
+typedef int (oomd_compare_t)(OomdCGroupContext * const *, OomdCGroupContext * const *);
+
+struct OomdCGroupContext {
+        char *path;
+
+        ResourcePressure memory_pressure;
+
+        uint64_t current_memory_usage;
+
+        uint64_t memory_min;
+        uint64_t memory_low;
+        uint64_t swap_usage;
+
+        uint64_t last_pgscan;
+        uint64_t pgscan;
+
+        /* These are only used by oomd_pressure_above for acting on high memory pressure. */
+        loadavg_t mem_pressure_limit;
+        usec_t last_hit_mem_pressure_limit;
+};
+
+struct OomdSystemContext {
+        uint64_t swap_total;
+        uint64_t swap_used;
+};
+
+OomdCGroupContext *oomd_cgroup_context_free(OomdCGroupContext *ctx);
+DEFINE_TRIVIAL_CLEANUP_FUNC(OomdCGroupContext*, oomd_cgroup_context_free);
+
+/* All hashmaps used with these functions are expected to be of the form
+ * key: cgroup paths -> value: OomdCGroupContext. */
+
+/* Scans all the OomdCGroupContexts in `h` and returns 1 and a set of pointers to those OomdCGroupContexts in `ret`
+ * if any of them have exceeded their supplied memory pressure limits for the `duration` length of time.
+ * `last_hit_mem_pressure_limit` is updated accordingly for each entry when the limit is exceeded, and when it returns
+ * below the limit.
+ * Returns 0 and sets `ret` to an empty set if no entries exceeded limits for `duration`.
+ * Returns -ENOMEM for allocation errors. */
+int oomd_pressure_above(Hashmap *h, usec_t duration, Set **ret);
+
+/* Sum up current OomdCGroupContexts' pgscan values and last interval's pgscan values in `h`. Returns true if the
+ * current sum is higher than the last interval's sum (there was some reclaim activity). */
+bool oomd_memory_reclaim(Hashmap *h);
+
+/* Returns true if the amount of swap free is below the percentage of swap specified by `threshold_percent`. */
+bool oomd_swap_free_below(const OomdSystemContext *ctx, uint64_t threshold_percent);
+
+static inline int compare_pgscan(OomdCGroupContext * const *c1, OomdCGroupContext * const *c2) {
+        assert(c1);
+        assert(c2);
+
+        if ((*c1)->pgscan > (*c2)->pgscan)
+                return -1;
+        else if ((*c1)->pgscan < (*c2)->pgscan)
+                return 1;
+        else
+                return 0;
+}
+
+static inline int compare_swap_usage(OomdCGroupContext * const *c1, OomdCGroupContext * const *c2) {
+        assert(c1);
+        assert(c2);
+
+        if ((*c1)->swap_usage > (*c2)->swap_usage)
+                return -1;
+        else if ((*c1)->swap_usage < (*c2)->swap_usage)
+                return 1;
+        else
+                return 0;
+}
+
+/* Get an array of OomdCGroupContexts from `h`, qsorted from largest to smallest values according to `compare_func`.
+ * If `prefix` is not NULL, only include OomdCGroupContexts whose paths start with prefix. Otherwise all paths are sorted.
+ * Returns the number of sorted items; negative on error. */
+int oomd_sort_cgroup_contexts(Hashmap *h, oomd_compare_t compare_func, const char *prefix, OomdCGroupContext ***ret);
+
+/* Returns a negative value on error, 0 if no processes were killed, or 1 if processes were killed. */
+int oomd_cgroup_kill(const char *path, bool recurse, bool dry_run);
+
+/* The following oomd_kill_by_* functions return 1 if processes were killed, or negative otherwise. */
+/* If `prefix` is supplied, only cgroups whose paths start with `prefix` are eligible candidates. Otherwise,
+ * everything in `h` is a candidate. */
+int oomd_kill_by_pgscan(Hashmap *h, const char *prefix, bool dry_run);
+int oomd_kill_by_swap_usage(Hashmap *h, bool dry_run);
+
+int oomd_cgroup_context_acquire(const char *path, OomdCGroupContext **ret);
+int oomd_system_context_acquire(const char *proc_swaps_path, OomdSystemContext *ret);
+
+/* Get the OomdCGroupContext of `path` and insert it into `new_h`. The key for the inserted context will be `path`.
+ *
+ * `old_h` is used to get data used to calculate prior interval information. `old_h` can be NULL in which case there
+ * was no prior data to reference. */
+int oomd_insert_cgroup_context(Hashmap *old_h, Hashmap *new_h, const char *path);
diff --git a/src/oom/test-oomd-util.c b/src/oom/test-oomd-util.c
new file mode 100644
index 0000000000..5503f8f2e0
--- /dev/null
+++ b/src/oom/test-oomd-util.c
@@ -0,0 +1,348 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+
+#include <unistd.h>
+
+#include "alloc-util.h"
+#include "cgroup-setup.h"
+#include "cgroup-util.h"
+#include "fileio.h"
+#include "fs-util.h"
+#include "oomd-util.h"
+#include "parse-util.h"
+#include "path-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "tests.h"
+
+static int fork_and_sleep(unsigned sleep_min) {
+        usec_t n, timeout, ts;
+
+        pid_t pid = fork();
+        assert_se(pid >= 0);
+
+        if (pid == 0) {
+                timeout = sleep_min * USEC_PER_MINUTE;
+                ts = now(CLOCK_MONOTONIC);
+                while (true) {
+                        n = now(CLOCK_MONOTONIC);
+                        if (ts + timeout < n) {
+                                log_error("Child timed out waiting to be killed");
+                                abort();
+                        }
+                        sleep(1);
+                }
+        }
+
+        return pid;
+}
+
+static void test_oomd_cgroup_kill(void) {
+        _cleanup_free_ char *cgroup_root = NULL, *cgroup = NULL;
+        int pid[2];
+
+        if (geteuid() != 0)
+                return (void) log_tests_skipped("not root");
+
+        if (cg_all_unified() <= 0)
+                return (void) log_tests_skipped("cgroups are not running in unified mode");
+
+        assert_se(cg_pid_get_path(NULL, 0, &cgroup_root) >= 0);
+
+        /* Create another cgroup below this one for the pids we forked off. We need this to be managed
+         * by the test so that pid1 doesn't delete it before we can read the xattrs. */
+        cgroup = path_join(cgroup_root, "oomdkilltest");
+        assert(cgroup);
+
+        /* If we don't have permissions to set xattrs we're likely in a userns or missing capabilities */
+        if (cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup, "user.oomd_test", "test", 4, 0) == -EPERM)
+                return (void) log_tests_skipped("no permissions to set user xattrs");
+
+        /* Do this twice to also check the increment behavior on the xattrs */
+        for (int i = 0; i < 2; i++) {
+                _cleanup_free_ char *v = NULL;
+                int r;
+
+                for (int j = 0; j < 2; j++) {
+                        pid[j] = fork_and_sleep(5);
+                        assert_se(cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup, pid[j]) >= 0);
+                }
+
+                r = oomd_cgroup_kill(cgroup, false /* recurse */, false /* dry run */);
+                if (r <= 0) {
+                        log_debug_errno(r, "Failed to kill processes under %s: %m", cgroup);
+                        abort();
+                }
+
+                /* Wait a bit since processes may take some time to be cleaned up. */
+                sleep(2);
+                assert_se(cg_is_empty(SYSTEMD_CGROUP_CONTROLLER, cgroup) == true);
+
+                assert_se(cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER, cgroup, "user.systemd_oomd_kill", &v) >= 0);
+                assert_se(memcmp(v, i == 0 ? "2" : "4", 2) == 0);
+        }
+}
+
+static void test_oomd_cgroup_context_acquire_and_insert(void) {
+        _cleanup_hashmap_free_ Hashmap *h1 = NULL, *h2 = NULL;
+        _cleanup_(oomd_cgroup_context_freep) OomdCGroupContext *ctx = NULL;
+        _cleanup_free_ char *cgroup = NULL;
+        OomdCGroupContext *c1, *c2;
+
+        if (geteuid() != 0)
+                return (void) log_tests_skipped("not root");
+
+        if (!is_pressure_supported())
+                return (void) log_tests_skipped("system does not support pressure");
+
+        if (cg_all_unified() <= 0)
+                return (void) log_tests_skipped("cgroups are not running in unified mode");
+
+        assert_se(cg_pid_get_path(NULL, 0, &cgroup) >= 0);
+
+        assert_se(oomd_cgroup_context_acquire(cgroup, &ctx) == 0);
+
+        assert_se(streq(ctx->path, cgroup));
+        assert_se(ctx->memory_pressure.avg10 == 0);
+        assert_se(ctx->memory_pressure.avg60 == 0);
+        assert_se(ctx->memory_pressure.avg300 == 0);
+        assert_se(ctx->memory_pressure.total == 0);
+        assert_se(ctx->current_memory_usage > 0);
+        assert_se(ctx->memory_min == 0);
+        assert_se(ctx->memory_low == 0);
+        assert_se(ctx->swap_usage == 0);
+        assert_se(ctx->last_pgscan == 0);
+        assert_se(ctx->pgscan == 0);
+        ctx = oomd_cgroup_context_free(ctx);
+
+        /* Test the root cgroup */
+        assert_se(oomd_cgroup_context_acquire("", &ctx) == 0);
+        assert_se(streq(ctx->path, "/"));
+        assert_se(ctx->current_memory_usage > 0);
+
+        /* Test hashmap inserts */
+        assert_se(h1 = hashmap_new(&oomd_cgroup_ctx_hash_ops));
+        assert_se(oomd_insert_cgroup_context(NULL, h1, cgroup) == 0);
+        c1 = hashmap_get(h1, cgroup);
+        assert_se(c1);
+
+         /* make sure certain values from h1 get updated in h2 */
+        c1->pgscan = 5555;
+        c1->mem_pressure_limit = 6789;
+        c1->last_hit_mem_pressure_limit = 42;
+        assert_se(h2 = hashmap_new(&oomd_cgroup_ctx_hash_ops));
+        assert_se(oomd_insert_cgroup_context(h1, h2, cgroup) == 0);
+        c1 = hashmap_get(h1, cgroup);
+        c2 = hashmap_get(h2, cgroup);
+        assert_se(c1);
+        assert_se(c2);
+        assert_se(c1 != c2);
+        assert_se(c2->last_pgscan == 5555);
+        assert_se(c2->mem_pressure_limit == 6789);
+        assert_se(c2->last_hit_mem_pressure_limit == 42);
+}
+
+static void test_oomd_system_context_acquire(void) {
+        _cleanup_(unlink_tempfilep) char path[] = "/oomdgetsysctxtestXXXXXX";
+        OomdSystemContext ctx;
+
+        if (geteuid() != 0)
+                return (void) log_tests_skipped("not root");
+
+        assert_se(mkstemp(path));
+
+        assert_se(oomd_system_context_acquire("/verylikelynonexistentpath", &ctx) == -ENOENT);
+
+        assert_se(oomd_system_context_acquire(path, &ctx) == 0);
+        assert_se(ctx.swap_total == 0);
+        assert_se(ctx.swap_used == 0);
+
+        assert_se(write_string_file(path, "some\nwords\nacross\nmultiple\nlines", WRITE_STRING_FILE_CREATE) == 0);
+        assert_se(oomd_system_context_acquire(path, &ctx) == 0);
+        assert_se(ctx.swap_total == 0);
+        assert_se(ctx.swap_used == 0);
+
+        assert_se(write_string_file(path, "Filename                                Type            Size    Used    Priority\n"
+                                          "/swapvol/swapfile                       file            18971644        0       -3\n"
+                                          "/dev/vda2                               partition       1999868 993780  -2", WRITE_STRING_FILE_CREATE) == 0);
+        assert_se(oomd_system_context_acquire(path, &ctx) == 0);
+        assert_se(ctx.swap_total == 21474828288);
+        assert_se(ctx.swap_used == 1017630720);
+}
+
+static void test_oomd_pressure_above(void) {
+        _cleanup_hashmap_free_ Hashmap *h1 = NULL, *h2 = NULL;
+        _cleanup_set_free_ Set *t1 = NULL, *t2 = NULL, *t3 = NULL;
+        OomdCGroupContext ctx[2], *c;
+        loadavg_t threshold;
+
+        assert_se(store_loadavg_fixed_point(80, 0, &threshold) == 0);
+
+        /* /herp.slice */
+        assert_se(store_loadavg_fixed_point(99, 99, &(ctx[0].memory_pressure.avg10)) == 0);
+        assert_se(store_loadavg_fixed_point(99, 99, &(ctx[0].memory_pressure.avg60)) == 0);
+        assert_se(store_loadavg_fixed_point(99, 99, &(ctx[0].memory_pressure.avg300)) == 0);
+        ctx[0].mem_pressure_limit = threshold;
+
+        /* /derp.slice */
+        assert_se(store_loadavg_fixed_point(1, 11, &(ctx[1].memory_pressure.avg10)) == 0);
+        assert_se(store_loadavg_fixed_point(1, 11, &(ctx[1].memory_pressure.avg60)) == 0);
+        assert_se(store_loadavg_fixed_point(1, 11, &(ctx[1].memory_pressure.avg300)) == 0);
+        ctx[1].mem_pressure_limit = threshold;
+
+
+        /* High memory pressure */
+        assert_se(h1 = hashmap_new(&string_hash_ops));
+        assert_se(hashmap_put(h1, "/herp.slice", &ctx[0]) >= 0);
+        assert_se(oomd_pressure_above(h1, 0 /* duration */, &t1) == 1);
+        assert_se(set_contains(t1, &ctx[0]) == true);
+        assert_se(c = hashmap_get(h1, "/herp.slice"));
+        assert_se(c->last_hit_mem_pressure_limit > 0);
+
+        /* Low memory pressure */
+        assert_se(h2 = hashmap_new(&string_hash_ops));
+        assert_se(hashmap_put(h2, "/derp.slice", &ctx[1]) >= 0);
+        assert_se(oomd_pressure_above(h2, 0 /* duration */, &t2) == 0);
+        assert_se(t2 == NULL);
+        assert_se(c = hashmap_get(h2, "/derp.slice"));
+        assert_se(c->last_hit_mem_pressure_limit == 0);
+
+        /* High memory pressure w/ multiple cgroups */
+        assert_se(hashmap_put(h1, "/derp.slice", &ctx[1]) >= 0);
+        assert_se(oomd_pressure_above(h1, 0 /* duration */, &t3) == 1);
+        assert_se(set_contains(t3, &ctx[0]) == true);
+        assert_se(set_size(t3) == 1);
+        assert_se(c = hashmap_get(h1, "/herp.slice"));
+        assert_se(c->last_hit_mem_pressure_limit > 0);
+        assert_se(c = hashmap_get(h1, "/derp.slice"));
+        assert_se(c->last_hit_mem_pressure_limit == 0);
+}
+
+static void test_oomd_memory_reclaim(void) {
+        _cleanup_hashmap_free_ Hashmap *h1 = NULL;
+        char **paths = STRV_MAKE("/0.slice",
+                                 "/1.slice",
+                                 "/2.slice",
+                                 "/3.slice",
+                                 "/4.slice");
+
+        OomdCGroupContext ctx[5] = {
+                { .path = paths[0],
+                  .last_pgscan = 100,
+                  .pgscan = 100 },
+                { .path = paths[1],
+                  .last_pgscan = 100,
+                  .pgscan = 100 },
+                { .path = paths[2],
+                  .last_pgscan = 77,
+                  .pgscan = 33 },
+                { .path = paths[3],
+                  .last_pgscan = UINT64_MAX,
+                  .pgscan = 100 },
+                { .path = paths[4],
+                  .last_pgscan = 100,
+                  .pgscan = UINT64_MAX },
+        };
+
+        assert_se(h1 = hashmap_new(&string_hash_ops));
+        assert_se(hashmap_put(h1, paths[0], &ctx[0]) >= 0);
+        assert_se(hashmap_put(h1, paths[1], &ctx[1]) >= 0);
+        assert_se(oomd_memory_reclaim(h1) == false);
+
+        assert_se(hashmap_put(h1, paths[2], &ctx[2]) >= 0);
+        assert_se(oomd_memory_reclaim(h1) == false);
+
+        assert_se(hashmap_put(h1, paths[4], &ctx[4]) >= 0);
+        assert_se(oomd_memory_reclaim(h1) == true);
+
+        assert_se(hashmap_put(h1, paths[3], &ctx[3]) >= 0);
+        assert_se(oomd_memory_reclaim(h1) == false);
+}
+
+static void test_oomd_swap_free_below(void) {
+        OomdSystemContext ctx = (OomdSystemContext) {
+                .swap_total = 20971512 * 1024U,
+                .swap_used = 20971440 * 1024U,
+        };
+        assert_se(oomd_swap_free_below(&ctx, 20) == true);
+
+        ctx = (OomdSystemContext) {
+                .swap_total = 20971512 * 1024U,
+                .swap_used = 3310136 * 1024U,
+        };
+        assert_se(oomd_swap_free_below(&ctx, 20) == false);
+}
+
+static void test_oomd_sort_cgroups(void) {
+        _cleanup_hashmap_free_ Hashmap *h = NULL;
+        _cleanup_free_ OomdCGroupContext **sorted_cgroups;
+        char **paths = STRV_MAKE("/herp.slice",
+                                 "/herp.slice/derp.scope",
+                                 "/herp.slice/derp.scope/sheep.service",
+                                 "/zupa.slice");
+
+        OomdCGroupContext ctx[4] = {
+                { .path = paths[0],
+                  .swap_usage = 20,
+                  .pgscan = 60 },
+                { .path = paths[1],
+                  .swap_usage = 60,
+                  .pgscan = 40 },
+                { .path = paths[2],
+                  .swap_usage = 40,
+                  .pgscan = 20 },
+                { .path = paths[3],
+                  .swap_usage = 10,
+                  .pgscan = 80 },
+        };
+
+        assert_se(h = hashmap_new(&string_hash_ops));
+
+        assert_se(hashmap_put(h, "/herp.slice", &ctx[0]) >= 0);
+        assert_se(hashmap_put(h, "/herp.slice/derp.scope", &ctx[1]) >= 0);
+        assert_se(hashmap_put(h, "/herp.slice/derp.scope/sheep.service", &ctx[2]) >= 0);
+        assert_se(hashmap_put(h, "/zupa.slice", &ctx[3]) >= 0);
+
+        assert_se(oomd_sort_cgroup_contexts(h, compare_swap_usage, NULL, &sorted_cgroups) == 4);
+        assert_se(sorted_cgroups[0] == &ctx[1]);
+        assert_se(sorted_cgroups[1] == &ctx[2]);
+        assert_se(sorted_cgroups[2] == &ctx[0]);
+        assert_se(sorted_cgroups[3] == &ctx[3]);
+        sorted_cgroups = mfree(sorted_cgroups);
+
+        assert_se(oomd_sort_cgroup_contexts(h, compare_pgscan, NULL, &sorted_cgroups) == 4);
+        assert_se(sorted_cgroups[0] == &ctx[3]);
+        assert_se(sorted_cgroups[1] == &ctx[0]);
+        assert_se(sorted_cgroups[2] == &ctx[1]);
+        assert_se(sorted_cgroups[3] == &ctx[2]);
+        sorted_cgroups = mfree(sorted_cgroups);
+
+        assert_se(oomd_sort_cgroup_contexts(h, compare_pgscan, "/herp.slice/derp.scope", &sorted_cgroups) == 2);
+        assert_se(sorted_cgroups[0] == &ctx[1]);
+        assert_se(sorted_cgroups[1] == &ctx[2]);
+        assert_se(sorted_cgroups[2] == 0);
+        assert_se(sorted_cgroups[3] == 0);
+        sorted_cgroups = mfree(sorted_cgroups);
+}
+
+int main(void) {
+        int r;
+
+        test_setup_logging(LOG_DEBUG);
+
+        test_oomd_system_context_acquire();
+        test_oomd_pressure_above();
+        test_oomd_memory_reclaim();
+        test_oomd_swap_free_below();
+        test_oomd_sort_cgroups();
+
+        /* The following tests operate on live cgroups */
+
+        r = enter_cgroup_root(NULL);
+        if (r < 0)
+                return log_tests_skipped_errno(r, "failed to enter a test cgroup scope");
+
+        test_oomd_cgroup_kill();
+        test_oomd_cgroup_context_acquire_and_insert();
+
+        return 0;
+}