oom: implement avoid/omit xattr support

There may be situations where a cgroup should be protected from killing
or deprioritized as a candidate. In FB oomd xattrs are used to bias oomd
away from supervisor cgroups and towards worker cgroups in container
tasks. On desktops this can be used to protect important units with
unpredictable resource consumption.

The patch allows systemd-oomd to understand 2 xattrs:
"user.oomd_avoid" and "user.oomd_omit". If systemd-oomd sees these
xattrs set to 1 on a candidate cgroup (i.e. while attempting to kill something)
AND the cgroup is owned by root, it will either deprioritize the cgroup as
a candidate (avoid) or remove it completely as a candidate (omit).

Usage is restricted to root owned cgroups to prevent situations where an
unprivileged user can set their own cgroups lower in the kill priority than
another user's (and prevent them from omitting their units from
systemd-oomd killing).

src/basic/cgroup-util.c

@@ -635,6 +635,20 @@ int cg_get_xattr_malloc(const char *controller, const char *path, const char *na
         return r;
 }
 
+int cg_get_xattr_bool(const char *controller, const char *path, const char *name) {
+        _cleanup_free_ char *val = NULL;
+        int r;
+
+        assert(path);
+        assert(name);
+
+        r = cg_get_xattr_malloc(controller, path, name, &val);
+        if (r < 0)
+                return r;
+
+        return parse_boolean(val);
+}
+
 int cg_remove_xattr(const char *controller, const char *path, const char *name) {
         _cleanup_free_ char *fs = NULL;
         int r;
@@ -1703,6 +1717,25 @@ int cg_get_attribute_as_bool(const char *controller, const char *path, const cha
         return 0;
 }
 
+int cg_get_owner(const char *controller, const char *path, uid_t *ret_uid) {
+        _cleanup_free_ char *f = NULL;
+        struct stat stats;
+        int r;
+
+        assert(ret_uid);
+
+        r = cg_get_path(controller, path, NULL, &f);
+        if (r < 0)
+                return r;
+
+        r = stat(f, &stats);
+        if (r < 0)
+                return -errno;
+
+        *ret_uid = stats.st_uid;
+        return 0;
+}
+
 int cg_get_keyed_attribute_full(
                 const char *controller,
                 const char *path,

src/basic/cgroup-util.h

@@ -212,10 +212,13 @@ int cg_get_attribute_as_uint64(const char *controller, const char *path, const c
 int cg_get_attribute_as_bool(const char *controller, const char *path, const char *attribute, bool *ret);
 
 int cg_set_access(const char *controller, const char *path, uid_t uid, gid_t gid);
+int cg_get_owner(const char *controller, const char *path, uid_t *ret_uid);
 
 int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags);
 int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size);
 int cg_get_xattr_malloc(const char *controller, const char *path, const char *name, char **ret);
+/* Returns negative on error, and 0 or 1 on success for the bool value */
+int cg_get_xattr_bool(const char *controller, const char *path, const char *name);
 int cg_remove_xattr(const char *controller, const char *path, const char *name);
 
 int cg_install_release_agent(const char *controller, const char *agent);

src/oom/oomd-util.c

@@ -3,7 +3,6 @@
 #include <sys/xattr.h>
 #include <unistd.h>
 
-#include "cgroup-util.h"
 #include "fd-util.h"
 #include "format-util.h"
 #include "oomd-util.h"
@@ -159,7 +158,8 @@ int oomd_sort_cgroup_contexts(Hashmap *h, oomd_compare_t compare_func, const cha
                 return -ENOMEM;
 
         HASHMAP_FOREACH(item, h) {
-                if (item->path && prefix && !path_startswith(item->path, prefix))
+                /* Skip over cgroups that are not valid candidates or are explicitly marked for omission */
+                if ((item->path && prefix && !path_startswith(item->path, prefix)) || item->preference == MANAGED_OOM_PREFERENCE_OMIT)
                         continue;
 
                 sorted[k++] = item;
@@ -219,9 +219,10 @@ int oomd_kill_by_pgscan(Hashmap *h, const char *prefix, bool dry_run) {
                 return r;
 
         for (int i = 0; i < r; i++) {
-                /* Skip cgroups with no reclaim and memory usage; it won't alleviate pressure */
+                /* Skip cgroups with no reclaim and memory usage; it won't alleviate pressure. */
+                /* Don't break since there might be "avoid" cgroups at the end. */
                 if (sorted[i]->pgscan == 0 && sorted[i]->current_memory_usage == 0)
-                        break;
+                        continue;
 
                 r = oomd_cgroup_kill(sorted[i]->path, true, dry_run);
                 if (r > 0 || r == -ENOMEM)
@@ -244,8 +245,10 @@ int oomd_kill_by_swap_usage(Hashmap *h, bool dry_run) {
         /* 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++) {
+                /* Skip over cgroups with no resource usage. Don't break since there might be "avoid"
+                 * cgroups at the end. */
                 if (sorted[i]->swap_usage == 0)
-                        break;
+                        continue;
 
                 r = oomd_cgroup_kill(sorted[i]->path, true, dry_run);
                 if (r > 0 || r == -ENOMEM)
@@ -259,6 +262,7 @@ 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;
+        uid_t uid;
         int r;
 
         assert(path);
@@ -269,6 +273,7 @@ int oomd_cgroup_context_acquire(const char *path, OomdCGroupContext **ret) {
                 return -ENOMEM;
 
         is_root = empty_or_root(path);
+        ctx->preference = MANAGED_OOM_PREFERENCE_NONE;
 
         r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, path, "memory.pressure", &p);
         if (r < 0)
@@ -278,6 +283,23 @@ int oomd_cgroup_context_acquire(const char *path, OomdCGroupContext **ret) {
         if (r < 0)
                 return log_debug_errno(r, "Error parsing memory pressure from %s: %m", p);
 
+        r = cg_get_owner(SYSTEMD_CGROUP_CONTROLLER, path, &uid);
+        if (r < 0)
+                log_debug_errno(r, "Failed to get owner/group from %s: %m", path);
+        else if (uid == 0) {
+                /* Ignore most errors when reading the xattr since it is usually unset and cgroup xattrs are only used
+                 * as an optional feature of systemd-oomd (and the system might not even support them). */
+                r = cg_get_xattr_bool(SYSTEMD_CGROUP_CONTROLLER, path, "user.oomd_avoid");
+                if (r == -ENOMEM)
+                        return r;
+                ctx->preference = r == 1 ? MANAGED_OOM_PREFERENCE_AVOID : ctx->preference;
+
+                r = cg_get_xattr_bool(SYSTEMD_CGROUP_CONTROLLER, path, "user.oomd_omit");
+                if (r == -ENOMEM)
+                        return r;
+                ctx->preference = r == 1 ? MANAGED_OOM_PREFERENCE_OMIT : ctx->preference;
+        }
+
         if (is_root) {
                 r = procfs_memory_get_used(&ctx->current_memory_usage);
                 if (r < 0)

src/oom/oomd-util.h

@@ -3,6 +3,7 @@
 
 #include <stdbool.h>
 
+#include "cgroup-util.h"
 #include "hashmap.h"
 #include "psi-util.h"
 
@@ -29,6 +30,8 @@ struct OomdCGroupContext {
         uint64_t last_pgscan;
         uint64_t pgscan;
 
+        ManagedOOMPreference preference;
+
         /* These are only used by oomd_pressure_above for acting on high memory pressure. */
         loadavg_t mem_pressure_limit;
         usec_t mem_pressure_duration_usec;
@@ -61,12 +64,18 @@ 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);
 
+/* The compare functions will sort from largest to smallest, putting all the contexts with "avoid" at the end
+ * (after the smallest values). */
 static inline int compare_pgscan_and_memory_usage(OomdCGroupContext * const *c1, OomdCGroupContext * const *c2) {
         int r;
 
         assert(c1);
         assert(c2);
 
+        r = CMP((*c1)->preference, (*c2)->preference);
+        if (r != 0)
+                return r;
+
         r = CMP((*c2)->pgscan, (*c1)->pgscan);
         if (r != 0)
                 return r;
@@ -75,9 +84,15 @@ static inline int compare_pgscan_and_memory_usage(OomdCGroupContext * const *c1,
 }
 
 static inline int compare_swap_usage(OomdCGroupContext * const *c1, OomdCGroupContext * const *c2) {
+        int r;
+
         assert(c1);
         assert(c2);
 
+        r = CMP((*c1)->preference, (*c2)->preference);
+        if (r != 0)
+                return r;
+
         return CMP((*c2)->swap_usage, (*c1)->swap_usage);
 }
 

src/oom/test-oomd-util.c

@@ -89,6 +89,8 @@ static void test_oomd_cgroup_context_acquire_and_insert(void) {
         _cleanup_(oomd_cgroup_context_freep) OomdCGroupContext *ctx = NULL;
         _cleanup_free_ char *cgroup = NULL;
         OomdCGroupContext *c1, *c2;
+        bool test_xattrs;
+        int r;
 
         if (geteuid() != 0)
                 return (void) log_tests_skipped("not root");
@@ -101,6 +103,16 @@ static void test_oomd_cgroup_context_acquire_and_insert(void) {
 
         assert_se(cg_pid_get_path(NULL, 0, &cgroup) >= 0);
 
+        /* If we don't have permissions to set xattrs we're likely in a userns or missing capabilities
+         * so skip the xattr portions of the test. */
+        r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup, "user.oomd_test", "1", 1, 0);
+        test_xattrs = !ERRNO_IS_PRIVILEGE(r) && !ERRNO_IS_NOT_SUPPORTED(r);
+
+        if (test_xattrs) {
+                assert_se(cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup, "user.oomd_omit", "1", 1, 0) >= 0);
+                assert_se(cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup, "user.oomd_avoid", "1", 1, 0) >= 0);
+        }
+
         assert_se(oomd_cgroup_context_acquire(cgroup, &ctx) == 0);
 
         assert_se(streq(ctx->path, cgroup));
@@ -110,12 +122,28 @@ static void test_oomd_cgroup_context_acquire_and_insert(void) {
         assert_se(ctx->swap_usage == 0);
         assert_se(ctx->last_pgscan == 0);
         assert_se(ctx->pgscan == 0);
+        /* omit takes precedence over avoid when both are set to true */
+        if (test_xattrs)
+                assert_se(ctx->preference == MANAGED_OOM_PREFERENCE_OMIT);
+        else
+                assert_se(ctx->preference == MANAGED_OOM_PREFERENCE_NONE);
+        ctx = oomd_cgroup_context_free(ctx);
+
+        /* also check when only avoid is set to true */
+        if (test_xattrs) {
+                assert_se(cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup, "user.oomd_omit", "0", 1, 0) >= 0);
+                assert_se(cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup, "user.oomd_avoid", "1", 1, 0) >= 0);
+        }
+        assert_se(oomd_cgroup_context_acquire(cgroup, &ctx) == 0);
+        if (test_xattrs)
+                assert_se(ctx->preference == MANAGED_OOM_PREFERENCE_AVOID);
         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);
+        assert_se(ctx->preference == MANAGED_OOM_PREFERENCE_NONE);
 
         /* Test hashmap inserts */
         assert_se(h1 = hashmap_new(&oomd_cgroup_ctx_hash_ops));
@@ -137,6 +165,14 @@ static void test_oomd_cgroup_context_acquire_and_insert(void) {
         assert_se(c2->last_pgscan == 5555);
         assert_se(c2->mem_pressure_limit == 6789);
         assert_se(c2->last_hit_mem_pressure_limit == 42);
+
+        /* Assert that avoid/omit are not set if the cgroup is not owned by root */
+        if (test_xattrs) {
+                ctx = oomd_cgroup_context_free(ctx);
+                assert_se(cg_set_access(SYSTEMD_CGROUP_CONTROLLER, cgroup, 65534, 0) >= 0);
+                assert_se(oomd_cgroup_context_acquire(cgroup, &ctx) == 0);
+                assert_se(ctx->preference == MANAGED_OOM_PREFERENCE_NONE);
+        }
 }
 
 static void test_oomd_system_context_acquire(void) {
@@ -287,9 +323,11 @@ static void test_oomd_sort_cgroups(void) {
         char **paths = STRV_MAKE("/herp.slice",
                                  "/herp.slice/derp.scope",
                                  "/herp.slice/derp.scope/sheep.service",
-                                 "/zupa.slice");
+                                 "/zupa.slice",
+                                 "/omitted.slice",
+                                 "/avoid.slice");
 
-        OomdCGroupContext ctx[4] = {
+        OomdCGroupContext ctx[6] = {
                 { .path = paths[0],
                   .swap_usage = 20,
                   .pgscan = 60,
@@ -306,6 +344,14 @@ static void test_oomd_sort_cgroups(void) {
                   .swap_usage = 10,
                   .pgscan = 80,
                   .current_memory_usage = 10 },
+                { .path = paths[4],
+                  .swap_usage = 90,
+                  .pgscan = 100,
+                  .preference = MANAGED_OOM_PREFERENCE_OMIT },
+                { .path = paths[5],
+                  .swap_usage = 99,
+                  .pgscan = 200,
+                  .preference = MANAGED_OOM_PREFERENCE_AVOID },
         };
 
         assert_se(h = hashmap_new(&string_hash_ops));
@@ -314,19 +360,23 @@ static void test_oomd_sort_cgroups(void) {
         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(hashmap_put(h, "/omitted.slice", &ctx[4]) >= 0);
+        assert_se(hashmap_put(h, "/avoid.slice", &ctx[5]) >= 0);
 
-        assert_se(oomd_sort_cgroup_contexts(h, compare_swap_usage, NULL, &sorted_cgroups) == 4);
+        assert_se(oomd_sort_cgroup_contexts(h, compare_swap_usage, NULL, &sorted_cgroups) == 5);
         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]);
+        assert_se(sorted_cgroups[4] == &ctx[5]);
         sorted_cgroups = mfree(sorted_cgroups);
 
-        assert_se(oomd_sort_cgroup_contexts(h, compare_pgscan_and_memory_usage, NULL, &sorted_cgroups) == 4);
+        assert_se(oomd_sort_cgroup_contexts(h, compare_pgscan_and_memory_usage, NULL, &sorted_cgroups) == 5);
         assert_se(sorted_cgroups[0] == &ctx[3]);
         assert_se(sorted_cgroups[1] == &ctx[0]);
         assert_se(sorted_cgroups[2] == &ctx[2]);
         assert_se(sorted_cgroups[3] == &ctx[1]);
+        assert_se(sorted_cgroups[4] == &ctx[5]);
         sorted_cgroups = mfree(sorted_cgroups);
 
         assert_se(oomd_sort_cgroup_contexts(h, compare_pgscan_and_memory_usage, "/herp.slice/derp.scope", &sorted_cgroups) == 2);
@@ -334,6 +384,8 @@ static void test_oomd_sort_cgroups(void) {
         assert_se(sorted_cgroups[1] == &ctx[1]);
         assert_se(sorted_cgroups[2] == 0);
         assert_se(sorted_cgroups[3] == 0);
+        assert_se(sorted_cgroups[4] == 0);
+        assert_se(sorted_cgroups[5] == 0);
         sorted_cgroups = mfree(sorted_cgroups);
 }
 

test/test-functions

@@ -124,6 +124,7 @@ BASICTOOLS=(
     rmdir
     sed
     seq
+    setfattr
     setfont
     setsid
     sfdisk

test/units/testsuite-56-testmunch.service

@@ -0,0 +1,7 @@
+[Unit]
+Description=Create some memory pressure
+
+[Service]
+MemoryHigh=2M
+Slice=testsuite-56-workload.slice
+ExecStart=/usr/lib/systemd/tests/testdata/units/testsuite-56-slowgrowth.sh

test/units/testsuite-56.sh

@@ -23,20 +23,43 @@ oomctl | grep "/testsuite-56-workload.slice"
 oomctl | grep "1.00%"
 oomctl | grep "Default Memory Pressure Duration: 5s"
 
-# systemd-oomd watches for elevated pressure for 30 seconds before acting.
+# systemd-oomd watches for elevated pressure for 5 seconds before acting.
-# It can take time to build up pressure so either wait 5 minutes or for the service to fail.
+# It can take time to build up pressure so either wait 2 minutes or for the service to fail.
-timeout=$(date -ud "5 minutes" +%s)
+timeout=$(date -ud "2 minutes" +%s)
 while [[ $(date -u +%s) -le $timeout ]]; do
     if ! systemctl status testsuite-56-testbloat.service; then
         break
     fi
-    sleep 15
+    sleep 5
 done
 
 # testbloat should be killed and testchill should be fine
 if systemctl status testsuite-56-testbloat.service; then exit 42; fi
 if ! systemctl status testsuite-56-testchill.service; then exit 24; fi
 
+# only run this portion of the test if we can set xattrs
+if setfattr -n user.xattr_test -v 1 /sys/fs/cgroup/; then
+    sleep 120 # wait for systemd-oomd kill cool down and elevated memory pressure to come down
+
+    systemctl start testsuite-56-testchill.service
+    systemctl start testsuite-56-testmunch.service
+    systemctl start testsuite-56-testbloat.service
+    setfattr -n user.oomd_avoid -v 1 /sys/fs/cgroup/testsuite.slice/testsuite-56.slice/testsuite-56-workload.slice/testsuite-56-testbloat.service
+
+    timeout=$(date -ud "2 minutes" +%s)
+    while [[ $(date -u +%s) -le $timeout ]]; do
+        if ! systemctl status testsuite-56-testmunch.service; then
+            break
+        fi
+        sleep 5
+    done
+
+    # testmunch should be killed since testbloat had the avoid xattr on it
+    if ! systemctl status testsuite-56-testbloat.service; then exit 25; fi
+    if systemctl status testsuite-56-testmunch.service; then exit 43; fi
+    if ! systemctl status testsuite-56-testchill.service; then exit 24; fi
+fi
+
 systemd-analyze log-level info
 
 echo OK > /testok