Merge pull request #19647 from ddstreet/test-oomd-failure

handle lack of kernel support for some cgroup files

src/basic/fileio.c

@@ -371,8 +371,6 @@ int read_virtual_file(const char *filename, size_t max_size, char **ret_contents
         int n_retries;
         bool truncated = false;
 
-        assert(ret_contents);
-
         /* Virtual filesystems such as sysfs or procfs use kernfs, and kernfs can work with two sorts of
          * virtual files. One sort uses "seq_file", and the results of the first read are buffered for the
          * second read. The other sort uses "raw" reads which always go direct to the device. In the latter
@@ -486,7 +484,9 @@ int read_virtual_file(const char *filename, size_t max_size, char **ret_contents
                 return -EBADMSG;
 
         buf[n] = 0;
-        *ret_contents = TAKE_PTR(buf);
+
+        if (ret_contents)
+                *ret_contents = TAKE_PTR(buf);
 
         return !truncated;
 }

src/basic/log.h

@@ -236,6 +236,29 @@ int log_emergency_level(void);
 #define log_error_errno(error, ...)     log_full_errno(LOG_ERR,     error, __VA_ARGS__)
 #define log_emergency_errno(error, ...) log_full_errno(log_emergency_level(), error, __VA_ARGS__)
 
+/* This logs at the specified level the first time it is called, and then
+ * logs at debug. If the specified level is debug, this logs only the first
+ * time it is called. */
+#define log_once(level, ...)                                             \
+        ({                                                               \
+                if (ONCE)                                                \
+                        log_full(level, __VA_ARGS__);                    \
+                else if (LOG_PRI(level) != LOG_DEBUG)                    \
+                        log_debug(__VA_ARGS__);                          \
+        })
+
+#define log_once_errno(level, error, ...)                                \
+        ({                                                               \
+                int _err = (error);                                      \
+                if (ONCE)                                                \
+                        _err = log_full_errno(level, _err, __VA_ARGS__); \
+                else if (LOG_PRI(level) != LOG_DEBUG)                    \
+                        _err = log_debug_errno(_err, __VA_ARGS__);       \
+                else                                                     \
+                        _err = -ERRNO_VALUE(_err);                       \
+                _err;                                                    \
+        })
+
 #if LOG_TRACE
 #  define log_trace(...) log_debug(__VA_ARGS__)
 #else

src/fundamental/macro-fundamental.h

@@ -50,6 +50,17 @@
 #define UNIQ_T(x, uniq) CONCATENATE(__unique_prefix_, CONCATENATE(x, uniq))
 #define UNIQ __COUNTER__
 
+/* Note that this works differently from pthread_once(): this macro does
+ * not synchronize code execution, i.e. code that is run conditionalized
+ * on this macro will run concurrently to all other code conditionalized
+ * the same way, there's no ordering or completion enforced. */
+#define ONCE __ONCE(UNIQ_T(_once_, UNIQ))
+#define __ONCE(o)                                                       \
+        ({                                                              \
+                static bool (o) = false;                                \
+                __sync_bool_compare_and_swap(&(o), false, true);        \
+        })
+
 #undef MAX
 #define MAX(a, b) __MAX(UNIQ, (a), UNIQ, (b))
 #define __MAX(aq, a, bq, b)                             \

src/oom/oomd-util.c

@@ -333,7 +333,11 @@ int oomd_cgroup_context_acquire(const char *path, OomdCGroupContext **ret) {
                         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)
+                if (r == -ENODATA)
+                        /* The kernel can be compiled without support for memory.swap.* files,
+                         * or it can be disabled with boot param 'swapaccount=0' */
+                        log_once(LOG_WARNING, "No kernel support for memory.swap.current from %s (try boot param swapaccount=1), ignoring.", path);
+                else 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);

src/shared/psi-util.c

@@ -4,6 +4,7 @@
 #include <unistd.h>
 
 #include "alloc-util.h"
+#include "errno-util.h"
 #include "extract-word.h"
 #include "fd-util.h"
 #include "fileio.h"
@@ -107,12 +108,18 @@ int read_resource_pressure(const char *path, PressureType type, ResourcePressure
 int is_pressure_supported(void) {
         const char *p;
 
-        FOREACH_STRING(p, "/proc/pressure/cpu", "/proc/pressure/io", "/proc/pressure/memory")
-                if (access(p, F_OK) < 0) {
-                        if (errno == ENOENT)
-                                return 0;
-                        return -errno;
-                }
+        /* The pressure files, both under /proc and in cgroups, will exist
+         * even if the kernel has PSI support disabled; we have to read
+         * the file to make sure it doesn't return -EOPNOTSUPP */
+        FOREACH_STRING(p, "/proc/pressure/cpu", "/proc/pressure/io", "/proc/pressure/memory") {
+                int r;
+
+                r = read_virtual_file(p, 0, NULL, NULL);
+                if (r == -ENOENT || ERRNO_IS_NOT_SUPPORTED(r))
+                        return 0;
+                if (r < 0)
+                        return r;
+        }
 
         return 1;
 }