proc_fuse: add psi(pressure stall information) procfs

Kernel support psi(pressure stall information) since 4.20
with procfs /proc/pressure/{io,cpu,memory} and
cgroupv2 {io.pressure, cpu.pressure, memory.pressure}.

This patch add read-only psi procfs,
and people can get pressure information now.
Full functional feature for monitoring are still under investigation.

Signed-off-by: Feng Sun <[email protected]>

Author: loyou

README.md

@@ -16,6 +16,9 @@ such as:
 /proc/swaps
 /proc/uptime
 /proc/slabinfo
+/proc/pressure/io
+/proc/pressure/cpu
+/proc/pressure/memory
 /sys/devices/system/cpu/online
 ```
 
@@ -109,6 +112,9 @@ docker run -it -m 256m --memory-swap 256m \
       -v /var/lib/lxcfs/proc/swaps:/proc/swaps:rw \
       -v /var/lib/lxcfs/proc/uptime:/proc/uptime:rw \
       -v /var/lib/lxcfs/proc/slabinfo:/proc/slabinfo:rw \
+      -v /var/lib/lxcfs/proc/pressure/io:/proc/pressure/io:rw \
+      -v /var/lib/lxcfs/proc/pressure/cpu:/proc/pressure/cpu:rw \
+      -v /var/lib/lxcfs/proc/pressure/memory:/proc/pressure/memory:rw \
       -v /var/lib/lxcfs/sys/devices/system/cpu:/sys/devices/system/cpu:rw \
       ubuntu:18.04 /bin/bash
  ```

src/api_extensions.h

@@ -23,6 +23,9 @@ static char *api_extensions[] = {
 	"proc_swaps",
 	"proc_uptime",
 	"proc_slabinfo",
+	"proc_pressure_io",
+	"proc_pressure_cpu",
+	"proc_pressure_memory",
 	"shared_pidns",
 	"cpuview_daemon",
 	"loadavg_daemon",

src/bindings.h

@@ -66,12 +66,24 @@ enum lxcfs_virt_t {
 
 	LXC_TYPE_SYS_DEVICES_SYSTEM_CPU_ONLINE,
 #define LXC_TYPE_SYS_DEVICES_SYSTEM_CPU_ONLINE_PATH "/sys/devices/system/cpu/online"
+
+	LXC_TYPE_PROC,
+	LXC_TYPE_PROC_PRESSURE,
+	LXC_TYPE_PROC_PRESSURE_IO,
+#define LXC_TYPE_PROC_PRESSURE_IO_PATH "/proc/pressure/io"
+
+	LXC_TYPE_PROC_PRESSURE_CPU,
+#define LXC_TYPE_PROC_PRESSURE_CPU_PATH "/proc/pressure/cpu"
+
+	LXC_TYPE_PROC_PRESSURE_MEMORY,
+#define LXC_TYPE_PROC_PRESSURE_MEMORY_PATH "/proc/pressure/memory"
 	LXC_TYPE_MAX,
 };
 
 /* Macros below used to check the class from the file types above */
 #define LXCFS_TYPE_CGROUP(type) (type >= LXC_TYPE_CGDIR && type <= LXC_TYPE_CGFILE)
-#define LXCFS_TYPE_PROC(type) (type >= LXC_TYPE_PROC_MEMINFO && type <= LXC_TYPE_PROC_SLABINFO)
+#define LXCFS_TYPE_PROC(type) ((type >= LXC_TYPE_PROC_MEMINFO && type <= LXC_TYPE_PROC_SLABINFO) || \
+							   (type >= LXC_TYPE_PROC && type <= LXC_TYPE_PROC_PRESSURE_MEMORY))
 #define LXCFS_TYPE_SYS(type) (type >= LXC_TYPE_SYS && type <= LXC_TYPE_SYS_DEVICES_SYSTEM_CPU_ONLINE)
 #define LXCFS_TYPE_OK(type) (type >= LXC_TYPE_CGDIR && type < LXC_TYPE_MAX)
 

src/cgroups/cgfsng.c

@@ -854,6 +854,54 @@ static bool cgfsng_can_use_cpuview(struct cgroup_ops *ops)
 	return true;
 }
 
+static int cgfsng_get_pressure_io_fd(struct cgroup_ops *ops, const char *cgroup)
+{
+	__do_free char *path = NULL;
+	struct hierarchy *h;
+
+	h = ops->get_hierarchy(ops, "blkio");
+	if (!h)
+		return -1;
+
+	if (faccessat(h->fd, "io.pressure", F_OK, 0))
+		return -1;
+
+	path = must_make_path_relative(cgroup, "io.pressure", NULL);
+	return openat(h->fd, path, O_RDWR | O_CLOEXEC | O_NOFOLLOW);
+}
+
+static int cgfsng_get_pressure_cpu_fd(struct cgroup_ops *ops, const char *cgroup)
+{
+	__do_free char *path = NULL;
+	struct hierarchy *h;
+
+	h = ops->get_hierarchy(ops, "cpu");
+	if (!h)
+		return -1;
+
+	if (faccessat(h->fd, "cpu.pressure", F_OK, 0))
+		return -1;
+
+	path = must_make_path_relative(cgroup, "cpu.pressure", NULL);
+	return openat(h->fd, path, O_RDWR | O_CLOEXEC | O_NOFOLLOW);
+}
+
+static int cgfsng_get_pressure_memory_fd(struct cgroup_ops *ops, const char *cgroup)
+{
+	__do_free char *path = NULL;
+	struct hierarchy *h;
+
+	h = ops->get_hierarchy(ops, "memory");
+	if (!h)
+		return -1;
+
+	if (faccessat(h->fd, "memory.pressure", F_OK, 0))
+		return -1;
+
+	path = must_make_path_relative(cgroup, "memory.pressure", NULL);
+	return openat(h->fd, path, O_RDWR | O_CLOEXEC | O_NOFOLLOW);
+}
+
 /* At startup, parse_hierarchies finds all the info we need about cgroup
  * mountpoints and current cgroups, and stores it in @d.
  */
@@ -1074,6 +1122,10 @@ struct cgroup_ops *cgfsng_ops_init(void)
 	cgfsng_ops->get_io_merged		= cgfsng_get_io_merged;
 	cgfsng_ops->get_io_wait_time		= cgfsng_get_io_wait_time;
 
+	/* psi */
+	cgfsng_ops->get_pressure_io_fd = cgfsng_get_pressure_io_fd;
+	cgfsng_ops->get_pressure_cpu_fd = cgfsng_get_pressure_cpu_fd;
+	cgfsng_ops->get_pressure_memory_fd = cgfsng_get_pressure_memory_fd;
 
 	return move_ptr(cgfsng_ops);
 }

src/cgroups/cgroup.h

@@ -155,7 +155,7 @@ struct cgroup_ops {
 			       char **value);
 	bool (*can_use_cpuview)(struct cgroup_ops *ops);
 
-	/* io */
+	/* blkio */
 	int (*get_io_service_bytes)(struct cgroup_ops *ops, const char *cgroup,
 				    char **value);
 	int (*get_io_service_time)(struct cgroup_ops *ops, const char *cgroup,
@@ -166,6 +166,11 @@ struct cgroup_ops {
 			     char **value);
 	int (*get_io_wait_time)(struct cgroup_ops *ops, const char *cgroup,
 				char **value);
+	/* psi */
+	int (*get_pressure_io_fd)(struct cgroup_ops *ops, const char *cgroup);
+	int (*get_pressure_cpu_fd)(struct cgroup_ops *ops, const char *cgroup);
+	int (*get_pressure_memory_fd)(struct cgroup_ops *ops,
+				      const char *cgroup);
 };
 
 extern struct cgroup_ops *cgroup_ops;

src/lxcfs.c

@@ -524,6 +524,20 @@ static int do_proc_open(const char *path, struct fuse_file_info *fi)
 	return __proc_open(path, fi);
 }
 
+static int do_proc_opendir(const char *path, struct fuse_file_info *fi)
+{
+	char *error;
+	int (*__proc_opendir)(const char *path, struct fuse_file_info *fi);
+
+	dlerror();
+	__proc_opendir = (int (*)(const char *path, struct fuse_file_info *fi))dlsym(dlopen_handle, "proc_opendir");
+	error = dlerror();
+	if (error)
+		return log_error(-1, "%s - Failed to find proc_opendir()", error);
+
+	return __proc_opendir(path, fi);
+}
+
 static int do_proc_access(const char *path, int mode)
 {
 	char *error;
@@ -608,6 +622,20 @@ static int do_proc_release(const char *path, struct fuse_file_info *fi)
 	return __proc_release(path, fi);
 }
 
+static int do_proc_releasedir(const char *path, struct fuse_file_info *fi)
+{
+	char *error;
+	int (*__proc_releasedir)(const char *path, struct fuse_file_info *fi);
+
+	dlerror();
+	__proc_releasedir = (int (*)(const char *path, struct fuse_file_info *)) dlsym(dlopen_handle, "proc_releasedir");
+	error = dlerror();
+	if (error)
+		return log_error(-1, "%s - Failed to find proc_releasedir()", error);
+
+	return __proc_releasedir(path, fi);
+}
+
 static int do_sys_release(const char *path, struct fuse_file_info *fi)
 {
 	char *error;
@@ -724,8 +752,12 @@ static int lxcfs_opendir(const char *path, struct fuse_file_info *fi)
 		return ret;
 	}
 
-	if (strcmp(path, "/proc") == 0)
-		return 0;
+	if (strncmp(path, "/proc", 5) == 0) {
+		up_users();
+		ret = do_proc_opendir(path, fi);
+		down_users();
+		return ret;
+	}
 
 	if (strncmp(path, "/sys", 4) == 0) {
 		up_users();
@@ -768,7 +800,7 @@ static int lxcfs_readdir(const char *path, void *buf, fuse_fill_dir_t filler,
 		return ret;
 	}
 
-	if (strcmp(path, "/proc") == 0) {
+	if (LXCFS_TYPE_PROC(type)) {
 		up_users();
 		ret = do_proc_readdir(path, buf, filler, offset, fi);
 		down_users();
@@ -837,12 +869,14 @@ static int lxcfs_releasedir(const char *path, struct fuse_file_info *fi)
 		return ret;
 	}
 
-	if (path) {
-		if (strcmp(path, "/") == 0)
-			return 0;
-		if (strcmp(path, "/proc") == 0)
-			return 0;
+	if (LXCFS_TYPE_PROC(type)) {
+		up_users();
+		ret = do_proc_releasedir(path, fi);
+		down_users();
+		return ret;
 	}
+	if (path && strcmp(path, "/") == 0)
+		return 0;
 
 	lxcfs_error("unknown file type: path=%s, type=%d, fi->fh=%" PRIu64,
 			path, type, fi->fh);