#define FUSE_USE_VERSION 26
+#define __STDC_FORMAT_MACROS
#include <dirent.h>
#include <errno.h>
#include <fcntl.h>
#include <fuse.h>
+#include <inttypes.h>
#include <libgen.h>
#include <pthread.h>
#include <sched.h>
#include <stdbool.h>
+#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/param.h>
#include <sys/socket.h>
#include <sys/syscall.h>
+#include <sys/sysinfo.h>
#include <sys/vfs.h>
#include "bindings.h"
#include "config.h" // for VERSION
+/* Maximum number for 64 bit integer is a string with 21 digits: 2^64 - 1 = 21 */
+#define LXCFS_NUMSTRLEN64 21
+
/* Define pivot_root() if missing from the C library */
#ifndef HAVE_PIVOT_ROOT
static int pivot_root(const char * new_root, const char * put_old)
LXC_TYPE_PROC_STAT,
LXC_TYPE_PROC_DISKSTATS,
LXC_TYPE_PROC_SWAPS,
+ LXC_TYPE_PROC_LOADAVG,
};
struct file_info {
int cached;
};
-/* reserve buffer size, for cpuall in /proc/stat */
-#define BUF_RESERVE_SIZE 256
+/* The function of hash table.*/
+#define LOAD_SIZE 100 /*the size of hash_table */
+#define FLUSH_TIME 5 /*the flush rate */
+#define DEPTH_DIR 3 /*the depth of per cgroup */
+/* The function of calculate loadavg .*/
+#define FSHIFT 11 /* nr of bits of precision */
+#define FIXED_1 (1<<FSHIFT) /* 1.0 as fixed-point */
+#define EXP_1 1884 /* 1/exp(5sec/1min) as fixed-point */
+#define EXP_5 2014 /* 1/exp(5sec/5min) */
+#define EXP_15 2037 /* 1/exp(5sec/15min) */
+#define LOAD_INT(x) ((x) >> FSHIFT)
+#define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100)
+/*
+ * This parameter is used for proc_loadavg_read().
+ * 1 means use loadavg, 0 means not use.
+ */
+static int loadavg = 0;
+static volatile sig_atomic_t loadavg_stop = 0;
+static int calc_hash(char *name)
+{
+ unsigned int hash = 0;
+ unsigned int x = 0;
+ /* ELFHash algorithm. */
+ while (*name) {
+ hash = (hash << 4) + *name++;
+ x = hash & 0xf0000000;
+ if (x != 0)
+ hash ^= (x >> 24);
+ hash &= ~x;
+ }
+ return ((hash & 0x7fffffff) % LOAD_SIZE);
+}
+
+struct load_node {
+ char *cg; /*cg */
+ unsigned long avenrun[3]; /* Load averages */
+ unsigned int run_pid;
+ unsigned int total_pid;
+ unsigned int last_pid;
+ int cfd; /* The file descriptor of the mounted cgroup */
+ struct load_node *next;
+ struct load_node **pre;
+};
+
+struct load_head {
+ /*
+ * The lock is about insert load_node and refresh load_node.To the first
+ * load_node of each hash bucket, insert and refresh in this hash bucket is
+ * mutually exclusive.
+ */
+ pthread_mutex_t lock;
+ /*
+ * The rdlock is about read loadavg and delete load_node.To each hash
+ * bucket, read and delete is mutually exclusive. But at the same time, we
+ * allow paratactic read operation. This rdlock is at list level.
+ */
+ pthread_rwlock_t rdlock;
+ /*
+ * The rilock is about read loadavg and insert load_node.To the first
+ * load_node of each hash bucket, read and insert is mutually exclusive.
+ * But at the same time, we allow paratactic read operation.
+ */
+ pthread_rwlock_t rilock;
+ struct load_node *next;
+};
+
+static struct load_head load_hash[LOAD_SIZE]; /* hash table */
+/*
+ * init_load initialize the hash table.
+ * Return 0 on success, return -1 on failure.
+ */
+static int init_load(void)
+{
+ int i;
+ int ret;
+
+ for (i = 0; i < LOAD_SIZE; i++) {
+ load_hash[i].next = NULL;
+ ret = pthread_mutex_init(&load_hash[i].lock, NULL);
+ if (ret != 0) {
+ lxcfs_error("%s\n", "Failed to initialize lock");
+ goto out3;
+ }
+ ret = pthread_rwlock_init(&load_hash[i].rdlock, NULL);
+ if (ret != 0) {
+ lxcfs_error("%s\n", "Failed to initialize rdlock");
+ goto out2;
+ }
+ ret = pthread_rwlock_init(&load_hash[i].rilock, NULL);
+ if (ret != 0) {
+ lxcfs_error("%s\n", "Failed to initialize rilock");
+ goto out1;
+ }
+ }
+ return 0;
+out1:
+ pthread_rwlock_destroy(&load_hash[i].rdlock);
+out2:
+ pthread_mutex_destroy(&load_hash[i].lock);
+out3:
+ while (i > 0) {
+ i--;
+ pthread_mutex_destroy(&load_hash[i].lock);
+ pthread_rwlock_destroy(&load_hash[i].rdlock);
+ pthread_rwlock_destroy(&load_hash[i].rilock);
+ }
+ return -1;
+}
+
+static void insert_node(struct load_node **n, int locate)
+{
+ struct load_node *f;
+
+ pthread_mutex_lock(&load_hash[locate].lock);
+ pthread_rwlock_wrlock(&load_hash[locate].rilock);
+ f = load_hash[locate].next;
+ load_hash[locate].next = *n;
+
+ (*n)->pre = &(load_hash[locate].next);
+ if (f)
+ f->pre = &((*n)->next);
+ (*n)->next = f;
+ pthread_mutex_unlock(&load_hash[locate].lock);
+ pthread_rwlock_unlock(&load_hash[locate].rilock);
+}
+/*
+ * locate_node() finds special node. Not return NULL means success.
+ * It should be noted that rdlock isn't unlocked at the end of code
+ * because this function is used to read special node. Delete is not
+ * allowed before read has ended.
+ * unlock rdlock only in proc_loadavg_read().
+ */
+static struct load_node *locate_node(char *cg, int locate)
+{
+ struct load_node *f = NULL;
+ int i = 0;
+
+ pthread_rwlock_rdlock(&load_hash[locate].rilock);
+ pthread_rwlock_rdlock(&load_hash[locate].rdlock);
+ if (load_hash[locate].next == NULL) {
+ pthread_rwlock_unlock(&load_hash[locate].rilock);
+ return f;
+ }
+ f = load_hash[locate].next;
+ pthread_rwlock_unlock(&load_hash[locate].rilock);
+ while (f && ((i = strcmp(f->cg, cg)) != 0))
+ f = f->next;
+ return f;
+}
+/* Delete the load_node n and return the next node of it. */
+static struct load_node *del_node(struct load_node *n, int locate)
+{
+ struct load_node *g;
+
+ pthread_rwlock_wrlock(&load_hash[locate].rdlock);
+ if (n->next == NULL) {
+ *(n->pre) = NULL;
+ } else {
+ *(n->pre) = n->next;
+ n->next->pre = n->pre;
+ }
+ g = n->next;
+ free(n->cg);
+ free(n);
+ pthread_rwlock_unlock(&load_hash[locate].rdlock);
+ return g;
+}
+
+static void load_free(void)
+{
+ int i;
+ struct load_node *f, *p;
+
+ for (i = 0; i < LOAD_SIZE; i++) {
+ pthread_mutex_lock(&load_hash[i].lock);
+ pthread_rwlock_wrlock(&load_hash[i].rilock);
+ pthread_rwlock_wrlock(&load_hash[i].rdlock);
+ if (load_hash[i].next == NULL) {
+ pthread_mutex_unlock(&load_hash[i].lock);
+ pthread_mutex_destroy(&load_hash[i].lock);
+ pthread_rwlock_unlock(&load_hash[i].rilock);
+ pthread_rwlock_destroy(&load_hash[i].rilock);
+ pthread_rwlock_unlock(&load_hash[i].rdlock);
+ pthread_rwlock_destroy(&load_hash[i].rdlock);
+ continue;
+ }
+ for (f = load_hash[i].next; f; ) {
+ free(f->cg);
+ p = f->next;
+ free(f);
+ f = p;
+ }
+ pthread_mutex_unlock(&load_hash[i].lock);
+ pthread_mutex_destroy(&load_hash[i].lock);
+ pthread_rwlock_unlock(&load_hash[i].rilock);
+ pthread_rwlock_destroy(&load_hash[i].rilock);
+ pthread_rwlock_unlock(&load_hash[i].rdlock);
+ pthread_rwlock_destroy(&load_hash[i].rdlock);
+ }
+}
+/* Reserve buffer size to account for file size changes. */
+#define BUF_RESERVE_SIZE 512
/*
* A table caching which pid is init for a pid namespace.
* another namespace using the *at() family of functions
* {openat(), fchownat(), ...}. */
static int *fd_hierarchies;
+static int cgroup_mount_ns_fd = -1;
static void unlock_mutex(pthread_mutex_t *l)
{
{
int i;
+ fprintf(stderr, "mount namespace: %d\n", cgroup_mount_ns_fd);
fprintf(stderr, "hierarchies:\n");
for (i = 0; i < num_hierarchies; i++) {
if (hierarchies[i])
fnam = alloca(len);
ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file);
if (ret < 0 || (size_t)ret >= len)
- return NULL;
+ return false;
fd = openat(cfd, fnam, O_RDONLY);
if (fd < 0)
- return NULL;
+ return false;
*value = slurp_file(fnam, fd);
return *value != NULL;
char *eol;
while (*memstat) {
- if (startswith(memstat, "cache")) {
- sscanf(memstat + 5, "%lu", cached);
+ if (startswith(memstat, "total_cache")) {
+ sscanf(memstat + 11, "%lu", cached);
*cached /= 1024;
- } else if (startswith(memstat, "active_anon")) {
- sscanf(memstat + 11, "%lu", active_anon);
+ } else if (startswith(memstat, "total_active_anon")) {
+ sscanf(memstat + 17, "%lu", active_anon);
*active_anon /= 1024;
- } else if (startswith(memstat, "inactive_anon")) {
- sscanf(memstat + 13, "%lu", inactive_anon);
+ } else if (startswith(memstat, "total_inactive_anon")) {
+ sscanf(memstat + 19, "%lu", inactive_anon);
*inactive_anon /= 1024;
- } else if (startswith(memstat, "active_file")) {
- sscanf(memstat + 11, "%lu", active_file);
+ } else if (startswith(memstat, "total_active_file")) {
+ sscanf(memstat + 17, "%lu", active_file);
*active_file /= 1024;
- } else if (startswith(memstat, "inactive_file")) {
- sscanf(memstat + 13, "%lu", inactive_file);
+ } else if (startswith(memstat, "total_inactive_file")) {
+ sscanf(memstat + 19, "%lu", inactive_file);
*inactive_file /= 1024;
- } else if (startswith(memstat, "unevictable")) {
- sscanf(memstat + 11, "%lu", unevictable);
+ } else if (startswith(memstat, "total_unevictable")) {
+ sscanf(memstat + 17, "%lu", unevictable);
*unevictable /= 1024;
}
eol = strchr(memstat, '\n');
snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage);
printme = lbuf;
} else if (startswith(line, "MemAvailable:")) {
- snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage);
+ snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage + cached);
printme = lbuf;
} else if (startswith(line, "SwapTotal:") && memswlimit > 0) {
sscanf(line+sizeof("SwapTotal:")-1, "%lu", &hostswtotal);
- if (hostswtotal < memswlimit - memlimit)
- memswlimit = hostswtotal + memlimit;
- snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit - memlimit);
+ if (hostswtotal < memswlimit)
+ memswlimit = hostswtotal;
+ snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit);
printme = lbuf;
} else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0) {
- unsigned long swaptotal = memswlimit - memlimit,
+ unsigned long swaptotal = memswlimit,
swapusage = memswusage - memusage,
swapfree = swapusage < swaptotal ? swaptotal - swapusage : 0;
snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree);
return rv;
}
+static uint64_t get_reaper_start_time(pid_t pid)
+{
+ int ret;
+ FILE *f;
+ uint64_t starttime;
+ /* strlen("/proc/") = 6
+ * +
+ * LXCFS_NUMSTRLEN64
+ * +
+ * strlen("/stat") = 5
+ * +
+ * \0 = 1
+ * */
+#define __PROC_PID_STAT_LEN (6 + LXCFS_NUMSTRLEN64 + 5 + 1)
+ char path[__PROC_PID_STAT_LEN];
+ pid_t qpid;
+
+ qpid = lookup_initpid_in_store(pid);
+ if (qpid <= 0) {
+ /* Caller can check for EINVAL on 0. */
+ errno = EINVAL;
+ return 0;
+ }
+
+ ret = snprintf(path, __PROC_PID_STAT_LEN, "/proc/%d/stat", qpid);
+ if (ret < 0 || ret >= __PROC_PID_STAT_LEN) {
+ /* Caller can check for EINVAL on 0. */
+ errno = EINVAL;
+ return 0;
+ }
+
+ f = fopen(path, "r");
+ if (!f) {
+ /* Caller can check for EINVAL on 0. */
+ errno = EINVAL;
+ return 0;
+ }
+
+ /* Note that the *scanf() argument supression requires that length
+ * modifiers such as "l" are omitted. Otherwise some compilers will yell
+ * at us. It's like telling someone you're not married and then asking
+ * if you can bring your wife to the party.
+ */
+ ret = fscanf(f, "%*d " /* (1) pid %d */
+ "%*s " /* (2) comm %s */
+ "%*c " /* (3) state %c */
+ "%*d " /* (4) ppid %d */
+ "%*d " /* (5) pgrp %d */
+ "%*d " /* (6) session %d */
+ "%*d " /* (7) tty_nr %d */
+ "%*d " /* (8) tpgid %d */
+ "%*u " /* (9) flags %u */
+ "%*u " /* (10) minflt %lu */
+ "%*u " /* (11) cminflt %lu */
+ "%*u " /* (12) majflt %lu */
+ "%*u " /* (13) cmajflt %lu */
+ "%*u " /* (14) utime %lu */
+ "%*u " /* (15) stime %lu */
+ "%*d " /* (16) cutime %ld */
+ "%*d " /* (17) cstime %ld */
+ "%*d " /* (18) priority %ld */
+ "%*d " /* (19) nice %ld */
+ "%*d " /* (20) num_threads %ld */
+ "%*d " /* (21) itrealvalue %ld */
+ "%" PRIu64, /* (22) starttime %llu */
+ &starttime);
+ if (ret != 1) {
+ fclose(f);
+ /* Caller can check for EINVAL on 0. */
+ errno = EINVAL;
+ return 0;
+ }
+
+ fclose(f);
+
+ errno = 0;
+ return starttime;
+}
+
+static uint64_t get_reaper_start_time_in_sec(pid_t pid)
+{
+ uint64_t clockticks;
+ int64_t ticks_per_sec;
+
+ clockticks = get_reaper_start_time(pid);
+ if (clockticks == 0 && errno == EINVAL) {
+ lxcfs_debug("failed to retrieve start time of pid %d\n", pid);
+ return 0;
+ }
+
+ ticks_per_sec = sysconf(_SC_CLK_TCK);
+ if (ticks_per_sec < 0 && errno == EINVAL) {
+ lxcfs_debug(
+ "%s\n",
+ "failed to determine number of clock ticks in a second");
+ return 0;
+ }
+
+ return (clockticks /= ticks_per_sec);
+}
+
+static uint64_t get_reaper_age(pid_t pid)
+{
+ uint64_t procstart, uptime, procage;
+
+ /* We need to substract the time the process has started since system
+ * boot minus the time when the system has started to get the actual
+ * reaper age.
+ */
+ procstart = get_reaper_start_time_in_sec(pid);
+ procage = procstart;
+ if (procstart > 0) {
+ int ret;
+ struct timespec spec;
+
+ ret = clock_gettime(CLOCK_BOOTTIME, &spec);
+ if (ret < 0)
+ return 0;
+ /* We could make this more precise here by using the tv_nsec
+ * field in the timespec struct and convert it to milliseconds
+ * and then create a double for the seconds and milliseconds but
+ * that seems more work than it is worth.
+ */
+ uptime = spec.tv_sec;
+ procage = uptime - procstart;
+ }
+
+ return procage;
+}
+
+#define CPUALL_MAX_SIZE (BUF_RESERVE_SIZE / 2)
static int proc_stat_read(char *buf, size_t size, off_t offset,
struct fuse_file_info *fi)
{
char *line = NULL;
size_t linelen = 0, total_len = 0, rv = 0;
int curcpu = -1; /* cpu numbering starts at 0 */
- unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0;
+ unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0;
unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0,
- irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0;
-#define CPUALL_MAX_SIZE BUF_RESERVE_SIZE
+ irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0, guest_nice_sum = 0;
char cpuall[CPUALL_MAX_SIZE];
/* reserve for cpu all */
char *cache = d->buf + CPUALL_MAX_SIZE;
cache_size -= l;
total_len += l;
- if (sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user, &nice, &system, &idle, &iowait, &irq,
- &softirq, &steal, &guest) != 9)
+ if (sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu",
+ &user,
+ &nice,
+ &system,
+ &idle,
+ &iowait,
+ &irq,
+ &softirq,
+ &steal,
+ &guest,
+ &guest_nice) != 10)
continue;
user_sum += user;
nice_sum += nice;
softirq_sum += softirq;
steal_sum += steal;
guest_sum += guest;
+ guest_nice_sum += guest_nice;
}
cache = d->buf;
- int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
- "cpu ", user_sum, nice_sum, system_sum, idle_sum, iowait_sum, irq_sum, softirq_sum, steal_sum, guest_sum);
- if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE){
+ int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "cpu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
+ user_sum,
+ nice_sum,
+ system_sum,
+ idle_sum,
+ iowait_sum,
+ irq_sum,
+ softirq_sum,
+ steal_sum,
+ guest_sum,
+ guest_nice_sum);
+ if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE) {
memcpy(cache, cpuall, cpuall_len);
cache += cpuall_len;
- } else{
+ } else {
/* shouldn't happen */
lxcfs_error("proc_stat_read copy cpuall failed, cpuall_len=%d.", cpuall_len);
cpuall_len = 0;
total_len += cpuall_len;
d->cached = 1;
d->size = total_len;
- if (total_len > size ) total_len = size;
+ if (total_len > size)
+ total_len = size;
memcpy(buf, d->buf, total_len);
rv = total_len;
return rv;
}
-static long int getreaperage(pid_t pid)
-{
- char fnam[100];
- struct stat sb;
- int ret;
- pid_t qpid;
-
- qpid = lookup_initpid_in_store(pid);
- if (qpid <= 0)
- return 0;
-
- ret = snprintf(fnam, 100, "/proc/%d", qpid);
- if (ret < 0 || ret >= 100)
- return 0;
-
- if (lstat(fnam, &sb) < 0)
- return 0;
-
- return time(NULL) - sb.st_ctime;
-}
-
+/* This function retrieves the busy time of a group of tasks by looking at
+ * cpuacct.usage. Unfortunately, this only makes sense when the container has
+ * been given it's own cpuacct cgroup. If not, this function will take the busy
+ * time of all other taks that do not actually belong to the container into
+ * account as well. If someone has a clever solution for this please send a
+ * patch!
+ */
static unsigned long get_reaper_busy(pid_t task)
{
pid_t initpid = lookup_initpid_in_store(task);
{
struct fuse_context *fc = fuse_get_context();
struct file_info *d = (struct file_info *)fi->fh;
- long int reaperage = getreaperage(fc->pid);
- unsigned long int busytime = get_reaper_busy(fc->pid), idletime;
+ unsigned long int busytime = get_reaper_busy(fc->pid);
char *cache = d->buf;
ssize_t total_len = 0;
+ uint64_t idletime, reaperage;
#if RELOADTEST
iwashere();
#endif
if (offset){
- if (offset > d->size)
- return -EINVAL;
if (!d->cached)
return 0;
+ if (offset > d->size)
+ return -EINVAL;
int left = d->size - offset;
total_len = left > size ? size: left;
memcpy(buf, cache + offset, total_len);
return total_len;
}
- idletime = reaperage - busytime;
- if (idletime > reaperage)
- idletime = reaperage;
+ reaperage = get_reaper_age(fc->pid);
+ /* To understand why this is done, please read the comment to the
+ * get_reaper_busy() function.
+ */
+ idletime = reaperage;
+ if (reaperage >= busytime)
+ idletime = reaperage - busytime;
- total_len = snprintf(d->buf, d->size, "%ld.0 %lu.0\n", reaperage, idletime);
- if (total_len < 0){
- perror("Error writing to cache");
+ total_len = snprintf(d->buf, d->buflen, "%"PRIu64".00 %"PRIu64".00\n", reaperage, idletime);
+ if (total_len < 0 || total_len >= d->buflen){
+ lxcfs_error("%s\n", "failed to write to cache");
return 0;
}
free(memswusage_str);
return rv;
}
+/*
+ * Find the process pid from cgroup path.
+ * eg:from /sys/fs/cgroup/cpu/docker/containerid/cgroup.procs to find the process pid.
+ * @pid_buf : put pid to pid_buf.
+ * @dpath : the path of cgroup. eg: /docker/containerid or /docker/containerid/child-cgroup ...
+ * @depth : the depth of cgroup in container.
+ * @sum : return the number of pid.
+ * @cfd : the file descriptor of the mounted cgroup. eg: /sys/fs/cgroup/cpu
+ */
+static int calc_pid(char ***pid_buf, char *dpath, int depth, int sum, int cfd)
+{
+ DIR *dir;
+ int fd;
+ struct dirent *file;
+ FILE *f = NULL;
+ size_t linelen = 0;
+ char *line = NULL;
+ int pd;
+ char *path_dir, *path;
+ char **pid;
+
+ /* path = dpath + "/cgroup.procs" + /0 */
+ do {
+ path = malloc(strlen(dpath) + 20);
+ } while (!path);
+
+ strcpy(path, dpath);
+ fd = openat(cfd, path, O_RDONLY);
+ if (fd < 0)
+ goto out;
+
+ dir = fdopendir(fd);
+ if (dir == NULL) {
+ close(fd);
+ goto out;
+ }
+
+ while (((file = readdir(dir)) != NULL) && depth > 0) {
+ if (strncmp(file->d_name, ".", 1) == 0)
+ continue;
+ if (strncmp(file->d_name, "..", 1) == 0)
+ continue;
+ if (file->d_type == DT_DIR) {
+ /* path + '/' + d_name +/0 */
+ do {
+ path_dir = malloc(strlen(path) + 2 + sizeof(file->d_name));
+ } while (!path_dir);
+ strcpy(path_dir, path);
+ strcat(path_dir, "/");
+ strcat(path_dir, file->d_name);
+ pd = depth - 1;
+ sum = calc_pid(pid_buf, path_dir, pd, sum, cfd);
+ free(path_dir);
+ }
+ }
+ closedir(dir);
+
+ strcat(path, "/cgroup.procs");
+ fd = openat(cfd, path, O_RDONLY);
+ if (fd < 0)
+ goto out;
+
+ f = fdopen(fd, "r");
+ if (!f) {
+ close(fd);
+ goto out;
+ }
+
+ while (getline(&line, &linelen, f) != -1) {
+ do {
+ pid = realloc(*pid_buf, sizeof(char *) * (sum + 1));
+ } while (!pid);
+ *pid_buf = pid;
+ do {
+ *(*pid_buf + sum) = malloc(strlen(line) + 1);
+ } while (*(*pid_buf + sum) == NULL);
+ strcpy(*(*pid_buf + sum), line);
+ sum++;
+ }
+ fclose(f);
+out:
+ free(path);
+ return sum;
+}
+/*
+ * calc_load calculates the load according to the following formula:
+ * load1 = load0 * exp + active * (1 - exp)
+ *
+ * @load1: the new loadavg.
+ * @load0: the former loadavg.
+ * @active: the total number of running pid at this moment.
+ * @exp: the fixed-point defined in the beginning.
+ */
+static unsigned long
+calc_load(unsigned long load, unsigned long exp, unsigned long active)
+{
+ unsigned long newload;
+
+ active = active > 0 ? active * FIXED_1 : 0;
+ newload = load * exp + active * (FIXED_1 - exp);
+ if (active >= load)
+ newload += FIXED_1 - 1;
+
+ return newload / FIXED_1;
+}
+
+/*
+ * Return 0 means that container p->cg is closed.
+ * Return -1 means that error occurred in refresh.
+ * Positive num equals the total number of pid.
+ */
+static int refresh_load(struct load_node *p, char *path)
+{
+ FILE *f = NULL;
+ char **idbuf;
+ char proc_path[256];
+ int i, ret, run_pid = 0, total_pid = 0, last_pid = 0;
+ char *line = NULL;
+ size_t linelen = 0;
+ int sum, length;
+ DIR *dp;
+ struct dirent *file;
+
+ do {
+ idbuf = malloc(sizeof(char *));
+ } while (!idbuf);
+ sum = calc_pid(&idbuf, path, DEPTH_DIR, 0, p->cfd);
+ /* normal exit */
+ if (sum == 0)
+ goto out;
+
+ for (i = 0; i < sum; i++) {
+ /*clean up '\n' */
+ length = strlen(idbuf[i])-1;
+ idbuf[i][length] = '\0';
+ ret = snprintf(proc_path, 256, "/proc/%s/task", idbuf[i]);
+ if (ret < 0 || ret > 255) {
+ lxcfs_error("%s\n", "snprintf() failed in refresh_load.");
+ i = sum;
+ sum = -1;
+ goto err_out;
+ }
+
+ dp = opendir(proc_path);
+ if (!dp) {
+ lxcfs_error("%s\n", "Open proc_path failed in refresh_load.");
+ continue;
+ }
+ while ((file = readdir(dp)) != NULL) {
+ if (strncmp(file->d_name, ".", 1) == 0)
+ continue;
+ if (strncmp(file->d_name, "..", 1) == 0)
+ continue;
+ total_pid++;
+ /* We make the biggest pid become last_pid.*/
+ ret = atof(file->d_name);
+ last_pid = (ret > last_pid) ? ret : last_pid;
+
+ ret = snprintf(proc_path, 256, "/proc/%s/task/%s/status", idbuf[i], file->d_name);
+ if (ret < 0 || ret > 255) {
+ lxcfs_error("%s\n", "snprintf() failed in refresh_load.");
+ i = sum;
+ sum = -1;
+ closedir(dp);
+ goto err_out;
+ }
+ f = fopen(proc_path, "r");
+ if (f != NULL) {
+ while (getline(&line, &linelen, f) != -1) {
+ /* Find State */
+ if ((line[0] == 'S') && (line[1] == 't'))
+ break;
+ }
+ if ((line[7] == 'R') || (line[7] == 'D'))
+ run_pid++;
+ fclose(f);
+ }
+ }
+ closedir(dp);
+ }
+ /*Calculate the loadavg.*/
+ p->avenrun[0] = calc_load(p->avenrun[0], EXP_1, run_pid);
+ p->avenrun[1] = calc_load(p->avenrun[1], EXP_5, run_pid);
+ p->avenrun[2] = calc_load(p->avenrun[2], EXP_15, run_pid);
+ p->run_pid = run_pid;
+ p->total_pid = total_pid;
+ p->last_pid = last_pid;
+
+ free(line);
+err_out:
+ for (; i > 0; i--)
+ free(idbuf[i-1]);
+out:
+ free(idbuf);
+ return sum;
+}
+/*
+ * Traverse the hash table and update it.
+ */
+void *load_begin(void *arg)
+{
+
+ char *path = NULL;
+ int i, sum, length, ret;
+ struct load_node *f;
+ int first_node;
+ clock_t time1, time2;
+
+ while (1) {
+ if (loadavg_stop == 1)
+ return NULL;
+
+ time1 = clock();
+ for (i = 0; i < LOAD_SIZE; i++) {
+ pthread_mutex_lock(&load_hash[i].lock);
+ if (load_hash[i].next == NULL) {
+ pthread_mutex_unlock(&load_hash[i].lock);
+ continue;
+ }
+ f = load_hash[i].next;
+ first_node = 1;
+ while (f) {
+ length = strlen(f->cg) + 2;
+ do {
+ /* strlen(f->cg) + '.' or '' + \0 */
+ path = malloc(length);
+ } while (!path);
+
+ ret = snprintf(path, length, "%s%s", *(f->cg) == '/' ? "." : "", f->cg);
+ if (ret < 0 || ret > length - 1) {
+ /* snprintf failed, ignore the node.*/
+ lxcfs_error("Refresh node %s failed for snprintf().\n", f->cg);
+ goto out;
+ }
+ sum = refresh_load(f, path);
+ if (sum == 0) {
+ f = del_node(f, i);
+ } else {
+out: f = f->next;
+ }
+ free(path);
+ /* load_hash[i].lock locks only on the first node.*/
+ if (first_node == 1) {
+ first_node = 0;
+ pthread_mutex_unlock(&load_hash[i].lock);
+ }
+ }
+ }
+
+ if (loadavg_stop == 1)
+ return NULL;
+
+ time2 = clock();
+ usleep(FLUSH_TIME * 1000000 - (int)((time2 - time1) * 1000000 / CLOCKS_PER_SEC));
+ }
+}
+
+static int proc_loadavg_read(char *buf, size_t size, off_t offset,
+ struct fuse_file_info *fi)
+{
+ struct fuse_context *fc = fuse_get_context();
+ struct file_info *d = (struct file_info *)fi->fh;
+ pid_t initpid;
+ char *cg;
+ size_t total_len = 0;
+ char *cache = d->buf;
+ struct load_node *n;
+ int hash;
+ int cfd;
+ unsigned long a, b, c;
+
+ if (offset) {
+ if (offset > d->size)
+ return -EINVAL;
+ if (!d->cached)
+ return 0;
+ int left = d->size - offset;
+ total_len = left > size ? size : left;
+ memcpy(buf, cache + offset, total_len);
+ return total_len;
+ }
+ if (!loadavg)
+ return read_file("/proc/loadavg", buf, size, d);
+
+ initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ cg = get_pid_cgroup(initpid, "cpu");
+ if (!cg)
+ return read_file("/proc/loadavg", buf, size, d);
+
+ prune_init_slice(cg);
+ hash = calc_hash(cg);
+ n = locate_node(cg, hash);
+
+ /* First time */
+ if (n == NULL) {
+ if (!find_mounted_controller("cpu", &cfd)) {
+ /*
+ * In locate_node() above, pthread_rwlock_unlock() isn't used
+ * because delete is not allowed before read has ended.
+ */
+ pthread_rwlock_unlock(&load_hash[hash].rdlock);
+ return 0;
+ }
+ do {
+ n = malloc(sizeof(struct load_node));
+ } while (!n);
+
+ do {
+ n->cg = malloc(strlen(cg)+1);
+ } while (!n->cg);
+ strcpy(n->cg, cg);
+ n->avenrun[0] = 0;
+ n->avenrun[1] = 0;
+ n->avenrun[2] = 0;
+ n->run_pid = 0;
+ n->total_pid = 1;
+ n->last_pid = initpid;
+ n->cfd = cfd;
+ insert_node(&n, hash);
+ }
+ a = n->avenrun[0] + (FIXED_1/200);
+ b = n->avenrun[1] + (FIXED_1/200);
+ c = n->avenrun[2] + (FIXED_1/200);
+ total_len = snprintf(d->buf, d->buflen, "%lu.%02lu %lu.%02lu %lu.%02lu %d/%d %d\n",
+ LOAD_INT(a), LOAD_FRAC(a),
+ LOAD_INT(b), LOAD_FRAC(b),
+ LOAD_INT(c), LOAD_FRAC(c),
+ n->run_pid, n->total_pid, n->last_pid);
+ pthread_rwlock_unlock(&load_hash[hash].rdlock);
+ if (total_len < 0 || total_len >= d->buflen) {
+ lxcfs_error("%s\n", "Failed to write to cache");
+ return 0;
+ }
+ d->size = (int)total_len;
+ d->cached = 1;
+
+ if (total_len > size)
+ total_len = size;
+ memcpy(buf, d->buf, total_len);
+ return total_len;
+}
+/* Return a positive number on success, return 0 on failure.*/
+pthread_t load_daemon(int load_use)
+{
+ int ret;
+ pthread_t pid;
+
+ ret = init_load();
+ if (ret == -1) {
+ lxcfs_error("%s\n", "Initialize hash_table fails in load_daemon!");
+ return 0;
+ }
+ ret = pthread_create(&pid, NULL, load_begin, NULL);
+ if (ret != 0) {
+ lxcfs_error("%s\n", "Create pthread fails in load_daemon!");
+ load_free();
+ return 0;
+ }
+ /* use loadavg, here loadavg = 1*/
+ loadavg = load_use;
+ return pid;
+}
+
+/* Returns 0 on success. */
+int stop_load_daemon(pthread_t pid)
+{
+ int s;
+
+ /* Signal the thread to gracefully stop */
+ loadavg_stop = 1;
+
+ s = pthread_join(pid, NULL); /* Make sure sub thread has been canceled. */
+ if (s != 0) {
+ lxcfs_error("%s\n", "stop_load_daemon error: failed to join");
+ return -1;
+ }
+
+ load_free();
+ loadavg_stop = 0;
+
+ return 0;
+}
static off_t get_procfile_size(const char *which)
{
strcmp(path, "/proc/uptime") == 0 ||
strcmp(path, "/proc/stat") == 0 ||
strcmp(path, "/proc/diskstats") == 0 ||
- strcmp(path, "/proc/swaps") == 0) {
+ strcmp(path, "/proc/swaps") == 0 ||
+ strcmp(path, "/proc/loadavg") == 0) {
sb->st_size = 0;
sb->st_mode = S_IFREG | 00444;
sb->st_nlink = 1;
filler(buf, "stat", NULL, 0) != 0 ||
filler(buf, "uptime", NULL, 0) != 0 ||
filler(buf, "diskstats", NULL, 0) != 0 ||
- filler(buf, "swaps", NULL, 0) != 0)
+ filler(buf, "swaps", NULL, 0) != 0 ||
+ filler(buf, "loadavg", NULL, 0) != 0)
return -EINVAL;
return 0;
}
type = LXC_TYPE_PROC_DISKSTATS;
else if (strcmp(path, "/proc/swaps") == 0)
type = LXC_TYPE_PROC_SWAPS;
+ else if (strcmp(path, "/proc/loadavg") == 0)
+ type = LXC_TYPE_PROC_LOADAVG;
if (type == -1)
return -ENOENT;
return proc_diskstats_read(buf, size, offset, fi);
case LXC_TYPE_PROC_SWAPS:
return proc_swaps_read(buf, size, offset, fi);
+ case LXC_TYPE_PROC_LOADAVG:
+ return proc_loadavg_read(buf, size, offset, fi);
default:
return -EINVAL;
}
return true;
}
+static int preserve_mnt_ns(int pid)
+{
+ int ret;
+ size_t len = sizeof("/proc/") + 21 + sizeof("/ns/mnt");
+ char path[len];
+
+ ret = snprintf(path, len, "/proc/%d/ns/mnt", pid);
+ if (ret < 0 || (size_t)ret >= len)
+ return -1;
+
+ return open(path, O_RDONLY | O_CLOEXEC);
+}
+
static bool cgfs_prepare_mounts(void)
{
if (!mkdir_p(BASEDIR, 0700)) {
return false;
}
+ cgroup_mount_ns_fd = preserve_mnt_ns(getpid());
+ if (cgroup_mount_ns_fd < 0) {
+ lxcfs_error("Failed to preserve mount namespace: %s.\n", strerror(errno));
+ return false;
+ }
+
if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) {
lxcfs_error("Failed to remount / private: %s.\n", strerror(errno));
return false;
for (i = 0; i < num_hierarchies; i++) {
char *controller = hierarchies[i];
+
clen = strlen(controller);
len = strlen(BASEDIR) + clen + 2;
target = malloc(len);
if (!target)
return false;
+
ret = snprintf(target, len, "%s/%s", BASEDIR, controller);
if (ret < 0 || ret >= len) {
free(target);
free(target);
return false;
}
- if (mount(controller, target, "cgroup", 0, controller) < 0) {
- lxcfs_error("Failed mounting cgroup %s\n", controller);
+ if (!strcmp(controller, "unified"))
+ ret = mount("none", target, "cgroup2", 0, NULL);
+ else
+ ret = mount(controller, target, "cgroup", 0, controller);
+ if (ret < 0) {
+ lxcfs_error("Failed mounting cgroup %s: %s\n", controller, strerror(errno));
free(target);
return false;
}
return true;
}
-static int preserve_ns(int pid)
-{
- int ret;
- size_t len = 5 /* /proc */ + 21 /* /int_as_str */ + 7 /* /ns/mnt */ + 1 /* \0 */;
- char path[len];
-
- ret = snprintf(path, len, "/proc/%d/ns/mnt", pid);
- if (ret < 0 || (size_t)ret >= len)
- return -1;
-
- return open(path, O_RDONLY | O_CLOEXEC);
-}
-
static void __attribute__((constructor)) collect_and_mount_subsystems(void)
{
FILE *f;
char cwd[MAXPATHLEN];
size_t len = 0;
int i, init_ns = -1;
+ bool found_unified = false;
if ((f = fopen("/proc/self/cgroup", "r")) == NULL) {
lxcfs_error("Error opening /proc/self/cgroup: %s\n", strerror(errno));
}
while (getline(&line, &len, f) != -1) {
- char *p, *p2;
+ char *idx, *p, *p2;
p = strchr(line, ':');
if (!p)
goto out;
+ idx = line;
*(p++) = '\0';
p2 = strrchr(p, ':');
* because it parses out the empty string "" and later on passes
* it to mount(). Let's skip such entries.
*/
- if (!strcmp(p, ""))
- continue;
+ if (!strcmp(p, "") && !strcmp(idx, "0") && !found_unified) {
+ found_unified = true;
+ p = "unified";
+ }
if (!store_hierarchy(line, p))
goto out;
}
/* Preserve initial namespace. */
- init_ns = preserve_ns(getpid());
+ init_ns = preserve_mnt_ns(getpid());
if (init_ns < 0) {
lxcfs_error("%s\n", "Failed to preserve initial mount namespace.");
goto out;
}
- fd_hierarchies = malloc(sizeof(int *) * num_hierarchies);
+ fd_hierarchies = malloc(sizeof(int) * num_hierarchies);
if (!fd_hierarchies) {
lxcfs_error("%s\n", strerror(errno));
goto out;
}
free(hierarchies);
free(fd_hierarchies);
+
+ if (cgroup_mount_ns_fd >= 0)
+ close(cgroup_mount_ns_fd);
}