/* lxcfs
*
- * Copyright © 2014,2015 Canonical, Inc
+ * Copyright © 2014-2016 Canonical, Inc
* Author: Serge Hallyn <serge.hallyn@ubuntu.com>
*
* See COPYING file for details.
#include <stdlib.h>
#include <libgen.h>
#include <sched.h>
+#include <pthread.h>
#include <linux/sched.h>
#include <sys/socket.h>
#include <sys/mount.h>
#define BUF_RESERVE_SIZE 256
/*
- * append pid to *src.
- * src: a pointer to a char* in which ot append the pid.
- * sz: the number of characters printed so far, minus trailing \0.
- * asz: the allocated size so far
- * pid: the pid to append
+ * A table caching which pid is init for a pid namespace.
+ * When looking up which pid is init for $qpid, we first
+ * 1. Stat /proc/$qpid/ns/pid.
+ * 2. Check whether the ino_t is in our store.
+ * a. if not, fork a child in qpid's ns to send us
+ * ucred.pid = 1, and read the initpid. Cache
+ * initpid and creation time for /proc/initpid
+ * in a new store entry.
+ * b. if so, verify that /proc/initpid still matches
+ * what we have saved. If not, clear the store
+ * entry and go back to a. If so, return the
+ * cached initpid.
*/
-static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid)
+struct pidns_init_store {
+ ino_t ino; // inode number for /proc/$pid/ns/pid
+ pid_t initpid; // the pid of nit in that ns
+ long int ctime; // the time at which /proc/$initpid was created
+ struct pidns_init_store *next;
+ long int lastcheck;
+};
+
+/* lol - look at how they are allocated in the kernel */
+#define PIDNS_HASH_SIZE 4096
+#define HASH(x) ((x) % PIDNS_HASH_SIZE)
+
+struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE];
+static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER;
+static void lock_mutex(pthread_mutex_t *l)
{
- char tmp[30];
+ int ret;
- int tmplen = sprintf(tmp, "%d\n", (int)pid);
+ if ((ret = pthread_mutex_lock(l)) != 0) {
+ fprintf(stderr, "pthread_mutex_lock returned:%d %s\n", ret, strerror(ret));
+ exit(1);
+ }
+}
- if (!*src || tmplen + *sz + 1 >= *asz) {
- char *tmp;
- do {
- tmp = realloc(*src, *asz + BUF_RESERVE_SIZE);
- } while (!tmp);
- *src = tmp;
- *asz += BUF_RESERVE_SIZE;
+static void unlock_mutex(pthread_mutex_t *l)
+{
+ int ret;
+
+ if ((ret = pthread_mutex_unlock(l)) != 0) {
+ fprintf(stderr, "pthread_mutex_unlock returned:%d %s\n", ret, strerror(ret));
+ exit(1);
}
- memcpy((*src) +*sz , tmp, tmplen);
- *sz += tmplen;
- (*src)[*sz] = '\0';
}
-static pid_t get_init_pid_for_task(pid_t task);
+static void store_lock(void)
+{
+ lock_mutex(&pidns_store_mutex);
+}
+
+static void store_unlock(void)
+{
+ unlock_mutex(&pidns_store_mutex);
+}
+
+/* Must be called under store_lock */
+static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb)
+{
+ struct stat initsb;
+ char fnam[100];
+
+ snprintf(fnam, 100, "/proc/%d", e->initpid);
+ if (stat(fnam, &initsb) < 0)
+ return false;
+#if DEBUG
+ fprintf(stderr, "comparing ctime %ld %ld for pid %d\n",
+ e->ctime, initsb.st_ctime, e->initpid);
+#endif
+ if (e->ctime != initsb.st_ctime)
+ return false;
+ return true;
+}
+
+/* Must be called under store_lock */
+static void remove_initpid(struct pidns_init_store *e)
+{
+ struct pidns_init_store *tmp;
+ int h;
+
+#if DEBUG
+ fprintf(stderr, "remove_initpid: removing entry for %d\n", e->initpid);
+#endif
+ h = HASH(e->ino);
+ if (pidns_hash_table[h] == e) {
+ pidns_hash_table[h] = e->next;
+ free(e);
+ return;
+ }
+
+ tmp = pidns_hash_table[h];
+ while (tmp) {
+ if (tmp->next == e) {
+ tmp->next = e->next;
+ free(e);
+ return;
+ }
+ tmp = tmp->next;
+ }
+}
+
+#define PURGE_SECS 5
+/* Must be called under store_lock */
+static void prune_initpid_store(void)
+{
+ static long int last_prune = 0;
+ struct pidns_init_store *e, *prev, *delme;
+ long int now, threshold;
+ int i;
+
+ if (!last_prune) {
+ last_prune = time(NULL);
+ return;
+ }
+ now = time(NULL);
+ if (now < last_prune + PURGE_SECS)
+ return;
+#if DEBUG
+ fprintf(stderr, "pruning\n");
+#endif
+ last_prune = now;
+ threshold = now - 2 * PURGE_SECS;
+
+ for (i = 0; i < PIDNS_HASH_SIZE; i++) {
+ for (prev = NULL, e = pidns_hash_table[i]; e; ) {
+ if (e->lastcheck < threshold) {
+#if DEBUG
+ fprintf(stderr, "Removing cached entry for %d\n", e->initpid);
+#endif
+ delme = e;
+ if (prev)
+ prev->next = e->next;
+ else
+ pidns_hash_table[i] = e->next;
+ e = e->next;
+ free(delme);
+ } else {
+ prev = e;
+ e = e->next;
+ }
+ }
+ }
+}
+
+/* Must be called under store_lock */
+static void save_initpid(struct stat *sb, pid_t pid)
+{
+ struct pidns_init_store *e;
+ char fpath[100];
+ struct stat procsb;
+ int h;
+
+#if DEBUG
+ fprintf(stderr, "save_initpid: adding entry for %d\n", pid);
+#endif
+ snprintf(fpath, 100, "/proc/%d", pid);
+ if (stat(fpath, &procsb) < 0)
+ return;
+ do {
+ e = malloc(sizeof(*e));
+ } while (!e);
+ e->ino = sb->st_ino;
+ e->initpid = pid;
+ e->ctime = procsb.st_ctime;
+ h = HASH(e->ino);
+ e->next = pidns_hash_table[h];
+ e->lastcheck = time(NULL);
+ pidns_hash_table[h] = e;
+}
+
+/*
+ * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store
+ * entry for the inode number and creation time. Verify that the init pid
+ * is still valid. If not, remove it. Return the entry if valid, NULL
+ * otherwise.
+ * Must be called under store_lock
+ */
+static struct pidns_init_store *lookup_verify_initpid(struct stat *sb)
+{
+ int h = HASH(sb->st_ino);
+ struct pidns_init_store *e = pidns_hash_table[h];
+
+ while (e) {
+ if (e->ino == sb->st_ino) {
+ if (initpid_still_valid(e, sb)) {
+ e->lastcheck = time(NULL);
+ return e;
+ }
+ remove_initpid(e);
+ return NULL;
+ }
+ e = e->next;
+ }
+
+ return NULL;
+}
+
+#define SEND_CREDS_OK 0
+#define SEND_CREDS_NOTSK 1
+#define SEND_CREDS_FAIL 2
+static bool recv_creds(int sock, struct ucred *cred, char *v);
+static int wait_for_pid(pid_t pid);
+static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst);
+
+/*
+ * fork a task which switches to @task's namespace and writes '1'.
+ * over a unix sock so we can read the task's reaper's pid in our
+ * namespace
+ */
+static void write_task_init_pid_exit(int sock, pid_t target)
+{
+ struct ucred cred;
+ char fnam[100];
+ pid_t pid;
+ char v;
+ int fd, ret;
+
+ ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target);
+ if (ret < 0 || ret >= sizeof(fnam))
+ _exit(1);
+
+ fd = open(fnam, O_RDONLY);
+ if (fd < 0) {
+ perror("write_task_init_pid_exit open of ns/pid");
+ _exit(1);
+ }
+ if (setns(fd, 0)) {
+ perror("write_task_init_pid_exit setns 1");
+ close(fd);
+ _exit(1);
+ }
+ pid = fork();
+ if (pid < 0)
+ _exit(1);
+ if (pid != 0) {
+ if (!wait_for_pid(pid))
+ _exit(1);
+ _exit(0);
+ }
+
+ /* we are the child */
+ cred.uid = 0;
+ cred.gid = 0;
+ cred.pid = 1;
+ v = '1';
+ if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK)
+ _exit(1);
+ _exit(0);
+}
+
+static pid_t get_init_pid_for_task(pid_t task)
+{
+ int sock[2];
+ pid_t pid;
+ pid_t ret = -1;
+ char v = '0';
+ struct ucred cred;
+
+ if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) {
+ perror("socketpair");
+ return -1;
+ }
+
+ pid = fork();
+ if (pid < 0)
+ goto out;
+ if (!pid) {
+ close(sock[1]);
+ write_task_init_pid_exit(sock[0], task);
+ _exit(0);
+ }
+
+ if (!recv_creds(sock[1], &cred, &v))
+ goto out;
+ ret = cred.pid;
+
+out:
+ close(sock[0]);
+ close(sock[1]);
+ if (pid > 0)
+ wait_for_pid(pid);
+ return ret;
+}
+
+static pid_t lookup_initpid_in_store(pid_t qpid)
+{
+ pid_t answer = 0;
+ struct stat sb;
+ struct pidns_init_store *e;
+ char fnam[100];
+
+ snprintf(fnam, 100, "/proc/%d/ns/pid", qpid);
+ store_lock();
+ if (stat(fnam, &sb) < 0)
+ goto out;
+ e = lookup_verify_initpid(&sb);
+ if (e) {
+ answer = e->initpid;
+ goto out;
+ }
+ answer = get_init_pid_for_task(qpid);
+ if (answer > 0)
+ save_initpid(&sb, answer);
+
+out:
+ /* we prune at end in case we are returning
+ * the value we were about to return */
+ prune_initpid_store();
+ store_unlock();
+ return answer;
+}
static int wait_for_pid(pid_t pid)
{
int status, ret;
+ if (pid <= 0)
+ return -1;
+
again:
ret = waitpid(pid, &status, 0);
if (ret == -1) {
return 0;
}
+
+/*
+ * append pid to *src.
+ * src: a pointer to a char* in which ot append the pid.
+ * sz: the number of characters printed so far, minus trailing \0.
+ * asz: the allocated size so far
+ * pid: the pid to append
+ */
+static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid)
+{
+ char tmp[30];
+
+ int tmplen = sprintf(tmp, "%d\n", (int)pid);
+
+ if (!*src || tmplen + *sz + 1 >= *asz) {
+ char *tmp;
+ do {
+ tmp = realloc(*src, *asz + BUF_RESERVE_SIZE);
+ } while (!tmp);
+ *src = tmp;
+ *asz += BUF_RESERVE_SIZE;
+ }
+ memcpy((*src) +*sz , tmp, tmplen);
+ *sz += tmplen;
+ (*src)[*sz] = '\0';
+}
+
/*
* Given a open file * to /proc/pid/{u,g}id_map, and an id
* valid in the caller's namespace, return the id mapped into
static void prune_init_slice(char *cg)
{
char *point;
- point = cg + strlen(cg) - strlen(INITSCOPE);
- if (point < cg)
- return;
+ size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE);
+
+ if (cg_len < initscope_len)
+ return;
+
+ point = cg + cg_len - initscope_len;
if (strcmp(point, INITSCOPE) == 0) {
if (point == cg)
*(point+1) = '\0';
}
/*
- * If caller is in /a/b/c/d, he may only act on things under cg=/a/b/c/d.
- * If caller is in /a, he may act on /a/b, but not on /b.
+ * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d.
+ * If pid is in /a, he may act on /a/b, but not on /b.
* if the answer is false and nextcg is not NULL, then *nextcg will point
* to a string containing the next cgroup directory under cg, which must be
* freed by the caller.
}
/*
- * If caller is in /a/b/c, he may see that /a exists, but not /b or /a/c.
+ * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c.
*/
static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg)
{
path2 = last;
}
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
/* check that cgcopy is either a child cgroup of cgdir, or listed in its keys.
* Then check that caller's cgroup is under path if last is a child
* cgroup, or cgdir if last is a file */
if (is_child_cgroup(controller, path1, path2)) {
- if (!caller_may_see_dir(fc->pid, controller, cgroup)) {
+ if (!caller_may_see_dir(initpid, controller, cgroup)) {
ret = -ENOENT;
goto out;
}
- if (!caller_is_in_ancestor(fc->pid, controller, cgroup, NULL)) {
+ if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) {
/* this is just /cgroup/controller, return it as a dir */
sb->st_mode = S_IFDIR | 00555;
sb->st_nlink = 2;
sb->st_gid = k->gid;
sb->st_size = 0;
free_key(k);
- if (!caller_is_in_ancestor(fc->pid, controller, path1, NULL)) {
+ if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) {
ret = -ENOENT;
goto out;
}
}
}
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
if (cgroup) {
- if (!caller_may_see_dir(fc->pid, controller, cgroup))
+ if (!caller_may_see_dir(initpid, controller, cgroup))
return -ENOENT;
if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY))
return -EACCES;
goto out;
}
- if (!caller_is_in_ancestor(fc->pid, d->controller, d->cgroup, &nextcg)) {
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) {
if (nextcg) {
int ret;
ret = filler(buf, nextcg, NULL, 0);
}
free_key(k);
- if (!caller_may_see_dir(fc->pid, controller, path1)) {
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ if (!caller_may_see_dir(initpid, controller, path1)) {
ret = -ENOENT;
goto out;
}
static bool wait_for_sock(int sock, int timeout)
{
struct epoll_event ev;
- int epfd, ret;
+ int epfd, ret, now, starttime, deltatime, saved_errno;
+
+ if ((starttime = time(NULL)) < 0)
+ return false;
- epfd = epoll_create(1);
- if (epfd < 0) {
+ if ((epfd = epoll_create(1)) < 0) {
fprintf(stderr, "Failed to create epoll socket: %m\n");
return false;
}
return false;
}
- ret = epoll_wait(epfd, &ev, 1, timeout);
- close(epfd);
+again:
+ if ((now = time(NULL)) < 0) {
+ close(epfd);
+ return false;
+ }
- if (ret == 0)
+ deltatime = (starttime + timeout) - now;
+ if (deltatime < 0) { // timeout
+ errno = 0;
+ close(epfd);
return false;
- if (ret < 0) {
- fprintf(stderr, "Failure during epoll_wait: %m\n");
+ }
+ ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1);
+ if (ret < 0 && errno == EINTR)
+ goto again;
+ saved_errno = errno;
+ close(epfd);
+
+ if (ret <= 0) {
+ errno = saved_errno;
return false;
}
return true;
return recv(sockfd, buf, len, MSG_DONTWAIT);
}
-#define SEND_CREDS_OK 0
-#define SEND_CREDS_NOTSK 1
-#define SEND_CREDS_FAIL 2
static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst)
{
struct msghdr msg = { 0 };
}
// give the child 1 second to be done forking and
- // write it's ack
+ // write its ack
if (!wait_for_sock(cpipe[0], 1))
goto again;
ret = read(cpipe[0], &v, 1);
else
path1 = cgdir;
- if (!caller_is_in_ancestor(fc->pid, controller, path1, &next)) {
- if (last && strcmp(next, last) == 0)
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ if (!caller_is_in_ancestor(initpid, controller, path1, &next)) {
+ if (!next)
+ ret = -EINVAL;
+ else if (last && strcmp(next, last) == 0)
ret = -EEXIST;
else
ret = -ENOENT;
ret = -EACCES;
goto out;
}
- if (!caller_is_in_ancestor(fc->pid, controller, path1, NULL)) {
+ if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) {
ret = -EACCES;
goto out;
}
goto out;
}
- if (!caller_is_in_ancestor(fc->pid, controller, cgroup, &next)) {
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) {
if (!last || strcmp(next, last) == 0)
ret = -EBUSY;
else
ret = -EACCES;
goto out;
}
- if (!caller_is_in_ancestor(fc->pid, controller, cgroup, NULL)) {
+ if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) {
ret = -EACCES;
goto out;
}
rv = 0;
goto err;
}
- if (l < cache_size) {
- cache += l;
- cache_size -= l;
- total_len += l;
- } else {
- cache += cache_size;
- total_len += cache_size;
- cache_size = 0;
- break;
- }
+ cache += l;
+ cache_size -= l;
+ total_len += l;
}
d->size = total_len;
return total_len;
}
- cg = get_pid_cgroup(fc->pid, "memory");
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ cg = get_pid_cgroup(initpid, "memory");
if (!cg)
return read_file("/proc/meminfo", buf, size, d);
return total_len;
}
- cg = get_pid_cgroup(fc->pid, "cpuset");
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ cg = get_pid_cgroup(initpid, "cpuset");
if (!cg)
return read_file("proc/cpuinfo", buf, size, d);
rv = 0;
goto err;
}
- if (l < cache_size){
- cache += l;
- cache_size -= l;
- total_len += l;
- }else{
- cache += cache_size;
- total_len += cache_size;
- cache_size = 0;
- break;
- }
+ cache += l;
+ cache_size -= l;
+ total_len += l;
}
continue;
}
rv = 0;
goto err;
}
- if (l < cache_size) {
- cache += l;
- cache_size -= l;
- total_len += l;
- } else {
- cache += cache_size;
- total_len += cache_size;
- cache_size = 0;
- break;
- }
+ cache += l;
+ cache_size -= l;
+ total_len += l;
}
}
return total_len;
}
- cg = get_pid_cgroup(fc->pid, "cpuset");
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ cg = get_pid_cgroup(initpid, "cpuset");
if (!cg)
return read_file("/proc/stat", buf, size, d);
rv = 0;
goto err;
}
- if (l < cache_size) {
- cache += l;
- cache_size -= l;
- total_len += l;
- continue;
- } else {
- //no more space, break it
- cache += cache_size;
- total_len += cache_size;
- cache_size = 0;
- break;
- }
+ cache += l;
+ cache_size -= l;
+ total_len += l;
+ continue;
}
if (sscanf(cpu_char, "%d", &cpu) != 1)
int ret;
pid_t qpid;
- qpid = get_init_pid_for_task(pid);
- if (qpid < 0)
+ qpid = lookup_initpid_in_store(pid);
+ if (qpid <= 0)
return 0;
ret = snprintf(fnam, 100, "/proc/%d", qpid);
return time(NULL) - sb.st_ctime;
}
-/*
- * fork a task which switches to @task's namespace and writes '1'.
- * over a unix sock so we can read the task's reaper's pid in our
- * namespace
- */
-void write_task_init_pid_exit(int sock, pid_t target)
-{
- struct ucred cred;
- char fnam[100];
- pid_t pid;
- char v;
- int fd, ret;
-
- ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target);
- if (ret < 0 || ret >= sizeof(fnam))
- _exit(1);
-
- fd = open(fnam, O_RDONLY);
- if (fd < 0) {
- perror("write_task_init_pid_exit open of ns/pid");
- _exit(1);
- }
- if (setns(fd, 0)) {
- perror("write_task_init_pid_exit setns 1");
- close(fd);
- _exit(1);
- }
- pid = fork();
- if (pid < 0)
- _exit(1);
- if (pid != 0) {
- wait_for_pid(pid);
- _exit(0);
- }
-
- /* we are the child */
- cred.uid = 0;
- cred.gid = 0;
- cred.pid = 1;
- v = '1';
- send_creds(sock, &cred, v, true);
- _exit(0);
-}
-
-static pid_t get_init_pid_for_task(pid_t task)
-{
- int sock[2];
- pid_t pid;
- pid_t ret = -1;
- char v = '0';
- struct ucred cred;
-
- if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) {
- perror("socketpair");
- return -1;
- }
-
- pid = fork();
- if (pid < 0)
- goto out;
- if (!pid) {
- close(sock[1]);
- write_task_init_pid_exit(sock[0], task);
- }
-
- if (!recv_creds(sock[1], &cred, &v))
- goto out;
- ret = cred.pid;
-
-out:
- close(sock[0]);
- close(sock[1]);
- wait_for_pid(pid);
- return ret;
-}
-
static unsigned long get_reaper_busy(pid_t task)
{
- pid_t init = get_init_pid_for_task(task);
+ pid_t initpid = lookup_initpid_in_store(task);
char *cgroup = NULL, *usage_str = NULL;
unsigned long usage = 0;
- if (init == -1)
+ if (initpid <= 0)
return 0;
- cgroup = get_pid_cgroup(init, "cpuacct");
+ cgroup = get_pid_cgroup(initpid, "cpuacct");
if (!cgroup)
goto out;
if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str))
return total_len;
}
- cg = get_pid_cgroup(fc->pid, "blkio");
+ pid_t initpid = lookup_initpid_in_store(fc->pid);
+ if (initpid <= 0)
+ initpid = fc->pid;
+ cg = get_pid_cgroup(initpid, "blkio");
if (!cg)
return read_file("/proc/diskstats", buf, size, d);