3 * Copyright © 2014-2016 Canonical, Inc
4 * Author: Serge Hallyn <serge.hallyn@ubuntu.com>
6 * See COPYING file for details.
9 #define FUSE_USE_VERSION 26
24 #include <linux/sched.h>
25 #include <sys/param.h>
26 #include <sys/socket.h>
27 #include <sys/mount.h>
28 #include <sys/epoll.h>
33 #include "config.h" // for VERSION
38 LXC_TYPE_PROC_MEMINFO
,
39 LXC_TYPE_PROC_CPUINFO
,
42 LXC_TYPE_PROC_DISKSTATS
,
51 char *buf
; // unused as of yet
53 int size
; //actual data size
57 /* reserve buffer size, for cpuall in /proc/stat */
58 #define BUF_RESERVE_SIZE 256
61 * A table caching which pid is init for a pid namespace.
62 * When looking up which pid is init for $qpid, we first
63 * 1. Stat /proc/$qpid/ns/pid.
64 * 2. Check whether the ino_t is in our store.
65 * a. if not, fork a child in qpid's ns to send us
66 * ucred.pid = 1, and read the initpid. Cache
67 * initpid and creation time for /proc/initpid
68 * in a new store entry.
69 * b. if so, verify that /proc/initpid still matches
70 * what we have saved. If not, clear the store
71 * entry and go back to a. If so, return the
74 struct pidns_init_store
{
75 ino_t ino
; // inode number for /proc/$pid/ns/pid
76 pid_t initpid
; // the pid of nit in that ns
77 long int ctime
; // the time at which /proc/$initpid was created
78 struct pidns_init_store
*next
;
82 /* lol - look at how they are allocated in the kernel */
83 #define PIDNS_HASH_SIZE 4096
84 #define HASH(x) ((x) % PIDNS_HASH_SIZE)
86 static struct pidns_init_store
*pidns_hash_table
[PIDNS_HASH_SIZE
];
87 static pthread_mutex_t pidns_store_mutex
= PTHREAD_MUTEX_INITIALIZER
;
88 static void lock_mutex(pthread_mutex_t
*l
)
92 if ((ret
= pthread_mutex_lock(l
)) != 0) {
93 fprintf(stderr
, "pthread_mutex_lock returned:%d %s\n", ret
, strerror(ret
));
98 static void unlock_mutex(pthread_mutex_t
*l
)
102 if ((ret
= pthread_mutex_unlock(l
)) != 0) {
103 fprintf(stderr
, "pthread_mutex_unlock returned:%d %s\n", ret
, strerror(ret
));
108 static void store_lock(void)
110 lock_mutex(&pidns_store_mutex
);
113 static void store_unlock(void)
115 unlock_mutex(&pidns_store_mutex
);
118 /* Must be called under store_lock */
119 static bool initpid_still_valid(struct pidns_init_store
*e
, struct stat
*nsfdsb
)
124 snprintf(fnam
, 100, "/proc/%d", e
->initpid
);
125 if (stat(fnam
, &initsb
) < 0)
128 fprintf(stderr
, "comparing ctime %ld %ld for pid %d\n",
129 e
->ctime
, initsb
.st_ctime
, e
->initpid
);
131 if (e
->ctime
!= initsb
.st_ctime
)
136 /* Must be called under store_lock */
137 static void remove_initpid(struct pidns_init_store
*e
)
139 struct pidns_init_store
*tmp
;
143 fprintf(stderr
, "remove_initpid: removing entry for %d\n", e
->initpid
);
146 if (pidns_hash_table
[h
] == e
) {
147 pidns_hash_table
[h
] = e
->next
;
152 tmp
= pidns_hash_table
[h
];
154 if (tmp
->next
== e
) {
164 /* Must be called under store_lock */
165 static void prune_initpid_store(void)
167 static long int last_prune
= 0;
168 struct pidns_init_store
*e
, *prev
, *delme
;
169 long int now
, threshold
;
173 last_prune
= time(NULL
);
177 if (now
< last_prune
+ PURGE_SECS
)
180 fprintf(stderr
, "pruning\n");
183 threshold
= now
- 2 * PURGE_SECS
;
185 for (i
= 0; i
< PIDNS_HASH_SIZE
; i
++) {
186 for (prev
= NULL
, e
= pidns_hash_table
[i
]; e
; ) {
187 if (e
->lastcheck
< threshold
) {
189 fprintf(stderr
, "Removing cached entry for %d\n", e
->initpid
);
193 prev
->next
= e
->next
;
195 pidns_hash_table
[i
] = e
->next
;
206 /* Must be called under store_lock */
207 static void save_initpid(struct stat
*sb
, pid_t pid
)
209 struct pidns_init_store
*e
;
215 fprintf(stderr
, "save_initpid: adding entry for %d\n", pid
);
217 snprintf(fpath
, 100, "/proc/%d", pid
);
218 if (stat(fpath
, &procsb
) < 0)
221 e
= malloc(sizeof(*e
));
225 e
->ctime
= procsb
.st_ctime
;
227 e
->next
= pidns_hash_table
[h
];
228 e
->lastcheck
= time(NULL
);
229 pidns_hash_table
[h
] = e
;
233 * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store
234 * entry for the inode number and creation time. Verify that the init pid
235 * is still valid. If not, remove it. Return the entry if valid, NULL
237 * Must be called under store_lock
239 static struct pidns_init_store
*lookup_verify_initpid(struct stat
*sb
)
241 int h
= HASH(sb
->st_ino
);
242 struct pidns_init_store
*e
= pidns_hash_table
[h
];
245 if (e
->ino
== sb
->st_ino
) {
246 if (initpid_still_valid(e
, sb
)) {
247 e
->lastcheck
= time(NULL
);
259 static int is_dir(const char *path
)
262 int ret
= stat(path
, &statbuf
);
263 if (ret
== 0 && S_ISDIR(statbuf
.st_mode
))
268 static char *must_copy_string(const char *str
)
280 static inline void drop_trailing_newlines(char *s
)
284 for (l
=strlen(s
); l
>0 && s
[l
-1] == '\n'; l
--)
288 #define BATCH_SIZE 50
289 static void dorealloc(char **mem
, size_t oldlen
, size_t newlen
)
291 int newbatches
= (newlen
/ BATCH_SIZE
) + 1;
292 int oldbatches
= (oldlen
/ BATCH_SIZE
) + 1;
294 if (!*mem
|| newbatches
> oldbatches
) {
297 tmp
= realloc(*mem
, newbatches
* BATCH_SIZE
);
302 static void append_line(char **contents
, size_t *len
, char *line
, ssize_t linelen
)
304 size_t newlen
= *len
+ linelen
;
305 dorealloc(contents
, *len
, newlen
+ 1);
306 memcpy(*contents
+ *len
, line
, linelen
+1);
310 static char *slurp_file(const char *from
, int fd
)
313 char *contents
= NULL
;
314 FILE *f
= fdopen(fd
, "r");
315 size_t len
= 0, fulllen
= 0;
321 while ((linelen
= getline(&line
, &len
, f
)) != -1) {
322 append_line(&contents
, &fulllen
, line
, linelen
);
327 drop_trailing_newlines(contents
);
332 static bool write_string(const char *fnam
, const char *string
)
337 if (!(f
= fopen(fnam
, "w")))
339 len
= strlen(string
);
340 ret
= fwrite(string
, 1, len
, f
);
342 fprintf(stderr
, "Error writing to file: %s\n", strerror(errno
));
347 fprintf(stderr
, "Error writing to file: %s\n", strerror(errno
));
360 static bool store_hierarchy(char *stridx
, char *h
)
362 if (num_hierarchies
% ALLOC_NUM
== 0) {
363 size_t n
= (num_hierarchies
/ ALLOC_NUM
) + 1;
365 char **tmp
= realloc(hierarchies
, n
* sizeof(char *));
367 fprintf(stderr
, "Out of memory\n");
373 hierarchies
[num_hierarchies
++] = must_copy_string(h
);
377 static void print_subsystems(void)
381 fprintf(stderr
, "hierarchies:\n");
382 for (i
= 0; i
< num_hierarchies
; i
++) {
384 fprintf(stderr
, " %d: %s\n", i
, hierarchies
[i
]);
388 static bool in_comma_list(const char *needle
, const char *haystack
)
390 const char *s
= haystack
, *e
;
391 size_t nlen
= strlen(needle
);
393 while (*s
&& (e
= index(s
, ','))) {
398 if (strncmp(needle
, s
, nlen
) == 0)
402 if (strcmp(needle
, s
) == 0)
407 /* do we need to do any massaging here? I'm not sure... */
408 /* Return the mounted controller and store the corresponding open file descriptor
409 * referring to the controller mountpoint in the private lxcfs namespace in
412 static char *find_mounted_controller(const char *controller
, int *cfd
)
416 for (i
= 0; i
< num_hierarchies
; i
++) {
419 if (strcmp(hierarchies
[i
], controller
) == 0) {
420 *cfd
= fd_hierarchies
[i
];
421 return hierarchies
[i
];
423 if (in_comma_list(controller
, hierarchies
[i
])) {
424 *cfd
= fd_hierarchies
[i
];
425 return hierarchies
[i
];
432 bool cgfs_set_value(const char *controller
, const char *cgroup
, const char *file
,
437 char *fnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
441 /* BASEDIR / tmpc / cgroup / file \0 */
442 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cgroup
) + strlen(file
) + 4;
444 snprintf(fnam
, len
, "%s/%s/%s/%s", BASEDIR
, tmpc
, cgroup
, file
);
446 return write_string(fnam
, value
);
449 // Chown all the files in the cgroup directory. We do this when we create
450 // a cgroup on behalf of a user.
451 static void chown_all_cgroup_files(const char *dirname
, uid_t uid
, gid_t gid
)
453 struct dirent dirent
, *direntp
;
454 char path
[MAXPATHLEN
];
459 len
= strlen(dirname
);
460 if (len
>= MAXPATHLEN
) {
461 fprintf(stderr
, "chown_all_cgroup_files: pathname too long: %s\n", dirname
);
465 d
= opendir(dirname
);
467 fprintf(stderr
, "chown_all_cgroup_files: failed to open %s\n", dirname
);
471 while (readdir_r(d
, &dirent
, &direntp
) == 0 && direntp
) {
472 if (!strcmp(direntp
->d_name
, ".") || !strcmp(direntp
->d_name
, ".."))
474 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
475 if (ret
< 0 || ret
>= MAXPATHLEN
) {
476 fprintf(stderr
, "chown_all_cgroup_files: pathname too long under %s\n", dirname
);
479 if (chown(path
, uid
, gid
) < 0)
480 fprintf(stderr
, "Failed to chown file %s to %u:%u", path
, uid
, gid
);
485 int cgfs_create(const char *controller
, const char *cg
, uid_t uid
, gid_t gid
)
489 char *dirnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
493 /* BASEDIR / tmpc / cg \0 */
494 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cg
) + 3;
495 dirnam
= alloca(len
);
496 snprintf(dirnam
, len
, "%s/%s/%s", BASEDIR
,tmpc
, cg
);
498 if (mkdir(dirnam
, 0755) < 0)
501 if (uid
== 0 && gid
== 0)
504 if (chown(dirnam
, uid
, gid
) < 0)
507 chown_all_cgroup_files(dirnam
, uid
, gid
);
512 static bool recursive_rmdir(const char *dirname
)
514 struct dirent dirent
, *direntp
;
517 char pathname
[MAXPATHLEN
];
519 dir
= opendir(dirname
);
522 fprintf(stderr
, "%s: failed to open %s: %s\n", __func__
, dirname
, strerror(errno
));
527 while (!readdir_r(dir
, &dirent
, &direntp
)) {
534 if (!strcmp(direntp
->d_name
, ".") ||
535 !strcmp(direntp
->d_name
, ".."))
538 rc
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
539 if (rc
< 0 || rc
>= MAXPATHLEN
) {
540 fprintf(stderr
, "pathname too long\n");
544 ret
= lstat(pathname
, &mystat
);
547 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
551 if (S_ISDIR(mystat
.st_mode
)) {
552 if (!recursive_rmdir(pathname
)) {
554 fprintf(stderr
, "Error removing %s\n", pathname
);
561 if (closedir(dir
) < 0) {
562 fprintf(stderr
, "%s: failed to close directory %s: %s\n", __func__
, dirname
, strerror(errno
));
566 if (rmdir(dirname
) < 0) {
568 fprintf(stderr
, "%s: failed to delete %s: %s\n", __func__
, dirname
, strerror(errno
));
576 bool cgfs_remove(const char *controller
, const char *cg
)
580 char *dirnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
584 /* BASEDIR / tmpc / cg \0 */
585 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cg
) + 3;
586 dirnam
= alloca(len
);
587 snprintf(dirnam
, len
, "%s/%s/%s", BASEDIR
,tmpc
, cg
);
588 return recursive_rmdir(dirnam
);
591 bool cgfs_chmod_file(const char *controller
, const char *file
, mode_t mode
)
595 char *pathname
, *tmpc
= find_mounted_controller(controller
, &cfd
);
599 /* BASEDIR / tmpc / file \0 */
600 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(file
) + 3;
601 pathname
= alloca(len
);
602 snprintf(pathname
, len
, "%s/%s/%s", BASEDIR
, tmpc
, file
);
603 if (chmod(pathname
, mode
) < 0)
608 static int chown_tasks_files(const char *dirname
, uid_t uid
, gid_t gid
)
613 len
= strlen(dirname
) + strlen("/cgroup.procs") + 1;
615 snprintf(fname
, len
, "%s/tasks", dirname
);
616 if (chown(fname
, uid
, gid
) != 0)
618 snprintf(fname
, len
, "%s/cgroup.procs", dirname
);
619 if (chown(fname
, uid
, gid
) != 0)
624 int cgfs_chown_file(const char *controller
, const char *file
, uid_t uid
, gid_t gid
)
628 char *pathname
, *tmpc
= find_mounted_controller(controller
, &cfd
);
632 /* BASEDIR / tmpc / file \0 */
633 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(file
) + 3;
634 pathname
= alloca(len
);
635 snprintf(pathname
, len
, "%s/%s/%s", BASEDIR
, tmpc
, file
);
636 if (chown(pathname
, uid
, gid
) < 0)
639 if (is_dir(pathname
))
640 // like cgmanager did, we want to chown the tasks file as well
641 return chown_tasks_files(pathname
, uid
, gid
);
646 FILE *open_pids_file(const char *controller
, const char *cgroup
)
650 char *pathname
, *tmpc
= find_mounted_controller(controller
, &cfd
);
654 /* BASEDIR / tmpc / cgroup / "cgroup.procs" \0 */
655 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cgroup
) + 4 + strlen("cgroup.procs");
656 pathname
= alloca(len
);
657 snprintf(pathname
, len
, "%s/%s/%s/cgroup.procs", BASEDIR
, tmpc
, cgroup
);
658 return fopen(pathname
, "w");
661 static bool cgfs_iterate_cgroup(const char *controller
, const char *cgroup
, bool directories
,
662 void ***list
, size_t typesize
,
663 void* (*iterator
)(const char*, const char*, const char*))
668 char pathname
[MAXPATHLEN
];
669 size_t sz
= 0, asz
= 0;
670 struct dirent
*dirent
;
673 tmpc
= find_mounted_controller(controller
, &cfd
);
678 /* Make sure we pass a relative path to openat(). */
679 len
= strlen(cgroup
) + 1 /* . */ + 1 /* \0 */;
681 ret
= snprintf(cg
, len
, "%s%s", *cgroup
== '/' ? "." : "", cgroup
);
682 if (ret
< 0 || (size_t)ret
>= len
) {
683 fprintf(stderr
, "%s: pathname too long under %s\n", __func__
, cgroup
);
687 fd
= openat(cfd
, cg
, O_DIRECTORY
);
695 while ((dirent
= readdir(dir
))) {
698 if (!strcmp(dirent
->d_name
, ".") ||
699 !strcmp(dirent
->d_name
, ".."))
702 ret
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", cg
, dirent
->d_name
);
703 if (ret
< 0 || ret
>= MAXPATHLEN
) {
704 fprintf(stderr
, "%s: pathname too long under %s\n", __func__
, cg
);
708 ret
= fstatat(cfd
, pathname
, &mystat
, AT_SYMLINK_NOFOLLOW
);
710 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
713 if ((!directories
&& !S_ISREG(mystat
.st_mode
)) ||
714 (directories
&& !S_ISDIR(mystat
.st_mode
)))
721 tmp
= realloc(*list
, asz
* typesize
);
725 (*list
)[sz
] = (*iterator
)(controller
, cg
, dirent
->d_name
);
726 (*list
)[sz
+1] = NULL
;
729 if (closedir(dir
) < 0) {
730 fprintf(stderr
, "%s: failed closedir for %s: %s\n", __func__
, cgroup
, strerror(errno
));
736 static void *make_children_list_entry(const char *controller
, const char *cgroup
, const char *dir_entry
)
740 dup
= strdup(dir_entry
);
745 bool cgfs_list_children(const char *controller
, const char *cgroup
, char ***list
)
747 return cgfs_iterate_cgroup(controller
, cgroup
, true, (void***)list
, sizeof(*list
), &make_children_list_entry
);
750 void free_key(struct cgfs_files
*k
)
758 void free_keys(struct cgfs_files
**keys
)
764 for (i
= 0; keys
[i
]; i
++) {
770 bool cgfs_get_value(const char *controller
, const char *cgroup
, const char *file
, char **value
)
774 char *fnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
778 /* . + /cgroup + / + file + \0 */
779 len
= strlen(cgroup
) + strlen(file
) + 3;
781 ret
= snprintf(fnam
, len
, "%s%s/%s", *cgroup
== '/' ? "." : "", cgroup
, file
);
782 if (ret
< 0 || (size_t)ret
>= len
)
785 fd
= openat(cfd
, fnam
, O_RDONLY
);
789 *value
= slurp_file(fnam
, fd
);
790 return *value
!= NULL
;
793 struct cgfs_files
*cgfs_get_key(const char *controller
, const char *cgroup
, const char *file
)
797 char *fnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
799 struct cgfs_files
*newkey
;
804 if (file
&& *file
== '/')
807 if (file
&& index(file
, '/'))
810 /* . + /cgroup + / + file + \0 */
811 len
= strlen(cgroup
) + 3;
813 len
+= strlen(file
) + 1;
815 snprintf(fnam
, len
, "%s%s%s%s", *cgroup
== '/' ? "." : "", cgroup
,
816 file
? "/" : "", file
? file
: "");
818 ret
= fstatat(cfd
, fnam
, &sb
, 0);
823 newkey
= malloc(sizeof(struct cgfs_files
));
826 newkey
->name
= must_copy_string(file
);
827 else if (rindex(cgroup
, '/'))
828 newkey
->name
= must_copy_string(rindex(cgroup
, '/'));
830 newkey
->name
= must_copy_string(cgroup
);
831 newkey
->uid
= sb
.st_uid
;
832 newkey
->gid
= sb
.st_gid
;
833 newkey
->mode
= sb
.st_mode
;
838 static void *make_key_list_entry(const char *controller
, const char *cgroup
, const char *dir_entry
)
840 struct cgfs_files
*entry
= cgfs_get_key(controller
, cgroup
, dir_entry
);
842 fprintf(stderr
, "%s: Error getting files under %s:%s\n",
843 __func__
, controller
, cgroup
);
848 bool cgfs_list_keys(const char *controller
, const char *cgroup
, struct cgfs_files
***keys
)
850 return cgfs_iterate_cgroup(controller
, cgroup
, false, (void***)keys
, sizeof(*keys
), &make_key_list_entry
);
853 bool is_child_cgroup(const char *controller
, const char *cgroup
, const char *f
)
857 char *fnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
863 /* BASEDIR / tmpc / cgroup / f \0 */
864 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cgroup
) + strlen(f
) + 4;
866 snprintf(fnam
, len
, "%s/%s/%s/%s", BASEDIR
, tmpc
, cgroup
, f
);
868 ret
= stat(fnam
, &sb
);
869 if (ret
< 0 || !S_ISDIR(sb
.st_mode
))
874 #define SEND_CREDS_OK 0
875 #define SEND_CREDS_NOTSK 1
876 #define SEND_CREDS_FAIL 2
877 static bool recv_creds(int sock
, struct ucred
*cred
, char *v
);
878 static int wait_for_pid(pid_t pid
);
879 static int send_creds(int sock
, struct ucred
*cred
, char v
, bool pingfirst
);
880 static int send_creds_clone_wrapper(void *arg
);
883 * clone a task which switches to @task's namespace and writes '1'.
884 * over a unix sock so we can read the task's reaper's pid in our
887 * Note: glibc's fork() does not respect pidns, which can lead to failed
888 * assertions inside glibc (and thus failed forks) if the child's pid in
889 * the pidns and the parent pid outside are identical. Using clone prevents
892 static void write_task_init_pid_exit(int sock
, pid_t target
)
897 size_t stack_size
= sysconf(_SC_PAGESIZE
);
898 void *stack
= alloca(stack_size
);
900 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", (int)target
);
901 if (ret
< 0 || ret
>= sizeof(fnam
))
904 fd
= open(fnam
, O_RDONLY
);
906 perror("write_task_init_pid_exit open of ns/pid");
910 perror("write_task_init_pid_exit setns 1");
914 pid
= clone(send_creds_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &sock
);
918 if (!wait_for_pid(pid
))
924 static int send_creds_clone_wrapper(void *arg
) {
927 int sock
= *(int *)arg
;
929 /* we are the child */
934 if (send_creds(sock
, &cred
, v
, true) != SEND_CREDS_OK
)
939 static pid_t
get_init_pid_for_task(pid_t task
)
947 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
948 perror("socketpair");
957 write_task_init_pid_exit(sock
[0], task
);
961 if (!recv_creds(sock
[1], &cred
, &v
))
973 static pid_t
lookup_initpid_in_store(pid_t qpid
)
977 struct pidns_init_store
*e
;
980 snprintf(fnam
, 100, "/proc/%d/ns/pid", qpid
);
982 if (stat(fnam
, &sb
) < 0)
984 e
= lookup_verify_initpid(&sb
);
989 answer
= get_init_pid_for_task(qpid
);
991 save_initpid(&sb
, answer
);
994 /* we prune at end in case we are returning
995 * the value we were about to return */
996 prune_initpid_store();
1001 static int wait_for_pid(pid_t pid
)
1009 ret
= waitpid(pid
, &status
, 0);
1017 if (!WIFEXITED(status
) || WEXITSTATUS(status
) != 0)
1024 * append pid to *src.
1025 * src: a pointer to a char* in which ot append the pid.
1026 * sz: the number of characters printed so far, minus trailing \0.
1027 * asz: the allocated size so far
1028 * pid: the pid to append
1030 static void must_strcat_pid(char **src
, size_t *sz
, size_t *asz
, pid_t pid
)
1034 int tmplen
= sprintf(tmp
, "%d\n", (int)pid
);
1036 if (!*src
|| tmplen
+ *sz
+ 1 >= *asz
) {
1039 tmp
= realloc(*src
, *asz
+ BUF_RESERVE_SIZE
);
1042 *asz
+= BUF_RESERVE_SIZE
;
1044 memcpy((*src
) +*sz
, tmp
, tmplen
+1); /* include the \0 */
1049 * Given a open file * to /proc/pid/{u,g}id_map, and an id
1050 * valid in the caller's namespace, return the id mapped into
1052 * Returns the mapped id, or -1 on error.
1055 convert_id_to_ns(FILE *idfile
, unsigned int in_id
)
1057 unsigned int nsuid
, // base id for a range in the idfile's namespace
1058 hostuid
, // base id for a range in the caller's namespace
1059 count
; // number of ids in this range
1063 fseek(idfile
, 0L, SEEK_SET
);
1064 while (fgets(line
, 400, idfile
)) {
1065 ret
= sscanf(line
, "%u %u %u\n", &nsuid
, &hostuid
, &count
);
1068 if (hostuid
+ count
< hostuid
|| nsuid
+ count
< nsuid
) {
1070 * uids wrapped around - unexpected as this is a procfile,
1073 fprintf(stderr
, "pid wrapparound at entry %u %u %u in %s\n",
1074 nsuid
, hostuid
, count
, line
);
1077 if (hostuid
<= in_id
&& hostuid
+count
> in_id
) {
1079 * now since hostuid <= in_id < hostuid+count, and
1080 * hostuid+count and nsuid+count do not wrap around,
1081 * we know that nsuid+(in_id-hostuid) which must be
1082 * less that nsuid+(count) must not wrap around
1084 return (in_id
- hostuid
) + nsuid
;
1093 * for is_privileged_over,
1094 * specify whether we require the calling uid to be root in his
1097 #define NS_ROOT_REQD true
1098 #define NS_ROOT_OPT false
1102 static bool is_privileged_over(pid_t pid
, uid_t uid
, uid_t victim
, bool req_ns_root
)
1104 char fpath
[PROCLEN
];
1106 bool answer
= false;
1109 if (victim
== -1 || uid
== -1)
1113 * If the request is one not requiring root in the namespace,
1114 * then having the same uid suffices. (i.e. uid 1000 has write
1115 * access to files owned by uid 1000
1117 if (!req_ns_root
&& uid
== victim
)
1120 ret
= snprintf(fpath
, PROCLEN
, "/proc/%d/uid_map", pid
);
1121 if (ret
< 0 || ret
>= PROCLEN
)
1123 FILE *f
= fopen(fpath
, "r");
1127 /* if caller's not root in his namespace, reject */
1128 nsuid
= convert_id_to_ns(f
, uid
);
1133 * If victim is not mapped into caller's ns, reject.
1134 * XXX I'm not sure this check is needed given that fuse
1135 * will be sending requests where the vfs has converted
1137 nsuid
= convert_id_to_ns(f
, victim
);
1148 static bool perms_include(int fmode
, mode_t req_mode
)
1152 switch (req_mode
& O_ACCMODE
) {
1160 r
= S_IROTH
| S_IWOTH
;
1165 return ((fmode
& r
) == r
);
1171 * querycg is /a/b/c/d/e
1174 static char *get_next_cgroup_dir(const char *taskcg
, const char *querycg
)
1178 if (strlen(taskcg
) <= strlen(querycg
)) {
1179 fprintf(stderr
, "%s: I was fed bad input\n", __func__
);
1183 if (strcmp(querycg
, "/") == 0)
1184 start
= strdup(taskcg
+ 1);
1186 start
= strdup(taskcg
+ strlen(querycg
) + 1);
1189 end
= strchr(start
, '/');
1195 static void stripnewline(char *x
)
1197 size_t l
= strlen(x
);
1198 if (l
&& x
[l
-1] == '\n')
1202 static char *get_pid_cgroup(pid_t pid
, const char *contrl
)
1207 char *answer
= NULL
;
1211 const char *h
= find_mounted_controller(contrl
, &cfd
);
1215 ret
= snprintf(fnam
, PROCLEN
, "/proc/%d/cgroup", pid
);
1216 if (ret
< 0 || ret
>= PROCLEN
)
1218 if (!(f
= fopen(fnam
, "r")))
1221 while (getline(&line
, &len
, f
) != -1) {
1225 c1
= strchr(line
, ':');
1229 c2
= strchr(c1
, ':');
1233 if (strcmp(c1
, h
) != 0)
1238 answer
= strdup(c2
);
1250 * check whether a fuse context may access a cgroup dir or file
1252 * If file is not null, it is a cgroup file to check under cg.
1253 * If file is null, then we are checking perms on cg itself.
1255 * For files we can check the mode of the list_keys result.
1256 * For cgroups, we must make assumptions based on the files under the
1257 * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups
1260 static bool fc_may_access(struct fuse_context
*fc
, const char *contrl
, const char *cg
, const char *file
, mode_t mode
)
1262 struct cgfs_files
*k
= NULL
;
1265 k
= cgfs_get_key(contrl
, cg
, file
);
1269 if (is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_OPT
)) {
1270 if (perms_include(k
->mode
>> 6, mode
)) {
1275 if (fc
->gid
== k
->gid
) {
1276 if (perms_include(k
->mode
>> 3, mode
)) {
1281 ret
= perms_include(k
->mode
, mode
);
1288 #define INITSCOPE "/init.scope"
1289 static void prune_init_slice(char *cg
)
1292 size_t cg_len
= strlen(cg
), initscope_len
= strlen(INITSCOPE
);
1294 if (cg_len
< initscope_len
)
1297 point
= cg
+ cg_len
- initscope_len
;
1298 if (strcmp(point
, INITSCOPE
) == 0) {
1307 * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d.
1308 * If pid is in /a, he may act on /a/b, but not on /b.
1309 * if the answer is false and nextcg is not NULL, then *nextcg will point
1310 * to a string containing the next cgroup directory under cg, which must be
1311 * freed by the caller.
1313 static bool caller_is_in_ancestor(pid_t pid
, const char *contrl
, const char *cg
, char **nextcg
)
1315 bool answer
= false;
1316 char *c2
= get_pid_cgroup(pid
, contrl
);
1321 prune_init_slice(c2
);
1324 * callers pass in '/' for root cgroup, otherwise they pass
1325 * in a cgroup without leading '/'
1327 linecmp
= *cg
== '/' ? c2
: c2
+1;
1328 if (strncmp(linecmp
, cg
, strlen(linecmp
)) != 0) {
1330 *nextcg
= get_next_cgroup_dir(linecmp
, cg
);
1342 * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c.
1344 static bool caller_may_see_dir(pid_t pid
, const char *contrl
, const char *cg
)
1346 bool answer
= false;
1348 size_t target_len
, task_len
;
1350 if (strcmp(cg
, "/") == 0)
1353 c2
= get_pid_cgroup(pid
, contrl
);
1356 prune_init_slice(c2
);
1359 target_len
= strlen(cg
);
1360 task_len
= strlen(task_cg
);
1361 if (task_len
== 0) {
1362 /* Task is in the root cg, it can see everything. This case is
1363 * not handled by the strmcps below, since they test for the
1364 * last /, but that is the first / that we've chopped off
1370 if (strcmp(cg
, task_cg
) == 0) {
1374 if (target_len
< task_len
) {
1375 /* looking up a parent dir */
1376 if (strncmp(task_cg
, cg
, target_len
) == 0 && task_cg
[target_len
] == '/')
1380 if (target_len
> task_len
) {
1381 /* looking up a child dir */
1382 if (strncmp(task_cg
, cg
, task_len
) == 0 && cg
[task_len
] == '/')
1393 * given /cgroup/freezer/a/b, return "freezer".
1394 * the returned char* should NOT be freed.
1396 static char *pick_controller_from_path(struct fuse_context
*fc
, const char *path
)
1399 char *contr
, *slash
;
1401 if (strlen(path
) < 9)
1403 if (*(path
+7) != '/')
1406 contr
= strdupa(p1
);
1409 slash
= strstr(contr
, "/");
1414 for (i
= 0; i
< num_hierarchies
; i
++) {
1415 if (hierarchies
[i
] && strcmp(hierarchies
[i
], contr
) == 0)
1416 return hierarchies
[i
];
1422 * Find the start of cgroup in /cgroup/controller/the/cgroup/path
1423 * Note that the returned value may include files (keynames) etc
1425 static const char *find_cgroup_in_path(const char *path
)
1429 if (strlen(path
) < 9)
1431 p1
= strstr(path
+8, "/");
1438 * split the last path element from the path in @cg.
1439 * @dir is newly allocated and should be freed, @last not
1441 static void get_cgdir_and_path(const char *cg
, char **dir
, char **last
)
1448 *last
= strrchr(cg
, '/');
1453 p
= strrchr(*dir
, '/');
1458 * FUSE ops for /cgroup
1461 int cg_getattr(const char *path
, struct stat
*sb
)
1463 struct timespec now
;
1464 struct fuse_context
*fc
= fuse_get_context();
1465 char * cgdir
= NULL
;
1466 char *last
= NULL
, *path1
, *path2
;
1467 struct cgfs_files
*k
= NULL
;
1469 const char *controller
= NULL
;
1476 memset(sb
, 0, sizeof(struct stat
));
1478 if (clock_gettime(CLOCK_REALTIME
, &now
) < 0)
1481 sb
->st_uid
= sb
->st_gid
= 0;
1482 sb
->st_atim
= sb
->st_mtim
= sb
->st_ctim
= now
;
1485 if (strcmp(path
, "/cgroup") == 0) {
1486 sb
->st_mode
= S_IFDIR
| 00755;
1491 controller
= pick_controller_from_path(fc
, path
);
1494 cgroup
= find_cgroup_in_path(path
);
1496 /* this is just /cgroup/controller, return it as a dir */
1497 sb
->st_mode
= S_IFDIR
| 00755;
1502 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1512 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1515 /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys.
1516 * Then check that caller's cgroup is under path if last is a child
1517 * cgroup, or cgdir if last is a file */
1519 if (is_child_cgroup(controller
, path1
, path2
)) {
1520 if (!caller_may_see_dir(initpid
, controller
, cgroup
)) {
1524 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, NULL
)) {
1525 /* this is just /cgroup/controller, return it as a dir */
1526 sb
->st_mode
= S_IFDIR
| 00555;
1531 if (!fc_may_access(fc
, controller
, cgroup
, NULL
, O_RDONLY
)) {
1536 // get uid, gid, from '/tasks' file and make up a mode
1537 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
1538 sb
->st_mode
= S_IFDIR
| 00755;
1539 k
= cgfs_get_key(controller
, cgroup
, NULL
);
1541 sb
->st_uid
= sb
->st_gid
= 0;
1543 sb
->st_uid
= k
->uid
;
1544 sb
->st_gid
= k
->gid
;
1552 if ((k
= cgfs_get_key(controller
, path1
, path2
)) != NULL
) {
1553 sb
->st_mode
= S_IFREG
| k
->mode
;
1555 sb
->st_uid
= k
->uid
;
1556 sb
->st_gid
= k
->gid
;
1559 if (!caller_is_in_ancestor(initpid
, controller
, path1
, NULL
)) {
1563 if (!fc_may_access(fc
, controller
, path1
, path2
, O_RDONLY
)) {
1576 int cg_opendir(const char *path
, struct fuse_file_info
*fi
)
1578 struct fuse_context
*fc
= fuse_get_context();
1580 struct file_info
*dir_info
;
1581 char *controller
= NULL
;
1586 if (strcmp(path
, "/cgroup") == 0) {
1590 // return list of keys for the controller, and list of child cgroups
1591 controller
= pick_controller_from_path(fc
, path
);
1595 cgroup
= find_cgroup_in_path(path
);
1597 /* this is just /cgroup/controller, return its contents */
1602 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1606 if (!caller_may_see_dir(initpid
, controller
, cgroup
))
1608 if (!fc_may_access(fc
, controller
, cgroup
, NULL
, O_RDONLY
))
1612 /* we'll free this at cg_releasedir */
1613 dir_info
= malloc(sizeof(*dir_info
));
1616 dir_info
->controller
= must_copy_string(controller
);
1617 dir_info
->cgroup
= must_copy_string(cgroup
);
1618 dir_info
->type
= LXC_TYPE_CGDIR
;
1619 dir_info
->buf
= NULL
;
1620 dir_info
->file
= NULL
;
1621 dir_info
->buflen
= 0;
1623 fi
->fh
= (unsigned long)dir_info
;
1627 int cg_readdir(const char *path
, void *buf
, fuse_fill_dir_t filler
, off_t offset
,
1628 struct fuse_file_info
*fi
)
1630 struct file_info
*d
= (struct file_info
*)fi
->fh
;
1631 struct cgfs_files
**list
= NULL
;
1633 char *nextcg
= NULL
;
1634 struct fuse_context
*fc
= fuse_get_context();
1635 char **clist
= NULL
;
1637 if (d
->type
!= LXC_TYPE_CGDIR
) {
1638 fprintf(stderr
, "Internal error: file cache info used in readdir\n");
1641 if (!d
->cgroup
&& !d
->controller
) {
1642 // ls /var/lib/lxcfs/cgroup - just show list of controllers
1645 for (i
= 0; i
< num_hierarchies
; i
++) {
1646 if (hierarchies
[i
] && filler(buf
, hierarchies
[i
], NULL
, 0) != 0) {
1653 if (!cgfs_list_keys(d
->controller
, d
->cgroup
, &list
)) {
1654 // not a valid cgroup
1659 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1662 if (!caller_is_in_ancestor(initpid
, d
->controller
, d
->cgroup
, &nextcg
)) {
1664 ret
= filler(buf
, nextcg
, NULL
, 0);
1675 for (i
= 0; list
[i
]; i
++) {
1676 if (filler(buf
, list
[i
]->name
, NULL
, 0) != 0) {
1682 // now get the list of child cgroups
1684 if (!cgfs_list_children(d
->controller
, d
->cgroup
, &clist
)) {
1689 for (i
= 0; clist
[i
]; i
++) {
1690 if (filler(buf
, clist
[i
], NULL
, 0) != 0) {
1701 for (i
= 0; clist
[i
]; i
++)
1708 static void do_release_file_info(struct fuse_file_info
*fi
)
1710 struct file_info
*f
= (struct file_info
*)fi
->fh
;
1717 free(f
->controller
);
1718 f
->controller
= NULL
;
1728 int cg_releasedir(const char *path
, struct fuse_file_info
*fi
)
1730 do_release_file_info(fi
);
1734 int cg_open(const char *path
, struct fuse_file_info
*fi
)
1737 char *last
= NULL
, *path1
, *path2
, * cgdir
= NULL
, *controller
;
1738 struct cgfs_files
*k
= NULL
;
1739 struct file_info
*file_info
;
1740 struct fuse_context
*fc
= fuse_get_context();
1746 controller
= pick_controller_from_path(fc
, path
);
1749 cgroup
= find_cgroup_in_path(path
);
1753 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1762 k
= cgfs_get_key(controller
, path1
, path2
);
1769 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1772 if (!caller_may_see_dir(initpid
, controller
, path1
)) {
1776 if (!fc_may_access(fc
, controller
, path1
, path2
, fi
->flags
)) {
1781 /* we'll free this at cg_release */
1782 file_info
= malloc(sizeof(*file_info
));
1787 file_info
->controller
= must_copy_string(controller
);
1788 file_info
->cgroup
= must_copy_string(path1
);
1789 file_info
->file
= must_copy_string(path2
);
1790 file_info
->type
= LXC_TYPE_CGFILE
;
1791 file_info
->buf
= NULL
;
1792 file_info
->buflen
= 0;
1794 fi
->fh
= (unsigned long)file_info
;
1802 int cg_access(const char *path
, int mode
)
1805 char *last
= NULL
, *path1
, *path2
, * cgdir
= NULL
, *controller
;
1806 struct cgfs_files
*k
= NULL
;
1807 struct fuse_context
*fc
= fuse_get_context();
1813 controller
= pick_controller_from_path(fc
, path
);
1816 cgroup
= find_cgroup_in_path(path
);
1818 // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not
1819 if ((mode
& W_OK
) == 0)
1824 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1833 k
= cgfs_get_key(controller
, path1
, path2
);
1835 if ((mode
& W_OK
) == 0)
1843 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1846 if (!caller_may_see_dir(initpid
, controller
, path1
)) {
1850 if (!fc_may_access(fc
, controller
, path1
, path2
, mode
)) {
1862 int cg_release(const char *path
, struct fuse_file_info
*fi
)
1864 do_release_file_info(fi
);
1868 #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP )
1870 static bool wait_for_sock(int sock
, int timeout
)
1872 struct epoll_event ev
;
1873 int epfd
, ret
, now
, starttime
, deltatime
, saved_errno
;
1875 if ((starttime
= time(NULL
)) < 0)
1878 if ((epfd
= epoll_create(1)) < 0) {
1879 fprintf(stderr
, "Failed to create epoll socket: %m\n");
1883 ev
.events
= POLLIN_SET
;
1885 if (epoll_ctl(epfd
, EPOLL_CTL_ADD
, sock
, &ev
) < 0) {
1886 fprintf(stderr
, "Failed adding socket to epoll: %m\n");
1892 if ((now
= time(NULL
)) < 0) {
1897 deltatime
= (starttime
+ timeout
) - now
;
1898 if (deltatime
< 0) { // timeout
1903 ret
= epoll_wait(epfd
, &ev
, 1, 1000*deltatime
+ 1);
1904 if (ret
< 0 && errno
== EINTR
)
1906 saved_errno
= errno
;
1910 errno
= saved_errno
;
1916 static int msgrecv(int sockfd
, void *buf
, size_t len
)
1918 if (!wait_for_sock(sockfd
, 2))
1920 return recv(sockfd
, buf
, len
, MSG_DONTWAIT
);
1923 static int send_creds(int sock
, struct ucred
*cred
, char v
, bool pingfirst
)
1925 struct msghdr msg
= { 0 };
1927 struct cmsghdr
*cmsg
;
1928 char cmsgbuf
[CMSG_SPACE(sizeof(*cred
))];
1933 if (msgrecv(sock
, buf
, 1) != 1) {
1934 fprintf(stderr
, "%s: Error getting reply from server over socketpair\n",
1936 return SEND_CREDS_FAIL
;
1940 msg
.msg_control
= cmsgbuf
;
1941 msg
.msg_controllen
= sizeof(cmsgbuf
);
1943 cmsg
= CMSG_FIRSTHDR(&msg
);
1944 cmsg
->cmsg_len
= CMSG_LEN(sizeof(struct ucred
));
1945 cmsg
->cmsg_level
= SOL_SOCKET
;
1946 cmsg
->cmsg_type
= SCM_CREDENTIALS
;
1947 memcpy(CMSG_DATA(cmsg
), cred
, sizeof(*cred
));
1949 msg
.msg_name
= NULL
;
1950 msg
.msg_namelen
= 0;
1954 iov
.iov_len
= sizeof(buf
);
1958 if (sendmsg(sock
, &msg
, 0) < 0) {
1959 fprintf(stderr
, "%s: failed at sendmsg: %s\n", __func__
,
1962 return SEND_CREDS_NOTSK
;
1963 return SEND_CREDS_FAIL
;
1966 return SEND_CREDS_OK
;
1969 static bool recv_creds(int sock
, struct ucred
*cred
, char *v
)
1971 struct msghdr msg
= { 0 };
1973 struct cmsghdr
*cmsg
;
1974 char cmsgbuf
[CMSG_SPACE(sizeof(*cred
))];
1985 if (setsockopt(sock
, SOL_SOCKET
, SO_PASSCRED
, &optval
, sizeof(optval
)) == -1) {
1986 fprintf(stderr
, "Failed to set passcred: %s\n", strerror(errno
));
1990 if (write(sock
, buf
, 1) != 1) {
1991 fprintf(stderr
, "Failed to start write on scm fd: %s\n", strerror(errno
));
1995 msg
.msg_name
= NULL
;
1996 msg
.msg_namelen
= 0;
1997 msg
.msg_control
= cmsgbuf
;
1998 msg
.msg_controllen
= sizeof(cmsgbuf
);
2001 iov
.iov_len
= sizeof(buf
);
2005 if (!wait_for_sock(sock
, 2)) {
2006 fprintf(stderr
, "Timed out waiting for scm_cred: %s\n",
2010 ret
= recvmsg(sock
, &msg
, MSG_DONTWAIT
);
2012 fprintf(stderr
, "Failed to receive scm_cred: %s\n",
2017 cmsg
= CMSG_FIRSTHDR(&msg
);
2019 if (cmsg
&& cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
)) &&
2020 cmsg
->cmsg_level
== SOL_SOCKET
&&
2021 cmsg
->cmsg_type
== SCM_CREDENTIALS
) {
2022 memcpy(cred
, CMSG_DATA(cmsg
), sizeof(*cred
));
2029 struct pid_ns_clone_args
{
2033 int (*wrapped
) (int, pid_t
); // pid_from_ns or pid_to_ns
2037 * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage
2038 * with clone(). This simply writes '1' as ACK back to the parent
2039 * before calling the actual wrapped function.
2041 static int pid_ns_clone_wrapper(void *arg
) {
2042 struct pid_ns_clone_args
* args
= (struct pid_ns_clone_args
*) arg
;
2045 close(args
->cpipe
[0]);
2046 if (write(args
->cpipe
[1], &b
, sizeof(char)) < 0) {
2047 fprintf(stderr
, "%s (child): error on write: %s\n",
2048 __func__
, strerror(errno
));
2050 close(args
->cpipe
[1]);
2051 return args
->wrapped(args
->sock
, args
->tpid
);
2055 * pid_to_ns - reads pids from a ucred over a socket, then writes the
2056 * int value back over the socket. This shifts the pid from the
2057 * sender's pidns into tpid's pidns.
2059 static int pid_to_ns(int sock
, pid_t tpid
)
2064 while (recv_creds(sock
, &cred
, &v
)) {
2067 if (write(sock
, &cred
.pid
, sizeof(pid_t
)) != sizeof(pid_t
))
2075 * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain
2076 * in your old pidns. Only children which you clone will be in the target
2077 * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to
2078 * actually convert pids.
2080 * Note: glibc's fork() does not respect pidns, which can lead to failed
2081 * assertions inside glibc (and thus failed forks) if the child's pid in
2082 * the pidns and the parent pid outside are identical. Using clone prevents
2085 static void pid_to_ns_wrapper(int sock
, pid_t tpid
)
2087 int newnsfd
= -1, ret
, cpipe
[2];
2092 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", tpid
);
2093 if (ret
< 0 || ret
>= sizeof(fnam
))
2095 newnsfd
= open(fnam
, O_RDONLY
);
2098 if (setns(newnsfd
, 0) < 0)
2102 if (pipe(cpipe
) < 0)
2105 struct pid_ns_clone_args args
= {
2109 .wrapped
= &pid_to_ns
2111 size_t stack_size
= sysconf(_SC_PAGESIZE
);
2112 void *stack
= alloca(stack_size
);
2114 cpid
= clone(pid_ns_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &args
);
2118 // give the child 1 second to be done forking and
2120 if (!wait_for_sock(cpipe
[0], 1))
2122 ret
= read(cpipe
[0], &v
, 1);
2123 if (ret
!= sizeof(char) || v
!= '1')
2126 if (!wait_for_pid(cpid
))
2132 * To read cgroup files with a particular pid, we will setns into the child
2133 * pidns, open a pipe, fork a child - which will be the first to really be in
2134 * the child ns - which does the cgfs_get_value and writes the data to the pipe.
2136 bool do_read_pids(pid_t tpid
, const char *contrl
, const char *cg
, const char *file
, char **d
)
2138 int sock
[2] = {-1, -1};
2139 char *tmpdata
= NULL
;
2141 pid_t qpid
, cpid
= -1;
2142 bool answer
= false;
2145 size_t sz
= 0, asz
= 0;
2147 if (!cgfs_get_value(contrl
, cg
, file
, &tmpdata
))
2151 * Now we read the pids from returned data one by one, pass
2152 * them into a child in the target namespace, read back the
2153 * translated pids, and put them into our to-return data
2156 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
2157 perror("socketpair");
2166 if (!cpid
) // child - exits when done
2167 pid_to_ns_wrapper(sock
[1], tpid
);
2169 char *ptr
= tmpdata
;
2172 while (sscanf(ptr
, "%d\n", &qpid
) == 1) {
2174 ret
= send_creds(sock
[0], &cred
, v
, true);
2176 if (ret
== SEND_CREDS_NOTSK
)
2178 if (ret
== SEND_CREDS_FAIL
)
2181 // read converted results
2182 if (!wait_for_sock(sock
[0], 2)) {
2183 fprintf(stderr
, "%s: timed out waiting for pid from child: %s\n",
2184 __func__
, strerror(errno
));
2187 if (read(sock
[0], &qpid
, sizeof(qpid
)) != sizeof(qpid
)) {
2188 fprintf(stderr
, "%s: error reading pid from child: %s\n",
2189 __func__
, strerror(errno
));
2192 must_strcat_pid(d
, &sz
, &asz
, qpid
);
2194 ptr
= strchr(ptr
, '\n');
2200 cred
.pid
= getpid();
2202 if (send_creds(sock
[0], &cred
, v
, true) != SEND_CREDS_OK
) {
2203 // failed to ask child to exit
2204 fprintf(stderr
, "%s: failed to ask child to exit: %s\n",
2205 __func__
, strerror(errno
));
2215 if (sock
[0] != -1) {
2222 int cg_read(const char *path
, char *buf
, size_t size
, off_t offset
,
2223 struct fuse_file_info
*fi
)
2225 struct fuse_context
*fc
= fuse_get_context();
2226 struct file_info
*f
= (struct file_info
*)fi
->fh
;
2227 struct cgfs_files
*k
= NULL
;
2232 if (f
->type
!= LXC_TYPE_CGFILE
) {
2233 fprintf(stderr
, "Internal error: directory cache info used in cg_read\n");
2246 if ((k
= cgfs_get_key(f
->controller
, f
->cgroup
, f
->file
)) == NULL
) {
2252 if (!fc_may_access(fc
, f
->controller
, f
->cgroup
, f
->file
, O_RDONLY
)) {
2257 if (strcmp(f
->file
, "tasks") == 0 ||
2258 strcmp(f
->file
, "/tasks") == 0 ||
2259 strcmp(f
->file
, "/cgroup.procs") == 0 ||
2260 strcmp(f
->file
, "cgroup.procs") == 0)
2261 // special case - we have to translate the pids
2262 r
= do_read_pids(fc
->pid
, f
->controller
, f
->cgroup
, f
->file
, &data
);
2264 r
= cgfs_get_value(f
->controller
, f
->cgroup
, f
->file
, &data
);
2278 memcpy(buf
, data
, s
);
2279 if (s
> 0 && s
< size
&& data
[s
-1] != '\n')
2289 static int pid_from_ns(int sock
, pid_t tpid
)
2299 if (!wait_for_sock(sock
, 2)) {
2300 fprintf(stderr
, "%s: timeout reading from parent\n", __func__
);
2303 if ((ret
= read(sock
, &vpid
, sizeof(pid_t
))) != sizeof(pid_t
)) {
2304 fprintf(stderr
, "%s: bad read from parent: %s\n",
2305 __func__
, strerror(errno
));
2308 if (vpid
== -1) // done
2312 if (send_creds(sock
, &cred
, v
, true) != SEND_CREDS_OK
) {
2314 cred
.pid
= getpid();
2315 if (send_creds(sock
, &cred
, v
, false) != SEND_CREDS_OK
)
2322 static void pid_from_ns_wrapper(int sock
, pid_t tpid
)
2324 int newnsfd
= -1, ret
, cpipe
[2];
2329 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", tpid
);
2330 if (ret
< 0 || ret
>= sizeof(fnam
))
2332 newnsfd
= open(fnam
, O_RDONLY
);
2335 if (setns(newnsfd
, 0) < 0)
2339 if (pipe(cpipe
) < 0)
2342 struct pid_ns_clone_args args
= {
2346 .wrapped
= &pid_from_ns
2348 size_t stack_size
= sysconf(_SC_PAGESIZE
);
2349 void *stack
= alloca(stack_size
);
2351 cpid
= clone(pid_ns_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &args
);
2355 // give the child 1 second to be done forking and
2357 if (!wait_for_sock(cpipe
[0], 1))
2359 ret
= read(cpipe
[0], &v
, 1);
2360 if (ret
!= sizeof(char) || v
!= '1')
2363 if (!wait_for_pid(cpid
))
2369 * Given host @uid, return the uid to which it maps in
2370 * @pid's user namespace, or -1 if none.
2372 bool hostuid_to_ns(uid_t uid
, pid_t pid
, uid_t
*answer
)
2377 sprintf(line
, "/proc/%d/uid_map", pid
);
2378 if ((f
= fopen(line
, "r")) == NULL
) {
2382 *answer
= convert_id_to_ns(f
, uid
);
2391 * get_pid_creds: get the real uid and gid of @pid from
2393 * (XXX should we use euid here?)
2395 void get_pid_creds(pid_t pid
, uid_t
*uid
, gid_t
*gid
)
2404 sprintf(line
, "/proc/%d/status", pid
);
2405 if ((f
= fopen(line
, "r")) == NULL
) {
2406 fprintf(stderr
, "Error opening %s: %s\n", line
, strerror(errno
));
2409 while (fgets(line
, 400, f
)) {
2410 if (strncmp(line
, "Uid:", 4) == 0) {
2411 if (sscanf(line
+4, "%u", &u
) != 1) {
2412 fprintf(stderr
, "bad uid line for pid %u\n", pid
);
2417 } else if (strncmp(line
, "Gid:", 4) == 0) {
2418 if (sscanf(line
+4, "%u", &g
) != 1) {
2419 fprintf(stderr
, "bad gid line for pid %u\n", pid
);
2430 * May the requestor @r move victim @v to a new cgroup?
2431 * This is allowed if
2432 * . they are the same task
2433 * . they are ownedy by the same uid
2434 * . @r is root on the host, or
2435 * . @v's uid is mapped into @r's where @r is root.
2437 bool may_move_pid(pid_t r
, uid_t r_uid
, pid_t v
)
2439 uid_t v_uid
, tmpuid
;
2446 get_pid_creds(v
, &v_uid
, &v_gid
);
2449 if (hostuid_to_ns(r_uid
, r
, &tmpuid
) && tmpuid
== 0
2450 && hostuid_to_ns(v_uid
, r
, &tmpuid
))
2455 static bool do_write_pids(pid_t tpid
, uid_t tuid
, const char *contrl
, const char *cg
,
2456 const char *file
, const char *buf
)
2458 int sock
[2] = {-1, -1};
2459 pid_t qpid
, cpid
= -1;
2460 FILE *pids_file
= NULL
;
2461 bool answer
= false, fail
= false;
2463 pids_file
= open_pids_file(contrl
, cg
);
2468 * write the pids to a socket, have helper in writer's pidns
2469 * call movepid for us
2471 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
2472 perror("socketpair");
2480 if (!cpid
) { // child
2482 pid_from_ns_wrapper(sock
[1], tpid
);
2485 const char *ptr
= buf
;
2486 while (sscanf(ptr
, "%d", &qpid
) == 1) {
2490 if (write(sock
[0], &qpid
, sizeof(qpid
)) != sizeof(qpid
)) {
2491 fprintf(stderr
, "%s: error writing pid to child: %s\n",
2492 __func__
, strerror(errno
));
2496 if (recv_creds(sock
[0], &cred
, &v
)) {
2498 if (!may_move_pid(tpid
, tuid
, cred
.pid
)) {
2502 if (fprintf(pids_file
, "%d", (int) cred
.pid
) < 0)
2507 ptr
= strchr(ptr
, '\n');
2513 /* All good, write the value */
2515 if (write(sock
[0], &qpid
,sizeof(qpid
)) != sizeof(qpid
))
2516 fprintf(stderr
, "Warning: failed to ask child to exit\n");
2524 if (sock
[0] != -1) {
2529 if (fclose(pids_file
) != 0)
2535 int cg_write(const char *path
, const char *buf
, size_t size
, off_t offset
,
2536 struct fuse_file_info
*fi
)
2538 struct fuse_context
*fc
= fuse_get_context();
2539 char *localbuf
= NULL
;
2540 struct cgfs_files
*k
= NULL
;
2541 struct file_info
*f
= (struct file_info
*)fi
->fh
;
2544 if (f
->type
!= LXC_TYPE_CGFILE
) {
2545 fprintf(stderr
, "Internal error: directory cache info used in cg_write\n");
2555 localbuf
= alloca(size
+1);
2556 localbuf
[size
] = '\0';
2557 memcpy(localbuf
, buf
, size
);
2559 if ((k
= cgfs_get_key(f
->controller
, f
->cgroup
, f
->file
)) == NULL
) {
2564 if (!fc_may_access(fc
, f
->controller
, f
->cgroup
, f
->file
, O_WRONLY
)) {
2569 if (strcmp(f
->file
, "tasks") == 0 ||
2570 strcmp(f
->file
, "/tasks") == 0 ||
2571 strcmp(f
->file
, "/cgroup.procs") == 0 ||
2572 strcmp(f
->file
, "cgroup.procs") == 0)
2573 // special case - we have to translate the pids
2574 r
= do_write_pids(fc
->pid
, fc
->uid
, f
->controller
, f
->cgroup
, f
->file
, localbuf
);
2576 r
= cgfs_set_value(f
->controller
, f
->cgroup
, f
->file
, localbuf
);
2586 int cg_chown(const char *path
, uid_t uid
, gid_t gid
)
2588 struct fuse_context
*fc
= fuse_get_context();
2589 char *cgdir
= NULL
, *last
= NULL
, *path1
, *path2
, *controller
;
2590 struct cgfs_files
*k
= NULL
;
2597 if (strcmp(path
, "/cgroup") == 0)
2600 controller
= pick_controller_from_path(fc
, path
);
2603 cgroup
= find_cgroup_in_path(path
);
2605 /* this is just /cgroup/controller */
2608 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2618 if (is_child_cgroup(controller
, path1
, path2
)) {
2619 // get uid, gid, from '/tasks' file and make up a mode
2620 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
2621 k
= cgfs_get_key(controller
, cgroup
, "tasks");
2624 k
= cgfs_get_key(controller
, path1
, path2
);
2632 * This being a fuse request, the uid and gid must be valid
2633 * in the caller's namespace. So we can just check to make
2634 * sure that the caller is root in his uid, and privileged
2635 * over the file's current owner.
2637 if (!is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_REQD
)) {
2642 ret
= cgfs_chown_file(controller
, cgroup
, uid
, gid
);
2651 int cg_chmod(const char *path
, mode_t mode
)
2653 struct fuse_context
*fc
= fuse_get_context();
2654 char * cgdir
= NULL
, *last
= NULL
, *path1
, *path2
, *controller
;
2655 struct cgfs_files
*k
= NULL
;
2662 if (strcmp(path
, "/cgroup") == 0)
2665 controller
= pick_controller_from_path(fc
, path
);
2668 cgroup
= find_cgroup_in_path(path
);
2670 /* this is just /cgroup/controller */
2673 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2683 if (is_child_cgroup(controller
, path1
, path2
)) {
2684 // get uid, gid, from '/tasks' file and make up a mode
2685 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
2686 k
= cgfs_get_key(controller
, cgroup
, "tasks");
2689 k
= cgfs_get_key(controller
, path1
, path2
);
2697 * This being a fuse request, the uid and gid must be valid
2698 * in the caller's namespace. So we can just check to make
2699 * sure that the caller is root in his uid, and privileged
2700 * over the file's current owner.
2702 if (!is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_OPT
)) {
2707 if (!cgfs_chmod_file(controller
, cgroup
, mode
)) {
2719 int cg_mkdir(const char *path
, mode_t mode
)
2721 struct fuse_context
*fc
= fuse_get_context();
2722 char *last
= NULL
, *path1
, *cgdir
= NULL
, *controller
, *next
= NULL
;
2730 controller
= pick_controller_from_path(fc
, path
);
2734 cgroup
= find_cgroup_in_path(path
);
2738 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2744 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2747 if (!caller_is_in_ancestor(initpid
, controller
, path1
, &next
)) {
2750 else if (last
&& strcmp(next
, last
) == 0)
2757 if (!fc_may_access(fc
, controller
, path1
, NULL
, O_RDWR
)) {
2761 if (!caller_is_in_ancestor(initpid
, controller
, path1
, NULL
)) {
2766 ret
= cgfs_create(controller
, cgroup
, fc
->uid
, fc
->gid
);
2774 int cg_rmdir(const char *path
)
2776 struct fuse_context
*fc
= fuse_get_context();
2777 char *last
= NULL
, *cgdir
= NULL
, *controller
, *next
= NULL
;
2784 controller
= pick_controller_from_path(fc
, path
);
2788 cgroup
= find_cgroup_in_path(path
);
2792 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2798 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2801 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, &next
)) {
2802 if (!last
|| strcmp(next
, last
) == 0)
2809 if (!fc_may_access(fc
, controller
, cgdir
, NULL
, O_WRONLY
)) {
2813 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, NULL
)) {
2818 if (!cgfs_remove(controller
, cgroup
)) {
2831 static bool startswith(const char *line
, const char *pref
)
2833 if (strncmp(line
, pref
, strlen(pref
)) == 0)
2838 static void get_mem_cached(char *memstat
, unsigned long *v
)
2844 if (startswith(memstat
, "total_cache")) {
2845 sscanf(memstat
+ 11, "%lu", v
);
2849 eol
= strchr(memstat
, '\n');
2856 static void get_blkio_io_value(char *str
, unsigned major
, unsigned minor
, char *iotype
, unsigned long *v
)
2862 snprintf(key
, 32, "%u:%u %s", major
, minor
, iotype
);
2864 size_t len
= strlen(key
);
2868 if (startswith(str
, key
)) {
2869 sscanf(str
+ len
, "%lu", v
);
2872 eol
= strchr(str
, '\n');
2879 static int read_file(const char *path
, char *buf
, size_t size
,
2880 struct file_info
*d
)
2882 size_t linelen
= 0, total_len
= 0, rv
= 0;
2884 char *cache
= d
->buf
;
2885 size_t cache_size
= d
->buflen
;
2886 FILE *f
= fopen(path
, "r");
2890 while (getline(&line
, &linelen
, f
) != -1) {
2891 ssize_t l
= snprintf(cache
, cache_size
, "%s", line
);
2893 perror("Error writing to cache");
2897 if (l
>= cache_size
) {
2898 fprintf(stderr
, "Internal error: truncated write to cache\n");
2907 d
->size
= total_len
;
2908 if (total_len
> size
)
2911 /* read from off 0 */
2912 memcpy(buf
, d
->buf
, total_len
);
2921 * FUSE ops for /proc
2924 static unsigned long get_memlimit(const char *cgroup
)
2926 char *memlimit_str
= NULL
;
2927 unsigned long memlimit
= -1;
2929 if (cgfs_get_value("memory", cgroup
, "memory.limit_in_bytes", &memlimit_str
))
2930 memlimit
= strtoul(memlimit_str
, NULL
, 10);
2937 static unsigned long get_min_memlimit(const char *cgroup
)
2939 char *copy
= strdupa(cgroup
);
2940 unsigned long memlimit
= 0, retlimit
;
2942 retlimit
= get_memlimit(copy
);
2944 while (strcmp(copy
, "/") != 0) {
2945 copy
= dirname(copy
);
2946 memlimit
= get_memlimit(copy
);
2947 if (memlimit
!= -1 && memlimit
< retlimit
)
2948 retlimit
= memlimit
;
2954 static int proc_meminfo_read(char *buf
, size_t size
, off_t offset
,
2955 struct fuse_file_info
*fi
)
2957 struct fuse_context
*fc
= fuse_get_context();
2958 struct file_info
*d
= (struct file_info
*)fi
->fh
;
2960 char *memusage_str
= NULL
, *memstat_str
= NULL
,
2961 *memswlimit_str
= NULL
, *memswusage_str
= NULL
,
2962 *memswlimit_default_str
= NULL
, *memswusage_default_str
= NULL
;
2963 unsigned long memlimit
= 0, memusage
= 0, memswlimit
= 0, memswusage
= 0,
2964 cached
= 0, hosttotal
= 0;
2966 size_t linelen
= 0, total_len
= 0, rv
= 0;
2967 char *cache
= d
->buf
;
2968 size_t cache_size
= d
->buflen
;
2972 if (offset
> d
->size
)
2976 int left
= d
->size
- offset
;
2977 total_len
= left
> size
? size
: left
;
2978 memcpy(buf
, cache
+ offset
, total_len
);
2982 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2985 cg
= get_pid_cgroup(initpid
, "memory");
2987 return read_file("/proc/meminfo", buf
, size
, d
);
2988 prune_init_slice(cg
);
2990 memlimit
= get_min_memlimit(cg
);
2991 if (!cgfs_get_value("memory", cg
, "memory.usage_in_bytes", &memusage_str
))
2993 if (!cgfs_get_value("memory", cg
, "memory.stat", &memstat_str
))
2996 // Following values are allowed to fail, because swapaccount might be turned
2997 // off for current kernel
2998 if(cgfs_get_value("memory", cg
, "memory.memsw.limit_in_bytes", &memswlimit_str
) &&
2999 cgfs_get_value("memory", cg
, "memory.memsw.usage_in_bytes", &memswusage_str
))
3001 /* If swapaccounting is turned on, then default value is assumed to be that of cgroup / */
3002 if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str
))
3004 if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str
))
3007 memswlimit
= strtoul(memswlimit_str
, NULL
, 10);
3008 memswusage
= strtoul(memswusage_str
, NULL
, 10);
3010 if (!strcmp(memswlimit_str
, memswlimit_default_str
))
3012 if (!strcmp(memswusage_str
, memswusage_default_str
))
3015 memswlimit
= memswlimit
/ 1024;
3016 memswusage
= memswusage
/ 1024;
3019 memusage
= strtoul(memusage_str
, NULL
, 10);
3023 get_mem_cached(memstat_str
, &cached
);
3025 f
= fopen("/proc/meminfo", "r");
3029 while (getline(&line
, &linelen
, f
) != -1) {
3031 char *printme
, lbuf
[100];
3033 memset(lbuf
, 0, 100);
3034 if (startswith(line
, "MemTotal:")) {
3035 sscanf(line
+14, "%lu", &hosttotal
);
3036 if (hosttotal
< memlimit
)
3037 memlimit
= hosttotal
;
3038 snprintf(lbuf
, 100, "MemTotal: %8lu kB\n", memlimit
);
3040 } else if (startswith(line
, "MemFree:")) {
3041 snprintf(lbuf
, 100, "MemFree: %8lu kB\n", memlimit
- memusage
);
3043 } else if (startswith(line
, "MemAvailable:")) {
3044 snprintf(lbuf
, 100, "MemAvailable: %8lu kB\n", memlimit
- memusage
);
3046 } else if (startswith(line
, "SwapTotal:") && memswlimit
> 0) {
3047 snprintf(lbuf
, 100, "SwapTotal: %8lu kB\n", memswlimit
- memlimit
);
3049 } else if (startswith(line
, "SwapFree:") && memswlimit
> 0 && memswusage
> 0) {
3050 snprintf(lbuf
, 100, "SwapFree: %8lu kB\n",
3051 (memswlimit
- memlimit
) - (memswusage
- memusage
));
3053 } else if (startswith(line
, "Slab:")) {
3054 snprintf(lbuf
, 100, "Slab: %8lu kB\n", 0UL);
3056 } else if (startswith(line
, "Buffers:")) {
3057 snprintf(lbuf
, 100, "Buffers: %8lu kB\n", 0UL);
3059 } else if (startswith(line
, "Cached:")) {
3060 snprintf(lbuf
, 100, "Cached: %8lu kB\n", cached
);
3062 } else if (startswith(line
, "SwapCached:")) {
3063 snprintf(lbuf
, 100, "SwapCached: %8lu kB\n", 0UL);
3068 l
= snprintf(cache
, cache_size
, "%s", printme
);
3070 perror("Error writing to cache");
3075 if (l
>= cache_size
) {
3076 fprintf(stderr
, "Internal error: truncated write to cache\n");
3087 d
->size
= total_len
;
3088 if (total_len
> size
) total_len
= size
;
3089 memcpy(buf
, d
->buf
, total_len
);
3098 free(memswlimit_str
);
3099 free(memswusage_str
);
3101 free(memswlimit_default_str
);
3102 free(memswusage_default_str
);
3107 * Read the cpuset.cpus for cg
3108 * Return the answer in a newly allocated string which must be freed
3110 static char *get_cpuset(const char *cg
)
3114 if (!cgfs_get_value("cpuset", cg
, "cpuset.cpus", &answer
))
3119 bool cpu_in_cpuset(int cpu
, const char *cpuset
);
3121 static bool cpuline_in_cpuset(const char *line
, const char *cpuset
)
3125 if (sscanf(line
, "processor : %d", &cpu
) != 1)
3127 return cpu_in_cpuset(cpu
, cpuset
);
3131 * check whether this is a '^processor" line in /proc/cpuinfo
3133 static bool is_processor_line(const char *line
)
3137 if (sscanf(line
, "processor : %d", &cpu
) == 1)
3142 static int proc_cpuinfo_read(char *buf
, size_t size
, off_t offset
,
3143 struct fuse_file_info
*fi
)
3145 struct fuse_context
*fc
= fuse_get_context();
3146 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3148 char *cpuset
= NULL
;
3150 size_t linelen
= 0, total_len
= 0, rv
= 0;
3151 bool am_printing
= false, firstline
= true, is_s390x
= false;
3152 int curcpu
= -1, cpu
;
3153 char *cache
= d
->buf
;
3154 size_t cache_size
= d
->buflen
;
3158 if (offset
> d
->size
)
3162 int left
= d
->size
- offset
;
3163 total_len
= left
> size
? size
: left
;
3164 memcpy(buf
, cache
+ offset
, total_len
);
3168 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3171 cg
= get_pid_cgroup(initpid
, "cpuset");
3173 return read_file("proc/cpuinfo", buf
, size
, d
);
3174 prune_init_slice(cg
);
3176 cpuset
= get_cpuset(cg
);
3180 f
= fopen("/proc/cpuinfo", "r");
3184 while (getline(&line
, &linelen
, f
) != -1) {
3188 if (strstr(line
, "IBM/S390") != NULL
) {
3194 if (strncmp(line
, "# processors:", 12) == 0)
3196 if (is_processor_line(line
)) {
3197 am_printing
= cpuline_in_cpuset(line
, cpuset
);
3200 l
= snprintf(cache
, cache_size
, "processor : %d\n", curcpu
);
3202 perror("Error writing to cache");
3206 if (l
>= cache_size
) {
3207 fprintf(stderr
, "Internal error: truncated write to cache\n");
3216 } else if (is_s390x
&& sscanf(line
, "processor %d:", &cpu
) == 1) {
3218 if (!cpu_in_cpuset(cpu
, cpuset
))
3221 p
= strchr(line
, ':');
3225 l
= snprintf(cache
, cache_size
, "processor %d:%s", curcpu
, p
);
3227 perror("Error writing to cache");
3231 if (l
>= cache_size
) {
3232 fprintf(stderr
, "Internal error: truncated write to cache\n");
3243 l
= snprintf(cache
, cache_size
, "%s", line
);
3245 perror("Error writing to cache");
3249 if (l
>= cache_size
) {
3250 fprintf(stderr
, "Internal error: truncated write to cache\n");
3261 char *origcache
= d
->buf
;
3264 d
->buf
= malloc(d
->buflen
);
3267 cache_size
= d
->buflen
;
3269 l
= snprintf(cache
, cache_size
, "vendor_id : IBM/S390\n");
3270 if (l
< 0 || l
>= cache_size
) {
3277 l
= snprintf(cache
, cache_size
, "# processors : %d\n", curcpu
+ 1);
3278 if (l
< 0 || l
>= cache_size
) {
3285 l
= snprintf(cache
, cache_size
, "%s", origcache
);
3287 if (l
< 0 || l
>= cache_size
)
3293 d
->size
= total_len
;
3294 if (total_len
> size
) total_len
= size
;
3296 /* read from off 0 */
3297 memcpy(buf
, d
->buf
, total_len
);
3308 static int proc_stat_read(char *buf
, size_t size
, off_t offset
,
3309 struct fuse_file_info
*fi
)
3311 struct fuse_context
*fc
= fuse_get_context();
3312 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3314 char *cpuset
= NULL
;
3316 size_t linelen
= 0, total_len
= 0, rv
= 0;
3317 int curcpu
= -1; /* cpu numbering starts at 0 */
3318 unsigned long user
= 0, nice
= 0, system
= 0, idle
= 0, iowait
= 0, irq
= 0, softirq
= 0, steal
= 0, guest
= 0;
3319 unsigned long user_sum
= 0, nice_sum
= 0, system_sum
= 0, idle_sum
= 0, iowait_sum
= 0,
3320 irq_sum
= 0, softirq_sum
= 0, steal_sum
= 0, guest_sum
= 0;
3321 #define CPUALL_MAX_SIZE BUF_RESERVE_SIZE
3322 char cpuall
[CPUALL_MAX_SIZE
];
3323 /* reserve for cpu all */
3324 char *cache
= d
->buf
+ CPUALL_MAX_SIZE
;
3325 size_t cache_size
= d
->buflen
- CPUALL_MAX_SIZE
;
3329 if (offset
> d
->size
)
3333 int left
= d
->size
- offset
;
3334 total_len
= left
> size
? size
: left
;
3335 memcpy(buf
, d
->buf
+ offset
, total_len
);
3339 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3342 cg
= get_pid_cgroup(initpid
, "cpuset");
3344 return read_file("/proc/stat", buf
, size
, d
);
3345 prune_init_slice(cg
);
3347 cpuset
= get_cpuset(cg
);
3351 f
= fopen("/proc/stat", "r");
3356 if (getline(&line
, &linelen
, f
) < 0) {
3357 fprintf(stderr
, "proc_stat_read read first line failed\n");
3361 while (getline(&line
, &linelen
, f
) != -1) {
3364 char cpu_char
[10]; /* That's a lot of cores */
3367 if (sscanf(line
, "cpu%9[^ ]", cpu_char
) != 1) {
3368 /* not a ^cpuN line containing a number N, just print it */
3369 l
= snprintf(cache
, cache_size
, "%s", line
);
3371 perror("Error writing to cache");
3375 if (l
>= cache_size
) {
3376 fprintf(stderr
, "Internal error: truncated write to cache\n");
3386 if (sscanf(cpu_char
, "%d", &cpu
) != 1)
3388 if (!cpu_in_cpuset(cpu
, cpuset
))
3392 c
= strchr(line
, ' ');
3395 l
= snprintf(cache
, cache_size
, "cpu%d%s", curcpu
, c
);
3397 perror("Error writing to cache");
3402 if (l
>= cache_size
) {
3403 fprintf(stderr
, "Internal error: truncated write to cache\n");
3412 if (sscanf(line
, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user
, &nice
, &system
, &idle
, &iowait
, &irq
,
3413 &softirq
, &steal
, &guest
) != 9)
3417 system_sum
+= system
;
3419 iowait_sum
+= iowait
;
3421 softirq_sum
+= softirq
;
3428 int cpuall_len
= snprintf(cpuall
, CPUALL_MAX_SIZE
, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
3429 "cpu ", user_sum
, nice_sum
, system_sum
, idle_sum
, iowait_sum
, irq_sum
, softirq_sum
, steal_sum
, guest_sum
);
3430 if (cpuall_len
> 0 && cpuall_len
< CPUALL_MAX_SIZE
){
3431 memcpy(cache
, cpuall
, cpuall_len
);
3432 cache
+= cpuall_len
;
3434 /* shouldn't happen */
3435 fprintf(stderr
, "proc_stat_read copy cpuall failed, cpuall_len=%d\n", cpuall_len
);
3439 memmove(cache
, d
->buf
+ CPUALL_MAX_SIZE
, total_len
);
3440 total_len
+= cpuall_len
;
3442 d
->size
= total_len
;
3443 if (total_len
> size
) total_len
= size
;
3445 memcpy(buf
, d
->buf
, total_len
);
3457 static long int getreaperage(pid_t pid
)
3464 qpid
= lookup_initpid_in_store(pid
);
3468 ret
= snprintf(fnam
, 100, "/proc/%d", qpid
);
3469 if (ret
< 0 || ret
>= 100)
3472 if (lstat(fnam
, &sb
) < 0)
3475 return time(NULL
) - sb
.st_ctime
;
3478 static unsigned long get_reaper_busy(pid_t task
)
3480 pid_t initpid
= lookup_initpid_in_store(task
);
3481 char *cgroup
= NULL
, *usage_str
= NULL
;
3482 unsigned long usage
= 0;
3487 cgroup
= get_pid_cgroup(initpid
, "cpuacct");
3490 prune_init_slice(cgroup
);
3491 if (!cgfs_get_value("cpuacct", cgroup
, "cpuacct.usage", &usage_str
))
3493 usage
= strtoul(usage_str
, NULL
, 10);
3494 usage
/= 1000000000;
3505 char *name
, *cwd
= get_current_dir_name();
3511 len
= strlen(cwd
) + strlen("/iwashere") + 1;
3513 snprintf(name
, len
, "%s/iwashere", cwd
);
3515 fd
= creat(name
, 0755);
3522 * We read /proc/uptime and reuse its second field.
3523 * For the first field, we use the mtime for the reaper for
3524 * the calling pid as returned by getreaperage
3526 static int proc_uptime_read(char *buf
, size_t size
, off_t offset
,
3527 struct fuse_file_info
*fi
)
3529 struct fuse_context
*fc
= fuse_get_context();
3530 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3531 long int reaperage
= getreaperage(fc
->pid
);
3532 unsigned long int busytime
= get_reaper_busy(fc
->pid
), idletime
;
3533 char *cache
= d
->buf
;
3534 ssize_t total_len
= 0;
3541 if (offset
> d
->size
)
3545 int left
= d
->size
- offset
;
3546 total_len
= left
> size
? size
: left
;
3547 memcpy(buf
, cache
+ offset
, total_len
);
3551 idletime
= reaperage
- busytime
;
3552 if (idletime
> reaperage
)
3553 idletime
= reaperage
;
3555 total_len
= snprintf(d
->buf
, d
->size
, "%ld.0 %lu.0\n", reaperage
, idletime
);
3557 perror("Error writing to cache");
3561 d
->size
= (int)total_len
;
3564 if (total_len
> size
) total_len
= size
;
3566 memcpy(buf
, d
->buf
, total_len
);
3570 static int proc_diskstats_read(char *buf
, size_t size
, off_t offset
,
3571 struct fuse_file_info
*fi
)
3574 struct fuse_context
*fc
= fuse_get_context();
3575 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3577 char *io_serviced_str
= NULL
, *io_merged_str
= NULL
, *io_service_bytes_str
= NULL
,
3578 *io_wait_time_str
= NULL
, *io_service_time_str
= NULL
;
3579 unsigned long read
= 0, write
= 0;
3580 unsigned long read_merged
= 0, write_merged
= 0;
3581 unsigned long read_sectors
= 0, write_sectors
= 0;
3582 unsigned long read_ticks
= 0, write_ticks
= 0;
3583 unsigned long ios_pgr
= 0, tot_ticks
= 0, rq_ticks
= 0;
3584 unsigned long rd_svctm
= 0, wr_svctm
= 0, rd_wait
= 0, wr_wait
= 0;
3585 char *cache
= d
->buf
;
3586 size_t cache_size
= d
->buflen
;
3588 size_t linelen
= 0, total_len
= 0, rv
= 0;
3589 unsigned int major
= 0, minor
= 0;
3594 if (offset
> d
->size
)
3598 int left
= d
->size
- offset
;
3599 total_len
= left
> size
? size
: left
;
3600 memcpy(buf
, cache
+ offset
, total_len
);
3604 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3607 cg
= get_pid_cgroup(initpid
, "blkio");
3609 return read_file("/proc/diskstats", buf
, size
, d
);
3610 prune_init_slice(cg
);
3612 if (!cgfs_get_value("blkio", cg
, "blkio.io_serviced_recursive", &io_serviced_str
))
3614 if (!cgfs_get_value("blkio", cg
, "blkio.io_merged_recursive", &io_merged_str
))
3616 if (!cgfs_get_value("blkio", cg
, "blkio.io_service_bytes_recursive", &io_service_bytes_str
))
3618 if (!cgfs_get_value("blkio", cg
, "blkio.io_wait_time_recursive", &io_wait_time_str
))
3620 if (!cgfs_get_value("blkio", cg
, "blkio.io_service_time_recursive", &io_service_time_str
))
3624 f
= fopen("/proc/diskstats", "r");
3628 while (getline(&line
, &linelen
, f
) != -1) {
3632 i
= sscanf(line
, "%u %u %71s", &major
, &minor
, dev_name
);
3636 get_blkio_io_value(io_serviced_str
, major
, minor
, "Read", &read
);
3637 get_blkio_io_value(io_serviced_str
, major
, minor
, "Write", &write
);
3638 get_blkio_io_value(io_merged_str
, major
, minor
, "Read", &read_merged
);
3639 get_blkio_io_value(io_merged_str
, major
, minor
, "Write", &write_merged
);
3640 get_blkio_io_value(io_service_bytes_str
, major
, minor
, "Read", &read_sectors
);
3641 read_sectors
= read_sectors
/512;
3642 get_blkio_io_value(io_service_bytes_str
, major
, minor
, "Write", &write_sectors
);
3643 write_sectors
= write_sectors
/512;
3645 get_blkio_io_value(io_service_time_str
, major
, minor
, "Read", &rd_svctm
);
3646 rd_svctm
= rd_svctm
/1000000;
3647 get_blkio_io_value(io_wait_time_str
, major
, minor
, "Read", &rd_wait
);
3648 rd_wait
= rd_wait
/1000000;
3649 read_ticks
= rd_svctm
+ rd_wait
;
3651 get_blkio_io_value(io_service_time_str
, major
, minor
, "Write", &wr_svctm
);
3652 wr_svctm
= wr_svctm
/1000000;
3653 get_blkio_io_value(io_wait_time_str
, major
, minor
, "Write", &wr_wait
);
3654 wr_wait
= wr_wait
/1000000;
3655 write_ticks
= wr_svctm
+ wr_wait
;
3657 get_blkio_io_value(io_service_time_str
, major
, minor
, "Total", &tot_ticks
);
3658 tot_ticks
= tot_ticks
/1000000;
3660 memset(lbuf
, 0, 256);
3661 if (read
|| write
|| read_merged
|| write_merged
|| read_sectors
|| write_sectors
|| read_ticks
|| write_ticks
)
3662 snprintf(lbuf
, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
3663 major
, minor
, dev_name
, read
, read_merged
, read_sectors
, read_ticks
,
3664 write
, write_merged
, write_sectors
, write_ticks
, ios_pgr
, tot_ticks
, rq_ticks
);
3668 l
= snprintf(cache
, cache_size
, "%s", lbuf
);
3670 perror("Error writing to fuse buf");
3674 if (l
>= cache_size
) {
3675 fprintf(stderr
, "Internal error: truncated write to cache\n");
3685 d
->size
= total_len
;
3686 if (total_len
> size
) total_len
= size
;
3687 memcpy(buf
, d
->buf
, total_len
);
3695 free(io_serviced_str
);
3696 free(io_merged_str
);
3697 free(io_service_bytes_str
);
3698 free(io_wait_time_str
);
3699 free(io_service_time_str
);
3703 static int proc_swaps_read(char *buf
, size_t size
, off_t offset
,
3704 struct fuse_file_info
*fi
)
3706 struct fuse_context
*fc
= fuse_get_context();
3707 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3709 char *memswlimit_str
= NULL
, *memlimit_str
= NULL
, *memusage_str
= NULL
, *memswusage_str
= NULL
,
3710 *memswlimit_default_str
= NULL
, *memswusage_default_str
= NULL
;
3711 unsigned long memswlimit
= 0, memlimit
= 0, memusage
= 0, memswusage
= 0, swap_total
= 0, swap_free
= 0;
3712 ssize_t total_len
= 0, rv
= 0;
3714 char *cache
= d
->buf
;
3717 if (offset
> d
->size
)
3721 int left
= d
->size
- offset
;
3722 total_len
= left
> size
? size
: left
;
3723 memcpy(buf
, cache
+ offset
, total_len
);
3727 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3730 cg
= get_pid_cgroup(initpid
, "memory");
3732 return read_file("/proc/swaps", buf
, size
, d
);
3733 prune_init_slice(cg
);
3735 if (!cgfs_get_value("memory", cg
, "memory.limit_in_bytes", &memlimit_str
))
3738 if (!cgfs_get_value("memory", cg
, "memory.usage_in_bytes", &memusage_str
))
3741 memlimit
= strtoul(memlimit_str
, NULL
, 10);
3742 memusage
= strtoul(memusage_str
, NULL
, 10);
3744 if (cgfs_get_value("memory", cg
, "memory.memsw.usage_in_bytes", &memswusage_str
) &&
3745 cgfs_get_value("memory", cg
, "memory.memsw.limit_in_bytes", &memswlimit_str
)) {
3747 /* If swap accounting is turned on, then default value is assumed to be that of cgroup / */
3748 if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str
))
3750 if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str
))
3753 memswlimit
= strtoul(memswlimit_str
, NULL
, 10);
3754 memswusage
= strtoul(memswusage_str
, NULL
, 10);
3756 if (!strcmp(memswlimit_str
, memswlimit_default_str
))
3758 if (!strcmp(memswusage_str
, memswusage_default_str
))
3761 swap_total
= (memswlimit
- memlimit
) / 1024;
3762 swap_free
= (memswusage
- memusage
) / 1024;
3765 total_len
= snprintf(d
->buf
, d
->size
, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
3767 /* When no mem + swap limit is specified or swapaccount=0*/
3771 FILE *f
= fopen("/proc/meminfo", "r");
3776 while (getline(&line
, &linelen
, f
) != -1) {
3777 if (startswith(line
, "SwapTotal:")) {
3778 sscanf(line
, "SwapTotal: %8lu kB", &swap_total
);
3779 } else if (startswith(line
, "SwapFree:")) {
3780 sscanf(line
, "SwapFree: %8lu kB", &swap_free
);
3788 if (swap_total
> 0) {
3789 l
= snprintf(d
->buf
+ total_len
, d
->size
- total_len
,
3790 "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ",
3791 swap_total
, swap_free
);
3795 if (total_len
< 0 || l
< 0) {
3796 perror("Error writing to cache");
3802 d
->size
= (int)total_len
;
3804 if (total_len
> size
) total_len
= size
;
3805 memcpy(buf
, d
->buf
, total_len
);
3810 free(memswlimit_str
);
3813 free(memswusage_str
);
3814 free(memswusage_default_str
);
3815 free(memswlimit_default_str
);
3819 static off_t
get_procfile_size(const char *which
)
3821 FILE *f
= fopen(which
, "r");
3824 ssize_t sz
, answer
= 0;
3828 while ((sz
= getline(&line
, &len
, f
)) != -1)
3836 int proc_getattr(const char *path
, struct stat
*sb
)
3838 struct timespec now
;
3840 memset(sb
, 0, sizeof(struct stat
));
3841 if (clock_gettime(CLOCK_REALTIME
, &now
) < 0)
3843 sb
->st_uid
= sb
->st_gid
= 0;
3844 sb
->st_atim
= sb
->st_mtim
= sb
->st_ctim
= now
;
3845 if (strcmp(path
, "/proc") == 0) {
3846 sb
->st_mode
= S_IFDIR
| 00555;
3850 if (strcmp(path
, "/proc/meminfo") == 0 ||
3851 strcmp(path
, "/proc/cpuinfo") == 0 ||
3852 strcmp(path
, "/proc/uptime") == 0 ||
3853 strcmp(path
, "/proc/stat") == 0 ||
3854 strcmp(path
, "/proc/diskstats") == 0 ||
3855 strcmp(path
, "/proc/swaps") == 0) {
3857 sb
->st_mode
= S_IFREG
| 00444;
3865 int proc_readdir(const char *path
, void *buf
, fuse_fill_dir_t filler
, off_t offset
,
3866 struct fuse_file_info
*fi
)
3868 if (filler(buf
, "cpuinfo", NULL
, 0) != 0 ||
3869 filler(buf
, "meminfo", NULL
, 0) != 0 ||
3870 filler(buf
, "stat", NULL
, 0) != 0 ||
3871 filler(buf
, "uptime", NULL
, 0) != 0 ||
3872 filler(buf
, "diskstats", NULL
, 0) != 0 ||
3873 filler(buf
, "swaps", NULL
, 0) != 0)
3878 int proc_open(const char *path
, struct fuse_file_info
*fi
)
3881 struct file_info
*info
;
3883 if (strcmp(path
, "/proc/meminfo") == 0)
3884 type
= LXC_TYPE_PROC_MEMINFO
;
3885 else if (strcmp(path
, "/proc/cpuinfo") == 0)
3886 type
= LXC_TYPE_PROC_CPUINFO
;
3887 else if (strcmp(path
, "/proc/uptime") == 0)
3888 type
= LXC_TYPE_PROC_UPTIME
;
3889 else if (strcmp(path
, "/proc/stat") == 0)
3890 type
= LXC_TYPE_PROC_STAT
;
3891 else if (strcmp(path
, "/proc/diskstats") == 0)
3892 type
= LXC_TYPE_PROC_DISKSTATS
;
3893 else if (strcmp(path
, "/proc/swaps") == 0)
3894 type
= LXC_TYPE_PROC_SWAPS
;
3898 info
= malloc(sizeof(*info
));
3902 memset(info
, 0, sizeof(*info
));
3905 info
->buflen
= get_procfile_size(path
) + BUF_RESERVE_SIZE
;
3907 info
->buf
= malloc(info
->buflen
);
3908 } while (!info
->buf
);
3909 memset(info
->buf
, 0, info
->buflen
);
3910 /* set actual size to buffer size */
3911 info
->size
= info
->buflen
;
3913 fi
->fh
= (unsigned long)info
;
3917 int proc_access(const char *path
, int mask
)
3919 /* these are all read-only */
3920 if ((mask
& ~R_OK
) != 0)
3925 int proc_release(const char *path
, struct fuse_file_info
*fi
)
3927 do_release_file_info(fi
);
3931 int proc_read(const char *path
, char *buf
, size_t size
, off_t offset
,
3932 struct fuse_file_info
*fi
)
3934 struct file_info
*f
= (struct file_info
*) fi
->fh
;
3937 case LXC_TYPE_PROC_MEMINFO
:
3938 return proc_meminfo_read(buf
, size
, offset
, fi
);
3939 case LXC_TYPE_PROC_CPUINFO
:
3940 return proc_cpuinfo_read(buf
, size
, offset
, fi
);
3941 case LXC_TYPE_PROC_UPTIME
:
3942 return proc_uptime_read(buf
, size
, offset
, fi
);
3943 case LXC_TYPE_PROC_STAT
:
3944 return proc_stat_read(buf
, size
, offset
, fi
);
3945 case LXC_TYPE_PROC_DISKSTATS
:
3946 return proc_diskstats_read(buf
, size
, offset
, fi
);
3947 case LXC_TYPE_PROC_SWAPS
:
3948 return proc_swaps_read(buf
, size
, offset
, fi
);
3954 static void __attribute__((constructor
)) collect_subsystems(void)
3960 if ((f
= fopen("/proc/self/cgroup", "r")) == NULL
) {
3961 fprintf(stderr
, "Error opening /proc/self/cgroup: %s\n", strerror(errno
));
3964 while (getline(&line
, &len
, f
) != -1) {
3967 p
= strchr(line
, ':');
3972 p2
= strrchr(p
, ':');
3977 /* With cgroupv2 /proc/self/cgroup can contain entries of the
3978 * form: 0::/ This will cause lxcfs to fail the cgroup mounts
3979 * because it parses out the empty string "" and later on passes
3980 * it to mount(). Let's skip such entries.
3985 if (!store_hierarchy(line
, p
))
3996 static void __attribute__((destructor
)) free_subsystems(void)
4000 for (i
= 0; i
< num_hierarchies
; i
++)
4002 free(hierarchies
[i
]);