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
)
313 char *contents
= NULL
;
314 FILE *f
= fopen(from
, "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
));
354 * hierarchies, i.e. 'cpu,cpuacct'
366 static bool store_hierarchy(char *stridx
, char *h
)
368 if (num_hierarchies
% ALLOC_NUM
== 0) {
369 size_t n
= (num_hierarchies
/ ALLOC_NUM
) + 1;
371 char **tmp
= realloc(hierarchies
, n
* sizeof(char *));
373 fprintf(stderr
, "Out of memory\n");
379 hierarchies
[num_hierarchies
++] = must_copy_string(h
);
383 static void print_subsystems(void)
387 fprintf(stderr
, "hierarchies:");
388 for (i
= 0; i
< num_hierarchies
; i
++) {
390 fprintf(stderr
, " %d: %s\n", i
, hierarchies
[i
]);
394 static bool in_comma_list(const char *needle
, const char *haystack
)
396 const char *s
= haystack
, *e
;
397 size_t nlen
= strlen(needle
);
399 while (*s
&& (e
= index(s
, ','))) {
404 if (strncmp(needle
, s
, nlen
) == 0)
408 if (strcmp(needle
, s
) == 0)
413 /* do we need to do any massaging here? I'm not sure... */
414 static char *find_mounted_controller(const char *controller
)
418 for (i
= 0; i
< num_hierarchies
; i
++) {
421 if (strcmp(hierarchies
[i
], controller
) == 0)
422 return hierarchies
[i
];
423 if (in_comma_list(controller
, hierarchies
[i
]))
424 return hierarchies
[i
];
430 bool cgfs_set_value(const char *controller
, const char *cgroup
, const char *file
,
434 char *fnam
, *tmpc
= find_mounted_controller(controller
);
438 /* basedir / tmpc / cgroup / file \0 */
439 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + strlen(file
) + 4;
441 snprintf(fnam
, len
, "%s/%s/%s/%s", basedir
, tmpc
, cgroup
, file
);
443 return write_string(fnam
, value
);
446 // Chown all the files in the cgroup directory. We do this when we create
447 // a cgroup on behalf of a user.
448 static void chown_all_cgroup_files(const char *dirname
, uid_t uid
, gid_t gid
)
450 struct dirent dirent
, *direntp
;
451 char path
[MAXPATHLEN
];
456 len
= strlen(dirname
);
457 if (len
>= MAXPATHLEN
) {
458 fprintf(stderr
, "chown_all_cgroup_files: pathname too long: %s\n", dirname
);
462 d
= opendir(dirname
);
464 fprintf(stderr
, "chown_all_cgroup_files: failed to open %s\n", dirname
);
468 while (readdir_r(d
, &dirent
, &direntp
) == 0 && direntp
) {
469 if (!strcmp(direntp
->d_name
, ".") || !strcmp(direntp
->d_name
, ".."))
471 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
472 if (ret
< 0 || ret
>= MAXPATHLEN
) {
473 fprintf(stderr
, "chown_all_cgroup_files: pathname too long under %s\n", dirname
);
476 if (chown(path
, uid
, gid
) < 0)
477 fprintf(stderr
, "Failed to chown file %s to %u:%u", path
, uid
, gid
);
482 int cgfs_create(const char *controller
, const char *cg
, uid_t uid
, gid_t gid
)
485 char *dirnam
, *tmpc
= find_mounted_controller(controller
);
489 /* basedir / tmpc / cg \0 */
490 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cg
) + 3;
491 dirnam
= alloca(len
);
492 snprintf(dirnam
, len
, "%s/%s/%s", basedir
,tmpc
, cg
);
494 if (mkdir(dirnam
, 0755) < 0)
497 if (uid
== 0 && gid
== 0)
500 if (chown(dirnam
, uid
, gid
) < 0)
503 chown_all_cgroup_files(dirnam
, uid
, gid
);
508 static bool recursive_rmdir(const char *dirname
)
510 struct dirent dirent
, *direntp
;
513 char pathname
[MAXPATHLEN
];
515 dir
= opendir(dirname
);
518 fprintf(stderr
, "%s: failed to open %s: %s\n", __func__
, dirname
, strerror(errno
));
523 while (!readdir_r(dir
, &dirent
, &direntp
)) {
530 if (!strcmp(direntp
->d_name
, ".") ||
531 !strcmp(direntp
->d_name
, ".."))
534 rc
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
535 if (rc
< 0 || rc
>= MAXPATHLEN
) {
536 fprintf(stderr
, "pathname too long\n");
540 ret
= lstat(pathname
, &mystat
);
543 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
547 if (S_ISDIR(mystat
.st_mode
)) {
548 if (!recursive_rmdir(pathname
)) {
550 fprintf(stderr
, "Error removing %s\n", pathname
);
557 if (closedir(dir
) < 0) {
558 fprintf(stderr
, "%s: failed to close directory %s: %s\n", __func__
, dirname
, strerror(errno
));
562 if (rmdir(dirname
) < 0) {
564 fprintf(stderr
, "%s: failed to delete %s: %s\n", __func__
, dirname
, strerror(errno
));
572 bool cgfs_remove(const char *controller
, const char *cg
)
575 char *dirnam
, *tmpc
= find_mounted_controller(controller
);
579 /* basedir / tmpc / cg \0 */
580 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cg
) + 3;
581 dirnam
= alloca(len
);
582 snprintf(dirnam
, len
, "%s/%s/%s", basedir
,tmpc
, cg
);
583 return recursive_rmdir(dirnam
);
586 bool cgfs_chmod_file(const char *controller
, const char *file
, mode_t mode
)
589 char *pathname
, *tmpc
= find_mounted_controller(controller
);
593 /* basedir / tmpc / file \0 */
594 len
= strlen(basedir
) + strlen(tmpc
) + strlen(file
) + 3;
595 pathname
= alloca(len
);
596 snprintf(pathname
, len
, "%s/%s/%s", basedir
, tmpc
, file
);
597 if (chmod(pathname
, mode
) < 0)
602 static int chown_tasks_files(const char *dirname
, uid_t uid
, gid_t gid
)
607 len
= strlen(dirname
) + strlen("/cgroup.procs") + 1;
609 snprintf(fname
, len
, "%s/tasks", dirname
);
610 if (chown(fname
, uid
, gid
) != 0)
612 snprintf(fname
, len
, "%s/cgroup.procs", dirname
);
613 if (chown(fname
, uid
, gid
) != 0)
618 int cgfs_chown_file(const char *controller
, const char *file
, uid_t uid
, gid_t gid
)
621 char *pathname
, *tmpc
= find_mounted_controller(controller
);
625 /* basedir / tmpc / file \0 */
626 len
= strlen(basedir
) + strlen(tmpc
) + strlen(file
) + 3;
627 pathname
= alloca(len
);
628 snprintf(pathname
, len
, "%s/%s/%s", basedir
, tmpc
, file
);
629 if (chown(pathname
, uid
, gid
) < 0)
632 if (is_dir(pathname
))
633 // like cgmanager did, we want to chown the tasks file as well
634 return chown_tasks_files(pathname
, uid
, gid
);
639 FILE *open_pids_file(const char *controller
, const char *cgroup
)
642 char *pathname
, *tmpc
= find_mounted_controller(controller
);
646 /* basedir / tmpc / cgroup / "cgroup.procs" \0 */
647 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + 4 + strlen("cgroup.procs");
648 pathname
= alloca(len
);
649 snprintf(pathname
, len
, "%s/%s/%s/cgroup.procs", basedir
, tmpc
, cgroup
);
650 return fopen(pathname
, "w");
653 static bool cgfs_iterate_cgroup(const char *controller
, const char *cgroup
, bool directories
,
654 void ***list
, size_t typesize
,
655 void* (*iterator
)(const char*, const char*, const char*))
658 char *dirname
, *tmpc
= find_mounted_controller(controller
);
659 char pathname
[MAXPATHLEN
];
660 size_t sz
= 0, asz
= 0;
661 struct dirent dirent
, *direntp
;
669 /* basedir / tmpc / cgroup \0 */
670 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + 3;
671 dirname
= alloca(len
);
672 snprintf(dirname
, len
, "%s/%s/%s", basedir
, tmpc
, cgroup
);
674 dir
= opendir(dirname
);
678 while (!readdir_r(dir
, &dirent
, &direntp
)) {
685 if (!strcmp(direntp
->d_name
, ".") ||
686 !strcmp(direntp
->d_name
, ".."))
689 rc
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
690 if (rc
< 0 || rc
>= MAXPATHLEN
) {
691 fprintf(stderr
, "%s: pathname too long under %s\n", __func__
, dirname
);
695 ret
= lstat(pathname
, &mystat
);
697 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
700 if ((!directories
&& !S_ISREG(mystat
.st_mode
)) ||
701 (directories
&& !S_ISDIR(mystat
.st_mode
)))
708 tmp
= realloc(*list
, asz
* typesize
);
712 (*list
)[sz
] = (*iterator
)(controller
, cgroup
, direntp
->d_name
);
713 (*list
)[sz
+1] = NULL
;
716 if (closedir(dir
) < 0) {
717 fprintf(stderr
, "%s: failed closedir for %s: %s\n", __func__
, dirname
, strerror(errno
));
723 static void *make_children_list_entry(const char *controller
, const char *cgroup
, const char *dir_entry
)
727 dup
= strdup(dir_entry
);
732 bool cgfs_list_children(const char *controller
, const char *cgroup
, char ***list
)
734 return cgfs_iterate_cgroup(controller
, cgroup
, true, (void***)list
, sizeof(*list
), &make_children_list_entry
);
737 void free_key(struct cgfs_files
*k
)
745 void free_keys(struct cgfs_files
**keys
)
751 for (i
= 0; keys
[i
]; i
++) {
757 bool cgfs_get_value(const char *controller
, const char *cgroup
, const char *file
, char **value
)
760 char *fnam
, *tmpc
= find_mounted_controller(controller
);
764 /* basedir / tmpc / cgroup / file \0 */
765 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + strlen(file
) + 4;
767 snprintf(fnam
, len
, "%s/%s/%s/%s", basedir
, tmpc
, cgroup
, file
);
769 *value
= slurp_file(fnam
);
770 return *value
!= NULL
;
773 struct cgfs_files
*cgfs_get_key(const char *controller
, const char *cgroup
, const char *file
)
776 char *fnam
, *tmpc
= find_mounted_controller(controller
);
778 struct cgfs_files
*newkey
;
784 if (file
&& *file
== '/')
787 if (file
&& index(file
, '/'))
790 /* basedir / tmpc / cgroup / file \0 */
791 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + 3;
793 len
+= strlen(file
) + 1;
795 snprintf(fnam
, len
, "%s/%s/%s%s%s", basedir
, tmpc
, cgroup
,
796 file
? "/" : "", file
? file
: "");
798 ret
= stat(fnam
, &sb
);
803 newkey
= malloc(sizeof(struct cgfs_files
));
806 newkey
->name
= must_copy_string(file
);
807 else if (rindex(cgroup
, '/'))
808 newkey
->name
= must_copy_string(rindex(cgroup
, '/'));
810 newkey
->name
= must_copy_string(cgroup
);
811 newkey
->uid
= sb
.st_uid
;
812 newkey
->gid
= sb
.st_gid
;
813 newkey
->mode
= sb
.st_mode
;
818 static void *make_key_list_entry(const char *controller
, const char *cgroup
, const char *dir_entry
)
820 struct cgfs_files
*entry
= cgfs_get_key(controller
, cgroup
, dir_entry
);
822 fprintf(stderr
, "%s: Error getting files under %s:%s\n",
823 __func__
, controller
, cgroup
);
828 bool cgfs_list_keys(const char *controller
, const char *cgroup
, struct cgfs_files
***keys
)
830 return cgfs_iterate_cgroup(controller
, cgroup
, false, (void***)keys
, sizeof(*keys
), &make_key_list_entry
);
833 bool is_child_cgroup(const char *controller
, const char *cgroup
, const char *f
)
835 char *fnam
, *tmpc
= find_mounted_controller(controller
);
841 /* basedir / tmpc / cgroup / f \0 */
842 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + strlen(f
) + 4;
844 snprintf(fnam
, len
, "%s/%s/%s/%s", basedir
, tmpc
, cgroup
, f
);
846 ret
= stat(fnam
, &sb
);
847 if (ret
< 0 || !S_ISDIR(sb
.st_mode
))
852 #define SEND_CREDS_OK 0
853 #define SEND_CREDS_NOTSK 1
854 #define SEND_CREDS_FAIL 2
855 static bool recv_creds(int sock
, struct ucred
*cred
, char *v
);
856 static int wait_for_pid(pid_t pid
);
857 static int send_creds(int sock
, struct ucred
*cred
, char v
, bool pingfirst
);
858 static int send_creds_clone_wrapper(void *arg
);
861 * clone a task which switches to @task's namespace and writes '1'.
862 * over a unix sock so we can read the task's reaper's pid in our
865 * Note: glibc's fork() does not respect pidns, which can lead to failed
866 * assertions inside glibc (and thus failed forks) if the child's pid in
867 * the pidns and the parent pid outside are identical. Using clone prevents
870 static void write_task_init_pid_exit(int sock
, pid_t target
)
875 size_t stack_size
= sysconf(_SC_PAGESIZE
);
876 void *stack
= alloca(stack_size
);
878 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", (int)target
);
879 if (ret
< 0 || ret
>= sizeof(fnam
))
882 fd
= open(fnam
, O_RDONLY
);
884 perror("write_task_init_pid_exit open of ns/pid");
888 perror("write_task_init_pid_exit setns 1");
892 pid
= clone(send_creds_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &sock
);
896 if (!wait_for_pid(pid
))
902 static int send_creds_clone_wrapper(void *arg
) {
905 int sock
= *(int *)arg
;
907 /* we are the child */
912 if (send_creds(sock
, &cred
, v
, true) != SEND_CREDS_OK
)
917 static pid_t
get_init_pid_for_task(pid_t task
)
925 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
926 perror("socketpair");
935 write_task_init_pid_exit(sock
[0], task
);
939 if (!recv_creds(sock
[1], &cred
, &v
))
951 static pid_t
lookup_initpid_in_store(pid_t qpid
)
955 struct pidns_init_store
*e
;
958 snprintf(fnam
, 100, "/proc/%d/ns/pid", qpid
);
960 if (stat(fnam
, &sb
) < 0)
962 e
= lookup_verify_initpid(&sb
);
967 answer
= get_init_pid_for_task(qpid
);
969 save_initpid(&sb
, answer
);
972 /* we prune at end in case we are returning
973 * the value we were about to return */
974 prune_initpid_store();
979 static int wait_for_pid(pid_t pid
)
987 ret
= waitpid(pid
, &status
, 0);
995 if (!WIFEXITED(status
) || WEXITSTATUS(status
) != 0)
1002 * append pid to *src.
1003 * src: a pointer to a char* in which ot append the pid.
1004 * sz: the number of characters printed so far, minus trailing \0.
1005 * asz: the allocated size so far
1006 * pid: the pid to append
1008 static void must_strcat_pid(char **src
, size_t *sz
, size_t *asz
, pid_t pid
)
1012 int tmplen
= sprintf(tmp
, "%d\n", (int)pid
);
1014 if (!*src
|| tmplen
+ *sz
+ 1 >= *asz
) {
1017 tmp
= realloc(*src
, *asz
+ BUF_RESERVE_SIZE
);
1020 *asz
+= BUF_RESERVE_SIZE
;
1022 memcpy((*src
) +*sz
, tmp
, tmplen
+1); /* include the \0 */
1027 * Given a open file * to /proc/pid/{u,g}id_map, and an id
1028 * valid in the caller's namespace, return the id mapped into
1030 * Returns the mapped id, or -1 on error.
1033 convert_id_to_ns(FILE *idfile
, unsigned int in_id
)
1035 unsigned int nsuid
, // base id for a range in the idfile's namespace
1036 hostuid
, // base id for a range in the caller's namespace
1037 count
; // number of ids in this range
1041 fseek(idfile
, 0L, SEEK_SET
);
1042 while (fgets(line
, 400, idfile
)) {
1043 ret
= sscanf(line
, "%u %u %u\n", &nsuid
, &hostuid
, &count
);
1046 if (hostuid
+ count
< hostuid
|| nsuid
+ count
< nsuid
) {
1048 * uids wrapped around - unexpected as this is a procfile,
1051 fprintf(stderr
, "pid wrapparound at entry %u %u %u in %s\n",
1052 nsuid
, hostuid
, count
, line
);
1055 if (hostuid
<= in_id
&& hostuid
+count
> in_id
) {
1057 * now since hostuid <= in_id < hostuid+count, and
1058 * hostuid+count and nsuid+count do not wrap around,
1059 * we know that nsuid+(in_id-hostuid) which must be
1060 * less that nsuid+(count) must not wrap around
1062 return (in_id
- hostuid
) + nsuid
;
1071 * for is_privileged_over,
1072 * specify whether we require the calling uid to be root in his
1075 #define NS_ROOT_REQD true
1076 #define NS_ROOT_OPT false
1080 static bool is_privileged_over(pid_t pid
, uid_t uid
, uid_t victim
, bool req_ns_root
)
1082 char fpath
[PROCLEN
];
1084 bool answer
= false;
1087 if (victim
== -1 || uid
== -1)
1091 * If the request is one not requiring root in the namespace,
1092 * then having the same uid suffices. (i.e. uid 1000 has write
1093 * access to files owned by uid 1000
1095 if (!req_ns_root
&& uid
== victim
)
1098 ret
= snprintf(fpath
, PROCLEN
, "/proc/%d/uid_map", pid
);
1099 if (ret
< 0 || ret
>= PROCLEN
)
1101 FILE *f
= fopen(fpath
, "r");
1105 /* if caller's not root in his namespace, reject */
1106 nsuid
= convert_id_to_ns(f
, uid
);
1111 * If victim is not mapped into caller's ns, reject.
1112 * XXX I'm not sure this check is needed given that fuse
1113 * will be sending requests where the vfs has converted
1115 nsuid
= convert_id_to_ns(f
, victim
);
1126 static bool perms_include(int fmode
, mode_t req_mode
)
1130 switch (req_mode
& O_ACCMODE
) {
1138 r
= S_IROTH
| S_IWOTH
;
1143 return ((fmode
& r
) == r
);
1149 * querycg is /a/b/c/d/e
1152 static char *get_next_cgroup_dir(const char *taskcg
, const char *querycg
)
1156 if (strlen(taskcg
) <= strlen(querycg
)) {
1157 fprintf(stderr
, "%s: I was fed bad input\n", __func__
);
1161 if (strcmp(querycg
, "/") == 0)
1162 start
= strdup(taskcg
+ 1);
1164 start
= strdup(taskcg
+ strlen(querycg
) + 1);
1167 end
= strchr(start
, '/');
1173 static void stripnewline(char *x
)
1175 size_t l
= strlen(x
);
1176 if (l
&& x
[l
-1] == '\n')
1180 static char *get_pid_cgroup(pid_t pid
, const char *contrl
)
1184 char *answer
= NULL
;
1188 const char *h
= find_mounted_controller(contrl
);
1192 ret
= snprintf(fnam
, PROCLEN
, "/proc/%d/cgroup", pid
);
1193 if (ret
< 0 || ret
>= PROCLEN
)
1195 if (!(f
= fopen(fnam
, "r")))
1198 while (getline(&line
, &len
, f
) != -1) {
1202 c1
= strchr(line
, ':');
1206 c2
= strchr(c1
, ':');
1210 if (strcmp(c1
, h
) != 0)
1215 answer
= strdup(c2
);
1227 * check whether a fuse context may access a cgroup dir or file
1229 * If file is not null, it is a cgroup file to check under cg.
1230 * If file is null, then we are checking perms on cg itself.
1232 * For files we can check the mode of the list_keys result.
1233 * For cgroups, we must make assumptions based on the files under the
1234 * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups
1237 static bool fc_may_access(struct fuse_context
*fc
, const char *contrl
, const char *cg
, const char *file
, mode_t mode
)
1239 struct cgfs_files
*k
= NULL
;
1242 k
= cgfs_get_key(contrl
, cg
, file
);
1246 if (is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_OPT
)) {
1247 if (perms_include(k
->mode
>> 6, mode
)) {
1252 if (fc
->gid
== k
->gid
) {
1253 if (perms_include(k
->mode
>> 3, mode
)) {
1258 ret
= perms_include(k
->mode
, mode
);
1265 #define INITSCOPE "/init.scope"
1266 static void prune_init_slice(char *cg
)
1269 size_t cg_len
= strlen(cg
), initscope_len
= strlen(INITSCOPE
);
1271 if (cg_len
< initscope_len
)
1274 point
= cg
+ cg_len
- initscope_len
;
1275 if (strcmp(point
, INITSCOPE
) == 0) {
1284 * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d.
1285 * If pid is in /a, he may act on /a/b, but not on /b.
1286 * if the answer is false and nextcg is not NULL, then *nextcg will point
1287 * to a string containing the next cgroup directory under cg, which must be
1288 * freed by the caller.
1290 static bool caller_is_in_ancestor(pid_t pid
, const char *contrl
, const char *cg
, char **nextcg
)
1292 bool answer
= false;
1293 char *c2
= get_pid_cgroup(pid
, contrl
);
1298 prune_init_slice(c2
);
1301 * callers pass in '/' for root cgroup, otherwise they pass
1302 * in a cgroup without leading '/'
1304 linecmp
= *cg
== '/' ? c2
: c2
+1;
1305 if (strncmp(linecmp
, cg
, strlen(linecmp
)) != 0) {
1307 *nextcg
= get_next_cgroup_dir(linecmp
, cg
);
1319 * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c.
1321 static bool caller_may_see_dir(pid_t pid
, const char *contrl
, const char *cg
)
1323 bool answer
= false;
1325 size_t target_len
, task_len
;
1327 if (strcmp(cg
, "/") == 0)
1330 c2
= get_pid_cgroup(pid
, contrl
);
1333 prune_init_slice(c2
);
1336 target_len
= strlen(cg
);
1337 task_len
= strlen(task_cg
);
1338 if (task_len
== 0) {
1339 /* Task is in the root cg, it can see everything. This case is
1340 * not handled by the strmcps below, since they test for the
1341 * last /, but that is the first / that we've chopped off
1347 if (strcmp(cg
, task_cg
) == 0) {
1351 if (target_len
< task_len
) {
1352 /* looking up a parent dir */
1353 if (strncmp(task_cg
, cg
, target_len
) == 0 && task_cg
[target_len
] == '/')
1357 if (target_len
> task_len
) {
1358 /* looking up a child dir */
1359 if (strncmp(task_cg
, cg
, task_len
) == 0 && cg
[task_len
] == '/')
1370 * given /cgroup/freezer/a/b, return "freezer".
1371 * the returned char* should NOT be freed.
1373 static char *pick_controller_from_path(struct fuse_context
*fc
, const char *path
)
1376 char *contr
, *slash
;
1378 if (strlen(path
) < 9)
1380 if (*(path
+7) != '/')
1383 contr
= strdupa(p1
);
1386 slash
= strstr(contr
, "/");
1391 for (i
= 0; i
< num_hierarchies
; i
++) {
1392 if (hierarchies
[i
] && strcmp(hierarchies
[i
], contr
) == 0)
1393 return hierarchies
[i
];
1399 * Find the start of cgroup in /cgroup/controller/the/cgroup/path
1400 * Note that the returned value may include files (keynames) etc
1402 static const char *find_cgroup_in_path(const char *path
)
1406 if (strlen(path
) < 9)
1408 p1
= strstr(path
+8, "/");
1415 * split the last path element from the path in @cg.
1416 * @dir is newly allocated and should be freed, @last not
1418 static void get_cgdir_and_path(const char *cg
, char **dir
, char **last
)
1425 *last
= strrchr(cg
, '/');
1430 p
= strrchr(*dir
, '/');
1435 * FUSE ops for /cgroup
1438 int cg_getattr(const char *path
, struct stat
*sb
)
1440 struct timespec now
;
1441 struct fuse_context
*fc
= fuse_get_context();
1442 char * cgdir
= NULL
;
1443 char *last
= NULL
, *path1
, *path2
;
1444 struct cgfs_files
*k
= NULL
;
1446 const char *controller
= NULL
;
1453 memset(sb
, 0, sizeof(struct stat
));
1455 if (clock_gettime(CLOCK_REALTIME
, &now
) < 0)
1458 sb
->st_uid
= sb
->st_gid
= 0;
1459 sb
->st_atim
= sb
->st_mtim
= sb
->st_ctim
= now
;
1462 if (strcmp(path
, "/cgroup") == 0) {
1463 sb
->st_mode
= S_IFDIR
| 00755;
1468 controller
= pick_controller_from_path(fc
, path
);
1471 cgroup
= find_cgroup_in_path(path
);
1473 /* this is just /cgroup/controller, return it as a dir */
1474 sb
->st_mode
= S_IFDIR
| 00755;
1479 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1489 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1492 /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys.
1493 * Then check that caller's cgroup is under path if last is a child
1494 * cgroup, or cgdir if last is a file */
1496 if (is_child_cgroup(controller
, path1
, path2
)) {
1497 if (!caller_may_see_dir(initpid
, controller
, cgroup
)) {
1501 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, NULL
)) {
1502 /* this is just /cgroup/controller, return it as a dir */
1503 sb
->st_mode
= S_IFDIR
| 00555;
1508 if (!fc_may_access(fc
, controller
, cgroup
, NULL
, O_RDONLY
)) {
1513 // get uid, gid, from '/tasks' file and make up a mode
1514 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
1515 sb
->st_mode
= S_IFDIR
| 00755;
1516 k
= cgfs_get_key(controller
, cgroup
, NULL
);
1518 sb
->st_uid
= sb
->st_gid
= 0;
1520 sb
->st_uid
= k
->uid
;
1521 sb
->st_gid
= k
->gid
;
1529 if ((k
= cgfs_get_key(controller
, path1
, path2
)) != NULL
) {
1530 sb
->st_mode
= S_IFREG
| k
->mode
;
1532 sb
->st_uid
= k
->uid
;
1533 sb
->st_gid
= k
->gid
;
1536 if (!caller_is_in_ancestor(initpid
, controller
, path1
, NULL
)) {
1540 if (!fc_may_access(fc
, controller
, path1
, path2
, O_RDONLY
)) {
1553 int cg_opendir(const char *path
, struct fuse_file_info
*fi
)
1555 struct fuse_context
*fc
= fuse_get_context();
1557 struct file_info
*dir_info
;
1558 char *controller
= NULL
;
1563 if (strcmp(path
, "/cgroup") == 0) {
1567 // return list of keys for the controller, and list of child cgroups
1568 controller
= pick_controller_from_path(fc
, path
);
1572 cgroup
= find_cgroup_in_path(path
);
1574 /* this is just /cgroup/controller, return its contents */
1579 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1583 if (!caller_may_see_dir(initpid
, controller
, cgroup
))
1585 if (!fc_may_access(fc
, controller
, cgroup
, NULL
, O_RDONLY
))
1589 /* we'll free this at cg_releasedir */
1590 dir_info
= malloc(sizeof(*dir_info
));
1593 dir_info
->controller
= must_copy_string(controller
);
1594 dir_info
->cgroup
= must_copy_string(cgroup
);
1595 dir_info
->type
= LXC_TYPE_CGDIR
;
1596 dir_info
->buf
= NULL
;
1597 dir_info
->file
= NULL
;
1598 dir_info
->buflen
= 0;
1600 fi
->fh
= (unsigned long)dir_info
;
1604 int cg_readdir(const char *path
, void *buf
, fuse_fill_dir_t filler
, off_t offset
,
1605 struct fuse_file_info
*fi
)
1607 struct file_info
*d
= (struct file_info
*)fi
->fh
;
1608 struct cgfs_files
**list
= NULL
;
1610 char *nextcg
= NULL
;
1611 struct fuse_context
*fc
= fuse_get_context();
1612 char **clist
= NULL
;
1614 if (d
->type
!= LXC_TYPE_CGDIR
) {
1615 fprintf(stderr
, "Internal error: file cache info used in readdir\n");
1618 if (!d
->cgroup
&& !d
->controller
) {
1619 // ls /var/lib/lxcfs/cgroup - just show list of controllers
1622 for (i
= 0; i
< num_hierarchies
; i
++) {
1623 if (hierarchies
[i
] && filler(buf
, hierarchies
[i
], NULL
, 0) != 0) {
1630 if (!cgfs_list_keys(d
->controller
, d
->cgroup
, &list
)) {
1631 // not a valid cgroup
1636 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1639 if (!caller_is_in_ancestor(initpid
, d
->controller
, d
->cgroup
, &nextcg
)) {
1641 ret
= filler(buf
, nextcg
, NULL
, 0);
1652 for (i
= 0; list
[i
]; i
++) {
1653 if (filler(buf
, list
[i
]->name
, NULL
, 0) != 0) {
1659 // now get the list of child cgroups
1661 if (!cgfs_list_children(d
->controller
, d
->cgroup
, &clist
)) {
1666 for (i
= 0; clist
[i
]; i
++) {
1667 if (filler(buf
, clist
[i
], NULL
, 0) != 0) {
1678 for (i
= 0; clist
[i
]; i
++)
1685 static void do_release_file_info(struct fuse_file_info
*fi
)
1687 struct file_info
*f
= (struct file_info
*)fi
->fh
;
1694 free(f
->controller
);
1695 f
->controller
= NULL
;
1705 int cg_releasedir(const char *path
, struct fuse_file_info
*fi
)
1707 do_release_file_info(fi
);
1711 int cg_open(const char *path
, struct fuse_file_info
*fi
)
1714 char *last
= NULL
, *path1
, *path2
, * cgdir
= NULL
, *controller
;
1715 struct cgfs_files
*k
= NULL
;
1716 struct file_info
*file_info
;
1717 struct fuse_context
*fc
= fuse_get_context();
1723 controller
= pick_controller_from_path(fc
, path
);
1726 cgroup
= find_cgroup_in_path(path
);
1730 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1739 k
= cgfs_get_key(controller
, path1
, path2
);
1746 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1749 if (!caller_may_see_dir(initpid
, controller
, path1
)) {
1753 if (!fc_may_access(fc
, controller
, path1
, path2
, fi
->flags
)) {
1758 /* we'll free this at cg_release */
1759 file_info
= malloc(sizeof(*file_info
));
1764 file_info
->controller
= must_copy_string(controller
);
1765 file_info
->cgroup
= must_copy_string(path1
);
1766 file_info
->file
= must_copy_string(path2
);
1767 file_info
->type
= LXC_TYPE_CGFILE
;
1768 file_info
->buf
= NULL
;
1769 file_info
->buflen
= 0;
1771 fi
->fh
= (unsigned long)file_info
;
1779 int cg_access(const char *path
, int mode
)
1782 char *last
= NULL
, *path1
, *path2
, * cgdir
= NULL
, *controller
;
1783 struct cgfs_files
*k
= NULL
;
1784 struct fuse_context
*fc
= fuse_get_context();
1790 controller
= pick_controller_from_path(fc
, path
);
1793 cgroup
= find_cgroup_in_path(path
);
1795 // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not
1796 if ((mode
& W_OK
) == 0)
1801 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1810 k
= cgfs_get_key(controller
, path1
, path2
);
1812 if ((mode
& W_OK
) == 0)
1820 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1823 if (!caller_may_see_dir(initpid
, controller
, path1
)) {
1827 if (!fc_may_access(fc
, controller
, path1
, path2
, mode
)) {
1839 int cg_release(const char *path
, struct fuse_file_info
*fi
)
1841 do_release_file_info(fi
);
1845 #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP )
1847 static bool wait_for_sock(int sock
, int timeout
)
1849 struct epoll_event ev
;
1850 int epfd
, ret
, now
, starttime
, deltatime
, saved_errno
;
1852 if ((starttime
= time(NULL
)) < 0)
1855 if ((epfd
= epoll_create(1)) < 0) {
1856 fprintf(stderr
, "Failed to create epoll socket: %m\n");
1860 ev
.events
= POLLIN_SET
;
1862 if (epoll_ctl(epfd
, EPOLL_CTL_ADD
, sock
, &ev
) < 0) {
1863 fprintf(stderr
, "Failed adding socket to epoll: %m\n");
1869 if ((now
= time(NULL
)) < 0) {
1874 deltatime
= (starttime
+ timeout
) - now
;
1875 if (deltatime
< 0) { // timeout
1880 ret
= epoll_wait(epfd
, &ev
, 1, 1000*deltatime
+ 1);
1881 if (ret
< 0 && errno
== EINTR
)
1883 saved_errno
= errno
;
1887 errno
= saved_errno
;
1893 static int msgrecv(int sockfd
, void *buf
, size_t len
)
1895 if (!wait_for_sock(sockfd
, 2))
1897 return recv(sockfd
, buf
, len
, MSG_DONTWAIT
);
1900 static int send_creds(int sock
, struct ucred
*cred
, char v
, bool pingfirst
)
1902 struct msghdr msg
= { 0 };
1904 struct cmsghdr
*cmsg
;
1905 char cmsgbuf
[CMSG_SPACE(sizeof(*cred
))];
1910 if (msgrecv(sock
, buf
, 1) != 1) {
1911 fprintf(stderr
, "%s: Error getting reply from server over socketpair\n",
1913 return SEND_CREDS_FAIL
;
1917 msg
.msg_control
= cmsgbuf
;
1918 msg
.msg_controllen
= sizeof(cmsgbuf
);
1920 cmsg
= CMSG_FIRSTHDR(&msg
);
1921 cmsg
->cmsg_len
= CMSG_LEN(sizeof(struct ucred
));
1922 cmsg
->cmsg_level
= SOL_SOCKET
;
1923 cmsg
->cmsg_type
= SCM_CREDENTIALS
;
1924 memcpy(CMSG_DATA(cmsg
), cred
, sizeof(*cred
));
1926 msg
.msg_name
= NULL
;
1927 msg
.msg_namelen
= 0;
1931 iov
.iov_len
= sizeof(buf
);
1935 if (sendmsg(sock
, &msg
, 0) < 0) {
1936 fprintf(stderr
, "%s: failed at sendmsg: %s\n", __func__
,
1939 return SEND_CREDS_NOTSK
;
1940 return SEND_CREDS_FAIL
;
1943 return SEND_CREDS_OK
;
1946 static bool recv_creds(int sock
, struct ucred
*cred
, char *v
)
1948 struct msghdr msg
= { 0 };
1950 struct cmsghdr
*cmsg
;
1951 char cmsgbuf
[CMSG_SPACE(sizeof(*cred
))];
1962 if (setsockopt(sock
, SOL_SOCKET
, SO_PASSCRED
, &optval
, sizeof(optval
)) == -1) {
1963 fprintf(stderr
, "Failed to set passcred: %s\n", strerror(errno
));
1967 if (write(sock
, buf
, 1) != 1) {
1968 fprintf(stderr
, "Failed to start write on scm fd: %s\n", strerror(errno
));
1972 msg
.msg_name
= NULL
;
1973 msg
.msg_namelen
= 0;
1974 msg
.msg_control
= cmsgbuf
;
1975 msg
.msg_controllen
= sizeof(cmsgbuf
);
1978 iov
.iov_len
= sizeof(buf
);
1982 if (!wait_for_sock(sock
, 2)) {
1983 fprintf(stderr
, "Timed out waiting for scm_cred: %s\n",
1987 ret
= recvmsg(sock
, &msg
, MSG_DONTWAIT
);
1989 fprintf(stderr
, "Failed to receive scm_cred: %s\n",
1994 cmsg
= CMSG_FIRSTHDR(&msg
);
1996 if (cmsg
&& cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
)) &&
1997 cmsg
->cmsg_level
== SOL_SOCKET
&&
1998 cmsg
->cmsg_type
== SCM_CREDENTIALS
) {
1999 memcpy(cred
, CMSG_DATA(cmsg
), sizeof(*cred
));
2006 struct pid_ns_clone_args
{
2010 int (*wrapped
) (int, pid_t
); // pid_from_ns or pid_to_ns
2014 * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage
2015 * with clone(). This simply writes '1' as ACK back to the parent
2016 * before calling the actual wrapped function.
2018 static int pid_ns_clone_wrapper(void *arg
) {
2019 struct pid_ns_clone_args
* args
= (struct pid_ns_clone_args
*) arg
;
2022 close(args
->cpipe
[0]);
2023 if (write(args
->cpipe
[1], &b
, sizeof(char)) < 0) {
2024 fprintf(stderr
, "%s (child): error on write: %s\n",
2025 __func__
, strerror(errno
));
2027 close(args
->cpipe
[1]);
2028 return args
->wrapped(args
->sock
, args
->tpid
);
2032 * pid_to_ns - reads pids from a ucred over a socket, then writes the
2033 * int value back over the socket. This shifts the pid from the
2034 * sender's pidns into tpid's pidns.
2036 static int pid_to_ns(int sock
, pid_t tpid
)
2041 while (recv_creds(sock
, &cred
, &v
)) {
2044 if (write(sock
, &cred
.pid
, sizeof(pid_t
)) != sizeof(pid_t
))
2052 * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain
2053 * in your old pidns. Only children which you clone will be in the target
2054 * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to
2055 * actually convert pids.
2057 * Note: glibc's fork() does not respect pidns, which can lead to failed
2058 * assertions inside glibc (and thus failed forks) if the child's pid in
2059 * the pidns and the parent pid outside are identical. Using clone prevents
2062 static void pid_to_ns_wrapper(int sock
, pid_t tpid
)
2064 int newnsfd
= -1, ret
, cpipe
[2];
2069 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", tpid
);
2070 if (ret
< 0 || ret
>= sizeof(fnam
))
2072 newnsfd
= open(fnam
, O_RDONLY
);
2075 if (setns(newnsfd
, 0) < 0)
2079 if (pipe(cpipe
) < 0)
2082 struct pid_ns_clone_args args
= {
2086 .wrapped
= &pid_to_ns
2088 size_t stack_size
= sysconf(_SC_PAGESIZE
);
2089 void *stack
= alloca(stack_size
);
2091 cpid
= clone(pid_ns_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &args
);
2095 // give the child 1 second to be done forking and
2097 if (!wait_for_sock(cpipe
[0], 1))
2099 ret
= read(cpipe
[0], &v
, 1);
2100 if (ret
!= sizeof(char) || v
!= '1')
2103 if (!wait_for_pid(cpid
))
2109 * To read cgroup files with a particular pid, we will setns into the child
2110 * pidns, open a pipe, fork a child - which will be the first to really be in
2111 * the child ns - which does the cgfs_get_value and writes the data to the pipe.
2113 bool do_read_pids(pid_t tpid
, const char *contrl
, const char *cg
, const char *file
, char **d
)
2115 int sock
[2] = {-1, -1};
2116 char *tmpdata
= NULL
;
2118 pid_t qpid
, cpid
= -1;
2119 bool answer
= false;
2122 size_t sz
= 0, asz
= 0;
2124 if (!cgfs_get_value(contrl
, cg
, file
, &tmpdata
))
2128 * Now we read the pids from returned data one by one, pass
2129 * them into a child in the target namespace, read back the
2130 * translated pids, and put them into our to-return data
2133 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
2134 perror("socketpair");
2143 if (!cpid
) // child - exits when done
2144 pid_to_ns_wrapper(sock
[1], tpid
);
2146 char *ptr
= tmpdata
;
2149 while (sscanf(ptr
, "%d\n", &qpid
) == 1) {
2151 ret
= send_creds(sock
[0], &cred
, v
, true);
2153 if (ret
== SEND_CREDS_NOTSK
)
2155 if (ret
== SEND_CREDS_FAIL
)
2158 // read converted results
2159 if (!wait_for_sock(sock
[0], 2)) {
2160 fprintf(stderr
, "%s: timed out waiting for pid from child: %s\n",
2161 __func__
, strerror(errno
));
2164 if (read(sock
[0], &qpid
, sizeof(qpid
)) != sizeof(qpid
)) {
2165 fprintf(stderr
, "%s: error reading pid from child: %s\n",
2166 __func__
, strerror(errno
));
2169 must_strcat_pid(d
, &sz
, &asz
, qpid
);
2171 ptr
= strchr(ptr
, '\n');
2177 cred
.pid
= getpid();
2179 if (send_creds(sock
[0], &cred
, v
, true) != SEND_CREDS_OK
) {
2180 // failed to ask child to exit
2181 fprintf(stderr
, "%s: failed to ask child to exit: %s\n",
2182 __func__
, strerror(errno
));
2192 if (sock
[0] != -1) {
2199 int cg_read(const char *path
, char *buf
, size_t size
, off_t offset
,
2200 struct fuse_file_info
*fi
)
2202 struct fuse_context
*fc
= fuse_get_context();
2203 struct file_info
*f
= (struct file_info
*)fi
->fh
;
2204 struct cgfs_files
*k
= NULL
;
2209 if (f
->type
!= LXC_TYPE_CGFILE
) {
2210 fprintf(stderr
, "Internal error: directory cache info used in cg_read\n");
2223 if ((k
= cgfs_get_key(f
->controller
, f
->cgroup
, f
->file
)) == NULL
) {
2229 if (!fc_may_access(fc
, f
->controller
, f
->cgroup
, f
->file
, O_RDONLY
)) {
2234 if (strcmp(f
->file
, "tasks") == 0 ||
2235 strcmp(f
->file
, "/tasks") == 0 ||
2236 strcmp(f
->file
, "/cgroup.procs") == 0 ||
2237 strcmp(f
->file
, "cgroup.procs") == 0)
2238 // special case - we have to translate the pids
2239 r
= do_read_pids(fc
->pid
, f
->controller
, f
->cgroup
, f
->file
, &data
);
2241 r
= cgfs_get_value(f
->controller
, f
->cgroup
, f
->file
, &data
);
2255 memcpy(buf
, data
, s
);
2256 if (s
> 0 && s
< size
&& data
[s
-1] != '\n')
2266 static int pid_from_ns(int sock
, pid_t tpid
)
2276 if (!wait_for_sock(sock
, 2)) {
2277 fprintf(stderr
, "%s: timeout reading from parent\n", __func__
);
2280 if ((ret
= read(sock
, &vpid
, sizeof(pid_t
))) != sizeof(pid_t
)) {
2281 fprintf(stderr
, "%s: bad read from parent: %s\n",
2282 __func__
, strerror(errno
));
2285 if (vpid
== -1) // done
2289 if (send_creds(sock
, &cred
, v
, true) != SEND_CREDS_OK
) {
2291 cred
.pid
= getpid();
2292 if (send_creds(sock
, &cred
, v
, false) != SEND_CREDS_OK
)
2299 static void pid_from_ns_wrapper(int sock
, pid_t tpid
)
2301 int newnsfd
= -1, ret
, cpipe
[2];
2306 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", tpid
);
2307 if (ret
< 0 || ret
>= sizeof(fnam
))
2309 newnsfd
= open(fnam
, O_RDONLY
);
2312 if (setns(newnsfd
, 0) < 0)
2316 if (pipe(cpipe
) < 0)
2319 struct pid_ns_clone_args args
= {
2323 .wrapped
= &pid_from_ns
2325 size_t stack_size
= sysconf(_SC_PAGESIZE
);
2326 void *stack
= alloca(stack_size
);
2328 cpid
= clone(pid_ns_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &args
);
2332 // give the child 1 second to be done forking and
2334 if (!wait_for_sock(cpipe
[0], 1))
2336 ret
= read(cpipe
[0], &v
, 1);
2337 if (ret
!= sizeof(char) || v
!= '1')
2340 if (!wait_for_pid(cpid
))
2346 * Given host @uid, return the uid to which it maps in
2347 * @pid's user namespace, or -1 if none.
2349 bool hostuid_to_ns(uid_t uid
, pid_t pid
, uid_t
*answer
)
2354 sprintf(line
, "/proc/%d/uid_map", pid
);
2355 if ((f
= fopen(line
, "r")) == NULL
) {
2359 *answer
= convert_id_to_ns(f
, uid
);
2368 * get_pid_creds: get the real uid and gid of @pid from
2370 * (XXX should we use euid here?)
2372 void get_pid_creds(pid_t pid
, uid_t
*uid
, gid_t
*gid
)
2381 sprintf(line
, "/proc/%d/status", pid
);
2382 if ((f
= fopen(line
, "r")) == NULL
) {
2383 fprintf(stderr
, "Error opening %s: %s\n", line
, strerror(errno
));
2386 while (fgets(line
, 400, f
)) {
2387 if (strncmp(line
, "Uid:", 4) == 0) {
2388 if (sscanf(line
+4, "%u", &u
) != 1) {
2389 fprintf(stderr
, "bad uid line for pid %u\n", pid
);
2394 } else if (strncmp(line
, "Gid:", 4) == 0) {
2395 if (sscanf(line
+4, "%u", &g
) != 1) {
2396 fprintf(stderr
, "bad gid line for pid %u\n", pid
);
2407 * May the requestor @r move victim @v to a new cgroup?
2408 * This is allowed if
2409 * . they are the same task
2410 * . they are ownedy by the same uid
2411 * . @r is root on the host, or
2412 * . @v's uid is mapped into @r's where @r is root.
2414 bool may_move_pid(pid_t r
, uid_t r_uid
, pid_t v
)
2416 uid_t v_uid
, tmpuid
;
2423 get_pid_creds(v
, &v_uid
, &v_gid
);
2426 if (hostuid_to_ns(r_uid
, r
, &tmpuid
) && tmpuid
== 0
2427 && hostuid_to_ns(v_uid
, r
, &tmpuid
))
2432 static bool do_write_pids(pid_t tpid
, uid_t tuid
, const char *contrl
, const char *cg
,
2433 const char *file
, const char *buf
)
2435 int sock
[2] = {-1, -1};
2436 pid_t qpid
, cpid
= -1;
2437 FILE *pids_file
= NULL
;
2438 bool answer
= false, fail
= false;
2440 pids_file
= open_pids_file(contrl
, cg
);
2445 * write the pids to a socket, have helper in writer's pidns
2446 * call movepid for us
2448 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
2449 perror("socketpair");
2457 if (!cpid
) { // child
2459 pid_from_ns_wrapper(sock
[1], tpid
);
2462 const char *ptr
= buf
;
2463 while (sscanf(ptr
, "%d", &qpid
) == 1) {
2467 if (write(sock
[0], &qpid
, sizeof(qpid
)) != sizeof(qpid
)) {
2468 fprintf(stderr
, "%s: error writing pid to child: %s\n",
2469 __func__
, strerror(errno
));
2473 if (recv_creds(sock
[0], &cred
, &v
)) {
2475 if (!may_move_pid(tpid
, tuid
, cred
.pid
)) {
2479 if (fprintf(pids_file
, "%d", (int) cred
.pid
) < 0)
2484 ptr
= strchr(ptr
, '\n');
2490 /* All good, write the value */
2492 if (write(sock
[0], &qpid
,sizeof(qpid
)) != sizeof(qpid
))
2493 fprintf(stderr
, "Warning: failed to ask child to exit\n");
2501 if (sock
[0] != -1) {
2506 if (fclose(pids_file
) != 0)
2512 int cg_write(const char *path
, const char *buf
, size_t size
, off_t offset
,
2513 struct fuse_file_info
*fi
)
2515 struct fuse_context
*fc
= fuse_get_context();
2516 char *localbuf
= NULL
;
2517 struct cgfs_files
*k
= NULL
;
2518 struct file_info
*f
= (struct file_info
*)fi
->fh
;
2521 if (f
->type
!= LXC_TYPE_CGFILE
) {
2522 fprintf(stderr
, "Internal error: directory cache info used in cg_write\n");
2532 localbuf
= alloca(size
+1);
2533 localbuf
[size
] = '\0';
2534 memcpy(localbuf
, buf
, size
);
2536 if ((k
= cgfs_get_key(f
->controller
, f
->cgroup
, f
->file
)) == NULL
) {
2541 if (!fc_may_access(fc
, f
->controller
, f
->cgroup
, f
->file
, O_WRONLY
)) {
2546 if (strcmp(f
->file
, "tasks") == 0 ||
2547 strcmp(f
->file
, "/tasks") == 0 ||
2548 strcmp(f
->file
, "/cgroup.procs") == 0 ||
2549 strcmp(f
->file
, "cgroup.procs") == 0)
2550 // special case - we have to translate the pids
2551 r
= do_write_pids(fc
->pid
, fc
->uid
, f
->controller
, f
->cgroup
, f
->file
, localbuf
);
2553 r
= cgfs_set_value(f
->controller
, f
->cgroup
, f
->file
, localbuf
);
2563 int cg_chown(const char *path
, uid_t uid
, gid_t gid
)
2565 struct fuse_context
*fc
= fuse_get_context();
2566 char *cgdir
= NULL
, *last
= NULL
, *path1
, *path2
, *controller
;
2567 struct cgfs_files
*k
= NULL
;
2574 if (strcmp(path
, "/cgroup") == 0)
2577 controller
= pick_controller_from_path(fc
, path
);
2580 cgroup
= find_cgroup_in_path(path
);
2582 /* this is just /cgroup/controller */
2585 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2595 if (is_child_cgroup(controller
, path1
, path2
)) {
2596 // get uid, gid, from '/tasks' file and make up a mode
2597 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
2598 k
= cgfs_get_key(controller
, cgroup
, "tasks");
2601 k
= cgfs_get_key(controller
, path1
, path2
);
2609 * This being a fuse request, the uid and gid must be valid
2610 * in the caller's namespace. So we can just check to make
2611 * sure that the caller is root in his uid, and privileged
2612 * over the file's current owner.
2614 if (!is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_REQD
)) {
2619 ret
= cgfs_chown_file(controller
, cgroup
, uid
, gid
);
2628 int cg_chmod(const char *path
, mode_t mode
)
2630 struct fuse_context
*fc
= fuse_get_context();
2631 char * cgdir
= NULL
, *last
= NULL
, *path1
, *path2
, *controller
;
2632 struct cgfs_files
*k
= NULL
;
2639 if (strcmp(path
, "/cgroup") == 0)
2642 controller
= pick_controller_from_path(fc
, path
);
2645 cgroup
= find_cgroup_in_path(path
);
2647 /* this is just /cgroup/controller */
2650 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2660 if (is_child_cgroup(controller
, path1
, path2
)) {
2661 // get uid, gid, from '/tasks' file and make up a mode
2662 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
2663 k
= cgfs_get_key(controller
, cgroup
, "tasks");
2666 k
= cgfs_get_key(controller
, path1
, path2
);
2674 * This being a fuse request, the uid and gid must be valid
2675 * in the caller's namespace. So we can just check to make
2676 * sure that the caller is root in his uid, and privileged
2677 * over the file's current owner.
2679 if (!is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_OPT
)) {
2684 if (!cgfs_chmod_file(controller
, cgroup
, mode
)) {
2696 int cg_mkdir(const char *path
, mode_t mode
)
2698 struct fuse_context
*fc
= fuse_get_context();
2699 char *last
= NULL
, *path1
, *cgdir
= NULL
, *controller
, *next
= NULL
;
2707 controller
= pick_controller_from_path(fc
, path
);
2711 cgroup
= find_cgroup_in_path(path
);
2715 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2721 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2724 if (!caller_is_in_ancestor(initpid
, controller
, path1
, &next
)) {
2727 else if (last
&& strcmp(next
, last
) == 0)
2734 if (!fc_may_access(fc
, controller
, path1
, NULL
, O_RDWR
)) {
2738 if (!caller_is_in_ancestor(initpid
, controller
, path1
, NULL
)) {
2743 ret
= cgfs_create(controller
, cgroup
, fc
->uid
, fc
->gid
);
2751 int cg_rmdir(const char *path
)
2753 struct fuse_context
*fc
= fuse_get_context();
2754 char *last
= NULL
, *cgdir
= NULL
, *controller
, *next
= NULL
;
2761 controller
= pick_controller_from_path(fc
, path
);
2765 cgroup
= find_cgroup_in_path(path
);
2769 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2775 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2778 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, &next
)) {
2779 if (!last
|| strcmp(next
, last
) == 0)
2786 if (!fc_may_access(fc
, controller
, cgdir
, NULL
, O_WRONLY
)) {
2790 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, NULL
)) {
2795 if (!cgfs_remove(controller
, cgroup
)) {
2808 static bool startswith(const char *line
, const char *pref
)
2810 if (strncmp(line
, pref
, strlen(pref
)) == 0)
2815 static void get_mem_cached(char *memstat
, unsigned long *v
)
2821 if (startswith(memstat
, "total_cache")) {
2822 sscanf(memstat
+ 11, "%lu", v
);
2826 eol
= strchr(memstat
, '\n');
2833 static void get_blkio_io_value(char *str
, unsigned major
, unsigned minor
, char *iotype
, unsigned long *v
)
2839 snprintf(key
, 32, "%u:%u %s", major
, minor
, iotype
);
2841 size_t len
= strlen(key
);
2845 if (startswith(str
, key
)) {
2846 sscanf(str
+ len
, "%lu", v
);
2849 eol
= strchr(str
, '\n');
2856 static int read_file(const char *path
, char *buf
, size_t size
,
2857 struct file_info
*d
)
2859 size_t linelen
= 0, total_len
= 0, rv
= 0;
2861 char *cache
= d
->buf
;
2862 size_t cache_size
= d
->buflen
;
2863 FILE *f
= fopen(path
, "r");
2867 while (getline(&line
, &linelen
, f
) != -1) {
2868 ssize_t l
= snprintf(cache
, cache_size
, "%s", line
);
2870 perror("Error writing to cache");
2874 if (l
>= cache_size
) {
2875 fprintf(stderr
, "Internal error: truncated write to cache\n");
2884 d
->size
= total_len
;
2885 if (total_len
> size
)
2888 /* read from off 0 */
2889 memcpy(buf
, d
->buf
, total_len
);
2898 * FUSE ops for /proc
2901 static unsigned long get_memlimit(const char *cgroup
)
2903 char *memlimit_str
= NULL
;
2904 unsigned long memlimit
= -1;
2906 if (cgfs_get_value("memory", cgroup
, "memory.limit_in_bytes", &memlimit_str
))
2907 memlimit
= strtoul(memlimit_str
, NULL
, 10);
2914 static unsigned long get_min_memlimit(const char *cgroup
)
2916 char *copy
= strdupa(cgroup
);
2917 unsigned long memlimit
= 0, retlimit
;
2919 retlimit
= get_memlimit(copy
);
2921 while (strcmp(copy
, "/") != 0) {
2922 copy
= dirname(copy
);
2923 memlimit
= get_memlimit(copy
);
2924 if (memlimit
!= -1 && memlimit
< retlimit
)
2925 retlimit
= memlimit
;
2931 static int proc_meminfo_read(char *buf
, size_t size
, off_t offset
,
2932 struct fuse_file_info
*fi
)
2934 struct fuse_context
*fc
= fuse_get_context();
2935 struct file_info
*d
= (struct file_info
*)fi
->fh
;
2937 char *memusage_str
= NULL
, *memstat_str
= NULL
,
2938 *memswlimit_str
= NULL
, *memswusage_str
= NULL
,
2939 *memswlimit_default_str
= NULL
, *memswusage_default_str
= NULL
;
2940 unsigned long memlimit
= 0, memusage
= 0, memswlimit
= 0, memswusage
= 0,
2941 cached
= 0, hosttotal
= 0;
2943 size_t linelen
= 0, total_len
= 0, rv
= 0;
2944 char *cache
= d
->buf
;
2945 size_t cache_size
= d
->buflen
;
2949 if (offset
> d
->size
)
2953 int left
= d
->size
- offset
;
2954 total_len
= left
> size
? size
: left
;
2955 memcpy(buf
, cache
+ offset
, total_len
);
2959 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2962 cg
= get_pid_cgroup(initpid
, "memory");
2964 return read_file("/proc/meminfo", buf
, size
, d
);
2965 prune_init_slice(cg
);
2967 memlimit
= get_min_memlimit(cg
);
2968 if (!cgfs_get_value("memory", cg
, "memory.usage_in_bytes", &memusage_str
))
2970 if (!cgfs_get_value("memory", cg
, "memory.stat", &memstat_str
))
2973 // Following values are allowed to fail, because swapaccount might be turned
2974 // off for current kernel
2975 if(cgfs_get_value("memory", cg
, "memory.memsw.limit_in_bytes", &memswlimit_str
) &&
2976 cgfs_get_value("memory", cg
, "memory.memsw.usage_in_bytes", &memswusage_str
))
2978 /* If swapaccounting is turned on, then default value is assumed to be that of cgroup / */
2979 if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str
))
2981 if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str
))
2984 memswlimit
= strtoul(memswlimit_str
, NULL
, 10);
2985 memswusage
= strtoul(memswusage_str
, NULL
, 10);
2987 if (!strcmp(memswlimit_str
, memswlimit_default_str
))
2989 if (!strcmp(memswusage_str
, memswusage_default_str
))
2992 memswlimit
= memswlimit
/ 1024;
2993 memswusage
= memswusage
/ 1024;
2996 memusage
= strtoul(memusage_str
, NULL
, 10);
3000 get_mem_cached(memstat_str
, &cached
);
3002 f
= fopen("/proc/meminfo", "r");
3006 while (getline(&line
, &linelen
, f
) != -1) {
3008 char *printme
, lbuf
[100];
3010 memset(lbuf
, 0, 100);
3011 if (startswith(line
, "MemTotal:")) {
3012 sscanf(line
+14, "%lu", &hosttotal
);
3013 if (hosttotal
< memlimit
)
3014 memlimit
= hosttotal
;
3015 snprintf(lbuf
, 100, "MemTotal: %8lu kB\n", memlimit
);
3017 } else if (startswith(line
, "MemFree:")) {
3018 snprintf(lbuf
, 100, "MemFree: %8lu kB\n", memlimit
- memusage
);
3020 } else if (startswith(line
, "MemAvailable:")) {
3021 snprintf(lbuf
, 100, "MemAvailable: %8lu kB\n", memlimit
- memusage
);
3023 } else if (startswith(line
, "SwapTotal:") && memswlimit
> 0) {
3024 snprintf(lbuf
, 100, "SwapTotal: %8lu kB\n", memswlimit
- memlimit
);
3026 } else if (startswith(line
, "SwapFree:") && memswlimit
> 0 && memswusage
> 0) {
3027 snprintf(lbuf
, 100, "SwapFree: %8lu kB\n",
3028 (memswlimit
- memlimit
) - (memswusage
- memusage
));
3030 } else if (startswith(line
, "Slab:")) {
3031 snprintf(lbuf
, 100, "Slab: %8lu kB\n", 0UL);
3033 } else if (startswith(line
, "Buffers:")) {
3034 snprintf(lbuf
, 100, "Buffers: %8lu kB\n", 0UL);
3036 } else if (startswith(line
, "Cached:")) {
3037 snprintf(lbuf
, 100, "Cached: %8lu kB\n", cached
);
3039 } else if (startswith(line
, "SwapCached:")) {
3040 snprintf(lbuf
, 100, "SwapCached: %8lu kB\n", 0UL);
3045 l
= snprintf(cache
, cache_size
, "%s", printme
);
3047 perror("Error writing to cache");
3052 if (l
>= cache_size
) {
3053 fprintf(stderr
, "Internal error: truncated write to cache\n");
3064 d
->size
= total_len
;
3065 if (total_len
> size
) total_len
= size
;
3066 memcpy(buf
, d
->buf
, total_len
);
3075 free(memswlimit_str
);
3076 free(memswusage_str
);
3078 free(memswlimit_default_str
);
3079 free(memswusage_default_str
);
3084 * Read the cpuset.cpus for cg
3085 * Return the answer in a newly allocated string which must be freed
3087 static char *get_cpuset(const char *cg
)
3091 if (!cgfs_get_value("cpuset", cg
, "cpuset.cpus", &answer
))
3096 bool cpu_in_cpuset(int cpu
, const char *cpuset
);
3098 static bool cpuline_in_cpuset(const char *line
, const char *cpuset
)
3102 if (sscanf(line
, "processor : %d", &cpu
) != 1)
3104 return cpu_in_cpuset(cpu
, cpuset
);
3108 * check whether this is a '^processor" line in /proc/cpuinfo
3110 static bool is_processor_line(const char *line
)
3114 if (sscanf(line
, "processor : %d", &cpu
) == 1)
3119 static int proc_cpuinfo_read(char *buf
, size_t size
, off_t offset
,
3120 struct fuse_file_info
*fi
)
3122 struct fuse_context
*fc
= fuse_get_context();
3123 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3125 char *cpuset
= NULL
;
3127 size_t linelen
= 0, total_len
= 0, rv
= 0;
3128 bool am_printing
= false, firstline
= true, is_s390x
= false;
3129 int curcpu
= -1, cpu
;
3130 char *cache
= d
->buf
;
3131 size_t cache_size
= d
->buflen
;
3135 if (offset
> d
->size
)
3139 int left
= d
->size
- offset
;
3140 total_len
= left
> size
? size
: left
;
3141 memcpy(buf
, cache
+ offset
, total_len
);
3145 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3148 cg
= get_pid_cgroup(initpid
, "cpuset");
3150 return read_file("proc/cpuinfo", buf
, size
, d
);
3151 prune_init_slice(cg
);
3153 cpuset
= get_cpuset(cg
);
3157 f
= fopen("/proc/cpuinfo", "r");
3161 while (getline(&line
, &linelen
, f
) != -1) {
3165 if (strstr(line
, "IBM/S390") != NULL
) {
3171 if (strncmp(line
, "# processors:", 12) == 0)
3173 if (is_processor_line(line
)) {
3174 am_printing
= cpuline_in_cpuset(line
, cpuset
);
3177 l
= snprintf(cache
, cache_size
, "processor : %d\n", curcpu
);
3179 perror("Error writing to cache");
3183 if (l
>= cache_size
) {
3184 fprintf(stderr
, "Internal error: truncated write to cache\n");
3193 } else if (is_s390x
&& sscanf(line
, "processor %d:", &cpu
) == 1) {
3195 if (!cpu_in_cpuset(cpu
, cpuset
))
3198 p
= strchr(line
, ':');
3202 l
= snprintf(cache
, cache_size
, "processor %d:%s", curcpu
, p
);
3204 perror("Error writing to cache");
3208 if (l
>= cache_size
) {
3209 fprintf(stderr
, "Internal error: truncated write to cache\n");
3220 l
= snprintf(cache
, cache_size
, "%s", line
);
3222 perror("Error writing to cache");
3226 if (l
>= cache_size
) {
3227 fprintf(stderr
, "Internal error: truncated write to cache\n");
3238 char *origcache
= d
->buf
;
3241 d
->buf
= malloc(d
->buflen
);
3244 cache_size
= d
->buflen
;
3246 l
= snprintf(cache
, cache_size
, "vendor_id : IBM/S390\n");
3247 if (l
< 0 || l
>= cache_size
) {
3254 l
= snprintf(cache
, cache_size
, "# processors : %d\n", curcpu
+ 1);
3255 if (l
< 0 || l
>= cache_size
) {
3262 l
= snprintf(cache
, cache_size
, "%s", origcache
);
3264 if (l
< 0 || l
>= cache_size
)
3270 d
->size
= total_len
;
3271 if (total_len
> size
) total_len
= size
;
3273 /* read from off 0 */
3274 memcpy(buf
, d
->buf
, total_len
);
3285 static int proc_stat_read(char *buf
, size_t size
, off_t offset
,
3286 struct fuse_file_info
*fi
)
3288 struct fuse_context
*fc
= fuse_get_context();
3289 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3291 char *cpuset
= NULL
;
3293 size_t linelen
= 0, total_len
= 0, rv
= 0;
3294 int curcpu
= -1; /* cpu numbering starts at 0 */
3295 unsigned long user
= 0, nice
= 0, system
= 0, idle
= 0, iowait
= 0, irq
= 0, softirq
= 0, steal
= 0, guest
= 0;
3296 unsigned long user_sum
= 0, nice_sum
= 0, system_sum
= 0, idle_sum
= 0, iowait_sum
= 0,
3297 irq_sum
= 0, softirq_sum
= 0, steal_sum
= 0, guest_sum
= 0;
3298 #define CPUALL_MAX_SIZE BUF_RESERVE_SIZE
3299 char cpuall
[CPUALL_MAX_SIZE
];
3300 /* reserve for cpu all */
3301 char *cache
= d
->buf
+ CPUALL_MAX_SIZE
;
3302 size_t cache_size
= d
->buflen
- CPUALL_MAX_SIZE
;
3306 if (offset
> d
->size
)
3310 int left
= d
->size
- offset
;
3311 total_len
= left
> size
? size
: left
;
3312 memcpy(buf
, d
->buf
+ offset
, total_len
);
3316 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3319 cg
= get_pid_cgroup(initpid
, "cpuset");
3321 return read_file("/proc/stat", buf
, size
, d
);
3322 prune_init_slice(cg
);
3324 cpuset
= get_cpuset(cg
);
3328 f
= fopen("/proc/stat", "r");
3333 if (getline(&line
, &linelen
, f
) < 0) {
3334 fprintf(stderr
, "proc_stat_read read first line failed\n");
3338 while (getline(&line
, &linelen
, f
) != -1) {
3341 char cpu_char
[10]; /* That's a lot of cores */
3344 if (sscanf(line
, "cpu%9[^ ]", cpu_char
) != 1) {
3345 /* not a ^cpuN line containing a number N, just print it */
3346 l
= snprintf(cache
, cache_size
, "%s", line
);
3348 perror("Error writing to cache");
3352 if (l
>= cache_size
) {
3353 fprintf(stderr
, "Internal error: truncated write to cache\n");
3363 if (sscanf(cpu_char
, "%d", &cpu
) != 1)
3365 if (!cpu_in_cpuset(cpu
, cpuset
))
3369 c
= strchr(line
, ' ');
3372 l
= snprintf(cache
, cache_size
, "cpu%d%s", curcpu
, c
);
3374 perror("Error writing to cache");
3379 if (l
>= cache_size
) {
3380 fprintf(stderr
, "Internal error: truncated write to cache\n");
3389 if (sscanf(line
, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user
, &nice
, &system
, &idle
, &iowait
, &irq
,
3390 &softirq
, &steal
, &guest
) != 9)
3394 system_sum
+= system
;
3396 iowait_sum
+= iowait
;
3398 softirq_sum
+= softirq
;
3405 int cpuall_len
= snprintf(cpuall
, CPUALL_MAX_SIZE
, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
3406 "cpu ", user_sum
, nice_sum
, system_sum
, idle_sum
, iowait_sum
, irq_sum
, softirq_sum
, steal_sum
, guest_sum
);
3407 if (cpuall_len
> 0 && cpuall_len
< CPUALL_MAX_SIZE
){
3408 memcpy(cache
, cpuall
, cpuall_len
);
3409 cache
+= cpuall_len
;
3411 /* shouldn't happen */
3412 fprintf(stderr
, "proc_stat_read copy cpuall failed, cpuall_len=%d\n", cpuall_len
);
3416 memmove(cache
, d
->buf
+ CPUALL_MAX_SIZE
, total_len
);
3417 total_len
+= cpuall_len
;
3419 d
->size
= total_len
;
3420 if (total_len
> size
) total_len
= size
;
3422 memcpy(buf
, d
->buf
, total_len
);
3434 static long int getreaperage(pid_t pid
)
3441 qpid
= lookup_initpid_in_store(pid
);
3445 ret
= snprintf(fnam
, 100, "/proc/%d", qpid
);
3446 if (ret
< 0 || ret
>= 100)
3449 if (lstat(fnam
, &sb
) < 0)
3452 return time(NULL
) - sb
.st_ctime
;
3455 static unsigned long get_reaper_busy(pid_t task
)
3457 pid_t initpid
= lookup_initpid_in_store(task
);
3458 char *cgroup
= NULL
, *usage_str
= NULL
;
3459 unsigned long usage
= 0;
3464 cgroup
= get_pid_cgroup(initpid
, "cpuacct");
3467 prune_init_slice(cgroup
);
3468 if (!cgfs_get_value("cpuacct", cgroup
, "cpuacct.usage", &usage_str
))
3470 usage
= strtoul(usage_str
, NULL
, 10);
3471 usage
/= 1000000000;
3482 char *name
, *cwd
= get_current_dir_name();
3488 len
= strlen(cwd
) + strlen("/iwashere") + 1;
3490 snprintf(name
, len
, "%s/iwashere", cwd
);
3492 fd
= creat(name
, 0755);
3499 * We read /proc/uptime and reuse its second field.
3500 * For the first field, we use the mtime for the reaper for
3501 * the calling pid as returned by getreaperage
3503 static int proc_uptime_read(char *buf
, size_t size
, off_t offset
,
3504 struct fuse_file_info
*fi
)
3506 struct fuse_context
*fc
= fuse_get_context();
3507 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3508 long int reaperage
= getreaperage(fc
->pid
);
3509 unsigned long int busytime
= get_reaper_busy(fc
->pid
), idletime
;
3510 char *cache
= d
->buf
;
3511 ssize_t total_len
= 0;
3518 if (offset
> d
->size
)
3522 int left
= d
->size
- offset
;
3523 total_len
= left
> size
? size
: left
;
3524 memcpy(buf
, cache
+ offset
, total_len
);
3528 idletime
= reaperage
- busytime
;
3529 if (idletime
> reaperage
)
3530 idletime
= reaperage
;
3532 total_len
= snprintf(d
->buf
, d
->size
, "%ld.0 %lu.0\n", reaperage
, idletime
);
3534 perror("Error writing to cache");
3538 d
->size
= (int)total_len
;
3541 if (total_len
> size
) total_len
= size
;
3543 memcpy(buf
, d
->buf
, total_len
);
3547 static int proc_diskstats_read(char *buf
, size_t size
, off_t offset
,
3548 struct fuse_file_info
*fi
)
3551 struct fuse_context
*fc
= fuse_get_context();
3552 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3554 char *io_serviced_str
= NULL
, *io_merged_str
= NULL
, *io_service_bytes_str
= NULL
,
3555 *io_wait_time_str
= NULL
, *io_service_time_str
= NULL
;
3556 unsigned long read
= 0, write
= 0;
3557 unsigned long read_merged
= 0, write_merged
= 0;
3558 unsigned long read_sectors
= 0, write_sectors
= 0;
3559 unsigned long read_ticks
= 0, write_ticks
= 0;
3560 unsigned long ios_pgr
= 0, tot_ticks
= 0, rq_ticks
= 0;
3561 unsigned long rd_svctm
= 0, wr_svctm
= 0, rd_wait
= 0, wr_wait
= 0;
3562 char *cache
= d
->buf
;
3563 size_t cache_size
= d
->buflen
;
3565 size_t linelen
= 0, total_len
= 0, rv
= 0;
3566 unsigned int major
= 0, minor
= 0;
3571 if (offset
> d
->size
)
3575 int left
= d
->size
- offset
;
3576 total_len
= left
> size
? size
: left
;
3577 memcpy(buf
, cache
+ offset
, total_len
);
3581 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3584 cg
= get_pid_cgroup(initpid
, "blkio");
3586 return read_file("/proc/diskstats", buf
, size
, d
);
3587 prune_init_slice(cg
);
3589 if (!cgfs_get_value("blkio", cg
, "blkio.io_serviced_recursive", &io_serviced_str
))
3591 if (!cgfs_get_value("blkio", cg
, "blkio.io_merged_recursive", &io_merged_str
))
3593 if (!cgfs_get_value("blkio", cg
, "blkio.io_service_bytes_recursive", &io_service_bytes_str
))
3595 if (!cgfs_get_value("blkio", cg
, "blkio.io_wait_time_recursive", &io_wait_time_str
))
3597 if (!cgfs_get_value("blkio", cg
, "blkio.io_service_time_recursive", &io_service_time_str
))
3601 f
= fopen("/proc/diskstats", "r");
3605 while (getline(&line
, &linelen
, f
) != -1) {
3609 i
= sscanf(line
, "%u %u %71s", &major
, &minor
, dev_name
);
3613 get_blkio_io_value(io_serviced_str
, major
, minor
, "Read", &read
);
3614 get_blkio_io_value(io_serviced_str
, major
, minor
, "Write", &write
);
3615 get_blkio_io_value(io_merged_str
, major
, minor
, "Read", &read_merged
);
3616 get_blkio_io_value(io_merged_str
, major
, minor
, "Write", &write_merged
);
3617 get_blkio_io_value(io_service_bytes_str
, major
, minor
, "Read", &read_sectors
);
3618 read_sectors
= read_sectors
/512;
3619 get_blkio_io_value(io_service_bytes_str
, major
, minor
, "Write", &write_sectors
);
3620 write_sectors
= write_sectors
/512;
3622 get_blkio_io_value(io_service_time_str
, major
, minor
, "Read", &rd_svctm
);
3623 rd_svctm
= rd_svctm
/1000000;
3624 get_blkio_io_value(io_wait_time_str
, major
, minor
, "Read", &rd_wait
);
3625 rd_wait
= rd_wait
/1000000;
3626 read_ticks
= rd_svctm
+ rd_wait
;
3628 get_blkio_io_value(io_service_time_str
, major
, minor
, "Write", &wr_svctm
);
3629 wr_svctm
= wr_svctm
/1000000;
3630 get_blkio_io_value(io_wait_time_str
, major
, minor
, "Write", &wr_wait
);
3631 wr_wait
= wr_wait
/1000000;
3632 write_ticks
= wr_svctm
+ wr_wait
;
3634 get_blkio_io_value(io_service_time_str
, major
, minor
, "Total", &tot_ticks
);
3635 tot_ticks
= tot_ticks
/1000000;
3637 memset(lbuf
, 0, 256);
3638 if (read
|| write
|| read_merged
|| write_merged
|| read_sectors
|| write_sectors
|| read_ticks
|| write_ticks
)
3639 snprintf(lbuf
, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
3640 major
, minor
, dev_name
, read
, read_merged
, read_sectors
, read_ticks
,
3641 write
, write_merged
, write_sectors
, write_ticks
, ios_pgr
, tot_ticks
, rq_ticks
);
3645 l
= snprintf(cache
, cache_size
, "%s", lbuf
);
3647 perror("Error writing to fuse buf");
3651 if (l
>= cache_size
) {
3652 fprintf(stderr
, "Internal error: truncated write to cache\n");
3662 d
->size
= total_len
;
3663 if (total_len
> size
) total_len
= size
;
3664 memcpy(buf
, d
->buf
, total_len
);
3672 free(io_serviced_str
);
3673 free(io_merged_str
);
3674 free(io_service_bytes_str
);
3675 free(io_wait_time_str
);
3676 free(io_service_time_str
);
3680 static int proc_swaps_read(char *buf
, size_t size
, off_t offset
,
3681 struct fuse_file_info
*fi
)
3683 struct fuse_context
*fc
= fuse_get_context();
3684 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3686 char *memswlimit_str
= NULL
, *memlimit_str
= NULL
, *memusage_str
= NULL
, *memswusage_str
= NULL
,
3687 *memswlimit_default_str
= NULL
, *memswusage_default_str
= NULL
;
3688 unsigned long memswlimit
= 0, memlimit
= 0, memusage
= 0, memswusage
= 0, swap_total
= 0, swap_free
= 0;
3689 ssize_t total_len
= 0, rv
= 0;
3691 char *cache
= d
->buf
;
3694 if (offset
> d
->size
)
3698 int left
= d
->size
- offset
;
3699 total_len
= left
> size
? size
: left
;
3700 memcpy(buf
, cache
+ offset
, total_len
);
3704 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3707 cg
= get_pid_cgroup(initpid
, "memory");
3709 return read_file("/proc/swaps", buf
, size
, d
);
3710 prune_init_slice(cg
);
3712 if (!cgfs_get_value("memory", cg
, "memory.limit_in_bytes", &memlimit_str
))
3715 if (!cgfs_get_value("memory", cg
, "memory.usage_in_bytes", &memusage_str
))
3718 memlimit
= strtoul(memlimit_str
, NULL
, 10);
3719 memusage
= strtoul(memusage_str
, NULL
, 10);
3721 if (cgfs_get_value("memory", cg
, "memory.memsw.usage_in_bytes", &memswusage_str
) &&
3722 cgfs_get_value("memory", cg
, "memory.memsw.limit_in_bytes", &memswlimit_str
)) {
3724 /* If swap accounting is turned on, then default value is assumed to be that of cgroup / */
3725 if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str
))
3727 if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str
))
3730 memswlimit
= strtoul(memswlimit_str
, NULL
, 10);
3731 memswusage
= strtoul(memswusage_str
, NULL
, 10);
3733 if (!strcmp(memswlimit_str
, memswlimit_default_str
))
3735 if (!strcmp(memswusage_str
, memswusage_default_str
))
3738 swap_total
= (memswlimit
- memlimit
) / 1024;
3739 swap_free
= (memswusage
- memusage
) / 1024;
3742 total_len
= snprintf(d
->buf
, d
->size
, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
3744 /* When no mem + swap limit is specified or swapaccount=0*/
3748 FILE *f
= fopen("/proc/meminfo", "r");
3753 while (getline(&line
, &linelen
, f
) != -1) {
3754 if (startswith(line
, "SwapTotal:")) {
3755 sscanf(line
, "SwapTotal: %8lu kB", &swap_total
);
3756 } else if (startswith(line
, "SwapFree:")) {
3757 sscanf(line
, "SwapFree: %8lu kB", &swap_free
);
3765 if (swap_total
> 0) {
3766 l
= snprintf(d
->buf
+ total_len
, d
->size
- total_len
,
3767 "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ",
3768 swap_total
, swap_free
);
3772 if (total_len
< 0 || l
< 0) {
3773 perror("Error writing to cache");
3779 d
->size
= (int)total_len
;
3781 if (total_len
> size
) total_len
= size
;
3782 memcpy(buf
, d
->buf
, total_len
);
3787 free(memswlimit_str
);
3790 free(memswusage_str
);
3791 free(memswusage_default_str
);
3792 free(memswlimit_default_str
);
3796 static off_t
get_procfile_size(const char *which
)
3798 FILE *f
= fopen(which
, "r");
3801 ssize_t sz
, answer
= 0;
3805 while ((sz
= getline(&line
, &len
, f
)) != -1)
3813 int proc_getattr(const char *path
, struct stat
*sb
)
3815 struct timespec now
;
3817 memset(sb
, 0, sizeof(struct stat
));
3818 if (clock_gettime(CLOCK_REALTIME
, &now
) < 0)
3820 sb
->st_uid
= sb
->st_gid
= 0;
3821 sb
->st_atim
= sb
->st_mtim
= sb
->st_ctim
= now
;
3822 if (strcmp(path
, "/proc") == 0) {
3823 sb
->st_mode
= S_IFDIR
| 00555;
3827 if (strcmp(path
, "/proc/meminfo") == 0 ||
3828 strcmp(path
, "/proc/cpuinfo") == 0 ||
3829 strcmp(path
, "/proc/uptime") == 0 ||
3830 strcmp(path
, "/proc/stat") == 0 ||
3831 strcmp(path
, "/proc/diskstats") == 0 ||
3832 strcmp(path
, "/proc/swaps") == 0) {
3834 sb
->st_mode
= S_IFREG
| 00444;
3842 int proc_readdir(const char *path
, void *buf
, fuse_fill_dir_t filler
, off_t offset
,
3843 struct fuse_file_info
*fi
)
3845 if (filler(buf
, "cpuinfo", NULL
, 0) != 0 ||
3846 filler(buf
, "meminfo", NULL
, 0) != 0 ||
3847 filler(buf
, "stat", NULL
, 0) != 0 ||
3848 filler(buf
, "uptime", NULL
, 0) != 0 ||
3849 filler(buf
, "diskstats", NULL
, 0) != 0 ||
3850 filler(buf
, "swaps", NULL
, 0) != 0)
3855 int proc_open(const char *path
, struct fuse_file_info
*fi
)
3858 struct file_info
*info
;
3860 if (strcmp(path
, "/proc/meminfo") == 0)
3861 type
= LXC_TYPE_PROC_MEMINFO
;
3862 else if (strcmp(path
, "/proc/cpuinfo") == 0)
3863 type
= LXC_TYPE_PROC_CPUINFO
;
3864 else if (strcmp(path
, "/proc/uptime") == 0)
3865 type
= LXC_TYPE_PROC_UPTIME
;
3866 else if (strcmp(path
, "/proc/stat") == 0)
3867 type
= LXC_TYPE_PROC_STAT
;
3868 else if (strcmp(path
, "/proc/diskstats") == 0)
3869 type
= LXC_TYPE_PROC_DISKSTATS
;
3870 else if (strcmp(path
, "/proc/swaps") == 0)
3871 type
= LXC_TYPE_PROC_SWAPS
;
3875 info
= malloc(sizeof(*info
));
3879 memset(info
, 0, sizeof(*info
));
3882 info
->buflen
= get_procfile_size(path
) + BUF_RESERVE_SIZE
;
3884 info
->buf
= malloc(info
->buflen
);
3885 } while (!info
->buf
);
3886 memset(info
->buf
, 0, info
->buflen
);
3887 /* set actual size to buffer size */
3888 info
->size
= info
->buflen
;
3890 fi
->fh
= (unsigned long)info
;
3894 int proc_access(const char *path
, int mask
)
3896 /* these are all read-only */
3897 if ((mask
& ~R_OK
) != 0)
3902 int proc_release(const char *path
, struct fuse_file_info
*fi
)
3904 do_release_file_info(fi
);
3908 int proc_read(const char *path
, char *buf
, size_t size
, off_t offset
,
3909 struct fuse_file_info
*fi
)
3911 struct file_info
*f
= (struct file_info
*) fi
->fh
;
3914 case LXC_TYPE_PROC_MEMINFO
:
3915 return proc_meminfo_read(buf
, size
, offset
, fi
);
3916 case LXC_TYPE_PROC_CPUINFO
:
3917 return proc_cpuinfo_read(buf
, size
, offset
, fi
);
3918 case LXC_TYPE_PROC_UPTIME
:
3919 return proc_uptime_read(buf
, size
, offset
, fi
);
3920 case LXC_TYPE_PROC_STAT
:
3921 return proc_stat_read(buf
, size
, offset
, fi
);
3922 case LXC_TYPE_PROC_DISKSTATS
:
3923 return proc_diskstats_read(buf
, size
, offset
, fi
);
3924 case LXC_TYPE_PROC_SWAPS
:
3925 return proc_swaps_read(buf
, size
, offset
, fi
);
3931 static void __attribute__((constructor
)) collect_subsystems(void)
3937 if ((f
= fopen("/proc/self/cgroup", "r")) == NULL
) {
3938 fprintf(stderr
, "Error opening /proc/self/cgroup: %s\n", strerror(errno
));
3941 while (getline(&line
, &len
, f
) != -1) {
3944 p
= strchr(line
, ':');
3949 p2
= strrchr(p
, ':');
3954 /* With cgroupv2 /proc/self/cgroup can contain entries of the
3955 * form: 0::/ This will cause lxcfs to fail the cgroup mounts
3956 * because it parses out the empty string "" and later on passes
3957 * it to mount(). Let's skip such entries.
3962 if (!store_hierarchy(line
, p
))
3973 static void __attribute__((destructor
)) free_subsystems(void)
3977 for (i
= 0; i
< num_hierarchies
; i
++)
3979 free(hierarchies
[i
]);