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>
32 #define GLIB_DISABLE_DEPRECATION_WARNINGS
33 #include <glib-object.h>
38 #include "config.h" // for VERSION
43 LXC_TYPE_PROC_MEMINFO
,
44 LXC_TYPE_PROC_CPUINFO
,
47 LXC_TYPE_PROC_DISKSTATS
,
55 char *buf
; // unused as of yet
57 int size
; //actual data size
61 /* reserve buffer size, for cpuall in /proc/stat */
62 #define BUF_RESERVE_SIZE 256
65 * A table caching which pid is init for a pid namespace.
66 * When looking up which pid is init for $qpid, we first
67 * 1. Stat /proc/$qpid/ns/pid.
68 * 2. Check whether the ino_t is in our store.
69 * a. if not, fork a child in qpid's ns to send us
70 * ucred.pid = 1, and read the initpid. Cache
71 * initpid and creation time for /proc/initpid
72 * in a new store entry.
73 * b. if so, verify that /proc/initpid still matches
74 * what we have saved. If not, clear the store
75 * entry and go back to a. If so, return the
78 struct pidns_init_store
{
79 ino_t ino
; // inode number for /proc/$pid/ns/pid
80 pid_t initpid
; // the pid of nit in that ns
81 long int ctime
; // the time at which /proc/$initpid was created
82 struct pidns_init_store
*next
;
86 /* lol - look at how they are allocated in the kernel */
87 #define PIDNS_HASH_SIZE 4096
88 #define HASH(x) ((x) % PIDNS_HASH_SIZE)
90 static struct pidns_init_store
*pidns_hash_table
[PIDNS_HASH_SIZE
];
91 static pthread_mutex_t pidns_store_mutex
= PTHREAD_MUTEX_INITIALIZER
;
92 static void lock_mutex(pthread_mutex_t
*l
)
96 if ((ret
= pthread_mutex_lock(l
)) != 0) {
97 fprintf(stderr
, "pthread_mutex_lock returned:%d %s\n", ret
, strerror(ret
));
102 static void unlock_mutex(pthread_mutex_t
*l
)
106 if ((ret
= pthread_mutex_unlock(l
)) != 0) {
107 fprintf(stderr
, "pthread_mutex_unlock returned:%d %s\n", ret
, strerror(ret
));
112 static void store_lock(void)
114 lock_mutex(&pidns_store_mutex
);
117 static void store_unlock(void)
119 unlock_mutex(&pidns_store_mutex
);
122 /* Must be called under store_lock */
123 static bool initpid_still_valid(struct pidns_init_store
*e
, struct stat
*nsfdsb
)
128 snprintf(fnam
, 100, "/proc/%d", e
->initpid
);
129 if (stat(fnam
, &initsb
) < 0)
132 fprintf(stderr
, "comparing ctime %ld %ld for pid %d\n",
133 e
->ctime
, initsb
.st_ctime
, e
->initpid
);
135 if (e
->ctime
!= initsb
.st_ctime
)
140 /* Must be called under store_lock */
141 static void remove_initpid(struct pidns_init_store
*e
)
143 struct pidns_init_store
*tmp
;
147 fprintf(stderr
, "remove_initpid: removing entry for %d\n", e
->initpid
);
150 if (pidns_hash_table
[h
] == e
) {
151 pidns_hash_table
[h
] = e
->next
;
156 tmp
= pidns_hash_table
[h
];
158 if (tmp
->next
== e
) {
168 /* Must be called under store_lock */
169 static void prune_initpid_store(void)
171 static long int last_prune
= 0;
172 struct pidns_init_store
*e
, *prev
, *delme
;
173 long int now
, threshold
;
177 last_prune
= time(NULL
);
181 if (now
< last_prune
+ PURGE_SECS
)
184 fprintf(stderr
, "pruning\n");
187 threshold
= now
- 2 * PURGE_SECS
;
189 for (i
= 0; i
< PIDNS_HASH_SIZE
; i
++) {
190 for (prev
= NULL
, e
= pidns_hash_table
[i
]; e
; ) {
191 if (e
->lastcheck
< threshold
) {
193 fprintf(stderr
, "Removing cached entry for %d\n", e
->initpid
);
197 prev
->next
= e
->next
;
199 pidns_hash_table
[i
] = e
->next
;
210 /* Must be called under store_lock */
211 static void save_initpid(struct stat
*sb
, pid_t pid
)
213 struct pidns_init_store
*e
;
219 fprintf(stderr
, "save_initpid: adding entry for %d\n", pid
);
221 snprintf(fpath
, 100, "/proc/%d", pid
);
222 if (stat(fpath
, &procsb
) < 0)
225 e
= malloc(sizeof(*e
));
229 e
->ctime
= procsb
.st_ctime
;
231 e
->next
= pidns_hash_table
[h
];
232 e
->lastcheck
= time(NULL
);
233 pidns_hash_table
[h
] = e
;
237 * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store
238 * entry for the inode number and creation time. Verify that the init pid
239 * is still valid. If not, remove it. Return the entry if valid, NULL
241 * Must be called under store_lock
243 static struct pidns_init_store
*lookup_verify_initpid(struct stat
*sb
)
245 int h
= HASH(sb
->st_ino
);
246 struct pidns_init_store
*e
= pidns_hash_table
[h
];
249 if (e
->ino
== sb
->st_ino
) {
250 if (initpid_still_valid(e
, sb
)) {
251 e
->lastcheck
= time(NULL
);
263 static int is_dir(const char *path
)
266 int ret
= stat(path
, &statbuf
);
267 if (ret
== 0 && S_ISDIR(statbuf
.st_mode
))
272 static char *must_copy_string(const char *str
)
284 static inline void drop_trailing_newlines(char *s
)
288 for (l
=strlen(s
); l
>0 && s
[l
-1] == '\n'; l
--)
292 #define BATCH_SIZE 50
293 static void dorealloc(char **mem
, size_t oldlen
, size_t newlen
)
295 int newbatches
= (newlen
/ BATCH_SIZE
) + 1;
296 int oldbatches
= (oldlen
/ BATCH_SIZE
) + 1;
298 if (!*mem
|| newbatches
> oldbatches
) {
301 tmp
= realloc(*mem
, newbatches
* BATCH_SIZE
);
306 static void append_line(char **contents
, size_t *len
, char *line
, ssize_t linelen
)
308 size_t newlen
= *len
+ linelen
;
309 dorealloc(contents
, *len
, newlen
+ 1);
310 memcpy(*contents
+ *len
, line
, linelen
+1);
314 static char *slurp_file(const char *from
)
317 char *contents
= NULL
;
318 FILE *f
= fopen(from
, "r");
319 size_t len
= 0, fulllen
= 0;
325 while ((linelen
= getline(&line
, &len
, f
)) != -1) {
326 append_line(&contents
, &fulllen
, line
, linelen
);
331 drop_trailing_newlines(contents
);
336 static bool write_string(const char *fnam
, const char *string
)
341 if (!(f
= fopen(fnam
, "w")))
343 len
= strlen(string
);
344 ret
= fwrite(string
, 1, len
, f
);
346 fprintf(stderr
, "Error writing to file: %s\n", strerror(errno
));
351 fprintf(stderr
, "Error writing to file: %s\n", strerror(errno
));
358 * hierarchies, i.e. 'cpu,cpuacct'
370 static bool store_hierarchy(char *stridx
, char *h
)
372 if (num_hierarchies
% ALLOC_NUM
== 0) {
373 size_t n
= (num_hierarchies
/ ALLOC_NUM
) + 1;
375 char **tmp
= realloc(hierarchies
, n
* sizeof(char *));
376 printf("allocated %d\n", n
);
378 fprintf(stderr
, "Out of memory\n");
384 hierarchies
[num_hierarchies
++] = must_copy_string(h
);
388 static void print_subsystems(void)
392 fprintf(stderr
, "hierarchies:");
393 for (i
= 0; i
< num_hierarchies
; i
++) {
395 fprintf(stderr
, " %d: %s\n", i
, hierarchies
[i
]);
399 static bool in_comma_list(const char *needle
, const char *haystack
)
401 const char *s
= haystack
, *e
;
402 size_t nlen
= strlen(needle
);
404 while (*s
&& (e
= index(s
, ','))) {
409 if (strncmp(needle
, s
, nlen
) == 0)
413 if (strcmp(needle
, s
) == 0)
418 /* do we need to do any massaging here? I'm not sure... */
419 static char *find_mounted_controller(const char *controller
)
423 for (i
= 0; i
< num_hierarchies
; i
++) {
426 if (strcmp(hierarchies
[i
], controller
) == 0)
427 return hierarchies
[i
];
428 if (in_comma_list(controller
, hierarchies
[i
]))
429 return hierarchies
[i
];
435 bool cgfs_set_value(const char *controller
, const char *cgroup
, const char *file
,
439 char *fnam
, *tmpc
= find_mounted_controller(controller
);
443 /* basedir / tmpc / cgroup / file \0 */
444 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + strlen(file
) + 4;
446 snprintf(fnam
, len
, "%s/%s/%s/%s", basedir
, tmpc
, cgroup
, file
);
448 return write_string(fnam
, value
);
451 // Chown all the files in the cgroup directory. We do this when we create
452 // a cgroup on behalf of a user.
453 static void chown_all_cgroup_files(const char *dirname
, uid_t uid
, gid_t gid
)
455 struct dirent dirent
, *direntp
;
456 char path
[MAXPATHLEN
];
461 len
= strlen(dirname
);
462 if (len
>= MAXPATHLEN
) {
463 fprintf(stderr
, "chown_all_cgroup_files: pathname too long: %s\n", dirname
);
467 d
= opendir(dirname
);
469 fprintf(stderr
, "chown_all_cgroup_files: failed to open %s\n", dirname
);
473 while (readdir_r(d
, &dirent
, &direntp
) == 0 && direntp
) {
474 if (!strcmp(direntp
->d_name
, ".") || !strcmp(direntp
->d_name
, ".."))
476 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
477 if (ret
< 0 || ret
>= MAXPATHLEN
) {
478 fprintf(stderr
, "chown_all_cgroup_files: pathname too long under %s\n", dirname
);
481 if (chown(path
, uid
, gid
) < 0)
482 fprintf(stderr
, "Failed to chown file %s to %u:%u", path
, uid
, gid
);
487 int cgfs_create(const char *controller
, const char *cg
, uid_t uid
, gid_t gid
)
490 char *dirnam
, *tmpc
= find_mounted_controller(controller
);
494 /* basedir / tmpc / cg \0 */
495 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cg
) + 3;
496 dirnam
= alloca(len
);
497 snprintf(dirnam
, len
, "%s/%s/%s", basedir
,tmpc
, cg
);
499 if (mkdir(dirnam
, 0755) < 0)
502 if (uid
== 0 && gid
== 0)
505 if (chown(dirnam
, uid
, gid
) < 0)
508 chown_all_cgroup_files(dirnam
, uid
, gid
);
513 static bool recursive_rmdir(const char *dirname
)
515 struct dirent dirent
, *direntp
;
518 char pathname
[MAXPATHLEN
];
520 dir
= opendir(dirname
);
523 fprintf(stderr
, "%s: failed to open %s: %s\n", __func__
, dirname
, strerror(errno
));
528 while (!readdir_r(dir
, &dirent
, &direntp
)) {
535 if (!strcmp(direntp
->d_name
, ".") ||
536 !strcmp(direntp
->d_name
, ".."))
539 rc
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
540 if (rc
< 0 || rc
>= MAXPATHLEN
) {
541 fprintf(stderr
, "pathname too long\n");
545 ret
= lstat(pathname
, &mystat
);
548 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
552 if (S_ISDIR(mystat
.st_mode
)) {
553 if (!recursive_rmdir(pathname
)) {
555 fprintf(stderr
, "Error removing %s\n", pathname
);
562 if (closedir(dir
) < 0) {
563 fprintf(stderr
, "%s: failed to close directory %s: %s\n", __func__
, dirname
, strerror(errno
));
567 if (rmdir(dirname
) < 0) {
569 fprintf(stderr
, "%s: failed to delete %s: %s\n", __func__
, dirname
, strerror(errno
));
577 bool cgfs_remove(const char *controller
, const char *cg
)
580 char *dirnam
, *tmpc
= find_mounted_controller(controller
);
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
)
594 char *pathname
, *tmpc
= find_mounted_controller(controller
);
598 /* basedir / tmpc / file \0 */
599 len
= strlen(basedir
) + strlen(tmpc
) + strlen(file
) + 3;
600 pathname
= alloca(len
);
601 snprintf(pathname
, len
, "%s/%s/%s", basedir
, tmpc
, file
);
602 if (chmod(pathname
, mode
) < 0)
607 static int chown_tasks_files(const char *dirname
, uid_t uid
, gid_t gid
)
612 len
= strlen(dirname
) + strlen("/cgroup.procs") + 1;
614 snprintf(fname
, len
, "%s/tasks", dirname
);
615 if (chown(fname
, uid
, gid
) != 0)
617 snprintf(fname
, len
, "%s/cgroup.procs", dirname
);
618 if (chown(fname
, uid
, gid
) != 0)
623 int cgfs_chown_file(const char *controller
, const char *file
, uid_t uid
, gid_t gid
)
626 char *pathname
, *tmpc
= find_mounted_controller(controller
);
630 /* basedir / tmpc / file \0 */
631 len
= strlen(basedir
) + strlen(tmpc
) + strlen(file
) + 3;
632 pathname
= alloca(len
);
633 snprintf(pathname
, len
, "%s/%s/%s", basedir
, tmpc
, file
);
634 if (chown(pathname
, uid
, gid
) < 0)
637 if (is_dir(pathname
))
638 // like cgmanager did, we want to chown the tasks file as well
639 return chown_tasks_files(pathname
, uid
, gid
);
644 FILE *open_pids_file(const char *controller
, const char *cgroup
)
647 char *pathname
, *tmpc
= find_mounted_controller(controller
);
651 /* basedir / tmpc / cgroup / "cgroup.procs" \0 */
652 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + 4 + strlen("cgroup.procs");
653 pathname
= alloca(len
);
654 snprintf(pathname
, len
, "%s/%s/%s/cgroup.procs", basedir
, tmpc
, cgroup
);
655 return fopen(pathname
, "w");
658 bool cgfs_list_children(const char *controller
, const char *cgroup
, char ***list
)
661 char *dirname
, *tmpc
= find_mounted_controller(controller
);
662 char pathname
[MAXPATHLEN
];
663 size_t sz
= 0, asz
= BATCH_SIZE
;
664 struct dirent dirent
, *direntp
;
669 *list
= malloc(asz
* sizeof(char *));
676 /* basedir / tmpc / cgroup \0 */
677 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + 3;
678 dirname
= alloca(len
);
679 snprintf(dirname
, len
, "%s/%s/%s", basedir
, tmpc
, cgroup
);
681 dir
= opendir(dirname
);
685 while (!readdir_r(dir
, &dirent
, &direntp
)) {
692 if (!strcmp(direntp
->d_name
, ".") ||
693 !strcmp(direntp
->d_name
, ".."))
696 rc
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
697 if (rc
< 0 || rc
>= MAXPATHLEN
) {
698 fprintf(stderr
, "%s: pathname too long under %s\n", __func__
, dirname
);
702 ret
= lstat(pathname
, &mystat
);
704 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
707 if (!S_ISDIR(mystat
.st_mode
))
714 tmp
= realloc(*list
, asz
* sizeof(char *));
719 (*list
)[sz
] = strdup(direntp
->d_name
);
720 } while (!(*list
)[sz
]);
721 (*list
)[sz
+1] = NULL
;
724 if (closedir(dir
) < 0) {
725 fprintf(stderr
, "%s: failed closedir for %s: %s\n", __func__
, dirname
, strerror(errno
));
731 void free_key(struct cgfs_files
*k
)
739 void free_keys(struct cgfs_files
**keys
)
745 for (i
= 0; keys
[i
]; i
++) {
751 bool cgfs_get_value(const char *controller
, const char *cgroup
, const char *file
, char **value
)
754 char *fnam
, *tmpc
= find_mounted_controller(controller
);
758 /* basedir / tmpc / cgroup / file \0 */
759 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + strlen(file
) + 4;
761 snprintf(fnam
, len
, "%s/%s/%s/%s", basedir
, tmpc
, cgroup
, file
);
763 *value
= slurp_file(fnam
);
764 return *value
!= NULL
;
767 struct cgfs_files
*cgfs_get_key(const char *controller
, const char *cgroup
, const char *file
)
770 char *fnam
, *tmpc
= find_mounted_controller(controller
);
772 struct cgfs_files
*newkey
;
778 if (file
&& *file
== '/')
781 if (file
&& index(file
, '/'))
784 /* basedir / tmpc / cgroup / file \0 */
785 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + 3;
787 len
+= strlen(file
) + 1;
789 snprintf(fnam
, len
, "%s/%s/%s%s%s", basedir
, tmpc
, cgroup
,
790 file
? "/" : "", file
? file
: "");
792 ret
= stat(fnam
, &sb
);
797 newkey
= malloc(sizeof(struct cgfs_files
));
800 newkey
->name
= must_copy_string(file
);
801 else if (rindex(cgroup
, '/'))
802 newkey
->name
= must_copy_string(rindex(cgroup
, '/'));
804 newkey
->name
= must_copy_string(cgroup
);
805 newkey
->uid
= sb
.st_uid
;
806 newkey
->gid
= sb
.st_gid
;
807 newkey
->mode
= sb
.st_mode
;
812 bool cgfs_list_keys(const char *controller
, const char *cgroup
, struct cgfs_files
***keys
)
815 char *dirname
, *tmpc
= find_mounted_controller(controller
);
816 char pathname
[MAXPATHLEN
];
817 size_t sz
= 0, asz
= 0;
818 struct dirent dirent
, *direntp
;
826 /* basedir / tmpc / cgroup \0 */
827 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + 3;
828 dirname
= alloca(len
);
829 snprintf(dirname
, len
, "%s/%s/%s", basedir
, tmpc
, cgroup
);
831 dir
= opendir(dirname
);
835 while (!readdir_r(dir
, &dirent
, &direntp
)) {
842 if (!strcmp(direntp
->d_name
, ".") ||
843 !strcmp(direntp
->d_name
, ".."))
846 rc
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
847 if (rc
< 0 || rc
>= MAXPATHLEN
) {
848 fprintf(stderr
, "%s: pathname too long under %s\n", __func__
, dirname
);
852 ret
= lstat(pathname
, &mystat
);
854 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
857 if (!S_ISREG(mystat
.st_mode
))
861 struct cgfs_files
**tmp
;
864 tmp
= realloc(*keys
, asz
* sizeof(struct cgfs_files
*));
868 (*keys
)[sz
] = cgfs_get_key(controller
, cgroup
, direntp
->d_name
);
869 (*keys
)[sz
+1] = NULL
;
871 fprintf(stderr
, "%s: Error getting files under %s:%s\n",
872 __func__
, controller
, cgroup
);
877 if (closedir(dir
) < 0) {
878 fprintf(stderr
, "%s: failed closedir for %s: %s\n", __func__
, dirname
, strerror(errno
));
884 bool is_child_cgroup(const char *controller
, const char *cgroup
, const char *f
)
886 char *fnam
, *tmpc
= find_mounted_controller(controller
);
892 /* basedir / tmpc / cgroup / f \0 */
893 len
= strlen(basedir
) + strlen(tmpc
) + strlen(cgroup
) + strlen(f
) + 4;
895 snprintf(fnam
, len
, "%s/%s/%s/%s", basedir
, tmpc
, cgroup
, f
);
897 ret
= stat(fnam
, &sb
);
898 if (ret
< 0 || !S_ISDIR(sb
.st_mode
))
903 #define SEND_CREDS_OK 0
904 #define SEND_CREDS_NOTSK 1
905 #define SEND_CREDS_FAIL 2
906 static bool recv_creds(int sock
, struct ucred
*cred
, char *v
);
907 static int wait_for_pid(pid_t pid
);
908 static int send_creds(int sock
, struct ucred
*cred
, char v
, bool pingfirst
);
911 * fork a task which switches to @task's namespace and writes '1'.
912 * over a unix sock so we can read the task's reaper's pid in our
915 static void write_task_init_pid_exit(int sock
, pid_t target
)
923 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", (int)target
);
924 if (ret
< 0 || ret
>= sizeof(fnam
))
927 fd
= open(fnam
, O_RDONLY
);
929 perror("write_task_init_pid_exit open of ns/pid");
933 perror("write_task_init_pid_exit setns 1");
941 if (!wait_for_pid(pid
))
946 /* we are the child */
951 if (send_creds(sock
, &cred
, v
, true) != SEND_CREDS_OK
)
956 static pid_t
get_init_pid_for_task(pid_t task
)
964 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
965 perror("socketpair");
974 write_task_init_pid_exit(sock
[0], task
);
978 if (!recv_creds(sock
[1], &cred
, &v
))
990 static pid_t
lookup_initpid_in_store(pid_t qpid
)
994 struct pidns_init_store
*e
;
997 snprintf(fnam
, 100, "/proc/%d/ns/pid", qpid
);
999 if (stat(fnam
, &sb
) < 0)
1001 e
= lookup_verify_initpid(&sb
);
1003 answer
= e
->initpid
;
1006 answer
= get_init_pid_for_task(qpid
);
1008 save_initpid(&sb
, answer
);
1011 /* we prune at end in case we are returning
1012 * the value we were about to return */
1013 prune_initpid_store();
1018 static int wait_for_pid(pid_t pid
)
1026 ret
= waitpid(pid
, &status
, 0);
1034 if (!WIFEXITED(status
) || WEXITSTATUS(status
) != 0)
1041 * append pid to *src.
1042 * src: a pointer to a char* in which ot append the pid.
1043 * sz: the number of characters printed so far, minus trailing \0.
1044 * asz: the allocated size so far
1045 * pid: the pid to append
1047 static void must_strcat_pid(char **src
, size_t *sz
, size_t *asz
, pid_t pid
)
1051 int tmplen
= sprintf(tmp
, "%d\n", (int)pid
);
1053 if (!*src
|| tmplen
+ *sz
+ 1 >= *asz
) {
1056 tmp
= realloc(*src
, *asz
+ BUF_RESERVE_SIZE
);
1059 *asz
+= BUF_RESERVE_SIZE
;
1061 memcpy((*src
) +*sz
, tmp
, tmplen
);
1067 * Given a open file * to /proc/pid/{u,g}id_map, and an id
1068 * valid in the caller's namespace, return the id mapped into
1070 * Returns the mapped id, or -1 on error.
1073 convert_id_to_ns(FILE *idfile
, unsigned int in_id
)
1075 unsigned int nsuid
, // base id for a range in the idfile's namespace
1076 hostuid
, // base id for a range in the caller's namespace
1077 count
; // number of ids in this range
1081 fseek(idfile
, 0L, SEEK_SET
);
1082 while (fgets(line
, 400, idfile
)) {
1083 ret
= sscanf(line
, "%u %u %u\n", &nsuid
, &hostuid
, &count
);
1086 if (hostuid
+ count
< hostuid
|| nsuid
+ count
< nsuid
) {
1088 * uids wrapped around - unexpected as this is a procfile,
1091 fprintf(stderr
, "pid wrapparound at entry %u %u %u in %s\n",
1092 nsuid
, hostuid
, count
, line
);
1095 if (hostuid
<= in_id
&& hostuid
+count
> in_id
) {
1097 * now since hostuid <= in_id < hostuid+count, and
1098 * hostuid+count and nsuid+count do not wrap around,
1099 * we know that nsuid+(in_id-hostuid) which must be
1100 * less that nsuid+(count) must not wrap around
1102 return (in_id
- hostuid
) + nsuid
;
1111 * for is_privileged_over,
1112 * specify whether we require the calling uid to be root in his
1115 #define NS_ROOT_REQD true
1116 #define NS_ROOT_OPT false
1120 static bool is_privileged_over(pid_t pid
, uid_t uid
, uid_t victim
, bool req_ns_root
)
1122 char fpath
[PROCLEN
];
1124 bool answer
= false;
1127 if (victim
== -1 || uid
== -1)
1131 * If the request is one not requiring root in the namespace,
1132 * then having the same uid suffices. (i.e. uid 1000 has write
1133 * access to files owned by uid 1000
1135 if (!req_ns_root
&& uid
== victim
)
1138 ret
= snprintf(fpath
, PROCLEN
, "/proc/%d/uid_map", pid
);
1139 if (ret
< 0 || ret
>= PROCLEN
)
1141 FILE *f
= fopen(fpath
, "r");
1145 /* if caller's not root in his namespace, reject */
1146 nsuid
= convert_id_to_ns(f
, uid
);
1151 * If victim is not mapped into caller's ns, reject.
1152 * XXX I'm not sure this check is needed given that fuse
1153 * will be sending requests where the vfs has converted
1155 nsuid
= convert_id_to_ns(f
, victim
);
1166 static bool perms_include(int fmode
, mode_t req_mode
)
1170 switch (req_mode
& O_ACCMODE
) {
1178 r
= S_IROTH
| S_IWOTH
;
1183 return ((fmode
& r
) == r
);
1189 * querycg is /a/b/c/d/e
1192 static char *get_next_cgroup_dir(const char *taskcg
, const char *querycg
)
1196 if (strlen(taskcg
) <= strlen(querycg
)) {
1197 fprintf(stderr
, "%s: I was fed bad input\n", __func__
);
1201 if (strcmp(querycg
, "/") == 0)
1202 start
= strdup(taskcg
+ 1);
1204 start
= strdup(taskcg
+ strlen(querycg
) + 1);
1207 end
= strchr(start
, '/');
1213 static void stripnewline(char *x
)
1215 size_t l
= strlen(x
);
1216 if (l
&& x
[l
-1] == '\n')
1220 static char *get_pid_cgroup(pid_t pid
, const char *contrl
)
1224 char *answer
= NULL
;
1228 const char *h
= find_mounted_controller(contrl
);
1232 ret
= snprintf(fnam
, PROCLEN
, "/proc/%d/cgroup", pid
);
1233 if (ret
< 0 || ret
>= PROCLEN
)
1235 if (!(f
= fopen(fnam
, "r")))
1238 while (getline(&line
, &len
, f
) != -1) {
1242 c1
= strchr(line
, ':');
1246 c2
= strchr(c1
, ':');
1250 if (strcmp(c1
, h
) != 0)
1255 answer
= strdup(c2
);
1267 * check whether a fuse context may access a cgroup dir or file
1269 * If file is not null, it is a cgroup file to check under cg.
1270 * If file is null, then we are checking perms on cg itself.
1272 * For files we can check the mode of the list_keys result.
1273 * For cgroups, we must make assumptions based on the files under the
1274 * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups
1277 static bool fc_may_access(struct fuse_context
*fc
, const char *contrl
, const char *cg
, const char *file
, mode_t mode
)
1279 struct cgfs_files
*k
= NULL
;
1282 k
= cgfs_get_key(contrl
, cg
, file
);
1286 if (is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_OPT
)) {
1287 if (perms_include(k
->mode
>> 6, mode
)) {
1292 if (fc
->gid
== k
->gid
) {
1293 if (perms_include(k
->mode
>> 3, mode
)) {
1298 ret
= perms_include(k
->mode
, mode
);
1305 #define INITSCOPE "/init.scope"
1306 static void prune_init_slice(char *cg
)
1309 size_t cg_len
= strlen(cg
), initscope_len
= strlen(INITSCOPE
);
1311 if (cg_len
< initscope_len
)
1314 point
= cg
+ cg_len
- initscope_len
;
1315 if (strcmp(point
, INITSCOPE
) == 0) {
1324 * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d.
1325 * If pid is in /a, he may act on /a/b, but not on /b.
1326 * if the answer is false and nextcg is not NULL, then *nextcg will point
1327 * to a string containing the next cgroup directory under cg, which must be
1328 * freed by the caller.
1330 static bool caller_is_in_ancestor(pid_t pid
, const char *contrl
, const char *cg
, char **nextcg
)
1332 bool answer
= false;
1333 char *c2
= get_pid_cgroup(pid
, contrl
);
1338 prune_init_slice(c2
);
1341 * callers pass in '/' for root cgroup, otherwise they pass
1342 * in a cgroup without leading '/'
1344 linecmp
= *cg
== '/' ? c2
: c2
+1;
1345 if (strncmp(linecmp
, cg
, strlen(linecmp
)) != 0) {
1347 *nextcg
= get_next_cgroup_dir(linecmp
, cg
);
1359 * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c.
1361 static bool caller_may_see_dir(pid_t pid
, const char *contrl
, const char *cg
)
1363 bool answer
= false;
1365 size_t target_len
, task_len
;
1367 if (strcmp(cg
, "/") == 0)
1370 c2
= get_pid_cgroup(pid
, contrl
);
1373 prune_init_slice(c2
);
1376 target_len
= strlen(cg
);
1377 task_len
= strlen(task_cg
);
1378 if (task_len
== 0) {
1379 /* Task is in the root cg, it can see everything. This case is
1380 * not handled by the strmcps below, since they test for the
1381 * last /, but that is the first / that we've chopped off
1387 if (strcmp(cg
, task_cg
) == 0) {
1391 if (target_len
< task_len
) {
1392 /* looking up a parent dir */
1393 if (strncmp(task_cg
, cg
, target_len
) == 0 && task_cg
[target_len
] == '/')
1397 if (target_len
> task_len
) {
1398 /* looking up a child dir */
1399 if (strncmp(task_cg
, cg
, task_len
) == 0 && cg
[task_len
] == '/')
1410 * given /cgroup/freezer/a/b, return "freezer".
1411 * the returned char* should NOT be freed.
1413 static char *pick_controller_from_path(struct fuse_context
*fc
, const char *path
)
1416 char *contr
, *slash
;
1418 if (strlen(path
) < 9)
1420 if (*(path
+7) != '/')
1423 contr
= strdupa(p1
);
1426 slash
= strstr(contr
, "/");
1431 for (i
= 0; i
< num_hierarchies
; i
++) {
1432 if (hierarchies
[i
] && strcmp(hierarchies
[i
], contr
) == 0)
1433 return hierarchies
[i
];
1439 * Find the start of cgroup in /cgroup/controller/the/cgroup/path
1440 * Note that the returned value may include files (keynames) etc
1442 static const char *find_cgroup_in_path(const char *path
)
1446 if (strlen(path
) < 9)
1448 p1
= strstr(path
+8, "/");
1455 * split the last path element from the path in @cg.
1456 * @dir is newly allocated and should be freed, @last not
1458 static void get_cgdir_and_path(const char *cg
, char **dir
, char **last
)
1465 *last
= strrchr(cg
, '/');
1470 p
= strrchr(*dir
, '/');
1475 * FUSE ops for /cgroup
1478 int cg_getattr(const char *path
, struct stat
*sb
)
1480 struct timespec now
;
1481 struct fuse_context
*fc
= fuse_get_context();
1482 char * cgdir
= NULL
;
1483 char *last
= NULL
, *path1
, *path2
;
1484 struct cgfs_files
*k
= NULL
;
1486 const char *controller
= NULL
;
1493 memset(sb
, 0, sizeof(struct stat
));
1495 if (clock_gettime(CLOCK_REALTIME
, &now
) < 0)
1498 sb
->st_uid
= sb
->st_gid
= 0;
1499 sb
->st_atim
= sb
->st_mtim
= sb
->st_ctim
= now
;
1502 if (strcmp(path
, "/cgroup") == 0) {
1503 sb
->st_mode
= S_IFDIR
| 00755;
1508 controller
= pick_controller_from_path(fc
, path
);
1511 cgroup
= find_cgroup_in_path(path
);
1513 /* this is just /cgroup/controller, return it as a dir */
1514 sb
->st_mode
= S_IFDIR
| 00755;
1519 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1529 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1532 /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys.
1533 * Then check that caller's cgroup is under path if last is a child
1534 * cgroup, or cgdir if last is a file */
1536 if (is_child_cgroup(controller
, path1
, path2
)) {
1537 if (!caller_may_see_dir(initpid
, controller
, cgroup
)) {
1541 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, NULL
)) {
1542 /* this is just /cgroup/controller, return it as a dir */
1543 sb
->st_mode
= S_IFDIR
| 00555;
1548 if (!fc_may_access(fc
, controller
, cgroup
, NULL
, O_RDONLY
)) {
1553 // get uid, gid, from '/tasks' file and make up a mode
1554 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
1555 sb
->st_mode
= S_IFDIR
| 00755;
1556 k
= cgfs_get_key(controller
, cgroup
, NULL
);
1558 sb
->st_uid
= sb
->st_gid
= 0;
1560 sb
->st_uid
= k
->uid
;
1561 sb
->st_gid
= k
->gid
;
1569 if ((k
= cgfs_get_key(controller
, path1
, path2
)) != NULL
) {
1570 sb
->st_mode
= S_IFREG
| k
->mode
;
1572 sb
->st_uid
= k
->uid
;
1573 sb
->st_gid
= k
->gid
;
1576 if (!caller_is_in_ancestor(initpid
, controller
, path1
, NULL
)) {
1580 if (!fc_may_access(fc
, controller
, path1
, path2
, O_RDONLY
)) {
1593 int cg_opendir(const char *path
, struct fuse_file_info
*fi
)
1595 struct fuse_context
*fc
= fuse_get_context();
1597 struct file_info
*dir_info
;
1598 char *controller
= NULL
;
1603 if (strcmp(path
, "/cgroup") == 0) {
1607 // return list of keys for the controller, and list of child cgroups
1608 controller
= pick_controller_from_path(fc
, path
);
1612 cgroup
= find_cgroup_in_path(path
);
1614 /* this is just /cgroup/controller, return its contents */
1619 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1623 if (!caller_may_see_dir(initpid
, controller
, cgroup
))
1625 if (!fc_may_access(fc
, controller
, cgroup
, NULL
, O_RDONLY
))
1629 /* we'll free this at cg_releasedir */
1630 dir_info
= malloc(sizeof(*dir_info
));
1633 dir_info
->controller
= must_copy_string(controller
);
1634 dir_info
->cgroup
= must_copy_string(cgroup
);
1635 dir_info
->type
= LXC_TYPE_CGDIR
;
1636 dir_info
->buf
= NULL
;
1637 dir_info
->file
= NULL
;
1638 dir_info
->buflen
= 0;
1640 fi
->fh
= (unsigned long)dir_info
;
1644 int cg_readdir(const char *path
, void *buf
, fuse_fill_dir_t filler
, off_t offset
,
1645 struct fuse_file_info
*fi
)
1647 struct file_info
*d
= (struct file_info
*)fi
->fh
;
1648 struct cgfs_files
**list
= NULL
;
1650 char *nextcg
= NULL
;
1651 struct fuse_context
*fc
= fuse_get_context();
1652 char **clist
= NULL
;
1654 if (d
->type
!= LXC_TYPE_CGDIR
) {
1655 fprintf(stderr
, "Internal error: file cache info used in readdir\n");
1658 if (!d
->cgroup
&& !d
->controller
) {
1659 // ls /var/lib/lxcfs/cgroup - just show list of controllers
1662 for (i
= 0; i
< num_hierarchies
; i
++) {
1663 if (hierarchies
[i
] && filler(buf
, hierarchies
[i
], NULL
, 0) != 0) {
1670 if (!cgfs_list_keys(d
->controller
, d
->cgroup
, &list
)) {
1671 // not a valid cgroup
1676 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1679 if (!caller_is_in_ancestor(initpid
, d
->controller
, d
->cgroup
, &nextcg
)) {
1681 ret
= filler(buf
, nextcg
, NULL
, 0);
1692 for (i
= 0; list
[i
]; i
++) {
1693 if (filler(buf
, list
[i
]->name
, NULL
, 0) != 0) {
1699 // now get the list of child cgroups
1701 if (!cgfs_list_children(d
->controller
, d
->cgroup
, &clist
)) {
1705 for (i
= 0; clist
[i
]; i
++) {
1706 if (filler(buf
, clist
[i
], NULL
, 0) != 0) {
1716 for (i
= 0; clist
[i
]; i
++)
1723 static void do_release_file_info(struct file_info
*f
)
1727 free(f
->controller
);
1734 int cg_releasedir(const char *path
, struct fuse_file_info
*fi
)
1736 struct file_info
*d
= (struct file_info
*)fi
->fh
;
1738 do_release_file_info(d
);
1742 int cg_open(const char *path
, struct fuse_file_info
*fi
)
1745 char *last
= NULL
, *path1
, *path2
, * cgdir
= NULL
, *controller
;
1746 struct cgfs_files
*k
= NULL
;
1747 struct file_info
*file_info
;
1748 struct fuse_context
*fc
= fuse_get_context();
1754 controller
= pick_controller_from_path(fc
, path
);
1757 cgroup
= find_cgroup_in_path(path
);
1761 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1770 k
= cgfs_get_key(controller
, path1
, path2
);
1777 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1780 if (!caller_may_see_dir(initpid
, controller
, path1
)) {
1784 if (!fc_may_access(fc
, controller
, path1
, path2
, fi
->flags
)) {
1785 // should never get here
1790 /* we'll free this at cg_release */
1791 file_info
= malloc(sizeof(*file_info
));
1796 file_info
->controller
= must_copy_string(controller
);
1797 file_info
->cgroup
= must_copy_string(path1
);
1798 file_info
->file
= must_copy_string(path2
);
1799 file_info
->type
= LXC_TYPE_CGFILE
;
1800 file_info
->buf
= NULL
;
1801 file_info
->buflen
= 0;
1803 fi
->fh
= (unsigned long)file_info
;
1811 int cg_release(const char *path
, struct fuse_file_info
*fi
)
1813 struct file_info
*f
= (struct file_info
*)fi
->fh
;
1815 do_release_file_info(f
);
1819 #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP )
1821 static bool wait_for_sock(int sock
, int timeout
)
1823 struct epoll_event ev
;
1824 int epfd
, ret
, now
, starttime
, deltatime
, saved_errno
;
1826 if ((starttime
= time(NULL
)) < 0)
1829 if ((epfd
= epoll_create(1)) < 0) {
1830 fprintf(stderr
, "Failed to create epoll socket: %m\n");
1834 ev
.events
= POLLIN_SET
;
1836 if (epoll_ctl(epfd
, EPOLL_CTL_ADD
, sock
, &ev
) < 0) {
1837 fprintf(stderr
, "Failed adding socket to epoll: %m\n");
1843 if ((now
= time(NULL
)) < 0) {
1848 deltatime
= (starttime
+ timeout
) - now
;
1849 if (deltatime
< 0) { // timeout
1854 ret
= epoll_wait(epfd
, &ev
, 1, 1000*deltatime
+ 1);
1855 if (ret
< 0 && errno
== EINTR
)
1857 saved_errno
= errno
;
1861 errno
= saved_errno
;
1867 static int msgrecv(int sockfd
, void *buf
, size_t len
)
1869 if (!wait_for_sock(sockfd
, 2))
1871 return recv(sockfd
, buf
, len
, MSG_DONTWAIT
);
1874 static int send_creds(int sock
, struct ucred
*cred
, char v
, bool pingfirst
)
1876 struct msghdr msg
= { 0 };
1878 struct cmsghdr
*cmsg
;
1879 char cmsgbuf
[CMSG_SPACE(sizeof(*cred
))];
1884 if (msgrecv(sock
, buf
, 1) != 1) {
1885 fprintf(stderr
, "%s: Error getting reply from server over socketpair\n",
1887 return SEND_CREDS_FAIL
;
1891 msg
.msg_control
= cmsgbuf
;
1892 msg
.msg_controllen
= sizeof(cmsgbuf
);
1894 cmsg
= CMSG_FIRSTHDR(&msg
);
1895 cmsg
->cmsg_len
= CMSG_LEN(sizeof(struct ucred
));
1896 cmsg
->cmsg_level
= SOL_SOCKET
;
1897 cmsg
->cmsg_type
= SCM_CREDENTIALS
;
1898 memcpy(CMSG_DATA(cmsg
), cred
, sizeof(*cred
));
1900 msg
.msg_name
= NULL
;
1901 msg
.msg_namelen
= 0;
1905 iov
.iov_len
= sizeof(buf
);
1909 if (sendmsg(sock
, &msg
, 0) < 0) {
1910 fprintf(stderr
, "%s: failed at sendmsg: %s\n", __func__
,
1913 return SEND_CREDS_NOTSK
;
1914 return SEND_CREDS_FAIL
;
1917 return SEND_CREDS_OK
;
1920 static bool recv_creds(int sock
, struct ucred
*cred
, char *v
)
1922 struct msghdr msg
= { 0 };
1924 struct cmsghdr
*cmsg
;
1925 char cmsgbuf
[CMSG_SPACE(sizeof(*cred
))];
1936 if (setsockopt(sock
, SOL_SOCKET
, SO_PASSCRED
, &optval
, sizeof(optval
)) == -1) {
1937 fprintf(stderr
, "Failed to set passcred: %s\n", strerror(errno
));
1941 if (write(sock
, buf
, 1) != 1) {
1942 fprintf(stderr
, "Failed to start write on scm fd: %s\n", strerror(errno
));
1946 msg
.msg_name
= NULL
;
1947 msg
.msg_namelen
= 0;
1948 msg
.msg_control
= cmsgbuf
;
1949 msg
.msg_controllen
= sizeof(cmsgbuf
);
1952 iov
.iov_len
= sizeof(buf
);
1956 if (!wait_for_sock(sock
, 2)) {
1957 fprintf(stderr
, "Timed out waiting for scm_cred: %s\n",
1961 ret
= recvmsg(sock
, &msg
, MSG_DONTWAIT
);
1963 fprintf(stderr
, "Failed to receive scm_cred: %s\n",
1968 cmsg
= CMSG_FIRSTHDR(&msg
);
1970 if (cmsg
&& cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
)) &&
1971 cmsg
->cmsg_level
== SOL_SOCKET
&&
1972 cmsg
->cmsg_type
== SCM_CREDENTIALS
) {
1973 memcpy(cred
, CMSG_DATA(cmsg
), sizeof(*cred
));
1982 * pid_to_ns - reads pids from a ucred over a socket, then writes the
1983 * int value back over the socket. This shifts the pid from the
1984 * sender's pidns into tpid's pidns.
1986 static void pid_to_ns(int sock
, pid_t tpid
)
1991 while (recv_creds(sock
, &cred
, &v
)) {
1994 if (write(sock
, &cred
.pid
, sizeof(pid_t
)) != sizeof(pid_t
))
2001 * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain
2002 * in your old pidns. Only children which you fork will be in the target
2003 * pidns. So the pid_to_ns_wrapper does the setns, then forks a child to
2004 * actually convert pids
2006 static void pid_to_ns_wrapper(int sock
, pid_t tpid
)
2008 int newnsfd
= -1, ret
, cpipe
[2];
2013 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", tpid
);
2014 if (ret
< 0 || ret
>= sizeof(fnam
))
2016 newnsfd
= open(fnam
, O_RDONLY
);
2019 if (setns(newnsfd
, 0) < 0)
2023 if (pipe(cpipe
) < 0)
2033 if (write(cpipe
[1], &b
, sizeof(char)) < 0) {
2034 fprintf(stderr
, "%s (child): erorr on write: %s\n",
2035 __func__
, strerror(errno
));
2038 pid_to_ns(sock
, tpid
);
2039 _exit(1); // not reached
2041 // give the child 1 second to be done forking and
2043 if (!wait_for_sock(cpipe
[0], 1))
2045 ret
= read(cpipe
[0], &v
, 1);
2046 if (ret
!= sizeof(char) || v
!= '1')
2049 if (!wait_for_pid(cpid
))
2055 * To read cgroup files with a particular pid, we will setns into the child
2056 * pidns, open a pipe, fork a child - which will be the first to really be in
2057 * the child ns - which does the cgfs_get_value and writes the data to the pipe.
2059 bool do_read_pids(pid_t tpid
, const char *contrl
, const char *cg
, const char *file
, char **d
)
2061 int sock
[2] = {-1, -1};
2062 char *tmpdata
= NULL
;
2064 pid_t qpid
, cpid
= -1;
2065 bool answer
= false;
2068 size_t sz
= 0, asz
= 0;
2070 if (!cgfs_get_value(contrl
, cg
, file
, &tmpdata
))
2074 * Now we read the pids from returned data one by one, pass
2075 * them into a child in the target namespace, read back the
2076 * translated pids, and put them into our to-return data
2079 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
2080 perror("socketpair");
2089 if (!cpid
) // child - exits when done
2090 pid_to_ns_wrapper(sock
[1], tpid
);
2092 char *ptr
= tmpdata
;
2095 while (sscanf(ptr
, "%d\n", &qpid
) == 1) {
2097 ret
= send_creds(sock
[0], &cred
, v
, true);
2099 if (ret
== SEND_CREDS_NOTSK
)
2101 if (ret
== SEND_CREDS_FAIL
)
2104 // read converted results
2105 if (!wait_for_sock(sock
[0], 2)) {
2106 fprintf(stderr
, "%s: timed out waiting for pid from child: %s\n",
2107 __func__
, strerror(errno
));
2110 if (read(sock
[0], &qpid
, sizeof(qpid
)) != sizeof(qpid
)) {
2111 fprintf(stderr
, "%s: error reading pid from child: %s\n",
2112 __func__
, strerror(errno
));
2115 must_strcat_pid(d
, &sz
, &asz
, qpid
);
2117 ptr
= strchr(ptr
, '\n');
2123 cred
.pid
= getpid();
2125 if (send_creds(sock
[0], &cred
, v
, true) != SEND_CREDS_OK
) {
2126 // failed to ask child to exit
2127 fprintf(stderr
, "%s: failed to ask child to exit: %s\n",
2128 __func__
, strerror(errno
));
2138 if (sock
[0] != -1) {
2145 int cg_read(const char *path
, char *buf
, size_t size
, off_t offset
,
2146 struct fuse_file_info
*fi
)
2148 struct fuse_context
*fc
= fuse_get_context();
2149 struct file_info
*f
= (struct file_info
*)fi
->fh
;
2150 struct cgfs_files
*k
= NULL
;
2155 if (f
->type
!= LXC_TYPE_CGFILE
) {
2156 fprintf(stderr
, "Internal error: directory cache info used in cg_read\n");
2169 if ((k
= cgfs_get_key(f
->controller
, f
->cgroup
, f
->file
)) == NULL
) {
2175 if (!fc_may_access(fc
, f
->controller
, f
->cgroup
, f
->file
, O_RDONLY
)) { // should never get here
2180 if (strcmp(f
->file
, "tasks") == 0 ||
2181 strcmp(f
->file
, "/tasks") == 0 ||
2182 strcmp(f
->file
, "/cgroup.procs") == 0 ||
2183 strcmp(f
->file
, "cgroup.procs") == 0)
2184 // special case - we have to translate the pids
2185 r
= do_read_pids(fc
->pid
, f
->controller
, f
->cgroup
, f
->file
, &data
);
2187 r
= cgfs_get_value(f
->controller
, f
->cgroup
, f
->file
, &data
);
2201 memcpy(buf
, data
, s
);
2202 if (s
> 0 && s
< size
&& data
[s
-1] != '\n')
2212 static void pid_from_ns(int sock
, pid_t tpid
)
2222 if (!wait_for_sock(sock
, 2)) {
2223 fprintf(stderr
, "%s: timeout reading from parent\n", __func__
);
2226 if ((ret
= read(sock
, &vpid
, sizeof(pid_t
))) != sizeof(pid_t
)) {
2227 fprintf(stderr
, "%s: bad read from parent: %s\n",
2228 __func__
, strerror(errno
));
2231 if (vpid
== -1) // done
2235 if (send_creds(sock
, &cred
, v
, true) != SEND_CREDS_OK
) {
2237 cred
.pid
= getpid();
2238 if (send_creds(sock
, &cred
, v
, false) != SEND_CREDS_OK
)
2245 static void pid_from_ns_wrapper(int sock
, pid_t tpid
)
2247 int newnsfd
= -1, ret
, cpipe
[2];
2252 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", tpid
);
2253 if (ret
< 0 || ret
>= sizeof(fnam
))
2255 newnsfd
= open(fnam
, O_RDONLY
);
2258 if (setns(newnsfd
, 0) < 0)
2262 if (pipe(cpipe
) < 0)
2274 if (write(cpipe
[1], &b
, sizeof(char)) < 0) {
2275 fprintf(stderr
, "%s (child): erorr on write: %s\n",
2276 __func__
, strerror(errno
));
2279 pid_from_ns(sock
, tpid
);
2282 // give the child 1 second to be done forking and
2284 if (!wait_for_sock(cpipe
[0], 1))
2286 ret
= read(cpipe
[0], &v
, 1);
2287 if (ret
!= sizeof(char) || v
!= '1') {
2291 if (!wait_for_pid(cpid
))
2296 kill(cpid
, SIGKILL
);
2302 * Given host @uid, return the uid to which it maps in
2303 * @pid's user namespace, or -1 if none.
2305 bool hostuid_to_ns(uid_t uid
, pid_t pid
, uid_t
*answer
)
2310 sprintf(line
, "/proc/%d/uid_map", pid
);
2311 if ((f
= fopen(line
, "r")) == NULL
) {
2315 *answer
= convert_id_to_ns(f
, uid
);
2324 * get_pid_creds: get the real uid and gid of @pid from
2326 * (XXX should we use euid here?)
2328 void get_pid_creds(pid_t pid
, uid_t
*uid
, gid_t
*gid
)
2337 sprintf(line
, "/proc/%d/status", pid
);
2338 if ((f
= fopen(line
, "r")) == NULL
) {
2339 fprintf(stderr
, "Error opening %s: %s\n", line
, strerror(errno
));
2342 while (fgets(line
, 400, f
)) {
2343 if (strncmp(line
, "Uid:", 4) == 0) {
2344 if (sscanf(line
+4, "%u", &u
) != 1) {
2345 fprintf(stderr
, "bad uid line for pid %u\n", pid
);
2350 } else if (strncmp(line
, "Gid:", 4) == 0) {
2351 if (sscanf(line
+4, "%u", &g
) != 1) {
2352 fprintf(stderr
, "bad gid line for pid %u\n", pid
);
2363 * May the requestor @r move victim @v to a new cgroup?
2364 * This is allowed if
2365 * . they are the same task
2366 * . they are ownedy by the same uid
2367 * . @r is root on the host, or
2368 * . @v's uid is mapped into @r's where @r is root.
2370 bool may_move_pid(pid_t r
, uid_t r_uid
, pid_t v
)
2372 uid_t v_uid
, tmpuid
;
2379 get_pid_creds(v
, &v_uid
, &v_gid
);
2382 if (hostuid_to_ns(r_uid
, r
, &tmpuid
) && tmpuid
== 0
2383 && hostuid_to_ns(v_uid
, r
, &tmpuid
))
2388 static bool do_write_pids(pid_t tpid
, uid_t tuid
, const char *contrl
, const char *cg
,
2389 const char *file
, const char *buf
)
2391 int sock
[2] = {-1, -1};
2392 pid_t qpid
, cpid
= -1;
2393 FILE *pids_file
= NULL
;
2394 bool answer
= false, fail
= false;
2396 pids_file
= open_pids_file(contrl
, cg
);
2401 * write the pids to a socket, have helper in writer's pidns
2402 * call movepid for us
2404 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
2405 perror("socketpair");
2413 if (!cpid
) { // child
2415 pid_from_ns_wrapper(sock
[1], tpid
);
2418 const char *ptr
= buf
;
2419 while (sscanf(ptr
, "%d", &qpid
) == 1) {
2423 if (write(sock
[0], &qpid
, sizeof(qpid
)) != sizeof(qpid
)) {
2424 fprintf(stderr
, "%s: error writing pid to child: %s\n",
2425 __func__
, strerror(errno
));
2429 if (recv_creds(sock
[0], &cred
, &v
)) {
2431 if (!may_move_pid(tpid
, tuid
, cred
.pid
)) {
2435 if (fprintf(pids_file
, "%d", (int) cred
.pid
) < 0)
2440 ptr
= strchr(ptr
, '\n');
2446 /* All good, write the value */
2448 if (write(sock
[0], &qpid
,sizeof(qpid
)) != sizeof(qpid
))
2449 fprintf(stderr
, "Warning: failed to ask child to exit\n");
2457 if (sock
[0] != -1) {
2462 if (fclose(pids_file
) != 0)
2468 int cg_write(const char *path
, const char *buf
, size_t size
, off_t offset
,
2469 struct fuse_file_info
*fi
)
2471 struct fuse_context
*fc
= fuse_get_context();
2472 char *localbuf
= NULL
;
2473 struct cgfs_files
*k
= NULL
;
2474 struct file_info
*f
= (struct file_info
*)fi
->fh
;
2477 if (f
->type
!= LXC_TYPE_CGFILE
) {
2478 fprintf(stderr
, "Internal error: directory cache info used in cg_write\n");
2488 localbuf
= alloca(size
+1);
2489 localbuf
[size
] = '\0';
2490 memcpy(localbuf
, buf
, size
);
2492 if ((k
= cgfs_get_key(f
->controller
, f
->cgroup
, f
->file
)) == NULL
) {
2497 if (!fc_may_access(fc
, f
->controller
, f
->cgroup
, f
->file
, O_WRONLY
)) {
2502 if (strcmp(f
->file
, "tasks") == 0 ||
2503 strcmp(f
->file
, "/tasks") == 0 ||
2504 strcmp(f
->file
, "/cgroup.procs") == 0 ||
2505 strcmp(f
->file
, "cgroup.procs") == 0)
2506 // special case - we have to translate the pids
2507 r
= do_write_pids(fc
->pid
, fc
->uid
, f
->controller
, f
->cgroup
, f
->file
, localbuf
);
2509 r
= cgfs_set_value(f
->controller
, f
->cgroup
, f
->file
, localbuf
);
2519 int cg_chown(const char *path
, uid_t uid
, gid_t gid
)
2521 struct fuse_context
*fc
= fuse_get_context();
2522 char *cgdir
= NULL
, *last
= NULL
, *path1
, *path2
, *controller
;
2523 struct cgfs_files
*k
= NULL
;
2530 if (strcmp(path
, "/cgroup") == 0)
2533 controller
= pick_controller_from_path(fc
, path
);
2536 cgroup
= find_cgroup_in_path(path
);
2538 /* this is just /cgroup/controller */
2541 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2551 if (is_child_cgroup(controller
, path1
, path2
)) {
2552 // get uid, gid, from '/tasks' file and make up a mode
2553 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
2554 k
= cgfs_get_key(controller
, cgroup
, "tasks");
2557 k
= cgfs_get_key(controller
, path1
, path2
);
2565 * This being a fuse request, the uid and gid must be valid
2566 * in the caller's namespace. So we can just check to make
2567 * sure that the caller is root in his uid, and privileged
2568 * over the file's current owner.
2570 if (!is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_REQD
)) {
2575 ret
= cgfs_chown_file(controller
, cgroup
, uid
, gid
);
2584 int cg_chmod(const char *path
, mode_t mode
)
2586 struct fuse_context
*fc
= fuse_get_context();
2587 char * cgdir
= NULL
, *last
= NULL
, *path1
, *path2
, *controller
;
2588 struct cgfs_files
*k
= NULL
;
2595 if (strcmp(path
, "/cgroup") == 0)
2598 controller
= pick_controller_from_path(fc
, path
);
2601 cgroup
= find_cgroup_in_path(path
);
2603 /* this is just /cgroup/controller */
2606 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2616 if (is_child_cgroup(controller
, path1
, path2
)) {
2617 // get uid, gid, from '/tasks' file and make up a mode
2618 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
2619 k
= cgfs_get_key(controller
, cgroup
, "tasks");
2622 k
= cgfs_get_key(controller
, path1
, path2
);
2630 * This being a fuse request, the uid and gid must be valid
2631 * in the caller's namespace. So we can just check to make
2632 * sure that the caller is root in his uid, and privileged
2633 * over the file's current owner.
2635 if (!is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_OPT
)) {
2640 if (!cgfs_chmod_file(controller
, cgroup
, mode
)) {
2652 int cg_mkdir(const char *path
, mode_t mode
)
2654 struct fuse_context
*fc
= fuse_get_context();
2655 char *last
= NULL
, *path1
, *cgdir
= NULL
, *controller
, *next
= NULL
;
2663 controller
= pick_controller_from_path(fc
, path
);
2667 cgroup
= find_cgroup_in_path(path
);
2671 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2677 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2680 if (!caller_is_in_ancestor(initpid
, controller
, path1
, &next
)) {
2683 else if (last
&& strcmp(next
, last
) == 0)
2690 if (!fc_may_access(fc
, controller
, path1
, NULL
, O_RDWR
)) {
2694 if (!caller_is_in_ancestor(initpid
, controller
, path1
, NULL
)) {
2699 ret
= cgfs_create(controller
, cgroup
, fc
->uid
, fc
->gid
);
2707 int cg_rmdir(const char *path
)
2709 struct fuse_context
*fc
= fuse_get_context();
2710 char *last
= NULL
, *cgdir
= NULL
, *controller
, *next
= NULL
;
2717 controller
= pick_controller_from_path(fc
, path
);
2721 cgroup
= find_cgroup_in_path(path
);
2725 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2731 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2734 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, &next
)) {
2735 if (!last
|| strcmp(next
, last
) == 0)
2742 if (!fc_may_access(fc
, controller
, cgdir
, NULL
, O_WRONLY
)) {
2746 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, NULL
)) {
2751 if (!cgfs_remove(controller
, cgroup
)) {
2764 static bool startswith(const char *line
, const char *pref
)
2766 if (strncmp(line
, pref
, strlen(pref
)) == 0)
2771 static void get_mem_cached(char *memstat
, unsigned long *v
)
2777 if (startswith(memstat
, "total_cache")) {
2778 sscanf(memstat
+ 11, "%lu", v
);
2782 eol
= strchr(memstat
, '\n');
2789 static void get_blkio_io_value(char *str
, unsigned major
, unsigned minor
, char *iotype
, unsigned long *v
)
2795 snprintf(key
, 32, "%u:%u %s", major
, minor
, iotype
);
2797 size_t len
= strlen(key
);
2801 if (startswith(str
, key
)) {
2802 sscanf(str
+ len
, "%lu", v
);
2805 eol
= strchr(str
, '\n');
2812 static int read_file(const char *path
, char *buf
, size_t size
,
2813 struct file_info
*d
)
2815 size_t linelen
= 0, total_len
= 0, rv
= 0;
2817 char *cache
= d
->buf
;
2818 size_t cache_size
= d
->buflen
;
2819 FILE *f
= fopen(path
, "r");
2823 while (getline(&line
, &linelen
, f
) != -1) {
2824 size_t l
= snprintf(cache
, cache_size
, "%s", line
);
2826 perror("Error writing to cache");
2830 if (l
>= cache_size
) {
2831 fprintf(stderr
, "Internal error: truncated write to cache\n");
2840 d
->size
= total_len
;
2841 if (total_len
> size
) total_len
= size
;
2843 /* read from off 0 */
2844 memcpy(buf
, d
->buf
, total_len
);
2853 * FUSE ops for /proc
2856 static unsigned long get_memlimit(const char *cgroup
)
2858 char *memlimit_str
= NULL
;
2859 unsigned long memlimit
= -1;
2861 if (cgfs_get_value("memory", cgroup
, "memory.limit_in_bytes", &memlimit_str
))
2862 memlimit
= strtoul(memlimit_str
, NULL
, 10);
2869 static unsigned long get_min_memlimit(const char *cgroup
)
2871 char *copy
= strdupa(cgroup
);
2872 unsigned long memlimit
= 0, retlimit
;
2874 retlimit
= get_memlimit(copy
);
2876 while (strcmp(copy
, "/") != 0) {
2877 copy
= dirname(copy
);
2878 memlimit
= get_memlimit(copy
);
2879 if (memlimit
!= -1 && memlimit
< retlimit
)
2880 retlimit
= memlimit
;
2886 static int proc_meminfo_read(char *buf
, size_t size
, off_t offset
,
2887 struct fuse_file_info
*fi
)
2889 struct fuse_context
*fc
= fuse_get_context();
2890 struct file_info
*d
= (struct file_info
*)fi
->fh
;
2892 char *memusage_str
= NULL
, *memstat_str
= NULL
,
2893 *memswlimit_str
= NULL
, *memswusage_str
= NULL
,
2894 *memswlimit_default_str
= NULL
, *memswusage_default_str
= NULL
;
2895 unsigned long memlimit
= 0, memusage
= 0, memswlimit
= 0, memswusage
= 0,
2896 cached
= 0, hosttotal
= 0;
2898 size_t linelen
= 0, total_len
= 0, rv
= 0;
2899 char *cache
= d
->buf
;
2900 size_t cache_size
= d
->buflen
;
2904 if (offset
> d
->size
)
2908 int left
= d
->size
- offset
;
2909 total_len
= left
> size
? size
: left
;
2910 memcpy(buf
, cache
+ offset
, total_len
);
2914 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2917 cg
= get_pid_cgroup(initpid
, "memory");
2919 return read_file("/proc/meminfo", buf
, size
, d
);
2921 memlimit
= get_min_memlimit(cg
);
2922 if (!cgfs_get_value("memory", cg
, "memory.usage_in_bytes", &memusage_str
))
2924 if (!cgfs_get_value("memory", cg
, "memory.stat", &memstat_str
))
2927 // Following values are allowed to fail, because swapaccount might be turned
2928 // off for current kernel
2929 if(cgfs_get_value("memory", cg
, "memory.memsw.limit_in_bytes", &memswlimit_str
) &&
2930 cgfs_get_value("memory", cg
, "memory.memsw.usage_in_bytes", &memswusage_str
))
2932 /* If swapaccounting is turned on, then default value is assumed to be that of cgroup / */
2933 if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str
))
2935 if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str
))
2938 memswlimit
= strtoul(memswlimit_str
, NULL
, 10);
2939 memswusage
= strtoul(memswusage_str
, NULL
, 10);
2941 if (!strcmp(memswlimit_str
, memswlimit_default_str
))
2943 if (!strcmp(memswusage_str
, memswusage_default_str
))
2946 memswlimit
= memswlimit
/ 1024;
2947 memswusage
= memswusage
/ 1024;
2950 memusage
= strtoul(memusage_str
, NULL
, 10);
2954 get_mem_cached(memstat_str
, &cached
);
2956 f
= fopen("/proc/meminfo", "r");
2960 while (getline(&line
, &linelen
, f
) != -1) {
2962 char *printme
, lbuf
[100];
2964 memset(lbuf
, 0, 100);
2965 if (startswith(line
, "MemTotal:")) {
2966 sscanf(line
+14, "%lu", &hosttotal
);
2967 if (hosttotal
< memlimit
)
2968 memlimit
= hosttotal
;
2969 snprintf(lbuf
, 100, "MemTotal: %8lu kB\n", memlimit
);
2971 } else if (startswith(line
, "MemFree:")) {
2972 snprintf(lbuf
, 100, "MemFree: %8lu kB\n", memlimit
- memusage
);
2974 } else if (startswith(line
, "MemAvailable:")) {
2975 snprintf(lbuf
, 100, "MemAvailable: %8lu kB\n", memlimit
- memusage
);
2977 } else if (startswith(line
, "SwapTotal:") && memswlimit
> 0) {
2978 snprintf(lbuf
, 100, "SwapTotal: %8lu kB\n", memswlimit
- memlimit
);
2980 } else if (startswith(line
, "SwapFree:") && memswlimit
> 0 && memswusage
> 0) {
2981 snprintf(lbuf
, 100, "SwapFree: %8lu kB\n",
2982 (memswlimit
- memlimit
) - (memswusage
- memusage
));
2984 } else if (startswith(line
, "Buffers:")) {
2985 snprintf(lbuf
, 100, "Buffers: %8lu kB\n", 0UL);
2987 } else if (startswith(line
, "Cached:")) {
2988 snprintf(lbuf
, 100, "Cached: %8lu kB\n", cached
);
2990 } else if (startswith(line
, "SwapCached:")) {
2991 snprintf(lbuf
, 100, "SwapCached: %8lu kB\n", 0UL);
2996 l
= snprintf(cache
, cache_size
, "%s", printme
);
2998 perror("Error writing to cache");
3003 if (l
>= cache_size
) {
3004 fprintf(stderr
, "Internal error: truncated write to cache\n");
3015 d
->size
= total_len
;
3016 if (total_len
> size
) total_len
= size
;
3017 memcpy(buf
, d
->buf
, total_len
);
3026 free(memswlimit_str
);
3027 free(memswusage_str
);
3029 free(memswlimit_default_str
);
3030 free(memswusage_default_str
);
3035 * Read the cpuset.cpus for cg
3036 * Return the answer in a newly allocated string which must be freed
3038 static char *get_cpuset(const char *cg
)
3042 if (!cgfs_get_value("cpuset", cg
, "cpuset.cpus", &answer
))
3047 bool cpu_in_cpuset(int cpu
, const char *cpuset
);
3049 static bool cpuline_in_cpuset(const char *line
, const char *cpuset
)
3053 if (sscanf(line
, "processor : %d", &cpu
) != 1)
3055 return cpu_in_cpuset(cpu
, cpuset
);
3059 * check whether this is a '^processor" line in /proc/cpuinfo
3061 static bool is_processor_line(const char *line
)
3065 if (sscanf(line
, "processor : %d", &cpu
) == 1)
3070 static int proc_cpuinfo_read(char *buf
, size_t size
, off_t offset
,
3071 struct fuse_file_info
*fi
)
3073 struct fuse_context
*fc
= fuse_get_context();
3074 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3076 char *cpuset
= NULL
;
3078 size_t linelen
= 0, total_len
= 0, rv
= 0;
3079 bool am_printing
= false;
3081 char *cache
= d
->buf
;
3082 size_t cache_size
= d
->buflen
;
3086 if (offset
> d
->size
)
3090 int left
= d
->size
- offset
;
3091 total_len
= left
> size
? size
: left
;
3092 memcpy(buf
, cache
+ offset
, total_len
);
3096 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3099 cg
= get_pid_cgroup(initpid
, "cpuset");
3101 return read_file("proc/cpuinfo", buf
, size
, d
);
3103 cpuset
= get_cpuset(cg
);
3107 f
= fopen("/proc/cpuinfo", "r");
3111 while (getline(&line
, &linelen
, f
) != -1) {
3113 if (is_processor_line(line
)) {
3114 am_printing
= cpuline_in_cpuset(line
, cpuset
);
3117 l
= snprintf(cache
, cache_size
, "processor : %d\n", curcpu
);
3119 perror("Error writing to cache");
3123 if (l
>= cache_size
) {
3124 fprintf(stderr
, "Internal error: truncated write to cache\n");
3135 l
= snprintf(cache
, cache_size
, "%s", line
);
3137 perror("Error writing to cache");
3141 if (l
>= cache_size
) {
3142 fprintf(stderr
, "Internal error: truncated write to cache\n");
3153 d
->size
= total_len
;
3154 if (total_len
> size
) total_len
= size
;
3156 /* read from off 0 */
3157 memcpy(buf
, d
->buf
, total_len
);
3168 static int proc_stat_read(char *buf
, size_t size
, off_t offset
,
3169 struct fuse_file_info
*fi
)
3171 struct fuse_context
*fc
= fuse_get_context();
3172 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3174 char *cpuset
= NULL
;
3176 size_t linelen
= 0, total_len
= 0, rv
= 0;
3177 int curcpu
= -1; /* cpu numbering starts at 0 */
3178 unsigned long user
= 0, nice
= 0, system
= 0, idle
= 0, iowait
= 0, irq
= 0, softirq
= 0, steal
= 0, guest
= 0;
3179 unsigned long user_sum
= 0, nice_sum
= 0, system_sum
= 0, idle_sum
= 0, iowait_sum
= 0,
3180 irq_sum
= 0, softirq_sum
= 0, steal_sum
= 0, guest_sum
= 0;
3181 #define CPUALL_MAX_SIZE BUF_RESERVE_SIZE
3182 char cpuall
[CPUALL_MAX_SIZE
];
3183 /* reserve for cpu all */
3184 char *cache
= d
->buf
+ CPUALL_MAX_SIZE
;
3185 size_t cache_size
= d
->buflen
- CPUALL_MAX_SIZE
;
3189 if (offset
> d
->size
)
3193 int left
= d
->size
- offset
;
3194 total_len
= left
> size
? size
: left
;
3195 memcpy(buf
, d
->buf
+ offset
, total_len
);
3199 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3202 cg
= get_pid_cgroup(initpid
, "cpuset");
3204 return read_file("/proc/stat", buf
, size
, d
);
3206 cpuset
= get_cpuset(cg
);
3210 f
= fopen("/proc/stat", "r");
3215 if (getline(&line
, &linelen
, f
) < 0) {
3216 fprintf(stderr
, "proc_stat_read read first line failed\n");
3220 while (getline(&line
, &linelen
, f
) != -1) {
3223 char cpu_char
[10]; /* That's a lot of cores */
3226 if (sscanf(line
, "cpu%9[^ ]", cpu_char
) != 1) {
3227 /* not a ^cpuN line containing a number N, just print it */
3228 l
= snprintf(cache
, cache_size
, "%s", line
);
3230 perror("Error writing to cache");
3234 if (l
>= cache_size
) {
3235 fprintf(stderr
, "Internal error: truncated write to cache\n");
3245 if (sscanf(cpu_char
, "%d", &cpu
) != 1)
3247 if (!cpu_in_cpuset(cpu
, cpuset
))
3251 c
= strchr(line
, ' ');
3254 l
= snprintf(cache
, cache_size
, "cpu%d%s", curcpu
, c
);
3256 perror("Error writing to cache");
3261 if (l
>= cache_size
) {
3262 fprintf(stderr
, "Internal error: truncated write to cache\n");
3271 if (sscanf(line
, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user
, &nice
, &system
, &idle
, &iowait
, &irq
,
3272 &softirq
, &steal
, &guest
) != 9)
3276 system_sum
+= system
;
3278 iowait_sum
+= iowait
;
3280 softirq_sum
+= softirq
;
3287 int cpuall_len
= snprintf(cpuall
, CPUALL_MAX_SIZE
, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
3288 "cpu ", user_sum
, nice_sum
, system_sum
, idle_sum
, iowait_sum
, irq_sum
, softirq_sum
, steal_sum
, guest_sum
);
3289 if (cpuall_len
> 0 && cpuall_len
< CPUALL_MAX_SIZE
){
3290 memcpy(cache
, cpuall
, cpuall_len
);
3291 cache
+= cpuall_len
;
3293 /* shouldn't happen */
3294 fprintf(stderr
, "proc_stat_read copy cpuall failed, cpuall_len=%d\n", cpuall_len
);
3298 memmove(cache
, d
->buf
+ CPUALL_MAX_SIZE
, total_len
);
3299 total_len
+= cpuall_len
;
3301 d
->size
= total_len
;
3302 if (total_len
> size
) total_len
= size
;
3304 memcpy(buf
, d
->buf
, total_len
);
3316 static long int getreaperage(pid_t pid
)
3323 qpid
= lookup_initpid_in_store(pid
);
3327 ret
= snprintf(fnam
, 100, "/proc/%d", qpid
);
3328 if (ret
< 0 || ret
>= 100)
3331 if (lstat(fnam
, &sb
) < 0)
3334 return time(NULL
) - sb
.st_ctime
;
3337 static unsigned long get_reaper_busy(pid_t task
)
3339 pid_t initpid
= lookup_initpid_in_store(task
);
3340 char *cgroup
= NULL
, *usage_str
= NULL
;
3341 unsigned long usage
= 0;
3346 cgroup
= get_pid_cgroup(initpid
, "cpuacct");
3349 if (!cgfs_get_value("cpuacct", cgroup
, "cpuacct.usage", &usage_str
))
3351 usage
= strtoul(usage_str
, NULL
, 10);
3352 usage
/= 1000000000;
3363 char *name
, *cwd
= get_current_dir_name();
3369 len
= strlen(cwd
) + strlen("/iwashere") + 1;
3371 snprintf(name
, len
, "%s/iwashere", cwd
);
3373 fd
= creat(name
, 0755);
3380 * We read /proc/uptime and reuse its second field.
3381 * For the first field, we use the mtime for the reaper for
3382 * the calling pid as returned by getreaperage
3384 static int proc_uptime_read(char *buf
, size_t size
, off_t offset
,
3385 struct fuse_file_info
*fi
)
3387 struct fuse_context
*fc
= fuse_get_context();
3388 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3389 long int reaperage
= getreaperage(fc
->pid
);
3390 unsigned long int busytime
= get_reaper_busy(fc
->pid
), idletime
;
3391 char *cache
= d
->buf
;
3392 size_t total_len
= 0;
3399 if (offset
> d
->size
)
3403 int left
= d
->size
- offset
;
3404 total_len
= left
> size
? size
: left
;
3405 memcpy(buf
, cache
+ offset
, total_len
);
3409 idletime
= reaperage
- busytime
;
3410 if (idletime
> reaperage
)
3411 idletime
= reaperage
;
3413 total_len
= snprintf(d
->buf
, d
->size
, "%ld.0 %lu.0\n", reaperage
, idletime
);
3415 perror("Error writing to cache");
3419 d
->size
= (int)total_len
;
3422 if (total_len
> size
) total_len
= size
;
3424 memcpy(buf
, d
->buf
, total_len
);
3428 static int proc_diskstats_read(char *buf
, size_t size
, off_t offset
,
3429 struct fuse_file_info
*fi
)
3432 struct fuse_context
*fc
= fuse_get_context();
3433 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3435 char *io_serviced_str
= NULL
, *io_merged_str
= NULL
, *io_service_bytes_str
= NULL
,
3436 *io_wait_time_str
= NULL
, *io_service_time_str
= NULL
;
3437 unsigned long read
= 0, write
= 0;
3438 unsigned long read_merged
= 0, write_merged
= 0;
3439 unsigned long read_sectors
= 0, write_sectors
= 0;
3440 unsigned long read_ticks
= 0, write_ticks
= 0;
3441 unsigned long ios_pgr
= 0, tot_ticks
= 0, rq_ticks
= 0;
3442 unsigned long rd_svctm
= 0, wr_svctm
= 0, rd_wait
= 0, wr_wait
= 0;
3443 char *cache
= d
->buf
;
3444 size_t cache_size
= d
->buflen
;
3446 size_t linelen
= 0, total_len
= 0, rv
= 0;
3447 unsigned int major
= 0, minor
= 0;
3452 if (offset
> d
->size
)
3456 int left
= d
->size
- offset
;
3457 total_len
= left
> size
? size
: left
;
3458 memcpy(buf
, cache
+ offset
, total_len
);
3462 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3465 cg
= get_pid_cgroup(initpid
, "blkio");
3467 return read_file("/proc/diskstats", buf
, size
, d
);
3469 if (!cgfs_get_value("blkio", cg
, "blkio.io_serviced", &io_serviced_str
))
3471 if (!cgfs_get_value("blkio", cg
, "blkio.io_merged", &io_merged_str
))
3473 if (!cgfs_get_value("blkio", cg
, "blkio.io_service_bytes", &io_service_bytes_str
))
3475 if (!cgfs_get_value("blkio", cg
, "blkio.io_wait_time", &io_wait_time_str
))
3477 if (!cgfs_get_value("blkio", cg
, "blkio.io_service_time", &io_service_time_str
))
3481 f
= fopen("/proc/diskstats", "r");
3485 while (getline(&line
, &linelen
, f
) != -1) {
3487 char *printme
, lbuf
[256];
3489 i
= sscanf(line
, "%u %u %71s", &major
, &minor
, dev_name
);
3491 get_blkio_io_value(io_serviced_str
, major
, minor
, "Read", &read
);
3492 get_blkio_io_value(io_serviced_str
, major
, minor
, "Write", &write
);
3493 get_blkio_io_value(io_merged_str
, major
, minor
, "Read", &read_merged
);
3494 get_blkio_io_value(io_merged_str
, major
, minor
, "Write", &write_merged
);
3495 get_blkio_io_value(io_service_bytes_str
, major
, minor
, "Read", &read_sectors
);
3496 read_sectors
= read_sectors
/512;
3497 get_blkio_io_value(io_service_bytes_str
, major
, minor
, "Write", &write_sectors
);
3498 write_sectors
= write_sectors
/512;
3500 get_blkio_io_value(io_service_time_str
, major
, minor
, "Read", &rd_svctm
);
3501 rd_svctm
= rd_svctm
/1000000;
3502 get_blkio_io_value(io_wait_time_str
, major
, minor
, "Read", &rd_wait
);
3503 rd_wait
= rd_wait
/1000000;
3504 read_ticks
= rd_svctm
+ rd_wait
;
3506 get_blkio_io_value(io_service_time_str
, major
, minor
, "Write", &wr_svctm
);
3507 wr_svctm
= wr_svctm
/1000000;
3508 get_blkio_io_value(io_wait_time_str
, major
, minor
, "Write", &wr_wait
);
3509 wr_wait
= wr_wait
/1000000;
3510 write_ticks
= wr_svctm
+ wr_wait
;
3512 get_blkio_io_value(io_service_time_str
, major
, minor
, "Total", &tot_ticks
);
3513 tot_ticks
= tot_ticks
/1000000;
3518 memset(lbuf
, 0, 256);
3519 if (read
|| write
|| read_merged
|| write_merged
|| read_sectors
|| write_sectors
|| read_ticks
|| write_ticks
) {
3520 snprintf(lbuf
, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
3521 major
, minor
, dev_name
, read
, read_merged
, read_sectors
, read_ticks
,
3522 write
, write_merged
, write_sectors
, write_ticks
, ios_pgr
, tot_ticks
, rq_ticks
);
3527 l
= snprintf(cache
, cache_size
, "%s", printme
);
3529 perror("Error writing to fuse buf");
3533 if (l
>= cache_size
) {
3534 fprintf(stderr
, "Internal error: truncated write to cache\n");
3544 d
->size
= total_len
;
3545 if (total_len
> size
) total_len
= size
;
3546 memcpy(buf
, d
->buf
, total_len
);
3554 free(io_serviced_str
);
3555 free(io_merged_str
);
3556 free(io_service_bytes_str
);
3557 free(io_wait_time_str
);
3558 free(io_service_time_str
);
3562 static off_t
get_procfile_size(const char *which
)
3564 FILE *f
= fopen(which
, "r");
3567 ssize_t sz
, answer
= 0;
3571 while ((sz
= getline(&line
, &len
, f
)) != -1)
3579 int proc_getattr(const char *path
, struct stat
*sb
)
3581 struct timespec now
;
3583 memset(sb
, 0, sizeof(struct stat
));
3584 if (clock_gettime(CLOCK_REALTIME
, &now
) < 0)
3586 sb
->st_uid
= sb
->st_gid
= 0;
3587 sb
->st_atim
= sb
->st_mtim
= sb
->st_ctim
= now
;
3588 if (strcmp(path
, "/proc") == 0) {
3589 sb
->st_mode
= S_IFDIR
| 00555;
3593 if (strcmp(path
, "/proc/meminfo") == 0 ||
3594 strcmp(path
, "/proc/cpuinfo") == 0 ||
3595 strcmp(path
, "/proc/uptime") == 0 ||
3596 strcmp(path
, "/proc/stat") == 0 ||
3597 strcmp(path
, "/proc/diskstats") == 0) {
3599 sb
->st_mode
= S_IFREG
| 00444;
3607 int proc_readdir(const char *path
, void *buf
, fuse_fill_dir_t filler
, off_t offset
,
3608 struct fuse_file_info
*fi
)
3610 if (filler(buf
, "cpuinfo", NULL
, 0) != 0 ||
3611 filler(buf
, "meminfo", NULL
, 0) != 0 ||
3612 filler(buf
, "stat", NULL
, 0) != 0 ||
3613 filler(buf
, "uptime", NULL
, 0) != 0 ||
3614 filler(buf
, "diskstats", NULL
, 0) != 0)
3619 int proc_open(const char *path
, struct fuse_file_info
*fi
)
3622 struct file_info
*info
;
3624 if (strcmp(path
, "/proc/meminfo") == 0)
3625 type
= LXC_TYPE_PROC_MEMINFO
;
3626 else if (strcmp(path
, "/proc/cpuinfo") == 0)
3627 type
= LXC_TYPE_PROC_CPUINFO
;
3628 else if (strcmp(path
, "/proc/uptime") == 0)
3629 type
= LXC_TYPE_PROC_UPTIME
;
3630 else if (strcmp(path
, "/proc/stat") == 0)
3631 type
= LXC_TYPE_PROC_STAT
;
3632 else if (strcmp(path
, "/proc/diskstats") == 0)
3633 type
= LXC_TYPE_PROC_DISKSTATS
;
3637 info
= malloc(sizeof(*info
));
3641 memset(info
, 0, sizeof(*info
));
3644 info
->buflen
= get_procfile_size(path
) + BUF_RESERVE_SIZE
;
3646 info
->buf
= malloc(info
->buflen
);
3647 } while (!info
->buf
);
3648 memset(info
->buf
, 0, info
->buflen
);
3649 /* set actual size to buffer size */
3650 info
->size
= info
->buflen
;
3652 fi
->fh
= (unsigned long)info
;
3656 int proc_release(const char *path
, struct fuse_file_info
*fi
)
3658 struct file_info
*f
= (struct file_info
*)fi
->fh
;
3660 do_release_file_info(f
);
3664 int proc_read(const char *path
, char *buf
, size_t size
, off_t offset
,
3665 struct fuse_file_info
*fi
)
3667 struct file_info
*f
= (struct file_info
*) fi
->fh
;
3670 case LXC_TYPE_PROC_MEMINFO
:
3671 return proc_meminfo_read(buf
, size
, offset
, fi
);
3672 case LXC_TYPE_PROC_CPUINFO
:
3673 return proc_cpuinfo_read(buf
, size
, offset
, fi
);
3674 case LXC_TYPE_PROC_UPTIME
:
3675 return proc_uptime_read(buf
, size
, offset
, fi
);
3676 case LXC_TYPE_PROC_STAT
:
3677 return proc_stat_read(buf
, size
, offset
, fi
);
3678 case LXC_TYPE_PROC_DISKSTATS
:
3679 return proc_diskstats_read(buf
, size
, offset
, fi
);
3685 static void __attribute__((constructor
)) collect_subsystems(void)
3691 if ((f
= fopen("/proc/self/cgroup", "r")) == NULL
) {
3692 fprintf(stderr
, "Error opening /proc/self/cgroup: %s\n", strerror(errno
));
3695 while (getline(&line
, &len
, f
) != -1) {
3698 p
= strchr(line
, ':');
3703 p2
= strrchr(p
, ':');
3708 if (!store_hierarchy(line
, p
))
3719 static void __attribute__((destructor
)) free_subsystems(void)
3723 for (i
= 0; i
< num_hierarchies
; i
++)
3725 free(hierarchies
[i
]);