]>
Commit | Line | Data |
---|---|---|
237e200e SH |
1 | /* lxcfs |
2 | * | |
3 | * Copyright © 2014-2016 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
5 | * | |
6 | * See COPYING file for details. | |
7 | */ | |
8 | ||
9 | #define FUSE_USE_VERSION 26 | |
10 | ||
237e200e | 11 | #include <dirent.h> |
29a73c2f | 12 | #include <errno.h> |
237e200e SH |
13 | #include <fcntl.h> |
14 | #include <fuse.h> | |
237e200e | 15 | #include <libgen.h> |
237e200e | 16 | #include <pthread.h> |
29a73c2f CB |
17 | #include <sched.h> |
18 | #include <stdbool.h> | |
19 | #include <stdio.h> | |
20 | #include <stdlib.h> | |
21 | #include <string.h> | |
22 | #include <time.h> | |
23 | #include <unistd.h> | |
24 | #include <wait.h> | |
d89504c4 | 25 | #include <linux/magic.h> |
237e200e | 26 | #include <linux/sched.h> |
29a73c2f CB |
27 | #include <sys/epoll.h> |
28 | #include <sys/mman.h> | |
29 | #include <sys/mount.h> | |
237e200e SH |
30 | #include <sys/param.h> |
31 | #include <sys/socket.h> | |
29a73c2f | 32 | #include <sys/syscall.h> |
d89504c4 | 33 | #include <sys/vfs.h> |
237e200e | 34 | |
237e200e | 35 | #include "bindings.h" |
237e200e SH |
36 | #include "config.h" // for VERSION |
37 | ||
29a73c2f CB |
38 | /* Define pivot_root() if missing from the C library */ |
39 | #ifndef HAVE_PIVOT_ROOT | |
40 | static int pivot_root(const char * new_root, const char * put_old) | |
41 | { | |
42 | #ifdef __NR_pivot_root | |
43 | return syscall(__NR_pivot_root, new_root, put_old); | |
44 | #else | |
45 | errno = ENOSYS; | |
46 | return -1; | |
47 | #endif | |
48 | } | |
49 | #else | |
50 | extern int pivot_root(const char * new_root, const char * put_old); | |
51 | #endif | |
52 | ||
237e200e SH |
53 | enum { |
54 | LXC_TYPE_CGDIR, | |
55 | LXC_TYPE_CGFILE, | |
56 | LXC_TYPE_PROC_MEMINFO, | |
57 | LXC_TYPE_PROC_CPUINFO, | |
58 | LXC_TYPE_PROC_UPTIME, | |
59 | LXC_TYPE_PROC_STAT, | |
60 | LXC_TYPE_PROC_DISKSTATS, | |
70dcc12e | 61 | LXC_TYPE_PROC_SWAPS, |
237e200e SH |
62 | }; |
63 | ||
64 | struct file_info { | |
65 | char *controller; | |
66 | char *cgroup; | |
67 | char *file; | |
68 | int type; | |
69 | char *buf; // unused as of yet | |
70 | int buflen; | |
71 | int size; //actual data size | |
72 | int cached; | |
73 | }; | |
74 | ||
f34de69a CB |
75 | /* Reserve buffer size to account for file size changes. */ |
76 | #define BUF_RESERVE_SIZE 512 | |
237e200e SH |
77 | |
78 | /* | |
79 | * A table caching which pid is init for a pid namespace. | |
80 | * When looking up which pid is init for $qpid, we first | |
81 | * 1. Stat /proc/$qpid/ns/pid. | |
82 | * 2. Check whether the ino_t is in our store. | |
83 | * a. if not, fork a child in qpid's ns to send us | |
84 | * ucred.pid = 1, and read the initpid. Cache | |
85 | * initpid and creation time for /proc/initpid | |
86 | * in a new store entry. | |
87 | * b. if so, verify that /proc/initpid still matches | |
88 | * what we have saved. If not, clear the store | |
89 | * entry and go back to a. If so, return the | |
90 | * cached initpid. | |
91 | */ | |
92 | struct pidns_init_store { | |
93 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
94 | pid_t initpid; // the pid of nit in that ns | |
95 | long int ctime; // the time at which /proc/$initpid was created | |
96 | struct pidns_init_store *next; | |
97 | long int lastcheck; | |
98 | }; | |
99 | ||
100 | /* lol - look at how they are allocated in the kernel */ | |
101 | #define PIDNS_HASH_SIZE 4096 | |
102 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
103 | ||
104 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
105 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
106 | static void lock_mutex(pthread_mutex_t *l) | |
107 | { | |
108 | int ret; | |
109 | ||
110 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
b8defc3d | 111 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
112 | exit(1); |
113 | } | |
114 | } | |
115 | ||
29a73c2f CB |
116 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. |
117 | * Number of hierarchies mounted. */ | |
118 | static int num_hierarchies; | |
119 | ||
120 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
121 | * Hierachies mounted {cpuset, blkio, ...}: | |
122 | * Initialized via __constructor__ collect_and_mount_subsystems(). */ | |
123 | static char **hierarchies; | |
124 | ||
125 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
126 | * Open file descriptors: | |
127 | * @fd_hierarchies[i] refers to cgroup @hierarchies[i]. They are mounted in a | |
128 | * private mount namespace. | |
129 | * Initialized via __constructor__ collect_and_mount_subsystems(). | |
130 | * @fd_hierarchies[i] can be used to perform file operations on the cgroup | |
131 | * mounts and respective files in the private namespace even when located in | |
132 | * another namespace using the *at() family of functions | |
133 | * {openat(), fchownat(), ...}. */ | |
134 | static int *fd_hierarchies; | |
135 | ||
237e200e SH |
136 | static void unlock_mutex(pthread_mutex_t *l) |
137 | { | |
138 | int ret; | |
139 | ||
140 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
b8defc3d | 141 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
142 | exit(1); |
143 | } | |
144 | } | |
145 | ||
146 | static void store_lock(void) | |
147 | { | |
148 | lock_mutex(&pidns_store_mutex); | |
149 | } | |
150 | ||
151 | static void store_unlock(void) | |
152 | { | |
153 | unlock_mutex(&pidns_store_mutex); | |
154 | } | |
155 | ||
156 | /* Must be called under store_lock */ | |
157 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
158 | { | |
159 | struct stat initsb; | |
160 | char fnam[100]; | |
161 | ||
162 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
163 | if (stat(fnam, &initsb) < 0) | |
164 | return false; | |
7dd6560a CB |
165 | |
166 | lxcfs_debug("Comparing ctime %ld == %ld for pid %d.\n", e->ctime, | |
167 | initsb.st_ctime, e->initpid); | |
168 | ||
237e200e SH |
169 | if (e->ctime != initsb.st_ctime) |
170 | return false; | |
171 | return true; | |
172 | } | |
173 | ||
174 | /* Must be called under store_lock */ | |
175 | static void remove_initpid(struct pidns_init_store *e) | |
176 | { | |
177 | struct pidns_init_store *tmp; | |
178 | int h; | |
179 | ||
7dd6560a CB |
180 | lxcfs_debug("Remove_initpid: removing entry for %d.\n", e->initpid); |
181 | ||
237e200e SH |
182 | h = HASH(e->ino); |
183 | if (pidns_hash_table[h] == e) { | |
184 | pidns_hash_table[h] = e->next; | |
185 | free(e); | |
186 | return; | |
187 | } | |
188 | ||
189 | tmp = pidns_hash_table[h]; | |
190 | while (tmp) { | |
191 | if (tmp->next == e) { | |
192 | tmp->next = e->next; | |
193 | free(e); | |
194 | return; | |
195 | } | |
196 | tmp = tmp->next; | |
197 | } | |
198 | } | |
199 | ||
200 | #define PURGE_SECS 5 | |
201 | /* Must be called under store_lock */ | |
202 | static void prune_initpid_store(void) | |
203 | { | |
204 | static long int last_prune = 0; | |
205 | struct pidns_init_store *e, *prev, *delme; | |
206 | long int now, threshold; | |
207 | int i; | |
208 | ||
209 | if (!last_prune) { | |
210 | last_prune = time(NULL); | |
211 | return; | |
212 | } | |
213 | now = time(NULL); | |
214 | if (now < last_prune + PURGE_SECS) | |
215 | return; | |
7dd6560a CB |
216 | |
217 | lxcfs_debug("%s\n", "Pruning."); | |
218 | ||
237e200e SH |
219 | last_prune = now; |
220 | threshold = now - 2 * PURGE_SECS; | |
221 | ||
222 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
223 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
224 | if (e->lastcheck < threshold) { | |
7dd6560a CB |
225 | |
226 | lxcfs_debug("Removing cached entry for %d.\n", e->initpid); | |
227 | ||
237e200e SH |
228 | delme = e; |
229 | if (prev) | |
230 | prev->next = e->next; | |
231 | else | |
232 | pidns_hash_table[i] = e->next; | |
233 | e = e->next; | |
234 | free(delme); | |
235 | } else { | |
236 | prev = e; | |
237 | e = e->next; | |
238 | } | |
239 | } | |
240 | } | |
241 | } | |
242 | ||
243 | /* Must be called under store_lock */ | |
244 | static void save_initpid(struct stat *sb, pid_t pid) | |
245 | { | |
246 | struct pidns_init_store *e; | |
247 | char fpath[100]; | |
248 | struct stat procsb; | |
249 | int h; | |
250 | ||
7dd6560a CB |
251 | lxcfs_debug("Save_initpid: adding entry for %d.\n", pid); |
252 | ||
237e200e SH |
253 | snprintf(fpath, 100, "/proc/%d", pid); |
254 | if (stat(fpath, &procsb) < 0) | |
255 | return; | |
256 | do { | |
257 | e = malloc(sizeof(*e)); | |
258 | } while (!e); | |
259 | e->ino = sb->st_ino; | |
260 | e->initpid = pid; | |
261 | e->ctime = procsb.st_ctime; | |
262 | h = HASH(e->ino); | |
263 | e->next = pidns_hash_table[h]; | |
264 | e->lastcheck = time(NULL); | |
265 | pidns_hash_table[h] = e; | |
266 | } | |
267 | ||
268 | /* | |
269 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
270 | * entry for the inode number and creation time. Verify that the init pid | |
271 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
272 | * otherwise. | |
273 | * Must be called under store_lock | |
274 | */ | |
275 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
276 | { | |
277 | int h = HASH(sb->st_ino); | |
278 | struct pidns_init_store *e = pidns_hash_table[h]; | |
279 | ||
280 | while (e) { | |
281 | if (e->ino == sb->st_ino) { | |
282 | if (initpid_still_valid(e, sb)) { | |
283 | e->lastcheck = time(NULL); | |
284 | return e; | |
285 | } | |
286 | remove_initpid(e); | |
287 | return NULL; | |
288 | } | |
289 | e = e->next; | |
290 | } | |
291 | ||
292 | return NULL; | |
293 | } | |
294 | ||
0f657ce3 | 295 | static int is_dir(const char *path, int fd) |
237e200e SH |
296 | { |
297 | struct stat statbuf; | |
0f657ce3 | 298 | int ret = fstatat(fd, path, &statbuf, fd); |
237e200e SH |
299 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) |
300 | return 1; | |
301 | return 0; | |
302 | } | |
303 | ||
304 | static char *must_copy_string(const char *str) | |
305 | { | |
306 | char *dup = NULL; | |
307 | if (!str) | |
308 | return NULL; | |
309 | do { | |
310 | dup = strdup(str); | |
311 | } while (!dup); | |
312 | ||
313 | return dup; | |
314 | } | |
315 | ||
316 | static inline void drop_trailing_newlines(char *s) | |
317 | { | |
318 | int l; | |
319 | ||
320 | for (l=strlen(s); l>0 && s[l-1] == '\n'; l--) | |
321 | s[l-1] = '\0'; | |
322 | } | |
323 | ||
324 | #define BATCH_SIZE 50 | |
325 | static void dorealloc(char **mem, size_t oldlen, size_t newlen) | |
326 | { | |
327 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
328 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
329 | ||
330 | if (!*mem || newbatches > oldbatches) { | |
331 | char *tmp; | |
332 | do { | |
333 | tmp = realloc(*mem, newbatches * BATCH_SIZE); | |
334 | } while (!tmp); | |
335 | *mem = tmp; | |
336 | } | |
337 | } | |
338 | static void append_line(char **contents, size_t *len, char *line, ssize_t linelen) | |
339 | { | |
340 | size_t newlen = *len + linelen; | |
341 | dorealloc(contents, *len, newlen + 1); | |
342 | memcpy(*contents + *len, line, linelen+1); | |
343 | *len = newlen; | |
344 | } | |
345 | ||
60f2ae53 | 346 | static char *slurp_file(const char *from, int fd) |
237e200e SH |
347 | { |
348 | char *line = NULL; | |
349 | char *contents = NULL; | |
60f2ae53 | 350 | FILE *f = fdopen(fd, "r"); |
237e200e SH |
351 | size_t len = 0, fulllen = 0; |
352 | ssize_t linelen; | |
353 | ||
354 | if (!f) | |
355 | return NULL; | |
356 | ||
357 | while ((linelen = getline(&line, &len, f)) != -1) { | |
358 | append_line(&contents, &fulllen, line, linelen); | |
359 | } | |
360 | fclose(f); | |
361 | ||
362 | if (contents) | |
363 | drop_trailing_newlines(contents); | |
364 | free(line); | |
365 | return contents; | |
366 | } | |
367 | ||
ba59ea09 | 368 | static bool write_string(const char *fnam, const char *string, int fd) |
237e200e SH |
369 | { |
370 | FILE *f; | |
371 | size_t len, ret; | |
372 | ||
ba59ea09 | 373 | if (!(f = fdopen(fd, "w"))) |
237e200e SH |
374 | return false; |
375 | len = strlen(string); | |
376 | ret = fwrite(string, 1, len, f); | |
377 | if (ret != len) { | |
b8defc3d | 378 | lxcfs_error("Error writing to file: %s\n", strerror(errno)); |
237e200e SH |
379 | fclose(f); |
380 | return false; | |
381 | } | |
382 | if (fclose(f) < 0) { | |
b8defc3d | 383 | lxcfs_error("Error writing to file: %s\n", strerror(errno)); |
237e200e SH |
384 | return false; |
385 | } | |
386 | return true; | |
387 | } | |
388 | ||
237e200e SH |
389 | struct cgfs_files { |
390 | char *name; | |
391 | uint32_t uid, gid; | |
392 | uint32_t mode; | |
393 | }; | |
394 | ||
0619767c | 395 | #define ALLOC_NUM 20 |
237e200e SH |
396 | static bool store_hierarchy(char *stridx, char *h) |
397 | { | |
0619767c SH |
398 | if (num_hierarchies % ALLOC_NUM == 0) { |
399 | size_t n = (num_hierarchies / ALLOC_NUM) + 1; | |
400 | n *= ALLOC_NUM; | |
401 | char **tmp = realloc(hierarchies, n * sizeof(char *)); | |
0619767c | 402 | if (!tmp) { |
b8defc3d | 403 | lxcfs_error("%s\n", strerror(errno)); |
0619767c SH |
404 | exit(1); |
405 | } | |
237e200e | 406 | hierarchies = tmp; |
237e200e | 407 | } |
f676eb79 | 408 | |
0619767c | 409 | hierarchies[num_hierarchies++] = must_copy_string(h); |
237e200e SH |
410 | return true; |
411 | } | |
412 | ||
413 | static void print_subsystems(void) | |
414 | { | |
415 | int i; | |
416 | ||
cc97d34c | 417 | fprintf(stderr, "hierarchies:\n"); |
237e200e SH |
418 | for (i = 0; i < num_hierarchies; i++) { |
419 | if (hierarchies[i]) | |
b8defc3d CB |
420 | fprintf(stderr, " %2d: fd: %3d: %s\n", i, |
421 | fd_hierarchies[i], hierarchies[i]); | |
237e200e SH |
422 | } |
423 | } | |
424 | ||
425 | static bool in_comma_list(const char *needle, const char *haystack) | |
426 | { | |
427 | const char *s = haystack, *e; | |
428 | size_t nlen = strlen(needle); | |
429 | ||
06081b29 | 430 | while (*s && (e = strchr(s, ','))) { |
237e200e SH |
431 | if (nlen != e - s) { |
432 | s = e + 1; | |
433 | continue; | |
434 | } | |
435 | if (strncmp(needle, s, nlen) == 0) | |
436 | return true; | |
437 | s = e + 1; | |
438 | } | |
439 | if (strcmp(needle, s) == 0) | |
440 | return true; | |
441 | return false; | |
442 | } | |
443 | ||
444 | /* do we need to do any massaging here? I'm not sure... */ | |
5dd3e6fd CB |
445 | /* Return the mounted controller and store the corresponding open file descriptor |
446 | * referring to the controller mountpoint in the private lxcfs namespace in | |
447 | * @cfd. | |
448 | */ | |
449 | static char *find_mounted_controller(const char *controller, int *cfd) | |
237e200e SH |
450 | { |
451 | int i; | |
452 | ||
453 | for (i = 0; i < num_hierarchies; i++) { | |
454 | if (!hierarchies[i]) | |
455 | continue; | |
5dd3e6fd CB |
456 | if (strcmp(hierarchies[i], controller) == 0) { |
457 | *cfd = fd_hierarchies[i]; | |
237e200e | 458 | return hierarchies[i]; |
5dd3e6fd CB |
459 | } |
460 | if (in_comma_list(controller, hierarchies[i])) { | |
461 | *cfd = fd_hierarchies[i]; | |
237e200e | 462 | return hierarchies[i]; |
5dd3e6fd | 463 | } |
237e200e SH |
464 | } |
465 | ||
466 | return NULL; | |
467 | } | |
468 | ||
469 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, | |
470 | const char *value) | |
471 | { | |
ba59ea09 | 472 | int ret, fd, cfd; |
237e200e | 473 | size_t len; |
f5a6d92e | 474 | char *fnam, *tmpc; |
237e200e | 475 | |
f5a6d92e | 476 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
477 | if (!tmpc) |
478 | return false; | |
f5a6d92e CB |
479 | |
480 | /* Make sure we pass a relative path to *at() family of functions. | |
481 | * . + /cgroup + / + file + \0 | |
482 | */ | |
ba59ea09 | 483 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 484 | fnam = alloca(len); |
ba59ea09 CB |
485 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
486 | if (ret < 0 || (size_t)ret >= len) | |
487 | return false; | |
488 | ||
489 | fd = openat(cfd, fnam, O_WRONLY); | |
490 | if (fd < 0) | |
491 | return false; | |
f676eb79 | 492 | |
ba59ea09 | 493 | return write_string(fnam, value, fd); |
237e200e SH |
494 | } |
495 | ||
496 | // Chown all the files in the cgroup directory. We do this when we create | |
497 | // a cgroup on behalf of a user. | |
f23fe717 | 498 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e | 499 | { |
f23fe717 | 500 | struct dirent *direntp; |
237e200e SH |
501 | char path[MAXPATHLEN]; |
502 | size_t len; | |
503 | DIR *d; | |
f23fe717 | 504 | int fd1, ret; |
237e200e SH |
505 | |
506 | len = strlen(dirname); | |
507 | if (len >= MAXPATHLEN) { | |
b8defc3d | 508 | lxcfs_error("Pathname too long: %s\n", dirname); |
237e200e SH |
509 | return; |
510 | } | |
511 | ||
f23fe717 CB |
512 | fd1 = openat(fd, dirname, O_DIRECTORY); |
513 | if (fd1 < 0) | |
514 | return; | |
515 | ||
516 | d = fdopendir(fd1); | |
237e200e | 517 | if (!d) { |
b8defc3d | 518 | lxcfs_error("Failed to open %s\n", dirname); |
237e200e SH |
519 | return; |
520 | } | |
521 | ||
f23fe717 | 522 | while ((direntp = readdir(d))) { |
237e200e SH |
523 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) |
524 | continue; | |
525 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
526 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 527 | lxcfs_error("Pathname too long under %s\n", dirname); |
237e200e SH |
528 | continue; |
529 | } | |
f23fe717 | 530 | if (fchownat(fd, path, uid, gid, 0) < 0) |
b8defc3d | 531 | lxcfs_error("Failed to chown file %s to %u:%u", path, uid, gid); |
237e200e SH |
532 | } |
533 | closedir(d); | |
534 | } | |
535 | ||
536 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
537 | { | |
5dd3e6fd | 538 | int cfd; |
237e200e | 539 | size_t len; |
f5a6d92e | 540 | char *dirnam, *tmpc; |
237e200e | 541 | |
f5a6d92e | 542 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
543 | if (!tmpc) |
544 | return -EINVAL; | |
f5a6d92e CB |
545 | |
546 | /* Make sure we pass a relative path to *at() family of functions. | |
547 | * . + /cg + \0 | |
548 | */ | |
f23fe717 | 549 | len = strlen(cg) + 2; |
237e200e | 550 | dirnam = alloca(len); |
f23fe717 | 551 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
237e200e | 552 | |
f23fe717 | 553 | if (mkdirat(cfd, dirnam, 0755) < 0) |
237e200e SH |
554 | return -errno; |
555 | ||
556 | if (uid == 0 && gid == 0) | |
557 | return 0; | |
558 | ||
f23fe717 | 559 | if (fchownat(cfd, dirnam, uid, gid, 0) < 0) |
237e200e SH |
560 | return -errno; |
561 | ||
f23fe717 | 562 | chown_all_cgroup_files(dirnam, uid, gid, cfd); |
237e200e SH |
563 | |
564 | return 0; | |
565 | } | |
566 | ||
7213ec5c | 567 | static bool recursive_rmdir(const char *dirname, int fd, const int cfd) |
237e200e | 568 | { |
b7672ded | 569 | struct dirent *direntp; |
237e200e SH |
570 | DIR *dir; |
571 | bool ret = false; | |
572 | char pathname[MAXPATHLEN]; | |
b7672ded CB |
573 | int dupfd; |
574 | ||
575 | dupfd = dup(fd); // fdopendir() does bad things once it uses an fd. | |
576 | if (dupfd < 0) | |
577 | return false; | |
237e200e | 578 | |
b7672ded | 579 | dir = fdopendir(dupfd); |
237e200e | 580 | if (!dir) { |
7dd6560a | 581 | lxcfs_debug("Failed to open %s: %s.\n", dirname, strerror(errno)); |
7213ec5c | 582 | close(dupfd); |
237e200e SH |
583 | return false; |
584 | } | |
585 | ||
b7672ded | 586 | while ((direntp = readdir(dir))) { |
237e200e SH |
587 | struct stat mystat; |
588 | int rc; | |
589 | ||
237e200e SH |
590 | if (!strcmp(direntp->d_name, ".") || |
591 | !strcmp(direntp->d_name, "..")) | |
592 | continue; | |
593 | ||
594 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
595 | if (rc < 0 || rc >= MAXPATHLEN) { | |
b8defc3d | 596 | lxcfs_error("%s\n", "Pathname too long."); |
237e200e SH |
597 | continue; |
598 | } | |
599 | ||
2e81a5e3 CB |
600 | rc = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
601 | if (rc) { | |
7dd6560a | 602 | lxcfs_debug("Failed to stat %s: %s.\n", pathname, strerror(errno)); |
237e200e SH |
603 | continue; |
604 | } | |
7dd6560a | 605 | if (S_ISDIR(mystat.st_mode)) |
2e81a5e3 | 606 | if (!recursive_rmdir(pathname, fd, cfd)) |
7dd6560a | 607 | lxcfs_debug("Error removing %s.\n", pathname); |
237e200e SH |
608 | } |
609 | ||
610 | ret = true; | |
611 | if (closedir(dir) < 0) { | |
b8defc3d | 612 | lxcfs_error("Failed to close directory %s: %s\n", dirname, strerror(errno)); |
237e200e SH |
613 | ret = false; |
614 | } | |
615 | ||
2e81a5e3 | 616 | if (unlinkat(cfd, dirname, AT_REMOVEDIR) < 0) { |
7dd6560a | 617 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); |
237e200e SH |
618 | ret = false; |
619 | } | |
7213ec5c CB |
620 | |
621 | close(dupfd); | |
237e200e SH |
622 | |
623 | return ret; | |
624 | } | |
625 | ||
626 | bool cgfs_remove(const char *controller, const char *cg) | |
627 | { | |
b7672ded | 628 | int fd, cfd; |
237e200e | 629 | size_t len; |
f5a6d92e | 630 | char *dirnam, *tmpc; |
7213ec5c | 631 | bool bret; |
237e200e | 632 | |
f5a6d92e | 633 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
634 | if (!tmpc) |
635 | return false; | |
f5a6d92e CB |
636 | |
637 | /* Make sure we pass a relative path to *at() family of functions. | |
638 | * . + /cg + \0 | |
639 | */ | |
b7672ded | 640 | len = strlen(cg) + 2; |
237e200e | 641 | dirnam = alloca(len); |
b7672ded CB |
642 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
643 | ||
644 | fd = openat(cfd, dirnam, O_DIRECTORY); | |
645 | if (fd < 0) | |
646 | return false; | |
647 | ||
7213ec5c CB |
648 | bret = recursive_rmdir(dirnam, fd, cfd); |
649 | close(fd); | |
650 | return bret; | |
237e200e SH |
651 | } |
652 | ||
653 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
654 | { | |
5dd3e6fd | 655 | int cfd; |
237e200e | 656 | size_t len; |
f5a6d92e | 657 | char *pathname, *tmpc; |
237e200e | 658 | |
f5a6d92e | 659 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
660 | if (!tmpc) |
661 | return false; | |
f5a6d92e CB |
662 | |
663 | /* Make sure we pass a relative path to *at() family of functions. | |
664 | * . + /file + \0 | |
665 | */ | |
534690b4 | 666 | len = strlen(file) + 2; |
237e200e | 667 | pathname = alloca(len); |
534690b4 CB |
668 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
669 | if (fchmodat(cfd, pathname, mode, 0) < 0) | |
237e200e SH |
670 | return false; |
671 | return true; | |
672 | } | |
673 | ||
0f657ce3 | 674 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e SH |
675 | { |
676 | size_t len; | |
677 | char *fname; | |
678 | ||
679 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
680 | fname = alloca(len); | |
681 | snprintf(fname, len, "%s/tasks", dirname); | |
0f657ce3 | 682 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
683 | return -errno; |
684 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
0f657ce3 | 685 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
686 | return -errno; |
687 | return 0; | |
688 | } | |
689 | ||
690 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
691 | { | |
5dd3e6fd | 692 | int cfd; |
237e200e | 693 | size_t len; |
f5a6d92e | 694 | char *pathname, *tmpc; |
237e200e | 695 | |
f5a6d92e | 696 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
697 | if (!tmpc) |
698 | return -EINVAL; | |
f5a6d92e CB |
699 | |
700 | /* Make sure we pass a relative path to *at() family of functions. | |
701 | * . + /file + \0 | |
702 | */ | |
0f657ce3 | 703 | len = strlen(file) + 2; |
237e200e | 704 | pathname = alloca(len); |
0f657ce3 CB |
705 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
706 | if (fchownat(cfd, pathname, uid, gid, 0) < 0) | |
237e200e SH |
707 | return -errno; |
708 | ||
0f657ce3 | 709 | if (is_dir(pathname, cfd)) |
237e200e | 710 | // like cgmanager did, we want to chown the tasks file as well |
0f657ce3 | 711 | return chown_tasks_files(pathname, uid, gid, cfd); |
237e200e SH |
712 | |
713 | return 0; | |
714 | } | |
715 | ||
716 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
717 | { | |
3ffd08ee | 718 | int fd, cfd; |
237e200e | 719 | size_t len; |
f5a6d92e | 720 | char *pathname, *tmpc; |
237e200e | 721 | |
f5a6d92e | 722 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
723 | if (!tmpc) |
724 | return NULL; | |
f5a6d92e CB |
725 | |
726 | /* Make sure we pass a relative path to *at() family of functions. | |
727 | * . + /cgroup + / "cgroup.procs" + \0 | |
728 | */ | |
3ffd08ee | 729 | len = strlen(cgroup) + strlen("cgroup.procs") + 3; |
237e200e | 730 | pathname = alloca(len); |
3ffd08ee CB |
731 | snprintf(pathname, len, "%s%s/cgroup.procs", *cgroup == '/' ? "." : "", cgroup); |
732 | ||
733 | fd = openat(cfd, pathname, O_WRONLY); | |
734 | if (fd < 0) | |
735 | return NULL; | |
736 | ||
737 | return fdopen(fd, "w"); | |
237e200e SH |
738 | } |
739 | ||
f366da65 WB |
740 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, |
741 | void ***list, size_t typesize, | |
742 | void* (*iterator)(const char*, const char*, const char*)) | |
237e200e | 743 | { |
4ea38a4c | 744 | int cfd, fd, ret; |
237e200e | 745 | size_t len; |
4ea38a4c | 746 | char *cg, *tmpc; |
237e200e | 747 | char pathname[MAXPATHLEN]; |
f366da65 | 748 | size_t sz = 0, asz = 0; |
4ea38a4c | 749 | struct dirent *dirent; |
237e200e | 750 | DIR *dir; |
237e200e | 751 | |
4ea38a4c | 752 | tmpc = find_mounted_controller(controller, &cfd); |
f366da65 | 753 | *list = NULL; |
237e200e | 754 | if (!tmpc) |
e97c834b | 755 | return false; |
237e200e | 756 | |
f5a6d92e | 757 | /* Make sure we pass a relative path to *at() family of functions. */ |
4ea38a4c CB |
758 | len = strlen(cgroup) + 1 /* . */ + 1 /* \0 */; |
759 | cg = alloca(len); | |
760 | ret = snprintf(cg, len, "%s%s", *cgroup == '/' ? "." : "", cgroup); | |
761 | if (ret < 0 || (size_t)ret >= len) { | |
b8defc3d | 762 | lxcfs_error("Pathname too long under %s\n", cgroup); |
4ea38a4c CB |
763 | return false; |
764 | } | |
237e200e | 765 | |
4ea38a4c CB |
766 | fd = openat(cfd, cg, O_DIRECTORY); |
767 | if (fd < 0) | |
768 | return false; | |
769 | ||
770 | dir = fdopendir(fd); | |
237e200e SH |
771 | if (!dir) |
772 | return false; | |
773 | ||
4ea38a4c | 774 | while ((dirent = readdir(dir))) { |
237e200e | 775 | struct stat mystat; |
237e200e | 776 | |
4ea38a4c CB |
777 | if (!strcmp(dirent->d_name, ".") || |
778 | !strcmp(dirent->d_name, "..")) | |
237e200e SH |
779 | continue; |
780 | ||
4ea38a4c CB |
781 | ret = snprintf(pathname, MAXPATHLEN, "%s/%s", cg, dirent->d_name); |
782 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 783 | lxcfs_error("Pathname too long under %s\n", cg); |
237e200e SH |
784 | continue; |
785 | } | |
786 | ||
4ea38a4c | 787 | ret = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
237e200e | 788 | if (ret) { |
b8defc3d | 789 | lxcfs_error("Failed to stat %s: %s\n", pathname, strerror(errno)); |
237e200e SH |
790 | continue; |
791 | } | |
f366da65 WB |
792 | if ((!directories && !S_ISREG(mystat.st_mode)) || |
793 | (directories && !S_ISDIR(mystat.st_mode))) | |
237e200e SH |
794 | continue; |
795 | ||
796 | if (sz+2 >= asz) { | |
f366da65 | 797 | void **tmp; |
237e200e SH |
798 | asz += BATCH_SIZE; |
799 | do { | |
f366da65 | 800 | tmp = realloc(*list, asz * typesize); |
237e200e SH |
801 | } while (!tmp); |
802 | *list = tmp; | |
803 | } | |
4ea38a4c | 804 | (*list)[sz] = (*iterator)(controller, cg, dirent->d_name); |
237e200e SH |
805 | (*list)[sz+1] = NULL; |
806 | sz++; | |
807 | } | |
808 | if (closedir(dir) < 0) { | |
b8defc3d | 809 | lxcfs_error("Failed closedir for %s: %s\n", cgroup, strerror(errno)); |
237e200e SH |
810 | return false; |
811 | } | |
812 | return true; | |
813 | } | |
814 | ||
f366da65 WB |
815 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
816 | { | |
817 | char *dup; | |
818 | do { | |
819 | dup = strdup(dir_entry); | |
820 | } while (!dup); | |
821 | return dup; | |
822 | } | |
823 | ||
824 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
825 | { | |
826 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
827 | } | |
828 | ||
237e200e SH |
829 | void free_key(struct cgfs_files *k) |
830 | { | |
831 | if (!k) | |
832 | return; | |
833 | free(k->name); | |
834 | free(k); | |
835 | } | |
836 | ||
837 | void free_keys(struct cgfs_files **keys) | |
838 | { | |
839 | int i; | |
840 | ||
841 | if (!keys) | |
842 | return; | |
843 | for (i = 0; keys[i]; i++) { | |
844 | free_key(keys[i]); | |
845 | } | |
846 | free(keys); | |
847 | } | |
848 | ||
849 | bool cgfs_get_value(const char *controller, const char *cgroup, const char *file, char **value) | |
850 | { | |
60f2ae53 | 851 | int ret, fd, cfd; |
237e200e | 852 | size_t len; |
f5a6d92e | 853 | char *fnam, *tmpc; |
237e200e | 854 | |
f5a6d92e | 855 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
856 | if (!tmpc) |
857 | return false; | |
f5a6d92e CB |
858 | |
859 | /* Make sure we pass a relative path to *at() family of functions. | |
860 | * . + /cgroup + / + file + \0 | |
861 | */ | |
60f2ae53 | 862 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 863 | fnam = alloca(len); |
60f2ae53 CB |
864 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
865 | if (ret < 0 || (size_t)ret >= len) | |
234a820c | 866 | return false; |
60f2ae53 CB |
867 | |
868 | fd = openat(cfd, fnam, O_RDONLY); | |
869 | if (fd < 0) | |
234a820c | 870 | return false; |
237e200e | 871 | |
60f2ae53 | 872 | *value = slurp_file(fnam, fd); |
237e200e SH |
873 | return *value != NULL; |
874 | } | |
875 | ||
876 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) | |
877 | { | |
4ea38a4c | 878 | int ret, cfd; |
237e200e | 879 | size_t len; |
f5a6d92e | 880 | char *fnam, *tmpc; |
237e200e SH |
881 | struct stat sb; |
882 | struct cgfs_files *newkey; | |
237e200e | 883 | |
f5a6d92e | 884 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
885 | if (!tmpc) |
886 | return false; | |
887 | ||
888 | if (file && *file == '/') | |
889 | file++; | |
890 | ||
06081b29 | 891 | if (file && strchr(file, '/')) |
237e200e SH |
892 | return NULL; |
893 | ||
f5a6d92e CB |
894 | /* Make sure we pass a relative path to *at() family of functions. |
895 | * . + /cgroup + / + file + \0 | |
896 | */ | |
4ea38a4c | 897 | len = strlen(cgroup) + 3; |
237e200e SH |
898 | if (file) |
899 | len += strlen(file) + 1; | |
900 | fnam = alloca(len); | |
4ea38a4c CB |
901 | snprintf(fnam, len, "%s%s%s%s", *cgroup == '/' ? "." : "", cgroup, |
902 | file ? "/" : "", file ? file : ""); | |
237e200e | 903 | |
4ea38a4c | 904 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
905 | if (ret < 0) |
906 | return NULL; | |
907 | ||
908 | do { | |
909 | newkey = malloc(sizeof(struct cgfs_files)); | |
910 | } while (!newkey); | |
911 | if (file) | |
912 | newkey->name = must_copy_string(file); | |
06081b29 CB |
913 | else if (strrchr(cgroup, '/')) |
914 | newkey->name = must_copy_string(strrchr(cgroup, '/')); | |
237e200e SH |
915 | else |
916 | newkey->name = must_copy_string(cgroup); | |
917 | newkey->uid = sb.st_uid; | |
918 | newkey->gid = sb.st_gid; | |
919 | newkey->mode = sb.st_mode; | |
920 | ||
921 | return newkey; | |
922 | } | |
923 | ||
f366da65 | 924 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
237e200e | 925 | { |
f366da65 WB |
926 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); |
927 | if (!entry) { | |
b8defc3d CB |
928 | lxcfs_error("Error getting files under %s:%s\n", controller, |
929 | cgroup); | |
237e200e | 930 | } |
f366da65 WB |
931 | return entry; |
932 | } | |
933 | ||
934 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
935 | { | |
936 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
237e200e SH |
937 | } |
938 | ||
939 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
5dd3e6fd CB |
940 | { |
941 | int cfd; | |
942 | size_t len; | |
f5a6d92e | 943 | char *fnam, *tmpc; |
237e200e SH |
944 | int ret; |
945 | struct stat sb; | |
946 | ||
f5a6d92e | 947 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
948 | if (!tmpc) |
949 | return false; | |
f5a6d92e CB |
950 | |
951 | /* Make sure we pass a relative path to *at() family of functions. | |
952 | * . + /cgroup + / + f + \0 | |
953 | */ | |
d04232f2 | 954 | len = strlen(cgroup) + strlen(f) + 3; |
237e200e | 955 | fnam = alloca(len); |
d04232f2 CB |
956 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, f); |
957 | if (ret < 0 || (size_t)ret >= len) | |
958 | return false; | |
237e200e | 959 | |
d04232f2 | 960 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
961 | if (ret < 0 || !S_ISDIR(sb.st_mode)) |
962 | return false; | |
f5a6d92e | 963 | |
237e200e SH |
964 | return true; |
965 | } | |
966 | ||
967 | #define SEND_CREDS_OK 0 | |
968 | #define SEND_CREDS_NOTSK 1 | |
969 | #define SEND_CREDS_FAIL 2 | |
970 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
971 | static int wait_for_pid(pid_t pid); | |
972 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
b10bdd6c | 973 | static int send_creds_clone_wrapper(void *arg); |
237e200e SH |
974 | |
975 | /* | |
b10bdd6c | 976 | * clone a task which switches to @task's namespace and writes '1'. |
237e200e SH |
977 | * over a unix sock so we can read the task's reaper's pid in our |
978 | * namespace | |
b10bdd6c FG |
979 | * |
980 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
981 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
982 | * the pidns and the parent pid outside are identical. Using clone prevents | |
983 | * this issue. | |
237e200e SH |
984 | */ |
985 | static void write_task_init_pid_exit(int sock, pid_t target) | |
986 | { | |
237e200e SH |
987 | char fnam[100]; |
988 | pid_t pid; | |
237e200e | 989 | int fd, ret; |
b10bdd6c FG |
990 | size_t stack_size = sysconf(_SC_PAGESIZE); |
991 | void *stack = alloca(stack_size); | |
237e200e SH |
992 | |
993 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
994 | if (ret < 0 || ret >= sizeof(fnam)) | |
995 | _exit(1); | |
996 | ||
997 | fd = open(fnam, O_RDONLY); | |
998 | if (fd < 0) { | |
999 | perror("write_task_init_pid_exit open of ns/pid"); | |
1000 | _exit(1); | |
1001 | } | |
1002 | if (setns(fd, 0)) { | |
1003 | perror("write_task_init_pid_exit setns 1"); | |
1004 | close(fd); | |
1005 | _exit(1); | |
1006 | } | |
b10bdd6c | 1007 | pid = clone(send_creds_clone_wrapper, stack + stack_size, SIGCHLD, &sock); |
237e200e SH |
1008 | if (pid < 0) |
1009 | _exit(1); | |
1010 | if (pid != 0) { | |
1011 | if (!wait_for_pid(pid)) | |
1012 | _exit(1); | |
1013 | _exit(0); | |
1014 | } | |
b10bdd6c FG |
1015 | } |
1016 | ||
1017 | static int send_creds_clone_wrapper(void *arg) { | |
1018 | struct ucred cred; | |
1019 | char v; | |
1020 | int sock = *(int *)arg; | |
237e200e SH |
1021 | |
1022 | /* we are the child */ | |
1023 | cred.uid = 0; | |
1024 | cred.gid = 0; | |
1025 | cred.pid = 1; | |
1026 | v = '1'; | |
1027 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
b10bdd6c FG |
1028 | return 1; |
1029 | return 0; | |
237e200e SH |
1030 | } |
1031 | ||
1032 | static pid_t get_init_pid_for_task(pid_t task) | |
1033 | { | |
1034 | int sock[2]; | |
1035 | pid_t pid; | |
1036 | pid_t ret = -1; | |
1037 | char v = '0'; | |
1038 | struct ucred cred; | |
1039 | ||
1040 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1041 | perror("socketpair"); | |
1042 | return -1; | |
1043 | } | |
1044 | ||
1045 | pid = fork(); | |
1046 | if (pid < 0) | |
1047 | goto out; | |
1048 | if (!pid) { | |
1049 | close(sock[1]); | |
1050 | write_task_init_pid_exit(sock[0], task); | |
1051 | _exit(0); | |
1052 | } | |
1053 | ||
1054 | if (!recv_creds(sock[1], &cred, &v)) | |
1055 | goto out; | |
1056 | ret = cred.pid; | |
1057 | ||
1058 | out: | |
1059 | close(sock[0]); | |
1060 | close(sock[1]); | |
1061 | if (pid > 0) | |
1062 | wait_for_pid(pid); | |
1063 | return ret; | |
1064 | } | |
1065 | ||
1066 | static pid_t lookup_initpid_in_store(pid_t qpid) | |
1067 | { | |
1068 | pid_t answer = 0; | |
1069 | struct stat sb; | |
1070 | struct pidns_init_store *e; | |
1071 | char fnam[100]; | |
1072 | ||
1073 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
1074 | store_lock(); | |
1075 | if (stat(fnam, &sb) < 0) | |
1076 | goto out; | |
1077 | e = lookup_verify_initpid(&sb); | |
1078 | if (e) { | |
1079 | answer = e->initpid; | |
1080 | goto out; | |
1081 | } | |
1082 | answer = get_init_pid_for_task(qpid); | |
1083 | if (answer > 0) | |
1084 | save_initpid(&sb, answer); | |
1085 | ||
1086 | out: | |
1087 | /* we prune at end in case we are returning | |
1088 | * the value we were about to return */ | |
1089 | prune_initpid_store(); | |
1090 | store_unlock(); | |
1091 | return answer; | |
1092 | } | |
1093 | ||
1094 | static int wait_for_pid(pid_t pid) | |
1095 | { | |
1096 | int status, ret; | |
1097 | ||
1098 | if (pid <= 0) | |
1099 | return -1; | |
1100 | ||
1101 | again: | |
1102 | ret = waitpid(pid, &status, 0); | |
1103 | if (ret == -1) { | |
1104 | if (errno == EINTR) | |
1105 | goto again; | |
1106 | return -1; | |
1107 | } | |
1108 | if (ret != pid) | |
1109 | goto again; | |
1110 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
1111 | return -1; | |
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | ||
1116 | /* | |
1117 | * append pid to *src. | |
1118 | * src: a pointer to a char* in which ot append the pid. | |
1119 | * sz: the number of characters printed so far, minus trailing \0. | |
1120 | * asz: the allocated size so far | |
1121 | * pid: the pid to append | |
1122 | */ | |
1123 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
1124 | { | |
1125 | char tmp[30]; | |
1126 | ||
1127 | int tmplen = sprintf(tmp, "%d\n", (int)pid); | |
1128 | ||
1129 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1130 | char *tmp; | |
1131 | do { | |
1132 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1133 | } while (!tmp); | |
1134 | *src = tmp; | |
1135 | *asz += BUF_RESERVE_SIZE; | |
1136 | } | |
bbfd0e33 | 1137 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ |
237e200e | 1138 | *sz += tmplen; |
237e200e SH |
1139 | } |
1140 | ||
1141 | /* | |
1142 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1143 | * valid in the caller's namespace, return the id mapped into | |
1144 | * pid's namespace. | |
1145 | * Returns the mapped id, or -1 on error. | |
1146 | */ | |
1147 | unsigned int | |
1148 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1149 | { | |
1150 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1151 | hostuid, // base id for a range in the caller's namespace | |
1152 | count; // number of ids in this range | |
1153 | char line[400]; | |
1154 | int ret; | |
1155 | ||
1156 | fseek(idfile, 0L, SEEK_SET); | |
1157 | while (fgets(line, 400, idfile)) { | |
1158 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1159 | if (ret != 3) | |
1160 | continue; | |
1161 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1162 | /* | |
1163 | * uids wrapped around - unexpected as this is a procfile, | |
1164 | * so just bail. | |
1165 | */ | |
b8defc3d | 1166 | lxcfs_error("pid wrapparound at entry %u %u %u in %s\n", |
237e200e SH |
1167 | nsuid, hostuid, count, line); |
1168 | return -1; | |
1169 | } | |
1170 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1171 | /* | |
1172 | * now since hostuid <= in_id < hostuid+count, and | |
1173 | * hostuid+count and nsuid+count do not wrap around, | |
1174 | * we know that nsuid+(in_id-hostuid) which must be | |
1175 | * less that nsuid+(count) must not wrap around | |
1176 | */ | |
1177 | return (in_id - hostuid) + nsuid; | |
1178 | } | |
1179 | } | |
1180 | ||
1181 | // no answer found | |
1182 | return -1; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * for is_privileged_over, | |
1187 | * specify whether we require the calling uid to be root in his | |
1188 | * namespace | |
1189 | */ | |
1190 | #define NS_ROOT_REQD true | |
1191 | #define NS_ROOT_OPT false | |
1192 | ||
1193 | #define PROCLEN 100 | |
1194 | ||
1195 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1196 | { | |
1197 | char fpath[PROCLEN]; | |
1198 | int ret; | |
1199 | bool answer = false; | |
1200 | uid_t nsuid; | |
1201 | ||
1202 | if (victim == -1 || uid == -1) | |
1203 | return false; | |
1204 | ||
1205 | /* | |
1206 | * If the request is one not requiring root in the namespace, | |
1207 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1208 | * access to files owned by uid 1000 | |
1209 | */ | |
1210 | if (!req_ns_root && uid == victim) | |
1211 | return true; | |
1212 | ||
1213 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1214 | if (ret < 0 || ret >= PROCLEN) | |
1215 | return false; | |
1216 | FILE *f = fopen(fpath, "r"); | |
1217 | if (!f) | |
1218 | return false; | |
1219 | ||
1220 | /* if caller's not root in his namespace, reject */ | |
1221 | nsuid = convert_id_to_ns(f, uid); | |
1222 | if (nsuid) | |
1223 | goto out; | |
1224 | ||
1225 | /* | |
1226 | * If victim is not mapped into caller's ns, reject. | |
1227 | * XXX I'm not sure this check is needed given that fuse | |
1228 | * will be sending requests where the vfs has converted | |
1229 | */ | |
1230 | nsuid = convert_id_to_ns(f, victim); | |
1231 | if (nsuid == -1) | |
1232 | goto out; | |
1233 | ||
1234 | answer = true; | |
1235 | ||
1236 | out: | |
1237 | fclose(f); | |
1238 | return answer; | |
1239 | } | |
1240 | ||
1241 | static bool perms_include(int fmode, mode_t req_mode) | |
1242 | { | |
1243 | mode_t r; | |
1244 | ||
1245 | switch (req_mode & O_ACCMODE) { | |
1246 | case O_RDONLY: | |
1247 | r = S_IROTH; | |
1248 | break; | |
1249 | case O_WRONLY: | |
1250 | r = S_IWOTH; | |
1251 | break; | |
1252 | case O_RDWR: | |
1253 | r = S_IROTH | S_IWOTH; | |
1254 | break; | |
1255 | default: | |
1256 | return false; | |
1257 | } | |
1258 | return ((fmode & r) == r); | |
1259 | } | |
1260 | ||
1261 | ||
1262 | /* | |
1263 | * taskcg is a/b/c | |
1264 | * querycg is /a/b/c/d/e | |
1265 | * we return 'd' | |
1266 | */ | |
1267 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1268 | { | |
1269 | char *start, *end; | |
1270 | ||
1271 | if (strlen(taskcg) <= strlen(querycg)) { | |
b8defc3d | 1272 | lxcfs_error("%s\n", "I was fed bad input."); |
237e200e SH |
1273 | return NULL; |
1274 | } | |
1275 | ||
06081b29 | 1276 | if ((strcmp(querycg, "/") == 0) || (strcmp(querycg, "./") == 0)) |
237e200e SH |
1277 | start = strdup(taskcg + 1); |
1278 | else | |
1279 | start = strdup(taskcg + strlen(querycg) + 1); | |
1280 | if (!start) | |
1281 | return NULL; | |
1282 | end = strchr(start, '/'); | |
1283 | if (end) | |
1284 | *end = '\0'; | |
1285 | return start; | |
1286 | } | |
1287 | ||
1288 | static void stripnewline(char *x) | |
1289 | { | |
1290 | size_t l = strlen(x); | |
1291 | if (l && x[l-1] == '\n') | |
1292 | x[l-1] = '\0'; | |
1293 | } | |
1294 | ||
1295 | static char *get_pid_cgroup(pid_t pid, const char *contrl) | |
1296 | { | |
5dd3e6fd | 1297 | int cfd; |
237e200e SH |
1298 | char fnam[PROCLEN]; |
1299 | FILE *f; | |
1300 | char *answer = NULL; | |
1301 | char *line = NULL; | |
1302 | size_t len = 0; | |
1303 | int ret; | |
5dd3e6fd | 1304 | const char *h = find_mounted_controller(contrl, &cfd); |
237e200e SH |
1305 | if (!h) |
1306 | return NULL; | |
1307 | ||
1308 | ret = snprintf(fnam, PROCLEN, "/proc/%d/cgroup", pid); | |
1309 | if (ret < 0 || ret >= PROCLEN) | |
1310 | return NULL; | |
1311 | if (!(f = fopen(fnam, "r"))) | |
1312 | return NULL; | |
1313 | ||
1314 | while (getline(&line, &len, f) != -1) { | |
1315 | char *c1, *c2; | |
1316 | if (!line[0]) | |
1317 | continue; | |
1318 | c1 = strchr(line, ':'); | |
1319 | if (!c1) | |
1320 | goto out; | |
1321 | c1++; | |
1322 | c2 = strchr(c1, ':'); | |
1323 | if (!c2) | |
1324 | goto out; | |
1325 | *c2 = '\0'; | |
1326 | if (strcmp(c1, h) != 0) | |
1327 | continue; | |
1328 | c2++; | |
1329 | stripnewline(c2); | |
1330 | do { | |
1331 | answer = strdup(c2); | |
1332 | } while (!answer); | |
1333 | break; | |
1334 | } | |
1335 | ||
1336 | out: | |
1337 | fclose(f); | |
1338 | free(line); | |
1339 | return answer; | |
1340 | } | |
1341 | ||
1342 | /* | |
1343 | * check whether a fuse context may access a cgroup dir or file | |
1344 | * | |
1345 | * If file is not null, it is a cgroup file to check under cg. | |
1346 | * If file is null, then we are checking perms on cg itself. | |
1347 | * | |
1348 | * For files we can check the mode of the list_keys result. | |
1349 | * For cgroups, we must make assumptions based on the files under the | |
1350 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1351 | * yet. | |
1352 | */ | |
1353 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1354 | { | |
1355 | struct cgfs_files *k = NULL; | |
1356 | bool ret = false; | |
1357 | ||
1358 | k = cgfs_get_key(contrl, cg, file); | |
1359 | if (!k) | |
1360 | return false; | |
1361 | ||
1362 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1363 | if (perms_include(k->mode >> 6, mode)) { | |
1364 | ret = true; | |
1365 | goto out; | |
1366 | } | |
1367 | } | |
1368 | if (fc->gid == k->gid) { | |
1369 | if (perms_include(k->mode >> 3, mode)) { | |
1370 | ret = true; | |
1371 | goto out; | |
1372 | } | |
1373 | } | |
1374 | ret = perms_include(k->mode, mode); | |
1375 | ||
1376 | out: | |
1377 | free_key(k); | |
1378 | return ret; | |
1379 | } | |
1380 | ||
1381 | #define INITSCOPE "/init.scope" | |
1382 | static void prune_init_slice(char *cg) | |
1383 | { | |
1384 | char *point; | |
1385 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1386 | ||
1387 | if (cg_len < initscope_len) | |
1388 | return; | |
1389 | ||
1390 | point = cg + cg_len - initscope_len; | |
1391 | if (strcmp(point, INITSCOPE) == 0) { | |
1392 | if (point == cg) | |
1393 | *(point+1) = '\0'; | |
1394 | else | |
1395 | *point = '\0'; | |
1396 | } | |
1397 | } | |
1398 | ||
1399 | /* | |
1400 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1401 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1402 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1403 | * to a string containing the next cgroup directory under cg, which must be | |
1404 | * freed by the caller. | |
1405 | */ | |
1406 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1407 | { | |
1408 | bool answer = false; | |
1409 | char *c2 = get_pid_cgroup(pid, contrl); | |
1410 | char *linecmp; | |
1411 | ||
1412 | if (!c2) | |
1413 | return false; | |
1414 | prune_init_slice(c2); | |
1415 | ||
1416 | /* | |
12c31268 CB |
1417 | * callers pass in '/' or './' (openat()) for root cgroup, otherwise |
1418 | * they pass in a cgroup without leading '/' | |
1419 | * | |
1420 | * The original line here was: | |
1421 | * linecmp = *cg == '/' ? c2 : c2+1; | |
1422 | * TODO: I'm not sure why you'd want to increment when *cg != '/'? | |
1423 | * Serge, do you know? | |
237e200e | 1424 | */ |
12c31268 CB |
1425 | if (*cg == '/' || !strncmp(cg, "./", 2)) |
1426 | linecmp = c2; | |
1427 | else | |
1428 | linecmp = c2 + 1; | |
237e200e SH |
1429 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { |
1430 | if (nextcg) { | |
1431 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1432 | } | |
1433 | goto out; | |
1434 | } | |
1435 | answer = true; | |
1436 | ||
1437 | out: | |
1438 | free(c2); | |
1439 | return answer; | |
1440 | } | |
1441 | ||
1442 | /* | |
1443 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1444 | */ | |
1445 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1446 | { | |
1447 | bool answer = false; | |
1448 | char *c2, *task_cg; | |
1449 | size_t target_len, task_len; | |
1450 | ||
f7bff426 | 1451 | if (strcmp(cg, "/") == 0 || strcmp(cg, "./") == 0) |
237e200e SH |
1452 | return true; |
1453 | ||
1454 | c2 = get_pid_cgroup(pid, contrl); | |
1455 | if (!c2) | |
1456 | return false; | |
1457 | prune_init_slice(c2); | |
1458 | ||
1459 | task_cg = c2 + 1; | |
1460 | target_len = strlen(cg); | |
1461 | task_len = strlen(task_cg); | |
1462 | if (task_len == 0) { | |
1463 | /* Task is in the root cg, it can see everything. This case is | |
1464 | * not handled by the strmcps below, since they test for the | |
1465 | * last /, but that is the first / that we've chopped off | |
1466 | * above. | |
1467 | */ | |
1468 | answer = true; | |
1469 | goto out; | |
1470 | } | |
1471 | if (strcmp(cg, task_cg) == 0) { | |
1472 | answer = true; | |
1473 | goto out; | |
1474 | } | |
1475 | if (target_len < task_len) { | |
1476 | /* looking up a parent dir */ | |
1477 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1478 | answer = true; | |
1479 | goto out; | |
1480 | } | |
1481 | if (target_len > task_len) { | |
1482 | /* looking up a child dir */ | |
1483 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1484 | answer = true; | |
1485 | goto out; | |
1486 | } | |
1487 | ||
1488 | out: | |
1489 | free(c2); | |
1490 | return answer; | |
1491 | } | |
1492 | ||
1493 | /* | |
1494 | * given /cgroup/freezer/a/b, return "freezer". | |
1495 | * the returned char* should NOT be freed. | |
1496 | */ | |
1497 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1498 | { | |
1499 | const char *p1; | |
1500 | char *contr, *slash; | |
1501 | ||
99142521 | 1502 | if (strlen(path) < 9) { |
e254948f | 1503 | errno = EACCES; |
237e200e | 1504 | return NULL; |
99142521 CB |
1505 | } |
1506 | if (*(path + 7) != '/') { | |
1507 | errno = EINVAL; | |
237e200e | 1508 | return NULL; |
99142521 | 1509 | } |
3adc421c | 1510 | p1 = path + 8; |
237e200e | 1511 | contr = strdupa(p1); |
99142521 CB |
1512 | if (!contr) { |
1513 | errno = ENOMEM; | |
237e200e | 1514 | return NULL; |
99142521 | 1515 | } |
237e200e SH |
1516 | slash = strstr(contr, "/"); |
1517 | if (slash) | |
1518 | *slash = '\0'; | |
1519 | ||
1520 | int i; | |
3adc421c | 1521 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
1522 | if (hierarchies[i] && strcmp(hierarchies[i], contr) == 0) |
1523 | return hierarchies[i]; | |
1524 | } | |
99142521 | 1525 | errno = ENOENT; |
237e200e SH |
1526 | return NULL; |
1527 | } | |
1528 | ||
1529 | /* | |
1530 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1531 | * Note that the returned value may include files (keynames) etc | |
1532 | */ | |
1533 | static const char *find_cgroup_in_path(const char *path) | |
1534 | { | |
1535 | const char *p1; | |
1536 | ||
bc70ba9b | 1537 | if (strlen(path) < 9) { |
e254948f | 1538 | errno = EACCES; |
237e200e | 1539 | return NULL; |
bc70ba9b CB |
1540 | } |
1541 | p1 = strstr(path + 8, "/"); | |
1542 | if (!p1) { | |
1543 | errno = EINVAL; | |
237e200e | 1544 | return NULL; |
bc70ba9b CB |
1545 | } |
1546 | errno = 0; | |
1547 | return p1 + 1; | |
237e200e SH |
1548 | } |
1549 | ||
1550 | /* | |
1551 | * split the last path element from the path in @cg. | |
1552 | * @dir is newly allocated and should be freed, @last not | |
1553 | */ | |
1554 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1555 | { | |
1556 | char *p; | |
1557 | ||
1558 | do { | |
1559 | *dir = strdup(cg); | |
1560 | } while (!*dir); | |
1561 | *last = strrchr(cg, '/'); | |
1562 | if (!*last) { | |
1563 | *last = NULL; | |
1564 | return; | |
1565 | } | |
1566 | p = strrchr(*dir, '/'); | |
1567 | *p = '\0'; | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * FUSE ops for /cgroup | |
1572 | */ | |
1573 | ||
1574 | int cg_getattr(const char *path, struct stat *sb) | |
1575 | { | |
1576 | struct timespec now; | |
1577 | struct fuse_context *fc = fuse_get_context(); | |
1578 | char * cgdir = NULL; | |
1579 | char *last = NULL, *path1, *path2; | |
1580 | struct cgfs_files *k = NULL; | |
1581 | const char *cgroup; | |
1582 | const char *controller = NULL; | |
1583 | int ret = -ENOENT; | |
1584 | ||
1585 | ||
1586 | if (!fc) | |
1587 | return -EIO; | |
1588 | ||
1589 | memset(sb, 0, sizeof(struct stat)); | |
1590 | ||
1591 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1592 | return -EINVAL; | |
1593 | ||
1594 | sb->st_uid = sb->st_gid = 0; | |
1595 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1596 | sb->st_size = 0; | |
1597 | ||
1598 | if (strcmp(path, "/cgroup") == 0) { | |
1599 | sb->st_mode = S_IFDIR | 00755; | |
1600 | sb->st_nlink = 2; | |
1601 | return 0; | |
1602 | } | |
1603 | ||
1604 | controller = pick_controller_from_path(fc, path); | |
1605 | if (!controller) | |
2f7036d0 | 1606 | return -errno; |
237e200e SH |
1607 | cgroup = find_cgroup_in_path(path); |
1608 | if (!cgroup) { | |
1609 | /* this is just /cgroup/controller, return it as a dir */ | |
1610 | sb->st_mode = S_IFDIR | 00755; | |
1611 | sb->st_nlink = 2; | |
1612 | return 0; | |
1613 | } | |
1614 | ||
1615 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1616 | ||
1617 | if (!last) { | |
1618 | path1 = "/"; | |
1619 | path2 = cgdir; | |
1620 | } else { | |
1621 | path1 = cgdir; | |
1622 | path2 = last; | |
1623 | } | |
1624 | ||
1625 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1626 | if (initpid <= 0) | |
1627 | initpid = fc->pid; | |
1628 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
1629 | * Then check that caller's cgroup is under path if last is a child | |
1630 | * cgroup, or cgdir if last is a file */ | |
1631 | ||
1632 | if (is_child_cgroup(controller, path1, path2)) { | |
1633 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
1634 | ret = -ENOENT; | |
1635 | goto out; | |
1636 | } | |
1637 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
1638 | /* this is just /cgroup/controller, return it as a dir */ | |
1639 | sb->st_mode = S_IFDIR | 00555; | |
1640 | sb->st_nlink = 2; | |
1641 | ret = 0; | |
1642 | goto out; | |
1643 | } | |
1644 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
1645 | ret = -EACCES; | |
1646 | goto out; | |
1647 | } | |
1648 | ||
1649 | // get uid, gid, from '/tasks' file and make up a mode | |
1650 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1651 | sb->st_mode = S_IFDIR | 00755; | |
1652 | k = cgfs_get_key(controller, cgroup, NULL); | |
1653 | if (!k) { | |
1654 | sb->st_uid = sb->st_gid = 0; | |
1655 | } else { | |
1656 | sb->st_uid = k->uid; | |
1657 | sb->st_gid = k->gid; | |
1658 | } | |
1659 | free_key(k); | |
1660 | sb->st_nlink = 2; | |
1661 | ret = 0; | |
1662 | goto out; | |
1663 | } | |
1664 | ||
1665 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
1666 | sb->st_mode = S_IFREG | k->mode; | |
1667 | sb->st_nlink = 1; | |
1668 | sb->st_uid = k->uid; | |
1669 | sb->st_gid = k->gid; | |
1670 | sb->st_size = 0; | |
1671 | free_key(k); | |
1672 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
1673 | ret = -ENOENT; | |
1674 | goto out; | |
1675 | } | |
237e200e SH |
1676 | ret = 0; |
1677 | } | |
1678 | ||
1679 | out: | |
1680 | free(cgdir); | |
1681 | return ret; | |
1682 | } | |
1683 | ||
1684 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
1685 | { | |
1686 | struct fuse_context *fc = fuse_get_context(); | |
1687 | const char *cgroup; | |
1688 | struct file_info *dir_info; | |
1689 | char *controller = NULL; | |
1690 | ||
1691 | if (!fc) | |
1692 | return -EIO; | |
1693 | ||
1694 | if (strcmp(path, "/cgroup") == 0) { | |
1695 | cgroup = NULL; | |
1696 | controller = NULL; | |
1697 | } else { | |
1698 | // return list of keys for the controller, and list of child cgroups | |
1699 | controller = pick_controller_from_path(fc, path); | |
1700 | if (!controller) | |
2f7036d0 | 1701 | return -errno; |
237e200e SH |
1702 | |
1703 | cgroup = find_cgroup_in_path(path); | |
1704 | if (!cgroup) { | |
1705 | /* this is just /cgroup/controller, return its contents */ | |
1706 | cgroup = "/"; | |
1707 | } | |
1708 | } | |
1709 | ||
1710 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1711 | if (initpid <= 0) | |
1712 | initpid = fc->pid; | |
1713 | if (cgroup) { | |
1714 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
1715 | return -ENOENT; | |
1716 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
1717 | return -EACCES; | |
1718 | } | |
1719 | ||
1720 | /* we'll free this at cg_releasedir */ | |
1721 | dir_info = malloc(sizeof(*dir_info)); | |
1722 | if (!dir_info) | |
1723 | return -ENOMEM; | |
1724 | dir_info->controller = must_copy_string(controller); | |
1725 | dir_info->cgroup = must_copy_string(cgroup); | |
1726 | dir_info->type = LXC_TYPE_CGDIR; | |
1727 | dir_info->buf = NULL; | |
1728 | dir_info->file = NULL; | |
1729 | dir_info->buflen = 0; | |
1730 | ||
1731 | fi->fh = (unsigned long)dir_info; | |
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1736 | struct fuse_file_info *fi) | |
1737 | { | |
1738 | struct file_info *d = (struct file_info *)fi->fh; | |
1739 | struct cgfs_files **list = NULL; | |
1740 | int i, ret; | |
1741 | char *nextcg = NULL; | |
1742 | struct fuse_context *fc = fuse_get_context(); | |
1743 | char **clist = NULL; | |
1744 | ||
d639f863 CB |
1745 | if (filler(buf, ".", NULL, 0) != 0 || filler(buf, "..", NULL, 0) != 0) |
1746 | return -EIO; | |
1747 | ||
237e200e | 1748 | if (d->type != LXC_TYPE_CGDIR) { |
b8defc3d | 1749 | lxcfs_error("%s\n", "Internal error: file cache info used in readdir."); |
237e200e SH |
1750 | return -EIO; |
1751 | } | |
1752 | if (!d->cgroup && !d->controller) { | |
1753 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
1754 | int i; | |
1755 | ||
1756 | for (i = 0; i < num_hierarchies; i++) { | |
1757 | if (hierarchies[i] && filler(buf, hierarchies[i], NULL, 0) != 0) { | |
1758 | return -EIO; | |
1759 | } | |
1760 | } | |
1761 | return 0; | |
1762 | } | |
1763 | ||
1764 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
1765 | // not a valid cgroup | |
1766 | ret = -EINVAL; | |
1767 | goto out; | |
1768 | } | |
1769 | ||
1770 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1771 | if (initpid <= 0) | |
1772 | initpid = fc->pid; | |
1773 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
1774 | if (nextcg) { | |
1775 | ret = filler(buf, nextcg, NULL, 0); | |
1776 | free(nextcg); | |
1777 | if (ret != 0) { | |
1778 | ret = -EIO; | |
1779 | goto out; | |
1780 | } | |
1781 | } | |
1782 | ret = 0; | |
1783 | goto out; | |
1784 | } | |
1785 | ||
1786 | for (i = 0; list[i]; i++) { | |
1787 | if (filler(buf, list[i]->name, NULL, 0) != 0) { | |
1788 | ret = -EIO; | |
1789 | goto out; | |
1790 | } | |
1791 | } | |
1792 | ||
1793 | // now get the list of child cgroups | |
1794 | ||
1795 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
1796 | ret = 0; | |
1797 | goto out; | |
1798 | } | |
f366da65 WB |
1799 | if (clist) { |
1800 | for (i = 0; clist[i]; i++) { | |
1801 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
1802 | ret = -EIO; | |
1803 | goto out; | |
1804 | } | |
237e200e SH |
1805 | } |
1806 | } | |
1807 | ret = 0; | |
1808 | ||
1809 | out: | |
1810 | free_keys(list); | |
1811 | if (clist) { | |
1812 | for (i = 0; clist[i]; i++) | |
1813 | free(clist[i]); | |
1814 | free(clist); | |
1815 | } | |
1816 | return ret; | |
1817 | } | |
1818 | ||
43215927 | 1819 | static void do_release_file_info(struct fuse_file_info *fi) |
237e200e | 1820 | { |
43215927 SH |
1821 | struct file_info *f = (struct file_info *)fi->fh; |
1822 | ||
237e200e SH |
1823 | if (!f) |
1824 | return; | |
43215927 SH |
1825 | |
1826 | fi->fh = 0; | |
1827 | ||
237e200e | 1828 | free(f->controller); |
43215927 | 1829 | f->controller = NULL; |
237e200e | 1830 | free(f->cgroup); |
43215927 | 1831 | f->cgroup = NULL; |
237e200e | 1832 | free(f->file); |
43215927 | 1833 | f->file = NULL; |
237e200e | 1834 | free(f->buf); |
43215927 | 1835 | f->buf = NULL; |
237e200e SH |
1836 | free(f); |
1837 | } | |
1838 | ||
1839 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
1840 | { | |
43215927 | 1841 | do_release_file_info(fi); |
237e200e SH |
1842 | return 0; |
1843 | } | |
1844 | ||
1845 | int cg_open(const char *path, struct fuse_file_info *fi) | |
1846 | { | |
1847 | const char *cgroup; | |
1848 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
1849 | struct cgfs_files *k = NULL; | |
1850 | struct file_info *file_info; | |
1851 | struct fuse_context *fc = fuse_get_context(); | |
1852 | int ret; | |
1853 | ||
1854 | if (!fc) | |
1855 | return -EIO; | |
1856 | ||
1857 | controller = pick_controller_from_path(fc, path); | |
1858 | if (!controller) | |
2f7036d0 | 1859 | return -errno; |
237e200e SH |
1860 | cgroup = find_cgroup_in_path(path); |
1861 | if (!cgroup) | |
bc70ba9b | 1862 | return -errno; |
237e200e SH |
1863 | |
1864 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1865 | if (!last) { | |
1866 | path1 = "/"; | |
1867 | path2 = cgdir; | |
1868 | } else { | |
1869 | path1 = cgdir; | |
1870 | path2 = last; | |
1871 | } | |
1872 | ||
1873 | k = cgfs_get_key(controller, path1, path2); | |
1874 | if (!k) { | |
1875 | ret = -EINVAL; | |
1876 | goto out; | |
1877 | } | |
1878 | free_key(k); | |
1879 | ||
1880 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1881 | if (initpid <= 0) | |
1882 | initpid = fc->pid; | |
1883 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1884 | ret = -ENOENT; | |
1885 | goto out; | |
1886 | } | |
1887 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
237e200e SH |
1888 | ret = -EACCES; |
1889 | goto out; | |
1890 | } | |
1891 | ||
1892 | /* we'll free this at cg_release */ | |
1893 | file_info = malloc(sizeof(*file_info)); | |
1894 | if (!file_info) { | |
1895 | ret = -ENOMEM; | |
1896 | goto out; | |
1897 | } | |
1898 | file_info->controller = must_copy_string(controller); | |
1899 | file_info->cgroup = must_copy_string(path1); | |
1900 | file_info->file = must_copy_string(path2); | |
1901 | file_info->type = LXC_TYPE_CGFILE; | |
1902 | file_info->buf = NULL; | |
1903 | file_info->buflen = 0; | |
1904 | ||
1905 | fi->fh = (unsigned long)file_info; | |
1906 | ret = 0; | |
1907 | ||
1908 | out: | |
1909 | free(cgdir); | |
1910 | return ret; | |
1911 | } | |
1912 | ||
bddbb106 SH |
1913 | int cg_access(const char *path, int mode) |
1914 | { | |
6f0f6b83 | 1915 | int ret; |
bddbb106 | 1916 | const char *cgroup; |
6f0f6b83 CB |
1917 | char *path1, *path2, *controller; |
1918 | char *last = NULL, *cgdir = NULL; | |
bddbb106 SH |
1919 | struct cgfs_files *k = NULL; |
1920 | struct fuse_context *fc = fuse_get_context(); | |
6f0f6b83 | 1921 | |
9873c5e8 | 1922 | if (strcmp(path, "/cgroup") == 0) |
6f0f6b83 | 1923 | return 0; |
bddbb106 SH |
1924 | |
1925 | if (!fc) | |
1926 | return -EIO; | |
1927 | ||
1928 | controller = pick_controller_from_path(fc, path); | |
1929 | if (!controller) | |
2f7036d0 | 1930 | return -errno; |
bddbb106 | 1931 | cgroup = find_cgroup_in_path(path); |
575316c4 SH |
1932 | if (!cgroup) { |
1933 | // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not | |
3f441bc7 SH |
1934 | if ((mode & W_OK) == 0) |
1935 | return 0; | |
1936 | return -EACCES; | |
575316c4 | 1937 | } |
bddbb106 SH |
1938 | |
1939 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1940 | if (!last) { | |
1941 | path1 = "/"; | |
1942 | path2 = cgdir; | |
1943 | } else { | |
1944 | path1 = cgdir; | |
1945 | path2 = last; | |
1946 | } | |
1947 | ||
1948 | k = cgfs_get_key(controller, path1, path2); | |
1949 | if (!k) { | |
3f441bc7 SH |
1950 | if ((mode & W_OK) == 0) |
1951 | ret = 0; | |
1952 | else | |
1953 | ret = -EACCES; | |
bddbb106 SH |
1954 | goto out; |
1955 | } | |
1956 | free_key(k); | |
1957 | ||
1958 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1959 | if (initpid <= 0) | |
1960 | initpid = fc->pid; | |
1961 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1962 | ret = -ENOENT; | |
1963 | goto out; | |
1964 | } | |
1965 | if (!fc_may_access(fc, controller, path1, path2, mode)) { | |
1966 | ret = -EACCES; | |
1967 | goto out; | |
1968 | } | |
1969 | ||
1970 | ret = 0; | |
1971 | ||
1972 | out: | |
1973 | free(cgdir); | |
1974 | return ret; | |
1975 | } | |
1976 | ||
237e200e SH |
1977 | int cg_release(const char *path, struct fuse_file_info *fi) |
1978 | { | |
43215927 | 1979 | do_release_file_info(fi); |
237e200e SH |
1980 | return 0; |
1981 | } | |
1982 | ||
1983 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
1984 | ||
1985 | static bool wait_for_sock(int sock, int timeout) | |
1986 | { | |
1987 | struct epoll_event ev; | |
1988 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
1989 | ||
1990 | if ((starttime = time(NULL)) < 0) | |
1991 | return false; | |
1992 | ||
1993 | if ((epfd = epoll_create(1)) < 0) { | |
b8defc3d | 1994 | lxcfs_error("%s\n", "Failed to create epoll socket: %m."); |
237e200e SH |
1995 | return false; |
1996 | } | |
1997 | ||
1998 | ev.events = POLLIN_SET; | |
1999 | ev.data.fd = sock; | |
2000 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
b8defc3d | 2001 | lxcfs_error("%s\n", "Failed adding socket to epoll: %m."); |
237e200e SH |
2002 | close(epfd); |
2003 | return false; | |
2004 | } | |
2005 | ||
2006 | again: | |
2007 | if ((now = time(NULL)) < 0) { | |
2008 | close(epfd); | |
2009 | return false; | |
2010 | } | |
2011 | ||
2012 | deltatime = (starttime + timeout) - now; | |
2013 | if (deltatime < 0) { // timeout | |
2014 | errno = 0; | |
2015 | close(epfd); | |
2016 | return false; | |
2017 | } | |
2018 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
2019 | if (ret < 0 && errno == EINTR) | |
2020 | goto again; | |
2021 | saved_errno = errno; | |
2022 | close(epfd); | |
2023 | ||
2024 | if (ret <= 0) { | |
2025 | errno = saved_errno; | |
2026 | return false; | |
2027 | } | |
2028 | return true; | |
2029 | } | |
2030 | ||
2031 | static int msgrecv(int sockfd, void *buf, size_t len) | |
2032 | { | |
2033 | if (!wait_for_sock(sockfd, 2)) | |
2034 | return -1; | |
2035 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
2036 | } | |
2037 | ||
2038 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
2039 | { | |
2040 | struct msghdr msg = { 0 }; | |
2041 | struct iovec iov; | |
2042 | struct cmsghdr *cmsg; | |
2043 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2044 | char buf[1]; | |
2045 | buf[0] = 'p'; | |
2046 | ||
2047 | if (pingfirst) { | |
2048 | if (msgrecv(sock, buf, 1) != 1) { | |
b8defc3d | 2049 | lxcfs_error("%s\n", "Error getting reply from server over socketpair."); |
237e200e SH |
2050 | return SEND_CREDS_FAIL; |
2051 | } | |
2052 | } | |
2053 | ||
2054 | msg.msg_control = cmsgbuf; | |
2055 | msg.msg_controllen = sizeof(cmsgbuf); | |
2056 | ||
2057 | cmsg = CMSG_FIRSTHDR(&msg); | |
2058 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
2059 | cmsg->cmsg_level = SOL_SOCKET; | |
2060 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
2061 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
2062 | ||
2063 | msg.msg_name = NULL; | |
2064 | msg.msg_namelen = 0; | |
2065 | ||
2066 | buf[0] = v; | |
2067 | iov.iov_base = buf; | |
2068 | iov.iov_len = sizeof(buf); | |
2069 | msg.msg_iov = &iov; | |
2070 | msg.msg_iovlen = 1; | |
2071 | ||
2072 | if (sendmsg(sock, &msg, 0) < 0) { | |
b8defc3d | 2073 | lxcfs_error("Failed at sendmsg: %s.\n",strerror(errno)); |
237e200e SH |
2074 | if (errno == 3) |
2075 | return SEND_CREDS_NOTSK; | |
2076 | return SEND_CREDS_FAIL; | |
2077 | } | |
2078 | ||
2079 | return SEND_CREDS_OK; | |
2080 | } | |
2081 | ||
2082 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
2083 | { | |
2084 | struct msghdr msg = { 0 }; | |
2085 | struct iovec iov; | |
2086 | struct cmsghdr *cmsg; | |
2087 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2088 | char buf[1]; | |
2089 | int ret; | |
2090 | int optval = 1; | |
2091 | ||
2092 | *v = '1'; | |
2093 | ||
2094 | cred->pid = -1; | |
2095 | cred->uid = -1; | |
2096 | cred->gid = -1; | |
2097 | ||
2098 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
b8defc3d | 2099 | lxcfs_error("Failed to set passcred: %s\n", strerror(errno)); |
237e200e SH |
2100 | return false; |
2101 | } | |
2102 | buf[0] = '1'; | |
2103 | if (write(sock, buf, 1) != 1) { | |
b8defc3d | 2104 | lxcfs_error("Failed to start write on scm fd: %s\n", strerror(errno)); |
237e200e SH |
2105 | return false; |
2106 | } | |
2107 | ||
2108 | msg.msg_name = NULL; | |
2109 | msg.msg_namelen = 0; | |
2110 | msg.msg_control = cmsgbuf; | |
2111 | msg.msg_controllen = sizeof(cmsgbuf); | |
2112 | ||
2113 | iov.iov_base = buf; | |
2114 | iov.iov_len = sizeof(buf); | |
2115 | msg.msg_iov = &iov; | |
2116 | msg.msg_iovlen = 1; | |
2117 | ||
2118 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2119 | lxcfs_error("Timed out waiting for scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2120 | return false; |
2121 | } | |
2122 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
2123 | if (ret < 0) { | |
b8defc3d | 2124 | lxcfs_error("Failed to receive scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2125 | return false; |
2126 | } | |
2127 | ||
2128 | cmsg = CMSG_FIRSTHDR(&msg); | |
2129 | ||
2130 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
2131 | cmsg->cmsg_level == SOL_SOCKET && | |
2132 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
2133 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
2134 | } | |
2135 | *v = buf[0]; | |
2136 | ||
2137 | return true; | |
2138 | } | |
2139 | ||
35174b0f FG |
2140 | struct pid_ns_clone_args { |
2141 | int *cpipe; | |
2142 | int sock; | |
2143 | pid_t tpid; | |
2144 | int (*wrapped) (int, pid_t); // pid_from_ns or pid_to_ns | |
2145 | }; | |
2146 | ||
2147 | /* | |
2148 | * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage | |
2149 | * with clone(). This simply writes '1' as ACK back to the parent | |
2150 | * before calling the actual wrapped function. | |
2151 | */ | |
2152 | static int pid_ns_clone_wrapper(void *arg) { | |
2153 | struct pid_ns_clone_args* args = (struct pid_ns_clone_args *) arg; | |
2154 | char b = '1'; | |
2155 | ||
2156 | close(args->cpipe[0]); | |
b8defc3d CB |
2157 | if (write(args->cpipe[1], &b, sizeof(char)) < 0) |
2158 | lxcfs_error("(child): error on write: %s.\n", strerror(errno)); | |
35174b0f FG |
2159 | close(args->cpipe[1]); |
2160 | return args->wrapped(args->sock, args->tpid); | |
2161 | } | |
237e200e SH |
2162 | |
2163 | /* | |
2164 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
2165 | * int value back over the socket. This shifts the pid from the | |
2166 | * sender's pidns into tpid's pidns. | |
2167 | */ | |
35174b0f | 2168 | static int pid_to_ns(int sock, pid_t tpid) |
237e200e SH |
2169 | { |
2170 | char v = '0'; | |
2171 | struct ucred cred; | |
2172 | ||
2173 | while (recv_creds(sock, &cred, &v)) { | |
2174 | if (v == '1') | |
35174b0f | 2175 | return 0; |
237e200e | 2176 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) |
35174b0f | 2177 | return 1; |
237e200e | 2178 | } |
35174b0f | 2179 | return 0; |
237e200e SH |
2180 | } |
2181 | ||
35174b0f | 2182 | |
237e200e SH |
2183 | /* |
2184 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
35174b0f FG |
2185 | * in your old pidns. Only children which you clone will be in the target |
2186 | * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to | |
2187 | * actually convert pids. | |
2188 | * | |
2189 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
2190 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
2191 | * the pidns and the parent pid outside are identical. Using clone prevents | |
2192 | * this issue. | |
237e200e SH |
2193 | */ |
2194 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
2195 | { | |
2196 | int newnsfd = -1, ret, cpipe[2]; | |
2197 | char fnam[100]; | |
2198 | pid_t cpid; | |
2199 | char v; | |
2200 | ||
2201 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2202 | if (ret < 0 || ret >= sizeof(fnam)) | |
2203 | _exit(1); | |
2204 | newnsfd = open(fnam, O_RDONLY); | |
2205 | if (newnsfd < 0) | |
2206 | _exit(1); | |
2207 | if (setns(newnsfd, 0) < 0) | |
2208 | _exit(1); | |
2209 | close(newnsfd); | |
2210 | ||
2211 | if (pipe(cpipe) < 0) | |
2212 | _exit(1); | |
2213 | ||
35174b0f FG |
2214 | struct pid_ns_clone_args args = { |
2215 | .cpipe = cpipe, | |
2216 | .sock = sock, | |
2217 | .tpid = tpid, | |
2218 | .wrapped = &pid_to_ns | |
2219 | }; | |
2220 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2221 | void *stack = alloca(stack_size); | |
2222 | ||
2223 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2224 | if (cpid < 0) |
2225 | _exit(1); | |
2226 | ||
237e200e SH |
2227 | // give the child 1 second to be done forking and |
2228 | // write its ack | |
2229 | if (!wait_for_sock(cpipe[0], 1)) | |
2230 | _exit(1); | |
2231 | ret = read(cpipe[0], &v, 1); | |
2232 | if (ret != sizeof(char) || v != '1') | |
2233 | _exit(1); | |
2234 | ||
2235 | if (!wait_for_pid(cpid)) | |
2236 | _exit(1); | |
2237 | _exit(0); | |
2238 | } | |
2239 | ||
2240 | /* | |
2241 | * To read cgroup files with a particular pid, we will setns into the child | |
2242 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2243 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2244 | */ | |
2245 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2246 | { | |
2247 | int sock[2] = {-1, -1}; | |
2248 | char *tmpdata = NULL; | |
2249 | int ret; | |
2250 | pid_t qpid, cpid = -1; | |
2251 | bool answer = false; | |
2252 | char v = '0'; | |
2253 | struct ucred cred; | |
2254 | size_t sz = 0, asz = 0; | |
2255 | ||
2256 | if (!cgfs_get_value(contrl, cg, file, &tmpdata)) | |
2257 | return false; | |
2258 | ||
2259 | /* | |
2260 | * Now we read the pids from returned data one by one, pass | |
2261 | * them into a child in the target namespace, read back the | |
2262 | * translated pids, and put them into our to-return data | |
2263 | */ | |
2264 | ||
2265 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2266 | perror("socketpair"); | |
2267 | free(tmpdata); | |
2268 | return false; | |
2269 | } | |
2270 | ||
2271 | cpid = fork(); | |
2272 | if (cpid == -1) | |
2273 | goto out; | |
2274 | ||
2275 | if (!cpid) // child - exits when done | |
2276 | pid_to_ns_wrapper(sock[1], tpid); | |
2277 | ||
2278 | char *ptr = tmpdata; | |
2279 | cred.uid = 0; | |
2280 | cred.gid = 0; | |
2281 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2282 | cred.pid = qpid; | |
2283 | ret = send_creds(sock[0], &cred, v, true); | |
2284 | ||
2285 | if (ret == SEND_CREDS_NOTSK) | |
2286 | goto next; | |
2287 | if (ret == SEND_CREDS_FAIL) | |
2288 | goto out; | |
2289 | ||
2290 | // read converted results | |
2291 | if (!wait_for_sock(sock[0], 2)) { | |
b8defc3d | 2292 | lxcfs_error("Timed out waiting for pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2293 | goto out; |
2294 | } | |
2295 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2296 | lxcfs_error("Error reading pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2297 | goto out; |
2298 | } | |
2299 | must_strcat_pid(d, &sz, &asz, qpid); | |
2300 | next: | |
2301 | ptr = strchr(ptr, '\n'); | |
2302 | if (!ptr) | |
2303 | break; | |
2304 | ptr++; | |
2305 | } | |
2306 | ||
2307 | cred.pid = getpid(); | |
2308 | v = '1'; | |
2309 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2310 | // failed to ask child to exit | |
b8defc3d | 2311 | lxcfs_error("Failed to ask child to exit: %s.\n", strerror(errno)); |
237e200e SH |
2312 | goto out; |
2313 | } | |
2314 | ||
2315 | answer = true; | |
2316 | ||
2317 | out: | |
2318 | free(tmpdata); | |
2319 | if (cpid != -1) | |
2320 | wait_for_pid(cpid); | |
2321 | if (sock[0] != -1) { | |
2322 | close(sock[0]); | |
2323 | close(sock[1]); | |
2324 | } | |
2325 | return answer; | |
2326 | } | |
2327 | ||
2328 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2329 | struct fuse_file_info *fi) | |
2330 | { | |
2331 | struct fuse_context *fc = fuse_get_context(); | |
2332 | struct file_info *f = (struct file_info *)fi->fh; | |
2333 | struct cgfs_files *k = NULL; | |
2334 | char *data = NULL; | |
2335 | int ret, s; | |
2336 | bool r; | |
2337 | ||
2338 | if (f->type != LXC_TYPE_CGFILE) { | |
b8defc3d | 2339 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_read."); |
237e200e SH |
2340 | return -EIO; |
2341 | } | |
2342 | ||
2343 | if (offset) | |
2344 | return 0; | |
2345 | ||
2346 | if (!fc) | |
2347 | return -EIO; | |
2348 | ||
2349 | if (!f->controller) | |
2350 | return -EINVAL; | |
2351 | ||
2352 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2353 | return -EINVAL; | |
2354 | } | |
2355 | free_key(k); | |
2356 | ||
2357 | ||
888f8f3c | 2358 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { |
237e200e SH |
2359 | ret = -EACCES; |
2360 | goto out; | |
2361 | } | |
2362 | ||
2363 | if (strcmp(f->file, "tasks") == 0 || | |
2364 | strcmp(f->file, "/tasks") == 0 || | |
2365 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2366 | strcmp(f->file, "cgroup.procs") == 0) | |
2367 | // special case - we have to translate the pids | |
2368 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2369 | else | |
2370 | r = cgfs_get_value(f->controller, f->cgroup, f->file, &data); | |
2371 | ||
2372 | if (!r) { | |
2373 | ret = -EINVAL; | |
2374 | goto out; | |
2375 | } | |
2376 | ||
2377 | if (!data) { | |
2378 | ret = 0; | |
2379 | goto out; | |
2380 | } | |
2381 | s = strlen(data); | |
2382 | if (s > size) | |
2383 | s = size; | |
2384 | memcpy(buf, data, s); | |
2385 | if (s > 0 && s < size && data[s-1] != '\n') | |
2386 | buf[s++] = '\n'; | |
2387 | ||
2388 | ret = s; | |
2389 | ||
2390 | out: | |
2391 | free(data); | |
2392 | return ret; | |
2393 | } | |
2394 | ||
35174b0f | 2395 | static int pid_from_ns(int sock, pid_t tpid) |
237e200e SH |
2396 | { |
2397 | pid_t vpid; | |
2398 | struct ucred cred; | |
2399 | char v; | |
2400 | int ret; | |
2401 | ||
2402 | cred.uid = 0; | |
2403 | cred.gid = 0; | |
2404 | while (1) { | |
2405 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2406 | lxcfs_error("%s\n", "Timeout reading from parent."); |
35174b0f | 2407 | return 1; |
237e200e SH |
2408 | } |
2409 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
b8defc3d | 2410 | lxcfs_error("Bad read from parent: %s.\n", strerror(errno)); |
35174b0f | 2411 | return 1; |
237e200e SH |
2412 | } |
2413 | if (vpid == -1) // done | |
2414 | break; | |
2415 | v = '0'; | |
2416 | cred.pid = vpid; | |
2417 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2418 | v = '1'; | |
2419 | cred.pid = getpid(); | |
2420 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
35174b0f | 2421 | return 1; |
237e200e SH |
2422 | } |
2423 | } | |
35174b0f | 2424 | return 0; |
237e200e SH |
2425 | } |
2426 | ||
2427 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2428 | { | |
2429 | int newnsfd = -1, ret, cpipe[2]; | |
2430 | char fnam[100]; | |
2431 | pid_t cpid; | |
2432 | char v; | |
2433 | ||
2434 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2435 | if (ret < 0 || ret >= sizeof(fnam)) | |
2436 | _exit(1); | |
2437 | newnsfd = open(fnam, O_RDONLY); | |
2438 | if (newnsfd < 0) | |
2439 | _exit(1); | |
2440 | if (setns(newnsfd, 0) < 0) | |
2441 | _exit(1); | |
2442 | close(newnsfd); | |
2443 | ||
2444 | if (pipe(cpipe) < 0) | |
2445 | _exit(1); | |
2446 | ||
35174b0f FG |
2447 | struct pid_ns_clone_args args = { |
2448 | .cpipe = cpipe, | |
2449 | .sock = sock, | |
2450 | .tpid = tpid, | |
2451 | .wrapped = &pid_from_ns | |
2452 | }; | |
f0f8b851 SH |
2453 | size_t stack_size = sysconf(_SC_PAGESIZE); |
2454 | void *stack = alloca(stack_size); | |
35174b0f FG |
2455 | |
2456 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2457 | if (cpid < 0) |
2458 | _exit(1); | |
2459 | ||
237e200e SH |
2460 | // give the child 1 second to be done forking and |
2461 | // write its ack | |
2462 | if (!wait_for_sock(cpipe[0], 1)) | |
f0f8b851 | 2463 | _exit(1); |
237e200e | 2464 | ret = read(cpipe[0], &v, 1); |
f0f8b851 SH |
2465 | if (ret != sizeof(char) || v != '1') |
2466 | _exit(1); | |
237e200e SH |
2467 | |
2468 | if (!wait_for_pid(cpid)) | |
2469 | _exit(1); | |
2470 | _exit(0); | |
237e200e SH |
2471 | } |
2472 | ||
2473 | /* | |
2474 | * Given host @uid, return the uid to which it maps in | |
2475 | * @pid's user namespace, or -1 if none. | |
2476 | */ | |
2477 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2478 | { | |
2479 | FILE *f; | |
2480 | char line[400]; | |
2481 | ||
2482 | sprintf(line, "/proc/%d/uid_map", pid); | |
2483 | if ((f = fopen(line, "r")) == NULL) { | |
2484 | return false; | |
2485 | } | |
2486 | ||
2487 | *answer = convert_id_to_ns(f, uid); | |
2488 | fclose(f); | |
2489 | ||
2490 | if (*answer == -1) | |
2491 | return false; | |
2492 | return true; | |
2493 | } | |
2494 | ||
2495 | /* | |
2496 | * get_pid_creds: get the real uid and gid of @pid from | |
2497 | * /proc/$$/status | |
2498 | * (XXX should we use euid here?) | |
2499 | */ | |
2500 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2501 | { | |
2502 | char line[400]; | |
2503 | uid_t u; | |
2504 | gid_t g; | |
2505 | FILE *f; | |
2506 | ||
2507 | *uid = -1; | |
2508 | *gid = -1; | |
2509 | sprintf(line, "/proc/%d/status", pid); | |
2510 | if ((f = fopen(line, "r")) == NULL) { | |
b8defc3d | 2511 | lxcfs_error("Error opening %s: %s\n", line, strerror(errno)); |
237e200e SH |
2512 | return; |
2513 | } | |
2514 | while (fgets(line, 400, f)) { | |
2515 | if (strncmp(line, "Uid:", 4) == 0) { | |
2516 | if (sscanf(line+4, "%u", &u) != 1) { | |
b8defc3d | 2517 | lxcfs_error("bad uid line for pid %u\n", pid); |
237e200e SH |
2518 | fclose(f); |
2519 | return; | |
2520 | } | |
2521 | *uid = u; | |
2522 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2523 | if (sscanf(line+4, "%u", &g) != 1) { | |
b8defc3d | 2524 | lxcfs_error("bad gid line for pid %u\n", pid); |
237e200e SH |
2525 | fclose(f); |
2526 | return; | |
2527 | } | |
2528 | *gid = g; | |
2529 | } | |
2530 | } | |
2531 | fclose(f); | |
2532 | } | |
2533 | ||
2534 | /* | |
2535 | * May the requestor @r move victim @v to a new cgroup? | |
2536 | * This is allowed if | |
2537 | * . they are the same task | |
2538 | * . they are ownedy by the same uid | |
2539 | * . @r is root on the host, or | |
2540 | * . @v's uid is mapped into @r's where @r is root. | |
2541 | */ | |
2542 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2543 | { | |
2544 | uid_t v_uid, tmpuid; | |
2545 | gid_t v_gid; | |
2546 | ||
2547 | if (r == v) | |
2548 | return true; | |
2549 | if (r_uid == 0) | |
2550 | return true; | |
2551 | get_pid_creds(v, &v_uid, &v_gid); | |
2552 | if (r_uid == v_uid) | |
2553 | return true; | |
2554 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2555 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2556 | return true; | |
2557 | return false; | |
2558 | } | |
2559 | ||
2560 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2561 | const char *file, const char *buf) | |
2562 | { | |
2563 | int sock[2] = {-1, -1}; | |
2564 | pid_t qpid, cpid = -1; | |
2565 | FILE *pids_file = NULL; | |
2566 | bool answer = false, fail = false; | |
2567 | ||
2568 | pids_file = open_pids_file(contrl, cg); | |
2569 | if (!pids_file) | |
2570 | return false; | |
2571 | ||
2572 | /* | |
2573 | * write the pids to a socket, have helper in writer's pidns | |
2574 | * call movepid for us | |
2575 | */ | |
2576 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2577 | perror("socketpair"); | |
2578 | goto out; | |
2579 | } | |
2580 | ||
2581 | cpid = fork(); | |
2582 | if (cpid == -1) | |
2583 | goto out; | |
2584 | ||
2585 | if (!cpid) { // child | |
2586 | fclose(pids_file); | |
2587 | pid_from_ns_wrapper(sock[1], tpid); | |
2588 | } | |
2589 | ||
2590 | const char *ptr = buf; | |
2591 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
2592 | struct ucred cred; | |
2593 | char v; | |
2594 | ||
2595 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2596 | lxcfs_error("Error writing pid to child: %s.\n", strerror(errno)); |
237e200e SH |
2597 | goto out; |
2598 | } | |
2599 | ||
2600 | if (recv_creds(sock[0], &cred, &v)) { | |
2601 | if (v == '0') { | |
2602 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
2603 | fail = true; | |
2604 | break; | |
2605 | } | |
2606 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
2607 | fail = true; | |
2608 | } | |
2609 | } | |
2610 | ||
2611 | ptr = strchr(ptr, '\n'); | |
2612 | if (!ptr) | |
2613 | break; | |
2614 | ptr++; | |
2615 | } | |
2616 | ||
2617 | /* All good, write the value */ | |
2618 | qpid = -1; | |
2619 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
b8defc3d | 2620 | lxcfs_error("%s\n", "Warning: failed to ask child to exit."); |
237e200e SH |
2621 | |
2622 | if (!fail) | |
2623 | answer = true; | |
2624 | ||
2625 | out: | |
2626 | if (cpid != -1) | |
2627 | wait_for_pid(cpid); | |
2628 | if (sock[0] != -1) { | |
2629 | close(sock[0]); | |
2630 | close(sock[1]); | |
2631 | } | |
2632 | if (pids_file) { | |
2633 | if (fclose(pids_file) != 0) | |
2634 | answer = false; | |
2635 | } | |
2636 | return answer; | |
2637 | } | |
2638 | ||
2639 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
2640 | struct fuse_file_info *fi) | |
2641 | { | |
2642 | struct fuse_context *fc = fuse_get_context(); | |
2643 | char *localbuf = NULL; | |
2644 | struct cgfs_files *k = NULL; | |
2645 | struct file_info *f = (struct file_info *)fi->fh; | |
2646 | bool r; | |
2647 | ||
2648 | if (f->type != LXC_TYPE_CGFILE) { | |
b8defc3d | 2649 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_write."); |
237e200e SH |
2650 | return -EIO; |
2651 | } | |
2652 | ||
2653 | if (offset) | |
2654 | return 0; | |
2655 | ||
2656 | if (!fc) | |
2657 | return -EIO; | |
2658 | ||
2659 | localbuf = alloca(size+1); | |
2660 | localbuf[size] = '\0'; | |
2661 | memcpy(localbuf, buf, size); | |
2662 | ||
2663 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2664 | size = -EINVAL; | |
2665 | goto out; | |
2666 | } | |
2667 | ||
2668 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
2669 | size = -EACCES; | |
2670 | goto out; | |
2671 | } | |
2672 | ||
2673 | if (strcmp(f->file, "tasks") == 0 || | |
2674 | strcmp(f->file, "/tasks") == 0 || | |
2675 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2676 | strcmp(f->file, "cgroup.procs") == 0) | |
2677 | // special case - we have to translate the pids | |
2678 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
2679 | else | |
2680 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
2681 | ||
2682 | if (!r) | |
2683 | size = -EINVAL; | |
2684 | ||
2685 | out: | |
2686 | free_key(k); | |
2687 | return size; | |
2688 | } | |
2689 | ||
2690 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
2691 | { | |
2692 | struct fuse_context *fc = fuse_get_context(); | |
2693 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2694 | struct cgfs_files *k = NULL; | |
2695 | const char *cgroup; | |
2696 | int ret; | |
2697 | ||
2698 | if (!fc) | |
2699 | return -EIO; | |
2700 | ||
2701 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 2702 | return -EPERM; |
237e200e SH |
2703 | |
2704 | controller = pick_controller_from_path(fc, path); | |
2705 | if (!controller) | |
bc70ba9b CB |
2706 | return errno == ENOENT ? -EPERM : -errno; |
2707 | ||
237e200e SH |
2708 | cgroup = find_cgroup_in_path(path); |
2709 | if (!cgroup) | |
2710 | /* this is just /cgroup/controller */ | |
bc70ba9b | 2711 | return -EPERM; |
237e200e SH |
2712 | |
2713 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2714 | ||
2715 | if (!last) { | |
2716 | path1 = "/"; | |
2717 | path2 = cgdir; | |
2718 | } else { | |
2719 | path1 = cgdir; | |
2720 | path2 = last; | |
2721 | } | |
2722 | ||
2723 | if (is_child_cgroup(controller, path1, path2)) { | |
2724 | // get uid, gid, from '/tasks' file and make up a mode | |
2725 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2726 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2727 | ||
2728 | } else | |
2729 | k = cgfs_get_key(controller, path1, path2); | |
2730 | ||
2731 | if (!k) { | |
2732 | ret = -EINVAL; | |
2733 | goto out; | |
2734 | } | |
2735 | ||
2736 | /* | |
2737 | * This being a fuse request, the uid and gid must be valid | |
2738 | * in the caller's namespace. So we can just check to make | |
2739 | * sure that the caller is root in his uid, and privileged | |
2740 | * over the file's current owner. | |
2741 | */ | |
2742 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
2743 | ret = -EACCES; | |
2744 | goto out; | |
2745 | } | |
2746 | ||
2747 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
2748 | ||
2749 | out: | |
2750 | free_key(k); | |
2751 | free(cgdir); | |
2752 | ||
2753 | return ret; | |
2754 | } | |
2755 | ||
2756 | int cg_chmod(const char *path, mode_t mode) | |
2757 | { | |
2758 | struct fuse_context *fc = fuse_get_context(); | |
2759 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2760 | struct cgfs_files *k = NULL; | |
2761 | const char *cgroup; | |
2762 | int ret; | |
2763 | ||
2764 | if (!fc) | |
2765 | return -EIO; | |
2766 | ||
2767 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 2768 | return -EPERM; |
237e200e SH |
2769 | |
2770 | controller = pick_controller_from_path(fc, path); | |
2771 | if (!controller) | |
bc70ba9b CB |
2772 | return errno == ENOENT ? -EPERM : -errno; |
2773 | ||
237e200e SH |
2774 | cgroup = find_cgroup_in_path(path); |
2775 | if (!cgroup) | |
2776 | /* this is just /cgroup/controller */ | |
bc70ba9b | 2777 | return -EPERM; |
237e200e SH |
2778 | |
2779 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2780 | ||
2781 | if (!last) { | |
2782 | path1 = "/"; | |
2783 | path2 = cgdir; | |
2784 | } else { | |
2785 | path1 = cgdir; | |
2786 | path2 = last; | |
2787 | } | |
2788 | ||
2789 | if (is_child_cgroup(controller, path1, path2)) { | |
2790 | // get uid, gid, from '/tasks' file and make up a mode | |
2791 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2792 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2793 | ||
2794 | } else | |
2795 | k = cgfs_get_key(controller, path1, path2); | |
2796 | ||
2797 | if (!k) { | |
2798 | ret = -EINVAL; | |
2799 | goto out; | |
2800 | } | |
2801 | ||
2802 | /* | |
2803 | * This being a fuse request, the uid and gid must be valid | |
2804 | * in the caller's namespace. So we can just check to make | |
2805 | * sure that the caller is root in his uid, and privileged | |
2806 | * over the file's current owner. | |
2807 | */ | |
2808 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
2809 | ret = -EPERM; | |
2810 | goto out; | |
2811 | } | |
2812 | ||
2813 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
2814 | ret = -EINVAL; | |
2815 | goto out; | |
2816 | } | |
2817 | ||
2818 | ret = 0; | |
2819 | out: | |
2820 | free_key(k); | |
2821 | free(cgdir); | |
2822 | return ret; | |
2823 | } | |
2824 | ||
2825 | int cg_mkdir(const char *path, mode_t mode) | |
2826 | { | |
2827 | struct fuse_context *fc = fuse_get_context(); | |
2828 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
2829 | const char *cgroup; | |
2830 | int ret; | |
2831 | ||
2832 | if (!fc) | |
2833 | return -EIO; | |
2834 | ||
237e200e SH |
2835 | controller = pick_controller_from_path(fc, path); |
2836 | if (!controller) | |
2f7036d0 | 2837 | return errno == ENOENT ? -EPERM : -errno; |
237e200e SH |
2838 | |
2839 | cgroup = find_cgroup_in_path(path); | |
2840 | if (!cgroup) | |
bc70ba9b | 2841 | return -errno; |
237e200e SH |
2842 | |
2843 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2844 | if (!last) | |
2845 | path1 = "/"; | |
2846 | else | |
2847 | path1 = cgdir; | |
2848 | ||
2849 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2850 | if (initpid <= 0) | |
2851 | initpid = fc->pid; | |
2852 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
2853 | if (!next) | |
2854 | ret = -EINVAL; | |
2855 | else if (last && strcmp(next, last) == 0) | |
2856 | ret = -EEXIST; | |
2857 | else | |
2f7036d0 | 2858 | ret = -EPERM; |
237e200e SH |
2859 | goto out; |
2860 | } | |
2861 | ||
2862 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
2863 | ret = -EACCES; | |
2864 | goto out; | |
2865 | } | |
2866 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
2867 | ret = -EACCES; | |
2868 | goto out; | |
2869 | } | |
2870 | ||
2871 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
2872 | ||
2873 | out: | |
2874 | free(cgdir); | |
2875 | free(next); | |
2876 | return ret; | |
2877 | } | |
2878 | ||
2879 | int cg_rmdir(const char *path) | |
2880 | { | |
2881 | struct fuse_context *fc = fuse_get_context(); | |
2882 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
2883 | const char *cgroup; | |
2884 | int ret; | |
2885 | ||
2886 | if (!fc) | |
2887 | return -EIO; | |
2888 | ||
2889 | controller = pick_controller_from_path(fc, path); | |
e254948f CB |
2890 | if (!controller) /* Someone's trying to delete "/cgroup". */ |
2891 | return -EPERM; | |
237e200e SH |
2892 | |
2893 | cgroup = find_cgroup_in_path(path); | |
e254948f CB |
2894 | if (!cgroup) /* Someone's trying to delete a controller e.g. "/blkio". */ |
2895 | return -EPERM; | |
237e200e SH |
2896 | |
2897 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2898 | if (!last) { | |
e254948f CB |
2899 | /* Someone's trying to delete a cgroup on the same level as the |
2900 | * "/lxc" cgroup e.g. rmdir "/cgroup/blkio/lxc" or | |
2901 | * rmdir "/cgroup/blkio/init.slice". | |
2902 | */ | |
2903 | ret = -EPERM; | |
237e200e SH |
2904 | goto out; |
2905 | } | |
2906 | ||
2907 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2908 | if (initpid <= 0) | |
2909 | initpid = fc->pid; | |
2910 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
de77249b | 2911 | if (!last || (next && (strcmp(next, last) == 0))) |
237e200e SH |
2912 | ret = -EBUSY; |
2913 | else | |
2914 | ret = -ENOENT; | |
2915 | goto out; | |
2916 | } | |
2917 | ||
2918 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
2919 | ret = -EACCES; | |
2920 | goto out; | |
2921 | } | |
2922 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
2923 | ret = -EACCES; | |
2924 | goto out; | |
2925 | } | |
2926 | ||
2927 | if (!cgfs_remove(controller, cgroup)) { | |
2928 | ret = -EINVAL; | |
2929 | goto out; | |
2930 | } | |
2931 | ||
2932 | ret = 0; | |
2933 | ||
2934 | out: | |
2935 | free(cgdir); | |
2936 | free(next); | |
2937 | return ret; | |
2938 | } | |
2939 | ||
2940 | static bool startswith(const char *line, const char *pref) | |
2941 | { | |
2942 | if (strncmp(line, pref, strlen(pref)) == 0) | |
2943 | return true; | |
2944 | return false; | |
2945 | } | |
2946 | ||
c6095b08 SH |
2947 | static void parse_memstat(char *memstat, unsigned long *cached, |
2948 | unsigned long *active_anon, unsigned long *inactive_anon, | |
2949 | unsigned long *active_file, unsigned long *inactive_file, | |
2950 | unsigned long *unevictable) | |
237e200e SH |
2951 | { |
2952 | char *eol; | |
2953 | ||
237e200e | 2954 | while (*memstat) { |
c6095b08 | 2955 | if (startswith(memstat, "cache")) { |
3f5ba3a0 | 2956 | sscanf(memstat + 5, "%lu", cached); |
c6095b08 SH |
2957 | *cached /= 1024; |
2958 | } else if (startswith(memstat, "active_anon")) { | |
2959 | sscanf(memstat + 11, "%lu", active_anon); | |
2960 | *active_anon /= 1024; | |
2961 | } else if (startswith(memstat, "inactive_anon")) { | |
3f5ba3a0 | 2962 | sscanf(memstat + 13, "%lu", inactive_anon); |
c6095b08 SH |
2963 | *inactive_anon /= 1024; |
2964 | } else if (startswith(memstat, "active_file")) { | |
2965 | sscanf(memstat + 11, "%lu", active_file); | |
2966 | *active_file /= 1024; | |
2967 | } else if (startswith(memstat, "inactive_file")) { | |
3f5ba3a0 | 2968 | sscanf(memstat + 13, "%lu", inactive_file); |
c6095b08 SH |
2969 | *inactive_file /= 1024; |
2970 | } else if (startswith(memstat, "unevictable")) { | |
2971 | sscanf(memstat + 11, "%lu", unevictable); | |
2972 | *unevictable /= 1024; | |
237e200e SH |
2973 | } |
2974 | eol = strchr(memstat, '\n'); | |
2975 | if (!eol) | |
2976 | return; | |
2977 | memstat = eol+1; | |
2978 | } | |
2979 | } | |
2980 | ||
2981 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
2982 | { | |
2983 | char *eol; | |
2984 | char key[32]; | |
2985 | ||
2986 | memset(key, 0, 32); | |
2987 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
2988 | ||
2989 | size_t len = strlen(key); | |
2990 | *v = 0; | |
2991 | ||
2992 | while (*str) { | |
2993 | if (startswith(str, key)) { | |
2994 | sscanf(str + len, "%lu", v); | |
2995 | return; | |
2996 | } | |
2997 | eol = strchr(str, '\n'); | |
2998 | if (!eol) | |
2999 | return; | |
3000 | str = eol+1; | |
3001 | } | |
3002 | } | |
3003 | ||
3004 | static int read_file(const char *path, char *buf, size_t size, | |
3005 | struct file_info *d) | |
3006 | { | |
3007 | size_t linelen = 0, total_len = 0, rv = 0; | |
3008 | char *line = NULL; | |
3009 | char *cache = d->buf; | |
3010 | size_t cache_size = d->buflen; | |
3011 | FILE *f = fopen(path, "r"); | |
3012 | if (!f) | |
3013 | return 0; | |
3014 | ||
3015 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3016 | ssize_t l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3017 | if (l < 0) { |
3018 | perror("Error writing to cache"); | |
3019 | rv = 0; | |
3020 | goto err; | |
3021 | } | |
3022 | if (l >= cache_size) { | |
b8defc3d | 3023 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3024 | rv = 0; |
3025 | goto err; | |
3026 | } | |
3027 | cache += l; | |
3028 | cache_size -= l; | |
3029 | total_len += l; | |
3030 | } | |
3031 | ||
3032 | d->size = total_len; | |
a262ddb7 CB |
3033 | if (total_len > size) |
3034 | total_len = size; | |
237e200e SH |
3035 | |
3036 | /* read from off 0 */ | |
3037 | memcpy(buf, d->buf, total_len); | |
3038 | rv = total_len; | |
3039 | err: | |
3040 | fclose(f); | |
3041 | free(line); | |
3042 | return rv; | |
3043 | } | |
3044 | ||
3045 | /* | |
3046 | * FUSE ops for /proc | |
3047 | */ | |
3048 | ||
018246ff | 3049 | static unsigned long get_memlimit(const char *cgroup, const char *file) |
237e200e SH |
3050 | { |
3051 | char *memlimit_str = NULL; | |
3052 | unsigned long memlimit = -1; | |
3053 | ||
018246ff | 3054 | if (cgfs_get_value("memory", cgroup, file, &memlimit_str)) |
237e200e SH |
3055 | memlimit = strtoul(memlimit_str, NULL, 10); |
3056 | ||
3057 | free(memlimit_str); | |
3058 | ||
3059 | return memlimit; | |
3060 | } | |
3061 | ||
018246ff | 3062 | static unsigned long get_min_memlimit(const char *cgroup, const char *file) |
237e200e SH |
3063 | { |
3064 | char *copy = strdupa(cgroup); | |
3065 | unsigned long memlimit = 0, retlimit; | |
3066 | ||
018246ff | 3067 | retlimit = get_memlimit(copy, file); |
237e200e SH |
3068 | |
3069 | while (strcmp(copy, "/") != 0) { | |
3070 | copy = dirname(copy); | |
018246ff | 3071 | memlimit = get_memlimit(copy, file); |
237e200e SH |
3072 | if (memlimit != -1 && memlimit < retlimit) |
3073 | retlimit = memlimit; | |
3074 | }; | |
3075 | ||
3076 | return retlimit; | |
3077 | } | |
3078 | ||
3079 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
3080 | struct fuse_file_info *fi) | |
3081 | { | |
3082 | struct fuse_context *fc = fuse_get_context(); | |
3083 | struct file_info *d = (struct file_info *)fi->fh; | |
3084 | char *cg; | |
3085 | char *memusage_str = NULL, *memstat_str = NULL, | |
018246ff | 3086 | *memswlimit_str = NULL, *memswusage_str = NULL; |
237e200e | 3087 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0, |
c6095b08 | 3088 | cached = 0, hosttotal = 0, active_anon = 0, inactive_anon = 0, |
594a10e6 WB |
3089 | active_file = 0, inactive_file = 0, unevictable = 0, |
3090 | hostswtotal = 0; | |
237e200e SH |
3091 | char *line = NULL; |
3092 | size_t linelen = 0, total_len = 0, rv = 0; | |
3093 | char *cache = d->buf; | |
3094 | size_t cache_size = d->buflen; | |
3095 | FILE *f = NULL; | |
3096 | ||
3097 | if (offset){ | |
3098 | if (offset > d->size) | |
3099 | return -EINVAL; | |
3100 | if (!d->cached) | |
3101 | return 0; | |
3102 | int left = d->size - offset; | |
3103 | total_len = left > size ? size: left; | |
3104 | memcpy(buf, cache + offset, total_len); | |
3105 | return total_len; | |
3106 | } | |
3107 | ||
3108 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3109 | if (initpid <= 0) | |
3110 | initpid = fc->pid; | |
3111 | cg = get_pid_cgroup(initpid, "memory"); | |
3112 | if (!cg) | |
3113 | return read_file("/proc/meminfo", buf, size, d); | |
6d2f6996 | 3114 | prune_init_slice(cg); |
237e200e | 3115 | |
018246ff | 3116 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); |
237e200e SH |
3117 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) |
3118 | goto err; | |
3119 | if (!cgfs_get_value("memory", cg, "memory.stat", &memstat_str)) | |
3120 | goto err; | |
3121 | ||
3122 | // Following values are allowed to fail, because swapaccount might be turned | |
3123 | // off for current kernel | |
3124 | if(cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str) && | |
3125 | cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str)) | |
3126 | { | |
018246ff | 3127 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); |
237e200e SH |
3128 | memswusage = strtoul(memswusage_str, NULL, 10); |
3129 | ||
237e200e SH |
3130 | memswlimit = memswlimit / 1024; |
3131 | memswusage = memswusage / 1024; | |
3132 | } | |
3133 | ||
3134 | memusage = strtoul(memusage_str, NULL, 10); | |
3135 | memlimit /= 1024; | |
3136 | memusage /= 1024; | |
3137 | ||
c6095b08 SH |
3138 | parse_memstat(memstat_str, &cached, &active_anon, |
3139 | &inactive_anon, &active_file, &inactive_file, | |
3140 | &unevictable); | |
237e200e SH |
3141 | |
3142 | f = fopen("/proc/meminfo", "r"); | |
3143 | if (!f) | |
3144 | goto err; | |
3145 | ||
3146 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3147 | ssize_t l; |
237e200e SH |
3148 | char *printme, lbuf[100]; |
3149 | ||
3150 | memset(lbuf, 0, 100); | |
3151 | if (startswith(line, "MemTotal:")) { | |
594a10e6 | 3152 | sscanf(line+sizeof("MemTotal:")-1, "%lu", &hosttotal); |
237e200e SH |
3153 | if (hosttotal < memlimit) |
3154 | memlimit = hosttotal; | |
3155 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
3156 | printme = lbuf; | |
3157 | } else if (startswith(line, "MemFree:")) { | |
3158 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
3159 | printme = lbuf; | |
3160 | } else if (startswith(line, "MemAvailable:")) { | |
3161 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage); | |
3162 | printme = lbuf; | |
3163 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0) { | |
594a10e6 WB |
3164 | sscanf(line+sizeof("SwapTotal:")-1, "%lu", &hostswtotal); |
3165 | if (hostswtotal < memswlimit - memlimit) | |
3166 | memswlimit = hostswtotal + memlimit; | |
237e200e SH |
3167 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit - memlimit); |
3168 | printme = lbuf; | |
3169 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0) { | |
b4665ce0 SH |
3170 | unsigned long swaptotal = memswlimit - memlimit, |
3171 | swapusage = memswusage - memusage, | |
3172 | swapfree = swapusage < swaptotal ? swaptotal - swapusage : 0; | |
3173 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree); | |
237e200e | 3174 | printme = lbuf; |
da35d72a SH |
3175 | } else if (startswith(line, "Slab:")) { |
3176 | snprintf(lbuf, 100, "Slab: %8lu kB\n", 0UL); | |
3177 | printme = lbuf; | |
237e200e SH |
3178 | } else if (startswith(line, "Buffers:")) { |
3179 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
3180 | printme = lbuf; | |
3181 | } else if (startswith(line, "Cached:")) { | |
3182 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
3183 | printme = lbuf; | |
3184 | } else if (startswith(line, "SwapCached:")) { | |
3185 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
3186 | printme = lbuf; | |
2f306ad3 | 3187 | } else if (startswith(line, "Active:")) { |
c6095b08 SH |
3188 | snprintf(lbuf, 100, "Active: %8lu kB\n", |
3189 | active_anon + active_file); | |
3190 | printme = lbuf; | |
2f306ad3 | 3191 | } else if (startswith(line, "Inactive:")) { |
c6095b08 SH |
3192 | snprintf(lbuf, 100, "Inactive: %8lu kB\n", |
3193 | inactive_anon + inactive_file); | |
3194 | printme = lbuf; | |
3195 | } else if (startswith(line, "Active(anon)")) { | |
3196 | snprintf(lbuf, 100, "Active(anon): %8lu kB\n", active_anon); | |
3197 | printme = lbuf; | |
3198 | } else if (startswith(line, "Inactive(anon)")) { | |
3199 | snprintf(lbuf, 100, "Inactive(anon): %8lu kB\n", inactive_anon); | |
3200 | printme = lbuf; | |
3201 | } else if (startswith(line, "Active(file)")) { | |
3202 | snprintf(lbuf, 100, "Active(file): %8lu kB\n", active_file); | |
3203 | printme = lbuf; | |
3204 | } else if (startswith(line, "Inactive(file)")) { | |
3205 | snprintf(lbuf, 100, "Inactive(file): %8lu kB\n", inactive_file); | |
3206 | printme = lbuf; | |
3207 | } else if (startswith(line, "Unevictable")) { | |
3208 | snprintf(lbuf, 100, "Unevictable: %8lu kB\n", unevictable); | |
3209 | printme = lbuf; | |
3210 | } else if (startswith(line, "SReclaimable")) { | |
3211 | snprintf(lbuf, 100, "SReclaimable: %8lu kB\n", 0UL); | |
3212 | printme = lbuf; | |
3213 | } else if (startswith(line, "SUnreclaim")) { | |
3214 | snprintf(lbuf, 100, "SUnreclaim: %8lu kB\n", 0UL); | |
3215 | printme = lbuf; | |
237e200e SH |
3216 | } else |
3217 | printme = line; | |
3218 | ||
3219 | l = snprintf(cache, cache_size, "%s", printme); | |
3220 | if (l < 0) { | |
3221 | perror("Error writing to cache"); | |
3222 | rv = 0; | |
3223 | goto err; | |
3224 | ||
3225 | } | |
3226 | if (l >= cache_size) { | |
b8defc3d | 3227 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3228 | rv = 0; |
3229 | goto err; | |
3230 | } | |
3231 | ||
3232 | cache += l; | |
3233 | cache_size -= l; | |
3234 | total_len += l; | |
3235 | } | |
3236 | ||
3237 | d->cached = 1; | |
3238 | d->size = total_len; | |
3239 | if (total_len > size ) total_len = size; | |
3240 | memcpy(buf, d->buf, total_len); | |
3241 | ||
3242 | rv = total_len; | |
3243 | err: | |
3244 | if (f) | |
3245 | fclose(f); | |
3246 | free(line); | |
3247 | free(cg); | |
3248 | free(memusage_str); | |
3249 | free(memswlimit_str); | |
3250 | free(memswusage_str); | |
3251 | free(memstat_str); | |
237e200e SH |
3252 | return rv; |
3253 | } | |
3254 | ||
3255 | /* | |
3256 | * Read the cpuset.cpus for cg | |
3257 | * Return the answer in a newly allocated string which must be freed | |
3258 | */ | |
3259 | static char *get_cpuset(const char *cg) | |
3260 | { | |
3261 | char *answer; | |
3262 | ||
3263 | if (!cgfs_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
3264 | return NULL; | |
3265 | return answer; | |
3266 | } | |
3267 | ||
3268 | bool cpu_in_cpuset(int cpu, const char *cpuset); | |
3269 | ||
3270 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) | |
3271 | { | |
3272 | int cpu; | |
3273 | ||
3274 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
3275 | return false; | |
3276 | return cpu_in_cpuset(cpu, cpuset); | |
3277 | } | |
3278 | ||
3279 | /* | |
3280 | * check whether this is a '^processor" line in /proc/cpuinfo | |
3281 | */ | |
3282 | static bool is_processor_line(const char *line) | |
3283 | { | |
3284 | int cpu; | |
3285 | ||
3286 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
3287 | return true; | |
3288 | return false; | |
3289 | } | |
3290 | ||
3291 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
3292 | struct fuse_file_info *fi) | |
3293 | { | |
3294 | struct fuse_context *fc = fuse_get_context(); | |
3295 | struct file_info *d = (struct file_info *)fi->fh; | |
3296 | char *cg; | |
3297 | char *cpuset = NULL; | |
3298 | char *line = NULL; | |
3299 | size_t linelen = 0, total_len = 0, rv = 0; | |
f676eb79 SH |
3300 | bool am_printing = false, firstline = true, is_s390x = false; |
3301 | int curcpu = -1, cpu; | |
237e200e SH |
3302 | char *cache = d->buf; |
3303 | size_t cache_size = d->buflen; | |
3304 | FILE *f = NULL; | |
3305 | ||
3306 | if (offset){ | |
3307 | if (offset > d->size) | |
3308 | return -EINVAL; | |
3309 | if (!d->cached) | |
3310 | return 0; | |
3311 | int left = d->size - offset; | |
3312 | total_len = left > size ? size: left; | |
3313 | memcpy(buf, cache + offset, total_len); | |
3314 | return total_len; | |
3315 | } | |
3316 | ||
3317 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3318 | if (initpid <= 0) | |
3319 | initpid = fc->pid; | |
3320 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3321 | if (!cg) | |
3322 | return read_file("proc/cpuinfo", buf, size, d); | |
6d2f6996 | 3323 | prune_init_slice(cg); |
237e200e SH |
3324 | |
3325 | cpuset = get_cpuset(cg); | |
3326 | if (!cpuset) | |
3327 | goto err; | |
3328 | ||
3329 | f = fopen("/proc/cpuinfo", "r"); | |
3330 | if (!f) | |
3331 | goto err; | |
3332 | ||
3333 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3334 | ssize_t l; |
f676eb79 SH |
3335 | if (firstline) { |
3336 | firstline = false; | |
3337 | if (strstr(line, "IBM/S390") != NULL) { | |
3338 | is_s390x = true; | |
3339 | am_printing = true; | |
5ed9d4e2 | 3340 | continue; |
f676eb79 SH |
3341 | } |
3342 | } | |
5ed9d4e2 SH |
3343 | if (strncmp(line, "# processors:", 12) == 0) |
3344 | continue; | |
237e200e SH |
3345 | if (is_processor_line(line)) { |
3346 | am_printing = cpuline_in_cpuset(line, cpuset); | |
3347 | if (am_printing) { | |
3348 | curcpu ++; | |
3349 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); | |
3350 | if (l < 0) { | |
3351 | perror("Error writing to cache"); | |
3352 | rv = 0; | |
3353 | goto err; | |
3354 | } | |
3355 | if (l >= cache_size) { | |
b8defc3d | 3356 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3357 | rv = 0; |
3358 | goto err; | |
3359 | } | |
3360 | cache += l; | |
3361 | cache_size -= l; | |
3362 | total_len += l; | |
3363 | } | |
3364 | continue; | |
f676eb79 SH |
3365 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { |
3366 | char *p; | |
3367 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3368 | continue; | |
3369 | curcpu ++; | |
3370 | p = strchr(line, ':'); | |
3371 | if (!p || !*p) | |
3372 | goto err; | |
3373 | p++; | |
5ed9d4e2 | 3374 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
f676eb79 SH |
3375 | if (l < 0) { |
3376 | perror("Error writing to cache"); | |
3377 | rv = 0; | |
3378 | goto err; | |
3379 | } | |
3380 | if (l >= cache_size) { | |
b8defc3d | 3381 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
f676eb79 SH |
3382 | rv = 0; |
3383 | goto err; | |
3384 | } | |
3385 | cache += l; | |
3386 | cache_size -= l; | |
3387 | total_len += l; | |
3388 | continue; | |
3389 | ||
237e200e SH |
3390 | } |
3391 | if (am_printing) { | |
3392 | l = snprintf(cache, cache_size, "%s", line); | |
3393 | if (l < 0) { | |
3394 | perror("Error writing to cache"); | |
3395 | rv = 0; | |
3396 | goto err; | |
3397 | } | |
3398 | if (l >= cache_size) { | |
b8defc3d | 3399 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3400 | rv = 0; |
3401 | goto err; | |
3402 | } | |
3403 | cache += l; | |
3404 | cache_size -= l; | |
3405 | total_len += l; | |
3406 | } | |
3407 | } | |
3408 | ||
5ed9d4e2 SH |
3409 | if (is_s390x) { |
3410 | char *origcache = d->buf; | |
a262ddb7 | 3411 | ssize_t l; |
5ed9d4e2 SH |
3412 | do { |
3413 | d->buf = malloc(d->buflen); | |
3414 | } while (!d->buf); | |
3415 | cache = d->buf; | |
3416 | cache_size = d->buflen; | |
3417 | total_len = 0; | |
3418 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
3419 | if (l < 0 || l >= cache_size) { | |
3420 | free(origcache); | |
3421 | goto err; | |
3422 | } | |
3423 | cache_size -= l; | |
3424 | cache += l; | |
3425 | total_len += l; | |
3426 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
3427 | if (l < 0 || l >= cache_size) { | |
3428 | free(origcache); | |
3429 | goto err; | |
3430 | } | |
3431 | cache_size -= l; | |
3432 | cache += l; | |
3433 | total_len += l; | |
3434 | l = snprintf(cache, cache_size, "%s", origcache); | |
3435 | free(origcache); | |
3436 | if (l < 0 || l >= cache_size) | |
3437 | goto err; | |
3438 | total_len += l; | |
3439 | } | |
3440 | ||
237e200e SH |
3441 | d->cached = 1; |
3442 | d->size = total_len; | |
3443 | if (total_len > size ) total_len = size; | |
3444 | ||
3445 | /* read from off 0 */ | |
3446 | memcpy(buf, d->buf, total_len); | |
3447 | rv = total_len; | |
3448 | err: | |
3449 | if (f) | |
3450 | fclose(f); | |
3451 | free(line); | |
3452 | free(cpuset); | |
3453 | free(cg); | |
3454 | return rv; | |
3455 | } | |
3456 | ||
9ac264cf JB |
3457 | static long int getreaperctime(pid_t pid) |
3458 | { | |
3459 | char fnam[100]; | |
3460 | struct stat sb; | |
3461 | int ret; | |
3462 | pid_t qpid; | |
3463 | ||
3464 | qpid = lookup_initpid_in_store(pid); | |
3465 | if (qpid <= 0) | |
3466 | return 0; | |
3467 | ||
3468 | ret = snprintf(fnam, 100, "/proc/%d", qpid); | |
3469 | if (ret < 0 || ret >= 100) | |
3470 | return 0; | |
3471 | ||
3472 | if (lstat(fnam, &sb) < 0) | |
3473 | return 0; | |
3474 | ||
3475 | return sb.st_ctime; | |
3476 | } | |
3477 | ||
f34de69a | 3478 | #define CPUALL_MAX_SIZE (BUF_RESERVE_SIZE / 2) |
237e200e SH |
3479 | static int proc_stat_read(char *buf, size_t size, off_t offset, |
3480 | struct fuse_file_info *fi) | |
3481 | { | |
3482 | struct fuse_context *fc = fuse_get_context(); | |
3483 | struct file_info *d = (struct file_info *)fi->fh; | |
3484 | char *cg; | |
3485 | char *cpuset = NULL; | |
3486 | char *line = NULL; | |
3487 | size_t linelen = 0, total_len = 0, rv = 0; | |
3488 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
7144f069 | 3489 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0; |
237e200e | 3490 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, |
7144f069 | 3491 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0, guest_nice_sum = 0; |
237e200e SH |
3492 | char cpuall[CPUALL_MAX_SIZE]; |
3493 | /* reserve for cpu all */ | |
3494 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
3495 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
3496 | FILE *f = NULL; | |
3497 | ||
3498 | if (offset){ | |
3499 | if (offset > d->size) | |
3500 | return -EINVAL; | |
3501 | if (!d->cached) | |
3502 | return 0; | |
3503 | int left = d->size - offset; | |
3504 | total_len = left > size ? size: left; | |
3505 | memcpy(buf, d->buf + offset, total_len); | |
3506 | return total_len; | |
3507 | } | |
3508 | ||
3509 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3510 | if (initpid <= 0) | |
3511 | initpid = fc->pid; | |
3512 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3513 | if (!cg) | |
3514 | return read_file("/proc/stat", buf, size, d); | |
6d2f6996 | 3515 | prune_init_slice(cg); |
237e200e SH |
3516 | |
3517 | cpuset = get_cpuset(cg); | |
3518 | if (!cpuset) | |
3519 | goto err; | |
3520 | ||
3521 | f = fopen("/proc/stat", "r"); | |
3522 | if (!f) | |
3523 | goto err; | |
3524 | ||
3525 | //skip first line | |
3526 | if (getline(&line, &linelen, f) < 0) { | |
b8defc3d | 3527 | lxcfs_error("%s\n", "proc_stat_read read first line failed."); |
237e200e SH |
3528 | goto err; |
3529 | } | |
3530 | ||
3531 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3532 | ssize_t l; |
237e200e SH |
3533 | int cpu; |
3534 | char cpu_char[10]; /* That's a lot of cores */ | |
3535 | char *c; | |
3536 | ||
b4665ce0 SH |
3537 | if (strlen(line) == 0) |
3538 | continue; | |
237e200e SH |
3539 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { |
3540 | /* not a ^cpuN line containing a number N, just print it */ | |
9ac264cf JB |
3541 | if (strncmp(line, "btime", 5) == 0) |
3542 | l = snprintf(cache, cache_size, "btime %ld\n", getreaperctime(fc->pid)); | |
3543 | else | |
3544 | l = snprintf(cache, cache_size, "%s", line); | |
237e200e SH |
3545 | if (l < 0) { |
3546 | perror("Error writing to cache"); | |
3547 | rv = 0; | |
3548 | goto err; | |
3549 | } | |
3550 | if (l >= cache_size) { | |
b8defc3d | 3551 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3552 | rv = 0; |
3553 | goto err; | |
3554 | } | |
3555 | cache += l; | |
3556 | cache_size -= l; | |
3557 | total_len += l; | |
3558 | continue; | |
3559 | } | |
3560 | ||
3561 | if (sscanf(cpu_char, "%d", &cpu) != 1) | |
3562 | continue; | |
3563 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3564 | continue; | |
3565 | curcpu ++; | |
3566 | ||
3567 | c = strchr(line, ' '); | |
3568 | if (!c) | |
3569 | continue; | |
3570 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
3571 | if (l < 0) { | |
3572 | perror("Error writing to cache"); | |
3573 | rv = 0; | |
3574 | goto err; | |
3575 | ||
3576 | } | |
3577 | if (l >= cache_size) { | |
b8defc3d | 3578 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3579 | rv = 0; |
3580 | goto err; | |
3581 | } | |
3582 | ||
3583 | cache += l; | |
3584 | cache_size -= l; | |
3585 | total_len += l; | |
3586 | ||
7144f069 CB |
3587 | if (sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", |
3588 | &user, | |
3589 | &nice, | |
3590 | &system, | |
3591 | &idle, | |
3592 | &iowait, | |
3593 | &irq, | |
3594 | &softirq, | |
3595 | &steal, | |
3596 | &guest, | |
3597 | &guest_nice) != 10) | |
237e200e SH |
3598 | continue; |
3599 | user_sum += user; | |
3600 | nice_sum += nice; | |
3601 | system_sum += system; | |
3602 | idle_sum += idle; | |
3603 | iowait_sum += iowait; | |
3604 | irq_sum += irq; | |
3605 | softirq_sum += softirq; | |
3606 | steal_sum += steal; | |
3607 | guest_sum += guest; | |
7144f069 | 3608 | guest_nice_sum += guest_nice; |
237e200e SH |
3609 | } |
3610 | ||
3611 | cache = d->buf; | |
3612 | ||
7144f069 CB |
3613 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "cpu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", |
3614 | user_sum, | |
3615 | nice_sum, | |
3616 | system_sum, | |
3617 | idle_sum, | |
3618 | iowait_sum, | |
3619 | irq_sum, | |
3620 | softirq_sum, | |
3621 | steal_sum, | |
3622 | guest_sum, | |
3623 | guest_nice_sum); | |
3624 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE) { | |
237e200e SH |
3625 | memcpy(cache, cpuall, cpuall_len); |
3626 | cache += cpuall_len; | |
7144f069 | 3627 | } else { |
237e200e | 3628 | /* shouldn't happen */ |
b8defc3d | 3629 | lxcfs_error("proc_stat_read copy cpuall failed, cpuall_len=%d.", cpuall_len); |
237e200e SH |
3630 | cpuall_len = 0; |
3631 | } | |
3632 | ||
3633 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
3634 | total_len += cpuall_len; | |
3635 | d->cached = 1; | |
3636 | d->size = total_len; | |
7144f069 CB |
3637 | if (total_len > size) |
3638 | total_len = size; | |
237e200e SH |
3639 | |
3640 | memcpy(buf, d->buf, total_len); | |
3641 | rv = total_len; | |
3642 | ||
3643 | err: | |
3644 | if (f) | |
3645 | fclose(f); | |
3646 | free(line); | |
3647 | free(cpuset); | |
3648 | free(cg); | |
3649 | return rv; | |
3650 | } | |
3651 | ||
3652 | static long int getreaperage(pid_t pid) | |
3653 | { | |
9ac264cf | 3654 | long int ctime; |
237e200e | 3655 | |
9ac264cf JB |
3656 | ctime = getreaperctime(pid); |
3657 | if (ctime) | |
3658 | return time(NULL) - ctime; | |
3659 | return ctime; | |
237e200e SH |
3660 | } |
3661 | ||
3662 | static unsigned long get_reaper_busy(pid_t task) | |
3663 | { | |
3664 | pid_t initpid = lookup_initpid_in_store(task); | |
3665 | char *cgroup = NULL, *usage_str = NULL; | |
3666 | unsigned long usage = 0; | |
3667 | ||
3668 | if (initpid <= 0) | |
3669 | return 0; | |
3670 | ||
3671 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
3672 | if (!cgroup) | |
3673 | goto out; | |
6d2f6996 | 3674 | prune_init_slice(cgroup); |
237e200e SH |
3675 | if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str)) |
3676 | goto out; | |
3677 | usage = strtoul(usage_str, NULL, 10); | |
3678 | usage /= 1000000000; | |
3679 | ||
3680 | out: | |
3681 | free(cgroup); | |
3682 | free(usage_str); | |
3683 | return usage; | |
3684 | } | |
3685 | ||
3686 | #if RELOADTEST | |
3687 | void iwashere(void) | |
3688 | { | |
237e200e SH |
3689 | int fd; |
3690 | ||
ec2b5e7c | 3691 | fd = creat("/tmp/lxcfs-iwashere", 0644); |
237e200e SH |
3692 | if (fd >= 0) |
3693 | close(fd); | |
3694 | } | |
3695 | #endif | |
3696 | ||
3697 | /* | |
3698 | * We read /proc/uptime and reuse its second field. | |
3699 | * For the first field, we use the mtime for the reaper for | |
3700 | * the calling pid as returned by getreaperage | |
3701 | */ | |
3702 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
3703 | struct fuse_file_info *fi) | |
3704 | { | |
3705 | struct fuse_context *fc = fuse_get_context(); | |
3706 | struct file_info *d = (struct file_info *)fi->fh; | |
3707 | long int reaperage = getreaperage(fc->pid); | |
3708 | unsigned long int busytime = get_reaper_busy(fc->pid), idletime; | |
3709 | char *cache = d->buf; | |
a262ddb7 | 3710 | ssize_t total_len = 0; |
237e200e SH |
3711 | |
3712 | #if RELOADTEST | |
3713 | iwashere(); | |
3714 | #endif | |
3715 | ||
3716 | if (offset){ | |
3717 | if (offset > d->size) | |
3718 | return -EINVAL; | |
3719 | if (!d->cached) | |
3720 | return 0; | |
3721 | int left = d->size - offset; | |
3722 | total_len = left > size ? size: left; | |
3723 | memcpy(buf, cache + offset, total_len); | |
3724 | return total_len; | |
3725 | } | |
3726 | ||
3727 | idletime = reaperage - busytime; | |
3728 | if (idletime > reaperage) | |
3729 | idletime = reaperage; | |
3730 | ||
3731 | total_len = snprintf(d->buf, d->size, "%ld.0 %lu.0\n", reaperage, idletime); | |
3732 | if (total_len < 0){ | |
3733 | perror("Error writing to cache"); | |
3734 | return 0; | |
3735 | } | |
3736 | ||
3737 | d->size = (int)total_len; | |
3738 | d->cached = 1; | |
3739 | ||
3740 | if (total_len > size) total_len = size; | |
3741 | ||
3742 | memcpy(buf, d->buf, total_len); | |
3743 | return total_len; | |
3744 | } | |
3745 | ||
3746 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
3747 | struct fuse_file_info *fi) | |
3748 | { | |
3749 | char dev_name[72]; | |
3750 | struct fuse_context *fc = fuse_get_context(); | |
3751 | struct file_info *d = (struct file_info *)fi->fh; | |
3752 | char *cg; | |
3753 | char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
3754 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
3755 | unsigned long read = 0, write = 0; | |
3756 | unsigned long read_merged = 0, write_merged = 0; | |
3757 | unsigned long read_sectors = 0, write_sectors = 0; | |
3758 | unsigned long read_ticks = 0, write_ticks = 0; | |
3759 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
3760 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
3761 | char *cache = d->buf; | |
3762 | size_t cache_size = d->buflen; | |
3763 | char *line = NULL; | |
3764 | size_t linelen = 0, total_len = 0, rv = 0; | |
3765 | unsigned int major = 0, minor = 0; | |
3766 | int i = 0; | |
3767 | FILE *f = NULL; | |
3768 | ||
3769 | if (offset){ | |
3770 | if (offset > d->size) | |
3771 | return -EINVAL; | |
3772 | if (!d->cached) | |
3773 | return 0; | |
3774 | int left = d->size - offset; | |
3775 | total_len = left > size ? size: left; | |
3776 | memcpy(buf, cache + offset, total_len); | |
3777 | return total_len; | |
3778 | } | |
3779 | ||
3780 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3781 | if (initpid <= 0) | |
3782 | initpid = fc->pid; | |
3783 | cg = get_pid_cgroup(initpid, "blkio"); | |
3784 | if (!cg) | |
3785 | return read_file("/proc/diskstats", buf, size, d); | |
6d2f6996 | 3786 | prune_init_slice(cg); |
237e200e | 3787 | |
2209fe50 | 3788 | if (!cgfs_get_value("blkio", cg, "blkio.io_serviced_recursive", &io_serviced_str)) |
237e200e | 3789 | goto err; |
2209fe50 | 3790 | if (!cgfs_get_value("blkio", cg, "blkio.io_merged_recursive", &io_merged_str)) |
237e200e | 3791 | goto err; |
2209fe50 | 3792 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_bytes_recursive", &io_service_bytes_str)) |
237e200e | 3793 | goto err; |
2209fe50 | 3794 | if (!cgfs_get_value("blkio", cg, "blkio.io_wait_time_recursive", &io_wait_time_str)) |
237e200e | 3795 | goto err; |
2209fe50 | 3796 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_time_recursive", &io_service_time_str)) |
237e200e SH |
3797 | goto err; |
3798 | ||
3799 | ||
3800 | f = fopen("/proc/diskstats", "r"); | |
3801 | if (!f) | |
3802 | goto err; | |
3803 | ||
3804 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3805 | ssize_t l; |
2209fe50 | 3806 | char lbuf[256]; |
237e200e SH |
3807 | |
3808 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
2209fe50 | 3809 | if (i != 3) |
237e200e | 3810 | continue; |
2209fe50 SH |
3811 | |
3812 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
3813 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
3814 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
3815 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
3816 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
3817 | read_sectors = read_sectors/512; | |
3818 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
3819 | write_sectors = write_sectors/512; | |
3820 | ||
3821 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
3822 | rd_svctm = rd_svctm/1000000; | |
3823 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
3824 | rd_wait = rd_wait/1000000; | |
3825 | read_ticks = rd_svctm + rd_wait; | |
3826 | ||
3827 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
3828 | wr_svctm = wr_svctm/1000000; | |
3829 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
3830 | wr_wait = wr_wait/1000000; | |
3831 | write_ticks = wr_svctm + wr_wait; | |
3832 | ||
3833 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
3834 | tot_ticks = tot_ticks/1000000; | |
237e200e SH |
3835 | |
3836 | memset(lbuf, 0, 256); | |
2db31eb6 SH |
3837 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) |
3838 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3839 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
3840 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
3841 | else | |
3842 | continue; | |
237e200e | 3843 | |
2209fe50 | 3844 | l = snprintf(cache, cache_size, "%s", lbuf); |
237e200e SH |
3845 | if (l < 0) { |
3846 | perror("Error writing to fuse buf"); | |
3847 | rv = 0; | |
3848 | goto err; | |
3849 | } | |
3850 | if (l >= cache_size) { | |
b8defc3d | 3851 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3852 | rv = 0; |
3853 | goto err; | |
3854 | } | |
3855 | cache += l; | |
3856 | cache_size -= l; | |
3857 | total_len += l; | |
3858 | } | |
3859 | ||
3860 | d->cached = 1; | |
3861 | d->size = total_len; | |
3862 | if (total_len > size ) total_len = size; | |
3863 | memcpy(buf, d->buf, total_len); | |
3864 | ||
3865 | rv = total_len; | |
3866 | err: | |
3867 | free(cg); | |
3868 | if (f) | |
3869 | fclose(f); | |
3870 | free(line); | |
3871 | free(io_serviced_str); | |
3872 | free(io_merged_str); | |
3873 | free(io_service_bytes_str); | |
3874 | free(io_wait_time_str); | |
3875 | free(io_service_time_str); | |
3876 | return rv; | |
3877 | } | |
3878 | ||
70dcc12e SH |
3879 | static int proc_swaps_read(char *buf, size_t size, off_t offset, |
3880 | struct fuse_file_info *fi) | |
3881 | { | |
3882 | struct fuse_context *fc = fuse_get_context(); | |
3883 | struct file_info *d = (struct file_info *)fi->fh; | |
3884 | char *cg = NULL; | |
018246ff | 3885 | char *memswlimit_str = NULL, *memlimit_str = NULL, *memusage_str = NULL, *memswusage_str = NULL; |
70dcc12e | 3886 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0, swap_total = 0, swap_free = 0; |
a262ddb7 CB |
3887 | ssize_t total_len = 0, rv = 0; |
3888 | ssize_t l = 0; | |
70dcc12e SH |
3889 | char *cache = d->buf; |
3890 | ||
3891 | if (offset) { | |
3892 | if (offset > d->size) | |
3893 | return -EINVAL; | |
3894 | if (!d->cached) | |
3895 | return 0; | |
3896 | int left = d->size - offset; | |
3897 | total_len = left > size ? size: left; | |
3898 | memcpy(buf, cache + offset, total_len); | |
3899 | return total_len; | |
3900 | } | |
3901 | ||
3902 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3903 | if (initpid <= 0) | |
3904 | initpid = fc->pid; | |
3905 | cg = get_pid_cgroup(initpid, "memory"); | |
3906 | if (!cg) | |
3907 | return read_file("/proc/swaps", buf, size, d); | |
6d2f6996 | 3908 | prune_init_slice(cg); |
70dcc12e | 3909 | |
018246ff | 3910 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); |
70dcc12e SH |
3911 | |
3912 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
3913 | goto err; | |
3914 | ||
70dcc12e SH |
3915 | memusage = strtoul(memusage_str, NULL, 10); |
3916 | ||
3917 | if (cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str) && | |
3918 | cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str)) { | |
3919 | ||
018246ff | 3920 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); |
70dcc12e SH |
3921 | memswusage = strtoul(memswusage_str, NULL, 10); |
3922 | ||
70dcc12e SH |
3923 | swap_total = (memswlimit - memlimit) / 1024; |
3924 | swap_free = (memswusage - memusage) / 1024; | |
3925 | } | |
3926 | ||
3927 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
3928 | ||
3929 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
3930 | if (!memswlimit) { | |
3931 | char *line = NULL; | |
3932 | size_t linelen = 0; | |
3933 | FILE *f = fopen("/proc/meminfo", "r"); | |
3934 | ||
3935 | if (!f) | |
3936 | goto err; | |
3937 | ||
3938 | while (getline(&line, &linelen, f) != -1) { | |
3939 | if (startswith(line, "SwapTotal:")) { | |
3940 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); | |
3941 | } else if (startswith(line, "SwapFree:")) { | |
3942 | sscanf(line, "SwapFree: %8lu kB", &swap_free); | |
3943 | } | |
3944 | } | |
3945 | ||
3946 | free(line); | |
3947 | fclose(f); | |
3948 | } | |
3949 | ||
3950 | if (swap_total > 0) { | |
a262ddb7 CB |
3951 | l = snprintf(d->buf + total_len, d->size - total_len, |
3952 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", | |
3953 | swap_total, swap_free); | |
3954 | total_len += l; | |
70dcc12e SH |
3955 | } |
3956 | ||
a262ddb7 | 3957 | if (total_len < 0 || l < 0) { |
70dcc12e SH |
3958 | perror("Error writing to cache"); |
3959 | rv = 0; | |
3960 | goto err; | |
3961 | } | |
3962 | ||
3963 | d->cached = 1; | |
3964 | d->size = (int)total_len; | |
3965 | ||
3966 | if (total_len > size) total_len = size; | |
3967 | memcpy(buf, d->buf, total_len); | |
3968 | rv = total_len; | |
3969 | ||
3970 | err: | |
3971 | free(cg); | |
3972 | free(memswlimit_str); | |
3973 | free(memlimit_str); | |
3974 | free(memusage_str); | |
3975 | free(memswusage_str); | |
70dcc12e SH |
3976 | return rv; |
3977 | } | |
3978 | ||
237e200e SH |
3979 | static off_t get_procfile_size(const char *which) |
3980 | { | |
3981 | FILE *f = fopen(which, "r"); | |
3982 | char *line = NULL; | |
3983 | size_t len = 0; | |
3984 | ssize_t sz, answer = 0; | |
3985 | if (!f) | |
3986 | return 0; | |
3987 | ||
3988 | while ((sz = getline(&line, &len, f)) != -1) | |
3989 | answer += sz; | |
3990 | fclose (f); | |
3991 | free(line); | |
3992 | ||
3993 | return answer; | |
3994 | } | |
3995 | ||
3996 | int proc_getattr(const char *path, struct stat *sb) | |
3997 | { | |
3998 | struct timespec now; | |
3999 | ||
4000 | memset(sb, 0, sizeof(struct stat)); | |
4001 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
4002 | return -EINVAL; | |
4003 | sb->st_uid = sb->st_gid = 0; | |
4004 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
4005 | if (strcmp(path, "/proc") == 0) { | |
4006 | sb->st_mode = S_IFDIR | 00555; | |
4007 | sb->st_nlink = 2; | |
4008 | return 0; | |
4009 | } | |
4010 | if (strcmp(path, "/proc/meminfo") == 0 || | |
4011 | strcmp(path, "/proc/cpuinfo") == 0 || | |
4012 | strcmp(path, "/proc/uptime") == 0 || | |
4013 | strcmp(path, "/proc/stat") == 0 || | |
70dcc12e SH |
4014 | strcmp(path, "/proc/diskstats") == 0 || |
4015 | strcmp(path, "/proc/swaps") == 0) { | |
237e200e SH |
4016 | sb->st_size = 0; |
4017 | sb->st_mode = S_IFREG | 00444; | |
4018 | sb->st_nlink = 1; | |
4019 | return 0; | |
4020 | } | |
4021 | ||
4022 | return -ENOENT; | |
4023 | } | |
4024 | ||
4025 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
4026 | struct fuse_file_info *fi) | |
4027 | { | |
d639f863 CB |
4028 | if (filler(buf, ".", NULL, 0) != 0 || |
4029 | filler(buf, "..", NULL, 0) != 0 || | |
4030 | filler(buf, "cpuinfo", NULL, 0) != 0 || | |
4031 | filler(buf, "meminfo", NULL, 0) != 0 || | |
4032 | filler(buf, "stat", NULL, 0) != 0 || | |
4033 | filler(buf, "uptime", NULL, 0) != 0 || | |
4034 | filler(buf, "diskstats", NULL, 0) != 0 || | |
4035 | filler(buf, "swaps", NULL, 0) != 0) | |
237e200e SH |
4036 | return -EINVAL; |
4037 | return 0; | |
4038 | } | |
4039 | ||
4040 | int proc_open(const char *path, struct fuse_file_info *fi) | |
4041 | { | |
4042 | int type = -1; | |
4043 | struct file_info *info; | |
4044 | ||
4045 | if (strcmp(path, "/proc/meminfo") == 0) | |
4046 | type = LXC_TYPE_PROC_MEMINFO; | |
4047 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
4048 | type = LXC_TYPE_PROC_CPUINFO; | |
4049 | else if (strcmp(path, "/proc/uptime") == 0) | |
4050 | type = LXC_TYPE_PROC_UPTIME; | |
4051 | else if (strcmp(path, "/proc/stat") == 0) | |
4052 | type = LXC_TYPE_PROC_STAT; | |
4053 | else if (strcmp(path, "/proc/diskstats") == 0) | |
4054 | type = LXC_TYPE_PROC_DISKSTATS; | |
70dcc12e SH |
4055 | else if (strcmp(path, "/proc/swaps") == 0) |
4056 | type = LXC_TYPE_PROC_SWAPS; | |
237e200e SH |
4057 | if (type == -1) |
4058 | return -ENOENT; | |
4059 | ||
4060 | info = malloc(sizeof(*info)); | |
4061 | if (!info) | |
4062 | return -ENOMEM; | |
4063 | ||
4064 | memset(info, 0, sizeof(*info)); | |
4065 | info->type = type; | |
4066 | ||
4067 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
4068 | do { | |
4069 | info->buf = malloc(info->buflen); | |
4070 | } while (!info->buf); | |
4071 | memset(info->buf, 0, info->buflen); | |
4072 | /* set actual size to buffer size */ | |
4073 | info->size = info->buflen; | |
4074 | ||
4075 | fi->fh = (unsigned long)info; | |
4076 | return 0; | |
4077 | } | |
4078 | ||
bddbb106 SH |
4079 | int proc_access(const char *path, int mask) |
4080 | { | |
e7849aa3 CB |
4081 | if (strcmp(path, "/proc") == 0 && access(path, R_OK) == 0) |
4082 | return 0; | |
4083 | ||
bddbb106 SH |
4084 | /* these are all read-only */ |
4085 | if ((mask & ~R_OK) != 0) | |
1b060d0a | 4086 | return -EACCES; |
bddbb106 SH |
4087 | return 0; |
4088 | } | |
4089 | ||
237e200e SH |
4090 | int proc_release(const char *path, struct fuse_file_info *fi) |
4091 | { | |
43215927 | 4092 | do_release_file_info(fi); |
237e200e SH |
4093 | return 0; |
4094 | } | |
4095 | ||
4096 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
4097 | struct fuse_file_info *fi) | |
4098 | { | |
4099 | struct file_info *f = (struct file_info *) fi->fh; | |
4100 | ||
4101 | switch (f->type) { | |
4102 | case LXC_TYPE_PROC_MEMINFO: | |
4103 | return proc_meminfo_read(buf, size, offset, fi); | |
4104 | case LXC_TYPE_PROC_CPUINFO: | |
4105 | return proc_cpuinfo_read(buf, size, offset, fi); | |
4106 | case LXC_TYPE_PROC_UPTIME: | |
4107 | return proc_uptime_read(buf, size, offset, fi); | |
4108 | case LXC_TYPE_PROC_STAT: | |
4109 | return proc_stat_read(buf, size, offset, fi); | |
4110 | case LXC_TYPE_PROC_DISKSTATS: | |
4111 | return proc_diskstats_read(buf, size, offset, fi); | |
70dcc12e SH |
4112 | case LXC_TYPE_PROC_SWAPS: |
4113 | return proc_swaps_read(buf, size, offset, fi); | |
237e200e SH |
4114 | default: |
4115 | return -EINVAL; | |
4116 | } | |
4117 | } | |
4118 | ||
29a73c2f CB |
4119 | /* |
4120 | * Functions needed to setup cgroups in the __constructor__. | |
29a73c2f CB |
4121 | */ |
4122 | ||
4123 | static bool mkdir_p(const char *dir, mode_t mode) | |
4124 | { | |
4125 | const char *tmp = dir; | |
4126 | const char *orig = dir; | |
4127 | char *makeme; | |
4128 | ||
4129 | do { | |
4130 | dir = tmp + strspn(tmp, "/"); | |
4131 | tmp = dir + strcspn(dir, "/"); | |
4132 | makeme = strndup(orig, dir - orig); | |
4133 | if (!makeme) | |
4134 | return false; | |
4135 | if (mkdir(makeme, mode) && errno != EEXIST) { | |
b8defc3d | 4136 | lxcfs_error("Failed to create directory '%s': %s.\n", |
29a73c2f CB |
4137 | makeme, strerror(errno)); |
4138 | free(makeme); | |
4139 | return false; | |
4140 | } | |
4141 | free(makeme); | |
4142 | } while(tmp != dir); | |
4143 | ||
4144 | return true; | |
4145 | } | |
4146 | ||
4147 | static bool umount_if_mounted(void) | |
4148 | { | |
4149 | if (umount2(BASEDIR, MNT_DETACH) < 0 && errno != EINVAL) { | |
b8defc3d | 4150 | lxcfs_error("Failed to unmount %s: %s.\n", BASEDIR, strerror(errno)); |
29a73c2f CB |
4151 | return false; |
4152 | } | |
4153 | return true; | |
4154 | } | |
4155 | ||
2283e240 CB |
4156 | /* __typeof__ should be safe to use with all compilers. */ |
4157 | typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic; | |
4158 | static bool has_fs_type(const struct statfs *fs, fs_type_magic magic_val) | |
4159 | { | |
4160 | return (fs->f_type == (fs_type_magic)magic_val); | |
4161 | } | |
4162 | ||
0a4dea41 CB |
4163 | /* |
4164 | * looking at fs/proc_namespace.c, it appears we can | |
4165 | * actually expect the rootfs entry to very specifically contain | |
4166 | * " - rootfs rootfs " | |
4167 | * IIUC, so long as we've chrooted so that rootfs is not our root, | |
4168 | * the rootfs entry should always be skipped in mountinfo contents. | |
4169 | */ | |
4170 | static bool is_on_ramfs(void) | |
4171 | { | |
4172 | FILE *f; | |
4173 | char *p, *p2; | |
4174 | char *line = NULL; | |
4175 | size_t len = 0; | |
4176 | int i; | |
4177 | ||
4178 | f = fopen("/proc/self/mountinfo", "r"); | |
4179 | if (!f) | |
4180 | return false; | |
4181 | ||
4182 | while (getline(&line, &len, f) != -1) { | |
4183 | for (p = line, i = 0; p && i < 4; i++) | |
4184 | p = strchr(p + 1, ' '); | |
4185 | if (!p) | |
4186 | continue; | |
4187 | p2 = strchr(p + 1, ' '); | |
4188 | if (!p2) | |
4189 | continue; | |
4190 | *p2 = '\0'; | |
4191 | if (strcmp(p + 1, "/") == 0) { | |
4192 | // this is '/'. is it the ramfs? | |
4193 | p = strchr(p2 + 1, '-'); | |
4194 | if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) { | |
4195 | free(line); | |
4196 | fclose(f); | |
4197 | return true; | |
4198 | } | |
4199 | } | |
4200 | } | |
4201 | free(line); | |
4202 | fclose(f); | |
4203 | return false; | |
4204 | } | |
4205 | ||
cc309f33 | 4206 | static int pivot_enter() |
0a4dea41 | 4207 | { |
cc309f33 CB |
4208 | int ret = -1, oldroot = -1, newroot = -1; |
4209 | ||
4210 | oldroot = open("/", O_DIRECTORY | O_RDONLY); | |
4211 | if (oldroot < 0) { | |
4212 | lxcfs_error("%s\n", "Failed to open old root for fchdir."); | |
4213 | return ret; | |
4214 | } | |
4215 | ||
4216 | newroot = open(ROOTDIR, O_DIRECTORY | O_RDONLY); | |
4217 | if (newroot < 0) { | |
4218 | lxcfs_error("%s\n", "Failed to open new root for fchdir."); | |
4219 | goto err; | |
4220 | } | |
4221 | ||
4222 | /* change into new root fs */ | |
4223 | if (fchdir(newroot) < 0) { | |
4224 | lxcfs_error("Failed to change directory to new rootfs: %s.\n", ROOTDIR); | |
4225 | goto err; | |
4226 | } | |
4227 | ||
0a4dea41 CB |
4228 | /* pivot_root into our new root fs */ |
4229 | if (pivot_root(".", ".") < 0) { | |
4230 | lxcfs_error("pivot_root() syscall failed: %s.\n", strerror(errno)); | |
cc309f33 | 4231 | goto err; |
0a4dea41 CB |
4232 | } |
4233 | ||
4234 | /* | |
4235 | * At this point the old-root is mounted on top of our new-root. | |
4236 | * To unmounted it we must not be chdir'd into it, so escape back | |
4237 | * to the old-root. | |
4238 | */ | |
4239 | if (fchdir(oldroot) < 0) { | |
4240 | lxcfs_error("%s\n", "Failed to enter old root."); | |
cc309f33 | 4241 | goto err; |
0a4dea41 CB |
4242 | } |
4243 | ||
4244 | if (umount2(".", MNT_DETACH) < 0) { | |
4245 | lxcfs_error("%s\n", "Failed to detach old root."); | |
cc309f33 | 4246 | goto err; |
0a4dea41 CB |
4247 | } |
4248 | ||
4249 | if (fchdir(newroot) < 0) { | |
4250 | lxcfs_error("%s\n", "Failed to re-enter new root."); | |
cc309f33 | 4251 | goto err; |
0a4dea41 CB |
4252 | } |
4253 | ||
cc309f33 CB |
4254 | ret = 0; |
4255 | ||
4256 | err: | |
4257 | if (oldroot > 0) | |
4258 | close(oldroot); | |
4259 | if (newroot > 0) | |
4260 | close(newroot); | |
4261 | ||
4262 | return ret; | |
0a4dea41 CB |
4263 | } |
4264 | ||
4265 | static int chroot_enter() | |
4266 | { | |
4267 | if (mount(ROOTDIR, "/", NULL, MS_REC | MS_BIND, NULL)) { | |
4268 | lxcfs_error("Failed to recursively bind-mount %s into /.", ROOTDIR); | |
4269 | return -1; | |
4270 | } | |
4271 | ||
4272 | if (chroot(".") < 0) { | |
4273 | lxcfs_error("Call to chroot() failed: %s.\n", strerror(errno)); | |
4274 | return -1; | |
4275 | } | |
4276 | ||
4277 | if (chdir("/") < 0) { | |
4278 | lxcfs_error("Failed to change directory: %s.\n", strerror(errno)); | |
4279 | return -1; | |
4280 | } | |
4281 | ||
4282 | return 0; | |
4283 | } | |
4284 | ||
0232cbac | 4285 | static int permute_and_enter(void) |
29a73c2f | 4286 | { |
0a4dea41 CB |
4287 | struct statfs sb; |
4288 | ||
4289 | if (statfs("/", &sb) < 0) { | |
4290 | lxcfs_error("%s\n", "Could not stat / mountpoint."); | |
cc309f33 | 4291 | return -1; |
0a4dea41 CB |
4292 | } |
4293 | ||
4294 | /* has_fs_type() is not reliable. When the ramfs is a tmpfs it will | |
4295 | * likely report TMPFS_MAGIC. Hence, when it reports no we still check | |
4296 | * /proc/1/mountinfo. */ | |
4297 | if (has_fs_type(&sb, RAMFS_MAGIC) || is_on_ramfs()) | |
4298 | return chroot_enter(); | |
29a73c2f | 4299 | |
cc309f33 | 4300 | if (pivot_enter() < 0) { |
0a4dea41 | 4301 | lxcfs_error("%s\n", "Could not perform pivot root."); |
cc309f33 | 4302 | return -1; |
29a73c2f CB |
4303 | } |
4304 | ||
cc309f33 | 4305 | return 0; |
29a73c2f CB |
4306 | } |
4307 | ||
4308 | /* Prepare our new clean root. */ | |
0232cbac | 4309 | static int permute_prepare(void) |
29a73c2f CB |
4310 | { |
4311 | if (mkdir(ROOTDIR, 0700) < 0 && errno != EEXIST) { | |
b8defc3d | 4312 | lxcfs_error("%s\n", "Failed to create directory for new root."); |
29a73c2f CB |
4313 | return -1; |
4314 | } | |
4315 | ||
4316 | if (mount("/", ROOTDIR, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 4317 | lxcfs_error("Failed to bind-mount / for new root: %s.\n", strerror(errno)); |
29a73c2f CB |
4318 | return -1; |
4319 | } | |
4320 | ||
4321 | if (mount(RUNTIME_PATH, ROOTDIR RUNTIME_PATH, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 4322 | lxcfs_error("Failed to bind-mount /run into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
4323 | return -1; |
4324 | } | |
4325 | ||
4326 | if (mount(BASEDIR, ROOTDIR BASEDIR, NULL, MS_REC | MS_MOVE, 0) < 0) { | |
b8defc3d | 4327 | printf("Failed to move " BASEDIR " into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
4328 | return -1; |
4329 | } | |
4330 | ||
4331 | return 0; | |
4332 | } | |
4333 | ||
0232cbac CB |
4334 | /* Calls chroot() on ramfs, pivot_root() in all other cases. */ |
4335 | static bool permute_root(void) | |
29a73c2f CB |
4336 | { |
4337 | /* Prepare new root. */ | |
0232cbac | 4338 | if (permute_prepare() < 0) |
29a73c2f CB |
4339 | return false; |
4340 | ||
4341 | /* Pivot into new root. */ | |
0232cbac | 4342 | if (permute_and_enter() < 0) |
29a73c2f CB |
4343 | return false; |
4344 | ||
4345 | return true; | |
4346 | } | |
4347 | ||
0a4dea41 | 4348 | static bool cgfs_prepare_mounts(void) |
29a73c2f CB |
4349 | { |
4350 | if (!mkdir_p(BASEDIR, 0700)) { | |
b8defc3d | 4351 | lxcfs_error("%s\n", "Failed to create lxcfs cgroup mountpoint."); |
29a73c2f CB |
4352 | return false; |
4353 | } | |
480262c9 | 4354 | |
29a73c2f | 4355 | if (!umount_if_mounted()) { |
b8defc3d | 4356 | lxcfs_error("%s\n", "Failed to clean up old lxcfs cgroup mountpoint."); |
480262c9 CB |
4357 | return false; |
4358 | } | |
4359 | ||
4360 | if (unshare(CLONE_NEWNS) < 0) { | |
b8defc3d | 4361 | lxcfs_error("Failed to unshare mount namespace: %s.\n", strerror(errno)); |
480262c9 CB |
4362 | return false; |
4363 | } | |
4364 | ||
4365 | if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) { | |
b8defc3d | 4366 | lxcfs_error("Failed to remount / private: %s.\n", strerror(errno)); |
29a73c2f CB |
4367 | return false; |
4368 | } | |
480262c9 | 4369 | |
29a73c2f | 4370 | if (mount("tmpfs", BASEDIR, "tmpfs", 0, "size=100000,mode=700") < 0) { |
b8defc3d | 4371 | lxcfs_error("%s\n", "Failed to mount tmpfs over lxcfs cgroup mountpoint."); |
29a73c2f CB |
4372 | return false; |
4373 | } | |
480262c9 | 4374 | |
29a73c2f CB |
4375 | return true; |
4376 | } | |
4377 | ||
0a4dea41 | 4378 | static bool cgfs_mount_hierarchies(void) |
29a73c2f CB |
4379 | { |
4380 | char *target; | |
4381 | size_t clen, len; | |
4382 | int i, ret; | |
4383 | ||
4384 | for (i = 0; i < num_hierarchies; i++) { | |
4385 | char *controller = hierarchies[i]; | |
4386 | clen = strlen(controller); | |
4387 | len = strlen(BASEDIR) + clen + 2; | |
4388 | target = malloc(len); | |
4389 | if (!target) | |
4390 | return false; | |
4391 | ret = snprintf(target, len, "%s/%s", BASEDIR, controller); | |
4392 | if (ret < 0 || ret >= len) { | |
4393 | free(target); | |
4394 | return false; | |
4395 | } | |
4396 | if (mkdir(target, 0755) < 0 && errno != EEXIST) { | |
4397 | free(target); | |
4398 | return false; | |
4399 | } | |
4400 | if (mount(controller, target, "cgroup", 0, controller) < 0) { | |
b8defc3d | 4401 | lxcfs_error("Failed mounting cgroup %s\n", controller); |
29a73c2f CB |
4402 | free(target); |
4403 | return false; | |
4404 | } | |
4405 | ||
4406 | fd_hierarchies[i] = open(target, O_DIRECTORY); | |
4407 | if (fd_hierarchies[i] < 0) { | |
4408 | free(target); | |
4409 | return false; | |
4410 | } | |
4411 | free(target); | |
4412 | } | |
4413 | return true; | |
4414 | } | |
4415 | ||
480262c9 | 4416 | static bool cgfs_setup_controllers(void) |
29a73c2f | 4417 | { |
0a4dea41 | 4418 | if (!cgfs_prepare_mounts()) |
29a73c2f | 4419 | return false; |
29a73c2f | 4420 | |
0a4dea41 | 4421 | if (!cgfs_mount_hierarchies()) { |
b8defc3d | 4422 | lxcfs_error("%s\n", "Failed to set up private lxcfs cgroup mounts."); |
29a73c2f CB |
4423 | return false; |
4424 | } | |
4425 | ||
0232cbac | 4426 | if (!permute_root()) |
29a73c2f CB |
4427 | return false; |
4428 | ||
4429 | return true; | |
4430 | } | |
4431 | ||
480262c9 CB |
4432 | static int preserve_ns(int pid) |
4433 | { | |
4434 | int ret; | |
4435 | size_t len = 5 /* /proc */ + 21 /* /int_as_str */ + 7 /* /ns/mnt */ + 1 /* \0 */; | |
4436 | char path[len]; | |
4437 | ||
4438 | ret = snprintf(path, len, "/proc/%d/ns/mnt", pid); | |
4439 | if (ret < 0 || (size_t)ret >= len) | |
4440 | return -1; | |
4441 | ||
4442 | return open(path, O_RDONLY | O_CLOEXEC); | |
4443 | } | |
4444 | ||
29a73c2f | 4445 | static void __attribute__((constructor)) collect_and_mount_subsystems(void) |
237e200e SH |
4446 | { |
4447 | FILE *f; | |
e58dab00 CB |
4448 | char *cret, *line = NULL; |
4449 | char cwd[MAXPATHLEN]; | |
237e200e | 4450 | size_t len = 0; |
480262c9 | 4451 | int i, init_ns = -1; |
237e200e SH |
4452 | |
4453 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) { | |
b8defc3d | 4454 | lxcfs_error("Error opening /proc/self/cgroup: %s\n", strerror(errno)); |
237e200e SH |
4455 | return; |
4456 | } | |
e58dab00 | 4457 | |
237e200e SH |
4458 | while (getline(&line, &len, f) != -1) { |
4459 | char *p, *p2; | |
4460 | ||
4461 | p = strchr(line, ':'); | |
4462 | if (!p) | |
4463 | goto out; | |
4464 | *(p++) = '\0'; | |
4465 | ||
4466 | p2 = strrchr(p, ':'); | |
4467 | if (!p2) | |
4468 | goto out; | |
4469 | *p2 = '\0'; | |
4470 | ||
a67719f6 CB |
4471 | /* With cgroupv2 /proc/self/cgroup can contain entries of the |
4472 | * form: 0::/ This will cause lxcfs to fail the cgroup mounts | |
4473 | * because it parses out the empty string "" and later on passes | |
4474 | * it to mount(). Let's skip such entries. | |
4475 | */ | |
4476 | if (!strcmp(p, "")) | |
4477 | continue; | |
4478 | ||
237e200e SH |
4479 | if (!store_hierarchy(line, p)) |
4480 | goto out; | |
4481 | } | |
4482 | ||
480262c9 CB |
4483 | /* Preserve initial namespace. */ |
4484 | init_ns = preserve_ns(getpid()); | |
b8defc3d CB |
4485 | if (init_ns < 0) { |
4486 | lxcfs_error("%s\n", "Failed to preserve initial mount namespace."); | |
480262c9 | 4487 | goto out; |
b8defc3d | 4488 | } |
480262c9 | 4489 | |
92c3ee11 | 4490 | fd_hierarchies = malloc(sizeof(int) * num_hierarchies); |
b8defc3d CB |
4491 | if (!fd_hierarchies) { |
4492 | lxcfs_error("%s\n", strerror(errno)); | |
29a73c2f | 4493 | goto out; |
b8defc3d | 4494 | } |
29a73c2f | 4495 | |
480262c9 CB |
4496 | for (i = 0; i < num_hierarchies; i++) |
4497 | fd_hierarchies[i] = -1; | |
4498 | ||
e58dab00 CB |
4499 | cret = getcwd(cwd, MAXPATHLEN); |
4500 | if (!cret) | |
4501 | lxcfs_debug("Could not retrieve current working directory: %s.\n", strerror(errno)); | |
4502 | ||
480262c9 CB |
4503 | /* This function calls unshare(CLONE_NEWNS) our initial mount namespace |
4504 | * to privately mount lxcfs cgroups. */ | |
b8defc3d CB |
4505 | if (!cgfs_setup_controllers()) { |
4506 | lxcfs_error("%s\n", "Failed to setup private cgroup mounts for lxcfs."); | |
29a73c2f | 4507 | goto out; |
b8defc3d | 4508 | } |
480262c9 | 4509 | |
b8defc3d CB |
4510 | if (setns(init_ns, 0) < 0) { |
4511 | lxcfs_error("Failed to switch back to initial mount namespace: %s.\n", strerror(errno)); | |
29a73c2f | 4512 | goto out; |
b8defc3d | 4513 | } |
29a73c2f | 4514 | |
e58dab00 CB |
4515 | if (!cret || chdir(cwd) < 0) |
4516 | lxcfs_debug("Could not change back to original working directory: %s.\n", strerror(errno)); | |
4517 | ||
237e200e SH |
4518 | print_subsystems(); |
4519 | ||
4520 | out: | |
4521 | free(line); | |
4522 | fclose(f); | |
480262c9 CB |
4523 | if (init_ns >= 0) |
4524 | close(init_ns); | |
237e200e SH |
4525 | } |
4526 | ||
4527 | static void __attribute__((destructor)) free_subsystems(void) | |
4528 | { | |
4529 | int i; | |
4530 | ||
b8defc3d CB |
4531 | lxcfs_debug("%s\n", "Running destructor for liblxcfs."); |
4532 | ||
29a73c2f | 4533 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
4534 | if (hierarchies[i]) |
4535 | free(hierarchies[i]); | |
480262c9 | 4536 | if (fd_hierarchies && fd_hierarchies[i] >= 0) |
29a73c2f CB |
4537 | close(fd_hierarchies[i]); |
4538 | } | |
237e200e | 4539 | free(hierarchies); |
480262c9 | 4540 | free(fd_hierarchies); |
237e200e | 4541 | } |