]>
Commit | Line | Data |
---|---|---|
758ad80c SH |
1 | /* lxcfs |
2 | * | |
3 | * Copyright © 2014 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2, as | |
8 | * published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along | |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 | */ | |
19 | ||
20 | /* | |
21 | * NOTES - make sure to run this as -s to avoid threading. | |
22 | * TODO - can we enforce that here from the code? | |
23 | */ | |
24 | #define FUSE_USE_VERSION 26 | |
25 | ||
2183082c | 26 | #include <stdio.h> |
758ad80c SH |
27 | #include <dirent.h> |
28 | #include <fcntl.h> | |
29 | #include <fuse.h> | |
30 | #include <unistd.h> | |
31 | #include <errno.h> | |
32 | #include <stdbool.h> | |
33 | #include <time.h> | |
34 | #include <string.h> | |
35 | #include <stdlib.h> | |
36 | #include <libgen.h> | |
37 | ||
38 | #include <nih/alloc.h> | |
39 | #include <nih/string.h> | |
40 | ||
41 | #include "cgmanager.h" | |
42 | ||
43 | struct lxcfs_state { | |
44 | /* | |
45 | * a null-terminated, nih-allocated list of the mounted subsystems. We | |
46 | * detect this at startup. | |
47 | */ | |
48 | char **subsystems; | |
49 | }; | |
50 | #define LXCFS_DATA ((struct lxcfs_state *) fuse_get_context()->private_data) | |
51 | ||
053a659d SH |
52 | /* |
53 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
54 | * valid in the caller's namespace, return the id mapped into | |
55 | * pid's namespace. | |
56 | * Returns the mapped id, or -1 on error. | |
57 | */ | |
58 | unsigned int | |
59 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
60 | { | |
61 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
62 | hostuid, // base id for a range in the caller's namespace | |
63 | count; // number of ids in this range | |
64 | char line[400]; | |
65 | int ret; | |
66 | ||
67 | fseek(idfile, 0L, SEEK_SET); | |
68 | while (fgets(line, 400, idfile)) { | |
69 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
70 | if (ret != 3) | |
71 | continue; | |
72 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
73 | /* | |
74 | * uids wrapped around - unexpected as this is a procfile, | |
75 | * so just bail. | |
76 | */ | |
77 | fprintf(stderr, "pid wrapparound at entry %u %u %u in %s", | |
78 | nsuid, hostuid, count, line); | |
79 | return -1; | |
80 | } | |
81 | if (hostuid <= in_id && hostuid+count > in_id) { | |
82 | /* | |
83 | * now since hostuid <= in_id < hostuid+count, and | |
84 | * hostuid+count and nsuid+count do not wrap around, | |
85 | * we know that nsuid+(in_id-hostuid) which must be | |
86 | * less that nsuid+(count) must not wrap around | |
87 | */ | |
88 | return (in_id - hostuid) + nsuid; | |
89 | } | |
90 | } | |
91 | ||
92 | // no answer found | |
93 | return -1; | |
94 | } | |
95 | ||
341b21ad SH |
96 | /* |
97 | * for is_privileged_over, | |
98 | * specify whether we require the calling uid to be root in his | |
99 | * namespace | |
100 | */ | |
101 | #define NS_ROOT_REQD true | |
102 | #define NS_ROOT_OPT false | |
103 | ||
104 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
758ad80c | 105 | { |
053a659d SH |
106 | nih_local char *fpath = NULL; |
107 | bool answer = false; | |
108 | uid_t nsuid; | |
109 | ||
341b21ad SH |
110 | if (victim == -1 || uid == -1) |
111 | return false; | |
112 | ||
113 | /* | |
114 | * If the request is one not requiring root in the namespace, | |
115 | * then having the same uid suffices. (i.e. uid 1000 has write | |
116 | * access to files owned by uid 1000 | |
117 | */ | |
118 | if (!req_ns_root && uid == victim) | |
758ad80c SH |
119 | return true; |
120 | ||
053a659d SH |
121 | fpath = NIH_MUST( nih_sprintf(NULL, "/proc/%d/uid_map", pid) ); |
122 | FILE *f = fopen(fpath, "r"); | |
123 | if (!f) | |
124 | return false; | |
125 | ||
341b21ad | 126 | /* if caller's not root in his namespace, reject */ |
053a659d SH |
127 | nsuid = convert_id_to_ns(f, uid); |
128 | if (nsuid) | |
129 | goto out; | |
130 | ||
341b21ad SH |
131 | /* |
132 | * If victim is not mapped into caller's ns, reject. | |
133 | * XXX I'm not sure this check is needed given that fuse | |
134 | * will be sending requests where the vfs has converted | |
135 | */ | |
053a659d SH |
136 | nsuid = convert_id_to_ns(f, victim); |
137 | if (nsuid == -1) | |
138 | goto out; | |
139 | ||
140 | answer = true; | |
141 | ||
142 | out: | |
143 | fclose(f); | |
144 | return answer; | |
758ad80c SH |
145 | } |
146 | ||
147 | static bool perms_include(int fmode, mode_t req_mode) | |
148 | { | |
2ad6d2bd SH |
149 | mode_t r; |
150 | ||
151 | switch (req_mode & O_ACCMODE) { | |
152 | case O_RDONLY: | |
153 | r = S_IROTH; | |
154 | break; | |
155 | case O_WRONLY: | |
156 | r = S_IWOTH; | |
157 | break; | |
158 | case O_RDWR: | |
159 | r = S_IROTH | S_IWOTH; | |
160 | break; | |
161 | default: | |
162 | return false; | |
163 | } | |
164 | return ((fmode & r) == r); | |
758ad80c SH |
165 | } |
166 | ||
167 | /* | |
168 | * check whether a fuse context may access a cgroup dir or file | |
169 | * | |
170 | * If file is not null, it is a cgroup file to check under cg. | |
171 | * If file is null, then we are checking perms on cg itself. | |
172 | * | |
173 | * For files we can check the mode of the list_keys result. | |
174 | * For cgroups, we must make assumptions based on the files under the | |
175 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
176 | * yet. | |
177 | */ | |
178 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
179 | { | |
180 | nih_local struct cgm_keys **list = NULL; | |
181 | int i; | |
182 | ||
183 | if (!file) | |
184 | file = "tasks"; | |
185 | ||
186 | if (*file == '/') | |
187 | file++; | |
188 | ||
189 | if (!cgm_list_keys(contrl, cg, &list)) | |
190 | return false; | |
191 | for (i = 0; list[i]; i++) { | |
192 | if (strcmp(list[i]->name, file) == 0) { | |
193 | struct cgm_keys *k = list[i]; | |
341b21ad | 194 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { |
758ad80c SH |
195 | if (perms_include(k->mode >> 6, mode)) |
196 | return true; | |
197 | } | |
198 | if (fc->gid == k->gid) { | |
199 | if (perms_include(k->mode >> 3, mode)) | |
200 | return true; | |
201 | } | |
202 | return perms_include(k->mode, mode); | |
203 | } | |
204 | } | |
205 | ||
206 | return false; | |
207 | } | |
208 | ||
209 | /* | |
210 | * given /cgroup/freezer/a/b, return "freezer". this will be nih-allocated | |
211 | * and needs to be nih_freed. | |
212 | */ | |
213 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
214 | { | |
215 | const char *p1; | |
216 | char *ret, *slash; | |
217 | ||
218 | if (strlen(path) < 9) | |
219 | return NULL; | |
220 | p1 = path+8; | |
221 | ret = nih_strdup(NULL, p1); | |
222 | if (!ret) | |
223 | return ret; | |
224 | slash = strstr(ret, "/"); | |
225 | if (slash) | |
226 | *slash = '\0'; | |
227 | ||
228 | /* verify that it is a subsystem */ | |
229 | char **list = LXCFS_DATA ? LXCFS_DATA->subsystems : NULL; | |
230 | int i; | |
231 | if (!list) { | |
232 | nih_free(ret); | |
233 | return NULL; | |
234 | } | |
235 | for (i = 0; list[i]; i++) { | |
236 | if (strcmp(list[i], ret) == 0) | |
237 | return ret; | |
238 | } | |
239 | nih_free(ret); | |
240 | return NULL; | |
241 | } | |
242 | ||
243 | /* | |
244 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
245 | * Note that the returned value may include files (keynames) etc | |
246 | */ | |
247 | static const char *find_cgroup_in_path(const char *path) | |
248 | { | |
249 | const char *p1; | |
250 | ||
251 | if (strlen(path) < 9) | |
252 | return NULL; | |
253 | p1 = strstr(path+8, "/"); | |
254 | if (!p1) | |
255 | return NULL; | |
256 | return p1+1; | |
257 | } | |
258 | ||
259 | static bool is_child_cgroup(const char *contr, const char *dir, const char *f) | |
260 | { | |
261 | nih_local char **list = NULL; | |
262 | int i; | |
263 | ||
264 | if (!f) | |
265 | return false; | |
266 | if (*f == '/') | |
267 | f++; | |
268 | ||
269 | if (!cgm_list_children(contr, dir, &list)) | |
270 | return false; | |
271 | for (i = 0; list[i]; i++) { | |
272 | if (strcmp(list[i], f) == 0) | |
273 | return true; | |
274 | } | |
275 | ||
276 | return false; | |
277 | } | |
278 | ||
279 | static struct cgm_keys *get_cgroup_key(const char *contr, const char *dir, const char *f) | |
280 | { | |
281 | nih_local struct cgm_keys **list = NULL; | |
282 | struct cgm_keys *k; | |
283 | int i; | |
284 | ||
285 | if (!f) | |
286 | return NULL; | |
287 | if (*f == '/') | |
288 | f++; | |
289 | if (!cgm_list_keys(contr, dir, &list)) | |
290 | return NULL; | |
291 | for (i = 0; list[i]; i++) { | |
292 | if (strcmp(list[i]->name, f) == 0) { | |
293 | k = NIH_MUST( nih_alloc(NULL, (sizeof(*k))) ); | |
294 | k->name = NIH_MUST( nih_strdup(k, list[i]->name) ); | |
295 | k->uid = list[i]->uid; | |
296 | k->gid = list[i]->gid; | |
297 | k->mode = list[i]->mode; | |
298 | return k; | |
299 | } | |
300 | } | |
301 | ||
302 | return NULL; | |
303 | } | |
304 | ||
305 | static void get_cgdir_and_path(const char *cg, char **dir, char **file) | |
306 | { | |
758ad80c SH |
307 | char *p; |
308 | ||
309 | *dir = NIH_MUST( nih_strdup(NULL, cg) ); | |
310 | *file = strrchr(cg, '/'); | |
311 | if (!*file) { | |
312 | *file = NULL; | |
313 | return; | |
314 | } | |
315 | p = strrchr(*dir, '/'); | |
316 | *p = '\0'; | |
317 | } | |
318 | ||
99978832 SH |
319 | static size_t get_file_size(const char *contrl, const char *cg, const char *f) |
320 | { | |
321 | nih_local char *data = NULL; | |
322 | size_t s; | |
323 | if (!cgm_get_value(contrl, cg, f, &data)) | |
324 | return -EINVAL; | |
325 | s = strlen(data); | |
326 | return s; | |
327 | } | |
2ad6d2bd | 328 | |
758ad80c | 329 | /* |
2ad6d2bd | 330 | * FUSE ops for /cgroup |
758ad80c | 331 | */ |
2ad6d2bd | 332 | |
758ad80c SH |
333 | static int cg_getattr(const char *path, struct stat *sb) |
334 | { | |
335 | struct timespec now; | |
336 | struct fuse_context *fc = fuse_get_context(); | |
337 | nih_local char * cgdir = NULL; | |
338 | char *fpath = NULL, *path1, *path2; | |
339 | nih_local struct cgm_keys *k = NULL; | |
340 | const char *cgroup; | |
341 | nih_local char *controller = NULL; | |
342 | ||
343 | ||
344 | if (!fc) | |
345 | return -EIO; | |
346 | ||
347 | memset(sb, 0, sizeof(struct stat)); | |
348 | ||
349 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
350 | return -EINVAL; | |
351 | ||
352 | sb->st_uid = sb->st_gid = 0; | |
353 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
354 | sb->st_size = 0; | |
355 | ||
356 | if (strcmp(path, "/cgroup") == 0) { | |
357 | sb->st_mode = S_IFDIR | 00755; | |
358 | sb->st_nlink = 2; | |
359 | return 0; | |
360 | } | |
361 | ||
362 | controller = pick_controller_from_path(fc, path); | |
363 | if (!controller) | |
364 | return -EIO; | |
758ad80c SH |
365 | cgroup = find_cgroup_in_path(path); |
366 | if (!cgroup) { | |
367 | /* this is just /cgroup/controller, return it as a dir */ | |
368 | sb->st_mode = S_IFDIR | 00755; | |
369 | sb->st_nlink = 2; | |
370 | return 0; | |
371 | } | |
341b21ad | 372 | |
758ad80c SH |
373 | get_cgdir_and_path(cgroup, &cgdir, &fpath); |
374 | ||
375 | if (!fpath) { | |
376 | path1 = "/"; | |
377 | path2 = cgdir; | |
378 | } else { | |
379 | path1 = cgdir; | |
380 | path2 = fpath; | |
381 | } | |
382 | ||
758ad80c SH |
383 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. |
384 | * Then check that caller's cgroup is under path if fpath is a child | |
385 | * cgroup, or cgdir if fpath is a file */ | |
386 | ||
387 | if (is_child_cgroup(controller, path1, path2)) { | |
388 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
389 | return -EPERM; | |
390 | ||
053a659d SH |
391 | // get uid, gid, from '/tasks' file and make up a mode |
392 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
393 | sb->st_mode = S_IFDIR | 00755; | |
394 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
395 | if (!k) { | |
053a659d SH |
396 | sb->st_uid = sb->st_gid = 0; |
397 | } else { | |
053a659d SH |
398 | sb->st_uid = k->uid; |
399 | sb->st_gid = k->gid; | |
400 | } | |
758ad80c SH |
401 | sb->st_nlink = 2; |
402 | return 0; | |
403 | } | |
404 | ||
405 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
406 | if (!fc_may_access(fc, controller, path1, path2, O_RDONLY)) | |
407 | return -EPERM; | |
408 | ||
758ad80c | 409 | sb->st_mode = S_IFREG | k->mode; |
053a659d | 410 | sb->st_nlink = 1; |
758ad80c SH |
411 | sb->st_uid = k->uid; |
412 | sb->st_gid = k->gid; | |
99978832 | 413 | sb->st_size = get_file_size(controller, path1, path2); |
758ad80c SH |
414 | return 0; |
415 | } | |
416 | ||
ab54b798 | 417 | return -ENOENT; |
758ad80c | 418 | } |
2183082c | 419 | |
758ad80c | 420 | static int cg_opendir(const char *path, struct fuse_file_info *fi) |
2183082c | 421 | { |
758ad80c SH |
422 | return 0; |
423 | } | |
424 | ||
758ad80c SH |
425 | static int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, |
426 | struct fuse_file_info *fi) | |
427 | { | |
428 | struct fuse_context *fc = fuse_get_context(); | |
429 | ||
430 | if (!fc) | |
431 | return -EIO; | |
432 | ||
433 | if (strcmp(path, "/cgroup") == 0) { | |
434 | // get list of controllers | |
435 | char **list = LXCFS_DATA ? LXCFS_DATA->subsystems : NULL; | |
436 | int i; | |
437 | ||
438 | if (!list) | |
439 | return -EIO; | |
440 | /* TODO - collect the list of controllers at fuse_init */ | |
441 | for (i = 0; list[i]; i++) { | |
442 | if (filler(buf, list[i], NULL, 0) != 0) { | |
443 | return -EIO; | |
444 | } | |
445 | } | |
446 | return 0; | |
447 | } | |
448 | ||
449 | // return list of keys for the controller, and list of child cgroups | |
450 | nih_local struct cgm_keys **list = NULL; | |
451 | const char *cgroup; | |
452 | nih_local char *controller = NULL; | |
453 | int i; | |
454 | ||
455 | controller = pick_controller_from_path(fc, path); | |
456 | if (!controller) | |
457 | return -EIO; | |
458 | ||
459 | cgroup = find_cgroup_in_path(path); | |
460 | if (!cgroup) { | |
461 | /* this is just /cgroup/controller, return its contents */ | |
462 | cgroup = "/"; | |
463 | } | |
464 | ||
465 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
466 | return -EPERM; | |
467 | ||
468 | if (!cgm_list_keys(controller, cgroup, &list)) | |
469 | return -EINVAL; | |
470 | for (i = 0; list[i]; i++) { | |
758ad80c SH |
471 | if (filler(buf, list[i]->name, NULL, 0) != 0) { |
472 | return -EIO; | |
473 | } | |
474 | } | |
475 | ||
476 | // now get the list of child cgroups | |
477 | nih_local char **clist; | |
478 | ||
479 | if (!cgm_list_children(controller, cgroup, &clist)) | |
480 | return 0; | |
481 | for (i = 0; clist[i]; i++) { | |
758ad80c SH |
482 | if (filler(buf, clist[i], NULL, 0) != 0) { |
483 | return -EIO; | |
484 | } | |
485 | } | |
486 | return 0; | |
487 | } | |
488 | ||
489 | static int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
490 | { | |
491 | return 0; | |
492 | } | |
493 | ||
99978832 SH |
494 | static int cg_open(const char *path, struct fuse_file_info *fi) |
495 | { | |
496 | nih_local char *controller = NULL; | |
497 | const char *cgroup; | |
498 | char *fpath = NULL, *path1, *path2; | |
499 | nih_local char * cgdir = NULL; | |
500 | nih_local struct cgm_keys *k = NULL; | |
501 | struct fuse_context *fc = fuse_get_context(); | |
502 | ||
503 | if (!fc) | |
504 | return -EIO; | |
505 | ||
506 | controller = pick_controller_from_path(fc, path); | |
507 | if (!controller) | |
508 | return -EIO; | |
509 | cgroup = find_cgroup_in_path(path); | |
510 | if (!cgroup) | |
511 | return -EINVAL; | |
512 | ||
513 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
514 | if (!fpath) { | |
515 | path1 = "/"; | |
516 | path2 = cgdir; | |
517 | } else { | |
518 | path1 = cgdir; | |
519 | path2 = fpath; | |
520 | } | |
521 | ||
522 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
523 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) | |
524 | return -EPERM; | |
525 | ||
526 | /* TODO - we want to cache this info for read/write */ | |
527 | return 0; | |
528 | } | |
529 | ||
530 | return -EINVAL; | |
531 | } | |
532 | ||
533 | static int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
534 | struct fuse_file_info *fi) | |
535 | { | |
536 | nih_local char *controller = NULL; | |
537 | const char *cgroup; | |
538 | char *fpath = NULL, *path1, *path2; | |
539 | struct fuse_context *fc = fuse_get_context(); | |
540 | nih_local char * cgdir = NULL; | |
541 | nih_local struct cgm_keys *k = NULL; | |
542 | ||
543 | if (offset) | |
544 | return -EIO; | |
545 | ||
546 | if (!fc) | |
547 | return -EIO; | |
548 | ||
549 | controller = pick_controller_from_path(fc, path); | |
550 | if (!controller) | |
551 | return -EIO; | |
552 | cgroup = find_cgroup_in_path(path); | |
553 | if (!cgroup) | |
554 | return -EINVAL; | |
555 | ||
556 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
557 | if (!fpath) { | |
558 | path1 = "/"; | |
559 | path2 = cgdir; | |
560 | } else { | |
561 | path1 = cgdir; | |
562 | path2 = fpath; | |
563 | } | |
564 | ||
565 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
566 | nih_local char *data = NULL; | |
567 | int s; | |
568 | ||
2ad6d2bd | 569 | if (!fc_may_access(fc, controller, path1, path2, O_RDONLY)) |
99978832 SH |
570 | return -EPERM; |
571 | ||
572 | if (!cgm_get_value(controller, path1, path2, &data)) | |
573 | return -EINVAL; | |
574 | ||
575 | s = strlen(data); | |
576 | if (s > size) | |
577 | s = size; | |
578 | memcpy(buf, data, s); | |
579 | ||
99978832 SH |
580 | return s; |
581 | } | |
582 | ||
583 | return -EINVAL; | |
584 | } | |
585 | ||
2ad6d2bd SH |
586 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, |
587 | struct fuse_file_info *fi) | |
588 | { | |
589 | nih_local char *controller = NULL; | |
590 | const char *cgroup; | |
591 | char *fpath = NULL, *path1, *path2; | |
592 | struct fuse_context *fc = fuse_get_context(); | |
593 | nih_local char * cgdir = NULL; | |
594 | nih_local struct cgm_keys *k = NULL; | |
595 | ||
2ad6d2bd SH |
596 | if (offset) |
597 | return -EIO; | |
598 | ||
599 | if (!fc) | |
600 | return -EIO; | |
601 | ||
602 | controller = pick_controller_from_path(fc, path); | |
603 | if (!controller) | |
604 | return -EIO; | |
605 | cgroup = find_cgroup_in_path(path); | |
606 | if (!cgroup) | |
607 | return -EINVAL; | |
608 | ||
609 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
610 | if (!fpath) { | |
611 | path1 = "/"; | |
612 | path2 = cgdir; | |
613 | } else { | |
614 | path1 = cgdir; | |
615 | path2 = fpath; | |
616 | } | |
617 | ||
618 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
619 | if (!fc_may_access(fc, controller, path1, path2, O_WRONLY)) | |
620 | return -EPERM; | |
621 | ||
622 | if (!cgm_set_value(controller, path1, path2, buf)) | |
623 | return -EINVAL; | |
624 | ||
625 | return size; | |
626 | } | |
627 | ||
628 | return -EINVAL; | |
629 | } | |
630 | ||
341b21ad SH |
631 | int cg_chown(const char *path, uid_t uid, gid_t gid) |
632 | { | |
633 | struct fuse_context *fc = fuse_get_context(); | |
634 | nih_local char * cgdir = NULL; | |
635 | char *fpath = NULL, *path1, *path2; | |
636 | nih_local struct cgm_keys *k = NULL; | |
637 | const char *cgroup; | |
638 | nih_local char *controller = NULL; | |
639 | ||
640 | ||
641 | if (!fc) | |
642 | return -EIO; | |
643 | ||
644 | if (strcmp(path, "/cgroup") == 0) | |
645 | return -EINVAL; | |
646 | ||
647 | controller = pick_controller_from_path(fc, path); | |
648 | if (!controller) | |
649 | return -EIO; | |
650 | cgroup = find_cgroup_in_path(path); | |
651 | if (!cgroup) | |
652 | /* this is just /cgroup/controller */ | |
653 | return -EINVAL; | |
654 | ||
655 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
656 | ||
657 | if (!fpath) { | |
658 | path1 = "/"; | |
659 | path2 = cgdir; | |
660 | } else { | |
661 | path1 = cgdir; | |
662 | path2 = fpath; | |
663 | } | |
664 | ||
665 | if (is_child_cgroup(controller, path1, path2)) { | |
666 | // get uid, gid, from '/tasks' file and make up a mode | |
667 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
668 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
669 | ||
670 | } else | |
671 | k = get_cgroup_key(controller, path1, path2); | |
672 | ||
673 | if (!k) | |
674 | return -EINVAL; | |
675 | ||
676 | /* | |
677 | * This being a fuse request, the uid and gid must be valid | |
678 | * in the caller's namespace. So we can just check to make | |
679 | * sure that the caller is root in his uid, and privileged | |
680 | * over the file's current owner. | |
681 | */ | |
682 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) | |
683 | return -EPERM; | |
684 | ||
685 | if (!cgm_chown_file(controller, cgroup, uid, gid)) | |
686 | return -EINVAL; | |
687 | return 0; | |
688 | } | |
2ad6d2bd | 689 | |
fd2e4e03 SH |
690 | int cg_chmod(const char *path, mode_t mode) |
691 | { | |
0a1bb5ea SH |
692 | struct fuse_context *fc = fuse_get_context(); |
693 | nih_local char * cgdir = NULL; | |
694 | char *fpath = NULL, *path1, *path2; | |
695 | nih_local struct cgm_keys *k = NULL; | |
696 | const char *cgroup; | |
697 | nih_local char *controller = NULL; | |
698 | ||
699 | if (!fc) | |
700 | return -EIO; | |
701 | ||
702 | if (strcmp(path, "/cgroup") == 0) | |
703 | return -EINVAL; | |
704 | ||
705 | controller = pick_controller_from_path(fc, path); | |
706 | if (!controller) | |
707 | return -EIO; | |
708 | cgroup = find_cgroup_in_path(path); | |
709 | if (!cgroup) | |
710 | /* this is just /cgroup/controller */ | |
711 | return -EINVAL; | |
712 | ||
713 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
714 | ||
715 | if (!fpath) { | |
716 | path1 = "/"; | |
717 | path2 = cgdir; | |
718 | } else { | |
719 | path1 = cgdir; | |
720 | path2 = fpath; | |
721 | } | |
722 | ||
723 | if (is_child_cgroup(controller, path1, path2)) { | |
724 | // get uid, gid, from '/tasks' file and make up a mode | |
725 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
726 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
727 | ||
728 | } else | |
729 | k = get_cgroup_key(controller, path1, path2); | |
730 | ||
731 | if (!k) | |
732 | return -EINVAL; | |
733 | ||
734 | /* | |
735 | * This being a fuse request, the uid and gid must be valid | |
736 | * in the caller's namespace. So we can just check to make | |
737 | * sure that the caller is root in his uid, and privileged | |
738 | * over the file's current owner. | |
739 | */ | |
740 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) | |
741 | return -EPERM; | |
742 | ||
743 | if (!cgm_chmod_file(controller, cgroup, mode)) | |
744 | return -EINVAL; | |
745 | return 0; | |
fd2e4e03 SH |
746 | } |
747 | ||
ab54b798 SH |
748 | int cg_mkdir(const char *path, mode_t mode) |
749 | { | |
750 | struct fuse_context *fc = fuse_get_context(); | |
751 | nih_local struct cgm_keys **list = NULL; | |
752 | char *fpath = NULL, *path1; | |
753 | nih_local char * cgdir = NULL; | |
754 | const char *cgroup; | |
755 | nih_local char *controller = NULL; | |
756 | ||
ab54b798 SH |
757 | if (!fc) |
758 | return -EIO; | |
759 | ||
760 | ||
761 | controller = pick_controller_from_path(fc, path); | |
762 | if (!controller) | |
763 | return -EIO; | |
764 | ||
765 | cgroup = find_cgroup_in_path(path); | |
766 | if (!cgroup) | |
767 | return -EIO; | |
768 | ||
769 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
770 | if (!fpath) | |
771 | path1 = "/"; | |
772 | else | |
773 | path1 = cgdir; | |
774 | ||
775 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) | |
776 | return -EPERM; | |
777 | ||
778 | ||
779 | if (!cgm_create(controller, cgroup, fc->uid, fc->gid)) | |
780 | return -EINVAL; | |
781 | ||
782 | return 0; | |
783 | } | |
784 | ||
50d8d5b5 SH |
785 | static int cg_rmdir(const char *path) |
786 | { | |
787 | struct fuse_context *fc = fuse_get_context(); | |
788 | nih_local struct cgm_keys **list = NULL; | |
789 | char *fpath = NULL; | |
790 | nih_local char * cgdir = NULL; | |
791 | const char *cgroup; | |
792 | nih_local char *controller = NULL; | |
793 | ||
794 | if (!fc) | |
795 | return -EIO; | |
796 | ||
797 | ||
798 | controller = pick_controller_from_path(fc, path); | |
799 | if (!controller) | |
800 | return -EIO; | |
801 | ||
802 | cgroup = find_cgroup_in_path(path); | |
803 | if (!cgroup) | |
804 | return -EIO; | |
805 | ||
806 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
807 | if (!fpath) | |
808 | return -EINVAL; | |
809 | ||
810 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) | |
811 | return -EPERM; | |
812 | ||
813 | if (!cgm_remove(controller, cgroup)) | |
814 | return -EINVAL; | |
815 | ||
816 | return 0; | |
817 | } | |
818 | ||
758ad80c | 819 | /* |
2ad6d2bd | 820 | * FUSE ops for /proc |
758ad80c | 821 | */ |
758ad80c SH |
822 | |
823 | static int proc_getattr(const char *path, struct stat *sb) | |
824 | { | |
825 | if (strcmp(path, "/proc") != 0) | |
826 | return -EINVAL; | |
827 | sb->st_mode = S_IFDIR | 00755; | |
828 | sb->st_nlink = 2; | |
829 | return 0; | |
830 | } | |
831 | ||
2ad6d2bd SH |
832 | /* |
833 | * FUSE ops for / | |
834 | * these just delegate to the /proc and /cgroup ops as | |
835 | * needed | |
836 | */ | |
758ad80c SH |
837 | |
838 | static int lxcfs_getattr(const char *path, struct stat *sb) | |
839 | { | |
840 | if (strcmp(path, "/") == 0) { | |
841 | sb->st_mode = S_IFDIR | 00755; | |
842 | sb->st_nlink = 2; | |
843 | return 0; | |
844 | } | |
845 | if (strncmp(path, "/cgroup", 7) == 0) { | |
846 | return cg_getattr(path, sb); | |
847 | } | |
848 | if (strncmp(path, "/proc", 7) == 0) { | |
849 | return proc_getattr(path, sb); | |
850 | } | |
851 | return -EINVAL; | |
852 | } | |
853 | ||
854 | static int lxcfs_opendir(const char *path, struct fuse_file_info *fi) | |
855 | { | |
856 | if (strcmp(path, "/") == 0) | |
857 | return 0; | |
858 | ||
859 | if (strncmp(path, "/cgroup", 7) == 0) { | |
860 | return cg_opendir(path, fi); | |
861 | } | |
862 | return -EINVAL; | |
863 | } | |
864 | ||
865 | static int lxcfs_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
866 | struct fuse_file_info *fi) | |
867 | { | |
868 | if (strcmp(path, "/") == 0) { | |
869 | if (filler(buf, "proc", NULL, 0) != 0 || | |
870 | filler(buf, "cgroup", NULL, 0) != 0) | |
871 | return -EINVAL; | |
872 | return 0; | |
873 | } | |
874 | if (strncmp(path, "/cgroup", 7) == 0) { | |
875 | return cg_readdir(path, buf, filler, offset, fi); | |
876 | } | |
877 | return -EINVAL; | |
878 | } | |
879 | ||
880 | static int lxcfs_releasedir(const char *path, struct fuse_file_info *fi) | |
881 | { | |
882 | if (strcmp(path, "/") == 0) | |
883 | return 0; | |
884 | if (strncmp(path, "/cgroup", 7) == 0) { | |
885 | return cg_releasedir(path, fi); | |
886 | } | |
887 | return -EINVAL; | |
888 | } | |
889 | ||
99978832 SH |
890 | static int lxcfs_open(const char *path, struct fuse_file_info *fi) |
891 | { | |
892 | if (strncmp(path, "/cgroup", 7) == 0) { | |
893 | return cg_open(path, fi); | |
894 | } | |
895 | ||
896 | return -EINVAL; | |
897 | } | |
898 | ||
899 | static int lxcfs_read(const char *path, char *buf, size_t size, off_t offset, | |
900 | struct fuse_file_info *fi) | |
901 | { | |
902 | if (strncmp(path, "/cgroup", 7) == 0) { | |
903 | return cg_read(path, buf, size, offset, fi); | |
904 | } | |
905 | ||
906 | return -EINVAL; | |
907 | } | |
908 | ||
2ad6d2bd SH |
909 | int lxcfs_write(const char *path, const char *buf, size_t size, off_t offset, |
910 | struct fuse_file_info *fi) | |
911 | { | |
912 | if (strncmp(path, "/cgroup", 7) == 0) { | |
913 | return cg_write(path, buf, size, offset, fi); | |
914 | } | |
915 | ||
916 | return -EINVAL; | |
917 | } | |
918 | ||
99978832 SH |
919 | static int lxcfs_flush(const char *path, struct fuse_file_info *fi) |
920 | { | |
921 | return 0; | |
922 | } | |
923 | ||
924 | static int lxcfs_release(const char *path, struct fuse_file_info *fi) | |
758ad80c | 925 | { |
99978832 SH |
926 | return 0; |
927 | } | |
928 | ||
929 | static int lxcfs_fsync(const char *path, int datasync, struct fuse_file_info *fi) | |
930 | { | |
931 | return 0; | |
758ad80c SH |
932 | } |
933 | ||
ab54b798 SH |
934 | int lxcfs_mkdir(const char *path, mode_t mode) |
935 | { | |
936 | if (strncmp(path, "/cgroup", 7) == 0) | |
937 | return cg_mkdir(path, mode); | |
938 | ||
939 | return -EINVAL; | |
940 | } | |
941 | ||
341b21ad SH |
942 | int lxcfs_chown(const char *path, uid_t uid, gid_t gid) |
943 | { | |
944 | if (strncmp(path, "/cgroup", 7) == 0) | |
945 | return cg_chown(path, uid, gid); | |
946 | ||
947 | return -EINVAL; | |
948 | } | |
949 | ||
2ad6d2bd SH |
950 | /* |
951 | * cat first does a truncate before doing ops->write. This doesn't | |
952 | * really make sense for cgroups. So just return 0 always but do | |
953 | * nothing. | |
954 | */ | |
955 | int lxcfs_truncate(const char *path, off_t newsize) | |
956 | { | |
957 | if (strncmp(path, "/cgroup", 7) == 0) | |
958 | return 0; | |
959 | return -EINVAL; | |
960 | } | |
961 | ||
50d8d5b5 SH |
962 | int lxcfs_rmdir(const char *path) |
963 | { | |
964 | if (strncmp(path, "/cgroup", 7) == 0) | |
965 | return cg_rmdir(path); | |
966 | return -EINVAL; | |
967 | } | |
968 | ||
fd2e4e03 SH |
969 | int lxcfs_chmod(const char *path, mode_t mode) |
970 | { | |
971 | if (strncmp(path, "/cgroup", 7) == 0) | |
972 | return cg_chmod(path, mode); | |
973 | return -EINVAL; | |
974 | } | |
975 | ||
758ad80c SH |
976 | const struct fuse_operations lxcfs_ops = { |
977 | .getattr = lxcfs_getattr, | |
978 | .readlink = NULL, | |
979 | .getdir = NULL, | |
980 | .mknod = NULL, | |
ab54b798 | 981 | .mkdir = lxcfs_mkdir, |
758ad80c | 982 | .unlink = NULL, |
50d8d5b5 | 983 | .rmdir = lxcfs_rmdir, |
758ad80c SH |
984 | .symlink = NULL, |
985 | .rename = NULL, | |
986 | .link = NULL, | |
fd2e4e03 | 987 | .chmod = lxcfs_chmod, |
341b21ad | 988 | .chown = lxcfs_chown, |
2ad6d2bd | 989 | .truncate = lxcfs_truncate, |
758ad80c | 990 | .utime = NULL, |
99978832 SH |
991 | |
992 | .open = lxcfs_open, | |
993 | .read = lxcfs_read, | |
994 | .release = lxcfs_release, | |
2ad6d2bd | 995 | .write = lxcfs_write, |
99978832 | 996 | |
758ad80c | 997 | .statfs = NULL, |
99978832 SH |
998 | .flush = lxcfs_flush, |
999 | .fsync = lxcfs_fsync, | |
758ad80c SH |
1000 | |
1001 | .setxattr = NULL, | |
1002 | .getxattr = NULL, | |
1003 | .listxattr = NULL, | |
1004 | .removexattr = NULL, | |
1005 | ||
1006 | .opendir = lxcfs_opendir, | |
1007 | .readdir = lxcfs_readdir, | |
1008 | .releasedir = lxcfs_releasedir, | |
1009 | ||
1010 | .fsyncdir = NULL, | |
1011 | .init = NULL, | |
1012 | .destroy = NULL, | |
1013 | .access = NULL, | |
1014 | .create = NULL, | |
1015 | .ftruncate = NULL, | |
1016 | .fgetattr = NULL, | |
1017 | }; | |
1018 | ||
99978832 | 1019 | static void usage(const char *me) |
758ad80c SH |
1020 | { |
1021 | fprintf(stderr, "Usage:\n"); | |
1022 | fprintf(stderr, "\n"); | |
1023 | fprintf(stderr, "%s [FUSE and mount options] mountpoint\n", me); | |
1024 | exit(1); | |
1025 | } | |
1026 | ||
99978832 | 1027 | static bool is_help(char *w) |
758ad80c SH |
1028 | { |
1029 | if (strcmp(w, "-h") == 0 || | |
1030 | strcmp(w, "--help") == 0 || | |
1031 | strcmp(w, "-help") == 0 || | |
1032 | strcmp(w, "help") == 0) | |
1033 | return true; | |
1034 | return false; | |
1035 | } | |
1036 | ||
1037 | int main(int argc, char *argv[]) | |
1038 | { | |
1039 | int ret; | |
1040 | struct lxcfs_state *d; | |
1041 | ||
1042 | if (argc < 2 || is_help(argv[1])) | |
1043 | usage(argv[0]); | |
1044 | ||
1045 | d = malloc(sizeof(*d)); | |
1046 | if (!d) | |
1047 | return -1; | |
1048 | ||
1049 | if (!cgm_escape_cgroup()) | |
1050 | fprintf(stderr, "WARNING: failed to escape to root cgroup\n"); | |
1051 | ||
1052 | if (!cgm_get_controllers(&d->subsystems)) | |
1053 | return -1; | |
1054 | ||
1055 | ret = fuse_main(argc, argv, &lxcfs_ops, d); | |
1056 | ||
1057 | return ret; | |
2183082c | 1058 | } |