]>
Commit | Line | Data |
---|---|---|
758ad80c SH |
1 | /* lxcfs |
2 | * | |
3 | * Copyright © 2014 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
5 | * | |
f2799430 | 6 | * See COPYING file for details. |
758ad80c SH |
7 | */ |
8 | ||
9 | /* | |
10 | * NOTES - make sure to run this as -s to avoid threading. | |
11 | * TODO - can we enforce that here from the code? | |
12 | */ | |
13 | #define FUSE_USE_VERSION 26 | |
14 | ||
2183082c | 15 | #include <stdio.h> |
758ad80c SH |
16 | #include <dirent.h> |
17 | #include <fcntl.h> | |
18 | #include <fuse.h> | |
19 | #include <unistd.h> | |
20 | #include <errno.h> | |
21 | #include <stdbool.h> | |
22 | #include <time.h> | |
23 | #include <string.h> | |
24 | #include <stdlib.h> | |
25 | #include <libgen.h> | |
26 | ||
27 | #include <nih/alloc.h> | |
28 | #include <nih/string.h> | |
29 | ||
30 | #include "cgmanager.h" | |
31 | ||
32 | struct lxcfs_state { | |
33 | /* | |
34 | * a null-terminated, nih-allocated list of the mounted subsystems. We | |
35 | * detect this at startup. | |
36 | */ | |
37 | char **subsystems; | |
38 | }; | |
39 | #define LXCFS_DATA ((struct lxcfs_state *) fuse_get_context()->private_data) | |
40 | ||
053a659d SH |
41 | /* |
42 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
43 | * valid in the caller's namespace, return the id mapped into | |
44 | * pid's namespace. | |
45 | * Returns the mapped id, or -1 on error. | |
46 | */ | |
47 | unsigned int | |
48 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
49 | { | |
50 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
51 | hostuid, // base id for a range in the caller's namespace | |
52 | count; // number of ids in this range | |
53 | char line[400]; | |
54 | int ret; | |
55 | ||
56 | fseek(idfile, 0L, SEEK_SET); | |
57 | while (fgets(line, 400, idfile)) { | |
58 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
59 | if (ret != 3) | |
60 | continue; | |
61 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
62 | /* | |
63 | * uids wrapped around - unexpected as this is a procfile, | |
64 | * so just bail. | |
65 | */ | |
66 | fprintf(stderr, "pid wrapparound at entry %u %u %u in %s", | |
67 | nsuid, hostuid, count, line); | |
68 | return -1; | |
69 | } | |
70 | if (hostuid <= in_id && hostuid+count > in_id) { | |
71 | /* | |
72 | * now since hostuid <= in_id < hostuid+count, and | |
73 | * hostuid+count and nsuid+count do not wrap around, | |
74 | * we know that nsuid+(in_id-hostuid) which must be | |
75 | * less that nsuid+(count) must not wrap around | |
76 | */ | |
77 | return (in_id - hostuid) + nsuid; | |
78 | } | |
79 | } | |
80 | ||
81 | // no answer found | |
82 | return -1; | |
83 | } | |
84 | ||
341b21ad SH |
85 | /* |
86 | * for is_privileged_over, | |
87 | * specify whether we require the calling uid to be root in his | |
88 | * namespace | |
89 | */ | |
90 | #define NS_ROOT_REQD true | |
91 | #define NS_ROOT_OPT false | |
92 | ||
93 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
758ad80c | 94 | { |
053a659d SH |
95 | nih_local char *fpath = NULL; |
96 | bool answer = false; | |
97 | uid_t nsuid; | |
98 | ||
341b21ad SH |
99 | if (victim == -1 || uid == -1) |
100 | return false; | |
101 | ||
102 | /* | |
103 | * If the request is one not requiring root in the namespace, | |
104 | * then having the same uid suffices. (i.e. uid 1000 has write | |
105 | * access to files owned by uid 1000 | |
106 | */ | |
107 | if (!req_ns_root && uid == victim) | |
758ad80c SH |
108 | return true; |
109 | ||
053a659d SH |
110 | fpath = NIH_MUST( nih_sprintf(NULL, "/proc/%d/uid_map", pid) ); |
111 | FILE *f = fopen(fpath, "r"); | |
112 | if (!f) | |
113 | return false; | |
114 | ||
341b21ad | 115 | /* if caller's not root in his namespace, reject */ |
053a659d SH |
116 | nsuid = convert_id_to_ns(f, uid); |
117 | if (nsuid) | |
118 | goto out; | |
119 | ||
341b21ad SH |
120 | /* |
121 | * If victim is not mapped into caller's ns, reject. | |
122 | * XXX I'm not sure this check is needed given that fuse | |
123 | * will be sending requests where the vfs has converted | |
124 | */ | |
053a659d SH |
125 | nsuid = convert_id_to_ns(f, victim); |
126 | if (nsuid == -1) | |
127 | goto out; | |
128 | ||
129 | answer = true; | |
130 | ||
131 | out: | |
132 | fclose(f); | |
133 | return answer; | |
758ad80c SH |
134 | } |
135 | ||
136 | static bool perms_include(int fmode, mode_t req_mode) | |
137 | { | |
2ad6d2bd SH |
138 | mode_t r; |
139 | ||
140 | switch (req_mode & O_ACCMODE) { | |
141 | case O_RDONLY: | |
142 | r = S_IROTH; | |
143 | break; | |
144 | case O_WRONLY: | |
145 | r = S_IWOTH; | |
146 | break; | |
147 | case O_RDWR: | |
148 | r = S_IROTH | S_IWOTH; | |
149 | break; | |
150 | default: | |
151 | return false; | |
152 | } | |
153 | return ((fmode & r) == r); | |
758ad80c SH |
154 | } |
155 | ||
3db25a35 SH |
156 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) |
157 | { | |
158 | char *start, *end; | |
159 | ||
160 | if (strlen(taskcg) <= strlen(querycg)) { | |
161 | fprintf(stderr, "%s: I was fed bad input\n", __func__); | |
162 | return NULL; | |
163 | } | |
164 | ||
165 | if (strcmp(querycg, "/") == 0) | |
166 | start = NIH_MUST( nih_strdup(NULL, taskcg + 1) ); | |
167 | else | |
168 | start = NIH_MUST( nih_strdup(NULL, taskcg + strlen(querycg) + 1) ); | |
169 | end = strchr(start, '/'); | |
170 | if (end) | |
171 | *end = '\0'; | |
172 | return start; | |
173 | } | |
174 | ||
758ad80c SH |
175 | /* |
176 | * check whether a fuse context may access a cgroup dir or file | |
177 | * | |
178 | * If file is not null, it is a cgroup file to check under cg. | |
179 | * If file is null, then we are checking perms on cg itself. | |
180 | * | |
181 | * For files we can check the mode of the list_keys result. | |
182 | * For cgroups, we must make assumptions based on the files under the | |
183 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
184 | * yet. | |
185 | */ | |
186 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
187 | { | |
188 | nih_local struct cgm_keys **list = NULL; | |
189 | int i; | |
190 | ||
191 | if (!file) | |
192 | file = "tasks"; | |
193 | ||
194 | if (*file == '/') | |
195 | file++; | |
196 | ||
197 | if (!cgm_list_keys(contrl, cg, &list)) | |
198 | return false; | |
199 | for (i = 0; list[i]; i++) { | |
200 | if (strcmp(list[i]->name, file) == 0) { | |
201 | struct cgm_keys *k = list[i]; | |
341b21ad | 202 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { |
758ad80c SH |
203 | if (perms_include(k->mode >> 6, mode)) |
204 | return true; | |
205 | } | |
206 | if (fc->gid == k->gid) { | |
207 | if (perms_include(k->mode >> 3, mode)) | |
208 | return true; | |
209 | } | |
210 | return perms_include(k->mode, mode); | |
211 | } | |
212 | } | |
213 | ||
214 | return false; | |
215 | } | |
216 | ||
3db25a35 SH |
217 | static void stripnewline(char *x) |
218 | { | |
219 | size_t l = strlen(x); | |
220 | if (l && x[l-1] == '\n') | |
221 | x[l-1] = '\0'; | |
222 | } | |
223 | ||
224 | /* | |
225 | * If caller is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
226 | * If caller is in /a, he may act on /a/b, but not on /b. | |
227 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
228 | * to a nih_alloc'd string containing the next cgroup directory under cg | |
229 | */ | |
230 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
231 | { | |
232 | nih_local char *fnam = NULL; | |
233 | FILE *f; | |
234 | bool answer = false; | |
235 | char *line = NULL; | |
236 | size_t len = 0; | |
237 | ||
238 | fnam = NIH_MUST( nih_sprintf(NULL, "/proc/%d/cgroup", pid) ); | |
239 | if (!(f = fopen(fnam, "r"))) | |
240 | return false; | |
241 | ||
242 | while (getline(&line, &len, f) != -1) { | |
243 | char *c1, *c2, *linecmp; | |
244 | if (!line[0]) | |
245 | continue; | |
246 | c1 = strchr(line, ':'); | |
247 | if (!c1) | |
248 | goto out; | |
249 | c1++; | |
250 | c2 = strchr(c1, ':'); | |
251 | if (!c2) | |
252 | goto out; | |
253 | *c2 = '\0'; | |
254 | if (strcmp(c1, contrl) != 0) | |
255 | continue; | |
256 | c2++; | |
257 | stripnewline(c2); | |
258 | /* | |
259 | * callers pass in '/' for root cgroup, otherwise they pass | |
260 | * in a cgroup without leading '/' | |
261 | */ | |
262 | linecmp = *cg == '/' ? c2 : c2+1; | |
263 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { | |
264 | if (nextcg) | |
265 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
266 | goto out; | |
267 | } | |
268 | answer = true; | |
269 | goto out; | |
270 | } | |
271 | ||
272 | out: | |
273 | fclose(f); | |
274 | free(line); | |
275 | return answer; | |
276 | } | |
277 | ||
758ad80c SH |
278 | /* |
279 | * given /cgroup/freezer/a/b, return "freezer". this will be nih-allocated | |
280 | * and needs to be nih_freed. | |
281 | */ | |
282 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
283 | { | |
284 | const char *p1; | |
285 | char *ret, *slash; | |
286 | ||
287 | if (strlen(path) < 9) | |
288 | return NULL; | |
289 | p1 = path+8; | |
290 | ret = nih_strdup(NULL, p1); | |
291 | if (!ret) | |
292 | return ret; | |
293 | slash = strstr(ret, "/"); | |
294 | if (slash) | |
295 | *slash = '\0'; | |
296 | ||
297 | /* verify that it is a subsystem */ | |
298 | char **list = LXCFS_DATA ? LXCFS_DATA->subsystems : NULL; | |
299 | int i; | |
300 | if (!list) { | |
301 | nih_free(ret); | |
302 | return NULL; | |
303 | } | |
304 | for (i = 0; list[i]; i++) { | |
305 | if (strcmp(list[i], ret) == 0) | |
306 | return ret; | |
307 | } | |
308 | nih_free(ret); | |
309 | return NULL; | |
310 | } | |
311 | ||
312 | /* | |
313 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
314 | * Note that the returned value may include files (keynames) etc | |
315 | */ | |
316 | static const char *find_cgroup_in_path(const char *path) | |
317 | { | |
318 | const char *p1; | |
319 | ||
320 | if (strlen(path) < 9) | |
321 | return NULL; | |
322 | p1 = strstr(path+8, "/"); | |
323 | if (!p1) | |
324 | return NULL; | |
325 | return p1+1; | |
326 | } | |
327 | ||
328 | static bool is_child_cgroup(const char *contr, const char *dir, const char *f) | |
329 | { | |
330 | nih_local char **list = NULL; | |
331 | int i; | |
332 | ||
333 | if (!f) | |
334 | return false; | |
335 | if (*f == '/') | |
336 | f++; | |
337 | ||
338 | if (!cgm_list_children(contr, dir, &list)) | |
339 | return false; | |
340 | for (i = 0; list[i]; i++) { | |
341 | if (strcmp(list[i], f) == 0) | |
342 | return true; | |
343 | } | |
344 | ||
345 | return false; | |
346 | } | |
347 | ||
348 | static struct cgm_keys *get_cgroup_key(const char *contr, const char *dir, const char *f) | |
349 | { | |
350 | nih_local struct cgm_keys **list = NULL; | |
351 | struct cgm_keys *k; | |
352 | int i; | |
353 | ||
354 | if (!f) | |
355 | return NULL; | |
356 | if (*f == '/') | |
357 | f++; | |
358 | if (!cgm_list_keys(contr, dir, &list)) | |
359 | return NULL; | |
360 | for (i = 0; list[i]; i++) { | |
361 | if (strcmp(list[i]->name, f) == 0) { | |
362 | k = NIH_MUST( nih_alloc(NULL, (sizeof(*k))) ); | |
363 | k->name = NIH_MUST( nih_strdup(k, list[i]->name) ); | |
364 | k->uid = list[i]->uid; | |
365 | k->gid = list[i]->gid; | |
366 | k->mode = list[i]->mode; | |
367 | return k; | |
368 | } | |
369 | } | |
370 | ||
371 | return NULL; | |
372 | } | |
373 | ||
374 | static void get_cgdir_and_path(const char *cg, char **dir, char **file) | |
375 | { | |
758ad80c SH |
376 | char *p; |
377 | ||
378 | *dir = NIH_MUST( nih_strdup(NULL, cg) ); | |
379 | *file = strrchr(cg, '/'); | |
380 | if (!*file) { | |
381 | *file = NULL; | |
382 | return; | |
383 | } | |
384 | p = strrchr(*dir, '/'); | |
385 | *p = '\0'; | |
386 | } | |
387 | ||
99978832 SH |
388 | static size_t get_file_size(const char *contrl, const char *cg, const char *f) |
389 | { | |
390 | nih_local char *data = NULL; | |
391 | size_t s; | |
392 | if (!cgm_get_value(contrl, cg, f, &data)) | |
393 | return -EINVAL; | |
394 | s = strlen(data); | |
395 | return s; | |
396 | } | |
2ad6d2bd | 397 | |
758ad80c | 398 | /* |
2ad6d2bd | 399 | * FUSE ops for /cgroup |
758ad80c | 400 | */ |
2ad6d2bd | 401 | |
758ad80c SH |
402 | static int cg_getattr(const char *path, struct stat *sb) |
403 | { | |
404 | struct timespec now; | |
405 | struct fuse_context *fc = fuse_get_context(); | |
406 | nih_local char * cgdir = NULL; | |
407 | char *fpath = NULL, *path1, *path2; | |
408 | nih_local struct cgm_keys *k = NULL; | |
409 | const char *cgroup; | |
410 | nih_local char *controller = NULL; | |
411 | ||
412 | ||
413 | if (!fc) | |
414 | return -EIO; | |
415 | ||
416 | memset(sb, 0, sizeof(struct stat)); | |
417 | ||
418 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
419 | return -EINVAL; | |
420 | ||
421 | sb->st_uid = sb->st_gid = 0; | |
422 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
423 | sb->st_size = 0; | |
424 | ||
425 | if (strcmp(path, "/cgroup") == 0) { | |
426 | sb->st_mode = S_IFDIR | 00755; | |
427 | sb->st_nlink = 2; | |
428 | return 0; | |
429 | } | |
430 | ||
431 | controller = pick_controller_from_path(fc, path); | |
432 | if (!controller) | |
433 | return -EIO; | |
758ad80c SH |
434 | cgroup = find_cgroup_in_path(path); |
435 | if (!cgroup) { | |
436 | /* this is just /cgroup/controller, return it as a dir */ | |
437 | sb->st_mode = S_IFDIR | 00755; | |
438 | sb->st_nlink = 2; | |
439 | return 0; | |
440 | } | |
341b21ad | 441 | |
758ad80c SH |
442 | get_cgdir_and_path(cgroup, &cgdir, &fpath); |
443 | ||
444 | if (!fpath) { | |
445 | path1 = "/"; | |
446 | path2 = cgdir; | |
447 | } else { | |
448 | path1 = cgdir; | |
449 | path2 = fpath; | |
450 | } | |
451 | ||
758ad80c SH |
452 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. |
453 | * Then check that caller's cgroup is under path if fpath is a child | |
454 | * cgroup, or cgdir if fpath is a file */ | |
455 | ||
456 | if (is_child_cgroup(controller, path1, path2)) { | |
f9a05025 SH |
457 | if (!caller_is_in_ancestor(fc->pid, controller, cgroup, NULL)) { |
458 | /* this is just /cgroup/controller, return it as a dir */ | |
459 | sb->st_mode = S_IFDIR | 00555; | |
460 | sb->st_nlink = 2; | |
461 | return 0; | |
462 | } | |
758ad80c | 463 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) |
f9a05025 | 464 | return -EACCES; |
758ad80c | 465 | |
053a659d SH |
466 | // get uid, gid, from '/tasks' file and make up a mode |
467 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
468 | sb->st_mode = S_IFDIR | 00755; | |
469 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
470 | if (!k) { | |
053a659d SH |
471 | sb->st_uid = sb->st_gid = 0; |
472 | } else { | |
053a659d SH |
473 | sb->st_uid = k->uid; |
474 | sb->st_gid = k->gid; | |
475 | } | |
758ad80c SH |
476 | sb->st_nlink = 2; |
477 | return 0; | |
478 | } | |
479 | ||
480 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
3db25a35 SH |
481 | if (!caller_is_in_ancestor(fc->pid, controller, path1, NULL)) |
482 | return -ENOENT; | |
758ad80c | 483 | if (!fc_may_access(fc, controller, path1, path2, O_RDONLY)) |
f9a05025 | 484 | return -EACCES; |
758ad80c | 485 | |
758ad80c | 486 | sb->st_mode = S_IFREG | k->mode; |
053a659d | 487 | sb->st_nlink = 1; |
758ad80c SH |
488 | sb->st_uid = k->uid; |
489 | sb->st_gid = k->gid; | |
99978832 | 490 | sb->st_size = get_file_size(controller, path1, path2); |
758ad80c SH |
491 | return 0; |
492 | } | |
493 | ||
ab54b798 | 494 | return -ENOENT; |
758ad80c | 495 | } |
2183082c | 496 | |
7f163b71 SH |
497 | /* |
498 | * TODO - cache these results in a table for use in opendir, free | |
499 | * in releasedir | |
500 | */ | |
758ad80c | 501 | static int cg_opendir(const char *path, struct fuse_file_info *fi) |
2183082c | 502 | { |
7f163b71 SH |
503 | struct fuse_context *fc = fuse_get_context(); |
504 | nih_local struct cgm_keys **list = NULL; | |
505 | const char *cgroup; | |
506 | nih_local char *controller = NULL; | |
7f163b71 SH |
507 | nih_local char *nextcg = NULL; |
508 | ||
509 | if (!fc) | |
510 | return -EIO; | |
511 | ||
512 | if (strcmp(path, "/cgroup") == 0) | |
513 | return 0; | |
514 | ||
515 | // return list of keys for the controller, and list of child cgroups | |
516 | controller = pick_controller_from_path(fc, path); | |
517 | if (!controller) | |
518 | return -EIO; | |
519 | ||
520 | cgroup = find_cgroup_in_path(path); | |
521 | if (!cgroup) { | |
522 | /* this is just /cgroup/controller, return its contents */ | |
523 | cgroup = "/"; | |
524 | } | |
525 | ||
526 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
527 | return -EACCES; | |
758ad80c SH |
528 | return 0; |
529 | } | |
530 | ||
758ad80c SH |
531 | static int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, |
532 | struct fuse_file_info *fi) | |
533 | { | |
534 | struct fuse_context *fc = fuse_get_context(); | |
535 | ||
536 | if (!fc) | |
537 | return -EIO; | |
538 | ||
539 | if (strcmp(path, "/cgroup") == 0) { | |
540 | // get list of controllers | |
541 | char **list = LXCFS_DATA ? LXCFS_DATA->subsystems : NULL; | |
542 | int i; | |
543 | ||
544 | if (!list) | |
545 | return -EIO; | |
7f163b71 | 546 | |
758ad80c SH |
547 | for (i = 0; list[i]; i++) { |
548 | if (filler(buf, list[i], NULL, 0) != 0) { | |
549 | return -EIO; | |
550 | } | |
551 | } | |
552 | return 0; | |
553 | } | |
554 | ||
555 | // return list of keys for the controller, and list of child cgroups | |
556 | nih_local struct cgm_keys **list = NULL; | |
557 | const char *cgroup; | |
558 | nih_local char *controller = NULL; | |
559 | int i; | |
3db25a35 | 560 | nih_local char *nextcg = NULL; |
758ad80c SH |
561 | |
562 | controller = pick_controller_from_path(fc, path); | |
563 | if (!controller) | |
564 | return -EIO; | |
565 | ||
566 | cgroup = find_cgroup_in_path(path); | |
567 | if (!cgroup) { | |
568 | /* this is just /cgroup/controller, return its contents */ | |
569 | cgroup = "/"; | |
570 | } | |
571 | ||
572 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
f9a05025 | 573 | return -EACCES; |
758ad80c SH |
574 | |
575 | if (!cgm_list_keys(controller, cgroup, &list)) | |
3db25a35 | 576 | // not a valid cgroup |
758ad80c | 577 | return -EINVAL; |
3db25a35 SH |
578 | |
579 | if (!caller_is_in_ancestor(fc->pid, controller, cgroup, &nextcg)) { | |
580 | if (nextcg) { | |
581 | int ret; | |
582 | ret = filler(buf, nextcg, NULL, 0); | |
583 | if (ret != 0) | |
584 | return -EIO; | |
585 | } | |
586 | return 0; | |
587 | } | |
588 | ||
758ad80c | 589 | for (i = 0; list[i]; i++) { |
758ad80c SH |
590 | if (filler(buf, list[i]->name, NULL, 0) != 0) { |
591 | return -EIO; | |
592 | } | |
593 | } | |
594 | ||
595 | // now get the list of child cgroups | |
596 | nih_local char **clist; | |
597 | ||
598 | if (!cgm_list_children(controller, cgroup, &clist)) | |
599 | return 0; | |
600 | for (i = 0; clist[i]; i++) { | |
758ad80c SH |
601 | if (filler(buf, clist[i], NULL, 0) != 0) { |
602 | return -EIO; | |
603 | } | |
604 | } | |
605 | return 0; | |
606 | } | |
607 | ||
608 | static int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
609 | { | |
610 | return 0; | |
611 | } | |
612 | ||
26faa701 SH |
613 | /* |
614 | * TODO - cache info here for read/write, release in cg_release. | |
615 | */ | |
99978832 SH |
616 | static int cg_open(const char *path, struct fuse_file_info *fi) |
617 | { | |
618 | nih_local char *controller = NULL; | |
619 | const char *cgroup; | |
620 | char *fpath = NULL, *path1, *path2; | |
621 | nih_local char * cgdir = NULL; | |
622 | nih_local struct cgm_keys *k = NULL; | |
623 | struct fuse_context *fc = fuse_get_context(); | |
624 | ||
625 | if (!fc) | |
626 | return -EIO; | |
627 | ||
628 | controller = pick_controller_from_path(fc, path); | |
629 | if (!controller) | |
630 | return -EIO; | |
631 | cgroup = find_cgroup_in_path(path); | |
632 | if (!cgroup) | |
633 | return -EINVAL; | |
634 | ||
635 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
636 | if (!fpath) { | |
637 | path1 = "/"; | |
638 | path2 = cgdir; | |
639 | } else { | |
640 | path1 = cgdir; | |
641 | path2 = fpath; | |
642 | } | |
643 | ||
644 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
645 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) | |
f9a05025 SH |
646 | // should never get here |
647 | return -EACCES; | |
99978832 | 648 | |
99978832 SH |
649 | return 0; |
650 | } | |
651 | ||
652 | return -EINVAL; | |
653 | } | |
654 | ||
655 | static int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
656 | struct fuse_file_info *fi) | |
657 | { | |
658 | nih_local char *controller = NULL; | |
659 | const char *cgroup; | |
660 | char *fpath = NULL, *path1, *path2; | |
661 | struct fuse_context *fc = fuse_get_context(); | |
662 | nih_local char * cgdir = NULL; | |
663 | nih_local struct cgm_keys *k = NULL; | |
664 | ||
665 | if (offset) | |
666 | return -EIO; | |
667 | ||
668 | if (!fc) | |
669 | return -EIO; | |
670 | ||
671 | controller = pick_controller_from_path(fc, path); | |
672 | if (!controller) | |
f9a05025 | 673 | return -EINVAL; |
99978832 SH |
674 | cgroup = find_cgroup_in_path(path); |
675 | if (!cgroup) | |
676 | return -EINVAL; | |
677 | ||
678 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
679 | if (!fpath) { | |
680 | path1 = "/"; | |
681 | path2 = cgdir; | |
682 | } else { | |
683 | path1 = cgdir; | |
684 | path2 = fpath; | |
685 | } | |
686 | ||
687 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
688 | nih_local char *data = NULL; | |
689 | int s; | |
690 | ||
2ad6d2bd | 691 | if (!fc_may_access(fc, controller, path1, path2, O_RDONLY)) |
f9a05025 SH |
692 | // should never get here |
693 | return -EACCES; | |
99978832 SH |
694 | |
695 | if (!cgm_get_value(controller, path1, path2, &data)) | |
696 | return -EINVAL; | |
697 | ||
698 | s = strlen(data); | |
699 | if (s > size) | |
700 | s = size; | |
701 | memcpy(buf, data, s); | |
702 | ||
99978832 SH |
703 | return s; |
704 | } | |
705 | ||
706 | return -EINVAL; | |
707 | } | |
708 | ||
2ad6d2bd SH |
709 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, |
710 | struct fuse_file_info *fi) | |
711 | { | |
712 | nih_local char *controller = NULL; | |
713 | const char *cgroup; | |
714 | char *fpath = NULL, *path1, *path2; | |
715 | struct fuse_context *fc = fuse_get_context(); | |
716 | nih_local char * cgdir = NULL; | |
717 | nih_local struct cgm_keys *k = NULL; | |
718 | ||
2ad6d2bd | 719 | if (offset) |
f9a05025 | 720 | return -EINVAL; |
2ad6d2bd SH |
721 | |
722 | if (!fc) | |
723 | return -EIO; | |
724 | ||
725 | controller = pick_controller_from_path(fc, path); | |
726 | if (!controller) | |
f9a05025 | 727 | return -EINVAL; |
2ad6d2bd SH |
728 | cgroup = find_cgroup_in_path(path); |
729 | if (!cgroup) | |
730 | return -EINVAL; | |
731 | ||
732 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
733 | if (!fpath) { | |
734 | path1 = "/"; | |
735 | path2 = cgdir; | |
736 | } else { | |
737 | path1 = cgdir; | |
738 | path2 = fpath; | |
739 | } | |
740 | ||
741 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
742 | if (!fc_may_access(fc, controller, path1, path2, O_WRONLY)) | |
f9a05025 | 743 | return -EACCES; |
2ad6d2bd SH |
744 | |
745 | if (!cgm_set_value(controller, path1, path2, buf)) | |
746 | return -EINVAL; | |
747 | ||
748 | return size; | |
749 | } | |
750 | ||
751 | return -EINVAL; | |
752 | } | |
753 | ||
341b21ad SH |
754 | int cg_chown(const char *path, uid_t uid, gid_t gid) |
755 | { | |
756 | struct fuse_context *fc = fuse_get_context(); | |
757 | nih_local char * cgdir = NULL; | |
758 | char *fpath = NULL, *path1, *path2; | |
759 | nih_local struct cgm_keys *k = NULL; | |
760 | const char *cgroup; | |
761 | nih_local char *controller = NULL; | |
762 | ||
763 | ||
764 | if (!fc) | |
765 | return -EIO; | |
766 | ||
767 | if (strcmp(path, "/cgroup") == 0) | |
768 | return -EINVAL; | |
769 | ||
770 | controller = pick_controller_from_path(fc, path); | |
771 | if (!controller) | |
f9a05025 | 772 | return -EINVAL; |
341b21ad SH |
773 | cgroup = find_cgroup_in_path(path); |
774 | if (!cgroup) | |
775 | /* this is just /cgroup/controller */ | |
776 | return -EINVAL; | |
777 | ||
778 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
779 | ||
780 | if (!fpath) { | |
781 | path1 = "/"; | |
782 | path2 = cgdir; | |
783 | } else { | |
784 | path1 = cgdir; | |
785 | path2 = fpath; | |
786 | } | |
787 | ||
788 | if (is_child_cgroup(controller, path1, path2)) { | |
789 | // get uid, gid, from '/tasks' file and make up a mode | |
790 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
791 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
792 | ||
793 | } else | |
794 | k = get_cgroup_key(controller, path1, path2); | |
795 | ||
796 | if (!k) | |
797 | return -EINVAL; | |
798 | ||
799 | /* | |
800 | * This being a fuse request, the uid and gid must be valid | |
801 | * in the caller's namespace. So we can just check to make | |
802 | * sure that the caller is root in his uid, and privileged | |
803 | * over the file's current owner. | |
804 | */ | |
805 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) | |
f9a05025 | 806 | return -EACCES; |
341b21ad SH |
807 | |
808 | if (!cgm_chown_file(controller, cgroup, uid, gid)) | |
809 | return -EINVAL; | |
810 | return 0; | |
811 | } | |
2ad6d2bd | 812 | |
fd2e4e03 SH |
813 | int cg_chmod(const char *path, mode_t mode) |
814 | { | |
0a1bb5ea SH |
815 | struct fuse_context *fc = fuse_get_context(); |
816 | nih_local char * cgdir = NULL; | |
817 | char *fpath = NULL, *path1, *path2; | |
818 | nih_local struct cgm_keys *k = NULL; | |
819 | const char *cgroup; | |
820 | nih_local char *controller = NULL; | |
821 | ||
822 | if (!fc) | |
823 | return -EIO; | |
824 | ||
825 | if (strcmp(path, "/cgroup") == 0) | |
826 | return -EINVAL; | |
827 | ||
828 | controller = pick_controller_from_path(fc, path); | |
829 | if (!controller) | |
f9a05025 | 830 | return -EINVAL; |
0a1bb5ea SH |
831 | cgroup = find_cgroup_in_path(path); |
832 | if (!cgroup) | |
833 | /* this is just /cgroup/controller */ | |
834 | return -EINVAL; | |
835 | ||
836 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
837 | ||
838 | if (!fpath) { | |
839 | path1 = "/"; | |
840 | path2 = cgdir; | |
841 | } else { | |
842 | path1 = cgdir; | |
843 | path2 = fpath; | |
844 | } | |
845 | ||
846 | if (is_child_cgroup(controller, path1, path2)) { | |
847 | // get uid, gid, from '/tasks' file and make up a mode | |
848 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
849 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
850 | ||
851 | } else | |
852 | k = get_cgroup_key(controller, path1, path2); | |
853 | ||
854 | if (!k) | |
855 | return -EINVAL; | |
856 | ||
857 | /* | |
858 | * This being a fuse request, the uid and gid must be valid | |
859 | * in the caller's namespace. So we can just check to make | |
860 | * sure that the caller is root in his uid, and privileged | |
861 | * over the file's current owner. | |
862 | */ | |
863 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) | |
864 | return -EPERM; | |
865 | ||
866 | if (!cgm_chmod_file(controller, cgroup, mode)) | |
867 | return -EINVAL; | |
868 | return 0; | |
fd2e4e03 SH |
869 | } |
870 | ||
ab54b798 SH |
871 | int cg_mkdir(const char *path, mode_t mode) |
872 | { | |
873 | struct fuse_context *fc = fuse_get_context(); | |
874 | nih_local struct cgm_keys **list = NULL; | |
875 | char *fpath = NULL, *path1; | |
876 | nih_local char * cgdir = NULL; | |
877 | const char *cgroup; | |
878 | nih_local char *controller = NULL; | |
879 | ||
ab54b798 SH |
880 | if (!fc) |
881 | return -EIO; | |
882 | ||
883 | ||
884 | controller = pick_controller_from_path(fc, path); | |
885 | if (!controller) | |
f9a05025 | 886 | return -EINVAL; |
ab54b798 SH |
887 | |
888 | cgroup = find_cgroup_in_path(path); | |
889 | if (!cgroup) | |
f9a05025 | 890 | return -EINVAL; |
ab54b798 SH |
891 | |
892 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
893 | if (!fpath) | |
894 | path1 = "/"; | |
895 | else | |
896 | path1 = cgdir; | |
897 | ||
898 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) | |
f9a05025 | 899 | return -EACCES; |
ab54b798 SH |
900 | |
901 | ||
902 | if (!cgm_create(controller, cgroup, fc->uid, fc->gid)) | |
903 | return -EINVAL; | |
904 | ||
905 | return 0; | |
906 | } | |
907 | ||
50d8d5b5 SH |
908 | static int cg_rmdir(const char *path) |
909 | { | |
910 | struct fuse_context *fc = fuse_get_context(); | |
911 | nih_local struct cgm_keys **list = NULL; | |
912 | char *fpath = NULL; | |
913 | nih_local char * cgdir = NULL; | |
914 | const char *cgroup; | |
915 | nih_local char *controller = NULL; | |
916 | ||
917 | if (!fc) | |
918 | return -EIO; | |
919 | ||
920 | ||
921 | controller = pick_controller_from_path(fc, path); | |
922 | if (!controller) | |
f9a05025 | 923 | return -EINVAL; |
50d8d5b5 SH |
924 | |
925 | cgroup = find_cgroup_in_path(path); | |
926 | if (!cgroup) | |
f9a05025 | 927 | return -EINVAL; |
50d8d5b5 SH |
928 | |
929 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
930 | if (!fpath) | |
931 | return -EINVAL; | |
932 | ||
933 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) | |
f9a05025 | 934 | return -EACCES; |
50d8d5b5 SH |
935 | |
936 | if (!cgm_remove(controller, cgroup)) | |
937 | return -EINVAL; | |
938 | ||
939 | return 0; | |
940 | } | |
941 | ||
2dc17609 SH |
942 | static bool startswith(const char *line, const char *pref) |
943 | { | |
944 | if (strncmp(line, pref, strlen(pref)) == 0) | |
945 | return true; | |
946 | return false; | |
947 | } | |
948 | ||
949 | static void get_mem_cached(char *memstat, unsigned long *v) | |
950 | { | |
951 | char *eol; | |
952 | ||
953 | *v = 0; | |
954 | while (*memstat) { | |
955 | if (startswith(memstat, "total_cache")) { | |
956 | sscanf(memstat + 11, "%lu", v); | |
957 | *v /= 1024; | |
958 | return; | |
959 | } | |
960 | eol = strchr(memstat, '\n'); | |
961 | if (!eol) | |
962 | return; | |
963 | memstat = eol+1; | |
964 | } | |
965 | } | |
966 | ||
967 | static char *get_pid_cgroup(pid_t pid, const char *contrl) | |
968 | { | |
969 | nih_local char *fnam = NULL; | |
970 | FILE *f; | |
971 | char *answer = NULL; | |
972 | char *line = NULL; | |
973 | size_t len = 0; | |
974 | ||
975 | fnam = NIH_MUST( nih_sprintf(NULL, "/proc/%d/cgroup", pid) ); | |
976 | if (!(f = fopen(fnam, "r"))) | |
977 | return false; | |
978 | ||
979 | while (getline(&line, &len, f) != -1) { | |
980 | char *c1, *c2; | |
981 | if (!line[0]) | |
982 | continue; | |
983 | c1 = strchr(line, ':'); | |
984 | if (!c1) | |
985 | goto out; | |
986 | c1++; | |
987 | c2 = strchr(c1, ':'); | |
988 | if (!c2) | |
989 | goto out; | |
990 | *c2 = '\0'; | |
991 | if (strcmp(c1, contrl) != 0) | |
992 | continue; | |
993 | c2++; | |
994 | stripnewline(c2); | |
995 | answer = NIH_MUST( nih_strdup(NULL, c2) ); | |
996 | goto out; | |
997 | } | |
998 | ||
999 | out: | |
1000 | fclose(f); | |
1001 | free(line); | |
1002 | return answer; | |
1003 | } | |
1004 | ||
758ad80c | 1005 | /* |
2ad6d2bd | 1006 | * FUSE ops for /proc |
758ad80c | 1007 | */ |
758ad80c | 1008 | |
23ce2127 SH |
1009 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, |
1010 | struct fuse_file_info *fi) | |
1011 | { | |
2dc17609 SH |
1012 | struct fuse_context *fc = fuse_get_context(); |
1013 | nih_local char *cg = get_pid_cgroup(fc->pid, "memory"); | |
1014 | nih_local char *memlimit_str = NULL, *memusage_str = NULL, *memstat_str = NULL; | |
1015 | unsigned long memlimit = 0, memusage = 0, cached = 0, hosttotal = 0; | |
1016 | char *line = NULL; | |
1017 | size_t linelen = 0, total_len = 0; | |
1018 | FILE *f; | |
1019 | ||
1020 | if (offset) | |
1021 | return -EINVAL; | |
1022 | ||
1023 | if (!cg) | |
1024 | return 0; | |
1025 | ||
1026 | if (!cgm_get_value("memory", cg, "memory.limit_in_bytes", &memlimit_str)) | |
1027 | return 0; | |
1028 | if (!cgm_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
1029 | return 0; | |
1030 | if (!cgm_get_value("memory", cg, "memory.stat", &memstat_str)) | |
1031 | return 0; | |
1032 | memlimit = strtoul(memlimit_str, NULL, 10); | |
1033 | memusage = strtoul(memusage_str, NULL, 10); | |
1034 | memlimit /= 1024; | |
1035 | memusage /= 1024; | |
1036 | get_mem_cached(memstat_str, &cached); | |
1037 | ||
1038 | f = fopen("/proc/meminfo", "r"); | |
1039 | if (!f) | |
1040 | return 0; | |
1041 | ||
1042 | while (getline(&line, &linelen, f) != -1) { | |
1043 | size_t l; | |
1044 | char *printme, lbuf[100]; | |
1045 | ||
1046 | memset(lbuf, 0, 100); | |
1047 | if (startswith(line, "MemTotal:")) { | |
1048 | sscanf(line+14, "%lu", &hosttotal); | |
1049 | if (hosttotal < memlimit) | |
1050 | memlimit = hosttotal; | |
1051 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
1052 | printme = lbuf; | |
1053 | } else if (startswith(line, "MemFree:")) { | |
1054 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
1055 | printme = lbuf; | |
1056 | } else if (startswith(line, "MemAvailable:")) { | |
1057 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage); | |
1058 | printme = lbuf; | |
1059 | } else if (startswith(line, "Buffers:")) { | |
1060 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
1061 | printme = lbuf; | |
1062 | } else if (startswith(line, "Cached:")) { | |
1063 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
1064 | printme = lbuf; | |
1065 | } else if (startswith(line, "SwapCached:")) { | |
1066 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
1067 | printme = lbuf; | |
1068 | } else | |
1069 | printme = line; | |
1070 | l = snprintf(buf, size, "%s", printme); | |
1071 | buf += l; | |
1072 | size -= l; | |
1073 | total_len += l; | |
1074 | } | |
1075 | ||
1076 | return total_len; | |
23ce2127 SH |
1077 | } |
1078 | ||
1079 | /* | |
1080 | * Read the cpuset.cpus for cg | |
1081 | * Return the answer in a nih_alloced string | |
1082 | */ | |
1083 | static char *get_cpuset(const char *cg) | |
1084 | { | |
1085 | char *answer; | |
1086 | ||
1087 | if (!cgm_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
1088 | return NULL; | |
1089 | return answer; | |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * Helper functions for cpuset_in-set | |
1094 | */ | |
1095 | char *cpuset_nexttok(const char *c) | |
1096 | { | |
1097 | char *r = strchr(c+1, ','); | |
1098 | if (r) | |
1099 | return r+1; | |
1100 | return NULL; | |
1101 | } | |
1102 | ||
1103 | int cpuset_getrange(const char *c, int *a, int *b) | |
1104 | { | |
1105 | int ret; | |
1106 | ||
1107 | ret = sscanf(c, "%d-%d", a, b); | |
1108 | return ret; | |
1109 | } | |
1110 | ||
1111 | /* | |
1112 | * cpusets are in format "1,2-3,4" | |
1113 | * iow, comma-delimited ranges | |
1114 | */ | |
aeb56147 | 1115 | static bool cpu_in_cpuset(int cpu, const char *cpuset) |
23ce2127 | 1116 | { |
23ce2127 SH |
1117 | const char *c; |
1118 | ||
23ce2127 SH |
1119 | for (c = cpuset; c; c = cpuset_nexttok(c)) { |
1120 | int a, b, ret; | |
1121 | ||
1122 | ret = cpuset_getrange(c, &a, &b); | |
1123 | if (ret == 1 && cpu == a) | |
1124 | return true; | |
1125 | if (ret != 2) // bad cpuset! | |
1126 | return false; | |
1127 | if (cpu >= a && cpu <= b) | |
1128 | return true; | |
1129 | } | |
1130 | ||
1131 | return false; | |
1132 | } | |
1133 | ||
aeb56147 SH |
1134 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) |
1135 | { | |
1136 | int cpu; | |
1137 | ||
1138 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
1139 | return false; | |
1140 | return cpu_in_cpuset(cpu, cpuset); | |
1141 | } | |
1142 | ||
23ce2127 SH |
1143 | /* |
1144 | * check whether this is a '^processor" line in /proc/cpuinfo | |
1145 | */ | |
1146 | static bool is_processor_line(const char *line) | |
1147 | { | |
1148 | int cpu; | |
1149 | ||
1150 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
1151 | return true; | |
1152 | return false; | |
1153 | } | |
1154 | ||
23ce2127 SH |
1155 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, |
1156 | struct fuse_file_info *fi) | |
1157 | { | |
1158 | struct fuse_context *fc = fuse_get_context(); | |
1159 | nih_local char *cg = get_pid_cgroup(fc->pid, "cpuset"); | |
1160 | nih_local char *cpuset = NULL; | |
1161 | char *line = NULL; | |
1162 | size_t linelen = 0, total_len = 0; | |
1163 | bool am_printing = false; | |
1164 | int curcpu = -1; | |
1165 | FILE *f; | |
1166 | ||
1167 | if (offset) | |
1168 | return -EINVAL; | |
1169 | ||
1170 | if (!cg) | |
1171 | return 0; | |
1172 | ||
1173 | cpuset = get_cpuset(cg); | |
1174 | if (!cpuset) | |
1175 | return 0; | |
1176 | ||
1177 | f = fopen("/proc/cpuinfo", "r"); | |
1178 | if (!f) | |
1179 | return 0; | |
1180 | ||
1181 | while (getline(&line, &linelen, f) != -1) { | |
1182 | size_t l; | |
1183 | if (is_processor_line(line)) { | |
aeb56147 | 1184 | am_printing = cpuline_in_cpuset(line, cpuset); |
23ce2127 SH |
1185 | if (am_printing) { |
1186 | curcpu ++; | |
1187 | l = snprintf(buf, size, "processor : %d\n", curcpu); | |
1188 | buf += l; | |
1189 | size -= l; | |
1190 | total_len += l; | |
1191 | } | |
1192 | continue; | |
1193 | } | |
1194 | if (am_printing) { | |
1195 | l = snprintf(buf, size, "%s", line); | |
1196 | buf += l; | |
1197 | size -= l; | |
1198 | total_len += l; | |
1199 | } | |
1200 | } | |
1201 | ||
1202 | return total_len; | |
1203 | } | |
1204 | ||
1205 | static int proc_stat_read(char *buf, size_t size, off_t offset, | |
1206 | struct fuse_file_info *fi) | |
1207 | { | |
aeb56147 SH |
1208 | struct fuse_context *fc = fuse_get_context(); |
1209 | nih_local char *cg = get_pid_cgroup(fc->pid, "cpuset"); | |
1210 | nih_local char *cpuset = NULL; | |
1211 | char *line = NULL; | |
1212 | size_t linelen = 0, total_len = 0; | |
1213 | int curcpu = 0; | |
1214 | FILE *f; | |
1215 | ||
1216 | if (offset) | |
1217 | return -EINVAL; | |
1218 | ||
1219 | if (!cg) | |
1220 | return 0; | |
1221 | ||
1222 | cpuset = get_cpuset(cg); | |
1223 | if (!cpuset) | |
1224 | return 0; | |
1225 | ||
1226 | f = fopen("/proc/stat", "r"); | |
1227 | if (!f) | |
1228 | return 0; | |
1229 | ||
1230 | while (getline(&line, &linelen, f) != -1) { | |
1231 | size_t l; | |
1232 | int cpu; | |
1233 | char *c; | |
1234 | ||
1235 | if (sscanf(line, "cpu%d", &cpu) != 1) { | |
1236 | /* not a ^cpu line, just print it */ | |
1237 | l = snprintf(buf, size, "%s", line); | |
1238 | buf += l; | |
1239 | size -= l; | |
1240 | total_len += l; | |
1241 | continue; | |
1242 | } | |
1243 | if (!cpu_in_cpuset(cpu, cpuset)) | |
1244 | continue; | |
1245 | curcpu ++; | |
1246 | ||
1247 | c = strchr(line, ' '); | |
1248 | if (!c) | |
1249 | continue; | |
1250 | l = snprintf(buf, size, "cpu%d %s", curcpu, c); | |
1251 | buf += l; | |
1252 | size -= l; | |
1253 | total_len += l; | |
1254 | } | |
1255 | ||
1256 | return total_len; | |
23ce2127 SH |
1257 | } |
1258 | ||
7bbf2246 SH |
1259 | /* |
1260 | * How to guess what to present for uptime? | |
1261 | * One thing we could do would be to take the date on the caller's | |
1262 | * memory.usage_in_bytes file, which should equal the time of creation | |
1263 | * of his cgroup. However, a task could be in a sub-cgroup of the | |
1264 | * container. The same problem exists if we try to look at the ages | |
1265 | * of processes in the caller's cgroup. | |
1266 | * | |
1267 | * So we'll fork a task that will enter the caller's pidns, mount a | |
1268 | * fresh procfs, get the age of /proc/1, and pass that back over a pipe. | |
1269 | * | |
1270 | * For the second uptime #, we'll do as Stéphane had done, just copy | |
1271 | * the number from /proc/uptime. Not sure how to best emulate 'idle' | |
1272 | * time. Maybe someone can come up with a good algorithm and submit a | |
1273 | * patch. Maybe something based on cpushare info? | |
1274 | */ | |
23ce2127 SH |
1275 | static int proc_uptime_read(char *buf, size_t size, off_t offset, |
1276 | struct fuse_file_info *fi) | |
1277 | { | |
1278 | return 0; | |
1279 | } | |
1280 | ||
1281 | static off_t get_procfile_size(const char *which) | |
1282 | { | |
1283 | FILE *f = fopen(which, "r"); | |
1284 | char *line = NULL; | |
1285 | size_t len = 0; | |
1286 | ssize_t sz, answer = 0; | |
1287 | if (!f) | |
1288 | return 0; | |
1289 | ||
1290 | while ((sz = getline(&line, &len, f)) != -1) | |
1291 | answer += sz; | |
1292 | fclose (f); | |
1293 | ||
1294 | return answer; | |
1295 | } | |
1296 | ||
758ad80c SH |
1297 | static int proc_getattr(const char *path, struct stat *sb) |
1298 | { | |
35629743 SH |
1299 | struct timespec now; |
1300 | ||
1301 | memset(sb, 0, sizeof(struct stat)); | |
1302 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1303 | return -EINVAL; | |
1304 | sb->st_uid = sb->st_gid = 0; | |
1305 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1306 | if (strcmp(path, "/proc") == 0) { | |
1307 | sb->st_mode = S_IFDIR | 00555; | |
1308 | sb->st_nlink = 2; | |
1309 | return 0; | |
1310 | } | |
1311 | if (strcmp(path, "/proc/meminfo") == 0 || | |
1312 | strcmp(path, "/proc/cpuinfo") == 0 || | |
1313 | strcmp(path, "/proc/uptime") == 0 || | |
1314 | strcmp(path, "/proc/stat") == 0) { | |
23ce2127 SH |
1315 | |
1316 | sb->st_size = get_procfile_size(path); | |
35629743 SH |
1317 | sb->st_mode = S_IFREG | 00444; |
1318 | sb->st_nlink = 1; | |
1319 | return 0; | |
1320 | } | |
1321 | ||
1322 | return -ENOENT; | |
1323 | } | |
1324 | ||
1325 | static int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1326 | struct fuse_file_info *fi) | |
1327 | { | |
1328 | if (filler(buf, "cpuinfo", NULL, 0) != 0 || | |
1329 | filler(buf, "meminfo", NULL, 0) != 0 || | |
1330 | filler(buf, "stat", NULL, 0) != 0 || | |
1331 | filler(buf, "uptime", NULL, 0) != 0) | |
758ad80c | 1332 | return -EINVAL; |
758ad80c SH |
1333 | return 0; |
1334 | } | |
1335 | ||
35629743 SH |
1336 | static int proc_open(const char *path, struct fuse_file_info *fi) |
1337 | { | |
1338 | if (strcmp(path, "/proc/meminfo") == 0 || | |
1339 | strcmp(path, "/proc/cpuinfo") == 0 || | |
1340 | strcmp(path, "/proc/uptime") == 0 || | |
1341 | strcmp(path, "/proc/stat") == 0) | |
1342 | return 0; | |
1343 | return -ENOENT; | |
1344 | } | |
1345 | ||
35629743 SH |
1346 | static int proc_read(const char *path, char *buf, size_t size, off_t offset, |
1347 | struct fuse_file_info *fi) | |
1348 | { | |
1349 | if (strcmp(path, "/proc/meminfo") == 0) | |
23ce2127 | 1350 | return proc_meminfo_read(buf, size, offset, fi); |
35629743 | 1351 | if (strcmp(path, "/proc/cpuinfo") == 0) |
23ce2127 | 1352 | return proc_cpuinfo_read(buf, size, offset, fi); |
35629743 | 1353 | if (strcmp(path, "/proc/uptime") == 0) |
23ce2127 | 1354 | return proc_uptime_read(buf, size, offset, fi); |
35629743 | 1355 | if (strcmp(path, "/proc/stat") == 0) |
23ce2127 | 1356 | return proc_stat_read(buf, size, offset, fi); |
35629743 SH |
1357 | return -EINVAL; |
1358 | } | |
1359 | ||
2ad6d2bd SH |
1360 | /* |
1361 | * FUSE ops for / | |
1362 | * these just delegate to the /proc and /cgroup ops as | |
1363 | * needed | |
1364 | */ | |
758ad80c SH |
1365 | |
1366 | static int lxcfs_getattr(const char *path, struct stat *sb) | |
1367 | { | |
1368 | if (strcmp(path, "/") == 0) { | |
1369 | sb->st_mode = S_IFDIR | 00755; | |
1370 | sb->st_nlink = 2; | |
1371 | return 0; | |
1372 | } | |
1373 | if (strncmp(path, "/cgroup", 7) == 0) { | |
1374 | return cg_getattr(path, sb); | |
1375 | } | |
35629743 | 1376 | if (strncmp(path, "/proc", 5) == 0) { |
758ad80c SH |
1377 | return proc_getattr(path, sb); |
1378 | } | |
1379 | return -EINVAL; | |
1380 | } | |
1381 | ||
1382 | static int lxcfs_opendir(const char *path, struct fuse_file_info *fi) | |
1383 | { | |
1384 | if (strcmp(path, "/") == 0) | |
1385 | return 0; | |
1386 | ||
1387 | if (strncmp(path, "/cgroup", 7) == 0) { | |
1388 | return cg_opendir(path, fi); | |
1389 | } | |
35629743 SH |
1390 | if (strcmp(path, "/proc") == 0) |
1391 | return 0; | |
1392 | return -ENOENT; | |
758ad80c SH |
1393 | } |
1394 | ||
1395 | static int lxcfs_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1396 | struct fuse_file_info *fi) | |
1397 | { | |
1398 | if (strcmp(path, "/") == 0) { | |
1399 | if (filler(buf, "proc", NULL, 0) != 0 || | |
1400 | filler(buf, "cgroup", NULL, 0) != 0) | |
1401 | return -EINVAL; | |
1402 | return 0; | |
1403 | } | |
35629743 | 1404 | if (strncmp(path, "/cgroup", 7) == 0) |
758ad80c | 1405 | return cg_readdir(path, buf, filler, offset, fi); |
35629743 SH |
1406 | if (strcmp(path, "/proc") == 0) |
1407 | return proc_readdir(path, buf, filler, offset, fi); | |
758ad80c SH |
1408 | return -EINVAL; |
1409 | } | |
1410 | ||
1411 | static int lxcfs_releasedir(const char *path, struct fuse_file_info *fi) | |
1412 | { | |
1413 | if (strcmp(path, "/") == 0) | |
1414 | return 0; | |
1415 | if (strncmp(path, "/cgroup", 7) == 0) { | |
1416 | return cg_releasedir(path, fi); | |
1417 | } | |
35629743 SH |
1418 | if (strcmp(path, "/proc") == 0) |
1419 | return 0; | |
758ad80c SH |
1420 | return -EINVAL; |
1421 | } | |
1422 | ||
99978832 SH |
1423 | static int lxcfs_open(const char *path, struct fuse_file_info *fi) |
1424 | { | |
35629743 | 1425 | if (strncmp(path, "/cgroup", 7) == 0) |
99978832 | 1426 | return cg_open(path, fi); |
35629743 SH |
1427 | if (strncmp(path, "/proc", 5) == 0) |
1428 | return proc_open(path, fi); | |
99978832 SH |
1429 | |
1430 | return -EINVAL; | |
1431 | } | |
1432 | ||
1433 | static int lxcfs_read(const char *path, char *buf, size_t size, off_t offset, | |
1434 | struct fuse_file_info *fi) | |
1435 | { | |
35629743 | 1436 | if (strncmp(path, "/cgroup", 7) == 0) |
99978832 | 1437 | return cg_read(path, buf, size, offset, fi); |
35629743 SH |
1438 | if (strncmp(path, "/proc", 5) == 0) |
1439 | return proc_read(path, buf, size, offset, fi); | |
99978832 SH |
1440 | |
1441 | return -EINVAL; | |
1442 | } | |
1443 | ||
2ad6d2bd SH |
1444 | int lxcfs_write(const char *path, const char *buf, size_t size, off_t offset, |
1445 | struct fuse_file_info *fi) | |
1446 | { | |
1447 | if (strncmp(path, "/cgroup", 7) == 0) { | |
1448 | return cg_write(path, buf, size, offset, fi); | |
1449 | } | |
1450 | ||
1451 | return -EINVAL; | |
1452 | } | |
1453 | ||
99978832 SH |
1454 | static int lxcfs_flush(const char *path, struct fuse_file_info *fi) |
1455 | { | |
1456 | return 0; | |
1457 | } | |
1458 | ||
1459 | static int lxcfs_release(const char *path, struct fuse_file_info *fi) | |
758ad80c | 1460 | { |
99978832 SH |
1461 | return 0; |
1462 | } | |
1463 | ||
1464 | static int lxcfs_fsync(const char *path, int datasync, struct fuse_file_info *fi) | |
1465 | { | |
1466 | return 0; | |
758ad80c SH |
1467 | } |
1468 | ||
ab54b798 SH |
1469 | int lxcfs_mkdir(const char *path, mode_t mode) |
1470 | { | |
1471 | if (strncmp(path, "/cgroup", 7) == 0) | |
1472 | return cg_mkdir(path, mode); | |
1473 | ||
1474 | return -EINVAL; | |
1475 | } | |
1476 | ||
341b21ad SH |
1477 | int lxcfs_chown(const char *path, uid_t uid, gid_t gid) |
1478 | { | |
1479 | if (strncmp(path, "/cgroup", 7) == 0) | |
1480 | return cg_chown(path, uid, gid); | |
1481 | ||
1482 | return -EINVAL; | |
1483 | } | |
1484 | ||
2ad6d2bd SH |
1485 | /* |
1486 | * cat first does a truncate before doing ops->write. This doesn't | |
1487 | * really make sense for cgroups. So just return 0 always but do | |
1488 | * nothing. | |
1489 | */ | |
1490 | int lxcfs_truncate(const char *path, off_t newsize) | |
1491 | { | |
1492 | if (strncmp(path, "/cgroup", 7) == 0) | |
1493 | return 0; | |
1494 | return -EINVAL; | |
1495 | } | |
1496 | ||
50d8d5b5 SH |
1497 | int lxcfs_rmdir(const char *path) |
1498 | { | |
1499 | if (strncmp(path, "/cgroup", 7) == 0) | |
1500 | return cg_rmdir(path); | |
1501 | return -EINVAL; | |
1502 | } | |
1503 | ||
fd2e4e03 SH |
1504 | int lxcfs_chmod(const char *path, mode_t mode) |
1505 | { | |
1506 | if (strncmp(path, "/cgroup", 7) == 0) | |
1507 | return cg_chmod(path, mode); | |
1508 | return -EINVAL; | |
1509 | } | |
1510 | ||
758ad80c SH |
1511 | const struct fuse_operations lxcfs_ops = { |
1512 | .getattr = lxcfs_getattr, | |
1513 | .readlink = NULL, | |
1514 | .getdir = NULL, | |
1515 | .mknod = NULL, | |
ab54b798 | 1516 | .mkdir = lxcfs_mkdir, |
758ad80c | 1517 | .unlink = NULL, |
50d8d5b5 | 1518 | .rmdir = lxcfs_rmdir, |
758ad80c SH |
1519 | .symlink = NULL, |
1520 | .rename = NULL, | |
1521 | .link = NULL, | |
fd2e4e03 | 1522 | .chmod = lxcfs_chmod, |
341b21ad | 1523 | .chown = lxcfs_chown, |
2ad6d2bd | 1524 | .truncate = lxcfs_truncate, |
758ad80c | 1525 | .utime = NULL, |
99978832 SH |
1526 | |
1527 | .open = lxcfs_open, | |
1528 | .read = lxcfs_read, | |
1529 | .release = lxcfs_release, | |
2ad6d2bd | 1530 | .write = lxcfs_write, |
99978832 | 1531 | |
758ad80c | 1532 | .statfs = NULL, |
99978832 SH |
1533 | .flush = lxcfs_flush, |
1534 | .fsync = lxcfs_fsync, | |
758ad80c SH |
1535 | |
1536 | .setxattr = NULL, | |
1537 | .getxattr = NULL, | |
1538 | .listxattr = NULL, | |
1539 | .removexattr = NULL, | |
1540 | ||
1541 | .opendir = lxcfs_opendir, | |
1542 | .readdir = lxcfs_readdir, | |
1543 | .releasedir = lxcfs_releasedir, | |
1544 | ||
1545 | .fsyncdir = NULL, | |
1546 | .init = NULL, | |
1547 | .destroy = NULL, | |
1548 | .access = NULL, | |
1549 | .create = NULL, | |
1550 | .ftruncate = NULL, | |
1551 | .fgetattr = NULL, | |
1552 | }; | |
1553 | ||
99978832 | 1554 | static void usage(const char *me) |
758ad80c SH |
1555 | { |
1556 | fprintf(stderr, "Usage:\n"); | |
1557 | fprintf(stderr, "\n"); | |
1558 | fprintf(stderr, "%s [FUSE and mount options] mountpoint\n", me); | |
1559 | exit(1); | |
1560 | } | |
1561 | ||
99978832 | 1562 | static bool is_help(char *w) |
758ad80c SH |
1563 | { |
1564 | if (strcmp(w, "-h") == 0 || | |
1565 | strcmp(w, "--help") == 0 || | |
1566 | strcmp(w, "-help") == 0 || | |
1567 | strcmp(w, "help") == 0) | |
1568 | return true; | |
1569 | return false; | |
1570 | } | |
1571 | ||
1572 | int main(int argc, char *argv[]) | |
1573 | { | |
1574 | int ret; | |
1575 | struct lxcfs_state *d; | |
1576 | ||
1577 | if (argc < 2 || is_help(argv[1])) | |
1578 | usage(argv[0]); | |
1579 | ||
1580 | d = malloc(sizeof(*d)); | |
1581 | if (!d) | |
1582 | return -1; | |
1583 | ||
1584 | if (!cgm_escape_cgroup()) | |
1585 | fprintf(stderr, "WARNING: failed to escape to root cgroup\n"); | |
1586 | ||
1587 | if (!cgm_get_controllers(&d->subsystems)) | |
1588 | return -1; | |
1589 | ||
1590 | ret = fuse_main(argc, argv, &lxcfs_ops, d); | |
1591 | ||
1592 | return ret; | |
2183082c | 1593 | } |