]>
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 | ||
758ad80c SH |
9 | #define FUSE_USE_VERSION 26 |
10 | ||
2183082c | 11 | #include <stdio.h> |
758ad80c SH |
12 | #include <dirent.h> |
13 | #include <fcntl.h> | |
14 | #include <fuse.h> | |
15 | #include <unistd.h> | |
16 | #include <errno.h> | |
17 | #include <stdbool.h> | |
18 | #include <time.h> | |
19 | #include <string.h> | |
20 | #include <stdlib.h> | |
21 | #include <libgen.h> | |
41bb9357 SH |
22 | #include <sched.h> |
23 | #include <linux/sched.h> | |
a05660a6 | 24 | #include <sys/socket.h> |
41bb9357 SH |
25 | #include <sys/mount.h> |
26 | #include <wait.h> | |
758ad80c SH |
27 | |
28 | #include <nih/alloc.h> | |
29 | #include <nih/string.h> | |
30 | ||
31 | #include "cgmanager.h" | |
2e9c0b32 | 32 | #include "config.h" // for VERSION |
758ad80c SH |
33 | |
34 | struct lxcfs_state { | |
35 | /* | |
36 | * a null-terminated, nih-allocated list of the mounted subsystems. We | |
37 | * detect this at startup. | |
38 | */ | |
39 | char **subsystems; | |
40 | }; | |
41 | #define LXCFS_DATA ((struct lxcfs_state *) fuse_get_context()->private_data) | |
42 | ||
443d13f5 SH |
43 | enum { |
44 | LXC_TYPE_CGDIR, | |
45 | LXC_TYPE_CGFILE, | |
46 | LXC_TYPE_PROC_MEMINFO, | |
47 | LXC_TYPE_PROC_CPUINFO, | |
48 | LXC_TYPE_PROC_UPTIME, | |
49 | LXC_TYPE_PROC_STAT, | |
50 | LXC_TYPE_PROC_DISKSTATS, | |
51 | }; | |
52 | ||
c688e1b3 SH |
53 | struct file_info { |
54 | char *controller; | |
55 | char *cgroup; | |
8f6e8f5e | 56 | char *file; |
443d13f5 | 57 | int type; |
c688e1b3 SH |
58 | char *buf; // unused as of yet |
59 | int buflen; | |
97f1f27b | 60 | int size; //actual data size |
c688e1b3 SH |
61 | }; |
62 | ||
97f1f27b YY |
63 | /* reserve buffer size, for cpuall in /proc/stat */ |
64 | #define BUF_RESERVE_SIZE 256 | |
65 | ||
bae07053 | 66 | static char *must_copy_string(void *parent, const char *str) |
c688e1b3 SH |
67 | { |
68 | if (!str) | |
69 | return NULL; | |
bae07053 | 70 | return NIH_MUST( nih_strdup(parent, str) ); |
c688e1b3 SH |
71 | } |
72 | ||
4775fba1 SH |
73 | /* |
74 | * TODO - return value should denote whether child exited with failure | |
75 | * so callers can return errors. Esp read/write of tasks and cgroup.procs | |
76 | */ | |
a05660a6 SH |
77 | static int wait_for_pid(pid_t pid) |
78 | { | |
79 | int status, ret; | |
80 | ||
81 | again: | |
82 | ret = waitpid(pid, &status, 0); | |
83 | if (ret == -1) { | |
84 | if (errno == EINTR) | |
85 | goto again; | |
86 | return -1; | |
87 | } | |
88 | if (ret != pid) | |
89 | goto again; | |
90 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
91 | return -1; | |
92 | return 0; | |
93 | } | |
94 | ||
053a659d SH |
95 | /* |
96 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
97 | * valid in the caller's namespace, return the id mapped into | |
98 | * pid's namespace. | |
99 | * Returns the mapped id, or -1 on error. | |
100 | */ | |
101 | unsigned int | |
102 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
103 | { | |
104 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
105 | hostuid, // base id for a range in the caller's namespace | |
106 | count; // number of ids in this range | |
107 | char line[400]; | |
108 | int ret; | |
109 | ||
110 | fseek(idfile, 0L, SEEK_SET); | |
111 | while (fgets(line, 400, idfile)) { | |
112 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
113 | if (ret != 3) | |
114 | continue; | |
115 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
116 | /* | |
117 | * uids wrapped around - unexpected as this is a procfile, | |
118 | * so just bail. | |
119 | */ | |
647c89e5 | 120 | fprintf(stderr, "pid wrapparound at entry %u %u %u in %s\n", |
053a659d SH |
121 | nsuid, hostuid, count, line); |
122 | return -1; | |
123 | } | |
124 | if (hostuid <= in_id && hostuid+count > in_id) { | |
125 | /* | |
126 | * now since hostuid <= in_id < hostuid+count, and | |
127 | * hostuid+count and nsuid+count do not wrap around, | |
128 | * we know that nsuid+(in_id-hostuid) which must be | |
129 | * less that nsuid+(count) must not wrap around | |
130 | */ | |
131 | return (in_id - hostuid) + nsuid; | |
132 | } | |
133 | } | |
134 | ||
135 | // no answer found | |
136 | return -1; | |
137 | } | |
138 | ||
341b21ad SH |
139 | /* |
140 | * for is_privileged_over, | |
141 | * specify whether we require the calling uid to be root in his | |
142 | * namespace | |
143 | */ | |
144 | #define NS_ROOT_REQD true | |
145 | #define NS_ROOT_OPT false | |
146 | ||
147 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
758ad80c | 148 | { |
053a659d SH |
149 | nih_local char *fpath = NULL; |
150 | bool answer = false; | |
151 | uid_t nsuid; | |
152 | ||
341b21ad SH |
153 | if (victim == -1 || uid == -1) |
154 | return false; | |
155 | ||
156 | /* | |
157 | * If the request is one not requiring root in the namespace, | |
158 | * then having the same uid suffices. (i.e. uid 1000 has write | |
159 | * access to files owned by uid 1000 | |
160 | */ | |
161 | if (!req_ns_root && uid == victim) | |
758ad80c SH |
162 | return true; |
163 | ||
053a659d SH |
164 | fpath = NIH_MUST( nih_sprintf(NULL, "/proc/%d/uid_map", pid) ); |
165 | FILE *f = fopen(fpath, "r"); | |
166 | if (!f) | |
167 | return false; | |
168 | ||
341b21ad | 169 | /* if caller's not root in his namespace, reject */ |
053a659d SH |
170 | nsuid = convert_id_to_ns(f, uid); |
171 | if (nsuid) | |
172 | goto out; | |
173 | ||
341b21ad SH |
174 | /* |
175 | * If victim is not mapped into caller's ns, reject. | |
176 | * XXX I'm not sure this check is needed given that fuse | |
177 | * will be sending requests where the vfs has converted | |
178 | */ | |
053a659d SH |
179 | nsuid = convert_id_to_ns(f, victim); |
180 | if (nsuid == -1) | |
181 | goto out; | |
182 | ||
183 | answer = true; | |
184 | ||
185 | out: | |
186 | fclose(f); | |
187 | return answer; | |
758ad80c SH |
188 | } |
189 | ||
190 | static bool perms_include(int fmode, mode_t req_mode) | |
191 | { | |
2ad6d2bd SH |
192 | mode_t r; |
193 | ||
194 | switch (req_mode & O_ACCMODE) { | |
195 | case O_RDONLY: | |
196 | r = S_IROTH; | |
197 | break; | |
198 | case O_WRONLY: | |
199 | r = S_IWOTH; | |
200 | break; | |
201 | case O_RDWR: | |
202 | r = S_IROTH | S_IWOTH; | |
203 | break; | |
204 | default: | |
205 | return false; | |
206 | } | |
207 | return ((fmode & r) == r); | |
758ad80c SH |
208 | } |
209 | ||
3db25a35 SH |
210 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) |
211 | { | |
212 | char *start, *end; | |
213 | ||
214 | if (strlen(taskcg) <= strlen(querycg)) { | |
215 | fprintf(stderr, "%s: I was fed bad input\n", __func__); | |
216 | return NULL; | |
217 | } | |
218 | ||
219 | if (strcmp(querycg, "/") == 0) | |
220 | start = NIH_MUST( nih_strdup(NULL, taskcg + 1) ); | |
221 | else | |
222 | start = NIH_MUST( nih_strdup(NULL, taskcg + strlen(querycg) + 1) ); | |
223 | end = strchr(start, '/'); | |
224 | if (end) | |
225 | *end = '\0'; | |
226 | return start; | |
227 | } | |
228 | ||
758ad80c SH |
229 | /* |
230 | * check whether a fuse context may access a cgroup dir or file | |
231 | * | |
232 | * If file is not null, it is a cgroup file to check under cg. | |
233 | * If file is null, then we are checking perms on cg itself. | |
234 | * | |
235 | * For files we can check the mode of the list_keys result. | |
236 | * For cgroups, we must make assumptions based on the files under the | |
237 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
238 | * yet. | |
239 | */ | |
240 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
241 | { | |
242 | nih_local struct cgm_keys **list = NULL; | |
243 | int i; | |
244 | ||
245 | if (!file) | |
246 | file = "tasks"; | |
247 | ||
248 | if (*file == '/') | |
249 | file++; | |
250 | ||
251 | if (!cgm_list_keys(contrl, cg, &list)) | |
252 | return false; | |
253 | for (i = 0; list[i]; i++) { | |
254 | if (strcmp(list[i]->name, file) == 0) { | |
255 | struct cgm_keys *k = list[i]; | |
341b21ad | 256 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { |
758ad80c SH |
257 | if (perms_include(k->mode >> 6, mode)) |
258 | return true; | |
259 | } | |
260 | if (fc->gid == k->gid) { | |
261 | if (perms_include(k->mode >> 3, mode)) | |
262 | return true; | |
263 | } | |
264 | return perms_include(k->mode, mode); | |
265 | } | |
266 | } | |
267 | ||
268 | return false; | |
269 | } | |
270 | ||
3db25a35 SH |
271 | static void stripnewline(char *x) |
272 | { | |
273 | size_t l = strlen(x); | |
274 | if (l && x[l-1] == '\n') | |
275 | x[l-1] = '\0'; | |
276 | } | |
277 | ||
278 | /* | |
279 | * If caller is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
280 | * If caller is in /a, he may act on /a/b, but not on /b. | |
281 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
282 | * to a nih_alloc'd string containing the next cgroup directory under cg | |
283 | */ | |
284 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
285 | { | |
286 | nih_local char *fnam = NULL; | |
287 | FILE *f; | |
288 | bool answer = false; | |
289 | char *line = NULL; | |
290 | size_t len = 0; | |
291 | ||
292 | fnam = NIH_MUST( nih_sprintf(NULL, "/proc/%d/cgroup", pid) ); | |
293 | if (!(f = fopen(fnam, "r"))) | |
294 | return false; | |
295 | ||
296 | while (getline(&line, &len, f) != -1) { | |
297 | char *c1, *c2, *linecmp; | |
298 | if (!line[0]) | |
299 | continue; | |
300 | c1 = strchr(line, ':'); | |
301 | if (!c1) | |
302 | goto out; | |
303 | c1++; | |
304 | c2 = strchr(c1, ':'); | |
305 | if (!c2) | |
306 | goto out; | |
307 | *c2 = '\0'; | |
308 | if (strcmp(c1, contrl) != 0) | |
309 | continue; | |
310 | c2++; | |
311 | stripnewline(c2); | |
312 | /* | |
313 | * callers pass in '/' for root cgroup, otherwise they pass | |
314 | * in a cgroup without leading '/' | |
315 | */ | |
316 | linecmp = *cg == '/' ? c2 : c2+1; | |
317 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { | |
318 | if (nextcg) | |
319 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
320 | goto out; | |
321 | } | |
322 | answer = true; | |
323 | goto out; | |
324 | } | |
325 | ||
326 | out: | |
327 | fclose(f); | |
328 | free(line); | |
329 | return answer; | |
330 | } | |
331 | ||
758ad80c SH |
332 | /* |
333 | * given /cgroup/freezer/a/b, return "freezer". this will be nih-allocated | |
334 | * and needs to be nih_freed. | |
335 | */ | |
336 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
337 | { | |
338 | const char *p1; | |
339 | char *ret, *slash; | |
340 | ||
341 | if (strlen(path) < 9) | |
342 | return NULL; | |
ac5d9d48 SH |
343 | if (*(path+7) != '/') |
344 | return NULL; | |
758ad80c SH |
345 | p1 = path+8; |
346 | ret = nih_strdup(NULL, p1); | |
347 | if (!ret) | |
348 | return ret; | |
349 | slash = strstr(ret, "/"); | |
350 | if (slash) | |
351 | *slash = '\0'; | |
352 | ||
353 | /* verify that it is a subsystem */ | |
354 | char **list = LXCFS_DATA ? LXCFS_DATA->subsystems : NULL; | |
355 | int i; | |
356 | if (!list) { | |
357 | nih_free(ret); | |
358 | return NULL; | |
359 | } | |
360 | for (i = 0; list[i]; i++) { | |
361 | if (strcmp(list[i], ret) == 0) | |
362 | return ret; | |
363 | } | |
364 | nih_free(ret); | |
365 | return NULL; | |
366 | } | |
367 | ||
368 | /* | |
369 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
370 | * Note that the returned value may include files (keynames) etc | |
371 | */ | |
372 | static const char *find_cgroup_in_path(const char *path) | |
373 | { | |
374 | const char *p1; | |
375 | ||
376 | if (strlen(path) < 9) | |
377 | return NULL; | |
378 | p1 = strstr(path+8, "/"); | |
379 | if (!p1) | |
380 | return NULL; | |
381 | return p1+1; | |
382 | } | |
383 | ||
384 | static bool is_child_cgroup(const char *contr, const char *dir, const char *f) | |
385 | { | |
386 | nih_local char **list = NULL; | |
387 | int i; | |
388 | ||
389 | if (!f) | |
390 | return false; | |
391 | if (*f == '/') | |
392 | f++; | |
393 | ||
394 | if (!cgm_list_children(contr, dir, &list)) | |
395 | return false; | |
396 | for (i = 0; list[i]; i++) { | |
397 | if (strcmp(list[i], f) == 0) | |
398 | return true; | |
399 | } | |
400 | ||
401 | return false; | |
402 | } | |
403 | ||
404 | static struct cgm_keys *get_cgroup_key(const char *contr, const char *dir, const char *f) | |
405 | { | |
406 | nih_local struct cgm_keys **list = NULL; | |
407 | struct cgm_keys *k; | |
408 | int i; | |
409 | ||
410 | if (!f) | |
411 | return NULL; | |
412 | if (*f == '/') | |
413 | f++; | |
414 | if (!cgm_list_keys(contr, dir, &list)) | |
415 | return NULL; | |
416 | for (i = 0; list[i]; i++) { | |
417 | if (strcmp(list[i]->name, f) == 0) { | |
418 | k = NIH_MUST( nih_alloc(NULL, (sizeof(*k))) ); | |
419 | k->name = NIH_MUST( nih_strdup(k, list[i]->name) ); | |
420 | k->uid = list[i]->uid; | |
421 | k->gid = list[i]->gid; | |
422 | k->mode = list[i]->mode; | |
423 | return k; | |
424 | } | |
425 | } | |
426 | ||
427 | return NULL; | |
428 | } | |
429 | ||
430 | static void get_cgdir_and_path(const char *cg, char **dir, char **file) | |
431 | { | |
758ad80c SH |
432 | char *p; |
433 | ||
434 | *dir = NIH_MUST( nih_strdup(NULL, cg) ); | |
435 | *file = strrchr(cg, '/'); | |
436 | if (!*file) { | |
437 | *file = NULL; | |
438 | return; | |
439 | } | |
440 | p = strrchr(*dir, '/'); | |
441 | *p = '\0'; | |
442 | } | |
443 | ||
444 | /* | |
2ad6d2bd | 445 | * FUSE ops for /cgroup |
758ad80c | 446 | */ |
2ad6d2bd | 447 | |
758ad80c SH |
448 | static int cg_getattr(const char *path, struct stat *sb) |
449 | { | |
450 | struct timespec now; | |
451 | struct fuse_context *fc = fuse_get_context(); | |
452 | nih_local char * cgdir = NULL; | |
453 | char *fpath = NULL, *path1, *path2; | |
454 | nih_local struct cgm_keys *k = NULL; | |
455 | const char *cgroup; | |
456 | nih_local char *controller = NULL; | |
457 | ||
458 | ||
459 | if (!fc) | |
460 | return -EIO; | |
461 | ||
462 | memset(sb, 0, sizeof(struct stat)); | |
463 | ||
464 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
465 | return -EINVAL; | |
466 | ||
467 | sb->st_uid = sb->st_gid = 0; | |
468 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
469 | sb->st_size = 0; | |
470 | ||
471 | if (strcmp(path, "/cgroup") == 0) { | |
472 | sb->st_mode = S_IFDIR | 00755; | |
473 | sb->st_nlink = 2; | |
474 | return 0; | |
475 | } | |
476 | ||
477 | controller = pick_controller_from_path(fc, path); | |
478 | if (!controller) | |
479 | return -EIO; | |
758ad80c SH |
480 | cgroup = find_cgroup_in_path(path); |
481 | if (!cgroup) { | |
482 | /* this is just /cgroup/controller, return it as a dir */ | |
483 | sb->st_mode = S_IFDIR | 00755; | |
484 | sb->st_nlink = 2; | |
485 | return 0; | |
486 | } | |
341b21ad | 487 | |
758ad80c SH |
488 | get_cgdir_and_path(cgroup, &cgdir, &fpath); |
489 | ||
490 | if (!fpath) { | |
491 | path1 = "/"; | |
492 | path2 = cgdir; | |
493 | } else { | |
494 | path1 = cgdir; | |
495 | path2 = fpath; | |
496 | } | |
497 | ||
758ad80c SH |
498 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. |
499 | * Then check that caller's cgroup is under path if fpath is a child | |
500 | * cgroup, or cgdir if fpath is a file */ | |
501 | ||
502 | if (is_child_cgroup(controller, path1, path2)) { | |
f9a05025 SH |
503 | if (!caller_is_in_ancestor(fc->pid, controller, cgroup, NULL)) { |
504 | /* this is just /cgroup/controller, return it as a dir */ | |
505 | sb->st_mode = S_IFDIR | 00555; | |
506 | sb->st_nlink = 2; | |
507 | return 0; | |
508 | } | |
758ad80c | 509 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) |
f9a05025 | 510 | return -EACCES; |
758ad80c | 511 | |
053a659d SH |
512 | // get uid, gid, from '/tasks' file and make up a mode |
513 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
514 | sb->st_mode = S_IFDIR | 00755; | |
515 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
516 | if (!k) { | |
053a659d SH |
517 | sb->st_uid = sb->st_gid = 0; |
518 | } else { | |
053a659d SH |
519 | sb->st_uid = k->uid; |
520 | sb->st_gid = k->gid; | |
521 | } | |
758ad80c SH |
522 | sb->st_nlink = 2; |
523 | return 0; | |
524 | } | |
525 | ||
526 | if ((k = get_cgroup_key(controller, path1, path2)) != NULL) { | |
3db25a35 SH |
527 | if (!caller_is_in_ancestor(fc->pid, controller, path1, NULL)) |
528 | return -ENOENT; | |
758ad80c | 529 | if (!fc_may_access(fc, controller, path1, path2, O_RDONLY)) |
f9a05025 | 530 | return -EACCES; |
758ad80c | 531 | |
758ad80c | 532 | sb->st_mode = S_IFREG | k->mode; |
053a659d | 533 | sb->st_nlink = 1; |
758ad80c SH |
534 | sb->st_uid = k->uid; |
535 | sb->st_gid = k->gid; | |
7253e0a4 | 536 | sb->st_size = 0; |
758ad80c SH |
537 | return 0; |
538 | } | |
539 | ||
ab54b798 | 540 | return -ENOENT; |
758ad80c | 541 | } |
2183082c | 542 | |
7f163b71 SH |
543 | /* |
544 | * TODO - cache these results in a table for use in opendir, free | |
545 | * in releasedir | |
546 | */ | |
758ad80c | 547 | static int cg_opendir(const char *path, struct fuse_file_info *fi) |
2183082c | 548 | { |
7f163b71 SH |
549 | struct fuse_context *fc = fuse_get_context(); |
550 | nih_local struct cgm_keys **list = NULL; | |
551 | const char *cgroup; | |
c688e1b3 | 552 | struct file_info *dir_info; |
7f163b71 | 553 | nih_local char *controller = NULL; |
7f163b71 SH |
554 | |
555 | if (!fc) | |
556 | return -EIO; | |
557 | ||
c688e1b3 SH |
558 | if (strcmp(path, "/cgroup") == 0) { |
559 | cgroup = NULL; | |
560 | controller = NULL; | |
561 | } else { | |
562 | // return list of keys for the controller, and list of child cgroups | |
563 | controller = pick_controller_from_path(fc, path); | |
564 | if (!controller) | |
565 | return -EIO; | |
7f163b71 | 566 | |
c688e1b3 SH |
567 | cgroup = find_cgroup_in_path(path); |
568 | if (!cgroup) { | |
569 | /* this is just /cgroup/controller, return its contents */ | |
570 | cgroup = "/"; | |
571 | } | |
7f163b71 SH |
572 | } |
573 | ||
3a6e1a76 | 574 | if (cgroup && !fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) |
7f163b71 | 575 | return -EACCES; |
c688e1b3 SH |
576 | |
577 | /* we'll free this at cg_releasedir */ | |
578 | dir_info = NIH_MUST( nih_alloc(NULL, sizeof(*dir_info)) ); | |
bae07053 SH |
579 | dir_info->controller = must_copy_string(dir_info, controller); |
580 | dir_info->cgroup = must_copy_string(dir_info, cgroup); | |
443d13f5 | 581 | dir_info->type = LXC_TYPE_CGDIR; |
c688e1b3 | 582 | dir_info->buf = NULL; |
8f6e8f5e | 583 | dir_info->file = NULL; |
c688e1b3 SH |
584 | dir_info->buflen = 0; |
585 | ||
586 | fi->fh = (unsigned long)dir_info; | |
758ad80c SH |
587 | return 0; |
588 | } | |
589 | ||
758ad80c SH |
590 | static int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, |
591 | struct fuse_file_info *fi) | |
592 | { | |
c688e1b3 SH |
593 | struct file_info *d = (struct file_info *)fi->fh; |
594 | nih_local struct cgm_keys **list = NULL; | |
595 | int i; | |
596 | nih_local char *nextcg = NULL; | |
758ad80c SH |
597 | struct fuse_context *fc = fuse_get_context(); |
598 | ||
443d13f5 | 599 | if (d->type != LXC_TYPE_CGDIR) { |
b845ad01 SH |
600 | fprintf(stderr, "Internal error: file cache info used in readdir\n"); |
601 | return -EIO; | |
602 | } | |
c688e1b3 SH |
603 | if (!d->cgroup && !d->controller) { |
604 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
758ad80c SH |
605 | char **list = LXCFS_DATA ? LXCFS_DATA->subsystems : NULL; |
606 | int i; | |
607 | ||
608 | if (!list) | |
609 | return -EIO; | |
7f163b71 | 610 | |
758ad80c SH |
611 | for (i = 0; list[i]; i++) { |
612 | if (filler(buf, list[i], NULL, 0) != 0) { | |
613 | return -EIO; | |
614 | } | |
615 | } | |
616 | return 0; | |
617 | } | |
618 | ||
c688e1b3 | 619 | if (!cgm_list_keys(d->controller, d->cgroup, &list)) |
3db25a35 | 620 | // not a valid cgroup |
758ad80c | 621 | return -EINVAL; |
3db25a35 | 622 | |
c688e1b3 | 623 | if (!caller_is_in_ancestor(fc->pid, d->controller, d->cgroup, &nextcg)) { |
3db25a35 SH |
624 | if (nextcg) { |
625 | int ret; | |
626 | ret = filler(buf, nextcg, NULL, 0); | |
627 | if (ret != 0) | |
628 | return -EIO; | |
629 | } | |
630 | return 0; | |
631 | } | |
632 | ||
758ad80c | 633 | for (i = 0; list[i]; i++) { |
758ad80c SH |
634 | if (filler(buf, list[i]->name, NULL, 0) != 0) { |
635 | return -EIO; | |
636 | } | |
637 | } | |
638 | ||
639 | // now get the list of child cgroups | |
422aa4a5 | 640 | nih_local char **clist = NULL; |
758ad80c | 641 | |
c688e1b3 | 642 | if (!cgm_list_children(d->controller, d->cgroup, &clist)) |
758ad80c SH |
643 | return 0; |
644 | for (i = 0; clist[i]; i++) { | |
758ad80c SH |
645 | if (filler(buf, clist[i], NULL, 0) != 0) { |
646 | return -EIO; | |
647 | } | |
648 | } | |
649 | return 0; | |
650 | } | |
651 | ||
8f6e8f5e SH |
652 | static void do_release_file_info(struct file_info *f) |
653 | { | |
bae07053 SH |
654 | /* |
655 | * all file_info fields which are nih_alloc()d with f as parent | |
656 | * will be automatically freed | |
657 | */ | |
8f6e8f5e SH |
658 | nih_free(f); |
659 | } | |
660 | ||
758ad80c SH |
661 | static int cg_releasedir(const char *path, struct fuse_file_info *fi) |
662 | { | |
c688e1b3 SH |
663 | struct file_info *d = (struct file_info *)fi->fh; |
664 | ||
8f6e8f5e | 665 | do_release_file_info(d); |
758ad80c SH |
666 | return 0; |
667 | } | |
668 | ||
99978832 SH |
669 | static int cg_open(const char *path, struct fuse_file_info *fi) |
670 | { | |
671 | nih_local char *controller = NULL; | |
672 | const char *cgroup; | |
673 | char *fpath = NULL, *path1, *path2; | |
674 | nih_local char * cgdir = NULL; | |
675 | nih_local struct cgm_keys *k = NULL; | |
8f6e8f5e | 676 | struct file_info *file_info; |
99978832 SH |
677 | struct fuse_context *fc = fuse_get_context(); |
678 | ||
679 | if (!fc) | |
680 | return -EIO; | |
681 | ||
682 | controller = pick_controller_from_path(fc, path); | |
683 | if (!controller) | |
684 | return -EIO; | |
685 | cgroup = find_cgroup_in_path(path); | |
686 | if (!cgroup) | |
687 | return -EINVAL; | |
688 | ||
689 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
690 | if (!fpath) { | |
691 | path1 = "/"; | |
692 | path2 = cgdir; | |
693 | } else { | |
694 | path1 = cgdir; | |
695 | path2 = fpath; | |
696 | } | |
697 | ||
8f6e8f5e SH |
698 | k = get_cgroup_key(controller, path1, path2); |
699 | if (!k) | |
700 | return -EINVAL; | |
99978832 | 701 | |
8f6e8f5e SH |
702 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) |
703 | // should never get here | |
704 | return -EACCES; | |
99978832 | 705 | |
8f6e8f5e SH |
706 | /* we'll free this at cg_release */ |
707 | file_info = NIH_MUST( nih_alloc(NULL, sizeof(*file_info)) ); | |
bae07053 SH |
708 | file_info->controller = must_copy_string(file_info, controller); |
709 | file_info->cgroup = must_copy_string(file_info, path1); | |
710 | file_info->file = must_copy_string(file_info, path2); | |
443d13f5 | 711 | file_info->type = LXC_TYPE_CGFILE; |
8f6e8f5e SH |
712 | file_info->buf = NULL; |
713 | file_info->buflen = 0; | |
714 | ||
715 | fi->fh = (unsigned long)file_info; | |
716 | return 0; | |
717 | } | |
718 | ||
719 | static int cg_release(const char *path, struct fuse_file_info *fi) | |
720 | { | |
721 | struct file_info *f = (struct file_info *)fi->fh; | |
722 | ||
723 | do_release_file_info(f); | |
724 | return 0; | |
99978832 SH |
725 | } |
726 | ||
a05660a6 SH |
727 | static int msgrecv(int sockfd, void *buf, size_t len) |
728 | { | |
729 | struct timeval tv; | |
730 | fd_set rfds; | |
731 | ||
732 | FD_ZERO(&rfds); | |
733 | FD_SET(sockfd, &rfds); | |
734 | tv.tv_sec = 2; | |
735 | tv.tv_usec = 0; | |
736 | ||
ea56f722 | 737 | if (select(sockfd+1, &rfds, NULL, NULL, &tv) <= 0) |
a05660a6 SH |
738 | return -1; |
739 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
740 | } | |
741 | ||
01e71852 SH |
742 | #define SEND_CREDS_OK 0 |
743 | #define SEND_CREDS_NOTSK 1 | |
744 | #define SEND_CREDS_FAIL 2 | |
745 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
a05660a6 SH |
746 | { |
747 | struct msghdr msg = { 0 }; | |
748 | struct iovec iov; | |
749 | struct cmsghdr *cmsg; | |
750 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
751 | char buf[1]; | |
752 | buf[0] = 'p'; | |
753 | ||
01e71852 SH |
754 | if (pingfirst) { |
755 | if (msgrecv(sock, buf, 1) != 1) { | |
1420baf8 | 756 | fprintf(stderr, "%s: Error getting reply from server over socketpair\n", |
01e71852 SH |
757 | __func__); |
758 | return SEND_CREDS_FAIL; | |
759 | } | |
a05660a6 SH |
760 | } |
761 | ||
762 | msg.msg_control = cmsgbuf; | |
763 | msg.msg_controllen = sizeof(cmsgbuf); | |
764 | ||
765 | cmsg = CMSG_FIRSTHDR(&msg); | |
766 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
767 | cmsg->cmsg_level = SOL_SOCKET; | |
768 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
769 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
770 | ||
771 | msg.msg_name = NULL; | |
772 | msg.msg_namelen = 0; | |
773 | ||
774 | buf[0] = v; | |
775 | iov.iov_base = buf; | |
776 | iov.iov_len = sizeof(buf); | |
777 | msg.msg_iov = &iov; | |
778 | msg.msg_iovlen = 1; | |
779 | ||
780 | if (sendmsg(sock, &msg, 0) < 0) { | |
1420baf8 | 781 | fprintf(stderr, "%s: failed at sendmsg: %s\n", __func__, |
a05660a6 SH |
782 | strerror(errno)); |
783 | if (errno == 3) | |
01e71852 SH |
784 | return SEND_CREDS_NOTSK; |
785 | return SEND_CREDS_FAIL; | |
a05660a6 SH |
786 | } |
787 | ||
01e71852 | 788 | return SEND_CREDS_OK; |
a05660a6 SH |
789 | } |
790 | ||
791 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
792 | { | |
793 | struct msghdr msg = { 0 }; | |
794 | struct iovec iov; | |
795 | struct cmsghdr *cmsg; | |
796 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
797 | char buf[1]; | |
798 | int ret; | |
799 | int optval = 1; | |
6ee867dc SH |
800 | struct timeval tv; |
801 | fd_set rfds; | |
a05660a6 SH |
802 | |
803 | *v = '1'; | |
804 | ||
805 | cred->pid = -1; | |
806 | cred->uid = -1; | |
807 | cred->gid = -1; | |
808 | ||
809 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
1420baf8 | 810 | fprintf(stderr, "Failed to set passcred: %s\n", strerror(errno)); |
a05660a6 SH |
811 | return false; |
812 | } | |
813 | buf[0] = '1'; | |
814 | if (write(sock, buf, 1) != 1) { | |
1420baf8 | 815 | fprintf(stderr, "Failed to start write on scm fd: %s\n", strerror(errno)); |
a05660a6 SH |
816 | return false; |
817 | } | |
818 | ||
819 | msg.msg_name = NULL; | |
820 | msg.msg_namelen = 0; | |
821 | msg.msg_control = cmsgbuf; | |
822 | msg.msg_controllen = sizeof(cmsgbuf); | |
823 | ||
824 | iov.iov_base = buf; | |
825 | iov.iov_len = sizeof(buf); | |
826 | msg.msg_iov = &iov; | |
827 | msg.msg_iovlen = 1; | |
828 | ||
6ee867dc SH |
829 | FD_ZERO(&rfds); |
830 | FD_SET(sock, &rfds); | |
831 | tv.tv_sec = 2; | |
832 | tv.tv_usec = 0; | |
ea56f722 | 833 | if (select(sock+1, &rfds, NULL, NULL, &tv) <= 0) { |
6ee867dc SH |
834 | fprintf(stderr, "Failed to select for scm_cred: %s\n", |
835 | strerror(errno)); | |
836 | return false; | |
837 | } | |
838 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
a05660a6 | 839 | if (ret < 0) { |
1420baf8 | 840 | fprintf(stderr, "Failed to receive scm_cred: %s\n", |
a05660a6 SH |
841 | strerror(errno)); |
842 | return false; | |
843 | } | |
844 | ||
845 | cmsg = CMSG_FIRSTHDR(&msg); | |
846 | ||
847 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
848 | cmsg->cmsg_level == SOL_SOCKET && | |
849 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
850 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
851 | } | |
852 | *v = buf[0]; | |
853 | ||
854 | return true; | |
855 | } | |
856 | ||
857 | ||
858 | /* | |
4775fba1 SH |
859 | * pid_to_ns - reads pids from a ucred over a socket, then writes the |
860 | * int value back over the socket. This shifts the pid from the | |
861 | * sender's pidns into tpid's pidns. | |
a05660a6 | 862 | */ |
4775fba1 | 863 | static void pid_to_ns(int sock, pid_t tpid) |
a05660a6 SH |
864 | { |
865 | char v = '0'; | |
866 | struct ucred cred; | |
867 | ||
868 | while (recv_creds(sock, &cred, &v)) { | |
869 | if (v == '1') | |
870 | exit(0); | |
a05660a6 SH |
871 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) |
872 | exit(1); | |
873 | } | |
874 | exit(0); | |
875 | } | |
876 | ||
877 | /* | |
4775fba1 | 878 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain |
a05660a6 | 879 | * in your old pidns. Only children which you fork will be in the target |
4775fba1 | 880 | * pidns. So the pid_to_ns_wrapper does the setns, then forks a child to |
a05660a6 SH |
881 | * actually convert pids |
882 | */ | |
4775fba1 | 883 | static void pid_to_ns_wrapper(int sock, pid_t tpid) |
a05660a6 | 884 | { |
ea56f722 | 885 | int newnsfd = -1, ret, cpipe[2]; |
a05660a6 SH |
886 | char fnam[100]; |
887 | pid_t cpid; | |
ea56f722 SH |
888 | struct timeval tv; |
889 | fd_set s; | |
890 | char v; | |
a05660a6 | 891 | |
c0adec85 SH |
892 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); |
893 | if (ret < 0 || ret >= sizeof(fnam)) | |
894 | exit(1); | |
a05660a6 SH |
895 | newnsfd = open(fnam, O_RDONLY); |
896 | if (newnsfd < 0) | |
897 | exit(1); | |
898 | if (setns(newnsfd, 0) < 0) | |
899 | exit(1); | |
900 | close(newnsfd); | |
901 | ||
ea56f722 SH |
902 | if (pipe(cpipe) < 0) |
903 | exit(1); | |
a05660a6 | 904 | |
ea56f722 SH |
905 | loop: |
906 | cpid = fork(); | |
a05660a6 SH |
907 | if (cpid < 0) |
908 | exit(1); | |
ea56f722 SH |
909 | |
910 | if (!cpid) { | |
911 | char b = '1'; | |
912 | close(cpipe[0]); | |
913 | if (write(cpipe[1], &b, sizeof(char)) < 0) { | |
914 | fprintf(stderr, "%s (child): erorr on write: %s\n", | |
915 | __func__, strerror(errno)); | |
916 | } | |
917 | close(cpipe[1]); | |
4775fba1 | 918 | pid_to_ns(sock, tpid); |
ea56f722 SH |
919 | } |
920 | // give the child 1 second to be done forking and | |
921 | // write it's ack | |
922 | FD_ZERO(&s); | |
923 | FD_SET(cpipe[0], &s); | |
924 | tv.tv_sec = 1; | |
925 | tv.tv_usec = 0; | |
926 | ret = select(cpipe[0]+1, &s, NULL, NULL, &tv); | |
927 | if (ret <= 0) | |
928 | goto again; | |
929 | ret = read(cpipe[0], &v, 1); | |
930 | if (ret != sizeof(char) || v != '1') { | |
931 | goto again; | |
932 | } | |
933 | ||
a05660a6 SH |
934 | if (!wait_for_pid(cpid)) |
935 | exit(1); | |
936 | exit(0); | |
ea56f722 SH |
937 | |
938 | again: | |
939 | kill(cpid, SIGKILL); | |
940 | wait_for_pid(cpid); | |
941 | goto loop; | |
a05660a6 SH |
942 | } |
943 | ||
944 | /* | |
945 | * To read cgroup files with a particular pid, we will setns into the child | |
946 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
947 | * the child ns - which does the cgm_get_value and writes the data to the pipe. | |
948 | */ | |
949 | static bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
950 | { | |
951 | int sock[2] = {-1, -1}; | |
952 | nih_local char *tmpdata = NULL; | |
953 | int ret; | |
954 | pid_t qpid, cpid = -1; | |
955 | bool answer = false; | |
956 | char v = '0'; | |
957 | struct ucred cred; | |
958 | struct timeval tv; | |
959 | fd_set s; | |
960 | ||
961 | if (!cgm_get_value(contrl, cg, file, &tmpdata)) | |
962 | return false; | |
963 | ||
964 | /* | |
965 | * Now we read the pids from returned data one by one, pass | |
966 | * them into a child in the target namespace, read back the | |
967 | * translated pids, and put them into our to-return data | |
968 | */ | |
969 | ||
970 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
971 | perror("socketpair"); | |
972 | exit(1); | |
973 | } | |
974 | ||
975 | cpid = fork(); | |
976 | if (cpid == -1) | |
977 | goto out; | |
978 | ||
979 | if (!cpid) // child | |
4775fba1 | 980 | pid_to_ns_wrapper(sock[1], tpid); |
a05660a6 SH |
981 | |
982 | char *ptr = tmpdata; | |
983 | cred.uid = 0; | |
984 | cred.gid = 0; | |
985 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
986 | cred.pid = qpid; | |
01e71852 SH |
987 | ret = send_creds(sock[0], &cred, v, true); |
988 | ||
989 | if (ret == SEND_CREDS_NOTSK) | |
990 | goto next; | |
991 | if (ret == SEND_CREDS_FAIL) | |
a05660a6 SH |
992 | goto out; |
993 | ||
994 | // read converted results | |
995 | FD_ZERO(&s); | |
996 | FD_SET(sock[0], &s); | |
6ee867dc | 997 | tv.tv_sec = 2; |
a05660a6 SH |
998 | tv.tv_usec = 0; |
999 | ret = select(sock[0]+1, &s, NULL, NULL, &tv); | |
1000 | if (ret <= 0) { | |
6ee867dc SH |
1001 | fprintf(stderr, "%s: select error waiting for pid from child: %s\n", |
1002 | __func__, strerror(errno)); | |
a05660a6 SH |
1003 | goto out; |
1004 | } | |
1005 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
6ee867dc SH |
1006 | fprintf(stderr, "%s: error reading pid from child: %s\n", |
1007 | __func__, strerror(errno)); | |
a05660a6 SH |
1008 | goto out; |
1009 | } | |
a05660a6 | 1010 | NIH_MUST( nih_strcat_sprintf(d, NULL, "%d\n", qpid) ); |
01e71852 | 1011 | next: |
a05660a6 SH |
1012 | ptr = strchr(ptr, '\n'); |
1013 | if (!ptr) | |
1014 | break; | |
1015 | ptr++; | |
1016 | } | |
1017 | ||
1018 | cred.pid = getpid(); | |
1019 | v = '1'; | |
01e71852 | 1020 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { |
a05660a6 | 1021 | // failed to ask child to exit |
6ee867dc SH |
1022 | fprintf(stderr, "%s: failed to ask child to exit: %s\n", |
1023 | __func__, strerror(errno)); | |
a05660a6 SH |
1024 | goto out; |
1025 | } | |
1026 | ||
1027 | answer = true; | |
1028 | ||
1029 | out: | |
1030 | if (cpid != -1) | |
1031 | wait_for_pid(cpid); | |
1032 | if (sock[0] != -1) { | |
1033 | close(sock[0]); | |
1034 | close(sock[1]); | |
1035 | } | |
1036 | return answer; | |
1037 | } | |
1038 | ||
99978832 SH |
1039 | static int cg_read(const char *path, char *buf, size_t size, off_t offset, |
1040 | struct fuse_file_info *fi) | |
1041 | { | |
99978832 | 1042 | struct fuse_context *fc = fuse_get_context(); |
8f6e8f5e | 1043 | struct file_info *f = (struct file_info *)fi->fh; |
99978832 SH |
1044 | nih_local struct cgm_keys *k = NULL; |
1045 | ||
443d13f5 | 1046 | if (f->type != LXC_TYPE_CGFILE) { |
b845ad01 SH |
1047 | fprintf(stderr, "Internal error: directory cache info used in cg_read\n"); |
1048 | return -EIO; | |
1049 | } | |
1050 | ||
99978832 | 1051 | if (offset) |
7253e0a4 | 1052 | return 0; |
99978832 SH |
1053 | |
1054 | if (!fc) | |
1055 | return -EIO; | |
1056 | ||
8f6e8f5e | 1057 | if (!f->controller) |
99978832 SH |
1058 | return -EINVAL; |
1059 | ||
8f6e8f5e | 1060 | if ((k = get_cgroup_key(f->controller, f->cgroup, f->file)) != NULL) { |
99978832 | 1061 | nih_local char *data = NULL; |
4775fba1 SH |
1062 | int s; |
1063 | bool r; | |
99978832 | 1064 | |
8f6e8f5e | 1065 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) |
f9a05025 SH |
1066 | // should never get here |
1067 | return -EACCES; | |
99978832 | 1068 | |
8f6e8f5e SH |
1069 | if (strcmp(f->file, "tasks") == 0 || |
1070 | strcmp(f->file, "/tasks") == 0 || | |
1071 | strcmp(f->file, "/cgroup.procs") == 0 || | |
1072 | strcmp(f->file, "cgroup.procs") == 0) | |
a05660a6 | 1073 | // special case - we have to translate the pids |
8f6e8f5e | 1074 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); |
a05660a6 | 1075 | else |
8f6e8f5e | 1076 | r = cgm_get_value(f->controller, f->cgroup, f->file, &data); |
a05660a6 | 1077 | |
4775fba1 | 1078 | if (!r) |
99978832 SH |
1079 | return -EINVAL; |
1080 | ||
4775fba1 SH |
1081 | if (!data) |
1082 | return 0; | |
99978832 SH |
1083 | s = strlen(data); |
1084 | if (s > size) | |
1085 | s = size; | |
1086 | memcpy(buf, data, s); | |
5ea0727e SH |
1087 | if (s > 0 && s < size && data[s-1] != '\n') |
1088 | buf[s++] = '\n'; | |
99978832 | 1089 | |
99978832 SH |
1090 | return s; |
1091 | } | |
1092 | ||
1093 | return -EINVAL; | |
1094 | } | |
1095 | ||
4775fba1 SH |
1096 | static void pid_from_ns(int sock, pid_t tpid) |
1097 | { | |
1098 | pid_t vpid; | |
1099 | struct ucred cred; | |
1100 | char v; | |
6ee867dc SH |
1101 | struct timeval tv; |
1102 | fd_set s; | |
1103 | int ret; | |
4775fba1 SH |
1104 | |
1105 | cred.uid = 0; | |
1106 | cred.gid = 0; | |
6ee867dc SH |
1107 | while (1) { |
1108 | FD_ZERO(&s); | |
1109 | FD_SET(sock, &s); | |
1110 | tv.tv_sec = 2; | |
1111 | tv.tv_usec = 0; | |
1112 | ret = select(sock+1, &s, NULL, NULL, &tv); | |
ea56f722 SH |
1113 | if (ret <= 0) { |
1114 | fprintf(stderr, "%s: bad select before read from parent: %s\n", | |
6ee867dc SH |
1115 | __func__, strerror(errno)); |
1116 | exit(1); | |
1117 | } | |
1118 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
1119 | fprintf(stderr, "%s: bad read from parent: %s\n", | |
1120 | __func__, strerror(errno)); | |
1121 | exit(1); | |
1122 | } | |
4775fba1 | 1123 | if (vpid == -1) // done |
01e71852 | 1124 | break; |
4775fba1 SH |
1125 | v = '0'; |
1126 | cred.pid = vpid; | |
01e71852 | 1127 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { |
4775fba1 SH |
1128 | v = '1'; |
1129 | cred.pid = getpid(); | |
01e71852 | 1130 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) |
4775fba1 SH |
1131 | exit(1); |
1132 | } | |
1133 | } | |
1134 | exit(0); | |
1135 | } | |
1136 | ||
1137 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
1138 | { | |
ea56f722 | 1139 | int newnsfd = -1, ret, cpipe[2]; |
4775fba1 SH |
1140 | char fnam[100]; |
1141 | pid_t cpid; | |
ea56f722 SH |
1142 | fd_set s; |
1143 | struct timeval tv; | |
1144 | char v; | |
4775fba1 | 1145 | |
c0adec85 SH |
1146 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); |
1147 | if (ret < 0 || ret >= sizeof(fnam)) | |
1148 | exit(1); | |
4775fba1 SH |
1149 | newnsfd = open(fnam, O_RDONLY); |
1150 | if (newnsfd < 0) | |
1151 | exit(1); | |
1152 | if (setns(newnsfd, 0) < 0) | |
1153 | exit(1); | |
1154 | close(newnsfd); | |
1155 | ||
ea56f722 SH |
1156 | if (pipe(cpipe) < 0) |
1157 | exit(1); | |
1158 | ||
1159 | loop: | |
4775fba1 SH |
1160 | cpid = fork(); |
1161 | ||
1162 | if (cpid < 0) | |
1163 | exit(1); | |
ea56f722 SH |
1164 | |
1165 | if (!cpid) { | |
1166 | char b = '1'; | |
1167 | close(cpipe[0]); | |
1168 | if (write(cpipe[1], &b, sizeof(char)) < 0) { | |
1169 | fprintf(stderr, "%s (child): erorr on write: %s\n", | |
1170 | __func__, strerror(errno)); | |
1171 | } | |
1172 | close(cpipe[1]); | |
4775fba1 | 1173 | pid_from_ns(sock, tpid); |
ea56f722 SH |
1174 | } |
1175 | ||
1176 | // give the child 1 second to be done forking and | |
1177 | // write it's ack | |
1178 | FD_ZERO(&s); | |
1179 | FD_SET(cpipe[0], &s); | |
1180 | tv.tv_sec = 1; | |
1181 | tv.tv_usec = 0; | |
1182 | ret = select(cpipe[0]+1, &s, NULL, NULL, &tv); | |
1183 | if (ret <= 0) | |
1184 | goto again; | |
1185 | ret = read(cpipe[0], &v, 1); | |
1186 | if (ret != sizeof(char) || v != '1') { | |
1187 | goto again; | |
1188 | } | |
1189 | ||
4775fba1 SH |
1190 | if (!wait_for_pid(cpid)) |
1191 | exit(1); | |
1192 | exit(0); | |
ea56f722 SH |
1193 | |
1194 | again: | |
1195 | kill(cpid, SIGKILL); | |
1196 | wait_for_pid(cpid); | |
1197 | goto loop; | |
4775fba1 SH |
1198 | } |
1199 | ||
1200 | static bool do_write_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, const char *buf) | |
1201 | { | |
1202 | int sock[2] = {-1, -1}; | |
1203 | pid_t qpid, cpid = -1; | |
1204 | bool answer = false, fail = false; | |
1205 | ||
1206 | /* | |
1207 | * write the pids to a socket, have helper in writer's pidns | |
1208 | * call movepid for us | |
1209 | */ | |
1210 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1211 | perror("socketpair"); | |
1212 | exit(1); | |
1213 | } | |
1214 | ||
1215 | cpid = fork(); | |
1216 | if (cpid == -1) | |
1217 | goto out; | |
1218 | ||
1219 | if (!cpid) // child | |
1220 | pid_from_ns_wrapper(sock[1], tpid); | |
1221 | ||
1222 | const char *ptr = buf; | |
1223 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
1224 | struct ucred cred; | |
1225 | char v; | |
1226 | ||
1227 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
6ee867dc SH |
1228 | fprintf(stderr, "%s: error writing pid to child: %s\n", |
1229 | __func__, strerror(errno)); | |
4775fba1 SH |
1230 | goto out; |
1231 | } | |
1232 | ||
01e71852 SH |
1233 | if (recv_creds(sock[0], &cred, &v)) { |
1234 | if (v == '0') { | |
1235 | if (!cgm_move_pid(contrl, cg, cred.pid)) | |
1236 | fail = true; | |
1237 | } | |
4775fba1 SH |
1238 | } |
1239 | ||
1240 | ptr = strchr(ptr, '\n'); | |
1241 | if (!ptr) | |
1242 | break; | |
1243 | ptr++; | |
1244 | } | |
1245 | ||
1246 | /* All good, write the value */ | |
1247 | qpid = -1; | |
1248 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
1420baf8 | 1249 | fprintf(stderr, "Warning: failed to ask child to exit\n"); |
4775fba1 SH |
1250 | |
1251 | if (!fail) | |
1252 | answer = true; | |
1253 | ||
1254 | out: | |
1255 | if (cpid != -1) | |
1256 | wait_for_pid(cpid); | |
1257 | if (sock[0] != -1) { | |
1258 | close(sock[0]); | |
1259 | close(sock[1]); | |
1260 | } | |
1261 | return answer; | |
1262 | } | |
1263 | ||
2ad6d2bd SH |
1264 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, |
1265 | struct fuse_file_info *fi) | |
1266 | { | |
2ad6d2bd | 1267 | struct fuse_context *fc = fuse_get_context(); |
47cbf0e5 | 1268 | nih_local char *localbuf = NULL; |
8f6e8f5e SH |
1269 | nih_local struct cgm_keys *k = NULL; |
1270 | struct file_info *f = (struct file_info *)fi->fh; | |
2ad6d2bd | 1271 | |
443d13f5 | 1272 | if (f->type != LXC_TYPE_CGFILE) { |
b845ad01 SH |
1273 | fprintf(stderr, "Internal error: directory cache info used in cg_write\n"); |
1274 | return -EIO; | |
1275 | } | |
1276 | ||
2ad6d2bd | 1277 | if (offset) |
7253e0a4 | 1278 | return 0; |
2ad6d2bd SH |
1279 | |
1280 | if (!fc) | |
1281 | return -EIO; | |
1282 | ||
47cbf0e5 SH |
1283 | localbuf = NIH_MUST( nih_alloc(NULL, size+1) ); |
1284 | localbuf[size] = '\0'; | |
1285 | memcpy(localbuf, buf, size); | |
2ad6d2bd | 1286 | |
8f6e8f5e | 1287 | if ((k = get_cgroup_key(f->controller, f->cgroup, f->file)) != NULL) { |
4775fba1 SH |
1288 | bool r; |
1289 | ||
8f6e8f5e | 1290 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) |
f9a05025 | 1291 | return -EACCES; |
2ad6d2bd | 1292 | |
8f6e8f5e SH |
1293 | if (strcmp(f->file, "tasks") == 0 || |
1294 | strcmp(f->file, "/tasks") == 0 || | |
1295 | strcmp(f->file, "/cgroup.procs") == 0 || | |
1296 | strcmp(f->file, "cgroup.procs") == 0) | |
4775fba1 | 1297 | // special case - we have to translate the pids |
8f6e8f5e | 1298 | r = do_write_pids(fc->pid, f->controller, f->cgroup, f->file, localbuf); |
4775fba1 | 1299 | else |
8f6e8f5e | 1300 | r = cgm_set_value(f->controller, f->cgroup, f->file, localbuf); |
4775fba1 SH |
1301 | |
1302 | if (!r) | |
2ad6d2bd SH |
1303 | return -EINVAL; |
1304 | ||
1305 | return size; | |
1306 | } | |
1307 | ||
1308 | return -EINVAL; | |
1309 | } | |
1310 | ||
341b21ad SH |
1311 | int cg_chown(const char *path, uid_t uid, gid_t gid) |
1312 | { | |
1313 | struct fuse_context *fc = fuse_get_context(); | |
1314 | nih_local char * cgdir = NULL; | |
1315 | char *fpath = NULL, *path1, *path2; | |
1316 | nih_local struct cgm_keys *k = NULL; | |
1317 | const char *cgroup; | |
1318 | nih_local char *controller = NULL; | |
1319 | ||
1320 | ||
1321 | if (!fc) | |
1322 | return -EIO; | |
1323 | ||
1324 | if (strcmp(path, "/cgroup") == 0) | |
1325 | return -EINVAL; | |
1326 | ||
1327 | controller = pick_controller_from_path(fc, path); | |
1328 | if (!controller) | |
f9a05025 | 1329 | return -EINVAL; |
341b21ad SH |
1330 | cgroup = find_cgroup_in_path(path); |
1331 | if (!cgroup) | |
1332 | /* this is just /cgroup/controller */ | |
1333 | return -EINVAL; | |
1334 | ||
1335 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
1336 | ||
1337 | if (!fpath) { | |
1338 | path1 = "/"; | |
1339 | path2 = cgdir; | |
1340 | } else { | |
1341 | path1 = cgdir; | |
1342 | path2 = fpath; | |
1343 | } | |
1344 | ||
1345 | if (is_child_cgroup(controller, path1, path2)) { | |
1346 | // get uid, gid, from '/tasks' file and make up a mode | |
1347 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1348 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
1349 | ||
1350 | } else | |
1351 | k = get_cgroup_key(controller, path1, path2); | |
1352 | ||
1353 | if (!k) | |
1354 | return -EINVAL; | |
1355 | ||
1356 | /* | |
1357 | * This being a fuse request, the uid and gid must be valid | |
1358 | * in the caller's namespace. So we can just check to make | |
1359 | * sure that the caller is root in his uid, and privileged | |
1360 | * over the file's current owner. | |
1361 | */ | |
1362 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) | |
f9a05025 | 1363 | return -EACCES; |
341b21ad SH |
1364 | |
1365 | if (!cgm_chown_file(controller, cgroup, uid, gid)) | |
1366 | return -EINVAL; | |
1367 | return 0; | |
1368 | } | |
2ad6d2bd | 1369 | |
fd2e4e03 SH |
1370 | int cg_chmod(const char *path, mode_t mode) |
1371 | { | |
0a1bb5ea SH |
1372 | struct fuse_context *fc = fuse_get_context(); |
1373 | nih_local char * cgdir = NULL; | |
1374 | char *fpath = NULL, *path1, *path2; | |
1375 | nih_local struct cgm_keys *k = NULL; | |
1376 | const char *cgroup; | |
1377 | nih_local char *controller = NULL; | |
1378 | ||
1379 | if (!fc) | |
1380 | return -EIO; | |
1381 | ||
1382 | if (strcmp(path, "/cgroup") == 0) | |
1383 | return -EINVAL; | |
1384 | ||
1385 | controller = pick_controller_from_path(fc, path); | |
1386 | if (!controller) | |
f9a05025 | 1387 | return -EINVAL; |
0a1bb5ea SH |
1388 | cgroup = find_cgroup_in_path(path); |
1389 | if (!cgroup) | |
1390 | /* this is just /cgroup/controller */ | |
1391 | return -EINVAL; | |
1392 | ||
1393 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
1394 | ||
1395 | if (!fpath) { | |
1396 | path1 = "/"; | |
1397 | path2 = cgdir; | |
1398 | } else { | |
1399 | path1 = cgdir; | |
1400 | path2 = fpath; | |
1401 | } | |
1402 | ||
1403 | if (is_child_cgroup(controller, path1, path2)) { | |
1404 | // get uid, gid, from '/tasks' file and make up a mode | |
1405 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1406 | k = get_cgroup_key(controller, cgroup, "tasks"); | |
1407 | ||
1408 | } else | |
1409 | k = get_cgroup_key(controller, path1, path2); | |
1410 | ||
1411 | if (!k) | |
1412 | return -EINVAL; | |
1413 | ||
1414 | /* | |
1415 | * This being a fuse request, the uid and gid must be valid | |
1416 | * in the caller's namespace. So we can just check to make | |
1417 | * sure that the caller is root in his uid, and privileged | |
1418 | * over the file's current owner. | |
1419 | */ | |
1420 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) | |
1421 | return -EPERM; | |
1422 | ||
1423 | if (!cgm_chmod_file(controller, cgroup, mode)) | |
1424 | return -EINVAL; | |
1425 | return 0; | |
fd2e4e03 SH |
1426 | } |
1427 | ||
ab54b798 SH |
1428 | int cg_mkdir(const char *path, mode_t mode) |
1429 | { | |
1430 | struct fuse_context *fc = fuse_get_context(); | |
1431 | nih_local struct cgm_keys **list = NULL; | |
1432 | char *fpath = NULL, *path1; | |
1433 | nih_local char * cgdir = NULL; | |
1434 | const char *cgroup; | |
1435 | nih_local char *controller = NULL; | |
1436 | ||
ab54b798 SH |
1437 | if (!fc) |
1438 | return -EIO; | |
1439 | ||
1440 | ||
1441 | controller = pick_controller_from_path(fc, path); | |
1442 | if (!controller) | |
f9a05025 | 1443 | return -EINVAL; |
ab54b798 SH |
1444 | |
1445 | cgroup = find_cgroup_in_path(path); | |
1446 | if (!cgroup) | |
f9a05025 | 1447 | return -EINVAL; |
ab54b798 SH |
1448 | |
1449 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
1450 | if (!fpath) | |
1451 | path1 = "/"; | |
1452 | else | |
1453 | path1 = cgdir; | |
1454 | ||
1455 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) | |
f9a05025 | 1456 | return -EACCES; |
ab54b798 SH |
1457 | |
1458 | ||
1459 | if (!cgm_create(controller, cgroup, fc->uid, fc->gid)) | |
1460 | return -EINVAL; | |
1461 | ||
1462 | return 0; | |
1463 | } | |
1464 | ||
50d8d5b5 SH |
1465 | static int cg_rmdir(const char *path) |
1466 | { | |
1467 | struct fuse_context *fc = fuse_get_context(); | |
1468 | nih_local struct cgm_keys **list = NULL; | |
1469 | char *fpath = NULL; | |
1470 | nih_local char * cgdir = NULL; | |
1471 | const char *cgroup; | |
1472 | nih_local char *controller = NULL; | |
1473 | ||
1474 | if (!fc) | |
1475 | return -EIO; | |
1476 | ||
1477 | ||
1478 | controller = pick_controller_from_path(fc, path); | |
1479 | if (!controller) | |
f9a05025 | 1480 | return -EINVAL; |
50d8d5b5 SH |
1481 | |
1482 | cgroup = find_cgroup_in_path(path); | |
1483 | if (!cgroup) | |
f9a05025 | 1484 | return -EINVAL; |
50d8d5b5 SH |
1485 | |
1486 | get_cgdir_and_path(cgroup, &cgdir, &fpath); | |
1487 | if (!fpath) | |
1488 | return -EINVAL; | |
1489 | ||
1490 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) | |
f9a05025 | 1491 | return -EACCES; |
50d8d5b5 SH |
1492 | |
1493 | if (!cgm_remove(controller, cgroup)) | |
1494 | return -EINVAL; | |
1495 | ||
1496 | return 0; | |
1497 | } | |
1498 | ||
2dc17609 SH |
1499 | static bool startswith(const char *line, const char *pref) |
1500 | { | |
1501 | if (strncmp(line, pref, strlen(pref)) == 0) | |
1502 | return true; | |
1503 | return false; | |
1504 | } | |
1505 | ||
1506 | static void get_mem_cached(char *memstat, unsigned long *v) | |
1507 | { | |
1508 | char *eol; | |
1509 | ||
1510 | *v = 0; | |
1511 | while (*memstat) { | |
1512 | if (startswith(memstat, "total_cache")) { | |
1513 | sscanf(memstat + 11, "%lu", v); | |
1514 | *v /= 1024; | |
1515 | return; | |
1516 | } | |
1517 | eol = strchr(memstat, '\n'); | |
1518 | if (!eol) | |
1519 | return; | |
1520 | memstat = eol+1; | |
1521 | } | |
1522 | } | |
1523 | ||
49878439 | 1524 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) |
2f919d9d | 1525 | { |
49878439 YY |
1526 | char *eol; |
1527 | char key[32]; | |
2f919d9d | 1528 | |
49878439 YY |
1529 | memset(key, 0, 32); |
1530 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
2f919d9d | 1531 | |
49878439 YY |
1532 | size_t len = strlen(key); |
1533 | *v = 0; | |
1534 | ||
1535 | while (*str) { | |
1536 | if (startswith(str, key)) { | |
2f919d9d SH |
1537 | sscanf(str + len, "%lu", v); |
1538 | return; | |
1539 | } | |
1540 | eol = strchr(str, '\n'); | |
49878439 | 1541 | if (!eol) |
2f919d9d | 1542 | return; |
49878439 YY |
1543 | str = eol+1; |
1544 | } | |
1545 | } | |
1546 | ||
2dc17609 SH |
1547 | static char *get_pid_cgroup(pid_t pid, const char *contrl) |
1548 | { | |
1549 | nih_local char *fnam = NULL; | |
1550 | FILE *f; | |
1551 | char *answer = NULL; | |
1552 | char *line = NULL; | |
1553 | size_t len = 0; | |
1554 | ||
1555 | fnam = NIH_MUST( nih_sprintf(NULL, "/proc/%d/cgroup", pid) ); | |
1556 | if (!(f = fopen(fnam, "r"))) | |
1557 | return false; | |
1558 | ||
1559 | while (getline(&line, &len, f) != -1) { | |
1560 | char *c1, *c2; | |
1561 | if (!line[0]) | |
1562 | continue; | |
1563 | c1 = strchr(line, ':'); | |
1564 | if (!c1) | |
1565 | goto out; | |
1566 | c1++; | |
1567 | c2 = strchr(c1, ':'); | |
1568 | if (!c2) | |
1569 | goto out; | |
1570 | *c2 = '\0'; | |
1571 | if (strcmp(c1, contrl) != 0) | |
1572 | continue; | |
1573 | c2++; | |
1574 | stripnewline(c2); | |
1575 | answer = NIH_MUST( nih_strdup(NULL, c2) ); | |
1576 | goto out; | |
1577 | } | |
1578 | ||
1579 | out: | |
1580 | fclose(f); | |
1581 | free(line); | |
1582 | return answer; | |
1583 | } | |
1584 | ||
758ad80c | 1585 | /* |
2ad6d2bd | 1586 | * FUSE ops for /proc |
758ad80c | 1587 | */ |
758ad80c | 1588 | |
23ce2127 SH |
1589 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, |
1590 | struct fuse_file_info *fi) | |
1591 | { | |
2dc17609 | 1592 | struct fuse_context *fc = fuse_get_context(); |
97f1f27b | 1593 | struct file_info *d = (struct file_info *)fi->fh; |
2dc17609 SH |
1594 | nih_local char *cg = get_pid_cgroup(fc->pid, "memory"); |
1595 | nih_local char *memlimit_str = NULL, *memusage_str = NULL, *memstat_str = NULL; | |
1596 | unsigned long memlimit = 0, memusage = 0, cached = 0, hosttotal = 0; | |
1597 | char *line = NULL; | |
1598 | size_t linelen = 0, total_len = 0; | |
97f1f27b YY |
1599 | char *cache = d->buf; |
1600 | size_t cache_size = d->buflen; | |
2dc17609 SH |
1601 | FILE *f; |
1602 | ||
97f1f27b YY |
1603 | if (offset){ |
1604 | if (offset > d->size) | |
1605 | return -EINVAL; | |
1606 | int left = d->size - offset; | |
1607 | total_len = left > size ? size: left; | |
1608 | memcpy(buf, cache + offset, total_len); | |
1609 | return total_len; | |
1610 | } | |
2dc17609 SH |
1611 | |
1612 | if (!cg) | |
1613 | return 0; | |
1614 | ||
1615 | if (!cgm_get_value("memory", cg, "memory.limit_in_bytes", &memlimit_str)) | |
1616 | return 0; | |
1617 | if (!cgm_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
1618 | return 0; | |
1619 | if (!cgm_get_value("memory", cg, "memory.stat", &memstat_str)) | |
1620 | return 0; | |
1621 | memlimit = strtoul(memlimit_str, NULL, 10); | |
1622 | memusage = strtoul(memusage_str, NULL, 10); | |
1623 | memlimit /= 1024; | |
1624 | memusage /= 1024; | |
1625 | get_mem_cached(memstat_str, &cached); | |
1626 | ||
1627 | f = fopen("/proc/meminfo", "r"); | |
1628 | if (!f) | |
1629 | return 0; | |
1630 | ||
1631 | while (getline(&line, &linelen, f) != -1) { | |
1632 | size_t l; | |
1633 | char *printme, lbuf[100]; | |
1634 | ||
1635 | memset(lbuf, 0, 100); | |
1636 | if (startswith(line, "MemTotal:")) { | |
1637 | sscanf(line+14, "%lu", &hosttotal); | |
1638 | if (hosttotal < memlimit) | |
1639 | memlimit = hosttotal; | |
1640 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
1641 | printme = lbuf; | |
1642 | } else if (startswith(line, "MemFree:")) { | |
1643 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
1644 | printme = lbuf; | |
1645 | } else if (startswith(line, "MemAvailable:")) { | |
1646 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage); | |
1647 | printme = lbuf; | |
1648 | } else if (startswith(line, "Buffers:")) { | |
1649 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
1650 | printme = lbuf; | |
1651 | } else if (startswith(line, "Cached:")) { | |
1652 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
1653 | printme = lbuf; | |
1654 | } else if (startswith(line, "SwapCached:")) { | |
1655 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
1656 | printme = lbuf; | |
1657 | } else | |
1658 | printme = line; | |
97f1f27b YY |
1659 | |
1660 | l = snprintf(cache, cache_size, "%s", printme); | |
1661 | cache += l; | |
1662 | cache_size -= l; | |
2f919d9d | 1663 | total_len += l; |
2dc17609 SH |
1664 | } |
1665 | ||
97f1f27b YY |
1666 | d->size = total_len; |
1667 | if (total_len > size ) total_len = size; | |
1668 | memcpy(buf, d->buf, total_len); | |
1669 | ||
92c84dc4 SH |
1670 | fclose(f); |
1671 | free(line); | |
2dc17609 | 1672 | return total_len; |
23ce2127 SH |
1673 | } |
1674 | ||
1675 | /* | |
1676 | * Read the cpuset.cpus for cg | |
1677 | * Return the answer in a nih_alloced string | |
1678 | */ | |
1679 | static char *get_cpuset(const char *cg) | |
1680 | { | |
1681 | char *answer; | |
1682 | ||
1683 | if (!cgm_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
1684 | return NULL; | |
1685 | return answer; | |
1686 | } | |
1687 | ||
fa47bb52 | 1688 | bool cpu_in_cpuset(int cpu, const char *cpuset); |
23ce2127 | 1689 | |
aeb56147 SH |
1690 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) |
1691 | { | |
1692 | int cpu; | |
1693 | ||
1694 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
1695 | return false; | |
1696 | return cpu_in_cpuset(cpu, cpuset); | |
1697 | } | |
1698 | ||
23ce2127 SH |
1699 | /* |
1700 | * check whether this is a '^processor" line in /proc/cpuinfo | |
1701 | */ | |
1702 | static bool is_processor_line(const char *line) | |
1703 | { | |
1704 | int cpu; | |
1705 | ||
1706 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
1707 | return true; | |
1708 | return false; | |
1709 | } | |
1710 | ||
23ce2127 SH |
1711 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, |
1712 | struct fuse_file_info *fi) | |
1713 | { | |
1714 | struct fuse_context *fc = fuse_get_context(); | |
97f1f27b | 1715 | struct file_info *d = (struct file_info *)fi->fh; |
23ce2127 SH |
1716 | nih_local char *cg = get_pid_cgroup(fc->pid, "cpuset"); |
1717 | nih_local char *cpuset = NULL; | |
1718 | char *line = NULL; | |
1719 | size_t linelen = 0, total_len = 0; | |
1720 | bool am_printing = false; | |
1721 | int curcpu = -1; | |
97f1f27b YY |
1722 | char *cache = d->buf; |
1723 | size_t cache_size = d->buflen; | |
23ce2127 SH |
1724 | FILE *f; |
1725 | ||
97f1f27b YY |
1726 | if (offset){ |
1727 | if (offset > d->size) | |
1728 | return -EINVAL; | |
1729 | int left = d->size - offset; | |
1730 | total_len = left > size ? size: left; | |
1731 | memcpy(buf, cache + offset, total_len); | |
2f919d9d | 1732 | return total_len; |
97f1f27b | 1733 | } |
23ce2127 SH |
1734 | |
1735 | if (!cg) | |
1736 | return 0; | |
1737 | ||
1738 | cpuset = get_cpuset(cg); | |
1739 | if (!cpuset) | |
1740 | return 0; | |
1741 | ||
1742 | f = fopen("/proc/cpuinfo", "r"); | |
1743 | if (!f) | |
1744 | return 0; | |
1745 | ||
1746 | while (getline(&line, &linelen, f) != -1) { | |
1747 | size_t l; | |
1748 | if (is_processor_line(line)) { | |
aeb56147 | 1749 | am_printing = cpuline_in_cpuset(line, cpuset); |
23ce2127 SH |
1750 | if (am_printing) { |
1751 | curcpu ++; | |
97f1f27b YY |
1752 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); |
1753 | if (l < cache_size){ | |
1754 | cache += l; | |
1755 | cache_size -= l; | |
1756 | total_len += l; | |
1757 | }else{ | |
1758 | cache += cache_size; | |
1759 | total_len += cache_size; | |
1760 | cache_size = 0; | |
1761 | break; | |
1762 | } | |
23ce2127 SH |
1763 | } |
1764 | continue; | |
1765 | } | |
1766 | if (am_printing) { | |
97f1f27b YY |
1767 | l = snprintf(cache, cache_size, "%s", line); |
1768 | if (l < cache_size) { | |
1769 | cache += l; | |
1770 | cache_size -= l; | |
1771 | total_len += l; | |
1772 | } else { | |
1773 | cache += cache_size; | |
1774 | total_len += cache_size; | |
1775 | cache_size = 0; | |
1776 | break; | |
1777 | } | |
23ce2127 SH |
1778 | } |
1779 | } | |
1780 | ||
97f1f27b YY |
1781 | d->size = total_len; |
1782 | if (total_len > size ) total_len = size; | |
1783 | ||
1784 | /* read from off 0 */ | |
1785 | memcpy(buf, d->buf, total_len); | |
1786 | ||
92c84dc4 SH |
1787 | fclose(f); |
1788 | free(line); | |
23ce2127 SH |
1789 | return total_len; |
1790 | } | |
1791 | ||
1792 | static int proc_stat_read(char *buf, size_t size, off_t offset, | |
1793 | struct fuse_file_info *fi) | |
1794 | { | |
aeb56147 | 1795 | struct fuse_context *fc = fuse_get_context(); |
97f1f27b | 1796 | struct file_info *d = (struct file_info *)fi->fh; |
aeb56147 SH |
1797 | nih_local char *cg = get_pid_cgroup(fc->pid, "cpuset"); |
1798 | nih_local char *cpuset = NULL; | |
1799 | char *line = NULL; | |
1800 | size_t linelen = 0, total_len = 0; | |
2a0fde62 | 1801 | int curcpu = -1; /* cpu numbering starts at 0 */ |
97f1f27b YY |
1802 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0; |
1803 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, | |
1804 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0; | |
1805 | #define CPUALL_MAX_SIZE BUF_RESERVE_SIZE | |
1806 | char cpuall[CPUALL_MAX_SIZE]; | |
1807 | /* reserve for cpu all */ | |
1808 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
1809 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
aeb56147 SH |
1810 | FILE *f; |
1811 | ||
97f1f27b YY |
1812 | if (offset){ |
1813 | if (offset > d->size) | |
1814 | return -EINVAL; | |
1815 | int left = d->size - offset; | |
1816 | total_len = left > size ? size: left; | |
1817 | memcpy(buf, d->buf + offset, total_len); | |
2f919d9d | 1818 | return total_len; |
97f1f27b | 1819 | } |
aeb56147 SH |
1820 | |
1821 | if (!cg) | |
1822 | return 0; | |
1823 | ||
1824 | cpuset = get_cpuset(cg); | |
1825 | if (!cpuset) | |
1826 | return 0; | |
1827 | ||
1828 | f = fopen("/proc/stat", "r"); | |
1829 | if (!f) | |
1830 | return 0; | |
1831 | ||
97f1f27b YY |
1832 | //skip first line |
1833 | if (getline(&line, &linelen, f) < 0) { | |
1834 | fprintf(stderr, "proc_stat_read read first line failed\n"); | |
1835 | goto out; | |
1836 | } | |
1837 | ||
aeb56147 SH |
1838 | while (getline(&line, &linelen, f) != -1) { |
1839 | size_t l; | |
1840 | int cpu; | |
2a0fde62 | 1841 | char cpu_char[10]; /* That's a lot of cores */ |
aeb56147 SH |
1842 | char *c; |
1843 | ||
2a0fde62 CB |
1844 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { |
1845 | /* not a ^cpuN line containing a number N, just print it */ | |
97f1f27b YY |
1846 | l = snprintf(cache, cache_size, "%s", line); |
1847 | if (l < cache_size){ | |
1848 | cache += l; | |
1849 | cache_size -= l; | |
1850 | total_len += l; | |
1851 | continue; | |
1852 | }else{ | |
1853 | //no more space, break it | |
1854 | cache += cache_size; | |
1855 | total_len += cache_size; | |
1856 | cache_size = 0; | |
1857 | break; | |
1858 | } | |
aeb56147 | 1859 | } |
2a0fde62 CB |
1860 | |
1861 | if (sscanf(cpu_char, "%d", &cpu) != 1) | |
1862 | continue; | |
aeb56147 SH |
1863 | if (!cpu_in_cpuset(cpu, cpuset)) |
1864 | continue; | |
1865 | curcpu ++; | |
1866 | ||
1867 | c = strchr(line, ' '); | |
1868 | if (!c) | |
1869 | continue; | |
25c5e8fb | 1870 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); |
97f1f27b YY |
1871 | cache += l; |
1872 | cache_size -= l; | |
aeb56147 | 1873 | total_len += l; |
2f919d9d | 1874 | |
97f1f27b YY |
1875 | if (sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user, &nice, &system, &idle, &iowait, &irq, |
1876 | &softirq, &steal, &guest) != 9) | |
1877 | continue; | |
1878 | user_sum += user; | |
1879 | nice_sum += nice; | |
1880 | system_sum += system; | |
1881 | idle_sum += idle; | |
1882 | iowait_sum += iowait; | |
1883 | irq_sum += irq; | |
1884 | softirq_sum += softirq; | |
1885 | steal_sum += steal; | |
2f919d9d | 1886 | guest_sum += guest; |
97f1f27b YY |
1887 | } |
1888 | ||
1889 | cache = d->buf; | |
1890 | ||
2f919d9d | 1891 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", |
97f1f27b YY |
1892 | "cpu ", user_sum, nice_sum, system_sum, idle_sum, iowait_sum, irq_sum, softirq_sum, steal_sum, guest_sum); |
1893 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE){ | |
1894 | memcpy(cache, cpuall, cpuall_len); | |
2f919d9d | 1895 | cache += cpuall_len; |
97f1f27b YY |
1896 | }else{ |
1897 | /* shouldn't happen */ | |
1898 | fprintf(stderr, "proc_stat_read copy cpuall failed, cpuall_len=%d\n", cpuall_len); | |
1899 | cpuall_len = 0; | |
1900 | } | |
1901 | ||
1902 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
1903 | total_len += cpuall_len; | |
1904 | d->size = total_len; | |
1905 | if (total_len > size ) total_len = size; | |
1906 | ||
1907 | memcpy(buf, d->buf, total_len); | |
97f1f27b | 1908 | out: |
92c84dc4 SH |
1909 | fclose(f); |
1910 | free(line); | |
aeb56147 | 1911 | return total_len; |
23ce2127 SH |
1912 | } |
1913 | ||
7bbf2246 SH |
1914 | /* |
1915 | * How to guess what to present for uptime? | |
1916 | * One thing we could do would be to take the date on the caller's | |
1917 | * memory.usage_in_bytes file, which should equal the time of creation | |
1918 | * of his cgroup. However, a task could be in a sub-cgroup of the | |
1919 | * container. The same problem exists if we try to look at the ages | |
1920 | * of processes in the caller's cgroup. | |
1921 | * | |
1922 | * So we'll fork a task that will enter the caller's pidns, mount a | |
1923 | * fresh procfs, get the age of /proc/1, and pass that back over a pipe. | |
1924 | * | |
1925 | * For the second uptime #, we'll do as Stéphane had done, just copy | |
1926 | * the number from /proc/uptime. Not sure how to best emulate 'idle' | |
1927 | * time. Maybe someone can come up with a good algorithm and submit a | |
1928 | * patch. Maybe something based on cpushare info? | |
1929 | */ | |
41bb9357 SH |
1930 | |
1931 | /* return age of the reaper for $pid, taken from ctime of its procdir */ | |
1932 | static long int get_pid1_time(pid_t pid) | |
1933 | { | |
1934 | char fnam[100]; | |
ea56f722 | 1935 | int fd, cpipe[2], ret; |
41bb9357 | 1936 | struct stat sb; |
ea56f722 SH |
1937 | pid_t cpid; |
1938 | struct timeval tv; | |
1939 | fd_set s; | |
1940 | char v; | |
41bb9357 SH |
1941 | |
1942 | if (unshare(CLONE_NEWNS)) | |
1943 | return 0; | |
1944 | ||
5ca64c2a SG |
1945 | if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL)) { |
1946 | perror("rslave mount failed"); | |
1947 | return 0; | |
1948 | } | |
1949 | ||
c0adec85 SH |
1950 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", pid); |
1951 | if (ret < 0 || ret >= sizeof(fnam)) | |
1952 | return 0; | |
1953 | ||
41bb9357 SH |
1954 | fd = open(fnam, O_RDONLY); |
1955 | if (fd < 0) { | |
1956 | perror("get_pid1_time open of ns/pid"); | |
1957 | return 0; | |
1958 | } | |
1959 | if (setns(fd, 0)) { | |
1960 | perror("get_pid1_time setns 1"); | |
1961 | close(fd); | |
1962 | return 0; | |
1963 | } | |
1964 | close(fd); | |
41bb9357 | 1965 | |
ea56f722 SH |
1966 | if (pipe(cpipe) < 0) |
1967 | exit(1); | |
41bb9357 | 1968 | |
ea56f722 SH |
1969 | loop: |
1970 | cpid = fork(); | |
1971 | if (cpid < 0) | |
41bb9357 | 1972 | return 0; |
ea56f722 SH |
1973 | |
1974 | if (!cpid) { | |
1975 | char b = '1'; | |
1976 | close(cpipe[0]); | |
1977 | if (write(cpipe[1], &b, sizeof(char)) < 0) { | |
1978 | fprintf(stderr, "%s (child): erorr on write: %s\n", | |
1979 | __func__, strerror(errno)); | |
1980 | } | |
1981 | close(cpipe[1]); | |
1982 | umount2("/proc", MNT_DETACH); | |
1983 | if (mount("proc", "/proc", "proc", 0, NULL)) { | |
1984 | perror("get_pid1_time mount"); | |
1985 | return 0; | |
1986 | } | |
1987 | ret = lstat("/proc/1", &sb); | |
1988 | if (ret) { | |
1989 | perror("get_pid1_time lstat"); | |
1990 | return 0; | |
1991 | } | |
1992 | return time(NULL) - sb.st_ctime; | |
41bb9357 | 1993 | } |
ea56f722 SH |
1994 | |
1995 | // give the child 1 second to be done forking and | |
1996 | // write it's ack | |
1997 | FD_ZERO(&s); | |
1998 | FD_SET(cpipe[0], &s); | |
1999 | tv.tv_sec = 1; | |
2000 | tv.tv_usec = 0; | |
2001 | ret = select(cpipe[0]+1, &s, NULL, NULL, &tv); | |
2002 | if (ret <= 0) | |
2003 | goto again; | |
2004 | ret = read(cpipe[0], &v, 1); | |
2005 | if (ret != sizeof(char) || v != '1') { | |
2006 | goto again; | |
41bb9357 | 2007 | } |
ea56f722 SH |
2008 | |
2009 | wait_for_pid(cpid); | |
2010 | exit(0); | |
2011 | ||
2012 | again: | |
2013 | kill(cpid, SIGKILL); | |
2014 | wait_for_pid(cpid); | |
2015 | goto loop; | |
41bb9357 SH |
2016 | } |
2017 | ||
2018 | static long int getreaperage(pid_t qpid) | |
2019 | { | |
2020 | int pid, mypipe[2], ret; | |
2021 | struct timeval tv; | |
2022 | fd_set s; | |
2023 | long int mtime, answer = 0; | |
2024 | ||
2025 | if (pipe(mypipe)) { | |
2026 | return 0; | |
2027 | } | |
2028 | ||
2029 | pid = fork(); | |
2030 | ||
2031 | if (!pid) { // child | |
2032 | mtime = get_pid1_time(qpid); | |
2033 | if (write(mypipe[1], &mtime, sizeof(mtime)) != sizeof(mtime)) | |
2034 | fprintf(stderr, "Warning: bad write from getreaperage\n"); | |
2035 | exit(0); | |
2036 | } | |
2037 | ||
2038 | close(mypipe[1]); | |
2039 | FD_ZERO(&s); | |
2040 | FD_SET(mypipe[0], &s); | |
2041 | tv.tv_sec = 1; | |
2042 | tv.tv_usec = 0; | |
2043 | ret = select(mypipe[0]+1, &s, NULL, NULL, &tv); | |
ea56f722 | 2044 | if (ret <= 0) { |
41bb9357 SH |
2045 | perror("select"); |
2046 | goto out; | |
2047 | } | |
2048 | if (!ret) { | |
1420baf8 | 2049 | fprintf(stderr, "timed out\n"); |
41bb9357 SH |
2050 | goto out; |
2051 | } | |
2052 | if (read(mypipe[0], &mtime, sizeof(mtime)) != sizeof(mtime)) { | |
2053 | perror("read"); | |
2054 | goto out; | |
2055 | } | |
2056 | answer = mtime; | |
2057 | ||
2058 | out: | |
2059 | wait_for_pid(pid); | |
2060 | close(mypipe[0]); | |
2061 | return answer; | |
2062 | } | |
2063 | ||
2064 | static long int getprocidle(void) | |
2065 | { | |
2066 | FILE *f = fopen("/proc/uptime", "r"); | |
2067 | long int age, idle; | |
92c84dc4 | 2068 | int ret; |
41bb9357 SH |
2069 | if (!f) |
2070 | return 0; | |
92c84dc4 SH |
2071 | ret = fscanf(f, "%ld %ld", &age, &idle); |
2072 | fclose(f); | |
2073 | if (ret != 2) | |
41bb9357 SH |
2074 | return 0; |
2075 | return idle; | |
2076 | } | |
2077 | ||
2078 | /* | |
2079 | * We read /proc/uptime and reuse its second field. | |
2080 | * For the first field, we use the mtime for the reaper for | |
2081 | * the calling pid as returned by getreaperage | |
2082 | */ | |
23ce2127 SH |
2083 | static int proc_uptime_read(char *buf, size_t size, off_t offset, |
2084 | struct fuse_file_info *fi) | |
2085 | { | |
41bb9357 | 2086 | struct fuse_context *fc = fuse_get_context(); |
97f1f27b | 2087 | struct file_info *d = (struct file_info *)fi->fh; |
41bb9357 SH |
2088 | long int reaperage = getreaperage(fc->pid);; |
2089 | long int idletime = getprocidle(); | |
97f1f27b | 2090 | size_t total_len = 0; |
41bb9357 | 2091 | |
97f1f27b YY |
2092 | if (offset){ |
2093 | if (offset > d->size) | |
2094 | return -EINVAL; | |
2095 | return 0; | |
2096 | } | |
2097 | ||
2098 | total_len = snprintf(buf, size, "%ld %ld\n", reaperage, idletime); | |
2099 | d->size = total_len; | |
2100 | return total_len; | |
23ce2127 SH |
2101 | } |
2102 | ||
49878439 YY |
2103 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, |
2104 | struct fuse_file_info *fi) | |
2105 | { | |
2106 | char dev_name[72]; | |
2107 | struct fuse_context *fc = fuse_get_context(); | |
97f1f27b | 2108 | struct file_info *d = (struct file_info *)fi->fh; |
49878439 YY |
2109 | nih_local char *cg = get_pid_cgroup(fc->pid, "blkio"); |
2110 | nih_local char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
2111 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
2112 | unsigned long read = 0, write = 0; | |
2113 | unsigned long read_merged = 0, write_merged = 0; | |
2114 | unsigned long read_sectors = 0, write_sectors = 0; | |
2115 | unsigned long read_ticks = 0, write_ticks = 0; | |
2116 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
2117 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
2118 | char *line = NULL; | |
2119 | size_t linelen = 0, total_len = 0; | |
2120 | unsigned int major = 0, minor = 0; | |
2121 | int i = 0; | |
2122 | FILE *f; | |
2123 | ||
97f1f27b YY |
2124 | if (offset){ |
2125 | if (offset > d->size) | |
2126 | return -EINVAL; | |
2127 | return 0; | |
2128 | } | |
49878439 YY |
2129 | |
2130 | if (!cg) | |
2131 | return 0; | |
2132 | ||
2133 | if (!cgm_get_value("blkio", cg, "blkio.io_serviced", &io_serviced_str)) | |
2134 | return 0; | |
2135 | if (!cgm_get_value("blkio", cg, "blkio.io_merged", &io_merged_str)) | |
2136 | return 0; | |
2137 | if (!cgm_get_value("blkio", cg, "blkio.io_service_bytes", &io_service_bytes_str)) | |
2138 | return 0; | |
2139 | if (!cgm_get_value("blkio", cg, "blkio.io_wait_time", &io_wait_time_str)) | |
2140 | return 0; | |
2141 | if (!cgm_get_value("blkio", cg, "blkio.io_service_time", &io_service_time_str)) | |
2142 | return 0; | |
2143 | ||
2144 | ||
2145 | f = fopen("/proc/diskstats", "r"); | |
2146 | if (!f) | |
2147 | return 0; | |
2148 | ||
2149 | while (getline(&line, &linelen, f) != -1) { | |
2150 | size_t l; | |
2151 | char *printme, lbuf[256]; | |
2152 | ||
c0adec85 | 2153 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); |
49878439 YY |
2154 | if(i == 3){ |
2155 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
2156 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
2157 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
2158 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
2159 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
2160 | read_sectors = read_sectors/512; | |
2161 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
2162 | write_sectors = write_sectors/512; | |
2f919d9d | 2163 | |
49878439 YY |
2164 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); |
2165 | rd_svctm = rd_svctm/1000000; | |
2166 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
2167 | rd_wait = rd_wait/1000000; | |
2168 | read_ticks = rd_svctm + rd_wait; | |
2169 | ||
2170 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
2171 | wr_svctm = wr_svctm/1000000; | |
2172 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
2173 | wr_wait = wr_wait/1000000; | |
2174 | write_ticks = wr_svctm + wr_wait; | |
2175 | ||
2176 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
2177 | tot_ticks = tot_ticks/1000000; | |
2178 | }else{ | |
2179 | continue; | |
2180 | } | |
2181 | ||
2182 | memset(lbuf, 0, 256); | |
2183 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) { | |
2f919d9d | 2184 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", |
49878439 YY |
2185 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, |
2186 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
2187 | printme = lbuf; | |
2188 | } else | |
2189 | continue; | |
2190 | ||
2191 | l = snprintf(buf, size, "%s", printme); | |
2192 | buf += l; | |
2193 | size -= l; | |
2194 | total_len += l; | |
2195 | } | |
2196 | ||
97f1f27b YY |
2197 | d->size = total_len; |
2198 | ||
49878439 YY |
2199 | fclose(f); |
2200 | free(line); | |
2201 | return total_len; | |
2202 | } | |
2203 | ||
23ce2127 SH |
2204 | static off_t get_procfile_size(const char *which) |
2205 | { | |
2206 | FILE *f = fopen(which, "r"); | |
2207 | char *line = NULL; | |
2208 | size_t len = 0; | |
2209 | ssize_t sz, answer = 0; | |
2210 | if (!f) | |
2211 | return 0; | |
2212 | ||
2213 | while ((sz = getline(&line, &len, f)) != -1) | |
2214 | answer += sz; | |
2215 | fclose (f); | |
92c84dc4 | 2216 | free(line); |
23ce2127 SH |
2217 | |
2218 | return answer; | |
2219 | } | |
2220 | ||
758ad80c SH |
2221 | static int proc_getattr(const char *path, struct stat *sb) |
2222 | { | |
35629743 SH |
2223 | struct timespec now; |
2224 | ||
2225 | memset(sb, 0, sizeof(struct stat)); | |
2226 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
2227 | return -EINVAL; | |
2228 | sb->st_uid = sb->st_gid = 0; | |
2229 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
2230 | if (strcmp(path, "/proc") == 0) { | |
2231 | sb->st_mode = S_IFDIR | 00555; | |
2232 | sb->st_nlink = 2; | |
2233 | return 0; | |
2234 | } | |
2235 | if (strcmp(path, "/proc/meminfo") == 0 || | |
2236 | strcmp(path, "/proc/cpuinfo") == 0 || | |
2237 | strcmp(path, "/proc/uptime") == 0 || | |
49878439 YY |
2238 | strcmp(path, "/proc/stat") == 0 || |
2239 | strcmp(path, "/proc/diskstats") == 0) { | |
7253e0a4 | 2240 | sb->st_size = 0; |
35629743 SH |
2241 | sb->st_mode = S_IFREG | 00444; |
2242 | sb->st_nlink = 1; | |
2243 | return 0; | |
2244 | } | |
2245 | ||
2246 | return -ENOENT; | |
2247 | } | |
2248 | ||
2249 | static int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
2250 | struct fuse_file_info *fi) | |
2251 | { | |
2252 | if (filler(buf, "cpuinfo", NULL, 0) != 0 || | |
2253 | filler(buf, "meminfo", NULL, 0) != 0 || | |
2254 | filler(buf, "stat", NULL, 0) != 0 || | |
49878439 YY |
2255 | filler(buf, "uptime", NULL, 0) != 0 || |
2256 | filler(buf, "diskstats", NULL, 0) != 0) | |
758ad80c | 2257 | return -EINVAL; |
758ad80c SH |
2258 | return 0; |
2259 | } | |
2260 | ||
35629743 SH |
2261 | static int proc_open(const char *path, struct fuse_file_info *fi) |
2262 | { | |
96fc5ee6 SH |
2263 | int type = -1; |
2264 | struct file_info *info; | |
2265 | ||
2266 | if (strcmp(path, "/proc/meminfo") == 0) | |
2267 | type = LXC_TYPE_PROC_MEMINFO; | |
2268 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
2269 | type = LXC_TYPE_PROC_CPUINFO; | |
2270 | else if (strcmp(path, "/proc/uptime") == 0) | |
2271 | type = LXC_TYPE_PROC_UPTIME; | |
2272 | else if (strcmp(path, "/proc/stat") == 0) | |
2273 | type = LXC_TYPE_PROC_STAT; | |
2274 | else if (strcmp(path, "/proc/diskstats") == 0) | |
2275 | type = LXC_TYPE_PROC_DISKSTATS; | |
2276 | if (type == -1) | |
2277 | return -ENOENT; | |
2278 | ||
2279 | info = NIH_MUST( nih_alloc(NULL, sizeof(*info)) ); | |
2280 | memset(info, 0, sizeof(*info)); | |
2281 | info->type = type; | |
2282 | ||
97f1f27b | 2283 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; |
25c5e8fb | 2284 | info->buf = NIH_MUST( nih_alloc(info, info->buflen) ); |
97f1f27b YY |
2285 | memset(info->buf, 0, info->buflen); |
2286 | /* set actual size to buffer size */ | |
2f919d9d | 2287 | info->size = info->buflen; |
97f1f27b | 2288 | |
96fc5ee6 SH |
2289 | fi->fh = (unsigned long)info; |
2290 | return 0; | |
2291 | } | |
2292 | ||
2293 | static int proc_release(const char *path, struct fuse_file_info *fi) | |
2294 | { | |
2295 | struct file_info *f = (struct file_info *)fi->fh; | |
2296 | ||
2297 | do_release_file_info(f); | |
2298 | return 0; | |
35629743 SH |
2299 | } |
2300 | ||
35629743 SH |
2301 | static int proc_read(const char *path, char *buf, size_t size, off_t offset, |
2302 | struct fuse_file_info *fi) | |
2303 | { | |
96fc5ee6 SH |
2304 | struct file_info *f = (struct file_info *) fi->fh; |
2305 | ||
2306 | switch (f->type) { | |
2f919d9d | 2307 | case LXC_TYPE_PROC_MEMINFO: |
23ce2127 | 2308 | return proc_meminfo_read(buf, size, offset, fi); |
96fc5ee6 | 2309 | case LXC_TYPE_PROC_CPUINFO: |
23ce2127 | 2310 | return proc_cpuinfo_read(buf, size, offset, fi); |
96fc5ee6 | 2311 | case LXC_TYPE_PROC_UPTIME: |
23ce2127 | 2312 | return proc_uptime_read(buf, size, offset, fi); |
96fc5ee6 | 2313 | case LXC_TYPE_PROC_STAT: |
23ce2127 | 2314 | return proc_stat_read(buf, size, offset, fi); |
96fc5ee6 | 2315 | case LXC_TYPE_PROC_DISKSTATS: |
49878439 | 2316 | return proc_diskstats_read(buf, size, offset, fi); |
96fc5ee6 SH |
2317 | default: |
2318 | return -EINVAL; | |
2319 | } | |
35629743 SH |
2320 | } |
2321 | ||
2ad6d2bd SH |
2322 | /* |
2323 | * FUSE ops for / | |
2324 | * these just delegate to the /proc and /cgroup ops as | |
2325 | * needed | |
2326 | */ | |
758ad80c SH |
2327 | |
2328 | static int lxcfs_getattr(const char *path, struct stat *sb) | |
2329 | { | |
2330 | if (strcmp(path, "/") == 0) { | |
2331 | sb->st_mode = S_IFDIR | 00755; | |
2332 | sb->st_nlink = 2; | |
2333 | return 0; | |
2334 | } | |
2335 | if (strncmp(path, "/cgroup", 7) == 0) { | |
2336 | return cg_getattr(path, sb); | |
2337 | } | |
35629743 | 2338 | if (strncmp(path, "/proc", 5) == 0) { |
758ad80c SH |
2339 | return proc_getattr(path, sb); |
2340 | } | |
2341 | return -EINVAL; | |
2342 | } | |
2343 | ||
2344 | static int lxcfs_opendir(const char *path, struct fuse_file_info *fi) | |
2345 | { | |
2346 | if (strcmp(path, "/") == 0) | |
2347 | return 0; | |
2348 | ||
2349 | if (strncmp(path, "/cgroup", 7) == 0) { | |
2350 | return cg_opendir(path, fi); | |
2351 | } | |
35629743 SH |
2352 | if (strcmp(path, "/proc") == 0) |
2353 | return 0; | |
2354 | return -ENOENT; | |
758ad80c SH |
2355 | } |
2356 | ||
2357 | static int lxcfs_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
2358 | struct fuse_file_info *fi) | |
2359 | { | |
2360 | if (strcmp(path, "/") == 0) { | |
2361 | if (filler(buf, "proc", NULL, 0) != 0 || | |
2362 | filler(buf, "cgroup", NULL, 0) != 0) | |
2363 | return -EINVAL; | |
2364 | return 0; | |
2365 | } | |
35629743 | 2366 | if (strncmp(path, "/cgroup", 7) == 0) |
758ad80c | 2367 | return cg_readdir(path, buf, filler, offset, fi); |
35629743 SH |
2368 | if (strcmp(path, "/proc") == 0) |
2369 | return proc_readdir(path, buf, filler, offset, fi); | |
758ad80c SH |
2370 | return -EINVAL; |
2371 | } | |
2372 | ||
2373 | static int lxcfs_releasedir(const char *path, struct fuse_file_info *fi) | |
2374 | { | |
2375 | if (strcmp(path, "/") == 0) | |
2376 | return 0; | |
2377 | if (strncmp(path, "/cgroup", 7) == 0) { | |
2378 | return cg_releasedir(path, fi); | |
2379 | } | |
35629743 SH |
2380 | if (strcmp(path, "/proc") == 0) |
2381 | return 0; | |
758ad80c SH |
2382 | return -EINVAL; |
2383 | } | |
2384 | ||
99978832 SH |
2385 | static int lxcfs_open(const char *path, struct fuse_file_info *fi) |
2386 | { | |
35629743 | 2387 | if (strncmp(path, "/cgroup", 7) == 0) |
99978832 | 2388 | return cg_open(path, fi); |
35629743 SH |
2389 | if (strncmp(path, "/proc", 5) == 0) |
2390 | return proc_open(path, fi); | |
99978832 SH |
2391 | |
2392 | return -EINVAL; | |
2393 | } | |
2394 | ||
2395 | static int lxcfs_read(const char *path, char *buf, size_t size, off_t offset, | |
2396 | struct fuse_file_info *fi) | |
2397 | { | |
35629743 | 2398 | if (strncmp(path, "/cgroup", 7) == 0) |
99978832 | 2399 | return cg_read(path, buf, size, offset, fi); |
35629743 SH |
2400 | if (strncmp(path, "/proc", 5) == 0) |
2401 | return proc_read(path, buf, size, offset, fi); | |
99978832 SH |
2402 | |
2403 | return -EINVAL; | |
2404 | } | |
2405 | ||
2ad6d2bd SH |
2406 | int lxcfs_write(const char *path, const char *buf, size_t size, off_t offset, |
2407 | struct fuse_file_info *fi) | |
2408 | { | |
2409 | if (strncmp(path, "/cgroup", 7) == 0) { | |
2410 | return cg_write(path, buf, size, offset, fi); | |
2411 | } | |
2412 | ||
2413 | return -EINVAL; | |
2414 | } | |
2415 | ||
99978832 SH |
2416 | static int lxcfs_flush(const char *path, struct fuse_file_info *fi) |
2417 | { | |
2418 | return 0; | |
2419 | } | |
2420 | ||
2421 | static int lxcfs_release(const char *path, struct fuse_file_info *fi) | |
758ad80c | 2422 | { |
8f6e8f5e SH |
2423 | if (strncmp(path, "/cgroup", 7) == 0) |
2424 | return cg_release(path, fi); | |
8f6e8f5e | 2425 | if (strncmp(path, "/proc", 5) == 0) |
96fc5ee6 | 2426 | return proc_release(path, fi); |
8f6e8f5e SH |
2427 | |
2428 | return -EINVAL; | |
99978832 SH |
2429 | } |
2430 | ||
2431 | static int lxcfs_fsync(const char *path, int datasync, struct fuse_file_info *fi) | |
2432 | { | |
2433 | return 0; | |
758ad80c SH |
2434 | } |
2435 | ||
ab54b798 SH |
2436 | int lxcfs_mkdir(const char *path, mode_t mode) |
2437 | { | |
2438 | if (strncmp(path, "/cgroup", 7) == 0) | |
2439 | return cg_mkdir(path, mode); | |
2440 | ||
2441 | return -EINVAL; | |
2442 | } | |
2443 | ||
341b21ad SH |
2444 | int lxcfs_chown(const char *path, uid_t uid, gid_t gid) |
2445 | { | |
2446 | if (strncmp(path, "/cgroup", 7) == 0) | |
2447 | return cg_chown(path, uid, gid); | |
2448 | ||
2449 | return -EINVAL; | |
2450 | } | |
2451 | ||
2ad6d2bd SH |
2452 | /* |
2453 | * cat first does a truncate before doing ops->write. This doesn't | |
2454 | * really make sense for cgroups. So just return 0 always but do | |
2455 | * nothing. | |
2456 | */ | |
2457 | int lxcfs_truncate(const char *path, off_t newsize) | |
2458 | { | |
2459 | if (strncmp(path, "/cgroup", 7) == 0) | |
2460 | return 0; | |
2461 | return -EINVAL; | |
2462 | } | |
2463 | ||
50d8d5b5 SH |
2464 | int lxcfs_rmdir(const char *path) |
2465 | { | |
2466 | if (strncmp(path, "/cgroup", 7) == 0) | |
2467 | return cg_rmdir(path); | |
2468 | return -EINVAL; | |
2469 | } | |
2470 | ||
fd2e4e03 SH |
2471 | int lxcfs_chmod(const char *path, mode_t mode) |
2472 | { | |
2473 | if (strncmp(path, "/cgroup", 7) == 0) | |
2474 | return cg_chmod(path, mode); | |
2475 | return -EINVAL; | |
2476 | } | |
2477 | ||
758ad80c SH |
2478 | const struct fuse_operations lxcfs_ops = { |
2479 | .getattr = lxcfs_getattr, | |
2480 | .readlink = NULL, | |
2481 | .getdir = NULL, | |
2482 | .mknod = NULL, | |
ab54b798 | 2483 | .mkdir = lxcfs_mkdir, |
758ad80c | 2484 | .unlink = NULL, |
50d8d5b5 | 2485 | .rmdir = lxcfs_rmdir, |
758ad80c SH |
2486 | .symlink = NULL, |
2487 | .rename = NULL, | |
2488 | .link = NULL, | |
fd2e4e03 | 2489 | .chmod = lxcfs_chmod, |
341b21ad | 2490 | .chown = lxcfs_chown, |
2ad6d2bd | 2491 | .truncate = lxcfs_truncate, |
758ad80c | 2492 | .utime = NULL, |
99978832 SH |
2493 | |
2494 | .open = lxcfs_open, | |
2495 | .read = lxcfs_read, | |
2496 | .release = lxcfs_release, | |
2ad6d2bd | 2497 | .write = lxcfs_write, |
99978832 | 2498 | |
758ad80c | 2499 | .statfs = NULL, |
99978832 SH |
2500 | .flush = lxcfs_flush, |
2501 | .fsync = lxcfs_fsync, | |
758ad80c SH |
2502 | |
2503 | .setxattr = NULL, | |
2504 | .getxattr = NULL, | |
2505 | .listxattr = NULL, | |
2506 | .removexattr = NULL, | |
2507 | ||
2508 | .opendir = lxcfs_opendir, | |
2509 | .readdir = lxcfs_readdir, | |
2510 | .releasedir = lxcfs_releasedir, | |
2511 | ||
2512 | .fsyncdir = NULL, | |
2513 | .init = NULL, | |
2514 | .destroy = NULL, | |
2515 | .access = NULL, | |
2516 | .create = NULL, | |
2517 | .ftruncate = NULL, | |
2518 | .fgetattr = NULL, | |
2519 | }; | |
2520 | ||
99978832 | 2521 | static void usage(const char *me) |
758ad80c SH |
2522 | { |
2523 | fprintf(stderr, "Usage:\n"); | |
2524 | fprintf(stderr, "\n"); | |
0b0f73db SH |
2525 | fprintf(stderr, "%s mountpoint\n", me); |
2526 | fprintf(stderr, "%s -h\n", me); | |
758ad80c SH |
2527 | exit(1); |
2528 | } | |
2529 | ||
99978832 | 2530 | static bool is_help(char *w) |
758ad80c SH |
2531 | { |
2532 | if (strcmp(w, "-h") == 0 || | |
2533 | strcmp(w, "--help") == 0 || | |
2534 | strcmp(w, "-help") == 0 || | |
2535 | strcmp(w, "help") == 0) | |
2536 | return true; | |
2537 | return false; | |
2538 | } | |
2539 | ||
0b0f73db SH |
2540 | void swallow_arg(int *argcp, char *argv[], char *which) |
2541 | { | |
2542 | int i; | |
2543 | ||
2544 | for (i = 1; argv[i]; i++) { | |
2545 | if (strcmp(argv[i], which) != 0) | |
2546 | continue; | |
2547 | for (; argv[i]; i++) { | |
2548 | argv[i] = argv[i+1]; | |
2549 | } | |
2550 | (*argcp)--; | |
2551 | return; | |
2552 | } | |
2553 | } | |
2554 | ||
2555 | void swallow_option(int *argcp, char *argv[], char *opt, char *v) | |
2556 | { | |
2557 | int i; | |
2558 | ||
2559 | for (i = 1; argv[i]; i++) { | |
2560 | if (!argv[i+1]) | |
2561 | continue; | |
2562 | if (strcmp(argv[i], opt) != 0) | |
2563 | continue; | |
2564 | if (strcmp(argv[i+1], v) != 0) { | |
2565 | fprintf(stderr, "Warning: unexpected fuse option %s\n", v); | |
2566 | exit(1); | |
2567 | } | |
2568 | for (; argv[i+1]; i++) { | |
2569 | argv[i] = argv[i+2]; | |
2570 | } | |
2571 | (*argcp) -= 2; | |
2572 | return; | |
2573 | } | |
2574 | } | |
2575 | ||
758ad80c SH |
2576 | int main(int argc, char *argv[]) |
2577 | { | |
c0adec85 | 2578 | int ret = -1; |
e5d26e0b | 2579 | struct lxcfs_state *d = NULL; |
0b0f73db SH |
2580 | /* |
2581 | * what we pass to fuse_main is: | |
2582 | * argv[0] -s -f -o allow_other,directio argv[1] NULL | |
2583 | */ | |
2584 | #define NARGS 7 | |
2585 | char *newargv[7]; | |
758ad80c | 2586 | |
0b0f73db SH |
2587 | /* accomodate older init scripts */ |
2588 | swallow_arg(&argc, argv, "-s"); | |
2589 | swallow_arg(&argc, argv, "-f"); | |
2590 | swallow_option(&argc, argv, "-o", "allow_other"); | |
2591 | ||
2e9c0b32 SH |
2592 | if (argc == 2 && strcmp(argv[1], "--version") == 0) { |
2593 | fprintf(stderr, "%s\n", VERSION); | |
2594 | exit(0); | |
2595 | } | |
0b0f73db | 2596 | if (argc != 2 || is_help(argv[1])) |
758ad80c SH |
2597 | usage(argv[0]); |
2598 | ||
0b0f73db SH |
2599 | d = NIH_MUST( malloc(sizeof(*d)) ); |
2600 | ||
2601 | newargv[0] = argv[0]; | |
2602 | newargv[1] = "-s"; | |
2603 | newargv[2] = "-f"; | |
2604 | newargv[3] = "-o"; | |
7253e0a4 | 2605 | newargv[4] = "allow_other,direct_io"; |
0b0f73db SH |
2606 | newargv[5] = argv[1]; |
2607 | newargv[6] = NULL; | |
758ad80c SH |
2608 | |
2609 | if (!cgm_escape_cgroup()) | |
2610 | fprintf(stderr, "WARNING: failed to escape to root cgroup\n"); | |
2611 | ||
2612 | if (!cgm_get_controllers(&d->subsystems)) | |
c0adec85 | 2613 | goto out; |
758ad80c | 2614 | |
0b0f73db | 2615 | ret = fuse_main(NARGS - 1, newargv, &lxcfs_ops, d); |
758ad80c | 2616 | |
c0adec85 | 2617 | out: |
e5d26e0b | 2618 | free(d); |
758ad80c | 2619 | return ret; |
2183082c | 2620 | } |