]>
Commit | Line | Data |
---|---|---|
237e200e SH |
1 | /* lxcfs |
2 | * | |
3 | * Copyright © 2014-2016 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
5 | * | |
6 | * See COPYING file for details. | |
7 | */ | |
8 | ||
9 | #define FUSE_USE_VERSION 26 | |
10 | ||
11 | #include <stdio.h> | |
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> | |
22 | #include <sched.h> | |
23 | #include <pthread.h> | |
24 | #include <linux/sched.h> | |
25 | #include <sys/param.h> | |
26 | #include <sys/socket.h> | |
27 | #include <sys/mount.h> | |
28 | #include <sys/epoll.h> | |
29 | #include <wait.h> | |
30 | ||
237e200e SH |
31 | #include "bindings.h" |
32 | ||
33 | #include "config.h" // for VERSION | |
34 | ||
35 | enum { | |
36 | LXC_TYPE_CGDIR, | |
37 | LXC_TYPE_CGFILE, | |
38 | LXC_TYPE_PROC_MEMINFO, | |
39 | LXC_TYPE_PROC_CPUINFO, | |
40 | LXC_TYPE_PROC_UPTIME, | |
41 | LXC_TYPE_PROC_STAT, | |
42 | LXC_TYPE_PROC_DISKSTATS, | |
70dcc12e | 43 | LXC_TYPE_PROC_SWAPS, |
237e200e SH |
44 | }; |
45 | ||
46 | struct file_info { | |
47 | char *controller; | |
48 | char *cgroup; | |
49 | char *file; | |
50 | int type; | |
51 | char *buf; // unused as of yet | |
52 | int buflen; | |
53 | int size; //actual data size | |
54 | int cached; | |
55 | }; | |
56 | ||
57 | /* reserve buffer size, for cpuall in /proc/stat */ | |
58 | #define BUF_RESERVE_SIZE 256 | |
59 | ||
60 | /* | |
61 | * A table caching which pid is init for a pid namespace. | |
62 | * When looking up which pid is init for $qpid, we first | |
63 | * 1. Stat /proc/$qpid/ns/pid. | |
64 | * 2. Check whether the ino_t is in our store. | |
65 | * a. if not, fork a child in qpid's ns to send us | |
66 | * ucred.pid = 1, and read the initpid. Cache | |
67 | * initpid and creation time for /proc/initpid | |
68 | * in a new store entry. | |
69 | * b. if so, verify that /proc/initpid still matches | |
70 | * what we have saved. If not, clear the store | |
71 | * entry and go back to a. If so, return the | |
72 | * cached initpid. | |
73 | */ | |
74 | struct pidns_init_store { | |
75 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
76 | pid_t initpid; // the pid of nit in that ns | |
77 | long int ctime; // the time at which /proc/$initpid was created | |
78 | struct pidns_init_store *next; | |
79 | long int lastcheck; | |
80 | }; | |
81 | ||
82 | /* lol - look at how they are allocated in the kernel */ | |
83 | #define PIDNS_HASH_SIZE 4096 | |
84 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
85 | ||
86 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
87 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
88 | static void lock_mutex(pthread_mutex_t *l) | |
89 | { | |
90 | int ret; | |
91 | ||
92 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
93 | fprintf(stderr, "pthread_mutex_lock returned:%d %s\n", ret, strerror(ret)); | |
94 | exit(1); | |
95 | } | |
96 | } | |
97 | ||
98 | static void unlock_mutex(pthread_mutex_t *l) | |
99 | { | |
100 | int ret; | |
101 | ||
102 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
103 | fprintf(stderr, "pthread_mutex_unlock returned:%d %s\n", ret, strerror(ret)); | |
104 | exit(1); | |
105 | } | |
106 | } | |
107 | ||
108 | static void store_lock(void) | |
109 | { | |
110 | lock_mutex(&pidns_store_mutex); | |
111 | } | |
112 | ||
113 | static void store_unlock(void) | |
114 | { | |
115 | unlock_mutex(&pidns_store_mutex); | |
116 | } | |
117 | ||
118 | /* Must be called under store_lock */ | |
119 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
120 | { | |
121 | struct stat initsb; | |
122 | char fnam[100]; | |
123 | ||
124 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
125 | if (stat(fnam, &initsb) < 0) | |
126 | return false; | |
127 | #if DEBUG | |
128 | fprintf(stderr, "comparing ctime %ld %ld for pid %d\n", | |
129 | e->ctime, initsb.st_ctime, e->initpid); | |
130 | #endif | |
131 | if (e->ctime != initsb.st_ctime) | |
132 | return false; | |
133 | return true; | |
134 | } | |
135 | ||
136 | /* Must be called under store_lock */ | |
137 | static void remove_initpid(struct pidns_init_store *e) | |
138 | { | |
139 | struct pidns_init_store *tmp; | |
140 | int h; | |
141 | ||
142 | #if DEBUG | |
143 | fprintf(stderr, "remove_initpid: removing entry for %d\n", e->initpid); | |
144 | #endif | |
145 | h = HASH(e->ino); | |
146 | if (pidns_hash_table[h] == e) { | |
147 | pidns_hash_table[h] = e->next; | |
148 | free(e); | |
149 | return; | |
150 | } | |
151 | ||
152 | tmp = pidns_hash_table[h]; | |
153 | while (tmp) { | |
154 | if (tmp->next == e) { | |
155 | tmp->next = e->next; | |
156 | free(e); | |
157 | return; | |
158 | } | |
159 | tmp = tmp->next; | |
160 | } | |
161 | } | |
162 | ||
163 | #define PURGE_SECS 5 | |
164 | /* Must be called under store_lock */ | |
165 | static void prune_initpid_store(void) | |
166 | { | |
167 | static long int last_prune = 0; | |
168 | struct pidns_init_store *e, *prev, *delme; | |
169 | long int now, threshold; | |
170 | int i; | |
171 | ||
172 | if (!last_prune) { | |
173 | last_prune = time(NULL); | |
174 | return; | |
175 | } | |
176 | now = time(NULL); | |
177 | if (now < last_prune + PURGE_SECS) | |
178 | return; | |
179 | #if DEBUG | |
180 | fprintf(stderr, "pruning\n"); | |
181 | #endif | |
182 | last_prune = now; | |
183 | threshold = now - 2 * PURGE_SECS; | |
184 | ||
185 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
186 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
187 | if (e->lastcheck < threshold) { | |
188 | #if DEBUG | |
189 | fprintf(stderr, "Removing cached entry for %d\n", e->initpid); | |
190 | #endif | |
191 | delme = e; | |
192 | if (prev) | |
193 | prev->next = e->next; | |
194 | else | |
195 | pidns_hash_table[i] = e->next; | |
196 | e = e->next; | |
197 | free(delme); | |
198 | } else { | |
199 | prev = e; | |
200 | e = e->next; | |
201 | } | |
202 | } | |
203 | } | |
204 | } | |
205 | ||
206 | /* Must be called under store_lock */ | |
207 | static void save_initpid(struct stat *sb, pid_t pid) | |
208 | { | |
209 | struct pidns_init_store *e; | |
210 | char fpath[100]; | |
211 | struct stat procsb; | |
212 | int h; | |
213 | ||
214 | #if DEBUG | |
215 | fprintf(stderr, "save_initpid: adding entry for %d\n", pid); | |
216 | #endif | |
217 | snprintf(fpath, 100, "/proc/%d", pid); | |
218 | if (stat(fpath, &procsb) < 0) | |
219 | return; | |
220 | do { | |
221 | e = malloc(sizeof(*e)); | |
222 | } while (!e); | |
223 | e->ino = sb->st_ino; | |
224 | e->initpid = pid; | |
225 | e->ctime = procsb.st_ctime; | |
226 | h = HASH(e->ino); | |
227 | e->next = pidns_hash_table[h]; | |
228 | e->lastcheck = time(NULL); | |
229 | pidns_hash_table[h] = e; | |
230 | } | |
231 | ||
232 | /* | |
233 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
234 | * entry for the inode number and creation time. Verify that the init pid | |
235 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
236 | * otherwise. | |
237 | * Must be called under store_lock | |
238 | */ | |
239 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
240 | { | |
241 | int h = HASH(sb->st_ino); | |
242 | struct pidns_init_store *e = pidns_hash_table[h]; | |
243 | ||
244 | while (e) { | |
245 | if (e->ino == sb->st_ino) { | |
246 | if (initpid_still_valid(e, sb)) { | |
247 | e->lastcheck = time(NULL); | |
248 | return e; | |
249 | } | |
250 | remove_initpid(e); | |
251 | return NULL; | |
252 | } | |
253 | e = e->next; | |
254 | } | |
255 | ||
256 | return NULL; | |
257 | } | |
258 | ||
259 | static int is_dir(const char *path) | |
260 | { | |
261 | struct stat statbuf; | |
262 | int ret = stat(path, &statbuf); | |
263 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) | |
264 | return 1; | |
265 | return 0; | |
266 | } | |
267 | ||
268 | static char *must_copy_string(const char *str) | |
269 | { | |
270 | char *dup = NULL; | |
271 | if (!str) | |
272 | return NULL; | |
273 | do { | |
274 | dup = strdup(str); | |
275 | } while (!dup); | |
276 | ||
277 | return dup; | |
278 | } | |
279 | ||
280 | static inline void drop_trailing_newlines(char *s) | |
281 | { | |
282 | int l; | |
283 | ||
284 | for (l=strlen(s); l>0 && s[l-1] == '\n'; l--) | |
285 | s[l-1] = '\0'; | |
286 | } | |
287 | ||
288 | #define BATCH_SIZE 50 | |
289 | static void dorealloc(char **mem, size_t oldlen, size_t newlen) | |
290 | { | |
291 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
292 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
293 | ||
294 | if (!*mem || newbatches > oldbatches) { | |
295 | char *tmp; | |
296 | do { | |
297 | tmp = realloc(*mem, newbatches * BATCH_SIZE); | |
298 | } while (!tmp); | |
299 | *mem = tmp; | |
300 | } | |
301 | } | |
302 | static void append_line(char **contents, size_t *len, char *line, ssize_t linelen) | |
303 | { | |
304 | size_t newlen = *len + linelen; | |
305 | dorealloc(contents, *len, newlen + 1); | |
306 | memcpy(*contents + *len, line, linelen+1); | |
307 | *len = newlen; | |
308 | } | |
309 | ||
310 | static char *slurp_file(const char *from) | |
311 | { | |
312 | char *line = NULL; | |
313 | char *contents = NULL; | |
314 | FILE *f = fopen(from, "r"); | |
315 | size_t len = 0, fulllen = 0; | |
316 | ssize_t linelen; | |
317 | ||
318 | if (!f) | |
319 | return NULL; | |
320 | ||
321 | while ((linelen = getline(&line, &len, f)) != -1) { | |
322 | append_line(&contents, &fulllen, line, linelen); | |
323 | } | |
324 | fclose(f); | |
325 | ||
326 | if (contents) | |
327 | drop_trailing_newlines(contents); | |
328 | free(line); | |
329 | return contents; | |
330 | } | |
331 | ||
332 | static bool write_string(const char *fnam, const char *string) | |
333 | { | |
334 | FILE *f; | |
335 | size_t len, ret; | |
336 | ||
337 | if (!(f = fopen(fnam, "w"))) | |
338 | return false; | |
339 | len = strlen(string); | |
340 | ret = fwrite(string, 1, len, f); | |
341 | if (ret != len) { | |
342 | fprintf(stderr, "Error writing to file: %s\n", strerror(errno)); | |
343 | fclose(f); | |
344 | return false; | |
345 | } | |
346 | if (fclose(f) < 0) { | |
347 | fprintf(stderr, "Error writing to file: %s\n", strerror(errno)); | |
348 | return false; | |
349 | } | |
350 | return true; | |
351 | } | |
352 | ||
353 | /* | |
354 | * hierarchies, i.e. 'cpu,cpuacct' | |
355 | */ | |
356 | char **hierarchies; | |
357 | int num_hierarchies; | |
358 | ||
359 | struct cgfs_files { | |
360 | char *name; | |
361 | uint32_t uid, gid; | |
362 | uint32_t mode; | |
363 | }; | |
364 | ||
0619767c | 365 | #define ALLOC_NUM 20 |
237e200e SH |
366 | static bool store_hierarchy(char *stridx, char *h) |
367 | { | |
0619767c SH |
368 | if (num_hierarchies % ALLOC_NUM == 0) { |
369 | size_t n = (num_hierarchies / ALLOC_NUM) + 1; | |
370 | n *= ALLOC_NUM; | |
371 | char **tmp = realloc(hierarchies, n * sizeof(char *)); | |
0619767c SH |
372 | if (!tmp) { |
373 | fprintf(stderr, "Out of memory\n"); | |
374 | exit(1); | |
375 | } | |
237e200e | 376 | hierarchies = tmp; |
237e200e SH |
377 | } |
378 | ||
0619767c | 379 | hierarchies[num_hierarchies++] = must_copy_string(h); |
237e200e SH |
380 | return true; |
381 | } | |
382 | ||
383 | static void print_subsystems(void) | |
384 | { | |
385 | int i; | |
386 | ||
387 | fprintf(stderr, "hierarchies:"); | |
388 | for (i = 0; i < num_hierarchies; i++) { | |
389 | if (hierarchies[i]) | |
390 | fprintf(stderr, " %d: %s\n", i, hierarchies[i]); | |
391 | } | |
392 | } | |
393 | ||
394 | static bool in_comma_list(const char *needle, const char *haystack) | |
395 | { | |
396 | const char *s = haystack, *e; | |
397 | size_t nlen = strlen(needle); | |
398 | ||
399 | while (*s && (e = index(s, ','))) { | |
400 | if (nlen != e - s) { | |
401 | s = e + 1; | |
402 | continue; | |
403 | } | |
404 | if (strncmp(needle, s, nlen) == 0) | |
405 | return true; | |
406 | s = e + 1; | |
407 | } | |
408 | if (strcmp(needle, s) == 0) | |
409 | return true; | |
410 | return false; | |
411 | } | |
412 | ||
413 | /* do we need to do any massaging here? I'm not sure... */ | |
414 | static char *find_mounted_controller(const char *controller) | |
415 | { | |
416 | int i; | |
417 | ||
418 | for (i = 0; i < num_hierarchies; i++) { | |
419 | if (!hierarchies[i]) | |
420 | continue; | |
421 | if (strcmp(hierarchies[i], controller) == 0) | |
422 | return hierarchies[i]; | |
423 | if (in_comma_list(controller, hierarchies[i])) | |
424 | return hierarchies[i]; | |
425 | } | |
426 | ||
427 | return NULL; | |
428 | } | |
429 | ||
430 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, | |
431 | const char *value) | |
432 | { | |
433 | size_t len; | |
434 | char *fnam, *tmpc = find_mounted_controller(controller); | |
435 | ||
436 | if (!tmpc) | |
437 | return false; | |
438 | /* basedir / tmpc / cgroup / file \0 */ | |
439 | len = strlen(basedir) + strlen(tmpc) + strlen(cgroup) + strlen(file) + 4; | |
440 | fnam = alloca(len); | |
441 | snprintf(fnam, len, "%s/%s/%s/%s", basedir, tmpc, cgroup, file); | |
442 | ||
443 | return write_string(fnam, value); | |
444 | } | |
445 | ||
446 | // Chown all the files in the cgroup directory. We do this when we create | |
447 | // a cgroup on behalf of a user. | |
448 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid) | |
449 | { | |
450 | struct dirent dirent, *direntp; | |
451 | char path[MAXPATHLEN]; | |
452 | size_t len; | |
453 | DIR *d; | |
454 | int ret; | |
455 | ||
456 | len = strlen(dirname); | |
457 | if (len >= MAXPATHLEN) { | |
458 | fprintf(stderr, "chown_all_cgroup_files: pathname too long: %s\n", dirname); | |
459 | return; | |
460 | } | |
461 | ||
462 | d = opendir(dirname); | |
463 | if (!d) { | |
464 | fprintf(stderr, "chown_all_cgroup_files: failed to open %s\n", dirname); | |
465 | return; | |
466 | } | |
467 | ||
468 | while (readdir_r(d, &dirent, &direntp) == 0 && direntp) { | |
469 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) | |
470 | continue; | |
471 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
472 | if (ret < 0 || ret >= MAXPATHLEN) { | |
473 | fprintf(stderr, "chown_all_cgroup_files: pathname too long under %s\n", dirname); | |
474 | continue; | |
475 | } | |
476 | if (chown(path, uid, gid) < 0) | |
477 | fprintf(stderr, "Failed to chown file %s to %u:%u", path, uid, gid); | |
478 | } | |
479 | closedir(d); | |
480 | } | |
481 | ||
482 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
483 | { | |
484 | size_t len; | |
485 | char *dirnam, *tmpc = find_mounted_controller(controller); | |
486 | ||
487 | if (!tmpc) | |
488 | return -EINVAL; | |
489 | /* basedir / tmpc / cg \0 */ | |
490 | len = strlen(basedir) + strlen(tmpc) + strlen(cg) + 3; | |
491 | dirnam = alloca(len); | |
492 | snprintf(dirnam, len, "%s/%s/%s", basedir,tmpc, cg); | |
493 | ||
494 | if (mkdir(dirnam, 0755) < 0) | |
495 | return -errno; | |
496 | ||
497 | if (uid == 0 && gid == 0) | |
498 | return 0; | |
499 | ||
500 | if (chown(dirnam, uid, gid) < 0) | |
501 | return -errno; | |
502 | ||
503 | chown_all_cgroup_files(dirnam, uid, gid); | |
504 | ||
505 | return 0; | |
506 | } | |
507 | ||
508 | static bool recursive_rmdir(const char *dirname) | |
509 | { | |
510 | struct dirent dirent, *direntp; | |
511 | DIR *dir; | |
512 | bool ret = false; | |
513 | char pathname[MAXPATHLEN]; | |
514 | ||
515 | dir = opendir(dirname); | |
516 | if (!dir) { | |
517 | #if DEBUG | |
518 | fprintf(stderr, "%s: failed to open %s: %s\n", __func__, dirname, strerror(errno)); | |
519 | #endif | |
520 | return false; | |
521 | } | |
522 | ||
523 | while (!readdir_r(dir, &dirent, &direntp)) { | |
524 | struct stat mystat; | |
525 | int rc; | |
526 | ||
527 | if (!direntp) | |
528 | break; | |
529 | ||
530 | if (!strcmp(direntp->d_name, ".") || | |
531 | !strcmp(direntp->d_name, "..")) | |
532 | continue; | |
533 | ||
534 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
535 | if (rc < 0 || rc >= MAXPATHLEN) { | |
536 | fprintf(stderr, "pathname too long\n"); | |
537 | continue; | |
538 | } | |
539 | ||
540 | ret = lstat(pathname, &mystat); | |
541 | if (ret) { | |
542 | #if DEBUG | |
543 | fprintf(stderr, "%s: failed to stat %s: %s\n", __func__, pathname, strerror(errno)); | |
544 | #endif | |
545 | continue; | |
546 | } | |
547 | if (S_ISDIR(mystat.st_mode)) { | |
548 | if (!recursive_rmdir(pathname)) { | |
549 | #if DEBUG | |
550 | fprintf(stderr, "Error removing %s\n", pathname); | |
551 | #endif | |
552 | } | |
553 | } | |
554 | } | |
555 | ||
556 | ret = true; | |
557 | if (closedir(dir) < 0) { | |
558 | fprintf(stderr, "%s: failed to close directory %s: %s\n", __func__, dirname, strerror(errno)); | |
559 | ret = false; | |
560 | } | |
561 | ||
562 | if (rmdir(dirname) < 0) { | |
563 | #if DEBUG | |
564 | fprintf(stderr, "%s: failed to delete %s: %s\n", __func__, dirname, strerror(errno)); | |
565 | #endif | |
566 | ret = false; | |
567 | } | |
568 | ||
569 | return ret; | |
570 | } | |
571 | ||
572 | bool cgfs_remove(const char *controller, const char *cg) | |
573 | { | |
574 | size_t len; | |
575 | char *dirnam, *tmpc = find_mounted_controller(controller); | |
576 | ||
577 | if (!tmpc) | |
578 | return false; | |
579 | /* basedir / tmpc / cg \0 */ | |
580 | len = strlen(basedir) + strlen(tmpc) + strlen(cg) + 3; | |
581 | dirnam = alloca(len); | |
582 | snprintf(dirnam, len, "%s/%s/%s", basedir,tmpc, cg); | |
583 | return recursive_rmdir(dirnam); | |
584 | } | |
585 | ||
586 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
587 | { | |
588 | size_t len; | |
589 | char *pathname, *tmpc = find_mounted_controller(controller); | |
590 | ||
591 | if (!tmpc) | |
592 | return false; | |
593 | /* basedir / tmpc / file \0 */ | |
594 | len = strlen(basedir) + strlen(tmpc) + strlen(file) + 3; | |
595 | pathname = alloca(len); | |
596 | snprintf(pathname, len, "%s/%s/%s", basedir, tmpc, file); | |
597 | if (chmod(pathname, mode) < 0) | |
598 | return false; | |
599 | return true; | |
600 | } | |
601 | ||
602 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid) | |
603 | { | |
604 | size_t len; | |
605 | char *fname; | |
606 | ||
607 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
608 | fname = alloca(len); | |
609 | snprintf(fname, len, "%s/tasks", dirname); | |
610 | if (chown(fname, uid, gid) != 0) | |
611 | return -errno; | |
612 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
613 | if (chown(fname, uid, gid) != 0) | |
614 | return -errno; | |
615 | return 0; | |
616 | } | |
617 | ||
618 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
619 | { | |
620 | size_t len; | |
621 | char *pathname, *tmpc = find_mounted_controller(controller); | |
622 | ||
623 | if (!tmpc) | |
624 | return -EINVAL; | |
625 | /* basedir / tmpc / file \0 */ | |
626 | len = strlen(basedir) + strlen(tmpc) + strlen(file) + 3; | |
627 | pathname = alloca(len); | |
628 | snprintf(pathname, len, "%s/%s/%s", basedir, tmpc, file); | |
629 | if (chown(pathname, uid, gid) < 0) | |
630 | return -errno; | |
631 | ||
632 | if (is_dir(pathname)) | |
633 | // like cgmanager did, we want to chown the tasks file as well | |
634 | return chown_tasks_files(pathname, uid, gid); | |
635 | ||
636 | return 0; | |
637 | } | |
638 | ||
639 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
640 | { | |
641 | size_t len; | |
642 | char *pathname, *tmpc = find_mounted_controller(controller); | |
643 | ||
644 | if (!tmpc) | |
645 | return NULL; | |
646 | /* basedir / tmpc / cgroup / "cgroup.procs" \0 */ | |
647 | len = strlen(basedir) + strlen(tmpc) + strlen(cgroup) + 4 + strlen("cgroup.procs"); | |
648 | pathname = alloca(len); | |
649 | snprintf(pathname, len, "%s/%s/%s/cgroup.procs", basedir, tmpc, cgroup); | |
650 | return fopen(pathname, "w"); | |
651 | } | |
652 | ||
f366da65 WB |
653 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, |
654 | void ***list, size_t typesize, | |
655 | void* (*iterator)(const char*, const char*, const char*)) | |
237e200e SH |
656 | { |
657 | size_t len; | |
658 | char *dirname, *tmpc = find_mounted_controller(controller); | |
659 | char pathname[MAXPATHLEN]; | |
f366da65 | 660 | size_t sz = 0, asz = 0; |
237e200e SH |
661 | struct dirent dirent, *direntp; |
662 | DIR *dir; | |
663 | int ret; | |
664 | ||
f366da65 | 665 | *list = NULL; |
237e200e | 666 | if (!tmpc) |
e97c834b | 667 | return false; |
237e200e SH |
668 | |
669 | /* basedir / tmpc / cgroup \0 */ | |
670 | len = strlen(basedir) + strlen(tmpc) + strlen(cgroup) + 3; | |
671 | dirname = alloca(len); | |
672 | snprintf(dirname, len, "%s/%s/%s", basedir, tmpc, cgroup); | |
673 | ||
674 | dir = opendir(dirname); | |
675 | if (!dir) | |
676 | return false; | |
677 | ||
678 | while (!readdir_r(dir, &dirent, &direntp)) { | |
679 | struct stat mystat; | |
680 | int rc; | |
681 | ||
682 | if (!direntp) | |
683 | break; | |
684 | ||
685 | if (!strcmp(direntp->d_name, ".") || | |
686 | !strcmp(direntp->d_name, "..")) | |
687 | continue; | |
688 | ||
689 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
690 | if (rc < 0 || rc >= MAXPATHLEN) { | |
691 | fprintf(stderr, "%s: pathname too long under %s\n", __func__, dirname); | |
692 | continue; | |
693 | } | |
694 | ||
695 | ret = lstat(pathname, &mystat); | |
696 | if (ret) { | |
697 | fprintf(stderr, "%s: failed to stat %s: %s\n", __func__, pathname, strerror(errno)); | |
698 | continue; | |
699 | } | |
f366da65 WB |
700 | if ((!directories && !S_ISREG(mystat.st_mode)) || |
701 | (directories && !S_ISDIR(mystat.st_mode))) | |
237e200e SH |
702 | continue; |
703 | ||
704 | if (sz+2 >= asz) { | |
f366da65 | 705 | void **tmp; |
237e200e SH |
706 | asz += BATCH_SIZE; |
707 | do { | |
f366da65 | 708 | tmp = realloc(*list, asz * typesize); |
237e200e SH |
709 | } while (!tmp); |
710 | *list = tmp; | |
711 | } | |
f366da65 | 712 | (*list)[sz] = (*iterator)(controller, cgroup, direntp->d_name); |
237e200e SH |
713 | (*list)[sz+1] = NULL; |
714 | sz++; | |
715 | } | |
716 | if (closedir(dir) < 0) { | |
717 | fprintf(stderr, "%s: failed closedir for %s: %s\n", __func__, dirname, strerror(errno)); | |
718 | return false; | |
719 | } | |
720 | return true; | |
721 | } | |
722 | ||
f366da65 WB |
723 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
724 | { | |
725 | char *dup; | |
726 | do { | |
727 | dup = strdup(dir_entry); | |
728 | } while (!dup); | |
729 | return dup; | |
730 | } | |
731 | ||
732 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
733 | { | |
734 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
735 | } | |
736 | ||
237e200e SH |
737 | void free_key(struct cgfs_files *k) |
738 | { | |
739 | if (!k) | |
740 | return; | |
741 | free(k->name); | |
742 | free(k); | |
743 | } | |
744 | ||
745 | void free_keys(struct cgfs_files **keys) | |
746 | { | |
747 | int i; | |
748 | ||
749 | if (!keys) | |
750 | return; | |
751 | for (i = 0; keys[i]; i++) { | |
752 | free_key(keys[i]); | |
753 | } | |
754 | free(keys); | |
755 | } | |
756 | ||
757 | bool cgfs_get_value(const char *controller, const char *cgroup, const char *file, char **value) | |
758 | { | |
759 | size_t len; | |
760 | char *fnam, *tmpc = find_mounted_controller(controller); | |
761 | ||
762 | if (!tmpc) | |
763 | return false; | |
764 | /* basedir / tmpc / cgroup / file \0 */ | |
765 | len = strlen(basedir) + strlen(tmpc) + strlen(cgroup) + strlen(file) + 4; | |
766 | fnam = alloca(len); | |
767 | snprintf(fnam, len, "%s/%s/%s/%s", basedir, tmpc, cgroup, file); | |
768 | ||
769 | *value = slurp_file(fnam); | |
770 | return *value != NULL; | |
771 | } | |
772 | ||
773 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) | |
774 | { | |
775 | size_t len; | |
776 | char *fnam, *tmpc = find_mounted_controller(controller); | |
777 | struct stat sb; | |
778 | struct cgfs_files *newkey; | |
779 | int ret; | |
780 | ||
781 | if (!tmpc) | |
782 | return false; | |
783 | ||
784 | if (file && *file == '/') | |
785 | file++; | |
786 | ||
787 | if (file && index(file, '/')) | |
788 | return NULL; | |
789 | ||
790 | /* basedir / tmpc / cgroup / file \0 */ | |
791 | len = strlen(basedir) + strlen(tmpc) + strlen(cgroup) + 3; | |
792 | if (file) | |
793 | len += strlen(file) + 1; | |
794 | fnam = alloca(len); | |
795 | snprintf(fnam, len, "%s/%s/%s%s%s", basedir, tmpc, cgroup, | |
796 | file ? "/" : "", file ? file : ""); | |
797 | ||
798 | ret = stat(fnam, &sb); | |
799 | if (ret < 0) | |
800 | return NULL; | |
801 | ||
802 | do { | |
803 | newkey = malloc(sizeof(struct cgfs_files)); | |
804 | } while (!newkey); | |
805 | if (file) | |
806 | newkey->name = must_copy_string(file); | |
807 | else if (rindex(cgroup, '/')) | |
808 | newkey->name = must_copy_string(rindex(cgroup, '/')); | |
809 | else | |
810 | newkey->name = must_copy_string(cgroup); | |
811 | newkey->uid = sb.st_uid; | |
812 | newkey->gid = sb.st_gid; | |
813 | newkey->mode = sb.st_mode; | |
814 | ||
815 | return newkey; | |
816 | } | |
817 | ||
f366da65 | 818 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
237e200e | 819 | { |
f366da65 WB |
820 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); |
821 | if (!entry) { | |
822 | fprintf(stderr, "%s: Error getting files under %s:%s\n", | |
823 | __func__, controller, cgroup); | |
237e200e | 824 | } |
f366da65 WB |
825 | return entry; |
826 | } | |
827 | ||
828 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
829 | { | |
830 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
237e200e SH |
831 | } |
832 | ||
833 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
834 | { size_t len; | |
835 | char *fnam, *tmpc = find_mounted_controller(controller); | |
836 | int ret; | |
837 | struct stat sb; | |
838 | ||
839 | if (!tmpc) | |
840 | return false; | |
841 | /* basedir / tmpc / cgroup / f \0 */ | |
842 | len = strlen(basedir) + strlen(tmpc) + strlen(cgroup) + strlen(f) + 4; | |
843 | fnam = alloca(len); | |
844 | snprintf(fnam, len, "%s/%s/%s/%s", basedir, tmpc, cgroup, f); | |
845 | ||
846 | ret = stat(fnam, &sb); | |
847 | if (ret < 0 || !S_ISDIR(sb.st_mode)) | |
848 | return false; | |
849 | return true; | |
850 | } | |
851 | ||
852 | #define SEND_CREDS_OK 0 | |
853 | #define SEND_CREDS_NOTSK 1 | |
854 | #define SEND_CREDS_FAIL 2 | |
855 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
856 | static int wait_for_pid(pid_t pid); | |
857 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
858 | ||
859 | /* | |
860 | * fork a task which switches to @task's namespace and writes '1'. | |
861 | * over a unix sock so we can read the task's reaper's pid in our | |
862 | * namespace | |
863 | */ | |
864 | static void write_task_init_pid_exit(int sock, pid_t target) | |
865 | { | |
866 | struct ucred cred; | |
867 | char fnam[100]; | |
868 | pid_t pid; | |
869 | char v; | |
870 | int fd, ret; | |
871 | ||
872 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
873 | if (ret < 0 || ret >= sizeof(fnam)) | |
874 | _exit(1); | |
875 | ||
876 | fd = open(fnam, O_RDONLY); | |
877 | if (fd < 0) { | |
878 | perror("write_task_init_pid_exit open of ns/pid"); | |
879 | _exit(1); | |
880 | } | |
881 | if (setns(fd, 0)) { | |
882 | perror("write_task_init_pid_exit setns 1"); | |
883 | close(fd); | |
884 | _exit(1); | |
885 | } | |
886 | pid = fork(); | |
887 | if (pid < 0) | |
888 | _exit(1); | |
889 | if (pid != 0) { | |
890 | if (!wait_for_pid(pid)) | |
891 | _exit(1); | |
892 | _exit(0); | |
893 | } | |
894 | ||
895 | /* we are the child */ | |
896 | cred.uid = 0; | |
897 | cred.gid = 0; | |
898 | cred.pid = 1; | |
899 | v = '1'; | |
900 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
901 | _exit(1); | |
902 | _exit(0); | |
903 | } | |
904 | ||
905 | static pid_t get_init_pid_for_task(pid_t task) | |
906 | { | |
907 | int sock[2]; | |
908 | pid_t pid; | |
909 | pid_t ret = -1; | |
910 | char v = '0'; | |
911 | struct ucred cred; | |
912 | ||
913 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
914 | perror("socketpair"); | |
915 | return -1; | |
916 | } | |
917 | ||
918 | pid = fork(); | |
919 | if (pid < 0) | |
920 | goto out; | |
921 | if (!pid) { | |
922 | close(sock[1]); | |
923 | write_task_init_pid_exit(sock[0], task); | |
924 | _exit(0); | |
925 | } | |
926 | ||
927 | if (!recv_creds(sock[1], &cred, &v)) | |
928 | goto out; | |
929 | ret = cred.pid; | |
930 | ||
931 | out: | |
932 | close(sock[0]); | |
933 | close(sock[1]); | |
934 | if (pid > 0) | |
935 | wait_for_pid(pid); | |
936 | return ret; | |
937 | } | |
938 | ||
939 | static pid_t lookup_initpid_in_store(pid_t qpid) | |
940 | { | |
941 | pid_t answer = 0; | |
942 | struct stat sb; | |
943 | struct pidns_init_store *e; | |
944 | char fnam[100]; | |
945 | ||
946 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
947 | store_lock(); | |
948 | if (stat(fnam, &sb) < 0) | |
949 | goto out; | |
950 | e = lookup_verify_initpid(&sb); | |
951 | if (e) { | |
952 | answer = e->initpid; | |
953 | goto out; | |
954 | } | |
955 | answer = get_init_pid_for_task(qpid); | |
956 | if (answer > 0) | |
957 | save_initpid(&sb, answer); | |
958 | ||
959 | out: | |
960 | /* we prune at end in case we are returning | |
961 | * the value we were about to return */ | |
962 | prune_initpid_store(); | |
963 | store_unlock(); | |
964 | return answer; | |
965 | } | |
966 | ||
967 | static int wait_for_pid(pid_t pid) | |
968 | { | |
969 | int status, ret; | |
970 | ||
971 | if (pid <= 0) | |
972 | return -1; | |
973 | ||
974 | again: | |
975 | ret = waitpid(pid, &status, 0); | |
976 | if (ret == -1) { | |
977 | if (errno == EINTR) | |
978 | goto again; | |
979 | return -1; | |
980 | } | |
981 | if (ret != pid) | |
982 | goto again; | |
983 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
984 | return -1; | |
985 | return 0; | |
986 | } | |
987 | ||
988 | ||
989 | /* | |
990 | * append pid to *src. | |
991 | * src: a pointer to a char* in which ot append the pid. | |
992 | * sz: the number of characters printed so far, minus trailing \0. | |
993 | * asz: the allocated size so far | |
994 | * pid: the pid to append | |
995 | */ | |
996 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
997 | { | |
998 | char tmp[30]; | |
999 | ||
1000 | int tmplen = sprintf(tmp, "%d\n", (int)pid); | |
1001 | ||
1002 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1003 | char *tmp; | |
1004 | do { | |
1005 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1006 | } while (!tmp); | |
1007 | *src = tmp; | |
1008 | *asz += BUF_RESERVE_SIZE; | |
1009 | } | |
bbfd0e33 | 1010 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ |
237e200e | 1011 | *sz += tmplen; |
237e200e SH |
1012 | } |
1013 | ||
1014 | /* | |
1015 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1016 | * valid in the caller's namespace, return the id mapped into | |
1017 | * pid's namespace. | |
1018 | * Returns the mapped id, or -1 on error. | |
1019 | */ | |
1020 | unsigned int | |
1021 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1022 | { | |
1023 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1024 | hostuid, // base id for a range in the caller's namespace | |
1025 | count; // number of ids in this range | |
1026 | char line[400]; | |
1027 | int ret; | |
1028 | ||
1029 | fseek(idfile, 0L, SEEK_SET); | |
1030 | while (fgets(line, 400, idfile)) { | |
1031 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1032 | if (ret != 3) | |
1033 | continue; | |
1034 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1035 | /* | |
1036 | * uids wrapped around - unexpected as this is a procfile, | |
1037 | * so just bail. | |
1038 | */ | |
1039 | fprintf(stderr, "pid wrapparound at entry %u %u %u in %s\n", | |
1040 | nsuid, hostuid, count, line); | |
1041 | return -1; | |
1042 | } | |
1043 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1044 | /* | |
1045 | * now since hostuid <= in_id < hostuid+count, and | |
1046 | * hostuid+count and nsuid+count do not wrap around, | |
1047 | * we know that nsuid+(in_id-hostuid) which must be | |
1048 | * less that nsuid+(count) must not wrap around | |
1049 | */ | |
1050 | return (in_id - hostuid) + nsuid; | |
1051 | } | |
1052 | } | |
1053 | ||
1054 | // no answer found | |
1055 | return -1; | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * for is_privileged_over, | |
1060 | * specify whether we require the calling uid to be root in his | |
1061 | * namespace | |
1062 | */ | |
1063 | #define NS_ROOT_REQD true | |
1064 | #define NS_ROOT_OPT false | |
1065 | ||
1066 | #define PROCLEN 100 | |
1067 | ||
1068 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1069 | { | |
1070 | char fpath[PROCLEN]; | |
1071 | int ret; | |
1072 | bool answer = false; | |
1073 | uid_t nsuid; | |
1074 | ||
1075 | if (victim == -1 || uid == -1) | |
1076 | return false; | |
1077 | ||
1078 | /* | |
1079 | * If the request is one not requiring root in the namespace, | |
1080 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1081 | * access to files owned by uid 1000 | |
1082 | */ | |
1083 | if (!req_ns_root && uid == victim) | |
1084 | return true; | |
1085 | ||
1086 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1087 | if (ret < 0 || ret >= PROCLEN) | |
1088 | return false; | |
1089 | FILE *f = fopen(fpath, "r"); | |
1090 | if (!f) | |
1091 | return false; | |
1092 | ||
1093 | /* if caller's not root in his namespace, reject */ | |
1094 | nsuid = convert_id_to_ns(f, uid); | |
1095 | if (nsuid) | |
1096 | goto out; | |
1097 | ||
1098 | /* | |
1099 | * If victim is not mapped into caller's ns, reject. | |
1100 | * XXX I'm not sure this check is needed given that fuse | |
1101 | * will be sending requests where the vfs has converted | |
1102 | */ | |
1103 | nsuid = convert_id_to_ns(f, victim); | |
1104 | if (nsuid == -1) | |
1105 | goto out; | |
1106 | ||
1107 | answer = true; | |
1108 | ||
1109 | out: | |
1110 | fclose(f); | |
1111 | return answer; | |
1112 | } | |
1113 | ||
1114 | static bool perms_include(int fmode, mode_t req_mode) | |
1115 | { | |
1116 | mode_t r; | |
1117 | ||
1118 | switch (req_mode & O_ACCMODE) { | |
1119 | case O_RDONLY: | |
1120 | r = S_IROTH; | |
1121 | break; | |
1122 | case O_WRONLY: | |
1123 | r = S_IWOTH; | |
1124 | break; | |
1125 | case O_RDWR: | |
1126 | r = S_IROTH | S_IWOTH; | |
1127 | break; | |
1128 | default: | |
1129 | return false; | |
1130 | } | |
1131 | return ((fmode & r) == r); | |
1132 | } | |
1133 | ||
1134 | ||
1135 | /* | |
1136 | * taskcg is a/b/c | |
1137 | * querycg is /a/b/c/d/e | |
1138 | * we return 'd' | |
1139 | */ | |
1140 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1141 | { | |
1142 | char *start, *end; | |
1143 | ||
1144 | if (strlen(taskcg) <= strlen(querycg)) { | |
1145 | fprintf(stderr, "%s: I was fed bad input\n", __func__); | |
1146 | return NULL; | |
1147 | } | |
1148 | ||
1149 | if (strcmp(querycg, "/") == 0) | |
1150 | start = strdup(taskcg + 1); | |
1151 | else | |
1152 | start = strdup(taskcg + strlen(querycg) + 1); | |
1153 | if (!start) | |
1154 | return NULL; | |
1155 | end = strchr(start, '/'); | |
1156 | if (end) | |
1157 | *end = '\0'; | |
1158 | return start; | |
1159 | } | |
1160 | ||
1161 | static void stripnewline(char *x) | |
1162 | { | |
1163 | size_t l = strlen(x); | |
1164 | if (l && x[l-1] == '\n') | |
1165 | x[l-1] = '\0'; | |
1166 | } | |
1167 | ||
1168 | static char *get_pid_cgroup(pid_t pid, const char *contrl) | |
1169 | { | |
1170 | char fnam[PROCLEN]; | |
1171 | FILE *f; | |
1172 | char *answer = NULL; | |
1173 | char *line = NULL; | |
1174 | size_t len = 0; | |
1175 | int ret; | |
1176 | const char *h = find_mounted_controller(contrl); | |
1177 | if (!h) | |
1178 | return NULL; | |
1179 | ||
1180 | ret = snprintf(fnam, PROCLEN, "/proc/%d/cgroup", pid); | |
1181 | if (ret < 0 || ret >= PROCLEN) | |
1182 | return NULL; | |
1183 | if (!(f = fopen(fnam, "r"))) | |
1184 | return NULL; | |
1185 | ||
1186 | while (getline(&line, &len, f) != -1) { | |
1187 | char *c1, *c2; | |
1188 | if (!line[0]) | |
1189 | continue; | |
1190 | c1 = strchr(line, ':'); | |
1191 | if (!c1) | |
1192 | goto out; | |
1193 | c1++; | |
1194 | c2 = strchr(c1, ':'); | |
1195 | if (!c2) | |
1196 | goto out; | |
1197 | *c2 = '\0'; | |
1198 | if (strcmp(c1, h) != 0) | |
1199 | continue; | |
1200 | c2++; | |
1201 | stripnewline(c2); | |
1202 | do { | |
1203 | answer = strdup(c2); | |
1204 | } while (!answer); | |
1205 | break; | |
1206 | } | |
1207 | ||
1208 | out: | |
1209 | fclose(f); | |
1210 | free(line); | |
1211 | return answer; | |
1212 | } | |
1213 | ||
1214 | /* | |
1215 | * check whether a fuse context may access a cgroup dir or file | |
1216 | * | |
1217 | * If file is not null, it is a cgroup file to check under cg. | |
1218 | * If file is null, then we are checking perms on cg itself. | |
1219 | * | |
1220 | * For files we can check the mode of the list_keys result. | |
1221 | * For cgroups, we must make assumptions based on the files under the | |
1222 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1223 | * yet. | |
1224 | */ | |
1225 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1226 | { | |
1227 | struct cgfs_files *k = NULL; | |
1228 | bool ret = false; | |
1229 | ||
1230 | k = cgfs_get_key(contrl, cg, file); | |
1231 | if (!k) | |
1232 | return false; | |
1233 | ||
1234 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1235 | if (perms_include(k->mode >> 6, mode)) { | |
1236 | ret = true; | |
1237 | goto out; | |
1238 | } | |
1239 | } | |
1240 | if (fc->gid == k->gid) { | |
1241 | if (perms_include(k->mode >> 3, mode)) { | |
1242 | ret = true; | |
1243 | goto out; | |
1244 | } | |
1245 | } | |
1246 | ret = perms_include(k->mode, mode); | |
1247 | ||
1248 | out: | |
1249 | free_key(k); | |
1250 | return ret; | |
1251 | } | |
1252 | ||
1253 | #define INITSCOPE "/init.scope" | |
1254 | static void prune_init_slice(char *cg) | |
1255 | { | |
1256 | char *point; | |
1257 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1258 | ||
1259 | if (cg_len < initscope_len) | |
1260 | return; | |
1261 | ||
1262 | point = cg + cg_len - initscope_len; | |
1263 | if (strcmp(point, INITSCOPE) == 0) { | |
1264 | if (point == cg) | |
1265 | *(point+1) = '\0'; | |
1266 | else | |
1267 | *point = '\0'; | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | /* | |
1272 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1273 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1274 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1275 | * to a string containing the next cgroup directory under cg, which must be | |
1276 | * freed by the caller. | |
1277 | */ | |
1278 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1279 | { | |
1280 | bool answer = false; | |
1281 | char *c2 = get_pid_cgroup(pid, contrl); | |
1282 | char *linecmp; | |
1283 | ||
1284 | if (!c2) | |
1285 | return false; | |
1286 | prune_init_slice(c2); | |
1287 | ||
1288 | /* | |
1289 | * callers pass in '/' for root cgroup, otherwise they pass | |
1290 | * in a cgroup without leading '/' | |
1291 | */ | |
1292 | linecmp = *cg == '/' ? c2 : c2+1; | |
1293 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { | |
1294 | if (nextcg) { | |
1295 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1296 | } | |
1297 | goto out; | |
1298 | } | |
1299 | answer = true; | |
1300 | ||
1301 | out: | |
1302 | free(c2); | |
1303 | return answer; | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1308 | */ | |
1309 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1310 | { | |
1311 | bool answer = false; | |
1312 | char *c2, *task_cg; | |
1313 | size_t target_len, task_len; | |
1314 | ||
1315 | if (strcmp(cg, "/") == 0) | |
1316 | return true; | |
1317 | ||
1318 | c2 = get_pid_cgroup(pid, contrl); | |
1319 | if (!c2) | |
1320 | return false; | |
1321 | prune_init_slice(c2); | |
1322 | ||
1323 | task_cg = c2 + 1; | |
1324 | target_len = strlen(cg); | |
1325 | task_len = strlen(task_cg); | |
1326 | if (task_len == 0) { | |
1327 | /* Task is in the root cg, it can see everything. This case is | |
1328 | * not handled by the strmcps below, since they test for the | |
1329 | * last /, but that is the first / that we've chopped off | |
1330 | * above. | |
1331 | */ | |
1332 | answer = true; | |
1333 | goto out; | |
1334 | } | |
1335 | if (strcmp(cg, task_cg) == 0) { | |
1336 | answer = true; | |
1337 | goto out; | |
1338 | } | |
1339 | if (target_len < task_len) { | |
1340 | /* looking up a parent dir */ | |
1341 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1342 | answer = true; | |
1343 | goto out; | |
1344 | } | |
1345 | if (target_len > task_len) { | |
1346 | /* looking up a child dir */ | |
1347 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1348 | answer = true; | |
1349 | goto out; | |
1350 | } | |
1351 | ||
1352 | out: | |
1353 | free(c2); | |
1354 | return answer; | |
1355 | } | |
1356 | ||
1357 | /* | |
1358 | * given /cgroup/freezer/a/b, return "freezer". | |
1359 | * the returned char* should NOT be freed. | |
1360 | */ | |
1361 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1362 | { | |
1363 | const char *p1; | |
1364 | char *contr, *slash; | |
1365 | ||
1366 | if (strlen(path) < 9) | |
1367 | return NULL; | |
1368 | if (*(path+7) != '/') | |
1369 | return NULL; | |
1370 | p1 = path+8; | |
1371 | contr = strdupa(p1); | |
1372 | if (!contr) | |
1373 | return NULL; | |
1374 | slash = strstr(contr, "/"); | |
1375 | if (slash) | |
1376 | *slash = '\0'; | |
1377 | ||
1378 | int i; | |
1379 | for (i = 0; i < num_hierarchies; i++) { | |
1380 | if (hierarchies[i] && strcmp(hierarchies[i], contr) == 0) | |
1381 | return hierarchies[i]; | |
1382 | } | |
1383 | return NULL; | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1388 | * Note that the returned value may include files (keynames) etc | |
1389 | */ | |
1390 | static const char *find_cgroup_in_path(const char *path) | |
1391 | { | |
1392 | const char *p1; | |
1393 | ||
1394 | if (strlen(path) < 9) | |
1395 | return NULL; | |
1396 | p1 = strstr(path+8, "/"); | |
1397 | if (!p1) | |
1398 | return NULL; | |
1399 | return p1+1; | |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * split the last path element from the path in @cg. | |
1404 | * @dir is newly allocated and should be freed, @last not | |
1405 | */ | |
1406 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1407 | { | |
1408 | char *p; | |
1409 | ||
1410 | do { | |
1411 | *dir = strdup(cg); | |
1412 | } while (!*dir); | |
1413 | *last = strrchr(cg, '/'); | |
1414 | if (!*last) { | |
1415 | *last = NULL; | |
1416 | return; | |
1417 | } | |
1418 | p = strrchr(*dir, '/'); | |
1419 | *p = '\0'; | |
1420 | } | |
1421 | ||
1422 | /* | |
1423 | * FUSE ops for /cgroup | |
1424 | */ | |
1425 | ||
1426 | int cg_getattr(const char *path, struct stat *sb) | |
1427 | { | |
1428 | struct timespec now; | |
1429 | struct fuse_context *fc = fuse_get_context(); | |
1430 | char * cgdir = NULL; | |
1431 | char *last = NULL, *path1, *path2; | |
1432 | struct cgfs_files *k = NULL; | |
1433 | const char *cgroup; | |
1434 | const char *controller = NULL; | |
1435 | int ret = -ENOENT; | |
1436 | ||
1437 | ||
1438 | if (!fc) | |
1439 | return -EIO; | |
1440 | ||
1441 | memset(sb, 0, sizeof(struct stat)); | |
1442 | ||
1443 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1444 | return -EINVAL; | |
1445 | ||
1446 | sb->st_uid = sb->st_gid = 0; | |
1447 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1448 | sb->st_size = 0; | |
1449 | ||
1450 | if (strcmp(path, "/cgroup") == 0) { | |
1451 | sb->st_mode = S_IFDIR | 00755; | |
1452 | sb->st_nlink = 2; | |
1453 | return 0; | |
1454 | } | |
1455 | ||
1456 | controller = pick_controller_from_path(fc, path); | |
1457 | if (!controller) | |
1458 | return -EIO; | |
1459 | cgroup = find_cgroup_in_path(path); | |
1460 | if (!cgroup) { | |
1461 | /* this is just /cgroup/controller, return it as a dir */ | |
1462 | sb->st_mode = S_IFDIR | 00755; | |
1463 | sb->st_nlink = 2; | |
1464 | return 0; | |
1465 | } | |
1466 | ||
1467 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1468 | ||
1469 | if (!last) { | |
1470 | path1 = "/"; | |
1471 | path2 = cgdir; | |
1472 | } else { | |
1473 | path1 = cgdir; | |
1474 | path2 = last; | |
1475 | } | |
1476 | ||
1477 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1478 | if (initpid <= 0) | |
1479 | initpid = fc->pid; | |
1480 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
1481 | * Then check that caller's cgroup is under path if last is a child | |
1482 | * cgroup, or cgdir if last is a file */ | |
1483 | ||
1484 | if (is_child_cgroup(controller, path1, path2)) { | |
1485 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
1486 | ret = -ENOENT; | |
1487 | goto out; | |
1488 | } | |
1489 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
1490 | /* this is just /cgroup/controller, return it as a dir */ | |
1491 | sb->st_mode = S_IFDIR | 00555; | |
1492 | sb->st_nlink = 2; | |
1493 | ret = 0; | |
1494 | goto out; | |
1495 | } | |
1496 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
1497 | ret = -EACCES; | |
1498 | goto out; | |
1499 | } | |
1500 | ||
1501 | // get uid, gid, from '/tasks' file and make up a mode | |
1502 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1503 | sb->st_mode = S_IFDIR | 00755; | |
1504 | k = cgfs_get_key(controller, cgroup, NULL); | |
1505 | if (!k) { | |
1506 | sb->st_uid = sb->st_gid = 0; | |
1507 | } else { | |
1508 | sb->st_uid = k->uid; | |
1509 | sb->st_gid = k->gid; | |
1510 | } | |
1511 | free_key(k); | |
1512 | sb->st_nlink = 2; | |
1513 | ret = 0; | |
1514 | goto out; | |
1515 | } | |
1516 | ||
1517 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
1518 | sb->st_mode = S_IFREG | k->mode; | |
1519 | sb->st_nlink = 1; | |
1520 | sb->st_uid = k->uid; | |
1521 | sb->st_gid = k->gid; | |
1522 | sb->st_size = 0; | |
1523 | free_key(k); | |
1524 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
1525 | ret = -ENOENT; | |
1526 | goto out; | |
1527 | } | |
1528 | if (!fc_may_access(fc, controller, path1, path2, O_RDONLY)) { | |
1529 | ret = -EACCES; | |
1530 | goto out; | |
1531 | } | |
1532 | ||
1533 | ret = 0; | |
1534 | } | |
1535 | ||
1536 | out: | |
1537 | free(cgdir); | |
1538 | return ret; | |
1539 | } | |
1540 | ||
1541 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
1542 | { | |
1543 | struct fuse_context *fc = fuse_get_context(); | |
1544 | const char *cgroup; | |
1545 | struct file_info *dir_info; | |
1546 | char *controller = NULL; | |
1547 | ||
1548 | if (!fc) | |
1549 | return -EIO; | |
1550 | ||
1551 | if (strcmp(path, "/cgroup") == 0) { | |
1552 | cgroup = NULL; | |
1553 | controller = NULL; | |
1554 | } else { | |
1555 | // return list of keys for the controller, and list of child cgroups | |
1556 | controller = pick_controller_from_path(fc, path); | |
1557 | if (!controller) | |
1558 | return -EIO; | |
1559 | ||
1560 | cgroup = find_cgroup_in_path(path); | |
1561 | if (!cgroup) { | |
1562 | /* this is just /cgroup/controller, return its contents */ | |
1563 | cgroup = "/"; | |
1564 | } | |
1565 | } | |
1566 | ||
1567 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1568 | if (initpid <= 0) | |
1569 | initpid = fc->pid; | |
1570 | if (cgroup) { | |
1571 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
1572 | return -ENOENT; | |
1573 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
1574 | return -EACCES; | |
1575 | } | |
1576 | ||
1577 | /* we'll free this at cg_releasedir */ | |
1578 | dir_info = malloc(sizeof(*dir_info)); | |
1579 | if (!dir_info) | |
1580 | return -ENOMEM; | |
1581 | dir_info->controller = must_copy_string(controller); | |
1582 | dir_info->cgroup = must_copy_string(cgroup); | |
1583 | dir_info->type = LXC_TYPE_CGDIR; | |
1584 | dir_info->buf = NULL; | |
1585 | dir_info->file = NULL; | |
1586 | dir_info->buflen = 0; | |
1587 | ||
1588 | fi->fh = (unsigned long)dir_info; | |
1589 | return 0; | |
1590 | } | |
1591 | ||
1592 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1593 | struct fuse_file_info *fi) | |
1594 | { | |
1595 | struct file_info *d = (struct file_info *)fi->fh; | |
1596 | struct cgfs_files **list = NULL; | |
1597 | int i, ret; | |
1598 | char *nextcg = NULL; | |
1599 | struct fuse_context *fc = fuse_get_context(); | |
1600 | char **clist = NULL; | |
1601 | ||
1602 | if (d->type != LXC_TYPE_CGDIR) { | |
1603 | fprintf(stderr, "Internal error: file cache info used in readdir\n"); | |
1604 | return -EIO; | |
1605 | } | |
1606 | if (!d->cgroup && !d->controller) { | |
1607 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
1608 | int i; | |
1609 | ||
1610 | for (i = 0; i < num_hierarchies; i++) { | |
1611 | if (hierarchies[i] && filler(buf, hierarchies[i], NULL, 0) != 0) { | |
1612 | return -EIO; | |
1613 | } | |
1614 | } | |
1615 | return 0; | |
1616 | } | |
1617 | ||
1618 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
1619 | // not a valid cgroup | |
1620 | ret = -EINVAL; | |
1621 | goto out; | |
1622 | } | |
1623 | ||
1624 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1625 | if (initpid <= 0) | |
1626 | initpid = fc->pid; | |
1627 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
1628 | if (nextcg) { | |
1629 | ret = filler(buf, nextcg, NULL, 0); | |
1630 | free(nextcg); | |
1631 | if (ret != 0) { | |
1632 | ret = -EIO; | |
1633 | goto out; | |
1634 | } | |
1635 | } | |
1636 | ret = 0; | |
1637 | goto out; | |
1638 | } | |
1639 | ||
1640 | for (i = 0; list[i]; i++) { | |
1641 | if (filler(buf, list[i]->name, NULL, 0) != 0) { | |
1642 | ret = -EIO; | |
1643 | goto out; | |
1644 | } | |
1645 | } | |
1646 | ||
1647 | // now get the list of child cgroups | |
1648 | ||
1649 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
1650 | ret = 0; | |
1651 | goto out; | |
1652 | } | |
f366da65 WB |
1653 | if (clist) { |
1654 | for (i = 0; clist[i]; i++) { | |
1655 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
1656 | ret = -EIO; | |
1657 | goto out; | |
1658 | } | |
237e200e SH |
1659 | } |
1660 | } | |
1661 | ret = 0; | |
1662 | ||
1663 | out: | |
1664 | free_keys(list); | |
1665 | if (clist) { | |
1666 | for (i = 0; clist[i]; i++) | |
1667 | free(clist[i]); | |
1668 | free(clist); | |
1669 | } | |
1670 | return ret; | |
1671 | } | |
1672 | ||
1673 | static void do_release_file_info(struct file_info *f) | |
1674 | { | |
1675 | if (!f) | |
1676 | return; | |
1677 | free(f->controller); | |
1678 | free(f->cgroup); | |
1679 | free(f->file); | |
1680 | free(f->buf); | |
1681 | free(f); | |
1682 | } | |
1683 | ||
1684 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
1685 | { | |
1686 | struct file_info *d = (struct file_info *)fi->fh; | |
1687 | ||
1688 | do_release_file_info(d); | |
1689 | return 0; | |
1690 | } | |
1691 | ||
1692 | int cg_open(const char *path, struct fuse_file_info *fi) | |
1693 | { | |
1694 | const char *cgroup; | |
1695 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
1696 | struct cgfs_files *k = NULL; | |
1697 | struct file_info *file_info; | |
1698 | struct fuse_context *fc = fuse_get_context(); | |
1699 | int ret; | |
1700 | ||
1701 | if (!fc) | |
1702 | return -EIO; | |
1703 | ||
1704 | controller = pick_controller_from_path(fc, path); | |
1705 | if (!controller) | |
1706 | return -EIO; | |
1707 | cgroup = find_cgroup_in_path(path); | |
1708 | if (!cgroup) | |
1709 | return -EINVAL; | |
1710 | ||
1711 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1712 | if (!last) { | |
1713 | path1 = "/"; | |
1714 | path2 = cgdir; | |
1715 | } else { | |
1716 | path1 = cgdir; | |
1717 | path2 = last; | |
1718 | } | |
1719 | ||
1720 | k = cgfs_get_key(controller, path1, path2); | |
1721 | if (!k) { | |
1722 | ret = -EINVAL; | |
1723 | goto out; | |
1724 | } | |
1725 | free_key(k); | |
1726 | ||
1727 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1728 | if (initpid <= 0) | |
1729 | initpid = fc->pid; | |
1730 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1731 | ret = -ENOENT; | |
1732 | goto out; | |
1733 | } | |
1734 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
1735 | // should never get here | |
1736 | ret = -EACCES; | |
1737 | goto out; | |
1738 | } | |
1739 | ||
1740 | /* we'll free this at cg_release */ | |
1741 | file_info = malloc(sizeof(*file_info)); | |
1742 | if (!file_info) { | |
1743 | ret = -ENOMEM; | |
1744 | goto out; | |
1745 | } | |
1746 | file_info->controller = must_copy_string(controller); | |
1747 | file_info->cgroup = must_copy_string(path1); | |
1748 | file_info->file = must_copy_string(path2); | |
1749 | file_info->type = LXC_TYPE_CGFILE; | |
1750 | file_info->buf = NULL; | |
1751 | file_info->buflen = 0; | |
1752 | ||
1753 | fi->fh = (unsigned long)file_info; | |
1754 | ret = 0; | |
1755 | ||
1756 | out: | |
1757 | free(cgdir); | |
1758 | return ret; | |
1759 | } | |
1760 | ||
1761 | int cg_release(const char *path, struct fuse_file_info *fi) | |
1762 | { | |
1763 | struct file_info *f = (struct file_info *)fi->fh; | |
1764 | ||
1765 | do_release_file_info(f); | |
1766 | return 0; | |
1767 | } | |
1768 | ||
1769 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
1770 | ||
1771 | static bool wait_for_sock(int sock, int timeout) | |
1772 | { | |
1773 | struct epoll_event ev; | |
1774 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
1775 | ||
1776 | if ((starttime = time(NULL)) < 0) | |
1777 | return false; | |
1778 | ||
1779 | if ((epfd = epoll_create(1)) < 0) { | |
1780 | fprintf(stderr, "Failed to create epoll socket: %m\n"); | |
1781 | return false; | |
1782 | } | |
1783 | ||
1784 | ev.events = POLLIN_SET; | |
1785 | ev.data.fd = sock; | |
1786 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
1787 | fprintf(stderr, "Failed adding socket to epoll: %m\n"); | |
1788 | close(epfd); | |
1789 | return false; | |
1790 | } | |
1791 | ||
1792 | again: | |
1793 | if ((now = time(NULL)) < 0) { | |
1794 | close(epfd); | |
1795 | return false; | |
1796 | } | |
1797 | ||
1798 | deltatime = (starttime + timeout) - now; | |
1799 | if (deltatime < 0) { // timeout | |
1800 | errno = 0; | |
1801 | close(epfd); | |
1802 | return false; | |
1803 | } | |
1804 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
1805 | if (ret < 0 && errno == EINTR) | |
1806 | goto again; | |
1807 | saved_errno = errno; | |
1808 | close(epfd); | |
1809 | ||
1810 | if (ret <= 0) { | |
1811 | errno = saved_errno; | |
1812 | return false; | |
1813 | } | |
1814 | return true; | |
1815 | } | |
1816 | ||
1817 | static int msgrecv(int sockfd, void *buf, size_t len) | |
1818 | { | |
1819 | if (!wait_for_sock(sockfd, 2)) | |
1820 | return -1; | |
1821 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
1822 | } | |
1823 | ||
1824 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
1825 | { | |
1826 | struct msghdr msg = { 0 }; | |
1827 | struct iovec iov; | |
1828 | struct cmsghdr *cmsg; | |
1829 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
1830 | char buf[1]; | |
1831 | buf[0] = 'p'; | |
1832 | ||
1833 | if (pingfirst) { | |
1834 | if (msgrecv(sock, buf, 1) != 1) { | |
1835 | fprintf(stderr, "%s: Error getting reply from server over socketpair\n", | |
1836 | __func__); | |
1837 | return SEND_CREDS_FAIL; | |
1838 | } | |
1839 | } | |
1840 | ||
1841 | msg.msg_control = cmsgbuf; | |
1842 | msg.msg_controllen = sizeof(cmsgbuf); | |
1843 | ||
1844 | cmsg = CMSG_FIRSTHDR(&msg); | |
1845 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
1846 | cmsg->cmsg_level = SOL_SOCKET; | |
1847 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
1848 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
1849 | ||
1850 | msg.msg_name = NULL; | |
1851 | msg.msg_namelen = 0; | |
1852 | ||
1853 | buf[0] = v; | |
1854 | iov.iov_base = buf; | |
1855 | iov.iov_len = sizeof(buf); | |
1856 | msg.msg_iov = &iov; | |
1857 | msg.msg_iovlen = 1; | |
1858 | ||
1859 | if (sendmsg(sock, &msg, 0) < 0) { | |
1860 | fprintf(stderr, "%s: failed at sendmsg: %s\n", __func__, | |
1861 | strerror(errno)); | |
1862 | if (errno == 3) | |
1863 | return SEND_CREDS_NOTSK; | |
1864 | return SEND_CREDS_FAIL; | |
1865 | } | |
1866 | ||
1867 | return SEND_CREDS_OK; | |
1868 | } | |
1869 | ||
1870 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
1871 | { | |
1872 | struct msghdr msg = { 0 }; | |
1873 | struct iovec iov; | |
1874 | struct cmsghdr *cmsg; | |
1875 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
1876 | char buf[1]; | |
1877 | int ret; | |
1878 | int optval = 1; | |
1879 | ||
1880 | *v = '1'; | |
1881 | ||
1882 | cred->pid = -1; | |
1883 | cred->uid = -1; | |
1884 | cred->gid = -1; | |
1885 | ||
1886 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
1887 | fprintf(stderr, "Failed to set passcred: %s\n", strerror(errno)); | |
1888 | return false; | |
1889 | } | |
1890 | buf[0] = '1'; | |
1891 | if (write(sock, buf, 1) != 1) { | |
1892 | fprintf(stderr, "Failed to start write on scm fd: %s\n", strerror(errno)); | |
1893 | return false; | |
1894 | } | |
1895 | ||
1896 | msg.msg_name = NULL; | |
1897 | msg.msg_namelen = 0; | |
1898 | msg.msg_control = cmsgbuf; | |
1899 | msg.msg_controllen = sizeof(cmsgbuf); | |
1900 | ||
1901 | iov.iov_base = buf; | |
1902 | iov.iov_len = sizeof(buf); | |
1903 | msg.msg_iov = &iov; | |
1904 | msg.msg_iovlen = 1; | |
1905 | ||
1906 | if (!wait_for_sock(sock, 2)) { | |
1907 | fprintf(stderr, "Timed out waiting for scm_cred: %s\n", | |
1908 | strerror(errno)); | |
1909 | return false; | |
1910 | } | |
1911 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
1912 | if (ret < 0) { | |
1913 | fprintf(stderr, "Failed to receive scm_cred: %s\n", | |
1914 | strerror(errno)); | |
1915 | return false; | |
1916 | } | |
1917 | ||
1918 | cmsg = CMSG_FIRSTHDR(&msg); | |
1919 | ||
1920 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
1921 | cmsg->cmsg_level == SOL_SOCKET && | |
1922 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
1923 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
1924 | } | |
1925 | *v = buf[0]; | |
1926 | ||
1927 | return true; | |
1928 | } | |
1929 | ||
1930 | ||
1931 | /* | |
1932 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
1933 | * int value back over the socket. This shifts the pid from the | |
1934 | * sender's pidns into tpid's pidns. | |
1935 | */ | |
1936 | static void pid_to_ns(int sock, pid_t tpid) | |
1937 | { | |
1938 | char v = '0'; | |
1939 | struct ucred cred; | |
1940 | ||
1941 | while (recv_creds(sock, &cred, &v)) { | |
1942 | if (v == '1') | |
1943 | _exit(0); | |
1944 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) | |
1945 | _exit(1); | |
1946 | } | |
1947 | _exit(0); | |
1948 | } | |
1949 | ||
1950 | /* | |
1951 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
1952 | * in your old pidns. Only children which you fork will be in the target | |
1953 | * pidns. So the pid_to_ns_wrapper does the setns, then forks a child to | |
1954 | * actually convert pids | |
1955 | */ | |
1956 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
1957 | { | |
1958 | int newnsfd = -1, ret, cpipe[2]; | |
1959 | char fnam[100]; | |
1960 | pid_t cpid; | |
1961 | char v; | |
1962 | ||
1963 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
1964 | if (ret < 0 || ret >= sizeof(fnam)) | |
1965 | _exit(1); | |
1966 | newnsfd = open(fnam, O_RDONLY); | |
1967 | if (newnsfd < 0) | |
1968 | _exit(1); | |
1969 | if (setns(newnsfd, 0) < 0) | |
1970 | _exit(1); | |
1971 | close(newnsfd); | |
1972 | ||
1973 | if (pipe(cpipe) < 0) | |
1974 | _exit(1); | |
1975 | ||
1976 | cpid = fork(); | |
1977 | if (cpid < 0) | |
1978 | _exit(1); | |
1979 | ||
1980 | if (!cpid) { | |
1981 | char b = '1'; | |
1982 | close(cpipe[0]); | |
1983 | if (write(cpipe[1], &b, sizeof(char)) < 0) { | |
1984 | fprintf(stderr, "%s (child): erorr on write: %s\n", | |
1985 | __func__, strerror(errno)); | |
1986 | } | |
1987 | close(cpipe[1]); | |
1988 | pid_to_ns(sock, tpid); | |
1989 | _exit(1); // not reached | |
1990 | } | |
1991 | // give the child 1 second to be done forking and | |
1992 | // write its ack | |
1993 | if (!wait_for_sock(cpipe[0], 1)) | |
1994 | _exit(1); | |
1995 | ret = read(cpipe[0], &v, 1); | |
1996 | if (ret != sizeof(char) || v != '1') | |
1997 | _exit(1); | |
1998 | ||
1999 | if (!wait_for_pid(cpid)) | |
2000 | _exit(1); | |
2001 | _exit(0); | |
2002 | } | |
2003 | ||
2004 | /* | |
2005 | * To read cgroup files with a particular pid, we will setns into the child | |
2006 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2007 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2008 | */ | |
2009 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2010 | { | |
2011 | int sock[2] = {-1, -1}; | |
2012 | char *tmpdata = NULL; | |
2013 | int ret; | |
2014 | pid_t qpid, cpid = -1; | |
2015 | bool answer = false; | |
2016 | char v = '0'; | |
2017 | struct ucred cred; | |
2018 | size_t sz = 0, asz = 0; | |
2019 | ||
2020 | if (!cgfs_get_value(contrl, cg, file, &tmpdata)) | |
2021 | return false; | |
2022 | ||
2023 | /* | |
2024 | * Now we read the pids from returned data one by one, pass | |
2025 | * them into a child in the target namespace, read back the | |
2026 | * translated pids, and put them into our to-return data | |
2027 | */ | |
2028 | ||
2029 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2030 | perror("socketpair"); | |
2031 | free(tmpdata); | |
2032 | return false; | |
2033 | } | |
2034 | ||
2035 | cpid = fork(); | |
2036 | if (cpid == -1) | |
2037 | goto out; | |
2038 | ||
2039 | if (!cpid) // child - exits when done | |
2040 | pid_to_ns_wrapper(sock[1], tpid); | |
2041 | ||
2042 | char *ptr = tmpdata; | |
2043 | cred.uid = 0; | |
2044 | cred.gid = 0; | |
2045 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2046 | cred.pid = qpid; | |
2047 | ret = send_creds(sock[0], &cred, v, true); | |
2048 | ||
2049 | if (ret == SEND_CREDS_NOTSK) | |
2050 | goto next; | |
2051 | if (ret == SEND_CREDS_FAIL) | |
2052 | goto out; | |
2053 | ||
2054 | // read converted results | |
2055 | if (!wait_for_sock(sock[0], 2)) { | |
2056 | fprintf(stderr, "%s: timed out waiting for pid from child: %s\n", | |
2057 | __func__, strerror(errno)); | |
2058 | goto out; | |
2059 | } | |
2060 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
2061 | fprintf(stderr, "%s: error reading pid from child: %s\n", | |
2062 | __func__, strerror(errno)); | |
2063 | goto out; | |
2064 | } | |
2065 | must_strcat_pid(d, &sz, &asz, qpid); | |
2066 | next: | |
2067 | ptr = strchr(ptr, '\n'); | |
2068 | if (!ptr) | |
2069 | break; | |
2070 | ptr++; | |
2071 | } | |
2072 | ||
2073 | cred.pid = getpid(); | |
2074 | v = '1'; | |
2075 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2076 | // failed to ask child to exit | |
2077 | fprintf(stderr, "%s: failed to ask child to exit: %s\n", | |
2078 | __func__, strerror(errno)); | |
2079 | goto out; | |
2080 | } | |
2081 | ||
2082 | answer = true; | |
2083 | ||
2084 | out: | |
2085 | free(tmpdata); | |
2086 | if (cpid != -1) | |
2087 | wait_for_pid(cpid); | |
2088 | if (sock[0] != -1) { | |
2089 | close(sock[0]); | |
2090 | close(sock[1]); | |
2091 | } | |
2092 | return answer; | |
2093 | } | |
2094 | ||
2095 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2096 | struct fuse_file_info *fi) | |
2097 | { | |
2098 | struct fuse_context *fc = fuse_get_context(); | |
2099 | struct file_info *f = (struct file_info *)fi->fh; | |
2100 | struct cgfs_files *k = NULL; | |
2101 | char *data = NULL; | |
2102 | int ret, s; | |
2103 | bool r; | |
2104 | ||
2105 | if (f->type != LXC_TYPE_CGFILE) { | |
2106 | fprintf(stderr, "Internal error: directory cache info used in cg_read\n"); | |
2107 | return -EIO; | |
2108 | } | |
2109 | ||
2110 | if (offset) | |
2111 | return 0; | |
2112 | ||
2113 | if (!fc) | |
2114 | return -EIO; | |
2115 | ||
2116 | if (!f->controller) | |
2117 | return -EINVAL; | |
2118 | ||
2119 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2120 | return -EINVAL; | |
2121 | } | |
2122 | free_key(k); | |
2123 | ||
2124 | ||
2125 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { // should never get here | |
2126 | ret = -EACCES; | |
2127 | goto out; | |
2128 | } | |
2129 | ||
2130 | if (strcmp(f->file, "tasks") == 0 || | |
2131 | strcmp(f->file, "/tasks") == 0 || | |
2132 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2133 | strcmp(f->file, "cgroup.procs") == 0) | |
2134 | // special case - we have to translate the pids | |
2135 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2136 | else | |
2137 | r = cgfs_get_value(f->controller, f->cgroup, f->file, &data); | |
2138 | ||
2139 | if (!r) { | |
2140 | ret = -EINVAL; | |
2141 | goto out; | |
2142 | } | |
2143 | ||
2144 | if (!data) { | |
2145 | ret = 0; | |
2146 | goto out; | |
2147 | } | |
2148 | s = strlen(data); | |
2149 | if (s > size) | |
2150 | s = size; | |
2151 | memcpy(buf, data, s); | |
2152 | if (s > 0 && s < size && data[s-1] != '\n') | |
2153 | buf[s++] = '\n'; | |
2154 | ||
2155 | ret = s; | |
2156 | ||
2157 | out: | |
2158 | free(data); | |
2159 | return ret; | |
2160 | } | |
2161 | ||
2162 | static void pid_from_ns(int sock, pid_t tpid) | |
2163 | { | |
2164 | pid_t vpid; | |
2165 | struct ucred cred; | |
2166 | char v; | |
2167 | int ret; | |
2168 | ||
2169 | cred.uid = 0; | |
2170 | cred.gid = 0; | |
2171 | while (1) { | |
2172 | if (!wait_for_sock(sock, 2)) { | |
2173 | fprintf(stderr, "%s: timeout reading from parent\n", __func__); | |
2174 | _exit(1); | |
2175 | } | |
2176 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
2177 | fprintf(stderr, "%s: bad read from parent: %s\n", | |
2178 | __func__, strerror(errno)); | |
2179 | _exit(1); | |
2180 | } | |
2181 | if (vpid == -1) // done | |
2182 | break; | |
2183 | v = '0'; | |
2184 | cred.pid = vpid; | |
2185 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2186 | v = '1'; | |
2187 | cred.pid = getpid(); | |
2188 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
2189 | _exit(1); | |
2190 | } | |
2191 | } | |
2192 | _exit(0); | |
2193 | } | |
2194 | ||
2195 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2196 | { | |
2197 | int newnsfd = -1, ret, cpipe[2]; | |
2198 | char fnam[100]; | |
2199 | pid_t cpid; | |
2200 | char v; | |
2201 | ||
2202 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2203 | if (ret < 0 || ret >= sizeof(fnam)) | |
2204 | _exit(1); | |
2205 | newnsfd = open(fnam, O_RDONLY); | |
2206 | if (newnsfd < 0) | |
2207 | _exit(1); | |
2208 | if (setns(newnsfd, 0) < 0) | |
2209 | _exit(1); | |
2210 | close(newnsfd); | |
2211 | ||
2212 | if (pipe(cpipe) < 0) | |
2213 | _exit(1); | |
2214 | ||
2215 | loop: | |
2216 | cpid = fork(); | |
2217 | ||
2218 | if (cpid < 0) | |
2219 | _exit(1); | |
2220 | ||
2221 | if (!cpid) { | |
2222 | char b = '1'; | |
2223 | close(cpipe[0]); | |
2224 | if (write(cpipe[1], &b, sizeof(char)) < 0) { | |
2225 | fprintf(stderr, "%s (child): erorr on write: %s\n", | |
2226 | __func__, strerror(errno)); | |
2227 | } | |
2228 | close(cpipe[1]); | |
2229 | pid_from_ns(sock, tpid); | |
2230 | } | |
2231 | ||
2232 | // give the child 1 second to be done forking and | |
2233 | // write its ack | |
2234 | if (!wait_for_sock(cpipe[0], 1)) | |
2235 | goto again; | |
2236 | ret = read(cpipe[0], &v, 1); | |
2237 | if (ret != sizeof(char) || v != '1') { | |
2238 | goto again; | |
2239 | } | |
2240 | ||
2241 | if (!wait_for_pid(cpid)) | |
2242 | _exit(1); | |
2243 | _exit(0); | |
2244 | ||
2245 | again: | |
2246 | kill(cpid, SIGKILL); | |
2247 | wait_for_pid(cpid); | |
2248 | goto loop; | |
2249 | } | |
2250 | ||
2251 | /* | |
2252 | * Given host @uid, return the uid to which it maps in | |
2253 | * @pid's user namespace, or -1 if none. | |
2254 | */ | |
2255 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2256 | { | |
2257 | FILE *f; | |
2258 | char line[400]; | |
2259 | ||
2260 | sprintf(line, "/proc/%d/uid_map", pid); | |
2261 | if ((f = fopen(line, "r")) == NULL) { | |
2262 | return false; | |
2263 | } | |
2264 | ||
2265 | *answer = convert_id_to_ns(f, uid); | |
2266 | fclose(f); | |
2267 | ||
2268 | if (*answer == -1) | |
2269 | return false; | |
2270 | return true; | |
2271 | } | |
2272 | ||
2273 | /* | |
2274 | * get_pid_creds: get the real uid and gid of @pid from | |
2275 | * /proc/$$/status | |
2276 | * (XXX should we use euid here?) | |
2277 | */ | |
2278 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2279 | { | |
2280 | char line[400]; | |
2281 | uid_t u; | |
2282 | gid_t g; | |
2283 | FILE *f; | |
2284 | ||
2285 | *uid = -1; | |
2286 | *gid = -1; | |
2287 | sprintf(line, "/proc/%d/status", pid); | |
2288 | if ((f = fopen(line, "r")) == NULL) { | |
2289 | fprintf(stderr, "Error opening %s: %s\n", line, strerror(errno)); | |
2290 | return; | |
2291 | } | |
2292 | while (fgets(line, 400, f)) { | |
2293 | if (strncmp(line, "Uid:", 4) == 0) { | |
2294 | if (sscanf(line+4, "%u", &u) != 1) { | |
2295 | fprintf(stderr, "bad uid line for pid %u\n", pid); | |
2296 | fclose(f); | |
2297 | return; | |
2298 | } | |
2299 | *uid = u; | |
2300 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2301 | if (sscanf(line+4, "%u", &g) != 1) { | |
2302 | fprintf(stderr, "bad gid line for pid %u\n", pid); | |
2303 | fclose(f); | |
2304 | return; | |
2305 | } | |
2306 | *gid = g; | |
2307 | } | |
2308 | } | |
2309 | fclose(f); | |
2310 | } | |
2311 | ||
2312 | /* | |
2313 | * May the requestor @r move victim @v to a new cgroup? | |
2314 | * This is allowed if | |
2315 | * . they are the same task | |
2316 | * . they are ownedy by the same uid | |
2317 | * . @r is root on the host, or | |
2318 | * . @v's uid is mapped into @r's where @r is root. | |
2319 | */ | |
2320 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2321 | { | |
2322 | uid_t v_uid, tmpuid; | |
2323 | gid_t v_gid; | |
2324 | ||
2325 | if (r == v) | |
2326 | return true; | |
2327 | if (r_uid == 0) | |
2328 | return true; | |
2329 | get_pid_creds(v, &v_uid, &v_gid); | |
2330 | if (r_uid == v_uid) | |
2331 | return true; | |
2332 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2333 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2334 | return true; | |
2335 | return false; | |
2336 | } | |
2337 | ||
2338 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2339 | const char *file, const char *buf) | |
2340 | { | |
2341 | int sock[2] = {-1, -1}; | |
2342 | pid_t qpid, cpid = -1; | |
2343 | FILE *pids_file = NULL; | |
2344 | bool answer = false, fail = false; | |
2345 | ||
2346 | pids_file = open_pids_file(contrl, cg); | |
2347 | if (!pids_file) | |
2348 | return false; | |
2349 | ||
2350 | /* | |
2351 | * write the pids to a socket, have helper in writer's pidns | |
2352 | * call movepid for us | |
2353 | */ | |
2354 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2355 | perror("socketpair"); | |
2356 | goto out; | |
2357 | } | |
2358 | ||
2359 | cpid = fork(); | |
2360 | if (cpid == -1) | |
2361 | goto out; | |
2362 | ||
2363 | if (!cpid) { // child | |
2364 | fclose(pids_file); | |
2365 | pid_from_ns_wrapper(sock[1], tpid); | |
2366 | } | |
2367 | ||
2368 | const char *ptr = buf; | |
2369 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
2370 | struct ucred cred; | |
2371 | char v; | |
2372 | ||
2373 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
2374 | fprintf(stderr, "%s: error writing pid to child: %s\n", | |
2375 | __func__, strerror(errno)); | |
2376 | goto out; | |
2377 | } | |
2378 | ||
2379 | if (recv_creds(sock[0], &cred, &v)) { | |
2380 | if (v == '0') { | |
2381 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
2382 | fail = true; | |
2383 | break; | |
2384 | } | |
2385 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
2386 | fail = true; | |
2387 | } | |
2388 | } | |
2389 | ||
2390 | ptr = strchr(ptr, '\n'); | |
2391 | if (!ptr) | |
2392 | break; | |
2393 | ptr++; | |
2394 | } | |
2395 | ||
2396 | /* All good, write the value */ | |
2397 | qpid = -1; | |
2398 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
2399 | fprintf(stderr, "Warning: failed to ask child to exit\n"); | |
2400 | ||
2401 | if (!fail) | |
2402 | answer = true; | |
2403 | ||
2404 | out: | |
2405 | if (cpid != -1) | |
2406 | wait_for_pid(cpid); | |
2407 | if (sock[0] != -1) { | |
2408 | close(sock[0]); | |
2409 | close(sock[1]); | |
2410 | } | |
2411 | if (pids_file) { | |
2412 | if (fclose(pids_file) != 0) | |
2413 | answer = false; | |
2414 | } | |
2415 | return answer; | |
2416 | } | |
2417 | ||
2418 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
2419 | struct fuse_file_info *fi) | |
2420 | { | |
2421 | struct fuse_context *fc = fuse_get_context(); | |
2422 | char *localbuf = NULL; | |
2423 | struct cgfs_files *k = NULL; | |
2424 | struct file_info *f = (struct file_info *)fi->fh; | |
2425 | bool r; | |
2426 | ||
2427 | if (f->type != LXC_TYPE_CGFILE) { | |
2428 | fprintf(stderr, "Internal error: directory cache info used in cg_write\n"); | |
2429 | return -EIO; | |
2430 | } | |
2431 | ||
2432 | if (offset) | |
2433 | return 0; | |
2434 | ||
2435 | if (!fc) | |
2436 | return -EIO; | |
2437 | ||
2438 | localbuf = alloca(size+1); | |
2439 | localbuf[size] = '\0'; | |
2440 | memcpy(localbuf, buf, size); | |
2441 | ||
2442 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2443 | size = -EINVAL; | |
2444 | goto out; | |
2445 | } | |
2446 | ||
2447 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
2448 | size = -EACCES; | |
2449 | goto out; | |
2450 | } | |
2451 | ||
2452 | if (strcmp(f->file, "tasks") == 0 || | |
2453 | strcmp(f->file, "/tasks") == 0 || | |
2454 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2455 | strcmp(f->file, "cgroup.procs") == 0) | |
2456 | // special case - we have to translate the pids | |
2457 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
2458 | else | |
2459 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
2460 | ||
2461 | if (!r) | |
2462 | size = -EINVAL; | |
2463 | ||
2464 | out: | |
2465 | free_key(k); | |
2466 | return size; | |
2467 | } | |
2468 | ||
2469 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
2470 | { | |
2471 | struct fuse_context *fc = fuse_get_context(); | |
2472 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2473 | struct cgfs_files *k = NULL; | |
2474 | const char *cgroup; | |
2475 | int ret; | |
2476 | ||
2477 | if (!fc) | |
2478 | return -EIO; | |
2479 | ||
2480 | if (strcmp(path, "/cgroup") == 0) | |
2481 | return -EINVAL; | |
2482 | ||
2483 | controller = pick_controller_from_path(fc, path); | |
2484 | if (!controller) | |
2485 | return -EINVAL; | |
2486 | cgroup = find_cgroup_in_path(path); | |
2487 | if (!cgroup) | |
2488 | /* this is just /cgroup/controller */ | |
2489 | return -EINVAL; | |
2490 | ||
2491 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2492 | ||
2493 | if (!last) { | |
2494 | path1 = "/"; | |
2495 | path2 = cgdir; | |
2496 | } else { | |
2497 | path1 = cgdir; | |
2498 | path2 = last; | |
2499 | } | |
2500 | ||
2501 | if (is_child_cgroup(controller, path1, path2)) { | |
2502 | // get uid, gid, from '/tasks' file and make up a mode | |
2503 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2504 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2505 | ||
2506 | } else | |
2507 | k = cgfs_get_key(controller, path1, path2); | |
2508 | ||
2509 | if (!k) { | |
2510 | ret = -EINVAL; | |
2511 | goto out; | |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | * This being a fuse request, the uid and gid must be valid | |
2516 | * in the caller's namespace. So we can just check to make | |
2517 | * sure that the caller is root in his uid, and privileged | |
2518 | * over the file's current owner. | |
2519 | */ | |
2520 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
2521 | ret = -EACCES; | |
2522 | goto out; | |
2523 | } | |
2524 | ||
2525 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
2526 | ||
2527 | out: | |
2528 | free_key(k); | |
2529 | free(cgdir); | |
2530 | ||
2531 | return ret; | |
2532 | } | |
2533 | ||
2534 | int cg_chmod(const char *path, mode_t mode) | |
2535 | { | |
2536 | struct fuse_context *fc = fuse_get_context(); | |
2537 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2538 | struct cgfs_files *k = NULL; | |
2539 | const char *cgroup; | |
2540 | int ret; | |
2541 | ||
2542 | if (!fc) | |
2543 | return -EIO; | |
2544 | ||
2545 | if (strcmp(path, "/cgroup") == 0) | |
2546 | return -EINVAL; | |
2547 | ||
2548 | controller = pick_controller_from_path(fc, path); | |
2549 | if (!controller) | |
2550 | return -EINVAL; | |
2551 | cgroup = find_cgroup_in_path(path); | |
2552 | if (!cgroup) | |
2553 | /* this is just /cgroup/controller */ | |
2554 | return -EINVAL; | |
2555 | ||
2556 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2557 | ||
2558 | if (!last) { | |
2559 | path1 = "/"; | |
2560 | path2 = cgdir; | |
2561 | } else { | |
2562 | path1 = cgdir; | |
2563 | path2 = last; | |
2564 | } | |
2565 | ||
2566 | if (is_child_cgroup(controller, path1, path2)) { | |
2567 | // get uid, gid, from '/tasks' file and make up a mode | |
2568 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2569 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2570 | ||
2571 | } else | |
2572 | k = cgfs_get_key(controller, path1, path2); | |
2573 | ||
2574 | if (!k) { | |
2575 | ret = -EINVAL; | |
2576 | goto out; | |
2577 | } | |
2578 | ||
2579 | /* | |
2580 | * This being a fuse request, the uid and gid must be valid | |
2581 | * in the caller's namespace. So we can just check to make | |
2582 | * sure that the caller is root in his uid, and privileged | |
2583 | * over the file's current owner. | |
2584 | */ | |
2585 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
2586 | ret = -EPERM; | |
2587 | goto out; | |
2588 | } | |
2589 | ||
2590 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
2591 | ret = -EINVAL; | |
2592 | goto out; | |
2593 | } | |
2594 | ||
2595 | ret = 0; | |
2596 | out: | |
2597 | free_key(k); | |
2598 | free(cgdir); | |
2599 | return ret; | |
2600 | } | |
2601 | ||
2602 | int cg_mkdir(const char *path, mode_t mode) | |
2603 | { | |
2604 | struct fuse_context *fc = fuse_get_context(); | |
2605 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
2606 | const char *cgroup; | |
2607 | int ret; | |
2608 | ||
2609 | if (!fc) | |
2610 | return -EIO; | |
2611 | ||
2612 | ||
2613 | controller = pick_controller_from_path(fc, path); | |
2614 | if (!controller) | |
2615 | return -EINVAL; | |
2616 | ||
2617 | cgroup = find_cgroup_in_path(path); | |
2618 | if (!cgroup) | |
2619 | return -EINVAL; | |
2620 | ||
2621 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2622 | if (!last) | |
2623 | path1 = "/"; | |
2624 | else | |
2625 | path1 = cgdir; | |
2626 | ||
2627 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2628 | if (initpid <= 0) | |
2629 | initpid = fc->pid; | |
2630 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
2631 | if (!next) | |
2632 | ret = -EINVAL; | |
2633 | else if (last && strcmp(next, last) == 0) | |
2634 | ret = -EEXIST; | |
2635 | else | |
2636 | ret = -ENOENT; | |
2637 | goto out; | |
2638 | } | |
2639 | ||
2640 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
2641 | ret = -EACCES; | |
2642 | goto out; | |
2643 | } | |
2644 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
2645 | ret = -EACCES; | |
2646 | goto out; | |
2647 | } | |
2648 | ||
2649 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
2650 | ||
2651 | out: | |
2652 | free(cgdir); | |
2653 | free(next); | |
2654 | return ret; | |
2655 | } | |
2656 | ||
2657 | int cg_rmdir(const char *path) | |
2658 | { | |
2659 | struct fuse_context *fc = fuse_get_context(); | |
2660 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
2661 | const char *cgroup; | |
2662 | int ret; | |
2663 | ||
2664 | if (!fc) | |
2665 | return -EIO; | |
2666 | ||
2667 | controller = pick_controller_from_path(fc, path); | |
2668 | if (!controller) | |
2669 | return -EINVAL; | |
2670 | ||
2671 | cgroup = find_cgroup_in_path(path); | |
2672 | if (!cgroup) | |
2673 | return -EINVAL; | |
2674 | ||
2675 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2676 | if (!last) { | |
2677 | ret = -EINVAL; | |
2678 | goto out; | |
2679 | } | |
2680 | ||
2681 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2682 | if (initpid <= 0) | |
2683 | initpid = fc->pid; | |
2684 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
2685 | if (!last || strcmp(next, last) == 0) | |
2686 | ret = -EBUSY; | |
2687 | else | |
2688 | ret = -ENOENT; | |
2689 | goto out; | |
2690 | } | |
2691 | ||
2692 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
2693 | ret = -EACCES; | |
2694 | goto out; | |
2695 | } | |
2696 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
2697 | ret = -EACCES; | |
2698 | goto out; | |
2699 | } | |
2700 | ||
2701 | if (!cgfs_remove(controller, cgroup)) { | |
2702 | ret = -EINVAL; | |
2703 | goto out; | |
2704 | } | |
2705 | ||
2706 | ret = 0; | |
2707 | ||
2708 | out: | |
2709 | free(cgdir); | |
2710 | free(next); | |
2711 | return ret; | |
2712 | } | |
2713 | ||
2714 | static bool startswith(const char *line, const char *pref) | |
2715 | { | |
2716 | if (strncmp(line, pref, strlen(pref)) == 0) | |
2717 | return true; | |
2718 | return false; | |
2719 | } | |
2720 | ||
2721 | static void get_mem_cached(char *memstat, unsigned long *v) | |
2722 | { | |
2723 | char *eol; | |
2724 | ||
2725 | *v = 0; | |
2726 | while (*memstat) { | |
2727 | if (startswith(memstat, "total_cache")) { | |
2728 | sscanf(memstat + 11, "%lu", v); | |
2729 | *v /= 1024; | |
2730 | return; | |
2731 | } | |
2732 | eol = strchr(memstat, '\n'); | |
2733 | if (!eol) | |
2734 | return; | |
2735 | memstat = eol+1; | |
2736 | } | |
2737 | } | |
2738 | ||
2739 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
2740 | { | |
2741 | char *eol; | |
2742 | char key[32]; | |
2743 | ||
2744 | memset(key, 0, 32); | |
2745 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
2746 | ||
2747 | size_t len = strlen(key); | |
2748 | *v = 0; | |
2749 | ||
2750 | while (*str) { | |
2751 | if (startswith(str, key)) { | |
2752 | sscanf(str + len, "%lu", v); | |
2753 | return; | |
2754 | } | |
2755 | eol = strchr(str, '\n'); | |
2756 | if (!eol) | |
2757 | return; | |
2758 | str = eol+1; | |
2759 | } | |
2760 | } | |
2761 | ||
2762 | static int read_file(const char *path, char *buf, size_t size, | |
2763 | struct file_info *d) | |
2764 | { | |
2765 | size_t linelen = 0, total_len = 0, rv = 0; | |
2766 | char *line = NULL; | |
2767 | char *cache = d->buf; | |
2768 | size_t cache_size = d->buflen; | |
2769 | FILE *f = fopen(path, "r"); | |
2770 | if (!f) | |
2771 | return 0; | |
2772 | ||
2773 | while (getline(&line, &linelen, f) != -1) { | |
2774 | size_t l = snprintf(cache, cache_size, "%s", line); | |
2775 | if (l < 0) { | |
2776 | perror("Error writing to cache"); | |
2777 | rv = 0; | |
2778 | goto err; | |
2779 | } | |
2780 | if (l >= cache_size) { | |
2781 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
2782 | rv = 0; | |
2783 | goto err; | |
2784 | } | |
2785 | cache += l; | |
2786 | cache_size -= l; | |
2787 | total_len += l; | |
2788 | } | |
2789 | ||
2790 | d->size = total_len; | |
2791 | if (total_len > size ) total_len = size; | |
2792 | ||
2793 | /* read from off 0 */ | |
2794 | memcpy(buf, d->buf, total_len); | |
2795 | rv = total_len; | |
2796 | err: | |
2797 | fclose(f); | |
2798 | free(line); | |
2799 | return rv; | |
2800 | } | |
2801 | ||
2802 | /* | |
2803 | * FUSE ops for /proc | |
2804 | */ | |
2805 | ||
2806 | static unsigned long get_memlimit(const char *cgroup) | |
2807 | { | |
2808 | char *memlimit_str = NULL; | |
2809 | unsigned long memlimit = -1; | |
2810 | ||
2811 | if (cgfs_get_value("memory", cgroup, "memory.limit_in_bytes", &memlimit_str)) | |
2812 | memlimit = strtoul(memlimit_str, NULL, 10); | |
2813 | ||
2814 | free(memlimit_str); | |
2815 | ||
2816 | return memlimit; | |
2817 | } | |
2818 | ||
2819 | static unsigned long get_min_memlimit(const char *cgroup) | |
2820 | { | |
2821 | char *copy = strdupa(cgroup); | |
2822 | unsigned long memlimit = 0, retlimit; | |
2823 | ||
2824 | retlimit = get_memlimit(copy); | |
2825 | ||
2826 | while (strcmp(copy, "/") != 0) { | |
2827 | copy = dirname(copy); | |
2828 | memlimit = get_memlimit(copy); | |
2829 | if (memlimit != -1 && memlimit < retlimit) | |
2830 | retlimit = memlimit; | |
2831 | }; | |
2832 | ||
2833 | return retlimit; | |
2834 | } | |
2835 | ||
2836 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
2837 | struct fuse_file_info *fi) | |
2838 | { | |
2839 | struct fuse_context *fc = fuse_get_context(); | |
2840 | struct file_info *d = (struct file_info *)fi->fh; | |
2841 | char *cg; | |
2842 | char *memusage_str = NULL, *memstat_str = NULL, | |
2843 | *memswlimit_str = NULL, *memswusage_str = NULL, | |
2844 | *memswlimit_default_str = NULL, *memswusage_default_str = NULL; | |
2845 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0, | |
2846 | cached = 0, hosttotal = 0; | |
2847 | char *line = NULL; | |
2848 | size_t linelen = 0, total_len = 0, rv = 0; | |
2849 | char *cache = d->buf; | |
2850 | size_t cache_size = d->buflen; | |
2851 | FILE *f = NULL; | |
2852 | ||
2853 | if (offset){ | |
2854 | if (offset > d->size) | |
2855 | return -EINVAL; | |
2856 | if (!d->cached) | |
2857 | return 0; | |
2858 | int left = d->size - offset; | |
2859 | total_len = left > size ? size: left; | |
2860 | memcpy(buf, cache + offset, total_len); | |
2861 | return total_len; | |
2862 | } | |
2863 | ||
2864 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2865 | if (initpid <= 0) | |
2866 | initpid = fc->pid; | |
2867 | cg = get_pid_cgroup(initpid, "memory"); | |
2868 | if (!cg) | |
2869 | return read_file("/proc/meminfo", buf, size, d); | |
2870 | ||
2871 | memlimit = get_min_memlimit(cg); | |
2872 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
2873 | goto err; | |
2874 | if (!cgfs_get_value("memory", cg, "memory.stat", &memstat_str)) | |
2875 | goto err; | |
2876 | ||
2877 | // Following values are allowed to fail, because swapaccount might be turned | |
2878 | // off for current kernel | |
2879 | if(cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str) && | |
2880 | cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str)) | |
2881 | { | |
2882 | /* If swapaccounting is turned on, then default value is assumed to be that of cgroup / */ | |
2883 | if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str)) | |
2884 | goto err; | |
2885 | if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str)) | |
2886 | goto err; | |
2887 | ||
2888 | memswlimit = strtoul(memswlimit_str, NULL, 10); | |
2889 | memswusage = strtoul(memswusage_str, NULL, 10); | |
2890 | ||
2891 | if (!strcmp(memswlimit_str, memswlimit_default_str)) | |
2892 | memswlimit = 0; | |
2893 | if (!strcmp(memswusage_str, memswusage_default_str)) | |
2894 | memswusage = 0; | |
2895 | ||
2896 | memswlimit = memswlimit / 1024; | |
2897 | memswusage = memswusage / 1024; | |
2898 | } | |
2899 | ||
2900 | memusage = strtoul(memusage_str, NULL, 10); | |
2901 | memlimit /= 1024; | |
2902 | memusage /= 1024; | |
2903 | ||
2904 | get_mem_cached(memstat_str, &cached); | |
2905 | ||
2906 | f = fopen("/proc/meminfo", "r"); | |
2907 | if (!f) | |
2908 | goto err; | |
2909 | ||
2910 | while (getline(&line, &linelen, f) != -1) { | |
2911 | size_t l; | |
2912 | char *printme, lbuf[100]; | |
2913 | ||
2914 | memset(lbuf, 0, 100); | |
2915 | if (startswith(line, "MemTotal:")) { | |
2916 | sscanf(line+14, "%lu", &hosttotal); | |
2917 | if (hosttotal < memlimit) | |
2918 | memlimit = hosttotal; | |
2919 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
2920 | printme = lbuf; | |
2921 | } else if (startswith(line, "MemFree:")) { | |
2922 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
2923 | printme = lbuf; | |
2924 | } else if (startswith(line, "MemAvailable:")) { | |
2925 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage); | |
2926 | printme = lbuf; | |
2927 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0) { | |
2928 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit - memlimit); | |
2929 | printme = lbuf; | |
2930 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0) { | |
2931 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", | |
2932 | (memswlimit - memlimit) - (memswusage - memusage)); | |
2933 | printme = lbuf; | |
2934 | } else if (startswith(line, "Buffers:")) { | |
2935 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
2936 | printme = lbuf; | |
2937 | } else if (startswith(line, "Cached:")) { | |
2938 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
2939 | printme = lbuf; | |
2940 | } else if (startswith(line, "SwapCached:")) { | |
2941 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
2942 | printme = lbuf; | |
2943 | } else | |
2944 | printme = line; | |
2945 | ||
2946 | l = snprintf(cache, cache_size, "%s", printme); | |
2947 | if (l < 0) { | |
2948 | perror("Error writing to cache"); | |
2949 | rv = 0; | |
2950 | goto err; | |
2951 | ||
2952 | } | |
2953 | if (l >= cache_size) { | |
2954 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
2955 | rv = 0; | |
2956 | goto err; | |
2957 | } | |
2958 | ||
2959 | cache += l; | |
2960 | cache_size -= l; | |
2961 | total_len += l; | |
2962 | } | |
2963 | ||
2964 | d->cached = 1; | |
2965 | d->size = total_len; | |
2966 | if (total_len > size ) total_len = size; | |
2967 | memcpy(buf, d->buf, total_len); | |
2968 | ||
2969 | rv = total_len; | |
2970 | err: | |
2971 | if (f) | |
2972 | fclose(f); | |
2973 | free(line); | |
2974 | free(cg); | |
2975 | free(memusage_str); | |
2976 | free(memswlimit_str); | |
2977 | free(memswusage_str); | |
2978 | free(memstat_str); | |
2979 | free(memswlimit_default_str); | |
2980 | free(memswusage_default_str); | |
2981 | return rv; | |
2982 | } | |
2983 | ||
2984 | /* | |
2985 | * Read the cpuset.cpus for cg | |
2986 | * Return the answer in a newly allocated string which must be freed | |
2987 | */ | |
2988 | static char *get_cpuset(const char *cg) | |
2989 | { | |
2990 | char *answer; | |
2991 | ||
2992 | if (!cgfs_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
2993 | return NULL; | |
2994 | return answer; | |
2995 | } | |
2996 | ||
2997 | bool cpu_in_cpuset(int cpu, const char *cpuset); | |
2998 | ||
2999 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) | |
3000 | { | |
3001 | int cpu; | |
3002 | ||
3003 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
3004 | return false; | |
3005 | return cpu_in_cpuset(cpu, cpuset); | |
3006 | } | |
3007 | ||
3008 | /* | |
3009 | * check whether this is a '^processor" line in /proc/cpuinfo | |
3010 | */ | |
3011 | static bool is_processor_line(const char *line) | |
3012 | { | |
3013 | int cpu; | |
3014 | ||
3015 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
3016 | return true; | |
3017 | return false; | |
3018 | } | |
3019 | ||
3020 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
3021 | struct fuse_file_info *fi) | |
3022 | { | |
3023 | struct fuse_context *fc = fuse_get_context(); | |
3024 | struct file_info *d = (struct file_info *)fi->fh; | |
3025 | char *cg; | |
3026 | char *cpuset = NULL; | |
3027 | char *line = NULL; | |
3028 | size_t linelen = 0, total_len = 0, rv = 0; | |
3029 | bool am_printing = false; | |
3030 | int curcpu = -1; | |
3031 | char *cache = d->buf; | |
3032 | size_t cache_size = d->buflen; | |
3033 | FILE *f = NULL; | |
3034 | ||
3035 | if (offset){ | |
3036 | if (offset > d->size) | |
3037 | return -EINVAL; | |
3038 | if (!d->cached) | |
3039 | return 0; | |
3040 | int left = d->size - offset; | |
3041 | total_len = left > size ? size: left; | |
3042 | memcpy(buf, cache + offset, total_len); | |
3043 | return total_len; | |
3044 | } | |
3045 | ||
3046 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3047 | if (initpid <= 0) | |
3048 | initpid = fc->pid; | |
3049 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3050 | if (!cg) | |
3051 | return read_file("proc/cpuinfo", buf, size, d); | |
3052 | ||
3053 | cpuset = get_cpuset(cg); | |
3054 | if (!cpuset) | |
3055 | goto err; | |
3056 | ||
3057 | f = fopen("/proc/cpuinfo", "r"); | |
3058 | if (!f) | |
3059 | goto err; | |
3060 | ||
3061 | while (getline(&line, &linelen, f) != -1) { | |
3062 | size_t l; | |
3063 | if (is_processor_line(line)) { | |
3064 | am_printing = cpuline_in_cpuset(line, cpuset); | |
3065 | if (am_printing) { | |
3066 | curcpu ++; | |
3067 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); | |
3068 | if (l < 0) { | |
3069 | perror("Error writing to cache"); | |
3070 | rv = 0; | |
3071 | goto err; | |
3072 | } | |
3073 | if (l >= cache_size) { | |
3074 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3075 | rv = 0; | |
3076 | goto err; | |
3077 | } | |
3078 | cache += l; | |
3079 | cache_size -= l; | |
3080 | total_len += l; | |
3081 | } | |
3082 | continue; | |
3083 | } | |
3084 | if (am_printing) { | |
3085 | l = snprintf(cache, cache_size, "%s", line); | |
3086 | if (l < 0) { | |
3087 | perror("Error writing to cache"); | |
3088 | rv = 0; | |
3089 | goto err; | |
3090 | } | |
3091 | if (l >= cache_size) { | |
3092 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3093 | rv = 0; | |
3094 | goto err; | |
3095 | } | |
3096 | cache += l; | |
3097 | cache_size -= l; | |
3098 | total_len += l; | |
3099 | } | |
3100 | } | |
3101 | ||
3102 | d->cached = 1; | |
3103 | d->size = total_len; | |
3104 | if (total_len > size ) total_len = size; | |
3105 | ||
3106 | /* read from off 0 */ | |
3107 | memcpy(buf, d->buf, total_len); | |
3108 | rv = total_len; | |
3109 | err: | |
3110 | if (f) | |
3111 | fclose(f); | |
3112 | free(line); | |
3113 | free(cpuset); | |
3114 | free(cg); | |
3115 | return rv; | |
3116 | } | |
3117 | ||
3118 | static int proc_stat_read(char *buf, size_t size, off_t offset, | |
3119 | struct fuse_file_info *fi) | |
3120 | { | |
3121 | struct fuse_context *fc = fuse_get_context(); | |
3122 | struct file_info *d = (struct file_info *)fi->fh; | |
3123 | char *cg; | |
3124 | char *cpuset = NULL; | |
3125 | char *line = NULL; | |
3126 | size_t linelen = 0, total_len = 0, rv = 0; | |
3127 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
3128 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0; | |
3129 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, | |
3130 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0; | |
3131 | #define CPUALL_MAX_SIZE BUF_RESERVE_SIZE | |
3132 | char cpuall[CPUALL_MAX_SIZE]; | |
3133 | /* reserve for cpu all */ | |
3134 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
3135 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
3136 | FILE *f = NULL; | |
3137 | ||
3138 | if (offset){ | |
3139 | if (offset > d->size) | |
3140 | return -EINVAL; | |
3141 | if (!d->cached) | |
3142 | return 0; | |
3143 | int left = d->size - offset; | |
3144 | total_len = left > size ? size: left; | |
3145 | memcpy(buf, d->buf + offset, total_len); | |
3146 | return total_len; | |
3147 | } | |
3148 | ||
3149 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3150 | if (initpid <= 0) | |
3151 | initpid = fc->pid; | |
3152 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3153 | if (!cg) | |
3154 | return read_file("/proc/stat", buf, size, d); | |
3155 | ||
3156 | cpuset = get_cpuset(cg); | |
3157 | if (!cpuset) | |
3158 | goto err; | |
3159 | ||
3160 | f = fopen("/proc/stat", "r"); | |
3161 | if (!f) | |
3162 | goto err; | |
3163 | ||
3164 | //skip first line | |
3165 | if (getline(&line, &linelen, f) < 0) { | |
3166 | fprintf(stderr, "proc_stat_read read first line failed\n"); | |
3167 | goto err; | |
3168 | } | |
3169 | ||
3170 | while (getline(&line, &linelen, f) != -1) { | |
3171 | size_t l; | |
3172 | int cpu; | |
3173 | char cpu_char[10]; /* That's a lot of cores */ | |
3174 | char *c; | |
3175 | ||
3176 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { | |
3177 | /* not a ^cpuN line containing a number N, just print it */ | |
3178 | l = snprintf(cache, cache_size, "%s", line); | |
3179 | if (l < 0) { | |
3180 | perror("Error writing to cache"); | |
3181 | rv = 0; | |
3182 | goto err; | |
3183 | } | |
3184 | if (l >= cache_size) { | |
3185 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3186 | rv = 0; | |
3187 | goto err; | |
3188 | } | |
3189 | cache += l; | |
3190 | cache_size -= l; | |
3191 | total_len += l; | |
3192 | continue; | |
3193 | } | |
3194 | ||
3195 | if (sscanf(cpu_char, "%d", &cpu) != 1) | |
3196 | continue; | |
3197 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3198 | continue; | |
3199 | curcpu ++; | |
3200 | ||
3201 | c = strchr(line, ' '); | |
3202 | if (!c) | |
3203 | continue; | |
3204 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
3205 | if (l < 0) { | |
3206 | perror("Error writing to cache"); | |
3207 | rv = 0; | |
3208 | goto err; | |
3209 | ||
3210 | } | |
3211 | if (l >= cache_size) { | |
3212 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3213 | rv = 0; | |
3214 | goto err; | |
3215 | } | |
3216 | ||
3217 | cache += l; | |
3218 | cache_size -= l; | |
3219 | total_len += l; | |
3220 | ||
3221 | if (sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user, &nice, &system, &idle, &iowait, &irq, | |
3222 | &softirq, &steal, &guest) != 9) | |
3223 | continue; | |
3224 | user_sum += user; | |
3225 | nice_sum += nice; | |
3226 | system_sum += system; | |
3227 | idle_sum += idle; | |
3228 | iowait_sum += iowait; | |
3229 | irq_sum += irq; | |
3230 | softirq_sum += softirq; | |
3231 | steal_sum += steal; | |
3232 | guest_sum += guest; | |
3233 | } | |
3234 | ||
3235 | cache = d->buf; | |
3236 | ||
3237 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3238 | "cpu ", user_sum, nice_sum, system_sum, idle_sum, iowait_sum, irq_sum, softirq_sum, steal_sum, guest_sum); | |
3239 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE){ | |
3240 | memcpy(cache, cpuall, cpuall_len); | |
3241 | cache += cpuall_len; | |
3242 | } else{ | |
3243 | /* shouldn't happen */ | |
3244 | fprintf(stderr, "proc_stat_read copy cpuall failed, cpuall_len=%d\n", cpuall_len); | |
3245 | cpuall_len = 0; | |
3246 | } | |
3247 | ||
3248 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
3249 | total_len += cpuall_len; | |
3250 | d->cached = 1; | |
3251 | d->size = total_len; | |
3252 | if (total_len > size ) total_len = size; | |
3253 | ||
3254 | memcpy(buf, d->buf, total_len); | |
3255 | rv = total_len; | |
3256 | ||
3257 | err: | |
3258 | if (f) | |
3259 | fclose(f); | |
3260 | free(line); | |
3261 | free(cpuset); | |
3262 | free(cg); | |
3263 | return rv; | |
3264 | } | |
3265 | ||
3266 | static long int getreaperage(pid_t pid) | |
3267 | { | |
3268 | char fnam[100]; | |
3269 | struct stat sb; | |
3270 | int ret; | |
3271 | pid_t qpid; | |
3272 | ||
3273 | qpid = lookup_initpid_in_store(pid); | |
3274 | if (qpid <= 0) | |
3275 | return 0; | |
3276 | ||
3277 | ret = snprintf(fnam, 100, "/proc/%d", qpid); | |
3278 | if (ret < 0 || ret >= 100) | |
3279 | return 0; | |
3280 | ||
3281 | if (lstat(fnam, &sb) < 0) | |
3282 | return 0; | |
3283 | ||
3284 | return time(NULL) - sb.st_ctime; | |
3285 | } | |
3286 | ||
3287 | static unsigned long get_reaper_busy(pid_t task) | |
3288 | { | |
3289 | pid_t initpid = lookup_initpid_in_store(task); | |
3290 | char *cgroup = NULL, *usage_str = NULL; | |
3291 | unsigned long usage = 0; | |
3292 | ||
3293 | if (initpid <= 0) | |
3294 | return 0; | |
3295 | ||
3296 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
3297 | if (!cgroup) | |
3298 | goto out; | |
3299 | if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str)) | |
3300 | goto out; | |
3301 | usage = strtoul(usage_str, NULL, 10); | |
3302 | usage /= 1000000000; | |
3303 | ||
3304 | out: | |
3305 | free(cgroup); | |
3306 | free(usage_str); | |
3307 | return usage; | |
3308 | } | |
3309 | ||
3310 | #if RELOADTEST | |
3311 | void iwashere(void) | |
3312 | { | |
3313 | char *name, *cwd = get_current_dir_name(); | |
3314 | size_t len; | |
3315 | int fd; | |
3316 | ||
3317 | if (!cwd) | |
3318 | exit(1); | |
3319 | len = strlen(cwd) + strlen("/iwashere") + 1; | |
3320 | name = alloca(len); | |
3321 | snprintf(name, len, "%s/iwashere", cwd); | |
3322 | free(cwd); | |
3323 | fd = creat(name, 0755); | |
3324 | if (fd >= 0) | |
3325 | close(fd); | |
3326 | } | |
3327 | #endif | |
3328 | ||
3329 | /* | |
3330 | * We read /proc/uptime and reuse its second field. | |
3331 | * For the first field, we use the mtime for the reaper for | |
3332 | * the calling pid as returned by getreaperage | |
3333 | */ | |
3334 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
3335 | struct fuse_file_info *fi) | |
3336 | { | |
3337 | struct fuse_context *fc = fuse_get_context(); | |
3338 | struct file_info *d = (struct file_info *)fi->fh; | |
3339 | long int reaperage = getreaperage(fc->pid); | |
3340 | unsigned long int busytime = get_reaper_busy(fc->pid), idletime; | |
3341 | char *cache = d->buf; | |
3342 | size_t total_len = 0; | |
3343 | ||
3344 | #if RELOADTEST | |
3345 | iwashere(); | |
3346 | #endif | |
3347 | ||
3348 | if (offset){ | |
3349 | if (offset > d->size) | |
3350 | return -EINVAL; | |
3351 | if (!d->cached) | |
3352 | return 0; | |
3353 | int left = d->size - offset; | |
3354 | total_len = left > size ? size: left; | |
3355 | memcpy(buf, cache + offset, total_len); | |
3356 | return total_len; | |
3357 | } | |
3358 | ||
3359 | idletime = reaperage - busytime; | |
3360 | if (idletime > reaperage) | |
3361 | idletime = reaperage; | |
3362 | ||
3363 | total_len = snprintf(d->buf, d->size, "%ld.0 %lu.0\n", reaperage, idletime); | |
3364 | if (total_len < 0){ | |
3365 | perror("Error writing to cache"); | |
3366 | return 0; | |
3367 | } | |
3368 | ||
3369 | d->size = (int)total_len; | |
3370 | d->cached = 1; | |
3371 | ||
3372 | if (total_len > size) total_len = size; | |
3373 | ||
3374 | memcpy(buf, d->buf, total_len); | |
3375 | return total_len; | |
3376 | } | |
3377 | ||
3378 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
3379 | struct fuse_file_info *fi) | |
3380 | { | |
3381 | char dev_name[72]; | |
3382 | struct fuse_context *fc = fuse_get_context(); | |
3383 | struct file_info *d = (struct file_info *)fi->fh; | |
3384 | char *cg; | |
3385 | char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
3386 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
3387 | unsigned long read = 0, write = 0; | |
3388 | unsigned long read_merged = 0, write_merged = 0; | |
3389 | unsigned long read_sectors = 0, write_sectors = 0; | |
3390 | unsigned long read_ticks = 0, write_ticks = 0; | |
3391 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
3392 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
3393 | char *cache = d->buf; | |
3394 | size_t cache_size = d->buflen; | |
3395 | char *line = NULL; | |
3396 | size_t linelen = 0, total_len = 0, rv = 0; | |
3397 | unsigned int major = 0, minor = 0; | |
3398 | int i = 0; | |
3399 | FILE *f = NULL; | |
3400 | ||
3401 | if (offset){ | |
3402 | if (offset > d->size) | |
3403 | return -EINVAL; | |
3404 | if (!d->cached) | |
3405 | return 0; | |
3406 | int left = d->size - offset; | |
3407 | total_len = left > size ? size: left; | |
3408 | memcpy(buf, cache + offset, total_len); | |
3409 | return total_len; | |
3410 | } | |
3411 | ||
3412 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3413 | if (initpid <= 0) | |
3414 | initpid = fc->pid; | |
3415 | cg = get_pid_cgroup(initpid, "blkio"); | |
3416 | if (!cg) | |
3417 | return read_file("/proc/diskstats", buf, size, d); | |
3418 | ||
3419 | if (!cgfs_get_value("blkio", cg, "blkio.io_serviced", &io_serviced_str)) | |
3420 | goto err; | |
3421 | if (!cgfs_get_value("blkio", cg, "blkio.io_merged", &io_merged_str)) | |
3422 | goto err; | |
3423 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_bytes", &io_service_bytes_str)) | |
3424 | goto err; | |
3425 | if (!cgfs_get_value("blkio", cg, "blkio.io_wait_time", &io_wait_time_str)) | |
3426 | goto err; | |
3427 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_time", &io_service_time_str)) | |
3428 | goto err; | |
3429 | ||
3430 | ||
3431 | f = fopen("/proc/diskstats", "r"); | |
3432 | if (!f) | |
3433 | goto err; | |
3434 | ||
3435 | while (getline(&line, &linelen, f) != -1) { | |
3436 | size_t l; | |
3437 | char *printme, lbuf[256]; | |
3438 | ||
3439 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
3440 | if(i == 3){ | |
3441 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
3442 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
3443 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
3444 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
3445 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
3446 | read_sectors = read_sectors/512; | |
3447 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
3448 | write_sectors = write_sectors/512; | |
3449 | ||
3450 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
3451 | rd_svctm = rd_svctm/1000000; | |
3452 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
3453 | rd_wait = rd_wait/1000000; | |
3454 | read_ticks = rd_svctm + rd_wait; | |
3455 | ||
3456 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
3457 | wr_svctm = wr_svctm/1000000; | |
3458 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
3459 | wr_wait = wr_wait/1000000; | |
3460 | write_ticks = wr_svctm + wr_wait; | |
3461 | ||
3462 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
3463 | tot_ticks = tot_ticks/1000000; | |
3464 | }else{ | |
3465 | continue; | |
3466 | } | |
3467 | ||
3468 | memset(lbuf, 0, 256); | |
3469 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) { | |
3470 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3471 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
3472 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
3473 | printme = lbuf; | |
3474 | } else | |
3475 | continue; | |
3476 | ||
3477 | l = snprintf(cache, cache_size, "%s", printme); | |
3478 | if (l < 0) { | |
3479 | perror("Error writing to fuse buf"); | |
3480 | rv = 0; | |
3481 | goto err; | |
3482 | } | |
3483 | if (l >= cache_size) { | |
3484 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3485 | rv = 0; | |
3486 | goto err; | |
3487 | } | |
3488 | cache += l; | |
3489 | cache_size -= l; | |
3490 | total_len += l; | |
3491 | } | |
3492 | ||
3493 | d->cached = 1; | |
3494 | d->size = total_len; | |
3495 | if (total_len > size ) total_len = size; | |
3496 | memcpy(buf, d->buf, total_len); | |
3497 | ||
3498 | rv = total_len; | |
3499 | err: | |
3500 | free(cg); | |
3501 | if (f) | |
3502 | fclose(f); | |
3503 | free(line); | |
3504 | free(io_serviced_str); | |
3505 | free(io_merged_str); | |
3506 | free(io_service_bytes_str); | |
3507 | free(io_wait_time_str); | |
3508 | free(io_service_time_str); | |
3509 | return rv; | |
3510 | } | |
3511 | ||
70dcc12e SH |
3512 | static int proc_swaps_read(char *buf, size_t size, off_t offset, |
3513 | struct fuse_file_info *fi) | |
3514 | { | |
3515 | struct fuse_context *fc = fuse_get_context(); | |
3516 | struct file_info *d = (struct file_info *)fi->fh; | |
3517 | char *cg = NULL; | |
3518 | char *memswlimit_str = NULL, *memlimit_str = NULL, *memusage_str = NULL, *memswusage_str = NULL, | |
3519 | *memswlimit_default_str = NULL, *memswusage_default_str = NULL; | |
3520 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0, swap_total = 0, swap_free = 0; | |
3521 | size_t total_len = 0, rv = 0; | |
3522 | char *cache = d->buf; | |
3523 | ||
3524 | if (offset) { | |
3525 | if (offset > d->size) | |
3526 | return -EINVAL; | |
3527 | if (!d->cached) | |
3528 | return 0; | |
3529 | int left = d->size - offset; | |
3530 | total_len = left > size ? size: left; | |
3531 | memcpy(buf, cache + offset, total_len); | |
3532 | return total_len; | |
3533 | } | |
3534 | ||
3535 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3536 | if (initpid <= 0) | |
3537 | initpid = fc->pid; | |
3538 | cg = get_pid_cgroup(initpid, "memory"); | |
3539 | if (!cg) | |
3540 | return read_file("/proc/swaps", buf, size, d); | |
3541 | ||
3542 | if (!cgfs_get_value("memory", cg, "memory.limit_in_bytes", &memlimit_str)) | |
3543 | goto err; | |
3544 | ||
3545 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
3546 | goto err; | |
3547 | ||
3548 | memlimit = strtoul(memlimit_str, NULL, 10); | |
3549 | memusage = strtoul(memusage_str, NULL, 10); | |
3550 | ||
3551 | if (cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str) && | |
3552 | cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str)) { | |
3553 | ||
3554 | /* If swap accounting is turned on, then default value is assumed to be that of cgroup / */ | |
3555 | if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str)) | |
3556 | goto err; | |
3557 | if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str)) | |
3558 | goto err; | |
3559 | ||
3560 | memswlimit = strtoul(memswlimit_str, NULL, 10); | |
3561 | memswusage = strtoul(memswusage_str, NULL, 10); | |
3562 | ||
3563 | if (!strcmp(memswlimit_str, memswlimit_default_str)) | |
3564 | memswlimit = 0; | |
3565 | if (!strcmp(memswusage_str, memswusage_default_str)) | |
3566 | memswusage = 0; | |
3567 | ||
3568 | swap_total = (memswlimit - memlimit) / 1024; | |
3569 | swap_free = (memswusage - memusage) / 1024; | |
3570 | } | |
3571 | ||
3572 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
3573 | ||
3574 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
3575 | if (!memswlimit) { | |
3576 | char *line = NULL; | |
3577 | size_t linelen = 0; | |
3578 | FILE *f = fopen("/proc/meminfo", "r"); | |
3579 | ||
3580 | if (!f) | |
3581 | goto err; | |
3582 | ||
3583 | while (getline(&line, &linelen, f) != -1) { | |
3584 | if (startswith(line, "SwapTotal:")) { | |
3585 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); | |
3586 | } else if (startswith(line, "SwapFree:")) { | |
3587 | sscanf(line, "SwapFree: %8lu kB", &swap_free); | |
3588 | } | |
3589 | } | |
3590 | ||
3591 | free(line); | |
3592 | fclose(f); | |
3593 | } | |
3594 | ||
3595 | if (swap_total > 0) { | |
42eba700 | 3596 | total_len += snprintf(d->buf + total_len, d->size - total_len, |
70dcc12e SH |
3597 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", |
3598 | swap_total, swap_free); | |
3599 | } | |
3600 | ||
3601 | if (total_len < 0) { | |
3602 | perror("Error writing to cache"); | |
3603 | rv = 0; | |
3604 | goto err; | |
3605 | } | |
3606 | ||
3607 | d->cached = 1; | |
3608 | d->size = (int)total_len; | |
3609 | ||
3610 | if (total_len > size) total_len = size; | |
3611 | memcpy(buf, d->buf, total_len); | |
3612 | rv = total_len; | |
3613 | ||
3614 | err: | |
3615 | free(cg); | |
3616 | free(memswlimit_str); | |
3617 | free(memlimit_str); | |
3618 | free(memusage_str); | |
3619 | free(memswusage_str); | |
3620 | free(memswusage_default_str); | |
3621 | free(memswlimit_default_str); | |
3622 | return rv; | |
3623 | } | |
3624 | ||
237e200e SH |
3625 | static off_t get_procfile_size(const char *which) |
3626 | { | |
3627 | FILE *f = fopen(which, "r"); | |
3628 | char *line = NULL; | |
3629 | size_t len = 0; | |
3630 | ssize_t sz, answer = 0; | |
3631 | if (!f) | |
3632 | return 0; | |
3633 | ||
3634 | while ((sz = getline(&line, &len, f)) != -1) | |
3635 | answer += sz; | |
3636 | fclose (f); | |
3637 | free(line); | |
3638 | ||
3639 | return answer; | |
3640 | } | |
3641 | ||
3642 | int proc_getattr(const char *path, struct stat *sb) | |
3643 | { | |
3644 | struct timespec now; | |
3645 | ||
3646 | memset(sb, 0, sizeof(struct stat)); | |
3647 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
3648 | return -EINVAL; | |
3649 | sb->st_uid = sb->st_gid = 0; | |
3650 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
3651 | if (strcmp(path, "/proc") == 0) { | |
3652 | sb->st_mode = S_IFDIR | 00555; | |
3653 | sb->st_nlink = 2; | |
3654 | return 0; | |
3655 | } | |
3656 | if (strcmp(path, "/proc/meminfo") == 0 || | |
3657 | strcmp(path, "/proc/cpuinfo") == 0 || | |
3658 | strcmp(path, "/proc/uptime") == 0 || | |
3659 | strcmp(path, "/proc/stat") == 0 || | |
70dcc12e SH |
3660 | strcmp(path, "/proc/diskstats") == 0 || |
3661 | strcmp(path, "/proc/swaps") == 0) { | |
237e200e SH |
3662 | sb->st_size = 0; |
3663 | sb->st_mode = S_IFREG | 00444; | |
3664 | sb->st_nlink = 1; | |
3665 | return 0; | |
3666 | } | |
3667 | ||
3668 | return -ENOENT; | |
3669 | } | |
3670 | ||
3671 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
3672 | struct fuse_file_info *fi) | |
3673 | { | |
3674 | if (filler(buf, "cpuinfo", NULL, 0) != 0 || | |
3675 | filler(buf, "meminfo", NULL, 0) != 0 || | |
3676 | filler(buf, "stat", NULL, 0) != 0 || | |
3677 | filler(buf, "uptime", NULL, 0) != 0 || | |
70dcc12e SH |
3678 | filler(buf, "diskstats", NULL, 0) != 0 || |
3679 | filler(buf, "swaps", NULL, 0) != 0) | |
237e200e SH |
3680 | return -EINVAL; |
3681 | return 0; | |
3682 | } | |
3683 | ||
3684 | int proc_open(const char *path, struct fuse_file_info *fi) | |
3685 | { | |
3686 | int type = -1; | |
3687 | struct file_info *info; | |
3688 | ||
3689 | if (strcmp(path, "/proc/meminfo") == 0) | |
3690 | type = LXC_TYPE_PROC_MEMINFO; | |
3691 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
3692 | type = LXC_TYPE_PROC_CPUINFO; | |
3693 | else if (strcmp(path, "/proc/uptime") == 0) | |
3694 | type = LXC_TYPE_PROC_UPTIME; | |
3695 | else if (strcmp(path, "/proc/stat") == 0) | |
3696 | type = LXC_TYPE_PROC_STAT; | |
3697 | else if (strcmp(path, "/proc/diskstats") == 0) | |
3698 | type = LXC_TYPE_PROC_DISKSTATS; | |
70dcc12e SH |
3699 | else if (strcmp(path, "/proc/swaps") == 0) |
3700 | type = LXC_TYPE_PROC_SWAPS; | |
237e200e SH |
3701 | if (type == -1) |
3702 | return -ENOENT; | |
3703 | ||
3704 | info = malloc(sizeof(*info)); | |
3705 | if (!info) | |
3706 | return -ENOMEM; | |
3707 | ||
3708 | memset(info, 0, sizeof(*info)); | |
3709 | info->type = type; | |
3710 | ||
3711 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
3712 | do { | |
3713 | info->buf = malloc(info->buflen); | |
3714 | } while (!info->buf); | |
3715 | memset(info->buf, 0, info->buflen); | |
3716 | /* set actual size to buffer size */ | |
3717 | info->size = info->buflen; | |
3718 | ||
3719 | fi->fh = (unsigned long)info; | |
3720 | return 0; | |
3721 | } | |
3722 | ||
3723 | int proc_release(const char *path, struct fuse_file_info *fi) | |
3724 | { | |
3725 | struct file_info *f = (struct file_info *)fi->fh; | |
3726 | ||
3727 | do_release_file_info(f); | |
3728 | return 0; | |
3729 | } | |
3730 | ||
3731 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
3732 | struct fuse_file_info *fi) | |
3733 | { | |
3734 | struct file_info *f = (struct file_info *) fi->fh; | |
3735 | ||
3736 | switch (f->type) { | |
3737 | case LXC_TYPE_PROC_MEMINFO: | |
3738 | return proc_meminfo_read(buf, size, offset, fi); | |
3739 | case LXC_TYPE_PROC_CPUINFO: | |
3740 | return proc_cpuinfo_read(buf, size, offset, fi); | |
3741 | case LXC_TYPE_PROC_UPTIME: | |
3742 | return proc_uptime_read(buf, size, offset, fi); | |
3743 | case LXC_TYPE_PROC_STAT: | |
3744 | return proc_stat_read(buf, size, offset, fi); | |
3745 | case LXC_TYPE_PROC_DISKSTATS: | |
3746 | return proc_diskstats_read(buf, size, offset, fi); | |
70dcc12e SH |
3747 | case LXC_TYPE_PROC_SWAPS: |
3748 | return proc_swaps_read(buf, size, offset, fi); | |
237e200e SH |
3749 | default: |
3750 | return -EINVAL; | |
3751 | } | |
3752 | } | |
3753 | ||
3754 | static void __attribute__((constructor)) collect_subsystems(void) | |
3755 | { | |
3756 | FILE *f; | |
3757 | char *line = NULL; | |
3758 | size_t len = 0; | |
3759 | ||
3760 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) { | |
3761 | fprintf(stderr, "Error opening /proc/self/cgroup: %s\n", strerror(errno)); | |
3762 | return; | |
3763 | } | |
3764 | while (getline(&line, &len, f) != -1) { | |
3765 | char *p, *p2; | |
3766 | ||
3767 | p = strchr(line, ':'); | |
3768 | if (!p) | |
3769 | goto out; | |
3770 | *(p++) = '\0'; | |
3771 | ||
3772 | p2 = strrchr(p, ':'); | |
3773 | if (!p2) | |
3774 | goto out; | |
3775 | *p2 = '\0'; | |
3776 | ||
3777 | if (!store_hierarchy(line, p)) | |
3778 | goto out; | |
3779 | } | |
3780 | ||
3781 | print_subsystems(); | |
3782 | ||
3783 | out: | |
3784 | free(line); | |
3785 | fclose(f); | |
3786 | } | |
3787 | ||
3788 | static void __attribute__((destructor)) free_subsystems(void) | |
3789 | { | |
3790 | int i; | |
3791 | ||
3792 | for (i = 0; i < num_hierarchies; i++) | |
3793 | if (hierarchies[i]) | |
3794 | free(hierarchies[i]); | |
3795 | free(hierarchies); | |
3796 | } |