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