]>
Commit | Line | Data |
---|---|---|
ccb4cabe SH |
1 | /* |
2 | * lxc: linux Container library | |
3 | * | |
4 | * Copyright © 2016 Canonical Ltd. | |
5 | * | |
6 | * Authors: | |
7 | * Serge Hallyn <serge.hallyn@ubuntu.com> | |
8 | * | |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, write to the Free Software | |
21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
22 | */ | |
23 | ||
24 | /* | |
25 | * cgfs-ng.c: this is a new, simplified implementation of a filesystem | |
26 | * cgroup backend. The original cgfs.c was designed to be as flexible | |
27 | * as possible. It would try to find cgroup filesystems no matter where | |
28 | * or how you had them mounted, and deduce the most usable mount for | |
29 | * each controller. It also was not designed for unprivileged use, as | |
30 | * that was reserved for cgmanager. | |
31 | * | |
32 | * This new implementation assumes that cgroup filesystems are mounted | |
33 | * under /sys/fs/cgroup/clist where clist is either the controller, or | |
34 | * a comman-separated list of controllers. | |
35 | */ | |
a54694f8 | 36 | |
ccb4cabe | 37 | #include "config.h" |
a54694f8 CB |
38 | |
39 | #include <ctype.h> | |
40 | #include <dirent.h> | |
41 | #include <errno.h> | |
42 | #include <grp.h> | |
43 | #include <stdint.h> | |
ccb4cabe SH |
44 | #include <stdio.h> |
45 | #include <stdlib.h> | |
a54694f8 | 46 | #include <string.h> |
ccb4cabe | 47 | #include <unistd.h> |
a54694f8 | 48 | #include <sys/types.h> |
ccb4cabe | 49 | |
d8e48992 | 50 | #include "bdev.h" |
ccb4cabe | 51 | #include "cgroup.h" |
ccb4cabe | 52 | #include "commands.h" |
a54694f8 CB |
53 | #include "log.h" |
54 | #include "utils.h" | |
ccb4cabe SH |
55 | |
56 | lxc_log_define(lxc_cgfsng, lxc); | |
57 | ||
58 | static struct cgroup_ops cgfsng_ops; | |
59 | ||
ccb4cabe SH |
60 | /* |
61 | * A descriptor for a mounted hierarchy | |
62 | * @controllers: either NULL, or a null-terminated list of all | |
63 | * the co-mounted controllers | |
64 | * @mountpoint: the mountpoint we will use. It will be either | |
65 | * /sys/fs/cgroup/controller or /sys/fs/cgroup/controllerlist | |
66 | * @base_cgroup: the cgroup under which the container cgroup path | |
67 | is created. This will be either the caller's cgroup (if not | |
68 | root), or init's cgroup (if root). | |
69 | */ | |
70 | struct hierarchy { | |
71 | char **controllers; | |
72 | char *mountpoint; | |
73 | char *base_cgroup; | |
74 | char *fullcgpath; | |
75 | }; | |
76 | ||
77 | /* | |
78 | * The cgroup data which is attached to the lxc_handler. | |
ccb4cabe SH |
79 | * @cgroup_pattern - a copy of the lxc.cgroup.pattern |
80 | * @container_cgroup - if not null, the cgroup which was created for | |
81 | * the container. For each hierarchy, it is created under the | |
82 | * @hierarchy->base_cgroup directory. Relative to the base_cgroup | |
83 | * it is the same for all hierarchies. | |
84 | * @name - the container name | |
85 | */ | |
86 | struct cgfsng_handler_data { | |
ccb4cabe SH |
87 | char *cgroup_pattern; |
88 | char *container_cgroup; // cgroup we created for the container | |
89 | char *name; // container name | |
90 | }; | |
91 | ||
457ca9aa SH |
92 | /* |
93 | * @hierarchies - a NULL-terminated array of struct hierarchy, one per | |
94 | * hierarchy. No duplicates. First sufficient, writeable mounted | |
95 | * hierarchy wins | |
96 | */ | |
97 | struct hierarchy **hierarchies; | |
98 | ||
99 | /* | |
100 | * @cgroup_use - a copy of the lxc.cgroup.use | |
101 | */ | |
102 | char *cgroup_use; | |
103 | ||
e4aeecf5 CB |
104 | /* |
105 | * @lxc_cgfsng_debug - whether to print debug info to stdout for the cgfsng | |
106 | * driver | |
107 | */ | |
108 | static bool lxc_cgfsng_debug; | |
109 | ||
ccb4cabe SH |
110 | static void free_string_list(char **clist) |
111 | { | |
112 | if (clist) { | |
113 | int i; | |
114 | ||
115 | for (i = 0; clist[i]; i++) | |
116 | free(clist[i]); | |
117 | free(clist); | |
118 | } | |
119 | } | |
120 | ||
ccb4cabe SH |
121 | /* Allocate a pointer, do not fail */ |
122 | static void *must_alloc(size_t sz) | |
123 | { | |
124 | return must_realloc(NULL, sz); | |
125 | } | |
126 | ||
ccb4cabe SH |
127 | /* |
128 | * This is a special case - return a copy of @entry | |
129 | * prepending 'name='. I.e. turn systemd into name=systemd. | |
130 | * Do not fail. | |
131 | */ | |
132 | static char *must_prefix_named(char *entry) | |
133 | { | |
134 | char *ret; | |
135 | size_t len = strlen(entry); | |
136 | ||
137 | ret = must_alloc(len + 6); | |
138 | snprintf(ret, len + 6, "name=%s", entry); | |
139 | return ret; | |
140 | } | |
141 | ||
142 | /* | |
143 | * Given a pointer to a null-terminated array of pointers, realloc to | |
144 | * add one entry, and point the new entry to NULL. Do not fail. Return | |
145 | * the index to the second-to-last entry - that is, the one which is | |
146 | * now available for use (keeping the list null-terminated). | |
147 | */ | |
148 | static int append_null_to_list(void ***list) | |
149 | { | |
150 | int newentry = 0; | |
151 | ||
152 | if (*list) | |
153 | for (; (*list)[newentry]; newentry++); | |
154 | ||
155 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
156 | (*list)[newentry + 1] = NULL; | |
157 | return newentry; | |
158 | } | |
159 | ||
160 | /* | |
161 | * Given a null-terminated array of strings, check whether @entry | |
162 | * is one of the strings | |
163 | */ | |
164 | static bool string_in_list(char **list, const char *entry) | |
165 | { | |
166 | int i; | |
167 | ||
168 | if (!list) | |
169 | return false; | |
170 | for (i = 0; list[i]; i++) | |
171 | if (strcmp(list[i], entry) == 0) | |
172 | return true; | |
173 | ||
174 | return false; | |
175 | } | |
176 | ||
177 | /* | |
178 | * append an entry to the clist. Do not fail. | |
179 | * *clist must be NULL the first time we are called. | |
180 | * | |
181 | * We also handle named subsystems here. Any controller which is not a | |
182 | * kernel subsystem, we prefix 'name='. Any which is both a kernel and | |
183 | * named subsystem, we refuse to use because we're not sure which we | |
184 | * have here. (TODO - we could work around this in some cases by just | |
185 | * remounting to be unambiguous, or by comparing mountpoint contents | |
186 | * with current cgroup) | |
187 | * | |
188 | * The last entry will always be NULL. | |
189 | */ | |
190 | static void must_append_controller(char **klist, char **nlist, char ***clist, char *entry) | |
191 | { | |
192 | int newentry; | |
193 | char *copy; | |
194 | ||
195 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) { | |
196 | ERROR("Refusing to use ambiguous controller '%s'", entry); | |
197 | ERROR("It is both a named and kernel subsystem"); | |
198 | return; | |
199 | } | |
200 | ||
201 | newentry = append_null_to_list((void ***)clist); | |
202 | ||
203 | if (strncmp(entry, "name=", 5) == 0) | |
204 | copy = must_copy_string(entry); | |
205 | else if (string_in_list(klist, entry)) | |
206 | copy = must_copy_string(entry); | |
207 | else | |
208 | copy = must_prefix_named(entry); | |
209 | ||
210 | (*clist)[newentry] = copy; | |
211 | } | |
212 | ||
ccb4cabe SH |
213 | static void free_handler_data(struct cgfsng_handler_data *d) |
214 | { | |
ccb4cabe SH |
215 | free(d->cgroup_pattern); |
216 | free(d->container_cgroup); | |
217 | free(d->name); | |
218 | free(d); | |
219 | } | |
220 | ||
221 | /* | |
222 | * Given a handler's cgroup data, return the struct hierarchy for the | |
223 | * controller @c, or NULL if there is none. | |
224 | */ | |
457ca9aa | 225 | struct hierarchy *get_hierarchy(const char *c) |
ccb4cabe SH |
226 | { |
227 | int i; | |
228 | ||
457ca9aa | 229 | if (!hierarchies) |
ccb4cabe | 230 | return NULL; |
457ca9aa SH |
231 | for (i = 0; hierarchies[i]; i++) { |
232 | if (string_in_list(hierarchies[i]->controllers, c)) | |
233 | return hierarchies[i]; | |
ccb4cabe SH |
234 | } |
235 | return NULL; | |
236 | } | |
237 | ||
a54694f8 CB |
238 | #define BATCH_SIZE 50 |
239 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
240 | { | |
241 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
242 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
243 | ||
244 | if (!*mem || newbatches > oldbatches) { | |
245 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
246 | } | |
247 | } | |
248 | ||
249 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
250 | { | |
251 | size_t full = oldlen + newlen; | |
252 | ||
253 | batch_realloc(dest, oldlen, full + 1); | |
254 | ||
255 | memcpy(*dest + oldlen, new, newlen + 1); | |
256 | } | |
257 | ||
258 | /* Slurp in a whole file */ | |
259 | static char *read_file(char *fnam) | |
260 | { | |
261 | FILE *f; | |
262 | char *line = NULL, *buf = NULL; | |
263 | size_t len = 0, fulllen = 0; | |
264 | int linelen; | |
265 | ||
266 | f = fopen(fnam, "r"); | |
267 | if (!f) | |
268 | return NULL; | |
269 | while ((linelen = getline(&line, &len, f)) != -1) { | |
270 | append_line(&buf, fulllen, line, linelen); | |
271 | fulllen += linelen; | |
272 | } | |
273 | fclose(f); | |
274 | free(line); | |
275 | return buf; | |
276 | } | |
277 | ||
278 | /* Taken over modified from the kernel sources. */ | |
279 | #define NBITS 32 /* bits in uint32_t */ | |
280 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
281 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
282 | ||
283 | static void set_bit(unsigned bit, uint32_t *bitarr) | |
284 | { | |
285 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
286 | } | |
287 | ||
288 | static void clear_bit(unsigned bit, uint32_t *bitarr) | |
289 | { | |
290 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
291 | } | |
292 | ||
293 | static bool is_set(unsigned bit, uint32_t *bitarr) | |
294 | { | |
295 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
296 | } | |
297 | ||
298 | /* Create cpumask from cpulist aka turn: | |
299 | * | |
300 | * 0,2-3 | |
301 | * | |
302 | * into bit array | |
303 | * | |
304 | * 1 0 1 1 | |
305 | */ | |
306 | static uint32_t *lxc_cpumask(char *buf, size_t nbits) | |
307 | { | |
308 | char *token; | |
309 | char *saveptr = NULL; | |
310 | size_t arrlen = BITS_TO_LONGS(nbits); | |
311 | uint32_t *bitarr = calloc(arrlen, sizeof(uint32_t)); | |
312 | if (!bitarr) | |
313 | return NULL; | |
314 | ||
315 | for (; (token = strtok_r(buf, ",", &saveptr)); buf = NULL) { | |
316 | errno = 0; | |
317 | unsigned start = strtoul(token, NULL, 0); | |
318 | unsigned end = start; | |
319 | ||
320 | char *range = strchr(token, '-'); | |
321 | if (range) | |
322 | end = strtoul(range + 1, NULL, 0); | |
323 | if (!(start <= end)) { | |
324 | free(bitarr); | |
325 | return NULL; | |
326 | } | |
327 | ||
328 | if (end >= nbits) { | |
329 | free(bitarr); | |
330 | return NULL; | |
331 | } | |
332 | ||
333 | while (start <= end) | |
334 | set_bit(start++, bitarr); | |
335 | } | |
336 | ||
337 | return bitarr; | |
338 | } | |
339 | ||
a54694f8 CB |
340 | /* Turn cpumask into simple, comma-separated cpulist. */ |
341 | static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
342 | { | |
343 | size_t i; | |
344 | int ret; | |
eab15c1e | 345 | char numstr[LXC_NUMSTRLEN64] = {0}; |
a54694f8 CB |
346 | char **cpulist = NULL; |
347 | ||
348 | for (i = 0; i <= nbits; i++) { | |
349 | if (is_set(i, bitarr)) { | |
eab15c1e CB |
350 | ret = snprintf(numstr, LXC_NUMSTRLEN64, "%zu", i); |
351 | if (ret < 0 || (size_t)ret >= LXC_NUMSTRLEN64) { | |
a54694f8 CB |
352 | lxc_free_array((void **)cpulist, free); |
353 | return NULL; | |
354 | } | |
355 | if (lxc_append_string(&cpulist, numstr) < 0) { | |
356 | lxc_free_array((void **)cpulist, free); | |
357 | return NULL; | |
358 | } | |
359 | } | |
360 | } | |
361 | return lxc_string_join(",", (const char **)cpulist, false); | |
362 | } | |
363 | ||
364 | static ssize_t get_max_cpus(char *cpulist) | |
365 | { | |
366 | char *c1, *c2; | |
367 | char *maxcpus = cpulist; | |
368 | size_t cpus = 0; | |
369 | ||
370 | c1 = strrchr(maxcpus, ','); | |
371 | if (c1) | |
372 | c1++; | |
373 | ||
374 | c2 = strrchr(maxcpus, '-'); | |
375 | if (c2) | |
376 | c2++; | |
377 | ||
378 | if (!c1 && !c2) | |
379 | c1 = maxcpus; | |
380 | else if (c1 > c2) | |
381 | c2 = c1; | |
382 | else if (c1 < c2) | |
383 | c1 = c2; | |
384 | else if (!c1 && c2) // The reverse case is obvs. not needed. | |
385 | c1 = c2; | |
386 | ||
387 | /* If the above logic is correct, c1 should always hold a valid string | |
388 | * here. | |
389 | */ | |
390 | ||
391 | errno = 0; | |
392 | cpus = strtoul(c1, NULL, 0); | |
393 | if (errno != 0) | |
394 | return -1; | |
395 | ||
396 | return cpus; | |
397 | } | |
398 | ||
6f9584d8 | 399 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
a54694f8 CB |
400 | static bool filter_and_set_cpus(char *path, bool am_initialized) |
401 | { | |
402 | char *lastslash, *fpath, oldv; | |
403 | int ret; | |
404 | ssize_t i; | |
405 | ||
406 | ssize_t maxposs = 0, maxisol = 0; | |
407 | char *cpulist = NULL, *posscpus = NULL, *isolcpus = NULL; | |
408 | uint32_t *possmask = NULL, *isolmask = NULL; | |
6f9584d8 | 409 | bool bret = false, flipped_bit = false; |
a54694f8 CB |
410 | |
411 | lastslash = strrchr(path, '/'); | |
412 | if (!lastslash) { // bug... this shouldn't be possible | |
6f9584d8 | 413 | ERROR("Invalid path: %s.", path); |
a54694f8 CB |
414 | return bret; |
415 | } | |
416 | oldv = *lastslash; | |
417 | *lastslash = '\0'; | |
418 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
419 | posscpus = read_file(fpath); | |
6f9584d8 CB |
420 | if (!posscpus) { |
421 | SYSERROR("Could not read file: %s.\n", fpath); | |
422 | goto on_error; | |
423 | } | |
a54694f8 CB |
424 | |
425 | /* Get maximum number of cpus found in possible cpuset. */ | |
426 | maxposs = get_max_cpus(posscpus); | |
427 | if (maxposs < 0) | |
6f9584d8 | 428 | goto on_error; |
a54694f8 | 429 | |
6f9584d8 CB |
430 | if (!file_exists(__ISOL_CPUS)) { |
431 | /* This system doesn't expose isolated cpus. */ | |
432 | DEBUG("Path: "__ISOL_CPUS" to read isolated cpus from does not exist.\n"); | |
65d29cbc CB |
433 | cpulist = posscpus; |
434 | /* No isolated cpus but we weren't already initialized by | |
435 | * someone. We should simply copy the parents cpuset.cpus | |
436 | * values. | |
437 | */ | |
438 | if (!am_initialized) { | |
439 | DEBUG("Copying cpuset of parent cgroup."); | |
440 | goto copy_parent; | |
441 | } | |
442 | /* No isolated cpus but we were already initialized by someone. | |
443 | * Nothing more to do for us. | |
444 | */ | |
6f9584d8 CB |
445 | goto on_success; |
446 | } | |
447 | ||
448 | isolcpus = read_file(__ISOL_CPUS); | |
449 | if (!isolcpus) { | |
450 | SYSERROR("Could not read file "__ISOL_CPUS); | |
451 | goto on_error; | |
452 | } | |
a54694f8 | 453 | if (!isdigit(isolcpus[0])) { |
6f9584d8 | 454 | DEBUG("No isolated cpus detected."); |
a54694f8 CB |
455 | cpulist = posscpus; |
456 | /* No isolated cpus but we weren't already initialized by | |
457 | * someone. We should simply copy the parents cpuset.cpus | |
458 | * values. | |
459 | */ | |
6f9584d8 CB |
460 | if (!am_initialized) { |
461 | DEBUG("Copying cpuset of parent cgroup."); | |
a54694f8 | 462 | goto copy_parent; |
6f9584d8 | 463 | } |
a54694f8 CB |
464 | /* No isolated cpus but we were already initialized by someone. |
465 | * Nothing more to do for us. | |
466 | */ | |
6f9584d8 | 467 | goto on_success; |
a54694f8 CB |
468 | } |
469 | ||
470 | /* Get maximum number of cpus found in isolated cpuset. */ | |
471 | maxisol = get_max_cpus(isolcpus); | |
472 | if (maxisol < 0) | |
6f9584d8 | 473 | goto on_error; |
a54694f8 CB |
474 | |
475 | if (maxposs < maxisol) | |
476 | maxposs = maxisol; | |
477 | maxposs++; | |
478 | ||
479 | possmask = lxc_cpumask(posscpus, maxposs); | |
6f9584d8 CB |
480 | if (!possmask) { |
481 | ERROR("Could not create cpumask for all possible cpus.\n"); | |
482 | goto on_error; | |
483 | } | |
a54694f8 CB |
484 | |
485 | isolmask = lxc_cpumask(isolcpus, maxposs); | |
6f9584d8 CB |
486 | if (!isolmask) { |
487 | ERROR("Could not create cpumask for all isolated cpus.\n"); | |
488 | goto on_error; | |
489 | } | |
a54694f8 CB |
490 | |
491 | for (i = 0; i <= maxposs; i++) { | |
492 | if (is_set(i, isolmask) && is_set(i, possmask)) { | |
6f9584d8 | 493 | flipped_bit = true; |
a54694f8 CB |
494 | clear_bit(i, possmask); |
495 | } | |
496 | } | |
497 | ||
6f9584d8 CB |
498 | if (!flipped_bit) { |
499 | DEBUG("No isolated cpus present in cpuset."); | |
500 | goto on_success; | |
501 | } | |
502 | DEBUG("Removed isolated cpus from cpuset."); | |
503 | ||
a54694f8 | 504 | cpulist = lxc_cpumask_to_cpulist(possmask, maxposs); |
6f9584d8 CB |
505 | if (!cpulist) { |
506 | ERROR("Could not create cpu list.\n"); | |
507 | goto on_error; | |
508 | } | |
a54694f8 CB |
509 | |
510 | copy_parent: | |
511 | *lastslash = oldv; | |
512 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
513 | ret = lxc_write_to_file(fpath, cpulist, strlen(cpulist), false); | |
6f9584d8 CB |
514 | if (ret < 0) { |
515 | SYSERROR("Could not write cpu list to: %s.\n", fpath); | |
516 | goto on_error; | |
517 | } | |
518 | ||
519 | on_success: | |
520 | bret = true; | |
a54694f8 | 521 | |
6f9584d8 | 522 | on_error: |
a54694f8 CB |
523 | free(fpath); |
524 | ||
525 | free(isolcpus); | |
526 | free(isolmask); | |
527 | ||
528 | if (posscpus != cpulist) | |
529 | free(posscpus); | |
530 | free(possmask); | |
531 | ||
532 | free(cpulist); | |
533 | return bret; | |
534 | } | |
535 | ||
e3a3fecf SH |
536 | /* Copy contents of parent(@path)/@file to @path/@file */ |
537 | static bool copy_parent_file(char *path, char *file) | |
538 | { | |
539 | char *lastslash, *value = NULL, *fpath, oldv; | |
540 | int len = 0; | |
541 | int ret; | |
542 | ||
543 | lastslash = strrchr(path, '/'); | |
544 | if (!lastslash) { // bug... this shouldn't be possible | |
545 | ERROR("cgfsng:copy_parent_file: bad path %s", path); | |
546 | return false; | |
547 | } | |
548 | oldv = *lastslash; | |
549 | *lastslash = '\0'; | |
550 | fpath = must_make_path(path, file, NULL); | |
551 | len = lxc_read_from_file(fpath, NULL, 0); | |
552 | if (len <= 0) | |
553 | goto bad; | |
554 | value = must_alloc(len + 1); | |
555 | if (lxc_read_from_file(fpath, value, len) != len) | |
556 | goto bad; | |
557 | free(fpath); | |
558 | *lastslash = oldv; | |
559 | fpath = must_make_path(path, file, NULL); | |
560 | ret = lxc_write_to_file(fpath, value, len, false); | |
561 | if (ret < 0) | |
562 | SYSERROR("Unable to write %s to %s", value, fpath); | |
563 | free(fpath); | |
564 | free(value); | |
565 | return ret >= 0; | |
566 | ||
567 | bad: | |
568 | SYSERROR("Error reading '%s'", fpath); | |
569 | free(fpath); | |
570 | free(value); | |
571 | return false; | |
572 | } | |
573 | ||
574 | /* | |
575 | * Initialize the cpuset hierarchy in first directory of @gname and | |
576 | * set cgroup.clone_children so that children inherit settings. | |
577 | * Since the h->base_path is populated by init or ourselves, we know | |
578 | * it is already initialized. | |
579 | */ | |
a54694f8 | 580 | static bool handle_cpuset_hierarchy(struct hierarchy *h, char *cgname) |
e3a3fecf SH |
581 | { |
582 | char *cgpath, *clonechildrenpath, v, *slash; | |
583 | ||
584 | if (!string_in_list(h->controllers, "cpuset")) | |
585 | return true; | |
586 | ||
587 | if (*cgname == '/') | |
588 | cgname++; | |
589 | slash = strchr(cgname, '/'); | |
590 | if (slash) | |
591 | *slash = '\0'; | |
592 | ||
593 | cgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); | |
594 | if (slash) | |
595 | *slash = '/'; | |
596 | if (mkdir(cgpath, 0755) < 0 && errno != EEXIST) { | |
597 | SYSERROR("Failed to create '%s'", cgpath); | |
598 | free(cgpath); | |
599 | return false; | |
600 | } | |
6f9584d8 | 601 | |
e3a3fecf SH |
602 | clonechildrenpath = must_make_path(cgpath, "cgroup.clone_children", NULL); |
603 | if (!file_exists(clonechildrenpath)) { /* unified hierarchy doesn't have clone_children */ | |
604 | free(clonechildrenpath); | |
605 | free(cgpath); | |
606 | return true; | |
607 | } | |
608 | if (lxc_read_from_file(clonechildrenpath, &v, 1) < 0) { | |
609 | SYSERROR("Failed to read '%s'", clonechildrenpath); | |
610 | free(clonechildrenpath); | |
611 | free(cgpath); | |
612 | return false; | |
613 | } | |
614 | ||
a54694f8 | 615 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ |
6f9584d8 CB |
616 | if (!filter_and_set_cpus(cgpath, v == '1')) { |
617 | SYSERROR("Failed to remove isolated cpus."); | |
618 | free(clonechildrenpath); | |
619 | free(cgpath); | |
a54694f8 | 620 | return false; |
6f9584d8 | 621 | } |
a54694f8 | 622 | |
e3a3fecf | 623 | if (v == '1') { /* already set for us by someone else */ |
6f9584d8 | 624 | DEBUG("\"cgroup.clone_children\" was already set to \"1\"."); |
e3a3fecf SH |
625 | free(clonechildrenpath); |
626 | free(cgpath); | |
627 | return true; | |
628 | } | |
629 | ||
630 | /* copy parent's settings */ | |
a54694f8 | 631 | if (!copy_parent_file(cgpath, "cpuset.mems")) { |
6f9584d8 | 632 | SYSERROR("Failed to copy \"cpuset.mems\" settings."); |
e3a3fecf SH |
633 | free(cgpath); |
634 | free(clonechildrenpath); | |
635 | return false; | |
636 | } | |
637 | free(cgpath); | |
638 | ||
639 | if (lxc_write_to_file(clonechildrenpath, "1", 1, false) < 0) { | |
640 | /* Set clone_children so children inherit our settings */ | |
641 | SYSERROR("Failed to write 1 to %s", clonechildrenpath); | |
642 | free(clonechildrenpath); | |
643 | return false; | |
644 | } | |
645 | free(clonechildrenpath); | |
646 | return true; | |
647 | } | |
648 | ||
ccb4cabe SH |
649 | /* |
650 | * Given two null-terminated lists of strings, return true if any string | |
651 | * is in both. | |
652 | */ | |
653 | static bool controller_lists_intersect(char **l1, char **l2) | |
654 | { | |
655 | int i; | |
656 | ||
657 | if (!l1 || !l2) | |
658 | return false; | |
659 | ||
660 | for (i = 0; l1[i]; i++) { | |
661 | if (string_in_list(l2, l1[i])) | |
662 | return true; | |
663 | } | |
664 | return false; | |
665 | } | |
666 | ||
667 | /* | |
668 | * For a null-terminated list of controllers @clist, return true if any of | |
669 | * those controllers is already listed the null-terminated list of | |
670 | * hierarchies @hlist. Realistically, if one is present, all must be present. | |
671 | */ | |
672 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
673 | { | |
674 | int i; | |
675 | ||
676 | if (!hlist) | |
677 | return false; | |
678 | for (i = 0; hlist[i]; i++) | |
679 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
680 | return true; | |
681 | return false; | |
682 | ||
683 | } | |
684 | ||
685 | /* | |
686 | * Return true if the controller @entry is found in the null-terminated | |
687 | * list of hierarchies @hlist | |
688 | */ | |
689 | static bool controller_found(struct hierarchy **hlist, char *entry) | |
690 | { | |
691 | int i; | |
692 | if (!hlist) | |
693 | return false; | |
694 | ||
695 | for (i = 0; hlist[i]; i++) | |
696 | if (string_in_list(hlist[i]->controllers, entry)) | |
697 | return true; | |
698 | return false; | |
699 | } | |
700 | ||
701 | /* | |
c30b61c3 SH |
702 | * Return true if all of the controllers which we require have been found. |
703 | * The required list is freezer and anything in * lxc.cgroup.use. | |
ccb4cabe | 704 | */ |
457ca9aa | 705 | static bool all_controllers_found(void) |
ccb4cabe SH |
706 | { |
707 | char *p, *saveptr = NULL; | |
457ca9aa | 708 | struct hierarchy ** hlist = hierarchies; |
ccb4cabe | 709 | |
ccb4cabe SH |
710 | if (!controller_found(hlist, "freezer")) { |
711 | ERROR("no freezer controller mountpoint found"); | |
712 | return false; | |
713 | } | |
714 | ||
457ca9aa | 715 | if (!cgroup_use) |
ccb4cabe | 716 | return true; |
457ca9aa | 717 | for (p = strtok_r(cgroup_use, ",", &saveptr); p; |
ccb4cabe SH |
718 | p = strtok_r(NULL, ",", &saveptr)) { |
719 | if (!controller_found(hlist, p)) { | |
720 | ERROR("no %s controller mountpoint found", p); | |
721 | return false; | |
722 | } | |
723 | } | |
724 | return true; | |
725 | } | |
726 | ||
727 | /* Return true if the fs type is fuse.lxcfs */ | |
728 | static bool is_lxcfs(const char *line) | |
729 | { | |
730 | char *p = strstr(line, " - "); | |
731 | if (!p) | |
732 | return false; | |
2f62fb00 | 733 | return strncmp(p, " - fuse.lxcfs ", 14) == 0; |
ccb4cabe SH |
734 | } |
735 | ||
736 | /* | |
737 | * Get the controllers from a mountinfo line | |
738 | * There are other ways we could get this info. For lxcfs, field 3 | |
739 | * is /cgroup/controller-list. For cgroupfs, we could parse the mount | |
740 | * options. But we simply assume that the mountpoint must be | |
741 | * /sys/fs/cgroup/controller-list | |
742 | */ | |
743 | static char **get_controllers(char **klist, char **nlist, char *line) | |
744 | { | |
745 | // the fourth field is /sys/fs/cgroup/comma-delimited-controller-list | |
746 | int i; | |
747 | char *p = line, *p2, *tok, *saveptr = NULL; | |
748 | char **aret = NULL; | |
749 | ||
750 | for (i = 0; i < 4; i++) { | |
235f1815 | 751 | p = strchr(p, ' '); |
ccb4cabe SH |
752 | if (!p) |
753 | return NULL; | |
754 | p++; | |
755 | } | |
756 | if (!p) | |
757 | return NULL; | |
758 | /* note - if we change how mountinfo works, then our caller | |
759 | * will need to verify /sys/fs/cgroup/ in this field */ | |
5059aae9 SH |
760 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) { |
761 | INFO("cgfsng: found hierarchy not under /sys/fs/cgroup: \"%s\"", p); | |
ccb4cabe | 762 | return NULL; |
5059aae9 | 763 | } |
ccb4cabe | 764 | p += 15; |
235f1815 | 765 | p2 = strchr(p, ' '); |
ccb4cabe SH |
766 | if (!p2) { |
767 | ERROR("corrupt mountinfo"); | |
768 | return NULL; | |
769 | } | |
770 | *p2 = '\0'; | |
771 | for (tok = strtok_r(p, ",", &saveptr); tok; | |
772 | tok = strtok_r(NULL, ",", &saveptr)) { | |
773 | must_append_controller(klist, nlist, &aret, tok); | |
774 | } | |
775 | ||
776 | return aret; | |
777 | } | |
778 | ||
779 | /* return true if the fstype is cgroup */ | |
780 | static bool is_cgroupfs(char *line) | |
781 | { | |
782 | char *p = strstr(line, " - "); | |
783 | if (!p) | |
784 | return false; | |
2f62fb00 | 785 | return strncmp(p, " - cgroup ", 10) == 0; |
ccb4cabe SH |
786 | } |
787 | ||
788 | /* Add a controller to our list of hierarchies */ | |
457ca9aa | 789 | static void add_controller(char **clist, char *mountpoint, char *base_cgroup) |
ccb4cabe SH |
790 | { |
791 | struct hierarchy *new; | |
792 | int newentry; | |
793 | ||
794 | new = must_alloc(sizeof(*new)); | |
795 | new->controllers = clist; | |
796 | new->mountpoint = mountpoint; | |
797 | new->base_cgroup = base_cgroup; | |
798 | new->fullcgpath = NULL; | |
799 | ||
457ca9aa SH |
800 | newentry = append_null_to_list((void ***)&hierarchies); |
801 | hierarchies[newentry] = new; | |
ccb4cabe SH |
802 | } |
803 | ||
804 | /* | |
805 | * Get a copy of the mountpoint from @line, which is a line from | |
806 | * /proc/self/mountinfo | |
807 | */ | |
808 | static char *get_mountpoint(char *line) | |
809 | { | |
810 | int i; | |
811 | char *p = line, *sret; | |
812 | size_t len; | |
813 | ||
814 | for (i = 0; i < 4; i++) { | |
235f1815 | 815 | p = strchr(p, ' '); |
ccb4cabe SH |
816 | if (!p) |
817 | return NULL; | |
818 | p++; | |
819 | } | |
820 | /* we've already stuck a \0 after the mountpoint */ | |
821 | len = strlen(p); | |
822 | sret = must_alloc(len + 1); | |
823 | memcpy(sret, p, len); | |
824 | sret[len] = '\0'; | |
825 | return sret; | |
826 | } | |
827 | ||
828 | /* | |
829 | * Given a multi-line string, return a null-terminated copy of the | |
830 | * current line. | |
831 | */ | |
832 | static char *copy_to_eol(char *p) | |
833 | { | |
235f1815 | 834 | char *p2 = strchr(p, '\n'), *sret; |
ccb4cabe SH |
835 | size_t len; |
836 | ||
837 | if (!p2) | |
838 | return NULL; | |
839 | ||
840 | len = p2 - p; | |
841 | sret = must_alloc(len + 1); | |
842 | memcpy(sret, p, len); | |
843 | sret[len] = '\0'; | |
844 | return sret; | |
845 | } | |
846 | ||
847 | /* | |
848 | * cgline: pointer to character after the first ':' in a line in a | |
849 | * \n-terminated /proc/self/cgroup file. Check whether * controller c is | |
850 | * present. | |
851 | */ | |
852 | static bool controller_in_clist(char *cgline, char *c) | |
853 | { | |
854 | char *tok, *saveptr = NULL, *eol, *tmp; | |
855 | size_t len; | |
856 | ||
235f1815 | 857 | eol = strchr(cgline, ':'); |
ccb4cabe SH |
858 | if (!eol) |
859 | return false; | |
860 | ||
861 | len = eol - cgline; | |
862 | tmp = alloca(len + 1); | |
863 | memcpy(tmp, cgline, len); | |
864 | tmp[len] = '\0'; | |
865 | ||
866 | for (tok = strtok_r(tmp, ",", &saveptr); tok; | |
867 | tok = strtok_r(NULL, ",", &saveptr)) { | |
868 | if (strcmp(tok, c) == 0) | |
869 | return true; | |
870 | } | |
871 | return false; | |
872 | } | |
873 | ||
874 | /* | |
875 | * @basecginfo is a copy of /proc/$$/cgroup. Return the current | |
876 | * cgroup for @controller | |
877 | */ | |
878 | static char *get_current_cgroup(char *basecginfo, char *controller) | |
879 | { | |
880 | char *p = basecginfo; | |
881 | ||
882 | while (1) { | |
235f1815 | 883 | p = strchr(p, ':'); |
ccb4cabe SH |
884 | if (!p) |
885 | return NULL; | |
886 | p++; | |
887 | if (controller_in_clist(p, controller)) { | |
235f1815 | 888 | p = strchr(p, ':'); |
ccb4cabe SH |
889 | if (!p) |
890 | return NULL; | |
891 | p++; | |
892 | return copy_to_eol(p); | |
893 | } | |
894 | ||
235f1815 | 895 | p = strchr(p, '\n'); |
ccb4cabe SH |
896 | if (!p) |
897 | return NULL; | |
898 | p++; | |
899 | } | |
900 | } | |
901 | ||
ccb4cabe SH |
902 | /* |
903 | * Given a hierarchy @mountpoint and base @path, verify that we can create | |
904 | * directories underneath it. | |
905 | */ | |
906 | static bool test_writeable(char *mountpoint, char *path) | |
907 | { | |
908 | char *fullpath = must_make_path(mountpoint, path, NULL); | |
909 | int ret; | |
910 | ||
911 | ret = access(fullpath, W_OK); | |
912 | free(fullpath); | |
913 | return ret == 0; | |
914 | } | |
915 | ||
916 | static void must_append_string(char ***list, char *entry) | |
917 | { | |
918 | int newentry = append_null_to_list((void ***)list); | |
919 | char *copy; | |
920 | ||
921 | copy = must_copy_string(entry); | |
922 | (*list)[newentry] = copy; | |
923 | } | |
924 | ||
925 | static void get_existing_subsystems(char ***klist, char ***nlist) | |
926 | { | |
927 | FILE *f; | |
928 | char *line = NULL; | |
929 | size_t len = 0; | |
930 | ||
931 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) | |
932 | return; | |
933 | while (getline(&line, &len, f) != -1) { | |
934 | char *p, *p2, *tok, *saveptr = NULL; | |
235f1815 | 935 | p = strchr(line, ':'); |
ccb4cabe SH |
936 | if (!p) |
937 | continue; | |
938 | p++; | |
235f1815 | 939 | p2 = strchr(p, ':'); |
ccb4cabe SH |
940 | if (!p2) |
941 | continue; | |
942 | *p2 = '\0'; | |
ff8d6ee9 CB |
943 | |
944 | /* If we have a mixture between cgroup v1 and cgroup v2 | |
945 | * hierarchies, then /proc/self/cgroup contains entries of the | |
946 | * form: | |
947 | * | |
948 | * 0::/some/path | |
949 | * | |
950 | * We need to skip those. | |
951 | */ | |
952 | if ((p2 - p) == 0) | |
953 | continue; | |
954 | ||
ccb4cabe SH |
955 | for (tok = strtok_r(p, ",", &saveptr); tok; |
956 | tok = strtok_r(NULL, ",", &saveptr)) { | |
957 | if (strncmp(tok, "name=", 5) == 0) | |
958 | must_append_string(nlist, tok); | |
959 | else | |
960 | must_append_string(klist, tok); | |
961 | } | |
962 | } | |
963 | ||
964 | free(line); | |
965 | fclose(f); | |
966 | } | |
967 | ||
968 | static void trim(char *s) | |
969 | { | |
970 | size_t len = strlen(s); | |
2c28d76b | 971 | while ((len > 1) && (s[len - 1] == '\n')) |
ccb4cabe SH |
972 | s[--len] = '\0'; |
973 | } | |
974 | ||
e4aeecf5 CB |
975 | static void lxc_cgfsng_print_handler_data(const struct cgfsng_handler_data *d) |
976 | { | |
977 | printf("Cgroup information:\n"); | |
978 | printf(" container name: %s\n", d->name ? d->name : "(null)"); | |
979 | printf(" lxc.cgroup.use: %s\n", cgroup_use ? cgroup_use : "(null)"); | |
980 | printf(" lxc.cgroup.pattern: %s\n", d->cgroup_pattern ? d->cgroup_pattern : "(null)"); | |
981 | printf(" cgroup: %s\n", d->container_cgroup ? d->container_cgroup : "(null)"); | |
982 | } | |
983 | ||
984 | static void lxc_cgfsng_print_hierarchies() | |
ccb4cabe | 985 | { |
a7b0cc4c | 986 | struct hierarchy **it; |
ccb4cabe | 987 | int i; |
41c33dbe | 988 | |
457ca9aa | 989 | if (!hierarchies) { |
e4aeecf5 | 990 | printf(" No hierarchies found."); |
ccb4cabe SH |
991 | return; |
992 | } | |
e4aeecf5 | 993 | printf(" Hierarchies:\n"); |
a7b0cc4c CB |
994 | for (i = 0, it = hierarchies; it && *it; it++, i++) { |
995 | char **cit; | |
ccb4cabe | 996 | int j; |
e4aeecf5 CB |
997 | printf(" %d: base_cgroup %s\n", i, (*it)->base_cgroup ? (*it)->base_cgroup : "(null)"); |
998 | printf(" mountpoint %s\n", (*it)->mountpoint ? (*it)->mountpoint : "(null)"); | |
999 | printf(" controllers:\n"); | |
a7b0cc4c | 1000 | for (j = 0, cit = (*it)->controllers; cit && *cit; cit++, j++) |
e4aeecf5 | 1001 | printf(" %d: %s\n", j, *cit); |
ccb4cabe SH |
1002 | } |
1003 | } | |
41c33dbe | 1004 | |
e4aeecf5 | 1005 | static void lxc_cgfsng_print_basecg_debuginfo(char *basecginfo, char **klist, char **nlist) |
41c33dbe SH |
1006 | { |
1007 | int k; | |
a7b0cc4c | 1008 | char **it; |
41c33dbe | 1009 | |
a7b0cc4c CB |
1010 | printf("basecginfo is:\n"); |
1011 | printf("%s\n", basecginfo); | |
41c33dbe | 1012 | |
a7b0cc4c CB |
1013 | for (k = 0, it = klist; it && *it; it++, k++) |
1014 | printf("kernel subsystem %d: %s\n", k, *it); | |
1015 | for (k = 0, it = nlist; it && *it; it++, k++) | |
1016 | printf("named subsystem %d: %s\n", k, *it); | |
41c33dbe | 1017 | } |
ccb4cabe | 1018 | |
e4aeecf5 CB |
1019 | static void lxc_cgfsng_print_debuginfo(const struct cgfsng_handler_data *d) |
1020 | { | |
1021 | lxc_cgfsng_print_handler_data(d); | |
1022 | lxc_cgfsng_print_hierarchies(); | |
1023 | } | |
1024 | ||
ccb4cabe SH |
1025 | /* |
1026 | * At startup, parse_hierarchies finds all the info we need about | |
1027 | * cgroup mountpoints and current cgroups, and stores it in @d. | |
1028 | */ | |
457ca9aa | 1029 | static bool parse_hierarchies(void) |
ccb4cabe SH |
1030 | { |
1031 | FILE *f; | |
1032 | char * line = NULL, *basecginfo; | |
1033 | char **klist = NULL, **nlist = NULL; | |
1034 | size_t len = 0; | |
1035 | ||
d30ec4cb SH |
1036 | /* |
1037 | * Root spawned containers escape the current cgroup, so use init's | |
1038 | * cgroups as our base in that case. | |
1039 | */ | |
ccb4cabe SH |
1040 | if (geteuid()) |
1041 | basecginfo = read_file("/proc/self/cgroup"); | |
1042 | else | |
1043 | basecginfo = read_file("/proc/1/cgroup"); | |
1044 | if (!basecginfo) | |
1045 | return false; | |
1046 | ||
1047 | if ((f = fopen("/proc/self/mountinfo", "r")) == NULL) { | |
d3b00a8f | 1048 | SYSERROR("Failed opening /proc/self/mountinfo"); |
ccb4cabe SH |
1049 | return false; |
1050 | } | |
1051 | ||
1052 | get_existing_subsystems(&klist, &nlist); | |
41c33dbe | 1053 | |
e4aeecf5 CB |
1054 | if (lxc_cgfsng_debug) |
1055 | lxc_cgfsng_print_basecg_debuginfo(basecginfo, klist, nlist); | |
ccb4cabe SH |
1056 | |
1057 | /* we support simple cgroup mounts and lxcfs mounts */ | |
1058 | while (getline(&line, &len, f) != -1) { | |
1059 | char **controller_list = NULL; | |
1060 | char *mountpoint, *base_cgroup; | |
1061 | ||
1062 | if (!is_lxcfs(line) && !is_cgroupfs(line)) | |
1063 | continue; | |
1064 | ||
1065 | controller_list = get_controllers(klist, nlist, line); | |
1066 | if (!controller_list) | |
1067 | continue; | |
1068 | ||
457ca9aa | 1069 | if (controller_list_is_dup(hierarchies, controller_list)) { |
ccb4cabe SH |
1070 | free(controller_list); |
1071 | continue; | |
1072 | } | |
1073 | ||
1074 | mountpoint = get_mountpoint(line); | |
1075 | if (!mountpoint) { | |
1076 | ERROR("Error reading mountinfo: bad line '%s'", line); | |
1077 | free_string_list(controller_list); | |
1078 | continue; | |
1079 | } | |
1080 | ||
1081 | base_cgroup = get_current_cgroup(basecginfo, controller_list[0]); | |
1082 | if (!base_cgroup) { | |
1083 | ERROR("Failed to find current cgroup for controller '%s'", controller_list[0]); | |
1084 | free_string_list(controller_list); | |
1085 | free(mountpoint); | |
1086 | continue; | |
1087 | } | |
1088 | trim(base_cgroup); | |
1089 | prune_init_scope(base_cgroup); | |
1090 | if (!test_writeable(mountpoint, base_cgroup)) { | |
1091 | free_string_list(controller_list); | |
1092 | free(mountpoint); | |
1093 | free(base_cgroup); | |
1094 | continue; | |
1095 | } | |
457ca9aa | 1096 | add_controller(controller_list, mountpoint, base_cgroup); |
ccb4cabe SH |
1097 | } |
1098 | ||
1099 | free_string_list(klist); | |
1100 | free_string_list(nlist); | |
1101 | ||
1102 | free(basecginfo); | |
1103 | ||
1104 | fclose(f); | |
1105 | free(line); | |
1106 | ||
e4aeecf5 CB |
1107 | if (lxc_cgfsng_debug) { |
1108 | printf("writeable subsystems:\n"); | |
1109 | lxc_cgfsng_print_hierarchies(); | |
1110 | } | |
1111 | ||
ccb4cabe SH |
1112 | /* verify that all controllers in cgroup.use and all crucial |
1113 | * controllers are accounted for | |
1114 | */ | |
5059aae9 SH |
1115 | if (!all_controllers_found()) { |
1116 | INFO("cgfsng: not all controllers were find, deferring to cgfs driver"); | |
ccb4cabe | 1117 | return false; |
5059aae9 | 1118 | } |
ccb4cabe SH |
1119 | |
1120 | return true; | |
1121 | } | |
1122 | ||
457ca9aa SH |
1123 | static bool collect_hierarchy_info(void) |
1124 | { | |
1125 | const char *tmp; | |
1126 | errno = 0; | |
1127 | tmp = lxc_global_config_value("lxc.cgroup.use"); | |
1128 | if (!cgroup_use && errno != 0) { // lxc.cgroup.use can be NULL | |
1129 | SYSERROR("cgfsng: error reading list of cgroups to use"); | |
1130 | return false; | |
1131 | } | |
1132 | cgroup_use = must_copy_string(tmp); | |
1133 | ||
1134 | return parse_hierarchies(); | |
1135 | } | |
1136 | ||
ccb4cabe SH |
1137 | static void *cgfsng_init(const char *name) |
1138 | { | |
1139 | struct cgfsng_handler_data *d; | |
457ca9aa | 1140 | const char *cgroup_pattern; |
ccb4cabe SH |
1141 | |
1142 | d = must_alloc(sizeof(*d)); | |
1143 | memset(d, 0, sizeof(*d)); | |
1144 | ||
1145 | d->name = must_copy_string(name); | |
1146 | ||
ccb4cabe SH |
1147 | cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); |
1148 | if (!cgroup_pattern) { // lxc.cgroup.pattern is only NULL on error | |
1149 | ERROR("Error getting cgroup pattern"); | |
1150 | goto out_free; | |
1151 | } | |
1152 | d->cgroup_pattern = must_copy_string(cgroup_pattern); | |
1153 | ||
e4aeecf5 CB |
1154 | if (lxc_cgfsng_debug) |
1155 | lxc_cgfsng_print_debuginfo(d); | |
ccb4cabe SH |
1156 | |
1157 | return d; | |
1158 | ||
1159 | out_free: | |
1160 | free_handler_data(d); | |
1161 | return NULL; | |
1162 | } | |
1163 | ||
ccb4cabe SH |
1164 | static int cgroup_rmdir(char *dirname) |
1165 | { | |
74f96976 | 1166 | struct dirent *direntp; |
ccb4cabe SH |
1167 | DIR *dir; |
1168 | int r = 0; | |
1169 | ||
1170 | dir = opendir(dirname); | |
1171 | if (!dir) | |
1172 | return -1; | |
1173 | ||
74f96976 | 1174 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
1175 | struct stat mystat; |
1176 | char *pathname; | |
1177 | ||
1178 | if (!direntp) | |
1179 | break; | |
1180 | ||
1181 | if (!strcmp(direntp->d_name, ".") || | |
1182 | !strcmp(direntp->d_name, "..")) | |
1183 | continue; | |
1184 | ||
1185 | pathname = must_make_path(dirname, direntp->d_name, NULL); | |
1186 | ||
1187 | if (lstat(pathname, &mystat)) { | |
1188 | if (!r) | |
1c9da8da | 1189 | WARN("failed to stat %s", pathname); |
ccb4cabe SH |
1190 | r = -1; |
1191 | goto next; | |
1192 | } | |
1193 | ||
1194 | if (!S_ISDIR(mystat.st_mode)) | |
1195 | goto next; | |
1196 | if (cgroup_rmdir(pathname) < 0) | |
1197 | r = -1; | |
1198 | next: | |
1199 | free(pathname); | |
1200 | } | |
1201 | ||
1202 | if (rmdir(dirname) < 0) { | |
1203 | if (!r) | |
13277ec4 | 1204 | WARN("failed to delete %s: %s", dirname, strerror(errno)); |
ccb4cabe SH |
1205 | r = -1; |
1206 | } | |
1207 | ||
1208 | if (closedir(dir) < 0) { | |
1209 | if (!r) | |
13277ec4 | 1210 | WARN("failed to delete %s: %s", dirname, strerror(errno)); |
ccb4cabe SH |
1211 | r = -1; |
1212 | } | |
1213 | return r; | |
1214 | } | |
1215 | ||
1216 | static int rmdir_wrapper(void *data) | |
1217 | { | |
1218 | char *path = data; | |
1219 | ||
1220 | if (setresgid(0,0,0) < 0) | |
1221 | SYSERROR("Failed to setgid to 0"); | |
1222 | if (setresuid(0,0,0) < 0) | |
1223 | SYSERROR("Failed to setuid to 0"); | |
1224 | if (setgroups(0, NULL) < 0) | |
1225 | SYSERROR("Failed to clear groups"); | |
1226 | ||
1227 | return cgroup_rmdir(path); | |
1228 | } | |
1229 | ||
1230 | void recursive_destroy(char *path, struct lxc_conf *conf) | |
1231 | { | |
1232 | int r; | |
1233 | if (conf && !lxc_list_empty(&conf->id_map)) | |
c9b7c33e | 1234 | r = userns_exec_1(conf, rmdir_wrapper, path, "rmdir_wrapper"); |
ccb4cabe SH |
1235 | else |
1236 | r = cgroup_rmdir(path); | |
1237 | ||
1238 | if (r < 0) | |
1c9da8da | 1239 | ERROR("Error destroying %s", path); |
ccb4cabe SH |
1240 | } |
1241 | ||
1242 | static void cgfsng_destroy(void *hdata, struct lxc_conf *conf) | |
1243 | { | |
1244 | struct cgfsng_handler_data *d = hdata; | |
1245 | ||
1246 | if (!d) | |
1247 | return; | |
1248 | ||
457ca9aa | 1249 | if (d->container_cgroup && hierarchies) { |
ccb4cabe | 1250 | int i; |
457ca9aa SH |
1251 | for (i = 0; hierarchies[i]; i++) { |
1252 | struct hierarchy *h = hierarchies[i]; | |
e2db2a89 | 1253 | if (h->fullcgpath) { |
ccb4cabe SH |
1254 | recursive_destroy(h->fullcgpath, conf); |
1255 | free(h->fullcgpath); | |
1256 | h->fullcgpath = NULL; | |
1257 | } | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | free_handler_data(d); | |
1262 | } | |
1263 | ||
1264 | struct cgroup_ops *cgfsng_ops_init(void) | |
1265 | { | |
e4aeecf5 CB |
1266 | if (getenv("LXC_DEBUG_CGFSNG")) |
1267 | lxc_cgfsng_debug = true; | |
1268 | ||
457ca9aa SH |
1269 | if (!collect_hierarchy_info()) |
1270 | return NULL; | |
e4aeecf5 | 1271 | |
ccb4cabe SH |
1272 | return &cgfsng_ops; |
1273 | } | |
1274 | ||
1275 | static bool create_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1276 | { | |
e3a3fecf | 1277 | h->fullcgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); |
6f9584d8 CB |
1278 | if (dir_exists(h->fullcgpath)) { // it must not already exist |
1279 | ERROR("Path \"%s\" already existed.", h->fullcgpath); | |
d8da679e | 1280 | return false; |
6f9584d8 CB |
1281 | } |
1282 | if (!handle_cpuset_hierarchy(h, cgname)) { | |
1283 | ERROR("Failed to handle cgroupfs v1 cpuset controller."); | |
e3a3fecf | 1284 | return false; |
6f9584d8 | 1285 | } |
e3a3fecf | 1286 | return mkdir_p(h->fullcgpath, 0755) == 0; |
ccb4cabe SH |
1287 | } |
1288 | ||
1289 | static void remove_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1290 | { | |
1291 | if (rmdir(h->fullcgpath) < 0) | |
1292 | SYSERROR("Failed to clean up cgroup %s from failed creation attempt", h->fullcgpath); | |
1293 | free(h->fullcgpath); | |
1294 | h->fullcgpath = NULL; | |
1295 | } | |
1296 | ||
1297 | /* | |
d30ec4cb | 1298 | * Try to create the same cgroup in all hierarchies. |
ccb4cabe SH |
1299 | * Start with cgroup_pattern; next cgroup_pattern-1, -2, ..., -999 |
1300 | */ | |
1301 | static inline bool cgfsng_create(void *hdata) | |
1302 | { | |
1303 | struct cgfsng_handler_data *d = hdata; | |
1304 | char *tmp, *cgname, *offset; | |
1305 | int i, idx = 0; | |
1306 | size_t len; | |
1307 | ||
1308 | if (!d) | |
1309 | return false; | |
1310 | if (d->container_cgroup) { | |
1311 | WARN("cgfsng_create called a second time"); | |
1312 | return false; | |
1313 | } | |
1314 | ||
1315 | tmp = lxc_string_replace("%n", d->name, d->cgroup_pattern); | |
1316 | if (!tmp) { | |
1317 | ERROR("Failed expanding cgroup name pattern"); | |
1318 | return false; | |
1319 | } | |
1320 | len = strlen(tmp) + 5; // leave room for -NNN\0 | |
1321 | cgname = must_alloc(len); | |
1322 | strcpy(cgname, tmp); | |
1323 | free(tmp); | |
1324 | offset = cgname + len - 5; | |
1325 | ||
1326 | again: | |
95adfe93 SH |
1327 | if (idx == 1000) { |
1328 | ERROR("Too many conflicting cgroup names"); | |
ccb4cabe | 1329 | goto out_free; |
95adfe93 | 1330 | } |
ccb4cabe SH |
1331 | if (idx) |
1332 | snprintf(offset, 5, "-%d", idx); | |
457ca9aa SH |
1333 | for (i = 0; hierarchies[i]; i++) { |
1334 | if (!create_path_for_hierarchy(hierarchies[i], cgname)) { | |
ccb4cabe | 1335 | int j; |
457ca9aa SH |
1336 | SYSERROR("Failed to create %s: %s", hierarchies[i]->fullcgpath, strerror(errno)); |
1337 | free(hierarchies[i]->fullcgpath); | |
1338 | hierarchies[i]->fullcgpath = NULL; | |
ccb4cabe | 1339 | for (j = 0; j < i; j++) |
457ca9aa | 1340 | remove_path_for_hierarchy(hierarchies[j], cgname); |
ccb4cabe SH |
1341 | idx++; |
1342 | goto again; | |
1343 | } | |
1344 | } | |
1345 | /* Done */ | |
1346 | d->container_cgroup = cgname; | |
1347 | return true; | |
1348 | ||
1349 | out_free: | |
1350 | free(cgname); | |
1351 | return false; | |
1352 | } | |
1353 | ||
ccb4cabe SH |
1354 | static bool cgfsng_enter(void *hdata, pid_t pid) |
1355 | { | |
ccb4cabe SH |
1356 | char pidstr[25]; |
1357 | int i, len; | |
1358 | ||
1359 | len = snprintf(pidstr, 25, "%d", pid); | |
1360 | if (len < 0 || len > 25) | |
1361 | return false; | |
1362 | ||
457ca9aa SH |
1363 | for (i = 0; hierarchies[i]; i++) { |
1364 | char *fullpath = must_make_path(hierarchies[i]->fullcgpath, | |
ccb4cabe SH |
1365 | "cgroup.procs", NULL); |
1366 | if (lxc_write_to_file(fullpath, pidstr, len, false) != 0) { | |
d3b00a8f | 1367 | SYSERROR("Failed to enter %s", fullpath); |
ccb4cabe SH |
1368 | free(fullpath); |
1369 | return false; | |
1370 | } | |
1371 | free(fullpath); | |
1372 | } | |
1373 | ||
1374 | return true; | |
1375 | } | |
1376 | ||
1377 | struct chown_data { | |
1378 | struct cgfsng_handler_data *d; | |
1379 | uid_t origuid; // target uid in parent namespace | |
1380 | }; | |
1381 | ||
c0888dfe SH |
1382 | /* |
1383 | * chgrp the container cgroups to container group. We leave | |
1384 | * the container owner as cgroup owner. So we must make the | |
1385 | * directories 775 so that the container can create sub-cgroups. | |
43647298 SH |
1386 | * |
1387 | * Also chown the tasks and cgroup.procs files. Those may not | |
1388 | * exist depending on kernel version. | |
c0888dfe | 1389 | */ |
ccb4cabe SH |
1390 | static int chown_cgroup_wrapper(void *data) |
1391 | { | |
1392 | struct chown_data *arg = data; | |
ccb4cabe SH |
1393 | uid_t destuid; |
1394 | int i; | |
1395 | ||
1396 | if (setresgid(0,0,0) < 0) | |
1397 | SYSERROR("Failed to setgid to 0"); | |
1398 | if (setresuid(0,0,0) < 0) | |
1399 | SYSERROR("Failed to setuid to 0"); | |
1400 | if (setgroups(0, NULL) < 0) | |
1401 | SYSERROR("Failed to clear groups"); | |
1402 | ||
1403 | destuid = get_ns_uid(arg->origuid); | |
1404 | ||
457ca9aa SH |
1405 | for (i = 0; hierarchies[i]; i++) { |
1406 | char *fullpath, *path = hierarchies[i]->fullcgpath; | |
43647298 SH |
1407 | |
1408 | if (chown(path, destuid, 0) < 0) { | |
ab8f5424 | 1409 | SYSERROR("Error chowning %s to %d", path, (int) destuid); |
ccb4cabe SH |
1410 | return -1; |
1411 | } | |
c0888dfe | 1412 | |
43647298 | 1413 | if (chmod(path, 0775) < 0) { |
ab8f5424 | 1414 | SYSERROR("Error chmoding %s", path); |
c0888dfe SH |
1415 | return -1; |
1416 | } | |
ccb4cabe | 1417 | |
ab8f5424 SH |
1418 | /* |
1419 | * Failures to chown these are inconvenient but not detrimental | |
1420 | * We leave these owned by the container launcher, so that container | |
1421 | * root can write to the files to attach. We chmod them 664 so that | |
1422 | * container systemd can write to the files (which systemd in wily | |
1423 | * insists on doing) | |
1424 | */ | |
43647298 SH |
1425 | fullpath = must_make_path(path, "tasks", NULL); |
1426 | if (chown(fullpath, destuid, 0) < 0 && errno != ENOENT) | |
13277ec4 | 1427 | WARN("Failed chowning %s to %d: %s", fullpath, (int) destuid, |
1428 | strerror(errno)); | |
ab8f5424 | 1429 | if (chmod(fullpath, 0664) < 0) |
13277ec4 | 1430 | WARN("Error chmoding %s: %s", path, strerror(errno)); |
43647298 SH |
1431 | free(fullpath); |
1432 | ||
1433 | fullpath = must_make_path(path, "cgroup.procs", NULL); | |
1434 | if (chown(fullpath, destuid, 0) < 0 && errno != ENOENT) | |
13277ec4 | 1435 | WARN("Failed chowning %s to %d: %s", fullpath, (int) destuid, |
1436 | strerror(errno)); | |
ab8f5424 | 1437 | if (chmod(fullpath, 0664) < 0) |
13277ec4 | 1438 | WARN("Error chmoding %s: %s", path, strerror(errno)); |
ccb4cabe SH |
1439 | free(fullpath); |
1440 | } | |
1441 | ||
1442 | return 0; | |
1443 | } | |
1444 | ||
1445 | static bool cgfsns_chown(void *hdata, struct lxc_conf *conf) | |
1446 | { | |
1447 | struct cgfsng_handler_data *d = hdata; | |
1448 | struct chown_data wrap; | |
1449 | ||
1450 | if (!d) | |
1451 | return false; | |
1452 | ||
1453 | if (lxc_list_empty(&conf->id_map)) | |
1454 | return true; | |
1455 | ||
1456 | wrap.d = d; | |
1457 | wrap.origuid = geteuid(); | |
1458 | ||
c9b7c33e CB |
1459 | if (userns_exec_1(conf, chown_cgroup_wrapper, &wrap, |
1460 | "chown_cgroup_wrapper") < 0) { | |
ccb4cabe SH |
1461 | ERROR("Error requesting cgroup chown in new namespace"); |
1462 | return false; | |
1463 | } | |
1464 | ||
1465 | return true; | |
1466 | } | |
1467 | ||
8aa1044f SH |
1468 | /* |
1469 | * We've safe-mounted a tmpfs as parent, so we don't need to protect against | |
1470 | * symlinks any more - just use mount | |
1471 | */ | |
1472 | ||
1473 | /* mount cgroup-full if requested */ | |
1474 | static int mount_cgroup_full(int type, struct hierarchy *h, char *dest, | |
1475 | char *container_cgroup) | |
1476 | { | |
1477 | if (type < LXC_AUTO_CGROUP_FULL_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) | |
1478 | return 0; | |
1479 | if (mount(h->mountpoint, dest, "cgroup", MS_BIND, NULL) < 0) { | |
1480 | SYSERROR("Error bind-mounting %s cgroup onto %s", h->mountpoint, | |
1481 | dest); | |
1482 | return -1; | |
1483 | } | |
1484 | if (type != LXC_AUTO_CGROUP_FULL_RW) { | |
5b6f9369 SH |
1485 | unsigned long flags = MS_BIND | MS_NOSUID | MS_NOEXEC | MS_NODEV | |
1486 | MS_REMOUNT | MS_RDONLY; | |
1487 | if (mount(NULL, dest, "cgroup", flags, NULL) < 0) { | |
8aa1044f SH |
1488 | SYSERROR("Error remounting %s readonly", dest); |
1489 | return -1; | |
1490 | } | |
1491 | } | |
1492 | ||
1493 | INFO("Bind mounted %s onto %s", h->mountpoint, dest); | |
1494 | if (type != LXC_AUTO_CGROUP_FULL_MIXED) | |
1495 | return 0; | |
1496 | ||
1497 | /* mount just the container path rw */ | |
1498 | char *source = must_make_path(h->mountpoint, h->base_cgroup, container_cgroup, NULL); | |
5b6f9369 | 1499 | char *rwpath = must_make_path(dest, h->base_cgroup, container_cgroup, NULL); |
8aa1044f | 1500 | if (mount(source, rwpath, "cgroup", MS_BIND, NULL) < 0) |
13277ec4 | 1501 | WARN("Failed to mount %s read-write: %s", rwpath, |
1502 | strerror(errno)); | |
8aa1044f SH |
1503 | INFO("Made %s read-write", rwpath); |
1504 | free(rwpath); | |
1505 | free(source); | |
1506 | return 0; | |
1507 | } | |
1508 | ||
1509 | /* cgroup-full:* is done, no need to create subdirs */ | |
1510 | static bool cg_mount_needs_subdirs(int type) | |
1511 | { | |
1512 | if (type >= LXC_AUTO_CGROUP_FULL_RO) | |
1513 | return false; | |
1514 | return true; | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * After $rootfs/sys/fs/container/controller/the/cg/path has been | |
1519 | * created, remount controller ro if needed and bindmount the | |
1520 | * cgroupfs onto controll/the/cg/path | |
1521 | */ | |
1522 | static int | |
1523 | do_secondstage_mounts_if_needed(int type, struct hierarchy *h, | |
1524 | char *controllerpath, char *cgpath, | |
1525 | const char *container_cgroup) | |
1526 | { | |
1527 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_MIXED) { | |
1528 | if (mount(controllerpath, controllerpath, "cgroup", MS_BIND, NULL) < 0) { | |
1529 | SYSERROR("Error bind-mounting %s", controllerpath); | |
1530 | return -1; | |
1531 | } | |
1532 | if (mount(controllerpath, controllerpath, "cgroup", | |
1533 | MS_REMOUNT | MS_BIND | MS_RDONLY, NULL) < 0) { | |
1534 | SYSERROR("Error remounting %s read-only", controllerpath); | |
1535 | return -1; | |
1536 | } | |
1537 | INFO("Remounted %s read-only", controllerpath); | |
1538 | } | |
1539 | char *sourcepath = must_make_path(h->mountpoint, h->base_cgroup, container_cgroup, NULL); | |
1540 | int flags = MS_BIND; | |
1541 | if (type == LXC_AUTO_CGROUP_RO) | |
1542 | flags |= MS_RDONLY; | |
1543 | INFO("Mounting %s onto %s", sourcepath, cgpath); | |
1544 | if (mount(sourcepath, cgpath, "cgroup", flags, NULL) < 0) { | |
1545 | free(sourcepath); | |
1546 | SYSERROR("Error mounting cgroup %s onto %s", h->controllers[0], | |
1547 | cgpath); | |
1548 | return -1; | |
1549 | } | |
1550 | free(sourcepath); | |
1551 | INFO("Completed second stage cgroup automounts for %s", cgpath); | |
1552 | return 0; | |
1553 | } | |
1554 | ||
ccb4cabe SH |
1555 | static bool cgfsng_mount(void *hdata, const char *root, int type) |
1556 | { | |
8aa1044f SH |
1557 | struct cgfsng_handler_data *d = hdata; |
1558 | char *tmpfspath = NULL; | |
1559 | bool retval = false; | |
a8de4c49 | 1560 | int i; |
8aa1044f SH |
1561 | |
1562 | if ((type & LXC_AUTO_CGROUP_MASK) == 0) | |
1563 | return true; | |
1564 | ||
ccb4cabe SH |
1565 | if (cgns_supported()) |
1566 | return true; | |
8aa1044f SH |
1567 | |
1568 | tmpfspath = must_make_path(root, "/sys/fs/cgroup", NULL); | |
1569 | ||
1570 | if (type == LXC_AUTO_CGROUP_NOSPEC) | |
1571 | type = LXC_AUTO_CGROUP_MIXED; | |
1572 | else if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
1573 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
1574 | ||
1575 | /* Mount tmpfs */ | |
1576 | if (safe_mount("cgroup_root", tmpfspath, "tmpfs", | |
1577 | MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_RELATIME, | |
1578 | "size=10240k,mode=755", | |
1579 | root) < 0) | |
1580 | goto bad; | |
1581 | ||
457ca9aa | 1582 | for (i = 0; hierarchies[i]; i++) { |
8aa1044f | 1583 | char *controllerpath, *path2; |
457ca9aa | 1584 | struct hierarchy *h = hierarchies[i]; |
8aa1044f SH |
1585 | char *controller = strrchr(h->mountpoint, '/'); |
1586 | int r; | |
1587 | ||
1588 | if (!controller) | |
1589 | continue; | |
1590 | controller++; | |
1591 | controllerpath = must_make_path(tmpfspath, controller, NULL); | |
1592 | if (dir_exists(controllerpath)) { | |
1593 | free(controllerpath); | |
1594 | continue; | |
1595 | } | |
1596 | if (mkdir(controllerpath, 0755) < 0) { | |
1597 | SYSERROR("Error creating cgroup path: %s", controllerpath); | |
1598 | free(controllerpath); | |
1599 | goto bad; | |
1600 | } | |
1601 | if (mount_cgroup_full(type, h, controllerpath, d->container_cgroup) < 0) { | |
1602 | free(controllerpath); | |
1603 | goto bad; | |
1604 | } | |
1605 | if (!cg_mount_needs_subdirs(type)) { | |
1606 | free(controllerpath); | |
1607 | continue; | |
1608 | } | |
ef4413fa | 1609 | path2 = must_make_path(controllerpath, h->base_cgroup, d->container_cgroup, NULL); |
8aa1044f SH |
1610 | if (mkdir_p(path2, 0755) < 0) { |
1611 | free(controllerpath); | |
1612 | goto bad; | |
1613 | } | |
2f62fb00 | 1614 | |
8aa1044f SH |
1615 | r = do_secondstage_mounts_if_needed(type, h, controllerpath, path2, |
1616 | d->container_cgroup); | |
1617 | free(controllerpath); | |
1618 | free(path2); | |
1619 | if (r < 0) | |
1620 | goto bad; | |
1621 | } | |
1622 | retval = true; | |
1623 | ||
1624 | bad: | |
1625 | free(tmpfspath); | |
1626 | return retval; | |
ccb4cabe SH |
1627 | } |
1628 | ||
1629 | static int recursive_count_nrtasks(char *dirname) | |
1630 | { | |
74f96976 | 1631 | struct dirent *direntp; |
ccb4cabe SH |
1632 | DIR *dir; |
1633 | int count = 0, ret; | |
1634 | char *path; | |
1635 | ||
1636 | dir = opendir(dirname); | |
1637 | if (!dir) | |
1638 | return 0; | |
1639 | ||
74f96976 | 1640 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
1641 | struct stat mystat; |
1642 | ||
1643 | if (!direntp) | |
1644 | break; | |
1645 | ||
1646 | if (!strcmp(direntp->d_name, ".") || | |
1647 | !strcmp(direntp->d_name, "..")) | |
1648 | continue; | |
1649 | ||
1650 | path = must_make_path(dirname, direntp->d_name, NULL); | |
1651 | ||
1652 | if (lstat(path, &mystat)) | |
1653 | goto next; | |
1654 | ||
1655 | if (!S_ISDIR(mystat.st_mode)) | |
1656 | goto next; | |
1657 | ||
1658 | count += recursive_count_nrtasks(path); | |
1659 | next: | |
1660 | free(path); | |
1661 | } | |
1662 | ||
1663 | path = must_make_path(dirname, "cgroup.procs", NULL); | |
1664 | ret = lxc_count_file_lines(path); | |
1665 | if (ret != -1) | |
1666 | count += ret; | |
1667 | free(path); | |
1668 | ||
1669 | (void) closedir(dir); | |
1670 | ||
1671 | return count; | |
1672 | } | |
1673 | ||
1674 | static int cgfsng_nrtasks(void *hdata) { | |
1675 | struct cgfsng_handler_data *d = hdata; | |
1676 | char *path; | |
1677 | int count; | |
1678 | ||
457ca9aa | 1679 | if (!d || !d->container_cgroup || !hierarchies) |
ccb4cabe | 1680 | return -1; |
457ca9aa | 1681 | path = must_make_path(hierarchies[0]->fullcgpath, NULL); |
ccb4cabe SH |
1682 | count = recursive_count_nrtasks(path); |
1683 | free(path); | |
1684 | return count; | |
1685 | } | |
1686 | ||
1687 | /* Only root needs to escape to the cgroup of its init */ | |
7103fe6f | 1688 | static bool cgfsng_escape() |
ccb4cabe | 1689 | { |
ccb4cabe SH |
1690 | int i; |
1691 | ||
1692 | if (geteuid()) | |
1693 | return true; | |
1694 | ||
457ca9aa SH |
1695 | for (i = 0; hierarchies[i]; i++) { |
1696 | char *fullpath = must_make_path(hierarchies[i]->mountpoint, | |
1697 | hierarchies[i]->base_cgroup, | |
ccb4cabe SH |
1698 | "cgroup.procs", NULL); |
1699 | if (lxc_write_to_file(fullpath, "0", 2, false) != 0) { | |
d3b00a8f | 1700 | SYSERROR("Failed to escape to %s", fullpath); |
ccb4cabe | 1701 | free(fullpath); |
6df334d1 | 1702 | return false; |
ccb4cabe SH |
1703 | } |
1704 | free(fullpath); | |
1705 | } | |
1706 | ||
6df334d1 | 1707 | return true; |
ccb4cabe SH |
1708 | } |
1709 | ||
36662416 TA |
1710 | static int cgfsng_num_hierarchies(void) |
1711 | { | |
1712 | int i; | |
1713 | ||
1714 | for (i = 0; hierarchies[i]; i++) | |
1715 | ; | |
1716 | ||
1717 | return i; | |
1718 | } | |
1719 | ||
1720 | static bool cgfsng_get_hierarchies(int n, char ***out) | |
1721 | { | |
1722 | int i; | |
1723 | ||
1724 | /* sanity check n */ | |
1725 | for (i = 0; i < n; i++) { | |
1726 | if (!hierarchies[i]) | |
1727 | return false; | |
1728 | } | |
1729 | ||
1730 | *out = hierarchies[i]->controllers; | |
1731 | ||
1732 | return true; | |
1733 | } | |
1734 | ||
ccb4cabe SH |
1735 | #define THAWED "THAWED" |
1736 | #define THAWED_LEN (strlen(THAWED)) | |
1737 | ||
1738 | static bool cgfsng_unfreeze(void *hdata) | |
1739 | { | |
ccb4cabe | 1740 | char *fullpath; |
457ca9aa | 1741 | struct hierarchy *h = get_hierarchy("freezer"); |
ccb4cabe | 1742 | |
457ca9aa | 1743 | if (!h) |
ccb4cabe SH |
1744 | return false; |
1745 | fullpath = must_make_path(h->fullcgpath, "freezer.state", NULL); | |
1746 | if (lxc_write_to_file(fullpath, THAWED, THAWED_LEN, false) != 0) { | |
1747 | free(fullpath); | |
1748 | return false; | |
1749 | } | |
1750 | free(fullpath); | |
1751 | return true; | |
1752 | } | |
1753 | ||
1754 | static const char *cgfsng_get_cgroup(void *hdata, const char *subsystem) | |
1755 | { | |
457ca9aa | 1756 | struct hierarchy *h = get_hierarchy(subsystem); |
ccb4cabe SH |
1757 | if (!h) |
1758 | return NULL; | |
1759 | ||
371f834d SH |
1760 | return h->fullcgpath ? h->fullcgpath + strlen(h->mountpoint) : NULL; |
1761 | } | |
1762 | ||
1763 | /* | |
1764 | * Given a cgroup path returned from lxc_cmd_get_cgroup_path, build a | |
1765 | * full path, which must be freed by the caller. | |
1766 | */ | |
1767 | static char *build_full_cgpath_from_monitorpath(struct hierarchy *h, | |
1768 | const char *inpath, | |
1769 | const char *filename) | |
1770 | { | |
371f834d | 1771 | return must_make_path(h->mountpoint, inpath, filename, NULL); |
ccb4cabe SH |
1772 | } |
1773 | ||
1774 | static bool cgfsng_attach(const char *name, const char *lxcpath, pid_t pid) | |
1775 | { | |
ccb4cabe SH |
1776 | char pidstr[25]; |
1777 | int i, len; | |
1778 | ||
1779 | len = snprintf(pidstr, 25, "%d", pid); | |
1780 | if (len < 0 || len > 25) | |
1781 | return false; | |
1782 | ||
457ca9aa | 1783 | for (i = 0; hierarchies[i]; i++) { |
ccb4cabe | 1784 | char *path, *fullpath; |
457ca9aa | 1785 | struct hierarchy *h = hierarchies[i]; |
ccb4cabe SH |
1786 | |
1787 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->controllers[0]); | |
1788 | if (!path) // not running | |
1789 | continue; | |
1790 | ||
371f834d SH |
1791 | fullpath = build_full_cgpath_from_monitorpath(h, path, "cgroup.procs"); |
1792 | free(path); | |
ccb4cabe SH |
1793 | if (lxc_write_to_file(fullpath, pidstr, len, false) != 0) { |
1794 | SYSERROR("Failed to attach %d to %s", (int)pid, fullpath); | |
1795 | free(fullpath); | |
ccb4cabe SH |
1796 | return false; |
1797 | } | |
ccb4cabe SH |
1798 | free(fullpath); |
1799 | } | |
1800 | ||
ccb4cabe SH |
1801 | return true; |
1802 | } | |
1803 | ||
1804 | /* | |
1805 | * Called externally (i.e. from 'lxc-cgroup') to query cgroup limits. | |
1806 | * Here we don't have a cgroup_data set up, so we ask the running | |
1807 | * container through the commands API for the cgroup path | |
1808 | */ | |
1809 | static int cgfsng_get(const char *filename, char *value, size_t len, const char *name, const char *lxcpath) | |
1810 | { | |
1811 | char *subsystem, *p, *path; | |
ccb4cabe SH |
1812 | struct hierarchy *h; |
1813 | int ret = -1; | |
1814 | ||
1815 | subsystem = alloca(strlen(filename) + 1); | |
1816 | strcpy(subsystem, filename); | |
1817 | if ((p = strchr(subsystem, '.')) != NULL) | |
1818 | *p = '\0'; | |
1819 | ||
1820 | path = lxc_cmd_get_cgroup_path(name, lxcpath, subsystem); | |
1821 | if (!path) // not running | |
1822 | return -1; | |
1823 | ||
457ca9aa | 1824 | h = get_hierarchy(subsystem); |
ccb4cabe | 1825 | if (h) { |
371f834d | 1826 | char *fullpath = build_full_cgpath_from_monitorpath(h, path, filename); |
ccb4cabe SH |
1827 | ret = lxc_read_from_file(fullpath, value, len); |
1828 | free(fullpath); | |
1829 | } | |
1830 | ||
ccb4cabe SH |
1831 | free(path); |
1832 | ||
1833 | return ret; | |
1834 | } | |
1835 | ||
1836 | /* | |
1837 | * Called externally (i.e. from 'lxc-cgroup') to set new cgroup limits. | |
1838 | * Here we don't have a cgroup_data set up, so we ask the running | |
1839 | * container through the commands API for the cgroup path | |
1840 | */ | |
1841 | static int cgfsng_set(const char *filename, const char *value, const char *name, const char *lxcpath) | |
1842 | { | |
1843 | char *subsystem, *p, *path; | |
ccb4cabe SH |
1844 | struct hierarchy *h; |
1845 | int ret = -1; | |
1846 | ||
1847 | subsystem = alloca(strlen(filename) + 1); | |
1848 | strcpy(subsystem, filename); | |
1849 | if ((p = strchr(subsystem, '.')) != NULL) | |
1850 | *p = '\0'; | |
1851 | ||
1852 | path = lxc_cmd_get_cgroup_path(name, lxcpath, subsystem); | |
1853 | if (!path) // not running | |
1854 | return -1; | |
1855 | ||
457ca9aa | 1856 | h = get_hierarchy(subsystem); |
ccb4cabe | 1857 | if (h) { |
371f834d | 1858 | char *fullpath = build_full_cgpath_from_monitorpath(h, path, filename); |
ccb4cabe SH |
1859 | ret = lxc_write_to_file(fullpath, value, strlen(value), false); |
1860 | free(fullpath); | |
1861 | } | |
1862 | ||
ccb4cabe SH |
1863 | free(path); |
1864 | ||
1865 | return ret; | |
1866 | } | |
1867 | ||
ccb4cabe SH |
1868 | /* |
1869 | * Called from setup_limits - here we have the container's cgroup_data because | |
1870 | * we created the cgroups | |
1871 | */ | |
1872 | static int lxc_cgroup_set_data(const char *filename, const char *value, struct cgfsng_handler_data *d) | |
1873 | { | |
1874 | char *subsystem = NULL, *p; | |
1875 | int ret = -1; | |
1876 | struct hierarchy *h; | |
1877 | ||
1878 | subsystem = alloca(strlen(filename) + 1); | |
1879 | strcpy(subsystem, filename); | |
1880 | if ((p = strchr(subsystem, '.')) != NULL) | |
1881 | *p = '\0'; | |
1882 | ||
457ca9aa | 1883 | h = get_hierarchy(subsystem); |
ccb4cabe SH |
1884 | if (h) { |
1885 | char *fullpath = must_make_path(h->fullcgpath, filename, NULL); | |
1886 | ret = lxc_write_to_file(fullpath, value, strlen(value), false); | |
1887 | free(fullpath); | |
1888 | } | |
1889 | return ret; | |
1890 | } | |
1891 | ||
1892 | static bool cgfsng_setup_limits(void *hdata, struct lxc_list *cgroup_settings, | |
1893 | bool do_devices) | |
1894 | { | |
1895 | struct cgfsng_handler_data *d = hdata; | |
1896 | struct lxc_list *iterator, *sorted_cgroup_settings, *next; | |
1897 | struct lxc_cgroup *cg; | |
ccb4cabe SH |
1898 | bool ret = false; |
1899 | ||
1900 | if (lxc_list_empty(cgroup_settings)) | |
1901 | return true; | |
1902 | ||
1903 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); | |
1904 | if (!sorted_cgroup_settings) { | |
1905 | return false; | |
1906 | } | |
1907 | ||
ccb4cabe SH |
1908 | lxc_list_for_each(iterator, sorted_cgroup_settings) { |
1909 | cg = iterator->elem; | |
1910 | ||
1911 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { | |
ccb4cabe SH |
1912 | if (lxc_cgroup_set_data(cg->subsystem, cg->value, d)) { |
1913 | if (do_devices && (errno == EACCES || errno == EPERM)) { | |
1914 | WARN("Error setting %s to %s for %s", | |
1915 | cg->subsystem, cg->value, d->name); | |
1916 | continue; | |
1917 | } | |
1918 | SYSERROR("Error setting %s to %s for %s", | |
1919 | cg->subsystem, cg->value, d->name); | |
1920 | goto out; | |
1921 | } | |
6a628f4a | 1922 | DEBUG("cgroup '%s' set to '%s'", cg->subsystem, cg->value); |
ccb4cabe | 1923 | } |
ccb4cabe SH |
1924 | } |
1925 | ||
1926 | ret = true; | |
1927 | INFO("cgroup has been setup"); | |
1928 | out: | |
ccb4cabe SH |
1929 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
1930 | lxc_list_del(iterator); | |
1931 | free(iterator); | |
1932 | } | |
1933 | free(sorted_cgroup_settings); | |
1934 | return ret; | |
1935 | } | |
1936 | ||
1937 | static struct cgroup_ops cgfsng_ops = { | |
1938 | .init = cgfsng_init, | |
1939 | .destroy = cgfsng_destroy, | |
1940 | .create = cgfsng_create, | |
1941 | .enter = cgfsng_enter, | |
ccb4cabe | 1942 | .escape = cgfsng_escape, |
36662416 TA |
1943 | .num_hierarchies = cgfsng_num_hierarchies, |
1944 | .get_hierarchies = cgfsng_get_hierarchies, | |
ccb4cabe SH |
1945 | .get_cgroup = cgfsng_get_cgroup, |
1946 | .get = cgfsng_get, | |
1947 | .set = cgfsng_set, | |
1948 | .unfreeze = cgfsng_unfreeze, | |
1949 | .setup_limits = cgfsng_setup_limits, | |
1950 | .name = "cgroupfs-ng", | |
1951 | .attach = cgfsng_attach, | |
1952 | .chown = cgfsns_chown, | |
1953 | .mount_cgroup = cgfsng_mount, | |
1954 | .nrtasks = cgfsng_nrtasks, | |
1955 | .driver = CGFSNG, | |
1956 | ||
1957 | /* unsupported */ | |
1958 | .create_legacy = NULL, | |
1959 | }; |