]>
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 | ||
121 | /* Re-alllocate a pointer, do not fail */ | |
122 | static void *must_realloc(void *orig, size_t sz) | |
123 | { | |
124 | void *ret; | |
125 | ||
126 | do { | |
127 | ret = realloc(orig, sz); | |
128 | } while (!ret); | |
129 | return ret; | |
130 | } | |
131 | ||
132 | /* Allocate a pointer, do not fail */ | |
133 | static void *must_alloc(size_t sz) | |
134 | { | |
135 | return must_realloc(NULL, sz); | |
136 | } | |
137 | ||
138 | /* return copy of string @entry; do not fail. */ | |
139 | static char *must_copy_string(const char *entry) | |
140 | { | |
141 | char *ret; | |
142 | ||
143 | if (!entry) | |
144 | return NULL; | |
145 | do { | |
146 | ret = strdup(entry); | |
147 | } while (!ret); | |
148 | return ret; | |
149 | } | |
150 | ||
151 | /* | |
152 | * This is a special case - return a copy of @entry | |
153 | * prepending 'name='. I.e. turn systemd into name=systemd. | |
154 | * Do not fail. | |
155 | */ | |
156 | static char *must_prefix_named(char *entry) | |
157 | { | |
158 | char *ret; | |
159 | size_t len = strlen(entry); | |
160 | ||
161 | ret = must_alloc(len + 6); | |
162 | snprintf(ret, len + 6, "name=%s", entry); | |
163 | return ret; | |
164 | } | |
165 | ||
166 | /* | |
167 | * Given a pointer to a null-terminated array of pointers, realloc to | |
168 | * add one entry, and point the new entry to NULL. Do not fail. Return | |
169 | * the index to the second-to-last entry - that is, the one which is | |
170 | * now available for use (keeping the list null-terminated). | |
171 | */ | |
172 | static int append_null_to_list(void ***list) | |
173 | { | |
174 | int newentry = 0; | |
175 | ||
176 | if (*list) | |
177 | for (; (*list)[newentry]; newentry++); | |
178 | ||
179 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
180 | (*list)[newentry + 1] = NULL; | |
181 | return newentry; | |
182 | } | |
183 | ||
184 | /* | |
185 | * Given a null-terminated array of strings, check whether @entry | |
186 | * is one of the strings | |
187 | */ | |
188 | static bool string_in_list(char **list, const char *entry) | |
189 | { | |
190 | int i; | |
191 | ||
192 | if (!list) | |
193 | return false; | |
194 | for (i = 0; list[i]; i++) | |
195 | if (strcmp(list[i], entry) == 0) | |
196 | return true; | |
197 | ||
198 | return false; | |
199 | } | |
200 | ||
201 | /* | |
202 | * append an entry to the clist. Do not fail. | |
203 | * *clist must be NULL the first time we are called. | |
204 | * | |
205 | * We also handle named subsystems here. Any controller which is not a | |
206 | * kernel subsystem, we prefix 'name='. Any which is both a kernel and | |
207 | * named subsystem, we refuse to use because we're not sure which we | |
208 | * have here. (TODO - we could work around this in some cases by just | |
209 | * remounting to be unambiguous, or by comparing mountpoint contents | |
210 | * with current cgroup) | |
211 | * | |
212 | * The last entry will always be NULL. | |
213 | */ | |
214 | static void must_append_controller(char **klist, char **nlist, char ***clist, char *entry) | |
215 | { | |
216 | int newentry; | |
217 | char *copy; | |
218 | ||
219 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) { | |
220 | ERROR("Refusing to use ambiguous controller '%s'", entry); | |
221 | ERROR("It is both a named and kernel subsystem"); | |
222 | return; | |
223 | } | |
224 | ||
225 | newentry = append_null_to_list((void ***)clist); | |
226 | ||
227 | if (strncmp(entry, "name=", 5) == 0) | |
228 | copy = must_copy_string(entry); | |
229 | else if (string_in_list(klist, entry)) | |
230 | copy = must_copy_string(entry); | |
231 | else | |
232 | copy = must_prefix_named(entry); | |
233 | ||
234 | (*clist)[newentry] = copy; | |
235 | } | |
236 | ||
ccb4cabe SH |
237 | static void free_handler_data(struct cgfsng_handler_data *d) |
238 | { | |
ccb4cabe SH |
239 | free(d->cgroup_pattern); |
240 | free(d->container_cgroup); | |
241 | free(d->name); | |
242 | free(d); | |
243 | } | |
244 | ||
245 | /* | |
246 | * Given a handler's cgroup data, return the struct hierarchy for the | |
247 | * controller @c, or NULL if there is none. | |
248 | */ | |
457ca9aa | 249 | struct hierarchy *get_hierarchy(const char *c) |
ccb4cabe SH |
250 | { |
251 | int i; | |
252 | ||
457ca9aa | 253 | if (!hierarchies) |
ccb4cabe | 254 | return NULL; |
457ca9aa SH |
255 | for (i = 0; hierarchies[i]; i++) { |
256 | if (string_in_list(hierarchies[i]->controllers, c)) | |
257 | return hierarchies[i]; | |
ccb4cabe SH |
258 | } |
259 | return NULL; | |
260 | } | |
261 | ||
e3a3fecf SH |
262 | static char *must_make_path(const char *first, ...) __attribute__((sentinel)); |
263 | ||
a54694f8 CB |
264 | #define BATCH_SIZE 50 |
265 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
266 | { | |
267 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
268 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
269 | ||
270 | if (!*mem || newbatches > oldbatches) { | |
271 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
272 | } | |
273 | } | |
274 | ||
275 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
276 | { | |
277 | size_t full = oldlen + newlen; | |
278 | ||
279 | batch_realloc(dest, oldlen, full + 1); | |
280 | ||
281 | memcpy(*dest + oldlen, new, newlen + 1); | |
282 | } | |
283 | ||
284 | /* Slurp in a whole file */ | |
285 | static char *read_file(char *fnam) | |
286 | { | |
287 | FILE *f; | |
288 | char *line = NULL, *buf = NULL; | |
289 | size_t len = 0, fulllen = 0; | |
290 | int linelen; | |
291 | ||
292 | f = fopen(fnam, "r"); | |
293 | if (!f) | |
294 | return NULL; | |
295 | while ((linelen = getline(&line, &len, f)) != -1) { | |
296 | append_line(&buf, fulllen, line, linelen); | |
297 | fulllen += linelen; | |
298 | } | |
299 | fclose(f); | |
300 | free(line); | |
301 | return buf; | |
302 | } | |
303 | ||
304 | /* Taken over modified from the kernel sources. */ | |
305 | #define NBITS 32 /* bits in uint32_t */ | |
306 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
307 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
308 | ||
309 | static void set_bit(unsigned bit, uint32_t *bitarr) | |
310 | { | |
311 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
312 | } | |
313 | ||
314 | static void clear_bit(unsigned bit, uint32_t *bitarr) | |
315 | { | |
316 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
317 | } | |
318 | ||
319 | static bool is_set(unsigned bit, uint32_t *bitarr) | |
320 | { | |
321 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
322 | } | |
323 | ||
324 | /* Create cpumask from cpulist aka turn: | |
325 | * | |
326 | * 0,2-3 | |
327 | * | |
328 | * into bit array | |
329 | * | |
330 | * 1 0 1 1 | |
331 | */ | |
332 | static uint32_t *lxc_cpumask(char *buf, size_t nbits) | |
333 | { | |
334 | char *token; | |
335 | char *saveptr = NULL; | |
336 | size_t arrlen = BITS_TO_LONGS(nbits); | |
337 | uint32_t *bitarr = calloc(arrlen, sizeof(uint32_t)); | |
338 | if (!bitarr) | |
339 | return NULL; | |
340 | ||
341 | for (; (token = strtok_r(buf, ",", &saveptr)); buf = NULL) { | |
342 | errno = 0; | |
343 | unsigned start = strtoul(token, NULL, 0); | |
344 | unsigned end = start; | |
345 | ||
346 | char *range = strchr(token, '-'); | |
347 | if (range) | |
348 | end = strtoul(range + 1, NULL, 0); | |
349 | if (!(start <= end)) { | |
350 | free(bitarr); | |
351 | return NULL; | |
352 | } | |
353 | ||
354 | if (end >= nbits) { | |
355 | free(bitarr); | |
356 | return NULL; | |
357 | } | |
358 | ||
359 | while (start <= end) | |
360 | set_bit(start++, bitarr); | |
361 | } | |
362 | ||
363 | return bitarr; | |
364 | } | |
365 | ||
a54694f8 CB |
366 | /* Turn cpumask into simple, comma-separated cpulist. */ |
367 | static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
368 | { | |
369 | size_t i; | |
370 | int ret; | |
eab15c1e | 371 | char numstr[LXC_NUMSTRLEN64] = {0}; |
a54694f8 CB |
372 | char **cpulist = NULL; |
373 | ||
374 | for (i = 0; i <= nbits; i++) { | |
375 | if (is_set(i, bitarr)) { | |
eab15c1e CB |
376 | ret = snprintf(numstr, LXC_NUMSTRLEN64, "%zu", i); |
377 | if (ret < 0 || (size_t)ret >= LXC_NUMSTRLEN64) { | |
a54694f8 CB |
378 | lxc_free_array((void **)cpulist, free); |
379 | return NULL; | |
380 | } | |
381 | if (lxc_append_string(&cpulist, numstr) < 0) { | |
382 | lxc_free_array((void **)cpulist, free); | |
383 | return NULL; | |
384 | } | |
385 | } | |
386 | } | |
387 | return lxc_string_join(",", (const char **)cpulist, false); | |
388 | } | |
389 | ||
390 | static ssize_t get_max_cpus(char *cpulist) | |
391 | { | |
392 | char *c1, *c2; | |
393 | char *maxcpus = cpulist; | |
394 | size_t cpus = 0; | |
395 | ||
396 | c1 = strrchr(maxcpus, ','); | |
397 | if (c1) | |
398 | c1++; | |
399 | ||
400 | c2 = strrchr(maxcpus, '-'); | |
401 | if (c2) | |
402 | c2++; | |
403 | ||
404 | if (!c1 && !c2) | |
405 | c1 = maxcpus; | |
406 | else if (c1 > c2) | |
407 | c2 = c1; | |
408 | else if (c1 < c2) | |
409 | c1 = c2; | |
410 | else if (!c1 && c2) // The reverse case is obvs. not needed. | |
411 | c1 = c2; | |
412 | ||
413 | /* If the above logic is correct, c1 should always hold a valid string | |
414 | * here. | |
415 | */ | |
416 | ||
417 | errno = 0; | |
418 | cpus = strtoul(c1, NULL, 0); | |
419 | if (errno != 0) | |
420 | return -1; | |
421 | ||
422 | return cpus; | |
423 | } | |
424 | ||
6f9584d8 | 425 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
a54694f8 CB |
426 | static bool filter_and_set_cpus(char *path, bool am_initialized) |
427 | { | |
428 | char *lastslash, *fpath, oldv; | |
429 | int ret; | |
430 | ssize_t i; | |
431 | ||
432 | ssize_t maxposs = 0, maxisol = 0; | |
433 | char *cpulist = NULL, *posscpus = NULL, *isolcpus = NULL; | |
434 | uint32_t *possmask = NULL, *isolmask = NULL; | |
6f9584d8 | 435 | bool bret = false, flipped_bit = false; |
a54694f8 CB |
436 | |
437 | lastslash = strrchr(path, '/'); | |
438 | if (!lastslash) { // bug... this shouldn't be possible | |
6f9584d8 | 439 | ERROR("Invalid path: %s.", path); |
a54694f8 CB |
440 | return bret; |
441 | } | |
442 | oldv = *lastslash; | |
443 | *lastslash = '\0'; | |
444 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
445 | posscpus = read_file(fpath); | |
6f9584d8 CB |
446 | if (!posscpus) { |
447 | SYSERROR("Could not read file: %s.\n", fpath); | |
448 | goto on_error; | |
449 | } | |
a54694f8 CB |
450 | |
451 | /* Get maximum number of cpus found in possible cpuset. */ | |
452 | maxposs = get_max_cpus(posscpus); | |
453 | if (maxposs < 0) | |
6f9584d8 | 454 | goto on_error; |
a54694f8 | 455 | |
6f9584d8 CB |
456 | if (!file_exists(__ISOL_CPUS)) { |
457 | /* This system doesn't expose isolated cpus. */ | |
458 | DEBUG("Path: "__ISOL_CPUS" to read isolated cpus from does not exist.\n"); | |
65d29cbc CB |
459 | cpulist = posscpus; |
460 | /* No isolated cpus but we weren't already initialized by | |
461 | * someone. We should simply copy the parents cpuset.cpus | |
462 | * values. | |
463 | */ | |
464 | if (!am_initialized) { | |
465 | DEBUG("Copying cpuset of parent cgroup."); | |
466 | goto copy_parent; | |
467 | } | |
468 | /* No isolated cpus but we were already initialized by someone. | |
469 | * Nothing more to do for us. | |
470 | */ | |
6f9584d8 CB |
471 | goto on_success; |
472 | } | |
473 | ||
474 | isolcpus = read_file(__ISOL_CPUS); | |
475 | if (!isolcpus) { | |
476 | SYSERROR("Could not read file "__ISOL_CPUS); | |
477 | goto on_error; | |
478 | } | |
a54694f8 | 479 | if (!isdigit(isolcpus[0])) { |
6f9584d8 | 480 | DEBUG("No isolated cpus detected."); |
a54694f8 CB |
481 | cpulist = posscpus; |
482 | /* No isolated cpus but we weren't already initialized by | |
483 | * someone. We should simply copy the parents cpuset.cpus | |
484 | * values. | |
485 | */ | |
6f9584d8 CB |
486 | if (!am_initialized) { |
487 | DEBUG("Copying cpuset of parent cgroup."); | |
a54694f8 | 488 | goto copy_parent; |
6f9584d8 | 489 | } |
a54694f8 CB |
490 | /* No isolated cpus but we were already initialized by someone. |
491 | * Nothing more to do for us. | |
492 | */ | |
6f9584d8 | 493 | goto on_success; |
a54694f8 CB |
494 | } |
495 | ||
496 | /* Get maximum number of cpus found in isolated cpuset. */ | |
497 | maxisol = get_max_cpus(isolcpus); | |
498 | if (maxisol < 0) | |
6f9584d8 | 499 | goto on_error; |
a54694f8 CB |
500 | |
501 | if (maxposs < maxisol) | |
502 | maxposs = maxisol; | |
503 | maxposs++; | |
504 | ||
505 | possmask = lxc_cpumask(posscpus, maxposs); | |
6f9584d8 CB |
506 | if (!possmask) { |
507 | ERROR("Could not create cpumask for all possible cpus.\n"); | |
508 | goto on_error; | |
509 | } | |
a54694f8 CB |
510 | |
511 | isolmask = lxc_cpumask(isolcpus, maxposs); | |
6f9584d8 CB |
512 | if (!isolmask) { |
513 | ERROR("Could not create cpumask for all isolated cpus.\n"); | |
514 | goto on_error; | |
515 | } | |
a54694f8 CB |
516 | |
517 | for (i = 0; i <= maxposs; i++) { | |
518 | if (is_set(i, isolmask) && is_set(i, possmask)) { | |
6f9584d8 | 519 | flipped_bit = true; |
a54694f8 CB |
520 | clear_bit(i, possmask); |
521 | } | |
522 | } | |
523 | ||
6f9584d8 CB |
524 | if (!flipped_bit) { |
525 | DEBUG("No isolated cpus present in cpuset."); | |
526 | goto on_success; | |
527 | } | |
528 | DEBUG("Removed isolated cpus from cpuset."); | |
529 | ||
a54694f8 | 530 | cpulist = lxc_cpumask_to_cpulist(possmask, maxposs); |
6f9584d8 CB |
531 | if (!cpulist) { |
532 | ERROR("Could not create cpu list.\n"); | |
533 | goto on_error; | |
534 | } | |
a54694f8 CB |
535 | |
536 | copy_parent: | |
537 | *lastslash = oldv; | |
538 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
539 | ret = lxc_write_to_file(fpath, cpulist, strlen(cpulist), false); | |
6f9584d8 CB |
540 | if (ret < 0) { |
541 | SYSERROR("Could not write cpu list to: %s.\n", fpath); | |
542 | goto on_error; | |
543 | } | |
544 | ||
545 | on_success: | |
546 | bret = true; | |
a54694f8 | 547 | |
6f9584d8 | 548 | on_error: |
a54694f8 CB |
549 | free(fpath); |
550 | ||
551 | free(isolcpus); | |
552 | free(isolmask); | |
553 | ||
554 | if (posscpus != cpulist) | |
555 | free(posscpus); | |
556 | free(possmask); | |
557 | ||
558 | free(cpulist); | |
559 | return bret; | |
560 | } | |
561 | ||
e3a3fecf SH |
562 | /* Copy contents of parent(@path)/@file to @path/@file */ |
563 | static bool copy_parent_file(char *path, char *file) | |
564 | { | |
565 | char *lastslash, *value = NULL, *fpath, oldv; | |
566 | int len = 0; | |
567 | int ret; | |
568 | ||
569 | lastslash = strrchr(path, '/'); | |
570 | if (!lastslash) { // bug... this shouldn't be possible | |
571 | ERROR("cgfsng:copy_parent_file: bad path %s", path); | |
572 | return false; | |
573 | } | |
574 | oldv = *lastslash; | |
575 | *lastslash = '\0'; | |
576 | fpath = must_make_path(path, file, NULL); | |
577 | len = lxc_read_from_file(fpath, NULL, 0); | |
578 | if (len <= 0) | |
579 | goto bad; | |
580 | value = must_alloc(len + 1); | |
581 | if (lxc_read_from_file(fpath, value, len) != len) | |
582 | goto bad; | |
583 | free(fpath); | |
584 | *lastslash = oldv; | |
585 | fpath = must_make_path(path, file, NULL); | |
586 | ret = lxc_write_to_file(fpath, value, len, false); | |
587 | if (ret < 0) | |
588 | SYSERROR("Unable to write %s to %s", value, fpath); | |
589 | free(fpath); | |
590 | free(value); | |
591 | return ret >= 0; | |
592 | ||
593 | bad: | |
594 | SYSERROR("Error reading '%s'", fpath); | |
595 | free(fpath); | |
596 | free(value); | |
597 | return false; | |
598 | } | |
599 | ||
600 | /* | |
601 | * Initialize the cpuset hierarchy in first directory of @gname and | |
602 | * set cgroup.clone_children so that children inherit settings. | |
603 | * Since the h->base_path is populated by init or ourselves, we know | |
604 | * it is already initialized. | |
605 | */ | |
a54694f8 | 606 | static bool handle_cpuset_hierarchy(struct hierarchy *h, char *cgname) |
e3a3fecf SH |
607 | { |
608 | char *cgpath, *clonechildrenpath, v, *slash; | |
609 | ||
610 | if (!string_in_list(h->controllers, "cpuset")) | |
611 | return true; | |
612 | ||
613 | if (*cgname == '/') | |
614 | cgname++; | |
615 | slash = strchr(cgname, '/'); | |
616 | if (slash) | |
617 | *slash = '\0'; | |
618 | ||
619 | cgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); | |
620 | if (slash) | |
621 | *slash = '/'; | |
622 | if (mkdir(cgpath, 0755) < 0 && errno != EEXIST) { | |
623 | SYSERROR("Failed to create '%s'", cgpath); | |
624 | free(cgpath); | |
625 | return false; | |
626 | } | |
6f9584d8 | 627 | |
e3a3fecf SH |
628 | clonechildrenpath = must_make_path(cgpath, "cgroup.clone_children", NULL); |
629 | if (!file_exists(clonechildrenpath)) { /* unified hierarchy doesn't have clone_children */ | |
630 | free(clonechildrenpath); | |
631 | free(cgpath); | |
632 | return true; | |
633 | } | |
634 | if (lxc_read_from_file(clonechildrenpath, &v, 1) < 0) { | |
635 | SYSERROR("Failed to read '%s'", clonechildrenpath); | |
636 | free(clonechildrenpath); | |
637 | free(cgpath); | |
638 | return false; | |
639 | } | |
640 | ||
a54694f8 | 641 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ |
6f9584d8 CB |
642 | if (!filter_and_set_cpus(cgpath, v == '1')) { |
643 | SYSERROR("Failed to remove isolated cpus."); | |
644 | free(clonechildrenpath); | |
645 | free(cgpath); | |
a54694f8 | 646 | return false; |
6f9584d8 | 647 | } |
a54694f8 | 648 | |
e3a3fecf | 649 | if (v == '1') { /* already set for us by someone else */ |
6f9584d8 | 650 | DEBUG("\"cgroup.clone_children\" was already set to \"1\"."); |
e3a3fecf SH |
651 | free(clonechildrenpath); |
652 | free(cgpath); | |
653 | return true; | |
654 | } | |
655 | ||
656 | /* copy parent's settings */ | |
a54694f8 | 657 | if (!copy_parent_file(cgpath, "cpuset.mems")) { |
6f9584d8 | 658 | SYSERROR("Failed to copy \"cpuset.mems\" settings."); |
e3a3fecf SH |
659 | free(cgpath); |
660 | free(clonechildrenpath); | |
661 | return false; | |
662 | } | |
663 | free(cgpath); | |
664 | ||
665 | if (lxc_write_to_file(clonechildrenpath, "1", 1, false) < 0) { | |
666 | /* Set clone_children so children inherit our settings */ | |
667 | SYSERROR("Failed to write 1 to %s", clonechildrenpath); | |
668 | free(clonechildrenpath); | |
669 | return false; | |
670 | } | |
671 | free(clonechildrenpath); | |
672 | return true; | |
673 | } | |
674 | ||
ccb4cabe SH |
675 | /* |
676 | * Given two null-terminated lists of strings, return true if any string | |
677 | * is in both. | |
678 | */ | |
679 | static bool controller_lists_intersect(char **l1, char **l2) | |
680 | { | |
681 | int i; | |
682 | ||
683 | if (!l1 || !l2) | |
684 | return false; | |
685 | ||
686 | for (i = 0; l1[i]; i++) { | |
687 | if (string_in_list(l2, l1[i])) | |
688 | return true; | |
689 | } | |
690 | return false; | |
691 | } | |
692 | ||
693 | /* | |
694 | * For a null-terminated list of controllers @clist, return true if any of | |
695 | * those controllers is already listed the null-terminated list of | |
696 | * hierarchies @hlist. Realistically, if one is present, all must be present. | |
697 | */ | |
698 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
699 | { | |
700 | int i; | |
701 | ||
702 | if (!hlist) | |
703 | return false; | |
704 | for (i = 0; hlist[i]; i++) | |
705 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
706 | return true; | |
707 | return false; | |
708 | ||
709 | } | |
710 | ||
711 | /* | |
712 | * Return true if the controller @entry is found in the null-terminated | |
713 | * list of hierarchies @hlist | |
714 | */ | |
715 | static bool controller_found(struct hierarchy **hlist, char *entry) | |
716 | { | |
717 | int i; | |
718 | if (!hlist) | |
719 | return false; | |
720 | ||
721 | for (i = 0; hlist[i]; i++) | |
722 | if (string_in_list(hlist[i]->controllers, entry)) | |
723 | return true; | |
724 | return false; | |
725 | } | |
726 | ||
727 | /* | |
c30b61c3 SH |
728 | * Return true if all of the controllers which we require have been found. |
729 | * The required list is freezer and anything in * lxc.cgroup.use. | |
ccb4cabe | 730 | */ |
457ca9aa | 731 | static bool all_controllers_found(void) |
ccb4cabe SH |
732 | { |
733 | char *p, *saveptr = NULL; | |
457ca9aa | 734 | struct hierarchy ** hlist = hierarchies; |
ccb4cabe | 735 | |
ccb4cabe SH |
736 | if (!controller_found(hlist, "freezer")) { |
737 | ERROR("no freezer controller mountpoint found"); | |
738 | return false; | |
739 | } | |
740 | ||
457ca9aa | 741 | if (!cgroup_use) |
ccb4cabe | 742 | return true; |
457ca9aa | 743 | for (p = strtok_r(cgroup_use, ",", &saveptr); p; |
ccb4cabe SH |
744 | p = strtok_r(NULL, ",", &saveptr)) { |
745 | if (!controller_found(hlist, p)) { | |
746 | ERROR("no %s controller mountpoint found", p); | |
747 | return false; | |
748 | } | |
749 | } | |
750 | return true; | |
751 | } | |
752 | ||
753 | /* Return true if the fs type is fuse.lxcfs */ | |
754 | static bool is_lxcfs(const char *line) | |
755 | { | |
756 | char *p = strstr(line, " - "); | |
757 | if (!p) | |
758 | return false; | |
2f62fb00 | 759 | return strncmp(p, " - fuse.lxcfs ", 14) == 0; |
ccb4cabe SH |
760 | } |
761 | ||
762 | /* | |
763 | * Get the controllers from a mountinfo line | |
764 | * There are other ways we could get this info. For lxcfs, field 3 | |
765 | * is /cgroup/controller-list. For cgroupfs, we could parse the mount | |
766 | * options. But we simply assume that the mountpoint must be | |
767 | * /sys/fs/cgroup/controller-list | |
768 | */ | |
769 | static char **get_controllers(char **klist, char **nlist, char *line) | |
770 | { | |
771 | // the fourth field is /sys/fs/cgroup/comma-delimited-controller-list | |
772 | int i; | |
773 | char *p = line, *p2, *tok, *saveptr = NULL; | |
774 | char **aret = NULL; | |
775 | ||
776 | for (i = 0; i < 4; i++) { | |
235f1815 | 777 | p = strchr(p, ' '); |
ccb4cabe SH |
778 | if (!p) |
779 | return NULL; | |
780 | p++; | |
781 | } | |
782 | if (!p) | |
783 | return NULL; | |
784 | /* note - if we change how mountinfo works, then our caller | |
785 | * will need to verify /sys/fs/cgroup/ in this field */ | |
5059aae9 SH |
786 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) { |
787 | INFO("cgfsng: found hierarchy not under /sys/fs/cgroup: \"%s\"", p); | |
ccb4cabe | 788 | return NULL; |
5059aae9 | 789 | } |
ccb4cabe | 790 | p += 15; |
235f1815 | 791 | p2 = strchr(p, ' '); |
ccb4cabe SH |
792 | if (!p2) { |
793 | ERROR("corrupt mountinfo"); | |
794 | return NULL; | |
795 | } | |
796 | *p2 = '\0'; | |
797 | for (tok = strtok_r(p, ",", &saveptr); tok; | |
798 | tok = strtok_r(NULL, ",", &saveptr)) { | |
799 | must_append_controller(klist, nlist, &aret, tok); | |
800 | } | |
801 | ||
802 | return aret; | |
803 | } | |
804 | ||
805 | /* return true if the fstype is cgroup */ | |
806 | static bool is_cgroupfs(char *line) | |
807 | { | |
808 | char *p = strstr(line, " - "); | |
809 | if (!p) | |
810 | return false; | |
2f62fb00 | 811 | return strncmp(p, " - cgroup ", 10) == 0; |
ccb4cabe SH |
812 | } |
813 | ||
814 | /* Add a controller to our list of hierarchies */ | |
457ca9aa | 815 | static void add_controller(char **clist, char *mountpoint, char *base_cgroup) |
ccb4cabe SH |
816 | { |
817 | struct hierarchy *new; | |
818 | int newentry; | |
819 | ||
820 | new = must_alloc(sizeof(*new)); | |
821 | new->controllers = clist; | |
822 | new->mountpoint = mountpoint; | |
823 | new->base_cgroup = base_cgroup; | |
824 | new->fullcgpath = NULL; | |
825 | ||
457ca9aa SH |
826 | newentry = append_null_to_list((void ***)&hierarchies); |
827 | hierarchies[newentry] = new; | |
ccb4cabe SH |
828 | } |
829 | ||
830 | /* | |
831 | * Get a copy of the mountpoint from @line, which is a line from | |
832 | * /proc/self/mountinfo | |
833 | */ | |
834 | static char *get_mountpoint(char *line) | |
835 | { | |
836 | int i; | |
837 | char *p = line, *sret; | |
838 | size_t len; | |
839 | ||
840 | for (i = 0; i < 4; i++) { | |
235f1815 | 841 | p = strchr(p, ' '); |
ccb4cabe SH |
842 | if (!p) |
843 | return NULL; | |
844 | p++; | |
845 | } | |
846 | /* we've already stuck a \0 after the mountpoint */ | |
847 | len = strlen(p); | |
848 | sret = must_alloc(len + 1); | |
849 | memcpy(sret, p, len); | |
850 | sret[len] = '\0'; | |
851 | return sret; | |
852 | } | |
853 | ||
854 | /* | |
855 | * Given a multi-line string, return a null-terminated copy of the | |
856 | * current line. | |
857 | */ | |
858 | static char *copy_to_eol(char *p) | |
859 | { | |
235f1815 | 860 | char *p2 = strchr(p, '\n'), *sret; |
ccb4cabe SH |
861 | size_t len; |
862 | ||
863 | if (!p2) | |
864 | return NULL; | |
865 | ||
866 | len = p2 - p; | |
867 | sret = must_alloc(len + 1); | |
868 | memcpy(sret, p, len); | |
869 | sret[len] = '\0'; | |
870 | return sret; | |
871 | } | |
872 | ||
873 | /* | |
874 | * cgline: pointer to character after the first ':' in a line in a | |
875 | * \n-terminated /proc/self/cgroup file. Check whether * controller c is | |
876 | * present. | |
877 | */ | |
878 | static bool controller_in_clist(char *cgline, char *c) | |
879 | { | |
880 | char *tok, *saveptr = NULL, *eol, *tmp; | |
881 | size_t len; | |
882 | ||
235f1815 | 883 | eol = strchr(cgline, ':'); |
ccb4cabe SH |
884 | if (!eol) |
885 | return false; | |
886 | ||
887 | len = eol - cgline; | |
888 | tmp = alloca(len + 1); | |
889 | memcpy(tmp, cgline, len); | |
890 | tmp[len] = '\0'; | |
891 | ||
892 | for (tok = strtok_r(tmp, ",", &saveptr); tok; | |
893 | tok = strtok_r(NULL, ",", &saveptr)) { | |
894 | if (strcmp(tok, c) == 0) | |
895 | return true; | |
896 | } | |
897 | return false; | |
898 | } | |
899 | ||
900 | /* | |
901 | * @basecginfo is a copy of /proc/$$/cgroup. Return the current | |
902 | * cgroup for @controller | |
903 | */ | |
904 | static char *get_current_cgroup(char *basecginfo, char *controller) | |
905 | { | |
906 | char *p = basecginfo; | |
907 | ||
908 | while (1) { | |
235f1815 | 909 | p = strchr(p, ':'); |
ccb4cabe SH |
910 | if (!p) |
911 | return NULL; | |
912 | p++; | |
913 | if (controller_in_clist(p, controller)) { | |
235f1815 | 914 | p = strchr(p, ':'); |
ccb4cabe SH |
915 | if (!p) |
916 | return NULL; | |
917 | p++; | |
918 | return copy_to_eol(p); | |
919 | } | |
920 | ||
235f1815 | 921 | p = strchr(p, '\n'); |
ccb4cabe SH |
922 | if (!p) |
923 | return NULL; | |
924 | p++; | |
925 | } | |
926 | } | |
927 | ||
ccb4cabe SH |
928 | /* |
929 | * Given a hierarchy @mountpoint and base @path, verify that we can create | |
930 | * directories underneath it. | |
931 | */ | |
932 | static bool test_writeable(char *mountpoint, char *path) | |
933 | { | |
934 | char *fullpath = must_make_path(mountpoint, path, NULL); | |
935 | int ret; | |
936 | ||
937 | ret = access(fullpath, W_OK); | |
938 | free(fullpath); | |
939 | return ret == 0; | |
940 | } | |
941 | ||
942 | static void must_append_string(char ***list, char *entry) | |
943 | { | |
944 | int newentry = append_null_to_list((void ***)list); | |
945 | char *copy; | |
946 | ||
947 | copy = must_copy_string(entry); | |
948 | (*list)[newentry] = copy; | |
949 | } | |
950 | ||
951 | static void get_existing_subsystems(char ***klist, char ***nlist) | |
952 | { | |
953 | FILE *f; | |
954 | char *line = NULL; | |
955 | size_t len = 0; | |
956 | ||
957 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) | |
958 | return; | |
959 | while (getline(&line, &len, f) != -1) { | |
960 | char *p, *p2, *tok, *saveptr = NULL; | |
235f1815 | 961 | p = strchr(line, ':'); |
ccb4cabe SH |
962 | if (!p) |
963 | continue; | |
964 | p++; | |
235f1815 | 965 | p2 = strchr(p, ':'); |
ccb4cabe SH |
966 | if (!p2) |
967 | continue; | |
968 | *p2 = '\0'; | |
ff8d6ee9 CB |
969 | |
970 | /* If we have a mixture between cgroup v1 and cgroup v2 | |
971 | * hierarchies, then /proc/self/cgroup contains entries of the | |
972 | * form: | |
973 | * | |
974 | * 0::/some/path | |
975 | * | |
976 | * We need to skip those. | |
977 | */ | |
978 | if ((p2 - p) == 0) | |
979 | continue; | |
980 | ||
ccb4cabe SH |
981 | for (tok = strtok_r(p, ",", &saveptr); tok; |
982 | tok = strtok_r(NULL, ",", &saveptr)) { | |
983 | if (strncmp(tok, "name=", 5) == 0) | |
984 | must_append_string(nlist, tok); | |
985 | else | |
986 | must_append_string(klist, tok); | |
987 | } | |
988 | } | |
989 | ||
990 | free(line); | |
991 | fclose(f); | |
992 | } | |
993 | ||
994 | static void trim(char *s) | |
995 | { | |
996 | size_t len = strlen(s); | |
2c28d76b | 997 | while ((len > 1) && (s[len - 1] == '\n')) |
ccb4cabe SH |
998 | s[--len] = '\0'; |
999 | } | |
1000 | ||
e4aeecf5 CB |
1001 | static void lxc_cgfsng_print_handler_data(const struct cgfsng_handler_data *d) |
1002 | { | |
1003 | printf("Cgroup information:\n"); | |
1004 | printf(" container name: %s\n", d->name ? d->name : "(null)"); | |
1005 | printf(" lxc.cgroup.use: %s\n", cgroup_use ? cgroup_use : "(null)"); | |
1006 | printf(" lxc.cgroup.pattern: %s\n", d->cgroup_pattern ? d->cgroup_pattern : "(null)"); | |
1007 | printf(" cgroup: %s\n", d->container_cgroup ? d->container_cgroup : "(null)"); | |
1008 | } | |
1009 | ||
1010 | static void lxc_cgfsng_print_hierarchies() | |
ccb4cabe | 1011 | { |
a7b0cc4c | 1012 | struct hierarchy **it; |
ccb4cabe | 1013 | int i; |
41c33dbe | 1014 | |
457ca9aa | 1015 | if (!hierarchies) { |
e4aeecf5 | 1016 | printf(" No hierarchies found."); |
ccb4cabe SH |
1017 | return; |
1018 | } | |
e4aeecf5 | 1019 | printf(" Hierarchies:\n"); |
a7b0cc4c CB |
1020 | for (i = 0, it = hierarchies; it && *it; it++, i++) { |
1021 | char **cit; | |
ccb4cabe | 1022 | int j; |
e4aeecf5 CB |
1023 | printf(" %d: base_cgroup %s\n", i, (*it)->base_cgroup ? (*it)->base_cgroup : "(null)"); |
1024 | printf(" mountpoint %s\n", (*it)->mountpoint ? (*it)->mountpoint : "(null)"); | |
1025 | printf(" controllers:\n"); | |
a7b0cc4c | 1026 | for (j = 0, cit = (*it)->controllers; cit && *cit; cit++, j++) |
e4aeecf5 | 1027 | printf(" %d: %s\n", j, *cit); |
ccb4cabe SH |
1028 | } |
1029 | } | |
41c33dbe | 1030 | |
e4aeecf5 | 1031 | static void lxc_cgfsng_print_basecg_debuginfo(char *basecginfo, char **klist, char **nlist) |
41c33dbe SH |
1032 | { |
1033 | int k; | |
a7b0cc4c | 1034 | char **it; |
41c33dbe | 1035 | |
a7b0cc4c CB |
1036 | printf("basecginfo is:\n"); |
1037 | printf("%s\n", basecginfo); | |
41c33dbe | 1038 | |
a7b0cc4c CB |
1039 | for (k = 0, it = klist; it && *it; it++, k++) |
1040 | printf("kernel subsystem %d: %s\n", k, *it); | |
1041 | for (k = 0, it = nlist; it && *it; it++, k++) | |
1042 | printf("named subsystem %d: %s\n", k, *it); | |
41c33dbe | 1043 | } |
ccb4cabe | 1044 | |
e4aeecf5 CB |
1045 | static void lxc_cgfsng_print_debuginfo(const struct cgfsng_handler_data *d) |
1046 | { | |
1047 | lxc_cgfsng_print_handler_data(d); | |
1048 | lxc_cgfsng_print_hierarchies(); | |
1049 | } | |
1050 | ||
ccb4cabe SH |
1051 | /* |
1052 | * At startup, parse_hierarchies finds all the info we need about | |
1053 | * cgroup mountpoints and current cgroups, and stores it in @d. | |
1054 | */ | |
457ca9aa | 1055 | static bool parse_hierarchies(void) |
ccb4cabe SH |
1056 | { |
1057 | FILE *f; | |
1058 | char * line = NULL, *basecginfo; | |
1059 | char **klist = NULL, **nlist = NULL; | |
1060 | size_t len = 0; | |
1061 | ||
d30ec4cb SH |
1062 | /* |
1063 | * Root spawned containers escape the current cgroup, so use init's | |
1064 | * cgroups as our base in that case. | |
1065 | */ | |
ccb4cabe SH |
1066 | if (geteuid()) |
1067 | basecginfo = read_file("/proc/self/cgroup"); | |
1068 | else | |
1069 | basecginfo = read_file("/proc/1/cgroup"); | |
1070 | if (!basecginfo) | |
1071 | return false; | |
1072 | ||
1073 | if ((f = fopen("/proc/self/mountinfo", "r")) == NULL) { | |
d3b00a8f | 1074 | SYSERROR("Failed opening /proc/self/mountinfo"); |
ccb4cabe SH |
1075 | return false; |
1076 | } | |
1077 | ||
1078 | get_existing_subsystems(&klist, &nlist); | |
41c33dbe | 1079 | |
e4aeecf5 CB |
1080 | if (lxc_cgfsng_debug) |
1081 | lxc_cgfsng_print_basecg_debuginfo(basecginfo, klist, nlist); | |
ccb4cabe SH |
1082 | |
1083 | /* we support simple cgroup mounts and lxcfs mounts */ | |
1084 | while (getline(&line, &len, f) != -1) { | |
1085 | char **controller_list = NULL; | |
1086 | char *mountpoint, *base_cgroup; | |
1087 | ||
1088 | if (!is_lxcfs(line) && !is_cgroupfs(line)) | |
1089 | continue; | |
1090 | ||
1091 | controller_list = get_controllers(klist, nlist, line); | |
1092 | if (!controller_list) | |
1093 | continue; | |
1094 | ||
457ca9aa | 1095 | if (controller_list_is_dup(hierarchies, controller_list)) { |
ccb4cabe SH |
1096 | free(controller_list); |
1097 | continue; | |
1098 | } | |
1099 | ||
1100 | mountpoint = get_mountpoint(line); | |
1101 | if (!mountpoint) { | |
1102 | ERROR("Error reading mountinfo: bad line '%s'", line); | |
1103 | free_string_list(controller_list); | |
1104 | continue; | |
1105 | } | |
1106 | ||
1107 | base_cgroup = get_current_cgroup(basecginfo, controller_list[0]); | |
1108 | if (!base_cgroup) { | |
1109 | ERROR("Failed to find current cgroup for controller '%s'", controller_list[0]); | |
1110 | free_string_list(controller_list); | |
1111 | free(mountpoint); | |
1112 | continue; | |
1113 | } | |
1114 | trim(base_cgroup); | |
1115 | prune_init_scope(base_cgroup); | |
1116 | if (!test_writeable(mountpoint, base_cgroup)) { | |
1117 | free_string_list(controller_list); | |
1118 | free(mountpoint); | |
1119 | free(base_cgroup); | |
1120 | continue; | |
1121 | } | |
457ca9aa | 1122 | add_controller(controller_list, mountpoint, base_cgroup); |
ccb4cabe SH |
1123 | } |
1124 | ||
1125 | free_string_list(klist); | |
1126 | free_string_list(nlist); | |
1127 | ||
1128 | free(basecginfo); | |
1129 | ||
1130 | fclose(f); | |
1131 | free(line); | |
1132 | ||
e4aeecf5 CB |
1133 | if (lxc_cgfsng_debug) { |
1134 | printf("writeable subsystems:\n"); | |
1135 | lxc_cgfsng_print_hierarchies(); | |
1136 | } | |
1137 | ||
ccb4cabe SH |
1138 | /* verify that all controllers in cgroup.use and all crucial |
1139 | * controllers are accounted for | |
1140 | */ | |
5059aae9 SH |
1141 | if (!all_controllers_found()) { |
1142 | INFO("cgfsng: not all controllers were find, deferring to cgfs driver"); | |
ccb4cabe | 1143 | return false; |
5059aae9 | 1144 | } |
ccb4cabe SH |
1145 | |
1146 | return true; | |
1147 | } | |
1148 | ||
457ca9aa SH |
1149 | static bool collect_hierarchy_info(void) |
1150 | { | |
1151 | const char *tmp; | |
1152 | errno = 0; | |
1153 | tmp = lxc_global_config_value("lxc.cgroup.use"); | |
1154 | if (!cgroup_use && errno != 0) { // lxc.cgroup.use can be NULL | |
1155 | SYSERROR("cgfsng: error reading list of cgroups to use"); | |
1156 | return false; | |
1157 | } | |
1158 | cgroup_use = must_copy_string(tmp); | |
1159 | ||
1160 | return parse_hierarchies(); | |
1161 | } | |
1162 | ||
ccb4cabe SH |
1163 | static void *cgfsng_init(const char *name) |
1164 | { | |
1165 | struct cgfsng_handler_data *d; | |
457ca9aa | 1166 | const char *cgroup_pattern; |
ccb4cabe SH |
1167 | |
1168 | d = must_alloc(sizeof(*d)); | |
1169 | memset(d, 0, sizeof(*d)); | |
1170 | ||
1171 | d->name = must_copy_string(name); | |
1172 | ||
ccb4cabe SH |
1173 | cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); |
1174 | if (!cgroup_pattern) { // lxc.cgroup.pattern is only NULL on error | |
1175 | ERROR("Error getting cgroup pattern"); | |
1176 | goto out_free; | |
1177 | } | |
1178 | d->cgroup_pattern = must_copy_string(cgroup_pattern); | |
1179 | ||
e4aeecf5 CB |
1180 | if (lxc_cgfsng_debug) |
1181 | lxc_cgfsng_print_debuginfo(d); | |
ccb4cabe SH |
1182 | |
1183 | return d; | |
1184 | ||
1185 | out_free: | |
1186 | free_handler_data(d); | |
1187 | return NULL; | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * Concatenate all passed-in strings into one path. Do not fail. If any piece is | |
1192 | * not prefixed with '/', add a '/'. | |
1193 | */ | |
1194 | static char *must_make_path(const char *first, ...) | |
1195 | { | |
1196 | va_list args; | |
1197 | char *cur, *dest; | |
1198 | size_t full_len = strlen(first); | |
1199 | ||
1200 | dest = must_copy_string(first); | |
1201 | ||
1202 | va_start(args, first); | |
1203 | while ((cur = va_arg(args, char *)) != NULL) { | |
1204 | full_len += strlen(cur); | |
1205 | if (cur[0] != '/') | |
1206 | full_len++; | |
1207 | dest = must_realloc(dest, full_len + 1); | |
1208 | if (cur[0] != '/') | |
1209 | strcat(dest, "/"); | |
1210 | strcat(dest, cur); | |
1211 | } | |
1212 | va_end(args); | |
1213 | ||
1214 | return dest; | |
1215 | } | |
1216 | ||
1217 | static int cgroup_rmdir(char *dirname) | |
1218 | { | |
74f96976 | 1219 | struct dirent *direntp; |
ccb4cabe SH |
1220 | DIR *dir; |
1221 | int r = 0; | |
1222 | ||
1223 | dir = opendir(dirname); | |
1224 | if (!dir) | |
1225 | return -1; | |
1226 | ||
74f96976 | 1227 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
1228 | struct stat mystat; |
1229 | char *pathname; | |
1230 | ||
1231 | if (!direntp) | |
1232 | break; | |
1233 | ||
1234 | if (!strcmp(direntp->d_name, ".") || | |
1235 | !strcmp(direntp->d_name, "..")) | |
1236 | continue; | |
1237 | ||
1238 | pathname = must_make_path(dirname, direntp->d_name, NULL); | |
1239 | ||
1240 | if (lstat(pathname, &mystat)) { | |
1241 | if (!r) | |
1c9da8da | 1242 | WARN("failed to stat %s", pathname); |
ccb4cabe SH |
1243 | r = -1; |
1244 | goto next; | |
1245 | } | |
1246 | ||
1247 | if (!S_ISDIR(mystat.st_mode)) | |
1248 | goto next; | |
1249 | if (cgroup_rmdir(pathname) < 0) | |
1250 | r = -1; | |
1251 | next: | |
1252 | free(pathname); | |
1253 | } | |
1254 | ||
1255 | if (rmdir(dirname) < 0) { | |
1256 | if (!r) | |
13277ec4 | 1257 | WARN("failed to delete %s: %s", dirname, strerror(errno)); |
ccb4cabe SH |
1258 | r = -1; |
1259 | } | |
1260 | ||
1261 | if (closedir(dir) < 0) { | |
1262 | if (!r) | |
13277ec4 | 1263 | WARN("failed to delete %s: %s", dirname, strerror(errno)); |
ccb4cabe SH |
1264 | r = -1; |
1265 | } | |
1266 | return r; | |
1267 | } | |
1268 | ||
1269 | static int rmdir_wrapper(void *data) | |
1270 | { | |
1271 | char *path = data; | |
1272 | ||
1273 | if (setresgid(0,0,0) < 0) | |
1274 | SYSERROR("Failed to setgid to 0"); | |
1275 | if (setresuid(0,0,0) < 0) | |
1276 | SYSERROR("Failed to setuid to 0"); | |
1277 | if (setgroups(0, NULL) < 0) | |
1278 | SYSERROR("Failed to clear groups"); | |
1279 | ||
1280 | return cgroup_rmdir(path); | |
1281 | } | |
1282 | ||
1283 | void recursive_destroy(char *path, struct lxc_conf *conf) | |
1284 | { | |
1285 | int r; | |
1286 | if (conf && !lxc_list_empty(&conf->id_map)) | |
c9b7c33e | 1287 | r = userns_exec_1(conf, rmdir_wrapper, path, "rmdir_wrapper"); |
ccb4cabe SH |
1288 | else |
1289 | r = cgroup_rmdir(path); | |
1290 | ||
1291 | if (r < 0) | |
1c9da8da | 1292 | ERROR("Error destroying %s", path); |
ccb4cabe SH |
1293 | } |
1294 | ||
1295 | static void cgfsng_destroy(void *hdata, struct lxc_conf *conf) | |
1296 | { | |
1297 | struct cgfsng_handler_data *d = hdata; | |
1298 | ||
1299 | if (!d) | |
1300 | return; | |
1301 | ||
457ca9aa | 1302 | if (d->container_cgroup && hierarchies) { |
ccb4cabe | 1303 | int i; |
457ca9aa SH |
1304 | for (i = 0; hierarchies[i]; i++) { |
1305 | struct hierarchy *h = hierarchies[i]; | |
e2db2a89 | 1306 | if (h->fullcgpath) { |
ccb4cabe SH |
1307 | recursive_destroy(h->fullcgpath, conf); |
1308 | free(h->fullcgpath); | |
1309 | h->fullcgpath = NULL; | |
1310 | } | |
1311 | } | |
1312 | } | |
1313 | ||
1314 | free_handler_data(d); | |
1315 | } | |
1316 | ||
1317 | struct cgroup_ops *cgfsng_ops_init(void) | |
1318 | { | |
e4aeecf5 CB |
1319 | if (getenv("LXC_DEBUG_CGFSNG")) |
1320 | lxc_cgfsng_debug = true; | |
1321 | ||
457ca9aa SH |
1322 | if (!collect_hierarchy_info()) |
1323 | return NULL; | |
e4aeecf5 | 1324 | |
ccb4cabe SH |
1325 | return &cgfsng_ops; |
1326 | } | |
1327 | ||
1328 | static bool create_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1329 | { | |
e3a3fecf | 1330 | h->fullcgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); |
6f9584d8 CB |
1331 | if (dir_exists(h->fullcgpath)) { // it must not already exist |
1332 | ERROR("Path \"%s\" already existed.", h->fullcgpath); | |
d8da679e | 1333 | return false; |
6f9584d8 CB |
1334 | } |
1335 | if (!handle_cpuset_hierarchy(h, cgname)) { | |
1336 | ERROR("Failed to handle cgroupfs v1 cpuset controller."); | |
e3a3fecf | 1337 | return false; |
6f9584d8 | 1338 | } |
e3a3fecf | 1339 | return mkdir_p(h->fullcgpath, 0755) == 0; |
ccb4cabe SH |
1340 | } |
1341 | ||
1342 | static void remove_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1343 | { | |
1344 | if (rmdir(h->fullcgpath) < 0) | |
1345 | SYSERROR("Failed to clean up cgroup %s from failed creation attempt", h->fullcgpath); | |
1346 | free(h->fullcgpath); | |
1347 | h->fullcgpath = NULL; | |
1348 | } | |
1349 | ||
1350 | /* | |
d30ec4cb | 1351 | * Try to create the same cgroup in all hierarchies. |
ccb4cabe SH |
1352 | * Start with cgroup_pattern; next cgroup_pattern-1, -2, ..., -999 |
1353 | */ | |
1354 | static inline bool cgfsng_create(void *hdata) | |
1355 | { | |
1356 | struct cgfsng_handler_data *d = hdata; | |
1357 | char *tmp, *cgname, *offset; | |
1358 | int i, idx = 0; | |
1359 | size_t len; | |
1360 | ||
1361 | if (!d) | |
1362 | return false; | |
1363 | if (d->container_cgroup) { | |
1364 | WARN("cgfsng_create called a second time"); | |
1365 | return false; | |
1366 | } | |
1367 | ||
1368 | tmp = lxc_string_replace("%n", d->name, d->cgroup_pattern); | |
1369 | if (!tmp) { | |
1370 | ERROR("Failed expanding cgroup name pattern"); | |
1371 | return false; | |
1372 | } | |
1373 | len = strlen(tmp) + 5; // leave room for -NNN\0 | |
1374 | cgname = must_alloc(len); | |
1375 | strcpy(cgname, tmp); | |
1376 | free(tmp); | |
1377 | offset = cgname + len - 5; | |
1378 | ||
1379 | again: | |
95adfe93 SH |
1380 | if (idx == 1000) { |
1381 | ERROR("Too many conflicting cgroup names"); | |
ccb4cabe | 1382 | goto out_free; |
95adfe93 | 1383 | } |
ccb4cabe SH |
1384 | if (idx) |
1385 | snprintf(offset, 5, "-%d", idx); | |
457ca9aa SH |
1386 | for (i = 0; hierarchies[i]; i++) { |
1387 | if (!create_path_for_hierarchy(hierarchies[i], cgname)) { | |
ccb4cabe | 1388 | int j; |
457ca9aa SH |
1389 | SYSERROR("Failed to create %s: %s", hierarchies[i]->fullcgpath, strerror(errno)); |
1390 | free(hierarchies[i]->fullcgpath); | |
1391 | hierarchies[i]->fullcgpath = NULL; | |
ccb4cabe | 1392 | for (j = 0; j < i; j++) |
457ca9aa | 1393 | remove_path_for_hierarchy(hierarchies[j], cgname); |
ccb4cabe SH |
1394 | idx++; |
1395 | goto again; | |
1396 | } | |
1397 | } | |
1398 | /* Done */ | |
1399 | d->container_cgroup = cgname; | |
1400 | return true; | |
1401 | ||
1402 | out_free: | |
1403 | free(cgname); | |
1404 | return false; | |
1405 | } | |
1406 | ||
ccb4cabe SH |
1407 | static bool cgfsng_enter(void *hdata, pid_t pid) |
1408 | { | |
ccb4cabe SH |
1409 | char pidstr[25]; |
1410 | int i, len; | |
1411 | ||
1412 | len = snprintf(pidstr, 25, "%d", pid); | |
1413 | if (len < 0 || len > 25) | |
1414 | return false; | |
1415 | ||
457ca9aa SH |
1416 | for (i = 0; hierarchies[i]; i++) { |
1417 | char *fullpath = must_make_path(hierarchies[i]->fullcgpath, | |
ccb4cabe SH |
1418 | "cgroup.procs", NULL); |
1419 | if (lxc_write_to_file(fullpath, pidstr, len, false) != 0) { | |
d3b00a8f | 1420 | SYSERROR("Failed to enter %s", fullpath); |
ccb4cabe SH |
1421 | free(fullpath); |
1422 | return false; | |
1423 | } | |
1424 | free(fullpath); | |
1425 | } | |
1426 | ||
1427 | return true; | |
1428 | } | |
1429 | ||
1430 | struct chown_data { | |
1431 | struct cgfsng_handler_data *d; | |
1432 | uid_t origuid; // target uid in parent namespace | |
1433 | }; | |
1434 | ||
c0888dfe SH |
1435 | /* |
1436 | * chgrp the container cgroups to container group. We leave | |
1437 | * the container owner as cgroup owner. So we must make the | |
1438 | * directories 775 so that the container can create sub-cgroups. | |
43647298 SH |
1439 | * |
1440 | * Also chown the tasks and cgroup.procs files. Those may not | |
1441 | * exist depending on kernel version. | |
c0888dfe | 1442 | */ |
ccb4cabe SH |
1443 | static int chown_cgroup_wrapper(void *data) |
1444 | { | |
1445 | struct chown_data *arg = data; | |
ccb4cabe SH |
1446 | uid_t destuid; |
1447 | int i; | |
1448 | ||
1449 | if (setresgid(0,0,0) < 0) | |
1450 | SYSERROR("Failed to setgid to 0"); | |
1451 | if (setresuid(0,0,0) < 0) | |
1452 | SYSERROR("Failed to setuid to 0"); | |
1453 | if (setgroups(0, NULL) < 0) | |
1454 | SYSERROR("Failed to clear groups"); | |
1455 | ||
1456 | destuid = get_ns_uid(arg->origuid); | |
1457 | ||
457ca9aa SH |
1458 | for (i = 0; hierarchies[i]; i++) { |
1459 | char *fullpath, *path = hierarchies[i]->fullcgpath; | |
43647298 SH |
1460 | |
1461 | if (chown(path, destuid, 0) < 0) { | |
ab8f5424 | 1462 | SYSERROR("Error chowning %s to %d", path, (int) destuid); |
ccb4cabe SH |
1463 | return -1; |
1464 | } | |
c0888dfe | 1465 | |
43647298 | 1466 | if (chmod(path, 0775) < 0) { |
ab8f5424 | 1467 | SYSERROR("Error chmoding %s", path); |
c0888dfe SH |
1468 | return -1; |
1469 | } | |
ccb4cabe | 1470 | |
ab8f5424 SH |
1471 | /* |
1472 | * Failures to chown these are inconvenient but not detrimental | |
1473 | * We leave these owned by the container launcher, so that container | |
1474 | * root can write to the files to attach. We chmod them 664 so that | |
1475 | * container systemd can write to the files (which systemd in wily | |
1476 | * insists on doing) | |
1477 | */ | |
43647298 SH |
1478 | fullpath = must_make_path(path, "tasks", NULL); |
1479 | if (chown(fullpath, destuid, 0) < 0 && errno != ENOENT) | |
13277ec4 | 1480 | WARN("Failed chowning %s to %d: %s", fullpath, (int) destuid, |
1481 | strerror(errno)); | |
ab8f5424 | 1482 | if (chmod(fullpath, 0664) < 0) |
13277ec4 | 1483 | WARN("Error chmoding %s: %s", path, strerror(errno)); |
43647298 SH |
1484 | free(fullpath); |
1485 | ||
1486 | fullpath = must_make_path(path, "cgroup.procs", NULL); | |
1487 | if (chown(fullpath, destuid, 0) < 0 && errno != ENOENT) | |
13277ec4 | 1488 | WARN("Failed chowning %s to %d: %s", fullpath, (int) destuid, |
1489 | strerror(errno)); | |
ab8f5424 | 1490 | if (chmod(fullpath, 0664) < 0) |
13277ec4 | 1491 | WARN("Error chmoding %s: %s", path, strerror(errno)); |
ccb4cabe SH |
1492 | free(fullpath); |
1493 | } | |
1494 | ||
1495 | return 0; | |
1496 | } | |
1497 | ||
1498 | static bool cgfsns_chown(void *hdata, struct lxc_conf *conf) | |
1499 | { | |
1500 | struct cgfsng_handler_data *d = hdata; | |
1501 | struct chown_data wrap; | |
1502 | ||
1503 | if (!d) | |
1504 | return false; | |
1505 | ||
1506 | if (lxc_list_empty(&conf->id_map)) | |
1507 | return true; | |
1508 | ||
1509 | wrap.d = d; | |
1510 | wrap.origuid = geteuid(); | |
1511 | ||
c9b7c33e CB |
1512 | if (userns_exec_1(conf, chown_cgroup_wrapper, &wrap, |
1513 | "chown_cgroup_wrapper") < 0) { | |
ccb4cabe SH |
1514 | ERROR("Error requesting cgroup chown in new namespace"); |
1515 | return false; | |
1516 | } | |
1517 | ||
1518 | return true; | |
1519 | } | |
1520 | ||
8aa1044f SH |
1521 | /* |
1522 | * We've safe-mounted a tmpfs as parent, so we don't need to protect against | |
1523 | * symlinks any more - just use mount | |
1524 | */ | |
1525 | ||
1526 | /* mount cgroup-full if requested */ | |
1527 | static int mount_cgroup_full(int type, struct hierarchy *h, char *dest, | |
1528 | char *container_cgroup) | |
1529 | { | |
1530 | if (type < LXC_AUTO_CGROUP_FULL_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) | |
1531 | return 0; | |
1532 | if (mount(h->mountpoint, dest, "cgroup", MS_BIND, NULL) < 0) { | |
1533 | SYSERROR("Error bind-mounting %s cgroup onto %s", h->mountpoint, | |
1534 | dest); | |
1535 | return -1; | |
1536 | } | |
1537 | if (type != LXC_AUTO_CGROUP_FULL_RW) { | |
5b6f9369 SH |
1538 | unsigned long flags = MS_BIND | MS_NOSUID | MS_NOEXEC | MS_NODEV | |
1539 | MS_REMOUNT | MS_RDONLY; | |
1540 | if (mount(NULL, dest, "cgroup", flags, NULL) < 0) { | |
8aa1044f SH |
1541 | SYSERROR("Error remounting %s readonly", dest); |
1542 | return -1; | |
1543 | } | |
1544 | } | |
1545 | ||
1546 | INFO("Bind mounted %s onto %s", h->mountpoint, dest); | |
1547 | if (type != LXC_AUTO_CGROUP_FULL_MIXED) | |
1548 | return 0; | |
1549 | ||
1550 | /* mount just the container path rw */ | |
1551 | char *source = must_make_path(h->mountpoint, h->base_cgroup, container_cgroup, NULL); | |
5b6f9369 | 1552 | char *rwpath = must_make_path(dest, h->base_cgroup, container_cgroup, NULL); |
8aa1044f | 1553 | if (mount(source, rwpath, "cgroup", MS_BIND, NULL) < 0) |
13277ec4 | 1554 | WARN("Failed to mount %s read-write: %s", rwpath, |
1555 | strerror(errno)); | |
8aa1044f SH |
1556 | INFO("Made %s read-write", rwpath); |
1557 | free(rwpath); | |
1558 | free(source); | |
1559 | return 0; | |
1560 | } | |
1561 | ||
1562 | /* cgroup-full:* is done, no need to create subdirs */ | |
1563 | static bool cg_mount_needs_subdirs(int type) | |
1564 | { | |
1565 | if (type >= LXC_AUTO_CGROUP_FULL_RO) | |
1566 | return false; | |
1567 | return true; | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * After $rootfs/sys/fs/container/controller/the/cg/path has been | |
1572 | * created, remount controller ro if needed and bindmount the | |
1573 | * cgroupfs onto controll/the/cg/path | |
1574 | */ | |
1575 | static int | |
1576 | do_secondstage_mounts_if_needed(int type, struct hierarchy *h, | |
1577 | char *controllerpath, char *cgpath, | |
1578 | const char *container_cgroup) | |
1579 | { | |
1580 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_MIXED) { | |
1581 | if (mount(controllerpath, controllerpath, "cgroup", MS_BIND, NULL) < 0) { | |
1582 | SYSERROR("Error bind-mounting %s", controllerpath); | |
1583 | return -1; | |
1584 | } | |
1585 | if (mount(controllerpath, controllerpath, "cgroup", | |
1586 | MS_REMOUNT | MS_BIND | MS_RDONLY, NULL) < 0) { | |
1587 | SYSERROR("Error remounting %s read-only", controllerpath); | |
1588 | return -1; | |
1589 | } | |
1590 | INFO("Remounted %s read-only", controllerpath); | |
1591 | } | |
1592 | char *sourcepath = must_make_path(h->mountpoint, h->base_cgroup, container_cgroup, NULL); | |
1593 | int flags = MS_BIND; | |
1594 | if (type == LXC_AUTO_CGROUP_RO) | |
1595 | flags |= MS_RDONLY; | |
1596 | INFO("Mounting %s onto %s", sourcepath, cgpath); | |
1597 | if (mount(sourcepath, cgpath, "cgroup", flags, NULL) < 0) { | |
1598 | free(sourcepath); | |
1599 | SYSERROR("Error mounting cgroup %s onto %s", h->controllers[0], | |
1600 | cgpath); | |
1601 | return -1; | |
1602 | } | |
1603 | free(sourcepath); | |
1604 | INFO("Completed second stage cgroup automounts for %s", cgpath); | |
1605 | return 0; | |
1606 | } | |
1607 | ||
ccb4cabe SH |
1608 | static bool cgfsng_mount(void *hdata, const char *root, int type) |
1609 | { | |
8aa1044f SH |
1610 | struct cgfsng_handler_data *d = hdata; |
1611 | char *tmpfspath = NULL; | |
1612 | bool retval = false; | |
a8de4c49 | 1613 | int i; |
8aa1044f SH |
1614 | |
1615 | if ((type & LXC_AUTO_CGROUP_MASK) == 0) | |
1616 | return true; | |
1617 | ||
ccb4cabe SH |
1618 | if (cgns_supported()) |
1619 | return true; | |
8aa1044f SH |
1620 | |
1621 | tmpfspath = must_make_path(root, "/sys/fs/cgroup", NULL); | |
1622 | ||
1623 | if (type == LXC_AUTO_CGROUP_NOSPEC) | |
1624 | type = LXC_AUTO_CGROUP_MIXED; | |
1625 | else if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
1626 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
1627 | ||
1628 | /* Mount tmpfs */ | |
1629 | if (safe_mount("cgroup_root", tmpfspath, "tmpfs", | |
1630 | MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_RELATIME, | |
1631 | "size=10240k,mode=755", | |
1632 | root) < 0) | |
1633 | goto bad; | |
1634 | ||
457ca9aa | 1635 | for (i = 0; hierarchies[i]; i++) { |
8aa1044f | 1636 | char *controllerpath, *path2; |
457ca9aa | 1637 | struct hierarchy *h = hierarchies[i]; |
8aa1044f SH |
1638 | char *controller = strrchr(h->mountpoint, '/'); |
1639 | int r; | |
1640 | ||
1641 | if (!controller) | |
1642 | continue; | |
1643 | controller++; | |
1644 | controllerpath = must_make_path(tmpfspath, controller, NULL); | |
1645 | if (dir_exists(controllerpath)) { | |
1646 | free(controllerpath); | |
1647 | continue; | |
1648 | } | |
1649 | if (mkdir(controllerpath, 0755) < 0) { | |
1650 | SYSERROR("Error creating cgroup path: %s", controllerpath); | |
1651 | free(controllerpath); | |
1652 | goto bad; | |
1653 | } | |
1654 | if (mount_cgroup_full(type, h, controllerpath, d->container_cgroup) < 0) { | |
1655 | free(controllerpath); | |
1656 | goto bad; | |
1657 | } | |
1658 | if (!cg_mount_needs_subdirs(type)) { | |
1659 | free(controllerpath); | |
1660 | continue; | |
1661 | } | |
ef4413fa | 1662 | path2 = must_make_path(controllerpath, h->base_cgroup, d->container_cgroup, NULL); |
8aa1044f SH |
1663 | if (mkdir_p(path2, 0755) < 0) { |
1664 | free(controllerpath); | |
1665 | goto bad; | |
1666 | } | |
2f62fb00 | 1667 | |
8aa1044f SH |
1668 | r = do_secondstage_mounts_if_needed(type, h, controllerpath, path2, |
1669 | d->container_cgroup); | |
1670 | free(controllerpath); | |
1671 | free(path2); | |
1672 | if (r < 0) | |
1673 | goto bad; | |
1674 | } | |
1675 | retval = true; | |
1676 | ||
1677 | bad: | |
1678 | free(tmpfspath); | |
1679 | return retval; | |
ccb4cabe SH |
1680 | } |
1681 | ||
1682 | static int recursive_count_nrtasks(char *dirname) | |
1683 | { | |
74f96976 | 1684 | struct dirent *direntp; |
ccb4cabe SH |
1685 | DIR *dir; |
1686 | int count = 0, ret; | |
1687 | char *path; | |
1688 | ||
1689 | dir = opendir(dirname); | |
1690 | if (!dir) | |
1691 | return 0; | |
1692 | ||
74f96976 | 1693 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
1694 | struct stat mystat; |
1695 | ||
1696 | if (!direntp) | |
1697 | break; | |
1698 | ||
1699 | if (!strcmp(direntp->d_name, ".") || | |
1700 | !strcmp(direntp->d_name, "..")) | |
1701 | continue; | |
1702 | ||
1703 | path = must_make_path(dirname, direntp->d_name, NULL); | |
1704 | ||
1705 | if (lstat(path, &mystat)) | |
1706 | goto next; | |
1707 | ||
1708 | if (!S_ISDIR(mystat.st_mode)) | |
1709 | goto next; | |
1710 | ||
1711 | count += recursive_count_nrtasks(path); | |
1712 | next: | |
1713 | free(path); | |
1714 | } | |
1715 | ||
1716 | path = must_make_path(dirname, "cgroup.procs", NULL); | |
1717 | ret = lxc_count_file_lines(path); | |
1718 | if (ret != -1) | |
1719 | count += ret; | |
1720 | free(path); | |
1721 | ||
1722 | (void) closedir(dir); | |
1723 | ||
1724 | return count; | |
1725 | } | |
1726 | ||
1727 | static int cgfsng_nrtasks(void *hdata) { | |
1728 | struct cgfsng_handler_data *d = hdata; | |
1729 | char *path; | |
1730 | int count; | |
1731 | ||
457ca9aa | 1732 | if (!d || !d->container_cgroup || !hierarchies) |
ccb4cabe | 1733 | return -1; |
457ca9aa | 1734 | path = must_make_path(hierarchies[0]->fullcgpath, NULL); |
ccb4cabe SH |
1735 | count = recursive_count_nrtasks(path); |
1736 | free(path); | |
1737 | return count; | |
1738 | } | |
1739 | ||
1740 | /* Only root needs to escape to the cgroup of its init */ | |
7103fe6f | 1741 | static bool cgfsng_escape() |
ccb4cabe | 1742 | { |
ccb4cabe SH |
1743 | int i; |
1744 | ||
1745 | if (geteuid()) | |
1746 | return true; | |
1747 | ||
457ca9aa SH |
1748 | for (i = 0; hierarchies[i]; i++) { |
1749 | char *fullpath = must_make_path(hierarchies[i]->mountpoint, | |
1750 | hierarchies[i]->base_cgroup, | |
ccb4cabe SH |
1751 | "cgroup.procs", NULL); |
1752 | if (lxc_write_to_file(fullpath, "0", 2, false) != 0) { | |
d3b00a8f | 1753 | SYSERROR("Failed to escape to %s", fullpath); |
ccb4cabe | 1754 | free(fullpath); |
6df334d1 | 1755 | return false; |
ccb4cabe SH |
1756 | } |
1757 | free(fullpath); | |
1758 | } | |
1759 | ||
6df334d1 | 1760 | return true; |
ccb4cabe SH |
1761 | } |
1762 | ||
36662416 TA |
1763 | static int cgfsng_num_hierarchies(void) |
1764 | { | |
1765 | int i; | |
1766 | ||
1767 | for (i = 0; hierarchies[i]; i++) | |
1768 | ; | |
1769 | ||
1770 | return i; | |
1771 | } | |
1772 | ||
1773 | static bool cgfsng_get_hierarchies(int n, char ***out) | |
1774 | { | |
1775 | int i; | |
1776 | ||
1777 | /* sanity check n */ | |
1778 | for (i = 0; i < n; i++) { | |
1779 | if (!hierarchies[i]) | |
1780 | return false; | |
1781 | } | |
1782 | ||
1783 | *out = hierarchies[i]->controllers; | |
1784 | ||
1785 | return true; | |
1786 | } | |
1787 | ||
ccb4cabe SH |
1788 | #define THAWED "THAWED" |
1789 | #define THAWED_LEN (strlen(THAWED)) | |
1790 | ||
1791 | static bool cgfsng_unfreeze(void *hdata) | |
1792 | { | |
ccb4cabe | 1793 | char *fullpath; |
457ca9aa | 1794 | struct hierarchy *h = get_hierarchy("freezer"); |
ccb4cabe | 1795 | |
457ca9aa | 1796 | if (!h) |
ccb4cabe SH |
1797 | return false; |
1798 | fullpath = must_make_path(h->fullcgpath, "freezer.state", NULL); | |
1799 | if (lxc_write_to_file(fullpath, THAWED, THAWED_LEN, false) != 0) { | |
1800 | free(fullpath); | |
1801 | return false; | |
1802 | } | |
1803 | free(fullpath); | |
1804 | return true; | |
1805 | } | |
1806 | ||
1807 | static const char *cgfsng_get_cgroup(void *hdata, const char *subsystem) | |
1808 | { | |
457ca9aa | 1809 | struct hierarchy *h = get_hierarchy(subsystem); |
ccb4cabe SH |
1810 | if (!h) |
1811 | return NULL; | |
1812 | ||
371f834d SH |
1813 | return h->fullcgpath ? h->fullcgpath + strlen(h->mountpoint) : NULL; |
1814 | } | |
1815 | ||
1816 | /* | |
1817 | * Given a cgroup path returned from lxc_cmd_get_cgroup_path, build a | |
1818 | * full path, which must be freed by the caller. | |
1819 | */ | |
1820 | static char *build_full_cgpath_from_monitorpath(struct hierarchy *h, | |
1821 | const char *inpath, | |
1822 | const char *filename) | |
1823 | { | |
1824 | /* | |
1825 | * XXX Remove this case after 2.0 release. It's for dealing with | |
1826 | * containers spawned under the old buggy cgfsng which wasn't around | |
1827 | * for long. | |
1828 | */ | |
1829 | if (strncmp(inpath, "/sys/fs/cgroup/", 15) == 0) | |
1830 | return must_make_path(inpath, filename, NULL); | |
1831 | return must_make_path(h->mountpoint, inpath, filename, NULL); | |
ccb4cabe SH |
1832 | } |
1833 | ||
1834 | static bool cgfsng_attach(const char *name, const char *lxcpath, pid_t pid) | |
1835 | { | |
ccb4cabe SH |
1836 | char pidstr[25]; |
1837 | int i, len; | |
1838 | ||
1839 | len = snprintf(pidstr, 25, "%d", pid); | |
1840 | if (len < 0 || len > 25) | |
1841 | return false; | |
1842 | ||
457ca9aa | 1843 | for (i = 0; hierarchies[i]; i++) { |
ccb4cabe | 1844 | char *path, *fullpath; |
457ca9aa | 1845 | struct hierarchy *h = hierarchies[i]; |
ccb4cabe SH |
1846 | |
1847 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->controllers[0]); | |
1848 | if (!path) // not running | |
1849 | continue; | |
1850 | ||
371f834d SH |
1851 | fullpath = build_full_cgpath_from_monitorpath(h, path, "cgroup.procs"); |
1852 | free(path); | |
ccb4cabe SH |
1853 | if (lxc_write_to_file(fullpath, pidstr, len, false) != 0) { |
1854 | SYSERROR("Failed to attach %d to %s", (int)pid, fullpath); | |
1855 | free(fullpath); | |
ccb4cabe SH |
1856 | return false; |
1857 | } | |
ccb4cabe SH |
1858 | free(fullpath); |
1859 | } | |
1860 | ||
ccb4cabe SH |
1861 | return true; |
1862 | } | |
1863 | ||
1864 | /* | |
1865 | * Called externally (i.e. from 'lxc-cgroup') to query cgroup limits. | |
1866 | * Here we don't have a cgroup_data set up, so we ask the running | |
1867 | * container through the commands API for the cgroup path | |
1868 | */ | |
1869 | static int cgfsng_get(const char *filename, char *value, size_t len, const char *name, const char *lxcpath) | |
1870 | { | |
1871 | char *subsystem, *p, *path; | |
ccb4cabe SH |
1872 | struct hierarchy *h; |
1873 | int ret = -1; | |
1874 | ||
1875 | subsystem = alloca(strlen(filename) + 1); | |
1876 | strcpy(subsystem, filename); | |
1877 | if ((p = strchr(subsystem, '.')) != NULL) | |
1878 | *p = '\0'; | |
1879 | ||
1880 | path = lxc_cmd_get_cgroup_path(name, lxcpath, subsystem); | |
1881 | if (!path) // not running | |
1882 | return -1; | |
1883 | ||
457ca9aa | 1884 | h = get_hierarchy(subsystem); |
ccb4cabe | 1885 | if (h) { |
371f834d | 1886 | char *fullpath = build_full_cgpath_from_monitorpath(h, path, filename); |
ccb4cabe SH |
1887 | ret = lxc_read_from_file(fullpath, value, len); |
1888 | free(fullpath); | |
1889 | } | |
1890 | ||
ccb4cabe SH |
1891 | free(path); |
1892 | ||
1893 | return ret; | |
1894 | } | |
1895 | ||
1896 | /* | |
1897 | * Called externally (i.e. from 'lxc-cgroup') to set new cgroup limits. | |
1898 | * Here we don't have a cgroup_data set up, so we ask the running | |
1899 | * container through the commands API for the cgroup path | |
1900 | */ | |
1901 | static int cgfsng_set(const char *filename, const char *value, const char *name, const char *lxcpath) | |
1902 | { | |
1903 | char *subsystem, *p, *path; | |
ccb4cabe SH |
1904 | struct hierarchy *h; |
1905 | int ret = -1; | |
1906 | ||
1907 | subsystem = alloca(strlen(filename) + 1); | |
1908 | strcpy(subsystem, filename); | |
1909 | if ((p = strchr(subsystem, '.')) != NULL) | |
1910 | *p = '\0'; | |
1911 | ||
1912 | path = lxc_cmd_get_cgroup_path(name, lxcpath, subsystem); | |
1913 | if (!path) // not running | |
1914 | return -1; | |
1915 | ||
457ca9aa | 1916 | h = get_hierarchy(subsystem); |
ccb4cabe | 1917 | if (h) { |
371f834d | 1918 | char *fullpath = build_full_cgpath_from_monitorpath(h, path, filename); |
ccb4cabe SH |
1919 | ret = lxc_write_to_file(fullpath, value, strlen(value), false); |
1920 | free(fullpath); | |
1921 | } | |
1922 | ||
ccb4cabe SH |
1923 | free(path); |
1924 | ||
1925 | return ret; | |
1926 | } | |
1927 | ||
ccb4cabe SH |
1928 | /* |
1929 | * Called from setup_limits - here we have the container's cgroup_data because | |
1930 | * we created the cgroups | |
1931 | */ | |
1932 | static int lxc_cgroup_set_data(const char *filename, const char *value, struct cgfsng_handler_data *d) | |
1933 | { | |
1934 | char *subsystem = NULL, *p; | |
1935 | int ret = -1; | |
1936 | struct hierarchy *h; | |
1937 | ||
1938 | subsystem = alloca(strlen(filename) + 1); | |
1939 | strcpy(subsystem, filename); | |
1940 | if ((p = strchr(subsystem, '.')) != NULL) | |
1941 | *p = '\0'; | |
1942 | ||
457ca9aa | 1943 | h = get_hierarchy(subsystem); |
ccb4cabe SH |
1944 | if (h) { |
1945 | char *fullpath = must_make_path(h->fullcgpath, filename, NULL); | |
1946 | ret = lxc_write_to_file(fullpath, value, strlen(value), false); | |
1947 | free(fullpath); | |
1948 | } | |
1949 | return ret; | |
1950 | } | |
1951 | ||
1952 | static bool cgfsng_setup_limits(void *hdata, struct lxc_list *cgroup_settings, | |
1953 | bool do_devices) | |
1954 | { | |
1955 | struct cgfsng_handler_data *d = hdata; | |
1956 | struct lxc_list *iterator, *sorted_cgroup_settings, *next; | |
1957 | struct lxc_cgroup *cg; | |
ccb4cabe SH |
1958 | bool ret = false; |
1959 | ||
1960 | if (lxc_list_empty(cgroup_settings)) | |
1961 | return true; | |
1962 | ||
1963 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); | |
1964 | if (!sorted_cgroup_settings) { | |
1965 | return false; | |
1966 | } | |
1967 | ||
ccb4cabe SH |
1968 | lxc_list_for_each(iterator, sorted_cgroup_settings) { |
1969 | cg = iterator->elem; | |
1970 | ||
1971 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { | |
ccb4cabe SH |
1972 | if (lxc_cgroup_set_data(cg->subsystem, cg->value, d)) { |
1973 | if (do_devices && (errno == EACCES || errno == EPERM)) { | |
1974 | WARN("Error setting %s to %s for %s", | |
1975 | cg->subsystem, cg->value, d->name); | |
1976 | continue; | |
1977 | } | |
1978 | SYSERROR("Error setting %s to %s for %s", | |
1979 | cg->subsystem, cg->value, d->name); | |
1980 | goto out; | |
1981 | } | |
1982 | } | |
1983 | ||
1984 | DEBUG("cgroup '%s' set to '%s'", cg->subsystem, cg->value); | |
1985 | } | |
1986 | ||
1987 | ret = true; | |
1988 | INFO("cgroup has been setup"); | |
1989 | out: | |
ccb4cabe SH |
1990 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
1991 | lxc_list_del(iterator); | |
1992 | free(iterator); | |
1993 | } | |
1994 | free(sorted_cgroup_settings); | |
1995 | return ret; | |
1996 | } | |
1997 | ||
1998 | static struct cgroup_ops cgfsng_ops = { | |
1999 | .init = cgfsng_init, | |
2000 | .destroy = cgfsng_destroy, | |
2001 | .create = cgfsng_create, | |
2002 | .enter = cgfsng_enter, | |
ccb4cabe | 2003 | .escape = cgfsng_escape, |
36662416 TA |
2004 | .num_hierarchies = cgfsng_num_hierarchies, |
2005 | .get_hierarchies = cgfsng_get_hierarchies, | |
ccb4cabe SH |
2006 | .get_cgroup = cgfsng_get_cgroup, |
2007 | .get = cgfsng_get, | |
2008 | .set = cgfsng_set, | |
2009 | .unfreeze = cgfsng_unfreeze, | |
2010 | .setup_limits = cgfsng_setup_limits, | |
2011 | .name = "cgroupfs-ng", | |
2012 | .attach = cgfsng_attach, | |
2013 | .chown = cgfsns_chown, | |
2014 | .mount_cgroup = cgfsng_mount, | |
2015 | .nrtasks = cgfsng_nrtasks, | |
2016 | .driver = CGFSNG, | |
2017 | ||
2018 | /* unsupported */ | |
2019 | .create_legacy = NULL, | |
2020 | }; |