]>
Commit | Line | Data |
---|---|---|
576f946d | 1 | /* |
2 | * lxc: linux Container library | |
3 | * | |
4 | * (C) Copyright IBM Corp. 2007, 2008 | |
5 | * | |
6 | * Authors: | |
9afe19d6 | 7 | * Daniel Lezcano <daniel.lezcano at free.fr> |
576f946d | 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 | |
250b1eec | 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
576f946d | 22 | */ |
d06245b8 NC |
23 | #include "config.h" |
24 | ||
576f946d | 25 | #include <stdio.h> |
576f946d | 26 | #include <stdlib.h> |
27 | #include <errno.h> | |
576f946d | 28 | #include <unistd.h> |
29 | #include <string.h> | |
341a9bd8 | 30 | #include <dirent.h> |
576f946d | 31 | #include <fcntl.h> |
b98f7d6e | 32 | #include <ctype.h> |
576f946d | 33 | #include <sys/types.h> |
34 | #include <sys/stat.h> | |
35 | #include <sys/param.h> | |
36 | #include <sys/inotify.h> | |
aae1f3c4 | 37 | #include <sys/mount.h> |
576f946d | 38 | #include <netinet/in.h> |
39 | #include <net/if.h> | |
40 | ||
e2bcd7db | 41 | #include "error.h" |
ae5c8b8e | 42 | #include "commands.h" |
b98f7d6e SH |
43 | #include "list.h" |
44 | #include "conf.h" | |
33ad9f1a | 45 | #include "utils.h" |
740d1928 | 46 | #include "bdev.h" |
f2363e38 ÇO |
47 | #include "log.h" |
48 | #include "cgroup.h" | |
49 | #include "start.h" | |
484ed030 | 50 | #include "state.h" |
36eb9bde | 51 | |
edaf8b1b SG |
52 | #if IS_BIONIC |
53 | #include <../include/lxcmntent.h> | |
54 | #else | |
55 | #include <mntent.h> | |
56 | #endif | |
57 | ||
4fb3cba5 DE |
58 | struct cgroup_hierarchy; |
59 | struct cgroup_meta_data; | |
60 | struct cgroup_mount_point; | |
61 | ||
62 | /* | |
63 | * cgroup_meta_data: the metadata about the cgroup infrastructure on this | |
64 | * host | |
65 | */ | |
66 | struct cgroup_meta_data { | |
67 | ptrdiff_t ref; /* simple refcount */ | |
68 | struct cgroup_hierarchy **hierarchies; | |
69 | struct cgroup_mount_point **mount_points; | |
70 | int maximum_hierarchy; | |
71 | }; | |
72 | ||
73 | /* | |
74 | * cgroup_hierarchy: describes a single cgroup hierarchy | |
75 | * (may have multiple mount points) | |
76 | */ | |
77 | struct cgroup_hierarchy { | |
78 | int index; | |
79 | bool used; /* false if the hierarchy should be ignored by lxc */ | |
80 | char **subsystems; | |
81 | struct cgroup_mount_point *rw_absolute_mount_point; | |
82 | struct cgroup_mount_point *ro_absolute_mount_point; | |
83 | struct cgroup_mount_point **all_mount_points; | |
84 | size_t all_mount_point_capacity; | |
85 | }; | |
86 | ||
87 | /* | |
88 | * cgroup_mount_point: a mount point to where a hierarchy | |
89 | * is mounted to | |
90 | */ | |
91 | struct cgroup_mount_point { | |
92 | struct cgroup_hierarchy *hierarchy; | |
93 | char *mount_point; | |
94 | char *mount_prefix; | |
95 | bool read_only; | |
96 | bool need_cpuset_init; | |
97 | }; | |
98 | ||
99 | /* | |
100 | * cgroup_process_info: describes the membership of a | |
101 | * process to the different cgroup | |
102 | * hierarchies | |
103 | * | |
104 | * Note this is the per-process info tracked by the cgfs_ops. | |
105 | * This is not used with cgmanager. | |
106 | */ | |
107 | struct cgroup_process_info { | |
108 | struct cgroup_process_info *next; | |
109 | struct cgroup_meta_data *meta_ref; | |
110 | struct cgroup_hierarchy *hierarchy; | |
111 | char *cgroup_path; | |
112 | char *cgroup_path_sub; | |
113 | char **created_paths; | |
114 | size_t created_paths_capacity; | |
115 | size_t created_paths_count; | |
116 | struct cgroup_mount_point *designated_mount_point; | |
117 | }; | |
118 | ||
119 | struct cgfs_data { | |
120 | char *name; | |
121 | const char *cgroup_pattern; | |
122 | struct cgroup_meta_data *meta; | |
123 | struct cgroup_process_info *info; | |
124 | }; | |
125 | ||
126 | lxc_log_define(lxc_cgfs, lxc); | |
576f946d | 127 | |
33ad9f1a CS |
128 | static struct cgroup_process_info *lxc_cgroup_process_info_getx(const char *proc_pid_cgroup_str, struct cgroup_meta_data *meta); |
129 | static char **subsystems_from_mount_options(const char *mount_options, char **kernel_list); | |
130 | static void lxc_cgroup_mount_point_free(struct cgroup_mount_point *mp); | |
131 | static void lxc_cgroup_hierarchy_free(struct cgroup_hierarchy *h); | |
132 | static bool is_valid_cgroup(const char *name); | |
33ad9f1a | 133 | static int create_cgroup(struct cgroup_mount_point *mp, const char *path); |
603c64c2 | 134 | static int remove_cgroup(struct cgroup_mount_point *mp, const char *path, bool recurse); |
33ad9f1a CS |
135 | static char *cgroup_to_absolute_path(struct cgroup_mount_point *mp, const char *path, const char *suffix); |
136 | static struct cgroup_process_info *find_info_for_subsystem(struct cgroup_process_info *info, const char *subsystem); | |
137 | static int do_cgroup_get(const char *cgroup_path, const char *sub_filename, char *value, size_t len); | |
138 | static int do_cgroup_set(const char *cgroup_path, const char *sub_filename, const char *value); | |
4fb3cba5 DE |
139 | static bool cgroup_devices_has_allow_or_deny(struct cgfs_data *d, char *v, bool for_allow); |
140 | static int do_setup_cgroup_limits(struct cgfs_data *d, struct lxc_list *cgroup_settings, bool do_devices); | |
33ad9f1a CS |
141 | static int cgroup_recursive_task_count(const char *cgroup_path); |
142 | static int count_lines(const char *fn); | |
1ea59ad2 | 143 | static int handle_cgroup_settings(struct cgroup_mount_point *mp, char *cgroup_path); |
d703c2b1 | 144 | static bool init_cpuset_if_needed(struct cgroup_mount_point *mp, const char *path); |
33ad9f1a | 145 | |
4fb3cba5 DE |
146 | static struct cgroup_meta_data *lxc_cgroup_load_meta2(const char **subsystem_whitelist); |
147 | static struct cgroup_meta_data *lxc_cgroup_get_meta(struct cgroup_meta_data *meta_data); | |
148 | static struct cgroup_meta_data *lxc_cgroup_put_meta(struct cgroup_meta_data *meta_data); | |
149 | ||
150 | /* free process membership information */ | |
151 | static void lxc_cgroup_process_info_free(struct cgroup_process_info *info); | |
152 | static void lxc_cgroup_process_info_free_and_remove(struct cgroup_process_info *info); | |
153 | ||
d4ef7c50 | 154 | static struct cgroup_ops cgfs_ops; |
d4ef7c50 | 155 | |
603c64c2 SH |
156 | static int cgroup_rmdir(char *dirname) |
157 | { | |
158 | struct dirent dirent, *direntp; | |
159 | int saved_errno = 0; | |
160 | DIR *dir; | |
161 | int ret, failed=0; | |
162 | char pathname[MAXPATHLEN]; | |
163 | ||
164 | dir = opendir(dirname); | |
165 | if (!dir) { | |
166 | ERROR("%s: failed to open %s", __func__, dirname); | |
167 | return -1; | |
168 | } | |
169 | ||
170 | while (!readdir_r(dir, &dirent, &direntp)) { | |
171 | struct stat mystat; | |
172 | int rc; | |
173 | ||
174 | if (!direntp) | |
175 | break; | |
176 | ||
177 | if (!strcmp(direntp->d_name, ".") || | |
178 | !strcmp(direntp->d_name, "..")) | |
179 | continue; | |
180 | ||
181 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
182 | if (rc < 0 || rc >= MAXPATHLEN) { | |
183 | ERROR("pathname too long"); | |
184 | failed=1; | |
185 | if (!saved_errno) | |
186 | saved_errno = -ENOMEM; | |
187 | continue; | |
188 | } | |
189 | ret = lstat(pathname, &mystat); | |
190 | if (ret) { | |
191 | SYSERROR("%s: failed to stat %s", __func__, pathname); | |
192 | failed=1; | |
193 | if (!saved_errno) | |
194 | saved_errno = errno; | |
195 | continue; | |
196 | } | |
197 | if (S_ISDIR(mystat.st_mode)) { | |
198 | if (cgroup_rmdir(pathname) < 0) { | |
199 | if (!saved_errno) | |
200 | saved_errno = errno; | |
201 | failed=1; | |
202 | } | |
203 | } | |
204 | } | |
205 | ||
206 | if (rmdir(dirname) < 0) { | |
207 | SYSERROR("%s: failed to delete %s", __func__, dirname); | |
208 | if (!saved_errno) | |
209 | saved_errno = errno; | |
210 | failed=1; | |
211 | } | |
212 | ||
213 | ret = closedir(dir); | |
214 | if (ret) { | |
215 | SYSERROR("%s: failed to close directory %s", __func__, dirname); | |
216 | if (!saved_errno) | |
217 | saved_errno = errno; | |
218 | failed=1; | |
219 | } | |
220 | ||
221 | errno = saved_errno; | |
222 | return failed ? -1 : 0; | |
223 | } | |
224 | ||
4fb3cba5 | 225 | static struct cgroup_meta_data *lxc_cgroup_load_meta() |
33ad9f1a CS |
226 | { |
227 | const char *cgroup_use = NULL; | |
228 | char **cgroup_use_list = NULL; | |
229 | struct cgroup_meta_data *md = NULL; | |
230 | int saved_errno; | |
231 | ||
232 | errno = 0; | |
593e8478 | 233 | cgroup_use = lxc_global_config_value("lxc.cgroup.use"); |
33ad9f1a CS |
234 | if (!cgroup_use && errno != 0) |
235 | return NULL; | |
236 | if (cgroup_use) { | |
237 | cgroup_use_list = lxc_string_split_and_trim(cgroup_use, ','); | |
238 | if (!cgroup_use_list) | |
239 | return NULL; | |
240 | } | |
576f946d | 241 | |
33ad9f1a CS |
242 | md = lxc_cgroup_load_meta2((const char **)cgroup_use_list); |
243 | saved_errno = errno; | |
244 | lxc_free_array((void **)cgroup_use_list, free); | |
245 | errno = saved_errno; | |
246 | return md; | |
247 | } | |
fd37327f | 248 | |
b653309a | 249 | /* Step 1: determine all kernel subsystems */ |
4fb3cba5 | 250 | static bool find_cgroup_subsystems(char ***kernel_subsystems) |
1d39a065 | 251 | { |
b653309a SH |
252 | FILE *proc_cgroups; |
253 | bool bret = false; | |
33ad9f1a CS |
254 | char *line = NULL; |
255 | size_t sz = 0; | |
b653309a SH |
256 | size_t kernel_subsystems_count = 0; |
257 | size_t kernel_subsystems_capacity = 0; | |
258 | int r; | |
1d39a065 | 259 | |
33ad9f1a CS |
260 | proc_cgroups = fopen_cloexec("/proc/cgroups", "r"); |
261 | if (!proc_cgroups) | |
b653309a | 262 | return false; |
1d39a065 | 263 | |
33ad9f1a CS |
264 | while (getline(&line, &sz, proc_cgroups) != -1) { |
265 | char *tab1; | |
266 | char *tab2; | |
267 | int hierarchy_number; | |
1d39a065 | 268 | |
33ad9f1a CS |
269 | if (line[0] == '#') |
270 | continue; | |
271 | if (!line[0]) | |
272 | continue; | |
1d39a065 | 273 | |
33ad9f1a CS |
274 | tab1 = strchr(line, '\t'); |
275 | if (!tab1) | |
8900b9eb | 276 | continue; |
33ad9f1a CS |
277 | *tab1++ = '\0'; |
278 | tab2 = strchr(tab1, '\t'); | |
279 | if (!tab2) | |
280 | continue; | |
281 | *tab2 = '\0'; | |
fd37327f | 282 | |
33ad9f1a CS |
283 | tab2 = NULL; |
284 | hierarchy_number = strtoul(tab1, &tab2, 10); | |
285 | if (!tab2 || *tab2) | |
286 | continue; | |
287 | (void)hierarchy_number; | |
288 | ||
b653309a | 289 | r = lxc_grow_array((void ***)kernel_subsystems, &kernel_subsystems_capacity, kernel_subsystems_count + 1, 12); |
33ad9f1a | 290 | if (r < 0) |
b653309a SH |
291 | goto out; |
292 | (*kernel_subsystems)[kernel_subsystems_count] = strdup(line); | |
293 | if (!(*kernel_subsystems)[kernel_subsystems_count]) | |
294 | goto out; | |
33ad9f1a | 295 | kernel_subsystems_count++; |
bcbd102c | 296 | } |
b653309a | 297 | bret = true; |
0d9f8e18 | 298 | |
b653309a | 299 | out: |
33ad9f1a | 300 | fclose(proc_cgroups); |
0ccf7c2a | 301 | free(line); |
b653309a SH |
302 | return bret; |
303 | } | |
304 | ||
305 | /* Step 2: determine all hierarchies (by reading /proc/self/cgroup), | |
306 | * since mount points don't specify hierarchy number and | |
307 | * /proc/cgroups does not contain named hierarchies | |
308 | */ | |
309 | static bool find_cgroup_hierarchies(struct cgroup_meta_data *meta_data, | |
310 | bool all_kernel_subsystems, bool all_named_subsystems, | |
311 | const char **subsystem_whitelist) | |
312 | { | |
313 | FILE *proc_self_cgroup; | |
314 | char *line = NULL; | |
315 | size_t sz = 0; | |
316 | int r; | |
317 | bool bret = false; | |
318 | size_t hierarchy_capacity = 0; | |
ef6e34ee | 319 | |
33ad9f1a CS |
320 | proc_self_cgroup = fopen_cloexec("/proc/self/cgroup", "r"); |
321 | /* if for some reason (because of setns() and pid namespace for example), | |
322 | * /proc/self is not valid, we try /proc/1/cgroup... */ | |
323 | if (!proc_self_cgroup) | |
324 | proc_self_cgroup = fopen_cloexec("/proc/1/cgroup", "r"); | |
325 | if (!proc_self_cgroup) | |
b653309a | 326 | return false; |
33ad9f1a CS |
327 | |
328 | while (getline(&line, &sz, proc_self_cgroup) != -1) { | |
329 | /* file format: hierarchy:subsystems:group, | |
330 | * we only extract hierarchy and subsystems | |
331 | * here */ | |
332 | char *colon1; | |
333 | char *colon2; | |
334 | int hierarchy_number; | |
335 | struct cgroup_hierarchy *h = NULL; | |
336 | char **p; | |
337 | ||
338 | if (!line[0]) | |
339 | continue; | |
ad08bbb7 | 340 | |
33ad9f1a CS |
341 | colon1 = strchr(line, ':'); |
342 | if (!colon1) | |
8900b9eb | 343 | continue; |
33ad9f1a CS |
344 | *colon1++ = '\0'; |
345 | colon2 = strchr(colon1, ':'); | |
346 | if (!colon2) | |
347 | continue; | |
348 | *colon2 = '\0'; | |
ad08bbb7 | 349 | |
33ad9f1a CS |
350 | colon2 = NULL; |
351 | hierarchy_number = strtoul(line, &colon2, 10); | |
352 | if (!colon2 || *colon2) | |
353 | continue; | |
576f946d | 354 | |
33ad9f1a CS |
355 | if (hierarchy_number > meta_data->maximum_hierarchy) { |
356 | /* lxc_grow_array will never shrink, so even if we find a lower | |
357 | * hierarchy number here, the array will never be smaller | |
358 | */ | |
359 | r = lxc_grow_array((void ***)&meta_data->hierarchies, &hierarchy_capacity, hierarchy_number + 1, 12); | |
360 | if (r < 0) | |
b653309a | 361 | goto out; |
5193cc3d | 362 | |
33ad9f1a CS |
363 | meta_data->maximum_hierarchy = hierarchy_number; |
364 | } | |
fd37327f | 365 | |
33ad9f1a CS |
366 | /* this shouldn't happen, we had this already */ |
367 | if (meta_data->hierarchies[hierarchy_number]) | |
b653309a | 368 | goto out; |
33ad9f1a CS |
369 | |
370 | h = calloc(1, sizeof(struct cgroup_hierarchy)); | |
371 | if (!h) | |
b653309a | 372 | goto out; |
33ad9f1a CS |
373 | |
374 | meta_data->hierarchies[hierarchy_number] = h; | |
375 | ||
376 | h->index = hierarchy_number; | |
377 | h->subsystems = lxc_string_split_and_trim(colon1, ','); | |
378 | if (!h->subsystems) | |
b653309a | 379 | goto out; |
33ad9f1a CS |
380 | /* see if this hierarchy should be considered */ |
381 | if (!all_kernel_subsystems || !all_named_subsystems) { | |
382 | for (p = h->subsystems; *p; p++) { | |
383 | if (!strncmp(*p, "name=", 5)) { | |
384 | if (all_named_subsystems || (subsystem_whitelist && lxc_string_in_array(*p, subsystem_whitelist))) { | |
385 | h->used = true; | |
386 | break; | |
387 | } | |
388 | } else { | |
389 | if (all_kernel_subsystems || (subsystem_whitelist && lxc_string_in_array(*p, subsystem_whitelist))) { | |
390 | h->used = true; | |
391 | break; | |
392 | } | |
393 | } | |
394 | } | |
395 | } else { | |
396 | /* we want all hierarchy anyway */ | |
397 | h->used = true; | |
ae5c8b8e | 398 | } |
ae5c8b8e | 399 | } |
b653309a | 400 | bret = true; |
0b9c21ab | 401 | |
b653309a | 402 | out: |
33ad9f1a | 403 | fclose(proc_self_cgroup); |
0ccf7c2a | 404 | free(line); |
b653309a SH |
405 | return bret; |
406 | } | |
407 | ||
408 | /* Step 3: determine all mount points of each hierarchy */ | |
409 | static bool find_hierarchy_mountpts( struct cgroup_meta_data *meta_data, char **kernel_subsystems) | |
410 | { | |
411 | bool bret = false; | |
412 | FILE *proc_self_mountinfo; | |
413 | char *line = NULL; | |
414 | size_t sz = 0; | |
415 | char **tokens = NULL; | |
416 | size_t mount_point_count = 0; | |
417 | size_t mount_point_capacity = 0; | |
418 | size_t token_capacity = 0; | |
419 | int r; | |
420 | ||
33ad9f1a CS |
421 | proc_self_mountinfo = fopen_cloexec("/proc/self/mountinfo", "r"); |
422 | /* if for some reason (because of setns() and pid namespace for example), | |
423 | * /proc/self is not valid, we try /proc/1/cgroup... */ | |
424 | if (!proc_self_mountinfo) | |
425 | proc_self_mountinfo = fopen_cloexec("/proc/1/mountinfo", "r"); | |
426 | if (!proc_self_mountinfo) | |
b653309a | 427 | return false; |
33ad9f1a CS |
428 | |
429 | while (getline(&line, &sz, proc_self_mountinfo) != -1) { | |
178938fe | 430 | char *token, *line_tok, *saveptr = NULL; |
33ad9f1a CS |
431 | size_t i, j, k; |
432 | struct cgroup_mount_point *mount_point; | |
433 | struct cgroup_hierarchy *h; | |
434 | char **subsystems; | |
435 | ||
436 | if (line[0] && line[strlen(line) - 1] == '\n') | |
437 | line[strlen(line) - 1] = '\0'; | |
438 | ||
178938fe | 439 | for (i = 0, line_tok = line; (token = strtok_r(line_tok, " ", &saveptr)); line_tok = NULL) { |
33ad9f1a CS |
440 | r = lxc_grow_array((void ***)&tokens, &token_capacity, i + 1, 64); |
441 | if (r < 0) | |
b653309a | 442 | goto out; |
33ad9f1a CS |
443 | tokens[i++] = token; |
444 | } | |
b98f7d6e | 445 | |
33ad9f1a CS |
446 | /* layout of /proc/self/mountinfo: |
447 | * 0: id | |
448 | * 1: parent id | |
449 | * 2: device major:minor | |
450 | * 3: mount prefix | |
8900b9eb | 451 | * 4: mount point |
33ad9f1a CS |
452 | * 5: per-mount options |
453 | * [optional X]: additional data | |
454 | * X+7: "-" | |
455 | * X+8: type | |
456 | * X+9: source | |
457 | * X+10: per-superblock options | |
458 | */ | |
459 | for (j = 6; j < i && tokens[j]; j++) | |
460 | if (!strcmp(tokens[j], "-")) | |
461 | break; | |
fd4f5a56 | 462 | |
33ad9f1a CS |
463 | /* could not find separator */ |
464 | if (j >= i || !tokens[j]) | |
465 | continue; | |
466 | /* there should be exactly three fields after | |
467 | * the separator | |
468 | */ | |
469 | if (i != j + 4) | |
470 | continue; | |
fd4f5a56 | 471 | |
33ad9f1a CS |
472 | /* not a cgroup filesystem */ |
473 | if (strcmp(tokens[j + 1], "cgroup") != 0) | |
474 | continue; | |
b98f7d6e | 475 | |
33ad9f1a CS |
476 | subsystems = subsystems_from_mount_options(tokens[j + 3], kernel_subsystems); |
477 | if (!subsystems) | |
b653309a | 478 | goto out; |
33ad9f1a CS |
479 | |
480 | h = NULL; | |
481 | for (k = 1; k <= meta_data->maximum_hierarchy; k++) { | |
482 | if (meta_data->hierarchies[k] && | |
483 | meta_data->hierarchies[k]->subsystems[0] && | |
484 | lxc_string_in_array(meta_data->hierarchies[k]->subsystems[0], (const char **)subsystems)) { | |
485 | /* TODO: we could also check if the lists really match completely, | |
486 | * just to have an additional sanity check */ | |
487 | h = meta_data->hierarchies[k]; | |
b98f7d6e | 488 | break; |
33ad9f1a | 489 | } |
b98f7d6e | 490 | } |
33ad9f1a CS |
491 | lxc_free_array((void **)subsystems, free); |
492 | ||
493 | r = lxc_grow_array((void ***)&meta_data->mount_points, &mount_point_capacity, mount_point_count + 1, 12); | |
494 | if (r < 0) | |
b653309a | 495 | goto out; |
33ad9f1a CS |
496 | |
497 | /* create mount point object */ | |
498 | mount_point = calloc(1, sizeof(*mount_point)); | |
499 | if (!mount_point) | |
b653309a | 500 | goto out; |
33ad9f1a CS |
501 | |
502 | meta_data->mount_points[mount_point_count++] = mount_point; | |
503 | ||
504 | mount_point->hierarchy = h; | |
505 | mount_point->mount_point = strdup(tokens[4]); | |
506 | mount_point->mount_prefix = strdup(tokens[3]); | |
507 | if (!mount_point->mount_point || !mount_point->mount_prefix) | |
b653309a | 508 | goto out; |
33ad9f1a CS |
509 | mount_point->read_only = !lxc_string_in_list("rw", tokens[5], ','); |
510 | ||
511 | if (!strcmp(mount_point->mount_prefix, "/")) { | |
512 | if (mount_point->read_only) { | |
513 | if (!h->ro_absolute_mount_point) | |
514 | h->ro_absolute_mount_point = mount_point; | |
515 | } else { | |
516 | if (!h->rw_absolute_mount_point) | |
517 | h->rw_absolute_mount_point = mount_point; | |
518 | } | |
b98f7d6e | 519 | } |
ae5c8b8e | 520 | |
33ad9f1a CS |
521 | k = lxc_array_len((void **)h->all_mount_points); |
522 | r = lxc_grow_array((void ***)&h->all_mount_points, &h->all_mount_point_capacity, k + 1, 4); | |
523 | if (r < 0) | |
b653309a | 524 | goto out; |
33ad9f1a | 525 | h->all_mount_points[k] = mount_point; |
fd4f5a56 | 526 | } |
b653309a SH |
527 | bret = true; |
528 | ||
529 | out: | |
b653309a | 530 | fclose(proc_self_mountinfo); |
b653309a | 531 | free(tokens); |
2cdafc54 | 532 | free(line); |
b653309a SH |
533 | return bret; |
534 | } | |
535 | ||
4fb3cba5 | 536 | static struct cgroup_meta_data *lxc_cgroup_load_meta2(const char **subsystem_whitelist) |
b653309a SH |
537 | { |
538 | bool all_kernel_subsystems = true; | |
539 | bool all_named_subsystems = false; | |
540 | struct cgroup_meta_data *meta_data = NULL; | |
541 | char **kernel_subsystems = NULL; | |
542 | int saved_errno = 0; | |
543 | ||
544 | /* if the subsystem whitelist is not specified, include all | |
545 | * hierarchies that contain kernel subsystems by default but | |
546 | * no hierarchies that only contain named subsystems | |
547 | * | |
548 | * if it is specified, the specifier @all will select all | |
549 | * hierarchies, @kernel will select all hierarchies with | |
550 | * kernel subsystems and @named will select all named | |
551 | * hierarchies | |
552 | */ | |
553 | all_kernel_subsystems = subsystem_whitelist ? | |
554 | (lxc_string_in_array("@kernel", subsystem_whitelist) || lxc_string_in_array("@all", subsystem_whitelist)) : | |
555 | true; | |
556 | all_named_subsystems = subsystem_whitelist ? | |
557 | (lxc_string_in_array("@named", subsystem_whitelist) || lxc_string_in_array("@all", subsystem_whitelist)) : | |
558 | false; | |
559 | ||
560 | meta_data = calloc(1, sizeof(struct cgroup_meta_data)); | |
561 | if (!meta_data) | |
562 | return NULL; | |
563 | meta_data->ref = 1; | |
564 | ||
565 | if (!find_cgroup_subsystems(&kernel_subsystems)) | |
566 | goto out_error; | |
567 | ||
568 | if (!find_cgroup_hierarchies(meta_data, all_kernel_subsystems, | |
569 | all_named_subsystems, subsystem_whitelist)) | |
570 | goto out_error; | |
571 | ||
572 | if (!find_hierarchy_mountpts(meta_data, kernel_subsystems)) | |
573 | goto out_error; | |
fd4f5a56 | 574 | |
33ad9f1a CS |
575 | /* oops, we couldn't find anything */ |
576 | if (!meta_data->hierarchies || !meta_data->mount_points) { | |
577 | errno = EINVAL; | |
578 | goto out_error; | |
ae5c8b8e | 579 | } |
fd4f5a56 | 580 | |
3a0abb3a | 581 | lxc_free_array((void **)kernel_subsystems, free); |
33ad9f1a CS |
582 | return meta_data; |
583 | ||
584 | out_error: | |
585 | saved_errno = errno; | |
33ad9f1a CS |
586 | lxc_free_array((void **)kernel_subsystems, free); |
587 | lxc_cgroup_put_meta(meta_data); | |
588 | errno = saved_errno; | |
589 | return NULL; | |
fd4f5a56 DL |
590 | } |
591 | ||
4fb3cba5 | 592 | static struct cgroup_meta_data *lxc_cgroup_get_meta(struct cgroup_meta_data *meta_data) |
e14f67a7 | 593 | { |
33ad9f1a CS |
594 | meta_data->ref++; |
595 | return meta_data; | |
596 | } | |
e14f67a7 | 597 | |
4fb3cba5 | 598 | static struct cgroup_meta_data *lxc_cgroup_put_meta(struct cgroup_meta_data *meta_data) |
33ad9f1a CS |
599 | { |
600 | size_t i; | |
601 | if (!meta_data) | |
602 | return NULL; | |
603 | if (--meta_data->ref > 0) | |
604 | return meta_data; | |
605 | lxc_free_array((void **)meta_data->mount_points, (lxc_free_fn)lxc_cgroup_mount_point_free); | |
606 | if (meta_data->hierarchies) { | |
607 | for (i = 0; i <= meta_data->maximum_hierarchy; i++) | |
608 | lxc_cgroup_hierarchy_free(meta_data->hierarchies[i]); | |
e14f67a7 | 609 | } |
33ad9f1a | 610 | free(meta_data->hierarchies); |
178938fe | 611 | free(meta_data); |
33ad9f1a | 612 | return NULL; |
e14f67a7 U |
613 | } |
614 | ||
4fb3cba5 | 615 | static struct cgroup_hierarchy *lxc_cgroup_find_hierarchy(struct cgroup_meta_data *meta_data, const char *subsystem) |
e14f67a7 | 616 | { |
33ad9f1a CS |
617 | size_t i; |
618 | for (i = 0; i <= meta_data->maximum_hierarchy; i++) { | |
619 | struct cgroup_hierarchy *h = meta_data->hierarchies[i]; | |
620 | if (h && lxc_string_in_array(subsystem, (const char **)h->subsystems)) | |
621 | return h; | |
e14f67a7 | 622 | } |
e14f67a7 U |
623 | return NULL; |
624 | } | |
625 | ||
4fb3cba5 | 626 | static struct cgroup_mount_point *lxc_cgroup_find_mount_point(struct cgroup_hierarchy *hierarchy, const char *group, bool should_be_writable) |
b98f7d6e | 627 | { |
33ad9f1a CS |
628 | struct cgroup_mount_point **mps; |
629 | struct cgroup_mount_point *current_result = NULL; | |
630 | ssize_t quality = -1; | |
b98f7d6e | 631 | |
33ad9f1a CS |
632 | /* trivial case */ |
633 | if (hierarchy->rw_absolute_mount_point) | |
634 | return hierarchy->rw_absolute_mount_point; | |
635 | if (!should_be_writable && hierarchy->ro_absolute_mount_point) | |
636 | return hierarchy->ro_absolute_mount_point; | |
b98f7d6e | 637 | |
33ad9f1a CS |
638 | for (mps = hierarchy->all_mount_points; mps && *mps; mps++) { |
639 | struct cgroup_mount_point *mp = *mps; | |
640 | size_t prefix_len = mp->mount_prefix ? strlen(mp->mount_prefix) : 0; | |
b98f7d6e | 641 | |
33ad9f1a CS |
642 | if (prefix_len == 1 && mp->mount_prefix[0] == '/') |
643 | prefix_len = 0; | |
b98f7d6e | 644 | |
33ad9f1a CS |
645 | if (should_be_writable && mp->read_only) |
646 | continue; | |
647 | ||
648 | if (!prefix_len || | |
649 | (strncmp(group, mp->mount_prefix, prefix_len) == 0 && | |
650 | (group[prefix_len] == '\0' || group[prefix_len] == '/'))) { | |
651 | /* search for the best quality match, i.e. the match with the | |
652 | * shortest prefix where this group is still contained | |
653 | */ | |
654 | if (quality == -1 || prefix_len < quality) { | |
655 | current_result = mp; | |
656 | quality = prefix_len; | |
657 | } | |
b98f7d6e SH |
658 | } |
659 | } | |
660 | ||
33ad9f1a CS |
661 | if (!current_result) |
662 | errno = ENOENT; | |
663 | return current_result; | |
b98f7d6e SH |
664 | } |
665 | ||
4fb3cba5 | 666 | static char *lxc_cgroup_find_abs_path(const char *subsystem, const char *group, bool should_be_writable, const char *suffix) |
b98f7d6e | 667 | { |
33ad9f1a CS |
668 | struct cgroup_meta_data *meta_data; |
669 | struct cgroup_hierarchy *h; | |
670 | struct cgroup_mount_point *mp; | |
671 | char *result; | |
672 | int saved_errno; | |
673 | ||
674 | meta_data = lxc_cgroup_load_meta(); | |
675 | if (!meta_data) | |
676 | return NULL; | |
b98f7d6e | 677 | |
33ad9f1a CS |
678 | h = lxc_cgroup_find_hierarchy(meta_data, subsystem); |
679 | if (!h) | |
680 | goto out_error; | |
b98f7d6e | 681 | |
33ad9f1a CS |
682 | mp = lxc_cgroup_find_mount_point(h, group, should_be_writable); |
683 | if (!mp) | |
684 | goto out_error; | |
b98f7d6e | 685 | |
33ad9f1a CS |
686 | result = cgroup_to_absolute_path(mp, group, suffix); |
687 | if (!result) | |
688 | goto out_error; | |
b98f7d6e | 689 | |
33ad9f1a CS |
690 | lxc_cgroup_put_meta(meta_data); |
691 | return result; | |
b98f7d6e | 692 | |
33ad9f1a CS |
693 | out_error: |
694 | saved_errno = errno; | |
695 | lxc_cgroup_put_meta(meta_data); | |
696 | errno = saved_errno; | |
697 | return NULL; | |
b98f7d6e SH |
698 | } |
699 | ||
4fb3cba5 | 700 | static struct cgroup_process_info *lxc_cgroup_process_info_get(pid_t pid, struct cgroup_meta_data *meta) |
fd4f5a56 | 701 | { |
33ad9f1a CS |
702 | char pid_buf[32]; |
703 | snprintf(pid_buf, 32, "/proc/%lu/cgroup", (unsigned long)pid); | |
704 | return lxc_cgroup_process_info_getx(pid_buf, meta); | |
c8f7c563 CS |
705 | } |
706 | ||
4fb3cba5 | 707 | static struct cgroup_process_info *lxc_cgroup_process_info_get_init(struct cgroup_meta_data *meta) |
c8f7c563 | 708 | { |
33ad9f1a CS |
709 | return lxc_cgroup_process_info_get(1, meta); |
710 | } | |
b98f7d6e | 711 | |
4fb3cba5 | 712 | static struct cgroup_process_info *lxc_cgroup_process_info_get_self(struct cgroup_meta_data *meta) |
33ad9f1a CS |
713 | { |
714 | struct cgroup_process_info *i; | |
715 | i = lxc_cgroup_process_info_getx("/proc/self/cgroup", meta); | |
716 | if (!i) | |
717 | i = lxc_cgroup_process_info_get(getpid(), meta); | |
718 | return i; | |
719 | } | |
ae5c8b8e | 720 | |
692ba18f SH |
721 | /* |
722 | * If a controller has ns cgroup mounted, then in that cgroup the handler->pid | |
723 | * is already in a new cgroup named after the pid. 'mnt' is passed in as | |
724 | * the full current cgroup. Say that is /sys/fs/cgroup/lxc/2975 and the container | |
725 | * name is c1. . We want to rename the cgroup directory to /sys/fs/cgroup/lxc/c1, | |
726 | * and return the string /sys/fs/cgroup/lxc/c1. | |
727 | */ | |
cea0552e | 728 | static char *cgroup_rename_nsgroup(const char *mountpath, const char *oldname, pid_t pid, const char *name) |
692ba18f SH |
729 | { |
730 | char *dir, *fulloldpath; | |
731 | char *newname, *fullnewpath; | |
cea0552e | 732 | int len, newlen, ret; |
692ba18f SH |
733 | |
734 | /* | |
735 | * if cgroup is mounted at /cgroup and task is in cgroup /ab/, pid 2375 and | |
736 | * name is c1, | |
737 | * dir: /ab | |
738 | * fulloldpath = /cgroup/ab/2375 | |
739 | * fullnewpath = /cgroup/ab/c1 | |
740 | * newname = /ab/c1 | |
741 | */ | |
742 | dir = alloca(strlen(oldname) + 1); | |
743 | strcpy(dir, oldname); | |
744 | ||
cea0552e SH |
745 | len = strlen(oldname) + strlen(mountpath) + 22; |
746 | fulloldpath = alloca(len); | |
747 | ret = snprintf(fulloldpath, len, "%s/%s/%ld", mountpath, oldname, (unsigned long)pid); | |
748 | if (ret < 0 || ret >= len) | |
749 | return NULL; | |
692ba18f SH |
750 | |
751 | len = strlen(dir) + strlen(name) + 2; | |
752 | newname = malloc(len); | |
753 | if (!newname) { | |
754 | SYSERROR("Out of memory"); | |
755 | return NULL; | |
756 | } | |
cea0552e SH |
757 | ret = snprintf(newname, len, "%s/%s", dir, name); |
758 | if (ret < 0 || ret >= len) { | |
759 | free(newname); | |
760 | return NULL; | |
761 | } | |
692ba18f | 762 | |
cea0552e SH |
763 | newlen = strlen(mountpath) + len + 2; |
764 | fullnewpath = alloca(newlen); | |
765 | ret = snprintf(fullnewpath, newlen, "%s/%s", mountpath, newname); | |
766 | if (ret < 0 || ret >= newlen) { | |
767 | free(newname); | |
768 | return NULL; | |
769 | } | |
692ba18f SH |
770 | |
771 | if (access(fullnewpath, F_OK) == 0) { | |
772 | if (rmdir(fullnewpath) != 0) { | |
773 | SYSERROR("container cgroup %s already exists.", fullnewpath); | |
774 | free(newname); | |
775 | return NULL; | |
776 | } | |
777 | } | |
778 | if (rename(fulloldpath, fullnewpath)) { | |
779 | SYSERROR("failed to rename cgroup %s->%s", fulloldpath, fullnewpath); | |
780 | free(newname); | |
781 | return NULL; | |
782 | } | |
783 | ||
784 | DEBUG("'%s' renamed to '%s'", oldname, newname); | |
785 | ||
786 | return newname; | |
787 | } | |
788 | ||
33ad9f1a | 789 | /* create a new cgroup */ |
4fb3cba5 | 790 | static struct cgroup_process_info *lxc_cgroupfs_create(const char *name, const char *path_pattern, struct cgroup_meta_data *meta_data, const char *sub_pattern) |
33ad9f1a | 791 | { |
001b026e | 792 | char **cgroup_path_components = NULL; |
33ad9f1a CS |
793 | char **p = NULL; |
794 | char *path_so_far = NULL; | |
795 | char **new_cgroup_paths = NULL; | |
796 | char **new_cgroup_paths_sub = NULL; | |
797 | struct cgroup_mount_point *mp; | |
798 | struct cgroup_hierarchy *h; | |
799 | struct cgroup_process_info *base_info = NULL; | |
800 | struct cgroup_process_info *info_ptr; | |
801 | int saved_errno; | |
802 | int r; | |
803 | unsigned suffix = 0; | |
804 | bool had_sub_pattern = false; | |
805 | size_t i; | |
ae5c8b8e | 806 | |
33ad9f1a CS |
807 | if (!is_valid_cgroup(name)) { |
808 | ERROR("Invalid cgroup name: '%s'", name); | |
809 | errno = EINVAL; | |
810 | return NULL; | |
ae5c8b8e SH |
811 | } |
812 | ||
33ad9f1a CS |
813 | if (!strstr(path_pattern, "%n")) { |
814 | ERROR("Invalid cgroup path pattern: '%s'; contains no %%n for specifying container name", path_pattern); | |
815 | errno = EINVAL; | |
816 | return NULL; | |
817 | } | |
fd37327f | 818 | |
33ad9f1a CS |
819 | /* we will modify the result of this operation directly, |
820 | * so we don't have to copy the data structure | |
821 | */ | |
822 | base_info = (path_pattern[0] == '/') ? | |
823 | lxc_cgroup_process_info_get_init(meta_data) : | |
824 | lxc_cgroup_process_info_get_self(meta_data); | |
825 | if (!base_info) | |
826 | return NULL; | |
c8f7c563 | 827 | |
33ad9f1a CS |
828 | new_cgroup_paths = calloc(meta_data->maximum_hierarchy + 1, sizeof(char *)); |
829 | if (!new_cgroup_paths) | |
830 | goto out_initial_error; | |
831 | ||
832 | new_cgroup_paths_sub = calloc(meta_data->maximum_hierarchy + 1, sizeof(char *)); | |
833 | if (!new_cgroup_paths_sub) | |
834 | goto out_initial_error; | |
835 | ||
836 | /* find mount points we can use */ | |
837 | for (info_ptr = base_info; info_ptr; info_ptr = info_ptr->next) { | |
838 | h = info_ptr->hierarchy; | |
839 | mp = lxc_cgroup_find_mount_point(h, info_ptr->cgroup_path, true); | |
840 | if (!mp) { | |
841 | ERROR("Could not find writable mount point for cgroup hierarchy %d while trying to create cgroup.", h->index); | |
842 | goto out_initial_error; | |
843 | } | |
844 | info_ptr->designated_mount_point = mp; | |
460a1cf0 | 845 | |
692ba18f SH |
846 | if (lxc_string_in_array("ns", (const char **)h->subsystems)) |
847 | continue; | |
2edb53c7 SH |
848 | if (handle_cgroup_settings(mp, info_ptr->cgroup_path) < 0) { |
849 | ERROR("Could not set clone_children to 1 for cpuset hierarchy in parent cgroup."); | |
33ad9f1a | 850 | goto out_initial_error; |
2edb53c7 | 851 | } |
33ad9f1a | 852 | } |
b98f7d6e | 853 | |
33ad9f1a CS |
854 | /* normalize the path */ |
855 | cgroup_path_components = lxc_normalize_path(path_pattern); | |
856 | if (!cgroup_path_components) | |
857 | goto out_initial_error; | |
858 | ||
859 | /* go through the path components to see if we can create them */ | |
860 | for (p = cgroup_path_components; *p || (sub_pattern && !had_sub_pattern); p++) { | |
861 | /* we only want to create the same component with -1, -2, etc. | |
862 | * if the component contains the container name itself, otherwise | |
863 | * it's not an error if it already exists | |
864 | */ | |
865 | char *p_eff = *p ? *p : (char *)sub_pattern; | |
866 | bool contains_name = strstr(p_eff, "%n"); | |
867 | char *current_component = NULL; | |
868 | char *current_subpath = NULL; | |
869 | char *current_entire_path = NULL; | |
870 | char *parts[3]; | |
871 | size_t j = 0; | |
872 | i = 0; | |
873 | ||
874 | /* if we are processing the subpattern, we want to make sure | |
875 | * loop is ended the next time around | |
876 | */ | |
877 | if (!*p) { | |
878 | had_sub_pattern = true; | |
879 | p--; | |
880 | } | |
b98f7d6e | 881 | |
33ad9f1a | 882 | goto find_name_on_this_level; |
4fb3cba5 | 883 | |
33ad9f1a CS |
884 | cleanup_name_on_this_level: |
885 | /* This is reached if we found a name clash. | |
886 | * In that case, remove the cgroup from all previous hierarchies | |
887 | */ | |
888 | for (j = 0, info_ptr = base_info; j < i && info_ptr; info_ptr = info_ptr->next, j++) { | |
603c64c2 | 889 | r = remove_cgroup(info_ptr->designated_mount_point, info_ptr->created_paths[info_ptr->created_paths_count - 1], false); |
33ad9f1a CS |
890 | if (r < 0) |
891 | WARN("could not clean up cgroup we created when trying to create container"); | |
892 | free(info_ptr->created_paths[info_ptr->created_paths_count - 1]); | |
893 | info_ptr->created_paths[--info_ptr->created_paths_count] = NULL; | |
894 | } | |
895 | if (current_component != current_subpath) | |
896 | free(current_subpath); | |
897 | if (current_component != p_eff) | |
898 | free(current_component); | |
899 | current_component = current_subpath = NULL; | |
900 | /* try again with another suffix */ | |
901 | ++suffix; | |
4fb3cba5 | 902 | |
33ad9f1a CS |
903 | find_name_on_this_level: |
904 | /* determine name of the path component we should create */ | |
905 | if (contains_name && suffix > 0) { | |
906 | char *buf = calloc(strlen(name) + 32, 1); | |
907 | if (!buf) | |
908 | goto out_initial_error; | |
909 | snprintf(buf, strlen(name) + 32, "%s-%u", name, suffix); | |
910 | current_component = lxc_string_replace("%n", buf, p_eff); | |
911 | free(buf); | |
912 | } else { | |
913 | current_component = contains_name ? lxc_string_replace("%n", name, p_eff) : p_eff; | |
914 | } | |
915 | parts[0] = path_so_far; | |
916 | parts[1] = current_component; | |
917 | parts[2] = NULL; | |
918 | current_subpath = path_so_far ? lxc_string_join("/", (const char **)parts, false) : current_component; | |
919 | ||
920 | /* Now go through each hierarchy and try to create the | |
921 | * corresponding cgroup | |
922 | */ | |
923 | for (i = 0, info_ptr = base_info; info_ptr; info_ptr = info_ptr->next, i++) { | |
924 | char *parts2[3]; | |
692ba18f SH |
925 | |
926 | if (lxc_string_in_array("ns", (const char **)info_ptr->hierarchy->subsystems)) | |
927 | continue; | |
33ad9f1a CS |
928 | current_entire_path = NULL; |
929 | ||
930 | parts2[0] = !strcmp(info_ptr->cgroup_path, "/") ? "" : info_ptr->cgroup_path; | |
931 | parts2[1] = current_subpath; | |
932 | parts2[2] = NULL; | |
933 | current_entire_path = lxc_string_join("/", (const char **)parts2, false); | |
934 | ||
935 | if (!*p) { | |
936 | /* we are processing the subpath, so only update that one */ | |
937 | free(new_cgroup_paths_sub[i]); | |
938 | new_cgroup_paths_sub[i] = strdup(current_entire_path); | |
939 | if (!new_cgroup_paths_sub[i]) | |
940 | goto cleanup_from_error; | |
941 | } else { | |
942 | /* remember which path was used on this controller */ | |
943 | free(new_cgroup_paths[i]); | |
944 | new_cgroup_paths[i] = strdup(current_entire_path); | |
945 | if (!new_cgroup_paths[i]) | |
946 | goto cleanup_from_error; | |
947 | } | |
fd4f5a56 | 948 | |
33ad9f1a CS |
949 | r = create_cgroup(info_ptr->designated_mount_point, current_entire_path); |
950 | if (r < 0 && errno == EEXIST && contains_name) { | |
951 | /* name clash => try new name with new suffix */ | |
952 | free(current_entire_path); | |
953 | current_entire_path = NULL; | |
954 | goto cleanup_name_on_this_level; | |
955 | } else if (r < 0 && errno != EEXIST) { | |
b38b62a6 | 956 | SYSERROR("Could not create cgroup '%s' in '%s'.", current_entire_path, info_ptr->designated_mount_point->mount_point); |
33ad9f1a CS |
957 | goto cleanup_from_error; |
958 | } else if (r == 0) { | |
959 | /* successfully created */ | |
960 | r = lxc_grow_array((void ***)&info_ptr->created_paths, &info_ptr->created_paths_capacity, info_ptr->created_paths_count + 1, 8); | |
961 | if (r < 0) | |
962 | goto cleanup_from_error; | |
d703c2b1 | 963 | if (!init_cpuset_if_needed(info_ptr->designated_mount_point, current_entire_path)) { |
b38b62a6 | 964 | ERROR("Failed to initialize cpuset for '%s' in '%s'.", current_entire_path, info_ptr->designated_mount_point->mount_point); |
d703c2b1 RV |
965 | goto cleanup_from_error; |
966 | } | |
33ad9f1a CS |
967 | info_ptr->created_paths[info_ptr->created_paths_count++] = current_entire_path; |
968 | } else { | |
969 | /* if we didn't create the cgroup, then we have to make sure that | |
970 | * further cgroups will be created properly | |
971 | */ | |
d703c2b1 | 972 | if (handle_cgroup_settings(info_ptr->designated_mount_point, info_ptr->cgroup_path) < 0) { |
f6ac3b9e | 973 | ERROR("Could not set clone_children to 1 for cpuset hierarchy in pre-existing cgroup."); |
33ad9f1a | 974 | goto cleanup_from_error; |
f6ac3b9e | 975 | } |
d703c2b1 RV |
976 | if (!init_cpuset_if_needed(info_ptr->designated_mount_point, info_ptr->cgroup_path)) { |
977 | ERROR("Failed to initialize cpuset in pre-existing '%s'.", info_ptr->cgroup_path); | |
978 | goto cleanup_from_error; | |
979 | } | |
33ad9f1a CS |
980 | |
981 | /* already existed but path component of pattern didn't contain '%n', | |
982 | * so this is not an error; but then we don't need current_entire_path | |
983 | * anymore... | |
984 | */ | |
985 | free(current_entire_path); | |
986 | current_entire_path = NULL; | |
987 | } | |
988 | } | |
fd4f5a56 | 989 | |
33ad9f1a CS |
990 | /* save path so far */ |
991 | free(path_so_far); | |
992 | path_so_far = strdup(current_subpath); | |
993 | if (!path_so_far) | |
994 | goto cleanup_from_error; | |
995 | ||
996 | /* cleanup */ | |
997 | if (current_component != current_subpath) | |
998 | free(current_subpath); | |
999 | if (current_component != p_eff) | |
1000 | free(current_component); | |
1001 | current_component = current_subpath = NULL; | |
1002 | continue; | |
4fb3cba5 | 1003 | |
33ad9f1a | 1004 | cleanup_from_error: |
ec64264d | 1005 | /* called if an error occurred in the loop, so we |
33ad9f1a CS |
1006 | * do some additional cleanup here |
1007 | */ | |
1008 | saved_errno = errno; | |
1009 | if (current_component != current_subpath) | |
1010 | free(current_subpath); | |
1011 | if (current_component != p_eff) | |
1012 | free(current_component); | |
1013 | free(current_entire_path); | |
1014 | errno = saved_errno; | |
1015 | goto out_initial_error; | |
fd4f5a56 DL |
1016 | } |
1017 | ||
33ad9f1a CS |
1018 | /* we're done, now update the paths */ |
1019 | for (i = 0, info_ptr = base_info; info_ptr; info_ptr = info_ptr->next, i++) { | |
47d8fb3b CS |
1020 | /* ignore legacy 'ns' subsystem here, lxc_cgroup_create_legacy |
1021 | * will take care of it | |
1022 | * Since we do a continue in above loop, new_cgroup_paths[i] is | |
1023 | * unset anyway, as is new_cgroup_paths_sub[i] | |
692ba18f | 1024 | */ |
47d8fb3b CS |
1025 | if (lxc_string_in_array("ns", (const char **)info_ptr->hierarchy->subsystems)) |
1026 | continue; | |
1027 | free(info_ptr->cgroup_path); | |
1028 | info_ptr->cgroup_path = new_cgroup_paths[i]; | |
1029 | info_ptr->cgroup_path_sub = new_cgroup_paths_sub[i]; | |
fd4f5a56 | 1030 | } |
33ad9f1a CS |
1031 | /* don't use lxc_free_array since we used the array members |
1032 | * to store them in our result... | |
1033 | */ | |
1034 | free(new_cgroup_paths); | |
1035 | free(new_cgroup_paths_sub); | |
1036 | free(path_so_far); | |
1037 | lxc_free_array((void **)cgroup_path_components, free); | |
1038 | return base_info; | |
1039 | ||
1040 | out_initial_error: | |
1041 | saved_errno = errno; | |
1042 | free(path_so_far); | |
1043 | lxc_cgroup_process_info_free_and_remove(base_info); | |
1044 | lxc_free_array((void **)new_cgroup_paths, free); | |
1045 | lxc_free_array((void **)new_cgroup_paths_sub, free); | |
1046 | lxc_free_array((void **)cgroup_path_components, free); | |
1047 | errno = saved_errno; | |
1048 | return NULL; | |
c8f7c563 CS |
1049 | } |
1050 | ||
4fb3cba5 | 1051 | static int lxc_cgroup_create_legacy(struct cgroup_process_info *base_info, const char *name, pid_t pid) |
47d8fb3b CS |
1052 | { |
1053 | struct cgroup_process_info *info_ptr; | |
1054 | int r; | |
1055 | ||
1056 | for (info_ptr = base_info; info_ptr; info_ptr = info_ptr->next) { | |
1057 | if (!lxc_string_in_array("ns", (const char **)info_ptr->hierarchy->subsystems)) | |
1058 | continue; | |
1059 | /* | |
1060 | * For any path which has ns cgroup mounted, handler->pid is already | |
1061 | * moved into a container called '%d % (handler->pid)'. Rename it to | |
1062 | * the cgroup name and record that. | |
1063 | */ | |
1064 | char *tmp = cgroup_rename_nsgroup((const char *)info_ptr->designated_mount_point->mount_point, | |
1065 | info_ptr->cgroup_path, pid, name); | |
1066 | if (!tmp) | |
1067 | return -1; | |
1068 | free(info_ptr->cgroup_path); | |
1069 | info_ptr->cgroup_path = tmp; | |
1070 | r = lxc_grow_array((void ***)&info_ptr->created_paths, &info_ptr->created_paths_capacity, info_ptr->created_paths_count + 1, 8); | |
1071 | if (r < 0) | |
1072 | return -1; | |
1073 | tmp = strdup(tmp); | |
1074 | if (!tmp) | |
1075 | return -1; | |
1076 | info_ptr->created_paths[info_ptr->created_paths_count++] = tmp; | |
1077 | } | |
1078 | return 0; | |
1079 | } | |
1080 | ||
33ad9f1a | 1081 | /* get the cgroup membership of a given container */ |
4fb3cba5 | 1082 | static struct cgroup_process_info *lxc_cgroup_get_container_info(const char *name, const char *lxcpath, struct cgroup_meta_data *meta_data) |
c8f7c563 | 1083 | { |
33ad9f1a CS |
1084 | struct cgroup_process_info *result = NULL; |
1085 | int saved_errno = 0; | |
1086 | size_t i; | |
1087 | struct cgroup_process_info **cptr = &result; | |
1088 | struct cgroup_process_info *entry = NULL; | |
1089 | char *path = NULL; | |
1090 | ||
1091 | for (i = 0; i <= meta_data->maximum_hierarchy; i++) { | |
1092 | struct cgroup_hierarchy *h = meta_data->hierarchies[i]; | |
1093 | if (!h || !h->used) | |
1094 | continue; | |
c8f7c563 | 1095 | |
33ad9f1a CS |
1096 | /* use the command interface to look for the cgroup */ |
1097 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->subsystems[0]); | |
c661b0a8 DE |
1098 | if (!path) { |
1099 | h->used = false; | |
1100 | WARN("Not attaching to cgroup %s unknown to %s %s", h->subsystems[0], lxcpath, name); | |
1101 | continue; | |
1102 | } | |
33ad9f1a CS |
1103 | |
1104 | entry = calloc(1, sizeof(struct cgroup_process_info)); | |
1105 | if (!entry) | |
1106 | goto out_error; | |
1107 | entry->meta_ref = lxc_cgroup_get_meta(meta_data); | |
1108 | entry->hierarchy = h; | |
1109 | entry->cgroup_path = path; | |
1110 | path = NULL; | |
1111 | ||
1112 | /* it is not an error if we don't find anything here, | |
1113 | * it is up to the caller to decide what to do in that | |
1114 | * case */ | |
1115 | entry->designated_mount_point = lxc_cgroup_find_mount_point(h, entry->cgroup_path, true); | |
1116 | ||
1117 | *cptr = entry; | |
1118 | cptr = &entry->next; | |
1119 | entry = NULL; | |
c8f7c563 CS |
1120 | } |
1121 | ||
33ad9f1a CS |
1122 | return result; |
1123 | out_error: | |
1124 | saved_errno = errno; | |
1125 | free(path); | |
1126 | lxc_cgroup_process_info_free(result); | |
1127 | lxc_cgroup_process_info_free(entry); | |
1128 | errno = saved_errno; | |
1129 | return NULL; | |
fd4f5a56 DL |
1130 | } |
1131 | ||
33ad9f1a | 1132 | /* move a processs to the cgroups specified by the membership */ |
4fb3cba5 | 1133 | static int lxc_cgroupfs_enter(struct cgroup_process_info *info, pid_t pid, bool enter_sub) |
4f17323e | 1134 | { |
33ad9f1a CS |
1135 | char pid_buf[32]; |
1136 | char *cgroup_tasks_fn; | |
1137 | int r; | |
1138 | struct cgroup_process_info *info_ptr; | |
1139 | ||
1140 | snprintf(pid_buf, 32, "%lu", (unsigned long)pid); | |
1141 | for (info_ptr = info; info_ptr; info_ptr = info_ptr->next) { | |
1142 | char *cgroup_path = (enter_sub && info_ptr->cgroup_path_sub) ? | |
1143 | info_ptr->cgroup_path_sub : | |
1144 | info_ptr->cgroup_path; | |
1145 | ||
1146 | if (!info_ptr->designated_mount_point) { | |
1147 | info_ptr->designated_mount_point = lxc_cgroup_find_mount_point(info_ptr->hierarchy, cgroup_path, true); | |
1148 | if (!info_ptr->designated_mount_point) { | |
1149 | SYSERROR("Could not add pid %lu to cgroup %s: internal error (couldn't find any writable mountpoint to cgroup filesystem)", (unsigned long)pid, cgroup_path); | |
1150 | return -1; | |
1151 | } | |
1152 | } | |
4f17323e | 1153 | |
33ad9f1a CS |
1154 | cgroup_tasks_fn = cgroup_to_absolute_path(info_ptr->designated_mount_point, cgroup_path, "/tasks"); |
1155 | if (!cgroup_tasks_fn) { | |
1156 | SYSERROR("Could not add pid %lu to cgroup %s: internal error", (unsigned long)pid, cgroup_path); | |
1157 | return -1; | |
1158 | } | |
4f17323e | 1159 | |
33ad9f1a | 1160 | r = lxc_write_to_file(cgroup_tasks_fn, pid_buf, strlen(pid_buf), false); |
5903da82 | 1161 | free(cgroup_tasks_fn); |
33ad9f1a CS |
1162 | if (r < 0) { |
1163 | SYSERROR("Could not add pid %lu to cgroup %s: internal error", (unsigned long)pid, cgroup_path); | |
1164 | return -1; | |
1165 | } | |
4f17323e CS |
1166 | } |
1167 | ||
33ad9f1a | 1168 | return 0; |
4f17323e CS |
1169 | } |
1170 | ||
33ad9f1a CS |
1171 | /* free process membership information */ |
1172 | void lxc_cgroup_process_info_free(struct cgroup_process_info *info) | |
fc7de561 | 1173 | { |
33ad9f1a CS |
1174 | struct cgroup_process_info *next; |
1175 | if (!info) | |
b98f7d6e | 1176 | return; |
33ad9f1a CS |
1177 | next = info->next; |
1178 | lxc_cgroup_put_meta(info->meta_ref); | |
1179 | free(info->cgroup_path); | |
1180 | free(info->cgroup_path_sub); | |
1181 | lxc_free_array((void **)info->created_paths, free); | |
1182 | free(info); | |
1183 | lxc_cgroup_process_info_free(next); | |
fc7de561 SH |
1184 | } |
1185 | ||
33ad9f1a CS |
1186 | /* free process membership information and remove cgroups that were created */ |
1187 | void lxc_cgroup_process_info_free_and_remove(struct cgroup_process_info *info) | |
b98f7d6e | 1188 | { |
33ad9f1a CS |
1189 | struct cgroup_process_info *next; |
1190 | char **pp; | |
1191 | if (!info) | |
1192 | return; | |
1193 | next = info->next; | |
603c64c2 | 1194 | { |
33ad9f1a CS |
1195 | struct cgroup_mount_point *mp = info->designated_mount_point; |
1196 | if (!mp) | |
1197 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, true); | |
1198 | if (mp) | |
1199 | /* ignore return value here, perhaps we created the | |
1200 | * '/lxc' cgroup in this container but another container | |
1201 | * is still running (for example) | |
1202 | */ | |
603c64c2 SH |
1203 | (void)remove_cgroup(mp, info->cgroup_path, true); |
1204 | } | |
1205 | for (pp = info->created_paths; pp && *pp; pp++); | |
1206 | for ((void)(pp && --pp); info->created_paths && pp >= info->created_paths; --pp) { | |
33ad9f1a | 1207 | free(*pp); |
b98f7d6e | 1208 | } |
33ad9f1a CS |
1209 | free(info->created_paths); |
1210 | lxc_cgroup_put_meta(info->meta_ref); | |
1211 | free(info->cgroup_path); | |
1212 | free(info->cgroup_path_sub); | |
1213 | free(info); | |
9431aa65 | 1214 | lxc_cgroup_process_info_free_and_remove(next); |
33ad9f1a | 1215 | } |
b98f7d6e | 1216 | |
4fb3cba5 | 1217 | static char *lxc_cgroup_get_hierarchy_path_data(const char *subsystem, struct cgfs_data *d) |
33ad9f1a | 1218 | { |
d4ef7c50 SH |
1219 | struct cgroup_process_info *info = d->info; |
1220 | info = find_info_for_subsystem(info, subsystem); | |
33ad9f1a CS |
1221 | if (!info) |
1222 | return NULL; | |
1223 | return info->cgroup_path; | |
b98f7d6e SH |
1224 | } |
1225 | ||
4fb3cba5 | 1226 | static char *lxc_cgroup_get_hierarchy_abs_path_data(const char *subsystem, struct cgfs_data *d) |
b98f7d6e | 1227 | { |
d4ef7c50 | 1228 | struct cgroup_process_info *info = d->info; |
33ad9f1a | 1229 | struct cgroup_mount_point *mp = NULL; |
d4ef7c50 SH |
1230 | |
1231 | info = find_info_for_subsystem(info, subsystem); | |
33ad9f1a CS |
1232 | if (!info) |
1233 | return NULL; | |
1234 | if (info->designated_mount_point) { | |
8900b9eb | 1235 | mp = info->designated_mount_point; |
33ad9f1a CS |
1236 | } else { |
1237 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, true); | |
1238 | if (!mp) | |
1239 | return NULL; | |
b98f7d6e | 1240 | } |
33ad9f1a | 1241 | return cgroup_to_absolute_path(mp, info->cgroup_path, NULL); |
b98f7d6e | 1242 | } |
55c76589 | 1243 | |
4fb3cba5 | 1244 | static char *lxc_cgroup_get_hierarchy_abs_path(const char *subsystem, const char *name, const char *lxcpath) |
9a93d992 | 1245 | { |
33ad9f1a CS |
1246 | struct cgroup_meta_data *meta; |
1247 | struct cgroup_process_info *base_info, *info; | |
1248 | struct cgroup_mount_point *mp; | |
1249 | char *result = NULL; | |
33ad9f1a CS |
1250 | |
1251 | meta = lxc_cgroup_load_meta(); | |
1252 | if (!meta) | |
9a93d992 | 1253 | return NULL; |
33ad9f1a CS |
1254 | base_info = lxc_cgroup_get_container_info(name, lxcpath, meta); |
1255 | if (!base_info) | |
178938fe | 1256 | goto out1; |
33ad9f1a CS |
1257 | info = find_info_for_subsystem(base_info, subsystem); |
1258 | if (!info) | |
178938fe | 1259 | goto out2; |
33ad9f1a | 1260 | if (info->designated_mount_point) { |
8900b9eb | 1261 | mp = info->designated_mount_point; |
33ad9f1a CS |
1262 | } else { |
1263 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, true); | |
1264 | if (!mp) | |
178938fe | 1265 | goto out3; |
33ad9f1a CS |
1266 | } |
1267 | result = cgroup_to_absolute_path(mp, info->cgroup_path, NULL); | |
178938fe | 1268 | out3: |
178938fe | 1269 | out2: |
33ad9f1a | 1270 | lxc_cgroup_process_info_free(base_info); |
178938fe | 1271 | out1: |
33ad9f1a | 1272 | lxc_cgroup_put_meta(meta); |
33ad9f1a CS |
1273 | return result; |
1274 | } | |
9a93d992 | 1275 | |
4fb3cba5 | 1276 | static int lxc_cgroup_set_data(const char *filename, const char *value, struct cgfs_data *d) |
33ad9f1a CS |
1277 | { |
1278 | char *subsystem = NULL, *p, *path; | |
1279 | int ret = -1; | |
9a93d992 | 1280 | |
33ad9f1a CS |
1281 | subsystem = alloca(strlen(filename) + 1); |
1282 | strcpy(subsystem, filename); | |
46cd2845 | 1283 | if ((p = strchr(subsystem, '.')) != NULL) |
33ad9f1a | 1284 | *p = '\0'; |
9a93d992 | 1285 | |
4fb3cba5 | 1286 | path = lxc_cgroup_get_hierarchy_abs_path_data(subsystem, d); |
33ad9f1a CS |
1287 | if (path) { |
1288 | ret = do_cgroup_set(path, filename, value); | |
1289 | free(path); | |
9a93d992 | 1290 | } |
33ad9f1a CS |
1291 | return ret; |
1292 | } | |
9a93d992 | 1293 | |
4fb3cba5 | 1294 | static int lxc_cgroupfs_set(const char *filename, const char *value, const char *name, const char *lxcpath) |
9a93d992 | 1295 | { |
33ad9f1a CS |
1296 | char *subsystem = NULL, *p, *path; |
1297 | int ret = -1; | |
9a93d992 | 1298 | |
33ad9f1a CS |
1299 | subsystem = alloca(strlen(filename) + 1); |
1300 | strcpy(subsystem, filename); | |
46cd2845 | 1301 | if ((p = strchr(subsystem, '.')) != NULL) |
33ad9f1a | 1302 | *p = '\0'; |
9a93d992 | 1303 | |
33ad9f1a CS |
1304 | path = lxc_cgroup_get_hierarchy_abs_path(subsystem, name, lxcpath); |
1305 | if (path) { | |
1306 | ret = do_cgroup_set(path, filename, value); | |
1307 | free(path); | |
1308 | } | |
b98f7d6e | 1309 | return ret; |
9a93d992 SH |
1310 | } |
1311 | ||
4fb3cba5 | 1312 | static int lxc_cgroupfs_get(const char *filename, char *value, size_t len, const char *name, const char *lxcpath) |
9a93d992 | 1313 | { |
33ad9f1a CS |
1314 | char *subsystem = NULL, *p, *path; |
1315 | int ret = -1; | |
1316 | ||
1317 | subsystem = alloca(strlen(filename) + 1); | |
1318 | strcpy(subsystem, filename); | |
46cd2845 | 1319 | if ((p = strchr(subsystem, '.')) != NULL) |
33ad9f1a CS |
1320 | *p = '\0'; |
1321 | ||
1322 | path = lxc_cgroup_get_hierarchy_abs_path(subsystem, name, lxcpath); | |
1323 | if (path) { | |
1324 | ret = do_cgroup_get(path, filename, value, len); | |
1325 | free(path); | |
9a93d992 | 1326 | } |
33ad9f1a | 1327 | return ret; |
9a93d992 SH |
1328 | } |
1329 | ||
4fb3cba5 | 1330 | static bool cgroupfs_mount_cgroup(void *hdata, const char *root, int type) |
aae1f3c4 CS |
1331 | { |
1332 | size_t bufsz = strlen(root) + sizeof("/sys/fs/cgroup"); | |
1333 | char *path = NULL; | |
1334 | char **parts = NULL; | |
1335 | char *dirname = NULL; | |
1336 | char *abs_path = NULL; | |
1337 | char *abs_path2 = NULL; | |
d4ef7c50 SH |
1338 | struct cgfs_data *cgfs_d; |
1339 | struct cgroup_process_info *info, *base_info; | |
aae1f3c4 CS |
1340 | int r, saved_errno = 0; |
1341 | ||
4fb3cba5 DE |
1342 | cgfs_d = hdata; |
1343 | if (!cgfs_d) | |
1344 | return false; | |
d4ef7c50 SH |
1345 | base_info = cgfs_d->info; |
1346 | ||
0769b82a CS |
1347 | /* If we get passed the _NOSPEC types, we default to _MIXED, since we don't |
1348 | * have access to the lxc_conf object at this point. It really should be up | |
1349 | * to the caller to fix this, but this doesn't really hurt. | |
1350 | */ | |
1351 | if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
1352 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
1353 | else if (type == LXC_AUTO_CGROUP_NOSPEC) | |
1354 | type = LXC_AUTO_CGROUP_MIXED; | |
1355 | ||
7997d7da CS |
1356 | if (type < LXC_AUTO_CGROUP_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) { |
1357 | ERROR("could not mount cgroups into container: invalid type specified internally"); | |
1358 | errno = EINVAL; | |
c476bdce | 1359 | return false; |
7997d7da CS |
1360 | } |
1361 | ||
aae1f3c4 CS |
1362 | path = calloc(1, bufsz); |
1363 | if (!path) | |
c476bdce | 1364 | return false; |
aae1f3c4 CS |
1365 | snprintf(path, bufsz, "%s/sys/fs/cgroup", root); |
1366 | r = mount("cgroup_root", path, "tmpfs", MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_RELATIME, "size=10240k,mode=755"); | |
1367 | if (r < 0) { | |
1368 | SYSERROR("could not mount tmpfs to /sys/fs/cgroup in the container"); | |
c476bdce | 1369 | return false; |
aae1f3c4 CS |
1370 | } |
1371 | ||
1372 | /* now mount all the hierarchies we care about */ | |
1373 | for (info = base_info; info; info = info->next) { | |
1374 | size_t subsystem_count, i; | |
1375 | struct cgroup_mount_point *mp = info->designated_mount_point; | |
1376 | if (!mp) | |
1377 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, true); | |
1378 | if (!mp) { | |
1379 | SYSERROR("could not find original mount point for cgroup hierarchy while trying to mount cgroup filesystem"); | |
1380 | goto out_error; | |
1381 | } | |
1382 | ||
1383 | subsystem_count = lxc_array_len((void **)info->hierarchy->subsystems); | |
1384 | parts = calloc(subsystem_count + 1, sizeof(char *)); | |
1385 | if (!parts) | |
1386 | goto out_error; | |
1387 | ||
1388 | for (i = 0; i < subsystem_count; i++) { | |
1389 | if (!strncmp(info->hierarchy->subsystems[i], "name=", 5)) | |
1390 | parts[i] = info->hierarchy->subsystems[i] + 5; | |
1391 | else | |
1392 | parts[i] = info->hierarchy->subsystems[i]; | |
1393 | } | |
1394 | dirname = lxc_string_join(",", (const char **)parts, false); | |
1395 | if (!dirname) | |
1396 | goto out_error; | |
1397 | ||
1398 | /* create subsystem directory */ | |
1399 | abs_path = lxc_append_paths(path, dirname); | |
1400 | if (!abs_path) | |
1401 | goto out_error; | |
1402 | r = mkdir_p(abs_path, 0755); | |
1403 | if (r < 0 && errno != EEXIST) { | |
1404 | SYSERROR("could not create cgroup subsystem directory /sys/fs/cgroup/%s", dirname); | |
1405 | goto out_error; | |
1406 | } | |
1407 | ||
aae1f3c4 CS |
1408 | abs_path2 = lxc_append_paths(abs_path, info->cgroup_path); |
1409 | if (!abs_path2) | |
1410 | goto out_error; | |
aae1f3c4 | 1411 | |
7997d7da CS |
1412 | if (type == LXC_AUTO_CGROUP_FULL_RO || type == LXC_AUTO_CGROUP_FULL_RW || type == LXC_AUTO_CGROUP_FULL_MIXED) { |
1413 | /* bind-mount the cgroup entire filesystem there */ | |
1414 | if (strcmp(mp->mount_prefix, "/") != 0) { | |
1415 | /* FIXME: maybe we should just try to remount the entire hierarchy | |
1416 | * with a regular mount command? may that works? */ | |
1417 | ERROR("could not automatically mount cgroup-full to /sys/fs/cgroup/%s: host has no mount point for this cgroup filesystem that has access to the root cgroup", dirname); | |
1418 | goto out_error; | |
1419 | } | |
1420 | r = mount(mp->mount_point, abs_path, "none", MS_BIND, 0); | |
1421 | if (r < 0) { | |
1422 | SYSERROR("error bind-mounting %s to %s", mp->mount_point, abs_path); | |
1423 | goto out_error; | |
1424 | } | |
f8f3c3c0 SG |
1425 | /* main cgroup path should be read-only */ |
1426 | if (type == LXC_AUTO_CGROUP_FULL_RO || type == LXC_AUTO_CGROUP_FULL_MIXED) { | |
1427 | r = mount(NULL, abs_path, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY, NULL); | |
1428 | if (r < 0) { | |
1429 | SYSERROR("error re-mounting %s readonly", abs_path); | |
1430 | goto out_error; | |
1431 | } | |
1432 | } | |
7997d7da CS |
1433 | /* own cgroup should be read-write */ |
1434 | if (type == LXC_AUTO_CGROUP_FULL_MIXED) { | |
1435 | r = mount(abs_path2, abs_path2, NULL, MS_BIND, NULL); | |
1436 | if (r < 0) { | |
1437 | SYSERROR("error bind-mounting %s onto itself", abs_path2); | |
1438 | goto out_error; | |
1439 | } | |
1440 | r = mount(NULL, abs_path2, NULL, MS_REMOUNT|MS_BIND, NULL); | |
1441 | if (r < 0) { | |
1442 | SYSERROR("error re-mounting %s readwrite", abs_path2); | |
1443 | goto out_error; | |
1444 | } | |
1445 | } | |
1446 | } else { | |
1447 | /* create path for container's cgroup */ | |
1448 | r = mkdir_p(abs_path2, 0755); | |
1449 | if (r < 0 && errno != EEXIST) { | |
1450 | SYSERROR("could not create cgroup directory /sys/fs/cgroup/%s%s", dirname, info->cgroup_path); | |
1451 | goto out_error; | |
1452 | } | |
aae1f3c4 | 1453 | |
b46f0553 CS |
1454 | /* for read-only and mixed cases, we have to bind-mount the tmpfs directory |
1455 | * that points to the hierarchy itself (i.e. /sys/fs/cgroup/cpu etc.) onto | |
1456 | * itself and then bind-mount it read-only, since we keep the tmpfs itself | |
1457 | * read-write (see comment below) | |
1458 | */ | |
1459 | if (type == LXC_AUTO_CGROUP_MIXED || type == LXC_AUTO_CGROUP_RO) { | |
1460 | r = mount(abs_path, abs_path, NULL, MS_BIND, NULL); | |
1461 | if (r < 0) { | |
1462 | SYSERROR("error bind-mounting %s onto itself", abs_path); | |
1463 | goto out_error; | |
1464 | } | |
1465 | r = mount(NULL, abs_path, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY, NULL); | |
1466 | if (r < 0) { | |
1467 | SYSERROR("error re-mounting %s readonly", abs_path); | |
1468 | goto out_error; | |
1469 | } | |
1470 | } | |
1471 | ||
7997d7da CS |
1472 | free(abs_path); |
1473 | abs_path = NULL; | |
1474 | ||
1475 | /* bind-mount container's cgroup to that directory */ | |
1476 | abs_path = cgroup_to_absolute_path(mp, info->cgroup_path, NULL); | |
1477 | if (!abs_path) | |
1478 | goto out_error; | |
1479 | r = mount(abs_path, abs_path2, "none", MS_BIND, 0); | |
1480 | if (r < 0) { | |
1481 | SYSERROR("error bind-mounting %s to %s", abs_path, abs_path2); | |
1482 | goto out_error; | |
1483 | } | |
1484 | if (type == LXC_AUTO_CGROUP_RO) { | |
1485 | r = mount(NULL, abs_path2, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY, NULL); | |
1486 | if (r < 0) { | |
1487 | SYSERROR("error re-mounting %s readonly", abs_path2); | |
1488 | goto out_error; | |
1489 | } | |
1490 | } | |
aae1f3c4 CS |
1491 | } |
1492 | ||
1493 | free(abs_path); | |
1494 | free(abs_path2); | |
1495 | abs_path = NULL; | |
1496 | abs_path2 = NULL; | |
1497 | ||
1498 | /* add symlinks for every single subsystem */ | |
1499 | if (subsystem_count > 1) { | |
1500 | for (i = 0; i < subsystem_count; i++) { | |
1501 | abs_path = lxc_append_paths(path, parts[i]); | |
1502 | if (!abs_path) | |
1503 | goto out_error; | |
1504 | r = symlink(dirname, abs_path); | |
1505 | if (r < 0) | |
1506 | WARN("could not create symlink %s -> %s in /sys/fs/cgroup of container", parts[i], dirname); | |
1507 | free(abs_path); | |
1508 | abs_path = NULL; | |
1509 | } | |
1510 | } | |
1511 | free(dirname); | |
1512 | free(parts); | |
1513 | dirname = NULL; | |
1514 | parts = NULL; | |
1515 | } | |
1516 | ||
b46f0553 CS |
1517 | /* We used to remount the entire tmpfs readonly if any :ro or |
1518 | * :mixed mode was specified. However, Ubuntu's mountall has the | |
1519 | * unfortunate behavior to block bootup if /sys/fs/cgroup is | |
1520 | * mounted read-only and cannot be remounted read-write. | |
1521 | * (mountall reads /lib/init/fstab and tries to (re-)mount all of | |
1522 | * these if they are not already mounted with the right options; | |
1523 | * it contains an entry for /sys/fs/cgroup. In case it can't do | |
1524 | * that, it prompts for the user to either manually fix it or | |
1525 | * boot anyway. But without user input, booting of the container | |
1526 | * hangs.) | |
1527 | * | |
1528 | * Instead of remounting the entire tmpfs readonly, we only | |
1529 | * remount the paths readonly that are part of the cgroup | |
1530 | * hierarchy. | |
f8f3c3c0 | 1531 | */ |
f8f3c3c0 | 1532 | |
aae1f3c4 CS |
1533 | free(path); |
1534 | ||
c476bdce | 1535 | return true; |
aae1f3c4 CS |
1536 | |
1537 | out_error: | |
1538 | saved_errno = errno; | |
1539 | free(path); | |
1540 | free(dirname); | |
1541 | free(parts); | |
1542 | free(abs_path); | |
1543 | free(abs_path2); | |
1544 | errno = saved_errno; | |
c476bdce | 1545 | return false; |
aae1f3c4 CS |
1546 | } |
1547 | ||
4fb3cba5 | 1548 | static int cgfs_nrtasks(void *hdata) |
33ad9f1a | 1549 | { |
4fb3cba5 DE |
1550 | struct cgfs_data *d = hdata; |
1551 | struct cgroup_process_info *info; | |
33ad9f1a CS |
1552 | struct cgroup_mount_point *mp = NULL; |
1553 | char *abs_path = NULL; | |
1554 | int ret; | |
460a1cf0 | 1555 | |
4fb3cba5 DE |
1556 | if (!d) { |
1557 | errno = ENOENT; | |
1558 | return -1; | |
1559 | } | |
1560 | ||
1561 | info = d->info; | |
33ad9f1a CS |
1562 | if (!info) { |
1563 | errno = ENOENT; | |
1564 | return -1; | |
b98f7d6e | 1565 | } |
c8f7c563 | 1566 | |
33ad9f1a | 1567 | if (info->designated_mount_point) { |
8900b9eb | 1568 | mp = info->designated_mount_point; |
33ad9f1a CS |
1569 | } else { |
1570 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, false); | |
1571 | if (!mp) | |
1572 | return -1; | |
c8f7c563 CS |
1573 | } |
1574 | ||
33ad9f1a CS |
1575 | abs_path = cgroup_to_absolute_path(mp, info->cgroup_path, NULL); |
1576 | if (!abs_path) | |
1577 | return -1; | |
1578 | ||
1579 | ret = cgroup_recursive_task_count(abs_path); | |
1580 | free(abs_path); | |
1581 | return ret; | |
c8f7c563 CS |
1582 | } |
1583 | ||
574c4428 QH |
1584 | static struct cgroup_process_info * |
1585 | lxc_cgroup_process_info_getx(const char *proc_pid_cgroup_str, | |
1586 | struct cgroup_meta_data *meta) | |
d08ba6ec | 1587 | { |
33ad9f1a CS |
1588 | struct cgroup_process_info *result = NULL; |
1589 | FILE *proc_pid_cgroup = NULL; | |
1590 | char *line = NULL; | |
1591 | size_t sz = 0; | |
1592 | int saved_errno = 0; | |
1593 | struct cgroup_process_info **cptr = &result; | |
1594 | struct cgroup_process_info *entry = NULL; | |
1595 | ||
1596 | proc_pid_cgroup = fopen_cloexec(proc_pid_cgroup_str, "r"); | |
1597 | if (!proc_pid_cgroup) | |
b98f7d6e | 1598 | return NULL; |
1ac470c0 | 1599 | |
33ad9f1a CS |
1600 | while (getline(&line, &sz, proc_pid_cgroup) != -1) { |
1601 | /* file format: hierarchy:subsystems:group */ | |
1602 | char *colon1; | |
1603 | char *colon2; | |
1604 | char *endptr; | |
1605 | int hierarchy_number; | |
1606 | struct cgroup_hierarchy *h = NULL; | |
fd4f5a56 | 1607 | |
33ad9f1a | 1608 | if (!line[0]) |
ae5c8b8e | 1609 | continue; |
b98f7d6e | 1610 | |
33ad9f1a CS |
1611 | if (line[strlen(line) - 1] == '\n') |
1612 | line[strlen(line) - 1] = '\0'; | |
1613 | ||
1614 | colon1 = strchr(line, ':'); | |
1615 | if (!colon1) | |
8900b9eb | 1616 | continue; |
33ad9f1a CS |
1617 | *colon1++ = '\0'; |
1618 | colon2 = strchr(colon1, ':'); | |
1619 | if (!colon2) | |
ae5c8b8e | 1620 | continue; |
33ad9f1a | 1621 | *colon2++ = '\0'; |
e4659536 | 1622 | |
33ad9f1a CS |
1623 | endptr = NULL; |
1624 | hierarchy_number = strtoul(line, &endptr, 10); | |
1625 | if (!endptr || *endptr) | |
9a93d992 | 1626 | continue; |
9a93d992 | 1627 | |
33ad9f1a CS |
1628 | if (hierarchy_number > meta->maximum_hierarchy) { |
1629 | /* we encountered a hierarchy we didn't have before, | |
1630 | * so probably somebody remounted some stuff in the | |
1631 | * mean time... | |
1632 | */ | |
1633 | errno = EAGAIN; | |
1634 | goto out_error; | |
b98f7d6e | 1635 | } |
33ad9f1a CS |
1636 | |
1637 | h = meta->hierarchies[hierarchy_number]; | |
1638 | if (!h) { | |
1639 | /* we encountered a hierarchy that was thought to be | |
1640 | * dead before, so probably somebody remounted some | |
1641 | * stuff in the mean time... | |
1642 | */ | |
1643 | errno = EAGAIN; | |
1644 | goto out_error; | |
b98f7d6e | 1645 | } |
33ad9f1a CS |
1646 | |
1647 | /* we are told that we should ignore this hierarchy */ | |
1648 | if (!h->used) | |
b98f7d6e | 1649 | continue; |
5193cc3d | 1650 | |
33ad9f1a CS |
1651 | entry = calloc(1, sizeof(struct cgroup_process_info)); |
1652 | if (!entry) | |
1653 | goto out_error; | |
fd4f5a56 | 1654 | |
33ad9f1a CS |
1655 | entry->meta_ref = lxc_cgroup_get_meta(meta); |
1656 | entry->hierarchy = h; | |
1657 | entry->cgroup_path = strdup(colon2); | |
1658 | if (!entry->cgroup_path) | |
1659 | goto out_error; | |
d08ba6ec | 1660 | |
33ad9f1a CS |
1661 | *cptr = entry; |
1662 | cptr = &entry->next; | |
1663 | entry = NULL; | |
b98f7d6e | 1664 | } |
b98f7d6e | 1665 | |
33ad9f1a CS |
1666 | fclose(proc_pid_cgroup); |
1667 | free(line); | |
1668 | return result; | |
1669 | ||
1670 | out_error: | |
1671 | saved_errno = errno; | |
1672 | if (proc_pid_cgroup) | |
1673 | fclose(proc_pid_cgroup); | |
1674 | lxc_cgroup_process_info_free(result); | |
1675 | lxc_cgroup_process_info_free(entry); | |
1676 | free(line); | |
1677 | errno = saved_errno; | |
ae5c8b8e | 1678 | return NULL; |
36b86299 DL |
1679 | } |
1680 | ||
574c4428 QH |
1681 | static char **subsystems_from_mount_options(const char *mount_options, |
1682 | char **kernel_list) | |
36b86299 | 1683 | { |
33ad9f1a CS |
1684 | char *token, *str, *saveptr = NULL; |
1685 | char **result = NULL; | |
1686 | size_t result_capacity = 0; | |
8900b9eb | 1687 | size_t result_count = 0; |
33ad9f1a CS |
1688 | int saved_errno; |
1689 | int r; | |
ef342abb | 1690 | |
33ad9f1a CS |
1691 | str = alloca(strlen(mount_options)+1); |
1692 | strcpy(str, mount_options); | |
1693 | for (; (token = strtok_r(str, ",", &saveptr)); str = NULL) { | |
1694 | /* we have a subsystem if it's either in the list of | |
1695 | * subsystems provided by the kernel OR if it starts | |
1696 | * with name= for named hierarchies | |
1697 | */ | |
1698 | if (!strncmp(token, "name=", 5) || lxc_string_in_array(token, (const char **)kernel_list)) { | |
1699 | r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 12); | |
1700 | if (r < 0) | |
1701 | goto out_free; | |
1702 | result[result_count + 1] = NULL; | |
1703 | result[result_count] = strdup(token); | |
1704 | if (!result[result_count]) | |
1705 | goto out_free; | |
1706 | result_count++; | |
1707 | } | |
ae5c8b8e | 1708 | } |
f0e64b8b | 1709 | |
33ad9f1a CS |
1710 | return result; |
1711 | ||
1712 | out_free: | |
1713 | saved_errno = errno; | |
1714 | lxc_free_array((void**)result, free); | |
1715 | errno = saved_errno; | |
1716 | return NULL; | |
b98f7d6e SH |
1717 | } |
1718 | ||
574c4428 | 1719 | static void lxc_cgroup_mount_point_free(struct cgroup_mount_point *mp) |
b98f7d6e | 1720 | { |
33ad9f1a CS |
1721 | if (!mp) |
1722 | return; | |
1723 | free(mp->mount_point); | |
1724 | free(mp->mount_prefix); | |
1725 | free(mp); | |
bcbd102c SH |
1726 | } |
1727 | ||
574c4428 | 1728 | static void lxc_cgroup_hierarchy_free(struct cgroup_hierarchy *h) |
341a9bd8 | 1729 | { |
33ad9f1a CS |
1730 | if (!h) |
1731 | return; | |
1732 | lxc_free_array((void **)h->subsystems, free); | |
8bfcb981 | 1733 | free(h->all_mount_points); |
33ad9f1a CS |
1734 | free(h); |
1735 | } | |
341a9bd8 | 1736 | |
574c4428 | 1737 | static bool is_valid_cgroup(const char *name) |
33ad9f1a CS |
1738 | { |
1739 | const char *p; | |
1740 | for (p = name; *p; p++) { | |
28bb9321 QH |
1741 | /* Use the ASCII printable characters range(32 - 127) |
1742 | * is reasonable, we kick out 32(SPACE) because it'll | |
1743 | * break legacy lxc-ls | |
1744 | */ | |
1745 | if (*p <= 32 || *p >= 127 || *p == '/') | |
33ad9f1a | 1746 | return false; |
341a9bd8 | 1747 | } |
33ad9f1a CS |
1748 | return strcmp(name, ".") != 0 && strcmp(name, "..") != 0; |
1749 | } | |
341a9bd8 | 1750 | |
574c4428 QH |
1751 | static int create_or_remove_cgroup(bool do_remove, |
1752 | struct cgroup_mount_point *mp, const char *path, int recurse) | |
33ad9f1a CS |
1753 | { |
1754 | int r, saved_errno = 0; | |
1755 | char *buf = cgroup_to_absolute_path(mp, path, NULL); | |
1756 | if (!buf) | |
1757 | return -1; | |
341a9bd8 | 1758 | |
33ad9f1a | 1759 | /* create or remove directory */ |
603c64c2 SH |
1760 | if (do_remove) { |
1761 | if (recurse) | |
1762 | r = cgroup_rmdir(buf); | |
1763 | else | |
1764 | r = rmdir(buf); | |
1765 | } else | |
1766 | r = mkdir(buf, 0777); | |
33ad9f1a CS |
1767 | saved_errno = errno; |
1768 | free(buf); | |
1769 | errno = saved_errno; | |
1770 | return r; | |
341a9bd8 | 1771 | } |
bcbd102c | 1772 | |
574c4428 | 1773 | static int create_cgroup(struct cgroup_mount_point *mp, const char *path) |
a6ddef61 | 1774 | { |
603c64c2 | 1775 | return create_or_remove_cgroup(false, mp, path, false); |
a6ddef61 MN |
1776 | } |
1777 | ||
574c4428 QH |
1778 | static int remove_cgroup(struct cgroup_mount_point *mp, |
1779 | const char *path, bool recurse) | |
576f946d | 1780 | { |
603c64c2 | 1781 | return create_or_remove_cgroup(true, mp, path, recurse); |
33ad9f1a | 1782 | } |
576f946d | 1783 | |
574c4428 QH |
1784 | static char *cgroup_to_absolute_path(struct cgroup_mount_point *mp, |
1785 | const char *path, const char *suffix) | |
33ad9f1a CS |
1786 | { |
1787 | /* first we have to make sure we subtract the mount point's prefix */ | |
1788 | char *prefix = mp->mount_prefix; | |
1789 | char *buf; | |
1790 | ssize_t len, rv; | |
1791 | ||
1792 | /* we want to make sure only absolute paths to cgroups are passed to us */ | |
1793 | if (path[0] != '/') { | |
1794 | errno = EINVAL; | |
1795 | return NULL; | |
1796 | } | |
b98f7d6e | 1797 | |
33ad9f1a CS |
1798 | if (prefix && !strcmp(prefix, "/")) |
1799 | prefix = NULL; | |
b98f7d6e | 1800 | |
33ad9f1a CS |
1801 | /* prefix doesn't match */ |
1802 | if (prefix && strncmp(prefix, path, strlen(prefix)) != 0) { | |
1803 | errno = EINVAL; | |
1804 | return NULL; | |
1805 | } | |
1806 | /* if prefix is /foo and path is /foobar */ | |
1807 | if (prefix && path[strlen(prefix)] != '/' && path[strlen(prefix)] != '\0') { | |
1808 | errno = EINVAL; | |
1809 | return NULL; | |
1810 | } | |
b98f7d6e | 1811 | |
33ad9f1a CS |
1812 | /* remove prefix from path */ |
1813 | path += prefix ? strlen(prefix) : 0; | |
b98f7d6e | 1814 | |
33ad9f1a CS |
1815 | len = strlen(mp->mount_point) + strlen(path) + (suffix ? strlen(suffix) : 0); |
1816 | buf = calloc(len + 1, 1); | |
50266dc6 DE |
1817 | if (!buf) |
1818 | return NULL; | |
33ad9f1a | 1819 | rv = snprintf(buf, len + 1, "%s%s%s", mp->mount_point, path, suffix ? suffix : ""); |
8900b9eb | 1820 | if (rv > len) { |
33ad9f1a CS |
1821 | free(buf); |
1822 | errno = ENOMEM; | |
8900b9eb | 1823 | return NULL; |
8b92dc3a | 1824 | } |
576f946d | 1825 | |
33ad9f1a | 1826 | return buf; |
e0f888d9 | 1827 | } |
283678ed | 1828 | |
574c4428 QH |
1829 | static struct cgroup_process_info * |
1830 | find_info_for_subsystem(struct cgroup_process_info *info, const char *subsystem) | |
283678ed | 1831 | { |
33ad9f1a CS |
1832 | struct cgroup_process_info *info_ptr; |
1833 | for (info_ptr = info; info_ptr; info_ptr = info_ptr->next) { | |
1834 | struct cgroup_hierarchy *h = info_ptr->hierarchy; | |
1835 | if (lxc_string_in_array(subsystem, (const char **)h->subsystems)) | |
1836 | return info_ptr; | |
b98f7d6e | 1837 | } |
33ad9f1a CS |
1838 | errno = ENOENT; |
1839 | return NULL; | |
1840 | } | |
283678ed | 1841 | |
574c4428 QH |
1842 | static int do_cgroup_get(const char *cgroup_path, const char *sub_filename, |
1843 | char *value, size_t len) | |
33ad9f1a CS |
1844 | { |
1845 | const char *parts[3] = { | |
1846 | cgroup_path, | |
1847 | sub_filename, | |
1848 | NULL | |
1849 | }; | |
1850 | char *filename; | |
1851 | int ret, saved_errno; | |
1852 | ||
1853 | filename = lxc_string_join("/", parts, false); | |
1854 | if (!filename) | |
1855 | return -1; | |
1856 | ||
1857 | ret = lxc_read_from_file(filename, value, len); | |
1858 | saved_errno = errno; | |
1859 | free(filename); | |
1860 | errno = saved_errno; | |
1861 | return ret; | |
283678ed | 1862 | } |
b113383b | 1863 | |
574c4428 QH |
1864 | static int do_cgroup_set(const char *cgroup_path, const char *sub_filename, |
1865 | const char *value) | |
b113383b | 1866 | { |
33ad9f1a CS |
1867 | const char *parts[3] = { |
1868 | cgroup_path, | |
1869 | sub_filename, | |
1870 | NULL | |
1871 | }; | |
1872 | char *filename; | |
1873 | int ret, saved_errno; | |
b113383b | 1874 | |
33ad9f1a CS |
1875 | filename = lxc_string_join("/", parts, false); |
1876 | if (!filename) | |
1877 | return -1; | |
b113383b | 1878 | |
33ad9f1a CS |
1879 | ret = lxc_write_to_file(filename, value, strlen(value), false); |
1880 | saved_errno = errno; | |
1881 | free(filename); | |
1882 | errno = saved_errno; | |
1883 | return ret; | |
b98f7d6e SH |
1884 | } |
1885 | ||
4fb3cba5 | 1886 | static int do_setup_cgroup_limits(struct cgfs_data *d, |
574c4428 | 1887 | struct lxc_list *cgroup_settings, bool do_devices) |
b98f7d6e | 1888 | { |
365d180a | 1889 | struct lxc_list *iterator, *sorted_cgroup_settings, *next; |
b98f7d6e SH |
1890 | struct lxc_cgroup *cg; |
1891 | int ret = -1; | |
1892 | ||
33ad9f1a | 1893 | if (lxc_list_empty(cgroup_settings)) |
b98f7d6e SH |
1894 | return 0; |
1895 | ||
aaf26830 | 1896 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); |
fac7c663 KT |
1897 | if (!sorted_cgroup_settings) { |
1898 | return -1; | |
1899 | } | |
aaf26830 KT |
1900 | |
1901 | lxc_list_for_each(iterator, sorted_cgroup_settings) { | |
b98f7d6e SH |
1902 | cg = iterator->elem; |
1903 | ||
33ad9f1a | 1904 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { |
b98f7d6e | 1905 | if (strcmp(cg->subsystem, "devices.deny") == 0 && |
4fb3cba5 | 1906 | cgroup_devices_has_allow_or_deny(d, cg->value, false)) |
b98f7d6e SH |
1907 | continue; |
1908 | if (strcmp(cg->subsystem, "devices.allow") == 0 && | |
4fb3cba5 | 1909 | cgroup_devices_has_allow_or_deny(d, cg->value, true)) |
b98f7d6e | 1910 | continue; |
4fb3cba5 | 1911 | if (lxc_cgroup_set_data(cg->subsystem, cg->value, d)) { |
959aee9c | 1912 | ERROR("Error setting %s to %s for %s", |
4fb3cba5 | 1913 | cg->subsystem, cg->value, d->name); |
b98f7d6e SH |
1914 | goto out; |
1915 | } | |
b113383b | 1916 | } |
b98f7d6e SH |
1917 | |
1918 | DEBUG("cgroup '%s' set to '%s'", cg->subsystem, cg->value); | |
b113383b SH |
1919 | } |
1920 | ||
b98f7d6e SH |
1921 | ret = 0; |
1922 | INFO("cgroup has been setup"); | |
1923 | out: | |
365d180a | 1924 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
aaf26830 KT |
1925 | lxc_list_del(iterator); |
1926 | free(iterator); | |
1927 | } | |
365d180a | 1928 | free(sorted_cgroup_settings); |
b113383b SH |
1929 | return ret; |
1930 | } | |
b98f7d6e | 1931 | |
4fb3cba5 | 1932 | static bool cgroup_devices_has_allow_or_deny(struct cgfs_data *d, |
574c4428 | 1933 | char *v, bool for_allow) |
33ad9f1a CS |
1934 | { |
1935 | char *path; | |
1936 | FILE *devices_list; | |
8900b9eb | 1937 | char *line = NULL; |
33ad9f1a CS |
1938 | size_t sz = 0; |
1939 | bool ret = !for_allow; | |
1940 | const char *parts[3] = { | |
1941 | NULL, | |
1942 | "devices.list", | |
1943 | NULL | |
1944 | }; | |
1945 | ||
1946 | // XXX FIXME if users could use something other than 'lxc.devices.deny = a'. | |
1947 | // not sure they ever do, but they *could* | |
1948 | // right now, I'm assuming they do NOT | |
1949 | if (!for_allow && strcmp(v, "a") != 0 && strcmp(v, "a *:* rwm") != 0) | |
1950 | return false; | |
1951 | ||
4fb3cba5 | 1952 | parts[0] = (const char *)lxc_cgroup_get_hierarchy_abs_path_data("devices", d); |
33ad9f1a CS |
1953 | if (!parts[0]) |
1954 | return false; | |
1955 | path = lxc_string_join("/", parts, false); | |
1956 | if (!path) { | |
1957 | free((void *)parts[0]); | |
1958 | return false; | |
1959 | } | |
1960 | ||
1961 | devices_list = fopen_cloexec(path, "r"); | |
1962 | if (!devices_list) { | |
1963 | free(path); | |
1964 | return false; | |
1965 | } | |
1966 | ||
1967 | while (getline(&line, &sz, devices_list) != -1) { | |
1968 | size_t len = strlen(line); | |
1969 | if (len > 0 && line[len-1] == '\n') | |
1970 | line[len-1] = '\0'; | |
1971 | if (strcmp(line, "a *:* rwm") == 0) { | |
1972 | ret = for_allow; | |
1973 | goto out; | |
1974 | } else if (for_allow && strcmp(line, v) == 0) { | |
1975 | ret = true; | |
8900b9eb | 1976 | goto out; |
33ad9f1a CS |
1977 | } |
1978 | } | |
1979 | ||
1980 | out: | |
1981 | fclose(devices_list); | |
1982 | free(line); | |
1983 | free(path); | |
1984 | return ret; | |
1985 | } | |
1986 | ||
574c4428 | 1987 | static int cgroup_recursive_task_count(const char *cgroup_path) |
b98f7d6e | 1988 | { |
33ad9f1a CS |
1989 | DIR *d; |
1990 | struct dirent *dent_buf; | |
1991 | struct dirent *dent; | |
8900b9eb | 1992 | ssize_t name_max; |
33ad9f1a CS |
1993 | int n = 0, r; |
1994 | ||
1995 | /* see man readdir_r(3) */ | |
1996 | name_max = pathconf(cgroup_path, _PC_NAME_MAX); | |
1997 | if (name_max <= 0) | |
1998 | name_max = 255; | |
1999 | dent_buf = malloc(offsetof(struct dirent, d_name) + name_max + 1); | |
2000 | if (!dent_buf) | |
2001 | return -1; | |
2002 | ||
2003 | d = opendir(cgroup_path); | |
034ef75d SH |
2004 | if (!d) { |
2005 | free(dent_buf); | |
33ad9f1a | 2006 | return 0; |
034ef75d | 2007 | } |
33ad9f1a CS |
2008 | |
2009 | while (readdir_r(d, dent_buf, &dent) == 0 && dent) { | |
2010 | const char *parts[3] = { | |
2011 | cgroup_path, | |
2012 | dent->d_name, | |
2013 | NULL | |
2014 | }; | |
2015 | char *sub_path; | |
2016 | struct stat st; | |
2017 | ||
2018 | if (!strcmp(dent->d_name, ".") || !strcmp(dent->d_name, "..")) | |
2019 | continue; | |
2020 | sub_path = lxc_string_join("/", parts, false); | |
2021 | if (!sub_path) { | |
2022 | closedir(d); | |
2023 | free(dent_buf); | |
2024 | return -1; | |
2025 | } | |
2026 | r = stat(sub_path, &st); | |
2027 | if (r < 0) { | |
2028 | closedir(d); | |
2029 | free(dent_buf); | |
2030 | free(sub_path); | |
2031 | return -1; | |
2032 | } | |
2033 | if (S_ISDIR(st.st_mode)) { | |
2034 | r = cgroup_recursive_task_count(sub_path); | |
2035 | if (r >= 0) | |
2036 | n += r; | |
2037 | } else if (!strcmp(dent->d_name, "tasks")) { | |
2038 | r = count_lines(sub_path); | |
2039 | if (r >= 0) | |
2040 | n += r; | |
2041 | } | |
2042 | free(sub_path); | |
2043 | } | |
2044 | closedir(d); | |
2045 | free(dent_buf); | |
2046 | ||
2047 | return n; | |
2048 | } | |
2049 | ||
574c4428 | 2050 | static int count_lines(const char *fn) |
33ad9f1a CS |
2051 | { |
2052 | FILE *f; | |
2053 | char *line = NULL; | |
2054 | size_t sz = 0; | |
2055 | int n = 0; | |
2056 | ||
2057 | f = fopen_cloexec(fn, "r"); | |
2058 | if (!f) | |
2059 | return -1; | |
2060 | ||
2061 | while (getline(&line, &sz, f) != -1) { | |
2062 | n++; | |
2063 | } | |
2064 | free(line); | |
2065 | fclose(f); | |
2066 | return n; | |
b98f7d6e SH |
2067 | } |
2068 | ||
574c4428 QH |
2069 | static int handle_cgroup_settings(struct cgroup_mount_point *mp, |
2070 | char *cgroup_path) | |
b98f7d6e | 2071 | { |
33ad9f1a | 2072 | int r, saved_errno = 0; |
7e7243e1 | 2073 | char buf[2]; |
1ea59ad2 | 2074 | |
934b1673 SH |
2075 | mp->need_cpuset_init = false; |
2076 | ||
1ea59ad2 SH |
2077 | /* If this is the memory cgroup, we want to enforce hierarchy. |
2078 | * But don't fail if for some reason we can't. | |
2079 | */ | |
2edb53c7 SH |
2080 | if (lxc_string_in_array("memory", (const char **)mp->hierarchy->subsystems)) { |
2081 | char *cc_path = cgroup_to_absolute_path(mp, cgroup_path, "/memory.use_hierarchy"); | |
2082 | if (cc_path) { | |
2083 | r = lxc_read_from_file(cc_path, buf, 1); | |
2084 | if (r < 1 || buf[0] != '1') { | |
2085 | r = lxc_write_to_file(cc_path, "1", 1, false); | |
2086 | if (r < 0) | |
a8916143 | 2087 | SYSERROR("failed to set memory.use_hierarchy to 1; continuing"); |
2edb53c7 | 2088 | } |
1ea59ad2 SH |
2089 | free(cc_path); |
2090 | } | |
2edb53c7 | 2091 | } |
1ea59ad2 | 2092 | |
33ad9f1a CS |
2093 | /* if this is a cpuset hierarchy, we have to set cgroup.clone_children in |
2094 | * the base cgroup, otherwise containers will start with an empty cpuset.mems | |
2095 | * and cpuset.cpus and then | |
2096 | */ | |
2edb53c7 SH |
2097 | if (lxc_string_in_array("cpuset", (const char **)mp->hierarchy->subsystems)) { |
2098 | char *cc_path = cgroup_to_absolute_path(mp, cgroup_path, "/cgroup.clone_children"); | |
d703c2b1 RV |
2099 | struct stat sb; |
2100 | ||
33ad9f1a | 2101 | if (!cc_path) |
2edb53c7 | 2102 | return -1; |
d703c2b1 RV |
2103 | /* cgroup.clone_children is not available when running under |
2104 | * older kernel versions; in this case, we'll initialize | |
2105 | * cpuset.cpus and cpuset.mems later, after the new cgroup | |
2106 | * was created | |
2107 | */ | |
2108 | if (stat(cc_path, &sb) != 0 && errno == ENOENT) { | |
934b1673 | 2109 | mp->need_cpuset_init = true; |
d703c2b1 RV |
2110 | free(cc_path); |
2111 | return 0; | |
2112 | } | |
7e7243e1 SH |
2113 | r = lxc_read_from_file(cc_path, buf, 1); |
2114 | if (r == 1 && buf[0] == '1') { | |
2115 | free(cc_path); | |
2edb53c7 | 2116 | return 0; |
7e7243e1 | 2117 | } |
33ad9f1a | 2118 | r = lxc_write_to_file(cc_path, "1", 1, false); |
2edb53c7 SH |
2119 | saved_errno = errno; |
2120 | free(cc_path); | |
2121 | errno = saved_errno; | |
2122 | return r < 0 ? -1 : 0; | |
33ad9f1a CS |
2123 | } |
2124 | return 0; | |
b98f7d6e | 2125 | } |
484ed030 | 2126 | |
934b1673 | 2127 | static int cgroup_read_from_file(const char *fn, char buf[], size_t bufsize) |
d703c2b1 RV |
2128 | { |
2129 | int ret = lxc_read_from_file(fn, buf, bufsize); | |
2130 | if (ret < 0) { | |
2131 | SYSERROR("failed to read %s", fn); | |
934b1673 | 2132 | return ret; |
d703c2b1 RV |
2133 | } |
2134 | if (ret == bufsize) { | |
934b1673 SH |
2135 | if (bufsize > 0) { |
2136 | /* obviously this wasn't empty */ | |
2137 | buf[bufsize-1] = '\0'; | |
2138 | return ret; | |
2139 | } | |
2140 | /* Callers don't do this, but regression/sanity check */ | |
2141 | ERROR("%s: was not expecting 0 bufsize", __func__); | |
2142 | return -1; | |
d703c2b1 RV |
2143 | } |
2144 | buf[ret] = '\0'; | |
934b1673 | 2145 | return ret; |
d703c2b1 RV |
2146 | } |
2147 | ||
2148 | static bool do_init_cpuset_file(struct cgroup_mount_point *mp, | |
2149 | const char *path, const char *name) | |
2150 | { | |
934b1673 SH |
2151 | char value[1024]; |
2152 | char *childfile, *parentfile = NULL, *tmp; | |
2153 | int ret; | |
2154 | bool ok = false; | |
2155 | ||
d703c2b1 RV |
2156 | childfile = cgroup_to_absolute_path(mp, path, name); |
2157 | if (!childfile) | |
2158 | return false; | |
2159 | ||
2160 | /* don't overwrite a non-empty value in the file */ | |
934b1673 SH |
2161 | ret = cgroup_read_from_file(childfile, value, sizeof(value)); |
2162 | if (ret < 0) | |
2163 | goto out; | |
d703c2b1 | 2164 | if (value[0] != '\0' && value[0] != '\n') { |
934b1673 SH |
2165 | ok = true; |
2166 | goto out; | |
d703c2b1 RV |
2167 | } |
2168 | ||
2169 | /* path to the same name in the parent cgroup */ | |
2170 | parentfile = strdup(path); | |
2171 | if (!parentfile) | |
934b1673 SH |
2172 | goto out; |
2173 | ||
d703c2b1 | 2174 | tmp = strrchr(parentfile, '/'); |
934b1673 SH |
2175 | if (!tmp) |
2176 | goto out; | |
d703c2b1 RV |
2177 | if (tmp == parentfile) |
2178 | tmp++; /* keep the '/' at the start */ | |
2179 | *tmp = '\0'; | |
2180 | tmp = parentfile; | |
2181 | parentfile = cgroup_to_absolute_path(mp, tmp, name); | |
2182 | free(tmp); | |
934b1673 SH |
2183 | if (!parentfile) |
2184 | goto out; | |
d703c2b1 RV |
2185 | |
2186 | /* copy from parent to child cgroup */ | |
934b1673 SH |
2187 | ret = cgroup_read_from_file(parentfile, value, sizeof(value)); |
2188 | if (ret < 0) | |
2189 | goto out; | |
2190 | if (ret == sizeof(value)) { | |
2191 | /* If anyone actually sees this error, we can address it */ | |
2192 | ERROR("parent cpuset value too long"); | |
2193 | goto out; | |
d703c2b1 RV |
2194 | } |
2195 | ok = (lxc_write_to_file(childfile, value, strlen(value), false) >= 0); | |
2196 | if (!ok) | |
2197 | SYSERROR("failed writing %s", childfile); | |
b1dad6f6 RV |
2198 | |
2199 | out: | |
f10fad2f | 2200 | free(parentfile); |
d703c2b1 | 2201 | free(childfile); |
d703c2b1 RV |
2202 | return ok; |
2203 | } | |
2204 | ||
2205 | static bool init_cpuset_if_needed(struct cgroup_mount_point *mp, | |
2206 | const char *path) | |
2207 | { | |
2208 | /* the files we have to handle here are only in cpuset hierarchies */ | |
2209 | if (!lxc_string_in_array("cpuset", | |
2210 | (const char **)mp->hierarchy->subsystems)) | |
2211 | return true; | |
2212 | ||
b1dad6f6 RV |
2213 | if (!mp->need_cpuset_init) |
2214 | return true; | |
2215 | ||
d703c2b1 RV |
2216 | return (do_init_cpuset_file(mp, path, "/cpuset.cpus") && |
2217 | do_init_cpuset_file(mp, path, "/cpuset.mems") ); | |
2218 | } | |
2219 | ||
4fb3cba5 | 2220 | struct cgroup_ops *cgfs_ops_init(void) |
484ed030 | 2221 | { |
4fb3cba5 | 2222 | return &cgfs_ops; |
d4ef7c50 | 2223 | } |
484ed030 | 2224 | |
4fb3cba5 | 2225 | static void *cgfs_init(const char *name) |
d4ef7c50 | 2226 | { |
4fb3cba5 | 2227 | struct cgfs_data *d; |
484ed030 | 2228 | |
4fb3cba5 DE |
2229 | d = malloc(sizeof(*d)); |
2230 | if (!d) | |
2231 | return NULL; | |
484ed030 | 2232 | |
4fb3cba5 DE |
2233 | memset(d, 0, sizeof(*d)); |
2234 | d->name = strdup(name); | |
2235 | if (!d->name) | |
2236 | goto err1; | |
2237 | ||
5e1c5795 | 2238 | d->cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); |
4fb3cba5 DE |
2239 | |
2240 | d->meta = lxc_cgroup_load_meta(); | |
2241 | if (!d->meta) { | |
2242 | ERROR("cgroupfs failed to detect cgroup metadata"); | |
2243 | goto err2; | |
2244 | } | |
2245 | return d; | |
2246 | ||
2247 | err2: | |
2248 | free(d->name); | |
2249 | err1: | |
2250 | free(d); | |
2251 | return NULL; | |
d4ef7c50 | 2252 | } |
484ed030 | 2253 | |
4fb3cba5 | 2254 | static void cgfs_destroy(void *hdata) |
d4ef7c50 | 2255 | { |
4fb3cba5 DE |
2256 | struct cgfs_data *d = hdata; |
2257 | ||
d4ef7c50 SH |
2258 | if (!d) |
2259 | return; | |
f10fad2f | 2260 | free(d->name); |
c55d4505 ME |
2261 | lxc_cgroup_process_info_free_and_remove(d->info); |
2262 | lxc_cgroup_put_meta(d->meta); | |
d4ef7c50 | 2263 | free(d); |
d4ef7c50 | 2264 | } |
484ed030 | 2265 | |
4fb3cba5 | 2266 | static inline bool cgfs_create(void *hdata) |
d4ef7c50 | 2267 | { |
4fb3cba5 DE |
2268 | struct cgfs_data *d = hdata; |
2269 | struct cgroup_process_info *i; | |
2270 | struct cgroup_meta_data *md; | |
484ed030 | 2271 | |
4fb3cba5 | 2272 | if (!d) |
d4ef7c50 | 2273 | return false; |
4fb3cba5 DE |
2274 | md = d->meta; |
2275 | i = lxc_cgroupfs_create(d->name, d->cgroup_pattern, md, NULL); | |
d4ef7c50 SH |
2276 | if (!i) |
2277 | return false; | |
2278 | d->info = i; | |
2279 | return true; | |
2280 | } | |
484ed030 | 2281 | |
4fb3cba5 | 2282 | static inline bool cgfs_enter(void *hdata, pid_t pid) |
d4ef7c50 | 2283 | { |
4fb3cba5 DE |
2284 | struct cgfs_data *d = hdata; |
2285 | struct cgroup_process_info *i; | |
d4ef7c50 | 2286 | int ret; |
4fb3cba5 DE |
2287 | |
2288 | if (!d) | |
2289 | return false; | |
2290 | i = d->info; | |
2291 | ret = lxc_cgroupfs_enter(i, pid, false); | |
484ed030 | 2292 | |
d4ef7c50 SH |
2293 | return ret == 0; |
2294 | } | |
2295 | ||
4fb3cba5 | 2296 | static inline bool cgfs_create_legacy(void *hdata, pid_t pid) |
d4ef7c50 | 2297 | { |
4fb3cba5 DE |
2298 | struct cgfs_data *d = hdata; |
2299 | struct cgroup_process_info *i; | |
2300 | ||
2301 | if (!d) | |
2302 | return false; | |
2303 | i = d->info; | |
2304 | if (lxc_cgroup_create_legacy(i, d->name, pid) < 0) { | |
2305 | ERROR("failed to create legacy ns cgroups for '%s'", d->name); | |
d4ef7c50 | 2306 | return false; |
484ed030 | 2307 | } |
d4ef7c50 SH |
2308 | return true; |
2309 | } | |
484ed030 | 2310 | |
4fb3cba5 | 2311 | static const char *cgfs_get_cgroup(void *hdata, const char *subsystem) |
d4ef7c50 | 2312 | { |
4fb3cba5 DE |
2313 | struct cgfs_data *d = hdata; |
2314 | ||
2315 | if (!d) | |
2316 | return NULL; | |
2317 | return lxc_cgroup_get_hierarchy_path_data(subsystem, d); | |
484ed030 SH |
2318 | } |
2319 | ||
2ba7a429 TA |
2320 | static const char *cgfs_canonical_path(void *hdata) |
2321 | { | |
2322 | struct cgfs_data *d = hdata; | |
2323 | struct cgroup_process_info *info_ptr; | |
2324 | char *path = NULL; | |
2325 | ||
2326 | if (!d) | |
2327 | return NULL; | |
2328 | ||
2329 | for (info_ptr = d->info; info_ptr; info_ptr = info_ptr->next) { | |
2330 | if (!path) | |
2331 | path = info_ptr->cgroup_path; | |
2332 | else if (strcmp(path, info_ptr->cgroup_path) != 0) { | |
2333 | ERROR("not all paths match %s, %s has path %s", path, | |
2334 | info_ptr->hierarchy->subsystems[0], info_ptr->cgroup_path); | |
2335 | return NULL; | |
2336 | } | |
2337 | } | |
2338 | ||
2339 | return path; | |
2340 | } | |
2341 | ||
4fb3cba5 | 2342 | static bool cgfs_unfreeze(void *hdata) |
0086f499 | 2343 | { |
4fb3cba5 | 2344 | struct cgfs_data *d = hdata; |
0086f499 SH |
2345 | char *cgabspath, *cgrelpath; |
2346 | int ret; | |
2347 | ||
4fb3cba5 DE |
2348 | if (!d) |
2349 | return false; | |
2350 | ||
2351 | cgrelpath = lxc_cgroup_get_hierarchy_path_data("freezer", d); | |
0086f499 SH |
2352 | cgabspath = lxc_cgroup_find_abs_path("freezer", cgrelpath, true, NULL); |
2353 | if (!cgabspath) | |
ecfcb3f0 | 2354 | return false; |
0086f499 SH |
2355 | |
2356 | ret = do_cgroup_set(cgabspath, "freezer.state", "THAWED"); | |
2357 | free(cgabspath); | |
ecfcb3f0 | 2358 | return ret == 0; |
0086f499 SH |
2359 | } |
2360 | ||
4fb3cba5 DE |
2361 | static bool cgroupfs_setup_limits(void *hdata, struct lxc_list *cgroup_conf, |
2362 | bool with_devices) | |
9daf6f5d | 2363 | { |
4fb3cba5 DE |
2364 | struct cgfs_data *d = hdata; |
2365 | ||
2366 | if (!d) | |
2367 | return false; | |
2368 | return do_setup_cgroup_limits(d, cgroup_conf, with_devices) == 0; | |
9daf6f5d SH |
2369 | } |
2370 | ||
4fb3cba5 | 2371 | static bool lxc_cgroupfs_attach(const char *name, const char *lxcpath, pid_t pid) |
5d897655 SH |
2372 | { |
2373 | struct cgroup_meta_data *meta_data; | |
2374 | struct cgroup_process_info *container_info; | |
2375 | int ret; | |
2376 | ||
2377 | meta_data = lxc_cgroup_load_meta(); | |
2378 | if (!meta_data) { | |
2379 | ERROR("could not move attached process %d to cgroup of container", pid); | |
2380 | return false; | |
2381 | } | |
2382 | ||
2383 | container_info = lxc_cgroup_get_container_info(name, lxcpath, meta_data); | |
2384 | lxc_cgroup_put_meta(meta_data); | |
2385 | if (!container_info) { | |
2386 | ERROR("could not move attached process %d to cgroup of container", pid); | |
2387 | return false; | |
2388 | } | |
2389 | ||
2390 | ret = lxc_cgroupfs_enter(container_info, pid, false); | |
2391 | lxc_cgroup_process_info_free(container_info); | |
2392 | if (ret < 0) { | |
2393 | ERROR("could not move attached process %d to cgroup of container", pid); | |
2394 | return false; | |
2395 | } | |
2396 | return true; | |
2397 | } | |
2398 | ||
d4ef7c50 | 2399 | static struct cgroup_ops cgfs_ops = { |
d4ef7c50 | 2400 | .init = cgfs_init, |
4fb3cba5 | 2401 | .destroy = cgfs_destroy, |
d4ef7c50 SH |
2402 | .create = cgfs_create, |
2403 | .enter = cgfs_enter, | |
2404 | .create_legacy = cgfs_create_legacy, | |
2405 | .get_cgroup = cgfs_get_cgroup, | |
2ba7a429 | 2406 | .canonical_path = cgfs_canonical_path, |
d4ef7c50 SH |
2407 | .get = lxc_cgroupfs_get, |
2408 | .set = lxc_cgroupfs_set, | |
4fb3cba5 | 2409 | .unfreeze = cgfs_unfreeze, |
9daf6f5d | 2410 | .setup_limits = cgroupfs_setup_limits, |
d4ef7c50 | 2411 | .name = "cgroupfs", |
5d897655 | 2412 | .attach = lxc_cgroupfs_attach, |
0996e18a | 2413 | .chown = NULL, |
c476bdce | 2414 | .mount_cgroup = cgroupfs_mount_cgroup, |
4fb3cba5 | 2415 | .nrtasks = cgfs_nrtasks, |
23befb18 | 2416 | .driver = CGFS, |
d4ef7c50 | 2417 | }; |