]>
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> |
8b276860 | 32 | #include <grp.h> |
b98f7d6e | 33 | #include <ctype.h> |
576f946d | 34 | #include <sys/types.h> |
35 | #include <sys/stat.h> | |
36 | #include <sys/param.h> | |
37 | #include <sys/inotify.h> | |
aae1f3c4 | 38 | #include <sys/mount.h> |
576f946d | 39 | #include <netinet/in.h> |
40 | #include <net/if.h> | |
41 | ||
e2bcd7db | 42 | #include "error.h" |
ae5c8b8e | 43 | #include "commands.h" |
b98f7d6e SH |
44 | #include "list.h" |
45 | #include "conf.h" | |
33ad9f1a | 46 | #include "utils.h" |
4ec31c52 | 47 | #include "bdev/bdev.h" |
f2363e38 ÇO |
48 | #include "log.h" |
49 | #include "cgroup.h" | |
50 | #include "start.h" | |
484ed030 | 51 | #include "state.h" |
36eb9bde | 52 | |
edaf8b1b SG |
53 | #if IS_BIONIC |
54 | #include <../include/lxcmntent.h> | |
55 | #else | |
56 | #include <mntent.h> | |
57 | #endif | |
58 | ||
4fb3cba5 DE |
59 | struct cgroup_hierarchy; |
60 | struct cgroup_meta_data; | |
61 | struct cgroup_mount_point; | |
62 | ||
63 | /* | |
64 | * cgroup_meta_data: the metadata about the cgroup infrastructure on this | |
65 | * host | |
66 | */ | |
67 | struct cgroup_meta_data { | |
68 | ptrdiff_t ref; /* simple refcount */ | |
69 | struct cgroup_hierarchy **hierarchies; | |
70 | struct cgroup_mount_point **mount_points; | |
71 | int maximum_hierarchy; | |
72 | }; | |
73 | ||
74 | /* | |
75 | * cgroup_hierarchy: describes a single cgroup hierarchy | |
76 | * (may have multiple mount points) | |
77 | */ | |
78 | struct cgroup_hierarchy { | |
79 | int index; | |
80 | bool used; /* false if the hierarchy should be ignored by lxc */ | |
81 | char **subsystems; | |
82 | struct cgroup_mount_point *rw_absolute_mount_point; | |
83 | struct cgroup_mount_point *ro_absolute_mount_point; | |
84 | struct cgroup_mount_point **all_mount_points; | |
85 | size_t all_mount_point_capacity; | |
86 | }; | |
87 | ||
88 | /* | |
89 | * cgroup_mount_point: a mount point to where a hierarchy | |
90 | * is mounted to | |
91 | */ | |
92 | struct cgroup_mount_point { | |
93 | struct cgroup_hierarchy *hierarchy; | |
94 | char *mount_point; | |
95 | char *mount_prefix; | |
96 | bool read_only; | |
97 | bool need_cpuset_init; | |
98 | }; | |
99 | ||
100 | /* | |
101 | * cgroup_process_info: describes the membership of a | |
102 | * process to the different cgroup | |
103 | * hierarchies | |
104 | * | |
105 | * Note this is the per-process info tracked by the cgfs_ops. | |
106 | * This is not used with cgmanager. | |
107 | */ | |
108 | struct cgroup_process_info { | |
109 | struct cgroup_process_info *next; | |
110 | struct cgroup_meta_data *meta_ref; | |
111 | struct cgroup_hierarchy *hierarchy; | |
112 | char *cgroup_path; | |
113 | char *cgroup_path_sub; | |
114 | char **created_paths; | |
115 | size_t created_paths_capacity; | |
116 | size_t created_paths_count; | |
117 | struct cgroup_mount_point *designated_mount_point; | |
118 | }; | |
119 | ||
120 | struct cgfs_data { | |
121 | char *name; | |
122 | const char *cgroup_pattern; | |
123 | struct cgroup_meta_data *meta; | |
124 | struct cgroup_process_info *info; | |
125 | }; | |
126 | ||
127 | lxc_log_define(lxc_cgfs, lxc); | |
576f946d | 128 | |
33ad9f1a CS |
129 | static struct cgroup_process_info *lxc_cgroup_process_info_getx(const char *proc_pid_cgroup_str, struct cgroup_meta_data *meta); |
130 | static char **subsystems_from_mount_options(const char *mount_options, char **kernel_list); | |
131 | static void lxc_cgroup_mount_point_free(struct cgroup_mount_point *mp); | |
132 | static void lxc_cgroup_hierarchy_free(struct cgroup_hierarchy *h); | |
133 | static bool is_valid_cgroup(const char *name); | |
33ad9f1a | 134 | static int create_cgroup(struct cgroup_mount_point *mp, const char *path); |
6a9e0f26 SH |
135 | static int remove_cgroup(struct cgroup_mount_point *mp, const char *path, bool recurse, |
136 | struct lxc_conf *conf); | |
33ad9f1a CS |
137 | static char *cgroup_to_absolute_path(struct cgroup_mount_point *mp, const char *path, const char *suffix); |
138 | static struct cgroup_process_info *find_info_for_subsystem(struct cgroup_process_info *info, const char *subsystem); | |
139 | static int do_cgroup_get(const char *cgroup_path, const char *sub_filename, char *value, size_t len); | |
140 | static int do_cgroup_set(const char *cgroup_path, const char *sub_filename, const char *value); | |
4fb3cba5 DE |
141 | static bool cgroup_devices_has_allow_or_deny(struct cgfs_data *d, char *v, bool for_allow); |
142 | static int do_setup_cgroup_limits(struct cgfs_data *d, struct lxc_list *cgroup_settings, bool do_devices); | |
33ad9f1a CS |
143 | static int cgroup_recursive_task_count(const char *cgroup_path); |
144 | static int count_lines(const char *fn); | |
1ea59ad2 | 145 | static int handle_cgroup_settings(struct cgroup_mount_point *mp, char *cgroup_path); |
d703c2b1 | 146 | static bool init_cpuset_if_needed(struct cgroup_mount_point *mp, const char *path); |
33ad9f1a | 147 | |
4fb3cba5 DE |
148 | static struct cgroup_meta_data *lxc_cgroup_load_meta2(const char **subsystem_whitelist); |
149 | static struct cgroup_meta_data *lxc_cgroup_get_meta(struct cgroup_meta_data *meta_data); | |
150 | static struct cgroup_meta_data *lxc_cgroup_put_meta(struct cgroup_meta_data *meta_data); | |
151 | ||
152 | /* free process membership information */ | |
153 | static void lxc_cgroup_process_info_free(struct cgroup_process_info *info); | |
6a9e0f26 SH |
154 | static void lxc_cgroup_process_info_free_and_remove(struct cgroup_process_info *info, |
155 | struct lxc_conf *conf); | |
4fb3cba5 | 156 | |
d4ef7c50 | 157 | static struct cgroup_ops cgfs_ops; |
d4ef7c50 | 158 | |
603c64c2 SH |
159 | static int cgroup_rmdir(char *dirname) |
160 | { | |
161 | struct dirent dirent, *direntp; | |
162 | int saved_errno = 0; | |
163 | DIR *dir; | |
164 | int ret, failed=0; | |
165 | char pathname[MAXPATHLEN]; | |
166 | ||
167 | dir = opendir(dirname); | |
168 | if (!dir) { | |
169 | ERROR("%s: failed to open %s", __func__, dirname); | |
170 | return -1; | |
171 | } | |
172 | ||
173 | while (!readdir_r(dir, &dirent, &direntp)) { | |
174 | struct stat mystat; | |
175 | int rc; | |
176 | ||
177 | if (!direntp) | |
178 | break; | |
179 | ||
180 | if (!strcmp(direntp->d_name, ".") || | |
181 | !strcmp(direntp->d_name, "..")) | |
182 | continue; | |
183 | ||
184 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
185 | if (rc < 0 || rc >= MAXPATHLEN) { | |
186 | ERROR("pathname too long"); | |
187 | failed=1; | |
188 | if (!saved_errno) | |
189 | saved_errno = -ENOMEM; | |
190 | continue; | |
191 | } | |
192 | ret = lstat(pathname, &mystat); | |
193 | if (ret) { | |
194 | SYSERROR("%s: failed to stat %s", __func__, pathname); | |
195 | failed=1; | |
196 | if (!saved_errno) | |
197 | saved_errno = errno; | |
198 | continue; | |
199 | } | |
200 | if (S_ISDIR(mystat.st_mode)) { | |
201 | if (cgroup_rmdir(pathname) < 0) { | |
202 | if (!saved_errno) | |
203 | saved_errno = errno; | |
204 | failed=1; | |
205 | } | |
206 | } | |
207 | } | |
208 | ||
209 | if (rmdir(dirname) < 0) { | |
210 | SYSERROR("%s: failed to delete %s", __func__, dirname); | |
211 | if (!saved_errno) | |
212 | saved_errno = errno; | |
213 | failed=1; | |
214 | } | |
215 | ||
216 | ret = closedir(dir); | |
217 | if (ret) { | |
218 | SYSERROR("%s: failed to close directory %s", __func__, dirname); | |
219 | if (!saved_errno) | |
220 | saved_errno = errno; | |
221 | failed=1; | |
222 | } | |
223 | ||
224 | errno = saved_errno; | |
225 | return failed ? -1 : 0; | |
226 | } | |
227 | ||
6a9e0f26 SH |
228 | static int rmdir_wrapper(void *data) |
229 | { | |
230 | char *path = data; | |
231 | ||
232 | if (setresgid(0,0,0) < 0) | |
233 | SYSERROR("Failed to setgid to 0"); | |
234 | if (setresuid(0,0,0) < 0) | |
235 | SYSERROR("Failed to setuid to 0"); | |
236 | if (setgroups(0, NULL) < 0) | |
237 | SYSERROR("Failed to clear groups"); | |
238 | ||
239 | return cgroup_rmdir(path); | |
240 | } | |
241 | ||
4fb3cba5 | 242 | static struct cgroup_meta_data *lxc_cgroup_load_meta() |
33ad9f1a CS |
243 | { |
244 | const char *cgroup_use = NULL; | |
245 | char **cgroup_use_list = NULL; | |
246 | struct cgroup_meta_data *md = NULL; | |
247 | int saved_errno; | |
248 | ||
249 | errno = 0; | |
593e8478 | 250 | cgroup_use = lxc_global_config_value("lxc.cgroup.use"); |
33ad9f1a CS |
251 | if (!cgroup_use && errno != 0) |
252 | return NULL; | |
253 | if (cgroup_use) { | |
254 | cgroup_use_list = lxc_string_split_and_trim(cgroup_use, ','); | |
255 | if (!cgroup_use_list) | |
256 | return NULL; | |
257 | } | |
576f946d | 258 | |
33ad9f1a CS |
259 | md = lxc_cgroup_load_meta2((const char **)cgroup_use_list); |
260 | saved_errno = errno; | |
261 | lxc_free_array((void **)cgroup_use_list, free); | |
262 | errno = saved_errno; | |
263 | return md; | |
264 | } | |
fd37327f | 265 | |
b653309a | 266 | /* Step 1: determine all kernel subsystems */ |
4fb3cba5 | 267 | static bool find_cgroup_subsystems(char ***kernel_subsystems) |
1d39a065 | 268 | { |
b653309a SH |
269 | FILE *proc_cgroups; |
270 | bool bret = false; | |
33ad9f1a CS |
271 | char *line = NULL; |
272 | size_t sz = 0; | |
b653309a SH |
273 | size_t kernel_subsystems_count = 0; |
274 | size_t kernel_subsystems_capacity = 0; | |
275 | int r; | |
1d39a065 | 276 | |
33ad9f1a CS |
277 | proc_cgroups = fopen_cloexec("/proc/cgroups", "r"); |
278 | if (!proc_cgroups) | |
b653309a | 279 | return false; |
1d39a065 | 280 | |
33ad9f1a CS |
281 | while (getline(&line, &sz, proc_cgroups) != -1) { |
282 | char *tab1; | |
283 | char *tab2; | |
284 | int hierarchy_number; | |
1d39a065 | 285 | |
33ad9f1a CS |
286 | if (line[0] == '#') |
287 | continue; | |
288 | if (!line[0]) | |
289 | continue; | |
1d39a065 | 290 | |
33ad9f1a CS |
291 | tab1 = strchr(line, '\t'); |
292 | if (!tab1) | |
8900b9eb | 293 | continue; |
33ad9f1a CS |
294 | *tab1++ = '\0'; |
295 | tab2 = strchr(tab1, '\t'); | |
296 | if (!tab2) | |
297 | continue; | |
298 | *tab2 = '\0'; | |
fd37327f | 299 | |
33ad9f1a CS |
300 | tab2 = NULL; |
301 | hierarchy_number = strtoul(tab1, &tab2, 10); | |
302 | if (!tab2 || *tab2) | |
303 | continue; | |
304 | (void)hierarchy_number; | |
305 | ||
b653309a | 306 | r = lxc_grow_array((void ***)kernel_subsystems, &kernel_subsystems_capacity, kernel_subsystems_count + 1, 12); |
33ad9f1a | 307 | if (r < 0) |
b653309a SH |
308 | goto out; |
309 | (*kernel_subsystems)[kernel_subsystems_count] = strdup(line); | |
310 | if (!(*kernel_subsystems)[kernel_subsystems_count]) | |
311 | goto out; | |
33ad9f1a | 312 | kernel_subsystems_count++; |
bcbd102c | 313 | } |
b653309a | 314 | bret = true; |
0d9f8e18 | 315 | |
b653309a | 316 | out: |
33ad9f1a | 317 | fclose(proc_cgroups); |
0ccf7c2a | 318 | free(line); |
b653309a SH |
319 | return bret; |
320 | } | |
321 | ||
322 | /* Step 2: determine all hierarchies (by reading /proc/self/cgroup), | |
323 | * since mount points don't specify hierarchy number and | |
324 | * /proc/cgroups does not contain named hierarchies | |
325 | */ | |
326 | static bool find_cgroup_hierarchies(struct cgroup_meta_data *meta_data, | |
327 | bool all_kernel_subsystems, bool all_named_subsystems, | |
328 | const char **subsystem_whitelist) | |
329 | { | |
330 | FILE *proc_self_cgroup; | |
331 | char *line = NULL; | |
332 | size_t sz = 0; | |
333 | int r; | |
334 | bool bret = false; | |
335 | size_t hierarchy_capacity = 0; | |
ef6e34ee | 336 | |
33ad9f1a CS |
337 | proc_self_cgroup = fopen_cloexec("/proc/self/cgroup", "r"); |
338 | /* if for some reason (because of setns() and pid namespace for example), | |
339 | * /proc/self is not valid, we try /proc/1/cgroup... */ | |
340 | if (!proc_self_cgroup) | |
341 | proc_self_cgroup = fopen_cloexec("/proc/1/cgroup", "r"); | |
342 | if (!proc_self_cgroup) | |
b653309a | 343 | return false; |
33ad9f1a CS |
344 | |
345 | while (getline(&line, &sz, proc_self_cgroup) != -1) { | |
346 | /* file format: hierarchy:subsystems:group, | |
347 | * we only extract hierarchy and subsystems | |
348 | * here */ | |
349 | char *colon1; | |
350 | char *colon2; | |
351 | int hierarchy_number; | |
352 | struct cgroup_hierarchy *h = NULL; | |
353 | char **p; | |
354 | ||
355 | if (!line[0]) | |
356 | continue; | |
ad08bbb7 | 357 | |
33ad9f1a CS |
358 | colon1 = strchr(line, ':'); |
359 | if (!colon1) | |
8900b9eb | 360 | continue; |
33ad9f1a CS |
361 | *colon1++ = '\0'; |
362 | colon2 = strchr(colon1, ':'); | |
363 | if (!colon2) | |
364 | continue; | |
365 | *colon2 = '\0'; | |
ad08bbb7 | 366 | |
33ad9f1a CS |
367 | colon2 = NULL; |
368 | hierarchy_number = strtoul(line, &colon2, 10); | |
369 | if (!colon2 || *colon2) | |
370 | continue; | |
576f946d | 371 | |
33ad9f1a CS |
372 | if (hierarchy_number > meta_data->maximum_hierarchy) { |
373 | /* lxc_grow_array will never shrink, so even if we find a lower | |
374 | * hierarchy number here, the array will never be smaller | |
375 | */ | |
376 | r = lxc_grow_array((void ***)&meta_data->hierarchies, &hierarchy_capacity, hierarchy_number + 1, 12); | |
377 | if (r < 0) | |
b653309a | 378 | goto out; |
5193cc3d | 379 | |
33ad9f1a CS |
380 | meta_data->maximum_hierarchy = hierarchy_number; |
381 | } | |
fd37327f | 382 | |
33ad9f1a CS |
383 | /* this shouldn't happen, we had this already */ |
384 | if (meta_data->hierarchies[hierarchy_number]) | |
b653309a | 385 | goto out; |
33ad9f1a CS |
386 | |
387 | h = calloc(1, sizeof(struct cgroup_hierarchy)); | |
388 | if (!h) | |
b653309a | 389 | goto out; |
33ad9f1a CS |
390 | |
391 | meta_data->hierarchies[hierarchy_number] = h; | |
392 | ||
393 | h->index = hierarchy_number; | |
394 | h->subsystems = lxc_string_split_and_trim(colon1, ','); | |
395 | if (!h->subsystems) | |
b653309a | 396 | goto out; |
33ad9f1a CS |
397 | /* see if this hierarchy should be considered */ |
398 | if (!all_kernel_subsystems || !all_named_subsystems) { | |
399 | for (p = h->subsystems; *p; p++) { | |
400 | if (!strncmp(*p, "name=", 5)) { | |
401 | if (all_named_subsystems || (subsystem_whitelist && lxc_string_in_array(*p, subsystem_whitelist))) { | |
402 | h->used = true; | |
403 | break; | |
404 | } | |
405 | } else { | |
406 | if (all_kernel_subsystems || (subsystem_whitelist && lxc_string_in_array(*p, subsystem_whitelist))) { | |
407 | h->used = true; | |
408 | break; | |
409 | } | |
410 | } | |
411 | } | |
412 | } else { | |
413 | /* we want all hierarchy anyway */ | |
414 | h->used = true; | |
ae5c8b8e | 415 | } |
ae5c8b8e | 416 | } |
b653309a | 417 | bret = true; |
0b9c21ab | 418 | |
b653309a | 419 | out: |
33ad9f1a | 420 | fclose(proc_self_cgroup); |
0ccf7c2a | 421 | free(line); |
b653309a SH |
422 | return bret; |
423 | } | |
424 | ||
425 | /* Step 3: determine all mount points of each hierarchy */ | |
426 | static bool find_hierarchy_mountpts( struct cgroup_meta_data *meta_data, char **kernel_subsystems) | |
427 | { | |
428 | bool bret = false; | |
429 | FILE *proc_self_mountinfo; | |
430 | char *line = NULL; | |
431 | size_t sz = 0; | |
432 | char **tokens = NULL; | |
433 | size_t mount_point_count = 0; | |
434 | size_t mount_point_capacity = 0; | |
435 | size_t token_capacity = 0; | |
436 | int r; | |
fcca16bc | 437 | bool is_cgns = cgns_supported(); |
b653309a | 438 | |
33ad9f1a CS |
439 | proc_self_mountinfo = fopen_cloexec("/proc/self/mountinfo", "r"); |
440 | /* if for some reason (because of setns() and pid namespace for example), | |
441 | * /proc/self is not valid, we try /proc/1/cgroup... */ | |
442 | if (!proc_self_mountinfo) | |
443 | proc_self_mountinfo = fopen_cloexec("/proc/1/mountinfo", "r"); | |
444 | if (!proc_self_mountinfo) | |
b653309a | 445 | return false; |
33ad9f1a CS |
446 | |
447 | while (getline(&line, &sz, proc_self_mountinfo) != -1) { | |
178938fe | 448 | char *token, *line_tok, *saveptr = NULL; |
33ad9f1a CS |
449 | size_t i, j, k; |
450 | struct cgroup_mount_point *mount_point; | |
451 | struct cgroup_hierarchy *h; | |
452 | char **subsystems; | |
836514a8 | 453 | bool is_lxcfs = false; |
33ad9f1a CS |
454 | |
455 | if (line[0] && line[strlen(line) - 1] == '\n') | |
456 | line[strlen(line) - 1] = '\0'; | |
457 | ||
178938fe | 458 | for (i = 0, line_tok = line; (token = strtok_r(line_tok, " ", &saveptr)); line_tok = NULL) { |
33ad9f1a CS |
459 | r = lxc_grow_array((void ***)&tokens, &token_capacity, i + 1, 64); |
460 | if (r < 0) | |
b653309a | 461 | goto out; |
33ad9f1a CS |
462 | tokens[i++] = token; |
463 | } | |
b98f7d6e | 464 | |
33ad9f1a CS |
465 | /* layout of /proc/self/mountinfo: |
466 | * 0: id | |
467 | * 1: parent id | |
468 | * 2: device major:minor | |
469 | * 3: mount prefix | |
8900b9eb | 470 | * 4: mount point |
33ad9f1a CS |
471 | * 5: per-mount options |
472 | * [optional X]: additional data | |
473 | * X+7: "-" | |
474 | * X+8: type | |
475 | * X+9: source | |
476 | * X+10: per-superblock options | |
477 | */ | |
478 | for (j = 6; j < i && tokens[j]; j++) | |
479 | if (!strcmp(tokens[j], "-")) | |
480 | break; | |
fd4f5a56 | 481 | |
33ad9f1a CS |
482 | /* could not find separator */ |
483 | if (j >= i || !tokens[j]) | |
484 | continue; | |
485 | /* there should be exactly three fields after | |
486 | * the separator | |
487 | */ | |
488 | if (i != j + 4) | |
489 | continue; | |
fd4f5a56 | 490 | |
33ad9f1a | 491 | /* not a cgroup filesystem */ |
836514a8 U |
492 | if (strcmp(tokens[j + 1], "cgroup") != 0) { |
493 | if (strcmp(tokens[j + 1], "fuse.lxcfs") != 0) | |
494 | continue; | |
495 | if (strncmp(tokens[4], "/sys/fs/cgroup/", 15) != 0) | |
496 | continue; | |
497 | is_lxcfs = true; | |
498 | char *curtok = tokens[4] + 15; | |
499 | subsystems = subsystems_from_mount_options(curtok, | |
500 | kernel_subsystems); | |
501 | } else | |
502 | subsystems = subsystems_from_mount_options(tokens[j + 3], | |
503 | kernel_subsystems); | |
33ad9f1a | 504 | if (!subsystems) |
b653309a | 505 | goto out; |
33ad9f1a CS |
506 | |
507 | h = NULL; | |
508 | for (k = 1; k <= meta_data->maximum_hierarchy; k++) { | |
509 | if (meta_data->hierarchies[k] && | |
510 | meta_data->hierarchies[k]->subsystems[0] && | |
511 | lxc_string_in_array(meta_data->hierarchies[k]->subsystems[0], (const char **)subsystems)) { | |
512 | /* TODO: we could also check if the lists really match completely, | |
513 | * just to have an additional sanity check */ | |
514 | h = meta_data->hierarchies[k]; | |
b98f7d6e | 515 | break; |
33ad9f1a | 516 | } |
b98f7d6e | 517 | } |
33ad9f1a CS |
518 | lxc_free_array((void **)subsystems, free); |
519 | ||
520 | r = lxc_grow_array((void ***)&meta_data->mount_points, &mount_point_capacity, mount_point_count + 1, 12); | |
521 | if (r < 0) | |
b653309a | 522 | goto out; |
33ad9f1a CS |
523 | |
524 | /* create mount point object */ | |
525 | mount_point = calloc(1, sizeof(*mount_point)); | |
526 | if (!mount_point) | |
b653309a | 527 | goto out; |
33ad9f1a CS |
528 | |
529 | meta_data->mount_points[mount_point_count++] = mount_point; | |
530 | ||
531 | mount_point->hierarchy = h; | |
fcca16bc | 532 | if (is_lxcfs || is_cgns) |
836514a8 U |
533 | mount_point->mount_prefix = strdup("/"); |
534 | else | |
535 | mount_point->mount_prefix = strdup(tokens[3]); | |
33ad9f1a | 536 | mount_point->mount_point = strdup(tokens[4]); |
33ad9f1a | 537 | if (!mount_point->mount_point || !mount_point->mount_prefix) |
b653309a | 538 | goto out; |
33ad9f1a CS |
539 | mount_point->read_only = !lxc_string_in_list("rw", tokens[5], ','); |
540 | ||
541 | if (!strcmp(mount_point->mount_prefix, "/")) { | |
542 | if (mount_point->read_only) { | |
543 | if (!h->ro_absolute_mount_point) | |
544 | h->ro_absolute_mount_point = mount_point; | |
545 | } else { | |
546 | if (!h->rw_absolute_mount_point) | |
547 | h->rw_absolute_mount_point = mount_point; | |
548 | } | |
b98f7d6e | 549 | } |
ae5c8b8e | 550 | |
33ad9f1a CS |
551 | k = lxc_array_len((void **)h->all_mount_points); |
552 | r = lxc_grow_array((void ***)&h->all_mount_points, &h->all_mount_point_capacity, k + 1, 4); | |
553 | if (r < 0) | |
b653309a | 554 | goto out; |
33ad9f1a | 555 | h->all_mount_points[k] = mount_point; |
fd4f5a56 | 556 | } |
b653309a SH |
557 | bret = true; |
558 | ||
559 | out: | |
b653309a | 560 | fclose(proc_self_mountinfo); |
b653309a | 561 | free(tokens); |
2cdafc54 | 562 | free(line); |
b653309a SH |
563 | return bret; |
564 | } | |
565 | ||
4fb3cba5 | 566 | static struct cgroup_meta_data *lxc_cgroup_load_meta2(const char **subsystem_whitelist) |
b653309a SH |
567 | { |
568 | bool all_kernel_subsystems = true; | |
569 | bool all_named_subsystems = false; | |
570 | struct cgroup_meta_data *meta_data = NULL; | |
571 | char **kernel_subsystems = NULL; | |
572 | int saved_errno = 0; | |
573 | ||
574 | /* if the subsystem whitelist is not specified, include all | |
575 | * hierarchies that contain kernel subsystems by default but | |
576 | * no hierarchies that only contain named subsystems | |
577 | * | |
578 | * if it is specified, the specifier @all will select all | |
579 | * hierarchies, @kernel will select all hierarchies with | |
580 | * kernel subsystems and @named will select all named | |
581 | * hierarchies | |
582 | */ | |
583 | all_kernel_subsystems = subsystem_whitelist ? | |
584 | (lxc_string_in_array("@kernel", subsystem_whitelist) || lxc_string_in_array("@all", subsystem_whitelist)) : | |
585 | true; | |
586 | all_named_subsystems = subsystem_whitelist ? | |
587 | (lxc_string_in_array("@named", subsystem_whitelist) || lxc_string_in_array("@all", subsystem_whitelist)) : | |
79c59e6b | 588 | true; |
b653309a SH |
589 | |
590 | meta_data = calloc(1, sizeof(struct cgroup_meta_data)); | |
591 | if (!meta_data) | |
592 | return NULL; | |
593 | meta_data->ref = 1; | |
594 | ||
595 | if (!find_cgroup_subsystems(&kernel_subsystems)) | |
596 | goto out_error; | |
597 | ||
598 | if (!find_cgroup_hierarchies(meta_data, all_kernel_subsystems, | |
599 | all_named_subsystems, subsystem_whitelist)) | |
600 | goto out_error; | |
601 | ||
602 | if (!find_hierarchy_mountpts(meta_data, kernel_subsystems)) | |
603 | goto out_error; | |
fd4f5a56 | 604 | |
33ad9f1a CS |
605 | /* oops, we couldn't find anything */ |
606 | if (!meta_data->hierarchies || !meta_data->mount_points) { | |
607 | errno = EINVAL; | |
608 | goto out_error; | |
ae5c8b8e | 609 | } |
fd4f5a56 | 610 | |
3a0abb3a | 611 | lxc_free_array((void **)kernel_subsystems, free); |
33ad9f1a CS |
612 | return meta_data; |
613 | ||
614 | out_error: | |
615 | saved_errno = errno; | |
33ad9f1a CS |
616 | lxc_free_array((void **)kernel_subsystems, free); |
617 | lxc_cgroup_put_meta(meta_data); | |
618 | errno = saved_errno; | |
619 | return NULL; | |
fd4f5a56 DL |
620 | } |
621 | ||
4fb3cba5 | 622 | static struct cgroup_meta_data *lxc_cgroup_get_meta(struct cgroup_meta_data *meta_data) |
e14f67a7 | 623 | { |
33ad9f1a CS |
624 | meta_data->ref++; |
625 | return meta_data; | |
626 | } | |
e14f67a7 | 627 | |
4fb3cba5 | 628 | static struct cgroup_meta_data *lxc_cgroup_put_meta(struct cgroup_meta_data *meta_data) |
33ad9f1a CS |
629 | { |
630 | size_t i; | |
631 | if (!meta_data) | |
632 | return NULL; | |
633 | if (--meta_data->ref > 0) | |
634 | return meta_data; | |
635 | lxc_free_array((void **)meta_data->mount_points, (lxc_free_fn)lxc_cgroup_mount_point_free); | |
636 | if (meta_data->hierarchies) { | |
637 | for (i = 0; i <= meta_data->maximum_hierarchy; i++) | |
638 | lxc_cgroup_hierarchy_free(meta_data->hierarchies[i]); | |
e14f67a7 | 639 | } |
33ad9f1a | 640 | free(meta_data->hierarchies); |
178938fe | 641 | free(meta_data); |
33ad9f1a | 642 | return NULL; |
e14f67a7 U |
643 | } |
644 | ||
4fb3cba5 | 645 | static struct cgroup_hierarchy *lxc_cgroup_find_hierarchy(struct cgroup_meta_data *meta_data, const char *subsystem) |
e14f67a7 | 646 | { |
33ad9f1a CS |
647 | size_t i; |
648 | for (i = 0; i <= meta_data->maximum_hierarchy; i++) { | |
649 | struct cgroup_hierarchy *h = meta_data->hierarchies[i]; | |
650 | if (h && lxc_string_in_array(subsystem, (const char **)h->subsystems)) | |
651 | return h; | |
e14f67a7 | 652 | } |
e14f67a7 U |
653 | return NULL; |
654 | } | |
655 | ||
d3f99e96 SH |
656 | static bool mountpoint_is_accessible(struct cgroup_mount_point *mp) |
657 | { | |
658 | return mp && access(mp->mount_point, F_OK) == 0; | |
659 | } | |
660 | ||
4fb3cba5 | 661 | static struct cgroup_mount_point *lxc_cgroup_find_mount_point(struct cgroup_hierarchy *hierarchy, const char *group, bool should_be_writable) |
b98f7d6e | 662 | { |
33ad9f1a CS |
663 | struct cgroup_mount_point **mps; |
664 | struct cgroup_mount_point *current_result = NULL; | |
665 | ssize_t quality = -1; | |
b98f7d6e | 666 | |
33ad9f1a | 667 | /* trivial case */ |
d3f99e96 | 668 | if (mountpoint_is_accessible(hierarchy->rw_absolute_mount_point)) |
33ad9f1a | 669 | return hierarchy->rw_absolute_mount_point; |
d3f99e96 | 670 | if (!should_be_writable && mountpoint_is_accessible(hierarchy->ro_absolute_mount_point)) |
33ad9f1a | 671 | return hierarchy->ro_absolute_mount_point; |
b98f7d6e | 672 | |
33ad9f1a CS |
673 | for (mps = hierarchy->all_mount_points; mps && *mps; mps++) { |
674 | struct cgroup_mount_point *mp = *mps; | |
675 | size_t prefix_len = mp->mount_prefix ? strlen(mp->mount_prefix) : 0; | |
b98f7d6e | 676 | |
33ad9f1a CS |
677 | if (prefix_len == 1 && mp->mount_prefix[0] == '/') |
678 | prefix_len = 0; | |
b98f7d6e | 679 | |
d3f99e96 SH |
680 | if (!mountpoint_is_accessible(mp)) |
681 | continue; | |
682 | ||
33ad9f1a CS |
683 | if (should_be_writable && mp->read_only) |
684 | continue; | |
685 | ||
686 | if (!prefix_len || | |
687 | (strncmp(group, mp->mount_prefix, prefix_len) == 0 && | |
688 | (group[prefix_len] == '\0' || group[prefix_len] == '/'))) { | |
689 | /* search for the best quality match, i.e. the match with the | |
690 | * shortest prefix where this group is still contained | |
691 | */ | |
692 | if (quality == -1 || prefix_len < quality) { | |
693 | current_result = mp; | |
694 | quality = prefix_len; | |
695 | } | |
b98f7d6e SH |
696 | } |
697 | } | |
698 | ||
33ad9f1a CS |
699 | if (!current_result) |
700 | errno = ENOENT; | |
701 | return current_result; | |
b98f7d6e SH |
702 | } |
703 | ||
4fb3cba5 | 704 | static char *lxc_cgroup_find_abs_path(const char *subsystem, const char *group, bool should_be_writable, const char *suffix) |
b98f7d6e | 705 | { |
33ad9f1a CS |
706 | struct cgroup_meta_data *meta_data; |
707 | struct cgroup_hierarchy *h; | |
708 | struct cgroup_mount_point *mp; | |
709 | char *result; | |
710 | int saved_errno; | |
711 | ||
712 | meta_data = lxc_cgroup_load_meta(); | |
713 | if (!meta_data) | |
714 | return NULL; | |
b98f7d6e | 715 | |
33ad9f1a CS |
716 | h = lxc_cgroup_find_hierarchy(meta_data, subsystem); |
717 | if (!h) | |
718 | goto out_error; | |
b98f7d6e | 719 | |
33ad9f1a CS |
720 | mp = lxc_cgroup_find_mount_point(h, group, should_be_writable); |
721 | if (!mp) | |
722 | goto out_error; | |
b98f7d6e | 723 | |
33ad9f1a CS |
724 | result = cgroup_to_absolute_path(mp, group, suffix); |
725 | if (!result) | |
726 | goto out_error; | |
b98f7d6e | 727 | |
33ad9f1a CS |
728 | lxc_cgroup_put_meta(meta_data); |
729 | return result; | |
b98f7d6e | 730 | |
33ad9f1a CS |
731 | out_error: |
732 | saved_errno = errno; | |
733 | lxc_cgroup_put_meta(meta_data); | |
734 | errno = saved_errno; | |
735 | return NULL; | |
b98f7d6e SH |
736 | } |
737 | ||
4fb3cba5 | 738 | static struct cgroup_process_info *lxc_cgroup_process_info_get(pid_t pid, struct cgroup_meta_data *meta) |
fd4f5a56 | 739 | { |
33ad9f1a CS |
740 | char pid_buf[32]; |
741 | snprintf(pid_buf, 32, "/proc/%lu/cgroup", (unsigned long)pid); | |
742 | return lxc_cgroup_process_info_getx(pid_buf, meta); | |
c8f7c563 CS |
743 | } |
744 | ||
4fb3cba5 | 745 | static struct cgroup_process_info *lxc_cgroup_process_info_get_init(struct cgroup_meta_data *meta) |
c8f7c563 | 746 | { |
33ad9f1a CS |
747 | return lxc_cgroup_process_info_get(1, meta); |
748 | } | |
b98f7d6e | 749 | |
4fb3cba5 | 750 | static struct cgroup_process_info *lxc_cgroup_process_info_get_self(struct cgroup_meta_data *meta) |
33ad9f1a CS |
751 | { |
752 | struct cgroup_process_info *i; | |
753 | i = lxc_cgroup_process_info_getx("/proc/self/cgroup", meta); | |
754 | if (!i) | |
755 | i = lxc_cgroup_process_info_get(getpid(), meta); | |
756 | return i; | |
757 | } | |
ae5c8b8e | 758 | |
692ba18f SH |
759 | /* |
760 | * If a controller has ns cgroup mounted, then in that cgroup the handler->pid | |
761 | * is already in a new cgroup named after the pid. 'mnt' is passed in as | |
762 | * the full current cgroup. Say that is /sys/fs/cgroup/lxc/2975 and the container | |
763 | * name is c1. . We want to rename the cgroup directory to /sys/fs/cgroup/lxc/c1, | |
764 | * and return the string /sys/fs/cgroup/lxc/c1. | |
765 | */ | |
cea0552e | 766 | static char *cgroup_rename_nsgroup(const char *mountpath, const char *oldname, pid_t pid, const char *name) |
692ba18f SH |
767 | { |
768 | char *dir, *fulloldpath; | |
769 | char *newname, *fullnewpath; | |
cea0552e | 770 | int len, newlen, ret; |
692ba18f SH |
771 | |
772 | /* | |
773 | * if cgroup is mounted at /cgroup and task is in cgroup /ab/, pid 2375 and | |
774 | * name is c1, | |
775 | * dir: /ab | |
776 | * fulloldpath = /cgroup/ab/2375 | |
777 | * fullnewpath = /cgroup/ab/c1 | |
778 | * newname = /ab/c1 | |
779 | */ | |
780 | dir = alloca(strlen(oldname) + 1); | |
781 | strcpy(dir, oldname); | |
782 | ||
cea0552e SH |
783 | len = strlen(oldname) + strlen(mountpath) + 22; |
784 | fulloldpath = alloca(len); | |
785 | ret = snprintf(fulloldpath, len, "%s/%s/%ld", mountpath, oldname, (unsigned long)pid); | |
786 | if (ret < 0 || ret >= len) | |
787 | return NULL; | |
692ba18f SH |
788 | |
789 | len = strlen(dir) + strlen(name) + 2; | |
790 | newname = malloc(len); | |
791 | if (!newname) { | |
792 | SYSERROR("Out of memory"); | |
793 | return NULL; | |
794 | } | |
cea0552e SH |
795 | ret = snprintf(newname, len, "%s/%s", dir, name); |
796 | if (ret < 0 || ret >= len) { | |
797 | free(newname); | |
798 | return NULL; | |
799 | } | |
692ba18f | 800 | |
cea0552e SH |
801 | newlen = strlen(mountpath) + len + 2; |
802 | fullnewpath = alloca(newlen); | |
803 | ret = snprintf(fullnewpath, newlen, "%s/%s", mountpath, newname); | |
804 | if (ret < 0 || ret >= newlen) { | |
805 | free(newname); | |
806 | return NULL; | |
807 | } | |
692ba18f SH |
808 | |
809 | if (access(fullnewpath, F_OK) == 0) { | |
810 | if (rmdir(fullnewpath) != 0) { | |
811 | SYSERROR("container cgroup %s already exists.", fullnewpath); | |
812 | free(newname); | |
813 | return NULL; | |
814 | } | |
815 | } | |
816 | if (rename(fulloldpath, fullnewpath)) { | |
817 | SYSERROR("failed to rename cgroup %s->%s", fulloldpath, fullnewpath); | |
818 | free(newname); | |
819 | return NULL; | |
820 | } | |
821 | ||
822 | DEBUG("'%s' renamed to '%s'", oldname, newname); | |
823 | ||
824 | return newname; | |
825 | } | |
826 | ||
ea439aac SH |
827 | static bool is_crucial_hierarchy(struct cgroup_hierarchy *h) |
828 | { | |
829 | char **p; | |
830 | ||
831 | for (p = h->subsystems; *p; p++) { | |
832 | if (is_crucial_cgroup_subsystem(*p)) | |
833 | return true; | |
834 | } | |
835 | return false; | |
836 | } | |
837 | ||
33ad9f1a | 838 | /* create a new cgroup */ |
4fb3cba5 | 839 | 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 | 840 | { |
001b026e | 841 | char **cgroup_path_components = NULL; |
33ad9f1a CS |
842 | char **p = NULL; |
843 | char *path_so_far = NULL; | |
844 | char **new_cgroup_paths = NULL; | |
845 | char **new_cgroup_paths_sub = NULL; | |
846 | struct cgroup_mount_point *mp; | |
847 | struct cgroup_hierarchy *h; | |
848 | struct cgroup_process_info *base_info = NULL; | |
849 | struct cgroup_process_info *info_ptr; | |
850 | int saved_errno; | |
851 | int r; | |
852 | unsigned suffix = 0; | |
853 | bool had_sub_pattern = false; | |
854 | size_t i; | |
ae5c8b8e | 855 | |
33ad9f1a CS |
856 | if (!is_valid_cgroup(name)) { |
857 | ERROR("Invalid cgroup name: '%s'", name); | |
858 | errno = EINVAL; | |
859 | return NULL; | |
ae5c8b8e SH |
860 | } |
861 | ||
33ad9f1a CS |
862 | if (!strstr(path_pattern, "%n")) { |
863 | ERROR("Invalid cgroup path pattern: '%s'; contains no %%n for specifying container name", path_pattern); | |
864 | errno = EINVAL; | |
865 | return NULL; | |
866 | } | |
fd37327f | 867 | |
33ad9f1a CS |
868 | /* we will modify the result of this operation directly, |
869 | * so we don't have to copy the data structure | |
870 | */ | |
871 | base_info = (path_pattern[0] == '/') ? | |
872 | lxc_cgroup_process_info_get_init(meta_data) : | |
873 | lxc_cgroup_process_info_get_self(meta_data); | |
874 | if (!base_info) | |
875 | return NULL; | |
c8f7c563 | 876 | |
33ad9f1a CS |
877 | new_cgroup_paths = calloc(meta_data->maximum_hierarchy + 1, sizeof(char *)); |
878 | if (!new_cgroup_paths) | |
879 | goto out_initial_error; | |
880 | ||
881 | new_cgroup_paths_sub = calloc(meta_data->maximum_hierarchy + 1, sizeof(char *)); | |
882 | if (!new_cgroup_paths_sub) | |
883 | goto out_initial_error; | |
884 | ||
885 | /* find mount points we can use */ | |
886 | for (info_ptr = base_info; info_ptr; info_ptr = info_ptr->next) { | |
887 | h = info_ptr->hierarchy; | |
888 | mp = lxc_cgroup_find_mount_point(h, info_ptr->cgroup_path, true); | |
889 | if (!mp) { | |
890 | ERROR("Could not find writable mount point for cgroup hierarchy %d while trying to create cgroup.", h->index); | |
891 | goto out_initial_error; | |
892 | } | |
893 | info_ptr->designated_mount_point = mp; | |
460a1cf0 | 894 | |
692ba18f SH |
895 | if (lxc_string_in_array("ns", (const char **)h->subsystems)) |
896 | continue; | |
2edb53c7 SH |
897 | if (handle_cgroup_settings(mp, info_ptr->cgroup_path) < 0) { |
898 | ERROR("Could not set clone_children to 1 for cpuset hierarchy in parent cgroup."); | |
33ad9f1a | 899 | goto out_initial_error; |
2edb53c7 | 900 | } |
33ad9f1a | 901 | } |
b98f7d6e | 902 | |
33ad9f1a CS |
903 | /* normalize the path */ |
904 | cgroup_path_components = lxc_normalize_path(path_pattern); | |
905 | if (!cgroup_path_components) | |
906 | goto out_initial_error; | |
907 | ||
908 | /* go through the path components to see if we can create them */ | |
909 | for (p = cgroup_path_components; *p || (sub_pattern && !had_sub_pattern); p++) { | |
910 | /* we only want to create the same component with -1, -2, etc. | |
911 | * if the component contains the container name itself, otherwise | |
912 | * it's not an error if it already exists | |
913 | */ | |
914 | char *p_eff = *p ? *p : (char *)sub_pattern; | |
915 | bool contains_name = strstr(p_eff, "%n"); | |
916 | char *current_component = NULL; | |
917 | char *current_subpath = NULL; | |
918 | char *current_entire_path = NULL; | |
919 | char *parts[3]; | |
920 | size_t j = 0; | |
921 | i = 0; | |
922 | ||
923 | /* if we are processing the subpattern, we want to make sure | |
924 | * loop is ended the next time around | |
925 | */ | |
926 | if (!*p) { | |
927 | had_sub_pattern = true; | |
928 | p--; | |
929 | } | |
b98f7d6e | 930 | |
33ad9f1a | 931 | goto find_name_on_this_level; |
4fb3cba5 | 932 | |
33ad9f1a CS |
933 | cleanup_name_on_this_level: |
934 | /* This is reached if we found a name clash. | |
935 | * In that case, remove the cgroup from all previous hierarchies | |
936 | */ | |
937 | for (j = 0, info_ptr = base_info; j < i && info_ptr; info_ptr = info_ptr->next, j++) { | |
77afbedf SH |
938 | if (info_ptr->created_paths_count < 1) |
939 | continue; | |
6a9e0f26 | 940 | r = remove_cgroup(info_ptr->designated_mount_point, info_ptr->created_paths[info_ptr->created_paths_count - 1], false, NULL); |
33ad9f1a CS |
941 | if (r < 0) |
942 | WARN("could not clean up cgroup we created when trying to create container"); | |
943 | free(info_ptr->created_paths[info_ptr->created_paths_count - 1]); | |
944 | info_ptr->created_paths[--info_ptr->created_paths_count] = NULL; | |
945 | } | |
946 | if (current_component != current_subpath) | |
947 | free(current_subpath); | |
948 | if (current_component != p_eff) | |
949 | free(current_component); | |
950 | current_component = current_subpath = NULL; | |
951 | /* try again with another suffix */ | |
952 | ++suffix; | |
4fb3cba5 | 953 | |
33ad9f1a CS |
954 | find_name_on_this_level: |
955 | /* determine name of the path component we should create */ | |
956 | if (contains_name && suffix > 0) { | |
957 | char *buf = calloc(strlen(name) + 32, 1); | |
958 | if (!buf) | |
959 | goto out_initial_error; | |
960 | snprintf(buf, strlen(name) + 32, "%s-%u", name, suffix); | |
961 | current_component = lxc_string_replace("%n", buf, p_eff); | |
962 | free(buf); | |
963 | } else { | |
964 | current_component = contains_name ? lxc_string_replace("%n", name, p_eff) : p_eff; | |
965 | } | |
966 | parts[0] = path_so_far; | |
967 | parts[1] = current_component; | |
968 | parts[2] = NULL; | |
969 | current_subpath = path_so_far ? lxc_string_join("/", (const char **)parts, false) : current_component; | |
970 | ||
971 | /* Now go through each hierarchy and try to create the | |
972 | * corresponding cgroup | |
973 | */ | |
974 | for (i = 0, info_ptr = base_info; info_ptr; info_ptr = info_ptr->next, i++) { | |
975 | char *parts2[3]; | |
692ba18f SH |
976 | |
977 | if (lxc_string_in_array("ns", (const char **)info_ptr->hierarchy->subsystems)) | |
978 | continue; | |
33ad9f1a CS |
979 | current_entire_path = NULL; |
980 | ||
981 | parts2[0] = !strcmp(info_ptr->cgroup_path, "/") ? "" : info_ptr->cgroup_path; | |
982 | parts2[1] = current_subpath; | |
983 | parts2[2] = NULL; | |
984 | current_entire_path = lxc_string_join("/", (const char **)parts2, false); | |
985 | ||
986 | if (!*p) { | |
987 | /* we are processing the subpath, so only update that one */ | |
988 | free(new_cgroup_paths_sub[i]); | |
989 | new_cgroup_paths_sub[i] = strdup(current_entire_path); | |
990 | if (!new_cgroup_paths_sub[i]) | |
991 | goto cleanup_from_error; | |
992 | } else { | |
993 | /* remember which path was used on this controller */ | |
994 | free(new_cgroup_paths[i]); | |
995 | new_cgroup_paths[i] = strdup(current_entire_path); | |
996 | if (!new_cgroup_paths[i]) | |
997 | goto cleanup_from_error; | |
998 | } | |
fd4f5a56 | 999 | |
33ad9f1a CS |
1000 | r = create_cgroup(info_ptr->designated_mount_point, current_entire_path); |
1001 | if (r < 0 && errno == EEXIST && contains_name) { | |
1002 | /* name clash => try new name with new suffix */ | |
1003 | free(current_entire_path); | |
1004 | current_entire_path = NULL; | |
1005 | goto cleanup_name_on_this_level; | |
1006 | } else if (r < 0 && errno != EEXIST) { | |
ea439aac SH |
1007 | if (is_crucial_hierarchy(info_ptr->hierarchy)) { |
1008 | SYSERROR("Could not create cgroup '%s' in '%s'.", current_entire_path, info_ptr->designated_mount_point->mount_point); | |
1009 | goto cleanup_from_error; | |
1010 | } | |
1011 | goto skip; | |
33ad9f1a CS |
1012 | } else if (r == 0) { |
1013 | /* successfully created */ | |
1014 | r = lxc_grow_array((void ***)&info_ptr->created_paths, &info_ptr->created_paths_capacity, info_ptr->created_paths_count + 1, 8); | |
1015 | if (r < 0) | |
1016 | goto cleanup_from_error; | |
d703c2b1 | 1017 | if (!init_cpuset_if_needed(info_ptr->designated_mount_point, current_entire_path)) { |
b38b62a6 | 1018 | ERROR("Failed to initialize cpuset for '%s' in '%s'.", current_entire_path, info_ptr->designated_mount_point->mount_point); |
d703c2b1 RV |
1019 | goto cleanup_from_error; |
1020 | } | |
33ad9f1a CS |
1021 | info_ptr->created_paths[info_ptr->created_paths_count++] = current_entire_path; |
1022 | } else { | |
1023 | /* if we didn't create the cgroup, then we have to make sure that | |
1024 | * further cgroups will be created properly | |
1025 | */ | |
d703c2b1 | 1026 | if (handle_cgroup_settings(info_ptr->designated_mount_point, info_ptr->cgroup_path) < 0) { |
f6ac3b9e | 1027 | ERROR("Could not set clone_children to 1 for cpuset hierarchy in pre-existing cgroup."); |
33ad9f1a | 1028 | goto cleanup_from_error; |
f6ac3b9e | 1029 | } |
d703c2b1 RV |
1030 | if (!init_cpuset_if_needed(info_ptr->designated_mount_point, info_ptr->cgroup_path)) { |
1031 | ERROR("Failed to initialize cpuset in pre-existing '%s'.", info_ptr->cgroup_path); | |
1032 | goto cleanup_from_error; | |
1033 | } | |
33ad9f1a | 1034 | |
ea439aac | 1035 | skip: |
33ad9f1a CS |
1036 | /* already existed but path component of pattern didn't contain '%n', |
1037 | * so this is not an error; but then we don't need current_entire_path | |
1038 | * anymore... | |
1039 | */ | |
1040 | free(current_entire_path); | |
1041 | current_entire_path = NULL; | |
1042 | } | |
1043 | } | |
fd4f5a56 | 1044 | |
33ad9f1a CS |
1045 | /* save path so far */ |
1046 | free(path_so_far); | |
1047 | path_so_far = strdup(current_subpath); | |
1048 | if (!path_so_far) | |
1049 | goto cleanup_from_error; | |
1050 | ||
1051 | /* cleanup */ | |
1052 | if (current_component != current_subpath) | |
1053 | free(current_subpath); | |
1054 | if (current_component != p_eff) | |
1055 | free(current_component); | |
1056 | current_component = current_subpath = NULL; | |
1057 | continue; | |
4fb3cba5 | 1058 | |
33ad9f1a | 1059 | cleanup_from_error: |
ec64264d | 1060 | /* called if an error occurred in the loop, so we |
33ad9f1a CS |
1061 | * do some additional cleanup here |
1062 | */ | |
1063 | saved_errno = errno; | |
1064 | if (current_component != current_subpath) | |
1065 | free(current_subpath); | |
1066 | if (current_component != p_eff) | |
1067 | free(current_component); | |
1068 | free(current_entire_path); | |
1069 | errno = saved_errno; | |
1070 | goto out_initial_error; | |
fd4f5a56 DL |
1071 | } |
1072 | ||
33ad9f1a CS |
1073 | /* we're done, now update the paths */ |
1074 | for (i = 0, info_ptr = base_info; info_ptr; info_ptr = info_ptr->next, i++) { | |
47d8fb3b CS |
1075 | /* ignore legacy 'ns' subsystem here, lxc_cgroup_create_legacy |
1076 | * will take care of it | |
1077 | * Since we do a continue in above loop, new_cgroup_paths[i] is | |
1078 | * unset anyway, as is new_cgroup_paths_sub[i] | |
692ba18f | 1079 | */ |
47d8fb3b CS |
1080 | if (lxc_string_in_array("ns", (const char **)info_ptr->hierarchy->subsystems)) |
1081 | continue; | |
1082 | free(info_ptr->cgroup_path); | |
1083 | info_ptr->cgroup_path = new_cgroup_paths[i]; | |
1084 | info_ptr->cgroup_path_sub = new_cgroup_paths_sub[i]; | |
fd4f5a56 | 1085 | } |
33ad9f1a CS |
1086 | /* don't use lxc_free_array since we used the array members |
1087 | * to store them in our result... | |
1088 | */ | |
1089 | free(new_cgroup_paths); | |
1090 | free(new_cgroup_paths_sub); | |
1091 | free(path_so_far); | |
1092 | lxc_free_array((void **)cgroup_path_components, free); | |
1093 | return base_info; | |
1094 | ||
1095 | out_initial_error: | |
1096 | saved_errno = errno; | |
1097 | free(path_so_far); | |
6a9e0f26 | 1098 | lxc_cgroup_process_info_free_and_remove(base_info, NULL); |
33ad9f1a CS |
1099 | lxc_free_array((void **)new_cgroup_paths, free); |
1100 | lxc_free_array((void **)new_cgroup_paths_sub, free); | |
1101 | lxc_free_array((void **)cgroup_path_components, free); | |
1102 | errno = saved_errno; | |
1103 | return NULL; | |
c8f7c563 CS |
1104 | } |
1105 | ||
4fb3cba5 | 1106 | static int lxc_cgroup_create_legacy(struct cgroup_process_info *base_info, const char *name, pid_t pid) |
47d8fb3b CS |
1107 | { |
1108 | struct cgroup_process_info *info_ptr; | |
1109 | int r; | |
1110 | ||
1111 | for (info_ptr = base_info; info_ptr; info_ptr = info_ptr->next) { | |
1112 | if (!lxc_string_in_array("ns", (const char **)info_ptr->hierarchy->subsystems)) | |
1113 | continue; | |
1114 | /* | |
1115 | * For any path which has ns cgroup mounted, handler->pid is already | |
1116 | * moved into a container called '%d % (handler->pid)'. Rename it to | |
1117 | * the cgroup name and record that. | |
1118 | */ | |
1119 | char *tmp = cgroup_rename_nsgroup((const char *)info_ptr->designated_mount_point->mount_point, | |
1120 | info_ptr->cgroup_path, pid, name); | |
1121 | if (!tmp) | |
1122 | return -1; | |
1123 | free(info_ptr->cgroup_path); | |
1124 | info_ptr->cgroup_path = tmp; | |
1125 | r = lxc_grow_array((void ***)&info_ptr->created_paths, &info_ptr->created_paths_capacity, info_ptr->created_paths_count + 1, 8); | |
1126 | if (r < 0) | |
1127 | return -1; | |
1128 | tmp = strdup(tmp); | |
1129 | if (!tmp) | |
1130 | return -1; | |
1131 | info_ptr->created_paths[info_ptr->created_paths_count++] = tmp; | |
1132 | } | |
1133 | return 0; | |
1134 | } | |
1135 | ||
33ad9f1a | 1136 | /* get the cgroup membership of a given container */ |
4fb3cba5 | 1137 | static struct cgroup_process_info *lxc_cgroup_get_container_info(const char *name, const char *lxcpath, struct cgroup_meta_data *meta_data) |
c8f7c563 | 1138 | { |
33ad9f1a CS |
1139 | struct cgroup_process_info *result = NULL; |
1140 | int saved_errno = 0; | |
1141 | size_t i; | |
1142 | struct cgroup_process_info **cptr = &result; | |
1143 | struct cgroup_process_info *entry = NULL; | |
1144 | char *path = NULL; | |
1145 | ||
1146 | for (i = 0; i <= meta_data->maximum_hierarchy; i++) { | |
1147 | struct cgroup_hierarchy *h = meta_data->hierarchies[i]; | |
1148 | if (!h || !h->used) | |
1149 | continue; | |
c8f7c563 | 1150 | |
33ad9f1a CS |
1151 | /* use the command interface to look for the cgroup */ |
1152 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->subsystems[0]); | |
c661b0a8 DE |
1153 | if (!path) { |
1154 | h->used = false; | |
c661b0a8 DE |
1155 | continue; |
1156 | } | |
33ad9f1a CS |
1157 | |
1158 | entry = calloc(1, sizeof(struct cgroup_process_info)); | |
1159 | if (!entry) | |
1160 | goto out_error; | |
1161 | entry->meta_ref = lxc_cgroup_get_meta(meta_data); | |
1162 | entry->hierarchy = h; | |
1163 | entry->cgroup_path = path; | |
1164 | path = NULL; | |
1165 | ||
1166 | /* it is not an error if we don't find anything here, | |
1167 | * it is up to the caller to decide what to do in that | |
1168 | * case */ | |
1169 | entry->designated_mount_point = lxc_cgroup_find_mount_point(h, entry->cgroup_path, true); | |
1170 | ||
1171 | *cptr = entry; | |
1172 | cptr = &entry->next; | |
1173 | entry = NULL; | |
c8f7c563 CS |
1174 | } |
1175 | ||
33ad9f1a CS |
1176 | return result; |
1177 | out_error: | |
1178 | saved_errno = errno; | |
1179 | free(path); | |
1180 | lxc_cgroup_process_info_free(result); | |
1181 | lxc_cgroup_process_info_free(entry); | |
1182 | errno = saved_errno; | |
1183 | return NULL; | |
fd4f5a56 DL |
1184 | } |
1185 | ||
33ad9f1a | 1186 | /* move a processs to the cgroups specified by the membership */ |
4fb3cba5 | 1187 | static int lxc_cgroupfs_enter(struct cgroup_process_info *info, pid_t pid, bool enter_sub) |
4f17323e | 1188 | { |
33ad9f1a CS |
1189 | char pid_buf[32]; |
1190 | char *cgroup_tasks_fn; | |
1191 | int r; | |
1192 | struct cgroup_process_info *info_ptr; | |
1193 | ||
1194 | snprintf(pid_buf, 32, "%lu", (unsigned long)pid); | |
1195 | for (info_ptr = info; info_ptr; info_ptr = info_ptr->next) { | |
1196 | char *cgroup_path = (enter_sub && info_ptr->cgroup_path_sub) ? | |
1197 | info_ptr->cgroup_path_sub : | |
1198 | info_ptr->cgroup_path; | |
1199 | ||
1200 | if (!info_ptr->designated_mount_point) { | |
1201 | info_ptr->designated_mount_point = lxc_cgroup_find_mount_point(info_ptr->hierarchy, cgroup_path, true); | |
1202 | if (!info_ptr->designated_mount_point) { | |
1203 | 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); | |
1204 | return -1; | |
1205 | } | |
1206 | } | |
4f17323e | 1207 | |
33ad9f1a CS |
1208 | cgroup_tasks_fn = cgroup_to_absolute_path(info_ptr->designated_mount_point, cgroup_path, "/tasks"); |
1209 | if (!cgroup_tasks_fn) { | |
1210 | SYSERROR("Could not add pid %lu to cgroup %s: internal error", (unsigned long)pid, cgroup_path); | |
1211 | return -1; | |
1212 | } | |
4f17323e | 1213 | |
33ad9f1a | 1214 | r = lxc_write_to_file(cgroup_tasks_fn, pid_buf, strlen(pid_buf), false); |
5903da82 | 1215 | free(cgroup_tasks_fn); |
ea439aac | 1216 | if (r < 0 && is_crucial_hierarchy(info_ptr->hierarchy)) { |
33ad9f1a CS |
1217 | SYSERROR("Could not add pid %lu to cgroup %s: internal error", (unsigned long)pid, cgroup_path); |
1218 | return -1; | |
1219 | } | |
4f17323e CS |
1220 | } |
1221 | ||
33ad9f1a | 1222 | return 0; |
4f17323e CS |
1223 | } |
1224 | ||
33ad9f1a CS |
1225 | /* free process membership information */ |
1226 | void lxc_cgroup_process_info_free(struct cgroup_process_info *info) | |
fc7de561 | 1227 | { |
33ad9f1a CS |
1228 | struct cgroup_process_info *next; |
1229 | if (!info) | |
b98f7d6e | 1230 | return; |
33ad9f1a CS |
1231 | next = info->next; |
1232 | lxc_cgroup_put_meta(info->meta_ref); | |
1233 | free(info->cgroup_path); | |
1234 | free(info->cgroup_path_sub); | |
1235 | lxc_free_array((void **)info->created_paths, free); | |
1236 | free(info); | |
1237 | lxc_cgroup_process_info_free(next); | |
fc7de561 SH |
1238 | } |
1239 | ||
33ad9f1a | 1240 | /* free process membership information and remove cgroups that were created */ |
6a9e0f26 | 1241 | void lxc_cgroup_process_info_free_and_remove(struct cgroup_process_info *info, struct lxc_conf *conf) |
b98f7d6e | 1242 | { |
33ad9f1a CS |
1243 | struct cgroup_process_info *next; |
1244 | char **pp; | |
1245 | if (!info) | |
1246 | return; | |
1247 | next = info->next; | |
603c64c2 | 1248 | { |
33ad9f1a CS |
1249 | struct cgroup_mount_point *mp = info->designated_mount_point; |
1250 | if (!mp) | |
1251 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, true); | |
1252 | if (mp) | |
1253 | /* ignore return value here, perhaps we created the | |
1254 | * '/lxc' cgroup in this container but another container | |
1255 | * is still running (for example) | |
1256 | */ | |
6a9e0f26 | 1257 | (void)remove_cgroup(mp, info->cgroup_path, true, conf); |
603c64c2 SH |
1258 | } |
1259 | for (pp = info->created_paths; pp && *pp; pp++); | |
1260 | for ((void)(pp && --pp); info->created_paths && pp >= info->created_paths; --pp) { | |
33ad9f1a | 1261 | free(*pp); |
b98f7d6e | 1262 | } |
33ad9f1a CS |
1263 | free(info->created_paths); |
1264 | lxc_cgroup_put_meta(info->meta_ref); | |
1265 | free(info->cgroup_path); | |
1266 | free(info->cgroup_path_sub); | |
1267 | free(info); | |
6a9e0f26 | 1268 | lxc_cgroup_process_info_free_and_remove(next, conf); |
33ad9f1a | 1269 | } |
b98f7d6e | 1270 | |
4fb3cba5 | 1271 | static char *lxc_cgroup_get_hierarchy_path_data(const char *subsystem, struct cgfs_data *d) |
33ad9f1a | 1272 | { |
d4ef7c50 SH |
1273 | struct cgroup_process_info *info = d->info; |
1274 | info = find_info_for_subsystem(info, subsystem); | |
33ad9f1a CS |
1275 | if (!info) |
1276 | return NULL; | |
f348e47c | 1277 | prune_init_scope(info->cgroup_path); |
33ad9f1a | 1278 | return info->cgroup_path; |
b98f7d6e SH |
1279 | } |
1280 | ||
4fb3cba5 | 1281 | static char *lxc_cgroup_get_hierarchy_abs_path_data(const char *subsystem, struct cgfs_data *d) |
b98f7d6e | 1282 | { |
d4ef7c50 | 1283 | struct cgroup_process_info *info = d->info; |
33ad9f1a | 1284 | struct cgroup_mount_point *mp = NULL; |
d4ef7c50 SH |
1285 | |
1286 | info = find_info_for_subsystem(info, subsystem); | |
33ad9f1a CS |
1287 | if (!info) |
1288 | return NULL; | |
1289 | if (info->designated_mount_point) { | |
8900b9eb | 1290 | mp = info->designated_mount_point; |
33ad9f1a CS |
1291 | } else { |
1292 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, true); | |
1293 | if (!mp) | |
1294 | return NULL; | |
b98f7d6e | 1295 | } |
33ad9f1a | 1296 | return cgroup_to_absolute_path(mp, info->cgroup_path, NULL); |
b98f7d6e | 1297 | } |
55c76589 | 1298 | |
4fb3cba5 | 1299 | static char *lxc_cgroup_get_hierarchy_abs_path(const char *subsystem, const char *name, const char *lxcpath) |
9a93d992 | 1300 | { |
33ad9f1a CS |
1301 | struct cgroup_meta_data *meta; |
1302 | struct cgroup_process_info *base_info, *info; | |
1303 | struct cgroup_mount_point *mp; | |
1304 | char *result = NULL; | |
33ad9f1a CS |
1305 | |
1306 | meta = lxc_cgroup_load_meta(); | |
1307 | if (!meta) | |
9a93d992 | 1308 | return NULL; |
33ad9f1a CS |
1309 | base_info = lxc_cgroup_get_container_info(name, lxcpath, meta); |
1310 | if (!base_info) | |
178938fe | 1311 | goto out1; |
33ad9f1a CS |
1312 | info = find_info_for_subsystem(base_info, subsystem); |
1313 | if (!info) | |
178938fe | 1314 | goto out2; |
33ad9f1a | 1315 | if (info->designated_mount_point) { |
8900b9eb | 1316 | mp = info->designated_mount_point; |
33ad9f1a CS |
1317 | } else { |
1318 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, true); | |
1319 | if (!mp) | |
178938fe | 1320 | goto out3; |
33ad9f1a CS |
1321 | } |
1322 | result = cgroup_to_absolute_path(mp, info->cgroup_path, NULL); | |
178938fe | 1323 | out3: |
178938fe | 1324 | out2: |
33ad9f1a | 1325 | lxc_cgroup_process_info_free(base_info); |
178938fe | 1326 | out1: |
33ad9f1a | 1327 | lxc_cgroup_put_meta(meta); |
33ad9f1a CS |
1328 | return result; |
1329 | } | |
9a93d992 | 1330 | |
4fb3cba5 | 1331 | static int lxc_cgroup_set_data(const char *filename, const char *value, struct cgfs_data *d) |
33ad9f1a CS |
1332 | { |
1333 | char *subsystem = NULL, *p, *path; | |
1334 | int ret = -1; | |
9a93d992 | 1335 | |
33ad9f1a CS |
1336 | subsystem = alloca(strlen(filename) + 1); |
1337 | strcpy(subsystem, filename); | |
46cd2845 | 1338 | if ((p = strchr(subsystem, '.')) != NULL) |
33ad9f1a | 1339 | *p = '\0'; |
9a93d992 | 1340 | |
4f875f70 | 1341 | errno = ENOENT; |
4fb3cba5 | 1342 | path = lxc_cgroup_get_hierarchy_abs_path_data(subsystem, d); |
33ad9f1a CS |
1343 | if (path) { |
1344 | ret = do_cgroup_set(path, filename, value); | |
4f875f70 | 1345 | int saved_errno = errno; |
33ad9f1a | 1346 | free(path); |
4f875f70 | 1347 | errno = saved_errno; |
9a93d992 | 1348 | } |
33ad9f1a CS |
1349 | return ret; |
1350 | } | |
9a93d992 | 1351 | |
4fb3cba5 | 1352 | static int lxc_cgroupfs_set(const char *filename, const char *value, const char *name, const char *lxcpath) |
9a93d992 | 1353 | { |
33ad9f1a CS |
1354 | char *subsystem = NULL, *p, *path; |
1355 | int ret = -1; | |
9a93d992 | 1356 | |
33ad9f1a CS |
1357 | subsystem = alloca(strlen(filename) + 1); |
1358 | strcpy(subsystem, filename); | |
46cd2845 | 1359 | if ((p = strchr(subsystem, '.')) != NULL) |
33ad9f1a | 1360 | *p = '\0'; |
9a93d992 | 1361 | |
33ad9f1a CS |
1362 | path = lxc_cgroup_get_hierarchy_abs_path(subsystem, name, lxcpath); |
1363 | if (path) { | |
1364 | ret = do_cgroup_set(path, filename, value); | |
1365 | free(path); | |
1366 | } | |
b98f7d6e | 1367 | return ret; |
9a93d992 SH |
1368 | } |
1369 | ||
4fb3cba5 | 1370 | static int lxc_cgroupfs_get(const char *filename, char *value, size_t len, const char *name, const char *lxcpath) |
9a93d992 | 1371 | { |
33ad9f1a CS |
1372 | char *subsystem = NULL, *p, *path; |
1373 | int ret = -1; | |
1374 | ||
1375 | subsystem = alloca(strlen(filename) + 1); | |
1376 | strcpy(subsystem, filename); | |
46cd2845 | 1377 | if ((p = strchr(subsystem, '.')) != NULL) |
33ad9f1a CS |
1378 | *p = '\0'; |
1379 | ||
1380 | path = lxc_cgroup_get_hierarchy_abs_path(subsystem, name, lxcpath); | |
1381 | if (path) { | |
1382 | ret = do_cgroup_get(path, filename, value, len); | |
1383 | free(path); | |
9a93d992 | 1384 | } |
33ad9f1a | 1385 | return ret; |
9a93d992 SH |
1386 | } |
1387 | ||
4fb3cba5 | 1388 | static bool cgroupfs_mount_cgroup(void *hdata, const char *root, int type) |
aae1f3c4 CS |
1389 | { |
1390 | size_t bufsz = strlen(root) + sizeof("/sys/fs/cgroup"); | |
1391 | char *path = NULL; | |
1392 | char **parts = NULL; | |
1393 | char *dirname = NULL; | |
1394 | char *abs_path = NULL; | |
1395 | char *abs_path2 = NULL; | |
d4ef7c50 SH |
1396 | struct cgfs_data *cgfs_d; |
1397 | struct cgroup_process_info *info, *base_info; | |
aae1f3c4 CS |
1398 | int r, saved_errno = 0; |
1399 | ||
4608594e SH |
1400 | if (cgns_supported()) |
1401 | return true; | |
1402 | ||
4fb3cba5 DE |
1403 | cgfs_d = hdata; |
1404 | if (!cgfs_d) | |
1405 | return false; | |
d4ef7c50 SH |
1406 | base_info = cgfs_d->info; |
1407 | ||
0769b82a CS |
1408 | /* If we get passed the _NOSPEC types, we default to _MIXED, since we don't |
1409 | * have access to the lxc_conf object at this point. It really should be up | |
1410 | * to the caller to fix this, but this doesn't really hurt. | |
1411 | */ | |
1412 | if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
1413 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
1414 | else if (type == LXC_AUTO_CGROUP_NOSPEC) | |
1415 | type = LXC_AUTO_CGROUP_MIXED; | |
1416 | ||
7997d7da CS |
1417 | if (type < LXC_AUTO_CGROUP_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) { |
1418 | ERROR("could not mount cgroups into container: invalid type specified internally"); | |
1419 | errno = EINVAL; | |
c476bdce | 1420 | return false; |
7997d7da CS |
1421 | } |
1422 | ||
aae1f3c4 CS |
1423 | path = calloc(1, bufsz); |
1424 | if (!path) | |
c476bdce | 1425 | return false; |
aae1f3c4 | 1426 | snprintf(path, bufsz, "%s/sys/fs/cgroup", root); |
592fd47a SH |
1427 | r = safe_mount("cgroup_root", path, "tmpfs", |
1428 | MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_RELATIME, | |
1429 | "size=10240k,mode=755", | |
1430 | root); | |
aae1f3c4 CS |
1431 | if (r < 0) { |
1432 | SYSERROR("could not mount tmpfs to /sys/fs/cgroup in the container"); | |
c476bdce | 1433 | return false; |
aae1f3c4 CS |
1434 | } |
1435 | ||
1436 | /* now mount all the hierarchies we care about */ | |
1437 | for (info = base_info; info; info = info->next) { | |
1438 | size_t subsystem_count, i; | |
1439 | struct cgroup_mount_point *mp = info->designated_mount_point; | |
d3f99e96 | 1440 | if (!mountpoint_is_accessible(mp)) |
aae1f3c4 | 1441 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, true); |
d3f99e96 | 1442 | |
aae1f3c4 CS |
1443 | if (!mp) { |
1444 | SYSERROR("could not find original mount point for cgroup hierarchy while trying to mount cgroup filesystem"); | |
1445 | goto out_error; | |
1446 | } | |
1447 | ||
1448 | subsystem_count = lxc_array_len((void **)info->hierarchy->subsystems); | |
1449 | parts = calloc(subsystem_count + 1, sizeof(char *)); | |
1450 | if (!parts) | |
1451 | goto out_error; | |
1452 | ||
1453 | for (i = 0; i < subsystem_count; i++) { | |
1454 | if (!strncmp(info->hierarchy->subsystems[i], "name=", 5)) | |
1455 | parts[i] = info->hierarchy->subsystems[i] + 5; | |
1456 | else | |
1457 | parts[i] = info->hierarchy->subsystems[i]; | |
1458 | } | |
1459 | dirname = lxc_string_join(",", (const char **)parts, false); | |
1460 | if (!dirname) | |
1461 | goto out_error; | |
1462 | ||
1463 | /* create subsystem directory */ | |
1464 | abs_path = lxc_append_paths(path, dirname); | |
1465 | if (!abs_path) | |
1466 | goto out_error; | |
1467 | r = mkdir_p(abs_path, 0755); | |
1468 | if (r < 0 && errno != EEXIST) { | |
1469 | SYSERROR("could not create cgroup subsystem directory /sys/fs/cgroup/%s", dirname); | |
1470 | goto out_error; | |
1471 | } | |
1472 | ||
aae1f3c4 CS |
1473 | abs_path2 = lxc_append_paths(abs_path, info->cgroup_path); |
1474 | if (!abs_path2) | |
1475 | goto out_error; | |
aae1f3c4 | 1476 | |
7997d7da CS |
1477 | if (type == LXC_AUTO_CGROUP_FULL_RO || type == LXC_AUTO_CGROUP_FULL_RW || type == LXC_AUTO_CGROUP_FULL_MIXED) { |
1478 | /* bind-mount the cgroup entire filesystem there */ | |
1479 | if (strcmp(mp->mount_prefix, "/") != 0) { | |
1480 | /* FIXME: maybe we should just try to remount the entire hierarchy | |
1481 | * with a regular mount command? may that works? */ | |
1482 | 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); | |
1483 | goto out_error; | |
1484 | } | |
1485 | r = mount(mp->mount_point, abs_path, "none", MS_BIND, 0); | |
1486 | if (r < 0) { | |
1487 | SYSERROR("error bind-mounting %s to %s", mp->mount_point, abs_path); | |
1488 | goto out_error; | |
1489 | } | |
f8f3c3c0 SG |
1490 | /* main cgroup path should be read-only */ |
1491 | if (type == LXC_AUTO_CGROUP_FULL_RO || type == LXC_AUTO_CGROUP_FULL_MIXED) { | |
1492 | r = mount(NULL, abs_path, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY, NULL); | |
1493 | if (r < 0) { | |
1494 | SYSERROR("error re-mounting %s readonly", abs_path); | |
1495 | goto out_error; | |
1496 | } | |
1497 | } | |
7997d7da CS |
1498 | /* own cgroup should be read-write */ |
1499 | if (type == LXC_AUTO_CGROUP_FULL_MIXED) { | |
1500 | r = mount(abs_path2, abs_path2, NULL, MS_BIND, NULL); | |
1501 | if (r < 0) { | |
1502 | SYSERROR("error bind-mounting %s onto itself", abs_path2); | |
1503 | goto out_error; | |
1504 | } | |
1505 | r = mount(NULL, abs_path2, NULL, MS_REMOUNT|MS_BIND, NULL); | |
1506 | if (r < 0) { | |
1507 | SYSERROR("error re-mounting %s readwrite", abs_path2); | |
1508 | goto out_error; | |
1509 | } | |
1510 | } | |
1511 | } else { | |
1512 | /* create path for container's cgroup */ | |
1513 | r = mkdir_p(abs_path2, 0755); | |
1514 | if (r < 0 && errno != EEXIST) { | |
1515 | SYSERROR("could not create cgroup directory /sys/fs/cgroup/%s%s", dirname, info->cgroup_path); | |
1516 | goto out_error; | |
1517 | } | |
aae1f3c4 | 1518 | |
b46f0553 CS |
1519 | /* for read-only and mixed cases, we have to bind-mount the tmpfs directory |
1520 | * that points to the hierarchy itself (i.e. /sys/fs/cgroup/cpu etc.) onto | |
1521 | * itself and then bind-mount it read-only, since we keep the tmpfs itself | |
1522 | * read-write (see comment below) | |
1523 | */ | |
1524 | if (type == LXC_AUTO_CGROUP_MIXED || type == LXC_AUTO_CGROUP_RO) { | |
1525 | r = mount(abs_path, abs_path, NULL, MS_BIND, NULL); | |
1526 | if (r < 0) { | |
1527 | SYSERROR("error bind-mounting %s onto itself", abs_path); | |
1528 | goto out_error; | |
1529 | } | |
1530 | r = mount(NULL, abs_path, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY, NULL); | |
1531 | if (r < 0) { | |
1532 | SYSERROR("error re-mounting %s readonly", abs_path); | |
1533 | goto out_error; | |
1534 | } | |
1535 | } | |
1536 | ||
7997d7da CS |
1537 | free(abs_path); |
1538 | abs_path = NULL; | |
1539 | ||
1540 | /* bind-mount container's cgroup to that directory */ | |
1541 | abs_path = cgroup_to_absolute_path(mp, info->cgroup_path, NULL); | |
1542 | if (!abs_path) | |
1543 | goto out_error; | |
1544 | r = mount(abs_path, abs_path2, "none", MS_BIND, 0); | |
ea439aac | 1545 | if (r < 0 && is_crucial_hierarchy(info->hierarchy)) { |
7997d7da CS |
1546 | SYSERROR("error bind-mounting %s to %s", abs_path, abs_path2); |
1547 | goto out_error; | |
1548 | } | |
1549 | if (type == LXC_AUTO_CGROUP_RO) { | |
1550 | r = mount(NULL, abs_path2, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY, NULL); | |
1551 | if (r < 0) { | |
1552 | SYSERROR("error re-mounting %s readonly", abs_path2); | |
1553 | goto out_error; | |
1554 | } | |
1555 | } | |
aae1f3c4 CS |
1556 | } |
1557 | ||
1558 | free(abs_path); | |
1559 | free(abs_path2); | |
1560 | abs_path = NULL; | |
1561 | abs_path2 = NULL; | |
1562 | ||
1563 | /* add symlinks for every single subsystem */ | |
1564 | if (subsystem_count > 1) { | |
1565 | for (i = 0; i < subsystem_count; i++) { | |
1566 | abs_path = lxc_append_paths(path, parts[i]); | |
1567 | if (!abs_path) | |
1568 | goto out_error; | |
1569 | r = symlink(dirname, abs_path); | |
1570 | if (r < 0) | |
1571 | WARN("could not create symlink %s -> %s in /sys/fs/cgroup of container", parts[i], dirname); | |
1572 | free(abs_path); | |
1573 | abs_path = NULL; | |
1574 | } | |
1575 | } | |
1576 | free(dirname); | |
1577 | free(parts); | |
1578 | dirname = NULL; | |
1579 | parts = NULL; | |
1580 | } | |
1581 | ||
b46f0553 CS |
1582 | /* We used to remount the entire tmpfs readonly if any :ro or |
1583 | * :mixed mode was specified. However, Ubuntu's mountall has the | |
1584 | * unfortunate behavior to block bootup if /sys/fs/cgroup is | |
1585 | * mounted read-only and cannot be remounted read-write. | |
1586 | * (mountall reads /lib/init/fstab and tries to (re-)mount all of | |
1587 | * these if they are not already mounted with the right options; | |
1588 | * it contains an entry for /sys/fs/cgroup. In case it can't do | |
1589 | * that, it prompts for the user to either manually fix it or | |
1590 | * boot anyway. But without user input, booting of the container | |
1591 | * hangs.) | |
1592 | * | |
1593 | * Instead of remounting the entire tmpfs readonly, we only | |
1594 | * remount the paths readonly that are part of the cgroup | |
1595 | * hierarchy. | |
f8f3c3c0 | 1596 | */ |
f8f3c3c0 | 1597 | |
aae1f3c4 CS |
1598 | free(path); |
1599 | ||
c476bdce | 1600 | return true; |
aae1f3c4 CS |
1601 | |
1602 | out_error: | |
1603 | saved_errno = errno; | |
1604 | free(path); | |
1605 | free(dirname); | |
1606 | free(parts); | |
1607 | free(abs_path); | |
1608 | free(abs_path2); | |
1609 | errno = saved_errno; | |
c476bdce | 1610 | return false; |
aae1f3c4 CS |
1611 | } |
1612 | ||
4fb3cba5 | 1613 | static int cgfs_nrtasks(void *hdata) |
33ad9f1a | 1614 | { |
4fb3cba5 DE |
1615 | struct cgfs_data *d = hdata; |
1616 | struct cgroup_process_info *info; | |
33ad9f1a CS |
1617 | struct cgroup_mount_point *mp = NULL; |
1618 | char *abs_path = NULL; | |
1619 | int ret; | |
460a1cf0 | 1620 | |
4fb3cba5 DE |
1621 | if (!d) { |
1622 | errno = ENOENT; | |
1623 | return -1; | |
1624 | } | |
1625 | ||
1626 | info = d->info; | |
33ad9f1a CS |
1627 | if (!info) { |
1628 | errno = ENOENT; | |
1629 | return -1; | |
b98f7d6e | 1630 | } |
c8f7c563 | 1631 | |
33ad9f1a | 1632 | if (info->designated_mount_point) { |
8900b9eb | 1633 | mp = info->designated_mount_point; |
33ad9f1a CS |
1634 | } else { |
1635 | mp = lxc_cgroup_find_mount_point(info->hierarchy, info->cgroup_path, false); | |
1636 | if (!mp) | |
1637 | return -1; | |
c8f7c563 CS |
1638 | } |
1639 | ||
33ad9f1a CS |
1640 | abs_path = cgroup_to_absolute_path(mp, info->cgroup_path, NULL); |
1641 | if (!abs_path) | |
1642 | return -1; | |
1643 | ||
1644 | ret = cgroup_recursive_task_count(abs_path); | |
1645 | free(abs_path); | |
1646 | return ret; | |
c8f7c563 CS |
1647 | } |
1648 | ||
574c4428 QH |
1649 | static struct cgroup_process_info * |
1650 | lxc_cgroup_process_info_getx(const char *proc_pid_cgroup_str, | |
1651 | struct cgroup_meta_data *meta) | |
d08ba6ec | 1652 | { |
33ad9f1a CS |
1653 | struct cgroup_process_info *result = NULL; |
1654 | FILE *proc_pid_cgroup = NULL; | |
1655 | char *line = NULL; | |
1656 | size_t sz = 0; | |
1657 | int saved_errno = 0; | |
1658 | struct cgroup_process_info **cptr = &result; | |
1659 | struct cgroup_process_info *entry = NULL; | |
1660 | ||
1661 | proc_pid_cgroup = fopen_cloexec(proc_pid_cgroup_str, "r"); | |
1662 | if (!proc_pid_cgroup) | |
b98f7d6e | 1663 | return NULL; |
1ac470c0 | 1664 | |
33ad9f1a CS |
1665 | while (getline(&line, &sz, proc_pid_cgroup) != -1) { |
1666 | /* file format: hierarchy:subsystems:group */ | |
1667 | char *colon1; | |
1668 | char *colon2; | |
1669 | char *endptr; | |
1670 | int hierarchy_number; | |
1671 | struct cgroup_hierarchy *h = NULL; | |
fd4f5a56 | 1672 | |
33ad9f1a | 1673 | if (!line[0]) |
ae5c8b8e | 1674 | continue; |
b98f7d6e | 1675 | |
33ad9f1a CS |
1676 | if (line[strlen(line) - 1] == '\n') |
1677 | line[strlen(line) - 1] = '\0'; | |
1678 | ||
1679 | colon1 = strchr(line, ':'); | |
1680 | if (!colon1) | |
8900b9eb | 1681 | continue; |
33ad9f1a CS |
1682 | *colon1++ = '\0'; |
1683 | colon2 = strchr(colon1, ':'); | |
1684 | if (!colon2) | |
ae5c8b8e | 1685 | continue; |
33ad9f1a | 1686 | *colon2++ = '\0'; |
e4659536 | 1687 | |
33ad9f1a CS |
1688 | endptr = NULL; |
1689 | hierarchy_number = strtoul(line, &endptr, 10); | |
1690 | if (!endptr || *endptr) | |
9a93d992 | 1691 | continue; |
9a93d992 | 1692 | |
33ad9f1a CS |
1693 | if (hierarchy_number > meta->maximum_hierarchy) { |
1694 | /* we encountered a hierarchy we didn't have before, | |
1695 | * so probably somebody remounted some stuff in the | |
1696 | * mean time... | |
1697 | */ | |
1698 | errno = EAGAIN; | |
1699 | goto out_error; | |
b98f7d6e | 1700 | } |
33ad9f1a CS |
1701 | |
1702 | h = meta->hierarchies[hierarchy_number]; | |
1703 | if (!h) { | |
1704 | /* we encountered a hierarchy that was thought to be | |
1705 | * dead before, so probably somebody remounted some | |
1706 | * stuff in the mean time... | |
1707 | */ | |
1708 | errno = EAGAIN; | |
1709 | goto out_error; | |
b98f7d6e | 1710 | } |
33ad9f1a CS |
1711 | |
1712 | /* we are told that we should ignore this hierarchy */ | |
1713 | if (!h->used) | |
b98f7d6e | 1714 | continue; |
5193cc3d | 1715 | |
33ad9f1a CS |
1716 | entry = calloc(1, sizeof(struct cgroup_process_info)); |
1717 | if (!entry) | |
1718 | goto out_error; | |
fd4f5a56 | 1719 | |
33ad9f1a CS |
1720 | entry->meta_ref = lxc_cgroup_get_meta(meta); |
1721 | entry->hierarchy = h; | |
1722 | entry->cgroup_path = strdup(colon2); | |
1723 | if (!entry->cgroup_path) | |
1724 | goto out_error; | |
3939a22a | 1725 | prune_init_scope(entry->cgroup_path); |
d08ba6ec | 1726 | |
33ad9f1a CS |
1727 | *cptr = entry; |
1728 | cptr = &entry->next; | |
1729 | entry = NULL; | |
b98f7d6e | 1730 | } |
b98f7d6e | 1731 | |
33ad9f1a CS |
1732 | fclose(proc_pid_cgroup); |
1733 | free(line); | |
1734 | return result; | |
1735 | ||
1736 | out_error: | |
1737 | saved_errno = errno; | |
1738 | if (proc_pid_cgroup) | |
1739 | fclose(proc_pid_cgroup); | |
1740 | lxc_cgroup_process_info_free(result); | |
1741 | lxc_cgroup_process_info_free(entry); | |
1742 | free(line); | |
1743 | errno = saved_errno; | |
ae5c8b8e | 1744 | return NULL; |
36b86299 DL |
1745 | } |
1746 | ||
574c4428 QH |
1747 | static char **subsystems_from_mount_options(const char *mount_options, |
1748 | char **kernel_list) | |
36b86299 | 1749 | { |
33ad9f1a CS |
1750 | char *token, *str, *saveptr = NULL; |
1751 | char **result = NULL; | |
1752 | size_t result_capacity = 0; | |
8900b9eb | 1753 | size_t result_count = 0; |
33ad9f1a CS |
1754 | int saved_errno; |
1755 | int r; | |
ef342abb | 1756 | |
33ad9f1a CS |
1757 | str = alloca(strlen(mount_options)+1); |
1758 | strcpy(str, mount_options); | |
1759 | for (; (token = strtok_r(str, ",", &saveptr)); str = NULL) { | |
1760 | /* we have a subsystem if it's either in the list of | |
1761 | * subsystems provided by the kernel OR if it starts | |
1762 | * with name= for named hierarchies | |
1763 | */ | |
836514a8 U |
1764 | r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 12); |
1765 | if (r < 0) | |
1766 | goto out_free; | |
1767 | result[result_count + 1] = NULL; | |
1768 | if (strncmp(token, "name=", 5) && !lxc_string_in_array(token, (const char **)kernel_list)) { | |
1769 | // this is eg 'systemd' but the mount will be 'name=systemd' | |
1770 | result[result_count] = malloc(strlen(token) + 6); | |
1771 | if (result[result_count]) | |
1772 | sprintf(result[result_count], "name=%s", token); | |
1773 | } else | |
33ad9f1a | 1774 | result[result_count] = strdup(token); |
836514a8 U |
1775 | if (!result[result_count]) |
1776 | goto out_free; | |
1777 | result_count++; | |
ae5c8b8e | 1778 | } |
f0e64b8b | 1779 | |
33ad9f1a CS |
1780 | return result; |
1781 | ||
1782 | out_free: | |
1783 | saved_errno = errno; | |
1784 | lxc_free_array((void**)result, free); | |
1785 | errno = saved_errno; | |
1786 | return NULL; | |
b98f7d6e SH |
1787 | } |
1788 | ||
574c4428 | 1789 | static void lxc_cgroup_mount_point_free(struct cgroup_mount_point *mp) |
b98f7d6e | 1790 | { |
33ad9f1a CS |
1791 | if (!mp) |
1792 | return; | |
1793 | free(mp->mount_point); | |
1794 | free(mp->mount_prefix); | |
1795 | free(mp); | |
bcbd102c SH |
1796 | } |
1797 | ||
574c4428 | 1798 | static void lxc_cgroup_hierarchy_free(struct cgroup_hierarchy *h) |
341a9bd8 | 1799 | { |
33ad9f1a CS |
1800 | if (!h) |
1801 | return; | |
1802 | lxc_free_array((void **)h->subsystems, free); | |
8bfcb981 | 1803 | free(h->all_mount_points); |
33ad9f1a CS |
1804 | free(h); |
1805 | } | |
341a9bd8 | 1806 | |
574c4428 | 1807 | static bool is_valid_cgroup(const char *name) |
33ad9f1a CS |
1808 | { |
1809 | const char *p; | |
1810 | for (p = name; *p; p++) { | |
28bb9321 QH |
1811 | /* Use the ASCII printable characters range(32 - 127) |
1812 | * is reasonable, we kick out 32(SPACE) because it'll | |
1813 | * break legacy lxc-ls | |
1814 | */ | |
1815 | if (*p <= 32 || *p >= 127 || *p == '/') | |
33ad9f1a | 1816 | return false; |
341a9bd8 | 1817 | } |
33ad9f1a CS |
1818 | return strcmp(name, ".") != 0 && strcmp(name, "..") != 0; |
1819 | } | |
341a9bd8 | 1820 | |
574c4428 | 1821 | static int create_or_remove_cgroup(bool do_remove, |
6a9e0f26 SH |
1822 | struct cgroup_mount_point *mp, const char *path, int recurse, |
1823 | struct lxc_conf *conf) | |
33ad9f1a CS |
1824 | { |
1825 | int r, saved_errno = 0; | |
1826 | char *buf = cgroup_to_absolute_path(mp, path, NULL); | |
1827 | if (!buf) | |
1828 | return -1; | |
341a9bd8 | 1829 | |
33ad9f1a | 1830 | /* create or remove directory */ |
603c64c2 | 1831 | if (do_remove) { |
01d59fe5 CB |
1832 | if (!dir_exists(buf)) |
1833 | return 0; | |
6a9e0f26 SH |
1834 | if (recurse) { |
1835 | if (conf && !lxc_list_empty(&conf->id_map)) | |
1836 | r = userns_exec_1(conf, rmdir_wrapper, buf); | |
1837 | else | |
1838 | r = cgroup_rmdir(buf); | |
1839 | } else | |
603c64c2 SH |
1840 | r = rmdir(buf); |
1841 | } else | |
1842 | r = mkdir(buf, 0777); | |
33ad9f1a CS |
1843 | saved_errno = errno; |
1844 | free(buf); | |
1845 | errno = saved_errno; | |
1846 | return r; | |
341a9bd8 | 1847 | } |
bcbd102c | 1848 | |
574c4428 | 1849 | static int create_cgroup(struct cgroup_mount_point *mp, const char *path) |
a6ddef61 | 1850 | { |
6a9e0f26 | 1851 | return create_or_remove_cgroup(false, mp, path, false, NULL); |
a6ddef61 MN |
1852 | } |
1853 | ||
574c4428 | 1854 | static int remove_cgroup(struct cgroup_mount_point *mp, |
6a9e0f26 | 1855 | const char *path, bool recurse, struct lxc_conf *conf) |
576f946d | 1856 | { |
6a9e0f26 | 1857 | return create_or_remove_cgroup(true, mp, path, recurse, conf); |
33ad9f1a | 1858 | } |
576f946d | 1859 | |
574c4428 QH |
1860 | static char *cgroup_to_absolute_path(struct cgroup_mount_point *mp, |
1861 | const char *path, const char *suffix) | |
33ad9f1a CS |
1862 | { |
1863 | /* first we have to make sure we subtract the mount point's prefix */ | |
1864 | char *prefix = mp->mount_prefix; | |
1865 | char *buf; | |
1866 | ssize_t len, rv; | |
1867 | ||
1868 | /* we want to make sure only absolute paths to cgroups are passed to us */ | |
1869 | if (path[0] != '/') { | |
1870 | errno = EINVAL; | |
1871 | return NULL; | |
1872 | } | |
b98f7d6e | 1873 | |
33ad9f1a CS |
1874 | if (prefix && !strcmp(prefix, "/")) |
1875 | prefix = NULL; | |
b98f7d6e | 1876 | |
33ad9f1a CS |
1877 | /* prefix doesn't match */ |
1878 | if (prefix && strncmp(prefix, path, strlen(prefix)) != 0) { | |
1879 | errno = EINVAL; | |
1880 | return NULL; | |
1881 | } | |
1882 | /* if prefix is /foo and path is /foobar */ | |
1883 | if (prefix && path[strlen(prefix)] != '/' && path[strlen(prefix)] != '\0') { | |
1884 | errno = EINVAL; | |
1885 | return NULL; | |
1886 | } | |
b98f7d6e | 1887 | |
33ad9f1a CS |
1888 | /* remove prefix from path */ |
1889 | path += prefix ? strlen(prefix) : 0; | |
b98f7d6e | 1890 | |
33ad9f1a CS |
1891 | len = strlen(mp->mount_point) + strlen(path) + (suffix ? strlen(suffix) : 0); |
1892 | buf = calloc(len + 1, 1); | |
50266dc6 DE |
1893 | if (!buf) |
1894 | return NULL; | |
33ad9f1a | 1895 | rv = snprintf(buf, len + 1, "%s%s%s", mp->mount_point, path, suffix ? suffix : ""); |
8900b9eb | 1896 | if (rv > len) { |
33ad9f1a CS |
1897 | free(buf); |
1898 | errno = ENOMEM; | |
8900b9eb | 1899 | return NULL; |
8b92dc3a | 1900 | } |
576f946d | 1901 | |
33ad9f1a | 1902 | return buf; |
e0f888d9 | 1903 | } |
283678ed | 1904 | |
574c4428 QH |
1905 | static struct cgroup_process_info * |
1906 | find_info_for_subsystem(struct cgroup_process_info *info, const char *subsystem) | |
283678ed | 1907 | { |
33ad9f1a CS |
1908 | struct cgroup_process_info *info_ptr; |
1909 | for (info_ptr = info; info_ptr; info_ptr = info_ptr->next) { | |
1910 | struct cgroup_hierarchy *h = info_ptr->hierarchy; | |
1911 | if (lxc_string_in_array(subsystem, (const char **)h->subsystems)) | |
1912 | return info_ptr; | |
b98f7d6e | 1913 | } |
33ad9f1a CS |
1914 | errno = ENOENT; |
1915 | return NULL; | |
1916 | } | |
283678ed | 1917 | |
574c4428 QH |
1918 | static int do_cgroup_get(const char *cgroup_path, const char *sub_filename, |
1919 | char *value, size_t len) | |
33ad9f1a CS |
1920 | { |
1921 | const char *parts[3] = { | |
1922 | cgroup_path, | |
1923 | sub_filename, | |
1924 | NULL | |
1925 | }; | |
1926 | char *filename; | |
1927 | int ret, saved_errno; | |
1928 | ||
1929 | filename = lxc_string_join("/", parts, false); | |
1930 | if (!filename) | |
1931 | return -1; | |
1932 | ||
1933 | ret = lxc_read_from_file(filename, value, len); | |
1934 | saved_errno = errno; | |
1935 | free(filename); | |
1936 | errno = saved_errno; | |
1937 | return ret; | |
283678ed | 1938 | } |
b113383b | 1939 | |
574c4428 QH |
1940 | static int do_cgroup_set(const char *cgroup_path, const char *sub_filename, |
1941 | const char *value) | |
b113383b | 1942 | { |
33ad9f1a CS |
1943 | const char *parts[3] = { |
1944 | cgroup_path, | |
1945 | sub_filename, | |
1946 | NULL | |
1947 | }; | |
1948 | char *filename; | |
1949 | int ret, saved_errno; | |
b113383b | 1950 | |
33ad9f1a CS |
1951 | filename = lxc_string_join("/", parts, false); |
1952 | if (!filename) | |
1953 | return -1; | |
b113383b | 1954 | |
33ad9f1a CS |
1955 | ret = lxc_write_to_file(filename, value, strlen(value), false); |
1956 | saved_errno = errno; | |
1957 | free(filename); | |
1958 | errno = saved_errno; | |
1959 | return ret; | |
b98f7d6e SH |
1960 | } |
1961 | ||
4fb3cba5 | 1962 | static int do_setup_cgroup_limits(struct cgfs_data *d, |
574c4428 | 1963 | struct lxc_list *cgroup_settings, bool do_devices) |
b98f7d6e | 1964 | { |
365d180a | 1965 | struct lxc_list *iterator, *sorted_cgroup_settings, *next; |
b98f7d6e SH |
1966 | struct lxc_cgroup *cg; |
1967 | int ret = -1; | |
1968 | ||
33ad9f1a | 1969 | if (lxc_list_empty(cgroup_settings)) |
b98f7d6e SH |
1970 | return 0; |
1971 | ||
aaf26830 | 1972 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); |
fac7c663 KT |
1973 | if (!sorted_cgroup_settings) { |
1974 | return -1; | |
1975 | } | |
aaf26830 KT |
1976 | |
1977 | lxc_list_for_each(iterator, sorted_cgroup_settings) { | |
b98f7d6e SH |
1978 | cg = iterator->elem; |
1979 | ||
33ad9f1a | 1980 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { |
b98f7d6e | 1981 | if (strcmp(cg->subsystem, "devices.deny") == 0 && |
4fb3cba5 | 1982 | cgroup_devices_has_allow_or_deny(d, cg->value, false)) |
b98f7d6e SH |
1983 | continue; |
1984 | if (strcmp(cg->subsystem, "devices.allow") == 0 && | |
4fb3cba5 | 1985 | cgroup_devices_has_allow_or_deny(d, cg->value, true)) |
b98f7d6e | 1986 | continue; |
4fb3cba5 | 1987 | if (lxc_cgroup_set_data(cg->subsystem, cg->value, d)) { |
dddf7c5b | 1988 | if (do_devices && (errno == EACCES || errno == EPERM)) { |
4f875f70 SH |
1989 | WARN("Error setting %s to %s for %s", |
1990 | cg->subsystem, cg->value, d->name); | |
1991 | continue; | |
1992 | } | |
dddf7c5b | 1993 | SYSERROR("Error setting %s to %s for %s", |
4fb3cba5 | 1994 | cg->subsystem, cg->value, d->name); |
b98f7d6e SH |
1995 | goto out; |
1996 | } | |
b113383b | 1997 | } |
b98f7d6e SH |
1998 | |
1999 | DEBUG("cgroup '%s' set to '%s'", cg->subsystem, cg->value); | |
b113383b SH |
2000 | } |
2001 | ||
b98f7d6e SH |
2002 | ret = 0; |
2003 | INFO("cgroup has been setup"); | |
2004 | out: | |
365d180a | 2005 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
aaf26830 KT |
2006 | lxc_list_del(iterator); |
2007 | free(iterator); | |
2008 | } | |
365d180a | 2009 | free(sorted_cgroup_settings); |
b113383b SH |
2010 | return ret; |
2011 | } | |
b98f7d6e | 2012 | |
4fb3cba5 | 2013 | static bool cgroup_devices_has_allow_or_deny(struct cgfs_data *d, |
574c4428 | 2014 | char *v, bool for_allow) |
33ad9f1a CS |
2015 | { |
2016 | char *path; | |
2017 | FILE *devices_list; | |
8900b9eb | 2018 | char *line = NULL; |
33ad9f1a CS |
2019 | size_t sz = 0; |
2020 | bool ret = !for_allow; | |
2021 | const char *parts[3] = { | |
2022 | NULL, | |
2023 | "devices.list", | |
2024 | NULL | |
2025 | }; | |
2026 | ||
2027 | // XXX FIXME if users could use something other than 'lxc.devices.deny = a'. | |
2028 | // not sure they ever do, but they *could* | |
2029 | // right now, I'm assuming they do NOT | |
2030 | if (!for_allow && strcmp(v, "a") != 0 && strcmp(v, "a *:* rwm") != 0) | |
2031 | return false; | |
2032 | ||
4fb3cba5 | 2033 | parts[0] = (const char *)lxc_cgroup_get_hierarchy_abs_path_data("devices", d); |
33ad9f1a CS |
2034 | if (!parts[0]) |
2035 | return false; | |
2036 | path = lxc_string_join("/", parts, false); | |
2037 | if (!path) { | |
2038 | free((void *)parts[0]); | |
2039 | return false; | |
2040 | } | |
2041 | ||
2042 | devices_list = fopen_cloexec(path, "r"); | |
2043 | if (!devices_list) { | |
2044 | free(path); | |
2045 | return false; | |
2046 | } | |
2047 | ||
2048 | while (getline(&line, &sz, devices_list) != -1) { | |
2049 | size_t len = strlen(line); | |
2050 | if (len > 0 && line[len-1] == '\n') | |
2051 | line[len-1] = '\0'; | |
2052 | if (strcmp(line, "a *:* rwm") == 0) { | |
2053 | ret = for_allow; | |
2054 | goto out; | |
2055 | } else if (for_allow && strcmp(line, v) == 0) { | |
2056 | ret = true; | |
8900b9eb | 2057 | goto out; |
33ad9f1a CS |
2058 | } |
2059 | } | |
2060 | ||
2061 | out: | |
2062 | fclose(devices_list); | |
2063 | free(line); | |
2064 | free(path); | |
2065 | return ret; | |
2066 | } | |
2067 | ||
574c4428 | 2068 | static int cgroup_recursive_task_count(const char *cgroup_path) |
b98f7d6e | 2069 | { |
33ad9f1a CS |
2070 | DIR *d; |
2071 | struct dirent *dent_buf; | |
2072 | struct dirent *dent; | |
8900b9eb | 2073 | ssize_t name_max; |
33ad9f1a CS |
2074 | int n = 0, r; |
2075 | ||
2076 | /* see man readdir_r(3) */ | |
2077 | name_max = pathconf(cgroup_path, _PC_NAME_MAX); | |
2078 | if (name_max <= 0) | |
2079 | name_max = 255; | |
2080 | dent_buf = malloc(offsetof(struct dirent, d_name) + name_max + 1); | |
2081 | if (!dent_buf) | |
2082 | return -1; | |
2083 | ||
2084 | d = opendir(cgroup_path); | |
034ef75d SH |
2085 | if (!d) { |
2086 | free(dent_buf); | |
33ad9f1a | 2087 | return 0; |
034ef75d | 2088 | } |
33ad9f1a CS |
2089 | |
2090 | while (readdir_r(d, dent_buf, &dent) == 0 && dent) { | |
2091 | const char *parts[3] = { | |
2092 | cgroup_path, | |
2093 | dent->d_name, | |
2094 | NULL | |
2095 | }; | |
2096 | char *sub_path; | |
2097 | struct stat st; | |
2098 | ||
2099 | if (!strcmp(dent->d_name, ".") || !strcmp(dent->d_name, "..")) | |
2100 | continue; | |
2101 | sub_path = lxc_string_join("/", parts, false); | |
2102 | if (!sub_path) { | |
2103 | closedir(d); | |
2104 | free(dent_buf); | |
2105 | return -1; | |
2106 | } | |
2107 | r = stat(sub_path, &st); | |
2108 | if (r < 0) { | |
2109 | closedir(d); | |
2110 | free(dent_buf); | |
2111 | free(sub_path); | |
2112 | return -1; | |
2113 | } | |
2114 | if (S_ISDIR(st.st_mode)) { | |
2115 | r = cgroup_recursive_task_count(sub_path); | |
2116 | if (r >= 0) | |
2117 | n += r; | |
2118 | } else if (!strcmp(dent->d_name, "tasks")) { | |
2119 | r = count_lines(sub_path); | |
2120 | if (r >= 0) | |
2121 | n += r; | |
2122 | } | |
2123 | free(sub_path); | |
2124 | } | |
2125 | closedir(d); | |
2126 | free(dent_buf); | |
2127 | ||
2128 | return n; | |
2129 | } | |
2130 | ||
574c4428 | 2131 | static int count_lines(const char *fn) |
33ad9f1a CS |
2132 | { |
2133 | FILE *f; | |
2134 | char *line = NULL; | |
2135 | size_t sz = 0; | |
2136 | int n = 0; | |
2137 | ||
2138 | f = fopen_cloexec(fn, "r"); | |
2139 | if (!f) | |
2140 | return -1; | |
2141 | ||
2142 | while (getline(&line, &sz, f) != -1) { | |
2143 | n++; | |
2144 | } | |
2145 | free(line); | |
2146 | fclose(f); | |
2147 | return n; | |
b98f7d6e SH |
2148 | } |
2149 | ||
574c4428 QH |
2150 | static int handle_cgroup_settings(struct cgroup_mount_point *mp, |
2151 | char *cgroup_path) | |
b98f7d6e | 2152 | { |
33ad9f1a | 2153 | int r, saved_errno = 0; |
7e7243e1 | 2154 | char buf[2]; |
1ea59ad2 | 2155 | |
934b1673 SH |
2156 | mp->need_cpuset_init = false; |
2157 | ||
1ea59ad2 SH |
2158 | /* If this is the memory cgroup, we want to enforce hierarchy. |
2159 | * But don't fail if for some reason we can't. | |
2160 | */ | |
2edb53c7 SH |
2161 | if (lxc_string_in_array("memory", (const char **)mp->hierarchy->subsystems)) { |
2162 | char *cc_path = cgroup_to_absolute_path(mp, cgroup_path, "/memory.use_hierarchy"); | |
2163 | if (cc_path) { | |
2164 | r = lxc_read_from_file(cc_path, buf, 1); | |
2165 | if (r < 1 || buf[0] != '1') { | |
2166 | r = lxc_write_to_file(cc_path, "1", 1, false); | |
2167 | if (r < 0) | |
a8916143 | 2168 | SYSERROR("failed to set memory.use_hierarchy to 1; continuing"); |
2edb53c7 | 2169 | } |
1ea59ad2 SH |
2170 | free(cc_path); |
2171 | } | |
2edb53c7 | 2172 | } |
1ea59ad2 | 2173 | |
33ad9f1a CS |
2174 | /* if this is a cpuset hierarchy, we have to set cgroup.clone_children in |
2175 | * the base cgroup, otherwise containers will start with an empty cpuset.mems | |
2176 | * and cpuset.cpus and then | |
2177 | */ | |
2edb53c7 SH |
2178 | if (lxc_string_in_array("cpuset", (const char **)mp->hierarchy->subsystems)) { |
2179 | char *cc_path = cgroup_to_absolute_path(mp, cgroup_path, "/cgroup.clone_children"); | |
d703c2b1 RV |
2180 | struct stat sb; |
2181 | ||
33ad9f1a | 2182 | if (!cc_path) |
2edb53c7 | 2183 | return -1; |
d703c2b1 RV |
2184 | /* cgroup.clone_children is not available when running under |
2185 | * older kernel versions; in this case, we'll initialize | |
2186 | * cpuset.cpus and cpuset.mems later, after the new cgroup | |
2187 | * was created | |
2188 | */ | |
2189 | if (stat(cc_path, &sb) != 0 && errno == ENOENT) { | |
934b1673 | 2190 | mp->need_cpuset_init = true; |
d703c2b1 RV |
2191 | free(cc_path); |
2192 | return 0; | |
2193 | } | |
7e7243e1 SH |
2194 | r = lxc_read_from_file(cc_path, buf, 1); |
2195 | if (r == 1 && buf[0] == '1') { | |
2196 | free(cc_path); | |
2edb53c7 | 2197 | return 0; |
7e7243e1 | 2198 | } |
33ad9f1a | 2199 | r = lxc_write_to_file(cc_path, "1", 1, false); |
2edb53c7 SH |
2200 | saved_errno = errno; |
2201 | free(cc_path); | |
2202 | errno = saved_errno; | |
2203 | return r < 0 ? -1 : 0; | |
33ad9f1a CS |
2204 | } |
2205 | return 0; | |
b98f7d6e | 2206 | } |
484ed030 | 2207 | |
934b1673 | 2208 | static int cgroup_read_from_file(const char *fn, char buf[], size_t bufsize) |
d703c2b1 RV |
2209 | { |
2210 | int ret = lxc_read_from_file(fn, buf, bufsize); | |
2211 | if (ret < 0) { | |
2212 | SYSERROR("failed to read %s", fn); | |
934b1673 | 2213 | return ret; |
d703c2b1 RV |
2214 | } |
2215 | if (ret == bufsize) { | |
934b1673 SH |
2216 | if (bufsize > 0) { |
2217 | /* obviously this wasn't empty */ | |
2218 | buf[bufsize-1] = '\0'; | |
2219 | return ret; | |
2220 | } | |
2221 | /* Callers don't do this, but regression/sanity check */ | |
2222 | ERROR("%s: was not expecting 0 bufsize", __func__); | |
2223 | return -1; | |
d703c2b1 RV |
2224 | } |
2225 | buf[ret] = '\0'; | |
934b1673 | 2226 | return ret; |
d703c2b1 RV |
2227 | } |
2228 | ||
2229 | static bool do_init_cpuset_file(struct cgroup_mount_point *mp, | |
2230 | const char *path, const char *name) | |
2231 | { | |
934b1673 SH |
2232 | char value[1024]; |
2233 | char *childfile, *parentfile = NULL, *tmp; | |
2234 | int ret; | |
2235 | bool ok = false; | |
2236 | ||
d703c2b1 RV |
2237 | childfile = cgroup_to_absolute_path(mp, path, name); |
2238 | if (!childfile) | |
2239 | return false; | |
2240 | ||
2241 | /* don't overwrite a non-empty value in the file */ | |
934b1673 SH |
2242 | ret = cgroup_read_from_file(childfile, value, sizeof(value)); |
2243 | if (ret < 0) | |
2244 | goto out; | |
d703c2b1 | 2245 | if (value[0] != '\0' && value[0] != '\n') { |
934b1673 SH |
2246 | ok = true; |
2247 | goto out; | |
d703c2b1 RV |
2248 | } |
2249 | ||
2250 | /* path to the same name in the parent cgroup */ | |
2251 | parentfile = strdup(path); | |
2252 | if (!parentfile) | |
934b1673 SH |
2253 | goto out; |
2254 | ||
d703c2b1 | 2255 | tmp = strrchr(parentfile, '/'); |
934b1673 SH |
2256 | if (!tmp) |
2257 | goto out; | |
d703c2b1 RV |
2258 | if (tmp == parentfile) |
2259 | tmp++; /* keep the '/' at the start */ | |
2260 | *tmp = '\0'; | |
2261 | tmp = parentfile; | |
2262 | parentfile = cgroup_to_absolute_path(mp, tmp, name); | |
2263 | free(tmp); | |
934b1673 SH |
2264 | if (!parentfile) |
2265 | goto out; | |
d703c2b1 RV |
2266 | |
2267 | /* copy from parent to child cgroup */ | |
934b1673 SH |
2268 | ret = cgroup_read_from_file(parentfile, value, sizeof(value)); |
2269 | if (ret < 0) | |
2270 | goto out; | |
2271 | if (ret == sizeof(value)) { | |
2272 | /* If anyone actually sees this error, we can address it */ | |
2273 | ERROR("parent cpuset value too long"); | |
2274 | goto out; | |
d703c2b1 RV |
2275 | } |
2276 | ok = (lxc_write_to_file(childfile, value, strlen(value), false) >= 0); | |
2277 | if (!ok) | |
2278 | SYSERROR("failed writing %s", childfile); | |
b1dad6f6 RV |
2279 | |
2280 | out: | |
f10fad2f | 2281 | free(parentfile); |
d703c2b1 | 2282 | free(childfile); |
d703c2b1 RV |
2283 | return ok; |
2284 | } | |
2285 | ||
2286 | static bool init_cpuset_if_needed(struct cgroup_mount_point *mp, | |
2287 | const char *path) | |
2288 | { | |
2289 | /* the files we have to handle here are only in cpuset hierarchies */ | |
2290 | if (!lxc_string_in_array("cpuset", | |
2291 | (const char **)mp->hierarchy->subsystems)) | |
2292 | return true; | |
2293 | ||
b1dad6f6 RV |
2294 | if (!mp->need_cpuset_init) |
2295 | return true; | |
2296 | ||
d703c2b1 RV |
2297 | return (do_init_cpuset_file(mp, path, "/cpuset.cpus") && |
2298 | do_init_cpuset_file(mp, path, "/cpuset.mems") ); | |
2299 | } | |
2300 | ||
4fb3cba5 | 2301 | struct cgroup_ops *cgfs_ops_init(void) |
484ed030 | 2302 | { |
4fb3cba5 | 2303 | return &cgfs_ops; |
d4ef7c50 | 2304 | } |
484ed030 | 2305 | |
4fb3cba5 | 2306 | static void *cgfs_init(const char *name) |
d4ef7c50 | 2307 | { |
4fb3cba5 | 2308 | struct cgfs_data *d; |
484ed030 | 2309 | |
4fb3cba5 DE |
2310 | d = malloc(sizeof(*d)); |
2311 | if (!d) | |
2312 | return NULL; | |
484ed030 | 2313 | |
4fb3cba5 DE |
2314 | memset(d, 0, sizeof(*d)); |
2315 | d->name = strdup(name); | |
2316 | if (!d->name) | |
2317 | goto err1; | |
2318 | ||
5e1c5795 | 2319 | d->cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); |
4fb3cba5 DE |
2320 | |
2321 | d->meta = lxc_cgroup_load_meta(); | |
2322 | if (!d->meta) { | |
2323 | ERROR("cgroupfs failed to detect cgroup metadata"); | |
2324 | goto err2; | |
2325 | } | |
2326 | return d; | |
2327 | ||
2328 | err2: | |
2329 | free(d->name); | |
2330 | err1: | |
2331 | free(d); | |
2332 | return NULL; | |
d4ef7c50 | 2333 | } |
484ed030 | 2334 | |
6a9e0f26 | 2335 | static void cgfs_destroy(void *hdata, struct lxc_conf *conf) |
d4ef7c50 | 2336 | { |
4fb3cba5 DE |
2337 | struct cgfs_data *d = hdata; |
2338 | ||
d4ef7c50 SH |
2339 | if (!d) |
2340 | return; | |
f10fad2f | 2341 | free(d->name); |
6a9e0f26 | 2342 | lxc_cgroup_process_info_free_and_remove(d->info, conf); |
c55d4505 | 2343 | lxc_cgroup_put_meta(d->meta); |
d4ef7c50 | 2344 | free(d); |
d4ef7c50 | 2345 | } |
484ed030 | 2346 | |
4fb3cba5 | 2347 | static inline bool cgfs_create(void *hdata) |
d4ef7c50 | 2348 | { |
4fb3cba5 DE |
2349 | struct cgfs_data *d = hdata; |
2350 | struct cgroup_process_info *i; | |
2351 | struct cgroup_meta_data *md; | |
484ed030 | 2352 | |
4fb3cba5 | 2353 | if (!d) |
d4ef7c50 | 2354 | return false; |
4fb3cba5 DE |
2355 | md = d->meta; |
2356 | i = lxc_cgroupfs_create(d->name, d->cgroup_pattern, md, NULL); | |
d4ef7c50 SH |
2357 | if (!i) |
2358 | return false; | |
2359 | d->info = i; | |
2360 | return true; | |
2361 | } | |
484ed030 | 2362 | |
4fb3cba5 | 2363 | static inline bool cgfs_enter(void *hdata, pid_t pid) |
d4ef7c50 | 2364 | { |
4fb3cba5 DE |
2365 | struct cgfs_data *d = hdata; |
2366 | struct cgroup_process_info *i; | |
d4ef7c50 | 2367 | int ret; |
4fb3cba5 DE |
2368 | |
2369 | if (!d) | |
2370 | return false; | |
2371 | i = d->info; | |
2372 | ret = lxc_cgroupfs_enter(i, pid, false); | |
484ed030 | 2373 | |
d4ef7c50 SH |
2374 | return ret == 0; |
2375 | } | |
2376 | ||
4fb3cba5 | 2377 | static inline bool cgfs_create_legacy(void *hdata, pid_t pid) |
d4ef7c50 | 2378 | { |
4fb3cba5 DE |
2379 | struct cgfs_data *d = hdata; |
2380 | struct cgroup_process_info *i; | |
2381 | ||
2382 | if (!d) | |
2383 | return false; | |
2384 | i = d->info; | |
2385 | if (lxc_cgroup_create_legacy(i, d->name, pid) < 0) { | |
2386 | ERROR("failed to create legacy ns cgroups for '%s'", d->name); | |
d4ef7c50 | 2387 | return false; |
484ed030 | 2388 | } |
d4ef7c50 SH |
2389 | return true; |
2390 | } | |
484ed030 | 2391 | |
4fb3cba5 | 2392 | static const char *cgfs_get_cgroup(void *hdata, const char *subsystem) |
d4ef7c50 | 2393 | { |
4fb3cba5 DE |
2394 | struct cgfs_data *d = hdata; |
2395 | ||
2396 | if (!d) | |
2397 | return NULL; | |
2398 | return lxc_cgroup_get_hierarchy_path_data(subsystem, d); | |
484ed030 SH |
2399 | } |
2400 | ||
2ba7a429 TA |
2401 | static const char *cgfs_canonical_path(void *hdata) |
2402 | { | |
2403 | struct cgfs_data *d = hdata; | |
2404 | struct cgroup_process_info *info_ptr; | |
2405 | char *path = NULL; | |
2406 | ||
2407 | if (!d) | |
2408 | return NULL; | |
2409 | ||
2410 | for (info_ptr = d->info; info_ptr; info_ptr = info_ptr->next) { | |
2411 | if (!path) | |
2412 | path = info_ptr->cgroup_path; | |
2413 | else if (strcmp(path, info_ptr->cgroup_path) != 0) { | |
2414 | ERROR("not all paths match %s, %s has path %s", path, | |
2415 | info_ptr->hierarchy->subsystems[0], info_ptr->cgroup_path); | |
2416 | return NULL; | |
2417 | } | |
2418 | } | |
2419 | ||
2420 | return path; | |
2421 | } | |
2422 | ||
06078509 TA |
2423 | static bool cgfs_escape(void) |
2424 | { | |
2425 | struct cgroup_meta_data *md; | |
2426 | int i; | |
2427 | bool ret = false; | |
2428 | ||
2429 | md = lxc_cgroup_load_meta(); | |
2430 | if (!md) | |
2431 | return false; | |
2432 | ||
2433 | for (i = 1; i <= md->maximum_hierarchy; i++) { | |
2434 | struct cgroup_hierarchy *h = md->hierarchies[i]; | |
2435 | struct cgroup_mount_point *mp; | |
2436 | char *tasks; | |
2437 | FILE *f; | |
2438 | int written; | |
2439 | ||
2440 | if (!h) { | |
2441 | WARN("not escaping hierarchy %d", i); | |
2442 | continue; | |
2443 | } | |
2444 | ||
2445 | mp = lxc_cgroup_find_mount_point(h, "/", true); | |
2446 | if (!mp) | |
2447 | goto out; | |
2448 | ||
2449 | tasks = cgroup_to_absolute_path(mp, "/", "tasks"); | |
2450 | if (!tasks) | |
2451 | goto out; | |
2452 | ||
2453 | f = fopen(tasks, "a"); | |
2454 | free(tasks); | |
2455 | if (!f) | |
2456 | goto out; | |
2457 | ||
2458 | written = fprintf(f, "%d\n", getpid()); | |
2459 | fclose(f); | |
2460 | if (written < 0) { | |
2461 | SYSERROR("writing tasks failed\n"); | |
2462 | goto out; | |
2463 | } | |
2464 | } | |
2465 | ||
2466 | ret = true; | |
2467 | out: | |
2468 | lxc_cgroup_put_meta(md); | |
2469 | return ret; | |
2470 | } | |
2471 | ||
4fb3cba5 | 2472 | static bool cgfs_unfreeze(void *hdata) |
0086f499 | 2473 | { |
4fb3cba5 | 2474 | struct cgfs_data *d = hdata; |
0086f499 SH |
2475 | char *cgabspath, *cgrelpath; |
2476 | int ret; | |
2477 | ||
4fb3cba5 DE |
2478 | if (!d) |
2479 | return false; | |
2480 | ||
2481 | cgrelpath = lxc_cgroup_get_hierarchy_path_data("freezer", d); | |
0086f499 SH |
2482 | cgabspath = lxc_cgroup_find_abs_path("freezer", cgrelpath, true, NULL); |
2483 | if (!cgabspath) | |
ecfcb3f0 | 2484 | return false; |
0086f499 SH |
2485 | |
2486 | ret = do_cgroup_set(cgabspath, "freezer.state", "THAWED"); | |
2487 | free(cgabspath); | |
ecfcb3f0 | 2488 | return ret == 0; |
0086f499 SH |
2489 | } |
2490 | ||
4fb3cba5 DE |
2491 | static bool cgroupfs_setup_limits(void *hdata, struct lxc_list *cgroup_conf, |
2492 | bool with_devices) | |
9daf6f5d | 2493 | { |
4fb3cba5 DE |
2494 | struct cgfs_data *d = hdata; |
2495 | ||
2496 | if (!d) | |
2497 | return false; | |
2498 | return do_setup_cgroup_limits(d, cgroup_conf, with_devices) == 0; | |
9daf6f5d SH |
2499 | } |
2500 | ||
4fb3cba5 | 2501 | static bool lxc_cgroupfs_attach(const char *name, const char *lxcpath, pid_t pid) |
5d897655 SH |
2502 | { |
2503 | struct cgroup_meta_data *meta_data; | |
2504 | struct cgroup_process_info *container_info; | |
2505 | int ret; | |
2506 | ||
2507 | meta_data = lxc_cgroup_load_meta(); | |
2508 | if (!meta_data) { | |
2509 | ERROR("could not move attached process %d to cgroup of container", pid); | |
2510 | return false; | |
2511 | } | |
2512 | ||
2513 | container_info = lxc_cgroup_get_container_info(name, lxcpath, meta_data); | |
2514 | lxc_cgroup_put_meta(meta_data); | |
2515 | if (!container_info) { | |
2516 | ERROR("could not move attached process %d to cgroup of container", pid); | |
2517 | return false; | |
2518 | } | |
2519 | ||
2520 | ret = lxc_cgroupfs_enter(container_info, pid, false); | |
2521 | lxc_cgroup_process_info_free(container_info); | |
2522 | if (ret < 0) { | |
2523 | ERROR("could not move attached process %d to cgroup of container", pid); | |
2524 | return false; | |
2525 | } | |
2526 | return true; | |
2527 | } | |
2528 | ||
8b276860 SH |
2529 | struct chown_data { |
2530 | const char *cgroup_path; | |
2531 | uid_t origuid; | |
2532 | }; | |
2533 | ||
2534 | /* | |
2535 | * TODO - someone should refactor this to unshare once passing all the paths | |
2536 | * to be chowned in one go | |
2537 | */ | |
2538 | static int chown_cgroup_wrapper(void *data) | |
2539 | { | |
2540 | struct chown_data *arg = data; | |
2541 | uid_t destuid; | |
2542 | char *fpath; | |
2543 | ||
8b276860 SH |
2544 | if (setresgid(0,0,0) < 0) |
2545 | SYSERROR("Failed to setgid to 0"); | |
2546 | if (setresuid(0,0,0) < 0) | |
2547 | SYSERROR("Failed to setuid to 0"); | |
2548 | if (setgroups(0, NULL) < 0) | |
2549 | SYSERROR("Failed to clear groups"); | |
2550 | destuid = get_ns_uid(arg->origuid); | |
2551 | ||
2552 | if (chown(arg->cgroup_path, destuid, 0) < 0) | |
2553 | SYSERROR("Failed chowning %s to %d", arg->cgroup_path, (int)destuid); | |
2554 | ||
2555 | fpath = lxc_append_paths(arg->cgroup_path, "tasks"); | |
2556 | if (!fpath) | |
2557 | return -1; | |
2558 | if (chown(fpath, destuid, 0) < 0) | |
2559 | SYSERROR("Error chowning %s\n", fpath); | |
2560 | free(fpath); | |
01d59fe5 | 2561 | |
8b276860 SH |
2562 | fpath = lxc_append_paths(arg->cgroup_path, "cgroup.procs"); |
2563 | if (!fpath) | |
2564 | return -1; | |
2565 | if (chown(fpath, destuid, 0) < 0) | |
2566 | SYSERROR("Error chowning %s", fpath); | |
2567 | free(fpath); | |
2568 | ||
2569 | return 0; | |
2570 | } | |
2571 | ||
2572 | static bool do_cgfs_chown(char *cgroup_path, struct lxc_conf *conf) | |
2573 | { | |
2574 | struct chown_data data; | |
2575 | char *fpath; | |
2576 | ||
01d59fe5 CB |
2577 | if (!dir_exists(cgroup_path)) |
2578 | return true; | |
2579 | ||
8b276860 SH |
2580 | if (lxc_list_empty(&conf->id_map)) |
2581 | /* If there's no mapping then we don't need to chown */ | |
2582 | return true; | |
2583 | ||
2584 | data.cgroup_path = cgroup_path; | |
2585 | data.origuid = geteuid(); | |
2586 | ||
2587 | /* Unpriv users can't chown it themselves, so chown from | |
2588 | * a child namespace mapping both our own and the target uid | |
2589 | */ | |
2590 | if (userns_exec_1(conf, chown_cgroup_wrapper, &data) < 0) { | |
2591 | ERROR("Error requesting cgroup chown in new namespace"); | |
2592 | return false; | |
2593 | } | |
2594 | ||
2595 | /* | |
2596 | * Now chmod 775 the directory else the container cannot create cgroups. | |
2597 | * This can't be done in the child namespace because it only group-owns | |
2598 | * the cgroup | |
2599 | */ | |
2600 | if (chmod(cgroup_path, 0775) < 0) { | |
2601 | SYSERROR("Error chmoding %s\n", cgroup_path); | |
2602 | return false; | |
2603 | } | |
2604 | fpath = lxc_append_paths(cgroup_path, "tasks"); | |
2605 | if (!fpath) | |
2606 | return false; | |
2607 | if (chmod(fpath, 0664) < 0) | |
2608 | SYSERROR("Error chmoding %s\n", fpath); | |
2609 | free(fpath); | |
2610 | fpath = lxc_append_paths(cgroup_path, "cgroup.procs"); | |
2611 | if (!fpath) | |
2612 | return false; | |
2613 | if (chmod(fpath, 0664) < 0) | |
2614 | SYSERROR("Error chmoding %s\n", fpath); | |
2615 | free(fpath); | |
2616 | ||
2617 | return true; | |
2618 | } | |
2619 | ||
2620 | static bool cgfs_chown(void *hdata, struct lxc_conf *conf) | |
2621 | { | |
2622 | struct cgfs_data *d = hdata; | |
2623 | struct cgroup_process_info *info_ptr; | |
2624 | char *cgpath; | |
2625 | bool r = true; | |
2626 | ||
2627 | if (!d) | |
2628 | return false; | |
2629 | ||
2630 | for (info_ptr = d->info; info_ptr; info_ptr = info_ptr->next) { | |
2631 | if (!info_ptr->designated_mount_point) { | |
2632 | info_ptr->designated_mount_point = lxc_cgroup_find_mount_point(info_ptr->hierarchy, info_ptr->cgroup_path, true); | |
2633 | if (!info_ptr->designated_mount_point) { | |
2634 | SYSERROR("Could not chown cgroup %s: internal error (couldn't find any writable mountpoint to cgroup filesystem)", info_ptr->cgroup_path); | |
2635 | return false; | |
2636 | } | |
2637 | } | |
2638 | ||
2639 | cgpath = cgroup_to_absolute_path(info_ptr->designated_mount_point, info_ptr->cgroup_path, NULL); | |
2640 | if (!cgpath) { | |
2641 | SYSERROR("Could not chown cgroup %s: internal error", info_ptr->cgroup_path); | |
2642 | continue; | |
2643 | } | |
2644 | r = do_cgfs_chown(cgpath, conf); | |
ea439aac | 2645 | if (!r && is_crucial_hierarchy(info_ptr->hierarchy)) { |
8b276860 SH |
2646 | ERROR("Failed chowning %s\n", cgpath); |
2647 | free(cgpath); | |
2648 | return false; | |
2649 | } | |
2650 | free(cgpath); | |
2651 | } | |
2652 | ||
2653 | return true; | |
2654 | } | |
2655 | ||
d4ef7c50 | 2656 | static struct cgroup_ops cgfs_ops = { |
d4ef7c50 | 2657 | .init = cgfs_init, |
4fb3cba5 | 2658 | .destroy = cgfs_destroy, |
d4ef7c50 SH |
2659 | .create = cgfs_create, |
2660 | .enter = cgfs_enter, | |
2661 | .create_legacy = cgfs_create_legacy, | |
2662 | .get_cgroup = cgfs_get_cgroup, | |
2ba7a429 | 2663 | .canonical_path = cgfs_canonical_path, |
06078509 | 2664 | .escape = cgfs_escape, |
d4ef7c50 SH |
2665 | .get = lxc_cgroupfs_get, |
2666 | .set = lxc_cgroupfs_set, | |
4fb3cba5 | 2667 | .unfreeze = cgfs_unfreeze, |
9daf6f5d | 2668 | .setup_limits = cgroupfs_setup_limits, |
d4ef7c50 | 2669 | .name = "cgroupfs", |
5d897655 | 2670 | .attach = lxc_cgroupfs_attach, |
8b276860 | 2671 | .chown = cgfs_chown, |
c476bdce | 2672 | .mount_cgroup = cgroupfs_mount_cgroup, |
4fb3cba5 | 2673 | .nrtasks = cgfs_nrtasks, |
23befb18 | 2674 | .driver = CGFS, |
d4ef7c50 | 2675 | }; |