]>
Commit | Line | Data |
---|---|---|
d6c9574f | 1 | /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/ |
8e274523 LP |
2 | |
3 | /*** | |
4 | This file is part of systemd. | |
5 | ||
4ad49000 | 6 | Copyright 2013 Lennart Poettering |
8e274523 LP |
7 | |
8 | systemd is free software; you can redistribute it and/or modify it | |
5430f7f2 LP |
9 | under the terms of the GNU Lesser General Public License as published by |
10 | the Free Software Foundation; either version 2.1 of the License, or | |
8e274523 LP |
11 | (at your option) any later version. |
12 | ||
13 | systemd is distributed in the hope that it will be useful, but | |
14 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
5430f7f2 | 16 | Lesser General Public License for more details. |
8e274523 | 17 | |
5430f7f2 | 18 | You should have received a copy of the GNU Lesser General Public License |
8e274523 LP |
19 | along with systemd; If not, see <http://www.gnu.org/licenses/>. |
20 | ***/ | |
21 | ||
c6c18be3 | 22 | #include <fcntl.h> |
e41969e3 | 23 | #include <fnmatch.h> |
8c6db833 | 24 | |
9eb977db | 25 | #include "path-util.h" |
9444b1f2 | 26 | #include "special.h" |
4ad49000 LP |
27 | #include "cgroup-util.h" |
28 | #include "cgroup.h" | |
8e274523 | 29 | |
4ad49000 LP |
30 | void cgroup_context_init(CGroupContext *c) { |
31 | assert(c); | |
32 | ||
33 | /* Initialize everything to the kernel defaults, assuming the | |
34 | * structure is preinitialized to 0 */ | |
35 | ||
36 | c->cpu_shares = 1024; | |
ddca82ac | 37 | c->memory_limit = (uint64_t) -1; |
4ad49000 LP |
38 | c->blockio_weight = 1000; |
39 | } | |
8e274523 | 40 | |
4ad49000 LP |
41 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
42 | assert(c); | |
43 | assert(a); | |
44 | ||
71fda00f | 45 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
46 | free(a->path); |
47 | free(a); | |
48 | } | |
49 | ||
50 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { | |
51 | assert(c); | |
52 | assert(w); | |
53 | ||
71fda00f | 54 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
55 | free(w->path); |
56 | free(w); | |
57 | } | |
58 | ||
59 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
60 | assert(c); | |
8e274523 | 61 | assert(b); |
8e274523 | 62 | |
71fda00f | 63 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
64 | free(b->path); |
65 | free(b); | |
66 | } | |
67 | ||
68 | void cgroup_context_done(CGroupContext *c) { | |
69 | assert(c); | |
70 | ||
71 | while (c->blockio_device_weights) | |
72 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
73 | ||
74 | while (c->blockio_device_bandwidths) | |
75 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
76 | ||
77 | while (c->device_allow) | |
78 | cgroup_context_free_device_allow(c, c->device_allow); | |
79 | } | |
80 | ||
81 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
82 | CGroupBlockIODeviceBandwidth *b; | |
83 | CGroupBlockIODeviceWeight *w; | |
84 | CGroupDeviceAllow *a; | |
85 | ||
86 | assert(c); | |
87 | assert(f); | |
88 | ||
89 | prefix = strempty(prefix); | |
90 | ||
91 | fprintf(f, | |
92 | "%sCPUAccounting=%s\n" | |
93 | "%sBlockIOAccounting=%s\n" | |
94 | "%sMemoryAccounting=%s\n" | |
95 | "%sCPUShares=%lu\n" | |
112a7f46 | 96 | "%sBlockIOWeight=%lu\n" |
4ad49000 | 97 | "%sMemoryLimit=%" PRIu64 "\n" |
4ad49000 LP |
98 | "%sDevicePolicy=%s\n", |
99 | prefix, yes_no(c->cpu_accounting), | |
100 | prefix, yes_no(c->blockio_accounting), | |
101 | prefix, yes_no(c->memory_accounting), | |
102 | prefix, c->cpu_shares, | |
103 | prefix, c->blockio_weight, | |
104 | prefix, c->memory_limit, | |
4ad49000 LP |
105 | prefix, cgroup_device_policy_to_string(c->device_policy)); |
106 | ||
107 | LIST_FOREACH(device_allow, a, c->device_allow) | |
108 | fprintf(f, | |
109 | "%sDeviceAllow=%s %s%s%s\n", | |
110 | prefix, | |
111 | a->path, | |
112 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
113 | ||
114 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
115 | fprintf(f, | |
8e7076ca | 116 | "%sBlockIODeviceWeight=%s %lu", |
4ad49000 LP |
117 | prefix, |
118 | w->path, | |
119 | w->weight); | |
120 | ||
121 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
122 | char buf[FORMAT_BYTES_MAX]; | |
123 | ||
124 | fprintf(f, | |
125 | "%s%s=%s %s\n", | |
126 | prefix, | |
127 | b->read ? "BlockIOReadBandwidth" : "BlockIOWriteBandwidth", | |
128 | b->path, | |
129 | format_bytes(buf, sizeof(buf), b->bandwidth)); | |
130 | } | |
131 | } | |
132 | ||
133 | static int lookup_blkio_device(const char *p, dev_t *dev) { | |
134 | struct stat st; | |
135 | int r; | |
136 | ||
137 | assert(p); | |
138 | assert(dev); | |
139 | ||
140 | r = stat(p, &st); | |
ab1f0633 | 141 | if (r < 0) { |
4ad49000 LP |
142 | log_warning("Couldn't stat device %s: %m", p); |
143 | return -errno; | |
ab1f0633 | 144 | } |
8e274523 | 145 | |
4ad49000 LP |
146 | if (S_ISBLK(st.st_mode)) |
147 | *dev = st.st_rdev; | |
148 | else if (major(st.st_dev) != 0) { | |
149 | /* If this is not a device node then find the block | |
150 | * device this file is stored on */ | |
151 | *dev = st.st_dev; | |
152 | ||
153 | /* If this is a partition, try to get the originating | |
154 | * block device */ | |
155 | block_get_whole_disk(*dev, dev); | |
156 | } else { | |
157 | log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p); | |
158 | return -ENODEV; | |
159 | } | |
8e274523 | 160 | |
8e274523 | 161 | return 0; |
8e274523 LP |
162 | } |
163 | ||
4ad49000 LP |
164 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
165 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
166 | struct stat st; | |
8c6db833 | 167 | int r; |
8e274523 | 168 | |
4ad49000 LP |
169 | assert(path); |
170 | assert(acc); | |
8e274523 | 171 | |
4ad49000 LP |
172 | if (stat(node, &st) < 0) { |
173 | log_warning("Couldn't stat device %s", node); | |
174 | return -errno; | |
175 | } | |
176 | ||
177 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
178 | log_warning("%s is not a device.", node); | |
179 | return -ENODEV; | |
180 | } | |
181 | ||
182 | sprintf(buf, | |
183 | "%c %u:%u %s", | |
184 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
185 | major(st.st_rdev), minor(st.st_rdev), | |
186 | acc); | |
187 | ||
188 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
189 | if (r < 0) | |
190 | log_warning("Failed to set devices.allow on %s: %s", path, strerror(-r)); | |
191 | ||
192 | return r; | |
8e274523 LP |
193 | } |
194 | ||
90060676 LP |
195 | static int whitelist_major(const char *path, const char *name, char type, const char *acc) { |
196 | _cleanup_fclose_ FILE *f = NULL; | |
197 | char line[LINE_MAX]; | |
198 | bool good = false; | |
199 | int r; | |
200 | ||
201 | assert(path); | |
202 | assert(acc); | |
203 | assert(type == 'b' || type == 'c'); | |
204 | ||
205 | f = fopen("/proc/devices", "re"); | |
206 | if (!f) { | |
207 | log_warning("Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
208 | return -errno; | |
209 | } | |
210 | ||
211 | FOREACH_LINE(line, f, goto fail) { | |
212 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; | |
213 | unsigned maj; | |
214 | ||
215 | truncate_nl(line); | |
216 | ||
217 | if (type == 'c' && streq(line, "Character devices:")) { | |
218 | good = true; | |
219 | continue; | |
220 | } | |
221 | ||
222 | if (type == 'b' && streq(line, "Block devices:")) { | |
223 | good = true; | |
224 | continue; | |
225 | } | |
226 | ||
227 | if (isempty(line)) { | |
228 | good = false; | |
229 | continue; | |
230 | } | |
231 | ||
232 | if (!good) | |
233 | continue; | |
234 | ||
235 | p = strstrip(line); | |
236 | ||
237 | w = strpbrk(p, WHITESPACE); | |
238 | if (!w) | |
239 | continue; | |
240 | *w = 0; | |
241 | ||
242 | r = safe_atou(p, &maj); | |
243 | if (r < 0) | |
244 | continue; | |
245 | if (maj <= 0) | |
246 | continue; | |
247 | ||
248 | w++; | |
249 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
250 | |
251 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
252 | continue; |
253 | ||
254 | sprintf(buf, | |
255 | "%c %u:* %s", | |
256 | type, | |
257 | maj, | |
258 | acc); | |
259 | ||
260 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
261 | if (r < 0) | |
262 | log_warning("Failed to set devices.allow on %s: %s", path, strerror(-r)); | |
263 | } | |
264 | ||
265 | return 0; | |
266 | ||
267 | fail: | |
268 | log_warning("Failed to read /proc/devices: %m"); | |
269 | return -errno; | |
270 | } | |
271 | ||
4ad49000 | 272 | void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const char *path) { |
01efdf13 | 273 | bool is_root; |
4ad49000 LP |
274 | int r; |
275 | ||
276 | assert(c); | |
277 | assert(path); | |
8e274523 | 278 | |
4ad49000 LP |
279 | if (mask == 0) |
280 | return; | |
8e274523 | 281 | |
01efdf13 LP |
282 | /* Some cgroup attributes are not support on the root cgroup, |
283 | * hence silently ignore */ | |
284 | is_root = isempty(path) || path_equal(path, "/"); | |
285 | ||
286 | if ((mask & CGROUP_CPU) && !is_root) { | |
4ad49000 | 287 | char buf[DECIMAL_STR_MAX(unsigned long) + 1]; |
8e274523 | 288 | |
4ad49000 LP |
289 | sprintf(buf, "%lu\n", c->cpu_shares); |
290 | r = cg_set_attribute("cpu", path, "cpu.shares", buf); | |
291 | if (r < 0) | |
292 | log_warning("Failed to set cpu.shares on %s: %s", path, strerror(-r)); | |
293 | } | |
294 | ||
295 | if (mask & CGROUP_BLKIO) { | |
296 | char buf[MAX3(DECIMAL_STR_MAX(unsigned long)+1, | |
297 | DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(unsigned long)*1, | |
298 | DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1)]; | |
299 | CGroupBlockIODeviceWeight *w; | |
300 | CGroupBlockIODeviceBandwidth *b; | |
301 | ||
01efdf13 LP |
302 | if (!is_root) { |
303 | sprintf(buf, "%lu\n", c->blockio_weight); | |
304 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); | |
305 | if (r < 0) | |
306 | log_warning("Failed to set blkio.weight on %s: %s", path, strerror(-r)); | |
4ad49000 | 307 | |
01efdf13 LP |
308 | /* FIXME: no way to reset this list */ |
309 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { | |
310 | dev_t dev; | |
4ad49000 | 311 | |
01efdf13 LP |
312 | r = lookup_blkio_device(w->path, &dev); |
313 | if (r < 0) | |
314 | continue; | |
8e274523 | 315 | |
01efdf13 LP |
316 | sprintf(buf, "%u:%u %lu", major(dev), minor(dev), w->weight); |
317 | r = cg_set_attribute("blkio", path, "blkio.weight_device", buf); | |
318 | if (r < 0) | |
319 | log_error("Failed to set blkio.weight_device on %s: %s", path, strerror(-r)); | |
320 | } | |
4ad49000 LP |
321 | } |
322 | ||
323 | /* FIXME: no way to reset this list */ | |
324 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
325 | const char *a; | |
326 | dev_t dev; | |
327 | ||
328 | r = lookup_blkio_device(b->path, &dev); | |
329 | if (r < 0) | |
330 | continue; | |
331 | ||
332 | a = b->read ? "blkio.throttle.read_bps_device" : "blkio.throttle.write_bps_device"; | |
333 | ||
334 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), b->bandwidth); | |
335 | r = cg_set_attribute("blkio", path, a, buf); | |
336 | if (r < 0) | |
337 | log_error("Failed to set %s on %s: %s", a, path, strerror(-r)); | |
d686d8a9 | 338 | } |
8e274523 LP |
339 | } |
340 | ||
4ad49000 | 341 | if (mask & CGROUP_MEMORY) { |
6a94f2e9 | 342 | if (c->memory_limit != (uint64_t) -1) { |
e58cec11 LP |
343 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
344 | ||
6a94f2e9 G |
345 | sprintf(buf, "%" PRIu64 "\n", c->memory_limit); |
346 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); | |
347 | } else | |
348 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", "-1"); | |
8e274523 | 349 | |
4ad49000 LP |
350 | if (r < 0) |
351 | log_error("Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r)); | |
4ad49000 | 352 | } |
8e274523 | 353 | |
01efdf13 | 354 | if ((mask & CGROUP_DEVICE) && !is_root) { |
4ad49000 | 355 | CGroupDeviceAllow *a; |
8e274523 | 356 | |
4ad49000 LP |
357 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
358 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
359 | else | |
360 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
361 | if (r < 0) | |
01efdf13 | 362 | log_warning("Failed to reset devices.list on %s: %s", path, strerror(-r)); |
fb385181 | 363 | |
4ad49000 LP |
364 | if (c->device_policy == CGROUP_CLOSED || |
365 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
366 | static const char auto_devices[] = | |
7d711efb LP |
367 | "/dev/null\0" "rwm\0" |
368 | "/dev/zero\0" "rwm\0" | |
369 | "/dev/full\0" "rwm\0" | |
370 | "/dev/random\0" "rwm\0" | |
371 | "/dev/urandom\0" "rwm\0" | |
372 | "/dev/tty\0" "rwm\0" | |
373 | "/dev/pts/ptmx\0" "rw\0"; /* /dev/pts/ptmx may not be duplicated, but accessed */ | |
4ad49000 LP |
374 | |
375 | const char *x, *y; | |
376 | ||
377 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
378 | whitelist_device(path, x, y); | |
7d711efb LP |
379 | |
380 | whitelist_major(path, "pts", 'c', "rw"); | |
381 | whitelist_major(path, "kdbus", 'c', "rw"); | |
382 | whitelist_major(path, "kdbus/*", 'c', "rw"); | |
4ad49000 LP |
383 | } |
384 | ||
385 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
386 | char acc[4]; | |
387 | unsigned k = 0; | |
388 | ||
389 | if (a->r) | |
390 | acc[k++] = 'r'; | |
391 | if (a->w) | |
392 | acc[k++] = 'w'; | |
393 | if (a->m) | |
394 | acc[k++] = 'm'; | |
fb385181 | 395 | |
4ad49000 LP |
396 | if (k == 0) |
397 | continue; | |
fb385181 | 398 | |
4ad49000 | 399 | acc[k++] = 0; |
90060676 LP |
400 | |
401 | if (startswith(a->path, "/dev/")) | |
402 | whitelist_device(path, a->path, acc); | |
403 | else if (startswith(a->path, "block-")) | |
404 | whitelist_major(path, a->path + 6, 'b', acc); | |
405 | else if (startswith(a->path, "char-")) | |
406 | whitelist_major(path, a->path + 5, 'c', acc); | |
407 | else | |
408 | log_debug("Ignoring device %s while writing cgroup attribute.", a->path); | |
4ad49000 LP |
409 | } |
410 | } | |
fb385181 LP |
411 | } |
412 | ||
4ad49000 LP |
413 | CGroupControllerMask cgroup_context_get_mask(CGroupContext *c) { |
414 | CGroupControllerMask mask = 0; | |
8e274523 | 415 | |
4ad49000 | 416 | /* Figure out which controllers we need */ |
8e274523 | 417 | |
4ad49000 LP |
418 | if (c->cpu_accounting || c->cpu_shares != 1024) |
419 | mask |= CGROUP_CPUACCT | CGROUP_CPU; | |
ecedd90f | 420 | |
4ad49000 LP |
421 | if (c->blockio_accounting || |
422 | c->blockio_weight != 1000 || | |
423 | c->blockio_device_weights || | |
424 | c->blockio_device_bandwidths) | |
425 | mask |= CGROUP_BLKIO; | |
ecedd90f | 426 | |
4ad49000 | 427 | if (c->memory_accounting || |
ddca82ac | 428 | c->memory_limit != (uint64_t) -1) |
4ad49000 | 429 | mask |= CGROUP_MEMORY; |
8e274523 | 430 | |
4ad49000 LP |
431 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
432 | mask |= CGROUP_DEVICE; | |
433 | ||
434 | return mask; | |
8e274523 LP |
435 | } |
436 | ||
bc432dc7 | 437 | CGroupControllerMask unit_get_cgroup_mask(Unit *u) { |
4ad49000 | 438 | CGroupContext *c; |
8e274523 | 439 | |
4ad49000 LP |
440 | c = unit_get_cgroup_context(u); |
441 | if (!c) | |
442 | return 0; | |
8e274523 | 443 | |
4ad49000 | 444 | return cgroup_context_get_mask(c); |
8e274523 LP |
445 | } |
446 | ||
bc432dc7 | 447 | CGroupControllerMask unit_get_members_mask(Unit *u) { |
4ad49000 | 448 | assert(u); |
bc432dc7 LP |
449 | |
450 | if (u->cgroup_members_mask_valid) | |
451 | return u->cgroup_members_mask; | |
452 | ||
453 | u->cgroup_members_mask = 0; | |
454 | ||
455 | if (u->type == UNIT_SLICE) { | |
456 | Unit *member; | |
457 | Iterator i; | |
458 | ||
459 | SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) { | |
460 | ||
461 | if (member == u) | |
462 | continue; | |
463 | ||
d4fdc205 | 464 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
465 | continue; |
466 | ||
467 | u->cgroup_members_mask |= | |
468 | unit_get_cgroup_mask(member) | | |
469 | unit_get_members_mask(member); | |
470 | } | |
471 | } | |
472 | ||
473 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 474 | return u->cgroup_members_mask; |
246aa6dd LP |
475 | } |
476 | ||
bc432dc7 LP |
477 | CGroupControllerMask unit_get_siblings_mask(Unit *u) { |
478 | CGroupControllerMask m; | |
479 | ||
4ad49000 | 480 | assert(u); |
246aa6dd | 481 | |
bc432dc7 LP |
482 | if (UNIT_ISSET(u->slice)) |
483 | m = unit_get_members_mask(UNIT_DEREF(u->slice)); | |
484 | else | |
485 | m = unit_get_cgroup_mask(u) | unit_get_members_mask(u); | |
4ad49000 LP |
486 | |
487 | /* Sibling propagation is only relevant for weight-based | |
488 | * controllers, so let's mask out everything else */ | |
bc432dc7 | 489 | return m & (CGROUP_CPU|CGROUP_BLKIO|CGROUP_CPUACCT); |
246aa6dd LP |
490 | } |
491 | ||
bc432dc7 | 492 | CGroupControllerMask unit_get_target_mask(Unit *u) { |
6414b7c9 DS |
493 | CGroupControllerMask mask; |
494 | ||
495 | mask = unit_get_cgroup_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); | |
496 | mask &= u->manager->cgroup_supported; | |
497 | ||
498 | return mask; | |
499 | } | |
500 | ||
501 | /* Recurse from a unit up through its containing slices, propagating | |
502 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 LP |
503 | void unit_update_cgroup_members_masks(Unit *u) { |
504 | CGroupControllerMask m; | |
505 | bool more; | |
506 | ||
507 | assert(u); | |
508 | ||
509 | /* Calculate subtree mask */ | |
510 | m = unit_get_cgroup_mask(u) | unit_get_members_mask(u); | |
511 | ||
512 | /* See if anything changed from the previous invocation. If | |
513 | * not, we're done. */ | |
514 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
515 | return; | |
516 | ||
517 | more = | |
518 | u->cgroup_subtree_mask_valid && | |
519 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
520 | ((~m & u->cgroup_subtree_mask) == 0); | |
521 | ||
522 | u->cgroup_subtree_mask = m; | |
523 | u->cgroup_subtree_mask_valid = true; | |
524 | ||
6414b7c9 DS |
525 | if (UNIT_ISSET(u->slice)) { |
526 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
527 | |
528 | if (more) | |
529 | /* There's more set now than before. We | |
530 | * propagate the new mask to the parent's mask | |
531 | * (not caring if it actually was valid or | |
532 | * not). */ | |
533 | ||
534 | s->cgroup_members_mask |= m; | |
535 | ||
536 | else | |
537 | /* There's less set now than before (or we | |
538 | * don't know), we need to recalculate | |
539 | * everything, so let's invalidate the | |
540 | * parent's members mask */ | |
541 | ||
542 | s->cgroup_members_mask_valid = false; | |
543 | ||
544 | /* And now make sure that this change also hits our | |
545 | * grandparents */ | |
546 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
547 | } |
548 | } | |
549 | ||
03b90d4b LP |
550 | static const char *migrate_callback(CGroupControllerMask mask, void *userdata) { |
551 | Unit *u = userdata; | |
552 | ||
553 | assert(mask != 0); | |
554 | assert(u); | |
555 | ||
556 | while (u) { | |
557 | if (u->cgroup_path && | |
558 | u->cgroup_realized && | |
559 | (u->cgroup_realized_mask & mask) == mask) | |
560 | return u->cgroup_path; | |
561 | ||
562 | u = UNIT_DEREF(u->slice); | |
563 | } | |
564 | ||
565 | return NULL; | |
566 | } | |
567 | ||
4ad49000 | 568 | static int unit_create_cgroups(Unit *u, CGroupControllerMask mask) { |
03b90d4b | 569 | _cleanup_free_ char *path = NULL; |
bc432dc7 | 570 | int r; |
64747e2d | 571 | |
4ad49000 | 572 | assert(u); |
64747e2d | 573 | |
4ad49000 LP |
574 | path = unit_default_cgroup_path(u); |
575 | if (!path) | |
a94042fa | 576 | return log_oom(); |
64747e2d | 577 | |
0a1eb06d | 578 | r = hashmap_put(u->manager->cgroup_unit, path, u); |
03b90d4b LP |
579 | if (r < 0) { |
580 | log_error(r == -EEXIST ? "cgroup %s exists already: %s" : "hashmap_put failed for %s: %s", path, strerror(-r)); | |
0a1eb06d | 581 | return r; |
b58b8e11 | 582 | } |
03b90d4b | 583 | if (r > 0) { |
b58b8e11 | 584 | u->cgroup_path = path; |
a94042fa | 585 | path = NULL; |
b58b8e11 HH |
586 | } |
587 | ||
03b90d4b LP |
588 | /* First, create our own group */ |
589 | r = cg_create_everywhere(u->manager->cgroup_supported, mask, u->cgroup_path); | |
590 | if (r < 0) { | |
591 | log_error("Failed to create cgroup %s: %s", u->cgroup_path, strerror(-r)); | |
592 | return r; | |
593 | } | |
594 | ||
595 | /* Keep track that this is now realized */ | |
4ad49000 | 596 | u->cgroup_realized = true; |
bc432dc7 | 597 | u->cgroup_realized_mask = mask; |
4ad49000 | 598 | |
03b90d4b LP |
599 | /* Then, possibly move things over */ |
600 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
601 | if (r < 0) | |
602 | log_warning("Failed to migrate cgroup from to %s: %s", u->cgroup_path, strerror(-r)); | |
603 | ||
64747e2d LP |
604 | return 0; |
605 | } | |
606 | ||
6414b7c9 | 607 | static bool unit_has_mask_realized(Unit *u, CGroupControllerMask mask) { |
bc432dc7 LP |
608 | assert(u); |
609 | ||
610 | return u->cgroup_realized && u->cgroup_realized_mask == mask; | |
6414b7c9 DS |
611 | } |
612 | ||
613 | /* Check if necessary controllers and attributes for a unit are in place. | |
614 | * | |
615 | * If so, do nothing. | |
616 | * If not, create paths, move processes over, and set attributes. | |
617 | * | |
618 | * Returns 0 on success and < 0 on failure. */ | |
0a1eb06d | 619 | static int unit_realize_cgroup_now(Unit *u) { |
4ad49000 | 620 | CGroupControllerMask mask; |
6414b7c9 | 621 | int r; |
64747e2d | 622 | |
4ad49000 | 623 | assert(u); |
64747e2d | 624 | |
4ad49000 | 625 | if (u->in_cgroup_queue) { |
71fda00f | 626 | LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u); |
4ad49000 LP |
627 | u->in_cgroup_queue = false; |
628 | } | |
64747e2d | 629 | |
6414b7c9 | 630 | mask = unit_get_target_mask(u); |
64747e2d | 631 | |
6414b7c9 | 632 | if (unit_has_mask_realized(u, mask)) |
0a1eb06d | 633 | return 0; |
64747e2d | 634 | |
4ad49000 | 635 | /* First, realize parents */ |
6414b7c9 DS |
636 | if (UNIT_ISSET(u->slice)) { |
637 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice)); | |
638 | if (r < 0) | |
639 | return r; | |
640 | } | |
4ad49000 LP |
641 | |
642 | /* And then do the real work */ | |
6414b7c9 DS |
643 | r = unit_create_cgroups(u, mask); |
644 | if (r < 0) | |
645 | return r; | |
646 | ||
647 | /* Finally, apply the necessary attributes. */ | |
648 | cgroup_context_apply(unit_get_cgroup_context(u), mask, u->cgroup_path); | |
649 | ||
650 | return 0; | |
64747e2d LP |
651 | } |
652 | ||
4ad49000 | 653 | static void unit_add_to_cgroup_queue(Unit *u) { |
ecedd90f | 654 | |
4ad49000 LP |
655 | if (u->in_cgroup_queue) |
656 | return; | |
8e274523 | 657 | |
71fda00f | 658 | LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u); |
4ad49000 LP |
659 | u->in_cgroup_queue = true; |
660 | } | |
8c6db833 | 661 | |
4ad49000 LP |
662 | unsigned manager_dispatch_cgroup_queue(Manager *m) { |
663 | Unit *i; | |
664 | unsigned n = 0; | |
6414b7c9 | 665 | int r; |
ecedd90f | 666 | |
4ad49000 LP |
667 | while ((i = m->cgroup_queue)) { |
668 | assert(i->in_cgroup_queue); | |
ecedd90f | 669 | |
6414b7c9 DS |
670 | r = unit_realize_cgroup_now(i); |
671 | if (r < 0) | |
672 | log_warning("Failed to realize cgroups for queued unit %s: %s", i->id, strerror(-r)); | |
0a1eb06d | 673 | |
4ad49000 LP |
674 | n++; |
675 | } | |
ecedd90f | 676 | |
4ad49000 | 677 | return n; |
8e274523 LP |
678 | } |
679 | ||
4ad49000 LP |
680 | static void unit_queue_siblings(Unit *u) { |
681 | Unit *slice; | |
ca949c9d | 682 | |
4ad49000 LP |
683 | /* This adds the siblings of the specified unit and the |
684 | * siblings of all parent units to the cgroup queue. (But | |
685 | * neither the specified unit itself nor the parents.) */ | |
686 | ||
687 | while ((slice = UNIT_DEREF(u->slice))) { | |
688 | Iterator i; | |
689 | Unit *m; | |
8f53a7b8 | 690 | |
4ad49000 LP |
691 | SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) { |
692 | if (m == u) | |
693 | continue; | |
8e274523 | 694 | |
6414b7c9 DS |
695 | /* Skip units that have a dependency on the slice |
696 | * but aren't actually in it. */ | |
4ad49000 | 697 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 698 | continue; |
8e274523 | 699 | |
6414b7c9 DS |
700 | /* No point in doing cgroup application for units |
701 | * without active processes. */ | |
702 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
703 | continue; | |
704 | ||
705 | /* If the unit doesn't need any new controllers | |
706 | * and has current ones realized, it doesn't need | |
707 | * any changes. */ | |
708 | if (unit_has_mask_realized(m, unit_get_target_mask(m))) | |
709 | continue; | |
710 | ||
4ad49000 | 711 | unit_add_to_cgroup_queue(m); |
50159e6a LP |
712 | } |
713 | ||
4ad49000 | 714 | u = slice; |
8e274523 | 715 | } |
4ad49000 LP |
716 | } |
717 | ||
0a1eb06d | 718 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
719 | CGroupContext *c; |
720 | ||
721 | assert(u); | |
722 | ||
723 | c = unit_get_cgroup_context(u); | |
724 | if (!c) | |
0a1eb06d | 725 | return 0; |
8e274523 | 726 | |
4ad49000 LP |
727 | /* So, here's the deal: when realizing the cgroups for this |
728 | * unit, we need to first create all parents, but there's more | |
729 | * actually: for the weight-based controllers we also need to | |
730 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 731 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
732 | * would become very uneven as each of their processes would |
733 | * get as much resources as all our group together. This call | |
734 | * will synchronously create the parent cgroups, but will | |
735 | * defer work on the siblings to the next event loop | |
736 | * iteration. */ | |
ca949c9d | 737 | |
4ad49000 LP |
738 | /* Add all sibling slices to the cgroup queue. */ |
739 | unit_queue_siblings(u); | |
740 | ||
6414b7c9 | 741 | /* And realize this one now (and apply the values) */ |
bc432dc7 | 742 | return unit_realize_cgroup_now(u); |
8e274523 LP |
743 | } |
744 | ||
4ad49000 | 745 | void unit_destroy_cgroup(Unit *u) { |
8e274523 LP |
746 | int r; |
747 | ||
4ad49000 | 748 | assert(u); |
8e274523 | 749 | |
4ad49000 LP |
750 | if (!u->cgroup_path) |
751 | return; | |
8e274523 | 752 | |
13b84ec7 | 753 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !unit_has_name(u, SPECIAL_ROOT_SLICE)); |
4ad49000 | 754 | if (r < 0) |
376dd21d | 755 | log_debug("Failed to destroy cgroup %s: %s", u->cgroup_path, strerror(-r)); |
8e274523 | 756 | |
0a1eb06d LP |
757 | hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); |
758 | ||
4ad49000 LP |
759 | free(u->cgroup_path); |
760 | u->cgroup_path = NULL; | |
761 | u->cgroup_realized = false; | |
bc432dc7 | 762 | u->cgroup_realized_mask = 0; |
0a1eb06d | 763 | |
8e274523 LP |
764 | } |
765 | ||
4ad49000 LP |
766 | pid_t unit_search_main_pid(Unit *u) { |
767 | _cleanup_fclose_ FILE *f = NULL; | |
768 | pid_t pid = 0, npid, mypid; | |
769 | ||
770 | assert(u); | |
771 | ||
772 | if (!u->cgroup_path) | |
773 | return 0; | |
774 | ||
775 | if (cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f) < 0) | |
776 | return 0; | |
777 | ||
778 | mypid = getpid(); | |
779 | while (cg_read_pid(f, &npid) > 0) { | |
780 | pid_t ppid; | |
781 | ||
782 | if (npid == pid) | |
783 | continue; | |
8e274523 | 784 | |
4ad49000 LP |
785 | /* Ignore processes that aren't our kids */ |
786 | if (get_parent_of_pid(npid, &ppid) >= 0 && ppid != mypid) | |
787 | continue; | |
8e274523 | 788 | |
4ad49000 LP |
789 | if (pid != 0) { |
790 | /* Dang, there's more than one daemonized PID | |
791 | in this group, so we don't know what process | |
792 | is the main process. */ | |
793 | pid = 0; | |
794 | break; | |
795 | } | |
8e274523 | 796 | |
4ad49000 | 797 | pid = npid; |
8e274523 LP |
798 | } |
799 | ||
4ad49000 | 800 | return pid; |
8e274523 LP |
801 | } |
802 | ||
8e274523 | 803 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 804 | _cleanup_free_ char *path = NULL; |
15c60e99 | 805 | char *e; |
8e274523 | 806 | int r; |
8e274523 LP |
807 | |
808 | assert(m); | |
809 | ||
e5a53dc7 | 810 | /* 0. Be nice to Ingo Molnar #628004 */ |
0c85a4f3 | 811 | if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) { |
e5a53dc7 LP |
812 | log_warning("No control group support available, not creating root group."); |
813 | return 0; | |
814 | } | |
815 | ||
35d2e7ec | 816 | /* 1. Determine hierarchy */ |
9444b1f2 LP |
817 | free(m->cgroup_root); |
818 | m->cgroup_root = NULL; | |
819 | ||
820 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); | |
9156e799 | 821 | if (r < 0) { |
12235040 | 822 | log_error("Cannot determine cgroup we are running in: %s", strerror(-r)); |
a32360f1 | 823 | return r; |
12235040 | 824 | } |
8e274523 | 825 | |
15c60e99 LP |
826 | /* LEGACY: Already in /system.slice? If so, let's cut this |
827 | * off. This is to support live upgrades from older systemd | |
828 | * versions where PID 1 was moved there. */ | |
9444b1f2 LP |
829 | if (m->running_as == SYSTEMD_SYSTEM) { |
830 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); | |
15c60e99 LP |
831 | if (!e) |
832 | e = endswith(m->cgroup_root, "/system"); | |
9444b1f2 LP |
833 | if (e) |
834 | *e = 0; | |
0baf24dd | 835 | } |
7ccfb64a | 836 | |
9444b1f2 LP |
837 | /* And make sure to store away the root value without trailing |
838 | * slash, even for the root dir, so that we can easily prepend | |
839 | * it everywhere. */ | |
840 | if (streq(m->cgroup_root, "/")) | |
841 | m->cgroup_root[0] = 0; | |
8e274523 | 842 | |
35d2e7ec | 843 | /* 2. Show data */ |
9444b1f2 | 844 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
3474ae3c | 845 | if (r < 0) { |
12235040 | 846 | log_error("Cannot find cgroup mount point: %s", strerror(-r)); |
a32360f1 | 847 | return r; |
12235040 | 848 | } |
8e274523 | 849 | |
c6c18be3 LP |
850 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path); |
851 | ||
35d2e7ec | 852 | /* 3. Install agent */ |
a32360f1 LP |
853 | if (m->running_as == SYSTEMD_SYSTEM) { |
854 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); | |
855 | if (r < 0) | |
856 | log_warning("Failed to install release agent, ignoring: %s", strerror(-r)); | |
857 | else if (r > 0) | |
858 | log_debug("Installed release agent."); | |
859 | else | |
860 | log_debug("Release agent already installed."); | |
861 | } | |
8e274523 | 862 | |
15c60e99 LP |
863 | /* 4. Make sure we are in the root cgroup */ |
864 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, 0); | |
9156e799 | 865 | if (r < 0) { |
8e274523 | 866 | log_error("Failed to create root cgroup hierarchy: %s", strerror(-r)); |
a32360f1 | 867 | return r; |
c6c18be3 LP |
868 | } |
869 | ||
35d2e7ec | 870 | /* 5. And pin it, so that it cannot be unmounted */ |
03e334a1 | 871 | safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 872 | |
9156e799 LP |
873 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); |
874 | if (r < 0) { | |
12235040 | 875 | log_error("Failed to open pin file: %m"); |
a32360f1 | 876 | return -errno; |
c6c18be3 LP |
877 | } |
878 | ||
4ad49000 LP |
879 | /* 6. Figure out which controllers are supported */ |
880 | m->cgroup_supported = cg_mask_supported(); | |
9156e799 | 881 | |
e58cec11 LP |
882 | /* 7. Always enable hierarchial support if it exists... */ |
883 | cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
884 | ||
a32360f1 | 885 | return 0; |
8e274523 LP |
886 | } |
887 | ||
c6c18be3 | 888 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
889 | assert(m); |
890 | ||
9444b1f2 LP |
891 | /* We can't really delete the group, since we are in it. But |
892 | * let's trim it. */ | |
893 | if (delete && m->cgroup_root) | |
894 | cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); | |
8e274523 | 895 | |
03e334a1 | 896 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 897 | |
9444b1f2 LP |
898 | free(m->cgroup_root); |
899 | m->cgroup_root = NULL; | |
8e274523 LP |
900 | } |
901 | ||
4ad49000 | 902 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 903 | char *p; |
4ad49000 | 904 | Unit *u; |
acb14d31 LP |
905 | |
906 | assert(m); | |
907 | assert(cgroup); | |
acb14d31 | 908 | |
4ad49000 LP |
909 | u = hashmap_get(m->cgroup_unit, cgroup); |
910 | if (u) | |
911 | return u; | |
acb14d31 | 912 | |
8e70580b | 913 | p = strdupa(cgroup); |
acb14d31 LP |
914 | for (;;) { |
915 | char *e; | |
916 | ||
917 | e = strrchr(p, '/'); | |
4ad49000 LP |
918 | if (e == p || !e) |
919 | return NULL; | |
acb14d31 LP |
920 | |
921 | *e = 0; | |
922 | ||
4ad49000 LP |
923 | u = hashmap_get(m->cgroup_unit, p); |
924 | if (u) | |
925 | return u; | |
acb14d31 LP |
926 | } |
927 | } | |
928 | ||
4ad49000 LP |
929 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { |
930 | _cleanup_free_ char *cgroup = NULL; | |
acb14d31 | 931 | int r; |
8e274523 | 932 | |
8c47c732 LP |
933 | assert(m); |
934 | ||
935 | if (pid <= 1) | |
936 | return NULL; | |
937 | ||
4ad49000 LP |
938 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup); |
939 | if (r < 0) | |
6dde1f33 LP |
940 | return NULL; |
941 | ||
4ad49000 | 942 | return manager_get_unit_by_cgroup(m, cgroup); |
6dde1f33 | 943 | } |
4fbf50b3 | 944 | |
4ad49000 LP |
945 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
946 | Unit *u; | |
947 | int r; | |
4fbf50b3 | 948 | |
4ad49000 LP |
949 | assert(m); |
950 | assert(cgroup); | |
4fbf50b3 | 951 | |
4ad49000 | 952 | u = manager_get_unit_by_cgroup(m, cgroup); |
b56c28c3 | 953 | if (u) { |
06025d91 LP |
954 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, true); |
955 | if (r > 0) { | |
956 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
957 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
b56c28c3 | 958 | |
06025d91 LP |
959 | unit_add_to_gc_queue(u); |
960 | } | |
b56c28c3 | 961 | } |
2633eb83 | 962 | |
4ad49000 | 963 | return 0; |
4fbf50b3 LP |
964 | } |
965 | ||
4ad49000 LP |
966 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
967 | [CGROUP_AUTO] = "auto", | |
968 | [CGROUP_CLOSED] = "closed", | |
969 | [CGROUP_STRICT] = "strict", | |
970 | }; | |
4fbf50b3 | 971 | |
4ad49000 | 972 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |