]>
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> |
8c6db833 | 23 | |
9eb977db | 24 | #include "path-util.h" |
9444b1f2 | 25 | #include "special.h" |
4ad49000 LP |
26 | #include "cgroup-util.h" |
27 | #include "cgroup.h" | |
8e274523 | 28 | |
4ad49000 LP |
29 | void cgroup_context_init(CGroupContext *c) { |
30 | assert(c); | |
31 | ||
32 | /* Initialize everything to the kernel defaults, assuming the | |
33 | * structure is preinitialized to 0 */ | |
34 | ||
35 | c->cpu_shares = 1024; | |
36 | c->memory_limit = c->memory_soft_limit = (uint64_t) -1; | |
37 | c->blockio_weight = 1000; | |
38 | } | |
8e274523 | 39 | |
4ad49000 LP |
40 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
41 | assert(c); | |
42 | assert(a); | |
43 | ||
44 | LIST_REMOVE(CGroupDeviceAllow, device_allow, c->device_allow, a); | |
45 | free(a->path); | |
46 | free(a); | |
47 | } | |
48 | ||
49 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { | |
50 | assert(c); | |
51 | assert(w); | |
52 | ||
53 | LIST_REMOVE(CGroupBlockIODeviceWeight, device_weights, c->blockio_device_weights, w); | |
54 | free(w->path); | |
55 | free(w); | |
56 | } | |
57 | ||
58 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
59 | assert(c); | |
8e274523 | 60 | assert(b); |
8e274523 | 61 | |
4ad49000 LP |
62 | LIST_REMOVE(CGroupBlockIODeviceBandwidth, device_bandwidths, c->blockio_device_bandwidths, b); |
63 | free(b->path); | |
64 | free(b); | |
65 | } | |
66 | ||
67 | void cgroup_context_done(CGroupContext *c) { | |
68 | assert(c); | |
69 | ||
70 | while (c->blockio_device_weights) | |
71 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
72 | ||
73 | while (c->blockio_device_bandwidths) | |
74 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
75 | ||
76 | while (c->device_allow) | |
77 | cgroup_context_free_device_allow(c, c->device_allow); | |
78 | } | |
79 | ||
80 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
81 | CGroupBlockIODeviceBandwidth *b; | |
82 | CGroupBlockIODeviceWeight *w; | |
83 | CGroupDeviceAllow *a; | |
84 | ||
85 | assert(c); | |
86 | assert(f); | |
87 | ||
88 | prefix = strempty(prefix); | |
89 | ||
90 | fprintf(f, | |
91 | "%sCPUAccounting=%s\n" | |
92 | "%sBlockIOAccounting=%s\n" | |
93 | "%sMemoryAccounting=%s\n" | |
94 | "%sCPUShares=%lu\n" | |
95 | "%sBlockIOWeight%lu\n" | |
96 | "%sMemoryLimit=%" PRIu64 "\n" | |
97 | "%sMemorySoftLimit=%" PRIu64 "\n" | |
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, | |
105 | prefix, c->memory_soft_limit, | |
106 | prefix, cgroup_device_policy_to_string(c->device_policy)); | |
107 | ||
108 | LIST_FOREACH(device_allow, a, c->device_allow) | |
109 | fprintf(f, | |
110 | "%sDeviceAllow=%s %s%s%s\n", | |
111 | prefix, | |
112 | a->path, | |
113 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
114 | ||
115 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
116 | fprintf(f, | |
8e7076ca | 117 | "%sBlockIODeviceWeight=%s %lu", |
4ad49000 LP |
118 | prefix, |
119 | w->path, | |
120 | w->weight); | |
121 | ||
122 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
123 | char buf[FORMAT_BYTES_MAX]; | |
124 | ||
125 | fprintf(f, | |
126 | "%s%s=%s %s\n", | |
127 | prefix, | |
128 | b->read ? "BlockIOReadBandwidth" : "BlockIOWriteBandwidth", | |
129 | b->path, | |
130 | format_bytes(buf, sizeof(buf), b->bandwidth)); | |
131 | } | |
132 | } | |
133 | ||
134 | static int lookup_blkio_device(const char *p, dev_t *dev) { | |
135 | struct stat st; | |
136 | int r; | |
137 | ||
138 | assert(p); | |
139 | assert(dev); | |
140 | ||
141 | r = stat(p, &st); | |
ab1f0633 | 142 | if (r < 0) { |
4ad49000 LP |
143 | log_warning("Couldn't stat device %s: %m", p); |
144 | return -errno; | |
ab1f0633 | 145 | } |
8e274523 | 146 | |
4ad49000 LP |
147 | if (S_ISBLK(st.st_mode)) |
148 | *dev = st.st_rdev; | |
149 | else if (major(st.st_dev) != 0) { | |
150 | /* If this is not a device node then find the block | |
151 | * device this file is stored on */ | |
152 | *dev = st.st_dev; | |
153 | ||
154 | /* If this is a partition, try to get the originating | |
155 | * block device */ | |
156 | block_get_whole_disk(*dev, dev); | |
157 | } else { | |
158 | log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p); | |
159 | return -ENODEV; | |
160 | } | |
8e274523 | 161 | |
8e274523 | 162 | return 0; |
8e274523 LP |
163 | } |
164 | ||
4ad49000 LP |
165 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
166 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
167 | struct stat st; | |
8c6db833 | 168 | int r; |
8e274523 | 169 | |
4ad49000 LP |
170 | assert(path); |
171 | assert(acc); | |
8e274523 | 172 | |
4ad49000 LP |
173 | if (stat(node, &st) < 0) { |
174 | log_warning("Couldn't stat device %s", node); | |
175 | return -errno; | |
176 | } | |
177 | ||
178 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
179 | log_warning("%s is not a device.", node); | |
180 | return -ENODEV; | |
181 | } | |
182 | ||
183 | sprintf(buf, | |
184 | "%c %u:%u %s", | |
185 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
186 | major(st.st_rdev), minor(st.st_rdev), | |
187 | acc); | |
188 | ||
189 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
190 | if (r < 0) | |
191 | log_warning("Failed to set devices.allow on %s: %s", path, strerror(-r)); | |
192 | ||
193 | return r; | |
8e274523 LP |
194 | } |
195 | ||
4ad49000 LP |
196 | void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const char *path) { |
197 | int r; | |
198 | ||
199 | assert(c); | |
200 | assert(path); | |
8e274523 | 201 | |
4ad49000 LP |
202 | if (mask == 0) |
203 | return; | |
8e274523 | 204 | |
4ad49000 LP |
205 | if (mask & CGROUP_CPU) { |
206 | char buf[DECIMAL_STR_MAX(unsigned long) + 1]; | |
8e274523 | 207 | |
4ad49000 LP |
208 | sprintf(buf, "%lu\n", c->cpu_shares); |
209 | r = cg_set_attribute("cpu", path, "cpu.shares", buf); | |
210 | if (r < 0) | |
211 | log_warning("Failed to set cpu.shares on %s: %s", path, strerror(-r)); | |
212 | } | |
213 | ||
214 | if (mask & CGROUP_BLKIO) { | |
215 | char buf[MAX3(DECIMAL_STR_MAX(unsigned long)+1, | |
216 | DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(unsigned long)*1, | |
217 | DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1)]; | |
218 | CGroupBlockIODeviceWeight *w; | |
219 | CGroupBlockIODeviceBandwidth *b; | |
220 | ||
221 | sprintf(buf, "%lu\n", c->blockio_weight); | |
222 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); | |
223 | if (r < 0) | |
224 | log_warning("Failed to set blkio.weight on %s: %s", path, strerror(-r)); | |
225 | ||
226 | /* FIXME: no way to reset this list */ | |
227 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { | |
228 | dev_t dev; | |
229 | ||
230 | r = lookup_blkio_device(w->path, &dev); | |
231 | if (r < 0) | |
232 | continue; | |
8e274523 | 233 | |
4ad49000 LP |
234 | sprintf(buf, "%u:%u %lu", major(dev), minor(dev), w->weight); |
235 | r = cg_set_attribute("blkio", path, "blkio.weight_device", buf); | |
236 | if (r < 0) | |
237 | log_error("Failed to set blkio.weight_device on %s: %s", path, strerror(-r)); | |
238 | } | |
239 | ||
240 | /* FIXME: no way to reset this list */ | |
241 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
242 | const char *a; | |
243 | dev_t dev; | |
244 | ||
245 | r = lookup_blkio_device(b->path, &dev); | |
246 | if (r < 0) | |
247 | continue; | |
248 | ||
249 | a = b->read ? "blkio.throttle.read_bps_device" : "blkio.throttle.write_bps_device"; | |
250 | ||
251 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), b->bandwidth); | |
252 | r = cg_set_attribute("blkio", path, a, buf); | |
253 | if (r < 0) | |
254 | log_error("Failed to set %s on %s: %s", a, path, strerror(-r)); | |
d686d8a9 | 255 | } |
8e274523 LP |
256 | } |
257 | ||
4ad49000 LP |
258 | if (mask & CGROUP_MEMORY) { |
259 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
8e274523 | 260 | |
4ad49000 LP |
261 | sprintf(buf, "%" PRIu64 "\n", c->memory_limit); |
262 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); | |
263 | if (r < 0) | |
264 | log_error("Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r)); | |
8e274523 | 265 | |
4ad49000 | 266 | sprintf(buf, "%" PRIu64 "\n", c->memory_soft_limit); |
15b4a754 | 267 | r = cg_set_attribute("memory", path, "memory.soft_limit_in_bytes", buf); |
4ad49000 | 268 | if (r < 0) |
84121bc2 | 269 | log_error("Failed to set memory.soft_limit_in_bytes on %s: %s", path, strerror(-r)); |
4ad49000 | 270 | } |
8e274523 | 271 | |
4ad49000 LP |
272 | if (mask & CGROUP_DEVICE) { |
273 | CGroupDeviceAllow *a; | |
8e274523 | 274 | |
4ad49000 LP |
275 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
276 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
277 | else | |
278 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
279 | if (r < 0) | |
280 | log_error("Failed to reset devices.list on %s: %s", path, strerror(-r)); | |
fb385181 | 281 | |
4ad49000 LP |
282 | if (c->device_policy == CGROUP_CLOSED || |
283 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
284 | static const char auto_devices[] = | |
285 | "/dev/null\0" "rw\0" | |
286 | "/dev/zero\0" "rw\0" | |
287 | "/dev/full\0" "rw\0" | |
288 | "/dev/random\0" "rw\0" | |
289 | "/dev/urandom\0" "rw\0"; | |
290 | ||
291 | const char *x, *y; | |
292 | ||
293 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
294 | whitelist_device(path, x, y); | |
295 | } | |
296 | ||
297 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
298 | char acc[4]; | |
299 | unsigned k = 0; | |
300 | ||
301 | if (a->r) | |
302 | acc[k++] = 'r'; | |
303 | if (a->w) | |
304 | acc[k++] = 'w'; | |
305 | if (a->m) | |
306 | acc[k++] = 'm'; | |
fb385181 | 307 | |
4ad49000 LP |
308 | if (k == 0) |
309 | continue; | |
fb385181 | 310 | |
4ad49000 LP |
311 | acc[k++] = 0; |
312 | whitelist_device(path, a->path, acc); | |
313 | } | |
314 | } | |
fb385181 LP |
315 | } |
316 | ||
4ad49000 LP |
317 | CGroupControllerMask cgroup_context_get_mask(CGroupContext *c) { |
318 | CGroupControllerMask mask = 0; | |
8e274523 | 319 | |
4ad49000 | 320 | /* Figure out which controllers we need */ |
8e274523 | 321 | |
4ad49000 LP |
322 | if (c->cpu_accounting || c->cpu_shares != 1024) |
323 | mask |= CGROUP_CPUACCT | CGROUP_CPU; | |
ecedd90f | 324 | |
4ad49000 LP |
325 | if (c->blockio_accounting || |
326 | c->blockio_weight != 1000 || | |
327 | c->blockio_device_weights || | |
328 | c->blockio_device_bandwidths) | |
329 | mask |= CGROUP_BLKIO; | |
ecedd90f | 330 | |
4ad49000 LP |
331 | if (c->memory_accounting || |
332 | c->memory_limit != (uint64_t) -1 || | |
333 | c->memory_soft_limit != (uint64_t) -1) | |
334 | mask |= CGROUP_MEMORY; | |
8e274523 | 335 | |
4ad49000 LP |
336 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
337 | mask |= CGROUP_DEVICE; | |
338 | ||
339 | return mask; | |
8e274523 LP |
340 | } |
341 | ||
4ad49000 LP |
342 | static CGroupControllerMask unit_get_cgroup_mask(Unit *u) { |
343 | CGroupContext *c; | |
8e274523 | 344 | |
4ad49000 LP |
345 | c = unit_get_cgroup_context(u); |
346 | if (!c) | |
347 | return 0; | |
8e274523 | 348 | |
4ad49000 | 349 | return cgroup_context_get_mask(c); |
8e274523 LP |
350 | } |
351 | ||
4ad49000 LP |
352 | static CGroupControllerMask unit_get_members_mask(Unit *u) { |
353 | CGroupControllerMask mask = 0; | |
354 | Unit *m; | |
355 | Iterator i; | |
246aa6dd | 356 | |
4ad49000 | 357 | assert(u); |
246aa6dd | 358 | |
4ad49000 | 359 | SET_FOREACH(m, u->dependencies[UNIT_BEFORE], i) { |
246aa6dd | 360 | |
4ad49000 | 361 | if (UNIT_DEREF(m->slice) != u) |
246aa6dd LP |
362 | continue; |
363 | ||
4ad49000 | 364 | mask |= unit_get_cgroup_mask(m) | unit_get_members_mask(m); |
246aa6dd LP |
365 | } |
366 | ||
4ad49000 | 367 | return mask; |
246aa6dd LP |
368 | } |
369 | ||
4ad49000 LP |
370 | static CGroupControllerMask unit_get_siblings_mask(Unit *u) { |
371 | assert(u); | |
246aa6dd | 372 | |
4ad49000 LP |
373 | if (!UNIT_ISSET(u->slice)) |
374 | return 0; | |
375 | ||
376 | /* Sibling propagation is only relevant for weight-based | |
377 | * controllers, so let's mask out everything else */ | |
378 | return unit_get_members_mask(UNIT_DEREF(u->slice)) & | |
379 | (CGROUP_CPU|CGROUP_BLKIO|CGROUP_CPUACCT); | |
246aa6dd LP |
380 | } |
381 | ||
4ad49000 LP |
382 | static int unit_create_cgroups(Unit *u, CGroupControllerMask mask) { |
383 | char *path = NULL; | |
384 | int r; | |
b58b8e11 | 385 | bool is_in_hash = false; |
64747e2d | 386 | |
4ad49000 | 387 | assert(u); |
64747e2d | 388 | |
4ad49000 LP |
389 | path = unit_default_cgroup_path(u); |
390 | if (!path) | |
391 | return -ENOMEM; | |
64747e2d | 392 | |
0a1eb06d | 393 | r = hashmap_put(u->manager->cgroup_unit, path, u); |
b58b8e11 HH |
394 | if (r == 0) |
395 | is_in_hash = true; | |
396 | ||
397 | if (r < 0) { | |
398 | free(path); | |
399 | log_error("cgroup %s exists already: %s", path, strerror(-r)); | |
0a1eb06d | 400 | return r; |
b58b8e11 | 401 | } |
0a1eb06d | 402 | |
4ad49000 LP |
403 | /* First, create our own group */ |
404 | r = cg_create_with_mask(mask, path); | |
405 | if (r < 0) | |
406 | log_error("Failed to create cgroup %s: %s", path, strerror(-r)); | |
64747e2d | 407 | |
4ad49000 LP |
408 | /* Then, possibly move things over */ |
409 | if (u->cgroup_path && !streq(path, u->cgroup_path)) { | |
410 | r = cg_migrate_with_mask(mask, u->cgroup_path, path); | |
64747e2d | 411 | if (r < 0) |
4ad49000 | 412 | log_error("Failed to migrate cgroup %s: %s", path, strerror(-r)); |
64747e2d LP |
413 | } |
414 | ||
b58b8e11 HH |
415 | if (!is_in_hash) { |
416 | /* And remember the new data */ | |
417 | free(u->cgroup_path); | |
418 | u->cgroup_path = path; | |
419 | } | |
420 | ||
4ad49000 LP |
421 | u->cgroup_realized = true; |
422 | u->cgroup_mask = mask; | |
423 | ||
64747e2d LP |
424 | return 0; |
425 | } | |
426 | ||
0a1eb06d | 427 | static int unit_realize_cgroup_now(Unit *u) { |
4ad49000 | 428 | CGroupControllerMask mask; |
64747e2d | 429 | |
4ad49000 | 430 | assert(u); |
64747e2d | 431 | |
4ad49000 LP |
432 | if (u->in_cgroup_queue) { |
433 | LIST_REMOVE(Unit, cgroup_queue, u->manager->cgroup_queue, u); | |
434 | u->in_cgroup_queue = false; | |
435 | } | |
64747e2d | 436 | |
4ad49000 LP |
437 | mask = unit_get_cgroup_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
438 | mask &= u->manager->cgroup_supported; | |
64747e2d | 439 | |
4ad49000 LP |
440 | if (u->cgroup_realized && |
441 | u->cgroup_mask == mask) | |
0a1eb06d | 442 | return 0; |
64747e2d | 443 | |
4ad49000 | 444 | /* First, realize parents */ |
3d040cf2 HH |
445 | if (UNIT_ISSET(u->slice)) |
446 | unit_realize_cgroup_now(UNIT_DEREF(u->slice)); | |
4ad49000 LP |
447 | |
448 | /* And then do the real work */ | |
0a1eb06d | 449 | return unit_create_cgroups(u, mask); |
64747e2d LP |
450 | } |
451 | ||
4ad49000 | 452 | static void unit_add_to_cgroup_queue(Unit *u) { |
ecedd90f | 453 | |
4ad49000 LP |
454 | if (u->in_cgroup_queue) |
455 | return; | |
8e274523 | 456 | |
4ad49000 LP |
457 | LIST_PREPEND(Unit, cgroup_queue, u->manager->cgroup_queue, u); |
458 | u->in_cgroup_queue = true; | |
459 | } | |
8c6db833 | 460 | |
4ad49000 LP |
461 | unsigned manager_dispatch_cgroup_queue(Manager *m) { |
462 | Unit *i; | |
463 | unsigned n = 0; | |
ecedd90f | 464 | |
4ad49000 LP |
465 | while ((i = m->cgroup_queue)) { |
466 | assert(i->in_cgroup_queue); | |
ecedd90f | 467 | |
0a1eb06d LP |
468 | if (unit_realize_cgroup_now(i) >= 0) |
469 | cgroup_context_apply(unit_get_cgroup_context(i), i->cgroup_mask, i->cgroup_path); | |
470 | ||
4ad49000 LP |
471 | n++; |
472 | } | |
ecedd90f | 473 | |
4ad49000 | 474 | return n; |
8e274523 LP |
475 | } |
476 | ||
4ad49000 LP |
477 | static void unit_queue_siblings(Unit *u) { |
478 | Unit *slice; | |
ca949c9d | 479 | |
4ad49000 LP |
480 | /* This adds the siblings of the specified unit and the |
481 | * siblings of all parent units to the cgroup queue. (But | |
482 | * neither the specified unit itself nor the parents.) */ | |
483 | ||
484 | while ((slice = UNIT_DEREF(u->slice))) { | |
485 | Iterator i; | |
486 | Unit *m; | |
8f53a7b8 | 487 | |
4ad49000 LP |
488 | SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) { |
489 | if (m == u) | |
490 | continue; | |
8e274523 | 491 | |
4ad49000 | 492 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 493 | continue; |
8e274523 | 494 | |
4ad49000 | 495 | unit_add_to_cgroup_queue(m); |
50159e6a LP |
496 | } |
497 | ||
4ad49000 | 498 | u = slice; |
8e274523 | 499 | } |
4ad49000 LP |
500 | } |
501 | ||
0a1eb06d | 502 | int unit_realize_cgroup(Unit *u) { |
4ad49000 | 503 | CGroupContext *c; |
0a1eb06d | 504 | int r; |
4ad49000 LP |
505 | |
506 | assert(u); | |
507 | ||
508 | c = unit_get_cgroup_context(u); | |
509 | if (!c) | |
0a1eb06d | 510 | return 0; |
8e274523 | 511 | |
4ad49000 LP |
512 | /* So, here's the deal: when realizing the cgroups for this |
513 | * unit, we need to first create all parents, but there's more | |
514 | * actually: for the weight-based controllers we also need to | |
515 | * make sure that all our siblings (i.e. units that are in the | |
516 | * same slice as we are) have cgroup too. Otherwise things | |
517 | * would become very uneven as each of their processes would | |
518 | * get as much resources as all our group together. This call | |
519 | * will synchronously create the parent cgroups, but will | |
520 | * defer work on the siblings to the next event loop | |
521 | * iteration. */ | |
ca949c9d | 522 | |
4ad49000 LP |
523 | /* Add all sibling slices to the cgroup queue. */ |
524 | unit_queue_siblings(u); | |
525 | ||
526 | /* And realize this one now */ | |
0a1eb06d | 527 | r = unit_realize_cgroup_now(u); |
4ad49000 LP |
528 | |
529 | /* And apply the values */ | |
0a1eb06d LP |
530 | if (r >= 0) |
531 | cgroup_context_apply(c, u->cgroup_mask, u->cgroup_path); | |
532 | ||
533 | return r; | |
8e274523 LP |
534 | } |
535 | ||
4ad49000 | 536 | void unit_destroy_cgroup(Unit *u) { |
8e274523 LP |
537 | int r; |
538 | ||
4ad49000 | 539 | assert(u); |
8e274523 | 540 | |
4ad49000 LP |
541 | if (!u->cgroup_path) |
542 | return; | |
8e274523 | 543 | |
8a841929 | 544 | r = cg_trim_with_mask(u->cgroup_mask, u->cgroup_path, !unit_has_name(u, SPECIAL_ROOT_SLICE)); |
4ad49000 | 545 | if (r < 0) |
376dd21d | 546 | log_debug("Failed to destroy cgroup %s: %s", u->cgroup_path, strerror(-r)); |
8e274523 | 547 | |
0a1eb06d LP |
548 | hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); |
549 | ||
4ad49000 LP |
550 | free(u->cgroup_path); |
551 | u->cgroup_path = NULL; | |
552 | u->cgroup_realized = false; | |
553 | u->cgroup_mask = 0; | |
0a1eb06d | 554 | |
8e274523 LP |
555 | } |
556 | ||
4ad49000 LP |
557 | pid_t unit_search_main_pid(Unit *u) { |
558 | _cleanup_fclose_ FILE *f = NULL; | |
559 | pid_t pid = 0, npid, mypid; | |
560 | ||
561 | assert(u); | |
562 | ||
563 | if (!u->cgroup_path) | |
564 | return 0; | |
565 | ||
566 | if (cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f) < 0) | |
567 | return 0; | |
568 | ||
569 | mypid = getpid(); | |
570 | while (cg_read_pid(f, &npid) > 0) { | |
571 | pid_t ppid; | |
572 | ||
573 | if (npid == pid) | |
574 | continue; | |
8e274523 | 575 | |
4ad49000 LP |
576 | /* Ignore processes that aren't our kids */ |
577 | if (get_parent_of_pid(npid, &ppid) >= 0 && ppid != mypid) | |
578 | continue; | |
8e274523 | 579 | |
4ad49000 LP |
580 | if (pid != 0) { |
581 | /* Dang, there's more than one daemonized PID | |
582 | in this group, so we don't know what process | |
583 | is the main process. */ | |
584 | pid = 0; | |
585 | break; | |
586 | } | |
8e274523 | 587 | |
4ad49000 | 588 | pid = npid; |
8e274523 LP |
589 | } |
590 | ||
4ad49000 | 591 | return pid; |
8e274523 LP |
592 | } |
593 | ||
8e274523 | 594 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 595 | _cleanup_free_ char *path = NULL; |
8e274523 | 596 | int r; |
9444b1f2 | 597 | char *e, *a; |
8e274523 LP |
598 | |
599 | assert(m); | |
600 | ||
e5a53dc7 | 601 | /* 0. Be nice to Ingo Molnar #628004 */ |
0c85a4f3 | 602 | if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) { |
e5a53dc7 LP |
603 | log_warning("No control group support available, not creating root group."); |
604 | return 0; | |
605 | } | |
606 | ||
35d2e7ec | 607 | /* 1. Determine hierarchy */ |
9444b1f2 LP |
608 | free(m->cgroup_root); |
609 | m->cgroup_root = NULL; | |
610 | ||
611 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); | |
9156e799 | 612 | if (r < 0) { |
12235040 | 613 | log_error("Cannot determine cgroup we are running in: %s", strerror(-r)); |
a32360f1 | 614 | return r; |
12235040 | 615 | } |
8e274523 | 616 | |
9444b1f2 LP |
617 | /* Already in /system.slice? If so, let's cut this off again */ |
618 | if (m->running_as == SYSTEMD_SYSTEM) { | |
619 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); | |
620 | if (e) | |
621 | *e = 0; | |
0baf24dd | 622 | } |
7ccfb64a | 623 | |
9444b1f2 LP |
624 | /* And make sure to store away the root value without trailing |
625 | * slash, even for the root dir, so that we can easily prepend | |
626 | * it everywhere. */ | |
627 | if (streq(m->cgroup_root, "/")) | |
628 | m->cgroup_root[0] = 0; | |
8e274523 | 629 | |
35d2e7ec | 630 | /* 2. Show data */ |
9444b1f2 | 631 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
3474ae3c | 632 | if (r < 0) { |
12235040 | 633 | log_error("Cannot find cgroup mount point: %s", strerror(-r)); |
a32360f1 | 634 | return r; |
12235040 | 635 | } |
8e274523 | 636 | |
c6c18be3 LP |
637 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path); |
638 | ||
35d2e7ec | 639 | /* 3. Install agent */ |
a32360f1 LP |
640 | if (m->running_as == SYSTEMD_SYSTEM) { |
641 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); | |
642 | if (r < 0) | |
643 | log_warning("Failed to install release agent, ignoring: %s", strerror(-r)); | |
644 | else if (r > 0) | |
645 | log_debug("Installed release agent."); | |
646 | else | |
647 | log_debug("Release agent already installed."); | |
648 | } | |
8e274523 | 649 | |
9444b1f2 | 650 | /* 4. Realize the system slice and put us in there */ |
be2c1bd2 LP |
651 | if (m->running_as == SYSTEMD_SYSTEM) { |
652 | a = strappenda(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); | |
653 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, a, 0); | |
654 | } else | |
655 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, 0); | |
9156e799 | 656 | if (r < 0) { |
8e274523 | 657 | log_error("Failed to create root cgroup hierarchy: %s", strerror(-r)); |
a32360f1 | 658 | return r; |
c6c18be3 LP |
659 | } |
660 | ||
35d2e7ec | 661 | /* 5. And pin it, so that it cannot be unmounted */ |
c6c18be3 LP |
662 | if (m->pin_cgroupfs_fd >= 0) |
663 | close_nointr_nofail(m->pin_cgroupfs_fd); | |
664 | ||
9156e799 LP |
665 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); |
666 | if (r < 0) { | |
12235040 | 667 | log_error("Failed to open pin file: %m"); |
a32360f1 | 668 | return -errno; |
c6c18be3 LP |
669 | } |
670 | ||
4ad49000 LP |
671 | /* 6. Figure out which controllers are supported */ |
672 | m->cgroup_supported = cg_mask_supported(); | |
9156e799 | 673 | |
a32360f1 | 674 | return 0; |
8e274523 LP |
675 | } |
676 | ||
c6c18be3 | 677 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
678 | assert(m); |
679 | ||
9444b1f2 LP |
680 | /* We can't really delete the group, since we are in it. But |
681 | * let's trim it. */ | |
682 | if (delete && m->cgroup_root) | |
683 | cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); | |
8e274523 | 684 | |
c6c18be3 LP |
685 | if (m->pin_cgroupfs_fd >= 0) { |
686 | close_nointr_nofail(m->pin_cgroupfs_fd); | |
687 | m->pin_cgroupfs_fd = -1; | |
688 | } | |
689 | ||
9444b1f2 LP |
690 | free(m->cgroup_root); |
691 | m->cgroup_root = NULL; | |
8e274523 LP |
692 | } |
693 | ||
4ad49000 | 694 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 695 | char *p; |
4ad49000 | 696 | Unit *u; |
acb14d31 LP |
697 | |
698 | assert(m); | |
699 | assert(cgroup); | |
acb14d31 | 700 | |
4ad49000 LP |
701 | u = hashmap_get(m->cgroup_unit, cgroup); |
702 | if (u) | |
703 | return u; | |
acb14d31 | 704 | |
8e70580b | 705 | p = strdupa(cgroup); |
acb14d31 LP |
706 | for (;;) { |
707 | char *e; | |
708 | ||
709 | e = strrchr(p, '/'); | |
4ad49000 LP |
710 | if (e == p || !e) |
711 | return NULL; | |
acb14d31 LP |
712 | |
713 | *e = 0; | |
714 | ||
4ad49000 LP |
715 | u = hashmap_get(m->cgroup_unit, p); |
716 | if (u) | |
717 | return u; | |
acb14d31 LP |
718 | } |
719 | } | |
720 | ||
4ad49000 LP |
721 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { |
722 | _cleanup_free_ char *cgroup = NULL; | |
acb14d31 | 723 | int r; |
8e274523 | 724 | |
8c47c732 LP |
725 | assert(m); |
726 | ||
727 | if (pid <= 1) | |
728 | return NULL; | |
729 | ||
4ad49000 LP |
730 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup); |
731 | if (r < 0) | |
6dde1f33 LP |
732 | return NULL; |
733 | ||
4ad49000 | 734 | return manager_get_unit_by_cgroup(m, cgroup); |
6dde1f33 | 735 | } |
4fbf50b3 | 736 | |
4ad49000 LP |
737 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
738 | Unit *u; | |
739 | int r; | |
4fbf50b3 | 740 | |
4ad49000 LP |
741 | assert(m); |
742 | assert(cgroup); | |
4fbf50b3 | 743 | |
4ad49000 | 744 | u = manager_get_unit_by_cgroup(m, cgroup); |
b56c28c3 | 745 | if (u) { |
06025d91 LP |
746 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, true); |
747 | if (r > 0) { | |
748 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
749 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
b56c28c3 | 750 | |
06025d91 LP |
751 | unit_add_to_gc_queue(u); |
752 | } | |
b56c28c3 | 753 | } |
2633eb83 | 754 | |
4ad49000 | 755 | return 0; |
4fbf50b3 LP |
756 | } |
757 | ||
4ad49000 LP |
758 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
759 | [CGROUP_AUTO] = "auto", | |
760 | [CGROUP_CLOSED] = "closed", | |
761 | [CGROUP_STRICT] = "strict", | |
762 | }; | |
4fbf50b3 | 763 | |
4ad49000 | 764 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |