]>
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; | |
ddca82ac | 36 | c->memory_limit = (uint64_t) -1; |
4ad49000 LP |
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 | ||
71fda00f | 44 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
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 | ||
71fda00f | 53 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
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 | |
71fda00f | 62 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
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" | |
112a7f46 | 95 | "%sBlockIOWeight=%lu\n" |
4ad49000 | 96 | "%sMemoryLimit=%" PRIu64 "\n" |
4ad49000 LP |
97 | "%sDevicePolicy=%s\n", |
98 | prefix, yes_no(c->cpu_accounting), | |
99 | prefix, yes_no(c->blockio_accounting), | |
100 | prefix, yes_no(c->memory_accounting), | |
101 | prefix, c->cpu_shares, | |
102 | prefix, c->blockio_weight, | |
103 | prefix, c->memory_limit, | |
4ad49000 LP |
104 | prefix, cgroup_device_policy_to_string(c->device_policy)); |
105 | ||
106 | LIST_FOREACH(device_allow, a, c->device_allow) | |
107 | fprintf(f, | |
108 | "%sDeviceAllow=%s %s%s%s\n", | |
109 | prefix, | |
110 | a->path, | |
111 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
112 | ||
113 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
114 | fprintf(f, | |
8e7076ca | 115 | "%sBlockIODeviceWeight=%s %lu", |
4ad49000 LP |
116 | prefix, |
117 | w->path, | |
118 | w->weight); | |
119 | ||
120 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
121 | char buf[FORMAT_BYTES_MAX]; | |
122 | ||
123 | fprintf(f, | |
124 | "%s%s=%s %s\n", | |
125 | prefix, | |
126 | b->read ? "BlockIOReadBandwidth" : "BlockIOWriteBandwidth", | |
127 | b->path, | |
128 | format_bytes(buf, sizeof(buf), b->bandwidth)); | |
129 | } | |
130 | } | |
131 | ||
132 | static int lookup_blkio_device(const char *p, dev_t *dev) { | |
133 | struct stat st; | |
134 | int r; | |
135 | ||
136 | assert(p); | |
137 | assert(dev); | |
138 | ||
139 | r = stat(p, &st); | |
ab1f0633 | 140 | if (r < 0) { |
4ad49000 LP |
141 | log_warning("Couldn't stat device %s: %m", p); |
142 | return -errno; | |
ab1f0633 | 143 | } |
8e274523 | 144 | |
4ad49000 LP |
145 | if (S_ISBLK(st.st_mode)) |
146 | *dev = st.st_rdev; | |
147 | else if (major(st.st_dev) != 0) { | |
148 | /* If this is not a device node then find the block | |
149 | * device this file is stored on */ | |
150 | *dev = st.st_dev; | |
151 | ||
152 | /* If this is a partition, try to get the originating | |
153 | * block device */ | |
154 | block_get_whole_disk(*dev, dev); | |
155 | } else { | |
156 | log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p); | |
157 | return -ENODEV; | |
158 | } | |
8e274523 | 159 | |
8e274523 | 160 | return 0; |
8e274523 LP |
161 | } |
162 | ||
4ad49000 LP |
163 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
164 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
165 | struct stat st; | |
8c6db833 | 166 | int r; |
8e274523 | 167 | |
4ad49000 LP |
168 | assert(path); |
169 | assert(acc); | |
8e274523 | 170 | |
4ad49000 LP |
171 | if (stat(node, &st) < 0) { |
172 | log_warning("Couldn't stat device %s", node); | |
173 | return -errno; | |
174 | } | |
175 | ||
176 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
177 | log_warning("%s is not a device.", node); | |
178 | return -ENODEV; | |
179 | } | |
180 | ||
181 | sprintf(buf, | |
182 | "%c %u:%u %s", | |
183 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
184 | major(st.st_rdev), minor(st.st_rdev), | |
185 | acc); | |
186 | ||
187 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
188 | if (r < 0) | |
189 | log_warning("Failed to set devices.allow on %s: %s", path, strerror(-r)); | |
190 | ||
191 | return r; | |
8e274523 LP |
192 | } |
193 | ||
4ad49000 LP |
194 | void cgroup_context_apply(CGroupContext *c, CGroupControllerMask mask, const char *path) { |
195 | int r; | |
196 | ||
197 | assert(c); | |
198 | assert(path); | |
8e274523 | 199 | |
4ad49000 LP |
200 | if (mask == 0) |
201 | return; | |
8e274523 | 202 | |
4ad49000 LP |
203 | if (mask & CGROUP_CPU) { |
204 | char buf[DECIMAL_STR_MAX(unsigned long) + 1]; | |
8e274523 | 205 | |
4ad49000 LP |
206 | sprintf(buf, "%lu\n", c->cpu_shares); |
207 | r = cg_set_attribute("cpu", path, "cpu.shares", buf); | |
208 | if (r < 0) | |
209 | log_warning("Failed to set cpu.shares on %s: %s", path, strerror(-r)); | |
210 | } | |
211 | ||
212 | if (mask & CGROUP_BLKIO) { | |
213 | char buf[MAX3(DECIMAL_STR_MAX(unsigned long)+1, | |
214 | DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(unsigned long)*1, | |
215 | DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1)]; | |
216 | CGroupBlockIODeviceWeight *w; | |
217 | CGroupBlockIODeviceBandwidth *b; | |
218 | ||
219 | sprintf(buf, "%lu\n", c->blockio_weight); | |
220 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); | |
221 | if (r < 0) | |
222 | log_warning("Failed to set blkio.weight on %s: %s", path, strerror(-r)); | |
223 | ||
224 | /* FIXME: no way to reset this list */ | |
225 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { | |
226 | dev_t dev; | |
227 | ||
228 | r = lookup_blkio_device(w->path, &dev); | |
229 | if (r < 0) | |
230 | continue; | |
8e274523 | 231 | |
4ad49000 LP |
232 | sprintf(buf, "%u:%u %lu", major(dev), minor(dev), w->weight); |
233 | r = cg_set_attribute("blkio", path, "blkio.weight_device", buf); | |
234 | if (r < 0) | |
235 | log_error("Failed to set blkio.weight_device on %s: %s", path, strerror(-r)); | |
236 | } | |
237 | ||
238 | /* FIXME: no way to reset this list */ | |
239 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
240 | const char *a; | |
241 | dev_t dev; | |
242 | ||
243 | r = lookup_blkio_device(b->path, &dev); | |
244 | if (r < 0) | |
245 | continue; | |
246 | ||
247 | a = b->read ? "blkio.throttle.read_bps_device" : "blkio.throttle.write_bps_device"; | |
248 | ||
249 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), b->bandwidth); | |
250 | r = cg_set_attribute("blkio", path, a, buf); | |
251 | if (r < 0) | |
252 | log_error("Failed to set %s on %s: %s", a, path, strerror(-r)); | |
d686d8a9 | 253 | } |
8e274523 LP |
254 | } |
255 | ||
4ad49000 | 256 | if (mask & CGROUP_MEMORY) { |
6a94f2e9 | 257 | if (c->memory_limit != (uint64_t) -1) { |
e58cec11 LP |
258 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
259 | ||
6a94f2e9 G |
260 | sprintf(buf, "%" PRIu64 "\n", c->memory_limit); |
261 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); | |
262 | } else | |
263 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", "-1"); | |
8e274523 | 264 | |
4ad49000 LP |
265 | if (r < 0) |
266 | log_error("Failed to set memory.limit_in_bytes on %s: %s", path, strerror(-r)); | |
4ad49000 | 267 | } |
8e274523 | 268 | |
4ad49000 LP |
269 | if (mask & CGROUP_DEVICE) { |
270 | CGroupDeviceAllow *a; | |
8e274523 | 271 | |
4ad49000 LP |
272 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
273 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
274 | else | |
275 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
276 | if (r < 0) | |
277 | log_error("Failed to reset devices.list on %s: %s", path, strerror(-r)); | |
fb385181 | 278 | |
4ad49000 LP |
279 | if (c->device_policy == CGROUP_CLOSED || |
280 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
281 | static const char auto_devices[] = | |
282 | "/dev/null\0" "rw\0" | |
283 | "/dev/zero\0" "rw\0" | |
284 | "/dev/full\0" "rw\0" | |
285 | "/dev/random\0" "rw\0" | |
286 | "/dev/urandom\0" "rw\0"; | |
287 | ||
288 | const char *x, *y; | |
289 | ||
290 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
291 | whitelist_device(path, x, y); | |
292 | } | |
293 | ||
294 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
295 | char acc[4]; | |
296 | unsigned k = 0; | |
297 | ||
298 | if (a->r) | |
299 | acc[k++] = 'r'; | |
300 | if (a->w) | |
301 | acc[k++] = 'w'; | |
302 | if (a->m) | |
303 | acc[k++] = 'm'; | |
fb385181 | 304 | |
4ad49000 LP |
305 | if (k == 0) |
306 | continue; | |
fb385181 | 307 | |
4ad49000 LP |
308 | acc[k++] = 0; |
309 | whitelist_device(path, a->path, acc); | |
310 | } | |
311 | } | |
fb385181 LP |
312 | } |
313 | ||
4ad49000 LP |
314 | CGroupControllerMask cgroup_context_get_mask(CGroupContext *c) { |
315 | CGroupControllerMask mask = 0; | |
8e274523 | 316 | |
4ad49000 | 317 | /* Figure out which controllers we need */ |
8e274523 | 318 | |
4ad49000 LP |
319 | if (c->cpu_accounting || c->cpu_shares != 1024) |
320 | mask |= CGROUP_CPUACCT | CGROUP_CPU; | |
ecedd90f | 321 | |
4ad49000 LP |
322 | if (c->blockio_accounting || |
323 | c->blockio_weight != 1000 || | |
324 | c->blockio_device_weights || | |
325 | c->blockio_device_bandwidths) | |
326 | mask |= CGROUP_BLKIO; | |
ecedd90f | 327 | |
4ad49000 | 328 | if (c->memory_accounting || |
ddca82ac | 329 | c->memory_limit != (uint64_t) -1) |
4ad49000 | 330 | mask |= CGROUP_MEMORY; |
8e274523 | 331 | |
4ad49000 LP |
332 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
333 | mask |= CGROUP_DEVICE; | |
334 | ||
335 | return mask; | |
8e274523 LP |
336 | } |
337 | ||
4ad49000 LP |
338 | static CGroupControllerMask unit_get_cgroup_mask(Unit *u) { |
339 | CGroupContext *c; | |
8e274523 | 340 | |
4ad49000 LP |
341 | c = unit_get_cgroup_context(u); |
342 | if (!c) | |
343 | return 0; | |
8e274523 | 344 | |
4ad49000 | 345 | return cgroup_context_get_mask(c); |
8e274523 LP |
346 | } |
347 | ||
4ad49000 LP |
348 | static CGroupControllerMask unit_get_members_mask(Unit *u) { |
349 | CGroupControllerMask mask = 0; | |
350 | Unit *m; | |
351 | Iterator i; | |
246aa6dd | 352 | |
4ad49000 | 353 | assert(u); |
246aa6dd | 354 | |
4ad49000 | 355 | SET_FOREACH(m, u->dependencies[UNIT_BEFORE], i) { |
246aa6dd | 356 | |
4ad49000 | 357 | if (UNIT_DEREF(m->slice) != u) |
246aa6dd LP |
358 | continue; |
359 | ||
4ad49000 | 360 | mask |= unit_get_cgroup_mask(m) | unit_get_members_mask(m); |
246aa6dd LP |
361 | } |
362 | ||
4ad49000 | 363 | return mask; |
246aa6dd LP |
364 | } |
365 | ||
4ad49000 LP |
366 | static CGroupControllerMask unit_get_siblings_mask(Unit *u) { |
367 | assert(u); | |
246aa6dd | 368 | |
4ad49000 LP |
369 | if (!UNIT_ISSET(u->slice)) |
370 | return 0; | |
371 | ||
372 | /* Sibling propagation is only relevant for weight-based | |
373 | * controllers, so let's mask out everything else */ | |
374 | return unit_get_members_mask(UNIT_DEREF(u->slice)) & | |
375 | (CGROUP_CPU|CGROUP_BLKIO|CGROUP_CPUACCT); | |
246aa6dd LP |
376 | } |
377 | ||
4ad49000 | 378 | static int unit_create_cgroups(Unit *u, CGroupControllerMask mask) { |
a94042fa | 379 | _cleanup_free_ char *path; |
4ad49000 | 380 | int r; |
a94042fa | 381 | bool was_in_hash = false; |
64747e2d | 382 | |
4ad49000 | 383 | assert(u); |
64747e2d | 384 | |
4ad49000 LP |
385 | path = unit_default_cgroup_path(u); |
386 | if (!path) | |
a94042fa | 387 | return log_oom(); |
64747e2d | 388 | |
0a1eb06d | 389 | r = hashmap_put(u->manager->cgroup_unit, path, u); |
b58b8e11 | 390 | if (r == 0) |
a94042fa ZJS |
391 | was_in_hash = true; |
392 | else if (r < 0) { | |
393 | log_error(r == -EEXIST ? | |
394 | "cgroup %s exists already: %s" : "hashmap_put failed for %s: %s", | |
395 | path, strerror(-r)); | |
0a1eb06d | 396 | return r; |
b58b8e11 | 397 | } |
0a1eb06d | 398 | |
4ad49000 | 399 | /* First, create our own group */ |
13b84ec7 | 400 | r = cg_create_everywhere(u->manager->cgroup_supported, mask, path); |
4ad49000 LP |
401 | if (r < 0) |
402 | log_error("Failed to create cgroup %s: %s", path, strerror(-r)); | |
64747e2d | 403 | |
4ad49000 | 404 | /* Then, possibly move things over */ |
13b84ec7 LP |
405 | if (u->cgroup_path) { |
406 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, path); | |
64747e2d | 407 | if (r < 0) |
a94042fa ZJS |
408 | log_error("Failed to migrate cgroup from %s to %s: %s", |
409 | u->cgroup_path, path, strerror(-r)); | |
64747e2d LP |
410 | } |
411 | ||
a94042fa ZJS |
412 | if (!was_in_hash) { |
413 | /* Remember the new data */ | |
b58b8e11 HH |
414 | free(u->cgroup_path); |
415 | u->cgroup_path = path; | |
a94042fa | 416 | path = NULL; |
b58b8e11 HH |
417 | } |
418 | ||
4ad49000 LP |
419 | u->cgroup_realized = true; |
420 | u->cgroup_mask = mask; | |
421 | ||
64747e2d LP |
422 | return 0; |
423 | } | |
424 | ||
0a1eb06d | 425 | static int unit_realize_cgroup_now(Unit *u) { |
4ad49000 | 426 | CGroupControllerMask mask; |
64747e2d | 427 | |
4ad49000 | 428 | assert(u); |
64747e2d | 429 | |
4ad49000 | 430 | if (u->in_cgroup_queue) { |
71fda00f | 431 | LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u); |
4ad49000 LP |
432 | u->in_cgroup_queue = false; |
433 | } | |
64747e2d | 434 | |
4ad49000 LP |
435 | mask = unit_get_cgroup_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
436 | mask &= u->manager->cgroup_supported; | |
64747e2d | 437 | |
4ad49000 LP |
438 | if (u->cgroup_realized && |
439 | u->cgroup_mask == mask) | |
0a1eb06d | 440 | return 0; |
64747e2d | 441 | |
4ad49000 | 442 | /* First, realize parents */ |
3d040cf2 HH |
443 | if (UNIT_ISSET(u->slice)) |
444 | unit_realize_cgroup_now(UNIT_DEREF(u->slice)); | |
4ad49000 LP |
445 | |
446 | /* And then do the real work */ | |
0a1eb06d | 447 | return unit_create_cgroups(u, mask); |
64747e2d LP |
448 | } |
449 | ||
4ad49000 | 450 | static void unit_add_to_cgroup_queue(Unit *u) { |
ecedd90f | 451 | |
4ad49000 LP |
452 | if (u->in_cgroup_queue) |
453 | return; | |
8e274523 | 454 | |
71fda00f | 455 | LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u); |
4ad49000 LP |
456 | u->in_cgroup_queue = true; |
457 | } | |
8c6db833 | 458 | |
4ad49000 LP |
459 | unsigned manager_dispatch_cgroup_queue(Manager *m) { |
460 | Unit *i; | |
461 | unsigned n = 0; | |
ecedd90f | 462 | |
4ad49000 LP |
463 | while ((i = m->cgroup_queue)) { |
464 | assert(i->in_cgroup_queue); | |
ecedd90f | 465 | |
0a1eb06d LP |
466 | if (unit_realize_cgroup_now(i) >= 0) |
467 | cgroup_context_apply(unit_get_cgroup_context(i), i->cgroup_mask, i->cgroup_path); | |
468 | ||
4ad49000 LP |
469 | n++; |
470 | } | |
ecedd90f | 471 | |
4ad49000 | 472 | return n; |
8e274523 LP |
473 | } |
474 | ||
4ad49000 LP |
475 | static void unit_queue_siblings(Unit *u) { |
476 | Unit *slice; | |
ca949c9d | 477 | |
4ad49000 LP |
478 | /* This adds the siblings of the specified unit and the |
479 | * siblings of all parent units to the cgroup queue. (But | |
480 | * neither the specified unit itself nor the parents.) */ | |
481 | ||
482 | while ((slice = UNIT_DEREF(u->slice))) { | |
483 | Iterator i; | |
484 | Unit *m; | |
8f53a7b8 | 485 | |
4ad49000 LP |
486 | SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) { |
487 | if (m == u) | |
488 | continue; | |
8e274523 | 489 | |
4ad49000 | 490 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 491 | continue; |
8e274523 | 492 | |
4ad49000 | 493 | unit_add_to_cgroup_queue(m); |
50159e6a LP |
494 | } |
495 | ||
4ad49000 | 496 | u = slice; |
8e274523 | 497 | } |
4ad49000 LP |
498 | } |
499 | ||
0a1eb06d | 500 | int unit_realize_cgroup(Unit *u) { |
4ad49000 | 501 | CGroupContext *c; |
0a1eb06d | 502 | int r; |
4ad49000 LP |
503 | |
504 | assert(u); | |
505 | ||
506 | c = unit_get_cgroup_context(u); | |
507 | if (!c) | |
0a1eb06d | 508 | return 0; |
8e274523 | 509 | |
4ad49000 LP |
510 | /* So, here's the deal: when realizing the cgroups for this |
511 | * unit, we need to first create all parents, but there's more | |
512 | * actually: for the weight-based controllers we also need to | |
513 | * make sure that all our siblings (i.e. units that are in the | |
f3669545 | 514 | * same slice as we are) have cgroups, too. Otherwise things |
4ad49000 LP |
515 | * would become very uneven as each of their processes would |
516 | * get as much resources as all our group together. This call | |
517 | * will synchronously create the parent cgroups, but will | |
518 | * defer work on the siblings to the next event loop | |
519 | * iteration. */ | |
ca949c9d | 520 | |
4ad49000 LP |
521 | /* Add all sibling slices to the cgroup queue. */ |
522 | unit_queue_siblings(u); | |
523 | ||
524 | /* And realize this one now */ | |
0a1eb06d | 525 | r = unit_realize_cgroup_now(u); |
4ad49000 LP |
526 | |
527 | /* And apply the values */ | |
0a1eb06d LP |
528 | if (r >= 0) |
529 | cgroup_context_apply(c, u->cgroup_mask, u->cgroup_path); | |
530 | ||
531 | return r; | |
8e274523 LP |
532 | } |
533 | ||
4ad49000 | 534 | void unit_destroy_cgroup(Unit *u) { |
8e274523 LP |
535 | int r; |
536 | ||
4ad49000 | 537 | assert(u); |
8e274523 | 538 | |
4ad49000 LP |
539 | if (!u->cgroup_path) |
540 | return; | |
8e274523 | 541 | |
13b84ec7 | 542 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !unit_has_name(u, SPECIAL_ROOT_SLICE)); |
4ad49000 | 543 | if (r < 0) |
376dd21d | 544 | log_debug("Failed to destroy cgroup %s: %s", u->cgroup_path, strerror(-r)); |
8e274523 | 545 | |
0a1eb06d LP |
546 | hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); |
547 | ||
4ad49000 LP |
548 | free(u->cgroup_path); |
549 | u->cgroup_path = NULL; | |
550 | u->cgroup_realized = false; | |
551 | u->cgroup_mask = 0; | |
0a1eb06d | 552 | |
8e274523 LP |
553 | } |
554 | ||
4ad49000 LP |
555 | pid_t unit_search_main_pid(Unit *u) { |
556 | _cleanup_fclose_ FILE *f = NULL; | |
557 | pid_t pid = 0, npid, mypid; | |
558 | ||
559 | assert(u); | |
560 | ||
561 | if (!u->cgroup_path) | |
562 | return 0; | |
563 | ||
564 | if (cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f) < 0) | |
565 | return 0; | |
566 | ||
567 | mypid = getpid(); | |
568 | while (cg_read_pid(f, &npid) > 0) { | |
569 | pid_t ppid; | |
570 | ||
571 | if (npid == pid) | |
572 | continue; | |
8e274523 | 573 | |
4ad49000 LP |
574 | /* Ignore processes that aren't our kids */ |
575 | if (get_parent_of_pid(npid, &ppid) >= 0 && ppid != mypid) | |
576 | continue; | |
8e274523 | 577 | |
4ad49000 LP |
578 | if (pid != 0) { |
579 | /* Dang, there's more than one daemonized PID | |
580 | in this group, so we don't know what process | |
581 | is the main process. */ | |
582 | pid = 0; | |
583 | break; | |
584 | } | |
8e274523 | 585 | |
4ad49000 | 586 | pid = npid; |
8e274523 LP |
587 | } |
588 | ||
4ad49000 | 589 | return pid; |
8e274523 LP |
590 | } |
591 | ||
8e274523 | 592 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 593 | _cleanup_free_ char *path = NULL; |
15c60e99 | 594 | char *e; |
8e274523 | 595 | int r; |
8e274523 LP |
596 | |
597 | assert(m); | |
598 | ||
e5a53dc7 | 599 | /* 0. Be nice to Ingo Molnar #628004 */ |
0c85a4f3 | 600 | if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) { |
e5a53dc7 LP |
601 | log_warning("No control group support available, not creating root group."); |
602 | return 0; | |
603 | } | |
604 | ||
35d2e7ec | 605 | /* 1. Determine hierarchy */ |
9444b1f2 LP |
606 | free(m->cgroup_root); |
607 | m->cgroup_root = NULL; | |
608 | ||
609 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); | |
9156e799 | 610 | if (r < 0) { |
12235040 | 611 | log_error("Cannot determine cgroup we are running in: %s", strerror(-r)); |
a32360f1 | 612 | return r; |
12235040 | 613 | } |
8e274523 | 614 | |
15c60e99 LP |
615 | /* LEGACY: Already in /system.slice? If so, let's cut this |
616 | * off. This is to support live upgrades from older systemd | |
617 | * versions where PID 1 was moved there. */ | |
9444b1f2 LP |
618 | if (m->running_as == SYSTEMD_SYSTEM) { |
619 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); | |
15c60e99 LP |
620 | if (!e) |
621 | e = endswith(m->cgroup_root, "/system"); | |
9444b1f2 LP |
622 | if (e) |
623 | *e = 0; | |
0baf24dd | 624 | } |
7ccfb64a | 625 | |
9444b1f2 LP |
626 | /* And make sure to store away the root value without trailing |
627 | * slash, even for the root dir, so that we can easily prepend | |
628 | * it everywhere. */ | |
629 | if (streq(m->cgroup_root, "/")) | |
630 | m->cgroup_root[0] = 0; | |
8e274523 | 631 | |
35d2e7ec | 632 | /* 2. Show data */ |
9444b1f2 | 633 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
3474ae3c | 634 | if (r < 0) { |
12235040 | 635 | log_error("Cannot find cgroup mount point: %s", strerror(-r)); |
a32360f1 | 636 | return r; |
12235040 | 637 | } |
8e274523 | 638 | |
c6c18be3 LP |
639 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path); |
640 | ||
35d2e7ec | 641 | /* 3. Install agent */ |
a32360f1 LP |
642 | if (m->running_as == SYSTEMD_SYSTEM) { |
643 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); | |
644 | if (r < 0) | |
645 | log_warning("Failed to install release agent, ignoring: %s", strerror(-r)); | |
646 | else if (r > 0) | |
647 | log_debug("Installed release agent."); | |
648 | else | |
649 | log_debug("Release agent already installed."); | |
650 | } | |
8e274523 | 651 | |
15c60e99 LP |
652 | /* 4. Make sure we are in the root cgroup */ |
653 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, 0); | |
9156e799 | 654 | if (r < 0) { |
8e274523 | 655 | log_error("Failed to create root cgroup hierarchy: %s", strerror(-r)); |
a32360f1 | 656 | return r; |
c6c18be3 LP |
657 | } |
658 | ||
35d2e7ec | 659 | /* 5. And pin it, so that it cannot be unmounted */ |
c6c18be3 LP |
660 | if (m->pin_cgroupfs_fd >= 0) |
661 | close_nointr_nofail(m->pin_cgroupfs_fd); | |
662 | ||
9156e799 LP |
663 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); |
664 | if (r < 0) { | |
12235040 | 665 | log_error("Failed to open pin file: %m"); |
a32360f1 | 666 | return -errno; |
c6c18be3 LP |
667 | } |
668 | ||
4ad49000 LP |
669 | /* 6. Figure out which controllers are supported */ |
670 | m->cgroup_supported = cg_mask_supported(); | |
9156e799 | 671 | |
e58cec11 LP |
672 | /* 7. Always enable hierarchial support if it exists... */ |
673 | cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
674 | ||
a32360f1 | 675 | return 0; |
8e274523 LP |
676 | } |
677 | ||
c6c18be3 | 678 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
679 | assert(m); |
680 | ||
9444b1f2 LP |
681 | /* We can't really delete the group, since we are in it. But |
682 | * let's trim it. */ | |
683 | if (delete && m->cgroup_root) | |
684 | cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); | |
8e274523 | 685 | |
c6c18be3 LP |
686 | if (m->pin_cgroupfs_fd >= 0) { |
687 | close_nointr_nofail(m->pin_cgroupfs_fd); | |
688 | m->pin_cgroupfs_fd = -1; | |
689 | } | |
690 | ||
9444b1f2 LP |
691 | free(m->cgroup_root); |
692 | m->cgroup_root = NULL; | |
8e274523 LP |
693 | } |
694 | ||
4ad49000 | 695 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 696 | char *p; |
4ad49000 | 697 | Unit *u; |
acb14d31 LP |
698 | |
699 | assert(m); | |
700 | assert(cgroup); | |
acb14d31 | 701 | |
4ad49000 LP |
702 | u = hashmap_get(m->cgroup_unit, cgroup); |
703 | if (u) | |
704 | return u; | |
acb14d31 | 705 | |
8e70580b | 706 | p = strdupa(cgroup); |
acb14d31 LP |
707 | for (;;) { |
708 | char *e; | |
709 | ||
710 | e = strrchr(p, '/'); | |
4ad49000 LP |
711 | if (e == p || !e) |
712 | return NULL; | |
acb14d31 LP |
713 | |
714 | *e = 0; | |
715 | ||
4ad49000 LP |
716 | u = hashmap_get(m->cgroup_unit, p); |
717 | if (u) | |
718 | return u; | |
acb14d31 LP |
719 | } |
720 | } | |
721 | ||
4ad49000 LP |
722 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { |
723 | _cleanup_free_ char *cgroup = NULL; | |
acb14d31 | 724 | int r; |
8e274523 | 725 | |
8c47c732 LP |
726 | assert(m); |
727 | ||
728 | if (pid <= 1) | |
729 | return NULL; | |
730 | ||
4ad49000 LP |
731 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup); |
732 | if (r < 0) | |
6dde1f33 LP |
733 | return NULL; |
734 | ||
4ad49000 | 735 | return manager_get_unit_by_cgroup(m, cgroup); |
6dde1f33 | 736 | } |
4fbf50b3 | 737 | |
4ad49000 LP |
738 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
739 | Unit *u; | |
740 | int r; | |
4fbf50b3 | 741 | |
4ad49000 LP |
742 | assert(m); |
743 | assert(cgroup); | |
4fbf50b3 | 744 | |
4ad49000 | 745 | u = manager_get_unit_by_cgroup(m, cgroup); |
b56c28c3 | 746 | if (u) { |
06025d91 LP |
747 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, true); |
748 | if (r > 0) { | |
749 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
750 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
b56c28c3 | 751 | |
06025d91 LP |
752 | unit_add_to_gc_queue(u); |
753 | } | |
b56c28c3 | 754 | } |
2633eb83 | 755 | |
4ad49000 | 756 | return 0; |
4fbf50b3 LP |
757 | } |
758 | ||
4ad49000 LP |
759 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
760 | [CGROUP_AUTO] = "auto", | |
761 | [CGROUP_CLOSED] = "closed", | |
762 | [CGROUP_STRICT] = "strict", | |
763 | }; | |
4fbf50b3 | 764 | |
4ad49000 | 765 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |