]>
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[] = | |
367 | "/dev/null\0" "rw\0" | |
368 | "/dev/zero\0" "rw\0" | |
369 | "/dev/full\0" "rw\0" | |
370 | "/dev/random\0" "rw\0" | |
371 | "/dev/urandom\0" "rw\0"; | |
372 | ||
373 | const char *x, *y; | |
374 | ||
375 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
376 | whitelist_device(path, x, y); | |
377 | } | |
378 | ||
379 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
380 | char acc[4]; | |
381 | unsigned k = 0; | |
382 | ||
383 | if (a->r) | |
384 | acc[k++] = 'r'; | |
385 | if (a->w) | |
386 | acc[k++] = 'w'; | |
387 | if (a->m) | |
388 | acc[k++] = 'm'; | |
fb385181 | 389 | |
4ad49000 LP |
390 | if (k == 0) |
391 | continue; | |
fb385181 | 392 | |
4ad49000 | 393 | acc[k++] = 0; |
90060676 LP |
394 | |
395 | if (startswith(a->path, "/dev/")) | |
396 | whitelist_device(path, a->path, acc); | |
397 | else if (startswith(a->path, "block-")) | |
398 | whitelist_major(path, a->path + 6, 'b', acc); | |
399 | else if (startswith(a->path, "char-")) | |
400 | whitelist_major(path, a->path + 5, 'c', acc); | |
401 | else | |
402 | log_debug("Ignoring device %s while writing cgroup attribute.", a->path); | |
4ad49000 LP |
403 | } |
404 | } | |
fb385181 LP |
405 | } |
406 | ||
4ad49000 LP |
407 | CGroupControllerMask cgroup_context_get_mask(CGroupContext *c) { |
408 | CGroupControllerMask mask = 0; | |
8e274523 | 409 | |
4ad49000 | 410 | /* Figure out which controllers we need */ |
8e274523 | 411 | |
4ad49000 LP |
412 | if (c->cpu_accounting || c->cpu_shares != 1024) |
413 | mask |= CGROUP_CPUACCT | CGROUP_CPU; | |
ecedd90f | 414 | |
4ad49000 LP |
415 | if (c->blockio_accounting || |
416 | c->blockio_weight != 1000 || | |
417 | c->blockio_device_weights || | |
418 | c->blockio_device_bandwidths) | |
419 | mask |= CGROUP_BLKIO; | |
ecedd90f | 420 | |
4ad49000 | 421 | if (c->memory_accounting || |
ddca82ac | 422 | c->memory_limit != (uint64_t) -1) |
4ad49000 | 423 | mask |= CGROUP_MEMORY; |
8e274523 | 424 | |
4ad49000 LP |
425 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
426 | mask |= CGROUP_DEVICE; | |
427 | ||
428 | return mask; | |
8e274523 LP |
429 | } |
430 | ||
bc432dc7 | 431 | CGroupControllerMask unit_get_cgroup_mask(Unit *u) { |
4ad49000 | 432 | CGroupContext *c; |
8e274523 | 433 | |
4ad49000 LP |
434 | c = unit_get_cgroup_context(u); |
435 | if (!c) | |
436 | return 0; | |
8e274523 | 437 | |
4ad49000 | 438 | return cgroup_context_get_mask(c); |
8e274523 LP |
439 | } |
440 | ||
bc432dc7 | 441 | CGroupControllerMask unit_get_members_mask(Unit *u) { |
4ad49000 | 442 | assert(u); |
bc432dc7 LP |
443 | |
444 | if (u->cgroup_members_mask_valid) | |
445 | return u->cgroup_members_mask; | |
446 | ||
447 | u->cgroup_members_mask = 0; | |
448 | ||
449 | if (u->type == UNIT_SLICE) { | |
450 | Unit *member; | |
451 | Iterator i; | |
452 | ||
453 | SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) { | |
454 | ||
455 | if (member == u) | |
456 | continue; | |
457 | ||
d4fdc205 | 458 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
459 | continue; |
460 | ||
461 | u->cgroup_members_mask |= | |
462 | unit_get_cgroup_mask(member) | | |
463 | unit_get_members_mask(member); | |
464 | } | |
465 | } | |
466 | ||
467 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 468 | return u->cgroup_members_mask; |
246aa6dd LP |
469 | } |
470 | ||
bc432dc7 LP |
471 | CGroupControllerMask unit_get_siblings_mask(Unit *u) { |
472 | CGroupControllerMask m; | |
473 | ||
4ad49000 | 474 | assert(u); |
246aa6dd | 475 | |
bc432dc7 LP |
476 | if (UNIT_ISSET(u->slice)) |
477 | m = unit_get_members_mask(UNIT_DEREF(u->slice)); | |
478 | else | |
479 | m = unit_get_cgroup_mask(u) | unit_get_members_mask(u); | |
4ad49000 LP |
480 | |
481 | /* Sibling propagation is only relevant for weight-based | |
482 | * controllers, so let's mask out everything else */ | |
bc432dc7 | 483 | return m & (CGROUP_CPU|CGROUP_BLKIO|CGROUP_CPUACCT); |
246aa6dd LP |
484 | } |
485 | ||
bc432dc7 | 486 | CGroupControllerMask unit_get_target_mask(Unit *u) { |
6414b7c9 DS |
487 | CGroupControllerMask mask; |
488 | ||
489 | mask = unit_get_cgroup_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); | |
490 | mask &= u->manager->cgroup_supported; | |
491 | ||
492 | return mask; | |
493 | } | |
494 | ||
495 | /* Recurse from a unit up through its containing slices, propagating | |
496 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 LP |
497 | void unit_update_cgroup_members_masks(Unit *u) { |
498 | CGroupControllerMask m; | |
499 | bool more; | |
500 | ||
501 | assert(u); | |
502 | ||
503 | /* Calculate subtree mask */ | |
504 | m = unit_get_cgroup_mask(u) | unit_get_members_mask(u); | |
505 | ||
506 | /* See if anything changed from the previous invocation. If | |
507 | * not, we're done. */ | |
508 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
509 | return; | |
510 | ||
511 | more = | |
512 | u->cgroup_subtree_mask_valid && | |
513 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
514 | ((~m & u->cgroup_subtree_mask) == 0); | |
515 | ||
516 | u->cgroup_subtree_mask = m; | |
517 | u->cgroup_subtree_mask_valid = true; | |
518 | ||
6414b7c9 DS |
519 | if (UNIT_ISSET(u->slice)) { |
520 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
521 | |
522 | if (more) | |
523 | /* There's more set now than before. We | |
524 | * propagate the new mask to the parent's mask | |
525 | * (not caring if it actually was valid or | |
526 | * not). */ | |
527 | ||
528 | s->cgroup_members_mask |= m; | |
529 | ||
530 | else | |
531 | /* There's less set now than before (or we | |
532 | * don't know), we need to recalculate | |
533 | * everything, so let's invalidate the | |
534 | * parent's members mask */ | |
535 | ||
536 | s->cgroup_members_mask_valid = false; | |
537 | ||
538 | /* And now make sure that this change also hits our | |
539 | * grandparents */ | |
540 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
541 | } |
542 | } | |
543 | ||
03b90d4b LP |
544 | static const char *migrate_callback(CGroupControllerMask mask, void *userdata) { |
545 | Unit *u = userdata; | |
546 | ||
547 | assert(mask != 0); | |
548 | assert(u); | |
549 | ||
550 | while (u) { | |
551 | if (u->cgroup_path && | |
552 | u->cgroup_realized && | |
553 | (u->cgroup_realized_mask & mask) == mask) | |
554 | return u->cgroup_path; | |
555 | ||
556 | u = UNIT_DEREF(u->slice); | |
557 | } | |
558 | ||
559 | return NULL; | |
560 | } | |
561 | ||
4ad49000 | 562 | static int unit_create_cgroups(Unit *u, CGroupControllerMask mask) { |
03b90d4b | 563 | _cleanup_free_ char *path = NULL; |
bc432dc7 | 564 | int r; |
64747e2d | 565 | |
4ad49000 | 566 | assert(u); |
64747e2d | 567 | |
4ad49000 LP |
568 | path = unit_default_cgroup_path(u); |
569 | if (!path) | |
a94042fa | 570 | return log_oom(); |
64747e2d | 571 | |
0a1eb06d | 572 | r = hashmap_put(u->manager->cgroup_unit, path, u); |
03b90d4b LP |
573 | if (r < 0) { |
574 | log_error(r == -EEXIST ? "cgroup %s exists already: %s" : "hashmap_put failed for %s: %s", path, strerror(-r)); | |
0a1eb06d | 575 | return r; |
b58b8e11 | 576 | } |
03b90d4b | 577 | if (r > 0) { |
b58b8e11 | 578 | u->cgroup_path = path; |
a94042fa | 579 | path = NULL; |
b58b8e11 HH |
580 | } |
581 | ||
03b90d4b LP |
582 | /* First, create our own group */ |
583 | r = cg_create_everywhere(u->manager->cgroup_supported, mask, u->cgroup_path); | |
584 | if (r < 0) { | |
585 | log_error("Failed to create cgroup %s: %s", u->cgroup_path, strerror(-r)); | |
586 | return r; | |
587 | } | |
588 | ||
589 | /* Keep track that this is now realized */ | |
4ad49000 | 590 | u->cgroup_realized = true; |
bc432dc7 | 591 | u->cgroup_realized_mask = mask; |
4ad49000 | 592 | |
03b90d4b LP |
593 | /* Then, possibly move things over */ |
594 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
595 | if (r < 0) | |
596 | log_warning("Failed to migrate cgroup from to %s: %s", u->cgroup_path, strerror(-r)); | |
597 | ||
64747e2d LP |
598 | return 0; |
599 | } | |
600 | ||
6414b7c9 | 601 | static bool unit_has_mask_realized(Unit *u, CGroupControllerMask mask) { |
bc432dc7 LP |
602 | assert(u); |
603 | ||
604 | return u->cgroup_realized && u->cgroup_realized_mask == mask; | |
6414b7c9 DS |
605 | } |
606 | ||
607 | /* Check if necessary controllers and attributes for a unit are in place. | |
608 | * | |
609 | * If so, do nothing. | |
610 | * If not, create paths, move processes over, and set attributes. | |
611 | * | |
612 | * Returns 0 on success and < 0 on failure. */ | |
0a1eb06d | 613 | static int unit_realize_cgroup_now(Unit *u) { |
4ad49000 | 614 | CGroupControllerMask mask; |
6414b7c9 | 615 | int r; |
64747e2d | 616 | |
4ad49000 | 617 | assert(u); |
64747e2d | 618 | |
4ad49000 | 619 | if (u->in_cgroup_queue) { |
71fda00f | 620 | LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u); |
4ad49000 LP |
621 | u->in_cgroup_queue = false; |
622 | } | |
64747e2d | 623 | |
6414b7c9 | 624 | mask = unit_get_target_mask(u); |
64747e2d | 625 | |
6414b7c9 | 626 | if (unit_has_mask_realized(u, mask)) |
0a1eb06d | 627 | return 0; |
64747e2d | 628 | |
4ad49000 | 629 | /* First, realize parents */ |
6414b7c9 DS |
630 | if (UNIT_ISSET(u->slice)) { |
631 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice)); | |
632 | if (r < 0) | |
633 | return r; | |
634 | } | |
4ad49000 LP |
635 | |
636 | /* And then do the real work */ | |
6414b7c9 DS |
637 | r = unit_create_cgroups(u, mask); |
638 | if (r < 0) | |
639 | return r; | |
640 | ||
641 | /* Finally, apply the necessary attributes. */ | |
642 | cgroup_context_apply(unit_get_cgroup_context(u), mask, u->cgroup_path); | |
643 | ||
644 | return 0; | |
64747e2d LP |
645 | } |
646 | ||
4ad49000 | 647 | static void unit_add_to_cgroup_queue(Unit *u) { |
ecedd90f | 648 | |
4ad49000 LP |
649 | if (u->in_cgroup_queue) |
650 | return; | |
8e274523 | 651 | |
71fda00f | 652 | LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u); |
4ad49000 LP |
653 | u->in_cgroup_queue = true; |
654 | } | |
8c6db833 | 655 | |
4ad49000 LP |
656 | unsigned manager_dispatch_cgroup_queue(Manager *m) { |
657 | Unit *i; | |
658 | unsigned n = 0; | |
6414b7c9 | 659 | int r; |
ecedd90f | 660 | |
4ad49000 LP |
661 | while ((i = m->cgroup_queue)) { |
662 | assert(i->in_cgroup_queue); | |
ecedd90f | 663 | |
6414b7c9 DS |
664 | r = unit_realize_cgroup_now(i); |
665 | if (r < 0) | |
666 | log_warning("Failed to realize cgroups for queued unit %s: %s", i->id, strerror(-r)); | |
0a1eb06d | 667 | |
4ad49000 LP |
668 | n++; |
669 | } | |
ecedd90f | 670 | |
4ad49000 | 671 | return n; |
8e274523 LP |
672 | } |
673 | ||
4ad49000 LP |
674 | static void unit_queue_siblings(Unit *u) { |
675 | Unit *slice; | |
ca949c9d | 676 | |
4ad49000 LP |
677 | /* This adds the siblings of the specified unit and the |
678 | * siblings of all parent units to the cgroup queue. (But | |
679 | * neither the specified unit itself nor the parents.) */ | |
680 | ||
681 | while ((slice = UNIT_DEREF(u->slice))) { | |
682 | Iterator i; | |
683 | Unit *m; | |
8f53a7b8 | 684 | |
4ad49000 LP |
685 | SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) { |
686 | if (m == u) | |
687 | continue; | |
8e274523 | 688 | |
6414b7c9 DS |
689 | /* Skip units that have a dependency on the slice |
690 | * but aren't actually in it. */ | |
4ad49000 | 691 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 692 | continue; |
8e274523 | 693 | |
6414b7c9 DS |
694 | /* No point in doing cgroup application for units |
695 | * without active processes. */ | |
696 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
697 | continue; | |
698 | ||
699 | /* If the unit doesn't need any new controllers | |
700 | * and has current ones realized, it doesn't need | |
701 | * any changes. */ | |
702 | if (unit_has_mask_realized(m, unit_get_target_mask(m))) | |
703 | continue; | |
704 | ||
4ad49000 | 705 | unit_add_to_cgroup_queue(m); |
50159e6a LP |
706 | } |
707 | ||
4ad49000 | 708 | u = slice; |
8e274523 | 709 | } |
4ad49000 LP |
710 | } |
711 | ||
0a1eb06d | 712 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
713 | CGroupContext *c; |
714 | ||
715 | assert(u); | |
716 | ||
717 | c = unit_get_cgroup_context(u); | |
718 | if (!c) | |
0a1eb06d | 719 | return 0; |
8e274523 | 720 | |
4ad49000 LP |
721 | /* So, here's the deal: when realizing the cgroups for this |
722 | * unit, we need to first create all parents, but there's more | |
723 | * actually: for the weight-based controllers we also need to | |
724 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 725 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
726 | * would become very uneven as each of their processes would |
727 | * get as much resources as all our group together. This call | |
728 | * will synchronously create the parent cgroups, but will | |
729 | * defer work on the siblings to the next event loop | |
730 | * iteration. */ | |
ca949c9d | 731 | |
4ad49000 LP |
732 | /* Add all sibling slices to the cgroup queue. */ |
733 | unit_queue_siblings(u); | |
734 | ||
6414b7c9 | 735 | /* And realize this one now (and apply the values) */ |
bc432dc7 | 736 | return unit_realize_cgroup_now(u); |
8e274523 LP |
737 | } |
738 | ||
4ad49000 | 739 | void unit_destroy_cgroup(Unit *u) { |
8e274523 LP |
740 | int r; |
741 | ||
4ad49000 | 742 | assert(u); |
8e274523 | 743 | |
4ad49000 LP |
744 | if (!u->cgroup_path) |
745 | return; | |
8e274523 | 746 | |
13b84ec7 | 747 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !unit_has_name(u, SPECIAL_ROOT_SLICE)); |
4ad49000 | 748 | if (r < 0) |
376dd21d | 749 | log_debug("Failed to destroy cgroup %s: %s", u->cgroup_path, strerror(-r)); |
8e274523 | 750 | |
0a1eb06d LP |
751 | hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); |
752 | ||
4ad49000 LP |
753 | free(u->cgroup_path); |
754 | u->cgroup_path = NULL; | |
755 | u->cgroup_realized = false; | |
bc432dc7 | 756 | u->cgroup_realized_mask = 0; |
0a1eb06d | 757 | |
8e274523 LP |
758 | } |
759 | ||
4ad49000 LP |
760 | pid_t unit_search_main_pid(Unit *u) { |
761 | _cleanup_fclose_ FILE *f = NULL; | |
762 | pid_t pid = 0, npid, mypid; | |
763 | ||
764 | assert(u); | |
765 | ||
766 | if (!u->cgroup_path) | |
767 | return 0; | |
768 | ||
769 | if (cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f) < 0) | |
770 | return 0; | |
771 | ||
772 | mypid = getpid(); | |
773 | while (cg_read_pid(f, &npid) > 0) { | |
774 | pid_t ppid; | |
775 | ||
776 | if (npid == pid) | |
777 | continue; | |
8e274523 | 778 | |
4ad49000 LP |
779 | /* Ignore processes that aren't our kids */ |
780 | if (get_parent_of_pid(npid, &ppid) >= 0 && ppid != mypid) | |
781 | continue; | |
8e274523 | 782 | |
4ad49000 LP |
783 | if (pid != 0) { |
784 | /* Dang, there's more than one daemonized PID | |
785 | in this group, so we don't know what process | |
786 | is the main process. */ | |
787 | pid = 0; | |
788 | break; | |
789 | } | |
8e274523 | 790 | |
4ad49000 | 791 | pid = npid; |
8e274523 LP |
792 | } |
793 | ||
4ad49000 | 794 | return pid; |
8e274523 LP |
795 | } |
796 | ||
8e274523 | 797 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 798 | _cleanup_free_ char *path = NULL; |
15c60e99 | 799 | char *e; |
8e274523 | 800 | int r; |
8e274523 LP |
801 | |
802 | assert(m); | |
803 | ||
e5a53dc7 | 804 | /* 0. Be nice to Ingo Molnar #628004 */ |
0c85a4f3 | 805 | if (path_is_mount_point("/sys/fs/cgroup/systemd", false) <= 0) { |
e5a53dc7 LP |
806 | log_warning("No control group support available, not creating root group."); |
807 | return 0; | |
808 | } | |
809 | ||
35d2e7ec | 810 | /* 1. Determine hierarchy */ |
9444b1f2 LP |
811 | free(m->cgroup_root); |
812 | m->cgroup_root = NULL; | |
813 | ||
814 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); | |
9156e799 | 815 | if (r < 0) { |
12235040 | 816 | log_error("Cannot determine cgroup we are running in: %s", strerror(-r)); |
a32360f1 | 817 | return r; |
12235040 | 818 | } |
8e274523 | 819 | |
15c60e99 LP |
820 | /* LEGACY: Already in /system.slice? If so, let's cut this |
821 | * off. This is to support live upgrades from older systemd | |
822 | * versions where PID 1 was moved there. */ | |
9444b1f2 LP |
823 | if (m->running_as == SYSTEMD_SYSTEM) { |
824 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); | |
15c60e99 LP |
825 | if (!e) |
826 | e = endswith(m->cgroup_root, "/system"); | |
9444b1f2 LP |
827 | if (e) |
828 | *e = 0; | |
0baf24dd | 829 | } |
7ccfb64a | 830 | |
9444b1f2 LP |
831 | /* And make sure to store away the root value without trailing |
832 | * slash, even for the root dir, so that we can easily prepend | |
833 | * it everywhere. */ | |
834 | if (streq(m->cgroup_root, "/")) | |
835 | m->cgroup_root[0] = 0; | |
8e274523 | 836 | |
35d2e7ec | 837 | /* 2. Show data */ |
9444b1f2 | 838 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
3474ae3c | 839 | if (r < 0) { |
12235040 | 840 | log_error("Cannot find cgroup mount point: %s", strerror(-r)); |
a32360f1 | 841 | return r; |
12235040 | 842 | } |
8e274523 | 843 | |
c6c18be3 LP |
844 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path); |
845 | ||
35d2e7ec | 846 | /* 3. Install agent */ |
a32360f1 LP |
847 | if (m->running_as == SYSTEMD_SYSTEM) { |
848 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); | |
849 | if (r < 0) | |
850 | log_warning("Failed to install release agent, ignoring: %s", strerror(-r)); | |
851 | else if (r > 0) | |
852 | log_debug("Installed release agent."); | |
853 | else | |
854 | log_debug("Release agent already installed."); | |
855 | } | |
8e274523 | 856 | |
15c60e99 LP |
857 | /* 4. Make sure we are in the root cgroup */ |
858 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, 0); | |
9156e799 | 859 | if (r < 0) { |
8e274523 | 860 | log_error("Failed to create root cgroup hierarchy: %s", strerror(-r)); |
a32360f1 | 861 | return r; |
c6c18be3 LP |
862 | } |
863 | ||
35d2e7ec | 864 | /* 5. And pin it, so that it cannot be unmounted */ |
c6c18be3 LP |
865 | if (m->pin_cgroupfs_fd >= 0) |
866 | close_nointr_nofail(m->pin_cgroupfs_fd); | |
867 | ||
9156e799 LP |
868 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); |
869 | if (r < 0) { | |
12235040 | 870 | log_error("Failed to open pin file: %m"); |
a32360f1 | 871 | return -errno; |
c6c18be3 LP |
872 | } |
873 | ||
4ad49000 LP |
874 | /* 6. Figure out which controllers are supported */ |
875 | m->cgroup_supported = cg_mask_supported(); | |
9156e799 | 876 | |
e58cec11 LP |
877 | /* 7. Always enable hierarchial support if it exists... */ |
878 | cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
879 | ||
a32360f1 | 880 | return 0; |
8e274523 LP |
881 | } |
882 | ||
c6c18be3 | 883 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
884 | assert(m); |
885 | ||
9444b1f2 LP |
886 | /* We can't really delete the group, since we are in it. But |
887 | * let's trim it. */ | |
888 | if (delete && m->cgroup_root) | |
889 | cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); | |
8e274523 | 890 | |
c6c18be3 LP |
891 | if (m->pin_cgroupfs_fd >= 0) { |
892 | close_nointr_nofail(m->pin_cgroupfs_fd); | |
893 | m->pin_cgroupfs_fd = -1; | |
894 | } | |
895 | ||
9444b1f2 LP |
896 | free(m->cgroup_root); |
897 | m->cgroup_root = NULL; | |
8e274523 LP |
898 | } |
899 | ||
4ad49000 | 900 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 901 | char *p; |
4ad49000 | 902 | Unit *u; |
acb14d31 LP |
903 | |
904 | assert(m); | |
905 | assert(cgroup); | |
acb14d31 | 906 | |
4ad49000 LP |
907 | u = hashmap_get(m->cgroup_unit, cgroup); |
908 | if (u) | |
909 | return u; | |
acb14d31 | 910 | |
8e70580b | 911 | p = strdupa(cgroup); |
acb14d31 LP |
912 | for (;;) { |
913 | char *e; | |
914 | ||
915 | e = strrchr(p, '/'); | |
4ad49000 LP |
916 | if (e == p || !e) |
917 | return NULL; | |
acb14d31 LP |
918 | |
919 | *e = 0; | |
920 | ||
4ad49000 LP |
921 | u = hashmap_get(m->cgroup_unit, p); |
922 | if (u) | |
923 | return u; | |
acb14d31 LP |
924 | } |
925 | } | |
926 | ||
4ad49000 LP |
927 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { |
928 | _cleanup_free_ char *cgroup = NULL; | |
acb14d31 | 929 | int r; |
8e274523 | 930 | |
8c47c732 LP |
931 | assert(m); |
932 | ||
933 | if (pid <= 1) | |
934 | return NULL; | |
935 | ||
4ad49000 LP |
936 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup); |
937 | if (r < 0) | |
6dde1f33 LP |
938 | return NULL; |
939 | ||
4ad49000 | 940 | return manager_get_unit_by_cgroup(m, cgroup); |
6dde1f33 | 941 | } |
4fbf50b3 | 942 | |
4ad49000 LP |
943 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
944 | Unit *u; | |
945 | int r; | |
4fbf50b3 | 946 | |
4ad49000 LP |
947 | assert(m); |
948 | assert(cgroup); | |
4fbf50b3 | 949 | |
4ad49000 | 950 | u = manager_get_unit_by_cgroup(m, cgroup); |
b56c28c3 | 951 | if (u) { |
06025d91 LP |
952 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, true); |
953 | if (r > 0) { | |
954 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
955 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
b56c28c3 | 956 | |
06025d91 LP |
957 | unit_add_to_gc_queue(u); |
958 | } | |
b56c28c3 | 959 | } |
2633eb83 | 960 | |
4ad49000 | 961 | return 0; |
4fbf50b3 LP |
962 | } |
963 | ||
4ad49000 LP |
964 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
965 | [CGROUP_AUTO] = "auto", | |
966 | [CGROUP_CLOSED] = "closed", | |
967 | [CGROUP_STRICT] = "strict", | |
968 | }; | |
4fbf50b3 | 969 | |
4ad49000 | 970 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |