]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/cgroup.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #3526 from fbuihuu/fix-console-log-color
[thirdparty/systemd.git] / src / core / cgroup.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2013 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <fcntl.h>
21 #include <fnmatch.h>
22
23 #include "alloc-util.h"
24 #include "cgroup-util.h"
25 #include "cgroup.h"
26 #include "fd-util.h"
27 #include "fileio.h"
28 #include "fs-util.h"
29 #include "parse-util.h"
30 #include "path-util.h"
31 #include "process-util.h"
32 #include "special.h"
33 #include "string-table.h"
34 #include "string-util.h"
35 #include "stdio-util.h"
36
37 #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
38
39 static void cgroup_compat_warn(void)
40 {
41 static bool cgroup_compat_warned = false;
42
43 if (cgroup_compat_warned)
44 return;
45
46 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details.");
47 cgroup_compat_warned = true;
48 }
49
50 #define log_cgroup_compat(unit, fmt, ...) do { \
51 cgroup_compat_warn(); \
52 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
53 } while (0)
54
55 void cgroup_context_init(CGroupContext *c) {
56 assert(c);
57
58 /* Initialize everything to the kernel defaults, assuming the
59 * structure is preinitialized to 0 */
60
61 c->cpu_shares = CGROUP_CPU_SHARES_INVALID;
62 c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID;
63 c->cpu_quota_per_sec_usec = USEC_INFINITY;
64
65 c->memory_high = CGROUP_LIMIT_MAX;
66 c->memory_max = CGROUP_LIMIT_MAX;
67
68 c->memory_limit = CGROUP_LIMIT_MAX;
69
70 c->io_weight = CGROUP_WEIGHT_INVALID;
71 c->startup_io_weight = CGROUP_WEIGHT_INVALID;
72
73 c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
74 c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
75
76 c->tasks_max = (uint64_t) -1;
77 }
78
79 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
80 assert(c);
81 assert(a);
82
83 LIST_REMOVE(device_allow, c->device_allow, a);
84 free(a->path);
85 free(a);
86 }
87
88 void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) {
89 assert(c);
90 assert(w);
91
92 LIST_REMOVE(device_weights, c->io_device_weights, w);
93 free(w->path);
94 free(w);
95 }
96
97 void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) {
98 assert(c);
99 assert(l);
100
101 LIST_REMOVE(device_limits, c->io_device_limits, l);
102 free(l->path);
103 free(l);
104 }
105
106 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
107 assert(c);
108 assert(w);
109
110 LIST_REMOVE(device_weights, c->blockio_device_weights, w);
111 free(w->path);
112 free(w);
113 }
114
115 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
116 assert(c);
117 assert(b);
118
119 LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
120 free(b->path);
121 free(b);
122 }
123
124 void cgroup_context_done(CGroupContext *c) {
125 assert(c);
126
127 while (c->io_device_weights)
128 cgroup_context_free_io_device_weight(c, c->io_device_weights);
129
130 while (c->io_device_limits)
131 cgroup_context_free_io_device_limit(c, c->io_device_limits);
132
133 while (c->blockio_device_weights)
134 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
135
136 while (c->blockio_device_bandwidths)
137 cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
138
139 while (c->device_allow)
140 cgroup_context_free_device_allow(c, c->device_allow);
141 }
142
143 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
144 CGroupIODeviceLimit *il;
145 CGroupIODeviceWeight *iw;
146 CGroupBlockIODeviceBandwidth *b;
147 CGroupBlockIODeviceWeight *w;
148 CGroupDeviceAllow *a;
149 char u[FORMAT_TIMESPAN_MAX];
150
151 assert(c);
152 assert(f);
153
154 prefix = strempty(prefix);
155
156 fprintf(f,
157 "%sCPUAccounting=%s\n"
158 "%sIOAccounting=%s\n"
159 "%sBlockIOAccounting=%s\n"
160 "%sMemoryAccounting=%s\n"
161 "%sTasksAccounting=%s\n"
162 "%sCPUShares=%" PRIu64 "\n"
163 "%sStartupCPUShares=%" PRIu64 "\n"
164 "%sCPUQuotaPerSecSec=%s\n"
165 "%sIOWeight=%" PRIu64 "\n"
166 "%sStartupIOWeight=%" PRIu64 "\n"
167 "%sBlockIOWeight=%" PRIu64 "\n"
168 "%sStartupBlockIOWeight=%" PRIu64 "\n"
169 "%sMemoryLow=%" PRIu64 "\n"
170 "%sMemoryHigh=%" PRIu64 "\n"
171 "%sMemoryMax=%" PRIu64 "\n"
172 "%sMemoryLimit=%" PRIu64 "\n"
173 "%sTasksMax=%" PRIu64 "\n"
174 "%sDevicePolicy=%s\n"
175 "%sDelegate=%s\n",
176 prefix, yes_no(c->cpu_accounting),
177 prefix, yes_no(c->io_accounting),
178 prefix, yes_no(c->blockio_accounting),
179 prefix, yes_no(c->memory_accounting),
180 prefix, yes_no(c->tasks_accounting),
181 prefix, c->cpu_shares,
182 prefix, c->startup_cpu_shares,
183 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
184 prefix, c->io_weight,
185 prefix, c->startup_io_weight,
186 prefix, c->blockio_weight,
187 prefix, c->startup_blockio_weight,
188 prefix, c->memory_low,
189 prefix, c->memory_high,
190 prefix, c->memory_max,
191 prefix, c->memory_limit,
192 prefix, c->tasks_max,
193 prefix, cgroup_device_policy_to_string(c->device_policy),
194 prefix, yes_no(c->delegate));
195
196 LIST_FOREACH(device_allow, a, c->device_allow)
197 fprintf(f,
198 "%sDeviceAllow=%s %s%s%s\n",
199 prefix,
200 a->path,
201 a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
202
203 LIST_FOREACH(device_weights, iw, c->io_device_weights)
204 fprintf(f,
205 "%sIODeviceWeight=%s %" PRIu64,
206 prefix,
207 iw->path,
208 iw->weight);
209
210 LIST_FOREACH(device_limits, il, c->io_device_limits) {
211 char buf[FORMAT_BYTES_MAX];
212 CGroupIOLimitType type;
213
214 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
215 if (il->limits[type] != cgroup_io_limit_defaults[type])
216 fprintf(f,
217 "%s%s=%s %s\n",
218 prefix,
219 cgroup_io_limit_type_to_string(type),
220 il->path,
221 format_bytes(buf, sizeof(buf), il->limits[type]));
222 }
223
224 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
225 fprintf(f,
226 "%sBlockIODeviceWeight=%s %" PRIu64,
227 prefix,
228 w->path,
229 w->weight);
230
231 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
232 char buf[FORMAT_BYTES_MAX];
233
234 if (b->rbps != CGROUP_LIMIT_MAX)
235 fprintf(f,
236 "%sBlockIOReadBandwidth=%s %s\n",
237 prefix,
238 b->path,
239 format_bytes(buf, sizeof(buf), b->rbps));
240 if (b->wbps != CGROUP_LIMIT_MAX)
241 fprintf(f,
242 "%sBlockIOWriteBandwidth=%s %s\n",
243 prefix,
244 b->path,
245 format_bytes(buf, sizeof(buf), b->wbps));
246 }
247 }
248
249 static int lookup_block_device(const char *p, dev_t *dev) {
250 struct stat st;
251 int r;
252
253 assert(p);
254 assert(dev);
255
256 r = stat(p, &st);
257 if (r < 0)
258 return log_warning_errno(errno, "Couldn't stat device %s: %m", p);
259
260 if (S_ISBLK(st.st_mode))
261 *dev = st.st_rdev;
262 else if (major(st.st_dev) != 0) {
263 /* If this is not a device node then find the block
264 * device this file is stored on */
265 *dev = st.st_dev;
266
267 /* If this is a partition, try to get the originating
268 * block device */
269 block_get_whole_disk(*dev, dev);
270 } else {
271 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p);
272 return -ENODEV;
273 }
274
275 return 0;
276 }
277
278 static int whitelist_device(const char *path, const char *node, const char *acc) {
279 char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
280 struct stat st;
281 int r;
282
283 assert(path);
284 assert(acc);
285
286 if (stat(node, &st) < 0) {
287 log_warning("Couldn't stat device %s", node);
288 return -errno;
289 }
290
291 if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
292 log_warning("%s is not a device.", node);
293 return -ENODEV;
294 }
295
296 sprintf(buf,
297 "%c %u:%u %s",
298 S_ISCHR(st.st_mode) ? 'c' : 'b',
299 major(st.st_rdev), minor(st.st_rdev),
300 acc);
301
302 r = cg_set_attribute("devices", path, "devices.allow", buf);
303 if (r < 0)
304 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
305 "Failed to set devices.allow on %s: %m", path);
306
307 return r;
308 }
309
310 static int whitelist_major(const char *path, const char *name, char type, const char *acc) {
311 _cleanup_fclose_ FILE *f = NULL;
312 char line[LINE_MAX];
313 bool good = false;
314 int r;
315
316 assert(path);
317 assert(acc);
318 assert(type == 'b' || type == 'c');
319
320 f = fopen("/proc/devices", "re");
321 if (!f)
322 return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
323
324 FOREACH_LINE(line, f, goto fail) {
325 char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w;
326 unsigned maj;
327
328 truncate_nl(line);
329
330 if (type == 'c' && streq(line, "Character devices:")) {
331 good = true;
332 continue;
333 }
334
335 if (type == 'b' && streq(line, "Block devices:")) {
336 good = true;
337 continue;
338 }
339
340 if (isempty(line)) {
341 good = false;
342 continue;
343 }
344
345 if (!good)
346 continue;
347
348 p = strstrip(line);
349
350 w = strpbrk(p, WHITESPACE);
351 if (!w)
352 continue;
353 *w = 0;
354
355 r = safe_atou(p, &maj);
356 if (r < 0)
357 continue;
358 if (maj <= 0)
359 continue;
360
361 w++;
362 w += strspn(w, WHITESPACE);
363
364 if (fnmatch(name, w, 0) != 0)
365 continue;
366
367 sprintf(buf,
368 "%c %u:* %s",
369 type,
370 maj,
371 acc);
372
373 r = cg_set_attribute("devices", path, "devices.allow", buf);
374 if (r < 0)
375 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
376 "Failed to set devices.allow on %s: %m", path);
377 }
378
379 return 0;
380
381 fail:
382 log_warning_errno(errno, "Failed to read /proc/devices: %m");
383 return -errno;
384 }
385
386 static bool cgroup_context_has_io_config(CGroupContext *c) {
387 return c->io_accounting ||
388 c->io_weight != CGROUP_WEIGHT_INVALID ||
389 c->startup_io_weight != CGROUP_WEIGHT_INVALID ||
390 c->io_device_weights ||
391 c->io_device_limits;
392 }
393
394 static bool cgroup_context_has_blockio_config(CGroupContext *c) {
395 return c->blockio_accounting ||
396 c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
397 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
398 c->blockio_device_weights ||
399 c->blockio_device_bandwidths;
400 }
401
402 static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) {
403 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
404 c->startup_io_weight != CGROUP_WEIGHT_INVALID)
405 return c->startup_io_weight;
406 else if (c->io_weight != CGROUP_WEIGHT_INVALID)
407 return c->io_weight;
408 else
409 return CGROUP_WEIGHT_DEFAULT;
410 }
411
412 static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) {
413 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
414 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
415 return c->startup_blockio_weight;
416 else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
417 return c->blockio_weight;
418 else
419 return CGROUP_BLKIO_WEIGHT_DEFAULT;
420 }
421
422 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) {
423 return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT,
424 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
425 }
426
427 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) {
428 return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT,
429 CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX);
430 }
431
432 static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) {
433 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
434 dev_t dev;
435 int r;
436
437 r = lookup_block_device(dev_path, &dev);
438 if (r < 0)
439 return;
440
441 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight);
442 r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf);
443 if (r < 0)
444 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
445 "Failed to set io.weight: %m");
446 }
447
448 static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) {
449 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
450 dev_t dev;
451 int r;
452
453 r = lookup_block_device(dev_path, &dev);
454 if (r < 0)
455 return;
456
457 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight);
458 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf);
459 if (r < 0)
460 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
461 "Failed to set blkio.weight_device: %m");
462 }
463
464 static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) {
465 char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)];
466 char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
467 CGroupIOLimitType type;
468 dev_t dev;
469 unsigned n = 0;
470 int r;
471
472 r = lookup_block_device(dev_path, &dev);
473 if (r < 0)
474 return 0;
475
476 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) {
477 if (limits[type] != cgroup_io_limit_defaults[type]) {
478 xsprintf(limit_bufs[type], "%" PRIu64, limits[type]);
479 n++;
480 } else {
481 xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0");
482 }
483 }
484
485 xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev),
486 limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX],
487 limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]);
488 r = cg_set_attribute("io", u->cgroup_path, "io.max", buf);
489 if (r < 0)
490 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
491 "Failed to set io.max: %m");
492 return n;
493 }
494
495 static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) {
496 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
497 dev_t dev;
498 unsigned n = 0;
499 int r;
500
501 r = lookup_block_device(dev_path, &dev);
502 if (r < 0)
503 return 0;
504
505 if (rbps != CGROUP_LIMIT_MAX)
506 n++;
507 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps);
508 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf);
509 if (r < 0)
510 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
511 "Failed to set blkio.throttle.read_bps_device: %m");
512
513 if (wbps != CGROUP_LIMIT_MAX)
514 n++;
515 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps);
516 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf);
517 if (r < 0)
518 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
519 "Failed to set blkio.throttle.write_bps_device: %m");
520
521 return n;
522 }
523
524 static bool cgroup_context_has_unified_memory_config(CGroupContext *c) {
525 return c->memory_low > 0 || c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX;
526 }
527
528 static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) {
529 char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
530 int r;
531
532 if (v != CGROUP_LIMIT_MAX)
533 xsprintf(buf, "%" PRIu64 "\n", v);
534
535 r = cg_set_attribute("memory", u->cgroup_path, file, buf);
536 if (r < 0)
537 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
538 "Failed to set %s: %m", file);
539 }
540
541 static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) {
542 const char *path;
543 CGroupContext *c;
544 bool is_root;
545 int r;
546
547 assert(u);
548
549 c = unit_get_cgroup_context(u);
550 path = u->cgroup_path;
551
552 assert(c);
553 assert(path);
554
555 if (mask == 0)
556 return;
557
558 /* Some cgroup attributes are not supported on the root cgroup,
559 * hence silently ignore */
560 is_root = isempty(path) || path_equal(path, "/");
561 if (is_root)
562 /* Make sure we don't try to display messages with an empty path. */
563 path = "/";
564
565 /* We generally ignore errors caused by read-only mounted
566 * cgroup trees (assuming we are running in a container then),
567 * and missing cgroups, i.e. EROFS and ENOENT. */
568
569 if ((mask & CGROUP_MASK_CPU) && !is_root) {
570 char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1];
571
572 sprintf(buf, "%" PRIu64 "\n",
573 IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID ? c->startup_cpu_shares :
574 c->cpu_shares != CGROUP_CPU_SHARES_INVALID ? c->cpu_shares : CGROUP_CPU_SHARES_DEFAULT);
575 r = cg_set_attribute("cpu", path, "cpu.shares", buf);
576 if (r < 0)
577 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
578 "Failed to set cpu.shares: %m");
579
580 sprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
581 r = cg_set_attribute("cpu", path, "cpu.cfs_period_us", buf);
582 if (r < 0)
583 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
584 "Failed to set cpu.cfs_period_us: %m");
585
586 if (c->cpu_quota_per_sec_usec != USEC_INFINITY) {
587 sprintf(buf, USEC_FMT "\n", c->cpu_quota_per_sec_usec * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC);
588 r = cg_set_attribute("cpu", path, "cpu.cfs_quota_us", buf);
589 } else
590 r = cg_set_attribute("cpu", path, "cpu.cfs_quota_us", "-1");
591 if (r < 0)
592 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
593 "Failed to set cpu.cfs_quota_us: %m");
594 }
595
596 if (mask & CGROUP_MASK_IO) {
597 bool has_io = cgroup_context_has_io_config(c);
598 bool has_blockio = cgroup_context_has_blockio_config(c);
599
600 if (!is_root) {
601 char buf[8+DECIMAL_STR_MAX(uint64_t)+1];
602 uint64_t weight;
603
604 if (has_io)
605 weight = cgroup_context_io_weight(c, state);
606 else if (has_blockio) {
607 uint64_t blkio_weight = cgroup_context_blkio_weight(c, state);
608
609 weight = cgroup_weight_blkio_to_io(blkio_weight);
610
611 log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64,
612 blkio_weight, weight);
613 } else
614 weight = CGROUP_WEIGHT_DEFAULT;
615
616 xsprintf(buf, "default %" PRIu64 "\n", weight);
617 r = cg_set_attribute("io", path, "io.weight", buf);
618 if (r < 0)
619 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
620 "Failed to set io.weight: %m");
621
622 if (has_io) {
623 CGroupIODeviceWeight *w;
624
625 /* FIXME: no way to reset this list */
626 LIST_FOREACH(device_weights, w, c->io_device_weights)
627 cgroup_apply_io_device_weight(u, w->path, w->weight);
628 } else if (has_blockio) {
629 CGroupBlockIODeviceWeight *w;
630
631 /* FIXME: no way to reset this list */
632 LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
633 weight = cgroup_weight_blkio_to_io(w->weight);
634
635 log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s",
636 w->weight, weight, w->path);
637
638 cgroup_apply_io_device_weight(u, w->path, weight);
639 }
640 }
641 }
642
643 /* Apply limits and free ones without config. */
644 if (has_io) {
645 CGroupIODeviceLimit *l, *next;
646
647 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
648 if (!cgroup_apply_io_device_limit(u, l->path, l->limits))
649 cgroup_context_free_io_device_limit(c, l);
650 }
651 } else if (has_blockio) {
652 CGroupBlockIODeviceBandwidth *b, *next;
653
654 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
655 uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX];
656 CGroupIOLimitType type;
657
658 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
659 limits[type] = cgroup_io_limit_defaults[type];
660
661 limits[CGROUP_IO_RBPS_MAX] = b->rbps;
662 limits[CGROUP_IO_WBPS_MAX] = b->wbps;
663
664 log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s",
665 b->rbps, b->wbps, b->path);
666
667 if (!cgroup_apply_io_device_limit(u, b->path, limits))
668 cgroup_context_free_blockio_device_bandwidth(c, b);
669 }
670 }
671 }
672
673 if (mask & CGROUP_MASK_BLKIO) {
674 bool has_io = cgroup_context_has_io_config(c);
675 bool has_blockio = cgroup_context_has_blockio_config(c);
676
677 if (!is_root) {
678 char buf[DECIMAL_STR_MAX(uint64_t)+1];
679 uint64_t weight;
680
681 if (has_blockio)
682 weight = cgroup_context_blkio_weight(c, state);
683 else if (has_io) {
684 uint64_t io_weight = cgroup_context_io_weight(c, state);
685
686 weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state));
687
688 log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64,
689 io_weight, weight);
690 } else
691 weight = CGROUP_BLKIO_WEIGHT_DEFAULT;
692
693 xsprintf(buf, "%" PRIu64 "\n", weight);
694 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
695 if (r < 0)
696 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
697 "Failed to set blkio.weight: %m");
698
699 if (has_blockio) {
700 CGroupBlockIODeviceWeight *w;
701
702 /* FIXME: no way to reset this list */
703 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
704 cgroup_apply_blkio_device_weight(u, w->path, w->weight);
705 } else if (has_io) {
706 CGroupIODeviceWeight *w;
707
708 /* FIXME: no way to reset this list */
709 LIST_FOREACH(device_weights, w, c->io_device_weights) {
710 weight = cgroup_weight_io_to_blkio(w->weight);
711
712 log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s",
713 w->weight, weight, w->path);
714
715 cgroup_apply_blkio_device_weight(u, w->path, weight);
716 }
717 }
718 }
719
720 /* Apply limits and free ones without config. */
721 if (has_blockio) {
722 CGroupBlockIODeviceBandwidth *b, *next;
723
724 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
725 if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps))
726 cgroup_context_free_blockio_device_bandwidth(c, b);
727 }
728 } else if (has_io) {
729 CGroupIODeviceLimit *l, *next;
730
731 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
732 log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s",
733 l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path);
734
735 if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]))
736 cgroup_context_free_io_device_limit(c, l);
737 }
738 }
739 }
740
741 if ((mask & CGROUP_MASK_MEMORY) && !is_root) {
742 if (cg_unified() > 0) {
743 uint64_t max = c->memory_max;
744
745 if (cgroup_context_has_unified_memory_config(c))
746 max = c->memory_max;
747 else {
748 max = c->memory_limit;
749
750 if (max != CGROUP_LIMIT_MAX)
751 log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max);
752 }
753
754 cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low);
755 cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high);
756 cgroup_apply_unified_memory_limit(u, "memory.max", max);
757 } else {
758 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
759
760 if (c->memory_limit != CGROUP_LIMIT_MAX)
761 xsprintf(buf, "%" PRIu64 "\n", c->memory_limit);
762 else {
763 xsprintf(buf, "%" PRIu64 "\n", c->memory_max);
764
765 if (c->memory_max != CGROUP_LIMIT_MAX)
766 log_cgroup_compat(u, "Applying MemoryMax %" PRIu64 " as MemoryLimit", c->memory_max);
767 }
768
769 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
770 if (r < 0)
771 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
772 "Failed to set memory.limit_in_bytes: %m");
773 }
774 }
775
776 if ((mask & CGROUP_MASK_DEVICES) && !is_root) {
777 CGroupDeviceAllow *a;
778
779 /* Changing the devices list of a populated cgroup
780 * might result in EINVAL, hence ignore EINVAL
781 * here. */
782
783 if (c->device_allow || c->device_policy != CGROUP_AUTO)
784 r = cg_set_attribute("devices", path, "devices.deny", "a");
785 else
786 r = cg_set_attribute("devices", path, "devices.allow", "a");
787 if (r < 0)
788 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
789 "Failed to reset devices.list: %m");
790
791 if (c->device_policy == CGROUP_CLOSED ||
792 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
793 static const char auto_devices[] =
794 "/dev/null\0" "rwm\0"
795 "/dev/zero\0" "rwm\0"
796 "/dev/full\0" "rwm\0"
797 "/dev/random\0" "rwm\0"
798 "/dev/urandom\0" "rwm\0"
799 "/dev/tty\0" "rwm\0"
800 "/dev/pts/ptmx\0" "rw\0"; /* /dev/pts/ptmx may not be duplicated, but accessed */
801
802 const char *x, *y;
803
804 NULSTR_FOREACH_PAIR(x, y, auto_devices)
805 whitelist_device(path, x, y);
806
807 whitelist_major(path, "pts", 'c', "rw");
808 whitelist_major(path, "kdbus", 'c', "rw");
809 whitelist_major(path, "kdbus/*", 'c', "rw");
810 }
811
812 LIST_FOREACH(device_allow, a, c->device_allow) {
813 char acc[4];
814 unsigned k = 0;
815
816 if (a->r)
817 acc[k++] = 'r';
818 if (a->w)
819 acc[k++] = 'w';
820 if (a->m)
821 acc[k++] = 'm';
822
823 if (k == 0)
824 continue;
825
826 acc[k++] = 0;
827
828 if (startswith(a->path, "/dev/"))
829 whitelist_device(path, a->path, acc);
830 else if (startswith(a->path, "block-"))
831 whitelist_major(path, a->path + 6, 'b', acc);
832 else if (startswith(a->path, "char-"))
833 whitelist_major(path, a->path + 5, 'c', acc);
834 else
835 log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path);
836 }
837 }
838
839 if ((mask & CGROUP_MASK_PIDS) && !is_root) {
840
841 if (c->tasks_max != (uint64_t) -1) {
842 char buf[DECIMAL_STR_MAX(uint64_t) + 2];
843
844 sprintf(buf, "%" PRIu64 "\n", c->tasks_max);
845 r = cg_set_attribute("pids", path, "pids.max", buf);
846 } else
847 r = cg_set_attribute("pids", path, "pids.max", "max");
848
849 if (r < 0)
850 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
851 "Failed to set pids.max: %m");
852 }
853 }
854
855 CGroupMask cgroup_context_get_mask(CGroupContext *c) {
856 CGroupMask mask = 0;
857
858 /* Figure out which controllers we need */
859
860 if (c->cpu_accounting ||
861 c->cpu_shares != CGROUP_CPU_SHARES_INVALID ||
862 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID ||
863 c->cpu_quota_per_sec_usec != USEC_INFINITY)
864 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
865
866 if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c))
867 mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
868
869 if (c->memory_accounting ||
870 c->memory_limit != CGROUP_LIMIT_MAX ||
871 cgroup_context_has_unified_memory_config(c))
872 mask |= CGROUP_MASK_MEMORY;
873
874 if (c->device_allow ||
875 c->device_policy != CGROUP_AUTO)
876 mask |= CGROUP_MASK_DEVICES;
877
878 if (c->tasks_accounting ||
879 c->tasks_max != (uint64_t) -1)
880 mask |= CGROUP_MASK_PIDS;
881
882 return mask;
883 }
884
885 CGroupMask unit_get_own_mask(Unit *u) {
886 CGroupContext *c;
887
888 /* Returns the mask of controllers the unit needs for itself */
889
890 c = unit_get_cgroup_context(u);
891 if (!c)
892 return 0;
893
894 /* If delegation is turned on, then turn on all cgroups,
895 * unless we are on the legacy hierarchy and the process we
896 * fork into it is known to drop privileges, and hence
897 * shouldn't get access to the controllers.
898 *
899 * Note that on the unified hierarchy it is safe to delegate
900 * controllers to unprivileged services. */
901
902 if (c->delegate) {
903 ExecContext *e;
904
905 e = unit_get_exec_context(u);
906 if (!e ||
907 exec_context_maintains_privileges(e) ||
908 cg_unified() > 0)
909 return _CGROUP_MASK_ALL;
910 }
911
912 return cgroup_context_get_mask(c);
913 }
914
915 CGroupMask unit_get_members_mask(Unit *u) {
916 assert(u);
917
918 /* Returns the mask of controllers all of the unit's children
919 * require, merged */
920
921 if (u->cgroup_members_mask_valid)
922 return u->cgroup_members_mask;
923
924 u->cgroup_members_mask = 0;
925
926 if (u->type == UNIT_SLICE) {
927 Unit *member;
928 Iterator i;
929
930 SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) {
931
932 if (member == u)
933 continue;
934
935 if (UNIT_DEREF(member->slice) != u)
936 continue;
937
938 u->cgroup_members_mask |=
939 unit_get_own_mask(member) |
940 unit_get_members_mask(member);
941 }
942 }
943
944 u->cgroup_members_mask_valid = true;
945 return u->cgroup_members_mask;
946 }
947
948 CGroupMask unit_get_siblings_mask(Unit *u) {
949 assert(u);
950
951 /* Returns the mask of controllers all of the unit's siblings
952 * require, i.e. the members mask of the unit's parent slice
953 * if there is one. */
954
955 if (UNIT_ISSET(u->slice))
956 return unit_get_members_mask(UNIT_DEREF(u->slice));
957
958 return unit_get_own_mask(u) | unit_get_members_mask(u);
959 }
960
961 CGroupMask unit_get_subtree_mask(Unit *u) {
962
963 /* Returns the mask of this subtree, meaning of the group
964 * itself and its children. */
965
966 return unit_get_own_mask(u) | unit_get_members_mask(u);
967 }
968
969 CGroupMask unit_get_target_mask(Unit *u) {
970 CGroupMask mask;
971
972 /* This returns the cgroup mask of all controllers to enable
973 * for a specific cgroup, i.e. everything it needs itself,
974 * plus all that its children need, plus all that its siblings
975 * need. This is primarily useful on the legacy cgroup
976 * hierarchy, where we need to duplicate each cgroup in each
977 * hierarchy that shall be enabled for it. */
978
979 mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
980 mask &= u->manager->cgroup_supported;
981
982 return mask;
983 }
984
985 CGroupMask unit_get_enable_mask(Unit *u) {
986 CGroupMask mask;
987
988 /* This returns the cgroup mask of all controllers to enable
989 * for the children of a specific cgroup. This is primarily
990 * useful for the unified cgroup hierarchy, where each cgroup
991 * controls which controllers are enabled for its children. */
992
993 mask = unit_get_members_mask(u);
994 mask &= u->manager->cgroup_supported;
995
996 return mask;
997 }
998
999 /* Recurse from a unit up through its containing slices, propagating
1000 * mask bits upward. A unit is also member of itself. */
1001 void unit_update_cgroup_members_masks(Unit *u) {
1002 CGroupMask m;
1003 bool more;
1004
1005 assert(u);
1006
1007 /* Calculate subtree mask */
1008 m = unit_get_subtree_mask(u);
1009
1010 /* See if anything changed from the previous invocation. If
1011 * not, we're done. */
1012 if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask)
1013 return;
1014
1015 more =
1016 u->cgroup_subtree_mask_valid &&
1017 ((m & ~u->cgroup_subtree_mask) != 0) &&
1018 ((~m & u->cgroup_subtree_mask) == 0);
1019
1020 u->cgroup_subtree_mask = m;
1021 u->cgroup_subtree_mask_valid = true;
1022
1023 if (UNIT_ISSET(u->slice)) {
1024 Unit *s = UNIT_DEREF(u->slice);
1025
1026 if (more)
1027 /* There's more set now than before. We
1028 * propagate the new mask to the parent's mask
1029 * (not caring if it actually was valid or
1030 * not). */
1031
1032 s->cgroup_members_mask |= m;
1033
1034 else
1035 /* There's less set now than before (or we
1036 * don't know), we need to recalculate
1037 * everything, so let's invalidate the
1038 * parent's members mask */
1039
1040 s->cgroup_members_mask_valid = false;
1041
1042 /* And now make sure that this change also hits our
1043 * grandparents */
1044 unit_update_cgroup_members_masks(s);
1045 }
1046 }
1047
1048 static const char *migrate_callback(CGroupMask mask, void *userdata) {
1049 Unit *u = userdata;
1050
1051 assert(mask != 0);
1052 assert(u);
1053
1054 while (u) {
1055 if (u->cgroup_path &&
1056 u->cgroup_realized &&
1057 (u->cgroup_realized_mask & mask) == mask)
1058 return u->cgroup_path;
1059
1060 u = UNIT_DEREF(u->slice);
1061 }
1062
1063 return NULL;
1064 }
1065
1066 char *unit_default_cgroup_path(Unit *u) {
1067 _cleanup_free_ char *escaped = NULL, *slice = NULL;
1068 int r;
1069
1070 assert(u);
1071
1072 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1073 return strdup(u->manager->cgroup_root);
1074
1075 if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) {
1076 r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice);
1077 if (r < 0)
1078 return NULL;
1079 }
1080
1081 escaped = cg_escape(u->id);
1082 if (!escaped)
1083 return NULL;
1084
1085 if (slice)
1086 return strjoin(u->manager->cgroup_root, "/", slice, "/", escaped, NULL);
1087 else
1088 return strjoin(u->manager->cgroup_root, "/", escaped, NULL);
1089 }
1090
1091 int unit_set_cgroup_path(Unit *u, const char *path) {
1092 _cleanup_free_ char *p = NULL;
1093 int r;
1094
1095 assert(u);
1096
1097 if (path) {
1098 p = strdup(path);
1099 if (!p)
1100 return -ENOMEM;
1101 } else
1102 p = NULL;
1103
1104 if (streq_ptr(u->cgroup_path, p))
1105 return 0;
1106
1107 if (p) {
1108 r = hashmap_put(u->manager->cgroup_unit, p, u);
1109 if (r < 0)
1110 return r;
1111 }
1112
1113 unit_release_cgroup(u);
1114
1115 u->cgroup_path = p;
1116 p = NULL;
1117
1118 return 1;
1119 }
1120
1121 int unit_watch_cgroup(Unit *u) {
1122 _cleanup_free_ char *events = NULL;
1123 int r;
1124
1125 assert(u);
1126
1127 if (!u->cgroup_path)
1128 return 0;
1129
1130 if (u->cgroup_inotify_wd >= 0)
1131 return 0;
1132
1133 /* Only applies to the unified hierarchy */
1134 r = cg_unified();
1135 if (r < 0)
1136 return log_unit_error_errno(u, r, "Failed detect wether the unified hierarchy is used: %m");
1137 if (r == 0)
1138 return 0;
1139
1140 /* Don't watch the root slice, it's pointless. */
1141 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1142 return 0;
1143
1144 r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops);
1145 if (r < 0)
1146 return log_oom();
1147
1148 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events);
1149 if (r < 0)
1150 return log_oom();
1151
1152 u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY);
1153 if (u->cgroup_inotify_wd < 0) {
1154
1155 if (errno == ENOENT) /* If the directory is already
1156 * gone we don't need to track
1157 * it, so this is not an error */
1158 return 0;
1159
1160 return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path);
1161 }
1162
1163 r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u);
1164 if (r < 0)
1165 return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m");
1166
1167 return 0;
1168 }
1169
1170 static int unit_create_cgroup(
1171 Unit *u,
1172 CGroupMask target_mask,
1173 CGroupMask enable_mask) {
1174
1175 CGroupContext *c;
1176 int r;
1177
1178 assert(u);
1179
1180 c = unit_get_cgroup_context(u);
1181 if (!c)
1182 return 0;
1183
1184 if (!u->cgroup_path) {
1185 _cleanup_free_ char *path = NULL;
1186
1187 path = unit_default_cgroup_path(u);
1188 if (!path)
1189 return log_oom();
1190
1191 r = unit_set_cgroup_path(u, path);
1192 if (r == -EEXIST)
1193 return log_unit_error_errno(u, r, "Control group %s exists already.", path);
1194 if (r < 0)
1195 return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path);
1196 }
1197
1198 /* First, create our own group */
1199 r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path);
1200 if (r < 0)
1201 return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path);
1202
1203 /* Start watching it */
1204 (void) unit_watch_cgroup(u);
1205
1206 /* Enable all controllers we need */
1207 r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path);
1208 if (r < 0)
1209 log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path);
1210
1211 /* Keep track that this is now realized */
1212 u->cgroup_realized = true;
1213 u->cgroup_realized_mask = target_mask;
1214 u->cgroup_enabled_mask = enable_mask;
1215
1216 if (u->type != UNIT_SLICE && !c->delegate) {
1217
1218 /* Then, possibly move things over, but not if
1219 * subgroups may contain processes, which is the case
1220 * for slice and delegation units. */
1221 r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u);
1222 if (r < 0)
1223 log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path);
1224 }
1225
1226 return 0;
1227 }
1228
1229 int unit_attach_pids_to_cgroup(Unit *u) {
1230 int r;
1231 assert(u);
1232
1233 r = unit_realize_cgroup(u);
1234 if (r < 0)
1235 return r;
1236
1237 r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u);
1238 if (r < 0)
1239 return r;
1240
1241 return 0;
1242 }
1243
1244 static bool unit_has_mask_realized(Unit *u, CGroupMask target_mask, CGroupMask enable_mask) {
1245 assert(u);
1246
1247 return u->cgroup_realized && u->cgroup_realized_mask == target_mask && u->cgroup_enabled_mask == enable_mask;
1248 }
1249
1250 /* Check if necessary controllers and attributes for a unit are in place.
1251 *
1252 * If so, do nothing.
1253 * If not, create paths, move processes over, and set attributes.
1254 *
1255 * Returns 0 on success and < 0 on failure. */
1256 static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
1257 CGroupMask target_mask, enable_mask;
1258 int r;
1259
1260 assert(u);
1261
1262 if (u->in_cgroup_queue) {
1263 LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u);
1264 u->in_cgroup_queue = false;
1265 }
1266
1267 target_mask = unit_get_target_mask(u);
1268 enable_mask = unit_get_enable_mask(u);
1269
1270 if (unit_has_mask_realized(u, target_mask, enable_mask))
1271 return 0;
1272
1273 /* First, realize parents */
1274 if (UNIT_ISSET(u->slice)) {
1275 r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
1276 if (r < 0)
1277 return r;
1278 }
1279
1280 /* And then do the real work */
1281 r = unit_create_cgroup(u, target_mask, enable_mask);
1282 if (r < 0)
1283 return r;
1284
1285 /* Finally, apply the necessary attributes. */
1286 cgroup_context_apply(u, target_mask, state);
1287
1288 return 0;
1289 }
1290
1291 static void unit_add_to_cgroup_queue(Unit *u) {
1292
1293 if (u->in_cgroup_queue)
1294 return;
1295
1296 LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u);
1297 u->in_cgroup_queue = true;
1298 }
1299
1300 unsigned manager_dispatch_cgroup_queue(Manager *m) {
1301 ManagerState state;
1302 unsigned n = 0;
1303 Unit *i;
1304 int r;
1305
1306 state = manager_state(m);
1307
1308 while ((i = m->cgroup_queue)) {
1309 assert(i->in_cgroup_queue);
1310
1311 r = unit_realize_cgroup_now(i, state);
1312 if (r < 0)
1313 log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id);
1314
1315 n++;
1316 }
1317
1318 return n;
1319 }
1320
1321 static void unit_queue_siblings(Unit *u) {
1322 Unit *slice;
1323
1324 /* This adds the siblings of the specified unit and the
1325 * siblings of all parent units to the cgroup queue. (But
1326 * neither the specified unit itself nor the parents.) */
1327
1328 while ((slice = UNIT_DEREF(u->slice))) {
1329 Iterator i;
1330 Unit *m;
1331
1332 SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) {
1333 if (m == u)
1334 continue;
1335
1336 /* Skip units that have a dependency on the slice
1337 * but aren't actually in it. */
1338 if (UNIT_DEREF(m->slice) != slice)
1339 continue;
1340
1341 /* No point in doing cgroup application for units
1342 * without active processes. */
1343 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
1344 continue;
1345
1346 /* If the unit doesn't need any new controllers
1347 * and has current ones realized, it doesn't need
1348 * any changes. */
1349 if (unit_has_mask_realized(m, unit_get_target_mask(m), unit_get_enable_mask(m)))
1350 continue;
1351
1352 unit_add_to_cgroup_queue(m);
1353 }
1354
1355 u = slice;
1356 }
1357 }
1358
1359 int unit_realize_cgroup(Unit *u) {
1360 assert(u);
1361
1362 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1363 return 0;
1364
1365 /* So, here's the deal: when realizing the cgroups for this
1366 * unit, we need to first create all parents, but there's more
1367 * actually: for the weight-based controllers we also need to
1368 * make sure that all our siblings (i.e. units that are in the
1369 * same slice as we are) have cgroups, too. Otherwise, things
1370 * would become very uneven as each of their processes would
1371 * get as much resources as all our group together. This call
1372 * will synchronously create the parent cgroups, but will
1373 * defer work on the siblings to the next event loop
1374 * iteration. */
1375
1376 /* Add all sibling slices to the cgroup queue. */
1377 unit_queue_siblings(u);
1378
1379 /* And realize this one now (and apply the values) */
1380 return unit_realize_cgroup_now(u, manager_state(u->manager));
1381 }
1382
1383 void unit_release_cgroup(Unit *u) {
1384 assert(u);
1385
1386 /* Forgets all cgroup details for this cgroup */
1387
1388 if (u->cgroup_path) {
1389 (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
1390 u->cgroup_path = mfree(u->cgroup_path);
1391 }
1392
1393 if (u->cgroup_inotify_wd >= 0) {
1394 if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0)
1395 log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id);
1396
1397 (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd));
1398 u->cgroup_inotify_wd = -1;
1399 }
1400 }
1401
1402 void unit_prune_cgroup(Unit *u) {
1403 int r;
1404 bool is_root_slice;
1405
1406 assert(u);
1407
1408 /* Removes the cgroup, if empty and possible, and stops watching it. */
1409
1410 if (!u->cgroup_path)
1411 return;
1412
1413 is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE);
1414
1415 r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice);
1416 if (r < 0) {
1417 log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path);
1418 return;
1419 }
1420
1421 if (is_root_slice)
1422 return;
1423
1424 unit_release_cgroup(u);
1425
1426 u->cgroup_realized = false;
1427 u->cgroup_realized_mask = 0;
1428 u->cgroup_enabled_mask = 0;
1429 }
1430
1431 int unit_search_main_pid(Unit *u, pid_t *ret) {
1432 _cleanup_fclose_ FILE *f = NULL;
1433 pid_t pid = 0, npid, mypid;
1434 int r;
1435
1436 assert(u);
1437 assert(ret);
1438
1439 if (!u->cgroup_path)
1440 return -ENXIO;
1441
1442 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f);
1443 if (r < 0)
1444 return r;
1445
1446 mypid = getpid();
1447 while (cg_read_pid(f, &npid) > 0) {
1448 pid_t ppid;
1449
1450 if (npid == pid)
1451 continue;
1452
1453 /* Ignore processes that aren't our kids */
1454 if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid)
1455 continue;
1456
1457 if (pid != 0)
1458 /* Dang, there's more than one daemonized PID
1459 in this group, so we don't know what process
1460 is the main process. */
1461
1462 return -ENODATA;
1463
1464 pid = npid;
1465 }
1466
1467 *ret = pid;
1468 return 0;
1469 }
1470
1471 static int unit_watch_pids_in_path(Unit *u, const char *path) {
1472 _cleanup_closedir_ DIR *d = NULL;
1473 _cleanup_fclose_ FILE *f = NULL;
1474 int ret = 0, r;
1475
1476 assert(u);
1477 assert(path);
1478
1479 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
1480 if (r < 0)
1481 ret = r;
1482 else {
1483 pid_t pid;
1484
1485 while ((r = cg_read_pid(f, &pid)) > 0) {
1486 r = unit_watch_pid(u, pid);
1487 if (r < 0 && ret >= 0)
1488 ret = r;
1489 }
1490
1491 if (r < 0 && ret >= 0)
1492 ret = r;
1493 }
1494
1495 r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
1496 if (r < 0) {
1497 if (ret >= 0)
1498 ret = r;
1499 } else {
1500 char *fn;
1501
1502 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1503 _cleanup_free_ char *p = NULL;
1504
1505 p = strjoin(path, "/", fn, NULL);
1506 free(fn);
1507
1508 if (!p)
1509 return -ENOMEM;
1510
1511 r = unit_watch_pids_in_path(u, p);
1512 if (r < 0 && ret >= 0)
1513 ret = r;
1514 }
1515
1516 if (r < 0 && ret >= 0)
1517 ret = r;
1518 }
1519
1520 return ret;
1521 }
1522
1523 int unit_watch_all_pids(Unit *u) {
1524 assert(u);
1525
1526 /* Adds all PIDs from our cgroup to the set of PIDs we
1527 * watch. This is a fallback logic for cases where we do not
1528 * get reliable cgroup empty notifications: we try to use
1529 * SIGCHLD as replacement. */
1530
1531 if (!u->cgroup_path)
1532 return -ENOENT;
1533
1534 if (cg_unified() > 0) /* On unified we can use proper notifications */
1535 return 0;
1536
1537 return unit_watch_pids_in_path(u, u->cgroup_path);
1538 }
1539
1540 int unit_notify_cgroup_empty(Unit *u) {
1541 int r;
1542
1543 assert(u);
1544
1545 if (!u->cgroup_path)
1546 return 0;
1547
1548 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
1549 if (r <= 0)
1550 return r;
1551
1552 unit_add_to_gc_queue(u);
1553
1554 if (UNIT_VTABLE(u)->notify_cgroup_empty)
1555 UNIT_VTABLE(u)->notify_cgroup_empty(u);
1556
1557 return 0;
1558 }
1559
1560 static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1561 Manager *m = userdata;
1562
1563 assert(s);
1564 assert(fd >= 0);
1565 assert(m);
1566
1567 for (;;) {
1568 union inotify_event_buffer buffer;
1569 struct inotify_event *e;
1570 ssize_t l;
1571
1572 l = read(fd, &buffer, sizeof(buffer));
1573 if (l < 0) {
1574 if (errno == EINTR || errno == EAGAIN)
1575 return 0;
1576
1577 return log_error_errno(errno, "Failed to read control group inotify events: %m");
1578 }
1579
1580 FOREACH_INOTIFY_EVENT(e, buffer, l) {
1581 Unit *u;
1582
1583 if (e->wd < 0)
1584 /* Queue overflow has no watch descriptor */
1585 continue;
1586
1587 if (e->mask & IN_IGNORED)
1588 /* The watch was just removed */
1589 continue;
1590
1591 u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd));
1592 if (!u) /* Not that inotify might deliver
1593 * events for a watch even after it
1594 * was removed, because it was queued
1595 * before the removal. Let's ignore
1596 * this here safely. */
1597 continue;
1598
1599 (void) unit_notify_cgroup_empty(u);
1600 }
1601 }
1602 }
1603
1604 int manager_setup_cgroup(Manager *m) {
1605 _cleanup_free_ char *path = NULL;
1606 CGroupController c;
1607 int r, unified;
1608 char *e;
1609
1610 assert(m);
1611
1612 /* 1. Determine hierarchy */
1613 m->cgroup_root = mfree(m->cgroup_root);
1614 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
1615 if (r < 0)
1616 return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
1617
1618 /* Chop off the init scope, if we are already located in it */
1619 e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1620
1621 /* LEGACY: Also chop off the system slice if we are in
1622 * it. This is to support live upgrades from older systemd
1623 * versions where PID 1 was moved there. Also see
1624 * cg_get_root_path(). */
1625 if (!e && MANAGER_IS_SYSTEM(m)) {
1626 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
1627 if (!e)
1628 e = endswith(m->cgroup_root, "/system"); /* even more legacy */
1629 }
1630 if (e)
1631 *e = 0;
1632
1633 /* And make sure to store away the root value without trailing
1634 * slash, even for the root dir, so that we can easily prepend
1635 * it everywhere. */
1636 while ((e = endswith(m->cgroup_root, "/")))
1637 *e = 0;
1638
1639 /* 2. Show data */
1640 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
1641 if (r < 0)
1642 return log_error_errno(r, "Cannot find cgroup mount point: %m");
1643
1644 unified = cg_unified();
1645 if (unified < 0)
1646 return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m");
1647 if (unified > 0)
1648 log_debug("Unified cgroup hierarchy is located at %s.", path);
1649 else
1650 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
1651
1652 if (!m->test_run) {
1653 const char *scope_path;
1654
1655 /* 3. Install agent */
1656 if (unified) {
1657
1658 /* In the unified hierarchy we can can get
1659 * cgroup empty notifications via inotify. */
1660
1661 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1662 safe_close(m->cgroup_inotify_fd);
1663
1664 m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1665 if (m->cgroup_inotify_fd < 0)
1666 return log_error_errno(errno, "Failed to create control group inotify object: %m");
1667
1668 r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m);
1669 if (r < 0)
1670 return log_error_errno(r, "Failed to watch control group inotify object: %m");
1671
1672 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
1673 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
1674 r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-5);
1675 if (r < 0)
1676 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
1677
1678 (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify");
1679
1680 } else if (MANAGER_IS_SYSTEM(m)) {
1681
1682 /* On the legacy hierarchy we only get
1683 * notifications via cgroup agents. (Which
1684 * isn't really reliable, since it does not
1685 * generate events when control groups with
1686 * children run empty. */
1687
1688 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
1689 if (r < 0)
1690 log_warning_errno(r, "Failed to install release agent, ignoring: %m");
1691 else if (r > 0)
1692 log_debug("Installed release agent.");
1693 else if (r == 0)
1694 log_debug("Release agent already installed.");
1695 }
1696
1697 /* 4. Make sure we are in the special "init.scope" unit in the root slice. */
1698 scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1699 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1700 if (r < 0)
1701 return log_error_errno(r, "Failed to create %s control group: %m", scope_path);
1702
1703 /* also, move all other userspace processes remaining
1704 * in the root cgroup into that scope. */
1705 r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, false);
1706 if (r < 0)
1707 log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m");
1708
1709 /* 5. And pin it, so that it cannot be unmounted */
1710 safe_close(m->pin_cgroupfs_fd);
1711 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
1712 if (m->pin_cgroupfs_fd < 0)
1713 return log_error_errno(errno, "Failed to open pin file: %m");
1714
1715 /* 6. Always enable hierarchical support if it exists... */
1716 if (!unified)
1717 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
1718 }
1719
1720 /* 7. Figure out which controllers are supported */
1721 r = cg_mask_supported(&m->cgroup_supported);
1722 if (r < 0)
1723 return log_error_errno(r, "Failed to determine supported controllers: %m");
1724
1725 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++)
1726 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c)));
1727
1728 return 0;
1729 }
1730
1731 void manager_shutdown_cgroup(Manager *m, bool delete) {
1732 assert(m);
1733
1734 /* We can't really delete the group, since we are in it. But
1735 * let's trim it. */
1736 if (delete && m->cgroup_root)
1737 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
1738
1739 m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit);
1740
1741 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1742 m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd);
1743
1744 m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
1745
1746 m->cgroup_root = mfree(m->cgroup_root);
1747 }
1748
1749 Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
1750 char *p;
1751 Unit *u;
1752
1753 assert(m);
1754 assert(cgroup);
1755
1756 u = hashmap_get(m->cgroup_unit, cgroup);
1757 if (u)
1758 return u;
1759
1760 p = strdupa(cgroup);
1761 for (;;) {
1762 char *e;
1763
1764 e = strrchr(p, '/');
1765 if (!e || e == p)
1766 return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE);
1767
1768 *e = 0;
1769
1770 u = hashmap_get(m->cgroup_unit, p);
1771 if (u)
1772 return u;
1773 }
1774 }
1775
1776 Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) {
1777 _cleanup_free_ char *cgroup = NULL;
1778 int r;
1779
1780 assert(m);
1781
1782 if (pid <= 0)
1783 return NULL;
1784
1785 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
1786 if (r < 0)
1787 return NULL;
1788
1789 return manager_get_unit_by_cgroup(m, cgroup);
1790 }
1791
1792 Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
1793 Unit *u;
1794
1795 assert(m);
1796
1797 if (pid <= 0)
1798 return NULL;
1799
1800 if (pid == 1)
1801 return hashmap_get(m->units, SPECIAL_INIT_SCOPE);
1802
1803 u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid));
1804 if (u)
1805 return u;
1806
1807 u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid));
1808 if (u)
1809 return u;
1810
1811 return manager_get_unit_by_pid_cgroup(m, pid);
1812 }
1813
1814 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
1815 Unit *u;
1816
1817 assert(m);
1818 assert(cgroup);
1819
1820 log_debug("Got cgroup empty notification for: %s", cgroup);
1821
1822 u = manager_get_unit_by_cgroup(m, cgroup);
1823 if (!u)
1824 return 0;
1825
1826 return unit_notify_cgroup_empty(u);
1827 }
1828
1829 int unit_get_memory_current(Unit *u, uint64_t *ret) {
1830 _cleanup_free_ char *v = NULL;
1831 int r;
1832
1833 assert(u);
1834 assert(ret);
1835
1836 if (!u->cgroup_path)
1837 return -ENODATA;
1838
1839 if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
1840 return -ENODATA;
1841
1842 if (cg_unified() <= 0)
1843 r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v);
1844 else
1845 r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v);
1846 if (r == -ENOENT)
1847 return -ENODATA;
1848 if (r < 0)
1849 return r;
1850
1851 return safe_atou64(v, ret);
1852 }
1853
1854 int unit_get_tasks_current(Unit *u, uint64_t *ret) {
1855 _cleanup_free_ char *v = NULL;
1856 int r;
1857
1858 assert(u);
1859 assert(ret);
1860
1861 if (!u->cgroup_path)
1862 return -ENODATA;
1863
1864 if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0)
1865 return -ENODATA;
1866
1867 r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v);
1868 if (r == -ENOENT)
1869 return -ENODATA;
1870 if (r < 0)
1871 return r;
1872
1873 return safe_atou64(v, ret);
1874 }
1875
1876 static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) {
1877 _cleanup_free_ char *v = NULL;
1878 uint64_t ns;
1879 int r;
1880
1881 assert(u);
1882 assert(ret);
1883
1884 if (!u->cgroup_path)
1885 return -ENODATA;
1886
1887 if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0)
1888 return -ENODATA;
1889
1890 r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v);
1891 if (r == -ENOENT)
1892 return -ENODATA;
1893 if (r < 0)
1894 return r;
1895
1896 r = safe_atou64(v, &ns);
1897 if (r < 0)
1898 return r;
1899
1900 *ret = ns;
1901 return 0;
1902 }
1903
1904 int unit_get_cpu_usage(Unit *u, nsec_t *ret) {
1905 nsec_t ns;
1906 int r;
1907
1908 r = unit_get_cpu_usage_raw(u, &ns);
1909 if (r < 0)
1910 return r;
1911
1912 if (ns > u->cpuacct_usage_base)
1913 ns -= u->cpuacct_usage_base;
1914 else
1915 ns = 0;
1916
1917 *ret = ns;
1918 return 0;
1919 }
1920
1921 int unit_reset_cpu_usage(Unit *u) {
1922 nsec_t ns;
1923 int r;
1924
1925 assert(u);
1926
1927 r = unit_get_cpu_usage_raw(u, &ns);
1928 if (r < 0) {
1929 u->cpuacct_usage_base = 0;
1930 return r;
1931 }
1932
1933 u->cpuacct_usage_base = ns;
1934 return 0;
1935 }
1936
1937 bool unit_cgroup_delegate(Unit *u) {
1938 CGroupContext *c;
1939
1940 assert(u);
1941
1942 c = unit_get_cgroup_context(u);
1943 if (!c)
1944 return false;
1945
1946 return c->delegate;
1947 }
1948
1949 void unit_invalidate_cgroup(Unit *u, CGroupMask m) {
1950 assert(u);
1951
1952 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1953 return;
1954
1955 if (m == 0)
1956 return;
1957
1958 /* always invalidate compat pairs together */
1959 if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO))
1960 m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
1961
1962 if ((u->cgroup_realized_mask & m) == 0)
1963 return;
1964
1965 u->cgroup_realized_mask &= ~m;
1966 unit_add_to_cgroup_queue(u);
1967 }
1968
1969 void manager_invalidate_startup_units(Manager *m) {
1970 Iterator i;
1971 Unit *u;
1972
1973 assert(m);
1974
1975 SET_FOREACH(u, m->startup_units, i)
1976 unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
1977 }
1978
1979 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
1980 [CGROUP_AUTO] = "auto",
1981 [CGROUP_CLOSED] = "closed",
1982 [CGROUP_STRICT] = "strict",
1983 };
1984
1985 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);
Unnamed repository; edit this file 'description' to name the repository.