]> 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
core: add "invocation ID" concept to service manager
[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 static bool cgroup_compat_warned = false;
41
42 if (cgroup_compat_warned)
43 return;
44
45 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details.");
46 cgroup_compat_warned = true;
47 }
48
49 #define log_cgroup_compat(unit, fmt, ...) do { \
50 cgroup_compat_warn(); \
51 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
52 } while (false)
53
54 void cgroup_context_init(CGroupContext *c) {
55 assert(c);
56
57 /* Initialize everything to the kernel defaults, assuming the
58 * structure is preinitialized to 0 */
59
60 c->cpu_weight = CGROUP_WEIGHT_INVALID;
61 c->startup_cpu_weight = CGROUP_WEIGHT_INVALID;
62 c->cpu_quota_per_sec_usec = USEC_INFINITY;
63
64 c->cpu_shares = CGROUP_CPU_SHARES_INVALID;
65 c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID;
66
67 c->memory_high = CGROUP_LIMIT_MAX;
68 c->memory_max = CGROUP_LIMIT_MAX;
69 c->memory_swap_max = CGROUP_LIMIT_MAX;
70
71 c->memory_limit = CGROUP_LIMIT_MAX;
72
73 c->io_weight = CGROUP_WEIGHT_INVALID;
74 c->startup_io_weight = CGROUP_WEIGHT_INVALID;
75
76 c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
77 c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
78
79 c->tasks_max = (uint64_t) -1;
80 }
81
82 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
83 assert(c);
84 assert(a);
85
86 LIST_REMOVE(device_allow, c->device_allow, a);
87 free(a->path);
88 free(a);
89 }
90
91 void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) {
92 assert(c);
93 assert(w);
94
95 LIST_REMOVE(device_weights, c->io_device_weights, w);
96 free(w->path);
97 free(w);
98 }
99
100 void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) {
101 assert(c);
102 assert(l);
103
104 LIST_REMOVE(device_limits, c->io_device_limits, l);
105 free(l->path);
106 free(l);
107 }
108
109 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
110 assert(c);
111 assert(w);
112
113 LIST_REMOVE(device_weights, c->blockio_device_weights, w);
114 free(w->path);
115 free(w);
116 }
117
118 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
119 assert(c);
120 assert(b);
121
122 LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
123 free(b->path);
124 free(b);
125 }
126
127 void cgroup_context_done(CGroupContext *c) {
128 assert(c);
129
130 while (c->io_device_weights)
131 cgroup_context_free_io_device_weight(c, c->io_device_weights);
132
133 while (c->io_device_limits)
134 cgroup_context_free_io_device_limit(c, c->io_device_limits);
135
136 while (c->blockio_device_weights)
137 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
138
139 while (c->blockio_device_bandwidths)
140 cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
141
142 while (c->device_allow)
143 cgroup_context_free_device_allow(c, c->device_allow);
144 }
145
146 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
147 CGroupIODeviceLimit *il;
148 CGroupIODeviceWeight *iw;
149 CGroupBlockIODeviceBandwidth *b;
150 CGroupBlockIODeviceWeight *w;
151 CGroupDeviceAllow *a;
152 char u[FORMAT_TIMESPAN_MAX];
153
154 assert(c);
155 assert(f);
156
157 prefix = strempty(prefix);
158
159 fprintf(f,
160 "%sCPUAccounting=%s\n"
161 "%sIOAccounting=%s\n"
162 "%sBlockIOAccounting=%s\n"
163 "%sMemoryAccounting=%s\n"
164 "%sTasksAccounting=%s\n"
165 "%sCPUWeight=%" PRIu64 "\n"
166 "%sStartupCPUWeight=%" PRIu64 "\n"
167 "%sCPUShares=%" PRIu64 "\n"
168 "%sStartupCPUShares=%" PRIu64 "\n"
169 "%sCPUQuotaPerSecSec=%s\n"
170 "%sIOWeight=%" PRIu64 "\n"
171 "%sStartupIOWeight=%" PRIu64 "\n"
172 "%sBlockIOWeight=%" PRIu64 "\n"
173 "%sStartupBlockIOWeight=%" PRIu64 "\n"
174 "%sMemoryLow=%" PRIu64 "\n"
175 "%sMemoryHigh=%" PRIu64 "\n"
176 "%sMemoryMax=%" PRIu64 "\n"
177 "%sMemorySwapMax=%" PRIu64 "\n"
178 "%sMemoryLimit=%" PRIu64 "\n"
179 "%sTasksMax=%" PRIu64 "\n"
180 "%sDevicePolicy=%s\n"
181 "%sDelegate=%s\n",
182 prefix, yes_no(c->cpu_accounting),
183 prefix, yes_no(c->io_accounting),
184 prefix, yes_no(c->blockio_accounting),
185 prefix, yes_no(c->memory_accounting),
186 prefix, yes_no(c->tasks_accounting),
187 prefix, c->cpu_weight,
188 prefix, c->startup_cpu_weight,
189 prefix, c->cpu_shares,
190 prefix, c->startup_cpu_shares,
191 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
192 prefix, c->io_weight,
193 prefix, c->startup_io_weight,
194 prefix, c->blockio_weight,
195 prefix, c->startup_blockio_weight,
196 prefix, c->memory_low,
197 prefix, c->memory_high,
198 prefix, c->memory_max,
199 prefix, c->memory_swap_max,
200 prefix, c->memory_limit,
201 prefix, c->tasks_max,
202 prefix, cgroup_device_policy_to_string(c->device_policy),
203 prefix, yes_no(c->delegate));
204
205 LIST_FOREACH(device_allow, a, c->device_allow)
206 fprintf(f,
207 "%sDeviceAllow=%s %s%s%s\n",
208 prefix,
209 a->path,
210 a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
211
212 LIST_FOREACH(device_weights, iw, c->io_device_weights)
213 fprintf(f,
214 "%sIODeviceWeight=%s %" PRIu64,
215 prefix,
216 iw->path,
217 iw->weight);
218
219 LIST_FOREACH(device_limits, il, c->io_device_limits) {
220 char buf[FORMAT_BYTES_MAX];
221 CGroupIOLimitType type;
222
223 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
224 if (il->limits[type] != cgroup_io_limit_defaults[type])
225 fprintf(f,
226 "%s%s=%s %s\n",
227 prefix,
228 cgroup_io_limit_type_to_string(type),
229 il->path,
230 format_bytes(buf, sizeof(buf), il->limits[type]));
231 }
232
233 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
234 fprintf(f,
235 "%sBlockIODeviceWeight=%s %" PRIu64,
236 prefix,
237 w->path,
238 w->weight);
239
240 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
241 char buf[FORMAT_BYTES_MAX];
242
243 if (b->rbps != CGROUP_LIMIT_MAX)
244 fprintf(f,
245 "%sBlockIOReadBandwidth=%s %s\n",
246 prefix,
247 b->path,
248 format_bytes(buf, sizeof(buf), b->rbps));
249 if (b->wbps != CGROUP_LIMIT_MAX)
250 fprintf(f,
251 "%sBlockIOWriteBandwidth=%s %s\n",
252 prefix,
253 b->path,
254 format_bytes(buf, sizeof(buf), b->wbps));
255 }
256 }
257
258 static int lookup_block_device(const char *p, dev_t *dev) {
259 struct stat st;
260 int r;
261
262 assert(p);
263 assert(dev);
264
265 r = stat(p, &st);
266 if (r < 0)
267 return log_warning_errno(errno, "Couldn't stat device %s: %m", p);
268
269 if (S_ISBLK(st.st_mode))
270 *dev = st.st_rdev;
271 else if (major(st.st_dev) != 0) {
272 /* If this is not a device node then find the block
273 * device this file is stored on */
274 *dev = st.st_dev;
275
276 /* If this is a partition, try to get the originating
277 * block device */
278 block_get_whole_disk(*dev, dev);
279 } else {
280 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p);
281 return -ENODEV;
282 }
283
284 return 0;
285 }
286
287 static int whitelist_device(const char *path, const char *node, const char *acc) {
288 char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
289 struct stat st;
290 int r;
291
292 assert(path);
293 assert(acc);
294
295 if (stat(node, &st) < 0) {
296 log_warning("Couldn't stat device %s", node);
297 return -errno;
298 }
299
300 if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
301 log_warning("%s is not a device.", node);
302 return -ENODEV;
303 }
304
305 sprintf(buf,
306 "%c %u:%u %s",
307 S_ISCHR(st.st_mode) ? 'c' : 'b',
308 major(st.st_rdev), minor(st.st_rdev),
309 acc);
310
311 r = cg_set_attribute("devices", path, "devices.allow", buf);
312 if (r < 0)
313 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
314 "Failed to set devices.allow on %s: %m", path);
315
316 return r;
317 }
318
319 static int whitelist_major(const char *path, const char *name, char type, const char *acc) {
320 _cleanup_fclose_ FILE *f = NULL;
321 char line[LINE_MAX];
322 bool good = false;
323 int r;
324
325 assert(path);
326 assert(acc);
327 assert(type == 'b' || type == 'c');
328
329 f = fopen("/proc/devices", "re");
330 if (!f)
331 return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
332
333 FOREACH_LINE(line, f, goto fail) {
334 char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w;
335 unsigned maj;
336
337 truncate_nl(line);
338
339 if (type == 'c' && streq(line, "Character devices:")) {
340 good = true;
341 continue;
342 }
343
344 if (type == 'b' && streq(line, "Block devices:")) {
345 good = true;
346 continue;
347 }
348
349 if (isempty(line)) {
350 good = false;
351 continue;
352 }
353
354 if (!good)
355 continue;
356
357 p = strstrip(line);
358
359 w = strpbrk(p, WHITESPACE);
360 if (!w)
361 continue;
362 *w = 0;
363
364 r = safe_atou(p, &maj);
365 if (r < 0)
366 continue;
367 if (maj <= 0)
368 continue;
369
370 w++;
371 w += strspn(w, WHITESPACE);
372
373 if (fnmatch(name, w, 0) != 0)
374 continue;
375
376 sprintf(buf,
377 "%c %u:* %s",
378 type,
379 maj,
380 acc);
381
382 r = cg_set_attribute("devices", path, "devices.allow", buf);
383 if (r < 0)
384 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
385 "Failed to set devices.allow on %s: %m", path);
386 }
387
388 return 0;
389
390 fail:
391 log_warning_errno(errno, "Failed to read /proc/devices: %m");
392 return -errno;
393 }
394
395 static bool cgroup_context_has_cpu_weight(CGroupContext *c) {
396 return c->cpu_weight != CGROUP_WEIGHT_INVALID ||
397 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID;
398 }
399
400 static bool cgroup_context_has_cpu_shares(CGroupContext *c) {
401 return c->cpu_shares != CGROUP_CPU_SHARES_INVALID ||
402 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID;
403 }
404
405 static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) {
406 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
407 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID)
408 return c->startup_cpu_weight;
409 else if (c->cpu_weight != CGROUP_WEIGHT_INVALID)
410 return c->cpu_weight;
411 else
412 return CGROUP_WEIGHT_DEFAULT;
413 }
414
415 static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) {
416 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
417 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID)
418 return c->startup_cpu_shares;
419 else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID)
420 return c->cpu_shares;
421 else
422 return CGROUP_CPU_SHARES_DEFAULT;
423 }
424
425 static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) {
426 char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)];
427 int r;
428
429 xsprintf(buf, "%" PRIu64 "\n", weight);
430 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf);
431 if (r < 0)
432 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
433 "Failed to set cpu.weight: %m");
434
435 if (quota != USEC_INFINITY)
436 xsprintf(buf, USEC_FMT " " USEC_FMT "\n",
437 quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC);
438 else
439 xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
440
441 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf);
442
443 if (r < 0)
444 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
445 "Failed to set cpu.max: %m");
446 }
447
448 static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) {
449 char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1];
450 int r;
451
452 xsprintf(buf, "%" PRIu64 "\n", shares);
453 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf);
454 if (r < 0)
455 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
456 "Failed to set cpu.shares: %m");
457
458 xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
459 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf);
460 if (r < 0)
461 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
462 "Failed to set cpu.cfs_period_us: %m");
463
464 if (quota != USEC_INFINITY) {
465 xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC);
466 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf);
467 } else
468 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1");
469 if (r < 0)
470 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
471 "Failed to set cpu.cfs_quota_us: %m");
472 }
473
474 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) {
475 return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT,
476 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
477 }
478
479 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) {
480 return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT,
481 CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX);
482 }
483
484 static bool cgroup_context_has_io_config(CGroupContext *c) {
485 return c->io_accounting ||
486 c->io_weight != CGROUP_WEIGHT_INVALID ||
487 c->startup_io_weight != CGROUP_WEIGHT_INVALID ||
488 c->io_device_weights ||
489 c->io_device_limits;
490 }
491
492 static bool cgroup_context_has_blockio_config(CGroupContext *c) {
493 return c->blockio_accounting ||
494 c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
495 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
496 c->blockio_device_weights ||
497 c->blockio_device_bandwidths;
498 }
499
500 static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) {
501 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
502 c->startup_io_weight != CGROUP_WEIGHT_INVALID)
503 return c->startup_io_weight;
504 else if (c->io_weight != CGROUP_WEIGHT_INVALID)
505 return c->io_weight;
506 else
507 return CGROUP_WEIGHT_DEFAULT;
508 }
509
510 static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) {
511 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
512 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
513 return c->startup_blockio_weight;
514 else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
515 return c->blockio_weight;
516 else
517 return CGROUP_BLKIO_WEIGHT_DEFAULT;
518 }
519
520 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) {
521 return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT,
522 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
523 }
524
525 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) {
526 return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT,
527 CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX);
528 }
529
530 static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) {
531 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
532 dev_t dev;
533 int r;
534
535 r = lookup_block_device(dev_path, &dev);
536 if (r < 0)
537 return;
538
539 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight);
540 r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf);
541 if (r < 0)
542 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
543 "Failed to set io.weight: %m");
544 }
545
546 static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) {
547 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
548 dev_t dev;
549 int r;
550
551 r = lookup_block_device(dev_path, &dev);
552 if (r < 0)
553 return;
554
555 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight);
556 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf);
557 if (r < 0)
558 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
559 "Failed to set blkio.weight_device: %m");
560 }
561
562 static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) {
563 char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)];
564 char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
565 CGroupIOLimitType type;
566 dev_t dev;
567 unsigned n = 0;
568 int r;
569
570 r = lookup_block_device(dev_path, &dev);
571 if (r < 0)
572 return 0;
573
574 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) {
575 if (limits[type] != cgroup_io_limit_defaults[type]) {
576 xsprintf(limit_bufs[type], "%" PRIu64, limits[type]);
577 n++;
578 } else {
579 xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0");
580 }
581 }
582
583 xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev),
584 limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX],
585 limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]);
586 r = cg_set_attribute("io", u->cgroup_path, "io.max", buf);
587 if (r < 0)
588 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
589 "Failed to set io.max: %m");
590 return n;
591 }
592
593 static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) {
594 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
595 dev_t dev;
596 unsigned n = 0;
597 int r;
598
599 r = lookup_block_device(dev_path, &dev);
600 if (r < 0)
601 return 0;
602
603 if (rbps != CGROUP_LIMIT_MAX)
604 n++;
605 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps);
606 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf);
607 if (r < 0)
608 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
609 "Failed to set blkio.throttle.read_bps_device: %m");
610
611 if (wbps != CGROUP_LIMIT_MAX)
612 n++;
613 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps);
614 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf);
615 if (r < 0)
616 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
617 "Failed to set blkio.throttle.write_bps_device: %m");
618
619 return n;
620 }
621
622 static bool cgroup_context_has_unified_memory_config(CGroupContext *c) {
623 return c->memory_low > 0 || c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || c->memory_swap_max != CGROUP_LIMIT_MAX;
624 }
625
626 static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) {
627 char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
628 int r;
629
630 if (v != CGROUP_LIMIT_MAX)
631 xsprintf(buf, "%" PRIu64 "\n", v);
632
633 r = cg_set_attribute("memory", u->cgroup_path, file, buf);
634 if (r < 0)
635 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
636 "Failed to set %s: %m", file);
637 }
638
639 static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) {
640 const char *path;
641 CGroupContext *c;
642 bool is_root;
643 int r;
644
645 assert(u);
646
647 c = unit_get_cgroup_context(u);
648 path = u->cgroup_path;
649
650 assert(c);
651 assert(path);
652
653 if (mask == 0)
654 return;
655
656 /* Some cgroup attributes are not supported on the root cgroup,
657 * hence silently ignore */
658 is_root = isempty(path) || path_equal(path, "/");
659 if (is_root)
660 /* Make sure we don't try to display messages with an empty path. */
661 path = "/";
662
663 /* We generally ignore errors caused by read-only mounted
664 * cgroup trees (assuming we are running in a container then),
665 * and missing cgroups, i.e. EROFS and ENOENT. */
666
667 if ((mask & CGROUP_MASK_CPU) && !is_root) {
668 bool has_weight = cgroup_context_has_cpu_weight(c);
669 bool has_shares = cgroup_context_has_cpu_shares(c);
670
671 if (cg_all_unified() > 0) {
672 uint64_t weight;
673
674 if (has_weight)
675 weight = cgroup_context_cpu_weight(c, state);
676 else if (has_shares) {
677 uint64_t shares = cgroup_context_cpu_shares(c, state);
678
679 weight = cgroup_cpu_shares_to_weight(shares);
680
681 log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s",
682 shares, weight, path);
683 } else
684 weight = CGROUP_WEIGHT_DEFAULT;
685
686 cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec);
687 } else {
688 uint64_t shares;
689
690 if (has_shares)
691 shares = cgroup_context_cpu_shares(c, state);
692 else if (has_weight) {
693 uint64_t weight = cgroup_context_cpu_weight(c, state);
694
695 shares = cgroup_cpu_weight_to_shares(weight);
696
697 log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s",
698 weight, shares, path);
699 } else
700 shares = CGROUP_CPU_SHARES_DEFAULT;
701
702 cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec);
703 }
704 }
705
706 if (mask & CGROUP_MASK_IO) {
707 bool has_io = cgroup_context_has_io_config(c);
708 bool has_blockio = cgroup_context_has_blockio_config(c);
709
710 if (!is_root) {
711 char buf[8+DECIMAL_STR_MAX(uint64_t)+1];
712 uint64_t weight;
713
714 if (has_io)
715 weight = cgroup_context_io_weight(c, state);
716 else if (has_blockio) {
717 uint64_t blkio_weight = cgroup_context_blkio_weight(c, state);
718
719 weight = cgroup_weight_blkio_to_io(blkio_weight);
720
721 log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64,
722 blkio_weight, weight);
723 } else
724 weight = CGROUP_WEIGHT_DEFAULT;
725
726 xsprintf(buf, "default %" PRIu64 "\n", weight);
727 r = cg_set_attribute("io", path, "io.weight", buf);
728 if (r < 0)
729 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
730 "Failed to set io.weight: %m");
731
732 if (has_io) {
733 CGroupIODeviceWeight *w;
734
735 /* FIXME: no way to reset this list */
736 LIST_FOREACH(device_weights, w, c->io_device_weights)
737 cgroup_apply_io_device_weight(u, w->path, w->weight);
738 } else if (has_blockio) {
739 CGroupBlockIODeviceWeight *w;
740
741 /* FIXME: no way to reset this list */
742 LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
743 weight = cgroup_weight_blkio_to_io(w->weight);
744
745 log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s",
746 w->weight, weight, w->path);
747
748 cgroup_apply_io_device_weight(u, w->path, weight);
749 }
750 }
751 }
752
753 /* Apply limits and free ones without config. */
754 if (has_io) {
755 CGroupIODeviceLimit *l, *next;
756
757 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
758 if (!cgroup_apply_io_device_limit(u, l->path, l->limits))
759 cgroup_context_free_io_device_limit(c, l);
760 }
761 } else if (has_blockio) {
762 CGroupBlockIODeviceBandwidth *b, *next;
763
764 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
765 uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX];
766 CGroupIOLimitType type;
767
768 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
769 limits[type] = cgroup_io_limit_defaults[type];
770
771 limits[CGROUP_IO_RBPS_MAX] = b->rbps;
772 limits[CGROUP_IO_WBPS_MAX] = b->wbps;
773
774 log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s",
775 b->rbps, b->wbps, b->path);
776
777 if (!cgroup_apply_io_device_limit(u, b->path, limits))
778 cgroup_context_free_blockio_device_bandwidth(c, b);
779 }
780 }
781 }
782
783 if (mask & CGROUP_MASK_BLKIO) {
784 bool has_io = cgroup_context_has_io_config(c);
785 bool has_blockio = cgroup_context_has_blockio_config(c);
786
787 if (!is_root) {
788 char buf[DECIMAL_STR_MAX(uint64_t)+1];
789 uint64_t weight;
790
791 if (has_blockio)
792 weight = cgroup_context_blkio_weight(c, state);
793 else if (has_io) {
794 uint64_t io_weight = cgroup_context_io_weight(c, state);
795
796 weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state));
797
798 log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64,
799 io_weight, weight);
800 } else
801 weight = CGROUP_BLKIO_WEIGHT_DEFAULT;
802
803 xsprintf(buf, "%" PRIu64 "\n", weight);
804 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
805 if (r < 0)
806 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
807 "Failed to set blkio.weight: %m");
808
809 if (has_blockio) {
810 CGroupBlockIODeviceWeight *w;
811
812 /* FIXME: no way to reset this list */
813 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
814 cgroup_apply_blkio_device_weight(u, w->path, w->weight);
815 } else if (has_io) {
816 CGroupIODeviceWeight *w;
817
818 /* FIXME: no way to reset this list */
819 LIST_FOREACH(device_weights, w, c->io_device_weights) {
820 weight = cgroup_weight_io_to_blkio(w->weight);
821
822 log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s",
823 w->weight, weight, w->path);
824
825 cgroup_apply_blkio_device_weight(u, w->path, weight);
826 }
827 }
828 }
829
830 /* Apply limits and free ones without config. */
831 if (has_blockio) {
832 CGroupBlockIODeviceBandwidth *b, *next;
833
834 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
835 if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps))
836 cgroup_context_free_blockio_device_bandwidth(c, b);
837 }
838 } else if (has_io) {
839 CGroupIODeviceLimit *l, *next;
840
841 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
842 log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s",
843 l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path);
844
845 if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]))
846 cgroup_context_free_io_device_limit(c, l);
847 }
848 }
849 }
850
851 if ((mask & CGROUP_MASK_MEMORY) && !is_root) {
852 if (cg_all_unified() > 0) {
853 uint64_t max = c->memory_max;
854 uint64_t swap_max = c->memory_swap_max;
855
856 if (cgroup_context_has_unified_memory_config(c)) {
857 max = c->memory_max;
858 swap_max = c->memory_swap_max;
859 } else {
860 max = c->memory_limit;
861
862 if (max != CGROUP_LIMIT_MAX)
863 log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max);
864 }
865
866 cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low);
867 cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high);
868 cgroup_apply_unified_memory_limit(u, "memory.max", max);
869 cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max);
870 } else {
871 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
872 uint64_t val = c->memory_limit;
873
874 if (val == CGROUP_LIMIT_MAX) {
875 val = c->memory_max;
876
877 if (val != CGROUP_LIMIT_MAX)
878 log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", c->memory_max);
879 }
880
881 if (val == CGROUP_LIMIT_MAX)
882 strncpy(buf, "-1\n", sizeof(buf));
883 else
884 xsprintf(buf, "%" PRIu64 "\n", val);
885
886 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
887 if (r < 0)
888 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
889 "Failed to set memory.limit_in_bytes: %m");
890 }
891 }
892
893 if ((mask & CGROUP_MASK_DEVICES) && !is_root) {
894 CGroupDeviceAllow *a;
895
896 /* Changing the devices list of a populated cgroup
897 * might result in EINVAL, hence ignore EINVAL
898 * here. */
899
900 if (c->device_allow || c->device_policy != CGROUP_AUTO)
901 r = cg_set_attribute("devices", path, "devices.deny", "a");
902 else
903 r = cg_set_attribute("devices", path, "devices.allow", "a");
904 if (r < 0)
905 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
906 "Failed to reset devices.list: %m");
907
908 if (c->device_policy == CGROUP_CLOSED ||
909 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
910 static const char auto_devices[] =
911 "/dev/null\0" "rwm\0"
912 "/dev/zero\0" "rwm\0"
913 "/dev/full\0" "rwm\0"
914 "/dev/random\0" "rwm\0"
915 "/dev/urandom\0" "rwm\0"
916 "/dev/tty\0" "rwm\0"
917 "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */
918 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
919 "/run/systemd/inaccessible/chr\0" "rwm\0"
920 "/run/systemd/inaccessible/blk\0" "rwm\0";
921
922 const char *x, *y;
923
924 NULSTR_FOREACH_PAIR(x, y, auto_devices)
925 whitelist_device(path, x, y);
926
927 whitelist_major(path, "pts", 'c', "rw");
928 whitelist_major(path, "kdbus", 'c', "rw");
929 whitelist_major(path, "kdbus/*", 'c', "rw");
930 }
931
932 LIST_FOREACH(device_allow, a, c->device_allow) {
933 char acc[4];
934 unsigned k = 0;
935
936 if (a->r)
937 acc[k++] = 'r';
938 if (a->w)
939 acc[k++] = 'w';
940 if (a->m)
941 acc[k++] = 'm';
942
943 if (k == 0)
944 continue;
945
946 acc[k++] = 0;
947
948 if (startswith(a->path, "/dev/"))
949 whitelist_device(path, a->path, acc);
950 else if (startswith(a->path, "block-"))
951 whitelist_major(path, a->path + 6, 'b', acc);
952 else if (startswith(a->path, "char-"))
953 whitelist_major(path, a->path + 5, 'c', acc);
954 else
955 log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path);
956 }
957 }
958
959 if ((mask & CGROUP_MASK_PIDS) && !is_root) {
960
961 if (c->tasks_max != CGROUP_LIMIT_MAX) {
962 char buf[DECIMAL_STR_MAX(uint64_t) + 2];
963
964 sprintf(buf, "%" PRIu64 "\n", c->tasks_max);
965 r = cg_set_attribute("pids", path, "pids.max", buf);
966 } else
967 r = cg_set_attribute("pids", path, "pids.max", "max");
968
969 if (r < 0)
970 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
971 "Failed to set pids.max: %m");
972 }
973 }
974
975 CGroupMask cgroup_context_get_mask(CGroupContext *c) {
976 CGroupMask mask = 0;
977
978 /* Figure out which controllers we need */
979
980 if (c->cpu_accounting ||
981 cgroup_context_has_cpu_weight(c) ||
982 cgroup_context_has_cpu_shares(c) ||
983 c->cpu_quota_per_sec_usec != USEC_INFINITY)
984 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
985
986 if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c))
987 mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
988
989 if (c->memory_accounting ||
990 c->memory_limit != CGROUP_LIMIT_MAX ||
991 cgroup_context_has_unified_memory_config(c))
992 mask |= CGROUP_MASK_MEMORY;
993
994 if (c->device_allow ||
995 c->device_policy != CGROUP_AUTO)
996 mask |= CGROUP_MASK_DEVICES;
997
998 if (c->tasks_accounting ||
999 c->tasks_max != (uint64_t) -1)
1000 mask |= CGROUP_MASK_PIDS;
1001
1002 return mask;
1003 }
1004
1005 CGroupMask unit_get_own_mask(Unit *u) {
1006 CGroupContext *c;
1007
1008 /* Returns the mask of controllers the unit needs for itself */
1009
1010 c = unit_get_cgroup_context(u);
1011 if (!c)
1012 return 0;
1013
1014 /* If delegation is turned on, then turn on all cgroups,
1015 * unless we are on the legacy hierarchy and the process we
1016 * fork into it is known to drop privileges, and hence
1017 * shouldn't get access to the controllers.
1018 *
1019 * Note that on the unified hierarchy it is safe to delegate
1020 * controllers to unprivileged services. */
1021
1022 if (c->delegate) {
1023 ExecContext *e;
1024
1025 e = unit_get_exec_context(u);
1026 if (!e ||
1027 exec_context_maintains_privileges(e) ||
1028 cg_all_unified() > 0)
1029 return _CGROUP_MASK_ALL;
1030 }
1031
1032 return cgroup_context_get_mask(c);
1033 }
1034
1035 CGroupMask unit_get_members_mask(Unit *u) {
1036 assert(u);
1037
1038 /* Returns the mask of controllers all of the unit's children
1039 * require, merged */
1040
1041 if (u->cgroup_members_mask_valid)
1042 return u->cgroup_members_mask;
1043
1044 u->cgroup_members_mask = 0;
1045
1046 if (u->type == UNIT_SLICE) {
1047 Unit *member;
1048 Iterator i;
1049
1050 SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) {
1051
1052 if (member == u)
1053 continue;
1054
1055 if (UNIT_DEREF(member->slice) != u)
1056 continue;
1057
1058 u->cgroup_members_mask |=
1059 unit_get_own_mask(member) |
1060 unit_get_members_mask(member);
1061 }
1062 }
1063
1064 u->cgroup_members_mask_valid = true;
1065 return u->cgroup_members_mask;
1066 }
1067
1068 CGroupMask unit_get_siblings_mask(Unit *u) {
1069 assert(u);
1070
1071 /* Returns the mask of controllers all of the unit's siblings
1072 * require, i.e. the members mask of the unit's parent slice
1073 * if there is one. */
1074
1075 if (UNIT_ISSET(u->slice))
1076 return unit_get_members_mask(UNIT_DEREF(u->slice));
1077
1078 return unit_get_own_mask(u) | unit_get_members_mask(u);
1079 }
1080
1081 CGroupMask unit_get_subtree_mask(Unit *u) {
1082
1083 /* Returns the mask of this subtree, meaning of the group
1084 * itself and its children. */
1085
1086 return unit_get_own_mask(u) | unit_get_members_mask(u);
1087 }
1088
1089 CGroupMask unit_get_target_mask(Unit *u) {
1090 CGroupMask mask;
1091
1092 /* This returns the cgroup mask of all controllers to enable
1093 * for a specific cgroup, i.e. everything it needs itself,
1094 * plus all that its children need, plus all that its siblings
1095 * need. This is primarily useful on the legacy cgroup
1096 * hierarchy, where we need to duplicate each cgroup in each
1097 * hierarchy that shall be enabled for it. */
1098
1099 mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
1100 mask &= u->manager->cgroup_supported;
1101
1102 return mask;
1103 }
1104
1105 CGroupMask unit_get_enable_mask(Unit *u) {
1106 CGroupMask mask;
1107
1108 /* This returns the cgroup mask of all controllers to enable
1109 * for the children of a specific cgroup. This is primarily
1110 * useful for the unified cgroup hierarchy, where each cgroup
1111 * controls which controllers are enabled for its children. */
1112
1113 mask = unit_get_members_mask(u);
1114 mask &= u->manager->cgroup_supported;
1115
1116 return mask;
1117 }
1118
1119 /* Recurse from a unit up through its containing slices, propagating
1120 * mask bits upward. A unit is also member of itself. */
1121 void unit_update_cgroup_members_masks(Unit *u) {
1122 CGroupMask m;
1123 bool more;
1124
1125 assert(u);
1126
1127 /* Calculate subtree mask */
1128 m = unit_get_subtree_mask(u);
1129
1130 /* See if anything changed from the previous invocation. If
1131 * not, we're done. */
1132 if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask)
1133 return;
1134
1135 more =
1136 u->cgroup_subtree_mask_valid &&
1137 ((m & ~u->cgroup_subtree_mask) != 0) &&
1138 ((~m & u->cgroup_subtree_mask) == 0);
1139
1140 u->cgroup_subtree_mask = m;
1141 u->cgroup_subtree_mask_valid = true;
1142
1143 if (UNIT_ISSET(u->slice)) {
1144 Unit *s = UNIT_DEREF(u->slice);
1145
1146 if (more)
1147 /* There's more set now than before. We
1148 * propagate the new mask to the parent's mask
1149 * (not caring if it actually was valid or
1150 * not). */
1151
1152 s->cgroup_members_mask |= m;
1153
1154 else
1155 /* There's less set now than before (or we
1156 * don't know), we need to recalculate
1157 * everything, so let's invalidate the
1158 * parent's members mask */
1159
1160 s->cgroup_members_mask_valid = false;
1161
1162 /* And now make sure that this change also hits our
1163 * grandparents */
1164 unit_update_cgroup_members_masks(s);
1165 }
1166 }
1167
1168 static const char *migrate_callback(CGroupMask mask, void *userdata) {
1169 Unit *u = userdata;
1170
1171 assert(mask != 0);
1172 assert(u);
1173
1174 while (u) {
1175 if (u->cgroup_path &&
1176 u->cgroup_realized &&
1177 (u->cgroup_realized_mask & mask) == mask)
1178 return u->cgroup_path;
1179
1180 u = UNIT_DEREF(u->slice);
1181 }
1182
1183 return NULL;
1184 }
1185
1186 char *unit_default_cgroup_path(Unit *u) {
1187 _cleanup_free_ char *escaped = NULL, *slice = NULL;
1188 int r;
1189
1190 assert(u);
1191
1192 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1193 return strdup(u->manager->cgroup_root);
1194
1195 if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) {
1196 r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice);
1197 if (r < 0)
1198 return NULL;
1199 }
1200
1201 escaped = cg_escape(u->id);
1202 if (!escaped)
1203 return NULL;
1204
1205 if (slice)
1206 return strjoin(u->manager->cgroup_root, "/", slice, "/", escaped, NULL);
1207 else
1208 return strjoin(u->manager->cgroup_root, "/", escaped, NULL);
1209 }
1210
1211 int unit_set_cgroup_path(Unit *u, const char *path) {
1212 _cleanup_free_ char *p = NULL;
1213 int r;
1214
1215 assert(u);
1216
1217 if (path) {
1218 p = strdup(path);
1219 if (!p)
1220 return -ENOMEM;
1221 } else
1222 p = NULL;
1223
1224 if (streq_ptr(u->cgroup_path, p))
1225 return 0;
1226
1227 if (p) {
1228 r = hashmap_put(u->manager->cgroup_unit, p, u);
1229 if (r < 0)
1230 return r;
1231 }
1232
1233 unit_release_cgroup(u);
1234
1235 u->cgroup_path = p;
1236 p = NULL;
1237
1238 return 1;
1239 }
1240
1241 int unit_watch_cgroup(Unit *u) {
1242 _cleanup_free_ char *events = NULL;
1243 int r;
1244
1245 assert(u);
1246
1247 if (!u->cgroup_path)
1248 return 0;
1249
1250 if (u->cgroup_inotify_wd >= 0)
1251 return 0;
1252
1253 /* Only applies to the unified hierarchy */
1254 r = cg_unified(SYSTEMD_CGROUP_CONTROLLER);
1255 if (r < 0)
1256 return log_unit_error_errno(u, r, "Failed detect whether the unified hierarchy is used: %m");
1257 if (r == 0)
1258 return 0;
1259
1260 /* Don't watch the root slice, it's pointless. */
1261 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1262 return 0;
1263
1264 r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops);
1265 if (r < 0)
1266 return log_oom();
1267
1268 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events);
1269 if (r < 0)
1270 return log_oom();
1271
1272 u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY);
1273 if (u->cgroup_inotify_wd < 0) {
1274
1275 if (errno == ENOENT) /* If the directory is already
1276 * gone we don't need to track
1277 * it, so this is not an error */
1278 return 0;
1279
1280 return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path);
1281 }
1282
1283 r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u);
1284 if (r < 0)
1285 return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m");
1286
1287 return 0;
1288 }
1289
1290 static int unit_create_cgroup(
1291 Unit *u,
1292 CGroupMask target_mask,
1293 CGroupMask enable_mask) {
1294
1295 CGroupContext *c;
1296 int r;
1297
1298 assert(u);
1299
1300 c = unit_get_cgroup_context(u);
1301 if (!c)
1302 return 0;
1303
1304 if (!u->cgroup_path) {
1305 _cleanup_free_ char *path = NULL;
1306
1307 path = unit_default_cgroup_path(u);
1308 if (!path)
1309 return log_oom();
1310
1311 r = unit_set_cgroup_path(u, path);
1312 if (r == -EEXIST)
1313 return log_unit_error_errno(u, r, "Control group %s exists already.", path);
1314 if (r < 0)
1315 return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path);
1316 }
1317
1318 /* First, create our own group */
1319 r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path);
1320 if (r < 0)
1321 return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path);
1322
1323 /* Start watching it */
1324 (void) unit_watch_cgroup(u);
1325
1326 /* Enable all controllers we need */
1327 r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path);
1328 if (r < 0)
1329 log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path);
1330
1331 /* Keep track that this is now realized */
1332 u->cgroup_realized = true;
1333 u->cgroup_realized_mask = target_mask;
1334 u->cgroup_enabled_mask = enable_mask;
1335
1336 if (u->type != UNIT_SLICE && !c->delegate) {
1337
1338 /* Then, possibly move things over, but not if
1339 * subgroups may contain processes, which is the case
1340 * for slice and delegation units. */
1341 r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u);
1342 if (r < 0)
1343 log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path);
1344 }
1345
1346 return 0;
1347 }
1348
1349 int unit_attach_pids_to_cgroup(Unit *u) {
1350 int r;
1351 assert(u);
1352
1353 r = unit_realize_cgroup(u);
1354 if (r < 0)
1355 return r;
1356
1357 r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u);
1358 if (r < 0)
1359 return r;
1360
1361 return 0;
1362 }
1363
1364 static void cgroup_xattr_apply(Unit *u) {
1365 char ids[SD_ID128_STRING_MAX];
1366 int r;
1367
1368 assert(u);
1369
1370 if (!MANAGER_IS_SYSTEM(u->manager))
1371 return;
1372
1373 if (sd_id128_is_null(u->invocation_id))
1374 return;
1375
1376 r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
1377 "trusted.invocation_id",
1378 sd_id128_to_string(u->invocation_id, ids), 32,
1379 0);
1380 if (r < 0)
1381 log_unit_warning_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path);
1382 }
1383
1384 static bool unit_has_mask_realized(Unit *u, CGroupMask target_mask, CGroupMask enable_mask) {
1385 assert(u);
1386
1387 return u->cgroup_realized && u->cgroup_realized_mask == target_mask && u->cgroup_enabled_mask == enable_mask;
1388 }
1389
1390 /* Check if necessary controllers and attributes for a unit are in place.
1391 *
1392 * If so, do nothing.
1393 * If not, create paths, move processes over, and set attributes.
1394 *
1395 * Returns 0 on success and < 0 on failure. */
1396 static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
1397 CGroupMask target_mask, enable_mask;
1398 int r;
1399
1400 assert(u);
1401
1402 if (u->in_cgroup_queue) {
1403 LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u);
1404 u->in_cgroup_queue = false;
1405 }
1406
1407 target_mask = unit_get_target_mask(u);
1408 enable_mask = unit_get_enable_mask(u);
1409
1410 if (unit_has_mask_realized(u, target_mask, enable_mask))
1411 return 0;
1412
1413 /* First, realize parents */
1414 if (UNIT_ISSET(u->slice)) {
1415 r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
1416 if (r < 0)
1417 return r;
1418 }
1419
1420 /* And then do the real work */
1421 r = unit_create_cgroup(u, target_mask, enable_mask);
1422 if (r < 0)
1423 return r;
1424
1425 /* Finally, apply the necessary attributes. */
1426 cgroup_context_apply(u, target_mask, state);
1427 cgroup_xattr_apply(u);
1428
1429 return 0;
1430 }
1431
1432 static void unit_add_to_cgroup_queue(Unit *u) {
1433
1434 if (u->in_cgroup_queue)
1435 return;
1436
1437 LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u);
1438 u->in_cgroup_queue = true;
1439 }
1440
1441 unsigned manager_dispatch_cgroup_queue(Manager *m) {
1442 ManagerState state;
1443 unsigned n = 0;
1444 Unit *i;
1445 int r;
1446
1447 state = manager_state(m);
1448
1449 while ((i = m->cgroup_queue)) {
1450 assert(i->in_cgroup_queue);
1451
1452 r = unit_realize_cgroup_now(i, state);
1453 if (r < 0)
1454 log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id);
1455
1456 n++;
1457 }
1458
1459 return n;
1460 }
1461
1462 static void unit_queue_siblings(Unit *u) {
1463 Unit *slice;
1464
1465 /* This adds the siblings of the specified unit and the
1466 * siblings of all parent units to the cgroup queue. (But
1467 * neither the specified unit itself nor the parents.) */
1468
1469 while ((slice = UNIT_DEREF(u->slice))) {
1470 Iterator i;
1471 Unit *m;
1472
1473 SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) {
1474 if (m == u)
1475 continue;
1476
1477 /* Skip units that have a dependency on the slice
1478 * but aren't actually in it. */
1479 if (UNIT_DEREF(m->slice) != slice)
1480 continue;
1481
1482 /* No point in doing cgroup application for units
1483 * without active processes. */
1484 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
1485 continue;
1486
1487 /* If the unit doesn't need any new controllers
1488 * and has current ones realized, it doesn't need
1489 * any changes. */
1490 if (unit_has_mask_realized(m, unit_get_target_mask(m), unit_get_enable_mask(m)))
1491 continue;
1492
1493 unit_add_to_cgroup_queue(m);
1494 }
1495
1496 u = slice;
1497 }
1498 }
1499
1500 int unit_realize_cgroup(Unit *u) {
1501 assert(u);
1502
1503 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1504 return 0;
1505
1506 /* So, here's the deal: when realizing the cgroups for this
1507 * unit, we need to first create all parents, but there's more
1508 * actually: for the weight-based controllers we also need to
1509 * make sure that all our siblings (i.e. units that are in the
1510 * same slice as we are) have cgroups, too. Otherwise, things
1511 * would become very uneven as each of their processes would
1512 * get as much resources as all our group together. This call
1513 * will synchronously create the parent cgroups, but will
1514 * defer work on the siblings to the next event loop
1515 * iteration. */
1516
1517 /* Add all sibling slices to the cgroup queue. */
1518 unit_queue_siblings(u);
1519
1520 /* And realize this one now (and apply the values) */
1521 return unit_realize_cgroup_now(u, manager_state(u->manager));
1522 }
1523
1524 void unit_release_cgroup(Unit *u) {
1525 assert(u);
1526
1527 /* Forgets all cgroup details for this cgroup */
1528
1529 if (u->cgroup_path) {
1530 (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
1531 u->cgroup_path = mfree(u->cgroup_path);
1532 }
1533
1534 if (u->cgroup_inotify_wd >= 0) {
1535 if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0)
1536 log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id);
1537
1538 (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd));
1539 u->cgroup_inotify_wd = -1;
1540 }
1541 }
1542
1543 void unit_prune_cgroup(Unit *u) {
1544 int r;
1545 bool is_root_slice;
1546
1547 assert(u);
1548
1549 /* Removes the cgroup, if empty and possible, and stops watching it. */
1550
1551 if (!u->cgroup_path)
1552 return;
1553
1554 (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */
1555
1556 is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE);
1557
1558 r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice);
1559 if (r < 0) {
1560 log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path);
1561 return;
1562 }
1563
1564 if (is_root_slice)
1565 return;
1566
1567 unit_release_cgroup(u);
1568
1569 u->cgroup_realized = false;
1570 u->cgroup_realized_mask = 0;
1571 u->cgroup_enabled_mask = 0;
1572 }
1573
1574 int unit_search_main_pid(Unit *u, pid_t *ret) {
1575 _cleanup_fclose_ FILE *f = NULL;
1576 pid_t pid = 0, npid, mypid;
1577 int r;
1578
1579 assert(u);
1580 assert(ret);
1581
1582 if (!u->cgroup_path)
1583 return -ENXIO;
1584
1585 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f);
1586 if (r < 0)
1587 return r;
1588
1589 mypid = getpid();
1590 while (cg_read_pid(f, &npid) > 0) {
1591 pid_t ppid;
1592
1593 if (npid == pid)
1594 continue;
1595
1596 /* Ignore processes that aren't our kids */
1597 if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid)
1598 continue;
1599
1600 if (pid != 0)
1601 /* Dang, there's more than one daemonized PID
1602 in this group, so we don't know what process
1603 is the main process. */
1604
1605 return -ENODATA;
1606
1607 pid = npid;
1608 }
1609
1610 *ret = pid;
1611 return 0;
1612 }
1613
1614 static int unit_watch_pids_in_path(Unit *u, const char *path) {
1615 _cleanup_closedir_ DIR *d = NULL;
1616 _cleanup_fclose_ FILE *f = NULL;
1617 int ret = 0, r;
1618
1619 assert(u);
1620 assert(path);
1621
1622 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
1623 if (r < 0)
1624 ret = r;
1625 else {
1626 pid_t pid;
1627
1628 while ((r = cg_read_pid(f, &pid)) > 0) {
1629 r = unit_watch_pid(u, pid);
1630 if (r < 0 && ret >= 0)
1631 ret = r;
1632 }
1633
1634 if (r < 0 && ret >= 0)
1635 ret = r;
1636 }
1637
1638 r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
1639 if (r < 0) {
1640 if (ret >= 0)
1641 ret = r;
1642 } else {
1643 char *fn;
1644
1645 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1646 _cleanup_free_ char *p = NULL;
1647
1648 p = strjoin(path, "/", fn, NULL);
1649 free(fn);
1650
1651 if (!p)
1652 return -ENOMEM;
1653
1654 r = unit_watch_pids_in_path(u, p);
1655 if (r < 0 && ret >= 0)
1656 ret = r;
1657 }
1658
1659 if (r < 0 && ret >= 0)
1660 ret = r;
1661 }
1662
1663 return ret;
1664 }
1665
1666 int unit_watch_all_pids(Unit *u) {
1667 assert(u);
1668
1669 /* Adds all PIDs from our cgroup to the set of PIDs we
1670 * watch. This is a fallback logic for cases where we do not
1671 * get reliable cgroup empty notifications: we try to use
1672 * SIGCHLD as replacement. */
1673
1674 if (!u->cgroup_path)
1675 return -ENOENT;
1676
1677 if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* On unified we can use proper notifications */
1678 return 0;
1679
1680 return unit_watch_pids_in_path(u, u->cgroup_path);
1681 }
1682
1683 int unit_notify_cgroup_empty(Unit *u) {
1684 int r;
1685
1686 assert(u);
1687
1688 if (!u->cgroup_path)
1689 return 0;
1690
1691 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
1692 if (r <= 0)
1693 return r;
1694
1695 unit_add_to_gc_queue(u);
1696
1697 if (UNIT_VTABLE(u)->notify_cgroup_empty)
1698 UNIT_VTABLE(u)->notify_cgroup_empty(u);
1699
1700 return 0;
1701 }
1702
1703 static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1704 Manager *m = userdata;
1705
1706 assert(s);
1707 assert(fd >= 0);
1708 assert(m);
1709
1710 for (;;) {
1711 union inotify_event_buffer buffer;
1712 struct inotify_event *e;
1713 ssize_t l;
1714
1715 l = read(fd, &buffer, sizeof(buffer));
1716 if (l < 0) {
1717 if (errno == EINTR || errno == EAGAIN)
1718 return 0;
1719
1720 return log_error_errno(errno, "Failed to read control group inotify events: %m");
1721 }
1722
1723 FOREACH_INOTIFY_EVENT(e, buffer, l) {
1724 Unit *u;
1725
1726 if (e->wd < 0)
1727 /* Queue overflow has no watch descriptor */
1728 continue;
1729
1730 if (e->mask & IN_IGNORED)
1731 /* The watch was just removed */
1732 continue;
1733
1734 u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd));
1735 if (!u) /* Not that inotify might deliver
1736 * events for a watch even after it
1737 * was removed, because it was queued
1738 * before the removal. Let's ignore
1739 * this here safely. */
1740 continue;
1741
1742 (void) unit_notify_cgroup_empty(u);
1743 }
1744 }
1745 }
1746
1747 int manager_setup_cgroup(Manager *m) {
1748 _cleanup_free_ char *path = NULL;
1749 CGroupController c;
1750 int r, all_unified, systemd_unified;
1751 char *e;
1752
1753 assert(m);
1754
1755 /* 1. Determine hierarchy */
1756 m->cgroup_root = mfree(m->cgroup_root);
1757 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
1758 if (r < 0)
1759 return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
1760
1761 /* Chop off the init scope, if we are already located in it */
1762 e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1763
1764 /* LEGACY: Also chop off the system slice if we are in
1765 * it. This is to support live upgrades from older systemd
1766 * versions where PID 1 was moved there. Also see
1767 * cg_get_root_path(). */
1768 if (!e && MANAGER_IS_SYSTEM(m)) {
1769 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
1770 if (!e)
1771 e = endswith(m->cgroup_root, "/system"); /* even more legacy */
1772 }
1773 if (e)
1774 *e = 0;
1775
1776 /* And make sure to store away the root value without trailing
1777 * slash, even for the root dir, so that we can easily prepend
1778 * it everywhere. */
1779 while ((e = endswith(m->cgroup_root, "/")))
1780 *e = 0;
1781
1782 /* 2. Show data */
1783 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
1784 if (r < 0)
1785 return log_error_errno(r, "Cannot find cgroup mount point: %m");
1786
1787 all_unified = cg_all_unified();
1788 systemd_unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER);
1789
1790 if (all_unified < 0 || systemd_unified < 0)
1791 return log_error_errno(all_unified < 0 ? all_unified : systemd_unified,
1792 "Couldn't determine if we are running in the unified hierarchy: %m");
1793
1794 if (all_unified > 0)
1795 log_debug("Unified cgroup hierarchy is located at %s.", path);
1796 else if (systemd_unified > 0)
1797 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path);
1798 else
1799 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
1800
1801 if (!m->test_run) {
1802 const char *scope_path;
1803
1804 /* 3. Install agent */
1805 if (systemd_unified) {
1806
1807 /* In the unified hierarchy we can get
1808 * cgroup empty notifications via inotify. */
1809
1810 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1811 safe_close(m->cgroup_inotify_fd);
1812
1813 m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1814 if (m->cgroup_inotify_fd < 0)
1815 return log_error_errno(errno, "Failed to create control group inotify object: %m");
1816
1817 r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m);
1818 if (r < 0)
1819 return log_error_errno(r, "Failed to watch control group inotify object: %m");
1820
1821 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
1822 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
1823 r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-5);
1824 if (r < 0)
1825 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
1826
1827 (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify");
1828
1829 } else if (MANAGER_IS_SYSTEM(m)) {
1830
1831 /* On the legacy hierarchy we only get
1832 * notifications via cgroup agents. (Which
1833 * isn't really reliable, since it does not
1834 * generate events when control groups with
1835 * children run empty. */
1836
1837 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
1838 if (r < 0)
1839 log_warning_errno(r, "Failed to install release agent, ignoring: %m");
1840 else if (r > 0)
1841 log_debug("Installed release agent.");
1842 else if (r == 0)
1843 log_debug("Release agent already installed.");
1844 }
1845
1846 /* 4. Make sure we are in the special "init.scope" unit in the root slice. */
1847 scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1848 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1849 if (r < 0)
1850 return log_error_errno(r, "Failed to create %s control group: %m", scope_path);
1851
1852 /* also, move all other userspace processes remaining
1853 * in the root cgroup into that scope. */
1854 r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1855 if (r < 0)
1856 log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m");
1857
1858 /* 5. And pin it, so that it cannot be unmounted */
1859 safe_close(m->pin_cgroupfs_fd);
1860 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
1861 if (m->pin_cgroupfs_fd < 0)
1862 return log_error_errno(errno, "Failed to open pin file: %m");
1863
1864 /* 6. Always enable hierarchical support if it exists... */
1865 if (!all_unified)
1866 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
1867 }
1868
1869 /* 7. Figure out which controllers are supported */
1870 r = cg_mask_supported(&m->cgroup_supported);
1871 if (r < 0)
1872 return log_error_errno(r, "Failed to determine supported controllers: %m");
1873
1874 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++)
1875 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c)));
1876
1877 return 0;
1878 }
1879
1880 void manager_shutdown_cgroup(Manager *m, bool delete) {
1881 assert(m);
1882
1883 /* We can't really delete the group, since we are in it. But
1884 * let's trim it. */
1885 if (delete && m->cgroup_root)
1886 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
1887
1888 m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit);
1889
1890 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1891 m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd);
1892
1893 m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
1894
1895 m->cgroup_root = mfree(m->cgroup_root);
1896 }
1897
1898 Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
1899 char *p;
1900 Unit *u;
1901
1902 assert(m);
1903 assert(cgroup);
1904
1905 u = hashmap_get(m->cgroup_unit, cgroup);
1906 if (u)
1907 return u;
1908
1909 p = strdupa(cgroup);
1910 for (;;) {
1911 char *e;
1912
1913 e = strrchr(p, '/');
1914 if (!e || e == p)
1915 return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE);
1916
1917 *e = 0;
1918
1919 u = hashmap_get(m->cgroup_unit, p);
1920 if (u)
1921 return u;
1922 }
1923 }
1924
1925 Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) {
1926 _cleanup_free_ char *cgroup = NULL;
1927 int r;
1928
1929 assert(m);
1930
1931 if (pid <= 0)
1932 return NULL;
1933
1934 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
1935 if (r < 0)
1936 return NULL;
1937
1938 return manager_get_unit_by_cgroup(m, cgroup);
1939 }
1940
1941 Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
1942 Unit *u;
1943
1944 assert(m);
1945
1946 if (pid <= 0)
1947 return NULL;
1948
1949 if (pid == 1)
1950 return hashmap_get(m->units, SPECIAL_INIT_SCOPE);
1951
1952 u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid));
1953 if (u)
1954 return u;
1955
1956 u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid));
1957 if (u)
1958 return u;
1959
1960 return manager_get_unit_by_pid_cgroup(m, pid);
1961 }
1962
1963 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
1964 Unit *u;
1965
1966 assert(m);
1967 assert(cgroup);
1968
1969 log_debug("Got cgroup empty notification for: %s", cgroup);
1970
1971 u = manager_get_unit_by_cgroup(m, cgroup);
1972 if (!u)
1973 return 0;
1974
1975 return unit_notify_cgroup_empty(u);
1976 }
1977
1978 int unit_get_memory_current(Unit *u, uint64_t *ret) {
1979 _cleanup_free_ char *v = NULL;
1980 int r;
1981
1982 assert(u);
1983 assert(ret);
1984
1985 if (!u->cgroup_path)
1986 return -ENODATA;
1987
1988 if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
1989 return -ENODATA;
1990
1991 if (cg_all_unified() <= 0)
1992 r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v);
1993 else
1994 r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v);
1995 if (r == -ENOENT)
1996 return -ENODATA;
1997 if (r < 0)
1998 return r;
1999
2000 return safe_atou64(v, ret);
2001 }
2002
2003 int unit_get_tasks_current(Unit *u, uint64_t *ret) {
2004 _cleanup_free_ char *v = NULL;
2005 int r;
2006
2007 assert(u);
2008 assert(ret);
2009
2010 if (!u->cgroup_path)
2011 return -ENODATA;
2012
2013 if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0)
2014 return -ENODATA;
2015
2016 r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v);
2017 if (r == -ENOENT)
2018 return -ENODATA;
2019 if (r < 0)
2020 return r;
2021
2022 return safe_atou64(v, ret);
2023 }
2024
2025 static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) {
2026 _cleanup_free_ char *v = NULL;
2027 uint64_t ns;
2028 int r;
2029
2030 assert(u);
2031 assert(ret);
2032
2033 if (!u->cgroup_path)
2034 return -ENODATA;
2035
2036 if (cg_all_unified() > 0) {
2037 const char *keys[] = { "usage_usec", NULL };
2038 _cleanup_free_ char *val = NULL;
2039 uint64_t us;
2040
2041 if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0)
2042 return -ENODATA;
2043
2044 r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val);
2045 if (r < 0)
2046 return r;
2047
2048 r = safe_atou64(val, &us);
2049 if (r < 0)
2050 return r;
2051
2052 ns = us * NSEC_PER_USEC;
2053 } else {
2054 if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0)
2055 return -ENODATA;
2056
2057 r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v);
2058 if (r == -ENOENT)
2059 return -ENODATA;
2060 if (r < 0)
2061 return r;
2062
2063 r = safe_atou64(v, &ns);
2064 if (r < 0)
2065 return r;
2066 }
2067
2068 *ret = ns;
2069 return 0;
2070 }
2071
2072 int unit_get_cpu_usage(Unit *u, nsec_t *ret) {
2073 nsec_t ns;
2074 int r;
2075
2076 assert(u);
2077
2078 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
2079 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
2080 * call this function with a NULL return value. */
2081
2082 r = unit_get_cpu_usage_raw(u, &ns);
2083 if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) {
2084 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
2085 * cached value. */
2086
2087 if (ret)
2088 *ret = u->cpu_usage_last;
2089 return 0;
2090 }
2091 if (r < 0)
2092 return r;
2093
2094 if (ns > u->cpu_usage_base)
2095 ns -= u->cpu_usage_base;
2096 else
2097 ns = 0;
2098
2099 u->cpu_usage_last = ns;
2100 if (ret)
2101 *ret = ns;
2102
2103 return 0;
2104 }
2105
2106 int unit_reset_cpu_usage(Unit *u) {
2107 nsec_t ns;
2108 int r;
2109
2110 assert(u);
2111
2112 u->cpu_usage_last = NSEC_INFINITY;
2113
2114 r = unit_get_cpu_usage_raw(u, &ns);
2115 if (r < 0) {
2116 u->cpu_usage_base = 0;
2117 return r;
2118 }
2119
2120 u->cpu_usage_base = ns;
2121 return 0;
2122 }
2123
2124 bool unit_cgroup_delegate(Unit *u) {
2125 CGroupContext *c;
2126
2127 assert(u);
2128
2129 c = unit_get_cgroup_context(u);
2130 if (!c)
2131 return false;
2132
2133 return c->delegate;
2134 }
2135
2136 void unit_invalidate_cgroup(Unit *u, CGroupMask m) {
2137 assert(u);
2138
2139 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2140 return;
2141
2142 if (m == 0)
2143 return;
2144
2145 /* always invalidate compat pairs together */
2146 if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO))
2147 m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
2148
2149 if ((u->cgroup_realized_mask & m) == 0)
2150 return;
2151
2152 u->cgroup_realized_mask &= ~m;
2153 unit_add_to_cgroup_queue(u);
2154 }
2155
2156 void manager_invalidate_startup_units(Manager *m) {
2157 Iterator i;
2158 Unit *u;
2159
2160 assert(m);
2161
2162 SET_FOREACH(u, m->startup_units, i)
2163 unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
2164 }
2165
2166 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
2167 [CGROUP_AUTO] = "auto",
2168 [CGROUP_CLOSED] = "closed",
2169 [CGROUP_STRICT] = "strict",
2170 };
2171
2172 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);
Unnamed repository; edit this file 'description' to name the repository.