]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/core/cgroup.c
projects / thirdparty / systemd.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
util: use SPECIAL_ROOT_SLICE macro where appropriate
[thirdparty/systemd.git] / src / core / cgroup.c
CommitLineData
8e274523
LP		 1/***
2 This file is part of systemd.
3
4ad49000 4 Copyright 2013 Lennart Poettering
8e274523
LP		 5
6 systemd is free software; you can redistribute it and/or modify it
5430f7f2
LP		 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
8e274523
LP		 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
5430f7f2 14 Lesser General Public License for more details.
8e274523 15
5430f7f2 16 You should have received a copy of the GNU Lesser General Public License
8e274523
LP		 17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18***/
19
c6c18be3 20#include <fcntl.h>
e41969e3 21#include <fnmatch.h>
8c6db833 22
b5efdb8a 23#include "alloc-util.h"
03a7b521 24#include "cgroup-util.h"
3ffd4af2
LP		 25#include "cgroup.h"
26#include "fd-util.h"
0d39fa9c 27#include "fileio.h"
77601719 28#include "fs-util.h"
6bedfcbb 29#include "parse-util.h"
9eb977db 30#include "path-util.h"
03a7b521 31#include "process-util.h"
9444b1f2 32#include "special.h"
8b43440b 33#include "string-table.h"
07630cea 34#include "string-util.h"
13c31542 35#include "stdio-util.h"
8e274523 36
9a054909
LP		 37#define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
38
2b40998d 39static void cgroup_compat_warn(void) {
128fadc9
TH		 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__); \
2b40998d 52 } while (false)
128fadc9 53
4ad49000
LP		 54void 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
66ebf6c0
TH		 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
d53d9474
LP		 64 c->cpu_shares = CGROUP_CPU_SHARES_INVALID;
65 c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID;
d53d9474 66
da4d897e
TH		 67 c->memory_high = CGROUP_LIMIT_MAX;
68 c->memory_max = CGROUP_LIMIT_MAX;
96e131ea 69 c->memory_swap_max = CGROUP_LIMIT_MAX;
da4d897e
TH		 70
71 c->memory_limit = CGROUP_LIMIT_MAX;
b2f8b02e 72
13c31542
TH		 73 c->io_weight = CGROUP_WEIGHT_INVALID;
74 c->startup_io_weight = CGROUP_WEIGHT_INVALID;
75
d53d9474
LP		 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;
4ad49000 80}
8e274523 81
4ad49000
LP		 82void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
83 assert(c);
84 assert(a);
85
71fda00f 86 LIST_REMOVE(device_allow, c->device_allow, a);
4ad49000
LP		 87 free(a->path);
88 free(a);
89}
90
13c31542
TH		 91void 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
100void 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
4ad49000
LP		 109void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
110 assert(c);
111 assert(w);
112
71fda00f 113 LIST_REMOVE(device_weights, c->blockio_device_weights, w);
4ad49000
LP		 114 free(w->path);
115 free(w);
116}
117
118void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
119 assert(c);
8e274523 120 assert(b);
8e274523 121
71fda00f 122 LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
4ad49000
LP		 123 free(b->path);
124 free(b);
125}
126
127void cgroup_context_done(CGroupContext *c) {
128 assert(c);
129
13c31542
TH		 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
4ad49000
LP		 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
146void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
13c31542
TH		 147 CGroupIODeviceLimit *il;
148 CGroupIODeviceWeight *iw;
4ad49000
LP		 149 CGroupBlockIODeviceBandwidth *b;
150 CGroupBlockIODeviceWeight *w;
151 CGroupDeviceAllow *a;
9a054909 152 char u[FORMAT_TIMESPAN_MAX];
4ad49000
LP		 153
154 assert(c);
155 assert(f);
156
157 prefix = strempty(prefix);
158
159 fprintf(f,
160 "%sCPUAccounting=%s\n"
13c31542 161 "%sIOAccounting=%s\n"
4ad49000
LP		 162 "%sBlockIOAccounting=%s\n"
163 "%sMemoryAccounting=%s\n"
d53d9474 164 "%sTasksAccounting=%s\n"
66ebf6c0
TH		 165 "%sCPUWeight=%" PRIu64 "\n"
166 "%sStartupCPUWeight=%" PRIu64 "\n"
d53d9474
LP		 167 "%sCPUShares=%" PRIu64 "\n"
168 "%sStartupCPUShares=%" PRIu64 "\n"
b2f8b02e 169 "%sCPUQuotaPerSecSec=%s\n"
13c31542
TH		 170 "%sIOWeight=%" PRIu64 "\n"
171 "%sStartupIOWeight=%" PRIu64 "\n"
d53d9474
LP		 172 "%sBlockIOWeight=%" PRIu64 "\n"
173 "%sStartupBlockIOWeight=%" PRIu64 "\n"
da4d897e
TH		 174 "%sMemoryLow=%" PRIu64 "\n"
175 "%sMemoryHigh=%" PRIu64 "\n"
176 "%sMemoryMax=%" PRIu64 "\n"
96e131ea 177 "%sMemorySwapMax=%" PRIu64 "\n"
4ad49000 178 "%sMemoryLimit=%" PRIu64 "\n"
03a7b521 179 "%sTasksMax=%" PRIu64 "\n"
a931ad47
LP		 180 "%sDevicePolicy=%s\n"
181 "%sDelegate=%s\n",
4ad49000 182 prefix, yes_no(c->cpu_accounting),
13c31542 183 prefix, yes_no(c->io_accounting),
4ad49000
LP		 184 prefix, yes_no(c->blockio_accounting),
185 prefix, yes_no(c->memory_accounting),
d53d9474 186 prefix, yes_no(c->tasks_accounting),
66ebf6c0
TH		 187 prefix, c->cpu_weight,
188 prefix, c->startup_cpu_weight,
4ad49000 189 prefix, c->cpu_shares,
95ae05c0 190 prefix, c->startup_cpu_shares,
b1d6dcf5 191 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
13c31542
TH		 192 prefix, c->io_weight,
193 prefix, c->startup_io_weight,
4ad49000 194 prefix, c->blockio_weight,
95ae05c0 195 prefix, c->startup_blockio_weight,
da4d897e
TH		 196 prefix, c->memory_low,
197 prefix, c->memory_high,
198 prefix, c->memory_max,
96e131ea 199 prefix, c->memory_swap_max,
4ad49000 200 prefix, c->memory_limit,
03a7b521 201 prefix, c->tasks_max,
a931ad47
LP		 202 prefix, cgroup_device_policy_to_string(c->device_policy),
203 prefix, yes_no(c->delegate));
4ad49000
LP		 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
13c31542
TH		 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];
9be57249
TH		 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]));
13c31542
TH		 231 }
232
4ad49000
LP		 233 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
234 fprintf(f,
d53d9474 235 "%sBlockIODeviceWeight=%s %" PRIu64,
4ad49000
LP		 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
979d0311
TH		 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));
4ad49000
LP		 255 }
256}
257
13c31542 258static int lookup_block_device(const char *p, dev_t *dev) {
4ad49000
LP		 259 struct stat st;
260 int r;
261
262 assert(p);
263 assert(dev);
264
265 r = stat(p, &st);
4a62c710
MS		 266 if (r < 0)
267 return log_warning_errno(errno, "Couldn't stat device %s: %m", p);
8e274523 268
4ad49000
LP		 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 }
8e274523 283
8e274523 284 return 0;
8e274523
LP		 285}
286
4ad49000
LP		 287static 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;
8c6db833 290 int r;
8e274523 291
4ad49000
LP		 292 assert(path);
293 assert(acc);
8e274523 294
4ad49000
LP		 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);
1aeab12b 312 if (r < 0)
077ba06e 313 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
714e2e1d 314 "Failed to set devices.allow on %s: %m", path);
4ad49000
LP		 315
316 return r;
8e274523
LP		 317}
318
90060676
LP		 319static 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");
4a62c710
MS		 330 if (!f)
331 return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
90060676
LP		 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);
e41969e3
LP		 372
373 if (fnmatch(name, w, 0) != 0)
90060676
LP		 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);
1aeab12b 383 if (r < 0)
077ba06e 384 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
714e2e1d 385 "Failed to set devices.allow on %s: %m", path);
90060676
LP		 386 }
387
388 return 0;
389
390fail:
56f64d95 391 log_warning_errno(errno, "Failed to read /proc/devices: %m");
90060676
LP		 392 return -errno;
393}
394
66ebf6c0
TH		 395static 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
400static 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
405static 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
415static 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
425static 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
448static 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
474static 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
479static 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
508c45da 484static bool cgroup_context_has_io_config(CGroupContext *c) {
538b4852
TH		 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
508c45da 492static bool cgroup_context_has_blockio_config(CGroupContext *c) {
538b4852
TH		 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
508c45da 500static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) {
64faf04c
TH		 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
508c45da 510static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) {
64faf04c
TH		 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
508c45da 520static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) {
538b4852
TH		 521 return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT,
522 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
523}
524
508c45da 525static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) {
538b4852
TH		 526 return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT,
527 CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX);
528}
529
f29ff115 530static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) {
64faf04c
TH		 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);
f29ff115 540 r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf);
64faf04c 541 if (r < 0)
f29ff115
TH		 542 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
543 "Failed to set io.weight: %m");
64faf04c
TH		 544}
545
f29ff115 546static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) {
64faf04c
TH		 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);
f29ff115 556 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf);
64faf04c 557 if (r < 0)
f29ff115
TH		 558 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
559 "Failed to set blkio.weight_device: %m");
64faf04c
TH		 560}
561
f29ff115 562static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) {
64faf04c
TH		 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]);
f29ff115 586 r = cg_set_attribute("io", u->cgroup_path, "io.max", buf);
64faf04c 587 if (r < 0)
f29ff115
TH		 588 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
589 "Failed to set io.max: %m");
64faf04c
TH		 590 return n;
591}
592
f29ff115 593static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) {
64faf04c
TH		 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);
f29ff115 606 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf);
64faf04c 607 if (r < 0)
f29ff115
TH		 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");
64faf04c
TH		 610
611 if (wbps != CGROUP_LIMIT_MAX)
612 n++;
613 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps);
f29ff115 614 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf);
64faf04c 615 if (r < 0)
f29ff115
TH		 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");
64faf04c
TH		 618
619 return n;
620}
621
da4d897e 622static bool cgroup_context_has_unified_memory_config(CGroupContext *c) {
96e131ea 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;
da4d897e
TH		 624}
625
f29ff115 626static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) {
da4d897e
TH		 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
f29ff115 633 r = cg_set_attribute("memory", u->cgroup_path, file, buf);
da4d897e 634 if (r < 0)
f29ff115
TH		 635 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
636 "Failed to set %s: %m", file);
da4d897e
TH		 637}
638
f29ff115
TH		 639static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) {
640 const char *path;
641 CGroupContext *c;
01efdf13 642 bool is_root;
4ad49000
LP		 643 int r;
644
f29ff115
TH		 645 assert(u);
646
647 c = unit_get_cgroup_context(u);
648 path = u->cgroup_path;
649
4ad49000
LP		 650 assert(c);
651 assert(path);
8e274523 652
4ad49000
LP		 653 if (mask == 0)
654 return;
8e274523 655
71c26873 656 /* Some cgroup attributes are not supported on the root cgroup,
01efdf13
LP		 657 * hence silently ignore */
658 is_root = isempty(path) || path_equal(path, "/");
6da13913
ZJS		 659 if (is_root)
660 /* Make sure we don't try to display messages with an empty path. */
661 path = "/";
01efdf13 662
714e2e1d
LP		 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
efdb0237 667 if ((mask & CGROUP_MASK_CPU) && !is_root) {
66ebf6c0
TH		 668 bool has_weight = cgroup_context_has_cpu_weight(c);
669 bool has_shares = cgroup_context_has_cpu_shares(c);
8e274523 670
ca2f6384 671 if (cg_all_unified() > 0) {
66ebf6c0 672 uint64_t weight;
b2f8b02e 673
66ebf6c0
TH		 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);
b2f8b02e 678
66ebf6c0
TH		 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 }
4ad49000
LP		 704 }
705
13c31542 706 if (mask & CGROUP_MASK_IO) {
538b4852
TH		 707 bool has_io = cgroup_context_has_io_config(c);
708 bool has_blockio = cgroup_context_has_blockio_config(c);
13c31542
TH		 709
710 if (!is_root) {
64faf04c
TH		 711 char buf[8+DECIMAL_STR_MAX(uint64_t)+1];
712 uint64_t weight;
13c31542 713
538b4852
TH		 714 if (has_io)
715 weight = cgroup_context_io_weight(c, state);
128fadc9
TH		 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
538b4852 724 weight = CGROUP_WEIGHT_DEFAULT;
13c31542
TH		 725
726 xsprintf(buf, "default %" PRIu64 "\n", weight);
727 r = cg_set_attribute("io", path, "io.weight", buf);
728 if (r < 0)
f29ff115
TH		 729 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
730 "Failed to set io.weight: %m");
13c31542 731
538b4852
TH		 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)
f29ff115 737 cgroup_apply_io_device_weight(u, w->path, w->weight);
538b4852
TH		 738 } else if (has_blockio) {
739 CGroupBlockIODeviceWeight *w;
740
741 /* FIXME: no way to reset this list */
128fadc9
TH		 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 }
538b4852 750 }
13c31542
TH		 751 }
752
64faf04c 753 /* Apply limits and free ones without config. */
538b4852
TH		 754 if (has_io) {
755 CGroupIODeviceLimit *l, *next;
756
757 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
f29ff115 758 if (!cgroup_apply_io_device_limit(u, l->path, l->limits))
538b4852
TH		 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
128fadc9
TH		 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
f29ff115 777 if (!cgroup_apply_io_device_limit(u, b->path, limits))
538b4852
TH		 778 cgroup_context_free_blockio_device_bandwidth(c, b);
779 }
13c31542
TH		 780 }
781 }
782
efdb0237 783 if (mask & CGROUP_MASK_BLKIO) {
538b4852
TH		 784 bool has_io = cgroup_context_has_io_config(c);
785 bool has_blockio = cgroup_context_has_blockio_config(c);
4ad49000 786
01efdf13 787 if (!is_root) {
64faf04c
TH		 788 char buf[DECIMAL_STR_MAX(uint64_t)+1];
789 uint64_t weight;
64faf04c 790
538b4852
TH		 791 if (has_blockio)
792 weight = cgroup_context_blkio_weight(c, state);
128fadc9
TH		 793 else if (has_io) {
794 uint64_t io_weight = cgroup_context_io_weight(c, state);
795
538b4852 796 weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state));
128fadc9
TH		 797
798 log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64,
799 io_weight, weight);
800 } else
538b4852 801 weight = CGROUP_BLKIO_WEIGHT_DEFAULT;
64faf04c
TH		 802
803 xsprintf(buf, "%" PRIu64 "\n", weight);
01efdf13 804 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
1aeab12b 805 if (r < 0)
f29ff115
TH		 806 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
807 "Failed to set blkio.weight: %m");
4ad49000 808
538b4852
TH		 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)
f29ff115 814 cgroup_apply_blkio_device_weight(u, w->path, w->weight);
538b4852
TH		 815 } else if (has_io) {
816 CGroupIODeviceWeight *w;
817
818 /* FIXME: no way to reset this list */
128fadc9
TH		 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 }
538b4852 827 }
4ad49000
LP		 828 }
829
64faf04c 830 /* Apply limits and free ones without config. */
538b4852
TH		 831 if (has_blockio) {
832 CGroupBlockIODeviceBandwidth *b, *next;
833
834 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
f29ff115 835 if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps))
538b4852
TH		 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) {
128fadc9
TH		 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
f29ff115 845 if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]))
538b4852
TH		 846 cgroup_context_free_io_device_limit(c, l);
847 }
d686d8a9 848 }
8e274523
LP		 849 }
850
efdb0237 851 if ((mask & CGROUP_MASK_MEMORY) && !is_root) {
ca2f6384 852 if (cg_all_unified() > 0) {
da4d897e 853 uint64_t max = c->memory_max;
96e131ea 854 uint64_t swap_max = c->memory_swap_max;
efdb0237 855
96e131ea 856 if (cgroup_context_has_unified_memory_config(c)) {
da4d897e 857 max = c->memory_max;
96e131ea
WC		 858 swap_max = c->memory_swap_max;
859 } else {
da4d897e 860 max = c->memory_limit;
efdb0237 861
128fadc9
TH		 862 if (max != CGROUP_LIMIT_MAX)
863 log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max);
864 }
865
f29ff115
TH		 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);
96e131ea 869 cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max);
efdb0237 870 } else {
da4d897e 871 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
78a4ee59 872 uint64_t val = c->memory_limit;
da4d897e 873
78a4ee59
DM		 874 if (val == CGROUP_LIMIT_MAX) {
875 val = c->memory_max;
8e274523 876
78a4ee59
DM		 877 if (val != CGROUP_LIMIT_MAX)
878 log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", c->memory_max);
128fadc9
TH		 879 }
880
78a4ee59
DM		 881 if (val == CGROUP_LIMIT_MAX)
882 strncpy(buf, "-1\n", sizeof(buf));
883 else
884 xsprintf(buf, "%" PRIu64 "\n", val);
885
da4d897e
TH		 886 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
887 if (r < 0)
f29ff115
TH		 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");
da4d897e 890 }
4ad49000 891 }
8e274523 892
3905f127 893 if ((mask & CGROUP_MASK_DEVICES) && !is_root) {
4ad49000 894 CGroupDeviceAllow *a;
8e274523 895
714e2e1d
LP		 896 /* Changing the devices list of a populated cgroup
897 * might result in EINVAL, hence ignore EINVAL
898 * here. */
899
4ad49000
LP		 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");
1aeab12b 904 if (r < 0)
f29ff115
TH		 905 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
906 "Failed to reset devices.list: %m");
fb385181 907
4ad49000
LP		 908 if (c->device_policy == CGROUP_CLOSED ||
909 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
910 static const char auto_devices[] =
7d711efb
LP		 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"
0d9e7991
AP		 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";
4ad49000
LP		 921
922 const char *x, *y;
923
924 NULSTR_FOREACH_PAIR(x, y, auto_devices)
925 whitelist_device(path, x, y);
7d711efb
LP		 926
927 whitelist_major(path, "pts", 'c', "rw");
928 whitelist_major(path, "kdbus", 'c', "rw");
929 whitelist_major(path, "kdbus/*", 'c', "rw");
4ad49000
LP		 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';
fb385181 942
4ad49000
LP		 943 if (k == 0)
944 continue;
fb385181 945
4ad49000 946 acc[k++] = 0;
90060676
LP		 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
f29ff115 955 log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path);
4ad49000
LP		 956 }
957 }
03a7b521
LP		 958
959 if ((mask & CGROUP_MASK_PIDS) && !is_root) {
960
f5058264 961 if (c->tasks_max != CGROUP_LIMIT_MAX) {
03a7b521
LP		 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)
f29ff115
TH		 970 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
971 "Failed to set pids.max: %m");
03a7b521 972 }
fb385181
LP		 973}
974
efdb0237
LP		 975CGroupMask cgroup_context_get_mask(CGroupContext *c) {
976 CGroupMask mask = 0;
8e274523 977
4ad49000 978 /* Figure out which controllers we need */
8e274523 979
b2f8b02e 980 if (c->cpu_accounting ||
66ebf6c0
TH		 981 cgroup_context_has_cpu_weight(c) ||
982 cgroup_context_has_cpu_shares(c) ||
3a43da28 983 c->cpu_quota_per_sec_usec != USEC_INFINITY)
efdb0237 984 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
ecedd90f 985
538b4852
TH		 986 if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c))
987 mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
ecedd90f 988
4ad49000 989 if (c->memory_accounting ||
da4d897e
TH		 990 c->memory_limit != CGROUP_LIMIT_MAX ||
991 cgroup_context_has_unified_memory_config(c))
efdb0237 992 mask |= CGROUP_MASK_MEMORY;
8e274523 993
a931ad47
LP		 994 if (c->device_allow ||
995 c->device_policy != CGROUP_AUTO)
3905f127 996 mask |= CGROUP_MASK_DEVICES;
4ad49000 997
03a7b521
LP		 998 if (c->tasks_accounting ||
999 c->tasks_max != (uint64_t) -1)
1000 mask |= CGROUP_MASK_PIDS;
1001
4ad49000 1002 return mask;
8e274523
LP		 1003}
1004
efdb0237 1005CGroupMask unit_get_own_mask(Unit *u) {
4ad49000 1006 CGroupContext *c;
8e274523 1007
efdb0237
LP		 1008 /* Returns the mask of controllers the unit needs for itself */
1009
4ad49000
LP		 1010 c = unit_get_cgroup_context(u);
1011 if (!c)
1012 return 0;
8e274523 1013
a931ad47 1014 /* If delegation is turned on, then turn on all cgroups,
19af675e
LP		 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. */
a931ad47
LP		 1021
1022 if (c->delegate) {
1023 ExecContext *e;
1024
1025 e = unit_get_exec_context(u);
19af675e
LP		 1026 if (!e ||
1027 exec_context_maintains_privileges(e) ||
ca2f6384 1028 cg_all_unified() > 0)
efdb0237 1029 return _CGROUP_MASK_ALL;
a931ad47
LP		 1030 }
1031
db785129 1032 return cgroup_context_get_mask(c);
8e274523
LP		 1033}
1034
efdb0237 1035CGroupMask unit_get_members_mask(Unit *u) {
4ad49000 1036 assert(u);
bc432dc7 1037
efdb0237
LP		 1038 /* Returns the mask of controllers all of the unit's children
1039 * require, merged */
1040
bc432dc7
LP		 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
d4fdc205 1055 if (UNIT_DEREF(member->slice) != u)
bc432dc7
LP		 1056 continue;
1057
1058 u->cgroup_members_mask |=
efdb0237 1059 unit_get_own_mask(member) |
bc432dc7
LP		 1060 unit_get_members_mask(member);
1061 }
1062 }
1063
1064 u->cgroup_members_mask_valid = true;
6414b7c9 1065 return u->cgroup_members_mask;
246aa6dd
LP		 1066}
1067
efdb0237 1068CGroupMask unit_get_siblings_mask(Unit *u) {
4ad49000 1069 assert(u);
246aa6dd 1070
efdb0237
LP		 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
bc432dc7 1075 if (UNIT_ISSET(u->slice))
637f421e 1076 return unit_get_members_mask(UNIT_DEREF(u->slice));
4ad49000 1077
efdb0237 1078 return unit_get_own_mask(u) | unit_get_members_mask(u);
246aa6dd
LP		 1079}
1080
efdb0237
LP		 1081CGroupMask 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
1089CGroupMask 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. */
6414b7c9 1098
efdb0237
LP		 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
1105CGroupMask 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);
6414b7c9
DS		 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. */
bc432dc7 1121void unit_update_cgroup_members_masks(Unit *u) {
efdb0237 1122 CGroupMask m;
bc432dc7
LP		 1123 bool more;
1124
1125 assert(u);
1126
1127 /* Calculate subtree mask */
efdb0237 1128 m = unit_get_subtree_mask(u);
bc432dc7
LP		 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
6414b7c9
DS		 1143 if (UNIT_ISSET(u->slice)) {
1144 Unit *s = UNIT_DEREF(u->slice);
bc432dc7
LP		 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);
6414b7c9
DS		 1165 }
1166}
1167
efdb0237 1168static const char *migrate_callback(CGroupMask mask, void *userdata) {
03b90d4b
LP		 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
efdb0237
LP		 1186char *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
1211int 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
1241int unit_watch_cgroup(Unit *u) {
ab2c3861 1242 _cleanup_free_ char *events = NULL;
efdb0237
LP		 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 */
5da38d07 1254 r = cg_unified(SYSTEMD_CGROUP_CONTROLLER);
efdb0237 1255 if (r < 0)
595bfe7d 1256 return log_unit_error_errno(u, r, "Failed detect whether the unified hierarchy is used: %m");
efdb0237
LP		 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
ab2c3861 1268 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events);
efdb0237
LP		 1269 if (r < 0)
1270 return log_oom();
1271
ab2c3861 1272 u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY);
efdb0237
LP		 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
1290static int unit_create_cgroup(
1291 Unit *u,
1292 CGroupMask target_mask,
1293 CGroupMask enable_mask) {
1294
0cd385d3 1295 CGroupContext *c;
bc432dc7 1296 int r;
64747e2d 1297
4ad49000 1298 assert(u);
64747e2d 1299
0cd385d3
LP		 1300 c = unit_get_cgroup_context(u);
1301 if (!c)
1302 return 0;
1303
7b3fd631
LP		 1304 if (!u->cgroup_path) {
1305 _cleanup_free_ char *path = NULL;
64747e2d 1306
7b3fd631
LP		 1307 path = unit_default_cgroup_path(u);
1308 if (!path)
1309 return log_oom();
1310
efdb0237
LP		 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);
b58b8e11
HH		 1316 }
1317
03b90d4b 1318 /* First, create our own group */
efdb0237 1319 r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path);
23bbb0de 1320 if (r < 0)
efdb0237
LP		 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);
03b90d4b
LP		 1330
1331 /* Keep track that this is now realized */
4ad49000 1332 u->cgroup_realized = true;
efdb0237 1333 u->cgroup_realized_mask = target_mask;
ccf78df1 1334 u->cgroup_enabled_mask = enable_mask;
4ad49000 1335
0cd385d3
LP		 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)
efdb0237 1343 log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path);
0cd385d3 1344 }
03b90d4b 1345
64747e2d
LP		 1346 return 0;
1347}
1348
7b3fd631
LP		 1349int 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
ccf78df1 1364static bool unit_has_mask_realized(Unit *u, CGroupMask target_mask, CGroupMask enable_mask) {
bc432dc7
LP		 1365 assert(u);
1366
ccf78df1 1367 return u->cgroup_realized && u->cgroup_realized_mask == target_mask && u->cgroup_enabled_mask == enable_mask;
6414b7c9
DS		 1368}
1369
1370/* Check if necessary controllers and attributes for a unit are in place.
1371 *
1372 * If so, do nothing.
1373 * If not, create paths, move processes over, and set attributes.
1374 *
1375 * Returns 0 on success and < 0 on failure. */
db785129 1376static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
efdb0237 1377 CGroupMask target_mask, enable_mask;
6414b7c9 1378 int r;
64747e2d 1379
4ad49000 1380 assert(u);
64747e2d 1381
4ad49000 1382 if (u->in_cgroup_queue) {
71fda00f 1383 LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u);
4ad49000
LP		 1384 u->in_cgroup_queue = false;
1385 }
64747e2d 1386
efdb0237 1387 target_mask = unit_get_target_mask(u);
ccf78df1
TH		 1388 enable_mask = unit_get_enable_mask(u);
1389
1390 if (unit_has_mask_realized(u, target_mask, enable_mask))
0a1eb06d 1391 return 0;
64747e2d 1392
4ad49000 1393 /* First, realize parents */
6414b7c9 1394 if (UNIT_ISSET(u->slice)) {
db785129 1395 r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
6414b7c9
DS		 1396 if (r < 0)
1397 return r;
1398 }
4ad49000
LP		 1399
1400 /* And then do the real work */
efdb0237 1401 r = unit_create_cgroup(u, target_mask, enable_mask);
6414b7c9
DS		 1402 if (r < 0)
1403 return r;
1404
1405 /* Finally, apply the necessary attributes. */
f29ff115 1406 cgroup_context_apply(u, target_mask, state);
6414b7c9
DS		 1407
1408 return 0;
64747e2d
LP		 1409}
1410
4ad49000 1411static void unit_add_to_cgroup_queue(Unit *u) {
ecedd90f 1412
4ad49000
LP		 1413 if (u->in_cgroup_queue)
1414 return;
8e274523 1415
71fda00f 1416 LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u);
4ad49000
LP		 1417 u->in_cgroup_queue = true;
1418}
8c6db833 1419
4ad49000 1420unsigned manager_dispatch_cgroup_queue(Manager *m) {
db785129 1421 ManagerState state;
4ad49000 1422 unsigned n = 0;
db785129 1423 Unit *i;
6414b7c9 1424 int r;
ecedd90f 1425
db785129
LP		 1426 state = manager_state(m);
1427
4ad49000
LP		 1428 while ((i = m->cgroup_queue)) {
1429 assert(i->in_cgroup_queue);
ecedd90f 1430
db785129 1431 r = unit_realize_cgroup_now(i, state);
6414b7c9 1432 if (r < 0)
efdb0237 1433 log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id);
0a1eb06d 1434
4ad49000
LP		 1435 n++;
1436 }
ecedd90f 1437
4ad49000 1438 return n;
8e274523
LP		 1439}
1440
4ad49000
LP		 1441static void unit_queue_siblings(Unit *u) {
1442 Unit *slice;
ca949c9d 1443
4ad49000
LP		 1444 /* This adds the siblings of the specified unit and the
1445 * siblings of all parent units to the cgroup queue. (But
1446 * neither the specified unit itself nor the parents.) */
1447
1448 while ((slice = UNIT_DEREF(u->slice))) {
1449 Iterator i;
1450 Unit *m;
8f53a7b8 1451
4ad49000
LP		 1452 SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) {
1453 if (m == u)
1454 continue;
8e274523 1455
6414b7c9
DS		 1456 /* Skip units that have a dependency on the slice
1457 * but aren't actually in it. */
4ad49000 1458 if (UNIT_DEREF(m->slice) != slice)
50159e6a 1459 continue;
8e274523 1460
6414b7c9
DS		 1461 /* No point in doing cgroup application for units
1462 * without active processes. */
1463 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
1464 continue;
1465
1466 /* If the unit doesn't need any new controllers
1467 * and has current ones realized, it doesn't need
1468 * any changes. */
ccf78df1 1469 if (unit_has_mask_realized(m, unit_get_target_mask(m), unit_get_enable_mask(m)))
6414b7c9
DS		 1470 continue;
1471
4ad49000 1472 unit_add_to_cgroup_queue(m);
50159e6a
LP		 1473 }
1474
4ad49000 1475 u = slice;
8e274523 1476 }
4ad49000
LP		 1477}
1478
0a1eb06d 1479int unit_realize_cgroup(Unit *u) {
4ad49000
LP		 1480 assert(u);
1481
35b7ff80 1482 if (!UNIT_HAS_CGROUP_CONTEXT(u))
0a1eb06d 1483 return 0;
8e274523 1484
4ad49000
LP		 1485 /* So, here's the deal: when realizing the cgroups for this
1486 * unit, we need to first create all parents, but there's more
1487 * actually: for the weight-based controllers we also need to
1488 * make sure that all our siblings (i.e. units that are in the
73e231ab 1489 * same slice as we are) have cgroups, too. Otherwise, things
4ad49000
LP		 1490 * would become very uneven as each of their processes would
1491 * get as much resources as all our group together. This call
1492 * will synchronously create the parent cgroups, but will
1493 * defer work on the siblings to the next event loop
1494 * iteration. */
ca949c9d 1495
4ad49000
LP		 1496 /* Add all sibling slices to the cgroup queue. */
1497 unit_queue_siblings(u);
1498
6414b7c9 1499 /* And realize this one now (and apply the values) */
db785129 1500 return unit_realize_cgroup_now(u, manager_state(u->manager));
8e274523
LP		 1501}
1502
efdb0237
LP		 1503void unit_release_cgroup(Unit *u) {
1504 assert(u);
1505
1506 /* Forgets all cgroup details for this cgroup */
1507
1508 if (u->cgroup_path) {
1509 (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
1510 u->cgroup_path = mfree(u->cgroup_path);
1511 }
1512
1513 if (u->cgroup_inotify_wd >= 0) {
1514 if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0)
1515 log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id);
1516
1517 (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd));
1518 u->cgroup_inotify_wd = -1;
1519 }
1520}
1521
1522void unit_prune_cgroup(Unit *u) {
8e274523 1523 int r;
efdb0237 1524 bool is_root_slice;
8e274523 1525
4ad49000 1526 assert(u);
8e274523 1527
efdb0237
LP		 1528 /* Removes the cgroup, if empty and possible, and stops watching it. */
1529
4ad49000
LP		 1530 if (!u->cgroup_path)
1531 return;
8e274523 1532
fe700f46
LP		 1533 (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */
1534
efdb0237
LP		 1535 is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE);
1536
1537 r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice);
dab5bf85 1538 if (r < 0) {
f29ff115 1539 log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path);
dab5bf85
RL		 1540 return;
1541 }
8e274523 1542
efdb0237
LP		 1543 if (is_root_slice)
1544 return;
1545
1546 unit_release_cgroup(u);
0a1eb06d 1547
4ad49000 1548 u->cgroup_realized = false;
bc432dc7 1549 u->cgroup_realized_mask = 0;
ccf78df1 1550 u->cgroup_enabled_mask = 0;
8e274523
LP		 1551}
1552
efdb0237 1553int unit_search_main_pid(Unit *u, pid_t *ret) {
4ad49000
LP		 1554 _cleanup_fclose_ FILE *f = NULL;
1555 pid_t pid = 0, npid, mypid;
efdb0237 1556 int r;
4ad49000
LP		 1557
1558 assert(u);
efdb0237 1559 assert(ret);
4ad49000
LP		 1560
1561 if (!u->cgroup_path)
efdb0237 1562 return -ENXIO;
4ad49000 1563
efdb0237
LP		 1564 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f);
1565 if (r < 0)
1566 return r;
4ad49000
LP		 1567
1568 mypid = getpid();
1569 while (cg_read_pid(f, &npid) > 0) {
1570 pid_t ppid;
1571
1572 if (npid == pid)
1573 continue;
8e274523 1574
4ad49000 1575 /* Ignore processes that aren't our kids */
6bc73acb 1576 if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid)
4ad49000 1577 continue;
8e274523 1578
efdb0237 1579 if (pid != 0)
4ad49000
LP		 1580 /* Dang, there's more than one daemonized PID
1581 in this group, so we don't know what process
1582 is the main process. */
efdb0237
LP		 1583
1584 return -ENODATA;
8e274523 1585
4ad49000 1586 pid = npid;
8e274523
LP		 1587 }
1588
efdb0237
LP		 1589 *ret = pid;
1590 return 0;
1591}
1592
1593static int unit_watch_pids_in_path(Unit *u, const char *path) {
b3c5bad3 1594 _cleanup_closedir_ DIR *d = NULL;
efdb0237
LP		 1595 _cleanup_fclose_ FILE *f = NULL;
1596 int ret = 0, r;
1597
1598 assert(u);
1599 assert(path);
1600
1601 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
1602 if (r < 0)
1603 ret = r;
1604 else {
1605 pid_t pid;
1606
1607 while ((r = cg_read_pid(f, &pid)) > 0) {
1608 r = unit_watch_pid(u, pid);
1609 if (r < 0 && ret >= 0)
1610 ret = r;
1611 }
1612
1613 if (r < 0 && ret >= 0)
1614 ret = r;
1615 }
1616
1617 r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
1618 if (r < 0) {
1619 if (ret >= 0)
1620 ret = r;
1621 } else {
1622 char *fn;
1623
1624 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1625 _cleanup_free_ char *p = NULL;
1626
1627 p = strjoin(path, "/", fn, NULL);
1628 free(fn);
1629
1630 if (!p)
1631 return -ENOMEM;
1632
1633 r = unit_watch_pids_in_path(u, p);
1634 if (r < 0 && ret >= 0)
1635 ret = r;
1636 }
1637
1638 if (r < 0 && ret >= 0)
1639 ret = r;
1640 }
1641
1642 return ret;
1643}
1644
1645int unit_watch_all_pids(Unit *u) {
1646 assert(u);
1647
1648 /* Adds all PIDs from our cgroup to the set of PIDs we
1649 * watch. This is a fallback logic for cases where we do not
1650 * get reliable cgroup empty notifications: we try to use
1651 * SIGCHLD as replacement. */
1652
1653 if (!u->cgroup_path)
1654 return -ENOENT;
1655
5da38d07 1656 if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* On unified we can use proper notifications */
efdb0237
LP		 1657 return 0;
1658
1659 return unit_watch_pids_in_path(u, u->cgroup_path);
1660}
1661
1662int unit_notify_cgroup_empty(Unit *u) {
1663 int r;
1664
1665 assert(u);
1666
1667 if (!u->cgroup_path)
1668 return 0;
1669
1670 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
1671 if (r <= 0)
1672 return r;
1673
1674 unit_add_to_gc_queue(u);
1675
1676 if (UNIT_VTABLE(u)->notify_cgroup_empty)
1677 UNIT_VTABLE(u)->notify_cgroup_empty(u);
1678
1679 return 0;
1680}
1681
1682static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1683 Manager *m = userdata;
1684
1685 assert(s);
1686 assert(fd >= 0);
1687 assert(m);
1688
1689 for (;;) {
1690 union inotify_event_buffer buffer;
1691 struct inotify_event *e;
1692 ssize_t l;
1693
1694 l = read(fd, &buffer, sizeof(buffer));
1695 if (l < 0) {
1696 if (errno == EINTR || errno == EAGAIN)
1697 return 0;
1698
1699 return log_error_errno(errno, "Failed to read control group inotify events: %m");
1700 }
1701
1702 FOREACH_INOTIFY_EVENT(e, buffer, l) {
1703 Unit *u;
1704
1705 if (e->wd < 0)
1706 /* Queue overflow has no watch descriptor */
1707 continue;
1708
1709 if (e->mask & IN_IGNORED)
1710 /* The watch was just removed */
1711 continue;
1712
1713 u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd));
1714 if (!u) /* Not that inotify might deliver
1715 * events for a watch even after it
1716 * was removed, because it was queued
1717 * before the removal. Let's ignore
1718 * this here safely. */
1719 continue;
1720
1721 (void) unit_notify_cgroup_empty(u);
1722 }
1723 }
8e274523
LP		 1724}
1725
8e274523 1726int manager_setup_cgroup(Manager *m) {
9444b1f2 1727 _cleanup_free_ char *path = NULL;
efdb0237 1728 CGroupController c;
5da38d07 1729 int r, all_unified, systemd_unified;
efdb0237 1730 char *e;
8e274523
LP		 1731
1732 assert(m);
1733
35d2e7ec 1734 /* 1. Determine hierarchy */
efdb0237 1735 m->cgroup_root = mfree(m->cgroup_root);
9444b1f2 1736 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
23bbb0de
MS		 1737 if (r < 0)
1738 return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
8e274523 1739
efdb0237
LP		 1740 /* Chop off the init scope, if we are already located in it */
1741 e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
0d8c31ff 1742
efdb0237
LP		 1743 /* LEGACY: Also chop off the system slice if we are in
1744 * it. This is to support live upgrades from older systemd
1745 * versions where PID 1 was moved there. Also see
1746 * cg_get_root_path(). */
463d0d15 1747 if (!e && MANAGER_IS_SYSTEM(m)) {
9444b1f2 1748 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
15c60e99 1749 if (!e)
efdb0237 1750 e = endswith(m->cgroup_root, "/system"); /* even more legacy */
0baf24dd 1751 }
efdb0237
LP		 1752 if (e)
1753 *e = 0;
7ccfb64a 1754
9444b1f2
LP		 1755 /* And make sure to store away the root value without trailing
1756 * slash, even for the root dir, so that we can easily prepend
1757 * it everywhere. */
efdb0237
LP		 1758 while ((e = endswith(m->cgroup_root, "/")))
1759 *e = 0;
8e274523 1760
35d2e7ec 1761 /* 2. Show data */
9444b1f2 1762 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
23bbb0de
MS		 1763 if (r < 0)
1764 return log_error_errno(r, "Cannot find cgroup mount point: %m");
8e274523 1765
5da38d07
TH		 1766 all_unified = cg_all_unified();
1767 systemd_unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER);
1768
1769 if (all_unified < 0 || systemd_unified < 0)
1770 return log_error_errno(all_unified < 0 ? all_unified : systemd_unified,
1771 "Couldn't determine if we are running in the unified hierarchy: %m");
1772
1773 if (all_unified > 0)
efdb0237 1774 log_debug("Unified cgroup hierarchy is located at %s.", path);
5da38d07
TH		 1775 else if (systemd_unified > 0)
1776 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path);
efdb0237
LP		 1777 else
1778 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
1779
0d8c31ff 1780 if (!m->test_run) {
efdb0237 1781 const char *scope_path;
c6c18be3 1782
0d8c31ff 1783 /* 3. Install agent */
5da38d07 1784 if (systemd_unified) {
efdb0237 1785
61233823 1786 /* In the unified hierarchy we can get
efdb0237
LP		 1787 * cgroup empty notifications via inotify. */
1788
1789 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1790 safe_close(m->cgroup_inotify_fd);
1791
1792 m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1793 if (m->cgroup_inotify_fd < 0)
1794 return log_error_errno(errno, "Failed to create control group inotify object: %m");
1795
1796 r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m);
1797 if (r < 0)
1798 return log_error_errno(r, "Failed to watch control group inotify object: %m");
1799
d8fdc620
LP		 1800 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
1801 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
1802 r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-5);
efdb0237
LP		 1803 if (r < 0)
1804 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
1805
1806 (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify");
1807
463d0d15 1808 } else if (MANAGER_IS_SYSTEM(m)) {
efdb0237
LP		 1809
1810 /* On the legacy hierarchy we only get
1811 * notifications via cgroup agents. (Which
1812 * isn't really reliable, since it does not
1813 * generate events when control groups with
1814 * children run empty. */
1815
0d8c31ff
ZJS		 1816 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
1817 if (r < 0)
da927ba9 1818 log_warning_errno(r, "Failed to install release agent, ignoring: %m");
0d8c31ff
ZJS		 1819 else if (r > 0)
1820 log_debug("Installed release agent.");
efdb0237 1821 else if (r == 0)
0d8c31ff
ZJS		 1822 log_debug("Release agent already installed.");
1823 }
8e274523 1824
efdb0237
LP		 1825 /* 4. Make sure we are in the special "init.scope" unit in the root slice. */
1826 scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1827 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
23bbb0de 1828 if (r < 0)
efdb0237
LP		 1829 return log_error_errno(r, "Failed to create %s control group: %m", scope_path);
1830
1831 /* also, move all other userspace processes remaining
1832 * in the root cgroup into that scope. */
1d98fef1 1833 r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
efdb0237
LP		 1834 if (r < 0)
1835 log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m");
c6c18be3 1836
0d8c31ff
ZJS		 1837 /* 5. And pin it, so that it cannot be unmounted */
1838 safe_close(m->pin_cgroupfs_fd);
0d8c31ff 1839 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
4a62c710
MS		 1840 if (m->pin_cgroupfs_fd < 0)
1841 return log_error_errno(errno, "Failed to open pin file: %m");
0d8c31ff 1842
cc98b302 1843 /* 6. Always enable hierarchical support if it exists... */
5da38d07 1844 if (!all_unified)
efdb0237 1845 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
c6c18be3
LP		 1846 }
1847
0d8c31ff 1848 /* 7. Figure out which controllers are supported */
efdb0237
LP		 1849 r = cg_mask_supported(&m->cgroup_supported);
1850 if (r < 0)
1851 return log_error_errno(r, "Failed to determine supported controllers: %m");
1852
1853 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++)
eee0a1e4 1854 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c)));
9156e799 1855
a32360f1 1856 return 0;
8e274523
LP		 1857}
1858
c6c18be3 1859void manager_shutdown_cgroup(Manager *m, bool delete) {
8e274523
LP		 1860 assert(m);
1861
9444b1f2
LP		 1862 /* We can't really delete the group, since we are in it. But
1863 * let's trim it. */
1864 if (delete && m->cgroup_root)
efdb0237
LP		 1865 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
1866
1867 m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit);
1868
1869 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1870 m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd);
8e274523 1871
03e334a1 1872 m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
c6c18be3 1873
efdb0237 1874 m->cgroup_root = mfree(m->cgroup_root);
8e274523
LP		 1875}
1876
4ad49000 1877Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
acb14d31 1878 char *p;
4ad49000 1879 Unit *u;
acb14d31
LP		 1880
1881 assert(m);
1882 assert(cgroup);
acb14d31 1883
4ad49000
LP		 1884 u = hashmap_get(m->cgroup_unit, cgroup);
1885 if (u)
1886 return u;
acb14d31 1887
8e70580b 1888 p = strdupa(cgroup);
acb14d31
LP		 1889 for (;;) {
1890 char *e;
1891
1892 e = strrchr(p, '/');
efdb0237
LP		 1893 if (!e || e == p)
1894 return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE);
acb14d31
LP		 1895
1896 *e = 0;
1897
4ad49000
LP		 1898 u = hashmap_get(m->cgroup_unit, p);
1899 if (u)
1900 return u;
acb14d31
LP		 1901 }
1902}
1903
b3ac818b 1904Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) {
4ad49000 1905 _cleanup_free_ char *cgroup = NULL;
acb14d31 1906 int r;
8e274523 1907
8c47c732
LP		 1908 assert(m);
1909
b3ac818b
LP		 1910 if (pid <= 0)
1911 return NULL;
1912
1913 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
1914 if (r < 0)
1915 return NULL;
1916
1917 return manager_get_unit_by_cgroup(m, cgroup);
1918}
1919
1920Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
1921 Unit *u;
1922
1923 assert(m);
1924
efdb0237 1925 if (pid <= 0)
8c47c732
LP		 1926 return NULL;
1927
efdb0237
LP		 1928 if (pid == 1)
1929 return hashmap_get(m->units, SPECIAL_INIT_SCOPE);
1930
fea72cc0 1931 u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid));
5fe8876b
LP		 1932 if (u)
1933 return u;
1934
fea72cc0 1935 u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid));
5fe8876b
LP		 1936 if (u)
1937 return u;
1938
b3ac818b 1939 return manager_get_unit_by_pid_cgroup(m, pid);
6dde1f33 1940}
4fbf50b3 1941
4ad49000
LP		 1942int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
1943 Unit *u;
4fbf50b3 1944
4ad49000
LP		 1945 assert(m);
1946 assert(cgroup);
4fbf50b3 1947
d8fdc620
LP		 1948 log_debug("Got cgroup empty notification for: %s", cgroup);
1949
4ad49000 1950 u = manager_get_unit_by_cgroup(m, cgroup);
5ad096b3
LP		 1951 if (!u)
1952 return 0;
b56c28c3 1953
efdb0237 1954 return unit_notify_cgroup_empty(u);
5ad096b3
LP		 1955}
1956
1957int unit_get_memory_current(Unit *u, uint64_t *ret) {
1958 _cleanup_free_ char *v = NULL;
1959 int r;
1960
1961 assert(u);
1962 assert(ret);
1963
1964 if (!u->cgroup_path)
1965 return -ENODATA;
1966
efdb0237 1967 if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
5ad096b3
LP		 1968 return -ENODATA;
1969
ca2f6384 1970 if (cg_all_unified() <= 0)
efdb0237
LP		 1971 r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v);
1972 else
1973 r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v);
5ad096b3
LP		 1974 if (r == -ENOENT)
1975 return -ENODATA;
1976 if (r < 0)
1977 return r;
1978
1979 return safe_atou64(v, ret);
1980}
1981
03a7b521
LP		 1982int unit_get_tasks_current(Unit *u, uint64_t *ret) {
1983 _cleanup_free_ char *v = NULL;
1984 int r;
1985
1986 assert(u);
1987 assert(ret);
1988
1989 if (!u->cgroup_path)
1990 return -ENODATA;
1991
1992 if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0)
1993 return -ENODATA;
1994
1995 r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v);
1996 if (r == -ENOENT)
1997 return -ENODATA;
1998 if (r < 0)
1999 return r;
2000
2001 return safe_atou64(v, ret);
2002}
2003
5ad096b3
LP		 2004static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) {
2005 _cleanup_free_ char *v = NULL;
2006 uint64_t ns;
2007 int r;
2008
2009 assert(u);
2010 assert(ret);
2011
2012 if (!u->cgroup_path)
2013 return -ENODATA;
2014
ca2f6384 2015 if (cg_all_unified() > 0) {
66ebf6c0
TH		 2016 const char *keys[] = { "usage_usec", NULL };
2017 _cleanup_free_ char *val = NULL;
2018 uint64_t us;
5ad096b3 2019
66ebf6c0
TH		 2020 if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0)
2021 return -ENODATA;
5ad096b3 2022
66ebf6c0
TH		 2023 r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val);
2024 if (r < 0)
2025 return r;
2026
2027 r = safe_atou64(val, &us);
2028 if (r < 0)
2029 return r;
2030
2031 ns = us * NSEC_PER_USEC;
2032 } else {
2033 if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0)
2034 return -ENODATA;
2035
2036 r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v);
2037 if (r == -ENOENT)
2038 return -ENODATA;
2039 if (r < 0)
2040 return r;
2041
2042 r = safe_atou64(v, &ns);
2043 if (r < 0)
2044 return r;
2045 }
5ad096b3
LP		 2046
2047 *ret = ns;
2048 return 0;
2049}
2050
2051int unit_get_cpu_usage(Unit *u, nsec_t *ret) {
2052 nsec_t ns;
2053 int r;
2054
fe700f46
LP		 2055 assert(u);
2056
2057 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
2058 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
2059 * call this function with a NULL return value. */
2060
5ad096b3 2061 r = unit_get_cpu_usage_raw(u, &ns);
fe700f46
LP		 2062 if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) {
2063 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
2064 * cached value. */
2065
2066 if (ret)
2067 *ret = u->cpu_usage_last;
2068 return 0;
2069 }
5ad096b3
LP		 2070 if (r < 0)
2071 return r;
2072
66ebf6c0
TH		 2073 if (ns > u->cpu_usage_base)
2074 ns -= u->cpu_usage_base;
5ad096b3
LP		 2075 else
2076 ns = 0;
2077
fe700f46
LP		 2078 u->cpu_usage_last = ns;
2079 if (ret)
2080 *ret = ns;
2081
5ad096b3
LP		 2082 return 0;
2083}
2084
2085int unit_reset_cpu_usage(Unit *u) {
2086 nsec_t ns;
2087 int r;
2088
2089 assert(u);
2090
fe700f46
LP		 2091 u->cpu_usage_last = NSEC_INFINITY;
2092
5ad096b3
LP		 2093 r = unit_get_cpu_usage_raw(u, &ns);
2094 if (r < 0) {
66ebf6c0 2095 u->cpu_usage_base = 0;
5ad096b3 2096 return r;
b56c28c3 2097 }
2633eb83 2098
66ebf6c0 2099 u->cpu_usage_base = ns;
4ad49000 2100 return 0;
4fbf50b3
LP		 2101}
2102
e9db43d5
LP		 2103bool unit_cgroup_delegate(Unit *u) {
2104 CGroupContext *c;
2105
2106 assert(u);
2107
2108 c = unit_get_cgroup_context(u);
2109 if (!c)
2110 return false;
2111
2112 return c->delegate;
2113}
2114
e7ab4d1a
LP		 2115void unit_invalidate_cgroup(Unit *u, CGroupMask m) {
2116 assert(u);
2117
2118 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2119 return;
2120
2121 if (m == 0)
2122 return;
2123
538b4852
TH		 2124 /* always invalidate compat pairs together */
2125 if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO))
2126 m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
2127
e7ab4d1a
LP		 2128 if ((u->cgroup_realized_mask & m) == 0)
2129 return;
2130
2131 u->cgroup_realized_mask &= ~m;
2132 unit_add_to_cgroup_queue(u);
2133}
2134
2135void manager_invalidate_startup_units(Manager *m) {
2136 Iterator i;
2137 Unit *u;
2138
2139 assert(m);
2140
2141 SET_FOREACH(u, m->startup_units, i)
13c31542 2142 unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
e7ab4d1a
LP		 2143}
2144
4ad49000
LP		 2145static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
2146 [CGROUP_AUTO] = "auto",
2147 [CGROUP_CLOSED] = "closed",
2148 [CGROUP_STRICT] = "strict",
2149};
4fbf50b3 2150
4ad49000 2151DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);
Unnamed repository; edit this file 'description' to name the repository.