]> 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
tree-wide: use IN_SET macro (#6977)
[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 "bpf-firewall.h"
25 #include "cgroup-util.h"
26 #include "cgroup.h"
27 #include "fd-util.h"
28 #include "fileio.h"
29 #include "fs-util.h"
30 #include "parse-util.h"
31 #include "path-util.h"
32 #include "process-util.h"
33 #include "special.h"
34 #include "stdio-util.h"
35 #include "string-table.h"
36 #include "string-util.h"
37
38 #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
39
40 static void cgroup_compat_warn(void) {
41 static bool cgroup_compat_warned = false;
42
43 if (cgroup_compat_warned)
44 return;
45
46 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details.");
47 cgroup_compat_warned = true;
48 }
49
50 #define log_cgroup_compat(unit, fmt, ...) do { \
51 cgroup_compat_warn(); \
52 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
53 } while (false)
54
55 void cgroup_context_init(CGroupContext *c) {
56 assert(c);
57
58 /* Initialize everything to the kernel defaults, assuming the
59 * structure is preinitialized to 0 */
60
61 c->cpu_weight = CGROUP_WEIGHT_INVALID;
62 c->startup_cpu_weight = CGROUP_WEIGHT_INVALID;
63 c->cpu_quota_per_sec_usec = USEC_INFINITY;
64
65 c->cpu_shares = CGROUP_CPU_SHARES_INVALID;
66 c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID;
67
68 c->memory_high = CGROUP_LIMIT_MAX;
69 c->memory_max = CGROUP_LIMIT_MAX;
70 c->memory_swap_max = CGROUP_LIMIT_MAX;
71
72 c->memory_limit = CGROUP_LIMIT_MAX;
73
74 c->io_weight = CGROUP_WEIGHT_INVALID;
75 c->startup_io_weight = CGROUP_WEIGHT_INVALID;
76
77 c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
78 c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
79
80 c->tasks_max = (uint64_t) -1;
81 }
82
83 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
84 assert(c);
85 assert(a);
86
87 LIST_REMOVE(device_allow, c->device_allow, a);
88 free(a->path);
89 free(a);
90 }
91
92 void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) {
93 assert(c);
94 assert(w);
95
96 LIST_REMOVE(device_weights, c->io_device_weights, w);
97 free(w->path);
98 free(w);
99 }
100
101 void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) {
102 assert(c);
103 assert(l);
104
105 LIST_REMOVE(device_limits, c->io_device_limits, l);
106 free(l->path);
107 free(l);
108 }
109
110 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
111 assert(c);
112 assert(w);
113
114 LIST_REMOVE(device_weights, c->blockio_device_weights, w);
115 free(w->path);
116 free(w);
117 }
118
119 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
120 assert(c);
121 assert(b);
122
123 LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
124 free(b->path);
125 free(b);
126 }
127
128 void cgroup_context_done(CGroupContext *c) {
129 assert(c);
130
131 while (c->io_device_weights)
132 cgroup_context_free_io_device_weight(c, c->io_device_weights);
133
134 while (c->io_device_limits)
135 cgroup_context_free_io_device_limit(c, c->io_device_limits);
136
137 while (c->blockio_device_weights)
138 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
139
140 while (c->blockio_device_bandwidths)
141 cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
142
143 while (c->device_allow)
144 cgroup_context_free_device_allow(c, c->device_allow);
145
146 c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow);
147 c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny);
148 }
149
150 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
151 CGroupIODeviceLimit *il;
152 CGroupIODeviceWeight *iw;
153 CGroupBlockIODeviceBandwidth *b;
154 CGroupBlockIODeviceWeight *w;
155 CGroupDeviceAllow *a;
156 IPAddressAccessItem *iaai;
157 char u[FORMAT_TIMESPAN_MAX];
158
159 assert(c);
160 assert(f);
161
162 prefix = strempty(prefix);
163
164 fprintf(f,
165 "%sCPUAccounting=%s\n"
166 "%sIOAccounting=%s\n"
167 "%sBlockIOAccounting=%s\n"
168 "%sMemoryAccounting=%s\n"
169 "%sTasksAccounting=%s\n"
170 "%sIPAccounting=%s\n"
171 "%sCPUWeight=%" PRIu64 "\n"
172 "%sStartupCPUWeight=%" PRIu64 "\n"
173 "%sCPUShares=%" PRIu64 "\n"
174 "%sStartupCPUShares=%" PRIu64 "\n"
175 "%sCPUQuotaPerSecSec=%s\n"
176 "%sIOWeight=%" PRIu64 "\n"
177 "%sStartupIOWeight=%" PRIu64 "\n"
178 "%sBlockIOWeight=%" PRIu64 "\n"
179 "%sStartupBlockIOWeight=%" PRIu64 "\n"
180 "%sMemoryLow=%" PRIu64 "\n"
181 "%sMemoryHigh=%" PRIu64 "\n"
182 "%sMemoryMax=%" PRIu64 "\n"
183 "%sMemorySwapMax=%" PRIu64 "\n"
184 "%sMemoryLimit=%" PRIu64 "\n"
185 "%sTasksMax=%" PRIu64 "\n"
186 "%sDevicePolicy=%s\n"
187 "%sDelegate=%s\n",
188 prefix, yes_no(c->cpu_accounting),
189 prefix, yes_no(c->io_accounting),
190 prefix, yes_no(c->blockio_accounting),
191 prefix, yes_no(c->memory_accounting),
192 prefix, yes_no(c->tasks_accounting),
193 prefix, yes_no(c->ip_accounting),
194 prefix, c->cpu_weight,
195 prefix, c->startup_cpu_weight,
196 prefix, c->cpu_shares,
197 prefix, c->startup_cpu_shares,
198 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
199 prefix, c->io_weight,
200 prefix, c->startup_io_weight,
201 prefix, c->blockio_weight,
202 prefix, c->startup_blockio_weight,
203 prefix, c->memory_low,
204 prefix, c->memory_high,
205 prefix, c->memory_max,
206 prefix, c->memory_swap_max,
207 prefix, c->memory_limit,
208 prefix, c->tasks_max,
209 prefix, cgroup_device_policy_to_string(c->device_policy),
210 prefix, yes_no(c->delegate));
211
212 LIST_FOREACH(device_allow, a, c->device_allow)
213 fprintf(f,
214 "%sDeviceAllow=%s %s%s%s\n",
215 prefix,
216 a->path,
217 a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
218
219 LIST_FOREACH(device_weights, iw, c->io_device_weights)
220 fprintf(f,
221 "%sIODeviceWeight=%s %" PRIu64,
222 prefix,
223 iw->path,
224 iw->weight);
225
226 LIST_FOREACH(device_limits, il, c->io_device_limits) {
227 char buf[FORMAT_BYTES_MAX];
228 CGroupIOLimitType type;
229
230 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
231 if (il->limits[type] != cgroup_io_limit_defaults[type])
232 fprintf(f,
233 "%s%s=%s %s\n",
234 prefix,
235 cgroup_io_limit_type_to_string(type),
236 il->path,
237 format_bytes(buf, sizeof(buf), il->limits[type]));
238 }
239
240 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
241 fprintf(f,
242 "%sBlockIODeviceWeight=%s %" PRIu64,
243 prefix,
244 w->path,
245 w->weight);
246
247 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
248 char buf[FORMAT_BYTES_MAX];
249
250 if (b->rbps != CGROUP_LIMIT_MAX)
251 fprintf(f,
252 "%sBlockIOReadBandwidth=%s %s\n",
253 prefix,
254 b->path,
255 format_bytes(buf, sizeof(buf), b->rbps));
256 if (b->wbps != CGROUP_LIMIT_MAX)
257 fprintf(f,
258 "%sBlockIOWriteBandwidth=%s %s\n",
259 prefix,
260 b->path,
261 format_bytes(buf, sizeof(buf), b->wbps));
262 }
263
264 LIST_FOREACH(items, iaai, c->ip_address_allow) {
265 _cleanup_free_ char *k = NULL;
266
267 (void) in_addr_to_string(iaai->family, &iaai->address, &k);
268 fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen);
269 }
270
271 LIST_FOREACH(items, iaai, c->ip_address_deny) {
272 _cleanup_free_ char *k = NULL;
273
274 (void) in_addr_to_string(iaai->family, &iaai->address, &k);
275 fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen);
276 }
277 }
278
279 static int lookup_block_device(const char *p, dev_t *dev) {
280 struct stat st;
281 int r;
282
283 assert(p);
284 assert(dev);
285
286 r = stat(p, &st);
287 if (r < 0)
288 return log_warning_errno(errno, "Couldn't stat device %s: %m", p);
289
290 if (S_ISBLK(st.st_mode))
291 *dev = st.st_rdev;
292 else if (major(st.st_dev) != 0) {
293 /* If this is not a device node then find the block
294 * device this file is stored on */
295 *dev = st.st_dev;
296
297 /* If this is a partition, try to get the originating
298 * block device */
299 block_get_whole_disk(*dev, dev);
300 } else {
301 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p);
302 return -ENODEV;
303 }
304
305 return 0;
306 }
307
308 static int whitelist_device(const char *path, const char *node, const char *acc) {
309 char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
310 struct stat st;
311 bool ignore_notfound;
312 int r;
313
314 assert(path);
315 assert(acc);
316
317 if (node[0] == '-') {
318 /* Non-existent paths starting with "-" must be silently ignored */
319 node++;
320 ignore_notfound = true;
321 } else
322 ignore_notfound = false;
323
324 if (stat(node, &st) < 0) {
325 if (errno == ENOENT && ignore_notfound)
326 return 0;
327
328 return log_warning_errno(errno, "Couldn't stat device %s: %m", node);
329 }
330
331 if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
332 log_warning("%s is not a device.", node);
333 return -ENODEV;
334 }
335
336 sprintf(buf,
337 "%c %u:%u %s",
338 S_ISCHR(st.st_mode) ? 'c' : 'b',
339 major(st.st_rdev), minor(st.st_rdev),
340 acc);
341
342 r = cg_set_attribute("devices", path, "devices.allow", buf);
343 if (r < 0)
344 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
345 "Failed to set devices.allow on %s: %m", path);
346
347 return r;
348 }
349
350 static int whitelist_major(const char *path, const char *name, char type, const char *acc) {
351 _cleanup_fclose_ FILE *f = NULL;
352 char line[LINE_MAX];
353 bool good = false;
354 int r;
355
356 assert(path);
357 assert(acc);
358 assert(IN_SET(type, 'b', 'c'));
359
360 f = fopen("/proc/devices", "re");
361 if (!f)
362 return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
363
364 FOREACH_LINE(line, f, goto fail) {
365 char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w;
366 unsigned maj;
367
368 truncate_nl(line);
369
370 if (type == 'c' && streq(line, "Character devices:")) {
371 good = true;
372 continue;
373 }
374
375 if (type == 'b' && streq(line, "Block devices:")) {
376 good = true;
377 continue;
378 }
379
380 if (isempty(line)) {
381 good = false;
382 continue;
383 }
384
385 if (!good)
386 continue;
387
388 p = strstrip(line);
389
390 w = strpbrk(p, WHITESPACE);
391 if (!w)
392 continue;
393 *w = 0;
394
395 r = safe_atou(p, &maj);
396 if (r < 0)
397 continue;
398 if (maj <= 0)
399 continue;
400
401 w++;
402 w += strspn(w, WHITESPACE);
403
404 if (fnmatch(name, w, 0) != 0)
405 continue;
406
407 sprintf(buf,
408 "%c %u:* %s",
409 type,
410 maj,
411 acc);
412
413 r = cg_set_attribute("devices", path, "devices.allow", buf);
414 if (r < 0)
415 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
416 "Failed to set devices.allow on %s: %m", path);
417 }
418
419 return 0;
420
421 fail:
422 return log_warning_errno(errno, "Failed to read /proc/devices: %m");
423 }
424
425 static bool cgroup_context_has_cpu_weight(CGroupContext *c) {
426 return c->cpu_weight != CGROUP_WEIGHT_INVALID ||
427 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID;
428 }
429
430 static bool cgroup_context_has_cpu_shares(CGroupContext *c) {
431 return c->cpu_shares != CGROUP_CPU_SHARES_INVALID ||
432 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID;
433 }
434
435 static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) {
436 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
437 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID)
438 return c->startup_cpu_weight;
439 else if (c->cpu_weight != CGROUP_WEIGHT_INVALID)
440 return c->cpu_weight;
441 else
442 return CGROUP_WEIGHT_DEFAULT;
443 }
444
445 static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) {
446 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
447 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID)
448 return c->startup_cpu_shares;
449 else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID)
450 return c->cpu_shares;
451 else
452 return CGROUP_CPU_SHARES_DEFAULT;
453 }
454
455 static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) {
456 char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)];
457 int r;
458
459 xsprintf(buf, "%" PRIu64 "\n", weight);
460 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf);
461 if (r < 0)
462 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
463 "Failed to set cpu.weight: %m");
464
465 if (quota != USEC_INFINITY)
466 xsprintf(buf, USEC_FMT " " USEC_FMT "\n",
467 quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC);
468 else
469 xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
470
471 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf);
472
473 if (r < 0)
474 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
475 "Failed to set cpu.max: %m");
476 }
477
478 static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) {
479 char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1];
480 int r;
481
482 xsprintf(buf, "%" PRIu64 "\n", shares);
483 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf);
484 if (r < 0)
485 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
486 "Failed to set cpu.shares: %m");
487
488 xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
489 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf);
490 if (r < 0)
491 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
492 "Failed to set cpu.cfs_period_us: %m");
493
494 if (quota != USEC_INFINITY) {
495 xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC);
496 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf);
497 } else
498 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1");
499 if (r < 0)
500 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
501 "Failed to set cpu.cfs_quota_us: %m");
502 }
503
504 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) {
505 return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT,
506 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
507 }
508
509 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) {
510 return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT,
511 CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX);
512 }
513
514 static bool cgroup_context_has_io_config(CGroupContext *c) {
515 return c->io_accounting ||
516 c->io_weight != CGROUP_WEIGHT_INVALID ||
517 c->startup_io_weight != CGROUP_WEIGHT_INVALID ||
518 c->io_device_weights ||
519 c->io_device_limits;
520 }
521
522 static bool cgroup_context_has_blockio_config(CGroupContext *c) {
523 return c->blockio_accounting ||
524 c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
525 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
526 c->blockio_device_weights ||
527 c->blockio_device_bandwidths;
528 }
529
530 static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) {
531 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
532 c->startup_io_weight != CGROUP_WEIGHT_INVALID)
533 return c->startup_io_weight;
534 else if (c->io_weight != CGROUP_WEIGHT_INVALID)
535 return c->io_weight;
536 else
537 return CGROUP_WEIGHT_DEFAULT;
538 }
539
540 static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) {
541 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
542 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
543 return c->startup_blockio_weight;
544 else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
545 return c->blockio_weight;
546 else
547 return CGROUP_BLKIO_WEIGHT_DEFAULT;
548 }
549
550 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) {
551 return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT,
552 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
553 }
554
555 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) {
556 return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT,
557 CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX);
558 }
559
560 static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) {
561 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
562 dev_t dev;
563 int r;
564
565 r = lookup_block_device(dev_path, &dev);
566 if (r < 0)
567 return;
568
569 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight);
570 r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf);
571 if (r < 0)
572 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
573 "Failed to set io.weight: %m");
574 }
575
576 static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) {
577 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
578 dev_t dev;
579 int r;
580
581 r = lookup_block_device(dev_path, &dev);
582 if (r < 0)
583 return;
584
585 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight);
586 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", 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 blkio.weight_device: %m");
590 }
591
592 static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) {
593 char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)];
594 char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
595 CGroupIOLimitType type;
596 dev_t dev;
597 unsigned n = 0;
598 int r;
599
600 r = lookup_block_device(dev_path, &dev);
601 if (r < 0)
602 return 0;
603
604 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) {
605 if (limits[type] != cgroup_io_limit_defaults[type]) {
606 xsprintf(limit_bufs[type], "%" PRIu64, limits[type]);
607 n++;
608 } else {
609 xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0");
610 }
611 }
612
613 xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev),
614 limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX],
615 limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]);
616 r = cg_set_attribute("io", u->cgroup_path, "io.max", buf);
617 if (r < 0)
618 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
619 "Failed to set io.max: %m");
620 return n;
621 }
622
623 static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) {
624 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
625 dev_t dev;
626 unsigned n = 0;
627 int r;
628
629 r = lookup_block_device(dev_path, &dev);
630 if (r < 0)
631 return 0;
632
633 if (rbps != CGROUP_LIMIT_MAX)
634 n++;
635 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps);
636 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf);
637 if (r < 0)
638 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
639 "Failed to set blkio.throttle.read_bps_device: %m");
640
641 if (wbps != CGROUP_LIMIT_MAX)
642 n++;
643 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps);
644 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf);
645 if (r < 0)
646 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
647 "Failed to set blkio.throttle.write_bps_device: %m");
648
649 return n;
650 }
651
652 static bool cgroup_context_has_unified_memory_config(CGroupContext *c) {
653 return c->memory_low > 0 || c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || c->memory_swap_max != CGROUP_LIMIT_MAX;
654 }
655
656 static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) {
657 char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
658 int r;
659
660 if (v != CGROUP_LIMIT_MAX)
661 xsprintf(buf, "%" PRIu64 "\n", v);
662
663 r = cg_set_attribute("memory", u->cgroup_path, file, buf);
664 if (r < 0)
665 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
666 "Failed to set %s: %m", file);
667 }
668
669 static void cgroup_apply_firewall(Unit *u, CGroupContext *c) {
670 int r;
671
672 if (u->type == UNIT_SLICE) /* Skip this for slice units, they are inner cgroup nodes, and since bpf/cgroup is
673 * not recursive we don't ever touch the bpf on them */
674 return;
675
676 r = bpf_firewall_compile(u);
677 if (r < 0)
678 return;
679
680 (void) bpf_firewall_install(u);
681 return;
682 }
683
684 static void cgroup_context_apply(
685 Unit *u,
686 CGroupMask apply_mask,
687 bool apply_bpf,
688 ManagerState state) {
689
690 const char *path;
691 CGroupContext *c;
692 bool is_root;
693 int r;
694
695 assert(u);
696
697 c = unit_get_cgroup_context(u);
698 path = u->cgroup_path;
699
700 assert(c);
701 assert(path);
702
703 /* Nothing to do? Exit early! */
704 if (apply_mask == 0 && !apply_bpf)
705 return;
706
707 /* Some cgroup attributes are not supported on the root cgroup,
708 * hence silently ignore */
709 is_root = isempty(path) || path_equal(path, "/");
710 if (is_root)
711 /* Make sure we don't try to display messages with an empty path. */
712 path = "/";
713
714 /* We generally ignore errors caused by read-only mounted
715 * cgroup trees (assuming we are running in a container then),
716 * and missing cgroups, i.e. EROFS and ENOENT. */
717
718 if ((apply_mask & CGROUP_MASK_CPU) && !is_root) {
719 bool has_weight, has_shares;
720
721 has_weight = cgroup_context_has_cpu_weight(c);
722 has_shares = cgroup_context_has_cpu_shares(c);
723
724 if (cg_all_unified() > 0) {
725 uint64_t weight;
726
727 if (has_weight)
728 weight = cgroup_context_cpu_weight(c, state);
729 else if (has_shares) {
730 uint64_t shares = cgroup_context_cpu_shares(c, state);
731
732 weight = cgroup_cpu_shares_to_weight(shares);
733
734 log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s",
735 shares, weight, path);
736 } else
737 weight = CGROUP_WEIGHT_DEFAULT;
738
739 cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec);
740 } else {
741 uint64_t shares;
742
743 if (has_weight) {
744 uint64_t weight = cgroup_context_cpu_weight(c, state);
745
746 shares = cgroup_cpu_weight_to_shares(weight);
747
748 log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s",
749 weight, shares, path);
750 } else if (has_shares)
751 shares = cgroup_context_cpu_shares(c, state);
752 else
753 shares = CGROUP_CPU_SHARES_DEFAULT;
754
755 cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec);
756 }
757 }
758
759 if (apply_mask & CGROUP_MASK_IO) {
760 bool has_io = cgroup_context_has_io_config(c);
761 bool has_blockio = cgroup_context_has_blockio_config(c);
762
763 if (!is_root) {
764 char buf[8+DECIMAL_STR_MAX(uint64_t)+1];
765 uint64_t weight;
766
767 if (has_io)
768 weight = cgroup_context_io_weight(c, state);
769 else if (has_blockio) {
770 uint64_t blkio_weight = cgroup_context_blkio_weight(c, state);
771
772 weight = cgroup_weight_blkio_to_io(blkio_weight);
773
774 log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64,
775 blkio_weight, weight);
776 } else
777 weight = CGROUP_WEIGHT_DEFAULT;
778
779 xsprintf(buf, "default %" PRIu64 "\n", weight);
780 r = cg_set_attribute("io", path, "io.weight", buf);
781 if (r < 0)
782 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
783 "Failed to set io.weight: %m");
784
785 if (has_io) {
786 CGroupIODeviceWeight *w;
787
788 /* FIXME: no way to reset this list */
789 LIST_FOREACH(device_weights, w, c->io_device_weights)
790 cgroup_apply_io_device_weight(u, w->path, w->weight);
791 } else if (has_blockio) {
792 CGroupBlockIODeviceWeight *w;
793
794 /* FIXME: no way to reset this list */
795 LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
796 weight = cgroup_weight_blkio_to_io(w->weight);
797
798 log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s",
799 w->weight, weight, w->path);
800
801 cgroup_apply_io_device_weight(u, w->path, weight);
802 }
803 }
804 }
805
806 /* Apply limits and free ones without config. */
807 if (has_io) {
808 CGroupIODeviceLimit *l, *next;
809
810 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
811 if (!cgroup_apply_io_device_limit(u, l->path, l->limits))
812 cgroup_context_free_io_device_limit(c, l);
813 }
814 } else if (has_blockio) {
815 CGroupBlockIODeviceBandwidth *b, *next;
816
817 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
818 uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX];
819 CGroupIOLimitType type;
820
821 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
822 limits[type] = cgroup_io_limit_defaults[type];
823
824 limits[CGROUP_IO_RBPS_MAX] = b->rbps;
825 limits[CGROUP_IO_WBPS_MAX] = b->wbps;
826
827 log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s",
828 b->rbps, b->wbps, b->path);
829
830 if (!cgroup_apply_io_device_limit(u, b->path, limits))
831 cgroup_context_free_blockio_device_bandwidth(c, b);
832 }
833 }
834 }
835
836 if (apply_mask & CGROUP_MASK_BLKIO) {
837 bool has_io = cgroup_context_has_io_config(c);
838 bool has_blockio = cgroup_context_has_blockio_config(c);
839
840 if (!is_root) {
841 char buf[DECIMAL_STR_MAX(uint64_t)+1];
842 uint64_t weight;
843
844 if (has_io) {
845 uint64_t io_weight = cgroup_context_io_weight(c, state);
846
847 weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state));
848
849 log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64,
850 io_weight, weight);
851 } else if (has_blockio)
852 weight = cgroup_context_blkio_weight(c, state);
853 else
854 weight = CGROUP_BLKIO_WEIGHT_DEFAULT;
855
856 xsprintf(buf, "%" PRIu64 "\n", weight);
857 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
858 if (r < 0)
859 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
860 "Failed to set blkio.weight: %m");
861
862 if (has_io) {
863 CGroupIODeviceWeight *w;
864
865 /* FIXME: no way to reset this list */
866 LIST_FOREACH(device_weights, w, c->io_device_weights) {
867 weight = cgroup_weight_io_to_blkio(w->weight);
868
869 log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s",
870 w->weight, weight, w->path);
871
872 cgroup_apply_blkio_device_weight(u, w->path, weight);
873 }
874 } else if (has_blockio) {
875 CGroupBlockIODeviceWeight *w;
876
877 /* FIXME: no way to reset this list */
878 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
879 cgroup_apply_blkio_device_weight(u, w->path, w->weight);
880 }
881 }
882
883 /* Apply limits and free ones without config. */
884 if (has_io) {
885 CGroupIODeviceLimit *l, *next;
886
887 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
888 log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s",
889 l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path);
890
891 if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]))
892 cgroup_context_free_io_device_limit(c, l);
893 }
894 } else if (has_blockio) {
895 CGroupBlockIODeviceBandwidth *b, *next;
896
897 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths)
898 if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps))
899 cgroup_context_free_blockio_device_bandwidth(c, b);
900 }
901 }
902
903 if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) {
904 if (cg_all_unified() > 0) {
905 uint64_t max, swap_max = CGROUP_LIMIT_MAX;
906
907 if (cgroup_context_has_unified_memory_config(c)) {
908 max = c->memory_max;
909 swap_max = c->memory_swap_max;
910 } else {
911 max = c->memory_limit;
912
913 if (max != CGROUP_LIMIT_MAX)
914 log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max);
915 }
916
917 cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low);
918 cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high);
919 cgroup_apply_unified_memory_limit(u, "memory.max", max);
920 cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max);
921 } else {
922 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
923 uint64_t val;
924
925 if (cgroup_context_has_unified_memory_config(c)) {
926 val = c->memory_max;
927 log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val);
928 } else
929 val = c->memory_limit;
930
931 if (val == CGROUP_LIMIT_MAX)
932 strncpy(buf, "-1\n", sizeof(buf));
933 else
934 xsprintf(buf, "%" PRIu64 "\n", val);
935
936 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
937 if (r < 0)
938 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
939 "Failed to set memory.limit_in_bytes: %m");
940 }
941 }
942
943 if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) {
944 CGroupDeviceAllow *a;
945
946 /* Changing the devices list of a populated cgroup
947 * might result in EINVAL, hence ignore EINVAL
948 * here. */
949
950 if (c->device_allow || c->device_policy != CGROUP_AUTO)
951 r = cg_set_attribute("devices", path, "devices.deny", "a");
952 else
953 r = cg_set_attribute("devices", path, "devices.allow", "a");
954 if (r < 0)
955 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
956 "Failed to reset devices.list: %m");
957
958 if (c->device_policy == CGROUP_CLOSED ||
959 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
960 static const char auto_devices[] =
961 "/dev/null\0" "rwm\0"
962 "/dev/zero\0" "rwm\0"
963 "/dev/full\0" "rwm\0"
964 "/dev/random\0" "rwm\0"
965 "/dev/urandom\0" "rwm\0"
966 "/dev/tty\0" "rwm\0"
967 "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */
968 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
969 "-/run/systemd/inaccessible/chr\0" "rwm\0"
970 "-/run/systemd/inaccessible/blk\0" "rwm\0";
971
972 const char *x, *y;
973
974 NULSTR_FOREACH_PAIR(x, y, auto_devices)
975 whitelist_device(path, x, y);
976
977 whitelist_major(path, "pts", 'c', "rw");
978 }
979
980 LIST_FOREACH(device_allow, a, c->device_allow) {
981 char acc[4], *val;
982 unsigned k = 0;
983
984 if (a->r)
985 acc[k++] = 'r';
986 if (a->w)
987 acc[k++] = 'w';
988 if (a->m)
989 acc[k++] = 'm';
990
991 if (k == 0)
992 continue;
993
994 acc[k++] = 0;
995
996 if (path_startswith(a->path, "/dev/"))
997 whitelist_device(path, a->path, acc);
998 else if ((val = startswith(a->path, "block-")))
999 whitelist_major(path, val, 'b', acc);
1000 else if ((val = startswith(a->path, "char-")))
1001 whitelist_major(path, val, 'c', acc);
1002 else
1003 log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path);
1004 }
1005 }
1006
1007 if ((apply_mask & CGROUP_MASK_PIDS) && !is_root) {
1008
1009 if (c->tasks_max != CGROUP_LIMIT_MAX) {
1010 char buf[DECIMAL_STR_MAX(uint64_t) + 2];
1011
1012 sprintf(buf, "%" PRIu64 "\n", c->tasks_max);
1013 r = cg_set_attribute("pids", path, "pids.max", buf);
1014 } else
1015 r = cg_set_attribute("pids", path, "pids.max", "max");
1016
1017 if (r < 0)
1018 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1019 "Failed to set pids.max: %m");
1020 }
1021
1022 if (apply_bpf)
1023 cgroup_apply_firewall(u, c);
1024 }
1025
1026 CGroupMask cgroup_context_get_mask(CGroupContext *c) {
1027 CGroupMask mask = 0;
1028
1029 /* Figure out which controllers we need */
1030
1031 if (c->cpu_accounting ||
1032 cgroup_context_has_cpu_weight(c) ||
1033 cgroup_context_has_cpu_shares(c) ||
1034 c->cpu_quota_per_sec_usec != USEC_INFINITY)
1035 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
1036
1037 if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c))
1038 mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
1039
1040 if (c->memory_accounting ||
1041 c->memory_limit != CGROUP_LIMIT_MAX ||
1042 cgroup_context_has_unified_memory_config(c))
1043 mask |= CGROUP_MASK_MEMORY;
1044
1045 if (c->device_allow ||
1046 c->device_policy != CGROUP_AUTO)
1047 mask |= CGROUP_MASK_DEVICES;
1048
1049 if (c->tasks_accounting ||
1050 c->tasks_max != (uint64_t) -1)
1051 mask |= CGROUP_MASK_PIDS;
1052
1053 return mask;
1054 }
1055
1056 CGroupMask unit_get_own_mask(Unit *u) {
1057 CGroupContext *c;
1058
1059 /* Returns the mask of controllers the unit needs for itself */
1060
1061 c = unit_get_cgroup_context(u);
1062 if (!c)
1063 return 0;
1064
1065 /* If delegation is turned on, then turn on all cgroups,
1066 * unless we are on the legacy hierarchy and the process we
1067 * fork into it is known to drop privileges, and hence
1068 * shouldn't get access to the controllers.
1069 *
1070 * Note that on the unified hierarchy it is safe to delegate
1071 * controllers to unprivileged services. */
1072
1073 if (c->delegate) {
1074 ExecContext *e;
1075
1076 e = unit_get_exec_context(u);
1077 if (!e ||
1078 exec_context_maintains_privileges(e) ||
1079 cg_all_unified() > 0)
1080 return _CGROUP_MASK_ALL;
1081 }
1082
1083 return cgroup_context_get_mask(c);
1084 }
1085
1086 CGroupMask unit_get_members_mask(Unit *u) {
1087 assert(u);
1088
1089 /* Returns the mask of controllers all of the unit's children
1090 * require, merged */
1091
1092 if (u->cgroup_members_mask_valid)
1093 return u->cgroup_members_mask;
1094
1095 u->cgroup_members_mask = 0;
1096
1097 if (u->type == UNIT_SLICE) {
1098 Unit *member;
1099 Iterator i;
1100
1101 SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) {
1102
1103 if (member == u)
1104 continue;
1105
1106 if (UNIT_DEREF(member->slice) != u)
1107 continue;
1108
1109 u->cgroup_members_mask |=
1110 unit_get_own_mask(member) |
1111 unit_get_members_mask(member);
1112 }
1113 }
1114
1115 u->cgroup_members_mask_valid = true;
1116 return u->cgroup_members_mask;
1117 }
1118
1119 CGroupMask unit_get_siblings_mask(Unit *u) {
1120 assert(u);
1121
1122 /* Returns the mask of controllers all of the unit's siblings
1123 * require, i.e. the members mask of the unit's parent slice
1124 * if there is one. */
1125
1126 if (UNIT_ISSET(u->slice))
1127 return unit_get_members_mask(UNIT_DEREF(u->slice));
1128
1129 return unit_get_own_mask(u) | unit_get_members_mask(u);
1130 }
1131
1132 CGroupMask unit_get_subtree_mask(Unit *u) {
1133
1134 /* Returns the mask of this subtree, meaning of the group
1135 * itself and its children. */
1136
1137 return unit_get_own_mask(u) | unit_get_members_mask(u);
1138 }
1139
1140 CGroupMask unit_get_target_mask(Unit *u) {
1141 CGroupMask mask;
1142
1143 /* This returns the cgroup mask of all controllers to enable
1144 * for a specific cgroup, i.e. everything it needs itself,
1145 * plus all that its children need, plus all that its siblings
1146 * need. This is primarily useful on the legacy cgroup
1147 * hierarchy, where we need to duplicate each cgroup in each
1148 * hierarchy that shall be enabled for it. */
1149
1150 mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
1151 mask &= u->manager->cgroup_supported;
1152
1153 return mask;
1154 }
1155
1156 CGroupMask unit_get_enable_mask(Unit *u) {
1157 CGroupMask mask;
1158
1159 /* This returns the cgroup mask of all controllers to enable
1160 * for the children of a specific cgroup. This is primarily
1161 * useful for the unified cgroup hierarchy, where each cgroup
1162 * controls which controllers are enabled for its children. */
1163
1164 mask = unit_get_members_mask(u);
1165 mask &= u->manager->cgroup_supported;
1166
1167 return mask;
1168 }
1169
1170 bool unit_get_needs_bpf(Unit *u) {
1171 CGroupContext *c;
1172 Unit *p;
1173 assert(u);
1174
1175 /* We never attach BPF to slice units, as they are inner cgroup nodes and cgroup/BPF is not recursive at the
1176 * moment. */
1177 if (u->type == UNIT_SLICE)
1178 return false;
1179
1180 c = unit_get_cgroup_context(u);
1181 if (!c)
1182 return false;
1183
1184 if (c->ip_accounting ||
1185 c->ip_address_allow ||
1186 c->ip_address_deny)
1187 return true;
1188
1189 /* If any parent slice has an IP access list defined, it applies too */
1190 for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) {
1191 c = unit_get_cgroup_context(p);
1192 if (!c)
1193 return false;
1194
1195 if (c->ip_address_allow ||
1196 c->ip_address_deny)
1197 return true;
1198 }
1199
1200 return false;
1201 }
1202
1203 /* Recurse from a unit up through its containing slices, propagating
1204 * mask bits upward. A unit is also member of itself. */
1205 void unit_update_cgroup_members_masks(Unit *u) {
1206 CGroupMask m;
1207 bool more;
1208
1209 assert(u);
1210
1211 /* Calculate subtree mask */
1212 m = unit_get_subtree_mask(u);
1213
1214 /* See if anything changed from the previous invocation. If
1215 * not, we're done. */
1216 if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask)
1217 return;
1218
1219 more =
1220 u->cgroup_subtree_mask_valid &&
1221 ((m & ~u->cgroup_subtree_mask) != 0) &&
1222 ((~m & u->cgroup_subtree_mask) == 0);
1223
1224 u->cgroup_subtree_mask = m;
1225 u->cgroup_subtree_mask_valid = true;
1226
1227 if (UNIT_ISSET(u->slice)) {
1228 Unit *s = UNIT_DEREF(u->slice);
1229
1230 if (more)
1231 /* There's more set now than before. We
1232 * propagate the new mask to the parent's mask
1233 * (not caring if it actually was valid or
1234 * not). */
1235
1236 s->cgroup_members_mask |= m;
1237
1238 else
1239 /* There's less set now than before (or we
1240 * don't know), we need to recalculate
1241 * everything, so let's invalidate the
1242 * parent's members mask */
1243
1244 s->cgroup_members_mask_valid = false;
1245
1246 /* And now make sure that this change also hits our
1247 * grandparents */
1248 unit_update_cgroup_members_masks(s);
1249 }
1250 }
1251
1252 static const char *migrate_callback(CGroupMask mask, void *userdata) {
1253 Unit *u = userdata;
1254
1255 assert(mask != 0);
1256 assert(u);
1257
1258 while (u) {
1259 if (u->cgroup_path &&
1260 u->cgroup_realized &&
1261 (u->cgroup_realized_mask & mask) == mask)
1262 return u->cgroup_path;
1263
1264 u = UNIT_DEREF(u->slice);
1265 }
1266
1267 return NULL;
1268 }
1269
1270 char *unit_default_cgroup_path(Unit *u) {
1271 _cleanup_free_ char *escaped = NULL, *slice = NULL;
1272 int r;
1273
1274 assert(u);
1275
1276 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1277 return strdup(u->manager->cgroup_root);
1278
1279 if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) {
1280 r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice);
1281 if (r < 0)
1282 return NULL;
1283 }
1284
1285 escaped = cg_escape(u->id);
1286 if (!escaped)
1287 return NULL;
1288
1289 if (slice)
1290 return strjoin(u->manager->cgroup_root, "/", slice, "/",
1291 escaped);
1292 else
1293 return strjoin(u->manager->cgroup_root, "/", escaped);
1294 }
1295
1296 int unit_set_cgroup_path(Unit *u, const char *path) {
1297 _cleanup_free_ char *p = NULL;
1298 int r;
1299
1300 assert(u);
1301
1302 if (path) {
1303 p = strdup(path);
1304 if (!p)
1305 return -ENOMEM;
1306 } else
1307 p = NULL;
1308
1309 if (streq_ptr(u->cgroup_path, p))
1310 return 0;
1311
1312 if (p) {
1313 r = hashmap_put(u->manager->cgroup_unit, p, u);
1314 if (r < 0)
1315 return r;
1316 }
1317
1318 unit_release_cgroup(u);
1319
1320 u->cgroup_path = p;
1321 p = NULL;
1322
1323 return 1;
1324 }
1325
1326 int unit_watch_cgroup(Unit *u) {
1327 _cleanup_free_ char *events = NULL;
1328 int r;
1329
1330 assert(u);
1331
1332 if (!u->cgroup_path)
1333 return 0;
1334
1335 if (u->cgroup_inotify_wd >= 0)
1336 return 0;
1337
1338 /* Only applies to the unified hierarchy */
1339 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1340 if (r < 0)
1341 return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1342 if (r == 0)
1343 return 0;
1344
1345 /* Don't watch the root slice, it's pointless. */
1346 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1347 return 0;
1348
1349 r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops);
1350 if (r < 0)
1351 return log_oom();
1352
1353 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events);
1354 if (r < 0)
1355 return log_oom();
1356
1357 u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY);
1358 if (u->cgroup_inotify_wd < 0) {
1359
1360 if (errno == ENOENT) /* If the directory is already
1361 * gone we don't need to track
1362 * it, so this is not an error */
1363 return 0;
1364
1365 return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path);
1366 }
1367
1368 r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u);
1369 if (r < 0)
1370 return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m");
1371
1372 return 0;
1373 }
1374
1375 static int unit_create_cgroup(
1376 Unit *u,
1377 CGroupMask target_mask,
1378 CGroupMask enable_mask,
1379 bool needs_bpf) {
1380
1381 CGroupContext *c;
1382 int r;
1383
1384 assert(u);
1385
1386 c = unit_get_cgroup_context(u);
1387 if (!c)
1388 return 0;
1389
1390 if (!u->cgroup_path) {
1391 _cleanup_free_ char *path = NULL;
1392
1393 path = unit_default_cgroup_path(u);
1394 if (!path)
1395 return log_oom();
1396
1397 r = unit_set_cgroup_path(u, path);
1398 if (r == -EEXIST)
1399 return log_unit_error_errno(u, r, "Control group %s exists already.", path);
1400 if (r < 0)
1401 return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path);
1402 }
1403
1404 /* First, create our own group */
1405 r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path);
1406 if (r < 0)
1407 return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path);
1408
1409 /* Start watching it */
1410 (void) unit_watch_cgroup(u);
1411
1412 /* Enable all controllers we need */
1413 r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path);
1414 if (r < 0)
1415 log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path);
1416
1417 /* Keep track that this is now realized */
1418 u->cgroup_realized = true;
1419 u->cgroup_realized_mask = target_mask;
1420 u->cgroup_enabled_mask = enable_mask;
1421 u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF;
1422
1423 if (u->type != UNIT_SLICE && !c->delegate) {
1424
1425 /* Then, possibly move things over, but not if
1426 * subgroups may contain processes, which is the case
1427 * for slice and delegation units. */
1428 r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u);
1429 if (r < 0)
1430 log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path);
1431 }
1432
1433 return 0;
1434 }
1435
1436 int unit_attach_pids_to_cgroup(Unit *u) {
1437 int r;
1438 assert(u);
1439
1440 r = unit_realize_cgroup(u);
1441 if (r < 0)
1442 return r;
1443
1444 r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u);
1445 if (r < 0)
1446 return r;
1447
1448 return 0;
1449 }
1450
1451 static void cgroup_xattr_apply(Unit *u) {
1452 char ids[SD_ID128_STRING_MAX];
1453 int r;
1454
1455 assert(u);
1456
1457 if (!MANAGER_IS_SYSTEM(u->manager))
1458 return;
1459
1460 if (sd_id128_is_null(u->invocation_id))
1461 return;
1462
1463 r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
1464 "trusted.invocation_id",
1465 sd_id128_to_string(u->invocation_id, ids), 32,
1466 0);
1467 if (r < 0)
1468 log_unit_warning_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path);
1469 }
1470
1471 static bool unit_has_mask_realized(
1472 Unit *u,
1473 CGroupMask target_mask,
1474 CGroupMask enable_mask,
1475 bool needs_bpf) {
1476
1477 assert(u);
1478
1479 return u->cgroup_realized &&
1480 u->cgroup_realized_mask == target_mask &&
1481 u->cgroup_enabled_mask == enable_mask &&
1482 ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) ||
1483 (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF));
1484 }
1485
1486 /* Check if necessary controllers and attributes for a unit are in place.
1487 *
1488 * If so, do nothing.
1489 * If not, create paths, move processes over, and set attributes.
1490 *
1491 * Returns 0 on success and < 0 on failure. */
1492 static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
1493 CGroupMask target_mask, enable_mask;
1494 bool needs_bpf, apply_bpf;
1495 int r;
1496
1497 assert(u);
1498
1499 if (u->in_cgroup_realize_queue) {
1500 LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
1501 u->in_cgroup_realize_queue = false;
1502 }
1503
1504 target_mask = unit_get_target_mask(u);
1505 enable_mask = unit_get_enable_mask(u);
1506 needs_bpf = unit_get_needs_bpf(u);
1507
1508 if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf))
1509 return 0;
1510
1511 /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously
1512 * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it
1513 * this will trickle down properly to cgroupfs. */
1514 apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF;
1515
1516 /* First, realize parents */
1517 if (UNIT_ISSET(u->slice)) {
1518 r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
1519 if (r < 0)
1520 return r;
1521 }
1522
1523 /* And then do the real work */
1524 r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf);
1525 if (r < 0)
1526 return r;
1527
1528 /* Finally, apply the necessary attributes. */
1529 cgroup_context_apply(u, target_mask, apply_bpf, state);
1530 cgroup_xattr_apply(u);
1531
1532 return 0;
1533 }
1534
1535 static void unit_add_to_cgroup_realize_queue(Unit *u) {
1536 assert(u);
1537
1538 if (u->in_cgroup_realize_queue)
1539 return;
1540
1541 LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
1542 u->in_cgroup_realize_queue = true;
1543 }
1544
1545 unsigned manager_dispatch_cgroup_realize_queue(Manager *m) {
1546 ManagerState state;
1547 unsigned n = 0;
1548 Unit *i;
1549 int r;
1550
1551 assert(m);
1552
1553 state = manager_state(m);
1554
1555 while ((i = m->cgroup_realize_queue)) {
1556 assert(i->in_cgroup_realize_queue);
1557
1558 r = unit_realize_cgroup_now(i, state);
1559 if (r < 0)
1560 log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id);
1561
1562 n++;
1563 }
1564
1565 return n;
1566 }
1567
1568 static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) {
1569 Unit *slice;
1570
1571 /* This adds the siblings of the specified unit and the
1572 * siblings of all parent units to the cgroup queue. (But
1573 * neither the specified unit itself nor the parents.) */
1574
1575 while ((slice = UNIT_DEREF(u->slice))) {
1576 Iterator i;
1577 Unit *m;
1578
1579 SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) {
1580 if (m == u)
1581 continue;
1582
1583 /* Skip units that have a dependency on the slice
1584 * but aren't actually in it. */
1585 if (UNIT_DEREF(m->slice) != slice)
1586 continue;
1587
1588 /* No point in doing cgroup application for units
1589 * without active processes. */
1590 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
1591 continue;
1592
1593 /* If the unit doesn't need any new controllers
1594 * and has current ones realized, it doesn't need
1595 * any changes. */
1596 if (unit_has_mask_realized(m,
1597 unit_get_target_mask(m),
1598 unit_get_enable_mask(m),
1599 unit_get_needs_bpf(m)))
1600 continue;
1601
1602 unit_add_to_cgroup_realize_queue(m);
1603 }
1604
1605 u = slice;
1606 }
1607 }
1608
1609 int unit_realize_cgroup(Unit *u) {
1610 assert(u);
1611
1612 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1613 return 0;
1614
1615 /* So, here's the deal: when realizing the cgroups for this
1616 * unit, we need to first create all parents, but there's more
1617 * actually: for the weight-based controllers we also need to
1618 * make sure that all our siblings (i.e. units that are in the
1619 * same slice as we are) have cgroups, too. Otherwise, things
1620 * would become very uneven as each of their processes would
1621 * get as much resources as all our group together. This call
1622 * will synchronously create the parent cgroups, but will
1623 * defer work on the siblings to the next event loop
1624 * iteration. */
1625
1626 /* Add all sibling slices to the cgroup queue. */
1627 unit_add_siblings_to_cgroup_realize_queue(u);
1628
1629 /* And realize this one now (and apply the values) */
1630 return unit_realize_cgroup_now(u, manager_state(u->manager));
1631 }
1632
1633 void unit_release_cgroup(Unit *u) {
1634 assert(u);
1635
1636 /* Forgets all cgroup details for this cgroup */
1637
1638 if (u->cgroup_path) {
1639 (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
1640 u->cgroup_path = mfree(u->cgroup_path);
1641 }
1642
1643 if (u->cgroup_inotify_wd >= 0) {
1644 if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0)
1645 log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id);
1646
1647 (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd));
1648 u->cgroup_inotify_wd = -1;
1649 }
1650 }
1651
1652 void unit_prune_cgroup(Unit *u) {
1653 int r;
1654 bool is_root_slice;
1655
1656 assert(u);
1657
1658 /* Removes the cgroup, if empty and possible, and stops watching it. */
1659
1660 if (!u->cgroup_path)
1661 return;
1662
1663 (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */
1664
1665 is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE);
1666
1667 r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice);
1668 if (r < 0) {
1669 log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path);
1670 return;
1671 }
1672
1673 if (is_root_slice)
1674 return;
1675
1676 unit_release_cgroup(u);
1677
1678 u->cgroup_realized = false;
1679 u->cgroup_realized_mask = 0;
1680 u->cgroup_enabled_mask = 0;
1681 }
1682
1683 int unit_search_main_pid(Unit *u, pid_t *ret) {
1684 _cleanup_fclose_ FILE *f = NULL;
1685 pid_t pid = 0, npid, mypid;
1686 int r;
1687
1688 assert(u);
1689 assert(ret);
1690
1691 if (!u->cgroup_path)
1692 return -ENXIO;
1693
1694 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f);
1695 if (r < 0)
1696 return r;
1697
1698 mypid = getpid_cached();
1699 while (cg_read_pid(f, &npid) > 0) {
1700 pid_t ppid;
1701
1702 if (npid == pid)
1703 continue;
1704
1705 /* Ignore processes that aren't our kids */
1706 if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid)
1707 continue;
1708
1709 if (pid != 0)
1710 /* Dang, there's more than one daemonized PID
1711 in this group, so we don't know what process
1712 is the main process. */
1713
1714 return -ENODATA;
1715
1716 pid = npid;
1717 }
1718
1719 *ret = pid;
1720 return 0;
1721 }
1722
1723 static int unit_watch_pids_in_path(Unit *u, const char *path) {
1724 _cleanup_closedir_ DIR *d = NULL;
1725 _cleanup_fclose_ FILE *f = NULL;
1726 int ret = 0, r;
1727
1728 assert(u);
1729 assert(path);
1730
1731 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
1732 if (r < 0)
1733 ret = r;
1734 else {
1735 pid_t pid;
1736
1737 while ((r = cg_read_pid(f, &pid)) > 0) {
1738 r = unit_watch_pid(u, pid);
1739 if (r < 0 && ret >= 0)
1740 ret = r;
1741 }
1742
1743 if (r < 0 && ret >= 0)
1744 ret = r;
1745 }
1746
1747 r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
1748 if (r < 0) {
1749 if (ret >= 0)
1750 ret = r;
1751 } else {
1752 char *fn;
1753
1754 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1755 _cleanup_free_ char *p = NULL;
1756
1757 p = strjoin(path, "/", fn);
1758 free(fn);
1759
1760 if (!p)
1761 return -ENOMEM;
1762
1763 r = unit_watch_pids_in_path(u, p);
1764 if (r < 0 && ret >= 0)
1765 ret = r;
1766 }
1767
1768 if (r < 0 && ret >= 0)
1769 ret = r;
1770 }
1771
1772 return ret;
1773 }
1774
1775 int unit_watch_all_pids(Unit *u) {
1776 int r;
1777
1778 assert(u);
1779
1780 /* Adds all PIDs from our cgroup to the set of PIDs we
1781 * watch. This is a fallback logic for cases where we do not
1782 * get reliable cgroup empty notifications: we try to use
1783 * SIGCHLD as replacement. */
1784
1785 if (!u->cgroup_path)
1786 return -ENOENT;
1787
1788 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1789 if (r < 0)
1790 return r;
1791 if (r > 0) /* On unified we can use proper notifications */
1792 return 0;
1793
1794 return unit_watch_pids_in_path(u, u->cgroup_path);
1795 }
1796
1797 static int on_cgroup_empty_event(sd_event_source *s, void *userdata) {
1798 Manager *m = userdata;
1799 Unit *u;
1800 int r;
1801
1802 assert(s);
1803 assert(m);
1804
1805 u = m->cgroup_empty_queue;
1806 if (!u)
1807 return 0;
1808
1809 assert(u->in_cgroup_empty_queue);
1810 u->in_cgroup_empty_queue = false;
1811 LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u);
1812
1813 if (m->cgroup_empty_queue) {
1814 /* More stuff queued, let's make sure we remain enabled */
1815 r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT);
1816 if (r < 0)
1817 log_debug_errno(r, "Failed to reenable cgroup empty event source: %m");
1818 }
1819
1820 unit_add_to_gc_queue(u);
1821
1822 if (UNIT_VTABLE(u)->notify_cgroup_empty)
1823 UNIT_VTABLE(u)->notify_cgroup_empty(u);
1824
1825 return 0;
1826 }
1827
1828 void unit_add_to_cgroup_empty_queue(Unit *u) {
1829 int r;
1830
1831 assert(u);
1832
1833 /* Note that there are four different ways how cgroup empty events reach us:
1834 *
1835 * 1. On the unified hierarchy we get an inotify event on the cgroup
1836 *
1837 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
1838 *
1839 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
1840 *
1841 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
1842 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
1843 *
1844 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
1845 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
1846 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
1847 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
1848 * case for scope units). */
1849
1850 if (u->in_cgroup_empty_queue)
1851 return;
1852
1853 /* Let's verify that the cgroup is really empty */
1854 if (!u->cgroup_path)
1855 return;
1856 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
1857 if (r < 0) {
1858 log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path);
1859 return;
1860 }
1861 if (r == 0)
1862 return;
1863
1864 LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
1865 u->in_cgroup_empty_queue = true;
1866
1867 /* Trigger the defer event */
1868 r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT);
1869 if (r < 0)
1870 log_debug_errno(r, "Failed to enable cgroup empty event source: %m");
1871 }
1872
1873 static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1874 Manager *m = userdata;
1875
1876 assert(s);
1877 assert(fd >= 0);
1878 assert(m);
1879
1880 for (;;) {
1881 union inotify_event_buffer buffer;
1882 struct inotify_event *e;
1883 ssize_t l;
1884
1885 l = read(fd, &buffer, sizeof(buffer));
1886 if (l < 0) {
1887 if (IN_SET(errno, EINTR, EAGAIN))
1888 return 0;
1889
1890 return log_error_errno(errno, "Failed to read control group inotify events: %m");
1891 }
1892
1893 FOREACH_INOTIFY_EVENT(e, buffer, l) {
1894 Unit *u;
1895
1896 if (e->wd < 0)
1897 /* Queue overflow has no watch descriptor */
1898 continue;
1899
1900 if (e->mask & IN_IGNORED)
1901 /* The watch was just removed */
1902 continue;
1903
1904 u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd));
1905 if (!u) /* Not that inotify might deliver
1906 * events for a watch even after it
1907 * was removed, because it was queued
1908 * before the removal. Let's ignore
1909 * this here safely. */
1910 continue;
1911
1912 unit_add_to_cgroup_empty_queue(u);
1913 }
1914 }
1915 }
1916
1917 int manager_setup_cgroup(Manager *m) {
1918 _cleanup_free_ char *path = NULL;
1919 const char *scope_path;
1920 CGroupController c;
1921 int r, all_unified;
1922 char *e;
1923
1924 assert(m);
1925
1926 /* 1. Determine hierarchy */
1927 m->cgroup_root = mfree(m->cgroup_root);
1928 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
1929 if (r < 0)
1930 return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
1931
1932 /* Chop off the init scope, if we are already located in it */
1933 e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1934
1935 /* LEGACY: Also chop off the system slice if we are in
1936 * it. This is to support live upgrades from older systemd
1937 * versions where PID 1 was moved there. Also see
1938 * cg_get_root_path(). */
1939 if (!e && MANAGER_IS_SYSTEM(m)) {
1940 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
1941 if (!e)
1942 e = endswith(m->cgroup_root, "/system"); /* even more legacy */
1943 }
1944 if (e)
1945 *e = 0;
1946
1947 /* And make sure to store away the root value without trailing
1948 * slash, even for the root dir, so that we can easily prepend
1949 * it everywhere. */
1950 while ((e = endswith(m->cgroup_root, "/")))
1951 *e = 0;
1952
1953 /* 2. Show data */
1954 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
1955 if (r < 0)
1956 return log_error_errno(r, "Cannot find cgroup mount point: %m");
1957
1958 r = cg_unified_flush();
1959 if (r < 0)
1960 return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m");
1961
1962 all_unified = cg_all_unified();
1963 if (r < 0)
1964 return log_error_errno(r, "Couldn't determine whether we are in all unified mode: %m");
1965 if (r > 0)
1966 log_debug("Unified cgroup hierarchy is located at %s.", path);
1967 else {
1968 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1969 if (r < 0)
1970 return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m");
1971 if (r > 0)
1972 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path);
1973 else
1974 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path);
1975 }
1976
1977 /* 3. Allocate cgroup empty defer event source */
1978 m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source);
1979 r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m);
1980 if (r < 0)
1981 return log_error_errno(r, "Failed to create cgroup empty event source: %m");
1982
1983 r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5);
1984 if (r < 0)
1985 return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m");
1986
1987 r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF);
1988 if (r < 0)
1989 return log_error_errno(r, "Failed to disable cgroup empty event source: %m");
1990
1991 (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty");
1992
1993 /* 4. Install notifier inotify object, or agent */
1994 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
1995
1996 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
1997
1998 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1999 safe_close(m->cgroup_inotify_fd);
2000
2001 m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
2002 if (m->cgroup_inotify_fd < 0)
2003 return log_error_errno(errno, "Failed to create control group inotify object: %m");
2004
2005 r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m);
2006 if (r < 0)
2007 return log_error_errno(r, "Failed to watch control group inotify object: %m");
2008
2009 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
2010 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
2011 r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4);
2012 if (r < 0)
2013 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
2014
2015 (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify");
2016
2017 } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) {
2018
2019 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
2020 * since it does not generate events when control groups with children run empty. */
2021
2022 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
2023 if (r < 0)
2024 log_warning_errno(r, "Failed to install release agent, ignoring: %m");
2025 else if (r > 0)
2026 log_debug("Installed release agent.");
2027 else if (r == 0)
2028 log_debug("Release agent already installed.");
2029 }
2030
2031 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
2032 scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
2033 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
2034 if (r < 0)
2035 return log_error_errno(r, "Failed to create %s control group: %m", scope_path);
2036
2037 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
2038 r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
2039 if (r < 0)
2040 log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m");
2041
2042 /* 6. And pin it, so that it cannot be unmounted */
2043 safe_close(m->pin_cgroupfs_fd);
2044 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
2045 if (m->pin_cgroupfs_fd < 0)
2046 return log_error_errno(errno, "Failed to open pin file: %m");
2047
2048 /* 7. Always enable hierarchical support if it exists... */
2049 if (!all_unified && m->test_run_flags == 0)
2050 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
2051
2052 /* 8. Figure out which controllers are supported, and log about it */
2053 r = cg_mask_supported(&m->cgroup_supported);
2054 if (r < 0)
2055 return log_error_errno(r, "Failed to determine supported controllers: %m");
2056 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++)
2057 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c)));
2058
2059 return 0;
2060 }
2061
2062 void manager_shutdown_cgroup(Manager *m, bool delete) {
2063 assert(m);
2064
2065 /* We can't really delete the group, since we are in it. But
2066 * let's trim it. */
2067 if (delete && m->cgroup_root)
2068 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
2069
2070 m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source);
2071
2072 m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit);
2073
2074 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
2075 m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd);
2076
2077 m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
2078
2079 m->cgroup_root = mfree(m->cgroup_root);
2080 }
2081
2082 Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
2083 char *p;
2084 Unit *u;
2085
2086 assert(m);
2087 assert(cgroup);
2088
2089 u = hashmap_get(m->cgroup_unit, cgroup);
2090 if (u)
2091 return u;
2092
2093 p = strdupa(cgroup);
2094 for (;;) {
2095 char *e;
2096
2097 e = strrchr(p, '/');
2098 if (!e || e == p)
2099 return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE);
2100
2101 *e = 0;
2102
2103 u = hashmap_get(m->cgroup_unit, p);
2104 if (u)
2105 return u;
2106 }
2107 }
2108
2109 Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) {
2110 _cleanup_free_ char *cgroup = NULL;
2111 int r;
2112
2113 assert(m);
2114
2115 if (pid <= 0)
2116 return NULL;
2117
2118 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
2119 if (r < 0)
2120 return NULL;
2121
2122 return manager_get_unit_by_cgroup(m, cgroup);
2123 }
2124
2125 Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
2126 Unit *u;
2127
2128 assert(m);
2129
2130 if (pid <= 0)
2131 return NULL;
2132
2133 if (pid == 1)
2134 return hashmap_get(m->units, SPECIAL_INIT_SCOPE);
2135
2136 u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid));
2137 if (u)
2138 return u;
2139
2140 u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid));
2141 if (u)
2142 return u;
2143
2144 return manager_get_unit_by_pid_cgroup(m, pid);
2145 }
2146
2147 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
2148 Unit *u;
2149
2150 assert(m);
2151 assert(cgroup);
2152
2153 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
2154 * or from the --system instance */
2155
2156 log_debug("Got cgroup empty notification for: %s", cgroup);
2157
2158 u = manager_get_unit_by_cgroup(m, cgroup);
2159 if (!u)
2160 return 0;
2161
2162 unit_add_to_cgroup_empty_queue(u);
2163 return 1;
2164 }
2165
2166 int unit_get_memory_current(Unit *u, uint64_t *ret) {
2167 _cleanup_free_ char *v = NULL;
2168 int r;
2169
2170 assert(u);
2171 assert(ret);
2172
2173 if (!UNIT_CGROUP_BOOL(u, memory_accounting))
2174 return -ENODATA;
2175
2176 if (!u->cgroup_path)
2177 return -ENODATA;
2178
2179 if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
2180 return -ENODATA;
2181
2182 r = cg_all_unified();
2183 if (r < 0)
2184 return r;
2185 if (r > 0)
2186 r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v);
2187 else
2188 r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v);
2189 if (r == -ENOENT)
2190 return -ENODATA;
2191 if (r < 0)
2192 return r;
2193
2194 return safe_atou64(v, ret);
2195 }
2196
2197 int unit_get_tasks_current(Unit *u, uint64_t *ret) {
2198 _cleanup_free_ char *v = NULL;
2199 int r;
2200
2201 assert(u);
2202 assert(ret);
2203
2204 if (!UNIT_CGROUP_BOOL(u, tasks_accounting))
2205 return -ENODATA;
2206
2207 if (!u->cgroup_path)
2208 return -ENODATA;
2209
2210 if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0)
2211 return -ENODATA;
2212
2213 r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v);
2214 if (r == -ENOENT)
2215 return -ENODATA;
2216 if (r < 0)
2217 return r;
2218
2219 return safe_atou64(v, ret);
2220 }
2221
2222 static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) {
2223 _cleanup_free_ char *v = NULL;
2224 uint64_t ns;
2225 int r;
2226
2227 assert(u);
2228 assert(ret);
2229
2230 if (!u->cgroup_path)
2231 return -ENODATA;
2232
2233 r = cg_all_unified();
2234 if (r < 0)
2235 return r;
2236 if (r > 0) {
2237 const char *keys[] = { "usage_usec", NULL };
2238 _cleanup_free_ char *val = NULL;
2239 uint64_t us;
2240
2241 if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0)
2242 return -ENODATA;
2243
2244 r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val);
2245 if (r < 0)
2246 return r;
2247
2248 r = safe_atou64(val, &us);
2249 if (r < 0)
2250 return r;
2251
2252 ns = us * NSEC_PER_USEC;
2253 } else {
2254 if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0)
2255 return -ENODATA;
2256
2257 r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v);
2258 if (r == -ENOENT)
2259 return -ENODATA;
2260 if (r < 0)
2261 return r;
2262
2263 r = safe_atou64(v, &ns);
2264 if (r < 0)
2265 return r;
2266 }
2267
2268 *ret = ns;
2269 return 0;
2270 }
2271
2272 int unit_get_cpu_usage(Unit *u, nsec_t *ret) {
2273 nsec_t ns;
2274 int r;
2275
2276 assert(u);
2277
2278 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
2279 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
2280 * call this function with a NULL return value. */
2281
2282 if (!UNIT_CGROUP_BOOL(u, cpu_accounting))
2283 return -ENODATA;
2284
2285 r = unit_get_cpu_usage_raw(u, &ns);
2286 if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) {
2287 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
2288 * cached value. */
2289
2290 if (ret)
2291 *ret = u->cpu_usage_last;
2292 return 0;
2293 }
2294 if (r < 0)
2295 return r;
2296
2297 if (ns > u->cpu_usage_base)
2298 ns -= u->cpu_usage_base;
2299 else
2300 ns = 0;
2301
2302 u->cpu_usage_last = ns;
2303 if (ret)
2304 *ret = ns;
2305
2306 return 0;
2307 }
2308
2309 int unit_get_ip_accounting(
2310 Unit *u,
2311 CGroupIPAccountingMetric metric,
2312 uint64_t *ret) {
2313
2314 uint64_t value;
2315 int fd, r;
2316
2317 assert(u);
2318 assert(metric >= 0);
2319 assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX);
2320 assert(ret);
2321
2322 /* IP accounting is currently not recursive, and hence we refuse to return any data for slice nodes. Slices are
2323 * inner cgroup nodes and hence have no processes directly attached, hence their counters would be zero
2324 * anyway. And if we block this now we can later open this up, if the kernel learns recursive BPF cgroup
2325 * filters. */
2326 if (u->type == UNIT_SLICE)
2327 return -ENODATA;
2328
2329 if (!UNIT_CGROUP_BOOL(u, ip_accounting))
2330 return -ENODATA;
2331
2332 fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ?
2333 u->ip_accounting_ingress_map_fd :
2334 u->ip_accounting_egress_map_fd;
2335
2336 if (fd < 0)
2337 return -ENODATA;
2338
2339 if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES))
2340 r = bpf_firewall_read_accounting(fd, &value, NULL);
2341 else
2342 r = bpf_firewall_read_accounting(fd, NULL, &value);
2343 if (r < 0)
2344 return r;
2345
2346 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
2347 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
2348 * ip_accounting_extra[] field, and add them in here transparently. */
2349
2350 *ret = value + u->ip_accounting_extra[metric];
2351
2352 return r;
2353 }
2354
2355 int unit_reset_cpu_accounting(Unit *u) {
2356 nsec_t ns;
2357 int r;
2358
2359 assert(u);
2360
2361 u->cpu_usage_last = NSEC_INFINITY;
2362
2363 r = unit_get_cpu_usage_raw(u, &ns);
2364 if (r < 0) {
2365 u->cpu_usage_base = 0;
2366 return r;
2367 }
2368
2369 u->cpu_usage_base = ns;
2370 return 0;
2371 }
2372
2373 int unit_reset_ip_accounting(Unit *u) {
2374 int r = 0, q = 0;
2375
2376 assert(u);
2377
2378 if (u->ip_accounting_ingress_map_fd >= 0)
2379 r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd);
2380
2381 if (u->ip_accounting_egress_map_fd >= 0)
2382 q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd);
2383
2384 zero(u->ip_accounting_extra);
2385
2386 return r < 0 ? r : q;
2387 }
2388
2389 void unit_invalidate_cgroup(Unit *u, CGroupMask m) {
2390 assert(u);
2391
2392 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2393 return;
2394
2395 if (m == 0)
2396 return;
2397
2398 /* always invalidate compat pairs together */
2399 if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO))
2400 m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
2401
2402 if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT))
2403 m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT;
2404
2405 if ((u->cgroup_realized_mask & m) == 0)
2406 return;
2407
2408 u->cgroup_realized_mask &= ~m;
2409 unit_add_to_cgroup_realize_queue(u);
2410 }
2411
2412 void unit_invalidate_cgroup_bpf(Unit *u) {
2413 assert(u);
2414
2415 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2416 return;
2417
2418 if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED)
2419 return;
2420
2421 u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED;
2422 unit_add_to_cgroup_realize_queue(u);
2423
2424 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
2425 * list of our children includes our own. */
2426 if (u->type == UNIT_SLICE) {
2427 Unit *member;
2428 Iterator i;
2429
2430 SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) {
2431 if (member == u)
2432 continue;
2433
2434 if (UNIT_DEREF(member->slice) != u)
2435 continue;
2436
2437 unit_invalidate_cgroup_bpf(member);
2438 }
2439 }
2440 }
2441
2442 void manager_invalidate_startup_units(Manager *m) {
2443 Iterator i;
2444 Unit *u;
2445
2446 assert(m);
2447
2448 SET_FOREACH(u, m->startup_units, i)
2449 unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
2450 }
2451
2452 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
2453 [CGROUP_AUTO] = "auto",
2454 [CGROUP_CLOSED] = "closed",
2455 [CGROUP_STRICT] = "strict",
2456 };
2457
2458 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);
Unnamed repository; edit this file 'description' to name the repository.