1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2013 Lennart Poettering
11 #include "alloc-util.h"
12 //#include "blockdev-util.h"
13 //#include "bpf-firewall.h"
14 //#include "bus-error.h"
15 #include "cgroup-util.h"
20 #include "parse-util.h"
21 #include "path-util.h"
22 #include "process-util.h"
23 //#include "procfs-util.h"
24 //#include "special.h"
25 #include "stdio-util.h"
26 #include "string-table.h"
27 #include "string-util.h"
30 #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
32 bool manager_owns_root_cgroup(Manager *m) {
35 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
36 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
37 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
38 * we run in any kind of container virtualization. */
40 if (detect_container() > 0)
43 return empty_or_root(m->cgroup_root);
46 #if 0 /// UNNEEDED by elogind
47 bool unit_has_root_cgroup(Unit *u) {
50 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
51 * the manager manages the root cgroup. */
53 if (!manager_owns_root_cgroup(u->manager))
56 return unit_has_name(u, SPECIAL_ROOT_SLICE);
59 static void cgroup_compat_warn(void) {
60 static bool cgroup_compat_warned = false;
62 if (cgroup_compat_warned)
65 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
66 "See cgroup-compat debug messages for details.");
68 cgroup_compat_warned = true;
71 #define log_cgroup_compat(unit, fmt, ...) do { \
72 cgroup_compat_warn(); \
73 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
76 void cgroup_context_init(CGroupContext *c) {
79 /* Initialize everything to the kernel defaults, assuming the
80 * structure is preinitialized to 0 */
82 c->cpu_weight = CGROUP_WEIGHT_INVALID;
83 c->startup_cpu_weight = CGROUP_WEIGHT_INVALID;
84 c->cpu_quota_per_sec_usec = USEC_INFINITY;
86 c->cpu_shares = CGROUP_CPU_SHARES_INVALID;
87 c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID;
89 c->memory_high = CGROUP_LIMIT_MAX;
90 c->memory_max = CGROUP_LIMIT_MAX;
91 c->memory_swap_max = CGROUP_LIMIT_MAX;
93 c->memory_limit = CGROUP_LIMIT_MAX;
95 c->io_weight = CGROUP_WEIGHT_INVALID;
96 c->startup_io_weight = CGROUP_WEIGHT_INVALID;
98 c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
99 c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
101 c->tasks_max = (uint64_t) -1;
104 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
108 LIST_REMOVE(device_allow, c->device_allow, a);
113 void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) {
117 LIST_REMOVE(device_weights, c->io_device_weights, w);
122 void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) {
126 LIST_REMOVE(device_limits, c->io_device_limits, l);
131 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
135 LIST_REMOVE(device_weights, c->blockio_device_weights, w);
140 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
144 LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
149 void cgroup_context_done(CGroupContext *c) {
152 while (c->io_device_weights)
153 cgroup_context_free_io_device_weight(c, c->io_device_weights);
155 while (c->io_device_limits)
156 cgroup_context_free_io_device_limit(c, c->io_device_limits);
158 while (c->blockio_device_weights)
159 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
161 while (c->blockio_device_bandwidths)
162 cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
164 while (c->device_allow)
165 cgroup_context_free_device_allow(c, c->device_allow);
167 c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow);
168 c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny);
171 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
172 CGroupIODeviceLimit *il;
173 CGroupIODeviceWeight *iw;
174 CGroupBlockIODeviceBandwidth *b;
175 CGroupBlockIODeviceWeight *w;
176 CGroupDeviceAllow *a;
177 IPAddressAccessItem *iaai;
178 char u[FORMAT_TIMESPAN_MAX];
183 prefix = strempty(prefix);
186 "%sCPUAccounting=%s\n"
187 "%sIOAccounting=%s\n"
188 "%sBlockIOAccounting=%s\n"
189 "%sMemoryAccounting=%s\n"
190 "%sTasksAccounting=%s\n"
191 "%sIPAccounting=%s\n"
192 "%sCPUWeight=%" PRIu64 "\n"
193 "%sStartupCPUWeight=%" PRIu64 "\n"
194 "%sCPUShares=%" PRIu64 "\n"
195 "%sStartupCPUShares=%" PRIu64 "\n"
196 "%sCPUQuotaPerSecSec=%s\n"
197 "%sIOWeight=%" PRIu64 "\n"
198 "%sStartupIOWeight=%" PRIu64 "\n"
199 "%sBlockIOWeight=%" PRIu64 "\n"
200 "%sStartupBlockIOWeight=%" PRIu64 "\n"
201 "%sMemoryLow=%" PRIu64 "\n"
202 "%sMemoryHigh=%" PRIu64 "\n"
203 "%sMemoryMax=%" PRIu64 "\n"
204 "%sMemorySwapMax=%" PRIu64 "\n"
205 "%sMemoryLimit=%" PRIu64 "\n"
206 "%sTasksMax=%" PRIu64 "\n"
207 "%sDevicePolicy=%s\n"
209 prefix, yes_no(c->cpu_accounting),
210 prefix, yes_no(c->io_accounting),
211 prefix, yes_no(c->blockio_accounting),
212 prefix, yes_no(c->memory_accounting),
213 prefix, yes_no(c->tasks_accounting),
214 prefix, yes_no(c->ip_accounting),
215 prefix, c->cpu_weight,
216 prefix, c->startup_cpu_weight,
217 prefix, c->cpu_shares,
218 prefix, c->startup_cpu_shares,
219 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
220 prefix, c->io_weight,
221 prefix, c->startup_io_weight,
222 prefix, c->blockio_weight,
223 prefix, c->startup_blockio_weight,
224 prefix, c->memory_low,
225 prefix, c->memory_high,
226 prefix, c->memory_max,
227 prefix, c->memory_swap_max,
228 prefix, c->memory_limit,
229 prefix, c->tasks_max,
230 prefix, cgroup_device_policy_to_string(c->device_policy),
231 prefix, yes_no(c->delegate));
234 _cleanup_free_ char *t = NULL;
236 (void) cg_mask_to_string(c->delegate_controllers, &t);
238 fprintf(f, "%sDelegateControllers=%s\n",
243 LIST_FOREACH(device_allow, a, c->device_allow)
245 "%sDeviceAllow=%s %s%s%s\n",
248 a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
250 LIST_FOREACH(device_weights, iw, c->io_device_weights)
252 "%sIODeviceWeight=%s %" PRIu64,
257 LIST_FOREACH(device_limits, il, c->io_device_limits) {
258 char buf[FORMAT_BYTES_MAX];
259 CGroupIOLimitType type;
261 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
262 if (il->limits[type] != cgroup_io_limit_defaults[type])
266 cgroup_io_limit_type_to_string(type),
268 format_bytes(buf, sizeof(buf), il->limits[type]));
271 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
273 "%sBlockIODeviceWeight=%s %" PRIu64,
278 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
279 char buf[FORMAT_BYTES_MAX];
281 if (b->rbps != CGROUP_LIMIT_MAX)
283 "%sBlockIOReadBandwidth=%s %s\n",
286 format_bytes(buf, sizeof(buf), b->rbps));
287 if (b->wbps != CGROUP_LIMIT_MAX)
289 "%sBlockIOWriteBandwidth=%s %s\n",
292 format_bytes(buf, sizeof(buf), b->wbps));
295 LIST_FOREACH(items, iaai, c->ip_address_allow) {
296 _cleanup_free_ char *k = NULL;
298 (void) in_addr_to_string(iaai->family, &iaai->address, &k);
299 fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen);
302 LIST_FOREACH(items, iaai, c->ip_address_deny) {
303 _cleanup_free_ char *k = NULL;
305 (void) in_addr_to_string(iaai->family, &iaai->address, &k);
306 fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen);
310 static int lookup_block_device(const char *p, dev_t *dev) {
316 if (stat(p, &st) < 0)
317 return log_warning_errno(errno, "Couldn't stat device %s: %m", p);
319 if (S_ISBLK(st.st_mode))
321 else if (major(st.st_dev) != 0) {
322 /* If this is not a device node then find the block
323 * device this file is stored on */
326 /* If this is a partition, try to get the originating
328 (void) block_get_whole_disk(*dev, dev);
330 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p);
337 static int whitelist_device(const char *path, const char *node, const char *acc) {
338 char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
340 bool ignore_notfound;
346 if (node[0] == '-') {
347 /* Non-existent paths starting with "-" must be silently ignored */
349 ignore_notfound = true;
351 ignore_notfound = false;
353 if (stat(node, &st) < 0) {
354 if (errno == ENOENT && ignore_notfound)
357 return log_warning_errno(errno, "Couldn't stat device %s: %m", node);
360 if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
361 log_warning("%s is not a device.", node);
367 S_ISCHR(st.st_mode) ? 'c' : 'b',
368 major(st.st_rdev), minor(st.st_rdev),
371 r = cg_set_attribute("devices", path, "devices.allow", buf);
373 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
374 "Failed to set devices.allow on %s: %m", path);
379 static int whitelist_major(const char *path, const char *name, char type, const char *acc) {
380 _cleanup_fclose_ FILE *f = NULL;
387 assert(IN_SET(type, 'b', 'c'));
389 f = fopen("/proc/devices", "re");
391 return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
393 FOREACH_LINE(line, f, goto fail) {
394 char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w;
399 if (type == 'c' && streq(line, "Character devices:")) {
404 if (type == 'b' && streq(line, "Block devices:")) {
419 w = strpbrk(p, WHITESPACE);
424 r = safe_atou(p, &maj);
431 w += strspn(w, WHITESPACE);
433 if (fnmatch(name, w, 0) != 0)
442 r = cg_set_attribute("devices", path, "devices.allow", buf);
444 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
445 "Failed to set devices.allow on %s: %m", path);
451 return log_warning_errno(errno, "Failed to read /proc/devices: %m");
454 static bool cgroup_context_has_cpu_weight(CGroupContext *c) {
455 return c->cpu_weight != CGROUP_WEIGHT_INVALID ||
456 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID;
459 static bool cgroup_context_has_cpu_shares(CGroupContext *c) {
460 return c->cpu_shares != CGROUP_CPU_SHARES_INVALID ||
461 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID;
464 static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) {
465 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
466 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID)
467 return c->startup_cpu_weight;
468 else if (c->cpu_weight != CGROUP_WEIGHT_INVALID)
469 return c->cpu_weight;
471 return CGROUP_WEIGHT_DEFAULT;
474 static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) {
475 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
476 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID)
477 return c->startup_cpu_shares;
478 else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID)
479 return c->cpu_shares;
481 return CGROUP_CPU_SHARES_DEFAULT;
484 static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) {
485 char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)];
488 xsprintf(buf, "%" PRIu64 "\n", weight);
489 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf);
491 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
492 "Failed to set cpu.weight: %m");
494 if (quota != USEC_INFINITY)
495 xsprintf(buf, USEC_FMT " " USEC_FMT "\n",
496 quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC);
498 xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
500 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf);
503 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
504 "Failed to set cpu.max: %m");
507 static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) {
508 char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1];
511 xsprintf(buf, "%" PRIu64 "\n", shares);
512 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf);
514 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
515 "Failed to set cpu.shares: %m");
517 xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
518 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf);
520 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
521 "Failed to set cpu.cfs_period_us: %m");
523 if (quota != USEC_INFINITY) {
524 xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC);
525 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf);
527 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1");
529 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
530 "Failed to set cpu.cfs_quota_us: %m");
533 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) {
534 return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT,
535 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
538 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) {
539 return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT,
540 CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX);
543 static bool cgroup_context_has_io_config(CGroupContext *c) {
544 return c->io_accounting ||
545 c->io_weight != CGROUP_WEIGHT_INVALID ||
546 c->startup_io_weight != CGROUP_WEIGHT_INVALID ||
547 c->io_device_weights ||
551 static bool cgroup_context_has_blockio_config(CGroupContext *c) {
552 return c->blockio_accounting ||
553 c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
554 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
555 c->blockio_device_weights ||
556 c->blockio_device_bandwidths;
559 static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) {
560 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
561 c->startup_io_weight != CGROUP_WEIGHT_INVALID)
562 return c->startup_io_weight;
563 else if (c->io_weight != CGROUP_WEIGHT_INVALID)
566 return CGROUP_WEIGHT_DEFAULT;
569 static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) {
570 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
571 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
572 return c->startup_blockio_weight;
573 else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
574 return c->blockio_weight;
576 return CGROUP_BLKIO_WEIGHT_DEFAULT;
579 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) {
580 return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT,
581 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
584 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) {
585 return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT,
586 CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX);
589 static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) {
590 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
594 r = lookup_block_device(dev_path, &dev);
598 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight);
599 r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf);
601 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
602 "Failed to set io.weight: %m");
605 static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) {
606 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
610 r = lookup_block_device(dev_path, &dev);
614 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight);
615 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf);
617 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
618 "Failed to set blkio.weight_device: %m");
621 static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) {
622 char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)];
623 char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
624 CGroupIOLimitType type;
629 r = lookup_block_device(dev_path, &dev);
633 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) {
634 if (limits[type] != cgroup_io_limit_defaults[type]) {
635 xsprintf(limit_bufs[type], "%" PRIu64, limits[type]);
638 xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0");
642 xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev),
643 limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX],
644 limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]);
645 r = cg_set_attribute("io", u->cgroup_path, "io.max", buf);
647 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
648 "Failed to set io.max: %m");
652 static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) {
653 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
658 r = lookup_block_device(dev_path, &dev);
662 if (rbps != CGROUP_LIMIT_MAX)
664 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps);
665 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf);
667 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
668 "Failed to set blkio.throttle.read_bps_device: %m");
670 if (wbps != CGROUP_LIMIT_MAX)
672 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps);
673 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf);
675 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
676 "Failed to set blkio.throttle.write_bps_device: %m");
681 static bool cgroup_context_has_unified_memory_config(CGroupContext *c) {
682 return c->memory_low > 0 || c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || c->memory_swap_max != CGROUP_LIMIT_MAX;
685 static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) {
686 char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
689 if (v != CGROUP_LIMIT_MAX)
690 xsprintf(buf, "%" PRIu64 "\n", v);
692 r = cg_set_attribute("memory", u->cgroup_path, file, buf);
694 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
695 "Failed to set %s: %m", file);
698 static void cgroup_apply_firewall(Unit *u) {
701 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
703 if (bpf_firewall_compile(u) < 0)
706 (void) bpf_firewall_install(u);
709 static void cgroup_context_apply(
711 CGroupMask apply_mask,
713 ManagerState state) {
722 /* Nothing to do? Exit early! */
723 if (apply_mask == 0 && !apply_bpf)
726 /* Some cgroup attributes are not supported on the root cgroup, hence silently ignore */
727 is_root = unit_has_root_cgroup(u);
729 assert_se(c = unit_get_cgroup_context(u));
730 assert_se(path = u->cgroup_path);
732 if (is_root) /* Make sure we don't try to display messages with an empty path. */
735 /* We generally ignore errors caused by read-only mounted
736 * cgroup trees (assuming we are running in a container then),
737 * and missing cgroups, i.e. EROFS and ENOENT. */
739 if ((apply_mask & CGROUP_MASK_CPU) && !is_root) {
740 bool has_weight, has_shares;
742 has_weight = cgroup_context_has_cpu_weight(c);
743 has_shares = cgroup_context_has_cpu_shares(c);
745 if (cg_all_unified() > 0) {
749 weight = cgroup_context_cpu_weight(c, state);
750 else if (has_shares) {
751 uint64_t shares = cgroup_context_cpu_shares(c, state);
753 weight = cgroup_cpu_shares_to_weight(shares);
755 log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s",
756 shares, weight, path);
758 weight = CGROUP_WEIGHT_DEFAULT;
760 cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec);
765 uint64_t weight = cgroup_context_cpu_weight(c, state);
767 shares = cgroup_cpu_weight_to_shares(weight);
769 log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s",
770 weight, shares, path);
771 } else if (has_shares)
772 shares = cgroup_context_cpu_shares(c, state);
774 shares = CGROUP_CPU_SHARES_DEFAULT;
776 cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec);
780 if (apply_mask & CGROUP_MASK_IO) {
781 bool has_io = cgroup_context_has_io_config(c);
782 bool has_blockio = cgroup_context_has_blockio_config(c);
785 char buf[8+DECIMAL_STR_MAX(uint64_t)+1];
789 weight = cgroup_context_io_weight(c, state);
790 else if (has_blockio) {
791 uint64_t blkio_weight = cgroup_context_blkio_weight(c, state);
793 weight = cgroup_weight_blkio_to_io(blkio_weight);
795 log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64,
796 blkio_weight, weight);
798 weight = CGROUP_WEIGHT_DEFAULT;
800 xsprintf(buf, "default %" PRIu64 "\n", weight);
801 r = cg_set_attribute("io", path, "io.weight", buf);
803 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
804 "Failed to set io.weight: %m");
807 CGroupIODeviceWeight *w;
809 /* FIXME: no way to reset this list */
810 LIST_FOREACH(device_weights, w, c->io_device_weights)
811 cgroup_apply_io_device_weight(u, w->path, w->weight);
812 } else if (has_blockio) {
813 CGroupBlockIODeviceWeight *w;
815 /* FIXME: no way to reset this list */
816 LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
817 weight = cgroup_weight_blkio_to_io(w->weight);
819 log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s",
820 w->weight, weight, w->path);
822 cgroup_apply_io_device_weight(u, w->path, weight);
827 /* Apply limits and free ones without config. */
829 CGroupIODeviceLimit *l, *next;
831 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
832 if (!cgroup_apply_io_device_limit(u, l->path, l->limits))
833 cgroup_context_free_io_device_limit(c, l);
835 } else if (has_blockio) {
836 CGroupBlockIODeviceBandwidth *b, *next;
838 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
839 uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX];
840 CGroupIOLimitType type;
842 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
843 limits[type] = cgroup_io_limit_defaults[type];
845 limits[CGROUP_IO_RBPS_MAX] = b->rbps;
846 limits[CGROUP_IO_WBPS_MAX] = b->wbps;
848 log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s",
849 b->rbps, b->wbps, b->path);
851 if (!cgroup_apply_io_device_limit(u, b->path, limits))
852 cgroup_context_free_blockio_device_bandwidth(c, b);
857 if (apply_mask & CGROUP_MASK_BLKIO) {
858 bool has_io = cgroup_context_has_io_config(c);
859 bool has_blockio = cgroup_context_has_blockio_config(c);
862 char buf[DECIMAL_STR_MAX(uint64_t)+1];
866 uint64_t io_weight = cgroup_context_io_weight(c, state);
868 weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state));
870 log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64,
872 } else if (has_blockio)
873 weight = cgroup_context_blkio_weight(c, state);
875 weight = CGROUP_BLKIO_WEIGHT_DEFAULT;
877 xsprintf(buf, "%" PRIu64 "\n", weight);
878 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
880 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
881 "Failed to set blkio.weight: %m");
884 CGroupIODeviceWeight *w;
886 /* FIXME: no way to reset this list */
887 LIST_FOREACH(device_weights, w, c->io_device_weights) {
888 weight = cgroup_weight_io_to_blkio(w->weight);
890 log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s",
891 w->weight, weight, w->path);
893 cgroup_apply_blkio_device_weight(u, w->path, weight);
895 } else if (has_blockio) {
896 CGroupBlockIODeviceWeight *w;
898 /* FIXME: no way to reset this list */
899 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
900 cgroup_apply_blkio_device_weight(u, w->path, w->weight);
904 /* Apply limits and free ones without config. */
906 CGroupIODeviceLimit *l, *next;
908 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
909 log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s",
910 l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path);
912 if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]))
913 cgroup_context_free_io_device_limit(c, l);
915 } else if (has_blockio) {
916 CGroupBlockIODeviceBandwidth *b, *next;
918 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths)
919 if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps))
920 cgroup_context_free_blockio_device_bandwidth(c, b);
924 if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) {
925 if (cg_all_unified() > 0) {
926 uint64_t max, swap_max = CGROUP_LIMIT_MAX;
928 if (cgroup_context_has_unified_memory_config(c)) {
930 swap_max = c->memory_swap_max;
932 max = c->memory_limit;
934 if (max != CGROUP_LIMIT_MAX)
935 log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max);
938 cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low);
939 cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high);
940 cgroup_apply_unified_memory_limit(u, "memory.max", max);
941 cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max);
943 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
946 if (cgroup_context_has_unified_memory_config(c)) {
948 log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val);
950 val = c->memory_limit;
952 if (val == CGROUP_LIMIT_MAX)
953 strncpy(buf, "-1\n", sizeof(buf));
955 xsprintf(buf, "%" PRIu64 "\n", val);
957 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
959 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
960 "Failed to set memory.limit_in_bytes: %m");
964 if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) {
965 CGroupDeviceAllow *a;
967 /* Changing the devices list of a populated cgroup
968 * might result in EINVAL, hence ignore EINVAL
971 if (c->device_allow || c->device_policy != CGROUP_AUTO)
972 r = cg_set_attribute("devices", path, "devices.deny", "a");
974 r = cg_set_attribute("devices", path, "devices.allow", "a");
976 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
977 "Failed to reset devices.list: %m");
979 if (c->device_policy == CGROUP_CLOSED ||
980 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
981 static const char auto_devices[] =
982 "/dev/null\0" "rwm\0"
983 "/dev/zero\0" "rwm\0"
984 "/dev/full\0" "rwm\0"
985 "/dev/random\0" "rwm\0"
986 "/dev/urandom\0" "rwm\0"
988 "/dev/ptmx\0" "rwm\0"
989 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
990 "-/run/systemd/inaccessible/chr\0" "rwm\0"
991 "-/run/systemd/inaccessible/blk\0" "rwm\0";
995 NULSTR_FOREACH_PAIR(x, y, auto_devices)
996 whitelist_device(path, x, y);
998 /* PTS (/dev/pts) devices may not be duplicated, but accessed */
999 whitelist_major(path, "pts", 'c', "rw");
1002 LIST_FOREACH(device_allow, a, c->device_allow) {
1018 if (path_startswith(a->path, "/dev/"))
1019 whitelist_device(path, a->path, acc);
1020 else if ((val = startswith(a->path, "block-")))
1021 whitelist_major(path, val, 'b', acc);
1022 else if ((val = startswith(a->path, "char-")))
1023 whitelist_major(path, val, 'c', acc);
1025 log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path);
1029 if (apply_mask & CGROUP_MASK_PIDS) {
1032 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1033 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1034 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1035 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1036 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1037 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1038 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1039 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1040 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1041 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1042 * which is desirable so that there's an offical way to release control of the sysctl from
1043 * systemd: set the limit to unbounded and reload. */
1045 if (c->tasks_max != CGROUP_LIMIT_MAX) {
1046 u->manager->sysctl_pid_max_changed = true;
1047 r = procfs_tasks_set_limit(c->tasks_max);
1048 } else if (u->manager->sysctl_pid_max_changed)
1049 r = procfs_tasks_set_limit(TASKS_MAX);
1054 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1055 "Failed to write to tasks limit sysctls: %m");
1058 if (c->tasks_max != CGROUP_LIMIT_MAX) {
1059 char buf[DECIMAL_STR_MAX(uint64_t) + 2];
1061 sprintf(buf, "%" PRIu64 "\n", c->tasks_max);
1062 r = cg_set_attribute("pids", path, "pids.max", buf);
1064 r = cg_set_attribute("pids", path, "pids.max", "max");
1066 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1067 "Failed to set pids.max: %m");
1072 cgroup_apply_firewall(u);
1075 CGroupMask cgroup_context_get_mask(CGroupContext *c) {
1076 CGroupMask mask = 0;
1078 /* Figure out which controllers we need */
1080 if (c->cpu_accounting ||
1081 cgroup_context_has_cpu_weight(c) ||
1082 cgroup_context_has_cpu_shares(c) ||
1083 c->cpu_quota_per_sec_usec != USEC_INFINITY)
1084 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
1086 if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c))
1087 mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
1089 if (c->memory_accounting ||
1090 c->memory_limit != CGROUP_LIMIT_MAX ||
1091 cgroup_context_has_unified_memory_config(c))
1092 mask |= CGROUP_MASK_MEMORY;
1094 if (c->device_allow ||
1095 c->device_policy != CGROUP_AUTO)
1096 mask |= CGROUP_MASK_DEVICES;
1098 if (c->tasks_accounting ||
1099 c->tasks_max != CGROUP_LIMIT_MAX)
1100 mask |= CGROUP_MASK_PIDS;
1105 CGroupMask unit_get_own_mask(Unit *u) {
1108 /* Returns the mask of controllers the unit needs for itself */
1110 c = unit_get_cgroup_context(u);
1114 return cgroup_context_get_mask(c) | unit_get_delegate_mask(u);
1117 CGroupMask unit_get_delegate_mask(Unit *u) {
1120 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1121 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1123 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1125 if (!unit_cgroup_delegate(u))
1128 if (cg_all_unified() <= 0) {
1131 e = unit_get_exec_context(u);
1132 if (e && !exec_context_maintains_privileges(e))
1136 assert_se(c = unit_get_cgroup_context(u));
1137 return c->delegate_controllers;
1140 CGroupMask unit_get_members_mask(Unit *u) {
1143 /* Returns the mask of controllers all of the unit's children require, merged */
1145 if (u->cgroup_members_mask_valid)
1146 return u->cgroup_members_mask;
1148 u->cgroup_members_mask = 0;
1150 if (u->type == UNIT_SLICE) {
1155 HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) {
1160 if (UNIT_DEREF(member->slice) != u)
1163 u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */
1167 u->cgroup_members_mask_valid = true;
1168 return u->cgroup_members_mask;
1171 CGroupMask unit_get_siblings_mask(Unit *u) {
1174 /* Returns the mask of controllers all of the unit's siblings
1175 * require, i.e. the members mask of the unit's parent slice
1176 * if there is one. */
1178 if (UNIT_ISSET(u->slice))
1179 return unit_get_members_mask(UNIT_DEREF(u->slice));
1181 return unit_get_subtree_mask(u); /* we are the top-level slice */
1184 CGroupMask unit_get_subtree_mask(Unit *u) {
1186 /* Returns the mask of this subtree, meaning of the group
1187 * itself and its children. */
1189 return unit_get_own_mask(u) | unit_get_members_mask(u);
1192 CGroupMask unit_get_target_mask(Unit *u) {
1195 /* This returns the cgroup mask of all controllers to enable
1196 * for a specific cgroup, i.e. everything it needs itself,
1197 * plus all that its children need, plus all that its siblings
1198 * need. This is primarily useful on the legacy cgroup
1199 * hierarchy, where we need to duplicate each cgroup in each
1200 * hierarchy that shall be enabled for it. */
1202 mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
1203 mask &= u->manager->cgroup_supported;
1208 CGroupMask unit_get_enable_mask(Unit *u) {
1211 /* This returns the cgroup mask of all controllers to enable
1212 * for the children of a specific cgroup. This is primarily
1213 * useful for the unified cgroup hierarchy, where each cgroup
1214 * controls which controllers are enabled for its children. */
1216 mask = unit_get_members_mask(u);
1217 mask &= u->manager->cgroup_supported;
1222 bool unit_get_needs_bpf(Unit *u) {
1227 c = unit_get_cgroup_context(u);
1231 if (c->ip_accounting ||
1232 c->ip_address_allow ||
1236 /* If any parent slice has an IP access list defined, it applies too */
1237 for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) {
1238 c = unit_get_cgroup_context(p);
1242 if (c->ip_address_allow ||
1250 /* Recurse from a unit up through its containing slices, propagating
1251 * mask bits upward. A unit is also member of itself. */
1252 void unit_update_cgroup_members_masks(Unit *u) {
1258 /* Calculate subtree mask */
1259 m = unit_get_subtree_mask(u);
1261 /* See if anything changed from the previous invocation. If
1262 * not, we're done. */
1263 if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask)
1267 u->cgroup_subtree_mask_valid &&
1268 ((m & ~u->cgroup_subtree_mask) != 0) &&
1269 ((~m & u->cgroup_subtree_mask) == 0);
1271 u->cgroup_subtree_mask = m;
1272 u->cgroup_subtree_mask_valid = true;
1274 if (UNIT_ISSET(u->slice)) {
1275 Unit *s = UNIT_DEREF(u->slice);
1278 /* There's more set now than before. We
1279 * propagate the new mask to the parent's mask
1280 * (not caring if it actually was valid or
1283 s->cgroup_members_mask |= m;
1286 /* There's less set now than before (or we
1287 * don't know), we need to recalculate
1288 * everything, so let's invalidate the
1289 * parent's members mask */
1291 s->cgroup_members_mask_valid = false;
1293 /* And now make sure that this change also hits our
1295 unit_update_cgroup_members_masks(s);
1299 const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) {
1301 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1305 if (u->cgroup_path &&
1306 u->cgroup_realized &&
1307 (u->cgroup_realized_mask & mask) == mask)
1308 return u->cgroup_path;
1310 u = UNIT_DEREF(u->slice);
1316 static const char *migrate_callback(CGroupMask mask, void *userdata) {
1317 return unit_get_realized_cgroup_path(userdata, mask);
1320 char *unit_default_cgroup_path(Unit *u) {
1321 _cleanup_free_ char *escaped = NULL, *slice = NULL;
1326 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1327 return strdup(u->manager->cgroup_root);
1329 if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) {
1330 r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice);
1335 escaped = cg_escape(u->id);
1340 return strjoin(u->manager->cgroup_root, "/", slice, "/",
1343 return strjoin(u->manager->cgroup_root, "/", escaped);
1346 int unit_set_cgroup_path(Unit *u, const char *path) {
1347 _cleanup_free_ char *p = NULL;
1359 if (streq_ptr(u->cgroup_path, p))
1363 r = hashmap_put(u->manager->cgroup_unit, p, u);
1368 unit_release_cgroup(u);
1370 u->cgroup_path = TAKE_PTR(p);
1375 int unit_watch_cgroup(Unit *u) {
1376 _cleanup_free_ char *events = NULL;
1381 if (!u->cgroup_path)
1384 if (u->cgroup_inotify_wd >= 0)
1387 /* Only applies to the unified hierarchy */
1388 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1390 return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1394 /* Don't watch the root slice, it's pointless. */
1395 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1398 r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops);
1402 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events);
1406 u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY);
1407 if (u->cgroup_inotify_wd < 0) {
1409 if (errno == ENOENT) /* If the directory is already
1410 * gone we don't need to track
1411 * it, so this is not an error */
1414 return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path);
1417 r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u);
1419 return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m");
1424 int unit_pick_cgroup_path(Unit *u) {
1425 _cleanup_free_ char *path = NULL;
1433 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1436 path = unit_default_cgroup_path(u);
1440 r = unit_set_cgroup_path(u, path);
1442 return log_unit_error_errno(u, r, "Control group %s exists already.", path);
1444 return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path);
1449 static int unit_create_cgroup(
1451 CGroupMask target_mask,
1452 CGroupMask enable_mask,
1460 c = unit_get_cgroup_context(u);
1464 /* Figure out our cgroup path */
1465 r = unit_pick_cgroup_path(u);
1469 /* First, create our own group */
1470 r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path);
1472 return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path);
1474 /* Start watching it */
1475 (void) unit_watch_cgroup(u);
1477 /* Enable all controllers we need */
1478 r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path);
1480 log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path);
1482 /* Keep track that this is now realized */
1483 u->cgroup_realized = true;
1484 u->cgroup_realized_mask = target_mask;
1485 u->cgroup_enabled_mask = enable_mask;
1486 u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF;
1488 if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) {
1490 /* Then, possibly move things over, but not if
1491 * subgroups may contain processes, which is the case
1492 * for slice and delegation units. */
1493 r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u);
1495 log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path);
1501 static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) {
1502 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1508 if (MANAGER_IS_SYSTEM(u->manager))
1511 if (!u->manager->system_bus)
1514 if (!u->cgroup_path)
1517 /* Determine this unit's cgroup path relative to our cgroup root */
1518 pp = path_startswith(u->cgroup_path, u->manager->cgroup_root);
1522 pp = strjoina("/", pp, suffix_path);
1523 path_kill_slashes(pp);
1525 r = sd_bus_call_method(u->manager->system_bus,
1526 "org.freedesktop.systemd1",
1527 "/org/freedesktop/systemd1",
1528 "org.freedesktop.systemd1.Manager",
1529 "AttachProcessesToUnit",
1532 NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid);
1534 return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r));
1539 int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) {
1540 CGroupMask delegated_mask;
1548 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1551 if (set_isempty(pids))
1554 r = unit_realize_cgroup(u);
1558 if (isempty(suffix_path))
1561 p = strjoina(u->cgroup_path, "/", suffix_path);
1563 delegated_mask = unit_get_delegate_mask(u);
1566 SET_FOREACH(pidp, pids, i) {
1567 pid_t pid = PTR_TO_PID(pidp);
1570 /* First, attach the PID to the main cgroup hierarchy */
1571 q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid);
1573 log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p);
1575 if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) {
1578 /* If we are in a user instance, and we can't move the process ourselves due to
1579 * permission problems, let's ask the system instance about it instead. Since it's more
1580 * privileged it might be able to move the process across the leaves of a subtree who's
1581 * top node is not owned by us. */
1583 z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path);
1585 log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p);
1587 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
1591 r = q; /* Remember first error */
1596 q = cg_all_unified();
1602 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
1603 * innermost realized one */
1605 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1606 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
1607 const char *realized;
1609 if (!(u->manager->cgroup_supported & bit))
1612 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
1613 if (delegated_mask & u->cgroup_realized_mask & bit) {
1614 q = cg_attach(cgroup_controller_to_string(c), p, pid);
1616 continue; /* Success! */
1618 log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to requested cgroup %s in controller %s, falling back to unit's cgroup: %m",
1619 pid, p, cgroup_controller_to_string(c));
1622 /* So this controller is either not delegate or realized, or something else weird happened. In
1623 * that case let's attach the PID at least to the closest cgroup up the tree that is
1625 realized = unit_get_realized_cgroup_path(u, bit);
1627 continue; /* Not even realized in the root slice? Then let's not bother */
1629 q = cg_attach(cgroup_controller_to_string(c), realized, pid);
1631 log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m",
1632 pid, realized, cgroup_controller_to_string(c));
1639 static void cgroup_xattr_apply(Unit *u) {
1640 char ids[SD_ID128_STRING_MAX];
1645 if (!MANAGER_IS_SYSTEM(u->manager))
1648 if (sd_id128_is_null(u->invocation_id))
1651 r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
1652 "trusted.invocation_id",
1653 sd_id128_to_string(u->invocation_id, ids), 32,
1656 log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path);
1659 static bool unit_has_mask_realized(
1661 CGroupMask target_mask,
1662 CGroupMask enable_mask,
1667 return u->cgroup_realized &&
1668 u->cgroup_realized_mask == target_mask &&
1669 u->cgroup_enabled_mask == enable_mask &&
1670 ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) ||
1671 (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF));
1674 static void unit_add_to_cgroup_realize_queue(Unit *u) {
1677 if (u->in_cgroup_realize_queue)
1680 LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
1681 u->in_cgroup_realize_queue = true;
1684 static void unit_remove_from_cgroup_realize_queue(Unit *u) {
1687 if (!u->in_cgroup_realize_queue)
1690 LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
1691 u->in_cgroup_realize_queue = false;
1694 /* Check if necessary controllers and attributes for a unit are in place.
1696 * If so, do nothing.
1697 * If not, create paths, move processes over, and set attributes.
1699 * Returns 0 on success and < 0 on failure. */
1700 static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
1701 CGroupMask target_mask, enable_mask;
1702 bool needs_bpf, apply_bpf;
1707 unit_remove_from_cgroup_realize_queue(u);
1709 target_mask = unit_get_target_mask(u);
1710 enable_mask = unit_get_enable_mask(u);
1711 needs_bpf = unit_get_needs_bpf(u);
1713 if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf))
1716 /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously
1717 * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it
1718 * this will trickle down properly to cgroupfs. */
1719 apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF;
1721 /* First, realize parents */
1722 if (UNIT_ISSET(u->slice)) {
1723 r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
1728 /* And then do the real work */
1729 r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf);
1733 /* Finally, apply the necessary attributes. */
1734 cgroup_context_apply(u, target_mask, apply_bpf, state);
1735 cgroup_xattr_apply(u);
1740 unsigned manager_dispatch_cgroup_realize_queue(Manager *m) {
1748 state = manager_state(m);
1750 while ((i = m->cgroup_realize_queue)) {
1751 assert(i->in_cgroup_realize_queue);
1753 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) {
1754 /* Maybe things changed, and the unit is not actually active anymore? */
1755 unit_remove_from_cgroup_realize_queue(i);
1759 r = unit_realize_cgroup_now(i, state);
1761 log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id);
1769 static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) {
1772 /* This adds the siblings of the specified unit and the
1773 * siblings of all parent units to the cgroup queue. (But
1774 * neither the specified unit itself nor the parents.) */
1776 while ((slice = UNIT_DEREF(u->slice))) {
1781 HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) {
1785 /* Skip units that have a dependency on the slice
1786 * but aren't actually in it. */
1787 if (UNIT_DEREF(m->slice) != slice)
1790 /* No point in doing cgroup application for units
1791 * without active processes. */
1792 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
1795 /* If the unit doesn't need any new controllers
1796 * and has current ones realized, it doesn't need
1798 if (unit_has_mask_realized(m,
1799 unit_get_target_mask(m),
1800 unit_get_enable_mask(m),
1801 unit_get_needs_bpf(m)))
1804 unit_add_to_cgroup_realize_queue(m);
1811 int unit_realize_cgroup(Unit *u) {
1814 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1817 /* So, here's the deal: when realizing the cgroups for this
1818 * unit, we need to first create all parents, but there's more
1819 * actually: for the weight-based controllers we also need to
1820 * make sure that all our siblings (i.e. units that are in the
1821 * same slice as we are) have cgroups, too. Otherwise, things
1822 * would become very uneven as each of their processes would
1823 * get as much resources as all our group together. This call
1824 * will synchronously create the parent cgroups, but will
1825 * defer work on the siblings to the next event loop
1828 /* Add all sibling slices to the cgroup queue. */
1829 unit_add_siblings_to_cgroup_realize_queue(u);
1831 /* And realize this one now (and apply the values) */
1832 return unit_realize_cgroup_now(u, manager_state(u->manager));
1835 void unit_release_cgroup(Unit *u) {
1838 /* Forgets all cgroup details for this cgroup */
1840 if (u->cgroup_path) {
1841 (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
1842 u->cgroup_path = mfree(u->cgroup_path);
1845 if (u->cgroup_inotify_wd >= 0) {
1846 if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0)
1847 log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id);
1849 (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd));
1850 u->cgroup_inotify_wd = -1;
1854 void unit_prune_cgroup(Unit *u) {
1860 /* Removes the cgroup, if empty and possible, and stops watching it. */
1862 if (!u->cgroup_path)
1865 (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */
1867 is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE);
1869 r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice);
1871 log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path);
1878 unit_release_cgroup(u);
1880 u->cgroup_realized = false;
1881 u->cgroup_realized_mask = 0;
1882 u->cgroup_enabled_mask = 0;
1885 int unit_search_main_pid(Unit *u, pid_t *ret) {
1886 _cleanup_fclose_ FILE *f = NULL;
1887 pid_t pid = 0, npid, mypid;
1893 if (!u->cgroup_path)
1896 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f);
1900 mypid = getpid_cached();
1901 while (cg_read_pid(f, &npid) > 0) {
1907 /* Ignore processes that aren't our kids */
1908 if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid)
1912 /* Dang, there's more than one daemonized PID
1913 in this group, so we don't know what process
1914 is the main process. */
1925 static int unit_watch_pids_in_path(Unit *u, const char *path) {
1926 _cleanup_closedir_ DIR *d = NULL;
1927 _cleanup_fclose_ FILE *f = NULL;
1933 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
1939 while ((r = cg_read_pid(f, &pid)) > 0) {
1940 r = unit_watch_pid(u, pid);
1941 if (r < 0 && ret >= 0)
1945 if (r < 0 && ret >= 0)
1949 r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
1956 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1957 _cleanup_free_ char *p = NULL;
1959 p = strjoin(path, "/", fn);
1965 r = unit_watch_pids_in_path(u, p);
1966 if (r < 0 && ret >= 0)
1970 if (r < 0 && ret >= 0)
1977 int unit_synthesize_cgroup_empty_event(Unit *u) {
1982 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
1983 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
1984 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
1986 if (!u->cgroup_path)
1989 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1992 if (r > 0) /* On unified we have reliable notifications, and don't need this */
1995 if (!set_isempty(u->pids))
1998 unit_add_to_cgroup_empty_queue(u);
2002 int unit_watch_all_pids(Unit *u) {
2007 /* Adds all PIDs from our cgroup to the set of PIDs we
2008 * watch. This is a fallback logic for cases where we do not
2009 * get reliable cgroup empty notifications: we try to use
2010 * SIGCHLD as replacement. */
2012 if (!u->cgroup_path)
2015 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
2018 if (r > 0) /* On unified we can use proper notifications */
2021 return unit_watch_pids_in_path(u, u->cgroup_path);
2024 static int on_cgroup_empty_event(sd_event_source *s, void *userdata) {
2025 Manager *m = userdata;
2032 u = m->cgroup_empty_queue;
2036 assert(u->in_cgroup_empty_queue);
2037 u->in_cgroup_empty_queue = false;
2038 LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u);
2040 if (m->cgroup_empty_queue) {
2041 /* More stuff queued, let's make sure we remain enabled */
2042 r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT);
2044 log_debug_errno(r, "Failed to reenable cgroup empty event source: %m");
2047 unit_add_to_gc_queue(u);
2049 if (UNIT_VTABLE(u)->notify_cgroup_empty)
2050 UNIT_VTABLE(u)->notify_cgroup_empty(u);
2055 void unit_add_to_cgroup_empty_queue(Unit *u) {
2060 /* Note that there are four different ways how cgroup empty events reach us:
2062 * 1. On the unified hierarchy we get an inotify event on the cgroup
2064 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2066 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2068 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2069 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2071 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2072 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2073 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2074 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2075 * case for scope units). */
2077 if (u->in_cgroup_empty_queue)
2080 /* Let's verify that the cgroup is really empty */
2081 if (!u->cgroup_path)
2083 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
2085 log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path);
2091 LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
2092 u->in_cgroup_empty_queue = true;
2094 /* Trigger the defer event */
2095 r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT);
2097 log_debug_errno(r, "Failed to enable cgroup empty event source: %m");
2100 static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
2101 Manager *m = userdata;
2108 union inotify_event_buffer buffer;
2109 struct inotify_event *e;
2112 l = read(fd, &buffer, sizeof(buffer));
2114 if (IN_SET(errno, EINTR, EAGAIN))
2117 return log_error_errno(errno, "Failed to read control group inotify events: %m");
2120 FOREACH_INOTIFY_EVENT(e, buffer, l) {
2124 /* Queue overflow has no watch descriptor */
2127 if (e->mask & IN_IGNORED)
2128 /* The watch was just removed */
2131 u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd));
2132 if (!u) /* Not that inotify might deliver
2133 * events for a watch even after it
2134 * was removed, because it was queued
2135 * before the removal. Let's ignore
2136 * this here safely. */
2139 unit_add_to_cgroup_empty_queue(u);
2145 int manager_setup_cgroup(Manager *m) {
2146 _cleanup_free_ char *path = NULL;
2147 const char *scope_path;
2150 #if 0 /// UNNEEDED by elogind
2156 /* 1. Determine hierarchy */
2157 m->cgroup_root = mfree(m->cgroup_root);
2158 #if 0 /// elogind is not init and must therefore search for PID 1 instead of self.
2159 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
2161 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &m->cgroup_root);
2164 return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
2166 #if 0 /// elogind does not support systemd scopes and slices
2167 /* Chop off the init scope, if we are already located in it */
2168 e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
2170 /* LEGACY: Also chop off the system slice if we are in
2171 * it. This is to support live upgrades from older systemd
2172 * versions where PID 1 was moved there. Also see
2173 * cg_get_root_path(). */
2174 if (!e && MANAGER_IS_SYSTEM(m)) {
2175 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
2177 e = endswith(m->cgroup_root, "/system"); /* even more legacy */
2183 log_debug_elogind("Cgroup Controller \"%s\" -> root \"%s\"",
2184 SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root);
2185 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
2186 * easily prepend it everywhere. */
2187 delete_trailing_chars(m->cgroup_root, "/");
2190 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
2192 return log_error_errno(r, "Cannot find cgroup mount point: %m");
2194 r = cg_unified_flush();
2196 return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m");
2198 all_unified = cg_all_unified();
2199 if (all_unified < 0)
2200 return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m");
2201 if (all_unified > 0)
2202 log_debug("Unified cgroup hierarchy is located at %s.", path);
2204 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
2206 return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m");
2208 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path);
2210 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path);
2213 #if 0 /// elogind is not init, and does not install the agent here.
2214 /* 3. Allocate cgroup empty defer event source */
2215 m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source);
2216 r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m);
2218 return log_error_errno(r, "Failed to create cgroup empty event source: %m");
2220 r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5);
2222 return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m");
2224 r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF);
2226 return log_error_errno(r, "Failed to disable cgroup empty event source: %m");
2228 (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty");
2230 /* 4. Install notifier inotify object, or agent */
2231 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
2233 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
2235 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
2236 safe_close(m->cgroup_inotify_fd);
2238 m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
2239 if (m->cgroup_inotify_fd < 0)
2240 return log_error_errno(errno, "Failed to create control group inotify object: %m");
2242 r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m);
2244 return log_error_errno(r, "Failed to watch control group inotify object: %m");
2246 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
2247 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
2248 r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4);
2250 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
2252 (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify");
2254 } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) {
2256 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
2257 * since it does not generate events when control groups with children run empty. */
2259 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
2261 log_warning_errno(r, "Failed to install release agent, ignoring: %m");
2263 log_debug("Installed release agent.");
2265 log_debug("Release agent already installed.");
2268 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
2269 scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
2270 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
2272 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
2273 r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
2275 log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m");
2278 * This method is in core, and normally called by systemd
2279 * being init. As elogind is never init, we can not install
2280 * our agent here. We do so when mounting our cgroup file
2281 * system, so only if elogind is its own tiny controller.
2282 * Further, elogind is not meant to run in systemd init scope. */
2283 if (MANAGER_IS_SYSTEM(m))
2284 // we are our own cgroup controller
2285 scope_path = strjoina("");
2286 else if (streq(m->cgroup_root, "/elogind"))
2287 // root already is our cgroup
2288 scope_path = strjoina(m->cgroup_root);
2290 // we have to create our own group
2291 scope_path = strjoina(m->cgroup_root, "/elogind");
2292 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
2294 log_debug_elogind("Created control group \"%s\"", scope_path);
2296 /* 6. And pin it, so that it cannot be unmounted */
2297 safe_close(m->pin_cgroupfs_fd);
2298 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
2299 if (m->pin_cgroupfs_fd < 0)
2300 return log_error_errno(errno, "Failed to open pin file: %m");
2302 #if 0 /// this is from the cgroup migration above that elogind does not need.
2303 } else if (r < 0 && !m->test_run_flags)
2304 return log_error_errno(r, "Failed to create %s control group: %m", scope_path);
2307 /* 7. Always enable hierarchical support if it exists... */
2308 if (!all_unified && m->test_run_flags == 0)
2309 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
2311 /* 8. Figure out which controllers are supported, and log about it */
2312 r = cg_mask_supported(&m->cgroup_supported);
2314 return log_error_errno(r, "Failed to determine supported controllers: %m");
2315 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++)
2316 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c)));
2321 void manager_shutdown_cgroup(Manager *m, bool delete) {
2324 #if 0 /// elogind is not init
2325 /* We can't really delete the group, since we are in it. But
2327 if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL)
2328 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
2330 m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source);
2332 m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit);
2334 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
2335 m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd);
2338 m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
2340 m->cgroup_root = mfree(m->cgroup_root);
2343 #if 0 /// UNNEEDED by elogind
2344 Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
2351 u = hashmap_get(m->cgroup_unit, cgroup);
2355 p = strdupa(cgroup);
2359 e = strrchr(p, '/');
2361 return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE);
2365 u = hashmap_get(m->cgroup_unit, p);
2371 Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) {
2372 _cleanup_free_ char *cgroup = NULL;
2376 if (!pid_is_valid(pid))
2379 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0)
2382 return manager_get_unit_by_cgroup(m, cgroup);
2385 Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
2390 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
2391 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
2392 * relevant one as children of the process will be assigned to that one, too, before all else. */
2394 if (!pid_is_valid(pid))
2397 if (pid == getpid_cached())
2398 return hashmap_get(m->units, SPECIAL_INIT_SCOPE);
2400 u = manager_get_unit_by_pid_cgroup(m, pid);
2404 u = hashmap_get(m->watch_pids, PID_TO_PTR(pid));
2408 array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid));
2416 #if 0 /// elogind must substitute this with its own variant
2417 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
2423 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
2424 * or from the --system instance */
2426 log_debug("Got cgroup empty notification for: %s", cgroup);
2428 u = manager_get_unit_by_cgroup(m, cgroup);
2432 unit_add_to_cgroup_empty_queue(u);
2436 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
2442 log_debug("Got cgroup empty notification for: %s", cgroup);
2444 s = hashmap_get(m->sessions, cgroup);
2447 session_finalize(s);
2450 log_warning("Session not found: %s", cgroup);
2455 #if 0 /// UNNEEDED by elogind
2456 int unit_get_memory_current(Unit *u, uint64_t *ret) {
2457 _cleanup_free_ char *v = NULL;
2463 if (!UNIT_CGROUP_BOOL(u, memory_accounting))
2466 if (!u->cgroup_path)
2469 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
2470 if (unit_has_root_cgroup(u))
2471 return procfs_memory_get_current(ret);
2473 if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
2476 r = cg_all_unified();
2480 r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v);
2482 r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v);
2488 return safe_atou64(v, ret);
2491 int unit_get_tasks_current(Unit *u, uint64_t *ret) {
2492 _cleanup_free_ char *v = NULL;
2498 if (!UNIT_CGROUP_BOOL(u, tasks_accounting))
2501 if (!u->cgroup_path)
2504 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
2505 if (unit_has_root_cgroup(u))
2506 return procfs_tasks_get_current(ret);
2508 if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0)
2511 r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v);
2517 return safe_atou64(v, ret);
2520 static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) {
2521 _cleanup_free_ char *v = NULL;
2528 if (!u->cgroup_path)
2531 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
2532 if (unit_has_root_cgroup(u))
2533 return procfs_cpu_get_usage(ret);
2535 r = cg_all_unified();
2539 _cleanup_free_ char *val = NULL;
2542 if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0)
2545 r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val);
2548 if (IN_SET(r, -ENOENT, -ENXIO))
2551 r = safe_atou64(val, &us);
2555 ns = us * NSEC_PER_USEC;
2557 if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0)
2560 r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v);
2566 r = safe_atou64(v, &ns);
2575 int unit_get_cpu_usage(Unit *u, nsec_t *ret) {
2581 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
2582 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
2583 * call this function with a NULL return value. */
2585 if (!UNIT_CGROUP_BOOL(u, cpu_accounting))
2588 r = unit_get_cpu_usage_raw(u, &ns);
2589 if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) {
2590 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
2594 *ret = u->cpu_usage_last;
2600 if (ns > u->cpu_usage_base)
2601 ns -= u->cpu_usage_base;
2605 u->cpu_usage_last = ns;
2612 int unit_get_ip_accounting(
2614 CGroupIPAccountingMetric metric,
2621 assert(metric >= 0);
2622 assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX);
2625 if (!UNIT_CGROUP_BOOL(u, ip_accounting))
2628 fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ?
2629 u->ip_accounting_ingress_map_fd :
2630 u->ip_accounting_egress_map_fd;
2634 if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES))
2635 r = bpf_firewall_read_accounting(fd, &value, NULL);
2637 r = bpf_firewall_read_accounting(fd, NULL, &value);
2641 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
2642 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
2643 * ip_accounting_extra[] field, and add them in here transparently. */
2645 *ret = value + u->ip_accounting_extra[metric];
2650 int unit_reset_cpu_accounting(Unit *u) {
2656 u->cpu_usage_last = NSEC_INFINITY;
2658 r = unit_get_cpu_usage_raw(u, &ns);
2660 u->cpu_usage_base = 0;
2664 u->cpu_usage_base = ns;
2668 int unit_reset_ip_accounting(Unit *u) {
2673 if (u->ip_accounting_ingress_map_fd >= 0)
2674 r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd);
2676 if (u->ip_accounting_egress_map_fd >= 0)
2677 q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd);
2679 zero(u->ip_accounting_extra);
2681 return r < 0 ? r : q;
2684 void unit_invalidate_cgroup(Unit *u, CGroupMask m) {
2687 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2693 /* always invalidate compat pairs together */
2694 if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO))
2695 m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
2697 if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT))
2698 m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT;
2700 if ((u->cgroup_realized_mask & m) == 0) /* NOP? */
2703 u->cgroup_realized_mask &= ~m;
2704 unit_add_to_cgroup_realize_queue(u);
2707 void unit_invalidate_cgroup_bpf(Unit *u) {
2710 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2713 if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) /* NOP? */
2716 u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED;
2717 unit_add_to_cgroup_realize_queue(u);
2719 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
2720 * list of our children includes our own. */
2721 if (u->type == UNIT_SLICE) {
2726 HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) {
2730 if (UNIT_DEREF(member->slice) != u)
2733 unit_invalidate_cgroup_bpf(member);
2738 bool unit_cgroup_delegate(Unit *u) {
2743 if (!UNIT_VTABLE(u)->can_delegate)
2746 c = unit_get_cgroup_context(u);
2753 void manager_invalidate_startup_units(Manager *m) {
2759 SET_FOREACH(u, m->startup_units, i)
2760 unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
2763 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
2764 [CGROUP_AUTO] = "auto",
2765 [CGROUP_CLOSED] = "closed",
2766 [CGROUP_STRICT] = "strict",
2769 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);