1 /* SPDX-License-Identifier: LGPL-2.1+ */
6 #include "alloc-util.h"
7 //#include "blockdev-util.h"
8 //#include "bpf-firewall.h"
9 //#include "btrfs-util.h"
10 //#include "bus-error.h"
11 #include "cgroup-util.h"
16 #include "parse-util.h"
17 #include "path-util.h"
18 #include "process-util.h"
19 //#include "procfs-util.h"
20 //#include "special.h"
21 #include "stdio-util.h"
22 #include "string-table.h"
23 #include "string-util.h"
26 #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
28 bool manager_owns_root_cgroup(Manager *m) {
31 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
32 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
33 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
34 * we run in any kind of container virtualization. */
36 if (detect_container() > 0)
39 return empty_or_root(m->cgroup_root);
42 #if 0 /// UNNEEDED by elogind
43 bool unit_has_root_cgroup(Unit *u) {
46 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
47 * the manager manages the root cgroup. */
49 if (!manager_owns_root_cgroup(u->manager))
52 return unit_has_name(u, SPECIAL_ROOT_SLICE);
55 static void cgroup_compat_warn(void) {
56 static bool cgroup_compat_warned = false;
58 if (cgroup_compat_warned)
61 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
62 "See cgroup-compat debug messages for details.");
64 cgroup_compat_warned = true;
67 #define log_cgroup_compat(unit, fmt, ...) do { \
68 cgroup_compat_warn(); \
69 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
72 void cgroup_context_init(CGroupContext *c) {
75 /* Initialize everything to the kernel defaults, assuming the
76 * structure is preinitialized to 0 */
78 c->cpu_weight = CGROUP_WEIGHT_INVALID;
79 c->startup_cpu_weight = CGROUP_WEIGHT_INVALID;
80 c->cpu_quota_per_sec_usec = USEC_INFINITY;
82 c->cpu_shares = CGROUP_CPU_SHARES_INVALID;
83 c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID;
85 c->memory_high = CGROUP_LIMIT_MAX;
86 c->memory_max = CGROUP_LIMIT_MAX;
87 c->memory_swap_max = CGROUP_LIMIT_MAX;
89 c->memory_limit = CGROUP_LIMIT_MAX;
91 c->io_weight = CGROUP_WEIGHT_INVALID;
92 c->startup_io_weight = CGROUP_WEIGHT_INVALID;
94 c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
95 c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
97 c->tasks_max = (uint64_t) -1;
100 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
104 LIST_REMOVE(device_allow, c->device_allow, a);
109 void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) {
113 LIST_REMOVE(device_weights, c->io_device_weights, w);
118 void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) {
122 LIST_REMOVE(device_limits, c->io_device_limits, l);
127 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
131 LIST_REMOVE(device_weights, c->blockio_device_weights, w);
136 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
140 LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
145 void cgroup_context_done(CGroupContext *c) {
148 while (c->io_device_weights)
149 cgroup_context_free_io_device_weight(c, c->io_device_weights);
151 while (c->io_device_limits)
152 cgroup_context_free_io_device_limit(c, c->io_device_limits);
154 while (c->blockio_device_weights)
155 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
157 while (c->blockio_device_bandwidths)
158 cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
160 while (c->device_allow)
161 cgroup_context_free_device_allow(c, c->device_allow);
163 c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow);
164 c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny);
167 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
168 CGroupIODeviceLimit *il;
169 CGroupIODeviceWeight *iw;
170 CGroupBlockIODeviceBandwidth *b;
171 CGroupBlockIODeviceWeight *w;
172 CGroupDeviceAllow *a;
173 IPAddressAccessItem *iaai;
174 char u[FORMAT_TIMESPAN_MAX];
179 prefix = strempty(prefix);
182 "%sCPUAccounting=%s\n"
183 "%sIOAccounting=%s\n"
184 "%sBlockIOAccounting=%s\n"
185 "%sMemoryAccounting=%s\n"
186 "%sTasksAccounting=%s\n"
187 "%sIPAccounting=%s\n"
188 "%sCPUWeight=%" PRIu64 "\n"
189 "%sStartupCPUWeight=%" PRIu64 "\n"
190 "%sCPUShares=%" PRIu64 "\n"
191 "%sStartupCPUShares=%" PRIu64 "\n"
192 "%sCPUQuotaPerSecSec=%s\n"
193 "%sIOWeight=%" PRIu64 "\n"
194 "%sStartupIOWeight=%" PRIu64 "\n"
195 "%sBlockIOWeight=%" PRIu64 "\n"
196 "%sStartupBlockIOWeight=%" PRIu64 "\n"
197 "%sMemoryLow=%" PRIu64 "\n"
198 "%sMemoryHigh=%" PRIu64 "\n"
199 "%sMemoryMax=%" PRIu64 "\n"
200 "%sMemorySwapMax=%" PRIu64 "\n"
201 "%sMemoryLimit=%" PRIu64 "\n"
202 "%sTasksMax=%" PRIu64 "\n"
203 "%sDevicePolicy=%s\n"
205 prefix, yes_no(c->cpu_accounting),
206 prefix, yes_no(c->io_accounting),
207 prefix, yes_no(c->blockio_accounting),
208 prefix, yes_no(c->memory_accounting),
209 prefix, yes_no(c->tasks_accounting),
210 prefix, yes_no(c->ip_accounting),
211 prefix, c->cpu_weight,
212 prefix, c->startup_cpu_weight,
213 prefix, c->cpu_shares,
214 prefix, c->startup_cpu_shares,
215 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
216 prefix, c->io_weight,
217 prefix, c->startup_io_weight,
218 prefix, c->blockio_weight,
219 prefix, c->startup_blockio_weight,
220 prefix, c->memory_low,
221 prefix, c->memory_high,
222 prefix, c->memory_max,
223 prefix, c->memory_swap_max,
224 prefix, c->memory_limit,
225 prefix, c->tasks_max,
226 prefix, cgroup_device_policy_to_string(c->device_policy),
227 prefix, yes_no(c->delegate));
230 _cleanup_free_ char *t = NULL;
232 (void) cg_mask_to_string(c->delegate_controllers, &t);
234 fprintf(f, "%sDelegateControllers=%s\n",
239 LIST_FOREACH(device_allow, a, c->device_allow)
241 "%sDeviceAllow=%s %s%s%s\n",
244 a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
246 LIST_FOREACH(device_weights, iw, c->io_device_weights)
248 "%sIODeviceWeight=%s %" PRIu64,
253 LIST_FOREACH(device_limits, il, c->io_device_limits) {
254 char buf[FORMAT_BYTES_MAX];
255 CGroupIOLimitType type;
257 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
258 if (il->limits[type] != cgroup_io_limit_defaults[type])
262 cgroup_io_limit_type_to_string(type),
264 format_bytes(buf, sizeof(buf), il->limits[type]));
267 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
269 "%sBlockIODeviceWeight=%s %" PRIu64,
274 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
275 char buf[FORMAT_BYTES_MAX];
277 if (b->rbps != CGROUP_LIMIT_MAX)
279 "%sBlockIOReadBandwidth=%s %s\n",
282 format_bytes(buf, sizeof(buf), b->rbps));
283 if (b->wbps != CGROUP_LIMIT_MAX)
285 "%sBlockIOWriteBandwidth=%s %s\n",
288 format_bytes(buf, sizeof(buf), b->wbps));
291 LIST_FOREACH(items, iaai, c->ip_address_allow) {
292 _cleanup_free_ char *k = NULL;
294 (void) in_addr_to_string(iaai->family, &iaai->address, &k);
295 fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen);
298 LIST_FOREACH(items, iaai, c->ip_address_deny) {
299 _cleanup_free_ char *k = NULL;
301 (void) in_addr_to_string(iaai->family, &iaai->address, &k);
302 fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen);
306 static int lookup_block_device(const char *p, dev_t *ret) {
313 if (stat(p, &st) < 0)
314 return log_warning_errno(errno, "Couldn't stat device '%s': %m", p);
316 if (S_ISBLK(st.st_mode))
318 else if (major(st.st_dev) != 0)
319 *ret = st.st_dev; /* If this is not a device node then use the block device this file is stored on */
321 /* If this is btrfs, getting the backing block device is a bit harder */
322 r = btrfs_get_block_device(p, ret);
323 if (r < 0 && r != -ENOTTY)
324 return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p);
326 log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p);
331 /* If this is a LUKS device, try to get the originating block device */
332 (void) block_get_originating(*ret, ret);
334 /* If this is a partition, try to get the originating block device */
335 (void) block_get_whole_disk(*ret, ret);
339 static int whitelist_device(const char *path, const char *node, const char *acc) {
340 char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
342 bool ignore_notfound;
348 if (node[0] == '-') {
349 /* Non-existent paths starting with "-" must be silently ignored */
351 ignore_notfound = true;
353 ignore_notfound = false;
355 if (stat(node, &st) < 0) {
356 if (errno == ENOENT && ignore_notfound)
359 return log_warning_errno(errno, "Couldn't stat device %s: %m", node);
362 if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
363 log_warning("%s is not a device.", node);
369 S_ISCHR(st.st_mode) ? 'c' : 'b',
370 major(st.st_rdev), minor(st.st_rdev),
373 r = cg_set_attribute("devices", path, "devices.allow", buf);
375 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
376 "Failed to set devices.allow on %s: %m", path);
381 static int whitelist_major(const char *path, const char *name, char type, const char *acc) {
382 _cleanup_fclose_ FILE *f = NULL;
389 assert(IN_SET(type, 'b', 'c'));
391 f = fopen("/proc/devices", "re");
393 return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
395 FOREACH_LINE(line, f, goto fail) {
396 char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w;
401 if (type == 'c' && streq(line, "Character devices:")) {
406 if (type == 'b' && streq(line, "Block devices:")) {
421 w = strpbrk(p, WHITESPACE);
426 r = safe_atou(p, &maj);
433 w += strspn(w, WHITESPACE);
435 if (fnmatch(name, w, 0) != 0)
444 r = cg_set_attribute("devices", path, "devices.allow", buf);
446 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
447 "Failed to set devices.allow on %s: %m", path);
453 return log_warning_errno(errno, "Failed to read /proc/devices: %m");
456 static bool cgroup_context_has_cpu_weight(CGroupContext *c) {
457 return c->cpu_weight != CGROUP_WEIGHT_INVALID ||
458 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID;
461 static bool cgroup_context_has_cpu_shares(CGroupContext *c) {
462 return c->cpu_shares != CGROUP_CPU_SHARES_INVALID ||
463 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID;
466 static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) {
467 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
468 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID)
469 return c->startup_cpu_weight;
470 else if (c->cpu_weight != CGROUP_WEIGHT_INVALID)
471 return c->cpu_weight;
473 return CGROUP_WEIGHT_DEFAULT;
476 static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) {
477 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
478 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID)
479 return c->startup_cpu_shares;
480 else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID)
481 return c->cpu_shares;
483 return CGROUP_CPU_SHARES_DEFAULT;
486 static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) {
487 char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)];
490 xsprintf(buf, "%" PRIu64 "\n", weight);
491 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf);
493 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
494 "Failed to set cpu.weight: %m");
496 if (quota != USEC_INFINITY)
497 xsprintf(buf, USEC_FMT " " USEC_FMT "\n",
498 quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC);
500 xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
502 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf);
505 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
506 "Failed to set cpu.max: %m");
509 static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) {
510 char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1];
513 xsprintf(buf, "%" PRIu64 "\n", shares);
514 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf);
516 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
517 "Failed to set cpu.shares: %m");
519 xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
520 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf);
522 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
523 "Failed to set cpu.cfs_period_us: %m");
525 if (quota != USEC_INFINITY) {
526 xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC);
527 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf);
529 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1");
531 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
532 "Failed to set cpu.cfs_quota_us: %m");
535 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) {
536 return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT,
537 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
540 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) {
541 return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT,
542 CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX);
545 static bool cgroup_context_has_io_config(CGroupContext *c) {
546 return c->io_accounting ||
547 c->io_weight != CGROUP_WEIGHT_INVALID ||
548 c->startup_io_weight != CGROUP_WEIGHT_INVALID ||
549 c->io_device_weights ||
553 static bool cgroup_context_has_blockio_config(CGroupContext *c) {
554 return c->blockio_accounting ||
555 c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
556 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
557 c->blockio_device_weights ||
558 c->blockio_device_bandwidths;
561 static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) {
562 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
563 c->startup_io_weight != CGROUP_WEIGHT_INVALID)
564 return c->startup_io_weight;
565 else if (c->io_weight != CGROUP_WEIGHT_INVALID)
568 return CGROUP_WEIGHT_DEFAULT;
571 static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) {
572 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
573 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
574 return c->startup_blockio_weight;
575 else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
576 return c->blockio_weight;
578 return CGROUP_BLKIO_WEIGHT_DEFAULT;
581 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) {
582 return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT,
583 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
586 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) {
587 return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT,
588 CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX);
591 static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) {
592 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
596 r = lookup_block_device(dev_path, &dev);
600 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight);
601 r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf);
603 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
604 "Failed to set io.weight: %m");
607 static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) {
608 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
612 r = lookup_block_device(dev_path, &dev);
616 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight);
617 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf);
619 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
620 "Failed to set blkio.weight_device: %m");
623 static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) {
624 char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)];
625 char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
626 CGroupIOLimitType type;
630 r = lookup_block_device(dev_path, &dev);
634 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
635 if (limits[type] != cgroup_io_limit_defaults[type])
636 xsprintf(limit_bufs[type], "%" PRIu64, limits[type]);
638 xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0");
640 xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev),
641 limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX],
642 limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]);
643 r = cg_set_attribute("io", u->cgroup_path, "io.max", buf);
645 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
646 "Failed to set io.max: %m");
649 static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) {
650 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
654 r = lookup_block_device(dev_path, &dev);
658 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps);
659 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf);
661 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
662 "Failed to set blkio.throttle.read_bps_device: %m");
664 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps);
665 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_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.write_bps_device: %m");
671 static bool cgroup_context_has_unified_memory_config(CGroupContext *c) {
672 return c->memory_low > 0 || c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || c->memory_swap_max != CGROUP_LIMIT_MAX;
675 static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) {
676 char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
679 if (v != CGROUP_LIMIT_MAX)
680 xsprintf(buf, "%" PRIu64 "\n", v);
682 r = cg_set_attribute("memory", u->cgroup_path, file, buf);
684 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
685 "Failed to set %s: %m", file);
688 static void cgroup_apply_firewall(Unit *u) {
691 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
693 if (bpf_firewall_compile(u) < 0)
696 (void) bpf_firewall_install(u);
699 static void cgroup_context_apply(
701 CGroupMask apply_mask,
703 ManagerState state) {
712 /* Nothing to do? Exit early! */
713 if (apply_mask == 0 && !apply_bpf)
716 /* Some cgroup attributes are not supported on the root cgroup, hence silently ignore */
717 is_root = unit_has_root_cgroup(u);
719 assert_se(c = unit_get_cgroup_context(u));
720 assert_se(path = u->cgroup_path);
722 if (is_root) /* Make sure we don't try to display messages with an empty path. */
725 /* We generally ignore errors caused by read-only mounted
726 * cgroup trees (assuming we are running in a container then),
727 * and missing cgroups, i.e. EROFS and ENOENT. */
729 if ((apply_mask & CGROUP_MASK_CPU) && !is_root) {
730 bool has_weight, has_shares;
732 has_weight = cgroup_context_has_cpu_weight(c);
733 has_shares = cgroup_context_has_cpu_shares(c);
735 if (cg_all_unified() > 0) {
739 weight = cgroup_context_cpu_weight(c, state);
740 else if (has_shares) {
741 uint64_t shares = cgroup_context_cpu_shares(c, state);
743 weight = cgroup_cpu_shares_to_weight(shares);
745 log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s",
746 shares, weight, path);
748 weight = CGROUP_WEIGHT_DEFAULT;
750 cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec);
755 uint64_t weight = cgroup_context_cpu_weight(c, state);
757 shares = cgroup_cpu_weight_to_shares(weight);
759 log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s",
760 weight, shares, path);
761 } else if (has_shares)
762 shares = cgroup_context_cpu_shares(c, state);
764 shares = CGROUP_CPU_SHARES_DEFAULT;
766 cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec);
770 if (apply_mask & CGROUP_MASK_IO) {
771 bool has_io = cgroup_context_has_io_config(c);
772 bool has_blockio = cgroup_context_has_blockio_config(c);
775 char buf[8+DECIMAL_STR_MAX(uint64_t)+1];
779 weight = cgroup_context_io_weight(c, state);
780 else if (has_blockio) {
781 uint64_t blkio_weight = cgroup_context_blkio_weight(c, state);
783 weight = cgroup_weight_blkio_to_io(blkio_weight);
785 log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64,
786 blkio_weight, weight);
788 weight = CGROUP_WEIGHT_DEFAULT;
790 xsprintf(buf, "default %" PRIu64 "\n", weight);
791 r = cg_set_attribute("io", path, "io.weight", buf);
793 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
794 "Failed to set io.weight: %m");
797 CGroupIODeviceWeight *w;
799 /* FIXME: no way to reset this list */
800 LIST_FOREACH(device_weights, w, c->io_device_weights)
801 cgroup_apply_io_device_weight(u, w->path, w->weight);
802 } else if (has_blockio) {
803 CGroupBlockIODeviceWeight *w;
805 /* FIXME: no way to reset this list */
806 LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
807 weight = cgroup_weight_blkio_to_io(w->weight);
809 log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s",
810 w->weight, weight, w->path);
812 cgroup_apply_io_device_weight(u, w->path, weight);
817 /* Apply limits and free ones without config. */
819 CGroupIODeviceLimit *l;
821 LIST_FOREACH(device_limits, l, c->io_device_limits)
822 cgroup_apply_io_device_limit(u, l->path, l->limits);
824 } else if (has_blockio) {
825 CGroupBlockIODeviceBandwidth *b;
827 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
828 uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX];
829 CGroupIOLimitType type;
831 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
832 limits[type] = cgroup_io_limit_defaults[type];
834 limits[CGROUP_IO_RBPS_MAX] = b->rbps;
835 limits[CGROUP_IO_WBPS_MAX] = b->wbps;
837 log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s",
838 b->rbps, b->wbps, b->path);
840 cgroup_apply_io_device_limit(u, b->path, limits);
845 if (apply_mask & CGROUP_MASK_BLKIO) {
846 bool has_io = cgroup_context_has_io_config(c);
847 bool has_blockio = cgroup_context_has_blockio_config(c);
850 char buf[DECIMAL_STR_MAX(uint64_t)+1];
854 uint64_t io_weight = cgroup_context_io_weight(c, state);
856 weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state));
858 log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64,
860 } else if (has_blockio)
861 weight = cgroup_context_blkio_weight(c, state);
863 weight = CGROUP_BLKIO_WEIGHT_DEFAULT;
865 xsprintf(buf, "%" PRIu64 "\n", weight);
866 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
868 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
869 "Failed to set blkio.weight: %m");
872 CGroupIODeviceWeight *w;
874 /* FIXME: no way to reset this list */
875 LIST_FOREACH(device_weights, w, c->io_device_weights) {
876 weight = cgroup_weight_io_to_blkio(w->weight);
878 log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s",
879 w->weight, weight, w->path);
881 cgroup_apply_blkio_device_weight(u, w->path, weight);
883 } else if (has_blockio) {
884 CGroupBlockIODeviceWeight *w;
886 /* FIXME: no way to reset this list */
887 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
888 cgroup_apply_blkio_device_weight(u, w->path, w->weight);
892 /* Apply limits and free ones without config. */
894 CGroupIODeviceLimit *l;
896 LIST_FOREACH(device_limits, l, c->io_device_limits) {
897 log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s",
898 l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path);
900 cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]);
902 } else if (has_blockio) {
903 CGroupBlockIODeviceBandwidth *b;
905 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths)
906 cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps);
910 if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) {
911 if (cg_all_unified() > 0) {
912 uint64_t max, swap_max = CGROUP_LIMIT_MAX;
914 if (cgroup_context_has_unified_memory_config(c)) {
916 swap_max = c->memory_swap_max;
918 max = c->memory_limit;
920 if (max != CGROUP_LIMIT_MAX)
921 log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max);
924 cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low);
925 cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high);
926 cgroup_apply_unified_memory_limit(u, "memory.max", max);
927 cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max);
929 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
932 if (cgroup_context_has_unified_memory_config(c)) {
934 log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val);
936 val = c->memory_limit;
938 if (val == CGROUP_LIMIT_MAX)
939 strncpy(buf, "-1\n", sizeof(buf));
941 xsprintf(buf, "%" PRIu64 "\n", val);
943 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
945 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
946 "Failed to set memory.limit_in_bytes: %m");
950 if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) {
951 CGroupDeviceAllow *a;
953 /* Changing the devices list of a populated cgroup
954 * might result in EINVAL, hence ignore EINVAL
957 if (c->device_allow || c->device_policy != CGROUP_AUTO)
958 r = cg_set_attribute("devices", path, "devices.deny", "a");
960 r = cg_set_attribute("devices", path, "devices.allow", "a");
962 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
963 "Failed to reset devices.list: %m");
965 if (c->device_policy == CGROUP_CLOSED ||
966 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
967 static const char auto_devices[] =
968 "/dev/null\0" "rwm\0"
969 "/dev/zero\0" "rwm\0"
970 "/dev/full\0" "rwm\0"
971 "/dev/random\0" "rwm\0"
972 "/dev/urandom\0" "rwm\0"
974 "/dev/ptmx\0" "rwm\0"
975 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
976 "-/run/systemd/inaccessible/chr\0" "rwm\0"
977 "-/run/systemd/inaccessible/blk\0" "rwm\0";
981 NULSTR_FOREACH_PAIR(x, y, auto_devices)
982 whitelist_device(path, x, y);
984 /* PTS (/dev/pts) devices may not be duplicated, but accessed */
985 whitelist_major(path, "pts", 'c', "rw");
988 LIST_FOREACH(device_allow, a, c->device_allow) {
1004 if (path_startswith(a->path, "/dev/"))
1005 whitelist_device(path, a->path, acc);
1006 else if ((val = startswith(a->path, "block-")))
1007 whitelist_major(path, val, 'b', acc);
1008 else if ((val = startswith(a->path, "char-")))
1009 whitelist_major(path, val, 'c', acc);
1011 log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path);
1015 if (apply_mask & CGROUP_MASK_PIDS) {
1018 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1019 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1020 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1021 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1022 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1023 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1024 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1025 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1026 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1027 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1028 * which is desirable so that there's an offical way to release control of the sysctl from
1029 * systemd: set the limit to unbounded and reload. */
1031 if (c->tasks_max != CGROUP_LIMIT_MAX) {
1032 u->manager->sysctl_pid_max_changed = true;
1033 r = procfs_tasks_set_limit(c->tasks_max);
1034 } else if (u->manager->sysctl_pid_max_changed)
1035 r = procfs_tasks_set_limit(TASKS_MAX);
1040 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1041 "Failed to write to tasks limit sysctls: %m");
1044 if (c->tasks_max != CGROUP_LIMIT_MAX) {
1045 char buf[DECIMAL_STR_MAX(uint64_t) + 2];
1047 sprintf(buf, "%" PRIu64 "\n", c->tasks_max);
1048 r = cg_set_attribute("pids", path, "pids.max", buf);
1050 r = cg_set_attribute("pids", path, "pids.max", "max");
1052 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1053 "Failed to set pids.max: %m");
1058 cgroup_apply_firewall(u);
1061 CGroupMask cgroup_context_get_mask(CGroupContext *c) {
1062 CGroupMask mask = 0;
1064 /* Figure out which controllers we need */
1066 if (c->cpu_accounting ||
1067 cgroup_context_has_cpu_weight(c) ||
1068 cgroup_context_has_cpu_shares(c) ||
1069 c->cpu_quota_per_sec_usec != USEC_INFINITY)
1070 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
1072 if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c))
1073 mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
1075 if (c->memory_accounting ||
1076 c->memory_limit != CGROUP_LIMIT_MAX ||
1077 cgroup_context_has_unified_memory_config(c))
1078 mask |= CGROUP_MASK_MEMORY;
1080 if (c->device_allow ||
1081 c->device_policy != CGROUP_AUTO)
1082 mask |= CGROUP_MASK_DEVICES;
1084 if (c->tasks_accounting ||
1085 c->tasks_max != CGROUP_LIMIT_MAX)
1086 mask |= CGROUP_MASK_PIDS;
1091 CGroupMask unit_get_own_mask(Unit *u) {
1094 /* Returns the mask of controllers the unit needs for itself */
1096 c = unit_get_cgroup_context(u);
1100 return cgroup_context_get_mask(c) | unit_get_delegate_mask(u);
1103 CGroupMask unit_get_delegate_mask(Unit *u) {
1106 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1107 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1109 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1111 if (!unit_cgroup_delegate(u))
1114 if (cg_all_unified() <= 0) {
1117 e = unit_get_exec_context(u);
1118 if (e && !exec_context_maintains_privileges(e))
1122 assert_se(c = unit_get_cgroup_context(u));
1123 return c->delegate_controllers;
1126 CGroupMask unit_get_members_mask(Unit *u) {
1129 /* Returns the mask of controllers all of the unit's children require, merged */
1131 if (u->cgroup_members_mask_valid)
1132 return u->cgroup_members_mask;
1134 u->cgroup_members_mask = 0;
1136 if (u->type == UNIT_SLICE) {
1141 HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) {
1146 if (UNIT_DEREF(member->slice) != u)
1149 u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */
1153 u->cgroup_members_mask_valid = true;
1154 return u->cgroup_members_mask;
1157 CGroupMask unit_get_siblings_mask(Unit *u) {
1160 /* Returns the mask of controllers all of the unit's siblings
1161 * require, i.e. the members mask of the unit's parent slice
1162 * if there is one. */
1164 if (UNIT_ISSET(u->slice))
1165 return unit_get_members_mask(UNIT_DEREF(u->slice));
1167 return unit_get_subtree_mask(u); /* we are the top-level slice */
1170 CGroupMask unit_get_subtree_mask(Unit *u) {
1172 /* Returns the mask of this subtree, meaning of the group
1173 * itself and its children. */
1175 return unit_get_own_mask(u) | unit_get_members_mask(u);
1178 CGroupMask unit_get_target_mask(Unit *u) {
1181 /* This returns the cgroup mask of all controllers to enable
1182 * for a specific cgroup, i.e. everything it needs itself,
1183 * plus all that its children need, plus all that its siblings
1184 * need. This is primarily useful on the legacy cgroup
1185 * hierarchy, where we need to duplicate each cgroup in each
1186 * hierarchy that shall be enabled for it. */
1188 mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
1189 mask &= u->manager->cgroup_supported;
1194 CGroupMask unit_get_enable_mask(Unit *u) {
1197 /* This returns the cgroup mask of all controllers to enable
1198 * for the children of a specific cgroup. This is primarily
1199 * useful for the unified cgroup hierarchy, where each cgroup
1200 * controls which controllers are enabled for its children. */
1202 mask = unit_get_members_mask(u);
1203 mask &= u->manager->cgroup_supported;
1208 bool unit_get_needs_bpf(Unit *u) {
1213 c = unit_get_cgroup_context(u);
1217 if (c->ip_accounting ||
1218 c->ip_address_allow ||
1222 /* If any parent slice has an IP access list defined, it applies too */
1223 for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) {
1224 c = unit_get_cgroup_context(p);
1228 if (c->ip_address_allow ||
1236 /* Recurse from a unit up through its containing slices, propagating
1237 * mask bits upward. A unit is also member of itself. */
1238 void unit_update_cgroup_members_masks(Unit *u) {
1244 /* Calculate subtree mask */
1245 m = unit_get_subtree_mask(u);
1247 /* See if anything changed from the previous invocation. If
1248 * not, we're done. */
1249 if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask)
1253 u->cgroup_subtree_mask_valid &&
1254 ((m & ~u->cgroup_subtree_mask) != 0) &&
1255 ((~m & u->cgroup_subtree_mask) == 0);
1257 u->cgroup_subtree_mask = m;
1258 u->cgroup_subtree_mask_valid = true;
1260 if (UNIT_ISSET(u->slice)) {
1261 Unit *s = UNIT_DEREF(u->slice);
1264 /* There's more set now than before. We
1265 * propagate the new mask to the parent's mask
1266 * (not caring if it actually was valid or
1269 s->cgroup_members_mask |= m;
1272 /* There's less set now than before (or we
1273 * don't know), we need to recalculate
1274 * everything, so let's invalidate the
1275 * parent's members mask */
1277 s->cgroup_members_mask_valid = false;
1279 /* And now make sure that this change also hits our
1281 unit_update_cgroup_members_masks(s);
1285 const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) {
1287 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1291 if (u->cgroup_path &&
1292 u->cgroup_realized &&
1293 FLAGS_SET(u->cgroup_realized_mask, mask))
1294 return u->cgroup_path;
1296 u = UNIT_DEREF(u->slice);
1302 static const char *migrate_callback(CGroupMask mask, void *userdata) {
1303 return unit_get_realized_cgroup_path(userdata, mask);
1306 char *unit_default_cgroup_path(Unit *u) {
1307 _cleanup_free_ char *escaped = NULL, *slice = NULL;
1312 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1313 return strdup(u->manager->cgroup_root);
1315 if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) {
1316 r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice);
1321 escaped = cg_escape(u->id);
1326 return strjoin(u->manager->cgroup_root, "/", slice, "/",
1329 return strjoin(u->manager->cgroup_root, "/", escaped);
1332 int unit_set_cgroup_path(Unit *u, const char *path) {
1333 _cleanup_free_ char *p = NULL;
1345 if (streq_ptr(u->cgroup_path, p))
1349 r = hashmap_put(u->manager->cgroup_unit, p, u);
1354 unit_release_cgroup(u);
1356 u->cgroup_path = TAKE_PTR(p);
1361 int unit_watch_cgroup(Unit *u) {
1362 _cleanup_free_ char *events = NULL;
1367 if (!u->cgroup_path)
1370 if (u->cgroup_inotify_wd >= 0)
1373 /* Only applies to the unified hierarchy */
1374 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1376 return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1380 /* Don't watch the root slice, it's pointless. */
1381 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1384 r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops);
1388 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events);
1392 u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY);
1393 if (u->cgroup_inotify_wd < 0) {
1395 if (errno == ENOENT) /* If the directory is already
1396 * gone we don't need to track
1397 * it, so this is not an error */
1400 return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path);
1403 r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u);
1405 return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m");
1410 int unit_pick_cgroup_path(Unit *u) {
1411 _cleanup_free_ char *path = NULL;
1419 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1422 path = unit_default_cgroup_path(u);
1426 r = unit_set_cgroup_path(u, path);
1428 return log_unit_error_errno(u, r, "Control group %s exists already.", path);
1430 return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path);
1435 static int unit_create_cgroup(
1437 CGroupMask target_mask,
1438 CGroupMask enable_mask,
1447 c = unit_get_cgroup_context(u);
1451 /* Figure out our cgroup path */
1452 r = unit_pick_cgroup_path(u);
1456 /* First, create our own group */
1457 r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path);
1459 return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path);
1462 /* Start watching it */
1463 (void) unit_watch_cgroup(u);
1465 /* Preserve enabled controllers in delegated units, adjust others. */
1466 if (created || !unit_cgroup_delegate(u)) {
1468 /* Enable all controllers we need */
1469 r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path);
1471 log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m",
1475 /* Keep track that this is now realized */
1476 u->cgroup_realized = true;
1477 u->cgroup_realized_mask = target_mask;
1478 u->cgroup_enabled_mask = enable_mask;
1479 u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF;
1481 if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) {
1483 /* Then, possibly move things over, but not if
1484 * subgroups may contain processes, which is the case
1485 * for slice and delegation units. */
1486 r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u);
1488 log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path);
1494 static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) {
1495 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1501 if (MANAGER_IS_SYSTEM(u->manager))
1504 if (!u->manager->system_bus)
1507 if (!u->cgroup_path)
1510 /* Determine this unit's cgroup path relative to our cgroup root */
1511 pp = path_startswith(u->cgroup_path, u->manager->cgroup_root);
1515 pp = strjoina("/", pp, suffix_path);
1516 path_simplify(pp, false);
1518 r = sd_bus_call_method(u->manager->system_bus,
1519 "org.freedesktop.systemd1",
1520 "/org/freedesktop/systemd1",
1521 "org.freedesktop.systemd1.Manager",
1522 "AttachProcessesToUnit",
1525 NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid);
1527 return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r));
1532 int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) {
1533 CGroupMask delegated_mask;
1541 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1544 if (set_isempty(pids))
1547 r = unit_realize_cgroup(u);
1551 if (isempty(suffix_path))
1554 p = strjoina(u->cgroup_path, "/", suffix_path);
1556 delegated_mask = unit_get_delegate_mask(u);
1559 SET_FOREACH(pidp, pids, i) {
1560 pid_t pid = PTR_TO_PID(pidp);
1563 /* First, attach the PID to the main cgroup hierarchy */
1564 q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid);
1566 log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p);
1568 if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) {
1571 /* If we are in a user instance, and we can't move the process ourselves due to
1572 * permission problems, let's ask the system instance about it instead. Since it's more
1573 * privileged it might be able to move the process across the leaves of a subtree who's
1574 * top node is not owned by us. */
1576 z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path);
1578 log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p);
1580 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
1584 r = q; /* Remember first error */
1589 q = cg_all_unified();
1595 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
1596 * innermost realized one */
1598 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
1599 CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
1600 const char *realized;
1602 if (!(u->manager->cgroup_supported & bit))
1605 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
1606 if (delegated_mask & u->cgroup_realized_mask & bit) {
1607 q = cg_attach(cgroup_controller_to_string(c), p, pid);
1609 continue; /* Success! */
1611 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",
1612 pid, p, cgroup_controller_to_string(c));
1615 /* So this controller is either not delegate or realized, or something else weird happened. In
1616 * that case let's attach the PID at least to the closest cgroup up the tree that is
1618 realized = unit_get_realized_cgroup_path(u, bit);
1620 continue; /* Not even realized in the root slice? Then let's not bother */
1622 q = cg_attach(cgroup_controller_to_string(c), realized, pid);
1624 log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m",
1625 pid, realized, cgroup_controller_to_string(c));
1632 static void cgroup_xattr_apply(Unit *u) {
1633 char ids[SD_ID128_STRING_MAX];
1638 if (!MANAGER_IS_SYSTEM(u->manager))
1641 if (sd_id128_is_null(u->invocation_id))
1644 r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
1645 "trusted.invocation_id",
1646 sd_id128_to_string(u->invocation_id, ids), 32,
1649 log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path);
1652 static bool unit_has_mask_realized(
1654 CGroupMask target_mask,
1655 CGroupMask enable_mask,
1660 return u->cgroup_realized &&
1661 u->cgroup_realized_mask == target_mask &&
1662 u->cgroup_enabled_mask == enable_mask &&
1663 ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) ||
1664 (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF));
1667 static void unit_add_to_cgroup_realize_queue(Unit *u) {
1670 if (u->in_cgroup_realize_queue)
1673 LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
1674 u->in_cgroup_realize_queue = true;
1677 static void unit_remove_from_cgroup_realize_queue(Unit *u) {
1680 if (!u->in_cgroup_realize_queue)
1683 LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
1684 u->in_cgroup_realize_queue = false;
1687 /* Check if necessary controllers and attributes for a unit are in place.
1689 * If so, do nothing.
1690 * If not, create paths, move processes over, and set attributes.
1692 * Returns 0 on success and < 0 on failure. */
1693 static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
1694 CGroupMask target_mask, enable_mask;
1695 bool needs_bpf, apply_bpf;
1700 unit_remove_from_cgroup_realize_queue(u);
1702 target_mask = unit_get_target_mask(u);
1703 enable_mask = unit_get_enable_mask(u);
1704 needs_bpf = unit_get_needs_bpf(u);
1706 if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf))
1709 /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously
1710 * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it
1711 * this will trickle down properly to cgroupfs. */
1712 apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF;
1714 /* First, realize parents */
1715 if (UNIT_ISSET(u->slice)) {
1716 r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
1721 /* And then do the real work */
1722 r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf);
1726 /* Finally, apply the necessary attributes. */
1727 cgroup_context_apply(u, target_mask, apply_bpf, state);
1728 cgroup_xattr_apply(u);
1733 unsigned manager_dispatch_cgroup_realize_queue(Manager *m) {
1741 state = manager_state(m);
1743 while ((i = m->cgroup_realize_queue)) {
1744 assert(i->in_cgroup_realize_queue);
1746 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) {
1747 /* Maybe things changed, and the unit is not actually active anymore? */
1748 unit_remove_from_cgroup_realize_queue(i);
1752 r = unit_realize_cgroup_now(i, state);
1754 log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id);
1762 static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) {
1765 /* This adds the siblings of the specified unit and the
1766 * siblings of all parent units to the cgroup queue. (But
1767 * neither the specified unit itself nor the parents.) */
1769 while ((slice = UNIT_DEREF(u->slice))) {
1774 HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) {
1778 /* Skip units that have a dependency on the slice
1779 * but aren't actually in it. */
1780 if (UNIT_DEREF(m->slice) != slice)
1783 /* No point in doing cgroup application for units
1784 * without active processes. */
1785 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
1788 /* If the unit doesn't need any new controllers
1789 * and has current ones realized, it doesn't need
1791 if (unit_has_mask_realized(m,
1792 unit_get_target_mask(m),
1793 unit_get_enable_mask(m),
1794 unit_get_needs_bpf(m)))
1797 unit_add_to_cgroup_realize_queue(m);
1804 int unit_realize_cgroup(Unit *u) {
1807 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1810 /* So, here's the deal: when realizing the cgroups for this
1811 * unit, we need to first create all parents, but there's more
1812 * actually: for the weight-based controllers we also need to
1813 * make sure that all our siblings (i.e. units that are in the
1814 * same slice as we are) have cgroups, too. Otherwise, things
1815 * would become very uneven as each of their processes would
1816 * get as much resources as all our group together. This call
1817 * will synchronously create the parent cgroups, but will
1818 * defer work on the siblings to the next event loop
1821 /* Add all sibling slices to the cgroup queue. */
1822 unit_add_siblings_to_cgroup_realize_queue(u);
1824 /* And realize this one now (and apply the values) */
1825 return unit_realize_cgroup_now(u, manager_state(u->manager));
1828 void unit_release_cgroup(Unit *u) {
1831 /* Forgets all cgroup details for this cgroup */
1833 if (u->cgroup_path) {
1834 (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
1835 u->cgroup_path = mfree(u->cgroup_path);
1838 if (u->cgroup_inotify_wd >= 0) {
1839 if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0)
1840 log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id);
1842 (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd));
1843 u->cgroup_inotify_wd = -1;
1847 void unit_prune_cgroup(Unit *u) {
1853 /* Removes the cgroup, if empty and possible, and stops watching it. */
1855 if (!u->cgroup_path)
1858 (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */
1860 is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE);
1862 r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice);
1864 log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path);
1871 unit_release_cgroup(u);
1873 u->cgroup_realized = false;
1874 u->cgroup_realized_mask = 0;
1875 u->cgroup_enabled_mask = 0;
1878 int unit_search_main_pid(Unit *u, pid_t *ret) {
1879 _cleanup_fclose_ FILE *f = NULL;
1880 pid_t pid = 0, npid, mypid;
1886 if (!u->cgroup_path)
1889 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f);
1893 mypid = getpid_cached();
1894 while (cg_read_pid(f, &npid) > 0) {
1900 /* Ignore processes that aren't our kids */
1901 if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid)
1905 /* Dang, there's more than one daemonized PID
1906 in this group, so we don't know what process
1907 is the main process. */
1918 static int unit_watch_pids_in_path(Unit *u, const char *path) {
1919 _cleanup_closedir_ DIR *d = NULL;
1920 _cleanup_fclose_ FILE *f = NULL;
1926 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
1932 while ((r = cg_read_pid(f, &pid)) > 0) {
1933 r = unit_watch_pid(u, pid);
1934 if (r < 0 && ret >= 0)
1938 if (r < 0 && ret >= 0)
1942 r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
1949 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1950 _cleanup_free_ char *p = NULL;
1952 p = strjoin(path, "/", fn);
1958 r = unit_watch_pids_in_path(u, p);
1959 if (r < 0 && ret >= 0)
1963 if (r < 0 && ret >= 0)
1970 int unit_synthesize_cgroup_empty_event(Unit *u) {
1975 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
1976 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
1977 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
1979 if (!u->cgroup_path)
1982 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1985 if (r > 0) /* On unified we have reliable notifications, and don't need this */
1988 if (!set_isempty(u->pids))
1991 unit_add_to_cgroup_empty_queue(u);
1995 int unit_watch_all_pids(Unit *u) {
2000 /* Adds all PIDs from our cgroup to the set of PIDs we
2001 * watch. This is a fallback logic for cases where we do not
2002 * get reliable cgroup empty notifications: we try to use
2003 * SIGCHLD as replacement. */
2005 if (!u->cgroup_path)
2008 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
2011 if (r > 0) /* On unified we can use proper notifications */
2014 return unit_watch_pids_in_path(u, u->cgroup_path);
2017 static int on_cgroup_empty_event(sd_event_source *s, void *userdata) {
2018 Manager *m = userdata;
2025 u = m->cgroup_empty_queue;
2029 assert(u->in_cgroup_empty_queue);
2030 u->in_cgroup_empty_queue = false;
2031 LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u);
2033 if (m->cgroup_empty_queue) {
2034 /* More stuff queued, let's make sure we remain enabled */
2035 r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT);
2037 log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m");
2040 unit_add_to_gc_queue(u);
2042 if (UNIT_VTABLE(u)->notify_cgroup_empty)
2043 UNIT_VTABLE(u)->notify_cgroup_empty(u);
2048 void unit_add_to_cgroup_empty_queue(Unit *u) {
2053 /* Note that there are four different ways how cgroup empty events reach us:
2055 * 1. On the unified hierarchy we get an inotify event on the cgroup
2057 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2059 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2061 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2062 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2064 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2065 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2066 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2067 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2068 * case for scope units). */
2070 if (u->in_cgroup_empty_queue)
2073 /* Let's verify that the cgroup is really empty */
2074 if (!u->cgroup_path)
2076 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
2078 log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path);
2084 LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
2085 u->in_cgroup_empty_queue = true;
2087 /* Trigger the defer event */
2088 r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT);
2090 log_debug_errno(r, "Failed to enable cgroup empty event source: %m");
2093 static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
2094 Manager *m = userdata;
2101 union inotify_event_buffer buffer;
2102 struct inotify_event *e;
2105 l = read(fd, &buffer, sizeof(buffer));
2107 if (IN_SET(errno, EINTR, EAGAIN))
2110 return log_error_errno(errno, "Failed to read control group inotify events: %m");
2113 FOREACH_INOTIFY_EVENT(e, buffer, l) {
2117 /* Queue overflow has no watch descriptor */
2120 if (e->mask & IN_IGNORED)
2121 /* The watch was just removed */
2124 u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd));
2125 if (!u) /* Not that inotify might deliver
2126 * events for a watch even after it
2127 * was removed, because it was queued
2128 * before the removal. Let's ignore
2129 * this here safely. */
2132 unit_add_to_cgroup_empty_queue(u);
2138 int manager_setup_cgroup(Manager *m) {
2139 _cleanup_free_ char *path = NULL;
2140 const char *scope_path;
2143 #if 0 /// UNNEEDED by elogind
2149 /* 1. Determine hierarchy */
2150 m->cgroup_root = mfree(m->cgroup_root);
2151 #if 0 /// elogind is not init and must therefore search for PID 1 instead of self.
2152 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
2154 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &m->cgroup_root);
2157 return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
2159 #if 0 /// elogind does not support systemd scopes and slices
2160 /* Chop off the init scope, if we are already located in it */
2161 e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
2163 /* LEGACY: Also chop off the system slice if we are in
2164 * it. This is to support live upgrades from older systemd
2165 * versions where PID 1 was moved there. Also see
2166 * cg_get_root_path(). */
2167 if (!e && MANAGER_IS_SYSTEM(m)) {
2168 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
2170 e = endswith(m->cgroup_root, "/system"); /* even more legacy */
2176 log_debug_elogind("Cgroup Controller \"%s\" -> root \"%s\"",
2177 SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root);
2178 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
2179 * easily prepend it everywhere. */
2180 delete_trailing_chars(m->cgroup_root, "/");
2183 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
2185 return log_error_errno(r, "Cannot find cgroup mount point: %m");
2187 r = cg_unified_flush();
2189 return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m");
2191 all_unified = cg_all_unified();
2192 if (all_unified < 0)
2193 return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m");
2194 if (all_unified > 0)
2195 log_debug("Unified cgroup hierarchy is located at %s.", path);
2197 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
2199 return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m");
2201 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path);
2203 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path);
2206 #if 0 /// elogind is not init, and does not install the agent here.
2207 /* 3. Allocate cgroup empty defer event source */
2208 m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source);
2209 r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m);
2211 return log_error_errno(r, "Failed to create cgroup empty event source: %m");
2213 r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5);
2215 return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m");
2217 r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF);
2219 return log_error_errno(r, "Failed to disable cgroup empty event source: %m");
2221 (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty");
2223 /* 4. Install notifier inotify object, or agent */
2224 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
2226 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
2228 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
2229 safe_close(m->cgroup_inotify_fd);
2231 m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
2232 if (m->cgroup_inotify_fd < 0)
2233 return log_error_errno(errno, "Failed to create control group inotify object: %m");
2235 r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m);
2237 return log_error_errno(r, "Failed to watch control group inotify object: %m");
2239 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
2240 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
2241 r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4);
2243 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
2245 (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify");
2247 } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) {
2249 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
2250 * since it does not generate events when control groups with children run empty. */
2252 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
2254 log_warning_errno(r, "Failed to install release agent, ignoring: %m");
2256 log_debug("Installed release agent.");
2258 log_debug("Release agent already installed.");
2261 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
2262 scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
2263 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
2265 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
2266 r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
2268 log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m");
2271 * This method is in core, and normally called by systemd
2272 * being init. As elogind is never init, we can not install
2273 * our agent here. We do so when mounting our cgroup file
2274 * system, so only if elogind is its own tiny controller.
2275 * Further, elogind is not meant to run in systemd init scope. */
2276 if (MANAGER_IS_SYSTEM(m))
2277 // we are our own cgroup controller
2278 scope_path = strjoina("");
2279 else if (streq(m->cgroup_root, "/elogind"))
2280 // root already is our cgroup
2281 scope_path = strjoina(m->cgroup_root);
2283 // we have to create our own group
2284 scope_path = strjoina(m->cgroup_root, "/elogind");
2285 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
2287 log_debug_elogind("Created control group \"%s\"", scope_path);
2289 /* 6. And pin it, so that it cannot be unmounted */
2290 safe_close(m->pin_cgroupfs_fd);
2291 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
2292 if (m->pin_cgroupfs_fd < 0)
2293 return log_error_errno(errno, "Failed to open pin file: %m");
2295 #if 0 /// this is from the cgroup migration above that elogind does not need.
2296 } else if (r < 0 && !m->test_run_flags)
2297 return log_error_errno(r, "Failed to create %s control group: %m", scope_path);
2300 /* 7. Always enable hierarchical support if it exists... */
2301 if (!all_unified && m->test_run_flags == 0)
2302 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
2304 /* 8. Figure out which controllers are supported, and log about it */
2305 r = cg_mask_supported(&m->cgroup_supported);
2307 return log_error_errno(r, "Failed to determine supported controllers: %m");
2308 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++)
2309 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c)));
2314 void manager_shutdown_cgroup(Manager *m, bool delete) {
2317 #if 0 /// elogind is not init
2318 /* We can't really delete the group, since we are in it. But
2320 if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL)
2321 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
2323 m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source);
2325 m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit);
2327 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
2328 m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd);
2331 m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
2333 m->cgroup_root = mfree(m->cgroup_root);
2336 #if 0 /// UNNEEDED by elogind
2337 Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
2344 u = hashmap_get(m->cgroup_unit, cgroup);
2348 p = strdupa(cgroup);
2352 e = strrchr(p, '/');
2354 return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE);
2358 u = hashmap_get(m->cgroup_unit, p);
2364 Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) {
2365 _cleanup_free_ char *cgroup = NULL;
2369 if (!pid_is_valid(pid))
2372 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0)
2375 return manager_get_unit_by_cgroup(m, cgroup);
2378 Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
2383 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
2384 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
2385 * relevant one as children of the process will be assigned to that one, too, before all else. */
2387 if (!pid_is_valid(pid))
2390 if (pid == getpid_cached())
2391 return hashmap_get(m->units, SPECIAL_INIT_SCOPE);
2393 u = manager_get_unit_by_pid_cgroup(m, pid);
2397 u = hashmap_get(m->watch_pids, PID_TO_PTR(pid));
2401 array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid));
2409 #if 0 /// elogind must substitute this with its own variant
2410 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
2416 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
2417 * or from the --system instance */
2419 log_debug("Got cgroup empty notification for: %s", cgroup);
2421 u = manager_get_unit_by_cgroup(m, cgroup);
2425 unit_add_to_cgroup_empty_queue(u);
2429 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
2435 log_debug("Got cgroup empty notification for: %s", cgroup);
2437 s = hashmap_get(m->sessions, cgroup);
2440 session_finalize(s);
2443 log_warning("Session not found: %s", cgroup);
2448 #if 0 /// UNNEEDED by elogind
2449 int unit_get_memory_current(Unit *u, uint64_t *ret) {
2450 _cleanup_free_ char *v = NULL;
2456 if (!UNIT_CGROUP_BOOL(u, memory_accounting))
2459 if (!u->cgroup_path)
2462 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
2463 if (unit_has_root_cgroup(u))
2464 return procfs_memory_get_current(ret);
2466 if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
2469 r = cg_all_unified();
2473 r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v);
2475 r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v);
2481 return safe_atou64(v, ret);
2484 int unit_get_tasks_current(Unit *u, uint64_t *ret) {
2485 _cleanup_free_ char *v = NULL;
2491 if (!UNIT_CGROUP_BOOL(u, tasks_accounting))
2494 if (!u->cgroup_path)
2497 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
2498 if (unit_has_root_cgroup(u))
2499 return procfs_tasks_get_current(ret);
2501 if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0)
2504 r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v);
2510 return safe_atou64(v, ret);
2513 static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) {
2514 _cleanup_free_ char *v = NULL;
2521 if (!u->cgroup_path)
2524 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
2525 if (unit_has_root_cgroup(u))
2526 return procfs_cpu_get_usage(ret);
2528 r = cg_all_unified();
2532 _cleanup_free_ char *val = NULL;
2535 if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0)
2538 r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val);
2541 if (IN_SET(r, -ENOENT, -ENXIO))
2544 r = safe_atou64(val, &us);
2548 ns = us * NSEC_PER_USEC;
2550 if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0)
2553 r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v);
2559 r = safe_atou64(v, &ns);
2568 int unit_get_cpu_usage(Unit *u, nsec_t *ret) {
2574 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
2575 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
2576 * call this function with a NULL return value. */
2578 if (!UNIT_CGROUP_BOOL(u, cpu_accounting))
2581 r = unit_get_cpu_usage_raw(u, &ns);
2582 if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) {
2583 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
2587 *ret = u->cpu_usage_last;
2593 if (ns > u->cpu_usage_base)
2594 ns -= u->cpu_usage_base;
2598 u->cpu_usage_last = ns;
2605 int unit_get_ip_accounting(
2607 CGroupIPAccountingMetric metric,
2614 assert(metric >= 0);
2615 assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX);
2618 if (!UNIT_CGROUP_BOOL(u, ip_accounting))
2621 fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ?
2622 u->ip_accounting_ingress_map_fd :
2623 u->ip_accounting_egress_map_fd;
2627 if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES))
2628 r = bpf_firewall_read_accounting(fd, &value, NULL);
2630 r = bpf_firewall_read_accounting(fd, NULL, &value);
2634 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
2635 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
2636 * ip_accounting_extra[] field, and add them in here transparently. */
2638 *ret = value + u->ip_accounting_extra[metric];
2643 int unit_reset_cpu_accounting(Unit *u) {
2649 u->cpu_usage_last = NSEC_INFINITY;
2651 r = unit_get_cpu_usage_raw(u, &ns);
2653 u->cpu_usage_base = 0;
2657 u->cpu_usage_base = ns;
2661 int unit_reset_ip_accounting(Unit *u) {
2666 if (u->ip_accounting_ingress_map_fd >= 0)
2667 r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd);
2669 if (u->ip_accounting_egress_map_fd >= 0)
2670 q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd);
2672 zero(u->ip_accounting_extra);
2674 return r < 0 ? r : q;
2677 void unit_invalidate_cgroup(Unit *u, CGroupMask m) {
2680 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2686 /* always invalidate compat pairs together */
2687 if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO))
2688 m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
2690 if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT))
2691 m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT;
2693 if ((u->cgroup_realized_mask & m) == 0) /* NOP? */
2696 u->cgroup_realized_mask &= ~m;
2697 unit_add_to_cgroup_realize_queue(u);
2700 void unit_invalidate_cgroup_bpf(Unit *u) {
2703 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2706 if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) /* NOP? */
2709 u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED;
2710 unit_add_to_cgroup_realize_queue(u);
2712 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
2713 * list of our children includes our own. */
2714 if (u->type == UNIT_SLICE) {
2719 HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) {
2723 if (UNIT_DEREF(member->slice) != u)
2726 unit_invalidate_cgroup_bpf(member);
2731 bool unit_cgroup_delegate(Unit *u) {
2736 if (!UNIT_VTABLE(u)->can_delegate)
2739 c = unit_get_cgroup_context(u);
2746 void manager_invalidate_startup_units(Manager *m) {
2752 SET_FOREACH(u, m->startup_units, i)
2753 unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
2756 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
2757 [CGROUP_AUTO] = "auto",
2758 [CGROUP_CLOSED] = "closed",
2759 [CGROUP_STRICT] = "strict",
2762 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);