2 This file is part of systemd.
4 Copyright 2013 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
23 #include "alloc-util.h"
24 #include "cgroup-util.h"
29 #include "parse-util.h"
30 #include "path-util.h"
31 #include "process-util.h"
32 //#include "special.h"
33 #include "string-table.h"
34 #include "string-util.h"
35 #include "stdio-util.h"
37 #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
39 #if 0 /// UNNEEDED by elogind
40 static void cgroup_compat_warn(void) {
41 static bool cgroup_compat_warned = false;
43 if (cgroup_compat_warned)
46 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details.");
47 cgroup_compat_warned = true;
50 #define log_cgroup_compat(unit, fmt, ...) do { \
51 cgroup_compat_warn(); \
52 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
55 void cgroup_context_init(CGroupContext *c) {
58 /* Initialize everything to the kernel defaults, assuming the
59 * structure is preinitialized to 0 */
61 c->cpu_shares = CGROUP_CPU_SHARES_INVALID;
62 c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID;
63 c->cpu_quota_per_sec_usec = USEC_INFINITY;
65 c->memory_high = CGROUP_LIMIT_MAX;
66 c->memory_max = CGROUP_LIMIT_MAX;
67 c->memory_swap_max = CGROUP_LIMIT_MAX;
69 c->memory_limit = CGROUP_LIMIT_MAX;
71 c->io_weight = CGROUP_WEIGHT_INVALID;
72 c->startup_io_weight = CGROUP_WEIGHT_INVALID;
74 c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
75 c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
77 c->tasks_max = (uint64_t) -1;
80 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
84 LIST_REMOVE(device_allow, c->device_allow, a);
89 void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) {
93 LIST_REMOVE(device_weights, c->io_device_weights, w);
98 void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) {
102 LIST_REMOVE(device_limits, c->io_device_limits, l);
107 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
111 LIST_REMOVE(device_weights, c->blockio_device_weights, w);
116 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
120 LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
125 void cgroup_context_done(CGroupContext *c) {
128 while (c->io_device_weights)
129 cgroup_context_free_io_device_weight(c, c->io_device_weights);
131 while (c->io_device_limits)
132 cgroup_context_free_io_device_limit(c, c->io_device_limits);
134 while (c->blockio_device_weights)
135 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
137 while (c->blockio_device_bandwidths)
138 cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
140 while (c->device_allow)
141 cgroup_context_free_device_allow(c, c->device_allow);
144 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
145 CGroupIODeviceLimit *il;
146 CGroupIODeviceWeight *iw;
147 CGroupBlockIODeviceBandwidth *b;
148 CGroupBlockIODeviceWeight *w;
149 CGroupDeviceAllow *a;
150 char u[FORMAT_TIMESPAN_MAX];
155 prefix = strempty(prefix);
158 "%sCPUAccounting=%s\n"
159 "%sIOAccounting=%s\n"
160 "%sBlockIOAccounting=%s\n"
161 "%sMemoryAccounting=%s\n"
162 "%sTasksAccounting=%s\n"
163 "%sCPUShares=%" PRIu64 "\n"
164 "%sStartupCPUShares=%" PRIu64 "\n"
165 "%sCPUQuotaPerSecSec=%s\n"
166 "%sIOWeight=%" PRIu64 "\n"
167 "%sStartupIOWeight=%" PRIu64 "\n"
168 "%sBlockIOWeight=%" PRIu64 "\n"
169 "%sStartupBlockIOWeight=%" PRIu64 "\n"
170 "%sMemoryLow=%" PRIu64 "\n"
171 "%sMemoryHigh=%" PRIu64 "\n"
172 "%sMemoryMax=%" PRIu64 "\n"
173 "%sMemorySwapMax=%" PRIu64 "\n"
174 "%sMemoryLimit=%" PRIu64 "\n"
175 "%sTasksMax=%" PRIu64 "\n"
176 "%sDevicePolicy=%s\n"
178 prefix, yes_no(c->cpu_accounting),
179 prefix, yes_no(c->io_accounting),
180 prefix, yes_no(c->blockio_accounting),
181 prefix, yes_no(c->memory_accounting),
182 prefix, yes_no(c->tasks_accounting),
183 prefix, c->cpu_shares,
184 prefix, c->startup_cpu_shares,
185 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
186 prefix, c->io_weight,
187 prefix, c->startup_io_weight,
188 prefix, c->blockio_weight,
189 prefix, c->startup_blockio_weight,
190 prefix, c->memory_low,
191 prefix, c->memory_high,
192 prefix, c->memory_max,
193 prefix, c->memory_swap_max,
194 prefix, c->memory_limit,
195 prefix, c->tasks_max,
196 prefix, cgroup_device_policy_to_string(c->device_policy),
197 prefix, yes_no(c->delegate));
199 LIST_FOREACH(device_allow, a, c->device_allow)
201 "%sDeviceAllow=%s %s%s%s\n",
204 a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
206 LIST_FOREACH(device_weights, iw, c->io_device_weights)
208 "%sIODeviceWeight=%s %" PRIu64,
213 LIST_FOREACH(device_limits, il, c->io_device_limits) {
214 char buf[FORMAT_BYTES_MAX];
215 CGroupIOLimitType type;
217 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
218 if (il->limits[type] != cgroup_io_limit_defaults[type])
222 cgroup_io_limit_type_to_string(type),
224 format_bytes(buf, sizeof(buf), il->limits[type]));
227 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
229 "%sBlockIODeviceWeight=%s %" PRIu64,
234 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
235 char buf[FORMAT_BYTES_MAX];
237 if (b->rbps != CGROUP_LIMIT_MAX)
239 "%sBlockIOReadBandwidth=%s %s\n",
242 format_bytes(buf, sizeof(buf), b->rbps));
243 if (b->wbps != CGROUP_LIMIT_MAX)
245 "%sBlockIOWriteBandwidth=%s %s\n",
248 format_bytes(buf, sizeof(buf), b->wbps));
252 static int lookup_block_device(const char *p, dev_t *dev) {
261 return log_warning_errno(errno, "Couldn't stat device %s: %m", p);
263 if (S_ISBLK(st.st_mode))
265 else if (major(st.st_dev) != 0) {
266 /* If this is not a device node then find the block
267 * device this file is stored on */
270 /* If this is a partition, try to get the originating
272 block_get_whole_disk(*dev, dev);
274 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p);
281 static int whitelist_device(const char *path, const char *node, const char *acc) {
282 char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
289 if (stat(node, &st) < 0) {
290 log_warning("Couldn't stat device %s", node);
294 if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
295 log_warning("%s is not a device.", node);
301 S_ISCHR(st.st_mode) ? 'c' : 'b',
302 major(st.st_rdev), minor(st.st_rdev),
305 r = cg_set_attribute("devices", path, "devices.allow", buf);
307 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
308 "Failed to set devices.allow on %s: %m", path);
313 static int whitelist_major(const char *path, const char *name, char type, const char *acc) {
314 _cleanup_fclose_ FILE *f = NULL;
321 assert(type == 'b' || type == 'c');
323 f = fopen("/proc/devices", "re");
325 return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
327 FOREACH_LINE(line, f, goto fail) {
328 char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w;
333 if (type == 'c' && streq(line, "Character devices:")) {
338 if (type == 'b' && streq(line, "Block devices:")) {
353 w = strpbrk(p, WHITESPACE);
358 r = safe_atou(p, &maj);
365 w += strspn(w, WHITESPACE);
367 if (fnmatch(name, w, 0) != 0)
376 r = cg_set_attribute("devices", path, "devices.allow", buf);
378 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
379 "Failed to set devices.allow on %s: %m", path);
385 log_warning_errno(errno, "Failed to read /proc/devices: %m");
389 static bool cgroup_context_has_io_config(CGroupContext *c) {
390 return c->io_accounting ||
391 c->io_weight != CGROUP_WEIGHT_INVALID ||
392 c->startup_io_weight != CGROUP_WEIGHT_INVALID ||
393 c->io_device_weights ||
397 static bool cgroup_context_has_blockio_config(CGroupContext *c) {
398 return c->blockio_accounting ||
399 c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
400 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
401 c->blockio_device_weights ||
402 c->blockio_device_bandwidths;
405 static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) {
406 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
407 c->startup_io_weight != CGROUP_WEIGHT_INVALID)
408 return c->startup_io_weight;
409 else if (c->io_weight != CGROUP_WEIGHT_INVALID)
412 return CGROUP_WEIGHT_DEFAULT;
415 static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) {
416 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
417 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
418 return c->startup_blockio_weight;
419 else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
420 return c->blockio_weight;
422 return CGROUP_BLKIO_WEIGHT_DEFAULT;
425 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) {
426 return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT,
427 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
430 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) {
431 return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT,
432 CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX);
435 static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) {
436 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
440 r = lookup_block_device(dev_path, &dev);
444 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight);
445 r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf);
447 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
448 "Failed to set io.weight: %m");
451 static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) {
452 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
456 r = lookup_block_device(dev_path, &dev);
460 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight);
461 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf);
463 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
464 "Failed to set blkio.weight_device: %m");
467 static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) {
468 char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)];
469 char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
470 CGroupIOLimitType type;
475 r = lookup_block_device(dev_path, &dev);
479 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) {
480 if (limits[type] != cgroup_io_limit_defaults[type]) {
481 xsprintf(limit_bufs[type], "%" PRIu64, limits[type]);
484 xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0");
488 xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev),
489 limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX],
490 limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]);
491 r = cg_set_attribute("io", u->cgroup_path, "io.max", buf);
493 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
494 "Failed to set io.max: %m");
498 static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) {
499 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
504 r = lookup_block_device(dev_path, &dev);
508 if (rbps != CGROUP_LIMIT_MAX)
510 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps);
511 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf);
513 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
514 "Failed to set blkio.throttle.read_bps_device: %m");
516 if (wbps != CGROUP_LIMIT_MAX)
518 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps);
519 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf);
521 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
522 "Failed to set blkio.throttle.write_bps_device: %m");
527 static bool cgroup_context_has_unified_memory_config(CGroupContext *c) {
528 return c->memory_low > 0 || c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || c->memory_swap_max != CGROUP_LIMIT_MAX;
531 static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) {
532 char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
535 if (v != CGROUP_LIMIT_MAX)
536 xsprintf(buf, "%" PRIu64 "\n", v);
538 r = cg_set_attribute("memory", u->cgroup_path, file, buf);
540 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
541 "Failed to set %s: %m", file);
544 static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) {
552 c = unit_get_cgroup_context(u);
553 path = u->cgroup_path;
561 /* Some cgroup attributes are not supported on the root cgroup,
562 * hence silently ignore */
563 is_root = isempty(path) || path_equal(path, "/");
565 /* Make sure we don't try to display messages with an empty path. */
568 /* We generally ignore errors caused by read-only mounted
569 * cgroup trees (assuming we are running in a container then),
570 * and missing cgroups, i.e. EROFS and ENOENT. */
572 if ((mask & CGROUP_MASK_CPU) && !is_root) {
573 char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1];
575 sprintf(buf, "%" PRIu64 "\n",
576 IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID ? c->startup_cpu_shares :
577 c->cpu_shares != CGROUP_CPU_SHARES_INVALID ? c->cpu_shares : CGROUP_CPU_SHARES_DEFAULT);
578 r = cg_set_attribute("cpu", path, "cpu.shares", buf);
580 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
581 "Failed to set cpu.shares: %m");
583 sprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
584 r = cg_set_attribute("cpu", path, "cpu.cfs_period_us", buf);
586 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
587 "Failed to set cpu.cfs_period_us: %m");
589 if (c->cpu_quota_per_sec_usec != USEC_INFINITY) {
590 sprintf(buf, USEC_FMT "\n", c->cpu_quota_per_sec_usec * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC);
591 r = cg_set_attribute("cpu", path, "cpu.cfs_quota_us", buf);
593 r = cg_set_attribute("cpu", path, "cpu.cfs_quota_us", "-1");
595 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
596 "Failed to set cpu.cfs_quota_us: %m");
599 if (mask & CGROUP_MASK_IO) {
600 bool has_io = cgroup_context_has_io_config(c);
601 bool has_blockio = cgroup_context_has_blockio_config(c);
604 char buf[8+DECIMAL_STR_MAX(uint64_t)+1];
608 weight = cgroup_context_io_weight(c, state);
609 else if (has_blockio) {
610 uint64_t blkio_weight = cgroup_context_blkio_weight(c, state);
612 weight = cgroup_weight_blkio_to_io(blkio_weight);
614 log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64,
615 blkio_weight, weight);
617 weight = CGROUP_WEIGHT_DEFAULT;
619 xsprintf(buf, "default %" PRIu64 "\n", weight);
620 r = cg_set_attribute("io", path, "io.weight", buf);
622 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
623 "Failed to set io.weight: %m");
626 CGroupIODeviceWeight *w;
628 /* FIXME: no way to reset this list */
629 LIST_FOREACH(device_weights, w, c->io_device_weights)
630 cgroup_apply_io_device_weight(u, w->path, w->weight);
631 } else if (has_blockio) {
632 CGroupBlockIODeviceWeight *w;
634 /* FIXME: no way to reset this list */
635 LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
636 weight = cgroup_weight_blkio_to_io(w->weight);
638 log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s",
639 w->weight, weight, w->path);
641 cgroup_apply_io_device_weight(u, w->path, weight);
646 /* Apply limits and free ones without config. */
648 CGroupIODeviceLimit *l, *next;
650 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
651 if (!cgroup_apply_io_device_limit(u, l->path, l->limits))
652 cgroup_context_free_io_device_limit(c, l);
654 } else if (has_blockio) {
655 CGroupBlockIODeviceBandwidth *b, *next;
657 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
658 uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX];
659 CGroupIOLimitType type;
661 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
662 limits[type] = cgroup_io_limit_defaults[type];
664 limits[CGROUP_IO_RBPS_MAX] = b->rbps;
665 limits[CGROUP_IO_WBPS_MAX] = b->wbps;
667 log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s",
668 b->rbps, b->wbps, b->path);
670 if (!cgroup_apply_io_device_limit(u, b->path, limits))
671 cgroup_context_free_blockio_device_bandwidth(c, b);
676 if (mask & CGROUP_MASK_BLKIO) {
677 bool has_io = cgroup_context_has_io_config(c);
678 bool has_blockio = cgroup_context_has_blockio_config(c);
681 char buf[DECIMAL_STR_MAX(uint64_t)+1];
685 weight = cgroup_context_blkio_weight(c, state);
687 uint64_t io_weight = cgroup_context_io_weight(c, state);
689 weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state));
691 log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64,
694 weight = CGROUP_BLKIO_WEIGHT_DEFAULT;
696 xsprintf(buf, "%" PRIu64 "\n", weight);
697 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
699 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
700 "Failed to set blkio.weight: %m");
703 CGroupBlockIODeviceWeight *w;
705 /* FIXME: no way to reset this list */
706 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
707 cgroup_apply_blkio_device_weight(u, w->path, w->weight);
709 CGroupIODeviceWeight *w;
711 /* FIXME: no way to reset this list */
712 LIST_FOREACH(device_weights, w, c->io_device_weights) {
713 weight = cgroup_weight_io_to_blkio(w->weight);
715 log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s",
716 w->weight, weight, w->path);
718 cgroup_apply_blkio_device_weight(u, w->path, weight);
723 /* Apply limits and free ones without config. */
725 CGroupBlockIODeviceBandwidth *b, *next;
727 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
728 if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps))
729 cgroup_context_free_blockio_device_bandwidth(c, b);
732 CGroupIODeviceLimit *l, *next;
734 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
735 log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s",
736 l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path);
738 if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]))
739 cgroup_context_free_io_device_limit(c, l);
744 if ((mask & CGROUP_MASK_MEMORY) && !is_root) {
745 if (cg_unified() > 0) {
746 uint64_t max = c->memory_max;
747 uint64_t swap_max = c->memory_swap_max;
749 if (cgroup_context_has_unified_memory_config(c)) {
751 swap_max = c->memory_swap_max;
753 max = c->memory_limit;
755 if (max != CGROUP_LIMIT_MAX)
756 log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max);
759 cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low);
760 cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high);
761 cgroup_apply_unified_memory_limit(u, "memory.max", max);
762 cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max);
764 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
765 uint64_t val = c->memory_limit;
767 if (val == CGROUP_LIMIT_MAX) {
770 if (val != CGROUP_LIMIT_MAX)
771 log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", c->memory_max);
774 if (val == CGROUP_LIMIT_MAX)
775 strncpy(buf, "-1\n", sizeof(buf));
777 xsprintf(buf, "%" PRIu64 "\n", val);
779 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
781 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
782 "Failed to set memory.limit_in_bytes: %m");
786 if ((mask & CGROUP_MASK_DEVICES) && !is_root) {
787 CGroupDeviceAllow *a;
789 /* Changing the devices list of a populated cgroup
790 * might result in EINVAL, hence ignore EINVAL
793 if (c->device_allow || c->device_policy != CGROUP_AUTO)
794 r = cg_set_attribute("devices", path, "devices.deny", "a");
796 r = cg_set_attribute("devices", path, "devices.allow", "a");
798 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
799 "Failed to reset devices.list: %m");
801 if (c->device_policy == CGROUP_CLOSED ||
802 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
803 static const char auto_devices[] =
804 "/dev/null\0" "rwm\0"
805 "/dev/zero\0" "rwm\0"
806 "/dev/full\0" "rwm\0"
807 "/dev/random\0" "rwm\0"
808 "/dev/urandom\0" "rwm\0"
810 "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */
811 /* Allow /run/elogind/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
812 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
813 "/run/systemd/inaccessible/chr\0" "rwm\0"
814 "/run/systemd/inaccessible/blk\0" "rwm\0";
818 NULSTR_FOREACH_PAIR(x, y, auto_devices)
819 whitelist_device(path, x, y);
821 whitelist_major(path, "pts", 'c', "rw");
822 whitelist_major(path, "kdbus", 'c', "rw");
823 whitelist_major(path, "kdbus/*", 'c', "rw");
826 LIST_FOREACH(device_allow, a, c->device_allow) {
842 if (startswith(a->path, "/dev/"))
843 whitelist_device(path, a->path, acc);
844 else if (startswith(a->path, "block-"))
845 whitelist_major(path, a->path + 6, 'b', acc);
846 else if (startswith(a->path, "char-"))
847 whitelist_major(path, a->path + 5, 'c', acc);
849 log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path);
853 if ((mask & CGROUP_MASK_PIDS) && !is_root) {
855 if (c->tasks_max != (uint64_t) -1) {
856 char buf[DECIMAL_STR_MAX(uint64_t) + 2];
858 sprintf(buf, "%" PRIu64 "\n", c->tasks_max);
859 r = cg_set_attribute("pids", path, "pids.max", buf);
861 r = cg_set_attribute("pids", path, "pids.max", "max");
864 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
865 "Failed to set pids.max: %m");
869 CGroupMask cgroup_context_get_mask(CGroupContext *c) {
872 /* Figure out which controllers we need */
874 if (c->cpu_accounting ||
875 c->cpu_shares != CGROUP_CPU_SHARES_INVALID ||
876 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID ||
877 c->cpu_quota_per_sec_usec != USEC_INFINITY)
878 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
880 if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c))
881 mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
883 if (c->memory_accounting ||
884 c->memory_limit != CGROUP_LIMIT_MAX ||
885 cgroup_context_has_unified_memory_config(c))
886 mask |= CGROUP_MASK_MEMORY;
888 if (c->device_allow ||
889 c->device_policy != CGROUP_AUTO)
890 mask |= CGROUP_MASK_DEVICES;
892 if (c->tasks_accounting ||
893 c->tasks_max != (uint64_t) -1)
894 mask |= CGROUP_MASK_PIDS;
899 CGroupMask unit_get_own_mask(Unit *u) {
902 /* Returns the mask of controllers the unit needs for itself */
904 c = unit_get_cgroup_context(u);
908 /* If delegation is turned on, then turn on all cgroups,
909 * unless we are on the legacy hierarchy and the process we
910 * fork into it is known to drop privileges, and hence
911 * shouldn't get access to the controllers.
913 * Note that on the unified hierarchy it is safe to delegate
914 * controllers to unprivileged services. */
919 e = unit_get_exec_context(u);
921 exec_context_maintains_privileges(e) ||
923 return _CGROUP_MASK_ALL;
926 return cgroup_context_get_mask(c);
929 CGroupMask unit_get_members_mask(Unit *u) {
932 /* Returns the mask of controllers all of the unit's children
935 if (u->cgroup_members_mask_valid)
936 return u->cgroup_members_mask;
938 u->cgroup_members_mask = 0;
940 if (u->type == UNIT_SLICE) {
944 SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) {
949 if (UNIT_DEREF(member->slice) != u)
952 u->cgroup_members_mask |=
953 unit_get_own_mask(member) |
954 unit_get_members_mask(member);
958 u->cgroup_members_mask_valid = true;
959 return u->cgroup_members_mask;
962 CGroupMask unit_get_siblings_mask(Unit *u) {
965 /* Returns the mask of controllers all of the unit's siblings
966 * require, i.e. the members mask of the unit's parent slice
967 * if there is one. */
969 if (UNIT_ISSET(u->slice))
970 return unit_get_members_mask(UNIT_DEREF(u->slice));
972 return unit_get_own_mask(u) | unit_get_members_mask(u);
975 CGroupMask unit_get_subtree_mask(Unit *u) {
977 /* Returns the mask of this subtree, meaning of the group
978 * itself and its children. */
980 return unit_get_own_mask(u) | unit_get_members_mask(u);
983 CGroupMask unit_get_target_mask(Unit *u) {
986 /* This returns the cgroup mask of all controllers to enable
987 * for a specific cgroup, i.e. everything it needs itself,
988 * plus all that its children need, plus all that its siblings
989 * need. This is primarily useful on the legacy cgroup
990 * hierarchy, where we need to duplicate each cgroup in each
991 * hierarchy that shall be enabled for it. */
993 mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
994 mask &= u->manager->cgroup_supported;
999 CGroupMask unit_get_enable_mask(Unit *u) {
1002 /* This returns the cgroup mask of all controllers to enable
1003 * for the children of a specific cgroup. This is primarily
1004 * useful for the unified cgroup hierarchy, where each cgroup
1005 * controls which controllers are enabled for its children. */
1007 mask = unit_get_members_mask(u);
1008 mask &= u->manager->cgroup_supported;
1013 /* Recurse from a unit up through its containing slices, propagating
1014 * mask bits upward. A unit is also member of itself. */
1015 void unit_update_cgroup_members_masks(Unit *u) {
1021 /* Calculate subtree mask */
1022 m = unit_get_subtree_mask(u);
1024 /* See if anything changed from the previous invocation. If
1025 * not, we're done. */
1026 if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask)
1030 u->cgroup_subtree_mask_valid &&
1031 ((m & ~u->cgroup_subtree_mask) != 0) &&
1032 ((~m & u->cgroup_subtree_mask) == 0);
1034 u->cgroup_subtree_mask = m;
1035 u->cgroup_subtree_mask_valid = true;
1037 if (UNIT_ISSET(u->slice)) {
1038 Unit *s = UNIT_DEREF(u->slice);
1041 /* There's more set now than before. We
1042 * propagate the new mask to the parent's mask
1043 * (not caring if it actually was valid or
1046 s->cgroup_members_mask |= m;
1049 /* There's less set now than before (or we
1050 * don't know), we need to recalculate
1051 * everything, so let's invalidate the
1052 * parent's members mask */
1054 s->cgroup_members_mask_valid = false;
1056 /* And now make sure that this change also hits our
1058 unit_update_cgroup_members_masks(s);
1062 static const char *migrate_callback(CGroupMask mask, void *userdata) {
1069 if (u->cgroup_path &&
1070 u->cgroup_realized &&
1071 (u->cgroup_realized_mask & mask) == mask)
1072 return u->cgroup_path;
1074 u = UNIT_DEREF(u->slice);
1080 char *unit_default_cgroup_path(Unit *u) {
1081 _cleanup_free_ char *escaped = NULL, *slice = NULL;
1086 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1087 return strdup(u->manager->cgroup_root);
1089 if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) {
1090 r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice);
1095 escaped = cg_escape(u->id);
1100 return strjoin(u->manager->cgroup_root, "/", slice, "/", escaped, NULL);
1102 return strjoin(u->manager->cgroup_root, "/", escaped, NULL);
1105 int unit_set_cgroup_path(Unit *u, const char *path) {
1106 _cleanup_free_ char *p = NULL;
1118 if (streq_ptr(u->cgroup_path, p))
1122 r = hashmap_put(u->manager->cgroup_unit, p, u);
1127 unit_release_cgroup(u);
1135 int unit_watch_cgroup(Unit *u) {
1136 _cleanup_free_ char *events = NULL;
1141 if (!u->cgroup_path)
1144 if (u->cgroup_inotify_wd >= 0)
1147 /* Only applies to the unified hierarchy */
1150 return log_unit_error_errno(u, r, "Failed detect whether the unified hierarchy is used: %m");
1154 /* Don't watch the root slice, it's pointless. */
1155 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1158 r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops);
1162 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events);
1166 u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY);
1167 if (u->cgroup_inotify_wd < 0) {
1169 if (errno == ENOENT) /* If the directory is already
1170 * gone we don't need to track
1171 * it, so this is not an error */
1174 return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path);
1177 r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u);
1179 return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m");
1184 static int unit_create_cgroup(
1186 CGroupMask target_mask,
1187 CGroupMask enable_mask) {
1194 c = unit_get_cgroup_context(u);
1198 if (!u->cgroup_path) {
1199 _cleanup_free_ char *path = NULL;
1201 path = unit_default_cgroup_path(u);
1205 r = unit_set_cgroup_path(u, path);
1207 return log_unit_error_errno(u, r, "Control group %s exists already.", path);
1209 return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path);
1212 /* First, create our own group */
1213 r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path);
1215 return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path);
1217 /* Start watching it */
1218 (void) unit_watch_cgroup(u);
1220 /* Enable all controllers we need */
1221 r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path);
1223 log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path);
1225 /* Keep track that this is now realized */
1226 u->cgroup_realized = true;
1227 u->cgroup_realized_mask = target_mask;
1228 u->cgroup_enabled_mask = enable_mask;
1230 if (u->type != UNIT_SLICE && !c->delegate) {
1232 /* Then, possibly move things over, but not if
1233 * subgroups may contain processes, which is the case
1234 * for slice and delegation units. */
1235 r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u);
1237 log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path);
1243 int unit_attach_pids_to_cgroup(Unit *u) {
1247 r = unit_realize_cgroup(u);
1251 r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u);
1258 static bool unit_has_mask_realized(Unit *u, CGroupMask target_mask, CGroupMask enable_mask) {
1261 return u->cgroup_realized && u->cgroup_realized_mask == target_mask && u->cgroup_enabled_mask == enable_mask;
1264 /* Check if necessary controllers and attributes for a unit are in place.
1266 * If so, do nothing.
1267 * If not, create paths, move processes over, and set attributes.
1269 * Returns 0 on success and < 0 on failure. */
1270 static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
1271 CGroupMask target_mask, enable_mask;
1276 if (u->in_cgroup_queue) {
1277 LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u);
1278 u->in_cgroup_queue = false;
1281 target_mask = unit_get_target_mask(u);
1282 enable_mask = unit_get_enable_mask(u);
1284 if (unit_has_mask_realized(u, target_mask, enable_mask))
1287 /* First, realize parents */
1288 if (UNIT_ISSET(u->slice)) {
1289 r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
1294 /* And then do the real work */
1295 r = unit_create_cgroup(u, target_mask, enable_mask);
1299 /* Finally, apply the necessary attributes. */
1300 cgroup_context_apply(u, target_mask, state);
1305 static void unit_add_to_cgroup_queue(Unit *u) {
1307 if (u->in_cgroup_queue)
1310 LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u);
1311 u->in_cgroup_queue = true;
1314 unsigned manager_dispatch_cgroup_queue(Manager *m) {
1320 state = manager_state(m);
1322 while ((i = m->cgroup_queue)) {
1323 assert(i->in_cgroup_queue);
1325 r = unit_realize_cgroup_now(i, state);
1327 log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id);
1335 static void unit_queue_siblings(Unit *u) {
1338 /* This adds the siblings of the specified unit and the
1339 * siblings of all parent units to the cgroup queue. (But
1340 * neither the specified unit itself nor the parents.) */
1342 while ((slice = UNIT_DEREF(u->slice))) {
1346 SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) {
1350 /* Skip units that have a dependency on the slice
1351 * but aren't actually in it. */
1352 if (UNIT_DEREF(m->slice) != slice)
1355 /* No point in doing cgroup application for units
1356 * without active processes. */
1357 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
1360 /* If the unit doesn't need any new controllers
1361 * and has current ones realized, it doesn't need
1363 if (unit_has_mask_realized(m, unit_get_target_mask(m), unit_get_enable_mask(m)))
1366 unit_add_to_cgroup_queue(m);
1373 int unit_realize_cgroup(Unit *u) {
1376 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1379 /* So, here's the deal: when realizing the cgroups for this
1380 * unit, we need to first create all parents, but there's more
1381 * actually: for the weight-based controllers we also need to
1382 * make sure that all our siblings (i.e. units that are in the
1383 * same slice as we are) have cgroups, too. Otherwise, things
1384 * would become very uneven as each of their processes would
1385 * get as much resources as all our group together. This call
1386 * will synchronously create the parent cgroups, but will
1387 * defer work on the siblings to the next event loop
1390 /* Add all sibling slices to the cgroup queue. */
1391 unit_queue_siblings(u);
1393 /* And realize this one now (and apply the values) */
1394 return unit_realize_cgroup_now(u, manager_state(u->manager));
1397 void unit_release_cgroup(Unit *u) {
1400 /* Forgets all cgroup details for this cgroup */
1402 if (u->cgroup_path) {
1403 (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
1404 u->cgroup_path = mfree(u->cgroup_path);
1407 if (u->cgroup_inotify_wd >= 0) {
1408 if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0)
1409 log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id);
1411 (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd));
1412 u->cgroup_inotify_wd = -1;
1416 void unit_prune_cgroup(Unit *u) {
1422 /* Removes the cgroup, if empty and possible, and stops watching it. */
1424 if (!u->cgroup_path)
1427 is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE);
1429 r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice);
1431 log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path);
1438 unit_release_cgroup(u);
1440 u->cgroup_realized = false;
1441 u->cgroup_realized_mask = 0;
1442 u->cgroup_enabled_mask = 0;
1445 int unit_search_main_pid(Unit *u, pid_t *ret) {
1446 _cleanup_fclose_ FILE *f = NULL;
1447 pid_t pid = 0, npid, mypid;
1453 if (!u->cgroup_path)
1456 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f);
1461 while (cg_read_pid(f, &npid) > 0) {
1467 /* Ignore processes that aren't our kids */
1468 if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid)
1472 /* Dang, there's more than one daemonized PID
1473 in this group, so we don't know what process
1474 is the main process. */
1485 static int unit_watch_pids_in_path(Unit *u, const char *path) {
1486 _cleanup_closedir_ DIR *d = NULL;
1487 _cleanup_fclose_ FILE *f = NULL;
1493 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
1499 while ((r = cg_read_pid(f, &pid)) > 0) {
1500 r = unit_watch_pid(u, pid);
1501 if (r < 0 && ret >= 0)
1505 if (r < 0 && ret >= 0)
1509 r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
1516 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1517 _cleanup_free_ char *p = NULL;
1519 p = strjoin(path, "/", fn, NULL);
1525 r = unit_watch_pids_in_path(u, p);
1526 if (r < 0 && ret >= 0)
1530 if (r < 0 && ret >= 0)
1537 int unit_watch_all_pids(Unit *u) {
1540 /* Adds all PIDs from our cgroup to the set of PIDs we
1541 * watch. This is a fallback logic for cases where we do not
1542 * get reliable cgroup empty notifications: we try to use
1543 * SIGCHLD as replacement. */
1545 if (!u->cgroup_path)
1548 if (cg_unified() > 0) /* On unified we can use proper notifications */
1551 return unit_watch_pids_in_path(u, u->cgroup_path);
1554 int unit_notify_cgroup_empty(Unit *u) {
1559 if (!u->cgroup_path)
1562 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
1566 unit_add_to_gc_queue(u);
1568 if (UNIT_VTABLE(u)->notify_cgroup_empty)
1569 UNIT_VTABLE(u)->notify_cgroup_empty(u);
1574 static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1575 Manager *m = userdata;
1582 union inotify_event_buffer buffer;
1583 struct inotify_event *e;
1586 l = read(fd, &buffer, sizeof(buffer));
1588 if (errno == EINTR || errno == EAGAIN)
1591 return log_error_errno(errno, "Failed to read control group inotify events: %m");
1594 FOREACH_INOTIFY_EVENT(e, buffer, l) {
1598 /* Queue overflow has no watch descriptor */
1601 if (e->mask & IN_IGNORED)
1602 /* The watch was just removed */
1605 u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd));
1606 if (!u) /* Not that inotify might deliver
1607 * events for a watch even after it
1608 * was removed, because it was queued
1609 * before the removal. Let's ignore
1610 * this here safely. */
1613 (void) unit_notify_cgroup_empty(u);
1619 int manager_setup_cgroup(Manager *m) {
1620 _cleanup_free_ char *path = NULL;
1627 /* 1. Determine hierarchy */
1628 m->cgroup_root = mfree(m->cgroup_root);
1629 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
1631 return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
1633 #if 0 /// elogind does not support systemd scopes and slices
1634 /* Chop off the init scope, if we are already located in it */
1635 e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1637 /* LEGACY: Also chop off the system slice if we are in
1638 * it. This is to support live upgrades from older systemd
1639 * versions where PID 1 was moved there. Also see
1640 * cg_get_root_path(). */
1641 if (!e && MANAGER_IS_SYSTEM(m)) {
1642 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
1644 e = endswith(m->cgroup_root, "/system"); /* even more legacy */
1650 /* And make sure to store away the root value without trailing
1651 * slash, even for the root dir, so that we can easily prepend
1653 while ((e = endswith(m->cgroup_root, "/")))
1655 log_debug_elogind("Cgroup Controller \"%s\" -> root \"%s\"",
1656 SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root);
1659 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
1661 return log_error_errno(r, "Cannot find cgroup mount point: %m");
1663 unified = cg_unified();
1665 return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m");
1667 log_debug("Unified cgroup hierarchy is located at %s.", path);
1669 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER ". File system hierarchy is at %s.", path);
1672 const char *scope_path;
1674 /* 3. Install agent */
1677 /* In the unified hierarchy we can get
1678 * cgroup empty notifications via inotify. */
1680 #if 0 /// elogind does not support the unified hierarchy, yet.
1681 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1682 safe_close(m->cgroup_inotify_fd);
1684 m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1685 if (m->cgroup_inotify_fd < 0)
1686 return log_error_errno(errno, "Failed to create control group inotify object: %m");
1688 r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m);
1690 return log_error_errno(r, "Failed to watch control group inotify object: %m");
1692 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
1693 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
1694 r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-5);
1696 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
1698 (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify");
1701 return log_error_errno(EOPNOTSUPP, "Unified cgroup hierarchy not supported: %m");
1703 } else if (MANAGER_IS_SYSTEM(m)) {
1705 /* On the legacy hierarchy we only get
1706 * notifications via cgroup agents. (Which
1707 * isn't really reliable, since it does not
1708 * generate events when control groups with
1709 * children run empty. */
1711 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
1713 log_warning_errno(r, "Failed to install release agent, ignoring: %m");
1715 log_debug("Installed release agent.");
1717 log_debug("Release agent already installed.");
1720 #if 0 /// elogind is not meant to run in systemd init scope
1721 /* 4. Make sure we are in the special "init.scope" unit in the root slice. */
1722 scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1723 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1725 if (streq(SYSTEMD_CGROUP_CONTROLLER, "name=elogind"))
1726 // we are our own cgroup controller
1727 scope_path = strjoina("");
1728 else if (streq(m->cgroup_root, "/elogind"))
1729 // root already is our cgroup
1730 scope_path = strjoina(m->cgroup_root);
1732 // we have to create our own group
1733 scope_path = strjoina(m->cgroup_root, "/elogind");
1734 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1737 return log_error_errno(r, "Failed to create %s control group: %m", scope_path);
1738 log_debug_elogind("Created control group \"%s\"", scope_path);
1740 /* also, move all other userspace processes remaining
1741 * in the root cgroup into that scope. */
1742 r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1744 log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m");
1746 /* 5. And pin it, so that it cannot be unmounted */
1747 safe_close(m->pin_cgroupfs_fd);
1748 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
1749 if (m->pin_cgroupfs_fd < 0)
1750 return log_error_errno(errno, "Failed to open pin file: %m");
1752 /* 6. Always enable hierarchical support if it exists... */
1754 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
1757 /* 7. Figure out which controllers are supported */
1758 r = cg_mask_supported(&m->cgroup_supported);
1760 return log_error_errno(r, "Failed to determine supported controllers: %m");
1762 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++)
1763 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c)));
1768 void manager_shutdown_cgroup(Manager *m, bool delete) {
1771 /* We can't really delete the group, since we are in it. But
1773 if (delete && m->cgroup_root)
1774 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
1776 #if 0 /// elogind does not support the unified hierarchy, yet.
1777 m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit);
1779 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1780 m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd);
1783 m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
1785 m->cgroup_root = mfree(m->cgroup_root);
1788 #if 0 /// UNNEEDED by elogind
1789 Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
1796 u = hashmap_get(m->cgroup_unit, cgroup);
1800 p = strdupa(cgroup);
1804 e = strrchr(p, '/');
1806 return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE);
1810 u = hashmap_get(m->cgroup_unit, p);
1816 Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) {
1817 _cleanup_free_ char *cgroup = NULL;
1825 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
1829 return manager_get_unit_by_cgroup(m, cgroup);
1832 Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
1841 return hashmap_get(m->units, SPECIAL_INIT_SCOPE);
1843 u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid));
1847 u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid));
1851 return manager_get_unit_by_pid_cgroup(m, pid);
1855 #if 0 /// elogind must substitute this with its own variant
1856 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
1862 log_debug("Got cgroup empty notification for: %s", cgroup);
1864 u = manager_get_unit_by_cgroup(m, cgroup);
1868 return unit_notify_cgroup_empty(u);
1871 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
1877 log_debug("Got cgroup empty notification for: %s", cgroup);
1879 s = hashmap_get(m->sessions, cgroup);
1882 session_finalize(s);
1885 log_warning("Session not found: %s", cgroup);
1891 #if 0 /// UNNEEDED by elogind
1892 int unit_get_memory_current(Unit *u, uint64_t *ret) {
1893 _cleanup_free_ char *v = NULL;
1899 if (!u->cgroup_path)
1902 if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
1905 if (cg_unified() <= 0)
1906 r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v);
1908 r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v);
1914 return safe_atou64(v, ret);
1917 int unit_get_tasks_current(Unit *u, uint64_t *ret) {
1918 _cleanup_free_ char *v = NULL;
1924 if (!u->cgroup_path)
1927 if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0)
1930 r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v);
1936 return safe_atou64(v, ret);
1939 static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) {
1940 _cleanup_free_ char *v = NULL;
1947 if (!u->cgroup_path)
1950 if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0)
1953 r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v);
1959 r = safe_atou64(v, &ns);
1967 int unit_get_cpu_usage(Unit *u, nsec_t *ret) {
1971 r = unit_get_cpu_usage_raw(u, &ns);
1975 if (ns > u->cpuacct_usage_base)
1976 ns -= u->cpuacct_usage_base;
1984 int unit_reset_cpu_usage(Unit *u) {
1990 r = unit_get_cpu_usage_raw(u, &ns);
1992 u->cpuacct_usage_base = 0;
1996 u->cpuacct_usage_base = ns;
2000 bool unit_cgroup_delegate(Unit *u) {
2005 c = unit_get_cgroup_context(u);
2012 void unit_invalidate_cgroup(Unit *u, CGroupMask m) {
2015 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2021 /* always invalidate compat pairs together */
2022 if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO))
2023 m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
2025 if ((u->cgroup_realized_mask & m) == 0)
2028 u->cgroup_realized_mask &= ~m;
2029 unit_add_to_cgroup_queue(u);
2032 void manager_invalidate_startup_units(Manager *m) {
2038 SET_FOREACH(u, m->startup_units, i)
2039 unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
2042 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
2043 [CGROUP_AUTO] = "auto",
2044 [CGROUP_CLOSED] = "closed",
2045 [CGROUP_STRICT] = "strict",
2048 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);