1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
5 #include "sd-messages.h"
8 #include "alloc-util.h"
9 #include "blockdev-util.h"
10 #include "bpf-devices.h"
11 #include "bpf-firewall.h"
12 #include "bpf-foreign.h"
13 #include "bpf-socket-bind.h"
14 #include "btrfs-util.h"
15 #include "bus-error.h"
16 #include "bus-locator.h"
17 #include "cgroup-setup.h"
18 #include "cgroup-util.h"
20 #include "devnum-util.h"
23 #include "firewall-util.h"
24 #include "in-addr-prefix-util.h"
25 #include "inotify-util.h"
27 #include "ip-protocol-list.h"
28 #include "limits-util.h"
29 #include "nulstr-util.h"
30 #include "parse-util.h"
31 #include "path-util.h"
32 #include "percent-util.h"
33 #include "process-util.h"
34 #include "procfs-util.h"
35 #include "restrict-ifaces.h"
38 #include "stdio-util.h"
39 #include "string-table.h"
40 #include "string-util.h"
44 #include "bpf-dlopen.h"
46 #include "bpf/restrict_fs/restrict-fs-skel.h"
49 #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
51 /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access
52 * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask
53 * out specific attributes from us. */
54 #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING)
56 uint64_t cgroup_tasks_max_resolve(const CGroupTasksMax
*tasks_max
) {
57 if (tasks_max
->scale
== 0)
58 return tasks_max
->value
;
60 return system_tasks_max_scale(tasks_max
->value
, tasks_max
->scale
);
63 bool manager_owns_host_root_cgroup(Manager
*m
) {
66 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
67 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
68 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
69 * we run in any kind of container virtualization. */
71 if (MANAGER_IS_USER(m
))
74 if (detect_container() > 0)
77 return empty_or_root(m
->cgroup_root
);
80 bool unit_has_startup_cgroup_constraints(Unit
*u
) {
83 /* Returns true if this unit has any directives which apply during
84 * startup/shutdown phases. */
88 c
= unit_get_cgroup_context(u
);
92 return c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
93 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
94 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
95 c
->startup_cpuset_cpus
.set
||
96 c
->startup_cpuset_mems
.set
||
97 c
->startup_memory_high_set
||
98 c
->startup_memory_max_set
||
99 c
->startup_memory_swap_max_set
||
100 c
->startup_memory_zswap_max_set
||
101 c
->startup_memory_low_set
;
104 bool unit_has_host_root_cgroup(Unit
*u
) {
107 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
108 * the manager manages the root cgroup. */
110 if (!manager_owns_host_root_cgroup(u
->manager
))
113 return unit_has_name(u
, SPECIAL_ROOT_SLICE
);
116 static int set_attribute_and_warn(Unit
*u
, const char *controller
, const char *attribute
, const char *value
) {
119 r
= cg_set_attribute(controller
, u
->cgroup_path
, attribute
, value
);
121 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%.*s': %m",
122 strna(attribute
), empty_to_root(u
->cgroup_path
), (int) strcspn(value
, NEWLINE
), value
);
127 static void cgroup_compat_warn(void) {
128 static bool cgroup_compat_warned
= false;
130 if (cgroup_compat_warned
)
133 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
134 "See cgroup-compat debug messages for details.");
136 cgroup_compat_warned
= true;
139 #define log_cgroup_compat(unit, fmt, ...) do { \
140 cgroup_compat_warn(); \
141 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
144 void cgroup_context_init(CGroupContext
*c
) {
147 /* Initialize everything to the kernel defaults. When initializing a bool member to 'true', make
148 * sure to serialize in execute-serialize.c using serialize_bool() instead of
149 * serialize_bool_elide(), as sd-executor will initialize here to 'true', but serialize_bool_elide()
150 * skips serialization if the value is 'false' (as that's the common default), so if the value at
151 * runtime is zero it would be lost after deserialization. Same when initializing uint64_t and other
152 * values, update/add a conditional serialization check. This is to minimize the amount of
153 * serialized data that is sent to the sd-executor, so that there is less work to do on the default
156 *c
= (CGroupContext
) {
157 .cpu_weight
= CGROUP_WEIGHT_INVALID
,
158 .startup_cpu_weight
= CGROUP_WEIGHT_INVALID
,
159 .cpu_quota_per_sec_usec
= USEC_INFINITY
,
160 .cpu_quota_period_usec
= USEC_INFINITY
,
162 .cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
163 .startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
165 .memory_high
= CGROUP_LIMIT_MAX
,
166 .startup_memory_high
= CGROUP_LIMIT_MAX
,
167 .memory_max
= CGROUP_LIMIT_MAX
,
168 .startup_memory_max
= CGROUP_LIMIT_MAX
,
169 .memory_swap_max
= CGROUP_LIMIT_MAX
,
170 .startup_memory_swap_max
= CGROUP_LIMIT_MAX
,
171 .memory_zswap_max
= CGROUP_LIMIT_MAX
,
172 .startup_memory_zswap_max
= CGROUP_LIMIT_MAX
,
174 .memory_limit
= CGROUP_LIMIT_MAX
,
176 .io_weight
= CGROUP_WEIGHT_INVALID
,
177 .startup_io_weight
= CGROUP_WEIGHT_INVALID
,
179 .blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
180 .startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
182 .tasks_max
= CGROUP_TASKS_MAX_UNSET
,
184 .moom_swap
= MANAGED_OOM_AUTO
,
185 .moom_mem_pressure
= MANAGED_OOM_AUTO
,
186 .moom_preference
= MANAGED_OOM_PREFERENCE_NONE
,
188 .memory_pressure_watch
= _CGROUP_PRESSURE_WATCH_INVALID
,
189 .memory_pressure_threshold_usec
= USEC_INFINITY
,
193 int cgroup_context_add_io_device_weight_dup(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
194 _cleanup_free_ CGroupIODeviceWeight
*n
= NULL
;
199 n
= new0(CGroupIODeviceWeight
, 1);
203 n
->path
= strdup(w
->path
);
206 n
->weight
= w
->weight
;
208 LIST_PREPEND(device_weights
, c
->io_device_weights
, TAKE_PTR(n
));
212 int cgroup_context_add_io_device_limit_dup(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
213 _cleanup_free_ CGroupIODeviceLimit
*n
= NULL
;
218 n
= new0(CGroupIODeviceLimit
, 1);
222 n
->path
= strdup(l
->path
);
226 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
227 n
->limits
[type
] = l
->limits
[type
];
229 LIST_PREPEND(device_limits
, c
->io_device_limits
, TAKE_PTR(n
));
233 int cgroup_context_add_io_device_latency_dup(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
234 _cleanup_free_ CGroupIODeviceLatency
*n
= NULL
;
239 n
= new0(CGroupIODeviceLatency
, 1);
243 n
->path
= strdup(l
->path
);
247 n
->target_usec
= l
->target_usec
;
249 LIST_PREPEND(device_latencies
, c
->io_device_latencies
, TAKE_PTR(n
));
253 int cgroup_context_add_block_io_device_weight_dup(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
254 _cleanup_free_ CGroupBlockIODeviceWeight
*n
= NULL
;
259 n
= new0(CGroupBlockIODeviceWeight
, 1);
263 n
->path
= strdup(w
->path
);
267 n
->weight
= w
->weight
;
269 LIST_PREPEND(device_weights
, c
->blockio_device_weights
, TAKE_PTR(n
));
273 int cgroup_context_add_block_io_device_bandwidth_dup(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
274 _cleanup_free_ CGroupBlockIODeviceBandwidth
*n
= NULL
;
279 n
= new0(CGroupBlockIODeviceBandwidth
, 1);
283 *n
= (CGroupBlockIODeviceBandwidth
) {
288 LIST_PREPEND(device_bandwidths
, c
->blockio_device_bandwidths
, TAKE_PTR(n
));
292 int cgroup_context_add_device_allow_dup(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
293 _cleanup_free_ CGroupDeviceAllow
*n
= NULL
;
298 n
= new0(CGroupDeviceAllow
, 1);
302 n
->path
= strdup(a
->path
);
306 n
->permissions
= a
->permissions
;
308 LIST_PREPEND(device_allow
, c
->device_allow
, TAKE_PTR(n
));
312 static int cgroup_context_add_socket_bind_item_dup(CGroupContext
*c
, CGroupSocketBindItem
*i
, CGroupSocketBindItem
*h
) {
313 _cleanup_free_ CGroupSocketBindItem
*n
= NULL
;
318 n
= new0(CGroupSocketBindItem
, 1);
322 *n
= (CGroupSocketBindItem
) {
323 .address_family
= i
->address_family
,
324 .ip_protocol
= i
->ip_protocol
,
325 .nr_ports
= i
->nr_ports
,
326 .port_min
= i
->port_min
,
329 LIST_PREPEND(socket_bind_items
, h
, TAKE_PTR(n
));
333 int cgroup_context_add_socket_bind_item_allow_dup(CGroupContext
*c
, CGroupSocketBindItem
*i
) {
334 return cgroup_context_add_socket_bind_item_dup(c
, i
, c
->socket_bind_allow
);
337 int cgroup_context_add_socket_bind_item_deny_dup(CGroupContext
*c
, CGroupSocketBindItem
*i
) {
338 return cgroup_context_add_socket_bind_item_dup(c
, i
, c
->socket_bind_deny
);
341 int cgroup_context_copy(CGroupContext
*dst
, const CGroupContext
*src
) {
342 struct in_addr_prefix
*i
;
349 dst
->cpu_accounting
= src
->cpu_accounting
;
350 dst
->io_accounting
= src
->io_accounting
;
351 dst
->blockio_accounting
= src
->blockio_accounting
;
352 dst
->memory_accounting
= src
->memory_accounting
;
353 dst
->tasks_accounting
= src
->tasks_accounting
;
354 dst
->ip_accounting
= src
->ip_accounting
;
356 dst
->memory_oom_group
= dst
->memory_oom_group
;
358 dst
->cpu_weight
= src
->cpu_weight
;
359 dst
->startup_cpu_weight
= src
->startup_cpu_weight
;
360 dst
->cpu_quota_per_sec_usec
= src
->cpu_quota_per_sec_usec
;
361 dst
->cpu_quota_period_usec
= src
->cpu_quota_period_usec
;
363 dst
->cpuset_cpus
= src
->cpuset_cpus
;
364 dst
->startup_cpuset_cpus
= src
->startup_cpuset_cpus
;
365 dst
->cpuset_mems
= src
->cpuset_mems
;
366 dst
->startup_cpuset_mems
= src
->startup_cpuset_mems
;
368 dst
->io_weight
= src
->io_weight
;
369 dst
->startup_io_weight
= src
->startup_io_weight
;
371 LIST_FOREACH_BACKWARDS(device_weights
, w
, LIST_FIND_TAIL(device_weights
, src
->io_device_weights
)) {
372 r
= cgroup_context_add_io_device_weight_dup(dst
, w
);
377 LIST_FOREACH_BACKWARDS(device_limits
, l
, LIST_FIND_TAIL(device_limits
, src
->io_device_limits
)) {
378 r
= cgroup_context_add_io_device_limit_dup(dst
, l
);
383 LIST_FOREACH_BACKWARDS(device_latencies
, l
, LIST_FIND_TAIL(device_latencies
, src
->io_device_latencies
)) {
384 r
= cgroup_context_add_io_device_latency_dup(dst
, l
);
389 dst
->default_memory_min
= src
->default_memory_min
;
390 dst
->default_memory_low
= src
->default_memory_low
;
391 dst
->default_startup_memory_low
= src
->default_startup_memory_low
;
392 dst
->memory_min
= src
->memory_min
;
393 dst
->memory_low
= src
->memory_low
;
394 dst
->startup_memory_low
= src
->startup_memory_low
;
395 dst
->memory_high
= src
->memory_high
;
396 dst
->startup_memory_high
= src
->startup_memory_high
;
397 dst
->memory_max
= src
->memory_max
;
398 dst
->startup_memory_max
= src
->startup_memory_max
;
399 dst
->memory_swap_max
= src
->memory_swap_max
;
400 dst
->startup_memory_swap_max
= src
->startup_memory_swap_max
;
401 dst
->memory_zswap_max
= src
->memory_zswap_max
;
402 dst
->startup_memory_zswap_max
= src
->startup_memory_zswap_max
;
404 dst
->default_memory_min_set
= src
->default_memory_min_set
;
405 dst
->default_memory_low_set
= src
->default_memory_low_set
;
406 dst
->default_startup_memory_low_set
= src
->default_startup_memory_low_set
;
407 dst
->memory_min_set
= src
->memory_min_set
;
408 dst
->memory_low_set
= src
->memory_low_set
;
409 dst
->startup_memory_low_set
= src
->startup_memory_low_set
;
410 dst
->startup_memory_high_set
= src
->startup_memory_high_set
;
411 dst
->startup_memory_max_set
= src
->startup_memory_max_set
;
412 dst
->startup_memory_swap_max_set
= src
->startup_memory_swap_max_set
;
413 dst
->startup_memory_zswap_max_set
= src
->startup_memory_zswap_max_set
;
415 SET_FOREACH(i
, src
->ip_address_allow
) {
416 r
= in_addr_prefix_add(&dst
->ip_address_allow
, i
);
421 SET_FOREACH(i
, src
->ip_address_deny
) {
422 r
= in_addr_prefix_add(&dst
->ip_address_deny
, i
);
427 dst
->ip_address_allow_reduced
= src
->ip_address_allow_reduced
;
428 dst
->ip_address_deny_reduced
= src
->ip_address_deny_reduced
;
430 if (!strv_isempty(src
->ip_filters_ingress
)) {
431 dst
->ip_filters_ingress
= strv_copy(src
->ip_filters_ingress
);
432 if (!dst
->ip_filters_ingress
)
436 if (!strv_isempty(src
->ip_filters_egress
)) {
437 dst
->ip_filters_egress
= strv_copy(src
->ip_filters_egress
);
438 if (!dst
->ip_filters_egress
)
442 LIST_FOREACH_BACKWARDS(programs
, l
, LIST_FIND_TAIL(programs
, src
->bpf_foreign_programs
)) {
443 r
= cgroup_context_add_bpf_foreign_program_dup(dst
, l
);
448 SET_FOREACH(iface
, src
->restrict_network_interfaces
) {
449 r
= set_put_strdup(&dst
->restrict_network_interfaces
, iface
);
453 dst
->restrict_network_interfaces_is_allow_list
= src
->restrict_network_interfaces_is_allow_list
;
455 dst
->cpu_shares
= src
->cpu_shares
;
456 dst
->startup_cpu_shares
= src
->startup_cpu_shares
;
458 dst
->blockio_weight
= src
->blockio_weight
;
459 dst
->startup_blockio_weight
= src
->startup_blockio_weight
;
461 LIST_FOREACH_BACKWARDS(device_weights
, l
, LIST_FIND_TAIL(device_weights
, src
->blockio_device_weights
)) {
462 r
= cgroup_context_add_block_io_device_weight_dup(dst
, l
);
467 LIST_FOREACH_BACKWARDS(device_bandwidths
, l
, LIST_FIND_TAIL(device_bandwidths
, src
->blockio_device_bandwidths
)) {
468 r
= cgroup_context_add_block_io_device_bandwidth_dup(dst
, l
);
473 dst
->memory_limit
= src
->memory_limit
;
475 dst
->device_policy
= src
->device_policy
;
476 LIST_FOREACH_BACKWARDS(device_allow
, l
, LIST_FIND_TAIL(device_allow
, src
->device_allow
)) {
477 r
= cgroup_context_add_device_allow_dup(dst
, l
);
482 LIST_FOREACH_BACKWARDS(socket_bind_items
, l
, LIST_FIND_TAIL(socket_bind_items
, src
->socket_bind_allow
)) {
483 r
= cgroup_context_add_socket_bind_item_allow_dup(dst
, l
);
489 LIST_FOREACH_BACKWARDS(socket_bind_items
, l
, LIST_FIND_TAIL(socket_bind_items
, src
->socket_bind_deny
)) {
490 r
= cgroup_context_add_socket_bind_item_deny_dup(dst
, l
);
495 dst
->tasks_max
= src
->tasks_max
;
500 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
504 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
509 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
513 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
518 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
522 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
527 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
531 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
536 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
540 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
545 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
549 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
554 void cgroup_context_remove_bpf_foreign_program(CGroupContext
*c
, CGroupBPFForeignProgram
*p
) {
558 LIST_REMOVE(programs
, c
->bpf_foreign_programs
, p
);
563 void cgroup_context_remove_socket_bind(CGroupSocketBindItem
**head
) {
566 LIST_CLEAR(socket_bind_items
, *head
, free
);
569 void cgroup_context_done(CGroupContext
*c
) {
572 while (c
->io_device_weights
)
573 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
575 while (c
->io_device_latencies
)
576 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
578 while (c
->io_device_limits
)
579 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
581 while (c
->blockio_device_weights
)
582 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
584 while (c
->blockio_device_bandwidths
)
585 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
587 while (c
->device_allow
)
588 cgroup_context_free_device_allow(c
, c
->device_allow
);
590 cgroup_context_remove_socket_bind(&c
->socket_bind_allow
);
591 cgroup_context_remove_socket_bind(&c
->socket_bind_deny
);
593 c
->ip_address_allow
= set_free(c
->ip_address_allow
);
594 c
->ip_address_deny
= set_free(c
->ip_address_deny
);
596 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
597 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
599 while (c
->bpf_foreign_programs
)
600 cgroup_context_remove_bpf_foreign_program(c
, c
->bpf_foreign_programs
);
602 c
->restrict_network_interfaces
= set_free_free(c
->restrict_network_interfaces
);
604 cpu_set_reset(&c
->cpuset_cpus
);
605 cpu_set_reset(&c
->startup_cpuset_cpus
);
606 cpu_set_reset(&c
->cpuset_mems
);
607 cpu_set_reset(&c
->startup_cpuset_mems
);
609 c
->delegate_subgroup
= mfree(c
->delegate_subgroup
);
611 nft_set_context_clear(&c
->nft_set_context
);
614 static int unit_get_kernel_memory_limit(Unit
*u
, const char *file
, uint64_t *ret
) {
617 if (!u
->cgroup_realized
)
620 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, file
, ret
);
623 static int unit_compare_memory_limit(Unit
*u
, const char *property_name
, uint64_t *ret_unit_value
, uint64_t *ret_kernel_value
) {
630 /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will
631 * return -ENODATA) on cgroup v1.
636 * 0: If the kernel memory setting doesn't match our configuration.
637 * >0: If the kernel memory setting matches our configuration.
639 * The following values are only guaranteed to be populated on return >=0:
641 * - ret_unit_value will contain our internal expected value for the unit, page-aligned.
642 * - ret_kernel_value will contain the actual value presented by the kernel. */
646 r
= cg_all_unified();
648 return log_debug_errno(r
, "Failed to determine cgroup hierarchy version: %m");
650 /* Unsupported on v1.
652 * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has
653 * silently masked the controller. */
657 /* The root slice doesn't have any controller files, so we can't compare anything. */
658 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
661 /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled,
662 * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To
663 * avoid specious errors in these scenarios, check that we even expect the memory controller to be
665 m
= unit_get_target_mask(u
);
666 if (!FLAGS_SET(m
, CGROUP_MASK_MEMORY
))
669 assert_se(c
= unit_get_cgroup_context(u
));
671 bool startup
= u
->manager
&& IN_SET(manager_state(u
->manager
), MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
);
673 if (streq(property_name
, "MemoryLow")) {
674 unit_value
= unit_get_ancestor_memory_low(u
);
676 } else if (startup
&& streq(property_name
, "StartupMemoryLow")) {
677 unit_value
= unit_get_ancestor_startup_memory_low(u
);
679 } else if (streq(property_name
, "MemoryMin")) {
680 unit_value
= unit_get_ancestor_memory_min(u
);
682 } else if (streq(property_name
, "MemoryHigh")) {
683 unit_value
= c
->memory_high
;
684 file
= "memory.high";
685 } else if (startup
&& streq(property_name
, "StartupMemoryHigh")) {
686 unit_value
= c
->startup_memory_high
;
687 file
= "memory.high";
688 } else if (streq(property_name
, "MemoryMax")) {
689 unit_value
= c
->memory_max
;
691 } else if (startup
&& streq(property_name
, "StartupMemoryMax")) {
692 unit_value
= c
->startup_memory_max
;
694 } else if (streq(property_name
, "MemorySwapMax")) {
695 unit_value
= c
->memory_swap_max
;
696 file
= "memory.swap.max";
697 } else if (startup
&& streq(property_name
, "StartupMemorySwapMax")) {
698 unit_value
= c
->startup_memory_swap_max
;
699 file
= "memory.swap.max";
700 } else if (streq(property_name
, "MemoryZSwapMax")) {
701 unit_value
= c
->memory_zswap_max
;
702 file
= "memory.zswap.max";
703 } else if (startup
&& streq(property_name
, "StartupMemoryZSwapMax")) {
704 unit_value
= c
->startup_memory_zswap_max
;
705 file
= "memory.zswap.max";
709 r
= unit_get_kernel_memory_limit(u
, file
, ret_kernel_value
);
711 return log_unit_debug_errno(u
, r
, "Failed to parse %s: %m", file
);
713 /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page
714 * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from
715 * our internal page-counting. To support those future kernels, just check the value itself first
716 * without any page-alignment. */
717 if (*ret_kernel_value
== unit_value
) {
718 *ret_unit_value
= unit_value
;
722 /* The current kernel behaviour, by comparison, is that even if you write a particular number of
723 * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel
724 * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is
726 if (unit_value
!= CGROUP_LIMIT_MAX
)
727 unit_value
= PAGE_ALIGN_DOWN(unit_value
);
729 *ret_unit_value
= unit_value
;
731 return *ret_kernel_value
== *ret_unit_value
;
734 #define FORMAT_CGROUP_DIFF_MAX 128
736 static char *format_cgroup_memory_limit_comparison(char *buf
, size_t l
, Unit
*u
, const char *property_name
) {
744 r
= unit_compare_memory_limit(u
, property_name
, &sval
, &kval
);
746 /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1).
747 * In the absence of reliably being able to detect whether memcg swap support is available or not,
748 * only complain if the error is not ENOENT. This is similarly the case for memory.zswap.max relying
749 * on CONFIG_ZSWAP. */
750 if (r
> 0 || IN_SET(r
, -ENODATA
, -EOWNERDEAD
) ||
751 (r
== -ENOENT
&& STR_IN_SET(property_name
,
753 "StartupMemorySwapMax",
755 "StartupMemoryZSwapMax")))
759 (void) snprintf(buf
, l
, " (error getting kernel value: %m)");
761 (void) snprintf(buf
, l
, " (different value in kernel: %" PRIu64
")", kval
);
766 const char *cgroup_device_permissions_to_string(CGroupDevicePermissions p
) {
767 static const char *table
[_CGROUP_DEVICE_PERMISSIONS_MAX
] = {
768 /* Lets simply define a table with every possible combination. As long as those are just 8 we
769 * can get away with it. If this ever grows to more we need to revisit this logic though. */
771 [CGROUP_DEVICE_READ
] = "r",
772 [CGROUP_DEVICE_WRITE
] = "w",
773 [CGROUP_DEVICE_MKNOD
] = "m",
774 [CGROUP_DEVICE_READ
|CGROUP_DEVICE_WRITE
] = "rw",
775 [CGROUP_DEVICE_READ
|CGROUP_DEVICE_MKNOD
] = "rm",
776 [CGROUP_DEVICE_WRITE
|CGROUP_DEVICE_MKNOD
] = "wm",
777 [CGROUP_DEVICE_READ
|CGROUP_DEVICE_WRITE
|CGROUP_DEVICE_MKNOD
] = "rwm",
780 if (p
< 0 || p
>= _CGROUP_DEVICE_PERMISSIONS_MAX
)
786 CGroupDevicePermissions
cgroup_device_permissions_from_string(const char *s
) {
787 CGroupDevicePermissions p
= 0;
790 return _CGROUP_DEVICE_PERMISSIONS_INVALID
;
792 for (const char *c
= s
; *c
; c
++) {
794 p
|= CGROUP_DEVICE_READ
;
796 p
|= CGROUP_DEVICE_WRITE
;
798 p
|= CGROUP_DEVICE_MKNOD
;
800 return _CGROUP_DEVICE_PERMISSIONS_INVALID
;
806 void cgroup_context_dump(Unit
*u
, FILE* f
, const char *prefix
) {
807 _cleanup_free_
char *disable_controllers_str
= NULL
, *delegate_controllers_str
= NULL
, *cpuset_cpus
= NULL
, *cpuset_mems
= NULL
, *startup_cpuset_cpus
= NULL
, *startup_cpuset_mems
= NULL
;
809 struct in_addr_prefix
*iaai
;
811 char cda
[FORMAT_CGROUP_DIFF_MAX
];
812 char cdb
[FORMAT_CGROUP_DIFF_MAX
];
813 char cdc
[FORMAT_CGROUP_DIFF_MAX
];
814 char cdd
[FORMAT_CGROUP_DIFF_MAX
];
815 char cde
[FORMAT_CGROUP_DIFF_MAX
];
816 char cdf
[FORMAT_CGROUP_DIFF_MAX
];
817 char cdg
[FORMAT_CGROUP_DIFF_MAX
];
818 char cdh
[FORMAT_CGROUP_DIFF_MAX
];
819 char cdi
[FORMAT_CGROUP_DIFF_MAX
];
820 char cdj
[FORMAT_CGROUP_DIFF_MAX
];
821 char cdk
[FORMAT_CGROUP_DIFF_MAX
];
826 assert_se(c
= unit_get_cgroup_context(u
));
828 prefix
= strempty(prefix
);
830 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
831 (void) cg_mask_to_string(c
->delegate_controllers
, &delegate_controllers_str
);
833 /* "Delegate=" means "yes, but no controllers". Show this as "(none)". */
834 const char *delegate_str
= delegate_controllers_str
?: c
->delegate
? "(none)" : "no";
836 cpuset_cpus
= cpu_set_to_range_string(&c
->cpuset_cpus
);
837 startup_cpuset_cpus
= cpu_set_to_range_string(&c
->startup_cpuset_cpus
);
838 cpuset_mems
= cpu_set_to_range_string(&c
->cpuset_mems
);
839 startup_cpuset_mems
= cpu_set_to_range_string(&c
->startup_cpuset_mems
);
842 "%sCPUAccounting: %s\n"
843 "%sIOAccounting: %s\n"
844 "%sBlockIOAccounting: %s\n"
845 "%sMemoryAccounting: %s\n"
846 "%sTasksAccounting: %s\n"
847 "%sIPAccounting: %s\n"
848 "%sCPUWeight: %" PRIu64
"\n"
849 "%sStartupCPUWeight: %" PRIu64
"\n"
850 "%sCPUShares: %" PRIu64
"\n"
851 "%sStartupCPUShares: %" PRIu64
"\n"
852 "%sCPUQuotaPerSecSec: %s\n"
853 "%sCPUQuotaPeriodSec: %s\n"
854 "%sAllowedCPUs: %s\n"
855 "%sStartupAllowedCPUs: %s\n"
856 "%sAllowedMemoryNodes: %s\n"
857 "%sStartupAllowedMemoryNodes: %s\n"
858 "%sIOWeight: %" PRIu64
"\n"
859 "%sStartupIOWeight: %" PRIu64
"\n"
860 "%sBlockIOWeight: %" PRIu64
"\n"
861 "%sStartupBlockIOWeight: %" PRIu64
"\n"
862 "%sDefaultMemoryMin: %" PRIu64
"\n"
863 "%sDefaultMemoryLow: %" PRIu64
"\n"
864 "%sMemoryMin: %" PRIu64
"%s\n"
865 "%sMemoryLow: %" PRIu64
"%s\n"
866 "%sStartupMemoryLow: %" PRIu64
"%s\n"
867 "%sMemoryHigh: %" PRIu64
"%s\n"
868 "%sStartupMemoryHigh: %" PRIu64
"%s\n"
869 "%sMemoryMax: %" PRIu64
"%s\n"
870 "%sStartupMemoryMax: %" PRIu64
"%s\n"
871 "%sMemorySwapMax: %" PRIu64
"%s\n"
872 "%sStartupMemorySwapMax: %" PRIu64
"%s\n"
873 "%sMemoryZSwapMax: %" PRIu64
"%s\n"
874 "%sStartupMemoryZSwapMax: %" PRIu64
"%s\n"
875 "%sMemoryLimit: %" PRIu64
"\n"
876 "%sTasksMax: %" PRIu64
"\n"
877 "%sDevicePolicy: %s\n"
878 "%sDisableControllers: %s\n"
880 "%sManagedOOMSwap: %s\n"
881 "%sManagedOOMMemoryPressure: %s\n"
882 "%sManagedOOMMemoryPressureLimit: " PERMYRIAD_AS_PERCENT_FORMAT_STR
"\n"
883 "%sManagedOOMPreference: %s\n"
884 "%sMemoryPressureWatch: %s\n"
885 "%sCoredumpReceive: %s\n",
886 prefix
, yes_no(c
->cpu_accounting
),
887 prefix
, yes_no(c
->io_accounting
),
888 prefix
, yes_no(c
->blockio_accounting
),
889 prefix
, yes_no(c
->memory_accounting
),
890 prefix
, yes_no(c
->tasks_accounting
),
891 prefix
, yes_no(c
->ip_accounting
),
892 prefix
, c
->cpu_weight
,
893 prefix
, c
->startup_cpu_weight
,
894 prefix
, c
->cpu_shares
,
895 prefix
, c
->startup_cpu_shares
,
896 prefix
, FORMAT_TIMESPAN(c
->cpu_quota_per_sec_usec
, 1),
897 prefix
, FORMAT_TIMESPAN(c
->cpu_quota_period_usec
, 1),
898 prefix
, strempty(cpuset_cpus
),
899 prefix
, strempty(startup_cpuset_cpus
),
900 prefix
, strempty(cpuset_mems
),
901 prefix
, strempty(startup_cpuset_mems
),
902 prefix
, c
->io_weight
,
903 prefix
, c
->startup_io_weight
,
904 prefix
, c
->blockio_weight
,
905 prefix
, c
->startup_blockio_weight
,
906 prefix
, c
->default_memory_min
,
907 prefix
, c
->default_memory_low
,
908 prefix
, c
->memory_min
, format_cgroup_memory_limit_comparison(cda
, sizeof(cda
), u
, "MemoryMin"),
909 prefix
, c
->memory_low
, format_cgroup_memory_limit_comparison(cdb
, sizeof(cdb
), u
, "MemoryLow"),
910 prefix
, c
->startup_memory_low
, format_cgroup_memory_limit_comparison(cdc
, sizeof(cdc
), u
, "StartupMemoryLow"),
911 prefix
, c
->memory_high
, format_cgroup_memory_limit_comparison(cdd
, sizeof(cdd
), u
, "MemoryHigh"),
912 prefix
, c
->startup_memory_high
, format_cgroup_memory_limit_comparison(cde
, sizeof(cde
), u
, "StartupMemoryHigh"),
913 prefix
, c
->memory_max
, format_cgroup_memory_limit_comparison(cdf
, sizeof(cdf
), u
, "MemoryMax"),
914 prefix
, c
->startup_memory_max
, format_cgroup_memory_limit_comparison(cdg
, sizeof(cdg
), u
, "StartupMemoryMax"),
915 prefix
, c
->memory_swap_max
, format_cgroup_memory_limit_comparison(cdh
, sizeof(cdh
), u
, "MemorySwapMax"),
916 prefix
, c
->startup_memory_swap_max
, format_cgroup_memory_limit_comparison(cdi
, sizeof(cdi
), u
, "StartupMemorySwapMax"),
917 prefix
, c
->memory_zswap_max
, format_cgroup_memory_limit_comparison(cdj
, sizeof(cdj
), u
, "MemoryZSwapMax"),
918 prefix
, c
->startup_memory_zswap_max
, format_cgroup_memory_limit_comparison(cdk
, sizeof(cdk
), u
, "StartupMemoryZSwapMax"),
919 prefix
, c
->memory_limit
,
920 prefix
, cgroup_tasks_max_resolve(&c
->tasks_max
),
921 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
922 prefix
, strempty(disable_controllers_str
),
923 prefix
, delegate_str
,
924 prefix
, managed_oom_mode_to_string(c
->moom_swap
),
925 prefix
, managed_oom_mode_to_string(c
->moom_mem_pressure
),
926 prefix
, PERMYRIAD_AS_PERCENT_FORMAT_VAL(UINT32_SCALE_TO_PERMYRIAD(c
->moom_mem_pressure_limit
)),
927 prefix
, managed_oom_preference_to_string(c
->moom_preference
),
928 prefix
, cgroup_pressure_watch_to_string(c
->memory_pressure_watch
),
929 prefix
, yes_no(c
->coredump_receive
));
931 if (c
->delegate_subgroup
)
932 fprintf(f
, "%sDelegateSubgroup: %s\n",
933 prefix
, c
->delegate_subgroup
);
935 if (c
->memory_pressure_threshold_usec
!= USEC_INFINITY
)
936 fprintf(f
, "%sMemoryPressureThresholdSec: %s\n",
937 prefix
, FORMAT_TIMESPAN(c
->memory_pressure_threshold_usec
, 1));
939 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
940 /* strna() below should be redundant, for avoiding -Werror=format-overflow= error. See #30223. */
942 "%sDeviceAllow: %s %s\n",
945 strna(cgroup_device_permissions_to_string(a
->permissions
)));
947 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
949 "%sIODeviceWeight: %s %" PRIu64
"\n",
954 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
956 "%sIODeviceLatencyTargetSec: %s %s\n",
959 FORMAT_TIMESPAN(l
->target_usec
, 1));
961 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
)
962 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
963 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
967 cgroup_io_limit_type_to_string(type
),
969 FORMAT_BYTES(il
->limits
[type
]));
971 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
973 "%sBlockIODeviceWeight: %s %" PRIu64
,
978 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
979 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
981 "%sBlockIOReadBandwidth: %s %s\n",
984 FORMAT_BYTES(b
->rbps
));
985 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
987 "%sBlockIOWriteBandwidth: %s %s\n",
990 FORMAT_BYTES(b
->wbps
));
993 SET_FOREACH(iaai
, c
->ip_address_allow
)
994 fprintf(f
, "%sIPAddressAllow: %s\n", prefix
,
995 IN_ADDR_PREFIX_TO_STRING(iaai
->family
, &iaai
->address
, iaai
->prefixlen
));
996 SET_FOREACH(iaai
, c
->ip_address_deny
)
997 fprintf(f
, "%sIPAddressDeny: %s\n", prefix
,
998 IN_ADDR_PREFIX_TO_STRING(iaai
->family
, &iaai
->address
, iaai
->prefixlen
));
1000 STRV_FOREACH(path
, c
->ip_filters_ingress
)
1001 fprintf(f
, "%sIPIngressFilterPath: %s\n", prefix
, *path
);
1002 STRV_FOREACH(path
, c
->ip_filters_egress
)
1003 fprintf(f
, "%sIPEgressFilterPath: %s\n", prefix
, *path
);
1005 LIST_FOREACH(programs
, p
, c
->bpf_foreign_programs
)
1006 fprintf(f
, "%sBPFProgram: %s:%s",
1007 prefix
, bpf_cgroup_attach_type_to_string(p
->attach_type
), p
->bpffs_path
);
1009 if (c
->socket_bind_allow
) {
1010 fprintf(f
, "%sSocketBindAllow: ", prefix
);
1011 cgroup_context_dump_socket_bind_items(c
->socket_bind_allow
, f
);
1015 if (c
->socket_bind_deny
) {
1016 fprintf(f
, "%sSocketBindDeny: ", prefix
);
1017 cgroup_context_dump_socket_bind_items(c
->socket_bind_deny
, f
);
1021 if (c
->restrict_network_interfaces
) {
1023 SET_FOREACH(iface
, c
->restrict_network_interfaces
)
1024 fprintf(f
, "%sRestrictNetworkInterfaces: %s\n", prefix
, iface
);
1027 FOREACH_ARRAY(nft_set
, c
->nft_set_context
.sets
, c
->nft_set_context
.n_sets
)
1028 fprintf(f
, "%sNFTSet: %s:%s:%s:%s\n", prefix
, nft_set_source_to_string(nft_set
->source
),
1029 nfproto_to_string(nft_set
->nfproto
), nft_set
->table
, nft_set
->set
);
1032 void cgroup_context_dump_socket_bind_item(const CGroupSocketBindItem
*item
, FILE *f
) {
1033 const char *family
, *colon1
, *protocol
= "", *colon2
= "";
1035 family
= strempty(af_to_ipv4_ipv6(item
->address_family
));
1036 colon1
= isempty(family
) ? "" : ":";
1038 if (item
->ip_protocol
!= 0) {
1039 protocol
= ip_protocol_to_tcp_udp(item
->ip_protocol
);
1043 if (item
->nr_ports
== 0)
1044 fprintf(f
, "%s%s%s%sany", family
, colon1
, protocol
, colon2
);
1045 else if (item
->nr_ports
== 1)
1046 fprintf(f
, "%s%s%s%s%" PRIu16
, family
, colon1
, protocol
, colon2
, item
->port_min
);
1048 uint16_t port_max
= item
->port_min
+ item
->nr_ports
- 1;
1049 fprintf(f
, "%s%s%s%s%" PRIu16
"-%" PRIu16
, family
, colon1
, protocol
, colon2
,
1050 item
->port_min
, port_max
);
1054 void cgroup_context_dump_socket_bind_items(const CGroupSocketBindItem
*items
, FILE *f
) {
1057 LIST_FOREACH(socket_bind_items
, bi
, items
) {
1063 cgroup_context_dump_socket_bind_item(bi
, f
);
1067 int cgroup_context_add_device_allow(CGroupContext
*c
, const char *dev
, CGroupDevicePermissions p
) {
1068 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
1069 _cleanup_free_
char *d
= NULL
;
1073 assert(p
>= 0 && p
< _CGROUP_DEVICE_PERMISSIONS_MAX
);
1076 p
= _CGROUP_DEVICE_PERMISSIONS_ALL
;
1078 a
= new(CGroupDeviceAllow
, 1);
1086 *a
= (CGroupDeviceAllow
) {
1087 .path
= TAKE_PTR(d
),
1091 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
1097 int cgroup_context_add_or_update_device_allow(CGroupContext
*c
, const char *dev
, CGroupDevicePermissions p
) {
1100 assert(p
>= 0 && p
< _CGROUP_DEVICE_PERMISSIONS_MAX
);
1103 p
= _CGROUP_DEVICE_PERMISSIONS_ALL
;
1105 LIST_FOREACH(device_allow
, b
, c
->device_allow
)
1106 if (path_equal(b
->path
, dev
)) {
1111 return cgroup_context_add_device_allow(c
, dev
, p
);
1114 int cgroup_context_add_bpf_foreign_program(CGroupContext
*c
, uint32_t attach_type
, const char *bpffs_path
) {
1115 CGroupBPFForeignProgram
*p
;
1116 _cleanup_free_
char *d
= NULL
;
1121 if (!path_is_normalized(bpffs_path
) || !path_is_absolute(bpffs_path
))
1122 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Path is not normalized: %m");
1124 d
= strdup(bpffs_path
);
1128 p
= new(CGroupBPFForeignProgram
, 1);
1132 *p
= (CGroupBPFForeignProgram
) {
1133 .attach_type
= attach_type
,
1134 .bpffs_path
= TAKE_PTR(d
),
1137 LIST_PREPEND(programs
, c
->bpf_foreign_programs
, TAKE_PTR(p
));
1142 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
1143 uint64_t unit_get_ancestor_##entry(Unit *u) { \
1146 /* 1. Is entry set in this unit? If so, use that. \
1147 * 2. Is the default for this entry set in any \
1148 * ancestor? If so, use that. \
1149 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
1153 c = unit_get_cgroup_context(u); \
1154 if (c && c->entry##_set) \
1157 while ((u = UNIT_GET_SLICE(u))) { \
1158 c = unit_get_cgroup_context(u); \
1159 if (c && c->default_##entry##_set) \
1160 return c->default_##entry; \
1163 /* We've reached the root, but nobody had default for \
1164 * this entry set, so set it to the kernel default. */ \
1165 return CGROUP_LIMIT_MIN; \
1168 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
1169 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(startup_memory_low
);
1170 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
1172 static void unit_set_xattr_graceful(Unit
*u
, const char *name
, const void *data
, size_t size
) {
1178 if (!u
->cgroup_path
)
1181 r
= cg_set_xattr(u
->cgroup_path
, name
, data
, size
, 0);
1183 log_unit_debug_errno(u
, r
, "Failed to set '%s' xattr on control group %s, ignoring: %m", name
, empty_to_root(u
->cgroup_path
));
1186 static void unit_remove_xattr_graceful(Unit
*u
, const char *name
) {
1192 if (!u
->cgroup_path
)
1195 r
= cg_remove_xattr(u
->cgroup_path
, name
);
1196 if (r
< 0 && !ERRNO_IS_XATTR_ABSENT(r
))
1197 log_unit_debug_errno(u
, r
, "Failed to remove '%s' xattr flag on control group %s, ignoring: %m", name
, empty_to_root(u
->cgroup_path
));
1200 static void cgroup_oomd_xattr_apply(Unit
*u
) {
1205 c
= unit_get_cgroup_context(u
);
1209 if (c
->moom_preference
== MANAGED_OOM_PREFERENCE_OMIT
)
1210 unit_set_xattr_graceful(u
, "user.oomd_omit", "1", 1);
1212 if (c
->moom_preference
== MANAGED_OOM_PREFERENCE_AVOID
)
1213 unit_set_xattr_graceful(u
, "user.oomd_avoid", "1", 1);
1215 if (c
->moom_preference
!= MANAGED_OOM_PREFERENCE_AVOID
)
1216 unit_remove_xattr_graceful(u
, "user.oomd_avoid");
1218 if (c
->moom_preference
!= MANAGED_OOM_PREFERENCE_OMIT
)
1219 unit_remove_xattr_graceful(u
, "user.oomd_omit");
1222 static int cgroup_log_xattr_apply(Unit
*u
) {
1224 size_t len
, allowed_patterns_len
, denied_patterns_len
;
1225 _cleanup_free_
char *patterns
= NULL
, *allowed_patterns
= NULL
, *denied_patterns
= NULL
;
1231 c
= unit_get_exec_context(u
);
1233 /* Some unit types have a cgroup context but no exec context, so we do not log
1234 * any error here to avoid confusion. */
1237 if (set_isempty(c
->log_filter_allowed_patterns
) && set_isempty(c
->log_filter_denied_patterns
)) {
1238 unit_remove_xattr_graceful(u
, "user.journald_log_filter_patterns");
1242 r
= set_make_nulstr(c
->log_filter_allowed_patterns
, &allowed_patterns
, &allowed_patterns_len
);
1244 return log_debug_errno(r
, "Failed to make nulstr from set: %m");
1246 r
= set_make_nulstr(c
->log_filter_denied_patterns
, &denied_patterns
, &denied_patterns_len
);
1248 return log_debug_errno(r
, "Failed to make nulstr from set: %m");
1250 /* Use nul character separated strings without trailing nul */
1251 allowed_patterns_len
= LESS_BY(allowed_patterns_len
, 1u);
1252 denied_patterns_len
= LESS_BY(denied_patterns_len
, 1u);
1254 len
= allowed_patterns_len
+ 1 + denied_patterns_len
;
1255 patterns
= new(char, len
);
1257 return log_oom_debug();
1259 last
= mempcpy_safe(patterns
, allowed_patterns
, allowed_patterns_len
);
1261 memcpy_safe(last
, denied_patterns
, denied_patterns_len
);
1263 unit_set_xattr_graceful(u
, "user.journald_log_filter_patterns", patterns
, len
);
1268 static void cgroup_invocation_id_xattr_apply(Unit
*u
) {
1273 b
= !sd_id128_is_null(u
->invocation_id
);
1274 FOREACH_STRING(xn
, "trusted.invocation_id", "user.invocation_id") {
1276 unit_set_xattr_graceful(u
, xn
, SD_ID128_TO_STRING(u
->invocation_id
), 32);
1278 unit_remove_xattr_graceful(u
, xn
);
1282 static void cgroup_coredump_xattr_apply(Unit
*u
) {
1287 c
= unit_get_cgroup_context(u
);
1291 if (unit_cgroup_delegate(u
) && c
->coredump_receive
)
1292 unit_set_xattr_graceful(u
, "user.coredump_receive", "1", 1);
1294 unit_remove_xattr_graceful(u
, "user.coredump_receive");
1297 static void cgroup_delegate_xattr_apply(Unit
*u
) {
1302 /* Indicate on the cgroup whether delegation is on, via an xattr. This is best-effort, as old kernels
1303 * didn't support xattrs on cgroups at all. Later they got support for setting 'trusted.*' xattrs,
1304 * and even later 'user.*' xattrs. We started setting this field when 'trusted.*' was added, and
1305 * given this is now pretty much API, let's continue to support that. But also set 'user.*' as well,
1306 * since it is readable by any user, not just CAP_SYS_ADMIN. This hence comes with slightly weaker
1307 * security (as users who got delegated cgroups could turn it off if they like), but this shouldn't
1308 * be a big problem given this communicates delegation state to clients, but the manager never reads
1310 b
= unit_cgroup_delegate(u
);
1311 FOREACH_STRING(xn
, "trusted.delegate", "user.delegate") {
1313 unit_set_xattr_graceful(u
, xn
, "1", 1);
1315 unit_remove_xattr_graceful(u
, xn
);
1319 static void cgroup_survive_xattr_apply(Unit
*u
) {
1324 if (u
->survive_final_kill_signal
) {
1327 "user.survive_final_kill_signal",
1331 /* user xattr support was added in kernel v5.7 */
1332 if (ERRNO_IS_NEG_NOT_SUPPORTED(r
))
1335 "trusted.survive_final_kill_signal",
1340 log_unit_debug_errno(u
,
1342 "Failed to set 'survive_final_kill_signal' xattr on control "
1343 "group %s, ignoring: %m",
1344 empty_to_root(u
->cgroup_path
));
1346 unit_remove_xattr_graceful(u
, "user.survive_final_kill_signal");
1347 unit_remove_xattr_graceful(u
, "trusted.survive_final_kill_signal");
1351 static void cgroup_xattr_apply(Unit
*u
) {
1354 /* The 'user.*' xattrs can be set from a user manager. */
1355 cgroup_oomd_xattr_apply(u
);
1356 cgroup_log_xattr_apply(u
);
1357 cgroup_coredump_xattr_apply(u
);
1359 if (!MANAGER_IS_SYSTEM(u
->manager
))
1362 cgroup_invocation_id_xattr_apply(u
);
1363 cgroup_delegate_xattr_apply(u
);
1364 cgroup_survive_xattr_apply(u
);
1367 static int lookup_block_device(const char *p
, dev_t
*ret
) {
1368 dev_t rdev
, dev
= 0;
1375 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
1376 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
1379 if (stat(p
, &st
) < 0)
1380 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
1386 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
1389 return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK
),
1390 "Device node '%s' is a character device, but block device needed.", p
);
1393 else if (major(dev
) != 0)
1394 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
1396 /* If this is btrfs, getting the backing block device is a bit harder */
1397 r
= btrfs_get_block_device(p
, ret
);
1399 return log_warning_errno(SYNTHETIC_ERRNO(ENODEV
),
1400 "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
1402 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
1405 /* If this is a LUKS/DM device, recursively try to get the originating block device */
1406 while (block_get_originating(*ret
, ret
) > 0);
1408 /* If this is a partition, try to get the originating block device */
1409 (void) block_get_whole_disk(*ret
, ret
);
1413 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
1414 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
1415 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
1418 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
1419 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
1420 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
1423 static bool cgroup_context_has_allowed_cpus(CGroupContext
*c
) {
1424 return c
->cpuset_cpus
.set
|| c
->startup_cpuset_cpus
.set
;
1427 static bool cgroup_context_has_allowed_mems(CGroupContext
*c
) {
1428 return c
->cpuset_mems
.set
|| c
->startup_cpuset_mems
.set
;
1431 uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
1434 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1435 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
1436 return c
->startup_cpu_weight
;
1437 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
1438 return c
->cpu_weight
;
1440 return CGROUP_WEIGHT_DEFAULT
;
1443 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
1444 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1445 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
1446 return c
->startup_cpu_shares
;
1447 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
1448 return c
->cpu_shares
;
1450 return CGROUP_CPU_SHARES_DEFAULT
;
1453 static CPUSet
*cgroup_context_allowed_cpus(CGroupContext
*c
, ManagerState state
) {
1454 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1455 c
->startup_cpuset_cpus
.set
)
1456 return &c
->startup_cpuset_cpus
;
1458 return &c
->cpuset_cpus
;
1461 static CPUSet
*cgroup_context_allowed_mems(CGroupContext
*c
, ManagerState state
) {
1462 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1463 c
->startup_cpuset_mems
.set
)
1464 return &c
->startup_cpuset_mems
;
1466 return &c
->cpuset_mems
;
1469 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
1470 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
1471 * need to be higher than that boundary. quota is specified in USecPerSec.
1472 * Additionally, period must be at most max_period. */
1475 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
1478 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
1481 if (quota
== USEC_INFINITY
)
1482 /* Always use default period for infinity quota. */
1483 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
1485 if (period
== USEC_INFINITY
)
1486 /* Default period was requested. */
1487 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
1489 /* Clamp to interval [1ms, 1s] */
1490 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
1492 if (new_period
!= period
) {
1493 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
,
1494 "Clamping CPU interval for cpu.max: period is now %s",
1495 FORMAT_TIMESPAN(new_period
, 1));
1496 u
->warned_clamping_cpu_quota_period
= true;
1502 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
1503 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
1505 if (weight
== CGROUP_WEIGHT_IDLE
)
1507 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1508 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
1511 static void cgroup_apply_unified_cpu_idle(Unit
*u
, uint64_t weight
) {
1514 const char *idle_val
;
1516 is_idle
= weight
== CGROUP_WEIGHT_IDLE
;
1517 idle_val
= one_zero(is_idle
);
1518 r
= cg_set_attribute("cpu", u
->cgroup_path
, "cpu.idle", idle_val
);
1519 if (r
< 0 && (r
!= -ENOENT
|| is_idle
))
1520 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%s': %m",
1521 "cpu.idle", empty_to_root(u
->cgroup_path
), idle_val
);
1524 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
1525 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
1527 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
1528 if (quota
!= USEC_INFINITY
)
1529 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
1530 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
1532 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
1533 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
1536 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
1537 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
1539 xsprintf(buf
, "%" PRIu64
"\n", shares
);
1540 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
1543 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
1544 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
1546 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
1548 xsprintf(buf
, USEC_FMT
"\n", period
);
1549 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
1551 if (quota
!= USEC_INFINITY
) {
1552 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
1553 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
1555 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
1558 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
1559 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
1560 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
1563 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
1564 /* we don't support idle in cgroupv1 */
1565 if (weight
== CGROUP_WEIGHT_IDLE
)
1566 return CGROUP_CPU_SHARES_MIN
;
1568 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
1569 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
1572 static void cgroup_apply_unified_cpuset(Unit
*u
, const CPUSet
*cpus
, const char *name
) {
1573 _cleanup_free_
char *buf
= NULL
;
1575 buf
= cpu_set_to_range_string(cpus
);
1581 (void) set_attribute_and_warn(u
, "cpuset", name
, buf
);
1584 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
1585 return c
->io_accounting
||
1586 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
1587 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
1588 c
->io_device_weights
||
1589 c
->io_device_latencies
||
1590 c
->io_device_limits
;
1593 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
1594 return c
->blockio_accounting
||
1595 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
1596 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
1597 c
->blockio_device_weights
||
1598 c
->blockio_device_bandwidths
;
1601 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
1602 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1603 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
1604 return c
->startup_io_weight
;
1605 if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
1606 return c
->io_weight
;
1607 return CGROUP_WEIGHT_DEFAULT
;
1610 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
1611 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1612 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
1613 return c
->startup_blockio_weight
;
1614 if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
1615 return c
->blockio_weight
;
1616 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
1619 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
1620 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
1621 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
1624 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
1625 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
1626 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
1629 static int set_bfq_weight(Unit
*u
, const char *controller
, dev_t dev
, uint64_t io_weight
) {
1630 static const char * const prop_names
[] = {
1634 "BlockIODeviceWeight",
1636 static bool warned
= false;
1637 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+STRLEN("\n")];
1639 uint64_t bfq_weight
;
1642 /* FIXME: drop this function when distro kernels properly support BFQ through "io.weight"
1643 * See also: https://github.com/systemd/systemd/pull/13335 and
1644 * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9. */
1645 p
= strjoina(controller
, ".bfq.weight");
1646 /* Adjust to kernel range is 1..1000, the default is 100. */
1647 bfq_weight
= BFQ_WEIGHT(io_weight
);
1650 xsprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), bfq_weight
);
1652 xsprintf(buf
, "%" PRIu64
"\n", bfq_weight
);
1654 r
= cg_set_attribute(controller
, u
->cgroup_path
, p
, buf
);
1656 /* FIXME: drop this when kernels prior
1657 * 795fe54c2a82 ("bfq: Add per-device weight") v5.4
1658 * are not interesting anymore. Old kernels will fail with EINVAL, while new kernels won't return
1659 * EINVAL on properly formatted input by us. Treat EINVAL accordingly. */
1660 if (r
== -EINVAL
&& major(dev
) > 0) {
1662 log_unit_warning(u
, "Kernel version does not accept per-device setting in %s.", p
);
1665 r
= -EOPNOTSUPP
; /* mask as unconfigured device */
1666 } else if (r
>= 0 && io_weight
!= bfq_weight
)
1667 log_unit_debug(u
, "%s=%" PRIu64
" scaled to %s=%" PRIu64
,
1668 prop_names
[2*(major(dev
) > 0) + streq(controller
, "blkio")],
1669 io_weight
, p
, bfq_weight
);
1673 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
1674 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1678 if (lookup_block_device(dev_path
, &dev
) < 0)
1681 r1
= set_bfq_weight(u
, "io", dev
, io_weight
);
1683 xsprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), io_weight
);
1684 r2
= cg_set_attribute("io", u
->cgroup_path
, "io.weight", buf
);
1686 /* Look at the configured device, when both fail, prefer io.weight errno. */
1687 r
= r2
== -EOPNOTSUPP
? r1
: r2
;
1690 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
),
1691 r
, "Failed to set 'io[.bfq].weight' attribute on '%s' to '%.*s': %m",
1692 empty_to_root(u
->cgroup_path
), (int) strcspn(buf
, NEWLINE
), buf
);
1695 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
1696 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1700 r
= lookup_block_device(dev_path
, &dev
);
1704 xsprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), blkio_weight
);
1705 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
1708 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
1709 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
1713 r
= lookup_block_device(dev_path
, &dev
);
1717 if (target
!= USEC_INFINITY
)
1718 xsprintf(buf
, DEVNUM_FORMAT_STR
" target=%" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), target
);
1720 xsprintf(buf
, DEVNUM_FORMAT_STR
" target=max\n", DEVNUM_FORMAT_VAL(dev
));
1722 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
1725 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
1726 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)],
1727 buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
1730 if (lookup_block_device(dev_path
, &dev
) < 0)
1733 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1734 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
1735 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
1737 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
1739 xsprintf(buf
, DEVNUM_FORMAT_STR
" rbps=%s wbps=%s riops=%s wiops=%s\n", DEVNUM_FORMAT_VAL(dev
),
1740 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
1741 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
1742 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
1745 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
1746 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1749 if (lookup_block_device(dev_path
, &dev
) < 0)
1752 sprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), rbps
);
1753 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
1755 sprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), wbps
);
1756 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
1759 static bool unit_has_unified_memory_config(Unit
*u
) {
1764 assert_se(c
= unit_get_cgroup_context(u
));
1766 return unit_get_ancestor_memory_min(u
) > 0 ||
1767 unit_get_ancestor_memory_low(u
) > 0 || unit_get_ancestor_startup_memory_low(u
) > 0 ||
1768 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->startup_memory_high_set
||
1769 c
->memory_max
!= CGROUP_LIMIT_MAX
|| c
->startup_memory_max_set
||
1770 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
|| c
->startup_memory_swap_max_set
||
1771 c
->memory_zswap_max
!= CGROUP_LIMIT_MAX
|| c
->startup_memory_zswap_max_set
;
1774 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
1775 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
1777 if (v
!= CGROUP_LIMIT_MAX
)
1778 xsprintf(buf
, "%" PRIu64
"\n", v
);
1780 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
1783 static void cgroup_apply_firewall(Unit
*u
) {
1786 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
1788 if (bpf_firewall_compile(u
) < 0)
1791 (void) bpf_firewall_load_custom(u
);
1792 (void) bpf_firewall_install(u
);
1795 void unit_modify_nft_set(Unit
*u
, bool add
) {
1800 if (!MANAGER_IS_SYSTEM(u
->manager
))
1803 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1806 if (cg_all_unified() <= 0)
1809 if (u
->cgroup_id
== 0)
1812 if (!u
->manager
->fw_ctx
) {
1813 r
= fw_ctx_new_full(&u
->manager
->fw_ctx
, /* init_tables= */ false);
1817 assert(u
->manager
->fw_ctx
);
1820 CGroupContext
*c
= ASSERT_PTR(unit_get_cgroup_context(u
));
1822 FOREACH_ARRAY(nft_set
, c
->nft_set_context
.sets
, c
->nft_set_context
.n_sets
) {
1823 if (nft_set
->source
!= NFT_SET_SOURCE_CGROUP
)
1826 uint64_t element
= u
->cgroup_id
;
1828 r
= nft_set_element_modify_any(u
->manager
->fw_ctx
, add
, nft_set
->nfproto
, nft_set
->table
, nft_set
->set
, &element
, sizeof(element
));
1830 log_warning_errno(r
, "Failed to %s NFT set: family %s, table %s, set %s, cgroup %" PRIu64
", ignoring: %m",
1831 add
? "add" : "delete", nfproto_to_string(nft_set
->nfproto
), nft_set
->table
, nft_set
->set
, u
->cgroup_id
);
1833 log_debug("%s NFT set: family %s, table %s, set %s, cgroup %" PRIu64
,
1834 add
? "Added" : "Deleted", nfproto_to_string(nft_set
->nfproto
), nft_set
->table
, nft_set
->set
, u
->cgroup_id
);
1838 static void cgroup_apply_socket_bind(Unit
*u
) {
1841 (void) bpf_socket_bind_install(u
);
1844 static void cgroup_apply_restrict_network_interfaces(Unit
*u
) {
1847 (void) restrict_network_interfaces_install(u
);
1850 static int cgroup_apply_devices(Unit
*u
) {
1851 _cleanup_(bpf_program_freep
) BPFProgram
*prog
= NULL
;
1854 CGroupDevicePolicy policy
;
1857 assert_se(c
= unit_get_cgroup_context(u
));
1858 assert_se(path
= u
->cgroup_path
);
1860 policy
= c
->device_policy
;
1862 if (cg_all_unified() > 0) {
1863 r
= bpf_devices_cgroup_init(&prog
, policy
, c
->device_allow
);
1865 return log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
1868 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore
1871 if (c
->device_allow
|| policy
!= CGROUP_DEVICE_POLICY_AUTO
)
1872 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
1874 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
1876 log_unit_full_errno(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1877 "Failed to reset devices.allow/devices.deny: %m");
1880 bool allow_list_static
= policy
== CGROUP_DEVICE_POLICY_CLOSED
||
1881 (policy
== CGROUP_DEVICE_POLICY_AUTO
&& c
->device_allow
);
1882 if (allow_list_static
)
1883 (void) bpf_devices_allow_list_static(prog
, path
);
1885 bool any
= allow_list_static
;
1886 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1889 if (a
->permissions
== 0)
1892 if (path_startswith(a
->path
, "/dev/"))
1893 r
= bpf_devices_allow_list_device(prog
, path
, a
->path
, a
->permissions
);
1894 else if ((val
= startswith(a
->path
, "block-")))
1895 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'b', a
->permissions
);
1896 else if ((val
= startswith(a
->path
, "char-")))
1897 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'c', a
->permissions
);
1899 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1908 log_unit_warning_errno(u
, SYNTHETIC_ERRNO(ENODEV
), "No devices matched by device filter.");
1910 /* The kernel verifier would reject a program we would build with the normal intro and outro
1911 but no allow-listing rules (outro would contain an unreachable instruction for successful
1913 policy
= CGROUP_DEVICE_POLICY_STRICT
;
1916 r
= bpf_devices_apply_policy(&prog
, policy
, any
, path
, &u
->bpf_device_control_installed
);
1918 static bool warned
= false;
1920 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1921 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1922 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1923 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1930 static void set_io_weight(Unit
*u
, uint64_t weight
) {
1931 char buf
[STRLEN("default \n")+DECIMAL_STR_MAX(uint64_t)];
1935 (void) set_bfq_weight(u
, "io", makedev(0, 0), weight
);
1937 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1938 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1941 static void set_blkio_weight(Unit
*u
, uint64_t weight
) {
1942 char buf
[STRLEN("\n")+DECIMAL_STR_MAX(uint64_t)];
1946 (void) set_bfq_weight(u
, "blkio", makedev(0, 0), weight
);
1948 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1949 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1952 static void cgroup_apply_bpf_foreign_program(Unit
*u
) {
1955 (void) bpf_foreign_install(u
);
1958 static void cgroup_context_apply(
1960 CGroupMask apply_mask
,
1961 ManagerState state
) {
1965 bool is_host_root
, is_local_root
;
1970 /* Nothing to do? Exit early! */
1971 if (apply_mask
== 0)
1974 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
1975 * attributes should only be managed for cgroups further down the tree. */
1976 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1977 is_host_root
= unit_has_host_root_cgroup(u
);
1979 assert_se(c
= unit_get_cgroup_context(u
));
1980 assert_se(path
= u
->cgroup_path
);
1982 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
1985 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
1986 * then), and missing cgroups, i.e. EROFS and ENOENT. */
1988 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
1989 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
1990 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
1991 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
1992 * we couldn't even write to them if we wanted to). */
1993 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
1995 if (cg_all_unified() > 0) {
1998 if (cgroup_context_has_cpu_weight(c
))
1999 weight
= cgroup_context_cpu_weight(c
, state
);
2000 else if (cgroup_context_has_cpu_shares(c
)) {
2003 shares
= cgroup_context_cpu_shares(c
, state
);
2004 weight
= cgroup_cpu_shares_to_weight(shares
);
2006 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
2007 shares
, weight
, path
);
2009 weight
= CGROUP_WEIGHT_DEFAULT
;
2011 cgroup_apply_unified_cpu_idle(u
, weight
);
2012 cgroup_apply_unified_cpu_weight(u
, weight
);
2013 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
2018 if (cgroup_context_has_cpu_weight(c
)) {
2021 weight
= cgroup_context_cpu_weight(c
, state
);
2022 shares
= cgroup_cpu_weight_to_shares(weight
);
2024 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
2025 weight
, shares
, path
);
2026 } else if (cgroup_context_has_cpu_shares(c
))
2027 shares
= cgroup_context_cpu_shares(c
, state
);
2029 shares
= CGROUP_CPU_SHARES_DEFAULT
;
2031 cgroup_apply_legacy_cpu_shares(u
, shares
);
2032 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
2036 if ((apply_mask
& CGROUP_MASK_CPUSET
) && !is_local_root
) {
2037 cgroup_apply_unified_cpuset(u
, cgroup_context_allowed_cpus(c
, state
), "cpuset.cpus");
2038 cgroup_apply_unified_cpuset(u
, cgroup_context_allowed_mems(c
, state
), "cpuset.mems");
2041 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
2042 * controller), and in case of containers we want to leave control of these attributes to the container manager
2043 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
2044 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
2045 bool has_io
, has_blockio
;
2048 has_io
= cgroup_context_has_io_config(c
);
2049 has_blockio
= cgroup_context_has_blockio_config(c
);
2052 weight
= cgroup_context_io_weight(c
, state
);
2053 else if (has_blockio
) {
2054 uint64_t blkio_weight
;
2056 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
2057 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
2059 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
2060 blkio_weight
, weight
);
2062 weight
= CGROUP_WEIGHT_DEFAULT
;
2064 set_io_weight(u
, weight
);
2067 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
2068 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
2070 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
2071 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
2073 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
2074 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
2076 } else if (has_blockio
) {
2077 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
2078 weight
= cgroup_weight_blkio_to_io(w
->weight
);
2080 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
2081 w
->weight
, weight
, w
->path
);
2083 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
2086 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
2087 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
2089 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
2090 limits
[type
] = cgroup_io_limit_defaults
[type
];
2092 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
2093 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
2095 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
2096 b
->rbps
, b
->wbps
, b
->path
);
2098 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
2103 if (apply_mask
& CGROUP_MASK_BLKIO
) {
2104 bool has_io
, has_blockio
;
2106 has_io
= cgroup_context_has_io_config(c
);
2107 has_blockio
= cgroup_context_has_blockio_config(c
);
2109 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
2110 * left to our container manager, too. */
2111 if (!is_local_root
) {
2117 io_weight
= cgroup_context_io_weight(c
, state
);
2118 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
2120 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
2122 } else if (has_blockio
)
2123 weight
= cgroup_context_blkio_weight(c
, state
);
2125 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
2127 set_blkio_weight(u
, weight
);
2130 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
2131 weight
= cgroup_weight_io_to_blkio(w
->weight
);
2133 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
2134 w
->weight
, weight
, w
->path
);
2136 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
2138 else if (has_blockio
)
2139 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
2140 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
2143 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
2144 * roots, as there we want the container manager to handle it */
2145 if (is_host_root
|| !is_local_root
) {
2147 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
2148 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
2149 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
2151 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
2153 else if (has_blockio
)
2154 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
2155 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
2159 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
2160 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
2161 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
2162 * write to this if we wanted to.) */
2163 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
2165 if (cg_all_unified() > 0) {
2166 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
, zswap_max
= CGROUP_LIMIT_MAX
, high
= CGROUP_LIMIT_MAX
;
2168 if (unit_has_unified_memory_config(u
)) {
2169 bool startup
= IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
);
2171 high
= startup
&& c
->startup_memory_high_set
? c
->startup_memory_high
: c
->memory_high
;
2172 max
= startup
&& c
->startup_memory_max_set
? c
->startup_memory_max
: c
->memory_max
;
2173 swap_max
= startup
&& c
->startup_memory_swap_max_set
? c
->startup_memory_swap_max
: c
->memory_swap_max
;
2174 zswap_max
= startup
&& c
->startup_memory_zswap_max_set
? c
->startup_memory_zswap_max
: c
->memory_zswap_max
;
2176 max
= c
->memory_limit
;
2178 if (max
!= CGROUP_LIMIT_MAX
)
2179 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
2182 cgroup_apply_unified_memory_limit(u
, "memory.min", unit_get_ancestor_memory_min(u
));
2183 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
2184 cgroup_apply_unified_memory_limit(u
, "memory.high", high
);
2185 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
2186 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
2187 cgroup_apply_unified_memory_limit(u
, "memory.zswap.max", zswap_max
);
2189 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
2192 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
2195 if (unit_has_unified_memory_config(u
)) {
2196 val
= c
->memory_max
;
2197 if (val
!= CGROUP_LIMIT_MAX
)
2198 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIu64
" as MemoryLimit=", val
);
2200 val
= c
->memory_limit
;
2202 if (val
== CGROUP_LIMIT_MAX
)
2203 strncpy(buf
, "-1\n", sizeof(buf
));
2205 xsprintf(buf
, "%" PRIu64
"\n", val
);
2207 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
2211 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
2212 * containers, where we leave this to the manager */
2213 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
2214 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
))
2215 (void) cgroup_apply_devices(u
);
2217 if (apply_mask
& CGROUP_MASK_PIDS
) {
2220 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
2221 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
2222 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
2223 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
2224 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
2225 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
2226 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
2227 * it also counts. But if the user never set a limit through us (i.e. we are the default of
2228 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
2229 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
2230 * which is desirable so that there's an official way to release control of the sysctl from
2231 * systemd: set the limit to unbounded and reload. */
2233 if (cgroup_tasks_max_isset(&c
->tasks_max
)) {
2234 u
->manager
->sysctl_pid_max_changed
= true;
2235 r
= procfs_tasks_set_limit(cgroup_tasks_max_resolve(&c
->tasks_max
));
2236 } else if (u
->manager
->sysctl_pid_max_changed
)
2237 r
= procfs_tasks_set_limit(TASKS_MAX
);
2241 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
2242 "Failed to write to tasks limit sysctls: %m");
2245 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
2246 * leave it for the container manager. */
2247 if (!is_local_root
) {
2248 if (cgroup_tasks_max_isset(&c
->tasks_max
)) {
2249 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
2251 xsprintf(buf
, "%" PRIu64
"\n", cgroup_tasks_max_resolve(&c
->tasks_max
));
2252 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
2254 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
2258 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
2259 cgroup_apply_firewall(u
);
2261 if (apply_mask
& CGROUP_MASK_BPF_FOREIGN
)
2262 cgroup_apply_bpf_foreign_program(u
);
2264 if (apply_mask
& CGROUP_MASK_BPF_SOCKET_BIND
)
2265 cgroup_apply_socket_bind(u
);
2267 if (apply_mask
& CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
)
2268 cgroup_apply_restrict_network_interfaces(u
);
2270 unit_modify_nft_set(u
, /* add = */ true);
2273 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
2277 c
= unit_get_cgroup_context(u
);
2281 if (c
->ip_accounting
||
2282 !set_isempty(c
->ip_address_allow
) ||
2283 !set_isempty(c
->ip_address_deny
) ||
2284 c
->ip_filters_ingress
||
2285 c
->ip_filters_egress
)
2288 /* If any parent slice has an IP access list defined, it applies too */
2289 for (Unit
*p
= UNIT_GET_SLICE(u
); p
; p
= UNIT_GET_SLICE(p
)) {
2290 c
= unit_get_cgroup_context(p
);
2294 if (!set_isempty(c
->ip_address_allow
) ||
2295 !set_isempty(c
->ip_address_deny
))
2302 static bool unit_get_needs_bpf_foreign_program(Unit
*u
) {
2306 c
= unit_get_cgroup_context(u
);
2310 return !!c
->bpf_foreign_programs
;
2313 static bool unit_get_needs_socket_bind(Unit
*u
) {
2317 c
= unit_get_cgroup_context(u
);
2321 return c
->socket_bind_allow
|| c
->socket_bind_deny
;
2324 static bool unit_get_needs_restrict_network_interfaces(Unit
*u
) {
2328 c
= unit_get_cgroup_context(u
);
2332 return !set_isempty(c
->restrict_network_interfaces
);
2335 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
2336 CGroupMask mask
= 0;
2341 assert_se(c
= unit_get_cgroup_context(u
));
2343 /* Figure out which controllers we need, based on the cgroup context object */
2345 if (c
->cpu_accounting
)
2346 mask
|= get_cpu_accounting_mask();
2348 if (cgroup_context_has_cpu_weight(c
) ||
2349 cgroup_context_has_cpu_shares(c
) ||
2350 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
2351 mask
|= CGROUP_MASK_CPU
;
2353 if (cgroup_context_has_allowed_cpus(c
) || cgroup_context_has_allowed_mems(c
))
2354 mask
|= CGROUP_MASK_CPUSET
;
2356 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
2357 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
2359 if (c
->memory_accounting
||
2360 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
2361 unit_has_unified_memory_config(u
))
2362 mask
|= CGROUP_MASK_MEMORY
;
2364 if (c
->device_allow
||
2365 c
->device_policy
!= CGROUP_DEVICE_POLICY_AUTO
)
2366 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
2368 if (c
->tasks_accounting
||
2369 cgroup_tasks_max_isset(&c
->tasks_max
))
2370 mask
|= CGROUP_MASK_PIDS
;
2372 return CGROUP_MASK_EXTEND_JOINED(mask
);
2375 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
2376 CGroupMask mask
= 0;
2378 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
2381 if (unit_get_needs_bpf_firewall(u
))
2382 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2384 if (unit_get_needs_bpf_foreign_program(u
))
2385 mask
|= CGROUP_MASK_BPF_FOREIGN
;
2387 if (unit_get_needs_socket_bind(u
))
2388 mask
|= CGROUP_MASK_BPF_SOCKET_BIND
;
2390 if (unit_get_needs_restrict_network_interfaces(u
))
2391 mask
|= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
;
2396 CGroupMask
unit_get_own_mask(Unit
*u
) {
2399 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
2400 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
2402 if (u
->load_state
!= UNIT_LOADED
)
2405 c
= unit_get_cgroup_context(u
);
2409 return unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
);
2412 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
2415 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
2416 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
2418 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
2420 if (!unit_cgroup_delegate(u
))
2423 if (cg_all_unified() <= 0) {
2426 e
= unit_get_exec_context(u
);
2427 if (e
&& !exec_context_maintains_privileges(e
))
2431 assert_se(c
= unit_get_cgroup_context(u
));
2432 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
2435 static CGroupMask
unit_get_subtree_mask(Unit
*u
) {
2437 /* Returns the mask of this subtree, meaning of the group
2438 * itself and its children. */
2440 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
2443 CGroupMask
unit_get_members_mask(Unit
*u
) {
2446 /* Returns the mask of controllers all of the unit's children require, merged */
2448 if (u
->cgroup_members_mask_valid
)
2449 return u
->cgroup_members_mask
; /* Use cached value if possible */
2451 u
->cgroup_members_mask
= 0;
2453 if (u
->type
== UNIT_SLICE
) {
2456 UNIT_FOREACH_DEPENDENCY(member
, u
, UNIT_ATOM_SLICE_OF
)
2457 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
2460 u
->cgroup_members_mask_valid
= true;
2461 return u
->cgroup_members_mask
;
2464 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
2468 /* Returns the mask of controllers all of the unit's siblings
2469 * require, i.e. the members mask of the unit's parent slice
2470 * if there is one. */
2472 slice
= UNIT_GET_SLICE(u
);
2474 return unit_get_members_mask(slice
);
2476 return unit_get_subtree_mask(u
); /* we are the top-level slice */
2479 static CGroupMask
unit_get_disable_mask(Unit
*u
) {
2482 c
= unit_get_cgroup_context(u
);
2486 return c
->disable_controllers
;
2489 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
2494 mask
= unit_get_disable_mask(u
);
2496 /* Returns the mask of controllers which are marked as forcibly
2497 * disabled in any ancestor unit or the unit in question. */
2499 slice
= UNIT_GET_SLICE(u
);
2501 mask
|= unit_get_ancestor_disable_mask(slice
);
2506 CGroupMask
unit_get_target_mask(Unit
*u
) {
2507 CGroupMask own_mask
, mask
;
2509 /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything
2510 * it needs itself, plus all that its children need, plus all that its siblings need. This is
2511 * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each
2512 * hierarchy that shall be enabled for it. */
2514 own_mask
= unit_get_own_mask(u
);
2516 if (own_mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
2517 emit_bpf_firewall_warning(u
);
2519 mask
= own_mask
| unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
2521 mask
&= u
->manager
->cgroup_supported
;
2522 mask
&= ~unit_get_ancestor_disable_mask(u
);
2527 CGroupMask
unit_get_enable_mask(Unit
*u
) {
2530 /* This returns the cgroup mask of all controllers to enable
2531 * for the children of a specific cgroup. This is primarily
2532 * useful for the unified cgroup hierarchy, where each cgroup
2533 * controls which controllers are enabled for its children. */
2535 mask
= unit_get_members_mask(u
);
2536 mask
&= u
->manager
->cgroup_supported
;
2537 mask
&= ~unit_get_ancestor_disable_mask(u
);
2542 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
2547 /* Recurse invalidate the member masks cache all the way up the tree */
2548 u
->cgroup_members_mask_valid
= false;
2550 slice
= UNIT_GET_SLICE(u
);
2552 unit_invalidate_cgroup_members_masks(slice
);
2555 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
2557 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
2561 if (u
->cgroup_path
&&
2562 u
->cgroup_realized
&&
2563 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
2564 return u
->cgroup_path
;
2566 u
= UNIT_GET_SLICE(u
);
2572 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
2573 /* If not realized at all, migrate to root ("").
2574 * It may happen if we're upgrading from older version that didn't clean up.
2576 return strempty(unit_get_realized_cgroup_path(userdata
, mask
));
2579 int unit_default_cgroup_path(const Unit
*u
, char **ret
) {
2580 _cleanup_free_
char *p
= NULL
;
2586 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
2587 p
= strdup(u
->manager
->cgroup_root
);
2589 _cleanup_free_
char *escaped
= NULL
, *slice_path
= NULL
;
2592 slice
= UNIT_GET_SLICE(u
);
2593 if (slice
&& !unit_has_name(slice
, SPECIAL_ROOT_SLICE
)) {
2594 r
= cg_slice_to_path(slice
->id
, &slice_path
);
2599 r
= cg_escape(u
->id
, &escaped
);
2603 p
= path_join(empty_to_root(u
->manager
->cgroup_root
), slice_path
, escaped
);
2612 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
2613 _cleanup_free_
char *p
= NULL
;
2618 if (streq_ptr(u
->cgroup_path
, path
))
2628 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
2633 unit_release_cgroup(u
);
2634 u
->cgroup_path
= TAKE_PTR(p
);
2639 int unit_watch_cgroup(Unit
*u
) {
2640 _cleanup_free_
char *events
= NULL
;
2645 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
2646 * cgroupv2 is available. */
2648 if (!u
->cgroup_path
)
2651 if (u
->cgroup_control_inotify_wd
>= 0)
2654 /* Only applies to the unified hierarchy */
2655 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2657 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
2661 /* No point in watch the top-level slice, it's never going to run empty. */
2662 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
2665 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
2669 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
2673 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
2674 if (u
->cgroup_control_inotify_wd
< 0) {
2676 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
2677 * is not an error */
2680 return log_unit_error_errno(u
, errno
, "Failed to add control inotify watch descriptor for control group %s: %m", empty_to_root(u
->cgroup_path
));
2683 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
2685 return log_unit_error_errno(u
, r
, "Failed to add control inotify watch descriptor for control group %s to hash map: %m", empty_to_root(u
->cgroup_path
));
2690 int unit_watch_cgroup_memory(Unit
*u
) {
2691 _cleanup_free_
char *events
= NULL
;
2697 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
2698 * cgroupv2 is available. */
2700 if (!u
->cgroup_path
)
2703 c
= unit_get_cgroup_context(u
);
2707 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
2708 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
2710 if (!c
->memory_accounting
)
2713 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
2714 * we also don't want to generate a log message for each parent cgroup of a process. */
2715 if (u
->type
== UNIT_SLICE
)
2718 if (u
->cgroup_memory_inotify_wd
>= 0)
2721 /* Only applies to the unified hierarchy */
2722 r
= cg_all_unified();
2724 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
2728 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
2732 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
2736 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
2737 if (u
->cgroup_memory_inotify_wd
< 0) {
2739 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
2740 * is not an error */
2743 return log_unit_error_errno(u
, errno
, "Failed to add memory inotify watch descriptor for control group %s: %m", empty_to_root(u
->cgroup_path
));
2746 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
2748 return log_unit_error_errno(u
, r
, "Failed to add memory inotify watch descriptor for control group %s to hash map: %m", empty_to_root(u
->cgroup_path
));
2753 int unit_pick_cgroup_path(Unit
*u
) {
2754 _cleanup_free_
char *path
= NULL
;
2762 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2765 r
= unit_default_cgroup_path(u
, &path
);
2767 return log_unit_error_errno(u
, r
, "Failed to generate default cgroup path: %m");
2769 r
= unit_set_cgroup_path(u
, path
);
2771 return log_unit_error_errno(u
, r
, "Control group %s exists already.", empty_to_root(path
));
2773 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", empty_to_root(path
));
2778 static int unit_update_cgroup(
2780 CGroupMask target_mask
,
2781 CGroupMask enable_mask
,
2782 ManagerState state
) {
2784 bool created
, is_root_slice
;
2785 CGroupMask migrate_mask
= 0;
2786 _cleanup_free_
char *cgroup_full_path
= NULL
;
2791 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2794 /* Figure out our cgroup path */
2795 r
= unit_pick_cgroup_path(u
);
2799 /* First, create our own group */
2800 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
2802 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", empty_to_root(u
->cgroup_path
));
2805 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
2806 uint64_t cgroup_id
= 0;
2808 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, NULL
, &cgroup_full_path
);
2810 r
= cg_path_get_cgroupid(cgroup_full_path
, &cgroup_id
);
2812 log_unit_full_errno(u
, ERRNO_IS_NOT_SUPPORTED(r
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2813 "Failed to get cgroup ID of cgroup %s, ignoring: %m", cgroup_full_path
);
2815 log_unit_warning_errno(u
, r
, "Failed to get full cgroup path on cgroup %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2817 u
->cgroup_id
= cgroup_id
;
2820 /* Start watching it */
2821 (void) unit_watch_cgroup(u
);
2822 (void) unit_watch_cgroup_memory(u
);
2824 /* For v2 we preserve enabled controllers in delegated units, adjust others,
2825 * for v1 we figure out which controller hierarchies need migration. */
2826 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
2827 CGroupMask result_mask
= 0;
2829 /* Enable all controllers we need */
2830 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
2832 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2834 /* Remember what's actually enabled now */
2835 u
->cgroup_enabled_mask
= result_mask
;
2837 migrate_mask
= u
->cgroup_realized_mask
^ target_mask
;
2840 /* Keep track that this is now realized */
2841 u
->cgroup_realized
= true;
2842 u
->cgroup_realized_mask
= target_mask
;
2844 /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling).
2846 * Unnecessary controller cgroups are trimmed (after emptied by upward migration).
2847 * We perform migration also with whole slices for cases when users don't care about leave
2848 * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing
2851 if (cg_all_unified() == 0) {
2852 r
= cg_migrate_v1_controllers(u
->manager
->cgroup_supported
, migrate_mask
, u
->cgroup_path
, migrate_callback
, u
);
2854 log_unit_warning_errno(u
, r
, "Failed to migrate controller cgroups from %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2856 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2857 r
= cg_trim_v1_controllers(u
->manager
->cgroup_supported
, ~target_mask
, u
->cgroup_path
, !is_root_slice
);
2859 log_unit_warning_errno(u
, r
, "Failed to delete controller cgroups %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2862 /* Set attributes */
2863 cgroup_context_apply(u
, target_mask
, state
);
2864 cgroup_xattr_apply(u
);
2866 /* For most units we expect that memory monitoring is set up before the unit is started and we won't
2867 * touch it after. For PID 1 this is different though, because we couldn't possibly do that given
2868 * that PID 1 runs before init.scope is even set up. Hence, whenever init.scope is realized, let's
2869 * try to open the memory pressure interface anew. */
2870 if (unit_has_name(u
, SPECIAL_INIT_SCOPE
))
2871 (void) manager_setup_memory_pressure_event_source(u
->manager
);
2876 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
2877 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2883 if (MANAGER_IS_SYSTEM(u
->manager
))
2886 if (!u
->manager
->system_bus
)
2889 if (!u
->cgroup_path
)
2892 /* Determine this unit's cgroup path relative to our cgroup root */
2893 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
2897 pp
= strjoina("/", pp
, suffix_path
);
2900 r
= bus_call_method(u
->manager
->system_bus
,
2902 "AttachProcessesToUnit",
2905 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
2907 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
2912 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
2913 _cleanup_free_
char *joined
= NULL
;
2914 CGroupMask delegated_mask
;
2921 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2924 if (set_isempty(pids
))
2927 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
2928 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
2929 r
= bpf_firewall_load_custom(u
);
2933 r
= unit_realize_cgroup(u
);
2937 if (isempty(suffix_path
))
2940 joined
= path_join(u
->cgroup_path
, suffix_path
);
2947 delegated_mask
= unit_get_delegate_mask(u
);
2950 SET_FOREACH(pid
, pids
) {
2952 /* Unfortunately we cannot add pids by pidfd to a cgroup. Hence we have to use PIDs instead,
2953 * which of course is racy. Let's shorten the race a bit though, and re-validate the PID
2954 * before we use it */
2955 r
= pidref_verify(pid
);
2957 log_unit_info_errno(u
, r
, "PID " PID_FMT
" vanished before we could move it to target cgroup '%s', skipping: %m", pid
->pid
, empty_to_root(p
));
2961 /* First, attach the PID to the main cgroup hierarchy */
2962 r
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
->pid
);
2964 bool again
= MANAGER_IS_USER(u
->manager
) && ERRNO_IS_PRIVILEGE(r
);
2966 log_unit_full_errno(u
, again
? LOG_DEBUG
: LOG_INFO
, r
,
2967 "Couldn't move process "PID_FMT
" to%s requested cgroup '%s': %m",
2968 pid
->pid
, again
? " directly" : "", empty_to_root(p
));
2973 /* If we are in a user instance, and we can't move the process ourselves due
2974 * to permission problems, let's ask the system instance about it instead.
2975 * Since it's more privileged it might be able to move the process across the
2976 * leaves of a subtree whose top node is not owned by us. */
2978 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
->pid
, suffix_path
);
2980 log_unit_info_errno(u
, z
, "Couldn't move process "PID_FMT
" to requested cgroup '%s' (directly or via the system bus): %m", pid
->pid
, empty_to_root(p
));
2983 ret
++; /* Count successful additions */
2984 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
2989 ret
= r
; /* Remember first error */
2992 } else if (ret
>= 0)
2993 ret
++; /* Count successful additions */
2995 r
= cg_all_unified();
3001 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
3002 * innermost realized one */
3004 for (CGroupController c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
3005 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
3006 const char *realized
;
3008 if (!(u
->manager
->cgroup_supported
& bit
))
3011 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
3012 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
3013 r
= cg_attach(cgroup_controller_to_string(c
), p
, pid
->pid
);
3015 continue; /* Success! */
3017 log_unit_debug_errno(u
, r
, "Failed to attach PID " PID_FMT
" to requested cgroup %s in controller %s, falling back to unit's cgroup: %m",
3018 pid
->pid
, empty_to_root(p
), cgroup_controller_to_string(c
));
3021 /* So this controller is either not delegate or realized, or something else weird happened. In
3022 * that case let's attach the PID at least to the closest cgroup up the tree that is
3024 realized
= unit_get_realized_cgroup_path(u
, bit
);
3026 continue; /* Not even realized in the root slice? Then let's not bother */
3028 r
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
->pid
);
3030 log_unit_debug_errno(u
, r
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
3031 pid
->pid
, realized
, cgroup_controller_to_string(c
));
3038 static bool unit_has_mask_realized(
3040 CGroupMask target_mask
,
3041 CGroupMask enable_mask
) {
3045 /* Returns true if this unit is fully realized. We check four things:
3047 * 1. Whether the cgroup was created at all
3048 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
3049 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
3050 * 4. Whether the invalidation mask is currently zero
3052 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
3053 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
3054 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
3055 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
3056 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
3057 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
3058 * simply don't matter. */
3060 return u
->cgroup_realized
&&
3061 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
3062 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
3063 u
->cgroup_invalidated_mask
== 0;
3066 static bool unit_has_mask_disables_realized(
3068 CGroupMask target_mask
,
3069 CGroupMask enable_mask
) {
3073 /* Returns true if all controllers which should be disabled are indeed disabled.
3075 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
3076 * already removed. */
3078 return !u
->cgroup_realized
||
3079 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
3080 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
3083 static bool unit_has_mask_enables_realized(
3085 CGroupMask target_mask
,
3086 CGroupMask enable_mask
) {
3090 /* Returns true if all controllers which should be enabled are indeed enabled.
3092 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
3093 * we want to add is already added. */
3095 return u
->cgroup_realized
&&
3096 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
3097 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
3100 void unit_add_to_cgroup_realize_queue(Unit
*u
) {
3103 if (u
->in_cgroup_realize_queue
)
3106 LIST_APPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
3107 u
->in_cgroup_realize_queue
= true;
3110 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
3113 if (!u
->in_cgroup_realize_queue
)
3116 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
3117 u
->in_cgroup_realize_queue
= false;
3120 /* Controllers can only be enabled breadth-first, from the root of the
3121 * hierarchy downwards to the unit in question. */
3122 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
3123 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
3129 /* First go deal with this unit's parent, or we won't be able to enable
3130 * any new controllers at this layer. */
3131 slice
= UNIT_GET_SLICE(u
);
3133 r
= unit_realize_cgroup_now_enable(slice
, state
);
3138 target_mask
= unit_get_target_mask(u
);
3139 enable_mask
= unit_get_enable_mask(u
);
3141 /* We can only enable in this direction, don't try to disable anything.
3143 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
3146 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
3147 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
3149 return unit_update_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
3152 /* Controllers can only be disabled depth-first, from the leaves of the
3153 * hierarchy upwards to the unit in question. */
3154 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
3159 if (u
->type
!= UNIT_SLICE
)
3162 UNIT_FOREACH_DEPENDENCY(m
, u
, UNIT_ATOM_SLICE_OF
) {
3163 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
3166 /* The cgroup for this unit might not actually be fully realised yet, in which case it isn't
3167 * holding any controllers open anyway. */
3168 if (!m
->cgroup_realized
)
3171 /* We must disable those below us first in order to release the controller. */
3172 if (m
->type
== UNIT_SLICE
)
3173 (void) unit_realize_cgroup_now_disable(m
, state
);
3175 target_mask
= unit_get_target_mask(m
);
3176 enable_mask
= unit_get_enable_mask(m
);
3178 /* We can only disable in this direction, don't try to enable anything. */
3179 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
3182 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
3183 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
3185 r
= unit_update_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
3193 /* Check if necessary controllers and attributes for a unit are in place.
3195 * - If so, do nothing.
3196 * - If not, create paths, move processes over, and set attributes.
3198 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
3199 * a depth-first way. As such the process looks like this:
3201 * Suppose we have a cgroup hierarchy which looks like this:
3214 * 1. We want to realise cgroup "d" now.
3215 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
3216 * 3. cgroup "k" just started requesting the memory controller.
3218 * To make this work we must do the following in order:
3220 * 1. Disable CPU controller in k, j
3221 * 2. Disable CPU controller in d
3222 * 3. Enable memory controller in root
3223 * 4. Enable memory controller in a
3224 * 5. Enable memory controller in d
3225 * 6. Enable memory controller in k
3227 * Notice that we need to touch j in one direction, but not the other. We also
3228 * don't go beyond d when disabling -- it's up to "a" to get realized if it
3229 * wants to disable further. The basic rules are therefore:
3231 * - If you're disabling something, you need to realise all of the cgroups from
3232 * your recursive descendants to the root. This starts from the leaves.
3233 * - If you're enabling something, you need to realise from the root cgroup
3234 * downwards, but you don't need to iterate your recursive descendants.
3236 * Returns 0 on success and < 0 on failure. */
3237 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
3238 CGroupMask target_mask
, enable_mask
;
3244 unit_remove_from_cgroup_realize_queue(u
);
3246 target_mask
= unit_get_target_mask(u
);
3247 enable_mask
= unit_get_enable_mask(u
);
3249 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
3252 /* Disable controllers below us, if there are any */
3253 r
= unit_realize_cgroup_now_disable(u
, state
);
3257 /* Enable controllers above us, if there are any */
3258 slice
= UNIT_GET_SLICE(u
);
3260 r
= unit_realize_cgroup_now_enable(slice
, state
);
3265 /* Now actually deal with the cgroup we were trying to realise and set attributes */
3266 r
= unit_update_cgroup(u
, target_mask
, enable_mask
, state
);
3270 /* Now, reset the invalidation mask */
3271 u
->cgroup_invalidated_mask
= 0;
3275 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
3283 state
= manager_state(m
);
3285 while ((i
= m
->cgroup_realize_queue
)) {
3286 assert(i
->in_cgroup_realize_queue
);
3288 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
3289 /* Maybe things changed, and the unit is not actually active anymore? */
3290 unit_remove_from_cgroup_realize_queue(i
);
3294 r
= unit_realize_cgroup_now(i
, state
);
3296 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
3304 void unit_add_family_to_cgroup_realize_queue(Unit
*u
) {
3306 assert(u
->type
== UNIT_SLICE
);
3308 /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all
3311 * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes
3312 * very weird if two units that own processes reside in the same slice, but one is realized in the
3313 * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does
3314 * not), because that means individual processes need to be scheduled against whole cgroups. Let's
3315 * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1
3316 * controller hierarchies too (if unit requires the controller to be realized).
3318 * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date
3324 /* Children of u likely changed when we're called */
3325 u
->cgroup_members_mask_valid
= false;
3327 UNIT_FOREACH_DEPENDENCY(m
, u
, UNIT_ATOM_SLICE_OF
) {
3329 /* No point in doing cgroup application for units without active processes. */
3330 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
3333 /* We only enqueue siblings if they were realized once at least, in the main
3335 if (!m
->cgroup_realized
)
3338 /* If the unit doesn't need any new controllers and has current ones
3339 * realized, it doesn't need any changes. */
3340 if (unit_has_mask_realized(m
,
3341 unit_get_target_mask(m
),
3342 unit_get_enable_mask(m
)))
3345 unit_add_to_cgroup_realize_queue(m
);
3348 /* Parent comes after children */
3349 unit_add_to_cgroup_realize_queue(u
);
3351 u
= UNIT_GET_SLICE(u
);
3355 int unit_realize_cgroup(Unit
*u
) {
3360 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3363 /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all
3364 * parents, but there's more actually: for the weight-based controllers we also need to make sure
3365 * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the
3366 * other hand, when a controller is removed from realized set, it may become unnecessary in siblings
3367 * and ancestors and they should be (de)realized too.
3369 * This call will defer work on the siblings and derealized ancestors to the next event loop
3370 * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */
3372 slice
= UNIT_GET_SLICE(u
);
3374 unit_add_family_to_cgroup_realize_queue(slice
);
3376 /* And realize this one now (and apply the values) */
3377 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
3380 void unit_release_cgroup(Unit
*u
) {
3383 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
3384 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
3386 if (u
->cgroup_path
) {
3387 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
3388 u
->cgroup_path
= mfree(u
->cgroup_path
);
3391 if (u
->cgroup_control_inotify_wd
>= 0) {
3392 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
3393 log_unit_debug_errno(u
, errno
, "Failed to remove cgroup control inotify watch %i for %s, ignoring: %m", u
->cgroup_control_inotify_wd
, u
->id
);
3395 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
3396 u
->cgroup_control_inotify_wd
= -1;
3399 if (u
->cgroup_memory_inotify_wd
>= 0) {
3400 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
3401 log_unit_debug_errno(u
, errno
, "Failed to remove cgroup memory inotify watch %i for %s, ignoring: %m", u
->cgroup_memory_inotify_wd
, u
->id
);
3403 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
3404 u
->cgroup_memory_inotify_wd
= -1;
3408 bool unit_maybe_release_cgroup(Unit
*u
) {
3413 if (!u
->cgroup_path
)
3416 /* Don't release the cgroup if there are still processes under it. If we get notified later when all the
3417 * processes exit (e.g. the processes were in D-state and exited after the unit was marked as failed)
3418 * we need the cgroup paths to continue to be tracked by the manager so they can be looked up and cleaned
3420 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
3422 log_unit_debug_errno(u
, r
, "Error checking if the cgroup is recursively empty, ignoring: %m");
3424 unit_release_cgroup(u
);
3431 void unit_prune_cgroup(Unit
*u
) {
3437 /* Removes the cgroup, if empty and possible, and stops watching it. */
3439 if (!u
->cgroup_path
)
3442 /* Cache the last CPU and memory usage values before we destroy the cgroup */
3443 (void) unit_get_cpu_usage(u
, /* ret = */ NULL
);
3445 for (CGroupMemoryAccountingMetric metric
= 0; metric
<= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST
; metric
++)
3446 (void) unit_get_memory_accounting(u
, metric
, /* ret = */ NULL
);
3449 (void) lsm_bpf_cleanup(u
); /* Remove cgroup from the global LSM BPF map */
3452 unit_modify_nft_set(u
, /* add = */ false);
3454 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
3456 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
3458 /* One reason we could have failed here is, that the cgroup still contains a process.
3459 * However, if the cgroup becomes removable at a later time, it might be removed when
3460 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
3461 * that the cgroup is still realized the next time it is started. Do not return early
3462 * on error, continue cleanup. */
3463 log_unit_full_errno(u
, r
== -EBUSY
? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to destroy cgroup %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
3468 if (!unit_maybe_release_cgroup(u
)) /* Returns true if the cgroup was released */
3471 u
->cgroup_realized
= false;
3472 u
->cgroup_realized_mask
= 0;
3473 u
->cgroup_enabled_mask
= 0;
3475 u
->bpf_device_control_installed
= bpf_program_free(u
->bpf_device_control_installed
);
3478 int unit_search_main_pid(Unit
*u
, PidRef
*ret
) {
3479 _cleanup_(pidref_done
) PidRef pidref
= PIDREF_NULL
;
3480 _cleanup_fclose_
FILE *f
= NULL
;
3486 if (!u
->cgroup_path
)
3489 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
3494 _cleanup_(pidref_done
) PidRef npidref
= PIDREF_NULL
;
3496 r
= cg_read_pidref(f
, &npidref
);
3502 if (pidref_equal(&pidref
, &npidref
)) /* seen already, cgroupfs reports duplicates! */
3505 if (pidref_is_my_child(&npidref
) <= 0) /* ignore processes further down the tree */
3508 if (pidref_is_set(&pidref
) != 0)
3509 /* Dang, there's more than one daemonized PID in this group, so we don't know what
3510 * process is the main process. */
3513 pidref
= TAKE_PIDREF(npidref
);
3516 if (!pidref_is_set(&pidref
))
3519 *ret
= TAKE_PIDREF(pidref
);
3523 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
3524 _cleanup_closedir_
DIR *d
= NULL
;
3525 _cleanup_fclose_
FILE *f
= NULL
;
3531 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
3536 _cleanup_(pidref_done
) PidRef pid
= PIDREF_NULL
;
3538 r
= cg_read_pidref(f
, &pid
);
3546 RET_GATHER(ret
, unit_watch_pidref(u
, &pid
, /* exclusive= */ false));
3550 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
3555 _cleanup_free_
char *fn
= NULL
, *p
= NULL
;
3557 r
= cg_read_subgroup(d
, &fn
);
3565 p
= path_join(empty_to_root(path
), fn
);
3569 RET_GATHER(ret
, unit_watch_pids_in_path(u
, p
));
3576 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
3581 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
3582 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
3583 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
3585 if (!u
->cgroup_path
)
3588 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
3591 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
3594 if (!set_isempty(u
->pids
))
3597 unit_add_to_cgroup_empty_queue(u
);
3601 int unit_watch_all_pids(Unit
*u
) {
3606 /* Adds all PIDs from our cgroup to the set of PIDs we
3607 * watch. This is a fallback logic for cases where we do not
3608 * get reliable cgroup empty notifications: we try to use
3609 * SIGCHLD as replacement. */
3611 if (!u
->cgroup_path
)
3614 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
3617 if (r
> 0) /* On unified we can use proper notifications */
3620 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
3623 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
3624 Manager
*m
= ASSERT_PTR(userdata
);
3630 u
= m
->cgroup_empty_queue
;
3634 assert(u
->in_cgroup_empty_queue
);
3635 u
->in_cgroup_empty_queue
= false;
3636 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
3638 if (m
->cgroup_empty_queue
) {
3639 /* More stuff queued, let's make sure we remain enabled */
3640 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
3642 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
3645 /* Update state based on OOM kills before we notify about cgroup empty event */
3646 (void) unit_check_oom(u
);
3647 (void) unit_check_oomd_kill(u
);
3649 unit_add_to_gc_queue(u
);
3651 if (IN_SET(unit_active_state(u
), UNIT_INACTIVE
, UNIT_FAILED
))
3652 unit_prune_cgroup(u
);
3653 else if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
3654 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
3659 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
3664 /* Note that there are four different ways how cgroup empty events reach us:
3666 * 1. On the unified hierarchy we get an inotify event on the cgroup
3668 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
3670 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
3672 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
3673 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
3675 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
3676 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
3677 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
3678 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
3679 * case for scope units). */
3681 if (u
->in_cgroup_empty_queue
)
3684 /* Let's verify that the cgroup is really empty */
3685 if (!u
->cgroup_path
)
3688 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
3690 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", empty_to_root(u
->cgroup_path
));
3696 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
3697 u
->in_cgroup_empty_queue
= true;
3699 /* Trigger the defer event */
3700 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
3702 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
3705 static void unit_remove_from_cgroup_empty_queue(Unit
*u
) {
3708 if (!u
->in_cgroup_empty_queue
)
3711 LIST_REMOVE(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
3712 u
->in_cgroup_empty_queue
= false;
3715 int unit_check_oomd_kill(Unit
*u
) {
3716 _cleanup_free_
char *value
= NULL
;
3721 if (!u
->cgroup_path
)
3724 r
= cg_all_unified();
3726 return log_unit_debug_errno(u
, r
, "Couldn't determine whether we are in all unified mode: %m");
3730 r
= cg_get_xattr_malloc(u
->cgroup_path
, "user.oomd_ooms", &value
);
3731 if (r
< 0 && !ERRNO_IS_XATTR_ABSENT(r
))
3734 if (!isempty(value
)) {
3735 r
= safe_atou64(value
, &n
);
3740 increased
= n
> u
->managed_oom_kill_last
;
3741 u
->managed_oom_kill_last
= n
;
3747 value
= mfree(value
);
3748 r
= cg_get_xattr_malloc(u
->cgroup_path
, "user.oomd_kill", &value
);
3749 if (r
>= 0 && !isempty(value
))
3750 (void) safe_atou64(value
, &n
);
3753 log_unit_struct(u
, LOG_NOTICE
,
3754 "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR
,
3755 LOG_UNIT_INVOCATION_ID(u
),
3756 LOG_UNIT_MESSAGE(u
, "systemd-oomd killed %"PRIu64
" process(es) in this unit.", n
),
3757 "N_PROCESSES=%" PRIu64
, n
);
3759 log_unit_struct(u
, LOG_NOTICE
,
3760 "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR
,
3761 LOG_UNIT_INVOCATION_ID(u
),
3762 LOG_UNIT_MESSAGE(u
, "systemd-oomd killed some process(es) in this unit."));
3764 unit_notify_cgroup_oom(u
, /* ManagedOOM= */ true);
3769 int unit_check_oom(Unit
*u
) {
3770 _cleanup_free_
char *oom_kill
= NULL
;
3775 if (!u
->cgroup_path
)
3778 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
3779 if (IN_SET(r
, -ENOENT
, -ENXIO
)) /* Handle gracefully if cgroup or oom_kill attribute don't exist */
3782 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
3784 r
= safe_atou64(oom_kill
, &c
);
3786 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
3789 increased
= c
> u
->oom_kill_last
;
3790 u
->oom_kill_last
= c
;
3795 log_unit_struct(u
, LOG_NOTICE
,
3796 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
3797 LOG_UNIT_INVOCATION_ID(u
),
3798 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
3800 unit_notify_cgroup_oom(u
, /* ManagedOOM= */ false);
3805 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
3806 Manager
*m
= ASSERT_PTR(userdata
);
3812 u
= m
->cgroup_oom_queue
;
3816 assert(u
->in_cgroup_oom_queue
);
3817 u
->in_cgroup_oom_queue
= false;
3818 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
3820 if (m
->cgroup_oom_queue
) {
3821 /* More stuff queued, let's make sure we remain enabled */
3822 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
3824 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
3827 (void) unit_check_oom(u
);
3828 unit_add_to_gc_queue(u
);
3833 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
3838 if (u
->in_cgroup_oom_queue
)
3840 if (!u
->cgroup_path
)
3843 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
3844 u
->in_cgroup_oom_queue
= true;
3846 /* Trigger the defer event */
3847 if (!u
->manager
->cgroup_oom_event_source
) {
3848 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
3850 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
3852 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
3856 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
3858 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
3862 (void) sd_event_source_set_description(s
, "cgroup-oom");
3863 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
3866 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
3868 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
3871 static int unit_check_cgroup_events(Unit
*u
) {
3872 char *values
[2] = {};
3877 if (!u
->cgroup_path
)
3880 r
= cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events",
3881 STRV_MAKE("populated", "frozen"), values
);
3885 /* The cgroup.events notifications can be merged together so act as we saw the given state for the
3886 * first time. The functions we call to handle given state are idempotent, which makes them
3887 * effectively remember the previous state. */
3889 if (streq(values
[0], "1"))
3890 unit_remove_from_cgroup_empty_queue(u
);
3892 unit_add_to_cgroup_empty_queue(u
);
3895 /* Disregard freezer state changes due to operations not initiated by us */
3896 if (values
[1] && IN_SET(u
->freezer_state
, FREEZER_FREEZING
, FREEZER_THAWING
)) {
3897 if (streq(values
[1], "0"))
3909 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
3910 Manager
*m
= ASSERT_PTR(userdata
);
3916 union inotify_event_buffer buffer
;
3919 l
= read(fd
, &buffer
, sizeof(buffer
));
3921 if (ERRNO_IS_TRANSIENT(errno
))
3924 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
3927 FOREACH_INOTIFY_EVENT_WARN(e
, buffer
, l
) {
3931 /* Queue overflow has no watch descriptor */
3934 if (e
->mask
& IN_IGNORED
)
3935 /* The watch was just removed */
3938 /* Note that inotify might deliver events for a watch even after it was removed,
3939 * because it was queued before the removal. Let's ignore this here safely. */
3941 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
3943 unit_check_cgroup_events(u
);
3945 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
3947 unit_add_to_cgroup_oom_queue(u
);
3952 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
3953 CGroupMask mask
= 0;
3956 /* BPF-based firewall */
3957 r
= bpf_firewall_supported();
3961 mask
|= CGROUP_MASK_BPF_FIREWALL
;
3963 /* BPF-based device access control */
3964 r
= bpf_devices_supported();
3968 mask
|= CGROUP_MASK_BPF_DEVICES
;
3970 /* BPF pinned prog */
3971 r
= bpf_foreign_supported();
3975 mask
|= CGROUP_MASK_BPF_FOREIGN
;
3977 /* BPF-based bind{4|6} hooks */
3978 r
= bpf_socket_bind_supported();
3982 mask
|= CGROUP_MASK_BPF_SOCKET_BIND
;
3984 /* BPF-based cgroup_skb/{egress|ingress} hooks */
3985 r
= restrict_network_interfaces_supported();
3989 mask
|= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
;
3995 int manager_setup_cgroup(Manager
*m
) {
3996 _cleanup_free_
char *path
= NULL
;
3997 const char *scope_path
;
4004 /* 1. Determine hierarchy */
4005 m
->cgroup_root
= mfree(m
->cgroup_root
);
4006 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
4008 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
4010 /* Chop off the init scope, if we are already located in it */
4011 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
4013 /* LEGACY: Also chop off the system slice if we are in
4014 * it. This is to support live upgrades from older systemd
4015 * versions where PID 1 was moved there. Also see
4016 * cg_get_root_path(). */
4017 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
4018 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
4020 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
4025 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
4026 * easily prepend it everywhere. */
4027 delete_trailing_chars(m
->cgroup_root
, "/");
4030 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
4032 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
4036 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
4038 all_unified
= cg_all_unified();
4039 if (all_unified
< 0)
4040 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
4041 if (all_unified
> 0)
4042 log_debug("Unified cgroup hierarchy is located at %s.", path
);
4044 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
4046 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
4048 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
4050 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
4053 /* 3. Allocate cgroup empty defer event source */
4054 m
->cgroup_empty_event_source
= sd_event_source_disable_unref(m
->cgroup_empty_event_source
);
4055 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
4057 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
4059 /* Schedule cgroup empty checks early, but after having processed service notification messages or
4060 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
4061 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
4062 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
4064 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
4066 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
4068 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
4070 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
4072 /* 4. Install notifier inotify object, or agent */
4073 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
4075 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
4077 m
->cgroup_inotify_event_source
= sd_event_source_disable_unref(m
->cgroup_inotify_event_source
);
4078 safe_close(m
->cgroup_inotify_fd
);
4080 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
4081 if (m
->cgroup_inotify_fd
< 0)
4082 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
4084 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
4086 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
4088 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
4089 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
4090 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
4091 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
4093 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
4095 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
4097 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
4099 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
4100 * since it does not generate events when control groups with children run empty. */
4102 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUPS_AGENT_PATH
);
4104 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
4106 log_debug("Installed release agent.");
4108 log_debug("Release agent already installed.");
4111 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
4112 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
4113 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
4115 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
4116 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
4118 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
4120 /* 6. And pin it, so that it cannot be unmounted */
4121 safe_close(m
->pin_cgroupfs_fd
);
4122 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
4123 if (m
->pin_cgroupfs_fd
< 0)
4124 return log_error_errno(errno
, "Failed to open pin file: %m");
4126 } else if (!MANAGER_IS_TEST_RUN(m
))
4127 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
4129 /* 7. Always enable hierarchical support if it exists... */
4130 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
4131 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
4133 /* 8. Figure out which controllers are supported */
4134 r
= cg_mask_supported_subtree(m
->cgroup_root
, &m
->cgroup_supported
);
4136 return log_error_errno(r
, "Failed to determine supported controllers: %m");
4138 /* 9. Figure out which bpf-based pseudo-controllers are supported */
4139 r
= cg_bpf_mask_supported(&mask
);
4141 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
4142 m
->cgroup_supported
|= mask
;
4144 /* 10. Log which controllers are supported */
4145 for (CGroupController c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
4146 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
),
4147 yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
4152 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
4155 /* We can't really delete the group, since we are in it. But
4157 if (delete && m
->cgroup_root
&& !FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
4158 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
4160 m
->cgroup_empty_event_source
= sd_event_source_disable_unref(m
->cgroup_empty_event_source
);
4162 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
4163 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
4165 m
->cgroup_inotify_event_source
= sd_event_source_disable_unref(m
->cgroup_inotify_event_source
);
4166 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
4168 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
4170 m
->cgroup_root
= mfree(m
->cgroup_root
);
4173 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
4180 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
4184 p
= strdupa_safe(cgroup
);
4188 e
= strrchr(p
, '/');
4190 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
4194 u
= hashmap_get(m
->cgroup_unit
, p
);
4200 Unit
*manager_get_unit_by_pidref_cgroup(Manager
*m
, PidRef
*pid
) {
4201 _cleanup_free_
char *cgroup
= NULL
;
4205 if (cg_pidref_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
4208 return manager_get_unit_by_cgroup(m
, cgroup
);
4211 Unit
*manager_get_unit_by_pidref_watching(Manager
*m
, PidRef
*pid
) {
4216 if (!pidref_is_set(pid
))
4219 u
= hashmap_get(m
->watch_pids
, pid
);
4223 array
= hashmap_get(m
->watch_pids_more
, pid
);
4230 Unit
*manager_get_unit_by_pidref(Manager
*m
, PidRef
*pid
) {
4235 /* Note that a process might be owned by multiple units, we return only one here, which is good
4236 * enough for most cases, though not strictly correct. We prefer the one reported by cgroup
4237 * membership, as that's the most relevant one as children of the process will be assigned to that
4238 * one, too, before all else. */
4240 if (!pidref_is_set(pid
))
4243 if (pidref_is_self(pid
))
4244 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
4248 u
= manager_get_unit_by_pidref_cgroup(m
, pid
);
4252 u
= manager_get_unit_by_pidref_watching(m
, pid
);
4259 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
4262 if (!pid_is_valid(pid
))
4265 return manager_get_unit_by_pidref(m
, &PIDREF_MAKE_FROM_PID(pid
));
4268 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
4274 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
4275 * or from the --system instance */
4277 log_debug("Got cgroup empty notification for: %s", cgroup
);
4279 u
= manager_get_unit_by_cgroup(m
, cgroup
);
4283 unit_add_to_cgroup_empty_queue(u
);
4287 int unit_get_memory_available(Unit
*u
, uint64_t *ret
) {
4288 uint64_t available
= UINT64_MAX
, current
= 0;
4293 /* If data from cgroups can be accessed, try to find out how much more memory a unit can
4294 * claim before hitting the configured cgroup limits (if any). Consider both MemoryHigh
4295 * and MemoryMax, and also any slice the unit might be nested below. */
4298 uint64_t unit_available
, unit_limit
= UINT64_MAX
;
4299 CGroupContext
*unit_context
;
4301 /* No point in continuing if we can't go any lower */
4305 unit_context
= unit_get_cgroup_context(u
);
4309 if (!u
->cgroup_path
)
4312 (void) unit_get_memory_current(u
, ¤t
);
4313 /* in case of error, previous current propagates as lower bound */
4315 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
4316 unit_limit
= physical_memory();
4317 else if (unit_context
->memory_max
== UINT64_MAX
&& unit_context
->memory_high
== UINT64_MAX
)
4319 unit_limit
= MIN3(unit_limit
, unit_context
->memory_max
, unit_context
->memory_high
);
4321 unit_available
= LESS_BY(unit_limit
, current
);
4322 available
= MIN(unit_available
, available
);
4323 } while ((u
= UNIT_GET_SLICE(u
)));
4330 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
4333 // FIXME: Merge this into unit_get_memory_accounting after support for cgroup v1 is dropped
4338 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
4341 if (!u
->cgroup_path
)
4344 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
4345 if (unit_has_host_root_cgroup(u
))
4346 return procfs_memory_get_used(ret
);
4348 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
4351 r
= cg_all_unified();
4355 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, r
> 0 ? "memory.current" : "memory.usage_in_bytes", ret
);
4358 int unit_get_memory_accounting(Unit
*u
, CGroupMemoryAccountingMetric metric
, uint64_t *ret
) {
4360 static const char* const attributes_table
[_CGROUP_MEMORY_ACCOUNTING_METRIC_MAX
] = {
4361 [CGROUP_MEMORY_PEAK
] = "memory.peak",
4362 [CGROUP_MEMORY_SWAP_CURRENT
] = "memory.swap.current",
4363 [CGROUP_MEMORY_SWAP_PEAK
] = "memory.swap.peak",
4364 [CGROUP_MEMORY_ZSWAP_CURRENT
] = "memory.zswap.current",
4368 bool updated
= false;
4372 assert(metric
>= 0);
4373 assert(metric
< _CGROUP_MEMORY_ACCOUNTING_METRIC_MAX
);
4375 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
4378 if (!u
->cgroup_path
)
4379 /* If the cgroup is already gone, we try to find the last cached value. */
4382 /* The root cgroup doesn't expose this information. */
4383 if (unit_has_host_root_cgroup(u
))
4386 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_MEMORY
))
4389 r
= cg_all_unified();
4395 r
= cg_get_attribute_as_uint64("memory", u
->cgroup_path
, attributes_table
[metric
], &bytes
);
4396 if (r
< 0 && r
!= -ENODATA
)
4401 if (metric
<= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST
) {
4402 uint64_t *last
= &u
->memory_accounting_last
[metric
];
4406 else if (*last
!= UINT64_MAX
)
4411 } else if (!updated
)
4420 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
4424 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
4427 if (!u
->cgroup_path
)
4430 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
4431 if (unit_has_host_root_cgroup(u
))
4432 return procfs_tasks_get_current(ret
);
4434 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
4437 return cg_get_attribute_as_uint64("pids", u
->cgroup_path
, "pids.current", ret
);
4440 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
4447 if (!u
->cgroup_path
)
4450 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
4451 if (unit_has_host_root_cgroup(u
))
4452 return procfs_cpu_get_usage(ret
);
4454 /* Requisite controllers for CPU accounting are not enabled */
4455 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
4458 r
= cg_all_unified();
4462 _cleanup_free_
char *val
= NULL
;
4465 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
4466 if (IN_SET(r
, -ENOENT
, -ENXIO
))
4471 r
= safe_atou64(val
, &us
);
4475 ns
= us
* NSEC_PER_USEC
;
4477 return cg_get_attribute_as_uint64("cpuacct", u
->cgroup_path
, "cpuacct.usage", ret
);
4483 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
4489 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
4490 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
4491 * call this function with a NULL return value. */
4493 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
4496 r
= unit_get_cpu_usage_raw(u
, &ns
);
4497 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
4498 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
4502 *ret
= u
->cpu_usage_last
;
4508 if (ns
> u
->cpu_usage_base
)
4509 ns
-= u
->cpu_usage_base
;
4513 u
->cpu_usage_last
= ns
;
4520 int unit_get_ip_accounting(
4522 CGroupIPAccountingMetric metric
,
4529 assert(metric
>= 0);
4530 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
4533 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
4536 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
4537 u
->ip_accounting_ingress_map_fd
:
4538 u
->ip_accounting_egress_map_fd
;
4542 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
4543 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
4545 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
4549 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
4550 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
4551 * ip_accounting_extra[] field, and add them in here transparently. */
4553 *ret
= value
+ u
->ip_accounting_extra
[metric
];
4558 static uint64_t unit_get_effective_limit_one(Unit
*u
, CGroupLimitType type
) {
4562 assert(UNIT_HAS_CGROUP_CONTEXT(u
));
4564 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
4566 case CGROUP_LIMIT_MEMORY_MAX
:
4567 case CGROUP_LIMIT_MEMORY_HIGH
:
4568 return physical_memory();
4569 case CGROUP_LIMIT_TASKS_MAX
:
4570 return system_tasks_max();
4572 assert_not_reached();
4575 cc
= unit_get_cgroup_context(u
);
4577 /* Note: on legacy/hybrid hierarchies memory_max stays CGROUP_LIMIT_MAX unless configured
4578 * explicitly. Effective value of MemoryLimit= (cgroup v1) is not implemented. */
4579 case CGROUP_LIMIT_MEMORY_MAX
:
4580 return cc
->memory_max
;
4581 case CGROUP_LIMIT_MEMORY_HIGH
:
4582 return cc
->memory_high
;
4583 case CGROUP_LIMIT_TASKS_MAX
:
4584 return cgroup_tasks_max_resolve(&cc
->tasks_max
);
4586 assert_not_reached();
4590 int unit_get_effective_limit(Unit
*u
, CGroupLimitType type
, uint64_t *ret
) {
4596 assert(type
< _CGROUP_LIMIT_TYPE_MAX
);
4598 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4601 infimum
= unit_get_effective_limit_one(u
, type
);
4602 for (Unit
*slice
= UNIT_GET_SLICE(u
); slice
; slice
= UNIT_GET_SLICE(slice
))
4603 infimum
= MIN(infimum
, unit_get_effective_limit_one(slice
, type
));
4609 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
4610 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
4611 [CGROUP_IO_READ_BYTES
] = "rbytes=",
4612 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
4613 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
4614 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
4616 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
4617 _cleanup_free_
char *path
= NULL
;
4618 _cleanup_fclose_
FILE *f
= NULL
;
4623 if (!u
->cgroup_path
)
4626 if (unit_has_host_root_cgroup(u
))
4627 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
4629 r
= cg_all_unified();
4632 if (r
== 0) /* TODO: support cgroupv1 */
4635 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
4638 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
4642 f
= fopen(path
, "re");
4647 _cleanup_free_
char *line
= NULL
;
4650 r
= read_line(f
, LONG_LINE_MAX
, &line
);
4657 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
4658 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
4661 _cleanup_free_
char *word
= NULL
;
4663 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
4669 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
4672 x
= startswith(word
, field_names
[i
]);
4676 r
= safe_atou64(x
, &w
);
4680 /* Sum up the stats of all devices */
4688 memcpy(ret
, acc
, sizeof(acc
));
4692 int unit_get_io_accounting(
4694 CGroupIOAccountingMetric metric
,
4698 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
4701 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
4703 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
4706 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
4709 r
= unit_get_io_accounting_raw(u
, raw
);
4710 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
4715 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
4716 /* Saturated subtraction */
4717 if (raw
[i
] > u
->io_accounting_base
[i
])
4718 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
4720 u
->io_accounting_last
[i
] = 0;
4725 *ret
= u
->io_accounting_last
[metric
];
4730 int unit_reset_cpu_accounting(Unit
*u
) {
4735 u
->cpu_usage_last
= NSEC_INFINITY
;
4737 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
4739 u
->cpu_usage_base
= 0;
4746 void unit_reset_memory_accounting_last(Unit
*u
) {
4749 FOREACH_ARRAY(i
, u
->memory_accounting_last
, ELEMENTSOF(u
->memory_accounting_last
))
4753 int unit_reset_ip_accounting(Unit
*u
) {
4758 if (u
->ip_accounting_ingress_map_fd
>= 0)
4759 RET_GATHER(r
, bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
));
4761 if (u
->ip_accounting_egress_map_fd
>= 0)
4762 RET_GATHER(r
, bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
));
4764 zero(u
->ip_accounting_extra
);
4769 void unit_reset_io_accounting_last(Unit
*u
) {
4772 FOREACH_ARRAY(i
, u
->io_accounting_last
, _CGROUP_IO_ACCOUNTING_METRIC_MAX
)
4776 int unit_reset_io_accounting(Unit
*u
) {
4781 unit_reset_io_accounting_last(u
);
4783 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
4785 zero(u
->io_accounting_base
);
4792 int unit_reset_accounting(Unit
*u
) {
4797 RET_GATHER(r
, unit_reset_cpu_accounting(u
));
4798 RET_GATHER(r
, unit_reset_io_accounting(u
));
4799 RET_GATHER(r
, unit_reset_ip_accounting(u
));
4800 unit_reset_memory_accounting_last(u
);
4805 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
4808 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4814 /* always invalidate compat pairs together */
4815 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
4816 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
4818 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
4819 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
4821 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
4824 u
->cgroup_invalidated_mask
|= m
;
4825 unit_add_to_cgroup_realize_queue(u
);
4828 void unit_invalidate_cgroup_bpf(Unit
*u
) {
4831 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4834 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
4837 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
4838 unit_add_to_cgroup_realize_queue(u
);
4840 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
4841 * list of our children includes our own. */
4842 if (u
->type
== UNIT_SLICE
) {
4845 UNIT_FOREACH_DEPENDENCY(member
, u
, UNIT_ATOM_SLICE_OF
)
4846 unit_invalidate_cgroup_bpf(member
);
4850 void unit_cgroup_catchup(Unit
*u
) {
4853 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4856 /* We dropped the inotify watch during reexec/reload, so we need to
4857 * check these as they may have changed.
4858 * Note that (currently) the kernel doesn't actually update cgroup
4859 * file modification times, so we can't just serialize and then check
4860 * the mtime for file(s) we are interested in. */
4861 (void) unit_check_cgroup_events(u
);
4862 unit_add_to_cgroup_oom_queue(u
);
4865 bool unit_cgroup_delegate(Unit
*u
) {
4870 if (!UNIT_VTABLE(u
)->can_delegate
)
4873 c
= unit_get_cgroup_context(u
);
4880 void manager_invalidate_startup_units(Manager
*m
) {
4885 SET_FOREACH(u
, m
->startup_units
)
4886 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
|CGROUP_MASK_CPUSET
);
4889 int unit_cgroup_freezer_action(Unit
*u
, FreezerAction action
) {
4890 _cleanup_free_
char *path
= NULL
;
4891 FreezerState target
, kernel
= _FREEZER_STATE_INVALID
;
4895 assert(IN_SET(action
, FREEZER_FREEZE
, FREEZER_THAW
));
4897 if (!cg_freezer_supported())
4900 /* Ignore all requests to thaw init.scope or -.slice and reject all requests to freeze them */
4901 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
) || unit_has_name(u
, SPECIAL_INIT_SCOPE
))
4902 return action
== FREEZER_FREEZE
? -EPERM
: 0;
4904 if (!u
->cgroup_realized
)
4907 if (action
== FREEZER_THAW
) {
4908 Unit
*slice
= UNIT_GET_SLICE(u
);
4911 r
= unit_cgroup_freezer_action(slice
, FREEZER_THAW
);
4913 return log_unit_error_errno(u
, r
, "Failed to thaw slice %s of unit: %m", slice
->id
);
4917 target
= action
== FREEZER_FREEZE
? FREEZER_FROZEN
: FREEZER_RUNNING
;
4919 r
= unit_freezer_state_kernel(u
, &kernel
);
4921 log_unit_debug_errno(u
, r
, "Failed to obtain cgroup freezer state: %m");
4923 if (target
== kernel
) {
4924 u
->freezer_state
= target
;
4925 if (action
== FREEZER_FREEZE
)
4931 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.freeze", &path
);
4935 log_unit_debug(u
, "%s unit.", action
== FREEZER_FREEZE
? "Freezing" : "Thawing");
4937 if (target
!= kernel
) {
4938 if (action
== FREEZER_FREEZE
)
4939 u
->freezer_state
= FREEZER_FREEZING
;
4941 u
->freezer_state
= FREEZER_THAWING
;
4944 r
= write_string_file(path
, one_zero(action
== FREEZER_FREEZE
), WRITE_STRING_FILE_DISABLE_BUFFER
);
4951 int unit_get_cpuset(Unit
*u
, CPUSet
*cpus
, const char *name
) {
4952 _cleanup_free_
char *v
= NULL
;
4958 if (!u
->cgroup_path
)
4961 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUSET
) == 0)
4964 r
= cg_all_unified();
4970 r
= cg_get_attribute("cpuset", u
->cgroup_path
, name
, &v
);
4976 return parse_cpu_set_full(v
, cpus
, false, NULL
, NULL
, 0, NULL
);
4979 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
4980 [CGROUP_DEVICE_POLICY_AUTO
] = "auto",
4981 [CGROUP_DEVICE_POLICY_CLOSED
] = "closed",
4982 [CGROUP_DEVICE_POLICY_STRICT
] = "strict",
4985 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);
4987 static const char* const freezer_action_table
[_FREEZER_ACTION_MAX
] = {
4988 [FREEZER_FREEZE
] = "freeze",
4989 [FREEZER_THAW
] = "thaw",
4992 DEFINE_STRING_TABLE_LOOKUP(freezer_action
, FreezerAction
);
4994 static const char* const cgroup_pressure_watch_table
[_CGROUP_PRESSURE_WATCH_MAX
] = {
4995 [CGROUP_PRESSURE_WATCH_OFF
] = "off",
4996 [CGROUP_PRESSURE_WATCH_AUTO
] = "auto",
4997 [CGROUP_PRESSURE_WATCH_ON
] = "on",
4998 [CGROUP_PRESSURE_WATCH_SKIP
] = "skip",
5001 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(cgroup_pressure_watch
, CGroupPressureWatch
, CGROUP_PRESSURE_WATCH_ON
);
5003 static const char* const cgroup_ip_accounting_metric_table
[_CGROUP_IP_ACCOUNTING_METRIC_MAX
] = {
5004 [CGROUP_IP_INGRESS_BYTES
] = "IPIngressBytes",
5005 [CGROUP_IP_EGRESS_BYTES
] = "IPEgressBytes",
5006 [CGROUP_IP_INGRESS_PACKETS
] = "IPIngressPackets",
5007 [CGROUP_IP_EGRESS_PACKETS
] = "IPEgressPackets",
5010 DEFINE_STRING_TABLE_LOOKUP(cgroup_ip_accounting_metric
, CGroupIPAccountingMetric
);
5012 static const char* const cgroup_io_accounting_metric_table
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
5013 [CGROUP_IO_READ_BYTES
] = "IOReadBytes",
5014 [CGROUP_IO_WRITE_BYTES
] = "IOWriteBytes",
5015 [CGROUP_IO_READ_OPERATIONS
] = "IOReadOperations",
5016 [CGROUP_IO_WRITE_OPERATIONS
] = "IOWriteOperations",
5019 DEFINE_STRING_TABLE_LOOKUP(cgroup_io_accounting_metric
, CGroupIOAccountingMetric
);
5021 static const char* const cgroup_memory_accounting_metric_table
[_CGROUP_MEMORY_ACCOUNTING_METRIC_MAX
] = {
5022 [CGROUP_MEMORY_PEAK
] = "MemoryPeak",
5023 [CGROUP_MEMORY_SWAP_CURRENT
] = "MemorySwapCurrent",
5024 [CGROUP_MEMORY_SWAP_PEAK
] = "MemorySwapPeak",
5025 [CGROUP_MEMORY_ZSWAP_CURRENT
] = "MemoryZSwapCurrent",
5028 DEFINE_STRING_TABLE_LOOKUP(cgroup_memory_accounting_metric
, CGroupMemoryAccountingMetric
);
5030 static const char *const cgroup_limit_type_table
[_CGROUP_LIMIT_TYPE_MAX
] = {
5031 [CGROUP_LIMIT_MEMORY_MAX
] = "EffectiveMemoryMax",
5032 [CGROUP_LIMIT_MEMORY_HIGH
] = "EffectiveMemoryHigh",
5033 [CGROUP_LIMIT_TASKS_MAX
] = "EffectiveTasksMax",
5036 DEFINE_STRING_TABLE_LOOKUP(cgroup_limit_type
, CGroupLimitType
);