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-restrict-ifaces.h"
14 #include "bpf-socket-bind.h"
15 #include "btrfs-util.h"
16 #include "bus-error.h"
17 #include "bus-locator.h"
18 #include "cgroup-setup.h"
19 #include "cgroup-util.h"
21 #include "devnum-util.h"
24 #include "firewall-util.h"
25 #include "in-addr-prefix-util.h"
26 #include "inotify-util.h"
28 #include "ip-protocol-list.h"
29 #include "limits-util.h"
30 #include "nulstr-util.h"
31 #include "parse-util.h"
32 #include "path-util.h"
33 #include "percent-util.h"
34 #include "process-util.h"
35 #include "procfs-util.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
, const CGroupIODeviceWeight
*w
) {
194 _cleanup_free_ CGroupIODeviceWeight
*n
= NULL
;
199 n
= new(CGroupIODeviceWeight
, 1);
203 *n
= (CGroupIODeviceWeight
) {
204 .path
= strdup(w
->path
),
210 LIST_PREPEND(device_weights
, c
->io_device_weights
, TAKE_PTR(n
));
214 int cgroup_context_add_io_device_limit_dup(CGroupContext
*c
, const CGroupIODeviceLimit
*l
) {
215 _cleanup_free_ CGroupIODeviceLimit
*n
= NULL
;
220 n
= new0(CGroupIODeviceLimit
, 1);
224 n
->path
= strdup(l
->path
);
228 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
229 n
->limits
[type
] = l
->limits
[type
];
231 LIST_PREPEND(device_limits
, c
->io_device_limits
, TAKE_PTR(n
));
235 int cgroup_context_add_io_device_latency_dup(CGroupContext
*c
, const CGroupIODeviceLatency
*l
) {
236 _cleanup_free_ CGroupIODeviceLatency
*n
= NULL
;
241 n
= new(CGroupIODeviceLatency
, 1);
245 *n
= (CGroupIODeviceLatency
) {
246 .path
= strdup(l
->path
),
247 .target_usec
= l
->target_usec
,
252 LIST_PREPEND(device_latencies
, c
->io_device_latencies
, TAKE_PTR(n
));
256 int cgroup_context_add_block_io_device_weight_dup(CGroupContext
*c
, const CGroupBlockIODeviceWeight
*w
) {
257 _cleanup_free_ CGroupBlockIODeviceWeight
*n
= NULL
;
262 n
= new(CGroupBlockIODeviceWeight
, 1);
266 *n
= (CGroupBlockIODeviceWeight
) {
267 .path
= strdup(w
->path
),
273 LIST_PREPEND(device_weights
, c
->blockio_device_weights
, TAKE_PTR(n
));
277 int cgroup_context_add_block_io_device_bandwidth_dup(CGroupContext
*c
, const CGroupBlockIODeviceBandwidth
*b
) {
278 _cleanup_free_ CGroupBlockIODeviceBandwidth
*n
= NULL
;
283 n
= new(CGroupBlockIODeviceBandwidth
, 1);
287 *n
= (CGroupBlockIODeviceBandwidth
) {
292 LIST_PREPEND(device_bandwidths
, c
->blockio_device_bandwidths
, TAKE_PTR(n
));
296 int cgroup_context_add_device_allow_dup(CGroupContext
*c
, const CGroupDeviceAllow
*a
) {
297 _cleanup_free_ CGroupDeviceAllow
*n
= NULL
;
302 n
= new(CGroupDeviceAllow
, 1);
306 *n
= (CGroupDeviceAllow
) {
307 .path
= strdup(a
->path
),
308 .permissions
= a
->permissions
,
313 LIST_PREPEND(device_allow
, c
->device_allow
, TAKE_PTR(n
));
317 static int cgroup_context_add_socket_bind_item_dup(CGroupContext
*c
, const CGroupSocketBindItem
*i
, CGroupSocketBindItem
*h
) {
318 _cleanup_free_ CGroupSocketBindItem
*n
= NULL
;
323 n
= new(CGroupSocketBindItem
, 1);
327 *n
= (CGroupSocketBindItem
) {
328 .address_family
= i
->address_family
,
329 .ip_protocol
= i
->ip_protocol
,
330 .nr_ports
= i
->nr_ports
,
331 .port_min
= i
->port_min
,
334 LIST_PREPEND(socket_bind_items
, h
, TAKE_PTR(n
));
338 int cgroup_context_add_socket_bind_item_allow_dup(CGroupContext
*c
, const CGroupSocketBindItem
*i
) {
339 return cgroup_context_add_socket_bind_item_dup(c
, i
, c
->socket_bind_allow
);
342 int cgroup_context_add_socket_bind_item_deny_dup(CGroupContext
*c
, const CGroupSocketBindItem
*i
) {
343 return cgroup_context_add_socket_bind_item_dup(c
, i
, c
->socket_bind_deny
);
346 int cgroup_context_copy(CGroupContext
*dst
, const CGroupContext
*src
) {
347 struct in_addr_prefix
*i
;
354 dst
->cpu_accounting
= src
->cpu_accounting
;
355 dst
->io_accounting
= src
->io_accounting
;
356 dst
->blockio_accounting
= src
->blockio_accounting
;
357 dst
->memory_accounting
= src
->memory_accounting
;
358 dst
->tasks_accounting
= src
->tasks_accounting
;
359 dst
->ip_accounting
= src
->ip_accounting
;
361 dst
->memory_oom_group
= src
->memory_oom_group
;
363 dst
->cpu_weight
= src
->cpu_weight
;
364 dst
->startup_cpu_weight
= src
->startup_cpu_weight
;
365 dst
->cpu_quota_per_sec_usec
= src
->cpu_quota_per_sec_usec
;
366 dst
->cpu_quota_period_usec
= src
->cpu_quota_period_usec
;
368 dst
->cpuset_cpus
= src
->cpuset_cpus
;
369 dst
->startup_cpuset_cpus
= src
->startup_cpuset_cpus
;
370 dst
->cpuset_mems
= src
->cpuset_mems
;
371 dst
->startup_cpuset_mems
= src
->startup_cpuset_mems
;
373 dst
->io_weight
= src
->io_weight
;
374 dst
->startup_io_weight
= src
->startup_io_weight
;
376 LIST_FOREACH_BACKWARDS(device_weights
, w
, LIST_FIND_TAIL(device_weights
, src
->io_device_weights
)) {
377 r
= cgroup_context_add_io_device_weight_dup(dst
, w
);
382 LIST_FOREACH_BACKWARDS(device_limits
, l
, LIST_FIND_TAIL(device_limits
, src
->io_device_limits
)) {
383 r
= cgroup_context_add_io_device_limit_dup(dst
, l
);
388 LIST_FOREACH_BACKWARDS(device_latencies
, l
, LIST_FIND_TAIL(device_latencies
, src
->io_device_latencies
)) {
389 r
= cgroup_context_add_io_device_latency_dup(dst
, l
);
394 dst
->default_memory_min
= src
->default_memory_min
;
395 dst
->default_memory_low
= src
->default_memory_low
;
396 dst
->default_startup_memory_low
= src
->default_startup_memory_low
;
397 dst
->memory_min
= src
->memory_min
;
398 dst
->memory_low
= src
->memory_low
;
399 dst
->startup_memory_low
= src
->startup_memory_low
;
400 dst
->memory_high
= src
->memory_high
;
401 dst
->startup_memory_high
= src
->startup_memory_high
;
402 dst
->memory_max
= src
->memory_max
;
403 dst
->startup_memory_max
= src
->startup_memory_max
;
404 dst
->memory_swap_max
= src
->memory_swap_max
;
405 dst
->startup_memory_swap_max
= src
->startup_memory_swap_max
;
406 dst
->memory_zswap_max
= src
->memory_zswap_max
;
407 dst
->startup_memory_zswap_max
= src
->startup_memory_zswap_max
;
409 dst
->default_memory_min_set
= src
->default_memory_min_set
;
410 dst
->default_memory_low_set
= src
->default_memory_low_set
;
411 dst
->default_startup_memory_low_set
= src
->default_startup_memory_low_set
;
412 dst
->memory_min_set
= src
->memory_min_set
;
413 dst
->memory_low_set
= src
->memory_low_set
;
414 dst
->startup_memory_low_set
= src
->startup_memory_low_set
;
415 dst
->startup_memory_high_set
= src
->startup_memory_high_set
;
416 dst
->startup_memory_max_set
= src
->startup_memory_max_set
;
417 dst
->startup_memory_swap_max_set
= src
->startup_memory_swap_max_set
;
418 dst
->startup_memory_zswap_max_set
= src
->startup_memory_zswap_max_set
;
420 SET_FOREACH(i
, src
->ip_address_allow
) {
421 r
= in_addr_prefix_add(&dst
->ip_address_allow
, i
);
426 SET_FOREACH(i
, src
->ip_address_deny
) {
427 r
= in_addr_prefix_add(&dst
->ip_address_deny
, i
);
432 dst
->ip_address_allow_reduced
= src
->ip_address_allow_reduced
;
433 dst
->ip_address_deny_reduced
= src
->ip_address_deny_reduced
;
435 if (!strv_isempty(src
->ip_filters_ingress
)) {
436 dst
->ip_filters_ingress
= strv_copy(src
->ip_filters_ingress
);
437 if (!dst
->ip_filters_ingress
)
441 if (!strv_isempty(src
->ip_filters_egress
)) {
442 dst
->ip_filters_egress
= strv_copy(src
->ip_filters_egress
);
443 if (!dst
->ip_filters_egress
)
447 LIST_FOREACH_BACKWARDS(programs
, l
, LIST_FIND_TAIL(programs
, src
->bpf_foreign_programs
)) {
448 r
= cgroup_context_add_bpf_foreign_program_dup(dst
, l
);
453 SET_FOREACH(iface
, src
->restrict_network_interfaces
) {
454 r
= set_put_strdup(&dst
->restrict_network_interfaces
, iface
);
458 dst
->restrict_network_interfaces_is_allow_list
= src
->restrict_network_interfaces_is_allow_list
;
460 dst
->cpu_shares
= src
->cpu_shares
;
461 dst
->startup_cpu_shares
= src
->startup_cpu_shares
;
463 dst
->blockio_weight
= src
->blockio_weight
;
464 dst
->startup_blockio_weight
= src
->startup_blockio_weight
;
466 LIST_FOREACH_BACKWARDS(device_weights
, l
, LIST_FIND_TAIL(device_weights
, src
->blockio_device_weights
)) {
467 r
= cgroup_context_add_block_io_device_weight_dup(dst
, l
);
472 LIST_FOREACH_BACKWARDS(device_bandwidths
, l
, LIST_FIND_TAIL(device_bandwidths
, src
->blockio_device_bandwidths
)) {
473 r
= cgroup_context_add_block_io_device_bandwidth_dup(dst
, l
);
478 dst
->memory_limit
= src
->memory_limit
;
480 dst
->device_policy
= src
->device_policy
;
481 LIST_FOREACH_BACKWARDS(device_allow
, l
, LIST_FIND_TAIL(device_allow
, src
->device_allow
)) {
482 r
= cgroup_context_add_device_allow_dup(dst
, l
);
487 LIST_FOREACH_BACKWARDS(socket_bind_items
, l
, LIST_FIND_TAIL(socket_bind_items
, src
->socket_bind_allow
)) {
488 r
= cgroup_context_add_socket_bind_item_allow_dup(dst
, l
);
494 LIST_FOREACH_BACKWARDS(socket_bind_items
, l
, LIST_FIND_TAIL(socket_bind_items
, src
->socket_bind_deny
)) {
495 r
= cgroup_context_add_socket_bind_item_deny_dup(dst
, l
);
500 dst
->tasks_max
= src
->tasks_max
;
505 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
509 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
514 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
518 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
523 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
527 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
532 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
536 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
541 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
545 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
550 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
554 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
559 void cgroup_context_remove_bpf_foreign_program(CGroupContext
*c
, CGroupBPFForeignProgram
*p
) {
563 LIST_REMOVE(programs
, c
->bpf_foreign_programs
, p
);
568 void cgroup_context_remove_socket_bind(CGroupSocketBindItem
**head
) {
571 LIST_CLEAR(socket_bind_items
, *head
, free
);
574 void cgroup_context_done(CGroupContext
*c
) {
577 while (c
->io_device_weights
)
578 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
580 while (c
->io_device_latencies
)
581 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
583 while (c
->io_device_limits
)
584 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
586 while (c
->blockio_device_weights
)
587 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
589 while (c
->blockio_device_bandwidths
)
590 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
592 while (c
->device_allow
)
593 cgroup_context_free_device_allow(c
, c
->device_allow
);
595 cgroup_context_remove_socket_bind(&c
->socket_bind_allow
);
596 cgroup_context_remove_socket_bind(&c
->socket_bind_deny
);
598 c
->ip_address_allow
= set_free(c
->ip_address_allow
);
599 c
->ip_address_deny
= set_free(c
->ip_address_deny
);
601 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
602 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
604 while (c
->bpf_foreign_programs
)
605 cgroup_context_remove_bpf_foreign_program(c
, c
->bpf_foreign_programs
);
607 c
->restrict_network_interfaces
= set_free_free(c
->restrict_network_interfaces
);
609 cpu_set_reset(&c
->cpuset_cpus
);
610 cpu_set_reset(&c
->startup_cpuset_cpus
);
611 cpu_set_reset(&c
->cpuset_mems
);
612 cpu_set_reset(&c
->startup_cpuset_mems
);
614 c
->delegate_subgroup
= mfree(c
->delegate_subgroup
);
616 nft_set_context_clear(&c
->nft_set_context
);
619 static int unit_get_kernel_memory_limit(Unit
*u
, const char *file
, uint64_t *ret
) {
622 if (!u
->cgroup_realized
)
625 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, file
, ret
);
628 static int unit_compare_memory_limit(Unit
*u
, const char *property_name
, uint64_t *ret_unit_value
, uint64_t *ret_kernel_value
) {
635 /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will
636 * return -ENODATA) on cgroup v1.
641 * 0: If the kernel memory setting doesn't match our configuration.
642 * >0: If the kernel memory setting matches our configuration.
644 * The following values are only guaranteed to be populated on return >=0:
646 * - ret_unit_value will contain our internal expected value for the unit, page-aligned.
647 * - ret_kernel_value will contain the actual value presented by the kernel. */
651 r
= cg_all_unified();
653 return log_debug_errno(r
, "Failed to determine cgroup hierarchy version: %m");
655 /* Unsupported on v1.
657 * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has
658 * silently masked the controller. */
662 /* The root slice doesn't have any controller files, so we can't compare anything. */
663 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
666 /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled,
667 * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To
668 * avoid specious errors in these scenarios, check that we even expect the memory controller to be
670 m
= unit_get_target_mask(u
);
671 if (!FLAGS_SET(m
, CGROUP_MASK_MEMORY
))
674 assert_se(c
= unit_get_cgroup_context(u
));
676 bool startup
= u
->manager
&& IN_SET(manager_state(u
->manager
), MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
);
678 if (streq(property_name
, "MemoryLow")) {
679 unit_value
= unit_get_ancestor_memory_low(u
);
681 } else if (startup
&& streq(property_name
, "StartupMemoryLow")) {
682 unit_value
= unit_get_ancestor_startup_memory_low(u
);
684 } else if (streq(property_name
, "MemoryMin")) {
685 unit_value
= unit_get_ancestor_memory_min(u
);
687 } else if (streq(property_name
, "MemoryHigh")) {
688 unit_value
= c
->memory_high
;
689 file
= "memory.high";
690 } else if (startup
&& streq(property_name
, "StartupMemoryHigh")) {
691 unit_value
= c
->startup_memory_high
;
692 file
= "memory.high";
693 } else if (streq(property_name
, "MemoryMax")) {
694 unit_value
= c
->memory_max
;
696 } else if (startup
&& streq(property_name
, "StartupMemoryMax")) {
697 unit_value
= c
->startup_memory_max
;
699 } else if (streq(property_name
, "MemorySwapMax")) {
700 unit_value
= c
->memory_swap_max
;
701 file
= "memory.swap.max";
702 } else if (startup
&& streq(property_name
, "StartupMemorySwapMax")) {
703 unit_value
= c
->startup_memory_swap_max
;
704 file
= "memory.swap.max";
705 } else if (streq(property_name
, "MemoryZSwapMax")) {
706 unit_value
= c
->memory_zswap_max
;
707 file
= "memory.zswap.max";
708 } else if (startup
&& streq(property_name
, "StartupMemoryZSwapMax")) {
709 unit_value
= c
->startup_memory_zswap_max
;
710 file
= "memory.zswap.max";
714 r
= unit_get_kernel_memory_limit(u
, file
, ret_kernel_value
);
716 return log_unit_debug_errno(u
, r
, "Failed to parse %s: %m", file
);
718 /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page
719 * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from
720 * our internal page-counting. To support those future kernels, just check the value itself first
721 * without any page-alignment. */
722 if (*ret_kernel_value
== unit_value
) {
723 *ret_unit_value
= unit_value
;
727 /* The current kernel behaviour, by comparison, is that even if you write a particular number of
728 * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel
729 * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is
731 if (unit_value
!= CGROUP_LIMIT_MAX
)
732 unit_value
= PAGE_ALIGN_DOWN(unit_value
);
734 *ret_unit_value
= unit_value
;
736 return *ret_kernel_value
== *ret_unit_value
;
739 #define FORMAT_CGROUP_DIFF_MAX 128
741 static char *format_cgroup_memory_limit_comparison(char *buf
, size_t l
, Unit
*u
, const char *property_name
) {
749 r
= unit_compare_memory_limit(u
, property_name
, &sval
, &kval
);
751 /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1).
752 * In the absence of reliably being able to detect whether memcg swap support is available or not,
753 * only complain if the error is not ENOENT. This is similarly the case for memory.zswap.max relying
754 * on CONFIG_ZSWAP. */
755 if (r
> 0 || IN_SET(r
, -ENODATA
, -EOWNERDEAD
) ||
756 (r
== -ENOENT
&& STR_IN_SET(property_name
,
758 "StartupMemorySwapMax",
760 "StartupMemoryZSwapMax")))
764 (void) snprintf(buf
, l
, " (error getting kernel value: %m)");
766 (void) snprintf(buf
, l
, " (different value in kernel: %" PRIu64
")", kval
);
771 const char *cgroup_device_permissions_to_string(CGroupDevicePermissions p
) {
772 static const char *table
[_CGROUP_DEVICE_PERMISSIONS_MAX
] = {
773 /* Lets simply define a table with every possible combination. As long as those are just 8 we
774 * can get away with it. If this ever grows to more we need to revisit this logic though. */
776 [CGROUP_DEVICE_READ
] = "r",
777 [CGROUP_DEVICE_WRITE
] = "w",
778 [CGROUP_DEVICE_MKNOD
] = "m",
779 [CGROUP_DEVICE_READ
|CGROUP_DEVICE_WRITE
] = "rw",
780 [CGROUP_DEVICE_READ
|CGROUP_DEVICE_MKNOD
] = "rm",
781 [CGROUP_DEVICE_WRITE
|CGROUP_DEVICE_MKNOD
] = "wm",
782 [CGROUP_DEVICE_READ
|CGROUP_DEVICE_WRITE
|CGROUP_DEVICE_MKNOD
] = "rwm",
785 if (p
< 0 || p
>= _CGROUP_DEVICE_PERMISSIONS_MAX
)
791 CGroupDevicePermissions
cgroup_device_permissions_from_string(const char *s
) {
792 CGroupDevicePermissions p
= 0;
795 return _CGROUP_DEVICE_PERMISSIONS_INVALID
;
797 for (const char *c
= s
; *c
; c
++) {
799 p
|= CGROUP_DEVICE_READ
;
801 p
|= CGROUP_DEVICE_WRITE
;
803 p
|= CGROUP_DEVICE_MKNOD
;
805 return _CGROUP_DEVICE_PERMISSIONS_INVALID
;
811 void cgroup_context_dump(Unit
*u
, FILE* f
, const char *prefix
) {
812 _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
;
814 struct in_addr_prefix
*iaai
;
816 char cda
[FORMAT_CGROUP_DIFF_MAX
];
817 char cdb
[FORMAT_CGROUP_DIFF_MAX
];
818 char cdc
[FORMAT_CGROUP_DIFF_MAX
];
819 char cdd
[FORMAT_CGROUP_DIFF_MAX
];
820 char cde
[FORMAT_CGROUP_DIFF_MAX
];
821 char cdf
[FORMAT_CGROUP_DIFF_MAX
];
822 char cdg
[FORMAT_CGROUP_DIFF_MAX
];
823 char cdh
[FORMAT_CGROUP_DIFF_MAX
];
824 char cdi
[FORMAT_CGROUP_DIFF_MAX
];
825 char cdj
[FORMAT_CGROUP_DIFF_MAX
];
826 char cdk
[FORMAT_CGROUP_DIFF_MAX
];
831 assert_se(c
= unit_get_cgroup_context(u
));
833 prefix
= strempty(prefix
);
835 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
836 (void) cg_mask_to_string(c
->delegate_controllers
, &delegate_controllers_str
);
838 /* "Delegate=" means "yes, but no controllers". Show this as "(none)". */
839 const char *delegate_str
= delegate_controllers_str
?: c
->delegate
? "(none)" : "no";
841 cpuset_cpus
= cpu_set_to_range_string(&c
->cpuset_cpus
);
842 startup_cpuset_cpus
= cpu_set_to_range_string(&c
->startup_cpuset_cpus
);
843 cpuset_mems
= cpu_set_to_range_string(&c
->cpuset_mems
);
844 startup_cpuset_mems
= cpu_set_to_range_string(&c
->startup_cpuset_mems
);
847 "%sCPUAccounting: %s\n"
848 "%sIOAccounting: %s\n"
849 "%sBlockIOAccounting: %s\n"
850 "%sMemoryAccounting: %s\n"
851 "%sTasksAccounting: %s\n"
852 "%sIPAccounting: %s\n"
853 "%sCPUWeight: %" PRIu64
"\n"
854 "%sStartupCPUWeight: %" PRIu64
"\n"
855 "%sCPUShares: %" PRIu64
"\n"
856 "%sStartupCPUShares: %" PRIu64
"\n"
857 "%sCPUQuotaPerSecSec: %s\n"
858 "%sCPUQuotaPeriodSec: %s\n"
859 "%sAllowedCPUs: %s\n"
860 "%sStartupAllowedCPUs: %s\n"
861 "%sAllowedMemoryNodes: %s\n"
862 "%sStartupAllowedMemoryNodes: %s\n"
863 "%sIOWeight: %" PRIu64
"\n"
864 "%sStartupIOWeight: %" PRIu64
"\n"
865 "%sBlockIOWeight: %" PRIu64
"\n"
866 "%sStartupBlockIOWeight: %" PRIu64
"\n"
867 "%sDefaultMemoryMin: %" PRIu64
"\n"
868 "%sDefaultMemoryLow: %" PRIu64
"\n"
869 "%sMemoryMin: %" PRIu64
"%s\n"
870 "%sMemoryLow: %" PRIu64
"%s\n"
871 "%sStartupMemoryLow: %" PRIu64
"%s\n"
872 "%sMemoryHigh: %" PRIu64
"%s\n"
873 "%sStartupMemoryHigh: %" PRIu64
"%s\n"
874 "%sMemoryMax: %" PRIu64
"%s\n"
875 "%sStartupMemoryMax: %" PRIu64
"%s\n"
876 "%sMemorySwapMax: %" PRIu64
"%s\n"
877 "%sStartupMemorySwapMax: %" PRIu64
"%s\n"
878 "%sMemoryZSwapMax: %" PRIu64
"%s\n"
879 "%sStartupMemoryZSwapMax: %" PRIu64
"%s\n"
880 "%sMemoryLimit: %" PRIu64
"\n"
881 "%sTasksMax: %" PRIu64
"\n"
882 "%sDevicePolicy: %s\n"
883 "%sDisableControllers: %s\n"
885 "%sManagedOOMSwap: %s\n"
886 "%sManagedOOMMemoryPressure: %s\n"
887 "%sManagedOOMMemoryPressureLimit: " PERMYRIAD_AS_PERCENT_FORMAT_STR
"\n"
888 "%sManagedOOMPreference: %s\n"
889 "%sMemoryPressureWatch: %s\n"
890 "%sCoredumpReceive: %s\n",
891 prefix
, yes_no(c
->cpu_accounting
),
892 prefix
, yes_no(c
->io_accounting
),
893 prefix
, yes_no(c
->blockio_accounting
),
894 prefix
, yes_no(c
->memory_accounting
),
895 prefix
, yes_no(c
->tasks_accounting
),
896 prefix
, yes_no(c
->ip_accounting
),
897 prefix
, c
->cpu_weight
,
898 prefix
, c
->startup_cpu_weight
,
899 prefix
, c
->cpu_shares
,
900 prefix
, c
->startup_cpu_shares
,
901 prefix
, FORMAT_TIMESPAN(c
->cpu_quota_per_sec_usec
, 1),
902 prefix
, FORMAT_TIMESPAN(c
->cpu_quota_period_usec
, 1),
903 prefix
, strempty(cpuset_cpus
),
904 prefix
, strempty(startup_cpuset_cpus
),
905 prefix
, strempty(cpuset_mems
),
906 prefix
, strempty(startup_cpuset_mems
),
907 prefix
, c
->io_weight
,
908 prefix
, c
->startup_io_weight
,
909 prefix
, c
->blockio_weight
,
910 prefix
, c
->startup_blockio_weight
,
911 prefix
, c
->default_memory_min
,
912 prefix
, c
->default_memory_low
,
913 prefix
, c
->memory_min
, format_cgroup_memory_limit_comparison(cda
, sizeof(cda
), u
, "MemoryMin"),
914 prefix
, c
->memory_low
, format_cgroup_memory_limit_comparison(cdb
, sizeof(cdb
), u
, "MemoryLow"),
915 prefix
, c
->startup_memory_low
, format_cgroup_memory_limit_comparison(cdc
, sizeof(cdc
), u
, "StartupMemoryLow"),
916 prefix
, c
->memory_high
, format_cgroup_memory_limit_comparison(cdd
, sizeof(cdd
), u
, "MemoryHigh"),
917 prefix
, c
->startup_memory_high
, format_cgroup_memory_limit_comparison(cde
, sizeof(cde
), u
, "StartupMemoryHigh"),
918 prefix
, c
->memory_max
, format_cgroup_memory_limit_comparison(cdf
, sizeof(cdf
), u
, "MemoryMax"),
919 prefix
, c
->startup_memory_max
, format_cgroup_memory_limit_comparison(cdg
, sizeof(cdg
), u
, "StartupMemoryMax"),
920 prefix
, c
->memory_swap_max
, format_cgroup_memory_limit_comparison(cdh
, sizeof(cdh
), u
, "MemorySwapMax"),
921 prefix
, c
->startup_memory_swap_max
, format_cgroup_memory_limit_comparison(cdi
, sizeof(cdi
), u
, "StartupMemorySwapMax"),
922 prefix
, c
->memory_zswap_max
, format_cgroup_memory_limit_comparison(cdj
, sizeof(cdj
), u
, "MemoryZSwapMax"),
923 prefix
, c
->startup_memory_zswap_max
, format_cgroup_memory_limit_comparison(cdk
, sizeof(cdk
), u
, "StartupMemoryZSwapMax"),
924 prefix
, c
->memory_limit
,
925 prefix
, cgroup_tasks_max_resolve(&c
->tasks_max
),
926 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
927 prefix
, strempty(disable_controllers_str
),
928 prefix
, delegate_str
,
929 prefix
, managed_oom_mode_to_string(c
->moom_swap
),
930 prefix
, managed_oom_mode_to_string(c
->moom_mem_pressure
),
931 prefix
, PERMYRIAD_AS_PERCENT_FORMAT_VAL(UINT32_SCALE_TO_PERMYRIAD(c
->moom_mem_pressure_limit
)),
932 prefix
, managed_oom_preference_to_string(c
->moom_preference
),
933 prefix
, cgroup_pressure_watch_to_string(c
->memory_pressure_watch
),
934 prefix
, yes_no(c
->coredump_receive
));
936 if (c
->delegate_subgroup
)
937 fprintf(f
, "%sDelegateSubgroup: %s\n",
938 prefix
, c
->delegate_subgroup
);
940 if (c
->memory_pressure_threshold_usec
!= USEC_INFINITY
)
941 fprintf(f
, "%sMemoryPressureThresholdSec: %s\n",
942 prefix
, FORMAT_TIMESPAN(c
->memory_pressure_threshold_usec
, 1));
944 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
945 /* strna() below should be redundant, for avoiding -Werror=format-overflow= error. See #30223. */
947 "%sDeviceAllow: %s %s\n",
950 strna(cgroup_device_permissions_to_string(a
->permissions
)));
952 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
954 "%sIODeviceWeight: %s %" PRIu64
"\n",
959 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
961 "%sIODeviceLatencyTargetSec: %s %s\n",
964 FORMAT_TIMESPAN(l
->target_usec
, 1));
966 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
)
967 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
968 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
972 cgroup_io_limit_type_to_string(type
),
974 FORMAT_BYTES(il
->limits
[type
]));
976 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
978 "%sBlockIODeviceWeight: %s %" PRIu64
,
983 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
984 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
986 "%sBlockIOReadBandwidth: %s %s\n",
989 FORMAT_BYTES(b
->rbps
));
990 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
992 "%sBlockIOWriteBandwidth: %s %s\n",
995 FORMAT_BYTES(b
->wbps
));
998 SET_FOREACH(iaai
, c
->ip_address_allow
)
999 fprintf(f
, "%sIPAddressAllow: %s\n", prefix
,
1000 IN_ADDR_PREFIX_TO_STRING(iaai
->family
, &iaai
->address
, iaai
->prefixlen
));
1001 SET_FOREACH(iaai
, c
->ip_address_deny
)
1002 fprintf(f
, "%sIPAddressDeny: %s\n", prefix
,
1003 IN_ADDR_PREFIX_TO_STRING(iaai
->family
, &iaai
->address
, iaai
->prefixlen
));
1005 STRV_FOREACH(path
, c
->ip_filters_ingress
)
1006 fprintf(f
, "%sIPIngressFilterPath: %s\n", prefix
, *path
);
1007 STRV_FOREACH(path
, c
->ip_filters_egress
)
1008 fprintf(f
, "%sIPEgressFilterPath: %s\n", prefix
, *path
);
1010 LIST_FOREACH(programs
, p
, c
->bpf_foreign_programs
)
1011 fprintf(f
, "%sBPFProgram: %s:%s",
1012 prefix
, bpf_cgroup_attach_type_to_string(p
->attach_type
), p
->bpffs_path
);
1014 if (c
->socket_bind_allow
) {
1015 fprintf(f
, "%sSocketBindAllow: ", prefix
);
1016 cgroup_context_dump_socket_bind_items(c
->socket_bind_allow
, f
);
1020 if (c
->socket_bind_deny
) {
1021 fprintf(f
, "%sSocketBindDeny: ", prefix
);
1022 cgroup_context_dump_socket_bind_items(c
->socket_bind_deny
, f
);
1026 if (c
->restrict_network_interfaces
) {
1028 SET_FOREACH(iface
, c
->restrict_network_interfaces
)
1029 fprintf(f
, "%sRestrictNetworkInterfaces: %s\n", prefix
, iface
);
1032 FOREACH_ARRAY(nft_set
, c
->nft_set_context
.sets
, c
->nft_set_context
.n_sets
)
1033 fprintf(f
, "%sNFTSet: %s:%s:%s:%s\n", prefix
, nft_set_source_to_string(nft_set
->source
),
1034 nfproto_to_string(nft_set
->nfproto
), nft_set
->table
, nft_set
->set
);
1037 void cgroup_context_dump_socket_bind_item(const CGroupSocketBindItem
*item
, FILE *f
) {
1038 const char *family
, *colon1
, *protocol
= "", *colon2
= "";
1040 family
= strempty(af_to_ipv4_ipv6(item
->address_family
));
1041 colon1
= isempty(family
) ? "" : ":";
1043 if (item
->ip_protocol
!= 0) {
1044 protocol
= ip_protocol_to_tcp_udp(item
->ip_protocol
);
1048 if (item
->nr_ports
== 0)
1049 fprintf(f
, "%s%s%s%sany", family
, colon1
, protocol
, colon2
);
1050 else if (item
->nr_ports
== 1)
1051 fprintf(f
, "%s%s%s%s%" PRIu16
, family
, colon1
, protocol
, colon2
, item
->port_min
);
1053 uint16_t port_max
= item
->port_min
+ item
->nr_ports
- 1;
1054 fprintf(f
, "%s%s%s%s%" PRIu16
"-%" PRIu16
, family
, colon1
, protocol
, colon2
,
1055 item
->port_min
, port_max
);
1059 void cgroup_context_dump_socket_bind_items(const CGroupSocketBindItem
*items
, FILE *f
) {
1062 LIST_FOREACH(socket_bind_items
, bi
, items
) {
1068 cgroup_context_dump_socket_bind_item(bi
, f
);
1072 int cgroup_context_add_device_allow(CGroupContext
*c
, const char *dev
, CGroupDevicePermissions p
) {
1073 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
1074 _cleanup_free_
char *d
= NULL
;
1078 assert(p
>= 0 && p
< _CGROUP_DEVICE_PERMISSIONS_MAX
);
1081 p
= _CGROUP_DEVICE_PERMISSIONS_ALL
;
1083 a
= new(CGroupDeviceAllow
, 1);
1091 *a
= (CGroupDeviceAllow
) {
1092 .path
= TAKE_PTR(d
),
1096 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
1102 int cgroup_context_add_or_update_device_allow(CGroupContext
*c
, const char *dev
, CGroupDevicePermissions p
) {
1105 assert(p
>= 0 && p
< _CGROUP_DEVICE_PERMISSIONS_MAX
);
1108 p
= _CGROUP_DEVICE_PERMISSIONS_ALL
;
1110 LIST_FOREACH(device_allow
, b
, c
->device_allow
)
1111 if (path_equal(b
->path
, dev
)) {
1116 return cgroup_context_add_device_allow(c
, dev
, p
);
1119 int cgroup_context_add_bpf_foreign_program(CGroupContext
*c
, uint32_t attach_type
, const char *bpffs_path
) {
1120 CGroupBPFForeignProgram
*p
;
1121 _cleanup_free_
char *d
= NULL
;
1126 if (!path_is_normalized(bpffs_path
) || !path_is_absolute(bpffs_path
))
1127 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Path is not normalized: %m");
1129 d
= strdup(bpffs_path
);
1133 p
= new(CGroupBPFForeignProgram
, 1);
1137 *p
= (CGroupBPFForeignProgram
) {
1138 .attach_type
= attach_type
,
1139 .bpffs_path
= TAKE_PTR(d
),
1142 LIST_PREPEND(programs
, c
->bpf_foreign_programs
, TAKE_PTR(p
));
1147 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
1148 uint64_t unit_get_ancestor_##entry(Unit *u) { \
1151 /* 1. Is entry set in this unit? If so, use that. \
1152 * 2. Is the default for this entry set in any \
1153 * ancestor? If so, use that. \
1154 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
1158 c = unit_get_cgroup_context(u); \
1159 if (c && c->entry##_set) \
1162 while ((u = UNIT_GET_SLICE(u))) { \
1163 c = unit_get_cgroup_context(u); \
1164 if (c && c->default_##entry##_set) \
1165 return c->default_##entry; \
1168 /* We've reached the root, but nobody had default for \
1169 * this entry set, so set it to the kernel default. */ \
1170 return CGROUP_LIMIT_MIN; \
1173 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
1174 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(startup_memory_low
);
1175 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
1177 static void unit_set_xattr_graceful(Unit
*u
, const char *name
, const void *data
, size_t size
) {
1183 if (!u
->cgroup_path
)
1186 r
= cg_set_xattr(u
->cgroup_path
, name
, data
, size
, 0);
1188 log_unit_debug_errno(u
, r
, "Failed to set '%s' xattr on control group %s, ignoring: %m", name
, empty_to_root(u
->cgroup_path
));
1191 static void unit_remove_xattr_graceful(Unit
*u
, const char *name
) {
1197 if (!u
->cgroup_path
)
1200 r
= cg_remove_xattr(u
->cgroup_path
, name
);
1201 if (r
< 0 && !ERRNO_IS_XATTR_ABSENT(r
))
1202 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
));
1205 static void cgroup_oomd_xattr_apply(Unit
*u
) {
1210 c
= unit_get_cgroup_context(u
);
1214 if (c
->moom_preference
== MANAGED_OOM_PREFERENCE_OMIT
)
1215 unit_set_xattr_graceful(u
, "user.oomd_omit", "1", 1);
1217 if (c
->moom_preference
== MANAGED_OOM_PREFERENCE_AVOID
)
1218 unit_set_xattr_graceful(u
, "user.oomd_avoid", "1", 1);
1220 if (c
->moom_preference
!= MANAGED_OOM_PREFERENCE_AVOID
)
1221 unit_remove_xattr_graceful(u
, "user.oomd_avoid");
1223 if (c
->moom_preference
!= MANAGED_OOM_PREFERENCE_OMIT
)
1224 unit_remove_xattr_graceful(u
, "user.oomd_omit");
1227 static int cgroup_log_xattr_apply(Unit
*u
) {
1229 size_t len
, allowed_patterns_len
, denied_patterns_len
;
1230 _cleanup_free_
char *patterns
= NULL
, *allowed_patterns
= NULL
, *denied_patterns
= NULL
;
1236 c
= unit_get_exec_context(u
);
1238 /* Some unit types have a cgroup context but no exec context, so we do not log
1239 * any error here to avoid confusion. */
1242 if (set_isempty(c
->log_filter_allowed_patterns
) && set_isempty(c
->log_filter_denied_patterns
)) {
1243 unit_remove_xattr_graceful(u
, "user.journald_log_filter_patterns");
1247 r
= set_make_nulstr(c
->log_filter_allowed_patterns
, &allowed_patterns
, &allowed_patterns_len
);
1249 return log_debug_errno(r
, "Failed to make nulstr from set: %m");
1251 r
= set_make_nulstr(c
->log_filter_denied_patterns
, &denied_patterns
, &denied_patterns_len
);
1253 return log_debug_errno(r
, "Failed to make nulstr from set: %m");
1255 /* Use nul character separated strings without trailing nul */
1256 allowed_patterns_len
= LESS_BY(allowed_patterns_len
, 1u);
1257 denied_patterns_len
= LESS_BY(denied_patterns_len
, 1u);
1259 len
= allowed_patterns_len
+ 1 + denied_patterns_len
;
1260 patterns
= new(char, len
);
1262 return log_oom_debug();
1264 last
= mempcpy_safe(patterns
, allowed_patterns
, allowed_patterns_len
);
1266 memcpy_safe(last
, denied_patterns
, denied_patterns_len
);
1268 unit_set_xattr_graceful(u
, "user.journald_log_filter_patterns", patterns
, len
);
1273 static void cgroup_invocation_id_xattr_apply(Unit
*u
) {
1278 b
= !sd_id128_is_null(u
->invocation_id
);
1279 FOREACH_STRING(xn
, "trusted.invocation_id", "user.invocation_id") {
1281 unit_set_xattr_graceful(u
, xn
, SD_ID128_TO_STRING(u
->invocation_id
), 32);
1283 unit_remove_xattr_graceful(u
, xn
);
1287 static void cgroup_coredump_xattr_apply(Unit
*u
) {
1292 c
= unit_get_cgroup_context(u
);
1296 if (unit_cgroup_delegate(u
) && c
->coredump_receive
)
1297 unit_set_xattr_graceful(u
, "user.coredump_receive", "1", 1);
1299 unit_remove_xattr_graceful(u
, "user.coredump_receive");
1302 static void cgroup_delegate_xattr_apply(Unit
*u
) {
1307 /* Indicate on the cgroup whether delegation is on, via an xattr. This is best-effort, as old kernels
1308 * didn't support xattrs on cgroups at all. Later they got support for setting 'trusted.*' xattrs,
1309 * and even later 'user.*' xattrs. We started setting this field when 'trusted.*' was added, and
1310 * given this is now pretty much API, let's continue to support that. But also set 'user.*' as well,
1311 * since it is readable by any user, not just CAP_SYS_ADMIN. This hence comes with slightly weaker
1312 * security (as users who got delegated cgroups could turn it off if they like), but this shouldn't
1313 * be a big problem given this communicates delegation state to clients, but the manager never reads
1315 b
= unit_cgroup_delegate(u
);
1316 FOREACH_STRING(xn
, "trusted.delegate", "user.delegate") {
1318 unit_set_xattr_graceful(u
, xn
, "1", 1);
1320 unit_remove_xattr_graceful(u
, xn
);
1324 static void cgroup_survive_xattr_apply(Unit
*u
) {
1329 if (u
->survive_final_kill_signal
) {
1332 "user.survive_final_kill_signal",
1336 /* user xattr support was added in kernel v5.7 */
1337 if (ERRNO_IS_NEG_NOT_SUPPORTED(r
))
1340 "trusted.survive_final_kill_signal",
1345 log_unit_debug_errno(u
,
1347 "Failed to set 'survive_final_kill_signal' xattr on control "
1348 "group %s, ignoring: %m",
1349 empty_to_root(u
->cgroup_path
));
1351 unit_remove_xattr_graceful(u
, "user.survive_final_kill_signal");
1352 unit_remove_xattr_graceful(u
, "trusted.survive_final_kill_signal");
1356 static void cgroup_xattr_apply(Unit
*u
) {
1359 /* The 'user.*' xattrs can be set from a user manager. */
1360 cgroup_oomd_xattr_apply(u
);
1361 cgroup_log_xattr_apply(u
);
1362 cgroup_coredump_xattr_apply(u
);
1364 if (!MANAGER_IS_SYSTEM(u
->manager
))
1367 cgroup_invocation_id_xattr_apply(u
);
1368 cgroup_delegate_xattr_apply(u
);
1369 cgroup_survive_xattr_apply(u
);
1372 static int lookup_block_device(const char *p
, dev_t
*ret
) {
1373 dev_t rdev
, dev
= 0;
1380 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
1381 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
1384 if (stat(p
, &st
) < 0)
1385 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
1391 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
1394 return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK
),
1395 "Device node '%s' is a character device, but block device needed.", p
);
1398 else if (major(dev
) != 0)
1399 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
1401 /* If this is btrfs, getting the backing block device is a bit harder */
1402 r
= btrfs_get_block_device(p
, ret
);
1404 return log_warning_errno(SYNTHETIC_ERRNO(ENODEV
),
1405 "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
1407 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
1410 /* If this is a LUKS/DM device, recursively try to get the originating block device */
1411 while (block_get_originating(*ret
, ret
) > 0);
1413 /* If this is a partition, try to get the originating block device */
1414 (void) block_get_whole_disk(*ret
, ret
);
1418 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
1419 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
1420 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
1423 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
1424 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
1425 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
1428 static bool cgroup_context_has_allowed_cpus(CGroupContext
*c
) {
1429 return c
->cpuset_cpus
.set
|| c
->startup_cpuset_cpus
.set
;
1432 static bool cgroup_context_has_allowed_mems(CGroupContext
*c
) {
1433 return c
->cpuset_mems
.set
|| c
->startup_cpuset_mems
.set
;
1436 uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
1439 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1440 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
1441 return c
->startup_cpu_weight
;
1442 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
1443 return c
->cpu_weight
;
1445 return CGROUP_WEIGHT_DEFAULT
;
1448 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
1449 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1450 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
1451 return c
->startup_cpu_shares
;
1452 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
1453 return c
->cpu_shares
;
1455 return CGROUP_CPU_SHARES_DEFAULT
;
1458 static CPUSet
*cgroup_context_allowed_cpus(CGroupContext
*c
, ManagerState state
) {
1459 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1460 c
->startup_cpuset_cpus
.set
)
1461 return &c
->startup_cpuset_cpus
;
1463 return &c
->cpuset_cpus
;
1466 static CPUSet
*cgroup_context_allowed_mems(CGroupContext
*c
, ManagerState state
) {
1467 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1468 c
->startup_cpuset_mems
.set
)
1469 return &c
->startup_cpuset_mems
;
1471 return &c
->cpuset_mems
;
1474 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
1475 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
1476 * need to be higher than that boundary. quota is specified in USecPerSec.
1477 * Additionally, period must be at most max_period. */
1480 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
1483 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
1486 if (quota
== USEC_INFINITY
)
1487 /* Always use default period for infinity quota. */
1488 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
1490 if (period
== USEC_INFINITY
)
1491 /* Default period was requested. */
1492 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
1494 /* Clamp to interval [1ms, 1s] */
1495 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
1497 if (new_period
!= period
) {
1498 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
,
1499 "Clamping CPU interval for cpu.max: period is now %s",
1500 FORMAT_TIMESPAN(new_period
, 1));
1501 u
->warned_clamping_cpu_quota_period
= true;
1507 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
1508 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
1510 if (weight
== CGROUP_WEIGHT_IDLE
)
1512 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1513 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
1516 static void cgroup_apply_unified_cpu_idle(Unit
*u
, uint64_t weight
) {
1519 const char *idle_val
;
1521 is_idle
= weight
== CGROUP_WEIGHT_IDLE
;
1522 idle_val
= one_zero(is_idle
);
1523 r
= cg_set_attribute("cpu", u
->cgroup_path
, "cpu.idle", idle_val
);
1524 if (r
< 0 && (r
!= -ENOENT
|| is_idle
))
1525 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%s': %m",
1526 "cpu.idle", empty_to_root(u
->cgroup_path
), idle_val
);
1529 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
1530 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
1532 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
1533 if (quota
!= USEC_INFINITY
)
1534 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
1535 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
1537 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
1538 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
1541 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
1542 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
1544 xsprintf(buf
, "%" PRIu64
"\n", shares
);
1545 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
1548 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
1549 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
1551 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
1553 xsprintf(buf
, USEC_FMT
"\n", period
);
1554 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
1556 if (quota
!= USEC_INFINITY
) {
1557 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
1558 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
1560 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
1563 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
1564 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
1565 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
1568 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
1569 /* we don't support idle in cgroupv1 */
1570 if (weight
== CGROUP_WEIGHT_IDLE
)
1571 return CGROUP_CPU_SHARES_MIN
;
1573 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
1574 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
1577 static void cgroup_apply_unified_cpuset(Unit
*u
, const CPUSet
*cpus
, const char *name
) {
1578 _cleanup_free_
char *buf
= NULL
;
1580 buf
= cpu_set_to_range_string(cpus
);
1586 (void) set_attribute_and_warn(u
, "cpuset", name
, buf
);
1589 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
1590 return c
->io_accounting
||
1591 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
1592 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
1593 c
->io_device_weights
||
1594 c
->io_device_latencies
||
1595 c
->io_device_limits
;
1598 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
1599 return c
->blockio_accounting
||
1600 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
1601 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
1602 c
->blockio_device_weights
||
1603 c
->blockio_device_bandwidths
;
1606 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
1607 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1608 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
1609 return c
->startup_io_weight
;
1610 if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
1611 return c
->io_weight
;
1612 return CGROUP_WEIGHT_DEFAULT
;
1615 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
1616 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1617 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
1618 return c
->startup_blockio_weight
;
1619 if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
1620 return c
->blockio_weight
;
1621 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
1624 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
1625 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
1626 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
1629 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
1630 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
1631 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
1634 static int set_bfq_weight(Unit
*u
, const char *controller
, dev_t dev
, uint64_t io_weight
) {
1635 static const char * const prop_names
[] = {
1639 "BlockIODeviceWeight",
1641 static bool warned
= false;
1642 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+STRLEN("\n")];
1644 uint64_t bfq_weight
;
1647 /* FIXME: drop this function when distro kernels properly support BFQ through "io.weight"
1648 * See also: https://github.com/systemd/systemd/pull/13335 and
1649 * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9. */
1650 p
= strjoina(controller
, ".bfq.weight");
1651 /* Adjust to kernel range is 1..1000, the default is 100. */
1652 bfq_weight
= BFQ_WEIGHT(io_weight
);
1655 xsprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), bfq_weight
);
1657 xsprintf(buf
, "%" PRIu64
"\n", bfq_weight
);
1659 r
= cg_set_attribute(controller
, u
->cgroup_path
, p
, buf
);
1661 /* FIXME: drop this when kernels prior
1662 * 795fe54c2a82 ("bfq: Add per-device weight") v5.4
1663 * are not interesting anymore. Old kernels will fail with EINVAL, while new kernels won't return
1664 * EINVAL on properly formatted input by us. Treat EINVAL accordingly. */
1665 if (r
== -EINVAL
&& major(dev
) > 0) {
1667 log_unit_warning(u
, "Kernel version does not accept per-device setting in %s.", p
);
1670 r
= -EOPNOTSUPP
; /* mask as unconfigured device */
1671 } else if (r
>= 0 && io_weight
!= bfq_weight
)
1672 log_unit_debug(u
, "%s=%" PRIu64
" scaled to %s=%" PRIu64
,
1673 prop_names
[2*(major(dev
) > 0) + streq(controller
, "blkio")],
1674 io_weight
, p
, bfq_weight
);
1678 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
1679 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1683 if (lookup_block_device(dev_path
, &dev
) < 0)
1686 r1
= set_bfq_weight(u
, "io", dev
, io_weight
);
1688 xsprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), io_weight
);
1689 r2
= cg_set_attribute("io", u
->cgroup_path
, "io.weight", buf
);
1691 /* Look at the configured device, when both fail, prefer io.weight errno. */
1692 r
= r2
== -EOPNOTSUPP
? r1
: r2
;
1695 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
),
1696 r
, "Failed to set 'io[.bfq].weight' attribute on '%s' to '%.*s': %m",
1697 empty_to_root(u
->cgroup_path
), (int) strcspn(buf
, NEWLINE
), buf
);
1700 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
1701 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1705 r
= lookup_block_device(dev_path
, &dev
);
1709 xsprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), blkio_weight
);
1710 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
1713 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
1714 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
1718 r
= lookup_block_device(dev_path
, &dev
);
1722 if (target
!= USEC_INFINITY
)
1723 xsprintf(buf
, DEVNUM_FORMAT_STR
" target=%" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), target
);
1725 xsprintf(buf
, DEVNUM_FORMAT_STR
" target=max\n", DEVNUM_FORMAT_VAL(dev
));
1727 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
1730 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
1731 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)],
1732 buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
1735 if (lookup_block_device(dev_path
, &dev
) < 0)
1738 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1739 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
1740 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
1742 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
1744 xsprintf(buf
, DEVNUM_FORMAT_STR
" rbps=%s wbps=%s riops=%s wiops=%s\n", DEVNUM_FORMAT_VAL(dev
),
1745 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
1746 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
1747 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
1750 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
1751 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1754 if (lookup_block_device(dev_path
, &dev
) < 0)
1757 sprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), rbps
);
1758 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
1760 sprintf(buf
, DEVNUM_FORMAT_STR
" %" PRIu64
"\n", DEVNUM_FORMAT_VAL(dev
), wbps
);
1761 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
1764 static bool unit_has_unified_memory_config(Unit
*u
) {
1769 assert_se(c
= unit_get_cgroup_context(u
));
1771 return unit_get_ancestor_memory_min(u
) > 0 ||
1772 unit_get_ancestor_memory_low(u
) > 0 || unit_get_ancestor_startup_memory_low(u
) > 0 ||
1773 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->startup_memory_high_set
||
1774 c
->memory_max
!= CGROUP_LIMIT_MAX
|| c
->startup_memory_max_set
||
1775 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
|| c
->startup_memory_swap_max_set
||
1776 c
->memory_zswap_max
!= CGROUP_LIMIT_MAX
|| c
->startup_memory_zswap_max_set
;
1779 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
1780 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
1782 if (v
!= CGROUP_LIMIT_MAX
)
1783 xsprintf(buf
, "%" PRIu64
"\n", v
);
1785 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
1788 static void cgroup_apply_firewall(Unit
*u
) {
1791 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
1793 if (bpf_firewall_compile(u
) < 0)
1796 (void) bpf_firewall_load_custom(u
);
1797 (void) bpf_firewall_install(u
);
1800 void unit_modify_nft_set(Unit
*u
, bool add
) {
1805 if (!MANAGER_IS_SYSTEM(u
->manager
))
1808 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1811 if (cg_all_unified() <= 0)
1814 if (u
->cgroup_id
== 0)
1817 if (!u
->manager
->fw_ctx
) {
1818 r
= fw_ctx_new_full(&u
->manager
->fw_ctx
, /* init_tables= */ false);
1822 assert(u
->manager
->fw_ctx
);
1825 CGroupContext
*c
= ASSERT_PTR(unit_get_cgroup_context(u
));
1827 FOREACH_ARRAY(nft_set
, c
->nft_set_context
.sets
, c
->nft_set_context
.n_sets
) {
1828 if (nft_set
->source
!= NFT_SET_SOURCE_CGROUP
)
1831 uint64_t element
= u
->cgroup_id
;
1833 r
= nft_set_element_modify_any(u
->manager
->fw_ctx
, add
, nft_set
->nfproto
, nft_set
->table
, nft_set
->set
, &element
, sizeof(element
));
1835 log_warning_errno(r
, "Failed to %s NFT set: family %s, table %s, set %s, cgroup %" PRIu64
", ignoring: %m",
1836 add
? "add" : "delete", nfproto_to_string(nft_set
->nfproto
), nft_set
->table
, nft_set
->set
, u
->cgroup_id
);
1838 log_debug("%s NFT set: family %s, table %s, set %s, cgroup %" PRIu64
,
1839 add
? "Added" : "Deleted", nfproto_to_string(nft_set
->nfproto
), nft_set
->table
, nft_set
->set
, u
->cgroup_id
);
1843 static void cgroup_apply_socket_bind(Unit
*u
) {
1846 (void) bpf_socket_bind_install(u
);
1849 static void cgroup_apply_restrict_network_interfaces(Unit
*u
) {
1852 (void) bpf_restrict_ifaces_install(u
);
1855 static int cgroup_apply_devices(Unit
*u
) {
1856 _cleanup_(bpf_program_freep
) BPFProgram
*prog
= NULL
;
1859 CGroupDevicePolicy policy
;
1862 assert_se(c
= unit_get_cgroup_context(u
));
1863 assert_se(path
= u
->cgroup_path
);
1865 policy
= c
->device_policy
;
1867 if (cg_all_unified() > 0) {
1868 r
= bpf_devices_cgroup_init(&prog
, policy
, c
->device_allow
);
1870 return log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
1873 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore
1876 if (c
->device_allow
|| policy
!= CGROUP_DEVICE_POLICY_AUTO
)
1877 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
1879 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
1881 log_unit_full_errno(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1882 "Failed to reset devices.allow/devices.deny: %m");
1885 bool allow_list_static
= policy
== CGROUP_DEVICE_POLICY_CLOSED
||
1886 (policy
== CGROUP_DEVICE_POLICY_AUTO
&& c
->device_allow
);
1887 if (allow_list_static
)
1888 (void) bpf_devices_allow_list_static(prog
, path
);
1890 bool any
= allow_list_static
;
1891 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1894 if (a
->permissions
== 0)
1897 if (path_startswith(a
->path
, "/dev/"))
1898 r
= bpf_devices_allow_list_device(prog
, path
, a
->path
, a
->permissions
);
1899 else if ((val
= startswith(a
->path
, "block-")))
1900 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'b', a
->permissions
);
1901 else if ((val
= startswith(a
->path
, "char-")))
1902 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'c', a
->permissions
);
1904 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1913 log_unit_warning_errno(u
, SYNTHETIC_ERRNO(ENODEV
), "No devices matched by device filter.");
1915 /* The kernel verifier would reject a program we would build with the normal intro and outro
1916 but no allow-listing rules (outro would contain an unreachable instruction for successful
1918 policy
= CGROUP_DEVICE_POLICY_STRICT
;
1921 r
= bpf_devices_apply_policy(&prog
, policy
, any
, path
, &u
->bpf_device_control_installed
);
1923 static bool warned
= false;
1925 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1926 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1927 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1928 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1935 static void set_io_weight(Unit
*u
, uint64_t weight
) {
1936 char buf
[STRLEN("default \n")+DECIMAL_STR_MAX(uint64_t)];
1940 (void) set_bfq_weight(u
, "io", makedev(0, 0), weight
);
1942 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1943 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1946 static void set_blkio_weight(Unit
*u
, uint64_t weight
) {
1947 char buf
[STRLEN("\n")+DECIMAL_STR_MAX(uint64_t)];
1951 (void) set_bfq_weight(u
, "blkio", makedev(0, 0), weight
);
1953 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1954 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1957 static void cgroup_apply_bpf_foreign_program(Unit
*u
) {
1960 (void) bpf_foreign_install(u
);
1963 static void cgroup_context_apply(
1965 CGroupMask apply_mask
,
1966 ManagerState state
) {
1970 bool is_host_root
, is_local_root
;
1975 /* Nothing to do? Exit early! */
1976 if (apply_mask
== 0)
1979 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
1980 * attributes should only be managed for cgroups further down the tree. */
1981 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1982 is_host_root
= unit_has_host_root_cgroup(u
);
1984 assert_se(c
= unit_get_cgroup_context(u
));
1985 assert_se(path
= u
->cgroup_path
);
1987 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
1990 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
1991 * then), and missing cgroups, i.e. EROFS and ENOENT. */
1993 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
1994 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
1995 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
1996 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
1997 * we couldn't even write to them if we wanted to). */
1998 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
2000 if (cg_all_unified() > 0) {
2003 if (cgroup_context_has_cpu_weight(c
))
2004 weight
= cgroup_context_cpu_weight(c
, state
);
2005 else if (cgroup_context_has_cpu_shares(c
)) {
2008 shares
= cgroup_context_cpu_shares(c
, state
);
2009 weight
= cgroup_cpu_shares_to_weight(shares
);
2011 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
2012 shares
, weight
, path
);
2014 weight
= CGROUP_WEIGHT_DEFAULT
;
2016 cgroup_apply_unified_cpu_idle(u
, weight
);
2017 cgroup_apply_unified_cpu_weight(u
, weight
);
2018 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
2023 if (cgroup_context_has_cpu_weight(c
)) {
2026 weight
= cgroup_context_cpu_weight(c
, state
);
2027 shares
= cgroup_cpu_weight_to_shares(weight
);
2029 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
2030 weight
, shares
, path
);
2031 } else if (cgroup_context_has_cpu_shares(c
))
2032 shares
= cgroup_context_cpu_shares(c
, state
);
2034 shares
= CGROUP_CPU_SHARES_DEFAULT
;
2036 cgroup_apply_legacy_cpu_shares(u
, shares
);
2037 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
2041 if ((apply_mask
& CGROUP_MASK_CPUSET
) && !is_local_root
) {
2042 cgroup_apply_unified_cpuset(u
, cgroup_context_allowed_cpus(c
, state
), "cpuset.cpus");
2043 cgroup_apply_unified_cpuset(u
, cgroup_context_allowed_mems(c
, state
), "cpuset.mems");
2046 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
2047 * controller), and in case of containers we want to leave control of these attributes to the container manager
2048 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
2049 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
2050 bool has_io
, has_blockio
;
2053 has_io
= cgroup_context_has_io_config(c
);
2054 has_blockio
= cgroup_context_has_blockio_config(c
);
2057 weight
= cgroup_context_io_weight(c
, state
);
2058 else if (has_blockio
) {
2059 uint64_t blkio_weight
;
2061 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
2062 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
2064 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
2065 blkio_weight
, weight
);
2067 weight
= CGROUP_WEIGHT_DEFAULT
;
2069 set_io_weight(u
, weight
);
2072 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
2073 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
2075 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
2076 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
2078 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
2079 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
2081 } else if (has_blockio
) {
2082 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
2083 weight
= cgroup_weight_blkio_to_io(w
->weight
);
2085 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
2086 w
->weight
, weight
, w
->path
);
2088 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
2091 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
2092 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
2094 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
2095 limits
[type
] = cgroup_io_limit_defaults
[type
];
2097 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
2098 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
2100 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
2101 b
->rbps
, b
->wbps
, b
->path
);
2103 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
2108 if (apply_mask
& CGROUP_MASK_BLKIO
) {
2109 bool has_io
, has_blockio
;
2111 has_io
= cgroup_context_has_io_config(c
);
2112 has_blockio
= cgroup_context_has_blockio_config(c
);
2114 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
2115 * left to our container manager, too. */
2116 if (!is_local_root
) {
2122 io_weight
= cgroup_context_io_weight(c
, state
);
2123 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
2125 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
2127 } else if (has_blockio
)
2128 weight
= cgroup_context_blkio_weight(c
, state
);
2130 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
2132 set_blkio_weight(u
, weight
);
2135 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
2136 weight
= cgroup_weight_io_to_blkio(w
->weight
);
2138 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
2139 w
->weight
, weight
, w
->path
);
2141 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
2143 else if (has_blockio
)
2144 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
2145 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
2148 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
2149 * roots, as there we want the container manager to handle it */
2150 if (is_host_root
|| !is_local_root
) {
2152 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
2153 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
2154 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
2156 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
2158 else if (has_blockio
)
2159 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
2160 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
2164 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
2165 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
2166 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
2167 * write to this if we wanted to.) */
2168 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
2170 if (cg_all_unified() > 0) {
2171 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
, zswap_max
= CGROUP_LIMIT_MAX
, high
= CGROUP_LIMIT_MAX
;
2173 if (unit_has_unified_memory_config(u
)) {
2174 bool startup
= IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
);
2176 high
= startup
&& c
->startup_memory_high_set
? c
->startup_memory_high
: c
->memory_high
;
2177 max
= startup
&& c
->startup_memory_max_set
? c
->startup_memory_max
: c
->memory_max
;
2178 swap_max
= startup
&& c
->startup_memory_swap_max_set
? c
->startup_memory_swap_max
: c
->memory_swap_max
;
2179 zswap_max
= startup
&& c
->startup_memory_zswap_max_set
? c
->startup_memory_zswap_max
: c
->memory_zswap_max
;
2181 max
= c
->memory_limit
;
2183 if (max
!= CGROUP_LIMIT_MAX
)
2184 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
2187 cgroup_apply_unified_memory_limit(u
, "memory.min", unit_get_ancestor_memory_min(u
));
2188 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
2189 cgroup_apply_unified_memory_limit(u
, "memory.high", high
);
2190 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
2191 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
2192 cgroup_apply_unified_memory_limit(u
, "memory.zswap.max", zswap_max
);
2194 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
2197 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
2200 if (unit_has_unified_memory_config(u
)) {
2201 val
= c
->memory_max
;
2202 if (val
!= CGROUP_LIMIT_MAX
)
2203 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIu64
" as MemoryLimit=", val
);
2205 val
= c
->memory_limit
;
2207 if (val
== CGROUP_LIMIT_MAX
)
2208 strncpy(buf
, "-1\n", sizeof(buf
));
2210 xsprintf(buf
, "%" PRIu64
"\n", val
);
2212 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
2216 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
2217 * containers, where we leave this to the manager */
2218 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
2219 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
))
2220 (void) cgroup_apply_devices(u
);
2222 if (apply_mask
& CGROUP_MASK_PIDS
) {
2225 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
2226 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
2227 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
2228 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
2229 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
2230 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
2231 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
2232 * it also counts. But if the user never set a limit through us (i.e. we are the default of
2233 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
2234 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
2235 * which is desirable so that there's an official way to release control of the sysctl from
2236 * systemd: set the limit to unbounded and reload. */
2238 if (cgroup_tasks_max_isset(&c
->tasks_max
)) {
2239 u
->manager
->sysctl_pid_max_changed
= true;
2240 r
= procfs_tasks_set_limit(cgroup_tasks_max_resolve(&c
->tasks_max
));
2241 } else if (u
->manager
->sysctl_pid_max_changed
)
2242 r
= procfs_tasks_set_limit(TASKS_MAX
);
2246 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
2247 "Failed to write to tasks limit sysctls: %m");
2250 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
2251 * leave it for the container manager. */
2252 if (!is_local_root
) {
2253 if (cgroup_tasks_max_isset(&c
->tasks_max
)) {
2254 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
2256 xsprintf(buf
, "%" PRIu64
"\n", cgroup_tasks_max_resolve(&c
->tasks_max
));
2257 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
2259 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
2263 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
2264 cgroup_apply_firewall(u
);
2266 if (apply_mask
& CGROUP_MASK_BPF_FOREIGN
)
2267 cgroup_apply_bpf_foreign_program(u
);
2269 if (apply_mask
& CGROUP_MASK_BPF_SOCKET_BIND
)
2270 cgroup_apply_socket_bind(u
);
2272 if (apply_mask
& CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
)
2273 cgroup_apply_restrict_network_interfaces(u
);
2275 unit_modify_nft_set(u
, /* add = */ true);
2278 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
2282 c
= unit_get_cgroup_context(u
);
2286 if (c
->ip_accounting
||
2287 !set_isempty(c
->ip_address_allow
) ||
2288 !set_isempty(c
->ip_address_deny
) ||
2289 c
->ip_filters_ingress
||
2290 c
->ip_filters_egress
)
2293 /* If any parent slice has an IP access list defined, it applies too */
2294 for (Unit
*p
= UNIT_GET_SLICE(u
); p
; p
= UNIT_GET_SLICE(p
)) {
2295 c
= unit_get_cgroup_context(p
);
2299 if (!set_isempty(c
->ip_address_allow
) ||
2300 !set_isempty(c
->ip_address_deny
))
2307 static bool unit_get_needs_bpf_foreign_program(Unit
*u
) {
2311 c
= unit_get_cgroup_context(u
);
2315 return !!c
->bpf_foreign_programs
;
2318 static bool unit_get_needs_socket_bind(Unit
*u
) {
2322 c
= unit_get_cgroup_context(u
);
2326 return c
->socket_bind_allow
|| c
->socket_bind_deny
;
2329 static bool unit_get_needs_restrict_network_interfaces(Unit
*u
) {
2333 c
= unit_get_cgroup_context(u
);
2337 return !set_isempty(c
->restrict_network_interfaces
);
2340 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
2341 CGroupMask mask
= 0;
2346 assert_se(c
= unit_get_cgroup_context(u
));
2348 /* Figure out which controllers we need, based on the cgroup context object */
2350 if (c
->cpu_accounting
)
2351 mask
|= get_cpu_accounting_mask();
2353 if (cgroup_context_has_cpu_weight(c
) ||
2354 cgroup_context_has_cpu_shares(c
) ||
2355 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
2356 mask
|= CGROUP_MASK_CPU
;
2358 if (cgroup_context_has_allowed_cpus(c
) || cgroup_context_has_allowed_mems(c
))
2359 mask
|= CGROUP_MASK_CPUSET
;
2361 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
2362 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
2364 if (c
->memory_accounting
||
2365 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
2366 unit_has_unified_memory_config(u
))
2367 mask
|= CGROUP_MASK_MEMORY
;
2369 if (c
->device_allow
||
2370 c
->device_policy
!= CGROUP_DEVICE_POLICY_AUTO
)
2371 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
2373 if (c
->tasks_accounting
||
2374 cgroup_tasks_max_isset(&c
->tasks_max
))
2375 mask
|= CGROUP_MASK_PIDS
;
2377 return CGROUP_MASK_EXTEND_JOINED(mask
);
2380 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
2381 CGroupMask mask
= 0;
2383 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
2386 if (unit_get_needs_bpf_firewall(u
))
2387 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2389 if (unit_get_needs_bpf_foreign_program(u
))
2390 mask
|= CGROUP_MASK_BPF_FOREIGN
;
2392 if (unit_get_needs_socket_bind(u
))
2393 mask
|= CGROUP_MASK_BPF_SOCKET_BIND
;
2395 if (unit_get_needs_restrict_network_interfaces(u
))
2396 mask
|= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
;
2401 CGroupMask
unit_get_own_mask(Unit
*u
) {
2404 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
2405 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
2407 if (u
->load_state
!= UNIT_LOADED
)
2410 c
= unit_get_cgroup_context(u
);
2414 return unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
);
2417 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
2420 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
2421 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
2423 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
2425 if (!unit_cgroup_delegate(u
))
2428 if (cg_all_unified() <= 0) {
2431 e
= unit_get_exec_context(u
);
2432 if (e
&& !exec_context_maintains_privileges(e
))
2436 assert_se(c
= unit_get_cgroup_context(u
));
2437 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
2440 static CGroupMask
unit_get_subtree_mask(Unit
*u
) {
2442 /* Returns the mask of this subtree, meaning of the group
2443 * itself and its children. */
2445 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
2448 CGroupMask
unit_get_members_mask(Unit
*u
) {
2451 /* Returns the mask of controllers all of the unit's children require, merged */
2453 if (u
->cgroup_members_mask_valid
)
2454 return u
->cgroup_members_mask
; /* Use cached value if possible */
2456 u
->cgroup_members_mask
= 0;
2458 if (u
->type
== UNIT_SLICE
) {
2461 UNIT_FOREACH_DEPENDENCY(member
, u
, UNIT_ATOM_SLICE_OF
)
2462 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
2465 u
->cgroup_members_mask_valid
= true;
2466 return u
->cgroup_members_mask
;
2469 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
2473 /* Returns the mask of controllers all of the unit's siblings
2474 * require, i.e. the members mask of the unit's parent slice
2475 * if there is one. */
2477 slice
= UNIT_GET_SLICE(u
);
2479 return unit_get_members_mask(slice
);
2481 return unit_get_subtree_mask(u
); /* we are the top-level slice */
2484 static CGroupMask
unit_get_disable_mask(Unit
*u
) {
2487 c
= unit_get_cgroup_context(u
);
2491 return c
->disable_controllers
;
2494 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
2499 mask
= unit_get_disable_mask(u
);
2501 /* Returns the mask of controllers which are marked as forcibly
2502 * disabled in any ancestor unit or the unit in question. */
2504 slice
= UNIT_GET_SLICE(u
);
2506 mask
|= unit_get_ancestor_disable_mask(slice
);
2511 CGroupMask
unit_get_target_mask(Unit
*u
) {
2512 CGroupMask own_mask
, mask
;
2514 /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything
2515 * it needs itself, plus all that its children need, plus all that its siblings need. This is
2516 * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each
2517 * hierarchy that shall be enabled for it. */
2519 own_mask
= unit_get_own_mask(u
);
2521 if (own_mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
2522 emit_bpf_firewall_warning(u
);
2524 mask
= own_mask
| unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
2526 mask
&= u
->manager
->cgroup_supported
;
2527 mask
&= ~unit_get_ancestor_disable_mask(u
);
2532 CGroupMask
unit_get_enable_mask(Unit
*u
) {
2535 /* This returns the cgroup mask of all controllers to enable
2536 * for the children of a specific cgroup. This is primarily
2537 * useful for the unified cgroup hierarchy, where each cgroup
2538 * controls which controllers are enabled for its children. */
2540 mask
= unit_get_members_mask(u
);
2541 mask
&= u
->manager
->cgroup_supported
;
2542 mask
&= ~unit_get_ancestor_disable_mask(u
);
2547 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
2552 /* Recurse invalidate the member masks cache all the way up the tree */
2553 u
->cgroup_members_mask_valid
= false;
2555 slice
= UNIT_GET_SLICE(u
);
2557 unit_invalidate_cgroup_members_masks(slice
);
2560 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
2562 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
2566 if (u
->cgroup_path
&&
2567 u
->cgroup_realized
&&
2568 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
2569 return u
->cgroup_path
;
2571 u
= UNIT_GET_SLICE(u
);
2577 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
2578 /* If not realized at all, migrate to root ("").
2579 * It may happen if we're upgrading from older version that didn't clean up.
2581 return strempty(unit_get_realized_cgroup_path(userdata
, mask
));
2584 int unit_default_cgroup_path(const Unit
*u
, char **ret
) {
2585 _cleanup_free_
char *p
= NULL
;
2591 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
2592 p
= strdup(u
->manager
->cgroup_root
);
2594 _cleanup_free_
char *escaped
= NULL
, *slice_path
= NULL
;
2597 slice
= UNIT_GET_SLICE(u
);
2598 if (slice
&& !unit_has_name(slice
, SPECIAL_ROOT_SLICE
)) {
2599 r
= cg_slice_to_path(slice
->id
, &slice_path
);
2604 r
= cg_escape(u
->id
, &escaped
);
2608 p
= path_join(empty_to_root(u
->manager
->cgroup_root
), slice_path
, escaped
);
2617 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
2618 _cleanup_free_
char *p
= NULL
;
2623 if (streq_ptr(u
->cgroup_path
, path
))
2633 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
2638 unit_release_cgroup(u
);
2639 u
->cgroup_path
= TAKE_PTR(p
);
2644 int unit_watch_cgroup(Unit
*u
) {
2645 _cleanup_free_
char *events
= NULL
;
2650 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
2651 * cgroupv2 is available. */
2653 if (!u
->cgroup_path
)
2656 if (u
->cgroup_control_inotify_wd
>= 0)
2659 /* Only applies to the unified hierarchy */
2660 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2662 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
2666 /* No point in watch the top-level slice, it's never going to run empty. */
2667 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
2670 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
2674 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
2678 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
2679 if (u
->cgroup_control_inotify_wd
< 0) {
2681 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
2682 * is not an error */
2685 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
));
2688 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
2690 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
));
2695 int unit_watch_cgroup_memory(Unit
*u
) {
2696 _cleanup_free_
char *events
= NULL
;
2702 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
2703 * cgroupv2 is available. */
2705 if (!u
->cgroup_path
)
2708 c
= unit_get_cgroup_context(u
);
2712 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
2713 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
2715 if (!c
->memory_accounting
)
2718 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
2719 * we also don't want to generate a log message for each parent cgroup of a process. */
2720 if (u
->type
== UNIT_SLICE
)
2723 if (u
->cgroup_memory_inotify_wd
>= 0)
2726 /* Only applies to the unified hierarchy */
2727 r
= cg_all_unified();
2729 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
2733 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
2737 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
2741 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
2742 if (u
->cgroup_memory_inotify_wd
< 0) {
2744 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
2745 * is not an error */
2748 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
));
2751 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
2753 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
));
2758 int unit_pick_cgroup_path(Unit
*u
) {
2759 _cleanup_free_
char *path
= NULL
;
2767 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2770 r
= unit_default_cgroup_path(u
, &path
);
2772 return log_unit_error_errno(u
, r
, "Failed to generate default cgroup path: %m");
2774 r
= unit_set_cgroup_path(u
, path
);
2776 return log_unit_error_errno(u
, r
, "Control group %s exists already.", empty_to_root(path
));
2778 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", empty_to_root(path
));
2783 static int unit_update_cgroup(
2785 CGroupMask target_mask
,
2786 CGroupMask enable_mask
,
2787 ManagerState state
) {
2789 bool created
, is_root_slice
;
2790 CGroupMask migrate_mask
= 0;
2791 _cleanup_free_
char *cgroup_full_path
= NULL
;
2796 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2799 /* Figure out our cgroup path */
2800 r
= unit_pick_cgroup_path(u
);
2804 /* First, create our own group */
2805 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
2807 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", empty_to_root(u
->cgroup_path
));
2810 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
2811 uint64_t cgroup_id
= 0;
2813 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, NULL
, &cgroup_full_path
);
2815 r
= cg_path_get_cgroupid(cgroup_full_path
, &cgroup_id
);
2817 log_unit_full_errno(u
, ERRNO_IS_NOT_SUPPORTED(r
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2818 "Failed to get cgroup ID of cgroup %s, ignoring: %m", cgroup_full_path
);
2820 log_unit_warning_errno(u
, r
, "Failed to get full cgroup path on cgroup %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2822 u
->cgroup_id
= cgroup_id
;
2825 /* Start watching it */
2826 (void) unit_watch_cgroup(u
);
2827 (void) unit_watch_cgroup_memory(u
);
2829 /* For v2 we preserve enabled controllers in delegated units, adjust others,
2830 * for v1 we figure out which controller hierarchies need migration. */
2831 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
2832 CGroupMask result_mask
= 0;
2834 /* Enable all controllers we need */
2835 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
2837 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2839 /* Remember what's actually enabled now */
2840 u
->cgroup_enabled_mask
= result_mask
;
2842 migrate_mask
= u
->cgroup_realized_mask
^ target_mask
;
2845 /* Keep track that this is now realized */
2846 u
->cgroup_realized
= true;
2847 u
->cgroup_realized_mask
= target_mask
;
2849 /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling).
2851 * Unnecessary controller cgroups are trimmed (after emptied by upward migration).
2852 * We perform migration also with whole slices for cases when users don't care about leave
2853 * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing
2856 if (cg_all_unified() == 0) {
2857 r
= cg_migrate_v1_controllers(u
->manager
->cgroup_supported
, migrate_mask
, u
->cgroup_path
, migrate_callback
, u
);
2859 log_unit_warning_errno(u
, r
, "Failed to migrate controller cgroups from %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2861 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2862 r
= cg_trim_v1_controllers(u
->manager
->cgroup_supported
, ~target_mask
, u
->cgroup_path
, !is_root_slice
);
2864 log_unit_warning_errno(u
, r
, "Failed to delete controller cgroups %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2867 /* Set attributes */
2868 cgroup_context_apply(u
, target_mask
, state
);
2869 cgroup_xattr_apply(u
);
2871 /* For most units we expect that memory monitoring is set up before the unit is started and we won't
2872 * touch it after. For PID 1 this is different though, because we couldn't possibly do that given
2873 * that PID 1 runs before init.scope is even set up. Hence, whenever init.scope is realized, let's
2874 * try to open the memory pressure interface anew. */
2875 if (unit_has_name(u
, SPECIAL_INIT_SCOPE
))
2876 (void) manager_setup_memory_pressure_event_source(u
->manager
);
2881 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
2882 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2888 if (MANAGER_IS_SYSTEM(u
->manager
))
2891 if (!u
->manager
->system_bus
)
2894 if (!u
->cgroup_path
)
2897 /* Determine this unit's cgroup path relative to our cgroup root */
2898 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
2902 pp
= strjoina("/", pp
, suffix_path
);
2905 r
= bus_call_method(u
->manager
->system_bus
,
2907 "AttachProcessesToUnit",
2910 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
2912 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
2917 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
2918 _cleanup_free_
char *joined
= NULL
;
2919 CGroupMask delegated_mask
;
2926 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2929 if (set_isempty(pids
))
2932 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
2933 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
2934 r
= bpf_firewall_load_custom(u
);
2938 r
= unit_realize_cgroup(u
);
2942 if (isempty(suffix_path
))
2945 joined
= path_join(u
->cgroup_path
, suffix_path
);
2952 delegated_mask
= unit_get_delegate_mask(u
);
2955 SET_FOREACH(pid
, pids
) {
2957 /* Unfortunately we cannot add pids by pidfd to a cgroup. Hence we have to use PIDs instead,
2958 * which of course is racy. Let's shorten the race a bit though, and re-validate the PID
2959 * before we use it */
2960 r
= pidref_verify(pid
);
2962 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
));
2966 /* First, attach the PID to the main cgroup hierarchy */
2967 r
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
->pid
);
2969 bool again
= MANAGER_IS_USER(u
->manager
) && ERRNO_IS_PRIVILEGE(r
);
2971 log_unit_full_errno(u
, again
? LOG_DEBUG
: LOG_INFO
, r
,
2972 "Couldn't move process "PID_FMT
" to%s requested cgroup '%s': %m",
2973 pid
->pid
, again
? " directly" : "", empty_to_root(p
));
2978 /* If we are in a user instance, and we can't move the process ourselves due
2979 * to permission problems, let's ask the system instance about it instead.
2980 * Since it's more privileged it might be able to move the process across the
2981 * leaves of a subtree whose top node is not owned by us. */
2983 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
->pid
, suffix_path
);
2985 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
));
2988 ret
++; /* Count successful additions */
2989 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
2994 ret
= r
; /* Remember first error */
2997 } else if (ret
>= 0)
2998 ret
++; /* Count successful additions */
3000 r
= cg_all_unified();
3006 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
3007 * innermost realized one */
3009 for (CGroupController c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
3010 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
3011 const char *realized
;
3013 if (!(u
->manager
->cgroup_supported
& bit
))
3016 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
3017 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
3018 r
= cg_attach(cgroup_controller_to_string(c
), p
, pid
->pid
);
3020 continue; /* Success! */
3022 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",
3023 pid
->pid
, empty_to_root(p
), cgroup_controller_to_string(c
));
3026 /* So this controller is either not delegate or realized, or something else weird happened. In
3027 * that case let's attach the PID at least to the closest cgroup up the tree that is
3029 realized
= unit_get_realized_cgroup_path(u
, bit
);
3031 continue; /* Not even realized in the root slice? Then let's not bother */
3033 r
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
->pid
);
3035 log_unit_debug_errno(u
, r
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
3036 pid
->pid
, realized
, cgroup_controller_to_string(c
));
3043 static bool unit_has_mask_realized(
3045 CGroupMask target_mask
,
3046 CGroupMask enable_mask
) {
3050 /* Returns true if this unit is fully realized. We check four things:
3052 * 1. Whether the cgroup was created at all
3053 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
3054 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
3055 * 4. Whether the invalidation mask is currently zero
3057 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
3058 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
3059 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
3060 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
3061 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
3062 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
3063 * simply don't matter. */
3065 return u
->cgroup_realized
&&
3066 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
3067 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
3068 u
->cgroup_invalidated_mask
== 0;
3071 static bool unit_has_mask_disables_realized(
3073 CGroupMask target_mask
,
3074 CGroupMask enable_mask
) {
3078 /* Returns true if all controllers which should be disabled are indeed disabled.
3080 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
3081 * already removed. */
3083 return !u
->cgroup_realized
||
3084 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
3085 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
3088 static bool unit_has_mask_enables_realized(
3090 CGroupMask target_mask
,
3091 CGroupMask enable_mask
) {
3095 /* Returns true if all controllers which should be enabled are indeed enabled.
3097 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
3098 * we want to add is already added. */
3100 return u
->cgroup_realized
&&
3101 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
3102 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
3105 void unit_add_to_cgroup_realize_queue(Unit
*u
) {
3108 if (u
->in_cgroup_realize_queue
)
3111 LIST_APPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
3112 u
->in_cgroup_realize_queue
= true;
3115 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
3118 if (!u
->in_cgroup_realize_queue
)
3121 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
3122 u
->in_cgroup_realize_queue
= false;
3125 /* Controllers can only be enabled breadth-first, from the root of the
3126 * hierarchy downwards to the unit in question. */
3127 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
3128 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
3134 /* First go deal with this unit's parent, or we won't be able to enable
3135 * any new controllers at this layer. */
3136 slice
= UNIT_GET_SLICE(u
);
3138 r
= unit_realize_cgroup_now_enable(slice
, state
);
3143 target_mask
= unit_get_target_mask(u
);
3144 enable_mask
= unit_get_enable_mask(u
);
3146 /* We can only enable in this direction, don't try to disable anything.
3148 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
3151 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
3152 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
3154 return unit_update_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
3157 /* Controllers can only be disabled depth-first, from the leaves of the
3158 * hierarchy upwards to the unit in question. */
3159 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
3164 if (u
->type
!= UNIT_SLICE
)
3167 UNIT_FOREACH_DEPENDENCY(m
, u
, UNIT_ATOM_SLICE_OF
) {
3168 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
3171 /* The cgroup for this unit might not actually be fully realised yet, in which case it isn't
3172 * holding any controllers open anyway. */
3173 if (!m
->cgroup_realized
)
3176 /* We must disable those below us first in order to release the controller. */
3177 if (m
->type
== UNIT_SLICE
)
3178 (void) unit_realize_cgroup_now_disable(m
, state
);
3180 target_mask
= unit_get_target_mask(m
);
3181 enable_mask
= unit_get_enable_mask(m
);
3183 /* We can only disable in this direction, don't try to enable anything. */
3184 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
3187 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
3188 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
3190 r
= unit_update_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
3198 /* Check if necessary controllers and attributes for a unit are in place.
3200 * - If so, do nothing.
3201 * - If not, create paths, move processes over, and set attributes.
3203 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
3204 * a depth-first way. As such the process looks like this:
3206 * Suppose we have a cgroup hierarchy which looks like this:
3219 * 1. We want to realise cgroup "d" now.
3220 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
3221 * 3. cgroup "k" just started requesting the memory controller.
3223 * To make this work we must do the following in order:
3225 * 1. Disable CPU controller in k, j
3226 * 2. Disable CPU controller in d
3227 * 3. Enable memory controller in root
3228 * 4. Enable memory controller in a
3229 * 5. Enable memory controller in d
3230 * 6. Enable memory controller in k
3232 * Notice that we need to touch j in one direction, but not the other. We also
3233 * don't go beyond d when disabling -- it's up to "a" to get realized if it
3234 * wants to disable further. The basic rules are therefore:
3236 * - If you're disabling something, you need to realise all of the cgroups from
3237 * your recursive descendants to the root. This starts from the leaves.
3238 * - If you're enabling something, you need to realise from the root cgroup
3239 * downwards, but you don't need to iterate your recursive descendants.
3241 * Returns 0 on success and < 0 on failure. */
3242 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
3243 CGroupMask target_mask
, enable_mask
;
3249 unit_remove_from_cgroup_realize_queue(u
);
3251 target_mask
= unit_get_target_mask(u
);
3252 enable_mask
= unit_get_enable_mask(u
);
3254 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
3257 /* Disable controllers below us, if there are any */
3258 r
= unit_realize_cgroup_now_disable(u
, state
);
3262 /* Enable controllers above us, if there are any */
3263 slice
= UNIT_GET_SLICE(u
);
3265 r
= unit_realize_cgroup_now_enable(slice
, state
);
3270 /* Now actually deal with the cgroup we were trying to realise and set attributes */
3271 r
= unit_update_cgroup(u
, target_mask
, enable_mask
, state
);
3275 /* Now, reset the invalidation mask */
3276 u
->cgroup_invalidated_mask
= 0;
3280 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
3288 state
= manager_state(m
);
3290 while ((i
= m
->cgroup_realize_queue
)) {
3291 assert(i
->in_cgroup_realize_queue
);
3293 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
3294 /* Maybe things changed, and the unit is not actually active anymore? */
3295 unit_remove_from_cgroup_realize_queue(i
);
3299 r
= unit_realize_cgroup_now(i
, state
);
3301 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
3309 void unit_add_family_to_cgroup_realize_queue(Unit
*u
) {
3311 assert(u
->type
== UNIT_SLICE
);
3313 /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all
3316 * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes
3317 * very weird if two units that own processes reside in the same slice, but one is realized in the
3318 * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does
3319 * not), because that means individual processes need to be scheduled against whole cgroups. Let's
3320 * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1
3321 * controller hierarchies too (if unit requires the controller to be realized).
3323 * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date
3329 /* Children of u likely changed when we're called */
3330 u
->cgroup_members_mask_valid
= false;
3332 UNIT_FOREACH_DEPENDENCY(m
, u
, UNIT_ATOM_SLICE_OF
) {
3334 /* No point in doing cgroup application for units without active processes. */
3335 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
3338 /* We only enqueue siblings if they were realized once at least, in the main
3340 if (!m
->cgroup_realized
)
3343 /* If the unit doesn't need any new controllers and has current ones
3344 * realized, it doesn't need any changes. */
3345 if (unit_has_mask_realized(m
,
3346 unit_get_target_mask(m
),
3347 unit_get_enable_mask(m
)))
3350 unit_add_to_cgroup_realize_queue(m
);
3353 /* Parent comes after children */
3354 unit_add_to_cgroup_realize_queue(u
);
3356 u
= UNIT_GET_SLICE(u
);
3360 int unit_realize_cgroup(Unit
*u
) {
3365 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3368 /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all
3369 * parents, but there's more actually: for the weight-based controllers we also need to make sure
3370 * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the
3371 * other hand, when a controller is removed from realized set, it may become unnecessary in siblings
3372 * and ancestors and they should be (de)realized too.
3374 * This call will defer work on the siblings and derealized ancestors to the next event loop
3375 * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */
3377 slice
= UNIT_GET_SLICE(u
);
3379 unit_add_family_to_cgroup_realize_queue(slice
);
3381 /* And realize this one now (and apply the values) */
3382 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
3385 void unit_release_cgroup(Unit
*u
) {
3388 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
3389 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
3391 if (u
->cgroup_path
) {
3392 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
3393 u
->cgroup_path
= mfree(u
->cgroup_path
);
3396 if (u
->cgroup_control_inotify_wd
>= 0) {
3397 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
3398 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
);
3400 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
3401 u
->cgroup_control_inotify_wd
= -1;
3404 if (u
->cgroup_memory_inotify_wd
>= 0) {
3405 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
3406 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
);
3408 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
3409 u
->cgroup_memory_inotify_wd
= -1;
3413 bool unit_maybe_release_cgroup(Unit
*u
) {
3418 if (!u
->cgroup_path
)
3421 /* Don't release the cgroup if there are still processes under it. If we get notified later when all the
3422 * processes exit (e.g. the processes were in D-state and exited after the unit was marked as failed)
3423 * we need the cgroup paths to continue to be tracked by the manager so they can be looked up and cleaned
3425 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
3427 log_unit_debug_errno(u
, r
, "Error checking if the cgroup is recursively empty, ignoring: %m");
3429 unit_release_cgroup(u
);
3436 void unit_prune_cgroup(Unit
*u
) {
3442 /* Removes the cgroup, if empty and possible, and stops watching it. */
3444 if (!u
->cgroup_path
)
3447 /* Cache the last CPU and memory usage values before we destroy the cgroup */
3448 (void) unit_get_cpu_usage(u
, /* ret = */ NULL
);
3450 for (CGroupMemoryAccountingMetric metric
= 0; metric
<= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST
; metric
++)
3451 (void) unit_get_memory_accounting(u
, metric
, /* ret = */ NULL
);
3454 (void) bpf_restrict_fs_cleanup(u
); /* Remove cgroup from the global LSM BPF map */
3457 unit_modify_nft_set(u
, /* add = */ false);
3459 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
3461 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
3463 /* One reason we could have failed here is, that the cgroup still contains a process.
3464 * However, if the cgroup becomes removable at a later time, it might be removed when
3465 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
3466 * that the cgroup is still realized the next time it is started. Do not return early
3467 * on error, continue cleanup. */
3468 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
));
3473 if (!unit_maybe_release_cgroup(u
)) /* Returns true if the cgroup was released */
3476 u
->cgroup_realized
= false;
3477 u
->cgroup_realized_mask
= 0;
3478 u
->cgroup_enabled_mask
= 0;
3480 u
->bpf_device_control_installed
= bpf_program_free(u
->bpf_device_control_installed
);
3483 int unit_search_main_pid(Unit
*u
, PidRef
*ret
) {
3484 _cleanup_(pidref_done
) PidRef pidref
= PIDREF_NULL
;
3485 _cleanup_fclose_
FILE *f
= NULL
;
3491 if (!u
->cgroup_path
)
3494 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
3499 _cleanup_(pidref_done
) PidRef npidref
= PIDREF_NULL
;
3501 r
= cg_read_pidref(f
, &npidref
);
3507 if (pidref_equal(&pidref
, &npidref
)) /* seen already, cgroupfs reports duplicates! */
3510 if (pidref_is_my_child(&npidref
) <= 0) /* ignore processes further down the tree */
3513 if (pidref_is_set(&pidref
) != 0)
3514 /* Dang, there's more than one daemonized PID in this group, so we don't know what
3515 * process is the main process. */
3518 pidref
= TAKE_PIDREF(npidref
);
3521 if (!pidref_is_set(&pidref
))
3524 *ret
= TAKE_PIDREF(pidref
);
3528 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
3529 _cleanup_closedir_
DIR *d
= NULL
;
3530 _cleanup_fclose_
FILE *f
= NULL
;
3536 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
3541 _cleanup_(pidref_done
) PidRef pid
= PIDREF_NULL
;
3543 r
= cg_read_pidref(f
, &pid
);
3551 RET_GATHER(ret
, unit_watch_pidref(u
, &pid
, /* exclusive= */ false));
3555 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
3560 _cleanup_free_
char *fn
= NULL
, *p
= NULL
;
3562 r
= cg_read_subgroup(d
, &fn
);
3570 p
= path_join(empty_to_root(path
), fn
);
3574 RET_GATHER(ret
, unit_watch_pids_in_path(u
, p
));
3581 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
3586 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
3587 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
3588 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
3590 if (!u
->cgroup_path
)
3593 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
3596 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
3599 if (!set_isempty(u
->pids
))
3602 unit_add_to_cgroup_empty_queue(u
);
3606 int unit_watch_all_pids(Unit
*u
) {
3611 /* Adds all PIDs from our cgroup to the set of PIDs we
3612 * watch. This is a fallback logic for cases where we do not
3613 * get reliable cgroup empty notifications: we try to use
3614 * SIGCHLD as replacement. */
3616 if (!u
->cgroup_path
)
3619 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
3622 if (r
> 0) /* On unified we can use proper notifications */
3625 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
3628 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
3629 Manager
*m
= ASSERT_PTR(userdata
);
3635 u
= m
->cgroup_empty_queue
;
3639 assert(u
->in_cgroup_empty_queue
);
3640 u
->in_cgroup_empty_queue
= false;
3641 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
3643 if (m
->cgroup_empty_queue
) {
3644 /* More stuff queued, let's make sure we remain enabled */
3645 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
3647 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
3650 /* Update state based on OOM kills before we notify about cgroup empty event */
3651 (void) unit_check_oom(u
);
3652 (void) unit_check_oomd_kill(u
);
3654 unit_add_to_gc_queue(u
);
3656 if (IN_SET(unit_active_state(u
), UNIT_INACTIVE
, UNIT_FAILED
))
3657 unit_prune_cgroup(u
);
3658 else if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
3659 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
3664 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
3669 /* Note that there are four different ways how cgroup empty events reach us:
3671 * 1. On the unified hierarchy we get an inotify event on the cgroup
3673 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
3675 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
3677 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
3678 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
3680 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
3681 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
3682 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
3683 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
3684 * case for scope units). */
3686 if (u
->in_cgroup_empty_queue
)
3689 /* Let's verify that the cgroup is really empty */
3690 if (!u
->cgroup_path
)
3693 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
3695 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", empty_to_root(u
->cgroup_path
));
3701 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
3702 u
->in_cgroup_empty_queue
= true;
3704 /* Trigger the defer event */
3705 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
3707 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
3710 static void unit_remove_from_cgroup_empty_queue(Unit
*u
) {
3713 if (!u
->in_cgroup_empty_queue
)
3716 LIST_REMOVE(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
3717 u
->in_cgroup_empty_queue
= false;
3720 int unit_check_oomd_kill(Unit
*u
) {
3721 _cleanup_free_
char *value
= NULL
;
3726 if (!u
->cgroup_path
)
3729 r
= cg_all_unified();
3731 return log_unit_debug_errno(u
, r
, "Couldn't determine whether we are in all unified mode: %m");
3735 r
= cg_get_xattr_malloc(u
->cgroup_path
, "user.oomd_ooms", &value
);
3736 if (r
< 0 && !ERRNO_IS_XATTR_ABSENT(r
))
3739 if (!isempty(value
)) {
3740 r
= safe_atou64(value
, &n
);
3745 increased
= n
> u
->managed_oom_kill_last
;
3746 u
->managed_oom_kill_last
= n
;
3752 value
= mfree(value
);
3753 r
= cg_get_xattr_malloc(u
->cgroup_path
, "user.oomd_kill", &value
);
3754 if (r
>= 0 && !isempty(value
))
3755 (void) safe_atou64(value
, &n
);
3758 log_unit_struct(u
, LOG_NOTICE
,
3759 "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR
,
3760 LOG_UNIT_INVOCATION_ID(u
),
3761 LOG_UNIT_MESSAGE(u
, "systemd-oomd killed %"PRIu64
" process(es) in this unit.", n
),
3762 "N_PROCESSES=%" PRIu64
, n
);
3764 log_unit_struct(u
, LOG_NOTICE
,
3765 "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR
,
3766 LOG_UNIT_INVOCATION_ID(u
),
3767 LOG_UNIT_MESSAGE(u
, "systemd-oomd killed some process(es) in this unit."));
3769 unit_notify_cgroup_oom(u
, /* ManagedOOM= */ true);
3774 int unit_check_oom(Unit
*u
) {
3775 _cleanup_free_
char *oom_kill
= NULL
;
3780 if (!u
->cgroup_path
)
3783 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
3784 if (IN_SET(r
, -ENOENT
, -ENXIO
)) /* Handle gracefully if cgroup or oom_kill attribute don't exist */
3787 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
3789 r
= safe_atou64(oom_kill
, &c
);
3791 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
3794 increased
= c
> u
->oom_kill_last
;
3795 u
->oom_kill_last
= c
;
3800 log_unit_struct(u
, LOG_NOTICE
,
3801 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
3802 LOG_UNIT_INVOCATION_ID(u
),
3803 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
3805 unit_notify_cgroup_oom(u
, /* ManagedOOM= */ false);
3810 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
3811 Manager
*m
= ASSERT_PTR(userdata
);
3817 u
= m
->cgroup_oom_queue
;
3821 assert(u
->in_cgroup_oom_queue
);
3822 u
->in_cgroup_oom_queue
= false;
3823 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
3825 if (m
->cgroup_oom_queue
) {
3826 /* More stuff queued, let's make sure we remain enabled */
3827 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
3829 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
3832 (void) unit_check_oom(u
);
3833 unit_add_to_gc_queue(u
);
3838 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
3843 if (u
->in_cgroup_oom_queue
)
3845 if (!u
->cgroup_path
)
3848 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
3849 u
->in_cgroup_oom_queue
= true;
3851 /* Trigger the defer event */
3852 if (!u
->manager
->cgroup_oom_event_source
) {
3853 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
3855 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
3857 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
3861 r
= sd_event_source_set_priority(s
, EVENT_PRIORITY_CGROUP_OOM
);
3863 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
3867 (void) sd_event_source_set_description(s
, "cgroup-oom");
3868 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
3871 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
3873 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
3876 static int unit_check_cgroup_events(Unit
*u
) {
3877 char *values
[2] = {};
3882 if (!u
->cgroup_path
)
3885 r
= cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events",
3886 STRV_MAKE("populated", "frozen"), values
);
3890 /* The cgroup.events notifications can be merged together so act as we saw the given state for the
3891 * first time. The functions we call to handle given state are idempotent, which makes them
3892 * effectively remember the previous state. */
3894 if (streq(values
[0], "1"))
3895 unit_remove_from_cgroup_empty_queue(u
);
3897 unit_add_to_cgroup_empty_queue(u
);
3900 /* Disregard freezer state changes due to operations not initiated by us */
3901 if (values
[1] && IN_SET(u
->freezer_state
, FREEZER_FREEZING
, FREEZER_THAWING
)) {
3902 if (streq(values
[1], "0"))
3914 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
3915 Manager
*m
= ASSERT_PTR(userdata
);
3921 union inotify_event_buffer buffer
;
3924 l
= read(fd
, &buffer
, sizeof(buffer
));
3926 if (ERRNO_IS_TRANSIENT(errno
))
3929 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
3932 FOREACH_INOTIFY_EVENT_WARN(e
, buffer
, l
) {
3936 /* Queue overflow has no watch descriptor */
3939 if (e
->mask
& IN_IGNORED
)
3940 /* The watch was just removed */
3943 /* Note that inotify might deliver events for a watch even after it was removed,
3944 * because it was queued before the removal. Let's ignore this here safely. */
3946 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
3948 unit_check_cgroup_events(u
);
3950 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
3952 unit_add_to_cgroup_oom_queue(u
);
3957 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
3958 CGroupMask mask
= 0;
3961 /* BPF-based firewall */
3962 r
= bpf_firewall_supported();
3966 mask
|= CGROUP_MASK_BPF_FIREWALL
;
3968 /* BPF-based device access control */
3969 r
= bpf_devices_supported();
3973 mask
|= CGROUP_MASK_BPF_DEVICES
;
3975 /* BPF pinned prog */
3976 r
= bpf_foreign_supported();
3980 mask
|= CGROUP_MASK_BPF_FOREIGN
;
3982 /* BPF-based bind{4|6} hooks */
3983 r
= bpf_socket_bind_supported();
3987 mask
|= CGROUP_MASK_BPF_SOCKET_BIND
;
3989 /* BPF-based cgroup_skb/{egress|ingress} hooks */
3990 r
= bpf_restrict_ifaces_supported();
3994 mask
|= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
;
4000 int manager_setup_cgroup(Manager
*m
) {
4001 _cleanup_free_
char *path
= NULL
;
4002 const char *scope_path
;
4009 /* 1. Determine hierarchy */
4010 m
->cgroup_root
= mfree(m
->cgroup_root
);
4011 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
4013 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
4015 /* Chop off the init scope, if we are already located in it */
4016 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
4018 /* LEGACY: Also chop off the system slice if we are in
4019 * it. This is to support live upgrades from older systemd
4020 * versions where PID 1 was moved there. Also see
4021 * cg_get_root_path(). */
4022 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
4023 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
4025 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
4030 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
4031 * easily prepend it everywhere. */
4032 delete_trailing_chars(m
->cgroup_root
, "/");
4035 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
4037 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
4041 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
4043 all_unified
= cg_all_unified();
4044 if (all_unified
< 0)
4045 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
4046 if (all_unified
> 0)
4047 log_debug("Unified cgroup hierarchy is located at %s.", path
);
4049 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
4051 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
4053 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
4055 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
4058 /* 3. Allocate cgroup empty defer event source */
4059 m
->cgroup_empty_event_source
= sd_event_source_disable_unref(m
->cgroup_empty_event_source
);
4060 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
4062 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
4064 /* Schedule cgroup empty checks early, but after having processed service notification messages or
4065 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
4066 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
4067 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, EVENT_PRIORITY_CGROUP_EMPTY
);
4069 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
4071 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
4073 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
4075 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
4077 /* 4. Install notifier inotify object, or agent */
4078 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
4080 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
4082 m
->cgroup_inotify_event_source
= sd_event_source_disable_unref(m
->cgroup_inotify_event_source
);
4083 safe_close(m
->cgroup_inotify_fd
);
4085 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
4086 if (m
->cgroup_inotify_fd
< 0)
4087 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
4089 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
4091 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
4093 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
4094 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
4095 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
4096 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, EVENT_PRIORITY_CGROUP_INOTIFY
);
4098 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
4100 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
4102 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
4104 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
4105 * since it does not generate events when control groups with children run empty. */
4107 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUPS_AGENT_PATH
);
4109 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
4111 log_debug("Installed release agent.");
4113 log_debug("Release agent already installed.");
4116 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
4117 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
4118 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
4120 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
4121 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
4123 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
4125 /* 6. And pin it, so that it cannot be unmounted */
4126 safe_close(m
->pin_cgroupfs_fd
);
4127 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
4128 if (m
->pin_cgroupfs_fd
< 0)
4129 return log_error_errno(errno
, "Failed to open pin file: %m");
4131 } else if (!MANAGER_IS_TEST_RUN(m
))
4132 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
4134 /* 7. Always enable hierarchical support if it exists... */
4135 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
4136 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
4138 /* 8. Figure out which controllers are supported */
4139 r
= cg_mask_supported_subtree(m
->cgroup_root
, &m
->cgroup_supported
);
4141 return log_error_errno(r
, "Failed to determine supported controllers: %m");
4143 /* 9. Figure out which bpf-based pseudo-controllers are supported */
4144 r
= cg_bpf_mask_supported(&mask
);
4146 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
4147 m
->cgroup_supported
|= mask
;
4149 /* 10. Log which controllers are supported */
4150 for (CGroupController c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
4151 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
),
4152 yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
4157 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
4160 /* We can't really delete the group, since we are in it. But
4162 if (delete && m
->cgroup_root
&& !FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
4163 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
4165 m
->cgroup_empty_event_source
= sd_event_source_disable_unref(m
->cgroup_empty_event_source
);
4167 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
4168 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
4170 m
->cgroup_inotify_event_source
= sd_event_source_disable_unref(m
->cgroup_inotify_event_source
);
4171 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
4173 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
4175 m
->cgroup_root
= mfree(m
->cgroup_root
);
4178 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
4185 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
4189 p
= strdupa_safe(cgroup
);
4193 e
= strrchr(p
, '/');
4195 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
4199 u
= hashmap_get(m
->cgroup_unit
, p
);
4205 Unit
*manager_get_unit_by_pidref_cgroup(Manager
*m
, const PidRef
*pid
) {
4206 _cleanup_free_
char *cgroup
= NULL
;
4210 if (cg_pidref_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
4213 return manager_get_unit_by_cgroup(m
, cgroup
);
4216 Unit
*manager_get_unit_by_pidref_watching(Manager
*m
, const PidRef
*pid
) {
4221 if (!pidref_is_set(pid
))
4224 u
= hashmap_get(m
->watch_pids
, pid
);
4228 array
= hashmap_get(m
->watch_pids_more
, pid
);
4235 Unit
*manager_get_unit_by_pidref(Manager
*m
, const PidRef
*pid
) {
4240 /* Note that a process might be owned by multiple units, we return only one here, which is good
4241 * enough for most cases, though not strictly correct. We prefer the one reported by cgroup
4242 * membership, as that's the most relevant one as children of the process will be assigned to that
4243 * one, too, before all else. */
4245 if (!pidref_is_set(pid
))
4248 if (pidref_is_self(pid
))
4249 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
4253 u
= manager_get_unit_by_pidref_cgroup(m
, pid
);
4257 u
= manager_get_unit_by_pidref_watching(m
, pid
);
4264 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
4267 if (!pid_is_valid(pid
))
4270 return manager_get_unit_by_pidref(m
, &PIDREF_MAKE_FROM_PID(pid
));
4273 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
4279 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
4280 * or from the --system instance */
4282 log_debug("Got cgroup empty notification for: %s", cgroup
);
4284 u
= manager_get_unit_by_cgroup(m
, cgroup
);
4288 unit_add_to_cgroup_empty_queue(u
);
4292 int unit_get_memory_available(Unit
*u
, uint64_t *ret
) {
4293 uint64_t available
= UINT64_MAX
, current
= 0;
4298 /* If data from cgroups can be accessed, try to find out how much more memory a unit can
4299 * claim before hitting the configured cgroup limits (if any). Consider both MemoryHigh
4300 * and MemoryMax, and also any slice the unit might be nested below. */
4303 uint64_t unit_available
, unit_limit
= UINT64_MAX
;
4304 CGroupContext
*unit_context
;
4306 /* No point in continuing if we can't go any lower */
4310 unit_context
= unit_get_cgroup_context(u
);
4314 if (!u
->cgroup_path
)
4317 (void) unit_get_memory_current(u
, ¤t
);
4318 /* in case of error, previous current propagates as lower bound */
4320 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
4321 unit_limit
= physical_memory();
4322 else if (unit_context
->memory_max
== UINT64_MAX
&& unit_context
->memory_high
== UINT64_MAX
)
4324 unit_limit
= MIN3(unit_limit
, unit_context
->memory_max
, unit_context
->memory_high
);
4326 unit_available
= LESS_BY(unit_limit
, current
);
4327 available
= MIN(unit_available
, available
);
4328 } while ((u
= UNIT_GET_SLICE(u
)));
4335 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
4338 // FIXME: Merge this into unit_get_memory_accounting after support for cgroup v1 is dropped
4343 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
4346 if (!u
->cgroup_path
)
4349 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
4350 if (unit_has_host_root_cgroup(u
))
4351 return procfs_memory_get_used(ret
);
4353 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
4356 r
= cg_all_unified();
4360 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, r
> 0 ? "memory.current" : "memory.usage_in_bytes", ret
);
4363 int unit_get_memory_accounting(Unit
*u
, CGroupMemoryAccountingMetric metric
, uint64_t *ret
) {
4365 static const char* const attributes_table
[_CGROUP_MEMORY_ACCOUNTING_METRIC_MAX
] = {
4366 [CGROUP_MEMORY_PEAK
] = "memory.peak",
4367 [CGROUP_MEMORY_SWAP_CURRENT
] = "memory.swap.current",
4368 [CGROUP_MEMORY_SWAP_PEAK
] = "memory.swap.peak",
4369 [CGROUP_MEMORY_ZSWAP_CURRENT
] = "memory.zswap.current",
4373 bool updated
= false;
4377 assert(metric
>= 0);
4378 assert(metric
< _CGROUP_MEMORY_ACCOUNTING_METRIC_MAX
);
4380 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
4383 if (!u
->cgroup_path
)
4384 /* If the cgroup is already gone, we try to find the last cached value. */
4387 /* The root cgroup doesn't expose this information. */
4388 if (unit_has_host_root_cgroup(u
))
4391 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_MEMORY
))
4394 r
= cg_all_unified();
4400 r
= cg_get_attribute_as_uint64("memory", u
->cgroup_path
, attributes_table
[metric
], &bytes
);
4401 if (r
< 0 && r
!= -ENODATA
)
4406 if (metric
<= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST
) {
4407 uint64_t *last
= &u
->memory_accounting_last
[metric
];
4411 else if (*last
!= UINT64_MAX
)
4416 } else if (!updated
)
4425 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
4429 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
4432 if (!u
->cgroup_path
)
4435 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
4436 if (unit_has_host_root_cgroup(u
))
4437 return procfs_tasks_get_current(ret
);
4439 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
4442 return cg_get_attribute_as_uint64("pids", u
->cgroup_path
, "pids.current", ret
);
4445 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
4452 if (!u
->cgroup_path
)
4455 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
4456 if (unit_has_host_root_cgroup(u
))
4457 return procfs_cpu_get_usage(ret
);
4459 /* Requisite controllers for CPU accounting are not enabled */
4460 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
4463 r
= cg_all_unified();
4467 _cleanup_free_
char *val
= NULL
;
4470 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
4471 if (IN_SET(r
, -ENOENT
, -ENXIO
))
4476 r
= safe_atou64(val
, &us
);
4480 ns
= us
* NSEC_PER_USEC
;
4482 return cg_get_attribute_as_uint64("cpuacct", u
->cgroup_path
, "cpuacct.usage", ret
);
4488 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
4494 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
4495 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
4496 * call this function with a NULL return value. */
4498 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
4501 r
= unit_get_cpu_usage_raw(u
, &ns
);
4502 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
4503 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
4507 *ret
= u
->cpu_usage_last
;
4513 if (ns
> u
->cpu_usage_base
)
4514 ns
-= u
->cpu_usage_base
;
4518 u
->cpu_usage_last
= ns
;
4525 int unit_get_ip_accounting(
4527 CGroupIPAccountingMetric metric
,
4534 assert(metric
>= 0);
4535 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
4538 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
4541 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
4542 u
->ip_accounting_ingress_map_fd
:
4543 u
->ip_accounting_egress_map_fd
;
4547 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
4548 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
4550 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
4554 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
4555 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
4556 * ip_accounting_extra[] field, and add them in here transparently. */
4558 *ret
= value
+ u
->ip_accounting_extra
[metric
];
4563 static uint64_t unit_get_effective_limit_one(Unit
*u
, CGroupLimitType type
) {
4567 assert(UNIT_HAS_CGROUP_CONTEXT(u
));
4569 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
4571 case CGROUP_LIMIT_MEMORY_MAX
:
4572 case CGROUP_LIMIT_MEMORY_HIGH
:
4573 return physical_memory();
4574 case CGROUP_LIMIT_TASKS_MAX
:
4575 return system_tasks_max();
4577 assert_not_reached();
4580 cc
= ASSERT_PTR(unit_get_cgroup_context(u
));
4582 /* Note: on legacy/hybrid hierarchies memory_max stays CGROUP_LIMIT_MAX unless configured
4583 * explicitly. Effective value of MemoryLimit= (cgroup v1) is not implemented. */
4584 case CGROUP_LIMIT_MEMORY_MAX
:
4585 return cc
->memory_max
;
4586 case CGROUP_LIMIT_MEMORY_HIGH
:
4587 return cc
->memory_high
;
4588 case CGROUP_LIMIT_TASKS_MAX
:
4589 return cgroup_tasks_max_resolve(&cc
->tasks_max
);
4591 assert_not_reached();
4595 int unit_get_effective_limit(Unit
*u
, CGroupLimitType type
, uint64_t *ret
) {
4601 assert(type
< _CGROUP_LIMIT_TYPE_MAX
);
4603 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4606 infimum
= unit_get_effective_limit_one(u
, type
);
4607 for (Unit
*slice
= UNIT_GET_SLICE(u
); slice
; slice
= UNIT_GET_SLICE(slice
))
4608 infimum
= MIN(infimum
, unit_get_effective_limit_one(slice
, type
));
4614 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
4615 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
4616 [CGROUP_IO_READ_BYTES
] = "rbytes=",
4617 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
4618 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
4619 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
4621 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
4622 _cleanup_free_
char *path
= NULL
;
4623 _cleanup_fclose_
FILE *f
= NULL
;
4628 if (!u
->cgroup_path
)
4631 if (unit_has_host_root_cgroup(u
))
4632 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
4634 r
= cg_all_unified();
4637 if (r
== 0) /* TODO: support cgroupv1 */
4640 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
4643 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
4647 f
= fopen(path
, "re");
4652 _cleanup_free_
char *line
= NULL
;
4655 r
= read_line(f
, LONG_LINE_MAX
, &line
);
4662 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
4663 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
4666 _cleanup_free_
char *word
= NULL
;
4668 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
4674 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
4677 x
= startswith(word
, field_names
[i
]);
4681 r
= safe_atou64(x
, &w
);
4685 /* Sum up the stats of all devices */
4693 memcpy(ret
, acc
, sizeof(acc
));
4697 int unit_get_io_accounting(
4699 CGroupIOAccountingMetric metric
,
4703 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
4706 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
4708 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
4711 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
4714 r
= unit_get_io_accounting_raw(u
, raw
);
4715 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
4720 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
4721 /* Saturated subtraction */
4722 if (raw
[i
] > u
->io_accounting_base
[i
])
4723 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
4725 u
->io_accounting_last
[i
] = 0;
4730 *ret
= u
->io_accounting_last
[metric
];
4735 int unit_reset_cpu_accounting(Unit
*u
) {
4740 u
->cpu_usage_last
= NSEC_INFINITY
;
4742 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
4744 u
->cpu_usage_base
= 0;
4751 void unit_reset_memory_accounting_last(Unit
*u
) {
4754 FOREACH_ARRAY(i
, u
->memory_accounting_last
, ELEMENTSOF(u
->memory_accounting_last
))
4758 int unit_reset_ip_accounting(Unit
*u
) {
4763 if (u
->ip_accounting_ingress_map_fd
>= 0)
4764 RET_GATHER(r
, bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
));
4766 if (u
->ip_accounting_egress_map_fd
>= 0)
4767 RET_GATHER(r
, bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
));
4769 zero(u
->ip_accounting_extra
);
4774 void unit_reset_io_accounting_last(Unit
*u
) {
4777 FOREACH_ARRAY(i
, u
->io_accounting_last
, _CGROUP_IO_ACCOUNTING_METRIC_MAX
)
4781 int unit_reset_io_accounting(Unit
*u
) {
4786 unit_reset_io_accounting_last(u
);
4788 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
4790 zero(u
->io_accounting_base
);
4797 int unit_reset_accounting(Unit
*u
) {
4802 RET_GATHER(r
, unit_reset_cpu_accounting(u
));
4803 RET_GATHER(r
, unit_reset_io_accounting(u
));
4804 RET_GATHER(r
, unit_reset_ip_accounting(u
));
4805 unit_reset_memory_accounting_last(u
);
4810 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
4813 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4819 /* always invalidate compat pairs together */
4820 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
4821 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
4823 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
4824 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
4826 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
4829 u
->cgroup_invalidated_mask
|= m
;
4830 unit_add_to_cgroup_realize_queue(u
);
4833 void unit_invalidate_cgroup_bpf(Unit
*u
) {
4836 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4839 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
4842 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
4843 unit_add_to_cgroup_realize_queue(u
);
4845 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
4846 * list of our children includes our own. */
4847 if (u
->type
== UNIT_SLICE
) {
4850 UNIT_FOREACH_DEPENDENCY(member
, u
, UNIT_ATOM_SLICE_OF
)
4851 unit_invalidate_cgroup_bpf(member
);
4855 void unit_cgroup_catchup(Unit
*u
) {
4858 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4861 /* We dropped the inotify watch during reexec/reload, so we need to
4862 * check these as they may have changed.
4863 * Note that (currently) the kernel doesn't actually update cgroup
4864 * file modification times, so we can't just serialize and then check
4865 * the mtime for file(s) we are interested in. */
4866 (void) unit_check_cgroup_events(u
);
4867 unit_add_to_cgroup_oom_queue(u
);
4870 bool unit_cgroup_delegate(Unit
*u
) {
4875 if (!UNIT_VTABLE(u
)->can_delegate
)
4878 c
= unit_get_cgroup_context(u
);
4885 void manager_invalidate_startup_units(Manager
*m
) {
4890 SET_FOREACH(u
, m
->startup_units
)
4891 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
|CGROUP_MASK_CPUSET
);
4894 int unit_cgroup_freezer_action(Unit
*u
, FreezerAction action
) {
4895 _cleanup_free_
char *path
= NULL
;
4896 FreezerState target
, kernel
= _FREEZER_STATE_INVALID
;
4900 assert(IN_SET(action
, FREEZER_FREEZE
, FREEZER_THAW
));
4902 if (!cg_freezer_supported())
4905 /* Ignore all requests to thaw init.scope or -.slice and reject all requests to freeze them */
4906 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
) || unit_has_name(u
, SPECIAL_INIT_SCOPE
))
4907 return action
== FREEZER_FREEZE
? -EPERM
: 0;
4909 if (!u
->cgroup_realized
)
4912 if (action
== FREEZER_THAW
) {
4913 Unit
*slice
= UNIT_GET_SLICE(u
);
4916 r
= unit_cgroup_freezer_action(slice
, FREEZER_THAW
);
4918 return log_unit_error_errno(u
, r
, "Failed to thaw slice %s of unit: %m", slice
->id
);
4922 target
= action
== FREEZER_FREEZE
? FREEZER_FROZEN
: FREEZER_RUNNING
;
4924 r
= unit_freezer_state_kernel(u
, &kernel
);
4926 log_unit_debug_errno(u
, r
, "Failed to obtain cgroup freezer state: %m");
4928 if (target
== kernel
) {
4929 u
->freezer_state
= target
;
4930 if (action
== FREEZER_FREEZE
)
4936 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.freeze", &path
);
4940 log_unit_debug(u
, "%s unit.", action
== FREEZER_FREEZE
? "Freezing" : "Thawing");
4942 if (target
!= kernel
) {
4943 if (action
== FREEZER_FREEZE
)
4944 u
->freezer_state
= FREEZER_FREEZING
;
4946 u
->freezer_state
= FREEZER_THAWING
;
4949 r
= write_string_file(path
, one_zero(action
== FREEZER_FREEZE
), WRITE_STRING_FILE_DISABLE_BUFFER
);
4956 int unit_get_cpuset(Unit
*u
, CPUSet
*cpus
, const char *name
) {
4957 _cleanup_free_
char *v
= NULL
;
4963 if (!u
->cgroup_path
)
4966 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUSET
) == 0)
4969 r
= cg_all_unified();
4975 r
= cg_get_attribute("cpuset", u
->cgroup_path
, name
, &v
);
4981 return parse_cpu_set_full(v
, cpus
, false, NULL
, NULL
, 0, NULL
);
4984 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
4985 [CGROUP_DEVICE_POLICY_AUTO
] = "auto",
4986 [CGROUP_DEVICE_POLICY_CLOSED
] = "closed",
4987 [CGROUP_DEVICE_POLICY_STRICT
] = "strict",
4990 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);
4992 static const char* const freezer_action_table
[_FREEZER_ACTION_MAX
] = {
4993 [FREEZER_FREEZE
] = "freeze",
4994 [FREEZER_THAW
] = "thaw",
4997 DEFINE_STRING_TABLE_LOOKUP(freezer_action
, FreezerAction
);
4999 static const char* const cgroup_pressure_watch_table
[_CGROUP_PRESSURE_WATCH_MAX
] = {
5000 [CGROUP_PRESSURE_WATCH_OFF
] = "off",
5001 [CGROUP_PRESSURE_WATCH_AUTO
] = "auto",
5002 [CGROUP_PRESSURE_WATCH_ON
] = "on",
5003 [CGROUP_PRESSURE_WATCH_SKIP
] = "skip",
5006 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(cgroup_pressure_watch
, CGroupPressureWatch
, CGROUP_PRESSURE_WATCH_ON
);
5008 static const char* const cgroup_ip_accounting_metric_table
[_CGROUP_IP_ACCOUNTING_METRIC_MAX
] = {
5009 [CGROUP_IP_INGRESS_BYTES
] = "IPIngressBytes",
5010 [CGROUP_IP_EGRESS_BYTES
] = "IPEgressBytes",
5011 [CGROUP_IP_INGRESS_PACKETS
] = "IPIngressPackets",
5012 [CGROUP_IP_EGRESS_PACKETS
] = "IPEgressPackets",
5015 DEFINE_STRING_TABLE_LOOKUP(cgroup_ip_accounting_metric
, CGroupIPAccountingMetric
);
5017 static const char* const cgroup_io_accounting_metric_table
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
5018 [CGROUP_IO_READ_BYTES
] = "IOReadBytes",
5019 [CGROUP_IO_WRITE_BYTES
] = "IOWriteBytes",
5020 [CGROUP_IO_READ_OPERATIONS
] = "IOReadOperations",
5021 [CGROUP_IO_WRITE_OPERATIONS
] = "IOWriteOperations",
5024 DEFINE_STRING_TABLE_LOOKUP(cgroup_io_accounting_metric
, CGroupIOAccountingMetric
);
5026 static const char* const cgroup_memory_accounting_metric_table
[_CGROUP_MEMORY_ACCOUNTING_METRIC_MAX
] = {
5027 [CGROUP_MEMORY_PEAK
] = "MemoryPeak",
5028 [CGROUP_MEMORY_SWAP_CURRENT
] = "MemorySwapCurrent",
5029 [CGROUP_MEMORY_SWAP_PEAK
] = "MemorySwapPeak",
5030 [CGROUP_MEMORY_ZSWAP_CURRENT
] = "MemoryZSwapCurrent",
5033 DEFINE_STRING_TABLE_LOOKUP(cgroup_memory_accounting_metric
, CGroupMemoryAccountingMetric
);
5035 static const char *const cgroup_effective_limit_type_table
[_CGROUP_LIMIT_TYPE_MAX
] = {
5036 [CGROUP_LIMIT_MEMORY_MAX
] = "EffectiveMemoryMax",
5037 [CGROUP_LIMIT_MEMORY_HIGH
] = "EffectiveMemoryHigh",
5038 [CGROUP_LIMIT_TASKS_MAX
] = "EffectiveTasksMax",
5041 DEFINE_STRING_TABLE_LOOKUP(cgroup_effective_limit_type
, CGroupLimitType
);