1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
5 #include "sd-messages.h"
8 #include "alloc-util.h"
9 #include "blockdev-util.h"
10 #include "bpf-devices.h"
11 #include "bpf-firewall.h"
12 #include "bpf-foreign.h"
13 #include "bpf-socket-bind.h"
14 #include "btrfs-util.h"
15 #include "bus-error.h"
16 #include "cgroup-setup.h"
17 #include "cgroup-util.h"
21 #include "in-addr-prefix-util.h"
22 #include "inotify-util.h"
24 #include "ip-protocol-list.h"
25 #include "limits-util.h"
26 #include "nulstr-util.h"
27 #include "parse-util.h"
28 #include "path-util.h"
29 #include "percent-util.h"
30 #include "process-util.h"
31 #include "procfs-util.h"
32 #include "restrict-ifaces.h"
34 #include "stat-util.h"
35 #include "stdio-util.h"
36 #include "string-table.h"
37 #include "string-util.h"
41 #include "bpf-dlopen.h"
43 #include "bpf/restrict_fs/restrict-fs-skel.h"
46 #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
48 /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access
49 * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask
50 * out specific attributes from us. */
51 #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING)
53 uint64_t tasks_max_resolve(const TasksMax
*tasks_max
) {
54 if (tasks_max
->scale
== 0)
55 return tasks_max
->value
;
57 return system_tasks_max_scale(tasks_max
->value
, tasks_max
->scale
);
60 bool manager_owns_host_root_cgroup(Manager
*m
) {
63 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
64 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
65 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
66 * we run in any kind of container virtualization. */
68 if (MANAGER_IS_USER(m
))
71 if (detect_container() > 0)
74 return empty_or_root(m
->cgroup_root
);
77 bool unit_has_startup_cgroup_constraints(Unit
*u
) {
80 /* Returns true if this unit has any directives which apply during
81 * startup/shutdown phases. */
85 c
= unit_get_cgroup_context(u
);
89 return c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
90 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
91 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
92 c
->startup_cpuset_cpus
.set
||
93 c
->startup_cpuset_mems
.set
;
96 bool unit_has_host_root_cgroup(Unit
*u
) {
99 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
100 * the manager manages the root cgroup. */
102 if (!manager_owns_host_root_cgroup(u
->manager
))
105 return unit_has_name(u
, SPECIAL_ROOT_SLICE
);
108 static int set_attribute_and_warn(Unit
*u
, const char *controller
, const char *attribute
, const char *value
) {
111 r
= cg_set_attribute(controller
, u
->cgroup_path
, attribute
, value
);
113 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%.*s': %m",
114 strna(attribute
), empty_to_root(u
->cgroup_path
), (int) strcspn(value
, NEWLINE
), value
);
119 static void cgroup_compat_warn(void) {
120 static bool cgroup_compat_warned
= false;
122 if (cgroup_compat_warned
)
125 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
126 "See cgroup-compat debug messages for details.");
128 cgroup_compat_warned
= true;
131 #define log_cgroup_compat(unit, fmt, ...) do { \
132 cgroup_compat_warn(); \
133 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
136 void cgroup_context_init(CGroupContext
*c
) {
139 /* Initialize everything to the kernel defaults. */
141 *c
= (CGroupContext
) {
142 .cpu_weight
= CGROUP_WEIGHT_INVALID
,
143 .startup_cpu_weight
= CGROUP_WEIGHT_INVALID
,
144 .cpu_quota_per_sec_usec
= USEC_INFINITY
,
145 .cpu_quota_period_usec
= USEC_INFINITY
,
147 .cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
148 .startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
150 .memory_high
= CGROUP_LIMIT_MAX
,
151 .memory_max
= CGROUP_LIMIT_MAX
,
152 .memory_swap_max
= CGROUP_LIMIT_MAX
,
154 .memory_limit
= CGROUP_LIMIT_MAX
,
156 .io_weight
= CGROUP_WEIGHT_INVALID
,
157 .startup_io_weight
= CGROUP_WEIGHT_INVALID
,
159 .blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
160 .startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
162 .tasks_max
= TASKS_MAX_UNSET
,
164 .moom_swap
= MANAGED_OOM_AUTO
,
165 .moom_mem_pressure
= MANAGED_OOM_AUTO
,
166 .moom_preference
= MANAGED_OOM_PREFERENCE_NONE
,
170 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
174 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
179 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
183 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
188 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
192 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
197 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
201 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
206 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
210 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
215 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
219 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
224 void cgroup_context_remove_bpf_foreign_program(CGroupContext
*c
, CGroupBPFForeignProgram
*p
) {
228 LIST_REMOVE(programs
, c
->bpf_foreign_programs
, p
);
233 void cgroup_context_remove_socket_bind(CGroupSocketBindItem
**head
) {
237 CGroupSocketBindItem
*h
= *head
;
238 LIST_REMOVE(socket_bind_items
, *head
, h
);
243 void cgroup_context_done(CGroupContext
*c
) {
246 while (c
->io_device_weights
)
247 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
249 while (c
->io_device_latencies
)
250 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
252 while (c
->io_device_limits
)
253 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
255 while (c
->blockio_device_weights
)
256 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
258 while (c
->blockio_device_bandwidths
)
259 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
261 while (c
->device_allow
)
262 cgroup_context_free_device_allow(c
, c
->device_allow
);
264 cgroup_context_remove_socket_bind(&c
->socket_bind_allow
);
265 cgroup_context_remove_socket_bind(&c
->socket_bind_deny
);
267 c
->ip_address_allow
= set_free(c
->ip_address_allow
);
268 c
->ip_address_deny
= set_free(c
->ip_address_deny
);
270 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
271 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
273 while (c
->bpf_foreign_programs
)
274 cgroup_context_remove_bpf_foreign_program(c
, c
->bpf_foreign_programs
);
276 c
->restrict_network_interfaces
= set_free(c
->restrict_network_interfaces
);
278 cpu_set_reset(&c
->cpuset_cpus
);
279 cpu_set_reset(&c
->startup_cpuset_cpus
);
280 cpu_set_reset(&c
->cpuset_mems
);
281 cpu_set_reset(&c
->startup_cpuset_mems
);
284 static int unit_get_kernel_memory_limit(Unit
*u
, const char *file
, uint64_t *ret
) {
287 if (!u
->cgroup_realized
)
290 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, file
, ret
);
293 static int unit_compare_memory_limit(Unit
*u
, const char *property_name
, uint64_t *ret_unit_value
, uint64_t *ret_kernel_value
) {
300 /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will
301 * return -ENODATA) on cgroup v1.
306 * 0: If the kernel memory setting doesn't match our configuration.
307 * >0: If the kernel memory setting matches our configuration.
309 * The following values are only guaranteed to be populated on return >=0:
311 * - ret_unit_value will contain our internal expected value for the unit, page-aligned.
312 * - ret_kernel_value will contain the actual value presented by the kernel. */
316 r
= cg_all_unified();
318 return log_debug_errno(r
, "Failed to determine cgroup hierarchy version: %m");
320 /* Unsupported on v1.
322 * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has
323 * silently masked the controller. */
327 /* The root slice doesn't have any controller files, so we can't compare anything. */
328 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
331 /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled,
332 * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To
333 * avoid specious errors in these scenarios, check that we even expect the memory controller to be
335 m
= unit_get_target_mask(u
);
336 if (!FLAGS_SET(m
, CGROUP_MASK_MEMORY
))
339 assert_se(c
= unit_get_cgroup_context(u
));
341 if (streq(property_name
, "MemoryLow")) {
342 unit_value
= unit_get_ancestor_memory_low(u
);
344 } else if (streq(property_name
, "MemoryMin")) {
345 unit_value
= unit_get_ancestor_memory_min(u
);
347 } else if (streq(property_name
, "MemoryHigh")) {
348 unit_value
= c
->memory_high
;
349 file
= "memory.high";
350 } else if (streq(property_name
, "MemoryMax")) {
351 unit_value
= c
->memory_max
;
353 } else if (streq(property_name
, "MemorySwapMax")) {
354 unit_value
= c
->memory_swap_max
;
355 file
= "memory.swap.max";
359 r
= unit_get_kernel_memory_limit(u
, file
, ret_kernel_value
);
361 return log_unit_debug_errno(u
, r
, "Failed to parse %s: %m", file
);
363 /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page
364 * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from
365 * our internal page-counting. To support those future kernels, just check the value itself first
366 * without any page-alignment. */
367 if (*ret_kernel_value
== unit_value
) {
368 *ret_unit_value
= unit_value
;
372 /* The current kernel behaviour, by comparison, is that even if you write a particular number of
373 * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel
374 * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is
376 if (unit_value
!= CGROUP_LIMIT_MAX
)
377 unit_value
= PAGE_ALIGN_DOWN(unit_value
);
379 *ret_unit_value
= unit_value
;
381 return *ret_kernel_value
== *ret_unit_value
;
384 #define FORMAT_CGROUP_DIFF_MAX 128
386 static char *format_cgroup_memory_limit_comparison(char *buf
, size_t l
, Unit
*u
, const char *property_name
) {
394 r
= unit_compare_memory_limit(u
, property_name
, &sval
, &kval
);
396 /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1).
397 * In the absence of reliably being able to detect whether memcg swap support is available or not,
398 * only complain if the error is not ENOENT. */
399 if (r
> 0 || IN_SET(r
, -ENODATA
, -EOWNERDEAD
) ||
400 (r
== -ENOENT
&& streq(property_name
, "MemorySwapMax"))) {
406 (void) snprintf(buf
, l
, " (error getting kernel value: %s)", strerror_safe(r
));
410 (void) snprintf(buf
, l
, " (different value in kernel: %" PRIu64
")", kval
);
415 void cgroup_context_dump(Unit
*u
, FILE* f
, const char *prefix
) {
416 _cleanup_free_
char *disable_controllers_str
= NULL
, *cpuset_cpus
= NULL
, *cpuset_mems
= NULL
, *startup_cpuset_cpus
= NULL
, *startup_cpuset_mems
= NULL
;
418 struct in_addr_prefix
*iaai
;
420 char cda
[FORMAT_CGROUP_DIFF_MAX
];
421 char cdb
[FORMAT_CGROUP_DIFF_MAX
];
422 char cdc
[FORMAT_CGROUP_DIFF_MAX
];
423 char cdd
[FORMAT_CGROUP_DIFF_MAX
];
424 char cde
[FORMAT_CGROUP_DIFF_MAX
];
429 assert_se(c
= unit_get_cgroup_context(u
));
431 prefix
= strempty(prefix
);
433 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
435 cpuset_cpus
= cpu_set_to_range_string(&c
->cpuset_cpus
);
436 startup_cpuset_cpus
= cpu_set_to_range_string(&c
->startup_cpuset_cpus
);
437 cpuset_mems
= cpu_set_to_range_string(&c
->cpuset_mems
);
438 startup_cpuset_mems
= cpu_set_to_range_string(&c
->startup_cpuset_mems
);
441 "%sCPUAccounting: %s\n"
442 "%sIOAccounting: %s\n"
443 "%sBlockIOAccounting: %s\n"
444 "%sMemoryAccounting: %s\n"
445 "%sTasksAccounting: %s\n"
446 "%sIPAccounting: %s\n"
447 "%sCPUWeight: %" PRIu64
"\n"
448 "%sStartupCPUWeight: %" PRIu64
"\n"
449 "%sCPUShares: %" PRIu64
"\n"
450 "%sStartupCPUShares: %" PRIu64
"\n"
451 "%sCPUQuotaPerSecSec: %s\n"
452 "%sCPUQuotaPeriodSec: %s\n"
453 "%sAllowedCPUs: %s\n"
454 "%sStartupAllowedCPUs: %s\n"
455 "%sAllowedMemoryNodes: %s\n"
456 "%sStartupAllowedMemoryNodes: %s\n"
457 "%sIOWeight: %" PRIu64
"\n"
458 "%sStartupIOWeight: %" PRIu64
"\n"
459 "%sBlockIOWeight: %" PRIu64
"\n"
460 "%sStartupBlockIOWeight: %" PRIu64
"\n"
461 "%sDefaultMemoryMin: %" PRIu64
"\n"
462 "%sDefaultMemoryLow: %" PRIu64
"\n"
463 "%sMemoryMin: %" PRIu64
"%s\n"
464 "%sMemoryLow: %" PRIu64
"%s\n"
465 "%sMemoryHigh: %" PRIu64
"%s\n"
466 "%sMemoryMax: %" PRIu64
"%s\n"
467 "%sMemorySwapMax: %" PRIu64
"%s\n"
468 "%sMemoryLimit: %" PRIu64
"\n"
469 "%sTasksMax: %" PRIu64
"\n"
470 "%sDevicePolicy: %s\n"
471 "%sDisableControllers: %s\n"
473 "%sManagedOOMSwap: %s\n"
474 "%sManagedOOMMemoryPressure: %s\n"
475 "%sManagedOOMMemoryPressureLimit: " PERMYRIAD_AS_PERCENT_FORMAT_STR
"\n"
476 "%sManagedOOMPreference: %s\n",
477 prefix
, yes_no(c
->cpu_accounting
),
478 prefix
, yes_no(c
->io_accounting
),
479 prefix
, yes_no(c
->blockio_accounting
),
480 prefix
, yes_no(c
->memory_accounting
),
481 prefix
, yes_no(c
->tasks_accounting
),
482 prefix
, yes_no(c
->ip_accounting
),
483 prefix
, c
->cpu_weight
,
484 prefix
, c
->startup_cpu_weight
,
485 prefix
, c
->cpu_shares
,
486 prefix
, c
->startup_cpu_shares
,
487 prefix
, FORMAT_TIMESPAN(c
->cpu_quota_per_sec_usec
, 1),
488 prefix
, FORMAT_TIMESPAN(c
->cpu_quota_period_usec
, 1),
489 prefix
, strempty(cpuset_cpus
),
490 prefix
, strempty(startup_cpuset_cpus
),
491 prefix
, strempty(cpuset_mems
),
492 prefix
, strempty(startup_cpuset_mems
),
493 prefix
, c
->io_weight
,
494 prefix
, c
->startup_io_weight
,
495 prefix
, c
->blockio_weight
,
496 prefix
, c
->startup_blockio_weight
,
497 prefix
, c
->default_memory_min
,
498 prefix
, c
->default_memory_low
,
499 prefix
, c
->memory_min
, format_cgroup_memory_limit_comparison(cda
, sizeof(cda
), u
, "MemoryMin"),
500 prefix
, c
->memory_low
, format_cgroup_memory_limit_comparison(cdb
, sizeof(cdb
), u
, "MemoryLow"),
501 prefix
, c
->memory_high
, format_cgroup_memory_limit_comparison(cdc
, sizeof(cdc
), u
, "MemoryHigh"),
502 prefix
, c
->memory_max
, format_cgroup_memory_limit_comparison(cdd
, sizeof(cdd
), u
, "MemoryMax"),
503 prefix
, c
->memory_swap_max
, format_cgroup_memory_limit_comparison(cde
, sizeof(cde
), u
, "MemorySwapMax"),
504 prefix
, c
->memory_limit
,
505 prefix
, tasks_max_resolve(&c
->tasks_max
),
506 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
507 prefix
, strempty(disable_controllers_str
),
508 prefix
, yes_no(c
->delegate
),
509 prefix
, managed_oom_mode_to_string(c
->moom_swap
),
510 prefix
, managed_oom_mode_to_string(c
->moom_mem_pressure
),
511 prefix
, PERMYRIAD_AS_PERCENT_FORMAT_VAL(UINT32_SCALE_TO_PERMYRIAD(c
->moom_mem_pressure_limit
)),
512 prefix
, managed_oom_preference_to_string(c
->moom_preference
));
515 _cleanup_free_
char *t
= NULL
;
517 (void) cg_mask_to_string(c
->delegate_controllers
, &t
);
519 fprintf(f
, "%sDelegateControllers: %s\n",
524 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
526 "%sDeviceAllow: %s %s%s%s\n",
529 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
531 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
533 "%sIODeviceWeight: %s %" PRIu64
"\n",
538 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
540 "%sIODeviceLatencyTargetSec: %s %s\n",
543 FORMAT_TIMESPAN(l
->target_usec
, 1));
545 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
)
546 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
547 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
551 cgroup_io_limit_type_to_string(type
),
553 FORMAT_BYTES(il
->limits
[type
]));
555 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
557 "%sBlockIODeviceWeight: %s %" PRIu64
,
562 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
563 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
565 "%sBlockIOReadBandwidth: %s %s\n",
568 FORMAT_BYTES(b
->rbps
));
569 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
571 "%sBlockIOWriteBandwidth: %s %s\n",
574 FORMAT_BYTES(b
->wbps
));
577 SET_FOREACH(iaai
, c
->ip_address_allow
) {
578 _cleanup_free_
char *k
= NULL
;
580 (void) in_addr_prefix_to_string(iaai
->family
, &iaai
->address
, iaai
->prefixlen
, &k
);
581 fprintf(f
, "%sIPAddressAllow: %s\n", prefix
, strnull(k
));
584 SET_FOREACH(iaai
, c
->ip_address_deny
) {
585 _cleanup_free_
char *k
= NULL
;
587 (void) in_addr_prefix_to_string(iaai
->family
, &iaai
->address
, iaai
->prefixlen
, &k
);
588 fprintf(f
, "%sIPAddressDeny: %s\n", prefix
, strnull(k
));
591 STRV_FOREACH(path
, c
->ip_filters_ingress
)
592 fprintf(f
, "%sIPIngressFilterPath: %s\n", prefix
, *path
);
594 STRV_FOREACH(path
, c
->ip_filters_egress
)
595 fprintf(f
, "%sIPEgressFilterPath: %s\n", prefix
, *path
);
597 LIST_FOREACH(programs
, p
, c
->bpf_foreign_programs
)
598 fprintf(f
, "%sBPFProgram: %s:%s",
599 prefix
, bpf_cgroup_attach_type_to_string(p
->attach_type
), p
->bpffs_path
);
601 if (c
->socket_bind_allow
) {
602 fprintf(f
, "%sSocketBindAllow:", prefix
);
603 LIST_FOREACH(socket_bind_items
, bi
, c
->socket_bind_allow
)
604 cgroup_context_dump_socket_bind_item(bi
, f
);
608 if (c
->socket_bind_deny
) {
609 fprintf(f
, "%sSocketBindDeny:", prefix
);
610 LIST_FOREACH(socket_bind_items
, bi
, c
->socket_bind_deny
)
611 cgroup_context_dump_socket_bind_item(bi
, f
);
615 if (c
->restrict_network_interfaces
) {
617 SET_FOREACH(iface
, c
->restrict_network_interfaces
)
618 fprintf(f
, "%sRestrictNetworkInterfaces: %s\n", prefix
, iface
);
622 void cgroup_context_dump_socket_bind_item(const CGroupSocketBindItem
*item
, FILE *f
) {
623 const char *family
, *colon1
, *protocol
= "", *colon2
= "";
625 family
= strempty(af_to_ipv4_ipv6(item
->address_family
));
626 colon1
= isempty(family
) ? "" : ":";
628 if (item
->ip_protocol
!= 0) {
629 protocol
= ip_protocol_to_tcp_udp(item
->ip_protocol
);
633 if (item
->nr_ports
== 0)
634 fprintf(f
, " %s%s%s%sany", family
, colon1
, protocol
, colon2
);
635 else if (item
->nr_ports
== 1)
636 fprintf(f
, " %s%s%s%s%" PRIu16
, family
, colon1
, protocol
, colon2
, item
->port_min
);
638 uint16_t port_max
= item
->port_min
+ item
->nr_ports
- 1;
639 fprintf(f
, " %s%s%s%s%" PRIu16
"-%" PRIu16
, family
, colon1
, protocol
, colon2
,
640 item
->port_min
, port_max
);
644 int cgroup_add_device_allow(CGroupContext
*c
, const char *dev
, const char *mode
) {
645 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
646 _cleanup_free_
char *d
= NULL
;
650 assert(isempty(mode
) || in_charset(mode
, "rwm"));
652 a
= new(CGroupDeviceAllow
, 1);
660 *a
= (CGroupDeviceAllow
) {
662 .r
= isempty(mode
) || strchr(mode
, 'r'),
663 .w
= isempty(mode
) || strchr(mode
, 'w'),
664 .m
= isempty(mode
) || strchr(mode
, 'm'),
667 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
673 int cgroup_add_bpf_foreign_program(CGroupContext
*c
, uint32_t attach_type
, const char *bpffs_path
) {
674 CGroupBPFForeignProgram
*p
;
675 _cleanup_free_
char *d
= NULL
;
680 if (!path_is_normalized(bpffs_path
) || !path_is_absolute(bpffs_path
))
681 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Path is not normalized: %m");
683 d
= strdup(bpffs_path
);
687 p
= new(CGroupBPFForeignProgram
, 1);
691 *p
= (CGroupBPFForeignProgram
) {
692 .attach_type
= attach_type
,
693 .bpffs_path
= TAKE_PTR(d
),
696 LIST_PREPEND(programs
, c
->bpf_foreign_programs
, TAKE_PTR(p
));
701 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
702 uint64_t unit_get_ancestor_##entry(Unit *u) { \
705 /* 1. Is entry set in this unit? If so, use that. \
706 * 2. Is the default for this entry set in any \
707 * ancestor? If so, use that. \
708 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
712 c = unit_get_cgroup_context(u); \
713 if (c && c->entry##_set) \
716 while ((u = UNIT_GET_SLICE(u))) { \
717 c = unit_get_cgroup_context(u); \
718 if (c && c->default_##entry##_set) \
719 return c->default_##entry; \
722 /* We've reached the root, but nobody had default for \
723 * this entry set, so set it to the kernel default. */ \
724 return CGROUP_LIMIT_MIN; \
727 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
728 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
730 static void unit_set_xattr_graceful(Unit
*u
, const char *cgroup_path
, const char *name
, const void *data
, size_t size
) {
740 cgroup_path
= u
->cgroup_path
;
743 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, cgroup_path
, name
, data
, size
, 0);
745 log_unit_debug_errno(u
, r
, "Failed to set '%s' xattr on control group %s, ignoring: %m", name
, empty_to_root(cgroup_path
));
748 static void unit_remove_xattr_graceful(Unit
*u
, const char *cgroup_path
, const char *name
) {
758 cgroup_path
= u
->cgroup_path
;
761 r
= cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER
, cgroup_path
, name
);
762 if (r
< 0 && r
!= -ENODATA
)
763 log_unit_debug_errno(u
, r
, "Failed to remove '%s' xattr flag on control group %s, ignoring: %m", name
, empty_to_root(cgroup_path
));
766 void cgroup_oomd_xattr_apply(Unit
*u
, const char *cgroup_path
) {
771 c
= unit_get_cgroup_context(u
);
775 if (c
->moom_preference
== MANAGED_OOM_PREFERENCE_OMIT
)
776 unit_set_xattr_graceful(u
, cgroup_path
, "user.oomd_omit", "1", 1);
778 if (c
->moom_preference
== MANAGED_OOM_PREFERENCE_AVOID
)
779 unit_set_xattr_graceful(u
, cgroup_path
, "user.oomd_avoid", "1", 1);
781 if (c
->moom_preference
!= MANAGED_OOM_PREFERENCE_AVOID
)
782 unit_remove_xattr_graceful(u
, cgroup_path
, "user.oomd_avoid");
784 if (c
->moom_preference
!= MANAGED_OOM_PREFERENCE_OMIT
)
785 unit_remove_xattr_graceful(u
, cgroup_path
, "user.oomd_omit");
788 static void cgroup_xattr_apply(Unit
*u
) {
794 if (!MANAGER_IS_SYSTEM(u
->manager
))
797 b
= !sd_id128_is_null(u
->invocation_id
);
798 FOREACH_STRING(xn
, "trusted.invocation_id", "user.invocation_id") {
800 unit_set_xattr_graceful(u
, NULL
, xn
, SD_ID128_TO_STRING(u
->invocation_id
), 32);
802 unit_remove_xattr_graceful(u
, NULL
, xn
);
805 /* Indicate on the cgroup whether delegation is on, via an xattr. This is best-effort, as old kernels
806 * didn't support xattrs on cgroups at all. Later they got support for setting 'trusted.*' xattrs,
807 * and even later 'user.*' xattrs. We started setting this field when 'trusted.*' was added, and
808 * given this is now pretty much API, let's continue to support that. But also set 'user.*' as well,
809 * since it is readable by any user, not just CAP_SYS_ADMIN. This hence comes with slightly weaker
810 * security (as users who got delegated cgroups could turn it off if they like), but this shouldn't
811 * be a big problem given this communicates delegation state to clients, but the manager never reads
813 b
= unit_cgroup_delegate(u
);
814 FOREACH_STRING(xn
, "trusted.delegate", "user.delegate") {
816 unit_set_xattr_graceful(u
, NULL
, xn
, "1", 1);
818 unit_remove_xattr_graceful(u
, NULL
, xn
);
821 cgroup_oomd_xattr_apply(u
, u
->cgroup_path
);
824 static int lookup_block_device(const char *p
, dev_t
*ret
) {
832 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
833 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
836 if (stat(p
, &st
) < 0)
837 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
843 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
846 return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK
),
847 "Device node '%s' is a character device, but block device needed.", p
);
850 else if (major(dev
) != 0)
851 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
853 /* If this is btrfs, getting the backing block device is a bit harder */
854 r
= btrfs_get_block_device(p
, ret
);
856 return log_warning_errno(SYNTHETIC_ERRNO(ENODEV
),
857 "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
859 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
862 /* If this is a LUKS/DM device, recursively try to get the originating block device */
863 while (block_get_originating(*ret
, ret
) > 0);
865 /* If this is a partition, try to get the originating block device */
866 (void) block_get_whole_disk(*ret
, ret
);
870 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
871 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
872 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
875 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
876 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
877 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
880 static bool cgroup_context_has_allowed_cpus(CGroupContext
*c
) {
881 return c
->cpuset_cpus
.set
|| c
->startup_cpuset_cpus
.set
;
884 static bool cgroup_context_has_allowed_mems(CGroupContext
*c
) {
885 return c
->cpuset_mems
.set
|| c
->startup_cpuset_mems
.set
;
888 static uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
889 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
890 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
891 return c
->startup_cpu_weight
;
892 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
893 return c
->cpu_weight
;
895 return CGROUP_WEIGHT_DEFAULT
;
898 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
899 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
900 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
901 return c
->startup_cpu_shares
;
902 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
903 return c
->cpu_shares
;
905 return CGROUP_CPU_SHARES_DEFAULT
;
908 static CPUSet
*cgroup_context_allowed_cpus(CGroupContext
*c
, ManagerState state
) {
909 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
910 c
->startup_cpuset_cpus
.set
)
911 return &c
->startup_cpuset_cpus
;
913 return &c
->cpuset_cpus
;
916 static CPUSet
*cgroup_context_allowed_mems(CGroupContext
*c
, ManagerState state
) {
917 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
918 c
->startup_cpuset_mems
.set
)
919 return &c
->startup_cpuset_mems
;
921 return &c
->cpuset_mems
;
924 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
925 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
926 * need to be higher than that boundary. quota is specified in USecPerSec.
927 * Additionally, period must be at most max_period. */
930 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
933 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
936 if (quota
== USEC_INFINITY
)
937 /* Always use default period for infinity quota. */
938 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
940 if (period
== USEC_INFINITY
)
941 /* Default period was requested. */
942 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
944 /* Clamp to interval [1ms, 1s] */
945 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
947 if (new_period
!= period
) {
948 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
,
949 "Clamping CPU interval for cpu.max: period is now %s",
950 FORMAT_TIMESPAN(new_period
, 1));
951 u
->warned_clamping_cpu_quota_period
= true;
957 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
958 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
960 xsprintf(buf
, "%" PRIu64
"\n", weight
);
961 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
964 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
965 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
967 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
968 if (quota
!= USEC_INFINITY
)
969 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
970 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
972 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
973 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
976 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
977 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
979 xsprintf(buf
, "%" PRIu64
"\n", shares
);
980 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
983 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
984 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
986 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
988 xsprintf(buf
, USEC_FMT
"\n", period
);
989 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
991 if (quota
!= USEC_INFINITY
) {
992 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
993 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
995 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
998 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
999 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
1000 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
1003 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
1004 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
1005 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
1008 static void cgroup_apply_unified_cpuset(Unit
*u
, const CPUSet
*cpus
, const char *name
) {
1009 _cleanup_free_
char *buf
= NULL
;
1011 buf
= cpu_set_to_range_string(cpus
);
1017 (void) set_attribute_and_warn(u
, "cpuset", name
, buf
);
1020 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
1021 return c
->io_accounting
||
1022 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
1023 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
1024 c
->io_device_weights
||
1025 c
->io_device_latencies
||
1026 c
->io_device_limits
;
1029 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
1030 return c
->blockio_accounting
||
1031 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
1032 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
1033 c
->blockio_device_weights
||
1034 c
->blockio_device_bandwidths
;
1037 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
1038 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1039 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
1040 return c
->startup_io_weight
;
1041 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
1042 return c
->io_weight
;
1044 return CGROUP_WEIGHT_DEFAULT
;
1047 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
1048 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
, MANAGER_STOPPING
) &&
1049 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
1050 return c
->startup_blockio_weight
;
1051 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
1052 return c
->blockio_weight
;
1054 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
1057 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
1058 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
1059 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
1062 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
1063 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
1064 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
1067 static void set_bfq_weight(Unit
*u
, const char *controller
, uint64_t io_weight
) {
1068 char buf
[DECIMAL_STR_MAX(uint64_t)+STRLEN("\n")];
1070 uint64_t bfq_weight
;
1072 /* FIXME: drop this function when distro kernels properly support BFQ through "io.weight"
1073 * See also: https://github.com/systemd/systemd/pull/13335 and
1074 * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9. */
1075 p
= strjoina(controller
, ".bfq.weight");
1076 /* Adjust to kernel range is 1..1000, the default is 100. */
1077 bfq_weight
= BFQ_WEIGHT(io_weight
);
1079 xsprintf(buf
, "%" PRIu64
"\n", bfq_weight
);
1081 if (set_attribute_and_warn(u
, controller
, p
, buf
) >= 0 && io_weight
!= bfq_weight
)
1082 log_unit_debug(u
, "%sIOWeight=%" PRIu64
" scaled to %s=%" PRIu64
,
1083 streq(controller
, "blkio") ? "Block" : "",
1084 io_weight
, p
, bfq_weight
);
1087 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
1088 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1092 r
= lookup_block_device(dev_path
, &dev
);
1096 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
1097 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1100 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
1101 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1105 r
= lookup_block_device(dev_path
, &dev
);
1109 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
1110 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
1113 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
1114 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
1118 r
= lookup_block_device(dev_path
, &dev
);
1122 if (target
!= USEC_INFINITY
)
1123 xsprintf(buf
, "%u:%u target=%" PRIu64
"\n", major(dev
), minor(dev
), target
);
1125 xsprintf(buf
, "%u:%u target=max\n", major(dev
), minor(dev
));
1127 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
1130 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
1131 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)],
1132 buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
1135 if (lookup_block_device(dev_path
, &dev
) < 0)
1138 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1139 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
1140 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
1142 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
1144 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
1145 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
1146 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
1147 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
1150 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
1151 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1154 if (lookup_block_device(dev_path
, &dev
) < 0)
1157 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
1158 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
1160 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
1161 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
1164 static bool unit_has_unified_memory_config(Unit
*u
) {
1169 assert_se(c
= unit_get_cgroup_context(u
));
1171 return unit_get_ancestor_memory_min(u
) > 0 || unit_get_ancestor_memory_low(u
) > 0 ||
1172 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
||
1173 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
;
1176 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
1177 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
1179 if (v
!= CGROUP_LIMIT_MAX
)
1180 xsprintf(buf
, "%" PRIu64
"\n", v
);
1182 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
1185 static void cgroup_apply_firewall(Unit
*u
) {
1188 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
1190 if (bpf_firewall_compile(u
) < 0)
1193 (void) bpf_firewall_load_custom(u
);
1194 (void) bpf_firewall_install(u
);
1197 static void cgroup_apply_socket_bind(Unit
*u
) {
1200 (void) bpf_socket_bind_install(u
);
1203 static void cgroup_apply_restrict_network_interfaces(Unit
*u
) {
1206 (void) restrict_network_interfaces_install(u
);
1209 static int cgroup_apply_devices(Unit
*u
) {
1210 _cleanup_(bpf_program_freep
) BPFProgram
*prog
= NULL
;
1213 CGroupDevicePolicy policy
;
1216 assert_se(c
= unit_get_cgroup_context(u
));
1217 assert_se(path
= u
->cgroup_path
);
1219 policy
= c
->device_policy
;
1221 if (cg_all_unified() > 0) {
1222 r
= bpf_devices_cgroup_init(&prog
, policy
, c
->device_allow
);
1224 return log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
1227 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore
1230 if (c
->device_allow
|| policy
!= CGROUP_DEVICE_POLICY_AUTO
)
1231 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
1233 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
1235 log_unit_full_errno(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1236 "Failed to reset devices.allow/devices.deny: %m");
1239 bool allow_list_static
= policy
== CGROUP_DEVICE_POLICY_CLOSED
||
1240 (policy
== CGROUP_DEVICE_POLICY_AUTO
&& c
->device_allow
);
1241 if (allow_list_static
)
1242 (void) bpf_devices_allow_list_static(prog
, path
);
1244 bool any
= allow_list_static
;
1245 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1259 if (path_startswith(a
->path
, "/dev/"))
1260 r
= bpf_devices_allow_list_device(prog
, path
, a
->path
, acc
);
1261 else if ((val
= startswith(a
->path
, "block-")))
1262 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'b', acc
);
1263 else if ((val
= startswith(a
->path
, "char-")))
1264 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'c', acc
);
1266 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1275 log_unit_warning_errno(u
, SYNTHETIC_ERRNO(ENODEV
), "No devices matched by device filter.");
1277 /* The kernel verifier would reject a program we would build with the normal intro and outro
1278 but no allow-listing rules (outro would contain an unreachable instruction for successful
1280 policy
= CGROUP_DEVICE_POLICY_STRICT
;
1283 r
= bpf_devices_apply_policy(&prog
, policy
, any
, path
, &u
->bpf_device_control_installed
);
1285 static bool warned
= false;
1287 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1288 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1289 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1290 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1297 static void set_io_weight(Unit
*u
, uint64_t weight
) {
1298 char buf
[STRLEN("default \n")+DECIMAL_STR_MAX(uint64_t)];
1302 set_bfq_weight(u
, "io", weight
);
1304 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1305 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1308 static void set_blkio_weight(Unit
*u
, uint64_t weight
) {
1309 char buf
[STRLEN("\n")+DECIMAL_STR_MAX(uint64_t)];
1313 set_bfq_weight(u
, "blkio", weight
);
1315 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1316 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1319 static void cgroup_apply_bpf_foreign_program(Unit
*u
) {
1322 (void) bpf_foreign_install(u
);
1325 static void cgroup_context_apply(
1327 CGroupMask apply_mask
,
1328 ManagerState state
) {
1332 bool is_host_root
, is_local_root
;
1337 /* Nothing to do? Exit early! */
1338 if (apply_mask
== 0)
1341 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
1342 * attributes should only be managed for cgroups further down the tree. */
1343 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1344 is_host_root
= unit_has_host_root_cgroup(u
);
1346 assert_se(c
= unit_get_cgroup_context(u
));
1347 assert_se(path
= u
->cgroup_path
);
1349 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
1352 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
1353 * then), and missing cgroups, i.e. EROFS and ENOENT. */
1355 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
1356 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
1357 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
1358 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
1359 * we couldn't even write to them if we wanted to). */
1360 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
1362 if (cg_all_unified() > 0) {
1365 if (cgroup_context_has_cpu_weight(c
))
1366 weight
= cgroup_context_cpu_weight(c
, state
);
1367 else if (cgroup_context_has_cpu_shares(c
)) {
1370 shares
= cgroup_context_cpu_shares(c
, state
);
1371 weight
= cgroup_cpu_shares_to_weight(shares
);
1373 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
1374 shares
, weight
, path
);
1376 weight
= CGROUP_WEIGHT_DEFAULT
;
1378 cgroup_apply_unified_cpu_weight(u
, weight
);
1379 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1384 if (cgroup_context_has_cpu_weight(c
)) {
1387 weight
= cgroup_context_cpu_weight(c
, state
);
1388 shares
= cgroup_cpu_weight_to_shares(weight
);
1390 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
1391 weight
, shares
, path
);
1392 } else if (cgroup_context_has_cpu_shares(c
))
1393 shares
= cgroup_context_cpu_shares(c
, state
);
1395 shares
= CGROUP_CPU_SHARES_DEFAULT
;
1397 cgroup_apply_legacy_cpu_shares(u
, shares
);
1398 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1402 if ((apply_mask
& CGROUP_MASK_CPUSET
) && !is_local_root
) {
1403 cgroup_apply_unified_cpuset(u
, cgroup_context_allowed_cpus(c
, state
), "cpuset.cpus");
1404 cgroup_apply_unified_cpuset(u
, cgroup_context_allowed_mems(c
, state
), "cpuset.mems");
1407 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
1408 * controller), and in case of containers we want to leave control of these attributes to the container manager
1409 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
1410 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
1411 bool has_io
, has_blockio
;
1414 has_io
= cgroup_context_has_io_config(c
);
1415 has_blockio
= cgroup_context_has_blockio_config(c
);
1418 weight
= cgroup_context_io_weight(c
, state
);
1419 else if (has_blockio
) {
1420 uint64_t blkio_weight
;
1422 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
1423 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
1425 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
1426 blkio_weight
, weight
);
1428 weight
= CGROUP_WEIGHT_DEFAULT
;
1430 set_io_weight(u
, weight
);
1433 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
1434 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
1436 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
1437 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
1439 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
1440 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
1442 } else if (has_blockio
) {
1443 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
1444 weight
= cgroup_weight_blkio_to_io(w
->weight
);
1446 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
1447 w
->weight
, weight
, w
->path
);
1449 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
1452 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
1453 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
1455 for (CGroupIOLimitType type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1456 limits
[type
] = cgroup_io_limit_defaults
[type
];
1458 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
1459 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
1461 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
1462 b
->rbps
, b
->wbps
, b
->path
);
1464 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
1469 if (apply_mask
& CGROUP_MASK_BLKIO
) {
1470 bool has_io
, has_blockio
;
1472 has_io
= cgroup_context_has_io_config(c
);
1473 has_blockio
= cgroup_context_has_blockio_config(c
);
1475 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
1476 * left to our container manager, too. */
1477 if (!is_local_root
) {
1483 io_weight
= cgroup_context_io_weight(c
, state
);
1484 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
1486 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
1488 } else if (has_blockio
)
1489 weight
= cgroup_context_blkio_weight(c
, state
);
1491 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
1493 set_blkio_weight(u
, weight
);
1496 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
1497 weight
= cgroup_weight_io_to_blkio(w
->weight
);
1499 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
1500 w
->weight
, weight
, w
->path
);
1502 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
1504 else if (has_blockio
)
1505 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
1506 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
1509 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
1510 * roots, as there we want the container manager to handle it */
1511 if (is_host_root
|| !is_local_root
) {
1513 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
1514 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
1515 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
1517 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
1519 else if (has_blockio
)
1520 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
1521 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
1525 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
1526 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
1527 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
1528 * write to this if we wanted to.) */
1529 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
1531 if (cg_all_unified() > 0) {
1532 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
;
1534 if (unit_has_unified_memory_config(u
)) {
1535 max
= c
->memory_max
;
1536 swap_max
= c
->memory_swap_max
;
1538 max
= c
->memory_limit
;
1540 if (max
!= CGROUP_LIMIT_MAX
)
1541 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
1544 cgroup_apply_unified_memory_limit(u
, "memory.min", unit_get_ancestor_memory_min(u
));
1545 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
1546 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
1547 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
1548 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
1550 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
1553 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1556 if (unit_has_unified_memory_config(u
)) {
1557 val
= c
->memory_max
;
1558 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIi64
" as MemoryLimit=", val
);
1560 val
= c
->memory_limit
;
1562 if (val
== CGROUP_LIMIT_MAX
)
1563 strncpy(buf
, "-1\n", sizeof(buf
));
1565 xsprintf(buf
, "%" PRIu64
"\n", val
);
1567 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
1571 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
1572 * containers, where we leave this to the manager */
1573 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
1574 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
))
1575 (void) cgroup_apply_devices(u
);
1577 if (apply_mask
& CGROUP_MASK_PIDS
) {
1580 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1581 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1582 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1583 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1584 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1585 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1586 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1587 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1588 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1589 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1590 * which is desirable so that there's an official way to release control of the sysctl from
1591 * systemd: set the limit to unbounded and reload. */
1593 if (tasks_max_isset(&c
->tasks_max
)) {
1594 u
->manager
->sysctl_pid_max_changed
= true;
1595 r
= procfs_tasks_set_limit(tasks_max_resolve(&c
->tasks_max
));
1596 } else if (u
->manager
->sysctl_pid_max_changed
)
1597 r
= procfs_tasks_set_limit(TASKS_MAX
);
1601 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
1602 "Failed to write to tasks limit sysctls: %m");
1605 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
1606 * leave it for the container manager. */
1607 if (!is_local_root
) {
1608 if (tasks_max_isset(&c
->tasks_max
)) {
1609 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1611 xsprintf(buf
, "%" PRIu64
"\n", tasks_max_resolve(&c
->tasks_max
));
1612 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
1614 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
1618 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
1619 cgroup_apply_firewall(u
);
1621 if (apply_mask
& CGROUP_MASK_BPF_FOREIGN
)
1622 cgroup_apply_bpf_foreign_program(u
);
1624 if (apply_mask
& CGROUP_MASK_BPF_SOCKET_BIND
)
1625 cgroup_apply_socket_bind(u
);
1627 if (apply_mask
& CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
)
1628 cgroup_apply_restrict_network_interfaces(u
);
1631 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
1635 c
= unit_get_cgroup_context(u
);
1639 if (c
->ip_accounting
||
1640 !set_isempty(c
->ip_address_allow
) ||
1641 !set_isempty(c
->ip_address_deny
) ||
1642 c
->ip_filters_ingress
||
1643 c
->ip_filters_egress
)
1646 /* If any parent slice has an IP access list defined, it applies too */
1647 for (Unit
*p
= UNIT_GET_SLICE(u
); p
; p
= UNIT_GET_SLICE(p
)) {
1648 c
= unit_get_cgroup_context(p
);
1652 if (!set_isempty(c
->ip_address_allow
) ||
1653 !set_isempty(c
->ip_address_deny
))
1660 static bool unit_get_needs_bpf_foreign_program(Unit
*u
) {
1664 c
= unit_get_cgroup_context(u
);
1668 return !LIST_IS_EMPTY(c
->bpf_foreign_programs
);
1671 static bool unit_get_needs_socket_bind(Unit
*u
) {
1675 c
= unit_get_cgroup_context(u
);
1679 return c
->socket_bind_allow
|| c
->socket_bind_deny
;
1682 static bool unit_get_needs_restrict_network_interfaces(Unit
*u
) {
1686 c
= unit_get_cgroup_context(u
);
1690 return !set_isempty(c
->restrict_network_interfaces
);
1693 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
1694 CGroupMask mask
= 0;
1699 assert_se(c
= unit_get_cgroup_context(u
));
1701 /* Figure out which controllers we need, based on the cgroup context object */
1703 if (c
->cpu_accounting
)
1704 mask
|= get_cpu_accounting_mask();
1706 if (cgroup_context_has_cpu_weight(c
) ||
1707 cgroup_context_has_cpu_shares(c
) ||
1708 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
1709 mask
|= CGROUP_MASK_CPU
;
1711 if (cgroup_context_has_allowed_cpus(c
) || cgroup_context_has_allowed_mems(c
))
1712 mask
|= CGROUP_MASK_CPUSET
;
1714 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
1715 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1717 if (c
->memory_accounting
||
1718 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
1719 unit_has_unified_memory_config(u
))
1720 mask
|= CGROUP_MASK_MEMORY
;
1722 if (c
->device_allow
||
1723 c
->device_policy
!= CGROUP_DEVICE_POLICY_AUTO
)
1724 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
1726 if (c
->tasks_accounting
||
1727 tasks_max_isset(&c
->tasks_max
))
1728 mask
|= CGROUP_MASK_PIDS
;
1730 return CGROUP_MASK_EXTEND_JOINED(mask
);
1733 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
1734 CGroupMask mask
= 0;
1736 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
1739 if (unit_get_needs_bpf_firewall(u
))
1740 mask
|= CGROUP_MASK_BPF_FIREWALL
;
1742 if (unit_get_needs_bpf_foreign_program(u
))
1743 mask
|= CGROUP_MASK_BPF_FOREIGN
;
1745 if (unit_get_needs_socket_bind(u
))
1746 mask
|= CGROUP_MASK_BPF_SOCKET_BIND
;
1748 if (unit_get_needs_restrict_network_interfaces(u
))
1749 mask
|= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
;
1754 CGroupMask
unit_get_own_mask(Unit
*u
) {
1757 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
1758 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
1760 if (u
->load_state
!= UNIT_LOADED
)
1763 c
= unit_get_cgroup_context(u
);
1767 return unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
);
1770 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
1773 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1774 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1776 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1778 if (!unit_cgroup_delegate(u
))
1781 if (cg_all_unified() <= 0) {
1784 e
= unit_get_exec_context(u
);
1785 if (e
&& !exec_context_maintains_privileges(e
))
1789 assert_se(c
= unit_get_cgroup_context(u
));
1790 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
1793 static CGroupMask
unit_get_subtree_mask(Unit
*u
) {
1795 /* Returns the mask of this subtree, meaning of the group
1796 * itself and its children. */
1798 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
1801 CGroupMask
unit_get_members_mask(Unit
*u
) {
1804 /* Returns the mask of controllers all of the unit's children require, merged */
1806 if (u
->cgroup_members_mask_valid
)
1807 return u
->cgroup_members_mask
; /* Use cached value if possible */
1809 u
->cgroup_members_mask
= 0;
1811 if (u
->type
== UNIT_SLICE
) {
1814 UNIT_FOREACH_DEPENDENCY(member
, u
, UNIT_ATOM_SLICE_OF
)
1815 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
1818 u
->cgroup_members_mask_valid
= true;
1819 return u
->cgroup_members_mask
;
1822 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
1826 /* Returns the mask of controllers all of the unit's siblings
1827 * require, i.e. the members mask of the unit's parent slice
1828 * if there is one. */
1830 slice
= UNIT_GET_SLICE(u
);
1832 return unit_get_members_mask(slice
);
1834 return unit_get_subtree_mask(u
); /* we are the top-level slice */
1837 static CGroupMask
unit_get_disable_mask(Unit
*u
) {
1840 c
= unit_get_cgroup_context(u
);
1844 return c
->disable_controllers
;
1847 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
1852 mask
= unit_get_disable_mask(u
);
1854 /* Returns the mask of controllers which are marked as forcibly
1855 * disabled in any ancestor unit or the unit in question. */
1857 slice
= UNIT_GET_SLICE(u
);
1859 mask
|= unit_get_ancestor_disable_mask(slice
);
1864 CGroupMask
unit_get_target_mask(Unit
*u
) {
1865 CGroupMask own_mask
, mask
;
1867 /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything
1868 * it needs itself, plus all that its children need, plus all that its siblings need. This is
1869 * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each
1870 * hierarchy that shall be enabled for it. */
1872 own_mask
= unit_get_own_mask(u
);
1874 if (own_mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
1875 emit_bpf_firewall_warning(u
);
1877 mask
= own_mask
| unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
1879 mask
&= u
->manager
->cgroup_supported
;
1880 mask
&= ~unit_get_ancestor_disable_mask(u
);
1885 CGroupMask
unit_get_enable_mask(Unit
*u
) {
1888 /* This returns the cgroup mask of all controllers to enable
1889 * for the children of a specific cgroup. This is primarily
1890 * useful for the unified cgroup hierarchy, where each cgroup
1891 * controls which controllers are enabled for its children. */
1893 mask
= unit_get_members_mask(u
);
1894 mask
&= u
->manager
->cgroup_supported
;
1895 mask
&= ~unit_get_ancestor_disable_mask(u
);
1900 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
1905 /* Recurse invalidate the member masks cache all the way up the tree */
1906 u
->cgroup_members_mask_valid
= false;
1908 slice
= UNIT_GET_SLICE(u
);
1910 unit_invalidate_cgroup_members_masks(slice
);
1913 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
1915 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1919 if (u
->cgroup_path
&&
1920 u
->cgroup_realized
&&
1921 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
1922 return u
->cgroup_path
;
1924 u
= UNIT_GET_SLICE(u
);
1930 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1931 /* If not realized at all, migrate to root ("").
1932 * It may happen if we're upgrading from older version that didn't clean up.
1934 return strempty(unit_get_realized_cgroup_path(userdata
, mask
));
1937 char *unit_default_cgroup_path(const Unit
*u
) {
1938 _cleanup_free_
char *escaped
= NULL
, *slice_path
= NULL
;
1944 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1945 return strdup(u
->manager
->cgroup_root
);
1947 slice
= UNIT_GET_SLICE(u
);
1948 if (slice
&& !unit_has_name(slice
, SPECIAL_ROOT_SLICE
)) {
1949 r
= cg_slice_to_path(slice
->id
, &slice_path
);
1954 escaped
= cg_escape(u
->id
);
1958 return path_join(empty_to_root(u
->manager
->cgroup_root
), slice_path
, escaped
);
1961 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1962 _cleanup_free_
char *p
= NULL
;
1967 if (streq_ptr(u
->cgroup_path
, path
))
1977 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1982 unit_release_cgroup(u
);
1983 u
->cgroup_path
= TAKE_PTR(p
);
1988 int unit_watch_cgroup(Unit
*u
) {
1989 _cleanup_free_
char *events
= NULL
;
1994 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
1995 * cgroupv2 is available. */
1997 if (!u
->cgroup_path
)
2000 if (u
->cgroup_control_inotify_wd
>= 0)
2003 /* Only applies to the unified hierarchy */
2004 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2006 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
2010 /* No point in watch the top-level slice, it's never going to run empty. */
2011 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
2014 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
2018 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
2022 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
2023 if (u
->cgroup_control_inotify_wd
< 0) {
2025 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
2026 * is not an error */
2029 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
));
2032 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
2034 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
));
2039 int unit_watch_cgroup_memory(Unit
*u
) {
2040 _cleanup_free_
char *events
= NULL
;
2046 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
2047 * cgroupv2 is available. */
2049 if (!u
->cgroup_path
)
2052 c
= unit_get_cgroup_context(u
);
2056 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
2057 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
2059 if (!c
->memory_accounting
)
2062 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
2063 * we also don't want to generate a log message for each parent cgroup of a process. */
2064 if (u
->type
== UNIT_SLICE
)
2067 if (u
->cgroup_memory_inotify_wd
>= 0)
2070 /* Only applies to the unified hierarchy */
2071 r
= cg_all_unified();
2073 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
2077 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
2081 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
2085 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
2086 if (u
->cgroup_memory_inotify_wd
< 0) {
2088 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
2089 * is not an error */
2092 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
));
2095 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
2097 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
));
2102 int unit_pick_cgroup_path(Unit
*u
) {
2103 _cleanup_free_
char *path
= NULL
;
2111 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2114 path
= unit_default_cgroup_path(u
);
2118 r
= unit_set_cgroup_path(u
, path
);
2120 return log_unit_error_errno(u
, r
, "Control group %s exists already.", empty_to_root(path
));
2122 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", empty_to_root(path
));
2127 static int unit_update_cgroup(
2129 CGroupMask target_mask
,
2130 CGroupMask enable_mask
,
2131 ManagerState state
) {
2133 bool created
, is_root_slice
;
2134 CGroupMask migrate_mask
= 0;
2135 _cleanup_free_
char *cgroup_full_path
= NULL
;
2140 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2143 /* Figure out our cgroup path */
2144 r
= unit_pick_cgroup_path(u
);
2148 /* First, create our own group */
2149 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
2151 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", empty_to_root(u
->cgroup_path
));
2154 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
2155 uint64_t cgroup_id
= 0;
2157 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, NULL
, &cgroup_full_path
);
2159 r
= cg_path_get_cgroupid(cgroup_full_path
, &cgroup_id
);
2161 log_unit_full_errno(u
, ERRNO_IS_NOT_SUPPORTED(r
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2162 "Failed to get cgroup ID of cgroup %s, ignoring: %m", cgroup_full_path
);
2164 log_unit_warning_errno(u
, r
, "Failed to get full cgroup path on cgroup %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2166 u
->cgroup_id
= cgroup_id
;
2169 /* Start watching it */
2170 (void) unit_watch_cgroup(u
);
2171 (void) unit_watch_cgroup_memory(u
);
2173 /* For v2 we preserve enabled controllers in delegated units, adjust others,
2174 * for v1 we figure out which controller hierarchies need migration. */
2175 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
2176 CGroupMask result_mask
= 0;
2178 /* Enable all controllers we need */
2179 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
2181 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2183 /* Remember what's actually enabled now */
2184 u
->cgroup_enabled_mask
= result_mask
;
2186 migrate_mask
= u
->cgroup_realized_mask
^ target_mask
;
2189 /* Keep track that this is now realized */
2190 u
->cgroup_realized
= true;
2191 u
->cgroup_realized_mask
= target_mask
;
2193 /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling).
2195 * Unnecessary controller cgroups are trimmed (after emptied by upward migration).
2196 * We perform migration also with whole slices for cases when users don't care about leave
2197 * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing
2200 if (cg_all_unified() == 0) {
2201 r
= cg_migrate_v1_controllers(u
->manager
->cgroup_supported
, migrate_mask
, u
->cgroup_path
, migrate_callback
, u
);
2203 log_unit_warning_errno(u
, r
, "Failed to migrate controller cgroups from %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2205 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2206 r
= cg_trim_v1_controllers(u
->manager
->cgroup_supported
, ~target_mask
, u
->cgroup_path
, !is_root_slice
);
2208 log_unit_warning_errno(u
, r
, "Failed to delete controller cgroups %s, ignoring: %m", empty_to_root(u
->cgroup_path
));
2211 /* Set attributes */
2212 cgroup_context_apply(u
, target_mask
, state
);
2213 cgroup_xattr_apply(u
);
2218 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
2219 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2225 if (MANAGER_IS_SYSTEM(u
->manager
))
2228 if (!u
->manager
->system_bus
)
2231 if (!u
->cgroup_path
)
2234 /* Determine this unit's cgroup path relative to our cgroup root */
2235 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
2239 pp
= strjoina("/", pp
, suffix_path
);
2242 r
= sd_bus_call_method(u
->manager
->system_bus
,
2243 "org.freedesktop.systemd1",
2244 "/org/freedesktop/systemd1",
2245 "org.freedesktop.systemd1.Manager",
2246 "AttachProcessesToUnit",
2249 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
2251 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
2256 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
2257 CGroupMask delegated_mask
;
2264 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2267 if (set_isempty(pids
))
2270 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
2271 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
2272 r
= bpf_firewall_load_custom(u
);
2276 r
= unit_realize_cgroup(u
);
2280 if (isempty(suffix_path
))
2283 p
= prefix_roota(u
->cgroup_path
, suffix_path
);
2285 delegated_mask
= unit_get_delegate_mask(u
);
2288 SET_FOREACH(pidp
, pids
) {
2289 pid_t pid
= PTR_TO_PID(pidp
);
2291 /* First, attach the PID to the main cgroup hierarchy */
2292 r
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
);
2294 bool again
= MANAGER_IS_USER(u
->manager
) && ERRNO_IS_PRIVILEGE(r
);
2296 log_unit_full_errno(u
, again
? LOG_DEBUG
: LOG_INFO
, r
,
2297 "Couldn't move process "PID_FMT
" to%s requested cgroup '%s': %m",
2298 pid
, again
? " directly" : "", empty_to_root(p
));
2303 /* If we are in a user instance, and we can't move the process ourselves due
2304 * to permission problems, let's ask the system instance about it instead.
2305 * Since it's more privileged it might be able to move the process across the
2306 * leaves of a subtree whose top node is not owned by us. */
2308 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
, suffix_path
);
2310 log_unit_info_errno(u
, z
, "Couldn't move process "PID_FMT
" to requested cgroup '%s' (directly or via the system bus): %m", pid
, empty_to_root(p
));
2313 ret
++; /* Count successful additions */
2314 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
2319 ret
= r
; /* Remember first error */
2322 } else if (ret
>= 0)
2323 ret
++; /* Count successful additions */
2325 r
= cg_all_unified();
2331 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
2332 * innermost realized one */
2334 for (CGroupController c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2335 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2336 const char *realized
;
2338 if (!(u
->manager
->cgroup_supported
& bit
))
2341 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
2342 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
2343 r
= cg_attach(cgroup_controller_to_string(c
), p
, pid
);
2345 continue; /* Success! */
2347 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",
2348 pid
, empty_to_root(p
), cgroup_controller_to_string(c
));
2351 /* So this controller is either not delegate or realized, or something else weird happened. In
2352 * that case let's attach the PID at least to the closest cgroup up the tree that is
2354 realized
= unit_get_realized_cgroup_path(u
, bit
);
2356 continue; /* Not even realized in the root slice? Then let's not bother */
2358 r
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
);
2360 log_unit_debug_errno(u
, r
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
2361 pid
, realized
, cgroup_controller_to_string(c
));
2368 static bool unit_has_mask_realized(
2370 CGroupMask target_mask
,
2371 CGroupMask enable_mask
) {
2375 /* Returns true if this unit is fully realized. We check four things:
2377 * 1. Whether the cgroup was created at all
2378 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
2379 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
2380 * 4. Whether the invalidation mask is currently zero
2382 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
2383 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
2384 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
2385 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
2386 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
2387 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
2388 * simply don't matter. */
2390 return u
->cgroup_realized
&&
2391 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
2392 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
2393 u
->cgroup_invalidated_mask
== 0;
2396 static bool unit_has_mask_disables_realized(
2398 CGroupMask target_mask
,
2399 CGroupMask enable_mask
) {
2403 /* Returns true if all controllers which should be disabled are indeed disabled.
2405 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
2406 * already removed. */
2408 return !u
->cgroup_realized
||
2409 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
2410 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
2413 static bool unit_has_mask_enables_realized(
2415 CGroupMask target_mask
,
2416 CGroupMask enable_mask
) {
2420 /* Returns true if all controllers which should be enabled are indeed enabled.
2422 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
2423 * we want to add is already added. */
2425 return u
->cgroup_realized
&&
2426 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
2427 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
2430 static void unit_add_to_cgroup_realize_queue(Unit
*u
) {
2433 if (u
->in_cgroup_realize_queue
)
2436 LIST_APPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2437 u
->in_cgroup_realize_queue
= true;
2440 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
2443 if (!u
->in_cgroup_realize_queue
)
2446 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2447 u
->in_cgroup_realize_queue
= false;
2450 /* Controllers can only be enabled breadth-first, from the root of the
2451 * hierarchy downwards to the unit in question. */
2452 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
2453 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2459 /* First go deal with this unit's parent, or we won't be able to enable
2460 * any new controllers at this layer. */
2461 slice
= UNIT_GET_SLICE(u
);
2463 r
= unit_realize_cgroup_now_enable(slice
, state
);
2468 target_mask
= unit_get_target_mask(u
);
2469 enable_mask
= unit_get_enable_mask(u
);
2471 /* We can only enable in this direction, don't try to disable anything.
2473 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
2476 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
2477 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
2479 return unit_update_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
2482 /* Controllers can only be disabled depth-first, from the leaves of the
2483 * hierarchy upwards to the unit in question. */
2484 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
2489 if (u
->type
!= UNIT_SLICE
)
2492 UNIT_FOREACH_DEPENDENCY(m
, u
, UNIT_ATOM_SLICE_OF
) {
2493 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2496 /* The cgroup for this unit might not actually be fully realised yet, in which case it isn't
2497 * holding any controllers open anyway. */
2498 if (!m
->cgroup_realized
)
2501 /* We must disable those below us first in order to release the controller. */
2502 if (m
->type
== UNIT_SLICE
)
2503 (void) unit_realize_cgroup_now_disable(m
, state
);
2505 target_mask
= unit_get_target_mask(m
);
2506 enable_mask
= unit_get_enable_mask(m
);
2508 /* We can only disable in this direction, don't try to enable anything. */
2509 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
2512 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
2513 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
2515 r
= unit_update_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
2523 /* Check if necessary controllers and attributes for a unit are in place.
2525 * - If so, do nothing.
2526 * - If not, create paths, move processes over, and set attributes.
2528 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
2529 * a depth-first way. As such the process looks like this:
2531 * Suppose we have a cgroup hierarchy which looks like this:
2544 * 1. We want to realise cgroup "d" now.
2545 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
2546 * 3. cgroup "k" just started requesting the memory controller.
2548 * To make this work we must do the following in order:
2550 * 1. Disable CPU controller in k, j
2551 * 2. Disable CPU controller in d
2552 * 3. Enable memory controller in root
2553 * 4. Enable memory controller in a
2554 * 5. Enable memory controller in d
2555 * 6. Enable memory controller in k
2557 * Notice that we need to touch j in one direction, but not the other. We also
2558 * don't go beyond d when disabling -- it's up to "a" to get realized if it
2559 * wants to disable further. The basic rules are therefore:
2561 * - If you're disabling something, you need to realise all of the cgroups from
2562 * your recursive descendants to the root. This starts from the leaves.
2563 * - If you're enabling something, you need to realise from the root cgroup
2564 * downwards, but you don't need to iterate your recursive descendants.
2566 * Returns 0 on success and < 0 on failure. */
2567 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
2568 CGroupMask target_mask
, enable_mask
;
2574 unit_remove_from_cgroup_realize_queue(u
);
2576 target_mask
= unit_get_target_mask(u
);
2577 enable_mask
= unit_get_enable_mask(u
);
2579 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
2582 /* Disable controllers below us, if there are any */
2583 r
= unit_realize_cgroup_now_disable(u
, state
);
2587 /* Enable controllers above us, if there are any */
2588 slice
= UNIT_GET_SLICE(u
);
2590 r
= unit_realize_cgroup_now_enable(slice
, state
);
2595 /* Now actually deal with the cgroup we were trying to realise and set attributes */
2596 r
= unit_update_cgroup(u
, target_mask
, enable_mask
, state
);
2600 /* Now, reset the invalidation mask */
2601 u
->cgroup_invalidated_mask
= 0;
2605 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
2613 state
= manager_state(m
);
2615 while ((i
= m
->cgroup_realize_queue
)) {
2616 assert(i
->in_cgroup_realize_queue
);
2618 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
2619 /* Maybe things changed, and the unit is not actually active anymore? */
2620 unit_remove_from_cgroup_realize_queue(i
);
2624 r
= unit_realize_cgroup_now(i
, state
);
2626 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
2634 void unit_add_family_to_cgroup_realize_queue(Unit
*u
) {
2636 assert(u
->type
== UNIT_SLICE
);
2638 /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all
2641 * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes
2642 * very weird if two units that own processes reside in the same slice, but one is realized in the
2643 * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does
2644 * not), because that means individual processes need to be scheduled against whole cgroups. Let's
2645 * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1
2646 * controller hierarchies too (if unit requires the controller to be realized).
2648 * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date
2654 /* Children of u likely changed when we're called */
2655 u
->cgroup_members_mask_valid
= false;
2657 UNIT_FOREACH_DEPENDENCY(m
, u
, UNIT_ATOM_SLICE_OF
) {
2659 /* No point in doing cgroup application for units without active processes. */
2660 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
2663 /* We only enqueue siblings if they were realized once at least, in the main
2665 if (!m
->cgroup_realized
)
2668 /* If the unit doesn't need any new controllers and has current ones
2669 * realized, it doesn't need any changes. */
2670 if (unit_has_mask_realized(m
,
2671 unit_get_target_mask(m
),
2672 unit_get_enable_mask(m
)))
2675 unit_add_to_cgroup_realize_queue(m
);
2678 /* Parent comes after children */
2679 unit_add_to_cgroup_realize_queue(u
);
2681 u
= UNIT_GET_SLICE(u
);
2685 int unit_realize_cgroup(Unit
*u
) {
2690 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2693 /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all
2694 * parents, but there's more actually: for the weight-based controllers we also need to make sure
2695 * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the
2696 * other hand, when a controller is removed from realized set, it may become unnecessary in siblings
2697 * and ancestors and they should be (de)realized too.
2699 * This call will defer work on the siblings and derealized ancestors to the next event loop
2700 * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */
2702 slice
= UNIT_GET_SLICE(u
);
2704 unit_add_family_to_cgroup_realize_queue(slice
);
2706 /* And realize this one now (and apply the values) */
2707 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
2710 void unit_release_cgroup(Unit
*u
) {
2713 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
2714 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
2716 if (u
->cgroup_path
) {
2717 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
2718 u
->cgroup_path
= mfree(u
->cgroup_path
);
2721 if (u
->cgroup_control_inotify_wd
>= 0) {
2722 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
2723 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
);
2725 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
2726 u
->cgroup_control_inotify_wd
= -1;
2729 if (u
->cgroup_memory_inotify_wd
>= 0) {
2730 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
2731 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
);
2733 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
2734 u
->cgroup_memory_inotify_wd
= -1;
2738 bool unit_maybe_release_cgroup(Unit
*u
) {
2743 if (!u
->cgroup_path
)
2746 /* Don't release the cgroup if there are still processes under it. If we get notified later when all the
2747 * processes exit (e.g. the processes were in D-state and exited after the unit was marked as failed)
2748 * we need the cgroup paths to continue to be tracked by the manager so they can be looked up and cleaned
2750 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2752 log_unit_debug_errno(u
, r
, "Error checking if the cgroup is recursively empty, ignoring: %m");
2754 unit_release_cgroup(u
);
2761 void unit_prune_cgroup(Unit
*u
) {
2767 /* Removes the cgroup, if empty and possible, and stops watching it. */
2769 if (!u
->cgroup_path
)
2772 (void) unit_get_cpu_usage(u
, NULL
); /* Cache the last CPU usage value before we destroy the cgroup */
2775 (void) lsm_bpf_cleanup(u
); /* Remove cgroup from the global LSM BPF map */
2778 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2780 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
2782 /* One reason we could have failed here is, that the cgroup still contains a process.
2783 * However, if the cgroup becomes removable at a later time, it might be removed when
2784 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
2785 * that the cgroup is still realized the next time it is started. Do not return early
2786 * on error, continue cleanup. */
2787 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
));
2792 if (!unit_maybe_release_cgroup(u
)) /* Returns true if the cgroup was released */
2795 u
->cgroup_realized
= false;
2796 u
->cgroup_realized_mask
= 0;
2797 u
->cgroup_enabled_mask
= 0;
2799 u
->bpf_device_control_installed
= bpf_program_free(u
->bpf_device_control_installed
);
2802 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
2803 _cleanup_fclose_
FILE *f
= NULL
;
2804 pid_t pid
= 0, npid
;
2810 if (!u
->cgroup_path
)
2813 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
2817 while (cg_read_pid(f
, &npid
) > 0) {
2822 if (pid_is_my_child(npid
) == 0)
2826 /* Dang, there's more than one daemonized PID
2827 in this group, so we don't know what process
2828 is the main process. */
2839 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
2840 _cleanup_closedir_
DIR *d
= NULL
;
2841 _cleanup_fclose_
FILE *f
= NULL
;
2847 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
2853 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
2854 r
= unit_watch_pid(u
, pid
, false);
2855 if (r
< 0 && ret
>= 0)
2859 if (r
< 0 && ret
>= 0)
2863 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
2870 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
2871 _cleanup_free_
char *p
= NULL
;
2873 p
= path_join(empty_to_root(path
), fn
);
2879 r
= unit_watch_pids_in_path(u
, p
);
2880 if (r
< 0 && ret
>= 0)
2884 if (r
< 0 && ret
>= 0)
2891 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
2896 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
2897 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
2898 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
2900 if (!u
->cgroup_path
)
2903 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2906 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
2909 if (!set_isempty(u
->pids
))
2912 unit_add_to_cgroup_empty_queue(u
);
2916 int unit_watch_all_pids(Unit
*u
) {
2921 /* Adds all PIDs from our cgroup to the set of PIDs we
2922 * watch. This is a fallback logic for cases where we do not
2923 * get reliable cgroup empty notifications: we try to use
2924 * SIGCHLD as replacement. */
2926 if (!u
->cgroup_path
)
2929 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2932 if (r
> 0) /* On unified we can use proper notifications */
2935 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
2938 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
2939 Manager
*m
= userdata
;
2946 u
= m
->cgroup_empty_queue
;
2950 assert(u
->in_cgroup_empty_queue
);
2951 u
->in_cgroup_empty_queue
= false;
2952 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
2954 if (m
->cgroup_empty_queue
) {
2955 /* More stuff queued, let's make sure we remain enabled */
2956 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2958 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
2961 unit_add_to_gc_queue(u
);
2963 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
2964 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
2969 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
2974 /* Note that there are four different ways how cgroup empty events reach us:
2976 * 1. On the unified hierarchy we get an inotify event on the cgroup
2978 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2980 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2982 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2983 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2985 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2986 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2987 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2988 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2989 * case for scope units). */
2991 if (u
->in_cgroup_empty_queue
)
2994 /* Let's verify that the cgroup is really empty */
2995 if (!u
->cgroup_path
)
2998 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
3000 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", empty_to_root(u
->cgroup_path
));
3006 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
3007 u
->in_cgroup_empty_queue
= true;
3009 /* Trigger the defer event */
3010 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
3012 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
3015 static void unit_remove_from_cgroup_empty_queue(Unit
*u
) {
3018 if (!u
->in_cgroup_empty_queue
)
3021 LIST_REMOVE(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
3022 u
->in_cgroup_empty_queue
= false;
3025 int unit_check_oomd_kill(Unit
*u
) {
3026 _cleanup_free_
char *value
= NULL
;
3031 if (!u
->cgroup_path
)
3034 r
= cg_all_unified();
3036 return log_unit_debug_errno(u
, r
, "Couldn't determine whether we are in all unified mode: %m");
3040 r
= cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "user.oomd_kill", &value
);
3041 if (r
< 0 && r
!= -ENODATA
)
3044 if (!isempty(value
)) {
3045 r
= safe_atou64(value
, &n
);
3050 increased
= n
> u
->managed_oom_kill_last
;
3051 u
->managed_oom_kill_last
= n
;
3057 log_unit_struct(u
, LOG_NOTICE
,
3058 "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR
,
3059 LOG_UNIT_INVOCATION_ID(u
),
3060 LOG_UNIT_MESSAGE(u
, "systemd-oomd killed %"PRIu64
" process(es) in this unit.", n
));
3065 int unit_check_oom(Unit
*u
) {
3066 _cleanup_free_
char *oom_kill
= NULL
;
3071 if (!u
->cgroup_path
)
3074 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
3075 if (IN_SET(r
, -ENOENT
, -ENXIO
)) /* Handle gracefully if cgroup or oom_kill attribute don't exist */
3078 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
3080 r
= safe_atou64(oom_kill
, &c
);
3082 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
3085 increased
= c
> u
->oom_kill_last
;
3086 u
->oom_kill_last
= c
;
3091 log_unit_struct(u
, LOG_NOTICE
,
3092 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
3093 LOG_UNIT_INVOCATION_ID(u
),
3094 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
3096 if (UNIT_VTABLE(u
)->notify_cgroup_oom
)
3097 UNIT_VTABLE(u
)->notify_cgroup_oom(u
);
3102 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
3103 Manager
*m
= userdata
;
3110 u
= m
->cgroup_oom_queue
;
3114 assert(u
->in_cgroup_oom_queue
);
3115 u
->in_cgroup_oom_queue
= false;
3116 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
3118 if (m
->cgroup_oom_queue
) {
3119 /* More stuff queued, let's make sure we remain enabled */
3120 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
3122 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
3125 (void) unit_check_oom(u
);
3129 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
3134 if (u
->in_cgroup_oom_queue
)
3136 if (!u
->cgroup_path
)
3139 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
3140 u
->in_cgroup_oom_queue
= true;
3142 /* Trigger the defer event */
3143 if (!u
->manager
->cgroup_oom_event_source
) {
3144 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
3146 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
3148 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
3152 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
3154 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
3158 (void) sd_event_source_set_description(s
, "cgroup-oom");
3159 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
3162 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
3164 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
3167 static int unit_check_cgroup_events(Unit
*u
) {
3168 char *values
[2] = {};
3173 if (!u
->cgroup_path
)
3176 r
= cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events",
3177 STRV_MAKE("populated", "frozen"), values
);
3181 /* The cgroup.events notifications can be merged together so act as we saw the given state for the
3182 * first time. The functions we call to handle given state are idempotent, which makes them
3183 * effectively remember the previous state. */
3185 if (streq(values
[0], "1"))
3186 unit_remove_from_cgroup_empty_queue(u
);
3188 unit_add_to_cgroup_empty_queue(u
);
3191 /* Disregard freezer state changes due to operations not initiated by us */
3192 if (values
[1] && IN_SET(u
->freezer_state
, FREEZER_FREEZING
, FREEZER_THAWING
)) {
3193 if (streq(values
[1], "0"))
3205 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
3206 Manager
*m
= userdata
;
3213 union inotify_event_buffer buffer
;
3214 struct inotify_event
*e
;
3217 l
= read(fd
, &buffer
, sizeof(buffer
));
3219 if (ERRNO_IS_TRANSIENT(errno
))
3222 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
3225 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
3229 /* Queue overflow has no watch descriptor */
3232 if (e
->mask
& IN_IGNORED
)
3233 /* The watch was just removed */
3236 /* Note that inotify might deliver events for a watch even after it was removed,
3237 * because it was queued before the removal. Let's ignore this here safely. */
3239 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
3241 unit_check_cgroup_events(u
);
3243 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
3245 unit_add_to_cgroup_oom_queue(u
);
3250 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
3251 CGroupMask mask
= 0;
3254 /* BPF-based firewall */
3255 r
= bpf_firewall_supported();
3259 mask
|= CGROUP_MASK_BPF_FIREWALL
;
3261 /* BPF-based device access control */
3262 r
= bpf_devices_supported();
3266 mask
|= CGROUP_MASK_BPF_DEVICES
;
3268 /* BPF pinned prog */
3269 r
= bpf_foreign_supported();
3273 mask
|= CGROUP_MASK_BPF_FOREIGN
;
3275 /* BPF-based bind{4|6} hooks */
3276 r
= bpf_socket_bind_supported();
3280 mask
|= CGROUP_MASK_BPF_SOCKET_BIND
;
3282 /* BPF-based cgroup_skb/{egress|ingress} hooks */
3283 r
= restrict_network_interfaces_supported();
3287 mask
|= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES
;
3293 int manager_setup_cgroup(Manager
*m
) {
3294 _cleanup_free_
char *path
= NULL
;
3295 const char *scope_path
;
3302 /* 1. Determine hierarchy */
3303 m
->cgroup_root
= mfree(m
->cgroup_root
);
3304 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
3306 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
3308 /* Chop off the init scope, if we are already located in it */
3309 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
3311 /* LEGACY: Also chop off the system slice if we are in
3312 * it. This is to support live upgrades from older systemd
3313 * versions where PID 1 was moved there. Also see
3314 * cg_get_root_path(). */
3315 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
3316 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
3318 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
3323 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
3324 * easily prepend it everywhere. */
3325 delete_trailing_chars(m
->cgroup_root
, "/");
3328 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
3330 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
3334 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
3336 all_unified
= cg_all_unified();
3337 if (all_unified
< 0)
3338 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
3339 if (all_unified
> 0)
3340 log_debug("Unified cgroup hierarchy is located at %s.", path
);
3342 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
3344 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
3346 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
3348 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
3351 /* 3. Allocate cgroup empty defer event source */
3352 m
->cgroup_empty_event_source
= sd_event_source_disable_unref(m
->cgroup_empty_event_source
);
3353 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
3355 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
3357 /* Schedule cgroup empty checks early, but after having processed service notification messages or
3358 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
3359 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
3360 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
3362 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
3364 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
3366 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
3368 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
3370 /* 4. Install notifier inotify object, or agent */
3371 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
3373 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
3375 m
->cgroup_inotify_event_source
= sd_event_source_disable_unref(m
->cgroup_inotify_event_source
);
3376 safe_close(m
->cgroup_inotify_fd
);
3378 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
3379 if (m
->cgroup_inotify_fd
< 0)
3380 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
3382 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
3384 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
3386 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
3387 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
3388 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
3389 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
3391 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
3393 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
3395 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
3397 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
3398 * since it does not generate events when control groups with children run empty. */
3400 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUPS_AGENT_PATH
);
3402 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
3404 log_debug("Installed release agent.");
3406 log_debug("Release agent already installed.");
3409 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
3410 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
3411 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
3413 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
3414 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
3416 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
3418 /* 6. And pin it, so that it cannot be unmounted */
3419 safe_close(m
->pin_cgroupfs_fd
);
3420 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
3421 if (m
->pin_cgroupfs_fd
< 0)
3422 return log_error_errno(errno
, "Failed to open pin file: %m");
3424 } else if (!MANAGER_IS_TEST_RUN(m
))
3425 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
3427 /* 7. Always enable hierarchical support if it exists... */
3428 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
3429 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
3431 /* 8. Figure out which controllers are supported */
3432 r
= cg_mask_supported_subtree(m
->cgroup_root
, &m
->cgroup_supported
);
3434 return log_error_errno(r
, "Failed to determine supported controllers: %m");
3436 /* 9. Figure out which bpf-based pseudo-controllers are supported */
3437 r
= cg_bpf_mask_supported(&mask
);
3439 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
3440 m
->cgroup_supported
|= mask
;
3442 /* 10. Log which controllers are supported */
3443 for (CGroupController c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
3444 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
),
3445 yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
3450 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
3453 /* We can't really delete the group, since we are in it. But
3455 if (delete && m
->cgroup_root
&& !FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
3456 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
3458 m
->cgroup_empty_event_source
= sd_event_source_disable_unref(m
->cgroup_empty_event_source
);
3460 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
3461 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
3463 m
->cgroup_inotify_event_source
= sd_event_source_disable_unref(m
->cgroup_inotify_event_source
);
3464 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
3466 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
3468 m
->cgroup_root
= mfree(m
->cgroup_root
);
3471 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
3478 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
3482 p
= strdupa_safe(cgroup
);
3486 e
= strrchr(p
, '/');
3488 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
3492 u
= hashmap_get(m
->cgroup_unit
, p
);
3498 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
3499 _cleanup_free_
char *cgroup
= NULL
;
3503 if (!pid_is_valid(pid
))
3506 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
3509 return manager_get_unit_by_cgroup(m
, cgroup
);
3512 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
3517 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
3518 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
3519 * relevant one as children of the process will be assigned to that one, too, before all else. */
3521 if (!pid_is_valid(pid
))
3524 if (pid
== getpid_cached())
3525 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
3527 u
= manager_get_unit_by_pid_cgroup(m
, pid
);
3531 u
= hashmap_get(m
->watch_pids
, PID_TO_PTR(pid
));
3535 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-pid
));
3542 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
3548 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
3549 * or from the --system instance */
3551 log_debug("Got cgroup empty notification for: %s", cgroup
);
3553 u
= manager_get_unit_by_cgroup(m
, cgroup
);
3557 unit_add_to_cgroup_empty_queue(u
);
3561 int unit_get_memory_available(Unit
*u
, uint64_t *ret
) {
3562 uint64_t unit_current
, available
= UINT64_MAX
;
3563 CGroupContext
*unit_context
;
3564 const char *memory_file
;
3570 /* If data from cgroups can be accessed, try to find out how much more memory a unit can
3571 * claim before hitting the configured cgroup limits (if any). Consider both MemoryHigh
3572 * and MemoryMax, and also any slice the unit might be nested below. */
3574 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3577 if (!u
->cgroup_path
)
3580 /* The root cgroup doesn't expose this information */
3581 if (unit_has_host_root_cgroup(u
))
3584 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3587 r
= cg_all_unified();
3590 memory_file
= r
> 0 ? "memory.current" : "memory.usage_in_bytes";
3592 r
= cg_get_attribute_as_uint64("memory", u
->cgroup_path
, memory_file
, &unit_current
);
3596 assert_se(unit_context
= unit_get_cgroup_context(u
));
3598 if (unit_context
->memory_max
!= UINT64_MAX
|| unit_context
->memory_high
!= UINT64_MAX
)
3599 available
= LESS_BY(MIN(unit_context
->memory_max
, unit_context
->memory_high
), unit_current
);
3601 for (Unit
*slice
= UNIT_GET_SLICE(u
); slice
; slice
= UNIT_GET_SLICE(slice
)) {
3602 uint64_t slice_current
, slice_available
= UINT64_MAX
;
3603 CGroupContext
*slice_context
;
3605 /* No point in continuing if we can't go any lower */
3609 if (!slice
->cgroup_path
)
3612 slice_context
= unit_get_cgroup_context(slice
);
3616 if (slice_context
->memory_max
== UINT64_MAX
&& slice_context
->memory_high
== UINT64_MAX
)
3619 r
= cg_get_attribute_as_uint64("memory", slice
->cgroup_path
, memory_file
, &slice_current
);
3623 slice_available
= LESS_BY(MIN(slice_context
->memory_max
, slice_context
->memory_high
), slice_current
);
3624 available
= MIN(slice_available
, available
);
3632 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
3638 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3641 if (!u
->cgroup_path
)
3644 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3645 if (unit_has_host_root_cgroup(u
))
3646 return procfs_memory_get_used(ret
);
3648 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3651 r
= cg_all_unified();
3655 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, r
> 0 ? "memory.current" : "memory.usage_in_bytes", ret
);
3658 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
3662 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
3665 if (!u
->cgroup_path
)
3668 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3669 if (unit_has_host_root_cgroup(u
))
3670 return procfs_tasks_get_current(ret
);
3672 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
3675 return cg_get_attribute_as_uint64("pids", u
->cgroup_path
, "pids.current", ret
);
3678 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
3685 if (!u
->cgroup_path
)
3688 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3689 if (unit_has_host_root_cgroup(u
))
3690 return procfs_cpu_get_usage(ret
);
3692 /* Requisite controllers for CPU accounting are not enabled */
3693 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
3696 r
= cg_all_unified();
3700 _cleanup_free_
char *val
= NULL
;
3703 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
3704 if (IN_SET(r
, -ENOENT
, -ENXIO
))
3709 r
= safe_atou64(val
, &us
);
3713 ns
= us
* NSEC_PER_USEC
;
3715 return cg_get_attribute_as_uint64("cpuacct", u
->cgroup_path
, "cpuacct.usage", ret
);
3721 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
3727 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
3728 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
3729 * call this function with a NULL return value. */
3731 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
3734 r
= unit_get_cpu_usage_raw(u
, &ns
);
3735 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
3736 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
3740 *ret
= u
->cpu_usage_last
;
3746 if (ns
> u
->cpu_usage_base
)
3747 ns
-= u
->cpu_usage_base
;
3751 u
->cpu_usage_last
= ns
;
3758 int unit_get_ip_accounting(
3760 CGroupIPAccountingMetric metric
,
3767 assert(metric
>= 0);
3768 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
3771 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
3774 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
3775 u
->ip_accounting_ingress_map_fd
:
3776 u
->ip_accounting_egress_map_fd
;
3780 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
3781 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
3783 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
3787 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
3788 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
3789 * ip_accounting_extra[] field, and add them in here transparently. */
3791 *ret
= value
+ u
->ip_accounting_extra
[metric
];
3796 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
3797 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
3798 [CGROUP_IO_READ_BYTES
] = "rbytes=",
3799 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
3800 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
3801 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
3803 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
3804 _cleanup_free_
char *path
= NULL
;
3805 _cleanup_fclose_
FILE *f
= NULL
;
3810 if (!u
->cgroup_path
)
3813 if (unit_has_host_root_cgroup(u
))
3814 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
3816 r
= cg_all_unified();
3819 if (r
== 0) /* TODO: support cgroupv1 */
3822 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
3825 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
3829 f
= fopen(path
, "re");
3834 _cleanup_free_
char *line
= NULL
;
3837 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3844 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
3845 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
3848 _cleanup_free_
char *word
= NULL
;
3850 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
3856 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3859 x
= startswith(word
, field_names
[i
]);
3863 r
= safe_atou64(x
, &w
);
3867 /* Sum up the stats of all devices */
3875 memcpy(ret
, acc
, sizeof(acc
));
3879 int unit_get_io_accounting(
3881 CGroupIOAccountingMetric metric
,
3885 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
3888 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
3890 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
3893 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3896 r
= unit_get_io_accounting_raw(u
, raw
);
3897 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3902 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3903 /* Saturated subtraction */
3904 if (raw
[i
] > u
->io_accounting_base
[i
])
3905 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
3907 u
->io_accounting_last
[i
] = 0;
3912 *ret
= u
->io_accounting_last
[metric
];
3917 int unit_reset_cpu_accounting(Unit
*u
) {
3922 u
->cpu_usage_last
= NSEC_INFINITY
;
3924 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
3926 u
->cpu_usage_base
= 0;
3933 int unit_reset_ip_accounting(Unit
*u
) {
3938 if (u
->ip_accounting_ingress_map_fd
>= 0)
3939 r
= bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
);
3941 if (u
->ip_accounting_egress_map_fd
>= 0)
3942 q
= bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
);
3944 zero(u
->ip_accounting_extra
);
3946 return r
< 0 ? r
: q
;
3949 int unit_reset_io_accounting(Unit
*u
) {
3954 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++)
3955 u
->io_accounting_last
[i
] = UINT64_MAX
;
3957 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
3959 zero(u
->io_accounting_base
);
3966 int unit_reset_accounting(Unit
*u
) {
3971 r
= unit_reset_cpu_accounting(u
);
3972 q
= unit_reset_io_accounting(u
);
3973 v
= unit_reset_ip_accounting(u
);
3975 return r
< 0 ? r
: q
< 0 ? q
: v
;
3978 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
3981 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3987 /* always invalidate compat pairs together */
3988 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
3989 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
3991 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
3992 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
3994 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
3997 u
->cgroup_invalidated_mask
|= m
;
3998 unit_add_to_cgroup_realize_queue(u
);
4001 void unit_invalidate_cgroup_bpf(Unit
*u
) {
4004 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4007 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
4010 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
4011 unit_add_to_cgroup_realize_queue(u
);
4013 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
4014 * list of our children includes our own. */
4015 if (u
->type
== UNIT_SLICE
) {
4018 UNIT_FOREACH_DEPENDENCY(member
, u
, UNIT_ATOM_SLICE_OF
)
4019 unit_invalidate_cgroup_bpf(member
);
4023 void unit_cgroup_catchup(Unit
*u
) {
4026 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
4029 /* We dropped the inotify watch during reexec/reload, so we need to
4030 * check these as they may have changed.
4031 * Note that (currently) the kernel doesn't actually update cgroup
4032 * file modification times, so we can't just serialize and then check
4033 * the mtime for file(s) we are interested in. */
4034 (void) unit_check_cgroup_events(u
);
4035 unit_add_to_cgroup_oom_queue(u
);
4038 bool unit_cgroup_delegate(Unit
*u
) {
4043 if (!UNIT_VTABLE(u
)->can_delegate
)
4046 c
= unit_get_cgroup_context(u
);
4053 void manager_invalidate_startup_units(Manager
*m
) {
4058 SET_FOREACH(u
, m
->startup_units
)
4059 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
|CGROUP_MASK_CPUSET
);
4062 static int unit_get_nice(Unit
*u
) {
4065 ec
= unit_get_exec_context(u
);
4066 return ec
? ec
->nice
: 0;
4069 static uint64_t unit_get_cpu_weight(Unit
*u
) {
4070 ManagerState state
= manager_state(u
->manager
);
4073 cc
= unit_get_cgroup_context(u
);
4074 return cc
? cgroup_context_cpu_weight(cc
, state
) : CGROUP_WEIGHT_DEFAULT
;
4077 int compare_job_priority(const void *a
, const void *b
) {
4078 const Job
*x
= a
, *y
= b
;
4080 uint64_t weight_x
, weight_y
;
4083 if ((ret
= CMP(x
->unit
->type
, y
->unit
->type
)) != 0)
4086 weight_x
= unit_get_cpu_weight(x
->unit
);
4087 weight_y
= unit_get_cpu_weight(y
->unit
);
4089 if ((ret
= CMP(weight_x
, weight_y
)) != 0)
4092 nice_x
= unit_get_nice(x
->unit
);
4093 nice_y
= unit_get_nice(y
->unit
);
4095 if ((ret
= CMP(nice_x
, nice_y
)) != 0)
4098 return strcmp(x
->unit
->id
, y
->unit
->id
);
4101 int unit_cgroup_freezer_action(Unit
*u
, FreezerAction action
) {
4102 _cleanup_free_
char *path
= NULL
;
4103 FreezerState target
, kernel
= _FREEZER_STATE_INVALID
;
4107 assert(IN_SET(action
, FREEZER_FREEZE
, FREEZER_THAW
));
4109 if (!cg_freezer_supported())
4112 if (!u
->cgroup_realized
)
4115 target
= action
== FREEZER_FREEZE
? FREEZER_FROZEN
: FREEZER_RUNNING
;
4117 r
= unit_freezer_state_kernel(u
, &kernel
);
4119 log_unit_debug_errno(u
, r
, "Failed to obtain cgroup freezer state: %m");
4121 if (target
== kernel
) {
4122 u
->freezer_state
= target
;
4126 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.freeze", &path
);
4130 log_unit_debug(u
, "%s unit.", action
== FREEZER_FREEZE
? "Freezing" : "Thawing");
4132 if (action
== FREEZER_FREEZE
)
4133 u
->freezer_state
= FREEZER_FREEZING
;
4135 u
->freezer_state
= FREEZER_THAWING
;
4137 r
= write_string_file(path
, one_zero(action
== FREEZER_FREEZE
), WRITE_STRING_FILE_DISABLE_BUFFER
);
4144 int unit_get_cpuset(Unit
*u
, CPUSet
*cpus
, const char *name
) {
4145 _cleanup_free_
char *v
= NULL
;
4151 if (!u
->cgroup_path
)
4154 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUSET
) == 0)
4157 r
= cg_all_unified();
4163 r
= cg_get_attribute("cpuset", u
->cgroup_path
, name
, &v
);
4169 return parse_cpu_set_full(v
, cpus
, false, NULL
, NULL
, 0, NULL
);
4172 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
4173 [CGROUP_DEVICE_POLICY_AUTO
] = "auto",
4174 [CGROUP_DEVICE_POLICY_CLOSED
] = "closed",
4175 [CGROUP_DEVICE_POLICY_STRICT
] = "strict",
4178 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);
4180 static const char* const freezer_action_table
[_FREEZER_ACTION_MAX
] = {
4181 [FREEZER_FREEZE
] = "freeze",
4182 [FREEZER_THAW
] = "thaw",
4185 DEFINE_STRING_TABLE_LOOKUP(freezer_action
, FreezerAction
);