1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
5 #include "sd-messages.h"
7 #include "alloc-util.h"
8 #include "blockdev-util.h"
9 #include "bpf-devices.h"
10 #include "bpf-firewall.h"
11 #include "btrfs-util.h"
12 #include "bus-error.h"
13 #include "cgroup-setup.h"
14 #include "cgroup-util.h"
20 #include "limits-util.h"
21 #include "nulstr-util.h"
22 #include "parse-util.h"
23 #include "path-util.h"
24 #include "percent-util.h"
25 #include "process-util.h"
26 #include "procfs-util.h"
28 #include "stat-util.h"
29 #include "stdio-util.h"
30 #include "string-table.h"
31 #include "string-util.h"
34 #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
36 /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access
37 * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask
38 * out specific attributes from us. */
39 #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING)
41 uint64_t tasks_max_resolve(const TasksMax
*tasks_max
) {
42 if (tasks_max
->scale
== 0)
43 return tasks_max
->value
;
45 return system_tasks_max_scale(tasks_max
->value
, tasks_max
->scale
);
48 bool manager_owns_host_root_cgroup(Manager
*m
) {
51 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
52 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
53 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
54 * we run in any kind of container virtualization. */
56 if (MANAGER_IS_USER(m
))
59 if (detect_container() > 0)
62 return empty_or_root(m
->cgroup_root
);
65 bool unit_has_host_root_cgroup(Unit
*u
) {
68 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
69 * the manager manages the root cgroup. */
71 if (!manager_owns_host_root_cgroup(u
->manager
))
74 return unit_has_name(u
, SPECIAL_ROOT_SLICE
);
77 static int set_attribute_and_warn(Unit
*u
, const char *controller
, const char *attribute
, const char *value
) {
80 r
= cg_set_attribute(controller
, u
->cgroup_path
, attribute
, value
);
82 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%.*s': %m",
83 strna(attribute
), isempty(u
->cgroup_path
) ? "/" : u
->cgroup_path
, (int) strcspn(value
, NEWLINE
), value
);
88 static void cgroup_compat_warn(void) {
89 static bool cgroup_compat_warned
= false;
91 if (cgroup_compat_warned
)
94 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
95 "See cgroup-compat debug messages for details.");
97 cgroup_compat_warned
= true;
100 #define log_cgroup_compat(unit, fmt, ...) do { \
101 cgroup_compat_warn(); \
102 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
105 void cgroup_context_init(CGroupContext
*c
) {
108 /* Initialize everything to the kernel defaults. */
110 *c
= (CGroupContext
) {
111 .cpu_weight
= CGROUP_WEIGHT_INVALID
,
112 .startup_cpu_weight
= CGROUP_WEIGHT_INVALID
,
113 .cpu_quota_per_sec_usec
= USEC_INFINITY
,
114 .cpu_quota_period_usec
= USEC_INFINITY
,
116 .cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
117 .startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
119 .memory_high
= CGROUP_LIMIT_MAX
,
120 .memory_max
= CGROUP_LIMIT_MAX
,
121 .memory_swap_max
= CGROUP_LIMIT_MAX
,
123 .memory_limit
= CGROUP_LIMIT_MAX
,
125 .io_weight
= CGROUP_WEIGHT_INVALID
,
126 .startup_io_weight
= CGROUP_WEIGHT_INVALID
,
128 .blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
129 .startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
131 .tasks_max
= TASKS_MAX_UNSET
,
133 .moom_swap
= MANAGED_OOM_AUTO
,
134 .moom_mem_pressure
= MANAGED_OOM_AUTO
,
135 .moom_preference
= MANAGED_OOM_PREFERENCE_NONE
,
139 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
143 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
148 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
152 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
157 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
161 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
166 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
170 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
175 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
179 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
184 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
188 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
193 void cgroup_context_done(CGroupContext
*c
) {
196 while (c
->io_device_weights
)
197 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
199 while (c
->io_device_latencies
)
200 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
202 while (c
->io_device_limits
)
203 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
205 while (c
->blockio_device_weights
)
206 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
208 while (c
->blockio_device_bandwidths
)
209 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
211 while (c
->device_allow
)
212 cgroup_context_free_device_allow(c
, c
->device_allow
);
214 c
->ip_address_allow
= ip_address_access_free_all(c
->ip_address_allow
);
215 c
->ip_address_deny
= ip_address_access_free_all(c
->ip_address_deny
);
217 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
218 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
220 cpu_set_reset(&c
->cpuset_cpus
);
221 cpu_set_reset(&c
->cpuset_mems
);
224 static int unit_get_kernel_memory_limit(Unit
*u
, const char *file
, uint64_t *ret
) {
227 if (!u
->cgroup_realized
)
230 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, file
, ret
);
233 static int unit_compare_memory_limit(Unit
*u
, const char *property_name
, uint64_t *ret_unit_value
, uint64_t *ret_kernel_value
) {
240 /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will
241 * return -ENODATA) on cgroup v1.
246 * 0: If the kernel memory setting doesn't match our configuration.
247 * >0: If the kernel memory setting matches our configuration.
249 * The following values are only guaranteed to be populated on return >=0:
251 * - ret_unit_value will contain our internal expected value for the unit, page-aligned.
252 * - ret_kernel_value will contain the actual value presented by the kernel. */
256 r
= cg_all_unified();
258 return log_debug_errno(r
, "Failed to determine cgroup hierarchy version: %m");
260 /* Unsupported on v1.
262 * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has
263 * silently masked the controller. */
267 /* The root slice doesn't have any controller files, so we can't compare anything. */
268 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
271 /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled,
272 * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To
273 * avoid specious errors in these scenarios, check that we even expect the memory controller to be
275 m
= unit_get_target_mask(u
);
276 if (!FLAGS_SET(m
, CGROUP_MASK_MEMORY
))
279 c
= unit_get_cgroup_context(u
);
282 if (streq(property_name
, "MemoryLow")) {
283 unit_value
= unit_get_ancestor_memory_low(u
);
285 } else if (streq(property_name
, "MemoryMin")) {
286 unit_value
= unit_get_ancestor_memory_min(u
);
288 } else if (streq(property_name
, "MemoryHigh")) {
289 unit_value
= c
->memory_high
;
290 file
= "memory.high";
291 } else if (streq(property_name
, "MemoryMax")) {
292 unit_value
= c
->memory_max
;
294 } else if (streq(property_name
, "MemorySwapMax")) {
295 unit_value
= c
->memory_swap_max
;
296 file
= "memory.swap.max";
300 r
= unit_get_kernel_memory_limit(u
, file
, ret_kernel_value
);
302 return log_unit_debug_errno(u
, r
, "Failed to parse %s: %m", file
);
304 /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page
305 * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from
306 * our internal page-counting. To support those future kernels, just check the value itself first
307 * without any page-alignment. */
308 if (*ret_kernel_value
== unit_value
) {
309 *ret_unit_value
= unit_value
;
313 /* The current kernel behaviour, by comparison, is that even if you write a particular number of
314 * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel
315 * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is
317 if (unit_value
!= CGROUP_LIMIT_MAX
)
318 unit_value
= PAGE_ALIGN_DOWN(unit_value
);
320 *ret_unit_value
= unit_value
;
322 return *ret_kernel_value
== *ret_unit_value
;
325 #define FORMAT_CGROUP_DIFF_MAX 128
327 static char *format_cgroup_memory_limit_comparison(char *buf
, size_t l
, Unit
*u
, const char *property_name
) {
335 r
= unit_compare_memory_limit(u
, property_name
, &sval
, &kval
);
337 /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1).
338 * In the absence of reliably being able to detect whether memcg swap support is available or not,
339 * only complain if the error is not ENOENT. */
340 if (r
> 0 || IN_SET(r
, -ENODATA
, -EOWNERDEAD
) ||
341 (r
== -ENOENT
&& streq(property_name
, "MemorySwapMax"))) {
347 snprintf(buf
, l
, " (error getting kernel value: %s)", strerror_safe(r
));
351 snprintf(buf
, l
, " (different value in kernel: %" PRIu64
")", kval
);
356 void cgroup_context_dump(Unit
*u
, FILE* f
, const char *prefix
) {
357 _cleanup_free_
char *disable_controllers_str
= NULL
, *cpuset_cpus
= NULL
, *cpuset_mems
= NULL
;
358 CGroupIODeviceLimit
*il
;
359 CGroupIODeviceWeight
*iw
;
360 CGroupIODeviceLatency
*l
;
361 CGroupBlockIODeviceBandwidth
*b
;
362 CGroupBlockIODeviceWeight
*w
;
363 CGroupDeviceAllow
*a
;
365 IPAddressAccessItem
*iaai
;
367 char q
[FORMAT_TIMESPAN_MAX
];
368 char v
[FORMAT_TIMESPAN_MAX
];
370 char cda
[FORMAT_CGROUP_DIFF_MAX
];
371 char cdb
[FORMAT_CGROUP_DIFF_MAX
];
372 char cdc
[FORMAT_CGROUP_DIFF_MAX
];
373 char cdd
[FORMAT_CGROUP_DIFF_MAX
];
374 char cde
[FORMAT_CGROUP_DIFF_MAX
];
379 c
= unit_get_cgroup_context(u
);
382 prefix
= strempty(prefix
);
384 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
386 cpuset_cpus
= cpu_set_to_range_string(&c
->cpuset_cpus
);
387 cpuset_mems
= cpu_set_to_range_string(&c
->cpuset_mems
);
390 "%sCPUAccounting: %s\n"
391 "%sIOAccounting: %s\n"
392 "%sBlockIOAccounting: %s\n"
393 "%sMemoryAccounting: %s\n"
394 "%sTasksAccounting: %s\n"
395 "%sIPAccounting: %s\n"
396 "%sCPUWeight: %" PRIu64
"\n"
397 "%sStartupCPUWeight: %" PRIu64
"\n"
398 "%sCPUShares: %" PRIu64
"\n"
399 "%sStartupCPUShares: %" PRIu64
"\n"
400 "%sCPUQuotaPerSecSec: %s\n"
401 "%sCPUQuotaPeriodSec: %s\n"
402 "%sAllowedCPUs: %s\n"
403 "%sAllowedMemoryNodes: %s\n"
404 "%sIOWeight: %" PRIu64
"\n"
405 "%sStartupIOWeight: %" PRIu64
"\n"
406 "%sBlockIOWeight: %" PRIu64
"\n"
407 "%sStartupBlockIOWeight: %" PRIu64
"\n"
408 "%sDefaultMemoryMin: %" PRIu64
"\n"
409 "%sDefaultMemoryLow: %" PRIu64
"\n"
410 "%sMemoryMin: %" PRIu64
"%s\n"
411 "%sMemoryLow: %" PRIu64
"%s\n"
412 "%sMemoryHigh: %" PRIu64
"%s\n"
413 "%sMemoryMax: %" PRIu64
"%s\n"
414 "%sMemorySwapMax: %" PRIu64
"%s\n"
415 "%sMemoryLimit: %" PRIu64
"\n"
416 "%sTasksMax: %" PRIu64
"\n"
417 "%sDevicePolicy: %s\n"
418 "%sDisableControllers: %s\n"
420 "%sManagedOOMSwap: %s\n"
421 "%sManagedOOMMemoryPressure: %s\n"
422 "%sManagedOOMMemoryPressureLimit: " PERMYRIAD_AS_PERCENT_FORMAT_STR
"\n"
423 "%sManagedOOMPreference: %s%%\n",
424 prefix
, yes_no(c
->cpu_accounting
),
425 prefix
, yes_no(c
->io_accounting
),
426 prefix
, yes_no(c
->blockio_accounting
),
427 prefix
, yes_no(c
->memory_accounting
),
428 prefix
, yes_no(c
->tasks_accounting
),
429 prefix
, yes_no(c
->ip_accounting
),
430 prefix
, c
->cpu_weight
,
431 prefix
, c
->startup_cpu_weight
,
432 prefix
, c
->cpu_shares
,
433 prefix
, c
->startup_cpu_shares
,
434 prefix
, format_timespan(q
, sizeof(q
), c
->cpu_quota_per_sec_usec
, 1),
435 prefix
, format_timespan(v
, sizeof(v
), c
->cpu_quota_period_usec
, 1),
436 prefix
, strempty(cpuset_cpus
),
437 prefix
, strempty(cpuset_mems
),
438 prefix
, c
->io_weight
,
439 prefix
, c
->startup_io_weight
,
440 prefix
, c
->blockio_weight
,
441 prefix
, c
->startup_blockio_weight
,
442 prefix
, c
->default_memory_min
,
443 prefix
, c
->default_memory_low
,
444 prefix
, c
->memory_min
, format_cgroup_memory_limit_comparison(cda
, sizeof(cda
), u
, "MemoryMin"),
445 prefix
, c
->memory_low
, format_cgroup_memory_limit_comparison(cdb
, sizeof(cdb
), u
, "MemoryLow"),
446 prefix
, c
->memory_high
, format_cgroup_memory_limit_comparison(cdc
, sizeof(cdc
), u
, "MemoryHigh"),
447 prefix
, c
->memory_max
, format_cgroup_memory_limit_comparison(cdd
, sizeof(cdd
), u
, "MemoryMax"),
448 prefix
, c
->memory_swap_max
, format_cgroup_memory_limit_comparison(cde
, sizeof(cde
), u
, "MemorySwapMax"),
449 prefix
, c
->memory_limit
,
450 prefix
, tasks_max_resolve(&c
->tasks_max
),
451 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
452 prefix
, strempty(disable_controllers_str
),
453 prefix
, yes_no(c
->delegate
),
454 prefix
, managed_oom_mode_to_string(c
->moom_swap
),
455 prefix
, managed_oom_mode_to_string(c
->moom_mem_pressure
),
456 prefix
, PERMYRIAD_AS_PERCENT_FORMAT_VAL(UINT32_SCALE_TO_PERMYRIAD(c
->moom_mem_pressure_limit
)),
457 prefix
, managed_oom_preference_to_string(c
->moom_preference
));
460 _cleanup_free_
char *t
= NULL
;
462 (void) cg_mask_to_string(c
->delegate_controllers
, &t
);
464 fprintf(f
, "%sDelegateControllers: %s\n",
469 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
471 "%sDeviceAllow: %s %s%s%s\n",
474 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
476 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
478 "%sIODeviceWeight: %s %" PRIu64
"\n",
483 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
485 "%sIODeviceLatencyTargetSec: %s %s\n",
488 format_timespan(q
, sizeof(q
), l
->target_usec
, 1));
490 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
491 char buf
[FORMAT_BYTES_MAX
];
492 CGroupIOLimitType type
;
494 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
495 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
499 cgroup_io_limit_type_to_string(type
),
501 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
504 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
506 "%sBlockIODeviceWeight: %s %" PRIu64
,
511 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
512 char buf
[FORMAT_BYTES_MAX
];
514 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
516 "%sBlockIOReadBandwidth: %s %s\n",
519 format_bytes(buf
, sizeof(buf
), b
->rbps
));
520 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
522 "%sBlockIOWriteBandwidth: %s %s\n",
525 format_bytes(buf
, sizeof(buf
), b
->wbps
));
528 LIST_FOREACH(items
, iaai
, c
->ip_address_allow
) {
529 _cleanup_free_
char *k
= NULL
;
531 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
532 fprintf(f
, "%sIPAddressAllow: %s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
535 LIST_FOREACH(items
, iaai
, c
->ip_address_deny
) {
536 _cleanup_free_
char *k
= NULL
;
538 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
539 fprintf(f
, "%sIPAddressDeny: %s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
542 STRV_FOREACH(path
, c
->ip_filters_ingress
)
543 fprintf(f
, "%sIPIngressFilterPath: %s\n", prefix
, *path
);
545 STRV_FOREACH(path
, c
->ip_filters_egress
)
546 fprintf(f
, "%sIPEgressFilterPath: %s\n", prefix
, *path
);
549 int cgroup_add_device_allow(CGroupContext
*c
, const char *dev
, const char *mode
) {
550 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
551 _cleanup_free_
char *d
= NULL
;
555 assert(isempty(mode
) || in_charset(mode
, "rwm"));
557 a
= new(CGroupDeviceAllow
, 1);
565 *a
= (CGroupDeviceAllow
) {
567 .r
= isempty(mode
) || strchr(mode
, 'r'),
568 .w
= isempty(mode
) || strchr(mode
, 'w'),
569 .m
= isempty(mode
) || strchr(mode
, 'm'),
572 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
578 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
579 uint64_t unit_get_ancestor_##entry(Unit *u) { \
582 /* 1. Is entry set in this unit? If so, use that. \
583 * 2. Is the default for this entry set in any \
584 * ancestor? If so, use that. \
585 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
589 c = unit_get_cgroup_context(u); \
590 if (c && c->entry##_set) \
593 while ((u = UNIT_DEREF(u->slice))) { \
594 c = unit_get_cgroup_context(u); \
595 if (c && c->default_##entry##_set) \
596 return c->default_##entry; \
599 /* We've reached the root, but nobody had default for \
600 * this entry set, so set it to the kernel default. */ \
601 return CGROUP_LIMIT_MIN; \
604 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
605 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
607 void cgroup_oomd_xattr_apply(Unit
*u
, const char *cgroup_path
) {
613 c
= unit_get_cgroup_context(u
);
617 if (c
->moom_preference
== MANAGED_OOM_PREFERENCE_OMIT
) {
618 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, cgroup_path
, "user.oomd_omit", "1", 1, 0);
620 log_unit_debug_errno(u
, r
, "Failed to set oomd_omit flag on control group %s, ignoring: %m", cgroup_path
);
623 if (c
->moom_preference
== MANAGED_OOM_PREFERENCE_AVOID
) {
624 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, cgroup_path
, "user.oomd_avoid", "1", 1, 0);
626 log_unit_debug_errno(u
, r
, "Failed to set oomd_avoid flag on control group %s, ignoring: %m", cgroup_path
);
629 if (c
->moom_preference
!= MANAGED_OOM_PREFERENCE_AVOID
) {
630 r
= cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER
, cgroup_path
, "user.oomd_avoid");
632 log_unit_debug_errno(u
, r
, "Failed to remove oomd_avoid flag on control group %s, ignoring: %m", cgroup_path
);
635 if (c
->moom_preference
!= MANAGED_OOM_PREFERENCE_OMIT
) {
636 r
= cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER
, cgroup_path
, "user.oomd_omit");
638 log_unit_debug_errno(u
, r
, "Failed to remove oomd_omit flag on control group %s, ignoring: %m", cgroup_path
);
642 static void cgroup_xattr_apply(Unit
*u
) {
643 char ids
[SD_ID128_STRING_MAX
];
648 if (!MANAGER_IS_SYSTEM(u
->manager
))
651 if (!sd_id128_is_null(u
->invocation_id
)) {
652 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
653 "trusted.invocation_id",
654 sd_id128_to_string(u
->invocation_id
, ids
), 32,
657 log_unit_debug_errno(u
, r
, "Failed to set invocation ID on control group %s, ignoring: %m", u
->cgroup_path
);
660 if (unit_cgroup_delegate(u
)) {
661 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
666 log_unit_debug_errno(u
, r
, "Failed to set delegate flag on control group %s, ignoring: %m", u
->cgroup_path
);
668 r
= cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "trusted.delegate");
670 log_unit_debug_errno(u
, r
, "Failed to remove delegate flag on control group %s, ignoring: %m", u
->cgroup_path
);
673 cgroup_oomd_xattr_apply(u
, u
->cgroup_path
);
676 static int lookup_block_device(const char *p
, dev_t
*ret
) {
684 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
685 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
688 if (stat(p
, &st
) < 0)
689 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
695 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
698 return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK
),
699 "Device node '%s' is a character device, but block device needed.", p
);
702 else if (major(dev
) != 0)
703 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
705 /* If this is btrfs, getting the backing block device is a bit harder */
706 r
= btrfs_get_block_device(p
, ret
);
708 return log_warning_errno(SYNTHETIC_ERRNO(ENODEV
),
709 "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
711 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
714 /* If this is a LUKS/DM device, recursively try to get the originating block device */
715 while (block_get_originating(*ret
, ret
) > 0);
717 /* If this is a partition, try to get the originating block device */
718 (void) block_get_whole_disk(*ret
, ret
);
722 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
723 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
724 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
727 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
728 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
729 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
732 static uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
733 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
734 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
735 return c
->startup_cpu_weight
;
736 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
737 return c
->cpu_weight
;
739 return CGROUP_WEIGHT_DEFAULT
;
742 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
743 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
744 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
745 return c
->startup_cpu_shares
;
746 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
747 return c
->cpu_shares
;
749 return CGROUP_CPU_SHARES_DEFAULT
;
752 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
753 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
754 * need to be higher than that boundary. quota is specified in USecPerSec.
755 * Additionally, period must be at most max_period. */
758 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
761 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
764 if (quota
== USEC_INFINITY
)
765 /* Always use default period for infinity quota. */
766 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
768 if (period
== USEC_INFINITY
)
769 /* Default period was requested. */
770 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
772 /* Clamp to interval [1ms, 1s] */
773 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
775 if (new_period
!= period
) {
776 char v
[FORMAT_TIMESPAN_MAX
];
777 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
,
778 "Clamping CPU interval for cpu.max: period is now %s",
779 format_timespan(v
, sizeof(v
), new_period
, 1));
780 u
->warned_clamping_cpu_quota_period
= true;
786 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
787 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
789 xsprintf(buf
, "%" PRIu64
"\n", weight
);
790 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
793 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
794 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
796 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
797 if (quota
!= USEC_INFINITY
)
798 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
799 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
801 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
802 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
805 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
806 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
808 xsprintf(buf
, "%" PRIu64
"\n", shares
);
809 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
812 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
813 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
815 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
817 xsprintf(buf
, USEC_FMT
"\n", period
);
818 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
820 if (quota
!= USEC_INFINITY
) {
821 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
822 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
824 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
827 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
828 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
829 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
832 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
833 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
834 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
837 static void cgroup_apply_unified_cpuset(Unit
*u
, const CPUSet
*cpus
, const char *name
) {
838 _cleanup_free_
char *buf
= NULL
;
840 buf
= cpu_set_to_range_string(cpus
);
846 (void) set_attribute_and_warn(u
, "cpuset", name
, buf
);
849 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
850 return c
->io_accounting
||
851 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
852 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
853 c
->io_device_weights
||
854 c
->io_device_latencies
||
858 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
859 return c
->blockio_accounting
||
860 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
861 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
862 c
->blockio_device_weights
||
863 c
->blockio_device_bandwidths
;
866 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
867 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
868 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
869 return c
->startup_io_weight
;
870 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
873 return CGROUP_WEIGHT_DEFAULT
;
876 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
877 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
878 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
879 return c
->startup_blockio_weight
;
880 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
881 return c
->blockio_weight
;
883 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
886 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
887 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
888 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
891 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
892 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
893 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
896 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
897 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
901 r
= lookup_block_device(dev_path
, &dev
);
905 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
906 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
909 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
910 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
914 r
= lookup_block_device(dev_path
, &dev
);
918 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
919 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
922 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
923 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
927 r
= lookup_block_device(dev_path
, &dev
);
931 if (target
!= USEC_INFINITY
)
932 xsprintf(buf
, "%u:%u target=%" PRIu64
"\n", major(dev
), minor(dev
), target
);
934 xsprintf(buf
, "%u:%u target=max\n", major(dev
), minor(dev
));
936 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
939 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
940 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
941 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
942 CGroupIOLimitType type
;
946 r
= lookup_block_device(dev_path
, &dev
);
950 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
951 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
952 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
954 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
956 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
957 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
958 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
959 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
962 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
963 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
967 r
= lookup_block_device(dev_path
, &dev
);
971 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
972 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
974 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
975 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
978 static bool unit_has_unified_memory_config(Unit
*u
) {
983 c
= unit_get_cgroup_context(u
);
986 return unit_get_ancestor_memory_min(u
) > 0 || unit_get_ancestor_memory_low(u
) > 0 ||
987 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
||
988 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
;
991 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
992 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
994 if (v
!= CGROUP_LIMIT_MAX
)
995 xsprintf(buf
, "%" PRIu64
"\n", v
);
997 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
1000 static void cgroup_apply_firewall(Unit
*u
) {
1003 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
1005 if (bpf_firewall_compile(u
) < 0)
1008 (void) bpf_firewall_load_custom(u
);
1009 (void) bpf_firewall_install(u
);
1012 static int cgroup_apply_devices(Unit
*u
) {
1013 _cleanup_(bpf_program_unrefp
) BPFProgram
*prog
= NULL
;
1016 CGroupDeviceAllow
*a
;
1017 CGroupDevicePolicy policy
;
1020 assert_se(c
= unit_get_cgroup_context(u
));
1021 assert_se(path
= u
->cgroup_path
);
1023 policy
= c
->device_policy
;
1025 if (cg_all_unified() > 0) {
1026 r
= bpf_devices_cgroup_init(&prog
, policy
, c
->device_allow
);
1028 return log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
1031 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore
1034 if (c
->device_allow
|| policy
!= CGROUP_DEVICE_POLICY_AUTO
)
1035 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
1037 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
1039 log_unit_full_errno(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1040 "Failed to reset devices.allow/devices.deny: %m");
1043 bool allow_list_static
= policy
== CGROUP_DEVICE_POLICY_CLOSED
||
1044 (policy
== CGROUP_DEVICE_POLICY_AUTO
&& c
->device_allow
);
1045 if (allow_list_static
)
1046 (void) bpf_devices_allow_list_static(prog
, path
);
1048 bool any
= allow_list_static
;
1049 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1063 if (path_startswith(a
->path
, "/dev/"))
1064 r
= bpf_devices_allow_list_device(prog
, path
, a
->path
, acc
);
1065 else if ((val
= startswith(a
->path
, "block-")))
1066 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'b', acc
);
1067 else if ((val
= startswith(a
->path
, "char-")))
1068 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'c', acc
);
1070 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1079 log_unit_warning_errno(u
, SYNTHETIC_ERRNO(ENODEV
), "No devices matched by device filter.");
1081 /* The kernel verifier would reject a program we would build with the normal intro and outro
1082 but no allow-listing rules (outro would contain an unreachable instruction for successful
1084 policy
= CGROUP_DEVICE_POLICY_STRICT
;
1087 r
= bpf_devices_apply_policy(prog
, policy
, any
, path
, &u
->bpf_device_control_installed
);
1089 static bool warned
= false;
1091 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1092 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1093 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1094 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1101 static void set_io_weight(Unit
*u
, const char *controller
, uint64_t weight
) {
1102 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
1105 p
= strjoina(controller
, ".weight");
1106 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1107 (void) set_attribute_and_warn(u
, controller
, p
, buf
);
1109 /* FIXME: drop this when distro kernels properly support BFQ through "io.weight"
1110 * See also: https://github.com/systemd/systemd/pull/13335 and
1111 * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9.
1112 * The range is 1..1000 apparently. */
1113 p
= strjoina(controller
, ".bfq.weight");
1114 xsprintf(buf
, "%" PRIu64
"\n", (weight
+ 9) / 10);
1115 (void) set_attribute_and_warn(u
, controller
, p
, buf
);
1118 static void cgroup_context_apply(
1120 CGroupMask apply_mask
,
1121 ManagerState state
) {
1125 bool is_host_root
, is_local_root
;
1130 /* Nothing to do? Exit early! */
1131 if (apply_mask
== 0)
1134 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
1135 * attributes should only be managed for cgroups further down the tree. */
1136 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1137 is_host_root
= unit_has_host_root_cgroup(u
);
1139 assert_se(c
= unit_get_cgroup_context(u
));
1140 assert_se(path
= u
->cgroup_path
);
1142 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
1145 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
1146 * then), and missing cgroups, i.e. EROFS and ENOENT. */
1148 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
1149 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
1150 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
1151 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
1152 * we couldn't even write to them if we wanted to). */
1153 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
1155 if (cg_all_unified() > 0) {
1158 if (cgroup_context_has_cpu_weight(c
))
1159 weight
= cgroup_context_cpu_weight(c
, state
);
1160 else if (cgroup_context_has_cpu_shares(c
)) {
1163 shares
= cgroup_context_cpu_shares(c
, state
);
1164 weight
= cgroup_cpu_shares_to_weight(shares
);
1166 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
1167 shares
, weight
, path
);
1169 weight
= CGROUP_WEIGHT_DEFAULT
;
1171 cgroup_apply_unified_cpu_weight(u
, weight
);
1172 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1177 if (cgroup_context_has_cpu_weight(c
)) {
1180 weight
= cgroup_context_cpu_weight(c
, state
);
1181 shares
= cgroup_cpu_weight_to_shares(weight
);
1183 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
1184 weight
, shares
, path
);
1185 } else if (cgroup_context_has_cpu_shares(c
))
1186 shares
= cgroup_context_cpu_shares(c
, state
);
1188 shares
= CGROUP_CPU_SHARES_DEFAULT
;
1190 cgroup_apply_legacy_cpu_shares(u
, shares
);
1191 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1195 if ((apply_mask
& CGROUP_MASK_CPUSET
) && !is_local_root
) {
1196 cgroup_apply_unified_cpuset(u
, &c
->cpuset_cpus
, "cpuset.cpus");
1197 cgroup_apply_unified_cpuset(u
, &c
->cpuset_mems
, "cpuset.mems");
1200 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
1201 * controller), and in case of containers we want to leave control of these attributes to the container manager
1202 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
1203 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
1204 bool has_io
, has_blockio
;
1207 has_io
= cgroup_context_has_io_config(c
);
1208 has_blockio
= cgroup_context_has_blockio_config(c
);
1211 weight
= cgroup_context_io_weight(c
, state
);
1212 else if (has_blockio
) {
1213 uint64_t blkio_weight
;
1215 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
1216 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
1218 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
1219 blkio_weight
, weight
);
1221 weight
= CGROUP_WEIGHT_DEFAULT
;
1223 set_io_weight(u
, "io", weight
);
1226 CGroupIODeviceLatency
*latency
;
1227 CGroupIODeviceLimit
*limit
;
1228 CGroupIODeviceWeight
*w
;
1230 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
1231 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
1233 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
1234 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
1236 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
1237 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
1239 } else if (has_blockio
) {
1240 CGroupBlockIODeviceWeight
*w
;
1241 CGroupBlockIODeviceBandwidth
*b
;
1243 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
1244 weight
= cgroup_weight_blkio_to_io(w
->weight
);
1246 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
1247 w
->weight
, weight
, w
->path
);
1249 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
1252 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
1253 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
1254 CGroupIOLimitType type
;
1256 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1257 limits
[type
] = cgroup_io_limit_defaults
[type
];
1259 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
1260 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
1262 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
1263 b
->rbps
, b
->wbps
, b
->path
);
1265 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
1270 if (apply_mask
& CGROUP_MASK_BLKIO
) {
1271 bool has_io
, has_blockio
;
1273 has_io
= cgroup_context_has_io_config(c
);
1274 has_blockio
= cgroup_context_has_blockio_config(c
);
1276 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
1277 * left to our container manager, too. */
1278 if (!is_local_root
) {
1284 io_weight
= cgroup_context_io_weight(c
, state
);
1285 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
1287 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
1289 } else if (has_blockio
)
1290 weight
= cgroup_context_blkio_weight(c
, state
);
1292 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
1294 set_io_weight(u
, "blkio", weight
);
1297 CGroupIODeviceWeight
*w
;
1299 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
1300 weight
= cgroup_weight_io_to_blkio(w
->weight
);
1302 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
1303 w
->weight
, weight
, w
->path
);
1305 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
1307 } else if (has_blockio
) {
1308 CGroupBlockIODeviceWeight
*w
;
1310 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
1311 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
1315 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
1316 * roots, as there we want the container manager to handle it */
1317 if (is_host_root
|| !is_local_root
) {
1319 CGroupIODeviceLimit
*l
;
1321 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
1322 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
1323 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
1325 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
1327 } else if (has_blockio
) {
1328 CGroupBlockIODeviceBandwidth
*b
;
1330 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
1331 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
1336 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
1337 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
1338 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
1339 * write to this if we wanted to.) */
1340 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
1342 if (cg_all_unified() > 0) {
1343 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
;
1345 if (unit_has_unified_memory_config(u
)) {
1346 max
= c
->memory_max
;
1347 swap_max
= c
->memory_swap_max
;
1349 max
= c
->memory_limit
;
1351 if (max
!= CGROUP_LIMIT_MAX
)
1352 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
1355 cgroup_apply_unified_memory_limit(u
, "memory.min", unit_get_ancestor_memory_min(u
));
1356 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
1357 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
1358 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
1359 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
1361 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
1364 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1367 if (unit_has_unified_memory_config(u
)) {
1368 val
= c
->memory_max
;
1369 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIi64
" as MemoryLimit=", val
);
1371 val
= c
->memory_limit
;
1373 if (val
== CGROUP_LIMIT_MAX
)
1374 strncpy(buf
, "-1\n", sizeof(buf
));
1376 xsprintf(buf
, "%" PRIu64
"\n", val
);
1378 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
1382 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
1383 * containers, where we leave this to the manager */
1384 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
1385 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
))
1386 (void) cgroup_apply_devices(u
);
1388 if (apply_mask
& CGROUP_MASK_PIDS
) {
1391 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1392 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1393 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1394 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1395 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1396 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1397 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1398 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1399 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1400 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1401 * which is desirable so that there's an official way to release control of the sysctl from
1402 * systemd: set the limit to unbounded and reload. */
1404 if (tasks_max_isset(&c
->tasks_max
)) {
1405 u
->manager
->sysctl_pid_max_changed
= true;
1406 r
= procfs_tasks_set_limit(tasks_max_resolve(&c
->tasks_max
));
1407 } else if (u
->manager
->sysctl_pid_max_changed
)
1408 r
= procfs_tasks_set_limit(TASKS_MAX
);
1412 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
1413 "Failed to write to tasks limit sysctls: %m");
1416 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
1417 * leave it for the container manager. */
1418 if (!is_local_root
) {
1419 if (tasks_max_isset(&c
->tasks_max
)) {
1420 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1422 xsprintf(buf
, "%" PRIu64
"\n", tasks_max_resolve(&c
->tasks_max
));
1423 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
1425 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
1429 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
1430 cgroup_apply_firewall(u
);
1433 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
1438 c
= unit_get_cgroup_context(u
);
1442 if (c
->ip_accounting
||
1443 c
->ip_address_allow
||
1444 c
->ip_address_deny
||
1445 c
->ip_filters_ingress
||
1446 c
->ip_filters_egress
)
1449 /* If any parent slice has an IP access list defined, it applies too */
1450 for (p
= UNIT_DEREF(u
->slice
); p
; p
= UNIT_DEREF(p
->slice
)) {
1451 c
= unit_get_cgroup_context(p
);
1455 if (c
->ip_address_allow
||
1463 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
1464 CGroupMask mask
= 0;
1469 c
= unit_get_cgroup_context(u
);
1473 /* Figure out which controllers we need, based on the cgroup context object */
1475 if (c
->cpu_accounting
)
1476 mask
|= get_cpu_accounting_mask();
1478 if (cgroup_context_has_cpu_weight(c
) ||
1479 cgroup_context_has_cpu_shares(c
) ||
1480 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
1481 mask
|= CGROUP_MASK_CPU
;
1483 if (c
->cpuset_cpus
.set
|| c
->cpuset_mems
.set
)
1484 mask
|= CGROUP_MASK_CPUSET
;
1486 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
1487 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1489 if (c
->memory_accounting
||
1490 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
1491 unit_has_unified_memory_config(u
))
1492 mask
|= CGROUP_MASK_MEMORY
;
1494 if (c
->device_allow
||
1495 c
->device_policy
!= CGROUP_DEVICE_POLICY_AUTO
)
1496 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
1498 if (c
->tasks_accounting
||
1499 tasks_max_isset(&c
->tasks_max
))
1500 mask
|= CGROUP_MASK_PIDS
;
1502 return CGROUP_MASK_EXTEND_JOINED(mask
);
1505 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
1506 CGroupMask mask
= 0;
1508 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
1511 if (unit_get_needs_bpf_firewall(u
))
1512 mask
|= CGROUP_MASK_BPF_FIREWALL
;
1517 CGroupMask
unit_get_own_mask(Unit
*u
) {
1520 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
1521 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
1523 if (u
->load_state
!= UNIT_LOADED
)
1526 c
= unit_get_cgroup_context(u
);
1530 return unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
);
1533 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
1536 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1537 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1539 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1541 if (!unit_cgroup_delegate(u
))
1544 if (cg_all_unified() <= 0) {
1547 e
= unit_get_exec_context(u
);
1548 if (e
&& !exec_context_maintains_privileges(e
))
1552 assert_se(c
= unit_get_cgroup_context(u
));
1553 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
1556 static CGroupMask
unit_get_subtree_mask(Unit
*u
) {
1558 /* Returns the mask of this subtree, meaning of the group
1559 * itself and its children. */
1561 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
1564 CGroupMask
unit_get_members_mask(Unit
*u
) {
1567 /* Returns the mask of controllers all of the unit's children require, merged */
1569 if (u
->cgroup_members_mask_valid
)
1570 return u
->cgroup_members_mask
; /* Use cached value if possible */
1572 u
->cgroup_members_mask
= 0;
1574 if (u
->type
== UNIT_SLICE
) {
1578 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
])
1579 if (UNIT_DEREF(member
->slice
) == u
)
1580 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
1583 u
->cgroup_members_mask_valid
= true;
1584 return u
->cgroup_members_mask
;
1587 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
1590 /* Returns the mask of controllers all of the unit's siblings
1591 * require, i.e. the members mask of the unit's parent slice
1592 * if there is one. */
1594 if (UNIT_ISSET(u
->slice
))
1595 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
1597 return unit_get_subtree_mask(u
); /* we are the top-level slice */
1600 static CGroupMask
unit_get_disable_mask(Unit
*u
) {
1603 c
= unit_get_cgroup_context(u
);
1607 return c
->disable_controllers
;
1610 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
1614 mask
= unit_get_disable_mask(u
);
1616 /* Returns the mask of controllers which are marked as forcibly
1617 * disabled in any ancestor unit or the unit in question. */
1619 if (UNIT_ISSET(u
->slice
))
1620 mask
|= unit_get_ancestor_disable_mask(UNIT_DEREF(u
->slice
));
1625 CGroupMask
unit_get_target_mask(Unit
*u
) {
1626 CGroupMask own_mask
, mask
;
1628 /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything
1629 * it needs itself, plus all that its children need, plus all that its siblings need. This is
1630 * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each
1631 * hierarchy that shall be enabled for it. */
1633 own_mask
= unit_get_own_mask(u
);
1635 if (own_mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
1636 emit_bpf_firewall_warning(u
);
1638 mask
= own_mask
| unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
1640 mask
&= u
->manager
->cgroup_supported
;
1641 mask
&= ~unit_get_ancestor_disable_mask(u
);
1646 CGroupMask
unit_get_enable_mask(Unit
*u
) {
1649 /* This returns the cgroup mask of all controllers to enable
1650 * for the children of a specific cgroup. This is primarily
1651 * useful for the unified cgroup hierarchy, where each cgroup
1652 * controls which controllers are enabled for its children. */
1654 mask
= unit_get_members_mask(u
);
1655 mask
&= u
->manager
->cgroup_supported
;
1656 mask
&= ~unit_get_ancestor_disable_mask(u
);
1661 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
1664 /* Recurse invalidate the member masks cache all the way up the tree */
1665 u
->cgroup_members_mask_valid
= false;
1667 if (UNIT_ISSET(u
->slice
))
1668 unit_invalidate_cgroup_members_masks(UNIT_DEREF(u
->slice
));
1671 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
1673 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1677 if (u
->cgroup_path
&&
1678 u
->cgroup_realized
&&
1679 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
1680 return u
->cgroup_path
;
1682 u
= UNIT_DEREF(u
->slice
);
1688 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1689 /* If not realized at all, migrate to root ("").
1690 * It may happen if we're upgrading from older version that didn't clean up.
1692 return strempty(unit_get_realized_cgroup_path(userdata
, mask
));
1695 char *unit_default_cgroup_path(const Unit
*u
) {
1696 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1701 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1702 return strdup(u
->manager
->cgroup_root
);
1704 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1705 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1710 escaped
= cg_escape(u
->id
);
1714 return path_join(empty_to_root(u
->manager
->cgroup_root
), slice
, escaped
);
1717 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1718 _cleanup_free_
char *p
= NULL
;
1723 if (streq_ptr(u
->cgroup_path
, path
))
1733 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1738 unit_release_cgroup(u
);
1739 u
->cgroup_path
= TAKE_PTR(p
);
1744 int unit_watch_cgroup(Unit
*u
) {
1745 _cleanup_free_
char *events
= NULL
;
1750 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
1751 * cgroupv2 is available. */
1753 if (!u
->cgroup_path
)
1756 if (u
->cgroup_control_inotify_wd
>= 0)
1759 /* Only applies to the unified hierarchy */
1760 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1762 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1766 /* No point in watch the top-level slice, it's never going to run empty. */
1767 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1770 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
1774 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1778 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1779 if (u
->cgroup_control_inotify_wd
< 0) {
1781 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1782 * is not an error */
1785 return log_unit_error_errno(u
, errno
, "Failed to add control inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1788 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
1790 return log_unit_error_errno(u
, r
, "Failed to add control inotify watch descriptor to hash map: %m");
1795 int unit_watch_cgroup_memory(Unit
*u
) {
1796 _cleanup_free_
char *events
= NULL
;
1802 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
1803 * cgroupv2 is available. */
1805 if (!u
->cgroup_path
)
1808 c
= unit_get_cgroup_context(u
);
1812 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
1813 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
1815 if (!c
->memory_accounting
)
1818 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
1819 * we also don't want to generate a log message for each parent cgroup of a process. */
1820 if (u
->type
== UNIT_SLICE
)
1823 if (u
->cgroup_memory_inotify_wd
>= 0)
1826 /* Only applies to the unified hierarchy */
1827 r
= cg_all_unified();
1829 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
1833 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
1837 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
1841 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1842 if (u
->cgroup_memory_inotify_wd
< 0) {
1844 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1845 * is not an error */
1848 return log_unit_error_errno(u
, errno
, "Failed to add memory inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1851 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
1853 return log_unit_error_errno(u
, r
, "Failed to add memory inotify watch descriptor to hash map: %m");
1858 int unit_pick_cgroup_path(Unit
*u
) {
1859 _cleanup_free_
char *path
= NULL
;
1867 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1870 path
= unit_default_cgroup_path(u
);
1874 r
= unit_set_cgroup_path(u
, path
);
1876 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1878 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1883 static int unit_update_cgroup(
1885 CGroupMask target_mask
,
1886 CGroupMask enable_mask
,
1887 ManagerState state
) {
1889 bool created
, is_root_slice
;
1890 CGroupMask migrate_mask
= 0;
1895 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1898 /* Figure out our cgroup path */
1899 r
= unit_pick_cgroup_path(u
);
1903 /* First, create our own group */
1904 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
1906 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
1909 /* Start watching it */
1910 (void) unit_watch_cgroup(u
);
1911 (void) unit_watch_cgroup_memory(u
);
1914 /* For v2 we preserve enabled controllers in delegated units, adjust others,
1915 * for v1 we figure out which controller hierarchies need migration. */
1916 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
1917 CGroupMask result_mask
= 0;
1919 /* Enable all controllers we need */
1920 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
1922 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
1924 /* Remember what's actually enabled now */
1925 u
->cgroup_enabled_mask
= result_mask
;
1927 migrate_mask
= u
->cgroup_realized_mask
^ target_mask
;
1930 /* Keep track that this is now realized */
1931 u
->cgroup_realized
= true;
1932 u
->cgroup_realized_mask
= target_mask
;
1934 /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling).
1936 * Unnecessary controller cgroups are trimmed (after emptied by upward migration).
1937 * We perform migration also with whole slices for cases when users don't care about leave
1938 * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing
1941 if (cg_all_unified() == 0) {
1942 r
= cg_migrate_v1_controllers(u
->manager
->cgroup_supported
, migrate_mask
, u
->cgroup_path
, migrate_callback
, u
);
1944 log_unit_warning_errno(u
, r
, "Failed to migrate controller cgroups from %s, ignoring: %m", u
->cgroup_path
);
1946 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1947 r
= cg_trim_v1_controllers(u
->manager
->cgroup_supported
, ~target_mask
, u
->cgroup_path
, !is_root_slice
);
1949 log_unit_warning_errno(u
, r
, "Failed to delete controller cgroups %s, ignoring: %m", u
->cgroup_path
);
1952 /* Set attributes */
1953 cgroup_context_apply(u
, target_mask
, state
);
1954 cgroup_xattr_apply(u
);
1959 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
1960 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1966 if (MANAGER_IS_SYSTEM(u
->manager
))
1969 if (!u
->manager
->system_bus
)
1972 if (!u
->cgroup_path
)
1975 /* Determine this unit's cgroup path relative to our cgroup root */
1976 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
1980 pp
= strjoina("/", pp
, suffix_path
);
1981 path_simplify(pp
, false);
1983 r
= sd_bus_call_method(u
->manager
->system_bus
,
1984 "org.freedesktop.systemd1",
1985 "/org/freedesktop/systemd1",
1986 "org.freedesktop.systemd1.Manager",
1987 "AttachProcessesToUnit",
1990 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
1992 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
1997 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
1998 CGroupMask delegated_mask
;
2005 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2008 if (set_isempty(pids
))
2011 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
2012 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
2013 r
= bpf_firewall_load_custom(u
);
2017 r
= unit_realize_cgroup(u
);
2021 if (isempty(suffix_path
))
2024 p
= prefix_roota(u
->cgroup_path
, suffix_path
);
2026 delegated_mask
= unit_get_delegate_mask(u
);
2029 SET_FOREACH(pidp
, pids
) {
2030 pid_t pid
= PTR_TO_PID(pidp
);
2033 /* First, attach the PID to the main cgroup hierarchy */
2034 q
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
);
2036 log_unit_debug_errno(u
, q
, "Couldn't move process " PID_FMT
" to requested cgroup '%s': %m", pid
, p
);
2038 if (MANAGER_IS_USER(u
->manager
) && ERRNO_IS_PRIVILEGE(q
)) {
2041 /* If we are in a user instance, and we can't move the process ourselves due to
2042 * permission problems, let's ask the system instance about it instead. Since it's more
2043 * privileged it might be able to move the process across the leaves of a subtree who's
2044 * top node is not owned by us. */
2046 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
, suffix_path
);
2048 log_unit_debug_errno(u
, z
, "Couldn't move process " PID_FMT
" to requested cgroup '%s' via the system bus either: %m", pid
, p
);
2050 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
2054 r
= q
; /* Remember first error */
2059 q
= cg_all_unified();
2065 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
2066 * innermost realized one */
2068 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2069 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2070 const char *realized
;
2072 if (!(u
->manager
->cgroup_supported
& bit
))
2075 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
2076 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
2077 q
= cg_attach(cgroup_controller_to_string(c
), p
, pid
);
2079 continue; /* Success! */
2081 log_unit_debug_errno(u
, q
, "Failed to attach PID " PID_FMT
" to requested cgroup %s in controller %s, falling back to unit's cgroup: %m",
2082 pid
, p
, cgroup_controller_to_string(c
));
2085 /* So this controller is either not delegate or realized, or something else weird happened. In
2086 * that case let's attach the PID at least to the closest cgroup up the tree that is
2088 realized
= unit_get_realized_cgroup_path(u
, bit
);
2090 continue; /* Not even realized in the root slice? Then let's not bother */
2092 q
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
);
2094 log_unit_debug_errno(u
, q
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
2095 pid
, realized
, cgroup_controller_to_string(c
));
2102 static bool unit_has_mask_realized(
2104 CGroupMask target_mask
,
2105 CGroupMask enable_mask
) {
2109 /* Returns true if this unit is fully realized. We check four things:
2111 * 1. Whether the cgroup was created at all
2112 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
2113 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
2114 * 4. Whether the invalidation mask is currently zero
2116 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
2117 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
2118 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
2119 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
2120 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
2121 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
2122 * simply don't matter. */
2124 return u
->cgroup_realized
&&
2125 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
2126 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
2127 u
->cgroup_invalidated_mask
== 0;
2130 static bool unit_has_mask_disables_realized(
2132 CGroupMask target_mask
,
2133 CGroupMask enable_mask
) {
2137 /* Returns true if all controllers which should be disabled are indeed disabled.
2139 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
2140 * already removed. */
2142 return !u
->cgroup_realized
||
2143 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
2144 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
2147 static bool unit_has_mask_enables_realized(
2149 CGroupMask target_mask
,
2150 CGroupMask enable_mask
) {
2154 /* Returns true if all controllers which should be enabled are indeed enabled.
2156 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
2157 * we want to add is already added. */
2159 return u
->cgroup_realized
&&
2160 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
2161 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
2164 static void unit_add_to_cgroup_realize_queue(Unit
*u
) {
2167 if (u
->in_cgroup_realize_queue
)
2170 LIST_APPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2171 u
->in_cgroup_realize_queue
= true;
2174 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
2177 if (!u
->in_cgroup_realize_queue
)
2180 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2181 u
->in_cgroup_realize_queue
= false;
2184 /* Controllers can only be enabled breadth-first, from the root of the
2185 * hierarchy downwards to the unit in question. */
2186 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
2187 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2192 /* First go deal with this unit's parent, or we won't be able to enable
2193 * any new controllers at this layer. */
2194 if (UNIT_ISSET(u
->slice
)) {
2195 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2200 target_mask
= unit_get_target_mask(u
);
2201 enable_mask
= unit_get_enable_mask(u
);
2203 /* We can only enable in this direction, don't try to disable anything.
2205 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
2208 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
2209 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
2211 return unit_update_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
2214 /* Controllers can only be disabled depth-first, from the leaves of the
2215 * hierarchy upwards to the unit in question. */
2216 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
2222 if (u
->type
!= UNIT_SLICE
)
2225 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
]) {
2226 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2229 if (UNIT_DEREF(m
->slice
) != u
)
2232 /* The cgroup for this unit might not actually be fully
2233 * realised yet, in which case it isn't holding any controllers
2235 if (!m
->cgroup_realized
)
2238 /* We must disable those below us first in order to release the
2240 if (m
->type
== UNIT_SLICE
)
2241 (void) unit_realize_cgroup_now_disable(m
, state
);
2243 target_mask
= unit_get_target_mask(m
);
2244 enable_mask
= unit_get_enable_mask(m
);
2246 /* We can only disable in this direction, don't try to enable
2248 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
2251 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
2252 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
2254 r
= unit_update_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
2262 /* Check if necessary controllers and attributes for a unit are in place.
2264 * - If so, do nothing.
2265 * - If not, create paths, move processes over, and set attributes.
2267 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
2268 * a depth-first way. As such the process looks like this:
2270 * Suppose we have a cgroup hierarchy which looks like this:
2283 * 1. We want to realise cgroup "d" now.
2284 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
2285 * 3. cgroup "k" just started requesting the memory controller.
2287 * To make this work we must do the following in order:
2289 * 1. Disable CPU controller in k, j
2290 * 2. Disable CPU controller in d
2291 * 3. Enable memory controller in root
2292 * 4. Enable memory controller in a
2293 * 5. Enable memory controller in d
2294 * 6. Enable memory controller in k
2296 * Notice that we need to touch j in one direction, but not the other. We also
2297 * don't go beyond d when disabling -- it's up to "a" to get realized if it
2298 * wants to disable further. The basic rules are therefore:
2300 * - If you're disabling something, you need to realise all of the cgroups from
2301 * your recursive descendants to the root. This starts from the leaves.
2302 * - If you're enabling something, you need to realise from the root cgroup
2303 * downwards, but you don't need to iterate your recursive descendants.
2305 * Returns 0 on success and < 0 on failure. */
2306 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
2307 CGroupMask target_mask
, enable_mask
;
2312 unit_remove_from_cgroup_realize_queue(u
);
2314 target_mask
= unit_get_target_mask(u
);
2315 enable_mask
= unit_get_enable_mask(u
);
2317 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
2320 /* Disable controllers below us, if there are any */
2321 r
= unit_realize_cgroup_now_disable(u
, state
);
2325 /* Enable controllers above us, if there are any */
2326 if (UNIT_ISSET(u
->slice
)) {
2327 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2332 /* Now actually deal with the cgroup we were trying to realise and set attributes */
2333 r
= unit_update_cgroup(u
, target_mask
, enable_mask
, state
);
2337 /* Now, reset the invalidation mask */
2338 u
->cgroup_invalidated_mask
= 0;
2342 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
2350 state
= manager_state(m
);
2352 while ((i
= m
->cgroup_realize_queue
)) {
2353 assert(i
->in_cgroup_realize_queue
);
2355 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
2356 /* Maybe things changed, and the unit is not actually active anymore? */
2357 unit_remove_from_cgroup_realize_queue(i
);
2361 r
= unit_realize_cgroup_now(i
, state
);
2363 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
2371 void unit_add_family_to_cgroup_realize_queue(Unit
*u
) {
2373 assert(u
->type
== UNIT_SLICE
);
2375 /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all
2378 * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes
2379 * very weird if two units that own processes reside in the same slice, but one is realized in the
2380 * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does
2381 * not), because that means individual processes need to be scheduled against whole cgroups. Let's
2382 * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1
2383 * controller hierarchies too (if unit requires the controller to be realized).
2385 * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date
2392 /* Children of u likely changed when we're called */
2393 u
->cgroup_members_mask_valid
= false;
2395 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
]) {
2396 /* Skip units that have a dependency on the slice but aren't actually in it. */
2397 if (UNIT_DEREF(m
->slice
) != u
)
2400 /* No point in doing cgroup application for units without active processes. */
2401 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
2404 /* We only enqueue siblings if they were realized once at least, in the main
2406 if (!m
->cgroup_realized
)
2409 /* If the unit doesn't need any new controllers and has current ones realized, it
2410 * doesn't need any changes. */
2411 if (unit_has_mask_realized(m
,
2412 unit_get_target_mask(m
),
2413 unit_get_enable_mask(m
)))
2416 unit_add_to_cgroup_realize_queue(m
);
2419 /* Parent comes after children */
2420 unit_add_to_cgroup_realize_queue(u
);
2421 } while ((u
= UNIT_DEREF(u
->slice
)));
2424 int unit_realize_cgroup(Unit
*u
) {
2427 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2430 /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all
2431 * parents, but there's more actually: for the weight-based controllers we also need to make sure
2432 * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the
2433 * other hand, when a controller is removed from realized set, it may become unnecessary in siblings
2434 * and ancestors and they should be (de)realized too.
2436 * This call will defer work on the siblings and derealized ancestors to the next event loop
2437 * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */
2439 if (UNIT_ISSET(u
->slice
))
2440 unit_add_family_to_cgroup_realize_queue(UNIT_DEREF(u
->slice
));
2442 /* And realize this one now (and apply the values) */
2443 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
2446 void unit_release_cgroup(Unit
*u
) {
2449 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
2450 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
2452 if (u
->cgroup_path
) {
2453 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
2454 u
->cgroup_path
= mfree(u
->cgroup_path
);
2457 if (u
->cgroup_control_inotify_wd
>= 0) {
2458 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
2459 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
);
2461 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
2462 u
->cgroup_control_inotify_wd
= -1;
2465 if (u
->cgroup_memory_inotify_wd
>= 0) {
2466 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
2467 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
);
2469 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
2470 u
->cgroup_memory_inotify_wd
= -1;
2474 bool unit_maybe_release_cgroup(Unit
*u
) {
2479 if (!u
->cgroup_path
)
2482 /* Don't release the cgroup if there are still processes under it. If we get notified later when all the
2483 * processes exit (e.g. the processes were in D-state and exited after the unit was marked as failed)
2484 * we need the cgroup paths to continue to be tracked by the manager so they can be looked up and cleaned
2486 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2488 log_unit_debug_errno(u
, r
, "Error checking if the cgroup is recursively empty, ignoring: %m");
2490 unit_release_cgroup(u
);
2497 void unit_prune_cgroup(Unit
*u
) {
2503 /* Removes the cgroup, if empty and possible, and stops watching it. */
2505 if (!u
->cgroup_path
)
2508 (void) unit_get_cpu_usage(u
, NULL
); /* Cache the last CPU usage value before we destroy the cgroup */
2510 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2512 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
2514 /* One reason we could have failed here is, that the cgroup still contains a process.
2515 * However, if the cgroup becomes removable at a later time, it might be removed when
2516 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
2517 * that the cgroup is still realized the next time it is started. Do not return early
2518 * on error, continue cleanup. */
2519 log_unit_full_errno(u
, r
== -EBUSY
? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
2524 if (!unit_maybe_release_cgroup(u
)) /* Returns true if the cgroup was released */
2527 u
->cgroup_realized
= false;
2528 u
->cgroup_realized_mask
= 0;
2529 u
->cgroup_enabled_mask
= 0;
2531 u
->bpf_device_control_installed
= bpf_program_unref(u
->bpf_device_control_installed
);
2534 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
2535 _cleanup_fclose_
FILE *f
= NULL
;
2536 pid_t pid
= 0, npid
;
2542 if (!u
->cgroup_path
)
2545 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
2549 while (cg_read_pid(f
, &npid
) > 0) {
2554 if (pid_is_my_child(npid
) == 0)
2558 /* Dang, there's more than one daemonized PID
2559 in this group, so we don't know what process
2560 is the main process. */
2571 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
2572 _cleanup_closedir_
DIR *d
= NULL
;
2573 _cleanup_fclose_
FILE *f
= NULL
;
2579 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
2585 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
2586 r
= unit_watch_pid(u
, pid
, false);
2587 if (r
< 0 && ret
>= 0)
2591 if (r
< 0 && ret
>= 0)
2595 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
2602 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
2603 _cleanup_free_
char *p
= NULL
;
2605 p
= path_join(empty_to_root(path
), fn
);
2611 r
= unit_watch_pids_in_path(u
, p
);
2612 if (r
< 0 && ret
>= 0)
2616 if (r
< 0 && ret
>= 0)
2623 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
2628 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
2629 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
2630 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
2632 if (!u
->cgroup_path
)
2635 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2638 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
2641 if (!set_isempty(u
->pids
))
2644 unit_add_to_cgroup_empty_queue(u
);
2648 int unit_watch_all_pids(Unit
*u
) {
2653 /* Adds all PIDs from our cgroup to the set of PIDs we
2654 * watch. This is a fallback logic for cases where we do not
2655 * get reliable cgroup empty notifications: we try to use
2656 * SIGCHLD as replacement. */
2658 if (!u
->cgroup_path
)
2661 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2664 if (r
> 0) /* On unified we can use proper notifications */
2667 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
2670 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
2671 Manager
*m
= userdata
;
2678 u
= m
->cgroup_empty_queue
;
2682 assert(u
->in_cgroup_empty_queue
);
2683 u
->in_cgroup_empty_queue
= false;
2684 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
2686 if (m
->cgroup_empty_queue
) {
2687 /* More stuff queued, let's make sure we remain enabled */
2688 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2690 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
2693 unit_add_to_gc_queue(u
);
2695 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
2696 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
2701 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
2706 /* Note that there are four different ways how cgroup empty events reach us:
2708 * 1. On the unified hierarchy we get an inotify event on the cgroup
2710 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2712 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2714 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2715 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2717 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2718 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2719 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2720 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2721 * case for scope units). */
2723 if (u
->in_cgroup_empty_queue
)
2726 /* Let's verify that the cgroup is really empty */
2727 if (!u
->cgroup_path
)
2730 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2732 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", u
->cgroup_path
);
2738 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2739 u
->in_cgroup_empty_queue
= true;
2741 /* Trigger the defer event */
2742 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
2744 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
2747 static void unit_remove_from_cgroup_empty_queue(Unit
*u
) {
2750 if (!u
->in_cgroup_empty_queue
)
2753 LIST_REMOVE(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2754 u
->in_cgroup_empty_queue
= false;
2757 int unit_check_oomd_kill(Unit
*u
) {
2758 _cleanup_free_
char *value
= NULL
;
2763 if (!u
->cgroup_path
)
2766 r
= cg_all_unified();
2768 return log_unit_debug_errno(u
, r
, "Couldn't determine whether we are in all unified mode: %m");
2772 r
= cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "user.oomd_kill", &value
);
2773 if (r
< 0 && r
!= -ENODATA
)
2776 if (!isempty(value
)) {
2777 r
= safe_atou64(value
, &n
);
2782 increased
= n
> u
->managed_oom_kill_last
;
2783 u
->managed_oom_kill_last
= n
;
2789 log_struct(LOG_NOTICE
,
2790 "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR
,
2792 LOG_UNIT_INVOCATION_ID(u
),
2793 LOG_UNIT_MESSAGE(u
, "systemd-oomd killed %"PRIu64
" process(es) in this unit.", n
));
2798 int unit_check_oom(Unit
*u
) {
2799 _cleanup_free_
char *oom_kill
= NULL
;
2804 if (!u
->cgroup_path
)
2807 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
2809 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
2811 r
= safe_atou64(oom_kill
, &c
);
2813 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
2815 increased
= c
> u
->oom_kill_last
;
2816 u
->oom_kill_last
= c
;
2821 log_struct(LOG_NOTICE
,
2822 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
2824 LOG_UNIT_INVOCATION_ID(u
),
2825 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
2827 if (UNIT_VTABLE(u
)->notify_cgroup_oom
)
2828 UNIT_VTABLE(u
)->notify_cgroup_oom(u
);
2833 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
2834 Manager
*m
= userdata
;
2841 u
= m
->cgroup_oom_queue
;
2845 assert(u
->in_cgroup_oom_queue
);
2846 u
->in_cgroup_oom_queue
= false;
2847 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
2849 if (m
->cgroup_oom_queue
) {
2850 /* More stuff queued, let's make sure we remain enabled */
2851 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2853 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
2856 (void) unit_check_oom(u
);
2860 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
2865 if (u
->in_cgroup_oom_queue
)
2867 if (!u
->cgroup_path
)
2870 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
2871 u
->in_cgroup_oom_queue
= true;
2873 /* Trigger the defer event */
2874 if (!u
->manager
->cgroup_oom_event_source
) {
2875 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
2877 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
2879 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
2883 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
2885 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
2889 (void) sd_event_source_set_description(s
, "cgroup-oom");
2890 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
2893 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
2895 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
2898 static int unit_check_cgroup_events(Unit
*u
) {
2899 char *values
[2] = {};
2904 r
= cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events",
2905 STRV_MAKE("populated", "frozen"), values
);
2909 /* The cgroup.events notifications can be merged together so act as we saw the given state for the
2910 * first time. The functions we call to handle given state are idempotent, which makes them
2911 * effectively remember the previous state. */
2913 if (streq(values
[0], "1"))
2914 unit_remove_from_cgroup_empty_queue(u
);
2916 unit_add_to_cgroup_empty_queue(u
);
2919 /* Disregard freezer state changes due to operations not initiated by us */
2920 if (values
[1] && IN_SET(u
->freezer_state
, FREEZER_FREEZING
, FREEZER_THAWING
)) {
2921 if (streq(values
[1], "0"))
2933 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
2934 Manager
*m
= userdata
;
2941 union inotify_event_buffer buffer
;
2942 struct inotify_event
*e
;
2945 l
= read(fd
, &buffer
, sizeof(buffer
));
2947 if (IN_SET(errno
, EINTR
, EAGAIN
))
2950 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
2953 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
2957 /* Queue overflow has no watch descriptor */
2960 if (e
->mask
& IN_IGNORED
)
2961 /* The watch was just removed */
2964 /* Note that inotify might deliver events for a watch even after it was removed,
2965 * because it was queued before the removal. Let's ignore this here safely. */
2967 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2969 unit_check_cgroup_events(u
);
2971 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2973 unit_add_to_cgroup_oom_queue(u
);
2978 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
2979 CGroupMask mask
= 0;
2982 /* BPF-based firewall */
2983 r
= bpf_firewall_supported();
2985 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2987 /* BPF-based device access control */
2988 r
= bpf_devices_supported();
2990 mask
|= CGROUP_MASK_BPF_DEVICES
;
2996 int manager_setup_cgroup(Manager
*m
) {
2997 _cleanup_free_
char *path
= NULL
;
2998 const char *scope_path
;
3006 /* 1. Determine hierarchy */
3007 m
->cgroup_root
= mfree(m
->cgroup_root
);
3008 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
3010 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
3012 /* Chop off the init scope, if we are already located in it */
3013 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
3015 /* LEGACY: Also chop off the system slice if we are in
3016 * it. This is to support live upgrades from older systemd
3017 * versions where PID 1 was moved there. Also see
3018 * cg_get_root_path(). */
3019 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
3020 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
3022 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
3027 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
3028 * easily prepend it everywhere. */
3029 delete_trailing_chars(m
->cgroup_root
, "/");
3032 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
3034 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
3038 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
3040 all_unified
= cg_all_unified();
3041 if (all_unified
< 0)
3042 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
3043 if (all_unified
> 0)
3044 log_debug("Unified cgroup hierarchy is located at %s.", path
);
3046 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
3048 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
3050 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
3052 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
3055 /* 3. Allocate cgroup empty defer event source */
3056 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
3057 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
3059 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
3061 /* Schedule cgroup empty checks early, but after having processed service notification messages or
3062 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
3063 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
3064 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
3066 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
3068 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
3070 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
3072 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
3074 /* 4. Install notifier inotify object, or agent */
3075 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
3077 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
3079 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
3080 safe_close(m
->cgroup_inotify_fd
);
3082 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
3083 if (m
->cgroup_inotify_fd
< 0)
3084 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
3086 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
3088 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
3090 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
3091 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
3092 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
3093 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
3095 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
3097 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
3099 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
3101 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
3102 * since it does not generate events when control groups with children run empty. */
3104 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUPS_AGENT_PATH
);
3106 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
3108 log_debug("Installed release agent.");
3110 log_debug("Release agent already installed.");
3113 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
3114 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
3115 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
3117 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
3118 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
3120 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
3122 /* 6. And pin it, so that it cannot be unmounted */
3123 safe_close(m
->pin_cgroupfs_fd
);
3124 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
3125 if (m
->pin_cgroupfs_fd
< 0)
3126 return log_error_errno(errno
, "Failed to open pin file: %m");
3128 } else if (!MANAGER_IS_TEST_RUN(m
))
3129 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
3131 /* 7. Always enable hierarchical support if it exists... */
3132 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
3133 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
3135 /* 8. Figure out which controllers are supported */
3136 r
= cg_mask_supported_subtree(m
->cgroup_root
, &m
->cgroup_supported
);
3138 return log_error_errno(r
, "Failed to determine supported controllers: %m");
3140 /* 9. Figure out which bpf-based pseudo-controllers are supported */
3141 r
= cg_bpf_mask_supported(&mask
);
3143 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
3144 m
->cgroup_supported
|= mask
;
3146 /* 10. Log which controllers are supported */
3147 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
3148 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
3153 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
3156 /* We can't really delete the group, since we are in it. But
3158 if (delete && m
->cgroup_root
&& m
->test_run_flags
!= MANAGER_TEST_RUN_MINIMAL
)
3159 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
3161 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
3163 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
3164 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
3166 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
3167 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
3169 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
3171 m
->cgroup_root
= mfree(m
->cgroup_root
);
3174 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
3181 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
3185 p
= strdupa(cgroup
);
3189 e
= strrchr(p
, '/');
3191 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
3195 u
= hashmap_get(m
->cgroup_unit
, p
);
3201 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
3202 _cleanup_free_
char *cgroup
= NULL
;
3206 if (!pid_is_valid(pid
))
3209 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
3212 return manager_get_unit_by_cgroup(m
, cgroup
);
3215 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
3220 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
3221 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
3222 * relevant one as children of the process will be assigned to that one, too, before all else. */
3224 if (!pid_is_valid(pid
))
3227 if (pid
== getpid_cached())
3228 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
3230 u
= manager_get_unit_by_pid_cgroup(m
, pid
);
3234 u
= hashmap_get(m
->watch_pids
, PID_TO_PTR(pid
));
3238 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-pid
));
3245 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
3251 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
3252 * or from the --system instance */
3254 log_debug("Got cgroup empty notification for: %s", cgroup
);
3256 u
= manager_get_unit_by_cgroup(m
, cgroup
);
3260 unit_add_to_cgroup_empty_queue(u
);
3264 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
3270 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3273 if (!u
->cgroup_path
)
3276 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3277 if (unit_has_host_root_cgroup(u
))
3278 return procfs_memory_get_used(ret
);
3280 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3283 r
= cg_all_unified();
3287 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, r
> 0 ? "memory.current" : "memory.usage_in_bytes", ret
);
3290 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
3294 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
3297 if (!u
->cgroup_path
)
3300 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3301 if (unit_has_host_root_cgroup(u
))
3302 return procfs_tasks_get_current(ret
);
3304 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
3307 return cg_get_attribute_as_uint64("pids", u
->cgroup_path
, "pids.current", ret
);
3310 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
3317 if (!u
->cgroup_path
)
3320 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3321 if (unit_has_host_root_cgroup(u
))
3322 return procfs_cpu_get_usage(ret
);
3324 /* Requisite controllers for CPU accounting are not enabled */
3325 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
3328 r
= cg_all_unified();
3332 _cleanup_free_
char *val
= NULL
;
3335 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
3336 if (IN_SET(r
, -ENOENT
, -ENXIO
))
3341 r
= safe_atou64(val
, &us
);
3345 ns
= us
* NSEC_PER_USEC
;
3347 return cg_get_attribute_as_uint64("cpuacct", u
->cgroup_path
, "cpuacct.usage", ret
);
3353 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
3359 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
3360 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
3361 * call this function with a NULL return value. */
3363 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
3366 r
= unit_get_cpu_usage_raw(u
, &ns
);
3367 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
3368 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
3372 *ret
= u
->cpu_usage_last
;
3378 if (ns
> u
->cpu_usage_base
)
3379 ns
-= u
->cpu_usage_base
;
3383 u
->cpu_usage_last
= ns
;
3390 int unit_get_ip_accounting(
3392 CGroupIPAccountingMetric metric
,
3399 assert(metric
>= 0);
3400 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
3403 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
3406 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
3407 u
->ip_accounting_ingress_map_fd
:
3408 u
->ip_accounting_egress_map_fd
;
3412 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
3413 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
3415 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
3419 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
3420 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
3421 * ip_accounting_extra[] field, and add them in here transparently. */
3423 *ret
= value
+ u
->ip_accounting_extra
[metric
];
3428 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
3429 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
3430 [CGROUP_IO_READ_BYTES
] = "rbytes=",
3431 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
3432 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
3433 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
3435 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
3436 _cleanup_free_
char *path
= NULL
;
3437 _cleanup_fclose_
FILE *f
= NULL
;
3442 if (!u
->cgroup_path
)
3445 if (unit_has_host_root_cgroup(u
))
3446 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
3448 r
= cg_all_unified();
3451 if (r
== 0) /* TODO: support cgroupv1 */
3454 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
3457 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
3461 f
= fopen(path
, "re");
3466 _cleanup_free_
char *line
= NULL
;
3469 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3476 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
3477 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
3480 _cleanup_free_
char *word
= NULL
;
3482 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
3488 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3491 x
= startswith(word
, field_names
[i
]);
3495 r
= safe_atou64(x
, &w
);
3499 /* Sum up the stats of all devices */
3507 memcpy(ret
, acc
, sizeof(acc
));
3511 int unit_get_io_accounting(
3513 CGroupIOAccountingMetric metric
,
3517 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
3520 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
3522 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
3525 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3528 r
= unit_get_io_accounting_raw(u
, raw
);
3529 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3534 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3535 /* Saturated subtraction */
3536 if (raw
[i
] > u
->io_accounting_base
[i
])
3537 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
3539 u
->io_accounting_last
[i
] = 0;
3544 *ret
= u
->io_accounting_last
[metric
];
3549 int unit_reset_cpu_accounting(Unit
*u
) {
3554 u
->cpu_usage_last
= NSEC_INFINITY
;
3556 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
3558 u
->cpu_usage_base
= 0;
3565 int unit_reset_ip_accounting(Unit
*u
) {
3570 if (u
->ip_accounting_ingress_map_fd
>= 0)
3571 r
= bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
);
3573 if (u
->ip_accounting_egress_map_fd
>= 0)
3574 q
= bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
);
3576 zero(u
->ip_accounting_extra
);
3578 return r
< 0 ? r
: q
;
3581 int unit_reset_io_accounting(Unit
*u
) {
3586 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++)
3587 u
->io_accounting_last
[i
] = UINT64_MAX
;
3589 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
3591 zero(u
->io_accounting_base
);
3598 int unit_reset_accounting(Unit
*u
) {
3603 r
= unit_reset_cpu_accounting(u
);
3604 q
= unit_reset_io_accounting(u
);
3605 v
= unit_reset_ip_accounting(u
);
3607 return r
< 0 ? r
: q
< 0 ? q
: v
;
3610 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
3613 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3619 /* always invalidate compat pairs together */
3620 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
3621 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
3623 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
3624 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
3626 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
3629 u
->cgroup_invalidated_mask
|= m
;
3630 unit_add_to_cgroup_realize_queue(u
);
3633 void unit_invalidate_cgroup_bpf(Unit
*u
) {
3636 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3639 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
3642 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
3643 unit_add_to_cgroup_realize_queue(u
);
3645 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
3646 * list of our children includes our own. */
3647 if (u
->type
== UNIT_SLICE
) {
3651 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
])
3652 if (UNIT_DEREF(member
->slice
) == u
)
3653 unit_invalidate_cgroup_bpf(member
);
3657 bool unit_cgroup_delegate(Unit
*u
) {
3662 if (!UNIT_VTABLE(u
)->can_delegate
)
3665 c
= unit_get_cgroup_context(u
);
3672 void manager_invalidate_startup_units(Manager
*m
) {
3677 SET_FOREACH(u
, m
->startup_units
)
3678 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
3681 static int unit_get_nice(Unit
*u
) {
3684 ec
= unit_get_exec_context(u
);
3685 return ec
? ec
->nice
: 0;
3688 static uint64_t unit_get_cpu_weight(Unit
*u
) {
3689 ManagerState state
= manager_state(u
->manager
);
3692 cc
= unit_get_cgroup_context(u
);
3693 return cc
? cgroup_context_cpu_weight(cc
, state
) : CGROUP_WEIGHT_DEFAULT
;
3696 int compare_job_priority(const void *a
, const void *b
) {
3697 const Job
*x
= a
, *y
= b
;
3699 uint64_t weight_x
, weight_y
;
3702 if ((ret
= CMP(x
->unit
->type
, y
->unit
->type
)) != 0)
3705 weight_x
= unit_get_cpu_weight(x
->unit
);
3706 weight_y
= unit_get_cpu_weight(y
->unit
);
3708 if ((ret
= CMP(weight_x
, weight_y
)) != 0)
3711 nice_x
= unit_get_nice(x
->unit
);
3712 nice_y
= unit_get_nice(y
->unit
);
3714 if ((ret
= CMP(nice_x
, nice_y
)) != 0)
3717 return strcmp(x
->unit
->id
, y
->unit
->id
);
3720 int unit_cgroup_freezer_action(Unit
*u
, FreezerAction action
) {
3721 _cleanup_free_
char *path
= NULL
;
3722 FreezerState target
, kernel
= _FREEZER_STATE_INVALID
;
3726 assert(IN_SET(action
, FREEZER_FREEZE
, FREEZER_THAW
));
3728 if (!cg_freezer_supported())
3731 if (!u
->cgroup_realized
)
3734 target
= action
== FREEZER_FREEZE
? FREEZER_FROZEN
: FREEZER_RUNNING
;
3736 r
= unit_freezer_state_kernel(u
, &kernel
);
3738 log_unit_debug_errno(u
, r
, "Failed to obtain cgroup freezer state: %m");
3740 if (target
== kernel
) {
3741 u
->freezer_state
= target
;
3745 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.freeze", &path
);
3749 log_unit_debug(u
, "%s unit.", action
== FREEZER_FREEZE
? "Freezing" : "Thawing");
3751 if (action
== FREEZER_FREEZE
)
3752 u
->freezer_state
= FREEZER_FREEZING
;
3754 u
->freezer_state
= FREEZER_THAWING
;
3756 r
= write_string_file(path
, one_zero(action
== FREEZER_FREEZE
), WRITE_STRING_FILE_DISABLE_BUFFER
);
3763 int unit_get_cpuset(Unit
*u
, CPUSet
*cpus
, const char *name
) {
3764 _cleanup_free_
char *v
= NULL
;
3770 if (!u
->cgroup_path
)
3773 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUSET
) == 0)
3776 r
= cg_all_unified();
3782 r
= cg_get_attribute("cpuset", u
->cgroup_path
, name
, &v
);
3788 return parse_cpu_set_full(v
, cpus
, false, NULL
, NULL
, 0, NULL
);
3791 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
3792 [CGROUP_DEVICE_POLICY_AUTO
] = "auto",
3793 [CGROUP_DEVICE_POLICY_CLOSED
] = "closed",
3794 [CGROUP_DEVICE_POLICY_STRICT
] = "strict",
3797 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);
3799 static const char* const freezer_action_table
[_FREEZER_ACTION_MAX
] = {
3800 [FREEZER_FREEZE
] = "freeze",
3801 [FREEZER_THAW
] = "thaw",
3804 DEFINE_STRING_TABLE_LOOKUP(freezer_action
, FreezerAction
);