1 /* SPDX-License-Identifier: LGPL-2.1+ */
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 "process-util.h"
25 #include "procfs-util.h"
27 #include "stat-util.h"
28 #include "stdio-util.h"
29 #include "string-table.h"
30 #include "string-util.h"
33 #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
35 /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access
36 * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask
37 * out specific attributes from us. */
38 #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING)
40 uint64_t tasks_max_resolve(const TasksMax
*tasks_max
) {
41 if (tasks_max
->scale
== 0)
42 return tasks_max
->value
;
44 return system_tasks_max_scale(tasks_max
->value
, tasks_max
->scale
);
47 bool manager_owns_host_root_cgroup(Manager
*m
) {
50 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
51 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
52 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
53 * we run in any kind of container virtualization. */
55 if (MANAGER_IS_USER(m
))
58 if (detect_container() > 0)
61 return empty_or_root(m
->cgroup_root
);
64 bool unit_has_host_root_cgroup(Unit
*u
) {
67 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
68 * the manager manages the root cgroup. */
70 if (!manager_owns_host_root_cgroup(u
->manager
))
73 return unit_has_name(u
, SPECIAL_ROOT_SLICE
);
76 static int set_attribute_and_warn(Unit
*u
, const char *controller
, const char *attribute
, const char *value
) {
79 r
= cg_set_attribute(controller
, u
->cgroup_path
, attribute
, value
);
81 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%.*s': %m",
82 strna(attribute
), isempty(u
->cgroup_path
) ? "/" : u
->cgroup_path
, (int) strcspn(value
, NEWLINE
), value
);
87 static void cgroup_compat_warn(void) {
88 static bool cgroup_compat_warned
= false;
90 if (cgroup_compat_warned
)
93 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
94 "See cgroup-compat debug messages for details.");
96 cgroup_compat_warned
= true;
99 #define log_cgroup_compat(unit, fmt, ...) do { \
100 cgroup_compat_warn(); \
101 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
104 void cgroup_context_init(CGroupContext
*c
) {
107 /* Initialize everything to the kernel defaults. */
109 *c
= (CGroupContext
) {
110 .cpu_weight
= CGROUP_WEIGHT_INVALID
,
111 .startup_cpu_weight
= CGROUP_WEIGHT_INVALID
,
112 .cpu_quota_per_sec_usec
= USEC_INFINITY
,
113 .cpu_quota_period_usec
= USEC_INFINITY
,
115 .cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
116 .startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
118 .memory_high
= CGROUP_LIMIT_MAX
,
119 .memory_max
= CGROUP_LIMIT_MAX
,
120 .memory_swap_max
= CGROUP_LIMIT_MAX
,
122 .memory_limit
= CGROUP_LIMIT_MAX
,
124 .io_weight
= CGROUP_WEIGHT_INVALID
,
125 .startup_io_weight
= CGROUP_WEIGHT_INVALID
,
127 .blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
128 .startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
130 .tasks_max
= TASKS_MAX_UNSET
,
134 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
138 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
143 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
147 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
152 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
156 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
161 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
165 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
170 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
174 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
179 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
183 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
188 void cgroup_context_done(CGroupContext
*c
) {
191 while (c
->io_device_weights
)
192 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
194 while (c
->io_device_latencies
)
195 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
197 while (c
->io_device_limits
)
198 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
200 while (c
->blockio_device_weights
)
201 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
203 while (c
->blockio_device_bandwidths
)
204 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
206 while (c
->device_allow
)
207 cgroup_context_free_device_allow(c
, c
->device_allow
);
209 c
->ip_address_allow
= ip_address_access_free_all(c
->ip_address_allow
);
210 c
->ip_address_deny
= ip_address_access_free_all(c
->ip_address_deny
);
212 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
213 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
215 cpu_set_reset(&c
->cpuset_cpus
);
216 cpu_set_reset(&c
->cpuset_mems
);
219 static int unit_get_kernel_memory_limit(Unit
*u
, const char *file
, uint64_t *ret
) {
222 if (!u
->cgroup_realized
)
225 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, file
, ret
);
228 static int unit_compare_memory_limit(Unit
*u
, const char *property_name
, uint64_t *ret_unit_value
, uint64_t *ret_kernel_value
) {
235 /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will
236 * return -ENODATA) on cgroup v1.
241 * 0: If the kernel memory setting doesn't match our configuration.
242 * >0: If the kernel memory setting matches our configuration.
244 * The following values are only guaranteed to be populated on return >=0:
246 * - ret_unit_value will contain our internal expected value for the unit, page-aligned.
247 * - ret_kernel_value will contain the actual value presented by the kernel. */
251 r
= cg_all_unified();
253 return log_debug_errno(r
, "Failed to determine cgroup hierarchy version: %m");
255 /* Unsupported on v1.
257 * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has
258 * silently masked the controller. */
262 /* The root slice doesn't have any controller files, so we can't compare anything. */
263 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
266 /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled,
267 * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To
268 * avoid specious errors in these scenarios, check that we even expect the memory controller to be
270 m
= unit_get_target_mask(u
);
271 if (!FLAGS_SET(m
, CGROUP_MASK_MEMORY
))
274 c
= unit_get_cgroup_context(u
);
277 if (streq(property_name
, "MemoryLow")) {
278 unit_value
= unit_get_ancestor_memory_low(u
);
280 } else if (streq(property_name
, "MemoryMin")) {
281 unit_value
= unit_get_ancestor_memory_min(u
);
283 } else if (streq(property_name
, "MemoryHigh")) {
284 unit_value
= c
->memory_high
;
285 file
= "memory.high";
286 } else if (streq(property_name
, "MemoryMax")) {
287 unit_value
= c
->memory_max
;
289 } else if (streq(property_name
, "MemorySwapMax")) {
290 unit_value
= c
->memory_swap_max
;
291 file
= "memory.swap.max";
295 r
= unit_get_kernel_memory_limit(u
, file
, ret_kernel_value
);
297 return log_unit_debug_errno(u
, r
, "Failed to parse %s: %m", file
);
299 /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page
300 * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from
301 * our internal page-counting. To support those future kernels, just check the value itself first
302 * without any page-alignment. */
303 if (*ret_kernel_value
== unit_value
) {
304 *ret_unit_value
= unit_value
;
308 /* The current kernel behaviour, by comparison, is that even if you write a particular number of
309 * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel
310 * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is
312 if (unit_value
!= CGROUP_LIMIT_MAX
)
313 unit_value
= PAGE_ALIGN_DOWN(unit_value
);
315 *ret_unit_value
= unit_value
;
317 return *ret_kernel_value
== *ret_unit_value
;
320 #define FORMAT_CGROUP_DIFF_MAX 128
322 static char *format_cgroup_memory_limit_comparison(char *buf
, size_t l
, Unit
*u
, const char *property_name
) {
330 r
= unit_compare_memory_limit(u
, property_name
, &sval
, &kval
);
332 /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1).
333 * In the absence of reliably being able to detect whether memcg swap support is available or not,
334 * only complain if the error is not ENOENT. */
335 if (r
> 0 || IN_SET(r
, -ENODATA
, -EOWNERDEAD
) ||
336 (r
== -ENOENT
&& streq(property_name
, "MemorySwapMax"))) {
342 snprintf(buf
, l
, " (error getting kernel value: %s)", strerror_safe(r
));
346 snprintf(buf
, l
, " (different value in kernel: %" PRIu64
")", kval
);
351 void cgroup_context_dump(Unit
*u
, FILE* f
, const char *prefix
) {
352 _cleanup_free_
char *disable_controllers_str
= NULL
, *cpuset_cpus
= NULL
, *cpuset_mems
= NULL
;
353 CGroupIODeviceLimit
*il
;
354 CGroupIODeviceWeight
*iw
;
355 CGroupIODeviceLatency
*l
;
356 CGroupBlockIODeviceBandwidth
*b
;
357 CGroupBlockIODeviceWeight
*w
;
358 CGroupDeviceAllow
*a
;
360 IPAddressAccessItem
*iaai
;
362 char q
[FORMAT_TIMESPAN_MAX
];
363 char v
[FORMAT_TIMESPAN_MAX
];
365 char cda
[FORMAT_CGROUP_DIFF_MAX
];
366 char cdb
[FORMAT_CGROUP_DIFF_MAX
];
367 char cdc
[FORMAT_CGROUP_DIFF_MAX
];
368 char cdd
[FORMAT_CGROUP_DIFF_MAX
];
369 char cde
[FORMAT_CGROUP_DIFF_MAX
];
374 c
= unit_get_cgroup_context(u
);
377 prefix
= strempty(prefix
);
379 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
381 cpuset_cpus
= cpu_set_to_range_string(&c
->cpuset_cpus
);
382 cpuset_mems
= cpu_set_to_range_string(&c
->cpuset_mems
);
385 "%sCPUAccounting: %s\n"
386 "%sIOAccounting: %s\n"
387 "%sBlockIOAccounting: %s\n"
388 "%sMemoryAccounting: %s\n"
389 "%sTasksAccounting: %s\n"
390 "%sIPAccounting: %s\n"
391 "%sCPUWeight: %" PRIu64
"\n"
392 "%sStartupCPUWeight: %" PRIu64
"\n"
393 "%sCPUShares: %" PRIu64
"\n"
394 "%sStartupCPUShares: %" PRIu64
"\n"
395 "%sCPUQuotaPerSecSec: %s\n"
396 "%sCPUQuotaPeriodSec: %s\n"
397 "%sAllowedCPUs: %s\n"
398 "%sAllowedMemoryNodes: %s\n"
399 "%sIOWeight: %" PRIu64
"\n"
400 "%sStartupIOWeight: %" PRIu64
"\n"
401 "%sBlockIOWeight: %" PRIu64
"\n"
402 "%sStartupBlockIOWeight: %" PRIu64
"\n"
403 "%sDefaultMemoryMin: %" PRIu64
"\n"
404 "%sDefaultMemoryLow: %" PRIu64
"\n"
405 "%sMemoryMin: %" PRIu64
"%s\n"
406 "%sMemoryLow: %" PRIu64
"%s\n"
407 "%sMemoryHigh: %" PRIu64
"%s\n"
408 "%sMemoryMax: %" PRIu64
"%s\n"
409 "%sMemorySwapMax: %" PRIu64
"%s\n"
410 "%sMemoryLimit: %" PRIu64
"\n"
411 "%sTasksMax: %" PRIu64
"\n"
412 "%sDevicePolicy: %s\n"
413 "%sDisableControllers: %s\n"
415 prefix
, yes_no(c
->cpu_accounting
),
416 prefix
, yes_no(c
->io_accounting
),
417 prefix
, yes_no(c
->blockio_accounting
),
418 prefix
, yes_no(c
->memory_accounting
),
419 prefix
, yes_no(c
->tasks_accounting
),
420 prefix
, yes_no(c
->ip_accounting
),
421 prefix
, c
->cpu_weight
,
422 prefix
, c
->startup_cpu_weight
,
423 prefix
, c
->cpu_shares
,
424 prefix
, c
->startup_cpu_shares
,
425 prefix
, format_timespan(q
, sizeof(q
), c
->cpu_quota_per_sec_usec
, 1),
426 prefix
, format_timespan(v
, sizeof(v
), c
->cpu_quota_period_usec
, 1),
427 prefix
, strempty(cpuset_cpus
),
428 prefix
, strempty(cpuset_mems
),
429 prefix
, c
->io_weight
,
430 prefix
, c
->startup_io_weight
,
431 prefix
, c
->blockio_weight
,
432 prefix
, c
->startup_blockio_weight
,
433 prefix
, c
->default_memory_min
,
434 prefix
, c
->default_memory_low
,
435 prefix
, c
->memory_min
, format_cgroup_memory_limit_comparison(cda
, sizeof(cda
), u
, "MemoryMin"),
436 prefix
, c
->memory_low
, format_cgroup_memory_limit_comparison(cdb
, sizeof(cdb
), u
, "MemoryLow"),
437 prefix
, c
->memory_high
, format_cgroup_memory_limit_comparison(cdc
, sizeof(cdc
), u
, "MemoryHigh"),
438 prefix
, c
->memory_max
, format_cgroup_memory_limit_comparison(cdd
, sizeof(cdd
), u
, "MemoryMax"),
439 prefix
, c
->memory_swap_max
, format_cgroup_memory_limit_comparison(cde
, sizeof(cde
), u
, "MemorySwapMax"),
440 prefix
, c
->memory_limit
,
441 prefix
, tasks_max_resolve(&c
->tasks_max
),
442 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
443 prefix
, strempty(disable_controllers_str
),
444 prefix
, yes_no(c
->delegate
));
447 _cleanup_free_
char *t
= NULL
;
449 (void) cg_mask_to_string(c
->delegate_controllers
, &t
);
451 fprintf(f
, "%sDelegateControllers: %s\n",
456 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
458 "%sDeviceAllow: %s %s%s%s\n",
461 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
463 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
465 "%sIODeviceWeight: %s %" PRIu64
"\n",
470 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
472 "%sIODeviceLatencyTargetSec: %s %s\n",
475 format_timespan(q
, sizeof(q
), l
->target_usec
, 1));
477 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
478 char buf
[FORMAT_BYTES_MAX
];
479 CGroupIOLimitType type
;
481 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
482 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
486 cgroup_io_limit_type_to_string(type
),
488 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
491 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
493 "%sBlockIODeviceWeight: %s %" PRIu64
,
498 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
499 char buf
[FORMAT_BYTES_MAX
];
501 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
503 "%sBlockIOReadBandwidth: %s %s\n",
506 format_bytes(buf
, sizeof(buf
), b
->rbps
));
507 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
509 "%sBlockIOWriteBandwidth: %s %s\n",
512 format_bytes(buf
, sizeof(buf
), b
->wbps
));
515 LIST_FOREACH(items
, iaai
, c
->ip_address_allow
) {
516 _cleanup_free_
char *k
= NULL
;
518 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
519 fprintf(f
, "%sIPAddressAllow: %s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
522 LIST_FOREACH(items
, iaai
, c
->ip_address_deny
) {
523 _cleanup_free_
char *k
= NULL
;
525 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
526 fprintf(f
, "%sIPAddressDeny: %s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
529 STRV_FOREACH(path
, c
->ip_filters_ingress
)
530 fprintf(f
, "%sIPIngressFilterPath: %s\n", prefix
, *path
);
532 STRV_FOREACH(path
, c
->ip_filters_egress
)
533 fprintf(f
, "%sIPEgressFilterPath: %s\n", prefix
, *path
);
536 int cgroup_add_device_allow(CGroupContext
*c
, const char *dev
, const char *mode
) {
537 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
538 _cleanup_free_
char *d
= NULL
;
542 assert(isempty(mode
) || in_charset(mode
, "rwm"));
544 a
= new(CGroupDeviceAllow
, 1);
552 *a
= (CGroupDeviceAllow
) {
554 .r
= isempty(mode
) || strchr(mode
, 'r'),
555 .w
= isempty(mode
) || strchr(mode
, 'w'),
556 .m
= isempty(mode
) || strchr(mode
, 'm'),
559 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
565 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
566 uint64_t unit_get_ancestor_##entry(Unit *u) { \
569 /* 1. Is entry set in this unit? If so, use that. \
570 * 2. Is the default for this entry set in any \
571 * ancestor? If so, use that. \
572 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
576 c = unit_get_cgroup_context(u); \
577 if (c && c->entry##_set) \
580 while ((u = UNIT_DEREF(u->slice))) { \
581 c = unit_get_cgroup_context(u); \
582 if (c && c->default_##entry##_set) \
583 return c->default_##entry; \
586 /* We've reached the root, but nobody had default for \
587 * this entry set, so set it to the kernel default. */ \
588 return CGROUP_LIMIT_MIN; \
591 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
592 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
594 static void cgroup_xattr_apply(Unit
*u
) {
595 char ids
[SD_ID128_STRING_MAX
];
600 if (!MANAGER_IS_SYSTEM(u
->manager
))
603 if (!sd_id128_is_null(u
->invocation_id
)) {
604 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
605 "trusted.invocation_id",
606 sd_id128_to_string(u
->invocation_id
, ids
), 32,
609 log_unit_debug_errno(u
, r
, "Failed to set invocation ID on control group %s, ignoring: %m", u
->cgroup_path
);
612 if (unit_cgroup_delegate(u
)) {
613 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
618 log_unit_debug_errno(u
, r
, "Failed to set delegate flag on control group %s, ignoring: %m", u
->cgroup_path
);
620 r
= cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "trusted.delegate");
622 log_unit_debug_errno(u
, r
, "Failed to remove delegate flag on control group %s, ignoring: %m", u
->cgroup_path
);
626 static int lookup_block_device(const char *p
, dev_t
*ret
) {
634 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
635 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
638 if (stat(p
, &st
) < 0)
639 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
645 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
648 return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK
),
649 "Device node '%s' is a character device, but block device needed.", p
);
652 else if (major(dev
) != 0)
653 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
655 /* If this is btrfs, getting the backing block device is a bit harder */
656 r
= btrfs_get_block_device(p
, ret
);
658 return log_warning_errno(SYNTHETIC_ERRNO(ENODEV
),
659 "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
661 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
664 /* If this is a LUKS/DM device, recursively try to get the originating block device */
665 while (block_get_originating(*ret
, ret
) > 0);
667 /* If this is a partition, try to get the originating block device */
668 (void) block_get_whole_disk(*ret
, ret
);
672 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
673 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
674 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
677 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
678 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
679 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
682 static uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
683 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
684 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
685 return c
->startup_cpu_weight
;
686 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
687 return c
->cpu_weight
;
689 return CGROUP_WEIGHT_DEFAULT
;
692 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
693 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
694 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
695 return c
->startup_cpu_shares
;
696 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
697 return c
->cpu_shares
;
699 return CGROUP_CPU_SHARES_DEFAULT
;
702 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
703 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
704 * need to be higher than that boundary. quota is specified in USecPerSec.
705 * Additionally, period must be at most max_period. */
708 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
711 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
714 if (quota
== USEC_INFINITY
)
715 /* Always use default period for infinity quota. */
716 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
718 if (period
== USEC_INFINITY
)
719 /* Default period was requested. */
720 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
722 /* Clamp to interval [1ms, 1s] */
723 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
725 if (new_period
!= period
) {
726 char v
[FORMAT_TIMESPAN_MAX
];
727 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
,
728 "Clamping CPU interval for cpu.max: period is now %s",
729 format_timespan(v
, sizeof(v
), new_period
, 1));
730 u
->warned_clamping_cpu_quota_period
= true;
736 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
737 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
739 xsprintf(buf
, "%" PRIu64
"\n", weight
);
740 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
743 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
744 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
746 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
747 if (quota
!= USEC_INFINITY
)
748 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
749 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
751 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
752 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
755 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
756 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
758 xsprintf(buf
, "%" PRIu64
"\n", shares
);
759 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
762 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
763 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
765 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
767 xsprintf(buf
, USEC_FMT
"\n", period
);
768 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
770 if (quota
!= USEC_INFINITY
) {
771 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
772 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
774 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
777 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
778 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
779 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
782 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
783 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
784 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
787 static void cgroup_apply_unified_cpuset(Unit
*u
, const CPUSet
*cpus
, const char *name
) {
788 _cleanup_free_
char *buf
= NULL
;
790 buf
= cpu_set_to_range_string(cpus
);
796 (void) set_attribute_and_warn(u
, "cpuset", name
, buf
);
799 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
800 return c
->io_accounting
||
801 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
802 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
803 c
->io_device_weights
||
804 c
->io_device_latencies
||
808 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
809 return c
->blockio_accounting
||
810 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
811 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
812 c
->blockio_device_weights
||
813 c
->blockio_device_bandwidths
;
816 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
817 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
818 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
819 return c
->startup_io_weight
;
820 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
823 return CGROUP_WEIGHT_DEFAULT
;
826 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
827 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
828 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
829 return c
->startup_blockio_weight
;
830 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
831 return c
->blockio_weight
;
833 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
836 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
837 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
838 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
841 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
842 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
843 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
846 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
847 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
851 r
= lookup_block_device(dev_path
, &dev
);
855 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
856 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
859 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
860 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
864 r
= lookup_block_device(dev_path
, &dev
);
868 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
869 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
872 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
873 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
877 r
= lookup_block_device(dev_path
, &dev
);
881 if (target
!= USEC_INFINITY
)
882 xsprintf(buf
, "%u:%u target=%" PRIu64
"\n", major(dev
), minor(dev
), target
);
884 xsprintf(buf
, "%u:%u target=max\n", major(dev
), minor(dev
));
886 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
889 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
890 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
891 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
892 CGroupIOLimitType type
;
896 r
= lookup_block_device(dev_path
, &dev
);
900 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
901 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
902 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
904 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
906 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
907 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
908 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
909 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
912 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
913 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
917 r
= lookup_block_device(dev_path
, &dev
);
921 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
922 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
924 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
925 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
928 static bool unit_has_unified_memory_config(Unit
*u
) {
933 c
= unit_get_cgroup_context(u
);
936 return unit_get_ancestor_memory_min(u
) > 0 || unit_get_ancestor_memory_low(u
) > 0 ||
937 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
||
938 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
;
941 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
942 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
944 if (v
!= CGROUP_LIMIT_MAX
)
945 xsprintf(buf
, "%" PRIu64
"\n", v
);
947 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
950 static void cgroup_apply_firewall(Unit
*u
) {
953 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
955 if (bpf_firewall_compile(u
) < 0)
958 (void) bpf_firewall_load_custom(u
);
959 (void) bpf_firewall_install(u
);
962 static int cgroup_apply_devices(Unit
*u
) {
963 _cleanup_(bpf_program_unrefp
) BPFProgram
*prog
= NULL
;
966 CGroupDeviceAllow
*a
;
967 CGroupDevicePolicy policy
;
970 assert_se(c
= unit_get_cgroup_context(u
));
971 assert_se(path
= u
->cgroup_path
);
973 policy
= c
->device_policy
;
975 if (cg_all_unified() > 0) {
976 r
= bpf_devices_cgroup_init(&prog
, policy
, c
->device_allow
);
978 return log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
981 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore
984 if (c
->device_allow
|| policy
!= CGROUP_DEVICE_POLICY_AUTO
)
985 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
987 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
989 log_unit_full_errno(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
990 "Failed to reset devices.allow/devices.deny: %m");
993 bool allow_list_static
= policy
== CGROUP_DEVICE_POLICY_CLOSED
||
994 (policy
== CGROUP_DEVICE_POLICY_AUTO
&& c
->device_allow
);
995 if (allow_list_static
)
996 (void) bpf_devices_allow_list_static(prog
, path
);
998 bool any
= allow_list_static
;
999 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1013 if (path_startswith(a
->path
, "/dev/"))
1014 r
= bpf_devices_allow_list_device(prog
, path
, a
->path
, acc
);
1015 else if ((val
= startswith(a
->path
, "block-")))
1016 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'b', acc
);
1017 else if ((val
= startswith(a
->path
, "char-")))
1018 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'c', acc
);
1020 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1029 log_unit_warning_errno(u
, SYNTHETIC_ERRNO(ENODEV
), "No devices matched by device filter.");
1031 /* The kernel verifier would reject a program we would build with the normal intro and outro
1032 but no allow-listing rules (outro would contain an unreachable instruction for successful
1034 policy
= CGROUP_DEVICE_POLICY_STRICT
;
1037 r
= bpf_devices_apply_policy(prog
, policy
, any
, path
, &u
->bpf_device_control_installed
);
1039 static bool warned
= false;
1041 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1042 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1043 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1044 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1051 static void cgroup_context_apply(
1053 CGroupMask apply_mask
,
1054 ManagerState state
) {
1058 bool is_host_root
, is_local_root
;
1063 /* Nothing to do? Exit early! */
1064 if (apply_mask
== 0)
1067 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
1068 * attributes should only be managed for cgroups further down the tree. */
1069 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1070 is_host_root
= unit_has_host_root_cgroup(u
);
1072 assert_se(c
= unit_get_cgroup_context(u
));
1073 assert_se(path
= u
->cgroup_path
);
1075 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
1078 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
1079 * then), and missing cgroups, i.e. EROFS and ENOENT. */
1081 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
1082 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
1083 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
1084 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
1085 * we couldn't even write to them if we wanted to). */
1086 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
1088 if (cg_all_unified() > 0) {
1091 if (cgroup_context_has_cpu_weight(c
))
1092 weight
= cgroup_context_cpu_weight(c
, state
);
1093 else if (cgroup_context_has_cpu_shares(c
)) {
1096 shares
= cgroup_context_cpu_shares(c
, state
);
1097 weight
= cgroup_cpu_shares_to_weight(shares
);
1099 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
1100 shares
, weight
, path
);
1102 weight
= CGROUP_WEIGHT_DEFAULT
;
1104 cgroup_apply_unified_cpu_weight(u
, weight
);
1105 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1110 if (cgroup_context_has_cpu_weight(c
)) {
1113 weight
= cgroup_context_cpu_weight(c
, state
);
1114 shares
= cgroup_cpu_weight_to_shares(weight
);
1116 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
1117 weight
, shares
, path
);
1118 } else if (cgroup_context_has_cpu_shares(c
))
1119 shares
= cgroup_context_cpu_shares(c
, state
);
1121 shares
= CGROUP_CPU_SHARES_DEFAULT
;
1123 cgroup_apply_legacy_cpu_shares(u
, shares
);
1124 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1128 if ((apply_mask
& CGROUP_MASK_CPUSET
) && !is_local_root
) {
1129 cgroup_apply_unified_cpuset(u
, &c
->cpuset_cpus
, "cpuset.cpus");
1130 cgroup_apply_unified_cpuset(u
, &c
->cpuset_mems
, "cpuset.mems");
1133 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
1134 * controller), and in case of containers we want to leave control of these attributes to the container manager
1135 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
1136 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
1137 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
1138 bool has_io
, has_blockio
;
1141 has_io
= cgroup_context_has_io_config(c
);
1142 has_blockio
= cgroup_context_has_blockio_config(c
);
1145 weight
= cgroup_context_io_weight(c
, state
);
1146 else if (has_blockio
) {
1147 uint64_t blkio_weight
;
1149 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
1150 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
1152 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
1153 blkio_weight
, weight
);
1155 weight
= CGROUP_WEIGHT_DEFAULT
;
1157 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1158 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1160 /* FIXME: drop this when distro kernels properly support BFQ through "io.weight"
1161 * See also: https://github.com/systemd/systemd/pull/13335 */
1162 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1163 (void) set_attribute_and_warn(u
, "io", "io.bfq.weight", buf
);
1166 CGroupIODeviceLatency
*latency
;
1167 CGroupIODeviceLimit
*limit
;
1168 CGroupIODeviceWeight
*w
;
1170 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
1171 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
1173 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
1174 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
1176 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
1177 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
1179 } else if (has_blockio
) {
1180 CGroupBlockIODeviceWeight
*w
;
1181 CGroupBlockIODeviceBandwidth
*b
;
1183 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
1184 weight
= cgroup_weight_blkio_to_io(w
->weight
);
1186 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
1187 w
->weight
, weight
, w
->path
);
1189 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
1192 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
1193 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
1194 CGroupIOLimitType type
;
1196 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1197 limits
[type
] = cgroup_io_limit_defaults
[type
];
1199 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
1200 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
1202 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
1203 b
->rbps
, b
->wbps
, b
->path
);
1205 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
1210 if (apply_mask
& CGROUP_MASK_BLKIO
) {
1211 bool has_io
, has_blockio
;
1213 has_io
= cgroup_context_has_io_config(c
);
1214 has_blockio
= cgroup_context_has_blockio_config(c
);
1216 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
1217 * left to our container manager, too. */
1218 if (!is_local_root
) {
1219 char buf
[DECIMAL_STR_MAX(uint64_t)+1];
1225 io_weight
= cgroup_context_io_weight(c
, state
);
1226 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
1228 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
1230 } else if (has_blockio
)
1231 weight
= cgroup_context_blkio_weight(c
, state
);
1233 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
1235 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1236 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1239 CGroupIODeviceWeight
*w
;
1241 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
1242 weight
= cgroup_weight_io_to_blkio(w
->weight
);
1244 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
1245 w
->weight
, weight
, w
->path
);
1247 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
1249 } else if (has_blockio
) {
1250 CGroupBlockIODeviceWeight
*w
;
1252 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
1253 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
1257 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
1258 * roots, as there we want the container manager to handle it */
1259 if (is_host_root
|| !is_local_root
) {
1261 CGroupIODeviceLimit
*l
;
1263 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
1264 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
1265 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
1267 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
1269 } else if (has_blockio
) {
1270 CGroupBlockIODeviceBandwidth
*b
;
1272 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
1273 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
1278 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
1279 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
1280 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
1281 * write to this if we wanted to.) */
1282 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
1284 if (cg_all_unified() > 0) {
1285 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
;
1287 if (unit_has_unified_memory_config(u
)) {
1288 max
= c
->memory_max
;
1289 swap_max
= c
->memory_swap_max
;
1291 max
= c
->memory_limit
;
1293 if (max
!= CGROUP_LIMIT_MAX
)
1294 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
1297 cgroup_apply_unified_memory_limit(u
, "memory.min", unit_get_ancestor_memory_min(u
));
1298 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
1299 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
1300 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
1301 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
1303 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
1306 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1309 if (unit_has_unified_memory_config(u
)) {
1310 val
= c
->memory_max
;
1311 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIi64
" as MemoryLimit=", val
);
1313 val
= c
->memory_limit
;
1315 if (val
== CGROUP_LIMIT_MAX
)
1316 strncpy(buf
, "-1\n", sizeof(buf
));
1318 xsprintf(buf
, "%" PRIu64
"\n", val
);
1320 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
1324 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
1325 * containers, where we leave this to the manager */
1326 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
1327 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
))
1328 (void) cgroup_apply_devices(u
);
1330 if (apply_mask
& CGROUP_MASK_PIDS
) {
1333 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1334 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1335 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1336 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1337 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1338 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1339 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1340 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1341 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1342 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1343 * which is desirable so that there's an official way to release control of the sysctl from
1344 * systemd: set the limit to unbounded and reload. */
1346 if (tasks_max_isset(&c
->tasks_max
)) {
1347 u
->manager
->sysctl_pid_max_changed
= true;
1348 r
= procfs_tasks_set_limit(tasks_max_resolve(&c
->tasks_max
));
1349 } else if (u
->manager
->sysctl_pid_max_changed
)
1350 r
= procfs_tasks_set_limit(TASKS_MAX
);
1354 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
1355 "Failed to write to tasks limit sysctls: %m");
1358 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
1359 * leave it for the container manager. */
1360 if (!is_local_root
) {
1361 if (tasks_max_isset(&c
->tasks_max
)) {
1362 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1364 xsprintf(buf
, "%" PRIu64
"\n", tasks_max_resolve(&c
->tasks_max
));
1365 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
1367 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
1371 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
1372 cgroup_apply_firewall(u
);
1375 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
1380 c
= unit_get_cgroup_context(u
);
1384 if (c
->ip_accounting
||
1385 c
->ip_address_allow
||
1386 c
->ip_address_deny
||
1387 c
->ip_filters_ingress
||
1388 c
->ip_filters_egress
)
1391 /* If any parent slice has an IP access list defined, it applies too */
1392 for (p
= UNIT_DEREF(u
->slice
); p
; p
= UNIT_DEREF(p
->slice
)) {
1393 c
= unit_get_cgroup_context(p
);
1397 if (c
->ip_address_allow
||
1405 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
1406 CGroupMask mask
= 0;
1411 c
= unit_get_cgroup_context(u
);
1415 /* Figure out which controllers we need, based on the cgroup context object */
1417 if (c
->cpu_accounting
)
1418 mask
|= get_cpu_accounting_mask();
1420 if (cgroup_context_has_cpu_weight(c
) ||
1421 cgroup_context_has_cpu_shares(c
) ||
1422 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
1423 mask
|= CGROUP_MASK_CPU
;
1425 if (c
->cpuset_cpus
.set
|| c
->cpuset_mems
.set
)
1426 mask
|= CGROUP_MASK_CPUSET
;
1428 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
1429 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1431 if (c
->memory_accounting
||
1432 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
1433 unit_has_unified_memory_config(u
))
1434 mask
|= CGROUP_MASK_MEMORY
;
1436 if (c
->device_allow
||
1437 c
->device_policy
!= CGROUP_DEVICE_POLICY_AUTO
)
1438 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
1440 if (c
->tasks_accounting
||
1441 tasks_max_isset(&c
->tasks_max
))
1442 mask
|= CGROUP_MASK_PIDS
;
1444 return CGROUP_MASK_EXTEND_JOINED(mask
);
1447 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
1448 CGroupMask mask
= 0;
1450 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
1453 if (unit_get_needs_bpf_firewall(u
))
1454 mask
|= CGROUP_MASK_BPF_FIREWALL
;
1459 CGroupMask
unit_get_own_mask(Unit
*u
) {
1462 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
1463 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
1465 if (u
->load_state
!= UNIT_LOADED
)
1468 c
= unit_get_cgroup_context(u
);
1472 return unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
);
1475 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
1478 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1479 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1481 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1483 if (!unit_cgroup_delegate(u
))
1486 if (cg_all_unified() <= 0) {
1489 e
= unit_get_exec_context(u
);
1490 if (e
&& !exec_context_maintains_privileges(e
))
1494 assert_se(c
= unit_get_cgroup_context(u
));
1495 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
1498 static CGroupMask
unit_get_subtree_mask(Unit
*u
) {
1500 /* Returns the mask of this subtree, meaning of the group
1501 * itself and its children. */
1503 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
1506 CGroupMask
unit_get_members_mask(Unit
*u
) {
1509 /* Returns the mask of controllers all of the unit's children require, merged */
1511 if (u
->cgroup_members_mask_valid
)
1512 return u
->cgroup_members_mask
; /* Use cached value if possible */
1514 u
->cgroup_members_mask
= 0;
1516 if (u
->type
== UNIT_SLICE
) {
1520 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
])
1521 if (UNIT_DEREF(member
->slice
) == u
)
1522 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
1525 u
->cgroup_members_mask_valid
= true;
1526 return u
->cgroup_members_mask
;
1529 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
1532 /* Returns the mask of controllers all of the unit's siblings
1533 * require, i.e. the members mask of the unit's parent slice
1534 * if there is one. */
1536 if (UNIT_ISSET(u
->slice
))
1537 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
1539 return unit_get_subtree_mask(u
); /* we are the top-level slice */
1542 static CGroupMask
unit_get_disable_mask(Unit
*u
) {
1545 c
= unit_get_cgroup_context(u
);
1549 return c
->disable_controllers
;
1552 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
1556 mask
= unit_get_disable_mask(u
);
1558 /* Returns the mask of controllers which are marked as forcibly
1559 * disabled in any ancestor unit or the unit in question. */
1561 if (UNIT_ISSET(u
->slice
))
1562 mask
|= unit_get_ancestor_disable_mask(UNIT_DEREF(u
->slice
));
1567 CGroupMask
unit_get_target_mask(Unit
*u
) {
1570 /* This returns the cgroup mask of all controllers to enable
1571 * for a specific cgroup, i.e. everything it needs itself,
1572 * plus all that its children need, plus all that its siblings
1573 * need. This is primarily useful on the legacy cgroup
1574 * hierarchy, where we need to duplicate each cgroup in each
1575 * hierarchy that shall be enabled for it. */
1577 mask
= unit_get_own_mask(u
) | unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
1579 if (mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
1580 emit_bpf_firewall_warning(u
);
1582 mask
&= u
->manager
->cgroup_supported
;
1583 mask
&= ~unit_get_ancestor_disable_mask(u
);
1588 CGroupMask
unit_get_enable_mask(Unit
*u
) {
1591 /* This returns the cgroup mask of all controllers to enable
1592 * for the children of a specific cgroup. This is primarily
1593 * useful for the unified cgroup hierarchy, where each cgroup
1594 * controls which controllers are enabled for its children. */
1596 mask
= unit_get_members_mask(u
);
1597 mask
&= u
->manager
->cgroup_supported
;
1598 mask
&= ~unit_get_ancestor_disable_mask(u
);
1603 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
1606 /* Recurse invalidate the member masks cache all the way up the tree */
1607 u
->cgroup_members_mask_valid
= false;
1609 if (UNIT_ISSET(u
->slice
))
1610 unit_invalidate_cgroup_members_masks(UNIT_DEREF(u
->slice
));
1613 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
1615 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1619 if (u
->cgroup_path
&&
1620 u
->cgroup_realized
&&
1621 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
1622 return u
->cgroup_path
;
1624 u
= UNIT_DEREF(u
->slice
);
1630 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1631 /* If not realized at all, migrate to root ("").
1632 * It may happen if we're upgrading from older version that didn't clean up.
1634 return strempty(unit_get_realized_cgroup_path(userdata
, mask
));
1637 char *unit_default_cgroup_path(const Unit
*u
) {
1638 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1643 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1644 return strdup(u
->manager
->cgroup_root
);
1646 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1647 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1652 escaped
= cg_escape(u
->id
);
1656 return path_join(empty_to_root(u
->manager
->cgroup_root
), slice
, escaped
);
1659 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1660 _cleanup_free_
char *p
= NULL
;
1665 if (streq_ptr(u
->cgroup_path
, path
))
1675 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1680 unit_release_cgroup(u
);
1681 u
->cgroup_path
= TAKE_PTR(p
);
1686 int unit_watch_cgroup(Unit
*u
) {
1687 _cleanup_free_
char *events
= NULL
;
1692 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
1693 * cgroupv2 is available. */
1695 if (!u
->cgroup_path
)
1698 if (u
->cgroup_control_inotify_wd
>= 0)
1701 /* Only applies to the unified hierarchy */
1702 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1704 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1708 /* No point in watch the top-level slice, it's never going to run empty. */
1709 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1712 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
1716 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1720 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1721 if (u
->cgroup_control_inotify_wd
< 0) {
1723 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1724 * is not an error */
1727 return log_unit_error_errno(u
, errno
, "Failed to add control inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1730 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
1732 return log_unit_error_errno(u
, r
, "Failed to add control inotify watch descriptor to hash map: %m");
1737 int unit_watch_cgroup_memory(Unit
*u
) {
1738 _cleanup_free_
char *events
= NULL
;
1744 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
1745 * cgroupv2 is available. */
1747 if (!u
->cgroup_path
)
1750 c
= unit_get_cgroup_context(u
);
1754 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
1755 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
1757 if (!c
->memory_accounting
)
1760 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
1761 * we also don't want to generate a log message for each parent cgroup of a process. */
1762 if (u
->type
== UNIT_SLICE
)
1765 if (u
->cgroup_memory_inotify_wd
>= 0)
1768 /* Only applies to the unified hierarchy */
1769 r
= cg_all_unified();
1771 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
1775 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
1779 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
1783 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1784 if (u
->cgroup_memory_inotify_wd
< 0) {
1786 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1787 * is not an error */
1790 return log_unit_error_errno(u
, errno
, "Failed to add memory inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1793 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
1795 return log_unit_error_errno(u
, r
, "Failed to add memory inotify watch descriptor to hash map: %m");
1800 int unit_pick_cgroup_path(Unit
*u
) {
1801 _cleanup_free_
char *path
= NULL
;
1809 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1812 path
= unit_default_cgroup_path(u
);
1816 r
= unit_set_cgroup_path(u
, path
);
1818 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1820 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1825 static int unit_update_cgroup(
1827 CGroupMask target_mask
,
1828 CGroupMask enable_mask
,
1829 ManagerState state
) {
1831 bool created
, is_root_slice
;
1832 CGroupMask migrate_mask
= 0;
1837 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1840 /* Figure out our cgroup path */
1841 r
= unit_pick_cgroup_path(u
);
1845 /* First, create our own group */
1846 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
1848 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
1851 /* Start watching it */
1852 (void) unit_watch_cgroup(u
);
1853 (void) unit_watch_cgroup_memory(u
);
1856 /* For v2 we preserve enabled controllers in delegated units, adjust others,
1857 * for v1 we figure out which controller hierarchies need migration. */
1858 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
1859 CGroupMask result_mask
= 0;
1861 /* Enable all controllers we need */
1862 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
1864 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
1866 /* Remember what's actually enabled now */
1867 u
->cgroup_enabled_mask
= result_mask
;
1869 migrate_mask
= u
->cgroup_realized_mask
^ target_mask
;
1872 /* Keep track that this is now realized */
1873 u
->cgroup_realized
= true;
1874 u
->cgroup_realized_mask
= target_mask
;
1876 /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling).
1878 * Unnecessary controller cgroups are trimmed (after emptied by upward migration).
1879 * We perform migration also with whole slices for cases when users don't care about leave
1880 * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing
1883 if (cg_all_unified() == 0) {
1884 r
= cg_migrate_v1_controllers(u
->manager
->cgroup_supported
, migrate_mask
, u
->cgroup_path
, migrate_callback
, u
);
1886 log_unit_warning_errno(u
, r
, "Failed to migrate controller cgroups from %s, ignoring: %m", u
->cgroup_path
);
1888 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1889 r
= cg_trim_v1_controllers(u
->manager
->cgroup_supported
, ~target_mask
, u
->cgroup_path
, !is_root_slice
);
1891 log_unit_warning_errno(u
, r
, "Failed to delete controller cgroups %s, ignoring: %m", u
->cgroup_path
);
1894 /* Set attributes */
1895 cgroup_context_apply(u
, target_mask
, state
);
1896 cgroup_xattr_apply(u
);
1901 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
1902 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1908 if (MANAGER_IS_SYSTEM(u
->manager
))
1911 if (!u
->manager
->system_bus
)
1914 if (!u
->cgroup_path
)
1917 /* Determine this unit's cgroup path relative to our cgroup root */
1918 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
1922 pp
= strjoina("/", pp
, suffix_path
);
1923 path_simplify(pp
, false);
1925 r
= sd_bus_call_method(u
->manager
->system_bus
,
1926 "org.freedesktop.systemd1",
1927 "/org/freedesktop/systemd1",
1928 "org.freedesktop.systemd1.Manager",
1929 "AttachProcessesToUnit",
1932 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
1934 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
1939 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
1940 CGroupMask delegated_mask
;
1947 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1950 if (set_isempty(pids
))
1953 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
1954 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
1955 r
= bpf_firewall_load_custom(u
);
1959 r
= unit_realize_cgroup(u
);
1963 if (isempty(suffix_path
))
1966 p
= prefix_roota(u
->cgroup_path
, suffix_path
);
1968 delegated_mask
= unit_get_delegate_mask(u
);
1971 SET_FOREACH(pidp
, pids
) {
1972 pid_t pid
= PTR_TO_PID(pidp
);
1975 /* First, attach the PID to the main cgroup hierarchy */
1976 q
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
);
1978 log_unit_debug_errno(u
, q
, "Couldn't move process " PID_FMT
" to requested cgroup '%s': %m", pid
, p
);
1980 if (MANAGER_IS_USER(u
->manager
) && ERRNO_IS_PRIVILEGE(q
)) {
1983 /* If we are in a user instance, and we can't move the process ourselves due to
1984 * permission problems, let's ask the system instance about it instead. Since it's more
1985 * privileged it might be able to move the process across the leaves of a subtree who's
1986 * top node is not owned by us. */
1988 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
, suffix_path
);
1990 log_unit_debug_errno(u
, z
, "Couldn't move process " PID_FMT
" to requested cgroup '%s' via the system bus either: %m", pid
, p
);
1992 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
1996 r
= q
; /* Remember first error */
2001 q
= cg_all_unified();
2007 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
2008 * innermost realized one */
2010 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2011 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2012 const char *realized
;
2014 if (!(u
->manager
->cgroup_supported
& bit
))
2017 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
2018 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
2019 q
= cg_attach(cgroup_controller_to_string(c
), p
, pid
);
2021 continue; /* Success! */
2023 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",
2024 pid
, p
, cgroup_controller_to_string(c
));
2027 /* So this controller is either not delegate or realized, or something else weird happened. In
2028 * that case let's attach the PID at least to the closest cgroup up the tree that is
2030 realized
= unit_get_realized_cgroup_path(u
, bit
);
2032 continue; /* Not even realized in the root slice? Then let's not bother */
2034 q
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
);
2036 log_unit_debug_errno(u
, q
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
2037 pid
, realized
, cgroup_controller_to_string(c
));
2044 static bool unit_has_mask_realized(
2046 CGroupMask target_mask
,
2047 CGroupMask enable_mask
) {
2051 /* Returns true if this unit is fully realized. We check four things:
2053 * 1. Whether the cgroup was created at all
2054 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
2055 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
2056 * 4. Whether the invalidation mask is currently zero
2058 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
2059 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
2060 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
2061 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
2062 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
2063 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
2064 * simply don't matter. */
2066 return u
->cgroup_realized
&&
2067 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
2068 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
2069 u
->cgroup_invalidated_mask
== 0;
2072 static bool unit_has_mask_disables_realized(
2074 CGroupMask target_mask
,
2075 CGroupMask enable_mask
) {
2079 /* Returns true if all controllers which should be disabled are indeed disabled.
2081 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
2082 * already removed. */
2084 return !u
->cgroup_realized
||
2085 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
2086 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
2089 static bool unit_has_mask_enables_realized(
2091 CGroupMask target_mask
,
2092 CGroupMask enable_mask
) {
2096 /* Returns true if all controllers which should be enabled are indeed enabled.
2098 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
2099 * we want to add is already added. */
2101 return u
->cgroup_realized
&&
2102 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
2103 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
2106 static void unit_add_to_cgroup_realize_queue(Unit
*u
) {
2109 if (u
->in_cgroup_realize_queue
)
2112 LIST_APPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2113 u
->in_cgroup_realize_queue
= true;
2116 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
2119 if (!u
->in_cgroup_realize_queue
)
2122 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2123 u
->in_cgroup_realize_queue
= false;
2126 /* Controllers can only be enabled breadth-first, from the root of the
2127 * hierarchy downwards to the unit in question. */
2128 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
2129 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2134 /* First go deal with this unit's parent, or we won't be able to enable
2135 * any new controllers at this layer. */
2136 if (UNIT_ISSET(u
->slice
)) {
2137 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2142 target_mask
= unit_get_target_mask(u
);
2143 enable_mask
= unit_get_enable_mask(u
);
2145 /* We can only enable in this direction, don't try to disable anything.
2147 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
2150 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
2151 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
2153 return unit_update_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
2156 /* Controllers can only be disabled depth-first, from the leaves of the
2157 * hierarchy upwards to the unit in question. */
2158 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
2164 if (u
->type
!= UNIT_SLICE
)
2167 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
]) {
2168 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2171 if (UNIT_DEREF(m
->slice
) != u
)
2174 /* The cgroup for this unit might not actually be fully
2175 * realised yet, in which case it isn't holding any controllers
2177 if (!m
->cgroup_realized
)
2180 /* We must disable those below us first in order to release the
2182 if (m
->type
== UNIT_SLICE
)
2183 (void) unit_realize_cgroup_now_disable(m
, state
);
2185 target_mask
= unit_get_target_mask(m
);
2186 enable_mask
= unit_get_enable_mask(m
);
2188 /* We can only disable in this direction, don't try to enable
2190 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
2193 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
2194 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
2196 r
= unit_update_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
2204 /* Check if necessary controllers and attributes for a unit are in place.
2206 * - If so, do nothing.
2207 * - If not, create paths, move processes over, and set attributes.
2209 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
2210 * a depth-first way. As such the process looks like this:
2212 * Suppose we have a cgroup hierarchy which looks like this:
2225 * 1. We want to realise cgroup "d" now.
2226 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
2227 * 3. cgroup "k" just started requesting the memory controller.
2229 * To make this work we must do the following in order:
2231 * 1. Disable CPU controller in k, j
2232 * 2. Disable CPU controller in d
2233 * 3. Enable memory controller in root
2234 * 4. Enable memory controller in a
2235 * 5. Enable memory controller in d
2236 * 6. Enable memory controller in k
2238 * Notice that we need to touch j in one direction, but not the other. We also
2239 * don't go beyond d when disabling -- it's up to "a" to get realized if it
2240 * wants to disable further. The basic rules are therefore:
2242 * - If you're disabling something, you need to realise all of the cgroups from
2243 * your recursive descendants to the root. This starts from the leaves.
2244 * - If you're enabling something, you need to realise from the root cgroup
2245 * downwards, but you don't need to iterate your recursive descendants.
2247 * Returns 0 on success and < 0 on failure. */
2248 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
2249 CGroupMask target_mask
, enable_mask
;
2254 unit_remove_from_cgroup_realize_queue(u
);
2256 target_mask
= unit_get_target_mask(u
);
2257 enable_mask
= unit_get_enable_mask(u
);
2259 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
2262 /* Disable controllers below us, if there are any */
2263 r
= unit_realize_cgroup_now_disable(u
, state
);
2267 /* Enable controllers above us, if there are any */
2268 if (UNIT_ISSET(u
->slice
)) {
2269 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2274 /* Now actually deal with the cgroup we were trying to realise and set attributes */
2275 r
= unit_update_cgroup(u
, target_mask
, enable_mask
, state
);
2279 /* Now, reset the invalidation mask */
2280 u
->cgroup_invalidated_mask
= 0;
2284 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
2292 state
= manager_state(m
);
2294 while ((i
= m
->cgroup_realize_queue
)) {
2295 assert(i
->in_cgroup_realize_queue
);
2297 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
2298 /* Maybe things changed, and the unit is not actually active anymore? */
2299 unit_remove_from_cgroup_realize_queue(i
);
2303 r
= unit_realize_cgroup_now(i
, state
);
2305 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
2313 void unit_add_family_to_cgroup_realize_queue(Unit
*u
) {
2315 assert(u
->type
== UNIT_SLICE
);
2317 /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all
2320 * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes
2321 * very weird if two units that own processes reside in the same slice, but one is realized in the
2322 * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does
2323 * not), because that means individual processes need to be scheduled against whole cgroups. Let's
2324 * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1
2325 * controller hierarchies too (if unit requires the controller to be realized).
2327 * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date
2334 /* Children of u likely changed when we're called */
2335 u
->cgroup_members_mask_valid
= false;
2337 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
]) {
2338 /* Skip units that have a dependency on the slice but aren't actually in it. */
2339 if (UNIT_DEREF(m
->slice
) != u
)
2342 /* No point in doing cgroup application for units without active processes. */
2343 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
2346 /* We only enqueue siblings if they were realized once at least, in the main
2348 if (!m
->cgroup_realized
)
2351 /* If the unit doesn't need any new controllers and has current ones realized, it
2352 * doesn't need any changes. */
2353 if (unit_has_mask_realized(m
,
2354 unit_get_target_mask(m
),
2355 unit_get_enable_mask(m
)))
2358 unit_add_to_cgroup_realize_queue(m
);
2361 /* Parent comes after children */
2362 unit_add_to_cgroup_realize_queue(u
);
2363 } while ((u
= UNIT_DEREF(u
->slice
)));
2366 int unit_realize_cgroup(Unit
*u
) {
2369 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2372 /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all
2373 * parents, but there's more actually: for the weight-based controllers we also need to make sure
2374 * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the
2375 * other hand, when a controller is removed from realized set, it may become unnecessary in siblings
2376 * and ancestors and they should be (de)realized too.
2378 * This call will defer work on the siblings and derealized ancestors to the next event loop
2379 * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */
2381 if (UNIT_ISSET(u
->slice
))
2382 unit_add_family_to_cgroup_realize_queue(UNIT_DEREF(u
->slice
));
2384 /* And realize this one now (and apply the values) */
2385 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
2388 void unit_release_cgroup(Unit
*u
) {
2391 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
2392 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
2394 if (u
->cgroup_path
) {
2395 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
2396 u
->cgroup_path
= mfree(u
->cgroup_path
);
2399 if (u
->cgroup_control_inotify_wd
>= 0) {
2400 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
2401 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
);
2403 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
2404 u
->cgroup_control_inotify_wd
= -1;
2407 if (u
->cgroup_memory_inotify_wd
>= 0) {
2408 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
2409 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
);
2411 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
2412 u
->cgroup_memory_inotify_wd
= -1;
2416 void unit_prune_cgroup(Unit
*u
) {
2422 /* Removes the cgroup, if empty and possible, and stops watching it. */
2424 if (!u
->cgroup_path
)
2427 (void) unit_get_cpu_usage(u
, NULL
); /* Cache the last CPU usage value before we destroy the cgroup */
2429 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2431 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
2433 /* One reason we could have failed here is, that the cgroup still contains a process.
2434 * However, if the cgroup becomes removable at a later time, it might be removed when
2435 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
2436 * that the cgroup is still realized the next time it is started. Do not return early
2437 * on error, continue cleanup. */
2438 log_unit_full_errno(u
, r
== -EBUSY
? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
2443 unit_release_cgroup(u
);
2445 u
->cgroup_realized
= false;
2446 u
->cgroup_realized_mask
= 0;
2447 u
->cgroup_enabled_mask
= 0;
2449 u
->bpf_device_control_installed
= bpf_program_unref(u
->bpf_device_control_installed
);
2452 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
2453 _cleanup_fclose_
FILE *f
= NULL
;
2454 pid_t pid
= 0, npid
;
2460 if (!u
->cgroup_path
)
2463 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
2467 while (cg_read_pid(f
, &npid
) > 0) {
2472 if (pid_is_my_child(npid
) == 0)
2476 /* Dang, there's more than one daemonized PID
2477 in this group, so we don't know what process
2478 is the main process. */
2489 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
2490 _cleanup_closedir_
DIR *d
= NULL
;
2491 _cleanup_fclose_
FILE *f
= NULL
;
2497 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
2503 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
2504 r
= unit_watch_pid(u
, pid
, false);
2505 if (r
< 0 && ret
>= 0)
2509 if (r
< 0 && ret
>= 0)
2513 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
2520 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
2521 _cleanup_free_
char *p
= NULL
;
2523 p
= path_join(empty_to_root(path
), fn
);
2529 r
= unit_watch_pids_in_path(u
, p
);
2530 if (r
< 0 && ret
>= 0)
2534 if (r
< 0 && ret
>= 0)
2541 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
2546 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
2547 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
2548 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
2550 if (!u
->cgroup_path
)
2553 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2556 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
2559 if (!set_isempty(u
->pids
))
2562 unit_add_to_cgroup_empty_queue(u
);
2566 int unit_watch_all_pids(Unit
*u
) {
2571 /* Adds all PIDs from our cgroup to the set of PIDs we
2572 * watch. This is a fallback logic for cases where we do not
2573 * get reliable cgroup empty notifications: we try to use
2574 * SIGCHLD as replacement. */
2576 if (!u
->cgroup_path
)
2579 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2582 if (r
> 0) /* On unified we can use proper notifications */
2585 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
2588 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
2589 Manager
*m
= userdata
;
2596 u
= m
->cgroup_empty_queue
;
2600 assert(u
->in_cgroup_empty_queue
);
2601 u
->in_cgroup_empty_queue
= false;
2602 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
2604 if (m
->cgroup_empty_queue
) {
2605 /* More stuff queued, let's make sure we remain enabled */
2606 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2608 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
2611 unit_add_to_gc_queue(u
);
2613 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
2614 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
2619 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
2624 /* Note that there are four different ways how cgroup empty events reach us:
2626 * 1. On the unified hierarchy we get an inotify event on the cgroup
2628 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2630 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2632 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2633 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2635 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2636 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2637 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2638 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2639 * case for scope units). */
2641 if (u
->in_cgroup_empty_queue
)
2644 /* Let's verify that the cgroup is really empty */
2645 if (!u
->cgroup_path
)
2648 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2650 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", u
->cgroup_path
);
2656 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2657 u
->in_cgroup_empty_queue
= true;
2659 /* Trigger the defer event */
2660 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
2662 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
2665 static void unit_remove_from_cgroup_empty_queue(Unit
*u
) {
2668 if (!u
->in_cgroup_empty_queue
)
2671 LIST_REMOVE(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2672 u
->in_cgroup_empty_queue
= false;
2675 int unit_check_oom(Unit
*u
) {
2676 _cleanup_free_
char *oom_kill
= NULL
;
2681 if (!u
->cgroup_path
)
2684 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
2686 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
2688 r
= safe_atou64(oom_kill
, &c
);
2690 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
2692 increased
= c
> u
->oom_kill_last
;
2693 u
->oom_kill_last
= c
;
2698 log_struct(LOG_NOTICE
,
2699 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
2701 LOG_UNIT_INVOCATION_ID(u
),
2702 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
2704 if (UNIT_VTABLE(u
)->notify_cgroup_oom
)
2705 UNIT_VTABLE(u
)->notify_cgroup_oom(u
);
2710 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
2711 Manager
*m
= userdata
;
2718 u
= m
->cgroup_oom_queue
;
2722 assert(u
->in_cgroup_oom_queue
);
2723 u
->in_cgroup_oom_queue
= false;
2724 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
2726 if (m
->cgroup_oom_queue
) {
2727 /* More stuff queued, let's make sure we remain enabled */
2728 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2730 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
2733 (void) unit_check_oom(u
);
2737 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
2742 if (u
->in_cgroup_oom_queue
)
2744 if (!u
->cgroup_path
)
2747 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
2748 u
->in_cgroup_oom_queue
= true;
2750 /* Trigger the defer event */
2751 if (!u
->manager
->cgroup_oom_event_source
) {
2752 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
2754 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
2756 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
2760 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
2762 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
2766 (void) sd_event_source_set_description(s
, "cgroup-oom");
2767 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
2770 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
2772 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
2775 static int unit_check_cgroup_events(Unit
*u
) {
2776 char *values
[2] = {};
2781 r
= cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events",
2782 STRV_MAKE("populated", "frozen"), values
);
2786 /* The cgroup.events notifications can be merged together so act as we saw the given state for the
2787 * first time. The functions we call to handle given state are idempotent, which makes them
2788 * effectively remember the previous state. */
2790 if (streq(values
[0], "1"))
2791 unit_remove_from_cgroup_empty_queue(u
);
2793 unit_add_to_cgroup_empty_queue(u
);
2796 /* Disregard freezer state changes due to operations not initiated by us */
2797 if (values
[1] && IN_SET(u
->freezer_state
, FREEZER_FREEZING
, FREEZER_THAWING
)) {
2798 if (streq(values
[1], "0"))
2810 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
2811 Manager
*m
= userdata
;
2818 union inotify_event_buffer buffer
;
2819 struct inotify_event
*e
;
2822 l
= read(fd
, &buffer
, sizeof(buffer
));
2824 if (IN_SET(errno
, EINTR
, EAGAIN
))
2827 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
2830 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
2834 /* Queue overflow has no watch descriptor */
2837 if (e
->mask
& IN_IGNORED
)
2838 /* The watch was just removed */
2841 /* Note that inotify might deliver events for a watch even after it was removed,
2842 * because it was queued before the removal. Let's ignore this here safely. */
2844 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2846 unit_check_cgroup_events(u
);
2848 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2850 unit_add_to_cgroup_oom_queue(u
);
2855 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
2856 CGroupMask mask
= 0;
2859 /* BPF-based firewall */
2860 r
= bpf_firewall_supported();
2862 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2864 /* BPF-based device access control */
2865 r
= bpf_devices_supported();
2867 mask
|= CGROUP_MASK_BPF_DEVICES
;
2873 int manager_setup_cgroup(Manager
*m
) {
2874 _cleanup_free_
char *path
= NULL
;
2875 const char *scope_path
;
2883 /* 1. Determine hierarchy */
2884 m
->cgroup_root
= mfree(m
->cgroup_root
);
2885 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
2887 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
2889 /* Chop off the init scope, if we are already located in it */
2890 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2892 /* LEGACY: Also chop off the system slice if we are in
2893 * it. This is to support live upgrades from older systemd
2894 * versions where PID 1 was moved there. Also see
2895 * cg_get_root_path(). */
2896 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
2897 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
2899 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
2904 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
2905 * easily prepend it everywhere. */
2906 delete_trailing_chars(m
->cgroup_root
, "/");
2909 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
2911 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
2915 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
2917 all_unified
= cg_all_unified();
2918 if (all_unified
< 0)
2919 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
2920 if (all_unified
> 0)
2921 log_debug("Unified cgroup hierarchy is located at %s.", path
);
2923 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2925 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
2927 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
2929 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
2932 /* 3. Allocate cgroup empty defer event source */
2933 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
2934 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
2936 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
2938 /* Schedule cgroup empty checks early, but after having processed service notification messages or
2939 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
2940 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
2941 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
2943 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
2945 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
2947 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
2949 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
2951 /* 4. Install notifier inotify object, or agent */
2952 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
2954 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
2956 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
2957 safe_close(m
->cgroup_inotify_fd
);
2959 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
2960 if (m
->cgroup_inotify_fd
< 0)
2961 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
2963 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
2965 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
2967 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
2968 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
2969 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
2970 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
2972 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
2974 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
2976 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
2978 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
2979 * since it does not generate events when control groups with children run empty. */
2981 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUP_AGENT_PATH
);
2983 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
2985 log_debug("Installed release agent.");
2987 log_debug("Release agent already installed.");
2990 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
2991 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2992 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
2994 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
2995 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
2997 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
2999 /* 6. And pin it, so that it cannot be unmounted */
3000 safe_close(m
->pin_cgroupfs_fd
);
3001 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
3002 if (m
->pin_cgroupfs_fd
< 0)
3003 return log_error_errno(errno
, "Failed to open pin file: %m");
3005 } else if (!MANAGER_IS_TEST_RUN(m
))
3006 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
3008 /* 7. Always enable hierarchical support if it exists... */
3009 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
3010 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
3012 /* 8. Figure out which controllers are supported */
3013 r
= cg_mask_supported(&m
->cgroup_supported
);
3015 return log_error_errno(r
, "Failed to determine supported controllers: %m");
3017 /* 9. Figure out which bpf-based pseudo-controllers are supported */
3018 r
= cg_bpf_mask_supported(&mask
);
3020 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
3021 m
->cgroup_supported
|= mask
;
3023 /* 10. Log which controllers are supported */
3024 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
3025 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
3030 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
3033 /* We can't really delete the group, since we are in it. But
3035 if (delete && m
->cgroup_root
&& m
->test_run_flags
!= MANAGER_TEST_RUN_MINIMAL
)
3036 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
3038 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
3040 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
3041 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
3043 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
3044 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
3046 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
3048 m
->cgroup_root
= mfree(m
->cgroup_root
);
3051 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
3058 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
3062 p
= strdupa(cgroup
);
3066 e
= strrchr(p
, '/');
3068 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
3072 u
= hashmap_get(m
->cgroup_unit
, p
);
3078 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
3079 _cleanup_free_
char *cgroup
= NULL
;
3083 if (!pid_is_valid(pid
))
3086 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
3089 return manager_get_unit_by_cgroup(m
, cgroup
);
3092 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
3097 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
3098 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
3099 * relevant one as children of the process will be assigned to that one, too, before all else. */
3101 if (!pid_is_valid(pid
))
3104 if (pid
== getpid_cached())
3105 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
3107 u
= manager_get_unit_by_pid_cgroup(m
, pid
);
3111 u
= hashmap_get(m
->watch_pids
, PID_TO_PTR(pid
));
3115 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-pid
));
3122 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
3128 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
3129 * or from the --system instance */
3131 log_debug("Got cgroup empty notification for: %s", cgroup
);
3133 u
= manager_get_unit_by_cgroup(m
, cgroup
);
3137 unit_add_to_cgroup_empty_queue(u
);
3141 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
3147 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3150 if (!u
->cgroup_path
)
3153 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3154 if (unit_has_host_root_cgroup(u
))
3155 return procfs_memory_get_used(ret
);
3157 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3160 r
= cg_all_unified();
3164 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, r
> 0 ? "memory.current" : "memory.usage_in_bytes", ret
);
3167 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
3171 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
3174 if (!u
->cgroup_path
)
3177 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3178 if (unit_has_host_root_cgroup(u
))
3179 return procfs_tasks_get_current(ret
);
3181 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
3184 return cg_get_attribute_as_uint64("pids", u
->cgroup_path
, "pids.current", ret
);
3187 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
3194 if (!u
->cgroup_path
)
3197 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3198 if (unit_has_host_root_cgroup(u
))
3199 return procfs_cpu_get_usage(ret
);
3201 /* Requisite controllers for CPU accounting are not enabled */
3202 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
3205 r
= cg_all_unified();
3209 _cleanup_free_
char *val
= NULL
;
3212 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
3213 if (IN_SET(r
, -ENOENT
, -ENXIO
))
3218 r
= safe_atou64(val
, &us
);
3222 ns
= us
* NSEC_PER_USEC
;
3224 return cg_get_attribute_as_uint64("cpuacct", u
->cgroup_path
, "cpuacct.usage", ret
);
3230 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
3236 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
3237 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
3238 * call this function with a NULL return value. */
3240 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
3243 r
= unit_get_cpu_usage_raw(u
, &ns
);
3244 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
3245 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
3249 *ret
= u
->cpu_usage_last
;
3255 if (ns
> u
->cpu_usage_base
)
3256 ns
-= u
->cpu_usage_base
;
3260 u
->cpu_usage_last
= ns
;
3267 int unit_get_ip_accounting(
3269 CGroupIPAccountingMetric metric
,
3276 assert(metric
>= 0);
3277 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
3280 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
3283 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
3284 u
->ip_accounting_ingress_map_fd
:
3285 u
->ip_accounting_egress_map_fd
;
3289 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
3290 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
3292 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
3296 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
3297 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
3298 * ip_accounting_extra[] field, and add them in here transparently. */
3300 *ret
= value
+ u
->ip_accounting_extra
[metric
];
3305 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
3306 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
3307 [CGROUP_IO_READ_BYTES
] = "rbytes=",
3308 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
3309 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
3310 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
3312 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
3313 _cleanup_free_
char *path
= NULL
;
3314 _cleanup_fclose_
FILE *f
= NULL
;
3319 if (!u
->cgroup_path
)
3322 if (unit_has_host_root_cgroup(u
))
3323 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
3325 r
= cg_all_unified();
3328 if (r
== 0) /* TODO: support cgroupv1 */
3331 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
3334 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
3338 f
= fopen(path
, "re");
3343 _cleanup_free_
char *line
= NULL
;
3346 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3353 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
3354 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
3357 _cleanup_free_
char *word
= NULL
;
3359 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
3365 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3368 x
= startswith(word
, field_names
[i
]);
3372 r
= safe_atou64(x
, &w
);
3376 /* Sum up the stats of all devices */
3384 memcpy(ret
, acc
, sizeof(acc
));
3388 int unit_get_io_accounting(
3390 CGroupIOAccountingMetric metric
,
3394 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
3397 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
3399 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
3402 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3405 r
= unit_get_io_accounting_raw(u
, raw
);
3406 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3411 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3412 /* Saturated subtraction */
3413 if (raw
[i
] > u
->io_accounting_base
[i
])
3414 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
3416 u
->io_accounting_last
[i
] = 0;
3421 *ret
= u
->io_accounting_last
[metric
];
3426 int unit_reset_cpu_accounting(Unit
*u
) {
3431 u
->cpu_usage_last
= NSEC_INFINITY
;
3433 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
3435 u
->cpu_usage_base
= 0;
3442 int unit_reset_ip_accounting(Unit
*u
) {
3447 if (u
->ip_accounting_ingress_map_fd
>= 0)
3448 r
= bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
);
3450 if (u
->ip_accounting_egress_map_fd
>= 0)
3451 q
= bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
);
3453 zero(u
->ip_accounting_extra
);
3455 return r
< 0 ? r
: q
;
3458 int unit_reset_io_accounting(Unit
*u
) {
3463 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++)
3464 u
->io_accounting_last
[i
] = UINT64_MAX
;
3466 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
3468 zero(u
->io_accounting_base
);
3475 int unit_reset_accounting(Unit
*u
) {
3480 r
= unit_reset_cpu_accounting(u
);
3481 q
= unit_reset_io_accounting(u
);
3482 v
= unit_reset_ip_accounting(u
);
3484 return r
< 0 ? r
: q
< 0 ? q
: v
;
3487 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
3490 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3496 /* always invalidate compat pairs together */
3497 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
3498 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
3500 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
3501 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
3503 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
3506 u
->cgroup_invalidated_mask
|= m
;
3507 unit_add_to_cgroup_realize_queue(u
);
3510 void unit_invalidate_cgroup_bpf(Unit
*u
) {
3513 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3516 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
3519 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
3520 unit_add_to_cgroup_realize_queue(u
);
3522 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
3523 * list of our children includes our own. */
3524 if (u
->type
== UNIT_SLICE
) {
3528 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
])
3529 if (UNIT_DEREF(member
->slice
) == u
)
3530 unit_invalidate_cgroup_bpf(member
);
3534 bool unit_cgroup_delegate(Unit
*u
) {
3539 if (!UNIT_VTABLE(u
)->can_delegate
)
3542 c
= unit_get_cgroup_context(u
);
3549 void manager_invalidate_startup_units(Manager
*m
) {
3554 SET_FOREACH(u
, m
->startup_units
)
3555 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
3558 static int unit_get_nice(Unit
*u
) {
3561 ec
= unit_get_exec_context(u
);
3562 return ec
? ec
->nice
: 0;
3565 static uint64_t unit_get_cpu_weight(Unit
*u
) {
3566 ManagerState state
= manager_state(u
->manager
);
3569 cc
= unit_get_cgroup_context(u
);
3570 return cc
? cgroup_context_cpu_weight(cc
, state
) : CGROUP_WEIGHT_DEFAULT
;
3573 int compare_job_priority(const void *a
, const void *b
) {
3574 const Job
*x
= a
, *y
= b
;
3576 uint64_t weight_x
, weight_y
;
3579 if ((ret
= CMP(x
->unit
->type
, y
->unit
->type
)) != 0)
3582 weight_x
= unit_get_cpu_weight(x
->unit
);
3583 weight_y
= unit_get_cpu_weight(y
->unit
);
3585 if ((ret
= CMP(weight_x
, weight_y
)) != 0)
3588 nice_x
= unit_get_nice(x
->unit
);
3589 nice_y
= unit_get_nice(y
->unit
);
3591 if ((ret
= CMP(nice_x
, nice_y
)) != 0)
3594 return strcmp(x
->unit
->id
, y
->unit
->id
);
3597 int unit_cgroup_freezer_action(Unit
*u
, FreezerAction action
) {
3598 _cleanup_free_
char *path
= NULL
;
3599 FreezerState target
, kernel
= _FREEZER_STATE_INVALID
;
3603 assert(IN_SET(action
, FREEZER_FREEZE
, FREEZER_THAW
));
3605 if (!cg_freezer_supported())
3608 if (!u
->cgroup_realized
)
3611 target
= action
== FREEZER_FREEZE
? FREEZER_FROZEN
: FREEZER_RUNNING
;
3613 r
= unit_freezer_state_kernel(u
, &kernel
);
3615 log_unit_debug_errno(u
, r
, "Failed to obtain cgroup freezer state: %m");
3617 if (target
== kernel
) {
3618 u
->freezer_state
= target
;
3622 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.freeze", &path
);
3626 log_unit_debug(u
, "%s unit.", action
== FREEZER_FREEZE
? "Freezing" : "Thawing");
3628 if (action
== FREEZER_FREEZE
)
3629 u
->freezer_state
= FREEZER_FREEZING
;
3631 u
->freezer_state
= FREEZER_THAWING
;
3633 r
= write_string_file(path
, one_zero(action
== FREEZER_FREEZE
), WRITE_STRING_FILE_DISABLE_BUFFER
);
3640 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
3641 [CGROUP_DEVICE_POLICY_AUTO
] = "auto",
3642 [CGROUP_DEVICE_POLICY_CLOSED
] = "closed",
3643 [CGROUP_DEVICE_POLICY_STRICT
] = "strict",
3646 int unit_get_cpuset(Unit
*u
, CPUSet
*cpus
, const char *name
) {
3647 _cleanup_free_
char *v
= NULL
;
3653 if (!u
->cgroup_path
)
3656 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUSET
) == 0)
3659 r
= cg_all_unified();
3665 r
= cg_get_attribute("cpuset", u
->cgroup_path
, name
, &v
);
3671 return parse_cpu_set_full(v
, cpus
, false, NULL
, NULL
, 0, NULL
);
3674 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);
3676 static const char* const freezer_action_table
[_FREEZER_ACTION_MAX
] = {
3677 [FREEZER_FREEZE
] = "freeze",
3678 [FREEZER_THAW
] = "thaw",
3681 DEFINE_STRING_TABLE_LOOKUP(freezer_action
, FreezerAction
);