1 /* SPDX-License-Identifier: LGPL-2.1+ */
6 #include "sd-messages.h"
8 #include "alloc-util.h"
9 #include "blockdev-util.h"
10 #include "bpf-devices.h"
11 #include "bpf-firewall.h"
12 #include "btrfs-util.h"
13 #include "bus-error.h"
14 #include "cgroup-setup.h"
15 #include "cgroup-util.h"
20 #include "nulstr-util.h"
21 #include "parse-util.h"
22 #include "path-util.h"
23 #include "process-util.h"
24 #include "procfs-util.h"
26 #include "stat-util.h"
27 #include "stdio-util.h"
28 #include "string-table.h"
29 #include "string-util.h"
32 #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
34 /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access
35 * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask
36 * out specific attributes from us. */
37 #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING)
39 bool manager_owns_host_root_cgroup(Manager
*m
) {
42 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
43 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
44 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
45 * we run in any kind of container virtualization. */
47 if (MANAGER_IS_USER(m
))
50 if (detect_container() > 0)
53 return empty_or_root(m
->cgroup_root
);
56 bool unit_has_host_root_cgroup(Unit
*u
) {
59 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
60 * the manager manages the root cgroup. */
62 if (!manager_owns_host_root_cgroup(u
->manager
))
65 return unit_has_name(u
, SPECIAL_ROOT_SLICE
);
68 static int set_attribute_and_warn(Unit
*u
, const char *controller
, const char *attribute
, const char *value
) {
71 r
= cg_set_attribute(controller
, u
->cgroup_path
, attribute
, value
);
73 log_unit_full(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%.*s': %m",
74 strna(attribute
), isempty(u
->cgroup_path
) ? "/" : u
->cgroup_path
, (int) strcspn(value
, NEWLINE
), value
);
79 static void cgroup_compat_warn(void) {
80 static bool cgroup_compat_warned
= false;
82 if (cgroup_compat_warned
)
85 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
86 "See cgroup-compat debug messages for details.");
88 cgroup_compat_warned
= true;
91 #define log_cgroup_compat(unit, fmt, ...) do { \
92 cgroup_compat_warn(); \
93 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
96 void cgroup_context_init(CGroupContext
*c
) {
99 /* Initialize everything to the kernel defaults. */
101 *c
= (CGroupContext
) {
102 .cpu_weight
= CGROUP_WEIGHT_INVALID
,
103 .startup_cpu_weight
= CGROUP_WEIGHT_INVALID
,
104 .cpu_quota_per_sec_usec
= USEC_INFINITY
,
105 .cpu_quota_period_usec
= USEC_INFINITY
,
107 .cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
108 .startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
110 .memory_high
= CGROUP_LIMIT_MAX
,
111 .memory_max
= CGROUP_LIMIT_MAX
,
112 .memory_swap_max
= CGROUP_LIMIT_MAX
,
114 .memory_limit
= CGROUP_LIMIT_MAX
,
116 .io_weight
= CGROUP_WEIGHT_INVALID
,
117 .startup_io_weight
= CGROUP_WEIGHT_INVALID
,
119 .blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
120 .startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
122 .tasks_max
= CGROUP_LIMIT_MAX
,
126 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
130 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
135 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
139 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
144 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
148 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
153 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
157 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
162 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
166 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
171 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
175 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
180 void cgroup_context_done(CGroupContext
*c
) {
183 while (c
->io_device_weights
)
184 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
186 while (c
->io_device_latencies
)
187 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
189 while (c
->io_device_limits
)
190 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
192 while (c
->blockio_device_weights
)
193 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
195 while (c
->blockio_device_bandwidths
)
196 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
198 while (c
->device_allow
)
199 cgroup_context_free_device_allow(c
, c
->device_allow
);
201 c
->ip_address_allow
= ip_address_access_free_all(c
->ip_address_allow
);
202 c
->ip_address_deny
= ip_address_access_free_all(c
->ip_address_deny
);
204 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
205 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
207 cpu_set_reset(&c
->cpuset_cpus
);
208 cpu_set_reset(&c
->cpuset_mems
);
211 void cgroup_context_dump(CGroupContext
*c
, FILE* f
, const char *prefix
) {
212 _cleanup_free_
char *disable_controllers_str
= NULL
;
213 _cleanup_free_
char *cpuset_cpus
= NULL
;
214 _cleanup_free_
char *cpuset_mems
= NULL
;
215 CGroupIODeviceLimit
*il
;
216 CGroupIODeviceWeight
*iw
;
217 CGroupIODeviceLatency
*l
;
218 CGroupBlockIODeviceBandwidth
*b
;
219 CGroupBlockIODeviceWeight
*w
;
220 CGroupDeviceAllow
*a
;
221 IPAddressAccessItem
*iaai
;
223 char u
[FORMAT_TIMESPAN_MAX
];
224 char v
[FORMAT_TIMESPAN_MAX
];
229 prefix
= strempty(prefix
);
231 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
233 cpuset_cpus
= cpu_set_to_range_string(&c
->cpuset_cpus
);
234 cpuset_mems
= cpu_set_to_range_string(&c
->cpuset_mems
);
237 "%sCPUAccounting=%s\n"
238 "%sIOAccounting=%s\n"
239 "%sBlockIOAccounting=%s\n"
240 "%sMemoryAccounting=%s\n"
241 "%sTasksAccounting=%s\n"
242 "%sIPAccounting=%s\n"
243 "%sCPUWeight=%" PRIu64
"\n"
244 "%sStartupCPUWeight=%" PRIu64
"\n"
245 "%sCPUShares=%" PRIu64
"\n"
246 "%sStartupCPUShares=%" PRIu64
"\n"
247 "%sCPUQuotaPerSecSec=%s\n"
248 "%sCPUQuotaPeriodSec=%s\n"
250 "%sAllowedMemoryNodes=%s\n"
251 "%sIOWeight=%" PRIu64
"\n"
252 "%sStartupIOWeight=%" PRIu64
"\n"
253 "%sBlockIOWeight=%" PRIu64
"\n"
254 "%sStartupBlockIOWeight=%" PRIu64
"\n"
255 "%sDefaultMemoryMin=%" PRIu64
"\n"
256 "%sDefaultMemoryLow=%" PRIu64
"\n"
257 "%sMemoryMin=%" PRIu64
"\n"
258 "%sMemoryLow=%" PRIu64
"\n"
259 "%sMemoryHigh=%" PRIu64
"\n"
260 "%sMemoryMax=%" PRIu64
"\n"
261 "%sMemorySwapMax=%" PRIu64
"\n"
262 "%sMemoryLimit=%" PRIu64
"\n"
263 "%sTasksMax=%" PRIu64
"\n"
264 "%sDevicePolicy=%s\n"
265 "%sDisableControllers=%s\n"
267 prefix
, yes_no(c
->cpu_accounting
),
268 prefix
, yes_no(c
->io_accounting
),
269 prefix
, yes_no(c
->blockio_accounting
),
270 prefix
, yes_no(c
->memory_accounting
),
271 prefix
, yes_no(c
->tasks_accounting
),
272 prefix
, yes_no(c
->ip_accounting
),
273 prefix
, c
->cpu_weight
,
274 prefix
, c
->startup_cpu_weight
,
275 prefix
, c
->cpu_shares
,
276 prefix
, c
->startup_cpu_shares
,
277 prefix
, format_timespan(u
, sizeof(u
), c
->cpu_quota_per_sec_usec
, 1),
278 prefix
, format_timespan(v
, sizeof(v
), c
->cpu_quota_period_usec
, 1),
281 prefix
, c
->io_weight
,
282 prefix
, c
->startup_io_weight
,
283 prefix
, c
->blockio_weight
,
284 prefix
, c
->startup_blockio_weight
,
285 prefix
, c
->default_memory_min
,
286 prefix
, c
->default_memory_low
,
287 prefix
, c
->memory_min
,
288 prefix
, c
->memory_low
,
289 prefix
, c
->memory_high
,
290 prefix
, c
->memory_max
,
291 prefix
, c
->memory_swap_max
,
292 prefix
, c
->memory_limit
,
293 prefix
, c
->tasks_max
,
294 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
295 prefix
, strempty(disable_controllers_str
),
296 prefix
, yes_no(c
->delegate
));
299 _cleanup_free_
char *t
= NULL
;
301 (void) cg_mask_to_string(c
->delegate_controllers
, &t
);
303 fprintf(f
, "%sDelegateControllers=%s\n",
308 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
310 "%sDeviceAllow=%s %s%s%s\n",
313 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
315 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
317 "%sIODeviceWeight=%s %" PRIu64
"\n",
322 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
324 "%sIODeviceLatencyTargetSec=%s %s\n",
327 format_timespan(u
, sizeof(u
), l
->target_usec
, 1));
329 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
330 char buf
[FORMAT_BYTES_MAX
];
331 CGroupIOLimitType type
;
333 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
334 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
338 cgroup_io_limit_type_to_string(type
),
340 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
343 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
345 "%sBlockIODeviceWeight=%s %" PRIu64
,
350 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
351 char buf
[FORMAT_BYTES_MAX
];
353 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
355 "%sBlockIOReadBandwidth=%s %s\n",
358 format_bytes(buf
, sizeof(buf
), b
->rbps
));
359 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
361 "%sBlockIOWriteBandwidth=%s %s\n",
364 format_bytes(buf
, sizeof(buf
), b
->wbps
));
367 LIST_FOREACH(items
, iaai
, c
->ip_address_allow
) {
368 _cleanup_free_
char *k
= NULL
;
370 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
371 fprintf(f
, "%sIPAddressAllow=%s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
374 LIST_FOREACH(items
, iaai
, c
->ip_address_deny
) {
375 _cleanup_free_
char *k
= NULL
;
377 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
378 fprintf(f
, "%sIPAddressDeny=%s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
381 STRV_FOREACH(path
, c
->ip_filters_ingress
)
382 fprintf(f
, "%sIPIngressFilterPath=%s\n", prefix
, *path
);
384 STRV_FOREACH(path
, c
->ip_filters_egress
)
385 fprintf(f
, "%sIPEgressFilterPath=%s\n", prefix
, *path
);
388 int cgroup_add_device_allow(CGroupContext
*c
, const char *dev
, const char *mode
) {
389 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
390 _cleanup_free_
char *d
= NULL
;
394 assert(isempty(mode
) || in_charset(mode
, "rwm"));
396 a
= new(CGroupDeviceAllow
, 1);
404 *a
= (CGroupDeviceAllow
) {
406 .r
= isempty(mode
) || strchr(mode
, 'r'),
407 .w
= isempty(mode
) || strchr(mode
, 'w'),
408 .m
= isempty(mode
) || strchr(mode
, 'm'),
411 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
417 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
418 uint64_t unit_get_ancestor_##entry(Unit *u) { \
421 /* 1. Is entry set in this unit? If so, use that. \
422 * 2. Is the default for this entry set in any \
423 * ancestor? If so, use that. \
424 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
428 c = unit_get_cgroup_context(u); \
429 if (c && c->entry##_set) \
432 while ((u = UNIT_DEREF(u->slice))) { \
433 c = unit_get_cgroup_context(u); \
434 if (c && c->default_##entry##_set) \
435 return c->default_##entry; \
438 /* We've reached the root, but nobody had default for \
439 * this entry set, so set it to the kernel default. */ \
440 return CGROUP_LIMIT_MIN; \
443 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
444 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
446 static void cgroup_xattr_apply(Unit
*u
) {
447 char ids
[SD_ID128_STRING_MAX
];
452 if (!MANAGER_IS_SYSTEM(u
->manager
))
455 if (sd_id128_is_null(u
->invocation_id
))
458 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
459 "trusted.invocation_id",
460 sd_id128_to_string(u
->invocation_id
, ids
), 32,
463 log_unit_debug_errno(u
, r
, "Failed to set invocation ID on control group %s, ignoring: %m", u
->cgroup_path
);
466 static int lookup_block_device(const char *p
, dev_t
*ret
) {
474 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
475 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
477 if (stat(p
, &st
) < 0)
478 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
479 rdev
= (dev_t
)st
.st_rdev
;
480 dev
= (dev_t
)st
.st_dev
;
483 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
486 log_warning("Device node '%s' is a character device, but block device needed.", p
);
488 } else if (S_ISBLK(mode
))
490 else if (major(dev
) != 0)
491 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
493 /* If this is btrfs, getting the backing block device is a bit harder */
494 r
= btrfs_get_block_device(p
, ret
);
495 if (r
< 0 && r
!= -ENOTTY
)
496 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
498 log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
503 /* If this is a LUKS device, try to get the originating block device */
504 (void) block_get_originating(*ret
, ret
);
506 /* If this is a partition, try to get the originating block device */
507 (void) block_get_whole_disk(*ret
, ret
);
511 static int whitelist_device(BPFProgram
*prog
, const char *path
, const char *node
, const char *acc
) {
519 /* Some special handling for /dev/block/%u:%u, /dev/char/%u:%u, /run/systemd/inaccessible/chr and
520 * /run/systemd/inaccessible/blk paths. Instead of stat()ing these we parse out the major/minor directly. This
521 * means clients can use these path without the device node actually around */
522 r
= device_path_parse_major_minor(node
, &mode
, &rdev
);
525 return log_warning_errno(r
, "Couldn't parse major/minor from device path '%s': %m", node
);
528 if (stat(node
, &st
) < 0)
529 return log_warning_errno(errno
, "Couldn't stat device %s: %m", node
);
531 if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
)) {
532 log_warning("%s is not a device.", node
);
535 rdev
= (dev_t
) st
.st_rdev
;
539 if (cg_all_unified() > 0) {
543 return cgroup_bpf_whitelist_device(prog
, S_ISCHR(mode
) ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
544 major(rdev
), minor(rdev
), acc
);
547 char buf
[2+DECIMAL_STR_MAX(dev_t
)*2+2+4];
551 S_ISCHR(mode
) ? 'c' : 'b',
552 major(rdev
), minor(rdev
),
555 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */
557 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
559 return log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
560 r
, "Failed to set devices.allow on %s: %m", path
);
566 static int whitelist_major(BPFProgram
*prog
, const char *path
, const char *name
, char type
, const char *acc
) {
567 _cleanup_fclose_
FILE *f
= NULL
;
568 char buf
[2+DECIMAL_STR_MAX(unsigned)+3+4];
575 assert(IN_SET(type
, 'b', 'c'));
577 if (streq(name
, "*")) {
578 /* If the name is a wildcard, then apply this list to all devices of this type */
580 if (cg_all_unified() > 0) {
584 (void) cgroup_bpf_whitelist_class(prog
, type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
, acc
);
586 xsprintf(buf
, "%c *:* %s", type
, acc
);
588 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
590 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
591 "Failed to set devices.allow on %s: %m", path
);
596 if (safe_atou(name
, &maj
) >= 0 && DEVICE_MAJOR_VALID(maj
)) {
597 /* The name is numeric and suitable as major. In that case, let's take is major, and create the entry
600 if (cg_all_unified() > 0) {
604 (void) cgroup_bpf_whitelist_major(prog
,
605 type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
608 xsprintf(buf
, "%c %u:* %s", type
, maj
, acc
);
610 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
612 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
613 "Failed to set devices.allow on %s: %m", path
);
619 f
= fopen("/proc/devices", "re");
621 return log_warning_errno(errno
, "Cannot open /proc/devices to resolve %s (%c): %m", name
, type
);
624 _cleanup_free_
char *line
= NULL
;
627 r
= read_line(f
, LONG_LINE_MAX
, &line
);
629 return log_warning_errno(r
, "Failed to read /proc/devices: %m");
633 if (type
== 'c' && streq(line
, "Character devices:")) {
638 if (type
== 'b' && streq(line
, "Block devices:")) {
653 w
= strpbrk(p
, WHITESPACE
);
658 r
= safe_atou(p
, &maj
);
665 w
+= strspn(w
, WHITESPACE
);
667 if (fnmatch(name
, w
, 0) != 0)
670 if (cg_all_unified() > 0) {
674 (void) cgroup_bpf_whitelist_major(prog
,
675 type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
684 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL
687 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
689 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
690 r
, "Failed to set devices.allow on %s: %m", path
);
697 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
698 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
699 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
702 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
703 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
704 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
707 static uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
708 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
709 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
710 return c
->startup_cpu_weight
;
711 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
712 return c
->cpu_weight
;
714 return CGROUP_WEIGHT_DEFAULT
;
717 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
718 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
719 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
720 return c
->startup_cpu_shares
;
721 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
722 return c
->cpu_shares
;
724 return CGROUP_CPU_SHARES_DEFAULT
;
727 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
728 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
729 * need to be higher than that boundary. quota is specified in USecPerSec.
730 * Additionally, period must be at most max_period. */
733 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
736 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
739 if (quota
== USEC_INFINITY
)
740 /* Always use default period for infinity quota. */
741 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
743 if (period
== USEC_INFINITY
)
744 /* Default period was requested. */
745 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
747 /* Clamp to interval [1ms, 1s] */
748 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
750 if (new_period
!= period
) {
751 char v
[FORMAT_TIMESPAN_MAX
];
752 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
, 0,
753 "Clamping CPU interval for cpu.max: period is now %s",
754 format_timespan(v
, sizeof(v
), new_period
, 1));
755 u
->warned_clamping_cpu_quota_period
= true;
761 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
762 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
764 xsprintf(buf
, "%" PRIu64
"\n", weight
);
765 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
768 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
769 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
771 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
772 if (quota
!= USEC_INFINITY
)
773 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
774 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
776 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
777 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
780 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
781 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
783 xsprintf(buf
, "%" PRIu64
"\n", shares
);
784 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
787 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
788 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
790 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
792 xsprintf(buf
, USEC_FMT
"\n", period
);
793 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
795 if (quota
!= USEC_INFINITY
) {
796 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
797 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
799 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
802 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
803 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
804 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
807 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
808 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
809 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
812 static void cgroup_apply_unified_cpuset(Unit
*u
, CPUSet cpus
, const char *name
) {
813 _cleanup_free_
char *buf
= NULL
;
815 buf
= cpu_set_to_range_string(&cpus
);
819 (void) set_attribute_and_warn(u
, "cpuset", name
, buf
);
822 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
823 return c
->io_accounting
||
824 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
825 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
826 c
->io_device_weights
||
827 c
->io_device_latencies
||
831 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
832 return c
->blockio_accounting
||
833 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
834 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
835 c
->blockio_device_weights
||
836 c
->blockio_device_bandwidths
;
839 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
840 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
841 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
842 return c
->startup_io_weight
;
843 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
846 return CGROUP_WEIGHT_DEFAULT
;
849 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
850 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
851 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
852 return c
->startup_blockio_weight
;
853 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
854 return c
->blockio_weight
;
856 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
859 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
860 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
861 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
864 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
865 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
866 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
869 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
870 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
874 r
= lookup_block_device(dev_path
, &dev
);
878 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
879 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
882 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
883 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
887 r
= lookup_block_device(dev_path
, &dev
);
891 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
892 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
895 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
896 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
900 r
= lookup_block_device(dev_path
, &dev
);
904 if (target
!= USEC_INFINITY
)
905 xsprintf(buf
, "%u:%u target=%" PRIu64
"\n", major(dev
), minor(dev
), target
);
907 xsprintf(buf
, "%u:%u target=max\n", major(dev
), minor(dev
));
909 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
912 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
913 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
914 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
915 CGroupIOLimitType type
;
919 r
= lookup_block_device(dev_path
, &dev
);
923 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
924 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
925 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
927 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
929 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
930 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
931 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
932 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
935 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
936 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
940 r
= lookup_block_device(dev_path
, &dev
);
944 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
945 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
947 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
948 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
951 static bool unit_has_unified_memory_config(Unit
*u
) {
956 c
= unit_get_cgroup_context(u
);
959 return unit_get_ancestor_memory_min(u
) > 0 || unit_get_ancestor_memory_low(u
) > 0 ||
960 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
||
961 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
;
964 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
965 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
967 if (v
!= CGROUP_LIMIT_MAX
)
968 xsprintf(buf
, "%" PRIu64
"\n", v
);
970 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
973 static void cgroup_apply_firewall(Unit
*u
) {
976 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
978 if (bpf_firewall_compile(u
) < 0)
981 (void) bpf_firewall_load_custom(u
);
982 (void) bpf_firewall_install(u
);
985 static void cgroup_context_apply(
987 CGroupMask apply_mask
,
988 ManagerState state
) {
992 bool is_host_root
, is_local_root
;
997 /* Nothing to do? Exit early! */
1001 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
1002 * attributes should only be managed for cgroups further down the tree. */
1003 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1004 is_host_root
= unit_has_host_root_cgroup(u
);
1006 assert_se(c
= unit_get_cgroup_context(u
));
1007 assert_se(path
= u
->cgroup_path
);
1009 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
1012 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
1013 * then), and missing cgroups, i.e. EROFS and ENOENT. */
1015 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
1016 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
1017 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
1018 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
1019 * we couldn't even write to them if we wanted to). */
1020 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
1022 if (cg_all_unified() > 0) {
1025 if (cgroup_context_has_cpu_weight(c
))
1026 weight
= cgroup_context_cpu_weight(c
, state
);
1027 else if (cgroup_context_has_cpu_shares(c
)) {
1030 shares
= cgroup_context_cpu_shares(c
, state
);
1031 weight
= cgroup_cpu_shares_to_weight(shares
);
1033 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
1034 shares
, weight
, path
);
1036 weight
= CGROUP_WEIGHT_DEFAULT
;
1038 cgroup_apply_unified_cpu_weight(u
, weight
);
1039 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1044 if (cgroup_context_has_cpu_weight(c
)) {
1047 weight
= cgroup_context_cpu_weight(c
, state
);
1048 shares
= cgroup_cpu_weight_to_shares(weight
);
1050 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
1051 weight
, shares
, path
);
1052 } else if (cgroup_context_has_cpu_shares(c
))
1053 shares
= cgroup_context_cpu_shares(c
, state
);
1055 shares
= CGROUP_CPU_SHARES_DEFAULT
;
1057 cgroup_apply_legacy_cpu_shares(u
, shares
);
1058 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1062 if ((apply_mask
& CGROUP_MASK_CPUSET
) && !is_local_root
) {
1063 cgroup_apply_unified_cpuset(u
, c
->cpuset_cpus
, "cpuset.cpus");
1064 cgroup_apply_unified_cpuset(u
, c
->cpuset_mems
, "cpuset.mems");
1067 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
1068 * controller), and in case of containers we want to leave control of these attributes to the container manager
1069 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
1070 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
1071 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
1072 bool has_io
, has_blockio
;
1075 has_io
= cgroup_context_has_io_config(c
);
1076 has_blockio
= cgroup_context_has_blockio_config(c
);
1079 weight
= cgroup_context_io_weight(c
, state
);
1080 else if (has_blockio
) {
1081 uint64_t blkio_weight
;
1083 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
1084 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
1086 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
1087 blkio_weight
, weight
);
1089 weight
= CGROUP_WEIGHT_DEFAULT
;
1091 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1092 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1094 /* FIXME: drop this when distro kernels properly support BFQ through "io.weight"
1095 * See also: https://github.com/systemd/systemd/pull/13335 */
1096 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1097 (void) set_attribute_and_warn(u
, "io", "io.bfq.weight", buf
);
1100 CGroupIODeviceLatency
*latency
;
1101 CGroupIODeviceLimit
*limit
;
1102 CGroupIODeviceWeight
*w
;
1104 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
1105 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
1107 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
1108 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
1110 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
1111 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
1113 } else if (has_blockio
) {
1114 CGroupBlockIODeviceWeight
*w
;
1115 CGroupBlockIODeviceBandwidth
*b
;
1117 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
1118 weight
= cgroup_weight_blkio_to_io(w
->weight
);
1120 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
1121 w
->weight
, weight
, w
->path
);
1123 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
1126 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
1127 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
1128 CGroupIOLimitType type
;
1130 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1131 limits
[type
] = cgroup_io_limit_defaults
[type
];
1133 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
1134 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
1136 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
1137 b
->rbps
, b
->wbps
, b
->path
);
1139 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
1144 if (apply_mask
& CGROUP_MASK_BLKIO
) {
1145 bool has_io
, has_blockio
;
1147 has_io
= cgroup_context_has_io_config(c
);
1148 has_blockio
= cgroup_context_has_blockio_config(c
);
1150 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
1151 * left to our container manager, too. */
1152 if (!is_local_root
) {
1153 char buf
[DECIMAL_STR_MAX(uint64_t)+1];
1159 io_weight
= cgroup_context_io_weight(c
, state
);
1160 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
1162 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
1164 } else if (has_blockio
)
1165 weight
= cgroup_context_blkio_weight(c
, state
);
1167 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
1169 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1170 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1173 CGroupIODeviceWeight
*w
;
1175 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
1176 weight
= cgroup_weight_io_to_blkio(w
->weight
);
1178 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
1179 w
->weight
, weight
, w
->path
);
1181 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
1183 } else if (has_blockio
) {
1184 CGroupBlockIODeviceWeight
*w
;
1186 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
1187 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
1191 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
1192 * roots, as there we want the container manager to handle it */
1193 if (is_host_root
|| !is_local_root
) {
1195 CGroupIODeviceLimit
*l
;
1197 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
1198 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
1199 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
1201 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
1203 } else if (has_blockio
) {
1204 CGroupBlockIODeviceBandwidth
*b
;
1206 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
1207 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
1212 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
1213 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
1214 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
1215 * write to this if we wanted to.) */
1216 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
1218 if (cg_all_unified() > 0) {
1219 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
;
1221 if (unit_has_unified_memory_config(u
)) {
1222 max
= c
->memory_max
;
1223 swap_max
= c
->memory_swap_max
;
1225 max
= c
->memory_limit
;
1227 if (max
!= CGROUP_LIMIT_MAX
)
1228 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
1231 cgroup_apply_unified_memory_limit(u
, "memory.min", unit_get_ancestor_memory_min(u
));
1232 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
1233 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
1234 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
1235 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
1237 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
1240 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1243 if (unit_has_unified_memory_config(u
)) {
1244 val
= c
->memory_max
;
1245 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIi64
" as MemoryLimit=", val
);
1247 val
= c
->memory_limit
;
1249 if (val
== CGROUP_LIMIT_MAX
)
1250 strncpy(buf
, "-1\n", sizeof(buf
));
1252 xsprintf(buf
, "%" PRIu64
"\n", val
);
1254 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
1258 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
1259 * containers, where we leave this to the manager */
1260 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
1261 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
)) {
1262 _cleanup_(bpf_program_unrefp
) BPFProgram
*prog
= NULL
;
1263 CGroupDeviceAllow
*a
;
1265 if (cg_all_unified() > 0) {
1266 r
= cgroup_init_device_bpf(&prog
, c
->device_policy
, c
->device_allow
);
1268 log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
1270 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL
1273 if (c
->device_allow
|| c
->device_policy
!= CGROUP_AUTO
)
1274 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
1276 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
1278 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1279 "Failed to reset devices.allow/devices.deny: %m");
1282 if (c
->device_policy
== CGROUP_CLOSED
||
1283 (c
->device_policy
== CGROUP_AUTO
&& c
->device_allow
)) {
1284 static const char auto_devices
[] =
1285 "/dev/null\0" "rwm\0"
1286 "/dev/zero\0" "rwm\0"
1287 "/dev/full\0" "rwm\0"
1288 "/dev/random\0" "rwm\0"
1289 "/dev/urandom\0" "rwm\0"
1290 "/dev/tty\0" "rwm\0"
1291 "/dev/ptmx\0" "rwm\0"
1292 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
1293 "/run/systemd/inaccessible/chr\0" "rwm\0"
1294 "/run/systemd/inaccessible/blk\0" "rwm\0";
1298 NULSTR_FOREACH_PAIR(x
, y
, auto_devices
)
1299 (void) whitelist_device(prog
, path
, x
, y
);
1301 /* PTS (/dev/pts) devices may not be duplicated, but accessed */
1302 (void) whitelist_major(prog
, path
, "pts", 'c', "rw");
1305 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1321 if (path_startswith(a
->path
, "/dev/"))
1322 (void) whitelist_device(prog
, path
, a
->path
, acc
);
1323 else if ((val
= startswith(a
->path
, "block-")))
1324 (void) whitelist_major(prog
, path
, val
, 'b', acc
);
1325 else if ((val
= startswith(a
->path
, "char-")))
1326 (void) whitelist_major(prog
, path
, val
, 'c', acc
);
1328 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1331 r
= cgroup_apply_device_bpf(u
, prog
, c
->device_policy
, c
->device_allow
);
1333 static bool warned
= false;
1335 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1336 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1337 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1338 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1344 if (apply_mask
& CGROUP_MASK_PIDS
) {
1347 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1348 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1349 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1350 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1351 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1352 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1353 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1354 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1355 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1356 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1357 * which is desirable so that there's an official way to release control of the sysctl from
1358 * systemd: set the limit to unbounded and reload. */
1360 if (c
->tasks_max
!= CGROUP_LIMIT_MAX
) {
1361 u
->manager
->sysctl_pid_max_changed
= true;
1362 r
= procfs_tasks_set_limit(c
->tasks_max
);
1363 } else if (u
->manager
->sysctl_pid_max_changed
)
1364 r
= procfs_tasks_set_limit(TASKS_MAX
);
1368 log_unit_full(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
1369 "Failed to write to tasks limit sysctls: %m");
1372 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
1373 * leave it for the container manager. */
1374 if (!is_local_root
) {
1375 if (c
->tasks_max
!= CGROUP_LIMIT_MAX
) {
1376 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
1378 sprintf(buf
, "%" PRIu64
"\n", c
->tasks_max
);
1379 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
1381 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
1385 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
1386 cgroup_apply_firewall(u
);
1389 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
1394 c
= unit_get_cgroup_context(u
);
1398 if (c
->ip_accounting
||
1399 c
->ip_address_allow
||
1400 c
->ip_address_deny
||
1401 c
->ip_filters_ingress
||
1402 c
->ip_filters_egress
)
1405 /* If any parent slice has an IP access list defined, it applies too */
1406 for (p
= UNIT_DEREF(u
->slice
); p
; p
= UNIT_DEREF(p
->slice
)) {
1407 c
= unit_get_cgroup_context(p
);
1411 if (c
->ip_address_allow
||
1419 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
1420 CGroupMask mask
= 0;
1425 c
= unit_get_cgroup_context(u
);
1429 /* Figure out which controllers we need, based on the cgroup context object */
1431 if (c
->cpu_accounting
)
1432 mask
|= get_cpu_accounting_mask();
1434 if (cgroup_context_has_cpu_weight(c
) ||
1435 cgroup_context_has_cpu_shares(c
) ||
1436 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
1437 mask
|= CGROUP_MASK_CPU
;
1439 if (c
->cpuset_cpus
.set
|| c
->cpuset_mems
.set
)
1440 mask
|= CGROUP_MASK_CPUSET
;
1442 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
1443 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1445 if (c
->memory_accounting
||
1446 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
1447 unit_has_unified_memory_config(u
))
1448 mask
|= CGROUP_MASK_MEMORY
;
1450 if (c
->device_allow
||
1451 c
->device_policy
!= CGROUP_AUTO
)
1452 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
1454 if (c
->tasks_accounting
||
1455 c
->tasks_max
!= CGROUP_LIMIT_MAX
)
1456 mask
|= CGROUP_MASK_PIDS
;
1458 return CGROUP_MASK_EXTEND_JOINED(mask
);
1461 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
1462 CGroupMask mask
= 0;
1464 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
1467 if (unit_get_needs_bpf_firewall(u
))
1468 mask
|= CGROUP_MASK_BPF_FIREWALL
;
1473 CGroupMask
unit_get_own_mask(Unit
*u
) {
1476 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
1477 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
1479 if (u
->load_state
!= UNIT_LOADED
)
1482 c
= unit_get_cgroup_context(u
);
1486 return (unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
)) & ~unit_get_ancestor_disable_mask(u
);
1489 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
1492 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1493 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1495 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1497 if (!unit_cgroup_delegate(u
))
1500 if (cg_all_unified() <= 0) {
1503 e
= unit_get_exec_context(u
);
1504 if (e
&& !exec_context_maintains_privileges(e
))
1508 assert_se(c
= unit_get_cgroup_context(u
));
1509 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
1512 CGroupMask
unit_get_members_mask(Unit
*u
) {
1515 /* Returns the mask of controllers all of the unit's children require, merged */
1517 if (u
->cgroup_members_mask_valid
)
1518 return u
->cgroup_members_mask
; /* Use cached value if possible */
1520 u
->cgroup_members_mask
= 0;
1522 if (u
->type
== UNIT_SLICE
) {
1527 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
], i
) {
1528 if (UNIT_DEREF(member
->slice
) == u
)
1529 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
1533 u
->cgroup_members_mask_valid
= true;
1534 return u
->cgroup_members_mask
;
1537 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
1540 /* Returns the mask of controllers all of the unit's siblings
1541 * require, i.e. the members mask of the unit's parent slice
1542 * if there is one. */
1544 if (UNIT_ISSET(u
->slice
))
1545 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
1547 return unit_get_subtree_mask(u
); /* we are the top-level slice */
1550 CGroupMask
unit_get_disable_mask(Unit
*u
) {
1553 c
= unit_get_cgroup_context(u
);
1557 return c
->disable_controllers
;
1560 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
1564 mask
= unit_get_disable_mask(u
);
1566 /* Returns the mask of controllers which are marked as forcibly
1567 * disabled in any ancestor unit or the unit in question. */
1569 if (UNIT_ISSET(u
->slice
))
1570 mask
|= unit_get_ancestor_disable_mask(UNIT_DEREF(u
->slice
));
1575 CGroupMask
unit_get_subtree_mask(Unit
*u
) {
1577 /* Returns the mask of this subtree, meaning of the group
1578 * itself and its children. */
1580 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
1583 CGroupMask
unit_get_target_mask(Unit
*u
) {
1586 /* This returns the cgroup mask of all controllers to enable
1587 * for a specific cgroup, i.e. everything it needs itself,
1588 * plus all that its children need, plus all that its siblings
1589 * need. This is primarily useful on the legacy cgroup
1590 * hierarchy, where we need to duplicate each cgroup in each
1591 * hierarchy that shall be enabled for it. */
1593 mask
= unit_get_own_mask(u
) | unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
1595 if (mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
1596 emit_bpf_firewall_warning(u
);
1598 mask
&= u
->manager
->cgroup_supported
;
1599 mask
&= ~unit_get_ancestor_disable_mask(u
);
1604 CGroupMask
unit_get_enable_mask(Unit
*u
) {
1607 /* This returns the cgroup mask of all controllers to enable
1608 * for the children of a specific cgroup. This is primarily
1609 * useful for the unified cgroup hierarchy, where each cgroup
1610 * controls which controllers are enabled for its children. */
1612 mask
= unit_get_members_mask(u
);
1613 mask
&= u
->manager
->cgroup_supported
;
1614 mask
&= ~unit_get_ancestor_disable_mask(u
);
1619 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
1622 /* Recurse invalidate the member masks cache all the way up the tree */
1623 u
->cgroup_members_mask_valid
= false;
1625 if (UNIT_ISSET(u
->slice
))
1626 unit_invalidate_cgroup_members_masks(UNIT_DEREF(u
->slice
));
1629 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
1631 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1635 if (u
->cgroup_path
&&
1636 u
->cgroup_realized
&&
1637 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
1638 return u
->cgroup_path
;
1640 u
= UNIT_DEREF(u
->slice
);
1646 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1647 return unit_get_realized_cgroup_path(userdata
, mask
);
1650 char *unit_default_cgroup_path(const Unit
*u
) {
1651 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1656 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1657 return strdup(u
->manager
->cgroup_root
);
1659 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1660 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1665 escaped
= cg_escape(u
->id
);
1669 return path_join(empty_to_root(u
->manager
->cgroup_root
), slice
, escaped
);
1672 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1673 _cleanup_free_
char *p
= NULL
;
1678 if (streq_ptr(u
->cgroup_path
, path
))
1688 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1693 unit_release_cgroup(u
);
1694 u
->cgroup_path
= TAKE_PTR(p
);
1699 int unit_watch_cgroup(Unit
*u
) {
1700 _cleanup_free_
char *events
= NULL
;
1705 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
1706 * cgroupv2 is available. */
1708 if (!u
->cgroup_path
)
1711 if (u
->cgroup_control_inotify_wd
>= 0)
1714 /* Only applies to the unified hierarchy */
1715 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1717 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1721 /* No point in watch the top-level slice, it's never going to run empty. */
1722 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1725 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
1729 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1733 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1734 if (u
->cgroup_control_inotify_wd
< 0) {
1736 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1737 * is not an error */
1740 return log_unit_error_errno(u
, errno
, "Failed to add control inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1743 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
1745 return log_unit_error_errno(u
, r
, "Failed to add control inotify watch descriptor to hash map: %m");
1750 int unit_watch_cgroup_memory(Unit
*u
) {
1751 _cleanup_free_
char *events
= NULL
;
1757 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
1758 * cgroupv2 is available. */
1760 if (!u
->cgroup_path
)
1763 c
= unit_get_cgroup_context(u
);
1767 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
1768 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
1770 if (!c
->memory_accounting
)
1773 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
1774 * we also don't want to generate a log message for each parent cgroup of a process. */
1775 if (u
->type
== UNIT_SLICE
)
1778 if (u
->cgroup_memory_inotify_wd
>= 0)
1781 /* Only applies to the unified hierarchy */
1782 r
= cg_all_unified();
1784 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
1788 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
1792 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
1796 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1797 if (u
->cgroup_memory_inotify_wd
< 0) {
1799 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1800 * is not an error */
1803 return log_unit_error_errno(u
, errno
, "Failed to add memory inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1806 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
1808 return log_unit_error_errno(u
, r
, "Failed to add memory inotify watch descriptor to hash map: %m");
1813 int unit_pick_cgroup_path(Unit
*u
) {
1814 _cleanup_free_
char *path
= NULL
;
1822 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1825 path
= unit_default_cgroup_path(u
);
1829 r
= unit_set_cgroup_path(u
, path
);
1831 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1833 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1838 static int unit_create_cgroup(
1840 CGroupMask target_mask
,
1841 CGroupMask enable_mask
,
1842 ManagerState state
) {
1849 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1852 /* Figure out our cgroup path */
1853 r
= unit_pick_cgroup_path(u
);
1857 /* First, create our own group */
1858 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
1860 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
1863 /* Start watching it */
1864 (void) unit_watch_cgroup(u
);
1865 (void) unit_watch_cgroup_memory(u
);
1867 /* Preserve enabled controllers in delegated units, adjust others. */
1868 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
1869 CGroupMask result_mask
= 0;
1871 /* Enable all controllers we need */
1872 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
1874 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
1876 /* If we just turned off a controller, this might release the controller for our parent too, let's
1877 * enqueue the parent for re-realization in that case again. */
1878 if (UNIT_ISSET(u
->slice
)) {
1879 CGroupMask turned_off
;
1881 turned_off
= (u
->cgroup_realized
? u
->cgroup_enabled_mask
& ~result_mask
: 0);
1882 if (turned_off
!= 0) {
1885 /* Force the parent to propagate the enable mask to the kernel again, by invalidating
1886 * the controller we just turned off. */
1888 for (parent
= UNIT_DEREF(u
->slice
); parent
; parent
= UNIT_DEREF(parent
->slice
))
1889 unit_invalidate_cgroup(parent
, turned_off
);
1893 /* Remember what's actually enabled now */
1894 u
->cgroup_enabled_mask
= result_mask
;
1897 /* Keep track that this is now realized */
1898 u
->cgroup_realized
= true;
1899 u
->cgroup_realized_mask
= target_mask
;
1901 if (u
->type
!= UNIT_SLICE
&& !unit_cgroup_delegate(u
)) {
1903 /* Then, possibly move things over, but not if
1904 * subgroups may contain processes, which is the case
1905 * for slice and delegation units. */
1906 r
= cg_migrate_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, u
->cgroup_path
, migrate_callback
, u
);
1908 log_unit_warning_errno(u
, r
, "Failed to migrate cgroup from to %s, ignoring: %m", u
->cgroup_path
);
1911 /* Set attributes */
1912 cgroup_context_apply(u
, target_mask
, state
);
1913 cgroup_xattr_apply(u
);
1918 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
1919 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1925 if (MANAGER_IS_SYSTEM(u
->manager
))
1928 if (!u
->manager
->system_bus
)
1931 if (!u
->cgroup_path
)
1934 /* Determine this unit's cgroup path relative to our cgroup root */
1935 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
1939 pp
= strjoina("/", pp
, suffix_path
);
1940 path_simplify(pp
, false);
1942 r
= sd_bus_call_method(u
->manager
->system_bus
,
1943 "org.freedesktop.systemd1",
1944 "/org/freedesktop/systemd1",
1945 "org.freedesktop.systemd1.Manager",
1946 "AttachProcessesToUnit",
1949 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
1951 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
1956 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
1957 CGroupMask delegated_mask
;
1965 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1968 if (set_isempty(pids
))
1971 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
1972 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
1973 r
= bpf_firewall_load_custom(u
);
1977 r
= unit_realize_cgroup(u
);
1981 if (isempty(suffix_path
))
1984 p
= prefix_roota(u
->cgroup_path
, suffix_path
);
1986 delegated_mask
= unit_get_delegate_mask(u
);
1989 SET_FOREACH(pidp
, pids
, i
) {
1990 pid_t pid
= PTR_TO_PID(pidp
);
1993 /* First, attach the PID to the main cgroup hierarchy */
1994 q
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
);
1996 log_unit_debug_errno(u
, q
, "Couldn't move process " PID_FMT
" to requested cgroup '%s': %m", pid
, p
);
1998 if (MANAGER_IS_USER(u
->manager
) && IN_SET(q
, -EPERM
, -EACCES
)) {
2001 /* If we are in a user instance, and we can't move the process ourselves due to
2002 * permission problems, let's ask the system instance about it instead. Since it's more
2003 * privileged it might be able to move the process across the leaves of a subtree who's
2004 * top node is not owned by us. */
2006 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
, suffix_path
);
2008 log_unit_debug_errno(u
, z
, "Couldn't move process " PID_FMT
" to requested cgroup '%s' via the system bus either: %m", pid
, p
);
2010 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
2014 r
= q
; /* Remember first error */
2019 q
= cg_all_unified();
2025 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
2026 * innermost realized one */
2028 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2029 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2030 const char *realized
;
2032 if (!(u
->manager
->cgroup_supported
& bit
))
2035 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
2036 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
2037 q
= cg_attach(cgroup_controller_to_string(c
), p
, pid
);
2039 continue; /* Success! */
2041 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",
2042 pid
, p
, cgroup_controller_to_string(c
));
2045 /* So this controller is either not delegate or realized, or something else weird happened. In
2046 * that case let's attach the PID at least to the closest cgroup up the tree that is
2048 realized
= unit_get_realized_cgroup_path(u
, bit
);
2050 continue; /* Not even realized in the root slice? Then let's not bother */
2052 q
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
);
2054 log_unit_debug_errno(u
, q
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
2055 pid
, realized
, cgroup_controller_to_string(c
));
2062 static bool unit_has_mask_realized(
2064 CGroupMask target_mask
,
2065 CGroupMask enable_mask
) {
2069 /* Returns true if this unit is fully realized. We check four things:
2071 * 1. Whether the cgroup was created at all
2072 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
2073 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
2074 * 4. Whether the invalidation mask is currently zero
2076 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
2077 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
2078 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
2079 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
2080 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
2081 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
2082 * simply don't matter. */
2084 return u
->cgroup_realized
&&
2085 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
2086 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
2087 u
->cgroup_invalidated_mask
== 0;
2090 static bool unit_has_mask_disables_realized(
2092 CGroupMask target_mask
,
2093 CGroupMask enable_mask
) {
2097 /* Returns true if all controllers which should be disabled are indeed disabled.
2099 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
2100 * already removed. */
2102 return !u
->cgroup_realized
||
2103 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
2104 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
2107 static bool unit_has_mask_enables_realized(
2109 CGroupMask target_mask
,
2110 CGroupMask enable_mask
) {
2114 /* Returns true if all controllers which should be enabled are indeed enabled.
2116 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
2117 * we want to add is already added. */
2119 return u
->cgroup_realized
&&
2120 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
2121 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
2124 void unit_add_to_cgroup_realize_queue(Unit
*u
) {
2127 if (u
->in_cgroup_realize_queue
)
2130 LIST_PREPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2131 u
->in_cgroup_realize_queue
= true;
2134 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
2137 if (!u
->in_cgroup_realize_queue
)
2140 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2141 u
->in_cgroup_realize_queue
= false;
2144 /* Controllers can only be enabled breadth-first, from the root of the
2145 * hierarchy downwards to the unit in question. */
2146 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
2147 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2152 /* First go deal with this unit's parent, or we won't be able to enable
2153 * any new controllers at this layer. */
2154 if (UNIT_ISSET(u
->slice
)) {
2155 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2160 target_mask
= unit_get_target_mask(u
);
2161 enable_mask
= unit_get_enable_mask(u
);
2163 /* We can only enable in this direction, don't try to disable anything.
2165 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
2168 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
2169 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
2171 return unit_create_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
2174 /* Controllers can only be disabled depth-first, from the leaves of the
2175 * hierarchy upwards to the unit in question. */
2176 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
2183 if (u
->type
!= UNIT_SLICE
)
2186 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
], i
) {
2187 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2190 if (UNIT_DEREF(m
->slice
) != u
)
2193 /* The cgroup for this unit might not actually be fully
2194 * realised yet, in which case it isn't holding any controllers
2196 if (!m
->cgroup_path
)
2199 /* We must disable those below us first in order to release the
2201 if (m
->type
== UNIT_SLICE
)
2202 (void) unit_realize_cgroup_now_disable(m
, state
);
2204 target_mask
= unit_get_target_mask(m
);
2205 enable_mask
= unit_get_enable_mask(m
);
2207 /* We can only disable in this direction, don't try to enable
2209 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
2212 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
2213 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
2215 r
= unit_create_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
2223 /* Check if necessary controllers and attributes for a unit are in place.
2225 * - If so, do nothing.
2226 * - If not, create paths, move processes over, and set attributes.
2228 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
2229 * a depth-first way. As such the process looks like this:
2231 * Suppose we have a cgroup hierarchy which looks like this:
2244 * 1. We want to realise cgroup "d" now.
2245 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
2246 * 3. cgroup "k" just started requesting the memory controller.
2248 * To make this work we must do the following in order:
2250 * 1. Disable CPU controller in k, j
2251 * 2. Disable CPU controller in d
2252 * 3. Enable memory controller in root
2253 * 4. Enable memory controller in a
2254 * 5. Enable memory controller in d
2255 * 6. Enable memory controller in k
2257 * Notice that we need to touch j in one direction, but not the other. We also
2258 * don't go beyond d when disabling -- it's up to "a" to get realized if it
2259 * wants to disable further. The basic rules are therefore:
2261 * - If you're disabling something, you need to realise all of the cgroups from
2262 * your recursive descendants to the root. This starts from the leaves.
2263 * - If you're enabling something, you need to realise from the root cgroup
2264 * downwards, but you don't need to iterate your recursive descendants.
2266 * Returns 0 on success and < 0 on failure. */
2267 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
2268 CGroupMask target_mask
, enable_mask
;
2273 unit_remove_from_cgroup_realize_queue(u
);
2275 target_mask
= unit_get_target_mask(u
);
2276 enable_mask
= unit_get_enable_mask(u
);
2278 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
2281 /* Disable controllers below us, if there are any */
2282 r
= unit_realize_cgroup_now_disable(u
, state
);
2286 /* Enable controllers above us, if there are any */
2287 if (UNIT_ISSET(u
->slice
)) {
2288 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2293 /* Now actually deal with the cgroup we were trying to realise and set attributes */
2294 r
= unit_create_cgroup(u
, target_mask
, enable_mask
, state
);
2298 /* Now, reset the invalidation mask */
2299 u
->cgroup_invalidated_mask
= 0;
2303 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
2311 state
= manager_state(m
);
2313 while ((i
= m
->cgroup_realize_queue
)) {
2314 assert(i
->in_cgroup_realize_queue
);
2316 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
2317 /* Maybe things changed, and the unit is not actually active anymore? */
2318 unit_remove_from_cgroup_realize_queue(i
);
2322 r
= unit_realize_cgroup_now(i
, state
);
2324 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
2332 static void unit_add_siblings_to_cgroup_realize_queue(Unit
*u
) {
2335 /* This adds the siblings of the specified unit and the
2336 * siblings of all parent units to the cgroup queue. (But
2337 * neither the specified unit itself nor the parents.) */
2339 while ((slice
= UNIT_DEREF(u
->slice
))) {
2344 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
], i
) {
2345 /* Skip units that have a dependency on the slice
2346 * but aren't actually in it. */
2347 if (UNIT_DEREF(m
->slice
) != slice
)
2350 /* No point in doing cgroup application for units
2351 * without active processes. */
2352 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
2355 /* If the unit doesn't need any new controllers
2356 * and has current ones realized, it doesn't need
2358 if (unit_has_mask_realized(m
,
2359 unit_get_target_mask(m
),
2360 unit_get_enable_mask(m
)))
2363 unit_add_to_cgroup_realize_queue(m
);
2370 int unit_realize_cgroup(Unit
*u
) {
2373 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2376 /* So, here's the deal: when realizing the cgroups for this
2377 * unit, we need to first create all parents, but there's more
2378 * actually: for the weight-based controllers we also need to
2379 * make sure that all our siblings (i.e. units that are in the
2380 * same slice as we are) have cgroups, too. Otherwise, things
2381 * would become very uneven as each of their processes would
2382 * get as much resources as all our group together. This call
2383 * will synchronously create the parent cgroups, but will
2384 * defer work on the siblings to the next event loop
2387 /* Add all sibling slices to the cgroup queue. */
2388 unit_add_siblings_to_cgroup_realize_queue(u
);
2390 /* And realize this one now (and apply the values) */
2391 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
2394 void unit_release_cgroup(Unit
*u
) {
2397 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
2398 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
2400 if (u
->cgroup_path
) {
2401 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
2402 u
->cgroup_path
= mfree(u
->cgroup_path
);
2405 if (u
->cgroup_control_inotify_wd
>= 0) {
2406 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
2407 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
);
2409 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
2410 u
->cgroup_control_inotify_wd
= -1;
2413 if (u
->cgroup_memory_inotify_wd
>= 0) {
2414 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
2415 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
);
2417 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
2418 u
->cgroup_memory_inotify_wd
= -1;
2422 void unit_prune_cgroup(Unit
*u
) {
2428 /* Removes the cgroup, if empty and possible, and stops watching it. */
2430 if (!u
->cgroup_path
)
2433 (void) unit_get_cpu_usage(u
, NULL
); /* Cache the last CPU usage value before we destroy the cgroup */
2435 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2437 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
2439 /* One reason we could have failed here is, that the cgroup still contains a process.
2440 * However, if the cgroup becomes removable at a later time, it might be removed when
2441 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
2442 * that the cgroup is still realized the next time it is started. Do not return early
2443 * on error, continue cleanup. */
2444 log_unit_full(u
, r
== -EBUSY
? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
2449 unit_release_cgroup(u
);
2451 u
->cgroup_realized
= false;
2452 u
->cgroup_realized_mask
= 0;
2453 u
->cgroup_enabled_mask
= 0;
2455 u
->bpf_device_control_installed
= bpf_program_unref(u
->bpf_device_control_installed
);
2458 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
2459 _cleanup_fclose_
FILE *f
= NULL
;
2460 pid_t pid
= 0, npid
;
2466 if (!u
->cgroup_path
)
2469 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
2473 while (cg_read_pid(f
, &npid
) > 0) {
2478 if (pid_is_my_child(npid
) == 0)
2482 /* Dang, there's more than one daemonized PID
2483 in this group, so we don't know what process
2484 is the main process. */
2495 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
2496 _cleanup_closedir_
DIR *d
= NULL
;
2497 _cleanup_fclose_
FILE *f
= NULL
;
2503 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
2509 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
2510 r
= unit_watch_pid(u
, pid
, false);
2511 if (r
< 0 && ret
>= 0)
2515 if (r
< 0 && ret
>= 0)
2519 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
2526 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
2527 _cleanup_free_
char *p
= NULL
;
2529 p
= path_join(empty_to_root(path
), fn
);
2535 r
= unit_watch_pids_in_path(u
, p
);
2536 if (r
< 0 && ret
>= 0)
2540 if (r
< 0 && ret
>= 0)
2547 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
2552 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
2553 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
2554 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
2556 if (!u
->cgroup_path
)
2559 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2562 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
2565 if (!set_isempty(u
->pids
))
2568 unit_add_to_cgroup_empty_queue(u
);
2572 int unit_watch_all_pids(Unit
*u
) {
2577 /* Adds all PIDs from our cgroup to the set of PIDs we
2578 * watch. This is a fallback logic for cases where we do not
2579 * get reliable cgroup empty notifications: we try to use
2580 * SIGCHLD as replacement. */
2582 if (!u
->cgroup_path
)
2585 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2588 if (r
> 0) /* On unified we can use proper notifications */
2591 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
2594 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
2595 Manager
*m
= userdata
;
2602 u
= m
->cgroup_empty_queue
;
2606 assert(u
->in_cgroup_empty_queue
);
2607 u
->in_cgroup_empty_queue
= false;
2608 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
2610 if (m
->cgroup_empty_queue
) {
2611 /* More stuff queued, let's make sure we remain enabled */
2612 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2614 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
2617 unit_add_to_gc_queue(u
);
2619 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
2620 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
2625 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
2630 /* Note that there are four different ways how cgroup empty events reach us:
2632 * 1. On the unified hierarchy we get an inotify event on the cgroup
2634 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2636 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2638 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2639 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2641 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2642 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2643 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2644 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2645 * case for scope units). */
2647 if (u
->in_cgroup_empty_queue
)
2650 /* Let's verify that the cgroup is really empty */
2651 if (!u
->cgroup_path
)
2653 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2655 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", u
->cgroup_path
);
2661 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2662 u
->in_cgroup_empty_queue
= true;
2664 /* Trigger the defer event */
2665 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
2667 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
2670 int unit_check_oom(Unit
*u
) {
2671 _cleanup_free_
char *oom_kill
= NULL
;
2676 if (!u
->cgroup_path
)
2679 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
2681 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
2683 r
= safe_atou64(oom_kill
, &c
);
2685 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
2687 increased
= c
> u
->oom_kill_last
;
2688 u
->oom_kill_last
= c
;
2693 log_struct(LOG_NOTICE
,
2694 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
2696 LOG_UNIT_INVOCATION_ID(u
),
2697 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
2699 if (UNIT_VTABLE(u
)->notify_cgroup_oom
)
2700 UNIT_VTABLE(u
)->notify_cgroup_oom(u
);
2705 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
2706 Manager
*m
= userdata
;
2713 u
= m
->cgroup_oom_queue
;
2717 assert(u
->in_cgroup_oom_queue
);
2718 u
->in_cgroup_oom_queue
= false;
2719 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
2721 if (m
->cgroup_oom_queue
) {
2722 /* More stuff queued, let's make sure we remain enabled */
2723 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2725 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
2728 (void) unit_check_oom(u
);
2732 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
2737 if (u
->in_cgroup_oom_queue
)
2739 if (!u
->cgroup_path
)
2742 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
2743 u
->in_cgroup_oom_queue
= true;
2745 /* Trigger the defer event */
2746 if (!u
->manager
->cgroup_oom_event_source
) {
2747 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
2749 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
2751 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
2755 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
2757 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
2761 (void) sd_event_source_set_description(s
, "cgroup-oom");
2762 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
2765 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
2767 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
2770 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
2771 Manager
*m
= userdata
;
2778 union inotify_event_buffer buffer
;
2779 struct inotify_event
*e
;
2782 l
= read(fd
, &buffer
, sizeof(buffer
));
2784 if (IN_SET(errno
, EINTR
, EAGAIN
))
2787 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
2790 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
2794 /* Queue overflow has no watch descriptor */
2797 if (e
->mask
& IN_IGNORED
)
2798 /* The watch was just removed */
2801 /* Note that inotify might deliver events for a watch even after it was removed,
2802 * because it was queued before the removal. Let's ignore this here safely. */
2804 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2806 unit_add_to_cgroup_empty_queue(u
);
2808 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2810 unit_add_to_cgroup_oom_queue(u
);
2815 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
2816 CGroupMask mask
= 0;
2819 /* BPF-based firewall */
2820 r
= bpf_firewall_supported();
2822 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2824 /* BPF-based device access control */
2825 r
= bpf_devices_supported();
2827 mask
|= CGROUP_MASK_BPF_DEVICES
;
2833 int manager_setup_cgroup(Manager
*m
) {
2834 _cleanup_free_
char *path
= NULL
;
2835 const char *scope_path
;
2843 /* 1. Determine hierarchy */
2844 m
->cgroup_root
= mfree(m
->cgroup_root
);
2845 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
2847 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
2849 /* Chop off the init scope, if we are already located in it */
2850 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2852 /* LEGACY: Also chop off the system slice if we are in
2853 * it. This is to support live upgrades from older systemd
2854 * versions where PID 1 was moved there. Also see
2855 * cg_get_root_path(). */
2856 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
2857 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
2859 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
2864 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
2865 * easily prepend it everywhere. */
2866 delete_trailing_chars(m
->cgroup_root
, "/");
2869 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
2871 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
2875 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
2877 all_unified
= cg_all_unified();
2878 if (all_unified
< 0)
2879 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
2880 if (all_unified
> 0)
2881 log_debug("Unified cgroup hierarchy is located at %s.", path
);
2883 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2885 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
2887 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
2889 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
2892 /* 3. Allocate cgroup empty defer event source */
2893 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
2894 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
2896 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
2898 /* Schedule cgroup empty checks early, but after having processed service notification messages or
2899 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
2900 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
2901 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
2903 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
2905 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
2907 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
2909 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
2911 /* 4. Install notifier inotify object, or agent */
2912 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
2914 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
2916 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
2917 safe_close(m
->cgroup_inotify_fd
);
2919 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
2920 if (m
->cgroup_inotify_fd
< 0)
2921 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
2923 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
2925 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
2927 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
2928 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
2929 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
2930 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
2932 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
2934 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
2936 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
2938 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
2939 * since it does not generate events when control groups with children run empty. */
2941 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUP_AGENT_PATH
);
2943 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
2945 log_debug("Installed release agent.");
2947 log_debug("Release agent already installed.");
2950 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
2951 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2952 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
2954 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
2955 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
2957 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
2959 /* 6. And pin it, so that it cannot be unmounted */
2960 safe_close(m
->pin_cgroupfs_fd
);
2961 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
2962 if (m
->pin_cgroupfs_fd
< 0)
2963 return log_error_errno(errno
, "Failed to open pin file: %m");
2965 } else if (!MANAGER_IS_TEST_RUN(m
))
2966 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
2968 /* 7. Always enable hierarchical support if it exists... */
2969 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
2970 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
2972 /* 8. Figure out which controllers are supported */
2973 r
= cg_mask_supported(&m
->cgroup_supported
);
2975 return log_error_errno(r
, "Failed to determine supported controllers: %m");
2977 /* 9. Figure out which bpf-based pseudo-controllers are supported */
2978 r
= cg_bpf_mask_supported(&mask
);
2980 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
2981 m
->cgroup_supported
|= mask
;
2983 /* 10. Log which controllers are supported */
2984 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
2985 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
2990 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
2993 /* We can't really delete the group, since we are in it. But
2995 if (delete && m
->cgroup_root
&& m
->test_run_flags
!= MANAGER_TEST_RUN_MINIMAL
)
2996 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
2998 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
3000 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
3001 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
3003 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
3004 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
3006 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
3008 m
->cgroup_root
= mfree(m
->cgroup_root
);
3011 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
3018 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
3022 p
= strdupa(cgroup
);
3026 e
= strrchr(p
, '/');
3028 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
3032 u
= hashmap_get(m
->cgroup_unit
, p
);
3038 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
3039 _cleanup_free_
char *cgroup
= NULL
;
3043 if (!pid_is_valid(pid
))
3046 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
3049 return manager_get_unit_by_cgroup(m
, cgroup
);
3052 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
3057 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
3058 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
3059 * relevant one as children of the process will be assigned to that one, too, before all else. */
3061 if (!pid_is_valid(pid
))
3064 if (pid
== getpid_cached())
3065 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
3067 u
= manager_get_unit_by_pid_cgroup(m
, pid
);
3071 u
= hashmap_get(m
->watch_pids
, PID_TO_PTR(pid
));
3075 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-pid
));
3082 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
3088 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
3089 * or from the --system instance */
3091 log_debug("Got cgroup empty notification for: %s", cgroup
);
3093 u
= manager_get_unit_by_cgroup(m
, cgroup
);
3097 unit_add_to_cgroup_empty_queue(u
);
3101 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
3102 _cleanup_free_
char *v
= NULL
;
3108 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3111 if (!u
->cgroup_path
)
3114 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3115 if (unit_has_host_root_cgroup(u
))
3116 return procfs_memory_get_used(ret
);
3118 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3121 r
= cg_all_unified();
3125 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.current", &v
);
3127 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.usage_in_bytes", &v
);
3133 return safe_atou64(v
, ret
);
3136 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
3137 _cleanup_free_
char *v
= NULL
;
3143 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
3146 if (!u
->cgroup_path
)
3149 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3150 if (unit_has_host_root_cgroup(u
))
3151 return procfs_tasks_get_current(ret
);
3153 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
3156 r
= cg_get_attribute("pids", u
->cgroup_path
, "pids.current", &v
);
3162 return safe_atou64(v
, ret
);
3165 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
3166 _cleanup_free_
char *v
= NULL
;
3173 if (!u
->cgroup_path
)
3176 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3177 if (unit_has_host_root_cgroup(u
))
3178 return procfs_cpu_get_usage(ret
);
3180 /* Requisite controllers for CPU accounting are not enabled */
3181 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
3184 r
= cg_all_unified();
3188 _cleanup_free_
char *val
= NULL
;
3191 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
3192 if (IN_SET(r
, -ENOENT
, -ENXIO
))
3197 r
= safe_atou64(val
, &us
);
3201 ns
= us
* NSEC_PER_USEC
;
3203 r
= cg_get_attribute("cpuacct", u
->cgroup_path
, "cpuacct.usage", &v
);
3209 r
= safe_atou64(v
, &ns
);
3218 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
3224 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
3225 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
3226 * call this function with a NULL return value. */
3228 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
3231 r
= unit_get_cpu_usage_raw(u
, &ns
);
3232 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
3233 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
3237 *ret
= u
->cpu_usage_last
;
3243 if (ns
> u
->cpu_usage_base
)
3244 ns
-= u
->cpu_usage_base
;
3248 u
->cpu_usage_last
= ns
;
3255 int unit_get_ip_accounting(
3257 CGroupIPAccountingMetric metric
,
3264 assert(metric
>= 0);
3265 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
3268 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
3271 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
3272 u
->ip_accounting_ingress_map_fd
:
3273 u
->ip_accounting_egress_map_fd
;
3277 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
3278 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
3280 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
3284 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
3285 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
3286 * ip_accounting_extra[] field, and add them in here transparently. */
3288 *ret
= value
+ u
->ip_accounting_extra
[metric
];
3293 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
3294 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
3295 [CGROUP_IO_READ_BYTES
] = "rbytes=",
3296 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
3297 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
3298 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
3300 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
3301 _cleanup_free_
char *path
= NULL
;
3302 _cleanup_fclose_
FILE *f
= NULL
;
3307 if (!u
->cgroup_path
)
3310 if (unit_has_host_root_cgroup(u
))
3311 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
3313 r
= cg_all_unified();
3316 if (r
== 0) /* TODO: support cgroupv1 */
3319 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
3322 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
3326 f
= fopen(path
, "re");
3331 _cleanup_free_
char *line
= NULL
;
3334 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3341 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
3342 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
3345 _cleanup_free_
char *word
= NULL
;
3347 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
3353 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3356 x
= startswith(word
, field_names
[i
]);
3360 r
= safe_atou64(x
, &w
);
3364 /* Sum up the stats of all devices */
3372 memcpy(ret
, acc
, sizeof(acc
));
3376 int unit_get_io_accounting(
3378 CGroupIOAccountingMetric metric
,
3382 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
3385 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
3387 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
3390 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3393 r
= unit_get_io_accounting_raw(u
, raw
);
3394 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3399 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3400 /* Saturated subtraction */
3401 if (raw
[i
] > u
->io_accounting_base
[i
])
3402 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
3404 u
->io_accounting_last
[i
] = 0;
3409 *ret
= u
->io_accounting_last
[metric
];
3414 int unit_reset_cpu_accounting(Unit
*u
) {
3419 u
->cpu_usage_last
= NSEC_INFINITY
;
3421 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
3423 u
->cpu_usage_base
= 0;
3430 int unit_reset_ip_accounting(Unit
*u
) {
3435 if (u
->ip_accounting_ingress_map_fd
>= 0)
3436 r
= bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
);
3438 if (u
->ip_accounting_egress_map_fd
>= 0)
3439 q
= bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
);
3441 zero(u
->ip_accounting_extra
);
3443 return r
< 0 ? r
: q
;
3446 int unit_reset_io_accounting(Unit
*u
) {
3451 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++)
3452 u
->io_accounting_last
[i
] = UINT64_MAX
;
3454 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
3456 zero(u
->io_accounting_base
);
3463 int unit_reset_accounting(Unit
*u
) {
3468 r
= unit_reset_cpu_accounting(u
);
3469 q
= unit_reset_io_accounting(u
);
3470 v
= unit_reset_ip_accounting(u
);
3472 return r
< 0 ? r
: q
< 0 ? q
: v
;
3475 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
3478 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3484 /* always invalidate compat pairs together */
3485 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
3486 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
3488 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
3489 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
3491 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
3494 u
->cgroup_invalidated_mask
|= m
;
3495 unit_add_to_cgroup_realize_queue(u
);
3498 void unit_invalidate_cgroup_bpf(Unit
*u
) {
3501 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3504 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
3507 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
3508 unit_add_to_cgroup_realize_queue(u
);
3510 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
3511 * list of our children includes our own. */
3512 if (u
->type
== UNIT_SLICE
) {
3517 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
], i
) {
3518 if (UNIT_DEREF(member
->slice
) == u
)
3519 unit_invalidate_cgroup_bpf(member
);
3524 bool unit_cgroup_delegate(Unit
*u
) {
3529 if (!UNIT_VTABLE(u
)->can_delegate
)
3532 c
= unit_get_cgroup_context(u
);
3539 void manager_invalidate_startup_units(Manager
*m
) {
3545 SET_FOREACH(u
, m
->startup_units
, i
)
3546 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
3549 static int unit_get_nice(Unit
*u
) {
3552 ec
= unit_get_exec_context(u
);
3553 return ec
? ec
->nice
: 0;
3556 static uint64_t unit_get_cpu_weight(Unit
*u
) {
3557 ManagerState state
= manager_state(u
->manager
);
3560 cc
= unit_get_cgroup_context(u
);
3561 return cc
? cgroup_context_cpu_weight(cc
, state
) : CGROUP_WEIGHT_DEFAULT
;
3564 int compare_job_priority(const void *a
, const void *b
) {
3565 const Job
*x
= a
, *y
= b
;
3567 uint64_t weight_x
, weight_y
;
3570 if ((ret
= CMP(x
->unit
->type
, y
->unit
->type
)) != 0)
3573 weight_x
= unit_get_cpu_weight(x
->unit
);
3574 weight_y
= unit_get_cpu_weight(y
->unit
);
3576 if ((ret
= CMP(weight_x
, weight_y
)) != 0)
3579 nice_x
= unit_get_nice(x
->unit
);
3580 nice_y
= unit_get_nice(y
->unit
);
3582 if ((ret
= CMP(nice_x
, nice_y
)) != 0)
3585 return strcmp(x
->unit
->id
, y
->unit
->id
);
3588 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
3589 [CGROUP_AUTO
] = "auto",
3590 [CGROUP_CLOSED
] = "closed",
3591 [CGROUP_STRICT
] = "strict",
3594 int unit_get_cpuset(Unit
*u
, CPUSet
*cpus
, const char *name
) {
3595 _cleanup_free_
char *v
= NULL
;
3601 if (!u
->cgroup_path
)
3604 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUSET
) == 0)
3607 r
= cg_all_unified();
3613 r
= cg_get_attribute("cpuset", u
->cgroup_path
, name
, &v
);
3619 return parse_cpu_set_full(v
, cpus
, false, NULL
, NULL
, 0, NULL
);
3622 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);