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-util.h"
19 #include "nulstr-util.h"
20 #include "parse-util.h"
21 #include "path-util.h"
22 #include "process-util.h"
23 #include "procfs-util.h"
25 #include "stat-util.h"
26 #include "stdio-util.h"
27 #include "string-table.h"
28 #include "string-util.h"
31 #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
33 /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access
34 * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask
35 * out specific attributes from us. */
36 #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING)
38 bool manager_owns_host_root_cgroup(Manager
*m
) {
41 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
42 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
43 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
44 * we run in any kind of container virtualization. */
46 if (MANAGER_IS_USER(m
))
49 if (detect_container() > 0)
52 return empty_or_root(m
->cgroup_root
);
55 bool unit_has_host_root_cgroup(Unit
*u
) {
58 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
59 * the manager manages the root cgroup. */
61 if (!manager_owns_host_root_cgroup(u
->manager
))
64 return unit_has_name(u
, SPECIAL_ROOT_SLICE
);
67 static int set_attribute_and_warn(Unit
*u
, const char *controller
, const char *attribute
, const char *value
) {
70 r
= cg_set_attribute(controller
, u
->cgroup_path
, attribute
, value
);
72 log_unit_full(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%.*s': %m",
73 strna(attribute
), isempty(u
->cgroup_path
) ? "/" : u
->cgroup_path
, (int) strcspn(value
, NEWLINE
), value
);
78 static void cgroup_compat_warn(void) {
79 static bool cgroup_compat_warned
= false;
81 if (cgroup_compat_warned
)
84 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
85 "See cgroup-compat debug messages for details.");
87 cgroup_compat_warned
= true;
90 #define log_cgroup_compat(unit, fmt, ...) do { \
91 cgroup_compat_warn(); \
92 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
95 void cgroup_context_init(CGroupContext
*c
) {
98 /* Initialize everything to the kernel defaults. */
100 *c
= (CGroupContext
) {
101 .cpu_weight
= CGROUP_WEIGHT_INVALID
,
102 .startup_cpu_weight
= CGROUP_WEIGHT_INVALID
,
103 .cpu_quota_per_sec_usec
= USEC_INFINITY
,
104 .cpu_quota_period_usec
= USEC_INFINITY
,
106 .cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
107 .startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
109 .memory_high
= CGROUP_LIMIT_MAX
,
110 .memory_max
= CGROUP_LIMIT_MAX
,
111 .memory_swap_max
= CGROUP_LIMIT_MAX
,
113 .memory_limit
= CGROUP_LIMIT_MAX
,
115 .io_weight
= CGROUP_WEIGHT_INVALID
,
116 .startup_io_weight
= CGROUP_WEIGHT_INVALID
,
118 .blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
119 .startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
121 .tasks_max
= CGROUP_LIMIT_MAX
,
125 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
129 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
134 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
138 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
143 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
147 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
152 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
156 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
161 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
165 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
170 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
174 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
179 void cgroup_context_done(CGroupContext
*c
) {
182 while (c
->io_device_weights
)
183 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
185 while (c
->io_device_latencies
)
186 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
188 while (c
->io_device_limits
)
189 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
191 while (c
->blockio_device_weights
)
192 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
194 while (c
->blockio_device_bandwidths
)
195 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
197 while (c
->device_allow
)
198 cgroup_context_free_device_allow(c
, c
->device_allow
);
200 c
->ip_address_allow
= ip_address_access_free_all(c
->ip_address_allow
);
201 c
->ip_address_deny
= ip_address_access_free_all(c
->ip_address_deny
);
203 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
204 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
207 void cgroup_context_dump(CGroupContext
*c
, FILE* f
, const char *prefix
) {
208 _cleanup_free_
char *disable_controllers_str
= NULL
;
209 CGroupIODeviceLimit
*il
;
210 CGroupIODeviceWeight
*iw
;
211 CGroupIODeviceLatency
*l
;
212 CGroupBlockIODeviceBandwidth
*b
;
213 CGroupBlockIODeviceWeight
*w
;
214 CGroupDeviceAllow
*a
;
215 IPAddressAccessItem
*iaai
;
217 char u
[FORMAT_TIMESPAN_MAX
];
218 char v
[FORMAT_TIMESPAN_MAX
];
223 prefix
= strempty(prefix
);
225 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
228 "%sCPUAccounting=%s\n"
229 "%sIOAccounting=%s\n"
230 "%sBlockIOAccounting=%s\n"
231 "%sMemoryAccounting=%s\n"
232 "%sTasksAccounting=%s\n"
233 "%sIPAccounting=%s\n"
234 "%sCPUWeight=%" PRIu64
"\n"
235 "%sStartupCPUWeight=%" PRIu64
"\n"
236 "%sCPUShares=%" PRIu64
"\n"
237 "%sStartupCPUShares=%" PRIu64
"\n"
238 "%sCPUQuotaPerSecSec=%s\n"
239 "%sCPUQuotaPeriodSec=%s\n"
240 "%sIOWeight=%" PRIu64
"\n"
241 "%sStartupIOWeight=%" PRIu64
"\n"
242 "%sBlockIOWeight=%" PRIu64
"\n"
243 "%sStartupBlockIOWeight=%" PRIu64
"\n"
244 "%sDefaultMemoryMin=%" PRIu64
"\n"
245 "%sDefaultMemoryLow=%" PRIu64
"\n"
246 "%sMemoryMin=%" PRIu64
"\n"
247 "%sMemoryLow=%" PRIu64
"\n"
248 "%sMemoryHigh=%" PRIu64
"\n"
249 "%sMemoryMax=%" PRIu64
"\n"
250 "%sMemorySwapMax=%" PRIu64
"\n"
251 "%sMemoryLimit=%" PRIu64
"\n"
252 "%sTasksMax=%" PRIu64
"\n"
253 "%sDevicePolicy=%s\n"
254 "%sDisableControllers=%s\n"
256 prefix
, yes_no(c
->cpu_accounting
),
257 prefix
, yes_no(c
->io_accounting
),
258 prefix
, yes_no(c
->blockio_accounting
),
259 prefix
, yes_no(c
->memory_accounting
),
260 prefix
, yes_no(c
->tasks_accounting
),
261 prefix
, yes_no(c
->ip_accounting
),
262 prefix
, c
->cpu_weight
,
263 prefix
, c
->startup_cpu_weight
,
264 prefix
, c
->cpu_shares
,
265 prefix
, c
->startup_cpu_shares
,
266 prefix
, format_timespan(u
, sizeof(u
), c
->cpu_quota_per_sec_usec
, 1),
267 prefix
, format_timespan(v
, sizeof(v
), c
->cpu_quota_period_usec
, 1),
268 prefix
, c
->io_weight
,
269 prefix
, c
->startup_io_weight
,
270 prefix
, c
->blockio_weight
,
271 prefix
, c
->startup_blockio_weight
,
272 prefix
, c
->default_memory_min
,
273 prefix
, c
->default_memory_low
,
274 prefix
, c
->memory_min
,
275 prefix
, c
->memory_low
,
276 prefix
, c
->memory_high
,
277 prefix
, c
->memory_max
,
278 prefix
, c
->memory_swap_max
,
279 prefix
, c
->memory_limit
,
280 prefix
, c
->tasks_max
,
281 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
282 prefix
, strempty(disable_controllers_str
),
283 prefix
, yes_no(c
->delegate
));
286 _cleanup_free_
char *t
= NULL
;
288 (void) cg_mask_to_string(c
->delegate_controllers
, &t
);
290 fprintf(f
, "%sDelegateControllers=%s\n",
295 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
297 "%sDeviceAllow=%s %s%s%s\n",
300 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
302 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
304 "%sIODeviceWeight=%s %" PRIu64
"\n",
309 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
311 "%sIODeviceLatencyTargetSec=%s %s\n",
314 format_timespan(u
, sizeof(u
), l
->target_usec
, 1));
316 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
317 char buf
[FORMAT_BYTES_MAX
];
318 CGroupIOLimitType type
;
320 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
321 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
325 cgroup_io_limit_type_to_string(type
),
327 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
330 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
332 "%sBlockIODeviceWeight=%s %" PRIu64
,
337 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
338 char buf
[FORMAT_BYTES_MAX
];
340 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
342 "%sBlockIOReadBandwidth=%s %s\n",
345 format_bytes(buf
, sizeof(buf
), b
->rbps
));
346 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
348 "%sBlockIOWriteBandwidth=%s %s\n",
351 format_bytes(buf
, sizeof(buf
), b
->wbps
));
354 LIST_FOREACH(items
, iaai
, c
->ip_address_allow
) {
355 _cleanup_free_
char *k
= NULL
;
357 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
358 fprintf(f
, "%sIPAddressAllow=%s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
361 LIST_FOREACH(items
, iaai
, c
->ip_address_deny
) {
362 _cleanup_free_
char *k
= NULL
;
364 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
365 fprintf(f
, "%sIPAddressDeny=%s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
368 STRV_FOREACH(path
, c
->ip_filters_ingress
)
369 fprintf(f
, "%sIPIngressFilterPath=%s\n", prefix
, *path
);
371 STRV_FOREACH(path
, c
->ip_filters_egress
)
372 fprintf(f
, "%sIPEgressFilterPath=%s\n", prefix
, *path
);
375 int cgroup_add_device_allow(CGroupContext
*c
, const char *dev
, const char *mode
) {
376 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
377 _cleanup_free_
char *d
= NULL
;
381 assert(isempty(mode
) || in_charset(mode
, "rwm"));
383 a
= new(CGroupDeviceAllow
, 1);
391 *a
= (CGroupDeviceAllow
) {
393 .r
= isempty(mode
) || strchr(mode
, 'r'),
394 .w
= isempty(mode
) || strchr(mode
, 'w'),
395 .m
= isempty(mode
) || strchr(mode
, 'm'),
398 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
404 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
405 uint64_t unit_get_ancestor_##entry(Unit *u) { \
408 /* 1. Is entry set in this unit? If so, use that. \
409 * 2. Is the default for this entry set in any \
410 * ancestor? If so, use that. \
411 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
415 c = unit_get_cgroup_context(u); \
416 if (c && c->entry##_set) \
419 while ((u = UNIT_DEREF(u->slice))) { \
420 c = unit_get_cgroup_context(u); \
421 if (c && c->default_##entry##_set) \
422 return c->default_##entry; \
425 /* We've reached the root, but nobody had default for \
426 * this entry set, so set it to the kernel default. */ \
427 return CGROUP_LIMIT_MIN; \
430 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
431 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
433 static void cgroup_xattr_apply(Unit
*u
) {
434 char ids
[SD_ID128_STRING_MAX
];
439 if (!MANAGER_IS_SYSTEM(u
->manager
))
442 if (sd_id128_is_null(u
->invocation_id
))
445 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
446 "trusted.invocation_id",
447 sd_id128_to_string(u
->invocation_id
, ids
), 32,
450 log_unit_debug_errno(u
, r
, "Failed to set invocation ID on control group %s, ignoring: %m", u
->cgroup_path
);
453 static int lookup_block_device(const char *p
, dev_t
*ret
) {
461 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
462 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
464 if (stat(p
, &st
) < 0)
465 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
466 rdev
= (dev_t
)st
.st_rdev
;
467 dev
= (dev_t
)st
.st_dev
;
470 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
473 log_warning("Device node '%s' is a character device, but block device needed.", p
);
475 } else if (S_ISBLK(mode
))
477 else if (major(dev
) != 0)
478 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
480 /* If this is btrfs, getting the backing block device is a bit harder */
481 r
= btrfs_get_block_device(p
, ret
);
482 if (r
< 0 && r
!= -ENOTTY
)
483 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
485 log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
490 /* If this is a LUKS device, try to get the originating block device */
491 (void) block_get_originating(*ret
, ret
);
493 /* If this is a partition, try to get the originating block device */
494 (void) block_get_whole_disk(*ret
, ret
);
498 static int whitelist_device(BPFProgram
*prog
, const char *path
, const char *node
, const char *acc
) {
506 /* Some special handling for /dev/block/%u:%u, /dev/char/%u:%u, /run/systemd/inaccessible/chr and
507 * /run/systemd/inaccessible/blk paths. Instead of stat()ing these we parse out the major/minor directly. This
508 * means clients can use these path without the device node actually around */
509 r
= device_path_parse_major_minor(node
, &mode
, &rdev
);
512 return log_warning_errno(r
, "Couldn't parse major/minor from device path '%s': %m", node
);
515 if (stat(node
, &st
) < 0)
516 return log_warning_errno(errno
, "Couldn't stat device %s: %m", node
);
518 if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
)) {
519 log_warning("%s is not a device.", node
);
522 rdev
= (dev_t
) st
.st_rdev
;
526 if (cg_all_unified() > 0) {
530 return cgroup_bpf_whitelist_device(prog
, S_ISCHR(mode
) ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
531 major(rdev
), minor(rdev
), acc
);
534 char buf
[2+DECIMAL_STR_MAX(dev_t
)*2+2+4];
538 S_ISCHR(mode
) ? 'c' : 'b',
539 major(rdev
), minor(rdev
),
542 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */
544 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
546 return log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
547 r
, "Failed to set devices.allow on %s: %m", path
);
553 static int whitelist_major(BPFProgram
*prog
, const char *path
, const char *name
, char type
, const char *acc
) {
554 _cleanup_fclose_
FILE *f
= NULL
;
555 char buf
[2+DECIMAL_STR_MAX(unsigned)+3+4];
562 assert(IN_SET(type
, 'b', 'c'));
564 if (streq(name
, "*")) {
565 /* If the name is a wildcard, then apply this list to all devices of this type */
567 if (cg_all_unified() > 0) {
571 (void) cgroup_bpf_whitelist_class(prog
, type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
, acc
);
573 xsprintf(buf
, "%c *:* %s", type
, acc
);
575 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
577 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
578 "Failed to set devices.allow on %s: %m", path
);
583 if (safe_atou(name
, &maj
) >= 0 && DEVICE_MAJOR_VALID(maj
)) {
584 /* The name is numeric and suitable as major. In that case, let's take is major, and create the entry
587 if (cg_all_unified() > 0) {
591 (void) cgroup_bpf_whitelist_major(prog
,
592 type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
595 xsprintf(buf
, "%c %u:* %s", type
, maj
, acc
);
597 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
599 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
600 "Failed to set devices.allow on %s: %m", path
);
606 f
= fopen("/proc/devices", "re");
608 return log_warning_errno(errno
, "Cannot open /proc/devices to resolve %s (%c): %m", name
, type
);
611 _cleanup_free_
char *line
= NULL
;
614 r
= read_line(f
, LONG_LINE_MAX
, &line
);
616 return log_warning_errno(r
, "Failed to read /proc/devices: %m");
620 if (type
== 'c' && streq(line
, "Character devices:")) {
625 if (type
== 'b' && streq(line
, "Block devices:")) {
640 w
= strpbrk(p
, WHITESPACE
);
645 r
= safe_atou(p
, &maj
);
652 w
+= strspn(w
, WHITESPACE
);
654 if (fnmatch(name
, w
, 0) != 0)
657 if (cg_all_unified() > 0) {
661 (void) cgroup_bpf_whitelist_major(prog
,
662 type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
671 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL
674 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
676 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
677 r
, "Failed to set devices.allow on %s: %m", path
);
684 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
685 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
686 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
689 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
690 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
691 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
694 static uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
695 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
696 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
697 return c
->startup_cpu_weight
;
698 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
699 return c
->cpu_weight
;
701 return CGROUP_WEIGHT_DEFAULT
;
704 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
705 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
706 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
707 return c
->startup_cpu_shares
;
708 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
709 return c
->cpu_shares
;
711 return CGROUP_CPU_SHARES_DEFAULT
;
714 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
715 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
716 * need to be higher than that boundary. quota is specified in USecPerSec.
717 * Additionally, period must be at most max_period. */
720 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
723 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
726 if (quota
== USEC_INFINITY
)
727 /* Always use default period for infinity quota. */
728 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
730 if (period
== USEC_INFINITY
)
731 /* Default period was requested. */
732 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
734 /* Clamp to interval [1ms, 1s] */
735 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
737 if (new_period
!= period
) {
738 char v
[FORMAT_TIMESPAN_MAX
];
739 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
, 0,
740 "Clamping CPU interval for cpu.max: period is now %s",
741 format_timespan(v
, sizeof(v
), new_period
, 1));
742 u
->warned_clamping_cpu_quota_period
= true;
748 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
749 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
751 xsprintf(buf
, "%" PRIu64
"\n", weight
);
752 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
755 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
756 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
758 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
759 if (quota
!= USEC_INFINITY
)
760 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
761 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
763 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
764 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
767 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
768 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
770 xsprintf(buf
, "%" PRIu64
"\n", shares
);
771 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
774 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
775 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
777 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
779 xsprintf(buf
, USEC_FMT
"\n", period
);
780 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
782 if (quota
!= USEC_INFINITY
) {
783 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
784 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
786 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
789 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
790 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
791 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
794 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
795 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
796 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
799 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
800 return c
->io_accounting
||
801 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
802 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
803 c
->io_device_weights
||
804 c
->io_device_latencies
||
808 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
809 return c
->blockio_accounting
||
810 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
811 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
812 c
->blockio_device_weights
||
813 c
->blockio_device_bandwidths
;
816 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
817 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
818 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
819 return c
->startup_io_weight
;
820 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
823 return CGROUP_WEIGHT_DEFAULT
;
826 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
827 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
828 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
829 return c
->startup_blockio_weight
;
830 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
831 return c
->blockio_weight
;
833 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
836 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
837 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
838 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
841 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
842 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
843 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
846 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
847 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
851 r
= lookup_block_device(dev_path
, &dev
);
855 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
856 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
859 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
860 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
864 r
= lookup_block_device(dev_path
, &dev
);
868 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
869 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
872 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
873 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
877 r
= lookup_block_device(dev_path
, &dev
);
881 if (target
!= USEC_INFINITY
)
882 xsprintf(buf
, "%u:%u target=%" PRIu64
"\n", major(dev
), minor(dev
), target
);
884 xsprintf(buf
, "%u:%u target=max\n", major(dev
), minor(dev
));
886 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
889 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
890 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
891 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
892 CGroupIOLimitType type
;
896 r
= lookup_block_device(dev_path
, &dev
);
900 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
901 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
902 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
904 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
906 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
907 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
908 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
909 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
912 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
913 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
917 r
= lookup_block_device(dev_path
, &dev
);
921 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
922 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
924 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
925 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
928 static bool unit_has_unified_memory_config(Unit
*u
) {
933 c
= unit_get_cgroup_context(u
);
936 return c
->memory_min
> 0 || unit_get_ancestor_memory_low(u
) > 0 ||
937 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
||
938 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
;
941 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
942 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
944 if (v
!= CGROUP_LIMIT_MAX
)
945 xsprintf(buf
, "%" PRIu64
"\n", v
);
947 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
950 static void cgroup_apply_firewall(Unit
*u
) {
953 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
955 if (bpf_firewall_compile(u
) < 0)
958 (void) bpf_firewall_load_custom(u
);
959 (void) bpf_firewall_install(u
);
962 static void cgroup_context_apply(
964 CGroupMask apply_mask
,
965 ManagerState state
) {
969 bool is_host_root
, is_local_root
;
974 /* Nothing to do? Exit early! */
978 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
979 * attributes should only be managed for cgroups further down the tree. */
980 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
981 is_host_root
= unit_has_host_root_cgroup(u
);
983 assert_se(c
= unit_get_cgroup_context(u
));
984 assert_se(path
= u
->cgroup_path
);
986 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
989 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
990 * then), and missing cgroups, i.e. EROFS and ENOENT. */
992 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
993 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
994 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
995 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
996 * we couldn't even write to them if we wanted to). */
997 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
999 if (cg_all_unified() > 0) {
1002 if (cgroup_context_has_cpu_weight(c
))
1003 weight
= cgroup_context_cpu_weight(c
, state
);
1004 else if (cgroup_context_has_cpu_shares(c
)) {
1007 shares
= cgroup_context_cpu_shares(c
, state
);
1008 weight
= cgroup_cpu_shares_to_weight(shares
);
1010 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
1011 shares
, weight
, path
);
1013 weight
= CGROUP_WEIGHT_DEFAULT
;
1015 cgroup_apply_unified_cpu_weight(u
, weight
);
1016 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1021 if (cgroup_context_has_cpu_weight(c
)) {
1024 weight
= cgroup_context_cpu_weight(c
, state
);
1025 shares
= cgroup_cpu_weight_to_shares(weight
);
1027 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
1028 weight
, shares
, path
);
1029 } else if (cgroup_context_has_cpu_shares(c
))
1030 shares
= cgroup_context_cpu_shares(c
, state
);
1032 shares
= CGROUP_CPU_SHARES_DEFAULT
;
1034 cgroup_apply_legacy_cpu_shares(u
, shares
);
1035 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1039 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
1040 * controller), and in case of containers we want to leave control of these attributes to the container manager
1041 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
1042 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
1043 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
1044 bool has_io
, has_blockio
;
1047 has_io
= cgroup_context_has_io_config(c
);
1048 has_blockio
= cgroup_context_has_blockio_config(c
);
1051 weight
= cgroup_context_io_weight(c
, state
);
1052 else if (has_blockio
) {
1053 uint64_t blkio_weight
;
1055 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
1056 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
1058 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
1059 blkio_weight
, weight
);
1061 weight
= CGROUP_WEIGHT_DEFAULT
;
1063 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1064 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1067 CGroupIODeviceLatency
*latency
;
1068 CGroupIODeviceLimit
*limit
;
1069 CGroupIODeviceWeight
*w
;
1071 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
1072 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
1074 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
1075 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
1077 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
1078 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
1080 } else if (has_blockio
) {
1081 CGroupBlockIODeviceWeight
*w
;
1082 CGroupBlockIODeviceBandwidth
*b
;
1084 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
1085 weight
= cgroup_weight_blkio_to_io(w
->weight
);
1087 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
1088 w
->weight
, weight
, w
->path
);
1090 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
1093 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
1094 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
1095 CGroupIOLimitType type
;
1097 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1098 limits
[type
] = cgroup_io_limit_defaults
[type
];
1100 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
1101 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
1103 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
1104 b
->rbps
, b
->wbps
, b
->path
);
1106 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
1111 if (apply_mask
& CGROUP_MASK_BLKIO
) {
1112 bool has_io
, has_blockio
;
1114 has_io
= cgroup_context_has_io_config(c
);
1115 has_blockio
= cgroup_context_has_blockio_config(c
);
1117 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
1118 * left to our container manager, too. */
1119 if (!is_local_root
) {
1120 char buf
[DECIMAL_STR_MAX(uint64_t)+1];
1126 io_weight
= cgroup_context_io_weight(c
, state
);
1127 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
1129 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
1131 } else if (has_blockio
)
1132 weight
= cgroup_context_blkio_weight(c
, state
);
1134 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
1136 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1137 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1140 CGroupIODeviceWeight
*w
;
1142 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
1143 weight
= cgroup_weight_io_to_blkio(w
->weight
);
1145 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
1146 w
->weight
, weight
, w
->path
);
1148 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
1150 } else if (has_blockio
) {
1151 CGroupBlockIODeviceWeight
*w
;
1153 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
1154 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
1158 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
1159 * roots, as there we want the container manager to handle it */
1160 if (is_host_root
|| !is_local_root
) {
1162 CGroupIODeviceLimit
*l
;
1164 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
1165 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
1166 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
1168 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
1170 } else if (has_blockio
) {
1171 CGroupBlockIODeviceBandwidth
*b
;
1173 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
1174 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
1179 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
1180 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
1181 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
1182 * write to this if we wanted to.) */
1183 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
1185 if (cg_all_unified() > 0) {
1186 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
;
1188 if (unit_has_unified_memory_config(u
)) {
1189 max
= c
->memory_max
;
1190 swap_max
= c
->memory_swap_max
;
1192 max
= c
->memory_limit
;
1194 if (max
!= CGROUP_LIMIT_MAX
)
1195 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
1198 cgroup_apply_unified_memory_limit(u
, "memory.min", c
->memory_min
);
1199 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
1200 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
1201 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
1202 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
1204 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
1207 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1210 if (unit_has_unified_memory_config(u
)) {
1211 val
= c
->memory_max
;
1212 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIi64
" as MemoryLimit=", val
);
1214 val
= c
->memory_limit
;
1216 if (val
== CGROUP_LIMIT_MAX
)
1217 strncpy(buf
, "-1\n", sizeof(buf
));
1219 xsprintf(buf
, "%" PRIu64
"\n", val
);
1221 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
1225 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
1226 * containers, where we leave this to the manager */
1227 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
1228 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
)) {
1229 _cleanup_(bpf_program_unrefp
) BPFProgram
*prog
= NULL
;
1230 CGroupDeviceAllow
*a
;
1232 if (cg_all_unified() > 0) {
1233 r
= cgroup_init_device_bpf(&prog
, c
->device_policy
, c
->device_allow
);
1235 log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
1237 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL
1240 if (c
->device_allow
|| c
->device_policy
!= CGROUP_AUTO
)
1241 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
1243 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
1245 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1246 "Failed to reset devices.allow/devices.deny: %m");
1249 if (c
->device_policy
== CGROUP_CLOSED
||
1250 (c
->device_policy
== CGROUP_AUTO
&& c
->device_allow
)) {
1251 static const char auto_devices
[] =
1252 "/dev/null\0" "rwm\0"
1253 "/dev/zero\0" "rwm\0"
1254 "/dev/full\0" "rwm\0"
1255 "/dev/random\0" "rwm\0"
1256 "/dev/urandom\0" "rwm\0"
1257 "/dev/tty\0" "rwm\0"
1258 "/dev/ptmx\0" "rwm\0"
1259 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
1260 "/run/systemd/inaccessible/chr\0" "rwm\0"
1261 "/run/systemd/inaccessible/blk\0" "rwm\0";
1265 NULSTR_FOREACH_PAIR(x
, y
, auto_devices
)
1266 (void) whitelist_device(prog
, path
, x
, y
);
1268 /* PTS (/dev/pts) devices may not be duplicated, but accessed */
1269 (void) whitelist_major(prog
, path
, "pts", 'c', "rw");
1272 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1288 if (path_startswith(a
->path
, "/dev/"))
1289 (void) whitelist_device(prog
, path
, a
->path
, acc
);
1290 else if ((val
= startswith(a
->path
, "block-")))
1291 (void) whitelist_major(prog
, path
, val
, 'b', acc
);
1292 else if ((val
= startswith(a
->path
, "char-")))
1293 (void) whitelist_major(prog
, path
, val
, 'c', acc
);
1295 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1298 r
= cgroup_apply_device_bpf(u
, prog
, c
->device_policy
, c
->device_allow
);
1300 static bool warned
= false;
1302 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1303 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1304 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1305 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1311 if (apply_mask
& CGROUP_MASK_PIDS
) {
1314 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1315 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1316 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1317 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1318 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1319 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1320 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1321 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1322 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1323 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1324 * which is desirable so that there's an official way to release control of the sysctl from
1325 * systemd: set the limit to unbounded and reload. */
1327 if (c
->tasks_max
!= CGROUP_LIMIT_MAX
) {
1328 u
->manager
->sysctl_pid_max_changed
= true;
1329 r
= procfs_tasks_set_limit(c
->tasks_max
);
1330 } else if (u
->manager
->sysctl_pid_max_changed
)
1331 r
= procfs_tasks_set_limit(TASKS_MAX
);
1335 log_unit_full(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
1336 "Failed to write to tasks limit sysctls: %m");
1339 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
1340 * leave it for the container manager. */
1341 if (!is_local_root
) {
1342 if (c
->tasks_max
!= CGROUP_LIMIT_MAX
) {
1343 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
1345 sprintf(buf
, "%" PRIu64
"\n", c
->tasks_max
);
1346 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
1348 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
1352 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
1353 cgroup_apply_firewall(u
);
1356 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
1361 c
= unit_get_cgroup_context(u
);
1365 if (c
->ip_accounting
||
1366 c
->ip_address_allow
||
1367 c
->ip_address_deny
||
1368 c
->ip_filters_ingress
||
1369 c
->ip_filters_egress
)
1372 /* If any parent slice has an IP access list defined, it applies too */
1373 for (p
= UNIT_DEREF(u
->slice
); p
; p
= UNIT_DEREF(p
->slice
)) {
1374 c
= unit_get_cgroup_context(p
);
1378 if (c
->ip_address_allow
||
1386 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
1387 CGroupMask mask
= 0;
1392 c
= unit_get_cgroup_context(u
);
1396 /* Figure out which controllers we need, based on the cgroup context object */
1398 if (c
->cpu_accounting
)
1399 mask
|= get_cpu_accounting_mask();
1401 if (cgroup_context_has_cpu_weight(c
) ||
1402 cgroup_context_has_cpu_shares(c
) ||
1403 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
1404 mask
|= CGROUP_MASK_CPU
;
1406 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
1407 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1409 if (c
->memory_accounting
||
1410 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
1411 unit_has_unified_memory_config(u
))
1412 mask
|= CGROUP_MASK_MEMORY
;
1414 if (c
->device_allow
||
1415 c
->device_policy
!= CGROUP_AUTO
)
1416 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
1418 if (c
->tasks_accounting
||
1419 c
->tasks_max
!= CGROUP_LIMIT_MAX
)
1420 mask
|= CGROUP_MASK_PIDS
;
1422 return CGROUP_MASK_EXTEND_JOINED(mask
);
1425 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
1426 CGroupMask mask
= 0;
1428 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
1431 if (unit_get_needs_bpf_firewall(u
))
1432 mask
|= CGROUP_MASK_BPF_FIREWALL
;
1437 CGroupMask
unit_get_own_mask(Unit
*u
) {
1440 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
1441 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
1443 if (u
->load_state
!= UNIT_LOADED
)
1446 c
= unit_get_cgroup_context(u
);
1450 return (unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
)) & ~unit_get_ancestor_disable_mask(u
);
1453 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
1456 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1457 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1459 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1461 if (!unit_cgroup_delegate(u
))
1464 if (cg_all_unified() <= 0) {
1467 e
= unit_get_exec_context(u
);
1468 if (e
&& !exec_context_maintains_privileges(e
))
1472 assert_se(c
= unit_get_cgroup_context(u
));
1473 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
1476 CGroupMask
unit_get_members_mask(Unit
*u
) {
1479 /* Returns the mask of controllers all of the unit's children require, merged */
1481 if (u
->cgroup_members_mask_valid
)
1482 return u
->cgroup_members_mask
; /* Use cached value if possible */
1484 u
->cgroup_members_mask
= 0;
1486 if (u
->type
== UNIT_SLICE
) {
1491 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
], i
) {
1492 if (UNIT_DEREF(member
->slice
) == u
)
1493 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
1497 u
->cgroup_members_mask_valid
= true;
1498 return u
->cgroup_members_mask
;
1501 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
1504 /* Returns the mask of controllers all of the unit's siblings
1505 * require, i.e. the members mask of the unit's parent slice
1506 * if there is one. */
1508 if (UNIT_ISSET(u
->slice
))
1509 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
1511 return unit_get_subtree_mask(u
); /* we are the top-level slice */
1514 CGroupMask
unit_get_disable_mask(Unit
*u
) {
1517 c
= unit_get_cgroup_context(u
);
1521 return c
->disable_controllers
;
1524 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
1528 mask
= unit_get_disable_mask(u
);
1530 /* Returns the mask of controllers which are marked as forcibly
1531 * disabled in any ancestor unit or the unit in question. */
1533 if (UNIT_ISSET(u
->slice
))
1534 mask
|= unit_get_ancestor_disable_mask(UNIT_DEREF(u
->slice
));
1539 CGroupMask
unit_get_subtree_mask(Unit
*u
) {
1541 /* Returns the mask of this subtree, meaning of the group
1542 * itself and its children. */
1544 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
1547 CGroupMask
unit_get_target_mask(Unit
*u
) {
1550 /* This returns the cgroup mask of all controllers to enable
1551 * for a specific cgroup, i.e. everything it needs itself,
1552 * plus all that its children need, plus all that its siblings
1553 * need. This is primarily useful on the legacy cgroup
1554 * hierarchy, where we need to duplicate each cgroup in each
1555 * hierarchy that shall be enabled for it. */
1557 mask
= unit_get_own_mask(u
) | unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
1559 if (mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
1560 emit_bpf_firewall_warning(u
);
1562 mask
&= u
->manager
->cgroup_supported
;
1563 mask
&= ~unit_get_ancestor_disable_mask(u
);
1568 CGroupMask
unit_get_enable_mask(Unit
*u
) {
1571 /* This returns the cgroup mask of all controllers to enable
1572 * for the children of a specific cgroup. This is primarily
1573 * useful for the unified cgroup hierarchy, where each cgroup
1574 * controls which controllers are enabled for its children. */
1576 mask
= unit_get_members_mask(u
);
1577 mask
&= u
->manager
->cgroup_supported
;
1578 mask
&= ~unit_get_ancestor_disable_mask(u
);
1583 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
1586 /* Recurse invalidate the member masks cache all the way up the tree */
1587 u
->cgroup_members_mask_valid
= false;
1589 if (UNIT_ISSET(u
->slice
))
1590 unit_invalidate_cgroup_members_masks(UNIT_DEREF(u
->slice
));
1593 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
1595 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1599 if (u
->cgroup_path
&&
1600 u
->cgroup_realized
&&
1601 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
1602 return u
->cgroup_path
;
1604 u
= UNIT_DEREF(u
->slice
);
1610 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1611 return unit_get_realized_cgroup_path(userdata
, mask
);
1614 char *unit_default_cgroup_path(const Unit
*u
) {
1615 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1620 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1621 return strdup(u
->manager
->cgroup_root
);
1623 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1624 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1629 escaped
= cg_escape(u
->id
);
1633 return path_join(empty_to_root(u
->manager
->cgroup_root
), slice
, escaped
);
1636 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1637 _cleanup_free_
char *p
= NULL
;
1642 if (streq_ptr(u
->cgroup_path
, path
))
1652 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1657 unit_release_cgroup(u
);
1658 u
->cgroup_path
= TAKE_PTR(p
);
1663 int unit_watch_cgroup(Unit
*u
) {
1664 _cleanup_free_
char *events
= NULL
;
1669 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
1670 * cgroupv2 is available. */
1672 if (!u
->cgroup_path
)
1675 if (u
->cgroup_control_inotify_wd
>= 0)
1678 /* Only applies to the unified hierarchy */
1679 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1681 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1685 /* No point in watch the top-level slice, it's never going to run empty. */
1686 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1689 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
1693 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1697 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1698 if (u
->cgroup_control_inotify_wd
< 0) {
1700 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1701 * is not an error */
1704 return log_unit_error_errno(u
, errno
, "Failed to add control inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1707 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
1709 return log_unit_error_errno(u
, r
, "Failed to add control inotify watch descriptor to hash map: %m");
1714 int unit_watch_cgroup_memory(Unit
*u
) {
1715 _cleanup_free_
char *events
= NULL
;
1721 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
1722 * cgroupv2 is available. */
1724 if (!u
->cgroup_path
)
1727 c
= unit_get_cgroup_context(u
);
1731 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
1732 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
1734 if (!c
->memory_accounting
)
1737 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
1738 * we also don't want to generate a log message for each parent cgroup of a process. */
1739 if (u
->type
== UNIT_SLICE
)
1742 if (u
->cgroup_memory_inotify_wd
>= 0)
1745 /* Only applies to the unified hierarchy */
1746 r
= cg_all_unified();
1748 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
1752 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
1756 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
1760 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1761 if (u
->cgroup_memory_inotify_wd
< 0) {
1763 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1764 * is not an error */
1767 return log_unit_error_errno(u
, errno
, "Failed to add memory inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1770 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
1772 return log_unit_error_errno(u
, r
, "Failed to add memory inotify watch descriptor to hash map: %m");
1777 int unit_pick_cgroup_path(Unit
*u
) {
1778 _cleanup_free_
char *path
= NULL
;
1786 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1789 path
= unit_default_cgroup_path(u
);
1793 r
= unit_set_cgroup_path(u
, path
);
1795 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1797 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1802 static int unit_create_cgroup(
1804 CGroupMask target_mask
,
1805 CGroupMask enable_mask
,
1806 ManagerState state
) {
1813 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1816 /* Figure out our cgroup path */
1817 r
= unit_pick_cgroup_path(u
);
1821 /* First, create our own group */
1822 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
1824 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
1827 /* Start watching it */
1828 (void) unit_watch_cgroup(u
);
1829 (void) unit_watch_cgroup_memory(u
);
1831 /* Preserve enabled controllers in delegated units, adjust others. */
1832 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
1833 CGroupMask result_mask
= 0;
1835 /* Enable all controllers we need */
1836 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
1838 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
1840 /* If we just turned off a controller, this might release the controller for our parent too, let's
1841 * enqueue the parent for re-realization in that case again. */
1842 if (UNIT_ISSET(u
->slice
)) {
1843 CGroupMask turned_off
;
1845 turned_off
= (u
->cgroup_realized
? u
->cgroup_enabled_mask
& ~result_mask
: 0);
1846 if (turned_off
!= 0) {
1849 /* Force the parent to propagate the enable mask to the kernel again, by invalidating
1850 * the controller we just turned off. */
1852 for (parent
= UNIT_DEREF(u
->slice
); parent
; parent
= UNIT_DEREF(parent
->slice
))
1853 unit_invalidate_cgroup(parent
, turned_off
);
1857 /* Remember what's actually enabled now */
1858 u
->cgroup_enabled_mask
= result_mask
;
1861 /* Keep track that this is now realized */
1862 u
->cgroup_realized
= true;
1863 u
->cgroup_realized_mask
= target_mask
;
1865 if (u
->type
!= UNIT_SLICE
&& !unit_cgroup_delegate(u
)) {
1867 /* Then, possibly move things over, but not if
1868 * subgroups may contain processes, which is the case
1869 * for slice and delegation units. */
1870 r
= cg_migrate_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, u
->cgroup_path
, migrate_callback
, u
);
1872 log_unit_warning_errno(u
, r
, "Failed to migrate cgroup from to %s, ignoring: %m", u
->cgroup_path
);
1875 /* Set attributes */
1876 cgroup_context_apply(u
, target_mask
, state
);
1877 cgroup_xattr_apply(u
);
1882 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
1883 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1889 if (MANAGER_IS_SYSTEM(u
->manager
))
1892 if (!u
->manager
->system_bus
)
1895 if (!u
->cgroup_path
)
1898 /* Determine this unit's cgroup path relative to our cgroup root */
1899 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
1903 pp
= strjoina("/", pp
, suffix_path
);
1904 path_simplify(pp
, false);
1906 r
= sd_bus_call_method(u
->manager
->system_bus
,
1907 "org.freedesktop.systemd1",
1908 "/org/freedesktop/systemd1",
1909 "org.freedesktop.systemd1.Manager",
1910 "AttachProcessesToUnit",
1913 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
1915 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
1920 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
1921 CGroupMask delegated_mask
;
1929 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1932 if (set_isempty(pids
))
1935 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
1936 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
1937 r
= bpf_firewall_load_custom(u
);
1941 r
= unit_realize_cgroup(u
);
1945 if (isempty(suffix_path
))
1948 p
= prefix_roota(u
->cgroup_path
, suffix_path
);
1950 delegated_mask
= unit_get_delegate_mask(u
);
1953 SET_FOREACH(pidp
, pids
, i
) {
1954 pid_t pid
= PTR_TO_PID(pidp
);
1957 /* First, attach the PID to the main cgroup hierarchy */
1958 q
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
);
1960 log_unit_debug_errno(u
, q
, "Couldn't move process " PID_FMT
" to requested cgroup '%s': %m", pid
, p
);
1962 if (MANAGER_IS_USER(u
->manager
) && IN_SET(q
, -EPERM
, -EACCES
)) {
1965 /* If we are in a user instance, and we can't move the process ourselves due to
1966 * permission problems, let's ask the system instance about it instead. Since it's more
1967 * privileged it might be able to move the process across the leaves of a subtree who's
1968 * top node is not owned by us. */
1970 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
, suffix_path
);
1972 log_unit_debug_errno(u
, z
, "Couldn't move process " PID_FMT
" to requested cgroup '%s' via the system bus either: %m", pid
, p
);
1974 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
1978 r
= q
; /* Remember first error */
1983 q
= cg_all_unified();
1989 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
1990 * innermost realized one */
1992 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
1993 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
1994 const char *realized
;
1996 if (!(u
->manager
->cgroup_supported
& bit
))
1999 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
2000 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
2001 q
= cg_attach(cgroup_controller_to_string(c
), p
, pid
);
2003 continue; /* Success! */
2005 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",
2006 pid
, p
, cgroup_controller_to_string(c
));
2009 /* So this controller is either not delegate or realized, or something else weird happened. In
2010 * that case let's attach the PID at least to the closest cgroup up the tree that is
2012 realized
= unit_get_realized_cgroup_path(u
, bit
);
2014 continue; /* Not even realized in the root slice? Then let's not bother */
2016 q
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
);
2018 log_unit_debug_errno(u
, q
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
2019 pid
, realized
, cgroup_controller_to_string(c
));
2026 static bool unit_has_mask_realized(
2028 CGroupMask target_mask
,
2029 CGroupMask enable_mask
) {
2033 /* Returns true if this unit is fully realized. We check four things:
2035 * 1. Whether the cgroup was created at all
2036 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
2037 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
2038 * 4. Whether the invalidation mask is currently zero
2040 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
2041 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
2042 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
2043 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
2044 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
2045 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
2046 * simply don't matter. */
2048 return u
->cgroup_realized
&&
2049 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
2050 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
2051 u
->cgroup_invalidated_mask
== 0;
2054 static bool unit_has_mask_disables_realized(
2056 CGroupMask target_mask
,
2057 CGroupMask enable_mask
) {
2061 /* Returns true if all controllers which should be disabled are indeed disabled.
2063 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
2064 * already removed. */
2066 return !u
->cgroup_realized
||
2067 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
2068 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
2071 static bool unit_has_mask_enables_realized(
2073 CGroupMask target_mask
,
2074 CGroupMask enable_mask
) {
2078 /* Returns true if all controllers which should be enabled are indeed enabled.
2080 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
2081 * we want to add is already added. */
2083 return u
->cgroup_realized
&&
2084 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
2085 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
2088 void unit_add_to_cgroup_realize_queue(Unit
*u
) {
2091 if (u
->in_cgroup_realize_queue
)
2094 LIST_PREPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2095 u
->in_cgroup_realize_queue
= true;
2098 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
2101 if (!u
->in_cgroup_realize_queue
)
2104 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2105 u
->in_cgroup_realize_queue
= false;
2108 /* Controllers can only be enabled breadth-first, from the root of the
2109 * hierarchy downwards to the unit in question. */
2110 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
2111 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2116 /* First go deal with this unit's parent, or we won't be able to enable
2117 * any new controllers at this layer. */
2118 if (UNIT_ISSET(u
->slice
)) {
2119 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2124 target_mask
= unit_get_target_mask(u
);
2125 enable_mask
= unit_get_enable_mask(u
);
2127 /* We can only enable in this direction, don't try to disable anything.
2129 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
2132 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
2133 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
2135 return unit_create_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
2138 /* Controllers can only be disabled depth-first, from the leaves of the
2139 * hierarchy upwards to the unit in question. */
2140 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
2147 if (u
->type
!= UNIT_SLICE
)
2150 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
], i
) {
2151 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2154 if (UNIT_DEREF(m
->slice
) != u
)
2157 /* The cgroup for this unit might not actually be fully
2158 * realised yet, in which case it isn't holding any controllers
2160 if (!m
->cgroup_path
)
2163 /* We must disable those below us first in order to release the
2165 if (m
->type
== UNIT_SLICE
)
2166 (void) unit_realize_cgroup_now_disable(m
, state
);
2168 target_mask
= unit_get_target_mask(m
);
2169 enable_mask
= unit_get_enable_mask(m
);
2171 /* We can only disable in this direction, don't try to enable
2173 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
2176 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
2177 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
2179 r
= unit_create_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
2187 /* Check if necessary controllers and attributes for a unit are in place.
2189 * - If so, do nothing.
2190 * - If not, create paths, move processes over, and set attributes.
2192 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
2193 * a depth-first way. As such the process looks like this:
2195 * Suppose we have a cgroup hierarchy which looks like this:
2208 * 1. We want to realise cgroup "d" now.
2209 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
2210 * 3. cgroup "k" just started requesting the memory controller.
2212 * To make this work we must do the following in order:
2214 * 1. Disable CPU controller in k, j
2215 * 2. Disable CPU controller in d
2216 * 3. Enable memory controller in root
2217 * 4. Enable memory controller in a
2218 * 5. Enable memory controller in d
2219 * 6. Enable memory controller in k
2221 * Notice that we need to touch j in one direction, but not the other. We also
2222 * don't go beyond d when disabling -- it's up to "a" to get realized if it
2223 * wants to disable further. The basic rules are therefore:
2225 * - If you're disabling something, you need to realise all of the cgroups from
2226 * your recursive descendants to the root. This starts from the leaves.
2227 * - If you're enabling something, you need to realise from the root cgroup
2228 * downwards, but you don't need to iterate your recursive descendants.
2230 * Returns 0 on success and < 0 on failure. */
2231 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
2232 CGroupMask target_mask
, enable_mask
;
2237 unit_remove_from_cgroup_realize_queue(u
);
2239 target_mask
= unit_get_target_mask(u
);
2240 enable_mask
= unit_get_enable_mask(u
);
2242 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
2245 /* Disable controllers below us, if there are any */
2246 r
= unit_realize_cgroup_now_disable(u
, state
);
2250 /* Enable controllers above us, if there are any */
2251 if (UNIT_ISSET(u
->slice
)) {
2252 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2257 /* Now actually deal with the cgroup we were trying to realise and set attributes */
2258 r
= unit_create_cgroup(u
, target_mask
, enable_mask
, state
);
2262 /* Now, reset the invalidation mask */
2263 u
->cgroup_invalidated_mask
= 0;
2267 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
2275 state
= manager_state(m
);
2277 while ((i
= m
->cgroup_realize_queue
)) {
2278 assert(i
->in_cgroup_realize_queue
);
2280 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
2281 /* Maybe things changed, and the unit is not actually active anymore? */
2282 unit_remove_from_cgroup_realize_queue(i
);
2286 r
= unit_realize_cgroup_now(i
, state
);
2288 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
2296 static void unit_add_siblings_to_cgroup_realize_queue(Unit
*u
) {
2299 /* This adds the siblings of the specified unit and the
2300 * siblings of all parent units to the cgroup queue. (But
2301 * neither the specified unit itself nor the parents.) */
2303 while ((slice
= UNIT_DEREF(u
->slice
))) {
2308 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
], i
) {
2309 /* Skip units that have a dependency on the slice
2310 * but aren't actually in it. */
2311 if (UNIT_DEREF(m
->slice
) != slice
)
2314 /* No point in doing cgroup application for units
2315 * without active processes. */
2316 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
2319 /* If the unit doesn't need any new controllers
2320 * and has current ones realized, it doesn't need
2322 if (unit_has_mask_realized(m
,
2323 unit_get_target_mask(m
),
2324 unit_get_enable_mask(m
)))
2327 unit_add_to_cgroup_realize_queue(m
);
2334 int unit_realize_cgroup(Unit
*u
) {
2337 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2340 /* So, here's the deal: when realizing the cgroups for this
2341 * unit, we need to first create all parents, but there's more
2342 * actually: for the weight-based controllers we also need to
2343 * make sure that all our siblings (i.e. units that are in the
2344 * same slice as we are) have cgroups, too. Otherwise, things
2345 * would become very uneven as each of their processes would
2346 * get as much resources as all our group together. This call
2347 * will synchronously create the parent cgroups, but will
2348 * defer work on the siblings to the next event loop
2351 /* Add all sibling slices to the cgroup queue. */
2352 unit_add_siblings_to_cgroup_realize_queue(u
);
2354 /* And realize this one now (and apply the values) */
2355 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
2358 void unit_release_cgroup(Unit
*u
) {
2361 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
2362 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
2364 if (u
->cgroup_path
) {
2365 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
2366 u
->cgroup_path
= mfree(u
->cgroup_path
);
2369 if (u
->cgroup_control_inotify_wd
>= 0) {
2370 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
2371 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
);
2373 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
2374 u
->cgroup_control_inotify_wd
= -1;
2377 if (u
->cgroup_memory_inotify_wd
>= 0) {
2378 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
2379 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
);
2381 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
2382 u
->cgroup_memory_inotify_wd
= -1;
2386 void unit_prune_cgroup(Unit
*u
) {
2392 /* Removes the cgroup, if empty and possible, and stops watching it. */
2394 if (!u
->cgroup_path
)
2397 (void) unit_get_cpu_usage(u
, NULL
); /* Cache the last CPU usage value before we destroy the cgroup */
2399 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2401 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
2403 /* One reason we could have failed here is, that the cgroup still contains a process.
2404 * However, if the cgroup becomes removable at a later time, it might be removed when
2405 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
2406 * that the cgroup is still realized the next time it is started. Do not return early
2407 * on error, continue cleanup. */
2408 log_unit_full(u
, r
== -EBUSY
? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
2413 unit_release_cgroup(u
);
2415 u
->cgroup_realized
= false;
2416 u
->cgroup_realized_mask
= 0;
2417 u
->cgroup_enabled_mask
= 0;
2419 u
->bpf_device_control_installed
= bpf_program_unref(u
->bpf_device_control_installed
);
2422 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
2423 _cleanup_fclose_
FILE *f
= NULL
;
2424 pid_t pid
= 0, npid
;
2430 if (!u
->cgroup_path
)
2433 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
2437 while (cg_read_pid(f
, &npid
) > 0) {
2442 if (pid_is_my_child(npid
) == 0)
2446 /* Dang, there's more than one daemonized PID
2447 in this group, so we don't know what process
2448 is the main process. */
2459 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
2460 _cleanup_closedir_
DIR *d
= NULL
;
2461 _cleanup_fclose_
FILE *f
= NULL
;
2467 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
2473 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
2474 r
= unit_watch_pid(u
, pid
, false);
2475 if (r
< 0 && ret
>= 0)
2479 if (r
< 0 && ret
>= 0)
2483 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
2490 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
2491 _cleanup_free_
char *p
= NULL
;
2493 p
= path_join(empty_to_root(path
), fn
);
2499 r
= unit_watch_pids_in_path(u
, p
);
2500 if (r
< 0 && ret
>= 0)
2504 if (r
< 0 && ret
>= 0)
2511 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
2516 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
2517 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
2518 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
2520 if (!u
->cgroup_path
)
2523 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2526 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
2529 if (!set_isempty(u
->pids
))
2532 unit_add_to_cgroup_empty_queue(u
);
2536 int unit_watch_all_pids(Unit
*u
) {
2541 /* Adds all PIDs from our cgroup to the set of PIDs we
2542 * watch. This is a fallback logic for cases where we do not
2543 * get reliable cgroup empty notifications: we try to use
2544 * SIGCHLD as replacement. */
2546 if (!u
->cgroup_path
)
2549 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2552 if (r
> 0) /* On unified we can use proper notifications */
2555 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
2558 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
2559 Manager
*m
= userdata
;
2566 u
= m
->cgroup_empty_queue
;
2570 assert(u
->in_cgroup_empty_queue
);
2571 u
->in_cgroup_empty_queue
= false;
2572 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
2574 if (m
->cgroup_empty_queue
) {
2575 /* More stuff queued, let's make sure we remain enabled */
2576 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2578 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
2581 unit_add_to_gc_queue(u
);
2583 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
2584 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
2589 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
2594 /* Note that there are four different ways how cgroup empty events reach us:
2596 * 1. On the unified hierarchy we get an inotify event on the cgroup
2598 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2600 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2602 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2603 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2605 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2606 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2607 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2608 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2609 * case for scope units). */
2611 if (u
->in_cgroup_empty_queue
)
2614 /* Let's verify that the cgroup is really empty */
2615 if (!u
->cgroup_path
)
2617 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2619 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", u
->cgroup_path
);
2625 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2626 u
->in_cgroup_empty_queue
= true;
2628 /* Trigger the defer event */
2629 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
2631 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
2634 int unit_check_oom(Unit
*u
) {
2635 _cleanup_free_
char *oom_kill
= NULL
;
2640 if (!u
->cgroup_path
)
2643 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
2645 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
2647 r
= safe_atou64(oom_kill
, &c
);
2649 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
2651 increased
= c
> u
->oom_kill_last
;
2652 u
->oom_kill_last
= c
;
2657 log_struct(LOG_NOTICE
,
2658 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
2660 LOG_UNIT_INVOCATION_ID(u
),
2661 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
2663 if (UNIT_VTABLE(u
)->notify_cgroup_oom
)
2664 UNIT_VTABLE(u
)->notify_cgroup_oom(u
);
2669 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
2670 Manager
*m
= userdata
;
2677 u
= m
->cgroup_oom_queue
;
2681 assert(u
->in_cgroup_oom_queue
);
2682 u
->in_cgroup_oom_queue
= false;
2683 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
2685 if (m
->cgroup_oom_queue
) {
2686 /* More stuff queued, let's make sure we remain enabled */
2687 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2689 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
2692 (void) unit_check_oom(u
);
2696 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
2701 if (u
->in_cgroup_oom_queue
)
2703 if (!u
->cgroup_path
)
2706 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
2707 u
->in_cgroup_oom_queue
= true;
2709 /* Trigger the defer event */
2710 if (!u
->manager
->cgroup_oom_event_source
) {
2711 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
2713 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
2715 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
2719 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
2721 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
2725 (void) sd_event_source_set_description(s
, "cgroup-oom");
2726 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
2729 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
2731 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
2734 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
2735 Manager
*m
= userdata
;
2742 union inotify_event_buffer buffer
;
2743 struct inotify_event
*e
;
2746 l
= read(fd
, &buffer
, sizeof(buffer
));
2748 if (IN_SET(errno
, EINTR
, EAGAIN
))
2751 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
2754 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
2758 /* Queue overflow has no watch descriptor */
2761 if (e
->mask
& IN_IGNORED
)
2762 /* The watch was just removed */
2765 /* Note that inotify might deliver events for a watch even after it was removed,
2766 * because it was queued before the removal. Let's ignore this here safely. */
2768 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2770 unit_add_to_cgroup_empty_queue(u
);
2772 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2774 unit_add_to_cgroup_oom_queue(u
);
2779 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
2780 CGroupMask mask
= 0;
2783 /* BPF-based firewall */
2784 r
= bpf_firewall_supported();
2786 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2788 /* BPF-based device access control */
2789 r
= bpf_devices_supported();
2791 mask
|= CGROUP_MASK_BPF_DEVICES
;
2797 int manager_setup_cgroup(Manager
*m
) {
2798 _cleanup_free_
char *path
= NULL
;
2799 const char *scope_path
;
2807 /* 1. Determine hierarchy */
2808 m
->cgroup_root
= mfree(m
->cgroup_root
);
2809 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
2811 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
2813 /* Chop off the init scope, if we are already located in it */
2814 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2816 /* LEGACY: Also chop off the system slice if we are in
2817 * it. This is to support live upgrades from older systemd
2818 * versions where PID 1 was moved there. Also see
2819 * cg_get_root_path(). */
2820 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
2821 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
2823 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
2828 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
2829 * easily prepend it everywhere. */
2830 delete_trailing_chars(m
->cgroup_root
, "/");
2833 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
2835 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
2837 r
= cg_unified_flush();
2839 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
2841 all_unified
= cg_all_unified();
2842 if (all_unified
< 0)
2843 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
2844 if (all_unified
> 0)
2845 log_debug("Unified cgroup hierarchy is located at %s.", path
);
2847 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2849 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
2851 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
2853 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
2856 /* 3. Allocate cgroup empty defer event source */
2857 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
2858 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
2860 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
2862 /* Schedule cgroup empty checks early, but after having processed service notification messages or
2863 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
2864 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
2865 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
2867 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
2869 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
2871 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
2873 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
2875 /* 4. Install notifier inotify object, or agent */
2876 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
2878 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
2880 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
2881 safe_close(m
->cgroup_inotify_fd
);
2883 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
2884 if (m
->cgroup_inotify_fd
< 0)
2885 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
2887 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
2889 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
2891 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
2892 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
2893 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
2894 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
2896 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
2898 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
2900 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
2902 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
2903 * since it does not generate events when control groups with children run empty. */
2905 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUP_AGENT_PATH
);
2907 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
2909 log_debug("Installed release agent.");
2911 log_debug("Release agent already installed.");
2914 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
2915 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2916 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
2918 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
2919 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
2921 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
2923 /* 6. And pin it, so that it cannot be unmounted */
2924 safe_close(m
->pin_cgroupfs_fd
);
2925 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
2926 if (m
->pin_cgroupfs_fd
< 0)
2927 return log_error_errno(errno
, "Failed to open pin file: %m");
2929 } else if (!MANAGER_IS_TEST_RUN(m
))
2930 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
2932 /* 7. Always enable hierarchical support if it exists... */
2933 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
2934 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
2936 /* 8. Figure out which controllers are supported */
2937 r
= cg_mask_supported(&m
->cgroup_supported
);
2939 return log_error_errno(r
, "Failed to determine supported controllers: %m");
2941 /* 9. Figure out which bpf-based pseudo-controllers are supported */
2942 r
= cg_bpf_mask_supported(&mask
);
2944 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
2945 m
->cgroup_supported
|= mask
;
2947 /* 10. Log which controllers are supported */
2948 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
2949 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
2954 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
2957 /* We can't really delete the group, since we are in it. But
2959 if (delete && m
->cgroup_root
&& m
->test_run_flags
!= MANAGER_TEST_RUN_MINIMAL
)
2960 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
2962 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
2964 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
2965 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
2967 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
2968 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
2970 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
2972 m
->cgroup_root
= mfree(m
->cgroup_root
);
2975 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
2982 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
2986 p
= strdupa(cgroup
);
2990 e
= strrchr(p
, '/');
2992 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
2996 u
= hashmap_get(m
->cgroup_unit
, p
);
3002 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
3003 _cleanup_free_
char *cgroup
= NULL
;
3007 if (!pid_is_valid(pid
))
3010 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
3013 return manager_get_unit_by_cgroup(m
, cgroup
);
3016 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
3021 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
3022 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
3023 * relevant one as children of the process will be assigned to that one, too, before all else. */
3025 if (!pid_is_valid(pid
))
3028 if (pid
== getpid_cached())
3029 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
3031 u
= manager_get_unit_by_pid_cgroup(m
, pid
);
3035 u
= hashmap_get(m
->watch_pids
, PID_TO_PTR(pid
));
3039 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-pid
));
3046 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
3052 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
3053 * or from the --system instance */
3055 log_debug("Got cgroup empty notification for: %s", cgroup
);
3057 u
= manager_get_unit_by_cgroup(m
, cgroup
);
3061 unit_add_to_cgroup_empty_queue(u
);
3065 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
3066 _cleanup_free_
char *v
= NULL
;
3072 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3075 if (!u
->cgroup_path
)
3078 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3079 if (unit_has_host_root_cgroup(u
))
3080 return procfs_memory_get_used(ret
);
3082 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3085 r
= cg_all_unified();
3089 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.current", &v
);
3091 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.usage_in_bytes", &v
);
3097 return safe_atou64(v
, ret
);
3100 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
3101 _cleanup_free_
char *v
= NULL
;
3107 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
3110 if (!u
->cgroup_path
)
3113 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3114 if (unit_has_host_root_cgroup(u
))
3115 return procfs_tasks_get_current(ret
);
3117 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
3120 r
= cg_get_attribute("pids", u
->cgroup_path
, "pids.current", &v
);
3126 return safe_atou64(v
, ret
);
3129 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
3130 _cleanup_free_
char *v
= NULL
;
3137 if (!u
->cgroup_path
)
3140 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3141 if (unit_has_host_root_cgroup(u
))
3142 return procfs_cpu_get_usage(ret
);
3144 /* Requisite controllers for CPU accounting are not enabled */
3145 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
3148 r
= cg_all_unified();
3152 _cleanup_free_
char *val
= NULL
;
3155 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
3156 if (IN_SET(r
, -ENOENT
, -ENXIO
))
3161 r
= safe_atou64(val
, &us
);
3165 ns
= us
* NSEC_PER_USEC
;
3167 r
= cg_get_attribute("cpuacct", u
->cgroup_path
, "cpuacct.usage", &v
);
3173 r
= safe_atou64(v
, &ns
);
3182 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
3188 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
3189 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
3190 * call this function with a NULL return value. */
3192 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
3195 r
= unit_get_cpu_usage_raw(u
, &ns
);
3196 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
3197 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
3201 *ret
= u
->cpu_usage_last
;
3207 if (ns
> u
->cpu_usage_base
)
3208 ns
-= u
->cpu_usage_base
;
3212 u
->cpu_usage_last
= ns
;
3219 int unit_get_ip_accounting(
3221 CGroupIPAccountingMetric metric
,
3228 assert(metric
>= 0);
3229 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
3232 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
3235 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
3236 u
->ip_accounting_ingress_map_fd
:
3237 u
->ip_accounting_egress_map_fd
;
3241 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
3242 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
3244 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
3248 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
3249 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
3250 * ip_accounting_extra[] field, and add them in here transparently. */
3252 *ret
= value
+ u
->ip_accounting_extra
[metric
];
3257 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
3258 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
3259 [CGROUP_IO_READ_BYTES
] = "rbytes=",
3260 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
3261 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
3262 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
3264 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
3265 _cleanup_free_
char *path
= NULL
;
3266 _cleanup_fclose_
FILE *f
= NULL
;
3271 if (!u
->cgroup_path
)
3274 if (unit_has_host_root_cgroup(u
))
3275 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
3277 r
= cg_all_unified();
3280 if (r
== 0) /* TODO: support cgroupv1 */
3283 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
3286 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
3290 f
= fopen(path
, "re");
3295 _cleanup_free_
char *line
= NULL
;
3298 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3305 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
3306 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
3309 _cleanup_free_
char *word
= NULL
;
3311 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
3317 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3320 x
= startswith(word
, field_names
[i
]);
3324 r
= safe_atou64(x
, &w
);
3328 /* Sum up the stats of all devices */
3336 memcpy(ret
, acc
, sizeof(acc
));
3340 int unit_get_io_accounting(
3342 CGroupIOAccountingMetric metric
,
3346 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
3349 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
3351 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
3354 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3357 r
= unit_get_io_accounting_raw(u
, raw
);
3358 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3363 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3364 /* Saturated subtraction */
3365 if (raw
[i
] > u
->io_accounting_base
[i
])
3366 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
3368 u
->io_accounting_last
[i
] = 0;
3373 *ret
= u
->io_accounting_last
[metric
];
3378 int unit_reset_cpu_accounting(Unit
*u
) {
3383 u
->cpu_usage_last
= NSEC_INFINITY
;
3385 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
3387 u
->cpu_usage_base
= 0;
3394 int unit_reset_ip_accounting(Unit
*u
) {
3399 if (u
->ip_accounting_ingress_map_fd
>= 0)
3400 r
= bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
);
3402 if (u
->ip_accounting_egress_map_fd
>= 0)
3403 q
= bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
);
3405 zero(u
->ip_accounting_extra
);
3407 return r
< 0 ? r
: q
;
3410 int unit_reset_io_accounting(Unit
*u
) {
3415 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++)
3416 u
->io_accounting_last
[i
] = UINT64_MAX
;
3418 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
3420 zero(u
->io_accounting_base
);
3427 int unit_reset_accounting(Unit
*u
) {
3432 r
= unit_reset_cpu_accounting(u
);
3433 q
= unit_reset_io_accounting(u
);
3434 v
= unit_reset_ip_accounting(u
);
3436 return r
< 0 ? r
: q
< 0 ? q
: v
;
3439 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
3442 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3448 /* always invalidate compat pairs together */
3449 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
3450 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
3452 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
3453 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
3455 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
3458 u
->cgroup_invalidated_mask
|= m
;
3459 unit_add_to_cgroup_realize_queue(u
);
3462 void unit_invalidate_cgroup_bpf(Unit
*u
) {
3465 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3468 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
3471 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
3472 unit_add_to_cgroup_realize_queue(u
);
3474 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
3475 * list of our children includes our own. */
3476 if (u
->type
== UNIT_SLICE
) {
3481 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
], i
) {
3482 if (UNIT_DEREF(member
->slice
) == u
)
3483 unit_invalidate_cgroup_bpf(member
);
3488 bool unit_cgroup_delegate(Unit
*u
) {
3493 if (!UNIT_VTABLE(u
)->can_delegate
)
3496 c
= unit_get_cgroup_context(u
);
3503 void manager_invalidate_startup_units(Manager
*m
) {
3509 SET_FOREACH(u
, m
->startup_units
, i
)
3510 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
3513 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
3514 [CGROUP_AUTO
] = "auto",
3515 [CGROUP_CLOSED
] = "closed",
3516 [CGROUP_STRICT
] = "strict",
3519 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);