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
);
204 void cgroup_context_dump(CGroupContext
*c
, FILE* f
, const char *prefix
) {
205 _cleanup_free_
char *disable_controllers_str
= NULL
;
206 CGroupIODeviceLimit
*il
;
207 CGroupIODeviceWeight
*iw
;
208 CGroupIODeviceLatency
*l
;
209 CGroupBlockIODeviceBandwidth
*b
;
210 CGroupBlockIODeviceWeight
*w
;
211 CGroupDeviceAllow
*a
;
212 IPAddressAccessItem
*iaai
;
213 char u
[FORMAT_TIMESPAN_MAX
];
214 char v
[FORMAT_TIMESPAN_MAX
];
219 prefix
= strempty(prefix
);
221 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
224 "%sCPUAccounting=%s\n"
225 "%sIOAccounting=%s\n"
226 "%sBlockIOAccounting=%s\n"
227 "%sMemoryAccounting=%s\n"
228 "%sTasksAccounting=%s\n"
229 "%sIPAccounting=%s\n"
230 "%sCPUWeight=%" PRIu64
"\n"
231 "%sStartupCPUWeight=%" PRIu64
"\n"
232 "%sCPUShares=%" PRIu64
"\n"
233 "%sStartupCPUShares=%" PRIu64
"\n"
234 "%sCPUQuotaPerSecSec=%s\n"
235 "%sCPUQuotaPeriodSec=%s\n"
236 "%sIOWeight=%" PRIu64
"\n"
237 "%sStartupIOWeight=%" PRIu64
"\n"
238 "%sBlockIOWeight=%" PRIu64
"\n"
239 "%sStartupBlockIOWeight=%" PRIu64
"\n"
240 "%sDefaultMemoryMin=%" PRIu64
"\n"
241 "%sDefaultMemoryLow=%" PRIu64
"\n"
242 "%sMemoryMin=%" PRIu64
"\n"
243 "%sMemoryLow=%" PRIu64
"\n"
244 "%sMemoryHigh=%" PRIu64
"\n"
245 "%sMemoryMax=%" PRIu64
"\n"
246 "%sMemorySwapMax=%" PRIu64
"\n"
247 "%sMemoryLimit=%" PRIu64
"\n"
248 "%sTasksMax=%" PRIu64
"\n"
249 "%sDevicePolicy=%s\n"
250 "%sDisableControllers=%s\n"
252 prefix
, yes_no(c
->cpu_accounting
),
253 prefix
, yes_no(c
->io_accounting
),
254 prefix
, yes_no(c
->blockio_accounting
),
255 prefix
, yes_no(c
->memory_accounting
),
256 prefix
, yes_no(c
->tasks_accounting
),
257 prefix
, yes_no(c
->ip_accounting
),
258 prefix
, c
->cpu_weight
,
259 prefix
, c
->startup_cpu_weight
,
260 prefix
, c
->cpu_shares
,
261 prefix
, c
->startup_cpu_shares
,
262 prefix
, format_timespan(u
, sizeof(u
), c
->cpu_quota_per_sec_usec
, 1),
263 prefix
, format_timespan(v
, sizeof(v
), c
->cpu_quota_period_usec
, 1),
264 prefix
, c
->io_weight
,
265 prefix
, c
->startup_io_weight
,
266 prefix
, c
->blockio_weight
,
267 prefix
, c
->startup_blockio_weight
,
268 prefix
, c
->default_memory_min
,
269 prefix
, c
->default_memory_low
,
270 prefix
, c
->memory_min
,
271 prefix
, c
->memory_low
,
272 prefix
, c
->memory_high
,
273 prefix
, c
->memory_max
,
274 prefix
, c
->memory_swap_max
,
275 prefix
, c
->memory_limit
,
276 prefix
, c
->tasks_max
,
277 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
278 prefix
, strnull(disable_controllers_str
),
279 prefix
, yes_no(c
->delegate
));
282 _cleanup_free_
char *t
= NULL
;
284 (void) cg_mask_to_string(c
->delegate_controllers
, &t
);
286 fprintf(f
, "%sDelegateControllers=%s\n",
291 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
293 "%sDeviceAllow=%s %s%s%s\n",
296 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
298 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
300 "%sIODeviceWeight=%s %" PRIu64
"\n",
305 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
307 "%sIODeviceLatencyTargetSec=%s %s\n",
310 format_timespan(u
, sizeof(u
), l
->target_usec
, 1));
312 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
313 char buf
[FORMAT_BYTES_MAX
];
314 CGroupIOLimitType type
;
316 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
317 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
321 cgroup_io_limit_type_to_string(type
),
323 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
326 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
328 "%sBlockIODeviceWeight=%s %" PRIu64
,
333 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
334 char buf
[FORMAT_BYTES_MAX
];
336 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
338 "%sBlockIOReadBandwidth=%s %s\n",
341 format_bytes(buf
, sizeof(buf
), b
->rbps
));
342 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
344 "%sBlockIOWriteBandwidth=%s %s\n",
347 format_bytes(buf
, sizeof(buf
), b
->wbps
));
350 LIST_FOREACH(items
, iaai
, c
->ip_address_allow
) {
351 _cleanup_free_
char *k
= NULL
;
353 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
354 fprintf(f
, "%sIPAddressAllow=%s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
357 LIST_FOREACH(items
, iaai
, c
->ip_address_deny
) {
358 _cleanup_free_
char *k
= NULL
;
360 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
361 fprintf(f
, "%sIPAddressDeny=%s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
365 int cgroup_add_device_allow(CGroupContext
*c
, const char *dev
, const char *mode
) {
366 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
367 _cleanup_free_
char *d
= NULL
;
371 assert(isempty(mode
) || in_charset(mode
, "rwm"));
373 a
= new(CGroupDeviceAllow
, 1);
381 *a
= (CGroupDeviceAllow
) {
383 .r
= isempty(mode
) || strchr(mode
, 'r'),
384 .w
= isempty(mode
) || strchr(mode
, 'w'),
385 .m
= isempty(mode
) || strchr(mode
, 'm'),
388 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
394 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
395 uint64_t unit_get_ancestor_##entry(Unit *u) { \
398 /* 1. Is entry set in this unit? If so, use that. \
399 * 2. Is the default for this entry set in any \
400 * ancestor? If so, use that. \
401 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
405 c = unit_get_cgroup_context(u); \
406 if (c && c->entry##_set) \
409 while ((u = UNIT_DEREF(u->slice))) { \
410 c = unit_get_cgroup_context(u); \
411 if (c && c->default_##entry##_set) \
412 return c->default_##entry; \
415 /* We've reached the root, but nobody had default for \
416 * this entry set, so set it to the kernel default. */ \
417 return CGROUP_LIMIT_MIN; \
420 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
421 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
423 static void cgroup_xattr_apply(Unit
*u
) {
424 char ids
[SD_ID128_STRING_MAX
];
429 if (!MANAGER_IS_SYSTEM(u
->manager
))
432 if (sd_id128_is_null(u
->invocation_id
))
435 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
436 "trusted.invocation_id",
437 sd_id128_to_string(u
->invocation_id
, ids
), 32,
440 log_unit_debug_errno(u
, r
, "Failed to set invocation ID on control group %s, ignoring: %m", u
->cgroup_path
);
443 static int lookup_block_device(const char *p
, dev_t
*ret
) {
451 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
452 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
454 if (stat(p
, &st
) < 0)
455 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
456 rdev
= (dev_t
)st
.st_rdev
;
457 dev
= (dev_t
)st
.st_dev
;
460 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
463 log_warning("Device node '%s' is a character device, but block device needed.", p
);
465 } else if (S_ISBLK(mode
))
467 else if (major(dev
) != 0)
468 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
470 /* If this is btrfs, getting the backing block device is a bit harder */
471 r
= btrfs_get_block_device(p
, ret
);
472 if (r
< 0 && r
!= -ENOTTY
)
473 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
475 log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
480 /* If this is a LUKS device, try to get the originating block device */
481 (void) block_get_originating(*ret
, ret
);
483 /* If this is a partition, try to get the originating block device */
484 (void) block_get_whole_disk(*ret
, ret
);
488 static int whitelist_device(BPFProgram
*prog
, const char *path
, const char *node
, const char *acc
) {
496 /* Some special handling for /dev/block/%u:%u, /dev/char/%u:%u, /run/systemd/inaccessible/chr and
497 * /run/systemd/inaccessible/blk paths. Instead of stat()ing these we parse out the major/minor directly. This
498 * means clients can use these path without the device node actually around */
499 r
= device_path_parse_major_minor(node
, &mode
, &rdev
);
502 return log_warning_errno(r
, "Couldn't parse major/minor from device path '%s': %m", node
);
505 if (stat(node
, &st
) < 0)
506 return log_warning_errno(errno
, "Couldn't stat device %s: %m", node
);
508 if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
)) {
509 log_warning("%s is not a device.", node
);
512 rdev
= (dev_t
) st
.st_rdev
;
516 if (cg_all_unified() > 0) {
520 return cgroup_bpf_whitelist_device(prog
, S_ISCHR(mode
) ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
521 major(rdev
), minor(rdev
), acc
);
524 char buf
[2+DECIMAL_STR_MAX(dev_t
)*2+2+4];
528 S_ISCHR(mode
) ? 'c' : 'b',
529 major(rdev
), minor(rdev
),
532 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */
534 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
536 return log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
537 r
, "Failed to set devices.allow on %s: %m", path
);
543 static int whitelist_major(BPFProgram
*prog
, const char *path
, const char *name
, char type
, const char *acc
) {
544 _cleanup_fclose_
FILE *f
= NULL
;
545 char buf
[2+DECIMAL_STR_MAX(unsigned)+3+4];
552 assert(IN_SET(type
, 'b', 'c'));
554 if (streq(name
, "*")) {
555 /* If the name is a wildcard, then apply this list to all devices of this type */
557 if (cg_all_unified() > 0) {
561 (void) cgroup_bpf_whitelist_class(prog
, type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
, acc
);
563 xsprintf(buf
, "%c *:* %s", type
, acc
);
565 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
567 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
568 "Failed to set devices.allow on %s: %m", path
);
573 if (safe_atou(name
, &maj
) >= 0 && DEVICE_MAJOR_VALID(maj
)) {
574 /* The name is numeric and suitable as major. In that case, let's take is major, and create the entry
577 if (cg_all_unified() > 0) {
581 (void) cgroup_bpf_whitelist_major(prog
,
582 type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
585 xsprintf(buf
, "%c %u:* %s", type
, maj
, acc
);
587 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
589 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
590 "Failed to set devices.allow on %s: %m", path
);
596 f
= fopen("/proc/devices", "re");
598 return log_warning_errno(errno
, "Cannot open /proc/devices to resolve %s (%c): %m", name
, type
);
601 _cleanup_free_
char *line
= NULL
;
604 r
= read_line(f
, LONG_LINE_MAX
, &line
);
606 return log_warning_errno(r
, "Failed to read /proc/devices: %m");
610 if (type
== 'c' && streq(line
, "Character devices:")) {
615 if (type
== 'b' && streq(line
, "Block devices:")) {
630 w
= strpbrk(p
, WHITESPACE
);
635 r
= safe_atou(p
, &maj
);
642 w
+= strspn(w
, WHITESPACE
);
644 if (fnmatch(name
, w
, 0) != 0)
647 if (cg_all_unified() > 0) {
651 (void) cgroup_bpf_whitelist_major(prog
,
652 type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
661 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL
664 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
666 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
667 r
, "Failed to set devices.allow on %s: %m", path
);
674 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
675 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
676 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
679 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
680 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
681 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
684 static uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
685 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
686 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
687 return c
->startup_cpu_weight
;
688 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
689 return c
->cpu_weight
;
691 return CGROUP_WEIGHT_DEFAULT
;
694 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
695 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
696 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
697 return c
->startup_cpu_shares
;
698 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
699 return c
->cpu_shares
;
701 return CGROUP_CPU_SHARES_DEFAULT
;
704 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
705 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
706 * need to be higher than that boundary. quota is specified in USecPerSec.
707 * Additionally, period must be at most max_period. */
710 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
713 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
716 if (quota
== USEC_INFINITY
)
717 /* Always use default period for infinity quota. */
718 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
720 if (period
== USEC_INFINITY
)
721 /* Default period was requested. */
722 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
724 /* Clamp to interval [1ms, 1s] */
725 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
727 if (new_period
!= period
) {
728 char v
[FORMAT_TIMESPAN_MAX
];
729 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
, 0,
730 "Clamping CPU interval for cpu.max: period is now %s",
731 format_timespan(v
, sizeof(v
), new_period
, 1));
732 u
->warned_clamping_cpu_quota_period
= true;
738 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
739 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
741 xsprintf(buf
, "%" PRIu64
"\n", weight
);
742 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
745 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
746 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
748 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
749 if (quota
!= USEC_INFINITY
)
750 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
751 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
753 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
754 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
757 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
758 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
760 xsprintf(buf
, "%" PRIu64
"\n", shares
);
761 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
764 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
765 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
767 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
769 xsprintf(buf
, USEC_FMT
"\n", period
);
770 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
772 if (quota
!= USEC_INFINITY
) {
773 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
774 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
776 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
779 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
780 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
781 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
784 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
785 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
786 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
789 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
790 return c
->io_accounting
||
791 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
792 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
793 c
->io_device_weights
||
794 c
->io_device_latencies
||
798 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
799 return c
->blockio_accounting
||
800 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
801 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
802 c
->blockio_device_weights
||
803 c
->blockio_device_bandwidths
;
806 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
807 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
808 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
809 return c
->startup_io_weight
;
810 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
813 return CGROUP_WEIGHT_DEFAULT
;
816 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
817 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
818 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
819 return c
->startup_blockio_weight
;
820 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
821 return c
->blockio_weight
;
823 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
826 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
827 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
828 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
831 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
832 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
833 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
836 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
837 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
841 r
= lookup_block_device(dev_path
, &dev
);
845 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
846 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
849 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
850 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
854 r
= lookup_block_device(dev_path
, &dev
);
858 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
859 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
862 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
863 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
867 r
= lookup_block_device(dev_path
, &dev
);
871 if (target
!= USEC_INFINITY
)
872 xsprintf(buf
, "%u:%u target=%" PRIu64
"\n", major(dev
), minor(dev
), target
);
874 xsprintf(buf
, "%u:%u target=max\n", major(dev
), minor(dev
));
876 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
879 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
880 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
881 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
882 CGroupIOLimitType type
;
886 r
= lookup_block_device(dev_path
, &dev
);
890 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
891 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
892 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
894 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
896 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
897 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
898 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
899 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
902 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
903 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
907 r
= lookup_block_device(dev_path
, &dev
);
911 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
912 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
914 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
915 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
918 static bool unit_has_unified_memory_config(Unit
*u
) {
923 c
= unit_get_cgroup_context(u
);
926 return c
->memory_min
> 0 || unit_get_ancestor_memory_low(u
) > 0 ||
927 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
||
928 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
;
931 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
932 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
934 if (v
!= CGROUP_LIMIT_MAX
)
935 xsprintf(buf
, "%" PRIu64
"\n", v
);
937 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
940 static void cgroup_apply_firewall(Unit
*u
) {
943 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
945 if (bpf_firewall_compile(u
) < 0)
948 (void) bpf_firewall_install(u
);
951 static void cgroup_context_apply(
953 CGroupMask apply_mask
,
954 ManagerState state
) {
958 bool is_host_root
, is_local_root
;
963 /* Nothing to do? Exit early! */
967 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
968 * attributes should only be managed for cgroups further down the tree. */
969 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
970 is_host_root
= unit_has_host_root_cgroup(u
);
972 assert_se(c
= unit_get_cgroup_context(u
));
973 assert_se(path
= u
->cgroup_path
);
975 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
978 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
979 * then), and missing cgroups, i.e. EROFS and ENOENT. */
981 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
982 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
983 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
984 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
985 * we couldn't even write to them if we wanted to). */
986 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
988 if (cg_all_unified() > 0) {
991 if (cgroup_context_has_cpu_weight(c
))
992 weight
= cgroup_context_cpu_weight(c
, state
);
993 else if (cgroup_context_has_cpu_shares(c
)) {
996 shares
= cgroup_context_cpu_shares(c
, state
);
997 weight
= cgroup_cpu_shares_to_weight(shares
);
999 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
1000 shares
, weight
, path
);
1002 weight
= CGROUP_WEIGHT_DEFAULT
;
1004 cgroup_apply_unified_cpu_weight(u
, weight
);
1005 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1010 if (cgroup_context_has_cpu_weight(c
)) {
1013 weight
= cgroup_context_cpu_weight(c
, state
);
1014 shares
= cgroup_cpu_weight_to_shares(weight
);
1016 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
1017 weight
, shares
, path
);
1018 } else if (cgroup_context_has_cpu_shares(c
))
1019 shares
= cgroup_context_cpu_shares(c
, state
);
1021 shares
= CGROUP_CPU_SHARES_DEFAULT
;
1023 cgroup_apply_legacy_cpu_shares(u
, shares
);
1024 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1028 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
1029 * controller), and in case of containers we want to leave control of these attributes to the container manager
1030 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
1031 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
1032 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
1033 bool has_io
, has_blockio
;
1036 has_io
= cgroup_context_has_io_config(c
);
1037 has_blockio
= cgroup_context_has_blockio_config(c
);
1040 weight
= cgroup_context_io_weight(c
, state
);
1041 else if (has_blockio
) {
1042 uint64_t blkio_weight
;
1044 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
1045 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
1047 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
1048 blkio_weight
, weight
);
1050 weight
= CGROUP_WEIGHT_DEFAULT
;
1052 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1053 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1056 CGroupIODeviceLatency
*latency
;
1057 CGroupIODeviceLimit
*limit
;
1058 CGroupIODeviceWeight
*w
;
1060 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
1061 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
1063 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
1064 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
1066 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
1067 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
1069 } else if (has_blockio
) {
1070 CGroupBlockIODeviceWeight
*w
;
1071 CGroupBlockIODeviceBandwidth
*b
;
1073 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
1074 weight
= cgroup_weight_blkio_to_io(w
->weight
);
1076 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
1077 w
->weight
, weight
, w
->path
);
1079 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
1082 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
1083 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
1084 CGroupIOLimitType type
;
1086 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1087 limits
[type
] = cgroup_io_limit_defaults
[type
];
1089 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
1090 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
1092 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
1093 b
->rbps
, b
->wbps
, b
->path
);
1095 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
1100 if (apply_mask
& CGROUP_MASK_BLKIO
) {
1101 bool has_io
, has_blockio
;
1103 has_io
= cgroup_context_has_io_config(c
);
1104 has_blockio
= cgroup_context_has_blockio_config(c
);
1106 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
1107 * left to our container manager, too. */
1108 if (!is_local_root
) {
1109 char buf
[DECIMAL_STR_MAX(uint64_t)+1];
1115 io_weight
= cgroup_context_io_weight(c
, state
);
1116 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
1118 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
1120 } else if (has_blockio
)
1121 weight
= cgroup_context_blkio_weight(c
, state
);
1123 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
1125 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1126 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1129 CGroupIODeviceWeight
*w
;
1131 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
1132 weight
= cgroup_weight_io_to_blkio(w
->weight
);
1134 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
1135 w
->weight
, weight
, w
->path
);
1137 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
1139 } else if (has_blockio
) {
1140 CGroupBlockIODeviceWeight
*w
;
1142 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
1143 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
1147 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
1148 * roots, as there we want the container manager to handle it */
1149 if (is_host_root
|| !is_local_root
) {
1151 CGroupIODeviceLimit
*l
;
1153 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
1154 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
1155 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
1157 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
1159 } else if (has_blockio
) {
1160 CGroupBlockIODeviceBandwidth
*b
;
1162 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
1163 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
1168 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
1169 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
1170 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
1171 * write to this if we wanted to.) */
1172 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
1174 if (cg_all_unified() > 0) {
1175 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
;
1177 if (unit_has_unified_memory_config(u
)) {
1178 max
= c
->memory_max
;
1179 swap_max
= c
->memory_swap_max
;
1181 max
= c
->memory_limit
;
1183 if (max
!= CGROUP_LIMIT_MAX
)
1184 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
1187 cgroup_apply_unified_memory_limit(u
, "memory.min", c
->memory_min
);
1188 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
1189 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
1190 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
1191 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
1193 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
1196 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1199 if (unit_has_unified_memory_config(u
)) {
1200 val
= c
->memory_max
;
1201 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIi64
" as MemoryLimit=", val
);
1203 val
= c
->memory_limit
;
1205 if (val
== CGROUP_LIMIT_MAX
)
1206 strncpy(buf
, "-1\n", sizeof(buf
));
1208 xsprintf(buf
, "%" PRIu64
"\n", val
);
1210 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
1214 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
1215 * containers, where we leave this to the manager */
1216 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
1217 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
)) {
1218 _cleanup_(bpf_program_unrefp
) BPFProgram
*prog
= NULL
;
1219 CGroupDeviceAllow
*a
;
1221 if (cg_all_unified() > 0) {
1222 r
= cgroup_init_device_bpf(&prog
, c
->device_policy
, c
->device_allow
);
1224 log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
1226 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL
1229 if (c
->device_allow
|| c
->device_policy
!= CGROUP_AUTO
)
1230 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
1232 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
1234 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1235 "Failed to reset devices.allow/devices.deny: %m");
1238 if (c
->device_policy
== CGROUP_CLOSED
||
1239 (c
->device_policy
== CGROUP_AUTO
&& c
->device_allow
)) {
1240 static const char auto_devices
[] =
1241 "/dev/null\0" "rwm\0"
1242 "/dev/zero\0" "rwm\0"
1243 "/dev/full\0" "rwm\0"
1244 "/dev/random\0" "rwm\0"
1245 "/dev/urandom\0" "rwm\0"
1246 "/dev/tty\0" "rwm\0"
1247 "/dev/ptmx\0" "rwm\0"
1248 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
1249 "/run/systemd/inaccessible/chr\0" "rwm\0"
1250 "/run/systemd/inaccessible/blk\0" "rwm\0";
1254 NULSTR_FOREACH_PAIR(x
, y
, auto_devices
)
1255 (void) whitelist_device(prog
, path
, x
, y
);
1257 /* PTS (/dev/pts) devices may not be duplicated, but accessed */
1258 (void) whitelist_major(prog
, path
, "pts", 'c', "rw");
1261 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1277 if (path_startswith(a
->path
, "/dev/"))
1278 (void) whitelist_device(prog
, path
, a
->path
, acc
);
1279 else if ((val
= startswith(a
->path
, "block-")))
1280 (void) whitelist_major(prog
, path
, val
, 'b', acc
);
1281 else if ((val
= startswith(a
->path
, "char-")))
1282 (void) whitelist_major(prog
, path
, val
, 'c', acc
);
1284 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1287 r
= cgroup_apply_device_bpf(u
, prog
, c
->device_policy
, c
->device_allow
);
1289 static bool warned
= false;
1291 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1292 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1293 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1294 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1300 if (apply_mask
& CGROUP_MASK_PIDS
) {
1303 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1304 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1305 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1306 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1307 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1308 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1309 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1310 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1311 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1312 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1313 * which is desirable so that there's an official way to release control of the sysctl from
1314 * systemd: set the limit to unbounded and reload. */
1316 if (c
->tasks_max
!= CGROUP_LIMIT_MAX
) {
1317 u
->manager
->sysctl_pid_max_changed
= true;
1318 r
= procfs_tasks_set_limit(c
->tasks_max
);
1319 } else if (u
->manager
->sysctl_pid_max_changed
)
1320 r
= procfs_tasks_set_limit(TASKS_MAX
);
1324 log_unit_full(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
1325 "Failed to write to tasks limit sysctls: %m");
1328 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
1329 * leave it for the container manager. */
1330 if (!is_local_root
) {
1331 if (c
->tasks_max
!= CGROUP_LIMIT_MAX
) {
1332 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
1334 sprintf(buf
, "%" PRIu64
"\n", c
->tasks_max
);
1335 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
1337 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
1341 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
1342 cgroup_apply_firewall(u
);
1345 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
1350 c
= unit_get_cgroup_context(u
);
1354 if (c
->ip_accounting
||
1355 c
->ip_address_allow
||
1359 /* If any parent slice has an IP access list defined, it applies too */
1360 for (p
= UNIT_DEREF(u
->slice
); p
; p
= UNIT_DEREF(p
->slice
)) {
1361 c
= unit_get_cgroup_context(p
);
1365 if (c
->ip_address_allow
||
1373 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
1374 CGroupMask mask
= 0;
1379 c
= unit_get_cgroup_context(u
);
1383 /* Figure out which controllers we need, based on the cgroup context object */
1385 if (c
->cpu_accounting
)
1386 mask
|= get_cpu_accounting_mask();
1388 if (cgroup_context_has_cpu_weight(c
) ||
1389 cgroup_context_has_cpu_shares(c
) ||
1390 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
1391 mask
|= CGROUP_MASK_CPU
;
1393 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
1394 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1396 if (c
->memory_accounting
||
1397 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
1398 unit_has_unified_memory_config(u
))
1399 mask
|= CGROUP_MASK_MEMORY
;
1401 if (c
->device_allow
||
1402 c
->device_policy
!= CGROUP_AUTO
)
1403 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
1405 if (c
->tasks_accounting
||
1406 c
->tasks_max
!= CGROUP_LIMIT_MAX
)
1407 mask
|= CGROUP_MASK_PIDS
;
1409 return CGROUP_MASK_EXTEND_JOINED(mask
);
1412 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
1413 CGroupMask mask
= 0;
1415 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
1418 if (unit_get_needs_bpf_firewall(u
))
1419 mask
|= CGROUP_MASK_BPF_FIREWALL
;
1424 CGroupMask
unit_get_own_mask(Unit
*u
) {
1427 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
1428 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
1430 if (u
->load_state
!= UNIT_LOADED
)
1433 c
= unit_get_cgroup_context(u
);
1437 return (unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
)) & ~unit_get_ancestor_disable_mask(u
);
1440 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
1443 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1444 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1446 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1448 if (!unit_cgroup_delegate(u
))
1451 if (cg_all_unified() <= 0) {
1454 e
= unit_get_exec_context(u
);
1455 if (e
&& !exec_context_maintains_privileges(e
))
1459 assert_se(c
= unit_get_cgroup_context(u
));
1460 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
1463 CGroupMask
unit_get_members_mask(Unit
*u
) {
1466 /* Returns the mask of controllers all of the unit's children require, merged */
1468 if (u
->cgroup_members_mask_valid
)
1469 return u
->cgroup_members_mask
; /* Use cached value if possible */
1471 u
->cgroup_members_mask
= 0;
1473 if (u
->type
== UNIT_SLICE
) {
1478 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
], i
) {
1479 if (UNIT_DEREF(member
->slice
) == u
)
1480 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
1484 u
->cgroup_members_mask_valid
= true;
1485 return u
->cgroup_members_mask
;
1488 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
1491 /* Returns the mask of controllers all of the unit's siblings
1492 * require, i.e. the members mask of the unit's parent slice
1493 * if there is one. */
1495 if (UNIT_ISSET(u
->slice
))
1496 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
1498 return unit_get_subtree_mask(u
); /* we are the top-level slice */
1501 CGroupMask
unit_get_disable_mask(Unit
*u
) {
1504 c
= unit_get_cgroup_context(u
);
1508 return c
->disable_controllers
;
1511 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
1515 mask
= unit_get_disable_mask(u
);
1517 /* Returns the mask of controllers which are marked as forcibly
1518 * disabled in any ancestor unit or the unit in question. */
1520 if (UNIT_ISSET(u
->slice
))
1521 mask
|= unit_get_ancestor_disable_mask(UNIT_DEREF(u
->slice
));
1526 CGroupMask
unit_get_subtree_mask(Unit
*u
) {
1528 /* Returns the mask of this subtree, meaning of the group
1529 * itself and its children. */
1531 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
1534 CGroupMask
unit_get_target_mask(Unit
*u
) {
1537 /* This returns the cgroup mask of all controllers to enable
1538 * for a specific cgroup, i.e. everything it needs itself,
1539 * plus all that its children need, plus all that its siblings
1540 * need. This is primarily useful on the legacy cgroup
1541 * hierarchy, where we need to duplicate each cgroup in each
1542 * hierarchy that shall be enabled for it. */
1544 mask
= unit_get_own_mask(u
) | unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
1546 if (mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
1547 emit_bpf_firewall_warning(u
);
1549 mask
&= u
->manager
->cgroup_supported
;
1550 mask
&= ~unit_get_ancestor_disable_mask(u
);
1555 CGroupMask
unit_get_enable_mask(Unit
*u
) {
1558 /* This returns the cgroup mask of all controllers to enable
1559 * for the children of a specific cgroup. This is primarily
1560 * useful for the unified cgroup hierarchy, where each cgroup
1561 * controls which controllers are enabled for its children. */
1563 mask
= unit_get_members_mask(u
);
1564 mask
&= u
->manager
->cgroup_supported
;
1565 mask
&= ~unit_get_ancestor_disable_mask(u
);
1570 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
1573 /* Recurse invalidate the member masks cache all the way up the tree */
1574 u
->cgroup_members_mask_valid
= false;
1576 if (UNIT_ISSET(u
->slice
))
1577 unit_invalidate_cgroup_members_masks(UNIT_DEREF(u
->slice
));
1580 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
1582 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1586 if (u
->cgroup_path
&&
1587 u
->cgroup_realized
&&
1588 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
1589 return u
->cgroup_path
;
1591 u
= UNIT_DEREF(u
->slice
);
1597 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1598 return unit_get_realized_cgroup_path(userdata
, mask
);
1601 char *unit_default_cgroup_path(const Unit
*u
) {
1602 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1607 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1608 return strdup(u
->manager
->cgroup_root
);
1610 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1611 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1616 escaped
= cg_escape(u
->id
);
1621 return strjoin(u
->manager
->cgroup_root
, "/", slice
, "/",
1624 return strjoin(u
->manager
->cgroup_root
, "/", escaped
);
1627 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1628 _cleanup_free_
char *p
= NULL
;
1633 if (streq_ptr(u
->cgroup_path
, path
))
1643 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1648 unit_release_cgroup(u
);
1649 u
->cgroup_path
= TAKE_PTR(p
);
1654 int unit_watch_cgroup(Unit
*u
) {
1655 _cleanup_free_
char *events
= NULL
;
1660 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
1661 * cgroupv2 is available. */
1663 if (!u
->cgroup_path
)
1666 if (u
->cgroup_control_inotify_wd
>= 0)
1669 /* Only applies to the unified hierarchy */
1670 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1672 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1676 /* No point in watch the top-level slice, it's never going to run empty. */
1677 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1680 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
1684 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1688 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1689 if (u
->cgroup_control_inotify_wd
< 0) {
1691 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1692 * is not an error */
1695 return log_unit_error_errno(u
, errno
, "Failed to add control inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1698 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
1700 return log_unit_error_errno(u
, r
, "Failed to add control inotify watch descriptor to hash map: %m");
1705 int unit_watch_cgroup_memory(Unit
*u
) {
1706 _cleanup_free_
char *events
= NULL
;
1712 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
1713 * cgroupv2 is available. */
1715 if (!u
->cgroup_path
)
1718 c
= unit_get_cgroup_context(u
);
1722 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
1723 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
1725 if (!c
->memory_accounting
)
1728 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
1729 * we also don't want to generate a log message for each parent cgroup of a process. */
1730 if (u
->type
== UNIT_SLICE
)
1733 if (u
->cgroup_memory_inotify_wd
>= 0)
1736 /* Only applies to the unified hierarchy */
1737 r
= cg_all_unified();
1739 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
1743 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
1747 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
1751 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1752 if (u
->cgroup_memory_inotify_wd
< 0) {
1754 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1755 * is not an error */
1758 return log_unit_error_errno(u
, errno
, "Failed to add memory inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1761 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
1763 return log_unit_error_errno(u
, r
, "Failed to add memory inotify watch descriptor to hash map: %m");
1768 int unit_pick_cgroup_path(Unit
*u
) {
1769 _cleanup_free_
char *path
= NULL
;
1777 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1780 path
= unit_default_cgroup_path(u
);
1784 r
= unit_set_cgroup_path(u
, path
);
1786 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1788 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1793 static int unit_create_cgroup(
1795 CGroupMask target_mask
,
1796 CGroupMask enable_mask
,
1797 ManagerState state
) {
1804 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1807 /* Figure out our cgroup path */
1808 r
= unit_pick_cgroup_path(u
);
1812 /* First, create our own group */
1813 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
1815 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
1818 /* Start watching it */
1819 (void) unit_watch_cgroup(u
);
1820 (void) unit_watch_cgroup_memory(u
);
1822 /* Preserve enabled controllers in delegated units, adjust others. */
1823 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
1824 CGroupMask result_mask
= 0;
1826 /* Enable all controllers we need */
1827 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
1829 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
1831 /* If we just turned off a controller, this might release the controller for our parent too, let's
1832 * enqueue the parent for re-realization in that case again. */
1833 if (UNIT_ISSET(u
->slice
)) {
1834 CGroupMask turned_off
;
1836 turned_off
= (u
->cgroup_realized
? u
->cgroup_enabled_mask
& ~result_mask
: 0);
1837 if (turned_off
!= 0) {
1840 /* Force the parent to propagate the enable mask to the kernel again, by invalidating
1841 * the controller we just turned off. */
1843 for (parent
= UNIT_DEREF(u
->slice
); parent
; parent
= UNIT_DEREF(parent
->slice
))
1844 unit_invalidate_cgroup(parent
, turned_off
);
1848 /* Remember what's actually enabled now */
1849 u
->cgroup_enabled_mask
= result_mask
;
1852 /* Keep track that this is now realized */
1853 u
->cgroup_realized
= true;
1854 u
->cgroup_realized_mask
= target_mask
;
1856 if (u
->type
!= UNIT_SLICE
&& !unit_cgroup_delegate(u
)) {
1858 /* Then, possibly move things over, but not if
1859 * subgroups may contain processes, which is the case
1860 * for slice and delegation units. */
1861 r
= cg_migrate_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, u
->cgroup_path
, migrate_callback
, u
);
1863 log_unit_warning_errno(u
, r
, "Failed to migrate cgroup from to %s, ignoring: %m", u
->cgroup_path
);
1866 /* Set attributes */
1867 cgroup_context_apply(u
, target_mask
, state
);
1868 cgroup_xattr_apply(u
);
1873 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
1874 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1880 if (MANAGER_IS_SYSTEM(u
->manager
))
1883 if (!u
->manager
->system_bus
)
1886 if (!u
->cgroup_path
)
1889 /* Determine this unit's cgroup path relative to our cgroup root */
1890 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
1894 pp
= strjoina("/", pp
, suffix_path
);
1895 path_simplify(pp
, false);
1897 r
= sd_bus_call_method(u
->manager
->system_bus
,
1898 "org.freedesktop.systemd1",
1899 "/org/freedesktop/systemd1",
1900 "org.freedesktop.systemd1.Manager",
1901 "AttachProcessesToUnit",
1904 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
1906 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
1911 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
1912 CGroupMask delegated_mask
;
1920 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1923 if (set_isempty(pids
))
1926 r
= unit_realize_cgroup(u
);
1930 if (isempty(suffix_path
))
1933 p
= strjoina(u
->cgroup_path
, "/", suffix_path
);
1935 delegated_mask
= unit_get_delegate_mask(u
);
1938 SET_FOREACH(pidp
, pids
, i
) {
1939 pid_t pid
= PTR_TO_PID(pidp
);
1942 /* First, attach the PID to the main cgroup hierarchy */
1943 q
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
);
1945 log_unit_debug_errno(u
, q
, "Couldn't move process " PID_FMT
" to requested cgroup '%s': %m", pid
, p
);
1947 if (MANAGER_IS_USER(u
->manager
) && IN_SET(q
, -EPERM
, -EACCES
)) {
1950 /* If we are in a user instance, and we can't move the process ourselves due to
1951 * permission problems, let's ask the system instance about it instead. Since it's more
1952 * privileged it might be able to move the process across the leaves of a subtree who's
1953 * top node is not owned by us. */
1955 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
, suffix_path
);
1957 log_unit_debug_errno(u
, z
, "Couldn't move process " PID_FMT
" to requested cgroup '%s' via the system bus either: %m", pid
, p
);
1959 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
1963 r
= q
; /* Remember first error */
1968 q
= cg_all_unified();
1974 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
1975 * innermost realized one */
1977 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
1978 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
1979 const char *realized
;
1981 if (!(u
->manager
->cgroup_supported
& bit
))
1984 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
1985 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
1986 q
= cg_attach(cgroup_controller_to_string(c
), p
, pid
);
1988 continue; /* Success! */
1990 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",
1991 pid
, p
, cgroup_controller_to_string(c
));
1994 /* So this controller is either not delegate or realized, or something else weird happened. In
1995 * that case let's attach the PID at least to the closest cgroup up the tree that is
1997 realized
= unit_get_realized_cgroup_path(u
, bit
);
1999 continue; /* Not even realized in the root slice? Then let's not bother */
2001 q
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
);
2003 log_unit_debug_errno(u
, q
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
2004 pid
, realized
, cgroup_controller_to_string(c
));
2011 static bool unit_has_mask_realized(
2013 CGroupMask target_mask
,
2014 CGroupMask enable_mask
) {
2018 /* Returns true if this unit is fully realized. We check four things:
2020 * 1. Whether the cgroup was created at all
2021 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
2022 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
2023 * 4. Whether the invalidation mask is currently zero
2025 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
2026 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
2027 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
2028 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
2029 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
2030 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
2031 * simply don't matter. */
2033 return u
->cgroup_realized
&&
2034 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
2035 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
2036 u
->cgroup_invalidated_mask
== 0;
2039 static bool unit_has_mask_disables_realized(
2041 CGroupMask target_mask
,
2042 CGroupMask enable_mask
) {
2046 /* Returns true if all controllers which should be disabled are indeed disabled.
2048 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
2049 * already removed. */
2051 return !u
->cgroup_realized
||
2052 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
2053 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
2056 static bool unit_has_mask_enables_realized(
2058 CGroupMask target_mask
,
2059 CGroupMask enable_mask
) {
2063 /* Returns true if all controllers which should be enabled are indeed enabled.
2065 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
2066 * we want to add is already added. */
2068 return u
->cgroup_realized
&&
2069 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
2070 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
2073 void unit_add_to_cgroup_realize_queue(Unit
*u
) {
2076 if (u
->in_cgroup_realize_queue
)
2079 LIST_PREPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2080 u
->in_cgroup_realize_queue
= true;
2083 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
2086 if (!u
->in_cgroup_realize_queue
)
2089 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2090 u
->in_cgroup_realize_queue
= false;
2093 /* Controllers can only be enabled breadth-first, from the root of the
2094 * hierarchy downwards to the unit in question. */
2095 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
2096 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2101 /* First go deal with this unit's parent, or we won't be able to enable
2102 * any new controllers at this layer. */
2103 if (UNIT_ISSET(u
->slice
)) {
2104 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2109 target_mask
= unit_get_target_mask(u
);
2110 enable_mask
= unit_get_enable_mask(u
);
2112 /* We can only enable in this direction, don't try to disable anything.
2114 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
2117 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
2118 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
2120 return unit_create_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
2123 /* Controllers can only be disabled depth-first, from the leaves of the
2124 * hierarchy upwards to the unit in question. */
2125 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
2132 if (u
->type
!= UNIT_SLICE
)
2135 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
], i
) {
2136 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2139 if (UNIT_DEREF(m
->slice
) != u
)
2142 /* The cgroup for this unit might not actually be fully
2143 * realised yet, in which case it isn't holding any controllers
2145 if (!m
->cgroup_path
)
2148 /* We must disable those below us first in order to release the
2150 if (m
->type
== UNIT_SLICE
)
2151 (void) unit_realize_cgroup_now_disable(m
, state
);
2153 target_mask
= unit_get_target_mask(m
);
2154 enable_mask
= unit_get_enable_mask(m
);
2156 /* We can only disable in this direction, don't try to enable
2158 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
2161 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
2162 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
2164 r
= unit_create_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
2172 /* Check if necessary controllers and attributes for a unit are in place.
2174 * - If so, do nothing.
2175 * - If not, create paths, move processes over, and set attributes.
2177 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
2178 * a depth-first way. As such the process looks like this:
2180 * Suppose we have a cgroup hierarchy which looks like this:
2193 * 1. We want to realise cgroup "d" now.
2194 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
2195 * 3. cgroup "k" just started requesting the memory controller.
2197 * To make this work we must do the following in order:
2199 * 1. Disable CPU controller in k, j
2200 * 2. Disable CPU controller in d
2201 * 3. Enable memory controller in root
2202 * 4. Enable memory controller in a
2203 * 5. Enable memory controller in d
2204 * 6. Enable memory controller in k
2206 * Notice that we need to touch j in one direction, but not the other. We also
2207 * don't go beyond d when disabling -- it's up to "a" to get realized if it
2208 * wants to disable further. The basic rules are therefore:
2210 * - If you're disabling something, you need to realise all of the cgroups from
2211 * your recursive descendants to the root. This starts from the leaves.
2212 * - If you're enabling something, you need to realise from the root cgroup
2213 * downwards, but you don't need to iterate your recursive descendants.
2215 * Returns 0 on success and < 0 on failure. */
2216 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
2217 CGroupMask target_mask
, enable_mask
;
2222 unit_remove_from_cgroup_realize_queue(u
);
2224 target_mask
= unit_get_target_mask(u
);
2225 enable_mask
= unit_get_enable_mask(u
);
2227 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
2230 /* Disable controllers below us, if there are any */
2231 r
= unit_realize_cgroup_now_disable(u
, state
);
2235 /* Enable controllers above us, if there are any */
2236 if (UNIT_ISSET(u
->slice
)) {
2237 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2242 /* Now actually deal with the cgroup we were trying to realise and set attributes */
2243 r
= unit_create_cgroup(u
, target_mask
, enable_mask
, state
);
2247 /* Now, reset the invalidation mask */
2248 u
->cgroup_invalidated_mask
= 0;
2252 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
2260 state
= manager_state(m
);
2262 while ((i
= m
->cgroup_realize_queue
)) {
2263 assert(i
->in_cgroup_realize_queue
);
2265 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
2266 /* Maybe things changed, and the unit is not actually active anymore? */
2267 unit_remove_from_cgroup_realize_queue(i
);
2271 r
= unit_realize_cgroup_now(i
, state
);
2273 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
2281 static void unit_add_siblings_to_cgroup_realize_queue(Unit
*u
) {
2284 /* This adds the siblings of the specified unit and the
2285 * siblings of all parent units to the cgroup queue. (But
2286 * neither the specified unit itself nor the parents.) */
2288 while ((slice
= UNIT_DEREF(u
->slice
))) {
2293 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
], i
) {
2294 /* Skip units that have a dependency on the slice
2295 * but aren't actually in it. */
2296 if (UNIT_DEREF(m
->slice
) != slice
)
2299 /* No point in doing cgroup application for units
2300 * without active processes. */
2301 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
2304 /* If the unit doesn't need any new controllers
2305 * and has current ones realized, it doesn't need
2307 if (unit_has_mask_realized(m
,
2308 unit_get_target_mask(m
),
2309 unit_get_enable_mask(m
)))
2312 unit_add_to_cgroup_realize_queue(m
);
2319 int unit_realize_cgroup(Unit
*u
) {
2322 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2325 /* So, here's the deal: when realizing the cgroups for this
2326 * unit, we need to first create all parents, but there's more
2327 * actually: for the weight-based controllers we also need to
2328 * make sure that all our siblings (i.e. units that are in the
2329 * same slice as we are) have cgroups, too. Otherwise, things
2330 * would become very uneven as each of their processes would
2331 * get as much resources as all our group together. This call
2332 * will synchronously create the parent cgroups, but will
2333 * defer work on the siblings to the next event loop
2336 /* Add all sibling slices to the cgroup queue. */
2337 unit_add_siblings_to_cgroup_realize_queue(u
);
2339 /* And realize this one now (and apply the values) */
2340 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
2343 void unit_release_cgroup(Unit
*u
) {
2346 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
2347 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
2349 if (u
->cgroup_path
) {
2350 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
2351 u
->cgroup_path
= mfree(u
->cgroup_path
);
2354 if (u
->cgroup_control_inotify_wd
>= 0) {
2355 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
2356 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
);
2358 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
2359 u
->cgroup_control_inotify_wd
= -1;
2362 if (u
->cgroup_memory_inotify_wd
>= 0) {
2363 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
2364 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
);
2366 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
2367 u
->cgroup_memory_inotify_wd
= -1;
2371 void unit_prune_cgroup(Unit
*u
) {
2377 /* Removes the cgroup, if empty and possible, and stops watching it. */
2379 if (!u
->cgroup_path
)
2382 (void) unit_get_cpu_usage(u
, NULL
); /* Cache the last CPU usage value before we destroy the cgroup */
2384 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2386 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
2388 log_unit_debug_errno(u
, r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
2395 unit_release_cgroup(u
);
2397 u
->cgroup_realized
= false;
2398 u
->cgroup_realized_mask
= 0;
2399 u
->cgroup_enabled_mask
= 0;
2401 u
->bpf_device_control_installed
= bpf_program_unref(u
->bpf_device_control_installed
);
2404 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
2405 _cleanup_fclose_
FILE *f
= NULL
;
2406 pid_t pid
= 0, npid
;
2412 if (!u
->cgroup_path
)
2415 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
2419 while (cg_read_pid(f
, &npid
) > 0) {
2424 if (pid_is_my_child(npid
) == 0)
2428 /* Dang, there's more than one daemonized PID
2429 in this group, so we don't know what process
2430 is the main process. */
2441 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
2442 _cleanup_closedir_
DIR *d
= NULL
;
2443 _cleanup_fclose_
FILE *f
= NULL
;
2449 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
2455 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
2456 r
= unit_watch_pid(u
, pid
, false);
2457 if (r
< 0 && ret
>= 0)
2461 if (r
< 0 && ret
>= 0)
2465 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
2472 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
2473 _cleanup_free_
char *p
= NULL
;
2475 p
= strjoin(path
, "/", fn
);
2481 r
= unit_watch_pids_in_path(u
, p
);
2482 if (r
< 0 && ret
>= 0)
2486 if (r
< 0 && ret
>= 0)
2493 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
2498 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
2499 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
2500 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
2502 if (!u
->cgroup_path
)
2505 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2508 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
2511 if (!set_isempty(u
->pids
))
2514 unit_add_to_cgroup_empty_queue(u
);
2518 int unit_watch_all_pids(Unit
*u
) {
2523 /* Adds all PIDs from our cgroup to the set of PIDs we
2524 * watch. This is a fallback logic for cases where we do not
2525 * get reliable cgroup empty notifications: we try to use
2526 * SIGCHLD as replacement. */
2528 if (!u
->cgroup_path
)
2531 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2534 if (r
> 0) /* On unified we can use proper notifications */
2537 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
2540 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
2541 Manager
*m
= userdata
;
2548 u
= m
->cgroup_empty_queue
;
2552 assert(u
->in_cgroup_empty_queue
);
2553 u
->in_cgroup_empty_queue
= false;
2554 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
2556 if (m
->cgroup_empty_queue
) {
2557 /* More stuff queued, let's make sure we remain enabled */
2558 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2560 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
2563 unit_add_to_gc_queue(u
);
2565 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
2566 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
2571 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
2576 /* Note that there are four different ways how cgroup empty events reach us:
2578 * 1. On the unified hierarchy we get an inotify event on the cgroup
2580 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2582 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2584 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2585 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2587 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2588 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2589 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2590 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2591 * case for scope units). */
2593 if (u
->in_cgroup_empty_queue
)
2596 /* Let's verify that the cgroup is really empty */
2597 if (!u
->cgroup_path
)
2599 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2601 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", u
->cgroup_path
);
2607 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2608 u
->in_cgroup_empty_queue
= true;
2610 /* Trigger the defer event */
2611 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
2613 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
2616 int unit_check_oom(Unit
*u
) {
2617 _cleanup_free_
char *oom_kill
= NULL
;
2622 if (!u
->cgroup_path
)
2625 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
2627 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
2629 r
= safe_atou64(oom_kill
, &c
);
2631 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
2633 increased
= c
> u
->oom_kill_last
;
2634 u
->oom_kill_last
= c
;
2639 log_struct(LOG_NOTICE
,
2640 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
2642 LOG_UNIT_INVOCATION_ID(u
),
2643 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
2645 if (UNIT_VTABLE(u
)->notify_cgroup_oom
)
2646 UNIT_VTABLE(u
)->notify_cgroup_oom(u
);
2651 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
2652 Manager
*m
= userdata
;
2659 u
= m
->cgroup_oom_queue
;
2663 assert(u
->in_cgroup_oom_queue
);
2664 u
->in_cgroup_oom_queue
= false;
2665 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
2667 if (m
->cgroup_oom_queue
) {
2668 /* More stuff queued, let's make sure we remain enabled */
2669 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2671 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
2674 (void) unit_check_oom(u
);
2678 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
2683 if (u
->in_cgroup_oom_queue
)
2685 if (!u
->cgroup_path
)
2688 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
2689 u
->in_cgroup_oom_queue
= true;
2691 /* Trigger the defer event */
2692 if (!u
->manager
->cgroup_oom_event_source
) {
2693 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
2695 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
2697 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
2701 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
2703 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
2707 (void) sd_event_source_set_description(s
, "cgroup-oom");
2708 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
2711 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
2713 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
2716 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
2717 Manager
*m
= userdata
;
2724 union inotify_event_buffer buffer
;
2725 struct inotify_event
*e
;
2728 l
= read(fd
, &buffer
, sizeof(buffer
));
2730 if (IN_SET(errno
, EINTR
, EAGAIN
))
2733 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
2736 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
2740 /* Queue overflow has no watch descriptor */
2743 if (e
->mask
& IN_IGNORED
)
2744 /* The watch was just removed */
2747 /* Note that inotify might deliver events for a watch even after it was removed,
2748 * because it was queued before the removal. Let's ignore this here safely. */
2750 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2752 unit_add_to_cgroup_empty_queue(u
);
2754 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2756 unit_add_to_cgroup_oom_queue(u
);
2761 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
2762 CGroupMask mask
= 0;
2765 /* BPF-based firewall */
2766 r
= bpf_firewall_supported();
2768 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2770 /* BPF-based device access control */
2771 r
= bpf_devices_supported();
2773 mask
|= CGROUP_MASK_BPF_DEVICES
;
2779 int manager_setup_cgroup(Manager
*m
) {
2780 _cleanup_free_
char *path
= NULL
;
2781 const char *scope_path
;
2789 /* 1. Determine hierarchy */
2790 m
->cgroup_root
= mfree(m
->cgroup_root
);
2791 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
2793 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
2795 /* Chop off the init scope, if we are already located in it */
2796 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2798 /* LEGACY: Also chop off the system slice if we are in
2799 * it. This is to support live upgrades from older systemd
2800 * versions where PID 1 was moved there. Also see
2801 * cg_get_root_path(). */
2802 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
2803 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
2805 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
2810 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
2811 * easily prepend it everywhere. */
2812 delete_trailing_chars(m
->cgroup_root
, "/");
2815 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
2817 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
2819 r
= cg_unified_flush();
2821 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
2823 all_unified
= cg_all_unified();
2824 if (all_unified
< 0)
2825 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
2826 if (all_unified
> 0)
2827 log_debug("Unified cgroup hierarchy is located at %s.", path
);
2829 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2831 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
2833 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
2835 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
2838 /* 3. Allocate cgroup empty defer event source */
2839 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
2840 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
2842 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
2844 /* Schedule cgroup empty checks early, but after having processed service notification messages or
2845 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
2846 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
2847 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
2849 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
2851 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
2853 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
2855 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
2857 /* 4. Install notifier inotify object, or agent */
2858 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
2860 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
2862 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
2863 safe_close(m
->cgroup_inotify_fd
);
2865 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
2866 if (m
->cgroup_inotify_fd
< 0)
2867 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
2869 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
2871 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
2873 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
2874 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
2875 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
2876 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
2878 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
2880 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
2882 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
2884 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
2885 * since it does not generate events when control groups with children run empty. */
2887 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUP_AGENT_PATH
);
2889 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
2891 log_debug("Installed release agent.");
2893 log_debug("Release agent already installed.");
2896 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
2897 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2898 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
2900 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
2901 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
2903 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
2905 /* 6. And pin it, so that it cannot be unmounted */
2906 safe_close(m
->pin_cgroupfs_fd
);
2907 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
2908 if (m
->pin_cgroupfs_fd
< 0)
2909 return log_error_errno(errno
, "Failed to open pin file: %m");
2911 } else if (!MANAGER_IS_TEST_RUN(m
))
2912 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
2914 /* 7. Always enable hierarchical support if it exists... */
2915 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
2916 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
2918 /* 8. Figure out which controllers are supported */
2919 r
= cg_mask_supported(&m
->cgroup_supported
);
2921 return log_error_errno(r
, "Failed to determine supported controllers: %m");
2923 /* 9. Figure out which bpf-based pseudo-controllers are supported */
2924 r
= cg_bpf_mask_supported(&mask
);
2926 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
2927 m
->cgroup_supported
|= mask
;
2929 /* 10. Log which controllers are supported */
2930 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
2931 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
2936 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
2939 /* We can't really delete the group, since we are in it. But
2941 if (delete && m
->cgroup_root
&& m
->test_run_flags
!= MANAGER_TEST_RUN_MINIMAL
)
2942 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
2944 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
2946 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
2947 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
2949 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
2950 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
2952 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
2954 m
->cgroup_root
= mfree(m
->cgroup_root
);
2957 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
2964 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
2968 p
= strdupa(cgroup
);
2972 e
= strrchr(p
, '/');
2974 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
2978 u
= hashmap_get(m
->cgroup_unit
, p
);
2984 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
2985 _cleanup_free_
char *cgroup
= NULL
;
2989 if (!pid_is_valid(pid
))
2992 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
2995 return manager_get_unit_by_cgroup(m
, cgroup
);
2998 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
3003 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
3004 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
3005 * relevant one as children of the process will be assigned to that one, too, before all else. */
3007 if (!pid_is_valid(pid
))
3010 if (pid
== getpid_cached())
3011 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
3013 u
= manager_get_unit_by_pid_cgroup(m
, pid
);
3017 u
= hashmap_get(m
->watch_pids
, PID_TO_PTR(pid
));
3021 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-pid
));
3028 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
3034 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
3035 * or from the --system instance */
3037 log_debug("Got cgroup empty notification for: %s", cgroup
);
3039 u
= manager_get_unit_by_cgroup(m
, cgroup
);
3043 unit_add_to_cgroup_empty_queue(u
);
3047 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
3048 _cleanup_free_
char *v
= NULL
;
3054 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3057 if (!u
->cgroup_path
)
3060 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3061 if (unit_has_host_root_cgroup(u
))
3062 return procfs_memory_get_used(ret
);
3064 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3067 r
= cg_all_unified();
3071 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.current", &v
);
3073 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.usage_in_bytes", &v
);
3079 return safe_atou64(v
, ret
);
3082 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
3083 _cleanup_free_
char *v
= NULL
;
3089 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
3092 if (!u
->cgroup_path
)
3095 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3096 if (unit_has_host_root_cgroup(u
))
3097 return procfs_tasks_get_current(ret
);
3099 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
3102 r
= cg_get_attribute("pids", u
->cgroup_path
, "pids.current", &v
);
3108 return safe_atou64(v
, ret
);
3111 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
3112 _cleanup_free_
char *v
= NULL
;
3119 if (!u
->cgroup_path
)
3122 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3123 if (unit_has_host_root_cgroup(u
))
3124 return procfs_cpu_get_usage(ret
);
3126 /* Requisite controllers for CPU accounting are not enabled */
3127 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
3130 r
= cg_all_unified();
3134 _cleanup_free_
char *val
= NULL
;
3137 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
3138 if (IN_SET(r
, -ENOENT
, -ENXIO
))
3143 r
= safe_atou64(val
, &us
);
3147 ns
= us
* NSEC_PER_USEC
;
3149 r
= cg_get_attribute("cpuacct", u
->cgroup_path
, "cpuacct.usage", &v
);
3155 r
= safe_atou64(v
, &ns
);
3164 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
3170 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
3171 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
3172 * call this function with a NULL return value. */
3174 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
3177 r
= unit_get_cpu_usage_raw(u
, &ns
);
3178 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
3179 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
3183 *ret
= u
->cpu_usage_last
;
3189 if (ns
> u
->cpu_usage_base
)
3190 ns
-= u
->cpu_usage_base
;
3194 u
->cpu_usage_last
= ns
;
3201 int unit_get_ip_accounting(
3203 CGroupIPAccountingMetric metric
,
3210 assert(metric
>= 0);
3211 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
3214 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
3217 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
3218 u
->ip_accounting_ingress_map_fd
:
3219 u
->ip_accounting_egress_map_fd
;
3223 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
3224 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
3226 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
3230 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
3231 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
3232 * ip_accounting_extra[] field, and add them in here transparently. */
3234 *ret
= value
+ u
->ip_accounting_extra
[metric
];
3239 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
3240 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
3241 [CGROUP_IO_READ_BYTES
] = "rbytes=",
3242 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
3243 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
3244 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
3246 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
3247 _cleanup_free_
char *path
= NULL
;
3248 _cleanup_fclose_
FILE *f
= NULL
;
3253 if (!u
->cgroup_path
)
3256 if (unit_has_host_root_cgroup(u
))
3257 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
3259 r
= cg_all_unified();
3262 if (r
== 0) /* TODO: support cgroupv1 */
3265 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
3268 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
3272 f
= fopen(path
, "re");
3277 _cleanup_free_
char *line
= NULL
;
3280 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3287 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
3288 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
3291 _cleanup_free_
char *word
= NULL
;
3293 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
3299 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3302 x
= startswith(word
, field_names
[i
]);
3306 r
= safe_atou64(x
, &w
);
3310 /* Sum up the stats of all devices */
3318 memcpy(ret
, acc
, sizeof(acc
));
3322 int unit_get_io_accounting(
3324 CGroupIOAccountingMetric metric
,
3328 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
3331 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
3333 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
3336 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3339 r
= unit_get_io_accounting_raw(u
, raw
);
3340 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3345 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3346 /* Saturated subtraction */
3347 if (raw
[i
] > u
->io_accounting_base
[i
])
3348 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
3350 u
->io_accounting_last
[i
] = 0;
3355 *ret
= u
->io_accounting_last
[metric
];
3360 int unit_reset_cpu_accounting(Unit
*u
) {
3365 u
->cpu_usage_last
= NSEC_INFINITY
;
3367 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
3369 u
->cpu_usage_base
= 0;
3376 int unit_reset_ip_accounting(Unit
*u
) {
3381 if (u
->ip_accounting_ingress_map_fd
>= 0)
3382 r
= bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
);
3384 if (u
->ip_accounting_egress_map_fd
>= 0)
3385 q
= bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
);
3387 zero(u
->ip_accounting_extra
);
3389 return r
< 0 ? r
: q
;
3392 int unit_reset_io_accounting(Unit
*u
) {
3397 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++)
3398 u
->io_accounting_last
[i
] = UINT64_MAX
;
3400 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
3402 zero(u
->io_accounting_base
);
3409 int unit_reset_accounting(Unit
*u
) {
3414 r
= unit_reset_cpu_accounting(u
);
3415 q
= unit_reset_io_accounting(u
);
3416 v
= unit_reset_ip_accounting(u
);
3418 return r
< 0 ? r
: q
< 0 ? q
: v
;
3421 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
3424 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3430 /* always invalidate compat pairs together */
3431 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
3432 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
3434 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
3435 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
3437 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
3440 u
->cgroup_invalidated_mask
|= m
;
3441 unit_add_to_cgroup_realize_queue(u
);
3444 void unit_invalidate_cgroup_bpf(Unit
*u
) {
3447 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3450 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
3453 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
3454 unit_add_to_cgroup_realize_queue(u
);
3456 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
3457 * list of our children includes our own. */
3458 if (u
->type
== UNIT_SLICE
) {
3463 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
], i
) {
3464 if (UNIT_DEREF(member
->slice
) == u
)
3465 unit_invalidate_cgroup_bpf(member
);
3470 bool unit_cgroup_delegate(Unit
*u
) {
3475 if (!UNIT_VTABLE(u
)->can_delegate
)
3478 c
= unit_get_cgroup_context(u
);
3485 void manager_invalidate_startup_units(Manager
*m
) {
3491 SET_FOREACH(u
, m
->startup_units
, i
)
3492 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
3495 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
3496 [CGROUP_AUTO
] = "auto",
3497 [CGROUP_CLOSED
] = "closed",
3498 [CGROUP_STRICT
] = "strict",
3501 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);