1 /* SPDX-License-Identifier: LGPL-2.1+ */
5 #include "sd-messages.h"
7 #include "alloc-util.h"
8 #include "blockdev-util.h"
9 #include "bpf-devices.h"
10 #include "bpf-firewall.h"
11 #include "btrfs-util.h"
12 #include "bus-error.h"
13 #include "cgroup-setup.h"
14 #include "cgroup-util.h"
20 #include "limits-util.h"
21 #include "nulstr-util.h"
22 #include "parse-util.h"
23 #include "path-util.h"
24 #include "process-util.h"
25 #include "procfs-util.h"
27 #include "stat-util.h"
28 #include "stdio-util.h"
29 #include "string-table.h"
30 #include "string-util.h"
33 #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
35 /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access
36 * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask
37 * out specific attributes from us. */
38 #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING)
40 uint64_t tasks_max_resolve(const TasksMax
*tasks_max
) {
41 if (tasks_max
->scale
== 0)
42 return tasks_max
->value
;
44 return system_tasks_max_scale(tasks_max
->value
, tasks_max
->scale
);
47 bool manager_owns_host_root_cgroup(Manager
*m
) {
50 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
51 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
52 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
53 * we run in any kind of container virtualization. */
55 if (MANAGER_IS_USER(m
))
58 if (detect_container() > 0)
61 return empty_or_root(m
->cgroup_root
);
64 bool unit_has_host_root_cgroup(Unit
*u
) {
67 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
68 * the manager manages the root cgroup. */
70 if (!manager_owns_host_root_cgroup(u
->manager
))
73 return unit_has_name(u
, SPECIAL_ROOT_SLICE
);
76 static int set_attribute_and_warn(Unit
*u
, const char *controller
, const char *attribute
, const char *value
) {
79 r
= cg_set_attribute(controller
, u
->cgroup_path
, attribute
, value
);
81 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%.*s': %m",
82 strna(attribute
), isempty(u
->cgroup_path
) ? "/" : u
->cgroup_path
, (int) strcspn(value
, NEWLINE
), value
);
87 static void cgroup_compat_warn(void) {
88 static bool cgroup_compat_warned
= false;
90 if (cgroup_compat_warned
)
93 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
94 "See cgroup-compat debug messages for details.");
96 cgroup_compat_warned
= true;
99 #define log_cgroup_compat(unit, fmt, ...) do { \
100 cgroup_compat_warn(); \
101 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
104 void cgroup_context_init(CGroupContext
*c
) {
107 /* Initialize everything to the kernel defaults. */
109 *c
= (CGroupContext
) {
110 .cpu_weight
= CGROUP_WEIGHT_INVALID
,
111 .startup_cpu_weight
= CGROUP_WEIGHT_INVALID
,
112 .cpu_quota_per_sec_usec
= USEC_INFINITY
,
113 .cpu_quota_period_usec
= USEC_INFINITY
,
115 .cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
116 .startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
118 .memory_high
= CGROUP_LIMIT_MAX
,
119 .memory_max
= CGROUP_LIMIT_MAX
,
120 .memory_swap_max
= CGROUP_LIMIT_MAX
,
122 .memory_limit
= CGROUP_LIMIT_MAX
,
124 .io_weight
= CGROUP_WEIGHT_INVALID
,
125 .startup_io_weight
= CGROUP_WEIGHT_INVALID
,
127 .blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
128 .startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
130 .tasks_max
= TASKS_MAX_UNSET
,
132 .moom_swap
= MANAGED_OOM_AUTO
,
133 .moom_mem_pressure
= MANAGED_OOM_AUTO
,
137 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
141 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
146 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
150 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
155 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
159 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
164 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
168 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
173 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
177 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
182 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
186 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
191 void cgroup_context_done(CGroupContext
*c
) {
194 while (c
->io_device_weights
)
195 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
197 while (c
->io_device_latencies
)
198 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
200 while (c
->io_device_limits
)
201 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
203 while (c
->blockio_device_weights
)
204 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
206 while (c
->blockio_device_bandwidths
)
207 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
209 while (c
->device_allow
)
210 cgroup_context_free_device_allow(c
, c
->device_allow
);
212 c
->ip_address_allow
= ip_address_access_free_all(c
->ip_address_allow
);
213 c
->ip_address_deny
= ip_address_access_free_all(c
->ip_address_deny
);
215 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
216 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
218 cpu_set_reset(&c
->cpuset_cpus
);
219 cpu_set_reset(&c
->cpuset_mems
);
222 static int unit_get_kernel_memory_limit(Unit
*u
, const char *file
, uint64_t *ret
) {
225 if (!u
->cgroup_realized
)
228 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, file
, ret
);
231 static int unit_compare_memory_limit(Unit
*u
, const char *property_name
, uint64_t *ret_unit_value
, uint64_t *ret_kernel_value
) {
238 /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will
239 * return -ENODATA) on cgroup v1.
244 * 0: If the kernel memory setting doesn't match our configuration.
245 * >0: If the kernel memory setting matches our configuration.
247 * The following values are only guaranteed to be populated on return >=0:
249 * - ret_unit_value will contain our internal expected value for the unit, page-aligned.
250 * - ret_kernel_value will contain the actual value presented by the kernel. */
254 r
= cg_all_unified();
256 return log_debug_errno(r
, "Failed to determine cgroup hierarchy version: %m");
258 /* Unsupported on v1.
260 * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has
261 * silently masked the controller. */
265 /* The root slice doesn't have any controller files, so we can't compare anything. */
266 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
269 /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled,
270 * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To
271 * avoid specious errors in these scenarios, check that we even expect the memory controller to be
273 m
= unit_get_target_mask(u
);
274 if (!FLAGS_SET(m
, CGROUP_MASK_MEMORY
))
277 c
= unit_get_cgroup_context(u
);
280 if (streq(property_name
, "MemoryLow")) {
281 unit_value
= unit_get_ancestor_memory_low(u
);
283 } else if (streq(property_name
, "MemoryMin")) {
284 unit_value
= unit_get_ancestor_memory_min(u
);
286 } else if (streq(property_name
, "MemoryHigh")) {
287 unit_value
= c
->memory_high
;
288 file
= "memory.high";
289 } else if (streq(property_name
, "MemoryMax")) {
290 unit_value
= c
->memory_max
;
292 } else if (streq(property_name
, "MemorySwapMax")) {
293 unit_value
= c
->memory_swap_max
;
294 file
= "memory.swap.max";
298 r
= unit_get_kernel_memory_limit(u
, file
, ret_kernel_value
);
300 return log_unit_debug_errno(u
, r
, "Failed to parse %s: %m", file
);
302 /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page
303 * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from
304 * our internal page-counting. To support those future kernels, just check the value itself first
305 * without any page-alignment. */
306 if (*ret_kernel_value
== unit_value
) {
307 *ret_unit_value
= unit_value
;
311 /* The current kernel behaviour, by comparison, is that even if you write a particular number of
312 * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel
313 * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is
315 if (unit_value
!= CGROUP_LIMIT_MAX
)
316 unit_value
= PAGE_ALIGN_DOWN(unit_value
);
318 *ret_unit_value
= unit_value
;
320 return *ret_kernel_value
== *ret_unit_value
;
323 #define FORMAT_CGROUP_DIFF_MAX 128
325 static char *format_cgroup_memory_limit_comparison(char *buf
, size_t l
, Unit
*u
, const char *property_name
) {
333 r
= unit_compare_memory_limit(u
, property_name
, &sval
, &kval
);
335 /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1).
336 * In the absence of reliably being able to detect whether memcg swap support is available or not,
337 * only complain if the error is not ENOENT. */
338 if (r
> 0 || IN_SET(r
, -ENODATA
, -EOWNERDEAD
) ||
339 (r
== -ENOENT
&& streq(property_name
, "MemorySwapMax"))) {
345 snprintf(buf
, l
, " (error getting kernel value: %s)", strerror_safe(r
));
349 snprintf(buf
, l
, " (different value in kernel: %" PRIu64
")", kval
);
354 void cgroup_context_dump(Unit
*u
, FILE* f
, const char *prefix
) {
355 _cleanup_free_
char *disable_controllers_str
= NULL
, *cpuset_cpus
= NULL
, *cpuset_mems
= NULL
;
356 CGroupIODeviceLimit
*il
;
357 CGroupIODeviceWeight
*iw
;
358 CGroupIODeviceLatency
*l
;
359 CGroupBlockIODeviceBandwidth
*b
;
360 CGroupBlockIODeviceWeight
*w
;
361 CGroupDeviceAllow
*a
;
363 IPAddressAccessItem
*iaai
;
365 char q
[FORMAT_TIMESPAN_MAX
];
366 char v
[FORMAT_TIMESPAN_MAX
];
368 char cda
[FORMAT_CGROUP_DIFF_MAX
];
369 char cdb
[FORMAT_CGROUP_DIFF_MAX
];
370 char cdc
[FORMAT_CGROUP_DIFF_MAX
];
371 char cdd
[FORMAT_CGROUP_DIFF_MAX
];
372 char cde
[FORMAT_CGROUP_DIFF_MAX
];
377 c
= unit_get_cgroup_context(u
);
380 prefix
= strempty(prefix
);
382 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
384 cpuset_cpus
= cpu_set_to_range_string(&c
->cpuset_cpus
);
385 cpuset_mems
= cpu_set_to_range_string(&c
->cpuset_mems
);
388 "%sCPUAccounting: %s\n"
389 "%sIOAccounting: %s\n"
390 "%sBlockIOAccounting: %s\n"
391 "%sMemoryAccounting: %s\n"
392 "%sTasksAccounting: %s\n"
393 "%sIPAccounting: %s\n"
394 "%sCPUWeight: %" PRIu64
"\n"
395 "%sStartupCPUWeight: %" PRIu64
"\n"
396 "%sCPUShares: %" PRIu64
"\n"
397 "%sStartupCPUShares: %" PRIu64
"\n"
398 "%sCPUQuotaPerSecSec: %s\n"
399 "%sCPUQuotaPeriodSec: %s\n"
400 "%sAllowedCPUs: %s\n"
401 "%sAllowedMemoryNodes: %s\n"
402 "%sIOWeight: %" PRIu64
"\n"
403 "%sStartupIOWeight: %" PRIu64
"\n"
404 "%sBlockIOWeight: %" PRIu64
"\n"
405 "%sStartupBlockIOWeight: %" PRIu64
"\n"
406 "%sDefaultMemoryMin: %" PRIu64
"\n"
407 "%sDefaultMemoryLow: %" PRIu64
"\n"
408 "%sMemoryMin: %" PRIu64
"%s\n"
409 "%sMemoryLow: %" PRIu64
"%s\n"
410 "%sMemoryHigh: %" PRIu64
"%s\n"
411 "%sMemoryMax: %" PRIu64
"%s\n"
412 "%sMemorySwapMax: %" PRIu64
"%s\n"
413 "%sMemoryLimit: %" PRIu64
"\n"
414 "%sTasksMax: %" PRIu64
"\n"
415 "%sDevicePolicy: %s\n"
416 "%sDisableControllers: %s\n"
418 "%sManagedOOMSwap: %s\n"
419 "%sManagedOOMMemoryPressure: %s\n"
420 "%sManagedOOMMemoryPressureLimitPercent: %d%%\n",
421 prefix
, yes_no(c
->cpu_accounting
),
422 prefix
, yes_no(c
->io_accounting
),
423 prefix
, yes_no(c
->blockio_accounting
),
424 prefix
, yes_no(c
->memory_accounting
),
425 prefix
, yes_no(c
->tasks_accounting
),
426 prefix
, yes_no(c
->ip_accounting
),
427 prefix
, c
->cpu_weight
,
428 prefix
, c
->startup_cpu_weight
,
429 prefix
, c
->cpu_shares
,
430 prefix
, c
->startup_cpu_shares
,
431 prefix
, format_timespan(q
, sizeof(q
), c
->cpu_quota_per_sec_usec
, 1),
432 prefix
, format_timespan(v
, sizeof(v
), c
->cpu_quota_period_usec
, 1),
433 prefix
, strempty(cpuset_cpus
),
434 prefix
, strempty(cpuset_mems
),
435 prefix
, c
->io_weight
,
436 prefix
, c
->startup_io_weight
,
437 prefix
, c
->blockio_weight
,
438 prefix
, c
->startup_blockio_weight
,
439 prefix
, c
->default_memory_min
,
440 prefix
, c
->default_memory_low
,
441 prefix
, c
->memory_min
, format_cgroup_memory_limit_comparison(cda
, sizeof(cda
), u
, "MemoryMin"),
442 prefix
, c
->memory_low
, format_cgroup_memory_limit_comparison(cdb
, sizeof(cdb
), u
, "MemoryLow"),
443 prefix
, c
->memory_high
, format_cgroup_memory_limit_comparison(cdc
, sizeof(cdc
), u
, "MemoryHigh"),
444 prefix
, c
->memory_max
, format_cgroup_memory_limit_comparison(cdd
, sizeof(cdd
), u
, "MemoryMax"),
445 prefix
, c
->memory_swap_max
, format_cgroup_memory_limit_comparison(cde
, sizeof(cde
), u
, "MemorySwapMax"),
446 prefix
, c
->memory_limit
,
447 prefix
, tasks_max_resolve(&c
->tasks_max
),
448 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
449 prefix
, strempty(disable_controllers_str
),
450 prefix
, yes_no(c
->delegate
),
451 prefix
, managed_oom_mode_to_string(c
->moom_swap
),
452 prefix
, managed_oom_mode_to_string(c
->moom_mem_pressure
),
453 prefix
, c
->moom_mem_pressure_limit
);
456 _cleanup_free_
char *t
= NULL
;
458 (void) cg_mask_to_string(c
->delegate_controllers
, &t
);
460 fprintf(f
, "%sDelegateControllers: %s\n",
465 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
467 "%sDeviceAllow: %s %s%s%s\n",
470 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
472 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
474 "%sIODeviceWeight: %s %" PRIu64
"\n",
479 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
481 "%sIODeviceLatencyTargetSec: %s %s\n",
484 format_timespan(q
, sizeof(q
), l
->target_usec
, 1));
486 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
487 char buf
[FORMAT_BYTES_MAX
];
488 CGroupIOLimitType type
;
490 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
491 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
495 cgroup_io_limit_type_to_string(type
),
497 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
500 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
502 "%sBlockIODeviceWeight: %s %" PRIu64
,
507 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
508 char buf
[FORMAT_BYTES_MAX
];
510 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
512 "%sBlockIOReadBandwidth: %s %s\n",
515 format_bytes(buf
, sizeof(buf
), b
->rbps
));
516 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
518 "%sBlockIOWriteBandwidth: %s %s\n",
521 format_bytes(buf
, sizeof(buf
), b
->wbps
));
524 LIST_FOREACH(items
, iaai
, c
->ip_address_allow
) {
525 _cleanup_free_
char *k
= NULL
;
527 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
528 fprintf(f
, "%sIPAddressAllow: %s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
531 LIST_FOREACH(items
, iaai
, c
->ip_address_deny
) {
532 _cleanup_free_
char *k
= NULL
;
534 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
535 fprintf(f
, "%sIPAddressDeny: %s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
538 STRV_FOREACH(path
, c
->ip_filters_ingress
)
539 fprintf(f
, "%sIPIngressFilterPath: %s\n", prefix
, *path
);
541 STRV_FOREACH(path
, c
->ip_filters_egress
)
542 fprintf(f
, "%sIPEgressFilterPath: %s\n", prefix
, *path
);
545 int cgroup_add_device_allow(CGroupContext
*c
, const char *dev
, const char *mode
) {
546 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
547 _cleanup_free_
char *d
= NULL
;
551 assert(isempty(mode
) || in_charset(mode
, "rwm"));
553 a
= new(CGroupDeviceAllow
, 1);
561 *a
= (CGroupDeviceAllow
) {
563 .r
= isempty(mode
) || strchr(mode
, 'r'),
564 .w
= isempty(mode
) || strchr(mode
, 'w'),
565 .m
= isempty(mode
) || strchr(mode
, 'm'),
568 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
574 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
575 uint64_t unit_get_ancestor_##entry(Unit *u) { \
578 /* 1. Is entry set in this unit? If so, use that. \
579 * 2. Is the default for this entry set in any \
580 * ancestor? If so, use that. \
581 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
585 c = unit_get_cgroup_context(u); \
586 if (c && c->entry##_set) \
589 while ((u = UNIT_DEREF(u->slice))) { \
590 c = unit_get_cgroup_context(u); \
591 if (c && c->default_##entry##_set) \
592 return c->default_##entry; \
595 /* We've reached the root, but nobody had default for \
596 * this entry set, so set it to the kernel default. */ \
597 return CGROUP_LIMIT_MIN; \
600 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
601 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
603 static void cgroup_xattr_apply(Unit
*u
) {
604 char ids
[SD_ID128_STRING_MAX
];
609 if (!MANAGER_IS_SYSTEM(u
->manager
))
612 if (!sd_id128_is_null(u
->invocation_id
)) {
613 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
614 "trusted.invocation_id",
615 sd_id128_to_string(u
->invocation_id
, ids
), 32,
618 log_unit_debug_errno(u
, r
, "Failed to set invocation ID on control group %s, ignoring: %m", u
->cgroup_path
);
621 if (unit_cgroup_delegate(u
)) {
622 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
627 log_unit_debug_errno(u
, r
, "Failed to set delegate flag on control group %s, ignoring: %m", u
->cgroup_path
);
629 r
= cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "trusted.delegate");
631 log_unit_debug_errno(u
, r
, "Failed to remove delegate flag on control group %s, ignoring: %m", u
->cgroup_path
);
635 static int lookup_block_device(const char *p
, dev_t
*ret
) {
643 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
644 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
647 if (stat(p
, &st
) < 0)
648 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
654 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
657 return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK
),
658 "Device node '%s' is a character device, but block device needed.", p
);
661 else if (major(dev
) != 0)
662 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
664 /* If this is btrfs, getting the backing block device is a bit harder */
665 r
= btrfs_get_block_device(p
, ret
);
667 return log_warning_errno(SYNTHETIC_ERRNO(ENODEV
),
668 "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
670 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
673 /* If this is a LUKS/DM device, recursively try to get the originating block device */
674 while (block_get_originating(*ret
, ret
) > 0);
676 /* If this is a partition, try to get the originating block device */
677 (void) block_get_whole_disk(*ret
, ret
);
681 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
682 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
683 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
686 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
687 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
688 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
691 static uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
692 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
693 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
694 return c
->startup_cpu_weight
;
695 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
696 return c
->cpu_weight
;
698 return CGROUP_WEIGHT_DEFAULT
;
701 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
702 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
703 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
704 return c
->startup_cpu_shares
;
705 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
706 return c
->cpu_shares
;
708 return CGROUP_CPU_SHARES_DEFAULT
;
711 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
712 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
713 * need to be higher than that boundary. quota is specified in USecPerSec.
714 * Additionally, period must be at most max_period. */
717 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
720 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
723 if (quota
== USEC_INFINITY
)
724 /* Always use default period for infinity quota. */
725 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
727 if (period
== USEC_INFINITY
)
728 /* Default period was requested. */
729 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
731 /* Clamp to interval [1ms, 1s] */
732 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
734 if (new_period
!= period
) {
735 char v
[FORMAT_TIMESPAN_MAX
];
736 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
,
737 "Clamping CPU interval for cpu.max: period is now %s",
738 format_timespan(v
, sizeof(v
), new_period
, 1));
739 u
->warned_clamping_cpu_quota_period
= true;
745 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
746 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
748 xsprintf(buf
, "%" PRIu64
"\n", weight
);
749 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
752 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
753 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
755 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
756 if (quota
!= USEC_INFINITY
)
757 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
758 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
760 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
761 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
764 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
765 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
767 xsprintf(buf
, "%" PRIu64
"\n", shares
);
768 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
771 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
772 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
774 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
776 xsprintf(buf
, USEC_FMT
"\n", period
);
777 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
779 if (quota
!= USEC_INFINITY
) {
780 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
781 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
783 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
786 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
787 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
788 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
791 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
792 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
793 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
796 static void cgroup_apply_unified_cpuset(Unit
*u
, const CPUSet
*cpus
, const char *name
) {
797 _cleanup_free_
char *buf
= NULL
;
799 buf
= cpu_set_to_range_string(cpus
);
805 (void) set_attribute_and_warn(u
, "cpuset", name
, buf
);
808 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
809 return c
->io_accounting
||
810 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
811 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
812 c
->io_device_weights
||
813 c
->io_device_latencies
||
817 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
818 return c
->blockio_accounting
||
819 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
820 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
821 c
->blockio_device_weights
||
822 c
->blockio_device_bandwidths
;
825 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
826 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
827 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
828 return c
->startup_io_weight
;
829 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
832 return CGROUP_WEIGHT_DEFAULT
;
835 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
836 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
837 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
838 return c
->startup_blockio_weight
;
839 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
840 return c
->blockio_weight
;
842 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
845 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
846 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
847 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
850 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
851 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
852 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
855 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
856 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
860 r
= lookup_block_device(dev_path
, &dev
);
864 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
865 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
868 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
869 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
873 r
= lookup_block_device(dev_path
, &dev
);
877 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
878 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
881 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
882 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
886 r
= lookup_block_device(dev_path
, &dev
);
890 if (target
!= USEC_INFINITY
)
891 xsprintf(buf
, "%u:%u target=%" PRIu64
"\n", major(dev
), minor(dev
), target
);
893 xsprintf(buf
, "%u:%u target=max\n", major(dev
), minor(dev
));
895 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
898 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
899 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
900 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
901 CGroupIOLimitType type
;
905 r
= lookup_block_device(dev_path
, &dev
);
909 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
910 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
911 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
913 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
915 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
916 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
917 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
918 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
921 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
922 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
926 r
= lookup_block_device(dev_path
, &dev
);
930 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
931 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
933 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
934 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
937 static bool unit_has_unified_memory_config(Unit
*u
) {
942 c
= unit_get_cgroup_context(u
);
945 return unit_get_ancestor_memory_min(u
) > 0 || unit_get_ancestor_memory_low(u
) > 0 ||
946 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
||
947 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
;
950 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
951 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
953 if (v
!= CGROUP_LIMIT_MAX
)
954 xsprintf(buf
, "%" PRIu64
"\n", v
);
956 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
959 static void cgroup_apply_firewall(Unit
*u
) {
962 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
964 if (bpf_firewall_compile(u
) < 0)
967 (void) bpf_firewall_load_custom(u
);
968 (void) bpf_firewall_install(u
);
971 static int cgroup_apply_devices(Unit
*u
) {
972 _cleanup_(bpf_program_unrefp
) BPFProgram
*prog
= NULL
;
975 CGroupDeviceAllow
*a
;
976 CGroupDevicePolicy policy
;
979 assert_se(c
= unit_get_cgroup_context(u
));
980 assert_se(path
= u
->cgroup_path
);
982 policy
= c
->device_policy
;
984 if (cg_all_unified() > 0) {
985 r
= bpf_devices_cgroup_init(&prog
, policy
, c
->device_allow
);
987 return log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
990 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore
993 if (c
->device_allow
|| policy
!= CGROUP_DEVICE_POLICY_AUTO
)
994 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
996 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
998 log_unit_full_errno(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
999 "Failed to reset devices.allow/devices.deny: %m");
1002 bool allow_list_static
= policy
== CGROUP_DEVICE_POLICY_CLOSED
||
1003 (policy
== CGROUP_DEVICE_POLICY_AUTO
&& c
->device_allow
);
1004 if (allow_list_static
)
1005 (void) bpf_devices_allow_list_static(prog
, path
);
1007 bool any
= allow_list_static
;
1008 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1022 if (path_startswith(a
->path
, "/dev/"))
1023 r
= bpf_devices_allow_list_device(prog
, path
, a
->path
, acc
);
1024 else if ((val
= startswith(a
->path
, "block-")))
1025 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'b', acc
);
1026 else if ((val
= startswith(a
->path
, "char-")))
1027 r
= bpf_devices_allow_list_major(prog
, path
, val
, 'c', acc
);
1029 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1038 log_unit_warning_errno(u
, SYNTHETIC_ERRNO(ENODEV
), "No devices matched by device filter.");
1040 /* The kernel verifier would reject a program we would build with the normal intro and outro
1041 but no allow-listing rules (outro would contain an unreachable instruction for successful
1043 policy
= CGROUP_DEVICE_POLICY_STRICT
;
1046 r
= bpf_devices_apply_policy(prog
, policy
, any
, path
, &u
->bpf_device_control_installed
);
1048 static bool warned
= false;
1050 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1051 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1052 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1053 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1060 static void cgroup_context_apply(
1062 CGroupMask apply_mask
,
1063 ManagerState state
) {
1067 bool is_host_root
, is_local_root
;
1072 /* Nothing to do? Exit early! */
1073 if (apply_mask
== 0)
1076 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
1077 * attributes should only be managed for cgroups further down the tree. */
1078 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1079 is_host_root
= unit_has_host_root_cgroup(u
);
1081 assert_se(c
= unit_get_cgroup_context(u
));
1082 assert_se(path
= u
->cgroup_path
);
1084 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
1087 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
1088 * then), and missing cgroups, i.e. EROFS and ENOENT. */
1090 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
1091 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
1092 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
1093 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
1094 * we couldn't even write to them if we wanted to). */
1095 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
1097 if (cg_all_unified() > 0) {
1100 if (cgroup_context_has_cpu_weight(c
))
1101 weight
= cgroup_context_cpu_weight(c
, state
);
1102 else if (cgroup_context_has_cpu_shares(c
)) {
1105 shares
= cgroup_context_cpu_shares(c
, state
);
1106 weight
= cgroup_cpu_shares_to_weight(shares
);
1108 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
1109 shares
, weight
, path
);
1111 weight
= CGROUP_WEIGHT_DEFAULT
;
1113 cgroup_apply_unified_cpu_weight(u
, weight
);
1114 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1119 if (cgroup_context_has_cpu_weight(c
)) {
1122 weight
= cgroup_context_cpu_weight(c
, state
);
1123 shares
= cgroup_cpu_weight_to_shares(weight
);
1125 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
1126 weight
, shares
, path
);
1127 } else if (cgroup_context_has_cpu_shares(c
))
1128 shares
= cgroup_context_cpu_shares(c
, state
);
1130 shares
= CGROUP_CPU_SHARES_DEFAULT
;
1132 cgroup_apply_legacy_cpu_shares(u
, shares
);
1133 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1137 if ((apply_mask
& CGROUP_MASK_CPUSET
) && !is_local_root
) {
1138 cgroup_apply_unified_cpuset(u
, &c
->cpuset_cpus
, "cpuset.cpus");
1139 cgroup_apply_unified_cpuset(u
, &c
->cpuset_mems
, "cpuset.mems");
1142 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
1143 * controller), and in case of containers we want to leave control of these attributes to the container manager
1144 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
1145 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
1146 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
1147 bool has_io
, has_blockio
;
1150 has_io
= cgroup_context_has_io_config(c
);
1151 has_blockio
= cgroup_context_has_blockio_config(c
);
1154 weight
= cgroup_context_io_weight(c
, state
);
1155 else if (has_blockio
) {
1156 uint64_t blkio_weight
;
1158 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
1159 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
1161 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
1162 blkio_weight
, weight
);
1164 weight
= CGROUP_WEIGHT_DEFAULT
;
1166 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1167 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1169 /* FIXME: drop this when distro kernels properly support BFQ through "io.weight"
1170 * See also: https://github.com/systemd/systemd/pull/13335 */
1171 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1172 (void) set_attribute_and_warn(u
, "io", "io.bfq.weight", buf
);
1175 CGroupIODeviceLatency
*latency
;
1176 CGroupIODeviceLimit
*limit
;
1177 CGroupIODeviceWeight
*w
;
1179 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
1180 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
1182 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
1183 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
1185 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
1186 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
1188 } else if (has_blockio
) {
1189 CGroupBlockIODeviceWeight
*w
;
1190 CGroupBlockIODeviceBandwidth
*b
;
1192 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
1193 weight
= cgroup_weight_blkio_to_io(w
->weight
);
1195 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
1196 w
->weight
, weight
, w
->path
);
1198 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
1201 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
1202 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
1203 CGroupIOLimitType type
;
1205 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1206 limits
[type
] = cgroup_io_limit_defaults
[type
];
1208 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
1209 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
1211 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
1212 b
->rbps
, b
->wbps
, b
->path
);
1214 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
1219 if (apply_mask
& CGROUP_MASK_BLKIO
) {
1220 bool has_io
, has_blockio
;
1222 has_io
= cgroup_context_has_io_config(c
);
1223 has_blockio
= cgroup_context_has_blockio_config(c
);
1225 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
1226 * left to our container manager, too. */
1227 if (!is_local_root
) {
1228 char buf
[DECIMAL_STR_MAX(uint64_t)+1];
1234 io_weight
= cgroup_context_io_weight(c
, state
);
1235 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
1237 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
1239 } else if (has_blockio
)
1240 weight
= cgroup_context_blkio_weight(c
, state
);
1242 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
1244 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1245 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1248 CGroupIODeviceWeight
*w
;
1250 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
1251 weight
= cgroup_weight_io_to_blkio(w
->weight
);
1253 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
1254 w
->weight
, weight
, w
->path
);
1256 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
1258 } else if (has_blockio
) {
1259 CGroupBlockIODeviceWeight
*w
;
1261 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
1262 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
1266 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
1267 * roots, as there we want the container manager to handle it */
1268 if (is_host_root
|| !is_local_root
) {
1270 CGroupIODeviceLimit
*l
;
1272 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
1273 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
1274 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
1276 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
1278 } else if (has_blockio
) {
1279 CGroupBlockIODeviceBandwidth
*b
;
1281 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
1282 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
1287 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
1288 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
1289 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
1290 * write to this if we wanted to.) */
1291 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
1293 if (cg_all_unified() > 0) {
1294 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
;
1296 if (unit_has_unified_memory_config(u
)) {
1297 max
= c
->memory_max
;
1298 swap_max
= c
->memory_swap_max
;
1300 max
= c
->memory_limit
;
1302 if (max
!= CGROUP_LIMIT_MAX
)
1303 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
1306 cgroup_apply_unified_memory_limit(u
, "memory.min", unit_get_ancestor_memory_min(u
));
1307 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
1308 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
1309 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
1310 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
1312 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
1315 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1318 if (unit_has_unified_memory_config(u
)) {
1319 val
= c
->memory_max
;
1320 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIi64
" as MemoryLimit=", val
);
1322 val
= c
->memory_limit
;
1324 if (val
== CGROUP_LIMIT_MAX
)
1325 strncpy(buf
, "-1\n", sizeof(buf
));
1327 xsprintf(buf
, "%" PRIu64
"\n", val
);
1329 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
1333 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
1334 * containers, where we leave this to the manager */
1335 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
1336 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
))
1337 (void) cgroup_apply_devices(u
);
1339 if (apply_mask
& CGROUP_MASK_PIDS
) {
1342 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1343 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1344 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1345 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1346 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1347 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1348 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1349 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1350 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1351 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1352 * which is desirable so that there's an official way to release control of the sysctl from
1353 * systemd: set the limit to unbounded and reload. */
1355 if (tasks_max_isset(&c
->tasks_max
)) {
1356 u
->manager
->sysctl_pid_max_changed
= true;
1357 r
= procfs_tasks_set_limit(tasks_max_resolve(&c
->tasks_max
));
1358 } else if (u
->manager
->sysctl_pid_max_changed
)
1359 r
= procfs_tasks_set_limit(TASKS_MAX
);
1363 log_unit_full_errno(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
1364 "Failed to write to tasks limit sysctls: %m");
1367 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
1368 * leave it for the container manager. */
1369 if (!is_local_root
) {
1370 if (tasks_max_isset(&c
->tasks_max
)) {
1371 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1373 xsprintf(buf
, "%" PRIu64
"\n", tasks_max_resolve(&c
->tasks_max
));
1374 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
1376 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
1380 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
1381 cgroup_apply_firewall(u
);
1384 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
1389 c
= unit_get_cgroup_context(u
);
1393 if (c
->ip_accounting
||
1394 c
->ip_address_allow
||
1395 c
->ip_address_deny
||
1396 c
->ip_filters_ingress
||
1397 c
->ip_filters_egress
)
1400 /* If any parent slice has an IP access list defined, it applies too */
1401 for (p
= UNIT_DEREF(u
->slice
); p
; p
= UNIT_DEREF(p
->slice
)) {
1402 c
= unit_get_cgroup_context(p
);
1406 if (c
->ip_address_allow
||
1414 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
1415 CGroupMask mask
= 0;
1420 c
= unit_get_cgroup_context(u
);
1424 /* Figure out which controllers we need, based on the cgroup context object */
1426 if (c
->cpu_accounting
)
1427 mask
|= get_cpu_accounting_mask();
1429 if (cgroup_context_has_cpu_weight(c
) ||
1430 cgroup_context_has_cpu_shares(c
) ||
1431 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
1432 mask
|= CGROUP_MASK_CPU
;
1434 if (c
->cpuset_cpus
.set
|| c
->cpuset_mems
.set
)
1435 mask
|= CGROUP_MASK_CPUSET
;
1437 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
1438 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1440 if (c
->memory_accounting
||
1441 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
1442 unit_has_unified_memory_config(u
))
1443 mask
|= CGROUP_MASK_MEMORY
;
1445 if (c
->device_allow
||
1446 c
->device_policy
!= CGROUP_DEVICE_POLICY_AUTO
)
1447 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
1449 if (c
->tasks_accounting
||
1450 tasks_max_isset(&c
->tasks_max
))
1451 mask
|= CGROUP_MASK_PIDS
;
1453 return CGROUP_MASK_EXTEND_JOINED(mask
);
1456 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
1457 CGroupMask mask
= 0;
1459 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
1462 if (unit_get_needs_bpf_firewall(u
))
1463 mask
|= CGROUP_MASK_BPF_FIREWALL
;
1468 CGroupMask
unit_get_own_mask(Unit
*u
) {
1471 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
1472 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
1474 if (u
->load_state
!= UNIT_LOADED
)
1477 c
= unit_get_cgroup_context(u
);
1481 return unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
);
1484 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
1487 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1488 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1490 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1492 if (!unit_cgroup_delegate(u
))
1495 if (cg_all_unified() <= 0) {
1498 e
= unit_get_exec_context(u
);
1499 if (e
&& !exec_context_maintains_privileges(e
))
1503 assert_se(c
= unit_get_cgroup_context(u
));
1504 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
1507 static CGroupMask
unit_get_subtree_mask(Unit
*u
) {
1509 /* Returns the mask of this subtree, meaning of the group
1510 * itself and its children. */
1512 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
1515 CGroupMask
unit_get_members_mask(Unit
*u
) {
1518 /* Returns the mask of controllers all of the unit's children require, merged */
1520 if (u
->cgroup_members_mask_valid
)
1521 return u
->cgroup_members_mask
; /* Use cached value if possible */
1523 u
->cgroup_members_mask
= 0;
1525 if (u
->type
== UNIT_SLICE
) {
1529 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
])
1530 if (UNIT_DEREF(member
->slice
) == u
)
1531 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
1534 u
->cgroup_members_mask_valid
= true;
1535 return u
->cgroup_members_mask
;
1538 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
1541 /* Returns the mask of controllers all of the unit's siblings
1542 * require, i.e. the members mask of the unit's parent slice
1543 * if there is one. */
1545 if (UNIT_ISSET(u
->slice
))
1546 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
1548 return unit_get_subtree_mask(u
); /* we are the top-level slice */
1551 static CGroupMask
unit_get_disable_mask(Unit
*u
) {
1554 c
= unit_get_cgroup_context(u
);
1558 return c
->disable_controllers
;
1561 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
1565 mask
= unit_get_disable_mask(u
);
1567 /* Returns the mask of controllers which are marked as forcibly
1568 * disabled in any ancestor unit or the unit in question. */
1570 if (UNIT_ISSET(u
->slice
))
1571 mask
|= unit_get_ancestor_disable_mask(UNIT_DEREF(u
->slice
));
1576 CGroupMask
unit_get_target_mask(Unit
*u
) {
1579 /* This returns the cgroup mask of all controllers to enable
1580 * for a specific cgroup, i.e. everything it needs itself,
1581 * plus all that its children need, plus all that its siblings
1582 * need. This is primarily useful on the legacy cgroup
1583 * hierarchy, where we need to duplicate each cgroup in each
1584 * hierarchy that shall be enabled for it. */
1586 mask
= unit_get_own_mask(u
) | unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
1588 if (mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
1589 emit_bpf_firewall_warning(u
);
1591 mask
&= u
->manager
->cgroup_supported
;
1592 mask
&= ~unit_get_ancestor_disable_mask(u
);
1597 CGroupMask
unit_get_enable_mask(Unit
*u
) {
1600 /* This returns the cgroup mask of all controllers to enable
1601 * for the children of a specific cgroup. This is primarily
1602 * useful for the unified cgroup hierarchy, where each cgroup
1603 * controls which controllers are enabled for its children. */
1605 mask
= unit_get_members_mask(u
);
1606 mask
&= u
->manager
->cgroup_supported
;
1607 mask
&= ~unit_get_ancestor_disable_mask(u
);
1612 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
1615 /* Recurse invalidate the member masks cache all the way up the tree */
1616 u
->cgroup_members_mask_valid
= false;
1618 if (UNIT_ISSET(u
->slice
))
1619 unit_invalidate_cgroup_members_masks(UNIT_DEREF(u
->slice
));
1622 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
1624 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1628 if (u
->cgroup_path
&&
1629 u
->cgroup_realized
&&
1630 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
1631 return u
->cgroup_path
;
1633 u
= UNIT_DEREF(u
->slice
);
1639 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1640 /* If not realized at all, migrate to root ("").
1641 * It may happen if we're upgrading from older version that didn't clean up.
1643 return strempty(unit_get_realized_cgroup_path(userdata
, mask
));
1646 char *unit_default_cgroup_path(const Unit
*u
) {
1647 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1652 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1653 return strdup(u
->manager
->cgroup_root
);
1655 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1656 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1661 escaped
= cg_escape(u
->id
);
1665 return path_join(empty_to_root(u
->manager
->cgroup_root
), slice
, escaped
);
1668 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1669 _cleanup_free_
char *p
= NULL
;
1674 if (streq_ptr(u
->cgroup_path
, path
))
1684 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1689 unit_release_cgroup(u
);
1690 u
->cgroup_path
= TAKE_PTR(p
);
1695 int unit_watch_cgroup(Unit
*u
) {
1696 _cleanup_free_
char *events
= NULL
;
1701 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
1702 * cgroupv2 is available. */
1704 if (!u
->cgroup_path
)
1707 if (u
->cgroup_control_inotify_wd
>= 0)
1710 /* Only applies to the unified hierarchy */
1711 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1713 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1717 /* No point in watch the top-level slice, it's never going to run empty. */
1718 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1721 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
1725 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1729 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1730 if (u
->cgroup_control_inotify_wd
< 0) {
1732 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1733 * is not an error */
1736 return log_unit_error_errno(u
, errno
, "Failed to add control inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1739 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
1741 return log_unit_error_errno(u
, r
, "Failed to add control inotify watch descriptor to hash map: %m");
1746 int unit_watch_cgroup_memory(Unit
*u
) {
1747 _cleanup_free_
char *events
= NULL
;
1753 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
1754 * cgroupv2 is available. */
1756 if (!u
->cgroup_path
)
1759 c
= unit_get_cgroup_context(u
);
1763 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
1764 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
1766 if (!c
->memory_accounting
)
1769 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
1770 * we also don't want to generate a log message for each parent cgroup of a process. */
1771 if (u
->type
== UNIT_SLICE
)
1774 if (u
->cgroup_memory_inotify_wd
>= 0)
1777 /* Only applies to the unified hierarchy */
1778 r
= cg_all_unified();
1780 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
1784 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
1788 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
1792 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1793 if (u
->cgroup_memory_inotify_wd
< 0) {
1795 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1796 * is not an error */
1799 return log_unit_error_errno(u
, errno
, "Failed to add memory inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1802 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
1804 return log_unit_error_errno(u
, r
, "Failed to add memory inotify watch descriptor to hash map: %m");
1809 int unit_pick_cgroup_path(Unit
*u
) {
1810 _cleanup_free_
char *path
= NULL
;
1818 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1821 path
= unit_default_cgroup_path(u
);
1825 r
= unit_set_cgroup_path(u
, path
);
1827 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1829 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1834 static int unit_update_cgroup(
1836 CGroupMask target_mask
,
1837 CGroupMask enable_mask
,
1838 ManagerState state
) {
1840 bool created
, is_root_slice
;
1841 CGroupMask migrate_mask
= 0;
1846 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1849 /* Figure out our cgroup path */
1850 r
= unit_pick_cgroup_path(u
);
1854 /* First, create our own group */
1855 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
1857 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
1860 /* Start watching it */
1861 (void) unit_watch_cgroup(u
);
1862 (void) unit_watch_cgroup_memory(u
);
1865 /* For v2 we preserve enabled controllers in delegated units, adjust others,
1866 * for v1 we figure out which controller hierarchies need migration. */
1867 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
1868 CGroupMask result_mask
= 0;
1870 /* Enable all controllers we need */
1871 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
1873 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
1875 /* Remember what's actually enabled now */
1876 u
->cgroup_enabled_mask
= result_mask
;
1878 migrate_mask
= u
->cgroup_realized_mask
^ target_mask
;
1881 /* Keep track that this is now realized */
1882 u
->cgroup_realized
= true;
1883 u
->cgroup_realized_mask
= target_mask
;
1885 /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling).
1887 * Unnecessary controller cgroups are trimmed (after emptied by upward migration).
1888 * We perform migration also with whole slices for cases when users don't care about leave
1889 * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing
1892 if (cg_all_unified() == 0) {
1893 r
= cg_migrate_v1_controllers(u
->manager
->cgroup_supported
, migrate_mask
, u
->cgroup_path
, migrate_callback
, u
);
1895 log_unit_warning_errno(u
, r
, "Failed to migrate controller cgroups from %s, ignoring: %m", u
->cgroup_path
);
1897 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1898 r
= cg_trim_v1_controllers(u
->manager
->cgroup_supported
, ~target_mask
, u
->cgroup_path
, !is_root_slice
);
1900 log_unit_warning_errno(u
, r
, "Failed to delete controller cgroups %s, ignoring: %m", u
->cgroup_path
);
1903 /* Set attributes */
1904 cgroup_context_apply(u
, target_mask
, state
);
1905 cgroup_xattr_apply(u
);
1910 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
1911 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1917 if (MANAGER_IS_SYSTEM(u
->manager
))
1920 if (!u
->manager
->system_bus
)
1923 if (!u
->cgroup_path
)
1926 /* Determine this unit's cgroup path relative to our cgroup root */
1927 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
1931 pp
= strjoina("/", pp
, suffix_path
);
1932 path_simplify(pp
, false);
1934 r
= sd_bus_call_method(u
->manager
->system_bus
,
1935 "org.freedesktop.systemd1",
1936 "/org/freedesktop/systemd1",
1937 "org.freedesktop.systemd1.Manager",
1938 "AttachProcessesToUnit",
1941 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
1943 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
1948 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
1949 CGroupMask delegated_mask
;
1956 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1959 if (set_isempty(pids
))
1962 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
1963 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
1964 r
= bpf_firewall_load_custom(u
);
1968 r
= unit_realize_cgroup(u
);
1972 if (isempty(suffix_path
))
1975 p
= prefix_roota(u
->cgroup_path
, suffix_path
);
1977 delegated_mask
= unit_get_delegate_mask(u
);
1980 SET_FOREACH(pidp
, pids
) {
1981 pid_t pid
= PTR_TO_PID(pidp
);
1984 /* First, attach the PID to the main cgroup hierarchy */
1985 q
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
);
1987 log_unit_debug_errno(u
, q
, "Couldn't move process " PID_FMT
" to requested cgroup '%s': %m", pid
, p
);
1989 if (MANAGER_IS_USER(u
->manager
) && ERRNO_IS_PRIVILEGE(q
)) {
1992 /* If we are in a user instance, and we can't move the process ourselves due to
1993 * permission problems, let's ask the system instance about it instead. Since it's more
1994 * privileged it might be able to move the process across the leaves of a subtree who's
1995 * top node is not owned by us. */
1997 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
, suffix_path
);
1999 log_unit_debug_errno(u
, z
, "Couldn't move process " PID_FMT
" to requested cgroup '%s' via the system bus either: %m", pid
, p
);
2001 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
2005 r
= q
; /* Remember first error */
2010 q
= cg_all_unified();
2016 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
2017 * innermost realized one */
2019 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2020 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2021 const char *realized
;
2023 if (!(u
->manager
->cgroup_supported
& bit
))
2026 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
2027 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
2028 q
= cg_attach(cgroup_controller_to_string(c
), p
, pid
);
2030 continue; /* Success! */
2032 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",
2033 pid
, p
, cgroup_controller_to_string(c
));
2036 /* So this controller is either not delegate or realized, or something else weird happened. In
2037 * that case let's attach the PID at least to the closest cgroup up the tree that is
2039 realized
= unit_get_realized_cgroup_path(u
, bit
);
2041 continue; /* Not even realized in the root slice? Then let's not bother */
2043 q
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
);
2045 log_unit_debug_errno(u
, q
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
2046 pid
, realized
, cgroup_controller_to_string(c
));
2053 static bool unit_has_mask_realized(
2055 CGroupMask target_mask
,
2056 CGroupMask enable_mask
) {
2060 /* Returns true if this unit is fully realized. We check four things:
2062 * 1. Whether the cgroup was created at all
2063 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
2064 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
2065 * 4. Whether the invalidation mask is currently zero
2067 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
2068 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
2069 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
2070 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
2071 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
2072 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
2073 * simply don't matter. */
2075 return u
->cgroup_realized
&&
2076 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
2077 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
2078 u
->cgroup_invalidated_mask
== 0;
2081 static bool unit_has_mask_disables_realized(
2083 CGroupMask target_mask
,
2084 CGroupMask enable_mask
) {
2088 /* Returns true if all controllers which should be disabled are indeed disabled.
2090 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
2091 * already removed. */
2093 return !u
->cgroup_realized
||
2094 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
2095 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
2098 static bool unit_has_mask_enables_realized(
2100 CGroupMask target_mask
,
2101 CGroupMask enable_mask
) {
2105 /* Returns true if all controllers which should be enabled are indeed enabled.
2107 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
2108 * we want to add is already added. */
2110 return u
->cgroup_realized
&&
2111 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
2112 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
2115 static void unit_add_to_cgroup_realize_queue(Unit
*u
) {
2118 if (u
->in_cgroup_realize_queue
)
2121 LIST_APPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2122 u
->in_cgroup_realize_queue
= true;
2125 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
2128 if (!u
->in_cgroup_realize_queue
)
2131 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2132 u
->in_cgroup_realize_queue
= false;
2135 /* Controllers can only be enabled breadth-first, from the root of the
2136 * hierarchy downwards to the unit in question. */
2137 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
2138 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2143 /* First go deal with this unit's parent, or we won't be able to enable
2144 * any new controllers at this layer. */
2145 if (UNIT_ISSET(u
->slice
)) {
2146 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2151 target_mask
= unit_get_target_mask(u
);
2152 enable_mask
= unit_get_enable_mask(u
);
2154 /* We can only enable in this direction, don't try to disable anything.
2156 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
2159 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
2160 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
2162 return unit_update_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
2165 /* Controllers can only be disabled depth-first, from the leaves of the
2166 * hierarchy upwards to the unit in question. */
2167 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
2173 if (u
->type
!= UNIT_SLICE
)
2176 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
]) {
2177 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2180 if (UNIT_DEREF(m
->slice
) != u
)
2183 /* The cgroup for this unit might not actually be fully
2184 * realised yet, in which case it isn't holding any controllers
2186 if (!m
->cgroup_realized
)
2189 /* We must disable those below us first in order to release the
2191 if (m
->type
== UNIT_SLICE
)
2192 (void) unit_realize_cgroup_now_disable(m
, state
);
2194 target_mask
= unit_get_target_mask(m
);
2195 enable_mask
= unit_get_enable_mask(m
);
2197 /* We can only disable in this direction, don't try to enable
2199 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
2202 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
2203 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
2205 r
= unit_update_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
2213 /* Check if necessary controllers and attributes for a unit are in place.
2215 * - If so, do nothing.
2216 * - If not, create paths, move processes over, and set attributes.
2218 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
2219 * a depth-first way. As such the process looks like this:
2221 * Suppose we have a cgroup hierarchy which looks like this:
2234 * 1. We want to realise cgroup "d" now.
2235 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
2236 * 3. cgroup "k" just started requesting the memory controller.
2238 * To make this work we must do the following in order:
2240 * 1. Disable CPU controller in k, j
2241 * 2. Disable CPU controller in d
2242 * 3. Enable memory controller in root
2243 * 4. Enable memory controller in a
2244 * 5. Enable memory controller in d
2245 * 6. Enable memory controller in k
2247 * Notice that we need to touch j in one direction, but not the other. We also
2248 * don't go beyond d when disabling -- it's up to "a" to get realized if it
2249 * wants to disable further. The basic rules are therefore:
2251 * - If you're disabling something, you need to realise all of the cgroups from
2252 * your recursive descendants to the root. This starts from the leaves.
2253 * - If you're enabling something, you need to realise from the root cgroup
2254 * downwards, but you don't need to iterate your recursive descendants.
2256 * Returns 0 on success and < 0 on failure. */
2257 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
2258 CGroupMask target_mask
, enable_mask
;
2263 unit_remove_from_cgroup_realize_queue(u
);
2265 target_mask
= unit_get_target_mask(u
);
2266 enable_mask
= unit_get_enable_mask(u
);
2268 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
2271 /* Disable controllers below us, if there are any */
2272 r
= unit_realize_cgroup_now_disable(u
, state
);
2276 /* Enable controllers above us, if there are any */
2277 if (UNIT_ISSET(u
->slice
)) {
2278 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2283 /* Now actually deal with the cgroup we were trying to realise and set attributes */
2284 r
= unit_update_cgroup(u
, target_mask
, enable_mask
, state
);
2288 /* Now, reset the invalidation mask */
2289 u
->cgroup_invalidated_mask
= 0;
2293 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
2301 state
= manager_state(m
);
2303 while ((i
= m
->cgroup_realize_queue
)) {
2304 assert(i
->in_cgroup_realize_queue
);
2306 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
2307 /* Maybe things changed, and the unit is not actually active anymore? */
2308 unit_remove_from_cgroup_realize_queue(i
);
2312 r
= unit_realize_cgroup_now(i
, state
);
2314 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
2322 void unit_add_family_to_cgroup_realize_queue(Unit
*u
) {
2324 assert(u
->type
== UNIT_SLICE
);
2326 /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all
2329 * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes
2330 * very weird if two units that own processes reside in the same slice, but one is realized in the
2331 * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does
2332 * not), because that means individual processes need to be scheduled against whole cgroups. Let's
2333 * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1
2334 * controller hierarchies too (if unit requires the controller to be realized).
2336 * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date
2343 /* Children of u likely changed when we're called */
2344 u
->cgroup_members_mask_valid
= false;
2346 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
]) {
2347 /* Skip units that have a dependency on the slice but aren't actually in it. */
2348 if (UNIT_DEREF(m
->slice
) != u
)
2351 /* No point in doing cgroup application for units without active processes. */
2352 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
2355 /* We only enqueue siblings if they were realized once at least, in the main
2357 if (!m
->cgroup_realized
)
2360 /* If the unit doesn't need any new controllers and has current ones realized, it
2361 * doesn't need any changes. */
2362 if (unit_has_mask_realized(m
,
2363 unit_get_target_mask(m
),
2364 unit_get_enable_mask(m
)))
2367 unit_add_to_cgroup_realize_queue(m
);
2370 /* Parent comes after children */
2371 unit_add_to_cgroup_realize_queue(u
);
2372 } while ((u
= UNIT_DEREF(u
->slice
)));
2375 int unit_realize_cgroup(Unit
*u
) {
2378 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2381 /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all
2382 * parents, but there's more actually: for the weight-based controllers we also need to make sure
2383 * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the
2384 * other hand, when a controller is removed from realized set, it may become unnecessary in siblings
2385 * and ancestors and they should be (de)realized too.
2387 * This call will defer work on the siblings and derealized ancestors to the next event loop
2388 * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */
2390 if (UNIT_ISSET(u
->slice
))
2391 unit_add_family_to_cgroup_realize_queue(UNIT_DEREF(u
->slice
));
2393 /* And realize this one now (and apply the values) */
2394 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
2397 void unit_release_cgroup(Unit
*u
) {
2400 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
2401 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
2403 if (u
->cgroup_path
) {
2404 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
2405 u
->cgroup_path
= mfree(u
->cgroup_path
);
2408 if (u
->cgroup_control_inotify_wd
>= 0) {
2409 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
2410 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
);
2412 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
2413 u
->cgroup_control_inotify_wd
= -1;
2416 if (u
->cgroup_memory_inotify_wd
>= 0) {
2417 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
2418 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
);
2420 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
2421 u
->cgroup_memory_inotify_wd
= -1;
2425 void unit_prune_cgroup(Unit
*u
) {
2431 /* Removes the cgroup, if empty and possible, and stops watching it. */
2433 if (!u
->cgroup_path
)
2436 (void) unit_get_cpu_usage(u
, NULL
); /* Cache the last CPU usage value before we destroy the cgroup */
2438 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2440 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
2442 /* One reason we could have failed here is, that the cgroup still contains a process.
2443 * However, if the cgroup becomes removable at a later time, it might be removed when
2444 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
2445 * that the cgroup is still realized the next time it is started. Do not return early
2446 * on error, continue cleanup. */
2447 log_unit_full_errno(u
, r
== -EBUSY
? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
2452 unit_release_cgroup(u
);
2454 u
->cgroup_realized
= false;
2455 u
->cgroup_realized_mask
= 0;
2456 u
->cgroup_enabled_mask
= 0;
2458 u
->bpf_device_control_installed
= bpf_program_unref(u
->bpf_device_control_installed
);
2461 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
2462 _cleanup_fclose_
FILE *f
= NULL
;
2463 pid_t pid
= 0, npid
;
2469 if (!u
->cgroup_path
)
2472 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
2476 while (cg_read_pid(f
, &npid
) > 0) {
2481 if (pid_is_my_child(npid
) == 0)
2485 /* Dang, there's more than one daemonized PID
2486 in this group, so we don't know what process
2487 is the main process. */
2498 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
2499 _cleanup_closedir_
DIR *d
= NULL
;
2500 _cleanup_fclose_
FILE *f
= NULL
;
2506 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
2512 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
2513 r
= unit_watch_pid(u
, pid
, false);
2514 if (r
< 0 && ret
>= 0)
2518 if (r
< 0 && ret
>= 0)
2522 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
2529 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
2530 _cleanup_free_
char *p
= NULL
;
2532 p
= path_join(empty_to_root(path
), fn
);
2538 r
= unit_watch_pids_in_path(u
, p
);
2539 if (r
< 0 && ret
>= 0)
2543 if (r
< 0 && ret
>= 0)
2550 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
2555 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
2556 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
2557 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
2559 if (!u
->cgroup_path
)
2562 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2565 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
2568 if (!set_isempty(u
->pids
))
2571 unit_add_to_cgroup_empty_queue(u
);
2575 int unit_watch_all_pids(Unit
*u
) {
2580 /* Adds all PIDs from our cgroup to the set of PIDs we
2581 * watch. This is a fallback logic for cases where we do not
2582 * get reliable cgroup empty notifications: we try to use
2583 * SIGCHLD as replacement. */
2585 if (!u
->cgroup_path
)
2588 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2591 if (r
> 0) /* On unified we can use proper notifications */
2594 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
2597 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
2598 Manager
*m
= userdata
;
2605 u
= m
->cgroup_empty_queue
;
2609 assert(u
->in_cgroup_empty_queue
);
2610 u
->in_cgroup_empty_queue
= false;
2611 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
2613 if (m
->cgroup_empty_queue
) {
2614 /* More stuff queued, let's make sure we remain enabled */
2615 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2617 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
2620 unit_add_to_gc_queue(u
);
2622 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
2623 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
2628 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
2633 /* Note that there are four different ways how cgroup empty events reach us:
2635 * 1. On the unified hierarchy we get an inotify event on the cgroup
2637 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2639 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2641 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2642 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2644 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2645 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2646 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2647 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2648 * case for scope units). */
2650 if (u
->in_cgroup_empty_queue
)
2653 /* Let's verify that the cgroup is really empty */
2654 if (!u
->cgroup_path
)
2657 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2659 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", u
->cgroup_path
);
2665 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2666 u
->in_cgroup_empty_queue
= true;
2668 /* Trigger the defer event */
2669 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
2671 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
2674 static void unit_remove_from_cgroup_empty_queue(Unit
*u
) {
2677 if (!u
->in_cgroup_empty_queue
)
2680 LIST_REMOVE(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2681 u
->in_cgroup_empty_queue
= false;
2684 int unit_check_oomd_kill(Unit
*u
) {
2685 _cleanup_free_
char *value
= NULL
;
2690 if (!u
->cgroup_path
)
2693 r
= cg_all_unified();
2695 return log_unit_debug_errno(u
, r
, "Couldn't determine whether we are in all unified mode: %m");
2699 r
= cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "user.systemd_oomd_kill", &value
);
2700 if (r
< 0 && r
!= -ENODATA
)
2703 if (!isempty(value
)) {
2704 r
= safe_atou64(value
, &n
);
2709 increased
= n
> u
->managed_oom_kill_last
;
2710 u
->managed_oom_kill_last
= n
;
2716 log_struct(LOG_NOTICE
,
2717 "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR
,
2719 LOG_UNIT_INVOCATION_ID(u
),
2720 LOG_UNIT_MESSAGE(u
, "systemd-oomd killed %"PRIu64
" process(es) in this unit.", n
));
2725 int unit_check_oom(Unit
*u
) {
2726 _cleanup_free_
char *oom_kill
= NULL
;
2731 if (!u
->cgroup_path
)
2734 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
2736 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
2738 r
= safe_atou64(oom_kill
, &c
);
2740 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
2742 increased
= c
> u
->oom_kill_last
;
2743 u
->oom_kill_last
= c
;
2748 log_struct(LOG_NOTICE
,
2749 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
2751 LOG_UNIT_INVOCATION_ID(u
),
2752 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
2754 if (UNIT_VTABLE(u
)->notify_cgroup_oom
)
2755 UNIT_VTABLE(u
)->notify_cgroup_oom(u
);
2760 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
2761 Manager
*m
= userdata
;
2768 u
= m
->cgroup_oom_queue
;
2772 assert(u
->in_cgroup_oom_queue
);
2773 u
->in_cgroup_oom_queue
= false;
2774 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
2776 if (m
->cgroup_oom_queue
) {
2777 /* More stuff queued, let's make sure we remain enabled */
2778 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2780 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
2783 (void) unit_check_oom(u
);
2787 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
2792 if (u
->in_cgroup_oom_queue
)
2794 if (!u
->cgroup_path
)
2797 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
2798 u
->in_cgroup_oom_queue
= true;
2800 /* Trigger the defer event */
2801 if (!u
->manager
->cgroup_oom_event_source
) {
2802 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
2804 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
2806 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
2810 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
2812 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
2816 (void) sd_event_source_set_description(s
, "cgroup-oom");
2817 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
2820 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
2822 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
2825 static int unit_check_cgroup_events(Unit
*u
) {
2826 char *values
[2] = {};
2831 r
= cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events",
2832 STRV_MAKE("populated", "frozen"), values
);
2836 /* The cgroup.events notifications can be merged together so act as we saw the given state for the
2837 * first time. The functions we call to handle given state are idempotent, which makes them
2838 * effectively remember the previous state. */
2840 if (streq(values
[0], "1"))
2841 unit_remove_from_cgroup_empty_queue(u
);
2843 unit_add_to_cgroup_empty_queue(u
);
2846 /* Disregard freezer state changes due to operations not initiated by us */
2847 if (values
[1] && IN_SET(u
->freezer_state
, FREEZER_FREEZING
, FREEZER_THAWING
)) {
2848 if (streq(values
[1], "0"))
2860 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
2861 Manager
*m
= userdata
;
2868 union inotify_event_buffer buffer
;
2869 struct inotify_event
*e
;
2872 l
= read(fd
, &buffer
, sizeof(buffer
));
2874 if (IN_SET(errno
, EINTR
, EAGAIN
))
2877 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
2880 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
2884 /* Queue overflow has no watch descriptor */
2887 if (e
->mask
& IN_IGNORED
)
2888 /* The watch was just removed */
2891 /* Note that inotify might deliver events for a watch even after it was removed,
2892 * because it was queued before the removal. Let's ignore this here safely. */
2894 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2896 unit_check_cgroup_events(u
);
2898 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2900 unit_add_to_cgroup_oom_queue(u
);
2905 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
2906 CGroupMask mask
= 0;
2909 /* BPF-based firewall */
2910 r
= bpf_firewall_supported();
2912 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2914 /* BPF-based device access control */
2915 r
= bpf_devices_supported();
2917 mask
|= CGROUP_MASK_BPF_DEVICES
;
2923 int manager_setup_cgroup(Manager
*m
) {
2924 _cleanup_free_
char *path
= NULL
;
2925 const char *scope_path
;
2933 /* 1. Determine hierarchy */
2934 m
->cgroup_root
= mfree(m
->cgroup_root
);
2935 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
2937 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
2939 /* Chop off the init scope, if we are already located in it */
2940 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
2942 /* LEGACY: Also chop off the system slice if we are in
2943 * it. This is to support live upgrades from older systemd
2944 * versions where PID 1 was moved there. Also see
2945 * cg_get_root_path(). */
2946 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
2947 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
2949 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
2954 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
2955 * easily prepend it everywhere. */
2956 delete_trailing_chars(m
->cgroup_root
, "/");
2959 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
2961 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
2965 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
2967 all_unified
= cg_all_unified();
2968 if (all_unified
< 0)
2969 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
2970 if (all_unified
> 0)
2971 log_debug("Unified cgroup hierarchy is located at %s.", path
);
2973 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2975 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
2977 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
2979 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
2982 /* 3. Allocate cgroup empty defer event source */
2983 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
2984 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
2986 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
2988 /* Schedule cgroup empty checks early, but after having processed service notification messages or
2989 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
2990 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
2991 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
2993 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
2995 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
2997 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
2999 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
3001 /* 4. Install notifier inotify object, or agent */
3002 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
3004 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
3006 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
3007 safe_close(m
->cgroup_inotify_fd
);
3009 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
3010 if (m
->cgroup_inotify_fd
< 0)
3011 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
3013 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
3015 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
3017 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
3018 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
3019 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
3020 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
3022 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
3024 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
3026 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
3028 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
3029 * since it does not generate events when control groups with children run empty. */
3031 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUP_AGENT_PATH
);
3033 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
3035 log_debug("Installed release agent.");
3037 log_debug("Release agent already installed.");
3040 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
3041 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
3042 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
3044 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
3045 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
3047 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
3049 /* 6. And pin it, so that it cannot be unmounted */
3050 safe_close(m
->pin_cgroupfs_fd
);
3051 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
3052 if (m
->pin_cgroupfs_fd
< 0)
3053 return log_error_errno(errno
, "Failed to open pin file: %m");
3055 } else if (!MANAGER_IS_TEST_RUN(m
))
3056 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
3058 /* 7. Always enable hierarchical support if it exists... */
3059 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
3060 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
3062 /* 8. Figure out which controllers are supported */
3063 r
= cg_mask_supported(&m
->cgroup_supported
);
3065 return log_error_errno(r
, "Failed to determine supported controllers: %m");
3067 /* 9. Figure out which bpf-based pseudo-controllers are supported */
3068 r
= cg_bpf_mask_supported(&mask
);
3070 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
3071 m
->cgroup_supported
|= mask
;
3073 /* 10. Log which controllers are supported */
3074 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
3075 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
3080 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
3083 /* We can't really delete the group, since we are in it. But
3085 if (delete && m
->cgroup_root
&& m
->test_run_flags
!= MANAGER_TEST_RUN_MINIMAL
)
3086 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
3088 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
3090 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
3091 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
3093 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
3094 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
3096 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
3098 m
->cgroup_root
= mfree(m
->cgroup_root
);
3101 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
3108 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
3112 p
= strdupa(cgroup
);
3116 e
= strrchr(p
, '/');
3118 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
3122 u
= hashmap_get(m
->cgroup_unit
, p
);
3128 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
3129 _cleanup_free_
char *cgroup
= NULL
;
3133 if (!pid_is_valid(pid
))
3136 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
3139 return manager_get_unit_by_cgroup(m
, cgroup
);
3142 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
3147 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
3148 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
3149 * relevant one as children of the process will be assigned to that one, too, before all else. */
3151 if (!pid_is_valid(pid
))
3154 if (pid
== getpid_cached())
3155 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
3157 u
= manager_get_unit_by_pid_cgroup(m
, pid
);
3161 u
= hashmap_get(m
->watch_pids
, PID_TO_PTR(pid
));
3165 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-pid
));
3172 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
3178 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
3179 * or from the --system instance */
3181 log_debug("Got cgroup empty notification for: %s", cgroup
);
3183 u
= manager_get_unit_by_cgroup(m
, cgroup
);
3187 unit_add_to_cgroup_empty_queue(u
);
3191 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
3197 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3200 if (!u
->cgroup_path
)
3203 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3204 if (unit_has_host_root_cgroup(u
))
3205 return procfs_memory_get_used(ret
);
3207 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3210 r
= cg_all_unified();
3214 return cg_get_attribute_as_uint64("memory", u
->cgroup_path
, r
> 0 ? "memory.current" : "memory.usage_in_bytes", ret
);
3217 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
3221 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
3224 if (!u
->cgroup_path
)
3227 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3228 if (unit_has_host_root_cgroup(u
))
3229 return procfs_tasks_get_current(ret
);
3231 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
3234 return cg_get_attribute_as_uint64("pids", u
->cgroup_path
, "pids.current", ret
);
3237 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
3244 if (!u
->cgroup_path
)
3247 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3248 if (unit_has_host_root_cgroup(u
))
3249 return procfs_cpu_get_usage(ret
);
3251 /* Requisite controllers for CPU accounting are not enabled */
3252 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
3255 r
= cg_all_unified();
3259 _cleanup_free_
char *val
= NULL
;
3262 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
3263 if (IN_SET(r
, -ENOENT
, -ENXIO
))
3268 r
= safe_atou64(val
, &us
);
3272 ns
= us
* NSEC_PER_USEC
;
3274 return cg_get_attribute_as_uint64("cpuacct", u
->cgroup_path
, "cpuacct.usage", ret
);
3280 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
3286 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
3287 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
3288 * call this function with a NULL return value. */
3290 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
3293 r
= unit_get_cpu_usage_raw(u
, &ns
);
3294 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
3295 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
3299 *ret
= u
->cpu_usage_last
;
3305 if (ns
> u
->cpu_usage_base
)
3306 ns
-= u
->cpu_usage_base
;
3310 u
->cpu_usage_last
= ns
;
3317 int unit_get_ip_accounting(
3319 CGroupIPAccountingMetric metric
,
3326 assert(metric
>= 0);
3327 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
3330 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
3333 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
3334 u
->ip_accounting_ingress_map_fd
:
3335 u
->ip_accounting_egress_map_fd
;
3339 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
3340 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
3342 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
3346 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
3347 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
3348 * ip_accounting_extra[] field, and add them in here transparently. */
3350 *ret
= value
+ u
->ip_accounting_extra
[metric
];
3355 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
3356 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
3357 [CGROUP_IO_READ_BYTES
] = "rbytes=",
3358 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
3359 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
3360 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
3362 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
3363 _cleanup_free_
char *path
= NULL
;
3364 _cleanup_fclose_
FILE *f
= NULL
;
3369 if (!u
->cgroup_path
)
3372 if (unit_has_host_root_cgroup(u
))
3373 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
3375 r
= cg_all_unified();
3378 if (r
== 0) /* TODO: support cgroupv1 */
3381 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
3384 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
3388 f
= fopen(path
, "re");
3393 _cleanup_free_
char *line
= NULL
;
3396 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3403 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
3404 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
3407 _cleanup_free_
char *word
= NULL
;
3409 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
3415 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3418 x
= startswith(word
, field_names
[i
]);
3422 r
= safe_atou64(x
, &w
);
3426 /* Sum up the stats of all devices */
3434 memcpy(ret
, acc
, sizeof(acc
));
3438 int unit_get_io_accounting(
3440 CGroupIOAccountingMetric metric
,
3444 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
3447 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
3449 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
3452 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3455 r
= unit_get_io_accounting_raw(u
, raw
);
3456 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3461 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3462 /* Saturated subtraction */
3463 if (raw
[i
] > u
->io_accounting_base
[i
])
3464 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
3466 u
->io_accounting_last
[i
] = 0;
3471 *ret
= u
->io_accounting_last
[metric
];
3476 int unit_reset_cpu_accounting(Unit
*u
) {
3481 u
->cpu_usage_last
= NSEC_INFINITY
;
3483 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
3485 u
->cpu_usage_base
= 0;
3492 int unit_reset_ip_accounting(Unit
*u
) {
3497 if (u
->ip_accounting_ingress_map_fd
>= 0)
3498 r
= bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
);
3500 if (u
->ip_accounting_egress_map_fd
>= 0)
3501 q
= bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
);
3503 zero(u
->ip_accounting_extra
);
3505 return r
< 0 ? r
: q
;
3508 int unit_reset_io_accounting(Unit
*u
) {
3513 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++)
3514 u
->io_accounting_last
[i
] = UINT64_MAX
;
3516 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
3518 zero(u
->io_accounting_base
);
3525 int unit_reset_accounting(Unit
*u
) {
3530 r
= unit_reset_cpu_accounting(u
);
3531 q
= unit_reset_io_accounting(u
);
3532 v
= unit_reset_ip_accounting(u
);
3534 return r
< 0 ? r
: q
< 0 ? q
: v
;
3537 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
3540 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3546 /* always invalidate compat pairs together */
3547 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
3548 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
3550 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
3551 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
3553 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
3556 u
->cgroup_invalidated_mask
|= m
;
3557 unit_add_to_cgroup_realize_queue(u
);
3560 void unit_invalidate_cgroup_bpf(Unit
*u
) {
3563 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3566 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
3569 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
3570 unit_add_to_cgroup_realize_queue(u
);
3572 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
3573 * list of our children includes our own. */
3574 if (u
->type
== UNIT_SLICE
) {
3578 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
])
3579 if (UNIT_DEREF(member
->slice
) == u
)
3580 unit_invalidate_cgroup_bpf(member
);
3584 bool unit_cgroup_delegate(Unit
*u
) {
3589 if (!UNIT_VTABLE(u
)->can_delegate
)
3592 c
= unit_get_cgroup_context(u
);
3599 void manager_invalidate_startup_units(Manager
*m
) {
3604 SET_FOREACH(u
, m
->startup_units
)
3605 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
3608 static int unit_get_nice(Unit
*u
) {
3611 ec
= unit_get_exec_context(u
);
3612 return ec
? ec
->nice
: 0;
3615 static uint64_t unit_get_cpu_weight(Unit
*u
) {
3616 ManagerState state
= manager_state(u
->manager
);
3619 cc
= unit_get_cgroup_context(u
);
3620 return cc
? cgroup_context_cpu_weight(cc
, state
) : CGROUP_WEIGHT_DEFAULT
;
3623 int compare_job_priority(const void *a
, const void *b
) {
3624 const Job
*x
= a
, *y
= b
;
3626 uint64_t weight_x
, weight_y
;
3629 if ((ret
= CMP(x
->unit
->type
, y
->unit
->type
)) != 0)
3632 weight_x
= unit_get_cpu_weight(x
->unit
);
3633 weight_y
= unit_get_cpu_weight(y
->unit
);
3635 if ((ret
= CMP(weight_x
, weight_y
)) != 0)
3638 nice_x
= unit_get_nice(x
->unit
);
3639 nice_y
= unit_get_nice(y
->unit
);
3641 if ((ret
= CMP(nice_x
, nice_y
)) != 0)
3644 return strcmp(x
->unit
->id
, y
->unit
->id
);
3647 int unit_cgroup_freezer_action(Unit
*u
, FreezerAction action
) {
3648 _cleanup_free_
char *path
= NULL
;
3649 FreezerState target
, kernel
= _FREEZER_STATE_INVALID
;
3653 assert(IN_SET(action
, FREEZER_FREEZE
, FREEZER_THAW
));
3655 if (!cg_freezer_supported())
3658 if (!u
->cgroup_realized
)
3661 target
= action
== FREEZER_FREEZE
? FREEZER_FROZEN
: FREEZER_RUNNING
;
3663 r
= unit_freezer_state_kernel(u
, &kernel
);
3665 log_unit_debug_errno(u
, r
, "Failed to obtain cgroup freezer state: %m");
3667 if (target
== kernel
) {
3668 u
->freezer_state
= target
;
3672 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.freeze", &path
);
3676 log_unit_debug(u
, "%s unit.", action
== FREEZER_FREEZE
? "Freezing" : "Thawing");
3678 if (action
== FREEZER_FREEZE
)
3679 u
->freezer_state
= FREEZER_FREEZING
;
3681 u
->freezer_state
= FREEZER_THAWING
;
3683 r
= write_string_file(path
, one_zero(action
== FREEZER_FREEZE
), WRITE_STRING_FILE_DISABLE_BUFFER
);
3690 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
3691 [CGROUP_DEVICE_POLICY_AUTO
] = "auto",
3692 [CGROUP_DEVICE_POLICY_CLOSED
] = "closed",
3693 [CGROUP_DEVICE_POLICY_STRICT
] = "strict",
3696 int unit_get_cpuset(Unit
*u
, CPUSet
*cpus
, const char *name
) {
3697 _cleanup_free_
char *v
= NULL
;
3703 if (!u
->cgroup_path
)
3706 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUSET
) == 0)
3709 r
= cg_all_unified();
3715 r
= cg_get_attribute("cpuset", u
->cgroup_path
, name
, &v
);
3721 return parse_cpu_set_full(v
, cpus
, false, NULL
, NULL
, 0, NULL
);
3724 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);
3726 static const char* const freezer_action_table
[_FREEZER_ACTION_MAX
] = {
3727 [FREEZER_FREEZE
] = "freeze",
3728 [FREEZER_THAW
] = "thaw",
3731 DEFINE_STRING_TABLE_LOOKUP(freezer_action
, FreezerAction
);