1 /* SPDX-License-Identifier: LGPL-2.1+ */
6 #include "sd-messages.h"
8 #include "alloc-util.h"
9 #include "blockdev-util.h"
10 #include "bpf-devices.h"
11 #include "bpf-firewall.h"
12 #include "btrfs-util.h"
13 #include "bus-error.h"
14 #include "cgroup-setup.h"
15 #include "cgroup-util.h"
20 #include "nulstr-util.h"
21 #include "parse-util.h"
22 #include "path-util.h"
23 #include "process-util.h"
24 #include "procfs-util.h"
26 #include "stat-util.h"
27 #include "stdio-util.h"
28 #include "string-table.h"
29 #include "string-util.h"
32 #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
34 /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access
35 * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask
36 * out specific attributes from us. */
37 #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING)
39 bool manager_owns_host_root_cgroup(Manager
*m
) {
42 /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the
43 * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's
44 * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if
45 * we run in any kind of container virtualization. */
47 if (MANAGER_IS_USER(m
))
50 if (detect_container() > 0)
53 return empty_or_root(m
->cgroup_root
);
56 bool unit_has_host_root_cgroup(Unit
*u
) {
59 /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and
60 * the manager manages the root cgroup. */
62 if (!manager_owns_host_root_cgroup(u
->manager
))
65 return unit_has_name(u
, SPECIAL_ROOT_SLICE
);
68 static int set_attribute_and_warn(Unit
*u
, const char *controller
, const char *attribute
, const char *value
) {
71 r
= cg_set_attribute(controller
, u
->cgroup_path
, attribute
, value
);
73 log_unit_full(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
, "Failed to set '%s' attribute on '%s' to '%.*s': %m",
74 strna(attribute
), isempty(u
->cgroup_path
) ? "/" : u
->cgroup_path
, (int) strcspn(value
, NEWLINE
), value
);
79 static void cgroup_compat_warn(void) {
80 static bool cgroup_compat_warned
= false;
82 if (cgroup_compat_warned
)
85 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. "
86 "See cgroup-compat debug messages for details.");
88 cgroup_compat_warned
= true;
91 #define log_cgroup_compat(unit, fmt, ...) do { \
92 cgroup_compat_warn(); \
93 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
96 void cgroup_context_init(CGroupContext
*c
) {
99 /* Initialize everything to the kernel defaults. */
101 *c
= (CGroupContext
) {
102 .cpu_weight
= CGROUP_WEIGHT_INVALID
,
103 .startup_cpu_weight
= CGROUP_WEIGHT_INVALID
,
104 .cpu_quota_per_sec_usec
= USEC_INFINITY
,
105 .cpu_quota_period_usec
= USEC_INFINITY
,
107 .cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
108 .startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
,
110 .memory_high
= CGROUP_LIMIT_MAX
,
111 .memory_max
= CGROUP_LIMIT_MAX
,
112 .memory_swap_max
= CGROUP_LIMIT_MAX
,
114 .memory_limit
= CGROUP_LIMIT_MAX
,
116 .io_weight
= CGROUP_WEIGHT_INVALID
,
117 .startup_io_weight
= CGROUP_WEIGHT_INVALID
,
119 .blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
120 .startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
,
122 .tasks_max
= CGROUP_LIMIT_MAX
,
126 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
130 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
135 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
139 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
144 void cgroup_context_free_io_device_latency(CGroupContext
*c
, CGroupIODeviceLatency
*l
) {
148 LIST_REMOVE(device_latencies
, c
->io_device_latencies
, l
);
153 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
157 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
162 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
166 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
171 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
175 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
180 void cgroup_context_done(CGroupContext
*c
) {
183 while (c
->io_device_weights
)
184 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
186 while (c
->io_device_latencies
)
187 cgroup_context_free_io_device_latency(c
, c
->io_device_latencies
);
189 while (c
->io_device_limits
)
190 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
192 while (c
->blockio_device_weights
)
193 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
195 while (c
->blockio_device_bandwidths
)
196 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
198 while (c
->device_allow
)
199 cgroup_context_free_device_allow(c
, c
->device_allow
);
201 c
->ip_address_allow
= ip_address_access_free_all(c
->ip_address_allow
);
202 c
->ip_address_deny
= ip_address_access_free_all(c
->ip_address_deny
);
204 c
->ip_filters_ingress
= strv_free(c
->ip_filters_ingress
);
205 c
->ip_filters_egress
= strv_free(c
->ip_filters_egress
);
207 cpu_set_reset(&c
->cpuset_cpus
);
208 cpu_set_reset(&c
->cpuset_mems
);
211 static int unit_get_kernel_memory_limit(Unit
*u
, const char *file
, uint64_t *ret
) {
212 _cleanup_free_
char *raw_kval
= NULL
;
218 if (!u
->cgroup_realized
)
221 r
= cg_get_attribute("memory", u
->cgroup_path
, file
, &raw_kval
);
225 if (streq(raw_kval
, "max")) {
226 *ret
= CGROUP_LIMIT_MAX
;
230 r
= safe_atou64(raw_kval
, &kval
);
239 static int unit_compare_memory_limit(Unit
*u
, const char *property_name
, uint64_t *ret_unit_value
, uint64_t *ret_kernel_value
) {
246 /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will
247 * return -ENODATA) on cgroup v1.
252 * 0: If the kernel memory setting doesn't match our configuration.
253 * >0: If the kernel memory setting matches our configuration.
255 * The following values are only guaranteed to be populated on return >=0:
257 * - ret_unit_value will contain our internal expected value for the unit, page-aligned.
258 * - ret_kernel_value will contain the actual value presented by the kernel. */
262 r
= cg_all_unified();
264 return log_debug_errno(r
, "Failed to determine cgroup hierarchy version: %m");
266 /* Unsupported on v1.
268 * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has
269 * silently masked the controller. */
273 /* The root slice doesn't have any controller files, so we can't compare anything. */
274 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
277 /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled,
278 * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To
279 * avoid specious errors in these scenarios, check that we even expect the memory controller to be
281 m
= unit_get_target_mask(u
);
282 if (!FLAGS_SET(m
, CGROUP_MASK_MEMORY
))
285 c
= unit_get_cgroup_context(u
);
288 if (streq(property_name
, "MemoryLow")) {
289 unit_value
= unit_get_ancestor_memory_low(u
);
291 } else if (streq(property_name
, "MemoryMin")) {
292 unit_value
= unit_get_ancestor_memory_min(u
);
294 } else if (streq(property_name
, "MemoryHigh")) {
295 unit_value
= c
->memory_high
;
296 file
= "memory.high";
297 } else if (streq(property_name
, "MemoryMax")) {
298 unit_value
= c
->memory_max
;
300 } else if (streq(property_name
, "MemorySwapMax")) {
301 unit_value
= c
->memory_swap_max
;
302 file
= "memory.swap.max";
306 r
= unit_get_kernel_memory_limit(u
, file
, ret_kernel_value
);
308 return log_unit_debug_errno(u
, r
, "Failed to parse %s: %m", file
);
310 /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page
311 * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from
312 * our internal page-counting. To support those future kernels, just check the value itself first
313 * without any page-alignment. */
314 if (*ret_kernel_value
== unit_value
) {
315 *ret_unit_value
= unit_value
;
319 /* The current kernel behaviour, by comparison, is that even if you write a particular number of
320 * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel
321 * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is
323 if (unit_value
!= CGROUP_LIMIT_MAX
)
324 unit_value
= PAGE_ALIGN_DOWN(unit_value
);
326 *ret_unit_value
= unit_value
;
328 return *ret_kernel_value
== *ret_unit_value
;
331 #define FORMAT_CGROUP_DIFF_MAX 128
333 static char *format_cgroup_memory_limit_comparison(char *buf
, size_t l
, Unit
*u
, const char *property_name
) {
341 r
= unit_compare_memory_limit(u
, property_name
, &sval
, &kval
);
343 /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1).
344 * In the absence of reliably being able to detect whether memcg swap support is available or not,
345 * only complain if the error is not ENOENT. */
346 if (r
> 0 || IN_SET(r
, -ENODATA
, -EOWNERDEAD
) ||
347 (r
== -ENOENT
&& streq(property_name
, "MemorySwapMax"))) {
353 snprintf(buf
, l
, " (error getting kernel value: %s)", strerror_safe(r
));
357 snprintf(buf
, l
, " (different value in kernel: %" PRIu64
")", kval
);
362 void cgroup_context_dump(Unit
*u
, FILE* f
, const char *prefix
) {
363 _cleanup_free_
char *disable_controllers_str
= NULL
, *cpuset_cpus
= NULL
, *cpuset_mems
= NULL
;
364 CGroupIODeviceLimit
*il
;
365 CGroupIODeviceWeight
*iw
;
366 CGroupIODeviceLatency
*l
;
367 CGroupBlockIODeviceBandwidth
*b
;
368 CGroupBlockIODeviceWeight
*w
;
369 CGroupDeviceAllow
*a
;
371 IPAddressAccessItem
*iaai
;
373 char q
[FORMAT_TIMESPAN_MAX
];
374 char v
[FORMAT_TIMESPAN_MAX
];
376 char cda
[FORMAT_CGROUP_DIFF_MAX
];
377 char cdb
[FORMAT_CGROUP_DIFF_MAX
];
378 char cdc
[FORMAT_CGROUP_DIFF_MAX
];
379 char cdd
[FORMAT_CGROUP_DIFF_MAX
];
380 char cde
[FORMAT_CGROUP_DIFF_MAX
];
385 c
= unit_get_cgroup_context(u
);
388 prefix
= strempty(prefix
);
390 (void) cg_mask_to_string(c
->disable_controllers
, &disable_controllers_str
);
392 cpuset_cpus
= cpu_set_to_range_string(&c
->cpuset_cpus
);
393 cpuset_mems
= cpu_set_to_range_string(&c
->cpuset_mems
);
396 "%sCPUAccounting: %s\n"
397 "%sIOAccounting: %s\n"
398 "%sBlockIOAccounting: %s\n"
399 "%sMemoryAccounting: %s\n"
400 "%sTasksAccounting: %s\n"
401 "%sIPAccounting: %s\n"
402 "%sCPUWeight: %" PRIu64
"\n"
403 "%sStartupCPUWeight: %" PRIu64
"\n"
404 "%sCPUShares: %" PRIu64
"\n"
405 "%sStartupCPUShares: %" PRIu64
"\n"
406 "%sCPUQuotaPerSecSec: %s\n"
407 "%sCPUQuotaPeriodSec: %s\n"
408 "%sAllowedCPUs: %s\n"
409 "%sAllowedMemoryNodes: %s\n"
410 "%sIOWeight: %" PRIu64
"\n"
411 "%sStartupIOWeight: %" PRIu64
"\n"
412 "%sBlockIOWeight: %" PRIu64
"\n"
413 "%sStartupBlockIOWeight: %" PRIu64
"\n"
414 "%sDefaultMemoryMin: %" PRIu64
"\n"
415 "%sDefaultMemoryLow: %" PRIu64
"\n"
416 "%sMemoryMin: %" PRIu64
"%s\n"
417 "%sMemoryLow: %" PRIu64
"%s\n"
418 "%sMemoryHigh: %" PRIu64
"%s\n"
419 "%sMemoryMax: %" PRIu64
"%s\n"
420 "%sMemorySwapMax: %" PRIu64
"%s\n"
421 "%sMemoryLimit: %" PRIu64
"\n"
422 "%sTasksMax: %" PRIu64
"\n"
423 "%sDevicePolicy: %s\n"
424 "%sDisableControllers: %s\n"
426 prefix
, yes_no(c
->cpu_accounting
),
427 prefix
, yes_no(c
->io_accounting
),
428 prefix
, yes_no(c
->blockio_accounting
),
429 prefix
, yes_no(c
->memory_accounting
),
430 prefix
, yes_no(c
->tasks_accounting
),
431 prefix
, yes_no(c
->ip_accounting
),
432 prefix
, c
->cpu_weight
,
433 prefix
, c
->startup_cpu_weight
,
434 prefix
, c
->cpu_shares
,
435 prefix
, c
->startup_cpu_shares
,
436 prefix
, format_timespan(q
, sizeof(q
), c
->cpu_quota_per_sec_usec
, 1),
437 prefix
, format_timespan(v
, sizeof(v
), c
->cpu_quota_period_usec
, 1),
438 prefix
, strempty(cpuset_cpus
),
439 prefix
, strempty(cpuset_mems
),
440 prefix
, c
->io_weight
,
441 prefix
, c
->startup_io_weight
,
442 prefix
, c
->blockio_weight
,
443 prefix
, c
->startup_blockio_weight
,
444 prefix
, c
->default_memory_min
,
445 prefix
, c
->default_memory_low
,
446 prefix
, c
->memory_min
, format_cgroup_memory_limit_comparison(cda
, sizeof(cda
), u
, "MemoryMin"),
447 prefix
, c
->memory_low
, format_cgroup_memory_limit_comparison(cdb
, sizeof(cdb
), u
, "MemoryLow"),
448 prefix
, c
->memory_high
, format_cgroup_memory_limit_comparison(cdc
, sizeof(cdc
), u
, "MemoryHigh"),
449 prefix
, c
->memory_max
, format_cgroup_memory_limit_comparison(cdd
, sizeof(cdd
), u
, "MemoryMax"),
450 prefix
, c
->memory_swap_max
, format_cgroup_memory_limit_comparison(cde
, sizeof(cde
), u
, "MemorySwapMax"),
451 prefix
, c
->memory_limit
,
452 prefix
, c
->tasks_max
,
453 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
454 prefix
, strempty(disable_controllers_str
),
455 prefix
, yes_no(c
->delegate
));
458 _cleanup_free_
char *t
= NULL
;
460 (void) cg_mask_to_string(c
->delegate_controllers
, &t
);
462 fprintf(f
, "%sDelegateControllers: %s\n",
467 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
469 "%sDeviceAllow: %s %s%s%s\n",
472 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
474 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
476 "%sIODeviceWeight: %s %" PRIu64
"\n",
481 LIST_FOREACH(device_latencies
, l
, c
->io_device_latencies
)
483 "%sIODeviceLatencyTargetSec: %s %s\n",
486 format_timespan(q
, sizeof(q
), l
->target_usec
, 1));
488 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
489 char buf
[FORMAT_BYTES_MAX
];
490 CGroupIOLimitType type
;
492 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
493 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
497 cgroup_io_limit_type_to_string(type
),
499 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
502 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
504 "%sBlockIODeviceWeight: %s %" PRIu64
,
509 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
510 char buf
[FORMAT_BYTES_MAX
];
512 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
514 "%sBlockIOReadBandwidth: %s %s\n",
517 format_bytes(buf
, sizeof(buf
), b
->rbps
));
518 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
520 "%sBlockIOWriteBandwidth: %s %s\n",
523 format_bytes(buf
, sizeof(buf
), b
->wbps
));
526 LIST_FOREACH(items
, iaai
, c
->ip_address_allow
) {
527 _cleanup_free_
char *k
= NULL
;
529 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
530 fprintf(f
, "%sIPAddressAllow: %s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
533 LIST_FOREACH(items
, iaai
, c
->ip_address_deny
) {
534 _cleanup_free_
char *k
= NULL
;
536 (void) in_addr_to_string(iaai
->family
, &iaai
->address
, &k
);
537 fprintf(f
, "%sIPAddressDeny: %s/%u\n", prefix
, strnull(k
), iaai
->prefixlen
);
540 STRV_FOREACH(path
, c
->ip_filters_ingress
)
541 fprintf(f
, "%sIPIngressFilterPath: %s\n", prefix
, *path
);
543 STRV_FOREACH(path
, c
->ip_filters_egress
)
544 fprintf(f
, "%sIPEgressFilterPath: %s\n", prefix
, *path
);
547 int cgroup_add_device_allow(CGroupContext
*c
, const char *dev
, const char *mode
) {
548 _cleanup_free_ CGroupDeviceAllow
*a
= NULL
;
549 _cleanup_free_
char *d
= NULL
;
553 assert(isempty(mode
) || in_charset(mode
, "rwm"));
555 a
= new(CGroupDeviceAllow
, 1);
563 *a
= (CGroupDeviceAllow
) {
565 .r
= isempty(mode
) || strchr(mode
, 'r'),
566 .w
= isempty(mode
) || strchr(mode
, 'w'),
567 .m
= isempty(mode
) || strchr(mode
, 'm'),
570 LIST_PREPEND(device_allow
, c
->device_allow
, a
);
576 #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \
577 uint64_t unit_get_ancestor_##entry(Unit *u) { \
580 /* 1. Is entry set in this unit? If so, use that. \
581 * 2. Is the default for this entry set in any \
582 * ancestor? If so, use that. \
583 * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \
587 c = unit_get_cgroup_context(u); \
588 if (c && c->entry##_set) \
591 while ((u = UNIT_DEREF(u->slice))) { \
592 c = unit_get_cgroup_context(u); \
593 if (c && c->default_##entry##_set) \
594 return c->default_##entry; \
597 /* We've reached the root, but nobody had default for \
598 * this entry set, so set it to the kernel default. */ \
599 return CGROUP_LIMIT_MIN; \
602 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low
);
603 UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min
);
605 static void cgroup_xattr_apply(Unit
*u
) {
606 char ids
[SD_ID128_STRING_MAX
];
611 if (!MANAGER_IS_SYSTEM(u
->manager
))
614 if (sd_id128_is_null(u
->invocation_id
))
617 r
= cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
,
618 "trusted.invocation_id",
619 sd_id128_to_string(u
->invocation_id
, ids
), 32,
622 log_unit_debug_errno(u
, r
, "Failed to set invocation ID on control group %s, ignoring: %m", u
->cgroup_path
);
625 static int lookup_block_device(const char *p
, dev_t
*ret
) {
633 r
= device_path_parse_major_minor(p
, &mode
, &rdev
);
634 if (r
== -ENODEV
) { /* not a parsable device node, need to go to disk */
636 if (stat(p
, &st
) < 0)
637 return log_warning_errno(errno
, "Couldn't stat device '%s': %m", p
);
638 rdev
= (dev_t
)st
.st_rdev
;
639 dev
= (dev_t
)st
.st_dev
;
642 return log_warning_errno(r
, "Failed to parse major/minor from path '%s': %m", p
);
645 log_warning("Device node '%s' is a character device, but block device needed.", p
);
647 } else if (S_ISBLK(mode
))
649 else if (major(dev
) != 0)
650 *ret
= dev
; /* If this is not a device node then use the block device this file is stored on */
652 /* If this is btrfs, getting the backing block device is a bit harder */
653 r
= btrfs_get_block_device(p
, ret
);
654 if (r
< 0 && r
!= -ENOTTY
)
655 return log_warning_errno(r
, "Failed to determine block device backing btrfs file system '%s': %m", p
);
657 log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p
);
662 /* If this is a LUKS device, try to get the originating block device */
663 (void) block_get_originating(*ret
, ret
);
665 /* If this is a partition, try to get the originating block device */
666 (void) block_get_whole_disk(*ret
, ret
);
670 static int whitelist_device(BPFProgram
*prog
, const char *path
, const char *node
, const char *acc
) {
678 /* Some special handling for /dev/block/%u:%u, /dev/char/%u:%u, /run/systemd/inaccessible/chr and
679 * /run/systemd/inaccessible/blk paths. Instead of stat()ing these we parse out the major/minor directly. This
680 * means clients can use these path without the device node actually around */
681 r
= device_path_parse_major_minor(node
, &mode
, &rdev
);
684 return log_warning_errno(r
, "Couldn't parse major/minor from device path '%s': %m", node
);
687 if (stat(node
, &st
) < 0)
688 return log_warning_errno(errno
, "Couldn't stat device %s: %m", node
);
690 if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
)) {
691 log_warning("%s is not a device.", node
);
694 rdev
= (dev_t
) st
.st_rdev
;
698 if (cg_all_unified() > 0) {
702 return cgroup_bpf_whitelist_device(prog
, S_ISCHR(mode
) ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
703 major(rdev
), minor(rdev
), acc
);
706 char buf
[2+DECIMAL_STR_MAX(dev_t
)*2+2+4];
710 S_ISCHR(mode
) ? 'c' : 'b',
711 major(rdev
), minor(rdev
),
714 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */
716 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
718 return log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
719 r
, "Failed to set devices.allow on %s: %m", path
);
725 static int whitelist_major(BPFProgram
*prog
, const char *path
, const char *name
, char type
, const char *acc
) {
726 _cleanup_fclose_
FILE *f
= NULL
;
727 char buf
[2+DECIMAL_STR_MAX(unsigned)+3+4];
734 assert(IN_SET(type
, 'b', 'c'));
736 if (streq(name
, "*")) {
737 /* If the name is a wildcard, then apply this list to all devices of this type */
739 if (cg_all_unified() > 0) {
743 (void) cgroup_bpf_whitelist_class(prog
, type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
, acc
);
745 xsprintf(buf
, "%c *:* %s", type
, acc
);
747 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
749 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
750 "Failed to set devices.allow on %s: %m", path
);
755 if (safe_atou(name
, &maj
) >= 0 && DEVICE_MAJOR_VALID(maj
)) {
756 /* The name is numeric and suitable as major. In that case, let's take is major, and create the entry
759 if (cg_all_unified() > 0) {
763 (void) cgroup_bpf_whitelist_major(prog
,
764 type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
767 xsprintf(buf
, "%c %u:* %s", type
, maj
, acc
);
769 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
771 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
772 "Failed to set devices.allow on %s: %m", path
);
778 f
= fopen("/proc/devices", "re");
780 return log_warning_errno(errno
, "Cannot open /proc/devices to resolve %s (%c): %m", name
, type
);
783 _cleanup_free_
char *line
= NULL
;
786 r
= read_line(f
, LONG_LINE_MAX
, &line
);
788 return log_warning_errno(r
, "Failed to read /proc/devices: %m");
792 if (type
== 'c' && streq(line
, "Character devices:")) {
797 if (type
== 'b' && streq(line
, "Block devices:")) {
812 w
= strpbrk(p
, WHITESPACE
);
817 r
= safe_atou(p
, &maj
);
824 w
+= strspn(w
, WHITESPACE
);
826 if (fnmatch(name
, w
, 0) != 0)
829 if (cg_all_unified() > 0) {
833 (void) cgroup_bpf_whitelist_major(prog
,
834 type
== 'c' ? BPF_DEVCG_DEV_CHAR
: BPF_DEVCG_DEV_BLOCK
,
843 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL
846 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
848 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
849 r
, "Failed to set devices.allow on %s: %m", path
);
856 static bool cgroup_context_has_cpu_weight(CGroupContext
*c
) {
857 return c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
||
858 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
;
861 static bool cgroup_context_has_cpu_shares(CGroupContext
*c
) {
862 return c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
863 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
;
866 static uint64_t cgroup_context_cpu_weight(CGroupContext
*c
, ManagerState state
) {
867 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
868 c
->startup_cpu_weight
!= CGROUP_WEIGHT_INVALID
)
869 return c
->startup_cpu_weight
;
870 else if (c
->cpu_weight
!= CGROUP_WEIGHT_INVALID
)
871 return c
->cpu_weight
;
873 return CGROUP_WEIGHT_DEFAULT
;
876 static uint64_t cgroup_context_cpu_shares(CGroupContext
*c
, ManagerState state
) {
877 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
878 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
879 return c
->startup_cpu_shares
;
880 else if (c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
)
881 return c
->cpu_shares
;
883 return CGROUP_CPU_SHARES_DEFAULT
;
886 usec_t
cgroup_cpu_adjust_period(usec_t period
, usec_t quota
, usec_t resolution
, usec_t max_period
) {
887 /* kernel uses a minimum resolution of 1ms, so both period and (quota * period)
888 * need to be higher than that boundary. quota is specified in USecPerSec.
889 * Additionally, period must be at most max_period. */
892 return MIN(MAX3(period
, resolution
, resolution
* USEC_PER_SEC
/ quota
), max_period
);
895 static usec_t
cgroup_cpu_adjust_period_and_log(Unit
*u
, usec_t period
, usec_t quota
) {
898 if (quota
== USEC_INFINITY
)
899 /* Always use default period for infinity quota. */
900 return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
902 if (period
== USEC_INFINITY
)
903 /* Default period was requested. */
904 period
= CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC
;
906 /* Clamp to interval [1ms, 1s] */
907 new_period
= cgroup_cpu_adjust_period(period
, quota
, USEC_PER_MSEC
, USEC_PER_SEC
);
909 if (new_period
!= period
) {
910 char v
[FORMAT_TIMESPAN_MAX
];
911 log_unit_full(u
, u
->warned_clamping_cpu_quota_period
? LOG_DEBUG
: LOG_WARNING
, 0,
912 "Clamping CPU interval for cpu.max: period is now %s",
913 format_timespan(v
, sizeof(v
), new_period
, 1));
914 u
->warned_clamping_cpu_quota_period
= true;
920 static void cgroup_apply_unified_cpu_weight(Unit
*u
, uint64_t weight
) {
921 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
923 xsprintf(buf
, "%" PRIu64
"\n", weight
);
924 (void) set_attribute_and_warn(u
, "cpu", "cpu.weight", buf
);
927 static void cgroup_apply_unified_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
928 char buf
[(DECIMAL_STR_MAX(usec_t
) + 1) * 2 + 1];
930 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
931 if (quota
!= USEC_INFINITY
)
932 xsprintf(buf
, USEC_FMT
" " USEC_FMT
"\n",
933 MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
), period
);
935 xsprintf(buf
, "max " USEC_FMT
"\n", period
);
936 (void) set_attribute_and_warn(u
, "cpu", "cpu.max", buf
);
939 static void cgroup_apply_legacy_cpu_shares(Unit
*u
, uint64_t shares
) {
940 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
942 xsprintf(buf
, "%" PRIu64
"\n", shares
);
943 (void) set_attribute_and_warn(u
, "cpu", "cpu.shares", buf
);
946 static void cgroup_apply_legacy_cpu_quota(Unit
*u
, usec_t quota
, usec_t period
) {
947 char buf
[DECIMAL_STR_MAX(usec_t
) + 2];
949 period
= cgroup_cpu_adjust_period_and_log(u
, period
, quota
);
951 xsprintf(buf
, USEC_FMT
"\n", period
);
952 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_period_us", buf
);
954 if (quota
!= USEC_INFINITY
) {
955 xsprintf(buf
, USEC_FMT
"\n", MAX(quota
* period
/ USEC_PER_SEC
, USEC_PER_MSEC
));
956 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", buf
);
958 (void) set_attribute_and_warn(u
, "cpu", "cpu.cfs_quota_us", "-1\n");
961 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares
) {
962 return CLAMP(shares
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_CPU_SHARES_DEFAULT
,
963 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
966 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight
) {
967 return CLAMP(weight
* CGROUP_CPU_SHARES_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
968 CGROUP_CPU_SHARES_MIN
, CGROUP_CPU_SHARES_MAX
);
971 static void cgroup_apply_unified_cpuset(Unit
*u
, const CPUSet
*cpus
, const char *name
) {
972 _cleanup_free_
char *buf
= NULL
;
974 buf
= cpu_set_to_range_string(cpus
);
980 (void) set_attribute_and_warn(u
, "cpuset", name
, buf
);
983 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
984 return c
->io_accounting
||
985 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
986 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
987 c
->io_device_weights
||
988 c
->io_device_latencies
||
992 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
993 return c
->blockio_accounting
||
994 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
995 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
996 c
->blockio_device_weights
||
997 c
->blockio_device_bandwidths
;
1000 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
1001 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
1002 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
1003 return c
->startup_io_weight
;
1004 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
1005 return c
->io_weight
;
1007 return CGROUP_WEIGHT_DEFAULT
;
1010 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
1011 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
1012 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
1013 return c
->startup_blockio_weight
;
1014 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
1015 return c
->blockio_weight
;
1017 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
1020 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
1021 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
1022 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
1025 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
1026 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
1027 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
1030 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
1031 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1035 r
= lookup_block_device(dev_path
, &dev
);
1039 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
1040 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1043 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
1044 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1048 r
= lookup_block_device(dev_path
, &dev
);
1052 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
1053 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight_device", buf
);
1056 static void cgroup_apply_io_device_latency(Unit
*u
, const char *dev_path
, usec_t target
) {
1057 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1];
1061 r
= lookup_block_device(dev_path
, &dev
);
1065 if (target
!= USEC_INFINITY
)
1066 xsprintf(buf
, "%u:%u target=%" PRIu64
"\n", major(dev
), minor(dev
), target
);
1068 xsprintf(buf
, "%u:%u target=max\n", major(dev
), minor(dev
));
1070 (void) set_attribute_and_warn(u
, "io", "io.latency", buf
);
1073 static void cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
1074 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
1075 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
1076 CGroupIOLimitType type
;
1080 r
= lookup_block_device(dev_path
, &dev
);
1084 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1085 if (limits
[type
] != cgroup_io_limit_defaults
[type
])
1086 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
1088 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
1090 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
1091 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
1092 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
1093 (void) set_attribute_and_warn(u
, "io", "io.max", buf
);
1096 static void cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
1097 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
1101 r
= lookup_block_device(dev_path
, &dev
);
1105 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
1106 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.read_bps_device", buf
);
1108 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
1109 (void) set_attribute_and_warn(u
, "blkio", "blkio.throttle.write_bps_device", buf
);
1112 static bool unit_has_unified_memory_config(Unit
*u
) {
1117 c
= unit_get_cgroup_context(u
);
1120 return unit_get_ancestor_memory_min(u
) > 0 || unit_get_ancestor_memory_low(u
) > 0 ||
1121 c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
||
1122 c
->memory_swap_max
!= CGROUP_LIMIT_MAX
;
1125 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
1126 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n";
1128 if (v
!= CGROUP_LIMIT_MAX
)
1129 xsprintf(buf
, "%" PRIu64
"\n", v
);
1131 (void) set_attribute_and_warn(u
, "memory", file
, buf
);
1134 static void cgroup_apply_firewall(Unit
*u
) {
1137 /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */
1139 if (bpf_firewall_compile(u
) < 0)
1142 (void) bpf_firewall_load_custom(u
);
1143 (void) bpf_firewall_install(u
);
1146 static void cgroup_context_apply(
1148 CGroupMask apply_mask
,
1149 ManagerState state
) {
1153 bool is_host_root
, is_local_root
;
1158 /* Nothing to do? Exit early! */
1159 if (apply_mask
== 0)
1162 /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other
1163 * attributes should only be managed for cgroups further down the tree. */
1164 is_local_root
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1165 is_host_root
= unit_has_host_root_cgroup(u
);
1167 assert_se(c
= unit_get_cgroup_context(u
));
1168 assert_se(path
= u
->cgroup_path
);
1170 if (is_local_root
) /* Make sure we don't try to display messages with an empty path. */
1173 /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container
1174 * then), and missing cgroups, i.e. EROFS and ENOENT. */
1176 /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but
1177 * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this
1178 * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of
1179 * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used
1180 * we couldn't even write to them if we wanted to). */
1181 if ((apply_mask
& CGROUP_MASK_CPU
) && !is_local_root
) {
1183 if (cg_all_unified() > 0) {
1186 if (cgroup_context_has_cpu_weight(c
))
1187 weight
= cgroup_context_cpu_weight(c
, state
);
1188 else if (cgroup_context_has_cpu_shares(c
)) {
1191 shares
= cgroup_context_cpu_shares(c
, state
);
1192 weight
= cgroup_cpu_shares_to_weight(shares
);
1194 log_cgroup_compat(u
, "Applying [Startup]CPUShares=%" PRIu64
" as [Startup]CPUWeight=%" PRIu64
" on %s",
1195 shares
, weight
, path
);
1197 weight
= CGROUP_WEIGHT_DEFAULT
;
1199 cgroup_apply_unified_cpu_weight(u
, weight
);
1200 cgroup_apply_unified_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1205 if (cgroup_context_has_cpu_weight(c
)) {
1208 weight
= cgroup_context_cpu_weight(c
, state
);
1209 shares
= cgroup_cpu_weight_to_shares(weight
);
1211 log_cgroup_compat(u
, "Applying [Startup]CPUWeight=%" PRIu64
" as [Startup]CPUShares=%" PRIu64
" on %s",
1212 weight
, shares
, path
);
1213 } else if (cgroup_context_has_cpu_shares(c
))
1214 shares
= cgroup_context_cpu_shares(c
, state
);
1216 shares
= CGROUP_CPU_SHARES_DEFAULT
;
1218 cgroup_apply_legacy_cpu_shares(u
, shares
);
1219 cgroup_apply_legacy_cpu_quota(u
, c
->cpu_quota_per_sec_usec
, c
->cpu_quota_period_usec
);
1223 if ((apply_mask
& CGROUP_MASK_CPUSET
) && !is_local_root
) {
1224 cgroup_apply_unified_cpuset(u
, &c
->cpuset_cpus
, "cpuset.cpus");
1225 cgroup_apply_unified_cpuset(u
, &c
->cpuset_mems
, "cpuset.mems");
1228 /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2
1229 * controller), and in case of containers we want to leave control of these attributes to the container manager
1230 * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */
1231 if ((apply_mask
& CGROUP_MASK_IO
) && !is_local_root
) {
1232 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
1233 bool has_io
, has_blockio
;
1236 has_io
= cgroup_context_has_io_config(c
);
1237 has_blockio
= cgroup_context_has_blockio_config(c
);
1240 weight
= cgroup_context_io_weight(c
, state
);
1241 else if (has_blockio
) {
1242 uint64_t blkio_weight
;
1244 blkio_weight
= cgroup_context_blkio_weight(c
, state
);
1245 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
1247 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight=%" PRIu64
" as [Startup]IOWeight=%" PRIu64
,
1248 blkio_weight
, weight
);
1250 weight
= CGROUP_WEIGHT_DEFAULT
;
1252 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
1253 (void) set_attribute_and_warn(u
, "io", "io.weight", buf
);
1255 /* FIXME: drop this when distro kernels properly support BFQ through "io.weight"
1256 * See also: https://github.com/systemd/systemd/pull/13335 */
1257 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1258 (void) set_attribute_and_warn(u
, "io", "io.bfq.weight", buf
);
1261 CGroupIODeviceLatency
*latency
;
1262 CGroupIODeviceLimit
*limit
;
1263 CGroupIODeviceWeight
*w
;
1265 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
1266 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
1268 LIST_FOREACH(device_limits
, limit
, c
->io_device_limits
)
1269 cgroup_apply_io_device_limit(u
, limit
->path
, limit
->limits
);
1271 LIST_FOREACH(device_latencies
, latency
, c
->io_device_latencies
)
1272 cgroup_apply_io_device_latency(u
, latency
->path
, latency
->target_usec
);
1274 } else if (has_blockio
) {
1275 CGroupBlockIODeviceWeight
*w
;
1276 CGroupBlockIODeviceBandwidth
*b
;
1278 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
1279 weight
= cgroup_weight_blkio_to_io(w
->weight
);
1281 log_cgroup_compat(u
, "Applying BlockIODeviceWeight=%" PRIu64
" as IODeviceWeight=%" PRIu64
" for %s",
1282 w
->weight
, weight
, w
->path
);
1284 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
1287 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
1288 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
1289 CGroupIOLimitType type
;
1291 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
1292 limits
[type
] = cgroup_io_limit_defaults
[type
];
1294 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
1295 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
1297 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax= for %s",
1298 b
->rbps
, b
->wbps
, b
->path
);
1300 cgroup_apply_io_device_limit(u
, b
->path
, limits
);
1305 if (apply_mask
& CGROUP_MASK_BLKIO
) {
1306 bool has_io
, has_blockio
;
1308 has_io
= cgroup_context_has_io_config(c
);
1309 has_blockio
= cgroup_context_has_blockio_config(c
);
1311 /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be
1312 * left to our container manager, too. */
1313 if (!is_local_root
) {
1314 char buf
[DECIMAL_STR_MAX(uint64_t)+1];
1320 io_weight
= cgroup_context_io_weight(c
, state
);
1321 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
1323 log_cgroup_compat(u
, "Applying [Startup]IOWeight=%" PRIu64
" as [Startup]BlockIOWeight=%" PRIu64
,
1325 } else if (has_blockio
)
1326 weight
= cgroup_context_blkio_weight(c
, state
);
1328 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
1330 xsprintf(buf
, "%" PRIu64
"\n", weight
);
1331 (void) set_attribute_and_warn(u
, "blkio", "blkio.weight", buf
);
1334 CGroupIODeviceWeight
*w
;
1336 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
1337 weight
= cgroup_weight_io_to_blkio(w
->weight
);
1339 log_cgroup_compat(u
, "Applying IODeviceWeight=%" PRIu64
" as BlockIODeviceWeight=%" PRIu64
" for %s",
1340 w
->weight
, weight
, w
->path
);
1342 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
1344 } else if (has_blockio
) {
1345 CGroupBlockIODeviceWeight
*w
;
1347 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
1348 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
1352 /* The bandwidth limits are something that make sense to be applied to the host's root but not container
1353 * roots, as there we want the container manager to handle it */
1354 if (is_host_root
|| !is_local_root
) {
1356 CGroupIODeviceLimit
*l
;
1358 LIST_FOREACH(device_limits
, l
, c
->io_device_limits
) {
1359 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth=%" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax= for %s",
1360 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
1362 cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]);
1364 } else if (has_blockio
) {
1365 CGroupBlockIODeviceBandwidth
*b
;
1367 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
)
1368 cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
);
1373 /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes'
1374 * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we
1375 * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even
1376 * write to this if we wanted to.) */
1377 if ((apply_mask
& CGROUP_MASK_MEMORY
) && !is_local_root
) {
1379 if (cg_all_unified() > 0) {
1380 uint64_t max
, swap_max
= CGROUP_LIMIT_MAX
;
1382 if (unit_has_unified_memory_config(u
)) {
1383 max
= c
->memory_max
;
1384 swap_max
= c
->memory_swap_max
;
1386 max
= c
->memory_limit
;
1388 if (max
!= CGROUP_LIMIT_MAX
)
1389 log_cgroup_compat(u
, "Applying MemoryLimit=%" PRIu64
" as MemoryMax=", max
);
1392 cgroup_apply_unified_memory_limit(u
, "memory.min", unit_get_ancestor_memory_min(u
));
1393 cgroup_apply_unified_memory_limit(u
, "memory.low", unit_get_ancestor_memory_low(u
));
1394 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
1395 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
1396 cgroup_apply_unified_memory_limit(u
, "memory.swap.max", swap_max
);
1398 (void) set_attribute_and_warn(u
, "memory", "memory.oom.group", one_zero(c
->memory_oom_group
));
1401 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
1404 if (unit_has_unified_memory_config(u
)) {
1405 val
= c
->memory_max
;
1406 log_cgroup_compat(u
, "Applying MemoryMax=%" PRIi64
" as MemoryLimit=", val
);
1408 val
= c
->memory_limit
;
1410 if (val
== CGROUP_LIMIT_MAX
)
1411 strncpy(buf
, "-1\n", sizeof(buf
));
1413 xsprintf(buf
, "%" PRIu64
"\n", val
);
1415 (void) set_attribute_and_warn(u
, "memory", "memory.limit_in_bytes", buf
);
1419 /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of
1420 * containers, where we leave this to the manager */
1421 if ((apply_mask
& (CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
)) &&
1422 (is_host_root
|| cg_all_unified() > 0 || !is_local_root
)) {
1423 _cleanup_(bpf_program_unrefp
) BPFProgram
*prog
= NULL
;
1424 CGroupDeviceAllow
*a
;
1426 if (cg_all_unified() > 0) {
1427 r
= cgroup_init_device_bpf(&prog
, c
->device_policy
, c
->device_allow
);
1429 log_unit_warning_errno(u
, r
, "Failed to initialize device control bpf program: %m");
1431 /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL
1434 if (c
->device_allow
|| c
->device_policy
!= CGROUP_AUTO
)
1435 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
1437 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
1439 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1440 "Failed to reset devices.allow/devices.deny: %m");
1443 if (c
->device_policy
== CGROUP_CLOSED
||
1444 (c
->device_policy
== CGROUP_AUTO
&& c
->device_allow
)) {
1445 static const char auto_devices
[] =
1446 "/dev/null\0" "rwm\0"
1447 "/dev/zero\0" "rwm\0"
1448 "/dev/full\0" "rwm\0"
1449 "/dev/random\0" "rwm\0"
1450 "/dev/urandom\0" "rwm\0"
1451 "/dev/tty\0" "rwm\0"
1452 "/dev/ptmx\0" "rwm\0"
1453 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
1454 "/run/systemd/inaccessible/chr\0" "rwm\0"
1455 "/run/systemd/inaccessible/blk\0" "rwm\0";
1459 NULSTR_FOREACH_PAIR(x
, y
, auto_devices
)
1460 (void) whitelist_device(prog
, path
, x
, y
);
1462 /* PTS (/dev/pts) devices may not be duplicated, but accessed */
1463 (void) whitelist_major(prog
, path
, "pts", 'c', "rw");
1466 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
1482 if (path_startswith(a
->path
, "/dev/"))
1483 (void) whitelist_device(prog
, path
, a
->path
, acc
);
1484 else if ((val
= startswith(a
->path
, "block-")))
1485 (void) whitelist_major(prog
, path
, val
, 'b', acc
);
1486 else if ((val
= startswith(a
->path
, "char-")))
1487 (void) whitelist_major(prog
, path
, val
, 'c', acc
);
1489 log_unit_debug(u
, "Ignoring device '%s' while writing cgroup attribute.", a
->path
);
1492 r
= cgroup_apply_device_bpf(u
, prog
, c
->device_policy
, c
->device_allow
);
1494 static bool warned
= false;
1496 log_full_errno(warned
? LOG_DEBUG
: LOG_WARNING
, r
,
1497 "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n"
1498 "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n"
1499 "(This warning is only shown for the first loaded unit using device ACL.)", u
->id
);
1505 if (apply_mask
& CGROUP_MASK_PIDS
) {
1508 /* So, the "pids" controller does not expose anything on the root cgroup, in order not to
1509 * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when
1510 * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a
1511 * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take
1512 * exclusive ownership of the sysctls, but we still want to honour things if the user sets
1513 * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit
1514 * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded)
1515 * it also counts. But if the user never set a limit through us (i.e. we are the default of
1516 * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on
1517 * the first time we set a limit. Note that this boolean is flushed out on manager reload,
1518 * which is desirable so that there's an official way to release control of the sysctl from
1519 * systemd: set the limit to unbounded and reload. */
1521 if (c
->tasks_max
!= CGROUP_LIMIT_MAX
) {
1522 u
->manager
->sysctl_pid_max_changed
= true;
1523 r
= procfs_tasks_set_limit(c
->tasks_max
);
1524 } else if (u
->manager
->sysctl_pid_max_changed
)
1525 r
= procfs_tasks_set_limit(TASKS_MAX
);
1529 log_unit_full(u
, LOG_LEVEL_CGROUP_WRITE(r
), r
,
1530 "Failed to write to tasks limit sysctls: %m");
1533 /* The attribute itself is not available on the host root cgroup, and in the container case we want to
1534 * leave it for the container manager. */
1535 if (!is_local_root
) {
1536 if (c
->tasks_max
!= CGROUP_LIMIT_MAX
) {
1537 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
1539 sprintf(buf
, "%" PRIu64
"\n", c
->tasks_max
);
1540 (void) set_attribute_and_warn(u
, "pids", "pids.max", buf
);
1542 (void) set_attribute_and_warn(u
, "pids", "pids.max", "max\n");
1546 if (apply_mask
& CGROUP_MASK_BPF_FIREWALL
)
1547 cgroup_apply_firewall(u
);
1550 static bool unit_get_needs_bpf_firewall(Unit
*u
) {
1555 c
= unit_get_cgroup_context(u
);
1559 if (c
->ip_accounting
||
1560 c
->ip_address_allow
||
1561 c
->ip_address_deny
||
1562 c
->ip_filters_ingress
||
1563 c
->ip_filters_egress
)
1566 /* If any parent slice has an IP access list defined, it applies too */
1567 for (p
= UNIT_DEREF(u
->slice
); p
; p
= UNIT_DEREF(p
->slice
)) {
1568 c
= unit_get_cgroup_context(p
);
1572 if (c
->ip_address_allow
||
1580 static CGroupMask
unit_get_cgroup_mask(Unit
*u
) {
1581 CGroupMask mask
= 0;
1586 c
= unit_get_cgroup_context(u
);
1590 /* Figure out which controllers we need, based on the cgroup context object */
1592 if (c
->cpu_accounting
)
1593 mask
|= get_cpu_accounting_mask();
1595 if (cgroup_context_has_cpu_weight(c
) ||
1596 cgroup_context_has_cpu_shares(c
) ||
1597 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
1598 mask
|= CGROUP_MASK_CPU
;
1600 if (c
->cpuset_cpus
.set
|| c
->cpuset_mems
.set
)
1601 mask
|= CGROUP_MASK_CPUSET
;
1603 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
1604 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1606 if (c
->memory_accounting
||
1607 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
1608 unit_has_unified_memory_config(u
))
1609 mask
|= CGROUP_MASK_MEMORY
;
1611 if (c
->device_allow
||
1612 c
->device_policy
!= CGROUP_AUTO
)
1613 mask
|= CGROUP_MASK_DEVICES
| CGROUP_MASK_BPF_DEVICES
;
1615 if (c
->tasks_accounting
||
1616 c
->tasks_max
!= CGROUP_LIMIT_MAX
)
1617 mask
|= CGROUP_MASK_PIDS
;
1619 return CGROUP_MASK_EXTEND_JOINED(mask
);
1622 static CGroupMask
unit_get_bpf_mask(Unit
*u
) {
1623 CGroupMask mask
= 0;
1625 /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children
1628 if (unit_get_needs_bpf_firewall(u
))
1629 mask
|= CGROUP_MASK_BPF_FIREWALL
;
1634 CGroupMask
unit_get_own_mask(Unit
*u
) {
1637 /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty
1638 * mask, as we shouldn't reflect it in the cgroup hierarchy then. */
1640 if (u
->load_state
!= UNIT_LOADED
)
1643 c
= unit_get_cgroup_context(u
);
1647 return (unit_get_cgroup_mask(u
) | unit_get_bpf_mask(u
) | unit_get_delegate_mask(u
)) & ~unit_get_ancestor_disable_mask(u
);
1650 CGroupMask
unit_get_delegate_mask(Unit
*u
) {
1653 /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the
1654 * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers.
1656 * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */
1658 if (!unit_cgroup_delegate(u
))
1661 if (cg_all_unified() <= 0) {
1664 e
= unit_get_exec_context(u
);
1665 if (e
&& !exec_context_maintains_privileges(e
))
1669 assert_se(c
= unit_get_cgroup_context(u
));
1670 return CGROUP_MASK_EXTEND_JOINED(c
->delegate_controllers
);
1673 CGroupMask
unit_get_members_mask(Unit
*u
) {
1676 /* Returns the mask of controllers all of the unit's children require, merged */
1678 if (u
->cgroup_members_mask_valid
)
1679 return u
->cgroup_members_mask
; /* Use cached value if possible */
1681 u
->cgroup_members_mask
= 0;
1683 if (u
->type
== UNIT_SLICE
) {
1688 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
], i
) {
1689 if (UNIT_DEREF(member
->slice
) == u
)
1690 u
->cgroup_members_mask
|= unit_get_subtree_mask(member
); /* note that this calls ourselves again, for the children */
1694 u
->cgroup_members_mask_valid
= true;
1695 return u
->cgroup_members_mask
;
1698 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
1701 /* Returns the mask of controllers all of the unit's siblings
1702 * require, i.e. the members mask of the unit's parent slice
1703 * if there is one. */
1705 if (UNIT_ISSET(u
->slice
))
1706 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
1708 return unit_get_subtree_mask(u
); /* we are the top-level slice */
1711 CGroupMask
unit_get_disable_mask(Unit
*u
) {
1714 c
= unit_get_cgroup_context(u
);
1718 return c
->disable_controllers
;
1721 CGroupMask
unit_get_ancestor_disable_mask(Unit
*u
) {
1725 mask
= unit_get_disable_mask(u
);
1727 /* Returns the mask of controllers which are marked as forcibly
1728 * disabled in any ancestor unit or the unit in question. */
1730 if (UNIT_ISSET(u
->slice
))
1731 mask
|= unit_get_ancestor_disable_mask(UNIT_DEREF(u
->slice
));
1736 CGroupMask
unit_get_subtree_mask(Unit
*u
) {
1738 /* Returns the mask of this subtree, meaning of the group
1739 * itself and its children. */
1741 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
1744 CGroupMask
unit_get_target_mask(Unit
*u
) {
1747 /* This returns the cgroup mask of all controllers to enable
1748 * for a specific cgroup, i.e. everything it needs itself,
1749 * plus all that its children need, plus all that its siblings
1750 * need. This is primarily useful on the legacy cgroup
1751 * hierarchy, where we need to duplicate each cgroup in each
1752 * hierarchy that shall be enabled for it. */
1754 mask
= unit_get_own_mask(u
) | unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
1756 if (mask
& CGROUP_MASK_BPF_FIREWALL
& ~u
->manager
->cgroup_supported
)
1757 emit_bpf_firewall_warning(u
);
1759 mask
&= u
->manager
->cgroup_supported
;
1760 mask
&= ~unit_get_ancestor_disable_mask(u
);
1765 CGroupMask
unit_get_enable_mask(Unit
*u
) {
1768 /* This returns the cgroup mask of all controllers to enable
1769 * for the children of a specific cgroup. This is primarily
1770 * useful for the unified cgroup hierarchy, where each cgroup
1771 * controls which controllers are enabled for its children. */
1773 mask
= unit_get_members_mask(u
);
1774 mask
&= u
->manager
->cgroup_supported
;
1775 mask
&= ~unit_get_ancestor_disable_mask(u
);
1780 void unit_invalidate_cgroup_members_masks(Unit
*u
) {
1783 /* Recurse invalidate the member masks cache all the way up the tree */
1784 u
->cgroup_members_mask_valid
= false;
1786 if (UNIT_ISSET(u
->slice
))
1787 unit_invalidate_cgroup_members_masks(UNIT_DEREF(u
->slice
));
1790 const char *unit_get_realized_cgroup_path(Unit
*u
, CGroupMask mask
) {
1792 /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */
1796 if (u
->cgroup_path
&&
1797 u
->cgroup_realized
&&
1798 FLAGS_SET(u
->cgroup_realized_mask
, mask
))
1799 return u
->cgroup_path
;
1801 u
= UNIT_DEREF(u
->slice
);
1807 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1808 return unit_get_realized_cgroup_path(userdata
, mask
);
1811 char *unit_default_cgroup_path(const Unit
*u
) {
1812 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1817 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1818 return strdup(u
->manager
->cgroup_root
);
1820 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1821 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1826 escaped
= cg_escape(u
->id
);
1830 return path_join(empty_to_root(u
->manager
->cgroup_root
), slice
, escaped
);
1833 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1834 _cleanup_free_
char *p
= NULL
;
1839 if (streq_ptr(u
->cgroup_path
, path
))
1849 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1854 unit_release_cgroup(u
);
1855 u
->cgroup_path
= TAKE_PTR(p
);
1860 int unit_watch_cgroup(Unit
*u
) {
1861 _cleanup_free_
char *events
= NULL
;
1866 /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if
1867 * cgroupv2 is available. */
1869 if (!u
->cgroup_path
)
1872 if (u
->cgroup_control_inotify_wd
>= 0)
1875 /* Only applies to the unified hierarchy */
1876 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1878 return log_error_errno(r
, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1882 /* No point in watch the top-level slice, it's never going to run empty. */
1883 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1886 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_control_inotify_wd_unit
, &trivial_hash_ops
);
1890 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1894 u
->cgroup_control_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1895 if (u
->cgroup_control_inotify_wd
< 0) {
1897 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1898 * is not an error */
1901 return log_unit_error_errno(u
, errno
, "Failed to add control inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1904 r
= hashmap_put(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
), u
);
1906 return log_unit_error_errno(u
, r
, "Failed to add control inotify watch descriptor to hash map: %m");
1911 int unit_watch_cgroup_memory(Unit
*u
) {
1912 _cleanup_free_
char *events
= NULL
;
1918 /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if
1919 * cgroupv2 is available. */
1921 if (!u
->cgroup_path
)
1924 c
= unit_get_cgroup_context(u
);
1928 /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie
1929 * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after
1931 if (!c
->memory_accounting
)
1934 /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and
1935 * we also don't want to generate a log message for each parent cgroup of a process. */
1936 if (u
->type
== UNIT_SLICE
)
1939 if (u
->cgroup_memory_inotify_wd
>= 0)
1942 /* Only applies to the unified hierarchy */
1943 r
= cg_all_unified();
1945 return log_error_errno(r
, "Failed to determine whether the memory controller is unified: %m");
1949 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_memory_inotify_wd_unit
, &trivial_hash_ops
);
1953 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "memory.events", &events
);
1957 u
->cgroup_memory_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1958 if (u
->cgroup_memory_inotify_wd
< 0) {
1960 if (errno
== ENOENT
) /* If the directory is already gone we don't need to track it, so this
1961 * is not an error */
1964 return log_unit_error_errno(u
, errno
, "Failed to add memory inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1967 r
= hashmap_put(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
), u
);
1969 return log_unit_error_errno(u
, r
, "Failed to add memory inotify watch descriptor to hash map: %m");
1974 int unit_pick_cgroup_path(Unit
*u
) {
1975 _cleanup_free_
char *path
= NULL
;
1983 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1986 path
= unit_default_cgroup_path(u
);
1990 r
= unit_set_cgroup_path(u
, path
);
1992 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1994 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1999 static int unit_create_cgroup(
2001 CGroupMask target_mask
,
2002 CGroupMask enable_mask
,
2003 ManagerState state
) {
2010 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2013 /* Figure out our cgroup path */
2014 r
= unit_pick_cgroup_path(u
);
2018 /* First, create our own group */
2019 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
2021 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
2024 /* Start watching it */
2025 (void) unit_watch_cgroup(u
);
2026 (void) unit_watch_cgroup_memory(u
);
2028 /* Preserve enabled controllers in delegated units, adjust others. */
2029 if (created
|| !u
->cgroup_realized
|| !unit_cgroup_delegate(u
)) {
2030 CGroupMask result_mask
= 0;
2032 /* Enable all controllers we need */
2033 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
, &result_mask
);
2035 log_unit_warning_errno(u
, r
, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
2037 /* If we just turned off a controller, this might release the controller for our parent too, let's
2038 * enqueue the parent for re-realization in that case again. */
2039 if (UNIT_ISSET(u
->slice
)) {
2040 CGroupMask turned_off
;
2042 turned_off
= (u
->cgroup_realized
? u
->cgroup_enabled_mask
& ~result_mask
: 0);
2043 if (turned_off
!= 0) {
2046 /* Force the parent to propagate the enable mask to the kernel again, by invalidating
2047 * the controller we just turned off. */
2049 for (parent
= UNIT_DEREF(u
->slice
); parent
; parent
= UNIT_DEREF(parent
->slice
))
2050 unit_invalidate_cgroup(parent
, turned_off
);
2054 /* Remember what's actually enabled now */
2055 u
->cgroup_enabled_mask
= result_mask
;
2058 /* Keep track that this is now realized */
2059 u
->cgroup_realized
= true;
2060 u
->cgroup_realized_mask
= target_mask
;
2062 if (u
->type
!= UNIT_SLICE
&& !unit_cgroup_delegate(u
)) {
2064 /* Then, possibly move things over, but not if
2065 * subgroups may contain processes, which is the case
2066 * for slice and delegation units. */
2067 r
= cg_migrate_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, u
->cgroup_path
, migrate_callback
, u
);
2069 log_unit_warning_errno(u
, r
, "Failed to migrate cgroup from to %s, ignoring: %m", u
->cgroup_path
);
2072 /* Set attributes */
2073 cgroup_context_apply(u
, target_mask
, state
);
2074 cgroup_xattr_apply(u
);
2079 static int unit_attach_pid_to_cgroup_via_bus(Unit
*u
, pid_t pid
, const char *suffix_path
) {
2080 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2086 if (MANAGER_IS_SYSTEM(u
->manager
))
2089 if (!u
->manager
->system_bus
)
2092 if (!u
->cgroup_path
)
2095 /* Determine this unit's cgroup path relative to our cgroup root */
2096 pp
= path_startswith(u
->cgroup_path
, u
->manager
->cgroup_root
);
2100 pp
= strjoina("/", pp
, suffix_path
);
2101 path_simplify(pp
, false);
2103 r
= sd_bus_call_method(u
->manager
->system_bus
,
2104 "org.freedesktop.systemd1",
2105 "/org/freedesktop/systemd1",
2106 "org.freedesktop.systemd1.Manager",
2107 "AttachProcessesToUnit",
2110 NULL
/* empty unit name means client's unit, i.e. us */, pp
, 1, (uint32_t) pid
);
2112 return log_unit_debug_errno(u
, r
, "Failed to attach unit process " PID_FMT
" via the bus: %s", pid
, bus_error_message(&error
, r
));
2117 int unit_attach_pids_to_cgroup(Unit
*u
, Set
*pids
, const char *suffix_path
) {
2118 CGroupMask delegated_mask
;
2126 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2129 if (set_isempty(pids
))
2132 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
2133 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
2134 r
= bpf_firewall_load_custom(u
);
2138 r
= unit_realize_cgroup(u
);
2142 if (isempty(suffix_path
))
2145 p
= prefix_roota(u
->cgroup_path
, suffix_path
);
2147 delegated_mask
= unit_get_delegate_mask(u
);
2150 SET_FOREACH(pidp
, pids
, i
) {
2151 pid_t pid
= PTR_TO_PID(pidp
);
2154 /* First, attach the PID to the main cgroup hierarchy */
2155 q
= cg_attach(SYSTEMD_CGROUP_CONTROLLER
, p
, pid
);
2157 log_unit_debug_errno(u
, q
, "Couldn't move process " PID_FMT
" to requested cgroup '%s': %m", pid
, p
);
2159 if (MANAGER_IS_USER(u
->manager
) && IN_SET(q
, -EPERM
, -EACCES
)) {
2162 /* If we are in a user instance, and we can't move the process ourselves due to
2163 * permission problems, let's ask the system instance about it instead. Since it's more
2164 * privileged it might be able to move the process across the leaves of a subtree who's
2165 * top node is not owned by us. */
2167 z
= unit_attach_pid_to_cgroup_via_bus(u
, pid
, suffix_path
);
2169 log_unit_debug_errno(u
, z
, "Couldn't move process " PID_FMT
" to requested cgroup '%s' via the system bus either: %m", pid
, p
);
2171 continue; /* When the bus thing worked via the bus we are fully done for this PID. */
2175 r
= q
; /* Remember first error */
2180 q
= cg_all_unified();
2186 /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the
2187 * innermost realized one */
2189 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++) {
2190 CGroupMask bit
= CGROUP_CONTROLLER_TO_MASK(c
);
2191 const char *realized
;
2193 if (!(u
->manager
->cgroup_supported
& bit
))
2196 /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */
2197 if (delegated_mask
& u
->cgroup_realized_mask
& bit
) {
2198 q
= cg_attach(cgroup_controller_to_string(c
), p
, pid
);
2200 continue; /* Success! */
2202 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",
2203 pid
, p
, cgroup_controller_to_string(c
));
2206 /* So this controller is either not delegate or realized, or something else weird happened. In
2207 * that case let's attach the PID at least to the closest cgroup up the tree that is
2209 realized
= unit_get_realized_cgroup_path(u
, bit
);
2211 continue; /* Not even realized in the root slice? Then let's not bother */
2213 q
= cg_attach(cgroup_controller_to_string(c
), realized
, pid
);
2215 log_unit_debug_errno(u
, q
, "Failed to attach PID " PID_FMT
" to realized cgroup %s in controller %s, ignoring: %m",
2216 pid
, realized
, cgroup_controller_to_string(c
));
2223 static bool unit_has_mask_realized(
2225 CGroupMask target_mask
,
2226 CGroupMask enable_mask
) {
2230 /* Returns true if this unit is fully realized. We check four things:
2232 * 1. Whether the cgroup was created at all
2233 * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1)
2234 * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2)
2235 * 4. Whether the invalidation mask is currently zero
2237 * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note
2238 * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for
2239 * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask
2240 * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they
2241 * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are
2242 * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they
2243 * simply don't matter. */
2245 return u
->cgroup_realized
&&
2246 ((u
->cgroup_realized_mask
^ target_mask
) & CGROUP_MASK_V1
) == 0 &&
2247 ((u
->cgroup_enabled_mask
^ enable_mask
) & CGROUP_MASK_V2
) == 0 &&
2248 u
->cgroup_invalidated_mask
== 0;
2251 static bool unit_has_mask_disables_realized(
2253 CGroupMask target_mask
,
2254 CGroupMask enable_mask
) {
2258 /* Returns true if all controllers which should be disabled are indeed disabled.
2260 * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is
2261 * already removed. */
2263 return !u
->cgroup_realized
||
2264 (FLAGS_SET(u
->cgroup_realized_mask
, target_mask
& CGROUP_MASK_V1
) &&
2265 FLAGS_SET(u
->cgroup_enabled_mask
, enable_mask
& CGROUP_MASK_V2
));
2268 static bool unit_has_mask_enables_realized(
2270 CGroupMask target_mask
,
2271 CGroupMask enable_mask
) {
2275 /* Returns true if all controllers which should be enabled are indeed enabled.
2277 * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything
2278 * we want to add is already added. */
2280 return u
->cgroup_realized
&&
2281 ((u
->cgroup_realized_mask
| target_mask
) & CGROUP_MASK_V1
) == (u
->cgroup_realized_mask
& CGROUP_MASK_V1
) &&
2282 ((u
->cgroup_enabled_mask
| enable_mask
) & CGROUP_MASK_V2
) == (u
->cgroup_enabled_mask
& CGROUP_MASK_V2
);
2285 void unit_add_to_cgroup_realize_queue(Unit
*u
) {
2288 if (u
->in_cgroup_realize_queue
)
2291 LIST_PREPEND(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2292 u
->in_cgroup_realize_queue
= true;
2295 static void unit_remove_from_cgroup_realize_queue(Unit
*u
) {
2298 if (!u
->in_cgroup_realize_queue
)
2301 LIST_REMOVE(cgroup_realize_queue
, u
->manager
->cgroup_realize_queue
, u
);
2302 u
->in_cgroup_realize_queue
= false;
2305 /* Controllers can only be enabled breadth-first, from the root of the
2306 * hierarchy downwards to the unit in question. */
2307 static int unit_realize_cgroup_now_enable(Unit
*u
, ManagerState state
) {
2308 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2313 /* First go deal with this unit's parent, or we won't be able to enable
2314 * any new controllers at this layer. */
2315 if (UNIT_ISSET(u
->slice
)) {
2316 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2321 target_mask
= unit_get_target_mask(u
);
2322 enable_mask
= unit_get_enable_mask(u
);
2324 /* We can only enable in this direction, don't try to disable anything.
2326 if (unit_has_mask_enables_realized(u
, target_mask
, enable_mask
))
2329 new_target_mask
= u
->cgroup_realized_mask
| target_mask
;
2330 new_enable_mask
= u
->cgroup_enabled_mask
| enable_mask
;
2332 return unit_create_cgroup(u
, new_target_mask
, new_enable_mask
, state
);
2335 /* Controllers can only be disabled depth-first, from the leaves of the
2336 * hierarchy upwards to the unit in question. */
2337 static int unit_realize_cgroup_now_disable(Unit
*u
, ManagerState state
) {
2344 if (u
->type
!= UNIT_SLICE
)
2347 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
], i
) {
2348 CGroupMask target_mask
, enable_mask
, new_target_mask
, new_enable_mask
;
2351 if (UNIT_DEREF(m
->slice
) != u
)
2354 /* The cgroup for this unit might not actually be fully
2355 * realised yet, in which case it isn't holding any controllers
2357 if (!m
->cgroup_path
)
2360 /* We must disable those below us first in order to release the
2362 if (m
->type
== UNIT_SLICE
)
2363 (void) unit_realize_cgroup_now_disable(m
, state
);
2365 target_mask
= unit_get_target_mask(m
);
2366 enable_mask
= unit_get_enable_mask(m
);
2368 /* We can only disable in this direction, don't try to enable
2370 if (unit_has_mask_disables_realized(m
, target_mask
, enable_mask
))
2373 new_target_mask
= m
->cgroup_realized_mask
& target_mask
;
2374 new_enable_mask
= m
->cgroup_enabled_mask
& enable_mask
;
2376 r
= unit_create_cgroup(m
, new_target_mask
, new_enable_mask
, state
);
2384 /* Check if necessary controllers and attributes for a unit are in place.
2386 * - If so, do nothing.
2387 * - If not, create paths, move processes over, and set attributes.
2389 * Controllers can only be *enabled* in a breadth-first way, and *disabled* in
2390 * a depth-first way. As such the process looks like this:
2392 * Suppose we have a cgroup hierarchy which looks like this:
2405 * 1. We want to realise cgroup "d" now.
2406 * 2. cgroup "a" has DisableControllers=cpu in the associated unit.
2407 * 3. cgroup "k" just started requesting the memory controller.
2409 * To make this work we must do the following in order:
2411 * 1. Disable CPU controller in k, j
2412 * 2. Disable CPU controller in d
2413 * 3. Enable memory controller in root
2414 * 4. Enable memory controller in a
2415 * 5. Enable memory controller in d
2416 * 6. Enable memory controller in k
2418 * Notice that we need to touch j in one direction, but not the other. We also
2419 * don't go beyond d when disabling -- it's up to "a" to get realized if it
2420 * wants to disable further. The basic rules are therefore:
2422 * - If you're disabling something, you need to realise all of the cgroups from
2423 * your recursive descendants to the root. This starts from the leaves.
2424 * - If you're enabling something, you need to realise from the root cgroup
2425 * downwards, but you don't need to iterate your recursive descendants.
2427 * Returns 0 on success and < 0 on failure. */
2428 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
2429 CGroupMask target_mask
, enable_mask
;
2434 unit_remove_from_cgroup_realize_queue(u
);
2436 target_mask
= unit_get_target_mask(u
);
2437 enable_mask
= unit_get_enable_mask(u
);
2439 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
2442 /* Disable controllers below us, if there are any */
2443 r
= unit_realize_cgroup_now_disable(u
, state
);
2447 /* Enable controllers above us, if there are any */
2448 if (UNIT_ISSET(u
->slice
)) {
2449 r
= unit_realize_cgroup_now_enable(UNIT_DEREF(u
->slice
), state
);
2454 /* Now actually deal with the cgroup we were trying to realise and set attributes */
2455 r
= unit_create_cgroup(u
, target_mask
, enable_mask
, state
);
2459 /* Now, reset the invalidation mask */
2460 u
->cgroup_invalidated_mask
= 0;
2464 unsigned manager_dispatch_cgroup_realize_queue(Manager
*m
) {
2472 state
= manager_state(m
);
2474 while ((i
= m
->cgroup_realize_queue
)) {
2475 assert(i
->in_cgroup_realize_queue
);
2477 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i
))) {
2478 /* Maybe things changed, and the unit is not actually active anymore? */
2479 unit_remove_from_cgroup_realize_queue(i
);
2483 r
= unit_realize_cgroup_now(i
, state
);
2485 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
2493 static void unit_add_siblings_to_cgroup_realize_queue(Unit
*u
) {
2496 /* This adds the siblings of the specified unit and the
2497 * siblings of all parent units to the cgroup queue. (But
2498 * neither the specified unit itself nor the parents.) */
2500 while ((slice
= UNIT_DEREF(u
->slice
))) {
2505 HASHMAP_FOREACH_KEY(v
, m
, u
->dependencies
[UNIT_BEFORE
], i
) {
2506 /* Skip units that have a dependency on the slice
2507 * but aren't actually in it. */
2508 if (UNIT_DEREF(m
->slice
) != slice
)
2511 /* No point in doing cgroup application for units
2512 * without active processes. */
2513 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
2516 /* If the unit doesn't need any new controllers
2517 * and has current ones realized, it doesn't need
2519 if (unit_has_mask_realized(m
,
2520 unit_get_target_mask(m
),
2521 unit_get_enable_mask(m
)))
2524 unit_add_to_cgroup_realize_queue(m
);
2531 int unit_realize_cgroup(Unit
*u
) {
2534 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
2537 /* So, here's the deal: when realizing the cgroups for this
2538 * unit, we need to first create all parents, but there's more
2539 * actually: for the weight-based controllers we also need to
2540 * make sure that all our siblings (i.e. units that are in the
2541 * same slice as we are) have cgroups, too. Otherwise, things
2542 * would become very uneven as each of their processes would
2543 * get as much resources as all our group together. This call
2544 * will synchronously create the parent cgroups, but will
2545 * defer work on the siblings to the next event loop
2548 /* Add all sibling slices to the cgroup queue. */
2549 unit_add_siblings_to_cgroup_realize_queue(u
);
2551 /* And realize this one now (and apply the values) */
2552 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
2555 void unit_release_cgroup(Unit
*u
) {
2558 /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call
2559 * when we close down everything for reexecution, where we really want to leave the cgroup in place. */
2561 if (u
->cgroup_path
) {
2562 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
2563 u
->cgroup_path
= mfree(u
->cgroup_path
);
2566 if (u
->cgroup_control_inotify_wd
>= 0) {
2567 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_control_inotify_wd
) < 0)
2568 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
);
2570 (void) hashmap_remove(u
->manager
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_control_inotify_wd
));
2571 u
->cgroup_control_inotify_wd
= -1;
2574 if (u
->cgroup_memory_inotify_wd
>= 0) {
2575 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_memory_inotify_wd
) < 0)
2576 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
);
2578 (void) hashmap_remove(u
->manager
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_memory_inotify_wd
));
2579 u
->cgroup_memory_inotify_wd
= -1;
2583 void unit_prune_cgroup(Unit
*u
) {
2589 /* Removes the cgroup, if empty and possible, and stops watching it. */
2591 if (!u
->cgroup_path
)
2594 (void) unit_get_cpu_usage(u
, NULL
); /* Cache the last CPU usage value before we destroy the cgroup */
2596 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
2598 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
2600 /* One reason we could have failed here is, that the cgroup still contains a process.
2601 * However, if the cgroup becomes removable at a later time, it might be removed when
2602 * the containing slice is stopped. So even if we failed now, this unit shouldn't assume
2603 * that the cgroup is still realized the next time it is started. Do not return early
2604 * on error, continue cleanup. */
2605 log_unit_full(u
, r
== -EBUSY
? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
2610 unit_release_cgroup(u
);
2612 u
->cgroup_realized
= false;
2613 u
->cgroup_realized_mask
= 0;
2614 u
->cgroup_enabled_mask
= 0;
2616 u
->bpf_device_control_installed
= bpf_program_unref(u
->bpf_device_control_installed
);
2619 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
2620 _cleanup_fclose_
FILE *f
= NULL
;
2621 pid_t pid
= 0, npid
;
2627 if (!u
->cgroup_path
)
2630 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
2634 while (cg_read_pid(f
, &npid
) > 0) {
2639 if (pid_is_my_child(npid
) == 0)
2643 /* Dang, there's more than one daemonized PID
2644 in this group, so we don't know what process
2645 is the main process. */
2656 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
2657 _cleanup_closedir_
DIR *d
= NULL
;
2658 _cleanup_fclose_
FILE *f
= NULL
;
2664 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
2670 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
2671 r
= unit_watch_pid(u
, pid
, false);
2672 if (r
< 0 && ret
>= 0)
2676 if (r
< 0 && ret
>= 0)
2680 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
2687 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
2688 _cleanup_free_
char *p
= NULL
;
2690 p
= path_join(empty_to_root(path
), fn
);
2696 r
= unit_watch_pids_in_path(u
, p
);
2697 if (r
< 0 && ret
>= 0)
2701 if (r
< 0 && ret
>= 0)
2708 int unit_synthesize_cgroup_empty_event(Unit
*u
) {
2713 /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility
2714 * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can
2715 * get as notification source as soon as we stopped having any useful PIDs to watch for. */
2717 if (!u
->cgroup_path
)
2720 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2723 if (r
> 0) /* On unified we have reliable notifications, and don't need this */
2726 if (!set_isempty(u
->pids
))
2729 unit_add_to_cgroup_empty_queue(u
);
2733 int unit_watch_all_pids(Unit
*u
) {
2738 /* Adds all PIDs from our cgroup to the set of PIDs we
2739 * watch. This is a fallback logic for cases where we do not
2740 * get reliable cgroup empty notifications: we try to use
2741 * SIGCHLD as replacement. */
2743 if (!u
->cgroup_path
)
2746 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
2749 if (r
> 0) /* On unified we can use proper notifications */
2752 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
2755 static int on_cgroup_empty_event(sd_event_source
*s
, void *userdata
) {
2756 Manager
*m
= userdata
;
2763 u
= m
->cgroup_empty_queue
;
2767 assert(u
->in_cgroup_empty_queue
);
2768 u
->in_cgroup_empty_queue
= false;
2769 LIST_REMOVE(cgroup_empty_queue
, m
->cgroup_empty_queue
, u
);
2771 if (m
->cgroup_empty_queue
) {
2772 /* More stuff queued, let's make sure we remain enabled */
2773 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2775 log_debug_errno(r
, "Failed to reenable cgroup empty event source, ignoring: %m");
2778 unit_add_to_gc_queue(u
);
2780 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
2781 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
2786 void unit_add_to_cgroup_empty_queue(Unit
*u
) {
2791 /* Note that there are four different ways how cgroup empty events reach us:
2793 * 1. On the unified hierarchy we get an inotify event on the cgroup
2795 * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket
2797 * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus
2799 * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as
2800 * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications.
2802 * Regardless which way we got the notification, we'll verify it here, and then add it to a separate
2803 * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use
2804 * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending
2805 * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the
2806 * case for scope units). */
2808 if (u
->in_cgroup_empty_queue
)
2811 /* Let's verify that the cgroup is really empty */
2812 if (!u
->cgroup_path
)
2814 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
2816 log_unit_debug_errno(u
, r
, "Failed to determine whether cgroup %s is empty: %m", u
->cgroup_path
);
2822 LIST_PREPEND(cgroup_empty_queue
, u
->manager
->cgroup_empty_queue
, u
);
2823 u
->in_cgroup_empty_queue
= true;
2825 /* Trigger the defer event */
2826 r
= sd_event_source_set_enabled(u
->manager
->cgroup_empty_event_source
, SD_EVENT_ONESHOT
);
2828 log_debug_errno(r
, "Failed to enable cgroup empty event source: %m");
2831 int unit_check_oom(Unit
*u
) {
2832 _cleanup_free_
char *oom_kill
= NULL
;
2837 if (!u
->cgroup_path
)
2840 r
= cg_get_keyed_attribute("memory", u
->cgroup_path
, "memory.events", STRV_MAKE("oom_kill"), &oom_kill
);
2842 return log_unit_debug_errno(u
, r
, "Failed to read oom_kill field of memory.events cgroup attribute: %m");
2844 r
= safe_atou64(oom_kill
, &c
);
2846 return log_unit_debug_errno(u
, r
, "Failed to parse oom_kill field: %m");
2848 increased
= c
> u
->oom_kill_last
;
2849 u
->oom_kill_last
= c
;
2854 log_struct(LOG_NOTICE
,
2855 "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR
,
2857 LOG_UNIT_INVOCATION_ID(u
),
2858 LOG_UNIT_MESSAGE(u
, "A process of this unit has been killed by the OOM killer."));
2860 if (UNIT_VTABLE(u
)->notify_cgroup_oom
)
2861 UNIT_VTABLE(u
)->notify_cgroup_oom(u
);
2866 static int on_cgroup_oom_event(sd_event_source
*s
, void *userdata
) {
2867 Manager
*m
= userdata
;
2874 u
= m
->cgroup_oom_queue
;
2878 assert(u
->in_cgroup_oom_queue
);
2879 u
->in_cgroup_oom_queue
= false;
2880 LIST_REMOVE(cgroup_oom_queue
, m
->cgroup_oom_queue
, u
);
2882 if (m
->cgroup_oom_queue
) {
2883 /* More stuff queued, let's make sure we remain enabled */
2884 r
= sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
2886 log_debug_errno(r
, "Failed to reenable cgroup oom event source, ignoring: %m");
2889 (void) unit_check_oom(u
);
2893 static void unit_add_to_cgroup_oom_queue(Unit
*u
) {
2898 if (u
->in_cgroup_oom_queue
)
2900 if (!u
->cgroup_path
)
2903 LIST_PREPEND(cgroup_oom_queue
, u
->manager
->cgroup_oom_queue
, u
);
2904 u
->in_cgroup_oom_queue
= true;
2906 /* Trigger the defer event */
2907 if (!u
->manager
->cgroup_oom_event_source
) {
2908 _cleanup_(sd_event_source_unrefp
) sd_event_source
*s
= NULL
;
2910 r
= sd_event_add_defer(u
->manager
->event
, &s
, on_cgroup_oom_event
, u
->manager
);
2912 log_error_errno(r
, "Failed to create cgroup oom event source: %m");
2916 r
= sd_event_source_set_priority(s
, SD_EVENT_PRIORITY_NORMAL
-8);
2918 log_error_errno(r
, "Failed to set priority of cgroup oom event source: %m");
2922 (void) sd_event_source_set_description(s
, "cgroup-oom");
2923 u
->manager
->cgroup_oom_event_source
= TAKE_PTR(s
);
2926 r
= sd_event_source_set_enabled(u
->manager
->cgroup_oom_event_source
, SD_EVENT_ONESHOT
);
2928 log_error_errno(r
, "Failed to enable cgroup oom event source: %m");
2931 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
2932 Manager
*m
= userdata
;
2939 union inotify_event_buffer buffer
;
2940 struct inotify_event
*e
;
2943 l
= read(fd
, &buffer
, sizeof(buffer
));
2945 if (IN_SET(errno
, EINTR
, EAGAIN
))
2948 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
2951 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
2955 /* Queue overflow has no watch descriptor */
2958 if (e
->mask
& IN_IGNORED
)
2959 /* The watch was just removed */
2962 /* Note that inotify might deliver events for a watch even after it was removed,
2963 * because it was queued before the removal. Let's ignore this here safely. */
2965 u
= hashmap_get(m
->cgroup_control_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2967 unit_add_to_cgroup_empty_queue(u
);
2969 u
= hashmap_get(m
->cgroup_memory_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
2971 unit_add_to_cgroup_oom_queue(u
);
2976 static int cg_bpf_mask_supported(CGroupMask
*ret
) {
2977 CGroupMask mask
= 0;
2980 /* BPF-based firewall */
2981 r
= bpf_firewall_supported();
2983 mask
|= CGROUP_MASK_BPF_FIREWALL
;
2985 /* BPF-based device access control */
2986 r
= bpf_devices_supported();
2988 mask
|= CGROUP_MASK_BPF_DEVICES
;
2994 int manager_setup_cgroup(Manager
*m
) {
2995 _cleanup_free_
char *path
= NULL
;
2996 const char *scope_path
;
3004 /* 1. Determine hierarchy */
3005 m
->cgroup_root
= mfree(m
->cgroup_root
);
3006 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
3008 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
3010 /* Chop off the init scope, if we are already located in it */
3011 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
3013 /* LEGACY: Also chop off the system slice if we are in
3014 * it. This is to support live upgrades from older systemd
3015 * versions where PID 1 was moved there. Also see
3016 * cg_get_root_path(). */
3017 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
3018 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
3020 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
3025 /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can
3026 * easily prepend it everywhere. */
3027 delete_trailing_chars(m
->cgroup_root
, "/");
3030 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
3032 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
3036 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
3038 all_unified
= cg_all_unified();
3039 if (all_unified
< 0)
3040 return log_error_errno(all_unified
, "Couldn't determine whether we are in all unified mode: %m");
3041 if (all_unified
> 0)
3042 log_debug("Unified cgroup hierarchy is located at %s.", path
);
3044 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
3046 return log_error_errno(r
, "Failed to determine whether systemd's own controller is in unified mode: %m");
3048 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path
);
3050 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY
". File system hierarchy is at %s.", path
);
3053 /* 3. Allocate cgroup empty defer event source */
3054 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
3055 r
= sd_event_add_defer(m
->event
, &m
->cgroup_empty_event_source
, on_cgroup_empty_event
, m
);
3057 return log_error_errno(r
, "Failed to create cgroup empty event source: %m");
3059 /* Schedule cgroup empty checks early, but after having processed service notification messages or
3060 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of
3061 * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */
3062 r
= sd_event_source_set_priority(m
->cgroup_empty_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
3064 return log_error_errno(r
, "Failed to set priority of cgroup empty event source: %m");
3066 r
= sd_event_source_set_enabled(m
->cgroup_empty_event_source
, SD_EVENT_OFF
);
3068 return log_error_errno(r
, "Failed to disable cgroup empty event source: %m");
3070 (void) sd_event_source_set_description(m
->cgroup_empty_event_source
, "cgroup-empty");
3072 /* 4. Install notifier inotify object, or agent */
3073 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
3075 /* In the unified hierarchy we can get cgroup empty notifications via inotify. */
3077 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
3078 safe_close(m
->cgroup_inotify_fd
);
3080 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
3081 if (m
->cgroup_inotify_fd
< 0)
3082 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
3084 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
3086 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
3088 /* Process cgroup empty notifications early. Note that when this event is dispatched it'll
3089 * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see
3090 * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
3091 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
3093 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
3095 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
3097 } else if (MANAGER_IS_SYSTEM(m
) && manager_owns_host_root_cgroup(m
) && !MANAGER_IS_TEST_RUN(m
)) {
3099 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
3100 * since it does not generate events when control groups with children run empty. */
3102 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUP_AGENT_PATH
);
3104 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
3106 log_debug("Installed release agent.");
3108 log_debug("Release agent already installed.");
3111 /* 5. Make sure we are in the special "init.scope" unit in the root slice. */
3112 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
3113 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
3115 /* Also, move all other userspace processes remaining in the root cgroup into that scope. */
3116 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
3118 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
3120 /* 6. And pin it, so that it cannot be unmounted */
3121 safe_close(m
->pin_cgroupfs_fd
);
3122 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
3123 if (m
->pin_cgroupfs_fd
< 0)
3124 return log_error_errno(errno
, "Failed to open pin file: %m");
3126 } else if (!MANAGER_IS_TEST_RUN(m
))
3127 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
3129 /* 7. Always enable hierarchical support if it exists... */
3130 if (!all_unified
&& !MANAGER_IS_TEST_RUN(m
))
3131 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
3133 /* 8. Figure out which controllers are supported */
3134 r
= cg_mask_supported(&m
->cgroup_supported
);
3136 return log_error_errno(r
, "Failed to determine supported controllers: %m");
3138 /* 9. Figure out which bpf-based pseudo-controllers are supported */
3139 r
= cg_bpf_mask_supported(&mask
);
3141 return log_error_errno(r
, "Failed to determine supported bpf-based pseudo-controllers: %m");
3142 m
->cgroup_supported
|= mask
;
3144 /* 10. Log which controllers are supported */
3145 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
3146 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
3151 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
3154 /* We can't really delete the group, since we are in it. But
3156 if (delete && m
->cgroup_root
&& m
->test_run_flags
!= MANAGER_TEST_RUN_MINIMAL
)
3157 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
3159 m
->cgroup_empty_event_source
= sd_event_source_unref(m
->cgroup_empty_event_source
);
3161 m
->cgroup_control_inotify_wd_unit
= hashmap_free(m
->cgroup_control_inotify_wd_unit
);
3162 m
->cgroup_memory_inotify_wd_unit
= hashmap_free(m
->cgroup_memory_inotify_wd_unit
);
3164 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
3165 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
3167 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
3169 m
->cgroup_root
= mfree(m
->cgroup_root
);
3172 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
3179 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
3183 p
= strdupa(cgroup
);
3187 e
= strrchr(p
, '/');
3189 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
3193 u
= hashmap_get(m
->cgroup_unit
, p
);
3199 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
3200 _cleanup_free_
char *cgroup
= NULL
;
3204 if (!pid_is_valid(pid
))
3207 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
) < 0)
3210 return manager_get_unit_by_cgroup(m
, cgroup
);
3213 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
3218 /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most
3219 * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most
3220 * relevant one as children of the process will be assigned to that one, too, before all else. */
3222 if (!pid_is_valid(pid
))
3225 if (pid
== getpid_cached())
3226 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
3228 u
= manager_get_unit_by_pid_cgroup(m
, pid
);
3232 u
= hashmap_get(m
->watch_pids
, PID_TO_PTR(pid
));
3236 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-pid
));
3243 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
3249 /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process
3250 * or from the --system instance */
3252 log_debug("Got cgroup empty notification for: %s", cgroup
);
3254 u
= manager_get_unit_by_cgroup(m
, cgroup
);
3258 unit_add_to_cgroup_empty_queue(u
);
3262 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
3263 _cleanup_free_
char *v
= NULL
;
3269 if (!UNIT_CGROUP_BOOL(u
, memory_accounting
))
3272 if (!u
->cgroup_path
)
3275 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3276 if (unit_has_host_root_cgroup(u
))
3277 return procfs_memory_get_used(ret
);
3279 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
3282 r
= cg_all_unified();
3286 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.current", &v
);
3288 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.usage_in_bytes", &v
);
3294 return safe_atou64(v
, ret
);
3297 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
3298 _cleanup_free_
char *v
= NULL
;
3304 if (!UNIT_CGROUP_BOOL(u
, tasks_accounting
))
3307 if (!u
->cgroup_path
)
3310 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3311 if (unit_has_host_root_cgroup(u
))
3312 return procfs_tasks_get_current(ret
);
3314 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
3317 r
= cg_get_attribute("pids", u
->cgroup_path
, "pids.current", &v
);
3323 return safe_atou64(v
, ret
);
3326 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
3327 _cleanup_free_
char *v
= NULL
;
3334 if (!u
->cgroup_path
)
3337 /* The root cgroup doesn't expose this information, let's get it from /proc instead */
3338 if (unit_has_host_root_cgroup(u
))
3339 return procfs_cpu_get_usage(ret
);
3341 /* Requisite controllers for CPU accounting are not enabled */
3342 if ((get_cpu_accounting_mask() & ~u
->cgroup_realized_mask
) != 0)
3345 r
= cg_all_unified();
3349 _cleanup_free_
char *val
= NULL
;
3352 r
= cg_get_keyed_attribute("cpu", u
->cgroup_path
, "cpu.stat", STRV_MAKE("usage_usec"), &val
);
3353 if (IN_SET(r
, -ENOENT
, -ENXIO
))
3358 r
= safe_atou64(val
, &us
);
3362 ns
= us
* NSEC_PER_USEC
;
3364 r
= cg_get_attribute("cpuacct", u
->cgroup_path
, "cpuacct.usage", &v
);
3370 r
= safe_atou64(v
, &ns
);
3379 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
3385 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
3386 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
3387 * call this function with a NULL return value. */
3389 if (!UNIT_CGROUP_BOOL(u
, cpu_accounting
))
3392 r
= unit_get_cpu_usage_raw(u
, &ns
);
3393 if (r
== -ENODATA
&& u
->cpu_usage_last
!= NSEC_INFINITY
) {
3394 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
3398 *ret
= u
->cpu_usage_last
;
3404 if (ns
> u
->cpu_usage_base
)
3405 ns
-= u
->cpu_usage_base
;
3409 u
->cpu_usage_last
= ns
;
3416 int unit_get_ip_accounting(
3418 CGroupIPAccountingMetric metric
,
3425 assert(metric
>= 0);
3426 assert(metric
< _CGROUP_IP_ACCOUNTING_METRIC_MAX
);
3429 if (!UNIT_CGROUP_BOOL(u
, ip_accounting
))
3432 fd
= IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_INGRESS_PACKETS
) ?
3433 u
->ip_accounting_ingress_map_fd
:
3434 u
->ip_accounting_egress_map_fd
;
3438 if (IN_SET(metric
, CGROUP_IP_INGRESS_BYTES
, CGROUP_IP_EGRESS_BYTES
))
3439 r
= bpf_firewall_read_accounting(fd
, &value
, NULL
);
3441 r
= bpf_firewall_read_accounting(fd
, NULL
, &value
);
3445 /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile
3446 * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the
3447 * ip_accounting_extra[] field, and add them in here transparently. */
3449 *ret
= value
+ u
->ip_accounting_extra
[metric
];
3454 static int unit_get_io_accounting_raw(Unit
*u
, uint64_t ret
[static _CGROUP_IO_ACCOUNTING_METRIC_MAX
]) {
3455 static const char *const field_names
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {
3456 [CGROUP_IO_READ_BYTES
] = "rbytes=",
3457 [CGROUP_IO_WRITE_BYTES
] = "wbytes=",
3458 [CGROUP_IO_READ_OPERATIONS
] = "rios=",
3459 [CGROUP_IO_WRITE_OPERATIONS
] = "wios=",
3461 uint64_t acc
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
] = {};
3462 _cleanup_free_
char *path
= NULL
;
3463 _cleanup_fclose_
FILE *f
= NULL
;
3468 if (!u
->cgroup_path
)
3471 if (unit_has_host_root_cgroup(u
))
3472 return -ENODATA
; /* TODO: return useful data for the top-level cgroup */
3474 r
= cg_all_unified();
3477 if (r
== 0) /* TODO: support cgroupv1 */
3480 if (!FLAGS_SET(u
->cgroup_realized_mask
, CGROUP_MASK_IO
))
3483 r
= cg_get_path("io", u
->cgroup_path
, "io.stat", &path
);
3487 f
= fopen(path
, "re");
3492 _cleanup_free_
char *line
= NULL
;
3495 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3502 p
+= strcspn(p
, WHITESPACE
); /* Skip over device major/minor */
3503 p
+= strspn(p
, WHITESPACE
); /* Skip over following whitespace */
3506 _cleanup_free_
char *word
= NULL
;
3508 r
= extract_first_word(&p
, &word
, NULL
, EXTRACT_RETAIN_ESCAPE
);
3514 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3517 x
= startswith(word
, field_names
[i
]);
3521 r
= safe_atou64(x
, &w
);
3525 /* Sum up the stats of all devices */
3533 memcpy(ret
, acc
, sizeof(acc
));
3537 int unit_get_io_accounting(
3539 CGroupIOAccountingMetric metric
,
3543 uint64_t raw
[_CGROUP_IO_ACCOUNTING_METRIC_MAX
];
3546 /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */
3548 if (!UNIT_CGROUP_BOOL(u
, io_accounting
))
3551 if (allow_cache
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3554 r
= unit_get_io_accounting_raw(u
, raw
);
3555 if (r
== -ENODATA
&& u
->io_accounting_last
[metric
] != UINT64_MAX
)
3560 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++) {
3561 /* Saturated subtraction */
3562 if (raw
[i
] > u
->io_accounting_base
[i
])
3563 u
->io_accounting_last
[i
] = raw
[i
] - u
->io_accounting_base
[i
];
3565 u
->io_accounting_last
[i
] = 0;
3570 *ret
= u
->io_accounting_last
[metric
];
3575 int unit_reset_cpu_accounting(Unit
*u
) {
3580 u
->cpu_usage_last
= NSEC_INFINITY
;
3582 r
= unit_get_cpu_usage_raw(u
, &u
->cpu_usage_base
);
3584 u
->cpu_usage_base
= 0;
3591 int unit_reset_ip_accounting(Unit
*u
) {
3596 if (u
->ip_accounting_ingress_map_fd
>= 0)
3597 r
= bpf_firewall_reset_accounting(u
->ip_accounting_ingress_map_fd
);
3599 if (u
->ip_accounting_egress_map_fd
>= 0)
3600 q
= bpf_firewall_reset_accounting(u
->ip_accounting_egress_map_fd
);
3602 zero(u
->ip_accounting_extra
);
3604 return r
< 0 ? r
: q
;
3607 int unit_reset_io_accounting(Unit
*u
) {
3612 for (CGroupIOAccountingMetric i
= 0; i
< _CGROUP_IO_ACCOUNTING_METRIC_MAX
; i
++)
3613 u
->io_accounting_last
[i
] = UINT64_MAX
;
3615 r
= unit_get_io_accounting_raw(u
, u
->io_accounting_base
);
3617 zero(u
->io_accounting_base
);
3624 int unit_reset_accounting(Unit
*u
) {
3629 r
= unit_reset_cpu_accounting(u
);
3630 q
= unit_reset_io_accounting(u
);
3631 v
= unit_reset_ip_accounting(u
);
3633 return r
< 0 ? r
: q
< 0 ? q
: v
;
3636 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
3639 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3645 /* always invalidate compat pairs together */
3646 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
3647 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
3649 if (m
& (CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
))
3650 m
|= CGROUP_MASK_CPU
| CGROUP_MASK_CPUACCT
;
3652 if (FLAGS_SET(u
->cgroup_invalidated_mask
, m
)) /* NOP? */
3655 u
->cgroup_invalidated_mask
|= m
;
3656 unit_add_to_cgroup_realize_queue(u
);
3659 void unit_invalidate_cgroup_bpf(Unit
*u
) {
3662 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
3665 if (u
->cgroup_invalidated_mask
& CGROUP_MASK_BPF_FIREWALL
) /* NOP? */
3668 u
->cgroup_invalidated_mask
|= CGROUP_MASK_BPF_FIREWALL
;
3669 unit_add_to_cgroup_realize_queue(u
);
3671 /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access
3672 * list of our children includes our own. */
3673 if (u
->type
== UNIT_SLICE
) {
3678 HASHMAP_FOREACH_KEY(v
, member
, u
->dependencies
[UNIT_BEFORE
], i
) {
3679 if (UNIT_DEREF(member
->slice
) == u
)
3680 unit_invalidate_cgroup_bpf(member
);
3685 bool unit_cgroup_delegate(Unit
*u
) {
3690 if (!UNIT_VTABLE(u
)->can_delegate
)
3693 c
= unit_get_cgroup_context(u
);
3700 void manager_invalidate_startup_units(Manager
*m
) {
3706 SET_FOREACH(u
, m
->startup_units
, i
)
3707 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
3710 static int unit_get_nice(Unit
*u
) {
3713 ec
= unit_get_exec_context(u
);
3714 return ec
? ec
->nice
: 0;
3717 static uint64_t unit_get_cpu_weight(Unit
*u
) {
3718 ManagerState state
= manager_state(u
->manager
);
3721 cc
= unit_get_cgroup_context(u
);
3722 return cc
? cgroup_context_cpu_weight(cc
, state
) : CGROUP_WEIGHT_DEFAULT
;
3725 int compare_job_priority(const void *a
, const void *b
) {
3726 const Job
*x
= a
, *y
= b
;
3728 uint64_t weight_x
, weight_y
;
3731 if ((ret
= CMP(x
->unit
->type
, y
->unit
->type
)) != 0)
3734 weight_x
= unit_get_cpu_weight(x
->unit
);
3735 weight_y
= unit_get_cpu_weight(y
->unit
);
3737 if ((ret
= CMP(weight_x
, weight_y
)) != 0)
3740 nice_x
= unit_get_nice(x
->unit
);
3741 nice_y
= unit_get_nice(y
->unit
);
3743 if ((ret
= CMP(nice_x
, nice_y
)) != 0)
3746 return strcmp(x
->unit
->id
, y
->unit
->id
);
3749 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
3750 [CGROUP_AUTO
] = "auto",
3751 [CGROUP_CLOSED
] = "closed",
3752 [CGROUP_STRICT
] = "strict",
3755 int unit_get_cpuset(Unit
*u
, CPUSet
*cpus
, const char *name
) {
3756 _cleanup_free_
char *v
= NULL
;
3762 if (!u
->cgroup_path
)
3765 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUSET
) == 0)
3768 r
= cg_all_unified();
3774 r
= cg_get_attribute("cpuset", u
->cgroup_path
, name
, &v
);
3780 return parse_cpu_set_full(v
, cpus
, false, NULL
, NULL
, 0, NULL
);
3783 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);