]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
8e274523 | 2 | |
c6c18be3 | 3 | #include <fcntl.h> |
8c6db833 | 4 | |
afcfaa69 LP |
5 | #include "sd-messages.h" |
6 | ||
a4817536 | 7 | #include "af-list.h" |
b5efdb8a | 8 | #include "alloc-util.h" |
18c528e9 | 9 | #include "blockdev-util.h" |
d8b4d14d | 10 | #include "bpf-devices.h" |
906c06f6 | 11 | #include "bpf-firewall.h" |
506ea51b | 12 | #include "bpf-foreign.h" |
cd09a5f3 | 13 | #include "bpf-socket-bind.h" |
45c2e068 | 14 | #include "btrfs-util.h" |
6592b975 | 15 | #include "bus-error.h" |
78fa2f91 | 16 | #include "bus-locator.h" |
fdb3deca | 17 | #include "cgroup-setup.h" |
03a7b521 | 18 | #include "cgroup-util.h" |
3ffd4af2 | 19 | #include "cgroup.h" |
7176f06c | 20 | #include "devnum-util.h" |
3ffd4af2 | 21 | #include "fd-util.h" |
0d39fa9c | 22 | #include "fileio.h" |
dc7d69b3 | 23 | #include "firewall-util.h" |
84ebe6f0 | 24 | #include "in-addr-prefix-util.h" |
9e5fd717 | 25 | #include "inotify-util.h" |
d9e45bc3 | 26 | #include "io-util.h" |
5587ce7f | 27 | #include "ip-protocol-list.h" |
3a0f06c4 | 28 | #include "limits-util.h" |
d9e45bc3 | 29 | #include "nulstr-util.h" |
6bedfcbb | 30 | #include "parse-util.h" |
9eb977db | 31 | #include "path-util.h" |
1ead0b2a | 32 | #include "percent-util.h" |
03a7b521 | 33 | #include "process-util.h" |
c36a69f4 | 34 | #include "procfs-util.h" |
6f50d4f7 | 35 | #include "restrict-ifaces.h" |
84c01612 | 36 | #include "set.h" |
9444b1f2 | 37 | #include "special.h" |
906c06f6 | 38 | #include "stdio-util.h" |
8b43440b | 39 | #include "string-table.h" |
07630cea | 40 | #include "string-util.h" |
cc6271f1 | 41 | #include "virt.h" |
8e274523 | 42 | |
b1994387 ILG |
43 | #if BPF_FRAMEWORK |
44 | #include "bpf-dlopen.h" | |
45 | #include "bpf-link.h" | |
46 | #include "bpf/restrict_fs/restrict-fs-skel.h" | |
47 | #endif | |
48 | ||
10f28641 | 49 | #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
9a054909 | 50 | |
39b9fefb LP |
51 | /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access |
52 | * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask | |
53 | * out specific attributes from us. */ | |
54 | #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING) | |
55 | ||
94f0b13b | 56 | uint64_t cgroup_tasks_max_resolve(const CGroupTasksMax *tasks_max) { |
3a0f06c4 ZJS |
57 | if (tasks_max->scale == 0) |
58 | return tasks_max->value; | |
59 | ||
60 | return system_tasks_max_scale(tasks_max->value, tasks_max->scale); | |
61 | } | |
62 | ||
611c4f8a | 63 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
64 | assert(m); |
65 | ||
66 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
67 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
68 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
69 | * we run in any kind of container virtualization. */ | |
70 | ||
28cfdc5a LP |
71 | if (MANAGER_IS_USER(m)) |
72 | return false; | |
73 | ||
cc6271f1 LP |
74 | if (detect_container() > 0) |
75 | return false; | |
76 | ||
57ea45e1 | 77 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
78 | } |
79 | ||
9dfb6a3a PM |
80 | bool unit_has_startup_cgroup_constraints(Unit *u) { |
81 | assert(u); | |
82 | ||
83 | /* Returns true if this unit has any directives which apply during | |
84 | * startup/shutdown phases. */ | |
85 | ||
86 | CGroupContext *c; | |
87 | ||
88 | c = unit_get_cgroup_context(u); | |
89 | if (!c) | |
90 | return false; | |
91 | ||
92 | return c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
93 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
94 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
95 | c->startup_cpuset_cpus.set || | |
53fda560 LB |
96 | c->startup_cpuset_mems.set || |
97 | c->startup_memory_high_set || | |
98 | c->startup_memory_max_set || | |
99 | c->startup_memory_swap_max_set|| | |
100 | c->startup_memory_zswap_max_set || | |
101 | c->startup_memory_low_set; | |
9dfb6a3a PM |
102 | } |
103 | ||
611c4f8a | 104 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
105 | assert(u); |
106 | ||
cc6271f1 LP |
107 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
108 | * the manager manages the root cgroup. */ | |
f3725e64 | 109 | |
611c4f8a | 110 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
111 | return false; |
112 | ||
cc6271f1 | 113 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
114 | } |
115 | ||
293d32df LP |
116 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
117 | int r; | |
118 | ||
119 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
120 | if (r < 0) | |
8ed6f81b | 121 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", |
6178e2f8 | 122 | strna(attribute), empty_to_root(u->cgroup_path), (int) strcspn(value, NEWLINE), value); |
293d32df LP |
123 | |
124 | return r; | |
125 | } | |
126 | ||
2b40998d | 127 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
128 | static bool cgroup_compat_warned = false; |
129 | ||
130 | if (cgroup_compat_warned) | |
131 | return; | |
132 | ||
cc6271f1 LP |
133 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
134 | "See cgroup-compat debug messages for details."); | |
135 | ||
128fadc9 TH |
136 | cgroup_compat_warned = true; |
137 | } | |
138 | ||
139 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
140 | cgroup_compat_warn(); \ | |
141 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 142 | } while (false) |
128fadc9 | 143 | |
4ad49000 LP |
144 | void cgroup_context_init(CGroupContext *c) { |
145 | assert(c); | |
146 | ||
154eb43f LB |
147 | /* Initialize everything to the kernel defaults. When initializing a bool member to 'true', make |
148 | * sure to serialize in execute-serialize.c using serialize_bool() instead of | |
149 | * serialize_bool_elide(), as sd-executor will initialize here to 'true', but serialize_bool_elide() | |
150 | * skips serialization if the value is 'false' (as that's the common default), so if the value at | |
151 | * runtime is zero it would be lost after deserialization. Same when initializing uint64_t and other | |
152 | * values, update/add a conditional serialization check. This is to minimize the amount of | |
153 | * serialized data that is sent to the sd-executor, so that there is less work to do on the default | |
154 | * cases. */ | |
4ad49000 | 155 | |
de8a711a LP |
156 | *c = (CGroupContext) { |
157 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
158 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
159 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
10f28641 | 160 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 161 | |
de8a711a LP |
162 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
163 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 164 | |
de8a711a | 165 | .memory_high = CGROUP_LIMIT_MAX, |
53fda560 | 166 | .startup_memory_high = CGROUP_LIMIT_MAX, |
de8a711a | 167 | .memory_max = CGROUP_LIMIT_MAX, |
53fda560 | 168 | .startup_memory_max = CGROUP_LIMIT_MAX, |
de8a711a | 169 | .memory_swap_max = CGROUP_LIMIT_MAX, |
53fda560 | 170 | .startup_memory_swap_max = CGROUP_LIMIT_MAX, |
d7fe0a67 | 171 | .memory_zswap_max = CGROUP_LIMIT_MAX, |
53fda560 | 172 | .startup_memory_zswap_max = CGROUP_LIMIT_MAX, |
da4d897e | 173 | |
de8a711a | 174 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 175 | |
de8a711a LP |
176 | .io_weight = CGROUP_WEIGHT_INVALID, |
177 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 178 | |
de8a711a LP |
179 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
180 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 181 | |
94f0b13b | 182 | .tasks_max = CGROUP_TASKS_MAX_UNSET, |
4d824a4e AZ |
183 | |
184 | .moom_swap = MANAGED_OOM_AUTO, | |
185 | .moom_mem_pressure = MANAGED_OOM_AUTO, | |
4e806bfa | 186 | .moom_preference = MANAGED_OOM_PREFERENCE_NONE, |
6bb00842 LP |
187 | |
188 | .memory_pressure_watch = _CGROUP_PRESSURE_WATCH_INVALID, | |
189 | .memory_pressure_threshold_usec = USEC_INFINITY, | |
de8a711a | 190 | }; |
4ad49000 | 191 | } |
8e274523 | 192 | |
9c02eb28 | 193 | int cgroup_context_add_io_device_weight_dup(CGroupContext *c, const CGroupIODeviceWeight *w) { |
84c01612 MS |
194 | _cleanup_free_ CGroupIODeviceWeight *n = NULL; |
195 | ||
196 | assert(c); | |
197 | assert(w); | |
198 | ||
9c02eb28 | 199 | n = new(CGroupIODeviceWeight, 1); |
84c01612 MS |
200 | if (!n) |
201 | return -ENOMEM; | |
202 | ||
9c02eb28 MY |
203 | *n = (CGroupIODeviceWeight) { |
204 | .path = strdup(w->path), | |
205 | .weight = w->weight, | |
206 | }; | |
84c01612 MS |
207 | if (!n->path) |
208 | return -ENOMEM; | |
84c01612 MS |
209 | |
210 | LIST_PREPEND(device_weights, c->io_device_weights, TAKE_PTR(n)); | |
211 | return 0; | |
212 | } | |
213 | ||
9c02eb28 | 214 | int cgroup_context_add_io_device_limit_dup(CGroupContext *c, const CGroupIODeviceLimit *l) { |
84c01612 MS |
215 | _cleanup_free_ CGroupIODeviceLimit *n = NULL; |
216 | ||
217 | assert(c); | |
218 | assert(l); | |
219 | ||
220 | n = new0(CGroupIODeviceLimit, 1); | |
cae58298 | 221 | if (!n) |
84c01612 MS |
222 | return -ENOMEM; |
223 | ||
224 | n->path = strdup(l->path); | |
225 | if (!n->path) | |
226 | return -ENOMEM; | |
227 | ||
228 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
229 | n->limits[type] = l->limits[type]; | |
230 | ||
231 | LIST_PREPEND(device_limits, c->io_device_limits, TAKE_PTR(n)); | |
232 | return 0; | |
233 | } | |
234 | ||
9c02eb28 | 235 | int cgroup_context_add_io_device_latency_dup(CGroupContext *c, const CGroupIODeviceLatency *l) { |
84c01612 MS |
236 | _cleanup_free_ CGroupIODeviceLatency *n = NULL; |
237 | ||
238 | assert(c); | |
239 | assert(l); | |
240 | ||
9c02eb28 | 241 | n = new(CGroupIODeviceLatency, 1); |
84c01612 MS |
242 | if (!n) |
243 | return -ENOMEM; | |
244 | ||
9c02eb28 MY |
245 | *n = (CGroupIODeviceLatency) { |
246 | .path = strdup(l->path), | |
247 | .target_usec = l->target_usec, | |
248 | }; | |
84c01612 MS |
249 | if (!n->path) |
250 | return -ENOMEM; | |
251 | ||
84c01612 MS |
252 | LIST_PREPEND(device_latencies, c->io_device_latencies, TAKE_PTR(n)); |
253 | return 0; | |
254 | } | |
255 | ||
9c02eb28 | 256 | int cgroup_context_add_block_io_device_weight_dup(CGroupContext *c, const CGroupBlockIODeviceWeight *w) { |
84c01612 MS |
257 | _cleanup_free_ CGroupBlockIODeviceWeight *n = NULL; |
258 | ||
259 | assert(c); | |
260 | assert(w); | |
261 | ||
9c02eb28 | 262 | n = new(CGroupBlockIODeviceWeight, 1); |
84c01612 MS |
263 | if (!n) |
264 | return -ENOMEM; | |
265 | ||
9c02eb28 MY |
266 | *n = (CGroupBlockIODeviceWeight) { |
267 | .path = strdup(w->path), | |
268 | .weight = w->weight, | |
269 | }; | |
84c01612 MS |
270 | if (!n->path) |
271 | return -ENOMEM; | |
272 | ||
84c01612 MS |
273 | LIST_PREPEND(device_weights, c->blockio_device_weights, TAKE_PTR(n)); |
274 | return 0; | |
275 | } | |
276 | ||
9c02eb28 | 277 | int cgroup_context_add_block_io_device_bandwidth_dup(CGroupContext *c, const CGroupBlockIODeviceBandwidth *b) { |
84c01612 MS |
278 | _cleanup_free_ CGroupBlockIODeviceBandwidth *n = NULL; |
279 | ||
280 | assert(c); | |
281 | assert(b); | |
282 | ||
9c02eb28 | 283 | n = new(CGroupBlockIODeviceBandwidth, 1); |
84c01612 MS |
284 | if (!n) |
285 | return -ENOMEM; | |
286 | ||
287 | *n = (CGroupBlockIODeviceBandwidth) { | |
288 | .rbps = b->rbps, | |
289 | .wbps = b->wbps, | |
290 | }; | |
291 | ||
292 | LIST_PREPEND(device_bandwidths, c->blockio_device_bandwidths, TAKE_PTR(n)); | |
293 | return 0; | |
294 | } | |
295 | ||
9c02eb28 | 296 | int cgroup_context_add_device_allow_dup(CGroupContext *c, const CGroupDeviceAllow *a) { |
84c01612 MS |
297 | _cleanup_free_ CGroupDeviceAllow *n = NULL; |
298 | ||
299 | assert(c); | |
300 | assert(a); | |
301 | ||
9c02eb28 | 302 | n = new(CGroupDeviceAllow, 1); |
84c01612 MS |
303 | if (!n) |
304 | return -ENOMEM; | |
305 | ||
9c02eb28 MY |
306 | *n = (CGroupDeviceAllow) { |
307 | .path = strdup(a->path), | |
308 | .permissions = a->permissions, | |
309 | }; | |
84c01612 MS |
310 | if (!n->path) |
311 | return -ENOMEM; | |
312 | ||
84c01612 MS |
313 | LIST_PREPEND(device_allow, c->device_allow, TAKE_PTR(n)); |
314 | return 0; | |
315 | } | |
316 | ||
9c02eb28 | 317 | static int cgroup_context_add_socket_bind_item_dup(CGroupContext *c, const CGroupSocketBindItem *i, CGroupSocketBindItem *h) { |
84c01612 MS |
318 | _cleanup_free_ CGroupSocketBindItem *n = NULL; |
319 | ||
320 | assert(c); | |
321 | assert(i); | |
322 | ||
9c02eb28 | 323 | n = new(CGroupSocketBindItem, 1); |
84c01612 MS |
324 | if (!n) |
325 | return -ENOMEM; | |
326 | ||
327 | *n = (CGroupSocketBindItem) { | |
328 | .address_family = i->address_family, | |
329 | .ip_protocol = i->ip_protocol, | |
330 | .nr_ports = i->nr_ports, | |
331 | .port_min = i->port_min, | |
332 | }; | |
333 | ||
334 | LIST_PREPEND(socket_bind_items, h, TAKE_PTR(n)); | |
335 | return 0; | |
336 | } | |
337 | ||
9c02eb28 | 338 | int cgroup_context_add_socket_bind_item_allow_dup(CGroupContext *c, const CGroupSocketBindItem *i) { |
84c01612 MS |
339 | return cgroup_context_add_socket_bind_item_dup(c, i, c->socket_bind_allow); |
340 | } | |
341 | ||
9c02eb28 | 342 | int cgroup_context_add_socket_bind_item_deny_dup(CGroupContext *c, const CGroupSocketBindItem *i) { |
84c01612 MS |
343 | return cgroup_context_add_socket_bind_item_dup(c, i, c->socket_bind_deny); |
344 | } | |
345 | ||
346 | int cgroup_context_copy(CGroupContext *dst, const CGroupContext *src) { | |
347 | struct in_addr_prefix *i; | |
348 | char *iface; | |
349 | int r; | |
350 | ||
351 | assert(src); | |
352 | assert(dst); | |
353 | ||
354 | dst->cpu_accounting = src->cpu_accounting; | |
355 | dst->io_accounting = src->io_accounting; | |
356 | dst->blockio_accounting = src->blockio_accounting; | |
357 | dst->memory_accounting = src->memory_accounting; | |
358 | dst->tasks_accounting = src->tasks_accounting; | |
359 | dst->ip_accounting = src->ip_accounting; | |
360 | ||
2717d36d | 361 | dst->memory_oom_group = src->memory_oom_group; |
84c01612 MS |
362 | |
363 | dst->cpu_weight = src->cpu_weight; | |
364 | dst->startup_cpu_weight = src->startup_cpu_weight; | |
365 | dst->cpu_quota_per_sec_usec = src->cpu_quota_per_sec_usec; | |
366 | dst->cpu_quota_period_usec = src->cpu_quota_period_usec; | |
367 | ||
368 | dst->cpuset_cpus = src->cpuset_cpus; | |
369 | dst->startup_cpuset_cpus = src->startup_cpuset_cpus; | |
370 | dst->cpuset_mems = src->cpuset_mems; | |
371 | dst->startup_cpuset_mems = src->startup_cpuset_mems; | |
372 | ||
373 | dst->io_weight = src->io_weight; | |
374 | dst->startup_io_weight = src->startup_io_weight; | |
375 | ||
376 | LIST_FOREACH_BACKWARDS(device_weights, w, LIST_FIND_TAIL(device_weights, src->io_device_weights)) { | |
377 | r = cgroup_context_add_io_device_weight_dup(dst, w); | |
378 | if (r < 0) | |
379 | return r; | |
380 | } | |
381 | ||
382 | LIST_FOREACH_BACKWARDS(device_limits, l, LIST_FIND_TAIL(device_limits, src->io_device_limits)) { | |
383 | r = cgroup_context_add_io_device_limit_dup(dst, l); | |
384 | if (r < 0) | |
385 | return r; | |
386 | } | |
387 | ||
388 | LIST_FOREACH_BACKWARDS(device_latencies, l, LIST_FIND_TAIL(device_latencies, src->io_device_latencies)) { | |
389 | r = cgroup_context_add_io_device_latency_dup(dst, l); | |
390 | if (r < 0) | |
391 | return r; | |
392 | } | |
393 | ||
394 | dst->default_memory_min = src->default_memory_min; | |
395 | dst->default_memory_low = src->default_memory_low; | |
396 | dst->default_startup_memory_low = src->default_startup_memory_low; | |
397 | dst->memory_min = src->memory_min; | |
398 | dst->memory_low = src->memory_low; | |
399 | dst->startup_memory_low = src->startup_memory_low; | |
400 | dst->memory_high = src->memory_high; | |
401 | dst->startup_memory_high = src->startup_memory_high; | |
402 | dst->memory_max = src->memory_max; | |
403 | dst->startup_memory_max = src->startup_memory_max; | |
404 | dst->memory_swap_max = src->memory_swap_max; | |
405 | dst->startup_memory_swap_max = src->startup_memory_swap_max; | |
406 | dst->memory_zswap_max = src->memory_zswap_max; | |
407 | dst->startup_memory_zswap_max = src->startup_memory_zswap_max; | |
408 | ||
409 | dst->default_memory_min_set = src->default_memory_min_set; | |
410 | dst->default_memory_low_set = src->default_memory_low_set; | |
411 | dst->default_startup_memory_low_set = src->default_startup_memory_low_set; | |
412 | dst->memory_min_set = src->memory_min_set; | |
413 | dst->memory_low_set = src->memory_low_set; | |
414 | dst->startup_memory_low_set = src->startup_memory_low_set; | |
415 | dst->startup_memory_high_set = src->startup_memory_high_set; | |
416 | dst->startup_memory_max_set = src->startup_memory_max_set; | |
417 | dst->startup_memory_swap_max_set = src->startup_memory_swap_max_set; | |
418 | dst->startup_memory_zswap_max_set = src->startup_memory_zswap_max_set; | |
419 | ||
420 | SET_FOREACH(i, src->ip_address_allow) { | |
421 | r = in_addr_prefix_add(&dst->ip_address_allow, i); | |
422 | if (r < 0) | |
423 | return r; | |
424 | } | |
425 | ||
426 | SET_FOREACH(i, src->ip_address_deny) { | |
427 | r = in_addr_prefix_add(&dst->ip_address_deny, i); | |
428 | if (r < 0) | |
429 | return r; | |
430 | } | |
431 | ||
432 | dst->ip_address_allow_reduced = src->ip_address_allow_reduced; | |
433 | dst->ip_address_deny_reduced = src->ip_address_deny_reduced; | |
434 | ||
435 | if (!strv_isempty(src->ip_filters_ingress)) { | |
436 | dst->ip_filters_ingress = strv_copy(src->ip_filters_ingress); | |
437 | if (!dst->ip_filters_ingress) | |
438 | return -ENOMEM; | |
439 | } | |
440 | ||
441 | if (!strv_isempty(src->ip_filters_egress)) { | |
442 | dst->ip_filters_egress = strv_copy(src->ip_filters_egress); | |
443 | if (!dst->ip_filters_egress) | |
444 | return -ENOMEM; | |
445 | } | |
446 | ||
447 | LIST_FOREACH_BACKWARDS(programs, l, LIST_FIND_TAIL(programs, src->bpf_foreign_programs)) { | |
448 | r = cgroup_context_add_bpf_foreign_program_dup(dst, l); | |
449 | if (r < 0) | |
450 | return r; | |
451 | } | |
452 | ||
453 | SET_FOREACH(iface, src->restrict_network_interfaces) { | |
454 | r = set_put_strdup(&dst->restrict_network_interfaces, iface); | |
455 | if (r < 0) | |
456 | return r; | |
457 | } | |
458 | dst->restrict_network_interfaces_is_allow_list = src->restrict_network_interfaces_is_allow_list; | |
459 | ||
460 | dst->cpu_shares = src->cpu_shares; | |
461 | dst->startup_cpu_shares = src->startup_cpu_shares; | |
462 | ||
463 | dst->blockio_weight = src->blockio_weight; | |
464 | dst->startup_blockio_weight = src->startup_blockio_weight; | |
465 | ||
466 | LIST_FOREACH_BACKWARDS(device_weights, l, LIST_FIND_TAIL(device_weights, src->blockio_device_weights)) { | |
467 | r = cgroup_context_add_block_io_device_weight_dup(dst, l); | |
468 | if (r < 0) | |
469 | return r; | |
470 | } | |
471 | ||
472 | LIST_FOREACH_BACKWARDS(device_bandwidths, l, LIST_FIND_TAIL(device_bandwidths, src->blockio_device_bandwidths)) { | |
473 | r = cgroup_context_add_block_io_device_bandwidth_dup(dst, l); | |
474 | if (r < 0) | |
475 | return r; | |
476 | } | |
477 | ||
478 | dst->memory_limit = src->memory_limit; | |
479 | ||
480 | dst->device_policy = src->device_policy; | |
481 | LIST_FOREACH_BACKWARDS(device_allow, l, LIST_FIND_TAIL(device_allow, src->device_allow)) { | |
482 | r = cgroup_context_add_device_allow_dup(dst, l); | |
483 | if (r < 0) | |
484 | return r; | |
485 | } | |
486 | ||
487 | LIST_FOREACH_BACKWARDS(socket_bind_items, l, LIST_FIND_TAIL(socket_bind_items, src->socket_bind_allow)) { | |
488 | r = cgroup_context_add_socket_bind_item_allow_dup(dst, l); | |
489 | if (r < 0) | |
490 | return r; | |
491 | ||
492 | } | |
493 | ||
494 | LIST_FOREACH_BACKWARDS(socket_bind_items, l, LIST_FIND_TAIL(socket_bind_items, src->socket_bind_deny)) { | |
495 | r = cgroup_context_add_socket_bind_item_deny_dup(dst, l); | |
496 | if (r < 0) | |
497 | return r; | |
498 | } | |
499 | ||
500 | dst->tasks_max = src->tasks_max; | |
501 | ||
502 | return 0; | |
503 | } | |
504 | ||
4ad49000 LP |
505 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
506 | assert(c); | |
507 | assert(a); | |
508 | ||
71fda00f | 509 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
510 | free(a->path); |
511 | free(a); | |
512 | } | |
513 | ||
13c31542 TH |
514 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
515 | assert(c); | |
516 | assert(w); | |
517 | ||
518 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
519 | free(w->path); | |
520 | free(w); | |
521 | } | |
522 | ||
6ae4283c TH |
523 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
524 | assert(c); | |
525 | assert(l); | |
526 | ||
527 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
528 | free(l->path); | |
529 | free(l); | |
530 | } | |
531 | ||
13c31542 TH |
532 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
533 | assert(c); | |
534 | assert(l); | |
535 | ||
536 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
537 | free(l->path); | |
538 | free(l); | |
539 | } | |
540 | ||
4ad49000 LP |
541 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
542 | assert(c); | |
543 | assert(w); | |
544 | ||
71fda00f | 545 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
546 | free(w->path); |
547 | free(w); | |
548 | } | |
549 | ||
550 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
551 | assert(c); | |
8e274523 | 552 | assert(b); |
8e274523 | 553 | |
71fda00f | 554 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
555 | free(b->path); |
556 | free(b); | |
557 | } | |
558 | ||
b894ef1b JK |
559 | void cgroup_context_remove_bpf_foreign_program(CGroupContext *c, CGroupBPFForeignProgram *p) { |
560 | assert(c); | |
561 | assert(p); | |
562 | ||
563 | LIST_REMOVE(programs, c->bpf_foreign_programs, p); | |
564 | free(p->bpffs_path); | |
565 | free(p); | |
566 | } | |
567 | ||
b18e9fc1 | 568 | void cgroup_context_remove_socket_bind(CGroupSocketBindItem **head) { |
b18e9fc1 JK |
569 | assert(head); |
570 | ||
9aad490e | 571 | LIST_CLEAR(socket_bind_items, *head, free); |
b18e9fc1 JK |
572 | } |
573 | ||
4ad49000 LP |
574 | void cgroup_context_done(CGroupContext *c) { |
575 | assert(c); | |
576 | ||
13c31542 TH |
577 | while (c->io_device_weights) |
578 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
579 | ||
6ae4283c TH |
580 | while (c->io_device_latencies) |
581 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
582 | ||
13c31542 TH |
583 | while (c->io_device_limits) |
584 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
585 | ||
4ad49000 LP |
586 | while (c->blockio_device_weights) |
587 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
588 | ||
589 | while (c->blockio_device_bandwidths) | |
590 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
591 | ||
592 | while (c->device_allow) | |
593 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f | 594 | |
b18e9fc1 JK |
595 | cgroup_context_remove_socket_bind(&c->socket_bind_allow); |
596 | cgroup_context_remove_socket_bind(&c->socket_bind_deny); | |
597 | ||
84ebe6f0 YW |
598 | c->ip_address_allow = set_free(c->ip_address_allow); |
599 | c->ip_address_deny = set_free(c->ip_address_deny); | |
fab34748 KL |
600 | |
601 | c->ip_filters_ingress = strv_free(c->ip_filters_ingress); | |
602 | c->ip_filters_egress = strv_free(c->ip_filters_egress); | |
047f5d63 | 603 | |
b894ef1b JK |
604 | while (c->bpf_foreign_programs) |
605 | cgroup_context_remove_bpf_foreign_program(c, c->bpf_foreign_programs); | |
606 | ||
9b412709 | 607 | c->restrict_network_interfaces = set_free_free(c->restrict_network_interfaces); |
6f50d4f7 | 608 | |
047f5d63 | 609 | cpu_set_reset(&c->cpuset_cpus); |
31d3a520 | 610 | cpu_set_reset(&c->startup_cpuset_cpus); |
047f5d63 | 611 | cpu_set_reset(&c->cpuset_mems); |
31d3a520 | 612 | cpu_set_reset(&c->startup_cpuset_mems); |
a8b993dc LP |
613 | |
614 | c->delegate_subgroup = mfree(c->delegate_subgroup); | |
dc7d69b3 TM |
615 | |
616 | nft_set_context_clear(&c->nft_set_context); | |
4ad49000 LP |
617 | } |
618 | ||
74b5fb27 | 619 | static int unit_get_kernel_memory_limit(Unit *u, const char *file, uint64_t *ret) { |
74b5fb27 CD |
620 | assert(u); |
621 | ||
622 | if (!u->cgroup_realized) | |
623 | return -EOWNERDEAD; | |
624 | ||
613328c3 | 625 | return cg_get_attribute_as_uint64("memory", u->cgroup_path, file, ret); |
74b5fb27 CD |
626 | } |
627 | ||
628 | static int unit_compare_memory_limit(Unit *u, const char *property_name, uint64_t *ret_unit_value, uint64_t *ret_kernel_value) { | |
629 | CGroupContext *c; | |
630 | CGroupMask m; | |
631 | const char *file; | |
632 | uint64_t unit_value; | |
633 | int r; | |
634 | ||
635 | /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will | |
636 | * return -ENODATA) on cgroup v1. | |
637 | * | |
638 | * Returns: | |
639 | * | |
640 | * <0: On error. | |
641 | * 0: If the kernel memory setting doesn't match our configuration. | |
642 | * >0: If the kernel memory setting matches our configuration. | |
643 | * | |
644 | * The following values are only guaranteed to be populated on return >=0: | |
645 | * | |
646 | * - ret_unit_value will contain our internal expected value for the unit, page-aligned. | |
647 | * - ret_kernel_value will contain the actual value presented by the kernel. */ | |
648 | ||
649 | assert(u); | |
650 | ||
651 | r = cg_all_unified(); | |
652 | if (r < 0) | |
653 | return log_debug_errno(r, "Failed to determine cgroup hierarchy version: %m"); | |
654 | ||
655 | /* Unsupported on v1. | |
656 | * | |
657 | * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has | |
658 | * silently masked the controller. */ | |
659 | if (r == 0) | |
660 | return -ENODATA; | |
661 | ||
662 | /* The root slice doesn't have any controller files, so we can't compare anything. */ | |
663 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
664 | return -ENODATA; | |
665 | ||
666 | /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled, | |
667 | * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To | |
668 | * avoid specious errors in these scenarios, check that we even expect the memory controller to be | |
669 | * enabled at all. */ | |
670 | m = unit_get_target_mask(u); | |
671 | if (!FLAGS_SET(m, CGROUP_MASK_MEMORY)) | |
672 | return -ENODATA; | |
673 | ||
806a9362 | 674 | assert_se(c = unit_get_cgroup_context(u)); |
74b5fb27 | 675 | |
53fda560 LB |
676 | bool startup = u->manager && IN_SET(manager_state(u->manager), MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING); |
677 | ||
74b5fb27 CD |
678 | if (streq(property_name, "MemoryLow")) { |
679 | unit_value = unit_get_ancestor_memory_low(u); | |
680 | file = "memory.low"; | |
53fda560 LB |
681 | } else if (startup && streq(property_name, "StartupMemoryLow")) { |
682 | unit_value = unit_get_ancestor_startup_memory_low(u); | |
683 | file = "memory.low"; | |
74b5fb27 CD |
684 | } else if (streq(property_name, "MemoryMin")) { |
685 | unit_value = unit_get_ancestor_memory_min(u); | |
686 | file = "memory.min"; | |
687 | } else if (streq(property_name, "MemoryHigh")) { | |
688 | unit_value = c->memory_high; | |
689 | file = "memory.high"; | |
53fda560 LB |
690 | } else if (startup && streq(property_name, "StartupMemoryHigh")) { |
691 | unit_value = c->startup_memory_high; | |
692 | file = "memory.high"; | |
74b5fb27 CD |
693 | } else if (streq(property_name, "MemoryMax")) { |
694 | unit_value = c->memory_max; | |
695 | file = "memory.max"; | |
53fda560 LB |
696 | } else if (startup && streq(property_name, "StartupMemoryMax")) { |
697 | unit_value = c->startup_memory_max; | |
698 | file = "memory.max"; | |
74b5fb27 CD |
699 | } else if (streq(property_name, "MemorySwapMax")) { |
700 | unit_value = c->memory_swap_max; | |
701 | file = "memory.swap.max"; | |
53fda560 LB |
702 | } else if (startup && streq(property_name, "StartupMemorySwapMax")) { |
703 | unit_value = c->startup_memory_swap_max; | |
704 | file = "memory.swap.max"; | |
d7fe0a67 PV |
705 | } else if (streq(property_name, "MemoryZSwapMax")) { |
706 | unit_value = c->memory_zswap_max; | |
707 | file = "memory.zswap.max"; | |
53fda560 LB |
708 | } else if (startup && streq(property_name, "StartupMemoryZSwapMax")) { |
709 | unit_value = c->startup_memory_zswap_max; | |
710 | file = "memory.zswap.max"; | |
74b5fb27 CD |
711 | } else |
712 | return -EINVAL; | |
713 | ||
714 | r = unit_get_kernel_memory_limit(u, file, ret_kernel_value); | |
715 | if (r < 0) | |
716 | return log_unit_debug_errno(u, r, "Failed to parse %s: %m", file); | |
717 | ||
718 | /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page | |
719 | * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from | |
720 | * our internal page-counting. To support those future kernels, just check the value itself first | |
721 | * without any page-alignment. */ | |
722 | if (*ret_kernel_value == unit_value) { | |
723 | *ret_unit_value = unit_value; | |
724 | return 1; | |
725 | } | |
726 | ||
727 | /* The current kernel behaviour, by comparison, is that even if you write a particular number of | |
728 | * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel | |
729 | * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is | |
730 | * cricket. */ | |
731 | if (unit_value != CGROUP_LIMIT_MAX) | |
732 | unit_value = PAGE_ALIGN_DOWN(unit_value); | |
733 | ||
734 | *ret_unit_value = unit_value; | |
735 | ||
736 | return *ret_kernel_value == *ret_unit_value; | |
737 | } | |
738 | ||
bc0623df CD |
739 | #define FORMAT_CGROUP_DIFF_MAX 128 |
740 | ||
741 | static char *format_cgroup_memory_limit_comparison(char *buf, size_t l, Unit *u, const char *property_name) { | |
742 | uint64_t kval, sval; | |
743 | int r; | |
744 | ||
745 | assert(u); | |
746 | assert(buf); | |
747 | assert(l > 0); | |
748 | ||
749 | r = unit_compare_memory_limit(u, property_name, &sval, &kval); | |
750 | ||
751 | /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1). | |
752 | * In the absence of reliably being able to detect whether memcg swap support is available or not, | |
d7fe0a67 PV |
753 | * only complain if the error is not ENOENT. This is similarly the case for memory.zswap.max relying |
754 | * on CONFIG_ZSWAP. */ | |
bc0623df | 755 | if (r > 0 || IN_SET(r, -ENODATA, -EOWNERDEAD) || |
53fda560 LB |
756 | (r == -ENOENT && STR_IN_SET(property_name, |
757 | "MemorySwapMax", | |
758 | "StartupMemorySwapMax", | |
759 | "MemoryZSwapMax", | |
760 | "StartupMemoryZSwapMax"))) | |
bc0623df | 761 | buf[0] = 0; |
38553034 ZJS |
762 | else if (r < 0) { |
763 | errno = -r; | |
764 | (void) snprintf(buf, l, " (error getting kernel value: %m)"); | |
765 | } else | |
766 | (void) snprintf(buf, l, " (different value in kernel: %" PRIu64 ")", kval); | |
bc0623df CD |
767 | |
768 | return buf; | |
769 | } | |
770 | ||
a1044811 LP |
771 | const char *cgroup_device_permissions_to_string(CGroupDevicePermissions p) { |
772 | static const char *table[_CGROUP_DEVICE_PERMISSIONS_MAX] = { | |
773 | /* Lets simply define a table with every possible combination. As long as those are just 8 we | |
774 | * can get away with it. If this ever grows to more we need to revisit this logic though. */ | |
775 | [0] = "", | |
776 | [CGROUP_DEVICE_READ] = "r", | |
777 | [CGROUP_DEVICE_WRITE] = "w", | |
778 | [CGROUP_DEVICE_MKNOD] = "m", | |
779 | [CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE] = "rw", | |
780 | [CGROUP_DEVICE_READ|CGROUP_DEVICE_MKNOD] = "rm", | |
781 | [CGROUP_DEVICE_WRITE|CGROUP_DEVICE_MKNOD] = "wm", | |
782 | [CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE|CGROUP_DEVICE_MKNOD] = "rwm", | |
783 | }; | |
784 | ||
785 | if (p < 0 || p >= _CGROUP_DEVICE_PERMISSIONS_MAX) | |
786 | return NULL; | |
787 | ||
788 | return table[p]; | |
789 | } | |
790 | ||
791 | CGroupDevicePermissions cgroup_device_permissions_from_string(const char *s) { | |
792 | CGroupDevicePermissions p = 0; | |
793 | ||
794 | if (!s) | |
795 | return _CGROUP_DEVICE_PERMISSIONS_INVALID; | |
796 | ||
797 | for (const char *c = s; *c; c++) { | |
798 | if (*c == 'r') | |
799 | p |= CGROUP_DEVICE_READ; | |
800 | else if (*c == 'w') | |
801 | p |= CGROUP_DEVICE_WRITE; | |
802 | else if (*c == 'm') | |
803 | p |= CGROUP_DEVICE_MKNOD; | |
804 | else | |
805 | return _CGROUP_DEVICE_PERMISSIONS_INVALID; | |
806 | } | |
807 | ||
808 | return p; | |
809 | } | |
810 | ||
bc0623df | 811 | void cgroup_context_dump(Unit *u, FILE* f, const char *prefix) { |
7b3693e4 | 812 | _cleanup_free_ char *disable_controllers_str = NULL, *delegate_controllers_str = NULL, *cpuset_cpus = NULL, *cpuset_mems = NULL, *startup_cpuset_cpus = NULL, *startup_cpuset_mems = NULL; |
bc0623df | 813 | CGroupContext *c; |
84ebe6f0 | 814 | struct in_addr_prefix *iaai; |
4ad49000 | 815 | |
bc0623df CD |
816 | char cda[FORMAT_CGROUP_DIFF_MAX]; |
817 | char cdb[FORMAT_CGROUP_DIFF_MAX]; | |
818 | char cdc[FORMAT_CGROUP_DIFF_MAX]; | |
819 | char cdd[FORMAT_CGROUP_DIFF_MAX]; | |
820 | char cde[FORMAT_CGROUP_DIFF_MAX]; | |
53fda560 LB |
821 | char cdf[FORMAT_CGROUP_DIFF_MAX]; |
822 | char cdg[FORMAT_CGROUP_DIFF_MAX]; | |
823 | char cdh[FORMAT_CGROUP_DIFF_MAX]; | |
824 | char cdi[FORMAT_CGROUP_DIFF_MAX]; | |
825 | char cdj[FORMAT_CGROUP_DIFF_MAX]; | |
826 | char cdk[FORMAT_CGROUP_DIFF_MAX]; | |
bc0623df CD |
827 | |
828 | assert(u); | |
4ad49000 LP |
829 | assert(f); |
830 | ||
806a9362 | 831 | assert_se(c = unit_get_cgroup_context(u)); |
bc0623df | 832 | |
4ad49000 LP |
833 | prefix = strempty(prefix); |
834 | ||
25cc30c4 | 835 | (void) cg_mask_to_string(c->disable_controllers, &disable_controllers_str); |
7b3693e4 | 836 | (void) cg_mask_to_string(c->delegate_controllers, &delegate_controllers_str); |
25cc30c4 | 837 | |
af2b151b ZJS |
838 | /* "Delegate=" means "yes, but no controllers". Show this as "(none)". */ |
839 | const char *delegate_str = delegate_controllers_str ?: c->delegate ? "(none)" : "no"; | |
840 | ||
047f5d63 | 841 | cpuset_cpus = cpu_set_to_range_string(&c->cpuset_cpus); |
31d3a520 | 842 | startup_cpuset_cpus = cpu_set_to_range_string(&c->startup_cpuset_cpus); |
047f5d63 | 843 | cpuset_mems = cpu_set_to_range_string(&c->cpuset_mems); |
31d3a520 | 844 | startup_cpuset_mems = cpu_set_to_range_string(&c->startup_cpuset_mems); |
047f5d63 | 845 | |
4ad49000 | 846 | fprintf(f, |
6dfb9282 CD |
847 | "%sCPUAccounting: %s\n" |
848 | "%sIOAccounting: %s\n" | |
849 | "%sBlockIOAccounting: %s\n" | |
850 | "%sMemoryAccounting: %s\n" | |
851 | "%sTasksAccounting: %s\n" | |
852 | "%sIPAccounting: %s\n" | |
853 | "%sCPUWeight: %" PRIu64 "\n" | |
854 | "%sStartupCPUWeight: %" PRIu64 "\n" | |
855 | "%sCPUShares: %" PRIu64 "\n" | |
856 | "%sStartupCPUShares: %" PRIu64 "\n" | |
857 | "%sCPUQuotaPerSecSec: %s\n" | |
858 | "%sCPUQuotaPeriodSec: %s\n" | |
859 | "%sAllowedCPUs: %s\n" | |
31d3a520 | 860 | "%sStartupAllowedCPUs: %s\n" |
6dfb9282 | 861 | "%sAllowedMemoryNodes: %s\n" |
31d3a520 | 862 | "%sStartupAllowedMemoryNodes: %s\n" |
6dfb9282 CD |
863 | "%sIOWeight: %" PRIu64 "\n" |
864 | "%sStartupIOWeight: %" PRIu64 "\n" | |
865 | "%sBlockIOWeight: %" PRIu64 "\n" | |
866 | "%sStartupBlockIOWeight: %" PRIu64 "\n" | |
867 | "%sDefaultMemoryMin: %" PRIu64 "\n" | |
868 | "%sDefaultMemoryLow: %" PRIu64 "\n" | |
bc0623df CD |
869 | "%sMemoryMin: %" PRIu64 "%s\n" |
870 | "%sMemoryLow: %" PRIu64 "%s\n" | |
53fda560 | 871 | "%sStartupMemoryLow: %" PRIu64 "%s\n" |
bc0623df | 872 | "%sMemoryHigh: %" PRIu64 "%s\n" |
53fda560 | 873 | "%sStartupMemoryHigh: %" PRIu64 "%s\n" |
bc0623df | 874 | "%sMemoryMax: %" PRIu64 "%s\n" |
53fda560 | 875 | "%sStartupMemoryMax: %" PRIu64 "%s\n" |
bc0623df | 876 | "%sMemorySwapMax: %" PRIu64 "%s\n" |
53fda560 | 877 | "%sStartupMemorySwapMax: %" PRIu64 "%s\n" |
d7fe0a67 | 878 | "%sMemoryZSwapMax: %" PRIu64 "%s\n" |
53fda560 | 879 | "%sStartupMemoryZSwapMax: %" PRIu64 "%s\n" |
6dfb9282 CD |
880 | "%sMemoryLimit: %" PRIu64 "\n" |
881 | "%sTasksMax: %" PRIu64 "\n" | |
882 | "%sDevicePolicy: %s\n" | |
883 | "%sDisableControllers: %s\n" | |
4d824a4e AZ |
884 | "%sDelegate: %s\n" |
885 | "%sManagedOOMSwap: %s\n" | |
886 | "%sManagedOOMMemoryPressure: %s\n" | |
d9d3f05d | 887 | "%sManagedOOMMemoryPressureLimit: " PERMYRIAD_AS_PERCENT_FORMAT_STR "\n" |
6bb00842 | 888 | "%sManagedOOMPreference: %s\n" |
6cf96ab4 NR |
889 | "%sMemoryPressureWatch: %s\n" |
890 | "%sCoredumpReceive: %s\n", | |
4ad49000 | 891 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 892 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
893 | prefix, yes_no(c->blockio_accounting), |
894 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 895 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 896 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
897 | prefix, c->cpu_weight, |
898 | prefix, c->startup_cpu_weight, | |
4ad49000 | 899 | prefix, c->cpu_shares, |
95ae05c0 | 900 | prefix, c->startup_cpu_shares, |
5291f26d ZJS |
901 | prefix, FORMAT_TIMESPAN(c->cpu_quota_per_sec_usec, 1), |
902 | prefix, FORMAT_TIMESPAN(c->cpu_quota_period_usec, 1), | |
85c3b278 | 903 | prefix, strempty(cpuset_cpus), |
31d3a520 | 904 | prefix, strempty(startup_cpuset_cpus), |
85c3b278 | 905 | prefix, strempty(cpuset_mems), |
31d3a520 | 906 | prefix, strempty(startup_cpuset_mems), |
13c31542 TH |
907 | prefix, c->io_weight, |
908 | prefix, c->startup_io_weight, | |
4ad49000 | 909 | prefix, c->blockio_weight, |
95ae05c0 | 910 | prefix, c->startup_blockio_weight, |
7ad5439e | 911 | prefix, c->default_memory_min, |
c52db42b | 912 | prefix, c->default_memory_low, |
bc0623df CD |
913 | prefix, c->memory_min, format_cgroup_memory_limit_comparison(cda, sizeof(cda), u, "MemoryMin"), |
914 | prefix, c->memory_low, format_cgroup_memory_limit_comparison(cdb, sizeof(cdb), u, "MemoryLow"), | |
53fda560 LB |
915 | prefix, c->startup_memory_low, format_cgroup_memory_limit_comparison(cdc, sizeof(cdc), u, "StartupMemoryLow"), |
916 | prefix, c->memory_high, format_cgroup_memory_limit_comparison(cdd, sizeof(cdd), u, "MemoryHigh"), | |
917 | prefix, c->startup_memory_high, format_cgroup_memory_limit_comparison(cde, sizeof(cde), u, "StartupMemoryHigh"), | |
918 | prefix, c->memory_max, format_cgroup_memory_limit_comparison(cdf, sizeof(cdf), u, "MemoryMax"), | |
919 | prefix, c->startup_memory_max, format_cgroup_memory_limit_comparison(cdg, sizeof(cdg), u, "StartupMemoryMax"), | |
920 | prefix, c->memory_swap_max, format_cgroup_memory_limit_comparison(cdh, sizeof(cdh), u, "MemorySwapMax"), | |
921 | prefix, c->startup_memory_swap_max, format_cgroup_memory_limit_comparison(cdi, sizeof(cdi), u, "StartupMemorySwapMax"), | |
922 | prefix, c->memory_zswap_max, format_cgroup_memory_limit_comparison(cdj, sizeof(cdj), u, "MemoryZSwapMax"), | |
923 | prefix, c->startup_memory_zswap_max, format_cgroup_memory_limit_comparison(cdk, sizeof(cdk), u, "StartupMemoryZSwapMax"), | |
4ad49000 | 924 | prefix, c->memory_limit, |
94f0b13b | 925 | prefix, cgroup_tasks_max_resolve(&c->tasks_max), |
a931ad47 | 926 | prefix, cgroup_device_policy_to_string(c->device_policy), |
f4c43a81 | 927 | prefix, strempty(disable_controllers_str), |
af2b151b | 928 | prefix, delegate_str, |
4d824a4e AZ |
929 | prefix, managed_oom_mode_to_string(c->moom_swap), |
930 | prefix, managed_oom_mode_to_string(c->moom_mem_pressure), | |
d9d3f05d | 931 | prefix, PERMYRIAD_AS_PERCENT_FORMAT_VAL(UINT32_SCALE_TO_PERMYRIAD(c->moom_mem_pressure_limit)), |
6bb00842 | 932 | prefix, managed_oom_preference_to_string(c->moom_preference), |
6cf96ab4 NR |
933 | prefix, cgroup_pressure_watch_to_string(c->memory_pressure_watch), |
934 | prefix, yes_no(c->coredump_receive)); | |
6bb00842 | 935 | |
a8b993dc LP |
936 | if (c->delegate_subgroup) |
937 | fprintf(f, "%sDelegateSubgroup: %s\n", | |
938 | prefix, c->delegate_subgroup); | |
939 | ||
6bb00842 LP |
940 | if (c->memory_pressure_threshold_usec != USEC_INFINITY) |
941 | fprintf(f, "%sMemoryPressureThresholdSec: %s\n", | |
942 | prefix, FORMAT_TIMESPAN(c->memory_pressure_threshold_usec, 1)); | |
4ad49000 LP |
943 | |
944 | LIST_FOREACH(device_allow, a, c->device_allow) | |
14338cca | 945 | /* strna() below should be redundant, for avoiding -Werror=format-overflow= error. See #30223. */ |
4ad49000 | 946 | fprintf(f, |
a1044811 | 947 | "%sDeviceAllow: %s %s\n", |
4ad49000 LP |
948 | prefix, |
949 | a->path, | |
14338cca | 950 | strna(cgroup_device_permissions_to_string(a->permissions))); |
4ad49000 | 951 | |
13c31542 TH |
952 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
953 | fprintf(f, | |
6dfb9282 | 954 | "%sIODeviceWeight: %s %" PRIu64 "\n", |
13c31542 TH |
955 | prefix, |
956 | iw->path, | |
957 | iw->weight); | |
958 | ||
6ae4283c TH |
959 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
960 | fprintf(f, | |
6dfb9282 | 961 | "%sIODeviceLatencyTargetSec: %s %s\n", |
6ae4283c TH |
962 | prefix, |
963 | l->path, | |
5291f26d | 964 | FORMAT_TIMESPAN(l->target_usec, 1)); |
6ae4283c | 965 | |
2b59bf51 | 966 | LIST_FOREACH(device_limits, il, c->io_device_limits) |
e8616626 | 967 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
9be57249 TH |
968 | if (il->limits[type] != cgroup_io_limit_defaults[type]) |
969 | fprintf(f, | |
6dfb9282 | 970 | "%s%s: %s %s\n", |
9be57249 TH |
971 | prefix, |
972 | cgroup_io_limit_type_to_string(type), | |
973 | il->path, | |
2b59bf51 | 974 | FORMAT_BYTES(il->limits[type])); |
13c31542 | 975 | |
4ad49000 LP |
976 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
977 | fprintf(f, | |
6dfb9282 | 978 | "%sBlockIODeviceWeight: %s %" PRIu64, |
4ad49000 LP |
979 | prefix, |
980 | w->path, | |
981 | w->weight); | |
982 | ||
983 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
979d0311 TH |
984 | if (b->rbps != CGROUP_LIMIT_MAX) |
985 | fprintf(f, | |
6dfb9282 | 986 | "%sBlockIOReadBandwidth: %s %s\n", |
979d0311 TH |
987 | prefix, |
988 | b->path, | |
2b59bf51 | 989 | FORMAT_BYTES(b->rbps)); |
979d0311 TH |
990 | if (b->wbps != CGROUP_LIMIT_MAX) |
991 | fprintf(f, | |
6dfb9282 | 992 | "%sBlockIOWriteBandwidth: %s %s\n", |
979d0311 TH |
993 | prefix, |
994 | b->path, | |
2b59bf51 | 995 | FORMAT_BYTES(b->wbps)); |
4ad49000 | 996 | } |
c21c9906 | 997 | |
c71384a9 ZJS |
998 | SET_FOREACH(iaai, c->ip_address_allow) |
999 | fprintf(f, "%sIPAddressAllow: %s\n", prefix, | |
1000 | IN_ADDR_PREFIX_TO_STRING(iaai->family, &iaai->address, iaai->prefixlen)); | |
1001 | SET_FOREACH(iaai, c->ip_address_deny) | |
1002 | fprintf(f, "%sIPAddressDeny: %s\n", prefix, | |
1003 | IN_ADDR_PREFIX_TO_STRING(iaai->family, &iaai->address, iaai->prefixlen)); | |
fab34748 KL |
1004 | |
1005 | STRV_FOREACH(path, c->ip_filters_ingress) | |
6dfb9282 | 1006 | fprintf(f, "%sIPIngressFilterPath: %s\n", prefix, *path); |
fab34748 | 1007 | STRV_FOREACH(path, c->ip_filters_egress) |
6dfb9282 | 1008 | fprintf(f, "%sIPEgressFilterPath: %s\n", prefix, *path); |
b894ef1b JK |
1009 | |
1010 | LIST_FOREACH(programs, p, c->bpf_foreign_programs) | |
1011 | fprintf(f, "%sBPFProgram: %s:%s", | |
1012 | prefix, bpf_cgroup_attach_type_to_string(p->attach_type), p->bpffs_path); | |
b18e9fc1 JK |
1013 | |
1014 | if (c->socket_bind_allow) { | |
b0bb3be1 FS |
1015 | fprintf(f, "%sSocketBindAllow: ", prefix); |
1016 | cgroup_context_dump_socket_bind_items(c->socket_bind_allow, f); | |
b18e9fc1 JK |
1017 | fputc('\n', f); |
1018 | } | |
1019 | ||
1020 | if (c->socket_bind_deny) { | |
b0bb3be1 FS |
1021 | fprintf(f, "%sSocketBindDeny: ", prefix); |
1022 | cgroup_context_dump_socket_bind_items(c->socket_bind_deny, f); | |
b18e9fc1 JK |
1023 | fputc('\n', f); |
1024 | } | |
6f50d4f7 MV |
1025 | |
1026 | if (c->restrict_network_interfaces) { | |
1027 | char *iface; | |
1028 | SET_FOREACH(iface, c->restrict_network_interfaces) | |
1029 | fprintf(f, "%sRestrictNetworkInterfaces: %s\n", prefix, iface); | |
1030 | } | |
dc7d69b3 TM |
1031 | |
1032 | FOREACH_ARRAY(nft_set, c->nft_set_context.sets, c->nft_set_context.n_sets) | |
1033 | fprintf(f, "%sNFTSet: %s:%s:%s:%s\n", prefix, nft_set_source_to_string(nft_set->source), | |
1034 | nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set); | |
b18e9fc1 JK |
1035 | } |
1036 | ||
1037 | void cgroup_context_dump_socket_bind_item(const CGroupSocketBindItem *item, FILE *f) { | |
5587ce7f | 1038 | const char *family, *colon1, *protocol = "", *colon2 = ""; |
a4817536 LP |
1039 | |
1040 | family = strempty(af_to_ipv4_ipv6(item->address_family)); | |
5587ce7f JK |
1041 | colon1 = isempty(family) ? "" : ":"; |
1042 | ||
1043 | if (item->ip_protocol != 0) { | |
1044 | protocol = ip_protocol_to_tcp_udp(item->ip_protocol); | |
1045 | colon2 = ":"; | |
1046 | } | |
b18e9fc1 JK |
1047 | |
1048 | if (item->nr_ports == 0) | |
b0bb3be1 | 1049 | fprintf(f, "%s%s%s%sany", family, colon1, protocol, colon2); |
b18e9fc1 | 1050 | else if (item->nr_ports == 1) |
b0bb3be1 | 1051 | fprintf(f, "%s%s%s%s%" PRIu16, family, colon1, protocol, colon2, item->port_min); |
b18e9fc1 JK |
1052 | else { |
1053 | uint16_t port_max = item->port_min + item->nr_ports - 1; | |
b0bb3be1 | 1054 | fprintf(f, "%s%s%s%s%" PRIu16 "-%" PRIu16, family, colon1, protocol, colon2, |
5587ce7f | 1055 | item->port_min, port_max); |
b18e9fc1 | 1056 | } |
4ad49000 LP |
1057 | } |
1058 | ||
b0bb3be1 FS |
1059 | void cgroup_context_dump_socket_bind_items(const CGroupSocketBindItem *items, FILE *f) { |
1060 | bool first = true; | |
1061 | ||
1062 | LIST_FOREACH(socket_bind_items, bi, items) { | |
1063 | if (first) | |
1064 | first = false; | |
1065 | else | |
1066 | fputc(' ', f); | |
1067 | ||
1068 | cgroup_context_dump_socket_bind_item(bi, f); | |
1069 | } | |
1070 | } | |
1071 | ||
a1044811 | 1072 | int cgroup_context_add_device_allow(CGroupContext *c, const char *dev, CGroupDevicePermissions p) { |
fd870bac YW |
1073 | _cleanup_free_ CGroupDeviceAllow *a = NULL; |
1074 | _cleanup_free_ char *d = NULL; | |
1075 | ||
1076 | assert(c); | |
1077 | assert(dev); | |
a1044811 LP |
1078 | assert(p >= 0 && p < _CGROUP_DEVICE_PERMISSIONS_MAX); |
1079 | ||
1080 | if (p == 0) | |
1081 | p = _CGROUP_DEVICE_PERMISSIONS_ALL; | |
fd870bac YW |
1082 | |
1083 | a = new(CGroupDeviceAllow, 1); | |
1084 | if (!a) | |
1085 | return -ENOMEM; | |
1086 | ||
1087 | d = strdup(dev); | |
1088 | if (!d) | |
1089 | return -ENOMEM; | |
1090 | ||
1091 | *a = (CGroupDeviceAllow) { | |
1092 | .path = TAKE_PTR(d), | |
a1044811 | 1093 | .permissions = p, |
fd870bac YW |
1094 | }; |
1095 | ||
1096 | LIST_PREPEND(device_allow, c->device_allow, a); | |
1097 | TAKE_PTR(a); | |
1098 | ||
1099 | return 0; | |
1100 | } | |
1101 | ||
a1044811 | 1102 | int cgroup_context_add_or_update_device_allow(CGroupContext *c, const char *dev, CGroupDevicePermissions p) { |
c3166b25 LB |
1103 | assert(c); |
1104 | assert(dev); | |
a1044811 LP |
1105 | assert(p >= 0 && p < _CGROUP_DEVICE_PERMISSIONS_MAX); |
1106 | ||
1107 | if (p == 0) | |
1108 | p = _CGROUP_DEVICE_PERMISSIONS_ALL; | |
c3166b25 LB |
1109 | |
1110 | LIST_FOREACH(device_allow, b, c->device_allow) | |
1111 | if (path_equal(b->path, dev)) { | |
a1044811 | 1112 | b->permissions = p; |
c3166b25 LB |
1113 | return 0; |
1114 | } | |
1115 | ||
a1044811 | 1116 | return cgroup_context_add_device_allow(c, dev, p); |
c3166b25 LB |
1117 | } |
1118 | ||
c6f2dca6 | 1119 | int cgroup_context_add_bpf_foreign_program(CGroupContext *c, uint32_t attach_type, const char *bpffs_path) { |
b894ef1b JK |
1120 | CGroupBPFForeignProgram *p; |
1121 | _cleanup_free_ char *d = NULL; | |
1122 | ||
1123 | assert(c); | |
1124 | assert(bpffs_path); | |
1125 | ||
1126 | if (!path_is_normalized(bpffs_path) || !path_is_absolute(bpffs_path)) | |
1127 | return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Path is not normalized: %m"); | |
1128 | ||
1129 | d = strdup(bpffs_path); | |
1130 | if (!d) | |
1131 | return log_oom(); | |
1132 | ||
1133 | p = new(CGroupBPFForeignProgram, 1); | |
1134 | if (!p) | |
1135 | return log_oom(); | |
1136 | ||
1137 | *p = (CGroupBPFForeignProgram) { | |
1138 | .attach_type = attach_type, | |
1139 | .bpffs_path = TAKE_PTR(d), | |
1140 | }; | |
1141 | ||
1142 | LIST_PREPEND(programs, c->bpf_foreign_programs, TAKE_PTR(p)); | |
1143 | ||
1144 | return 0; | |
1145 | } | |
1146 | ||
6264b85e CD |
1147 | #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \ |
1148 | uint64_t unit_get_ancestor_##entry(Unit *u) { \ | |
1149 | CGroupContext *c; \ | |
1150 | \ | |
1151 | /* 1. Is entry set in this unit? If so, use that. \ | |
1152 | * 2. Is the default for this entry set in any \ | |
1153 | * ancestor? If so, use that. \ | |
1154 | * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \ | |
1155 | \ | |
1156 | assert(u); \ | |
1157 | \ | |
1158 | c = unit_get_cgroup_context(u); \ | |
c5322608 | 1159 | if (c && c->entry##_set) \ |
6264b85e CD |
1160 | return c->entry; \ |
1161 | \ | |
12f64221 | 1162 | while ((u = UNIT_GET_SLICE(u))) { \ |
6264b85e | 1163 | c = unit_get_cgroup_context(u); \ |
c5322608 | 1164 | if (c && c->default_##entry##_set) \ |
6264b85e CD |
1165 | return c->default_##entry; \ |
1166 | } \ | |
1167 | \ | |
1168 | /* We've reached the root, but nobody had default for \ | |
1169 | * this entry set, so set it to the kernel default. */ \ | |
1170 | return CGROUP_LIMIT_MIN; \ | |
c52db42b CD |
1171 | } |
1172 | ||
6264b85e | 1173 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low); |
53fda560 | 1174 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(startup_memory_low); |
7ad5439e | 1175 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min); |
6264b85e | 1176 | |
17d047f5 | 1177 | static void unit_set_xattr_graceful(Unit *u, const char *name, const void *data, size_t size) { |
1fa3b6c2 LP |
1178 | int r; |
1179 | ||
1180 | assert(u); | |
1181 | assert(name); | |
1182 | ||
17d047f5 LP |
1183 | if (!u->cgroup_path) |
1184 | return; | |
1fa3b6c2 | 1185 | |
17d047f5 | 1186 | r = cg_set_xattr(u->cgroup_path, name, data, size, 0); |
1fa3b6c2 | 1187 | if (r < 0) |
17d047f5 | 1188 | log_unit_debug_errno(u, r, "Failed to set '%s' xattr on control group %s, ignoring: %m", name, empty_to_root(u->cgroup_path)); |
1fa3b6c2 LP |
1189 | } |
1190 | ||
17d047f5 | 1191 | static void unit_remove_xattr_graceful(Unit *u, const char *name) { |
1fa3b6c2 LP |
1192 | int r; |
1193 | ||
1194 | assert(u); | |
1195 | assert(name); | |
1196 | ||
17d047f5 LP |
1197 | if (!u->cgroup_path) |
1198 | return; | |
1fa3b6c2 | 1199 | |
17d047f5 | 1200 | r = cg_remove_xattr(u->cgroup_path, name); |
00675c36 | 1201 | if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r)) |
17d047f5 | 1202 | log_unit_debug_errno(u, r, "Failed to remove '%s' xattr flag on control group %s, ignoring: %m", name, empty_to_root(u->cgroup_path)); |
1fa3b6c2 LP |
1203 | } |
1204 | ||
64c71f4f | 1205 | static void cgroup_oomd_xattr_apply(Unit *u) { |
4e806bfa | 1206 | CGroupContext *c; |
4e806bfa AZ |
1207 | |
1208 | assert(u); | |
1209 | ||
1210 | c = unit_get_cgroup_context(u); | |
1211 | if (!c) | |
1212 | return; | |
1213 | ||
1fa3b6c2 | 1214 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_OMIT) |
17d047f5 | 1215 | unit_set_xattr_graceful(u, "user.oomd_omit", "1", 1); |
4e806bfa | 1216 | |
1fa3b6c2 | 1217 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_AVOID) |
17d047f5 | 1218 | unit_set_xattr_graceful(u, "user.oomd_avoid", "1", 1); |
4e806bfa | 1219 | |
1fa3b6c2 | 1220 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_AVOID) |
17d047f5 | 1221 | unit_remove_xattr_graceful(u, "user.oomd_avoid"); |
4e806bfa | 1222 | |
1fa3b6c2 | 1223 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_OMIT) |
17d047f5 | 1224 | unit_remove_xattr_graceful(u, "user.oomd_omit"); |
4e806bfa AZ |
1225 | } |
1226 | ||
64c71f4f | 1227 | static int cgroup_log_xattr_apply(Unit *u) { |
523ea123 QD |
1228 | ExecContext *c; |
1229 | size_t len, allowed_patterns_len, denied_patterns_len; | |
1230 | _cleanup_free_ char *patterns = NULL, *allowed_patterns = NULL, *denied_patterns = NULL; | |
48d85160 | 1231 | char *last; |
523ea123 QD |
1232 | int r; |
1233 | ||
1234 | assert(u); | |
1235 | ||
1236 | c = unit_get_exec_context(u); | |
1237 | if (!c) | |
1238 | /* Some unit types have a cgroup context but no exec context, so we do not log | |
1239 | * any error here to avoid confusion. */ | |
1240 | return 0; | |
1241 | ||
1242 | if (set_isempty(c->log_filter_allowed_patterns) && set_isempty(c->log_filter_denied_patterns)) { | |
17d047f5 | 1243 | unit_remove_xattr_graceful(u, "user.journald_log_filter_patterns"); |
523ea123 QD |
1244 | return 0; |
1245 | } | |
1246 | ||
1247 | r = set_make_nulstr(c->log_filter_allowed_patterns, &allowed_patterns, &allowed_patterns_len); | |
1248 | if (r < 0) | |
1249 | return log_debug_errno(r, "Failed to make nulstr from set: %m"); | |
1250 | ||
1251 | r = set_make_nulstr(c->log_filter_denied_patterns, &denied_patterns, &denied_patterns_len); | |
1252 | if (r < 0) | |
1253 | return log_debug_errno(r, "Failed to make nulstr from set: %m"); | |
1254 | ||
1255 | /* Use nul character separated strings without trailing nul */ | |
1256 | allowed_patterns_len = LESS_BY(allowed_patterns_len, 1u); | |
1257 | denied_patterns_len = LESS_BY(denied_patterns_len, 1u); | |
1258 | ||
1259 | len = allowed_patterns_len + 1 + denied_patterns_len; | |
1260 | patterns = new(char, len); | |
1261 | if (!patterns) | |
1262 | return log_oom_debug(); | |
1263 | ||
48d85160 QD |
1264 | last = mempcpy_safe(patterns, allowed_patterns, allowed_patterns_len); |
1265 | *(last++) = '\xff'; | |
1266 | memcpy_safe(last, denied_patterns, denied_patterns_len); | |
523ea123 | 1267 | |
17d047f5 | 1268 | unit_set_xattr_graceful(u, "user.journald_log_filter_patterns", patterns, len); |
523ea123 QD |
1269 | |
1270 | return 0; | |
1271 | } | |
1272 | ||
d46510de | 1273 | static void cgroup_invocation_id_xattr_apply(Unit *u) { |
d9bc1c36 | 1274 | bool b; |
0d2d6fbf CD |
1275 | |
1276 | assert(u); | |
1277 | ||
1fa3b6c2 LP |
1278 | b = !sd_id128_is_null(u->invocation_id); |
1279 | FOREACH_STRING(xn, "trusted.invocation_id", "user.invocation_id") { | |
1280 | if (b) | |
17d047f5 | 1281 | unit_set_xattr_graceful(u, xn, SD_ID128_TO_STRING(u->invocation_id), 32); |
1fa3b6c2 | 1282 | else |
17d047f5 | 1283 | unit_remove_xattr_graceful(u, xn); |
3288ea8f | 1284 | } |
d46510de LP |
1285 | } |
1286 | ||
6cf96ab4 NR |
1287 | static void cgroup_coredump_xattr_apply(Unit *u) { |
1288 | CGroupContext *c; | |
1289 | ||
1290 | assert(u); | |
1291 | ||
1292 | c = unit_get_cgroup_context(u); | |
1293 | if (!c) | |
1294 | return; | |
1295 | ||
1296 | if (unit_cgroup_delegate(u) && c->coredump_receive) | |
1297 | unit_set_xattr_graceful(u, "user.coredump_receive", "1", 1); | |
1298 | else | |
1299 | unit_remove_xattr_graceful(u, "user.coredump_receive"); | |
1300 | } | |
1301 | ||
d46510de LP |
1302 | static void cgroup_delegate_xattr_apply(Unit *u) { |
1303 | bool b; | |
1304 | ||
1305 | assert(u); | |
0d2d6fbf | 1306 | |
d9bc1c36 LP |
1307 | /* Indicate on the cgroup whether delegation is on, via an xattr. This is best-effort, as old kernels |
1308 | * didn't support xattrs on cgroups at all. Later they got support for setting 'trusted.*' xattrs, | |
1309 | * and even later 'user.*' xattrs. We started setting this field when 'trusted.*' was added, and | |
1310 | * given this is now pretty much API, let's continue to support that. But also set 'user.*' as well, | |
1311 | * since it is readable by any user, not just CAP_SYS_ADMIN. This hence comes with slightly weaker | |
1312 | * security (as users who got delegated cgroups could turn it off if they like), but this shouldn't | |
1313 | * be a big problem given this communicates delegation state to clients, but the manager never reads | |
1314 | * it. */ | |
1315 | b = unit_cgroup_delegate(u); | |
1316 | FOREACH_STRING(xn, "trusted.delegate", "user.delegate") { | |
1fa3b6c2 | 1317 | if (b) |
17d047f5 | 1318 | unit_set_xattr_graceful(u, xn, "1", 1); |
1fa3b6c2 | 1319 | else |
17d047f5 | 1320 | unit_remove_xattr_graceful(u, xn); |
3288ea8f | 1321 | } |
d46510de LP |
1322 | } |
1323 | ||
1324 | static void cgroup_survive_xattr_apply(Unit *u) { | |
1325 | int r; | |
1326 | ||
1327 | assert(u); | |
559214cb LB |
1328 | |
1329 | if (u->survive_final_kill_signal) { | |
bd1791b5 LP |
1330 | r = cg_set_xattr( |
1331 | u->cgroup_path, | |
1332 | "user.survive_final_kill_signal", | |
1333 | "1", | |
1334 | 1, | |
1335 | /* flags= */ 0); | |
559214cb LB |
1336 | /* user xattr support was added in kernel v5.7 */ |
1337 | if (ERRNO_IS_NEG_NOT_SUPPORTED(r)) | |
bd1791b5 | 1338 | r = cg_set_xattr( |
559214cb LB |
1339 | u->cgroup_path, |
1340 | "trusted.survive_final_kill_signal", | |
1341 | "1", | |
1342 | 1, | |
1343 | /* flags= */ 0); | |
1344 | if (r < 0) | |
1345 | log_unit_debug_errno(u, | |
1346 | r, | |
1347 | "Failed to set 'survive_final_kill_signal' xattr on control " | |
1348 | "group %s, ignoring: %m", | |
1349 | empty_to_root(u->cgroup_path)); | |
1350 | } else { | |
17d047f5 LP |
1351 | unit_remove_xattr_graceful(u, "user.survive_final_kill_signal"); |
1352 | unit_remove_xattr_graceful(u, "trusted.survive_final_kill_signal"); | |
559214cb | 1353 | } |
0d2d6fbf CD |
1354 | } |
1355 | ||
d46510de LP |
1356 | static void cgroup_xattr_apply(Unit *u) { |
1357 | assert(u); | |
1358 | ||
1359 | /* The 'user.*' xattrs can be set from a user manager. */ | |
1360 | cgroup_oomd_xattr_apply(u); | |
1361 | cgroup_log_xattr_apply(u); | |
6cf96ab4 | 1362 | cgroup_coredump_xattr_apply(u); |
d46510de LP |
1363 | |
1364 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1365 | return; | |
1366 | ||
1367 | cgroup_invocation_id_xattr_apply(u); | |
1368 | cgroup_delegate_xattr_apply(u); | |
1369 | cgroup_survive_xattr_apply(u); | |
1370 | } | |
1371 | ||
45c2e068 | 1372 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
1373 | dev_t rdev, dev = 0; |
1374 | mode_t mode; | |
45c2e068 | 1375 | int r; |
4ad49000 LP |
1376 | |
1377 | assert(p); | |
45c2e068 | 1378 | assert(ret); |
4ad49000 | 1379 | |
f5855697 | 1380 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 1381 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 1382 | struct stat st; |
57f1030b | 1383 | |
d5aecba6 LP |
1384 | if (stat(p, &st) < 0) |
1385 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
57f1030b | 1386 | |
f5855697 | 1387 | mode = st.st_mode; |
a0d6590c LP |
1388 | rdev = st.st_rdev; |
1389 | dev = st.st_dev; | |
d5aecba6 LP |
1390 | } else if (r < 0) |
1391 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
1392 | ||
57f1030b LP |
1393 | if (S_ISCHR(mode)) |
1394 | return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK), | |
1395 | "Device node '%s' is a character device, but block device needed.", p); | |
1396 | if (S_ISBLK(mode)) | |
f5855697 YS |
1397 | *ret = rdev; |
1398 | else if (major(dev) != 0) | |
1399 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
1400 | else { |
1401 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
1402 | r = btrfs_get_block_device(p, ret); | |
57f1030b LP |
1403 | if (r == -ENOTTY) |
1404 | return log_warning_errno(SYNTHETIC_ERRNO(ENODEV), | |
1405 | "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
1406 | if (r < 0) | |
45c2e068 | 1407 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); |
4ad49000 | 1408 | } |
8e274523 | 1409 | |
b7cf4b4e BB |
1410 | /* If this is a LUKS/DM device, recursively try to get the originating block device */ |
1411 | while (block_get_originating(*ret, ret) > 0); | |
45c2e068 LP |
1412 | |
1413 | /* If this is a partition, try to get the originating block device */ | |
1414 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 1415 | return 0; |
8e274523 LP |
1416 | } |
1417 | ||
66ebf6c0 TH |
1418 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
1419 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
1420 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
1421 | } | |
1422 | ||
1423 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
1424 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
1425 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
1426 | } | |
1427 | ||
31d3a520 PM |
1428 | static bool cgroup_context_has_allowed_cpus(CGroupContext *c) { |
1429 | return c->cpuset_cpus.set || c->startup_cpuset_cpus.set; | |
1430 | } | |
1431 | ||
1432 | static bool cgroup_context_has_allowed_mems(CGroupContext *c) { | |
1433 | return c->cpuset_mems.set || c->startup_cpuset_mems.set; | |
1434 | } | |
1435 | ||
a8157796 LP |
1436 | uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { |
1437 | assert(c); | |
1438 | ||
9dfb6a3a | 1439 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
66ebf6c0 TH |
1440 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) |
1441 | return c->startup_cpu_weight; | |
1442 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
1443 | return c->cpu_weight; | |
1444 | else | |
1445 | return CGROUP_WEIGHT_DEFAULT; | |
1446 | } | |
1447 | ||
1448 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
9dfb6a3a | 1449 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
66ebf6c0 TH |
1450 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) |
1451 | return c->startup_cpu_shares; | |
1452 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
1453 | return c->cpu_shares; | |
1454 | else | |
1455 | return CGROUP_CPU_SHARES_DEFAULT; | |
1456 | } | |
1457 | ||
31d3a520 | 1458 | static CPUSet *cgroup_context_allowed_cpus(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1459 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
31d3a520 PM |
1460 | c->startup_cpuset_cpus.set) |
1461 | return &c->startup_cpuset_cpus; | |
1462 | else | |
1463 | return &c->cpuset_cpus; | |
1464 | } | |
1465 | ||
1466 | static CPUSet *cgroup_context_allowed_mems(CGroupContext *c, ManagerState state) { | |
9dfb6a3a | 1467 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
31d3a520 PM |
1468 | c->startup_cpuset_mems.set) |
1469 | return &c->startup_cpuset_mems; | |
1470 | else | |
1471 | return &c->cpuset_mems; | |
1472 | } | |
1473 | ||
10f28641 FB |
1474 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
1475 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
1476 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
1477 | * Additionally, period must be at most max_period. */ | |
1478 | assert(quota > 0); | |
1479 | ||
1480 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
1481 | } | |
1482 | ||
1483 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
1484 | usec_t new_period; | |
1485 | ||
1486 | if (quota == USEC_INFINITY) | |
1487 | /* Always use default period for infinity quota. */ | |
1488 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
1489 | ||
1490 | if (period == USEC_INFINITY) | |
1491 | /* Default period was requested. */ | |
1492 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
1493 | ||
1494 | /* Clamp to interval [1ms, 1s] */ | |
1495 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
1496 | ||
1497 | if (new_period != period) { | |
8ed6f81b | 1498 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, |
10f28641 | 1499 | "Clamping CPU interval for cpu.max: period is now %s", |
5291f26d | 1500 | FORMAT_TIMESPAN(new_period, 1)); |
527ede0c | 1501 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
1502 | } |
1503 | ||
1504 | return new_period; | |
1505 | } | |
1506 | ||
52fecf20 LP |
1507 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
1508 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 | 1509 | |
c8340822 | 1510 | if (weight == CGROUP_WEIGHT_IDLE) |
1511 | return; | |
66ebf6c0 | 1512 | xsprintf(buf, "%" PRIu64 "\n", weight); |
293d32df | 1513 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
1514 | } |
1515 | ||
c8340822 | 1516 | static void cgroup_apply_unified_cpu_idle(Unit *u, uint64_t weight) { |
1517 | int r; | |
1518 | bool is_idle; | |
1519 | const char *idle_val; | |
1520 | ||
1521 | is_idle = weight == CGROUP_WEIGHT_IDLE; | |
1522 | idle_val = one_zero(is_idle); | |
1523 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.idle", idle_val); | |
1524 | if (r < 0 && (r != -ENOENT || is_idle)) | |
1525 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%s': %m", | |
1526 | "cpu.idle", empty_to_root(u->cgroup_path), idle_val); | |
1527 | } | |
1528 | ||
10f28641 | 1529 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 1530 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 1531 | |
10f28641 | 1532 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
1533 | if (quota != USEC_INFINITY) |
1534 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 1535 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 1536 | else |
10f28641 | 1537 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 1538 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
1539 | } |
1540 | ||
52fecf20 LP |
1541 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
1542 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
1543 | |
1544 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 1545 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
1546 | } |
1547 | ||
10f28641 | 1548 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 1549 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 1550 | |
10f28641 FB |
1551 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
1552 | ||
1553 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 1554 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
1555 | |
1556 | if (quota != USEC_INFINITY) { | |
10f28641 | 1557 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 1558 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 1559 | } else |
589a5f7a | 1560 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
1561 | } |
1562 | ||
1563 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
1564 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
1565 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
1566 | } | |
1567 | ||
1568 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
c8340822 | 1569 | /* we don't support idle in cgroupv1 */ |
1570 | if (weight == CGROUP_WEIGHT_IDLE) | |
1571 | return CGROUP_CPU_SHARES_MIN; | |
1572 | ||
66ebf6c0 TH |
1573 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
1574 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
1575 | } | |
1576 | ||
2cea199e | 1577 | static void cgroup_apply_unified_cpuset(Unit *u, const CPUSet *cpus, const char *name) { |
047f5d63 PH |
1578 | _cleanup_free_ char *buf = NULL; |
1579 | ||
2cea199e | 1580 | buf = cpu_set_to_range_string(cpus); |
c259ac9a LP |
1581 | if (!buf) { |
1582 | log_oom(); | |
1583 | return; | |
1584 | } | |
047f5d63 PH |
1585 | |
1586 | (void) set_attribute_and_warn(u, "cpuset", name, buf); | |
1587 | } | |
1588 | ||
508c45da | 1589 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
1590 | return c->io_accounting || |
1591 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
1592 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
1593 | c->io_device_weights || | |
6ae4283c | 1594 | c->io_device_latencies || |
538b4852 TH |
1595 | c->io_device_limits; |
1596 | } | |
1597 | ||
508c45da | 1598 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
1599 | return c->blockio_accounting || |
1600 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
1601 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
1602 | c->blockio_device_weights || | |
1603 | c->blockio_device_bandwidths; | |
1604 | } | |
1605 | ||
508c45da | 1606 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1607 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
64faf04c TH |
1608 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) |
1609 | return c->startup_io_weight; | |
d38655d7 | 1610 | if (c->io_weight != CGROUP_WEIGHT_INVALID) |
64faf04c | 1611 | return c->io_weight; |
d38655d7 | 1612 | return CGROUP_WEIGHT_DEFAULT; |
64faf04c TH |
1613 | } |
1614 | ||
508c45da | 1615 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1616 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
64faf04c TH |
1617 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) |
1618 | return c->startup_blockio_weight; | |
d38655d7 | 1619 | if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) |
64faf04c | 1620 | return c->blockio_weight; |
d38655d7 | 1621 | return CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1622 | } |
1623 | ||
508c45da | 1624 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
1625 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
1626 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
1627 | } | |
1628 | ||
508c45da | 1629 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
1630 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
1631 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
1632 | } | |
1633 | ||
3e6eafdd | 1634 | static int set_bfq_weight(Unit *u, const char *controller, dev_t dev, uint64_t io_weight) { |
1cf4a685 MK |
1635 | static const char * const prop_names[] = { |
1636 | "IOWeight", | |
1637 | "BlockIOWeight", | |
1638 | "IODeviceWeight", | |
1639 | "BlockIODeviceWeight", | |
1640 | }; | |
8d75f60e | 1641 | static bool warned = false; |
9f0c0c4e | 1642 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+STRLEN("\n")]; |
bec17e80 MK |
1643 | const char *p; |
1644 | uint64_t bfq_weight; | |
8d75f60e | 1645 | int r; |
bec17e80 MK |
1646 | |
1647 | /* FIXME: drop this function when distro kernels properly support BFQ through "io.weight" | |
1648 | * See also: https://github.com/systemd/systemd/pull/13335 and | |
1649 | * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9. */ | |
1650 | p = strjoina(controller, ".bfq.weight"); | |
1651 | /* Adjust to kernel range is 1..1000, the default is 100. */ | |
1652 | bfq_weight = BFQ_WEIGHT(io_weight); | |
1653 | ||
9f0c0c4e | 1654 | if (major(dev) > 0) |
ec61371f | 1655 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), bfq_weight); |
9f0c0c4e MK |
1656 | else |
1657 | xsprintf(buf, "%" PRIu64 "\n", bfq_weight); | |
bec17e80 | 1658 | |
8d75f60e MK |
1659 | r = cg_set_attribute(controller, u->cgroup_path, p, buf); |
1660 | ||
1661 | /* FIXME: drop this when kernels prior | |
1662 | * 795fe54c2a82 ("bfq: Add per-device weight") v5.4 | |
1663 | * are not interesting anymore. Old kernels will fail with EINVAL, while new kernels won't return | |
1664 | * EINVAL on properly formatted input by us. Treat EINVAL accordingly. */ | |
3e6eafdd MK |
1665 | if (r == -EINVAL && major(dev) > 0) { |
1666 | if (!warned) { | |
1667 | log_unit_warning(u, "Kernel version does not accept per-device setting in %s.", p); | |
1668 | warned = true; | |
1669 | } | |
1670 | r = -EOPNOTSUPP; /* mask as unconfigured device */ | |
1671 | } else if (r >= 0 && io_weight != bfq_weight) | |
1cf4a685 MK |
1672 | log_unit_debug(u, "%s=%" PRIu64 " scaled to %s=%" PRIu64, |
1673 | prop_names[2*(major(dev) > 0) + streq(controller, "blkio")], | |
bec17e80 | 1674 | io_weight, p, bfq_weight); |
3e6eafdd | 1675 | return r; |
bec17e80 MK |
1676 | } |
1677 | ||
f29ff115 | 1678 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
1679 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1680 | dev_t dev; | |
3e6eafdd | 1681 | int r, r1, r2; |
64faf04c | 1682 | |
3e6eafdd | 1683 | if (lookup_block_device(dev_path, &dev) < 0) |
64faf04c TH |
1684 | return; |
1685 | ||
3e6eafdd | 1686 | r1 = set_bfq_weight(u, "io", dev, io_weight); |
9f0c0c4e | 1687 | |
ec61371f | 1688 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), io_weight); |
3e6eafdd MK |
1689 | r2 = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
1690 | ||
1691 | /* Look at the configured device, when both fail, prefer io.weight errno. */ | |
1692 | r = r2 == -EOPNOTSUPP ? r1 : r2; | |
1693 | ||
1694 | if (r < 0) | |
1695 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), | |
1696 | r, "Failed to set 'io[.bfq].weight' attribute on '%s' to '%.*s': %m", | |
1697 | empty_to_root(u->cgroup_path), (int) strcspn(buf, NEWLINE), buf); | |
64faf04c TH |
1698 | } |
1699 | ||
f29ff115 | 1700 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
1701 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1702 | dev_t dev; | |
1703 | int r; | |
1704 | ||
1705 | r = lookup_block_device(dev_path, &dev); | |
1706 | if (r < 0) | |
1707 | return; | |
1708 | ||
ec61371f | 1709 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), blkio_weight); |
293d32df | 1710 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
1711 | } |
1712 | ||
6ae4283c TH |
1713 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
1714 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
1715 | dev_t dev; | |
1716 | int r; | |
1717 | ||
1718 | r = lookup_block_device(dev_path, &dev); | |
1719 | if (r < 0) | |
1720 | return; | |
1721 | ||
1722 | if (target != USEC_INFINITY) | |
ec61371f | 1723 | xsprintf(buf, DEVNUM_FORMAT_STR " target=%" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), target); |
6ae4283c | 1724 | else |
ec61371f | 1725 | xsprintf(buf, DEVNUM_FORMAT_STR " target=max\n", DEVNUM_FORMAT_VAL(dev)); |
6ae4283c | 1726 | |
293d32df | 1727 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
1728 | } |
1729 | ||
17ae2780 | 1730 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
4c1f9343 ZJS |
1731 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)], |
1732 | buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
64faf04c | 1733 | dev_t dev; |
64faf04c | 1734 | |
4c1f9343 | 1735 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1736 | return; |
64faf04c | 1737 | |
4c1f9343 | 1738 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
17ae2780 | 1739 | if (limits[type] != cgroup_io_limit_defaults[type]) |
64faf04c | 1740 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 1741 | else |
64faf04c | 1742 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c | 1743 | |
ec61371f | 1744 | xsprintf(buf, DEVNUM_FORMAT_STR " rbps=%s wbps=%s riops=%s wiops=%s\n", DEVNUM_FORMAT_VAL(dev), |
64faf04c TH |
1745 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], |
1746 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 1747 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
1748 | } |
1749 | ||
17ae2780 | 1750 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
1751 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1752 | dev_t dev; | |
64faf04c | 1753 | |
4c1f9343 | 1754 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1755 | return; |
64faf04c | 1756 | |
ec61371f | 1757 | sprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), rbps); |
293d32df | 1758 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 1759 | |
ec61371f | 1760 | sprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), wbps); |
293d32df | 1761 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
1762 | } |
1763 | ||
c52db42b CD |
1764 | static bool unit_has_unified_memory_config(Unit *u) { |
1765 | CGroupContext *c; | |
1766 | ||
1767 | assert(u); | |
1768 | ||
806a9362 | 1769 | assert_se(c = unit_get_cgroup_context(u)); |
c52db42b | 1770 | |
53fda560 LB |
1771 | return unit_get_ancestor_memory_min(u) > 0 || |
1772 | unit_get_ancestor_memory_low(u) > 0 || unit_get_ancestor_startup_memory_low(u) > 0 || | |
1773 | c->memory_high != CGROUP_LIMIT_MAX || c->startup_memory_high_set || | |
1774 | c->memory_max != CGROUP_LIMIT_MAX || c->startup_memory_max_set || | |
1775 | c->memory_swap_max != CGROUP_LIMIT_MAX || c->startup_memory_swap_max_set || | |
1776 | c->memory_zswap_max != CGROUP_LIMIT_MAX || c->startup_memory_zswap_max_set; | |
da4d897e TH |
1777 | } |
1778 | ||
f29ff115 | 1779 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 1780 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
1781 | |
1782 | if (v != CGROUP_LIMIT_MAX) | |
1783 | xsprintf(buf, "%" PRIu64 "\n", v); | |
1784 | ||
293d32df | 1785 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
1786 | } |
1787 | ||
0f2d84d2 | 1788 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
1789 | assert(u); |
1790 | ||
acf7f253 | 1791 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 1792 | |
acf7f253 | 1793 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
1794 | return; |
1795 | ||
fab34748 | 1796 | (void) bpf_firewall_load_custom(u); |
906c06f6 | 1797 | (void) bpf_firewall_install(u); |
906c06f6 DM |
1798 | } |
1799 | ||
49b6babb | 1800 | void unit_modify_nft_set(Unit *u, bool add) { |
dc7d69b3 | 1801 | int r; |
dc7d69b3 TM |
1802 | |
1803 | assert(u); | |
1804 | ||
1805 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1806 | return; | |
1807 | ||
49b6babb LP |
1808 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
1809 | return; | |
1810 | ||
dc7d69b3 TM |
1811 | if (cg_all_unified() <= 0) |
1812 | return; | |
1813 | ||
dc7d69b3 TM |
1814 | if (u->cgroup_id == 0) |
1815 | return; | |
1816 | ||
1817 | if (!u->manager->fw_ctx) { | |
1818 | r = fw_ctx_new_full(&u->manager->fw_ctx, /* init_tables= */ false); | |
1819 | if (r < 0) | |
1820 | return; | |
1821 | ||
1822 | assert(u->manager->fw_ctx); | |
1823 | } | |
1824 | ||
49b6babb LP |
1825 | CGroupContext *c = ASSERT_PTR(unit_get_cgroup_context(u)); |
1826 | ||
dc7d69b3 | 1827 | FOREACH_ARRAY(nft_set, c->nft_set_context.sets, c->nft_set_context.n_sets) { |
b2082753 TM |
1828 | if (nft_set->source != NFT_SET_SOURCE_CGROUP) |
1829 | continue; | |
1830 | ||
dc7d69b3 TM |
1831 | uint64_t element = u->cgroup_id; |
1832 | ||
1833 | r = nft_set_element_modify_any(u->manager->fw_ctx, add, nft_set->nfproto, nft_set->table, nft_set->set, &element, sizeof(element)); | |
1834 | if (r < 0) | |
1835 | log_warning_errno(r, "Failed to %s NFT set: family %s, table %s, set %s, cgroup %" PRIu64 ", ignoring: %m", | |
1836 | add? "add" : "delete", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, u->cgroup_id); | |
1837 | else | |
1838 | log_debug("%s NFT set: family %s, table %s, set %s, cgroup %" PRIu64, | |
1839 | add? "Added" : "Deleted", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, u->cgroup_id); | |
1840 | } | |
1841 | } | |
1842 | ||
a8e5eb17 JK |
1843 | static void cgroup_apply_socket_bind(Unit *u) { |
1844 | assert(u); | |
1845 | ||
cd09a5f3 | 1846 | (void) bpf_socket_bind_install(u); |
a8e5eb17 JK |
1847 | } |
1848 | ||
6f50d4f7 MV |
1849 | static void cgroup_apply_restrict_network_interfaces(Unit *u) { |
1850 | assert(u); | |
1851 | ||
1852 | (void) restrict_network_interfaces_install(u); | |
1853 | } | |
1854 | ||
8b139557 | 1855 | static int cgroup_apply_devices(Unit *u) { |
76dc1725 | 1856 | _cleanup_(bpf_program_freep) BPFProgram *prog = NULL; |
8b139557 ZJS |
1857 | const char *path; |
1858 | CGroupContext *c; | |
45669ae2 | 1859 | CGroupDevicePolicy policy; |
8b139557 ZJS |
1860 | int r; |
1861 | ||
1862 | assert_se(c = unit_get_cgroup_context(u)); | |
1863 | assert_se(path = u->cgroup_path); | |
1864 | ||
45669ae2 ZJS |
1865 | policy = c->device_policy; |
1866 | ||
8b139557 | 1867 | if (cg_all_unified() > 0) { |
45669ae2 | 1868 | r = bpf_devices_cgroup_init(&prog, policy, c->device_allow); |
8b139557 ZJS |
1869 | if (r < 0) |
1870 | return log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1871 | ||
1872 | } else { | |
1873 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore | |
1874 | * EINVAL here. */ | |
1875 | ||
45669ae2 | 1876 | if (c->device_allow || policy != CGROUP_DEVICE_POLICY_AUTO) |
8b139557 ZJS |
1877 | r = cg_set_attribute("devices", path, "devices.deny", "a"); |
1878 | else | |
1879 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1880 | if (r < 0) | |
8ed6f81b YW |
1881 | log_unit_full_errno(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1882 | "Failed to reset devices.allow/devices.deny: %m"); | |
8b139557 ZJS |
1883 | } |
1884 | ||
6b000af4 | 1885 | bool allow_list_static = policy == CGROUP_DEVICE_POLICY_CLOSED || |
45669ae2 | 1886 | (policy == CGROUP_DEVICE_POLICY_AUTO && c->device_allow); |
6b000af4 LP |
1887 | if (allow_list_static) |
1888 | (void) bpf_devices_allow_list_static(prog, path); | |
8b139557 | 1889 | |
6b000af4 | 1890 | bool any = allow_list_static; |
8b139557 | 1891 | LIST_FOREACH(device_allow, a, c->device_allow) { |
a1044811 LP |
1892 | const char *val; |
1893 | ||
1894 | if (a->permissions == 0) | |
8b139557 | 1895 | continue; |
8b139557 ZJS |
1896 | |
1897 | if (path_startswith(a->path, "/dev/")) | |
a1044811 | 1898 | r = bpf_devices_allow_list_device(prog, path, a->path, a->permissions); |
8b139557 | 1899 | else if ((val = startswith(a->path, "block-"))) |
a1044811 | 1900 | r = bpf_devices_allow_list_major(prog, path, val, 'b', a->permissions); |
8b139557 | 1901 | else if ((val = startswith(a->path, "char-"))) |
a1044811 | 1902 | r = bpf_devices_allow_list_major(prog, path, val, 'c', a->permissions); |
45669ae2 | 1903 | else { |
8b139557 | 1904 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
45669ae2 ZJS |
1905 | continue; |
1906 | } | |
1907 | ||
1908 | if (r >= 0) | |
1909 | any = true; | |
1910 | } | |
1911 | ||
1912 | if (prog && !any) { | |
1913 | log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENODEV), "No devices matched by device filter."); | |
1914 | ||
1915 | /* The kernel verifier would reject a program we would build with the normal intro and outro | |
6b000af4 | 1916 | but no allow-listing rules (outro would contain an unreachable instruction for successful |
45669ae2 ZJS |
1917 | return). */ |
1918 | policy = CGROUP_DEVICE_POLICY_STRICT; | |
8b139557 ZJS |
1919 | } |
1920 | ||
76dc1725 | 1921 | r = bpf_devices_apply_policy(&prog, policy, any, path, &u->bpf_device_control_installed); |
8b139557 ZJS |
1922 | if (r < 0) { |
1923 | static bool warned = false; | |
1924 | ||
1925 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1926 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1927 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1928 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1929 | ||
1930 | warned = true; | |
1931 | } | |
1932 | return r; | |
1933 | } | |
1934 | ||
17283ce7 YW |
1935 | static void set_io_weight(Unit *u, uint64_t weight) { |
1936 | char buf[STRLEN("default \n")+DECIMAL_STR_MAX(uint64_t)]; | |
17283ce7 YW |
1937 | |
1938 | assert(u); | |
29eb0eef | 1939 | |
3e6eafdd | 1940 | (void) set_bfq_weight(u, "io", makedev(0, 0), weight); |
29eb0eef | 1941 | |
29eb0eef | 1942 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
17283ce7 YW |
1943 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
1944 | } | |
1945 | ||
1946 | static void set_blkio_weight(Unit *u, uint64_t weight) { | |
1947 | char buf[STRLEN("\n")+DECIMAL_STR_MAX(uint64_t)]; | |
17283ce7 YW |
1948 | |
1949 | assert(u); | |
29eb0eef | 1950 | |
3e6eafdd | 1951 | (void) set_bfq_weight(u, "blkio", makedev(0, 0), weight); |
17283ce7 YW |
1952 | |
1953 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
1954 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); | |
29eb0eef ZJS |
1955 | } |
1956 | ||
506ea51b JK |
1957 | static void cgroup_apply_bpf_foreign_program(Unit *u) { |
1958 | assert(u); | |
1959 | ||
1960 | (void) bpf_foreign_install(u); | |
1961 | } | |
1962 | ||
906c06f6 DM |
1963 | static void cgroup_context_apply( |
1964 | Unit *u, | |
1965 | CGroupMask apply_mask, | |
906c06f6 DM |
1966 | ManagerState state) { |
1967 | ||
f29ff115 TH |
1968 | const char *path; |
1969 | CGroupContext *c; | |
52fecf20 | 1970 | bool is_host_root, is_local_root; |
4ad49000 LP |
1971 | int r; |
1972 | ||
f29ff115 TH |
1973 | assert(u); |
1974 | ||
906c06f6 | 1975 | /* Nothing to do? Exit early! */ |
17f14955 | 1976 | if (apply_mask == 0) |
4ad49000 | 1977 | return; |
8e274523 | 1978 | |
52fecf20 LP |
1979 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
1980 | * attributes should only be managed for cgroups further down the tree. */ | |
1981 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
1982 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
1983 | |
1984 | assert_se(c = unit_get_cgroup_context(u)); | |
1985 | assert_se(path = u->cgroup_path); | |
1986 | ||
52fecf20 | 1987 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 1988 | path = "/"; |
01efdf13 | 1989 | |
be2c0327 LP |
1990 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
1991 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 1992 | |
be2c0327 LP |
1993 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
1994 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
1995 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 1996 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
1997 | * we couldn't even write to them if we wanted to). */ |
1998 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 1999 | |
b4cccbc1 | 2000 | if (cg_all_unified() > 0) { |
be2c0327 | 2001 | uint64_t weight; |
b2f8b02e | 2002 | |
be2c0327 LP |
2003 | if (cgroup_context_has_cpu_weight(c)) |
2004 | weight = cgroup_context_cpu_weight(c, state); | |
2005 | else if (cgroup_context_has_cpu_shares(c)) { | |
2006 | uint64_t shares; | |
66ebf6c0 | 2007 | |
be2c0327 LP |
2008 | shares = cgroup_context_cpu_shares(c, state); |
2009 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 2010 | |
be2c0327 LP |
2011 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
2012 | shares, weight, path); | |
2013 | } else | |
2014 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 2015 | |
c8340822 | 2016 | cgroup_apply_unified_cpu_idle(u, weight); |
be2c0327 | 2017 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 2018 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 2019 | |
52fecf20 | 2020 | } else { |
be2c0327 | 2021 | uint64_t shares; |
52fecf20 | 2022 | |
be2c0327 LP |
2023 | if (cgroup_context_has_cpu_weight(c)) { |
2024 | uint64_t weight; | |
52fecf20 | 2025 | |
be2c0327 LP |
2026 | weight = cgroup_context_cpu_weight(c, state); |
2027 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 2028 | |
be2c0327 LP |
2029 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
2030 | weight, shares, path); | |
2031 | } else if (cgroup_context_has_cpu_shares(c)) | |
2032 | shares = cgroup_context_cpu_shares(c, state); | |
2033 | else | |
2034 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 2035 | |
be2c0327 | 2036 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 2037 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 2038 | } |
4ad49000 LP |
2039 | } |
2040 | ||
047f5d63 | 2041 | if ((apply_mask & CGROUP_MASK_CPUSET) && !is_local_root) { |
31d3a520 PM |
2042 | cgroup_apply_unified_cpuset(u, cgroup_context_allowed_cpus(c, state), "cpuset.cpus"); |
2043 | cgroup_apply_unified_cpuset(u, cgroup_context_allowed_mems(c, state), "cpuset.mems"); | |
047f5d63 PH |
2044 | } |
2045 | ||
4e1dfa45 | 2046 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
2047 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
2048 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
2049 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
52fecf20 LP |
2050 | bool has_io, has_blockio; |
2051 | uint64_t weight; | |
13c31542 | 2052 | |
52fecf20 LP |
2053 | has_io = cgroup_context_has_io_config(c); |
2054 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 2055 | |
52fecf20 LP |
2056 | if (has_io) |
2057 | weight = cgroup_context_io_weight(c, state); | |
2058 | else if (has_blockio) { | |
2059 | uint64_t blkio_weight; | |
128fadc9 | 2060 | |
52fecf20 LP |
2061 | blkio_weight = cgroup_context_blkio_weight(c, state); |
2062 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 2063 | |
67e2ea15 | 2064 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
2065 | blkio_weight, weight); |
2066 | } else | |
2067 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 2068 | |
17283ce7 | 2069 | set_io_weight(u, weight); |
2dbc45ae | 2070 | |
52fecf20 | 2071 | if (has_io) { |
52fecf20 LP |
2072 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
2073 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 2074 | |
52fecf20 LP |
2075 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
2076 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 2077 | |
52fecf20 LP |
2078 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
2079 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 2080 | |
52fecf20 | 2081 | } else if (has_blockio) { |
52fecf20 LP |
2082 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
2083 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 2084 | |
67e2ea15 | 2085 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 2086 | w->weight, weight, w->path); |
538b4852 | 2087 | |
52fecf20 LP |
2088 | cgroup_apply_io_device_weight(u, w->path, weight); |
2089 | } | |
538b4852 | 2090 | |
17ae2780 | 2091 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 | 2092 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
538b4852 | 2093 | |
e8616626 | 2094 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
538b4852 TH |
2095 | limits[type] = cgroup_io_limit_defaults[type]; |
2096 | ||
2097 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
2098 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
2099 | ||
67e2ea15 | 2100 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
2101 | b->rbps, b->wbps, b->path); |
2102 | ||
17ae2780 | 2103 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 2104 | } |
13c31542 TH |
2105 | } |
2106 | } | |
2107 | ||
906c06f6 | 2108 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 2109 | bool has_io, has_blockio; |
4ad49000 | 2110 | |
52fecf20 LP |
2111 | has_io = cgroup_context_has_io_config(c); |
2112 | has_blockio = cgroup_context_has_blockio_config(c); | |
2113 | ||
2114 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
2115 | * left to our container manager, too. */ | |
2116 | if (!is_local_root) { | |
64faf04c | 2117 | uint64_t weight; |
64faf04c | 2118 | |
7d862ab8 | 2119 | if (has_io) { |
52fecf20 | 2120 | uint64_t io_weight; |
128fadc9 | 2121 | |
52fecf20 | 2122 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 2123 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 2124 | |
67e2ea15 | 2125 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 2126 | io_weight, weight); |
7d862ab8 TH |
2127 | } else if (has_blockio) |
2128 | weight = cgroup_context_blkio_weight(c, state); | |
2129 | else | |
538b4852 | 2130 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c | 2131 | |
17283ce7 | 2132 | set_blkio_weight(u, weight); |
35e7a62c | 2133 | |
03677889 | 2134 | if (has_io) |
128fadc9 TH |
2135 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
2136 | weight = cgroup_weight_io_to_blkio(w->weight); | |
2137 | ||
67e2ea15 | 2138 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
2139 | w->weight, weight, w->path); |
2140 | ||
2141 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
2142 | } | |
03677889 | 2143 | else if (has_blockio) |
7d862ab8 TH |
2144 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
2145 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
4ad49000 LP |
2146 | } |
2147 | ||
5238e957 | 2148 | /* The bandwidth limits are something that make sense to be applied to the host's root but not container |
52fecf20 LP |
2149 | * roots, as there we want the container manager to handle it */ |
2150 | if (is_host_root || !is_local_root) { | |
03677889 | 2151 | if (has_io) |
52fecf20 | 2152 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 2153 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 2154 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 2155 | |
52fecf20 LP |
2156 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
2157 | } | |
03677889 | 2158 | else if (has_blockio) |
52fecf20 LP |
2159 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
2160 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
d686d8a9 | 2161 | } |
8e274523 LP |
2162 | } |
2163 | ||
be2c0327 LP |
2164 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
2165 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 2166 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
2167 | * write to this if we wanted to.) */ |
2168 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 2169 | |
52fecf20 | 2170 | if (cg_all_unified() > 0) { |
53fda560 | 2171 | uint64_t max, swap_max = CGROUP_LIMIT_MAX, zswap_max = CGROUP_LIMIT_MAX, high = CGROUP_LIMIT_MAX; |
be2c0327 | 2172 | |
c52db42b | 2173 | if (unit_has_unified_memory_config(u)) { |
53fda560 LB |
2174 | bool startup = IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING); |
2175 | ||
2176 | high = startup && c->startup_memory_high_set ? c->startup_memory_high : c->memory_high; | |
2177 | max = startup && c->startup_memory_max_set ? c->startup_memory_max : c->memory_max; | |
2178 | swap_max = startup && c->startup_memory_swap_max_set ? c->startup_memory_swap_max : c->memory_swap_max; | |
2179 | zswap_max = startup && c->startup_memory_zswap_max_set ? c->startup_memory_zswap_max : c->memory_zswap_max; | |
be2c0327 LP |
2180 | } else { |
2181 | max = c->memory_limit; | |
efdb0237 | 2182 | |
be2c0327 LP |
2183 | if (max != CGROUP_LIMIT_MAX) |
2184 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 2185 | } |
da4d897e | 2186 | |
64fe532e | 2187 | cgroup_apply_unified_memory_limit(u, "memory.min", unit_get_ancestor_memory_min(u)); |
c52db42b | 2188 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
53fda560 | 2189 | cgroup_apply_unified_memory_limit(u, "memory.high", high); |
be2c0327 LP |
2190 | cgroup_apply_unified_memory_limit(u, "memory.max", max); |
2191 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
d7fe0a67 | 2192 | cgroup_apply_unified_memory_limit(u, "memory.zswap.max", zswap_max); |
128fadc9 | 2193 | |
afcfaa69 LP |
2194 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
2195 | ||
be2c0327 LP |
2196 | } else { |
2197 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
2198 | uint64_t val; | |
52fecf20 | 2199 | |
c52db42b | 2200 | if (unit_has_unified_memory_config(u)) { |
be2c0327 | 2201 | val = c->memory_max; |
b7a41491 MK |
2202 | if (val != CGROUP_LIMIT_MAX) |
2203 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIu64 " as MemoryLimit=", val); | |
be2c0327 LP |
2204 | } else |
2205 | val = c->memory_limit; | |
78a4ee59 | 2206 | |
be2c0327 LP |
2207 | if (val == CGROUP_LIMIT_MAX) |
2208 | strncpy(buf, "-1\n", sizeof(buf)); | |
2209 | else | |
2210 | xsprintf(buf, "%" PRIu64 "\n", val); | |
2211 | ||
2212 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 2213 | } |
4ad49000 | 2214 | } |
8e274523 | 2215 | |
4e1dfa45 | 2216 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
2217 | * containers, where we leave this to the manager */ |
2218 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
8b139557 ZJS |
2219 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) |
2220 | (void) cgroup_apply_devices(u); | |
03a7b521 | 2221 | |
00b5974f LP |
2222 | if (apply_mask & CGROUP_MASK_PIDS) { |
2223 | ||
52fecf20 | 2224 | if (is_host_root) { |
00b5974f LP |
2225 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
2226 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
2227 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
2228 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
2229 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
2230 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
2231 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
2232 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
2233 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
2234 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
5238e957 | 2235 | * which is desirable so that there's an official way to release control of the sysctl from |
00b5974f LP |
2236 | * systemd: set the limit to unbounded and reload. */ |
2237 | ||
94f0b13b | 2238 | if (cgroup_tasks_max_isset(&c->tasks_max)) { |
00b5974f | 2239 | u->manager->sysctl_pid_max_changed = true; |
94f0b13b | 2240 | r = procfs_tasks_set_limit(cgroup_tasks_max_resolve(&c->tasks_max)); |
00b5974f LP |
2241 | } else if (u->manager->sysctl_pid_max_changed) |
2242 | r = procfs_tasks_set_limit(TASKS_MAX); | |
2243 | else | |
2244 | r = 0; | |
00b5974f | 2245 | if (r < 0) |
8ed6f81b YW |
2246 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
2247 | "Failed to write to tasks limit sysctls: %m"); | |
52fecf20 | 2248 | } |
03a7b521 | 2249 | |
52fecf20 LP |
2250 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
2251 | * leave it for the container manager. */ | |
2252 | if (!is_local_root) { | |
94f0b13b | 2253 | if (cgroup_tasks_max_isset(&c->tasks_max)) { |
3a0f06c4 | 2254 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
03a7b521 | 2255 | |
94f0b13b | 2256 | xsprintf(buf, "%" PRIu64 "\n", cgroup_tasks_max_resolve(&c->tasks_max)); |
293d32df | 2257 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 2258 | } else |
589a5f7a | 2259 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 2260 | } |
03a7b521 | 2261 | } |
906c06f6 | 2262 | |
17f14955 | 2263 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 2264 | cgroup_apply_firewall(u); |
506ea51b JK |
2265 | |
2266 | if (apply_mask & CGROUP_MASK_BPF_FOREIGN) | |
2267 | cgroup_apply_bpf_foreign_program(u); | |
a8e5eb17 JK |
2268 | |
2269 | if (apply_mask & CGROUP_MASK_BPF_SOCKET_BIND) | |
2270 | cgroup_apply_socket_bind(u); | |
6f50d4f7 MV |
2271 | |
2272 | if (apply_mask & CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES) | |
2273 | cgroup_apply_restrict_network_interfaces(u); | |
dc7d69b3 | 2274 | |
49b6babb | 2275 | unit_modify_nft_set(u, /* add = */ true); |
fb385181 LP |
2276 | } |
2277 | ||
16492445 LP |
2278 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
2279 | CGroupContext *c; | |
16492445 LP |
2280 | assert(u); |
2281 | ||
2282 | c = unit_get_cgroup_context(u); | |
2283 | if (!c) | |
2284 | return false; | |
2285 | ||
2286 | if (c->ip_accounting || | |
84ebe6f0 YW |
2287 | !set_isempty(c->ip_address_allow) || |
2288 | !set_isempty(c->ip_address_deny) || | |
fab34748 KL |
2289 | c->ip_filters_ingress || |
2290 | c->ip_filters_egress) | |
16492445 LP |
2291 | return true; |
2292 | ||
2293 | /* If any parent slice has an IP access list defined, it applies too */ | |
e8616626 | 2294 | for (Unit *p = UNIT_GET_SLICE(u); p; p = UNIT_GET_SLICE(p)) { |
16492445 LP |
2295 | c = unit_get_cgroup_context(p); |
2296 | if (!c) | |
2297 | return false; | |
2298 | ||
84ebe6f0 YW |
2299 | if (!set_isempty(c->ip_address_allow) || |
2300 | !set_isempty(c->ip_address_deny)) | |
16492445 LP |
2301 | return true; |
2302 | } | |
2303 | ||
2304 | return false; | |
2305 | } | |
2306 | ||
506ea51b JK |
2307 | static bool unit_get_needs_bpf_foreign_program(Unit *u) { |
2308 | CGroupContext *c; | |
2309 | assert(u); | |
2310 | ||
2311 | c = unit_get_cgroup_context(u); | |
2312 | if (!c) | |
2313 | return false; | |
2314 | ||
64903d18 | 2315 | return !!c->bpf_foreign_programs; |
506ea51b JK |
2316 | } |
2317 | ||
a8e5eb17 JK |
2318 | static bool unit_get_needs_socket_bind(Unit *u) { |
2319 | CGroupContext *c; | |
2320 | assert(u); | |
2321 | ||
2322 | c = unit_get_cgroup_context(u); | |
2323 | if (!c) | |
2324 | return false; | |
2325 | ||
11ab01e4 | 2326 | return c->socket_bind_allow || c->socket_bind_deny; |
a8e5eb17 JK |
2327 | } |
2328 | ||
6f50d4f7 MV |
2329 | static bool unit_get_needs_restrict_network_interfaces(Unit *u) { |
2330 | CGroupContext *c; | |
2331 | assert(u); | |
2332 | ||
2333 | c = unit_get_cgroup_context(u); | |
2334 | if (!c) | |
2335 | return false; | |
2336 | ||
2337 | return !set_isempty(c->restrict_network_interfaces); | |
2338 | } | |
2339 | ||
c52db42b | 2340 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 2341 | CGroupMask mask = 0; |
c52db42b CD |
2342 | CGroupContext *c; |
2343 | ||
2344 | assert(u); | |
2345 | ||
806a9362 | 2346 | assert_se(c = unit_get_cgroup_context(u)); |
c710d3b4 | 2347 | |
fae9bc29 | 2348 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 2349 | |
fae9bc29 | 2350 | if (c->cpu_accounting) |
f98c2585 | 2351 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
2352 | |
2353 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 2354 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 2355 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 2356 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 2357 | |
31d3a520 | 2358 | if (cgroup_context_has_allowed_cpus(c) || cgroup_context_has_allowed_mems(c)) |
047f5d63 PH |
2359 | mask |= CGROUP_MASK_CPUSET; |
2360 | ||
538b4852 TH |
2361 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
2362 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 2363 | |
4ad49000 | 2364 | if (c->memory_accounting || |
da4d897e | 2365 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 2366 | unit_has_unified_memory_config(u)) |
efdb0237 | 2367 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 2368 | |
a931ad47 | 2369 | if (c->device_allow || |
084870f9 | 2370 | c->device_policy != CGROUP_DEVICE_POLICY_AUTO) |
084c7007 | 2371 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 2372 | |
03a7b521 | 2373 | if (c->tasks_accounting || |
94f0b13b | 2374 | cgroup_tasks_max_isset(&c->tasks_max)) |
03a7b521 LP |
2375 | mask |= CGROUP_MASK_PIDS; |
2376 | ||
fae9bc29 | 2377 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
2378 | } |
2379 | ||
53aea74a | 2380 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
2381 | CGroupMask mask = 0; |
2382 | ||
fae9bc29 LP |
2383 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
2384 | * too. */ | |
2385 | ||
17f14955 RG |
2386 | if (unit_get_needs_bpf_firewall(u)) |
2387 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2388 | ||
506ea51b JK |
2389 | if (unit_get_needs_bpf_foreign_program(u)) |
2390 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
2391 | ||
a8e5eb17 JK |
2392 | if (unit_get_needs_socket_bind(u)) |
2393 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
2394 | ||
6f50d4f7 MV |
2395 | if (unit_get_needs_restrict_network_interfaces(u)) |
2396 | mask |= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES; | |
2397 | ||
17f14955 RG |
2398 | return mask; |
2399 | } | |
2400 | ||
efdb0237 | 2401 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 2402 | CGroupContext *c; |
8e274523 | 2403 | |
442ce775 LP |
2404 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
2405 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
2406 | ||
2407 | if (u->load_state != UNIT_LOADED) | |
2408 | return 0; | |
efdb0237 | 2409 | |
4ad49000 LP |
2410 | c = unit_get_cgroup_context(u); |
2411 | if (!c) | |
2412 | return 0; | |
8e274523 | 2413 | |
12b975e0 | 2414 | return unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u); |
02638280 LP |
2415 | } |
2416 | ||
2417 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
2418 | CGroupContext *c; | |
2419 | ||
2420 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
2421 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 2422 | * |
02638280 | 2423 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 2424 | |
1d9cc876 | 2425 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
2426 | return 0; |
2427 | ||
2428 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
2429 | ExecContext *e; |
2430 | ||
2431 | e = unit_get_exec_context(u); | |
02638280 LP |
2432 | if (e && !exec_context_maintains_privileges(e)) |
2433 | return 0; | |
a931ad47 LP |
2434 | } |
2435 | ||
1d9cc876 | 2436 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 2437 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
2438 | } |
2439 | ||
d9ef5944 MK |
2440 | static CGroupMask unit_get_subtree_mask(Unit *u) { |
2441 | ||
2442 | /* Returns the mask of this subtree, meaning of the group | |
2443 | * itself and its children. */ | |
2444 | ||
2445 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
2446 | } | |
2447 | ||
efdb0237 | 2448 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 2449 | assert(u); |
bc432dc7 | 2450 | |
02638280 | 2451 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 2452 | |
bc432dc7 | 2453 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 2454 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 2455 | |
64e844e5 | 2456 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
2457 | |
2458 | if (u->type == UNIT_SLICE) { | |
2459 | Unit *member; | |
bc432dc7 | 2460 | |
d219a2b0 | 2461 | UNIT_FOREACH_DEPENDENCY(member, u, UNIT_ATOM_SLICE_OF) |
15ed3c3a | 2462 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
2463 | } |
2464 | ||
2465 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 2466 | return u->cgroup_members_mask; |
246aa6dd LP |
2467 | } |
2468 | ||
efdb0237 | 2469 | CGroupMask unit_get_siblings_mask(Unit *u) { |
12f64221 | 2470 | Unit *slice; |
4ad49000 | 2471 | assert(u); |
246aa6dd | 2472 | |
efdb0237 LP |
2473 | /* Returns the mask of controllers all of the unit's siblings |
2474 | * require, i.e. the members mask of the unit's parent slice | |
2475 | * if there is one. */ | |
2476 | ||
12f64221 LP |
2477 | slice = UNIT_GET_SLICE(u); |
2478 | if (slice) | |
2479 | return unit_get_members_mask(slice); | |
4ad49000 | 2480 | |
64e844e5 | 2481 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
2482 | } |
2483 | ||
d9ef5944 | 2484 | static CGroupMask unit_get_disable_mask(Unit *u) { |
4f6f62e4 CD |
2485 | CGroupContext *c; |
2486 | ||
2487 | c = unit_get_cgroup_context(u); | |
2488 | if (!c) | |
2489 | return 0; | |
2490 | ||
2491 | return c->disable_controllers; | |
2492 | } | |
2493 | ||
2494 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
2495 | CGroupMask mask; | |
12f64221 | 2496 | Unit *slice; |
4f6f62e4 CD |
2497 | |
2498 | assert(u); | |
2499 | mask = unit_get_disable_mask(u); | |
2500 | ||
2501 | /* Returns the mask of controllers which are marked as forcibly | |
2502 | * disabled in any ancestor unit or the unit in question. */ | |
2503 | ||
12f64221 LP |
2504 | slice = UNIT_GET_SLICE(u); |
2505 | if (slice) | |
2506 | mask |= unit_get_ancestor_disable_mask(slice); | |
4f6f62e4 CD |
2507 | |
2508 | return mask; | |
2509 | } | |
2510 | ||
efdb0237 | 2511 | CGroupMask unit_get_target_mask(Unit *u) { |
a437c5e4 | 2512 | CGroupMask own_mask, mask; |
efdb0237 | 2513 | |
a437c5e4 LP |
2514 | /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything |
2515 | * it needs itself, plus all that its children need, plus all that its siblings need. This is | |
2516 | * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each | |
efdb0237 | 2517 | * hierarchy that shall be enabled for it. */ |
6414b7c9 | 2518 | |
a437c5e4 | 2519 | own_mask = unit_get_own_mask(u); |
84d2744b | 2520 | |
a437c5e4 | 2521 | if (own_mask & CGROUP_MASK_BPF_FIREWALL & ~u->manager->cgroup_supported) |
84d2744b ZJS |
2522 | emit_bpf_firewall_warning(u); |
2523 | ||
a437c5e4 LP |
2524 | mask = own_mask | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
2525 | ||
efdb0237 | 2526 | mask &= u->manager->cgroup_supported; |
c72703e2 | 2527 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
2528 | |
2529 | return mask; | |
2530 | } | |
2531 | ||
2532 | CGroupMask unit_get_enable_mask(Unit *u) { | |
2533 | CGroupMask mask; | |
2534 | ||
2535 | /* This returns the cgroup mask of all controllers to enable | |
2536 | * for the children of a specific cgroup. This is primarily | |
2537 | * useful for the unified cgroup hierarchy, where each cgroup | |
2538 | * controls which controllers are enabled for its children. */ | |
2539 | ||
2540 | mask = unit_get_members_mask(u); | |
6414b7c9 | 2541 | mask &= u->manager->cgroup_supported; |
c72703e2 | 2542 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
2543 | |
2544 | return mask; | |
2545 | } | |
2546 | ||
5af88058 | 2547 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
12f64221 LP |
2548 | Unit *slice; |
2549 | ||
bc432dc7 LP |
2550 | assert(u); |
2551 | ||
5af88058 LP |
2552 | /* Recurse invalidate the member masks cache all the way up the tree */ |
2553 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 2554 | |
12f64221 LP |
2555 | slice = UNIT_GET_SLICE(u); |
2556 | if (slice) | |
2557 | unit_invalidate_cgroup_members_masks(slice); | |
6414b7c9 DS |
2558 | } |
2559 | ||
6592b975 | 2560 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 2561 | |
6592b975 | 2562 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
2563 | |
2564 | while (u) { | |
6592b975 | 2565 | |
03b90d4b LP |
2566 | if (u->cgroup_path && |
2567 | u->cgroup_realized && | |
d94a24ca | 2568 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
2569 | return u->cgroup_path; |
2570 | ||
12f64221 | 2571 | u = UNIT_GET_SLICE(u); |
03b90d4b LP |
2572 | } |
2573 | ||
2574 | return NULL; | |
2575 | } | |
2576 | ||
6592b975 | 2577 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
7b639614 MK |
2578 | /* If not realized at all, migrate to root (""). |
2579 | * It may happen if we're upgrading from older version that didn't clean up. | |
2580 | */ | |
2581 | return strempty(unit_get_realized_cgroup_path(userdata, mask)); | |
6592b975 LP |
2582 | } |
2583 | ||
1a56b0c0 LP |
2584 | int unit_default_cgroup_path(const Unit *u, char **ret) { |
2585 | _cleanup_free_ char *p = NULL; | |
efdb0237 LP |
2586 | int r; |
2587 | ||
2588 | assert(u); | |
1a56b0c0 | 2589 | assert(ret); |
efdb0237 LP |
2590 | |
2591 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1a56b0c0 LP |
2592 | p = strdup(u->manager->cgroup_root); |
2593 | else { | |
2594 | _cleanup_free_ char *escaped = NULL, *slice_path = NULL; | |
2595 | Unit *slice; | |
efdb0237 | 2596 | |
1a56b0c0 LP |
2597 | slice = UNIT_GET_SLICE(u); |
2598 | if (slice && !unit_has_name(slice, SPECIAL_ROOT_SLICE)) { | |
2599 | r = cg_slice_to_path(slice->id, &slice_path); | |
2600 | if (r < 0) | |
2601 | return r; | |
2602 | } | |
2603 | ||
2604 | r = cg_escape(u->id, &escaped); | |
efdb0237 | 2605 | if (r < 0) |
1a56b0c0 | 2606 | return r; |
efdb0237 | 2607 | |
1a56b0c0 LP |
2608 | p = path_join(empty_to_root(u->manager->cgroup_root), slice_path, escaped); |
2609 | } | |
2610 | if (!p) | |
2611 | return -ENOMEM; | |
efdb0237 | 2612 | |
1a56b0c0 LP |
2613 | *ret = TAKE_PTR(p); |
2614 | return 0; | |
efdb0237 LP |
2615 | } |
2616 | ||
2617 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
2618 | _cleanup_free_ char *p = NULL; | |
2619 | int r; | |
2620 | ||
2621 | assert(u); | |
2622 | ||
5210387e LP |
2623 | if (streq_ptr(u->cgroup_path, path)) |
2624 | return 0; | |
2625 | ||
efdb0237 LP |
2626 | if (path) { |
2627 | p = strdup(path); | |
2628 | if (!p) | |
2629 | return -ENOMEM; | |
5210387e | 2630 | } |
efdb0237 LP |
2631 | |
2632 | if (p) { | |
2633 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
2634 | if (r < 0) | |
2635 | return r; | |
2636 | } | |
2637 | ||
2638 | unit_release_cgroup(u); | |
ae2a15bc | 2639 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
2640 | |
2641 | return 1; | |
2642 | } | |
2643 | ||
2644 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 2645 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
2646 | int r; |
2647 | ||
2648 | assert(u); | |
2649 | ||
0bb814c2 LP |
2650 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
2651 | * cgroupv2 is available. */ | |
2652 | ||
efdb0237 LP |
2653 | if (!u->cgroup_path) |
2654 | return 0; | |
2655 | ||
0bb814c2 | 2656 | if (u->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
2657 | return 0; |
2658 | ||
2659 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 2660 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2661 | if (r < 0) |
2662 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
2663 | if (r == 0) | |
efdb0237 LP |
2664 | return 0; |
2665 | ||
0bb814c2 | 2666 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
2667 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
2668 | return 0; | |
2669 | ||
0bb814c2 | 2670 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
2671 | if (r < 0) |
2672 | return log_oom(); | |
2673 | ||
ab2c3861 | 2674 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
2675 | if (r < 0) |
2676 | return log_oom(); | |
2677 | ||
0bb814c2 LP |
2678 | u->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
2679 | if (u->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 2680 | |
0bb814c2 LP |
2681 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
2682 | * is not an error */ | |
efdb0237 LP |
2683 | return 0; |
2684 | ||
6178e2f8 | 2685 | return log_unit_error_errno(u, errno, "Failed to add control inotify watch descriptor for control group %s: %m", empty_to_root(u->cgroup_path)); |
efdb0237 LP |
2686 | } |
2687 | ||
0bb814c2 | 2688 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd), u); |
efdb0237 | 2689 | if (r < 0) |
6178e2f8 | 2690 | return log_unit_error_errno(u, r, "Failed to add control inotify watch descriptor for control group %s to hash map: %m", empty_to_root(u->cgroup_path)); |
efdb0237 LP |
2691 | |
2692 | return 0; | |
2693 | } | |
2694 | ||
afcfaa69 LP |
2695 | int unit_watch_cgroup_memory(Unit *u) { |
2696 | _cleanup_free_ char *events = NULL; | |
2697 | CGroupContext *c; | |
2698 | int r; | |
2699 | ||
2700 | assert(u); | |
2701 | ||
2702 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
2703 | * cgroupv2 is available. */ | |
2704 | ||
2705 | if (!u->cgroup_path) | |
2706 | return 0; | |
2707 | ||
2708 | c = unit_get_cgroup_context(u); | |
2709 | if (!c) | |
2710 | return 0; | |
2711 | ||
2712 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
2713 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
2714 | * all. */ | |
2715 | if (!c->memory_accounting) | |
2716 | return 0; | |
2717 | ||
2718 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
2719 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
2720 | if (u->type == UNIT_SLICE) | |
2721 | return 0; | |
2722 | ||
2723 | if (u->cgroup_memory_inotify_wd >= 0) | |
2724 | return 0; | |
2725 | ||
2726 | /* Only applies to the unified hierarchy */ | |
2727 | r = cg_all_unified(); | |
2728 | if (r < 0) | |
2729 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
2730 | if (r == 0) | |
2731 | return 0; | |
2732 | ||
2733 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
2734 | if (r < 0) | |
2735 | return log_oom(); | |
2736 | ||
2737 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "memory.events", &events); | |
2738 | if (r < 0) | |
2739 | return log_oom(); | |
2740 | ||
2741 | u->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); | |
2742 | if (u->cgroup_memory_inotify_wd < 0) { | |
2743 | ||
2744 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
2745 | * is not an error */ | |
2746 | return 0; | |
2747 | ||
6178e2f8 | 2748 | return log_unit_error_errno(u, errno, "Failed to add memory inotify watch descriptor for control group %s: %m", empty_to_root(u->cgroup_path)); |
afcfaa69 LP |
2749 | } |
2750 | ||
2751 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd), u); | |
2752 | if (r < 0) | |
6178e2f8 | 2753 | return log_unit_error_errno(u, r, "Failed to add memory inotify watch descriptor for control group %s to hash map: %m", empty_to_root(u->cgroup_path)); |
afcfaa69 LP |
2754 | |
2755 | return 0; | |
2756 | } | |
2757 | ||
a4634b21 LP |
2758 | int unit_pick_cgroup_path(Unit *u) { |
2759 | _cleanup_free_ char *path = NULL; | |
2760 | int r; | |
2761 | ||
2762 | assert(u); | |
2763 | ||
2764 | if (u->cgroup_path) | |
2765 | return 0; | |
2766 | ||
2767 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2768 | return -EINVAL; | |
2769 | ||
1a56b0c0 LP |
2770 | r = unit_default_cgroup_path(u, &path); |
2771 | if (r < 0) | |
2772 | return log_unit_error_errno(u, r, "Failed to generate default cgroup path: %m"); | |
a4634b21 LP |
2773 | |
2774 | r = unit_set_cgroup_path(u, path); | |
2775 | if (r == -EEXIST) | |
6178e2f8 | 2776 | return log_unit_error_errno(u, r, "Control group %s exists already.", empty_to_root(path)); |
a4634b21 | 2777 | if (r < 0) |
6178e2f8 | 2778 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", empty_to_root(path)); |
a4634b21 LP |
2779 | |
2780 | return 0; | |
2781 | } | |
2782 | ||
7b639614 | 2783 | static int unit_update_cgroup( |
efdb0237 LP |
2784 | Unit *u, |
2785 | CGroupMask target_mask, | |
0d2d6fbf CD |
2786 | CGroupMask enable_mask, |
2787 | ManagerState state) { | |
efdb0237 | 2788 | |
7b639614 MK |
2789 | bool created, is_root_slice; |
2790 | CGroupMask migrate_mask = 0; | |
184b4f78 | 2791 | _cleanup_free_ char *cgroup_full_path = NULL; |
27adcc97 | 2792 | int r; |
64747e2d | 2793 | |
4ad49000 | 2794 | assert(u); |
64747e2d | 2795 | |
27c4ed79 | 2796 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
2797 | return 0; |
2798 | ||
a4634b21 LP |
2799 | /* Figure out our cgroup path */ |
2800 | r = unit_pick_cgroup_path(u); | |
2801 | if (r < 0) | |
2802 | return r; | |
b58b8e11 | 2803 | |
03b90d4b | 2804 | /* First, create our own group */ |
efdb0237 | 2805 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 2806 | if (r < 0) |
6178e2f8 | 2807 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", empty_to_root(u->cgroup_path)); |
490c5a37 | 2808 | created = r; |
efdb0237 | 2809 | |
184b4f78 | 2810 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
1b420223 LP |
2811 | uint64_t cgroup_id = 0; |
2812 | ||
184b4f78 ILG |
2813 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, NULL, &cgroup_full_path); |
2814 | if (r == 0) { | |
2815 | r = cg_path_get_cgroupid(cgroup_full_path, &cgroup_id); | |
2816 | if (r < 0) | |
1b420223 LP |
2817 | log_unit_full_errno(u, ERRNO_IS_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r, |
2818 | "Failed to get cgroup ID of cgroup %s, ignoring: %m", cgroup_full_path); | |
184b4f78 ILG |
2819 | } else |
2820 | log_unit_warning_errno(u, r, "Failed to get full cgroup path on cgroup %s, ignoring: %m", empty_to_root(u->cgroup_path)); | |
2821 | ||
2822 | u->cgroup_id = cgroup_id; | |
2823 | } | |
2824 | ||
efdb0237 LP |
2825 | /* Start watching it */ |
2826 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 2827 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 2828 | |
7b639614 MK |
2829 | /* For v2 we preserve enabled controllers in delegated units, adjust others, |
2830 | * for v1 we figure out which controller hierarchies need migration. */ | |
1fd3a10c | 2831 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 2832 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
2833 | |
2834 | /* Enable all controllers we need */ | |
27adcc97 | 2835 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 2836 | if (r < 0) |
6178e2f8 | 2837 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
27adcc97 | 2838 | |
27adcc97 LP |
2839 | /* Remember what's actually enabled now */ |
2840 | u->cgroup_enabled_mask = result_mask; | |
7b639614 MK |
2841 | |
2842 | migrate_mask = u->cgroup_realized_mask ^ target_mask; | |
65be7e06 | 2843 | } |
03b90d4b LP |
2844 | |
2845 | /* Keep track that this is now realized */ | |
4ad49000 | 2846 | u->cgroup_realized = true; |
efdb0237 | 2847 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 2848 | |
7b639614 MK |
2849 | /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling). |
2850 | * | |
2851 | * Unnecessary controller cgroups are trimmed (after emptied by upward migration). | |
2852 | * We perform migration also with whole slices for cases when users don't care about leave | |
2853 | * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing | |
2854 | * delegated units. | |
2855 | */ | |
2856 | if (cg_all_unified() == 0) { | |
2857 | r = cg_migrate_v1_controllers(u->manager->cgroup_supported, migrate_mask, u->cgroup_path, migrate_callback, u); | |
2858 | if (r < 0) | |
6178e2f8 | 2859 | log_unit_warning_errno(u, r, "Failed to migrate controller cgroups from %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
0cd385d3 | 2860 | |
7b639614 MK |
2861 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2862 | r = cg_trim_v1_controllers(u->manager->cgroup_supported, ~target_mask, u->cgroup_path, !is_root_slice); | |
0cd385d3 | 2863 | if (r < 0) |
6178e2f8 | 2864 | log_unit_warning_errno(u, r, "Failed to delete controller cgroups %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
0cd385d3 | 2865 | } |
03b90d4b | 2866 | |
0d2d6fbf CD |
2867 | /* Set attributes */ |
2868 | cgroup_context_apply(u, target_mask, state); | |
2869 | cgroup_xattr_apply(u); | |
2870 | ||
29e6b0c1 LP |
2871 | /* For most units we expect that memory monitoring is set up before the unit is started and we won't |
2872 | * touch it after. For PID 1 this is different though, because we couldn't possibly do that given | |
2873 | * that PID 1 runs before init.scope is even set up. Hence, whenever init.scope is realized, let's | |
2874 | * try to open the memory pressure interface anew. */ | |
2875 | if (unit_has_name(u, SPECIAL_INIT_SCOPE)) | |
2876 | (void) manager_setup_memory_pressure_event_source(u->manager); | |
2877 | ||
64747e2d LP |
2878 | return 0; |
2879 | } | |
2880 | ||
6592b975 LP |
2881 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
2882 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2883 | char *pp; | |
7b3fd631 | 2884 | int r; |
6592b975 | 2885 | |
7b3fd631 LP |
2886 | assert(u); |
2887 | ||
6592b975 LP |
2888 | if (MANAGER_IS_SYSTEM(u->manager)) |
2889 | return -EINVAL; | |
2890 | ||
2891 | if (!u->manager->system_bus) | |
2892 | return -EIO; | |
2893 | ||
2894 | if (!u->cgroup_path) | |
2895 | return -EINVAL; | |
2896 | ||
2897 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
2898 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
2899 | if (!pp) | |
2900 | return -EINVAL; | |
2901 | ||
2902 | pp = strjoina("/", pp, suffix_path); | |
4ff361cc | 2903 | path_simplify(pp); |
6592b975 | 2904 | |
78fa2f91 | 2905 | r = bus_call_method(u->manager->system_bus, |
2906 | bus_systemd_mgr, | |
2907 | "AttachProcessesToUnit", | |
2908 | &error, NULL, | |
2909 | "ssau", | |
2910 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 2911 | if (r < 0) |
6592b975 LP |
2912 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
2913 | ||
2914 | return 0; | |
2915 | } | |
2916 | ||
2917 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
8e7e4a73 | 2918 | _cleanup_free_ char *joined = NULL; |
6592b975 LP |
2919 | CGroupMask delegated_mask; |
2920 | const char *p; | |
495e75ed | 2921 | PidRef *pid; |
db4229d1 | 2922 | int ret, r; |
6592b975 LP |
2923 | |
2924 | assert(u); | |
2925 | ||
2926 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2927 | return -EINVAL; | |
2928 | ||
2929 | if (set_isempty(pids)) | |
2930 | return 0; | |
7b3fd631 | 2931 | |
fab34748 KL |
2932 | /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable. |
2933 | * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */ | |
2934 | r = bpf_firewall_load_custom(u); | |
2935 | if (r < 0) | |
2936 | return r; | |
2937 | ||
6592b975 | 2938 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
2939 | if (r < 0) |
2940 | return r; | |
2941 | ||
6592b975 LP |
2942 | if (isempty(suffix_path)) |
2943 | p = u->cgroup_path; | |
8e7e4a73 LP |
2944 | else { |
2945 | joined = path_join(u->cgroup_path, suffix_path); | |
2946 | if (!joined) | |
2947 | return -ENOMEM; | |
2948 | ||
2949 | p = joined; | |
2950 | } | |
6592b975 LP |
2951 | |
2952 | delegated_mask = unit_get_delegate_mask(u); | |
2953 | ||
db4229d1 | 2954 | ret = 0; |
495e75ed LP |
2955 | SET_FOREACH(pid, pids) { |
2956 | ||
2957 | /* Unfortunately we cannot add pids by pidfd to a cgroup. Hence we have to use PIDs instead, | |
2958 | * which of course is racy. Let's shorten the race a bit though, and re-validate the PID | |
2959 | * before we use it */ | |
2960 | r = pidref_verify(pid); | |
2961 | if (r < 0) { | |
2962 | log_unit_info_errno(u, r, "PID " PID_FMT " vanished before we could move it to target cgroup '%s', skipping: %m", pid->pid, empty_to_root(p)); | |
2963 | continue; | |
2964 | } | |
6592b975 LP |
2965 | |
2966 | /* First, attach the PID to the main cgroup hierarchy */ | |
495e75ed | 2967 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid->pid); |
db4229d1 LP |
2968 | if (r < 0) { |
2969 | bool again = MANAGER_IS_USER(u->manager) && ERRNO_IS_PRIVILEGE(r); | |
6592b975 | 2970 | |
db4229d1 | 2971 | log_unit_full_errno(u, again ? LOG_DEBUG : LOG_INFO, r, |
7a2ba407 | 2972 | "Couldn't move process "PID_FMT" to%s requested cgroup '%s': %m", |
495e75ed | 2973 | pid->pid, again ? " directly" : "", empty_to_root(p)); |
7a2ba407 ZJS |
2974 | |
2975 | if (again) { | |
6592b975 LP |
2976 | int z; |
2977 | ||
7a2ba407 ZJS |
2978 | /* If we are in a user instance, and we can't move the process ourselves due |
2979 | * to permission problems, let's ask the system instance about it instead. | |
2980 | * Since it's more privileged it might be able to move the process across the | |
2981 | * leaves of a subtree whose top node is not owned by us. */ | |
6592b975 | 2982 | |
495e75ed | 2983 | z = unit_attach_pid_to_cgroup_via_bus(u, pid->pid, suffix_path); |
6592b975 | 2984 | if (z < 0) |
495e75ed | 2985 | log_unit_info_errno(u, z, "Couldn't move process "PID_FMT" to requested cgroup '%s' (directly or via the system bus): %m", pid->pid, empty_to_root(p)); |
c65417a0 JW |
2986 | else { |
2987 | if (ret >= 0) | |
2988 | ret++; /* Count successful additions */ | |
6592b975 | 2989 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ |
c65417a0 | 2990 | } |
6592b975 LP |
2991 | } |
2992 | ||
db4229d1 LP |
2993 | if (ret >= 0) |
2994 | ret = r; /* Remember first error */ | |
6592b975 LP |
2995 | |
2996 | continue; | |
8d3e4ac7 LP |
2997 | } else if (ret >= 0) |
2998 | ret++; /* Count successful additions */ | |
6592b975 | 2999 | |
db4229d1 LP |
3000 | r = cg_all_unified(); |
3001 | if (r < 0) | |
3002 | return r; | |
3003 | if (r > 0) | |
6592b975 LP |
3004 | continue; |
3005 | ||
3006 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
3007 | * innermost realized one */ | |
3008 | ||
e8616626 | 3009 | for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
6592b975 LP |
3010 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); |
3011 | const char *realized; | |
3012 | ||
3013 | if (!(u->manager->cgroup_supported & bit)) | |
3014 | continue; | |
3015 | ||
3016 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
3017 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
495e75ed | 3018 | r = cg_attach(cgroup_controller_to_string(c), p, pid->pid); |
db4229d1 | 3019 | if (r >= 0) |
6592b975 LP |
3020 | continue; /* Success! */ |
3021 | ||
db4229d1 | 3022 | log_unit_debug_errno(u, r, "Failed to attach PID " PID_FMT " to requested cgroup %s in controller %s, falling back to unit's cgroup: %m", |
495e75ed | 3023 | pid->pid, empty_to_root(p), cgroup_controller_to_string(c)); |
6592b975 LP |
3024 | } |
3025 | ||
3026 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
3027 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
3028 | * realized. */ | |
3029 | realized = unit_get_realized_cgroup_path(u, bit); | |
3030 | if (!realized) | |
3031 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
3032 | ||
495e75ed | 3033 | r = cg_attach(cgroup_controller_to_string(c), realized, pid->pid); |
db4229d1 LP |
3034 | if (r < 0) |
3035 | log_unit_debug_errno(u, r, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
495e75ed | 3036 | pid->pid, realized, cgroup_controller_to_string(c)); |
6592b975 LP |
3037 | } |
3038 | } | |
3039 | ||
db4229d1 | 3040 | return ret; |
7b3fd631 LP |
3041 | } |
3042 | ||
906c06f6 DM |
3043 | static bool unit_has_mask_realized( |
3044 | Unit *u, | |
3045 | CGroupMask target_mask, | |
17f14955 | 3046 | CGroupMask enable_mask) { |
906c06f6 | 3047 | |
bc432dc7 LP |
3048 | assert(u); |
3049 | ||
d5095dcd LP |
3050 | /* Returns true if this unit is fully realized. We check four things: |
3051 | * | |
3052 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
3053 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
3054 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
3055 | * 4. Whether the invalidation mask is currently zero |
3056 | * | |
3057 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
3058 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
3059 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
3060 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
3061 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
3062 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
3063 | * simply don't matter. */ | |
3064 | ||
906c06f6 | 3065 | return u->cgroup_realized && |
d5095dcd LP |
3066 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
3067 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 3068 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
3069 | } |
3070 | ||
4f6f62e4 CD |
3071 | static bool unit_has_mask_disables_realized( |
3072 | Unit *u, | |
3073 | CGroupMask target_mask, | |
3074 | CGroupMask enable_mask) { | |
3075 | ||
3076 | assert(u); | |
3077 | ||
3078 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
3079 | * | |
3080 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
3081 | * already removed. */ | |
3082 | ||
3083 | return !u->cgroup_realized || | |
3084 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
3085 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
3086 | } | |
3087 | ||
a57669d2 CD |
3088 | static bool unit_has_mask_enables_realized( |
3089 | Unit *u, | |
3090 | CGroupMask target_mask, | |
3091 | CGroupMask enable_mask) { | |
3092 | ||
3093 | assert(u); | |
3094 | ||
3095 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
3096 | * | |
3097 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
3098 | * we want to add is already added. */ | |
3099 | ||
3100 | return u->cgroup_realized && | |
c72703e2 CD |
3101 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
3102 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
3103 | } |
3104 | ||
020b2e41 | 3105 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
3106 | assert(u); |
3107 | ||
3108 | if (u->in_cgroup_realize_queue) | |
3109 | return; | |
3110 | ||
a479c21e | 3111 | LIST_APPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); |
2aa57a65 LP |
3112 | u->in_cgroup_realize_queue = true; |
3113 | } | |
3114 | ||
3115 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
3116 | assert(u); | |
3117 | ||
3118 | if (!u->in_cgroup_realize_queue) | |
3119 | return; | |
3120 | ||
3121 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
3122 | u->in_cgroup_realize_queue = false; | |
3123 | } | |
3124 | ||
a57669d2 CD |
3125 | /* Controllers can only be enabled breadth-first, from the root of the |
3126 | * hierarchy downwards to the unit in question. */ | |
3127 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
3128 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
12f64221 | 3129 | Unit *slice; |
a57669d2 CD |
3130 | int r; |
3131 | ||
3132 | assert(u); | |
3133 | ||
3134 | /* First go deal with this unit's parent, or we won't be able to enable | |
3135 | * any new controllers at this layer. */ | |
12f64221 LP |
3136 | slice = UNIT_GET_SLICE(u); |
3137 | if (slice) { | |
3138 | r = unit_realize_cgroup_now_enable(slice, state); | |
a57669d2 CD |
3139 | if (r < 0) |
3140 | return r; | |
3141 | } | |
3142 | ||
3143 | target_mask = unit_get_target_mask(u); | |
3144 | enable_mask = unit_get_enable_mask(u); | |
3145 | ||
3146 | /* We can only enable in this direction, don't try to disable anything. | |
3147 | */ | |
3148 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
3149 | return 0; | |
3150 | ||
3151 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
3152 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
3153 | ||
7b639614 | 3154 | return unit_update_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
3155 | } |
3156 | ||
4f6f62e4 CD |
3157 | /* Controllers can only be disabled depth-first, from the leaves of the |
3158 | * hierarchy upwards to the unit in question. */ | |
3159 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
4f6f62e4 | 3160 | Unit *m; |
4f6f62e4 CD |
3161 | |
3162 | assert(u); | |
3163 | ||
3164 | if (u->type != UNIT_SLICE) | |
3165 | return 0; | |
3166 | ||
d219a2b0 | 3167 | UNIT_FOREACH_DEPENDENCY(m, u, UNIT_ATOM_SLICE_OF) { |
4f6f62e4 CD |
3168 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; |
3169 | int r; | |
3170 | ||
defe63b0 LP |
3171 | /* The cgroup for this unit might not actually be fully realised yet, in which case it isn't |
3172 | * holding any controllers open anyway. */ | |
d9ef5944 | 3173 | if (!m->cgroup_realized) |
4f6f62e4 CD |
3174 | continue; |
3175 | ||
defe63b0 | 3176 | /* We must disable those below us first in order to release the controller. */ |
4f6f62e4 CD |
3177 | if (m->type == UNIT_SLICE) |
3178 | (void) unit_realize_cgroup_now_disable(m, state); | |
3179 | ||
3180 | target_mask = unit_get_target_mask(m); | |
3181 | enable_mask = unit_get_enable_mask(m); | |
3182 | ||
defe63b0 | 3183 | /* We can only disable in this direction, don't try to enable anything. */ |
4f6f62e4 CD |
3184 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) |
3185 | continue; | |
3186 | ||
3187 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
3188 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
3189 | ||
7b639614 | 3190 | r = unit_update_cgroup(m, new_target_mask, new_enable_mask, state); |
4f6f62e4 CD |
3191 | if (r < 0) |
3192 | return r; | |
3193 | } | |
3194 | ||
3195 | return 0; | |
3196 | } | |
a57669d2 | 3197 | |
6414b7c9 DS |
3198 | /* Check if necessary controllers and attributes for a unit are in place. |
3199 | * | |
a57669d2 CD |
3200 | * - If so, do nothing. |
3201 | * - If not, create paths, move processes over, and set attributes. | |
3202 | * | |
3203 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
3204 | * a depth-first way. As such the process looks like this: | |
3205 | * | |
3206 | * Suppose we have a cgroup hierarchy which looks like this: | |
3207 | * | |
3208 | * root | |
3209 | * / \ | |
3210 | * / \ | |
3211 | * / \ | |
3212 | * a b | |
3213 | * / \ / \ | |
3214 | * / \ / \ | |
3215 | * c d e f | |
3216 | * / \ / \ / \ / \ | |
3217 | * h i j k l m n o | |
3218 | * | |
3219 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 3220 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
3221 | * 3. cgroup "k" just started requesting the memory controller. |
3222 | * | |
3223 | * To make this work we must do the following in order: | |
3224 | * | |
3225 | * 1. Disable CPU controller in k, j | |
3226 | * 2. Disable CPU controller in d | |
3227 | * 3. Enable memory controller in root | |
3228 | * 4. Enable memory controller in a | |
3229 | * 5. Enable memory controller in d | |
3230 | * 6. Enable memory controller in k | |
3231 | * | |
3232 | * Notice that we need to touch j in one direction, but not the other. We also | |
3233 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
3234 | * wants to disable further. The basic rules are therefore: | |
3235 | * | |
3236 | * - If you're disabling something, you need to realise all of the cgroups from | |
3237 | * your recursive descendants to the root. This starts from the leaves. | |
3238 | * - If you're enabling something, you need to realise from the root cgroup | |
3239 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
3240 | * |
3241 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 3242 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 3243 | CGroupMask target_mask, enable_mask; |
12f64221 | 3244 | Unit *slice; |
6414b7c9 | 3245 | int r; |
64747e2d | 3246 | |
4ad49000 | 3247 | assert(u); |
64747e2d | 3248 | |
2aa57a65 | 3249 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 3250 | |
efdb0237 | 3251 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
3252 | enable_mask = unit_get_enable_mask(u); |
3253 | ||
17f14955 | 3254 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 3255 | return 0; |
64747e2d | 3256 | |
4f6f62e4 CD |
3257 | /* Disable controllers below us, if there are any */ |
3258 | r = unit_realize_cgroup_now_disable(u, state); | |
3259 | if (r < 0) | |
3260 | return r; | |
3261 | ||
3262 | /* Enable controllers above us, if there are any */ | |
12f64221 LP |
3263 | slice = UNIT_GET_SLICE(u); |
3264 | if (slice) { | |
3265 | r = unit_realize_cgroup_now_enable(slice, state); | |
6414b7c9 DS |
3266 | if (r < 0) |
3267 | return r; | |
3268 | } | |
4ad49000 | 3269 | |
0d2d6fbf | 3270 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
7b639614 | 3271 | r = unit_update_cgroup(u, target_mask, enable_mask, state); |
6414b7c9 DS |
3272 | if (r < 0) |
3273 | return r; | |
3274 | ||
c2baf11c LP |
3275 | /* Now, reset the invalidation mask */ |
3276 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 3277 | return 0; |
64747e2d LP |
3278 | } |
3279 | ||
91a6073e | 3280 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 3281 | ManagerState state; |
4ad49000 | 3282 | unsigned n = 0; |
db785129 | 3283 | Unit *i; |
6414b7c9 | 3284 | int r; |
ecedd90f | 3285 | |
91a6073e LP |
3286 | assert(m); |
3287 | ||
db785129 LP |
3288 | state = manager_state(m); |
3289 | ||
91a6073e LP |
3290 | while ((i = m->cgroup_realize_queue)) { |
3291 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 3292 | |
2aa57a65 LP |
3293 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
3294 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
3295 | unit_remove_from_cgroup_realize_queue(i); | |
3296 | continue; | |
3297 | } | |
3298 | ||
db785129 | 3299 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 3300 | if (r < 0) |
efdb0237 | 3301 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 3302 | |
4ad49000 LP |
3303 | n++; |
3304 | } | |
ecedd90f | 3305 | |
4ad49000 | 3306 | return n; |
8e274523 LP |
3307 | } |
3308 | ||
4c591f39 MK |
3309 | void unit_add_family_to_cgroup_realize_queue(Unit *u) { |
3310 | assert(u); | |
3311 | assert(u->type == UNIT_SLICE); | |
ca949c9d | 3312 | |
4c591f39 MK |
3313 | /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all |
3314 | * its ancestors. | |
3315 | * | |
3316 | * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes | |
3317 | * very weird if two units that own processes reside in the same slice, but one is realized in the | |
3318 | * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does | |
3319 | * not), because that means individual processes need to be scheduled against whole cgroups. Let's | |
3320 | * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1 | |
3321 | * controller hierarchies too (if unit requires the controller to be realized). | |
e1e98911 | 3322 | * |
4c591f39 MK |
3323 | * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date |
3324 | * masks. */ | |
3325 | ||
3326 | do { | |
4ad49000 | 3327 | Unit *m; |
8f53a7b8 | 3328 | |
4c591f39 MK |
3329 | /* Children of u likely changed when we're called */ |
3330 | u->cgroup_members_mask_valid = false; | |
f23ba94d | 3331 | |
d219a2b0 | 3332 | UNIT_FOREACH_DEPENDENCY(m, u, UNIT_ATOM_SLICE_OF) { |
8e274523 | 3333 | |
65f6b6bd | 3334 | /* No point in doing cgroup application for units without active processes. */ |
6414b7c9 DS |
3335 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) |
3336 | continue; | |
3337 | ||
e1e98911 LP |
3338 | /* We only enqueue siblings if they were realized once at least, in the main |
3339 | * hierarchy. */ | |
3340 | if (!m->cgroup_realized) | |
3341 | continue; | |
3342 | ||
defe63b0 LP |
3343 | /* If the unit doesn't need any new controllers and has current ones |
3344 | * realized, it doesn't need any changes. */ | |
906c06f6 DM |
3345 | if (unit_has_mask_realized(m, |
3346 | unit_get_target_mask(m), | |
17f14955 | 3347 | unit_get_enable_mask(m))) |
6414b7c9 DS |
3348 | continue; |
3349 | ||
91a6073e | 3350 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
3351 | } |
3352 | ||
4c591f39 MK |
3353 | /* Parent comes after children */ |
3354 | unit_add_to_cgroup_realize_queue(u); | |
12f64221 LP |
3355 | |
3356 | u = UNIT_GET_SLICE(u); | |
3357 | } while (u); | |
4ad49000 LP |
3358 | } |
3359 | ||
0a1eb06d | 3360 | int unit_realize_cgroup(Unit *u) { |
12f64221 LP |
3361 | Unit *slice; |
3362 | ||
4ad49000 LP |
3363 | assert(u); |
3364 | ||
35b7ff80 | 3365 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 3366 | return 0; |
8e274523 | 3367 | |
4c591f39 MK |
3368 | /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all |
3369 | * parents, but there's more actually: for the weight-based controllers we also need to make sure | |
3370 | * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the | |
3371 | * other hand, when a controller is removed from realized set, it may become unnecessary in siblings | |
3372 | * and ancestors and they should be (de)realized too. | |
3373 | * | |
3374 | * This call will defer work on the siblings and derealized ancestors to the next event loop | |
3375 | * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */ | |
ca949c9d | 3376 | |
12f64221 LP |
3377 | slice = UNIT_GET_SLICE(u); |
3378 | if (slice) | |
3379 | unit_add_family_to_cgroup_realize_queue(slice); | |
4ad49000 | 3380 | |
6414b7c9 | 3381 | /* And realize this one now (and apply the values) */ |
db785129 | 3382 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
3383 | } |
3384 | ||
efdb0237 LP |
3385 | void unit_release_cgroup(Unit *u) { |
3386 | assert(u); | |
3387 | ||
8a0d5388 LP |
3388 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
3389 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
3390 | |
3391 | if (u->cgroup_path) { | |
3392 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
3393 | u->cgroup_path = mfree(u->cgroup_path); | |
3394 | } | |
3395 | ||
0bb814c2 LP |
3396 | if (u->cgroup_control_inotify_wd >= 0) { |
3397 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_control_inotify_wd) < 0) | |
3398 | 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); | |
efdb0237 | 3399 | |
0bb814c2 LP |
3400 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd)); |
3401 | u->cgroup_control_inotify_wd = -1; | |
efdb0237 | 3402 | } |
afcfaa69 LP |
3403 | |
3404 | if (u->cgroup_memory_inotify_wd >= 0) { | |
3405 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_memory_inotify_wd) < 0) | |
3406 | 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); | |
3407 | ||
3408 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd)); | |
3409 | u->cgroup_memory_inotify_wd = -1; | |
3410 | } | |
efdb0237 LP |
3411 | } |
3412 | ||
e08dabfe AZ |
3413 | bool unit_maybe_release_cgroup(Unit *u) { |
3414 | int r; | |
3415 | ||
3416 | assert(u); | |
3417 | ||
3418 | if (!u->cgroup_path) | |
3419 | return true; | |
3420 | ||
3421 | /* Don't release the cgroup if there are still processes under it. If we get notified later when all the | |
3422 | * processes exit (e.g. the processes were in D-state and exited after the unit was marked as failed) | |
3423 | * we need the cgroup paths to continue to be tracked by the manager so they can be looked up and cleaned | |
3424 | * up later. */ | |
3425 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
3426 | if (r < 0) | |
3427 | log_unit_debug_errno(u, r, "Error checking if the cgroup is recursively empty, ignoring: %m"); | |
3428 | else if (r == 1) { | |
3429 | unit_release_cgroup(u); | |
3430 | return true; | |
3431 | } | |
3432 | ||
3433 | return false; | |
3434 | } | |
3435 | ||
efdb0237 | 3436 | void unit_prune_cgroup(Unit *u) { |
8e274523 | 3437 | int r; |
efdb0237 | 3438 | bool is_root_slice; |
8e274523 | 3439 | |
4ad49000 | 3440 | assert(u); |
8e274523 | 3441 | |
efdb0237 LP |
3442 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
3443 | ||
4ad49000 LP |
3444 | if (!u->cgroup_path) |
3445 | return; | |
8e274523 | 3446 | |
ad009380 MY |
3447 | /* Cache the last CPU and memory usage values before we destroy the cgroup */ |
3448 | (void) unit_get_cpu_usage(u, /* ret = */ NULL); | |
3449 | ||
3450 | for (CGroupMemoryAccountingMetric metric = 0; metric <= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST; metric++) | |
3451 | (void) unit_get_memory_accounting(u, metric, /* ret = */ NULL); | |
fe700f46 | 3452 | |
b1994387 ILG |
3453 | #if BPF_FRAMEWORK |
3454 | (void) lsm_bpf_cleanup(u); /* Remove cgroup from the global LSM BPF map */ | |
3455 | #endif | |
3456 | ||
49b6babb | 3457 | unit_modify_nft_set(u, /* add = */ false); |
dc7d69b3 | 3458 | |
efdb0237 LP |
3459 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
3460 | ||
3461 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
0219b352 DB |
3462 | if (r < 0) |
3463 | /* One reason we could have failed here is, that the cgroup still contains a process. | |
3464 | * However, if the cgroup becomes removable at a later time, it might be removed when | |
3465 | * the containing slice is stopped. So even if we failed now, this unit shouldn't assume | |
3466 | * that the cgroup is still realized the next time it is started. Do not return early | |
3467 | * on error, continue cleanup. */ | |
6178e2f8 | 3468 | log_unit_full_errno(u, r == -EBUSY ? LOG_DEBUG : LOG_WARNING, r, "Failed to destroy cgroup %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
8e274523 | 3469 | |
efdb0237 LP |
3470 | if (is_root_slice) |
3471 | return; | |
3472 | ||
e08dabfe AZ |
3473 | if (!unit_maybe_release_cgroup(u)) /* Returns true if the cgroup was released */ |
3474 | return; | |
0a1eb06d | 3475 | |
4ad49000 | 3476 | u->cgroup_realized = false; |
bc432dc7 | 3477 | u->cgroup_realized_mask = 0; |
ccf78df1 | 3478 | u->cgroup_enabled_mask = 0; |
084c7007 | 3479 | |
76dc1725 | 3480 | u->bpf_device_control_installed = bpf_program_free(u->bpf_device_control_installed); |
8e274523 LP |
3481 | } |
3482 | ||
495e75ed LP |
3483 | int unit_search_main_pid(Unit *u, PidRef *ret) { |
3484 | _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL; | |
4ad49000 | 3485 | _cleanup_fclose_ FILE *f = NULL; |
efdb0237 | 3486 | int r; |
4ad49000 LP |
3487 | |
3488 | assert(u); | |
efdb0237 | 3489 | assert(ret); |
4ad49000 LP |
3490 | |
3491 | if (!u->cgroup_path) | |
efdb0237 | 3492 | return -ENXIO; |
4ad49000 | 3493 | |
efdb0237 LP |
3494 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
3495 | if (r < 0) | |
3496 | return r; | |
4ad49000 | 3497 | |
495e75ed LP |
3498 | for (;;) { |
3499 | _cleanup_(pidref_done) PidRef npidref = PIDREF_NULL; | |
4ad49000 | 3500 | |
495e75ed LP |
3501 | r = cg_read_pidref(f, &npidref); |
3502 | if (r < 0) | |
3503 | return r; | |
3504 | if (r == 0) | |
3505 | break; | |
8e274523 | 3506 | |
495e75ed | 3507 | if (pidref_equal(&pidref, &npidref)) /* seen already, cgroupfs reports duplicates! */ |
4ad49000 | 3508 | continue; |
8e274523 | 3509 | |
6774be42 | 3510 | if (pidref_is_my_child(&npidref) <= 0) /* ignore processes further down the tree */ |
495e75ed | 3511 | continue; |
efdb0237 | 3512 | |
495e75ed LP |
3513 | if (pidref_is_set(&pidref) != 0) |
3514 | /* Dang, there's more than one daemonized PID in this group, so we don't know what | |
3515 | * process is the main process. */ | |
efdb0237 | 3516 | return -ENODATA; |
8e274523 | 3517 | |
495e75ed | 3518 | pidref = TAKE_PIDREF(npidref); |
8e274523 LP |
3519 | } |
3520 | ||
495e75ed LP |
3521 | if (!pidref_is_set(&pidref)) |
3522 | return -ENODATA; | |
3523 | ||
3524 | *ret = TAKE_PIDREF(pidref); | |
efdb0237 LP |
3525 | return 0; |
3526 | } | |
3527 | ||
3528 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 3529 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
3530 | _cleanup_fclose_ FILE *f = NULL; |
3531 | int ret = 0, r; | |
3532 | ||
3533 | assert(u); | |
3534 | assert(path); | |
3535 | ||
3536 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
3537 | if (r < 0) | |
495e75ed | 3538 | RET_GATHER(ret, r); |
efdb0237 | 3539 | else { |
495e75ed LP |
3540 | for (;;) { |
3541 | _cleanup_(pidref_done) PidRef pid = PIDREF_NULL; | |
3542 | ||
3543 | r = cg_read_pidref(f, &pid); | |
3544 | if (r == 0) | |
3545 | break; | |
3546 | if (r < 0) { | |
3547 | RET_GATHER(ret, r); | |
3548 | break; | |
3549 | } | |
efdb0237 | 3550 | |
495e75ed | 3551 | RET_GATHER(ret, unit_watch_pidref(u, &pid, /* exclusive= */ false)); |
efdb0237 | 3552 | } |
efdb0237 LP |
3553 | } |
3554 | ||
3555 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
495e75ed LP |
3556 | if (r < 0) |
3557 | RET_GATHER(ret, r); | |
3558 | else { | |
3559 | for (;;) { | |
3560 | _cleanup_free_ char *fn = NULL, *p = NULL; | |
3561 | ||
3562 | r = cg_read_subgroup(d, &fn); | |
3563 | if (r == 0) | |
3564 | break; | |
3565 | if (r < 0) { | |
3566 | RET_GATHER(ret, r); | |
3567 | break; | |
3568 | } | |
efdb0237 | 3569 | |
95b21cff | 3570 | p = path_join(empty_to_root(path), fn); |
efdb0237 LP |
3571 | if (!p) |
3572 | return -ENOMEM; | |
3573 | ||
495e75ed | 3574 | RET_GATHER(ret, unit_watch_pids_in_path(u, p)); |
efdb0237 | 3575 | } |
efdb0237 LP |
3576 | } |
3577 | ||
3578 | return ret; | |
3579 | } | |
3580 | ||
11aef522 LP |
3581 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
3582 | int r; | |
3583 | ||
3584 | assert(u); | |
3585 | ||
3586 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
3587 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
3588 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
3589 | ||
3590 | if (!u->cgroup_path) | |
3591 | return -ENOENT; | |
3592 | ||
3593 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
3594 | if (r < 0) | |
3595 | return r; | |
3596 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
3597 | return 0; | |
3598 | ||
3599 | if (!set_isempty(u->pids)) | |
3600 | return 0; | |
3601 | ||
3602 | unit_add_to_cgroup_empty_queue(u); | |
3603 | return 0; | |
3604 | } | |
3605 | ||
efdb0237 | 3606 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
3607 | int r; |
3608 | ||
efdb0237 LP |
3609 | assert(u); |
3610 | ||
3611 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
3612 | * watch. This is a fallback logic for cases where we do not | |
3613 | * get reliable cgroup empty notifications: we try to use | |
3614 | * SIGCHLD as replacement. */ | |
3615 | ||
3616 | if (!u->cgroup_path) | |
3617 | return -ENOENT; | |
3618 | ||
c22800e4 | 3619 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
3620 | if (r < 0) |
3621 | return r; | |
3622 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
3623 | return 0; |
3624 | ||
3625 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
3626 | } | |
3627 | ||
09e24654 | 3628 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
99534007 | 3629 | Manager *m = ASSERT_PTR(userdata); |
09e24654 | 3630 | Unit *u; |
efdb0237 LP |
3631 | int r; |
3632 | ||
09e24654 | 3633 | assert(s); |
efdb0237 | 3634 | |
09e24654 LP |
3635 | u = m->cgroup_empty_queue; |
3636 | if (!u) | |
efdb0237 LP |
3637 | return 0; |
3638 | ||
09e24654 LP |
3639 | assert(u->in_cgroup_empty_queue); |
3640 | u->in_cgroup_empty_queue = false; | |
3641 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
3642 | ||
3643 | if (m->cgroup_empty_queue) { | |
3644 | /* More stuff queued, let's make sure we remain enabled */ | |
3645 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
3646 | if (r < 0) | |
19a691a9 | 3647 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 3648 | } |
efdb0237 | 3649 | |
f7829525 NK |
3650 | /* Update state based on OOM kills before we notify about cgroup empty event */ |
3651 | (void) unit_check_oom(u); | |
3652 | (void) unit_check_oomd_kill(u); | |
3653 | ||
efdb0237 LP |
3654 | unit_add_to_gc_queue(u); |
3655 | ||
380dd177 RP |
3656 | if (IN_SET(unit_active_state(u), UNIT_INACTIVE, UNIT_FAILED)) |
3657 | unit_prune_cgroup(u); | |
3658 | else if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
efdb0237 LP |
3659 | UNIT_VTABLE(u)->notify_cgroup_empty(u); |
3660 | ||
3661 | return 0; | |
3662 | } | |
3663 | ||
09e24654 LP |
3664 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
3665 | int r; | |
3666 | ||
3667 | assert(u); | |
3668 | ||
3669 | /* Note that there are four different ways how cgroup empty events reach us: | |
3670 | * | |
3671 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
3672 | * | |
3673 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
3674 | * | |
3675 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
3676 | * | |
3677 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
3678 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
3679 | * | |
3680 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
3681 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
3682 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
3683 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
3684 | * case for scope units). */ | |
3685 | ||
3686 | if (u->in_cgroup_empty_queue) | |
3687 | return; | |
3688 | ||
3689 | /* Let's verify that the cgroup is really empty */ | |
3690 | if (!u->cgroup_path) | |
3691 | return; | |
e1e98911 | 3692 | |
09e24654 LP |
3693 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); |
3694 | if (r < 0) { | |
6178e2f8 | 3695 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", empty_to_root(u->cgroup_path)); |
09e24654 LP |
3696 | return; |
3697 | } | |
3698 | if (r == 0) | |
3699 | return; | |
3700 | ||
3701 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
3702 | u->in_cgroup_empty_queue = true; | |
3703 | ||
3704 | /* Trigger the defer event */ | |
3705 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
3706 | if (r < 0) | |
3707 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
3708 | } | |
3709 | ||
d9e45bc3 MS |
3710 | static void unit_remove_from_cgroup_empty_queue(Unit *u) { |
3711 | assert(u); | |
3712 | ||
3713 | if (!u->in_cgroup_empty_queue) | |
3714 | return; | |
3715 | ||
3716 | LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
3717 | u->in_cgroup_empty_queue = false; | |
3718 | } | |
3719 | ||
fe8d22fb AZ |
3720 | int unit_check_oomd_kill(Unit *u) { |
3721 | _cleanup_free_ char *value = NULL; | |
3722 | bool increased; | |
3723 | uint64_t n = 0; | |
3724 | int r; | |
3725 | ||
3726 | if (!u->cgroup_path) | |
3727 | return 0; | |
3728 | ||
3729 | r = cg_all_unified(); | |
3730 | if (r < 0) | |
3731 | return log_unit_debug_errno(u, r, "Couldn't determine whether we are in all unified mode: %m"); | |
3732 | else if (r == 0) | |
3733 | return 0; | |
3734 | ||
bd1791b5 | 3735 | r = cg_get_xattr_malloc(u->cgroup_path, "user.oomd_ooms", &value); |
00675c36 | 3736 | if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r)) |
fe8d22fb AZ |
3737 | return r; |
3738 | ||
3739 | if (!isempty(value)) { | |
3740 | r = safe_atou64(value, &n); | |
3741 | if (r < 0) | |
3742 | return r; | |
3743 | } | |
3744 | ||
3745 | increased = n > u->managed_oom_kill_last; | |
3746 | u->managed_oom_kill_last = n; | |
3747 | ||
3748 | if (!increased) | |
3749 | return 0; | |
3750 | ||
38c41427 NK |
3751 | n = 0; |
3752 | value = mfree(value); | |
bd1791b5 | 3753 | r = cg_get_xattr_malloc(u->cgroup_path, "user.oomd_kill", &value); |
38c41427 NK |
3754 | if (r >= 0 && !isempty(value)) |
3755 | (void) safe_atou64(value, &n); | |
3756 | ||
fe8d22fb | 3757 | if (n > 0) |
c2503e35 RH |
3758 | log_unit_struct(u, LOG_NOTICE, |
3759 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
3760 | LOG_UNIT_INVOCATION_ID(u), | |
38c41427 NK |
3761 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed %"PRIu64" process(es) in this unit.", n), |
3762 | "N_PROCESSES=%" PRIu64, n); | |
3763 | else | |
3764 | log_unit_struct(u, LOG_NOTICE, | |
3765 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
3766 | LOG_UNIT_INVOCATION_ID(u), | |
3767 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed some process(es) in this unit.")); | |
3768 | ||
3769 | unit_notify_cgroup_oom(u, /* ManagedOOM= */ true); | |
fe8d22fb AZ |
3770 | |
3771 | return 1; | |
3772 | } | |
3773 | ||
2ba6ae6b | 3774 | int unit_check_oom(Unit *u) { |
afcfaa69 LP |
3775 | _cleanup_free_ char *oom_kill = NULL; |
3776 | bool increased; | |
3777 | uint64_t c; | |
3778 | int r; | |
3779 | ||
3780 | if (!u->cgroup_path) | |
3781 | return 0; | |
3782 | ||
3783 | r = cg_get_keyed_attribute("memory", u->cgroup_path, "memory.events", STRV_MAKE("oom_kill"), &oom_kill); | |
fc594dee LP |
3784 | if (IN_SET(r, -ENOENT, -ENXIO)) /* Handle gracefully if cgroup or oom_kill attribute don't exist */ |
3785 | c = 0; | |
3786 | else if (r < 0) | |
afcfaa69 | 3787 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); |
fc594dee LP |
3788 | else { |
3789 | r = safe_atou64(oom_kill, &c); | |
3790 | if (r < 0) | |
3791 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
3792 | } | |
afcfaa69 LP |
3793 | |
3794 | increased = c > u->oom_kill_last; | |
3795 | u->oom_kill_last = c; | |
3796 | ||
3797 | if (!increased) | |
3798 | return 0; | |
3799 | ||
c2503e35 RH |
3800 | log_unit_struct(u, LOG_NOTICE, |
3801 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
3802 | LOG_UNIT_INVOCATION_ID(u), | |
3803 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
afcfaa69 | 3804 | |
38c41427 | 3805 | unit_notify_cgroup_oom(u, /* ManagedOOM= */ false); |
afcfaa69 LP |
3806 | |
3807 | return 1; | |
3808 | } | |
3809 | ||
3810 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
99534007 | 3811 | Manager *m = ASSERT_PTR(userdata); |
afcfaa69 LP |
3812 | Unit *u; |
3813 | int r; | |
3814 | ||
3815 | assert(s); | |
afcfaa69 LP |
3816 | |
3817 | u = m->cgroup_oom_queue; | |
3818 | if (!u) | |
3819 | return 0; | |
3820 | ||
3821 | assert(u->in_cgroup_oom_queue); | |
3822 | u->in_cgroup_oom_queue = false; | |
3823 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
3824 | ||
3825 | if (m->cgroup_oom_queue) { | |
3826 | /* More stuff queued, let's make sure we remain enabled */ | |
3827 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
3828 | if (r < 0) | |
3829 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
3830 | } | |
3831 | ||
3832 | (void) unit_check_oom(u); | |
935f8042 LP |
3833 | unit_add_to_gc_queue(u); |
3834 | ||
afcfaa69 LP |
3835 | return 0; |
3836 | } | |
3837 | ||
3838 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
3839 | int r; | |
3840 | ||
3841 | assert(u); | |
3842 | ||
3843 | if (u->in_cgroup_oom_queue) | |
3844 | return; | |
3845 | if (!u->cgroup_path) | |
3846 | return; | |
3847 | ||
3848 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
3849 | u->in_cgroup_oom_queue = true; | |
3850 | ||
3851 | /* Trigger the defer event */ | |
3852 | if (!u->manager->cgroup_oom_event_source) { | |
3853 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
3854 | ||
3855 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
3856 | if (r < 0) { | |
3857 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
3858 | return; | |
3859 | } | |
3860 | ||
d42b61d2 | 3861 | r = sd_event_source_set_priority(s, EVENT_PRIORITY_CGROUP_OOM); |
afcfaa69 LP |
3862 | if (r < 0) { |
3863 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
3864 | return; | |
3865 | } | |
3866 | ||
3867 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
3868 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
3869 | } | |
3870 | ||
3871 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
3872 | if (r < 0) | |
3873 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
3874 | } | |
3875 | ||
d9e45bc3 MS |
3876 | static int unit_check_cgroup_events(Unit *u) { |
3877 | char *values[2] = {}; | |
3878 | int r; | |
3879 | ||
3880 | assert(u); | |
3881 | ||
869f52f2 DS |
3882 | if (!u->cgroup_path) |
3883 | return 0; | |
3884 | ||
d9e45bc3 MS |
3885 | r = cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", |
3886 | STRV_MAKE("populated", "frozen"), values); | |
3887 | if (r < 0) | |
3888 | return r; | |
3889 | ||
3890 | /* The cgroup.events notifications can be merged together so act as we saw the given state for the | |
3891 | * first time. The functions we call to handle given state are idempotent, which makes them | |
3892 | * effectively remember the previous state. */ | |
3893 | if (values[0]) { | |
3894 | if (streq(values[0], "1")) | |
3895 | unit_remove_from_cgroup_empty_queue(u); | |
3896 | else | |
3897 | unit_add_to_cgroup_empty_queue(u); | |
3898 | } | |
3899 | ||
3900 | /* Disregard freezer state changes due to operations not initiated by us */ | |
3901 | if (values[1] && IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_THAWING)) { | |
3902 | if (streq(values[1], "0")) | |
3903 | unit_thawed(u); | |
3904 | else | |
3905 | unit_frozen(u); | |
3906 | } | |
3907 | ||
3908 | free(values[0]); | |
3909 | free(values[1]); | |
3910 | ||
3911 | return 0; | |
3912 | } | |
3913 | ||
efdb0237 | 3914 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
99534007 | 3915 | Manager *m = ASSERT_PTR(userdata); |
efdb0237 LP |
3916 | |
3917 | assert(s); | |
3918 | assert(fd >= 0); | |
efdb0237 LP |
3919 | |
3920 | for (;;) { | |
3921 | union inotify_event_buffer buffer; | |
efdb0237 LP |
3922 | ssize_t l; |
3923 | ||
3924 | l = read(fd, &buffer, sizeof(buffer)); | |
3925 | if (l < 0) { | |
8add30a0 | 3926 | if (ERRNO_IS_TRANSIENT(errno)) |
efdb0237 LP |
3927 | return 0; |
3928 | ||
3929 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
3930 | } | |
3931 | ||
00adc340 | 3932 | FOREACH_INOTIFY_EVENT_WARN(e, buffer, l) { |
efdb0237 LP |
3933 | Unit *u; |
3934 | ||
3935 | if (e->wd < 0) | |
3936 | /* Queue overflow has no watch descriptor */ | |
3937 | continue; | |
3938 | ||
3939 | if (e->mask & IN_IGNORED) | |
3940 | /* The watch was just removed */ | |
3941 | continue; | |
3942 | ||
afcfaa69 LP |
3943 | /* Note that inotify might deliver events for a watch even after it was removed, |
3944 | * because it was queued before the removal. Let's ignore this here safely. */ | |
3945 | ||
0bb814c2 | 3946 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 | 3947 | if (u) |
d9e45bc3 | 3948 | unit_check_cgroup_events(u); |
efdb0237 | 3949 | |
afcfaa69 LP |
3950 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
3951 | if (u) | |
3952 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
3953 | } |
3954 | } | |
8e274523 LP |
3955 | } |
3956 | ||
17f14955 RG |
3957 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
3958 | CGroupMask mask = 0; | |
3959 | int r; | |
3960 | ||
3961 | /* BPF-based firewall */ | |
3962 | r = bpf_firewall_supported(); | |
ad13559e YW |
3963 | if (r < 0) |
3964 | return r; | |
17f14955 RG |
3965 | if (r > 0) |
3966 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
3967 | ||
084c7007 RG |
3968 | /* BPF-based device access control */ |
3969 | r = bpf_devices_supported(); | |
ad13559e YW |
3970 | if (r < 0) |
3971 | return r; | |
084c7007 RG |
3972 | if (r > 0) |
3973 | mask |= CGROUP_MASK_BPF_DEVICES; | |
3974 | ||
506ea51b JK |
3975 | /* BPF pinned prog */ |
3976 | r = bpf_foreign_supported(); | |
ad13559e YW |
3977 | if (r < 0) |
3978 | return r; | |
506ea51b JK |
3979 | if (r > 0) |
3980 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
3981 | ||
a8e5eb17 | 3982 | /* BPF-based bind{4|6} hooks */ |
cd09a5f3 | 3983 | r = bpf_socket_bind_supported(); |
ad13559e YW |
3984 | if (r < 0) |
3985 | return r; | |
a8e5eb17 JK |
3986 | if (r > 0) |
3987 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
3988 | ||
6f50d4f7 MV |
3989 | /* BPF-based cgroup_skb/{egress|ingress} hooks */ |
3990 | r = restrict_network_interfaces_supported(); | |
ad13559e YW |
3991 | if (r < 0) |
3992 | return r; | |
6f50d4f7 MV |
3993 | if (r > 0) |
3994 | mask |= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES; | |
3995 | ||
17f14955 RG |
3996 | *ret = mask; |
3997 | return 0; | |
3998 | } | |
3999 | ||
8e274523 | 4000 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 4001 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 4002 | const char *scope_path; |
b4cccbc1 | 4003 | int r, all_unified; |
17f14955 | 4004 | CGroupMask mask; |
efdb0237 | 4005 | char *e; |
8e274523 LP |
4006 | |
4007 | assert(m); | |
4008 | ||
35d2e7ec | 4009 | /* 1. Determine hierarchy */ |
efdb0237 | 4010 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 4011 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
4012 | if (r < 0) |
4013 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 4014 | |
efdb0237 LP |
4015 | /* Chop off the init scope, if we are already located in it */ |
4016 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 4017 | |
efdb0237 LP |
4018 | /* LEGACY: Also chop off the system slice if we are in |
4019 | * it. This is to support live upgrades from older systemd | |
4020 | * versions where PID 1 was moved there. Also see | |
4021 | * cg_get_root_path(). */ | |
463d0d15 | 4022 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 4023 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 4024 | if (!e) |
efdb0237 | 4025 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 4026 | } |
efdb0237 LP |
4027 | if (e) |
4028 | *e = 0; | |
7ccfb64a | 4029 | |
7546145e LP |
4030 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
4031 | * easily prepend it everywhere. */ | |
4032 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 4033 | |
35d2e7ec | 4034 | /* 2. Show data */ |
9444b1f2 | 4035 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
4036 | if (r < 0) |
4037 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 4038 | |
d4d99bc6 | 4039 | r = cg_unified(); |
415fc41c TH |
4040 | if (r < 0) |
4041 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 4042 | |
b4cccbc1 | 4043 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
4044 | if (all_unified < 0) |
4045 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
4046 | if (all_unified > 0) | |
efdb0237 | 4047 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 4048 | else { |
c22800e4 | 4049 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
4050 | if (r < 0) |
4051 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
4052 | if (r > 0) | |
4053 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
4054 | else | |
4055 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
4056 | } | |
efdb0237 | 4057 | |
09e24654 | 4058 | /* 3. Allocate cgroup empty defer event source */ |
5dcadb4c | 4059 | m->cgroup_empty_event_source = sd_event_source_disable_unref(m->cgroup_empty_event_source); |
09e24654 LP |
4060 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); |
4061 | if (r < 0) | |
4062 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
4063 | ||
cbe83389 LP |
4064 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
4065 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
4066 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
d42b61d2 | 4067 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, EVENT_PRIORITY_CGROUP_EMPTY); |
09e24654 LP |
4068 | if (r < 0) |
4069 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
4070 | ||
4071 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
4072 | if (r < 0) | |
4073 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
4074 | ||
4075 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
4076 | ||
4077 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 4078 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 4079 | |
09e24654 | 4080 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 4081 | |
5dcadb4c | 4082 | m->cgroup_inotify_event_source = sd_event_source_disable_unref(m->cgroup_inotify_event_source); |
10bd3e2e | 4083 | safe_close(m->cgroup_inotify_fd); |
efdb0237 | 4084 | |
10bd3e2e LP |
4085 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
4086 | if (m->cgroup_inotify_fd < 0) | |
4087 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 4088 | |
10bd3e2e LP |
4089 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
4090 | if (r < 0) | |
4091 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 4092 | |
cbe83389 LP |
4093 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
4094 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
4095 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
d42b61d2 | 4096 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, EVENT_PRIORITY_CGROUP_INOTIFY); |
10bd3e2e LP |
4097 | if (r < 0) |
4098 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 4099 | |
10bd3e2e | 4100 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 4101 | |
611c4f8a | 4102 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 4103 | |
10bd3e2e LP |
4104 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
4105 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 4106 | |
ce906769 | 4107 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUPS_AGENT_PATH); |
23bbb0de | 4108 | if (r < 0) |
10bd3e2e LP |
4109 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
4110 | else if (r > 0) | |
4111 | log_debug("Installed release agent."); | |
4112 | else if (r == 0) | |
4113 | log_debug("Release agent already installed."); | |
4114 | } | |
efdb0237 | 4115 | |
09e24654 | 4116 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
4117 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
4118 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
4119 | if (r >= 0) { |
4120 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
4121 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
4122 | if (r < 0) | |
4123 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 4124 | |
aa77e234 MS |
4125 | /* 6. And pin it, so that it cannot be unmounted */ |
4126 | safe_close(m->pin_cgroupfs_fd); | |
4127 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
4128 | if (m->pin_cgroupfs_fd < 0) | |
4129 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 4130 | |
638cece4 | 4131 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 4132 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 4133 | |
09e24654 | 4134 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 4135 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 4136 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 4137 | |
17f14955 | 4138 | /* 8. Figure out which controllers are supported */ |
0fa7b500 | 4139 | r = cg_mask_supported_subtree(m->cgroup_root, &m->cgroup_supported); |
efdb0237 LP |
4140 | if (r < 0) |
4141 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
4142 | |
4143 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
4144 | r = cg_bpf_mask_supported(&mask); | |
4145 | if (r < 0) | |
4146 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
4147 | m->cgroup_supported |= mask; | |
4148 | ||
4149 | /* 10. Log which controllers are supported */ | |
e8616626 ZJS |
4150 | for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
4151 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), | |
4152 | yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); | |
9156e799 | 4153 | |
a32360f1 | 4154 | return 0; |
8e274523 LP |
4155 | } |
4156 | ||
c6c18be3 | 4157 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
4158 | assert(m); |
4159 | ||
9444b1f2 LP |
4160 | /* We can't really delete the group, since we are in it. But |
4161 | * let's trim it. */ | |
5dd2f5ff | 4162 | if (delete && m->cgroup_root && !FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL)) |
efdb0237 LP |
4163 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
4164 | ||
5dcadb4c | 4165 | m->cgroup_empty_event_source = sd_event_source_disable_unref(m->cgroup_empty_event_source); |
09e24654 | 4166 | |
0bb814c2 | 4167 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 4168 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 | 4169 | |
5dcadb4c | 4170 | m->cgroup_inotify_event_source = sd_event_source_disable_unref(m->cgroup_inotify_event_source); |
efdb0237 | 4171 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); |
8e274523 | 4172 | |
03e334a1 | 4173 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 4174 | |
efdb0237 | 4175 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
4176 | } |
4177 | ||
4ad49000 | 4178 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 4179 | char *p; |
4ad49000 | 4180 | Unit *u; |
acb14d31 LP |
4181 | |
4182 | assert(m); | |
4183 | assert(cgroup); | |
acb14d31 | 4184 | |
4ad49000 LP |
4185 | u = hashmap_get(m->cgroup_unit, cgroup); |
4186 | if (u) | |
4187 | return u; | |
acb14d31 | 4188 | |
2f82562b | 4189 | p = strdupa_safe(cgroup); |
acb14d31 LP |
4190 | for (;;) { |
4191 | char *e; | |
4192 | ||
4193 | e = strrchr(p, '/'); | |
efdb0237 LP |
4194 | if (!e || e == p) |
4195 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
4196 | |
4197 | *e = 0; | |
4198 | ||
4ad49000 LP |
4199 | u = hashmap_get(m->cgroup_unit, p); |
4200 | if (u) | |
4201 | return u; | |
acb14d31 LP |
4202 | } |
4203 | } | |
4204 | ||
495e75ed | 4205 | Unit *manager_get_unit_by_pidref_cgroup(Manager *m, PidRef *pid) { |
4ad49000 | 4206 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 4207 | |
8c47c732 LP |
4208 | assert(m); |
4209 | ||
a9062242 | 4210 | if (cg_pidref_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
4211 | return NULL; |
4212 | ||
4213 | return manager_get_unit_by_cgroup(m, cgroup); | |
4214 | } | |
4215 | ||
495e75ed | 4216 | Unit *manager_get_unit_by_pidref_watching(Manager *m, PidRef *pid) { |
62a76913 | 4217 | Unit *u, **array; |
b3ac818b LP |
4218 | |
4219 | assert(m); | |
4220 | ||
495e75ed LP |
4221 | if (!pidref_is_set(pid)) |
4222 | return NULL; | |
62a76913 | 4223 | |
495e75ed LP |
4224 | u = hashmap_get(m->watch_pids, pid); |
4225 | if (u) | |
4226 | return u; | |
4227 | ||
4228 | array = hashmap_get(m->watch_pids_more, pid); | |
4229 | if (array) | |
4230 | return array[0]; | |
4231 | ||
4232 | return NULL; | |
4233 | } | |
4234 | ||
4235 | Unit *manager_get_unit_by_pidref(Manager *m, PidRef *pid) { | |
4236 | Unit *u; | |
4237 | ||
4238 | assert(m); | |
4239 | ||
4240 | /* Note that a process might be owned by multiple units, we return only one here, which is good | |
4241 | * enough for most cases, though not strictly correct. We prefer the one reported by cgroup | |
4242 | * membership, as that's the most relevant one as children of the process will be assigned to that | |
4243 | * one, too, before all else. */ | |
4244 | ||
4245 | if (!pidref_is_set(pid)) | |
8c47c732 LP |
4246 | return NULL; |
4247 | ||
a7a87769 | 4248 | if (pidref_is_self(pid)) |
efdb0237 | 4249 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
495e75ed LP |
4250 | if (pid->pid == 1) |
4251 | return NULL; | |
efdb0237 | 4252 | |
495e75ed | 4253 | u = manager_get_unit_by_pidref_cgroup(m, pid); |
5fe8876b LP |
4254 | if (u) |
4255 | return u; | |
4256 | ||
495e75ed | 4257 | u = manager_get_unit_by_pidref_watching(m, pid); |
5fe8876b LP |
4258 | if (u) |
4259 | return u; | |
4260 | ||
62a76913 | 4261 | return NULL; |
6dde1f33 | 4262 | } |
4fbf50b3 | 4263 | |
495e75ed LP |
4264 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { |
4265 | assert(m); | |
4266 | ||
4267 | if (!pid_is_valid(pid)) | |
4268 | return NULL; | |
4269 | ||
4270 | return manager_get_unit_by_pidref(m, &PIDREF_MAKE_FROM_PID(pid)); | |
4271 | } | |
4272 | ||
4ad49000 LP |
4273 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
4274 | Unit *u; | |
4fbf50b3 | 4275 | |
4ad49000 LP |
4276 | assert(m); |
4277 | assert(cgroup); | |
4fbf50b3 | 4278 | |
09e24654 LP |
4279 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
4280 | * or from the --system instance */ | |
4281 | ||
d8fdc620 LP |
4282 | log_debug("Got cgroup empty notification for: %s", cgroup); |
4283 | ||
4ad49000 | 4284 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
4285 | if (!u) |
4286 | return 0; | |
b56c28c3 | 4287 | |
09e24654 LP |
4288 | unit_add_to_cgroup_empty_queue(u); |
4289 | return 1; | |
5ad096b3 LP |
4290 | } |
4291 | ||
93ff34e4 | 4292 | int unit_get_memory_available(Unit *u, uint64_t *ret) { |
8db929a1 | 4293 | uint64_t available = UINT64_MAX, current = 0; |
93ff34e4 LB |
4294 | |
4295 | assert(u); | |
4296 | assert(ret); | |
4297 | ||
4298 | /* If data from cgroups can be accessed, try to find out how much more memory a unit can | |
4299 | * claim before hitting the configured cgroup limits (if any). Consider both MemoryHigh | |
4300 | * and MemoryMax, and also any slice the unit might be nested below. */ | |
4301 | ||
727cea76 | 4302 | do { |
8db929a1 | 4303 | uint64_t unit_available, unit_limit = UINT64_MAX; |
727cea76 | 4304 | CGroupContext *unit_context; |
93ff34e4 LB |
4305 | |
4306 | /* No point in continuing if we can't go any lower */ | |
4307 | if (available == 0) | |
4308 | break; | |
4309 | ||
727cea76 MK |
4310 | unit_context = unit_get_cgroup_context(u); |
4311 | if (!unit_context) | |
4312 | return -ENODATA; | |
93ff34e4 | 4313 | |
727cea76 | 4314 | if (!u->cgroup_path) |
93ff34e4 LB |
4315 | continue; |
4316 | ||
8db929a1 MK |
4317 | (void) unit_get_memory_current(u, ¤t); |
4318 | /* in case of error, previous current propagates as lower bound */ | |
4319 | ||
727cea76 MK |
4320 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
4321 | unit_limit = physical_memory(); | |
4322 | else if (unit_context->memory_max == UINT64_MAX && unit_context->memory_high == UINT64_MAX) | |
93ff34e4 | 4323 | continue; |
727cea76 | 4324 | unit_limit = MIN3(unit_limit, unit_context->memory_max, unit_context->memory_high); |
93ff34e4 | 4325 | |
8db929a1 | 4326 | unit_available = LESS_BY(unit_limit, current); |
727cea76 MK |
4327 | available = MIN(unit_available, available); |
4328 | } while ((u = UNIT_GET_SLICE(u))); | |
93ff34e4 LB |
4329 | |
4330 | *ret = available; | |
4331 | ||
4332 | return 0; | |
4333 | } | |
4334 | ||
5ad096b3 | 4335 | int unit_get_memory_current(Unit *u, uint64_t *ret) { |
5ad096b3 LP |
4336 | int r; |
4337 | ||
9824ab1f MY |
4338 | // FIXME: Merge this into unit_get_memory_accounting after support for cgroup v1 is dropped |
4339 | ||
5ad096b3 LP |
4340 | assert(u); |
4341 | assert(ret); | |
4342 | ||
2e4025c0 | 4343 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
4344 | return -ENODATA; |
4345 | ||
5ad096b3 LP |
4346 | if (!u->cgroup_path) |
4347 | return -ENODATA; | |
4348 | ||
1f73aa00 | 4349 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 4350 | if (unit_has_host_root_cgroup(u)) |
c482724a | 4351 | return procfs_memory_get_used(ret); |
1f73aa00 | 4352 | |
efdb0237 | 4353 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
4354 | return -ENODATA; |
4355 | ||
b4cccbc1 LP |
4356 | r = cg_all_unified(); |
4357 | if (r < 0) | |
4358 | return r; | |
5ad096b3 | 4359 | |
613328c3 | 4360 | return cg_get_attribute_as_uint64("memory", u->cgroup_path, r > 0 ? "memory.current" : "memory.usage_in_bytes", ret); |
5ad096b3 LP |
4361 | } |
4362 | ||
9824ab1f MY |
4363 | int unit_get_memory_accounting(Unit *u, CGroupMemoryAccountingMetric metric, uint64_t *ret) { |
4364 | ||
4365 | static const char* const attributes_table[_CGROUP_MEMORY_ACCOUNTING_METRIC_MAX] = { | |
4366 | [CGROUP_MEMORY_PEAK] = "memory.peak", | |
4367 | [CGROUP_MEMORY_SWAP_CURRENT] = "memory.swap.current", | |
4368 | [CGROUP_MEMORY_SWAP_PEAK] = "memory.swap.peak", | |
4369 | [CGROUP_MEMORY_ZSWAP_CURRENT] = "memory.zswap.current", | |
4370 | }; | |
4371 | ||
4372 | uint64_t bytes; | |
f17b07f4 | 4373 | bool updated = false; |
6c71db76 FS |
4374 | int r; |
4375 | ||
4376 | assert(u); | |
9824ab1f MY |
4377 | assert(metric >= 0); |
4378 | assert(metric < _CGROUP_MEMORY_ACCOUNTING_METRIC_MAX); | |
6c71db76 | 4379 | |
37533c94 FS |
4380 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
4381 | return -ENODATA; | |
4382 | ||
6c71db76 | 4383 | if (!u->cgroup_path) |
f17b07f4 | 4384 | /* If the cgroup is already gone, we try to find the last cached value. */ |
a8aed6a9 | 4385 | goto finish; |
6c71db76 FS |
4386 | |
4387 | /* The root cgroup doesn't expose this information. */ | |
4388 | if (unit_has_host_root_cgroup(u)) | |
4389 | return -ENODATA; | |
4390 | ||
9824ab1f | 4391 | if (!FLAGS_SET(u->cgroup_realized_mask, CGROUP_MASK_MEMORY)) |
6c71db76 FS |
4392 | return -ENODATA; |
4393 | ||
4394 | r = cg_all_unified(); | |
4395 | if (r < 0) | |
4396 | return r; | |
9824ab1f | 4397 | if (r == 0) |
6c71db76 FS |
4398 | return -ENODATA; |
4399 | ||
9824ab1f | 4400 | r = cg_get_attribute_as_uint64("memory", u->cgroup_path, attributes_table[metric], &bytes); |
f17b07f4 | 4401 | if (r < 0 && r != -ENODATA) |
9824ab1f | 4402 | return r; |
f17b07f4 | 4403 | updated = r >= 0; |
6c71db76 | 4404 | |
a8aed6a9 MY |
4405 | finish: |
4406 | if (metric <= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST) { | |
4407 | uint64_t *last = &u->memory_accounting_last[metric]; | |
6c71db76 | 4408 | |
a8aed6a9 MY |
4409 | if (updated) |
4410 | *last = bytes; | |
4411 | else if (*last != UINT64_MAX) | |
4412 | bytes = *last; | |
4413 | else | |
4414 | return -ENODATA; | |
f17b07f4 | 4415 | |
a8aed6a9 | 4416 | } else if (!updated) |
f17b07f4 | 4417 | return -ENODATA; |
6c71db76 | 4418 | |
6c71db76 FS |
4419 | if (ret) |
4420 | *ret = bytes; | |
4421 | ||
4422 | return 0; | |
4423 | } | |
4424 | ||
03a7b521 | 4425 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
03a7b521 LP |
4426 | assert(u); |
4427 | assert(ret); | |
4428 | ||
2e4025c0 | 4429 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
4430 | return -ENODATA; |
4431 | ||
03a7b521 LP |
4432 | if (!u->cgroup_path) |
4433 | return -ENODATA; | |
4434 | ||
c36a69f4 | 4435 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 4436 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
4437 | return procfs_tasks_get_current(ret); |
4438 | ||
1f73aa00 LP |
4439 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
4440 | return -ENODATA; | |
4441 | ||
613328c3 | 4442 | return cg_get_attribute_as_uint64("pids", u->cgroup_path, "pids.current", ret); |
03a7b521 LP |
4443 | } |
4444 | ||
5ad096b3 | 4445 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
5ad096b3 LP |
4446 | uint64_t ns; |
4447 | int r; | |
4448 | ||
4449 | assert(u); | |
4450 | assert(ret); | |
4451 | ||
4452 | if (!u->cgroup_path) | |
4453 | return -ENODATA; | |
4454 | ||
1f73aa00 | 4455 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 4456 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
4457 | return procfs_cpu_get_usage(ret); |
4458 | ||
f98c2585 CD |
4459 | /* Requisite controllers for CPU accounting are not enabled */ |
4460 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
4461 | return -ENODATA; | |
4462 | ||
92a99304 LP |
4463 | r = cg_all_unified(); |
4464 | if (r < 0) | |
4465 | return r; | |
b4cccbc1 | 4466 | if (r > 0) { |
66ebf6c0 TH |
4467 | _cleanup_free_ char *val = NULL; |
4468 | uint64_t us; | |
5ad096b3 | 4469 | |
b734a4ff | 4470 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
4471 | if (IN_SET(r, -ENOENT, -ENXIO)) |
4472 | return -ENODATA; | |
d742f4b5 LP |
4473 | if (r < 0) |
4474 | return r; | |
66ebf6c0 TH |
4475 | |
4476 | r = safe_atou64(val, &us); | |
4477 | if (r < 0) | |
4478 | return r; | |
4479 | ||
4480 | ns = us * NSEC_PER_USEC; | |
613328c3 AZ |
4481 | } else |
4482 | return cg_get_attribute_as_uint64("cpuacct", u->cgroup_path, "cpuacct.usage", ret); | |
5ad096b3 LP |
4483 | |
4484 | *ret = ns; | |
4485 | return 0; | |
4486 | } | |
4487 | ||
4488 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
4489 | nsec_t ns; | |
4490 | int r; | |
4491 | ||
fe700f46 LP |
4492 | assert(u); |
4493 | ||
4494 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
4495 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
4496 | * call this function with a NULL return value. */ | |
4497 | ||
2e4025c0 | 4498 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
4499 | return -ENODATA; |
4500 | ||
5ad096b3 | 4501 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
4502 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
4503 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
4504 | * cached value. */ | |
4505 | ||
4506 | if (ret) | |
4507 | *ret = u->cpu_usage_last; | |
4508 | return 0; | |
4509 | } | |
5ad096b3 LP |
4510 | if (r < 0) |
4511 | return r; | |
4512 | ||
66ebf6c0 TH |
4513 | if (ns > u->cpu_usage_base) |
4514 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
4515 | else |
4516 | ns = 0; | |
4517 | ||
fe700f46 LP |
4518 | u->cpu_usage_last = ns; |
4519 | if (ret) | |
4520 | *ret = ns; | |
4521 | ||
5ad096b3 LP |
4522 | return 0; |
4523 | } | |
4524 | ||
906c06f6 DM |
4525 | int unit_get_ip_accounting( |
4526 | Unit *u, | |
4527 | CGroupIPAccountingMetric metric, | |
4528 | uint64_t *ret) { | |
4529 | ||
6b659ed8 | 4530 | uint64_t value; |
906c06f6 DM |
4531 | int fd, r; |
4532 | ||
4533 | assert(u); | |
4534 | assert(metric >= 0); | |
4535 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
4536 | assert(ret); | |
4537 | ||
2e4025c0 | 4538 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
4539 | return -ENODATA; |
4540 | ||
906c06f6 DM |
4541 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
4542 | u->ip_accounting_ingress_map_fd : | |
4543 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
4544 | if (fd < 0) |
4545 | return -ENODATA; | |
4546 | ||
4547 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 4548 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 4549 | else |
6b659ed8 LP |
4550 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
4551 | if (r < 0) | |
4552 | return r; | |
4553 | ||
4554 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
4555 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
4556 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
4557 | ||
4558 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
4559 | |
4560 | return r; | |
4561 | } | |
4562 | ||
4fb0d2dc MK |
4563 | static uint64_t unit_get_effective_limit_one(Unit *u, CGroupLimitType type) { |
4564 | CGroupContext *cc; | |
4565 | ||
4566 | assert(u); | |
4567 | assert(UNIT_HAS_CGROUP_CONTEXT(u)); | |
4568 | ||
93f8e88d MK |
4569 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
4570 | switch (type) { | |
4571 | case CGROUP_LIMIT_MEMORY_MAX: | |
4572 | case CGROUP_LIMIT_MEMORY_HIGH: | |
4573 | return physical_memory(); | |
4574 | case CGROUP_LIMIT_TASKS_MAX: | |
4575 | return system_tasks_max(); | |
4576 | default: | |
4577 | assert_not_reached(); | |
4578 | } | |
4579 | ||
c658ad79 | 4580 | cc = ASSERT_PTR(unit_get_cgroup_context(u)); |
4fb0d2dc MK |
4581 | switch (type) { |
4582 | /* Note: on legacy/hybrid hierarchies memory_max stays CGROUP_LIMIT_MAX unless configured | |
4583 | * explicitly. Effective value of MemoryLimit= (cgroup v1) is not implemented. */ | |
4584 | case CGROUP_LIMIT_MEMORY_MAX: | |
4585 | return cc->memory_max; | |
4586 | case CGROUP_LIMIT_MEMORY_HIGH: | |
4587 | return cc->memory_high; | |
4588 | case CGROUP_LIMIT_TASKS_MAX: | |
4589 | return cgroup_tasks_max_resolve(&cc->tasks_max); | |
4590 | default: | |
4591 | assert_not_reached(); | |
4592 | } | |
4593 | } | |
4594 | ||
4595 | int unit_get_effective_limit(Unit *u, CGroupLimitType type, uint64_t *ret) { | |
4596 | uint64_t infimum; | |
4597 | ||
4598 | assert(u); | |
4599 | assert(ret); | |
4600 | assert(type >= 0); | |
4601 | assert(type < _CGROUP_LIMIT_TYPE_MAX); | |
4602 | ||
4603 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4604 | return -EINVAL; | |
4605 | ||
4606 | infimum = unit_get_effective_limit_one(u, type); | |
4607 | for (Unit *slice = UNIT_GET_SLICE(u); slice; slice = UNIT_GET_SLICE(slice)) | |
4608 | infimum = MIN(infimum, unit_get_effective_limit_one(slice, type)); | |
4609 | ||
4610 | *ret = infimum; | |
4611 | return 0; | |
4612 | } | |
4613 | ||
fbe14fc9 LP |
4614 | static int unit_get_io_accounting_raw(Unit *u, uint64_t ret[static _CGROUP_IO_ACCOUNTING_METRIC_MAX]) { |
4615 | static const char *const field_names[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
4616 | [CGROUP_IO_READ_BYTES] = "rbytes=", | |
4617 | [CGROUP_IO_WRITE_BYTES] = "wbytes=", | |
4618 | [CGROUP_IO_READ_OPERATIONS] = "rios=", | |
4619 | [CGROUP_IO_WRITE_OPERATIONS] = "wios=", | |
4620 | }; | |
4621 | uint64_t acc[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {}; | |
4622 | _cleanup_free_ char *path = NULL; | |
4623 | _cleanup_fclose_ FILE *f = NULL; | |
4624 | int r; | |
4625 | ||
4626 | assert(u); | |
4627 | ||
4628 | if (!u->cgroup_path) | |
4629 | return -ENODATA; | |
4630 | ||
4631 | if (unit_has_host_root_cgroup(u)) | |
4632 | return -ENODATA; /* TODO: return useful data for the top-level cgroup */ | |
4633 | ||
4634 | r = cg_all_unified(); | |
4635 | if (r < 0) | |
4636 | return r; | |
4637 | if (r == 0) /* TODO: support cgroupv1 */ | |
4638 | return -ENODATA; | |
4639 | ||
4640 | if (!FLAGS_SET(u->cgroup_realized_mask, CGROUP_MASK_IO)) | |
4641 | return -ENODATA; | |
4642 | ||
4643 | r = cg_get_path("io", u->cgroup_path, "io.stat", &path); | |
4644 | if (r < 0) | |
4645 | return r; | |
4646 | ||
4647 | f = fopen(path, "re"); | |
4648 | if (!f) | |
4649 | return -errno; | |
4650 | ||
4651 | for (;;) { | |
4652 | _cleanup_free_ char *line = NULL; | |
4653 | const char *p; | |
4654 | ||
4655 | r = read_line(f, LONG_LINE_MAX, &line); | |
4656 | if (r < 0) | |
4657 | return r; | |
4658 | if (r == 0) | |
4659 | break; | |
4660 | ||
4661 | p = line; | |
4662 | p += strcspn(p, WHITESPACE); /* Skip over device major/minor */ | |
4663 | p += strspn(p, WHITESPACE); /* Skip over following whitespace */ | |
4664 | ||
4665 | for (;;) { | |
4666 | _cleanup_free_ char *word = NULL; | |
4667 | ||
4668 | r = extract_first_word(&p, &word, NULL, EXTRACT_RETAIN_ESCAPE); | |
4669 | if (r < 0) | |
4670 | return r; | |
4671 | if (r == 0) | |
4672 | break; | |
4673 | ||
4674 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
4675 | const char *x; | |
4676 | ||
4677 | x = startswith(word, field_names[i]); | |
4678 | if (x) { | |
4679 | uint64_t w; | |
4680 | ||
4681 | r = safe_atou64(x, &w); | |
4682 | if (r < 0) | |
4683 | return r; | |
4684 | ||
4685 | /* Sum up the stats of all devices */ | |
4686 | acc[i] += w; | |
4687 | break; | |
4688 | } | |
4689 | } | |
4690 | } | |
4691 | } | |
4692 | ||
4693 | memcpy(ret, acc, sizeof(acc)); | |
4694 | return 0; | |
4695 | } | |
4696 | ||
4697 | int unit_get_io_accounting( | |
4698 | Unit *u, | |
4699 | CGroupIOAccountingMetric metric, | |
4700 | bool allow_cache, | |
4701 | uint64_t *ret) { | |
4702 | ||
4703 | uint64_t raw[_CGROUP_IO_ACCOUNTING_METRIC_MAX]; | |
4704 | int r; | |
4705 | ||
4706 | /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */ | |
4707 | ||
4708 | if (!UNIT_CGROUP_BOOL(u, io_accounting)) | |
4709 | return -ENODATA; | |
4710 | ||
4711 | if (allow_cache && u->io_accounting_last[metric] != UINT64_MAX) | |
4712 | goto done; | |
4713 | ||
4714 | r = unit_get_io_accounting_raw(u, raw); | |
4715 | if (r == -ENODATA && u->io_accounting_last[metric] != UINT64_MAX) | |
4716 | goto done; | |
4717 | if (r < 0) | |
4718 | return r; | |
4719 | ||
4720 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
4721 | /* Saturated subtraction */ | |
4722 | if (raw[i] > u->io_accounting_base[i]) | |
4723 | u->io_accounting_last[i] = raw[i] - u->io_accounting_base[i]; | |
4724 | else | |
4725 | u->io_accounting_last[i] = 0; | |
4726 | } | |
4727 | ||
4728 | done: | |
4729 | if (ret) | |
4730 | *ret = u->io_accounting_last[metric]; | |
4731 | ||
4732 | return 0; | |
4733 | } | |
4734 | ||
906c06f6 | 4735 | int unit_reset_cpu_accounting(Unit *u) { |
5ad096b3 LP |
4736 | int r; |
4737 | ||
4738 | assert(u); | |
4739 | ||
fe700f46 LP |
4740 | u->cpu_usage_last = NSEC_INFINITY; |
4741 | ||
0bbff7d6 | 4742 | r = unit_get_cpu_usage_raw(u, &u->cpu_usage_base); |
5ad096b3 | 4743 | if (r < 0) { |
66ebf6c0 | 4744 | u->cpu_usage_base = 0; |
5ad096b3 | 4745 | return r; |
b56c28c3 | 4746 | } |
2633eb83 | 4747 | |
4ad49000 | 4748 | return 0; |
4fbf50b3 LP |
4749 | } |
4750 | ||
d4bdc202 MY |
4751 | void unit_reset_memory_accounting_last(Unit *u) { |
4752 | assert(u); | |
4753 | ||
4754 | FOREACH_ARRAY(i, u->memory_accounting_last, ELEMENTSOF(u->memory_accounting_last)) | |
4755 | *i = UINT64_MAX; | |
4756 | } | |
4757 | ||
906c06f6 | 4758 | int unit_reset_ip_accounting(Unit *u) { |
cbd2abbb | 4759 | int r = 0; |
906c06f6 DM |
4760 | |
4761 | assert(u); | |
4762 | ||
4763 | if (u->ip_accounting_ingress_map_fd >= 0) | |
cbd2abbb | 4764 | RET_GATHER(r, bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd)); |
906c06f6 DM |
4765 | |
4766 | if (u->ip_accounting_egress_map_fd >= 0) | |
cbd2abbb | 4767 | RET_GATHER(r, bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd)); |
906c06f6 | 4768 | |
6b659ed8 LP |
4769 | zero(u->ip_accounting_extra); |
4770 | ||
cbd2abbb | 4771 | return r; |
906c06f6 DM |
4772 | } |
4773 | ||
d4bdc202 MY |
4774 | void unit_reset_io_accounting_last(Unit *u) { |
4775 | assert(u); | |
4776 | ||
4777 | FOREACH_ARRAY(i, u->io_accounting_last, _CGROUP_IO_ACCOUNTING_METRIC_MAX) | |
4778 | *i = UINT64_MAX; | |
4779 | } | |
4780 | ||
fbe14fc9 LP |
4781 | int unit_reset_io_accounting(Unit *u) { |
4782 | int r; | |
4783 | ||
4784 | assert(u); | |
4785 | ||
d4bdc202 | 4786 | unit_reset_io_accounting_last(u); |
fbe14fc9 LP |
4787 | |
4788 | r = unit_get_io_accounting_raw(u, u->io_accounting_base); | |
4789 | if (r < 0) { | |
4790 | zero(u->io_accounting_base); | |
4791 | return r; | |
4792 | } | |
4793 | ||
4794 | return 0; | |
4795 | } | |
4796 | ||
9b2559a1 | 4797 | int unit_reset_accounting(Unit *u) { |
cbd2abbb | 4798 | int r = 0; |
9b2559a1 LP |
4799 | |
4800 | assert(u); | |
4801 | ||
cbd2abbb MY |
4802 | RET_GATHER(r, unit_reset_cpu_accounting(u)); |
4803 | RET_GATHER(r, unit_reset_io_accounting(u)); | |
4804 | RET_GATHER(r, unit_reset_ip_accounting(u)); | |
d4bdc202 | 4805 | unit_reset_memory_accounting_last(u); |
9b2559a1 | 4806 | |
cbd2abbb | 4807 | return r; |
9b2559a1 LP |
4808 | } |
4809 | ||
e7ab4d1a LP |
4810 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
4811 | assert(u); | |
4812 | ||
4813 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4814 | return; | |
4815 | ||
4816 | if (m == 0) | |
4817 | return; | |
4818 | ||
538b4852 TH |
4819 | /* always invalidate compat pairs together */ |
4820 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
4821 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
4822 | ||
7cce4fb7 LP |
4823 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
4824 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
4825 | ||
e00068e7 | 4826 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
4827 | return; |
4828 | ||
e00068e7 | 4829 | u->cgroup_invalidated_mask |= m; |
91a6073e | 4830 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
4831 | } |
4832 | ||
906c06f6 DM |
4833 | void unit_invalidate_cgroup_bpf(Unit *u) { |
4834 | assert(u); | |
4835 | ||
4836 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4837 | return; | |
4838 | ||
17f14955 | 4839 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
4840 | return; |
4841 | ||
17f14955 | 4842 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 4843 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
4844 | |
4845 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
4846 | * list of our children includes our own. */ | |
4847 | if (u->type == UNIT_SLICE) { | |
4848 | Unit *member; | |
906c06f6 | 4849 | |
d219a2b0 | 4850 | UNIT_FOREACH_DEPENDENCY(member, u, UNIT_ATOM_SLICE_OF) |
15ed3c3a | 4851 | unit_invalidate_cgroup_bpf(member); |
906c06f6 DM |
4852 | } |
4853 | } | |
4854 | ||
869f52f2 DS |
4855 | void unit_cgroup_catchup(Unit *u) { |
4856 | assert(u); | |
4857 | ||
4858 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4859 | return; | |
4860 | ||
4861 | /* We dropped the inotify watch during reexec/reload, so we need to | |
4862 | * check these as they may have changed. | |
4863 | * Note that (currently) the kernel doesn't actually update cgroup | |
4864 | * file modification times, so we can't just serialize and then check | |
4865 | * the mtime for file(s) we are interested in. */ | |
4866 | (void) unit_check_cgroup_events(u); | |
4867 | unit_add_to_cgroup_oom_queue(u); | |
4868 | } | |
4869 | ||
1d9cc876 LP |
4870 | bool unit_cgroup_delegate(Unit *u) { |
4871 | CGroupContext *c; | |
4872 | ||
4873 | assert(u); | |
4874 | ||
4875 | if (!UNIT_VTABLE(u)->can_delegate) | |
4876 | return false; | |
4877 | ||
4878 | c = unit_get_cgroup_context(u); | |
4879 | if (!c) | |
4880 | return false; | |
4881 | ||
4882 | return c->delegate; | |
4883 | } | |
4884 | ||
e7ab4d1a | 4885 | void manager_invalidate_startup_units(Manager *m) { |
e7ab4d1a LP |
4886 | Unit *u; |
4887 | ||
4888 | assert(m); | |
4889 | ||
90e74a66 | 4890 | SET_FOREACH(u, m->startup_units) |
9dfb6a3a | 4891 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO|CGROUP_MASK_CPUSET); |
e7ab4d1a LP |
4892 | } |
4893 | ||
d9e45bc3 MS |
4894 | int unit_cgroup_freezer_action(Unit *u, FreezerAction action) { |
4895 | _cleanup_free_ char *path = NULL; | |
4896 | FreezerState target, kernel = _FREEZER_STATE_INVALID; | |
7fcd2697 | 4897 | int r, ret; |
d9e45bc3 MS |
4898 | |
4899 | assert(u); | |
4900 | assert(IN_SET(action, FREEZER_FREEZE, FREEZER_THAW)); | |
4901 | ||
9a1e90ae MS |
4902 | if (!cg_freezer_supported()) |
4903 | return 0; | |
4904 | ||
a14137d9 | 4905 | /* Ignore all requests to thaw init.scope or -.slice and reject all requests to freeze them */ |
4906 | if (unit_has_name(u, SPECIAL_ROOT_SLICE) || unit_has_name(u, SPECIAL_INIT_SCOPE)) | |
4907 | return action == FREEZER_FREEZE ? -EPERM : 0; | |
4908 | ||
d9e45bc3 MS |
4909 | if (!u->cgroup_realized) |
4910 | return -EBUSY; | |
4911 | ||
a14137d9 | 4912 | if (action == FREEZER_THAW) { |
4913 | Unit *slice = UNIT_GET_SLICE(u); | |
4914 | ||
4915 | if (slice) { | |
4916 | r = unit_cgroup_freezer_action(slice, FREEZER_THAW); | |
4917 | if (r < 0) | |
4918 | return log_unit_error_errno(u, r, "Failed to thaw slice %s of unit: %m", slice->id); | |
4919 | } | |
4920 | } | |
4921 | ||
d9e45bc3 MS |
4922 | target = action == FREEZER_FREEZE ? FREEZER_FROZEN : FREEZER_RUNNING; |
4923 | ||
4924 | r = unit_freezer_state_kernel(u, &kernel); | |
4925 | if (r < 0) | |
4926 | log_unit_debug_errno(u, r, "Failed to obtain cgroup freezer state: %m"); | |
4927 | ||
4928 | if (target == kernel) { | |
4929 | u->freezer_state = target; | |
7fcd2697 | 4930 | if (action == FREEZER_FREEZE) |
4931 | return 0; | |
4932 | ret = 0; | |
4933 | } else | |
4934 | ret = 1; | |
d9e45bc3 MS |
4935 | |
4936 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.freeze", &path); | |
4937 | if (r < 0) | |
4938 | return r; | |
4939 | ||
4940 | log_unit_debug(u, "%s unit.", action == FREEZER_FREEZE ? "Freezing" : "Thawing"); | |
4941 | ||
7fcd2697 | 4942 | if (target != kernel) { |
4943 | if (action == FREEZER_FREEZE) | |
4944 | u->freezer_state = FREEZER_FREEZING; | |
4945 | else | |
4946 | u->freezer_state = FREEZER_THAWING; | |
4947 | } | |
d9e45bc3 MS |
4948 | |
4949 | r = write_string_file(path, one_zero(action == FREEZER_FREEZE), WRITE_STRING_FILE_DISABLE_BUFFER); | |
4950 | if (r < 0) | |
4951 | return r; | |
4952 | ||
7fcd2697 | 4953 | return ret; |
d9e45bc3 MS |
4954 | } |
4955 | ||
047f5d63 PH |
4956 | int unit_get_cpuset(Unit *u, CPUSet *cpus, const char *name) { |
4957 | _cleanup_free_ char *v = NULL; | |
4958 | int r; | |
4959 | ||
4960 | assert(u); | |
4961 | assert(cpus); | |
4962 | ||
4963 | if (!u->cgroup_path) | |
4964 | return -ENODATA; | |
4965 | ||
4966 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUSET) == 0) | |
4967 | return -ENODATA; | |
4968 | ||
4969 | r = cg_all_unified(); | |
4970 | if (r < 0) | |
4971 | return r; | |
4972 | if (r == 0) | |
4973 | return -ENODATA; | |
48fd01e5 LP |
4974 | |
4975 | r = cg_get_attribute("cpuset", u->cgroup_path, name, &v); | |
047f5d63 PH |
4976 | if (r == -ENOENT) |
4977 | return -ENODATA; | |
4978 | if (r < 0) | |
4979 | return r; | |
4980 | ||
4981 | return parse_cpu_set_full(v, cpus, false, NULL, NULL, 0, NULL); | |
4982 | } | |
4983 | ||
4e806bfa AZ |
4984 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
4985 | [CGROUP_DEVICE_POLICY_AUTO] = "auto", | |
4986 | [CGROUP_DEVICE_POLICY_CLOSED] = "closed", | |
4987 | [CGROUP_DEVICE_POLICY_STRICT] = "strict", | |
4988 | }; | |
4989 | ||
4ad49000 | 4990 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |
d9e45bc3 MS |
4991 | |
4992 | static const char* const freezer_action_table[_FREEZER_ACTION_MAX] = { | |
4993 | [FREEZER_FREEZE] = "freeze", | |
4994 | [FREEZER_THAW] = "thaw", | |
4995 | }; | |
4996 | ||
4997 | DEFINE_STRING_TABLE_LOOKUP(freezer_action, FreezerAction); | |
6bb00842 LP |
4998 | |
4999 | static const char* const cgroup_pressure_watch_table[_CGROUP_PRESSURE_WATCH_MAX] = { | |
5000 | [CGROUP_PRESSURE_WATCH_OFF] = "off", | |
5001 | [CGROUP_PRESSURE_WATCH_AUTO] = "auto", | |
5002 | [CGROUP_PRESSURE_WATCH_ON] = "on", | |
5003 | [CGROUP_PRESSURE_WATCH_SKIP] = "skip", | |
5004 | }; | |
5005 | ||
5006 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(cgroup_pressure_watch, CGroupPressureWatch, CGROUP_PRESSURE_WATCH_ON); | |
435996e6 DDM |
5007 | |
5008 | static const char* const cgroup_ip_accounting_metric_table[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = { | |
5009 | [CGROUP_IP_INGRESS_BYTES] = "IPIngressBytes", | |
5010 | [CGROUP_IP_EGRESS_BYTES] = "IPEgressBytes", | |
5011 | [CGROUP_IP_INGRESS_PACKETS] = "IPIngressPackets", | |
5012 | [CGROUP_IP_EGRESS_PACKETS] = "IPEgressPackets", | |
5013 | }; | |
5014 | ||
5015 | DEFINE_STRING_TABLE_LOOKUP(cgroup_ip_accounting_metric, CGroupIPAccountingMetric); | |
5016 | ||
5017 | static const char* const cgroup_io_accounting_metric_table[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
5018 | [CGROUP_IO_READ_BYTES] = "IOReadBytes", | |
5019 | [CGROUP_IO_WRITE_BYTES] = "IOWriteBytes", | |
5020 | [CGROUP_IO_READ_OPERATIONS] = "IOReadOperations", | |
5021 | [CGROUP_IO_WRITE_OPERATIONS] = "IOWriteOperations", | |
5022 | }; | |
5023 | ||
5024 | DEFINE_STRING_TABLE_LOOKUP(cgroup_io_accounting_metric, CGroupIOAccountingMetric); | |
9824ab1f MY |
5025 | |
5026 | static const char* const cgroup_memory_accounting_metric_table[_CGROUP_MEMORY_ACCOUNTING_METRIC_MAX] = { | |
5027 | [CGROUP_MEMORY_PEAK] = "MemoryPeak", | |
5028 | [CGROUP_MEMORY_SWAP_CURRENT] = "MemorySwapCurrent", | |
5029 | [CGROUP_MEMORY_SWAP_PEAK] = "MemorySwapPeak", | |
5030 | [CGROUP_MEMORY_ZSWAP_CURRENT] = "MemoryZSwapCurrent", | |
5031 | }; | |
5032 | ||
5033 | DEFINE_STRING_TABLE_LOOKUP(cgroup_memory_accounting_metric, CGroupMemoryAccountingMetric); | |
4fb0d2dc | 5034 | |
8ad61489 | 5035 | static const char *const cgroup_effective_limit_type_table[_CGROUP_LIMIT_TYPE_MAX] = { |
4fb0d2dc MK |
5036 | [CGROUP_LIMIT_MEMORY_MAX] = "EffectiveMemoryMax", |
5037 | [CGROUP_LIMIT_MEMORY_HIGH] = "EffectiveMemoryHigh", | |
5038 | [CGROUP_LIMIT_TASKS_MAX] = "EffectiveTasksMax", | |
5039 | }; | |
5040 | ||
8ad61489 | 5041 | DEFINE_STRING_TABLE_LOOKUP(cgroup_effective_limit_type, CGroupLimitType); |