]>
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" |
62e22490 | 13 | #include "bpf-restrict-ifaces.h" |
cd09a5f3 | 14 | #include "bpf-socket-bind.h" |
45c2e068 | 15 | #include "btrfs-util.h" |
6592b975 | 16 | #include "bus-error.h" |
78fa2f91 | 17 | #include "bus-locator.h" |
fdb3deca | 18 | #include "cgroup-setup.h" |
03a7b521 | 19 | #include "cgroup-util.h" |
3ffd4af2 | 20 | #include "cgroup.h" |
7176f06c | 21 | #include "devnum-util.h" |
3ffd4af2 | 22 | #include "fd-util.h" |
0d39fa9c | 23 | #include "fileio.h" |
dc7d69b3 | 24 | #include "firewall-util.h" |
84ebe6f0 | 25 | #include "in-addr-prefix-util.h" |
9e5fd717 | 26 | #include "inotify-util.h" |
d9e45bc3 | 27 | #include "io-util.h" |
5587ce7f | 28 | #include "ip-protocol-list.h" |
3a0f06c4 | 29 | #include "limits-util.h" |
d9e45bc3 | 30 | #include "nulstr-util.h" |
6bedfcbb | 31 | #include "parse-util.h" |
9eb977db | 32 | #include "path-util.h" |
1ead0b2a | 33 | #include "percent-util.h" |
03a7b521 | 34 | #include "process-util.h" |
c36a69f4 | 35 | #include "procfs-util.h" |
84c01612 | 36 | #include "set.h" |
9cc54544 | 37 | #include "serialize.h" |
9444b1f2 | 38 | #include "special.h" |
906c06f6 | 39 | #include "stdio-util.h" |
8b43440b | 40 | #include "string-table.h" |
07630cea | 41 | #include "string-util.h" |
cc6271f1 | 42 | #include "virt.h" |
8e274523 | 43 | |
b1994387 ILG |
44 | #if BPF_FRAMEWORK |
45 | #include "bpf-dlopen.h" | |
46 | #include "bpf-link.h" | |
47 | #include "bpf/restrict_fs/restrict-fs-skel.h" | |
48 | #endif | |
49 | ||
10f28641 | 50 | #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
9a054909 | 51 | |
39b9fefb LP |
52 | /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access |
53 | * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask | |
54 | * out specific attributes from us. */ | |
55 | #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING) | |
56 | ||
94f0b13b | 57 | uint64_t cgroup_tasks_max_resolve(const CGroupTasksMax *tasks_max) { |
3a0f06c4 ZJS |
58 | if (tasks_max->scale == 0) |
59 | return tasks_max->value; | |
60 | ||
61 | return system_tasks_max_scale(tasks_max->value, tasks_max->scale); | |
62 | } | |
63 | ||
611c4f8a | 64 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
65 | assert(m); |
66 | ||
67 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
68 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
69 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
70 | * we run in any kind of container virtualization. */ | |
71 | ||
28cfdc5a LP |
72 | if (MANAGER_IS_USER(m)) |
73 | return false; | |
74 | ||
cc6271f1 LP |
75 | if (detect_container() > 0) |
76 | return false; | |
77 | ||
57ea45e1 | 78 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
79 | } |
80 | ||
9dfb6a3a PM |
81 | bool unit_has_startup_cgroup_constraints(Unit *u) { |
82 | assert(u); | |
83 | ||
84 | /* Returns true if this unit has any directives which apply during | |
85 | * startup/shutdown phases. */ | |
86 | ||
87 | CGroupContext *c; | |
88 | ||
89 | c = unit_get_cgroup_context(u); | |
90 | if (!c) | |
91 | return false; | |
92 | ||
93 | return c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
94 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
95 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
96 | c->startup_cpuset_cpus.set || | |
53fda560 LB |
97 | c->startup_cpuset_mems.set || |
98 | c->startup_memory_high_set || | |
99 | c->startup_memory_max_set || | |
100 | c->startup_memory_swap_max_set|| | |
101 | c->startup_memory_zswap_max_set || | |
102 | c->startup_memory_low_set; | |
9dfb6a3a PM |
103 | } |
104 | ||
611c4f8a | 105 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
106 | assert(u); |
107 | ||
cc6271f1 LP |
108 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
109 | * the manager manages the root cgroup. */ | |
f3725e64 | 110 | |
611c4f8a | 111 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
112 | return false; |
113 | ||
cc6271f1 | 114 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
115 | } |
116 | ||
293d32df LP |
117 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
118 | int r; | |
119 | ||
9cc54544 LP |
120 | assert(u); |
121 | ||
122 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
123 | if (!crt || !crt->cgroup_path) | |
124 | return -EOWNERDEAD; | |
125 | ||
126 | r = cg_set_attribute(controller, crt->cgroup_path, attribute, value); | |
293d32df | 127 | if (r < 0) |
8ed6f81b | 128 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", |
9cc54544 | 129 | strna(attribute), empty_to_root(crt->cgroup_path), (int) strcspn(value, NEWLINE), value); |
293d32df LP |
130 | |
131 | return r; | |
132 | } | |
133 | ||
2b40998d | 134 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
135 | static bool cgroup_compat_warned = false; |
136 | ||
137 | if (cgroup_compat_warned) | |
138 | return; | |
139 | ||
cc6271f1 LP |
140 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
141 | "See cgroup-compat debug messages for details."); | |
142 | ||
128fadc9 TH |
143 | cgroup_compat_warned = true; |
144 | } | |
145 | ||
146 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
147 | cgroup_compat_warn(); \ | |
148 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 149 | } while (false) |
128fadc9 | 150 | |
4ad49000 LP |
151 | void cgroup_context_init(CGroupContext *c) { |
152 | assert(c); | |
153 | ||
154eb43f LB |
154 | /* Initialize everything to the kernel defaults. When initializing a bool member to 'true', make |
155 | * sure to serialize in execute-serialize.c using serialize_bool() instead of | |
156 | * serialize_bool_elide(), as sd-executor will initialize here to 'true', but serialize_bool_elide() | |
157 | * skips serialization if the value is 'false' (as that's the common default), so if the value at | |
158 | * runtime is zero it would be lost after deserialization. Same when initializing uint64_t and other | |
159 | * values, update/add a conditional serialization check. This is to minimize the amount of | |
160 | * serialized data that is sent to the sd-executor, so that there is less work to do on the default | |
161 | * cases. */ | |
4ad49000 | 162 | |
de8a711a LP |
163 | *c = (CGroupContext) { |
164 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
165 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
166 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
10f28641 | 167 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 168 | |
de8a711a LP |
169 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
170 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 171 | |
de8a711a | 172 | .memory_high = CGROUP_LIMIT_MAX, |
53fda560 | 173 | .startup_memory_high = CGROUP_LIMIT_MAX, |
de8a711a | 174 | .memory_max = CGROUP_LIMIT_MAX, |
53fda560 | 175 | .startup_memory_max = CGROUP_LIMIT_MAX, |
de8a711a | 176 | .memory_swap_max = CGROUP_LIMIT_MAX, |
53fda560 | 177 | .startup_memory_swap_max = CGROUP_LIMIT_MAX, |
d7fe0a67 | 178 | .memory_zswap_max = CGROUP_LIMIT_MAX, |
53fda560 | 179 | .startup_memory_zswap_max = CGROUP_LIMIT_MAX, |
da4d897e | 180 | |
de8a711a | 181 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 182 | |
de8a711a LP |
183 | .io_weight = CGROUP_WEIGHT_INVALID, |
184 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 185 | |
de8a711a LP |
186 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
187 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 188 | |
94f0b13b | 189 | .tasks_max = CGROUP_TASKS_MAX_UNSET, |
4d824a4e AZ |
190 | |
191 | .moom_swap = MANAGED_OOM_AUTO, | |
192 | .moom_mem_pressure = MANAGED_OOM_AUTO, | |
4e806bfa | 193 | .moom_preference = MANAGED_OOM_PREFERENCE_NONE, |
6bb00842 LP |
194 | |
195 | .memory_pressure_watch = _CGROUP_PRESSURE_WATCH_INVALID, | |
196 | .memory_pressure_threshold_usec = USEC_INFINITY, | |
de8a711a | 197 | }; |
4ad49000 | 198 | } |
8e274523 | 199 | |
9c02eb28 | 200 | int cgroup_context_add_io_device_weight_dup(CGroupContext *c, const CGroupIODeviceWeight *w) { |
84c01612 MS |
201 | _cleanup_free_ CGroupIODeviceWeight *n = NULL; |
202 | ||
203 | assert(c); | |
204 | assert(w); | |
205 | ||
9c02eb28 | 206 | n = new(CGroupIODeviceWeight, 1); |
84c01612 MS |
207 | if (!n) |
208 | return -ENOMEM; | |
209 | ||
9c02eb28 MY |
210 | *n = (CGroupIODeviceWeight) { |
211 | .path = strdup(w->path), | |
212 | .weight = w->weight, | |
213 | }; | |
84c01612 MS |
214 | if (!n->path) |
215 | return -ENOMEM; | |
84c01612 MS |
216 | |
217 | LIST_PREPEND(device_weights, c->io_device_weights, TAKE_PTR(n)); | |
218 | return 0; | |
219 | } | |
220 | ||
9c02eb28 | 221 | int cgroup_context_add_io_device_limit_dup(CGroupContext *c, const CGroupIODeviceLimit *l) { |
84c01612 MS |
222 | _cleanup_free_ CGroupIODeviceLimit *n = NULL; |
223 | ||
224 | assert(c); | |
225 | assert(l); | |
226 | ||
227 | n = new0(CGroupIODeviceLimit, 1); | |
cae58298 | 228 | if (!n) |
84c01612 MS |
229 | return -ENOMEM; |
230 | ||
231 | n->path = strdup(l->path); | |
232 | if (!n->path) | |
233 | return -ENOMEM; | |
234 | ||
235 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
236 | n->limits[type] = l->limits[type]; | |
237 | ||
238 | LIST_PREPEND(device_limits, c->io_device_limits, TAKE_PTR(n)); | |
239 | return 0; | |
240 | } | |
241 | ||
9c02eb28 | 242 | int cgroup_context_add_io_device_latency_dup(CGroupContext *c, const CGroupIODeviceLatency *l) { |
84c01612 MS |
243 | _cleanup_free_ CGroupIODeviceLatency *n = NULL; |
244 | ||
245 | assert(c); | |
246 | assert(l); | |
247 | ||
9c02eb28 | 248 | n = new(CGroupIODeviceLatency, 1); |
84c01612 MS |
249 | if (!n) |
250 | return -ENOMEM; | |
251 | ||
9c02eb28 MY |
252 | *n = (CGroupIODeviceLatency) { |
253 | .path = strdup(l->path), | |
254 | .target_usec = l->target_usec, | |
255 | }; | |
84c01612 MS |
256 | if (!n->path) |
257 | return -ENOMEM; | |
258 | ||
84c01612 MS |
259 | LIST_PREPEND(device_latencies, c->io_device_latencies, TAKE_PTR(n)); |
260 | return 0; | |
261 | } | |
262 | ||
9c02eb28 | 263 | int cgroup_context_add_block_io_device_weight_dup(CGroupContext *c, const CGroupBlockIODeviceWeight *w) { |
84c01612 MS |
264 | _cleanup_free_ CGroupBlockIODeviceWeight *n = NULL; |
265 | ||
266 | assert(c); | |
267 | assert(w); | |
268 | ||
9c02eb28 | 269 | n = new(CGroupBlockIODeviceWeight, 1); |
84c01612 MS |
270 | if (!n) |
271 | return -ENOMEM; | |
272 | ||
9c02eb28 MY |
273 | *n = (CGroupBlockIODeviceWeight) { |
274 | .path = strdup(w->path), | |
275 | .weight = w->weight, | |
276 | }; | |
84c01612 MS |
277 | if (!n->path) |
278 | return -ENOMEM; | |
279 | ||
84c01612 MS |
280 | LIST_PREPEND(device_weights, c->blockio_device_weights, TAKE_PTR(n)); |
281 | return 0; | |
282 | } | |
283 | ||
9c02eb28 | 284 | int cgroup_context_add_block_io_device_bandwidth_dup(CGroupContext *c, const CGroupBlockIODeviceBandwidth *b) { |
84c01612 MS |
285 | _cleanup_free_ CGroupBlockIODeviceBandwidth *n = NULL; |
286 | ||
287 | assert(c); | |
288 | assert(b); | |
289 | ||
9c02eb28 | 290 | n = new(CGroupBlockIODeviceBandwidth, 1); |
84c01612 MS |
291 | if (!n) |
292 | return -ENOMEM; | |
293 | ||
294 | *n = (CGroupBlockIODeviceBandwidth) { | |
295 | .rbps = b->rbps, | |
296 | .wbps = b->wbps, | |
297 | }; | |
298 | ||
299 | LIST_PREPEND(device_bandwidths, c->blockio_device_bandwidths, TAKE_PTR(n)); | |
300 | return 0; | |
301 | } | |
302 | ||
9c02eb28 | 303 | int cgroup_context_add_device_allow_dup(CGroupContext *c, const CGroupDeviceAllow *a) { |
84c01612 MS |
304 | _cleanup_free_ CGroupDeviceAllow *n = NULL; |
305 | ||
306 | assert(c); | |
307 | assert(a); | |
308 | ||
9c02eb28 | 309 | n = new(CGroupDeviceAllow, 1); |
84c01612 MS |
310 | if (!n) |
311 | return -ENOMEM; | |
312 | ||
9c02eb28 MY |
313 | *n = (CGroupDeviceAllow) { |
314 | .path = strdup(a->path), | |
315 | .permissions = a->permissions, | |
316 | }; | |
84c01612 MS |
317 | if (!n->path) |
318 | return -ENOMEM; | |
319 | ||
84c01612 MS |
320 | LIST_PREPEND(device_allow, c->device_allow, TAKE_PTR(n)); |
321 | return 0; | |
322 | } | |
323 | ||
9c02eb28 | 324 | static int cgroup_context_add_socket_bind_item_dup(CGroupContext *c, const CGroupSocketBindItem *i, CGroupSocketBindItem *h) { |
84c01612 MS |
325 | _cleanup_free_ CGroupSocketBindItem *n = NULL; |
326 | ||
327 | assert(c); | |
328 | assert(i); | |
329 | ||
9c02eb28 | 330 | n = new(CGroupSocketBindItem, 1); |
84c01612 MS |
331 | if (!n) |
332 | return -ENOMEM; | |
333 | ||
334 | *n = (CGroupSocketBindItem) { | |
335 | .address_family = i->address_family, | |
336 | .ip_protocol = i->ip_protocol, | |
337 | .nr_ports = i->nr_ports, | |
338 | .port_min = i->port_min, | |
339 | }; | |
340 | ||
341 | LIST_PREPEND(socket_bind_items, h, TAKE_PTR(n)); | |
342 | return 0; | |
343 | } | |
344 | ||
9c02eb28 | 345 | int cgroup_context_add_socket_bind_item_allow_dup(CGroupContext *c, const CGroupSocketBindItem *i) { |
84c01612 MS |
346 | return cgroup_context_add_socket_bind_item_dup(c, i, c->socket_bind_allow); |
347 | } | |
348 | ||
9c02eb28 | 349 | int cgroup_context_add_socket_bind_item_deny_dup(CGroupContext *c, const CGroupSocketBindItem *i) { |
84c01612 MS |
350 | return cgroup_context_add_socket_bind_item_dup(c, i, c->socket_bind_deny); |
351 | } | |
352 | ||
353 | int cgroup_context_copy(CGroupContext *dst, const CGroupContext *src) { | |
354 | struct in_addr_prefix *i; | |
355 | char *iface; | |
356 | int r; | |
357 | ||
358 | assert(src); | |
359 | assert(dst); | |
360 | ||
361 | dst->cpu_accounting = src->cpu_accounting; | |
362 | dst->io_accounting = src->io_accounting; | |
363 | dst->blockio_accounting = src->blockio_accounting; | |
364 | dst->memory_accounting = src->memory_accounting; | |
365 | dst->tasks_accounting = src->tasks_accounting; | |
366 | dst->ip_accounting = src->ip_accounting; | |
367 | ||
2717d36d | 368 | dst->memory_oom_group = src->memory_oom_group; |
84c01612 MS |
369 | |
370 | dst->cpu_weight = src->cpu_weight; | |
371 | dst->startup_cpu_weight = src->startup_cpu_weight; | |
372 | dst->cpu_quota_per_sec_usec = src->cpu_quota_per_sec_usec; | |
373 | dst->cpu_quota_period_usec = src->cpu_quota_period_usec; | |
374 | ||
375 | dst->cpuset_cpus = src->cpuset_cpus; | |
376 | dst->startup_cpuset_cpus = src->startup_cpuset_cpus; | |
377 | dst->cpuset_mems = src->cpuset_mems; | |
378 | dst->startup_cpuset_mems = src->startup_cpuset_mems; | |
379 | ||
380 | dst->io_weight = src->io_weight; | |
381 | dst->startup_io_weight = src->startup_io_weight; | |
382 | ||
383 | LIST_FOREACH_BACKWARDS(device_weights, w, LIST_FIND_TAIL(device_weights, src->io_device_weights)) { | |
384 | r = cgroup_context_add_io_device_weight_dup(dst, w); | |
385 | if (r < 0) | |
386 | return r; | |
387 | } | |
388 | ||
389 | LIST_FOREACH_BACKWARDS(device_limits, l, LIST_FIND_TAIL(device_limits, src->io_device_limits)) { | |
390 | r = cgroup_context_add_io_device_limit_dup(dst, l); | |
391 | if (r < 0) | |
392 | return r; | |
393 | } | |
394 | ||
395 | LIST_FOREACH_BACKWARDS(device_latencies, l, LIST_FIND_TAIL(device_latencies, src->io_device_latencies)) { | |
396 | r = cgroup_context_add_io_device_latency_dup(dst, l); | |
397 | if (r < 0) | |
398 | return r; | |
399 | } | |
400 | ||
401 | dst->default_memory_min = src->default_memory_min; | |
402 | dst->default_memory_low = src->default_memory_low; | |
403 | dst->default_startup_memory_low = src->default_startup_memory_low; | |
404 | dst->memory_min = src->memory_min; | |
405 | dst->memory_low = src->memory_low; | |
406 | dst->startup_memory_low = src->startup_memory_low; | |
407 | dst->memory_high = src->memory_high; | |
408 | dst->startup_memory_high = src->startup_memory_high; | |
409 | dst->memory_max = src->memory_max; | |
410 | dst->startup_memory_max = src->startup_memory_max; | |
411 | dst->memory_swap_max = src->memory_swap_max; | |
412 | dst->startup_memory_swap_max = src->startup_memory_swap_max; | |
413 | dst->memory_zswap_max = src->memory_zswap_max; | |
414 | dst->startup_memory_zswap_max = src->startup_memory_zswap_max; | |
415 | ||
416 | dst->default_memory_min_set = src->default_memory_min_set; | |
417 | dst->default_memory_low_set = src->default_memory_low_set; | |
418 | dst->default_startup_memory_low_set = src->default_startup_memory_low_set; | |
419 | dst->memory_min_set = src->memory_min_set; | |
420 | dst->memory_low_set = src->memory_low_set; | |
421 | dst->startup_memory_low_set = src->startup_memory_low_set; | |
422 | dst->startup_memory_high_set = src->startup_memory_high_set; | |
423 | dst->startup_memory_max_set = src->startup_memory_max_set; | |
424 | dst->startup_memory_swap_max_set = src->startup_memory_swap_max_set; | |
425 | dst->startup_memory_zswap_max_set = src->startup_memory_zswap_max_set; | |
426 | ||
427 | SET_FOREACH(i, src->ip_address_allow) { | |
428 | r = in_addr_prefix_add(&dst->ip_address_allow, i); | |
429 | if (r < 0) | |
430 | return r; | |
431 | } | |
432 | ||
433 | SET_FOREACH(i, src->ip_address_deny) { | |
434 | r = in_addr_prefix_add(&dst->ip_address_deny, i); | |
435 | if (r < 0) | |
436 | return r; | |
437 | } | |
438 | ||
439 | dst->ip_address_allow_reduced = src->ip_address_allow_reduced; | |
440 | dst->ip_address_deny_reduced = src->ip_address_deny_reduced; | |
441 | ||
442 | if (!strv_isempty(src->ip_filters_ingress)) { | |
443 | dst->ip_filters_ingress = strv_copy(src->ip_filters_ingress); | |
444 | if (!dst->ip_filters_ingress) | |
445 | return -ENOMEM; | |
446 | } | |
447 | ||
448 | if (!strv_isempty(src->ip_filters_egress)) { | |
449 | dst->ip_filters_egress = strv_copy(src->ip_filters_egress); | |
450 | if (!dst->ip_filters_egress) | |
451 | return -ENOMEM; | |
452 | } | |
453 | ||
454 | LIST_FOREACH_BACKWARDS(programs, l, LIST_FIND_TAIL(programs, src->bpf_foreign_programs)) { | |
455 | r = cgroup_context_add_bpf_foreign_program_dup(dst, l); | |
456 | if (r < 0) | |
457 | return r; | |
458 | } | |
459 | ||
460 | SET_FOREACH(iface, src->restrict_network_interfaces) { | |
461 | r = set_put_strdup(&dst->restrict_network_interfaces, iface); | |
462 | if (r < 0) | |
463 | return r; | |
464 | } | |
465 | dst->restrict_network_interfaces_is_allow_list = src->restrict_network_interfaces_is_allow_list; | |
466 | ||
467 | dst->cpu_shares = src->cpu_shares; | |
468 | dst->startup_cpu_shares = src->startup_cpu_shares; | |
469 | ||
470 | dst->blockio_weight = src->blockio_weight; | |
471 | dst->startup_blockio_weight = src->startup_blockio_weight; | |
472 | ||
473 | LIST_FOREACH_BACKWARDS(device_weights, l, LIST_FIND_TAIL(device_weights, src->blockio_device_weights)) { | |
474 | r = cgroup_context_add_block_io_device_weight_dup(dst, l); | |
475 | if (r < 0) | |
476 | return r; | |
477 | } | |
478 | ||
479 | LIST_FOREACH_BACKWARDS(device_bandwidths, l, LIST_FIND_TAIL(device_bandwidths, src->blockio_device_bandwidths)) { | |
480 | r = cgroup_context_add_block_io_device_bandwidth_dup(dst, l); | |
481 | if (r < 0) | |
482 | return r; | |
483 | } | |
484 | ||
485 | dst->memory_limit = src->memory_limit; | |
486 | ||
487 | dst->device_policy = src->device_policy; | |
488 | LIST_FOREACH_BACKWARDS(device_allow, l, LIST_FIND_TAIL(device_allow, src->device_allow)) { | |
489 | r = cgroup_context_add_device_allow_dup(dst, l); | |
490 | if (r < 0) | |
491 | return r; | |
492 | } | |
493 | ||
494 | LIST_FOREACH_BACKWARDS(socket_bind_items, l, LIST_FIND_TAIL(socket_bind_items, src->socket_bind_allow)) { | |
495 | r = cgroup_context_add_socket_bind_item_allow_dup(dst, l); | |
496 | if (r < 0) | |
497 | return r; | |
498 | ||
499 | } | |
500 | ||
501 | LIST_FOREACH_BACKWARDS(socket_bind_items, l, LIST_FIND_TAIL(socket_bind_items, src->socket_bind_deny)) { | |
502 | r = cgroup_context_add_socket_bind_item_deny_dup(dst, l); | |
503 | if (r < 0) | |
504 | return r; | |
505 | } | |
506 | ||
507 | dst->tasks_max = src->tasks_max; | |
508 | ||
509 | return 0; | |
510 | } | |
511 | ||
4ad49000 LP |
512 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
513 | assert(c); | |
514 | assert(a); | |
515 | ||
71fda00f | 516 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
517 | free(a->path); |
518 | free(a); | |
519 | } | |
520 | ||
13c31542 TH |
521 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
522 | assert(c); | |
523 | assert(w); | |
524 | ||
525 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
526 | free(w->path); | |
527 | free(w); | |
528 | } | |
529 | ||
6ae4283c TH |
530 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
531 | assert(c); | |
532 | assert(l); | |
533 | ||
534 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
535 | free(l->path); | |
536 | free(l); | |
537 | } | |
538 | ||
13c31542 TH |
539 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
540 | assert(c); | |
541 | assert(l); | |
542 | ||
543 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
544 | free(l->path); | |
545 | free(l); | |
546 | } | |
547 | ||
4ad49000 LP |
548 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
549 | assert(c); | |
550 | assert(w); | |
551 | ||
71fda00f | 552 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
553 | free(w->path); |
554 | free(w); | |
555 | } | |
556 | ||
557 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
558 | assert(c); | |
8e274523 | 559 | assert(b); |
8e274523 | 560 | |
71fda00f | 561 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
562 | free(b->path); |
563 | free(b); | |
564 | } | |
565 | ||
b894ef1b JK |
566 | void cgroup_context_remove_bpf_foreign_program(CGroupContext *c, CGroupBPFForeignProgram *p) { |
567 | assert(c); | |
568 | assert(p); | |
569 | ||
570 | LIST_REMOVE(programs, c->bpf_foreign_programs, p); | |
571 | free(p->bpffs_path); | |
572 | free(p); | |
573 | } | |
574 | ||
b18e9fc1 | 575 | void cgroup_context_remove_socket_bind(CGroupSocketBindItem **head) { |
b18e9fc1 JK |
576 | assert(head); |
577 | ||
9aad490e | 578 | LIST_CLEAR(socket_bind_items, *head, free); |
b18e9fc1 JK |
579 | } |
580 | ||
4ad49000 LP |
581 | void cgroup_context_done(CGroupContext *c) { |
582 | assert(c); | |
583 | ||
13c31542 TH |
584 | while (c->io_device_weights) |
585 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
586 | ||
6ae4283c TH |
587 | while (c->io_device_latencies) |
588 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
589 | ||
13c31542 TH |
590 | while (c->io_device_limits) |
591 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
592 | ||
4ad49000 LP |
593 | while (c->blockio_device_weights) |
594 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
595 | ||
596 | while (c->blockio_device_bandwidths) | |
597 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
598 | ||
599 | while (c->device_allow) | |
600 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f | 601 | |
b18e9fc1 JK |
602 | cgroup_context_remove_socket_bind(&c->socket_bind_allow); |
603 | cgroup_context_remove_socket_bind(&c->socket_bind_deny); | |
604 | ||
84ebe6f0 YW |
605 | c->ip_address_allow = set_free(c->ip_address_allow); |
606 | c->ip_address_deny = set_free(c->ip_address_deny); | |
fab34748 KL |
607 | |
608 | c->ip_filters_ingress = strv_free(c->ip_filters_ingress); | |
609 | c->ip_filters_egress = strv_free(c->ip_filters_egress); | |
047f5d63 | 610 | |
b894ef1b JK |
611 | while (c->bpf_foreign_programs) |
612 | cgroup_context_remove_bpf_foreign_program(c, c->bpf_foreign_programs); | |
613 | ||
9b412709 | 614 | c->restrict_network_interfaces = set_free_free(c->restrict_network_interfaces); |
6f50d4f7 | 615 | |
047f5d63 | 616 | cpu_set_reset(&c->cpuset_cpus); |
31d3a520 | 617 | cpu_set_reset(&c->startup_cpuset_cpus); |
047f5d63 | 618 | cpu_set_reset(&c->cpuset_mems); |
31d3a520 | 619 | cpu_set_reset(&c->startup_cpuset_mems); |
a8b993dc LP |
620 | |
621 | c->delegate_subgroup = mfree(c->delegate_subgroup); | |
dc7d69b3 TM |
622 | |
623 | nft_set_context_clear(&c->nft_set_context); | |
4ad49000 LP |
624 | } |
625 | ||
74b5fb27 | 626 | static int unit_get_kernel_memory_limit(Unit *u, const char *file, uint64_t *ret) { |
74b5fb27 CD |
627 | assert(u); |
628 | ||
9cc54544 LP |
629 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
630 | if (!crt || !crt->cgroup_path) | |
74b5fb27 CD |
631 | return -EOWNERDEAD; |
632 | ||
9cc54544 | 633 | return cg_get_attribute_as_uint64("memory", crt->cgroup_path, file, ret); |
74b5fb27 CD |
634 | } |
635 | ||
636 | static int unit_compare_memory_limit(Unit *u, const char *property_name, uint64_t *ret_unit_value, uint64_t *ret_kernel_value) { | |
637 | CGroupContext *c; | |
638 | CGroupMask m; | |
639 | const char *file; | |
640 | uint64_t unit_value; | |
641 | int r; | |
642 | ||
643 | /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will | |
644 | * return -ENODATA) on cgroup v1. | |
645 | * | |
646 | * Returns: | |
647 | * | |
648 | * <0: On error. | |
649 | * 0: If the kernel memory setting doesn't match our configuration. | |
650 | * >0: If the kernel memory setting matches our configuration. | |
651 | * | |
652 | * The following values are only guaranteed to be populated on return >=0: | |
653 | * | |
654 | * - ret_unit_value will contain our internal expected value for the unit, page-aligned. | |
655 | * - ret_kernel_value will contain the actual value presented by the kernel. */ | |
656 | ||
657 | assert(u); | |
658 | ||
659 | r = cg_all_unified(); | |
660 | if (r < 0) | |
661 | return log_debug_errno(r, "Failed to determine cgroup hierarchy version: %m"); | |
662 | ||
663 | /* Unsupported on v1. | |
664 | * | |
665 | * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has | |
666 | * silently masked the controller. */ | |
667 | if (r == 0) | |
668 | return -ENODATA; | |
669 | ||
670 | /* The root slice doesn't have any controller files, so we can't compare anything. */ | |
671 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
672 | return -ENODATA; | |
673 | ||
674 | /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled, | |
675 | * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To | |
676 | * avoid specious errors in these scenarios, check that we even expect the memory controller to be | |
677 | * enabled at all. */ | |
678 | m = unit_get_target_mask(u); | |
679 | if (!FLAGS_SET(m, CGROUP_MASK_MEMORY)) | |
680 | return -ENODATA; | |
681 | ||
806a9362 | 682 | assert_se(c = unit_get_cgroup_context(u)); |
74b5fb27 | 683 | |
53fda560 LB |
684 | bool startup = u->manager && IN_SET(manager_state(u->manager), MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING); |
685 | ||
74b5fb27 CD |
686 | if (streq(property_name, "MemoryLow")) { |
687 | unit_value = unit_get_ancestor_memory_low(u); | |
688 | file = "memory.low"; | |
53fda560 LB |
689 | } else if (startup && streq(property_name, "StartupMemoryLow")) { |
690 | unit_value = unit_get_ancestor_startup_memory_low(u); | |
691 | file = "memory.low"; | |
74b5fb27 CD |
692 | } else if (streq(property_name, "MemoryMin")) { |
693 | unit_value = unit_get_ancestor_memory_min(u); | |
694 | file = "memory.min"; | |
695 | } else if (streq(property_name, "MemoryHigh")) { | |
696 | unit_value = c->memory_high; | |
697 | file = "memory.high"; | |
53fda560 LB |
698 | } else if (startup && streq(property_name, "StartupMemoryHigh")) { |
699 | unit_value = c->startup_memory_high; | |
700 | file = "memory.high"; | |
74b5fb27 CD |
701 | } else if (streq(property_name, "MemoryMax")) { |
702 | unit_value = c->memory_max; | |
703 | file = "memory.max"; | |
53fda560 LB |
704 | } else if (startup && streq(property_name, "StartupMemoryMax")) { |
705 | unit_value = c->startup_memory_max; | |
706 | file = "memory.max"; | |
74b5fb27 CD |
707 | } else if (streq(property_name, "MemorySwapMax")) { |
708 | unit_value = c->memory_swap_max; | |
709 | file = "memory.swap.max"; | |
53fda560 LB |
710 | } else if (startup && streq(property_name, "StartupMemorySwapMax")) { |
711 | unit_value = c->startup_memory_swap_max; | |
712 | file = "memory.swap.max"; | |
d7fe0a67 PV |
713 | } else if (streq(property_name, "MemoryZSwapMax")) { |
714 | unit_value = c->memory_zswap_max; | |
715 | file = "memory.zswap.max"; | |
53fda560 LB |
716 | } else if (startup && streq(property_name, "StartupMemoryZSwapMax")) { |
717 | unit_value = c->startup_memory_zswap_max; | |
718 | file = "memory.zswap.max"; | |
74b5fb27 CD |
719 | } else |
720 | return -EINVAL; | |
721 | ||
722 | r = unit_get_kernel_memory_limit(u, file, ret_kernel_value); | |
723 | if (r < 0) | |
724 | return log_unit_debug_errno(u, r, "Failed to parse %s: %m", file); | |
725 | ||
726 | /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page | |
727 | * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from | |
728 | * our internal page-counting. To support those future kernels, just check the value itself first | |
729 | * without any page-alignment. */ | |
730 | if (*ret_kernel_value == unit_value) { | |
731 | *ret_unit_value = unit_value; | |
732 | return 1; | |
733 | } | |
734 | ||
735 | /* The current kernel behaviour, by comparison, is that even if you write a particular number of | |
736 | * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel | |
737 | * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is | |
738 | * cricket. */ | |
739 | if (unit_value != CGROUP_LIMIT_MAX) | |
740 | unit_value = PAGE_ALIGN_DOWN(unit_value); | |
741 | ||
742 | *ret_unit_value = unit_value; | |
743 | ||
744 | return *ret_kernel_value == *ret_unit_value; | |
745 | } | |
746 | ||
bc0623df CD |
747 | #define FORMAT_CGROUP_DIFF_MAX 128 |
748 | ||
3f236f24 | 749 | static char *format_cgroup_memory_limit_comparison(Unit *u, const char *property_name, char *buf, size_t l) { |
bc0623df CD |
750 | uint64_t kval, sval; |
751 | int r; | |
752 | ||
753 | assert(u); | |
3f236f24 | 754 | assert(property_name); |
bc0623df CD |
755 | assert(buf); |
756 | assert(l > 0); | |
757 | ||
758 | r = unit_compare_memory_limit(u, property_name, &sval, &kval); | |
759 | ||
760 | /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1). | |
761 | * In the absence of reliably being able to detect whether memcg swap support is available or not, | |
d7fe0a67 PV |
762 | * only complain if the error is not ENOENT. This is similarly the case for memory.zswap.max relying |
763 | * on CONFIG_ZSWAP. */ | |
bc0623df | 764 | if (r > 0 || IN_SET(r, -ENODATA, -EOWNERDEAD) || |
53fda560 LB |
765 | (r == -ENOENT && STR_IN_SET(property_name, |
766 | "MemorySwapMax", | |
767 | "StartupMemorySwapMax", | |
768 | "MemoryZSwapMax", | |
769 | "StartupMemoryZSwapMax"))) | |
bc0623df | 770 | buf[0] = 0; |
38553034 ZJS |
771 | else if (r < 0) { |
772 | errno = -r; | |
773 | (void) snprintf(buf, l, " (error getting kernel value: %m)"); | |
774 | } else | |
775 | (void) snprintf(buf, l, " (different value in kernel: %" PRIu64 ")", kval); | |
bc0623df CD |
776 | |
777 | return buf; | |
778 | } | |
779 | ||
a1044811 LP |
780 | const char *cgroup_device_permissions_to_string(CGroupDevicePermissions p) { |
781 | static const char *table[_CGROUP_DEVICE_PERMISSIONS_MAX] = { | |
782 | /* Lets simply define a table with every possible combination. As long as those are just 8 we | |
783 | * can get away with it. If this ever grows to more we need to revisit this logic though. */ | |
784 | [0] = "", | |
785 | [CGROUP_DEVICE_READ] = "r", | |
786 | [CGROUP_DEVICE_WRITE] = "w", | |
787 | [CGROUP_DEVICE_MKNOD] = "m", | |
788 | [CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE] = "rw", | |
789 | [CGROUP_DEVICE_READ|CGROUP_DEVICE_MKNOD] = "rm", | |
790 | [CGROUP_DEVICE_WRITE|CGROUP_DEVICE_MKNOD] = "wm", | |
791 | [CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE|CGROUP_DEVICE_MKNOD] = "rwm", | |
792 | }; | |
793 | ||
794 | if (p < 0 || p >= _CGROUP_DEVICE_PERMISSIONS_MAX) | |
795 | return NULL; | |
796 | ||
797 | return table[p]; | |
798 | } | |
799 | ||
800 | CGroupDevicePermissions cgroup_device_permissions_from_string(const char *s) { | |
801 | CGroupDevicePermissions p = 0; | |
802 | ||
803 | if (!s) | |
804 | return _CGROUP_DEVICE_PERMISSIONS_INVALID; | |
805 | ||
806 | for (const char *c = s; *c; c++) { | |
807 | if (*c == 'r') | |
808 | p |= CGROUP_DEVICE_READ; | |
809 | else if (*c == 'w') | |
810 | p |= CGROUP_DEVICE_WRITE; | |
811 | else if (*c == 'm') | |
812 | p |= CGROUP_DEVICE_MKNOD; | |
813 | else | |
814 | return _CGROUP_DEVICE_PERMISSIONS_INVALID; | |
815 | } | |
816 | ||
817 | return p; | |
818 | } | |
819 | ||
bc0623df | 820 | void cgroup_context_dump(Unit *u, FILE* f, const char *prefix) { |
7b3693e4 | 821 | _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 | 822 | CGroupContext *c; |
84ebe6f0 | 823 | struct in_addr_prefix *iaai; |
3f236f24 LP |
824 | char cda[FORMAT_CGROUP_DIFF_MAX], cdb[FORMAT_CGROUP_DIFF_MAX], cdc[FORMAT_CGROUP_DIFF_MAX], cdd[FORMAT_CGROUP_DIFF_MAX], |
825 | cde[FORMAT_CGROUP_DIFF_MAX], cdf[FORMAT_CGROUP_DIFF_MAX], cdg[FORMAT_CGROUP_DIFF_MAX], cdh[FORMAT_CGROUP_DIFF_MAX], | |
826 | cdi[FORMAT_CGROUP_DIFF_MAX], cdj[FORMAT_CGROUP_DIFF_MAX], 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, |
3f236f24 LP |
913 | prefix, c->memory_min, format_cgroup_memory_limit_comparison(u, "MemoryMin", cda, sizeof(cda)), |
914 | prefix, c->memory_low, format_cgroup_memory_limit_comparison(u, "MemoryLow", cdb, sizeof(cdb)), | |
915 | prefix, c->startup_memory_low, format_cgroup_memory_limit_comparison(u, "StartupMemoryLow", cdc, sizeof(cdc)), | |
916 | prefix, c->memory_high, format_cgroup_memory_limit_comparison(u, "MemoryHigh", cdd, sizeof(cdd)), | |
917 | prefix, c->startup_memory_high, format_cgroup_memory_limit_comparison(u, "StartupMemoryHigh", cde, sizeof(cde)), | |
918 | prefix, c->memory_max, format_cgroup_memory_limit_comparison(u, "MemoryMax", cdf, sizeof(cdf)), | |
919 | prefix, c->startup_memory_max, format_cgroup_memory_limit_comparison(u, "StartupMemoryMax", cdg, sizeof(cdg)), | |
920 | prefix, c->memory_swap_max, format_cgroup_memory_limit_comparison(u, "MemorySwapMax", cdh, sizeof(cdh)), | |
921 | prefix, c->startup_memory_swap_max, format_cgroup_memory_limit_comparison(u, "StartupMemorySwapMax", cdi, sizeof(cdi)), | |
922 | prefix, c->memory_zswap_max, format_cgroup_memory_limit_comparison(u, "MemoryZSwapMax", cdj, sizeof(cdj)), | |
923 | prefix, c->startup_memory_zswap_max, format_cgroup_memory_limit_comparison(u, "StartupMemoryZSwapMax", cdk, sizeof(cdk)), | |
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 | ||
9cc54544 LP |
1183 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
1184 | if (!crt || !crt->cgroup_path) | |
17d047f5 | 1185 | return; |
1fa3b6c2 | 1186 | |
9cc54544 | 1187 | r = cg_set_xattr(crt->cgroup_path, name, data, size, 0); |
1fa3b6c2 | 1188 | if (r < 0) |
9cc54544 | 1189 | log_unit_debug_errno(u, r, "Failed to set '%s' xattr on control group %s, ignoring: %m", name, empty_to_root(crt->cgroup_path)); |
1fa3b6c2 LP |
1190 | } |
1191 | ||
17d047f5 | 1192 | static void unit_remove_xattr_graceful(Unit *u, const char *name) { |
1fa3b6c2 LP |
1193 | int r; |
1194 | ||
1195 | assert(u); | |
1196 | assert(name); | |
1197 | ||
9cc54544 LP |
1198 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
1199 | if (!crt || !crt->cgroup_path) | |
17d047f5 | 1200 | return; |
1fa3b6c2 | 1201 | |
9cc54544 | 1202 | r = cg_remove_xattr(crt->cgroup_path, name); |
00675c36 | 1203 | if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r)) |
9cc54544 | 1204 | log_unit_debug_errno(u, r, "Failed to remove '%s' xattr flag on control group %s, ignoring: %m", name, empty_to_root(crt->cgroup_path)); |
1fa3b6c2 LP |
1205 | } |
1206 | ||
64c71f4f | 1207 | static void cgroup_oomd_xattr_apply(Unit *u) { |
4e806bfa | 1208 | CGroupContext *c; |
4e806bfa AZ |
1209 | |
1210 | assert(u); | |
1211 | ||
1212 | c = unit_get_cgroup_context(u); | |
1213 | if (!c) | |
1214 | return; | |
1215 | ||
1fa3b6c2 | 1216 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_OMIT) |
17d047f5 | 1217 | unit_set_xattr_graceful(u, "user.oomd_omit", "1", 1); |
4e806bfa | 1218 | |
1fa3b6c2 | 1219 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_AVOID) |
17d047f5 | 1220 | unit_set_xattr_graceful(u, "user.oomd_avoid", "1", 1); |
4e806bfa | 1221 | |
1fa3b6c2 | 1222 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_AVOID) |
17d047f5 | 1223 | unit_remove_xattr_graceful(u, "user.oomd_avoid"); |
4e806bfa | 1224 | |
1fa3b6c2 | 1225 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_OMIT) |
17d047f5 | 1226 | unit_remove_xattr_graceful(u, "user.oomd_omit"); |
4e806bfa AZ |
1227 | } |
1228 | ||
64c71f4f | 1229 | static int cgroup_log_xattr_apply(Unit *u) { |
523ea123 QD |
1230 | ExecContext *c; |
1231 | size_t len, allowed_patterns_len, denied_patterns_len; | |
1232 | _cleanup_free_ char *patterns = NULL, *allowed_patterns = NULL, *denied_patterns = NULL; | |
48d85160 | 1233 | char *last; |
523ea123 QD |
1234 | int r; |
1235 | ||
1236 | assert(u); | |
1237 | ||
1238 | c = unit_get_exec_context(u); | |
1239 | if (!c) | |
1240 | /* Some unit types have a cgroup context but no exec context, so we do not log | |
1241 | * any error here to avoid confusion. */ | |
1242 | return 0; | |
1243 | ||
1244 | if (set_isempty(c->log_filter_allowed_patterns) && set_isempty(c->log_filter_denied_patterns)) { | |
17d047f5 | 1245 | unit_remove_xattr_graceful(u, "user.journald_log_filter_patterns"); |
523ea123 QD |
1246 | return 0; |
1247 | } | |
1248 | ||
1249 | r = set_make_nulstr(c->log_filter_allowed_patterns, &allowed_patterns, &allowed_patterns_len); | |
1250 | if (r < 0) | |
1251 | return log_debug_errno(r, "Failed to make nulstr from set: %m"); | |
1252 | ||
1253 | r = set_make_nulstr(c->log_filter_denied_patterns, &denied_patterns, &denied_patterns_len); | |
1254 | if (r < 0) | |
1255 | return log_debug_errno(r, "Failed to make nulstr from set: %m"); | |
1256 | ||
1257 | /* Use nul character separated strings without trailing nul */ | |
1258 | allowed_patterns_len = LESS_BY(allowed_patterns_len, 1u); | |
1259 | denied_patterns_len = LESS_BY(denied_patterns_len, 1u); | |
1260 | ||
1261 | len = allowed_patterns_len + 1 + denied_patterns_len; | |
1262 | patterns = new(char, len); | |
1263 | if (!patterns) | |
1264 | return log_oom_debug(); | |
1265 | ||
48d85160 QD |
1266 | last = mempcpy_safe(patterns, allowed_patterns, allowed_patterns_len); |
1267 | *(last++) = '\xff'; | |
1268 | memcpy_safe(last, denied_patterns, denied_patterns_len); | |
523ea123 | 1269 | |
17d047f5 | 1270 | unit_set_xattr_graceful(u, "user.journald_log_filter_patterns", patterns, len); |
523ea123 QD |
1271 | |
1272 | return 0; | |
1273 | } | |
1274 | ||
d46510de | 1275 | static void cgroup_invocation_id_xattr_apply(Unit *u) { |
d9bc1c36 | 1276 | bool b; |
0d2d6fbf CD |
1277 | |
1278 | assert(u); | |
1279 | ||
1fa3b6c2 LP |
1280 | b = !sd_id128_is_null(u->invocation_id); |
1281 | FOREACH_STRING(xn, "trusted.invocation_id", "user.invocation_id") { | |
1282 | if (b) | |
17d047f5 | 1283 | unit_set_xattr_graceful(u, xn, SD_ID128_TO_STRING(u->invocation_id), 32); |
1fa3b6c2 | 1284 | else |
17d047f5 | 1285 | unit_remove_xattr_graceful(u, xn); |
3288ea8f | 1286 | } |
d46510de LP |
1287 | } |
1288 | ||
6cf96ab4 NR |
1289 | static void cgroup_coredump_xattr_apply(Unit *u) { |
1290 | CGroupContext *c; | |
1291 | ||
1292 | assert(u); | |
1293 | ||
1294 | c = unit_get_cgroup_context(u); | |
1295 | if (!c) | |
1296 | return; | |
1297 | ||
1298 | if (unit_cgroup_delegate(u) && c->coredump_receive) | |
1299 | unit_set_xattr_graceful(u, "user.coredump_receive", "1", 1); | |
1300 | else | |
1301 | unit_remove_xattr_graceful(u, "user.coredump_receive"); | |
1302 | } | |
1303 | ||
d46510de LP |
1304 | static void cgroup_delegate_xattr_apply(Unit *u) { |
1305 | bool b; | |
1306 | ||
1307 | assert(u); | |
0d2d6fbf | 1308 | |
d9bc1c36 LP |
1309 | /* Indicate on the cgroup whether delegation is on, via an xattr. This is best-effort, as old kernels |
1310 | * didn't support xattrs on cgroups at all. Later they got support for setting 'trusted.*' xattrs, | |
1311 | * and even later 'user.*' xattrs. We started setting this field when 'trusted.*' was added, and | |
1312 | * given this is now pretty much API, let's continue to support that. But also set 'user.*' as well, | |
1313 | * since it is readable by any user, not just CAP_SYS_ADMIN. This hence comes with slightly weaker | |
1314 | * security (as users who got delegated cgroups could turn it off if they like), but this shouldn't | |
1315 | * be a big problem given this communicates delegation state to clients, but the manager never reads | |
1316 | * it. */ | |
1317 | b = unit_cgroup_delegate(u); | |
1318 | FOREACH_STRING(xn, "trusted.delegate", "user.delegate") { | |
1fa3b6c2 | 1319 | if (b) |
17d047f5 | 1320 | unit_set_xattr_graceful(u, xn, "1", 1); |
1fa3b6c2 | 1321 | else |
17d047f5 | 1322 | unit_remove_xattr_graceful(u, xn); |
3288ea8f | 1323 | } |
d46510de LP |
1324 | } |
1325 | ||
1326 | static void cgroup_survive_xattr_apply(Unit *u) { | |
1327 | int r; | |
1328 | ||
1329 | assert(u); | |
559214cb | 1330 | |
9cc54544 LP |
1331 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
1332 | if (!crt) | |
1333 | return; | |
1334 | ||
559214cb | 1335 | if (u->survive_final_kill_signal) { |
bd1791b5 | 1336 | r = cg_set_xattr( |
9cc54544 | 1337 | crt->cgroup_path, |
bd1791b5 LP |
1338 | "user.survive_final_kill_signal", |
1339 | "1", | |
1340 | 1, | |
1341 | /* flags= */ 0); | |
559214cb LB |
1342 | /* user xattr support was added in kernel v5.7 */ |
1343 | if (ERRNO_IS_NEG_NOT_SUPPORTED(r)) | |
bd1791b5 | 1344 | r = cg_set_xattr( |
9cc54544 | 1345 | crt->cgroup_path, |
559214cb LB |
1346 | "trusted.survive_final_kill_signal", |
1347 | "1", | |
1348 | 1, | |
1349 | /* flags= */ 0); | |
1350 | if (r < 0) | |
1351 | log_unit_debug_errno(u, | |
1352 | r, | |
1353 | "Failed to set 'survive_final_kill_signal' xattr on control " | |
1354 | "group %s, ignoring: %m", | |
9cc54544 | 1355 | empty_to_root(crt->cgroup_path)); |
559214cb | 1356 | } else { |
17d047f5 LP |
1357 | unit_remove_xattr_graceful(u, "user.survive_final_kill_signal"); |
1358 | unit_remove_xattr_graceful(u, "trusted.survive_final_kill_signal"); | |
559214cb | 1359 | } |
0d2d6fbf CD |
1360 | } |
1361 | ||
d46510de LP |
1362 | static void cgroup_xattr_apply(Unit *u) { |
1363 | assert(u); | |
1364 | ||
1365 | /* The 'user.*' xattrs can be set from a user manager. */ | |
1366 | cgroup_oomd_xattr_apply(u); | |
1367 | cgroup_log_xattr_apply(u); | |
6cf96ab4 | 1368 | cgroup_coredump_xattr_apply(u); |
d46510de LP |
1369 | |
1370 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1371 | return; | |
1372 | ||
1373 | cgroup_invocation_id_xattr_apply(u); | |
1374 | cgroup_delegate_xattr_apply(u); | |
1375 | cgroup_survive_xattr_apply(u); | |
1376 | } | |
1377 | ||
45c2e068 | 1378 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
1379 | dev_t rdev, dev = 0; |
1380 | mode_t mode; | |
45c2e068 | 1381 | int r; |
4ad49000 LP |
1382 | |
1383 | assert(p); | |
45c2e068 | 1384 | assert(ret); |
4ad49000 | 1385 | |
f5855697 | 1386 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 1387 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 1388 | struct stat st; |
57f1030b | 1389 | |
d5aecba6 LP |
1390 | if (stat(p, &st) < 0) |
1391 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
57f1030b | 1392 | |
f5855697 | 1393 | mode = st.st_mode; |
a0d6590c LP |
1394 | rdev = st.st_rdev; |
1395 | dev = st.st_dev; | |
d5aecba6 LP |
1396 | } else if (r < 0) |
1397 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
1398 | ||
57f1030b LP |
1399 | if (S_ISCHR(mode)) |
1400 | return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK), | |
1401 | "Device node '%s' is a character device, but block device needed.", p); | |
1402 | if (S_ISBLK(mode)) | |
f5855697 YS |
1403 | *ret = rdev; |
1404 | else if (major(dev) != 0) | |
1405 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
1406 | else { |
1407 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
1408 | r = btrfs_get_block_device(p, ret); | |
57f1030b LP |
1409 | if (r == -ENOTTY) |
1410 | return log_warning_errno(SYNTHETIC_ERRNO(ENODEV), | |
1411 | "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
1412 | if (r < 0) | |
45c2e068 | 1413 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); |
4ad49000 | 1414 | } |
8e274523 | 1415 | |
b7cf4b4e BB |
1416 | /* If this is a LUKS/DM device, recursively try to get the originating block device */ |
1417 | while (block_get_originating(*ret, ret) > 0); | |
45c2e068 LP |
1418 | |
1419 | /* If this is a partition, try to get the originating block device */ | |
1420 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 1421 | return 0; |
8e274523 LP |
1422 | } |
1423 | ||
66ebf6c0 TH |
1424 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
1425 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
1426 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
1427 | } | |
1428 | ||
1429 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
1430 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
1431 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
1432 | } | |
1433 | ||
31d3a520 PM |
1434 | static bool cgroup_context_has_allowed_cpus(CGroupContext *c) { |
1435 | return c->cpuset_cpus.set || c->startup_cpuset_cpus.set; | |
1436 | } | |
1437 | ||
1438 | static bool cgroup_context_has_allowed_mems(CGroupContext *c) { | |
1439 | return c->cpuset_mems.set || c->startup_cpuset_mems.set; | |
1440 | } | |
1441 | ||
a8157796 LP |
1442 | uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { |
1443 | assert(c); | |
1444 | ||
9dfb6a3a | 1445 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
66ebf6c0 TH |
1446 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) |
1447 | return c->startup_cpu_weight; | |
1448 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
1449 | return c->cpu_weight; | |
1450 | else | |
1451 | return CGROUP_WEIGHT_DEFAULT; | |
1452 | } | |
1453 | ||
1454 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
9dfb6a3a | 1455 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
66ebf6c0 TH |
1456 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) |
1457 | return c->startup_cpu_shares; | |
1458 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
1459 | return c->cpu_shares; | |
1460 | else | |
1461 | return CGROUP_CPU_SHARES_DEFAULT; | |
1462 | } | |
1463 | ||
31d3a520 | 1464 | static CPUSet *cgroup_context_allowed_cpus(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1465 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
31d3a520 PM |
1466 | c->startup_cpuset_cpus.set) |
1467 | return &c->startup_cpuset_cpus; | |
1468 | else | |
1469 | return &c->cpuset_cpus; | |
1470 | } | |
1471 | ||
1472 | static CPUSet *cgroup_context_allowed_mems(CGroupContext *c, ManagerState state) { | |
9dfb6a3a | 1473 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
31d3a520 PM |
1474 | c->startup_cpuset_mems.set) |
1475 | return &c->startup_cpuset_mems; | |
1476 | else | |
1477 | return &c->cpuset_mems; | |
1478 | } | |
1479 | ||
10f28641 FB |
1480 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
1481 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
1482 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
1483 | * Additionally, period must be at most max_period. */ | |
1484 | assert(quota > 0); | |
1485 | ||
1486 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
1487 | } | |
1488 | ||
1489 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
1490 | usec_t new_period; | |
1491 | ||
9cc54544 LP |
1492 | assert(u); |
1493 | ||
1494 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
1495 | if (!crt) | |
1496 | return USEC_INFINITY; | |
1497 | ||
10f28641 FB |
1498 | if (quota == USEC_INFINITY) |
1499 | /* Always use default period for infinity quota. */ | |
1500 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
1501 | ||
1502 | if (period == USEC_INFINITY) | |
1503 | /* Default period was requested. */ | |
1504 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
1505 | ||
1506 | /* Clamp to interval [1ms, 1s] */ | |
1507 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
1508 | ||
1509 | if (new_period != period) { | |
9cc54544 | 1510 | log_unit_full(u, crt->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, |
10f28641 | 1511 | "Clamping CPU interval for cpu.max: period is now %s", |
5291f26d | 1512 | FORMAT_TIMESPAN(new_period, 1)); |
9cc54544 | 1513 | crt->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
1514 | } |
1515 | ||
1516 | return new_period; | |
1517 | } | |
1518 | ||
52fecf20 LP |
1519 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
1520 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 | 1521 | |
c8340822 | 1522 | if (weight == CGROUP_WEIGHT_IDLE) |
1523 | return; | |
66ebf6c0 | 1524 | xsprintf(buf, "%" PRIu64 "\n", weight); |
293d32df | 1525 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
1526 | } |
1527 | ||
c8340822 | 1528 | static void cgroup_apply_unified_cpu_idle(Unit *u, uint64_t weight) { |
1529 | int r; | |
1530 | bool is_idle; | |
1531 | const char *idle_val; | |
1532 | ||
9cc54544 LP |
1533 | assert(u); |
1534 | ||
1535 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
1536 | if (!crt || !crt->cgroup_path) | |
1537 | return; | |
1538 | ||
c8340822 | 1539 | is_idle = weight == CGROUP_WEIGHT_IDLE; |
1540 | idle_val = one_zero(is_idle); | |
9cc54544 | 1541 | r = cg_set_attribute("cpu", crt->cgroup_path, "cpu.idle", idle_val); |
c8340822 | 1542 | if (r < 0 && (r != -ENOENT || is_idle)) |
1543 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%s': %m", | |
9cc54544 | 1544 | "cpu.idle", empty_to_root(crt->cgroup_path), idle_val); |
c8340822 | 1545 | } |
1546 | ||
10f28641 | 1547 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 1548 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 1549 | |
9cc54544 LP |
1550 | assert(u); |
1551 | ||
10f28641 | 1552 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
1553 | if (quota != USEC_INFINITY) |
1554 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 1555 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 1556 | else |
10f28641 | 1557 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 1558 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
1559 | } |
1560 | ||
52fecf20 LP |
1561 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
1562 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
1563 | |
1564 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 1565 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
1566 | } |
1567 | ||
10f28641 | 1568 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 1569 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 1570 | |
10f28641 FB |
1571 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
1572 | ||
1573 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 1574 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
1575 | |
1576 | if (quota != USEC_INFINITY) { | |
10f28641 | 1577 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 1578 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 1579 | } else |
589a5f7a | 1580 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
1581 | } |
1582 | ||
1583 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
1584 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
1585 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
1586 | } | |
1587 | ||
1588 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
c8340822 | 1589 | /* we don't support idle in cgroupv1 */ |
1590 | if (weight == CGROUP_WEIGHT_IDLE) | |
1591 | return CGROUP_CPU_SHARES_MIN; | |
1592 | ||
66ebf6c0 TH |
1593 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
1594 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
1595 | } | |
1596 | ||
2cea199e | 1597 | static void cgroup_apply_unified_cpuset(Unit *u, const CPUSet *cpus, const char *name) { |
047f5d63 PH |
1598 | _cleanup_free_ char *buf = NULL; |
1599 | ||
2cea199e | 1600 | buf = cpu_set_to_range_string(cpus); |
c259ac9a LP |
1601 | if (!buf) { |
1602 | log_oom(); | |
1603 | return; | |
1604 | } | |
047f5d63 PH |
1605 | |
1606 | (void) set_attribute_and_warn(u, "cpuset", name, buf); | |
1607 | } | |
1608 | ||
508c45da | 1609 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
1610 | return c->io_accounting || |
1611 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
1612 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
1613 | c->io_device_weights || | |
6ae4283c | 1614 | c->io_device_latencies || |
538b4852 TH |
1615 | c->io_device_limits; |
1616 | } | |
1617 | ||
508c45da | 1618 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
1619 | return c->blockio_accounting || |
1620 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
1621 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
1622 | c->blockio_device_weights || | |
1623 | c->blockio_device_bandwidths; | |
1624 | } | |
1625 | ||
508c45da | 1626 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1627 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
64faf04c TH |
1628 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) |
1629 | return c->startup_io_weight; | |
d38655d7 | 1630 | if (c->io_weight != CGROUP_WEIGHT_INVALID) |
64faf04c | 1631 | return c->io_weight; |
d38655d7 | 1632 | return CGROUP_WEIGHT_DEFAULT; |
64faf04c TH |
1633 | } |
1634 | ||
508c45da | 1635 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1636 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
64faf04c TH |
1637 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) |
1638 | return c->startup_blockio_weight; | |
d38655d7 | 1639 | if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) |
64faf04c | 1640 | return c->blockio_weight; |
d38655d7 | 1641 | return CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1642 | } |
1643 | ||
508c45da | 1644 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
1645 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
1646 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
1647 | } | |
1648 | ||
508c45da | 1649 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
1650 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
1651 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
1652 | } | |
1653 | ||
3e6eafdd | 1654 | static int set_bfq_weight(Unit *u, const char *controller, dev_t dev, uint64_t io_weight) { |
1cf4a685 MK |
1655 | static const char * const prop_names[] = { |
1656 | "IOWeight", | |
1657 | "BlockIOWeight", | |
1658 | "IODeviceWeight", | |
1659 | "BlockIODeviceWeight", | |
1660 | }; | |
8d75f60e | 1661 | static bool warned = false; |
9f0c0c4e | 1662 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+STRLEN("\n")]; |
bec17e80 MK |
1663 | const char *p; |
1664 | uint64_t bfq_weight; | |
8d75f60e | 1665 | int r; |
bec17e80 | 1666 | |
9cc54544 LP |
1667 | assert(u); |
1668 | ||
1669 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
1670 | if (!crt || !crt->cgroup_path) | |
1671 | return -EOWNERDEAD; | |
1672 | ||
bec17e80 MK |
1673 | /* FIXME: drop this function when distro kernels properly support BFQ through "io.weight" |
1674 | * See also: https://github.com/systemd/systemd/pull/13335 and | |
1675 | * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9. */ | |
1676 | p = strjoina(controller, ".bfq.weight"); | |
1677 | /* Adjust to kernel range is 1..1000, the default is 100. */ | |
1678 | bfq_weight = BFQ_WEIGHT(io_weight); | |
1679 | ||
9f0c0c4e | 1680 | if (major(dev) > 0) |
ec61371f | 1681 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), bfq_weight); |
9f0c0c4e MK |
1682 | else |
1683 | xsprintf(buf, "%" PRIu64 "\n", bfq_weight); | |
bec17e80 | 1684 | |
9cc54544 | 1685 | r = cg_set_attribute(controller, crt->cgroup_path, p, buf); |
8d75f60e MK |
1686 | |
1687 | /* FIXME: drop this when kernels prior | |
1688 | * 795fe54c2a82 ("bfq: Add per-device weight") v5.4 | |
1689 | * are not interesting anymore. Old kernels will fail with EINVAL, while new kernels won't return | |
1690 | * EINVAL on properly formatted input by us. Treat EINVAL accordingly. */ | |
3e6eafdd MK |
1691 | if (r == -EINVAL && major(dev) > 0) { |
1692 | if (!warned) { | |
1693 | log_unit_warning(u, "Kernel version does not accept per-device setting in %s.", p); | |
1694 | warned = true; | |
1695 | } | |
1696 | r = -EOPNOTSUPP; /* mask as unconfigured device */ | |
1697 | } else if (r >= 0 && io_weight != bfq_weight) | |
1cf4a685 MK |
1698 | log_unit_debug(u, "%s=%" PRIu64 " scaled to %s=%" PRIu64, |
1699 | prop_names[2*(major(dev) > 0) + streq(controller, "blkio")], | |
bec17e80 | 1700 | io_weight, p, bfq_weight); |
3e6eafdd | 1701 | return r; |
bec17e80 MK |
1702 | } |
1703 | ||
f29ff115 | 1704 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
1705 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1706 | dev_t dev; | |
3e6eafdd | 1707 | int r, r1, r2; |
64faf04c | 1708 | |
9cc54544 LP |
1709 | assert(u); |
1710 | ||
1711 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
1712 | if (!crt || !crt->cgroup_path) | |
1713 | return; | |
1714 | ||
3e6eafdd | 1715 | if (lookup_block_device(dev_path, &dev) < 0) |
64faf04c TH |
1716 | return; |
1717 | ||
3e6eafdd | 1718 | r1 = set_bfq_weight(u, "io", dev, io_weight); |
9f0c0c4e | 1719 | |
ec61371f | 1720 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), io_weight); |
9cc54544 | 1721 | r2 = cg_set_attribute("io", crt->cgroup_path, "io.weight", buf); |
3e6eafdd MK |
1722 | |
1723 | /* Look at the configured device, when both fail, prefer io.weight errno. */ | |
1724 | r = r2 == -EOPNOTSUPP ? r1 : r2; | |
1725 | ||
1726 | if (r < 0) | |
1727 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), | |
1728 | r, "Failed to set 'io[.bfq].weight' attribute on '%s' to '%.*s': %m", | |
9cc54544 | 1729 | empty_to_root(crt->cgroup_path), (int) strcspn(buf, NEWLINE), buf); |
64faf04c TH |
1730 | } |
1731 | ||
f29ff115 | 1732 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
1733 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1734 | dev_t dev; | |
1735 | int r; | |
1736 | ||
1737 | r = lookup_block_device(dev_path, &dev); | |
1738 | if (r < 0) | |
1739 | return; | |
1740 | ||
ec61371f | 1741 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), blkio_weight); |
293d32df | 1742 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
1743 | } |
1744 | ||
6ae4283c TH |
1745 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
1746 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
1747 | dev_t dev; | |
1748 | int r; | |
1749 | ||
1750 | r = lookup_block_device(dev_path, &dev); | |
1751 | if (r < 0) | |
1752 | return; | |
1753 | ||
1754 | if (target != USEC_INFINITY) | |
ec61371f | 1755 | xsprintf(buf, DEVNUM_FORMAT_STR " target=%" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), target); |
6ae4283c | 1756 | else |
ec61371f | 1757 | xsprintf(buf, DEVNUM_FORMAT_STR " target=max\n", DEVNUM_FORMAT_VAL(dev)); |
6ae4283c | 1758 | |
293d32df | 1759 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
1760 | } |
1761 | ||
17ae2780 | 1762 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
4c1f9343 ZJS |
1763 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)], |
1764 | buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
64faf04c | 1765 | dev_t dev; |
64faf04c | 1766 | |
4c1f9343 | 1767 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1768 | return; |
64faf04c | 1769 | |
4c1f9343 | 1770 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
17ae2780 | 1771 | if (limits[type] != cgroup_io_limit_defaults[type]) |
64faf04c | 1772 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 1773 | else |
64faf04c | 1774 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c | 1775 | |
ec61371f | 1776 | xsprintf(buf, DEVNUM_FORMAT_STR " rbps=%s wbps=%s riops=%s wiops=%s\n", DEVNUM_FORMAT_VAL(dev), |
64faf04c TH |
1777 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], |
1778 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 1779 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
1780 | } |
1781 | ||
17ae2780 | 1782 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
1783 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1784 | dev_t dev; | |
64faf04c | 1785 | |
4c1f9343 | 1786 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1787 | return; |
64faf04c | 1788 | |
ec61371f | 1789 | sprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), rbps); |
293d32df | 1790 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 1791 | |
ec61371f | 1792 | sprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), wbps); |
293d32df | 1793 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
1794 | } |
1795 | ||
c52db42b CD |
1796 | static bool unit_has_unified_memory_config(Unit *u) { |
1797 | CGroupContext *c; | |
1798 | ||
1799 | assert(u); | |
1800 | ||
806a9362 | 1801 | assert_se(c = unit_get_cgroup_context(u)); |
c52db42b | 1802 | |
53fda560 LB |
1803 | return unit_get_ancestor_memory_min(u) > 0 || |
1804 | unit_get_ancestor_memory_low(u) > 0 || unit_get_ancestor_startup_memory_low(u) > 0 || | |
1805 | c->memory_high != CGROUP_LIMIT_MAX || c->startup_memory_high_set || | |
1806 | c->memory_max != CGROUP_LIMIT_MAX || c->startup_memory_max_set || | |
1807 | c->memory_swap_max != CGROUP_LIMIT_MAX || c->startup_memory_swap_max_set || | |
1808 | c->memory_zswap_max != CGROUP_LIMIT_MAX || c->startup_memory_zswap_max_set; | |
da4d897e TH |
1809 | } |
1810 | ||
f29ff115 | 1811 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 1812 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
1813 | |
1814 | if (v != CGROUP_LIMIT_MAX) | |
1815 | xsprintf(buf, "%" PRIu64 "\n", v); | |
1816 | ||
293d32df | 1817 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
1818 | } |
1819 | ||
0f2d84d2 | 1820 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
1821 | assert(u); |
1822 | ||
acf7f253 | 1823 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 1824 | |
acf7f253 | 1825 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
1826 | return; |
1827 | ||
fab34748 | 1828 | (void) bpf_firewall_load_custom(u); |
906c06f6 | 1829 | (void) bpf_firewall_install(u); |
906c06f6 DM |
1830 | } |
1831 | ||
49b6babb | 1832 | void unit_modify_nft_set(Unit *u, bool add) { |
dc7d69b3 | 1833 | int r; |
dc7d69b3 TM |
1834 | |
1835 | assert(u); | |
1836 | ||
1837 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1838 | return; | |
1839 | ||
49b6babb LP |
1840 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
1841 | return; | |
1842 | ||
dc7d69b3 TM |
1843 | if (cg_all_unified() <= 0) |
1844 | return; | |
1845 | ||
9cc54544 LP |
1846 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
1847 | if (!crt || crt->cgroup_id == 0) | |
dc7d69b3 TM |
1848 | return; |
1849 | ||
1850 | if (!u->manager->fw_ctx) { | |
1851 | r = fw_ctx_new_full(&u->manager->fw_ctx, /* init_tables= */ false); | |
1852 | if (r < 0) | |
1853 | return; | |
1854 | ||
1855 | assert(u->manager->fw_ctx); | |
1856 | } | |
1857 | ||
49b6babb LP |
1858 | CGroupContext *c = ASSERT_PTR(unit_get_cgroup_context(u)); |
1859 | ||
dc7d69b3 | 1860 | FOREACH_ARRAY(nft_set, c->nft_set_context.sets, c->nft_set_context.n_sets) { |
b2082753 TM |
1861 | if (nft_set->source != NFT_SET_SOURCE_CGROUP) |
1862 | continue; | |
1863 | ||
9cc54544 | 1864 | uint64_t element = crt->cgroup_id; |
dc7d69b3 TM |
1865 | |
1866 | r = nft_set_element_modify_any(u->manager->fw_ctx, add, nft_set->nfproto, nft_set->table, nft_set->set, &element, sizeof(element)); | |
1867 | if (r < 0) | |
1868 | log_warning_errno(r, "Failed to %s NFT set: family %s, table %s, set %s, cgroup %" PRIu64 ", ignoring: %m", | |
9cc54544 | 1869 | add? "add" : "delete", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, crt->cgroup_id); |
dc7d69b3 TM |
1870 | else |
1871 | log_debug("%s NFT set: family %s, table %s, set %s, cgroup %" PRIu64, | |
9cc54544 | 1872 | add? "Added" : "Deleted", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, crt->cgroup_id); |
dc7d69b3 TM |
1873 | } |
1874 | } | |
1875 | ||
a8e5eb17 JK |
1876 | static void cgroup_apply_socket_bind(Unit *u) { |
1877 | assert(u); | |
1878 | ||
cd09a5f3 | 1879 | (void) bpf_socket_bind_install(u); |
a8e5eb17 JK |
1880 | } |
1881 | ||
6f50d4f7 MV |
1882 | static void cgroup_apply_restrict_network_interfaces(Unit *u) { |
1883 | assert(u); | |
1884 | ||
62e22490 | 1885 | (void) bpf_restrict_ifaces_install(u); |
6f50d4f7 MV |
1886 | } |
1887 | ||
8b139557 | 1888 | static int cgroup_apply_devices(Unit *u) { |
76dc1725 | 1889 | _cleanup_(bpf_program_freep) BPFProgram *prog = NULL; |
8b139557 | 1890 | CGroupContext *c; |
45669ae2 | 1891 | CGroupDevicePolicy policy; |
8b139557 ZJS |
1892 | int r; |
1893 | ||
1894 | assert_se(c = unit_get_cgroup_context(u)); | |
9cc54544 LP |
1895 | |
1896 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
1897 | if (!crt || !crt->cgroup_path) | |
1898 | return -EOWNERDEAD; | |
8b139557 | 1899 | |
45669ae2 ZJS |
1900 | policy = c->device_policy; |
1901 | ||
8b139557 | 1902 | if (cg_all_unified() > 0) { |
45669ae2 | 1903 | r = bpf_devices_cgroup_init(&prog, policy, c->device_allow); |
8b139557 ZJS |
1904 | if (r < 0) |
1905 | return log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1906 | ||
1907 | } else { | |
1908 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore | |
1909 | * EINVAL here. */ | |
1910 | ||
45669ae2 | 1911 | if (c->device_allow || policy != CGROUP_DEVICE_POLICY_AUTO) |
9cc54544 | 1912 | r = cg_set_attribute("devices", crt->cgroup_path, "devices.deny", "a"); |
8b139557 | 1913 | else |
9cc54544 | 1914 | r = cg_set_attribute("devices", crt->cgroup_path, "devices.allow", "a"); |
8b139557 | 1915 | if (r < 0) |
8ed6f81b YW |
1916 | log_unit_full_errno(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1917 | "Failed to reset devices.allow/devices.deny: %m"); | |
8b139557 ZJS |
1918 | } |
1919 | ||
6b000af4 | 1920 | bool allow_list_static = policy == CGROUP_DEVICE_POLICY_CLOSED || |
45669ae2 | 1921 | (policy == CGROUP_DEVICE_POLICY_AUTO && c->device_allow); |
8b139557 | 1922 | |
958b73be LP |
1923 | bool any = false; |
1924 | if (allow_list_static) { | |
9cc54544 | 1925 | r = bpf_devices_allow_list_static(prog, crt->cgroup_path); |
958b73be LP |
1926 | if (r > 0) |
1927 | any = true; | |
1928 | } | |
1929 | ||
8b139557 | 1930 | LIST_FOREACH(device_allow, a, c->device_allow) { |
a1044811 LP |
1931 | const char *val; |
1932 | ||
1933 | if (a->permissions == 0) | |
8b139557 | 1934 | continue; |
8b139557 ZJS |
1935 | |
1936 | if (path_startswith(a->path, "/dev/")) | |
9cc54544 | 1937 | r = bpf_devices_allow_list_device(prog, crt->cgroup_path, a->path, a->permissions); |
8b139557 | 1938 | else if ((val = startswith(a->path, "block-"))) |
9cc54544 | 1939 | r = bpf_devices_allow_list_major(prog, crt->cgroup_path, val, 'b', a->permissions); |
8b139557 | 1940 | else if ((val = startswith(a->path, "char-"))) |
9cc54544 | 1941 | r = bpf_devices_allow_list_major(prog, crt->cgroup_path, val, 'c', a->permissions); |
45669ae2 | 1942 | else { |
8b139557 | 1943 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
45669ae2 ZJS |
1944 | continue; |
1945 | } | |
1946 | ||
958b73be | 1947 | if (r > 0) |
45669ae2 ZJS |
1948 | any = true; |
1949 | } | |
1950 | ||
1951 | if (prog && !any) { | |
1952 | log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENODEV), "No devices matched by device filter."); | |
1953 | ||
1954 | /* The kernel verifier would reject a program we would build with the normal intro and outro | |
6b000af4 | 1955 | but no allow-listing rules (outro would contain an unreachable instruction for successful |
45669ae2 ZJS |
1956 | return). */ |
1957 | policy = CGROUP_DEVICE_POLICY_STRICT; | |
8b139557 ZJS |
1958 | } |
1959 | ||
9cc54544 | 1960 | r = bpf_devices_apply_policy(&prog, policy, any, crt->cgroup_path, &crt->bpf_device_control_installed); |
8b139557 ZJS |
1961 | if (r < 0) { |
1962 | static bool warned = false; | |
1963 | ||
1964 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1965 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1966 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1967 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1968 | ||
1969 | warned = true; | |
1970 | } | |
1971 | return r; | |
1972 | } | |
1973 | ||
17283ce7 YW |
1974 | static void set_io_weight(Unit *u, uint64_t weight) { |
1975 | char buf[STRLEN("default \n")+DECIMAL_STR_MAX(uint64_t)]; | |
17283ce7 YW |
1976 | |
1977 | assert(u); | |
29eb0eef | 1978 | |
3e6eafdd | 1979 | (void) set_bfq_weight(u, "io", makedev(0, 0), weight); |
29eb0eef | 1980 | |
29eb0eef | 1981 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
17283ce7 YW |
1982 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
1983 | } | |
1984 | ||
1985 | static void set_blkio_weight(Unit *u, uint64_t weight) { | |
1986 | char buf[STRLEN("\n")+DECIMAL_STR_MAX(uint64_t)]; | |
17283ce7 YW |
1987 | |
1988 | assert(u); | |
29eb0eef | 1989 | |
3e6eafdd | 1990 | (void) set_bfq_weight(u, "blkio", makedev(0, 0), weight); |
17283ce7 YW |
1991 | |
1992 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
1993 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); | |
29eb0eef ZJS |
1994 | } |
1995 | ||
506ea51b JK |
1996 | static void cgroup_apply_bpf_foreign_program(Unit *u) { |
1997 | assert(u); | |
1998 | ||
1999 | (void) bpf_foreign_install(u); | |
2000 | } | |
2001 | ||
906c06f6 DM |
2002 | static void cgroup_context_apply( |
2003 | Unit *u, | |
2004 | CGroupMask apply_mask, | |
906c06f6 DM |
2005 | ManagerState state) { |
2006 | ||
9cc54544 | 2007 | bool is_host_root, is_local_root; |
f29ff115 TH |
2008 | const char *path; |
2009 | CGroupContext *c; | |
4ad49000 LP |
2010 | int r; |
2011 | ||
f29ff115 TH |
2012 | assert(u); |
2013 | ||
906c06f6 | 2014 | /* Nothing to do? Exit early! */ |
17f14955 | 2015 | if (apply_mask == 0) |
4ad49000 | 2016 | return; |
8e274523 | 2017 | |
52fecf20 LP |
2018 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
2019 | * attributes should only be managed for cgroups further down the tree. */ | |
2020 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
2021 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
2022 | |
2023 | assert_se(c = unit_get_cgroup_context(u)); | |
9cc54544 LP |
2024 | |
2025 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
2026 | if (!crt || !crt->cgroup_path) | |
2027 | return; | |
2028 | ||
2029 | path = crt->cgroup_path; | |
f3725e64 | 2030 | |
52fecf20 | 2031 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 2032 | path = "/"; |
01efdf13 | 2033 | |
be2c0327 LP |
2034 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
2035 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 2036 | |
be2c0327 LP |
2037 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
2038 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
2039 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 2040 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
2041 | * we couldn't even write to them if we wanted to). */ |
2042 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 2043 | |
b4cccbc1 | 2044 | if (cg_all_unified() > 0) { |
be2c0327 | 2045 | uint64_t weight; |
b2f8b02e | 2046 | |
be2c0327 LP |
2047 | if (cgroup_context_has_cpu_weight(c)) |
2048 | weight = cgroup_context_cpu_weight(c, state); | |
2049 | else if (cgroup_context_has_cpu_shares(c)) { | |
2050 | uint64_t shares; | |
66ebf6c0 | 2051 | |
be2c0327 LP |
2052 | shares = cgroup_context_cpu_shares(c, state); |
2053 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 2054 | |
be2c0327 LP |
2055 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
2056 | shares, weight, path); | |
2057 | } else | |
2058 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 2059 | |
c8340822 | 2060 | cgroup_apply_unified_cpu_idle(u, weight); |
be2c0327 | 2061 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 2062 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 2063 | |
52fecf20 | 2064 | } else { |
be2c0327 | 2065 | uint64_t shares; |
52fecf20 | 2066 | |
be2c0327 LP |
2067 | if (cgroup_context_has_cpu_weight(c)) { |
2068 | uint64_t weight; | |
52fecf20 | 2069 | |
be2c0327 LP |
2070 | weight = cgroup_context_cpu_weight(c, state); |
2071 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 2072 | |
be2c0327 LP |
2073 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
2074 | weight, shares, path); | |
2075 | } else if (cgroup_context_has_cpu_shares(c)) | |
2076 | shares = cgroup_context_cpu_shares(c, state); | |
2077 | else | |
2078 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 2079 | |
be2c0327 | 2080 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 2081 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 2082 | } |
4ad49000 LP |
2083 | } |
2084 | ||
047f5d63 | 2085 | if ((apply_mask & CGROUP_MASK_CPUSET) && !is_local_root) { |
31d3a520 PM |
2086 | cgroup_apply_unified_cpuset(u, cgroup_context_allowed_cpus(c, state), "cpuset.cpus"); |
2087 | cgroup_apply_unified_cpuset(u, cgroup_context_allowed_mems(c, state), "cpuset.mems"); | |
047f5d63 PH |
2088 | } |
2089 | ||
4e1dfa45 | 2090 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
2091 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
2092 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
2093 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
52fecf20 LP |
2094 | bool has_io, has_blockio; |
2095 | uint64_t weight; | |
13c31542 | 2096 | |
52fecf20 LP |
2097 | has_io = cgroup_context_has_io_config(c); |
2098 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 2099 | |
52fecf20 LP |
2100 | if (has_io) |
2101 | weight = cgroup_context_io_weight(c, state); | |
2102 | else if (has_blockio) { | |
2103 | uint64_t blkio_weight; | |
128fadc9 | 2104 | |
52fecf20 LP |
2105 | blkio_weight = cgroup_context_blkio_weight(c, state); |
2106 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 2107 | |
67e2ea15 | 2108 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
2109 | blkio_weight, weight); |
2110 | } else | |
2111 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 2112 | |
17283ce7 | 2113 | set_io_weight(u, weight); |
2dbc45ae | 2114 | |
52fecf20 | 2115 | if (has_io) { |
52fecf20 LP |
2116 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
2117 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 2118 | |
52fecf20 LP |
2119 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
2120 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 2121 | |
52fecf20 LP |
2122 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
2123 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 2124 | |
52fecf20 | 2125 | } else if (has_blockio) { |
52fecf20 LP |
2126 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
2127 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 2128 | |
67e2ea15 | 2129 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 2130 | w->weight, weight, w->path); |
538b4852 | 2131 | |
52fecf20 LP |
2132 | cgroup_apply_io_device_weight(u, w->path, weight); |
2133 | } | |
538b4852 | 2134 | |
17ae2780 | 2135 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 | 2136 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
538b4852 | 2137 | |
e8616626 | 2138 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
538b4852 TH |
2139 | limits[type] = cgroup_io_limit_defaults[type]; |
2140 | ||
2141 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
2142 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
2143 | ||
67e2ea15 | 2144 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
2145 | b->rbps, b->wbps, b->path); |
2146 | ||
17ae2780 | 2147 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 2148 | } |
13c31542 TH |
2149 | } |
2150 | } | |
2151 | ||
906c06f6 | 2152 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 2153 | bool has_io, has_blockio; |
4ad49000 | 2154 | |
52fecf20 LP |
2155 | has_io = cgroup_context_has_io_config(c); |
2156 | has_blockio = cgroup_context_has_blockio_config(c); | |
2157 | ||
2158 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
2159 | * left to our container manager, too. */ | |
2160 | if (!is_local_root) { | |
64faf04c | 2161 | uint64_t weight; |
64faf04c | 2162 | |
7d862ab8 | 2163 | if (has_io) { |
52fecf20 | 2164 | uint64_t io_weight; |
128fadc9 | 2165 | |
52fecf20 | 2166 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 2167 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 2168 | |
67e2ea15 | 2169 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 2170 | io_weight, weight); |
7d862ab8 TH |
2171 | } else if (has_blockio) |
2172 | weight = cgroup_context_blkio_weight(c, state); | |
2173 | else | |
538b4852 | 2174 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c | 2175 | |
17283ce7 | 2176 | set_blkio_weight(u, weight); |
35e7a62c | 2177 | |
03677889 | 2178 | if (has_io) |
128fadc9 TH |
2179 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
2180 | weight = cgroup_weight_io_to_blkio(w->weight); | |
2181 | ||
67e2ea15 | 2182 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
2183 | w->weight, weight, w->path); |
2184 | ||
2185 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
2186 | } | |
03677889 | 2187 | else if (has_blockio) |
7d862ab8 TH |
2188 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
2189 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
4ad49000 LP |
2190 | } |
2191 | ||
5238e957 | 2192 | /* The bandwidth limits are something that make sense to be applied to the host's root but not container |
52fecf20 LP |
2193 | * roots, as there we want the container manager to handle it */ |
2194 | if (is_host_root || !is_local_root) { | |
03677889 | 2195 | if (has_io) |
52fecf20 | 2196 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 2197 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 2198 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 2199 | |
52fecf20 LP |
2200 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
2201 | } | |
03677889 | 2202 | else if (has_blockio) |
52fecf20 LP |
2203 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
2204 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
d686d8a9 | 2205 | } |
8e274523 LP |
2206 | } |
2207 | ||
be2c0327 LP |
2208 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
2209 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 2210 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
2211 | * write to this if we wanted to.) */ |
2212 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 2213 | |
52fecf20 | 2214 | if (cg_all_unified() > 0) { |
53fda560 | 2215 | uint64_t max, swap_max = CGROUP_LIMIT_MAX, zswap_max = CGROUP_LIMIT_MAX, high = CGROUP_LIMIT_MAX; |
be2c0327 | 2216 | |
c52db42b | 2217 | if (unit_has_unified_memory_config(u)) { |
53fda560 LB |
2218 | bool startup = IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING); |
2219 | ||
2220 | high = startup && c->startup_memory_high_set ? c->startup_memory_high : c->memory_high; | |
2221 | max = startup && c->startup_memory_max_set ? c->startup_memory_max : c->memory_max; | |
2222 | swap_max = startup && c->startup_memory_swap_max_set ? c->startup_memory_swap_max : c->memory_swap_max; | |
2223 | zswap_max = startup && c->startup_memory_zswap_max_set ? c->startup_memory_zswap_max : c->memory_zswap_max; | |
be2c0327 LP |
2224 | } else { |
2225 | max = c->memory_limit; | |
efdb0237 | 2226 | |
be2c0327 LP |
2227 | if (max != CGROUP_LIMIT_MAX) |
2228 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 2229 | } |
da4d897e | 2230 | |
64fe532e | 2231 | cgroup_apply_unified_memory_limit(u, "memory.min", unit_get_ancestor_memory_min(u)); |
c52db42b | 2232 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
53fda560 | 2233 | cgroup_apply_unified_memory_limit(u, "memory.high", high); |
be2c0327 LP |
2234 | cgroup_apply_unified_memory_limit(u, "memory.max", max); |
2235 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
d7fe0a67 | 2236 | cgroup_apply_unified_memory_limit(u, "memory.zswap.max", zswap_max); |
128fadc9 | 2237 | |
afcfaa69 LP |
2238 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
2239 | ||
be2c0327 LP |
2240 | } else { |
2241 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
2242 | uint64_t val; | |
52fecf20 | 2243 | |
c52db42b | 2244 | if (unit_has_unified_memory_config(u)) { |
be2c0327 | 2245 | val = c->memory_max; |
b7a41491 MK |
2246 | if (val != CGROUP_LIMIT_MAX) |
2247 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIu64 " as MemoryLimit=", val); | |
be2c0327 LP |
2248 | } else |
2249 | val = c->memory_limit; | |
78a4ee59 | 2250 | |
be2c0327 LP |
2251 | if (val == CGROUP_LIMIT_MAX) |
2252 | strncpy(buf, "-1\n", sizeof(buf)); | |
2253 | else | |
2254 | xsprintf(buf, "%" PRIu64 "\n", val); | |
2255 | ||
2256 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 2257 | } |
4ad49000 | 2258 | } |
8e274523 | 2259 | |
4e1dfa45 | 2260 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
2261 | * containers, where we leave this to the manager */ |
2262 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
8b139557 ZJS |
2263 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) |
2264 | (void) cgroup_apply_devices(u); | |
03a7b521 | 2265 | |
00b5974f LP |
2266 | if (apply_mask & CGROUP_MASK_PIDS) { |
2267 | ||
52fecf20 | 2268 | if (is_host_root) { |
00b5974f LP |
2269 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
2270 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
2271 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
2272 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
2273 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
2274 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
2275 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
2276 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
2277 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
2278 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
5238e957 | 2279 | * which is desirable so that there's an official way to release control of the sysctl from |
00b5974f LP |
2280 | * systemd: set the limit to unbounded and reload. */ |
2281 | ||
94f0b13b | 2282 | if (cgroup_tasks_max_isset(&c->tasks_max)) { |
00b5974f | 2283 | u->manager->sysctl_pid_max_changed = true; |
94f0b13b | 2284 | r = procfs_tasks_set_limit(cgroup_tasks_max_resolve(&c->tasks_max)); |
00b5974f LP |
2285 | } else if (u->manager->sysctl_pid_max_changed) |
2286 | r = procfs_tasks_set_limit(TASKS_MAX); | |
2287 | else | |
2288 | r = 0; | |
00b5974f | 2289 | if (r < 0) |
8ed6f81b YW |
2290 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
2291 | "Failed to write to tasks limit sysctls: %m"); | |
52fecf20 | 2292 | } |
03a7b521 | 2293 | |
52fecf20 LP |
2294 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
2295 | * leave it for the container manager. */ | |
2296 | if (!is_local_root) { | |
94f0b13b | 2297 | if (cgroup_tasks_max_isset(&c->tasks_max)) { |
3a0f06c4 | 2298 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
03a7b521 | 2299 | |
94f0b13b | 2300 | xsprintf(buf, "%" PRIu64 "\n", cgroup_tasks_max_resolve(&c->tasks_max)); |
293d32df | 2301 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 2302 | } else |
589a5f7a | 2303 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 2304 | } |
03a7b521 | 2305 | } |
906c06f6 | 2306 | |
17f14955 | 2307 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 2308 | cgroup_apply_firewall(u); |
506ea51b JK |
2309 | |
2310 | if (apply_mask & CGROUP_MASK_BPF_FOREIGN) | |
2311 | cgroup_apply_bpf_foreign_program(u); | |
a8e5eb17 JK |
2312 | |
2313 | if (apply_mask & CGROUP_MASK_BPF_SOCKET_BIND) | |
2314 | cgroup_apply_socket_bind(u); | |
6f50d4f7 MV |
2315 | |
2316 | if (apply_mask & CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES) | |
2317 | cgroup_apply_restrict_network_interfaces(u); | |
dc7d69b3 | 2318 | |
49b6babb | 2319 | unit_modify_nft_set(u, /* add = */ true); |
fb385181 LP |
2320 | } |
2321 | ||
16492445 LP |
2322 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
2323 | CGroupContext *c; | |
16492445 LP |
2324 | assert(u); |
2325 | ||
2326 | c = unit_get_cgroup_context(u); | |
2327 | if (!c) | |
2328 | return false; | |
2329 | ||
2330 | if (c->ip_accounting || | |
84ebe6f0 YW |
2331 | !set_isempty(c->ip_address_allow) || |
2332 | !set_isempty(c->ip_address_deny) || | |
fab34748 KL |
2333 | c->ip_filters_ingress || |
2334 | c->ip_filters_egress) | |
16492445 LP |
2335 | return true; |
2336 | ||
2337 | /* If any parent slice has an IP access list defined, it applies too */ | |
e8616626 | 2338 | for (Unit *p = UNIT_GET_SLICE(u); p; p = UNIT_GET_SLICE(p)) { |
16492445 LP |
2339 | c = unit_get_cgroup_context(p); |
2340 | if (!c) | |
2341 | return false; | |
2342 | ||
84ebe6f0 YW |
2343 | if (!set_isempty(c->ip_address_allow) || |
2344 | !set_isempty(c->ip_address_deny)) | |
16492445 LP |
2345 | return true; |
2346 | } | |
2347 | ||
2348 | return false; | |
2349 | } | |
2350 | ||
506ea51b JK |
2351 | static bool unit_get_needs_bpf_foreign_program(Unit *u) { |
2352 | CGroupContext *c; | |
2353 | assert(u); | |
2354 | ||
2355 | c = unit_get_cgroup_context(u); | |
2356 | if (!c) | |
2357 | return false; | |
2358 | ||
64903d18 | 2359 | return !!c->bpf_foreign_programs; |
506ea51b JK |
2360 | } |
2361 | ||
a8e5eb17 JK |
2362 | static bool unit_get_needs_socket_bind(Unit *u) { |
2363 | CGroupContext *c; | |
2364 | assert(u); | |
2365 | ||
2366 | c = unit_get_cgroup_context(u); | |
2367 | if (!c) | |
2368 | return false; | |
2369 | ||
11ab01e4 | 2370 | return c->socket_bind_allow || c->socket_bind_deny; |
a8e5eb17 JK |
2371 | } |
2372 | ||
6f50d4f7 MV |
2373 | static bool unit_get_needs_restrict_network_interfaces(Unit *u) { |
2374 | CGroupContext *c; | |
2375 | assert(u); | |
2376 | ||
2377 | c = unit_get_cgroup_context(u); | |
2378 | if (!c) | |
2379 | return false; | |
2380 | ||
2381 | return !set_isempty(c->restrict_network_interfaces); | |
2382 | } | |
2383 | ||
c52db42b | 2384 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 2385 | CGroupMask mask = 0; |
c52db42b CD |
2386 | CGroupContext *c; |
2387 | ||
2388 | assert(u); | |
2389 | ||
806a9362 | 2390 | assert_se(c = unit_get_cgroup_context(u)); |
c710d3b4 | 2391 | |
fae9bc29 | 2392 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 2393 | |
fae9bc29 | 2394 | if (c->cpu_accounting) |
f98c2585 | 2395 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
2396 | |
2397 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 2398 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 2399 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 2400 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 2401 | |
31d3a520 | 2402 | if (cgroup_context_has_allowed_cpus(c) || cgroup_context_has_allowed_mems(c)) |
047f5d63 PH |
2403 | mask |= CGROUP_MASK_CPUSET; |
2404 | ||
538b4852 TH |
2405 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
2406 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 2407 | |
4ad49000 | 2408 | if (c->memory_accounting || |
da4d897e | 2409 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 2410 | unit_has_unified_memory_config(u)) |
efdb0237 | 2411 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 2412 | |
a931ad47 | 2413 | if (c->device_allow || |
084870f9 | 2414 | c->device_policy != CGROUP_DEVICE_POLICY_AUTO) |
084c7007 | 2415 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 2416 | |
03a7b521 | 2417 | if (c->tasks_accounting || |
94f0b13b | 2418 | cgroup_tasks_max_isset(&c->tasks_max)) |
03a7b521 LP |
2419 | mask |= CGROUP_MASK_PIDS; |
2420 | ||
fae9bc29 | 2421 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
2422 | } |
2423 | ||
53aea74a | 2424 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
2425 | CGroupMask mask = 0; |
2426 | ||
fae9bc29 LP |
2427 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
2428 | * too. */ | |
2429 | ||
17f14955 RG |
2430 | if (unit_get_needs_bpf_firewall(u)) |
2431 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2432 | ||
506ea51b JK |
2433 | if (unit_get_needs_bpf_foreign_program(u)) |
2434 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
2435 | ||
a8e5eb17 JK |
2436 | if (unit_get_needs_socket_bind(u)) |
2437 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
2438 | ||
6f50d4f7 MV |
2439 | if (unit_get_needs_restrict_network_interfaces(u)) |
2440 | mask |= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES; | |
2441 | ||
17f14955 RG |
2442 | return mask; |
2443 | } | |
2444 | ||
efdb0237 | 2445 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 2446 | CGroupContext *c; |
8e274523 | 2447 | |
442ce775 LP |
2448 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
2449 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
2450 | ||
2451 | if (u->load_state != UNIT_LOADED) | |
2452 | return 0; | |
efdb0237 | 2453 | |
4ad49000 LP |
2454 | c = unit_get_cgroup_context(u); |
2455 | if (!c) | |
2456 | return 0; | |
8e274523 | 2457 | |
12b975e0 | 2458 | return unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u); |
02638280 LP |
2459 | } |
2460 | ||
2461 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
2462 | CGroupContext *c; | |
2463 | ||
2464 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
2465 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 2466 | * |
02638280 | 2467 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 2468 | |
1d9cc876 | 2469 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
2470 | return 0; |
2471 | ||
2472 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
2473 | ExecContext *e; |
2474 | ||
2475 | e = unit_get_exec_context(u); | |
02638280 LP |
2476 | if (e && !exec_context_maintains_privileges(e)) |
2477 | return 0; | |
a931ad47 LP |
2478 | } |
2479 | ||
1d9cc876 | 2480 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 2481 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
2482 | } |
2483 | ||
d9ef5944 MK |
2484 | static CGroupMask unit_get_subtree_mask(Unit *u) { |
2485 | ||
2486 | /* Returns the mask of this subtree, meaning of the group | |
2487 | * itself and its children. */ | |
2488 | ||
2489 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
2490 | } | |
2491 | ||
efdb0237 | 2492 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 2493 | assert(u); |
bc432dc7 | 2494 | |
02638280 | 2495 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 2496 | |
9cc54544 LP |
2497 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
2498 | if (crt && crt->cgroup_members_mask_valid) | |
2499 | return crt->cgroup_members_mask; /* Use cached value if possible */ | |
bc432dc7 | 2500 | |
9cc54544 | 2501 | CGroupMask m = 0; |
bc432dc7 LP |
2502 | if (u->type == UNIT_SLICE) { |
2503 | Unit *member; | |
bc432dc7 | 2504 | |
d219a2b0 | 2505 | UNIT_FOREACH_DEPENDENCY(member, u, UNIT_ATOM_SLICE_OF) |
9cc54544 | 2506 | m |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
2507 | } |
2508 | ||
9cc54544 LP |
2509 | if (crt) { |
2510 | crt->cgroup_members_mask = m; | |
2511 | crt->cgroup_members_mask_valid = true; | |
2512 | } | |
2513 | ||
2514 | return m; | |
246aa6dd LP |
2515 | } |
2516 | ||
efdb0237 | 2517 | CGroupMask unit_get_siblings_mask(Unit *u) { |
12f64221 | 2518 | Unit *slice; |
4ad49000 | 2519 | assert(u); |
246aa6dd | 2520 | |
efdb0237 LP |
2521 | /* Returns the mask of controllers all of the unit's siblings |
2522 | * require, i.e. the members mask of the unit's parent slice | |
2523 | * if there is one. */ | |
2524 | ||
12f64221 LP |
2525 | slice = UNIT_GET_SLICE(u); |
2526 | if (slice) | |
2527 | return unit_get_members_mask(slice); | |
4ad49000 | 2528 | |
64e844e5 | 2529 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
2530 | } |
2531 | ||
d9ef5944 | 2532 | static CGroupMask unit_get_disable_mask(Unit *u) { |
4f6f62e4 CD |
2533 | CGroupContext *c; |
2534 | ||
2535 | c = unit_get_cgroup_context(u); | |
2536 | if (!c) | |
2537 | return 0; | |
2538 | ||
2539 | return c->disable_controllers; | |
2540 | } | |
2541 | ||
2542 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
2543 | CGroupMask mask; | |
12f64221 | 2544 | Unit *slice; |
4f6f62e4 CD |
2545 | |
2546 | assert(u); | |
2547 | mask = unit_get_disable_mask(u); | |
2548 | ||
2549 | /* Returns the mask of controllers which are marked as forcibly | |
2550 | * disabled in any ancestor unit or the unit in question. */ | |
2551 | ||
12f64221 LP |
2552 | slice = UNIT_GET_SLICE(u); |
2553 | if (slice) | |
2554 | mask |= unit_get_ancestor_disable_mask(slice); | |
4f6f62e4 CD |
2555 | |
2556 | return mask; | |
2557 | } | |
2558 | ||
efdb0237 | 2559 | CGroupMask unit_get_target_mask(Unit *u) { |
a437c5e4 | 2560 | CGroupMask own_mask, mask; |
efdb0237 | 2561 | |
a437c5e4 LP |
2562 | /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything |
2563 | * it needs itself, plus all that its children need, plus all that its siblings need. This is | |
2564 | * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each | |
efdb0237 | 2565 | * hierarchy that shall be enabled for it. */ |
6414b7c9 | 2566 | |
a437c5e4 | 2567 | own_mask = unit_get_own_mask(u); |
84d2744b | 2568 | |
a437c5e4 | 2569 | if (own_mask & CGROUP_MASK_BPF_FIREWALL & ~u->manager->cgroup_supported) |
84d2744b ZJS |
2570 | emit_bpf_firewall_warning(u); |
2571 | ||
a437c5e4 LP |
2572 | mask = own_mask | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
2573 | ||
efdb0237 | 2574 | mask &= u->manager->cgroup_supported; |
c72703e2 | 2575 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
2576 | |
2577 | return mask; | |
2578 | } | |
2579 | ||
2580 | CGroupMask unit_get_enable_mask(Unit *u) { | |
2581 | CGroupMask mask; | |
2582 | ||
2583 | /* This returns the cgroup mask of all controllers to enable | |
2584 | * for the children of a specific cgroup. This is primarily | |
2585 | * useful for the unified cgroup hierarchy, where each cgroup | |
2586 | * controls which controllers are enabled for its children. */ | |
2587 | ||
2588 | mask = unit_get_members_mask(u); | |
6414b7c9 | 2589 | mask &= u->manager->cgroup_supported; |
c72703e2 | 2590 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
2591 | |
2592 | return mask; | |
2593 | } | |
2594 | ||
5af88058 | 2595 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
12f64221 LP |
2596 | Unit *slice; |
2597 | ||
bc432dc7 LP |
2598 | assert(u); |
2599 | ||
9cc54544 LP |
2600 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
2601 | if (!crt) | |
2602 | return; | |
2603 | ||
5af88058 | 2604 | /* Recurse invalidate the member masks cache all the way up the tree */ |
9cc54544 | 2605 | crt->cgroup_members_mask_valid = false; |
bc432dc7 | 2606 | |
12f64221 LP |
2607 | slice = UNIT_GET_SLICE(u); |
2608 | if (slice) | |
2609 | unit_invalidate_cgroup_members_masks(slice); | |
6414b7c9 DS |
2610 | } |
2611 | ||
6592b975 | 2612 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 2613 | |
6592b975 | 2614 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
2615 | |
2616 | while (u) { | |
9cc54544 LP |
2617 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
2618 | if (crt && | |
2619 | crt->cgroup_path && | |
2620 | crt->cgroup_realized && | |
2621 | FLAGS_SET(crt->cgroup_realized_mask, mask)) | |
2622 | return crt->cgroup_path; | |
03b90d4b | 2623 | |
12f64221 | 2624 | u = UNIT_GET_SLICE(u); |
03b90d4b LP |
2625 | } |
2626 | ||
2627 | return NULL; | |
2628 | } | |
2629 | ||
6592b975 | 2630 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
7b639614 MK |
2631 | /* If not realized at all, migrate to root (""). |
2632 | * It may happen if we're upgrading from older version that didn't clean up. | |
2633 | */ | |
2634 | return strempty(unit_get_realized_cgroup_path(userdata, mask)); | |
6592b975 LP |
2635 | } |
2636 | ||
1a56b0c0 LP |
2637 | int unit_default_cgroup_path(const Unit *u, char **ret) { |
2638 | _cleanup_free_ char *p = NULL; | |
efdb0237 LP |
2639 | int r; |
2640 | ||
2641 | assert(u); | |
1a56b0c0 | 2642 | assert(ret); |
efdb0237 LP |
2643 | |
2644 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1a56b0c0 LP |
2645 | p = strdup(u->manager->cgroup_root); |
2646 | else { | |
2647 | _cleanup_free_ char *escaped = NULL, *slice_path = NULL; | |
2648 | Unit *slice; | |
efdb0237 | 2649 | |
1a56b0c0 LP |
2650 | slice = UNIT_GET_SLICE(u); |
2651 | if (slice && !unit_has_name(slice, SPECIAL_ROOT_SLICE)) { | |
2652 | r = cg_slice_to_path(slice->id, &slice_path); | |
2653 | if (r < 0) | |
2654 | return r; | |
2655 | } | |
2656 | ||
2657 | r = cg_escape(u->id, &escaped); | |
efdb0237 | 2658 | if (r < 0) |
1a56b0c0 | 2659 | return r; |
efdb0237 | 2660 | |
1a56b0c0 LP |
2661 | p = path_join(empty_to_root(u->manager->cgroup_root), slice_path, escaped); |
2662 | } | |
2663 | if (!p) | |
2664 | return -ENOMEM; | |
efdb0237 | 2665 | |
1a56b0c0 LP |
2666 | *ret = TAKE_PTR(p); |
2667 | return 0; | |
efdb0237 LP |
2668 | } |
2669 | ||
2670 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
2671 | _cleanup_free_ char *p = NULL; | |
9cc54544 | 2672 | CGroupRuntime *crt; |
efdb0237 LP |
2673 | int r; |
2674 | ||
2675 | assert(u); | |
2676 | ||
9cc54544 LP |
2677 | crt = unit_get_cgroup_runtime(u); |
2678 | ||
2679 | if (crt && streq_ptr(crt->cgroup_path, path)) | |
5210387e LP |
2680 | return 0; |
2681 | ||
9cc54544 LP |
2682 | unit_release_cgroup(u); |
2683 | ||
2684 | crt = unit_setup_cgroup_runtime(u); | |
2685 | if (!crt) | |
2686 | return -ENOMEM; | |
2687 | ||
efdb0237 LP |
2688 | if (path) { |
2689 | p = strdup(path); | |
2690 | if (!p) | |
2691 | return -ENOMEM; | |
efdb0237 | 2692 | |
efdb0237 LP |
2693 | r = hashmap_put(u->manager->cgroup_unit, p, u); |
2694 | if (r < 0) | |
2695 | return r; | |
2696 | } | |
2697 | ||
9cc54544 LP |
2698 | assert(!crt->cgroup_path); |
2699 | crt->cgroup_path = TAKE_PTR(p); | |
efdb0237 LP |
2700 | |
2701 | return 1; | |
2702 | } | |
2703 | ||
2704 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 2705 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
2706 | int r; |
2707 | ||
2708 | assert(u); | |
2709 | ||
0bb814c2 LP |
2710 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
2711 | * cgroupv2 is available. */ | |
2712 | ||
9cc54544 LP |
2713 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
2714 | if (!crt || !crt->cgroup_path) | |
efdb0237 LP |
2715 | return 0; |
2716 | ||
9cc54544 | 2717 | if (crt->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
2718 | return 0; |
2719 | ||
2720 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 2721 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2722 | if (r < 0) |
2723 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
2724 | if (r == 0) | |
efdb0237 LP |
2725 | return 0; |
2726 | ||
0bb814c2 | 2727 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
2728 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
2729 | return 0; | |
2730 | ||
0bb814c2 | 2731 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
2732 | if (r < 0) |
2733 | return log_oom(); | |
2734 | ||
9cc54544 | 2735 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, crt->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
2736 | if (r < 0) |
2737 | return log_oom(); | |
2738 | ||
9cc54544 LP |
2739 | crt->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
2740 | if (crt->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 2741 | |
0bb814c2 LP |
2742 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
2743 | * is not an error */ | |
efdb0237 LP |
2744 | return 0; |
2745 | ||
9cc54544 | 2746 | return log_unit_error_errno(u, errno, "Failed to add control inotify watch descriptor for control group %s: %m", empty_to_root(crt->cgroup_path)); |
efdb0237 LP |
2747 | } |
2748 | ||
9cc54544 | 2749 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(crt->cgroup_control_inotify_wd), u); |
efdb0237 | 2750 | if (r < 0) |
9cc54544 | 2751 | 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(crt->cgroup_path)); |
efdb0237 LP |
2752 | |
2753 | return 0; | |
2754 | } | |
2755 | ||
afcfaa69 LP |
2756 | int unit_watch_cgroup_memory(Unit *u) { |
2757 | _cleanup_free_ char *events = NULL; | |
afcfaa69 LP |
2758 | int r; |
2759 | ||
2760 | assert(u); | |
2761 | ||
2762 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
2763 | * cgroupv2 is available. */ | |
2764 | ||
9cc54544 LP |
2765 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
2766 | if (!crt || !crt->cgroup_path) | |
afcfaa69 LP |
2767 | return 0; |
2768 | ||
9cc54544 | 2769 | CGroupContext *c = unit_get_cgroup_context(u); |
afcfaa69 LP |
2770 | if (!c) |
2771 | return 0; | |
2772 | ||
2773 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
2774 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
2775 | * all. */ | |
2776 | if (!c->memory_accounting) | |
2777 | return 0; | |
2778 | ||
2779 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
2780 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
2781 | if (u->type == UNIT_SLICE) | |
2782 | return 0; | |
2783 | ||
9cc54544 | 2784 | if (crt->cgroup_memory_inotify_wd >= 0) |
afcfaa69 LP |
2785 | return 0; |
2786 | ||
2787 | /* Only applies to the unified hierarchy */ | |
2788 | r = cg_all_unified(); | |
2789 | if (r < 0) | |
2790 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
2791 | if (r == 0) | |
2792 | return 0; | |
2793 | ||
2794 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
2795 | if (r < 0) | |
2796 | return log_oom(); | |
2797 | ||
9cc54544 | 2798 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, crt->cgroup_path, "memory.events", &events); |
afcfaa69 LP |
2799 | if (r < 0) |
2800 | return log_oom(); | |
2801 | ||
9cc54544 LP |
2802 | crt->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
2803 | if (crt->cgroup_memory_inotify_wd < 0) { | |
afcfaa69 LP |
2804 | |
2805 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
2806 | * is not an error */ | |
2807 | return 0; | |
2808 | ||
9cc54544 | 2809 | return log_unit_error_errno(u, errno, "Failed to add memory inotify watch descriptor for control group %s: %m", empty_to_root(crt->cgroup_path)); |
afcfaa69 LP |
2810 | } |
2811 | ||
9cc54544 | 2812 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(crt->cgroup_memory_inotify_wd), u); |
afcfaa69 | 2813 | if (r < 0) |
9cc54544 | 2814 | 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(crt->cgroup_path)); |
afcfaa69 LP |
2815 | |
2816 | return 0; | |
2817 | } | |
2818 | ||
a4634b21 LP |
2819 | int unit_pick_cgroup_path(Unit *u) { |
2820 | _cleanup_free_ char *path = NULL; | |
2821 | int r; | |
2822 | ||
2823 | assert(u); | |
2824 | ||
a4634b21 LP |
2825 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
2826 | return -EINVAL; | |
2827 | ||
9cc54544 LP |
2828 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); |
2829 | if (!crt) | |
2830 | return -ENOMEM; | |
2831 | if (crt->cgroup_path) | |
2832 | return 0; | |
2833 | ||
1a56b0c0 LP |
2834 | r = unit_default_cgroup_path(u, &path); |
2835 | if (r < 0) | |
2836 | return log_unit_error_errno(u, r, "Failed to generate default cgroup path: %m"); | |
a4634b21 LP |
2837 | |
2838 | r = unit_set_cgroup_path(u, path); | |
2839 | if (r == -EEXIST) | |
6178e2f8 | 2840 | return log_unit_error_errno(u, r, "Control group %s exists already.", empty_to_root(path)); |
a4634b21 | 2841 | if (r < 0) |
6178e2f8 | 2842 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", empty_to_root(path)); |
a4634b21 LP |
2843 | |
2844 | return 0; | |
2845 | } | |
2846 | ||
7b639614 | 2847 | static int unit_update_cgroup( |
efdb0237 LP |
2848 | Unit *u, |
2849 | CGroupMask target_mask, | |
0d2d6fbf CD |
2850 | CGroupMask enable_mask, |
2851 | ManagerState state) { | |
efdb0237 | 2852 | |
7b639614 MK |
2853 | bool created, is_root_slice; |
2854 | CGroupMask migrate_mask = 0; | |
184b4f78 | 2855 | _cleanup_free_ char *cgroup_full_path = NULL; |
27adcc97 | 2856 | int r; |
64747e2d | 2857 | |
4ad49000 | 2858 | assert(u); |
64747e2d | 2859 | |
27c4ed79 | 2860 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
2861 | return 0; |
2862 | ||
a4634b21 LP |
2863 | /* Figure out our cgroup path */ |
2864 | r = unit_pick_cgroup_path(u); | |
2865 | if (r < 0) | |
2866 | return r; | |
b58b8e11 | 2867 | |
9cc54544 LP |
2868 | CGroupRuntime *crt = ASSERT_PTR(unit_get_cgroup_runtime(u)); |
2869 | ||
03b90d4b | 2870 | /* First, create our own group */ |
9cc54544 | 2871 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, crt->cgroup_path); |
23bbb0de | 2872 | if (r < 0) |
9cc54544 | 2873 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", empty_to_root(crt->cgroup_path)); |
490c5a37 | 2874 | created = r; |
efdb0237 | 2875 | |
184b4f78 | 2876 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
1b420223 LP |
2877 | uint64_t cgroup_id = 0; |
2878 | ||
9cc54544 | 2879 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, crt->cgroup_path, NULL, &cgroup_full_path); |
184b4f78 ILG |
2880 | if (r == 0) { |
2881 | r = cg_path_get_cgroupid(cgroup_full_path, &cgroup_id); | |
2882 | if (r < 0) | |
1b420223 LP |
2883 | log_unit_full_errno(u, ERRNO_IS_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r, |
2884 | "Failed to get cgroup ID of cgroup %s, ignoring: %m", cgroup_full_path); | |
184b4f78 | 2885 | } else |
9cc54544 | 2886 | log_unit_warning_errno(u, r, "Failed to get full cgroup path on cgroup %s, ignoring: %m", empty_to_root(crt->cgroup_path)); |
184b4f78 | 2887 | |
9cc54544 | 2888 | crt->cgroup_id = cgroup_id; |
184b4f78 ILG |
2889 | } |
2890 | ||
efdb0237 LP |
2891 | /* Start watching it */ |
2892 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 2893 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 2894 | |
7b639614 MK |
2895 | /* For v2 we preserve enabled controllers in delegated units, adjust others, |
2896 | * for v1 we figure out which controller hierarchies need migration. */ | |
9cc54544 | 2897 | if (created || !crt->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 2898 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
2899 | |
2900 | /* Enable all controllers we need */ | |
9cc54544 | 2901 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, crt->cgroup_path, &result_mask); |
65be7e06 | 2902 | if (r < 0) |
9cc54544 | 2903 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", empty_to_root(crt->cgroup_path)); |
27adcc97 | 2904 | |
27adcc97 | 2905 | /* Remember what's actually enabled now */ |
9cc54544 | 2906 | crt->cgroup_enabled_mask = result_mask; |
7b639614 | 2907 | |
9cc54544 | 2908 | migrate_mask = crt->cgroup_realized_mask ^ target_mask; |
65be7e06 | 2909 | } |
03b90d4b LP |
2910 | |
2911 | /* Keep track that this is now realized */ | |
9cc54544 LP |
2912 | crt->cgroup_realized = true; |
2913 | crt->cgroup_realized_mask = target_mask; | |
4ad49000 | 2914 | |
7b639614 MK |
2915 | /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling). |
2916 | * | |
2917 | * Unnecessary controller cgroups are trimmed (after emptied by upward migration). | |
2918 | * We perform migration also with whole slices for cases when users don't care about leave | |
2919 | * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing | |
2920 | * delegated units. | |
2921 | */ | |
2922 | if (cg_all_unified() == 0) { | |
9cc54544 | 2923 | r = cg_migrate_v1_controllers(u->manager->cgroup_supported, migrate_mask, crt->cgroup_path, migrate_callback, u); |
7b639614 | 2924 | if (r < 0) |
9cc54544 | 2925 | log_unit_warning_errno(u, r, "Failed to migrate controller cgroups from %s, ignoring: %m", empty_to_root(crt->cgroup_path)); |
0cd385d3 | 2926 | |
7b639614 | 2927 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
9cc54544 | 2928 | r = cg_trim_v1_controllers(u->manager->cgroup_supported, ~target_mask, crt->cgroup_path, !is_root_slice); |
0cd385d3 | 2929 | if (r < 0) |
9cc54544 | 2930 | log_unit_warning_errno(u, r, "Failed to delete controller cgroups %s, ignoring: %m", empty_to_root(crt->cgroup_path)); |
0cd385d3 | 2931 | } |
03b90d4b | 2932 | |
0d2d6fbf CD |
2933 | /* Set attributes */ |
2934 | cgroup_context_apply(u, target_mask, state); | |
2935 | cgroup_xattr_apply(u); | |
2936 | ||
29e6b0c1 LP |
2937 | /* For most units we expect that memory monitoring is set up before the unit is started and we won't |
2938 | * touch it after. For PID 1 this is different though, because we couldn't possibly do that given | |
2939 | * that PID 1 runs before init.scope is even set up. Hence, whenever init.scope is realized, let's | |
2940 | * try to open the memory pressure interface anew. */ | |
2941 | if (unit_has_name(u, SPECIAL_INIT_SCOPE)) | |
2942 | (void) manager_setup_memory_pressure_event_source(u->manager); | |
2943 | ||
64747e2d LP |
2944 | return 0; |
2945 | } | |
2946 | ||
6592b975 LP |
2947 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
2948 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2949 | char *pp; | |
7b3fd631 | 2950 | int r; |
6592b975 | 2951 | |
7b3fd631 LP |
2952 | assert(u); |
2953 | ||
6592b975 LP |
2954 | if (MANAGER_IS_SYSTEM(u->manager)) |
2955 | return -EINVAL; | |
2956 | ||
2957 | if (!u->manager->system_bus) | |
2958 | return -EIO; | |
2959 | ||
9cc54544 LP |
2960 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
2961 | if (!crt || !crt->cgroup_path) | |
2962 | return -EOWNERDEAD; | |
6592b975 LP |
2963 | |
2964 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
9cc54544 | 2965 | pp = path_startswith(crt->cgroup_path, u->manager->cgroup_root); |
6592b975 LP |
2966 | if (!pp) |
2967 | return -EINVAL; | |
2968 | ||
2969 | pp = strjoina("/", pp, suffix_path); | |
4ff361cc | 2970 | path_simplify(pp); |
6592b975 | 2971 | |
78fa2f91 | 2972 | r = bus_call_method(u->manager->system_bus, |
2973 | bus_systemd_mgr, | |
2974 | "AttachProcessesToUnit", | |
2975 | &error, NULL, | |
2976 | "ssau", | |
2977 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 2978 | if (r < 0) |
6592b975 LP |
2979 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
2980 | ||
2981 | return 0; | |
2982 | } | |
2983 | ||
2984 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
8e7e4a73 | 2985 | _cleanup_free_ char *joined = NULL; |
6592b975 LP |
2986 | CGroupMask delegated_mask; |
2987 | const char *p; | |
495e75ed | 2988 | PidRef *pid; |
db4229d1 | 2989 | int ret, r; |
6592b975 LP |
2990 | |
2991 | assert(u); | |
2992 | ||
2993 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2994 | return -EINVAL; | |
2995 | ||
2996 | if (set_isempty(pids)) | |
2997 | return 0; | |
7b3fd631 | 2998 | |
fab34748 KL |
2999 | /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable. |
3000 | * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */ | |
3001 | r = bpf_firewall_load_custom(u); | |
3002 | if (r < 0) | |
3003 | return r; | |
3004 | ||
6592b975 | 3005 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
3006 | if (r < 0) |
3007 | return r; | |
3008 | ||
9cc54544 LP |
3009 | CGroupRuntime *crt = ASSERT_PTR(unit_get_cgroup_runtime(u)); |
3010 | ||
6592b975 | 3011 | if (isempty(suffix_path)) |
9cc54544 | 3012 | p = crt->cgroup_path; |
8e7e4a73 | 3013 | else { |
9cc54544 | 3014 | joined = path_join(crt->cgroup_path, suffix_path); |
8e7e4a73 LP |
3015 | if (!joined) |
3016 | return -ENOMEM; | |
3017 | ||
3018 | p = joined; | |
3019 | } | |
6592b975 LP |
3020 | |
3021 | delegated_mask = unit_get_delegate_mask(u); | |
3022 | ||
db4229d1 | 3023 | ret = 0; |
495e75ed LP |
3024 | SET_FOREACH(pid, pids) { |
3025 | ||
3026 | /* Unfortunately we cannot add pids by pidfd to a cgroup. Hence we have to use PIDs instead, | |
3027 | * which of course is racy. Let's shorten the race a bit though, and re-validate the PID | |
3028 | * before we use it */ | |
3029 | r = pidref_verify(pid); | |
3030 | if (r < 0) { | |
3031 | 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)); | |
3032 | continue; | |
3033 | } | |
6592b975 LP |
3034 | |
3035 | /* First, attach the PID to the main cgroup hierarchy */ | |
495e75ed | 3036 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid->pid); |
db4229d1 LP |
3037 | if (r < 0) { |
3038 | bool again = MANAGER_IS_USER(u->manager) && ERRNO_IS_PRIVILEGE(r); | |
6592b975 | 3039 | |
db4229d1 | 3040 | log_unit_full_errno(u, again ? LOG_DEBUG : LOG_INFO, r, |
7a2ba407 | 3041 | "Couldn't move process "PID_FMT" to%s requested cgroup '%s': %m", |
495e75ed | 3042 | pid->pid, again ? " directly" : "", empty_to_root(p)); |
7a2ba407 ZJS |
3043 | |
3044 | if (again) { | |
6592b975 LP |
3045 | int z; |
3046 | ||
7a2ba407 ZJS |
3047 | /* If we are in a user instance, and we can't move the process ourselves due |
3048 | * to permission problems, let's ask the system instance about it instead. | |
3049 | * Since it's more privileged it might be able to move the process across the | |
3050 | * leaves of a subtree whose top node is not owned by us. */ | |
6592b975 | 3051 | |
495e75ed | 3052 | z = unit_attach_pid_to_cgroup_via_bus(u, pid->pid, suffix_path); |
6592b975 | 3053 | if (z < 0) |
495e75ed | 3054 | 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 |
3055 | else { |
3056 | if (ret >= 0) | |
3057 | ret++; /* Count successful additions */ | |
6592b975 | 3058 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ |
c65417a0 | 3059 | } |
6592b975 LP |
3060 | } |
3061 | ||
db4229d1 LP |
3062 | if (ret >= 0) |
3063 | ret = r; /* Remember first error */ | |
6592b975 LP |
3064 | |
3065 | continue; | |
8d3e4ac7 LP |
3066 | } else if (ret >= 0) |
3067 | ret++; /* Count successful additions */ | |
6592b975 | 3068 | |
db4229d1 LP |
3069 | r = cg_all_unified(); |
3070 | if (r < 0) | |
3071 | return r; | |
3072 | if (r > 0) | |
6592b975 LP |
3073 | continue; |
3074 | ||
3075 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
3076 | * innermost realized one */ | |
3077 | ||
e8616626 | 3078 | for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
6592b975 LP |
3079 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); |
3080 | const char *realized; | |
3081 | ||
3082 | if (!(u->manager->cgroup_supported & bit)) | |
3083 | continue; | |
3084 | ||
3085 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
9cc54544 | 3086 | if (delegated_mask & crt->cgroup_realized_mask & bit) { |
495e75ed | 3087 | r = cg_attach(cgroup_controller_to_string(c), p, pid->pid); |
db4229d1 | 3088 | if (r >= 0) |
6592b975 LP |
3089 | continue; /* Success! */ |
3090 | ||
db4229d1 | 3091 | 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 | 3092 | pid->pid, empty_to_root(p), cgroup_controller_to_string(c)); |
6592b975 LP |
3093 | } |
3094 | ||
3095 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
3096 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
3097 | * realized. */ | |
3098 | realized = unit_get_realized_cgroup_path(u, bit); | |
3099 | if (!realized) | |
3100 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
3101 | ||
495e75ed | 3102 | r = cg_attach(cgroup_controller_to_string(c), realized, pid->pid); |
db4229d1 LP |
3103 | if (r < 0) |
3104 | log_unit_debug_errno(u, r, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
495e75ed | 3105 | pid->pid, realized, cgroup_controller_to_string(c)); |
6592b975 LP |
3106 | } |
3107 | } | |
3108 | ||
db4229d1 | 3109 | return ret; |
7b3fd631 LP |
3110 | } |
3111 | ||
906c06f6 DM |
3112 | static bool unit_has_mask_realized( |
3113 | Unit *u, | |
3114 | CGroupMask target_mask, | |
17f14955 | 3115 | CGroupMask enable_mask) { |
906c06f6 | 3116 | |
bc432dc7 LP |
3117 | assert(u); |
3118 | ||
9cc54544 LP |
3119 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3120 | if (!crt) | |
3121 | return false; | |
3122 | ||
d5095dcd LP |
3123 | /* Returns true if this unit is fully realized. We check four things: |
3124 | * | |
3125 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
3126 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
3127 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
3128 | * 4. Whether the invalidation mask is currently zero |
3129 | * | |
3130 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
3131 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
3132 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
3133 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
3134 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
3135 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
3136 | * simply don't matter. */ | |
3137 | ||
9cc54544 LP |
3138 | return crt->cgroup_realized && |
3139 | ((crt->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && | |
3140 | ((crt->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
3141 | crt->cgroup_invalidated_mask == 0; | |
6414b7c9 DS |
3142 | } |
3143 | ||
4f6f62e4 CD |
3144 | static bool unit_has_mask_disables_realized( |
3145 | Unit *u, | |
3146 | CGroupMask target_mask, | |
3147 | CGroupMask enable_mask) { | |
3148 | ||
3149 | assert(u); | |
3150 | ||
9cc54544 LP |
3151 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3152 | if (!crt) | |
3153 | return true; | |
3154 | ||
4f6f62e4 CD |
3155 | /* Returns true if all controllers which should be disabled are indeed disabled. |
3156 | * | |
3157 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
3158 | * already removed. */ | |
3159 | ||
9cc54544 LP |
3160 | return !crt->cgroup_realized || |
3161 | (FLAGS_SET(crt->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
3162 | FLAGS_SET(crt->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
4f6f62e4 CD |
3163 | } |
3164 | ||
a57669d2 CD |
3165 | static bool unit_has_mask_enables_realized( |
3166 | Unit *u, | |
3167 | CGroupMask target_mask, | |
3168 | CGroupMask enable_mask) { | |
3169 | ||
3170 | assert(u); | |
3171 | ||
9cc54544 LP |
3172 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3173 | if (!crt) | |
3174 | return false; | |
3175 | ||
a57669d2 CD |
3176 | /* Returns true if all controllers which should be enabled are indeed enabled. |
3177 | * | |
3178 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
3179 | * we want to add is already added. */ | |
3180 | ||
9cc54544 LP |
3181 | return crt->cgroup_realized && |
3182 | ((crt->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (crt->cgroup_realized_mask & CGROUP_MASK_V1) && | |
3183 | ((crt->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (crt->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
3184 | } |
3185 | ||
020b2e41 | 3186 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
3187 | assert(u); |
3188 | ||
3189 | if (u->in_cgroup_realize_queue) | |
3190 | return; | |
3191 | ||
a479c21e | 3192 | LIST_APPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); |
2aa57a65 LP |
3193 | u->in_cgroup_realize_queue = true; |
3194 | } | |
3195 | ||
3196 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
3197 | assert(u); | |
3198 | ||
3199 | if (!u->in_cgroup_realize_queue) | |
3200 | return; | |
3201 | ||
3202 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
3203 | u->in_cgroup_realize_queue = false; | |
3204 | } | |
3205 | ||
a57669d2 CD |
3206 | /* Controllers can only be enabled breadth-first, from the root of the |
3207 | * hierarchy downwards to the unit in question. */ | |
3208 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
3209 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
12f64221 | 3210 | Unit *slice; |
a57669d2 CD |
3211 | int r; |
3212 | ||
3213 | assert(u); | |
3214 | ||
3215 | /* First go deal with this unit's parent, or we won't be able to enable | |
3216 | * any new controllers at this layer. */ | |
12f64221 LP |
3217 | slice = UNIT_GET_SLICE(u); |
3218 | if (slice) { | |
3219 | r = unit_realize_cgroup_now_enable(slice, state); | |
a57669d2 CD |
3220 | if (r < 0) |
3221 | return r; | |
3222 | } | |
3223 | ||
3224 | target_mask = unit_get_target_mask(u); | |
3225 | enable_mask = unit_get_enable_mask(u); | |
3226 | ||
3227 | /* We can only enable in this direction, don't try to disable anything. | |
3228 | */ | |
3229 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
3230 | return 0; | |
3231 | ||
9cc54544 LP |
3232 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3233 | ||
3234 | new_target_mask = (crt ? crt->cgroup_realized_mask : 0) | target_mask; | |
3235 | new_enable_mask = (crt ? crt->cgroup_enabled_mask : 0) | enable_mask; | |
a57669d2 | 3236 | |
7b639614 | 3237 | return unit_update_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
3238 | } |
3239 | ||
4f6f62e4 CD |
3240 | /* Controllers can only be disabled depth-first, from the leaves of the |
3241 | * hierarchy upwards to the unit in question. */ | |
3242 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
4f6f62e4 | 3243 | Unit *m; |
4f6f62e4 CD |
3244 | |
3245 | assert(u); | |
3246 | ||
3247 | if (u->type != UNIT_SLICE) | |
3248 | return 0; | |
3249 | ||
d219a2b0 | 3250 | UNIT_FOREACH_DEPENDENCY(m, u, UNIT_ATOM_SLICE_OF) { |
4f6f62e4 CD |
3251 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; |
3252 | int r; | |
3253 | ||
9cc54544 LP |
3254 | CGroupRuntime *rt = unit_get_cgroup_runtime(m); |
3255 | if (!rt) | |
3256 | continue; | |
3257 | ||
defe63b0 LP |
3258 | /* The cgroup for this unit might not actually be fully realised yet, in which case it isn't |
3259 | * holding any controllers open anyway. */ | |
9cc54544 | 3260 | if (!rt->cgroup_realized) |
4f6f62e4 CD |
3261 | continue; |
3262 | ||
defe63b0 | 3263 | /* We must disable those below us first in order to release the controller. */ |
4f6f62e4 CD |
3264 | if (m->type == UNIT_SLICE) |
3265 | (void) unit_realize_cgroup_now_disable(m, state); | |
3266 | ||
3267 | target_mask = unit_get_target_mask(m); | |
3268 | enable_mask = unit_get_enable_mask(m); | |
3269 | ||
defe63b0 | 3270 | /* We can only disable in this direction, don't try to enable anything. */ |
4f6f62e4 CD |
3271 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) |
3272 | continue; | |
3273 | ||
9cc54544 LP |
3274 | new_target_mask = rt->cgroup_realized_mask & target_mask; |
3275 | new_enable_mask = rt->cgroup_enabled_mask & enable_mask; | |
4f6f62e4 | 3276 | |
7b639614 | 3277 | r = unit_update_cgroup(m, new_target_mask, new_enable_mask, state); |
4f6f62e4 CD |
3278 | if (r < 0) |
3279 | return r; | |
3280 | } | |
3281 | ||
3282 | return 0; | |
3283 | } | |
a57669d2 | 3284 | |
6414b7c9 DS |
3285 | /* Check if necessary controllers and attributes for a unit are in place. |
3286 | * | |
a57669d2 CD |
3287 | * - If so, do nothing. |
3288 | * - If not, create paths, move processes over, and set attributes. | |
3289 | * | |
3290 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
3291 | * a depth-first way. As such the process looks like this: | |
3292 | * | |
3293 | * Suppose we have a cgroup hierarchy which looks like this: | |
3294 | * | |
3295 | * root | |
3296 | * / \ | |
3297 | * / \ | |
3298 | * / \ | |
3299 | * a b | |
3300 | * / \ / \ | |
3301 | * / \ / \ | |
3302 | * c d e f | |
3303 | * / \ / \ / \ / \ | |
3304 | * h i j k l m n o | |
3305 | * | |
3306 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 3307 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
3308 | * 3. cgroup "k" just started requesting the memory controller. |
3309 | * | |
3310 | * To make this work we must do the following in order: | |
3311 | * | |
3312 | * 1. Disable CPU controller in k, j | |
3313 | * 2. Disable CPU controller in d | |
3314 | * 3. Enable memory controller in root | |
3315 | * 4. Enable memory controller in a | |
3316 | * 5. Enable memory controller in d | |
3317 | * 6. Enable memory controller in k | |
3318 | * | |
3319 | * Notice that we need to touch j in one direction, but not the other. We also | |
3320 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
3321 | * wants to disable further. The basic rules are therefore: | |
3322 | * | |
3323 | * - If you're disabling something, you need to realise all of the cgroups from | |
3324 | * your recursive descendants to the root. This starts from the leaves. | |
3325 | * - If you're enabling something, you need to realise from the root cgroup | |
3326 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
3327 | * |
3328 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 3329 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 3330 | CGroupMask target_mask, enable_mask; |
12f64221 | 3331 | Unit *slice; |
6414b7c9 | 3332 | int r; |
64747e2d | 3333 | |
4ad49000 | 3334 | assert(u); |
64747e2d | 3335 | |
2aa57a65 | 3336 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 3337 | |
efdb0237 | 3338 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
3339 | enable_mask = unit_get_enable_mask(u); |
3340 | ||
17f14955 | 3341 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 3342 | return 0; |
64747e2d | 3343 | |
4f6f62e4 CD |
3344 | /* Disable controllers below us, if there are any */ |
3345 | r = unit_realize_cgroup_now_disable(u, state); | |
3346 | if (r < 0) | |
3347 | return r; | |
3348 | ||
3349 | /* Enable controllers above us, if there are any */ | |
12f64221 LP |
3350 | slice = UNIT_GET_SLICE(u); |
3351 | if (slice) { | |
3352 | r = unit_realize_cgroup_now_enable(slice, state); | |
6414b7c9 DS |
3353 | if (r < 0) |
3354 | return r; | |
3355 | } | |
4ad49000 | 3356 | |
0d2d6fbf | 3357 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
7b639614 | 3358 | r = unit_update_cgroup(u, target_mask, enable_mask, state); |
6414b7c9 DS |
3359 | if (r < 0) |
3360 | return r; | |
3361 | ||
9cc54544 LP |
3362 | CGroupRuntime *crt = ASSERT_PTR(unit_get_cgroup_runtime(u)); |
3363 | ||
c2baf11c | 3364 | /* Now, reset the invalidation mask */ |
9cc54544 | 3365 | crt->cgroup_invalidated_mask = 0; |
6414b7c9 | 3366 | return 0; |
64747e2d LP |
3367 | } |
3368 | ||
91a6073e | 3369 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 3370 | ManagerState state; |
4ad49000 | 3371 | unsigned n = 0; |
db785129 | 3372 | Unit *i; |
6414b7c9 | 3373 | int r; |
ecedd90f | 3374 | |
91a6073e LP |
3375 | assert(m); |
3376 | ||
db785129 LP |
3377 | state = manager_state(m); |
3378 | ||
91a6073e LP |
3379 | while ((i = m->cgroup_realize_queue)) { |
3380 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 3381 | |
2aa57a65 LP |
3382 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
3383 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
3384 | unit_remove_from_cgroup_realize_queue(i); | |
3385 | continue; | |
3386 | } | |
3387 | ||
db785129 | 3388 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 3389 | if (r < 0) |
efdb0237 | 3390 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 3391 | |
4ad49000 LP |
3392 | n++; |
3393 | } | |
ecedd90f | 3394 | |
4ad49000 | 3395 | return n; |
8e274523 LP |
3396 | } |
3397 | ||
4c591f39 MK |
3398 | void unit_add_family_to_cgroup_realize_queue(Unit *u) { |
3399 | assert(u); | |
3400 | assert(u->type == UNIT_SLICE); | |
ca949c9d | 3401 | |
4c591f39 MK |
3402 | /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all |
3403 | * its ancestors. | |
3404 | * | |
3405 | * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes | |
3406 | * very weird if two units that own processes reside in the same slice, but one is realized in the | |
3407 | * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does | |
3408 | * not), because that means individual processes need to be scheduled against whole cgroups. Let's | |
3409 | * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1 | |
3410 | * controller hierarchies too (if unit requires the controller to be realized). | |
e1e98911 | 3411 | * |
4c591f39 MK |
3412 | * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date |
3413 | * masks. */ | |
3414 | ||
3415 | do { | |
9cc54544 | 3416 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
8f53a7b8 | 3417 | |
4c591f39 | 3418 | /* Children of u likely changed when we're called */ |
9cc54544 LP |
3419 | if (crt) |
3420 | crt->cgroup_members_mask_valid = false; | |
f23ba94d | 3421 | |
9cc54544 | 3422 | Unit *m; |
d219a2b0 | 3423 | UNIT_FOREACH_DEPENDENCY(m, u, UNIT_ATOM_SLICE_OF) { |
8e274523 | 3424 | |
65f6b6bd | 3425 | /* No point in doing cgroup application for units without active processes. */ |
6414b7c9 DS |
3426 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) |
3427 | continue; | |
3428 | ||
e1e98911 LP |
3429 | /* We only enqueue siblings if they were realized once at least, in the main |
3430 | * hierarchy. */ | |
9cc54544 LP |
3431 | crt = unit_get_cgroup_runtime(m); |
3432 | if (!crt || !crt->cgroup_realized) | |
e1e98911 LP |
3433 | continue; |
3434 | ||
defe63b0 LP |
3435 | /* If the unit doesn't need any new controllers and has current ones |
3436 | * realized, it doesn't need any changes. */ | |
906c06f6 DM |
3437 | if (unit_has_mask_realized(m, |
3438 | unit_get_target_mask(m), | |
17f14955 | 3439 | unit_get_enable_mask(m))) |
6414b7c9 DS |
3440 | continue; |
3441 | ||
91a6073e | 3442 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
3443 | } |
3444 | ||
4c591f39 MK |
3445 | /* Parent comes after children */ |
3446 | unit_add_to_cgroup_realize_queue(u); | |
12f64221 LP |
3447 | |
3448 | u = UNIT_GET_SLICE(u); | |
3449 | } while (u); | |
4ad49000 LP |
3450 | } |
3451 | ||
0a1eb06d | 3452 | int unit_realize_cgroup(Unit *u) { |
12f64221 LP |
3453 | Unit *slice; |
3454 | ||
4ad49000 LP |
3455 | assert(u); |
3456 | ||
35b7ff80 | 3457 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 3458 | return 0; |
8e274523 | 3459 | |
4c591f39 MK |
3460 | /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all |
3461 | * parents, but there's more actually: for the weight-based controllers we also need to make sure | |
3462 | * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the | |
3463 | * other hand, when a controller is removed from realized set, it may become unnecessary in siblings | |
3464 | * and ancestors and they should be (de)realized too. | |
3465 | * | |
3466 | * This call will defer work on the siblings and derealized ancestors to the next event loop | |
3467 | * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */ | |
ca949c9d | 3468 | |
12f64221 LP |
3469 | slice = UNIT_GET_SLICE(u); |
3470 | if (slice) | |
3471 | unit_add_family_to_cgroup_realize_queue(slice); | |
4ad49000 | 3472 | |
6414b7c9 | 3473 | /* And realize this one now (and apply the values) */ |
db785129 | 3474 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
3475 | } |
3476 | ||
efdb0237 LP |
3477 | void unit_release_cgroup(Unit *u) { |
3478 | assert(u); | |
3479 | ||
8a0d5388 LP |
3480 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
3481 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 | 3482 | |
9cc54544 LP |
3483 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3484 | if (!crt) | |
3485 | return; | |
3486 | ||
3487 | if (crt->cgroup_path) { | |
3488 | (void) hashmap_remove(u->manager->cgroup_unit, crt->cgroup_path); | |
3489 | crt->cgroup_path = mfree(crt->cgroup_path); | |
efdb0237 LP |
3490 | } |
3491 | ||
9cc54544 LP |
3492 | if (crt->cgroup_control_inotify_wd >= 0) { |
3493 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, crt->cgroup_control_inotify_wd) < 0) | |
3494 | log_unit_debug_errno(u, errno, "Failed to remove cgroup control inotify watch %i for %s, ignoring: %m", crt->cgroup_control_inotify_wd, u->id); | |
efdb0237 | 3495 | |
9cc54544 LP |
3496 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(crt->cgroup_control_inotify_wd)); |
3497 | crt->cgroup_control_inotify_wd = -1; | |
efdb0237 | 3498 | } |
afcfaa69 | 3499 | |
9cc54544 LP |
3500 | if (crt->cgroup_memory_inotify_wd >= 0) { |
3501 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, crt->cgroup_memory_inotify_wd) < 0) | |
3502 | log_unit_debug_errno(u, errno, "Failed to remove cgroup memory inotify watch %i for %s, ignoring: %m", crt->cgroup_memory_inotify_wd, u->id); | |
afcfaa69 | 3503 | |
9cc54544 LP |
3504 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(crt->cgroup_memory_inotify_wd)); |
3505 | crt->cgroup_memory_inotify_wd = -1; | |
afcfaa69 | 3506 | } |
9cc54544 LP |
3507 | |
3508 | *(CGroupRuntime**) ((uint8_t*) u + UNIT_VTABLE(u)->cgroup_runtime_offset) = cgroup_runtime_free(crt); | |
3509 | } | |
3510 | ||
3511 | int unit_cgroup_is_empty(Unit *u) { | |
3512 | int r; | |
3513 | ||
3514 | assert(u); | |
3515 | ||
3516 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
3517 | if (!crt) | |
3518 | return -ENXIO; | |
3519 | if (!crt->cgroup_path) | |
3520 | return -EOWNERDEAD; | |
3521 | ||
3522 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, crt->cgroup_path); | |
3523 | if (r < 0) | |
3524 | return log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty, ignoring: %m", empty_to_root(crt->cgroup_path)); | |
3525 | ||
3526 | return r; | |
efdb0237 LP |
3527 | } |
3528 | ||
e08dabfe AZ |
3529 | bool unit_maybe_release_cgroup(Unit *u) { |
3530 | int r; | |
3531 | ||
3532 | assert(u); | |
3533 | ||
9cc54544 LP |
3534 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3535 | if (!crt || !crt->cgroup_path) | |
e08dabfe AZ |
3536 | return true; |
3537 | ||
9cc54544 LP |
3538 | /* Don't release the cgroup if there are still processes under it. If we get notified later when all |
3539 | * the processes exit (e.g. the processes were in D-state and exited after the unit was marked as | |
3540 | * failed) we need the cgroup paths to continue to be tracked by the manager so they can be looked up | |
3541 | * and cleaned up later. */ | |
3542 | r = unit_cgroup_is_empty(u); | |
3543 | if (r == 1) { | |
e08dabfe AZ |
3544 | unit_release_cgroup(u); |
3545 | return true; | |
3546 | } | |
3547 | ||
3548 | return false; | |
3549 | } | |
3550 | ||
efdb0237 | 3551 | void unit_prune_cgroup(Unit *u) { |
8e274523 | 3552 | int r; |
efdb0237 | 3553 | bool is_root_slice; |
8e274523 | 3554 | |
4ad49000 | 3555 | assert(u); |
8e274523 | 3556 | |
efdb0237 | 3557 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
9cc54544 LP |
3558 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3559 | if (!crt || !crt->cgroup_path) | |
4ad49000 | 3560 | return; |
8e274523 | 3561 | |
ad009380 MY |
3562 | /* Cache the last CPU and memory usage values before we destroy the cgroup */ |
3563 | (void) unit_get_cpu_usage(u, /* ret = */ NULL); | |
3564 | ||
3565 | for (CGroupMemoryAccountingMetric metric = 0; metric <= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST; metric++) | |
3566 | (void) unit_get_memory_accounting(u, metric, /* ret = */ NULL); | |
fe700f46 | 3567 | |
b1994387 | 3568 | #if BPF_FRAMEWORK |
352ec23c | 3569 | (void) bpf_restrict_fs_cleanup(u); /* Remove cgroup from the global LSM BPF map */ |
b1994387 ILG |
3570 | #endif |
3571 | ||
49b6babb | 3572 | unit_modify_nft_set(u, /* add = */ false); |
dc7d69b3 | 3573 | |
efdb0237 LP |
3574 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
3575 | ||
9cc54544 | 3576 | r = cg_trim_everywhere(u->manager->cgroup_supported, crt->cgroup_path, !is_root_slice); |
0219b352 DB |
3577 | if (r < 0) |
3578 | /* One reason we could have failed here is, that the cgroup still contains a process. | |
3579 | * However, if the cgroup becomes removable at a later time, it might be removed when | |
3580 | * the containing slice is stopped. So even if we failed now, this unit shouldn't assume | |
3581 | * that the cgroup is still realized the next time it is started. Do not return early | |
3582 | * on error, continue cleanup. */ | |
9cc54544 | 3583 | log_unit_full_errno(u, r == -EBUSY ? LOG_DEBUG : LOG_WARNING, r, "Failed to destroy cgroup %s, ignoring: %m", empty_to_root(crt->cgroup_path)); |
8e274523 | 3584 | |
efdb0237 LP |
3585 | if (is_root_slice) |
3586 | return; | |
3587 | ||
e08dabfe AZ |
3588 | if (!unit_maybe_release_cgroup(u)) /* Returns true if the cgroup was released */ |
3589 | return; | |
0a1eb06d | 3590 | |
9cc54544 LP |
3591 | crt = unit_get_cgroup_runtime(u); /* The above might have destroyed the runtime object, let's see if it's still there */ |
3592 | if (!crt) | |
3593 | return; | |
3594 | ||
3595 | crt->cgroup_realized = false; | |
3596 | crt->cgroup_realized_mask = 0; | |
3597 | crt->cgroup_enabled_mask = 0; | |
084c7007 | 3598 | |
9cc54544 | 3599 | crt->bpf_device_control_installed = bpf_program_free(crt->bpf_device_control_installed); |
8e274523 LP |
3600 | } |
3601 | ||
495e75ed LP |
3602 | int unit_search_main_pid(Unit *u, PidRef *ret) { |
3603 | _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL; | |
4ad49000 | 3604 | _cleanup_fclose_ FILE *f = NULL; |
efdb0237 | 3605 | int r; |
4ad49000 LP |
3606 | |
3607 | assert(u); | |
efdb0237 | 3608 | assert(ret); |
4ad49000 | 3609 | |
9cc54544 LP |
3610 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3611 | if (!crt || !crt->cgroup_path) | |
efdb0237 | 3612 | return -ENXIO; |
4ad49000 | 3613 | |
9cc54544 | 3614 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, crt->cgroup_path, &f); |
efdb0237 LP |
3615 | if (r < 0) |
3616 | return r; | |
4ad49000 | 3617 | |
495e75ed LP |
3618 | for (;;) { |
3619 | _cleanup_(pidref_done) PidRef npidref = PIDREF_NULL; | |
4ad49000 | 3620 | |
495e75ed LP |
3621 | r = cg_read_pidref(f, &npidref); |
3622 | if (r < 0) | |
3623 | return r; | |
3624 | if (r == 0) | |
3625 | break; | |
8e274523 | 3626 | |
495e75ed | 3627 | if (pidref_equal(&pidref, &npidref)) /* seen already, cgroupfs reports duplicates! */ |
4ad49000 | 3628 | continue; |
8e274523 | 3629 | |
6774be42 | 3630 | if (pidref_is_my_child(&npidref) <= 0) /* ignore processes further down the tree */ |
495e75ed | 3631 | continue; |
efdb0237 | 3632 | |
495e75ed LP |
3633 | if (pidref_is_set(&pidref) != 0) |
3634 | /* Dang, there's more than one daemonized PID in this group, so we don't know what | |
3635 | * process is the main process. */ | |
efdb0237 | 3636 | return -ENODATA; |
8e274523 | 3637 | |
495e75ed | 3638 | pidref = TAKE_PIDREF(npidref); |
8e274523 LP |
3639 | } |
3640 | ||
495e75ed LP |
3641 | if (!pidref_is_set(&pidref)) |
3642 | return -ENODATA; | |
3643 | ||
3644 | *ret = TAKE_PIDREF(pidref); | |
efdb0237 LP |
3645 | return 0; |
3646 | } | |
3647 | ||
3648 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 3649 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
3650 | _cleanup_fclose_ FILE *f = NULL; |
3651 | int ret = 0, r; | |
3652 | ||
3653 | assert(u); | |
3654 | assert(path); | |
3655 | ||
3656 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
3657 | if (r < 0) | |
495e75ed | 3658 | RET_GATHER(ret, r); |
efdb0237 | 3659 | else { |
495e75ed LP |
3660 | for (;;) { |
3661 | _cleanup_(pidref_done) PidRef pid = PIDREF_NULL; | |
3662 | ||
3663 | r = cg_read_pidref(f, &pid); | |
3664 | if (r == 0) | |
3665 | break; | |
3666 | if (r < 0) { | |
3667 | RET_GATHER(ret, r); | |
3668 | break; | |
3669 | } | |
efdb0237 | 3670 | |
495e75ed | 3671 | RET_GATHER(ret, unit_watch_pidref(u, &pid, /* exclusive= */ false)); |
efdb0237 | 3672 | } |
efdb0237 LP |
3673 | } |
3674 | ||
3675 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
495e75ed LP |
3676 | if (r < 0) |
3677 | RET_GATHER(ret, r); | |
3678 | else { | |
3679 | for (;;) { | |
3680 | _cleanup_free_ char *fn = NULL, *p = NULL; | |
3681 | ||
3682 | r = cg_read_subgroup(d, &fn); | |
3683 | if (r == 0) | |
3684 | break; | |
3685 | if (r < 0) { | |
3686 | RET_GATHER(ret, r); | |
3687 | break; | |
3688 | } | |
efdb0237 | 3689 | |
95b21cff | 3690 | p = path_join(empty_to_root(path), fn); |
efdb0237 LP |
3691 | if (!p) |
3692 | return -ENOMEM; | |
3693 | ||
495e75ed | 3694 | RET_GATHER(ret, unit_watch_pids_in_path(u, p)); |
efdb0237 | 3695 | } |
efdb0237 LP |
3696 | } |
3697 | ||
3698 | return ret; | |
3699 | } | |
3700 | ||
11aef522 LP |
3701 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
3702 | int r; | |
3703 | ||
3704 | assert(u); | |
3705 | ||
3706 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
3707 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
3708 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
3709 | ||
9cc54544 LP |
3710 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3711 | if (!crt || !crt->cgroup_path) | |
11aef522 LP |
3712 | return -ENOENT; |
3713 | ||
3714 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
3715 | if (r < 0) | |
3716 | return r; | |
3717 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
3718 | return 0; | |
3719 | ||
3720 | if (!set_isempty(u->pids)) | |
3721 | return 0; | |
3722 | ||
3723 | unit_add_to_cgroup_empty_queue(u); | |
3724 | return 0; | |
3725 | } | |
3726 | ||
efdb0237 | 3727 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
3728 | int r; |
3729 | ||
efdb0237 LP |
3730 | assert(u); |
3731 | ||
3732 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
3733 | * watch. This is a fallback logic for cases where we do not | |
3734 | * get reliable cgroup empty notifications: we try to use | |
3735 | * SIGCHLD as replacement. */ | |
3736 | ||
9cc54544 LP |
3737 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3738 | if (!crt || !crt->cgroup_path) | |
efdb0237 LP |
3739 | return -ENOENT; |
3740 | ||
c22800e4 | 3741 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
3742 | if (r < 0) |
3743 | return r; | |
3744 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
3745 | return 0; |
3746 | ||
9cc54544 | 3747 | return unit_watch_pids_in_path(u, crt->cgroup_path); |
efdb0237 LP |
3748 | } |
3749 | ||
09e24654 | 3750 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
99534007 | 3751 | Manager *m = ASSERT_PTR(userdata); |
09e24654 | 3752 | Unit *u; |
efdb0237 LP |
3753 | int r; |
3754 | ||
09e24654 | 3755 | assert(s); |
efdb0237 | 3756 | |
09e24654 LP |
3757 | u = m->cgroup_empty_queue; |
3758 | if (!u) | |
efdb0237 LP |
3759 | return 0; |
3760 | ||
09e24654 LP |
3761 | assert(u->in_cgroup_empty_queue); |
3762 | u->in_cgroup_empty_queue = false; | |
3763 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
3764 | ||
3765 | if (m->cgroup_empty_queue) { | |
3766 | /* More stuff queued, let's make sure we remain enabled */ | |
3767 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
3768 | if (r < 0) | |
19a691a9 | 3769 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 3770 | } |
efdb0237 | 3771 | |
f7829525 NK |
3772 | /* Update state based on OOM kills before we notify about cgroup empty event */ |
3773 | (void) unit_check_oom(u); | |
3774 | (void) unit_check_oomd_kill(u); | |
3775 | ||
efdb0237 LP |
3776 | unit_add_to_gc_queue(u); |
3777 | ||
380dd177 RP |
3778 | if (IN_SET(unit_active_state(u), UNIT_INACTIVE, UNIT_FAILED)) |
3779 | unit_prune_cgroup(u); | |
3780 | else if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
efdb0237 LP |
3781 | UNIT_VTABLE(u)->notify_cgroup_empty(u); |
3782 | ||
3783 | return 0; | |
3784 | } | |
3785 | ||
09e24654 LP |
3786 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
3787 | int r; | |
3788 | ||
3789 | assert(u); | |
3790 | ||
3791 | /* Note that there are four different ways how cgroup empty events reach us: | |
3792 | * | |
3793 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
3794 | * | |
3795 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
3796 | * | |
3797 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
3798 | * | |
3799 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
3800 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
3801 | * | |
3802 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
3803 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
3804 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
3805 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
3806 | * case for scope units). */ | |
3807 | ||
3808 | if (u->in_cgroup_empty_queue) | |
3809 | return; | |
3810 | ||
3811 | /* Let's verify that the cgroup is really empty */ | |
9cc54544 LP |
3812 | r = unit_cgroup_is_empty(u); |
3813 | if (r <= 0) | |
09e24654 LP |
3814 | return; |
3815 | ||
3816 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
3817 | u->in_cgroup_empty_queue = true; | |
3818 | ||
3819 | /* Trigger the defer event */ | |
3820 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
3821 | if (r < 0) | |
3822 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
3823 | } | |
3824 | ||
d9e45bc3 MS |
3825 | static void unit_remove_from_cgroup_empty_queue(Unit *u) { |
3826 | assert(u); | |
3827 | ||
3828 | if (!u->in_cgroup_empty_queue) | |
3829 | return; | |
3830 | ||
3831 | LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
3832 | u->in_cgroup_empty_queue = false; | |
3833 | } | |
3834 | ||
fe8d22fb AZ |
3835 | int unit_check_oomd_kill(Unit *u) { |
3836 | _cleanup_free_ char *value = NULL; | |
3837 | bool increased; | |
3838 | uint64_t n = 0; | |
3839 | int r; | |
3840 | ||
9cc54544 LP |
3841 | assert(u); |
3842 | ||
3843 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
3844 | if (!crt || !crt->cgroup_path) | |
fe8d22fb AZ |
3845 | return 0; |
3846 | ||
3847 | r = cg_all_unified(); | |
3848 | if (r < 0) | |
3849 | return log_unit_debug_errno(u, r, "Couldn't determine whether we are in all unified mode: %m"); | |
3850 | else if (r == 0) | |
3851 | return 0; | |
3852 | ||
9cc54544 | 3853 | r = cg_get_xattr_malloc(crt->cgroup_path, "user.oomd_ooms", &value); |
00675c36 | 3854 | if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r)) |
fe8d22fb AZ |
3855 | return r; |
3856 | ||
3857 | if (!isempty(value)) { | |
3858 | r = safe_atou64(value, &n); | |
3859 | if (r < 0) | |
3860 | return r; | |
3861 | } | |
3862 | ||
9cc54544 LP |
3863 | increased = n > crt->managed_oom_kill_last; |
3864 | crt->managed_oom_kill_last = n; | |
fe8d22fb AZ |
3865 | |
3866 | if (!increased) | |
3867 | return 0; | |
3868 | ||
38c41427 NK |
3869 | n = 0; |
3870 | value = mfree(value); | |
9cc54544 | 3871 | r = cg_get_xattr_malloc(crt->cgroup_path, "user.oomd_kill", &value); |
38c41427 NK |
3872 | if (r >= 0 && !isempty(value)) |
3873 | (void) safe_atou64(value, &n); | |
3874 | ||
fe8d22fb | 3875 | if (n > 0) |
c2503e35 RH |
3876 | log_unit_struct(u, LOG_NOTICE, |
3877 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
3878 | LOG_UNIT_INVOCATION_ID(u), | |
38c41427 NK |
3879 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed %"PRIu64" process(es) in this unit.", n), |
3880 | "N_PROCESSES=%" PRIu64, n); | |
3881 | else | |
3882 | log_unit_struct(u, LOG_NOTICE, | |
3883 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
3884 | LOG_UNIT_INVOCATION_ID(u), | |
3885 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed some process(es) in this unit.")); | |
3886 | ||
3887 | unit_notify_cgroup_oom(u, /* ManagedOOM= */ true); | |
fe8d22fb AZ |
3888 | |
3889 | return 1; | |
3890 | } | |
3891 | ||
2ba6ae6b | 3892 | int unit_check_oom(Unit *u) { |
afcfaa69 LP |
3893 | _cleanup_free_ char *oom_kill = NULL; |
3894 | bool increased; | |
3895 | uint64_t c; | |
3896 | int r; | |
3897 | ||
9cc54544 LP |
3898 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
3899 | if (!crt || !crt->cgroup_path) | |
afcfaa69 LP |
3900 | return 0; |
3901 | ||
9cc54544 LP |
3902 | r = cg_get_keyed_attribute( |
3903 | "memory", | |
3904 | crt->cgroup_path, | |
3905 | "memory.events", | |
3906 | STRV_MAKE("oom_kill"), | |
3907 | &oom_kill); | |
fc594dee LP |
3908 | if (IN_SET(r, -ENOENT, -ENXIO)) /* Handle gracefully if cgroup or oom_kill attribute don't exist */ |
3909 | c = 0; | |
3910 | else if (r < 0) | |
afcfaa69 | 3911 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); |
fc594dee LP |
3912 | else { |
3913 | r = safe_atou64(oom_kill, &c); | |
3914 | if (r < 0) | |
3915 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
3916 | } | |
afcfaa69 | 3917 | |
9cc54544 LP |
3918 | increased = c > crt->oom_kill_last; |
3919 | crt->oom_kill_last = c; | |
afcfaa69 LP |
3920 | |
3921 | if (!increased) | |
3922 | return 0; | |
3923 | ||
c2503e35 RH |
3924 | log_unit_struct(u, LOG_NOTICE, |
3925 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
3926 | LOG_UNIT_INVOCATION_ID(u), | |
3927 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
afcfaa69 | 3928 | |
38c41427 | 3929 | unit_notify_cgroup_oom(u, /* ManagedOOM= */ false); |
afcfaa69 LP |
3930 | |
3931 | return 1; | |
3932 | } | |
3933 | ||
3934 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
99534007 | 3935 | Manager *m = ASSERT_PTR(userdata); |
afcfaa69 LP |
3936 | Unit *u; |
3937 | int r; | |
3938 | ||
3939 | assert(s); | |
afcfaa69 LP |
3940 | |
3941 | u = m->cgroup_oom_queue; | |
3942 | if (!u) | |
3943 | return 0; | |
3944 | ||
3945 | assert(u->in_cgroup_oom_queue); | |
3946 | u->in_cgroup_oom_queue = false; | |
3947 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
3948 | ||
3949 | if (m->cgroup_oom_queue) { | |
3950 | /* More stuff queued, let's make sure we remain enabled */ | |
3951 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
3952 | if (r < 0) | |
3953 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
3954 | } | |
3955 | ||
3956 | (void) unit_check_oom(u); | |
935f8042 LP |
3957 | unit_add_to_gc_queue(u); |
3958 | ||
afcfaa69 LP |
3959 | return 0; |
3960 | } | |
3961 | ||
3962 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
3963 | int r; | |
3964 | ||
3965 | assert(u); | |
3966 | ||
3967 | if (u->in_cgroup_oom_queue) | |
3968 | return; | |
9cc54544 LP |
3969 | |
3970 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
3971 | if (!crt || !crt->cgroup_path) | |
afcfaa69 LP |
3972 | return; |
3973 | ||
3974 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
3975 | u->in_cgroup_oom_queue = true; | |
3976 | ||
3977 | /* Trigger the defer event */ | |
3978 | if (!u->manager->cgroup_oom_event_source) { | |
3979 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
3980 | ||
3981 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
3982 | if (r < 0) { | |
3983 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
3984 | return; | |
3985 | } | |
3986 | ||
d42b61d2 | 3987 | r = sd_event_source_set_priority(s, EVENT_PRIORITY_CGROUP_OOM); |
afcfaa69 LP |
3988 | if (r < 0) { |
3989 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
3990 | return; | |
3991 | } | |
3992 | ||
3993 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
3994 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
3995 | } | |
3996 | ||
3997 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
3998 | if (r < 0) | |
3999 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
4000 | } | |
4001 | ||
d9e45bc3 MS |
4002 | static int unit_check_cgroup_events(Unit *u) { |
4003 | char *values[2] = {}; | |
4004 | int r; | |
4005 | ||
4006 | assert(u); | |
4007 | ||
9cc54544 LP |
4008 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4009 | if (!crt || !crt->cgroup_path) | |
869f52f2 DS |
4010 | return 0; |
4011 | ||
9cc54544 LP |
4012 | r = cg_get_keyed_attribute_graceful( |
4013 | SYSTEMD_CGROUP_CONTROLLER, | |
4014 | crt->cgroup_path, | |
4015 | "cgroup.events", | |
4016 | STRV_MAKE("populated", "frozen"), | |
4017 | values); | |
d9e45bc3 MS |
4018 | if (r < 0) |
4019 | return r; | |
4020 | ||
4021 | /* The cgroup.events notifications can be merged together so act as we saw the given state for the | |
4022 | * first time. The functions we call to handle given state are idempotent, which makes them | |
4023 | * effectively remember the previous state. */ | |
4024 | if (values[0]) { | |
4025 | if (streq(values[0], "1")) | |
4026 | unit_remove_from_cgroup_empty_queue(u); | |
4027 | else | |
4028 | unit_add_to_cgroup_empty_queue(u); | |
4029 | } | |
4030 | ||
16b6af6a AV |
4031 | /* Disregard freezer state changes due to operations not initiated by us. |
4032 | * See: https://github.com/systemd/systemd/pull/13512/files#r416469963 and | |
4033 | * https://github.com/systemd/systemd/pull/13512#issuecomment-573007207 */ | |
4034 | if (values[1] && IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_FREEZING_BY_PARENT, FREEZER_THAWING)) { | |
d9e45bc3 MS |
4035 | if (streq(values[1], "0")) |
4036 | unit_thawed(u); | |
4037 | else | |
4038 | unit_frozen(u); | |
4039 | } | |
4040 | ||
4041 | free(values[0]); | |
4042 | free(values[1]); | |
4043 | ||
4044 | return 0; | |
4045 | } | |
4046 | ||
efdb0237 | 4047 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
99534007 | 4048 | Manager *m = ASSERT_PTR(userdata); |
efdb0237 LP |
4049 | |
4050 | assert(s); | |
4051 | assert(fd >= 0); | |
efdb0237 LP |
4052 | |
4053 | for (;;) { | |
4054 | union inotify_event_buffer buffer; | |
efdb0237 LP |
4055 | ssize_t l; |
4056 | ||
4057 | l = read(fd, &buffer, sizeof(buffer)); | |
4058 | if (l < 0) { | |
8add30a0 | 4059 | if (ERRNO_IS_TRANSIENT(errno)) |
efdb0237 LP |
4060 | return 0; |
4061 | ||
4062 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
4063 | } | |
4064 | ||
00adc340 | 4065 | FOREACH_INOTIFY_EVENT_WARN(e, buffer, l) { |
efdb0237 LP |
4066 | Unit *u; |
4067 | ||
4068 | if (e->wd < 0) | |
4069 | /* Queue overflow has no watch descriptor */ | |
4070 | continue; | |
4071 | ||
4072 | if (e->mask & IN_IGNORED) | |
4073 | /* The watch was just removed */ | |
4074 | continue; | |
4075 | ||
afcfaa69 LP |
4076 | /* Note that inotify might deliver events for a watch even after it was removed, |
4077 | * because it was queued before the removal. Let's ignore this here safely. */ | |
4078 | ||
0bb814c2 | 4079 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 | 4080 | if (u) |
d9e45bc3 | 4081 | unit_check_cgroup_events(u); |
efdb0237 | 4082 | |
afcfaa69 LP |
4083 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
4084 | if (u) | |
4085 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
4086 | } |
4087 | } | |
8e274523 LP |
4088 | } |
4089 | ||
17f14955 RG |
4090 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
4091 | CGroupMask mask = 0; | |
4092 | int r; | |
4093 | ||
4094 | /* BPF-based firewall */ | |
4095 | r = bpf_firewall_supported(); | |
ad13559e YW |
4096 | if (r < 0) |
4097 | return r; | |
17f14955 RG |
4098 | if (r > 0) |
4099 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
4100 | ||
084c7007 RG |
4101 | /* BPF-based device access control */ |
4102 | r = bpf_devices_supported(); | |
ad13559e YW |
4103 | if (r < 0) |
4104 | return r; | |
084c7007 RG |
4105 | if (r > 0) |
4106 | mask |= CGROUP_MASK_BPF_DEVICES; | |
4107 | ||
506ea51b JK |
4108 | /* BPF pinned prog */ |
4109 | r = bpf_foreign_supported(); | |
ad13559e YW |
4110 | if (r < 0) |
4111 | return r; | |
506ea51b JK |
4112 | if (r > 0) |
4113 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
4114 | ||
a8e5eb17 | 4115 | /* BPF-based bind{4|6} hooks */ |
cd09a5f3 | 4116 | r = bpf_socket_bind_supported(); |
ad13559e YW |
4117 | if (r < 0) |
4118 | return r; | |
a8e5eb17 JK |
4119 | if (r > 0) |
4120 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
4121 | ||
6f50d4f7 | 4122 | /* BPF-based cgroup_skb/{egress|ingress} hooks */ |
62e22490 | 4123 | r = bpf_restrict_ifaces_supported(); |
ad13559e YW |
4124 | if (r < 0) |
4125 | return r; | |
6f50d4f7 MV |
4126 | if (r > 0) |
4127 | mask |= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES; | |
4128 | ||
17f14955 RG |
4129 | *ret = mask; |
4130 | return 0; | |
4131 | } | |
4132 | ||
8e274523 | 4133 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 4134 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 4135 | const char *scope_path; |
b4cccbc1 | 4136 | int r, all_unified; |
17f14955 | 4137 | CGroupMask mask; |
efdb0237 | 4138 | char *e; |
8e274523 LP |
4139 | |
4140 | assert(m); | |
4141 | ||
35d2e7ec | 4142 | /* 1. Determine hierarchy */ |
efdb0237 | 4143 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 4144 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
4145 | if (r < 0) |
4146 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 4147 | |
efdb0237 LP |
4148 | /* Chop off the init scope, if we are already located in it */ |
4149 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 4150 | |
efdb0237 LP |
4151 | /* LEGACY: Also chop off the system slice if we are in |
4152 | * it. This is to support live upgrades from older systemd | |
4153 | * versions where PID 1 was moved there. Also see | |
4154 | * cg_get_root_path(). */ | |
463d0d15 | 4155 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 4156 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 4157 | if (!e) |
efdb0237 | 4158 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 4159 | } |
efdb0237 LP |
4160 | if (e) |
4161 | *e = 0; | |
7ccfb64a | 4162 | |
7546145e LP |
4163 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
4164 | * easily prepend it everywhere. */ | |
4165 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 4166 | |
35d2e7ec | 4167 | /* 2. Show data */ |
9444b1f2 | 4168 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
4169 | if (r < 0) |
4170 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 4171 | |
d4d99bc6 | 4172 | r = cg_unified(); |
415fc41c TH |
4173 | if (r < 0) |
4174 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 4175 | |
b4cccbc1 | 4176 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
4177 | if (all_unified < 0) |
4178 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
4179 | if (all_unified > 0) | |
efdb0237 | 4180 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 4181 | else { |
c22800e4 | 4182 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
4183 | if (r < 0) |
4184 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
4185 | if (r > 0) | |
4186 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
4187 | else | |
4188 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
4189 | } | |
efdb0237 | 4190 | |
09e24654 | 4191 | /* 3. Allocate cgroup empty defer event source */ |
5dcadb4c | 4192 | m->cgroup_empty_event_source = sd_event_source_disable_unref(m->cgroup_empty_event_source); |
09e24654 LP |
4193 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); |
4194 | if (r < 0) | |
4195 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
4196 | ||
cbe83389 LP |
4197 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
4198 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
4199 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
d42b61d2 | 4200 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, EVENT_PRIORITY_CGROUP_EMPTY); |
09e24654 LP |
4201 | if (r < 0) |
4202 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
4203 | ||
4204 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
4205 | if (r < 0) | |
4206 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
4207 | ||
4208 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
4209 | ||
4210 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 4211 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 4212 | |
09e24654 | 4213 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 4214 | |
5dcadb4c | 4215 | m->cgroup_inotify_event_source = sd_event_source_disable_unref(m->cgroup_inotify_event_source); |
10bd3e2e | 4216 | safe_close(m->cgroup_inotify_fd); |
efdb0237 | 4217 | |
10bd3e2e LP |
4218 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
4219 | if (m->cgroup_inotify_fd < 0) | |
4220 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 4221 | |
10bd3e2e LP |
4222 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
4223 | if (r < 0) | |
4224 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 4225 | |
cbe83389 LP |
4226 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
4227 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
4228 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
d42b61d2 | 4229 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, EVENT_PRIORITY_CGROUP_INOTIFY); |
10bd3e2e LP |
4230 | if (r < 0) |
4231 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 4232 | |
10bd3e2e | 4233 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 4234 | |
611c4f8a | 4235 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 4236 | |
10bd3e2e LP |
4237 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
4238 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 4239 | |
ce906769 | 4240 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUPS_AGENT_PATH); |
23bbb0de | 4241 | if (r < 0) |
10bd3e2e LP |
4242 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
4243 | else if (r > 0) | |
4244 | log_debug("Installed release agent."); | |
4245 | else if (r == 0) | |
4246 | log_debug("Release agent already installed."); | |
4247 | } | |
efdb0237 | 4248 | |
09e24654 | 4249 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
4250 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
4251 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
4252 | if (r >= 0) { |
4253 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
4254 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
4255 | if (r < 0) | |
4256 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 4257 | |
aa77e234 MS |
4258 | /* 6. And pin it, so that it cannot be unmounted */ |
4259 | safe_close(m->pin_cgroupfs_fd); | |
4260 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
4261 | if (m->pin_cgroupfs_fd < 0) | |
4262 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 4263 | |
638cece4 | 4264 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 4265 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 4266 | |
09e24654 | 4267 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 4268 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 4269 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 4270 | |
17f14955 | 4271 | /* 8. Figure out which controllers are supported */ |
0fa7b500 | 4272 | r = cg_mask_supported_subtree(m->cgroup_root, &m->cgroup_supported); |
efdb0237 LP |
4273 | if (r < 0) |
4274 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
4275 | |
4276 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
4277 | r = cg_bpf_mask_supported(&mask); | |
4278 | if (r < 0) | |
4279 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
4280 | m->cgroup_supported |= mask; | |
4281 | ||
4282 | /* 10. Log which controllers are supported */ | |
e8616626 ZJS |
4283 | for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
4284 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), | |
4285 | yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); | |
9156e799 | 4286 | |
a32360f1 | 4287 | return 0; |
8e274523 LP |
4288 | } |
4289 | ||
c6c18be3 | 4290 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
4291 | assert(m); |
4292 | ||
9444b1f2 LP |
4293 | /* We can't really delete the group, since we are in it. But |
4294 | * let's trim it. */ | |
5dd2f5ff | 4295 | if (delete && m->cgroup_root && !FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL)) |
efdb0237 LP |
4296 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
4297 | ||
5dcadb4c | 4298 | m->cgroup_empty_event_source = sd_event_source_disable_unref(m->cgroup_empty_event_source); |
09e24654 | 4299 | |
0bb814c2 | 4300 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 4301 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 | 4302 | |
5dcadb4c | 4303 | m->cgroup_inotify_event_source = sd_event_source_disable_unref(m->cgroup_inotify_event_source); |
efdb0237 | 4304 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); |
8e274523 | 4305 | |
03e334a1 | 4306 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 4307 | |
efdb0237 | 4308 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
4309 | } |
4310 | ||
4ad49000 | 4311 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 4312 | char *p; |
4ad49000 | 4313 | Unit *u; |
acb14d31 LP |
4314 | |
4315 | assert(m); | |
4316 | assert(cgroup); | |
acb14d31 | 4317 | |
4ad49000 LP |
4318 | u = hashmap_get(m->cgroup_unit, cgroup); |
4319 | if (u) | |
4320 | return u; | |
acb14d31 | 4321 | |
2f82562b | 4322 | p = strdupa_safe(cgroup); |
acb14d31 LP |
4323 | for (;;) { |
4324 | char *e; | |
4325 | ||
4326 | e = strrchr(p, '/'); | |
efdb0237 LP |
4327 | if (!e || e == p) |
4328 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
4329 | |
4330 | *e = 0; | |
4331 | ||
4ad49000 LP |
4332 | u = hashmap_get(m->cgroup_unit, p); |
4333 | if (u) | |
4334 | return u; | |
acb14d31 LP |
4335 | } |
4336 | } | |
4337 | ||
d70dfe1b | 4338 | Unit *manager_get_unit_by_pidref_cgroup(Manager *m, const PidRef *pid) { |
4ad49000 | 4339 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 4340 | |
8c47c732 LP |
4341 | assert(m); |
4342 | ||
a9062242 | 4343 | if (cg_pidref_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
4344 | return NULL; |
4345 | ||
4346 | return manager_get_unit_by_cgroup(m, cgroup); | |
4347 | } | |
4348 | ||
d70dfe1b | 4349 | Unit *manager_get_unit_by_pidref_watching(Manager *m, const PidRef *pid) { |
62a76913 | 4350 | Unit *u, **array; |
b3ac818b LP |
4351 | |
4352 | assert(m); | |
4353 | ||
495e75ed LP |
4354 | if (!pidref_is_set(pid)) |
4355 | return NULL; | |
62a76913 | 4356 | |
495e75ed LP |
4357 | u = hashmap_get(m->watch_pids, pid); |
4358 | if (u) | |
4359 | return u; | |
4360 | ||
4361 | array = hashmap_get(m->watch_pids_more, pid); | |
4362 | if (array) | |
4363 | return array[0]; | |
4364 | ||
4365 | return NULL; | |
4366 | } | |
4367 | ||
d70dfe1b | 4368 | Unit *manager_get_unit_by_pidref(Manager *m, const PidRef *pid) { |
495e75ed LP |
4369 | Unit *u; |
4370 | ||
4371 | assert(m); | |
4372 | ||
4373 | /* Note that a process might be owned by multiple units, we return only one here, which is good | |
4374 | * enough for most cases, though not strictly correct. We prefer the one reported by cgroup | |
4375 | * membership, as that's the most relevant one as children of the process will be assigned to that | |
4376 | * one, too, before all else. */ | |
4377 | ||
4378 | if (!pidref_is_set(pid)) | |
8c47c732 LP |
4379 | return NULL; |
4380 | ||
a7a87769 | 4381 | if (pidref_is_self(pid)) |
efdb0237 | 4382 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
495e75ed LP |
4383 | if (pid->pid == 1) |
4384 | return NULL; | |
efdb0237 | 4385 | |
495e75ed | 4386 | u = manager_get_unit_by_pidref_cgroup(m, pid); |
5fe8876b LP |
4387 | if (u) |
4388 | return u; | |
4389 | ||
495e75ed | 4390 | u = manager_get_unit_by_pidref_watching(m, pid); |
5fe8876b LP |
4391 | if (u) |
4392 | return u; | |
4393 | ||
62a76913 | 4394 | return NULL; |
6dde1f33 | 4395 | } |
4fbf50b3 | 4396 | |
495e75ed LP |
4397 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { |
4398 | assert(m); | |
4399 | ||
4400 | if (!pid_is_valid(pid)) | |
4401 | return NULL; | |
4402 | ||
4403 | return manager_get_unit_by_pidref(m, &PIDREF_MAKE_FROM_PID(pid)); | |
4404 | } | |
4405 | ||
4ad49000 LP |
4406 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
4407 | Unit *u; | |
4fbf50b3 | 4408 | |
4ad49000 LP |
4409 | assert(m); |
4410 | assert(cgroup); | |
4fbf50b3 | 4411 | |
09e24654 LP |
4412 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
4413 | * or from the --system instance */ | |
4414 | ||
d8fdc620 LP |
4415 | log_debug("Got cgroup empty notification for: %s", cgroup); |
4416 | ||
4ad49000 | 4417 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
4418 | if (!u) |
4419 | return 0; | |
b56c28c3 | 4420 | |
09e24654 LP |
4421 | unit_add_to_cgroup_empty_queue(u); |
4422 | return 1; | |
5ad096b3 LP |
4423 | } |
4424 | ||
93ff34e4 | 4425 | int unit_get_memory_available(Unit *u, uint64_t *ret) { |
8db929a1 | 4426 | uint64_t available = UINT64_MAX, current = 0; |
93ff34e4 LB |
4427 | |
4428 | assert(u); | |
4429 | assert(ret); | |
4430 | ||
4431 | /* If data from cgroups can be accessed, try to find out how much more memory a unit can | |
4432 | * claim before hitting the configured cgroup limits (if any). Consider both MemoryHigh | |
4433 | * and MemoryMax, and also any slice the unit might be nested below. */ | |
4434 | ||
727cea76 | 4435 | do { |
8db929a1 | 4436 | uint64_t unit_available, unit_limit = UINT64_MAX; |
727cea76 | 4437 | CGroupContext *unit_context; |
93ff34e4 LB |
4438 | |
4439 | /* No point in continuing if we can't go any lower */ | |
4440 | if (available == 0) | |
4441 | break; | |
4442 | ||
727cea76 MK |
4443 | unit_context = unit_get_cgroup_context(u); |
4444 | if (!unit_context) | |
4445 | return -ENODATA; | |
93ff34e4 | 4446 | |
9cc54544 LP |
4447 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4448 | if (!crt || !crt->cgroup_path) | |
93ff34e4 LB |
4449 | continue; |
4450 | ||
8db929a1 MK |
4451 | (void) unit_get_memory_current(u, ¤t); |
4452 | /* in case of error, previous current propagates as lower bound */ | |
4453 | ||
727cea76 MK |
4454 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
4455 | unit_limit = physical_memory(); | |
4456 | else if (unit_context->memory_max == UINT64_MAX && unit_context->memory_high == UINT64_MAX) | |
93ff34e4 | 4457 | continue; |
727cea76 | 4458 | unit_limit = MIN3(unit_limit, unit_context->memory_max, unit_context->memory_high); |
93ff34e4 | 4459 | |
8db929a1 | 4460 | unit_available = LESS_BY(unit_limit, current); |
727cea76 MK |
4461 | available = MIN(unit_available, available); |
4462 | } while ((u = UNIT_GET_SLICE(u))); | |
93ff34e4 LB |
4463 | |
4464 | *ret = available; | |
4465 | ||
4466 | return 0; | |
4467 | } | |
4468 | ||
5ad096b3 | 4469 | int unit_get_memory_current(Unit *u, uint64_t *ret) { |
5ad096b3 LP |
4470 | int r; |
4471 | ||
9824ab1f MY |
4472 | // FIXME: Merge this into unit_get_memory_accounting after support for cgroup v1 is dropped |
4473 | ||
5ad096b3 LP |
4474 | assert(u); |
4475 | assert(ret); | |
4476 | ||
2e4025c0 | 4477 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
4478 | return -ENODATA; |
4479 | ||
9cc54544 LP |
4480 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4481 | if (!crt || !crt->cgroup_path) | |
5ad096b3 LP |
4482 | return -ENODATA; |
4483 | ||
1f73aa00 | 4484 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 4485 | if (unit_has_host_root_cgroup(u)) |
c482724a | 4486 | return procfs_memory_get_used(ret); |
1f73aa00 | 4487 | |
9cc54544 | 4488 | if ((crt->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
4489 | return -ENODATA; |
4490 | ||
b4cccbc1 LP |
4491 | r = cg_all_unified(); |
4492 | if (r < 0) | |
4493 | return r; | |
5ad096b3 | 4494 | |
9cc54544 | 4495 | return cg_get_attribute_as_uint64("memory", crt->cgroup_path, r > 0 ? "memory.current" : "memory.usage_in_bytes", ret); |
5ad096b3 LP |
4496 | } |
4497 | ||
9824ab1f MY |
4498 | int unit_get_memory_accounting(Unit *u, CGroupMemoryAccountingMetric metric, uint64_t *ret) { |
4499 | ||
4500 | static const char* const attributes_table[_CGROUP_MEMORY_ACCOUNTING_METRIC_MAX] = { | |
4501 | [CGROUP_MEMORY_PEAK] = "memory.peak", | |
4502 | [CGROUP_MEMORY_SWAP_CURRENT] = "memory.swap.current", | |
4503 | [CGROUP_MEMORY_SWAP_PEAK] = "memory.swap.peak", | |
4504 | [CGROUP_MEMORY_ZSWAP_CURRENT] = "memory.zswap.current", | |
4505 | }; | |
4506 | ||
4507 | uint64_t bytes; | |
f17b07f4 | 4508 | bool updated = false; |
6c71db76 FS |
4509 | int r; |
4510 | ||
4511 | assert(u); | |
9824ab1f MY |
4512 | assert(metric >= 0); |
4513 | assert(metric < _CGROUP_MEMORY_ACCOUNTING_METRIC_MAX); | |
6c71db76 | 4514 | |
37533c94 FS |
4515 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
4516 | return -ENODATA; | |
4517 | ||
9cc54544 LP |
4518 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4519 | if (!crt) | |
4520 | return -ENODATA; | |
4521 | if (!crt->cgroup_path) | |
f17b07f4 | 4522 | /* If the cgroup is already gone, we try to find the last cached value. */ |
a8aed6a9 | 4523 | goto finish; |
6c71db76 FS |
4524 | |
4525 | /* The root cgroup doesn't expose this information. */ | |
4526 | if (unit_has_host_root_cgroup(u)) | |
4527 | return -ENODATA; | |
4528 | ||
9cc54544 | 4529 | if (!FLAGS_SET(crt->cgroup_realized_mask, CGROUP_MASK_MEMORY)) |
6c71db76 FS |
4530 | return -ENODATA; |
4531 | ||
4532 | r = cg_all_unified(); | |
4533 | if (r < 0) | |
4534 | return r; | |
9824ab1f | 4535 | if (r == 0) |
6c71db76 FS |
4536 | return -ENODATA; |
4537 | ||
9cc54544 | 4538 | r = cg_get_attribute_as_uint64("memory", crt->cgroup_path, attributes_table[metric], &bytes); |
f17b07f4 | 4539 | if (r < 0 && r != -ENODATA) |
9824ab1f | 4540 | return r; |
f17b07f4 | 4541 | updated = r >= 0; |
6c71db76 | 4542 | |
a8aed6a9 MY |
4543 | finish: |
4544 | if (metric <= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST) { | |
9cc54544 | 4545 | uint64_t *last = &crt->memory_accounting_last[metric]; |
6c71db76 | 4546 | |
a8aed6a9 MY |
4547 | if (updated) |
4548 | *last = bytes; | |
4549 | else if (*last != UINT64_MAX) | |
4550 | bytes = *last; | |
4551 | else | |
4552 | return -ENODATA; | |
f17b07f4 | 4553 | |
a8aed6a9 | 4554 | } else if (!updated) |
f17b07f4 | 4555 | return -ENODATA; |
6c71db76 | 4556 | |
6c71db76 FS |
4557 | if (ret) |
4558 | *ret = bytes; | |
4559 | ||
4560 | return 0; | |
4561 | } | |
4562 | ||
03a7b521 | 4563 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
03a7b521 LP |
4564 | assert(u); |
4565 | assert(ret); | |
4566 | ||
2e4025c0 | 4567 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
4568 | return -ENODATA; |
4569 | ||
9cc54544 LP |
4570 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4571 | if (!crt || !crt->cgroup_path) | |
03a7b521 LP |
4572 | return -ENODATA; |
4573 | ||
c36a69f4 | 4574 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 4575 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
4576 | return procfs_tasks_get_current(ret); |
4577 | ||
9cc54544 | 4578 | if ((crt->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
1f73aa00 LP |
4579 | return -ENODATA; |
4580 | ||
9cc54544 | 4581 | return cg_get_attribute_as_uint64("pids", crt->cgroup_path, "pids.current", ret); |
03a7b521 LP |
4582 | } |
4583 | ||
5ad096b3 | 4584 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
5ad096b3 LP |
4585 | uint64_t ns; |
4586 | int r; | |
4587 | ||
4588 | assert(u); | |
4589 | assert(ret); | |
4590 | ||
9cc54544 LP |
4591 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4592 | if (!crt || !crt->cgroup_path) | |
5ad096b3 LP |
4593 | return -ENODATA; |
4594 | ||
1f73aa00 | 4595 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 4596 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
4597 | return procfs_cpu_get_usage(ret); |
4598 | ||
f98c2585 | 4599 | /* Requisite controllers for CPU accounting are not enabled */ |
9cc54544 | 4600 | if ((get_cpu_accounting_mask() & ~crt->cgroup_realized_mask) != 0) |
f98c2585 CD |
4601 | return -ENODATA; |
4602 | ||
92a99304 LP |
4603 | r = cg_all_unified(); |
4604 | if (r < 0) | |
4605 | return r; | |
b4cccbc1 | 4606 | if (r > 0) { |
66ebf6c0 TH |
4607 | _cleanup_free_ char *val = NULL; |
4608 | uint64_t us; | |
5ad096b3 | 4609 | |
9cc54544 | 4610 | r = cg_get_keyed_attribute("cpu", crt->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
4611 | if (IN_SET(r, -ENOENT, -ENXIO)) |
4612 | return -ENODATA; | |
d742f4b5 LP |
4613 | if (r < 0) |
4614 | return r; | |
66ebf6c0 TH |
4615 | |
4616 | r = safe_atou64(val, &us); | |
4617 | if (r < 0) | |
4618 | return r; | |
4619 | ||
4620 | ns = us * NSEC_PER_USEC; | |
613328c3 | 4621 | } else |
9cc54544 | 4622 | return cg_get_attribute_as_uint64("cpuacct", crt->cgroup_path, "cpuacct.usage", ret); |
5ad096b3 LP |
4623 | |
4624 | *ret = ns; | |
4625 | return 0; | |
4626 | } | |
4627 | ||
4628 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
4629 | nsec_t ns; | |
4630 | int r; | |
4631 | ||
fe700f46 LP |
4632 | assert(u); |
4633 | ||
4634 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
4635 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
4636 | * call this function with a NULL return value. */ | |
4637 | ||
9cc54544 LP |
4638 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4639 | if (!crt || !crt->cgroup_path) | |
4640 | return -ENODATA; | |
4641 | ||
2e4025c0 | 4642 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
4643 | return -ENODATA; |
4644 | ||
5ad096b3 | 4645 | r = unit_get_cpu_usage_raw(u, &ns); |
9cc54544 | 4646 | if (r == -ENODATA && crt->cpu_usage_last != NSEC_INFINITY) { |
fe700f46 LP |
4647 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our |
4648 | * cached value. */ | |
4649 | ||
4650 | if (ret) | |
9cc54544 | 4651 | *ret = crt->cpu_usage_last; |
fe700f46 LP |
4652 | return 0; |
4653 | } | |
5ad096b3 LP |
4654 | if (r < 0) |
4655 | return r; | |
4656 | ||
9cc54544 LP |
4657 | if (ns > crt->cpu_usage_base) |
4658 | ns -= crt->cpu_usage_base; | |
5ad096b3 LP |
4659 | else |
4660 | ns = 0; | |
4661 | ||
9cc54544 | 4662 | crt->cpu_usage_last = ns; |
fe700f46 LP |
4663 | if (ret) |
4664 | *ret = ns; | |
4665 | ||
5ad096b3 LP |
4666 | return 0; |
4667 | } | |
4668 | ||
906c06f6 DM |
4669 | int unit_get_ip_accounting( |
4670 | Unit *u, | |
4671 | CGroupIPAccountingMetric metric, | |
4672 | uint64_t *ret) { | |
4673 | ||
6b659ed8 | 4674 | uint64_t value; |
906c06f6 DM |
4675 | int fd, r; |
4676 | ||
4677 | assert(u); | |
4678 | assert(metric >= 0); | |
4679 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
4680 | assert(ret); | |
4681 | ||
2e4025c0 | 4682 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
4683 | return -ENODATA; |
4684 | ||
9cc54544 LP |
4685 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4686 | if (!crt || !crt->cgroup_path) | |
4687 | return -ENODATA; | |
4688 | ||
906c06f6 | 4689 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
9cc54544 LP |
4690 | crt->ip_accounting_ingress_map_fd : |
4691 | crt->ip_accounting_egress_map_fd; | |
906c06f6 DM |
4692 | if (fd < 0) |
4693 | return -ENODATA; | |
4694 | ||
4695 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 4696 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 4697 | else |
6b659ed8 LP |
4698 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
4699 | if (r < 0) | |
4700 | return r; | |
4701 | ||
4702 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
4703 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
4704 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
4705 | ||
9cc54544 | 4706 | *ret = value + crt->ip_accounting_extra[metric]; |
906c06f6 DM |
4707 | |
4708 | return r; | |
4709 | } | |
4710 | ||
4fb0d2dc MK |
4711 | static uint64_t unit_get_effective_limit_one(Unit *u, CGroupLimitType type) { |
4712 | CGroupContext *cc; | |
4713 | ||
4714 | assert(u); | |
4715 | assert(UNIT_HAS_CGROUP_CONTEXT(u)); | |
4716 | ||
93f8e88d MK |
4717 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
4718 | switch (type) { | |
4719 | case CGROUP_LIMIT_MEMORY_MAX: | |
4720 | case CGROUP_LIMIT_MEMORY_HIGH: | |
4721 | return physical_memory(); | |
4722 | case CGROUP_LIMIT_TASKS_MAX: | |
4723 | return system_tasks_max(); | |
4724 | default: | |
4725 | assert_not_reached(); | |
4726 | } | |
4727 | ||
c658ad79 | 4728 | cc = ASSERT_PTR(unit_get_cgroup_context(u)); |
4fb0d2dc MK |
4729 | switch (type) { |
4730 | /* Note: on legacy/hybrid hierarchies memory_max stays CGROUP_LIMIT_MAX unless configured | |
4731 | * explicitly. Effective value of MemoryLimit= (cgroup v1) is not implemented. */ | |
4732 | case CGROUP_LIMIT_MEMORY_MAX: | |
4733 | return cc->memory_max; | |
4734 | case CGROUP_LIMIT_MEMORY_HIGH: | |
4735 | return cc->memory_high; | |
4736 | case CGROUP_LIMIT_TASKS_MAX: | |
4737 | return cgroup_tasks_max_resolve(&cc->tasks_max); | |
4738 | default: | |
4739 | assert_not_reached(); | |
4740 | } | |
4741 | } | |
4742 | ||
4743 | int unit_get_effective_limit(Unit *u, CGroupLimitType type, uint64_t *ret) { | |
4744 | uint64_t infimum; | |
4745 | ||
4746 | assert(u); | |
4747 | assert(ret); | |
4748 | assert(type >= 0); | |
4749 | assert(type < _CGROUP_LIMIT_TYPE_MAX); | |
4750 | ||
4751 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4752 | return -EINVAL; | |
4753 | ||
4754 | infimum = unit_get_effective_limit_one(u, type); | |
4755 | for (Unit *slice = UNIT_GET_SLICE(u); slice; slice = UNIT_GET_SLICE(slice)) | |
4756 | infimum = MIN(infimum, unit_get_effective_limit_one(slice, type)); | |
4757 | ||
4758 | *ret = infimum; | |
4759 | return 0; | |
4760 | } | |
4761 | ||
fbe14fc9 LP |
4762 | static int unit_get_io_accounting_raw(Unit *u, uint64_t ret[static _CGROUP_IO_ACCOUNTING_METRIC_MAX]) { |
4763 | static const char *const field_names[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
4764 | [CGROUP_IO_READ_BYTES] = "rbytes=", | |
4765 | [CGROUP_IO_WRITE_BYTES] = "wbytes=", | |
4766 | [CGROUP_IO_READ_OPERATIONS] = "rios=", | |
4767 | [CGROUP_IO_WRITE_OPERATIONS] = "wios=", | |
4768 | }; | |
4769 | uint64_t acc[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {}; | |
4770 | _cleanup_free_ char *path = NULL; | |
4771 | _cleanup_fclose_ FILE *f = NULL; | |
4772 | int r; | |
4773 | ||
4774 | assert(u); | |
4775 | ||
9cc54544 LP |
4776 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4777 | if (!crt || !crt->cgroup_path) | |
fbe14fc9 LP |
4778 | return -ENODATA; |
4779 | ||
4780 | if (unit_has_host_root_cgroup(u)) | |
4781 | return -ENODATA; /* TODO: return useful data for the top-level cgroup */ | |
4782 | ||
4783 | r = cg_all_unified(); | |
4784 | if (r < 0) | |
4785 | return r; | |
4786 | if (r == 0) /* TODO: support cgroupv1 */ | |
4787 | return -ENODATA; | |
4788 | ||
9cc54544 | 4789 | if (!FLAGS_SET(crt->cgroup_realized_mask, CGROUP_MASK_IO)) |
fbe14fc9 LP |
4790 | return -ENODATA; |
4791 | ||
9cc54544 | 4792 | r = cg_get_path("io", crt->cgroup_path, "io.stat", &path); |
fbe14fc9 LP |
4793 | if (r < 0) |
4794 | return r; | |
4795 | ||
4796 | f = fopen(path, "re"); | |
4797 | if (!f) | |
4798 | return -errno; | |
4799 | ||
4800 | for (;;) { | |
4801 | _cleanup_free_ char *line = NULL; | |
4802 | const char *p; | |
4803 | ||
4804 | r = read_line(f, LONG_LINE_MAX, &line); | |
4805 | if (r < 0) | |
4806 | return r; | |
4807 | if (r == 0) | |
4808 | break; | |
4809 | ||
4810 | p = line; | |
4811 | p += strcspn(p, WHITESPACE); /* Skip over device major/minor */ | |
4812 | p += strspn(p, WHITESPACE); /* Skip over following whitespace */ | |
4813 | ||
4814 | for (;;) { | |
4815 | _cleanup_free_ char *word = NULL; | |
4816 | ||
4817 | r = extract_first_word(&p, &word, NULL, EXTRACT_RETAIN_ESCAPE); | |
4818 | if (r < 0) | |
4819 | return r; | |
4820 | if (r == 0) | |
4821 | break; | |
4822 | ||
4823 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
4824 | const char *x; | |
4825 | ||
4826 | x = startswith(word, field_names[i]); | |
4827 | if (x) { | |
4828 | uint64_t w; | |
4829 | ||
4830 | r = safe_atou64(x, &w); | |
4831 | if (r < 0) | |
4832 | return r; | |
4833 | ||
4834 | /* Sum up the stats of all devices */ | |
4835 | acc[i] += w; | |
4836 | break; | |
4837 | } | |
4838 | } | |
4839 | } | |
4840 | } | |
4841 | ||
4842 | memcpy(ret, acc, sizeof(acc)); | |
4843 | return 0; | |
4844 | } | |
4845 | ||
4846 | int unit_get_io_accounting( | |
4847 | Unit *u, | |
4848 | CGroupIOAccountingMetric metric, | |
4849 | bool allow_cache, | |
4850 | uint64_t *ret) { | |
4851 | ||
4852 | uint64_t raw[_CGROUP_IO_ACCOUNTING_METRIC_MAX]; | |
4853 | int r; | |
4854 | ||
4855 | /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */ | |
4856 | ||
4857 | if (!UNIT_CGROUP_BOOL(u, io_accounting)) | |
4858 | return -ENODATA; | |
4859 | ||
9cc54544 LP |
4860 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4861 | if (!crt || !crt->cgroup_path) | |
4862 | return -ENODATA; | |
4863 | ||
4864 | if (allow_cache && crt->io_accounting_last[metric] != UINT64_MAX) | |
fbe14fc9 LP |
4865 | goto done; |
4866 | ||
4867 | r = unit_get_io_accounting_raw(u, raw); | |
9cc54544 | 4868 | if (r == -ENODATA && crt->io_accounting_last[metric] != UINT64_MAX) |
fbe14fc9 LP |
4869 | goto done; |
4870 | if (r < 0) | |
4871 | return r; | |
4872 | ||
4873 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
4874 | /* Saturated subtraction */ | |
9cc54544 LP |
4875 | if (raw[i] > crt->io_accounting_base[i]) |
4876 | crt->io_accounting_last[i] = raw[i] - crt->io_accounting_base[i]; | |
fbe14fc9 | 4877 | else |
9cc54544 | 4878 | crt->io_accounting_last[i] = 0; |
fbe14fc9 LP |
4879 | } |
4880 | ||
4881 | done: | |
4882 | if (ret) | |
9cc54544 | 4883 | *ret = crt->io_accounting_last[metric]; |
fbe14fc9 LP |
4884 | |
4885 | return 0; | |
4886 | } | |
4887 | ||
906c06f6 | 4888 | int unit_reset_cpu_accounting(Unit *u) { |
5ad096b3 LP |
4889 | int r; |
4890 | ||
4891 | assert(u); | |
4892 | ||
9cc54544 LP |
4893 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4894 | if (!crt || !crt->cgroup_path) | |
4895 | return 0; | |
4896 | ||
4897 | crt->cpu_usage_last = NSEC_INFINITY; | |
fe700f46 | 4898 | |
9cc54544 | 4899 | r = unit_get_cpu_usage_raw(u, &crt->cpu_usage_base); |
5ad096b3 | 4900 | if (r < 0) { |
9cc54544 | 4901 | crt->cpu_usage_base = 0; |
5ad096b3 | 4902 | return r; |
b56c28c3 | 4903 | } |
2633eb83 | 4904 | |
4ad49000 | 4905 | return 0; |
4fbf50b3 LP |
4906 | } |
4907 | ||
d4bdc202 MY |
4908 | void unit_reset_memory_accounting_last(Unit *u) { |
4909 | assert(u); | |
4910 | ||
9cc54544 LP |
4911 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4912 | if (!crt || !crt->cgroup_path) | |
4913 | return; | |
4914 | ||
4915 | FOREACH_ARRAY(i, crt->memory_accounting_last, ELEMENTSOF(crt->memory_accounting_last)) | |
d4bdc202 MY |
4916 | *i = UINT64_MAX; |
4917 | } | |
4918 | ||
906c06f6 | 4919 | int unit_reset_ip_accounting(Unit *u) { |
cbd2abbb | 4920 | int r = 0; |
906c06f6 DM |
4921 | |
4922 | assert(u); | |
4923 | ||
9cc54544 LP |
4924 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4925 | if (!crt || !crt->cgroup_path) | |
4926 | return 0; | |
4927 | ||
4928 | if (crt->ip_accounting_ingress_map_fd >= 0) | |
4929 | RET_GATHER(r, bpf_firewall_reset_accounting(crt->ip_accounting_ingress_map_fd)); | |
906c06f6 | 4930 | |
9cc54544 LP |
4931 | if (crt->ip_accounting_egress_map_fd >= 0) |
4932 | RET_GATHER(r, bpf_firewall_reset_accounting(crt->ip_accounting_egress_map_fd)); | |
906c06f6 | 4933 | |
9cc54544 | 4934 | zero(crt->ip_accounting_extra); |
6b659ed8 | 4935 | |
cbd2abbb | 4936 | return r; |
906c06f6 DM |
4937 | } |
4938 | ||
d4bdc202 MY |
4939 | void unit_reset_io_accounting_last(Unit *u) { |
4940 | assert(u); | |
4941 | ||
9cc54544 LP |
4942 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4943 | if (!crt || !crt->cgroup_path) | |
4944 | return; | |
4945 | ||
4946 | FOREACH_ARRAY(i, crt->io_accounting_last, _CGROUP_IO_ACCOUNTING_METRIC_MAX) | |
d4bdc202 MY |
4947 | *i = UINT64_MAX; |
4948 | } | |
4949 | ||
fbe14fc9 LP |
4950 | int unit_reset_io_accounting(Unit *u) { |
4951 | int r; | |
4952 | ||
4953 | assert(u); | |
4954 | ||
9cc54544 LP |
4955 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4956 | if (!crt || !crt->cgroup_path) | |
4957 | return 0; | |
4958 | ||
d4bdc202 | 4959 | unit_reset_io_accounting_last(u); |
fbe14fc9 | 4960 | |
9cc54544 | 4961 | r = unit_get_io_accounting_raw(u, crt->io_accounting_base); |
fbe14fc9 | 4962 | if (r < 0) { |
9cc54544 | 4963 | zero(crt->io_accounting_base); |
fbe14fc9 LP |
4964 | return r; |
4965 | } | |
4966 | ||
4967 | return 0; | |
4968 | } | |
4969 | ||
9b2559a1 | 4970 | int unit_reset_accounting(Unit *u) { |
cbd2abbb | 4971 | int r = 0; |
9b2559a1 LP |
4972 | |
4973 | assert(u); | |
4974 | ||
cbd2abbb MY |
4975 | RET_GATHER(r, unit_reset_cpu_accounting(u)); |
4976 | RET_GATHER(r, unit_reset_io_accounting(u)); | |
4977 | RET_GATHER(r, unit_reset_ip_accounting(u)); | |
d4bdc202 | 4978 | unit_reset_memory_accounting_last(u); |
9b2559a1 | 4979 | |
cbd2abbb | 4980 | return r; |
9b2559a1 LP |
4981 | } |
4982 | ||
e7ab4d1a LP |
4983 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
4984 | assert(u); | |
4985 | ||
4986 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4987 | return; | |
4988 | ||
9cc54544 LP |
4989 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
4990 | if (!crt) | |
4991 | return; | |
4992 | ||
e7ab4d1a LP |
4993 | if (m == 0) |
4994 | return; | |
4995 | ||
538b4852 TH |
4996 | /* always invalidate compat pairs together */ |
4997 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
4998 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
4999 | ||
7cce4fb7 LP |
5000 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
5001 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
5002 | ||
9cc54544 | 5003 | if (FLAGS_SET(crt->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
5004 | return; |
5005 | ||
9cc54544 | 5006 | crt->cgroup_invalidated_mask |= m; |
91a6073e | 5007 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
5008 | } |
5009 | ||
906c06f6 DM |
5010 | void unit_invalidate_cgroup_bpf(Unit *u) { |
5011 | assert(u); | |
5012 | ||
5013 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
5014 | return; | |
5015 | ||
9cc54544 LP |
5016 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
5017 | if (!crt) | |
906c06f6 DM |
5018 | return; |
5019 | ||
9cc54544 LP |
5020 | if (crt->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
5021 | return; | |
5022 | ||
5023 | crt->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; | |
91a6073e | 5024 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
5025 | |
5026 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
5027 | * list of our children includes our own. */ | |
5028 | if (u->type == UNIT_SLICE) { | |
5029 | Unit *member; | |
906c06f6 | 5030 | |
d219a2b0 | 5031 | UNIT_FOREACH_DEPENDENCY(member, u, UNIT_ATOM_SLICE_OF) |
15ed3c3a | 5032 | unit_invalidate_cgroup_bpf(member); |
906c06f6 DM |
5033 | } |
5034 | } | |
5035 | ||
869f52f2 DS |
5036 | void unit_cgroup_catchup(Unit *u) { |
5037 | assert(u); | |
5038 | ||
5039 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
5040 | return; | |
5041 | ||
5042 | /* We dropped the inotify watch during reexec/reload, so we need to | |
5043 | * check these as they may have changed. | |
5044 | * Note that (currently) the kernel doesn't actually update cgroup | |
5045 | * file modification times, so we can't just serialize and then check | |
5046 | * the mtime for file(s) we are interested in. */ | |
5047 | (void) unit_check_cgroup_events(u); | |
5048 | unit_add_to_cgroup_oom_queue(u); | |
5049 | } | |
5050 | ||
1d9cc876 LP |
5051 | bool unit_cgroup_delegate(Unit *u) { |
5052 | CGroupContext *c; | |
5053 | ||
5054 | assert(u); | |
5055 | ||
5056 | if (!UNIT_VTABLE(u)->can_delegate) | |
5057 | return false; | |
5058 | ||
5059 | c = unit_get_cgroup_context(u); | |
5060 | if (!c) | |
5061 | return false; | |
5062 | ||
5063 | return c->delegate; | |
5064 | } | |
5065 | ||
e7ab4d1a | 5066 | void manager_invalidate_startup_units(Manager *m) { |
e7ab4d1a LP |
5067 | Unit *u; |
5068 | ||
5069 | assert(m); | |
5070 | ||
90e74a66 | 5071 | SET_FOREACH(u, m->startup_units) |
9dfb6a3a | 5072 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO|CGROUP_MASK_CPUSET); |
e7ab4d1a LP |
5073 | } |
5074 | ||
16b6af6a AV |
5075 | static int unit_cgroup_freezer_kernel_state(Unit *u, FreezerState *ret) { |
5076 | _cleanup_free_ char *val = NULL; | |
5077 | FreezerState s; | |
5078 | int r; | |
d9e45bc3 MS |
5079 | |
5080 | assert(u); | |
16b6af6a | 5081 | assert(ret); |
d9e45bc3 | 5082 | |
9cc54544 LP |
5083 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
5084 | if (!crt || !crt->cgroup_path) | |
5085 | return -EOWNERDEAD; | |
5086 | ||
5087 | r = cg_get_keyed_attribute( | |
5088 | SYSTEMD_CGROUP_CONTROLLER, | |
5089 | crt->cgroup_path, | |
5090 | "cgroup.events", | |
5091 | STRV_MAKE("frozen"), | |
5092 | &val); | |
16b6af6a AV |
5093 | if (IN_SET(r, -ENOENT, -ENXIO)) |
5094 | return -ENODATA; | |
5095 | if (r < 0) | |
5096 | return r; | |
9a1e90ae | 5097 | |
16b6af6a AV |
5098 | if (streq(val, "0")) |
5099 | s = FREEZER_RUNNING; | |
5100 | else if (streq(val, "1")) | |
5101 | s = FREEZER_FROZEN; | |
5102 | else { | |
5103 | log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL), "Unexpected cgroup frozen state: %s", val); | |
5104 | s = _FREEZER_STATE_INVALID; | |
5105 | } | |
a14137d9 | 5106 | |
16b6af6a AV |
5107 | *ret = s; |
5108 | return 0; | |
5109 | } | |
d9e45bc3 | 5110 | |
16b6af6a AV |
5111 | int unit_cgroup_freezer_action(Unit *u, FreezerAction action) { |
5112 | _cleanup_free_ char *path = NULL; | |
5113 | FreezerState target, current, next; | |
5114 | int r; | |
a14137d9 | 5115 | |
16b6af6a AV |
5116 | assert(u); |
5117 | assert(IN_SET(action, FREEZER_FREEZE, FREEZER_PARENT_FREEZE, | |
5118 | FREEZER_THAW, FREEZER_PARENT_THAW)); | |
5119 | ||
9cc54544 | 5120 | if (!cg_freezer_supported()) |
16b6af6a | 5121 | return 0; |
a14137d9 | 5122 | |
9cc54544 LP |
5123 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
5124 | if (!crt || !crt->cgroup_realized) | |
5125 | return -EBUSY; | |
5126 | ||
16b6af6a | 5127 | unit_next_freezer_state(u, action, &next, &target); |
d9e45bc3 | 5128 | |
16b6af6a | 5129 | r = unit_cgroup_freezer_kernel_state(u, ¤t); |
d9e45bc3 | 5130 | if (r < 0) |
16b6af6a | 5131 | return r; |
d9e45bc3 | 5132 | |
16b6af6a AV |
5133 | if (current == target) |
5134 | next = freezer_state_finish(next); | |
5135 | else if (IN_SET(next, FREEZER_FROZEN, FREEZER_FROZEN_BY_PARENT, FREEZER_RUNNING)) { | |
5136 | /* We're transitioning into a finished state, which implies that the cgroup's | |
5137 | * current state already matches the target and thus we'd return 0. But, reality | |
5138 | * shows otherwise. This indicates that our freezer_state tracking has diverged | |
5139 | * from the real state of the cgroup, which can happen if someone meddles with the | |
5140 | * cgroup from underneath us. This really shouldn't happen during normal operation, | |
5141 | * though. So, let's warn about it and fix up the state to be valid */ | |
5142 | ||
5143 | log_unit_warning(u, "Unit wants to transition to %s freezer state but cgroup is unexpectedly %s, fixing up.", | |
5144 | freezer_state_to_string(next), freezer_state_to_string(current) ?: "(invalid)"); | |
5145 | ||
5146 | if (next == FREEZER_FROZEN) | |
5147 | next = FREEZER_FREEZING; | |
5148 | else if (next == FREEZER_FROZEN_BY_PARENT) | |
5149 | next = FREEZER_FREEZING_BY_PARENT; | |
5150 | else if (next == FREEZER_RUNNING) | |
5151 | next = FREEZER_THAWING; | |
5152 | } | |
d9e45bc3 | 5153 | |
9cc54544 | 5154 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, crt->cgroup_path, "cgroup.freeze", &path); |
d9e45bc3 MS |
5155 | if (r < 0) |
5156 | return r; | |
5157 | ||
16b6af6a AV |
5158 | log_unit_debug(u, "Unit freezer state was %s, now %s.", |
5159 | freezer_state_to_string(u->freezer_state), | |
5160 | freezer_state_to_string(next)); | |
d9e45bc3 | 5161 | |
16b6af6a | 5162 | r = write_string_file(path, one_zero(target == FREEZER_FROZEN), WRITE_STRING_FILE_DISABLE_BUFFER); |
d9e45bc3 MS |
5163 | if (r < 0) |
5164 | return r; | |
5165 | ||
16b6af6a AV |
5166 | u->freezer_state = next; |
5167 | return target != current; | |
d9e45bc3 MS |
5168 | } |
5169 | ||
047f5d63 PH |
5170 | int unit_get_cpuset(Unit *u, CPUSet *cpus, const char *name) { |
5171 | _cleanup_free_ char *v = NULL; | |
5172 | int r; | |
5173 | ||
5174 | assert(u); | |
5175 | assert(cpus); | |
5176 | ||
9cc54544 LP |
5177 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); |
5178 | if (!crt || !crt->cgroup_path) | |
047f5d63 PH |
5179 | return -ENODATA; |
5180 | ||
9cc54544 | 5181 | if ((crt->cgroup_realized_mask & CGROUP_MASK_CPUSET) == 0) |
047f5d63 PH |
5182 | return -ENODATA; |
5183 | ||
5184 | r = cg_all_unified(); | |
5185 | if (r < 0) | |
5186 | return r; | |
5187 | if (r == 0) | |
5188 | return -ENODATA; | |
48fd01e5 | 5189 | |
9cc54544 | 5190 | r = cg_get_attribute("cpuset", crt->cgroup_path, name, &v); |
047f5d63 PH |
5191 | if (r == -ENOENT) |
5192 | return -ENODATA; | |
5193 | if (r < 0) | |
5194 | return r; | |
5195 | ||
5196 | return parse_cpu_set_full(v, cpus, false, NULL, NULL, 0, NULL); | |
5197 | } | |
5198 | ||
9cc54544 LP |
5199 | CGroupRuntime *cgroup_runtime_new(void) { |
5200 | _cleanup_(cgroup_runtime_freep) CGroupRuntime *crt = NULL; | |
5201 | ||
5202 | crt = new(CGroupRuntime, 1); | |
5203 | if (!crt) | |
5204 | return NULL; | |
5205 | ||
5206 | *crt = (CGroupRuntime) { | |
5207 | .cpu_usage_last = NSEC_INFINITY, | |
5208 | ||
5209 | .cgroup_control_inotify_wd = -1, | |
5210 | .cgroup_memory_inotify_wd = -1, | |
5211 | ||
5212 | .ip_accounting_ingress_map_fd = -EBADF, | |
5213 | .ip_accounting_egress_map_fd = -EBADF, | |
5214 | ||
5215 | .ipv4_allow_map_fd = -EBADF, | |
5216 | .ipv6_allow_map_fd = -EBADF, | |
5217 | .ipv4_deny_map_fd = -EBADF, | |
5218 | .ipv6_deny_map_fd = -EBADF, | |
5219 | ||
5220 | .cgroup_invalidated_mask = _CGROUP_MASK_ALL, | |
5221 | }; | |
5222 | ||
5223 | FOREACH_ARRAY(i, crt->memory_accounting_last, ELEMENTSOF(crt->memory_accounting_last)) | |
5224 | *i = UINT64_MAX; | |
5225 | FOREACH_ARRAY(i, crt->io_accounting_base, ELEMENTSOF(crt->io_accounting_base)) | |
5226 | *i = UINT64_MAX; | |
5227 | FOREACH_ARRAY(i, crt->io_accounting_last, ELEMENTSOF(crt->io_accounting_last)) | |
5228 | *i = UINT64_MAX; | |
5229 | FOREACH_ARRAY(i, crt->ip_accounting_extra, ELEMENTSOF(crt->ip_accounting_extra)) | |
5230 | *i = UINT64_MAX; | |
5231 | ||
5232 | return TAKE_PTR(crt); | |
5233 | } | |
5234 | ||
5235 | CGroupRuntime *cgroup_runtime_free(CGroupRuntime *crt) { | |
5236 | if (!crt) | |
5237 | return NULL; | |
5238 | ||
5239 | fdset_free(crt->initial_socket_bind_link_fds); | |
5240 | #if BPF_FRAMEWORK | |
5241 | bpf_link_free(crt->ipv4_socket_bind_link); | |
5242 | bpf_link_free(crt->ipv6_socket_bind_link); | |
5243 | #endif | |
5244 | hashmap_free(crt->bpf_foreign_by_key); | |
5245 | ||
5246 | bpf_program_free(crt->bpf_device_control_installed); | |
5247 | ||
5248 | #if BPF_FRAMEWORK | |
5249 | bpf_link_free(crt->restrict_ifaces_ingress_bpf_link); | |
5250 | bpf_link_free(crt->restrict_ifaces_egress_bpf_link); | |
5251 | #endif | |
5252 | fdset_free(crt->initial_restric_ifaces_link_fds); | |
5253 | ||
5254 | safe_close(crt->ipv4_allow_map_fd); | |
5255 | safe_close(crt->ipv6_allow_map_fd); | |
5256 | safe_close(crt->ipv4_deny_map_fd); | |
5257 | safe_close(crt->ipv6_deny_map_fd); | |
5258 | ||
5259 | bpf_program_free(crt->ip_bpf_ingress); | |
5260 | bpf_program_free(crt->ip_bpf_ingress_installed); | |
5261 | bpf_program_free(crt->ip_bpf_egress); | |
5262 | bpf_program_free(crt->ip_bpf_egress_installed); | |
5263 | ||
5264 | set_free(crt->ip_bpf_custom_ingress); | |
5265 | set_free(crt->ip_bpf_custom_ingress_installed); | |
5266 | set_free(crt->ip_bpf_custom_egress); | |
5267 | set_free(crt->ip_bpf_custom_egress_installed); | |
5268 | ||
5269 | fdset_free(crt->initial_socket_bind_link_fds); | |
5270 | fdset_free(crt->initial_restric_ifaces_link_fds); | |
5271 | ||
5272 | free(crt->cgroup_path); | |
5273 | ||
5274 | return mfree(crt); | |
5275 | } | |
5276 | ||
5277 | static const char* const ip_accounting_metric_field_table[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = { | |
5278 | [CGROUP_IP_INGRESS_BYTES] = "ip-accounting-ingress-bytes", | |
5279 | [CGROUP_IP_INGRESS_PACKETS] = "ip-accounting-ingress-packets", | |
5280 | [CGROUP_IP_EGRESS_BYTES] = "ip-accounting-egress-bytes", | |
5281 | [CGROUP_IP_EGRESS_PACKETS] = "ip-accounting-egress-packets", | |
5282 | }; | |
5283 | ||
5284 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP(ip_accounting_metric_field, CGroupIPAccountingMetric); | |
5285 | ||
5286 | static const char* const io_accounting_metric_field_base_table[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
5287 | [CGROUP_IO_READ_BYTES] = "io-accounting-read-bytes-base", | |
5288 | [CGROUP_IO_WRITE_BYTES] = "io-accounting-write-bytes-base", | |
5289 | [CGROUP_IO_READ_OPERATIONS] = "io-accounting-read-operations-base", | |
5290 | [CGROUP_IO_WRITE_OPERATIONS] = "io-accounting-write-operations-base", | |
5291 | }; | |
5292 | ||
5293 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP(io_accounting_metric_field_base, CGroupIOAccountingMetric); | |
5294 | ||
5295 | static const char* const io_accounting_metric_field_last_table[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
5296 | [CGROUP_IO_READ_BYTES] = "io-accounting-read-bytes-last", | |
5297 | [CGROUP_IO_WRITE_BYTES] = "io-accounting-write-bytes-last", | |
5298 | [CGROUP_IO_READ_OPERATIONS] = "io-accounting-read-operations-last", | |
5299 | [CGROUP_IO_WRITE_OPERATIONS] = "io-accounting-write-operations-last", | |
5300 | }; | |
5301 | ||
5302 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP(io_accounting_metric_field_last, CGroupIOAccountingMetric); | |
5303 | ||
5304 | static const char* const memory_accounting_metric_field_last_table[_CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST + 1] = { | |
5305 | [CGROUP_MEMORY_PEAK] = "memory-accounting-peak", | |
5306 | [CGROUP_MEMORY_SWAP_PEAK] = "memory-accounting-swap-peak", | |
5307 | }; | |
5308 | ||
5309 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP(memory_accounting_metric_field_last, CGroupMemoryAccountingMetric); | |
5310 | ||
5311 | static int serialize_cgroup_mask(FILE *f, const char *key, CGroupMask mask) { | |
5312 | _cleanup_free_ char *s = NULL; | |
5313 | int r; | |
5314 | ||
5315 | assert(f); | |
5316 | assert(key); | |
5317 | ||
5318 | if (mask == 0) | |
5319 | return 0; | |
5320 | ||
5321 | r = cg_mask_to_string(mask, &s); | |
5322 | if (r < 0) | |
5323 | return log_error_errno(r, "Failed to format cgroup mask: %m"); | |
5324 | ||
5325 | return serialize_item(f, key, s); | |
5326 | } | |
5327 | ||
5328 | int cgroup_runtime_serialize(Unit *u, FILE *f, FDSet *fds) { | |
5329 | int r; | |
5330 | ||
5331 | assert(u); | |
5332 | assert(f); | |
5333 | assert(fds); | |
5334 | ||
5335 | CGroupRuntime *crt = unit_get_cgroup_runtime(u); | |
5336 | if (!crt) | |
5337 | return 0; | |
5338 | ||
5339 | (void) serialize_item_format(f, "cpu-usage-base", "%" PRIu64, crt->cpu_usage_base); | |
5340 | if (crt->cpu_usage_last != NSEC_INFINITY) | |
5341 | (void) serialize_item_format(f, "cpu-usage-last", "%" PRIu64, crt->cpu_usage_last); | |
5342 | ||
5343 | if (crt->managed_oom_kill_last > 0) | |
5344 | (void) serialize_item_format(f, "managed-oom-kill-last", "%" PRIu64, crt->managed_oom_kill_last); | |
5345 | ||
5346 | if (crt->oom_kill_last > 0) | |
5347 | (void) serialize_item_format(f, "oom-kill-last", "%" PRIu64, crt->oom_kill_last); | |
5348 | ||
5349 | for (CGroupMemoryAccountingMetric metric = 0; metric <= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST; metric++) { | |
5350 | uint64_t v; | |
5351 | ||
5352 | r = unit_get_memory_accounting(u, metric, &v); | |
5353 | if (r >= 0) | |
5354 | (void) serialize_item_format(f, memory_accounting_metric_field_last_to_string(metric), "%" PRIu64, v); | |
5355 | } | |
5356 | ||
5357 | for (CGroupIPAccountingMetric m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) { | |
5358 | uint64_t v; | |
5359 | ||
5360 | r = unit_get_ip_accounting(u, m, &v); | |
5361 | if (r >= 0) | |
5362 | (void) serialize_item_format(f, ip_accounting_metric_field_to_string(m), "%" PRIu64, v); | |
5363 | } | |
5364 | ||
5365 | for (CGroupIOAccountingMetric im = 0; im < _CGROUP_IO_ACCOUNTING_METRIC_MAX; im++) { | |
5366 | (void) serialize_item_format(f, io_accounting_metric_field_base_to_string(im), "%" PRIu64, crt->io_accounting_base[im]); | |
5367 | ||
5368 | if (crt->io_accounting_last[im] != UINT64_MAX) | |
5369 | (void) serialize_item_format(f, io_accounting_metric_field_last_to_string(im), "%" PRIu64, crt->io_accounting_last[im]); | |
5370 | } | |
5371 | ||
5372 | if (crt->cgroup_path) | |
5373 | (void) serialize_item(f, "cgroup", crt->cgroup_path); | |
5374 | if (crt->cgroup_id != 0) | |
5375 | (void) serialize_item_format(f, "cgroup-id", "%" PRIu64, crt->cgroup_id); | |
5376 | ||
5377 | (void) serialize_bool(f, "cgroup-realized", crt->cgroup_realized); | |
5378 | (void) serialize_cgroup_mask(f, "cgroup-realized-mask", crt->cgroup_realized_mask); | |
5379 | (void) serialize_cgroup_mask(f, "cgroup-enabled-mask", crt->cgroup_enabled_mask); | |
5380 | (void) serialize_cgroup_mask(f, "cgroup-invalidated-mask", crt->cgroup_invalidated_mask); | |
5381 | ||
5382 | (void) bpf_socket_bind_serialize(u, f, fds); | |
5383 | ||
5384 | (void) bpf_program_serialize_attachment(f, fds, "ip-bpf-ingress-installed", crt->ip_bpf_ingress_installed); | |
5385 | (void) bpf_program_serialize_attachment(f, fds, "ip-bpf-egress-installed", crt->ip_bpf_egress_installed); | |
5386 | (void) bpf_program_serialize_attachment(f, fds, "bpf-device-control-installed", crt->bpf_device_control_installed); | |
5387 | (void) bpf_program_serialize_attachment_set(f, fds, "ip-bpf-custom-ingress-installed", crt->ip_bpf_custom_ingress_installed); | |
5388 | (void) bpf_program_serialize_attachment_set(f, fds, "ip-bpf-custom-egress-installed", crt->ip_bpf_custom_egress_installed); | |
5389 | ||
5390 | (void) bpf_restrict_ifaces_serialize(u, f, fds); | |
5391 | ||
5392 | return 0; | |
5393 | } | |
5394 | ||
5395 | #define MATCH_DESERIALIZE(u, key, l, v, parse_func, target) \ | |
5396 | ({ \ | |
5397 | bool _deserialize_matched = streq(l, key); \ | |
5398 | if (_deserialize_matched) { \ | |
5399 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); \ | |
5400 | if (!crt) \ | |
5401 | log_oom_debug(); \ | |
5402 | else { \ | |
5403 | int _deserialize_r = parse_func(v); \ | |
5404 | if (_deserialize_r < 0) \ | |
5405 | log_unit_debug_errno(u, _deserialize_r, \ | |
5406 | "Failed to parse \"%s=%s\", ignoring.", l, v); \ | |
5407 | else \ | |
5408 | crt->target = _deserialize_r; \ | |
5409 | } \ | |
5410 | } \ | |
5411 | _deserialize_matched; \ | |
5412 | }) | |
5413 | ||
5414 | #define MATCH_DESERIALIZE_IMMEDIATE(u, key, l, v, parse_func, target) \ | |
5415 | ({ \ | |
5416 | bool _deserialize_matched = streq(l, key); \ | |
5417 | if (_deserialize_matched) { \ | |
5418 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); \ | |
5419 | if (!crt) \ | |
5420 | log_oom_debug(); \ | |
5421 | else { \ | |
5422 | int _deserialize_r = parse_func(v, &crt->target); \ | |
5423 | if (_deserialize_r < 0) \ | |
5424 | log_unit_debug_errno(u, _deserialize_r, \ | |
5425 | "Failed to parse \"%s=%s\", ignoring", l, v); \ | |
5426 | } \ | |
5427 | } \ | |
5428 | _deserialize_matched; \ | |
5429 | }) | |
5430 | ||
5431 | #define MATCH_DESERIALIZE_METRIC(u, key, l, v, parse_func, target) \ | |
5432 | ({ \ | |
5433 | bool _deserialize_matched = streq(l, key); \ | |
5434 | if (_deserialize_matched) { \ | |
5435 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); \ | |
5436 | if (!crt) \ | |
5437 | log_oom_debug(); \ | |
5438 | else { \ | |
5439 | int _deserialize_r = parse_func(v); \ | |
5440 | if (_deserialize_r < 0) \ | |
5441 | log_unit_debug_errno(u, _deserialize_r, \ | |
5442 | "Failed to parse \"%s=%s\", ignoring.", l, v); \ | |
5443 | else \ | |
5444 | crt->target = _deserialize_r; \ | |
5445 | } \ | |
5446 | } \ | |
5447 | _deserialize_matched; \ | |
5448 | }) | |
5449 | ||
5450 | int cgroup_runtime_deserialize_one(Unit *u, const char *key, const char *value, FDSet *fds) { | |
5451 | int r; | |
5452 | ||
5453 | assert(u); | |
5454 | assert(value); | |
5455 | ||
5456 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
5457 | return 0; | |
5458 | ||
5459 | if (MATCH_DESERIALIZE_IMMEDIATE(u, "cpu-usage-base", key, value, safe_atou64, cpu_usage_base) || | |
5460 | MATCH_DESERIALIZE_IMMEDIATE(u, "cpuacct-usage-base", key, value, safe_atou64, cpu_usage_base)) | |
5461 | return 1; | |
5462 | ||
5463 | if (MATCH_DESERIALIZE_IMMEDIATE(u, "cpu-usage-last", key, value, safe_atou64, cpu_usage_last)) | |
5464 | return 1; | |
5465 | ||
5466 | if (MATCH_DESERIALIZE_IMMEDIATE(u, "managed-oom-kill-last", key, value, safe_atou64, managed_oom_kill_last)) | |
5467 | return 1; | |
5468 | ||
5469 | if (MATCH_DESERIALIZE_IMMEDIATE(u, "oom-kill-last", key, value, safe_atou64, oom_kill_last)) | |
5470 | return 1; | |
5471 | ||
5472 | if (streq(key, "cgroup")) { | |
5473 | r = unit_set_cgroup_path(u, value); | |
5474 | if (r < 0) | |
5475 | log_unit_debug_errno(u, r, "Failed to set cgroup path %s, ignoring: %m", value); | |
5476 | ||
5477 | (void) unit_watch_cgroup(u); | |
5478 | (void) unit_watch_cgroup_memory(u); | |
5479 | return 1; | |
5480 | } | |
5481 | ||
5482 | if (MATCH_DESERIALIZE_IMMEDIATE(u, "cgroup-id", key, value, safe_atou64, cgroup_id)) | |
5483 | return 1; | |
5484 | ||
5485 | if (MATCH_DESERIALIZE(u, "cgroup-realized", key, value, parse_boolean, cgroup_realized)) | |
5486 | return 1; | |
5487 | ||
5488 | if (MATCH_DESERIALIZE_IMMEDIATE(u, "cgroup-realized-mask", key, value, cg_mask_from_string, cgroup_realized_mask)) | |
5489 | return 1; | |
5490 | ||
5491 | if (MATCH_DESERIALIZE_IMMEDIATE(u, "cgroup-enabled-mask", key, value, cg_mask_from_string, cgroup_enabled_mask)) | |
5492 | return 1; | |
5493 | ||
5494 | if (MATCH_DESERIALIZE_IMMEDIATE(u, "cgroup-invalidated-mask", key, value, cg_mask_from_string, cgroup_invalidated_mask)) | |
5495 | return 1; | |
5496 | ||
5497 | if (STR_IN_SET(key, "ipv4-socket-bind-bpf-link-fd", "ipv6-socket-bind-bpf-link-fd")) { | |
5498 | int fd; | |
5499 | ||
5500 | fd = deserialize_fd(fds, value); | |
5501 | if (fd >= 0) | |
5502 | (void) bpf_socket_bind_add_initial_link_fd(u, fd); | |
5503 | ||
5504 | return 1; | |
5505 | } | |
5506 | ||
5507 | if (STR_IN_SET(key, | |
5508 | "ip-bpf-ingress-installed", "ip-bpf-egress-installed", | |
5509 | "bpf-device-control-installed", | |
5510 | "ip-bpf-custom-ingress-installed", "ip-bpf-custom-egress-installed")) { | |
5511 | ||
5512 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); | |
5513 | if (!crt) | |
5514 | log_oom_debug(); | |
5515 | else { | |
5516 | if (streq(key, "ip-bpf-ingress-installed")) | |
5517 | (void) bpf_program_deserialize_attachment(value, fds, &crt->ip_bpf_ingress_installed); | |
5518 | ||
5519 | if (streq(key, "ip-bpf-egress-installed")) | |
5520 | (void) bpf_program_deserialize_attachment(value, fds, &crt->ip_bpf_egress_installed); | |
5521 | ||
5522 | if (streq(key, "bpf-device-control-installed")) | |
5523 | (void) bpf_program_deserialize_attachment(value, fds, &crt->bpf_device_control_installed); | |
5524 | ||
5525 | if (streq(key, "ip-bpf-custom-ingress-installed")) | |
5526 | (void) bpf_program_deserialize_attachment_set(value, fds, &crt->ip_bpf_custom_ingress_installed); | |
5527 | ||
5528 | if (streq(key, "ip-bpf-custom-egress-installed")) | |
5529 | (void) bpf_program_deserialize_attachment_set(value, fds, &crt->ip_bpf_custom_egress_installed); | |
5530 | } | |
5531 | ||
5532 | return 1; | |
5533 | } | |
5534 | ||
5535 | if (streq(key, "restrict-ifaces-bpf-fd")) { | |
5536 | int fd; | |
5537 | ||
5538 | fd = deserialize_fd(fds, value); | |
5539 | if (fd >= 0) | |
5540 | (void) bpf_restrict_ifaces_add_initial_link_fd(u, fd); | |
5541 | return 1; | |
5542 | } | |
5543 | ||
5544 | CGroupMemoryAccountingMetric mm = memory_accounting_metric_field_last_from_string(key); | |
5545 | if (mm >= 0) { | |
5546 | uint64_t c; | |
5547 | ||
5548 | r = safe_atou64(value, &c); | |
5549 | if (r < 0) | |
5550 | log_unit_debug(u, "Failed to parse memory accounting last value %s, ignoring.", value); | |
5551 | else { | |
5552 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); | |
5553 | if (!crt) | |
5554 | log_oom_debug(); | |
5555 | else | |
5556 | crt->memory_accounting_last[mm] = c; | |
5557 | } | |
5558 | ||
5559 | return 1; | |
5560 | } | |
5561 | ||
5562 | CGroupIPAccountingMetric ipm = ip_accounting_metric_field_from_string(key); | |
5563 | if (ipm >= 0) { | |
5564 | uint64_t c; | |
5565 | ||
5566 | r = safe_atou64(value, &c); | |
5567 | if (r < 0) | |
5568 | log_unit_debug(u, "Failed to parse IP accounting value %s, ignoring.", value); | |
5569 | else { | |
5570 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); | |
5571 | if (!crt) | |
5572 | log_oom_debug(); | |
5573 | else | |
5574 | crt->ip_accounting_extra[ipm] = c; | |
5575 | } | |
5576 | ||
5577 | return 1; | |
5578 | } | |
5579 | ||
5580 | CGroupIOAccountingMetric iom = io_accounting_metric_field_base_from_string(key); | |
5581 | if (iom >= 0) { | |
5582 | uint64_t c; | |
5583 | ||
5584 | r = safe_atou64(value, &c); | |
5585 | if (r < 0) | |
5586 | log_unit_debug(u, "Failed to parse IO accounting base value %s, ignoring.", value); | |
5587 | else { | |
5588 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); | |
5589 | if (!crt) | |
5590 | log_oom_debug(); | |
5591 | else | |
5592 | crt->io_accounting_base[iom] = c; | |
5593 | } | |
5594 | ||
5595 | return 1; | |
5596 | } | |
5597 | ||
5598 | iom = io_accounting_metric_field_last_from_string(key); | |
5599 | if (iom >= 0) { | |
5600 | uint64_t c; | |
5601 | ||
5602 | r = safe_atou64(value, &c); | |
5603 | if (r < 0) | |
5604 | log_unit_debug(u, "Failed to parse IO accounting last value %s, ignoring.", value); | |
5605 | else { | |
5606 | CGroupRuntime *crt = unit_setup_cgroup_runtime(u); | |
5607 | if (!crt) | |
5608 | log_oom_debug(); | |
5609 | else | |
5610 | crt->io_accounting_last[iom] = c; | |
5611 | } | |
5612 | return 1; | |
5613 | } | |
5614 | ||
5615 | return 0; | |
5616 | } | |
5617 | ||
4e806bfa AZ |
5618 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
5619 | [CGROUP_DEVICE_POLICY_AUTO] = "auto", | |
5620 | [CGROUP_DEVICE_POLICY_CLOSED] = "closed", | |
5621 | [CGROUP_DEVICE_POLICY_STRICT] = "strict", | |
5622 | }; | |
5623 | ||
4ad49000 | 5624 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |
d9e45bc3 | 5625 | |
6bb00842 | 5626 | static const char* const cgroup_pressure_watch_table[_CGROUP_PRESSURE_WATCH_MAX] = { |
16b6af6a | 5627 | [CGROUP_PRESSURE_WATCH_OFF] = "off", |
6bb00842 | 5628 | [CGROUP_PRESSURE_WATCH_AUTO] = "auto", |
16b6af6a | 5629 | [CGROUP_PRESSURE_WATCH_ON] = "on", |
6bb00842 LP |
5630 | [CGROUP_PRESSURE_WATCH_SKIP] = "skip", |
5631 | }; | |
5632 | ||
5633 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(cgroup_pressure_watch, CGroupPressureWatch, CGROUP_PRESSURE_WATCH_ON); | |
435996e6 DDM |
5634 | |
5635 | static const char* const cgroup_ip_accounting_metric_table[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = { | |
5636 | [CGROUP_IP_INGRESS_BYTES] = "IPIngressBytes", | |
5637 | [CGROUP_IP_EGRESS_BYTES] = "IPEgressBytes", | |
5638 | [CGROUP_IP_INGRESS_PACKETS] = "IPIngressPackets", | |
5639 | [CGROUP_IP_EGRESS_PACKETS] = "IPEgressPackets", | |
5640 | }; | |
5641 | ||
5642 | DEFINE_STRING_TABLE_LOOKUP(cgroup_ip_accounting_metric, CGroupIPAccountingMetric); | |
5643 | ||
5644 | static const char* const cgroup_io_accounting_metric_table[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
5645 | [CGROUP_IO_READ_BYTES] = "IOReadBytes", | |
5646 | [CGROUP_IO_WRITE_BYTES] = "IOWriteBytes", | |
5647 | [CGROUP_IO_READ_OPERATIONS] = "IOReadOperations", | |
5648 | [CGROUP_IO_WRITE_OPERATIONS] = "IOWriteOperations", | |
5649 | }; | |
5650 | ||
5651 | DEFINE_STRING_TABLE_LOOKUP(cgroup_io_accounting_metric, CGroupIOAccountingMetric); | |
9824ab1f MY |
5652 | |
5653 | static const char* const cgroup_memory_accounting_metric_table[_CGROUP_MEMORY_ACCOUNTING_METRIC_MAX] = { | |
5654 | [CGROUP_MEMORY_PEAK] = "MemoryPeak", | |
5655 | [CGROUP_MEMORY_SWAP_CURRENT] = "MemorySwapCurrent", | |
5656 | [CGROUP_MEMORY_SWAP_PEAK] = "MemorySwapPeak", | |
5657 | [CGROUP_MEMORY_ZSWAP_CURRENT] = "MemoryZSwapCurrent", | |
5658 | }; | |
5659 | ||
5660 | DEFINE_STRING_TABLE_LOOKUP(cgroup_memory_accounting_metric, CGroupMemoryAccountingMetric); | |
4fb0d2dc | 5661 | |
8ad61489 | 5662 | static const char *const cgroup_effective_limit_type_table[_CGROUP_LIMIT_TYPE_MAX] = { |
4fb0d2dc MK |
5663 | [CGROUP_LIMIT_MEMORY_MAX] = "EffectiveMemoryMax", |
5664 | [CGROUP_LIMIT_MEMORY_HIGH] = "EffectiveMemoryHigh", | |
5665 | [CGROUP_LIMIT_TASKS_MAX] = "EffectiveTasksMax", | |
5666 | }; | |
5667 | ||
8ad61489 | 5668 | DEFINE_STRING_TABLE_LOOKUP(cgroup_effective_limit_type, CGroupLimitType); |